Building a bootc `anaconda-iso` or `bootc-installer` from Fedora fails with unset vendor

[supakeen]

When doing:

sudo image-builder build --bootc-ref quay.io/fedora/fedora-bootc:rawhide --bootc-default-fs ext4 anaconda-iso

We seem to not set the vendor for the org.osbuild.grub2.iso stage causing the build to fail with:

Traceback (most recent call last):
File "/run/osbuild/bin/org.osbuild.grub2.iso", line 136, in <module>
    ret = main(args["tree"], args["options"])
  File "/run/osbuild/bin/org.osbuild.grub2.iso", line 70, in main
    vendor = options["vendor"]
             ~~~~~~~^^^^^^^^^^
KeyError: 'vendor'

Note that this also happens with bootc-image-builder and only affects installer image types; not disk images.

[supakeen]

I tried this out with the bootc-installer image type as well; and it also fails if the installer image is based on Fedora. While building my Fedora-based installer environment I saw:

STEP 3/4: RUN dnf reinstall -y shim-x64
Updating and loading repositories:
 Fedora - Rawhide - Developmental packa 100% |  11.8 MiB/s |  21.8 MiB |  00m02s
 Fedora rawhide openh264 (From Cisco) - 100% |   1.9 KiB/s |   4.1 KiB |  00m02s
Repositories loaded.
Package               Arch   Version  Repository      Size
Reinstalling:
 shim-x64             x86_64 0:16.1-5 rawhide      1.9 MiB
   replacing shim-x64 x86_64 0:16.1-5 <unknown>    1.9 MiB

Transaction Summary:
 Reinstalling:       1 package
 Replacing:          1 package

Total size of inbound packages is 482 KiB. Need to download 482 KiB.
After this operation, 0 B extra will be used (install 2 MiB, remove 2 MiB).
[1/1] shim-x64-0:16.1-5.x86_64          100% |   3.8 MiB/s | 482.3 KiB |  00m00s
--------------------------------------------------------------------------------
[1/1] Total                             100% |   3.7 MiB/s | 482.3 KiB |  00m00s
Running transaction
[1/4] Verify package files              100% |   1.0 KiB/s |   1.0   B |  00m00s
[2/4] Prepare transaction               100% |  60.0   B/s |   2.0   B |  00m00s
[3/4] Reinstalling shim-x64-0:16.1-5.x8 100% | 385.6 MiB/s |   1.9 MiB |  00m00s
[4/4] Removing shim-x64-0:16.1-5.x86_64 100% | 106.0   B/s |  13.0   B |  00m00s
>>> Running %posttrans scriptlet: shim-x64-0:16.1-5.x86_64
>>> Finished %posttrans scriptlet: shim-x64-0:16.1-5.x86_64
>>> Scriptlet output:
>>> cp: cannot create directory '/boot/efi/EFI/fedora': No such file or directory
>>> cp: cannot create directory '/boot/efi/EFI/BOOT': No such file or directory
>>> 

During the build; since we inspect /boot/efi to find the UEFI vendor and these files never exist this is likely why things break on Fedora. This also makes our 'workaround' in the example Containerfiles no longer work; the behavior is the same. Pre-creating the /boot/efi/EFI directory in the Containerfile before reinstalling shim makes the files end up where they're supposed to be; however, we still fail to find the vendor.

This is probably related to osbuild/osbuild#2202 but instead of passing the source directory to the grub2.iso stage we should likely also look in /usr/lib/bootupd/updates to determine the UEFI vendor during manifest generation?

However, that directory is also empty (aside from some JSON which doesn't contain the vendor).

This used to work so something somewhere must have changed, perhaps in shim packaging, bootupd, or bootc?

[supakeen]

The shim reinstall not working seems to only affect rawhide; perhaps related to shim-16.1 versus shim-15.8. Us not finding the vendor affects all Fedora's.