Commit | Line | Data |
---|---|---|
9d09e663 N |
1 | /* |
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> | |
9 | ||
10 | #include "md.h" | |
11 | #include "raid5.h" | |
9d09e663 N |
12 | #include "bitmap.h" |
13 | ||
3e8dbb7f AK |
14 | #include <linux/device-mapper.h> |
15 | ||
9d09e663 N |
16 | #define DM_MSG_PREFIX "raid" |
17 | ||
18 | /* | |
b12d437b JB |
19 | * The following flags are used by dm-raid.c to set up the array state. |
20 | * They must be cleared before md_run is called. | |
9d09e663 | 21 | */ |
b12d437b | 22 | #define FirstUse 10 /* rdev flag */ |
9d09e663 N |
23 | |
24 | struct raid_dev { | |
25 | /* | |
26 | * Two DM devices, one to hold metadata and one to hold the | |
27 | * actual data/parity. The reason for this is to not confuse | |
28 | * ti->len and give more flexibility in altering size and | |
29 | * characteristics. | |
30 | * | |
31 | * While it is possible for this device to be associated | |
32 | * with a different physical device than the data_dev, it | |
33 | * is intended for it to be the same. | |
34 | * |--------- Physical Device ---------| | |
35 | * |- meta_dev -|------ data_dev ------| | |
36 | */ | |
37 | struct dm_dev *meta_dev; | |
38 | struct dm_dev *data_dev; | |
39 | struct mdk_rdev_s rdev; | |
40 | }; | |
41 | ||
42 | /* | |
43 | * Flags for rs->print_flags field. | |
44 | */ | |
13c87583 JB |
45 | #define DMPF_SYNC 0x1 |
46 | #define DMPF_NOSYNC 0x2 | |
47 | #define DMPF_REBUILD 0x4 | |
48 | #define DMPF_DAEMON_SLEEP 0x8 | |
49 | #define DMPF_MIN_RECOVERY_RATE 0x10 | |
50 | #define DMPF_MAX_RECOVERY_RATE 0x20 | |
51 | #define DMPF_MAX_WRITE_BEHIND 0x40 | |
52 | #define DMPF_STRIPE_CACHE 0x80 | |
c1084561 | 53 | #define DMPF_REGION_SIZE 0X100 |
9d09e663 N |
54 | struct raid_set { |
55 | struct dm_target *ti; | |
56 | ||
57 | uint64_t print_flags; | |
58 | ||
59 | struct mddev_s md; | |
60 | struct raid_type *raid_type; | |
61 | struct dm_target_callbacks callbacks; | |
62 | ||
63 | struct raid_dev dev[0]; | |
64 | }; | |
65 | ||
66 | /* Supported raid types and properties. */ | |
67 | static struct raid_type { | |
68 | const char *name; /* RAID algorithm. */ | |
69 | const char *descr; /* Descriptor text for logging. */ | |
70 | const unsigned parity_devs; /* # of parity devices. */ | |
71 | const unsigned minimal_devs; /* minimal # of devices in set. */ | |
72 | const unsigned level; /* RAID level. */ | |
73 | const unsigned algorithm; /* RAID algorithm. */ | |
74 | } raid_types[] = { | |
75 | {"raid4", "RAID4 (dedicated parity disk)", 1, 2, 5, ALGORITHM_PARITY_0}, | |
76 | {"raid5_la", "RAID5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC}, | |
77 | {"raid5_ra", "RAID5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC}, | |
78 | {"raid5_ls", "RAID5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC}, | |
79 | {"raid5_rs", "RAID5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC}, | |
80 | {"raid6_zr", "RAID6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART}, | |
81 | {"raid6_nr", "RAID6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART}, | |
82 | {"raid6_nc", "RAID6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE} | |
83 | }; | |
84 | ||
85 | static struct raid_type *get_raid_type(char *name) | |
86 | { | |
87 | int i; | |
88 | ||
89 | for (i = 0; i < ARRAY_SIZE(raid_types); i++) | |
90 | if (!strcmp(raid_types[i].name, name)) | |
91 | return &raid_types[i]; | |
92 | ||
93 | return NULL; | |
94 | } | |
95 | ||
96 | static struct raid_set *context_alloc(struct dm_target *ti, struct raid_type *raid_type, unsigned raid_devs) | |
97 | { | |
98 | unsigned i; | |
99 | struct raid_set *rs; | |
