AFLplusplus - FRIDA

[Jon Wilson]

I have just made this PR looking to add FRIDA to AFLplusplus as an option for pure binary fuzzing:

https://github.com/AFLplusplus/AFLplusplus/pull/833

I think it should be possible to achieve feature parity with QEMU as well as providing a very similar user-experience. It might even be possible to share some features such as libqasan etc down the line.

What is the best forum for discussing new features/design decisions?

Also if anyone has any input on how best to performance test this against QEMU, then it would be very welcome. At present I have built the libpng benchmark for fuzzbench and using the sample images in libpng for a corpus, this is the results I get with a 30 second run...

QEMU 4500 execs/sec
FRIDA 1500 execs/sec

I've not been able to replicate the 12x performance gain over QEMU reported in utils/afl_frida, but I'm not sure if this is down to a different in testing methodology or something I have screwed up.

Any comments or thoughts would be most welcome.

Thanks.

Jon