NOTE: If you've been using
vintage_net
v0.6.x
or earlier, we split out network technology support out to separate libraries inv0.7.0
. You'll need to add those libraries to yourmix
dependency list and rename some atoms. Configurations stored on deployed devices will be automatically updated. See the v0.7.0 release notes for details.
VintageNet
is network configuration library built specifically for Nerves
Project devices. It has the following features:
- Ethernet and WiFi support included. Extendible to other technologies
- Default configurations specified in your Application config
- Runtime updates to configurations are persisted and applied on next boot (configurations are obfuscated by default to hide WiFi passphrases)
- Simple subscription to network status change events
- Connect to multiple networks at a time and prioritize which interfaces are used (Ethernet over WiFi over cellular)
- Internet connection monitoring and failure detection
- Predictable network interface names
TL;DR: Don't care about any of this and just want the string to copy/paste to set up networking? See the VintageNet Cookbook.
The following network configurations are supported:
- Wired Ethernet, IPv4 DHCP
- Wired Ethernet, IPv4 static IP
- WiFi password-less and WEP
- WPA2 PSK and EAP
- USB gadget mode Ethernet, IPv4 DHCP server to supply host IP address
- Cellular networks (see
vintage_net_qmi
andvintage_net_mobile
for details) - WiFi AP mode
- IPv6 - Partially supported. SLAAC configuration works.
vintage_net
takes a different approach to networking from nerves_network
.
Its focus is on building and applying network configurations. Where
nerves_network
provided configurable state machines, vintage_net
turns
human-readable configurations into everything from configuration files and calls
to ip
to starting up networking GenServers
and routing table updates. This makes it easier to add support for new network
technologies and features. While Elixir and Erlang were great to implement
network protocols in, it was frequently more practical to reuse embedded Linux
implementations. Importantly, though, vintage_net
monitors Linux daemons under
its OTP supervision tree so failures on both the "C" and Elixir sides propagate
in the expected ways.
Another important difference is that VintageNet
doesn't attempt to make
incremental modifications to configurations. It completely tears down an
interface's connection and then brings up new configurations in a fresh state.
Network reconfiguration is assumed to be an infrequent event so while this can
cause a hiccup in the network connectivity, it removes state machine code that
made nerves_network
hard to maintain.
First, if you're modifying an existing project, you will need to remove
nerves_network
and nerves_init_gadget
. vintage_net
doesn't work with
either of them. You'll get an error if any project references those packages.
There are two routes to integrating vintage_net
:
- Use nerves_pack.
nerves_pack
is likenerves_init_gadget
, but forvintage_net
. - Copy and paste from the Nerves hello WiFi example
The next step is to make sure that your Nerves system is compatible. The official Nerves systems released after 12/11/2019 work without modification. If rolling your own Nerves port, you will need the following Linux kernel options enabled:
CONFIG_IP_ADVANCED_ROUTER=y
CONFIG_IP_MULTIPLE_TABLES=y
Then make sure that you have the following Busybox options enabled:
CONFIG_IFCONFIG=y
-ifconfig
ifconfigCONFIG_UDHCPC=y
-udhcpc
DHCP ClientCONFIG_UDHCPD=y
-udhcpd
DHCP Server (optional)
Finally, you'll need to choose what network connection technologies that you
want available in your firmware. If using nerves_pack
, you'll get support for
wired Ethernet, WiFi, and USB gadget networking automatically. Otherwise, add
one or more of the following to your dependency list:
vintage_net_ethernet
- Standard wired Ethernetvintage_net_wifi
- Client configurations for 802.11 WiFivintage_net_direct
- Direct connections like those used for USB gadgetvintage_net_qmi
- Support USB-connected cellular modemsvintage_net_mobile
- Support UART-connected cellular modems
VintageNet
has many application configuration keys. Most defaults are fine. At
a minimum, you'll want to specify a default configuration and default regulatory
domain if using WiFi. In your main config.exs
, add the following:
config :vintage_net,
regulatory_domain: "US",
config: [
{"eth0", %{type: VintageNetEthernet, ipv4: %{method: :dhcp}}},
{"wlan0", %{type: VintageNetWiFi}}
]
This sets the regulatory domain to the US (set to your ISO 3166-1 alpha-2 country code. This code is passed on to the drivers for WiFi and other wireless networking technologies so that they comply with local regulations. If you need a global default, set to "00" or don't set at all. Unfortunately, this may mean that an access point isn't visible if it is running on a frequency that's allowed in your country, but not globally.
The config
section is a list of network configurations. The one shown above
configures DHCP on wired Ethernet and minimally starts up a WiFi LAN so that
it's possible to scan for networks. The typical setup is to provide generic
defaults here. Static IP addresses, WiFi SSIDs and credentials are more
appropriately configured at run-time. VintageNet
persists configurations too.
