IPv6 Prefix Target Generation through Pattern and Distribution Learning using Vision-Transformer and Guided-Diffusion

IPv6 network scanning is an essential active technology for network management. Existing target generation algorithms enable the usability of IPv6 scanning technology across the whole network. Under prefix target generation, they are affected by non-target prefix data, resulting in severe performance degradation. Prefix target generation mainly faces the challenges of complex and diverse addressing patterns and huge 64-bit prefix space. In this paper, we first propose a prefix target generation algorithm - 6PTG to address these challenges. 6PTG mines the knowledge of addressing patterns in massive unlabeled addresses based on unsupervised Vision Transformer Autoencoder(ViT-AE) and introduces it into the generation stage. Based on Guided-Diffusion, 6PTG captures the mapping between prefixes and active address distributions in a unified model to avoid the need to train the model separately for each prefix. In natural network environments, the 6PTG achieves an average hit rate of 52 % and a generation rate of 24 % under prefixes of varying active scales. Compared to transition-adapted whole-network-level algorithms, it achieves a 2.1x and 7.6x improvement on the existing public dataset IPv6 Hitlist. 6PTG effectively fills the gap in relevant research.