The public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources,
gathering and maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection
of information, including suggestions for reducing this burden to Department of Defense, Washington Headquarters Services, Directorate for Information Operations and Reports (0704–0188),
1215 Jefferson Davis Highway, Suite 1204, Arlington, VA 22202–4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to any penalty
for failing to comply with a collection of information if it does not display a currently valid OMB control number. PLEASE DO NOT RETURN YOUR FORM TO THE ABOVE
ADDRESS.

1. REPORT DATE (DD–MM–YYYY)

2. REPORT TYPE

31–8–2013

Technical Report

3. DATES COVERED (From — To)

2012-10-01—2013-09-15

4. TITLE AND SUBTITLE

5a. CONTRACT NUMBER

5b. GRANT NUMBER

Bulk Extractor 1.4 User’s Manual
5c. PROGRAM ELEMENT NUMBER

6. AUTHOR(S)

5d. PROJECT NUMBER

5e. TASK NUMBER

Jessica R. Bradley, Simson L. Garfinkel
5f. WORK UNIT NUMBER

7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES)

8. PERFORMING ORGANIZATION REPORT
NUMBER

Naval Postgraduate School
Monterey, CA 93943

NPS-CS-13-006

9. SPONSORING / MONITORING AGENCY NAME(S) AND ADDRESS(ES)

10. SPONSOR/MONITOR’S ACRONYM(S)

Defense Intelligence Agency

11. SPONSOR/MONITOR’S REPORT
NUMBER(S)

12. DISTRIBUTION / AVAILABILITY STATEMENT

Approved for public release; distribution is unlimited
13. SUPPLEMENTARY NOTES

The views expressed in this report are those of the author and do not reflect the official policy or position of the Department of
Defense or the U.S. Government.
14. ABSTRACT

bulk extractor is a program that extracts features such as email addresses, credit card numbers, URLs, and other types of
information from digital evidence files. It is a useful forensic investigation tool for many tasks such as malware and intrusion
investigations, identity investigations and cyber investigations, as well as analyzing imagery and password cracking. This
document provides instructions on how to use the program in Windows, Linux and Mac environments.

15. SUBJECT TERMS

bulk extractor, User’s Manual
16. SECURITY CLASSIFICATION OF:
a. REPORT

b. ABSTRACT

c. THIS PAGE

Unclassified Unclassified Unclassified
NSN 7540-01-280-5500

17. LIMITATION OF
ABSTRACT

18. NUMBER
OF
PAGES

19a. NAME OF RESPONSIBLE PERSON
19b. TELEPHONE NUMBER (include area code)

Standard Form 298 (Rev. 8–98)
Prescribed by ANSI Std. Z39.18

THIS PAGE INTENTIONALLY LEFT BLANK

bulk extractor

USERS MANUAL
Quickstart Guide Included

August 28, 2013

Authored by:
Jessica R. Bradley
Simson L. Garfinkel

One Page Quickstart for Linux & Mac Users
This page provides a very brief introduction to downloading, installing and running bulk_extractor.
1. If you do not already have one, obtain a disk image on which to run bulk_extractor.
Sample images can be downloaded from http://digitalcorpora.org/corpora/
disk-images. Suggestions include nps-2009-domexusers and
nps-2009-ubnist1.gen3.
2. Download the latest version of bulk_extractor. It can be obtained from http://digitalcorpora.
org/downloads/bulk_extractor/. The file is called bulk_extractor-x.y.z.tar.gz
where x.y.z is the latest version.
3. Un-tar and un-zip the file. In the newly created bulk_extractor-x.y directory, run the
following commands:
./configure
make
sudo make install
[Refer to Subsubsection 3.1.1 Installing on Linux or Mac. Note, for full functionality,
some users may need to first download and install dependent library files. Instructions are
outlined in the referenced section.]
4. To run bulk_extractor from the command line, type the following instructions:
bulk_extractor -o output mydisk.raw
In the above instructions, output is directory that will be created to store bulk_extractor
results. It can not already exist. The input mydisk.raw is the disk image to be processed.
[See Subsection 3.2 Run bulk_extractor from Command Line]
5. To run bulk_extractor from the Bulk Extractor Viewer, navigate to the directory called
/java_gui in the bulk_extractor folder and run the following command:
./BEViewer
In the Bulk Extractor Viewer, click on the Gear/down arrow icon as depicted below.

A window will pop up and the first two input boxes allow you to select an Image File
and specify an Output Feature Directory to create. Enter both of those and then select
the button at the bottom of the window titled "Start bulk_extractor" to run bulk_extractor.
[See Subsection 3.3 Run bulk_extractor from Bulk Extractor Viewer ]
6. Whether bulk_extractor was run from the command line or the Bulk Extractor Viewer
tool, after the run the resulting output files will be contained in the specified output
directory. Open that directory and verify files have been created. There should be 15-25
files. Some will be empty and others will be populated with data.
7. Users can join the google email users group for more information and help with any issues
encountered. Email bulk_extractor-users+subscribe@googlegroups.com with a blank
message to join.
ii

One Page Quickstart for Windows Users
This page provides a very brief introduction to downloading, installing and running bulk_extractor.
1. If you do not already have one, obtain a disk image on which to run bulk_extractor.
Sample images can be downloaded from http://digitalcorpora.org/corpora/
disk-images. Suggestions include nps-2009-domexusers and
nps-2009-ubnist1.gen3.
2. Download the latest version of the bulk_extractor Windows installer. It can be obtained from http://digitalcorpora.org/downloads/bulk_extractor. The file
to download is called bulk_extractor-x.y.z-windowsinstaller.exe where x.y.z
is the latest version number. Run the installer file. This will automatically install
bulk_extractor on your machine. The automatic installation includes the complete
bulk_extractor system as well as the Bulk Extractor Viewer tool. [See Subsubsection 3.1.2 Installing on Windows]
3. To run bulk_extractor from the command line, type the following instructions:
bulk_extractor -o output mydisk.raw
In the above instructions, output is directory that will be created to store bulk_extractor
results. It can not already exist. The input mydisk.raw is the disk image to be processed.
[See Subsection 3.2 Run bulk_extractor from Command Line]
4. To run bulk_extractor from the Bulk Extractor Viewer, run the program Bulk Extractor
X.Y from the Start Menu.
In the Bulk Extractor Viewer, click on the Gear/down arrow icon as depicted below.

A window will pop up and the first two input boxes allow you to select an Image File
and specify an Output Feature Directory to create. Enter both of those and then select
the button at the bottom of the window titled "Start bulk_extractor" to run bulk_extractor.
[See Subsection 3.3 Run bulk_extractor from Bulk Extractor Viewer ]
5. Whether bulk_extractor was run from the command line or the Bulk Extractor Viewer
tool, after the run the resulting output files will be contained in the specified output
directory. Open that directory and verify files have been created. There should be 15-25
files. Some will be empty and others will be populated with data.
6. Users can join the google email users group for more information and help with any issues
encountered. Email bulk_extractor-users+subscribe@googlegroups.com with a blank
message to join.

iii

Contents
1 Introduction
1.1 Overview of bulk_extractor . . . . . . .
1.1.1 A bulk_extractor Success Story
1.2 Purpose of this Manual . . . . . . . . .
1.3 Conventions Used in this Manual . . . .

.
.
.
.

2 How bulk_extractor Works

1
1
2
3
3
3

3 Running bulk_extractor
3.1 Installation Guide . . . . . . . . . . . . . . . . .
3.1.1 Installing on Linux or Mac . . . . . . . .
3.1.2 Installing on Windows . . . . . . . . . .
3.2 Run bulk_extractor from Command Line . . . .
3.3 Run bulk_extractor from Bulk Extractor Viewer

.
.
.
.
.

6
6
6
7
9
11

4 Processing Data
4.1 Types of Input Data . . . . . . . . . . . . . . .
4.2 Scanners . . . . . . . . . . . . . . . . . . . . .
4.3 Carving . . . . . . . . . . . . . . . . . . . . .
4.4 Suppressing False Positives . . . . . . . . . . .
4.5 Using an Alert List . . . . . . . . . . . . . . .
4.6 The Importance of Compressed Data Processing

.
.
.
.
.
.

18
18
19
22
23
25
25

5 Use Cases for bulk_extractor
5.1 Malware Investigations . . . . . . . . . . . . . . . . . . .
5.2 Cyber Investigations . . . . . . . . . . . . . . . . . . . .
5.3 Identity Investigations . . . . . . . . . . . . . . . . . . . .
5.4 Password Cracking . . . . . . . . . . . . . . . . . . . . .
5.5 Analyzing Imagery Information . . . . . . . . . . . . . .
5.6 Using bulk_extractor in a Highly Specialized Environment

.
.
.
.
.
.

26
26
27
28
29
31
31

.
.
.
.
.
.

6 Tuning bulk_extractor

7 Post Processing Capabilities
7.1 bulk_diff.py: Difference Between Runs . . . . . . . . . . . . . . . . . . . . .
7.2 identify_filenames.py: Identify File Origin of Features . . . . . . . . . . . . .

32
32
33

8 Worked Examples
8.1 Encoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

33
33

9 2009-M57 Patents Scenario
9.1 Run bulk_extractor with the Data .
9.2 Digital Media Triage . . . . . . .
9.3 Analyzing Imagery . . . . . . . .
9.4 Password Cracking . . . . . . . .
9.5 Post Processing . . . . . . . . . .

34
34
36
41
42
44

.
.
.
.
.

10 NPS DOMEX Users Image
10.1 Malware Investigations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.2 Cyber Investigations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

45
47
49

11 Troubleshooting

12 Related Reading

Appendices

A Output of bulk_extractor Help Command

1 Introduction
1.1 Overview of bulk_extractor
bulk_extractor is a program that extracts features such as email addresses, credit card numbers,
URLs, and other types of information from digital evidence files. It is a useful forensic investigation tool for many tasks such as malware and intrusion investigations, identity investigations
and cyber investigations, as well as analyzing imagery and password cracking. The program
provides several unusual capabilities including:
• It finds email addresses, URLs and credit card numbers that other tools miss because it
can process compressed data (like ZIP, PDF and GZIP files) and incomplete or partially
corrupted data. It can carve JPEGs, office documents and other kinds of files out of
fragments of compressed data. It will detect and carve encrypted RAR files.
• It builds word lists based on all of the words found within the data, even those in compressed files that are in unallocated space. Those word lists can be useful for password
cracking.
• It is multi-threaded; running bulk_extractor on a computer with twice the number of cores
typically makes it complete a run in half the time.
• It creates histograms showing the most common email addresses, URLs, domains, search
terms and other kinds of information on the drive.
bulk_extractor operates on disk images, files or a directory of files and extracts useful information without parsing the file system or file system structures. The input is split into pages
and processed by one or more scanners. The results are stored in feature files that can be
easily inspected, parsed, or processed with other automated tools. bulk_extractor also creates
histograms of features that it finds. This is useful because features such as email addresses and
internet search terms that are more common tend to be important.
In addition to the capabilities described above, bulk_extractor also includes
• A graphical user interface, Bulk Extractor Viewer, for browsing features stored in feature
files and for launching bulk_extractor scans
• A small number of python programs for performing additional analysis on feature files
bulk_extractor 1.4 detects and optimistically decompresses data in ZIP, GZIP, RAR, and Microsoft’s Hibernation files. This has proven useful, for example, in recovering email addresses
from fragments of compressed files found in unallocated space.
bulk_extractor contains a simple but effective mechanism for protecting against decompression
bombs. It also has capabilities specifically designed for Windows and malware analysis including
decoders for the Windows PE, Linux ELF, VCARD, Base16, Base64 and Windows directory
formats.
bulk_extractor gets its speed through the use of compiled search expressions and multi-threading.
The search expressions are written as pre-compiled regular expressions, essentially allowing
bulk_extractor to perform searches on disparate terms in parallel. Threading is accomplished
through the use of an analysis thread pool. After the features have been extracted, bulk_extractor
builds a histogram of email addresses, Google search terms, and other extracted features. Stop
1

lists can also be used to remove features not relevant to a case.
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.
This means that an 8-core machine will process a disk image roughly 8 times faster than a
1-core machine. bulk_extractor is also thorough. It automatically detects, decompresses, and
recursively re-processes data that has been compressed with a variety of algorithms. Our testing
has shown there is a significant amount of compressed data in the unallocated regions of file
systems missed by most forensics tools that are commonly in use today[5]. Another advantage of
ignoring file systems is that bulk_extractor can be used to process any kind of digital media. The
program has been used to process hard drives, SSDs, optical media, camera cards, cell phones,
network packet dumps, and other kinds of digital information.
Between 2005 and 2008, the bulk_extractor team interviewed law enforcement regarding their
use of forensic tools. Law enforcement officers wanted a highly automated tool for finding
email addresses and credit card numbers (including track 2 information), phone numbers, GPS
coordinates and EXIF information from JPEGs, search terms (extracted from URLs), and all
words that were present on the disk (for password cracking). The tool needed to run on Windows,
Linux and Mac-based systems with no user interaction. It also had to operate on raw disk images,
split-raw volumes and E01 files files. The tool needed to run at the maximum I/O speed of
the physical drive and never crash. Through these interviews, the initial requirements for the
bulk_extractor system were developed. Over the past five years, we have worked to create the
tool that those officers desired.

1.1.1

A bulk_extractor Success Story

One early bulk_extractor success story comes from the City of San Luis Obispo Police Department in the Spring of 2010. The District Attorney filed charges against two individuals for
credit card Fraud and possession of materials to commit credit card fraud. The defendants were
arrested with a computer. Defense attorneys were expected to argue that the defendants were
unsophisticated and lacked knowledge to commit the crime. The examiner was given a 250 GB
drive the day before the preliminary hearing; typically it would take several days to conduct a
proper forensic investigation of that much data.
bulk_extractor found actionable evidence in only two and a half hours including the following
information:
• There were over 10,000 credit card numbers on the hard drive (illegal materials). Over
1000 of the credit card numbers were unique.
• The most common email address belonged to the primary defendant (evidence of possession).
• The most commonly occurring internet search engine queries concerned credit card fraud
and bank identification numbers (evidence of intent).
• The most commonly visited websites were in a foreign country whose primary language
is spoken by the defendant (evidence of flight risk).
Armed with this data, the defendants were held without bail.

As bulk_extractor has been deployed and used in different applications, it has evolved to meet
additional requirements. This manual describes use cases for the bulk_extractor system and
demonstrates how users can take full advantage of all of its capabilities.

1.2 Purpose of this Manual
This Users Manual is intended to be useful to new, intermediate and experienced users of
bulk_extractor. It provides an in-depth review of the functionality included in bulk_extractor
and shows how to access and utilize features through both command line operation and the Bulk
Extractor Viewer. This manual includes working examples with links to the input data (disk
images) used, giving users the opportunity to work through the examples and utilize all aspects
of the system.

