blob: c2907d836e4e563ba50834e2524ae91919ddbd81 [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"
NeilBrown9d09e662011-01-13 20:00:02 +000014#include "bitmap.h"
15
Alasdair G Kergon3e8dbb72011-08-02 12:32:03 +010016#include <linux/device-mapper.h>
17
NeilBrown9d09e662011-01-13 20:00:02 +000018#define DM_MSG_PREFIX "raid"
19
20/*
Jonathan Brassowb12d4372011-08-02 12:32:07 +010021 * The following flags are used by dm-raid.c to set up the array state.
22 * They must be cleared before md_run is called.
NeilBrown9d09e662011-01-13 20:00:02 +000023 */
Jonathan Brassowb12d4372011-08-02 12:32:07 +010024#define FirstUse 10 /* rdev flag */
NeilBrown9d09e662011-01-13 20:00:02 +000025
26struct raid_dev {
27 /*
28 * Two DM devices, one to hold metadata and one to hold the
29 * actual data/parity. The reason for this is to not confuse
30 * ti->len and give more flexibility in altering size and
31 * characteristics.
32 *
33 * While it is possible for this device to be associated
34 * with a different physical device than the data_dev, it
35 * is intended for it to be the same.
36 * |--------- Physical Device ---------|
37 * |- meta_dev -|------ data_dev ------|
38 */
39 struct dm_dev *meta_dev;
40 struct dm_dev *data_dev;
NeilBrown3cb03002011-10-11 16:45:26 +110041 struct md_rdev rdev;
NeilBrown9d09e662011-01-13 20:00:02 +000042};
43
44/*
45 * Flags for rs->print_flags field.
46 */
Jonathan Brassow13c87582011-08-02 12:32:03 +010047#define DMPF_SYNC 0x1
48#define DMPF_NOSYNC 0x2
49#define DMPF_REBUILD 0x4
50#define DMPF_DAEMON_SLEEP 0x8
51#define DMPF_MIN_RECOVERY_RATE 0x10
52#define DMPF_MAX_RECOVERY_RATE 0x20
53#define DMPF_MAX_WRITE_BEHIND 0x40
54#define DMPF_STRIPE_CACHE 0x80
Jonathan Brassowc1084562011-08-02 12:32:07 +010055#define DMPF_REGION_SIZE 0X100
NeilBrown9d09e662011-01-13 20:00:02 +000056struct raid_set {
57 struct dm_target *ti;
58
59 uint64_t print_flags;
60
NeilBrownfd01b882011-10-11 16:47:53 +110061 struct mddev md;
NeilBrown9d09e662011-01-13 20:00:02 +000062 struct raid_type *raid_type;
63 struct dm_target_callbacks callbacks;
64
65 struct raid_dev dev[0];
66};
67
68/* Supported raid types and properties. */
69static struct raid_type {
70 const char *name; /* RAID algorithm. */
71 const char *descr; /* Descriptor text for logging. */
72 const unsigned parity_devs; /* # of parity devices. */
73 const unsigned minimal_devs; /* minimal # of devices in set. */
74 const unsigned level; /* RAID level. */
75 const unsigned algorithm; /* RAID algorithm. */
76} raid_types[] = {
Jonathan Brassow32737272011-08-02 12:32:07 +010077 {"raid1", "RAID1 (mirroring)", 0, 2, 1, 0 /* NONE */},
NeilBrown9d09e662011-01-13 20:00:02 +000078 {"raid4", "RAID4 (dedicated parity disk)", 1, 2, 5, ALGORITHM_PARITY_0},
79 {"raid5_la", "RAID5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
80 {"raid5_ra", "RAID5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
81 {"raid5_ls", "RAID5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
82 {"raid5_rs", "RAID5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
83 {"raid6_zr", "RAID6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
84 {"raid6_nr", "RAID6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
85 {"raid6_nc", "RAID6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE}
86};
87
88static struct raid_type *get_raid_type(char *name)
89{
90 int i;
91
92 for (i = 0; i < ARRAY_SIZE(raid_types); i++)
93 if (!strcmp(raid_types[i].name, name))
94 return &raid_types[i];
95
96 return NULL;
97}
98
99static struct raid_set *context_alloc(struct dm_target *ti, struct raid_type *raid_type, unsigned raid_devs)
100{
101 unsigned i;
102 struct raid_set *rs;
103 sector_t sectors_per_dev;
104
105 if (raid_devs <= raid_type->parity_devs) {
106 ti->error = "Insufficient number of devices";
107 return ERR_PTR(-EINVAL);
108 }
109
110 sectors_per_dev = ti->len;
Jonathan Brassow32737272011-08-02 12:32:07 +0100111 if ((raid_type->level > 1) &&
112 sector_div(sectors_per_dev, (raid_devs - raid_type->parity_devs))) {
NeilBrown9d09e662011-01-13 20:00:02 +0000113 ti->error = "Target length not divisible by number of data devices";
114 return ERR_PTR(-EINVAL);
115 }
116
117 rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
118 if (!rs) {
119 ti->error = "Cannot allocate raid context";
120 return ERR_PTR(-ENOMEM);
121 }
122
123 mddev_init(&rs->md);
124
125 rs->ti = ti;
126 rs->raid_type = raid_type;
127 rs->md.raid_disks = raid_devs;
128 rs->md.level = raid_type->level;
129 rs->md.new_level = rs->md.level;
130 rs->md.dev_sectors = sectors_per_dev;
131 rs->md.layout = raid_type->algorithm;
132 rs->md.new_layout = rs->md.layout;
133 rs->md.delta_disks = 0;
134 rs->md.recovery_cp = 0;
135
136 for (i = 0; i < raid_devs; i++)
137 md_rdev_init(&rs->dev[i].rdev);
138
139 /*
140 * Remaining items to be initialized by further RAID params:
141 * rs->md.persistent
142 * rs->md.external
143 * rs->md.chunk_sectors
144 * rs->md.new_chunk_sectors
145 */
146
147 return rs;
148}
149
150static void context_free(struct raid_set *rs)
151{
152 int i;
153
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100154 for (i = 0; i < rs->md.raid_disks; i++) {
155 if (rs->dev[i].meta_dev)
156 dm_put_device(rs->ti, rs->dev[i].meta_dev);
157 if (rs->dev[i].rdev.sb_page)
158 put_page(rs->dev[i].rdev.sb_page);
159 rs->dev[i].rdev.sb_page = NULL;
160 rs->dev[i].rdev.sb_loaded = 0;
NeilBrown9d09e662011-01-13 20:00:02 +0000161 if (rs->dev[i].data_dev)
162 dm_put_device(rs->ti, rs->dev[i].data_dev);
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100163 }
NeilBrown9d09e662011-01-13 20:00:02 +0000164
165 kfree(rs);
166}
167
168/*
169 * For every device we have two words
170 * <meta_dev>: meta device name or '-' if missing
171 * <data_dev>: data device name or '-' if missing
172 *
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100173 * The following are permitted:
174 * - -
175 * - <data_dev>
176 * <meta_dev> <data_dev>
177 *
178 * The following is not allowed:
179 * <meta_dev> -
180 *
181 * This code parses those words. If there is a failure,
182 * the caller must use context_free to unwind the operations.
