blob: 9a01d1e4c78302a0fef677e2c516f2bf3daeca70 [file] [log] [blame]
NeilBrown9d09e662011-01-13 20:00:02 +00001/*
2 * Copyright (C) 2010-2011 Neil Brown
3 * Copyright (C) 2010-2011 Red Hat, Inc. All rights reserved.
4 *
5 * This file is released under the GPL.
6 */
7
8#include <linux/slab.h>
Paul Gortmaker056075c2011-07-03 13:58:33 -04009#include <linux/module.h>
NeilBrown9d09e662011-01-13 20:00:02 +000010
11#include "md.h"
Jonathan Brassow32737272011-08-02 12:32:07 +010012#include "raid1.h"
NeilBrown9d09e662011-01-13 20:00:02 +000013#include "raid5.h"
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -050014#include "raid10.h"
NeilBrown9d09e662011-01-13 20:00:02 +000015#include "bitmap.h"
16
Alasdair G Kergon3e8dbb72011-08-02 12:32:03 +010017#include <linux/device-mapper.h>
18
NeilBrown9d09e662011-01-13 20:00:02 +000019#define DM_MSG_PREFIX "raid"
20
21/*
Jonathan Brassowb12d4372011-08-02 12:32:07 +010022 * The following flags are used by dm-raid.c to set up the array state.
23 * They must be cleared before md_run is called.
NeilBrown9d09e662011-01-13 20:00:02 +000024 */
Jonathan Brassowb12d4372011-08-02 12:32:07 +010025#define FirstUse 10 /* rdev flag */
NeilBrown9d09e662011-01-13 20:00:02 +000026
27struct raid_dev {
28 /*
29 * Two DM devices, one to hold metadata and one to hold the
30 * actual data/parity. The reason for this is to not confuse
31 * ti->len and give more flexibility in altering size and
32 * characteristics.
33 *
34 * While it is possible for this device to be associated
35 * with a different physical device than the data_dev, it
36 * is intended for it to be the same.
37 * |--------- Physical Device ---------|
38 * |- meta_dev -|------ data_dev ------|
39 */
40 struct dm_dev *meta_dev;
41 struct dm_dev *data_dev;
NeilBrown3cb03002011-10-11 16:45:26 +110042 struct md_rdev rdev;
NeilBrown9d09e662011-01-13 20:00:02 +000043};
44
45/*
46 * Flags for rs->print_flags field.
47 */
Jonathan Brassow13c87582011-08-02 12:32:03 +010048#define DMPF_SYNC 0x1
49#define DMPF_NOSYNC 0x2
50#define DMPF_REBUILD 0x4
51#define DMPF_DAEMON_SLEEP 0x8
52#define DMPF_MIN_RECOVERY_RATE 0x10
53#define DMPF_MAX_RECOVERY_RATE 0x20
54#define DMPF_MAX_WRITE_BEHIND 0x40
55#define DMPF_STRIPE_CACHE 0x80
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -050056#define DMPF_REGION_SIZE 0x100
57#define DMPF_RAID10_COPIES 0x200
58#define DMPF_RAID10_FORMAT 0x400
59
NeilBrown9d09e662011-01-13 20:00:02 +000060struct raid_set {
61 struct dm_target *ti;
62
Jonathan Brassow34f8ac6d2012-01-27 14:53:53 -060063 uint32_t bitmap_loaded;
64 uint32_t print_flags;
NeilBrown9d09e662011-01-13 20:00:02 +000065
NeilBrownfd01b882011-10-11 16:47:53 +110066 struct mddev md;
NeilBrown9d09e662011-01-13 20:00:02 +000067 struct raid_type *raid_type;
68 struct dm_target_callbacks callbacks;
69
70 struct raid_dev dev[0];
71};
72
73/* Supported raid types and properties. */
74static struct raid_type {
75 const char *name; /* RAID algorithm. */
76 const char *descr; /* Descriptor text for logging. */
77 const unsigned parity_devs; /* # of parity devices. */
78 const unsigned minimal_devs; /* minimal # of devices in set. */
79 const unsigned level; /* RAID level. */
80 const unsigned algorithm; /* RAID algorithm. */
81} raid_types[] = {
Jonathan Brassow32737272011-08-02 12:32:07 +010082 {"raid1", "RAID1 (mirroring)", 0, 2, 1, 0 /* NONE */},
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -050083 {"raid10", "RAID10 (striped mirrors)", 0, 2, 10, UINT_MAX /* Varies */},
NeilBrown9d09e662011-01-13 20:00:02 +000084 {"raid4", "RAID4 (dedicated parity disk)", 1, 2, 5, ALGORITHM_PARITY_0},
85 {"raid5_la", "RAID5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
86 {"raid5_ra", "RAID5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
87 {"raid5_ls", "RAID5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
88 {"raid5_rs", "RAID5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
89 {"raid6_zr", "RAID6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
90 {"raid6_nr", "RAID6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
91 {"raid6_nc", "RAID6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE}
92};
93
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -050094static unsigned raid10_md_layout_to_copies(int layout)
95{
96 return layout & 0xFF;
97}
98
99static int raid10_format_to_md_layout(char *format, unsigned copies)
100{
101 /* 1 "far" copy, and 'copies' "near" copies */
102 return (1 << 8) | (copies & 0xFF);
103}
104
NeilBrown9d09e662011-01-13 20:00:02 +0000105static struct raid_type *get_raid_type(char *name)
106{
107 int i;
108
109 for (i = 0; i < ARRAY_SIZE(raid_types); i++)
110 if (!strcmp(raid_types[i].name, name))
111 return &raid_types[i];
112
113 return NULL;
114}
115
116static struct raid_set *context_alloc(struct dm_target *ti, struct raid_type *raid_type, unsigned raid_devs)
117{
118 unsigned i;
119 struct raid_set *rs;
NeilBrown9d09e662011-01-13 20:00:02 +0000120
121 if (raid_devs <= raid_type->parity_devs) {
122 ti->error = "Insufficient number of devices";
123 return ERR_PTR(-EINVAL);
124 }
125
NeilBrown9d09e662011-01-13 20:00:02 +0000126 rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
127 if (!rs) {
128 ti->error = "Cannot allocate raid context";
129 return ERR_PTR(-ENOMEM);
130 }
131
132 mddev_init(&rs->md);
133
134 rs->ti = ti;
135 rs->raid_type = raid_type;
136 rs->md.raid_disks = raid_devs;
137 rs->md.level = raid_type->level;
138 rs->md.new_level = rs->md.level;
NeilBrown9d09e662011-01-13 20:00:02 +0000139 rs->md.layout = raid_type->algorithm;
140 rs->md.new_layout = rs->md.layout;
141 rs->md.delta_disks = 0;
142 rs->md.recovery_cp = 0;
143
144 for (i = 0; i < raid_devs; i++)
145 md_rdev_init(&rs->dev[i].rdev);
146
147 /*
148 * Remaining items to be initialized by further RAID params:
149 * rs->md.persistent
150 * rs->md.external
151 * rs->md.chunk_sectors
152 * rs->md.new_chunk_sectors
Jonathan E Brassowc039c332012-07-27 15:08:04 +0100153 * rs->md.dev_sectors
NeilBrown9d09e662011-01-13 20:00:02 +0000154 */
155
156 return rs;
157}
158
159static void context_free(struct raid_set *rs)
160{
161 int i;
162
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100163 for (i = 0; i < rs->md.raid_disks; i++) {
164 if (rs->dev[i].meta_dev)
165 dm_put_device(rs->ti, rs->dev[i].meta_dev);
NeilBrown545c8792012-05-22 13:54:30 +1000166 md_rdev_clear(&rs->dev[i].rdev);
NeilBrown9d09e662011-01-13 20:00:02 +0000167 if (rs->dev[i].data_dev)
168 dm_put_device(rs->ti, rs->dev[i].data_dev);
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100169 }
NeilBrown9d09e662011-01-13 20:00:02 +0000170
171 kfree(rs);
172}
173
174/*
175 * For every device we have two words
176 * <meta_dev>: meta device name or '-' if missing
177 * <data_dev>: data device name or '-' if missing
178 *
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100179 * The following are permitted:
180 * - -
181 * - <data_dev>
182 * <meta_dev> <data_dev>
183 *
184 * The following is not allowed:
185 * <meta_dev> -
186 *
187 * This code parses those words. If there is a failure,
188 * the caller must use context_free to unwind the operations.
