KioskConfig/configuration.nix

{ config, pkgs, radiantwaveApp, upgradeFlakeRef, radiantwavePlymouth, ... }:

let
  # The published app is a store path with context (builtins.storePath in
  # flake.nix), so Nix substitutes it from the channel's Attic cache and the
  # system closure depends on it.
  radiantApp = radiantwaveApp;
  # When the live-updater module is enabled, RadiantWave launches through the
  # mutable /var/lib/radiantwave/current
  # pointer so it can be swapped without a full nixos-rebuild. Otherwise it
  # launches straight from the immutable system closure as before.
  launchBin =
    if config.services.radiantwaveUpdater.enable or false
    then "/var/lib/radiantwave/current/bin/radiantwave"
    else "/run/current-system/sw/bin/radiantwave";
  myKey = [ "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIL4qWNZnYU5JaqR9c2qh6oFZktzVNBOx42PfDaGofCL6 cnclack@pm.me macbook" ];

  # Headscale preauth key. Paste/replace here.
  #
  # This MUST be a *reusable* key with a long expiry (1 year is good), created
  # with:
  #   headscale preauthkeys create --user <user-id> --reusable --expiration 8760h
  #
  # A single-use key only ever enrolls one node, which would mean a fresh key
  # plus a rebuild for every kiosk that joins the fleet. A reusable key lets any
  # number of kiosks enroll unattended, and the timer below retries the join so
  # a node that misses its window at boot heals on its own.
  headscaleAuthKey = "hskey-auth-UfU_eS_sdOu6-CjrqrVdDJFRFsPnaTtK5sFuCiVlffJAS7uQ1wX1_65MNk205WGsF4clOiBD3ZOV-";
in
{
  ############################################################
  # System configuration
  ############################################################

  # Included packages
  environment.systemPackages = [
    pkgs.cage pkgs.rsync pkgs.dbus pkgs.mesa pkgs.git pkgs.tailscale pkgs.sqlite pkgs.wireplumber
    radiantApp
  ];

  # This populates the EFI partition and tells the NUC how to start NixOS.
  boot.loader.systemd-boot.enable = true;
  boot.loader.efi.canTouchEfiVariables = true;
  # Boot straight into the splash screen without waiting on the bootloader menu.
  boot.loader.timeout = 0;

  # Hardware-specific: NUC 14 needs these for the NVMe and Graphics
  boot.initrd.availableKernelModules = [ "nvme" "xhci_pci" "ahci" "usbhid" "usb_storage" "sd_mod" "i915" "drm" "drm_kms_helper" ];

  # --- RadiantWave splash: Plymouth from boot until the app's first frame ---
  boot.plymouth.enable = true;
  boot.plymouth.theme = "radiantwave";
  boot.plymouth.themePackages = [ (radiantwavePlymouth pkgs) ];

  # Silence the kernel/initrd output behind the splash. The module adds
  # "splash" itself; we keep "quiet" plus show_status=false and loglevel 0.
  boot.consoleLogLevel = 0;
  boot.initrd.verbose = false;
  boot.kernelParams = [ "quiet" "splash" "systemd.show_status=false" ];

  # Hold the splash until the app is ready to take over the display, instead of
  # quitting at multi-user.target (which would flash a black/console screen
  # during the login shell's network wait). The kiosk login shell runs
  # `sudo -n plymouth quit --wait` immediately before exec'ing cage.
  systemd.services.plymouth-quit.wantedBy = pkgs.lib.mkForce [ ];
  systemd.services.plymouth-quit-wait.wantedBy = pkgs.lib.mkForce [ ];

  # Allow the kiosk user to quit Plymouth (NOPASSWD, command-scoped).
  security.sudo.extraRules = [
    {
      users = [ "kiosk" ];
      commands = [
        {
          command = "/run/current-system/sw/bin/plymouth";
          options = [ "NOPASSWD" ];
        }
      ];
    }
  ];

  # Let the kiosk app apply pending updates by restarting its own TTY
  # session (re-autologin -> fresh cage + RadiantWave). Scoped to exactly
  # one unit and verb; everything else keeps stock (denied) policy.
  security.polkit.enable = true;
  environment.etc."polkit-1/rules.d/10-kiosk-getty.rules".text = ''
    polkit.addRule(function(action, subject) {
      if (subject.user !== "kiosk") return null;
      if (action.id !== "org.freedesktop.systemd1.manage-units") return null;
      var unit = action.lookup("unit");
      var verb = action.lookup("verb");
      if ((unit === "getty@tty1.service" || unit === "getty@tty1") && verb === "restart") {
        return polkit.Result.YES;
      }
    });
  '';

  hardware.graphics.enable = true;

  time.timeZone = "America/Chicago";

  networking.networkmanager.enable = true;