1.3 Conventions Used in this Manual
This manual uses standard formatting conventions to highlight file names, directory names and
example commands. The conventions for those specific types are described in this section.
Names of programs including the post-processing tools native to bulk_extractor and third-party
tools are shown in bold, as in tcpflow.
File names are displayed in a fixed width font. They will appear as filename.txt within the
text throughout the manual.
Directory names are displayed in italics. They appear as directoryname/ within the text. The
only exception is for directory names that are part of an example command. Directory names
referenced in example commands appear in the example command format.
Scanner names are denoted with bold, italicized text. They are always specified in lower-case,
because that is how they are referred in the options and usage information for bulk_extractor.
Names will appear as scannername.
This manual contains example commands that should be typed in by the user. A command
entered at the terminal is shown like this:
command
The first character on the line is the terminal prompt, and should not be typed. The black square
is used as the standard prompt in this manual, although the prompt shown on a users screen will
vary according to the system they are using.

How bulk_extractor Works

bulk_extractor finds email addresses, URLS, and CCNs that other tools miss. This is due in
part to the fact that bulk_extractor optimistically decompresses and re-analyzes all data (e.g. zip
fragments, gzip browser cache runs). The decompression operates on incomplete and corrupted
data until decompression fails. bulk_extractor can also build word lists for password cracking
There are three phases of operation in bulk_extractor: feature extraction, histogram creation,
post processing as shown in Figure 1. The output feature files contain extracted data designed
3

Disk image
files
...

EXTRACT FEATURES

HISTOGRAM
CREATION

.E01
.aff
.dd
.000, .001

POST PROCESSING

DONE

report.xml
— log file
telephone.txt — list of phone numbers with context
telephone_histogram.txt — histogram of phone numbers
vcard/
— directory of VCARDs
...

Figure 1: Three Phases of bulk_extractor Operation
for easy processing by third party programs or use in spreadsheet tools. The bulk_extractor
histogram system automatically summarizes features.
Features files are written using the feature recording system. As features are discovered, they
are sent to the feature recorder and recorded in the appropriate file. Multiple scanners might
write to the same feature file. For example, the exif scanner searches the file formats used
by digital cameras and finds GPS coordinates in images. Those findings are written to the
output file gps.txt by the gps feature recorder. A separate scanner, the gps scanner, searches
Garmin Trackpoint data and also finds GPS coordinates and writes them to gps.txt. It is
worth noting that some scanners also find more than one type of feature and write to several
feature files. For example, the email scanner looks for email addresses, domains, URLs and
RFC822 headers and writes them to email.txt, domain.txt, url.txt, rfc822.txt and
ether.txt respectively.
A feature file contains rows of features. Each row is typically comprised of an offset, a feature,
and the feature in evidence context although scanners are free to store whatever information they
wish. A few lines of an email feature file might look like the following:
OFFSET
48198832
48200361
48413823

FEATURE
domexuser2@gmail.com
domexuser2@live.com
siege@preoccupied.net

FEATURE IN EVIDENCE CONTEXT
__domexuser2@gmail.com/Home
__domexuser2@live.com_l

The types of features displayed in the feature file will vary depending on what type of feature is
being stored. However, all feature files use the same format with each row corresponding to one
found instance of a feature and three columns describing the related data (offset, feature, and
feature in evidence context).
Histograms are a powerful tool for understanding certain kinds of evidence. A histogram of
emails allows us to rapidly determine the drive’s primary user, the user’s organization, primary
4

correspondents and other email addresses. The feature recording system automatically makes
histograms as data are processed. When the scanner writes to the feature recording system, the
relevant histograms are automatically updated.
A histogram file will, in general, look like the following file excerpt:
n=875
n=651
n=605
...
n=288
n=281
n=226
n=225
n=218
n=210
n=201
n=186

mozilla@kewis.ch (utf16=3)
charlie@m57.biz (utf16=120)
ajbanck@planet.nl
mattwillis@gmail.com
garths@oeone.com
michael.buettner@sun.com (utf16=2)
bugzilla@babylonsounds.com
berend.cornelius@sun.com
ips@mail.ips.es
mschroeder@mozilla.x-home.org
pat@m57.biz (utf16=1)

Each line shows a feature and the number of times that feature was found by bulk_extractor (the
histogram indicates how many times the item was found coded as UTF-16). Features are stored
in the file in order of occurrence with most frequent features appearing at the top of the file and
least frequent displayed at the bottom.
bulk_extractor has multiple scanners that extract features. Each scanner runs in an arbitrary order.
Scanners can be enabled or disabled which can be useful for debugging and speed optimization.
Some scanners are recursive and actually expand the data they are exploring, thereby creating
more data that bulk_extractor can analyze. These blocks are called sbufs. The "s" stands for the
word safe. All access to data in the sbuf is bounds-checked, so buffer overflow events are very
unlikely. The sbuf data structure is one of the reasons that bulk_extractor is so crash resistant.
Recursion is used for, among other things, decompressing ZLIB and Windows HIBERFILE,
extracting text from PDFs and handling compressed browser cache data.
The recursion process requires a new way to describe offsets. To do this, bulk_extractor
introduces the concept of the “forensic path.” The forensic path is a description of the origination
of a piece of data. It might come from, for example, a flat file, a data stream, or a decompression
of some type of data. Consider an HTTP stream that contains a GZIP-compressed email as
shown in Figure 2. A series of scanners will first find the ZLIB compressed regions in the HTTP
stream that contain the email, decompress them, and then find the features in that email which
may include email addresses, names and phone numbers. Using this method, bulk_extractor
can find email addresses in compressed data. The forensic path for the email addresses found
indicate that it originated in an email, that was GZIP compressed and found in an HTTP stream.
The forensic path of the email addresses features found might be represented as follows:
11052168704-GZIP-3437
11052168704-GZIP-3475
11052168704-GZIP-3512

live.com
live.com
live.com

eMn=’domexuser@live.com’;var
pMn=’domexuser@live.com’;var
eCk=’domexuser@live.com’;var

srf_sDispM
srf_sDreCk
srf_sFT=’<

The full functionality of bulk_extractor is provided both through command line operation and the
GUI tool, Bulk Extractor Viewer. Both modes of operation work for Linux, Mac and Windows.
The following section describes how to download, install and run bulk_extractor using either the
command line or the Bulk Extractor Viewer.

Figure 2: Forensic path of features found in email lead back to HTTP Stream

Running bulk_extractor

bulk_extractor is a command line tool with an accompanying graphical user interface tool, Bulk
Extractor Viewer. All of the command line functionality of bulk_extractor is also available in
the Bulk Extractor Viewer. Users can access the functionality in whichever way they prefer. In
this manual we review the bulk_extractor user options in both formats.
bulk_extractor can be run on a Linux, MacOS or Windows system. The fastest way to run
bulk_extractor is using Linux on a Linux system. Running on Windows provides the same
results, but the run will typically take 40

3.1 Installation Guide
Installation instructions vary for Linux/Mac users and Windows users. The following sections
explain how to install bulk_extractor on those systems.
3.1.1

Installing on Linux or Mac

Before compiling bulk_extractor for your platform, you may need to install other packages on
your system which bulk_extractor requires to compile cleanly and with a full set of capabilities.
Dependencies for Linux Fedora
This command should add the appropriate packages:
sudo yum update
sudo yum groupinstall development-tools
sudo yum install flex

Dependencies for Linux Debian Testing (wheezy) or Ubuntu 13.0
The following command should add the appropriate libraries:
sudo apt-get -y install gcc g++ flex libewf-dev
Dependencies for Mac Systems
Mac users must first install Apple’s Xcode development system. Other components should be
downloaded using the MacPorts system. If you do not have MacPorts, go to the App store and
download and install it. It is free. Once it is installed, try:
sudo port install flex autoconf automake libewf-devel
Mac users should note that libewf-devel may not be available in ports. If it is not, download and
un-tar the libewf source, cd into the directory and run:
./configure
make
sudo make install
Download and Install bulk_extractor
Next, download the latest version of bulk_extractor. The software can be downloaded from
http://digitalcorpora.org/downloads/bulk_extractor/. The file to download will
be called bulk_extractor-x.y.z.tar.gz where x.y.z is the latest version. As of publication
of this manual, the latest version of bulk_extractor is 1.4.0.
After downloading the file, un-tar it. Then, in the newly created bulk_extractor-x.y.z directory,
run the following commands to install bulk_extractor in /usr/local/bin (by default):
./configure
make
sudo make install
With these instructions, the following directory will not be installed:
• plugins/ - This is for C/C++ developers only. You can develop your own bulk_extractor
plugins which will then be run at run-time with the bulk_extractor executable. Refer to the
bulk_extractor Programmers Manual for Developing Scanner Plug-ins [3] for more
information.
Instructions on running bulk_extractor from the command line can be found in Subsection 3.2.
The Bulk Extractor Viewer tool is installed as part of the above installation process. Specific
instructions on running it can be found in Subsection 3.3.
3.1.2

Installing on Windows

Windows users should download the Windows Installer for bulk_extractor. The file to download
is located at http://digitalcorpora.org/downloads/bulk_extractor/executables/
and is called bulk_extractor-x.y.z-windowsinstaller.exe where x.y.z is the latest version number (1.4.0 - as of publication of this manual).
Next, run the bulk_extractor-x.y.z-windowsinstaller.exe file. This will automatically install bulk_extractor on your machine. Because this file is not used by many Windows
7

Figure 3: Anti-virus software, such as Symantec, often tries to block download of the installer
file
users, some anti-virus systems will try to manual delete it on download or block the download
as shown in Figure 3. Be aware that you may have to work around your anti-virus system.
Additionally, some Windows versions will try to prevent you from running it. Figure 4 shows
the message Windows 8 displays when trying to run the installer. To run anyway, click on “More
info” and then select “Run Anyway.”
When the installer file is executed, the installation will begin and show a dialog like the one
shown in Figure 5. Most users will not require the 32-bit installation and can un-check that box
if disk space is an issue. Click on “Install” and the installer will install bulk_extractor on your
system and then notify you when it is complete.
The automatic installation includes the Bulk Extractor Viewer tool as well as the complete

Figure 4: Windows 8 warning when trying to run the installer
8

Figure 5: Dialog appears when the user executes the Windows Installer
bulk_extractor system that can be run from the command line. Java 7 or above must be installed
on the machine for the Bulk Extractor Viewer to run. Instructions on running bulk_extractor
from the command line can be found in Subsection 3.2. Instructions on running it from the
Bulk Extractor Viewer are located in Subsection 3.3.

3.2 Run bulk_extractor from Command Line
The two main parameters required to run bulk_extractor are an output directory and a disk image.
The output directory must be a directory that does not already exist. The disk image can be either
a file such as a disk image or a directory of individual files. bulk_extractor cannot process a
directory of disk images.
In the following instructions, output is the name of the directory that will be created to store the
bulk_extractor output. The file mydisk.raw is the name of the disk image that will be extracted
by bulk_extractor.
To run bulk_extractor from the command line on any machine, type the following command:
bulk_extractor -o output mydisk.raw
The above command on any of the supported operating systems assumes that the disk image
mydisk.raw is located in the directory where the command is being executed. However, you can
point bulk_extractor to a disk image found elsewhere on your machine by explicitly entering the
path to that image.
The following text shows the output that is produced when bulk_extractor is run on the file
nps-2010-emails.E01. The information printed indicates the version number, input file, output directory and disk size. The screen is updated as bulk_extractor runs with status information.
bulk_extractor then prints performance information and the number of features found when the
run is complete.
C:\>bulk_extractor -o bulk_extractor\Output\nps-2010-emails bulk_extractor\In
putData\nps-2010-emails.E01

bulk_extractor version: 1.4.0-beta4
Input file: bulk_extractor\InputData\nps-2010-emails.E01
Output directory: bulk_extractor\Output\nps-2010-emails
Disk Size: 10485760
Threads: 4
All data are read; waiting for threads to finish...
Time elapsed waiting for 1 thread to finish:
(timeout in 60 min .)
Time elapsed waiting for 1 thread to finish:
6 sec (timeout in 59 min 54 sec.)
Thread 0: Processing 0
All Threads Finished!
Producer time spent waiting: 0 sec.
Average consumer time spent waiting: 8.32332 sec.
Phase 2. Shutting down scanners
Phase 3. Creating Histograms
ccn histogram...
ccn_track2 histogram...
domain histogram...
email histogram...
ether histogram...
find histogram...
ip histogram...
lightgrep histogram...
tcp histogram...
telephone histogram...
url histogram...
url microsoft-live...
url services...
url facebook-address...
url facebook-id...
url searches...Elapsed time: 11.1603 sec.
Overall performance: 0.939557 MBytes/sec
Total email features found: 67

Note that bulk_extractor has automatically selected to use 4 threads; this is because the program
was run on a computer with 4 cores. In general, bulk_extractor automatically determines the
correct number of cores to use. It is not necessary to set the number of threads to use.
After running bulk_extractor, examine the output directory specified by name in the run command. There should now be a number of generated output files in that directory. There are several
categories of output created for each bulk_extractor run. First, there are feature files grouped by
category, which contain the features found and include the path, feature and context. Second,
there are histogram files that allow users to quickly see the features grouped by the frequency
in which they occur. Certain kids of files, such as JPEGs and KML files, may be carved into
directories. Finally, bulk_extractor creates a file report.xml, in DFXML format, that captures
the provenance of the run. After bulk_extractor has been run, all of these files will be found in
the output directory specified by the user.
The text below shows the results of running the command ls -s within the output directory from
the bulk_extractor run on the disk image nps-2010-emails.E01. The numbers next to the
file names indicate the file size and show that several of the files, including email.txt and
domain.txt, were populated with features during the run.
C:\bulk_extractor\Output\nps-2010-emails>ls -s

total 303
0 aes_keys.txt
0 alerts.txt
0 ccn.txt
0 ccn_histogram.txt
0 ccn_track2.txt
0 ccn_track2_histogram.txt
64 domain.txt
1 domain_histogram.txt
0 elf.txt
16 email.txt
4 email_histogram.txt

0
0
0
0
8
0
0
0
0
0
96

kml.txt
lightgrep.txt
lightgrep_histogram.txt
rar.txt
report.xml
rfc822.txt
tcp.txt
tcp_histogram.txt
telephone.txt
telephone_histogram.txt
url.txt

0
0
1
0
0
0
0
0
0
0
8
0

ether.txt
ether_histogram.txt
exif.txt
find.txt
find_histogram.txt
gps.txt
hex.txt
ip.txt
ip_histogram.txt
jpeg
jpeg.txt
json.txt

0
0
4
0
0
1
0
12
0
0
88

url_facebook-address.txt
url_facebook-id.txt
url_histogram.txt
url_microsoft-live.txt
url_searches.txt
url_services.txt
vcard.txt
windirs.txt
winpe.txt
winprefetch.txt
zip.txt

There are numerous feature files produced by bulk_extractor for each run. A feature file is a
tab delimited file that show a feature on each row. Each row includes a path, a feature and the
context. The files are in UTF-8 format.
Any of the feature files created by bulk_extractor may have an accompanying *_stopped.txt
file found in the output directory. This file will show all stopped entries of that type that have
been found so that users can examine those files to make sure nothing critical has been hidden.
A stopped features is a feature that appears in a stop list. The stop list is a list of features that
are not of concern for a particular investigation. For example, users may input a stop list file to
bulk_extractor that contains numerous email addresses that should be ignored and not marked as
a found feature. Rather than throwing away those results when they are found, bulk_extractor
will create a file, named email_stopped.txt that shows all email addresses from the stop list
that were found during the run. The stopped email addresses will not show up in the email.txt
file. More information on creating and using stop lists can be found in Subsection 4.4.
While the above commands are all that is required for basic operation, there are numerous usage
options that allow the user to affect input and output, tuning, path processing mode, debugging,
and control of scanners. All of those options are described when bulk_extractor is run with the
-h (help) option. It is important to note that the overwhelming tendency of users is to use many
of these options, however, that is not generally recommended. Most of the time, the best way to
run bulk_extractor is with no options specified other than -o to specify the output directory. For
best performance and results in general users should avoid adding them in. Only advanced users
in specific cases should use these options.
Running bulk_extractor with only the -h option specified produces the output shown in Appendix A. To run any optional usage options, they should be inserted before the input and output
options are specified. Specifically, the order should look like the following:
bulk_extractor [Usage Options] -o output mydisk.raw
The specific order in which multiple usage options are specified matters. Some of the options are
discussed within the following sections for specific use cases, other options are for programmer
or experimental use. In general, avoid using the options unless indicated for a specific purpose.

