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
19 driver provides virtual devices that are created from one or more
20 independent underlying devices. This array of devices often contains
21 redundancy and the devices are often disk drives, hence the acronym RAID
22 which stands for a Redundant Array of Independent Disks.
27 4 (striped array with parity device),
28 5 (striped array with distributed parity information),
29 6 (striped array with distributed dual redundancy information), and
30 10 (striped and mirrored).
31 If some number of underlying devices fails while using one of these
32 levels, the array will continue to function; this number is one for
33 RAID levels 4 and 5, two for RAID level 6, and all but one (N-1) for
34 RAID level 1, and dependent on configuration for level 10.
37 also supports a number of pseudo RAID (non-redundant) configurations
38 including RAID0 (striped array), LINEAR (catenated array),
39 MULTIPATH (a set of different interfaces to the same device),
40 and FAULTY (a layer over a single device into which errors can be injected).
43 Each device in an array may have some
45 stored in the device. This metadata is sometimes called a
47 The metadata records information about the structure and state of the array.
48 This allows the array to be reliably re-assembled after a shutdown.
50 From Linux kernel version 2.6.10,
52 provides support for two different formats of metadata, and
53 other formats can be added. Prior to this release, only one format is
56 The common format \(em known as version 0.90 \(em has
57 a superblock that is 4K long and is written into a 64K aligned block that
58 starts at least 64K and less than 128K from the end of the device
59 (i.e. to get the address of the superblock round the size of the
60 device down to a multiple of 64K and then subtract 64K).
61 The available size of each device is the amount of space before the
62 super block, so between 64K and 128K is lost when a device in
63 incorporated into an MD array.
64 This superblock stores multi-byte fields in a processor-dependent
65 manner, so arrays cannot easily be moved between computers with
68 The new format \(em known as version 1 \(em has a superblock that is
69 normally 1K long, but can be longer. It is normally stored between 8K
70 and 12K from the end of the device, on a 4K boundary, though
71 variations can be stored at the start of the device (version 1.1) or 4K from
72 the start of the device (version 1.2).
73 This metadata format stores multibyte data in a
74 processor-independent format and supports up to hundreds of
75 component devices (version 0.90 only supports 28).
77 The metadata contains, among other things:
80 The manner in which the devices are arranged into the array
81 (linear, raid0, raid1, raid4, raid5, raid10, multipath).
84 a 128 bit Universally Unique Identifier that identifies the array that
88 When a version 0.90 array is being reshaped (e.g. adding extra devices
89 to a RAID5), the version number is temporarily set to 0.91. This
90 ensures that if the reshape process is stopped in the middle (e.g. by
91 a system crash) and the machine boots into an older kernel that does
92 not support reshaping, then the array will not be assembled (which
93 would cause data corruption) but will be left untouched until a kernel
94 that can complete the reshape processes is used.
96 .SS ARRAYS WITHOUT METADATA
97 While it is usually best to create arrays with superblocks so that
98 they can be assembled reliably, there are some circumstances when an
99 array without superblocks is preferred. These include:
102 Early versions of the
104 driver only supported Linear and Raid0 configurations and did not use
105 a superblock (which is less critical with these configurations).
106 While such arrays should be rebuilt with superblocks if possible,
108 continues to support them.
111 Being a largely transparent layer over a different device, the FAULTY
112 personality doesn't gain anything from having a superblock.
115 It is often possible to detect devices which are different paths to
116 the same storage directly rather than having a distinctive superblock
117 written to the device and searched for on all paths. In this case,
118 a MULTIPATH array with no superblock makes sense.
121 In some configurations it might be desired to create a raid1
122 configuration that does not use a superblock, and to maintain the state of
123 the array elsewhere. While not encouraged for general us, it does
124 have special-purpose uses and is supported.
126 .SS ARRAYS WITH EXTERNAL METADATA
128 From release 2.6.28, the
130 driver supports arrays with externally managed metadata. That is,
131 the metadata is not managed by the kernel by rather by a user-space
132 program which is external to the kernel. This allows support for a
133 variety of metadata formats without cluttering the kernel with lots of
137 is able to communicate with the user-space program through various
138 sysfs attributes so that it can make appropriate changes to the
139 metadata \- for example to make a device as faulty. When necessary,
141 will wait for the program to acknowledge the event by writing to a
145 contains more detail about this interaction.
148 Many metadata formats use a single block of metadata to describe a
149 number of different arrays which all use the same set of devices.
