Notice

This document is for a development version of Ceph.

mount.ceph -- mount a Ceph file system

Synopsis

mount.ceph name*@*fsid.*fs_name*=/[subdir] dir [-o options ]

Description

mount.ceph is a helper for mounting the Ceph file system on a Linux host. It serves to resolve monitor hostname(s) into IP addresses and read authentication keys from disk; the Linux kernel client component does most of the real work. To mount a Ceph file system use:

mount.ceph name@07fe3187-00d9-42a3-814b-72a4d5e7d5be.fs_name=/ /mnt/mycephfs -o mon_addr=1.2.3.4

where “name” is the RADOS client name (referred to hereafter as “RADOS user”, and meaning any individual or system actor such as an application).

Mount helper can fill in the cluster FSID by reading the ceph configuration file. Its recommended to call the mount helper via mount(8) as per:

mount -t ceph name@.fs_name=/ /mnt/mycephfs -o mon_addr=1.2.3.4

Note that the dot . still needs to be a part of the device string in this case.

The first argument is the device part of the mount command. It includes the RADOS user for authentication, the file system name and a path within CephFS that will be mounted at the mount point.

Monitor addresses can be passed using mon_addr mount option. Multiple monitor addresses can be passed by separating addresses with a slash (/). Only one monitor is needed to mount successfully; the client will learn about all monitors from any responsive monitor. However, it is a good idea to specify more than one in case the one happens to be down at the time of mount. Monitor addresses takes the form ip_address[:port]. If the port is not specified, the Ceph default of 6789 is assumed.

If monitor addresses are not specified, then mount.ceph will attempt to determine monitor addresses using local configuration files and/or DNS SRV records. In similar way, if authentication is enabled on Ceph cluster (which is done using CephX) and options secret and secretfile are not specified in the command, the mount helper will spawn a child process that will use the standard Ceph library routines to find a keyring and fetch the secret from it (including the monitor address and FSID if those not specified).

A sub-directory of the file system can be mounted by specifying the (absolute) path to the sub-directory right after “=” in the device part of the mount command.

Mount helper application conventions dictate that the first two options are device to be mounted and the mount point for that device. Options must be passed only after these fixed arguments.

Options

Basic

conf

Path to a ceph.conf file. This is used to initialize the Ceph context for autodiscovery of monitor addresses and auth secrets. The default is to use the standard search path for ceph.conf files.

mount_timeout

int (seconds), Default: 60

ms_mode=<legacy|crc|secure|prefer-crc|prefer-secure>

Set the connection mode that the client uses for transport. The available modes are:

legacy: use messenger v1 protocol to talk to the cluster
crc: use messenger v2, without on-the-wire encryption
secure: use messenger v2, with on-the-wire encryption
prefer-crc: crc mode, if denied agree to secure mode
prefer-secure: secure mode, if denied agree to crc mode

mon_addr

Monitor address of the cluster in the form of ip_address[:port]

fsid

Cluster FSID. This can be found using ceph fsid command.

secret

secret key for use with CephX. This option is insecure because it exposes the secret on the command line. To avoid this, use the secretfile option.

secretfile

path to file containing the secret key to use with CephX

recover_session=<no|clean>

Set auto reconnect mode in the case where the client is blocklisted. The available modes are no and clean. The default is no.

no: never attempt to reconnect when client detects that it has been blocklisted. Blocklisted clients will not attempt to reconnect and their operations will fail too.
clean: client reconnects to the Ceph cluster automatically when it detects that it has been blocklisted. During reconnect, client drops dirty data/metadata, invalidates page caches and writable file handles. After reconnect, file locks become stale because the MDS loses track of them. If an inode contains any stale file locks, read/write on the inode is not allowed until applications release all stale file locks.

command: fs=<fs-name> Specify the non-default file system to be mounted, when using the old syntax.
command: mds_namespace=<fs-name> A synonym of “fs=” (Deprecated).

