Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1=======
2dm-raid
3=======
4
5The device-mapper RAID (dm-raid) target provides a bridge from DM to MD.
6It allows the MD RAID drivers to be accessed using a device-mapper
7interface.
8
9
10Mapping Table Interface
11-----------------------
12The target is named "raid" and it accepts the following parameters::
13
14 <raid_type> <#raid_params> <raid_params> \
15 <#raid_devs> <metadata_dev0> <dev0> [.. <metadata_devN> <devN>]
16
17<raid_type>:
18
19 ============= ===============================================================
20 raid0 RAID0 striping (no resilience)
21 raid1 RAID1 mirroring
22 raid4 RAID4 with dedicated last parity disk
23 raid5_n RAID5 with dedicated last parity disk supporting takeover from/to raid1
24 Same as raid4
25
26 - Transitory layout for takeover from/to raid1
27 raid5_la RAID5 left asymmetric
28
29 - rotating parity 0 with data continuation
30 raid5_ra RAID5 right asymmetric
31
32 - rotating parity N with data continuation
33 raid5_ls RAID5 left symmetric
34
35 - rotating parity 0 with data restart
36 raid5_rs RAID5 right symmetric
37
38 - rotating parity N with data restart
39 raid6_zr RAID6 zero restart
40
41 - rotating parity zero (left-to-right) with data restart
42 raid6_nr RAID6 N restart
43
44 - rotating parity N (right-to-left) with data restart
45 raid6_nc RAID6 N continue
46
47 - rotating parity N (right-to-left) with data continuation
48 raid6_n_6 RAID6 with dedicate parity disks
49
50 - parity and Q-syndrome on the last 2 disks;
51 layout for takeover from/to raid0/raid4/raid5_n
52 raid6_la_6 Same as "raid_la" plus dedicated last Q-syndrome disk supporting takeover from/to raid5
53
54 - layout for takeover from raid5_la from/to raid6
55 raid6_ra_6 Same as "raid5_ra" dedicated last Q-syndrome disk
56
57 - layout for takeover from raid5_ra from/to raid6
58 raid6_ls_6 Same as "raid5_ls" dedicated last Q-syndrome disk
59
60 - layout for takeover from raid5_ls from/to raid6
61 raid6_rs_6 Same as "raid5_rs" dedicated last Q-syndrome disk
62
63 - layout for takeover from raid5_rs from/to raid6
64 raid10 Various RAID10 inspired algorithms chosen by additional params
65 (see raid10_format and raid10_copies below)
66
67 - RAID10: Striped Mirrors (aka 'Striping on top of mirrors')
68 - RAID1E: Integrated Adjacent Stripe Mirroring
69 - RAID1E: Integrated Offset Stripe Mirroring
70 - and other similar RAID10 variants
71 ============= ===============================================================
72
73 Reference: Chapter 4 of
74 https://www.snia.org/sites/default/files/SNIA_DDF_Technical_Position_v2.0.pdf
75
76<#raid_params>: The number of parameters that follow.
77
78<raid_params> consists of
79
80 Mandatory parameters:
81 <chunk_size>:
82 Chunk size in sectors. This parameter is often known as
83 "stripe size". It is the only mandatory parameter and
84 is placed first.
85
86 followed by optional parameters (in any order):
87 [sync|nosync]
88 Force or prevent RAID initialization.
89
90 [rebuild <idx>]
91 Rebuild drive number 'idx' (first drive is 0).
92
93 [daemon_sleep <ms>]
94 Interval between runs of the bitmap daemon that
95 clear bits. A longer interval means less bitmap I/O but
96 resyncing after a failure is likely to take longer.
97
98 [min_recovery_rate <kB/sec/disk>]
99 Throttle RAID initialization
100 [max_recovery_rate <kB/sec/disk>]
101 Throttle RAID initialization
102 [write_mostly <idx>]
103 Mark drive index 'idx' write-mostly.
104 [max_write_behind <sectors>]
105 See '--write-behind=' (man mdadm)
106 [stripe_cache <sectors>]
107 Stripe cache size (RAID 4/5/6 only)
108 [region_size <sectors>]
109 The region_size multiplied by the number of regions is the
110 logical size of the array. The bitmap records the device
111 synchronisation state for each region.