Details on network configuration are described later.
The following table describes the other application config keys.
Key | Description |
---|---|
config | A list of default network configurations |
tmpdir | Path to a temporary directory for VintageNet |
udhcpc_handler | Module for handling notifications from udhcpc |
resolvconf | Path to /etc/resolv.conf |
persistence | Module for persisting network configurations |
persistence_dir | Path to a directory for storing persisted configurations |
persistence_secret | A 16-byte secret or a function or MFArgs (module, function, arguments tuple) for getting a secret |
internet_host_list | IP address or hostnames and ports to try to connect to for checking Internet connectivity. Defaults to a list of large public DNS providers. E.g., [{{1, 1, 1, 1}, 53}] . |
regulatory_domain | ISO 3166-1 alpha-2 country (00 for global, US , etc.) |
additional_name_servers | List of DNS servers to be used in addition to any supplied by an interface. E.g., [{1, 1, 1, 1}, {8, 8, 8, 8}] |
route_metric_fun | Customize how network interfaces are prioritized. See VintageNet.Route.DefaultMetric.compute_metric/2 |
VintageNet
supports several network technologies out of the box and
third-party libraries can provide more via the VintageNet.Technology
behaviour.
Configurations are Elixir maps. These are specified in three places:
- The
vintage_net
application config (e.g., yourconfig.exs
) - Locally saved configuration (see the
VintageNet.Persistence
behaviour for replacing the default) - Calling
VintageNet.configure/2
to change the configuration at run-time
When vintage_net
starts, it applies saved configurations first and if any
thing is wrong with those configs, it reverts to the application config. A good
practice is to have safe defaults for all network interfaces in the application
config.
The only required key in the configuration maps is :type
. All other keys
follow from the type. :type
should be set to a module that implements the
VintageNet.Technology
behaviour. The following are common technologies:
VintageNetEthernet
- Standard wired EthernetVintageNetWiFi
- Client configurations for 802.11 WiFiVintageNetDirect
- Direct connections like those used for USB gadget connectionsVintageNet.Technology.Null
- An empty configuration useful for turning off a configuration
See the links above for specific documentation.
By default, VintageNet stores network configuration to disk. If you are
migrating from nerves_network
you may already have a persistence
implementation. To disable the default persistence, configure vintage_net
as
follows:
config :vintage_net,
persistence: VintageNet.Persistence.Null
Debugging networking issues is not fun. When you're starting out with
vintage_net
, it is highly recommended to connect to your target using a method
that doesn't require networking to work. This could be a UART connection to an
IEx console on a Nerves device or maybe just hooking up a keyboard and monitor.
If having trouble, first check VintageNet.info()
to verify the configuration
and connection status:
iex> VintageNet.info
VintageNet 0.3.0
All interfaces: ["eth0", "lo", "tap0", "wlan0"]
Available interfaces: ["eth0", "wlan0"]
Interface eth0
Type: VintageNetEthernet
Present: true
State: :configured
Connection: :internet
Configuration:
%{ipv4: %{method: :dhcp}, type: VintageNetEthernet}
Interface wlan0
Type: VintageNetWiFi
Present: true
State: :configured
Connection: :internet
Configuration:
%{
ipv4: %{method: :dhcp},
type: VintageNetWiFi,
wifi: %{
key_mgmt: :wpa_psk,
mode: :infrastructure,
psk: "******",
ssid: "MyLAN"
}
}
If you're using Toolshed, try running the following:
iex> ifconfig
lo: flags=[:up, :loopback, :running]
inet 127.0.0.1 netmask 255.0.0.0
inet ::1 netmask ffff:ffff:ffff:ffff:ffff:ffff:ffff:ffff
hwaddr 00:00:00:00:00:00
eth0: flags=[:up, :broadcast, :running, :multicast]
inet 192.168.9.131 netmask 255.255.255.0 broadcast 192.168.9.255
inet fe80::6264:5ff:fee1:4045 netmask ffff:ffff:ffff:ffff::
hwaddr 60:64:05:e1:40:45
wlan0: flags=[:up, :broadcast, :running, :multicast]
inet 192.168.9.175 netmask 255.255.255.0 broadcast 192.168.9.255
inet fe80::20c:e7ff:fe11:3d46 netmask ffff:ffff:ffff:ffff::
hwaddr 00:0c:e7:11:3d:46
Or ping:
iex> ping "nerves-project.com"
Press enter to stop
Response from nerves-project.com (96.126.123.244): time=48.87ms
Response from nerves-project.com (96.126.123.244): time=42.856ms
Response from nerves-project.com (96.126.123.244): time=43.097ms
You can also specify an interface to use with ping
:
iex> ping "nerves-project.com", ifname: "wlan0"
Press enter to stop
Response from nerves-project.com (96.126.123.244): time=57.817ms
Response from nerves-project.com (96.126.123.244): time=46.796ms
iex> ping "nerves-project.com", ifname: "eth0"
Press enter to stop
Response from nerves-project.com (96.126.123.244): time=47.923ms
Response from nerves-project.com (96.126.123.244): time=48.688ms
If it looks like nothing is working, check the logs. On Nerves devices, this
is frequently done by calling RingLogger.next
or RingLogger.attach
.