NeilBrown9d09e662011-01-13 20:00:02 +0000189 */
190static int dev_parms(struct raid_set *rs, char **argv)
191{
192 int i;
193 int rebuild = 0;
194 int metadata_available = 0;
195 int ret = 0;
196
197 for (i = 0; i < rs->md.raid_disks; i++, argv += 2) {
198 rs->dev[i].rdev.raid_disk = i;
199
200 rs->dev[i].meta_dev = NULL;
201 rs->dev[i].data_dev = NULL;
202
203 /*
204 * There are no offsets, since there is a separate device
205 * for data and metadata.
206 */
207 rs->dev[i].rdev.data_offset = 0;
208 rs->dev[i].rdev.mddev = &rs->md;
209
210 if (strcmp(argv[0], "-")) {
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100211 ret = dm_get_device(rs->ti, argv[0],
212 dm_table_get_mode(rs->ti->table),
213 &rs->dev[i].meta_dev);
214 rs->ti->error = "RAID metadata device lookup failure";
215 if (ret)
216 return ret;
217
218 rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
219 if (!rs->dev[i].rdev.sb_page)
220 return -ENOMEM;
NeilBrown9d09e662011-01-13 20:00:02 +0000221 }
222
223 if (!strcmp(argv[1], "-")) {
224 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
225 (!rs->dev[i].rdev.recovery_offset)) {
226 rs->ti->error = "Drive designated for rebuild not specified";
227 return -EINVAL;
228 }
229
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100230 rs->ti->error = "No data device supplied with metadata device";
231 if (rs->dev[i].meta_dev)
232 return -EINVAL;
233
NeilBrown9d09e662011-01-13 20:00:02 +0000234 continue;
235 }
236
237 ret = dm_get_device(rs->ti, argv[1],
238 dm_table_get_mode(rs->ti->table),
239 &rs->dev[i].data_dev);
240 if (ret) {
241 rs->ti->error = "RAID device lookup failure";
242 return ret;
243 }
244
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100245 if (rs->dev[i].meta_dev) {
246 metadata_available = 1;
247 rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
248 }
NeilBrown9d09e662011-01-13 20:00:02 +0000249 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
250 list_add(&rs->dev[i].rdev.same_set, &rs->md.disks);
251 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
252 rebuild++;
253 }
254
255 if (metadata_available) {
256 rs->md.external = 0;
257 rs->md.persistent = 1;
258 rs->md.major_version = 2;
259 } else if (rebuild && !rs->md.recovery_cp) {
260 /*
261 * Without metadata, we will not be able to tell if the array
262 * is in-sync or not - we must assume it is not. Therefore,
263 * it is impossible to rebuild a drive.
264 *
265 * Even if there is metadata, the on-disk information may
266 * indicate that the array is not in-sync and it will then
267 * fail at that time.
268 *
269 * User could specify 'nosync' option if desperate.
270 */
271 DMERR("Unable to rebuild drive while array is not in-sync");
272 rs->ti->error = "RAID device lookup failure";
273 return -EINVAL;
274 }
275
276 return 0;
277}
278
279/*
Jonathan Brassowc1084562011-08-02 12:32:07 +0100280 * validate_region_size
281 * @rs
282 * @region_size: region size in sectors. If 0, pick a size (4MiB default).
283 *
284 * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
285 * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
286 *
287 * Returns: 0 on success, -EINVAL on failure.
288 */
289static int validate_region_size(struct raid_set *rs, unsigned long region_size)
290{
291 unsigned long min_region_size = rs->ti->len / (1 << 21);
292
293 if (!region_size) {
294 /*
295 * Choose a reasonable default. All figures in sectors.
296 */
297 if (min_region_size > (1 << 13)) {
Jonathan Brassow3a0f9aa2012-12-21 20:23:33 +0000298 /* If not a power of 2, make it the next power of 2 */
299 if (min_region_size & (min_region_size - 1))
300 region_size = 1 << fls(region_size);
Jonathan Brassowc1084562011-08-02 12:32:07 +0100301 DMINFO("Choosing default region size of %lu sectors",
302 region_size);
Jonathan Brassowc1084562011-08-02 12:32:07 +0100303 } else {
304 DMINFO("Choosing default region size of 4MiB");
305 region_size = 1 << 13; /* sectors */
306 }
307 } else {
308 /*
309 * Validate user-supplied value.
310 */
311 if (region_size > rs->ti->len) {
312 rs->ti->error = "Supplied region size is too large";
313 return -EINVAL;
314 }
315
316 if (region_size < min_region_size) {
317 DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
318 region_size, min_region_size);
319 rs->ti->error = "Supplied region size is too small";
320 return -EINVAL;
321 }
322
323 if (!is_power_of_2(region_size)) {
324 rs->ti->error = "Region size is not a power of 2";
325 return -EINVAL;
326 }
327
328 if (region_size < rs->md.chunk_sectors) {
329 rs->ti->error = "Region size is smaller than the chunk size";
330 return -EINVAL;
331 }
332 }
333
334 /*
335 * Convert sectors to bytes.
336 */
337 rs->md.bitmap_info.chunksize = (region_size << 9);
338
339 return 0;
340}
341
342/*
Jonathan Brassow55ebbb52013-01-22 21:42:18 -0600343 * validate_raid_redundancy
Jonathan Brassoweb649122012-10-11 13:40:09 +1100344 * @rs
345 *
Jonathan Brassow55ebbb52013-01-22 21:42:18 -0600346 * Determine if there are enough devices in the array that haven't
347 * failed (or are being rebuilt) to form a usable array.
Jonathan Brassoweb649122012-10-11 13:40:09 +1100348 *
349 * Returns: 0 on success, -EINVAL on failure.
350 */
Jonathan Brassow55ebbb52013-01-22 21:42:18 -0600351static int validate_raid_redundancy(struct raid_set *rs)
Jonathan Brassoweb649122012-10-11 13:40:09 +1100352{
353 unsigned i, rebuild_cnt = 0;
Jonathan Brassow4ec1e362012-10-11 13:40:24 +1100354 unsigned rebuilds_per_group, copies, d;
Jonathan Brassoweb649122012-10-11 13:40:09 +1100355
Jonathan Brassoweb649122012-10-11 13:40:09 +1100356 for (i = 0; i < rs->md.raid_disks; i++)
Jonathan Brassow55ebbb52013-01-22 21:42:18 -0600357 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
358 !rs->dev[i].rdev.sb_page)
Jonathan Brassoweb649122012-10-11 13:40:09 +1100359 rebuild_cnt++;
360
361 switch (rs->raid_type->level) {
362 case 1:
363 if (rebuild_cnt >= rs->md.raid_disks)
364 goto too_many;
365 break;
366 case 4:
367 case 5:
368 case 6:
369 if (rebuild_cnt > rs->raid_type->parity_devs)
370 goto too_many;
371 break;
372 case 10:
Jonathan Brassow4ec1e362012-10-11 13:40:24 +1100373 copies = raid10_md_layout_to_copies(rs->md.layout);
374 if (rebuild_cnt < copies)
375 break;
376
377 /*
378 * It is possible to have a higher rebuild count for RAID10,
379 * as long as the failed devices occur in different mirror
380 * groups (i.e. different stripes).
