Containerized OpenWrt image builder

What
- Note
Why
How
Examples
- Building a x86_64 image and running it in qemu
Building an OpenWrt snapshot release
Author
License

What

Easily and quickly build OpenWrt custom images (e.g. for your embedded device or a Raspberry PI) using a self-contained docker container or a nix-shell and the OpenWrt image builder. On the builder host, Docker, podman/buildah (for dockerless operation) or nix-shell is the only requirement. Supports latest OpenWrt release (23.05.x).

Note

The OpenWrt imagebuilder uses pre-compiled packages to build the final image. Go here if you are looking for a docker images to compile OpenWrt completely from source.

Why

customized and optimized (size) images with your personal configurations
full automatic image creation (could be run in CI)
repeatable builds
easy configuration, fast build

How

Using docker

$ git clone https://github.com/jandelgado/lede-dockerbuilder.git
$ cd lede-dockerbuilder
$ ./builder.sh build-docker-image example-nexx-wt3020.conf
$ ./builder.sh build example-nexx-wt3020.conf

The build-docker-image command will first build the docker image containing the actual image builder. The resulting docker image is per default tagged with openwrt-imagebuilder:<Release>-<Target>-<Subtarget>. The build command will afterwards run a container, which builds the actual OpenWrt image. The final OpenWrt image will be available in the output/ directory.

Using nix-shell

$ git clone https://github.com/jandelgado/lede-dockerbuilder.git
$ cd lede-dockerbuilder
$ ./builder.sh build example-nexx-wt3020.conf --nix

Using nix-shell does not require building a container image or starting a container first, therefore it is usually faster.

Usage

Dockerized LEDE/OpenWRT image builder.

Usage: $1 COMMAND CONFIGFILE [OPTIONS]
  COMMAND is one of:
    build-docker-image - build the docker image (run once first)
    profiles           - show available profiles for current configuration
    build              - start container and build the LEDE/OpenWRT image
    shell              - start shell in the build dir 
  CONFIGFILE           - configuraton file to use

  OPTIONS:
  -o OUTPUT_DIR        - output directory (default $OUTPUT_DIR)
  --docker-opts OPTS   - additional options to pass to docker run
                         (can occur multiple times)
  -f ROOTFS_OVERLAY    - rootfs-overlay directory (default $ROOTFS_OVERLAY)
  --sudo               - call container tool with sudo
  --podman             - use buildah and podman to build and run container
  --nerdctl            - use nerdctl to build and run container
  --docker             - use docker to build and run container (default)
  --nix                - build using nix-shell

  command line options -o, -f override config file settings.

Example:
  # build the builder docker image first
  ./builder.sh build-docker-image example-glinet-gl-ar750.conf

  # now build the OpenWrt image, overriding output and rootfs locations
  ./builder.sh build example-glinet-gl-ar750.conf -o output -f myrootfs

  # show available profiles for the arch/target/subtarget of the given configuration
  ./builder.sh profiles example-glinet-gl-ar750.conf

  # pass additional docker options: mount downloads to host directory during build
  ./builder.sh build example-glinet-gl-ar750.conf --docker-opts "-v=$(pwd)/dl:/lede/imagebuilder/dl:z"

  # use nix to build the OpenWrt image, no need to build a container first
  ./builder.sh build example-x86_64.conf --nix

Builder runtime

By default docker will be used to build and run the container.
When called with --podman option, lede-dockerbuilder will use buildah and podman to build and run the container.
When called with --nerdctl option, lede-dockerbuilder will use nerdctl to build and run the container.
Use the --sudo option to run the container command with sudo.
Use the --nix option to run the build in a nix-shell (instead of using a container runtime)

When using a container builder like docker, the build container will be newly created on every build. When using the nix builder, the build environment will be reused, which is ususally faster. By default, the nix build environments are installed in the .build directory, relative to the builder.sh script. This can be overriden with the NIX_BUILD_DIR environment variable.

