The layer that decides whether a pile of disks deserves trust

Turning a stack of disks into one large pool that can be relied on depends, more than people expect, on the file system laid over them. The hardware can only say that a disk answered; it cannot say whether what came back is what was written. OpenZFS is an open source advanced file system and volume manager that gathers snapshots, checksums, compression, deduplication and self repair into one piece of software, published under the CDDL, and it is the storage floor under several of the well known systems for keeping files on your own machine. It descends from a file system that came out of the workstation era, the community edition now runs on Linux and on BSD, and the stable line, of which 2.3.8 appeared in June 2026, is still maintained while the feature work has moved on to the next branch.

Integrity is the whole argument

An ordinary file system writes data without knowing whether the write went right, and reads it back without asking whether it is still correct. This one computes a checksum for every block, verifies that checksum whenever the block is read, and when it finds damage it repairs the block on the spot from a redundant copy or from the parity information in the pool. Silent corruption, the kind nobody notices until the day a backup is needed, therefore has nowhere to hide. Copy on write means the system is never halfway through a write: pull the power and the pool is still consistent, because the old state is still whole on the other side of the change. Snapshots and clones cost almost nothing, a snapshot being finished the moment it is asked for and taking no room of its own, so a file deleted by mistake or encrypted by something unpleasant can be taken back to any point in time.

What the current line added

Three things asked for over years arrived together. A redundant pool built from several disks can now be enlarged by adding a single disk, instead of being rebuilt as a whole by the old rule, so the owner of a small machine at home can grow the space without moving the data anywhere else first. Deduplication became an order of magnitude quicker, with a memory cost that is easier to live with. And direct input and output lets reads and writes go around the cache, which on a fast solid state disk means the hardware is no longer held back by the layer in front of it. The command line tool can print its answers as JSON, which makes it easy to feed a script or a monitor, and the limit on the length of a file name was raised to 1023 characters. A later small release also added a command that applies settings such as compression and record size to data that is already on the pool.

Who it is for, and how it is used

It suits a machine built at home to hold files, a small laboratory, a backup server and the host of a set of virtual machines: the turnkey storage systems that wrap it work out of the box, and on the mainstream Linux distributions it is a package to install. Because its licence does not sit comfortably inside the kernel, it exists as a module rather than as part of it, and taking the version the distribution has settled on is safer than chasing the newest. Two cautions are worth passing on. Deduplication and heavy compression want memory, so leave room in the plan before switching them on; and a machine doing real work should run the stable line rather than the newest experiment.

Maintenance, mostly quiet

Day to day it asks little of its keeper. A scheduled verification pass can walk the whole pool and check every block against its checksum without anyone watching, a failing disk is replaced and the data is rebuilt onto the new one, and none of this requires the machine to be taken out of service. A snapshot taken before an upgrade is a whole plan for going back, for the cost of a few seconds of attention. For anyone who cares about the data more than about the excitement of new features, this remains close to the reference answer among the things that cost nothing.