381 *
382 * Right now, we only allow for "near" copies. When other
383 * formats are added, we will have to check those too.
384 *
385 * When checking "near" format, make sure no adjacent devices
386 * have failed beyond what can be handled. In addition to the
387 * simple case where the number of devices is a multiple of the
388 * number of copies, we must also handle cases where the number
389 * of devices is not a multiple of the number of copies.
390 * E.g. dev1 dev2 dev3 dev4 dev5
391 * A A B B C
392 * C D D E E
393 */
Jonathan Brassow4ec1e362012-10-11 13:40:24 +1100394 for (i = 0; i < rs->md.raid_disks * copies; i++) {
Jonathan Brassow55ebbb52013-01-22 21:42:18 -0600395 if (!(i % copies))
396 rebuilds_per_group = 0;
Jonathan Brassow4ec1e362012-10-11 13:40:24 +1100397 d = i % rs->md.raid_disks;
Jonathan Brassow55ebbb52013-01-22 21:42:18 -0600398 if ((!rs->dev[d].rdev.sb_page ||
399 !test_bit(In_sync, &rs->dev[d].rdev.flags)) &&
Jonathan Brassow4ec1e362012-10-11 13:40:24 +1100400 (++rebuilds_per_group >= copies))
401 goto too_many;
Jonathan Brassow4ec1e362012-10-11 13:40:24 +1100402 }
403 break;
Jonathan Brassoweb649122012-10-11 13:40:09 +1100404 default:
Jonathan Brassow55ebbb52013-01-22 21:42:18 -0600405 if (rebuild_cnt)
406 return -EINVAL;
Jonathan Brassoweb649122012-10-11 13:40:09 +1100407 }
408
409 return 0;
410
411too_many:
Jonathan Brassoweb649122012-10-11 13:40:09 +1100412 return -EINVAL;
413}
414
415/*
NeilBrown9d09e662011-01-13 20:00:02 +0000416 * Possible arguments are...
NeilBrown9d09e662011-01-13 20:00:02 +0000417 * <chunk_size> [optional_args]
418 *
Jonathan Brassow32737272011-08-02 12:32:07 +0100419 * Argument definitions
420 * <chunk_size> The number of sectors per disk that
421 * will form the "stripe"
422 * [[no]sync] Force or prevent recovery of the
423 * entire array
NeilBrown9d09e662011-01-13 20:00:02 +0000424 * [rebuild <idx>] Rebuild the drive indicated by the index
Jonathan Brassow32737272011-08-02 12:32:07 +0100425 * [daemon_sleep <ms>] Time between bitmap daemon work to
426 * clear bits
NeilBrown9d09e662011-01-13 20:00:02 +0000427 * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
428 * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
Jonathan Brassow46bed2b2011-08-02 12:32:07 +0100429 * [write_mostly <idx>] Indicate a write mostly drive via index
NeilBrown9d09e662011-01-13 20:00:02 +0000430 * [max_write_behind <sectors>] See '-write-behind=' (man mdadm)
431 * [stripe_cache <sectors>] Stripe cache size for higher RAIDs
Jonathan Brassowc1084562011-08-02 12:32:07 +0100432 * [region_size <sectors>] Defines granularity of bitmap
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -0500433 *
434 * RAID10-only options:
435 * [raid10_copies <# copies>] Number of copies. (Default: 2)
436 * [raid10_format <near>] Layout algorithm. (Default: near)
NeilBrown9d09e662011-01-13 20:00:02 +0000437 */
438static int parse_raid_params(struct raid_set *rs, char **argv,
439 unsigned num_raid_params)
440{
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -0500441 char *raid10_format = "near";
442 unsigned raid10_copies = 2;
Jonathan Brassoweb649122012-10-11 13:40:09 +1100443 unsigned i;
Jonathan Brassowc1084562011-08-02 12:32:07 +0100444 unsigned long value, region_size = 0;
Jonathan E Brassowc039c332012-07-27 15:08:04 +0100445 sector_t sectors_per_dev = rs->ti->len;
Mike Snitzer542f9032012-07-27 15:08:00 +0100446 sector_t max_io_len;
NeilBrown9d09e662011-01-13 20:00:02 +0000447 char *key;
448
449 /*
450 * First, parse the in-order required arguments
Jonathan Brassow32737272011-08-02 12:32:07 +0100451 * "chunk_size" is the only argument of this type.
NeilBrown9d09e662011-01-13 20:00:02 +0000452 */
Jonathan Brassow32737272011-08-02 12:32:07 +0100453 if ((strict_strtoul(argv[0], 10, &value) < 0)) {
NeilBrown9d09e662011-01-13 20:00:02 +0000454 rs->ti->error = "Bad chunk size";
455 return -EINVAL;
Jonathan Brassow32737272011-08-02 12:32:07 +0100456 } else if (rs->raid_type->level == 1) {
457 if (value)
458 DMERR("Ignoring chunk size parameter for RAID 1");
459 value = 0;
460 } else if (!is_power_of_2(value)) {
461 rs->ti->error = "Chunk size must be a power of 2";
462 return -EINVAL;
463 } else if (value < 8) {
464 rs->ti->error = "Chunk size value is too small";
465 return -EINVAL;
NeilBrown9d09e662011-01-13 20:00:02 +0000466 }
467
468 rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
469 argv++;
470 num_raid_params--;
471
472 /*
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100473 * We set each individual device as In_sync with a completed
474 * 'recovery_offset'. If there has been a device failure or
475 * replacement then one of the following cases applies:
476 *
477 * 1) User specifies 'rebuild'.
478 * - Device is reset when param is read.
479 * 2) A new device is supplied.
480 * - No matching superblock found, resets device.
481 * 3) Device failure was transient and returns on reload.
482 * - Failure noticed, resets device for bitmap replay.
483 * 4) Device hadn't completed recovery after previous failure.
484 * - Superblock is read and overrides recovery_offset.
485 *
486 * What is found in the superblocks of the devices is always
487 * authoritative, unless 'rebuild' or '[no]sync' was specified.