NeilBrown9d09e662011-01-13 20:00:02 +0000183 */
184static int dev_parms(struct raid_set *rs, char **argv)
185{
186 int i;
187 int rebuild = 0;
188 int metadata_available = 0;
189 int ret = 0;
190
191 for (i = 0; i < rs->md.raid_disks; i++, argv += 2) {
192 rs->dev[i].rdev.raid_disk = i;
193
194 rs->dev[i].meta_dev = NULL;
195 rs->dev[i].data_dev = NULL;
196
197 /*
198 * There are no offsets, since there is a separate device
199 * for data and metadata.
200 */
201 rs->dev[i].rdev.data_offset = 0;
202 rs->dev[i].rdev.mddev = &rs->md;
203
204 if (strcmp(argv[0], "-")) {
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100205 ret = dm_get_device(rs->ti, argv[0],
206 dm_table_get_mode(rs->ti->table),
207 &rs->dev[i].meta_dev);
208 rs->ti->error = "RAID metadata device lookup failure";
209 if (ret)
210 return ret;
211
212 rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
213 if (!rs->dev[i].rdev.sb_page)
214 return -ENOMEM;
NeilBrown9d09e662011-01-13 20:00:02 +0000215 }
216
217 if (!strcmp(argv[1], "-")) {
218 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
219 (!rs->dev[i].rdev.recovery_offset)) {
220 rs->ti->error = "Drive designated for rebuild not specified";
221 return -EINVAL;
222 }
223
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100224 rs->ti->error = "No data device supplied with metadata device";
225 if (rs->dev[i].meta_dev)
226 return -EINVAL;
227
NeilBrown9d09e662011-01-13 20:00:02 +0000228 continue;
229 }
230
231 ret = dm_get_device(rs->ti, argv[1],
232 dm_table_get_mode(rs->ti->table),
233 &rs->dev[i].data_dev);
234 if (ret) {
235 rs->ti->error = "RAID device lookup failure";
236 return ret;
237 }
238
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100239 if (rs->dev[i].meta_dev) {
240 metadata_available = 1;
241 rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
242 }
NeilBrown9d09e662011-01-13 20:00:02 +0000243 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
244 list_add(&rs->dev[i].rdev.same_set, &rs->md.disks);
245 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
246 rebuild++;
247 }
248
249 if (metadata_available) {
250 rs->md.external = 0;
251 rs->md.persistent = 1;
252 rs->md.major_version = 2;
253 } else if (rebuild && !rs->md.recovery_cp) {
254 /*
255 * Without metadata, we will not be able to tell if the array
256 * is in-sync or not - we must assume it is not. Therefore,
257 * it is impossible to rebuild a drive.
258 *
259 * Even if there is metadata, the on-disk information may
260 * indicate that the array is not in-sync and it will then
261 * fail at that time.
262 *
263 * User could specify 'nosync' option if desperate.
264 */
265 DMERR("Unable to rebuild drive while array is not in-sync");
266 rs->ti->error = "RAID device lookup failure";
267 return -EINVAL;
268 }
269
270 return 0;
271}
272
273/*
Jonathan Brassowc1084562011-08-02 12:32:07 +0100274 * validate_region_size
275 * @rs
276 * @region_size: region size in sectors. If 0, pick a size (4MiB default).
277 *
278 * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
279 * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
280 *
281 * Returns: 0 on success, -EINVAL on failure.
282 */
283static int validate_region_size(struct raid_set *rs, unsigned long region_size)
284{
285 unsigned long min_region_size = rs->ti->len / (1 << 21);
286
287 if (!region_size) {
288 /*
289 * Choose a reasonable default. All figures in sectors.
290 */
291 if (min_region_size > (1 << 13)) {
292 DMINFO("Choosing default region size of %lu sectors",
293 region_size);
294 region_size = min_region_size;
295 } else {
296 DMINFO("Choosing default region size of 4MiB");
297 region_size = 1 << 13; /* sectors */
298 }
299 } else {
300 /*
301 * Validate user-supplied value.
302 */
303 if (region_size > rs->ti->len) {
304 rs->ti->error = "Supplied region size is too large";
305 return -EINVAL;
306 }
307
308 if (region_size < min_region_size) {
309 DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
310 region_size, min_region_size);
311 rs->ti->error = "Supplied region size is too small";
312 return -EINVAL;
313 }
314
315 if (!is_power_of_2(region_size)) {
316 rs->ti->error = "Region size is not a power of 2";
317 return -EINVAL;
318 }
319
320 if (region_size < rs->md.chunk_sectors) {
321 rs->ti->error = "Region size is smaller than the chunk size";
322 return -EINVAL;
323 }
324 }
325
326 /*
327 * Convert sectors to bytes.
328 */
329 rs->md.bitmap_info.chunksize = (region_size << 9);
330
331 return 0;
332}
333
334/*
NeilBrown9d09e662011-01-13 20:00:02 +0000335 * Possible arguments are...
