Skip to content

wolframroesler/afl-demo

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Fuzz Testing Demonstration

This is a simple demonstration of how to fuzz-test a C++ program with the AFL ("American Fuzzy Lop") fuzz tester.

AFL screen shot

What Is Fuzz Testing?

The goal of unit testing is to feed well-known input into your program and compare its output to the expected results.

The goal of fuzz testing (or fuzzing) is to feed random data into your program until it crashes. Every crash is an indication of undefined behavior and thus of a potential security hole.

Files

  • main.cpp: Our demonstration program.
  • CMakeLists.txt: Build configuration.
  • testcases: Example input for the program.
  • fuzz.sh: Runs the AFL fuzzer.

Running The Demonstration Program

The demonstration program consists of an (intentionally buggy) implementation of a URI decoder function and a simple main function that calls it with input read from stdin. A URI decoder converts hex sequences like %2f into the appropriate characters (in this case, /).

Build The Program

$ cd afl-demo
$ mkdir build
$ cd $_
$ cmake ..
$ make

Try It Out

The afldemo program reads URI encoded input from stdin and decodes it to stdout.

$ cd afl-demo/build
$ ./afldemo <<<Hello+World%21
Hello World!

Looks ok, doesn't it? Now wait and see ...

Fuzzing The Demonstration Program

Throughout the following, replace 2.51b by the appropriate version of AFL (that is, what afl-latest.tgz unpacks into).

Download And Build AFL

$ cd your-preferred-directory
$ mkdir afl
$ cd $_
$ wget http://lcamtuf.coredump.cx/afl/releases/afl-latest.tgz
$ tar xzf afl-latest.tgz
$ cd afl-2.51b/
$ make

Create Initial Test Cases

To get AFL started, give it some legal input that your programm processes. Create one file per input string in the testcasesdirectory.

$ cd afl-demo
$ mkdir testcases
$ cd $_
$ echo "your first test input" >01.txt
$ echo "your second test input" >02.txt

Build The Instrumented Test Program And Run The Fuzzer

$ cd afl-demo
$ mkdir aflbuild
$ cd $_
$ CC=/path/to/afl/afl-2.51b/afl-gcc CXX=/path/to/afl/afl-2.51b/afl-g++ cmake ..
$ make
$ /path/to/afl/afl-2.51b/afl-fuzz -i ../testcases -o ../findings ./afldemo

Or, when using this demonstration project, use the provided shell script that builds the instrumented executable and runs the fuzzer. Edit fuzz.sh first and set variable AFL to the directory in which you built AFL. Then:

$ cd afl-demo
$ ./fuzz.sh

Evaluating The Fuzzer Results

Let the fuzzer run until it finds one or more crashes or hangs, then terminate it with ^C. Fuzzing can take a considerable time (hours or days depending on the complexity of your program; for this demonstration, however, the first crashes are fund after about ten minutes on my machine).

You find the input that made the program crash or hang in the findings/crashes and findings/hangs directories. Redirect these files into your program to reproduce a crash or hang.

$ ls -l findings/crashes/
total 3432
-rw-------  1 wolfram  staff    657 20 Jul 15:19 README.txt
-rw-------  1 wolfram  staff  19311 20 Jul 15:19 id:000000,sig:11,src:000002+000031,op:splice,rep:128
-rw-------  1 wolfram  staff   1288 20 Jul 15:19 id:000001,sig:11,src:000002+000028,op:splice,rep:32
-rw-------  1 wolfram  staff    292 20 Jul 15:19 id:000002,sig:11,src:000002+000028,op:splice,rep:128
-rw-------  1 wolfram  staff  23322 20 Jul 15:19 id:000003,sig:11,src:000002+000028,op:splice,rep:16
-rw-------  1 wolfram  staff  14905 20 Jul 15:19 id:000004,sig:11,src:000002+000013,op:splice,rep:128
-rw-------  1 wolfram  staff   1535 20 Jul 15:19 id:000005,sig:11,src:000002+000013,op:splice,rep:64
…
$ build/afldemo <findings/crashes/id:000000,sig:11,src:000002+000031,op:splice,rep:128
Bus error: 10

Debug and fix your program until it processes the offending input cleanly (it may be a good idea to add AFL's findings to your suite of unit tests). Repeat (you may want to rm -fr findings for a fresh start).

Fuzzing Black-Box Binaries

To fuzz-test a program for which you don't have the source code, use the QEMU tool. Linux only.

In the following example, $AFL is the directory in which AFL was built.

$ sudo apt install libtool-bin
$ sudo apt install libglib2.0-dev

$ cd $AFL/qemu_mode
$ ./build_qemu_support.sh

$ cd your/testing/directory
$ AFL_SKIP_CPUFREQ=1 $AFL/afl-fuzz -Q -i testcases -o findings /path/to/executable

Fuzzing Java programs is tricky because you want to fuzz the actual program, not the Java runtime package. One solution (albeit abandoned and outdated): https://github.com/floyd-fuh/AFL_GCJ_Fuzzing_Simple

Links

AFL homepage: http://lcamtuf.coredump.cx/afl/

About fuzz testing: https://en.wikipedia.org/wiki/Fuzzing

The OSS Fuzz project: https://github.com/google/oss-fuzz


Wolfram Rösler • wolfram@roesler-ac.dehttps://twitter.com/wolframroeslerhttps://github.com/wolframroesler

About

Simple demonstration for how to fuzz test a C++ program with AFL.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published