vmix.nix/nixos/vms/config.nix

{ config, pkgs, lib, vmixLib, ... }:
with lib;
with vmixLib.network;
let
  vmixCfg = config.vmix;
  mkShortIfaceName = prefix: seed: "${prefix}-${builtins.substring 0 8 (builtins.hashString "sha256" seed)}";

  mkServices4aVMInNamespace = spaceName: vmName: cfg:
    let
      vmCfg = cfg // { name = vmName; };
      netCfg = vmCfg.networks;

      mkTap4aLan = lanName: tapCfg:
      let
        tapInterfaceName = mkShortIfaceName "vt" "${spaceName}-${vmCfg.name}-${lanName}";
        lanInterfaceName = "brx-${lanName}";
      in
      {
        name = lanName;
        iface = tapInterfaceName;
        mac = tapCfg.mac;
        create = ''
          ip tuntap add dev ${tapInterfaceName} mode tap
          ip link set dev ${tapInterfaceName} up
          ip link set dev ${tapInterfaceName} master ${lanInterfaceName}
        '' + lib.optionalString (tapCfg.ip != null) '' 
          # Make the static ip somehow part of the script so nixOs thinks the service has changed when the IP changes, which will trigger a VM restart.
          # So whenever IP changes, VM will restart automatically
          # The IP is actually bein assigned to VM by dnsmasq, when VM requests it via DHCP 
          # Static IP - ${tapCfg.ip}
        '';
        delete = ''
          ip link del ${tapInterfaceName}
        '';
      };

      mkMacvtap = macvtapName: macvtapVmCfg:
      let
        macvtapNetworkCfg = config.vmix.namespaces.${spaceName}.networks.macvtaps.${macvtapName};
        macvtapInterfaceName = mkShortIfaceName "mt" "${spaceName}-${vmCfg.name}-${macvtapNetworkCfg.uplink.iface}-${macvtapName}";
        uplinkNamespaceArg = lib.optionalString (macvtapNetworkCfg.uplink.namespace != null) "-n ${macvtapNetworkCfg.uplink.namespace}";
      in
      {
        name = macvtapName;
        iface = macvtapInterfaceName;
        mac = macvtapVmCfg.mac;

        create = ''
          ip ${uplinkNamespaceArg} link add link ${macvtapNetworkCfg.uplink.iface} name ${macvtapInterfaceName} type macvtap mode bridge
          ${lib.optionalString (macvtapVmCfg.mac != null) "ip ${uplinkNamespaceArg} link set dev ${macvtapInterfaceName} address ${macvtapVmCfg.mac}"}
          ip ${uplinkNamespaceArg} link set ${macvtapInterfaceName} netns ${spaceName}.vmix
          ip -n ${spaceName}.vmix link set dev ${macvtapInterfaceName} up
        '';
        delete = ''
          ip -n ${spaceName}.vmix link del ${macvtapInterfaceName}
        '';
      };

      allTaps = (mapAttrsToList mkTap4aLan netCfg.lans);
      allMacvtaps = (mapAttrsToList mkMacvtap netCfg.macvtaps);

      createTapsforLansScript = pkgs.writeShellScript "${vmCfg.name}-taps-vmix" (
        concatStringsSep "\n" (builtins.map (tap: tap.create) allTaps)
      );

      deleteTapsforLansScript = pkgs.writeShellScript "${vmCfg.name}-taps-vmix" (
        concatStringsSep "\n" (builtins.map (tap: tap.delete) allTaps)
      );

      createMacvTapsScript = pkgs.writeShellScript "${vmCfg.name}-taps-vmix" (
        concatStringsSep "\n" (builtins.map (macvtap: macvtap.create) allMacvtaps)
      );

      deleteMacvTapsScript = pkgs.writeShellScript "${vmCfg.name}-taps-vmix" (
        concatStringsSep "\n" (builtins.map (macvtap: macvtap.delete) allMacvtaps)
      );

      hasOsDisk = vmCfg.disks.os.file != null;

      # Auto-detect Windows from _vmixOsType marker on the disk image
      isWindows = vmCfg.windows.enable || (hasOsDisk && (vmCfg.disks.os.file._vmixOsType or "linux") == "windows");

