Introduction

Distribution installers, cloud instantiation, image builds for particular devices, or any other way to deploy an operating system put its desired network configuration into YAML configuration file(s). During early boot, the netplan “network renderer” runs which reads /{lib,etc,run}/netplan/*.yaml and writes configuration to /run to hand off control of devices to the specified networking daemon.

•

Configured devices get handled by systemd-networkd by default, unless explicitly marked as managed by a specific renderer (NetworkManager)

•

Devices not covered by the network config do not get touched at all.

•

Usable in initramfs (few dependencies and fast)

•

No persistent generated config, only original YAML config

•

Parser supports multiple config files to allow applications like libvirt or lxd to package up expected network config (virbr0, lxdbr0), or to change the global default policy to use NetworkManager for everything.

•

Retains the flexibility to change backends/policy later or adjust to removing NetworkManager, as generated configuration is ephemeral.

General structure

netplan's configuration files use the YAML (http://yaml.org/spec/1.1/current.html) format. All /{lib,etc,run}/netplan/*.yaml are considered. Lexicographically later files (regardless of in which directory they are) amend (new mapping keys) or override (same mapping keys) previous ones. A file in /run/netplan completely shadows a file with same name in /etc/netplan, and a file in either of those directories shadows a file with the same name in /lib/netplan.

The top-level node in a netplan configuration file is a network: mapping that contains version: 2 (the YAML currently being used by curtin, MaaS, etc. is version 1), and then device definitions grouped by their type, such as ethernets:, wifis:, or bridges:. These are the types that our renderer can understand and are supported by our backends.

Each type block contains device definitions as a map where the keys (called “configuration IDs”) are defined as below.

Device configuration IDs

The key names below the per-device-type definition maps (like ethernets:) are called “ID”s. They must be unique throughout the entire set of configuration files. Their primary purpose is to serve as anchor names for composite devices, for example to enumerate the members of a bridge that is currently being defined.

There are two physically/structurally different classes of device definitions, and the ID field has a different interpretation for each:

Physical devices

(Examples: ethernet, wifi) These can dynamically come and go between reboots and even during runtime (hotplugging). In the generic case, they can be selected by match: rules on desired properties, such as name/name pattern, MAC address, driver, or device paths. In general these will match any number of devices (unless they refer to properties which are unique such as the full path or MAC address), so without further knowledge about the hardware these will always be considered as a group.

Virtual devices

(Examples: veth, bridge, bond) These are fully under the control of the config file(s) and the network stack. I. e. these devices are being created instead of matched. Thus match: and set-name: are not applicable for these, and the ID field is the name of the created virtual device.

Common properties for physical device types

match (mapping)

This selects a subset of available physical devices by various hardware properties. The following configuration will then apply to all matching devices, as soon as they appear. All specified properties must match.

set-name (scalar)

When matching on unique properties such as path or MAC, or with additional assumptions such as “there will only ever be one wifi device”, match rules can be written so that they only match one device. Then this property can be used to give that device a more specific/desirable/nicer name than the default from udev's ifnames. Any additional device that satisfies the match rules will then fail to get renamed and keep the original kernel name (and dmesg will show an error).

wakeonlan (bool)

Enable wake on LAN. Off by default.

Common properties for all device types

renderer (scalar)

Use the given networking backend for this definition. Currently supported are networkd and NetworkManager. This property can be specified globally in networks:, for a device type (in e. g. ethernets:) or for a particular device definition. Default is networkd.

dhcp4 (bool)

Enable DHCP for IPv4. Off by default.

dhcp6 (bool)

Enable DHCP for IPv6. Off by default. This covers both stateless DHCP - where the DHCP server supplies information like DNS nameservers but not the IP address - and stateful DHCP, where the server provides both the address and the other information.

ipv6-privacy (bool)

Enable IPv6 Privacy Extensions (RFC 4941) for the specified interface, and prefer temporary addresses. Defaults to false - no privacy extensions. There is currently no way to have a private address but prefer the public address.

link-local (sequence of scalars)

