* Use interface-name to exclude veth
The type veth is not a valid type (see [1] for how to obtain a list of
valid device types. Use `driver` to filter veth.
Note: It seems that NetworkManager did not manage veth so far, so this
change seems not to be relevant in practice.
Co-authored-by: Pascal Vizeli <pascal.vizeli@syshack.ch>
The latest version of OS Agent sets haos.wipe=1 as kernel argument to
trigger a device wipe. Let systemd pickup this kernel command line
argument and start haos-wipe.service.
This rather complex architecture allows to add other triggers in the
future, e.g. a button read in the boot loader.
* Disable systemd-logind support for udisks2
Currently udisks2 uses systemd-logind to prevent the system from
rebooting or similar operations while udisks operations are ongoing.
Unfortunately this stops us from using udisks2 during early boot since
systemd-logind is not ready at this point. Make the dependency
configureable so we can opt-out of using systemd-logind.
* Make dbus.service/socket and udisks2.service/socket available early
Disable default dependencies. This avoids those services to be ordered
after sysinit.target, and makes them available before local-fs.target
is reached. All mounts like mnt-data.mount are ordered before
local-fs.target, so breaking this dependency allows to use D-Bus before
mounting local file systems.
This seems fine when using the system bus directly from /run (instead of
/var/run, which is anyway a symlink to /run normally). It seems that
udisks misses /var/lib/udisks2 but it seems not to be required for the
features used so far.
So far the exit code has been evaluated, which seems to be non-zero even
with a regular term signal. With that systemd assumed the service is in
a failed state, when in fact this seems the regular behavior of dropbear
when shutting it down.
* Rename NetworkManager default profile
Rename the NetworkManager default profile to "Home Assistant OS
default". Improve documentation on how to reset to default
configuration.
* Add --cpu-rt-runtime to allow Docker allocate real-time CPU time (#1235)
* Enable Supervisor's CPU bandwith allocation feature (#1235)
Since we have CONFIG_RT_GROUP_SCHED enabled in the Home Assistant OS
kernel the Supervisor needs to enable CPU bandwith allocation for
Add-Ons which need real-time scheduling. Set the appropriate environment
variable.
Currently Linux has a limit of IGMP memberships of 20. When trying to
add membership to more than that, Linux fails with:
OSError: [Errno 105] No buffer space available
Allowing more memberships should not really be problematic as memory is
allocated dynamically when membership is actually added.
However, there is a protocol limit of how many memberships a host can be
in. The number of memberships needs to fit in a single group report
datagram of 64kB. In total 5459 group records fit in a datagram, but due
to IP header options this might be slightly smaller in practise.
(see https://github.com/home-assistant/core/issues/45957).
Use a limit of 1024, which should be plenty of headroom in both
directions.
Related to: https://github.com/home-assistant/core/issues/45957
It seems that Busybox shell (ash) cannot calculate the disk size
properly probably due to integer overflow. Use jq to calculate the last
usable LBA which seems to be able to handle large integers.
There are incident reports on the internet where poeple report that
fsck.(v)fat actually leads to problems rather file system fixes. Around
the time when Home Assistant OS added fsck.fat for the boot partition,
reports of empty boot partitions or file with weired filenames started
to appear. This could be caused by fsck.fat.
Disable fsck on the boot partition.
Use udev rules to set the CPU online. For memory, we let the kernel
bring memory online automatically. This is preferred as udev rule
processing might be delayed in a low memory situation, see:
https://lwn.net/Articles/668944/
Partition handling for disks with 4k sectors broke partition resizing
when using MBR disk label. It seems that sfdisk doesn't calculate the
last LBA for diks with MBR label. Calculate the last usable LBA ourselfs
in the MBR case.
The calculation whether to resize the partition only works with disks
with 512 byte sector size. Use values provided by sfdisk exclusively to
make sure comparing the same sector size.
Furthermore, it seems that sgdisk does not like sfdisk's backup GPT
placement:
$ sgdisk -e /dev/zram1
Warning! Secondary partition table overlaps the last partition by 250 blocks!
Today it seems sfdisk can handle GPT quite well. Use sfdisk for all
operations in hassos-expand.
The supervisor container requires the "hassio-supervisor" AppArmor
profile. Make sure our AppArmor service hassos-apparmor is a dependency
of the hassos-supervisor.service.
* Use systemd-growfs instead of resize2fs (#1106)
Since systemd 236 systemd has a built-in file system growing mechanism.
The mechanism relies on the kernels online file system resize
capabilities instead of the external resize2fs utility. Online resizing
is supposedly much faster since the kernel takes care of things.
This also makes sure that external file systems get resized which
previously have not been taken care of.