      # Linux VMs: apply customizeImage with 9p fstab and machine-id setup
      linuxOsImage = vmixLib.linux.customizeImage vmCfg.disks.os.file {
        name = vmCfg.name;
        commands = ''
          truncate /etc/machine-id
          run-command systemd-machine-id-setup
          run-command ssh-keygen -A
          run ${vmixLib.linux.scriptsNFiles.add-9p-mounts-to-fstab vmCfg.shares}
        '';
      };

      # Windows VMs: use disk image as-is (customization done at image build time)
      storeImage = if !hasOsDisk then null
        else if isWindows then vmCfg.disks.os.file
        else linuxOsImage;

      # When persist = true, QEMU needs a mutable disk outside /nix/store.
      # The store image is copied to persistPath on first boot.
      osDiskPath = if !hasOsDisk then null
        else if vmCfg.disks.os.persist then vmCfg.disks.os.persistPath
        else toString storeImage;

      # Script to seed the persistent disk from the store image on first boot
      seedPersistentDiskScript = pkgs.writeShellScript "${vmCfg.name}-seed-disk-vmix" ''
        PERSIST_PATH="${vmCfg.disks.os.persistPath}"
        if [ ! -f "$PERSIST_PATH" ]; then
          echo "Seeding persistent disk from store image..."
          mkdir -p "$(dirname "$PERSIST_PATH")"
          cp --no-preserve=mode "${toString storeImage}" "$PERSIST_PATH"
          chmod 600 "$PERSIST_PATH"
        fi
      '';

      persistExecStartPre = lib.optional (hasOsDisk && vmCfg.disks.os.persist) seedPersistentDiskScript;

      # QEMU expects single-letter boot codes (e.g. c,d,n), while vmix uses readable names.
      bootOrderQemu =
      let
        bootDeviceAliases = {
          os = "c";
          iso = "d";
          net = "n";
          floppy = "a";
        };
      in
        concatStrings (builtins.map (device: bootDeviceAliases.${device}) vmCfg.boot.order);

      spiceUsbRedirArgs =
        if vmCfg.spice.enable && vmCfg.spice.usbRedir.enable then
          concatStringsSep " \\\n          " ([
            "-device qemu-xhci,id=spice-usb-xhci"
          ] ++ (concatMap (i: [
            "-chardev spicevmc,name=usbredir,id=spice-usbredirchardev${toString i}"
            "-device usb-redir,chardev=spice-usbredirchardev${toString i},id=spice-usbredirdev${toString i}"
          ]) (range 1 vmCfg.spice.usbRedir.channels)))
        else
          "";

      vncArgs = concatStringsSep "," (
        [
          "${vmCfg.vnc.addr}:${toString (vmCfg.vnc.port - 5900)}"
          "share=${vmCfg.vnc.sharePolicy}"
        ]
        ++ optional (vmCfg.vnc.websocketPort != null) "websocket=${toString vmCfg.vnc.websocketPort}"
        ++ optional (vmCfg.vnc.passwordFile != null) "password-secret=vnc-pass-${vmCfg.name}"
      );

