Now that xz 5.4 is out and erofs-utils is built with LZMA support, it is
possible to compress the EROFS image with LZMA for higher compression.
`mkfs.erofs` trows a few warnings about using experimental features, but
they should not be an issue.
Nothing changes for the releng profile, for now at least.
The `bootstrap` build mode never calls `_export_gpg_publickey`, so even if
the GPG key is passed with the `-g` option and thus the `gpg_key` variable
is set, the `${work_dir}/pubkey.gpg` file will not exist.
This has not caused any issue so far because the `ARCHISO_GNUPG_FD` file
descriptor opens the file for both reading and writing, which means the
file gets created if it does not exist.
Assign the exported public key file name to a `gpg_publickey` variable in
`_export_gpg_publickey` and check for it when the file is used.
Since the exist status of the gpg command cannot be checked, look for the
exported public key file instead.
Make sure existing sig files are deleted before creating new ones and make sure to not sign any sig files.
This allows retrying failed mkarchiso runs without ending up with files such as vmlinuz.ipxe.sig.ipxe.sig.
Fixes#198
(gitlab ci)
Added a CA structure to the codesigning certificates.
This to test the functionality of optional CA being in the signing message.
(mkarchiso)
Removed the ``sign_netboot_artifacts`` variable and instead
we'll now rely on ``if [[ -v cert_list ]]; then``.
Added ``ARCHISO_TLS_FD`` and ``ARCHISO_TLSCA_FD`` environment variables
to override the certificates used. This is so that third party CA's can
be used during building in a meaningful way without distrupting the
CA trust that is shipped by default.
_cms_sign_artifact() was added which signs the rootfs using OpenSSL CMS.
The files will be saved as "${artifact}.cms.sig". That would be for instance
"${isofs_dir}/${install_dir}/${arch}/airootfs.sfs.cms.sig".
archlinux-keyring-wkd-sync.service needs an initialized pacman keyring to work.
Add BindsTo=etc-pacman.d-gnupg.mount to stop pacman-init.service if the mount unit suddenly enters inactive state.
--disable-shim-lock is required to support Secure Boot with custom signatures without using shim.
Otherwise GRUB will trow an error when trying to boot a kernel:
error: shim_lock protocol not found.
error: you need to load the kernel first.
The modules GRUB will use need to be preloaded otherwise the EFI binaries cannot be signed and used for Secure Boot.
See https://bugs.archlinux.org/task/71382.
GRUB will trow en error:
error: verification requested but nobody cares
These changes are done to support Secure Boot using custom keys (not shim) by simply extracting the boot loader
(BOOTx64.EFI and BOOTIA32.EFI), kernel, UEFI shell, signing them and then repacking the ISO.
For example.
Extract the files:
$ osirrox -indev archlinux-YYYY.MM.DD-x86_64.iso \
-extract_boot_images ./ \
-extract /EFI/BOOT/BOOTx64.EFI BOOTx64.EFI \
-extract /EFI/BOOT/BOOTIA32.EFI BOOTIA32.EFI \
-extract /shellx64.efi shellx64.efi \
-extract /shellia32.efi shellia32.efi \
-extract /arch/boot/x86_64/vmlinuz-linux vmlinuz-linux
Make the files writable:
$ chmod +w BOOTx64.EFI BOOTIA32.EFI shellx64.efi shellia32.efi vmlinuz-linux
Sign the files:
$ sbsign --key db.key --cert db.crt --output BOOTx64.EFI BOOTx64.EFI
$ sbsign --key db.key --cert db.crt --output BOOTIA32.EFI BOOTIA32.EFI
$ sbsign --key db.key --cert db.crt --output shellx64.efi shellx64.efi
$ sbsign --key db.key --cert db.crt --output shellia32.efi shellia32.efi
$ sbsign --key db.key --cert db.crt --output vmlinuz-linux vmlinuz-linux
Copy the boot loader and UEFI shell to the EFI system partition image:
$ mcopy -D oO -i eltorito_img2_uefi.img BOOTx64.EFI BOOTIA32.EFI ::/EFI/BOOT/
$ mcopy -D oO -i eltorito_img2_uefi.img shellx64.efi shellia32.efi ::/
Repack the ISO using the modified El Torito UEFI boot image and add the signed boot loader files, UEFI shell and
kernel to ISO9660:
$ xorriso -indev archlinux-YYYY.MM.DD-x86_64.iso \
-outdev archlinux-YYYY.MM.DD-x86_64-Secure_Boot.iso \
-boot_image any replay \
-append_partition 2 0xef eltorito_img2_uefi.img \
-map BOOTx64.EFI /EFI/BOOT/BOOTx64.EFI \
-map BOOTIA32.EFI /EFI/BOOT/BOOTIA32.EFI \
-map shellx64.efi /shellx64.efi \
-map shellia32.efi /shellia32.efi \
-map vmlinuz-linux /arch/boot/x86_64/vmlinuz-linux
Boot the resulting archlinux-YYYY.MM.DD-x86_64-Secure_Boot.iso.
Do not limit file copying to only grub.cfg and instead copy all GRUB configuration files and assets to both the ISO9660 and FAT image.
This will allow for including custom images, fonts, etc.
To easily match all non-configuration files (i.e. files without the .cfg extension), bash's extended glob feature will be enabled.
Actions common to multiple _make_bootmode_uefi-*.grub are split off into dedicated functions:
* _make_common_bootmode_grub_copy_to_efibootimg,
* _make_common_bootmode_grub_copy_to_isofs,
* _make_common_bootmode_grub_cfg.
Use the same du command in all efiboot_imgsize variable assignments.
Fixes#185.
The glibc 2.35-6 package ships with the C.UTF-8 locale included.
This means there is now a UTF-8 locale available by default and en_US.UTF-8, which requires editing /etc/locale.gen and running locale-gen, is not needed anymore.
Implements #175.
* open-vm-tools package, vmtoolsd.service and vmware-vmblock-fuse.service for VMware.
* hyperv package, hv_fcopy_daemon.service, hv_kvp_daemon.service and hv_vss_daemon.service for Hyper-V.
Related to #118.
When using tmpfs, it is possible that parts of it end up getting put in swap space (only if there is one).
This may not be desired, so use ramfs instead.
When booting the ISO, you can observe a message that systemd-gpt-auto-generator has failed:
systemd-gpt-auto-generator[197]: Reading EFI variable /sys/firmware/efi/efivars/LoaderDevicePartUUID-4a67b082-0a4c-41cf-b6c7-440b29bb8c4f.
systemd-gpt-auto-generator[197]: open("/sys/firmware/efi/efivars/LoaderDevicePartUUID-4a67b082-0a4c-41cf-b6c7-440b29bb8c4f") failed: No such file or directory
systemd-gpt-auto-generator[197]: EFI loader partition unknown, exiting.
systemd-gpt-auto-generator[197]: (The boot loader did not set EFI variable LoaderDevicePartUUID.)
systemd-gpt-auto-generator[197]: Failed to open device: No such device
Seeing as it started to appear relatively recently, it may be a systemd bug.
Since we do not want any GPT partition automounting in the live environment anyway, systemd-gpt-auto-generator can simply be disabled.
Fixes#164.
By default, syslinux interprets TFTP paths as relative to the location of the *.c32 modules, regardless whether a path starts with a slash. Without the `::` that I added to all of these paths, syslinux cannot find these files. It searches for them in /%INSTALL_DIR%/syslinux//%INSTALL_DIR%/boot/x86_64/vmlinuz-linux, etc.
