Using path identifiers for tag names causes ambiguity as it's not
(immediately) clear whether a reference to it is a path or a tagname.
This causes hard to read/interpret log files and can lead to subtle and
hard to detect bugs.
The same tag can refer to different things based on the context
(partition/device/mountpoint/etc), so just use the 'tag-' prefix.
On the previous 'mount: /' step, I also added that it's to be mounted on
"dirname: '/'" to make it (more) explicit.
And removed it again as it seems you need to specify both 'dirname' AND
'mount-on' or neither. See https://bugs.debian.org/1023321
The firmware partition holds a copy of the initramfs and the kernel and
over the years we have seen a steady increase in its sizes.
Resizing the firmware partition later on is cumbersome as the root
partition follows directly, so it's better to make the firmware
partition not too small. A size of 508MB should be enough to accommodate
4-5 kernels+initramfs, which seems desirable.
Previously one had to calculate how large the /boot/firmware partition
would be, but expressing it directly in MiB units is much clearer.
This also has the benefit that the /boot/firmware partition's size would
not change if the total image size would be changed.
Such a change should be a deliberate decision and not some side-effect.
As that 'side-effect' did happen since first submitting this patch,
revert the /boot/firmware partition's size back to 300MB.
Signed-off-by: Diederik de Haas <didi.debian@cknow.org>
The wireless firmware is in the firmware-brcm80211 package, but that is
still in 'non-free', so add 'non-free' back to the sources until all
the needed firmware packages are in 'non-free-firmware'.
https://salsa.debian.org/kernel-team/firmware-nonfree/-/merge_requests/36
is where the move to non-free-firmware is proposed, but not yet merged.
When the image was build also determines which package versions got
installed in the generated image and could help explain why a user has
problems with the downloaded image.
Signed-off-by: Diederik de Haas <didi.debian@cknow.org>
Regression introduced in f89f71560d2ca1bd60d97dbb26b89782657d56ae:
the sed call modifies /etc/default/raspi-firmware, which used to be
/etc/default/raspi3-firmware; while not ideal, working on
/etc/default/raspi*-firmware shouldn't interfere on unrelated files.
It used to be pulled this way (up to Bullseye), via systemd:
Depends: […] systemd-timesyncd | time-daemon […]
Starting with Bookworm, this was downgraded to:
Recommends: […] systemd-timesyncd | time-daemon
Install it all the time: NTP support is important on Raspberry Pi
devices, which usually don't feature an RTC.
But be careful since Buster had systemd itself provide that feature (no
separate systemd-timesyncd package yet).
Thanks, David Tomaschik!
Commit 26a7de63b0 in master:
/etc/machine-id needs to exist and be empty on buster, while bullseye
needs this file not to exist at all. For now, treat both bullseye and
bookworm the same way.
See https://salsa.debian.org/raspi-team/image-specs/-/issues/57 for
detailed background.
Summary, e.g. on the Pi 4:
- fresh build and first boot means:
console=ttyS1,115200 console=tty0
- after dpkg-reconfigure raspi-firmware has run, with the default
settings:
console=tty0 console=ttyS1,115200
Having some consistency across boots seems desirable (esp. when the Pi
fails to boot and the hints are on a serial console which might not be
wired), so insert the console= parameter for the serial console right
before the root= parameter.
Currently, the /etc/kernel/postinst.d/z50-raspi-firmware hook uses:
${pre_cmdline} root=$ROOTPART […]
and console= parameters are inserted via ${pre_cmdline}, so inserting
the serial console before root= should get us the same results.
Without this, the block device holding the root filesystem would be
resolved at the first boot when reconfiguring raspi-firmware (e.g.
/dev/mmcblk1p2) which would then make the system fail to boot if it
ever shows up under a different name (e.g. /dev/mmcblk0p2).
Set ROOTPART parameter explicitly to stick to label-based booting.
The logic wrt /etc/machine-id changed between Buster and Bullseye.
While on Bullseye the file should not exist, on Buster the file must
exist, but be empty, in order to generate a new machine-id on first
boot.
It seems that /var/lib/dbus/machine-id is a symlink to /etc/machine-id
on Buster, while a separate file on Bullseye, so nothing needs to be
done with that file/symlink.
Signed-off-by: Diederik de Haas <didi.debian@cknow.org>
This means the generated recipes are getting two empty lines if there
are no such commands (that's the case for everyone right now), but this
emphasizes the existence of this placeholder, the same way as for its
__EXTRA_ROOT_SHELL_CMDS__ twin.
Group raspi-firmware and firmware-brcm80211 together, and make the
firmware package a regular list item in the master YAML file (making
editors happy about it).
Of course, this means that in all generated recipes, linux-image and
raspi*-firmware switch places.
This is a proof of concept rather than an ideal, final situation.
It can be used this way:
for v in 1 2 3 4 ; do
for s in buster bullseye; do
./generate-recipe.py $v $s
done
done
and it has been verified to produce very similar results compared to the
existing many-sed approach.
Differences are as follows:
- Missing newline after some backports stanza, due to the removal of
the other APT line. There's already MR#51 that aims at fixing some
newline-related issues anyway, so this can be addressed separately.
- Less schizophrenia in the generated sources.list for buster/4, as we
are now only showing a reason for enabling the backports, instead
of starting by explaining why backports are disabled by default.
- Dropping APT::Default-Release = buster in the buster/4 case, which
is no longer needed as we are pulling things from buster-backports
rather than pulling them from unstable (see 57e90df103).
- No longer trying to fix the firmware package name by throwing a
broken sed at rpi-reconfigure-raspi-firmware.service in the buster/4
case: the syntax was buggy and fixing it would have made us try to
replace raspi-firmware with raspi-firmware/buster-backports, while
the correct thing to do is to not touch it in the first place
(raspi-firmware is the correct name for the firmware package, pulled
from buster-backports).
As a side effect, this transforms the existing __EXTRA_SHELL_CMDS__ into
a slightly more explicit __EXTRA_ROOT_SHELL_CMDS__ which now has its
__EXTRA_CHROOT_SHELL_CMDS__ twin. That's the entry point that was
missing and made 45cb5619d4 necessary in the past.
This is needed to use Debian repos served over https, but also a LOT of
other programs, like reportbug, which want to communicate securely.
Also sorted the list of packages alphabatically as I couldn't find a
reason for the current order and then a logical sort order is better.
Signed-off-by: Diederik de Haas <didi.debian@cknow.org>
Switch away from using a systemd service for the initial root resize.
Instead, we resize the root partition and filesystem in the initrd.
To simplify things, the initrd script will check whether it should resize
the partition on every boot. It does this by checking if the entire disk
(ignoring an empty 4MB) is in use. However, the scripts themselves are
deleted from the system after the initrd is generated. After the image
is installed, the resize script should exist only in the initrd. When the
kernel gets upgraded (eg, for a security update) or a new initrd is generated
due to a package install, the new initrd will not contain the resize script.
At that point, nothing will remain from the image's initial resize
bootstrapping process.
This process (but not the scripts) is similar to what cloud-initramfs-growroot
does. However, that particular package has an indirect dependency on Python,
and we don't necessarily want that overhead in our images just for resizing.
Why almost? Because Rpi0w uses ttyS1 instead of ttyAMA0 desipte being part of the RPi1 family...
...But it will work fine for the _second_ boot onwards, if things go according to plan.
For all other RPi models, it should work from the first boot on.