100 | sector_t sectors_per_dev; | |
101 | ||
102 | if (raid_devs <= raid_type->parity_devs) { | |
103 | ti->error = "Insufficient number of devices"; | |
104 | return ERR_PTR(-EINVAL); | |
105 | } | |
106 | ||
107 | sectors_per_dev = ti->len; | |
108 | if (sector_div(sectors_per_dev, (raid_devs - raid_type->parity_devs))) { | |
109 | ti->error = "Target length not divisible by number of data devices"; | |
110 | return ERR_PTR(-EINVAL); | |
111 | } | |
112 | ||
113 | rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL); | |
114 | if (!rs) { | |
115 | ti->error = "Cannot allocate raid context"; | |
116 | return ERR_PTR(-ENOMEM); | |
117 | } | |
118 | ||
119 | mddev_init(&rs->md); | |
120 | ||
121 | rs->ti = ti; | |
122 | rs->raid_type = raid_type; | |
123 | rs->md.raid_disks = raid_devs; | |
124 | rs->md.level = raid_type->level; | |
125 | rs->md.new_level = rs->md.level; | |
126 | rs->md.dev_sectors = sectors_per_dev; | |
127 | rs->md.layout = raid_type->algorithm; | |
128 | rs->md.new_layout = rs->md.layout; | |
129 | rs->md.delta_disks = 0; | |
130 | rs->md.recovery_cp = 0; | |
131 | ||
132 | for (i = 0; i < raid_devs; i++) | |
133 | md_rdev_init(&rs->dev[i].rdev); | |
134 | ||
135 | /* | |
136 | * Remaining items to be initialized by further RAID params: | |
137 | * rs->md.persistent | |
138 | * rs->md.external | |
139 | * rs->md.chunk_sectors | |
140 | * rs->md.new_chunk_sectors | |
141 | */ | |
142 | ||
143 | return rs; | |
144 | } | |
145 | ||
146 | static void context_free(struct raid_set *rs) | |
147 | { | |
148 | int i; | |
149 | ||
b12d437b JB |
150 | for (i = 0; i < rs->md.raid_disks; i++) { |
151 | if (rs->dev[i].meta_dev) | |
152 | dm_put_device(rs->ti, rs->dev[i].meta_dev); | |
153 | if (rs->dev[i].rdev.sb_page) | |
154 | put_page(rs->dev[i].rdev.sb_page); | |
155 | rs->dev[i].rdev.sb_page = NULL; | |
156 | rs->dev[i].rdev.sb_loaded = 0; | |
9d09e663 N |
157 | if (rs->dev[i].data_dev) |
158 | dm_put_device(rs->ti, rs->dev[i].data_dev); | |
b12d437b | 159 | } |
9d09e663 N |
160 | |
161 | kfree(rs); | |
162 | } | |
163 | ||
164 | /* | |
165 | * For every device we have two words | |
166 | * <meta_dev>: meta device name or '-' if missing | |
167 | * <data_dev>: data device name or '-' if missing | |
168 | * | |
b12d437b JB |
169 | * The following are permitted: |
170 | * - - | |
171 | * - <data_dev> | |
172 | * <meta_dev> <data_dev> | |
173 | * | |
174 | * The following is not allowed: | |
175 | * <meta_dev> - | |
176 | * | |
177 | * This code parses those words. If there is a failure, | |
178 | * the caller must use context_free to unwind the operations. | |
9d09e663 N |
179 | */ |
180 | static int dev_parms(struct raid_set *rs, char **argv) | |
181 | { | |
182 | int i; | |
183 | int rebuild = 0; | |
184 | int metadata_available = 0; | |
185 | int ret = 0; | |
186 | ||
187 | for (i = 0; i < rs->md.raid_disks; i++, argv += 2) { | |
188 | rs->dev[i].rdev.raid_disk = i; | |
189 | ||
190 | rs->dev[i].meta_dev = NULL; | |
191 | rs->dev[i].data_dev = NULL; | |
192 | ||
193 | /* | |
194 | * There are no offsets, since there is a separate device | |
195 | * for data and metadata. | |
196 | */ | |
197 | rs->dev[i].rdev.data_offset = 0; | |
198 | rs->dev[i].rdev.mddev = &rs->md; | |
199 | ||
200 | if (strcmp(argv[0], "-")) { | |
b12d437b JB |
201 | ret = dm_get_device(rs->ti, argv[0], |
202 | dm_table_get_mode(rs->ti->table), | |
203 | &rs->dev[i].meta_dev); | |
204 | rs->ti->error = "RAID metadata device lookup failure"; | |
205 | if (ret) | |
206 | return ret; | |
207 | ||
208 | rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL); | |
