Documentation/filesystems/xfs.txt

   1
   2 The SGI XFS Filesystem
   3 ======================
   4
   5 XFS is a high performance journaling filesystem which originated
   6 on the SGI IRIX platform.  It is completely multi-threaded, can
   7 support large files and large filesystems, extended attributes,
   8 variable block sizes, is extent based, and makes extensive use of
   9 Btrees (directories, extents, free space) to aid both performance
  10 and scalability.
  11
  12 Refer to the documentation at http://oss.sgi.com/projects/xfs/
  13 for further details.  This implementation is on-disk compatible
  14 with the IRIX version of XFS.
  15
  16
  17 Mount Options
  18 =============
  19
  20 When mounting an XFS filesystem, the following options are accepted.
  21 For boolean mount options, the names with the (*) suffix is the
  22 default behaviour.
  23
  24   allocsize=size
  25         Sets the buffered I/O end-of-file preallocation size when
  26         doing delayed allocation writeout (default size is 64KiB).
  27         Valid values for this option are page size (typically 4KiB)
  28         through to 1GiB, inclusive, in power-of-2 increments.
  29
  30         The default behaviour is for dynamic end-of-file
  31         preallocation size, which uses a set of heuristics to
  32         optimise the preallocation size based on the current
  33         allocation patterns within the file and the access patterns
  34         to the file. Specifying a fixed allocsize value turns off
  35         the dynamic behaviour.
  36
  37   attr2
  38   noattr2
  39         The options enable/disable an "opportunistic" improvement to
  40         be made in the way inline extended attributes are stored
  41         on-disk.  When the new form is used for the first time when
  42         attr2 is selected (either when setting or removing extended
  43         attributes) the on-disk superblock feature bit field will be
  44         updated to reflect this format being in use.
  45
  46         The default behaviour is determined by the on-disk feature
  47         bit indicating that attr2 behaviour is active. If either
  48         mount option it set, then that becomes the new default used
  49         by the filesystem.
  50
  51         CRC enabled filesystems always use the attr2 format, and so
  52         will reject the noattr2 mount option if it is set.
  53
  54   discard
  55   nodiscard (*)
  56         Enable/disable the issuing of commands to let the block
  57         device reclaim space freed by the filesystem.  This is
  58         useful for SSD devices, thinly provisioned LUNs and virtual
  59         machine images, but may have a performance impact.
  60
  61         Note: It is currently recommended that you use the fstrim
  62         application to discard unused blocks rather than the discard
  63         mount option because the performance impact of this option
  64         is quite severe.
  65
  66   grpid/bsdgroups
  67   nogrpid/sysvgroups (*)
  68         These options define what group ID a newly created file
  69         gets.  When grpid is set, it takes the group ID of the
  70         directory in which it is created; otherwise it takes the
  71         fsgid of the current process, unless the directory has the
  72         setgid bit set, in which case it takes the gid from the
  73         parent directory, and also gets the setgid bit set if it is
  74         a directory itself.
  75
  76   filestreams
  77         Make the data allocator use the filestreams allocation mode
  78         across the entire filesystem rather than just on directories
  79         configured to use it.
  80
  81   ikeep
  82   noikeep (*)
  83         When ikeep is specified, XFS does not delete empty inode
  84         clusters and keeps them around on disk.  When noikeep is
  85         specified, empty inode clusters are returned to the free
  86         space pool.
  87
  88   inode32
  89   inode64 (*)
  90         When inode32 is specified, it indicates that XFS limits
  91         inode creation to locations which will not result in inode
  92         numbers with more than 32 bits of significance.
  93
  94         When inode64 is specified, it indicates that XFS is allowed
  95         to create inodes at any location in the filesystem,
  96         including those which will result in inode numbers occupying
  97         more than 32 bits of significance.