150 In this case it is helpful for the kernel to know about the full set
151 of devices as a whole. This set is known to md as a
155 array with externally managed metadata and with device offset and size
156 so that it just covers the metadata part of the devices. The
157 remainder of each device is available to be incorporated into various
162 A linear array simply catenates the available space on each
163 drive to form one large virtual drive.
165 One advantage of this arrangement over the more common RAID0
166 arrangement is that the array may be reconfigured at a later time with
167 an extra drive, so the array is made bigger without disturbing the
168 data that is on the array. This can even be done on a live
171 If a chunksize is given with a LINEAR array, the usable space on each
172 device is rounded down to a multiple of this chunksize.
176 A RAID0 array (which has zero redundancy) is also known as a
178 A RAID0 array is configured at creation with a
180 which must be a power of two (prior to Linux 2.6.31), and at least 4
183 The RAID0 driver assigns the first chunk of the array to the first
184 device, the second chunk to the second device, and so on until all
185 drives have been assigned one chunk. This collection of chunks forms a
187 Further chunks are gathered into stripes in the same way, and are
188 assigned to the remaining space in the drives.
190 If devices in the array are not all the same size, then once the
191 smallest device has been exhausted, the RAID0 driver starts
192 collecting chunks into smaller stripes that only span the drives which
193 still have remaining space.
198 A RAID1 array is also known as a mirrored set (though mirrors tend to
199 provide reflected images, which RAID1 does not) or a plex.
201 Once initialised, each device in a RAID1 array contains exactly the
202 same data. Changes are written to all devices in parallel. Data is
203 read from any one device. The driver attempts to distribute read
204 requests across all devices to maximise performance.
206 All devices in a RAID1 array should be the same size. If they are
207 not, then only the amount of space available on the smallest device is
208 used (any extra space on other devices is wasted).
210 Note that the read balancing done by the driver does not make the RAID1
211 performance profile be the same as for RAID0; a single stream of
212 sequential input will not be accelerated (e.g. a single dd), but
213 multiple sequential streams or a random workload will use more than one
214 spindle. In theory, having an N-disk RAID1 will allow N sequential
215 threads to read from all disks.
217 Individual devices in a RAID1 can be marked as "write-mostly".
218 This drives are excluded from the normal read balancing and will only
219 be read from when there is no other option. This can be useful for
220 devices connected over a slow link.
224 A RAID4 array is like a RAID0 array with an extra device for storing
225 parity. This device is the last of the active devices in the
226 array. Unlike RAID0, RAID4 also requires that all stripes span all
227 drives, so extra space on devices that are larger than the smallest is
230 When any block in a RAID4 array is modified, the parity block for that
231 stripe (i.e. the block in the parity device at the same device offset
232 as the stripe) is also modified so that the parity block always
233 contains the "parity" for the whole stripe. I.e. its content is
234 equivalent to the result of performing an exclusive-or operation
235 between all the data blocks in the stripe.
237 This allows the array to continue to function if one device fails.
238 The data that was on that device can be calculated as needed from the
239 parity block and the other data blocks.
243 RAID5 is very similar to RAID4. The difference is that the parity
244 blocks for each stripe, instead of being on a single device, are
245 distributed across all devices. This allows more parallelism when
246 writing, as two different block updates will quite possibly affect
247 parity blocks on different devices so there is less contention.
249 This also allows more parallelism when reading, as read requests are
250 distributed over all the devices in the array instead of all but one.
254 RAID6 is similar to RAID5, but can handle the loss of any \fItwo\fP
255 devices without data loss. Accordingly, it requires N+2 drives to
256 store N drives worth of data.
258 The performance for RAID6 is slightly lower but comparable to RAID5 in
259 normal mode and single disk failure mode. It is very slow in dual
260 disk failure mode, however.
264 RAID10 provides a combination of RAID1 and RAID0, and is sometimes known
265 as RAID1+0. Every datablock is duplicated some number of times, and
266 the resulting collection of datablocks are distributed over multiple
269 When configuring a RAID10 array, it is necessary to specify the number
270 of replicas of each data block that are required (this will normally
271 be 2) and whether the replicas should be 'near', 'offset' or 'far'.
272 (Note that the 'offset' layout is only available from 2.6.18).
274 When 'near' replicas are chosen, the multiple copies of a given chunk
275 are laid out consecutively across the stripes of the array, so the two
276 copies of a datablock will likely be at the same offset on two
279 When 'far' replicas are chosen, the multiple copies of a given chunk
280 are laid out quite distant from each other. The first copy of all
281 data blocks will be striped across the early part of all drives in
282 RAID0 fashion, and then the next copy of all blocks will be striped
283 across a later section of all drives, always ensuring that all copies
284 of any given block are on different drives.