112
113 [raid10_copies <# copies>], [raid10_format <near|far|offset>]
114 These two options are used to alter the default layout of
115 a RAID10 configuration. The number of copies is can be
116 specified, but the default is 2. There are also three
117 variations to how the copies are laid down - the default
118 is "near". Near copies are what most people think of with
119 respect to mirroring. If these options are left unspecified,
120 or 'raid10_copies 2' and/or 'raid10_format near' are given,
121 then the layouts for 2, 3 and 4 devices are:
122
123 ======== ========== ==============
124 2 drives 3 drives 4 drives
125 ======== ========== ==============
126 A1 A1 A1 A1 A2 A1 A1 A2 A2
127 A2 A2 A2 A3 A3 A3 A3 A4 A4
128 A3 A3 A4 A4 A5 A5 A5 A6 A6
129 A4 A4 A5 A6 A6 A7 A7 A8 A8
130 .. .. .. .. .. .. .. .. ..
131 ======== ========== ==============
132
133 The 2-device layout is equivalent 2-way RAID1. The 4-device
134 layout is what a traditional RAID10 would look like. The
135 3-device layout is what might be called a 'RAID1E - Integrated
136 Adjacent Stripe Mirroring'.
137
138 If 'raid10_copies 2' and 'raid10_format far', then the layouts
139 for 2, 3 and 4 devices are:
140
141 ======== ============ ===================
142 2 drives 3 drives 4 drives
143 ======== ============ ===================
144 A1 A2 A1 A2 A3 A1 A2 A3 A4
145 A3 A4 A4 A5 A6 A5 A6 A7 A8
146 A5 A6 A7 A8 A9 A9 A10 A11 A12
147 .. .. .. .. .. .. .. .. ..
148 A2 A1 A3 A1 A2 A2 A1 A4 A3
149 A4 A3 A6 A4 A5 A6 A5 A8 A7
150 A6 A5 A9 A7 A8 A10 A9 A12 A11
151 .. .. .. .. .. .. .. .. ..
152 ======== ============ ===================
153
154 If 'raid10_copies 2' and 'raid10_format offset', then the
155 layouts for 2, 3 and 4 devices are:
156
157 ======== ========== ================
158 2 drives 3 drives 4 drives
159 ======== ========== ================
160 A1 A2 A1 A2 A3 A1 A2 A3 A4
161 A2 A1 A3 A1 A2 A2 A1 A4 A3
162 A3 A4 A4 A5 A6 A5 A6 A7 A8
163 A4 A3 A6 A4 A5 A6 A5 A8 A7
164 A5 A6 A7 A8 A9 A9 A10 A11 A12
165 A6 A5 A9 A7 A8 A10 A9 A12 A11
166 .. .. .. .. .. .. .. .. ..
167 ======== ========== ================
168
169 Here we see layouts closely akin to 'RAID1E - Integrated
170 Offset Stripe Mirroring'.
171
172 [delta_disks <N>]
173 The delta_disks option value (-251 < N < +251) triggers
174 device removal (negative value) or device addition (positive
175 value) to any reshape supporting raid levels 4/5/6 and 10.
176 RAID levels 4/5/6 allow for addition and removal of devices
177 (metadata and data device tuple), raid10_near and raid10_offset
178 only allow for device addition. raid10_far does not support any
179 reshaping at all.
180 A minimum of devices have to be kept to enforce resilience,
181 which is 3 devices for raid4/5 and 4 devices for raid6.
182
183 [data_offset <sectors>]
184 This option value defines the offset into each data device
185 where the data starts. This is used to provide out-of-place
186 reshaping space to avoid writing over data while
187 changing the layout of stripes, hence an interruption/crash
188 may happen at any time without the risk of losing data.
189 E.g. when adding devices to an existing raid set during
190 forward reshaping, the out-of-place space will be allocated
191 at the beginning of each raid device. The kernel raid4/5/6/10
192 MD personalities supporting such device addition will read the data from
193 the existing first stripes (those with smaller number of stripes)
194 starting at data_offset to fill up a new stripe with the larger
195 number of stripes, calculate the redundancy blocks (CRC/Q-syndrome)
196 and write that new stripe to offset 0. Same will be applied to all
197 N-1 other new stripes. This out-of-place scheme is used to change
198 the RAID type (i.e. the allocation algorithm) as well, e.g.
199 changing from raid5_ls to raid5_n.