  98
  99         inode32 is provided for backwards compatibility with older
 100         systems and applications, since 64 bits inode numbers might
 101         cause problems for some applications that cannot handle
 102         large inode numbers.  If applications are in use which do
 103         not handle inode numbers bigger than 32 bits, the inode32
 104         option should be specified.
 105
 106
 107   largeio
 108   nolargeio (*)
 109         If "nolargeio" is specified, the optimal I/O reported in
 110         st_blksize by stat(2) will be as small as possible to allow
 111         user applications to avoid inefficient read/modify/write
 112         I/O.  This is typically the page size of the machine, as
 113         this is the granularity of the page cache.
 114
 115         If "largeio" specified, a filesystem that was created with a
 116         "swidth" specified will return the "swidth" value (in bytes)
 117         in st_blksize. If the filesystem does not have a "swidth"
 118         specified but does specify an "allocsize" then "allocsize"
 119         (in bytes) will be returned instead. Otherwise the behaviour
 120         is the same as if "nolargeio" was specified.
 121
 122   logbufs=value
 123         Set the number of in-memory log buffers.  Valid numbers
 124         range from 2-8 inclusive.
 125
 126         The default value is 8 buffers.
 127
 128         If the memory cost of 8 log buffers is too high on small
 129         systems, then it may be reduced at some cost to performance
 130         on metadata intensive workloads. The logbsize option below
 131         controls the size of each buffer and so is also relevant to
 132         this case.
 133
 134   logbsize=value
 135         Set the size of each in-memory log buffer.  The size may be
 136         specified in bytes, or in kilobytes with a "k" suffix.
 137         Valid sizes for version 1 and version 2 logs are 16384 (16k)
 138         and 32768 (32k).  Valid sizes for version 2 logs also
 139         include 65536 (64k), 131072 (128k) and 262144 (256k). The
 140         logbsize must be an integer multiple of the log
 141         stripe unit configured at mkfs time.
 142
 143         The default value for for version 1 logs is 32768, while the
 144         default value for version 2 logs is MAX(32768, log_sunit).
 145
 146   logdev=device and rtdev=device
 147         Use an external log (metadata journal) and/or real-time device.
 148         An XFS filesystem has up to three parts: a data section, a log
 149         section, and a real-time section.  The real-time section is
 150         optional, and the log section can be separate from the data
 151         section or contained within it.
 152
 153   noalign
 154         Data allocations will not be aligned at stripe unit
 155         boundaries. This is only relevant to filesystems created
 156         with non-zero data alignment parameters (sunit, swidth) by
 157         mkfs.
 158
 159   norecovery
 160         The filesystem will be mounted without running log recovery.
 161         If the filesystem was not cleanly unmounted, it is likely to
 162         be inconsistent when mounted in "norecovery" mode.
 163         Some files or directories may not be accessible because of this.
 164         Filesystems mounted "norecovery" must be mounted read-only or
 165         the mount will fail.
 166
 167   nouuid
 168         Don't check for double mounted file systems using the file
 169         system uuid.  This is useful to mount LVM snapshot volumes,
 170         and often used in combination with "norecovery" for mounting
 171         read-only snapshots.
 172
 173   noquota
 174         Forcibly turns off all quota accounting and enforcement
 175         within the filesystem.
 176
 177   uquota/usrquota/uqnoenforce/quota
 178         User disk quota accounting enabled, and limits (optionally)
 179         enforced.  Refer to xfs_quota(8) for further details.
 180
 181   gquota/grpquota/gqnoenforce
 182         Group disk quota accounting enabled and limits (optionally)
 183         enforced.  Refer to xfs_quota(8) for further details.
 184
 185   pquota/prjquota/pqnoenforce
 186         Project disk quota accounting enabled and limits (optionally)
 187         enforced.  Refer to xfs_quota(8) for further details.
 188
 189   sunit=value and swidth=value
 190         Used to specify the stripe unit and width for a RAID device
 191         or a stripe volume.  "value" must be specified in 512-byte
 192         block units. These options are only relevant to filesystems
 193         that were created with non-zero data alignment parameters.