286 The 'far' arrangement can give sequential read performance equal to
287 that of a RAID0 array, but at the cost of reduced write performance.
289 When 'offset' replicas are chosen, the multiple copies of a given
290 chunk are laid out on consecutive drives and at consecutive offsets.
291 Effectively each stripe is duplicated and the copies are offset by one
292 device. This should give similar read characteristics to 'far' if a
293 suitably large chunk size is used, but without as much seeking for
296 It should be noted that the number of devices in a RAID10 array need
297 not be a multiple of the number of replica of each data block; however,
298 there must be at least as many devices as replicas.
300 If, for example, an array is created with 5 devices and 2 replicas,
301 then space equivalent to 2.5 of the devices will be available, and
302 every block will be stored on two different devices.
304 Finally, it is possible to have an array with both 'near' and 'far'
305 copies. If an array is configured with 2 near copies and 2 far
306 copies, then there will be a total of 4 copies of each block, each on
307 a different drive. This is an artifact of the implementation and is
308 unlikely to be of real value.
312 MULTIPATH is not really a RAID at all as there is only one real device
313 in a MULTIPATH md array. However there are multiple access points
314 (paths) to this device, and one of these paths might fail, so there
315 are some similarities.
317 A MULTIPATH array is composed of a number of logically different
318 devices, often fibre channel interfaces, that all refer the the same
319 real device. If one of these interfaces fails (e.g. due to cable
320 problems), the multipath driver will attempt to redirect requests to
323 The MULTIPATH drive is not receiving any ongoing development and
324 should be considered a legacy driver. The device-mapper based
325 multipath drivers should be preferred for new installations.
328 The FAULTY md module is provided for testing purposes. A faulty array
329 has exactly one component device and is normally assembled without a
330 superblock, so the md array created provides direct access to all of
331 the data in the component device.
333 The FAULTY module may be requested to simulate faults to allow testing
334 of other md levels or of filesystems. Faults can be chosen to trigger
335 on read requests or write requests, and can be transient (a subsequent
336 read/write at the address will probably succeed) or persistent
337 (subsequent read/write of the same address will fail). Further, read
338 faults can be "fixable" meaning that they persist until a write
339 request at the same address.
341 Fault types can be requested with a period. In this case, the fault
342 will recur repeatedly after the given number of requests of the
343 relevant type. For example if persistent read faults have a period of
344 100, then every 100th read request would generate a fault, and the
345 faulty sector would be recorded so that subsequent reads on that
346 sector would also fail.
348 There is a limit to the number of faulty sectors that are remembered.
349 Faults generated after this limit is exhausted are treated as
352 The list of faulty sectors can be flushed, and the active list of
353 failure modes can be cleared.
357 When changes are made to a RAID1, RAID4, RAID5, RAID6, or RAID10 array
358 there is a possibility of inconsistency for short periods of time as
359 each update requires at least two block to be written to different
360 devices, and these writes probably won't happen at exactly the same
361 time. Thus if a system with one of these arrays is shutdown in the
362 middle of a write operation (e.g. due to power failure), the array may
365 To handle this situation, the md driver marks an array as "dirty"
366 before writing any data to it, and marks it as "clean" when the array
367 is being disabled, e.g. at shutdown. If the md driver finds an array
368 to be dirty at startup, it proceeds to correct any possibly
369 inconsistency. For RAID1, this involves copying the contents of the
370 first drive onto all other drives. For RAID4, RAID5 and RAID6 this
371 involves recalculating the parity for each stripe and making sure that
372 the parity block has the correct data. For RAID10 it involves copying
373 one of the replicas of each block onto all the others. This process,
374 known as "resynchronising" or "resync" is performed in the background.
375 The array can still be used, though possibly with reduced performance.
377 If a RAID4, RAID5 or RAID6 array is degraded (missing at least one
378 drive, two for RAID6) when it is restarted after an unclean shutdown, it cannot
379 recalculate parity, and so it is possible that data might be
380 undetectably corrupted. The 2.4 md driver
382 alert the operator to this condition. The 2.6 md driver will fail to
383 start an array in this condition without manual intervention, though
384 this behaviour can be overridden by a kernel parameter.
388 If the md driver detects a write error on a device in a RAID1, RAID4,
389 RAID5, RAID6, or RAID10 array, it immediately disables that device
390 (marking it as faulty) and continues operation on the remaining
391 devices. If there are spare drives, the driver will start recreating
392 on one of the spare drives the data which was on that failed drive,
393 either by copying a working drive in a RAID1 configuration, or by
394 doing calculations with the parity block on RAID4, RAID5 or RAID6, or
395 by finding and copying originals for RAID10.
