This repository contains a library (afl.c, afl.h) based on the llvm runtime from AFL. This library makes it easy to provide custom coverage feedback to AFL, replacing AFLs custom instrumentation which uses branch coverage as its main metric.

A very simple example (provided in main.c) looks like this:

int main() {
	afl_init();

	switch(getc(stdin)) {
		case 'a':
			afl_set('a', 1);
		case 'b':
			afl_set('b', 1);
		case 'c':
			afl_set('c', 1);
		default:
			afl_set(0, 1);
	}
}

Here afl_init initialized the shared memory region and the pipes that are used to communicate with AFL. It also starts the fork server which will allow AFL to execute your code after the call to afl_init over and over again without reloading your binary.

The provided example program now reads one byte from stdin and gives different feedback depending on its content. Thus we expect AFL to come up with 4 different inputs which correspond to 4 different paths. One should correspond to a, one to b, one to c and the other to something other than a, b or c.

To run the example, you can use the included Makefile. You will probably need to adjust AFL_DIR to point to your local checkout of AFL.

Then just type: > make run

This starts AFL which will begin fuzzing your binary. Pretty quickly the number of paths should go up to 4 after which you can stop AFL (Ctrl + c).

Now you can have a look at out/queue to inspect the inputs that AFL produced.

That's it. Happy Fuzzing!