md.txt 24.1 KB
Newer Older
L
Linus Torvalds 已提交
1
Tools that manage md devices can be found at
2
   http://www.kernel.org/pub/linux/utils/raid/ 
L
Linus Torvalds 已提交
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53


Boot time assembly of RAID arrays
---------------------------------

You can boot with your md device with the following kernel command
lines:

for old raid arrays without persistent superblocks:
  md=<md device no.>,<raid level>,<chunk size factor>,<fault level>,dev0,dev1,...,devn

for raid arrays with persistent superblocks
  md=<md device no.>,dev0,dev1,...,devn
or, to assemble a partitionable array:
  md=d<md device no.>,dev0,dev1,...,devn
  
md device no. = the number of the md device ... 
              0 means md0, 
	      1 md1,
	      2 md2,
	      3 md3,
	      4 md4

raid level = -1 linear mode
              0 striped mode
	      other modes are only supported with persistent super blocks

chunk size factor = (raid-0 and raid-1 only)
              Set  the chunk size as 4k << n.
	      
fault level = totally ignored
			    
dev0-devn: e.g. /dev/hda1,/dev/hdc1,/dev/sda1,/dev/sdb1
			    
A possible loadlin line (Harald Hoyer <HarryH@Royal.Net>)  looks like this:

e:\loadlin\loadlin e:\zimage root=/dev/md0 md=0,0,4,0,/dev/hdb2,/dev/hdc3 ro


Boot time autodetection of RAID arrays
--------------------------------------

When md is compiled into the kernel (not as module), partitions of
type 0xfd are scanned and automatically assembled into RAID arrays.
This autodetection may be suppressed with the kernel parameter
"raid=noautodetect".  As of kernel 2.6.9, only drives with a type 0
superblock can be autodetected and run at boot time.

The kernel parameter "raid=partitionable" (or "raid=part") means
that all auto-detected arrays are assembled as partitionable.

54 55 56 57 58 59 60 61 62 63 64
Boot time assembly of degraded/dirty arrays
-------------------------------------------

If a raid5 or raid6 array is both dirty and degraded, it could have
undetectable data corruption.  This is because the fact that it is
'dirty' means that the parity cannot be trusted, and the fact that it
is degraded means that some datablocks are missing and cannot reliably
be reconstructed (due to no parity).

For this reason, md will normally refuse to start such an array.  This
requires the sysadmin to take action to explicitly start the array
65
despite possible corruption.  This is normally done with
66 67 68 69 70 71 72 73 74 75 76 77
   mdadm --assemble --force ....

This option is not really available if the array has the root
filesystem on it.  In order to support this booting from such an
array, md supports a module parameter "start_dirty_degraded" which,
when set to 1, bypassed the checks and will allows dirty degraded
arrays to be started.

So, to boot with a root filesystem of a dirty degraded raid[56], use

   md-mod.start_dirty_degraded=1

L
Linus Torvalds 已提交
78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138

Superblock formats
------------------

The md driver can support a variety of different superblock formats.
Currently, it supports superblock formats "0.90.0" and the "md-1" format
introduced in the 2.5 development series.

The kernel will autodetect which format superblock is being used.

Superblock format '0' is treated differently to others for legacy
reasons - it is the original superblock format.


General Rules - apply for all superblock formats
------------------------------------------------

An array is 'created' by writing appropriate superblocks to all
devices.

It is 'assembled' by associating each of these devices with an
particular md virtual device.  Once it is completely assembled, it can
be accessed.

An array should be created by a user-space tool.  This will write
superblocks to all devices.  It will usually mark the array as
'unclean', or with some devices missing so that the kernel md driver
can create appropriate redundancy (copying in raid1, parity
calculation in raid4/5).

When an array is assembled, it is first initialized with the
SET_ARRAY_INFO ioctl.  This contains, in particular, a major and minor
version number.  The major version number selects which superblock
format is to be used.  The minor number might be used to tune handling
of the format, such as suggesting where on each device to look for the
superblock.

Then each device is added using the ADD_NEW_DISK ioctl.  This
provides, in particular, a major and minor number identifying the
device to add.

The array is started with the RUN_ARRAY ioctl.

Once started, new devices can be added.  They should have an
appropriate superblock written to them, and then passed be in with
ADD_NEW_DISK.

Devices that have failed or are not yet active can be detached from an
array using HOT_REMOVE_DISK.