  ############################################################
  # Tailscale (self-hosted Headscale control plane)
  ############################################################
  # The node joins the RadiantWave Headscale server at
  # hs.radiantwavetech.com using the *hardware serial* as its Tailscale
  # node name (not networking.hostName), so a kiosk is always reachable at
  # <serial> even though every variant shares one hostName per branch.
  #
  # The preauth key is set in `headscaleAuthKey` in the let block at the
  # top of this file. NOTE: it ends up in the world-readable /nix/store, so
  # keep this file out of public repos while a live key is in it.
  services.tailscale = {
    enable = true;
    useRoutingFeatures = "client";
  };

  # Reach the kiosk's services (SSH, etc.) over the tailnet without extra
  # per-port rules. Membership is gated by the preauth key.
  networking.firewall.trustedInterfaces = [ "tailscale0" ];

  systemd.services.radiantwave-tailscale-auth = {
    description = "Join the RadiantWave tailnet using the hardware serial as node name";
    after = [ "tailscaled.service" "network-online.target" ];
    wants = [ "tailscaled.service" "network-online.target" ];
    wantedBy = [ "multi-user.target" ];
    serviceConfig = {
      Type = "oneshot";
      # Deliberately no RemainAfterExit: the unit has to fall back to inactive
      # so the timer below can trigger it again. The script is idempotent, so
      # re-running it on an already-joined kiosk is a cheap no-op.
    };
    path = [ pkgs.tailscale pkgs.coreutils pkgs.gnused pkgs.gnugrep ];
    script = ''
      set -eu

      # Hardware serial. x86 exposes it via DMI; Raspberry Pi exposes it in
      # the device tree (null-terminated), mirrored under /proc/device-tree.
      serial=""
      for f in \
        /sys/class/dmi/id/product_serial \
        /sys/class/dmi/id/board_serial \
        /sys/firmware/devicetree/base/serial-number \
        /proc/device-tree/serial-number
      do
        if [ -r "$f" ]; then
          serial="$(tr -d '\000' < "$f" | tr -d '\n\r')"
          [ -n "$serial" ] && break
        fi
      done

      # Tailscale node names are RFC 1035 labels: lowercase, [a-z0-9-], <=63.
      serial="$(printf '%s' "$serial" \
        | tr '[:upper:]' '[:lower:]' \
        | tr -c 'a-z0-9-' '-' \
        | sed -e 's/-\{1,\}/-/g' -e 's/^-//' -e 's/-$//' \
        | cut -c1-63)"
      [ -n "$serial" ] || serial="radiantwave-unknown"

      # Already joined? Bail out early. This is what makes the periodic
      # re-invocation from the timer below cheap and safe to run forever.
      # The JSON is pretty-printed with a space, so allow optional whitespace.
      state="$(tailscale status --json 2>/dev/null \
        | sed -n 's/.*"BackendState": *"\([A-Za-z]*\)".*/\1/p')"
      if [ "$state" = "Running" ]; then
        echo "radiantwave-tailscale-auth: already joined as '$serial'"
        exit 0
      fi

      echo "radiantwave-tailscale-auth: joining as '$serial' (state: ''${state:-unknown})"
      for _ in $(seq 1 30); do
        if tailscale up \
             --auth-key=${pkgs.lib.escapeShellArg headscaleAuthKey} \
             --login-server=https://hs.radiantwavetech.com \
             --hostname="$serial" \
             --accept-dns=false
        then
          echo "radiantwave-tailscale-auth: connected"
          exit 0
        fi
        sleep 2
      done

      # Best-effort by design: do NOT exit non-zero here.
      #
      # This unit is part of the new generation that switch-to-configuration
      # starts, and *any* failing unit among those aborts the switch with
      # status 4/NOPERMISSION. While the preauth key was expired that is exactly
      # what happened: this unit failed, which failed `switch-to-configuration`,
      # which failed nixos-upgrade.service -- so the tailscale key silently
      # stopped *all* system upgrades on the fleet every 10 minutes.
      #
      # The retry timer below is what actually recovers the join, so a failed
      # attempt here must never take down the system switch with it.
      echo "radiantwave-tailscale-auth: could not join tailnet after retries;" >&2
      echo "radiantwave-tailscale-auth: leaving the join to the retry timer" >&2
      exit 0
    '';
  };