488 */
489 for (i = 0; i < rs->md.raid_disks; i++) {
490 set_bit(In_sync, &rs->dev[i].rdev.flags);
491 rs->dev[i].rdev.recovery_offset = MaxSector;
492 }
493
494 /*
NeilBrown9d09e662011-01-13 20:00:02 +0000495 * Second, parse the unordered optional arguments
496 */
NeilBrown9d09e662011-01-13 20:00:02 +0000497 for (i = 0; i < num_raid_params; i++) {
Jonathan Brassow13c87582011-08-02 12:32:03 +0100498 if (!strcasecmp(argv[i], "nosync")) {
NeilBrown9d09e662011-01-13 20:00:02 +0000499 rs->md.recovery_cp = MaxSector;
500 rs->print_flags |= DMPF_NOSYNC;
NeilBrown9d09e662011-01-13 20:00:02 +0000501 continue;
502 }
Jonathan Brassow13c87582011-08-02 12:32:03 +0100503 if (!strcasecmp(argv[i], "sync")) {
NeilBrown9d09e662011-01-13 20:00:02 +0000504 rs->md.recovery_cp = 0;
505 rs->print_flags |= DMPF_SYNC;
NeilBrown9d09e662011-01-13 20:00:02 +0000506 continue;
507 }
508
509 /* The rest of the optional arguments come in key/value pairs */
510 if ((i + 1) >= num_raid_params) {
511 rs->ti->error = "Wrong number of raid parameters given";
512 return -EINVAL;
513 }
514
515 key = argv[i++];
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -0500516
517 /* Parameters that take a string value are checked here. */
518 if (!strcasecmp(key, "raid10_format")) {
519 if (rs->raid_type->level != 10) {
520 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
521 return -EINVAL;
522 }
523 if (strcmp("near", argv[i])) {
524 rs->ti->error = "Invalid 'raid10_format' value given";
525 return -EINVAL;
526 }
527 raid10_format = argv[i];
528 rs->print_flags |= DMPF_RAID10_FORMAT;
529 continue;
530 }
531
NeilBrown9d09e662011-01-13 20:00:02 +0000532 if (strict_strtoul(argv[i], 10, &value) < 0) {
533 rs->ti->error = "Bad numerical argument given in raid params";
534 return -EINVAL;
535 }
536
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -0500537 /* Parameters that take a numeric value are checked here */
Jonathan Brassow13c87582011-08-02 12:32:03 +0100538 if (!strcasecmp(key, "rebuild")) {
Jonathan Brassow73861992012-10-11 13:40:36 +1100539 if (value >= rs->md.raid_disks) {
NeilBrown9d09e662011-01-13 20:00:02 +0000540 rs->ti->error = "Invalid rebuild index given";
541 return -EINVAL;
542 }
543 clear_bit(In_sync, &rs->dev[value].rdev.flags);
544 rs->dev[value].rdev.recovery_offset = 0;
Jonathan Brassow13c87582011-08-02 12:32:03 +0100545 rs->print_flags |= DMPF_REBUILD;
Jonathan Brassow46bed2b2011-08-02 12:32:07 +0100546 } else if (!strcasecmp(key, "write_mostly")) {
547 if (rs->raid_type->level != 1) {
548 rs->ti->error = "write_mostly option is only valid for RAID1";
549 return -EINVAL;
550 }
Jonthan Brassow82324802011-09-25 23:26:19 +0100551 if (value >= rs->md.raid_disks) {
Jonathan Brassow46bed2b2011-08-02 12:32:07 +0100552 rs->ti->error = "Invalid write_mostly drive index given";
553 return -EINVAL;
554 }
555 set_bit(WriteMostly, &rs->dev[value].rdev.flags);
Jonathan Brassow13c87582011-08-02 12:32:03 +0100556 } else if (!strcasecmp(key, "max_write_behind")) {
Jonathan Brassow46bed2b2011-08-02 12:32:07 +0100557 if (rs->raid_type->level != 1) {
558 rs->ti->error = "max_write_behind option is only valid for RAID1";
559 return -EINVAL;
560 }
NeilBrown9d09e662011-01-13 20:00:02 +0000561 rs->print_flags |= DMPF_MAX_WRITE_BEHIND;
562
563 /*
564 * In device-mapper, we specify things in sectors, but
565 * MD records this value in kB
566 */
567 value /= 2;
568 if (value > COUNTER_MAX) {
569 rs->ti->error = "Max write-behind limit out of range";
570 return -EINVAL;
571 }
572 rs->md.bitmap_info.max_write_behind = value;
Jonathan Brassow13c87582011-08-02 12:32:03 +0100573 } else if (!strcasecmp(key, "daemon_sleep")) {
NeilBrown9d09e662011-01-13 20:00:02 +0000574 rs->print_flags |= DMPF_DAEMON_SLEEP;
575 if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
576 rs->ti->error = "daemon sleep period out of range";
577 return -EINVAL;
578 }
579 rs->md.bitmap_info.daemon_sleep = value;
Jonathan Brassow13c87582011-08-02 12:32:03 +0100580 } else if (!strcasecmp(key, "stripe_cache")) {
NeilBrown9d09e662011-01-13 20:00:02 +0000581 rs->print_flags |= DMPF_STRIPE_CACHE;
582
583 /*
584 * In device-mapper, we specify things in sectors, but
585 * MD records this value in kB
586 */
587 value /= 2;
588
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -0500589 if ((rs->raid_type->level != 5) &&
590 (rs->raid_type->level != 6)) {
NeilBrown9d09e662011-01-13 20:00:02 +0000591 rs->ti->error = "Inappropriate argument: stripe_cache";
592 return -EINVAL;
593 }
594 if (raid5_set_cache_size(&rs->md, (int)value)) {
595 rs->ti->error = "Bad stripe_cache size";
596 return -EINVAL;
597 }
Jonathan Brassow13c87582011-08-02 12:32:03 +0100598 } else if (!strcasecmp(key, "min_recovery_rate")) {
NeilBrown9d09e662011-01-13 20:00:02 +0000599 rs->print_flags |= DMPF_MIN_RECOVERY_RATE;
600 if (value > INT_MAX) {
601 rs->ti->error = "min_recovery_rate out of range";
602 return -EINVAL;
603 }
604 rs->md.sync_speed_min = (int)value;
Jonathan Brassow13c87582011-08-02 12:32:03 +0100605 } else if (!strcasecmp(key, "max_recovery_rate")) {
NeilBrown9d09e662011-01-13 20:00:02 +0000606 rs->print_flags |= DMPF_MAX_RECOVERY_RATE;
607 if (value > INT_MAX) {
608 rs->ti->error = "max_recovery_rate out of range";
609 return -EINVAL;
610 }
611 rs->md.sync_speed_max = (int)value;
Jonathan Brassowc1084562011-08-02 12:32:07 +0100612 } else if (!strcasecmp(key, "region_size")) {
613 rs->print_flags |= DMPF_REGION_SIZE;
614 region_size = value;
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -0500615 } else if (!strcasecmp(key, "raid10_copies") &&
616 (rs->raid_type->level == 10)) {
617 if ((value < 2) || (value > 0xFF)) {
618 rs->ti->error = "Bad value for 'raid10_copies'";
619 return -EINVAL;
620 }
621 rs->print_flags |= DMPF_RAID10_COPIES;
622 raid10_copies = value;
NeilBrown9d09e662011-01-13 20:00:02 +0000623 } else {
624 DMERR("Unable to parse RAID parameter: %s", key);
625 rs->ti->error = "Unable to parse RAID parameters";
626 return -EINVAL;
627 }
628 }
629
Jonathan Brassowc1084562011-08-02 12:32:07 +0100630 if (validate_region_size(rs, region_size))
631 return -EINVAL;
632
633 if (rs->md.chunk_sectors)
Mike Snitzer542f9032012-07-27 15:08:00 +0100634 max_io_len = rs->md.chunk_sectors;
Jonathan Brassowc1084562011-08-02 12:32:07 +0100635 else
Mike Snitzer542f9032012-07-27 15:08:00 +0100636 max_io_len = region_size;
Jonathan Brassowc1084562011-08-02 12:32:07 +0100637
Mike Snitzer542f9032012-07-27 15:08:00 +0100638 if (dm_set_target_max_io_len(rs->ti, max_io_len))
639 return -EINVAL;
Jonathan Brassow32737272011-08-02 12:32:07 +0100640
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -0500641 if (rs->raid_type->level == 10) {
642 if (raid10_copies > rs->md.raid_disks) {
643 rs->ti->error = "Not enough devices to satisfy specification";
644 return -EINVAL;
645 }
646
647 /* (Len * #mirrors) / #devices */
648 sectors_per_dev = rs->ti->len * raid10_copies;
649 sector_div(sectors_per_dev, rs->md.raid_disks);
650
651 rs->md.layout = raid10_format_to_md_layout(raid10_format,
652 raid10_copies);
653 rs->md.new_layout = rs->md.layout;
654 } else if ((rs->raid_type->level > 1) &&
655 sector_div(sectors_per_dev,
656 (rs->md.raid_disks - rs->raid_type->parity_devs))) {
Jonathan E Brassowc039c332012-07-27 15:08:04 +0100657 rs->ti->error = "Target length not divisible by number of data devices";
658 return -EINVAL;
659 }
660 rs->md.dev_sectors = sectors_per_dev;
661
NeilBrown9d09e662011-01-13 20:00:02 +0000662 /* Assume there are no metadata devices until the drives are parsed */
663 rs->md.persistent = 0;
664 rs->md.external = 1;
665
666 return 0;
667}
668
669static void do_table_event(struct work_struct *ws)
670{
671 struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
672
673 dm_table_event(rs->ti->table);
674}
675
676static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
677{
678 struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
679
Jonathan Brassow32737272011-08-02 12:32:07 +0100680 if (rs->raid_type->level == 1)
681 return md_raid1_congested(&rs->md, bits);
682
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -0500683 if (rs->raid_type->level == 10)
684 return md_raid10_congested(&rs->md, bits);
685
NeilBrown9d09e662011-01-13 20:00:02 +0000686 return md_raid5_congested(&rs->md, bits);
687}
688
NeilBrown9d09e662011-01-13 20:00:02 +0000689/*
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100690 * This structure is never routinely used by userspace, unlike md superblocks.
