manila/doc/source/devref/cephfs_driver.rst
Ramana Raja 23075e6c0b Refactor and rename CephFSNativeDriver
Refactor CephFSNativeDriver as a driver class using protocol helper
classes. The helper classes would handle protocol specific driver
actions such as controlling access and fetching share's export
locations. For now, the driver uses a protocol helper to support
CephFS's native protocol. The driver can be made to support other
NAS protocols later on by adding protocol helper classes.

Since the driver would not just support the native protocol
rename the driver's file name and its driver class as
`driver` and `CephFSDriver` respectively. The driver would by
default support the native protocol, and can be referred to
by its previous class name and module name.

DocImpact

Partially-impelements: blueprint cephfs-nfs-support

Change-Id: I8a33be1df4864131435b794e791cc2d651fbe741
2017-04-18 10:45:25 +05:30

11 KiB

CephFS driver

The CephFS driver enables manila to export shared filesystems to guests using the Ceph network protocol. Guests require a Ceph client in order to mount the filesystem.

Access is controlled via Ceph's cephx authentication system. When a user requests share access for an ID, Ceph creates a corresponding Ceph auth ID and a secret key, if they do not already exist, and authorizes the ID to access the share. The client can then mount the share using the ID and the secret key.

To learn more about configuring Ceph clients to access the shares created using this driver, please see the Ceph documentation( http://docs.ceph.com/docs/master/cephfs/). If you choose to use the kernel client rather than the FUSE client, the share size limits set in manila may not be obeyed.

Supported Operations

The following operations are supported with CephFS backend:

  • Create/delete CephFS share
  • Allow/deny CephFS share access
    • Only cephx access type is supported for CephFS protocol.
    • read-only access level is supported in Newton or later versions of manila.
    • read-write access level is supported in Mitaka or later versions of manila.
  • Extend/shrink share
  • Create/delete snapshot
  • Create/delete consistency group (CG)
  • Create/delete CG snapshot

Prerequisites

  • Mitaka or later versions of manila.
  • Jewel or later versions of Ceph.
  • A Ceph cluster with a filesystem configured ( http://docs.ceph.com/docs/master/cephfs/createfs/)
  • ceph-common package installed in the servers running the manila-share service.
  • Ceph client installed in the guest, preferably the FUSE based client, ceph-fuse.
  • Network connectivity between your Ceph cluster's public network and the servers running the manila-share service.
  • Network connectivity between your Ceph cluster's public network and guests.

Important

A manila share backed onto CephFS is only as good as the underlying filesystem. Take care when configuring your Ceph cluster, and consult the latest guidance on the use of CephFS in the Ceph documentation ( http://docs.ceph.com/docs/master/cephfs/)

Authorize the driver to communicate with Ceph

Run the following commands to create a Ceph identity for manila to use:

read -d '' MON_CAPS << EOF
allow r,
allow command "auth del",
allow command "auth caps",
allow command "auth get",
allow command "auth get-or-create"
EOF

ceph auth get-or-create client.manila -o manila.keyring \
mds 'allow *' \
osd 'allow rw' \
mon "$MON_CAPS"

manila.keyring, along with your ceph.conf file, will then need to be placed on the server running the manila-share service.

Enable snapshots in Ceph if you want to use them in manila:

ceph mds set allow_new_snaps true --yes-i-really-mean-it

In the server running the manila-share service, you can place the ceph.conf and manila.keyring files in the /etc/ceph directory. Set the same owner for the manila-share process and the manila.keyring file. Add the following section to the ceph.conf file.

[client.manila]
client mount uid = 0
client mount gid = 0
log file = /opt/stack/logs/ceph-client.manila.log
admin socket = /opt/stack/status/stack/ceph-$name.$pid.asok
keyring = /etc/ceph/manila.keyring

It is advisable to modify the Ceph client's admin socket file and log file locations so that they are co-located with manila services's pid files and log files respectively.

Configure CephFS backend in manila.conf

Add CephFS to enabled_share_protocols (enforced at manila api layer). In this example we leave NFS and CIFS enabled, although you can remove these if you will only use CephFS:

enabled_share_protocols = NFS,CIFS,CEPHFS

Create a section like this to define a CephFS backend:

[cephfs1]
driver_handles_share_servers = False
share_backend_name = CEPHFS1
share_driver = manila.share.drivers.cephfs.driver.CephFSDriver
cephfs_conf_path = /etc/ceph/ceph.conf
cephfs_auth_id = manila
cephfs_cluster_name = ceph
cephfs_enable_snapshots = True

Set driver-handles-share-servers to False as the driver does not manage the lifecycle of share-servers. To let the driver perform snapshot related operations, set cephfs_enable_snapshots to True.

