CephFS namespaces can be exported over NFS protocol using the NFS-Ganesha NFS server.


  • Ceph file system (preferably latest stable luminous or higher versions)

  • In the NFS server host machine, ‘libcephfs2’ (preferably latest stable luminous or higher), ‘nfs-ganesha’ and ‘nfs-ganesha-ceph’ packages (latest ganesha v2.5 stable or higher versions)

  • NFS-Ganesha server host connected to the Ceph public network

Configuring NFS-Ganesha to export CephFS

NFS-Ganesha provides a File System Abstraction Layer (FSAL) to plug in different storage backends. FSAL_CEPH is the plugin FSAL for CephFS. For each NFS-Ganesha export, FSAL_CEPH uses a libcephfs client, user-space CephFS client, to mount the CephFS path that NFS-Ganesha exports.

Setting up NFS-Ganesha with CephFS, involves setting up NFS-Ganesha’s configuration file, and also setting up a Ceph configuration file and cephx access credentials for the Ceph clients created by NFS-Ganesha to access CephFS.

NFS-Ganesha configuration

A sample ganesha.conf configured with FSAL_CEPH can be found here, It is suitable for a standalone NFS-Ganesha server, or an active/passive configuration of NFS-Ganesha servers managed by some sort of clustering software (e.g., Pacemaker). Important details about the options are added as comments in the sample conf. There are options to do the following:

  • minimize Ganesha caching wherever possible since the libcephfs clients (of FSAL_CEPH) also cache aggressively

  • read from Ganesha config files stored in RADOS objects

  • store client recovery data in RADOS OMAP key-value interface

  • mandate NFSv4.1+ access

  • enable read delegations (need at least v13.0.1 ‘libcephfs2’ package and v2.6.0 stable ‘nfs-ganesha’ and ‘nfs-ganesha-ceph’ packages)

Configuration for libcephfs clients

Required ceph.conf for libcephfs clients includes:

  • a [client] section with mon_host option set to let the clients connect to the Ceph cluster’s monitors, usually generated via ceph config generate-minimal-conf, e.g.,

            mon host = [v2:,v1:], [v2:,v1:], [v2:,v1:]

Mount using NFSv4 clients

It is preferred to mount the NFS-Ganesha exports using NFSv4.1+ protocols to get the benefit of sessions.

Conventions for mounting NFS resources are platform-specific. The following conventions work on Linux and some Unix platforms:

From the command line:

mount -t nfs -o nfsvers=4.1,proto=tcp <ganesha-host-name>:<ganesha-pseudo-path> <mount-point>

Current limitations

  • Per running ganesha daemon, FSAL_CEPH can only export one Ceph file system although multiple directories in a Ceph file system may be exported.

Exporting over NFS clusters deployed using rook

This tutorial assumes you have a kubernetes cluster deployed. If not minikube can be used to setup a single node cluster. In this tutorial minikube is used.


Configuration of this tutorial should not be used in a a real production cluster. For the purpose of simplification, the security aspects of Ceph are overlooked in this setup.

Rook Setup And Cluster Deployment

Clone the rook repository:

git clone

Deploy the rook operator:

cd cluster/examples/kubernetes/ceph
kubectl create -f common.yaml
kubectl create -f operator.yaml


Nautilus release or latest Ceph image should be used.

Before proceding check if the pods are running:

kubectl -n rook-ceph get pod


For troubleshooting on any pod use:

kubectl describe -n rook-ceph pod <pod-name>

If using minikube cluster change the dataDirHostPath to /data/rook in cluster-test.yaml file. This is to make sure data persists across reboots.

Deploy the ceph cluster:

kubectl create -f cluster-test.yaml

To interact with Ceph Daemons, let’s deploy toolbox:

kubectl create -f ./toolbox.yaml

Exec into the rook-ceph-tools pod:

kubectl -n rook-ceph exec -it $(kubectl -n rook-ceph get pod -l "app=rook-ceph-tools" -o jsonpath='{.items[0]}') bash

Check if you have one Ceph monitor, manager, OSD running and cluster is healthy:

[root@minikube /]# ceph -s
        id:     3a30f44c-a9ce-4c26-9f25-cc6fd23128d0
        health: HEALTH_OK

        mon: 1 daemons, quorum a (age 14m)
        mgr: a(active, since 13m)
        osd: 1 osds: 1 up (since 13m), 1 in (since 13m)

        pools:   0 pools, 0 pgs
        objects: 0 objects, 0 B
        usage:   5.0 GiB used, 11 GiB / 16 GiB avail


Single monitor should never be used in real production deployment. As it can cause single point of failure.