NeilBrown9d09e662011-01-13 20:00:02 +0000336 * <chunk_size> [optional_args]
337 *
Jonathan Brassow32737272011-08-02 12:32:07 +0100338 * Argument definitions
339 * <chunk_size> The number of sectors per disk that
340 * will form the "stripe"
341 * [[no]sync] Force or prevent recovery of the
342 * entire array
NeilBrown9d09e662011-01-13 20:00:02 +0000343 * [rebuild <idx>] Rebuild the drive indicated by the index
Jonathan Brassow32737272011-08-02 12:32:07 +0100344 * [daemon_sleep <ms>] Time between bitmap daemon work to
345 * clear bits
NeilBrown9d09e662011-01-13 20:00:02 +0000346 * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
347 * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
Jonathan Brassow46bed2b2011-08-02 12:32:07 +0100348 * [write_mostly <idx>] Indicate a write mostly drive via index
NeilBrown9d09e662011-01-13 20:00:02 +0000349 * [max_write_behind <sectors>] See '-write-behind=' (man mdadm)
350 * [stripe_cache <sectors>] Stripe cache size for higher RAIDs
Jonathan Brassowc1084562011-08-02 12:32:07 +0100351 * [region_size <sectors>] Defines granularity of bitmap
NeilBrown9d09e662011-01-13 20:00:02 +0000352 */
353static int parse_raid_params(struct raid_set *rs, char **argv,
354 unsigned num_raid_params)
355{
356 unsigned i, rebuild_cnt = 0;
Jonathan Brassowc1084562011-08-02 12:32:07 +0100357 unsigned long value, region_size = 0;
NeilBrown9d09e662011-01-13 20:00:02 +0000358 char *key;
359
360 /*
361 * First, parse the in-order required arguments
Jonathan Brassow32737272011-08-02 12:32:07 +0100362 * "chunk_size" is the only argument of this type.
NeilBrown9d09e662011-01-13 20:00:02 +0000363 */
Jonathan Brassow32737272011-08-02 12:32:07 +0100364 if ((strict_strtoul(argv[0], 10, &value) < 0)) {
NeilBrown9d09e662011-01-13 20:00:02 +0000365 rs->ti->error = "Bad chunk size";
366 return -EINVAL;
Jonathan Brassow32737272011-08-02 12:32:07 +0100367 } else if (rs->raid_type->level == 1) {
368 if (value)
369 DMERR("Ignoring chunk size parameter for RAID 1");
370 value = 0;
371 } else if (!is_power_of_2(value)) {
372 rs->ti->error = "Chunk size must be a power of 2";
373 return -EINVAL;
374 } else if (value < 8) {
375 rs->ti->error = "Chunk size value is too small";
376 return -EINVAL;
NeilBrown9d09e662011-01-13 20:00:02 +0000377 }
378
379 rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
380 argv++;
381 num_raid_params--;
382
383 /*
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100384 * We set each individual device as In_sync with a completed
385 * 'recovery_offset'. If there has been a device failure or
386 * replacement then one of the following cases applies:
387 *
388 * 1) User specifies 'rebuild'.
389 * - Device is reset when param is read.
390 * 2) A new device is supplied.
391 * - No matching superblock found, resets device.
392 * 3) Device failure was transient and returns on reload.
393 * - Failure noticed, resets device for bitmap replay.
394 * 4) Device hadn't completed recovery after previous failure.
395 * - Superblock is read and overrides recovery_offset.
396 *
397 * What is found in the superblocks of the devices is always
398 * authoritative, unless 'rebuild' or '[no]sync' was specified.
399 */
400 for (i = 0; i < rs->md.raid_disks; i++) {
401 set_bit(In_sync, &rs->dev[i].rdev.flags);
402 rs->dev[i].rdev.recovery_offset = MaxSector;
403 }
404
405 /*
NeilBrown9d09e662011-01-13 20:00:02 +0000406 * Second, parse the unordered optional arguments
407 */
NeilBrown9d09e662011-01-13 20:00:02 +0000408 for (i = 0; i < num_raid_params; i++) {
Jonathan Brassow13c87582011-08-02 12:32:03 +0100409 if (!strcasecmp(argv[i], "nosync")) {
NeilBrown9d09e662011-01-13 20:00:02 +0000410 rs->md.recovery_cp = MaxSector;
411 rs->print_flags |= DMPF_NOSYNC;
NeilBrown9d09e662011-01-13 20:00:02 +0000412 continue;
413 }
Jonathan Brassow13c87582011-08-02 12:32:03 +0100414 if (!strcasecmp(argv[i], "sync")) {
NeilBrown9d09e662011-01-13 20:00:02 +0000415 rs->md.recovery_cp = 0;
416 rs->print_flags |= DMPF_SYNC;
NeilBrown9d09e662011-01-13 20:00:02 +0000417 continue;
418 }
419
420 /* The rest of the optional arguments come in key/value pairs */
421 if ((i + 1) >= num_raid_params) {
422 rs->ti->error = "Wrong number of raid parameters given";
423 return -EINVAL;
424 }
425
426 key = argv[i++];
427 if (strict_strtoul(argv[i], 10, &value) < 0) {
428 rs->ti->error = "Bad numerical argument given in raid params";
429 return -EINVAL;
430 }
431
Jonathan Brassow13c87582011-08-02 12:32:03 +0100432 if (!strcasecmp(key, "rebuild")) {
Jonathan Brassow32737272011-08-02 12:32:07 +0100433 rebuild_cnt++;
434 if (((rs->raid_type->level != 1) &&
435 (rebuild_cnt > rs->raid_type->parity_devs)) ||
436 ((rs->raid_type->level == 1) &&
437 (rebuild_cnt > (rs->md.raid_disks - 1)))) {
438 rs->ti->error = "Too many rebuild devices specified for given RAID type";
NeilBrown9d09e662011-01-13 20:00:02 +0000439 return -EINVAL;
440 }
441 if (value > rs->md.raid_disks) {
442 rs->ti->error = "Invalid rebuild index given";
443 return -EINVAL;
444 }
445 clear_bit(In_sync, &rs->dev[value].rdev.flags);
446 rs->dev[value].rdev.recovery_offset = 0;
Jonathan Brassow13c87582011-08-02 12:32:03 +0100447 rs->print_flags |= DMPF_REBUILD;
Jonathan Brassow46bed2b2011-08-02 12:32:07 +0100448 } else if (!strcasecmp(key, "write_mostly")) {
449 if (rs->raid_type->level != 1) {
450 rs->ti->error = "write_mostly option is only valid for RAID1";
451 return -EINVAL;
452 }
Jonthan Brassow82324802011-09-25 23:26:19 +0100453 if (value >= rs->md.raid_disks) {
Jonathan Brassow46bed2b2011-08-02 12:32:07 +0100454 rs->ti->error = "Invalid write_mostly drive index given";
455 return -EINVAL;
456 }
457 set_bit(WriteMostly, &rs->dev[value].rdev.flags);
Jonathan Brassow13c87582011-08-02 12:32:03 +0100458 } else if (!strcasecmp(key, "max_write_behind")) {
Jonathan Brassow46bed2b2011-08-02 12:32:07 +0100459 if (rs->raid_type->level != 1) {
460 rs->ti->error = "max_write_behind option is only valid for RAID1";
461 return -EINVAL;
462 }
NeilBrown9d09e662011-01-13 20:00:02 +0000463 rs->print_flags |= DMPF_MAX_WRITE_BEHIND;
464
465 /*
466 * In device-mapper, we specify things in sectors, but
467 * MD records this value in kB
468 */
469 value /= 2;
470 if (value > COUNTER_MAX) {
471 rs->ti->error = "Max write-behind limit out of range";
472 return -EINVAL;
473 }
474 rs->md.bitmap_info.max_write_behind = value;
Jonathan Brassow13c87582011-08-02 12:32:03 +0100475 } else if (!strcasecmp(key, "daemon_sleep")) {
NeilBrown9d09e662011-01-13 20:00:02 +0000476 rs->print_flags |= DMPF_DAEMON_SLEEP;
477 if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
478 rs->ti->error = "daemon sleep period out of range";
479 return -EINVAL;
480 }
481 rs->md.bitmap_info.daemon_sleep = value;
Jonathan Brassow13c87582011-08-02 12:32:03 +0100482 } else if (!strcasecmp(key, "stripe_cache")) {
NeilBrown9d09e662011-01-13 20:00:02 +0000483 rs->print_flags |= DMPF_STRIPE_CACHE;
484
485 /*
486 * In device-mapper, we specify things in sectors, but
487 * MD records this value in kB
488 */
489 value /= 2;
490
491 if (rs->raid_type->level < 5) {
492 rs->ti->error = "Inappropriate argument: stripe_cache";
493 return -EINVAL;
494 }
495 if (raid5_set_cache_size(&rs->md, (int)value)) {
496 rs->ti->error = "Bad stripe_cache size";
497 return -EINVAL;
498 }
Jonathan Brassow13c87582011-08-02 12:32:03 +0100499 } else if (!strcasecmp(key, "min_recovery_rate")) {
NeilBrown9d09e662011-01-13 20:00:02 +0000500 rs->print_flags |= DMPF_MIN_RECOVERY_RATE;
501 if (value > INT_MAX) {
502 rs->ti->error = "min_recovery_rate out of range";
503 return -EINVAL;
504 }
505 rs->md.sync_speed_min = (int)value;
Jonathan Brassow13c87582011-08-02 12:32:03 +0100506 } else if (!strcasecmp(key, "max_recovery_rate")) {
NeilBrown9d09e662011-01-13 20:00:02 +0000507 rs->print_flags |= DMPF_MAX_RECOVERY_RATE;
508 if (value > INT_MAX) {
509 rs->ti->error = "max_recovery_rate out of range";
510 return -EINVAL;
511 }
512 rs->md.sync_speed_max = (int)value;
Jonathan Brassowc1084562011-08-02 12:32:07 +0100513 } else if (!strcasecmp(key, "region_size")) {
514 rs->print_flags |= DMPF_REGION_SIZE;
515 region_size = value;
NeilBrown9d09e662011-01-13 20:00:02 +0000516 } else {
517 DMERR("Unable to parse RAID parameter: %s", key);
518 rs->ti->error = "Unable to parse RAID parameters";
519 return -EINVAL;
520 }
521 }
522
Jonathan Brassowc1084562011-08-02 12:32:07 +0100523 if (validate_region_size(rs, region_size))
524 return -EINVAL;
525
526 if (rs->md.chunk_sectors)
527 rs->ti->split_io = rs->md.chunk_sectors;
528 else
529 rs->ti->split_io = region_size;
530
Jonathan Brassow32737272011-08-02 12:32:07 +0100531 if (rs->md.chunk_sectors)
532 rs->ti->split_io = rs->md.chunk_sectors;
533 else
534 rs->ti->split_io = region_size;
535
NeilBrown9d09e662011-01-13 20:00:02 +0000536 /* Assume there are no metadata devices until the drives are parsed */
537 rs->md.persistent = 0;
538 rs->md.external = 1;
539
540 return 0;
541}
542
543static void do_table_event(struct work_struct *ws)
544{
545 struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
546
547 dm_table_event(rs->ti->table);
548}
549
550static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
551{
552 struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
553
Jonathan Brassow32737272011-08-02 12:32:07 +0100554 if (rs->raid_type->level == 1)
555 return md_raid1_congested(&rs->md, bits);
556
NeilBrown9d09e662011-01-13 20:00:02 +0000557 return md_raid5_congested(&rs->md, bits);
558}
559
NeilBrown9d09e662011-01-13 20:00:02 +0000560/*
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100561 * This structure is never routinely used by userspace, unlike md superblocks.
562 * Devices with this superblock should only ever be accessed via device-mapper.
563 */
564#define DM_RAID_MAGIC 0x64526D44
565struct dm_raid_superblock {
566 __le32 magic; /* "DmRd" */
567 __le32 features; /* Used to indicate possible future changes */
568
569 __le32 num_devices; /* Number of devices in this array. (Max 64) */
570 __le32 array_position; /* The position of this drive in the array */
571
572 __le64 events; /* Incremented by md when superblock updated */
573 __le64 failed_devices; /* Bit field of devices to indicate failures */
574
575 /*
576 * This offset tracks the progress of the repair or replacement of
577 * an individual drive.
578 */
579 __le64 disk_recovery_offset;
580
581 /*
582 * This offset tracks the progress of the initial array
583 * synchronisation/parity calculation.
