Skip to main content

Importing VMware VMs

FluxVM imports an exported VMware VM (OVA, or OVF plus VMDKs) into raw disks and fixes the guest offline so it boots on virtio. No libguestfs or virt-v2v is involved; the repair goes through guestkit.

fluxctl import-image /srv/exports/web01.ova --name web01
# or over the API
curl -X POST "$API/v1/images/import" -H "Authorization: Bearer $TOKEN" \
  -d '{"source": "/srv/exports/web01.ova", "name": "web01"}'

The source path is on the FluxVM host. When policy.allowed_image_dirs is set the source must be under one of those directories. The route is admin-only.

What it does​

  1. Extracts the OVA (plain files with bare names only; anything with a path is rejected) and parses the OVF: vCPUs, memory, NIC count, firmware (efi/bios), guest OS type and the disks in controller order.
  2. Converts every disk to raw under <state_dir>/images/imported/<name>/disk<N>.raw. Sparse and stream-optimized VMDKs are read natively; other formats go through qemu-img. disk0.raw is the boot disk.
  3. With repair (default on), mounts the boot disk and:
    • disables vmtoolsd, open-vm-tools, vmware-tools, vgauth and related units. remove_vmware_tools: true also purges the packages with dpkg/rpm (no network needed);
    • adds virtio_blk, virtio_scsi, virtio_net, virtio_pci and virtio_console to the dracut or initramfs-tools config and rebuilds the initramfs for every installed kernel;
    • rewrites /dev/sdX and /dev/hdX to /dev/vdX in /etc/fstab, /etc/default/grub and grub.cfg (UUID= and LABEL= entries are left alone);
    • removes 70-persistent-net.rules and adds a DHCP fallback for any en* interface (netplan, or a NetworkManager keyfile on RHEL-family guests). Windows guests get the virtio-win drivers instead (see below).
  4. Returns the disk paths, the OVF summary, a repair report and a suggested POST /v1/vms body.
{
  "image": "/var/lib/fluxvm/images/imported/web01/disk0.raw",
  "extra_disks": ["/var/lib/fluxvm/images/imported/web01/disk1.raw"],
  "ovf": {"name": "web01", "vcpus": 4, "memory_mib": 8192, "firmware": "efi", "os": "ubuntu64Guest", "nics": 1, "disks": [...]},
  "repair": {
    "os_type": "linux", "distro": "ubuntu",
    "actions": ["disabled open-vm-tools.service", "/etc/fstab: 2 /dev/sdX reference(s) moved to /dev/vdX", "rebuilt initramfs (update-initramfs -u -k all)"],
    "warnings": []
  },
  "suggested": {"name": "web01", "image": ".../disk0.raw", "backend": "qemu", "vcpus": 4, "memory_mib": 8192,
                "firmware": "/usr/share/OVMF/OVMF_CODE.fd",
                "data_disks": [{"name": "disk1", "backing": ".../disk1.raw"}]}
}

Drop notes (if any) from suggested and send it to POST /v1/vms. Every disk after the boot disk is in data_disks: each VM gets its own qcow2 overlay on the imported file, created before the first boot and attached as a SCSI disk (serial diskN), so several VMs can be created from one import without sharing writes. For a UEFI source, firmware is filled from qemu_ovmf_code when it is configured; otherwise a note asks for it.

Read the warnings​

  • Static IPs: a config file naming a VMware NIC (ens192, ens160, ...) is reported. The virtio NIC gets a different name, so edit it or rely on the DHCP fallback.
  • Initramfs rebuild failed: the virtio config is still written. Boot with a rescue kernel or run the printed command on first boot.
  • Windows without virtio_win_dir: nothing is injected. Configure it, or install the virtio-win drivers (viostor, vioscsi, netkvm) in the VM before exporting it.

Windows guests​

Set virtio_win_dir in the FluxVM config to an extracted virtio-win ISO:

virtio_win_dir = "/usr/share/virtio-win"

With repair on, the import mounts the Windows system volume and, for viostor (virtio-blk boot disk), vioscsi, NetKVM and vioserial (guest agent channel):

  • picks the driver build for the guest: 2k25/2k22/2k19/2k16/2k12R2 for Windows Server by product name, w11/w10/w8.1/w8/w7 by version, always amd64;
  • copies the .inf, .sys, .cat and .dll files to C:\Windows\Drivers\VirtIO and each .sys to System32\drivers;
  • appends that directory to the SOFTWARE hive DevicePath;
  • registers the driver service from the INF and binds its PCI hardware IDs in the SYSTEM hive CriticalDeviceDatabase, so Windows loads it at the first boot without a PnP install.

Each driver shows up in repair.actions as injected <driver> from <dir>. A missing driver is a warning. A missing viostor gets its own warning, because the guest can't find its boot disk on virtio-blk without it.

MCP​

fluxctl mcp serve --allow-write exposes image_import with the same arguments. See mcp.md.