]> git.ipfire.org Git - thirdparty/mdadm.git/blame - md.4
Assemble(): don't dup_super() before we need it.
[thirdparty/mdadm.git] / md.4
CommitLineData
519561f7
NB
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.
56eb10c0
NB
7.TH MD 4
8.SH NAME
93e790af 9md \- Multiple Device driver aka Linux Software RAID
56eb10c0
NB
10.SH SYNOPSIS
11.BI /dev/md n
12.br
13.BI /dev/md/ n
e0fe762a
N
14.br
15.BR /dev/md/ name
56eb10c0
NB
16.SH DESCRIPTION
17The
18.B md
19driver provides virtual devices that are created from one or more
e0d19036 20independent underlying devices. This array of devices often contains
02b76eea
NB
21redundancy and the devices are often disk drives, hence the acronym RAID
22which stands for a Redundant Array of Independent Disks.
56eb10c0
NB
23.PP
24.B md
599e5a36
NB
25supports RAID levels
261 (mirroring),
274 (striped array with parity device),
285 (striped array with distributed parity information),
296 (striped array with distributed dual redundancy information), and
3010 (striped and mirrored).
31If some number of underlying devices fails while using one of these
98c6faba
NB
32levels, the array will continue to function; this number is one for
33RAID levels 4 and 5, two for RAID level 6, and all but one (N-1) for
93e790af 34RAID level 1, and dependent on configuration for level 10.
56eb10c0
NB
35.PP
36.B md
e0d19036 37also supports a number of pseudo RAID (non-redundant) configurations
570c0542
NB
38including RAID0 (striped array), LINEAR (catenated array),
39MULTIPATH (a set of different interfaces to the same device),
40and FAULTY (a layer over a single device into which errors can be injected).
56eb10c0 41
e0fe762a
N
42.SS MD METADATA
43Each device in an array may have some
44.I metadata
45stored in the device. This metadata is sometimes called a
46.BR superblock .
47The metadata records information about the structure and state of the array.
570c0542 48This allows the array to be reliably re-assembled after a shutdown.
56eb10c0 49
570c0542
NB
50From Linux kernel version 2.6.10,
51.B md
e0fe762a 52provides support for two different formats of metadata, and
570c0542
NB
53other formats can be added. Prior to this release, only one format is
54supported.
55
b3f1c093 56The common format \(em known as version 0.90 \(em has
570c0542 57a superblock that is 4K long and is written into a 64K aligned block that
11a3e71d 58starts at least 64K and less than 128K from the end of the device
56eb10c0
NB
59(i.e. to get the address of the superblock round the size of the
60device down to a multiple of 64K and then subtract 64K).
11a3e71d 61The available size of each device is the amount of space before the
56eb10c0
NB
62super block, so between 64K and 128K is lost when a device in
63incorporated into an MD array.
93e790af 64This superblock stores multi-byte fields in a processor-dependent
570c0542
NB
65manner, so arrays cannot easily be moved between computers with
66different processors.
67
b3f1c093 68The new format \(em known as version 1 \(em has a superblock that is
570c0542
NB
69normally 1K long, but can be longer. It is normally stored between 8K
70and 12K from the end of the device, on a 4K boundary, though
71variations can be stored at the start of the device (version 1.1) or 4K from
72the start of the device (version 1.2).
e0fe762a 73This metadata format stores multibyte data in a
93e790af 74processor-independent format and supports up to hundreds of
570c0542 75component devices (version 0.90 only supports 28).
56eb10c0 76
e0fe762a 77The metadata contains, among other things:
56eb10c0
NB
78.TP
79LEVEL
11a3e71d 80The manner in which the devices are arranged into the array
599e5a36 81(linear, raid0, raid1, raid4, raid5, raid10, multipath).
56eb10c0
NB
82.TP
83UUID
84a 128 bit Universally Unique Identifier that identifies the array that
93e790af 85contains this device.
56eb10c0 86
e0fe762a 87.PP
2a940e36
NB
88When a version 0.90 array is being reshaped (e.g. adding extra devices
89to a RAID5), the version number is temporarily set to 0.91. This
90ensures that if the reshape process is stopped in the middle (e.g. by
91a system crash) and the machine boots into an older kernel that does
92not support reshaping, then the array will not be assembled (which
93would cause data corruption) but will be left untouched until a kernel
94that can complete the reshape processes is used.
