Divak: Non-invasive Characterization of Out-of-Bounds Write Vulnerabilities

Abstract

Despite the high level of automation that fuzzing has brought into the vulnerability research process, the assessment of a discovered vulnerability's implications mostly requires human expertise and intuition. A promising approach to reduce such a manual effort is the automatic extraction of vulnerability characteristics that provide vital clues for exploitability. In this work, we focus on out-of-bounds write vulnerabilities and investigate how to automatically distill the set of source codelevel objects affected by such unintended writes. As this poses unique challenges with regard to the invasiveness of the analysis methods, we propose a novel approach that enables monitoring a compiled program for spatial memory safety violations without the need for heavy instrumentation. We implement Divak, a prototype of our design, and we evaluate it on both benchmarks and real-world vulnerabilities, showing that its detection and characterization capabilities outperform instrumentationbased tools in several scenarios, at the cost of an increased overhead.