repartition: Correctly detect rootfs on loop device

[wjt]

After resizing the main partition, we have the following logic to invoke partprobe if it is on a loop device:

# Loop devices need a prod
if [ -n "$using_loop" ]; then
  partprobe $root_disk
  udevadm settle
fi

But empirically this block does not run on a fresh dual-boot installation: the kernel still believed the partition has its old size until I manually ran partprobe. As a result eos-firstboot fails to grow the main partition:

eos-firstboot[528]: resize2fs 1.47.0 (5-Feb-2023)
eos-firstboot[528]: The filesystem is already 11053568 (4k) blocks long.  Nothing to do!

$using_loop is conditionally set as follows:

case ${orig_root_part} in
  /dev/loop?p?)
    using_loop=1
    ;;
esac

$orig_root_part is set as follows:

  orig_root_part=$(systemctl show -p What sysroot.mount)
  orig_root_part=${orig_root_part#What=}

I ran this command once the system had booted and the output is:

What=/dev/disk/endless-image3

This is not too surprising because eos-image-boot-generator has the following:

cat <<EOF >${GENERATOR_DIR}/sysroot.mount
[Unit]
Before=initrd-root-fs.target
Requires=systemd-fsck-root.service
After=systemd-fsck-root.service
[Mount]
What=/dev/disk/endless-image3
Where=/sysroot
Options=$rwflag
EOF

So the fix is to follow any symbolic links so we can inspect the real device path. Conveniently we already do this:

# Identify root partition device node and parent disk
root_part=$(readlink -f "${orig_root_part}")

https://phabricator.endlessm.com/T35432