CAREER: New Abstractions for Sensitive Data Management in Modern Operating Systems

现代操作系统（OS）中数据存储的发展为细粒度数据保护带来了挑战。虽然传统操作系统提供简单的，相对较低级别的数据抽象-文件和目录-现代操作系统，包括Android和iOS，嵌入更高级别的抽象，如关系数据库和对象关系模型。新的抽象使文件结构和访问模式复杂化，极大地挑战了现有的保护系统，如加密文件系统，可否认文件系统，防病毒和异常检测器，这些系统落后于时代，继续在文件级别运行。在这个项目中，我们正在研究新的数据保护抽象，更好地适应现代操作系统。一个示例是逻辑数据对象（LDO）。LDO对应于特定于应用程序的资源（如电子邮件、文档或银行账户），并且包括与之相关的所有数据，无论其如何或在何处持久化（例如，数据库中的行、对象关系模型中的对象、文件系统中的文件等）。保护系统使用LDO来获取关于数据的丰富语义，以改进其有效性。使用LDO，我们正在构建HideIt，这是一个细粒度的对象隐藏系统，允许用户通过其未修改的应用程序的熟悉UI选择任意对象-例如个人电子邮件，文档，银行帐户-并隐藏或取消隐藏它们。通过创建新的、方便的保护抽象，并向学生和更广泛的社区传授这些知识，我们希望推广一种负责任的数据管理方法，让用户谨慎地管理他们的数据，最大限度地减少其受到攻击的风险。

The evolution of data storage in modern operating systems (OSes) brings challenges for fine-grained data protection. While traditional OSes offer simple, relatively low-level data abstractions -- files and directories -- modern operating systems, including Android and iOS, embed much higher-level abstractions, such as relational databases and object-relational models. The new abstractions complicate file structures and access patterns, greatly challenging existing protection systems, such as encrypted file systems, deniable file systems, antiviruses, and anomaly detectors, which, fallen behind the times, continue to operate at file level.In this project, we are investigating new data protection abstractions that are better attuned to modern OSes. One example is a logical data object (LDO). An LDO corresponds to an application-specific resource -- such as an email, a document, or a bank account -- and includes all the data related to it, no matter how or where it is persisted (e.g., rows in databases, objects in object-relational models, files in the file system, etc.). Protection systems use LDOs to acquire rich semantics about the data to refine their effectiveness. Using LDOs, we are building HideIt, a fine-grained object hiding system that lets users select, through the familiar UIs of their unmodified applications, arbitrary objects -- such as individual emails, documents, bank accounts -- and hide or unhide them. By creating new, convenient protection abstractions, and teaching students and the broader community about them, we hope to promote a responsible approach to data management, in which users manage their data carefully, minimizing its exposure to attacks.