Date of Award

3-22-2018

Document Type

Thesis

Degree Name

Master of Science in Electrical Engineering

Department

Department of Electrical and Computer Engineering

First Advisor

Timothy Carbino, PhD.

Abstract

A Fuzz test is an approach used to discover vulnerabilities by intentionally sending invalid inputs to a system for the purpose of triggering some type of fault or unintended effect that renders the system vulnerable to an exploit. Fuzz testing is an important cyber-testing technique used to find and fix vulnerabilities before they are exploited. The fuzzing of military data links presents a particular challenge because existing fuzzing tools cannot be easily applied to these systems. As a result, the tools and techniques used to fuzz these links vary widely in sophistication and effectiveness. Because of the infinite, or nearly infinite, number of possible fuzzed messages that can be sent on a military data link, measuring the coverage of a fuzz test is not straightforward. This thesis proposes an understandable and meaningful metric for protocol fuzz testing called ExCov. This metric computes the coverage of a fuzz test set from a probabilistic model of vulnerability occurrence and defines coverage as the expected percent of existing vulnerabilities discovered by a set of test cases. This metric enables the acquisitions community to more succinctly write weapons system requirements for cyber security. Furthermore, it quantifies the number of faults and vulnerabilities that are expected to be found by a set of test cases, which provides decision makers with valuable information to make more informed choices on whether or not to perform additional testing. As a result, industry will be better equipped to determine cost and effort when performing cyber vulnerability testing. In addition, industry will also be able to more concretely represent the results of the cyber testing they perform. ExCov was implemented in a suite of tools called ExFuzz, and these tools were used to compare and contrast military data link fuzz testing techniques that are in use today. By assessing these current methods using the ExCov metric, optimal bit flip probabilities for the mutative fuzzing of three custom protocols was found. A generative fuzzer was also built based on the metric and was shown to outperform mutative and manual generation strategies in nearly every case.

AFIT Designator

AFIT-ENG-MS-18-M-063

DTIC Accession Number

AD1056216

Download

Share

COinS