209 | if (!rs->dev[i].rdev.sb_page) | |
210 | return -ENOMEM; | |
9d09e663 N |
211 | } |
212 | ||
213 | if (!strcmp(argv[1], "-")) { | |
214 | if (!test_bit(In_sync, &rs->dev[i].rdev.flags) && | |
215 | (!rs->dev[i].rdev.recovery_offset)) { | |
216 | rs->ti->error = "Drive designated for rebuild not specified"; | |
217 | return -EINVAL; | |
218 | } | |
219 | ||
b12d437b JB |
220 | rs->ti->error = "No data device supplied with metadata device"; |
221 | if (rs->dev[i].meta_dev) | |
222 | return -EINVAL; | |
223 | ||
9d09e663 N |
224 | continue; |
225 | } | |
226 | ||
227 | ret = dm_get_device(rs->ti, argv[1], | |
228 | dm_table_get_mode(rs->ti->table), | |
229 | &rs->dev[i].data_dev); | |
230 | if (ret) { | |
231 | rs->ti->error = "RAID device lookup failure"; | |
232 | return ret; | |
233 | } | |
234 | ||
b12d437b JB |
235 | if (rs->dev[i].meta_dev) { |
236 | metadata_available = 1; | |
237 | rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev; | |
238 | } | |
9d09e663 N |
239 | rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev; |
240 | list_add(&rs->dev[i].rdev.same_set, &rs->md.disks); | |
241 | if (!test_bit(In_sync, &rs->dev[i].rdev.flags)) | |
242 | rebuild++; | |
243 | } | |
244 | ||
245 | if (metadata_available) { | |
246 | rs->md.external = 0; | |
247 | rs->md.persistent = 1; | |
248 | rs->md.major_version = 2; | |
249 | } else if (rebuild && !rs->md.recovery_cp) { | |
250 | /* | |
251 | * Without metadata, we will not be able to tell if the array | |
252 | * is in-sync or not - we must assume it is not. Therefore, | |
253 | * it is impossible to rebuild a drive. | |
254 | * | |
255 | * Even if there is metadata, the on-disk information may | |
256 | * indicate that the array is not in-sync and it will then | |
257 | * fail at that time. | |
258 | * | |
259 | * User could specify 'nosync' option if desperate. | |
260 | */ | |
261 | DMERR("Unable to rebuild drive while array is not in-sync"); | |
262 | rs->ti->error = "RAID device lookup failure"; | |
263 | return -EINVAL; | |
264 | } | |
265 | ||
266 | return 0; | |
267 | } | |
268 | ||
c1084561 JB |
269 | /* |
270 | * validate_region_size | |
271 | * @rs | |
272 | * @region_size: region size in sectors. If 0, pick a size (4MiB default). | |
273 | * | |
274 | * Set rs->md.bitmap_info.chunksize (which really refers to 'region size'). | |
275 | * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap. | |
276 | * | |
277 | * Returns: 0 on success, -EINVAL on failure. | |
278 | */ | |
279 | static int validate_region_size(struct raid_set *rs, unsigned long region_size) | |
280 | { | |
281 | unsigned long min_region_size = rs->ti->len / (1 << 21); | |
282 | ||
283 | if (!region_size) { | |
284 | /* | |
285 | * Choose a reasonable default. All figures in sectors. | |
286 | */ | |
287 | if (min_region_size > (1 << 13)) { | |
288 | DMINFO("Choosing default region size of %lu sectors", | |
289 | region_size); | |
290 | region_size = min_region_size; | |
291 | } else { | |
292 | DMINFO("Choosing default region size of 4MiB"); | |
293 | region_size = 1 << 13; /* sectors */ | |
294 | } | |
295 | } else { | |
296 | /* | |
297 | * Validate user-supplied value. | |
298 | */ | |
299 | if (region_size > rs->ti->len) { | |
300 | rs->ti->error = "Supplied region size is too large"; | |
301 | return -EINVAL; | |
302 | } | |
303 | ||
304 | if (region_size < min_region_size) { | |
305 | DMERR("Supplied region_size (%lu sectors) below minimum (%lu)", | |
306 | region_size, min_region_size); | |
307 | rs->ti->error = "Supplied region size is too small"; | |
308 | return -EINVAL; | |
309 | } | |
310 | ||
311 | if (!is_power_of_2(region_size)) { | |