95
e0fe762a 96.SS ARRAYS WITHOUT METADATA
570c0542 97While it is usually best to create arrays with superblocks so that
93e790af
SW
98they can be assembled reliably, there are some circumstances when an
99array without superblocks is preferred. These include:
570c0542
NB
100.TP
101LEGACY ARRAYS
11a3e71d
NB
102Early versions of the
103.B md
570c0542
NB
104driver only supported Linear and Raid0 configurations and did not use
105a superblock (which is less critical with these configurations).
106While such arrays should be rebuilt with superblocks if possible,
11a3e71d 107.B md
570c0542
NB
108continues to support them.
109.TP
110FAULTY
111Being a largely transparent layer over a different device, the FAULTY
112personality doesn't gain anything from having a superblock.
113.TP
114MULTIPATH
115It is often possible to detect devices which are different paths to
116the same storage directly rather than having a distinctive superblock
117written to the device and searched for on all paths. In this case,
118a MULTIPATH array with no superblock makes sense.
119.TP
120RAID1
121In some configurations it might be desired to create a raid1
93e790af 122configuration that does not use a superblock, and to maintain the state of
095407fa 123the array elsewhere. While not encouraged for general use, it does
addc80c4 124have special-purpose uses and is supported.
11a3e71d 125
e0fe762a
N
126.SS ARRAYS WITH EXTERNAL METADATA
127
128From release 2.6.28, the
129.I md
130driver supports arrays with externally managed metadata. That is,
1e49aaa0 131the metadata is not managed by the kernel but rather by a user-space
e0fe762a
N
132program which is external to the kernel. This allows support for a
133variety of metadata formats without cluttering the kernel with lots of
134details.
135.PP
136.I md
137is able to communicate with the user-space program through various
138sysfs attributes so that it can make appropriate changes to the
1b17b4e4 139metadata \- for example to mark a device as faulty. When necessary,
e0fe762a
N
140.I md
141will wait for the program to acknowledge the event by writing to a
142sysfs attribute.
143The manual page for
144.IR mdmon (8)
145contains more detail about this interaction.
146
147.SS CONTAINERS
148Many metadata formats use a single block of metadata to describe a
149number of different arrays which all use the same set of devices.
150In this case it is helpful for the kernel to know about the full set
151of devices as a whole. This set is known to md as a
152.IR container .
153A container is an
154.I md
155array with externally managed metadata and with device offset and size
156so that it just covers the metadata part of the devices. The
157remainder of each device is available to be incorporated into various
158arrays.
159
56eb10c0 160.SS LINEAR
11a3e71d
NB
161
162A linear array simply catenates the available space on each
93e790af 163drive to form one large virtual drive.
11a3e71d
NB
164
165One advantage of this arrangement over the more common RAID0
166arrangement is that the array may be reconfigured at a later time with
93e790af
SW
167an extra drive, so the array is made bigger without disturbing the
168data that is on the array. This can even be done on a live
11a3e71d
NB
169array.
170
599e5a36
NB
171If a chunksize is given with a LINEAR array, the usable space on each
172device is rounded down to a multiple of this chunksize.
11a3e71d 173
56eb10c0 174.SS RAID0
11a3e71d
NB
175
176A RAID0 array (which has zero redundancy) is also known as a
177striped array.
e0d19036
NB
178A RAID0 array is configured at creation with a
179.B "Chunk Size"
e0fe762a
N
180which must be a power of two (prior to Linux 2.6.31), and at least 4
181kibibytes.
e0d19036 182
2d465520 183The RAID0 driver assigns the first chunk of the array to the first
e0d19036 184device, the second chunk to the second device, and so on until all
e0fe762a 185drives have been assigned one chunk. This collection of chunks forms a
e0d19036 186.BR stripe .
93e790af 187Further chunks are gathered into stripes in the same way, and are
e0d19036
NB
188assigned to the remaining space in the drives.
189
2d465520
NB
190If devices in the array are not all the same size, then once the
191smallest device has been exhausted, the RAID0 driver starts
e0d19036
NB
192collecting chunks into smaller stripes that only span the drives which
193still have remaining space.
194
195
56eb10c0 196.SS RAID1
e0d19036
NB
197
198A RAID1 array is also known as a mirrored set (though mirrors tend to
5787fa49 199provide reflected images, which RAID1 does not) or a plex.
e0d19036
NB
200
201Once initialised, each device in a RAID1 array contains exactly the
202same data. Changes are written to all devices in parallel. Data is
203read from any one device. The driver attempts to distribute read
204requests across all devices to maximise performance.
205
206All devices in a RAID1 array should be the same size. If they are
207not, then only the amount of space available on the smallest device is
93e790af 208used (any extra space on other devices is wasted).
e0d19036 209
3dacb890
IP
210Note that the read balancing done by the driver does not make the RAID1
211performance profile be the same as for RAID0; a single stream of
212sequential input will not be accelerated (e.g. a single dd), but
213multiple sequential streams or a random workload will use more than one
214spindle. In theory, having an N-disk RAID1 will allow N sequential
215threads to read from all disks.
216
e0fe762a 217Individual devices in a RAID1 can be marked as "write-mostly".
1b17b4e4 218These drives are excluded from the normal read balancing and will only
e0fe762a
N
219be read from when there is no other option. This can be useful for
220devices connected over a slow link.
221
56eb10c0 222.SS RAID4
e0d19036
NB
223
224A RAID4 array is like a RAID0 array with an extra device for storing
aa88f531
NB
225parity. This device is the last of the active devices in the
226array. Unlike RAID0, RAID4 also requires that all stripes span all
e0d19036
NB
227drives, so extra space on devices that are larger than the smallest is
228wasted.
229
93e790af 230When any block in a RAID4 array is modified, the parity block for that
e0d19036
NB
231stripe (i.e. the block in the parity device at the same device offset
232as the stripe) is also modified so that the parity block always
93e790af 233contains the "parity" for the whole stripe. I.e. its content is
e0d19036
NB
234equivalent to the result of performing an exclusive-or operation
235between all the data blocks in the stripe.
236
237This allows the array to continue to function if one device fails.
238The data that was on that device can be calculated as needed from the
239parity block and the other data blocks.
240
56eb10c0 241.SS RAID5
e0d19036
NB
242
243RAID5 is very similar to RAID4. The difference is that the parity
244blocks for each stripe, instead of being on a single device, are
245distributed across all devices. This allows more parallelism when
93e790af 246writing, as two different block updates will quite possibly affect
e0d19036
NB
247parity blocks on different devices so there is less contention.
248
93e790af 249This also allows more parallelism when reading, as read requests are
e0d19036
NB
250distributed over all the devices in the array instead of all but one.
251
98c6faba
NB
252.SS RAID6
253
254RAID6 is similar to RAID5, but can handle the loss of any \fItwo\fP
255devices without data loss. Accordingly, it requires N+2 drives to
256store N drives worth of data.
257
258The performance for RAID6 is slightly lower but comparable to RAID5 in
259normal mode and single disk failure mode. It is very slow in dual
260disk failure mode, however.
261
599e5a36
NB
262.SS RAID10
263
93e790af 264RAID10 provides a combination of RAID1 and RAID0, and is sometimes known
599e5a36
NB
265as RAID1+0. Every datablock is duplicated some number of times, and
266the resulting collection of datablocks are distributed over multiple
267drives.
268
93e790af 269When configuring a RAID10 array, it is necessary to specify the number
599e5a36 270of replicas of each data block that are required (this will normally
b578481c
NB
271be 2) and whether the replicas should be 'near', 'offset' or 'far'.
272(Note that the 'offset' layout is only available from 2.6.18).
599e5a36
NB
273
274When 'near' replicas are chosen, the multiple copies of a given chunk
275are laid out consecutively across the stripes of the array, so the two
276copies of a datablock will likely be at the same offset on two
277adjacent devices.
278
279When 'far' replicas are chosen, the multiple copies of a given chunk
280are laid out quite distant from each other. The first copy of all
281data blocks will be striped across the early part of all drives in
282RAID0 fashion, and then the next copy of all blocks will be striped
283across a later section of all drives, always ensuring that all copies
284of any given block are on different drives.
285
286The 'far' arrangement can give sequential read performance equal to
91c00388 287that of a RAID0 array, but at the cost of reduced write performance.
599e5a36 288
b578481c
NB
289When 'offset' replicas are chosen, the multiple copies of a given
290chunk are laid out on consecutive drives and at consecutive offsets.
291Effectively each stripe is duplicated and the copies are offset by one
292device. This should give similar read characteristics to 'far' if a
293suitably large chunk size is used, but without as much seeking for
294writes.
295
599e5a36 296It should be noted that the number of devices in a RAID10 array need
93e790af 297not be a multiple of the number of replica of each data block; however,
599e5a36
NB
298there must be at least as many devices as replicas.
299
300If, for example, an array is created with 5 devices and 2 replicas,
301then space equivalent to 2.5 of the devices will be available, and
302every block will be stored on two different devices.
303
304Finally, it is possible to have an array with both 'near' and 'far'
93e790af 305copies. If an array is configured with 2 near copies and 2 far
599e5a36
NB
306copies, then there will be a total of 4 copies of each block, each on
307a different drive. This is an artifact of the implementation and is
308unlikely to be of real value.
309
bf40ab85 310.SS MULTIPATH
e0d19036
NB
311
312MULTIPATH is not really a RAID at all as there is only one real device
313in a MULTIPATH md array. However there are multiple access points
314(paths) to this device, and one of these paths might fail, so there
315are some similarities.
316
a9d69660 317A MULTIPATH array is composed of a number of logically different
2d465520
NB
318devices, often fibre channel interfaces, that all refer the the same
319real device. If one of these interfaces fails (e.g. due to cable
a9d69660 320problems), the multipath driver will attempt to redirect requests to
e0fe762a
N
321another interface.
322
323The MULTIPATH drive is not receiving any ongoing development and
324should be considered a legacy driver. The device-mapper based
325multipath drivers should be preferred for new installations.
e0d19036 326
b5e64645
NB
327.SS FAULTY
328The FAULTY md module is provided for testing purposes. A faulty array
329has exactly one component device and is normally assembled without a
330superblock, so the md array created provides direct access to all of
331the data in the component device.
332
333The FAULTY module may be requested to simulate faults to allow testing
a9d69660 334of other md levels or of filesystems. Faults can be chosen to trigger
b5e64645 335on read requests or write requests, and can be transient (a subsequent
addc80c4 336read/write at the address will probably succeed) or persistent
b5e64645
NB
337(subsequent read/write of the same address will fail). Further, read
338faults can be "fixable" meaning that they persist until a write
339request at the same address.
340
93e790af 341Fault types can be requested with a period. In this case, the fault
a9d69660
NB
342will recur repeatedly after the given number of requests of the
343relevant type. For example if persistent read faults have a period of
344100, then every 100th read request would generate a fault, and the
b5e64645
NB
345faulty sector would be recorded so that subsequent reads on that
346sector would also fail.
347
348There is a limit to the number of faulty sectors that are remembered.
349Faults generated after this limit is exhausted are treated as
350transient.
351
a9d69660 352The list of faulty sectors can be flushed, and the active list of
b5e64645 353failure modes can be cleared.
e0d19036
NB
354
355.SS UNCLEAN SHUTDOWN
356
599e5a36
NB
357When changes are made to a RAID1, RAID4, RAID5, RAID6, or RAID10 array
358there is a possibility of inconsistency for short periods of time as
93e790af
SW
359each update requires at least two block to be written to different
360devices, and these writes probably won't happen at exactly the same
599e5a36
NB
361time. Thus if a system with one of these arrays is shutdown in the
362middle of a write operation (e.g. due to power failure), the array may
363not be consistent.
e0d19036 364
2d465520 365To handle this situation, the md driver marks an array as "dirty"
e0d19036 366before writing any data to it, and marks it as "clean" when the array
98c6faba
NB
367is being disabled, e.g. at shutdown. If the md driver finds an array
368to be dirty at startup, it proceeds to correct any possibly
369inconsistency. For RAID1, this involves copying the contents of the
370first drive onto all other drives. For RAID4, RAID5 and RAID6 this
371involves recalculating the parity for each stripe and making sure that
599e5a36
NB
372the parity block has the correct data. For RAID10 it involves copying
373one of the replicas of each block onto all the others. This process,
374known as "resynchronising" or "resync" is performed in the background.
375The array can still be used, though possibly with reduced performance.
98c6faba
NB
376
377If a RAID4, RAID5 or RAID6 array is degraded (missing at least one
93e790af 378drive, two for RAID6) when it is restarted after an unclean shutdown, it cannot
98c6faba
NB
379recalculate parity, and so it is possible that data might be
380undetectably corrupted. The 2.4 md driver
e0d19036 381.B does not
addc80c4
NB
382alert the operator to this condition. The 2.6 md driver will fail to
383start an array in this condition without manual intervention, though
35cc5be4 384this behaviour can be overridden by a kernel parameter.
e0d19036
NB
385
386.SS RECOVERY
387
addc80c4 388If the md driver detects a write error on a device in a RAID1, RAID4,
599e5a36
NB
389RAID5, RAID6, or RAID10 array, it immediately disables that device
390(marking it as faulty) and continues operation on the remaining
93e790af
SW
391devices. If there are spare drives, the driver will start recreating
392on one of the spare drives the data which was on that failed drive,
599e5a36
NB
393either by copying a working drive in a RAID1 configuration, or by
394doing calculations with the parity block on RAID4, RAID5 or RAID6, or
93e790af 395by finding and copying originals for RAID10.
e0d19036 396
addc80c4
NB
397In kernels prior to about 2.6.15, a read error would cause the same
398effect as a write error. In later kernels, a read-error will instead
399cause md to attempt a recovery by overwriting the bad block. i.e. it
400will find the correct data from elsewhere, write it over the block
401that failed, and then try to read it back again. If either the write
402or the re-read fail, md will treat the error the same way that a write
93e790af 403error is treated, and will fail the whole device.
addc80c4 404
2d465520 405While this recovery process is happening, the md driver will monitor
e0d19036
NB
406accesses to the array and will slow down the rate of recovery if other
407activity is happening, so that normal access to the array will not be
408unduly affected. When no other activity is happening, the recovery
409process proceeds at full speed. The actual speed targets for the two
410different situations can be controlled by the
411.B speed_limit_min
412and
413.B speed_limit_max
414control files mentioned below.
415
1cc44574
N
416.SS SCRUBBING AND MISMATCHES
417
418As storage devices can develop bad blocks at any time it is valuable
419to regularly read all blocks on all devices in an array so as to catch
420such bad blocks early. This process is called
421.IR scrubbing .
422
423md arrays can be scrubbed by writing either
424.I check
425or
426.I repair
427to the file
428.I md/sync_action
429in the
430.I sysfs
431directory for the device.
432
c93e9d68 433Requesting a scrub will cause
1cc44574
N
434.I md
435to read every block on every device in the array, and check that the
c93e9d68
N
436data is consistent. For RAID1 and RAID10, this means checking that the copies
437are identical. For RAID4, RAID5, RAID6 this means checking that the
438parity block is (or blocks are) correct.
1cc44574
N
439
440If a read error is detected during this process, the normal read-error
441handling causes correct data to be found from other devices and to be
442written back to the faulty device. In many case this will
443effectively
444.I fix
445the bad block.
446
447If all blocks read successfully but are found to not be consistent,
448then this is regarded as a
449.IR mismatch .
450
451If
452.I check
453was used, then no action is taken to handle the mismatch, it is simply
454recorded.
455If
456.I repair
457was used, then a mismatch will be repaired in the same way that
458.I resync
c93e9d68 459repairs arrays. For RAID5/RAID6 new parity blocks are written. For RAID1/RAID10,
1cc44574
N
460all but one block are overwritten with the content of that one block.
461
462A count of mismatches is recorded in the
463.I sysfs
464file
465.IR md/mismatch_cnt .
466This is set to zero when a
c93e9d68 467scrub starts and is incremented whenever a sector is
1cc44574
N
468found that is a mismatch.
469.I md
470normally works in units much larger than a single sector and when it
1e49aaa0 471finds a mismatch, it does not determine exactly how many actual sectors were
c93e9d68
N
472affected but simply adds the number of sectors in the IO unit that was
473used. So a value of 128 could simply mean that a single 64KB check
474found an error (128 x 512bytes = 64KB).
475
476If an array is created by
477.I mdadm
478with
1cc44574
N
479.I \-\-assume\-clean
480then a subsequent check could be expected to find some mismatches.
481
482On a truly clean RAID5 or RAID6 array, any mismatches should indicate
483a hardware problem at some level - software issues should never cause
484such a mismatch.
485
486However on RAID1 and RAID10 it is possible for software issues to
487cause a mismatch to be reported. This does not necessarily mean that
488the data on the array is corrupted. It could simply be that the
489system does not care what is stored on that part of the array - it is
490unused space.
491
492The most likely cause for an unexpected mismatch on RAID1 or RAID10
493occurs if a swap partition or swap file is stored on the array.
494
495When the swap subsystem wants to write a page of memory out, it flags
496the page as 'clean' in the memory manager and requests the swap device
497to write it out. It is quite possible that the memory will be
498changed while the write-out is happening. In that case the 'clean'
499flag will be found to be clear when the write completes and so the
500swap subsystem will simply forget that the swapout had been attempted,
c93e9d68 501and will possibly choose a different page to write out.
1cc44574 502
c93e9d68 503If the swap device was on RAID1 (or RAID10), then the data is sent
1cc44574 504from memory to a device twice (or more depending on the number of
c93e9d68
N
505devices in the array). Thus it is possible that the memory gets changed
506between the times it is sent, so different data can be written to
507the different devices in the array. This will be detected by
1cc44574
N
508.I check
509as a mismatch. However it does not reflect any corruption as the
510block where this mismatch occurs is being treated by the swap system as
511being empty, and the data will never be read from that block.
512
513It is conceivable for a similar situation to occur on non-swap files,
514though it is less likely.
515
516Thus the
517.I mismatch_cnt
518value can not be interpreted very reliably on RAID1 or RAID10,
519especially when the device is used for swap.
520
521
599e5a36
NB
522.SS BITMAP WRITE-INTENT LOGGING
523
524From Linux 2.6.13,
525.I md
526supports a bitmap based write-intent log. If configured, the bitmap
527is used to record which blocks of the array may be out of sync.
528Before any write request is honoured, md will make sure that the
529corresponding bit in the log is set. After a period of time with no
530writes to an area of the array, the corresponding bit will be cleared.
531
532This bitmap is used for two optimisations.
533
1afe1167 534Firstly, after an unclean shutdown, the resync process will consult
599e5a36 535the bitmap and only resync those blocks that correspond to bits in the
1afe1167 536bitmap that are set. This can dramatically reduce resync time.
599e5a36
NB
537
538Secondly, when a drive fails and is removed from the array, md stops
539clearing bits in the intent log. If that same drive is re-added to
540the array, md will notice and will only recover the sections of the
541drive that are covered by bits in the intent log that are set. This
542can allow a device to be temporarily removed and reinserted without
543causing an enormous recovery cost.
544
545The intent log can be stored in a file on a separate device, or it can
546be stored near the superblocks of an array which has superblocks.
547
93e790af 548It is possible to add an intent log to an active array, or remove an
addc80c4 549intent log if one is present.
599e5a36
NB
550
551In 2.6.13, intent bitmaps are only supported with RAID1. Other levels
addc80c4 552with redundancy are supported from 2.6.15.
599e5a36
NB
553
554.SS WRITE-BEHIND
555
556From Linux 2.6.14,
557.I md
addc80c4 558supports WRITE-BEHIND on RAID1 arrays.
599e5a36
NB
559
560This allows certain devices in the array to be flagged as
561.IR write-mostly .
562MD will only read from such devices if there is no
563other option.
564
565If a write-intent bitmap is also provided, write requests to
566write-mostly devices will be treated as write-behind requests and md
567will not wait for writes to those requests to complete before
568reporting the write as complete to the filesystem.
569
570This allows for a RAID1 with WRITE-BEHIND to be used to mirror data
8f21823f 571over a slow link to a remote computer (providing the link isn't too
599e5a36
NB
572slow). The extra latency of the remote link will not slow down normal
573operations, but the remote system will still have a reasonably
574up-to-date copy of all data.
575
addc80c4
NB
576.SS RESTRIPING
577
578.IR Restriping ,
579also known as
580.IR Reshaping ,
581is the processes of re-arranging the data stored in each stripe into a
582new layout. This might involve changing the number of devices in the
93e790af 583array (so the stripes are wider), changing the chunk size (so stripes
addc80c4 584are deeper or shallower), or changing the arrangement of data and
93e790af 585parity (possibly changing the raid level, e.g. 1 to 5 or 5 to 6).
addc80c4 586
c64881d7
N
587As of Linux 2.6.35, md can reshape a RAID4, RAID5, or RAID6 array to
588have a different number of devices (more or fewer) and to have a
589different layout or chunk size. It can also convert between these
590different RAID levels. It can also convert between RAID0 and RAID10,
591and between RAID0 and RAID4 or RAID5.
592Other possibilities may follow in future kernels.
addc80c4
NB
593
594During any stripe process there is a 'critical section' during which
35cc5be4 595live data is being overwritten on disk. For the operation of
addc80c4
NB
596increasing the number of drives in a raid5, this critical section
597covers the first few stripes (the number being the product of the old
598and new number of devices). After this critical section is passed,
599data is only written to areas of the array which no longer hold live
b3f1c093 600data \(em the live data has already been located away.
addc80c4 601
c64881d7
N
602For a reshape which reduces the number of devices, the 'critical
603section' is at the end of the reshape process.
604
addc80c4
NB
605md is not able to ensure data preservation if there is a crash
606(e.g. power failure) during the critical section. If md is asked to
607start an array which failed during a critical section of restriping,
608it will fail to start the array.
609
610To deal with this possibility, a user-space program must
611.IP \(bu 4
612Disable writes to that section of the array (using the
613.B sysfs
614interface),
615.IP \(bu 4
93e790af 616take a copy of the data somewhere (i.e. make a backup),
addc80c4 617.IP \(bu 4
93e790af 618allow the process to continue and invalidate the backup and restore
addc80c4
NB
619write access once the critical section is passed, and
620.IP \(bu 4
93e790af 621provide for restoring the critical data before restarting the array
addc80c4
NB
622after a system crash.
623.PP
624
625.B mdadm
93e790af 626versions from 2.4 do this for growing a RAID5 array.
addc80c4
NB
627
628For operations that do not change the size of the array, like simply
629increasing chunk size, or converting RAID5 to RAID6 with one extra
93e790af
SW
630device, the entire process is the critical section. In this case, the
631restripe will need to progress in stages, as a section is suspended,
c64881d7 632backed up, restriped, and released.
addc80c4
NB
633
634.SS SYSFS INTERFACE
93e790af 635Each block device appears as a directory in
addc80c4 636.I sysfs
93e790af 637(which is usually mounted at
addc80c4
NB
638.BR /sys ).
639For MD devices, this directory will contain a subdirectory called
640.B md
641which contains various files for providing access to information about
642the array.
643
644This interface is documented more fully in the file
645.B Documentation/md.txt
646which is distributed with the kernel sources. That file should be
647consulted for full documentation. The following are just a selection
648of attribute files that are available.
649
650.TP
651.B md/sync_speed_min
652This value, if set, overrides the system-wide setting in
653.B /proc/sys/dev/raid/speed_limit_min
654for this array only.
655Writing the value
93e790af
SW
656.B "system"
657to this file will cause the system-wide setting to have effect.
addc80c4
NB
658
659.TP
660.B md/sync_speed_max
661This is the partner of
662.B md/sync_speed_min
663and overrides
1e49aaa0 664.B /proc/sys/dev/raid/speed_limit_max
addc80c4
NB
665described below.
666
667.TP
668.B md/sync_action
669This can be used to monitor and control the resync/recovery process of
670MD.
671In particular, writing "check" here will cause the array to read all
672data block and check that they are consistent (e.g. parity is correct,
673or all mirror replicas are the same). Any discrepancies found are
674.B NOT
675corrected.
676
677A count of problems found will be stored in
678.BR md/mismatch_count .
679
680Alternately, "repair" can be written which will cause the same check
681to be performed, but any errors will be corrected.
682
683Finally, "idle" can be written to stop the check/repair process.
684
685.TP
686.B md/stripe_cache_size
687This is only available on RAID5 and RAID6. It records the size (in
688pages per device) of the stripe cache which is used for synchronising
800053d6
DW
689all write operations to the array and all read operations if the array
690is degraded. The default is 256. Valid values are 17 to 32768.
addc80c4 691Increasing this number can increase performance in some situations, at
800053d6
DW
692some cost in system memory. Note, setting this value too high can
693result in an "out of memory" condition for the system.