691 * Devices with this superblock should only ever be accessed via device-mapper.
692 */
693#define DM_RAID_MAGIC 0x64526D44
694struct dm_raid_superblock {
695 __le32 magic; /* "DmRd" */
696 __le32 features; /* Used to indicate possible future changes */
697
698 __le32 num_devices; /* Number of devices in this array. (Max 64) */
699 __le32 array_position; /* The position of this drive in the array */
700
701 __le64 events; /* Incremented by md when superblock updated */
702 __le64 failed_devices; /* Bit field of devices to indicate failures */
703
704 /*
705 * This offset tracks the progress of the repair or replacement of
706 * an individual drive.
707 */
708 __le64 disk_recovery_offset;
709
710 /*
711 * This offset tracks the progress of the initial array
712 * synchronisation/parity calculation.
713 */
714 __le64 array_resync_offset;
715
716 /*
717 * RAID characteristics
718 */
719 __le32 level;
720 __le32 layout;
721 __le32 stripe_sectors;
722
723 __u8 pad[452]; /* Round struct to 512 bytes. */
724 /* Always set to 0 when writing. */
725} __packed;
726
NeilBrown3cb03002011-10-11 16:45:26 +1100727static int read_disk_sb(struct md_rdev *rdev, int size)
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100728{
729 BUG_ON(!rdev->sb_page);
730
731 if (rdev->sb_loaded)
732 return 0;
733
734 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, 1)) {
Jonathan E Brassow04475682012-03-28 18:41:26 +0100735 DMERR("Failed to read superblock of device at position %d",
736 rdev->raid_disk);
Jonathan Brassowc32fb9e2012-05-22 13:55:31 +1000737 md_error(rdev->mddev, rdev);
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100738 return -EINVAL;
739 }
740
741 rdev->sb_loaded = 1;
742
743 return 0;
744}
745
NeilBrownfd01b882011-10-11 16:47:53 +1100746static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100747{
Jonathan Brassow81f382f2012-05-22 13:55:30 +1000748 int i;
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100749 uint64_t failed_devices;
750 struct dm_raid_superblock *sb;
Jonathan Brassow81f382f2012-05-22 13:55:30 +1000751 struct raid_set *rs = container_of(mddev, struct raid_set, md);
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100752
753 sb = page_address(rdev->sb_page);
754 failed_devices = le64_to_cpu(sb->failed_devices);
755
Jonathan Brassow81f382f2012-05-22 13:55:30 +1000756 for (i = 0; i < mddev->raid_disks; i++)
757 if (!rs->dev[i].data_dev ||
758 test_bit(Faulty, &(rs->dev[i].rdev.flags)))
759 failed_devices |= (1ULL << i);
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100760
761 memset(sb, 0, sizeof(*sb));
762
763 sb->magic = cpu_to_le32(DM_RAID_MAGIC);
764 sb->features = cpu_to_le32(0); /* No features yet */
765
766 sb->num_devices = cpu_to_le32(mddev->raid_disks);
767 sb->array_position = cpu_to_le32(rdev->raid_disk);
768
769 sb->events = cpu_to_le64(mddev->events);
770 sb->failed_devices = cpu_to_le64(failed_devices);
771
772 sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
773 sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
774
775 sb->level = cpu_to_le32(mddev->level);
776 sb->layout = cpu_to_le32(mddev->layout);
777 sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
778}
779
780/*
781 * super_load
782 *
783 * This function creates a superblock if one is not found on the device
784 * and will decide which superblock to use if there's a choice.
785 *
786 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
787 */
NeilBrown3cb03002011-10-11 16:45:26 +1100788static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100789{
790 int ret;
791 struct dm_raid_superblock *sb;
792 struct dm_raid_superblock *refsb;
793 uint64_t events_sb, events_refsb;
794
795 rdev->sb_start = 0;
796 rdev->sb_size = sizeof(*sb);
797
798 ret = read_disk_sb(rdev, rdev->sb_size);
799 if (ret)
800 return ret;
801
802 sb = page_address(rdev->sb_page);
Jonathan E Brassow3aa3b2b2012-03-07 19:09:47 +0000803
804 /*
805 * Two cases that we want to write new superblocks and rebuild:
806 * 1) New device (no matching magic number)
807 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
808 */
809 if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
810 (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100811 super_sync(rdev->mddev, rdev);
812
813 set_bit(FirstUse, &rdev->flags);
814
815 /* Force writing of superblocks to disk */
816 set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags);
817
818 /* Any superblock is better than none, choose that if given */
819 return refdev ? 0 : 1;
820 }
821
822 if (!refdev)
823 return 1;
824
825 events_sb = le64_to_cpu(sb->events);
826
827 refsb = page_address(refdev->sb_page);
828 events_refsb = le64_to_cpu(refsb->events);
829
830 return (events_sb > events_refsb) ? 1 : 0;
831}
832
NeilBrownfd01b882011-10-11 16:47:53 +1100833static int super_init_validation(struct mddev *mddev, struct md_rdev *rdev)
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100834{
835 int role;
836 struct raid_set *rs = container_of(mddev, struct raid_set, md);
837 uint64_t events_sb;
838 uint64_t failed_devices;
839 struct dm_raid_superblock *sb;
840 uint32_t new_devs = 0;
841 uint32_t rebuilds = 0;
NeilBrowndafb20f2012-03-19 12:46:39 +1100842 struct md_rdev *r;
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100843 struct dm_raid_superblock *sb2;
844
845 sb = page_address(rdev->sb_page);
846 events_sb = le64_to_cpu(sb->events);
847 failed_devices = le64_to_cpu(sb->failed_devices);
848
849 /*
850 * Initialise to 1 if this is a new superblock.
851 */
852 mddev->events = events_sb ? : 1;
853
854 /*
855 * Reshaping is not currently allowed
856 */
857 if ((le32_to_cpu(sb->level) != mddev->level) ||
858 (le32_to_cpu(sb->layout) != mddev->layout) ||
859 (le32_to_cpu(sb->stripe_sectors) != mddev->chunk_sectors)) {
860 DMERR("Reshaping arrays not yet supported.");
861 return -EINVAL;
862 }
863
864 /* We can only change the number of devices in RAID1 right now */
865 if ((rs->raid_type->level != 1) &&
866 (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) {
867 DMERR("Reshaping arrays not yet supported.");
868 return -EINVAL;
869 }
870
871 if (!(rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC)))
872 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
873
874 /*
875 * During load, we set FirstUse if a new superblock was written.
876 * There are two reasons we might not have a superblock:
877 * 1) The array is brand new - in which case, all of the
878 * devices must have their In_sync bit set. Also,
879 * recovery_cp must be 0, unless forced.
880 * 2) This is a new device being added to an old array
881 * and the new device needs to be rebuilt - in which
882 * case the In_sync bit will /not/ be set and
883 * recovery_cp must be MaxSector.
