Research On Active Vulnerability Mining Of Industrial Control Protocols Based On Generative Adversarial Learning

The frequency and sophistication of cyber-threats towards Industrial Control Systems(ICSs)is increasing.The Industrial Control Protocol(ICP)is the cornerstone of the communication of the ICS,hence ensuring the safety and security of ICPs is of great significance.Security developers come to realize the safety and security of ICSs cannot considered separately without the safety and security of ICPs,and they should be considered jointly.However,the industrial control network environments applicable to ICPs have strong diversities,which are difficult for testers to formulate a series of universal security rules.Fuzz testing(fuzzing),as one of the active vulnerability mining methods,has become the main method of detecting vulnerabilities in ICPs.Fuzzy testing technology is often used to mine vulnerabilities of network protocols.Traditional fuzzy testing techniques are of great significance to ensure the safety and security of ICPs,and many of them have specific tools.However,most traditional fuzzing methods consist in big measure of the specification of ICPs.Analyzing the specifications of ICPs and coding strategies on the basis of the specifications that apply to fuzzy testing techniques is a complex and tedious task.In this study,we propose an intelligent protocol fuzzing methodology based on improved Deep Convolution Generative Adversarial Network(DCGAN)to solve the aforementioned problems.Compared with traditional methods,our framework can generate massive fake but plausible test protocol messages automatically in a short time without protocol specifications.On this basis,we propose a new simple and smart sequence generation neural network framework,which is based on Improved Wasserstein GANs(WGAN-GP)and self-attention mechanism,to further improve the performance of the fuzzing model and increase the probability of triggering vulnerabilities.Compared with other deep learning works for fuzzing,our method is more parallelizable and requires significantly less time to train.Moreover,we put forward a series of performance metrics to evaluate different models in the field of fuzzing for ICPs.On the basis of the two aforementioned models,an automated and intelligent fuzzing framework for application,called Hex GANFuzzer,is designed.It can be applied to both public and private ICPs,and testers need not learn the message format of the special protocol artificially,which outperforms many previous works.Several typical ICPs,including Modbus,Ether CAT and MQTT,are applied to test the effectiveness and efficiency of our framework.Extensive experiments shows that our framework not only outperforms the existing traditional fuzzing methods and deep learning based fuzzing methods in convenience but also demonstrates significant improvements on test effectiveness and efficiency.