1 .\" Copyright Neil Brown and others.
2 .\" This program is free software; you can redistribute it and/or modify
3 .\" it under the terms of the GNU General Public License as published by
4 .\" the Free Software Foundation; either version 2 of the License, or
5 .\" (at your option) any later version.
6 .\" See file COPYING in distribution for details.
9 md \- Multiple Device driver aka Linux Software RAID
17 driver provides virtual devices that are created from one or more
18 independent underlying devices. This array of devices often contains
19 redundancy and the devices are often disk drives, hence the acronym RAID
20 which stands for a Redundant Array of Independent Disks.
25 4 (striped array with parity device),
26 5 (striped array with distributed parity information),
27 6 (striped array with distributed dual redundancy information), and
28 10 (striped and mirrored).
29 If some number of underlying devices fails while using one of these
30 levels, the array will continue to function; this number is one for
31 RAID levels 4 and 5, two for RAID level 6, and all but one (N-1) for
32 RAID level 1, and dependent on configuration for level 10.
35 also supports a number of pseudo RAID (non-redundant) configurations
36 including RAID0 (striped array), LINEAR (catenated array),
37 MULTIPATH (a set of different interfaces to the same device),
38 and FAULTY (a layer over a single device into which errors can be injected).
41 Each device in an array may have a
43 which records information about the structure and state of the array.
44 This allows the array to be reliably re-assembled after a shutdown.
46 From Linux kernel version 2.6.10,
48 provides support for two different formats of this superblock, and
49 other formats can be added. Prior to this release, only one format is
52 The common format \(em known as version 0.90 \(em has
53 a superblock that is 4K long and is written into a 64K aligned block that
54 starts at least 64K and less than 128K from the end of the device
55 (i.e. to get the address of the superblock round the size of the
56 device down to a multiple of 64K and then subtract 64K).
57 The available size of each device is the amount of space before the
58 super block, so between 64K and 128K is lost when a device in
59 incorporated into an MD array.
60 This superblock stores multi-byte fields in a processor-dependent
61 manner, so arrays cannot easily be moved between computers with
64 The new format \(em known as version 1 \(em has a superblock that is
65 normally 1K long, but can be longer. It is normally stored between 8K
66 and 12K from the end of the device, on a 4K boundary, though
67 variations can be stored at the start of the device (version 1.1) or 4K from
68 the start of the device (version 1.2).
69 This superblock format stores multibyte data in a
70 processor-independent format and supports up to hundreds of
71 component devices (version 0.90 only supports 28).
73 The superblock contains, among other things:
76 The manner in which the devices are arranged into the array
77 (linear, raid0, raid1, raid4, raid5, raid10, multipath).
80 a 128 bit Universally Unique Identifier that identifies the array that
83 When a version 0.90 array is being reshaped (e.g. adding extra devices
84 to a RAID5), the version number is temporarily set to 0.91. This
85 ensures that if the reshape process is stopped in the middle (e.g. by
86 a system crash) and the machine boots into an older kernel that does
87 not support reshaping, then the array will not be assembled (which
88 would cause data corruption) but will be left untouched until a kernel
89 that can complete the reshape processes is used.
91 .SS ARRAYS WITHOUT SUPERBLOCKS
92 While it is usually best to create arrays with superblocks so that
93 they can be assembled reliably, there are some circumstances when an
94 array without superblocks is preferred. These include:
99 driver only supported Linear and Raid0 configurations and did not use
100 a superblock (which is less critical with these configurations).
101 While such arrays should be rebuilt with superblocks if possible,
103 continues to support them.
106 Being a largely transparent layer over a different device, the FAULTY
107 personality doesn't gain anything from having a superblock.
110 It is often possible to detect devices which are different paths to
111 the same storage directly rather than having a distinctive superblock
112 written to the device and searched for on all paths. In this case,
113 a MULTIPATH array with no superblock makes sense.
116 In some configurations it might be desired to create a raid1
117 configuration that does not use a superblock, and to maintain the state of
118 the array elsewhere. While not encouraged for general us, it does
119 have special-purpose uses and is supported.
123 A linear array simply catenates the available space on each
124 drive to form one large virtual drive.
126 One advantage of this arrangement over the more common RAID0
127 arrangement is that the array may be reconfigured at a later time with
128 an extra drive, so the array is made bigger without disturbing the
129 data that is on the array. This can even be done on a live
132 If a chunksize is given with a LINEAR array, the usable space on each
133 device is rounded down to a multiple of this chunksize.
