(Edited by @cgwalters)

@p5 experienced this: bootc-dev/bootc#1350 (comment)

Reproducer

• Set up Fedora Silverblue 42 via Anaconda with LUKS ("encrypt my disks"), otherwise accepting the other defaults (btrfs, etc.) or set up CentOS Stream 10 (again via Anaconda) using https://docs.fedoraproject.org/en-US/bootc/bare-metal/ (the stock quay.io/centos-bootc/centos-bootc:stream10 image is fine)
• Have ostree 2025.4 or newer

# Can be replaced with anything that gives a new userspace filesystem without changing the kernel
$ rpm-ostree install strace
$ ostree admin prepare-soft-reboot 0
$ systemctl reboot

The system ends up in emergency mode.

Non-reproducers (successful scenarios)

This does not reproduce when the system is provisioned without LUKS. It also doesn't reproduce on multiple instatations of CoreOS I've tried to set up (the defaults, using LUKS for / via tpm2 binding, having /var as a separate partition and both of those...)

Key findings

• One thing I notice with the Silverblue-LUKS scenario is that a simple reboot has multiple errors...we fail to unmount etc and sysroot and tear down the LUKS device. I haven't yet determined those are strictly related...because in a soft reboot scenario we actually expect to keep the LUKS device set up. Some of that is now being discussed in Fix unmount of /etc and /sysroot #3513
• The reason we go into emergency mode is because var.mount fails...but actually before that journald fails to start - and I think when journald fails to start we don't get any logs from that? (Needs investigation)