Advanced

cap_release_safety

int, Default: calculated

caps_wanted_delay_max

int, cap release delay, Default: 60

caps_wanted_delay_min

int, cap release delay, Default: 5

dirstat

funky cat dirname for stats, Default: off

nodirstat

no funky cat dirname for stats

ip

my ip

noasyncreaddir

no dcache readdir

nocrc

no data crc on writes

noshare

create a new client instance, instead of sharing an existing instance of a client mounting the same cluster

osdkeepalive

int, Default: 5

osd_idle_ttl

int (seconds), Default: 60

rasize

int (bytes), max readahead. Default: 8388608 (8192*1024)

rbytes

Report the recursive size of the directory contents for st_size on directories. Default: off

norbytes

Do not report the recursive size of the directory contents for st_size on directories.

readdir_max_bytes

int, Default: 524288 (512*1024)

readdir_max_entries

int, Default: 1024

rsize

int (bytes), max read size. Default: 16777216 (16*1024*1024)

snapdirname

string, set the name of the hidden snapdir. Default: .snap

write_congestion_kb

int (kb), max writeback in flight. scale with available memory. Default: calculated from available memory

wsize

int (bytes), max write size. Default: 16777216 (16*1024*1024) (writeback uses smaller of wsize and stripe unit)

wsync

Execute all namespace operations synchronously. This ensures that the namespace operation will only complete after receiving a reply from the MDS.

nowsync

Allow the client to do namespace operations asynchronously. When this option is enabled, a namespace operation may complete before the MDS replies, if it has sufficient capabilities to do so. This has been the default since kernel version 5.16.

crush_location=x

Specify the location of the client in terms of CRUSH hierarchy (since 5.8). This is a set of key-value pairs separated from each other by ‘|’, with keys separated from values by ‘:’. Note that ‘|’ may need to be quoted or escaped to avoid it being interpreted as a pipe by the shell. The key is the bucket type name (e.g. rack, datacenter or region with default bucket types) and the value is the bucket name. For example, to indicate that the client is local to rack “myrack”, data center “mydc” and region “myregion”:

crush_location=rack:myrack|datacenter:mydc|region:myregion

Each key-value pair stands on its own: “myrack” doesn’t need to reside in “mydc”, which in turn doesn’t need to reside in “myregion”. The location is not a path to the root of the hierarchy but rather a set of nodes that are matched independently. “Multipath” locations are supported, so it is possible to indicate locality for multiple parallel hierarchies:

crush_location=rack:myrack1|rack:myrack2|datacenter:mydc

read_from_replica=<no|balance|localize>

no: Disable replica reads, always pick the primary OSD (since 5.8, default).
balance: When a replicated pool receives a read request, pick a random OSD from the PG’s acting set to serve it (since 5.8).

This mode is safe for general use only since Octopus (i.e. after “ceph osd require-osd-release octopus”). Otherwise it should be limited to read-only workloads such as snapshots.
localize: When a replicated pool receives a read request, pick the most local OSD to serve it (since 5.8). The locality metric is calculated against the location of the client given with crush_location; a match with the lowest-valued bucket type wins. For example, an OSD in a matching rack is closer than an OSD in a matching data center, which in turn is closer than an OSD in a matching region.

This mode is safe for general use only since Octopus (i.e. after “ceph osd require-osd-release octopus”). Otherwise it should be limited to read-only workloads such as snapshots.

Examples

Mount the full file system:

mount -t ceph fs_user@.mycephfs2=/ /mnt/mycephfs

Mount only part of the namespace/file system:

mount.ceph fs_user@.mycephfs2=/some/directory/in/cephfs /mnt/mycephfs

Pass the monitor host’s IP address, optionally:

mount.ceph fs_user@.mycephfs2=/ /mnt/mycephfs -o mon_addr=192.168.0.1

Pass the port along with IP address if it’s running on a non-standard port:

mount.ceph fs_user@.mycephfs2=/ /mnt/mycephfs -o mon_addr=192.168.0.1:7000

If there are multiple monitors, pass each address separated by a /:

mount.ceph fs_user@.mycephfs2=/ /mnt/mycephfs -o mon_addr=192.168.0.1/192.168.0.2/192.168.0.3

Pass secret key for CephX user optionally:

mount.ceph fs_user@.mycephfs2=/ /mnt/mycephfs -o secret=AQATSKdNGBnwLhAAnNDKnH65FmVKpXZJVasUeQ==

Pass file containing secret key to avoid leaving secret key in shell’s command history:

mount.ceph fs_user@.mycephfs2=/ /mnt/mycephfs -o secretfile=/etc/ceph/fs_username.secret

If authentication is disabled on Ceph cluster, omit the credential related option:

mount.ceph fs_user@.mycephfs2=/ /mnt/mycephfs

To mount using the old syntax:

mount -t ceph 192.168.0.1:/ /mnt/mycephfs

Availability

mount.ceph is part of Ceph, a massively scalable, open-source, distributed storage system. Please refer to the Ceph documentation at https://docs.ceph.com for more information.

Feature Availability

The recover_session= option was added to mainline Linux kernels in v5.4. wsync and nowsync were added in v5.7.