200
201 [journal_dev <dev>]
202 This option adds a journal device to raid4/5/6 raid sets and
203 uses it to close the 'write hole' caused by the non-atomic updates
204 to the component devices which can cause data loss during recovery.
205 The journal device is used as writethrough thus causing writes to
206 be throttled versus non-journaled raid4/5/6 sets.
207 Takeover/reshape is not possible with a raid4/5/6 journal device;
208 it has to be deconfigured before requesting these.
209
210 [journal_mode <mode>]
211 This option sets the caching mode on journaled raid4/5/6 raid sets
212 (see 'journal_dev <dev>' above) to 'writethrough' or 'writeback'.
213 If 'writeback' is selected the journal device has to be resilient
214 and must not suffer from the 'write hole' problem itself (e.g. use
215 raid1 or raid10) to avoid a single point of failure.
216
217<#raid_devs>: The number of devices composing the array.
218 Each device consists of two entries. The first is the device
219 containing the metadata (if any); the second is the one containing the
220 data. A Maximum of 64 metadata/data device entries are supported
221 up to target version 1.8.0.
222 1.9.0 supports up to 253 which is enforced by the used MD kernel runtime.
223
224 If a drive has failed or is missing at creation time, a '-' can be
225 given for both the metadata and data drives for a given position.
226
227
228Example Tables
229--------------
230
231::
232
233 # RAID4 - 4 data drives, 1 parity (no metadata devices)
234 # No metadata devices specified to hold superblock/bitmap info
235 # Chunk size of 1MiB
236 # (Lines separated for easy reading)
237
238 0 1960893648 raid \
239 raid4 1 2048 \
240 5 - 8:17 - 8:33 - 8:49 - 8:65 - 8:81
241
242 # RAID4 - 4 data drives, 1 parity (with metadata devices)
243 # Chunk size of 1MiB, force RAID initialization,
244 # min recovery rate at 20 kiB/sec/disk
245
246 0 1960893648 raid \
247 raid4 4 2048 sync min_recovery_rate 20 \
248 5 8:17 8:18 8:33 8:34 8:49 8:50 8:65 8:66 8:81 8:82
249
250
251Status Output
252-------------
253'dmsetup table' displays the table used to construct the mapping.
254The optional parameters are always printed in the order listed
255above with "sync" or "nosync" always output ahead of the other
256arguments, regardless of the order used when originally loading the table.
257Arguments that can be repeated are ordered by value.
258
259
260'dmsetup status' yields information on the state and health of the array.
261The output is as follows (normally a single line, but expanded here for
262clarity)::
263
264 1: <s> <l> raid \
265 2: <raid_type> <#devices> <health_chars> \
266 3: <sync_ratio> <sync_action> <mismatch_cnt>
267
268Line 1 is the standard output produced by device-mapper.
269
270Line 2 & 3 are produced by the raid target and are best explained by example::
271
272 0 1960893648 raid raid4 5 AAAAA 2/490221568 init 0
273
274Here we can see the RAID type is raid4, there are 5 devices - all of
275which are 'A'live, and the array is 2/490221568 complete with its initial
276recovery. Here is a fuller description of the individual fields:
277
278 =============== =========================================================
279 <raid_type> Same as the <raid_type> used to create the array.
280 <health_chars> One char for each device, indicating:
281
282 - 'A' = alive and in-sync
283 - 'a' = alive but not in-sync
284 - 'D' = dead/failed.
285 <sync_ratio> The ratio indicating how much of the array has undergone
286 the process described by 'sync_action'. If the
287 'sync_action' is "check" or "repair", then the process
288 of "resync" or "recover" can be considered complete.
289 <sync_action> One of the following possible states:
290
291 idle
292 - No synchronization action is being performed.
293 frozen
294 - The current action has been halted.
295 resync
296 - Array is undergoing its initial synchronization
297 or is resynchronizing after an unclean shutdown
298 (possibly aided by a bitmap).
299 recover
300 - A device in the array is being rebuilt or
301 replaced.
302 check
303 - A user-initiated full check of the array is
304 being performed. All blocks are read and
305 checked for consistency. The number of
306 discrepancies found are recorded in
307 <mismatch_cnt>. No changes are made to the
308 array by this action.
309 repair
310 - The same as "check", but discrepancies are
311 corrected.
312 reshape
313 - The array is undergoing a reshape.
