1RAID arrays
2===========
3
4Boot time assembly of RAID arrays
5---------------------------------
6
7Tools that manage md devices can be found at
8   http://www.kernel.org/pub/linux/utils/raid/
9
10
11You can boot with your md device with the following kernel command
12lines:
13
14for old raid arrays without persistent superblocks::
15
16  md=<md device no.>,<raid level>,<chunk size factor>,<fault level>,dev0,dev1,...,devn
17
18for raid arrays with persistent superblocks::
19
20  md=<md device no.>,dev0,dev1,...,devn
21
22or, to assemble a partitionable array::
23
24  md=d<md device no.>,dev0,dev1,...,devn
25
26``md device no.``
27+++++++++++++++++
28
29The number of the md device
30
31================= =========
32``md device no.`` device
33================= =========
34              0		md0
35	      1		md1
36	      2		md2
37	      3		md3
38	      4		md4
39================= =========
40
41``raid level``
42++++++++++++++
43
44level of the RAID array
45
46=============== =============
47``raid level``  level
48=============== =============
49-1		linear mode
500		striped mode
51=============== =============
52
53other modes are only supported with persistent super blocks
54
55``chunk size factor``
56+++++++++++++++++++++
57
58(raid-0 and raid-1 only)
59
60Set  the chunk size as 4k << n.
61
62``fault level``
63+++++++++++++++
64
65Totally ignored
66
67``dev0`` to ``devn``
68++++++++++++++++++++
69
70e.g. ``/dev/hda1``, ``/dev/hdc1``, ``/dev/sda1``, ``/dev/sdb1``
71
72A possible loadlin line (Harald Hoyer <HarryH@Royal.Net>)  looks like this::
73
74	e:\loadlin\loadlin e:\zimage root=/dev/md0 md=0,0,4,0,/dev/hdb2,/dev/hdc3 ro
75
76
77Boot time autodetection of RAID arrays
78--------------------------------------
79
80When md is compiled into the kernel (not as module), partitions of
81type 0xfd are scanned and automatically assembled into RAID arrays.
82This autodetection may be suppressed with the kernel parameter
83``raid=noautodetect``.  As of kernel 2.6.9, only drives with a type 0
84superblock can be autodetected and run at boot time.
85
86The kernel parameter ``raid=partitionable`` (or ``raid=part``) means
87that all auto-detected arrays are assembled as partitionable.
88
89Boot time assembly of degraded/dirty arrays
90-------------------------------------------
91
92If a raid5 or raid6 array is both dirty and degraded, it could have
93undetectable data corruption.  This is because the fact that it is
94``dirty`` means that the parity cannot be trusted, and the fact that it
95is degraded means that some datablocks are missing and cannot reliably
96be reconstructed (due to no parity).
97
98For this reason, md will normally refuse to start such an array.  This
99requires the sysadmin to take action to explicitly start the array
100despite possible corruption.  This is normally done with::
101
102   mdadm --assemble --force ....
103
104This option is not really available if the array has the root
105filesystem on it.  In order to support this booting from such an
106array, md supports a module parameter ``start_dirty_degraded`` which,
107when set to 1, bypassed the checks and will allows dirty degraded
108arrays to be started.
109
110So, to boot with a root filesystem of a dirty degraded raid 5 or 6, use::
111
112   md-mod.start_dirty_degraded=1
113
114
115Superblock formats
116------------------
117
118The md driver can support a variety of different superblock formats.
119Currently, it supports superblock formats ``0.90.0`` and the ``md-1`` format
120introduced in the 2.5 development series.
121
122The kernel will autodetect which format superblock is being used.
123
124Superblock format ``0`` is treated differently to others for legacy
125reasons - it is the original superblock format.
126
127
128General Rules - apply for all superblock formats
129------------------------------------------------
130
131An array is ``created`` by writing appropriate superblocks to all
132devices.
133
134It is ``assembled`` by associating each of these devices with an
135particular md virtual device.  Once it is completely assembled, it can
136be accessed.
137
138An array should be created by a user-space tool.  This will write
139superblocks to all devices.  It will usually mark the array as
140``unclean``, or with some devices missing so that the kernel md driver
141can create appropriate redundancy (copying in raid 1, parity
142calculation in raid 4/5).
