Advanced Fuzzing and Crash Analysis

This class is designed to introduce students to the best tools and technology available for automating vulnerability discovery and crash triage with a focus on delivering a practical approach to applying this technology in real deployments.

Through an applied understanding of introductory program analysis and binary translation, techniques for finding various bug classes and methods for improved crash debugging will be discussed. We will take a deep dive into fuzzing, covering all aspects of this practical approach to finding bugs. As the most approachable and versatile of the available tools, the student will apply various fuzzing techniques to several real-world pieces of software. Students will learn strategies for analyzing attack surface, writing grammars, and generating effective corpus. We will explore in detail the latest innovations such as harnessing code coverage for guided evolutionary fuzzing and symbolic reasoning for concolic fuzzing.

We approach crash analysis through the lens of scriptable debuggers and program analysis. We will apply tools like reverse debugging and memory debuggers to assist in interactively diagnosing root cause of crashes. Then we will leverage the power of dynamic taint tracking and graph slicing to help isolate the path of user controlled input in the program and identify the exact input bytes influencing a crash. Lastly, we will look at possible ways to determine the impact of a vulnerability.

This class will focus on x86/x64 architecture and target file parsers, network parsers and browsers on both Windows and Linux environments.

Key Learning Objectives

• Learn an effective strategy for using the latest tools & technology to discover vulnerabilities
• Master the latest fuzzing techniques for file, network, and browser fuzzing
• Learn grammar fuzzing, evolutionary fuzzing, in-memory fuzzing, and symbolic fuzzing
• Best practices for corpus generation, fuzzer deployment, and targeting
• Leverage dynamic binary translation for efficient tracing and deep program inspection
• Learn how to leverage time travel debugging for crash triage on Linux and Windows
• Introduction to intermediate languages for program analysis
• Apply powerful techniques like taint analysis and graph slicing towards crash analysis

Topics Covered

• Analysis of generational and mutational fuzzing
  – Attack surface analysis
  – Effective mutation engines
  – Effective corpus generation
  – Protocol and file format grammars
  – Crash detection
• Fuzzing file and network parsers with coverage guided fuzzing
  – Fuzz any Ubuntu/Debian package with AFL
  – Modifying targets and writing harnesses with LibFuzzer
  – Fuzzing closed source parsers with QEMU and Dyninst
• Best practices for high performance fuzzing
  – System configuration
  – Corpus generation techniques
  – Cross-fuzzing difficult parsers
• Dynamic Binary Translation for Fuzzing and Triage
  – Effectively instrument Linux and Windows with binary translation
  – Introduction to Valgrind, Dr. Memory, and Address Sanitizer
  – Introduction to PIN, DynamoRIO, and Dyninst internals
  – Identifying hook locations with Debuggers and DBI
  – Fuzzing kernels and other architectures with QEMU
• Fuzzing parsers with WinAFL
  – Optimizing harnesses for exported APIs
  – Hooking closed source command line applications
  – Deep hooks into private library functions with global state
  – Fuzzing internal data streams in complex OLE objects
• Fuzzing browsers with evolutionary grammar fuzzing
  – Understanding grammars and object models
  – Fuzzing object models with dynamic grammar fuzzing
  – Improving grammar fuzzers with feedback metrics
• Time Travel Debugging
  – Introduction to time travel debugging
  – Crash analysis with reverse debugging on Linux
  – Crash analysis with reverse debugging on Windows
• Taint assisted root cause analysis
  – Introduction to dynamic taint analysis
  – Taint slicing for root cause analysis
• Symbolic and Concolic Execution
  – Introduction to constraint solving
  – Concolic execution for test case generation
  – Hybrid fuzzing with concolic execution