314 <mismatch_cnt> The number of discrepancies found between mirror copies
315 in RAID1/10 or wrong parity values found in RAID4/5/6.
316 This value is valid only after a "check" of the array
317 is performed. A healthy array has a 'mismatch_cnt' of 0.
318 <data_offset> The current data offset to the start of the user data on
319 each component device of a raid set (see the respective
320 raid parameter to support out-of-place reshaping).
321 <journal_char> - 'A' - active write-through journal device.
322 - 'a' - active write-back journal device.
323 - 'D' - dead journal device.
324 - '-' - no journal device.
325 =============== =========================================================
326
327
328Message Interface
329-----------------
330The dm-raid target will accept certain actions through the 'message' interface.
331('man dmsetup' for more information on the message interface.) These actions
332include:
333
334 ========= ================================================
335 "idle" Halt the current sync action.
336 "frozen" Freeze the current sync action.
337 "resync" Initiate/continue a resync.
338 "recover" Initiate/continue a recover process.
339 "check" Initiate a check (i.e. a "scrub") of the array.
340 "repair" Initiate a repair of the array.
341 ========= ================================================
342
343
344Discard Support
345---------------
346The implementation of discard support among hardware vendors varies.
347When a block is discarded, some storage devices will return zeroes when
348the block is read. These devices set the 'discard_zeroes_data'
349attribute. Other devices will return random data. Confusingly, some
350devices that advertise 'discard_zeroes_data' will not reliably return
351zeroes when discarded blocks are read! Since RAID 4/5/6 uses blocks
352from a number of devices to calculate parity blocks and (for performance
353reasons) relies on 'discard_zeroes_data' being reliable, it is important
354that the devices be consistent. Blocks may be discarded in the middle
355of a RAID 4/5/6 stripe and if subsequent read results are not
356consistent, the parity blocks may be calculated differently at any time;
357making the parity blocks useless for redundancy. It is important to
358understand how your hardware behaves with discards if you are going to
359enable discards with RAID 4/5/6.
360
361Since the behavior of storage devices is unreliable in this respect,
362even when reporting 'discard_zeroes_data', by default RAID 4/5/6
363discard support is disabled -- this ensures data integrity at the
364expense of losing some performance.
365
366Storage devices that properly support 'discard_zeroes_data' are
367increasingly whitelisted in the kernel and can thus be trusted.
368
369For trusted devices, the following dm-raid module parameter can be set
370to safely enable discard support for RAID 4/5/6:
371
372 'devices_handle_discards_safely'
373
374
375Takeover/Reshape Support
376------------------------
377The target natively supports these two types of MDRAID conversions:
378
379o Takeover: Converts an array from one RAID level to another
380
381o Reshape: Changes the internal layout while maintaining the current RAID level
382
383Each operation is only valid under specific constraints imposed by the existing array's layout and configuration.
384
385
386Takeover:
387linear -> raid1 with N >= 2 mirrors
388raid0 -> raid4 (add dedicated parity device)
389raid0 -> raid5 (add dedicated parity device)
390raid0 -> raid10 with near layout and N >= 2 mirror groups (raid0 stripes have to become first member within mirror groups)
391raid1 -> linear
392raid1 -> raid5 with 2 mirrors
393raid4 -> raid5 w/ rotating parity
394raid5 with dedicated parity device -> raid4
395raid5 -> raid6 (with dedicated Q-syndrome)
396raid6 (with dedicated Q-syndrome) -> raid5
397raid10 with near layout and even number of disks -> raid0 (select any in-sync device from each mirror group)
398
399Reshape:
400linear: not possible
401raid0: not possible
402raid1: change number of mirrors
403raid4: add and remove stripes (minimum 3), change stripesize
404raid5: add and remove stripes (minimum 3, special case 2 for raid1 takeover), change rotating parity algorithms, change stripesize
405raid6: add and remove stripes (minimum 4), change rotating syndrome algorithms, change stripesize
406raid10 near: add stripes (minimum 4), change stripesize, no stripe removal possible, change to offset layout
407raid10 offset: add stripes, change stripesize, no stripe removal possible, change to near layout
408raid10 far: not possible
409
410Table line examples:
411
412### raid1 -> raid5
413#
414# 2 devices limitation in raid1.
415# raid5 personality is able to just map 2 like raid1.
