gear
gear copied to clipboard
Initial work on Fuzzer randomization
Resolves #3988.
Notes
At the moment this PR contains simple implementation, I am still working on making this randomization better. At the moment some runs show success rates of 80-90% and others show 30%. Coverage is also increased but ever so slightly, I am guessing more testing with syscall injections is needed but for this I need to figure out how to randomize them properly.
@techraed request for review now. I've adjusted parameters to match master fuzzer success/failure rate and potentially have better success rate. Still need to verify this on dedicated server though
@techraed
Great!!!
What's with checks on a server - is everything satisfiable?
As we discussed in DM compared to master this branch runs 10-15x more programs compared to master
| master | ap/random-fuzzer | |
|---|---|---|
| Success | 14795 | 256865 |
| Traps | 19246 | 110748 |
Also as you can see we beat success rate of master fuzzer which is a significant improvement.
- ap/random-fuzzer: 69.87% success, 30.12% traps
- master: 43.46% success, 56.53% traps
EDIT: fix numbers