143
144When an array is assembled, it is first initialized with the
145SET_ARRAY_INFO ioctl.  This contains, in particular, a major and minor
146version number.  The major version number selects which superblock
147format is to be used.  The minor number might be used to tune handling
148of the format, such as suggesting where on each device to look for the
149superblock.
150
151Then each device is added using the ADD_NEW_DISK ioctl.  This
152provides, in particular, a major and minor number identifying the
153device to add.
154
155The array is started with the RUN_ARRAY ioctl.
156
157Once started, new devices can be added.  They should have an
158appropriate superblock written to them, and then be passed in with
159ADD_NEW_DISK.
160
161Devices that have failed or are not yet active can be detached from an
162array using HOT_REMOVE_DISK.
163
164
165Specific Rules that apply to format-0 super block arrays, and arrays with no superblock (non-persistent)
166--------------------------------------------------------------------------------------------------------
167
168An array can be ``created`` by describing the array (level, chunksize
169etc) in a SET_ARRAY_INFO ioctl.  This must have ``major_version==0`` and
170``raid_disks != 0``.
171
172Then uninitialized devices can be added with ADD_NEW_DISK.  The
173structure passed to ADD_NEW_DISK must specify the state of the device
174and its role in the array.
175
176Once started with RUN_ARRAY, uninitialized spares can be added with
177HOT_ADD_DISK.
178
179
180MD devices in sysfs
181-------------------
182
183md devices appear in sysfs (``/sys``) as regular block devices,
184e.g.::
185
186   /sys/block/md0
187
188Each ``md`` device will contain a subdirectory called ``md`` which
189contains further md-specific information about the device.
190
191All md devices contain:
192
193  level
194     a text file indicating the ``raid level``. e.g. raid0, raid1,
195     raid5, linear, multipath, faulty.
196     If no raid level has been set yet (array is still being
197     assembled), the value will reflect whatever has been written
198     to it, which may be a name like the above, or may be a number
199     such as ``0``, ``5``, etc.
200
201  raid_disks
202     a text file with a simple number indicating the number of devices
203     in a fully functional array.  If this is not yet known, the file
204     will be empty.  If an array is being resized this will contain
205     the new number of devices.
206     Some raid levels allow this value to be set while the array is
207     active.  This will reconfigure the array.   Otherwise it can only
208     be set while assembling an array.
209     A change to this attribute will not be permitted if it would
210     reduce the size of the array.  To reduce the number of drives
211     in an e.g. raid5, the array size must first be reduced by
212     setting the ``array_size`` attribute.
213
214  chunk_size
215     This is the size in bytes for ``chunks`` and is only relevant to
216     raid levels that involve striping (0,4,5,6,10). The address space
217     of the array is conceptually divided into chunks and consecutive
218     chunks are striped onto neighbouring devices.
219     The size should be at least PAGE_SIZE (4k) and should be a power
220     of 2.  This can only be set while assembling an array
221
222  layout
223     The ``layout`` for the array for the particular level.  This is
224     simply a number that is interpretted differently by different
225     levels.  It can be written while assembling an array.
226
227  array_size
228     This can be used to artificially constrain the available space in
229     the array to be less than is actually available on the combined
230     devices.  Writing a number (in Kilobytes) which is less than
231     the available size will set the size.  Any reconfiguration of the
232     array (e.g. adding devices) will not cause the size to change.
233     Writing the word ``default`` will cause the effective size of the
234     array to be whatever size is actually available based on
235     ``level``, ``chunk_size`` and ``component_size``.
236
237     This can be used to reduce the size of the array before reducing
238     the number of devices in a raid4/5/6, or to support external
239     metadata formats which mandate such clipping.
240
241  reshape_position
242     This is either ``none`` or a sector number within the devices of
243     the array where ``reshape`` is up to.  If this is set, the three
244     attributes mentioned above (raid_disks, chunk_size, layout) can
245     potentially have 2 values, an old and a new value.  If these
246     values differ, reading the attribute returns::
247
248        new (old)
249
250     and writing will effect the ``new`` value, leaving the ``old``
251     unchanged.
252
253  component_size
254     For arrays with data redundancy (i.e. not raid0, linear, faulty,
255     multipath), all components must be the same size - or at least
256     there must a size that they all provide space for.  This is a key
257     part or the geometry of the array.  It is measured in sectors
258     and can be read from here.  Writing to this value may resize
259     the array if the personality supports it (raid1, raid5, raid6),
260     and if the component drives are large enough.
