Towards Automatically Extracting the Relational Model of an Operating System

Several recent projects have argued that users would benefit from having a relational interface to operating systems state. In this paper, we report on the progress of a tool to automate this data modeling process. While automating schema inference in the general case appears to be intractable, we argue that it is feasible in the limited domain of operating system. Our approach leverages static analysis on the tightly-controlled coding styles common in OS kernels and hybridized techniques from the programming languages and databases literature for inferring and propagating constraints. We use a simple static analysis to extract relation names, types, and arity from constrained operating system source code patterns. We then leverage LLMs and iterative refinement to identify constraints. Our prototype implementation automates the extraction of 10% of the operating system structure into the relational model.