      qemuStartVMScript = pkgs.writeShellScript "${vmCfg.name}-qemu-vmix" ''
        ${optionalString vmCfg.vnc.enable ''
          ${optionalString (vmCfg.vnc.passwordFile != null) ''
            if [ ! -r ${escapeShellArg vmCfg.vnc.passwordFile} ]; then
              echo "VNC password file is not readable: ${vmCfg.vnc.passwordFile}" >&2
              exit 1
            fi
            if [ ! -s ${escapeShellArg vmCfg.vnc.passwordFile} ]; then
              echo "VNC password file is empty: ${vmCfg.vnc.passwordFile}" >&2
              exit 1
            fi
          ''}
        ''}
        ${optionalString vmCfg.spice.enable ''
          ${optionalString (vmCfg.spice.passwordFile != null) ''
            if [ ! -r ${escapeShellArg vmCfg.spice.passwordFile} ]; then
              echo "SPICE password file is not readable: ${vmCfg.spice.passwordFile}" >&2
              exit 1
            fi
            if [ ! -s ${escapeShellArg vmCfg.spice.passwordFile} ]; then
              echo "SPICE password file is empty: ${vmCfg.spice.passwordFile}" >&2
              exit 1
            fi
          ''}
        ''}
        exec qemu-system-${vmCfg.arch} \
          ${if vmCfg.nographic && vmCfg.pci.passthrough != [] then "-display none -vga none" else optionalString vmCfg.nographic "-nographic"} \
          ${optionalString (vmCfg.vnc.enable && vmCfg.vnc.passwordFile != null) "-object secret,id=vnc-pass-${vmCfg.name},file=${escapeShellArg vmCfg.vnc.passwordFile}"} \
          ${optionalString vmCfg.vnc.enable "-vnc ${vncArgs}"} \
          ${optionalString (vmCfg.spice.enable && vmCfg.spice.passwordFile != null) "-object secret,id=spice-pass-${vmCfg.name},file=${escapeShellArg vmCfg.spice.passwordFile}"} \
          ${optionalString vmCfg.spice.enable "-spice addr=${vmCfg.spice.addr},port=${toString vmCfg.spice.port}${optionalString (vmCfg.spice.passwordFile == null) ",disable-ticketing=on"}${optionalString (vmCfg.spice.passwordFile != null) ",password-secret=spice-pass-${vmCfg.name}"}"} \
          ${optionalString vmCfg.spice.enable "-vga ${vmCfg.spice.displayDevice}"} \
          ${optionalString (vmCfg.spice.enable && vmCfg.spice.agent.enable) "-device virtio-serial-pci -chardev spicevmc,id=vdagent,debug=0,name=vdagent -device virtserialport,chardev=vdagent,name=com.redhat.spice.0"} \
          ${# Guest agent channel — prevents qemu-ga from spinning when virtio-win guest tools are installed
            optionalString isWindows "${optionalString (!vmCfg.spice.enable || !vmCfg.spice.agent.enable) "-device virtio-serial-pci"} -chardev socket,path=/tmp/qga-${vmCfg.name}.sock,server=on,wait=off,id=qga0 -device virtserialport,chardev=qga0,name=org.qemu.guest_agent.0"} \
          ${spiceUsbRedirArgs} \
          ${optionalString vmCfg.kvm "-accel kvm"} \
          -name ${vmCfg.name} \
          -m ${toString vmCfg.mem.size} \
          ${optionalString vmCfg.mem.balloon "-device virtio-balloon-pci"} \
          -smp cores=${toString vmCfg.cpu.cores} \
          -cpu ${vmCfg.cpu.model}${optionalString vmCfg.cpu.hideVirtualized ",kvm=off,hv_vendor_id=1234567890ab,-hypervisor"} \
          -machine type=${vmCfg.pc.type}${optionalString vmCfg.cpu.hideVirtualized ",kernel_irqchip=on"} \
          ${optionalString vmCfg.bios.efi "-bios ${pkgs.OVMF.fd}/FV/OVMF.fd"} \
          ${optionalString vmCfg.bios.tpm "-chardev socket,id=chrtpm,path=/tmp/mytpm-sock -tpmdev emulator,id=tpm0,chardev=chrtpm -device tpm-tis,tpmdev=tpm0"} \
          ${# Windows: localtime RTC, USB tablet for mouse, disable S3/S4 sleep
            optionalString isWindows ''
          -rtc base=localtime,clock=host \
          -device qemu-xhci -device usb-tablet \
          -global ICH9-LMB.disable_s3=1 -global ICH9-LMB.disable_s4=1 \