261
262  metadata_version
263     This indicates the format that is being used to record metadata
264     about the array.  It can be 0.90 (traditional format), 1.0, 1.1,
265     1.2 (newer format in varying locations) or ``none`` indicating that
266     the kernel isn't managing metadata at all.
267     Alternately it can be ``external:`` followed by a string which
268     is set by user-space.  This indicates that metadata is managed
269     by a user-space program.  Any device failure or other event that
270     requires a metadata update will cause array activity to be
271     suspended until the event is acknowledged.
272
273  resync_start
274     The point at which resync should start.  If no resync is needed,
275     this will be a very large number (or ``none`` since 2.6.30-rc1).  At
276     array creation it will default to 0, though starting the array as
277     ``clean`` will set it much larger.
278
279  new_dev
280     This file can be written but not read.  The value written should
281     be a block device number as major:minor.  e.g. 8:0
282     This will cause that device to be attached to the array, if it is
283     available.  It will then appear at md/dev-XXX (depending on the
284     name of the device) and further configuration is then possible.
285
286  safe_mode_delay
287     When an md array has seen no write requests for a certain period
288     of time, it will be marked as ``clean``.  When another write
289     request arrives, the array is marked as ``dirty`` before the write
290     commences.  This is known as ``safe_mode``.
291     The ``certain period`` is controlled by this file which stores the
292     period as a number of seconds.  The default is 200msec (0.200).
293     Writing a value of 0 disables safemode.
294
295  array_state
296     This file contains a single word which describes the current
297     state of the array.  In many cases, the state can be set by
298     writing the word for the desired state, however some states
299     cannot be explicitly set, and some transitions are not allowed.
300
301     Select/poll works on this file.  All changes except between
302     Active_idle and active (which can be frequent and are not
303     very interesting) are notified.  active->active_idle is
304     reported if the metadata is externally managed.
305
306     clear
307         No devices, no size, no level
308
309         Writing is equivalent to STOP_ARRAY ioctl
310
311     inactive
312         May have some settings, but array is not active
313         all IO results in error
314
315         When written, doesn't tear down array, but just stops it
316
317     suspended (not supported yet)
318         All IO requests will block. The array can be reconfigured.
319
320         Writing this, if accepted, will block until array is quiessent
321
322     readonly
323         no resync can happen.  no superblocks get written.
324
325         Write requests fail
326
327     read-auto
328         like readonly, but behaves like ``clean`` on a write request.
329
330     clean
331         no pending writes, but otherwise active.
332
333         When written to inactive array, starts without resync
334
335         If a write request arrives then
336         if metadata is known, mark ``dirty`` and switch to ``active``.
337         if not known, block and switch to write-pending
338
339         If written to an active array that has pending writes, then fails.
340     active
341         fully active: IO and resync can be happening.
342         When written to inactive array, starts with resync
343
344     write-pending
345         clean, but writes are blocked waiting for ``active`` to be written.
346
347     active-idle
348         like active, but no writes have been seen for a while (safe_mode_delay).
349
350  bitmap/location
351     This indicates where the write-intent bitmap for the array is
352     stored.
353
354     It can be one of ``none``, ``file`` or ``[+-]N``.
355     ``file`` may later be extended to ``file:/file/name``
356     ``[+-]N`` means that many sectors from the start of the metadata.
357
358     This is replicated on all devices.  For arrays with externally
359     managed metadata, the offset is from the beginning of the
360     device.
361
362  bitmap/chunksize
363     The size, in bytes, of the chunk which will be represented by a
364     single bit.  For RAID456, it is a portion of an individual
365     device. For RAID10, it is a portion of the array.  For RAID1, it
366     is both (they come to the same thing).
367
368  bitmap/time_base
369     The time, in seconds, between looking for bits in the bitmap to
370     be cleared. In the current implementation, a bit will be cleared
371     between 2 and 3 times ``time_base`` after all the covered blocks
372     are known to be in-sync.
373
374  bitmap/backlog
375     When write-mostly devices are active in a RAID1, write requests
376     to those devices proceed in the background - the filesystem (or
377     other user of the device) does not have to wait for them.