 194
 195         The sunit and swidth parameters specified must be compatible
 196         with the existing filesystem alignment characteristics.  In
 197         general, that means the only valid changes to sunit are
 198         increasing it by a power-of-2 multiple. Valid swidth values
 199         are any integer multiple of a valid sunit value.
 200
 201         Typically the only time these mount options are necessary if
 202         after an underlying RAID device has had it's geometry
 203         modified, such as adding a new disk to a RAID5 lun and
 204         reshaping it.
 205
 206   swalloc
 207         Data allocations will be rounded up to stripe width boundaries
 208         when the current end of file is being extended and the file
 209         size is larger than the stripe width size.
 210
 211   wsync
 212         When specified, all filesystem namespace operations are
 213         executed synchronously. This ensures that when the namespace
 214         operation (create, unlink, etc) completes, the change to the
 215         namespace is on stable storage. This is useful in HA setups
 216         where failover must not result in clients seeing
 217         inconsistent namespace presentation during or after a
 218         failover event.
 219
 220
 221 Deprecated Mount Options
 222 ========================
 223
 224   Name                          Removal Schedule
 225   ----                          ----------------
 226   barrier                       no earlier than v4.15
 227   nobarrier                     no earlier than v4.15
 228
 229
 230 Removed Mount Options
 231 =====================
 232
 233   Name                          Removed
 234   ----                          -------
 235   delaylog/nodelaylog           v4.0
 236   ihashsize                     v4.0
 237   irixsgid                      v4.0
 238   osyncisdsync/osyncisosync     v4.0
 239
 240
 241 sysctls
 242 =======
 243
 244 The following sysctls are available for the XFS filesystem:
 245
 246   fs.xfs.stats_clear            (Min: 0  Default: 0  Max: 1)
 247         Setting this to "1" clears accumulated XFS statistics
 248         in /proc/fs/xfs/stat.  It then immediately resets to "0".
 249
 250   fs.xfs.xfssyncd_centisecs     (Min: 100  Default: 3000  Max: 720000)
 251         The interval at which the filesystem flushes metadata
 252         out to disk and runs internal cache cleanup routines.
 253
 254   fs.xfs.filestream_centisecs   (Min: 1  Default: 3000  Max: 360000)
 255         The interval at which the filesystem ages filestreams cache
 256         references and returns timed-out AGs back to the free stream
 257         pool.
 258
 259   fs.xfs.speculative_prealloc_lifetime
 260                 (Units: seconds   Min: 1  Default: 300  Max: 86400)
 261         The interval at which the background scanning for inodes
 262         with unused speculative preallocation runs. The scan
 263         removes unused preallocation from clean inodes and releases
 264         the unused space back to the free pool.
 265
 266   fs.xfs.error_level            (Min: 0  Default: 3  Max: 11)
 267         A volume knob for error reporting when internal errors occur.
 268         This will generate detailed messages & backtraces for filesystem
 269         shutdowns, for example.  Current threshold values are:
 270
 271                 XFS_ERRLEVEL_OFF:       0
 272                 XFS_ERRLEVEL_LOW:       1
 273                 XFS_ERRLEVEL_HIGH:      5
 274
 275   fs.xfs.panic_mask             (Min: 0  Default: 0  Max: 255)
 276         Causes certain error conditions to call BUG(). Value is a bitmask;
 277         OR together the tags which represent errors which should cause panics:
 278
 279                 XFS_NO_PTAG                     0
 280                 XFS_PTAG_IFLUSH                 0x00000001
 281                 XFS_PTAG_LOGRES                 0x00000002
 282                 XFS_PTAG_AILDELETE              0x00000004
 283                 XFS_PTAG_ERROR_REPORT           0x00000008
 284                 XFS_PTAG_SHUTDOWN_CORRUPT       0x00000010
 285                 XFS_PTAG_SHUTDOWN_IOERROR       0x00000020
 286                 XFS_PTAG_SHUTDOWN_LOGERROR      0x00000040
 287                 XFS_PTAG_FSBLOCK_ZERO           0x00000080
 288
 289         This option is intended for debugging only.
 290
 291   fs.xfs.irix_symlink_mode      (Min: 0  Default: 0  Max: 1)
 292         Controls whether symlinks are created with mode 0777 (default)
 293         or whether their mode is affected by the umask (irix mode).