584 */
585 __le64 array_resync_offset;
586
587 /*
588 * RAID characteristics
589 */
590 __le32 level;
591 __le32 layout;
592 __le32 stripe_sectors;
593
594 __u8 pad[452]; /* Round struct to 512 bytes. */
595 /* Always set to 0 when writing. */
596} __packed;
597
NeilBrown3cb03002011-10-11 16:45:26 +1100598static int read_disk_sb(struct md_rdev *rdev, int size)
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100599{
600 BUG_ON(!rdev->sb_page);
601
602 if (rdev->sb_loaded)
603 return 0;
604
605 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, 1)) {
606 DMERR("Failed to read device superblock");
607 return -EINVAL;
608 }
609
610 rdev->sb_loaded = 1;
611
612 return 0;
613}
614
NeilBrownfd01b882011-10-11 16:47:53 +1100615static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100616{
NeilBrown3cb03002011-10-11 16:45:26 +1100617 struct md_rdev *r, *t;
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100618 uint64_t failed_devices;
619 struct dm_raid_superblock *sb;
620
621 sb = page_address(rdev->sb_page);
622 failed_devices = le64_to_cpu(sb->failed_devices);
623
624 rdev_for_each(r, t, mddev)
625 if ((r->raid_disk >= 0) && test_bit(Faulty, &r->flags))
626 failed_devices |= (1ULL << r->raid_disk);
627
628 memset(sb, 0, sizeof(*sb));
629
630 sb->magic = cpu_to_le32(DM_RAID_MAGIC);
631 sb->features = cpu_to_le32(0); /* No features yet */
632
633 sb->num_devices = cpu_to_le32(mddev->raid_disks);
634 sb->array_position = cpu_to_le32(rdev->raid_disk);
635
636 sb->events = cpu_to_le64(mddev->events);
637 sb->failed_devices = cpu_to_le64(failed_devices);
638
639 sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
640 sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
641
642 sb->level = cpu_to_le32(mddev->level);
643 sb->layout = cpu_to_le32(mddev->layout);
644 sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
645}
646
647/*
648 * super_load
649 *
650 * This function creates a superblock if one is not found on the device
651 * and will decide which superblock to use if there's a choice.
652 *
653 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
654 */
NeilBrown3cb03002011-10-11 16:45:26 +1100655static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100656{
657 int ret;
658 struct dm_raid_superblock *sb;
659 struct dm_raid_superblock *refsb;
660 uint64_t events_sb, events_refsb;
661
662 rdev->sb_start = 0;
663 rdev->sb_size = sizeof(*sb);
664
665 ret = read_disk_sb(rdev, rdev->sb_size);
666 if (ret)
667 return ret;
668
669 sb = page_address(rdev->sb_page);
670 if (sb->magic != cpu_to_le32(DM_RAID_MAGIC)) {
671 super_sync(rdev->mddev, rdev);
672
673 set_bit(FirstUse, &rdev->flags);
674
675 /* Force writing of superblocks to disk */
676 set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags);
677
678 /* Any superblock is better than none, choose that if given */
679 return refdev ? 0 : 1;
680 }
681
682 if (!refdev)
683 return 1;
684
685 events_sb = le64_to_cpu(sb->events);
686
687 refsb = page_address(refdev->sb_page);
688 events_refsb = le64_to_cpu(refsb->events);
689
690 return (events_sb > events_refsb) ? 1 : 0;
691}
692
NeilBrownfd01b882011-10-11 16:47:53 +1100693static int super_init_validation(struct mddev *mddev, struct md_rdev *rdev)
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100694{
695 int role;
696 struct raid_set *rs = container_of(mddev, struct raid_set, md);
697 uint64_t events_sb;
698 uint64_t failed_devices;
699 struct dm_raid_superblock *sb;
700 uint32_t new_devs = 0;
701 uint32_t rebuilds = 0;
NeilBrown3cb03002011-10-11 16:45:26 +1100702 struct md_rdev *r, *t;
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100703 struct dm_raid_superblock *sb2;
704
705 sb = page_address(rdev->sb_page);
706 events_sb = le64_to_cpu(sb->events);
707 failed_devices = le64_to_cpu(sb->failed_devices);
708
709 /*
710 * Initialise to 1 if this is a new superblock.
711 */
712 mddev->events = events_sb ? : 1;
713
714 /*
715 * Reshaping is not currently allowed
716 */
717 if ((le32_to_cpu(sb->level) != mddev->level) ||
718 (le32_to_cpu(sb->layout) != mddev->layout) ||
719 (le32_to_cpu(sb->stripe_sectors) != mddev->chunk_sectors)) {
720 DMERR("Reshaping arrays not yet supported.");
721 return -EINVAL;
722 }
723
724 /* We can only change the number of devices in RAID1 right now */
725 if ((rs->raid_type->level != 1) &&
726 (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) {
727 DMERR("Reshaping arrays not yet supported.");
728 return -EINVAL;
729 }
730
731 if (!(rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC)))
732 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
733
734 /*
735 * During load, we set FirstUse if a new superblock was written.
736 * There are two reasons we might not have a superblock:
737 * 1) The array is brand new - in which case, all of the
738 * devices must have their In_sync bit set. Also,
739 * recovery_cp must be 0, unless forced.
740 * 2) This is a new device being added to an old array
741 * and the new device needs to be rebuilt - in which
742 * case the In_sync bit will /not/ be set and
743 * recovery_cp must be MaxSector.
744 */
745 rdev_for_each(r, t, mddev) {
746 if (!test_bit(In_sync, &r->flags)) {
747 if (!test_bit(FirstUse, &r->flags))
748 DMERR("Superblock area of "
749 "rebuild device %d should have been "
750 "cleared.", r->raid_disk);
751 set_bit(FirstUse, &r->flags);
752 rebuilds++;
753 } else if (test_bit(FirstUse, &r->flags))
754 new_devs++;
755 }
756
757 if (!rebuilds) {
758 if (new_devs == mddev->raid_disks) {
759 DMINFO("Superblocks created for new array");
760 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
761 } else if (new_devs) {
762 DMERR("New device injected "
763 "into existing array without 'rebuild' "
764 "parameter specified");
765 return -EINVAL;
766 }
767 } else if (new_devs) {
768 DMERR("'rebuild' devices cannot be "
769 "injected into an array with other first-time devices");
770 return -EINVAL;
771 } else if (mddev->recovery_cp != MaxSector) {
772 DMERR("'rebuild' specified while array is not in-sync");
773 return -EINVAL;
774 }
775
776 /*
777 * Now we set the Faulty bit for those devices that are
778 * recorded in the superblock as failed.
