truenas/truenas-csi v1.1.0

v1.1.0 — NVMe-oF/TCP support

on GitHub

This release adds NVMe over Fabrics (TCP) as a third storage protocol alongside
NFS and iSCSI — higher-performance, lower-latency block storage backed by ZFS ZVOLs.

Requires TrueNAS SCALE 25.10.0 or newer. The driver verifies the server's
advertised API versions on startup and fails fast on older releases.

🚀 NVMe-oF/TCP

Select it per-StorageClass with parameters.protocol: nvmeof.

• Block and filesystem volume modes (ext4 / xfs)
• Volume expansion, snapshots, and clones — full parity with iSCSI
• DH-CHAP authentication — per-StorageClass nvmeof.hostNQN + nvmeof.dhchapKey,
  with optional mutual auth (nvmeof.dhchapCtrlKey) and hash / DH-group overrides.
  Subsystems with auth are created with allow_any_host=false; without it, the
  posture matches iSCSI-without-ACLs.
• Shared TCP listener per portal (:4420) and a shared nvmet.host per StorageClass
  with reference-counted cleanup; controller restarts reconstruct NVMe-oF volumes and
  disambiguate them from iSCSI ZVOLs
• Examples: examples/storageclass-nvmeof.yaml, examples/storageclass-nvmeof-dhchap.yaml

Prerequisites: enable the NVMe-oF service on TrueNAS (Services → NVMe-oF). The node
DaemonSet loads nvme_tcp / nvme_fabrics on the host. Set the portal via
TRUENAS_NVMEOF_PORTAL (or nvmeofPortal in the operator CR); it defaults to the
TrueNAS host on port 4420.

⚠️ Note for existing deployments

Pool-based node topology (topology.truenas.io/pool) has been removed so Pods using
volumes from multiple pools schedule correctly (#29). PersistentVolumes created before
v1.1.0 keep their stale pool nodeAffinity — they still schedule on existing nodes, but
recreate the PV (or remove the affinity) to drop the constraint.

📦 Images

Full Changelog: v1.0.0...v1.1.0