Most homelab Proxmox boxes start with everything on local disk, and that’s fine right up until there’s a second box, or a NAS already doing the storage job sitting idle on the network. Proxmox can mount that NAS directly as a storage backend through Datacenter > Storage > Add, with no Ceph cluster and no replication job required. The catch is that NFS, iSCSI, and SMB aren’t interchangeable once they’re wired in. Each one supports a different subset of content types, behaves differently under a live VM, and fails in a different way when the network hiccups. Picking the wrong one for the job looks fine on day one and turns into a surprise the first time something tries to snapshot a VM or the NAS reboots mid-write.

This is a different layer than the clustering and Ceph articles already on this site. Ceph and storage replication are about keeping VM disks redundant across multiple Proxmox nodes. This is about pointing Proxmox at storage that already exists somewhere else on the network and isn’t trying to be a redundant, Proxmox-aware pool at all, just a destination.

NFS: the easiest one to reason about

NFS is the backend most homelabs reach for first, because most NAS appliances (TrueNAS, Unraid, Synology, a plain Linux box) already speak it, and Proxmox’s NFS backend is genuinely full-featured. Add it under Datacenter > Storage > Add > NFS, point it at the NAS’s IP and export path, and Proxmox mounts it automatically on every node in the cluster.

What makes NFS the most flexible of the three: Proxmox can store almost every content type on it, VM disk images (as qcow2 files, not raw block), ISO images, backup archives, and container templates, all from the same export. qcow2 on NFS also supports snapshots, because the snapshot logic lives in the qcow2 file format itself rather than depending on the underlying filesystem, which is not true of the other two options below.

The gotchas sit on the NAS side, not the Proxmox side:

  • Export permissions. The NAS’s NFS export has to explicitly allow the Proxmox host’s IP (or subnet), and in many setups needs no_root_squash set, because Proxmox mounts and writes as root. Skip this and the storage will add successfully in the GUI but fail the moment something tries to actually write to it.
  • NFS version matters for locking behavior. NFSv4 handles file locking more predictably than NFSv3 under concurrent access, which matters more once more than one node is writing to the same export. Pin the version explicitly in the storage definition rather than letting negotiation pick one silently.
  • Latency, not throughput, is usually the real limit. A gigabit link has plenty of throughput for most homelab VM disk I/O, but NFS adds round-trip latency per operation that a local NVMe drive never has. A VM that’s fine running from local SSD can feel noticeably sluggish on NFS for latency-sensitive workloads (databases doing lots of small synchronous writes), even when the network is nowhere near saturated.

SMB/CIFS: fine for files, a poor choice for live VM disks

Proxmox also supports CIFS as a storage backend, and it’s worth adding specifically because plenty of existing home NAS setups are SMB-first (Windows-adjacent households, or a NAS where CIFS shares were already set up for other reasons before Proxmox entered the picture). It’s genuinely useful for ISO images, backup archives, and container templates, content that’s written once and read back wholesale, with no concurrent fine-grained access.

Where it gets shaky is VM disk images. CIFS’s file-locking model wasn’t designed around one process holding a large file open and doing constant small random writes into it the way a running VM does, and permission/ownership mapping between the Proxmox host’s Linux UID and whatever the SMB server expects adds another place for things to go subtly wrong (a VM disk that works fine until a permissions quirk on the NAS side corrupts a write). It’s not that it never works, plenty of homelabs run VM disks on CIFS without incident, it’s that NFS does the same job more predictably for the same NAS hardware, so there’s rarely a good reason to pick CIFS for VM disks specifically when NFS is on the table. Reach for CIFS when the NAS genuinely only speaks SMB, or when the use is backups/ISOs/templates and nothing more demanding than that.

iSCSI: block storage, which changes the whole model

iSCSI is a different kind of backend entirely. NFS and CIFS hand Proxmox a filesystem it can drop files into. iSCSI hands Proxmox a raw block device, a LUN, with no filesystem on it at all. That one difference changes almost everything about how it has to be used.

Because there’s no filesystem, Proxmox can’t just drop multiple VM disk image files onto an iSCSI LUN the way it does on NFS. The standard pattern is layering LVM on top of the iSCSI LUN, with each VM disk becoming its own LVM logical volume carved out of that one big block device. This works and it’s a legitimate, sometimes-necessary pattern (SAN gear and some higher-end NAS boxes offer iSCSI, not NFS, as their primary block-storage interface), but it comes with real limitations on plain LVM: no snapshot support without LVM-thin provisioning, and even then snapshot behavior and performance under LVM-thin isn’t the same as qcow2-on-NFS or ZFS’s native snapshots. If snapshots matter and the NAS can do NFS instead, NFS is usually the simpler answer.

iSCSI’s other practical considerations:

  • CHAP authentication should be enabled rather than left open, since anything on the same VLAN that can reach the target IP can otherwise attempt to connect to the LUN.
  • Multipathing is worth setting up if the NAS and switch support it, iSCSI over a single NIC path has no redundancy if that link or port dies mid-session, unlike NFS where a network blip is usually just a stall, not a hard failure.
  • One LUN, one consumer, usually. Having multiple hosts write to the same iSCSI LUN without a cluster-aware filesystem or LVM-with-clustering support on top is a fast way to corrupt data. This is a block device, not a shared filesystem, and it doesn’t arbitrate concurrent writers on its own.

Picking one

For most homelabs adding an existing NAS to Proxmox as a storage backend, NFS is the right default: broadest content-type support, snapshot support via qcow2, and the simplest mental model (it’s just a network filesystem). Reach for CIFS specifically when the NAS only offers SMB, and scope it to backups, ISOs, and templates rather than live VM disks. Reach for iSCSI only when the storage device is block-only by design (a SAN, or NAS software that doesn’t expose NFS), accept the LVM layer and the snapshot tradeoff that comes with it, and take multipathing and CHAP seriously since you’re now depending on one network path to a raw block device with none of NFS’s built-in slack for a bad moment on the wire.

None of these three turn a single NAS into redundant storage. They just move where the files live. If the NAS itself dies, the Proxmox storage backend dies with it, same as local disk would, just further away. That’s a separate problem from the one this solves, and it’s why Ceph and storage replication exist as their own, heavier answers when actual redundancy across nodes is the goal rather than simply centralizing storage you already have.