3.3 Run bulk_extractor from Bulk Extractor Viewer
On a Linux or Mac system, go to the directory where the Bulk Extractor Viewer is installed or
specify the full path name to the jar file. It will be in the location where the bulk_extractor code
was installed and in the sub-directory labeled java_gui. From that directory, run the following
command to start the Bulk Extractor Viewer:
11

Figure 6: What Bulk Extractor Viewer looks like when it is started
./BEViewer
Windows users should go to the Start menu and choose Programs->Bulk_Extractor x.y.z->BE
Viewer with Bulk_extractor x.y.z (64-bit). If the 64-bit version can not be run on your machine,
you can choose the 32-bit version but it is not recommended for most users.
When the Bulk Extractor Viewer starts up, it will look like Figure 6. The look and feel may
vary slightly according to the specific operating system but all options should appear similar. To
run bulk_extractor from the viewer, click on the icon that looks like a gear with a down arrow.
It is next to the Print icon below the Tools menu. Clicking on this icon will bring up the “Run
bulk_extractor” Window as shown in Figure 7.
Next, in the “Run bulk_extractor” window select the Image File and Output Feature Directory to run bulk_extractor. Figure 8 shows an example where the user has selected the file
nps-2010-emails.E01 as input and is going to create a directory called nps-2010-charlieoutput in the parent directory C:\bulk_extractor\Output. Note that figures may vary slightly in
future versions of bulk_extractor but the major functionality will remain the same.
After selecting the input and output directories, click on the button at the bottom of the “Run
bulk_extractor” window labeled “Start bulk_extractor.” This will bring up the window shown in
12

Figure 7: Clicking on gear icon brings up this “Run bulk_extractor” Window

Figure 8: After selecting an Image File for input, the user must select an output directory to
create

Figure 9: Status window that shows what happens as bulk_extractor runs and indicates when
bulk_extractor is complete
Figure 9 that updates as bulk_extractor is running, providing status information during the run
and after the run is complete.
When the run is complete, a dialog will pop-up indicating the results are ready to be viewed.
Figure 10 shows this dialog. Click the “Ok” button which will return you to the main Bulk
Extractor Viewer window to view the results of the run. The “Reports” window on the left will
now show the newly created report. In this example, the report is called “nps-2010-emails-output.”
Clicking once on this report name will expand the report and show all of the files that have been
created as shown in Figure 11.
Clicking on one of the files will bring that file up in the “Feature File” window in the middle
of the screen. In the example, the user clicked on email.txt to view the email feature file.
Clicking on one of the features, in this case rtf_text@textedit.com, shows the feature in context
within the feature file on the right hand side of the window as shown in Figure 12.

Figure 10: Dialog indicating the run of bulk_extractor is complete and results are ready to be
viewed

Figure 11: Reports window shows the newly created report and all of the files created in that
report

Figure 12: While viewing the feature file, the user can select a feature to view with it”s full
context in the feature file as shown in the right hand side of the window

Figure 13: User can view histograms of features, referenced feature files and specific features in
context
The user can also view histogram files in the Bulk Extractor Viewer. Clicking on the file,
email_histogram.txt in the Reports window on the left hand side will bring up the contents
of the histogram file in the middle window. It will also display the referenced feature file in the
window below the histogram file. In this case, the referenced feature file is email.txt. Clicking
on a feature in the histogram, in this example rtf_text@textedit.com, will display the feature in
context as found within the feature file on the right hand side of the screen as shown in Figure 13.

Processing Data

4.1 Types of Input Data
The bulk_extractor system can handle multiple image formats including E01, raw, split raw and
individual disk files as well as raw devices or files. It can also operate on memory and packet
captures, although packet captures will be more completely extracted if you pre-process them
with tcpflow.
The scanners all serve different functions and look for different types of information. Often,
a feature will be stored in a format not easily accessible and will require multiple scanners to
18

extract the feature data. For example, some PDF files contain text data but the PDF format
is not directly searchable by the scanner that finds email addresses or the scanner that looks
for keywords. bulk_extractor resolves this by having the two scanners work together. The pdf
scanner will first extract all of the text from the PDF and then the other scanners will look at the
extracted text for features. This is important to remember when turning scanners off and on, as
scanners work together to retrieve the features from the disk image. The types of information
examined, extracted or carved by the existing bulk_extractor scanners are as described in Table
1, along with the scanners that process them and the specific sections where they are referenced
in this manual.

4.2 Scanners
There are multiple scanners deployed with the bulk_extractor system. For a detailed list of the
scanners installed with your version of bulk_extractor, run the following command:
bulk_extractor -H
This command will show all of the scanners installed with additional information included about
each scanner. Specifically, there is a description for each scanner, a list of the features it finds
and any relevant flags. A sample of the output is below:
Scanner Name: accts
flags: NONE
Scanner Interface version: 3
Author: Simson L. Garfinkel
Description: scans for CCNs, track 2, and phone #s
Scanner Version: 1.0
Feature Names: alerts ccn ccn_track2 telephone
Scanner Name: base16
flags: SCANNER_RECURSE
Scanner Interface version: 3
Author: Simson L. Garfinkel
Description: Base16 (hex) scanner
Scanner Version: 1.0
Feature Names: hex
...
Scanner Name: wordlist
flags: SCANNER_DISABLED
Scanner Interface version: 3
Author:
Description:
Scanner Version:
Feature Names: wordlist

This output shows that the accts scanner looks for credit card numbers, credit card track 2
information and phone numbers and finds the feature names alerts, ccn, ccn_track2 and telephone. This means it writes to the feature files alerts.txt, ccn.txt, ccn_track2.txt, and
telephone.txt.
The output also shows that the base16 scanner is a recursive scanner (indicated by the flag
SCANNER_RECURSE) meaning it expands data or finds new data for other scanners to process.
It also writes to the file hex.txt.
Finally, the output shows that the wordlist scanner is disabled by default (indicated by the flag
SCANNER_DISABLED). This means that if the user would like to use the wordlist scanner, it
19

Table 1: Input Data Processed by the Scanners
Scanner
Name
base16

Data Type

base64

Subsection 4.6 and
Subsection 5.2
Executable and Linkable Format (ELF) Subsection 5.1
EXIF structures from JPEGS (and carv- Subsection 5.5
ing of JPEG files)
GZIP files and ZLIB-compressed GZIP Subsection 4.6 and
streams
Subsection 5.2
In-memory AES keys from their key Subsection 5.2
schedules
JavaScript Object Notation files and Subsection 5.1
objects downloaded from web servers,
as well as JSON-like objects found in
source code
JPEG carving. Default is only encoded Subsection 4.3 and
JPEGs are carved. JPEGs without EX- Subsection 5.5
IFs are also carved
KML files (carved)
Subsection 5.3
RAR components in unencrypted Subsection 4.3
archives are decrypted and processed.
Encrypted RAR file are carved.
Text from PDF files (extracted for pro- Subsection 4.6
cessing not carved)
Windows FAT32 and NTFS directory Subsection 5.2
entries
Windows Hibernation File Fragments Subsection 4.6
(decompressed and processed, not
carved)
Windows Prefetch files, file fragments Subsection 5.1
(processed)
Windows Preinstallation Environment Subsection 5.1
(PE) Executables (.exe and .dll files notated with MD5 hash of first 4k)
vCard files (carved)
Subsection 5.3
XML from Garmin GPS devices (pro- Subsection 5.3
cessed)
ZIP files and zlib streams (processed, Subsection 4.3 and
and optionally carved)
Subsection 4.6

elf
exif
gzip
aes
json

jpeg
kml
rar
pdf
windirs
hiber
winprefetch
winpe
vcard
gps
zip

Base 16 (hex) encoded data (includes
MD5 codes embedded in the data)
Base 64 code

Section Discussed
in Manual
Subsection 5.2

will have to be specifically enabled. The wordlist scanner is useful for password cracking and is
discussed in Subsection 5.4.

In general, most users will not need to enable or disable scanners. The default settings installed
with the bulk_extractor system work best for the majority of users. However, individual scanners
can be enabled or disabled for different purposes. To enable the wordlist scanner, which is
disabled by default, use the following command:
bulk_extractor -e wordlist -o output diskimage.raw
Additionally, users can disable a scanner that is enabled by default. Most of the scanners are
enabled by default. To disable the accts scanner, which is very CPU intensive, run the following
command:
bulk_extractor -x accts -o output diskimage.raw
The command -E disables all scanners, then enables the one that follows the option. For example,
to disable all scanners except the aes scanner, use the following command:
bulk_extractor -E aes -o output diskimage.raw
The options -E, -e and -x are all processed in order. So, the following command will also disable
all scanners and then enable the aes scanner:
bulk_extractor -x all -e aes -o output diskimage.raw
Some of the scanners installed with bulk_extractor have parameters that can be set and utilized
by advanced users for different purposes. Those parameters are also described in the -H output
described above (as well as the -h output) and include the following:
Settable Options (and their defaults):
-S work_start_work_end=YES
Record work start and end of each scanner in
report.xml file ()
-S enable_histograms=YES
Disable generation of histograms ()
-S debug_histogram_malloc_fail_frequency=0
Set >0 to make histogram maker
fail with memory allocations ()
-S hash_alg=md5
Specifies hash algorithm to be used for all hash
calculations ()
-S word_min=6
Minimum word size (wordlist)
-S word_max=14
Maximum word size (wordlist)
-S max_word_outfile_size=100000000
Maximum size of the words output file
(wordlist)
-S exif_debug=0
debug exif decoder (exif)
-S jpeg_carve_mode=1
0=carve none; 1=carve encoded; 2=carve all (exif)
-S min_jpeg_size=1000
Smallest JPEG stream that will be carved (exif)
-S zip_min_uncompr_size=6
Minimum size of a ZIP uncompressed object (zip)
-S zip_max_uncompr_size=268435456
Maximum size of a ZIP uncompressed
object (zip)
-S zip_name_len_max=1024
Maximum name of a ZIP component filename (zip)
-S rar_find_components=YES
Search for RAR components (rar)
-S raw_find_volumes=YES
Search for RAR volumes (rar)
-S gzip_max_uncompr_size=268435456
maximum size for decompressing GZIP
objects (gzip)
-S pdf_dump=NO
Dump the contents of PDF buffers (pdf)
-S opt_weird_file_size=157286400
Weird file size (windirs)
-S opt_weird_file_size2=536870912
Weird file size2 (windirs)
-S opt_max_cluster=67108864
Ignore clusters larger than this (windirs)
-S opt_max_cluster2=268435456
Ignore clusters larger than this (windirs)
-S opt_max_bits_in_attrib=3
Ignore FAT32 entries with more attributes set
than this (windirs)
-S opt_max_weird_count=2
Ignore FAT32 entries with more things weird than
this (windirs)
-S opt_last_year=2020
Ignore FAT32 entries with a later year than this (windirs)
-S bulk_block_size=512
Block size (in bytes) for bulk data analysis (bulk)
-S DFRWS2012=NO
True if running DFRWS2012 challenge code (bulk)
-S xor_mask=255
XOR mask string, in decimal (xor)

To use any of these options, the user should specify the -S with the name=value pair when
running bulk_extractor as in the following example:
bulk_extractor -S name=value -o output diskimage.raw
As with the other scanner and bulk_extractor usage options, most users will not have to use any
of these options.

4.3 Carving
File carving is a special kind of carving in which files are recovered. File carving is useful for
both data recovery and forensic investigations because it can recover files when sectors containing
file system metadata are either overwritten or damaged [4]. Currently, bulk_extractor provides
carving of contiguous JPEG, ZIP and RAR files. To carve fragmented files we recommend
PhotoRec (free) or Adroit Photo Recovery( commercial). Additionally, Forensics Toolkit and
EnCase Forensic provide some carving capability on fragmented files.
Carved results are stored in two different places. First, a file listing all the files that are carved
are written to a corresponding .txt file: JPEG files to jpeg.txt, ZIP files to unzip.txt and
RAR files to unrar.txt. Second, the carved JPEG, ZIP and RAR files are placed in binned
directories that are named /jpeg, /unzip and /unrar respectively. For example, all carved JPEGs
will go in the directory /jpeg. The output files are further binned with 1000 files in each directory.
The directory names are 3 decimal digits. If there are more than 999,000 carved files of one
type, then the next set of directories are named with 4 digits. File names for JPEGs are the
forensicpath.jpg. File names for the ZIP carver are the forensicpath_filename. If the ZIP file
name has slashes in it (denoting directories), they are turned into ’_’ (underbars). For example,
the file mydocs/output/specialfile will be named mydocs_output_specialfile.
Table 2: There are three carving modes in bulk_extractor that are specified separately for each
file type, JPEG, ZIP and RAR.
Mode
0
1
2

Mode Description
Do not carve files of the specified type.
Only carve encoded files of the specified type
Carve everything of the specified type.

