TrustGraph: Trusted Graphics Subsystem for High Assurance Systems

Abstract

High assurance MILS and MLS systems require strict limitation of the interactions between different security compartments based on a security policy. Virtualization can be used to provide a high degree of separation in such systems. Even with perfect isolation, however, the I/O devices are shared between different security compartments. Among the I/O controllers, the graphics subsystem is the largest and the most complex. This paper describes the design and implementation of TrustGraph, a trusted graphics subsystem for high assurance systems. First, we explain the threats and attacks possible against an unsecured graphics subsystem. We then describe the design of TrustGraph, the security principles it is built upon, and its implementation. Finally, we verify our implementation through different levels of verification which include functionality testing for simple operations, attack testing for security mechanisms, and formal verification for the critical components of the implementation. An analysis of the graphics API covert channel attack is presented, its channel capacity is measured, and the capacity is reduced using the idea of fuzzy time.

Constraints on autonomous use of standard GPU components for asynchronous observations and intrusion detection

“…This renders their usage difficult in terms of host intrusion detection. The necessity of an isolated graphics system was mentioned by Okhravi and Nicol [28]. They describe the design and implementation of a trusted graphics subsystem for high assurance systems.…”