137 A RAID0 array (which has zero redundancy) is also known as a
139 A RAID0 array is configured at creation with a
141 which must be a power of two, and at least 4 kibibytes.
143 The RAID0 driver assigns the first chunk of the array to the first
144 device, the second chunk to the second device, and so on until all
145 drives have been assigned one chunk. This collection of chunks forms
148 Further chunks are gathered into stripes in the same way, and are
149 assigned to the remaining space in the drives.
151 If devices in the array are not all the same size, then once the
152 smallest device has been exhausted, the RAID0 driver starts
153 collecting chunks into smaller stripes that only span the drives which
154 still have remaining space.
159 A RAID1 array is also known as a mirrored set (though mirrors tend to
160 provide reflected images, which RAID1 does not) or a plex.
162 Once initialised, each device in a RAID1 array contains exactly the
163 same data. Changes are written to all devices in parallel. Data is
164 read from any one device. The driver attempts to distribute read
165 requests across all devices to maximise performance.
167 All devices in a RAID1 array should be the same size. If they are
168 not, then only the amount of space available on the smallest device is
169 used (any extra space on other devices is wasted).
171 Note that the read balancing done by the driver does not make the RAID1
172 performance profile be the same as for RAID0; a single stream of
173 sequential input will not be accelerated (e.g. a single dd), but
174 multiple sequential streams or a random workload will use more than one
175 spindle. In theory, having an N-disk RAID1 will allow N sequential
176 threads to read from all disks.
180 A RAID4 array is like a RAID0 array with an extra device for storing
181 parity. This device is the last of the active devices in the
182 array. Unlike RAID0, RAID4 also requires that all stripes span all
183 drives, so extra space on devices that are larger than the smallest is
186 When any block in a RAID4 array is modified, the parity block for that
187 stripe (i.e. the block in the parity device at the same device offset
188 as the stripe) is also modified so that the parity block always
189 contains the "parity" for the whole stripe. I.e. its content is
190 equivalent to the result of performing an exclusive-or operation
191 between all the data blocks in the stripe.
193 This allows the array to continue to function if one device fails.
194 The data that was on that device can be calculated as needed from the
195 parity block and the other data blocks.
199 RAID5 is very similar to RAID4. The difference is that the parity
200 blocks for each stripe, instead of being on a single device, are
201 distributed across all devices. This allows more parallelism when
202 writing, as two different block updates will quite possibly affect
203 parity blocks on different devices so there is less contention.
205 This also allows more parallelism when reading, as read requests are
206 distributed over all the devices in the array instead of all but one.
210 RAID6 is similar to RAID5, but can handle the loss of any \fItwo\fP
211 devices without data loss. Accordingly, it requires N+2 drives to
212 store N drives worth of data.
214 The performance for RAID6 is slightly lower but comparable to RAID5 in
215 normal mode and single disk failure mode. It is very slow in dual
216 disk failure mode, however.
220 RAID10 provides a combination of RAID1 and RAID0, and is sometimes known
221 as RAID1+0. Every datablock is duplicated some number of times, and
222 the resulting collection of datablocks are distributed over multiple
225 When configuring a RAID10 array, it is necessary to specify the number
226 of replicas of each data block that are required (this will normally
227 be 2) and whether the replicas should be 'near', 'offset' or 'far'.
228 (Note that the 'offset' layout is only available from 2.6.18).
230 When 'near' replicas are chosen, the multiple copies of a given chunk
231 are laid out consecutively across the stripes of the array, so the two
232 copies of a datablock will likely be at the same offset on two
235 When 'far' replicas are chosen, the multiple copies of a given chunk
236 are laid out quite distant from each other. The first copy of all
237 data blocks will be striped across the early part of all drives in
238 RAID0 fashion, and then the next copy of all blocks will be striped
239 across a later section of all drives, always ensuring that all copies
240 of any given block are on different drives.
242 The 'far' arrangement can give sequential read performance equal to
243 that of a RAID0 array, but at the cost of reduced write performance.
245 When 'offset' replicas are chosen, the multiple copies of a given
246 chunk are laid out on consecutive drives and at consecutive offsets.