378     ``backlog`` sets a limit on the number of concurrent background
379     writes.  If there are more than this, new writes will by
380     synchronous.
381
382  bitmap/metadata
383     This can be either ``internal`` or ``external``.
384
385     ``internal``
386       is the default and means the metadata for the bitmap
387       is stored in the first 256 bytes of the allocated space and is
388       managed by the md module.
389
390     ``external``
391       means that bitmap metadata is managed externally to
392       the kernel (i.e. by some userspace program)
393
394  bitmap/can_clear
395     This is either ``true`` or ``false``.  If ``true``, then bits in the
396     bitmap will be cleared when the corresponding blocks are thought
397     to be in-sync.  If ``false``, bits will never be cleared.
398     This is automatically set to ``false`` if a write happens on a
399     degraded array, or if the array becomes degraded during a write.
400     When metadata is managed externally, it should be set to true
401     once the array becomes non-degraded, and this fact has been
402     recorded in the metadata.
403
404  consistency_policy
405     This indicates how the array maintains consistency in case of unexpected
406     shutdown. It can be:
407
408     none
409       Array has no redundancy information, e.g. raid0, linear.
410
411     resync
412       Full resync is performed and all redundancy is regenerated when the
413       array is started after unclean shutdown.
414
415     bitmap
416       Resync assisted by a write-intent bitmap.
417
418     journal
419       For raid4/5/6, journal device is used to log transactions and replay
420       after unclean shutdown.
421
422     ppl
423       For raid5 only, Partial Parity Log is used to close the write hole and
424       eliminate resync.
425
426     The accepted values when writing to this file are ``ppl`` and ``resync``,
427     used to enable and disable PPL.
428
429
430As component devices are added to an md array, they appear in the ``md``
431directory as new directories named::
432
433      dev-XXX
434
435where ``XXX`` is a name that the kernel knows for the device, e.g. hdb1.
436Each directory contains:
437
438      block
439        a symlink to the block device in /sys/block, e.g.::
440
441	     /sys/block/md0/md/dev-hdb1/block -> ../../../../block/hdb/hdb1
442
443      super
444        A file containing an image of the superblock read from, or
445        written to, that device.
446
447      state
448	A file recording the current state of the device in the array
449	which can be a comma separated list of:
450
451	      faulty
452			device has been kicked from active use due to
453			a detected fault, or it has unacknowledged bad
454			blocks
455
456	      in_sync
457			device is a fully in-sync member of the array
458
459	      writemostly
460			device will only be subject to read
461			requests if there are no other options.
462
463			This applies only to raid1 arrays.
464
465	      blocked
466			device has failed, and the failure hasn't been
467			acknowledged yet by the metadata handler.
468
469			Writes that would write to this device if
470			it were not faulty are blocked.
471
472	      spare
473			device is working, but not a full member.
474
475			This includes spares that are in the process
476			of being recovered to
477
478	      write_error
479			device has ever seen a write error.
480
481	      want_replacement
482			device is (mostly) working but probably
483			should be replaced, either due to errors or
484			due to user request.
485
486	      replacement
487			device is a replacement for another active
488			device with same raid_disk.
489
490
491	This list may grow in future.
492
493	This can be written to.
494
495	Writing ``faulty``  simulates a failure on the device.
496
497	Writing ``remove`` removes the device from the array.
498
499	Writing ``writemostly`` sets the writemostly flag.
500
501	Writing ``-writemostly`` clears the writemostly flag.
502
503	Writing ``blocked`` sets the ``blocked`` flag.
504
505	Writing ``-blocked`` clears the ``blocked`` flags and allows writes
506	to complete and possibly simulates an error.
507
508	Writing ``in_sync`` sets the in_sync flag.
509
510	Writing ``write_error`` sets writeerrorseen flag.
511
512	Writing ``-write_error`` clears writeerrorseen flag.
513
514	Writing ``want_replacement`` is allowed at any time except to a
515	replacement device or a spare.  It sets the flag.
516
517	Writing ``-want_replacement`` is allowed at any time.  It clears
518	the flag.
519
520	Writing ``replacement`` or ``-replacement`` is only allowed before
521	starting the array.  It sets or clears the flag.
522
523
524	This file responds to select/poll. Any change to ``faulty``
525	or ``blocked`` causes an event.