312 | rs->ti->error = "Region size is not a power of 2"; | |
313 | return -EINVAL; | |
314 | } | |
315 | ||
316 | if (region_size < rs->md.chunk_sectors) { | |
317 | rs->ti->error = "Region size is smaller than the chunk size"; | |
318 | return -EINVAL; | |
319 | } | |
320 | } | |
321 | ||
322 | /* | |
323 | * Convert sectors to bytes. | |
324 | */ | |
325 | rs->md.bitmap_info.chunksize = (region_size << 9); | |
326 | ||
327 | return 0; | |
328 | } | |
329 | ||
9d09e663 N |
330 | /* |
331 | * Possible arguments are... | |
332 | * RAID456: | |
333 | * <chunk_size> [optional_args] | |
334 | * | |
335 | * Optional args: | |
336 | * [[no]sync] Force or prevent recovery of the entire array | |
337 | * [rebuild <idx>] Rebuild the drive indicated by the index | |
338 | * [daemon_sleep <ms>] Time between bitmap daemon work to clear bits | |
339 | * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization | |
340 | * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization | |
46bed2b5 | 341 | * [write_mostly <idx>] Indicate a write mostly drive via index |
9d09e663 N |
342 | * [max_write_behind <sectors>] See '-write-behind=' (man mdadm) |
343 | * [stripe_cache <sectors>] Stripe cache size for higher RAIDs | |
c1084561 | 344 | * [region_size <sectors>] Defines granularity of bitmap |
9d09e663 N |
345 | */ |
346 | static int parse_raid_params(struct raid_set *rs, char **argv, | |
347 | unsigned num_raid_params) | |
348 | { | |
349 | unsigned i, rebuild_cnt = 0; | |
c1084561 | 350 | unsigned long value, region_size = 0; |
9d09e663 N |
351 | char *key; |
352 | ||
353 | /* | |
354 | * First, parse the in-order required arguments | |
355 | */ | |
356 | if ((strict_strtoul(argv[0], 10, &value) < 0) || | |
357 | !is_power_of_2(value) || (value < 8)) { | |
358 | rs->ti->error = "Bad chunk size"; | |
359 | return -EINVAL; | |
360 | } | |
361 | ||
362 | rs->md.new_chunk_sectors = rs->md.chunk_sectors = value; | |
363 | argv++; | |
364 | num_raid_params--; | |
365 | ||
366 | /* | |
b12d437b JB |
367 | * We set each individual device as In_sync with a completed |
368 | * 'recovery_offset'. If there has been a device failure or | |
369 | * replacement then one of the following cases applies: | |
370 | * | |
371 | * 1) User specifies 'rebuild'. | |
372 | * - Device is reset when param is read. | |
373 | * 2) A new device is supplied. | |
374 | * - No matching superblock found, resets device. | |
375 | * 3) Device failure was transient and returns on reload. | |
376 | * - Failure noticed, resets device for bitmap replay. | |
377 | * 4) Device hadn't completed recovery after previous failure. | |
378 | * - Superblock is read and overrides recovery_offset. | |
379 | * | |
380 | * What is found in the superblocks of the devices is always | |
381 | * authoritative, unless 'rebuild' or '[no]sync' was specified. | |
9d09e663 | 382 | */ |
b12d437b | 383 | for (i = 0; i < rs->md.raid_disks; i++) { |
9d09e663 | 384 | set_bit(In_sync, &rs->dev[i].rdev.flags); |
b12d437b JB |
385 | rs->dev[i].rdev.recovery_offset = MaxSector; |
386 | } | |
9d09e663 | 387 | |
b12d437b JB |
388 | /* |
389 | * Second, parse the unordered optional arguments | |
390 | */ | |
9d09e663 | 391 | for (i = 0; i < num_raid_params; i++) { |
13c87583 | 392 | if (!strcasecmp(argv[i], "nosync")) { |
9d09e663 N |
393 | rs->md.recovery_cp = MaxSector; |
394 | rs->print_flags |= DMPF_NOSYNC; | |
9d09e663 N |
395 | continue; |
396 | } | |
13c87583 | 397 | if (!strcasecmp(argv[i], "sync")) { |
9d09e663 N |
398 | rs->md.recovery_cp = 0; |
399 | rs->print_flags |= DMPF_SYNC; | |
9d09e663 N |
400 | continue; |
401 | } | |