Specific Rules that apply to format-0 super block arrays, and
       arrays with no superblock (non-persistent).
-------------------------------------------------------------

An array can be 'created' by describing the array (level, chunksize
etc) in a SET_ARRAY_INFO ioctl.  This must has major_version==0 and
raid_disks != 0.

Then uninitialized devices can be added with ADD_NEW_DISK.  The
structure passed to ADD_NEW_DISK must specify the state of the device
139
and its role in the array.
L
Linus Torvalds 已提交
140 141 142

Once started with RUN_ARRAY, uninitialized spares can be added with
HOT_ADD_DISK.
143 144 145 146 147 148 149 150 151 152 153 154 155 156



MD devices in sysfs
-------------------
md devices appear in sysfs (/sys) as regular block devices,
e.g.
   /sys/block/md0

Each 'md' device will contain a subdirectory called 'md' which
contains further md-specific information about the device.

All md devices contain:
  level
157 158
     a text file indicating the 'raid level'. e.g. raid0, raid1,
     raid5, linear, multipath, faulty.
159
     If no raid level has been set yet (array is still being
160 161 162
     assembled), the value will reflect whatever has been written
     to it, which may be a name like the above, or may be a number
     such as '0', '5', etc.
163 164 165 166

  raid_disks
     a text file with a simple number indicating the number of devices
     in a fully functional array.  If this is not yet known, the file
N
NeilBrown 已提交
167 168 169 170 171 172 173 174 175
     will be empty.  If an array is being resized this will contain
     the new number of devices.
     Some raid levels allow this value to be set while the array is
     active.  This will reconfigure the array.   Otherwise it can only
     be set while assembling an array.
     A change to this attribute will not be permitted if it would
     reduce the size of the array.  To reduce the number of drives
     in an e.g. raid5, the array size must first be reduced by
     setting the 'array_size' attribute.
176

177
  chunk_size
N
NeilBrown 已提交
178 179
     This is the size in bytes for 'chunks' and is only relevant to
     raid levels that involve striping (0,4,5,6,10). The address space
180 181
     of the array is conceptually divided into chunks and consecutive
     chunks are striped onto neighbouring devices.
182
     The size should be at least PAGE_SIZE (4k) and should be a power
183 184
     of 2.  This can only be set while assembling an array

185 186 187 188 189
  layout
     The "layout" for the array for the particular level.  This is
     simply a number that is interpretted differently by different
     levels.  It can be written while assembling an array.

N
NeilBrown 已提交
190 191 192 193 194 195 196 197 198 199 200 201 202 203
  array_size
     This can be used to artificially constrain the available space in
     the array to be less than is actually available on the combined
     devices.  Writing a number (in Kilobytes) which is less than
     the available size will set the size.  Any reconfiguration of the
     array (e.g. adding devices) will not cause the size to change.
     Writing the word 'default' will cause the effective size of the
     array to be whatever size is actually available based on
     'level', 'chunk_size' and 'component_size'.

     This can be used to reduce the size of the array before reducing
     the number of devices in a raid4/5/6, or to support external
     metadata formats which mandate such clipping.

204 205 206 207 208 209 210 211 212 213
  reshape_position
     This is either "none" or a sector number within the devices of
     the array where "reshape" is up to.  If this is set, the three
     attributes mentioned above (raid_disks, chunk_size, layout) can
     potentially have 2 values, an old and a new value.  If these
     values differ, reading the attribute returns
        new (old)
     and writing will effect the 'new' value, leaving the 'old'
     unchanged.

214 215 216 217 218 219 220 221 222
  component_size
     For arrays with data redundancy (i.e. not raid0, linear, faulty,
     multipath), all components must be the same size - or at least
     there must a size that they all provide space for.  This is a key
     part or the geometry of the array.  It is measured in sectors
     and can be read from here.  Writing to this value may resize
     the array if the personality supports it (raid1, raid5, raid6),
     and if the component drives are large enough.

223 224 225 226 227
  metadata_version
     This indicates the format that is being used to record metadata
     about the array.  It can be 0.90 (traditional format), 1.0, 1.1,
     1.2 (newer format in varying locations) or "none" indicating that
     the kernel isn't managing metadata at all.
N
NeilBrown 已提交
228 229 230 231 232
     Alternately it can be "external:" followed by a string which
     is set by user-space.  This indicates that metadata is managed
     by a user-space program.  Any device failure or other event that
     requires a metadata update will cause array activity to be
     suspended until the event is acknowledged.
233

234 235
  resync_start
     The point at which resync should start.  If no resync is needed,
236 237 238
     this will be a very large number (or 'none' since 2.6.30-rc1).  At
     array creation it will default to 0, though starting the array as
     'clean' will set it much larger.
239

240 241 242 243 244 245 246
   new_dev
     This file can be written but not read.  The value written should
     be a block device number as major:minor.  e.g. 8:0
     This will cause that device to be attached to the array, if it is
     available.  It will then appear at md/dev-XXX (depending on the
     name of the device) and further configuration is then possible.