Configure the link-local addresses to bring up. Valid options are `ipv4' and `ipv6', which respectively allow enabling IPv4 and IPv6 link local addressing. If this field is not defined, the default is to enable only IPv6 link-local addresses. If the field is defined but configured as an empty set, IPv6 link-local addresses are disabled as well as IPv4 link- local addresses.

critical (bool)

(networkd backend only) Designate the connection as “critical to the system”, meaning that special care will be taken by systemd-networkd to not release the IP from DHCP when it the daemon is restarted.

dhcp-identifier (scalar)

When set to `mac'; pass that setting over to systemd-networkd to use the device's MAC address as a unique identifier rather than a RFC4361-compliant Client ID. This has no effect when NetworkManager is used as a renderer.

dhcp4-overrides (mapping)

(networkd backend only) Overrides default DHCP behavior; see the DHCP Overrides section below.

dhcp6-overrides (mapping)

(networkd backend only) Overrides default DHCP behavior; see the DHCP Overrides section below.

accept-ra (bool)

Accept Router Advertisement that would have the kernel configure IPv6 by itself. When enabled, accept Router Advertisements. When disabled, do not respond to Router Advertisements. If unset use the host kernel default setting.

addresses (sequence of scalars)

Add static addresses to the interface in addition to the ones received through DHCP or RA. Each sequence entry is in CIDR notation, i. e. of the form addr/prefixlen. addr is an IPv4 or IPv6 address as recognized by inet_pton(3) and prefixlen the number of bits of the subnet.

gateway4, gateway6 (scalar)

Set default gateway for IPv4/6, for manual address configuration. This requires setting addresses too. Gateway IPs must be in a form recognized by inet_pton(3).

nameservers (mapping)

Set DNS servers and search domains, for manual address configuration. There are two supported fields: addresses: is a list of IPv4 or IPv6 addresses similar to gateway*, and search: is a list of search domains.

macaddress (scalar)

Set the device's MAC address. The MAC address must be in the form “XX:XX:XX:XX:XX:XX”.

mtu (scalar)

Set the Maximum Transmission Unit for the interface. The default is 1500. Valid values depend on your network interface.

optional (bool)

An optional device is not required for booting. Normally, networkd will wait some time for device to become configured before proceeding with booting. However, if a device is marked as optional, networkd will not wait for it. This is only supported by networkd, and the default is false.

optional-addresses (sequence of scalars)

Specify types of addresses that are not required for a device to be considered online. This changes the behavior of backends at boot time to avoid waiting for addresses that are marked optional, and thus consider the interface as “usable” sooner. This does not disable these addresses, which will be brought up anyway.

routes (mapping)

Configure static routing for the device; see the Routing section below.

routing-policy (mapping)

Configure policy routing for the device; see the Routing section below.

DHCP Overrides

Several DHCP behavior overrides are available. Most currently only have any effect when using the networkd backend, with the exception of use-routes and route-metric.

Overrides only have an effect if the corresponding dhcp4 or dhcp6 is set to true.

If both dhcp4 and dhcp6 are true, the networkd backend requires that dhcp4-overrides and dhcp6-overrides contain the same keys and values. If the values do not match, an error will be shown and the network configuration will not be applied.

When using the NetworkManager backend, different values may be specified for dhcp4-overrides and dhcp6-overrides, and will be applied to the DHCP client processes as specified in the netplan YAML.

: The dhcp4-overrides and dhcp6-overrides mappings override the default DHCP behavior.

 ``use-dns`` (bool)
 :    Default: ``true``. When ``true``, the DNS servers received from the
      DHCP server will be used and take precedence over any statically
      configured ones. Currently only has an effect on the ``networkd``
      backend.
 ``use-ntp`` (bool)
 :    Default: ``true``. When ``true``, the NTP servers received from the
      DHCP server will be used by systemd-timesyncd and take precedence
      over any statically configured ones. Currently only has an effect on
      the ``networkd`` backend.
 ``send-hostname`` (bool)
 :    Default: ``true``. When ``true``, the machine's hostname will be sent
      to the DHCP server. Currently only has an effect on the ``networkd``
      backend.
 ``use-hostname`` (bool)
 :    Default: ``true``. When ``true``, the hostname received from the DHCP
      server will be set as the transient hostname of the system. Currently
      only has an effect on the ``networkd`` backend.
 ``use-mtu`` (bool)
 :    Default: ``true``. When ``true``, the MTU received from the DHCP
      server will be set as the MTU of the network interface. When ``false``,
      the MTU advertised by the DHCP server will be ignored. Currently only
      has an effect on the ``networkd`` backend.
 ``hostname`` (scalar)
 :    Use this value for the hostname which is sent to the DHCP server,
      instead of machine's hostname. Currently only has an effect on the
      ``networkd`` backend.
 ``use-routes`` (bool)
 :    Default: ``true``. When ``true``, the routes received from the DHCP
      server will be installed in the routing table normally. When set to
      ``false``, routes from the DHCP server will be ignored: in this case,
      the user is responsible for adding static routes if necessary for
      correct network operation. This allows users to avoid installing a
      default gateway for interfaces configured via DHCP. Available for
      both the ``networkd`` and ``NetworkManager`` backends.
 ``route-metric`` (scalar)
 :    Use this value for default metric for automatically-added routes.
      Use this to prioritize routes for devices by setting a higher metric
      on a preferred interface. Available for both the ``networkd`` and
      ``NetworkManager`` backends.

Routing

Complex routing is possible with netplan. Standard static routes as well as policy routing using routing tables are supported via the networkd backend.

These options are available for all types of interfaces.

routes (mapping)

The routes block defines standard static routes for an interface. At least to and via must be specified.

routing-policy (mapping)

The routing-policy block defines extra routing policy for a network, where traffic may be handled specially based on the source IP, firewall marking, etc.

Authentication

Netplan supports advanced authentication settings for ethernet and wifi interfaces, as well as individual wifi networks, by means of the auth block.

auth (mapping)

Specifies authentication settings for a device of type ethernets:, or an access-points: entry on a wifis: device.

Properties for device type ethernets:

Ethernet device definitions do not support any specific properties beyond the common ones described above.

Properties for device type wifis:

Note that systemd-networkd does not natively support wifi, so you need wpasupplicant installed if you let the networkd renderer handle wifi.

access-points (mapping)

This provides pre-configured connections to NetworkManager. Note that users can of course select other access points/SSIDs. The keys of the mapping are the SSIDs, and the values are mappings with the following supported properties:

Properties for device type bridges:

interfaces (sequence of scalars)

All devices matching this ID list will be added to the bridge. This may be an empty list, in which case the bridge will be brought online with no member interfaces.

parameters (mapping)

Customization parameters for special bridging options. Time intervals may need to be expressed as a number of seconds or milliseconds: the default value type is specified below. If necessary, time intervals can be qualified using a time suffix (such as “s” for seconds, “ms” for milliseconds) to allow for more control over its behavior.

Properties for device type bonds:

interfaces (sequence of scalars)

All devices matching this ID list will be added to the bond.

parameters (mapping)

Customization parameters for special bonding options. Time intervals may need to be expressed as a number of seconds or milliseconds: the default value type is specified below. If necessary, time intervals can be qualified using a time suffix (such as “s” for seconds, “ms” for milliseconds) to allow for more control over its behavior.

Properties for device type tunnels:

Tunnels allow traffic to pass as if it was between systems on the same local network, although systems may be far from each other but reachable via the Internet. They may be used to support IPv6 traffic on a network where the ISP does not provide the service, or to extend and “connect” separate local networks. Please see https://en.wikipedia.org/wiki/Tunneling_protocol for more general information about tunnels.

mode (scalar)

Defines the tunnel mode. Valid options are sit, gre, ip6gre, ipip, ipip6, ip6ip6, vti, and vti6. Additionally, the networkd backend also supports gretap and ip6gretap modes. In addition, the NetworkManager backend supports isatap tunnels.

local (scalar)

Defines the address of the local endpoint of the tunnel.

remote (scalar)

Defines the address of the remote endpoint of the tunnel.

key (scalar or mapping)

Define keys to use for the tunnel. The key can be a number or a dotted quad (an IPv4 address). It is used for identification of IP transforms. This is only required for vti and vti6 when using the networkd backend, and for gre or ip6gre tunnels when using the NetworkManager backend.

