关键词: 跨平台, 漏洞关联, 特征选择, 神经网络, 二分图匹配

Abstract: Nowadays, most IOT vendors use the similar code to compile firmware for devices based on various CPU architectures. However, the prior vulnerability searching methods are limited to the same platform, which can’t be directly extended to the cross-platform case, and the cross-platform studies have just started. In this paper, we propose an algorithm to search vulnerabilities of firmware in a cross-platform model based on neural network and local calling structure matching. Firstly we extract the selected compared features from the call graphs, the basic attributes and the control flow graphs of the two compared functions as the input of the neural network, and gain the calculated results. Then we match the call sub-graphs of the compared functions with the results of the previous step as weight to improve the accuracy. The experimental results on the open source code OpenSSL demonstrate our method has better performance than the prior cross-platform vulnerability searching method with the Top1 increasing from 32.1% to 76.49% in the searching pattern from ARM to MIPS. The searching ranks of the common five vulnerabilities in OpenSSL are all No.1 rank. Moreover, we search the common four vulnerabilities in the firmware of the 372 types of D-Link routers and the results show good performance too.

Key words: cross-platform, vulnerability search, feature selection, neural network, bipartite matching