1
# From: http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=blob_plain;f=Documentation/md.txt;hb=v2.6.31
3
Tools that manage md devices can be found at
4
http://www.<country>.kernel.org/pub/linux/utils/raid/....
7
Boot time assembly of RAID arrays
8
---------------------------------
10
You can boot with your md device with the following kernel command
13
for old raid arrays without persistent superblocks:
14
md=<md device no.>,<raid level>,<chunk size factor>,<fault level>,dev0,dev1,...,devn
16
for raid arrays with persistent superblocks
17
md=<md device no.>,dev0,dev1,...,devn
18
or, to assemble a partitionable array:
19
md=d<md device no.>,dev0,dev1,...,devn
21
md device no. = the number of the md device ...
28
raid level = -1 linear mode
30
other modes are only supported with persistent super blocks
32
chunk size factor = (raid-0 and raid-1 only)
33
Set the chunk size as 4k << n.
35
fault level = totally ignored
37
dev0-devn: e.g. /dev/hda1,/dev/hdc1,/dev/sda1,/dev/sdb1
39
A possible loadlin line (Harald Hoyer <HarryH@Royal.Net>) looks like this:
41
e:\loadlin\loadlin e:\zimage root=/dev/md0 md=0,0,4,0,/dev/hdb2,/dev/hdc3 ro
44
Boot time autodetection of RAID arrays
45
--------------------------------------
47
When md is compiled into the kernel (not as module), partitions of
48
type 0xfd are scanned and automatically assembled into RAID arrays.
49
This autodetection may be suppressed with the kernel parameter
50
"raid=noautodetect". As of kernel 2.6.9, only drives with a type 0
51
superblock can be autodetected and run at boot time.
53
The kernel parameter "raid=partitionable" (or "raid=part") means
54
that all auto-detected arrays are assembled as partitionable.
56
Boot time assembly of degraded/dirty arrays
57
-------------------------------------------
59
If a raid5 or raid6 array is both dirty and degraded, it could have
60
undetectable data corruption. This is because the fact that it is
61
'dirty' means that the parity cannot be trusted, and the fact that it
62
is degraded means that some datablocks are missing and cannot reliably
63
be reconstructed (due to no parity).
65
For this reason, md will normally refuse to start such an array. This
66
requires the sysadmin to take action to explicitly start the array
67
despite possible corruption. This is normally done with
68
mdadm --assemble --force ....
70
This option is not really available if the array has the root
71
filesystem on it. In order to support this booting from such an
72
array, md supports a module parameter "start_dirty_degraded" which,
73
when set to 1, bypassed the checks and will allows dirty degraded
76
So, to boot with a root filesystem of a dirty degraded raid[56], use
78
md-mod.start_dirty_degraded=1
84
The md driver can support a variety of different superblock formats.
85
Currently, it supports superblock formats "0.90.0" and the "md-1" format
86
introduced in the 2.5 development series.
88
The kernel will autodetect which format superblock is being used.
90
Superblock format '0' is treated differently to others for legacy
91
reasons - it is the original superblock format.
94
General Rules - apply for all superblock formats
95
------------------------------------------------
97
An array is 'created' by writing appropriate superblocks to all
100
It is 'assembled' by associating each of these devices with an
101
particular md virtual device. Once it is completely assembled, it can
104
An array should be created by a user-space tool. This will write
105
superblocks to all devices. It will usually mark the array as
106
'unclean', or with some devices missing so that the kernel md driver
107
can create appropriate redundancy (copying in raid1, parity
108
calculation in raid4/5).
110
When an array is assembled, it is first initialized with the
111
SET_ARRAY_INFO ioctl. This contains, in particular, a major and minor
112
version number. The major version number selects which superblock
113
format is to be used. The minor number might be used to tune handling
114
of the format, such as suggesting where on each device to look for the
117
Then each device is added using the ADD_NEW_DISK ioctl. This
118
provides, in particular, a major and minor number identifying the
121
The array is started with the RUN_ARRAY ioctl.
123
Once started, new devices can be added. They should have an
124
appropriate superblock written to them, and then passed be in with
127
Devices that have failed or are not yet active can be detached from an
128
array using HOT_REMOVE_DISK.
131
Specific Rules that apply to format-0 super block arrays, and
132
arrays with no superblock (non-persistent).