Configuration file

The configuration file is quiet self-explanatory. The following parameters are mandatory (prefixed with LEDE_ for historical reasons, config works also with OpenWrt):

LEDE_TARGET - Target architecture
LEDE_SUBTARGET - Sub target architecture
LEDE_RELEASE - Release to use
LEDE_PROFILE - Profile to use
LEDE_PACKAGES - list of packages to include/exclude. Prepend package to be excluded with -
LEDE_DISABLED_SERVICES - list of services to disable on startup in /etc/init.d

LEDE_TARGET, LEDE_SUBTARGET and LEDE_RELEASE are used to construct the URL of the image builder binary well as for the construction for the tag of the docker image.

You can find the proper values by browsing the OpenWrt website e.g. here and here.

In addition the following optional parameters can be set, to further control output and image creation:

OUTPUT_DIR - path where resulting images are stored. Defaults to output in the scripts directory (can be overridden by -o parameter). Will be automatically created.
ROOTFS_OVERLAY - path of the root file system overlay directory. Defaults to rootfs-overlay in the scripts directory (can be overridden by -f parameter).
LEDE_BUILDER_URL - URL of the LEDE/OpenWrt image builder to use, override if you do not wish to use the default builder (https://downloads.openwrt.org/releases/$LEDE_RELEASE/targets/$LEDE_TARGET/$LEDE_SUBTARGET/openwrt-imagebuilder-$LEDE_RELEASE-$LEDE_TARGET-$LEDE_SUBTARGET.Linux-x86_64.tar.xz)
REPOSITORIES_CONF - optional file file to use instead of the default repositories.conf. The file will be mounted in the container. Look at the official documentation for more information.

Use the BASEDIR_CONFIG_FILE variable to set locations of OUTPUT_DIR or ROOTFS_OVERLAY relative to the configuration files location. This allows self-contained projects outside of the lede-dockerbuilder folder. If e.g. ROOTFS_OVERLAY=$BASEDIR_CONFIG_FILE/rootfs-overlay is set, then the rootfs-overlay directory is expected to be in the same directory as the configuration file.

Example configuration for my NEXX WT3020 router, where I have an encrypted USB disk attached so I can use it as a simple NAS with samba and ftp:

# LEDE profile to use: NEXX WT3020
LEDE_PROFILE=nexx_wt3020-8m
LEDE_RELEASE=23.05.0
LEDE_TARGET=ramips
LEDE_SUBTARGET=mt7620

# list packages to include in LEDE image. prepend packages to deinstall with "-".
# 
# include all packages to build a mobile NAS supporting disk encryption:
# ksmbd (samba4 is too large now for the WT3020's 8MB), cryptsetup.
# see https://github.com/namjaejeon/ksmbd-tools for ksmbd info.
LEDE_PACKAGES="ksmbd-server lsblk block-mount\
    kmod-usb-storage-uas kmod-scsi-core ntfs-3g\
    kmod-nls-iso8859-1 cryptsetup kmod-crypto-xts\
    kmod-mt76 kmod-usb2 kmod-usb-ohci kmod-usb-core kmod-dm kmod-crypto-ecb\
    kmod-crypto-misc kmod-crypto-cbc kmod-crypto-crc32c kmod-crypto-hash\
    kmod-crypto-user\
    -ppp -kmod-ppp -kmod-pppoe -kmod-pppox -ppp-mod-pppoe -opkg\
    -ip6tables -odhcp6c -kmod-ipv6 -kmod-ip6tables -odhcpd-ipv6only"

# optionally override OUTPUT_DIR and ROOTFS_OVERLAY directory location here

File system overlay

Place any files and folders that should be copied to the root file system of the resulting image to the directory pointed to by ROOTFS_OVERLAY (default: rootfs-overlay/), which can be overridden by the -f command line option.

Example directory structure

The following is an example directoy layout, which I use to create a customized OpenWrt image for my NEXX WT3020 router (including the generated output).