Create a Ceph File System

Using ceph-mgr volumes module, we will create a ceph file system:

[root@minikube /]# ceph fs volume create myfs

By default replicated size for OSD is 3. Since we are using only one OSD. It can cause error. Let’s fix this up by setting replicated size to 1.:

[root@minikube /]# ceph osd pool set cephfs.myfs.meta size 1
[root@minikube /]# ceph osd pool set size 1


The replicated size should never be less than 3 in real production deployment.

Check Cluster status again:

[root@minikube /]# ceph -s
    id:     3a30f44c-a9ce-4c26-9f25-cc6fd23128d0
    health: HEALTH_OK

    mon: 1 daemons, quorum a (age 27m)
    mgr: a(active, since 27m)
    mds: myfs:1 {0=myfs-a=up:active} 1 up:standby-replay
    osd: 1 osds: 1 up (since 56m), 1 in (since 56m)

    pools:   2 pools, 24 pgs
    objects: 22 objects, 2.2 KiB
    usage:   5.1 GiB used, 11 GiB / 16 GiB avail
    pgs:     24 active+clean

    client:   639 B/s rd, 1 op/s rd, 0 op/s wr

Create a NFS-Ganesha Server Cluster

Add Storage for NFS-Ganesha Servers to prevent recovery conflicts:

[root@minikube /]# ceph osd pool create nfs-ganesha 64
pool 'nfs-ganesha' created
[root@minikube /]# ceph osd pool set nfs-ganesha size 1
[root@minikube /]# ceph orch nfs add mynfs nfs-ganesha ganesha

Here we have created a NFS-Ganesha cluster called “mynfs” in “ganesha” namespace with “nfs-ganesha” OSD pool.

Scale out NFS-Ganesha cluster:

[root@minikube /]# ceph orch nfs update mynfs 2

Configure NFS-Ganesha Exports

Initially rook creates ClusterIP service for the dashboard. With this service type, only the pods in same kubernetes cluster can access it.

Expose Ceph Dashboard port:

kubectl patch service -n rook-ceph -p '{"spec":{"type": "NodePort"}}' rook-ceph-mgr-dashboard
kubectl get service -n rook-ceph rook-ceph-mgr-dashboard
NAME                      TYPE       CLUSTER-IP       EXTERNAL-IP   PORT(S)          AGE
rook-ceph-mgr-dashboard   NodePort   <none>        8443:31727/TCP   117m

This makes the dashboard reachable outside kubernetes cluster and the service type is changed to NodePort service.

Create JSON file for dashboard:

$ cat ~/export.json
      "cluster_id": "mynfs",
      "path": "/",
      "fsal": {"name": "CEPH", "user_id":"admin", "fs_name": "myfs", "sec_label_xattr": null},
      "pseudo": "/cephfs",
      "tag": null,
      "access_type": "RW",
      "squash": "no_root_squash",
      "protocols": [4],
      "transports": ["TCP"],
      "security_label": true,
      "daemons": ["mynfs.a", "mynfs.b"],
      "clients": []


Don’t use this JSON file for real production deployment. As here the ganesha servers are given client-admin access rights.

We need to download and run this script to pass the JSON file contents. Dashboard creates NFS-Ganesha export file based on this JSON file.:

./ POST /api/nfs-ganesha/export "$(cat <json-file-path>)"

Expose the NFS Servers:

kubectl patch service -n rook-ceph -p '{"spec":{"type": "NodePort"}}' rook-ceph-nfs-mynfs-a
kubectl patch service -n rook-ceph -p '{"spec":{"type": "NodePort"}}' rook-ceph-nfs-mynfs-b
kubectl get services -n rook-ceph rook-ceph-nfs-mynfs-a rook-ceph-nfs-mynfs-b
NAME                    TYPE       CLUSTER-IP       EXTERNAL-IP   PORT(S)          AGE
rook-ceph-nfs-mynfs-a   NodePort   <none>        2049:31013/TCP   72m
rook-ceph-nfs-mynfs-b   NodePort     <none>        2049:31587/TCP   63m


Ports are chosen at random by Kubernetes from a certain range. Specific port number can be added to nodePort field in spec.

Testing access to NFS Servers

Open a root shell on the host and mount one of the NFS servers:

mkdir -p /mnt/rook
mount -t nfs -o port=31013 $(minikube ip):/cephfs /mnt/rook

Normal file operations can be performed on /mnt/rook if the mount is successful.


If minikube is used then VM host is the only client for the servers. In a real kubernetes cluster, multiple hosts can be used as clients, only when kubernetes cluster node IP addresses are accessible to them.