RAProducer: Efficiently Diagnose and Reproduce Data Race Bugs for Binaries via Trace Analysis

ISSTA 2021,

Ming Yuan , Yeseop Lee , Chao Zhang , Yun Li , Yan Cai , Bodong Zhao .


A growing number of bugs have been reported by vulnerability discovery solutions. Among them, some bugs are hard to diagnose or reproduce, including data race bugs caused by thread interleavings. Few solutions are able to well address this issue, due to the huge space of interleavings to explore. What’s worse, in security analysis scenarios, analysts usually have no access to the source code of target programs and have troubles in comprehending them.

In this paper, we propose a general solution RAProducer to efficiently diagnose and reproduce data race bugs, for both user-land binary programs and kernels without source code. The efficiency of RAProducer is achieved by analyzing the execution trace of the given PoC (proof-of-concept) sample to recognize race- and bug-related elements (including locks and shared variables), which greatly facilitate narrowing down the huge search space of data race spots and thread interleavings. We have implemented a prototype of RAProducer and evaluated it on 7 kernel and 10 user-land data race bugs. Results show that RAProducer successfully reproduces all these bugs. More importantly, it enables us to diagnose 2 extra real-world bugs which were left unconfirmed for a long time. It is also efficient as it reduces candidate data race spots of each bug to a small set, and narrows down the thread interleaving greatly. RAProducer is also more effective in reproducing real-world data race bugs than other state-of-the-art solutions.