* Drop HA OS specific file system resizing
Since we have systemd-growfs in place now we can drop our file system
resizing code.
* Make sure /dev/disk/by-label/hassos-data is present after resizing
Note: systemd will retry mnt-data.mount later, so at least in theory
this shouldn't really matter. However, the journal has a lot of churn
due to that reordering.
It seems that on certain setups the default DNS over TLS mode
"opportunistic" causes delays of ~10s when trying to resolve names. This
is probably caused by providers and/or firewall setups not properly rejecting
connections on port 853.
It seems that also other distributions (such as Arch Linux) still
disable DNS over TLS currently. Side step issues with DNS over TLS by
disabling it for now.
Old Laptops are a popular choice to run Home Assistant: They have low
power consumption, are relatively fast and cheap to come by. However,
closing their lid by default puts a Linux system into suspend. This is
not what the typical user of Home Assistant OS wants. Ignore lid
activity in any state by default.
When we write the update to the boot partiton, there is nothing which
makes sure that data is written to disk. This leaves a rather large
window (probably around 30s) where a machine reset/poweroff can lead
to a corrupted boot partition. Use the sync mount option to minimize the
corruption window.
Note that sync is not ideal for flash drives normally. But since we
write very little and typically only on OS update to the boot partition,
this shouldn't be a problem.
* Avoid waiting for external drive unnecessarily
Even though the condition to start hassos-data.service is not met (the
file /mnt/overlay/data-move is not there by default), it seems that
systemd waits for the dependencies for hassos-data.service. Don't
Require or Wants any dependencies which might not be present by
default.
* Use systemd to wait for partition using partlabel device
* Use sfdisk which allows to wipe filesystem signatures
Even though we zap the partition table using sgdisk, the file system
superblock (which contains the file system label) does survive. This
can cause problems when trying to reuse a disk previously already
labeled using hassos-data: It might take precendence on next boot
over the existing data partition on the eMMC.
Make sure to clean all file system signatures using sfdisk.
* Make the datactl command more robust
Validate target disk (partition) size to avoid a copy attempt which will
fail. If e2image operation fails, make sure the leftover copy is not
regonized as data partition.
* Fix hassos-data service device unit dependencies
In case the data partition is missing avoid using the Docker command.
The Docker command triggers a socket activation, which in turn makes
systemd wait for the data partition. This blocks entry into the shell
forever.
Just enter the shell in case data partition is not mounted.
* Rewrite datactl command
Prepare the target partition as part of the datactl command. Rely on
partlabel for the target disk since we are always using GPT on the
target disk. Use systemd and partlabel mechanism to wait and find
the target data disk. Keep using the file system label to identify
the source disk.
Also use e2image instead of raw dd to move data. This should
speed up the processes significantly.
* Fix corner case when reusing same disk again
* Remove busybox Linux module support
Since systemd relies on the upstream Linux kernel module handling
utility "kmod" the busybox implementations are not required. Already
today the official "kmod" utility takes precedence:
haos # ls -la /usr/sbin/*mod*
lrwxrwxrwx 1 root root 11 Nov 11 11:32 /usr/sbin/depmod -> ../bin/kmod
lrwxrwxrwx 1 root root 11 Nov 11 11:32 /usr/sbin/insmod -> ../bin/kmod
lrwxrwxrwx 1 root root 11 Nov 11 11:32 /usr/sbin/lsmod -> ../bin/kmod
lrwxrwxrwx 1 root root 11 Nov 11 11:32 /usr/sbin/modinfo -> ../bin/kmod
lrwxrwxrwx 1 root root 11 Nov 11 11:32 /usr/sbin/modprobe -> ../bin/kmod
lrwxrwxrwx 1 root root 11 Nov 11 11:32 /usr/sbin/rmmod -> ../bin/kmod
* Move modprobe configuration alsa-base.conf to correct location
The official modprobe package from kmod checks three locations:
/etc/modprobe.d/, /lib/modprobe.d/ and /run/modprobe.d/. Since usr-move
/lib is a symlink to /usr/lib, the correct location for distribution
provided modprobe files is /usr/lib/modprobe.d.
* Add resolved.conf to disable stub resolver and DNSSEC
There are Add-Ons which try to bind port 53 on all interfaces including
127.0.0.53. Disable the stub resolver to make them continue working. We
don't need the resolver currently anyway.
Also disable DNSSEC to make sure the baords can access a NTP time server
even when their time is incorrect (since DNSSEC validation may fail).
This is a known chicken-egg problem with systemd-resolved/systemd-timesyncd
and might be addressed in a future version, with what we can reenable
DNSSEC:
https://github.com/systemd/systemd/issues/5873
* Make sure resolve gets added only once to nsswitch.conf
Only add resolve to nsswitch.conf if not already present.
