Towards Reliable and Scalable Linux Kernel CVE Attribution in Automated Static Firmware Analyses (2209.05217v1)

Abstract: In vulnerability assessments, software component-based CVE attribution is a common method to identify possibly vulnerable systems at scale. However, such version-centric approaches yield high false-positive rates for binary distributed Linux kernels in firmware images. Not filtering included vulnerable components is a reason for unreliable matching, as heterogeneous hardware properties, modularity, and numerous development streams result in a plethora of vendor-customized builds. To make a step towards increased result reliability while retaining scalability of the analysis method, we enrich version-based CVE matching with kernel-specific build data from binary images using automated static firmware analysis. We open source an attribution pipeline that gathers kernel configuration and target architecture to dry build the present kernel version and filter CVEs based on affected file references in record descriptions. In a case study with 127 router firmware images, we show that in comparison to naive version matching, our approach identifies 68% of all version CVE matches as false-positives and reliably removes them from the result set. For 12% of all matches it provides additional evidence of issue applicability. For 19.4%, our approach does not improve reliability because required file references in CVEs are missing.