Efficient greybox fuzzing framework with tabu search

Recently, fuzzing has emerged as a prominent technique for automatically detecting a wide range of security vulnerabilities in cybersecurity. Nevertheless, existing fuzzers remain limited in seed scheduling and energy allocation, as they typically rely on coarse-grained criteria or computationally intensive methods to evaluate seed quality. In this paper, we propose a novel adaptive fuzzing framework, namely TabuFuzz, which aims to prioritize seeds with a higher likelihood of generating interesting variants through mutation and allocate more energy to that seed. Specifically, TabuFuzz utilizes a tabu search (TS) algorithm to model seed scheduling, enabling the precise selection of promising seeds. In addition, we design an adaptive energy allocation strategy that exploits fine-grained dynamic execution information, thereby enhancing overall fuzzing efficiency. We implement TabuFuzz on top of AFL and evaluate it on nine widely-used real-world programs. Across six independent 24-hour trials, TabuFuzz achieved higher observed path exploration and crash discovery than the evaluated baselines on most target programs, with improvements of 19.89% and 160.87% over AFL, respectively. Comparative analyses further show that TabuFuzz recorded higher numbers of paths and crashes than representative fuzzers such as FairFuzz, MOPT, EcoFuzz, and AFL + + under the reported experimental setup.