nixpkgs/nixos/modules/tasks/network-interfaces.nix

{ config, lib, pkgs, ... }:

with lib;

let

  cfg = config.networking;
  interfaces = attrValues cfg.interfaces;
  hasVirtuals = any (i: i.virtual) interfaces;
  hasBonds = cfg.bonds != { };

  interfaceOpts = { name, ... }: {

    options = {

      name = mkOption {
        example = "eth0";
        type = types.str;
        description = "Name of the interface.";
      };

      ipAddress = mkOption {
        default = null;
        example = "10.0.0.1";
        type = types.nullOr (types.str);
        description = ''
          IP address of the interface.  Leave empty to configure the
          interface using DHCP.
        '';
      };

      prefixLength = mkOption {
        default = null;
        example = 24;
        type = types.nullOr types.int;
        description = ''
          Subnet mask of the interface, specified as the number of
          bits in the prefix (<literal>24</literal>).
        '';
      };

      subnetMask = mkOption {
        default = "";
        example = "255.255.255.0";
        type = types.str;
        description = ''
          Subnet mask of the interface, specified as a bitmask.
          This is deprecated; use <option>prefixLength</option>
          instead.
        '';
      };

      ipv6Address = mkOption {
        default = null;
        example = "2001:1470:fffd:2098::e006";
        type = types.nullOr types.string;
        description = ''
          IPv6 address of the interface.  Leave empty to configure the
          interface using NDP.
        '';
      };

      ipv6prefixLength = mkOption {
        default = 64;
        example = 64;
        type = types.int;
        description = ''
          Subnet mask of the interface, specified as the number of
          bits in the prefix (<literal>64</literal>).
        '';
      };

      macAddress = mkOption {
        default = null;
        example = "00:11:22:33:44:55";
        type = types.nullOr (types.str);
        description = ''
          MAC address of the interface. Leave empty to use the default.
        '';
      };

      mtu = mkOption {
        default = null;
        example = 9000;
        type = types.nullOr types.int;
        description = ''
          MTU size for packets leaving the interface. Leave empty to use the default.
        '';
      };

      virtual = mkOption {
        default = false;
        type = types.bool;
        description = ''
          Whether this interface is virtual and should be created by tunctl.
          This is mainly useful for creating bridges between a host a virtual
          network such as VPN or a virtual machine.

          Defaults to tap device, unless interface contains "tun" in its name.
        '';
      };

      virtualOwner = mkOption {
        default = "root";
        type = types.str;
        description = ''
          In case of a virtual device, the user who owns it.
        '';
      };

      proxyARP = mkOption {
        default = false;
        type = types.bool;
        description = ''
          Turn on proxy_arp for this device (and proxy_ndp for ipv6).
          This is mainly useful for creating pseudo-bridges between a real
          interface and a virtual network such as VPN or a virtual machine for
          interfaces that don't support real bridging (most wlan interfaces).
          As ARP proxying acts slightly above the link-layer, below-ip traffic
          isn't bridged, so things like DHCP won't work. The advantage above
          using NAT lies in the fact that no IP addresses are shared, so all
          hosts are reachable/routeable.

          WARNING: turns on ip-routing, so if you have multiple interfaces, you
          should think of the consequence and setup firewall rules to limit this.
        '';
      };

    };

    config = {
      name = mkDefault name;
    };

  };

in

{

  ###### interface

  options = {

    networking.hostName = mkOption {
      default = "nixos";
      description = ''
        The name of the machine.  Leave it empty if you want to obtain
        it from a DHCP server (if using DHCP).
      '';
    };

    networking.enableIPv6 = mkOption {
      default = true;
      description = ''
        Whether to enable support for IPv6.
      '';
    };

    networking.defaultGateway = mkOption {
      default = "";
      example = "131.211.84.1";
      description = ''
        The default gateway.  It can be left empty if it is auto-detected through DHCP.
      '';
    };

    networking.defaultGatewayWindowSize = mkOption {
      default = null;
      example = 524288;
      type = types.nullOr types.int;
      description = ''
        The window size of the default gateway. It limits maximal data bursts that TCP peers
        are allowed to send to us.
      '';
    };

    networking.nameservers = mkOption {
      default = [];
      example = ["130.161.158.4" "130.161.33.17"];
      description = ''
        The list of nameservers.  It can be left empty if it is auto-detected through DHCP.
      '';
    };

    networking.domain = mkOption {
      default = "";
      example = "home";
      description = ''
        The domain.  It can be left empty if it is auto-detected through DHCP.
      '';
    };

    networking.localCommands = mkOption {
      default = "";
      example = "text=anything; echo You can put $text here.";
      description = ''
        Shell commands to be executed at the end of the
        <literal>network-setup</literal> systemd service.  Note that if
        you are using DHCP to obtain the network configuration,
        interfaces may not be fully configured yet.
      '';
    };

    networking.interfaces = mkOption {
      default = {};
      example =
        { eth0 = {
            ipAddress = "131.211.84.78";
            subnetMask = "255.255.255.128";
          };
        };
      description = ''
        The configuration for each network interface.  If
        <option>networking.useDHCP</option> is true, then every
        interface not listed here will be configured using DHCP.
      '';
      type = types.loaOf types.optionSet;
      options = [ interfaceOpts ];
    };

    networking.bridges = mkOption {
      default = { };
      example =
        { br0.interfaces = [ "eth0" "eth1" ];
          br1.interfaces = [ "eth2" "wlan0" ];
        };
      description =
        ''
          This option allows you to define Ethernet bridge devices
          that connect physical networks together.  The value of this
          option is an attribute set.  Each attribute specifies a
          bridge, with the attribute name specifying the name of the
          bridge's network interface.
        '';

      type = types.attrsOf types.optionSet;

      options = {

        interfaces = mkOption {
          example = [ "eth0" "eth1" ];
          type = types.listOf types.string;
          description =
            "The physical network interfaces connected by the bridge.";
        };

      };

    };

    networking.bonds = mkOption {
      default = { };
      example = {
        bond0 = {
          interfaces = [ "eth0" "wlan0" ];
          miimon = 100;
          mode = "active-backup";
        };
        fatpipe.interfaces = [ "enp4s0f0" "enp4s0f1" "enp5s0f0" "enp5s0f1" ];
      };
      description = ''
        This option allows you to define bond devices that aggregate multiple,
        underlying networking interfaces together. The value of this option is
        an attribute set. Each attribute specifies a bond, with the attribute
        name specifying the name of the bond's network interface
      '';

      type = types.attrsOf types.optionSet;

      options = {

        interfaces = mkOption {
          example = [ "enp4s0f0" "enp4s0f1" "wlan0" ];
          type = types.listOf types.string;
          description = "The interfaces to bond together";
        };

        miimon = mkOption {
          default = null;
          example = 100;
          type = types.nullOr types.int;
          description = ''
            Miimon is the number of millisecond in between each round of polling
            by the device driver for failed links. By default polling is not
            enabled and the driver is trusted to properly detect and handle
            failure scenarios.
          '';
        };

        mode = mkOption {
          default = null;
          example = "active-backup";
          type = types.nullOr types.string;
          description = ''
            The mode which the bond will be running. The default mode for
            the bonding driver is balance-rr, optimizing for throughput.
            More information about valid modes can be found at
            https://www.kernel.org/doc/Documentation/networking/bonding.txt
          '';
        };

      };
    };

    networking.vlans = mkOption {
      default = { };
      example = {
        vlan0 = {
          id = 3;
          interface = "enp3s0";
        };
        vlan1 = {
          id = 1;
          interface = "wlan0";
        };
      };
      description =
        ''
          This option allows you to define vlan devices that tag packets
          on top of a physical interface. The value of this option is an
          attribute set. Each attribute specifies a vlan, with the name
          specifying the name of the vlan interface.
        '';

      type = types.attrsOf types.optionSet;

      options = {

        id = mkOption {
          example = 1;
          type = types.int;
          description = "The vlan identifier";
        };

        interface = mkOption {
          example = "enp4s0";
          type = types.string;
          description = "The interface the vlan will transmit packets through.";
        };

      };
    };

    networking.useDHCP = mkOption {
      type = types.bool;
      default = true;
      description = ''
        Whether to use DHCP to obtain an IP address and other
        configuration for all network interfaces that are not manually
        configured.
      '';
    };

  };


  ###### implementation

  config = {

    boot.kernelModules = [ ]
      ++ optional cfg.enableIPv6 "ipv6"
      ++ optional hasVirtuals "tun"
      ++ optional hasBonds "bonding";

    boot.extraModprobeConfig =
      # This setting is intentional as it prevents default bond devices
      # from being created.
      optionalString hasBonds "options bonding max_bonds=0";

    environment.systemPackages =
      [ pkgs.host
        pkgs.iproute
        pkgs.iputils
        pkgs.nettools
        pkgs.wirelesstools
        pkgs.iw
        pkgs.rfkill
        pkgs.openresolv
      ]
      ++ optional (cfg.bridges != {}) pkgs.bridge_utils
      ++ optional hasVirtuals pkgs.tunctl
      ++ optional cfg.enableIPv6 pkgs.ndisc6;

    security.setuidPrograms = [ "ping" "ping6" ];

    systemd.targets."network-interfaces" =
      { description = "All Network Interfaces";
        wantedBy = [ "network.target" ];
        unitConfig.X-StopOnReconfiguration = true;
      };

    systemd.services =
      let

        networkSetup =
          { description = "Networking Setup";

            after = [ "network-interfaces.target" ];
            before = [ "network.target" ];
            wantedBy = [ "network.target" ];

            unitConfig.ConditionCapability = "CAP_NET_ADMIN";

            path = [ pkgs.iproute ];

            serviceConfig.Type = "oneshot";
            serviceConfig.RemainAfterExit = true;

            script =
              ''
                # Set the static DNS configuration, if given.
                ${pkgs.openresolv}/sbin/resolvconf -m 1 -a static <<EOF
                ${optionalString (cfg.nameservers != [] && cfg.domain != "") ''
                  domain ${cfg.domain}
                ''}
                ${flip concatMapStrings cfg.nameservers (ns: ''
                  nameserver ${ns}
                '')}
                EOF

                # Disable or enable IPv6.
                ${optionalString (!config.boot.isContainer) ''
                  if [ -e /proc/sys/net/ipv6/conf/all/disable_ipv6 ]; then
                    echo ${if cfg.enableIPv6 then "0" else "1"} > /proc/sys/net/ipv6/conf/all/disable_ipv6
                  fi
                ''}

                # Set the default gateway.
                ${optionalString (cfg.defaultGateway != "") ''
                  # FIXME: get rid of "|| true" (necessary to make it idempotent).
                  ip route add default via "${cfg.defaultGateway}" ${
                    optionalString (cfg.defaultGatewayWindowSize != null)
                      "window ${cfg.defaultGatewayWindowSize}"} || true
                ''}

                # Turn on forwarding if any interface has enabled proxy_arp.
                ${optionalString (any (i: i.proxyARP) interfaces) ''
                  echo 1 > /proc/sys/net/ipv4/ip_forward
                ''}

                # Run any user-specified commands.
                ${cfg.localCommands}
              '';
          };

        # For each interface <foo>, create a job ‘<foo>-cfg.service"
        # that performs static configuration.  It has a "wants"
        # dependency on ‘<foo>.service’, which is supposed to create
        # the interface and need not exist (i.e. for hardware
        # interfaces).  It has a binds-to dependency on the actual
        # network device, so it only gets started after the interface
        # has appeared, and it's stopped when the interface
        # disappears.
        configureInterface = i: nameValuePair "${i.name}-cfg"
          (let mask =
                if i.prefixLength != null then toString i.prefixLength else
                if i.subnetMask != "" then i.subnetMask else "32";
               staticIPv6 = cfg.enableIPv6 && i.ipv6Address != null;
          in
          { description = "Configuration of ${i.name}";
            wantedBy = [ "network-interfaces.target" ];
            bindsTo = [ "sys-subsystem-net-devices-${i.name}.device" ];
            after = [ "sys-subsystem-net-devices-${i.name}.device" ];
            serviceConfig.Type = "oneshot";
            serviceConfig.RemainAfterExit = true;
            path = [ pkgs.iproute pkgs.gawk ];
            script =
              ''
                echo "bringing up interface..."
                ip link set "${i.name}" up
              ''
              + optionalString (i.macAddress != null)
                ''
                  echo "setting MAC address to ${i.macAddress}..."
                  ip link set "${i.name}" address "${i.macAddress}"
                ''
              + optionalString (i.mtu != null)
                ''
                  echo "setting MTU to ${toString i.mtu}..."
                  ip link set "${i.name}" mtu "${toString i.mtu}"
                ''
              + optionalString (i.ipAddress != null)
                ''
                  cur=$(ip -4 -o a show dev "${i.name}" | awk '{print $4}')
                  # Only do a flush/add if it's necessary.  This is
                  # useful when the Nix store is accessed via this
                  # interface (e.g. in a QEMU VM test).
                  if [ "$cur" != "${i.ipAddress}/${mask}" ]; then
                    echo "configuring interface..."
                    ip -4 addr flush dev "${i.name}"
                    ip -4 addr add "${i.ipAddress}/${mask}" dev "${i.name}"
                    restart_network_setup=true
                  else
                    echo "skipping configuring interface"
                  fi
                ''
              + optionalString (staticIPv6)
                ''
                  # Only do a flush/add if it's necessary.  This is
                  # useful when the Nix store is accessed via this
                  # interface (e.g. in a QEMU VM test).
                  if ! ip -6 -o a show dev "${i.name}" | grep "${i.ipv6Address}/${toString i.ipv6prefixLength}"; then
                    echo "configuring interface..."
                    ip -6 addr flush dev "${i.name}"
                    ip -6 addr add "${i.ipv6Address}/${toString i.ipv6prefixLength}" dev "${i.name}"
                    restart_network_setup=true
                  else
                    echo "skipping configuring interface"
                  fi
                ''
              + optionalString (i.ipAddress != null || staticIPv6)
                ''
                  if [ restart_network_setup = true ]; then
                    # Ensure that the default gateway remains set.
                    # (Flushing this interface may have removed it.)
                    ${config.systemd.package}/bin/systemctl try-restart --no-block network-setup.service
                  fi
                  ${config.systemd.package}/bin/systemctl start ip-up.target
                ''
              + optionalString i.proxyARP
                ''
                  echo 1 > /proc/sys/net/ipv4/conf/${i.name}/proxy_arp
                ''
              + optionalString (i.proxyARP && cfg.enableIPv6)
                ''
                  echo 1 > /proc/sys/net/ipv6/conf/${i.name}/proxy_ndp
                '';
          });

        createTunDevice = i: nameValuePair "${i.name}"
          { description = "Virtual Network Interface ${i.name}";
            requires = [ "dev-net-tun.device" ];
            after = [ "dev-net-tun.device" ];
            wantedBy = [ "network.target" ];
            requiredBy = [ "sys-subsystem-net-devices-${i.name}.device" ];
            serviceConfig =
              { Type = "oneshot";
                RemainAfterExit = true;
                ExecStart = "${pkgs.tunctl}/bin/tunctl -t '${i.name}' -u '${i.virtualOwner}'";
                ExecStop = "${pkgs.tunctl}/bin/tunctl -d '${i.name}'";
              };
          };

        createBridgeDevice = n: v:
          let
            deps = map (i: "sys-subsystem-net-devices-${i}.device") v.interfaces;
          in
          { description = "Bridge Interface ${n}";
            wantedBy = [ "network.target" "sys-subsystem-net-devices-${n}.device" ];
            bindsTo = deps;
            after = deps;
            serviceConfig.Type = "oneshot";
            serviceConfig.RemainAfterExit = true;
            path = [ pkgs.bridge_utils pkgs.iproute ];
            script =
              ''
                # Remove Dead Interfaces
                ip link show "${n}" >/dev/null 2>&1 && ip link delete "${n}"

                brctl addbr "${n}"

                # Set bridge's hello time to 0 to avoid startup delays.
                brctl setfd "${n}" 0

                ${flip concatMapStrings v.interfaces (i: ''
                  brctl addif "${n}" "${i}"
                  ip link set "${i}" up
                  ip addr flush dev "${i}"

                  echo "bringing up network device ${n}..."
                  ip link set "${n}" up
                '')}

                # !!! Should delete (brctl delif) any interfaces that
                # no longer belong to the bridge.
              '';
            postStop =
              ''
                ip link set "${n}" down
                brctl delbr "${n}"
              '';
          };

        createBondDevice = n: v:
          let
            deps = map (i: "sys-subsystem-net-devices-${i}.device") v.interfaces;
          in
          { description = "Bond Interface ${n}";
            wantedBy = [ "network.target" "sys-subsystem-net-devices-${n}.device" ];
            bindsTo = deps;
            after = deps;
            serviceConfig.Type = "oneshot";
            serviceConfig.RemainAfterExit = true;
            path = [ pkgs.ifenslave pkgs.iproute ];
            script = ''
              # Remove Dead Interfaces
              ip link show "${n}" >/dev/null 2>&1 && ip link delete "${n}"

              ip link add "${n}" type bond

              # !!! There must be a better way to wait for the interface
              while [ ! -d /sys/class/net/${n} ]; do sleep 0.1; done;

              # Set the miimon and mode options
              ${optionalString (v.miimon != null)
                "echo ${toString v.miimon} > /sys/class/net/${n}/bonding/miimon"}
              ${optionalString (v.mode != null)
                "echo \"${v.mode}\" > /sys/class/net/${n}/bonding/mode"}

              # Bring up the bridge and enslave the specified interfaces
              ip link set "${n}" up
              ${flip concatMapStrings v.interfaces (i: ''
                ifenslave "${n}" "${i}"
              '')}
            '';
            postStop = ''
              ip link set "${n}" down
              ifenslave -d "${n}"
              ip link delete "${n}"
            '';
          };

        createVlanDevice = n: v:
          let
            deps = [ "sys-subsystem-net-devices-${v.interface}.device" ];
          in
          { description = "Vlan Interface ${n}";
            wantedBy = [ "network.target" "sys-subsystem-net-devices-${n}.device" ];
            bindsTo = deps;
            after = deps;
            serviceConfig.Type = "oneshot";
            serviceConfig.RemainAfterExit = true;
            path = [ pkgs.iproute ];
            script = ''
              # Remove Dead Interfaces
              ip link show "${n}" >/dev/null 2>&1 && ip link delete "${n}"
              ip link add link "${v.interface}" "${n}" type vlan id "${toString v.id}"
              ip link set "${n}" up
            '';
            postStop = ''
              ip link delete "${n}"
            '';
          };

      in listToAttrs (
           map configureInterface interfaces ++
           map createTunDevice (filter (i: i.virtual) interfaces))
         // mapAttrs createBridgeDevice cfg.bridges
         // mapAttrs createBondDevice cfg.bonds
         // mapAttrs createVlanDevice cfg.vlans
         // { "network-setup" = networkSetup; };

    # Set the host and domain names in the activation script.  Don't
    # clear it if it's not configured in the NixOS configuration,
    # since it may have been set by dhcpcd in the meantime.
    system.activationScripts.hostname =
      optionalString (config.networking.hostName != "") ''
        hostname "${config.networking.hostName}"
      '';
    system.activationScripts.domain =
      optionalString (config.networking.domain != "") ''
        domainname "${config.networking.domain}"
      '';

    services.udev.extraRules =
      ''
        KERNEL=="tun", TAG+="systemd"
      '';

  };

}