Practical Reverse Engineering: x86, x64, ARM, Windows Kernel, Reversing Tools, and Obfuscation

Practical Reverse Engineering: x86, x64, ARM, Windows Kernel, Reversing Tools, and ObfuscationReviews
Author: Bruce Dang
Pub Date: 2014
ISBN: 978-1118787311
Pages: 384
Language: English
Format: PDF/EPUB/MOBI
Size: 10 Mb

Download

Analyzing how hacks are done, so as to stop them in the future Reverse engineering is the process of analyzing hardware or software and understanding it, without having access to the source code or design documents. Hackers are able to reverse engineer systems and exploit what they find with scary results. Now the good guys can use the same tools to thwart these threats. Practical Reverse Engineering goes under the hood of reverse engineering for security analysts, security engineers, and system programmers, so they can learn how to use these same processes to stop hackers in their tracks. The book covers x86, x64, and ARM (the first book to cover all three); Windows kernel-mode code rootkits and drivers; virtual machine protection techniques; and much more. Best of all, it offers a systematic approach to the material, with plenty of hands-on exercises and real-world examples. Offers a systematic approach to understanding reverse engineering, with hands-on exercises and real-world examples Covers x86, x64, and advanced RISC machine (ARM) architectures as well as deobfuscation and virtual machine protection techniques Provides special coverage of Windows kernel-mode code (rootkits/drivers), a topic not often covered elsewhere, and explains how to analyze drivers step by step Demystifies topics that have a steep learning curve Includes a bonus chapter on reverse engineering tools Practical Reverse Engineering: Using x86, x64, ARM, Windows Kernel, and Reversing Tools provides crucial, up-to-date guidance for a broad range of IT professionals.

+

Obfuscation

Speaking in the abstract, “obfuscation” can be viewed in terms of program transformations. The goal of such methods is to take as input a program, and produce as output a new program that has the same computational effect as the original program (formally speaking, this property is called semantic equivalence or computational equivalence), but at the same time it is “more difficult” to analyze.

The notion of “difficulty of analysis” has long been defined informally, without any backing mathematical rigor. For example, it is widely believed that—insofar as a human analyst is concerned—a program’s size is an indicator of the difficulty in analyzing it. A program that consumes 20,000 instructions in performing a single operation might be thought to be “more difficult” to analyze than one that takes one instruction to perform the same operation. Such assumptions are dubious and have attracted the scrutiny of theoreticians (such as that by Mila Dalla Preda and Barak et al.2).

Several models have been proposed to represent an obfuscator, and (in a dual way) a deobfuscator. These models are useful to improve the design of obfuscation tools and to reason about their robustness, through adapted criteria. Among them, two models are of special interest.

The first model is suited for the analysis of cryptographic mechanisms, in the so-called white box attack context. This model defines an attacker as a probabilistic algorithm that tries to deduce a pertinent property from a protected program. More precisely, it tries to extract information other than what can be trivially deduced from the analysis of the program’s inputs and outputs. This information is pertinent in the sense that it enables the attacker to bypass a security function or represents itself as critical data of the protected program. In a dual way, an obfuscator is defined in this model as a virtual black box’s probabilistic generator, an ideal obfuscator ensuring that the protected program analysis does not provide more information than the analysis of its input and output distributions.

Another way to formalize an attacker is to define the reverse engineering action as an abstract interpretation of the concrete semantics of the protected program. Such a definition is naturally suited to the static analysis of the program’s data flow, which is a first step before the application of optimization transformations. In a dual way, an obfuscator is defined in the abstract interpretation model as a specialized compiler, parameterized by some semantic properties that are not preserved.

The goal of these modeling attempts is to get some objective criteria relative to the effective robustness of obfuscation transformations. Indeed, many problems that were once thought to be difficult can be efficiently attacked via judicious application of code analysis techniques. Many methods that have arisen in the context of more conventional topics in programming language theory (such as compilers and formal verification) can be repurposed for the sake of defeating obfuscation.

This chapter begins with a survey of existing obfuscation techniques as commonly found in real-world situations. It then covers the various available methods and tools developed to analyze and possibly break obfuscation code. Finally, it provides an example of a difficult, modern obfuscation scheme, and details its circumvention using state-of-the-art analysis techniques.

Control-Based Obfuscation

When reverse engineering compiler-generated code, reverse engineers are able to rely on the predictability of the compiler’s translations of control flow constructs. In doing so, they can quickly ascertain the control flow structure of the original code at a level of abstraction higher than assembly language. Along the way, the reverse engineer relies upon a host of assumptions about how compilers generate code. In a pure compiled program, all code in a basic block will be most often sequentially located (heavy compiler optimizations can possibly render this basic premise null and void). Temporally related blocks usually will, too. A CALL instruction always corresponds to the invocation of some function. The RET instruction, too, will almost always signify the end of some function and its return to its caller. Indirect jumps, such as for implementing switch statements, appear infrequently and follow standard schemas.
Control-based obfuscation attacks these planks of standard reverse engineering, in a way that complicates both static and dynamic analyses. Standard static analysis tools make similar assumptions as human reverse engineers, in particular:

Destruction of Sequential and Temporal Locality

As stated, and as understood intrinsically by those who reverse engineer compiled code, the instructions within a single, compiled basic block lie in one straight-line sequence. This property is called sequential locality. Furthermore, compiler optimizers attempt to put basic blocks that are related to one another (for example, a block and its successors) nearby, for the purpose of maximizing instruction cache locality and reducing the number of branches in the compiled output. We call this property the sequential locality of temporally related code. When you reverse engineer compiled code, these properties customarily hold true. One learns in analyzing such code that all of the code responsible for a single unit of functionality will be neatly contained in a single region, and that the proximate control-flow neighbors will be nearby and similarly sequentially located.

A very old technique in program obfuscation is to introduce unconditional branches to destroy this aspect of familiarity that reverse engineers organically obtain through typical endeavors. Here is a simple example: