3 md \- Multiple Device driver aka Linux Software Raid
11 driver provides virtual devices that are created from one or more
12 independent underlying devices. This array of devices often contains
13 redundancy, and hence the acronym RAID which stands for a Redundant
14 Array of Independent Devices.
19 4 (striped array with parity device),
20 5 (striped array with distributed parity information),
21 6 (striped array with distributed dual redundancy information), and
22 10 (striped and mirrored).
23 If some number of underlying devices fails while using one of these
24 levels, the array will continue to function; this number is one for
25 RAID levels 4 and 5, two for RAID level 6, and all but one (N-1) for
26 RAID level 1, and dependant on configuration for level 10.
29 also supports a number of pseudo RAID (non-redundant) configurations
30 including RAID0 (striped array), LINEAR (catenated array),
31 MULTIPATH (a set of different interfaces to the same device),
32 and FAULTY (a layer over a single device into which errors can be injected).
35 Each device in an array may have a
37 which records information about the structure and state of the array.
38 This allows the array to be reliably re-assembled after a shutdown.
40 From Linux kernel version 2.6.10,
42 provides support for two different formats of this superblock, and
43 other formats can be added. Prior to this release, only one format is
46 The common format - known as version 0.90 - has
47 a superblock that is 4K long and is written into a 64K aligned block that
48 starts at least 64K and less than 128K from the end of the device
49 (i.e. to get the address of the superblock round the size of the
50 device down to a multiple of 64K and then subtract 64K).
51 The available size of each device is the amount of space before the
52 super block, so between 64K and 128K is lost when a device in
53 incorporated into an MD array.
54 This superblock stores multi-byte fields in a processor-dependant
55 manner, so arrays cannot easily be moved between computers with
58 The new format - known as version 1 - has a superblock that is
59 normally 1K long, but can be longer. It is normally stored between 8K
60 and 12K from the end of the device, on a 4K boundary, though
61 variations can be stored at the start of the device (version 1.1) or 4K from
62 the start of the device (version 1.2).
63 This superblock format stores multibyte data in a
64 processor-independent format and has supports up to hundreds of
65 component devices (version 0.90 only supports 28).
67 The superblock contains, among other things:
70 The manner in which the devices are arranged into the array
71 (linear, raid0, raid1, raid4, raid5, raid10, multipath).
74 a 128 bit Universally Unique Identifier that identifies the array that
75 this device is part of.
77 .SS ARRAYS WITHOUT SUPERBLOCKS
78 While it is usually best to create arrays with superblocks so that
79 they can be assembled reliably, there are some circumstances where an
80 array without superblocks in preferred. This include:
85 driver only supported Linear and Raid0 configurations and did not use
86 a superblock (which is less critical with these configurations).
87 While such arrays should be rebuilt with superblocks if possible,
89 continues to support them.
92 Being a largely transparent layer over a different device, the FAULTY
93 personality doesn't gain anything from having a superblock.
96 It is often possible to detect devices which are different paths to
97 the same storage directly rather than having a distinctive superblock
98 written to the device and searched for on all paths. In this case,
99 a MULTIPATH array with no superblock makes sense.
102 In some configurations it might be desired to create a raid1
103 configuration that does use a superblock, and to maintain the state of
104 the array elsewhere. While not encouraged for general us, it does
105 have special-purpose uses and is supported.
109 A linear array simply catenates the available space on each
110 drive together to form one large virtual drive.
112 One advantage of this arrangement over the more common RAID0
113 arrangement is that the array may be reconfigured at a later time with
114 an extra drive and so the array is made bigger without disturbing the
115 data that is on the array. However this cannot be done on a live
118 If a chunksize is given with a LINEAR array, the usable space on each
119 device is rounded down to a multiple of this chunksize.
123 A RAID0 array (which has zero redundancy) is also known as a
125 A RAID0 array is configured at creation with a
127 which must be a power of two, and at least 4 kibibytes.
129 The RAID0 driver assigns the first chunk of the array to the first
130 device, the second chunk to the second device, and so on until all
131 drives have been assigned one chunk. This collection of chunks forms
134 Further chunks are gathered into stripes in the same way which are
135 assigned to the remaining space in the drives.
137 If devices in the array are not all the same size, then once the
138 smallest device has been exhausted, the RAID0 driver starts
139 collecting chunks into smaller stripes that only span the drives which
140 still have remaining space.
145 A RAID1 array is also known as a mirrored set (though mirrors tend to
146 provide reflected images, which RAID1 does not) or a plex.
148 Once initialised, each device in a RAID1 array contains exactly the
149 same data. Changes are written to all devices in parallel. Data is
150 read from any one device. The driver attempts to distribute read
151 requests across all devices to maximise performance.
153 All devices in a RAID1 array should be the same size. If they are
154 not, then only the amount of space available on the smallest device is
155 used. Any extra space on other devices is wasted.
159 A RAID4 array is like a RAID0 array with an extra device for storing
160 parity. This device is the last of the active devices in the
161 array. Unlike RAID0, RAID4 also requires that all stripes span all
162 drives, so extra space on devices that are larger than the smallest is
165 When any block in a RAID4 array is modified the parity block for that
166 stripe (i.e. the block in the parity device at the same device offset
167 as the stripe) is also modified so that the parity block always
168 contains the "parity" for the whole stripe. i.e. its contents is
169 equivalent to the result of performing an exclusive-or operation
170 between all the data blocks in the stripe.
172 This allows the array to continue to function if one device fails.
173 The data that was on that device can be calculated as needed from the
174 parity block and the other data blocks.
178 RAID5 is very similar to RAID4. The difference is that the parity
179 blocks for each stripe, instead of being on a single device, are
180 distributed across all devices. This allows more parallelism when
181 writing as two different block updates will quite possibly affect
182 parity blocks on different devices so there is less contention.
184 This also allows more parallelism when reading as read requests are
185 distributed over all the devices in the array instead of all but one.
189 RAID6 is similar to RAID5, but can handle the loss of any \fItwo\fP
190 devices without data loss. Accordingly, it requires N+2 drives to
191 store N drives worth of data.
193 The performance for RAID6 is slightly lower but comparable to RAID5 in
194 normal mode and single disk failure mode. It is very slow in dual
195 disk failure mode, however.
199 RAID10 provides a combination of RAID1 and RAID0, and sometimes known
200 as RAID1+0. Every datablock is duplicated some number of times, and
201 the resulting collection of datablocks are distributed over multiple
204 When configuring a RAID10 array it is necessary to specify the number
205 of replicas of each data block that are required (this will normally
206 be 2) and whether the replicas should be 'near', 'offset' or 'far'.
207 (Note that the 'offset' layout is only available from 2.6.18).
209 When 'near' replicas are chosen, the multiple copies of a given chunk
210 are laid out consecutively across the stripes of the array, so the two
211 copies of a datablock will likely be at the same offset on two
214 When 'far' replicas are chosen, the multiple copies of a given chunk
215 are laid out quite distant from each other. The first copy of all
216 data blocks will be striped across the early part of all drives in
217 RAID0 fashion, and then the next copy of all blocks will be striped
218 across a later section of all drives, always ensuring that all copies
219 of any given block are on different drives.
221 The 'far' arrangement can give sequential read performance equal to
222 that of a RAID0 array, but at the cost of degraded write performance.
224 When 'offset' replicas are chosen, the multiple copies of a given
225 chunk are laid out on consecutive drives and at consecutive offsets.
226 Effectively each stripe is duplicated and the copies are offset by one
227 device. This should give similar read characteristics to 'far' if a
228 suitably large chunk size is used, but without as much seeking for
231 It should be noted that the number of devices in a RAID10 array need
232 not be a multiple of the number of replica of each data block, those
233 there must be at least as many devices as replicas.
235 If, for example, an array is created with 5 devices and 2 replicas,
236 then space equivalent to 2.5 of the devices will be available, and
237 every block will be stored on two different devices.
239 Finally, it is possible to have an array with both 'near' and 'far'
240 copies. If and array is configured with 2 near copies and 2 far
241 copies, then there will be a total of 4 copies of each block, each on
242 a different drive. This is an artifact of the implementation and is
243 unlikely to be of real value.
247 MULTIPATH is not really a RAID at all as there is only one real device
248 in a MULTIPATH md array. However there are multiple access points
249 (paths) to this device, and one of these paths might fail, so there
250 are some similarities.
252 A MULTIPATH array is composed of a number of logically different
253 devices, often fibre channel interfaces, that all refer the the same
254 real device. If one of these interfaces fails (e.g. due to cable
255 problems), the multipath driver will attempt to redirect requests to
259 The FAULTY md module is provided for testing purposes. A faulty array
260 has exactly one component device and is normally assembled without a
261 superblock, so the md array created provides direct access to all of
262 the data in the component device.
264 The FAULTY module may be requested to simulate faults to allow testing
265 of other md levels or of filesystems. Faults can be chosen to trigger
266 on read requests or write requests, and can be transient (a subsequent
267 read/write at the address will probably succeed) or persistent
268 (subsequent read/write of the same address will fail). Further, read
269 faults can be "fixable" meaning that they persist until a write
270 request at the same address.
272 Fault types can be requested with a period. In this case the fault
273 will recur repeatedly after the given number of requests of the
274 relevant type. For example if persistent read faults have a period of
275 100, then every 100th read request would generate a fault, and the
276 faulty sector would be recorded so that subsequent reads on that
277 sector would also fail.
279 There is a limit to the number of faulty sectors that are remembered.
280 Faults generated after this limit is exhausted are treated as
283 The list of faulty sectors can be flushed, and the active list of
284 failure modes can be cleared.
288 When changes are made to a RAID1, RAID4, RAID5, RAID6, or RAID10 array
289 there is a possibility of inconsistency for short periods of time as
290 each update requires are least two block to be written to different
291 devices, and these writes probably wont happen at exactly the same
292 time. Thus if a system with one of these arrays is shutdown in the
293 middle of a write operation (e.g. due to power failure), the array may
296 To handle this situation, the md driver marks an array as "dirty"
297 before writing any data to it, and marks it as "clean" when the array
298 is being disabled, e.g. at shutdown. If the md driver finds an array
299 to be dirty at startup, it proceeds to correct any possibly
300 inconsistency. For RAID1, this involves copying the contents of the
301 first drive onto all other drives. For RAID4, RAID5 and RAID6 this
302 involves recalculating the parity for each stripe and making sure that
303 the parity block has the correct data. For RAID10 it involves copying
304 one of the replicas of each block onto all the others. This process,
305 known as "resynchronising" or "resync" is performed in the background.
306 The array can still be used, though possibly with reduced performance.
308 If a RAID4, RAID5 or RAID6 array is degraded (missing at least one
309 drive) when it is restarted after an unclean shutdown, it cannot
310 recalculate parity, and so it is possible that data might be
311 undetectably corrupted. The 2.4 md driver
313 alert the operator to this condition. The 2.6 md driver will fail to
314 start an array in this condition without manual intervention, though
315 this behaviour can be over-ridden by a kernel parameter.
319 If the md driver detects a write error on a device in a RAID1, RAID4,
320 RAID5, RAID6, or RAID10 array, it immediately disables that device
321 (marking it as faulty) and continues operation on the remaining
322 devices. If there is a spare drive, the driver will start recreating
323 on one of the spare drives the data what was on that failed drive,
324 either by copying a working drive in a RAID1 configuration, or by
325 doing calculations with the parity block on RAID4, RAID5 or RAID6, or
326 by finding a copying originals for RAID10.
328 In kernels prior to about 2.6.15, a read error would cause the same
329 effect as a write error. In later kernels, a read-error will instead
330 cause md to attempt a recovery by overwriting the bad block. i.e. it
331 will find the correct data from elsewhere, write it over the block
332 that failed, and then try to read it back again. If either the write
333 or the re-read fail, md will treat the error the same way that a write
334 error is treated and will fail the whole device.
336 While this recovery process is happening, the md driver will monitor
337 accesses to the array and will slow down the rate of recovery if other
338 activity is happening, so that normal access to the array will not be
339 unduly affected. When no other activity is happening, the recovery
340 process proceeds at full speed. The actual speed targets for the two
341 different situations can be controlled by the
345 control files mentioned below.
347 .SS BITMAP WRITE-INTENT LOGGING
351 supports a bitmap based write-intent log. If configured, the bitmap
352 is used to record which blocks of the array may be out of sync.
353 Before any write request is honoured, md will make sure that the
354 corresponding bit in the log is set. After a period of time with no
355 writes to an area of the array, the corresponding bit will be cleared.
357 This bitmap is used for two optimisations.
359 Firstly, after an unclear shutdown, the resync process will consult
360 the bitmap and only resync those blocks that correspond to bits in the
361 bitmap that are set. This can dramatically increase resync time.
363 Secondly, when a drive fails and is removed from the array, md stops
364 clearing bits in the intent log. If that same drive is re-added to
365 the array, md will notice and will only recover the sections of the
366 drive that are covered by bits in the intent log that are set. This
367 can allow a device to be temporarily removed and reinserted without
368 causing an enormous recovery cost.
370 The intent log can be stored in a file on a separate device, or it can
371 be stored near the superblocks of an array which has superblocks.
373 It is possible to add an intent log or an active array, or remove an
374 intent log if one is present.
376 In 2.6.13, intent bitmaps are only supported with RAID1. Other levels
377 with redundancy are supported from 2.6.15.
383 supports WRITE-BEHIND on RAID1 arrays.
385 This allows certain devices in the array to be flagged as
387 MD will only read from such devices if there is no
390 If a write-intent bitmap is also provided, write requests to
391 write-mostly devices will be treated as write-behind requests and md
392 will not wait for writes to those requests to complete before
393 reporting the write as complete to the filesystem.
395 This allows for a RAID1 with WRITE-BEHIND to be used to mirror data
396 over a slow link to a remove computer (providing the link isn't too
397 slow). The extra latency of the remote link will not slow down normal
398 operations, but the remote system will still have a reasonably
399 up-to-date copy of all data.
406 is the processes of re-arranging the data stored in each stripe into a
407 new layout. This might involve changing the number of devices in the
408 array (so the stripes are wider) changing the chunk size (so stripes
409 are deeper or shallower), or changing the arrangement of data and
410 parity, possibly changing the raid level (e.g. 1 to 5 or 5 to 6).
412 As of Linux 2.6.17, md can reshape a raid5 array to have more
413 devices. Other possibilities may follow in future kernels.
415 During any stripe process there is a 'critical section' during which
416 live data is being over-written on disk. For the operation of
417 increasing the number of drives in a raid5, this critical section
418 covers the first few stripes (the number being the product of the old
419 and new number of devices). After this critical section is passed,
420 data is only written to areas of the array which no longer hold live
421 data - the live data has already been located away.
423 md is not able to ensure data preservation if there is a crash
424 (e.g. power failure) during the critical section. If md is asked to
425 start an array which failed during a critical section of restriping,
426 it will fail to start the array.
428 To deal with this possibility, a user-space program must
430 Disable writes to that section of the array (using the
434 Take a copy of the data somewhere (i.e. make a backup)
436 Allow the process to continue and invalidate the backup and restore
437 write access once the critical section is passed, and
439 Provide for restoring the critical data before restarting the array
440 after a system crash.
444 version 2.4 and later will do this for growing a RAID5 array.
446 For operations that do not change the size of the array, like simply
447 increasing chunk size, or converting RAID5 to RAID6 with one extra
448 device, the entire process is the critical section. In this case the
449 restripe will need to progress in stages as a section is suspended,
451 restriped, and released. This is not yet implemented.
