xref: /linux/Documentation/filesystems/ceph.rst (revision d27bb0246e5356dbef4d923e72c680bf893885a8)
1.. SPDX-License-Identifier: GPL-2.0
2
3============================
4Ceph Distributed File System
5============================
6
7Ceph is a distributed network file system designed to provide good
8performance, reliability, and scalability.
9
10Basic features include:
11
12 * POSIX semantics
13 * Seamless scaling from 1 to many thousands of nodes
14 * High availability and reliability.  No single point of failure.
15 * N-way replication of data across storage nodes
16 * Fast recovery from node failures
17 * Automatic rebalancing of data on node addition/removal
18 * Easy deployment: most FS components are userspace daemons
19
20Also,
21
22 * Flexible snapshots (on any directory)
23 * Recursive accounting (nested files, directories, bytes)
24
25In contrast to cluster filesystems like GFS, OCFS2, and GPFS that rely
26on symmetric access by all clients to shared block devices, Ceph
27separates data and metadata management into independent server
28clusters, similar to Lustre.  Unlike Lustre, however, metadata and
29storage nodes run entirely as user space daemons.  File data is striped
30across storage nodes in large chunks to distribute workload and
31facilitate high throughputs.  When storage nodes fail, data is
32re-replicated in a distributed fashion by the storage nodes themselves
33(with some minimal coordination from a cluster monitor), making the
34system extremely efficient and scalable.
35
36Metadata servers effectively form a large, consistent, distributed
37in-memory cache above the file namespace that is extremely scalable,
38dynamically redistributes metadata in response to workload changes,
39and can tolerate arbitrary (well, non-Byzantine) node failures.  The
40metadata server takes a somewhat unconventional approach to metadata
41storage to significantly improve performance for common workloads.  In
42particular, inodes with only a single link are embedded in
43directories, allowing entire directories of dentries and inodes to be
44loaded into its cache with a single I/O operation.  The contents of
45extremely large directories can be fragmented and managed by
46independent metadata servers, allowing scalable concurrent access.
47
48The system offers automatic data rebalancing/migration when scaling
49from a small cluster of just a few nodes to many hundreds, without
50requiring an administrator carve the data set into static volumes or
51go through the tedious process of migrating data between servers.
52When the file system approaches full, new nodes can be easily added
53and things will "just work."
54
55Ceph includes flexible snapshot mechanism that allows a user to create
56a snapshot on any subdirectory (and its nested contents) in the
57system.  Snapshot creation and deletion are as simple as 'mkdir
58.snap/foo' and 'rmdir .snap/foo'.
59
60Snapshot names have two limitations:
61
62* They can not start with an underscore ('_'), as these names are reserved
63  for internal usage by the MDS.
64* They can not exceed 240 characters in size.  This is because the MDS makes
65  use of long snapshot names internally, which follow the format:
66  `_<SNAPSHOT-NAME>_<INODE-NUMBER>`.  Since filenames in general can't have
67  more than 255 characters, and `<node-id>` takes 13 characters, the long
68  snapshot names can take as much as 255 - 1 - 1 - 13 = 240.
69
70Ceph also provides some recursive accounting on directories for nested
71files and bytes.  That is, a 'getfattr -d foo' on any directory in the
72system will reveal the total number of nested regular files and
73subdirectories, and a summation of all nested file sizes.  This makes
74the identification of large disk space consumers relatively quick, as
75no 'du' or similar recursive scan of the file system is required.
76
77Finally, Ceph also allows quotas to be set on any directory in the system.
78The quota can restrict the number of bytes or the number of files stored
79beneath that point in the directory hierarchy.  Quotas can be set using
80extended attributes 'ceph.quota.max_files' and 'ceph.quota.max_bytes', eg::
81
82 setfattr -n ceph.quota.max_bytes -v 100000000 /some/dir
83 getfattr -n ceph.quota.max_bytes /some/dir
84
85A limitation of the current quotas implementation is that it relies on the
86cooperation of the client mounting the file system to stop writers when a
87limit is reached.  A modified or adversarial client cannot be prevented
88from writing as much data as it needs.