247 Effectively each stripe is duplicated and the copies are offset by one
248 device. This should give similar read characteristics to 'far' if a
249 suitably large chunk size is used, but without as much seeking for
252 It should be noted that the number of devices in a RAID10 array need
253 not be a multiple of the number of replica of each data block; however,
254 there must be at least as many devices as replicas.
256 If, for example, an array is created with 5 devices and 2 replicas,
257 then space equivalent to 2.5 of the devices will be available, and
258 every block will be stored on two different devices.
260 Finally, it is possible to have an array with both 'near' and 'far'
261 copies. If an array is configured with 2 near copies and 2 far
262 copies, then there will be a total of 4 copies of each block, each on
263 a different drive. This is an artifact of the implementation and is
264 unlikely to be of real value.
268 MULTIPATH is not really a RAID at all as there is only one real device
269 in a MULTIPATH md array. However there are multiple access points
270 (paths) to this device, and one of these paths might fail, so there
271 are some similarities.
273 A MULTIPATH array is composed of a number of logically different
274 devices, often fibre channel interfaces, that all refer the the same
275 real device. If one of these interfaces fails (e.g. due to cable
276 problems), the multipath driver will attempt to redirect requests to
280 The FAULTY md module is provided for testing purposes. A faulty array
281 has exactly one component device and is normally assembled without a
282 superblock, so the md array created provides direct access to all of
283 the data in the component device.
285 The FAULTY module may be requested to simulate faults to allow testing
286 of other md levels or of filesystems. Faults can be chosen to trigger
287 on read requests or write requests, and can be transient (a subsequent
288 read/write at the address will probably succeed) or persistent
289 (subsequent read/write of the same address will fail). Further, read
290 faults can be "fixable" meaning that they persist until a write
291 request at the same address.
293 Fault types can be requested with a period. In this case, the fault
294 will recur repeatedly after the given number of requests of the
295 relevant type. For example if persistent read faults have a period of
296 100, then every 100th read request would generate a fault, and the
297 faulty sector would be recorded so that subsequent reads on that
298 sector would also fail.
300 There is a limit to the number of faulty sectors that are remembered.
301 Faults generated after this limit is exhausted are treated as
304 The list of faulty sectors can be flushed, and the active list of
305 failure modes can be cleared.
309 When changes are made to a RAID1, RAID4, RAID5, RAID6, or RAID10 array
310 there is a possibility of inconsistency for short periods of time as
311 each update requires at least two block to be written to different
312 devices, and these writes probably won't happen at exactly the same
313 time. Thus if a system with one of these arrays is shutdown in the
314 middle of a write operation (e.g. due to power failure), the array may
317 To handle this situation, the md driver marks an array as "dirty"
318 before writing any data to it, and marks it as "clean" when the array
319 is being disabled, e.g. at shutdown. If the md driver finds an array
320 to be dirty at startup, it proceeds to correct any possibly
321 inconsistency. For RAID1, this involves copying the contents of the
322 first drive onto all other drives. For RAID4, RAID5 and RAID6 this
323 involves recalculating the parity for each stripe and making sure that
324 the parity block has the correct data. For RAID10 it involves copying
325 one of the replicas of each block onto all the others. This process,
326 known as "resynchronising" or "resync" is performed in the background.
327 The array can still be used, though possibly with reduced performance.
329 If a RAID4, RAID5 or RAID6 array is degraded (missing at least one
330 drive, two for RAID6) when it is restarted after an unclean shutdown, it cannot
331 recalculate parity, and so it is possible that data might be
332 undetectably corrupted. The 2.4 md driver
334 alert the operator to this condition. The 2.6 md driver will fail to
335 start an array in this condition without manual intervention, though
336 this behaviour can be overridden by a kernel parameter.
340 If the md driver detects a write error on a device in a RAID1, RAID4,
341 RAID5, RAID6, or RAID10 array, it immediately disables that device
342 (marking it as faulty) and continues operation on the remaining
343 devices. If there are spare drives, the driver will start recreating
344 on one of the spare drives the data which was on that failed drive,
345 either by copying a working drive in a RAID1 configuration, or by
346 doing calculations with the parity block on RAID4, RAID5 or RAID6, or
347 by finding and copying originals for RAID10.
