Inspired by other projects that run hostapd
in a Docker container. This goes one step further and boots a full network OS intended for embedded devices called OpenWrt, so you can manage all aspects of your network from a user-friendly web UI.
For Raspberry Pi-specific instructions, see Building on Raspberry Pi.
- docker
- iw
- iproute2
- envsubst (part of
gettext
orgettext-base
package) - dhcpcd
Pre-built images are available on Docker Hub at oofnik/openwrt
. Alternatively, build the image yourself using the make build
target:
$ make build
If you want additional OpenWrt packages to be present in the base image, add them to the Dockerfile. Otherwise you can install them with opkg
after bringing up the container.
A searchable package list is available on openwrt.org.
Initial configuration is performed using a config file, openwrt.conf
. Values read from this file at runtime are used to generate OpenWrt format config files from templates in etc/config/*.tpl
.
You can use the included openwrt.conf.example
as a baseline, which explains the values.
It is also possible to make persistent changes in the UI and download a backup of your full router configuration by navigating to System > Backup / Flash Firmware and clicking Backup.
Prepare your openwrt.conf
file as explained above and execute the make run
target:
$ make run
If you arrive at * Ready
, point your browser to http://openwrt.home (or whatever you set in LAN_DOMAIN
) and you should be presented with the login page. The default login is root
with the password set as ROOT_PW
.
To shut down the router, press Ctrl+C
. Any settings you configured or additional packages you installed will persist until you run make clean
, which will delete the container.
$ make install
Install and uninstall targets for systemd
have been included in the Makefile.
Installing will create and enable a service pointing to wherever you cloned this directory and execute run.sh
on boot.
$ make clean
This will delete the container and all associated Docker networks so you can start fresh if you screw something up.
In order for WLAN clients to see one another, OpenWrt bridges all interfaces in the LAN zone and sets hairpin mode (aka reflective relay) on the WLAN interface, meaning packets arriving on that interface can be 'reflected' back out through the same interface.
run.sh
tries to handle this if WIFI_HAIRPIN
is set to true, and prints a warning if it fails.
Hairpin mode may not be needed in all cases, but if you experience an issue where Wi-Fi clients are unable to see each other despite AP isolation being disabled, this may fix it.
For hostapd
running inside the container to have access to the physical wireless device, we need to set the device's network namespace to the PID of the running container. This causes the interface to 'disappear' from the primary network namespace for the duration of the container's parent process. run.sh
checks if the host is using NetworkManager to manage the wifi interface, and tries to steal it away if so.
Additional containers that are run alongside OpenWrt on the same physical host are directly accessible on the LAN. No port forwarding to the host is necessary. It's recommended to add static hostnames to be able to resolve local services on your LAN.
See Monitoring with InfluxDB + Grafana for example.
Read the upgrade guide.
Logs are redirected to stdout
so the Docker daemon can process them. They are accessible with:
$ docker logs ${CONTAINER} [-f]
As an alternative to installing debug packages inside your router, it's possible to execute commands available to the host inside the network namespace. A symlink is created in /var/run/netns/<container_name>
for convenience:
$ sudo ip netns exec ${CONTAINER} tcpdump -vvi any