ZFS, Actually Storage you can finally reason about.

Zvols

definition
depth 2/3 − deeper + surface

If you run virtual machines—computers simulated in software—you've probably assumed each machine's disk is a big file sitting in a folder somewhere. On most systems, it is. On ZFS you can go hunting for that file and find nothing—because there is no file.

ZFS can serve up a slice of the pool as if it were a blank physical disk. That slice is a zvol. You hand it to anything that wants a whole disk of its own—most often a virtual machine—and it formats and uses it exactly as if it were real hardware. Underneath the pretense it's still ZFS: fingerprint-checked, freezable in time, copyable to another machine, like everything else in the pool.

Where you meet them: Proxmox carves one zvol per VM disk (that's why zfs list on a Proxmox host shows .../vm-103-disk-0 entries where you expected files). The practical grammar: created with zfs create -V 32G tank/vm-disk; snapshots and send/recv work exactly as for datasets—which is how a VM disk gets hourly history and offsite replication with the same machinery as everything else.

Two zvol-specific knobs worth knowing exist: volblocksize is the recordsize question in block-device form—fixed at creation, workload-matched (virtualizers set sane defaults); and thin vs thick (-s for sparse) decides whether the pool promises the space up front or lets you overcommit—overcommitted thin zvols on a filling pool are a classic slow-motion incident.