Examples:

tunnels:
  tun0:
    mode: gre
    local: ...
    remote: ...
    keys:
      input: 1234
      output: 5678
tunnels:
  tun0:
    mode: vti6
    local: ...
    remote: ...
    key: 59568549

keys (scalar or mapping)

Alternate name for the key field. See above.

Properties for device type vlans:

id (scalar)

VLAN ID, a number between 0 and 4094.

link (scalar)

netplan ID of the underlying device definition on which this VLAN gets created.

Example:

ethernets:
  eno1: {...}
vlans:
  en-intra:
    id: 1
    link: eno1
    dhcp4: yes
  en-vpn:
    id: 2
    link: eno1
    address: ...

Examples

Configure an ethernet device with networkd, identified by its name, and enable DHCP:

network:
  version: 2
  ethernets:
    eno1:
      dhcp4: true

This is an example of a static-configured interface with multiple IPv4 addresses and multiple gateways with networkd, with equal route metric levels, and static DNS nameservers (Google DNS for this example):

network:
  version: 2
  renderer: networkd
  ethernets:
    eno1:
      addresses:
      - 10.0.0.10/24
      - 11.0.0.11/24
      nameservers:
        addresses:
          - 8.8.8.8
          - 8.8.4.4
      routes:
      - to: 0.0.0.0/0
        via: 10.0.0.1
        metric: 100
      - to: 0.0.0.0/0
        via: 11.0.0.1
        metric: 100

This is a complex example which shows most available features:

network:
  version: 2
  # if specified, can only realistically have that value, as networkd cannot
  # render wifi/3G.
  renderer: NetworkManager
  ethernets:
    # opaque ID for physical interfaces, only referred to by other stanzas
    id0:
      match:
        macaddress: 00:11:22:33:44:55
      wakeonlan: true
      dhcp4: true
      addresses:
        - 192.168.14.2/24
        - 192.168.14.3/24
        - "2001:1::1/64"
      gateway4: 192.168.14.1
      gateway6: "2001:1::2"
      nameservers:
        search: [foo.local, bar.local]
        addresses: [8.8.8.8]
      routes:
        - to: 0.0.0.0/0
          via: 11.0.0.1
          table: 70
          on-link: true
          metric: 3
      routing-policy:
        - to: 10.0.0.0/8
          from: 192.168.14.2/24
          table: 70
          priority: 100
        - to: 20.0.0.0/8
          from: 192.168.14.3/24
          table: 70
          priority: 50
      # only networkd can render on-link routes and routing policies
      renderer: networkd
    lom:
      match:
        driver: ixgbe
      # you are responsible for setting tight enough match rules
      # that only match one device if you use set-name
      set-name: lom1
      dhcp6: true
    switchports:
      # all cards on second PCI bus unconfigured by
      # themselves, will be added to br0 below
      # note: globbing is not supported by NetworkManager
      match:
        name: enp2*
      mtu: 1280
  wifis:
    all-wlans:
      # useful on a system where you know there is
      # only ever going to be one device
      match: {}
      access-points:
        "Joe's home":
          # mode defaults to "infrastructure" (client)
          password: "s3kr1t"
    # this creates an AP on wlp1s0 using hostapd
    # no match rules, thus the ID is the interface name
    wlp1s0:
      access-points:
        "guest":
           mode: ap
           # no WPA config implies default of open
  bridges:
    # the key name is the name for virtual (created) interfaces
    # no match: and set-name: allowed
    br0:
      # IDs of the components; switchports expands into multiple interfaces
      interfaces: [wlp1s0, switchports]
      dhcp4: true