779 */
780 rdev_for_each(r, t, mddev) {
781 if (!r->sb_page)
782 continue;
783 sb2 = page_address(r->sb_page);
784 sb2->failed_devices = 0;
785
786 /*
787 * Check for any device re-ordering.
788 */
789 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
790 role = le32_to_cpu(sb2->array_position);
791 if (role != r->raid_disk) {
792 if (rs->raid_type->level != 1) {
793 rs->ti->error = "Cannot change device "
794 "positions in RAID array";
795 return -EINVAL;
796 }
797 DMINFO("RAID1 device #%d now at position #%d",
798 role, r->raid_disk);
799 }
800
801 /*
802 * Partial recovery is performed on
803 * returning failed devices.
804 */
805 if (failed_devices & (1 << role))
806 set_bit(Faulty, &r->flags);
807 }
808 }
809
810 return 0;
811}
812
NeilBrownfd01b882011-10-11 16:47:53 +1100813static int super_validate(struct mddev *mddev, struct md_rdev *rdev)
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100814{
815 struct dm_raid_superblock *sb = page_address(rdev->sb_page);
816
817 /*
818 * If mddev->events is not set, we know we have not yet initialized
819 * the array.
820 */
821 if (!mddev->events && super_init_validation(mddev, rdev))
822 return -EINVAL;
823
824 mddev->bitmap_info.offset = 4096 >> 9; /* Enable bitmap creation */
825 rdev->mddev->bitmap_info.default_offset = 4096 >> 9;
826 if (!test_bit(FirstUse, &rdev->flags)) {
827 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
828 if (rdev->recovery_offset != MaxSector)
829 clear_bit(In_sync, &rdev->flags);
830 }
831
832 /*
833 * If a device comes back, set it as not In_sync and no longer faulty.
834 */
835 if (test_bit(Faulty, &rdev->flags)) {
836 clear_bit(Faulty, &rdev->flags);
837 clear_bit(In_sync, &rdev->flags);
838 rdev->saved_raid_disk = rdev->raid_disk;
839 rdev->recovery_offset = 0;
840 }
841
842 clear_bit(FirstUse, &rdev->flags);
843
844 return 0;
845}
846
847/*
848 * Analyse superblocks and select the freshest.
849 */
850static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
851{
852 int ret;
NeilBrown3cb03002011-10-11 16:45:26 +1100853 struct md_rdev *rdev, *freshest, *tmp;
NeilBrownfd01b882011-10-11 16:47:53 +1100854 struct mddev *mddev = &rs->md;
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100855
856 freshest = NULL;
857 rdev_for_each(rdev, tmp, mddev) {
858 if (!rdev->meta_bdev)
859 continue;
860
861 ret = super_load(rdev, freshest);
862
863 switch (ret) {
864 case 1:
865 freshest = rdev;
866 break;
867 case 0:
868 break;
869 default:
870 ti->error = "Failed to load superblock";
871 return ret;
872 }
873 }
874
875 if (!freshest)
876 return 0;
877
878 /*
879 * Validation of the freshest device provides the source of
880 * validation for the remaining devices.
881 */
882 ti->error = "Unable to assemble array: Invalid superblocks";
883 if (super_validate(mddev, freshest))
884 return -EINVAL;
885
886 rdev_for_each(rdev, tmp, mddev)
887 if ((rdev != freshest) && super_validate(mddev, rdev))
888 return -EINVAL;
889
890 return 0;
891}
892
893/*
NeilBrown9d09e662011-01-13 20:00:02 +0000894 * Construct a RAID4/5/6 mapping:
895 * Args:
896 * <raid_type> <#raid_params> <raid_params> \
897 * <#raid_devs> { <meta_dev1> <dev1> .. <meta_devN> <devN> }
898 *
NeilBrown9d09e662011-01-13 20:00:02 +0000899 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
900 * details on possible <raid_params>.
901 */
902static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv)
903{
904 int ret;
905 struct raid_type *rt;
906 unsigned long num_raid_params, num_raid_devs;
907 struct raid_set *rs = NULL;
908
909 /* Must have at least <raid_type> <#raid_params> */
910 if (argc < 2) {
911 ti->error = "Too few arguments";
912 return -EINVAL;
913 }
914
915 /* raid type */
916 rt = get_raid_type(argv[0]);
917 if (!rt) {
918 ti->error = "Unrecognised raid_type";
919 return -EINVAL;
920 }
921 argc--;
922 argv++;
923
924 /* number of RAID parameters */
925 if (strict_strtoul(argv[0], 10, &num_raid_params) < 0) {
926 ti->error = "Cannot understand number of RAID parameters";
927 return -EINVAL;
928 }
929 argc--;
930 argv++;
931
932 /* Skip over RAID params for now and find out # of devices */
933 if (num_raid_params + 1 > argc) {
934 ti->error = "Arguments do not agree with counts given";
935 return -EINVAL;
936 }