As the above table describes, there are three carving modes in bulk_extractor that can be specified separately for each file type, JPEG, ZIP or RAR. The first mode, mode 0, explicitly tells
bulk_extractor not to carve files of that type. The second mode, mode 1, is on by default and
tells bulk_extractor to carve only encoded files of that type. If the user is running the ZIP carver
in mode 1 and there is a simple ZIP file, it will not get carved. However, if there is an encoded
attachment of that file (like Base64) it will get carved. The final mode, mode 2, will carve
everything of that type. There is no way to specify which types of files (particular extensions)
will get carved and which will not in mode 2. For example, bulk_extractor will carve both JPEGs
and doc files. It carves whatever is encountered.
To specify the carving modes for bulk_extractor, command line arguments can be specified. To
modify the JPEG carving modes, type the following where carve mode 1=default value that does
not need to be specified (carve encoded), 0=no carving or 2=carve everything:
bulk_extractor -S jpeg_carve_mode=1 -o output diskimage.raw
22

To modify the ZIP carving modes, type the following where carve mode 1=default value that
does not need to be specified (carve encoded), 0=no carving or 2=carve everything:
bulk_extractor -S zip_carve_mode=1 -o output diskimage.raw
To modify the RAR carving modes, type the following where carve mode 1=default value that
does not need to be specified (carve encoded), 0=no carving or 2=carve everything:
bulk_extractor -S rar_carve_mode=1 -o output diskimage.raw
Any combination of the carving mode options can be specified for a given run. The carvers can
run in any combination of modes. For example, the JPEG carver can be run in mode 2 while the
RAR carving is turned off in mode 1 and the ZIP carver carves only encoded files in mode 1.
Because bulk_extractor can carve files and preserve original file extensions, there is a real possibility that bulk_extractor might be carving out malware. There is no protection in bulk_extractor
against putting malware in a file on your hard drive. Users running bulk_extractor to look for
malware should turn off all anti-virus software because the anti-virus program will think its
creating malware and stop it. Then the user should should carefully scan the results looking for
malware before re-enabling the anti-virus.

4.4 Suppressing False Positives
Modern operating systems are filled with email addresses. They come from Windows binaries,
SSL certificates and sample documents. Most of these email addresses, particularly those that
occur the most frequently, such as someone@example.com, are not relevant to the case. It is
important to be able to suppress those email addresses not relevant to the case. To address this
problem, bulk_extractor provides two approaches.
First, bulk_extractor allows users to build a stop list or use an existing one available for download.
These stop lists are used to recognize and dismiss the email addresses that are native to the
Operating System. This approach works well for email addresses that are clearly invalid, such
as someone@example.com. For most email addresses, however, you will want to stop them in
some circumstances but not others. For example, there are over 20,000 Linux developers, you
want to stop their email addresses in program binaries, not in email messages.
To address this problem, bulk_extractor uses context-sensitive stop lists. Instead of a stop list of
features, this approach uses the feature+context. The following example is an excerpt from a
context-sensitive stop list file.
ubuntu-users@lists.ubuntu.com
Maint\x0A935261357\x09ubuntu-users@lists.ubuntu.com\x0
ubuntu-motu@lists.ubuntu.com
untu_\x0A923867047\x09ubuntu-motu@lists.ubuntu.com\x09
pschiffe@redhat.com
Peter Schiffer - 0.8-1.1N\x94/\xC0phpdevel@echospace.com : Vlad Krupin \x0AMAINTENANCE:
anholt@freebsd.org
34-GZIP-1021192\x09anholt@freebsd.org\x09r: EricAnholt
ubuntu-motu@lists.ubuntu.com
http\x0A938966489\x09ubuntu-motu@lists.ubuntu.com\x09

The context for the feature is the 8 characters on either side of the feature. Each “stop list” entry
is the feature+context. This ignores Linux developer email addresses in Linux binaries. The
email address will be ignored if found in that context but reported if it appears in a different
context.

Figure 14: Email Histogram Results With and Without the Context-Sensitive Stop List. Results
from the Domexusers HD image.
There is a context-sensitive stop list for Microsoft Windows XP, 2000, 2003, Vista and several
Linux systems. The total stop list is 70 MB and includes 628,792 features in a 9 MB zip file.
The context-sensitive stop list prunes many of the OS-supplied features. By applying it to the
domexusers HD image (the image can be downloaded at http://http://digitalcorpora.
org/corp/nps/drives/nps-2009-domexusers/, the number of emails found went from
9,143 down to 4,459. This significantly reduces the amount of work to be done by the investigator.
Figure 14 shows how the histogram of email addresses differs when bulk_extractor is run with
and without the context-sensitive stop list. The context-sensitive stop list built for the various
operating systems described above can be downloaded from http://digitalcorpora.org/
downloads/bulk_extractor. The file will have the words “stoplist” in it somewhere. The
current version as of publication of this manual is called bulk_extractor-3-stoplist.zip.
It should be noted that bulk_extractor does allow the users to create stop lists that are not context
sensitive. A stop list can simply be a list of words that the user wants bulk_extractor to ignore.
For example, the following three lines would constitute a valid stop list file:
abc@google.com
ignore@microsoft.com
www.google.com

However for the reasons stated above, it is recommended that users rely on context-sensitive
stop lists when available to reduce the time required to analyze the results of a bulk_extractor run.
Stopped results are not completely hidden from users. If stopped feature are discovered, they
will be written to the appropriate category feature file with the extension _stopped.txt.
For example, stopped domain names that are found in the disk image will be written to
domain_stopped.txt in the output directory. The stopped files serve the purpose of allowing users to verify that bulk_extractor is functioning properly and that the lists they have
written are being processed correctly.
24

4.5 Using an Alert List
Users may have specific words, or feature in a given context, that are important to their investigation. The alert list allows bulk_extractor to specifically alert or flag the user when those concepts
are found. Alert lists can contain a list of words or a feature file. The feature file operates much
in the same way as the feature files used for context-sensitive stop lists. It will provide a feature
but alert on that feature only when it’s found in the specified context.
A sample alert list file might look like the following:
abc@google.com
SilentFury2012
www.maliciousintent.com

While this list does not appear to help in any particular investigation, it demonstrates that you
can specify distinct words that are important to their analysis. Results containing the alert list
information are found in the file alert.txt in the run output directory.

4.6 The Importance of Compressed Data Processing
Many forensic tools frequently miss case-critical data because they do not examine certain
classes of compressed data found. For example, a recent study of 1400 drives found thousands
of email addresses that were compressed but in unallocated space[5]. Without looking at all
the data on each drive and optimistically decompressing it, those features would be missed.
Compressed email addresses, such as those in a GZIP file, do not look like email addresses to a
scanner; they must first be decompressed to be identified. Although some of these features are
from software distributions, many are not. Table 3 shows the kinds of encodings that can be
decoded by bulk_extractor [5].
Table 3: The kinds of encodings that can be decoded by bulk_extractor and the amount of
context required for the decoding
Encoding
GZIP
BASE64
HIBER

PDF
ZIP

Can be decoded when bulk_extractor finds
The beginning of a zlib-compressed stream
The beginning of a BASE64-encoded stream
Any fragment of a hibernation file can generally be decompressed, as each Windows 4k page is separately compressed
and the beginning of each compressed page in the hibernation file is indicated by a well-known sequence
Any PDF stream compressed with ZLIB bracketed by
stream and endstream
The local file header of a ZIP-file component

The reason that users must be aware of this is because users have a tendency to want to enable and
disable scanners for specific uses; They can unintentionally damage their results. For example,
if a user only wants to find email addresses, they may try to turn off all scanners except the
email scanner. This will find some email addresses. However, it will miss the email addresses on
the media that are only present in compressed data. This is because scanners such as zip, rar
and gzip will not be running. Those scanners each work on a different type of compressed data.
For example, the gzip scanner will find GZIP compressed data, decompress it and then pass it
other scanners to search for features. In that way, GZIP compressed emails can be processed by
25

bulk_extractor.
The pdf scanner is another type of scanner that finds text that otherwise wouldn’t be found.
While PDF files are human readable, they are not readable but many software tools and scanners because of their formatting. The pdf scanner extracts some kinds of text found within
PDFs and then passes that text on to other scanners for further processing. Many typical disk
images include PDF files, so most users will want to have this scanner enabled (as it is by default).
Finally, the hiber scanner decompresses Windows hibernation files. If the disk image being
analyzed is from a Windows system, bulk_extractor users will want that turned on (as it is by
default). The scanner is very fast, however, so it will not significantly decrease performance on
non-Windows drives.

5 Use Cases for bulk_extractor
There are many digital forensic use cases for bulk_extractor— more than we can enumerate
within this manual. In this section we highlight some of the most common uses of the system.
Each case discusses which output files, including feature files and histograms, are most relevant
to these types of investigations. In Section 8, Worked Examples, we provide more detailed
walk-throughs and refer back to these use cases with more detailed output file information.

5.1

Malware Investigations

Malware is a programmatic intrusion. When performing a malware investigation, users will want
to look at executables, information that has been downloaded from web-based applications and
windows directory entries (for Windows-specific investigations). bulk_extractor enables this in
several ways.
First, bulk_extractor finds evidence of virtually all executables on the hard drive including those
by themselves, those contained in ZIP files, and those that are compressed. It does not give you
the hash value of the full file, rather, it gives the hash of just the first 4KB of the file. Our research
has shown that the first 4KB is predictive because most executables have a distinct hash value
for the first 4KB of the file [6]. Additionally, many of these files may be fragmented and looking
at the first 4KB will still provide information relevant to an investigation because fragmentation
is unlikely to happen before the first 4KB. The full hash of a fragmented file is not available in
bulk_extractor.
Several output feature files produced by bulk_extractor contain relevant and important information about executables. These files include:
• elf.txt — This file (produced by the elf scanner) contains information about ELF
executables that can be used to target Linux and Mac systems.
• winprefetch.txt — This file (produced by the winprefetch scanner lists the current
and deleted files found in the Windows prefetch directory.
The XML in these feature files is too complicated to review without using other applications.
The recommended way to analyze the executable output is to use a third party tool that analyzes
executables or pull the results into a spreadsheet. In a spreadsheet, one column could contain
the hash values and those values can be compared against a database of executable hashes.
26

There is also a python tool that comes with bulk_extractor called identify_filenames.py that can
be used to get the full filename of the file. The python tool is discussed in more detail in Section 7.
For Windows specific malware investigations, the files winpe.txt and winprefetch.txt are
very useful. They are produced by the winpe and winprefetch scanners respectively. Windows
Prefetch shows files that have been prefetched in the Windows prefetch directory and shows the
deleted files that were found in unallocated space. The Windows PE feature file shows entries
related to the Windows executable files.
JSON, the JavaScript Object Notation, is a lightweight data-interchange format. Websites tend
to download a lot of information using JSON. The output file json.txt, produced by the
json scanner, can be useful for malware investigations and analysis of web-based applications.
If a website has downloaded information in JSON format, the JSON scanner may find that
information in the browser cache.

5.2 Cyber Investigations
Cyber investigations may scan a wide variety of information types. A few unifying features
of these investigations are the need to find encryption keys, hash values and information about
ethernet packets. bulk_extractor provides several scanners that produce feature files containing
this information.
For encryption information, the following feature files may be useful:
• aes.txt — AES is an encryption system. Many implementations leave keys in memory
that can be found using an algorithm invented at Princeton University. bulk_extractor
provides an improved version of that algorithm to find AES keys in the aes scanner. When
it scans memory, such as swap files or decompressed hibernation files, it will identify the
AES keys. The keys can be used for software that will decrypt AES encrypted material.
• hex.txt — The base16 scanner decodes information that is stored in Base16, breaking
it into the corresponding hexidecimal values. This is useful if you are looking for AES
keys or SHA1 hashes. This scanner only writes blocks that are of size 128 and 256
because they are the sizes used for encryption keys. The feature file is helpful if the
investigator is looking for people who have emailed encryption keys or hash values in a
cyber investigation.
Additionally, the base64 scanner is important for cyber investigations because it looks mostly at
email attachments that are coded in Base64. The information found in these attachments will be
analyzed by other scanners looking for specific features.
The windirs scanner finds Windows FAT32 and NTFS directory entries and will also be useful
for cyber investigations involving Windows machines, as they may be indicators of times that
activity took place.
Finally, the files ether.txt, ip.txt, tcp.txt and domain.txt are all produced by the net
scanner. It searches for ethernet packets and memory structures associated with network data
structures in memory. It is important to note that tcp connections have a lot of false positives and
many of the information found by this scanner will be false. Investigators should be careful with
the interpretation of these feature files for that reason.

5.3 Identity Investigations
Identity investigations may be looking for a wide variety of information including email addresses, credit card information, telephone numbers, geographical information and keywords.
For example, if the investigator is trying to find out of who a person is and who their associates
are, they will be looking at phone numbers, search terms to see what they are doing and emails
to see who they are communicating with.
The accts scanner is very useful for identity investigations. It produces several feature files with
identity information including:
• ccn.txt — credit card numbers
• ccn_track2.txt - credit card track two information - relevant information if someone
is trying to make physical fake credit cards
• pii.txt - personally identifiable information including birth dates and social numbers
• telephone.txt - telephone numbers
The kml and gps scanner both produce GPS information that give information about a person
in a certain area or link to what they have been doing in a certain area. Both of these scanners
write to gps.txt. KML is a format used by Google Earth and Google Map files. This scanner
searches in that formatted data for GPS coordinates. The gps scanner looks at Garmin Trackpoint
formatted information and finds GPS coordinates in that data.
The email scanner looks for email addresses in all data and writes that to email.txt. The
vcard scanner looks at vCard data, an electronic business card format, and finds names, email
addresses and phone numbers to write to the respective feature file.
The are multiple url files including url.txt, url_facebook-address, url_facebook-id,
url_microsoft-live, url_searches.txt and url_services.txt that are produced by
the email scanner. They are useful for looking at what websites a person has visited as well as
the people they are associating with.
An important aspect of identity investigations (as well as other types) is the ability to search
the data for a list of keywords. bulk_extractor provides the capability to do that through two
different means. First, the find scanner is a simple regular expression finder that uses regular
expressions. The find scanner looks through the data for anything listed in the global find list.
The format of the find list should be rows of regular expressions while any line beginning with a
# is considered a comment. The following is an excerpt from a sample find list file:
# This is a comment line
\b\d{1,3}\.\d{1,3}\.\d{1,3}\.\d{1,3}\b
# another comment line
/^[a-z0-9_-]{3,16}$/

The first regular expression from the above example, beginning with \b, looks for the following
in order: a word boundary followed a digit repeated between 1-3 times, a digit repeated between
1-3 times, a digit repeated 1-3 times, a ’.’, a digit repeated 1-3 times, a digit repeated 1-3 times
and the end of the word boundary. That regular expression would find, for example, the sequence
2219.889 separated out from other text by a word boundary.