397 In kernels prior to about 2.6.15, a read error would cause the same
398 effect as a write error. In later kernels, a read-error will instead
399 cause md to attempt a recovery by overwriting the bad block. i.e. it
400 will find the correct data from elsewhere, write it over the block
401 that failed, and then try to read it back again. If either the write
402 or the re-read fail, md will treat the error the same way that a write
403 error is treated, and will fail the whole device.
405 While this recovery process is happening, the md driver will monitor
406 accesses to the array and will slow down the rate of recovery if other
407 activity is happening, so that normal access to the array will not be
408 unduly affected. When no other activity is happening, the recovery
409 process proceeds at full speed. The actual speed targets for the two
410 different situations can be controlled by the
414 control files mentioned below.
416 .SS BITMAP WRITE-INTENT LOGGING
420 supports a bitmap based write-intent log. If configured, the bitmap
421 is used to record which blocks of the array may be out of sync.
422 Before any write request is honoured, md will make sure that the
423 corresponding bit in the log is set. After a period of time with no
424 writes to an area of the array, the corresponding bit will be cleared.
426 This bitmap is used for two optimisations.
428 Firstly, after an unclean shutdown, the resync process will consult
429 the bitmap and only resync those blocks that correspond to bits in the
430 bitmap that are set. This can dramatically reduce resync time.
432 Secondly, when a drive fails and is removed from the array, md stops
433 clearing bits in the intent log. If that same drive is re-added to
434 the array, md will notice and will only recover the sections of the
435 drive that are covered by bits in the intent log that are set. This
436 can allow a device to be temporarily removed and reinserted without
437 causing an enormous recovery cost.
439 The intent log can be stored in a file on a separate device, or it can
440 be stored near the superblocks of an array which has superblocks.
442 It is possible to add an intent log to an active array, or remove an
443 intent log if one is present.
445 In 2.6.13, intent bitmaps are only supported with RAID1. Other levels
446 with redundancy are supported from 2.6.15.
452 supports WRITE-BEHIND on RAID1 arrays.
454 This allows certain devices in the array to be flagged as
456 MD will only read from such devices if there is no
459 If a write-intent bitmap is also provided, write requests to
460 write-mostly devices will be treated as write-behind requests and md
461 will not wait for writes to those requests to complete before
462 reporting the write as complete to the filesystem.
464 This allows for a RAID1 with WRITE-BEHIND to be used to mirror data
465 over a slow link to a remote computer (providing the link isn't too
466 slow). The extra latency of the remote link will not slow down normal
467 operations, but the remote system will still have a reasonably
468 up-to-date copy of all data.
475 is the processes of re-arranging the data stored in each stripe into a
476 new layout. This might involve changing the number of devices in the
477 array (so the stripes are wider), changing the chunk size (so stripes
478 are deeper or shallower), or changing the arrangement of data and
479 parity (possibly changing the raid level, e.g. 1 to 5 or 5 to 6).
481 As of Linux 2.6.17, md can reshape a raid5 array to have more
482 devices. Other possibilities may follow in future kernels.
484 During any stripe process there is a 'critical section' during which
485 live data is being overwritten on disk. For the operation of
486 increasing the number of drives in a raid5, this critical section
487 covers the first few stripes (the number being the product of the old
488 and new number of devices). After this critical section is passed,
489 data is only written to areas of the array which no longer hold live
490 data \(em the live data has already been located away.
492 md is not able to ensure data preservation if there is a crash
493 (e.g. power failure) during the critical section. If md is asked to
494 start an array which failed during a critical section of restriping,
495 it will fail to start the array.
497 To deal with this possibility, a user-space program must
499 Disable writes to that section of the array (using the
503 take a copy of the data somewhere (i.e. make a backup),
505 allow the process to continue and invalidate the backup and restore
506 write access once the critical section is passed, and
508 provide for restoring the critical data before restarting the array
509 after a system crash.
513 versions from 2.4 do this for growing a RAID5 array.
515 For operations that do not change the size of the array, like simply
516 increasing chunk size, or converting RAID5 to RAID6 with one extra
517 device, the entire process is the critical section. In this case, the
518 restripe will need to progress in stages, as a section is suspended,
520 restriped, and released; this is not yet implemented.
523 Each block device appears as a directory in
525 (which is usually mounted at
527 For MD devices, this directory will contain a subdirectory called
529 which contains various files for providing access to information about
532 This interface is documented more fully in the file
533 .B Documentation/md.txt
534 which is distributed with the kernel sources. That file should be
535 consulted for full documentation. The following are just a selection
536 of attribute files that are available.