At a last resort, please open a GitHub issue. We would be glad to help. We only have one ask and that is that you get us started with an improvement to our documentation or code so that the next person to run into the issue will have an easier time. Thanks!
VintageNet
maintains a key/value store for retrieving information on
networking information:
iex> VintageNet.get(["interface", "eth0", "connection"])
:internet
iex> VintageNet.get_by_prefix([])
[
{["interface", "eth0", "connection"], :internet},
{["interface", "eth0", "state"], :configured},
{["interface", "eth0", "type"], VintageNetEthernet},
{["interface", "wlan0", "connection"], :internet},
{["interface", "wlan0", "state"], :configured},
{["interface", "wlan0", "type"], VintageNetWiFi}
]
You can also subscribe to keys and receive a message every time it or one its child keys changes:
iex> VintageNet.subscribe(["interface", "eth0"])
:ok
iex> flush
{VintageNet, ["interface", "eth0", "state"], :configuring, :configured, %{old_timestamp: 123456, new_timestamp: 124456}}
The message format is {VintageNet, name, old_value, new_value, metadata}
Metadata fields include at least the following:
:old_timestamp
- the timestamp for when the property first had theold_value
:new_timestamp
- the timestamp when the property changed from theold_value
to thenew_value
Timestamps are all captured by calling System.monotonic_time/0
so they're in :native
time units.
Use System.convert_time_unit/3
to convert to the usual time units.
Property | Values | Description |
---|---|---|
available_interfaces |
[eth0, ...] |
Currently available network interfaces in priority order. E.g., the first one is used by default |
connection |
:disconnected , :lan , :internet |
The overall network connection status. This is the best status of all interfaces. |
name_servers |
[%{address: ..., from: []}] |
Name server addresses and where VintageNet learned about them |
All network interface properties can be found under ["interface", ifname]
in
the PropertyTable
. The following table lists out properties common to all
interfaces:
Property | Values | Description |
---|---|---|
type |
VintageNetEthernet , etc. |
The type of the interface |
config |
%{...} |
The configuration for this interface |
state |
:configured , :configuring , etc. |
The state of the interface from VintageNet 's point of view. |
hw_path |
"/devices/platform/ocp/4a100000.ethernet" |
This is how Linux internally views the connections going to the interface. |
connection |
:disconnected , :lan , :internet |
This provides a determination of the Internet connection status |
lower_up |
true or false |
This indicates whether the physical layer is "up". E.g., a cable is connected or WiFi associated |
mac_address |
"11:22:33:44:55:66" | The interface's MAC address as a string |
addresses |
[address_info] | This is a list of all of the addresses assigned to this interface |
dhcp_options |
%{...} |
When DHCP is in use, the processed response information and options is stored here. See t:VintageNet.DHCP.Options.t/0 |
Specific types of interfaces provide more parameters.
When using more than one of the same type of interface, it's possible for Linux
to reorder their naming. For example, if you have two USB WiFi adapters, one
will be named wlan0
and the other wlan1
. Which one is first depends on
things like when the adapter is found and when kernel modules are loaded. This
can vary between boots and cause a lot of confusion.
The solution is to rename network interfaces based on characteristics of the
interface - such as how it's connected. Then application software refers to the
new name rather than names like wlan0
. This is a common problem, and
VintageNet provides support for automatically renaming network interfaces.
If you're used to systemd
's approach to naming interfaces, be aware that
VintageNet's approach is different: systemd
has an
algorithm
for generating names (e.g., enp4s0
) automatically. VintageNet requires you to
provide the names to use (e.q., internet0
, lan0
, etc.) and how they map to
hardware. If VintageNet is confronted with a network interface that is connected
in a way that it doesn't know about, it will do nothing.
IMPORTANT: Do not mix and match predictable interface names and non-predictable interface names (
wlan*
,eth*
,usb*
,wwan*
). It is confusing and VintageNet will fight you.
Before switching to predictable names, find out how your network interfaces are connected. For example, this device has an Ethernet interface and two USB WiFi dongles:
iex> VintageNet.match(["interface", :_, "hw_path"])
[
{["interface", "eth0", "hw_path"], "/devices/platform/ocp/4a100000.ethernet"},
{["interface", "lo", "hw_path"], "/devices/virtual"},
{["interface", "wlan0", "hw_path"], "/devices/platform/ocp/47400000.usb/47401c00.usb/musb-hdrc.1/usb2/2-1/2-1:1.0"},
{["interface", "wlan1", "hw_path"], "/devices/platform/ocp/47400000.usb/47401400.usb/musb-hdrc.0/usb1/1-1/1-1:1.4"}
]
Now update your config.exs
with the mappings with the :ifnames
key. Be sure
to also update the default configuration with the new interface names.