* Use /run as default location for lock files for U-Boot tools
While there is a command line parameter to set the lock file explicitly,
there are other tools invoking fw_setenv (in particular rauc) which do
not set the lock file. Using /run by default makes fw_setenv use the
correct lock file in all situations.
* Don't explicitly set lock file location
Since we patch U-Boot tools to use /run by default setting it explicitly
is unnecessary.
* Update buildroot-patches for 2020.11-rc1 buildroot
* Update buildroot to 2020.11-rc1
Signed-off-by: Stefan Agner <stefan@agner.ch>
* Don't rely on sfdisk --list-free output
The --list-free (-F) argument does not allow machine readable mode. And
it seems that the output format changes over time (different spacing,
using size postfixes instead of raw blocks).
Use sfdisk json output and calculate free partition space ourselfs. This
works for 2.35 and 2.36 and is more robust since we rely on output which
is meant for scripts to parse.
* Migrate defconfigs for Buildroot 2020.11-rc1
In particular, rename BR2_TARGET_UBOOT_BOOT_SCRIPT(_SOURCE) to
BR2_PACKAGE_HOST_UBOOT_TOOLS_BOOT_SCRIPT(_SOURCE).
* Rebase/remove systemd patches for systemd 246
* Drop apparmor/libapparmor from buildroot-external
* hassos-persists: use /run as directory for lockfiles
The U-Boot tools use /var/lock by default which is not created any more
by systemd by default (it is under tmpfiles legacy.conf, which we no
longer install).
* Disable systemd-update-done.service
The service is not suited for pure read-only systems. In particular the
service needs to be able to write a file in /etc and /var. Remove the
service. Note: This is a static service and cannot be removed using
systemd-preset.
* Disable apparmor.service for now
The service loads all default profiles. Some might actually cause
problems. E.g. the profile for ping seems not to match our setup for
/etc/resolv.conf:
[85503.634653] audit: type=1400 audit(1605286002.684:236): apparmor="DENIED" operation="open" profile="ping" name="/run/resolv.conf" pid=27585 comm="ping" requested_mask="r" denied_mask="r" fsuid=0 ouid=0
Drop AVAHI and use systemd-resolved to announce hostname via mDNS
and LLMNR. Also continue to offer the _workstation._tcp.local service
since it is used by the CoreDNS mDNS plug-in.
In case a container image is corrupted `docker inspect` might fail:
# docker inspect --format='{{.Id}}' "${SUPERVISOR_IMAGE}"
Error response from daemon: readlink /mnt/data/docker/overlay2: invalid argument
In that same state the `docker images` command still shows the images.
Since `docker inspect` returns an error SUPERVISOR_IMAGE_ID will be empty
and a simple `docker pull` will be attempted. That does not suffice to
recover from a corrupted container image.
Use `docker images` to get the image ids and make sure to delete all
image ids found by that command.
Also don't use RuntimeDirectory since it deletes the runtime directory
between the service start attempts which defeats the purpose.
* Simplify self healing capabilities of Supervisor service
Instead of relying on time based information on how long the container
has been running use a startup marker file to infer if the last startup
has been successful.
* Update buildroot-external/rootfs-overlay/usr/sbin/hassos-supervisor
Co-authored-by: Pascal Vizeli <pascal.vizeli@syshack.ch>
Co-authored-by: Pascal Vizeli <pascal.vizeli@syshack.ch>
* automatically fsck to repair partitions
* add fsck.fat so rpi boot partition can be repaired
* Use Wants= instead of Requires=
Co-authored-by: Pascal Vizeli <pascal.vizeli@syshack.ch>
* add dosfstools to all images
* run hassos-data and hassos-expand after fsck
Co-authored-by: Pascal Vizeli <pascal.vizeli@syshack.ch>
The Docker socket path is /run/docker.sock. Also only one path can be
used per property. This fixes the supervisor service, which currently
refuses to start due to missing Docker socket.
The new readline utilty used by the CLI add-on requires the size of the
terminal to be set. Use the resize command to initialize terminal size
on login if we are running on a serial terminal.
The hassos-expand script calls sfdisk to find free disk space. It seems
that today it considers the space before the first partition as free:
$ sudo sfdisk -Fq /dev/sdi
Start End Sectors Size
2048 16383 14336 7M
This causes the script to always resize. It seems not to cause harm to
the partition table (it does not resize really). However, the call to
partx seems to confuse systemd and kill the mnt-data.mount process
(presumably because udev causes remove/add events for the by-label
device units).
Consider everything below 8MiB to not be worthy of a size change. This
avoids missdetection and resize attempts where there is no need.
