TL;DR
This paper introduces a novel framework that allows cloud storage providers to perform data reduction (compression and deduplication) on encrypted data while maintaining end-to-end confidentiality. This is achieved by a "Trusted Decrypter" (TD) component within the cloud provider that handles decryption, data reduction, and re-encryption, ensuring data remains secure even after client offboarding.
What Problem Does This Paper Address?
Cloud storage offers cost savings and scalability, but users are increasingly concerned about data confidentiality. End-to-end encryption, where data is encrypted on the client-side before being sent to the cloud, is a common solution. However, this encryption renders traditional data reduction techniques like compression and deduplication ineffective. These techniques are crucial for reducing storage costs (often by 20-70%), so their inability to work with encrypted data leads to increased storage capacity requirements and higher costs. The paper aims to bridge this gap by enabling both strong confidentiality and efficient data reduction in cloud storage.
Methods
The core of the proposed solution is a Trusted Decrypter (TD) module integrated into the cloud storage provider's infrastructure. Here's how it works:
- Client-Side Encryption: Data is encrypted by the client's storage application using a master encryption key (MK) and ephemeral file encryption keys (FKs). FKs are then wrapped with the MK and stored as metadata.
- Trusted Decrypter (TD):
- Write Requests: When data arrives, the TD intercepts it. It retrieves the wrapped FK from the auxiliary repository, uses the client's MK (obtained securely from a key repository) to unwrap the FK, and then decrypts the data. The TD then applies compression and deduplication. Finally, it re-encrypts the processed data with a new, randomly generated auxiliary key and stores it. The TD updates its internal indirection table (mapping logical to physical addresses) and deduplication table.
- Read Requests: The TD retrieves the encrypted data, uses the auxiliary information to decrypt it, and then re-encrypts it with the original FK before returning it to the client.
- Secure Data Reduction: The paper proposes algorithms for secure compression and deduplication that work on encrypted data. Crucially, cross-tenant deduplication is disabled to prevent collusion attacks between tenants and the cloud provider.
- Key Management: The system relies on a hierarchical key management scheme, with master keys (MKs) managed by the client's key repository and ephemeral file keys (FKs) managed by the TD.
- Security Provisions: The TD is designed to have a small, verifiable codebase and can leverage hardware security primitives (like TPMs or TEEs) to ensure its integrity and protect sensitive operations.
Datasets
Not discussed in the paper. The evaluation uses synthetic data and performance measurements based on system configurations.
Key Insights and Ideas
- The "Trusted Decrypter" paradigm: This is the central innovation. By placing a trusted component within the cloud provider's infrastructure, the system can reconcile the client's need for end-to-end encryption with the provider's need for data reduction.
- Decoupling Encryption and Data Reduction: The TD acts as an intermediary, allowing data reduction to occur on decrypted data temporarily before it's re-encrypted. This is key to making compression and deduplication effective.
- Security through Isolation and Attestation: The TD's security relies on its small attack surface, secure coding practices, and the ability to attest its integrity to the key repository. This is crucial for building trust in a component that handles sensitive data.
- Intra-tenant Deduplication is Sufficient: The paper argues that disabling cross-tenant deduplication is a necessary security measure to prevent collusion attacks, and that intra-tenant deduplication still provides significant storage savings.
- Secure Offboarding: The framework ensures that when a tenant leaves, their data is inaccessible because the TD no longer has access to the necessary master keys.
Key Results
- Performance: The paper demonstrates that the overhead of cryptographic operations (encryption, decryption, key unwrapping) is manageable, especially when overlapped with disk I/O. For HDD-based systems, cryptographic operations add minimal latency. For SSDs, the latency is more pronounced but still potentially hideable.
- Storage Savings: The framework enables significant storage capacity savings through compression and deduplication, which would otherwise be lost with standard end-to-end encryption.
- Security Analysis: The framework is shown to be secure against storage access adversaries and law enforcement agencies, provided certain assumptions about key management and TD integrity hold. Secure deletion is also addressed.
Limitations and Future Work
- Trust in the TD: The security of the entire system hinges on the trustworthiness of the TD. While the paper proposes methods to secure it (small TCB, attestation), achieving this in practice is a significant engineering challenge.
- Performance with SSDs: While manageable, the cryptographic overhead on SSDs is more significant than on HDDs, suggesting potential areas for optimization.
- Future Work: The authors suggest investigating methodologies for parallel trusted decrypters to further improve performance and scalability. They also note that a full analysis of the tradeoff between secure deletion and deduplication, especially for deduplicated pages, is out of scope.
Summary updated