Hashing Fuzzing: Introducing Input Diversity to Improve Crash Detection

The utility of a test set program inputs is strongly influenced by its diversity and size. Syntax coverage has become standard proxy for diversity. Although more sophisticated measures exist, such as proximity sample to uniform distribution, methods use them tend be type dependent. We r-wise hash functions create novel, semantics preserving, testability transformation C programs that we call HashFuzz. Use HashFuzz improves the sets produced instrumentation-based fuzzers. evaluate effect on eight from Google Fuzzer Test Suite using four state-of-the-art fuzzers have been widely used in previous research. demonstrate pronounced improvements performance transformed across all used. These include strong every case, maintenance or small improvement branch – up 4.8 perent best significant unique crash detection numbers between 28 97 increases compared untransformed programs.