884 */
NeilBrowndafb20f2012-03-19 12:46:39 +1100885 rdev_for_each(r, mddev) {
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100886 if (!test_bit(In_sync, &r->flags)) {
Jonathan E Brassow3aa3b2b2012-03-07 19:09:47 +0000887 DMINFO("Device %d specified for rebuild: "
888 "Clearing superblock", r->raid_disk);
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100889 rebuilds++;
890 } else if (test_bit(FirstUse, &r->flags))
891 new_devs++;
892 }
893
894 if (!rebuilds) {
895 if (new_devs == mddev->raid_disks) {
896 DMINFO("Superblocks created for new array");
897 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
898 } else if (new_devs) {
899 DMERR("New device injected "
900 "into existing array without 'rebuild' "
901 "parameter specified");
902 return -EINVAL;
903 }
904 } else if (new_devs) {
905 DMERR("'rebuild' devices cannot be "
906 "injected into an array with other first-time devices");
907 return -EINVAL;
908 } else if (mddev->recovery_cp != MaxSector) {
909 DMERR("'rebuild' specified while array is not in-sync");
910 return -EINVAL;
911 }
912
913 /*
914 * Now we set the Faulty bit for those devices that are
915 * recorded in the superblock as failed.
916 */
NeilBrowndafb20f2012-03-19 12:46:39 +1100917 rdev_for_each(r, mddev) {
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100918 if (!r->sb_page)
919 continue;
920 sb2 = page_address(r->sb_page);
921 sb2->failed_devices = 0;
922
923 /*
924 * Check for any device re-ordering.
925 */
926 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
927 role = le32_to_cpu(sb2->array_position);
928 if (role != r->raid_disk) {
929 if (rs->raid_type->level != 1) {
930 rs->ti->error = "Cannot change device "
931 "positions in RAID array";
932 return -EINVAL;
933 }
934 DMINFO("RAID1 device #%d now at position #%d",
935 role, r->raid_disk);
936 }
937
938 /*
939 * Partial recovery is performed on
940 * returning failed devices.
941 */
942 if (failed_devices & (1 << role))
943 set_bit(Faulty, &r->flags);
944 }
945 }
946
947 return 0;
948}
949
NeilBrownfd01b882011-10-11 16:47:53 +1100950static int super_validate(struct mddev *mddev, struct md_rdev *rdev)
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100951{
952 struct dm_raid_superblock *sb = page_address(rdev->sb_page);
953
954 /*
955 * If mddev->events is not set, we know we have not yet initialized
956 * the array.
957 */
958 if (!mddev->events && super_init_validation(mddev, rdev))
959 return -EINVAL;
960
961 mddev->bitmap_info.offset = 4096 >> 9; /* Enable bitmap creation */
962 rdev->mddev->bitmap_info.default_offset = 4096 >> 9;
963 if (!test_bit(FirstUse, &rdev->flags)) {
964 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
965 if (rdev->recovery_offset != MaxSector)
966 clear_bit(In_sync, &rdev->flags);
967 }
968
969 /*
970 * If a device comes back, set it as not In_sync and no longer faulty.
971 */
972 if (test_bit(Faulty, &rdev->flags)) {
973 clear_bit(Faulty, &rdev->flags);
974 clear_bit(In_sync, &rdev->flags);
975 rdev->saved_raid_disk = rdev->raid_disk;
976 rdev->recovery_offset = 0;
977 }
978
979 clear_bit(FirstUse, &rdev->flags);
980
981 return 0;
982}
983
984/*
985 * Analyse superblocks and select the freshest.
986 */
987static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
988{
989 int ret;
Jonathan E Brassow04475682012-03-28 18:41:26 +0100990 struct raid_dev *dev;
Jonathan Brassowa9ad8522012-04-24 10:23:13 +1000991 struct md_rdev *rdev, *tmp, *freshest;
NeilBrownfd01b882011-10-11 16:47:53 +1100992 struct mddev *mddev = &rs->md;
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100993
994 freshest = NULL;
Jonathan Brassowa9ad8522012-04-24 10:23:13 +1000995 rdev_for_each_safe(rdev, tmp, mddev) {
Jonathan Brassow761becf2012-10-11 13:42:19 +1100996 /*
997 * Skipping super_load due to DMPF_SYNC will cause
998 * the array to undergo initialization again as
999 * though it were new. This is the intended effect
1000 * of the "sync" directive.
1001 *
1002 * When reshaping capability is added, we must ensure
1003 * that the "sync" directive is disallowed during the
1004 * reshape.
1005 */
1006 if (rs->print_flags & DMPF_SYNC)
1007 continue;
1008
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001009 if (!rdev->meta_bdev)
1010 continue;
1011
1012 ret = super_load(rdev, freshest);
1013
1014 switch (ret) {
1015 case 1:
1016 freshest = rdev;
1017 break;
1018 case 0:
1019 break;
1020 default:
Jonathan E Brassow04475682012-03-28 18:41:26 +01001021 dev = container_of(rdev, struct raid_dev, rdev);
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06001022 if (dev->meta_dev)
1023 dm_put_device(ti, dev->meta_dev);
Jonathan E Brassow04475682012-03-28 18:41:26 +01001024
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06001025 dev->meta_dev = NULL;
1026 rdev->meta_bdev = NULL;
Jonathan E Brassow04475682012-03-28 18:41:26 +01001027
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06001028 if (rdev->sb_page)
1029 put_page(rdev->sb_page);
Jonathan E Brassow04475682012-03-28 18:41:26 +01001030
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06001031 rdev->sb_page = NULL;
Jonathan E Brassow04475682012-03-28 18:41:26 +01001032
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06001033 rdev->sb_loaded = 0;
Jonathan E Brassow04475682012-03-28 18:41:26 +01001034
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06001035 /*
1036 * We might be able to salvage the data device
1037 * even though the meta device has failed. For
1038 * now, we behave as though '- -' had been
1039 * set for this device in the table.
1040 */
1041 if (dev->data_dev)
1042 dm_put_device(ti, dev->data_dev);
Jonathan E Brassow04475682012-03-28 18:41:26 +01001043
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06001044 dev->data_dev = NULL;
1045 rdev->bdev = NULL;
Jonathan E Brassow04475682012-03-28 18:41:26 +01001046
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06001047 list_del(&rdev->same_set);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001048 }
1049 }
1050
1051 if (!freshest)
1052 return 0;
1053
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06001054 if (validate_raid_redundancy(rs)) {
1055 rs->ti->error = "Insufficient redundancy to activate array";
1056 return -EINVAL;
1057 }
1058
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001059 /*
1060 * Validation of the freshest device provides the source of
1061 * validation for the remaining devices.
1062 */
1063 ti->error = "Unable to assemble array: Invalid superblocks";
1064 if (super_validate(mddev, freshest))
1065 return -EINVAL;
1066
NeilBrowndafb20f2012-03-19 12:46:39 +11001067 rdev_for_each(rdev, mddev)
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001068 if ((rdev != freshest) && super_validate(mddev, rdev))
1069 return -EINVAL;
1070
1071 return 0;
1072}
1073
1074/*
NeilBrown9d09e662011-01-13 20:00:02 +00001075 * Construct a RAID4/5/6 mapping:
1076 * Args:
1077 * <raid_type> <#raid_params> <raid_params> \
1078 * <#raid_devs> { <meta_dev1> <dev1> .. <meta_devN> <devN> }
1079 *
NeilBrown9d09e662011-01-13 20:00:02 +00001080 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
1081 * details on possible <raid_params>.
