Default ConfidentialInstanceType to SEV-SNP on supported machines when confidentialCompute is Enabled

[bgartzi]

/kind feature

Describe the solution you'd like
A patch that opens the possibility of configuring AMD SEV-SNP confidential computing instances was merged (see #1410).
This was implemented by extending confidentialCompute, a GCPMachineSpec attribute, to accept more values than just the prior Enabled/Disabled.
Another similar effort is #1426, which aims at TDX on c3 machines (Intel's equivalent technology).

When only Enabled/Disabled was allowed, only AMD SEV machines could be configured. This set compute.Instance's ConfidentialInstanceConfig.EnableConfidentialCompute to true and relied on google API's internal defaults to provision a SEV machine. Those defaults do not seem to be documented [0], though.

AMD SEV-SNP extends previous AMD SEV by covering some of the existing attack channels [1]. That is, AMD SEV-SNP is considered to be a safer alternative following AMD SEV.
However, confidential instance type relies on Google API defaults, which default to SEV (I guess that for backward compatibility reasons).
Wouldn't be a safer choice (from the user perspective) to default to AMD SEV-SNP when confidentialCompute: Enabled is configured for a machine (n2d) that supports both SEV-SNP and SEV?

However, this could have some backward compatibility implications from my point of view. At the moment (as prior to #1410), setting confidentialCompute: Enabled for a n2d machine results on a SEV confidential instance. If the default for n2d and Enabled was changed, the same configuration would result in, although safer on paper, a different SEV-SNP confidential instance.

In case it's needed, I could share a draft patch if that would explain the intentions better.

0. https://pkg.go.dev/google.golang.org/api@v0.201.0/compute/v1#ConfidentialInstanceConfig
1. https://www.amd.com/content/dam/amd/en/documents/epyc-business-docs/white-papers/SEV-SNP-strengthening-vm-isolation-with-integrity-protection-and-more.pdf

[bgartzi]

cc @JoelSpeed,@damdo,@uril,@richardcase who helped me on #1410

[JoelSpeed]

This sounds like a breaking change to me, I think we should probably stick with the behaviour as is.

If particular users would prefer, they could implement policies (like the upcoming mutatingadmissionpolicy, or existing validatingadmissionpolicy) to either mutate or prevent the value from being Enabled, and forcing users to pick the more secure value

[bgartzi]

Thanks for confirming @JoelSpeed!

So If I understood correctly,

• validatingadmissionpolicy would simply force some users to for example, avoid confidentialCompute: Enabled. In kubernetes-for-dummies (me) terms,
• mutatingadmissionpolicy would let users introduce tricks such as mutating Enabled into AmdEncryptedVirtualizationNestedPaging when possible. However, this will be available in kubernetes 1.32, so it wouldn't be possible as of now, right?

And finally, just to understand the whole picture a bit better: this would be left for users to configure, so cluster admins should configure these policies themself, right?. I mean, this is not a solution that could be built in a downstream kubernetes distro, but left for users to decide by themself whether or not to configure it.

[JoelSpeed]

Yes, I think you've understood what I was suggesting.

In theory, you could build the policy into many levels.

• We could include this in the code here, but it would be a breaking change
• Distros could include VAPs/MAPs to enforce it at their level, if they so desired
• Cluster Admins could add VAPs/MAPs to enforce it, if their platform does not

Leaving the behaviour as is, and perhaps documenting if appropriate, is the most flexible option AFAICT

[k8s-triage-robot]

The Kubernetes project currently lacks enough contributors to adequately respond to all issues.

This bot triages un-triaged issues according to the following rules:

• After 90d of inactivity, lifecycle/stale is applied
• After 30d of inactivity since lifecycle/stale was applied, lifecycle/rotten is applied
• After 30d of inactivity since lifecycle/rotten was applied, the issue is closed

You can:

• Mark this issue as fresh with /remove-lifecycle stale
• Close this issue with /close
• Offer to help out with Issue Triage

Please send feedback to sig-contributor-experience at kubernetes/community.

/lifecycle stale

[bgartzi]

/close

[k8s-ci-robot]

@bgartzi: Closing this issue.

In response to this:

/close

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository.