Linux guests on vz

On an Apple Silicon Mac, a Linux guest can run on Apple’s Virtualization.framework instead of QEMU — the same vz backend that hosts macOS guests, selected per VM with --backend vz.

vee create dev --template server --distro ubuntu --backend vz --vsock
vee ssh dev

Any cloud-image template works unchanged (server, devbox, github-runner, …) — the template builds exactly as it would for QEMU, and the backend flag switches the machine that boots it.

The main reason is vsock. The bundled macOS QEMU is a plain-HVF build with no vsock device, and QEMU’s vsock is vhost-only (a Linux/KVM feature), so a QEMU Linux guest on a macOS host can never have a host↔guest vsock channel. Virtualization.framework provides one natively (VZVirtioSocketDevice), so vsock: true works for Linux guests on macOS hosts — with virtio disks, networking, and EFI boot as native framework devices alongside.

• EFI boot from a whole-disk raw image. The template’s qcow2 cloud-image overlay is flattened to a raw disk once, at create time (qemu-img is required for that step: brew install qemu).
• cloud-init provisioning via the normal cidata.iso seed.
• NAT networking with a deterministic MAC; vee ssh and vee ip resolve the guest address from the host’s DHCP leases.
• A boot console log — the guest’s virtio console is captured to serial.log in the VM directory.
• vsock with --vsock, driven through the helper control protocol.

vz Linux guests are headless (no SPICE or GPU — use the QEMU backend for graphical guests), NAT-only (no bridge networking, no ssh_port forwarding), and currently have no virtiofs shares, TPM, raw device passthrough, or nested virtualization.

See docs/linux-vz.md for the full details.