  # The join only needs to happen once per node, but it fails for reasons
  # outside our control: the network may not be usable yet at boot, tailscaled
  # may be restarted later, or the preauth key may be rotated out from under
  # the fleet. A timer re-runs the (idempotent) join so a kiosk recovers on its
  # own instead of needing a manual `systemctl restart`.
  #
  # This replaces the old behavior where the one-shot unit failed once at boot
  # and stayed failed forever (Restart=no, NRestarts=0).
  systemd.timers.radiantwave-tailscale-auth = {
    description = "Periodically ensure the kiosk is joined to the tailnet";
    wantedBy = [ "timers.target" ];
    timerConfig = {
      # The join is attempted at boot via multi-user.target above; this is the
      # first retry in case that one lost the race with the network.
      OnBootSec = "5min";
      # Measured from when the unit last went inactive, so a slow join doesn't
      # immediately retrigger.
      OnUnitInactiveSec = "10min";
      Unit = "radiantwave-tailscale-auth.service";
    };
  };

  system.autoUpgrade = {
    enable = true;
    flake = upgradeFlakeRef;
    dates = "*:00/10";
    # --impure: the flake reads the channel's published manifest at eval time
    # (builtins.fetchurl) to pick the app store path for the next generation.
    flags = [ "--refresh" "--impure" "-L" ];
    operation = "switch";
  };

  # Virtual machine declaration
  virtualisation.vmVariant = {
    boot.loader.systemd-boot.enable = pkgs.lib.mkForce false;
    boot.loader.efi.canTouchEfiVariables = pkgs.lib.mkForce false;
    virtualisation.forwardPorts = [
      { from = "host"; host.port = 2222; guest.port = 22; }
    ];
    virtualisation.qemu.options = [
      "-device intel-hda"
      "-device hda-duplex"
      "-vga virtio"
    ];
    virtualisation = {
      cores = 8;
      memorySize = 8192;
    };
    services.logrotate.enable = false;
  };
  virtualisation.diskSize = 10485760;

  # NixOS settings.
  nix.settings.experimental-features = [ "nix-command" "flakes" ];
  nix.settings.accept-flake-config = true;
  # The app baked into each generation comes from the channel's published
  # manifest, fetched at eval time with `builtins.fetchurl` (see flake.nix).
  # Disable the tarball cache so a rebuild during autoUpgrade always reads the
  # current manifest instead of a copy up to an hour old.
  nix.settings.tarball-ttl = 0;
  nix.settings.extra-substituters = [
    "https://attic.radiantwavetech.com/radiantwave"
    "https://attic.radiantwavetech.com/radiantwave-dev"
  ];
  nix.settings.extra-trusted-public-keys = [
    "radiantwave:QADrjpBWJssTZ3KmiZtcWEf4uAm66oOlHmtQI3jkrgA="
    "radiantwave-dev:B8FhLu3MQ9JfeEIHIKNrKxnP+3akzzkqMbTZR4hRfj0="
  ];
  system.stateVersion = "24.05";

  ############################################################
  # User configuration
  ############################################################

  users.users.root.openssh.authorizedKeys.keys = myKey;
  users.users.kiosk = {
    isNormalUser = true;
    uid = 1000;
    extraGroups = ["audio" "video" "input" "networkmanager" "wheel" "seat"];
    openssh.authorizedKeys.keys = myKey;
  };

  services.getty.autologinUser = "kiosk";

  # --- The Kiosk Startup Script ---
  environment.loginShellInit = ''
   if [ "$(tty)" = "/dev/tty1" ]; then
      export GODEBUG=netdns=cgo
      export SSL_CERT_FILE=/etc/ssl/certs/ca-certificates.crt

      # Smart Network Wait
      for i in {1..10}; do
        if ${pkgs.iputils}/bin/ping -c 1 8.8.8.8 &> /dev/null; then
          break
        fi
        sleep 1
      done

      # Graphics fixes
      export SDL_VIDEODRIVER=wayland
      export QT_QPA_PLATFORM=wayland
      export XDG_SESSION_TYPE=wayland
      export WLR_RENDERER=pixman
      export WLR_NO_HARDWARE_CURSORS=1