349 In kernels prior to about 2.6.15, a read error would cause the same
350 effect as a write error. In later kernels, a read-error will instead
351 cause md to attempt a recovery by overwriting the bad block. i.e. it
352 will find the correct data from elsewhere, write it over the block
353 that failed, and then try to read it back again. If either the write
354 or the re-read fail, md will treat the error the same way that a write
355 error is treated, and will fail the whole device.
357 While this recovery process is happening, the md driver will monitor
358 accesses to the array and will slow down the rate of recovery if other
359 activity is happening, so that normal access to the array will not be
360 unduly affected. When no other activity is happening, the recovery
361 process proceeds at full speed. The actual speed targets for the two
362 different situations can be controlled by the
366 control files mentioned below.
368 .SS BITMAP WRITE-INTENT LOGGING
372 supports a bitmap based write-intent log. If configured, the bitmap
373 is used to record which blocks of the array may be out of sync.
374 Before any write request is honoured, md will make sure that the
375 corresponding bit in the log is set. After a period of time with no
376 writes to an area of the array, the corresponding bit will be cleared.
378 This bitmap is used for two optimisations.
380 Firstly, after an unclean shutdown, the resync process will consult
381 the bitmap and only resync those blocks that correspond to bits in the
382 bitmap that are set. This can dramatically reduce resync time.
384 Secondly, when a drive fails and is removed from the array, md stops
385 clearing bits in the intent log. If that same drive is re-added to
386 the array, md will notice and will only recover the sections of the
387 drive that are covered by bits in the intent log that are set. This
388 can allow a device to be temporarily removed and reinserted without
389 causing an enormous recovery cost.
391 The intent log can be stored in a file on a separate device, or it can
392 be stored near the superblocks of an array which has superblocks.
394 It is possible to add an intent log to an active array, or remove an
395 intent log if one is present.
397 In 2.6.13, intent bitmaps are only supported with RAID1. Other levels
398 with redundancy are supported from 2.6.15.
404 supports WRITE-BEHIND on RAID1 arrays.
406 This allows certain devices in the array to be flagged as
408 MD will only read from such devices if there is no
411 If a write-intent bitmap is also provided, write requests to
412 write-mostly devices will be treated as write-behind requests and md
413 will not wait for writes to those requests to complete before
414 reporting the write as complete to the filesystem.
416 This allows for a RAID1 with WRITE-BEHIND to be used to mirror data
417 over a slow link to a remote computer (providing the link isn't too
418 slow). The extra latency of the remote link will not slow down normal
419 operations, but the remote system will still have a reasonably
420 up-to-date copy of all data.
427 is the processes of re-arranging the data stored in each stripe into a
428 new layout. This might involve changing the number of devices in the
429 array (so the stripes are wider), changing the chunk size (so stripes
430 are deeper or shallower), or changing the arrangement of data and
431 parity (possibly changing the raid level, e.g. 1 to 5 or 5 to 6).
433 As of Linux 2.6.17, md can reshape a raid5 array to have more
434 devices. Other possibilities may follow in future kernels.
436 During any stripe process there is a 'critical section' during which
437 live data is being overwritten on disk. For the operation of
438 increasing the number of drives in a raid5, this critical section
439 covers the first few stripes (the number being the product of the old
440 and new number of devices). After this critical section is passed,
441 data is only written to areas of the array which no longer hold live
442 data \(em the live data has already been located away.
444 md is not able to ensure data preservation if there is a crash
445 (e.g. power failure) during the critical section. If md is asked to
446 start an array which failed during a critical section of restriping,
447 it will fail to start the array.
449 To deal with this possibility, a user-space program must
451 Disable writes to that section of the array (using the
455 take a copy of the data somewhere (i.e. make a backup),
457 allow the process to continue and invalidate the backup and restore
458 write access once the critical section is passed, and
460 provide for restoring the critical data before restarting the array
461 after a system crash.
465 versions from 2.4 do this for growing a RAID5 array.
467 For operations that do not change the size of the array, like simply
468 increasing chunk size, or converting RAID5 to RAID6 with one extra
469 device, the entire process is the critical section. In this case, the
470 restripe will need to progress in stages, as a section is suspended,
472 restriped, and released; this is not yet implemented.
475 Each block device appears as a directory in
477 (which is usually mounted at
479 For MD devices, this directory will contain a subdirectory called
481 which contains various files for providing access to information about
484 This interface is documented more fully in the file
485 .B Documentation/md.txt
486 which is distributed with the kernel sources. That file should be
487 consulted for full documentation. The following are just a selection
488 of attribute files that are available.