540 This value, if set, overrides the system-wide setting in
541 .B /proc/sys/dev/raid/speed_limit_min
545 to this file will cause the system-wide setting to have effect.
549 This is the partner of
552 .B /proc/sys/dev/raid/spool_limit_max
557 This can be used to monitor and control the resync/recovery process of
559 In particular, writing "check" here will cause the array to read all
560 data block and check that they are consistent (e.g. parity is correct,
561 or all mirror replicas are the same). Any discrepancies found are
565 A count of problems found will be stored in
566 .BR md/mismatch_count .
568 Alternately, "repair" can be written which will cause the same check
569 to be performed, but any errors will be corrected.
571 Finally, "idle" can be written to stop the check/repair process.
574 .B md/stripe_cache_size
575 This is only available on RAID5 and RAID6. It records the size (in
576 pages per device) of the stripe cache which is used for synchronising
577 all write operations to the array and all read operations if the array
578 is degraded. The default is 256. Valid values are 17 to 32768.
579 Increasing this number can increase performance in some situations, at
580 some cost in system memory. Note, setting this value too high can
581 result in an "out of memory" condition for the system.
583 memory_consumed = system_page_size * nr_disks * stripe_cache_size
586 .B md/preread_bypass_threshold
587 This is only available on RAID5 and RAID6. This variable sets the
588 number of times MD will service a full-stripe-write before servicing a
589 stripe that requires some "prereading". For fairness this defaults to
590 1. Valid values are 0 to stripe_cache_size. Setting this to 0
591 maximizes sequential-write throughput at the cost of fairness to threads
592 doing small or random writes.
594 .SS KERNEL PARAMETERS
596 The md driver recognised several different kernel parameters.
599 This will disable the normal detection of md arrays that happens at
600 boot time. If a drive is partitioned with MS-DOS style partitions,
601 then if any of the 4 main partitions has a partition type of 0xFD,
602 then that partition will normally be inspected to see if it is part of
603 an MD array, and if any full arrays are found, they are started. This
604 kernel parameter disables this behaviour.
607 .B raid=partitionable
610 These are available in 2.6 and later kernels only. They indicate that
611 autodetected MD arrays should be created as partitionable arrays, with
612 a different major device number to the original non-partitionable md
613 arrays. The device number is listed as
621 .B /sys/module/md_mod/parameters/start_ro
622 This tells md to start all arrays in read-only mode. This is a soft
623 read-only that will automatically switch to read-write on the first
624 write request. However until that write request, nothing is written
625 to any device by md, and in particular, no resync or recovery
626 operation is started.
629 .B md_mod.start_dirty_degraded=1
631 .B /sys/module/md_mod/parameters/start_dirty_degraded
632 As mentioned above, md will not normally start a RAID4, RAID5, or
633 RAID6 that is both dirty and degraded as this situation can imply
634 hidden data loss. This can be awkward if the root filesystem is
635 affected. Using this module parameter allows such arrays to be started
636 at boot time. It should be understood that there is a real (though
637 small) risk of data corruption in this situation.
640 .BI md= n , dev , dev ,...
642 .BI md=d n , dev , dev ,...
643 This tells the md driver to assemble
645 from the listed devices. It is only necessary to start the device
646 holding the root filesystem this way. Other arrays are best started
647 once the system is booted.
651 immediately after the
653 indicates that a partitionable device (e.g.
655 should be created rather than the original non-partitionable device.
658 .BI md= n , l , c , i , dev...
659 This tells the md driver to assemble a legacy RAID0 or LINEAR array
660 without a superblock.
662 gives the md device number,
664 gives the level, 0 for RAID0 or -1 for LINEAR,
666 gives the chunk size as a base-2 logarithm offset by twelve, so 0
667 means 4K, 1 means 8K.
669 is ignored (legacy support).
674 Contains information about the status of currently running array.
676 .B /proc/sys/dev/raid/speed_limit_min
677 A readable and writable file that reflects the current "goal" rebuild
678 speed for times when non-rebuild activity is current on an array.
679 The speed is in Kibibytes per second, and is a per-device rate, not a
680 per-array rate (which means that an array with more disks will shuffle
681 more data for a given speed). The default is 1000.
684 .B /proc/sys/dev/raid/speed_limit_max
685 A readable and writable file that reflects the current "goal" rebuild
686 speed for times when no non-rebuild activity is current on an array.
687 The default is 200,000.