In the following, we describe a variety of ideas that could be implemented for future AFL++ versions.
For GSOC2020 interested students please see AFLplusplus#208
Currently, AFL++'s mutation does not have deeper knowledge about the fuzzed binary, apart from feedback, even though the developer may have insights about the target.
A developer may choose to provide dictionaries and implement own mutations in python or C, but an easy mutator that behaves according to a given grammar, does not exist.
State-of-the-art research on grammar fuzzing has some problems in their implementations like code quality, scalability, or ease of use and other common issues of the academic code.
We aim to develop a pluggable grammar mutator for afl++ that combines various results.
Mentor: andreafioraldi
Work on the MOpt mutator that is already in AFL++.
This is an excellent mutations scheduler based on Particle Swarm Optimization but the current implementation schedule only the mutations that were present on AFL.
AFL++ added a lot of optional mutators like the Input-2-State one based on Redqueen, the Radamsa mutator, the Custom mutator (the user can define its own mutator) and the work is to generalize MOpt for all the current and future mutators.
Mentor: vanhauser-thc or andreafioraldi
Either Port the patch to the upcoming Ubuntu LTS 20.04 default kernel and provide a qemu-kvm image or find a different userspace snapshot solution that has a good performance and is reliable, e.g. with docker. perf-fuzz The perf-fuzz kernel can be found at https://github.com/sslab-gatech/perf-fuzz There also is/was a FreeBSD project at https://github.com/veracode-research/freebsd-perf-fuzz
This enables snapshot fuzzing on Linux with an incredible performance!
Mentor: any Idea/Issue tracker: AFLplusplus#248
First tests to use QEMU 4 for binary-only AFL++ showed that caching behavior changed, which vastly decreases fuzzing speeds.
This is the cause why, right now, we cannot switch to QEMU 4.2.
Understanding the current instrumentation and fixing the current caching issues will be needed.
Mentor: andreafioraldi
Currently, AFL++ can be used for source code fuzzing and traditional binaries. With the rise of WASM as compile target, however, a novel way of instrumentation needs to be implemented for binaries compiled to Webassembly. This can either be done by inserting instrumentation directly into the WASM AST, or by patching feedback into a WASM VMs of choice, similar to the current Unicorn instrumentation.
Mentor: any
Something with machine learning, better than NEUZZ :-) Either improve a single mutator thorugh learning of many different bugs (a bug class) or gather deep insights about a single target beforehand (CFG, DFG, VFG, ...?) and improve performance for a single target.
Mentor: domenukk
Right now, afl-fuzz is single threaded, cannot safely be embedded in tools, and not multi-threaded. It makes use of a large number of globals, must always be the parent process and exec child processes. Instead, afl-fuzz could be refactored to contain no global state and globals. This allows for different use cases that could be implemented during this project. Note that in the mean time a lot has happened here already, but e.g. making it all work and implement multithreading in afl-fuzz ... there is still quite some work to do.
Mentor: hexcoder- or vanhauser-thc
AFL++ supports collison-free maps using an LTO (link-time-optimization) pass. This should be possible to implement for QEMU and Unicorn instrumentations. As the forkserver parent caches just in time translated translation blocks, adding a simple counter between jumps should be doable.
Note: this is already in development for qemu by Andrea, so for people who want to contribute it might make more sense to port his solution to unicorn.
Mentor: andreafioraldi or domenukk Issue/idea tracker: AFLplusplus#237
Finally, we are open to proposals! Create an issue at https://github.com/AFLplusplus/AFLplusplus/issues and let's discuss :-)