Trusted execution environments (TEEs) are widely used, and their kernel security has become a significant area of focus. Fuzzing, a powerful technique for detecting vulnerabilities in operating systems, has increasingly been applied to the security analysis of TEEs. However, conventional fuzzing tools cannot be directly used for TEE kernels due to their isolation. Coverage-guided fuzzers often discard test cases that trigger new states but cover the same code, which limits their effectiveness in discovering vulnerabilities. To address these challenges, we propose a state-aware fuzzing method specifically designed for TEE kernels. Initially, we develop a modeling and tracing approach to represent the program state through state-variable values, overcoming the limitations of coverage-guided fuzzers. Subsequently, we propose a new communication scheme to address the issues resulting from the isolation of TEEs. Additionally, new seed preservation and selection algorithms are put forward to better guide the fuzzer in exploring vulnerabilities. Finally, we employ the N-gram model to enhance the test case generation process and optimize the framework's performance. We have implemented a prototype called Trusty-Statefuzz and evaluated it on Fuchsia, our self-developed microkernel operating system Nebula, and the OP-TEE. The evaluation results show that Trusty-Statefuzz is effective at detecting both new code and vulnerabilities. Trusty-Statefuzz discovers nine unknown vulnerabilities and 23 known vulnerabilities. Additionally, it achieves 13% higher code coverage and 27% higher state coverage than the state-of-the-art fuzzer Syzkaller.
区块链技术凭借其独有的去中心化、不可篡改、可追溯等特点,为社会发展中的信任问题、存证问题、数据治理问题等提供了全新的解决思路.而智能合约作为区块链的核心支撑技术,通过编写去中心化应用将区块链的应用范围从单一数字货币领域扩展至其他泛金融领域.但随着智能合约在区块链中应用的不断发展,其面临的安全问题也正日益突出,因此对智能合约的安全漏洞问题进行研究尤为重要.首先介绍了整数溢出漏洞、重入攻击漏洞等11种智能合约安全漏洞问题及其防范策略;然后讨论了形式化验证、符号执行、模糊测试、污点分析4种漏洞检测方法及其对应的检测工具;最后在总结现有漏洞检测工作不足的基础上对未来研究方向进行了展望.
联邦学习作为一种分布式机器学习框架,可以在不泄露用户数据的前提下完成模型训练.然而,最近的攻击表明,在训练过程中仅仅保持数据的局部性并不能提供足够的隐私保障.因此,为了解决联邦学习训练过程中存在的隐私保护问题,提出了一种基于BERT的文本分类模型,该模型将差异隐私(DP)和联邦学习(FL)相结合,在联邦学习参数的传递过程中保证联邦模型训练过程免受推理攻击的影响.最终实验表明,提出的方法在能够保护隐私的同时仍可保证较高的模型准确率.
