​
Security Architecture

The security architecture of dStack KMS is built on a foundation of hardware-backed key derivation and a rigorous trust model. Every key is produced using HKDF-SHA256, seeded by unique hardware attestation data and domain-separating context. This ensures that each key is cryptographically unique, isolated, and never persisted outside the TEE.

​
Hardware-Backed Key Derivation

At the core of dStack KMS is a hardware-rooted key derivation model. Each key is produced via HKDF-SHA256, seeded by unique hardware attestation data—specifically, TDX measurements and a domain-separating salt (“RATLS”) plus context data (such as application identifiers). See the key derivation implementation in ra-tls/src/kdf.rs#L19. Key derivation is hierarchical: parent keys can deterministically derive child keys for distinct application, instance, or usage contexts (ra-tls/src/kdf.rs#L34). This ensures every derived key is cryptographically unique and isolated—even if the same KMS instance manages thousands of workloads, each application’s secrets remain independently protected.

​
Trust Model

dStack KMS’s security model is grounded in four main principles:
  1. Hardware Root of Trust: All keys are derived from CPU and runtime measurements (MRTD, RTMR0–RTMR3).
  2. No Key Persistence: Keys are regenerated as needed from attestation data and are never stored outside TEE memory.
  3. Attestation Gated Access: Only workloads that have passed strict TDX quote and event log validation (ra-tls/src/attestation.rs#L291) can access keys.
  4. Determinism: The same measurements and context will always yield the same derived key—supporting reproducibility and scalable, stateless key management.
The security architecture is the backbone of dStack KMS. The next section explores the security properties and guarantees that result from this design.