937
938 if ((strict_strtoul(argv[num_raid_params], 10, &num_raid_devs) < 0) ||
939 (num_raid_devs >= INT_MAX)) {
940 ti->error = "Cannot understand number of raid devices";
941 return -EINVAL;
942 }
943
944 rs = context_alloc(ti, rt, (unsigned)num_raid_devs);
945 if (IS_ERR(rs))
946 return PTR_ERR(rs);
947
948 ret = parse_raid_params(rs, argv, (unsigned)num_raid_params);
949 if (ret)
950 goto bad;
951
952 ret = -EINVAL;
953
954 argc -= num_raid_params + 1; /* +1: we already have num_raid_devs */
955 argv += num_raid_params + 1;
956
957 if (argc != (num_raid_devs * 2)) {
958 ti->error = "Supplied RAID devices does not match the count given";
959 goto bad;
960 }
961
962 ret = dev_parms(rs, argv);
963 if (ret)
964 goto bad;
965
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100966 rs->md.sync_super = super_sync;
967 ret = analyse_superblocks(ti, rs);
968 if (ret)
969 goto bad;
970
NeilBrown9d09e662011-01-13 20:00:02 +0000971 INIT_WORK(&rs->md.event_work, do_table_event);
NeilBrown9d09e662011-01-13 20:00:02 +0000972 ti->private = rs;
973
974 mutex_lock(&rs->md.reconfig_mutex);
975 ret = md_run(&rs->md);
976 rs->md.in_sync = 0; /* Assume already marked dirty */
977 mutex_unlock(&rs->md.reconfig_mutex);
978
979 if (ret) {
980 ti->error = "Fail to run raid array";
981 goto bad;
982 }
983
984 rs->callbacks.congested_fn = raid_is_congested;
NeilBrown9d09e662011-01-13 20:00:02 +0000985 dm_table_add_target_callbacks(ti->table, &rs->callbacks);
986
Jonathan Brassow32737272011-08-02 12:32:07 +0100987 mddev_suspend(&rs->md);
NeilBrown9d09e662011-01-13 20:00:02 +0000988 return 0;
989
990bad:
991 context_free(rs);
992
993 return ret;
994}
995
996static void raid_dtr(struct dm_target *ti)
997{
998 struct raid_set *rs = ti->private;
999
1000 list_del_init(&rs->callbacks.list);
1001 md_stop(&rs->md);
1002 context_free(rs);
1003}
1004
1005static int raid_map(struct dm_target *ti, struct bio *bio, union map_info *map_context)
1006{
1007 struct raid_set *rs = ti->private;
NeilBrownfd01b882011-10-11 16:47:53 +11001008 struct mddev *mddev = &rs->md;
NeilBrown9d09e662011-01-13 20:00:02 +00001009
1010 mddev->pers->make_request(mddev, bio);
1011
1012 return DM_MAPIO_SUBMITTED;
1013}
1014
1015static int raid_status(struct dm_target *ti, status_type_t type,
1016 char *result, unsigned maxlen)
1017{
1018 struct raid_set *rs = ti->private;
1019 unsigned raid_param_cnt = 1; /* at least 1 for chunksize */
1020 unsigned sz = 0;
Jonathan E Brassow2e727c32011-10-31 20:21:26 +00001021 int i, array_in_sync = 0;
NeilBrown9d09e662011-01-13 20:00:02 +00001022 sector_t sync;
1023
1024 switch (type) {
1025 case STATUSTYPE_INFO:
1026 DMEMIT("%s %d ", rs->raid_type->name, rs->md.raid_disks);
1027
NeilBrown9d09e662011-01-13 20:00:02 +00001028 if (test_bit(MD_RECOVERY_RUNNING, &rs->md.recovery))
1029 sync = rs->md.curr_resync_completed;
1030 else
1031 sync = rs->md.recovery_cp;
1032
Jonathan E Brassow2e727c32011-10-31 20:21:26 +00001033 if (sync >= rs->md.resync_max_sectors) {
1034 array_in_sync = 1;
NeilBrown9d09e662011-01-13 20:00:02 +00001035 sync = rs->md.resync_max_sectors;
Jonathan E Brassow2e727c32011-10-31 20:21:26 +00001036 } else {
1037 /*
1038 * The array may be doing an initial sync, or it may
1039 * be rebuilding individual components. If all the
1040 * devices are In_sync, then it is the array that is
1041 * being initialized.
1042 */
1043 for (i = 0; i < rs->md.raid_disks; i++)
1044 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
1045 array_in_sync = 1;
1046 }
1047 /*
1048 * Status characters:
1049 * 'D' = Dead/Failed device
1050 * 'a' = Alive but not in-sync
1051 * 'A' = Alive and in-sync
1052 */
1053 for (i = 0; i < rs->md.raid_disks; i++) {
1054 if (test_bit(Faulty, &rs->dev[i].rdev.flags))
1055 DMEMIT("D");
1056 else if (!array_in_sync ||
1057 !test_bit(In_sync, &rs->dev[i].rdev.flags))
1058 DMEMIT("a");
1059 else
1060 DMEMIT("A");
1061 }
NeilBrown9d09e662011-01-13 20:00:02 +00001062
Jonathan E Brassow2e727c32011-10-31 20:21:26 +00001063 /*
1064 * In-sync ratio:
1065 * The in-sync ratio shows the progress of:
1066 * - Initializing the array
1067 * - Rebuilding a subset of devices of the array
1068 * The user can distinguish between the two by referring
1069 * to the status characters.
