VIDEZZO: Dependency-aware Virtual Device Fuzzing
Qiang Liu,Flavio Toffalini,Yajin Zhou,Mathias Payer
DOI: https://doi.org/10.1109/sp46215.2023.10179354
2023-01-01
Abstract:A virtual machine interacts with its host environment through virtual devices, driven by virtual device messages, e.g., I/O operations. By issuing crafted messages, an adversary can exploit a vulnerability in a virtual device to escape the virtual machine, gaining host access. Even though hundreds of bugs in virtual devices have been discovered, coverage-based virtual device fuzzers hardly consider intra-message dependencies (a field in a virtual device message may be dependent on another field) and inter-message dependencies (a message may depend on a previously issued message), thus resulting in limited scalability or efficiency. VIDEZZO, our new dependency-aware fuzzing framework for virtual devices, overcomes the limitations of existing virtual device fuzzers by annotating intra-message dependencies with a lightweight grammar, and by self-learning inter-message dependencies with new mutation rules. Specifically, VIDEZZO annotates message dependencies and applies three categories of message mutators. This approach avoids heavy manual effort to analyze specifications and speeds up the slow exploration by satisfying dependencies, resulting in a scalable and efficient fuzzer that boosts bug discovery in virtual devices. In our evaluation, VIDEZZO covers two hypervisors, four architectures, five device categories, and 28 virtual devices, and reaches competitive coverage faster. Moreover, VIDEZZO successfully finds 24 existing and 28 new bugs across diverse bug types. We are actively engaging with the community with 7 of our submitted patches already accepted.