247 248 249
   safe_mode_delay
     When an md array has seen no write requests for a certain period
     of time, it will be marked as 'clean'.  When another write
250 251
     request arrives, the array is marked as 'dirty' before the write
     commences.  This is known as 'safe_mode'.
252 253 254 255
     The 'certain period' is controlled by this file which stores the
     period as a number of seconds.  The default is 200msec (0.200).
     Writing a value of 0 disables safemode.

256 257 258 259 260 261
   array_state
     This file contains a single word which describes the current
     state of the array.  In many cases, the state can be set by
     writing the word for the desired state, however some states
     cannot be explicitly set, and some transitions are not allowed.

262 263 264 265 266
     Select/poll works on this file.  All changes except between
     	active_idle and active (which can be frequent and are not
	very interesting) are notified.  active->active_idle is
	reported if the metadata is externally managed.

267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298
     clear
         No devices, no size, no level
         Writing is equivalent to STOP_ARRAY ioctl
     inactive
         May have some settings, but array is not active
            all IO results in error
         When written, doesn't tear down array, but just stops it
     suspended (not supported yet)
         All IO requests will block. The array can be reconfigured.
         Writing this, if accepted, will block until array is quiessent
     readonly
         no resync can happen.  no superblocks get written.
         write requests fail
     read-auto
         like readonly, but behaves like 'clean' on a write request.

     clean - no pending writes, but otherwise active.
         When written to inactive array, starts without resync
         If a write request arrives then
           if metadata is known, mark 'dirty' and switch to 'active'.
           if not known, block and switch to write-pending
         If written to an active array that has pending writes, then fails.
     active
         fully active: IO and resync can be happening.
         When written to inactive array, starts with resync

     write-pending
         clean, but writes are blocked waiting for 'active' to be written.

     active-idle
         like active, but no writes have been seen for a while (safe_mode_delay).

299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324
  bitmap/location
     This indicates where the write-intent bitmap for the array is
     stored.
     It can be one of "none", "file" or "[+-]N".
     "file" may later be extended to "file:/file/name"
     "[+-]N" means that many sectors from the start of the metadata.
       This is replicated on all devices.  For arrays with externally
       managed metadata, the offset is from the beginning of the
       device.
  bitmap/chunksize
     The size, in bytes, of the chunk which will be represented by a
     single bit.  For RAID456, it is a portion of an individual
     device. For RAID10, it is a portion of the array.  For RAID1, it
     is both (they come to the same thing).
  bitmap/time_base
     The time, in seconds, between looking for bits in the bitmap to
     be cleared. In the current implementation, a bit will be cleared
     between 2 and 3 times "time_base" after all the covered blocks
     are known to be in-sync.
  bitmap/backlog
     When write-mostly devices are active in a RAID1, write requests
     to those devices proceed in the background - the filesystem (or
     other user of the device) does not have to wait for them.
     'backlog' sets a limit on the number of concurrent background
     writes.  If there are more than this, new writes will by
     synchronous.
325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340
  bitmap/metadata
     This can be either 'internal' or 'external'.
     'internal' is the default and means the metadata for the bitmap
     is stored in the first 256 bytes of the allocated space and is
     managed by the md module.
     'external' means that bitmap metadata is managed externally to
     the kernel (i.e. by some userspace program)
  bitmap/can_clear
     This is either 'true' or 'false'.  If 'true', then bits in the
     bitmap will be cleared when the corresponding blocks are thought
     to be in-sync.  If 'false', bits will never be cleared.
     This is automatically set to 'false' if a write happens on a
     degraded array, or if the array becomes degraded during a write.
     When metadata is managed externally, it should be set to true
     once the array becomes non-degraded, and this fact has been
     recorded in the metadata.
341 342 343
     
     
     
344

345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364
As component devices are added to an md array, they appear in the 'md'
directory as new directories named
      dev-XXX
where XXX is a name that the kernel knows for the device, e.g. hdb1.
Each directory contains:

      block
        a symlink to the block device in /sys/block, e.g.
	     /sys/block/md0/md/dev-hdb1/block -> ../../../../block/hdb/hdb1

      super
        A file containing an image of the superblock read from, or
        written to, that device.