402 | ||
403 | /* The rest of the optional arguments come in key/value pairs */ | |
404 | if ((i + 1) >= num_raid_params) { | |
405 | rs->ti->error = "Wrong number of raid parameters given"; | |
406 | return -EINVAL; | |
407 | } | |
408 | ||
409 | key = argv[i++]; | |
410 | if (strict_strtoul(argv[i], 10, &value) < 0) { | |
411 | rs->ti->error = "Bad numerical argument given in raid params"; | |
412 | return -EINVAL; | |
413 | } | |
414 | ||
13c87583 | 415 | if (!strcasecmp(key, "rebuild")) { |
9d09e663 N |
416 | if (++rebuild_cnt > rs->raid_type->parity_devs) { |
417 | rs->ti->error = "Too many rebuild drives given"; | |
418 | return -EINVAL; | |
419 | } | |
420 | if (value > rs->md.raid_disks) { | |
421 | rs->ti->error = "Invalid rebuild index given"; | |
422 | return -EINVAL; | |
423 | } | |
424 | clear_bit(In_sync, &rs->dev[value].rdev.flags); | |
425 | rs->dev[value].rdev.recovery_offset = 0; | |
13c87583 | 426 | rs->print_flags |= DMPF_REBUILD; |
46bed2b5 JB |
427 | } else if (!strcasecmp(key, "write_mostly")) { |
428 | if (rs->raid_type->level != 1) { | |
429 | rs->ti->error = "write_mostly option is only valid for RAID1"; | |
430 | return -EINVAL; | |
431 | } | |
432 | if (value > rs->md.raid_disks) { | |
433 | rs->ti->error = "Invalid write_mostly drive index given"; | |
434 | return -EINVAL; | |
435 | } | |
436 | set_bit(WriteMostly, &rs->dev[value].rdev.flags); | |
13c87583 | 437 | } else if (!strcasecmp(key, "max_write_behind")) { |
46bed2b5 JB |
438 | if (rs->raid_type->level != 1) { |
439 | rs->ti->error = "max_write_behind option is only valid for RAID1"; | |
440 | return -EINVAL; | |
441 | } | |
9d09e663 N |
442 | rs->print_flags |= DMPF_MAX_WRITE_BEHIND; |
443 | ||
444 | /* | |
445 | * In device-mapper, we specify things in sectors, but | |
446 | * MD records this value in kB | |
447 | */ | |
448 | value /= 2; | |
449 | if (value > COUNTER_MAX) { | |
450 | rs->ti->error = "Max write-behind limit out of range"; | |
451 | return -EINVAL; | |
452 | } | |
453 | rs->md.bitmap_info.max_write_behind = value; | |
13c87583 | 454 | } else if (!strcasecmp(key, "daemon_sleep")) { |
9d09e663 N |
455 | rs->print_flags |= DMPF_DAEMON_SLEEP; |
456 | if (!value || (value > MAX_SCHEDULE_TIMEOUT)) { | |
457 | rs->ti->error = "daemon sleep period out of range"; | |
458 | return -EINVAL; | |
459 | } | |
460 | rs->md.bitmap_info.daemon_sleep = value; | |
13c87583 | 461 | } else if (!strcasecmp(key, "stripe_cache")) { |
9d09e663 N |
462 | rs->print_flags |= DMPF_STRIPE_CACHE; |
463 | ||
464 | /* | |
465 | * In device-mapper, we specify things in sectors, but | |
466 | * MD records this value in kB | |
467 | */ | |
468 | value /= 2; | |
469 | ||
470 | if (rs->raid_type->level < 5) { | |
471 | rs->ti->error = "Inappropriate argument: stripe_cache"; | |
472 | return -EINVAL; | |
473 | } | |
474 | if (raid5_set_cache_size(&rs->md, (int)value)) { | |
475 | rs->ti->error = "Bad stripe_cache size"; | |
476 | return -EINVAL; | |
477 | } | |
13c87583 | 478 | } else if (!strcasecmp(key, "min_recovery_rate")) { |
9d09e663 N |
479 | rs->print_flags |= DMPF_MIN_RECOVERY_RATE; |
480 | if (value > INT_MAX) { | |
481 | rs->ti->error = "min_recovery_rate out of range"; | |
482 | return -EINVAL; | |
483 | } | |
484 | rs->md.sync_speed_min = (int)value; | |
13c87583 | 485 | } else if (!strcasecmp(key, "max_recovery_rate")) { |
9d09e663 N |
486 | rs->print_flags |= DMPF_MAX_RECOVERY_RATE; |
487 | if (value > INT_MAX) { | |
488 | rs->ti->error = "max_recovery_rate out of range"; | |
489 | return -EINVAL; | |
490 | } | |
491 | rs->md.sync_speed_max = (int)value; | |
c1084561 JB |