133
-------------------------------------------------------------
135
An array can be 'created' by describing the array (level, chunksize
136
etc) in a SET_ARRAY_INFO ioctl. This must has major_version==0 and
139
Then uninitialized devices can be added with ADD_NEW_DISK. The
140
structure passed to ADD_NEW_DISK must specify the state of the device
141
and it's role in the array.
143
Once started with RUN_ARRAY, uninitialized spares can be added with
150
md devices appear in sysfs (/sys) as regular block devices,
154
Each 'md' device will contain a subdirectory called 'md' which
155
contains further md-specific information about the device.
157
All md devices contain:
159
a text file indicating the 'raid level'. e.g. raid0, raid1,
160
raid5, linear, multipath, faulty.
161
If no raid level has been set yet (array is still being
162
assembled), the value will reflect whatever has been written
163
to it, which may be a name like the above, or may be a number
164
such as '0', '5', etc.
167
a text file with a simple number indicating the number of devices
168
in a fully functional array. If this is not yet known, the file
169
will be empty. If an array is being resized this will contain
170
the new number of devices.
171
Some raid levels allow this value to be set while the array is
172
active. This will reconfigure the array. Otherwise it can only
173
be set while assembling an array.
174
A change to this attribute will not be permitted if it would
175
reduce the size of the array. To reduce the number of drives
176
in an e.g. raid5, the array size must first be reduced by
177
setting the 'array_size' attribute.
180
This is the size in bytes for 'chunks' and is only relevant to
181
raid levels that involve striping (0,4,5,6,10). The address space
182
of the array is conceptually divided into chunks and consecutive
183
chunks are striped onto neighbouring devices.
184
The size should be at least PAGE_SIZE (4k) and should be a power
185
of 2. This can only be set while assembling an array
188
The "layout" for the array for the particular level. This is
189
simply a number that is interpretted differently by different
190
levels. It can be written while assembling an array.
193
This can be used to artificially constrain the available space in
194
the array to be less than is actually available on the combined
195
devices. Writing a number (in Kilobytes) which is less than
196
the available size will set the size. Any reconfiguration of the
197
array (e.g. adding devices) will not cause the size to change.
198
Writing the word 'default' will cause the effective size of the
199
array to be whatever size is actually available based on
200
'level', 'chunk_size' and 'component_size'.
202
This can be used to reduce the size of the array before reducing
203
the number of devices in a raid4/5/6, or to support external
204
metadata formats which mandate such clipping.
207
This is either "none" or a sector number within the devices of
208
the array where "reshape" is up to. If this is set, the three
209
attributes mentioned above (raid_disks, chunk_size, layout) can
210
potentially have 2 values, an old and a new value. If these
211
values differ, reading the attribute returns
213
and writing will effect the 'new' value, leaving the 'old'
217
For arrays with data redundancy (i.e. not raid0, linear, faulty,
218
multipath), all components must be the same size - or at least
219
there must a size that they all provide space for. This is a key
220
part or the geometry of the array. It is measured in sectors
221
and can be read from here. Writing to this value may resize
222
the array if the personality supports it (raid1, raid5, raid6),
223
and if the component drives are large enough.
226
This indicates the format that is being used to record metadata
227
about the array. It can be 0.90 (traditional format), 1.0, 1.1,
228
1.2 (newer format in varying locations) or "none" indicating that
229
the kernel isn't managing metadata at all.
230
Alternately it can be "external:" followed by a string which
231
is set by user-space. This indicates that metadata is managed
232
by a user-space program. Any device failure or other event that
233
requires a metadata update will cause array activity to be
234
suspended until the event is acknowledged.
237
The point at which resync should start. If no resync is needed,
238
this will be a very large number. At array creation it will
239
default to 0, though starting the array as 'clean' will
243
This file can be written but not read. The value written should
244
be a block device number as major:minor. e.g. 8:0
245
This will cause that device to be attached to the array, if it is
246
available. It will then appear at md/dev-XXX (depending on the
247
name of the device) and further configuration is then possible.
250
When an md array has seen no write requests for a certain period
251
of time, it will be marked as 'clean'. When another write
252
request arrives, the array is marked as 'dirty' before the write
253
commences. This is known as 'safe_mode'.