The second regular expression from the above example, beginning with / looks for the following
in order: a ’/’, the beginning of a line, repeats of any character in lowercase a-z, 0-9, ’_’, or ’-’,
repeated 3 to 16 times, and the end of the line followed by ’\.’ That expression would find, for
example, the following sequence:
\
284284284284
/
Regular expressions can be used to represent character and number sequences (or ranges of
values) that might be of particular importance to an investigation.
The find list is sent in as input to bulk_extractor using the “-F findlist” option. To run
bulk_extractor with a find list, the following basic parameters are required (where findlist.txt
is the name of the find list):
bulk_extractor -F findlist.txt -o output mydisk.raw
Another scanner, the lightgrep scanner provides the same functionality as the find scanner but it
is much faster and provides more functionality. It is also a regular expression scanner that looks
through the buffers and matches in the global find list. A syntax sheet of regular expressions that
might be helpful to users in creating a find list to be used by the Lightgrep Scanner is shown in
Figure 15.
The lightgrep scanner uses the Lightgrep library from Lightbox Technologies. An open source
version of that library can be downloaded from https://github.com/LightboxTech/
liblightgrep. Installation instructions are also available at the download site. The lightgrep scanner is preferable because it looks for all regular expressions at once, on the first pass
through the data. The find scanner actually looks for each expression in the find list one at a
time. For example, if the find list is a list of medical terms and diagnoses and bulk_extractor
is searching medical records, the find scanner looks for each term in each piece of data on one
pass through, one at a time. A list of 35 expressions would require 35 passes through the data.
The lightgrep scanner will search a given buffer for all of the medical terms at once, in one pass
through.
If the Lightgrep library is installed and the find list is provided to bulk_extractor, it will run the
lightgrep scanner. If not, it will use the find scanner. Neither scanner needs to be enabled by the
user specifically, calling bulk_extractor with the find list will automatically enable the appropriate
scanner. However, we do not recommend using the find list without the Lightgrep library — it
will make bulk_extractor run very slowly because each find search will be sequentially executed.
This will provide an exponential slow-down.
Investigators looking for identity information may rely heavily on the find list to search for
specific names, numbers or keywords relevant to the investigation. The features found by the find
or lightgrep scanner will be written to the files find.txt and lightgrep.txt respectively.

5.4

Password Cracking

If an investigator is looking to crack a password, the wordlist scanner can be useful. It generates
a list of all the words found on the disk that are between 6 and 14 characters. Users can change
the minimum and maximum size of words by specifying options at run-time but we have found
this size range to be optimal for most applications. Because the wordlist scanner is disabled by
29

⇤ except

U+0000 (NUL) and metacharacters
extension; not part of PCRE.
any of: adefnprstwDPSW1234567890

† Lightgrep
‡ except

2 Named Character Classes

.
\d
\D
\s
\S
\w
\W
\p{property}
\P{property}

any character
[0-9] (= ASCII digits)
[^0-9]
[\t\n\f\r ] (= ASCII whitespace)
[^\t\n\f\r ]
[0-9A-Za-z_] (= ASCII words)
[^0-9A-Za-z_]
any character having property
any character lacking property

3 Character Classes

[stu ]
any character in stu
[^stu ]
any character not in stu
where stu is. . .
c
a character
a-b
a character range, inclusive
\zhh
a byte
\zhh-\zhh
a byte range, inclusive
[S]
a character class
ST
S [ T (union)
S&&T
S \ T (intersection)
S--T
S T (di erence)
S~~T
S 4 T (symmetric di erence, XOR)
8 EnCase GREP Syntax

c
\xhh
\whhhh
\c
.
#
[a-b]
[S]
[^S]
(S)
S*
S+
S?
S{n,m}
ST
S|T

the character c (except metacharacters)
U+00hh, 2 hexadecimal digits h
U+hhhh, 4 hexadecimal digits h
the character c
any character
[0-9] (= ASCII digits)
any character in the range a–b
any character in S
any character not in S
grouping
repeat S 0 or more times (max 255)
repeat S 1 or more times (max 255)
repeat S 0 or 1 or time
repeat S n–m times (max 255)
matches S, then matches T
matches S or T

9 Importing from EnCase into Lightgrep

\whhhh
#
S*
S+

! \xhhhh
! \d
! S{0,255}
! S{1,255}

S* and S+ are limited to
255 repetitions by EnCase;
Lightgrep preserves this in
imported patterns.

\w is limited to BMP characters ( U+10000) only.

Some people, when confronted with a problem, think “I know,
I’ll use regular expressions.” Now they have two problems.
—JWZ in alt.religion.emacs, 12 August 1997

Lightgrep Cheat Sheet
Notes & Examples

Characters:
.*?\x00
(= null-terminated string)
\z50\z4B\z03\z04
(= ZIP signature)
\N{EURO SIGN}, \N{NO-BREAK SPACE}
\x{042F} (= CYRILLIC CAPITAL LETTER YA)
\+12\.5% (= escaping metacharacters)

Grouping: Operators bind tightly. Use (aa)+,
not aa+, to match pairs of a’s.
Ordered alternation: a|ab matches a twice in
aab. Left alternatives preferred to right.

Repetition: Greedy operators match as much
as possible. Reluctant operators match as little
as possible. a+a matches all of aaaa; a+?a
matches the first aa, then the second aa.

.+ will (uselessly) match the entire input.
Prefer reluctant operators when possible.

Character classes:
[abc] = a, b, or c
[^a] = anything but a
[A-Z] = A to Z
[A\-Z]
= A, Z, or hyphen (!)
[A-Zaeiou] = capitals
or lowercase vowels
[.+*?\]]
= ., +, *, ?, or ]
[Q\z00-\z7F]
= Q or 7-bit bytes
[[abcd][bce]]
= a, b, c, d, or e
[[abcd]&&[bce]]
= b or c
[[abcd]--[bce]]
= a or d
[[abcd]~~[bce]]
= a, d, or e
[\p{Greek}\d]
= Greek or digits
[^\p{Greek}7]
= neither Greek nor 7
[\p{Greek}&&\p{Ll}]
= lowercase Greek

4 Grouping

(S)

makes any pattern S atomic

ST
S|T

matches S, then matches T
matches S or T , preferring S

5 Concatenation & Alternation
6 Repetition

S*
S+
S?
S{n,}
S{n,m}
S*?
S+?
S??
S{n,}?
S{n,m}?

Greedy

the character c⇤
U+0007 (BEL) bell
U+001B (ESC) escape
U+000C (FF) form feed
U+000A (NL) newline
U+000D (CR) carriage return
U+0009 (TAB) horizontal tab
U+ooo, 1–3 octal digits o,  0377
U+00hh, 2 hexadecimal digits h
U+hhhhhh, 1–6 hex digits h
the byte 0xhh (not the character!)†
the character called name
same as \x{hhhhhh}
the character c‡

Reluctant

1 Single Characters
c
\a
\e
\f
\n
\r
\t
\ooo
\xhh
\x{hhhhhh}
\zhh
\N{name}
\N{U+hhhhhh}
\c

7 Selected Unicode Properties

Repeats S. . .
0 or more times (= S{0,})
1 or more times (= S{1,})
0 or 1 time (= S{0,1})
n or more times
n–m times, inclusive
0 or more times (= S{0,})
1 or more times (= S{1,})
0 or 1 time (= S{0,1})
n or more times
n–m times, inclusive

Any
Assigned
Alphabetic
White_Space
Uppercase
Lowercase
ASCII
Noncharacter_Code_Point
Name=name
Default_Ignorable_Code_Point
General_Category=category
L, Letter
P, Punctuation
Lu, Uppercase Letter
Pc, Connector Punctuation
Ll, Lowercase Letter
Pd, Dash Punctuation
Lt, Titlecase Letter
Ps, Open Punctuation
Lm, Modifier Letter
Pe, Close Punctuation
Lo, Other Letter
Pi, Initial Punctuation
M, Mark
Pf, Final Punctuation
Mn, Non-Spacing Mark
Po, Other Punctuation
Me, Enclosing Mark
Z, Separator
N, Number
Zs, Space Separator
Nd, Decimal Digit Number Zl, Line Separator
Nl, Letter Number
Zp, Paragraph Separator
No, Other Number
C, Other
S, Symbol
Cc, Control
Sm, Math Symbol
Cf, Format
Sc, Currency Symbol
Cs, Surrogate
Sk, Modifier Symbol
Co, Private Use
So, Other Symbol
Cn, Not Assigned
Script=script
Operators need not be
Common Latin Greek Cyrillic Armenian Hebrew Araescaped inside charbic Syraic Thaana Devanagari Bengali Gurmukhi Guacter classes.
jarati Oriya Tamil Telugu Kannada Malayalam Sinhala Thai Lao Tibetan Myanmar Georgian Hangul
Ethiopic Cherokee Ogham Runic Khmer Mongolian
Lightgrep Search
Hiragana Katakana Bopomofo Han Yi Old_Italic
for EnCase R
Gothic Inherited Tagalog Hanunoo Buhid Tagbanwa
Limbu Tai_Le Linear_B Ugaritic Shavian Osmanya
Fast Search for
Cypriot Buginese Coptic New_Tai_Lue Glagolitic
Forensics
Tifinagh Syloti_Nagri Old_Persian Kharoshthi Balinese Cuneiform Phoenician Phags_Pa Nko Sudanese
www.lightgrep.com Lepcha . . .
See Unicode Standard for more.
Email addresses: [a-z\d!#$%&’*+/=?^_‘{|}~-][a-z\d!#$%&’*+/=?^_‘{|}~.-]{0,63}
@[a-z\d.-]{1,253}\.[a-z\d-]{2,22}
Hostnames: ([a-z\d]([a-z\d_-]{0,61}[a-z\d])?\.){2,5}[a-z\d][a-z\d-]{1,22}
N. American phone numbers: \(?\d{3}[ ).-]{0,2}\d{3}[ .-]?\d{4}\D
Visa, MasterCard: \d{4}([ -]?\d{4}){3}
American Express: 3[47]\d{2}[ -]?\d{6}[ -]?\d{5}
Diners Club: 3[08]\d{2}[ -]?\d{6}[ -]?\d{4}
EMF header: \z01\z00\z00\z00.{36}\z20EMF
JPEG: \zFF\zD8\zFF[\zC4\zDB\zE0-\zEF\zFE] Footer: \zFF\zD9
GIF: GIF8[79] Footer: \z00\z3B
BMP: BM.{4}\z00\z00\z00\z00.{4}\z28
PNG: \z89\z50\z4E\z47 Footer: \z49\z45\z4E\z44\zAE\z42\z60\z82
ZIP: \z50\z4B\z03\z04 Footer: \z50\z4B\z05\z06
RAR: \z52\z61\z72\z21\z1a\z07\z00...[\z00-\z7F]
Footer: \z88\zC4\z3D\z7B\z00\z40\z07\z00
GZIP: \z1F\z8B\z08
MS O ce 97–03: \zD0\zCF\z11\zE0\zA1\zB1\z1A\zE1
LNK: \z4c\z00\z00\z00\z01\z14\z02\z00
PDF: \z25\z50\z44\z46\z2D\z31 Footer: \z25\z45\z4F\z46

Figure 15: Guide to Syntax Used by Lightgrep Scanner

default, users must specifically enable it at run-time when needed. To do that, run the following
command:
bulk_extractor -e wordlist -o output mydisk.raw
This will produce two files useful for password cracking, wordlist_histogram.txt and
wordlist.txt. These files will contain large words that can be used to recommend passwords.

5.5 Analyzing Imagery Information
In an investigator needs to specifically analyze imagery, for something such as a child pornography investigation, the exif scanner would be useful. It finds JPEGs on the disk image and then
carves the encoded ones that might be in, for example, ZIP files or hibernation files. It writes the
output of this carving to jpeg.txt.

5.6 Using bulk_extractor in a Highly Specialized Environment
If using bulk_extractor in a specialized environment, two specific features might be useful.
The first is the option to include a banner on each output file created by bulk_extractor. The
banner file, specified in the example command below as banner.txt could include a security
classification of the output data. When bulk_extractor is run with the command specified below,
the data in the banner file will be printed at the top of each output file produced.
bulk_extractor -b banner.txt -o output mydisk.raw
The second feature might be useful to users in a specialized environment is the ability to develop
plug-ins. Plug-ins in bulk_extractor are external scanners that an individual or organization
can run in addition to the open source capabilities provided with the bulk_extractor system.
The plug-in system gives the full power of bulk_extractor to external developers, as all of
bulk_extractor’s native scanners are written with the plug-in system. This power gives third party
developers the ability to utilize proprietary or security protected algorithms and information
in bulk_extractor scanners. It is worth noting that all scanners installed with bulk_extractor
use the plug-in system, bulk_extractor is really just a framework for running plug-ins. The
separate publication Programmers Manual for Developing Scanner Plug-ins [3] provides
specific details on how to develop and use plug-ins with bulk_extractor.

6 Tuning bulk_extractor
All data that bulk_extractor processes is divided into buffers called sbufs. Buffers created from
disk images are created with a pre-determined size (bufsize). The buffer includes a page and an
overlap area. As shown in Figure 16, the pages overlap with each other in the red region. The
red overlap region is called the margin. bulk_extractor scans the pages one-by-one looking for
features. Pages overlap with each other so that bulk_extractor won’t miss any features that cross
from one page into another across boundaries.
Users may be looking for potentially large features that are bigger than the buffer size or that
overlap into the margin. In that case, they may want to adjust the margin size or buffer size. For
example, if the input data includes a 30 MB ZIP file (possibly a software program), bulk_extractor
won’t find features in the program because it overlaps the margins. To find features of that size,
the margin size must be increased.

Disk Image

pagesize
bufsize

Figure 16: Image Processor divides the disk image into buffers. Each buffer is the size of
a page (pagesize) with a buffer overlap in an area called the margin. (marginsize is equal to
bufsize-pagesize). The buffers overlap with each other to ensure all information is processed.
To adjust the page size, the following usage options need to be included where NN should be set
to the size (default page size is 16777216):
bulk_extractor -G NN -o output mydisk.raw
To adjust the margin size, the following usage options need to be included where NN should be
set to the size (default margin size is 4194304):
bulk_extractor -g NN -o output mydisk.raw
bulk_extractor provides many other tuning capabilities that are primarily recommended for users
doing advanced research. Many of those options relate to specifying file sizes for input or output,
specifying block sizes, dumping the contents of a buffer or ignoring certain entries. Those
options are all found in the output of the -h input to bulk_extractor and listed in Appendix A.

