Information leakage of multi-threaded programs

Abstract

Quantitative information flow is an important technique for measuring information leakage of a program. It is widely used in analyzing anonymity protocols and timing channels. One area of interest is to quantify the information leakage of multi-threaded programs, which has not been well-studied in prior work. In this paper, an automated trace-based approach is proposed to precisely quantify information leakage of shared-memory multi-threaded programs. The approach takes into account the effect of schedulers and leakage in intermediate states of executions. The programs are modeled by Markovian processes. Then, variants of information leakage, including expected, bounded time, maximum, and minimum leakages are measured. The validity of the approach is demonstrated by implementing it in a tool PRISM-Leak, which is built upon PRISM, a probabilistic model checker. Finally, two case studies are utilized to analyze and compare the approach against state-of-the-art leakage quantification tools.