├── builder.sh
├── docker
│   ├── Dockerfile
│   └── etc
│        └── entrypoint.sh
├── example.cfg
├── example-openwrt.cfg
├── output
│   ├── openwrt-xx.yy.z-ramips-mt7620-device-wt3020-8m.manifest
│   ├── openwrt-xx.yy.z-ramips-mt7620-wt3020-8M-squashfs-factory.bin
│   ├── openwrt-xx.yy.z-ramips-mt7620-wt3020-8M-squashfs-sysupgrade.bin
│   └── sha256sums
├── README.md
└── rootfs-overlay
    ├── etc
    │   ├── config
    │   │   ├── dhcp
    │   │   ├── dropbear
    │   │   ├── firewall
    │   │   ├── network
    │   │   ├── samba
    │   │   ├── system
    │   │   ├── wireless
    │   ├── dropbear
    │   │   └── authorized_keys
    │   ├── hotplug.d
    │   │   └── block
    │   │       └── 10-mount
    │   ├── passwd
    │   ├── rc.local
    │   ├── shadow
    │   └── vsftpd.conf
    ├── README.md
    └── usr
        └── local
            └── bin
                └── fix_sta_ap.sh

Debugging

Run ./builder.sh shell CONFIGFILE to get a shell into the docker container, e.g. ./builder.sh shell example.cfg.

Examples

These examples evolved from images I use myself.

image with LUCI web GUI for the Raspberry PI 2. Just ~8MB gziped. I use this image on my home dnsmasq/openvpn 'server'.
image with LUCI web GUI and adblocker for the Raspberry PI 4
image for the TP-Link WR1043ND
image with samba, vsftpd and encrypted usb disk for NEXX-WT3020. Is the predessor of ...
image with samba, vsftpd and encrypted usb disk for GINET-GL-M300N V2. Is the predessor of ...
image with samba, vsftpd, adblock and encrypted usb disk for GINET-GL-AR750. This is my travel router setup where I have an encrypted USB disk connected to the router, accessible through SMB or FTP and have an adblocker running. Useful if you travel much.

To build an example run ./builder.sh build <config-file>, e.g.

$ ./builder.sh build example-rpi2.conf

The resulting image can be found in the output/ directory. The OpenWrt wiki describes how to flash the new image in detail.

Building a x86_64 image and running it in qemu

The example-x86_64.conf file can be used to build a x86_64 based OpenWrt image which can also be run in qemu, e.g. if you need a virtual router/firewall.

First build the image with builder.sh build example-x86_64.conf, then unpack the resulting image with e.g. gunzip output/openwrt-23.05.0-x86-64-generic-ext4-combined.img.gz. Finally the image can be started with qemu (or simply use run_in_qemu.sh)

qemu-system-x86_64 \
    -enable-kvm \
    -nographic \
    -device ide-hd,drive=d0,bus=ide.0 \
    -netdev user,id=hn0 \
    -device virtio-net-pci,netdev=hn0,id=wan \
    -netdev user,id=hn1,hostfwd=tcp::5555-:22001 \
    -device virtio-net-pci,netdev=hn1,id=lan \
    -drive id=d0,if=none,file=output/openwrt-19.07.0-x86-64-combined-ext4.img

The hostfwd=... part can be omitted and is used in case you redirect port 22001 on your WAN adapter to port 22 of the LAN adapter, in case you want to access SSH in the VM from your qemu-host. Check the /etc/config/firewall file for details.

Qemu will assign the IP address 10.0.2.15/24 to the WAN interface (eth1) and OpenWrt the address 192.168.1.1/24 to the LAN (br-lan bridge with eth0) interface.

Note: press CTRL-A X to exit qemu.

Building an OpenWrt snapshot release

To build a snapshot release, set LEDE_RELEASE to snapshots and let LEDE_BUILDER_URL point to the image builder in the snapshot dir, e.g.

LEDE_RELEASE=snapshots
LEDE_BUILDER_URL="https://downloads.openwrt.org/$LEDE_RELEASE/targets/$LEDE_TARGET/$LEDE_SUBTARGET/openwrt-imagebuilder-$LEDE_TARGET-$LEDE_SUBTARGET.Linux-x86_64.tar.xz"

See the this example which builds an x86_64 image using the snapshot release.

Author

License

Apache License 2.0