feat(chopper): add microvm host support - #118
Merged
Merged
Conversation
Bring in microvm.nix and the plumbing a guest needs, without declaring one yet. Adding a guest is now a file in hosts/chopper/microvm/ plus an import. common.nix carries the settings every guest should share and reaches each VM through extraModules, so per-guest files only need what differs: the address, the tap interface and the vsock CID. Guests attach a tap interface to br0 through the existing vm-* match, so no host networking changes are required, and boot against the host's nix store over virtiofs with posixAcl disabled, since zroot has acltype=off and a read-only store carries no ACLs. Each guest gets a small state volume at /var/lib for the things it must not lose: ssh host keys, and the tailscale node identity, which under tailnet lock has to be signed by hand if it is ever regenerated. Bulk storage stays a per-guest concern and belongs on a nested volume, so recreating it cannot take the identity with it. Volumes are only sized at creation and cannot be resized. The image path is relative, so each guest gets its own under /var/lib/microvms/<name>. ts_key reaches guests as a systemd credential rather than a share, since virtiofs-mounting the host's /run/secrets breaks whenever sops rotates it. qemu opens it as the microvm user, so the key is readable by a tailscale-key group instead of root alone, the way security.acme grants cert access. It is an OAuth client secret, so nodes register ephemeral and unapproved unless told otherwise, and a tag is mandatory. Guests register with systemd-machined for visibility through machinectl and run sshd over VSOCK, so `microvm -s <name>` reaches one without a network path to it.
This was referenced Aug 30, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Brings in microvm.nix and everything a guest needs, without declaring one. Adding a guest becomes a file in
hosts/chopper/microvm/plus an import.The shape was worked out by actually running a guest — see #115, which adds an echoip VM on top of this and is kept as a draft for reference. Everything here was verified on chopper with that guest before being split out.
What's included
common.nixcarries the settings every guest should share and reaches each VM throughextraModules, so per-guest files only need what differs: the address, the tap interface, and the vsock CID.Networking. Guests attach a tap interface to
br0through the existingvm-*match invirtualisation.nix, so no host networking changes are needed. Verified:tailscale pingfrom chopper to a guest resolved direct over the LAN at 1ms rather than falling back to a DERP relay.State. Each guest gets a small volume at
/var/libfor what it must not lose — ssh host keys, and the tailscale node identity, which under tailnet lock has to be signed by hand if regenerated. Bulk storage stays a per-guest concern on a nested volume, so recreating it can't take the identity with it. Volumes are only sized at creation and there is no resize upstream, so/var/libis deliberately small and stable. The image path is relative, so each guest gets its own under/var/lib/microvms/<name>.Secrets.
ts_keyreaches guests as a systemd credential rather than a share; virtiofs-mounting the host's/run/secretsis a known-broken pattern, since virtiofsd resolves the symlink once and loses the contents when sops rotates a generation. qemu opens the credential as themicrovmuser, so the key is granted to atailscale-keygroup rather than being reowned — the same shapesecurity.acmeuses for cert access.It's an OAuth client secret, which means nodes register ephemeral and unapproved unless told otherwise, and a tag is mandatory. Both are set explicitly.
Access. Guests register with systemd-machined for visibility through
machinectl, and run sshd over VSOCK somicrovm -s <name>reaches one with no network path — useful for guests that are deliberately unreachable.Effect of merging this alone
No guests are declared, so nothing new runs. The host module creates the
microvmuser and the systemd templates, andts_keymoves from0400 root:rootto0440 root:tailscale-key. Every existing consumer of that secret reads it as root, so the mode change takes nothing away from them.