1082 */
1083static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv)
1084{
1085 int ret;
1086 struct raid_type *rt;
1087 unsigned long num_raid_params, num_raid_devs;
1088 struct raid_set *rs = NULL;
1089
1090 /* Must have at least <raid_type> <#raid_params> */
1091 if (argc < 2) {
1092 ti->error = "Too few arguments";
1093 return -EINVAL;
1094 }
1095
1096 /* raid type */
1097 rt = get_raid_type(argv[0]);
1098 if (!rt) {
1099 ti->error = "Unrecognised raid_type";
1100 return -EINVAL;
1101 }
1102 argc--;
1103 argv++;
1104
1105 /* number of RAID parameters */
1106 if (strict_strtoul(argv[0], 10, &num_raid_params) < 0) {
1107 ti->error = "Cannot understand number of RAID parameters";
1108 return -EINVAL;
1109 }
1110 argc--;
1111 argv++;
1112
1113 /* Skip over RAID params for now and find out # of devices */
1114 if (num_raid_params + 1 > argc) {
1115 ti->error = "Arguments do not agree with counts given";
1116 return -EINVAL;
1117 }
1118
1119 if ((strict_strtoul(argv[num_raid_params], 10, &num_raid_devs) < 0) ||
1120 (num_raid_devs >= INT_MAX)) {
1121 ti->error = "Cannot understand number of raid devices";
1122 return -EINVAL;
1123 }
1124
1125 rs = context_alloc(ti, rt, (unsigned)num_raid_devs);
1126 if (IS_ERR(rs))
1127 return PTR_ERR(rs);
1128
1129 ret = parse_raid_params(rs, argv, (unsigned)num_raid_params);
1130 if (ret)
1131 goto bad;
1132
1133 ret = -EINVAL;
1134
1135 argc -= num_raid_params + 1; /* +1: we already have num_raid_devs */
1136 argv += num_raid_params + 1;
1137
1138 if (argc != (num_raid_devs * 2)) {
1139 ti->error = "Supplied RAID devices does not match the count given";
1140 goto bad;
1141 }
1142
1143 ret = dev_parms(rs, argv);
1144 if (ret)
1145 goto bad;
1146
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001147 rs->md.sync_super = super_sync;
1148 ret = analyse_superblocks(ti, rs);
1149 if (ret)
1150 goto bad;
1151
NeilBrown9d09e662011-01-13 20:00:02 +00001152 INIT_WORK(&rs->md.event_work, do_table_event);
NeilBrown9d09e662011-01-13 20:00:02 +00001153 ti->private = rs;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00001154 ti->num_flush_bios = 1;
NeilBrown9d09e662011-01-13 20:00:02 +00001155
1156 mutex_lock(&rs->md.reconfig_mutex);
1157 ret = md_run(&rs->md);
1158 rs->md.in_sync = 0; /* Assume already marked dirty */
1159 mutex_unlock(&rs->md.reconfig_mutex);
1160
1161 if (ret) {
1162 ti->error = "Fail to run raid array";
1163 goto bad;
1164 }
1165
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -05001166 if (ti->len != rs->md.array_sectors) {
1167 ti->error = "Array size does not match requested target length";
1168 ret = -EINVAL;
1169 goto size_mismatch;
1170 }
NeilBrown9d09e662011-01-13 20:00:02 +00001171 rs->callbacks.congested_fn = raid_is_congested;
NeilBrown9d09e662011-01-13 20:00:02 +00001172 dm_table_add_target_callbacks(ti->table, &rs->callbacks);
1173
Jonathan Brassow32737272011-08-02 12:32:07 +01001174 mddev_suspend(&rs->md);
NeilBrown9d09e662011-01-13 20:00:02 +00001175 return 0;
1176
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -05001177size_mismatch:
1178 md_stop(&rs->md);
NeilBrown9d09e662011-01-13 20:00:02 +00001179bad:
1180 context_free(rs);
1181
1182 return ret;
1183}
1184
1185static void raid_dtr(struct dm_target *ti)
1186{
1187 struct raid_set *rs = ti->private;
1188
1189 list_del_init(&rs->callbacks.list);
1190 md_stop(&rs->md);
1191 context_free(rs);
1192}
1193
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00001194static int raid_map(struct dm_target *ti, struct bio *bio)
NeilBrown9d09e662011-01-13 20:00:02 +00001195{
1196 struct raid_set *rs = ti->private;
NeilBrownfd01b882011-10-11 16:47:53 +11001197 struct mddev *mddev = &rs->md;
NeilBrown9d09e662011-01-13 20:00:02 +00001198
1199 mddev->pers->make_request(mddev, bio);
1200
1201 return DM_MAPIO_SUBMITTED;
1202}
1203
Mikulas Patockafd7c0922013-03-01 22:45:44 +00001204static void raid_status(struct dm_target *ti, status_type_t type,
1205 unsigned status_flags, char *result, unsigned maxlen)
NeilBrown9d09e662011-01-13 20:00:02 +00001206{
1207 struct raid_set *rs = ti->private;
1208 unsigned raid_param_cnt = 1; /* at least 1 for chunksize */
1209 unsigned sz = 0;
Jonathan E Brassow2e727c32011-10-31 20:21:26 +00001210 int i, array_in_sync = 0;
NeilBrown9d09e662011-01-13 20:00:02 +00001211 sector_t sync;
1212
1213 switch (type) {
1214 case STATUSTYPE_INFO:
1215 DMEMIT("%s %d ", rs->raid_type->name, rs->md.raid_disks);
1216
NeilBrown9d09e662011-01-13 20:00:02 +00001217 if (test_bit(MD_RECOVERY_RUNNING, &rs->md.recovery))
1218 sync = rs->md.curr_resync_completed;
1219 else
1220 sync = rs->md.recovery_cp;
1221
Jonathan E Brassow2e727c32011-10-31 20:21:26 +00001222 if (sync >= rs->md.resync_max_sectors) {
1223 array_in_sync = 1;
NeilBrown9d09e662011-01-13 20:00:02 +00001224 sync = rs->md.resync_max_sectors;
Jonathan E Brassow2e727c32011-10-31 20:21:26 +00001225 } else {
1226 /*
1227 * The array may be doing an initial sync, or it may
1228 * be rebuilding individual components. If all the
1229 * devices are In_sync, then it is the array that is
1230 * being initialized.
1231 */
1232 for (i = 0; i < rs->md.raid_disks; i++)
1233 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
1234 array_in_sync = 1;
1235 }
1236 /*
1237 * Status characters:
1238 * 'D' = Dead/Failed device
1239 * 'a' = Alive but not in-sync
1240 * 'A' = Alive and in-sync
1241 */
1242 for (i = 0; i < rs->md.raid_disks; i++) {
1243 if (test_bit(Faulty, &rs->dev[i].rdev.flags))
1244 DMEMIT("D");
1245 else if (!array_in_sync ||
1246 !test_bit(In_sync, &rs->dev[i].rdev.flags))
1247 DMEMIT("a");
1248 else
1249 DMEMIT("A");
1250 }
NeilBrown9d09e662011-01-13 20:00:02 +00001251
Jonathan E Brassow2e727c32011-10-31 20:21:26 +00001252 /*
1253 * In-sync ratio:
1254 * The in-sync ratio shows the progress of:
1255 * - Initializing the array
1256 * - Rebuilding a subset of devices of the array
1257 * The user can distinguish between the two by referring
1258 * to the status characters.