      state
        A file recording the current state of the device in the array
	which can be a comma separated list of
	      faulty   - device has been kicked from active use due to
                         a detected fault
	      in_sync  - device is a fully in-sync member of the array
365 366 367
	      writemostly - device will only be subject to read
		         requests if there are no other options.
			 This applies only to raid1 arrays.
368 369 370 371
	      blocked  - device has failed, metadata is "external",
	                 and the failure hasn't been acknowledged yet.
			 Writes that would write to this device if
			 it were not faulty are blocked.
372 373
	      spare    - device is working, but not a full member.
			 This includes spares that are in the process
374 375
			 of being recovered to
	This list may grow in future.
376 377 378
	This can be written to.
	Writing "faulty"  simulates a failure on the device.
	Writing "remove" removes the device from the array.
379 380
	Writing "writemostly" sets the writemostly flag.
	Writing "-writemostly" clears the writemostly flag.
381
	Writing "blocked" sets the "blocked" flag.
382
	Writing "-blocked" clears the "blocked" flag and allows writes
383
		to complete.
384
	Writing "in_sync" sets the in_sync flag.
385 386 387

	This file responds to select/poll. Any change to 'faulty'
	or 'blocked' causes an event.
388

389 390 391 392 393 394 395 396 397 398 399
      errors
	An approximate count of read errors that have been detected on
	this device but have not caused the device to be evicted from
	the array (either because they were corrected or because they
	happened while the array was read-only).  When using version-1
	metadata, this value persists across restarts of the array.

	This value can be written while assembling an array thus
	providing an ongoing count for arrays with metadata managed by
	userspace.

400 401 402 403
      slot
        This gives the role that the device has in the array.  It will
	either be 'none' if the device is not active in the array
        (i.e. is a spare or has failed) or an integer less than the
404
	'raid_disks' number for the array indicating which position
405 406 407
	it currently fills.  This can only be set while assembling an
	array.  A device for which this is set is assumed to be working.

408 409 410 411 412 413
      offset
        This gives the location in the device (in sectors from the
        start) where data from the array will be stored.  Any part of
        the device before this offset us not touched, unless it is
        used for storing metadata (Formats 1.1 and 1.2).

414 415 416 417 418
      size
        The amount of the device, after the offset, that can be used
        for storage of data.  This will normally be the same as the
	component_size.  This can be written while assembling an
        array.  If a value less than the current component_size is
N
Neil Brown 已提交
419
        written, it will be rejected.
420

421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438
      recovery_start

        When the device is not 'in_sync', this records the number of
	sectors from the start of the device which are known to be
	correct.  This is normally zero, but during a recovery
	operation is will steadily increase, and if the recovery is
	interrupted, restoring this value can cause recovery to
	avoid repeating the earlier blocks.  With v1.x metadata, this
	value is saved and restored automatically.

	This can be set whenever the device is not an active member of
	the array, either before the array is activated, or before
	the 'slot' is set.

	Setting this to 'none' is equivalent to setting 'in_sync'.
	Setting to any other value also clears the 'in_sync' flag.
	

439 440 441 442 443 444

An active md device will also contain and entry for each active device
in the array.  These are named

    rdNN

445
where 'NN' is the position in the array, starting from 0.
446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485
So for a 3 drive array there will be rd0, rd1, rd2.
These are symbolic links to the appropriate 'dev-XXX' entry.
Thus, for example,
       cat /sys/block/md*/md/rd*/state
will show 'in_sync' on every line.



Active md devices for levels that support data redundancy (1,4,5,6)
also have

   sync_action
     a text file that can be used to monitor and control the rebuild
     process.  It contains one word which can be one of:
       resync        - redundancy is being recalculated after unclean
                       shutdown or creation
       recover       - a hot spare is being built to replace a
                       failed/missing device
       idle          - nothing is happening
       check         - A full check of redundancy was requested and is
                       happening.  This reads all block and checks
                       them. A repair may also happen for some raid
                       levels.
       repair        - A full check and repair is happening.  This is
                       similar to 'resync', but was requested by the
                       user, and the write-intent bitmap is NOT used to
		       optimise the process.

      This file is writable, and each of the strings that could be
      read are meaningful for writing.

