一般化之合作型模糊測試框架

Abstract

隨著近年來模糊測試技術的快速發展，目前已出現眾多針對不同問題所設計的模糊測試工具。每種工具皆有其優勢與限制，而對於一個特定目標程式而言，在未進行實驗並分析結果之前，往往難以判斷最適合的工具是何者。即使找出有效的工具，如何分配各自的資源亦是一項不易解決的問題。

合作型模糊測試（Collaborative Fuzzing）旨在讓多個模護測試工具彼此合作以提升效率。然而，現有的合作方案多僅著重於特定面向，例如運算資源分配或種子同步，且往往建立在固定策略之上。儘管這些方法有效，卻缺乏足夠的彈性來支援多樣化或未來新興的合作策略。

本研究提出一個通用的協同合作型模糊測試架構，並實作一套工具以驗證此架構。該架構抽象出合作機制中的核心元件，並提供模組化與可擴充的設計，方便整合不同策略與工具。

我們從通用性、可維護性、可擴充性與犧牲的效能等面向，對該工具進行評估。實驗結果顯示，我們的工具能簡化合作型模糊測試的實作流程，並降低整合新元件的開發負擔。我們期望此研究能促進合作型模糊測試領域的未來發展，加速新策略的原型設計與評估。

As fuzzing techniques have advanced significantly in recent years, a wide range of fuzzers has been developed to address different challenges in fuzzing. Each fuzzer has its own strengths and limitations. Given a target program, it is often difficult to determine the optimal fuzzer without running experiments and analyzing the results. Even when effective fuzzers are identified, determining how to allocate resources among them remains a non-trivial problem.

Collaborative fuzzing addresses this by enabling multiple fuzzers to work together. However, most existing approaches focus on specific aspects---such as resource allocation or seed synchronization---and are often designed around fixed strategies. While effective, these tools lack the flexibility to accommodate alternative or evolving collaboration strategies.

In this work, we propose a general framework for collaborative fuzzing and implement a tool based on it. Our framework abstracts the core components of collaboration and provides a modular, extensible structure for integrating diverse strategies.

We evaluate our tool in terms of generality, maintainability, extensibility, and performance overhead. Our results show that it simplifies the implementation of collaborative fuzzing and reduces the effort required to integrate new components. We hope this work will facilitate future research in collaborative fuzzing by making it easier to prototype and evaluate new strategies.