Secure and efficient processing of outsourced data structures using trusted execution environments

In recent years, more and more companies make use of cloud computing; in other words, they outsource data storage and data processing to a third party, the cloud provider. From cloud computing, the companies expect, for example, cost reductions, fast deployment time, and improved security. However, security also presents a significant challenge as demonstrated by many cloud computing–related data breaches. Whether it is due to failing security measures, government interventions, or internal attackers, data leakages can have severe consequences, e.g., revenue loss, damage to brand reputation, and loss of intellectual property. A valid strategy to mitigate these consequences is data encryption during storage, transport, and processing. Nevertheless, the outsourced data processing should combine the following three properties: strong security, high efficiency, and arbitrary processing capabilities. Many approaches for outsourced data processing based purely on cryptography are available. For instance, encrypted storage of outsourced data, property-preserving encryption, fully homomorphic encryption, searchable encryption, and functional encryption. However, all of these approaches fail in at least one of the three mentioned properties. Besides approaches purely based on cryptography, some approaches use a trusted execution environment (TEE) to process data at a cloud provider. TEEs provide an isolated processing environment for user-defined code and data, i.e., the confidentiality and integrity of code and data processed in this environment are protected against other software and physical accesses. Additionally, TEEs promise efficient data processing. Various research papers use TEEs to protect objects at different levels of granularity. On the one end of the range, TEEs can protect entire (legacy) applications. This approach facilitates the development effort for protected applications as it requires only minor changes. However, the downsides of this approach are that the attack surface is large, it is difficult to capture the exact leakage, and it might not even be possible as the isolated environment of commercially available TEEs is limited. On the other end of the range, TEEs can protect individual, stateless operations, which are called from otherwise unchanged applications. This approach does not suffer from the problems stated before, but it leaks the (encrypted) result of each operation and the detailed control flow through the application. It is difficult to capture the leakage of this approach, because it depends on the processed operation and the operation’s location in the code. In this dissertation, we propose a trade-off between both approaches: the TEE-based processing of data structures. In this approach, otherwise unchanged applications call a TEE for self-contained data structure operations and receive encrypted results. We examine three data structures: TEE-protected B+-trees, TEE-protected database dictionaries, and TEE-protected file systems. Using these data structures, we design three secure and efficient systems: an outsourced system for index searches; an outsourced, dictionary-encoding–based, column-oriented, in-memory database supporting analytic queries on large datasets; and an outsourced system for group file sharing supporting large and dynamic groups. Due to our approach, the systems have a small attack surface, a low likelihood of security-relevant bugs, and a data owner can easily perform a (formal) code verification of the sensitive code. At the same time, we prevent low-level leakage of individual operation results. For all systems, we present a thorough security evaluation showing lower bounds of security. Additionally, we use prototype implementations to present upper bounds on performance. For our implementations, we use a widely available TEE that has a limited isolated environment—Intel Software Guard Extensions. By comparing our systems to related work, we show that they provide a favorable trade-off regarding security and efficiency.