Creating new BE’s using snapshots and clones can get messy with dependencies all over the place. I had an epiphany the other day, why not create a snapshot on the current dataset, send that snapshot to a new dataset within the same zpool (or elsewhere), and subsequently destroy the (two!) snapshots?

Instant transfer of data, and no strings attached to either dataset once the snapshots have been removed.

zfs snapshot zroot/ROOT/current-be@transfer
zfs send     zroot/ROOT/current-be@transfer | zfs receive -uv zroot/ROOT/new-be
zfs destroy  zroot/ROOT/current-be@transfer
zfs destroy  zroot/ROOT/new-be@transfer

Maybe a new zfs copy command should be created to simplify the typing when copying datasets within a zpool.

Proceed with:

  • Mount the new BE on, say, /new-be.
  • Running make -C /usr/src DESTDIR=/new-be installworld installkernel, ensures new versions overwriting old versions.
  • Running mergemaster -Fi -D /new-be, ensures configuration changes are merged into the new BE.
  • Running make -C /usr/src DESTDIR=/new-be delete-old, ensures obsolete stuff are removed.
  • Inherit the mountpoint property for new BE, or explicitly set the mountpoint property to legacy.
  • Update the bootfs property on the zpool to point to the new BE, and reboot to begin using the new BE.

If you’re not short on disk space, the above suggestion might be an acceptable compromise.