          ''} \
          ${optionalString hasOsDisk "-drive file=${osDiskPath},format=qcow2,if=virtio${optionalString (vmCfg.disks.os.persist == false) ",snapshot=on"}"} \
          ${optionalString (vmCfg.disks.iso.file != null) "-drive file=${toString vmCfg.disks.iso.file},media=cdrom,readonly=on"} \
          ${concatMapStrings (diskCfg: ''
            -drive file=${toString diskCfg.file},format=${diskCfg.format},if=${vmCfg.disks.bus} \
          '') (attrValues vmCfg.disks.add)} \
          ${concatStrings (mapAttrsToList (shareName: shareCfg: ''
            -virtfs local,path=${toString shareCfg.source},security_model=passthrough,mount_tag=${shareName} \
          '') vmCfg.shares)} \
          ${optionalString cfg.networks.user.enable "
            -netdev user,id=user \
            -device virtio-net-pci,netdev=user \
          "} \
          ${concatMapStrings (tapCfg: ''
            -device virtio-net-pci,netdev=lan-${tapCfg.name},mac=${tapCfg.mac} \
            -netdev tap,id=lan-${tapCfg.name},ifname=${tapCfg.iface},script=no,downscript=no \
          '') allTaps} \
          ${concatStrings (imap1 (i: macvtap: ''
            -device virtio-net-pci,netdev=macvtap-${macvtap.name},mac=$(ip l show ${macvtap.iface} | awk '/link\/ether/{print $2}') \
            -netdev tap,id=macvtap-${macvtap.name},fd=${toString (i+2)} ${toString (i+2)}<>/dev/tap$(ip l show ${macvtap.iface} | awk -F':' '/${macvtap.iface}/{print $1}') \
          '') allMacvtaps)} \
          ${concatStrings (imap1 (i: pciAddr: ''
            -device pcie-root-port,id=pci-passthrough${toString i},chassis=${toString i},slot=${toString i} \
            -device vfio-pci,host=${pciAddr},bus=pci-passthrough${toString i}${optionalString (i == 1) ",x-vga=on${optionalString (vmCfg.pci.romFile != null) ",romfile=${vmCfg.pci.romFile}"}"} \
          '') vmCfg.pci.passthrough)} \
          ${concatMapStrings (usbDev: ''
            -device usb-host,vendorid=0x${usbDev.vendorId},productid=0x${usbDev.productId} \
          '') vmCfg.usb.hostDevices} \
          ${optionalString (vmCfg.boot.menu == true) "-boot menu=on"} \
          ${optionalString (length vmCfg.boot.order > 0) "-boot order=${bootOrderQemu}"} \
      '';
  in
  lib.optionalAttrs (cfg.enable) {
    "vm.vmix@${vmCfg.name}" = rec {
      bindsTo = [ "net.vmix@${spaceName}.target" "macvtaps.vm.vmix@${vmCfg.name}.service" ];
      unitConfig.JoinsNamespaceOf = "ns.net.vmix@${spaceName}.service";
      after = bindsTo;
      path = with pkgs; [ iproute2 qemu gawk coreutils ];
      serviceConfig = {
        ExecStartPre = persistExecStartPre ++ [ createTapsforLansScript ];
        ExecStart = qemuStartVMScript;
        ExecStopPost = deleteTapsforLansScript;
        PrivateTmp = true;
        ProtectSystem = true;
        ProtectHome = true;
        PrivateNetwork = true;
      } // lib.optionalAttrs (vmCfg.pci.passthrough != []) {
        # VFIO passthrough needs raw device access — relax sandboxing
        ProtectSystem = lib.mkForce false;
        SupplementaryGroups = [ "kvm" ];
      };
      wantedBy = lib.mkIf vmCfg.autostart [ "multi-user.target" ]; 
    };

    "macvtaps.vm.vmix@${vmCfg.name}" = rec {
      bindsTo = [ "net.vmix@${spaceName}.target" ];
      after = bindsTo;
      partOf = [ "vm.vmix@${vmCfg.name}.service" ];
      path = with pkgs; [ iproute2 ];
      serviceConfig = {
        Type = "oneshot";
        RemainAfterExit = true;
        ExecStart = createMacvTapsScript;
        ExecStop = deleteMacvTapsScript;
      };
    };
  };

  vmServices = concatMapAttrs (spaceName: namespaceCfg: (concatMapAttrs (mkServices4aVMInNamespace spaceName) namespaceCfg.vms)) vmixCfg.namespaces;
in
{
  config.systemd.services = vmServices;
}