nixos-router/modules/settings/default.nix

# A router configured by a settings file instead of Nix code, so that it can
# be changed on the box itself (by the web UI or by hand) and rebuilt there.
#
# settings.json holds the `router` options as JSON plus a few system settings:
#
#   {
#     "system": { "hostName": "router", "timeZone": "UTC", "sshKeys": [ "ssh-ed25519 ..." ],
#                 "stateVersion": "26.11" },
#     "router": { "wan": { "interface": "enp1s0" }, "lans": { "lan": { ... } } }
#   }
#
# The module system validates it like any configuration: unknown keys, wrong
# types and the router's assertions all fail the build. Secrets stay out of it
# (files under /var/lib/router, referenced by path).
#
# On the device the configuration lives in /var/lib/router/config: a flake
# (pinned to this router's source), settings.json and an optional local.nix.
# `router-apply` builds it and activates it with commit-confirm: unless
# `router-confirm` follows within router.rollback.timeout, the previous system
# comes back.
{
  config,
  lib,
  pkgs,
  ...
}:
let
  cfg = config.routerSettings;
  profile = "/nix/var/nix/profiles/system";
  settings = lib.importJSON cfg.file;
  system = settings.system or { };

  pending = "/run/router-pending-settings.json";
  confirmed = "${cfg.stateDirectory}/settings.confirmed.json";

  apply = pkgs.writeShellApplication {
    name = "router-apply";
    runtimeInputs = [
      config.nix.package
      pkgs.coreutils
    ];
    text = ''
      timeout=''${1:-${config.router.rollback.timeout}}
      # Remember what is being built: confirming records it (the web UI shows
      # settings that differ from it as not yet applied).
      building=$(mktemp /run/router-building.XXXXXX)
      cp ${cfg.directory}/settings.json "$building"
      echo "router: building ${cfg.directory}"
      new=$(nix build --no-link --print-out-paths --out-link /run/router-pending \
        "path:${cfg.directory}#nixosConfigurations.router.config.system.build.toplevel")
      if [ "$new" = "$(readlink -f /run/current-system)" ]; then
        echo "router: no changes"
        if [ "$new" = "$(readlink -f ${profile})" ]; then
          install -m 0600 "$building" ${confirmed}
        fi
        rm -f "$building"
        exit 0
      fi
      mv "$building" ${pending}
      router-arm "$timeout"
      echo "router: activating $new"
      "$new/bin/switch-to-configuration" test
    '';
  };

  webui = pkgs.callPackage ../../pkgs/router-webui/package.nix { };
in
{
  # Outside `router`: the settings file defines that whole tree.
  options.routerSettings = {
    file = lib.mkOption {
      type = lib.types.path;
      description = "The settings file (JSON) this system is built from.";
    };
    directory = lib.mkOption {
      type = lib.types.str;
      default = "/var/lib/router/config";
      description = "Where the configuration (flake and settings) lives on the device.";
    };
    stateDirectory = lib.mkOption {
      type = lib.types.str;
      default = "/var/lib/router";
      description = "Secrets and web UI state (password hash, certificate).";
    };
    webui = {
      enable = lib.mkOption {
        type = lib.types.bool;
        default = true;
        description = "Serve the web UI (HTTPS) to LANs and tunnels with `allowManagement`.";
      };
      port = lib.mkOption {
        type = lib.types.port;
        default = 443;
      };
    };
  };

  config = {
    router = lib.mkMerge [
      (settings.router // { enable = true; })
      {
        rollback.onConfirm = ''
          if [ -e ${pending} ]; then mv ${pending} ${confirmed}; fi
        '';
      }
      (lib.mkIf cfg.webui.enable { managementPorts = [ cfg.webui.port ]; })
    ];

    networking.hostName = system.hostName or "router";
    time.timeZone = system.timeZone or "UTC";
    users.users.root.openssh.authorizedKeys.keys = system.sshKeys or [ ];
    system.stateVersion = system.stateVersion or config.system.nixos.release;

    nix.settings.experimental-features = [
      "nix-command"
      "flakes"
    ];

    environment.systemPackages = [
      apply
      webui
    ];

    # The web UI starts these rather than running anything itself, so a switch
    # that restarts the UI does not cut them short; a switch must not restart
    # or stop them either (router-apply is what runs the switch).
    systemd.services.router-apply = {
      description = "Build and activate the router configuration";
      restartIfChanged = false;
      stopIfChanged = false;
      path = [ "/run/current-system/sw" ];
      serviceConfig = {
        Type = "oneshot";
        ExecStart = lib.getExe apply;
        TimeoutStartSec = "1h";
      };
    };
    systemd.services.router-confirm = {
      description = "Confirm the active router configuration";
      restartIfChanged = false;
      stopIfChanged = false;
      path = [ "/run/current-system/sw" ];
      serviceConfig = {
        Type = "oneshot";
        ExecStart = "/run/current-system/sw/bin/router-confirm";
      };
    };

    systemd.services.router-webui = lib.mkIf cfg.webui.enable {
      description = "Router web UI";
      wantedBy = [ "multi-user.target" ];
      after = [ "network.target" ];
      path = [ config.systemd.package ];
      serviceConfig = {
        ExecStart = "${lib.getExe webui} -listen :${toString cfg.webui.port} -config ${cfg.directory} -state ${cfg.stateDirectory}";
        Restart = "on-failure";
        # It edits settings and secrets and talks to systemd; nothing else.
        ProtectSystem = "strict";
        ReadWritePaths = [ cfg.stateDirectory ];
        ProtectHome = true;
        PrivateTmp = true;
        PrivateDevices = true;
        NoNewPrivileges = true;
        ProtectKernelTunables = true;
        ProtectKernelModules = true;
        ProtectControlGroups = true;
        RestrictAddressFamilies = [
          "AF_UNIX"
          "AF_INET"
          "AF_INET6"
          "AF_NETLINK"
        ];
        LockPersonality = true;
        MemoryDenyWriteExecute = true;
      };
    };

    # Rebuilding on the device must work without internet (the uplink may be
    # what is being configured). What a changed configuration rebuilds is all
    # generated by NixOS, but those builds need these packages; found by
    # diffing the derivation graphs of typical changes (pool, addresses, host
    # name, PPPoE on a VLAN, a VLAN LAN, IPv6 PD/ULA, a port forward, DNS
    # records) against the installed system.
    system.extraDependencies = with pkgs; [
      stdenvNoCC
      findXMLCatalogs
      jq.dev
      kmod.dev
      libcap-text-verifier
      libxslt.bin
      # nftables checks its ruleset at build time inside lkl.
      libredirect
      lklWithFirewall.lib
      lndir
      makeShellWrapper
      mdadm
      nukeReferences
      # Switching the uplink to PPPoE is a typical first change.
      ppp
    ];
  };
}