Matching machine-code functions in executables within one product line via bioinformatic sequence alignment

In this paper we evaluate whether different executables from the same software product line have similar sequences of machine-code functions. We provide a method of creating matchings of machine-code functions using alignment techniques known from bioinformatics. We map, per function, vectors of code metrics to symbols from an alphabet using machine learning techniques, and construct sequence alignments using off-the-shelf alignment tools. Our evaluation of alignments of glibc versions, musl optimizations, different RedBoot platforms and architectures, and the Linux kernel shows that the above statement holds in all cases except for differing architectures. Our method can therefore be used to match functions in executables for most variations within one product line.