SIP Shaker: Software Integrity Protection Composition
This addresses the challenge of efficiently securing software against powerful adversaries, though it is incremental as it builds on existing protection measures.
The paper tackles the problem of composing software integrity protections against Man-At-The-End attackers by developing a framework that uses Integer Linear Programming to handle conflicts and optimize for security and performance, reducing overhead by approximately 39% and achieving a 5-fold decrease compared to state-of-the-art heuristics.
Man-At-The-End (MATE) attackers are almighty adversaries against whom there exists no silver-bullet countermeasure. To raise the bar, a wide range of protection measures were proposed in the literature each of which adds resilience against certain attacks on certain digital assets of a program. Intuitively, composing a set of protections (rather than applying just one of them) can mitigate a wider range of attacks and hence offer a higher level of security. Despite the potential benefits, very limited research has been done on the composition of protections. Naive compositions could lead to conflicts which, in turn, limit the application of protections, raise false alarms, and worse yet, yield corrupted binaries. More importantly, inadequate compositions of such protections are not tailored for the program at hand and thus the offered security and performance are sub-optimal. In this paper, we first lay out a set of generic constraints for a conflict-free composition of protections. Then, we develop a composition framework based on a defense graph in which nodes and edges capture protections, their relations, and constraints. The conflicts problem together with optimization requirements are then translated into a set of integer constraints. We then use Integer Linear Programming (ILP) to handle conflicts while optimizing for a higher security and lower overhead. To measure the overhead, we use a set of real-world programs (MiBench dataset and open source games). Our evaluation results indicate that our composition framework reduces the overhead by $\approx$ 39% while maximizing the coverage. Moreover, our approach yields a 5-fold decrease in overhead compared to state-of-the-art heuristics.