       'idle' will stop an active resync/recovery etc.  There is no
           guarantee that another resync/recovery may not be automatically
	   started again, though some event will be needed to trigger
           this.
	'resync' or 'recovery' can be used to restart the
           corresponding operation if it was stopped with 'idle'.
	'check' and 'repair' will start the appropriate process
           providing the current state is 'idle'.

486 487 488 489 490 491
      This file responds to select/poll.  Any important change in the value
      triggers a poll event.  Sometimes the value will briefly be
      "recover" if a recovery seems to be needed, but cannot be
      achieved. In that case, the transition to "recover" isn't
      notified, but the transition away is.

492 493 494 495 496 497 498
   degraded
      This contains a count of the number of devices by which the
      arrays is degraded.  So an optimal array with show '0'.  A
      single failed/missing drive will show '1', etc.
      This file responds to select/poll, any increase or decrease
      in the count of missing devices will trigger an event.

499 500 501 502 503 504 505 506 507
   mismatch_count
      When performing 'check' and 'repair', and possibly when
      performing 'resync', md will count the number of errors that are
      found.  The count in 'mismatch_cnt' is the number of sectors
      that were re-written, or (for 'check') would have been
      re-written.  As most raid levels work in units of pages rather
      than sectors, this my be larger than the number of actual errors
      by a factor of the number of sectors in a page.

508 509 510 511 512 513 514 515 516
   bitmap_set_bits
      If the array has a write-intent bitmap, then writing to this
      attribute can set bits in the bitmap, indicating that a resync
      would need to check the corresponding blocks. Either individual
      numbers or start-end pairs can be written.  Multiple numbers
      can be separated by a space.
      Note that the numbers are 'bit' numbers, not 'block' numbers.
      They should be scaled by the bitmap_chunksize.

517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533
   sync_speed_min
   sync_speed_max
     This are similar to /proc/sys/dev/raid/speed_limit_{min,max}
     however they only apply to the particular array.
     If no value has been written to these, of if the word 'system'
     is written, then the system-wide value is used.  If a value,
     in kibibytes-per-second is written, then it is used.
     When the files are read, they show the currently active value
     followed by "(local)" or "(system)" depending on whether it is
     a locally set or system-wide value.

   sync_completed
     This shows the number of sectors that have been completed of
     whatever the current sync_action is, followed by the number of
     sectors in total that could need to be processed.  The two
     numbers are separated by a '/'  thus effectively showing one
     value, a fraction of the process that is complete.
534 535 536 537 538 539 540 541 542 543
     A 'select' on this attribute will return when resync completes,
     when it reaches the current sync_max (below) and possibly at
     other times.

   sync_max
     This is a number of sectors at which point a resync/recovery
     process will pause.  When a resync is active, the value can
     only ever be increased, never decreased.  The value of 'max'
     effectively disables the limit.

544 545 546 547 548 549 550 551 552 553 554

   sync_speed
     This shows the current actual speed, in K/sec, of the current
     sync_action.  It is averaged over the last 30 seconds.

   suspend_lo
   suspend_hi
     The two values, given as numbers of sectors, indicate a range
     within the array where IO will be blocked.  This is currently
     only supported for raid4/5/6.

555 556 557
   sync_min
   sync_max
     The two values, given as numbers of sectors, indicate a range
N
NeilBrown 已提交
558
     within the array where 'check'/'repair' will operate. Must be
559 560 561 562 563 564
     a multiple of chunk_size. When it reaches "sync_max" it will
     pause, rather than complete.
     You can use 'select' or 'poll' on "sync_completed" to wait for
     that number to reach sync_max.  Then you can either increase
     "sync_max", or can write 'idle' to "sync_action".

565

566 567 568 569 570 571 572 573 574 575 576 577
Each active md device may also have attributes specific to the
personality module that manages it.
These are specific to the implementation of the module and could
change substantially if the implementation changes.

These currently include

  stripe_cache_size  (currently raid5 only)
      number of entries in the stripe cache.  This is writable, but
      there are upper and lower limits (32768, 16).  Default is 128.
  strip_cache_active (currently raid5 only)
      number of active entries in the stripe cache
578 579 580 581 582 583
  preread_bypass_threshold (currently raid5 only)
      number of times a stripe requiring preread will be bypassed by
      a stripe that does not require preread.  For fairness defaults
      to 1.  Setting this to 0 disables bypass accounting and
      requires preread stripes to wait until all full-width stripe-
      writes are complete.  Valid values are 0 to stripe_cache_size.