raid1 boot_device mirroring with partitions > 2TB breaks install

[stephankoelle]

Describe the bug
When using raid1 boot_device mirroring together with partition sizes > 2TB there is file corruption.

Reproduction steps
Steps to reproduce the behavior:

1. using a host with two NVMEs > 2 TB

variant: fcos
version: 1.4.0
boot_device:
  mirror:
    devices:
      - /dev/nvme0n1
      - /dev/nvme1n1

3. /usr/bin/rpm-ostree install --apply-live --allow-inactive --idempotent htop
4. error: Running tmpfiles for /run and /var: Running systemd worker: exit status: 127

After that system is broker, no further ssh login possible (everything gives "invalid ELF header")

Expected behavior
boot_device mirroring should work with "arbitrary" sizes partition

Actual behavior
System is completely broker, needs re-install where partition is bigger then "size_mib: 2097000". Have not tested the exact value, OK: size_mib: 2097000 , failure: size_mib: 2621000

System details

• Bare Metal & QEMU (with mounted NVMEs)
• Fedora CoreOS 35.20220424.3.0"

Ignition config

variant: fcos
version: 1.4.0
boot_device:
  mirror:
    devices:
      - /dev/sda
      - /dev/sdb

Additional information
Work around, reduce size of root partition

storage:
  disks:
    - device: /dev/nvme0n1
      partitions:
        - label: root-1
          #defect 2621000 => error: Running tmpfiles for /run and /var: Running systemd worker: exit status: 127
          #ok 2097000
          size_mib: 2097000
    - device: /dev/nvme1n1
      partitions:
        - label: root-2
          size_mib: 2097000

Forum link with more information: https://discussion.fedoraproject.org/t/fedora-coreos-installation-works-preparing-etc-invalid-byte-sequence-in-conversion-input-on-rpm-ostree-install-afterwards/38772/14

[travier]

See #1183 as it might be related.

[bgilbert]

boot_device directives recreate the root filesystem, so #1183 is unlikely to be related.