Exploração de código Assembly Intel X64 oti mizado para o fuzzer AFL

Abstract

In recent decades, there has been an undeniable increase in the popularization of digital technology worldwide, along with a growing dependency on it in all sectors of society. However, the level of concern with cybersecurity, by companies and related professionals, usually does not align with the current recommended security standards. Consequently, companies in the sector have frequently observed undesirable occurrences of security breaches, denial of service, leakage of sensitive data, and all sorts of unwanted actions. In this context, it becomes evident the need for the development of technologies that allow, if not eliminating, at least significantly mitigating the attack surface of modern commercial applications. In this study, an empirical investigation was conducted to assess the optimization potential at the assembly level for a contemporary fuzzer application. By employing the AVX2 set of vectorized instructions, parallelism at the CPU instruction level was leveraged to optimize the main function present in the studied fuzzer, resulting in substantial performance improvements for this specific function. Consequently, once integrated into the application, a reduction of 37.37% in the execution time overhead of the optimized function was achieved in the best-case scenario, while in the worst-case scenario, the overhead reduction was of 16.14%.