Adversarial Logic

Julien Vanegue

Abstract: We introduce Adversarial Logic, an extension of Incorrectness Logic [1] with an explicit Dolev-Yao [2] adversary to statically analyze the severity of security vulnerabilities in the under-approximate setting. Adversarial logic is built on the ability to separate logical facts known to the adversary from facts solely known to the program under analysis. This flavor of program incorrectness can be used to analyze software in which error behavior occurs at deeper levels of interaction between the program and its environment, such as subtle cases of information disclosure requiring multiple program executions to be uncovered. We introduce the Oscillating Bit Protocol, an example algorithm where such a vulnerability can be detected using adversarial logic while remaining elusive to other frameworks. We define a flavor of symbolic execution in which the adversary guides the introduction of symbolic variables and the checking of attack assertions. Additionally, we introduce equivalence testing, an under-approximate version of program equivalence only proven on specific program paths and used to extract differences between comparable implementations. We provide a denotational semantics for adversarial logic and prove its soundness, thereby guaranteeing that extracted attack paths are true positives.

External IDs:dblp:conf/sas/Vanegue22