Post Processing Capabilities

There are two Python programs useful for post-processing the bulk_extractor output. Those
programs are bulk_diff.py and identify_filenames.py. To run either of these programs, you
must have Python version 2.7 or higher installed on your system. On Linux and Mac systems, the bulk_extractor python programs are located in the directory ./python under the main
bulk_extractor installation.

7.1 bulk_diff.py: Difference Between Runs
The program bulk_diff.py takes the results of two bulk_extractor runs and shows the differences
between the two runs. This program essentially tells the difference between two disk images.
It will note the different features that are found by bulk_extractor between one image and the
next. It can be used, for example, to easily tell whether or not a computer user has been visiting
websites they are not supposed to by comparing a disk image from their computer from one
week to the next. To run the program, users should type the following, where pre and post are
both locations of two bulk_extractor output directories:
bulk_diff.py
Note, Linux and Mac users may have to type python2.7, python3, or python3.3 before the
command, indicating the version of Python installed on your machine. An example use of the
bulk_diff.py program can be found in Section 8.
32

7.2 identify_filenames.py: Identify File Origin of Features
The program identify_filenames.py operates on the results of bulk_extractor run and identifies
the filenames (where possible) of the features that were found on the disk image. The user can
run this program on one or all of the features file produced by a given run. It can be used, for
example, to find the full content of an email when references to its contents are found in one of
the feature files. Often email features are relevant to an investigation and an investigator would
like to be able to view the full email.
To run this program, users will need the program fiwalk installed on their machine or have
a DFXML file generated by fiwalk that corresponds to the disk image. fiwalk is part of the
SleuthKit and can be installed by installing Sleuthkit, available at http://www.sleuthkit.
org/.
The identify_filenames.py program provides various usage options but to run the program on
all feature files produced by a bulk_extractor run, the user should type the following (where
“bulkoutputdirectory” is the directory containing the output of a bulk_extractor run and “idoutput”
will contain the annotated feature files after the program runs):
identify_filenames.py --all bulkoutputdirectory idoutput
Note, Linux and Mac users may have to type python2.7, python3, or python3.3 before the
command, indicating the version of Python installed on your machine. An example use of the
bulk_diff.py program can be found in Section 8.

8 Worked Examples
The worked examples provided are intended to further illustrate how to use bulk_extractor to
answer specific questions and conduct investigatons. Each example uses a different, publicly
available dataset and can be replicated by readers of this manual.

8.1 Encoding
We describe the encoding system here in order to prepare users to view the feature files produced
by bulk_extractor. Unicode is the international standard used by all modern computer systems
to define a mapping between information stored inside a computer and the letters, digits, and
symbols that are displayed on the screens or printed on paper. UTF-8 is a variable width encoding
that can represent every character in the Unicode character set. It was designed for backward
compatibility with ASCII and to avoid the complications of endianness and byte order marks in
UTF-16 and UTF-32. Feature files in bulk_extractor are all coded in UTF-8 format. This means
that the odd looking symbols, such as accented characters (è ), funny symbols ( · · · ) and the
occasional Chinese character ( 路) that may show up in the files are legitimate. Glyphs from
language, for example ,Cyrillic (X) or Arabic ( è) may show up in features files as all foreign
languages can be coded in UTF-8 format. It is perfectly appropriate and typical to open up a

feature file and see characters that the user may not recognize.

9 2009-M57 Patents Scenario
The 2009-M57-Patents scenario tracks the first four weeks of corporate history of the (fictional)
M57 Patents company. The company started operation on Friday, November 13th, 2009, and
ceased operation on Saturday, December 12, 2009. This specific scenario was built to be used as
a teaching tool both as a disk forensics exercise and as a network forensics exercise. The scenario
data is also useful for computer forensics research because the hard drive of each computer
and each computers memory were imaged every day. In this example, we are not particularly
interested in the exercises related to illegal activity, exfiltration and eavesdropping; they do
however provide interesting components for us to examine in the example data[2].

9.1 Run bulk_extractor with the Data
For this example, we downloaded and utilized one of the disk images from the 2009-M57Patents Scenario. Those images are available at http://digitalcorpora.org/corp/nps/
scenarios/2009-m57-patents/drives-redacted/. The file used throughout this example is called charlie-2009-12-11.E01. Running bulk_extractor on the command line
produces the following output (text input by the user is bold):
C:\bulk_extractor>bulk_extractor -o ../Output/charlie-2009-12-11 charlie-2009-12-11.E01

bulk_extractor version: 1.4.0-beta4
Input file: charlie-2009-12-11.E01
Output directory: ../Output/charlie-2009-12-11
Disk Size: 10239860736
Threads: 4
8:02:08 Offset 67MB (0.66%) Done in 1:21:23 at 09:23:31
8:02:34 Offset 150MB (1.47%) Done in 1:05:18 at 09:07:52
8:03:03 Offset 234MB (2.29%) Done in 1:01:39 at 09:04:42
8:03:49 Offset 318MB (3.11%) Done in 1:09:19 at 09:13:08
...
9:06:23 Offset 10049MB (98.14%) Done in 0:01:13 at 09:07:36
9:06:59 Offset 10133MB (98.96%) Done in 0:00:41 at 09:07:40
9:07:29 Offset 10217MB (99.78%) Done in 0:00:08 at 09:07:37
All data are read; waiting for threads to finish...
Time elapsed waiting for 4 threads to finish:
(timeout in 60 min .)
Time elapsed waiting for 3 threads to finish:
7 sec (timeout in 59 min 53 sec.)
Thread 0: Processing 10200547328
Thread 2: Processing 10217324544
Thread 3: Processing 10234101760
Time elapsed waiting for 2 threads to finish:
13 sec (timeout in 59 min 47 sec.)
Thread 0: Processing 10200547328
Thread 2: Processing 10217324544
All Threads Finished!
Producer time spent waiting: 3645.8 sec.
Average consumer time spent waiting: 3.67321 sec.
*******************************************

Figure 17: Screenshot from Windows Explorer of the Output Directory Created by the
bulk_extractor run
** bulk_extractor is probably CPU bound. **
Run on a computer with more cores **
**
to get better performance.
**
**
*******************************************
Phase 2. Shutting down scanners
Phase 3. Creating Histograms
ccn histogram...
ccn_track2 histogram...
domain histogram...
email histogram...
ether histogram...
find histogram...
ip histogram...
lightgrep histogram...
tcp histogram...
telephone histogram...
url histogram...
url microsoft-live...
url services...
url facebook-address...
url facebook-id...
url searches...Elapsed time: 3991.77 sec.
Overall performance: 2.56524 MBytes/sec
Total email features found: 15277

All of the results from the bulk_extractor run are stored in the output directory charlie-200912-11. The contents of that directory after the run include the feature files, histogram files and
carved output. Figure 17 is a screenshot of the Windows output directory. Additionally, the
following output shows a list of the files, directories and their sizes under Linux:
C:\bulk_extractor\charlie-2009-12-11>ls -s -F

1 aes_keys.txt
0 alerts.txt

0 kml.txt
0 lightgrep.txt

4
1
0
0
23028
192
0
1696
36
24
1
508
0
0
0
0
32
4
12
504
1896

ccn.txt
ccn_histogram.txt
ccn_track2.txt
ccn_track2_histogram.txt
domain.txt
domain_histogram.txt
elf.txt
email.txt
email_histogram.txt
ether.txt
ether_histogram.txt
exif.txt
find.txt
find_histogram.txt
gps.txt
hex.txt
ip.txt
ip_histogram.txt
jpeg/
jpeg.txt
json.txt

0
196
1
108
3728
20
4
60
8
70108
1
0
6684
0
12
156
0
16432
20800
1864
29624

lightgrep_histogram.txt
packets.pcap
rar.txt
report.xml
rfc822.txt
tcp.txt
tcp_histogram.txt
telephone.txt
telephone_histogram.txt
url.txt
url_facebook-address.txt
url_facebook-id.txt
url_histogram.txt
url_microsoft-live.txt
url_searches.txt
url_services.txt
vcard.txt
windirs.txt
winpe.txt
winprefetch.txt
zip.txt

Many of the feature files and histograms are populated with data. Additionally, there were some
JPEG files carved and placed in the jpeg directory. In the following sections, we demonstrate how
to look at these results to discover more information about the disk user and the files contained
on the disk image.

9.2 Digital Media Triage
Digital media triage is the process of using the results of a rapid and automated analysis of the
media, performed when the media is first encountered to determine if the media is likely to have
information of intelligence value and, therefore, should be prioritized for immediate analysis.
bulk_extractor performs bulk data analysis to help investigators quickly decide which piece of
digital media is the most relevant and useful to an investigation. Thus, bulk_extractor can be
used to aid in investigations (through the identification of new leads and social networks) rather
than just aiding in conviction-support (through the identification of illegal materials)[4].
In this example, we look at the charlie-2009-12-11.E01 image to quickly assess what kinds
of information useful to an investigation might be present on the disk. For the purposes of this
example, we will assume we are investigating corporate fraud and trying to discover the answers
to the following questions:
• Who are the users of the drive?
• Who is this person communicating with?
• What kinds of websites have they have been visiting most often?
• What search terms are used?
To answer many of these questions, we look at the identify information on the drive including
email addresses, credit card information, search terms, Facebook IDs, domain names and vCard
data. The output files created by bulk_extractor contain all of this type of information that was
found on the disk image.

The scenario setup leads us to believe that Charlie is the user of the this drive (based on the name
of the disk image). First, we look at email.txt to find information about the email addresses
contained on the disk. The first two lines of the email features found are the following (each
block of text represents one long line of offset, feature and context):
50395384
n\x00o\x00m\x00b\x00r\x00e\x00_\x001\x002\x003\x00@\x00h\x00o\x00t
\x00m\x00a\x00i\x00l\x00.\x00c\x00o\x00m\x00
e\x00m\x00p\x00l\x00o\x00\x00\x0A\x00
\x09\x00n\x00o\x00m\x00b\x00r\x00e\x00_\x001\x002\x003\x00@\x00h\x00o\x00t\x00m
\x00a\x00i\x00l\x00.\x00c\x00o\x00m\x00\x0A\x00\x09\x00m\x00i\x00n\x00o\x00m\x00b\x00
50395432
m\x00i\x00n\x00o\x00m\x00b\x00r\x00e\x00@\x00m\x00s\x00n\x00.\x00c
\x00o\x00m\x00 i\x00l\x00.\x00c\x00o\x00m\x00\x0A\x00\x09\x00m\x00i\x00n\x00o\x00m
\x00b \x00r\x00e\x00@\x00m\x00s\x00n\x00.\x00c\x00o\x00m\x00\x0A\x00\x09\x00e\x00j
\x00e\x00m\x00p\x00l\x00

It is important to note that UTF-16 formatted text is escaped with \x00. This means that "\x00t
\x00e \x00x \x00t" translates to "text." The first two features found are "nombre_123@hotmail.com"
and "minombre@msn.com." Both of the offset values, 50395384 and 50395432, are early on
the disk. At this point, there is no way to know if either of these email addresses are of any
significance unless they happen to belong to a suspect or person related to the investigation. The
first set of email features found appear on the disk printed in UTF-16 formatted text, like the
lines above.
Further down in the feature file, we find the following:
9263459 charlie@m57.biz 21)(88=Charlie )(89\x0D\x0A
9263497 pat@m57.biz
=Pat McGoo )(8B=WELCOME TO

=Pat

Finding Charlie’s email address on the computer begins to further confirm the assumption that
this is his computer. The email_histogram.txt file provides important information. It shows
the most frequently occurring email addresses found on the disk. The following is an excerpt
from that top of that file:
n=875
n=651
n=605
n=411
n=395
n=379
n=356
n=312

mozilla@kewis.ch
(utf16=3)
charlie@m57.biz (utf16=120)
ajbanck@planet.nl
mikep@oeone.com
belhaire@ief.u-psud.fr
premium-server@thawte.com
(utf16=11)
lilmatt@mozilla.com
cedric.corazza@wanadoo.fr

This histogram output shows us that Charlie’s email address is the second most frequently occurring name on the disk. It would likely be the first but, as described in the scenario description,
this company has only been in business for three weeks and its employees are new users of
the computers. Looking at this histogram file also gives us some insight into who the user of
this disk is communicating with. Those email addresses occurring most frequently that are
not part of the software installed on the machine (such as ajbanck@planet.nl) might indicate
addresses of people with whom the drive user is corresponding or they may result from other
software or web pages that were downloaded. (In this case, the email is from a Firefox extension.)
The file domain.txt provides a list of all the "domains" and host names that were found. The
sources include URLS, email and dotted quads. Much of the beginning of the feature file is
populated with microsoft.com domains. This is largely due to the fact that the disk image is from
a Windows machine. Further down in the file we find the following:

53878576
53879083
53880076
53880536

www.uspto.gov
www.uspto.gov
ebiz1.uspto.gov
ebiz1.uspto.gov

Tech
568-4494\x0D\x0A
Microsoft
568-2294\x0D\x0A
Other sta
568-3503\x0D\x0A
Priority
668-7975\x0D\x0A
Text Tele

The next set of "telephone" numbers are clearly bogus numbers:
3649684174
3649684741
3649818237
3649818274

008-017-0108
000-031-0009
000-001-0005
000-004-0002

WA,98366,1,4031-008-017-0108,City of Port Or
98337,0.13,3768-000-031-0009,Kitsap County C
8312,2.25,"3768-000-001-0005, 3768-000-003-0
0-003-003, 3768-000-004-0002, 3768-000-005-0

Finally, many of the numbers found are legitimate ones. These numbers were all found in GZIP
compressed data:
3772517888-GZIP-28322 (831) 373-5555 onterey - (831) 373-5555
(831) 899-8300
(831) 899-8300
\x0A ’6543210123456788’ Inexact Rounde
4909069861 6543210123456788 \x0Addadd7541 add ’6543210123456788’ 0.5
4909069897 6543210123456788 5
-> ’6543210123456788’ Inexact Rounde
4909069947 6543210123456788 \x0Addadd7542 add ’6543210123456788’ 0.500000001
5304221350 5678901234560000 +4
-> 5678901234560000\x0D\x0Addshi052 shift
5612375618 6543210123456788 \x0D\x0Aaddx6240 add ’6543210123456788’ 0.499999999
5612375654 6543210123456788 499999999
-> ’6543210123456788’ Inexact Rounde
5612375703 6543210123456788 \x0D\x0Aaddx6241 add ’6543210123456788’ 0.5
5612375739 6543210123456788 5 -> ’6543210123456788’ Inexact Rounde
5612375788 6543210123456788 \x0D\x0Aaddx6242 add ’6543210123456788’ 0.500000001
5612715901 5700122152274696 div4036 divide 5700122152274696
5700122152251

In the above example, ‘525273347458642687’ looks like it could be a valid credit card number
from the context (\x0A is a new line). The number ‘4015751530102097’ looks like a random
number in a piece of Java Script. Note that both of those numbers were compressed; the offset
indicates they were found in GZIP streams (shown as a number followed by ‘-GZIP’). The
numbers whose context include “Inexact Rounde” are actually from Python source code and not
credit card numbers at all. Again, the ccn.txt tends to alert on a lot of false positives.
The ccn_track2.txt file did not find any information in this disk image but is also useful for
credit card fraud and identity theft investigations. It will contain credit card track 2 information
found on the disk image.
Using the files produced by bulk_extractor described above, an investigator can quickly review
a disk image for important information that is relevant to an investigation and find actionable
intelligence quickly.