 294
 295   fs.xfs.irix_sgid_inherit      (Min: 0  Default: 0  Max: 1)
 296         Controls files created in SGID directories.
 297         If the group ID of the new file does not match the effective group
 298         ID or one of the supplementary group IDs of the parent dir, the
 299         ISGID bit is cleared if the irix_sgid_inherit compatibility sysctl
 300         is set.
 301
 302   fs.xfs.inherit_sync           (Min: 0  Default: 1  Max: 1)
 303         Setting this to "1" will cause the "sync" flag set
 304         by the xfs_io(8) chattr command on a directory to be
 305         inherited by files in that directory.
 306
 307   fs.xfs.inherit_nodump         (Min: 0  Default: 1  Max: 1)
 308         Setting this to "1" will cause the "nodump" flag set
 309         by the xfs_io(8) chattr command on a directory to be
 310         inherited by files in that directory.
 311
 312   fs.xfs.inherit_noatime        (Min: 0  Default: 1  Max: 1)
 313         Setting this to "1" will cause the "noatime" flag set
 314         by the xfs_io(8) chattr command on a directory to be
 315         inherited by files in that directory.
 316
 317   fs.xfs.inherit_nosymlinks     (Min: 0  Default: 1  Max: 1)
 318         Setting this to "1" will cause the "nosymlinks" flag set
 319         by the xfs_io(8) chattr command on a directory to be
 320         inherited by files in that directory.
 321
 322   fs.xfs.inherit_nodefrag       (Min: 0  Default: 1  Max: 1)
 323         Setting this to "1" will cause the "nodefrag" flag set
 324         by the xfs_io(8) chattr command on a directory to be
 325         inherited by files in that directory.
 326
 327   fs.xfs.rotorstep              (Min: 1  Default: 1  Max: 256)
 328         In "inode32" allocation mode, this option determines how many
 329         files the allocator attempts to allocate in the same allocation
 330         group before moving to the next allocation group.  The intent
 331         is to control the rate at which the allocator moves between
 332         allocation groups when allocating extents for new files.
 333
 334 Deprecated Sysctls
 335 ==================
 336
 337 None at present.
 338
 339
 340 Removed Sysctls
 341 ===============
 342
 343   Name                          Removed
 344   ----                          -------
 345   fs.xfs.xfsbufd_centisec       v4.0
 346   fs.xfs.age_buffer_centisecs   v4.0
 347
 348
 349 Error handling
 350 ==============
 351
 352 XFS can act differently according to the type of error found during its
 353 operation. The implementation introduces the following concepts to the error
 354 handler:
 355
 356  -failure speed:
 357         Defines how fast XFS should propagate an error upwards when a specific
 358         error is found during the filesystem operation. It can propagate
 359         immediately, after a defined number of retries, after a set time period,
 360         or simply retry forever.
 361
 362  -error classes:
 363         Specifies the subsystem the error configuration will apply to, such as
 364         metadata IO or memory allocation. Different subsystems will have
 365         different error handlers for which behaviour can be configured.
 366
 367  -error handlers:
 368         Defines the behavior for a specific error.
 369
 370 The filesystem behavior during an error can be set via sysfs files. Each
 371 error handler works independently - the first condition met by an error handler
 372 for a specific class will cause the error to be propagated rather than reset and
 373 retried.
 374
 375 The action taken by the filesystem when the error is propagated is context
 376 dependent - it may cause a shut down in the case of an unrecoverable error,
 377 it may be reported back to userspace, or it may even be ignored because
 378 there's nothing useful we can with the error or anyone we can report it to (e.g.
