tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1
2The SGI XFS Filesystem
3======================
4
5XFS is a high performance journaling filesystem which originated
6on the SGI IRIX platform. It is completely multi-threaded, can
7support large files and large filesystems, extended attributes,
8variable block sizes, is extent based, and makes extensive use of
9Btrees (directories, extents, free space) to aid both performance
10and scalability.
11
12Refer to the documentation at http://oss.sgi.com/projects/xfs/
13for further details. This implementation is on-disk compatible
14with the IRIX version of XFS.
15
16
17Mount Options
18=============
19
20When mounting an XFS filesystem, the following options are accepted.
21
22 allocsize=size
23 Sets the buffered I/O end-of-file preallocation size when
24 doing delayed allocation writeout (default size is 64KiB).
25 Valid values for this option are page size (typically 4KiB)
26 through to 1GiB, inclusive, in power-of-2 increments.
27
28 attr2/noattr2
29 The options enable/disable (default is disabled for backward
30 compatibility on-disk) an "opportunistic" improvement to be
31 made in the way inline extended attributes are stored on-disk.
32 When the new form is used for the first time (by setting or
33 removing extended attributes) the on-disk superblock feature
34 bit field will be updated to reflect this format being in use.
35
36 barrier
37 Enables the use of block layer write barriers for writes into
38 the journal and unwritten extent conversion. This allows for
39 drive level write caching to be enabled, for devices that
40 support write barriers.
41
42 discard
43 Issue command to let the block device reclaim space freed by the
44 filesystem. This is useful for SSD devices, thinly provisioned
45 LUNs and virtual machine images, but may have a performance
46 impact.
47
48 dmapi
49 Enable the DMAPI (Data Management API) event callouts.
50 Use with the "mtpt" option.
51
52 grpid/bsdgroups and nogrpid/sysvgroups
53 These options define what group ID a newly created file gets.
54 When grpid is set, it takes the group ID of the directory in
55 which it is created; otherwise (the default) it takes the fsgid
56 of the current process, unless the directory has the setgid bit
57 set, in which case it takes the gid from the parent directory,
58 and also gets the setgid bit set if it is a directory itself.
59
60 ihashsize=value
61 In memory inode hashes have been removed, so this option has
62 no function as of August 2007. Option is deprecated.
63
64 ikeep/noikeep
65 When ikeep is specified, XFS does not delete empty inode clusters
66 and keeps them around on disk. ikeep is the traditional XFS
67 behaviour. When noikeep is specified, empty inode clusters
68 are returned to the free space pool. The default is noikeep for
69 non-DMAPI mounts, while ikeep is the default when DMAPI is in use.
70
71 inode64
72 Indicates that XFS is allowed to create inodes at any location
73 in the filesystem, including those which will result in inode
74 numbers occupying more than 32 bits of significance. This is
75 the default allocation option. Applications which do not handle
76 inode numbers bigger than 32 bits, should use inode32 option.
77
78 inode32
79 Indicates that XFS is limited to create inodes at locations which
80 will not result in inode numbers with more than 32 bits of
81 significance. This is provided for backwards compatibility, since
82 64 bits inode numbers might cause problems for some applications
83 that cannot handle large inode numbers.
84
85 largeio/nolargeio
86 If "nolargeio" is specified, the optimal I/O reported in
87 st_blksize by stat(2) will be as small as possible to allow user
88 applications to avoid inefficient read/modify/write I/O.
89 If "largeio" specified, a filesystem that has a "swidth" specified
90 will return the "swidth" value (in bytes) in st_blksize. If the
91 filesystem does not have a "swidth" specified but does specify
92 an "allocsize" then "allocsize" (in bytes) will be returned
93 instead.
94 If neither of these two options are specified, then filesystem
95 will behave as if "nolargeio" was specified.
96
97 logbufs=value
98 Set the number of in-memory log buffers. Valid numbers range
99 from 2-8 inclusive.
100 The default value is 8 buffers for filesystems with a
101 blocksize of 64KiB, 4 buffers for filesystems with a blocksize
102 of 32KiB, 3 buffers for filesystems with a blocksize of 16KiB
103 and 2 buffers for all other configurations. Increasing the
104 number of buffers may increase performance on some workloads
105 at the cost of the memory used for the additional log buffers
106 and their associated control structures.
107
108 logbsize=value
109 Set the size of each in-memory log buffer.
110 Size may be specified in bytes, or in kilobytes with a "k" suffix.
111 Valid sizes for version 1 and version 2 logs are 16384 (16k) and
112 32768 (32k). Valid sizes for version 2 logs also include
113 65536 (64k), 131072 (128k) and 262144 (256k).
114 The default value for machines with more than 32MiB of memory
115 is 32768, machines with less memory use 16384 by default.
116
117 logdev=device and rtdev=device
118 Use an external log (metadata journal) and/or real-time device.
119 An XFS filesystem has up to three parts: a data section, a log
120 section, and a real-time section. The real-time section is
121 optional, and the log section can be separate from the data
122 section or contained within it.
123
124 mtpt=mountpoint
125 Use with the "dmapi" option. The value specified here will be
126 included in the DMAPI mount event, and should be the path of
127 the actual mountpoint that is used.
128
129 noalign
130 Data allocations will not be aligned at stripe unit boundaries.
131
132 noatime
133 Access timestamps are not updated when a file is read.
134
135 norecovery
136 The filesystem will be mounted without running log recovery.
137 If the filesystem was not cleanly unmounted, it is likely to
138 be inconsistent when mounted in "norecovery" mode.