1070 */
NeilBrown9d09e662011-01-13 20:00:02 +00001071 DMEMIT(" %llu/%llu",
1072 (unsigned long long) sync,
1073 (unsigned long long) rs->md.resync_max_sectors);
1074
1075 break;
1076 case STATUSTYPE_TABLE:
1077 /* The string you would use to construct this array */
Jonathan Brassow46bed2b2011-08-02 12:32:07 +01001078 for (i = 0; i < rs->md.raid_disks; i++) {
Jonathan Brassow13c87582011-08-02 12:32:03 +01001079 if ((rs->print_flags & DMPF_REBUILD) &&
1080 rs->dev[i].data_dev &&
NeilBrown9d09e662011-01-13 20:00:02 +00001081 !test_bit(In_sync, &rs->dev[i].rdev.flags))
Jonathan Brassow13c87582011-08-02 12:32:03 +01001082 raid_param_cnt += 2; /* for rebuilds */
Jonathan Brassow46bed2b2011-08-02 12:32:07 +01001083 if (rs->dev[i].data_dev &&
1084 test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1085 raid_param_cnt += 2;
1086 }
NeilBrown9d09e662011-01-13 20:00:02 +00001087
Jonathan Brassow13c87582011-08-02 12:32:03 +01001088 raid_param_cnt += (hweight64(rs->print_flags & ~DMPF_REBUILD) * 2);
NeilBrown9d09e662011-01-13 20:00:02 +00001089 if (rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC))
1090 raid_param_cnt--;
1091
1092 DMEMIT("%s %u %u", rs->raid_type->name,
1093 raid_param_cnt, rs->md.chunk_sectors);
1094
1095 if ((rs->print_flags & DMPF_SYNC) &&
1096 (rs->md.recovery_cp == MaxSector))
1097 DMEMIT(" sync");
1098 if (rs->print_flags & DMPF_NOSYNC)
1099 DMEMIT(" nosync");
1100
1101 for (i = 0; i < rs->md.raid_disks; i++)
Jonathan Brassow13c87582011-08-02 12:32:03 +01001102 if ((rs->print_flags & DMPF_REBUILD) &&
1103 rs->dev[i].data_dev &&
NeilBrown9d09e662011-01-13 20:00:02 +00001104 !test_bit(In_sync, &rs->dev[i].rdev.flags))
1105 DMEMIT(" rebuild %u", i);
1106
1107 if (rs->print_flags & DMPF_DAEMON_SLEEP)
1108 DMEMIT(" daemon_sleep %lu",
1109 rs->md.bitmap_info.daemon_sleep);
1110
1111 if (rs->print_flags & DMPF_MIN_RECOVERY_RATE)
1112 DMEMIT(" min_recovery_rate %d", rs->md.sync_speed_min);
1113
1114 if (rs->print_flags & DMPF_MAX_RECOVERY_RATE)
1115 DMEMIT(" max_recovery_rate %d", rs->md.sync_speed_max);
1116
Jonathan Brassow46bed2b2011-08-02 12:32:07 +01001117 for (i = 0; i < rs->md.raid_disks; i++)
1118 if (rs->dev[i].data_dev &&
1119 test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1120 DMEMIT(" write_mostly %u", i);
1121
NeilBrown9d09e662011-01-13 20:00:02 +00001122 if (rs->print_flags & DMPF_MAX_WRITE_BEHIND)
1123 DMEMIT(" max_write_behind %lu",
1124 rs->md.bitmap_info.max_write_behind);
1125
1126 if (rs->print_flags & DMPF_STRIPE_CACHE) {
NeilBrownd1688a62011-10-11 16:49:52 +11001127 struct r5conf *conf = rs->md.private;
NeilBrown9d09e662011-01-13 20:00:02 +00001128
1129 /* convert from kiB to sectors */
1130 DMEMIT(" stripe_cache %d",
1131 conf ? conf->max_nr_stripes * 2 : 0);
1132 }
1133
Jonathan Brassowc1084562011-08-02 12:32:07 +01001134 if (rs->print_flags & DMPF_REGION_SIZE)
1135 DMEMIT(" region_size %lu",
1136 rs->md.bitmap_info.chunksize >> 9);
1137
NeilBrown9d09e662011-01-13 20:00:02 +00001138 DMEMIT(" %d", rs->md.raid_disks);
1139 for (i = 0; i < rs->md.raid_disks; i++) {
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001140 if (rs->dev[i].meta_dev)
1141 DMEMIT(" %s", rs->dev[i].meta_dev->name);
1142 else
1143 DMEMIT(" -");
NeilBrown9d09e662011-01-13 20:00:02 +00001144
1145 if (rs->dev[i].data_dev)
1146 DMEMIT(" %s", rs->dev[i].data_dev->name);
1147 else
1148 DMEMIT(" -");
1149 }
1150 }
1151
1152 return 0;
1153}
1154
1155static int raid_iterate_devices(struct dm_target *ti, iterate_devices_callout_fn fn, void *data)
1156{
1157 struct raid_set *rs = ti->private;
1158 unsigned i;
1159 int ret = 0;
1160
1161 for (i = 0; !ret && i < rs->md.raid_disks; i++)
1162 if (rs->dev[i].data_dev)
1163 ret = fn(ti,
1164 rs->dev[i].data_dev,
1165 0, /* No offset on data devs */
1166 rs->md.dev_sectors,
1167 data);
1168
1169 return ret;
1170}
1171
1172static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
1173{
1174 struct raid_set *rs = ti->private;
1175 unsigned chunk_size = rs->md.chunk_sectors << 9;
NeilBrownd1688a62011-10-11 16:49:52 +11001176 struct r5conf *conf = rs->md.private;
NeilBrown9d09e662011-01-13 20:00:02 +00001177
1178 blk_limits_io_min(limits, chunk_size);
1179 blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded));
1180}
1181
1182static void raid_presuspend(struct dm_target *ti)
1183{
1184 struct raid_set *rs = ti->private;
1185
1186 md_stop_writes(&rs->md);
1187}
1188
1189static void raid_postsuspend(struct dm_target *ti)
1190{
1191 struct raid_set *rs = ti->private;
1192
1193 mddev_suspend(&rs->md);
1194}
1195
1196static void raid_resume(struct dm_target *ti)
1197{
1198 struct raid_set *rs = ti->private;
1199
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001200 bitmap_load(&rs->md);
NeilBrown9d09e662011-01-13 20:00:02 +00001201 mddev_resume(&rs->md);
1202}
1203
1204static struct target_type raid_target = {
1205 .name = "raid",
Jonathan Brassow32737272011-08-02 12:32:07 +01001206 .version = {1, 1, 0},
NeilBrown9d09e662011-01-13 20:00:02 +00001207 .module = THIS_MODULE,
1208 .ctr = raid_ctr,
1209 .dtr = raid_dtr,
1210 .map = raid_map,
1211 .status = raid_status,
1212 .iterate_devices = raid_iterate_devices,
1213 .io_hints = raid_io_hints,
1214 .presuspend = raid_presuspend,
1215 .postsuspend = raid_postsuspend,
1216 .resume = raid_resume,
1217};
1218
1219static int __init dm_raid_init(void)
1220{
1221 return dm_register_target(&raid_target);
1222}
1223
1224static void __exit dm_raid_exit(void)
1225{
1226 dm_unregister_target(&raid_target);
1227}
1228
1229module_init(dm_raid_init);
1230module_exit(dm_raid_exit);
1231
1232MODULE_DESCRIPTION(DM_NAME " raid4/5/6 target");
1233MODULE_ALIAS("dm-raid4");
1234MODULE_ALIAS("dm-raid5");
1235MODULE_ALIAS("dm-raid6");
1236MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
1237MODULE_LICENSE("GPL");