面向HDF5格式预训练模型的模糊测试方法  

Fuzz Testing Framework for HDF5 Format Pre-trained Model

作　　者：严飞[1] 瞿铸枫 张立强[1] YAN Fei;QU Zhufeng;ZHANG Liqiang(Key Laboratory of Aerospace Information Security and Trusted Computing of Ministry of Education,School of Cyber Science and Engineering,Wuhan University,Wuhan 430072,China)

机构地区：[1]武汉大学国家网络安全学院空天信息安全与可信计算教育部重点实验室,湖北武汉430072

出　　处：《郑州大学学报（理学版）》2023年第1期1-7,共7页Journal of Zhengzhou University:Natural Science Edition

基　　金：国家自然科学基金项目(61272452);国家重点基础研究发展计划(973计划)项目(2014CB340601);湖北省重点研发计划项目(2020BAA003);苏州市前瞻性应用研究项目(SYG201845)。

摘　　要：使用模糊测试对HDF5文件格式的相关程序与工具集进行漏洞检测,并对模糊测试在HDF5输入上的性能优化方案进行研究。通过轻量级文件结构分析,精简模糊测试的确定性变异阶段,从而将模糊测试的注意力集中在更有价值的区域,减少无意义的变异与执行尝试次数;提出一系列HDF5文件格式敏感的变异策略,在模糊测试的随机变异阶段,使变异生成的输入更可能被程序的解析逻辑所接受,从而探索更深层代码。相比传统模糊测试框架,实现的原型框架HDFL可以保证极小的覆盖率与崩溃数量损耗,提高模糊测试的效率。The fuzz testing was used to detect the vulnerabilities of related programs and toolsets of the HDF5 file format.And the performance optimization strategies of fuzz testing on HDF5 input was studied.Through lightweight analysis of the file structure,the deterministic variation stage of fuzzy test was simplified,so as to focus the attention of fuzz testing on the more valuable areas,and to reduce the number of meaningless mutations and execution attempts.A series of HDF5 file format sensitive mutation strategies were proposed,which made the input generated by the mutation more likely to be accepted by the checking logic of the program during the havoc mutation stage of the fuzz testing,so as to explore deeper code.Compared with the traditional fuzz testing framework,the realized prototype framework HDFL could guarantee extremely small coverage and crash loss,and improve the efficiency of fuzz testing.

关 键 词：模糊测试 HDF5 深度学习模型 漏洞检测 

分 类 号：TN915.08[电子电信—通信与信息系统]