694
695memory_consumed = system_page_size * nr_disks * stripe_cache_size
addc80c4 696
a5ee6dfb
DW
697.TP
698.B md/preread_bypass_threshold
699This is only available on RAID5 and RAID6. This variable sets the
700number of times MD will service a full-stripe-write before servicing a
701stripe that requires some "prereading". For fairness this defaults to
800053d6
DW
7021. Valid values are 0 to stripe_cache_size. Setting this to 0
703maximizes sequential-write throughput at the cost of fairness to threads
704doing small or random writes.
addc80c4 705
5787fa49
NB
706.SS KERNEL PARAMETERS
707
addc80c4 708The md driver recognised several different kernel parameters.
5787fa49
NB
709.TP
710.B raid=noautodetect
711This will disable the normal detection of md arrays that happens at
712boot time. If a drive is partitioned with MS-DOS style partitions,
713then if any of the 4 main partitions has a partition type of 0xFD,
714then that partition will normally be inspected to see if it is part of
715an MD array, and if any full arrays are found, they are started. This
addc80c4 716kernel parameter disables this behaviour.
5787fa49 717
a9d69660
NB
718.TP
719.B raid=partitionable
720.TP
721.B raid=part
722These are available in 2.6 and later kernels only. They indicate that
723autodetected MD arrays should be created as partitionable arrays, with
724a different major device number to the original non-partitionable md
725arrays. The device number is listed as
726.I mdp
727in
728.IR /proc/devices .
729
addc80c4
NB
730.TP
731.B md_mod.start_ro=1
e0fe762a
N
732.TP
733.B /sys/module/md_mod/parameters/start_ro
addc80c4
NB
734This tells md to start all arrays in read-only mode. This is a soft
735read-only that will automatically switch to read-write on the first
736write request. However until that write request, nothing is written
737to any device by md, and in particular, no resync or recovery
738operation is started.
739
740.TP
741.B md_mod.start_dirty_degraded=1
e0fe762a
N
742.TP
743.B /sys/module/md_mod/parameters/start_dirty_degraded
addc80c4
NB
744As mentioned above, md will not normally start a RAID4, RAID5, or
745RAID6 that is both dirty and degraded as this situation can imply
746hidden data loss. This can be awkward if the root filesystem is
93e790af 747affected. Using this module parameter allows such arrays to be started
addc80c4
NB
748at boot time. It should be understood that there is a real (though
749small) risk of data corruption in this situation.
a9d69660 750
5787fa49
NB
751.TP
752.BI md= n , dev , dev ,...
a9d69660
NB
753.TP
754.BI md=d n , dev , dev ,...
5787fa49
NB
755This tells the md driver to assemble
756.B /dev/md n
757from the listed devices. It is only necessary to start the device
758holding the root filesystem this way. Other arrays are best started
759once the system is booted.
760
a9d69660
NB
761In 2.6 kernels, the
762.B d
763immediately after the
764.B =
765indicates that a partitionable device (e.g.
766.BR /dev/md/d0 )
767should be created rather than the original non-partitionable device.
768
5787fa49
NB
769.TP
770.BI md= n , l , c , i , dev...
771This tells the md driver to assemble a legacy RAID0 or LINEAR array
772without a superblock.
773.I n
774gives the md device number,
775.I l
776gives the level, 0 for RAID0 or -1 for LINEAR,
777.I c
778gives the chunk size as a base-2 logarithm offset by twelve, so 0
779means 4K, 1 means 8K.
780.I i
781is ignored (legacy support).
e0d19036 782
56eb10c0
NB
783.SH FILES
784.TP
785.B /proc/mdstat
786Contains information about the status of currently running array.
787.TP
788.B /proc/sys/dev/raid/speed_limit_min
93e790af 789A readable and writable file that reflects the current "goal" rebuild
56eb10c0
NB
790speed for times when non-rebuild activity is current on an array.
791The speed is in Kibibytes per second, and is a per-device rate, not a
93e790af 792per-array rate (which means that an array with more disks will shuffle
e0fe762a 793more data for a given speed). The default is 1000.
56eb10c0
NB
794
795.TP
796.B /proc/sys/dev/raid/speed_limit_max
93e790af 797A readable and writable file that reflects the current "goal" rebuild
56eb10c0 798speed for times when no non-rebuild activity is current on an array.
e0fe762a 799The default is 200,000.
56eb10c0
NB
800
801.SH SEE ALSO
802.BR mdadm (8),
803.BR mkraid (8).