很多应用程序需要使用通用的或者自定义的网络协议来进行通信，而模糊测试作为目前发现软件漏洞的一种重要方法，测试网络协议时却并不方便和高效。本项目提出让协议的各方协同参与模糊测试的新方法，以减少测试前准备工作，并面向协同以及协议的特征重新设计了模糊测试的关键技术，以提高对协议进行模糊测试的效率。主要研究内容包括：将协同测试时各协同方的交互关系、交互内容结合进执行覆盖统计中以全面地记录执行行为；利用马尔可夫链对消息的执行覆盖进行建模，并采用进化算法中的适应值共享技术来保持消息的多样性，从而实现高效地面向协同地对交互消息进行更改；自动识别协议中可能存在的消息认证字段检查并使其失效，以使模糊测试能测试到认证检查后面的逻辑；对网络协议实现中比较容易出现的数据竞争、越权等漏洞设计专门的模糊测试方法。

Many applications are using common or custom network protocols to communicate with each other, and fuzz testing is an important method for discovering software vulnerabilities. However, it is not convenient and efficient to fuzz network protocols now. This project proposes a collaboration-based fuzzing method, where all the participants of the network protocol are fuzzed together, to reduce the preparation work required before, and redesigns the key techniques of fuzz testing based on the characteristics of collaboration and network protocol, to improve the efficiency of protocol fuzzing. The main research contents are as follows. First, we combine the interaction relationship and interaction message content into the execution coverage calculations to perform comprehensive execution behavior recording during the collaborative fuzz testing. Second, we use the Markov chain to model the execution coverage of the messages, and adopt the fitness sharing technique in evolutionary algorithms to maintain the diversity of messages, to achieve efficient collaboration-oriented message mutation. Third, we propose to automatically identify and disable the message authentication checks that may exist in the protocol to make the fuzzing can test codes after the checks. At last, we propose special fuzzing strategies for detecting data races and privilege violations, since they are common vulnerabilities in network protocol implementations.