Trust No One: Bringing Confidential Computing to Containers
summary
Confidential computing for containers uses VM-based TEEs (TDX/SEV/etc.) plus attestation to encrypt memory and CPU state and remove the host from the guest TCB. Proof-of-concept uses Kata Containers + cloud-hypervisor + KVM: runc’s inability to talk to KVM and host image mounts are blockers. Proposed fixes are image layer encryption or signatures, and moving image handling into the guest or a hybrid offload with a remote attestation service that vends keys after measured boot; prototypes work but duplicate pulls make the naive design impractical.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 0:35Existing Trust Computing Base
- 1:30What is Trust No One
- 2:12How do we get there
- 3:00Protecting data
- 4:19Verifying software stack
- 5:43Hardware dependencies
- 7:02Software dependencies
- 8:05Blockers
- 9:40Solutions
- 10:25KATA Containers
- 12:18Service Offload
- 15:26Walkthrough
- 18:56What does this mean for the user
- 22:33Gaps
- 25:00Summary
related talks
29mConfidential Containers: The Next Era of Cloud Data Security
31mConfidential Containers – Sensitive Data and Privacy in Cloud Native Environments
50mConfidential Containers: Why, How, and Where Are We?
38mConfidential Containers Explained
30mTrust No One: Secure Storage With Confidential Containers
28mSealed for All: Multi-Party Confidential Computing, No Trust
40mDon’t Trust Opaque Clouds, Cryptographically Verify Instead!
36mEverybody is talking about Confidential Computing - the minimum Devs should know