9.3 Analyzing Imagery
The scenario described in the M57 Patents data is not necessarily relevant to an imagery investigation. However, there is imagery information on the disk. We use that information here
to demonstrate how imagery information can be analyzed by an investigator using bulk_extractor.
The file in the output directory, jpeg.txt, lists all JPEGs that were found on the disk whether
they were carved or not. bulk_extractor was run with default values meaning that only encoded
JPEGs were carved. The following excerpt from the JPEG file shows information about JPEGs
found on the disk image:
54798824
../Output/charlie-2009-12-11/jpeg/54783488.jpg
../Output/charlie-2009-12-11/jpeg/54783488.jpg15336
13823ce7c21587d31f6eb4474612e660

The JPEG described above was not carved because it was not encoded. However, the first
section “../Output/charlie-2009-12-11/jpeg/54783488.jpg” shows where the file would be found
in the output directories if it had been carved. The next section of information also gives
the file size, the hash type (in this case ‘md5’) and the hash value of the file (in this case
13823ce7c21587d31f6eb4474612e660). Note that this may not match the hash value of the file
in the original file system as bulk_extractor cannot properly carve fragmented files.
Information about encoded JPEGs can also be found in the jpeg.txt file. The following excerpt
shows a description of a JPEG found in a GZIP format on the disk:
3771686400-GZIP-8332
../Output/charlie-2009-12-11/jpeg/3771686400-GZIP-0.jpg
../Output/charlie-2009-12-11/jpeg/3771686400-GZIP-0.jpg
8332
5b77035c983b04996774370f735ea72a

The JPEG described above was carved and can be found in the /jpeg output directory in the file
named 3771686400-GZIP-0.jpg. The file also gives information about the filesize, hash type
and hash ID. That file is shown in the directory output shown below along with all of the encoded
JPEGs that were found on the disk image and were carved. The contents of the /jpeg directory
are as follows:
10037939712-GZIP-0.jpg
10117679783-ZIP-0.jpg

5324841013-ZIP-0.jpg
6039195136-GZIP-0.jpg

Figure 18: A JPEG carved from encoded data on the M57 Patents disk image
3761630720-GZIP-0.jpg
3764534784-GZIP-0.jpg
3771686400-GZIP-0.jpg
3771706880-GZIP-0.jpg
3771715072-GZIP-0.jpg
3771723264-GZIP-0.jpg
3771735552-GZIP-0.jpg
3771792896-GZIP-0.jpg
3771809280-GZIP-0.jpg
3771833856-GZIP-0.jpg
3771858432-GZIP-0.jpg
429788672-GZIP-0.jpg
5310405287-ZIP-0.jpg

6039215616-GZIP-0.jpg
6039223808-GZIP-0.jpg
6039232000-GZIP-0.jpg
6039244288-GZIP-0.jpg
6039301632-GZIP-0.jpg
6039318016-GZIP-0.jpg
6883925636-ZIP-0.jpg
6884040324-ZIP-0.jpg
6884056948-ZIP-0.jpg
7276064256-GZIP-0.jpg
7279128576-GZIP-0.jpg
8877243047-ZIP-0.jpg
9948655104-GZIP-0.jpg

All of these JPEG files can be viewed and used by investigators. The filename is the forensic
path of where the JPEG was found. The file 3771686400-GZIP-0.jpg mentioned above is
shown in Figure 18.

9.4 Password Cracking
The wordlist generates a list of all the words found on the disk that are between 6 and 14
characters long. The word list that is generated by the scanner can be very useful in determining
combinations of words to use for password cracking. The scanner is enabled by default because
it slows down the bulk_extractor run significantly. To show the word list in this example,
bulk_extractor was run again on the M57 Patents scenario data with the wordlist scanner enabled.
Running bulk_extractor on the command line with it enabled produces the following output:
C:\be\>bulk_extractor -e wordlist -o ../Output/charlie-wordlist charlie-2009-12-11.E01

bulk_extractor version: 1.4.0-beta4
Input file: charlie-2009-12-11.E01
Output directory: ../Output/charlie-wordlist
Disk Size: 10239860736
Threads: 4
12:58:46 Offset 67MB (0.66%) Done in 1:14:55 at 14:13:41
...
14:03:24 Offset 10217MB (99.78%) Done in 0:00:08 at 14:03:32
All data are read; waiting for threads to finish...
Time elapsed waiting for 4 threads to finish:
(timeout in 60 min .)
Time elapsed waiting for 4 threads to finish:
8 sec (timeout in 59 min 52 sec.)
Thread 0: Processing 10200547328
Thread 1: Processing 10234101760
Thread 2: Processing 10183770112
Thread 3: Processing 10217324544
Time elapsed waiting for 1 thread to finish:
14 sec (timeout in 59 min 46 sec.)

Thread 3: Processing 10217324544
All Threads Finished!
Producer time spent waiting: 3627.92 sec.
Average consumer time spent waiting: 4.1518 sec.
*******************************************
** bulk_extractor is probably CPU bound. **
Run on a computer with more cores **
**
to get better performance.
**
**
*******************************************
Phase 2. Shutting down scanners
Phase 3. Uniquifying and recombining wordlist
Phase 3. Creating Histograms
ccn histogram...
ccn_track2 histogram...
domain histogram...
email histogram...
ether histogram...
find histogram...
ip histogram...
lightgrep histogram...
tcp histogram...
telephone histogram...
url histogram...
url microsoft-live...
url services...
url facebook-address...
url facebook-id
url searches...Elapsed time: 4065.09 sec.
Overall performance: 2.51898 MBytes/sec
Total email features found: 152775

Note that it took 3991.71 seconds to run bulk_extractor without the wordlist scanner enabled and,
in this case, it took 4065.09 seconds with wordlist enabled. The new output directory contains
a file called wordlist.txt. That file has both filenames and words in it. The following is an
excerpt from that file:
50497556
50497624
50497692
50497760
50497828
50497896
50497987
50498009
50498050
50498057
50498090
50498097
50498153
50498181
50498229
50498238
50498335

usemodem.jpg
usemsn.jpg
usemsnnow.jpg
welcome.htm
whereNow.htm
xmlutil.js
^Photoshop
Resolution
Global
Lighting
Global
Altitude
Copyright
Japanese
Halftone
Settings
Transfer

The wordlist contains ALL words found on the disk between 6 and 14 characters long. Automated
programs can be used to generate passwords from combinations of these words. The wordlist
scanner also generates a split wordlist containing the same words found in the wordlist.txt
file with all words deduplicated, sorted by size and alphabetized. The following is an excerpt
from the file wordlist_split_000.txt generated from the disk image:
concluded|1
concluder/2
concluder/M
concluir/XQ
conclurai/x
conclusion,
conclusion.
conclusione
conclusions
conclusive,

The split wordlist is the file that is typically fed to password cracking software.

9.5 Post Processing
The programs identify_filenames.py and bulk_diff.py can provide further insight into the data
contained on the disk image. The identify_filenames.py program can be used on the feature
files produced from the bulk_extractor run to show the file location of the features that were
found. Running the program on all of the feature files produced by the bulk_extractor run
produces the following output (where charlie-2009-12-11 is the bulk_extractor output directory
and charlieAnnotatedOutput is where all the annotated files are written):
C:\be\>identify_filenames.py -all charlie-2009-12-11 charlieAnnotatedOutput

Reading file map by running fiwalk on charlie-2009-12-11.E01
Processed 1000 fileobjects in DFXML file
Processed 2000 fileobjects in DFXML file
...
Processed 39000 fileobjects in DFXML file
Processed 40000 fileobjects in DFXML file
feature_file: aes_keys.txt
feature_file: ccn.txt
feature_file: domain.txt
feature_file: email.txt
feature_file: ether.txt
feature_file: exif.txt
feature_file: ip.txt
feature_file: jpeg.txt
feature_file: json.txt
feature_file: rar.txt
feature_file: rfc822.txt
feature_file: telephone.txt
feature_file: url.txt
feature_file: windirs.txt
feature_file: winpe.txt
feature_file: winprefetch.txt
feature_file: zip.txt
******************************
754038 **
** Total Features:
754038 **
** Total Located:
******************************

Note, in this example that fiwalk is installed on the computer running the identify_filenames.py
program. The directory charlieAnnotatedOutput contains all of the annotated feature files,
showing the file location of the features. The directory contents are as follows:
annotated_aes_keys.txt
annotated_ccn.txt
annotated_domain.txt
annotated_email.txt
annotated_ether.txt
annotated_exif.txt
annotated_ip.txt
annotated_jpeg.txt
annotated_json.txt

annotated_rar.txt
annotated_rfc822.txt
annotated_telephone.txt
annotated_url.txt
annotated_windirs.txt
annotated_winpe.txt
annotated_winprefetch.txt
annotated_zip.txt

The annotated files display the feature with the file in which the feature was found (where it was
identified by the program). The following is an excerpt from the annotated_email.txt file:

27767966 pat@m57.biz
m: "Pat McGoo" \x0D\x0ATo: \x0D\x0ATo: ,\x0D\x0A\x09,\x0D\x0A\x09\x0D\x0AX-ASG-OrigSu Documents and Settings/Charlie/Application Data/Thunderbird/Profiles/4zy34x9h.def
ault/Mail/Local Folders/Inbox dcb794e350bd198c4279614eae6c8b76

The email address "pat@m57biz" was found in the file Documents and Settings/Charlie/
Application Data/Thunderbird/Profiles/4zy34x9h.default/Mail/Local Folders/Inbox

and investigators can refer to that location on the disk image to view the full text.
The program bulk_diff.py shows the difference between two bulk_extractor runs. In this case,
we used a disk image from the same user ("charlie") taken almost a month before the disk image
that has been used throughout this example. The disk image we have been using throughout
this example is dated December 11, 2009. The older disk image we downloaded for comparison is dated November 17, 2009. The earlier disk image data is stored in a file named
charlie-2009-11-17.E01 and can be downloaded from http://digitalcorpora.org/
corp/nps/scenarios/2009-m57-patents/drives-redacted/.
After running bulk_extractor using the earlier disk image, we ran the program bulk_diff.py on
the output of that disk image and on the output of the charlie-2009-12-11.E01 run. To run,
we typed the following, piping the output of the program to a file called bulkdiffoutput.txt:
bulk_diff.py /charlie-2009-11-17 /charlie-2009-12-11 > bulkdiffoutput.txt
The output shows the features differences on the disk image. The following is an excerpt of that
output:
domain_histogram.txt:
#in PRE
#in POST
Value
-----------------------------------------------------------------------------------401
4,470
4,069
patft.uspto.gov
181
3,151
2,970
www.wipo.int
295
3,157
2,862
www.google.com
0
2,537
2,537
l.yimg.com

The output specifically shows the differences in the histograms between the two runs across all
of the histogram files that were created. The excerpt above shows that "charlie" (the disk user)
visited the domain "patft.uspto.gov" frequently between the time the two images were recorder.
It was found 4,069 more times in the later disk image than in the one taken earlier. It also shows
that the domain "l.yimg.com" was not found on the earlier disk image but was found 2,537 times
on the later disk image. The results are sorted by the amount of the difference. This means that
features that are most different appear first. This can be very helpful because those features
generally give the most insight into the disk users activity over that period of time.

NPS DOMEX Users Image

NPS Test Disk Images are a set of disk images that have been created for testing computer
forensic tools. These images are free of non-public Personally Identifiable Information (PII) and
45

are approved for release to the general public. The NPS-created data in the images is public
domain and free of any copyright restriction; the images may contain some copyrighted data that
was made available by the copyright holder. These copyrights, where known, are noted in the
files themselves[1].
The NPS DOMEX users image is a disk image of a Windows XP SP3 system that has two users,
domexuser1 and domexuser2, who communicate with a third user (domexuser3) via IM and email.
The data is available for download at http://digitalcorpora.org/corp/nps/drives/
nps-2009-domexusers/. For this example, we use the file nps-2009-domexusers.E01
which includes the full system including the Microsoft Windows executables. Running bulk_extractor
on the command line produces the following output:
C:\be\>bulk_extractor -o ../Output/nps-2009-domexusers nps-2009-domexusers.E01

bulk_extractor version: 1.4.0-beta4
Input file: nps-2009-domexusers.E01
Output directory: ../Output/nps-2009-domexusers2
Disk Size: 42949672960
Threads: 4
16:50:53 Offset 67MB (0.16%) Done in 4:23:43 at 21:14:36
16:51:19 Offset 150MB (0.35%) Done in 3:58:37 at 20:49:56
...
16:13:12 Offset 42849MB (99.77%) Done in 0:00:11 at 16:13:23
16:13:13 Offset 42932MB (99.96%) Done in 0:00:01 at 16:13:14
All data are read; waiting for threads to finish...
Time elapsed waiting for 3 threads to finish:
(timeout in 60 min .)
Time elapsed waiting for 1 thread to finish:
6 sec (timeout in 59 min 54 sec.)
Thread 0: Processing 42932895744
Time elapsed waiting for 1 thread to finish:
12 sec (timeout in 59 min 48 sec.)
Thread 0: Processing 42932895744
All Threads Finished!
Producer time spent waiting: 4254.07 sec.
Average consumer time spent waiting: 89.309 sec.
*******************************************
** bulk_extractor is probably CPU bound. **
Run on a computer with more cores **
**
to get better performance.
**
**
*******************************************
Phase 2. Shutting down scanners
Phase 3. Creating Histograms
ccn histogram...
ccn_track2 histogram...
domain histogram...
email histogram...
ether histogram...
find histogram...
ip histogram...
lightgrep histogram...
tcp histogram...
telephone histogram...
url histogram...
url microsoft-live...
url services...
url facebook-address...
url facebook-id...
url searches...Elapsed time: 4846.74 sec.
Overall performance: 8.86156 MBytes/sec
Total email features found: 8774

All of the results from the bulk_extractor run are stored in the output directory nps-2009-domex.
The contents of that directory after the run are as follows:
1 aes_keys.txt
0 alerts.txt
1 ccn.txt

1 kml.txt
0 lightgrep.txt
0 lightgrep_histogram.txt

1
0
0
7364
44
0
1528
32
1
1
152
0
0
0
0
4
1
20
380
316

ccn_histogram.txt
ccn_track2.txt
ccn_track2_histogram.txt
domain.txt
domain_histogram.txt
elf.txt
email.txt
email_histogram.txt
ether.txt
ether_histogram.txt
exif.txt
find.txt
find_histogram.txt
gps.txt
hex.txt
ip.txt
ip_histogram.txt
jpeg/
jpeg.txt
json.txt

4
1
424
536
1
1
48
4
51888
0
0
1240
0
4
32
0
15228
26516
1312
1956

packets.pcap
rar.txt
report.xml
rfc822.txt
tcp.txt
tcp_histogram.txt
telephone.txt
telephone_histogram.txt
url.txt
url_facebook-address.txt
url_facebook-id.txt
url_histogram.txt
url_microsoft-live.txt
url_searches.txt
url_services.txt
vcard.txt
windirs.txt
winpe.txt
winprefetch.txt
zip.txt

For this example, we will focus on the files that are most important to malware investigations
and cyber investigations, showing how those files can be interpreted and used by investigators.

