Automating Vulnerability Discovery in Critical Applications
Created September 2017
Vulnerabilities are too pervasive in software-based systems to find them all manually. CERT researchers develop automated tools that discover and mitigate software vulnerabilities and transfer them to security researchers, procurement specialists, and software vendors.
Automating Vulnerability Discovery
There are many places in the software lifecycle where software vulnerabilities can be discovered and mitigated. We develop new automated tools and techniques and put them in the hands of security researchers, procurement specialists, and software vendors to help them improve and evaluate the security of the software ecosystem used by the U.S. Department of Defense (DoD) and the U.S. government (USG).
These tools and techniques are intended to be used by vendors during software development. Still, vulnerabilities exist in delivered software. In a project called “Automating Vulnerability Discovery,” we focused on the ability to automatically discover vulnerabilities after software is developed and shipped, without assistance from the software vendor. This approach makes the problem more challenging because we must analyze programs at the binary level. But this requirement is critical in practice because many vendors are reluctant to share access to their products’ source code, which they regard as sensitive and proprietary.
There Are Too Many Vulnerabilities Not to Automate
Vulnerabilities are pervasive in software-based systems and protocols, both in traditional IT networks and in those that support critical U.S. infrastructure. Software is written and updated frequently, and there are not enough human analysts to keep pace with the speed of production. Automation is the only way to keep up.
Led by Dr. Edward Schwarz, the Vulnerability Discovery project aims to reduce vulnerabilities in critical DoD and USG systems by improving techniques for automated vulnerability discovery. If vendors adopt our techniques in their software development processes, the DoD will acquire software applications that are hardened—and more secure—before and after they are deployed into the DoD infrastructure.
Our Approach: Automate Discovery, Prioritize Results, Address the Most Important Vulnerabilities First
Our approach to vulnerability discovery builds on an extensive body of work at Carnegie Mellon University and the SEI's CERT Division. This work includes the following techniques:
- Black-box mutational fuzzing—Mutational fuzzing randomly introduces mutations into starting inputs called seed files. It then runs these modified inputs on the program to see if it causes a crash. Fuzzing is one of the most effective vulnerability discovery techniques in practice, and the SEI maintains its own black-box fuzzer, the CERT Basic Fuzzing Framework.
- Vulnerability uniqueness determination—A primary challenge in vulnerability discovery is understanding which vulnerability triggered a crash and how serious that vulnerability is. For example, in one of our recent experiments, a major vendor's software produced over 40,000 exploitable crashes, but these crashes were caused by only a few vulnerabilities. We use a new technique that applies software patches to identify the scope of a vulnerability, which allows us to precisely count the number of vulnerabilities that a particular vulnerability discovery technique identifies.
Combining Fuzzing with Concolic Execution
ForAllSecure is a start-up company on a mission to make software safer. The ForAllSecure team builds software security tools for developers, enterprises, and end users that automatically find vulnerabilities in software at the binary level, using research transitioned from the Carnegie Mellon University CyLab.
ForAllSecure is one of the world’s leading experts on binary concolic execution and is widely known for its Mayhem concolic executor, which recently won DARPA’s Cyber Grand Challenge competition. Concolic execution is a powerful form of symbolic program analysis that describes program executions as logical formulas and solves them to trigger and test new fragments of program code. Concolic execution is known for its ability to trigger code that is difficult to reach, allowing it to find vulnerabilities that other techniques such as fuzzing might miss. This power comes with a price, however, as concolic execution tends to be slow and has not scaled to large programs.
By joining our expertise in fuzzing with ForAllSecure’s expertise in concolic execution, we have been collaborating on a new technique that brings the best of both fuzzing and concolic execution. Fuzzing brings scalability, speed, and the ability to discover vulnerabilities in large, complex programs. Concolic execution allows analysts to omit seed files, making the process of vulnerability discovery easier than ever. It also allows software analysts to test and detect vulnerabilities in code that is difficult for a fuzzer to reach.
Software and Tools
October 22, 2015 Poster
A poster for the 2015 Research Review presentation on discovering vulnerabilities in software.read
August 20, 2014 Conference Paper
Alexandre Rebert (Carnegie Mellon University and ForAllSecure, Inc.)Sang Kil Cha (Carnegie Mellon University)Thanassis Avgerinos (Carnegie Mellon University)
In this paper, we focus on how to mathematically formulate and reason about one critical aspect in fuzzing: how best to pick seed files to maximize the total number of bugs found during a fuzz campaign.read
August 01, 2012 Technical Note
In this report, the authors describe an algorithm for automating the selection of seed files and other parameters used in black-box fuzz testing.read