1259 */
NeilBrown9d09e662011-01-13 20:00:02 +00001260 DMEMIT(" %llu/%llu",
1261 (unsigned long long) sync,
1262 (unsigned long long) rs->md.resync_max_sectors);
1263
1264 break;
1265 case STATUSTYPE_TABLE:
1266 /* The string you would use to construct this array */
Jonathan Brassow46bed2b2011-08-02 12:32:07 +01001267 for (i = 0; i < rs->md.raid_disks; i++) {
Jonathan Brassow13c87582011-08-02 12:32:03 +01001268 if ((rs->print_flags & DMPF_REBUILD) &&
1269 rs->dev[i].data_dev &&
NeilBrown9d09e662011-01-13 20:00:02 +00001270 !test_bit(In_sync, &rs->dev[i].rdev.flags))
Jonathan Brassow13c87582011-08-02 12:32:03 +01001271 raid_param_cnt += 2; /* for rebuilds */
Jonathan Brassow46bed2b2011-08-02 12:32:07 +01001272 if (rs->dev[i].data_dev &&
1273 test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1274 raid_param_cnt += 2;
1275 }
NeilBrown9d09e662011-01-13 20:00:02 +00001276
Jonathan Brassow34f8ac6d2012-01-27 14:53:53 -06001277 raid_param_cnt += (hweight32(rs->print_flags & ~DMPF_REBUILD) * 2);
NeilBrown9d09e662011-01-13 20:00:02 +00001278 if (rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC))
1279 raid_param_cnt--;
1280
1281 DMEMIT("%s %u %u", rs->raid_type->name,
1282 raid_param_cnt, rs->md.chunk_sectors);
1283
1284 if ((rs->print_flags & DMPF_SYNC) &&
1285 (rs->md.recovery_cp == MaxSector))
1286 DMEMIT(" sync");
1287 if (rs->print_flags & DMPF_NOSYNC)
1288 DMEMIT(" nosync");
1289
1290 for (i = 0; i < rs->md.raid_disks; i++)
Jonathan Brassow13c87582011-08-02 12:32:03 +01001291 if ((rs->print_flags & DMPF_REBUILD) &&
1292 rs->dev[i].data_dev &&
NeilBrown9d09e662011-01-13 20:00:02 +00001293 !test_bit(In_sync, &rs->dev[i].rdev.flags))
1294 DMEMIT(" rebuild %u", i);
1295
1296 if (rs->print_flags & DMPF_DAEMON_SLEEP)
1297 DMEMIT(" daemon_sleep %lu",
1298 rs->md.bitmap_info.daemon_sleep);
1299
1300 if (rs->print_flags & DMPF_MIN_RECOVERY_RATE)
1301 DMEMIT(" min_recovery_rate %d", rs->md.sync_speed_min);
1302
1303 if (rs->print_flags & DMPF_MAX_RECOVERY_RATE)
1304 DMEMIT(" max_recovery_rate %d", rs->md.sync_speed_max);
1305
Jonathan Brassow46bed2b2011-08-02 12:32:07 +01001306 for (i = 0; i < rs->md.raid_disks; i++)
1307 if (rs->dev[i].data_dev &&
1308 test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1309 DMEMIT(" write_mostly %u", i);
1310
NeilBrown9d09e662011-01-13 20:00:02 +00001311 if (rs->print_flags & DMPF_MAX_WRITE_BEHIND)
1312 DMEMIT(" max_write_behind %lu",
1313 rs->md.bitmap_info.max_write_behind);
1314
1315 if (rs->print_flags & DMPF_STRIPE_CACHE) {
NeilBrownd1688a62011-10-11 16:49:52 +11001316 struct r5conf *conf = rs->md.private;
NeilBrown9d09e662011-01-13 20:00:02 +00001317
1318 /* convert from kiB to sectors */
1319 DMEMIT(" stripe_cache %d",
1320 conf ? conf->max_nr_stripes * 2 : 0);
1321 }
1322
Jonathan Brassowc1084562011-08-02 12:32:07 +01001323 if (rs->print_flags & DMPF_REGION_SIZE)
1324 DMEMIT(" region_size %lu",
1325 rs->md.bitmap_info.chunksize >> 9);
1326
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -05001327 if (rs->print_flags & DMPF_RAID10_COPIES)
1328 DMEMIT(" raid10_copies %u",
1329 raid10_md_layout_to_copies(rs->md.layout));
1330
1331 if (rs->print_flags & DMPF_RAID10_FORMAT)
1332 DMEMIT(" raid10_format near");
1333
NeilBrown9d09e662011-01-13 20:00:02 +00001334 DMEMIT(" %d", rs->md.raid_disks);
1335 for (i = 0; i < rs->md.raid_disks; i++) {
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001336 if (rs->dev[i].meta_dev)
1337 DMEMIT(" %s", rs->dev[i].meta_dev->name);
1338 else
1339 DMEMIT(" -");
NeilBrown9d09e662011-01-13 20:00:02 +00001340
1341 if (rs->dev[i].data_dev)
1342 DMEMIT(" %s", rs->dev[i].data_dev->name);
1343 else
1344 DMEMIT(" -");
1345 }
1346 }
NeilBrown9d09e662011-01-13 20:00:02 +00001347}
1348
1349static int raid_iterate_devices(struct dm_target *ti, iterate_devices_callout_fn fn, void *data)
1350{
1351 struct raid_set *rs = ti->private;
1352 unsigned i;
1353 int ret = 0;
1354
1355 for (i = 0; !ret && i < rs->md.raid_disks; i++)
1356 if (rs->dev[i].data_dev)
1357 ret = fn(ti,
1358 rs->dev[i].data_dev,
1359 0, /* No offset on data devs */
1360 rs->md.dev_sectors,
1361 data);
1362
1363 return ret;
1364}
1365
1366static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
1367{
1368 struct raid_set *rs = ti->private;
1369 unsigned chunk_size = rs->md.chunk_sectors << 9;
NeilBrownd1688a62011-10-11 16:49:52 +11001370 struct r5conf *conf = rs->md.private;
NeilBrown9d09e662011-01-13 20:00:02 +00001371
1372 blk_limits_io_min(limits, chunk_size);
1373 blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded));
1374}
1375
1376static void raid_presuspend(struct dm_target *ti)
1377{
1378 struct raid_set *rs = ti->private;
1379
1380 md_stop_writes(&rs->md);
1381}
1382
1383static void raid_postsuspend(struct dm_target *ti)
1384{
1385 struct raid_set *rs = ti->private;
1386
1387 mddev_suspend(&rs->md);
1388}
1389
1390static void raid_resume(struct dm_target *ti)
1391{
1392 struct raid_set *rs = ti->private;
1393
Jonathan Brassow81f382f2012-05-22 13:55:30 +10001394 set_bit(MD_CHANGE_DEVS, &rs->md.flags);
Jonathan Brassow34f8ac6d2012-01-27 14:53:53 -06001395 if (!rs->bitmap_loaded) {
1396 bitmap_load(&rs->md);
1397 rs->bitmap_loaded = 1;
Jonathan Brassow47525e52012-05-22 13:55:29 +10001398 }
Jonathan Brassow34f8ac6d2012-01-27 14:53:53 -06001399
Jonathan Brassow47525e52012-05-22 13:55:29 +10001400 clear_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
NeilBrown9d09e662011-01-13 20:00:02 +00001401 mddev_resume(&rs->md);
1402}
1403
1404static struct target_type raid_target = {
1405 .name = "raid",
Mikulas Patockafd7c0922013-03-01 22:45:44 +00001406 .version = {1, 4, 2},
NeilBrown9d09e662011-01-13 20:00:02 +00001407 .module = THIS_MODULE,
1408 .ctr = raid_ctr,
1409 .dtr = raid_dtr,
1410 .map = raid_map,
1411 .status = raid_status,
1412 .iterate_devices = raid_iterate_devices,
1413 .io_hints = raid_io_hints,
1414 .presuspend = raid_presuspend,
1415 .postsuspend = raid_postsuspend,
1416 .resume = raid_resume,
1417};
1418
1419static int __init dm_raid_init(void)
1420{
1421 return dm_register_target(&raid_target);
1422}
1423
1424static void __exit dm_raid_exit(void)
1425{
1426 dm_unregister_target(&raid_target);
1427}
1428
1429module_init(dm_raid_init);
1430module_exit(dm_raid_exit);
1431
1432MODULE_DESCRIPTION(DM_NAME " raid4/5/6 target");
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -05001433MODULE_ALIAS("dm-raid1");
1434MODULE_ALIAS("dm-raid10");
NeilBrown9d09e662011-01-13 20:00:02 +00001435MODULE_ALIAS("dm-raid4");
1436MODULE_ALIAS("dm-raid5");
1437MODULE_ALIAS("dm-raid6");
1438MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
1439MODULE_LICENSE("GPL");