139 Some files or directories may not be accessible because of this.
140 Filesystems mounted "norecovery" must be mounted read-only or
141 the mount will fail.
142
143 nouuid
144 Don't check for double mounted file systems using the file system uuid.
145 This is useful to mount LVM snapshot volumes.
146
147 uquota/usrquota/uqnoenforce/quota
148 User disk quota accounting enabled, and limits (optionally)
149 enforced. Refer to xfs_quota(8) for further details.
150
151 gquota/grpquota/gqnoenforce
152 Group disk quota accounting enabled and limits (optionally)
153 enforced. Refer to xfs_quota(8) for further details.
154
155 pquota/prjquota/pqnoenforce
156 Project disk quota accounting enabled and limits (optionally)
157 enforced. Refer to xfs_quota(8) for further details.
158
159 sunit=value and swidth=value
160 Used to specify the stripe unit and width for a RAID device or
161 a stripe volume. "value" must be specified in 512-byte block
162 units.
163 If this option is not specified and the filesystem was made on
164 a stripe volume or the stripe width or unit were specified for
165 the RAID device at mkfs time, then the mount system call will
166 restore the value from the superblock. For filesystems that
167 are made directly on RAID devices, these options can be used
168 to override the information in the superblock if the underlying
169 disk layout changes after the filesystem has been created.
170 The "swidth" option is required if the "sunit" option has been
171 specified, and must be a multiple of the "sunit" value.
172
173 swalloc
174 Data allocations will be rounded up to stripe width boundaries
175 when the current end of file is being extended and the file
176 size is larger than the stripe width size.
177
178
179sysctls
180=======
181
182The following sysctls are available for the XFS filesystem:
183
184 fs.xfs.stats_clear (Min: 0 Default: 0 Max: 1)
185 Setting this to "1" clears accumulated XFS statistics
186 in /proc/fs/xfs/stat. It then immediately resets to "0".
187
188 fs.xfs.xfssyncd_centisecs (Min: 100 Default: 3000 Max: 720000)
189 The interval at which the xfssyncd thread flushes metadata
190 out to disk. This thread will flush log activity out, and
191 do some processing on unlinked inodes.
192
193 fs.xfs.xfsbufd_centisecs (Min: 50 Default: 100 Max: 3000)
194 The interval at which xfsbufd scans the dirty metadata buffers list.
195
196 fs.xfs.age_buffer_centisecs (Min: 100 Default: 1500 Max: 720000)
197 The age at which xfsbufd flushes dirty metadata buffers to disk.
198
199 fs.xfs.error_level (Min: 0 Default: 3 Max: 11)
200 A volume knob for error reporting when internal errors occur.
201 This will generate detailed messages & backtraces for filesystem
202 shutdowns, for example. Current threshold values are:
203
204 XFS_ERRLEVEL_OFF: 0
205 XFS_ERRLEVEL_LOW: 1
206 XFS_ERRLEVEL_HIGH: 5
207
208 fs.xfs.panic_mask (Min: 0 Default: 0 Max: 127)
209 Causes certain error conditions to call BUG(). Value is a bitmask;
210 AND together the tags which represent errors which should cause panics:
211
212 XFS_NO_PTAG 0
213 XFS_PTAG_IFLUSH 0x00000001
214 XFS_PTAG_LOGRES 0x00000002
215 XFS_PTAG_AILDELETE 0x00000004
216 XFS_PTAG_ERROR_REPORT 0x00000008
217 XFS_PTAG_SHUTDOWN_CORRUPT 0x00000010
218 XFS_PTAG_SHUTDOWN_IOERROR 0x00000020
219 XFS_PTAG_SHUTDOWN_LOGERROR 0x00000040
220
221 This option is intended for debugging only.
222
223 fs.xfs.irix_symlink_mode (Min: 0 Default: 0 Max: 1)
224 Controls whether symlinks are created with mode 0777 (default)
225 or whether their mode is affected by the umask (irix mode).
226
227 fs.xfs.irix_sgid_inherit (Min: 0 Default: 0 Max: 1)
228 Controls files created in SGID directories.
229 If the group ID of the new file does not match the effective group
230 ID or one of the supplementary group IDs of the parent dir, the
231 ISGID bit is cleared if the irix_sgid_inherit compatibility sysctl
232 is set.
233
234 fs.xfs.inherit_sync (Min: 0 Default: 1 Max: 1)
235 Setting this to "1" will cause the "sync" flag set
236 by the xfs_io(8) chattr command on a directory to be
237 inherited by files in that directory.
238
239 fs.xfs.inherit_nodump (Min: 0 Default: 1 Max: 1)
240 Setting this to "1" will cause the "nodump" flag set
241 by the xfs_io(8) chattr command on a directory to be
242 inherited by files in that directory.
243
244 fs.xfs.inherit_noatime (Min: 0 Default: 1 Max: 1)
245 Setting this to "1" will cause the "noatime" flag set
246 by the xfs_io(8) chattr command on a directory to be
247 inherited by files in that directory.
248
249 fs.xfs.inherit_nosymlinks (Min: 0 Default: 1 Max: 1)
250 Setting this to "1" will cause the "nosymlinks" flag set
251 by the xfs_io(8) chattr command on a directory to be
252 inherited by files in that directory.
253
254 fs.xfs.rotorstep (Min: 1 Default: 1 Max: 256)
255 In "inode32" allocation mode, this option determines how many
256 files the allocator attempts to allocate in the same allocation
257 group before moving to the next allocation group. The intent
258 is to control the rate at which the allocator moves between
259 allocation groups when allocating extents for new files.