13 June 2016 (v0.1.1) Current kernel: 4.1.21
Building pre-built images is hard. The curent status quo is setting up the image on your device and then using dd
to create the device image.
All app files live in the /srv
directory.
- Build directories are set up as below:
/srv/ ->
- src/ # All sources are downloaded/cloned into this directory
- img/ # The final image "rpi2-picake.img" lives in this directory
- tmpmnt/ ->
- boot # The boot partition of the image is mounted to this directory
- root # The root partition of the image is mounted to this directory
-
The latest Arch Arm image for the RPi 2 is downloaded into
/srv/builddir/src/ArchLinuxARM-rpi-2-latest.tar.gz
. -
The final disk image is allocated at
/srv/builddir/img/rpi2-picake.img
. (Note: The default size of 3G is specified inbuild.sh
. This is an arbitrary size that is an artefact of piCake's original use-case.) -
fdisk is used with a bash HEREDOC to partition the .img file into a 100MB root partition and a [size of image file - 100]MB root partition. (Note: The empty lines in the bash HEREDOC passed to fdisk are INTENTIONAL as they pass a /newline to accept the FDISK default.)
-
A loop device is created (this is why the --privileged flag is required on the docker container) and pointed at the rpi2-picake.img file. We save the loopdevice in the $LOOPDEV bash variable.
-
kpartx
is used to detect partitions from the loop device and create devices in/dev/mapper
.dmsetup
is then used to the prepare the devices for mounting. The --noudevsync tag is used because docker containers don't have access to udev. -
We format and create the filesystems in
/dev/mapper
with FAT32 for theboot
parttion and EXT4 for theroot
partition. -
Then, we mount the
boot
partition at/srv/buiddir/tmpmnt/boot
and theroot
partition at/srv/buiddir/tmpmnt/root
. -
We use
bsdtar
(to preserve permissions) to extract the contents of the downloaded arch image into/srv/buiddir/tmpmnt/root
. Then, we move the contents of/srv/buiddir/tmpmnt/root/boot
to/srv/buiddir/tmpmnt/boot
to ensure that the boot files are allocated properly. -
We use
proot
to chroot into the image, passing in the boot and root partitions mounted in/srv/builddir/tmpmnt
as parameters.proot
is syntatic sugar as it mounts the dns configuration and the/proc
and/dev
directories among others. It also conveniently mounts the host fs at/host-rootfs
in the chroot. However, it is also perfectly legit to rewrite this to use vanilla chroot. -
We
source
the env file at./src/config/env
in order to pass variables defined in the current Bash shell to the Bash shell invoked byproot
. This is a bit of a hack, but I haven't found a better way so far. Then wesource
-
We define
$APPDIR
at the top of this file. By default it is/srv/app
. -
We run
pacman -Syu
to pull in a couple of base packages. Currently we pull in:
crda, iw, wireless-regdb, wpa_supplicant
for Wifivim, sudo, python2, wget, git, curl, ca-certificates-mozilla, db, dbus, ncurses, openresolv, openssl, xfsprogs, unzip
for general utility, updating ssh and easy of use when SSH-ing into the Pi.bind
to provide the on-box DNS for the self-hosted Wifi hotspot.hostapd
to create the self-hosted wifi networkdhcp
to provide dhcp address assignment on the self-hosted wifi network.
-
We install a custom-built kernel from
./src/kernel/linux-raspberrypi-4.1.15-1-armv7h.pkg.tar.xz
. This custom-build of the kernel includes a wifi patch that enables the rt2x00 wifi card to host multiple wifi networks i.e. to host a wifi network while being a client to another wifi network. This is NOT REQUIRED if you do not need virtual wifi capability. If you choose to comment this line out though, you MUST update thehostapd
config files to point to the correct device. -
We copy
./src/config
,./src/deps
and./src/scripts
. -
If you have a public key defined in the
$PRIV_KEY
env variable, it will be inserted in/root/.ssh/id_rsa
and/home/alarm/.ssh/id_rsa
. -
Now, we will execute the executable scripts in
./src/scripts/install.d
bysource
ing them into the current script. Currently, there are NO guarantees about the order in which the scripts are run. -
Now, we copy the systemd init files to
/etc/systemd/system
and create an env config file in/etc/systemd/system/picake.service.d
. We copy the$APPDIR
variable into here to enable changing the APPDIR in one location only (at the top ofarch-build.sh
). The symbolic links duplicate what happens whensystemctl enable <service>
is run as systemd detects when it is in a chroot environment and does not run, thussystemctl
does not work.
-
We try to detach all devices listed by
dmsetup ls
. Currently this detaches ALL devices, not just the two used by the current build. This is because I have yet to figure out how to share the$BOOTPART
and$ROOTPART
variables to theclean.sh
script. This is due to Drone's build process running each script in a new Bash shell. -
The loop device is also removed using
losetup -d
.
-
We use
pigz
which is a version of gzip optimized to utilize multiple cores to gzip up the image. -
The image is then deployed to the release directory, in this case
/host-release
which is mounted from/srv/app/releases/picake
.
If you fork this repository, but want to keep it in sync with updates pushed to the master, don't forget to add this repo as an upstream
branch to your current fork. Instructions here: https://help.github.com/articles/configuring-a-remote-for-a-fork/
Then, you can sync your fork at any time like so: https://help.github.com/articles/syncing-a-fork/
- Command-line client to build images.
- Investigate removing dependency on Drone
- Save $BOOTPART and $ROOTPART to a temporary build file to remove only the relevant devices in
clean.sh
. - Make fdisk commands more explicit i.e. don't accept defaults, specify all values.
- Move to Alpine Linux for the build container.
- Consider Alpine as a base for the RPi base.
Code originally by Asyrique Thevendran, 2016. Heavily inspired by this blog post
Name credits: My brother, Aqiel Thevendran.