Using a loadtime metaobject protocol to enforce access control policies upon user-level compiled code
This thesis evaluates the use of a loadtime metaobject protocol as a practical mechanism for enforcing access control policies upon applications distributed as user-level compiled code. Enforcing access control policies upon user-level compiled code is necessary because there are many situations where users are vulnerable to security breaches because they download and run potentially untrustworthy applications provided in the form of user-level compiled code. These applications might be distributed applications so access control for both local and distributed resources is required. Examples of potentially untrustworthy applications are Browser plug-ins, software patches, new applications, or Internet computing applications such as SETI@home. Even applications from trusted sources might be malicious or simply contain bugs that can be exploited by attackers so access control policies must be imposed to prevent the misuse of resources. Additionally, system administrators might wish to enforce access control policies upon these applications to ensure that users use them in accordance with local security requirements. Unfortunately, applications developed externally may not include the necessary enforcement code to allow the specification of organisation-specific access control policies. Operating system security mechanisms are too coarse-grained to enforce security policies on applications implemented as user-level code. Mechanisms that control access to both user-level and operating system-level resources are required for access control policies but operating system mechanisms only focus on controlling access to system-level objects. Conventional object-oriented software engineering can be used to use existing security architectures to enforce access control on user-level resources as well as system-resources. Common techniques are to insert enforcement within libraries or applications, use inheritance and proxies. However, these all provide a poor separation of concerns and cannot be used with compiled code. In-lined reference monitors provide a good separation of concerns and meet criteria for good security engineering. They use object code rewriting to control access to both userlevel and system-level objects by in-lining reference monitor code into user-level compiled code. However, their focus is upon replacing existing security architectures and current implementations do not address distributed access control policies. Another approach that does provide a good separation of concerns and allows reuse of existing security architectures are metaobject protocols. These allow constrained changes to be made to the semantics of code and therefore can be used to implement access control policies for both local and distributed resources. Loadtime metaobject protocols allow metaobject protocols to be used with compiled code because they rewrite base level classes and insert meta-level interceptions. However, these have not been demonstrated to meet requirements for good security engineering such as complete mediation. Also current implementations do not provide distributed access control. This thesis implements a loadtime metaobject protocol for the Java programming language. The design of the metaobject protocol specifically addresses separation of concerns, least privilege, complete mediation and economy of mechanism. The implementation of the metaobject protocol, called Kava, has been evaluated by implementing diverse security policies in two case studies involving third-party standalone and distributed applications. These case studies are used as the basis of inferences about general suitability of using loadtime reflection for enforcing access control policies upon user-level compiled code.