254
The 'certain period' is controlled by this file which stores the
255
period as a number of seconds. The default is 200msec (0.200).
256
Writing a value of 0 disables safemode.
259
This file contains a single word which describes the current
260
state of the array. In many cases, the state can be set by
261
writing the word for the desired state, however some states
262
cannot be explicitly set, and some transitions are not allowed.
264
Select/poll works on this file. All changes except between
265
active_idle and active (which can be frequent and are not
266
very interesting) are notified. active->active_idle is
267
reported if the metadata is externally managed.
270
No devices, no size, no level
271
Writing is equivalent to STOP_ARRAY ioctl
273
May have some settings, but array is not active
274
all IO results in error
275
When written, doesn't tear down array, but just stops it
276
suspended (not supported yet)
277
All IO requests will block. The array can be reconfigured.
278
Writing this, if accepted, will block until array is quiessent
280
no resync can happen. no superblocks get written.
283
like readonly, but behaves like 'clean' on a write request.
285
clean - no pending writes, but otherwise active.
286
When written to inactive array, starts without resync
287
If a write request arrives then
288
if metadata is known, mark 'dirty' and switch to 'active'.
289
if not known, block and switch to write-pending
290
If written to an active array that has pending writes, then fails.
292
fully active: IO and resync can be happening.
293
When written to inactive array, starts with resync
296
clean, but writes are blocked waiting for 'active' to be written.
299
like active, but no writes have been seen for a while (safe_mode_delay).
302
As component devices are added to an md array, they appear in the 'md'
303
directory as new directories named
305
where XXX is a name that the kernel knows for the device, e.g. hdb1.
306
Each directory contains:
309
a symlink to the block device in /sys/block, e.g.
310
/sys/block/md0/md/dev-hdb1/block -> ../../../../block/hdb/hdb1
313
A file containing an image of the superblock read from, or
314
written to, that device.
317
A file recording the current state of the device in the array
318
which can be a comma separated list of
319
faulty - device has been kicked from active use due to
321
in_sync - device is a fully in-sync member of the array
322
writemostly - device will only be subject to read
323
requests if there are no other options.
324
This applies only to raid1 arrays.
325
blocked - device has failed, metadata is "external",
326
and the failure hasn't been acknowledged yet.
327
Writes that would write to this device if
328
it were not faulty are blocked.
329
spare - device is working, but not a full member.
330
This includes spares that are in the process
331
of being recovered to
332
This list may grow in future.
333
This can be written to.
334
Writing "faulty" simulates a failure on the device.
335
Writing "remove" removes the device from the array.
336
Writing "writemostly" sets the writemostly flag.
337
Writing "-writemostly" clears the writemostly flag.
338
Writing "blocked" sets the "blocked" flag.
339
Writing "-blocked" clear the "blocked" flag and allows writes
342
This file responds to select/poll. Any change to 'faulty'
343
or 'blocked' causes an event.
346
An approximate count of read errors that have been detected on
347
this device but have not caused the device to be evicted from
348
the array (either because they were corrected or because they
349
happened while the array was read-only). When using version-1
350
metadata, this value persists across restarts of the array.
352
This value can be written while assembling an array thus
353
providing an ongoing count for arrays with metadata managed by
357
This gives the role that the device has in the array. It will
358
either be 'none' if the device is not active in the array
359
(i.e. is a spare or has failed) or an integer less than the
360
'raid_disks' number for the array indicating which position
361
it currently fills. This can only be set while assembling an
362
array. A device for which this is set is assumed to be working.
365
This gives the location in the device (in sectors from the
366
start) where data from the array will be stored. Any part of
367
the device before this offset us not touched, unless it is
368
used for storing metadata (Formats 1.1 and 1.2).
371
The amount of the device, after the offset, that can be used
372
for storage of data. This will normally be the same as the
373
component_size. This can be written while assembling an
374
array. If a value less than the current component_size is
375
written, it will be rejected.
378
An active md device will also contain and entry for each active device
379
in the array. These are named
383
where 'NN' is the position in the array, starting from 0.
384
So for a 3 drive array there will be rd0, rd1, rd2.
385
These are symbolic links to the appropriate 'dev-XXX' entry.
387
cat /sys/block/md*/md/rd*/state
388
will show 'in_sync' on every line.
