Title:Corpus Christi: Establishing Replicability when Sharing the Bread is Not Allowed

Abstract:In this paper, we provide practical tools to improve the scientific soundness of firmware corpora beyond the state of the art. We identify binary analysis challenges that significantly impact corpus creation. We use them to derive a framework of key corpus requirements that nurture the scientific goals of replicability and representativeness. We apply the framework to 44 top tier papers and collect 704 data points to show that there is currently no common ground on corpus creation. We discover in otherwise excellent work, that incomplete documentation and inflated corpus sizes blur visions on representativeness and hinder replicability. Our results show that the strict framework provides useful and practical guidelines that can identify miniscule step stones in corpus creation with significant impact on soundness.
Finally, we show that it is possible to meet all requirements: We provide a new corpus called LFwC. It is designed for large-scale static analyses on Linux-based firmware and consists of 10,913 high-quality images, covering 2,365 network appliances. We share rich meta data and scripts for replicability with the community. We verify unpacking, perform deduplication, identify contents, and provide bug ground truth. We identify ISAs and Linux kernels. All samples can be unpacked with the open source tool FACT.