dracut_functions.sh: make sure device UUIDs are correct before creating systemd device units

[Cyanoxygen]

Sometimes the information read from udev may be outdated, for example a disk has just got repartitioned. We don't know why this happens, but it is better to verify the unique identifiers before returning from get_persistent_dev().

We compare the ID provided by get_persistent_dev() and the probed result of blkid. If they don't match, we use the latter one.

This inconsistency is observed during initramfs generation stage of the OS installation, and can not be reproduced reliably, and may be platform dependent. It results in an unbootable OS, stuck at initrd trying to wait a nonexistant UUID.

This pull request adds a verification to confirm the device UUID we are going to wait for during the initramfs actually is correct. @MingcongBai can provide some details on the issue this PR is going to fix.

Changes

• dracut-functions.sh: verify device UUID before creating device units for the initramfs

Checklist

• I have tested it locally
• I have reviewed and updated any documentation if relevant
• I am providing new code and test(s) for it

[MingcongBai]

This issue is actually quite weird, sometimes as though U-Dev had some sort of "cache" in place for devices. I ran into this whilst installing AOSC OS, which runs dracut at the end of installation in a chroot - the disk was repartitioned, but the resulting initramfs contains /dev/disk entries that reflects the disk before it was repartitioned.

I can confirm that this patch fixes the issue I ran into, but it could also be worth looking into why U-Dev did not refresh the device paths immediately.

[Cyanoxygen]

why U-Dev did not refresh the device paths immediately.

Then we should dive into systemd-udevd then, or should we manually trigger the refresh?

[Cyanoxygen]

Oh well. A quickly written script showed that udev sometimes doesn't drop the old records, or won't update the symlinks in /dev/disk on time.

I have a whole bunch of broken ones too:

So, reading from udev may be unreliable, we can either add an additional step to make sure they are correct, or switch to using blkid --probe entirely.

[Conan-Kudo]

Is there a bug filed with systemd about this?

[Cyanoxygen]

Is there a bug filed with systemd about this?

No AFAIK. Do you think we should treat this as an upstream issue?

[jozzsi]

Do you think we should treat this as an upstream issue?

First yes, please.

[Conan-Kudo]

Marking as draft until we get feedback from systemd folks.