526
527      errors
528	An approximate count of read errors that have been detected on
529	this device but have not caused the device to be evicted from
530	the array (either because they were corrected or because they
531	happened while the array was read-only).  When using version-1
532	metadata, this value persists across restarts of the array.
533
534	This value can be written while assembling an array thus
535	providing an ongoing count for arrays with metadata managed by
536	userspace.
537
538      slot
539        This gives the role that the device has in the array.  It will
540	either be ``none`` if the device is not active in the array
541        (i.e. is a spare or has failed) or an integer less than the
542	``raid_disks`` number for the array indicating which position
543	it currently fills.  This can only be set while assembling an
544	array.  A device for which this is set is assumed to be working.
545
546      offset
547        This gives the location in the device (in sectors from the
548        start) where data from the array will be stored.  Any part of
549        the device before this offset is not touched, unless it is
550        used for storing metadata (Formats 1.1 and 1.2).
551
552      size
553        The amount of the device, after the offset, that can be used
554        for storage of data.  This will normally be the same as the
555	component_size.  This can be written while assembling an
556        array.  If a value less than the current component_size is
557        written, it will be rejected.
558
559      recovery_start
560        When the device is not ``in_sync``, this records the number of
561	sectors from the start of the device which are known to be
562	correct.  This is normally zero, but during a recovery
563	operation it will steadily increase, and if the recovery is
564	interrupted, restoring this value can cause recovery to
565	avoid repeating the earlier blocks.  With v1.x metadata, this
566	value is saved and restored automatically.
567
568	This can be set whenever the device is not an active member of
569	the array, either before the array is activated, or before
570	the ``slot`` is set.
571
572	Setting this to ``none`` is equivalent to setting ``in_sync``.
573	Setting to any other value also clears the ``in_sync`` flag.
574
575      bad_blocks
576	This gives the list of all known bad blocks in the form of
577	start address and length (in sectors respectively). If output
578	is too big to fit in a page, it will be truncated. Writing
579	``sector length`` to this file adds new acknowledged (i.e.
580	recorded to disk safely) bad blocks.
581
582      unacknowledged_bad_blocks
583	This gives the list of known-but-not-yet-saved-to-disk bad
584	blocks in the same form of ``bad_blocks``. If output is too big
585	to fit in a page, it will be truncated. Writing to this file
586	adds bad blocks without acknowledging them. This is largely
587	for testing.
588
589      ppl_sector, ppl_size
590        Location and size (in sectors) of the space used for Partial Parity Log
591        on this device.
592
593
594An active md device will also contain an entry for each active device
595in the array.  These are named::
596
597    rdNN
598
599where ``NN`` is the position in the array, starting from 0.
600So for a 3 drive array there will be rd0, rd1, rd2.
601These are symbolic links to the appropriate ``dev-XXX`` entry.
602Thus, for example::
603
604       cat /sys/block/md*/md/rd*/state
605
606will show ``in_sync`` on every line.
607
608
609
610Active md devices for levels that support data redundancy (1,4,5,6,10)
611also have
612
613   sync_action
614     a text file that can be used to monitor and control the rebuild
615     process.  It contains one word which can be one of:
616
617       resync
618		redundancy is being recalculated after unclean
619                shutdown or creation
620
621       recover
622		a hot spare is being built to replace a
623		failed/missing device
624
625       idle
626		nothing is happening
627       check
628		A full check of redundancy was requested and is
629                happening.  This reads all blocks and checks
630                them. A repair may also happen for some raid
631                levels.
632
633       repair
634		A full check and repair is happening.  This is
635		similar to ``resync``, but was requested by the
636                user, and the write-intent bitmap is NOT used to
637		optimise the process.
638
639      This file is writable, and each of the strings that could be
640      read are meaningful for writing.
641
642	``idle`` will stop an active resync/recovery etc.  There is no
643	guarantee that another resync/recovery may not be automatically
644	started again, though some event will be needed to trigger
645	this.
646
647	``resync`` or ``recovery`` can be used to restart the
648        corresponding operation if it was stopped with ``idle``.
649
650	``check`` and ``repair`` will start the appropriate process
651	providing the current state is ``idle``.
