ZFS, Actually Storage you can finally reason about.

Mountpoints

concept
depth 3/3 − deeper + surface

A mountpoint is where a piece of storage shows up in your folder tree— the address at which its contents appear. On most systems, those addresses live in a settings file on the machine (on Linux, /etc/fstab), and people trained there go looking for it and find it empty. Actually, ZFS handles this itself: where a dataset appears is a setting stored on the dataset, inside the pool. One command changes it, and children follow their parent unless told otherwise.

The deeper shift is worth pausing on: the machine's settings file describes storage to that machine. A ZFS pool carries its own description—so the layout travels with the data, not with the computer it happens to be plugged into.

Working with it:

  • zfs get mountpoint tank/media—see where and why (the SOURCE column: inherited or local). Default is inherited path-building: tank/a/b mounts at /tank/a/b until some ancestor overrides.
  • zfs set mountpoint=...—takes effect immediately (it remounts), and children follow by inheritance.
  • The escape hatches, for the cases the automation shouldn't own: mountpoint=legacy hands one dataset back to fstab/mount (useful when boot ordering or another system insists), and canmount=off makes a dataset a pure structural parent—a policy node in the tree with nothing mounted (the trick behind tidy hierarchies like tank/services/*).

The payoff scenario: move the disks to a new machine, zpool import, and the entire mount layout arrives with them—nothing to reconstruct, no fstab archaeology (see Import and Export).

The one real collision to know about: ZFS refuses to mount over a non-empty directory. If something wrote files at /srv/media while the dataset wasn't mounted (a service that started too early, a restore to the wrong layer), the next mount fails or the data hides underneath—the classic "my files vanished" that's actually "my files are under the mount." The overlay property and systemd ordering exist for this boundary; the diagnostic habit is zfs mount + looking underneath before assuming loss.

Against its neighbors: fstab still owns swap and non-ZFS filesystems; /etc/zfs/zpool.cache (or its zgenhostid successor machinery) is just the host remembering which pools to import at boot—not mount config. The division of labor: the pool knows its layout; the host only knows which pools are its.