492 This value, if set, overrides the system-wide setting in
493 .B /proc/sys/dev/raid/speed_limit_min
497 to this file will cause the system-wide setting to have effect.
501 This is the partner of
504 .B /proc/sys/dev/raid/spool_limit_max
509 This can be used to monitor and control the resync/recovery process of
511 In particular, writing "check" here will cause the array to read all
512 data block and check that they are consistent (e.g. parity is correct,
513 or all mirror replicas are the same). Any discrepancies found are
517 A count of problems found will be stored in
518 .BR md/mismatch_count .
520 Alternately, "repair" can be written which will cause the same check
521 to be performed, but any errors will be corrected.
523 Finally, "idle" can be written to stop the check/repair process.
526 .B md/stripe_cache_size
527 This is only available on RAID5 and RAID6. It records the size (in
528 pages per device) of the stripe cache which is used for synchronising
529 all write operations to the array and all read operations if the array
530 is degraded. The default is 256. Valid values are 17 to 32768.
531 Increasing this number can increase performance in some situations, at
532 some cost in system memory. Note, setting this value too high can
533 result in an "out of memory" condition for the system.
535 memory_consumed = system_page_size * nr_disks * stripe_cache_size
538 .B md/preread_bypass_threshold
539 This is only available on RAID5 and RAID6. This variable sets the
540 number of times MD will service a full-stripe-write before servicing a
541 stripe that requires some "prereading". For fairness this defaults to
542 1. Valid values are 0 to stripe_cache_size. Setting this to 0
543 maximizes sequential-write throughput at the cost of fairness to threads
544 doing small or random writes.
546 .SS KERNEL PARAMETERS
548 The md driver recognised several different kernel parameters.
551 This will disable the normal detection of md arrays that happens at
552 boot time. If a drive is partitioned with MS-DOS style partitions,
553 then if any of the 4 main partitions has a partition type of 0xFD,
554 then that partition will normally be inspected to see if it is part of
555 an MD array, and if any full arrays are found, they are started. This
556 kernel parameter disables this behaviour.
559 .B raid=partitionable
562 These are available in 2.6 and later kernels only. They indicate that
563 autodetected MD arrays should be created as partitionable arrays, with
564 a different major device number to the original non-partitionable md
565 arrays. The device number is listed as
572 This tells md to start all arrays in read-only mode. This is a soft
573 read-only that will automatically switch to read-write on the first
574 write request. However until that write request, nothing is written
575 to any device by md, and in particular, no resync or recovery
576 operation is started.
579 .B md_mod.start_dirty_degraded=1
580 As mentioned above, md will not normally start a RAID4, RAID5, or
581 RAID6 that is both dirty and degraded as this situation can imply
582 hidden data loss. This can be awkward if the root filesystem is
583 affected. Using this module parameter allows such arrays to be started
584 at boot time. It should be understood that there is a real (though
585 small) risk of data corruption in this situation.
588 .BI md= n , dev , dev ,...
590 .BI md=d n , dev , dev ,...
591 This tells the md driver to assemble
593 from the listed devices. It is only necessary to start the device
594 holding the root filesystem this way. Other arrays are best started
595 once the system is booted.
599 immediately after the
601 indicates that a partitionable device (e.g.
603 should be created rather than the original non-partitionable device.
606 .BI md= n , l , c , i , dev...
607 This tells the md driver to assemble a legacy RAID0 or LINEAR array
608 without a superblock.
610 gives the md device number,
612 gives the level, 0 for RAID0 or -1 for LINEAR,
614 gives the chunk size as a base-2 logarithm offset by twelve, so 0
615 means 4K, 1 means 8K.
617 is ignored (legacy support).
622 Contains information about the status of currently running array.
624 .B /proc/sys/dev/raid/speed_limit_min
625 A readable and writable file that reflects the current "goal" rebuild
626 speed for times when non-rebuild activity is current on an array.
627 The speed is in Kibibytes per second, and is a per-device rate, not a
628 per-array rate (which means that an array with more disks will shuffle
629 more data for a given speed). The default is 100.
632 .B /proc/sys/dev/raid/speed_limit_max
633 A readable and writable file that reflects the current "goal" rebuild
634 speed for times when no non-rebuild activity is current on an array.
635 The default is 100,000.