Modelling Access Propagation in Dynamic Systems

Abstract

Access control is a critical feature of many systems, including networks of services, processes within a computer, and objects within a running process. The security consequences of a particular architecture or access control policy are often difficult to determine, especially where some components are not under our control, where components are created dynamically, or where access policies are updated dynamically. The SERSCIS Access Modeller (SAM) takes a model of a system and explores how access can propagate through it. It can both prove defined safety properties and discover unwanted properties. By defining expected behaviours, recording the results as a baseline, and then introducing untrusted actors, SAM can discover a wide variety of design flaws. SAM is designed to handle dynamic systems (i.e., at runtime, new objects are created and access policies modified) and systems where some objects are not trusted. It extends previous approaches such as Scollar and Authodox to provide a programmer-friendly syntax for specifying behaviour, and allows modelling of services with mutually suspicious clients. Taking the Confused Deputy example from Authodox we show that SAM detects the attack automatically; using a web-based backup service, we show how to model RBAC systems, detecting a missing validation check; and using a proxy certificate system, we show how to extend it to model new access mechanisms. On discovering that a library fails to follow an RFC precisely, we re-evaluate our existing models under the new assumption and discover that the proxy certificate design is not safe with this library.