vm-provisioning

VM Provisioning on Unraid

Architecture

Unraid uses three layers for virtualization:

Layer	Role
KVM	Kernel hypervisor module
QEMU	Hardware emulation (disk, network, display)
libvirt	Management daemon, XML domain definitions, virsh CLI

CRITICAL: The GraphQL API cannot create VMs. It only supports lifecycle operations (start, stop, pause) on existing VMs. Use libvirt for all VM provisioning.

Storage Layout

Path	Contents
/mnt/user/system/libvirt/libvirt.img	Loop-mounted image containing all VM XML definitions
/mnt/user/domains/<VM_NAME>/vdisk1.img	VM disk images (qcow2 or raw)
/mnt/user/isos/	ISO images for installation or cloud-init

virsh Commands

Command	Purpose
virsh list --all	List all VMs with state
virsh define myvm.xml	Register VM from XML file
virsh undefine MyVM	Remove VM definition (keeps disk)
virsh start MyVM	Start a defined VM
virsh shutdown MyVM	Graceful ACPI shutdown
virsh destroy MyVM	Force power off
virsh dumpxml MyVM	Export current XML definition
virsh edit MyVM	Edit XML in-place
virsh snapshot-create-as --domain MyVM --name snap1 --disk-only --atomic	Create disk snapshot
virsh vol-create-as domains disk.qcow2 80G --format qcow2	Create storage volume
virsh nodedev-detach pci_0000_01_00_0	Detach PCI device for passthrough
virsh nodedev-reattach pci_0000_01_00_0	Reattach PCI device to host

Python libvirt API

import libvirt

# Local connection (on Unraid host)
conn = libvirt.open("qemu:///system")

# Remote connection (over SSH)
conn = libvirt.open("qemu+ssh://root@192.168.1.100/system")

# Define and start a VM
with open("myvm.xml") as f:
    xml = f.read()
dom = conn.defineXML(xml)
dom.create()  # start the VM

# List all domains
for dom in conn.listAllDomains():
    print(dom.name(), dom.state())

conn.close()

Cloud-Init NoCloud Workflow

Provision headless Linux VMs without interactive installation:

1. Create VM Disk from Cloud Image

qemu-img create -f qcow2 \
  -b /mnt/user/isos/jammy-server-cloudimg-amd64.img \
  -F qcow2 /mnt/user/domains/MyVM/vdisk1.img 40G

2. Create user-data

#cloud-config
hostname: myvm
users:
  - name: admin
    sudo: ALL=(ALL) NOPASSWD:ALL
    ssh_authorized_keys:
      - ssh-ed25519 AAAA... user@host
packages:
  - qemu-guest-agent
runcmd:
  - systemctl enable --now qemu-guest-agent

3. Create meta-data

instance-id: myvm-001
local-hostname: myvm

4. Generate Cloud-Init ISO

genisoimage -output /mnt/user/isos/myvm-cloud-init.iso \
  -volid cidata -rational-rock -joliet \
  /tmp/user-data /tmp/meta-data

5. Define and Start VM

Attach the ISO as a CD-ROM in the VM XML (see the reference template), then:

virsh define /tmp/myvm.xml
virsh start MyVM

The VM boots, applies cloud-init configuration, and is accessible via SSH.

GPU Passthrough

VFIO Bindings

Edit /boot/config/vfio-pci.cfg to bind PCI devices at boot:

BIND=0000:01:00.0|10de:2484
BIND=0000:01:00.1|10de:228b

Format: BIND=<PCI_ADDRESS>|<VENDOR>:<DEVICE_ID>

Kernel Parameters

Edit /boot/syslinux/syslinux.cfg to add IOMMU and device stubbing:

append ... iommu=pt intel_iommu=on vfio-pci.ids=10de:2484,10de:228b

Runtime Binding

virsh nodedev-detach pci_0000_01_00_0
virsh nodedev-detach pci_0000_01_00_1

XML hostdev Block

Add to the VM XML <devices> section:

<hostdev mode="subsystem" type="pci" managed="yes">
  <source>
    <address domain="0x0000" bus="0x01" slot="0x00" function="0x0"/>
  </source>
</hostdev>

IOMMU Group Enumeration

for d in /sys/kernel/iommu_groups/*/devices/*; do
  n="${d##*/}"; g="${d%/devices/*}"; g="${g##*/}"
  echo "IOMMU Group $g: $(lspci -nns "$n")"
done

VM Cloning

Full Clone

virt-clone --original=SourceVM --name=CloneVM --auto-clone

Creates independent copies of all disks. The clone can run simultaneously with the source.

Linked Clone (COW)

qemu-img create -f qcow2 \
  -b /mnt/user/domains/SourceVM/vdisk1.img \
  -F qcow2 /mnt/user/domains/CloneVM/vdisk1.img

Instant creation, shares base image. Only stores differences. The backing file must remain accessible.

XML Template Reference

See ${CLAUDE_PLUGIN_ROOT}/skills/vm-provisioning/references/vm-xml-template.md for a complete annotated libvirt domain XML template with UEFI boot, cloud-init, GPU passthrough, and USB passthrough examples.