Smart fuzzing

Abstract

Fuzzing is an automated form of black-box testing where (semi-)random input is passed to a program under examination. This testing technique has been successful in finding security vulnerabilities and has attracted a large amount of research activities.<br />The challenge when building an effective fuzzing tool is to generate test cases that cover as much of the code as possi- ble. Otherwise, only a small fraction of the application's functionality might be exercised.<br />In this thesis, we present LiFT, a Lightweight Fuzzing Tool. LiFT incor- porates several novel techniques to select appropriate inputs that increase test coverage. First, our system leverages specifications that describe the expected program inputs. Second, our tool monitors the execution of the program under test and selects the most promising input candidate based on feedback from previous test runs. Finally, the fuzzer uses a simple but efficient form of data tracking to identify corner cases that are difficult to trigger with random inputs.