Research on Linux Kernel Fuzz Testing Technology Based on Generative Adversarial Networks

Currently, Linux kernel fuzz testing techniques primarily rely on system call templates to generate test cases.However, this approach suffers from significant drawbacks, including high manual effort, poor adaptability to evolving vulnerabilities, and limited test case effectiveness.Given the increasing demand for robust security in Linux-based systems, more intelligent and automated fuzz testing methods are essential for improving vulnerability detection efficiency.To address these issues, this paper combines deep learning technology and fuzz testing to design a framework called DeepLiFF (Deep Learning-based Linux Kernel Fuzzing Framework) for vulnerability discovery in the Linux kernel.We propose a test case generation algorithm based on generative adversarial networks (GANs), which extracts features from high-quality test cases and trains models using GANs to generate even higher-quality test cases.DeepLiFuzzer has broad applications in operating system security auditing, vulnerability detection, and automated security testing.Experimental results show that, compared to existing tools such as syzkaller and Healer, DeepLiFuzzer improves code branch coverage by 25.49% and 7.03%, respectively, and increases the number of discovered bugs by 3.6 times and 43.75%, demonstrating its effectiveness in enhancing Linux kernel fuzz testing.