As overlay mounts its responsible of mounting some of the RW paths that
bind mounts are using (for example /etc) we need to depend on it and let
it finish before running ht ebind mounts
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
- Mount the needed base mounts (/proc /dev /sys /tmp)
- Use our own console for yip (required to add the PATH under uki)
- Order the DAG in a proper way (was out of order and not working)
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
This makes immucore run more in parallel rather than block everything
else. We just tell sysroot.mount that eventually it will be mounted and
to wait for a bit.
This allows us to be more flexible where to run and run in parallel in
cases like cdrom in which we may do things but we need the sysroot to be
mounted already but not from us.
Also adds the initramfs stage directly in immucore and merges all the dracut config into one
Dont create sysroot, just add a timeout override so it waits for us
Dont block on the service, just make sure to finish before
initrd.target
Fix mounts from cmdline
More proper log
Store logs under the /run/immucore dir
Store rootfs and initramfs logs separated
Do not log the full stages in INFO level
Run initramfs stage in immucore directly on boot and cd/netboot
Drop systemd requirement from dracut module
Signed-off-by: Itxaka itxaka.garcia@spectrocloud.com
* Add tests and fix some issues that arised from testing
Mainly around the cmdargs and how many items it returns.
Also drop the iso target and jobs as its not necessary now
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
* Lint
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
---------
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
Introduce support for the same old cos stanzas but with immucore naming
so we can control them better in the future
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
Load the base overlay config from the cos-layout.env file. This means we
have to depend the base overlay mount to the load-config step, which
moves things a bit in the DAG.
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
* Fix recovery
- change TargetLabel into TargetDevice to hav ethe full device instead
of just the label
- log error if we cant mount
- Get state partition to mount depending on recovery or active/passive
- Get state patition fs on the fly
- Set proper label device to mount on sysroot based on our boot state
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
* Fix test
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
---------
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
This patch allows setting the root mount as RW indicated by setting
rd.cos.debugrw in the cmdline
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
Original cos-immutable-rootfs sets the sentinel to livecd if it finds
the rd.cos.disable in the cmdline.
We should do the same. Unfortunately the state sdk doesnt support that
yet so we have to do it on our side until its supported
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
* Full rework
- Extract steps to a different file
- Simplify dag for easy understanding
- Load dag based on our boot process
- Simplify steps to not depend on useless stuff
- Better logging
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
* Move sentinel file to the dag
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
* Adapt tests
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
---------
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
Unfortunately ghw detects mountpoint only by device, thus kairos-agent
does the same.
If we expect to make this work we need to mount the devices by the dev
instead of label until ghw supports detecting the devices mountpoint
ALSO via label.
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
- Set debug from ENV var as well
- Only set sentinel if we are not on a dry-run so we can examine the
DAG
- Set the DAG with Init so tmpfs is part of it
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
Looks like we need to have the oem partition loaded for some cases, and
during rootfs we should pick it up.
Currently we only mount it under /sysroot/oem. IIRC in
cos-immutable-rootfs it was mounted under /oem directly, the rootfs
stage service run and then when cos-layout was launched it made sure to
unmount /oem so it could mount it under /sysroot/oem
In our case its a bit more complicated, we may need to mount it under
/oem before the rootfs run, then unmount it and mount it in the proper
place. Im hoping we can get away with just creating the symlink
Signed-off-by: Itxaka <itxaka.garcia@spectrocloud.com>
- Set global debug log
- Set caller in the log if debug
- Fatal if cannot get target image
- clean up slices from cos-layoout.env
- Add generic log with more context info
- dump runtime to log on start
- Set proper default paths if overlay is empty (fixes recovery)
- Fix ErrAlreadyMounted to not trigger failures
Signed-off-by: Itxaka <itxaka@spectrocloud.com>
It was nice when things break all over but its much more stable now.
Lets add logging with head and with proper enrichment of the data being
sent
Signed-off-by: Itxaka <itxaka@spectrocloud.com>
- set debug level via cmdline
- set default RW paths needed ALWAYS to boot a working system, even
with no config files(recovery)
- Check for loop device before mounting it to avoid duplicated mounts
- Force overlay and bind arrays to uniqueness
- Rename OverlayDir to OverlayDirs to better represent the content
Signed-off-by: Itxaka <itxaka@spectrocloud.com>
So we can inherit the rootfs where to store stuff, no matter if / or
/sysroot.
also drop fstabfile as its alwasy gonna be the same file but with a
different rootfs
Signed-off-by: Itxaka <itxaka@spectrocloud.com>
- drop vfs
- less hardcoded stuff, use the runtime status
- Remove things that syscalls dont understand (defaults, auto)
Signed-off-by: Itxaka <itxaka@spectrocloud.com>