10.1

Malware Investigations

In a malware investigation, investigators are looking for information about programmatic intrusions. In this example, we examine all files that provide information about executables, Windows
directory entries and information downloaded from web-based applications. We recommend that
"-e xor" be enabled for malware investigations.
The file windirs.txt provides information about FAT32 and NTFS directories. It contains
most of the disk entries. The following is an excerpt showing one line from the file:
281954816
A0001801.dll
2008-10-21T00:45:51Z8224
2008-10-21T00:45:51Z2008-10-21T00:45:51Z
A0001801.dll1000000000000
01234373392008-10-21T00:45:51Z
11201731

The line from the file gives information about the disk entry A0001801.dll. It provides some
data about the file including the file size, file creation time (ctime) and time of last file modification (mtime). It is important to note that the error rate for FAT32 entries is high and those entries
should be ignored if the drive is not FAT.
For investigations on Windows disk images, such as the nps-2009-domexusers, the file
winpe.txt shows Windows executables related to the Windows Preinstallation Environment.
These file entries contain very long lines. The following is one line from the file:
42753536 87d84154e7789013878c6340a4d2d445

The first number is the offset and tells you were to find the file. Most executables are not
fragmented. The second is the MD5 has of the first 4k of the file that can be used to deduplicate
and look up the file in the hash database. Finally, the bulk of the information is contained in
the XML block that breaks out all of the Windows PE header information. It contains
information about the File header, the characteristics of the file, Windows header information
and section header information.
The file winprefetch.txt contains the information from carved files Windows Prefetch
that were discovered anywhere on the drive. bulk_extractor will carve the Prefetch files from
unallocated space. This extremely useful because Prefetch files are frequently deleted. A single
line in the prefetch output file is also very long. The following is only the beginning of one line
from the file:
55758336
MSIEXEC.EXE
Windows
XPMSIEXEC.EXE152
2008-10-30T03:17:27Z14
\x5CDEVICE\x5CHARDDISKVOLUME1\x5CWINDOWS\x5CSYSTEM32\x5CNTDLL.DLL
\x5CDEVICE\x5CHARDDISKVOLUME1\x5CWINDOWS\x5CSYSTEM32\x5CKERNEL32.DLL
...

Printing the line out here would cover almost two pages. It includes a lot of information about the
Prefetch file including the name of the executable, the name of the DLLs, the directory of DLLs,
the atime, the number of runs, the serial number, and the ctime. The Prefetch file is searchable
and useable by investigators searching for EXEs or DLLs related to a malware investigation.
JSON is the JavaScript Object Notation (used in Facebook, etc). The file json.txt provides the
offset, JSON and MD5 hash of the JSON information found on the disk image. bulk_extractor is
great at finding JSON in compressed streams and HIBER files. The following are a few lines
from the JSON file:
62836579 {"ask":["Ask"],"delicious":["Del.icio.us"],"digg":["Digg"],"email":["Email"],

"favorites":["Favorites"],"facebook":["Facebook"],"fark":["Fark"],"furl":["Furl"],
"google":["Google"],"live":["Live"],"myspace":["MySpace"],"myweb":["Yahoo MyWeb"
,"yahoo-myweb"],"newsvine":["Newsvine"],"reddit":["Reddit"],"sk*rt":["Sk*rt","skrt"],
"slashdot":["Slashdot"],"stumbleupon":["StumbleUpon","su"],"stylehive":["Stylehive"],
"tailrank":["Tailrank","tailrank2"],"technorati":["Technorati"],"thisnext":
["ThisNext"],"twitter":["Twitter"],"ballhype":["BallHype"],"yardbarker":
["Yardbarker"],"kaboodle":["Kaboodle"],"more":["More ..."]}
26d3b8c5010f4d39250dab3a1c1b839e
62842797 ["6jb4","3j1d","v1me","gu83","uefc","fq1j","r5l7","ftho","gdq9","717h",
"24b7","d0en","ads7","m9b4","n0lq","42c3","p5mp","7hbi","f0g6","7v98","mv86",
"d0ns","9a8a","64gg","jogl","cehp","mu2r","6h7h","sntb","94ds","n1fv","3a2i",
"3end","l42s","a9j","q3dj","s150","di3s","3nu5","sk74","e39d","mkvj","482d","kfej",
"nlcv","eroi","m6ee","rvaa","9nis","ef6b","g00q","b4hp","kbpq","bm4l","f7iu",
"e5gb","1sbj","rk0a","ck86","1etp","26sr","fivt","3v95","foqq","vtmj","canb",
"bchv","ku35","q4p9","gdkt","gng8","mdb9","ejjg","27k9","30mf","nene",
"smmm","q204","83ot","6kbr","df1o","1q0j","nh32","ebso","d6t5","f2dp",
"3sqp","i4cs","6k7b","a1pv","ki2l","1f7","d6lv","u7r5","9t0e","5h0l","j8kn",
"7akj","9tj","jmu3","1ir1"]
5a04af7518ad74c497c9e74b7025736e
64044544-GZIP-610 ["Top","Left","Right","Bottom"] 5354ef6838974b1979e49ee379883c56

Some of the JSON features found, such as the one located at ’62836579’, are comprised of a lot
of information in the notation. Other JSON features are very short, such as the feature located at
in the GZIP compressed stream at ’64044544-GZIP-610.’ All of the lines contain the MD5 hash
of the JSON that is used for deduplication.
The file elf.txt typically contains information about ELF executables, which is the executable
file format for Linux and Android systems. The sample corpus used in this example is from a
Windows machine and does not contain any ELF executables.

10.2

Cyber Investigations

Cyber investigations cover a wide variety of areas. However, most involve looking for encryption
keys, hash values or information about ethernet packets. bulk_extractor finds all of those things
on the disk and writes them to different output files. Of note, bulk_extractor also finds information in Base64 encoding and decompresses fragments of Windows Hibernation files. There are
not specific files created for that processing; the information found in data with these encodings
will be processed by other scanners and stored in the appropriate feature files. The fact that a
feature came from encoded data will be indicated in the forensic path. The information contained
therein may very well be relevant to cyber investigations.
AES encryption implementation system sometimes leaves keys in memory and bulk_extractor
finds those keys, usually in RAM, Swap or hibernation files. The keys can sometimes be used to
decrypt AES encrypted material. The file aes.txt contains the keys that are found. There was
only one AES key found on the nps-2009-domexusers disk image. The following is the line
that describes it from the keys file including the offset, key and key size descriptor (AES256):
1608580652
28 90 90 5e f7 ce b4 a7 2b 7d d9 45 d8 b0 56 99 97 f4 42
33 35 f1 54 9a 79 36 e7 1c 94 02 28 78 AES256

The file hex.txt contains extracted hexidecimal strings of a special length. The block sizes
cotained within it are either 128 or 256 due to the fact that those are the sizes used for encryption
keys and hash values. The disk image used in this example does not have any of those and the

file is blank.
bulk_extractor produces network information including PCAP files, Ethernet addresses, and
TCP/IP connections. The files ether.txt and ether_histogram.txt provide a list of
ethernet addresses from packets and ASCII. These are the addresses found on the disk and
located in ether.txt:
2435863552
2435863552
2435865088
2435865088
22637986225

00:0C:29:26:BB:CD
00:50:56:E0:FE:24
00:0C:29:26:BB:CD
00:50:56:E0:FE:24
00:80:C7:8F:6C:96

(ether_dhost)
(ether_shost)
(ether_dhost)
(ether_shost)
apter.\x0AExample: 00:80:C7:8F:6C:96\x00\x00

The file ether_histogram.txt groups these ethernet addresses in a histogram:
n=2
n=2
n=1

00:0C:29:26:BB:CD
00:50:56:E0:FE:24
00:80:C7:8F:6C:96

Packets likely traveled from 00:0C:29:26:BB:CD to 00:50:56:E0:FE:24. The other usage has
Ethernet addresses in UTF-16 format.
The file ip.txt contains IP addresses from packet carving, not from dotted quads. The following
is an excerpt from that file:
2435865102
2435865102
2805534669
8694397397
9047318477
9446959573
11295228937

inet_ntop win32
inet_ntop win32
123.12.0.192
135.5.0.234
123.12.0.192
135.5.0.234
1.70.0.1

struct ip L (src) cksum-ok
struct ip R (dst) cksum-ok
sockaddr_in
sockaddr_in
sockaddr_in
sockaddr_in
sockaddr_in

The L or R in the ’struct ip’ information indicates Local or Remote. This line also includes the IP
checksum is ok. The value could also be listed as "cksum-bad" to indicate it is bad. Bad checksums may indicate a false positive and not a legitimate IP address. Finally, the "sockaddr_in"
indicates the IP address is from a "sockaddr_in" structure. The file ip_histogram.txt removes the random noise that is found in the ip.txt. Here is an excerpt from the histogram
file:
n=5
n=4
n=4
n=3
n=2
n=2

2.172.0.101
123.12.0.192
inet_ntop win32
135.5.0.234
209.85.147.109
65.55.15.242

The file packets.pcap is a pcap file made from carved packet. To view that file, use any packet
analysis tool you like (such as tcpdump). Only packets carved from a PCAP file will have the
correct packet time stamp; others will given a time in 1970.
Finally, the file tcp.txt contains details about TCP (and UDP) network flows. It contains more
detail than ip.txt but investigators should be careful of false positives, as there are often many
in this file. The following are the two lines found in that file:
2435863566
2435865102

inet_ntop win32:80 -> inet_ntop win32:1034 (TCP)
inet_ntop win32:80 -> inet_ntop win32:1034 (TCP)

Size: 1472
Size: 1252

The file tcp_histogram.txt often provides further insight into the tcp information found
on the disk image. In this case, it does not because there were only two features found. It is
important to note that the histogram file still contains a lot of false positives.

Troubleshooting

Every forensic tool crashes at times because the tools are routinely used with data fragments,
non-standard codings, etc. One major issue is that the evidence that makes the tool crash typically
cannot be shared with the developer. The bulk_extractor system implements checkpointing
to protect the user and the results. bulk_extractor checkpoints the current page in the file
report.xml. After a crash, the user can just hit the up-arrow at the command line prompt and
return. bulk_extractor will restart at the next page.
All bulk_extractor users should join the bulk_extractor users Google group for more information and help with any issues encountered. To join, send an email to bulk_extractorusers+subscribe@googlegroups.com.
For the most part, the only kind of debugging bulk_extractor users should be doing is turning off
scanners. If bulk_extractor crashes repeatedly on a data set, the scanners can all be disabled and
then turned back on, one by one, until it crashes again. Then, the user can report the specific
scanner that made bulk_extractor crash on their disk image. In general, users who experience
crashes should feel free to report issues and problems to the developers via the Google users
group.
Users running the 32-bit version of bulk_extractor may occasionally encounter memory allocation
errors. This problem is more likely to occur on machines with a greater number of cores. Our
testing has shown this to be an issue using one of our test data sets on a 32-bit machine with 12
cores. In the user encounters memory allocation errors with bulk_extractor they will likely see
an error similar to the following:
bulk_extractor scan error: ’std::exception Scanner: gzip Exception:
std::bad_alloc sbuf.pos0: (|21894266880) bufsize=20971520’

Memory allocation errors such as the one shown above will contain the phrase “bad_alloc” somewhere in the message. If the user encounters this error, they should try running bulk_extractor
with fewer threads. For example, the following command will run bulk_extractor with only 4
threads (the -j option changes this parameter):
bulk_extractor -j 4 -o output mydisk.raw
Reducing the number of threads and re-running the program should eliminate the problem.
Users may encounter errors if they are processing a large disk image and trying to write the
output of bulk_extractor to an output file directory on a smaller drive. In that case the user might
see an error similiar to the following:
bulk_extractor version: 1.4.0-beta6
Input file: G:\nps-2011-2tb\nps-2011-2tb.E01
Output directory: C:\Users\Mark Richer\Documents\BE Testing\OFD nps-2011-2tb 64bit
Disk Size: 2000054960128
Threads: 12
DISK FULL
DISK FULL

DISK FULL
*** carve: Cannot write(pos=7,0 len=24724184): No space left on device
DISK FULL
DISK FULL
DISK FULL
DISK FULL
DISK FULL
*** carve: Cannot write(pos=7,0 len=24724198): No space left on device
*** carve: Cannot write(pos=7,0 len=49160): No space left on device
*** carve: Cannot create C:\Users\Mark Richer\Documents\BE Testing\OFD nps-2011-2tb
64bit/kml/000/426602508288-ZIP-0.kml: No space left on device
Could not make directory C:\Users\Mark Richer\Documents\BE Testing\OFD nps-2011-2tb
64bit/kml/001: No space left on device
Phase 3. Creating Histograms
Cannot open histogram output file: C:\Users\Mark Richer\Documents\BE Testing\OFD
nps-2011-2tb 64bit/ccn_track2_histogram.txt
Elapsed time: 45111.4 sec.
Overall performance: 44.3359 MBytes/sec
Total email features found: 6716934

If this situation is encountered, the solution is to run bulk_extractor with an output directory on a
machine with more available disk space so that bulk_extractor has room to create all the output
files and directories required.

584102 A584102

Navigation menu

Versions of this User Manual:

Views

Navigation