454 All block devices appear as a directory in
458 For MD devices, this directory will contain a subdirectory called
460 which contains various files for providing access to information about
463 This interface is documented more fully in the file
464 .B Documentation/md.txt
465 which is distributed with the kernel sources. That file should be
466 consulted for full documentation. The following are just a selection
467 of attribute files that are available.
471 This value, if set, overrides the system-wide setting in
472 .B /proc/sys/dev/raid/speed_limit_min
476 to this file cause the system-wide setting to have effect.
480 This is the partner of
483 .B /proc/sys/dev/raid/spool_limit_max
488 This can be used to monitor and control the resync/recovery process of
490 In particular, writing "check" here will cause the array to read all
491 data block and check that they are consistent (e.g. parity is correct,
492 or all mirror replicas are the same). Any discrepancies found are
496 A count of problems found will be stored in
497 .BR md/mismatch_count .
499 Alternately, "repair" can be written which will cause the same check
500 to be performed, but any errors will be corrected.
502 Finally, "idle" can be written to stop the check/repair process.
505 .B md/stripe_cache_size
506 This is only available on RAID5 and RAID6. It records the size (in
507 pages per device) of the stripe cache which is used for synchronising
508 all read and write operations to the array. The default is 128.
509 Increasing this number can increase performance in some situations, at
510 some cost in system memory.
513 .SS KERNEL PARAMETERS
515 The md driver recognised several different kernel parameters.
518 This will disable the normal detection of md arrays that happens at
519 boot time. If a drive is partitioned with MS-DOS style partitions,
520 then if any of the 4 main partitions has a partition type of 0xFD,
521 then that partition will normally be inspected to see if it is part of
522 an MD array, and if any full arrays are found, they are started. This
523 kernel parameter disables this behaviour.
526 .B raid=partitionable
529 These are available in 2.6 and later kernels only. They indicate that
530 autodetected MD arrays should be created as partitionable arrays, with
531 a different major device number to the original non-partitionable md
532 arrays. The device number is listed as
539 This tells md to start all arrays in read-only mode. This is a soft
540 read-only that will automatically switch to read-write on the first
541 write request. However until that write request, nothing is written
542 to any device by md, and in particular, no resync or recovery
543 operation is started.
546 .B md_mod.start_dirty_degraded=1
547 As mentioned above, md will not normally start a RAID4, RAID5, or
548 RAID6 that is both dirty and degraded as this situation can imply
549 hidden data loss. This can be awkward if the root filesystem is
550 affected. Using the module parameter allows such arrays to be started
551 at boot time. It should be understood that there is a real (though
552 small) risk of data corruption in this situation.
555 .BI md= n , dev , dev ,...
557 .BI md=d n , dev , dev ,...
558 This tells the md driver to assemble
560 from the listed devices. It is only necessary to start the device
561 holding the root filesystem this way. Other arrays are best started
562 once the system is booted.
566 immediately after the
568 indicates that a partitionable device (e.g.
570 should be created rather than the original non-partitionable device.
573 .BI md= n , l , c , i , dev...
574 This tells the md driver to assemble a legacy RAID0 or LINEAR array
575 without a superblock.
577 gives the md device number,
579 gives the level, 0 for RAID0 or -1 for LINEAR,
581 gives the chunk size as a base-2 logarithm offset by twelve, so 0
582 means 4K, 1 means 8K.
584 is ignored (legacy support).
589 Contains information about the status of currently running array.
591 .B /proc/sys/dev/raid/speed_limit_min
592 A readable and writable file that reflects the current goal rebuild
593 speed for times when non-rebuild activity is current on an array.
594 The speed is in Kibibytes per second, and is a per-device rate, not a
595 per-array rate (which means that an array with more disc will shuffle
596 more data for a given speed). The default is 100.
599 .B /proc/sys/dev/raid/speed_limit_max
600 A readable and writable file that reflects the current goal rebuild
601 speed for times when no non-rebuild activity is current on an array.
602 The default is 100,000.