3 ms·
afl-fuzz can be parallelized fairly easily. The exchanged data amounts to newly-discovered, interesting inputs that then seed the subsequent fuzzing work.
by f- 12y ago
afl-fuzz can be parallelized fairly easily. The exchanged data amounts to newly-discovered, interesting inputs that then seed the subsequent fuzzing work.
- dalke 12y agoGreat work with afl. I tried it out last week, and found two segfaults and a stack smash detection in one program. I tried it on another program, only to have gcc crash with an internal error. :( By parallelized, do you mean on the same machine or across a distributed cluster? If they only share the same set of interesting inputs, won't the different nodes also end up searching much of the same space? .... Hmm, no I see how I could be wrong. With interesting seeds, boring space is easy to re-identify, so there's a trivial amount of duplicate work, and the rest is spent just trying to find something new and interesting.