Fuzzing@Home: Distributed Fuzzing on Untrusted Heterogeneous Clients

Abstract

Fuzzing is a practical technique to automatically find vulnerabilities in software. It is well-suited to running at scale with distributed computing platforms thanks to its parallelizability. Therefore, individual researchers and companies typically setup fuzzing platforms on multiple servers and run fuzzers in parallel. However, as such resources are private, they suffer from financial and physical limits. In this paper, we propose Fuzzing@Home; the first public collaborative fuzzing network, based on heterogeneous machines owned by potentially untrusted users. Using our system, multiple organizations (or individuals) can easily collaborate to fuzz a software of common interest in an efficient way. One can participate and earn economic benefits if the fuzzing network is tied to a bug-bounty program, or simply donate spare computing power as a volunteer. If the network compensates collaborators, system fairness becomes an issue. In this light, we devise a system to make the fuzzing results verifiable and devise cheat detection techniques to ensure integrity and fairness in collaboration. In terms of performance, we devise a technique to effectively sync the global coverage state, hence minimizing the overhead for verifying computation results. Finally, to increase participation, Fuzzing@Home uses WebAssembly to run fuzzers inside the web browser engine, allowing anyone to instantly join a fuzzing network with a single click on their mobile phone, tablet, or any modern computing device. To evaluate our system, we bootstrapped Fuzzing@Home with 72 open-source projects and ran experimental fuzzing networks for 330 days with 826 collaborators as beta testers.