Model-Based Fuzzing for Finding Kernel Vulnerabilities
Abstract
Kernel vulnerabilities are significant threats to computer security as attackers use them to obtain unauthorized root privilege and bypass security mitigations. In this project, we extend IMF, a state-of-the-art kernel fuzzing technique on macOS, to find vulnerabilities on Windows kernel. Unlike other OSes, Windows has numerous undocumented system calls, which make it difficult to generate a valid sequence of system calls for fuzzing. We propose a novel way to analyze Windows kernel binaries to figure out API specifications of undocumented system calls. We then use the inferred API specifications to generate a C program that automatically fuzzes Windows kernel by calling a randomly generated sequence of system calls.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 23, 2020
- Accession Number
- AD1095728
Entities
People
- Sang K. Cha
Organizations
- KAIST