协议(科学)
计算机科学
模糊测试
人工智能
专家系统
钥匙(锁)
数据挖掘
通信协议
融合
组分(热力学)
会话(web分析)
作者
Yu Zhang,Kai-Long Zhu,Qian Chen,Zixiong Li,Yuliang Lu,Zhaoqi Hu,Jie Peng
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-05-13
卷期号:: 3133-3138
标识
DOI:10.1109/cscwd68734.2026.11582580
摘要
The effectiveness of stateful network protocol fuzzing is fundamentally constrained by the diversity-validity trade-off in seed generation. While seeds derived from real-world traffic lack diversity and limit state-space exploration, those generated from RFC specifications often exhibit low validity, leading to early parser rejection. To resolve this fundamental challenge, we propose FusionSeed, a novel framework that systematically fuses knowledge from both RFC documents and network traffic to generate high-quality seeds. FusionSeed integrates three key innovations: dual-source knowledge extraction, Probabilistic Context-Sensitive Grammar-guided LLM generation, and an iterative refinement pipeline. Evaluations demonstrate that FusionSeed outperforms state-of-the-art baselines, achieving an average 18.30% higher code coverage and 21.51% more unique crashes. Our approach establishes a new paradigm for generating high-validity, high-diversity seeds for stateful protocol fuzzing.
科研通智能强力驱动
Strongly Powered by AbleSci AI