Verifying Non-blocking Data Structures with Manual Memory Management

Abstract

Verification of concurrent data structures is one of the most challenging tasks in software verification. The topic has received considerable attention over the course of the last decade. Nevertheless, human-driven techniques remain cumbersome and notoriously difficult while automated approaches suffer from limited applicability. This is particularly true in the absence of garbage collection. The intricacy of non-blocking manual memory management (manual memory reclamation) paired with the complexity of concurrent data structures has so far made automated verification prohibitive. We tackle the challenge of automated verification of non-blocking data structures which manually manage their memory. To that end, we contribute several insights that greatly simplify the verification task. The guiding theme of those simplifications are semantic reductions. We show that the verification of a data structure's complicated target semantics can be conducted in a simpler and smaller semantics which is more amenable to automatic techniques. Some of our reductions rely on good conduct properties of the data structure. The properties we use are derived from practice, for instance, by exploiting common programming patterns. Furthermore, we also show how to automatically check for those properties under the smaller semantics. The main contributions are: (i) A compositional verification approach that verifies the memory management and the data structure separately. (ii) A notion of weak ownership that applies when memory is reclaimed and reused, bridging the gap between garbage collection and manual memory management (iii) A notion of pointer races and harmful ABAs the absence of which ensures that the memory management does not influence the data structure, i.e., it behaves as if executed under garbage collection. Notably, we show that a check for pointer races and harmful ABAs only needs to consider executions where at most a single address is reused. (iv) A notion of strong pointer races the absence of which entails the absence of ordinary pointer races and harmful ABAs. We devise a highly-efficient type check for strong pointer races. After a successful type check, the actual verification can be performed under garbage collection using an off-the-shelf verifier. (v) Experimental evaluations of the aforementioned contributions. We are the first to fully automatically verify practical non-blocking data structures with manual memory management.