Secure platforms for enforcing contextual access control

Aditi Gupta, Purdue University

Abstract

Advances in technology and wide scale deployment of networking enabled portable devices such as smartphones has made it possible to provide pervasive access to sensitive data to authorized individuals from any location. While this has certainly made data more accessible, it has also increased the risk of data theft as the data may be accessed from potentially unsafe locations in the presence of untrusted parties. The smartphones come with various embedded sensors that can provide rich contextual information such as sensing the presence of other users in a context. Frequent context profiling can also allow a mobile device to learn its surroundings and infer the familiarity and safety of a context. This can be used to further strengthen the access control policies enforced on a mobile device. Incorporating contextual factors into access control decisions requires that one must be able to trust the information provided by these context sensors. This requires that the underlying operating system and hardware be well protected against attacks from malicious adversaries. In this work, we explore how contextual factors can be leveraged to infer the safety of a context. We use a context profiling technique to gradually learn a context's profile, infer its familiarity and safety and then use this information in the enforcement of contextual access policies. While intuitive security configurations may be suitable for non-critical applications, other security-critical applications require a more rigorous definition and enforcement of contextual policies. We thus propose a formal model for proximity that allows one to define whether two users are in proximity in a given context and then extend the traditional RBAC model by incorporating these proximity constraints. Trusted enforcement of contextual access control requires that the underlying platform be secured against various attacks such as code reuse attacks. To mitigate these attacks, we propose a binary diversification approach that randomizes the target executable with every run. We also propose a defense framework based on control flow analysis that detects, diagnoses and responds to code reuse attacks in real time.

Degree

Ph.D.

Advisors

Bertino, Purdue University.

Subject Area

Computer science

COinS