Fuzzing-101

Do you want to learn how to fuzz like a real expert, but don't know how to start?

If so, this is the course for you!

10 real targets, 10 exercises. Are you able to solve all 10?

Structure

Exercise No.	Target	CVEs to find	Time estimated	Main topics
Exercise 1	Xpdf	CVE-2019-13288	120 mins	Afl-clang-fast, Afl-fuzz, GDB
Exercise 2	libexif	CVE-2009-3895, CVE-2012-2836	6 hours	Afl-clang-lto, Fuzz libraries, Eclipse IDE
Exercise 3	TCPdump	CVE-2017-13028	4 hours	ASan, Sanitizers
Exercise 4	LibTIFF	CVE-2016-9297	3 hours	Code coverage, LCOV
Exercise 5	Libxml2	CVE-2017-9048	3 hours	Dictionaries, Basic parallelization, Fuzzing command-line arguments
Exercise 6	GIMP	CVE-2016-4994, Bonus bugs	7 hours	Persistent fuzzing, Fuzzing interactive applications
Exercise 7	VLC media player	CVE-2019-14776	6 hours	Partial instrumentation, Fuzzing harness
Exercise 8	Adobe Reader		8 hours	Fuzzing closed-source applications, QEMU instrumentation
Exercise 9	7-Zip	CVE-2016-2334	8 hours	WinAFL, Fuzzing Windows Applications
Exercise 10 (Final Challenge)	Google Chrome / V8	CVE-2019-5847	8 hours	Fuzzilli, Fuzzing Javascript engines

Changelog

• 02/14/2022: Fixed some 'wget' typos in Exercise 5
• 11/25/2021: Exercise 3 updated with some fixes.

Who is the course intended for?

• Anyone wishing to learn fuzzing basics
• Anyone who wants to learn how to find vulnerabilities in real software projects.

Requirements

• All you need for this course is a running Linux system with an internet connection. You will find a suitable VMware image in the exercises.
• At least basic Linux skills are highly recommended.
• All the exercises have been tested on Ubuntu 20.04.2 LTS. You can download it from here
• In this course we're going to use AFL++, a newer and superior fork of Michał "lcamtuf" Zalewski's AFL, for solving the fuzzing exercises.

What is fuzzing?

Fuzz testing (or fuzzing) is an automated software testing technique that is based on feeding the program with random/mutated input values and monitoring it for exceptions/crashes.

AFL, libFuzzer and HonggFuzz are three of the most successful fuzzers when it comes to real world applications. All three are examples of Coverage-guided evolutionary fuzzers.

Coverage-guided evolutionary fuzzer

• Evolutionary: is a metaheuristic approach inspired by evolutionary algorithms, which basically consists in the evolution and mutation of the initial subset (seeds) over time, by using a selection criteria (ex. coverage).

• Coverage-guided: To increase the chance of finding new crashes, coverage-guided fuzzers gather and compare code coverage data between different inputs (usually through instrumentation) and pick those inputs which lead to new execution paths.

Simplification of the coverage gathering process of a coverage-guided evolutionary fuzzer

Thanks

Thanks for their help:

• Xavier RENE-CORAIL
• Alan Vivona
• Jason White
• Octavio Gianatiempo
• van Hauser
• Marc Poulhiès
• Xu Hanyu
• tclan126
• epi052
• Jeremias Gomes