      # Pin system audio to 100% and unmute before launching RadiantWave
      ${pkgs.systemd}/bin/systemctl --user start radiantwave-audio-init.service || true

      # Release the splash screen right before the compositor takes over the display
      sudo -n plymouth quit --wait || true

      exec dbus-run-session cage -- ${launchBin} -d -s > /tmp/kiosk.log 2>&1
    fi
  '';

  ############################################################
  # Service configuration
  ############################################################

  services.pipewire = { enable = true; alsa.enable = true; pulse.enable = true; };

  # New sinks with no stored volume default to ~40% in WirePlumber 0.5+.
  # Force every newly-created sink to 100% so switching the default sink
  # never changes the volume.
  services.pipewire.wireplumber.extraConfig."99-kiosk-audio" = {
    "wireplumber.settings"."device.routes.default-sink-volume" = 1.0;
  };

  # --- Pin all system audio sinks to 100% before RadiantWave launches ---
  # User unit (PipeWire runs per-user in the kiosk session), owned by this
  # system config. Ordered after PipeWire/WirePlumber; the retry loop covers
  # sink readiness at boot. Pins every connected sink (not just the default)
  # and unmutes them, so switching the default sink never changes the volume.
  # RadiantWave's loginShellInit waits on this before exec'ing cage.
  systemd.user.services.radiantwave-audio-init = {
    description = "Pin all system audio sinks to 100% and unmute";
    after = [ "pipewire.service" "wireplumber.service" ];
    wants = [ "pipewire.service" "wireplumber.service" ];
    serviceConfig = {
      Type = "oneshot";
    };
    script = ''
      # Wait until WirePlumber exposes at least one sink.
      # `wpctl list` output is tab-separated: ID, name, media/object type,
      # then `*` if the node is the current default. `cut -f1` grabs the ID.
      sinks=""
      for i in {1..30}; do
        sinks=$(${pkgs.wireplumber}/bin/wpctl list audio sinks 2>/dev/null \
          | ${pkgs.coreutils}/bin/cut -f1)
        if [ -n "$sinks" ]; then
          break
        fi
        sleep 1
      done

      # Pin and unmute every connected sink
      for sink_id in $sinks; do
        ${pkgs.wireplumber}/bin/wpctl set-volume "$sink_id" 1.0 2>/dev/null || true
        ${pkgs.wireplumber}/bin/wpctl set-mute "$sink_id" 0 2>/dev/null || true
      done

      # Fallback in case the sink list came up empty
      ${pkgs.wireplumber}/bin/wpctl set-volume @DEFAULT_AUDIO_SINK@ 1.0 2>/dev/null || true
      ${pkgs.wireplumber}/bin/wpctl set-mute @DEFAULT_AUDIO_SINK@ 0 2>/dev/null || true

      # Persist 100% as WirePlumber's default for any brand-new sink so it
      # comes up at full volume even before this service runs on next boot.
      ${pkgs.wireplumber}/bin/wpctl settings device.routes.default-sink-volume 1.0 --save 2>/dev/null || true
    '';
  };

  services.seatd.enable = true;
  services.openssh.enable = true;
  services.openssh.settings.PermitRootLogin = "prohibit-password";

  # Networking Network-Manager
  services.resolved.enable = false;
  systemd.services.NetworkManager-wait-online.enable = false;
  systemd.network.wait-online.enable = false;

  # --- Set ownership of Radiantwave files ---
  systemd.tmpfiles.rules = [
    "d /home/kiosk/.local/radiantwave 0770 kiosk kiosk -"
  ];

  # --- Backup service: runs before nixos-upgrade ---
  systemd.services.radiantwave-backup = {
    description = "Backup RadiantWave user data before upgrade";
    before = [ "nixos-upgrade.service" ];
    wantedBy = [ "nixos-upgrade.service" ];
    serviceConfig = {
      Type = "oneshot";
      User = "kiosk";
    };
    script = ''
      BACKUP_DIR="/home/kiosk/.local/radiantwave-backup"
      SOURCE_DIR="/home/kiosk/.local/radiantwave"

      mkdir -p "$BACKUP_DIR"

      ${pkgs.rsync}/bin/rsync -a \
        --exclude='audio/' \
        "$SOURCE_DIR/" "$BACKUP_DIR/"

      echo "Backup completed at $(date)" > "$BACKUP_DIR/.backup-timestamp"
    '';
  };
}