89
90Mount Syntax
91============
92
93The basic mount syntax is::
94
95 # mount -t ceph user@fsid.fs_name=/[subdir] mnt -o mon_addr=monip1[:port][/monip2[:port]]
96
97You only need to specify a single monitor, as the client will get the
98full list when it connects.  (However, if the monitor you specify
99happens to be down, the mount won't succeed.)  The port can be left
100off if the monitor is using the default.  So if the monitor is at
1011.2.3.4::
102
103 # mount -t ceph cephuser@07fe3187-00d9-42a3-814b-72a4d5e7d5be.cephfs=/ /mnt/ceph -o mon_addr=1.2.3.4
104
105is sufficient.  If /sbin/mount.ceph is installed, a hostname can be
106used instead of an IP address and the cluster FSID can be left out
107(as the mount helper will fill it in by reading the ceph configuration
108file)::
109
110  # mount -t ceph cephuser@cephfs=/ /mnt/ceph -o mon_addr=mon-addr
111
112Multiple monitor addresses can be passed by separating each address with a slash (`/`)::
113
114  # mount -t ceph cephuser@cephfs=/ /mnt/ceph -o mon_addr=192.168.1.100/192.168.1.101
115
116When using the mount helper, monitor address can be read from ceph
117configuration file if available. Note that, the cluster FSID (passed as part
118of the device string) is validated by checking it with the FSID reported by
119the monitor.
120
121Mount Options
122=============
123
124  mon_addr=ip_address[:port][/ip_address[:port]]
125	Monitor address to the cluster. This is used to bootstrap the
126        connection to the cluster. Once connection is established, the
127        monitor addresses in the monitor map are followed.
128
129  fsid=cluster-id
130	FSID of the cluster (from `ceph fsid` command).
131
132  ip=A.B.C.D[:N]
133	Specify the IP and/or port the client should bind to locally.
134	There is normally not much reason to do this.  If the IP is not
135	specified, the client's IP address is determined by looking at the
136	address its connection to the monitor originates from.
137
138  wsize=X
139	Specify the maximum write size in bytes.  Default: 64 MB.
140
141  rsize=X
142	Specify the maximum read size in bytes.  Default: 64 MB.
143
144  rasize=X
145	Specify the maximum readahead size in bytes.  Default: 8 MB.
146
147  mount_timeout=X
148	Specify the timeout value for mount (in seconds), in the case
149	of a non-responsive Ceph file system.  The default is 60
150	seconds.
151
152  caps_max=X
153	Specify the maximum number of caps to hold. Unused caps are released
154	when number of caps exceeds the limit. The default is 0 (no limit)
155
156  rbytes
157	When stat() is called on a directory, set st_size to 'rbytes',
158	the summation of file sizes over all files nested beneath that
159	directory.  This is the default.
160
161  norbytes
162	When stat() is called on a directory, set st_size to the
163	number of entries in that directory.
164
165  nocrc
166	Disable CRC32C calculation for data writes.  If set, the storage node
167	must rely on TCP's error correction to detect data corruption
168	in the data payload.
169
170  dcache
171        Use the dcache contents to perform negative lookups and
172        readdir when the client has the entire directory contents in
173        its cache.  (This does not change correctness; the client uses
174        cached metadata only when a lease or capability ensures it is
175        valid.)
176
177  nodcache
178        Do not use the dcache as above.  This avoids a significant amount of
179        complex code, sacrificing performance without affecting correctness,
180        and is useful for tracking down bugs.
181
182  noasyncreaddir
183	Do not use the dcache as above for readdir.
184
185  noquotadf
186        Report overall filesystem usage in statfs instead of using the root
187        directory quota.
188
189  nocopyfrom
190        Don't use the RADOS 'copy-from' operation to perform remote object
191        copies.  Currently, it's only used in copy_file_range, which will revert
192        to the default VFS implementation if this option is used.
193
194  recover_session=<no|clean>
195	Set auto reconnect mode in the case where the client is blocklisted. The
196	available modes are "no" and "clean". The default is "no".
197
198	* no: never attempt to reconnect when client detects that it has been
199	  blocklisted. Operations will generally fail after being blocklisted.
200
201	* clean: client reconnects to the ceph cluster automatically when it
202	  detects that it has been blocklisted. During reconnect, client drops
203	  dirty data/metadata, invalidates page caches and writable file handles.
204	  After reconnect, file locks become stale because the MDS loses track
205	  of them. If an inode contains any stale file locks, read/write on the
206	  inode is not allowed until applications release all stale file locks.
207
208More Information
209================
210
211For more information on Ceph, see the home page at
212	https://ceph.com/
213
214The Linux kernel client source tree is available at
215	- https://github.com/ceph/ceph-client.git
216
217and the source for the full system is at
218	https://github.com/ceph/ceph.git
219