An Industrial Network Protocol Fuzzing Framework Based on Deep Adversarial Networks

Abstract

Fuzzing test is one of the major vulnerability methods in recent years and is also widely used for testing protocols. However, most industrial control network protocol norms are unknown and running in a isolateded environment, so it is difficult to directly apply reverse engineering to construct protocol syntax rules. In this paper, we propose an approach to automatically generate test cases by Deep learning to parse the protocol format and learn the grammar, using Sequence Generative Adversarial Networks (SeqGAN) to train a generative model on real protocol messages to learn the protocol grammar, and then generating spurious but plausible messages as fuzzing test cases. Based on this approach, we present a fuzzing framework to automatic test industrial control protocols. Compared to traditional methods, our approach does not rely on protocol specifications and can be applied to most of protocols. The final experimental data also shows that our framework can effectively learn the features of the target protocols and successfully trigger some exceptions.