416# Reshape after takeover to change to full raid5 layout
417
418 0 1960886272 raid raid1 3 0 region_size 2048 2 /dev/dm-0 /dev/dm-1 /dev/dm-2 /dev/dm-3
419
420# dm-0 and dm-2 are e.g. 4MiB large metadata devices, dm-1 and dm-3 have to be at least 1960886272 big.
421#
422# Table line to takeover to raid5
423
424 0 1960886272 raid raid5 3 0 region_size 2048 2 /dev/dm-0 /dev/dm-1 /dev/dm-2 /dev/dm-3
425
426# Add required out-of-place reshape space to the beginniong of the given 2 data devices,
427# allocate another metadata/data device tuple with the same sizes for the parity space
428# and zero the first 4K of the metadata device.
429#
430# Example table of the out-of-place reshape space addition for one data device, e.g. dm-1
431
432 0 8192 linear 8:0 0 1960903888 # <- must be free space segment
433 8192 1960886272 linear 8:0 0 2048 # previous data segment
434
435# Mapping table for e.g. raid5_rs reshape causing the size of the raid device to double-fold once the reshape finishes.
436# Check the status output (e.g. "dmsetup status $RaidDev") for progress.
437
438 0 $((2 * 1960886272)) raid raid5 7 0 region_size 2048 data_offset 8192 delta_disk 1 2 /dev/dm-0 /dev/dm-1 /dev/dm-2 /dev/dm-3
439
440
441Version History
442---------------
443
444::
445
446 1.0.0 Initial version. Support for RAID 4/5/6
447 1.1.0 Added support for RAID 1
448 1.2.0 Handle creation of arrays that contain failed devices.
449 1.3.0 Added support for RAID 10
450 1.3.1 Allow device replacement/rebuild for RAID 10
451 1.3.2 Fix/improve redundancy checking for RAID10
452 1.4.0 Non-functional change. Removes arg from mapping function.
453 1.4.1 RAID10 fix redundancy validation checks (commit 55ebbb5).
454 1.4.2 Add RAID10 "far" and "offset" algorithm support.
455 1.5.0 Add message interface to allow manipulation of the sync_action.
456 New status (STATUSTYPE_INFO) fields: sync_action and mismatch_cnt.
457 1.5.1 Add ability to restore transiently failed devices on resume.
458 1.5.2 'mismatch_cnt' is zero unless [last_]sync_action is "check".
459 1.6.0 Add discard support (and devices_handle_discard_safely module param).
460 1.7.0 Add support for MD RAID0 mappings.
461 1.8.0 Explicitly check for compatible flags in the superblock metadata
462 and reject to start the raid set if any are set by a newer
463 target version, thus avoiding data corruption on a raid set
464 with a reshape in progress.
465 1.9.0 Add support for RAID level takeover/reshape/region size
466 and set size reduction.
467 1.9.1 Fix activation of existing RAID 4/10 mapped devices
468 1.9.2 Don't emit '- -' on the status table line in case the constructor
469 fails reading a superblock. Correctly emit 'maj:min1 maj:min2' and
470 'D' on the status line. If '- -' is passed into the constructor, emit
471 '- -' on the table line and '-' as the status line health character.
472 1.10.0 Add support for raid4/5/6 journal device
473 1.10.1 Fix data corruption on reshape request
474 1.11.0 Fix table line argument order
475 (wrong raid10_copies/raid10_format sequence)
476 1.11.1 Add raid4/5/6 journal write-back support via journal_mode option
477 1.12.1 Fix for MD deadlock between mddev_suspend() and md_write_start() available
478 1.13.0 Fix dev_health status at end of "recover" (was 'a', now 'A')
479 1.13.1 Fix deadlock caused by early md_stop_writes(). Also fix size an
480 state races.
481 1.13.2 Fix raid redundancy validation and avoid keeping raid set frozen
482 1.14.0 Fix reshape race on small devices. Fix stripe adding reshape
483 deadlock/potential data corruption. Update superblock when
484 specific devices are requested via rebuild. Fix RAID leg
485 rebuild errors.
486 1.15.0 Fix size extensions not being synchronized in case of new MD bitmap
487 pages allocated; also fix those not occurring after previous reductions
488 1.15.1 Fix argument count and arguments for rebuild/write_mostly/journal_(dev|mode)
489 on the status line.