I used "Guided - use entire disk and set up encrypted LVM" On April 25, 2023 12:55:27 PM CDT, Pascal Hambourg <pas...@plouf.fr.eu.org> wrote: >Hello, > >On 25/04/2023 at 01:29, Adam wrote: >> >> I installed from a thumb drive, to another thumb drive, on a computer >> that had an nvme drive that should not have been touched. The installer >> overwrote data on the nvme drive despite the target being /dev/sda. I >> manually mounted the installed system (the target thumb drive) on another >> computer, figured out what happened (ESP was empty) and fixed it so I could >> submit a bug report from the thumb drive that failed to install properly. >> >> This is similar to the other UEFI installation problems, but it did >> not install to the MBR, and it did not install any files on the correct >> ESP, thus is is a separate issue. >> >> The smoking gun for understanding what went wrong was in /etc/fstab, where >> there were two comments: >> >> # /boot was on /dev/sda2 during installation >> # /boot/efi was on /dev/nvme0n1p1 during installation > >From the partition layout I assume you used guided partitioning with LVM >(without encryption). Guided partitioning is supposed to not use any >partitions outside the selected disk by calling clean_method() defined in >partman-auto/lib/recipes.sh. This is what I observe with non-LVM schemes, but >the two LVM schemes have issues. Here is a summary of my observations: > >Guided - use the largest continuous free space > calls clean_method() in partman-auto/autopartition > does not run partman-efi/init.d/efi > does not use existing EFI or swap partitions on other disks (good) > >Guided - use entire disk > calls clean_method() in partman-auto/autopartition > does not run partman-efi/init.d/efi > does not use existing EFI or swap partitions on other disks (good) > >Guided - use entire disk and set up LVM > does not call clean_method() > runs partman-efi/init.d/efi > uses existing EFI and swap partitions on other disks (bad) > >Guided - use entire disk and set up encrypted LVM > calls clean_method() in partman-auto-crypto/autopartition-crypto > runs partman-efi/init.d/efi > uses existing EFI partitions on other disks (bad) > does not use existing swap partitions on other disks (good) > >partman-efi/init.d/efi detects possible EFI partitions and sets method "efi" >on them. > >As you can see, the issue also affects swap partitions (and they will be >reformatted with new UUIDs, which can be harmful if they are used by another >system). > >Note: partman-auto-lvm used to call clean_method() in lib/auto-lvm.sh but it >was removed by commit cfc6797f6f561b87069160ba7c375c5b487b7c1e with code >factoring. > >Suggested fix is two-fold: > >1) Call clean_method() at the beginning of partman-auto-lvm/autopartition-lvm, >as is done in partman-auto/autopartition and >partman-auto-crypto/autopartition-crypto. This should solve the issue for swap >partitions but is not enough for ESPs. > >2) In partman-efi/init.d/efi, set method "efi" only once, as is done with swap >partitions in partman-basicfilesystems/init.d/autouse_swap. >I already submitted two patch versions for #1034208 "Partman may reset user's >choice for ESP partitions use" as a follow-up to Steve's latest fixes for >#834373 and #1033913. > >Caveat: I don't know if these changes could have any negative impact on >preseeded automatic partitioning.
-- Sent from my iPod. Please excuse my brevity.