Continuing the example, imagine that one WiFi adapter supports 802.11 meshing
and it's guaranteed to be in one USB port on the device. The other USB port can
have any of a few types of USB WiFi modules. We need to use predictable naming
in this case so that meshing is only setup on the adapter that supports it.
config :vintage_net,
ifnames: [
%{
hw_path: "/devices/platform/ocp/4a100000.ethernet",
ifname: "ethernet0"
},
%{
hw_path: "/devices/platform/ocp/47400000.usb/47401c00.usb/musb-hdrc.1/usb2/2-1/2-1:1.0",
ifname: "primary_wifi"
},
%{
hw_path: "/devices/platform/ocp/47400000.usb/47401400.usb/musb-hdrc.0/usb1/1-1/1-1:1.4",
ifname: "mesh_wifi"
}
],
config: [
{"ethernet0", %{type: VintageNetEthernet}},
{"primary_wifi", %{type: VintageNetWiFi}},
{"mesh_wifi", %{type: VintageNetWiFi}}
]
IMPORTANT: VintageNet has rules about renaming interfaces to prevent confusing errors. Below is a list of reasons it will not rename an interface
hw_path
matches/devices/virtual
(such aslo0
,ppp0
etc.)- A second interface's
hw_path
matches an interface that has already been renamed. This should never happen.
VintageNet can check whether a network interface can reach the Internet. This has a few uses:
- Selecting which network interface is used when a device has more than one. A common example is a device with a backup cellular connection.
- Automatically recovering a network interface that has lost connectivity. Some times bouncing the network interface actually works, so doing this automatically can sometimes revive a remote device.
- Letting the application know the status of the network connection to provide more helpful information about what's happening.
The logic for declaring that the Internet is available is:
- Is there a TCP socket in use on the network interface that has sent and received data from a host that's not on the same subnet? If yes, then the device is Internet-connected.
- Get the list of Internet servers to check. See below for the list.
- Resolve any domain names in the list. If DNS isn't working, remove them from the list.
- Pick a random IP address from the remaining list and "ping" it. Technically, VintageNet tries to connect over TCP to a specified port, and if it either connects successfully or gets a port closed response, then the device is Internet-connected.
- Wait a bit and then go back to step 1.
The list of Internet servers to check is critically important. VintageNet uses
the :internet_host_list
key in the application environment for this. The
default setting has many popular name servers in it. The idea being that if you
can't reach a name server, the Internet probably isn't going to work well.
If you are deploying to locations with locked down networks, you'll find that the default setting to test name servers won't work. It is not uncommon to find a network that blocks popular name servers like 8.8.8.8.
The recommendation is to set the :internet_host_list
to include your backend
server. If VintageNet can reach it, then presumably your application works and
having VintageNet declare the internet reachable via that network interface is
correct.
For example,
config :vintage_net,
internet_host_list: [{"abcdefghijk-ats.iot.us-east-1.amazonaws.com", 443}]
The use of the connectivity checker is specified by the technology. Both the
VintageNetEthernet
and VintageNetWiFi
use the internet connectivity checker.
This is selected by adding the VintageNet.Connectivity.InternetChecker
GenServer to the :child_specs
configuration returned by the technology. E.g.,
child_specs: [{VintageNet.Connectivity.InternetChecker, "eth0"}]
. Most users
do not need to be concerned about this.
Some devices require additional work to be done for them to become available. Examples of this are:
- Setting a GPIO to enable power to the module
- Loading a Linux kernel module that is not automatically loaded via the default mechanisms
- Running
usb_modeswitch
to change the USB interface to the appropriate state - Performing an initialization step such as loading firmware
Similarly, when the network interface is no longer being used, it can be nice to undo any steps above.
This process is referred to as power management in VintageNet even though the
implementation may not actually affect power use. To use it, implement the
VintageNet.PowerManager
behaviour and register the implementation in your
config.exs
.
Additionally, VintageNet runs a watchdog-like service for network devices that
supply VintageNet.PowerManager
implementations. If the watchdog is not pet
within the timeout period (user-specified and defaults to 60 seconds),
VintageNet powers the device off and and on. The VintageNet power management
code supports mandatory minimum on and off times to prevent damage to hardware
and also minimize pointless power cycling of hardware.
While many network devices are fairly reliable and powering off and on seems unnecessary, it can save a trip to the field or a full device reboot.
VintageNet.info/1
shows the power management state for network interfaces that
are using this feature.