Fuzzing is arguably the most practical approach for detecting security bugs in software, but a non-trivial extent of efforts is required for its adoption. To be effective, high-quality fuzz drivers should be first formulated with a proper sequence of APIs that can exhaustively explore the program states. To alleviate this burden, existing solutions attempt to generate fuzz drivers either by inferring the valid sequences of APIs from the consumer code (i.e., actual uses of APIs) or by directly extracting them from sample executions. Unfortunately, all existing approaches suffer from a common problem: the observed API sequences, either statically inferred or dynamically monitored, are intermingled with custom application logics. However, we observed that the unit tests are carefully crafted by the actual designer of the APIs to validate their proper usages, and importantly, it is a common practice to write the unit tests during their development (e.g., over 70% of popular GitHub projects).In this paper, we propose, UTopia, an open-source tool and analysis algorithm that can automatically synthesize effective fuzz drivers from existing unit tests with near-zero human involvement. To demonstrate its effectiveness, we applied UTopia to 55 open-source project libraries, including Tizen and Node.js, and automatically generated 5K fuzz drivers from 8K eligible unit tests. In addition, we executed the generated fuzzers for approximately 5 million per-core hours and discovered 123 bugs. More importantly, 2.4K of the generated fuzz drivers were adopted to the continuous integration process of the Tizen project, indicating the quality of the synthesized fuzz driver. The proposed tool and results are publicly available and maintained for a broader adoption among both researchers and practitioners.

https://ieeexplore.ieee.org/document/10179394