652
653      This file responds to select/poll.  Any important change in the value
654      triggers a poll event.  Sometimes the value will briefly be
655      ``recover`` if a recovery seems to be needed, but cannot be
656      achieved. In that case, the transition to ``recover`` isn't
657      notified, but the transition away is.
658
659   degraded
660      This contains a count of the number of devices by which the
661      arrays is degraded.  So an optimal array will show ``0``.  A
662      single failed/missing drive will show ``1``, etc.
663
664      This file responds to select/poll, any increase or decrease
665      in the count of missing devices will trigger an event.
666
667   mismatch_count
668      When performing ``check`` and ``repair``, and possibly when
669      performing ``resync``, md will count the number of errors that are
670      found.  The count in ``mismatch_cnt`` is the number of sectors
671      that were re-written, or (for ``check``) would have been
672      re-written.  As most raid levels work in units of pages rather
673      than sectors, this may be larger than the number of actual errors
674      by a factor of the number of sectors in a page.
675
676   bitmap_set_bits
677      If the array has a write-intent bitmap, then writing to this
678      attribute can set bits in the bitmap, indicating that a resync
679      would need to check the corresponding blocks. Either individual
680      numbers or start-end pairs can be written.  Multiple numbers
681      can be separated by a space.
682
683      Note that the numbers are ``bit`` numbers, not ``block`` numbers.
684      They should be scaled by the bitmap_chunksize.
685
686   sync_speed_min, sync_speed_max
687     This are similar to ``/proc/sys/dev/raid/speed_limit_{min,max}``
688     however they only apply to the particular array.
689
690     If no value has been written to these, or if the word ``system``
691     is written, then the system-wide value is used.  If a value,
692     in kibibytes-per-second is written, then it is used.
693
694     When the files are read, they show the currently active value
695     followed by ``(local)`` or ``(system)`` depending on whether it is
696     a locally set or system-wide value.
697
698   sync_completed
699     This shows the number of sectors that have been completed of
700     whatever the current sync_action is, followed by the number of
701     sectors in total that could need to be processed.  The two
702     numbers are separated by a ``/``  thus effectively showing one
703     value, a fraction of the process that is complete.
704
705     A ``select`` on this attribute will return when resync completes,
706     when it reaches the current sync_max (below) and possibly at
707     other times.
708
709   sync_speed
710     This shows the current actual speed, in K/sec, of the current
711     sync_action.  It is averaged over the last 30 seconds.
712
713   suspend_lo, suspend_hi
714     The two values, given as numbers of sectors, indicate a range
715     within the array where IO will be blocked.  This is currently
716     only supported for raid4/5/6.
717
718   sync_min, sync_max
719     The two values, given as numbers of sectors, indicate a range
720     within the array where ``check``/``repair`` will operate. Must be
721     a multiple of chunk_size. When it reaches ``sync_max`` it will
722     pause, rather than complete.
723     You can use ``select`` or ``poll`` on ``sync_completed`` to wait for
724     that number to reach sync_max.  Then you can either increase
725     ``sync_max``, or can write ``idle`` to ``sync_action``.
726
727     The value of ``max`` for ``sync_max`` effectively disables the limit.
728     When a resync is active, the value can only ever be increased,
729     never decreased.
730     The value of ``0`` is the minimum for ``sync_min``.
731
732
733
734Each active md device may also have attributes specific to the
735personality module that manages it.
736These are specific to the implementation of the module and could
737change substantially if the implementation changes.
738
739These currently include:
740
741  stripe_cache_size  (currently raid5 only)
742      number of entries in the stripe cache.  This is writable, but
743      there are upper and lower limits (32768, 17).  Default is 256.
744
745  strip_cache_active (currently raid5 only)
746      number of active entries in the stripe cache
747
748  preread_bypass_threshold (currently raid5 only)
749      number of times a stripe requiring preread will be bypassed by
750      a stripe that does not require preread.  For fairness defaults
751      to 1.  Setting this to 0 disables bypass accounting and
752      requires preread stripes to wait until all full-width stripe-
753      writes are complete.  Valid values are 0 to stripe_cache_size.
754
755  journal_mode (currently raid5 only)
756      The cache mode for raid5. raid5 could include an extra disk for
757      caching. The mode can be "write-throuth" and "write-back". The
758      default is "write-through".
759
760  ppl_write_hint
761      NVMe stream ID to be set for each PPL write request.
762