Capability-based access control for cyber physical systems

Abstract

Cyber Physical Systems (CPS) couple digital systems with the physical environment, creating technical, usability, and economic security challenges beyond those of information systems. Their distributed and hierarchical nature, real-time and safety-critical requirements, and limited resources create new vulnerability classes and severely constrain the security solution space. This dissertation explores these challenges, focusing on Industrial Control Systems (ICS), but demonstrating broader applicability to the whole domain.

We begin by systematising the usability and economic challenges to secure ICS. We fingerprint and track more than 10\,000 Internet-connected devices over four years and show the population is growing, continuously-connected, and unpatched. We then explore adversarial interest in this vulnerable population. We track 150\,000 botnet hosts, sift 70 million underground forum posts, and perform the largest ICS honeypot study to date to demonstrate that the cybercrime community has little competence or interest in the domain. We show that the current heterogeneity, cost, and level of expertise required for large-scale attacks on ICS are economic deterrents when targets in the IoT domain are available.

The ICS landscape is changing, however, and we demonstrate the imminent convergence with the IoT domain as inexpensive hardware, commodity operating Cyber Physical Systems (CPS) couple digital systems with the physical environment, creating technical, usability, and economic security challenges beyond those of information systems. Their distributed and hierarchical nature, real-time and safety-critical requirements, and limited resources create new vulnerability classes and severely constrain the security solution space. This dissertation explores these challenges, focusing on Industrial Control Systems (ICS), but demonstrating broader applicability to the whole domain.

We begin by systematising the usability and economic challenges to secure ICS. We fingerprint and track more than 10,000 Internet-connected devices over four years and show the population is growing, continuously-connected, and unpatched. We then explore adversarial interest in this vulnerable population. We track 150,000 botnet hosts, sift 70 million underground forum posts, and perform the largest ICS honeypot study to date to demonstrate that the cybercrime community has little competence or interest in the domain. We show that the current heterogeneity, cost, and level of expertise required for large-scale attacks on ICS are economic deterrents when targets in the IoT domain are available.

The ICS landscape is changing, however, and we demonstrate the imminent convergence with the IoT domain as inexpensive hardware, commodity operating systems, and wireless connectivity become standard. Industry's security solution is boundary defence, pushing privilege to firewalls and anomaly detectors; however, this propagates rather than minimises privilege and leaves the hierarchy vulnerable to a single boundary compromise.

In contrast, we propose, implement, and evaluate a security architecture based on distributed capabilities. Specifically, we show that object capabilities, representing physical resources, can be constructed, delegated, and used anywhere in a distributed CPS by composing hardware-enforced architectural capabilities and cryptographic network tokens. Our architecture provides defence-in-depth, minimising privilege at every level of the CPS hierarchy, and both supports and adds integrity protection to legacy CPS protocols. We implement distributed capabilities in robotics and ICS demonstrators, and we show that our architecture adds negligible overhead to realistic integrations and can be implemented without significant modification to existing source code. In contrast, we propose, implement, and evaluate a security architecture based on distributed capabilities. Specifically, we show that object capabilities, representing physical resources, can be constructed, delegated, and used anywhere in a distributed CPS by composing hardware-enforced architectural capabilities and cryptographic network tokens. Our architecture provides defence-in-depth, minimising privilege at every level of the CPS hierarchy, and both supports and adds integrity protection to legacy CPS protocols. We implement distributed capabilities in robotics and ICS demonstrators, and we show that our architecture adds negligible overhead to realistic integrations and can be implemented without significant modification to existing source code.