392
Active md devices for levels that support data redundancy (1,4,5,6)
396
a text file that can be used to monitor and control the rebuild
397
process. It contains one word which can be one of:
398
resync - redundancy is being recalculated after unclean
400
recover - a hot spare is being built to replace a
401
failed/missing device
402
idle - nothing is happening
403
check - A full check of redundancy was requested and is
404
happening. This reads all block and checks
405
them. A repair may also happen for some raid
407
repair - A full check and repair is happening. This is
408
similar to 'resync', but was requested by the
409
user, and the write-intent bitmap is NOT used to
410
optimise the process.
412
This file is writable, and each of the strings that could be
413
read are meaningful for writing.
415
'idle' will stop an active resync/recovery etc. There is no
416
guarantee that another resync/recovery may not be automatically
417
started again, though some event will be needed to trigger
419
'resync' or 'recovery' can be used to restart the
420
corresponding operation if it was stopped with 'idle'.
421
'check' and 'repair' will start the appropriate process
422
providing the current state is 'idle'.
424
This file responds to select/poll. Any important change in the value
425
triggers a poll event. Sometimes the value will briefly be
426
"recover" if a recovery seems to be needed, but cannot be
427
achieved. In that case, the transition to "recover" isn't
428
notified, but the transition away is.
431
This contains a count of the number of devices by which the
432
arrays is degraded. So an optimal array with show '0'. A
433
single failed/missing drive will show '1', etc.
434
This file responds to select/poll, any increase or decrease
435
in the count of missing devices will trigger an event.
438
When performing 'check' and 'repair', and possibly when
439
performing 'resync', md will count the number of errors that are
440
found. The count in 'mismatch_cnt' is the number of sectors
441
that were re-written, or (for 'check') would have been
442
re-written. As most raid levels work in units of pages rather
443
than sectors, this my be larger than the number of actual errors
444
by a factor of the number of sectors in a page.
447
If the array has a write-intent bitmap, then writing to this
448
attribute can set bits in the bitmap, indicating that a resync
449
would need to check the corresponding blocks. Either individual
450
numbers or start-end pairs can be written. Multiple numbers
451
can be separated by a space.
452
Note that the numbers are 'bit' numbers, not 'block' numbers.
453
They should be scaled by the bitmap_chunksize.
457
This are similar to /proc/sys/dev/raid/speed_limit_{min,max}
458
however they only apply to the particular array.
459
If no value has been written to these, of if the word 'system'
460
is written, then the system-wide value is used. If a value,
461
in kibibytes-per-second is written, then it is used.
462
When the files are read, they show the currently active value
463
followed by "(local)" or "(system)" depending on whether it is
464
a locally set or system-wide value.
467
This shows the number of sectors that have been completed of
468
whatever the current sync_action is, followed by the number of
469
sectors in total that could need to be processed. The two
470
numbers are separated by a '/' thus effectively showing one
471
value, a fraction of the process that is complete.
472
A 'select' on this attribute will return when resync completes,
473
when it reaches the current sync_max (below) and possibly at
477
This is a number of sectors at which point a resync/recovery
478
process will pause. When a resync is active, the value can
479
only ever be increased, never decreased. The value of 'max'
480
effectively disables the limit.
484
This shows the current actual speed, in K/sec, of the current
485
sync_action. It is averaged over the last 30 seconds.
489
The two values, given as numbers of sectors, indicate a range
490
within the array where IO will be blocked. This is currently
491
only supported for raid4/5/6.
494
Each active md device may also have attributes specific to the
495
personality module that manages it.
496
These are specific to the implementation of the module and could
497
change substantially if the implementation changes.
499
These currently include
501
stripe_cache_size (currently raid5 only)
502
number of entries in the stripe cache. This is writable, but
503
there are upper and lower limits (32768, 16). Default is 128.
504
strip_cache_active (currently raid5 only)
505
number of active entries in the stripe cache
506
preread_bypass_threshold (currently raid5 only)
507
number of times a stripe requiring preread will be bypassed by
508
a stripe that does not require preread. For fairness defaults
509
to 1. Setting this to 0 disables bypass accounting and
510
requires preread stripes to wait until all full-width stripe-
511
writes are complete. Valid values are 0 to stripe_cache_size.