A scalable approach for vulnerability discovery based on security patches

Software vulnerability has long been considered an important threat to the system safety. A vulnerability often gets reproduced due to the frequent code reuse by programmers. Security patches are often not propagated to all code clones, however they could be leveraged to discover unknown vulnerabilities. Static auditing approaches are frequently proposed to scan code for security flaws, unfortunately, they often generate too many false positives. While dynamic execution analysis can precisely report vulnerabilities, they are in effective in path exploration which limits them to scale to large programs. In this paper, we propose a scalable approach to discover vulnerabilities in real world programs based on released security patches. We use a fast and scalable syntax-based way to find code clones and then, we verify the code clones using concolic testing to dramatically decrease the false positives. Besides, we mitigate the path explosion problem by backward data tracing in concolic execution. We conducted experiments with real world open source projects (Linux Ubuntu OS distributions and program packages) and we reported 7 real vulnerabilities out of 63 code clones found in Ubuntu 14.04 LTS. In one step further, we have confirmed more code clone vulnerabilities in various versions of programs including Apache and Rsyslog. Meanwhile, we also tested the effectiveness of vulnerability verification with test cases from Juliet Test Suite. The result showed that our verification method achieved 98% accuracy with 0 false positives.