 379 during unmount).
 380
 381 The configuration files are organized into the following hierarchy for each
 382 mounted filesystem:
 383
 384   /sys/fs/xfs/<dev>/error/<class>/<error>/
 385
 386 Where:
 387   <dev>
 388         The short device name of the mounted filesystem. This is the same device
 389         name that shows up in XFS kernel error messages as "XFS(<dev>): ..."
 390
 391   <class>
 392         The subsystem the error configuration belongs to. As of 4.9, the defined
 393         classes are:
 394
 395                 - "metadata": applies metadata buffer write IO
 396
 397   <error>
 398         The individual error handler configurations.
 399
 400
 401 Each filesystem has "global" error configuration options defined in their top
 402 level directory:
 403
 404   /sys/fs/xfs/<dev>/error/
 405
 406   fail_at_unmount               (Min:  0  Default:  1  Max: 1)
 407         Defines the filesystem error behavior at unmount time.
 408
 409         If set to a value of 1, XFS will override all other error configurations
 410         during unmount and replace them with "immediate fail" characteristics.
 411         i.e. no retries, no retry timeout. This will always allow unmount to
 412         succeed when there are persistent errors present.
 413
 414         If set to 0, the configured retry behaviour will continue until all
 415         retries and/or timeouts have been exhausted. This will delay unmount
 416         completion when there are persistent errors, and it may prevent the
 417         filesystem from ever unmounting fully in the case of "retry forever"
 418         handler configurations.
 419
 420         Note: there is no guarantee that fail_at_unmount can be set whilst an
 421         unmount is in progress. It is possible that the sysfs entries are
 422         removed by the unmounting filesystem before a "retry forever" error
 423         handler configuration causes unmount to hang, and hence the filesystem
 424         must be configured appropriately before unmount begins to prevent
 425         unmount hangs.
 426
 427 Each filesystem has specific error class handlers that define the error
 428 propagation behaviour for specific errors. There is also a "default" error
 429 handler defined, which defines the behaviour for all errors that don't have
 430 specific handlers defined. Where multiple retry constraints are configuredi for
 431 a single error, the first retry configuration that expires will cause the error
 432 to be propagated. The handler configurations are found in the directory:
 433
 434   /sys/fs/xfs/<dev>/error/<class>/<error>/
 435
 436   max_retries                   (Min: -1  Default: Varies  Max: INTMAX)
 437         Defines the allowed number of retries of a specific error before
 438         the filesystem will propagate the error. The retry count for a given
 439         error context (e.g. a specific metadata buffer) is reset every time
 440         there is a successful completion of the operation.
 441
 442         Setting the value to "-1" will cause XFS to retry forever for this
 443         specific error.
 444
 445         Setting the value to "0" will cause XFS to fail immediately when the
 446         specific error is reported.
 447
 448         Setting the value to "N" (where 0 < N < Max) will make XFS retry the
 449         operation "N" times before propagating the error.
 450
 451   retry_timeout_seconds         (Min:  -1  Default:  Varies  Max: 1 day)
 452         Define the amount of time (in seconds) that the filesystem is
 453         allowed to retry its operations when the specific error is
 454         found.
 455
 456         Setting the value to "-1" will allow XFS to retry forever for this
 457         specific error.
 458
 459         Setting the value to "0" will cause XFS to fail immediately when the
 460         specific error is reported.
 461
 462         Setting the value to "N" (where 0 < N < Max) will allow XFS to retry the
 463         operation for up to "N" seconds before propagating the error.
 464
 465 Note: The default behaviour for a specific error handler is dependent on both
 466 the class and error context. For example, the default values for
 467 "metadata/ENODEV" are "0" rather than "-1" so that this error handler defaults
 468 to "fail immediately" behaviour. This is done because ENODEV is a fatal,
 469 unrecoverable error no matter how many times the metadata IO is retried.