Then edit enabled_share_backends to point to the driver's backend section using the section name. In this example we are also including another backend ("generic1"), you would include whatever other backends you have configured.

Note

For Mitaka, Newton, and Ocata releases, the share_driver path was manila.share.drivers.cephfs.cephfs_native.CephFSNativeDriver

enabled_share_backends = generic1, cephfs1

Creating shares

The default share type may have driver_handles_share_servers set to True. Configure a share type suitable for cephfs:

manila type-create cephfstype false

Then create yourself a share:

manila create --share-type cephfstype --name cephshare1 cephfs 1

Note the export location of the share:

manila share-export-location-list cephshare1

The export location of the share contains the Ceph monitor (mon) addresses and ports, and the path to be mounted. It is of the form, {mon ip addr:port}[,{mon ip addr:port}]:{path to be mounted}

Allowing access to shares

Allow Ceph auth ID alice access to the share using cephx access type.

manila access-allow cephshare1 cephx alice

Note the access status, and the access/secret key of alice.

manila access-list cephshare1

Note

In Mitaka release, the secret key is not exposed by any manila API. The Ceph storage admin needs to pass the secret key to the guest out of band of manila. You can refer to the link below to see how the storage admin could obtain the secret key of an ID. http://docs.ceph.com/docs/jewel/rados/operations/user-management/#get-a-user

Alternatively, the cloud admin can create Ceph auth IDs for each of the tenants. The users can then request manila to authorize the pre-created Ceph auth IDs, whose secret keys are already shared with them out of band of manila, to access the shares.

Following is a command that the cloud admin could run from the server running the manila-share service to create a Ceph auth ID and get its keyring file.

ceph --name=client.manila --keyring=/etc/ceph/manila.keyring auth \
get-or-create client.alice -o alice.keyring

For more details, please see the Ceph documentation. http://docs.ceph.com/docs/jewel/rados/operations/user-management/#add-a-user

Mounting shares using FUSE client

Using the secret key of the authorized ID alice create a keyring file, alice.keyring like:

[client.alice]
        key = AQA8+ANW/4ZWNRAAOtWJMFPEihBA1unFImJczA==

Using the mon IP addresses from the share's export location, create a configuration file, ceph.conf like:

[client]
        client quota = true
        mon host = 192.168.1.7:6789, 192.168.1.8:6789, 192.168.1.9:6789

Finally, mount the filesystem, substituting the filenames of the keyring and configuration files you just created, and substituting the path to be mounted from the share's export location:

sudo ceph-fuse ~/mnt \
--id=alice \
--conf=./ceph.conf \
--keyring=./alice.keyring \
--client-mountpoint=/volumes/_nogroup/4c55ad20-9c55-4a5e-9233-8ac64566b98c

Known restrictions

Consider the driver as a building block for supporting multi-tenant workloads in the future. However, it can be used in private cloud deployments.

  • The guests have direct access to Ceph's public network.
  • The snapshot support of the driver is disabled by default. cephfs_enable_snapshots configuration option needs to be set to True to allow snapshot operations.
  • Snapshots are read-only. A user can read a snapshot's contents from the .snap/{manila-snapshot-id}_{unknown-id} folder within the mounted share.
  • To restrict share sizes, CephFS uses quotas that are enforced in the client side. The CephFS clients are relied on to respect quotas.

Mitaka release

  • The secret-key of a Ceph auth ID required to mount a share is not exposed to an user by a manila API. To workaround this, the storage admin would need to pass the key out of band of manila, or the user would need to use the Ceph ID and key already created and shared with her by the cloud admin.

Security

  • Each share's data is mapped to a distinct Ceph RADOS namespace. A guest is restricted to access only that particular RADOS namespace. http://docs.ceph.com/docs/master/cephfs/file-layouts/

  • An additional level of resource isolation can be provided by mapping a share's contents to a separate RADOS pool. This layout would be be preferred only for cloud deployments with a limited number of shares needing strong resource separation. You can do this by setting a share type specification, cephfs:data_isolated for the share type used by the cephfs driver.

    manila type-key cephfstype set cephfs:data_isolated=True
  • As mentioned earlier, untrusted manila guests pose security risks to the Ceph storage cluster as they would have direct access to the cluster's public network.

The manila.share.drivers.cephfs.driver Module

manila.share.drivers.cephfs.driver