492 | } else if (!strcasecmp(key, "region_size")) { |
493 | rs->print_flags |= DMPF_REGION_SIZE; | |
494 | region_size = value; | |
9d09e663 N |
495 | } else { |
496 | DMERR("Unable to parse RAID parameter: %s", key); | |
497 | rs->ti->error = "Unable to parse RAID parameters"; | |
498 | return -EINVAL; | |
499 | } | |
500 | } | |
501 | ||
c1084561 JB |
502 | if (validate_region_size(rs, region_size)) |
503 | return -EINVAL; | |
504 | ||
505 | if (rs->md.chunk_sectors) | |
506 | rs->ti->split_io = rs->md.chunk_sectors; | |
507 | else | |
508 | rs->ti->split_io = region_size; | |
509 | ||
9d09e663 N |
510 | /* Assume there are no metadata devices until the drives are parsed */ |
511 | rs->md.persistent = 0; | |
512 | rs->md.external = 1; | |
513 | ||
514 | return 0; | |
515 | } | |
516 | ||
517 | static void do_table_event(struct work_struct *ws) | |
518 | { | |
519 | struct raid_set *rs = container_of(ws, struct raid_set, md.event_work); | |
520 | ||
521 | dm_table_event(rs->ti->table); | |
522 | } | |
523 | ||
524 | static int raid_is_congested(struct dm_target_callbacks *cb, int bits) | |
525 | { | |
526 | struct raid_set *rs = container_of(cb, struct raid_set, callbacks); | |
527 | ||
528 | return md_raid5_congested(&rs->md, bits); | |
529 | } | |
530 | ||
b12d437b JB |
531 | /* |
532 | * This structure is never routinely used by userspace, unlike md superblocks. | |
533 | * Devices with this superblock should only ever be accessed via device-mapper. | |
534 | */ | |
535 | #define DM_RAID_MAGIC 0x64526D44 | |
536 | struct dm_raid_superblock { | |
537 | __le32 magic; /* "DmRd" */ | |
538 | __le32 features; /* Used to indicate possible future changes */ | |
539 | ||
540 | __le32 num_devices; /* Number of devices in this array. (Max 64) */ | |
541 | __le32 array_position; /* The position of this drive in the array */ | |
542 | ||
543 | __le64 events; /* Incremented by md when superblock updated */ | |
544 | __le64 failed_devices; /* Bit field of devices to indicate failures */ | |
545 | ||
546 | /* | |
547 | * This offset tracks the progress of the repair or replacement of | |
548 | * an individual drive. | |
549 | */ | |
550 | __le64 disk_recovery_offset; | |
551 | ||
552 | /* | |
553 | * This offset tracks the progress of the initial array | |
554 | * synchronisation/parity calculation. | |
555 | */ | |
556 | __le64 array_resync_offset; | |
557 | ||
558 | /* | |
559 | * RAID characteristics | |
560 | */ | |
561 | __le32 level; | |
562 | __le32 layout; | |
563 | __le32 stripe_sectors; | |
564 | ||
565 | __u8 pad[452]; /* Round struct to 512 bytes. */ | |
566 | /* Always set to 0 when writing. */ | |
567 | } __packed; | |
568 | ||
569 | static int read_disk_sb(mdk_rdev_t *rdev, int size) | |
570 | { | |
571 | BUG_ON(!rdev->sb_page); | |
572 | ||
573 | if (rdev->sb_loaded) | |
574 | return 0; | |
575 | ||
576 | if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, 1)) { | |
577 | DMERR("Failed to read device superblock"); | |
578 | return -EINVAL; | |
579 | } | |
580 | ||
581 | rdev->sb_loaded = 1; | |
582 | ||
583 | return 0; | |
584 | } | |
585 | ||
586 | static void super_sync(mddev_t *mddev, mdk_rdev_t *rdev) | |
587 | { | |
588 | mdk_rdev_t *r, *t; | |
589 | uint64_t failed_devices; | |
590 | struct dm_raid_superblock *sb; | |
591 | ||
592 | sb = page_address(rdev->sb_page); | |
593 | failed_devices = le64_to_cpu(sb->failed_devices); | |
594 | ||
595 | rdev_for_each(r, t, mddev) | |
596 | if ((r->raid_disk >= 0) && test_bit(Faulty, &r->flags)) | |
597 | failed_devices |= (1ULL << r->raid_disk); | |
598 | ||
599 | memset(sb, 0, sizeof(*sb)); | |
600 | ||
601 | sb->magic = cpu_to_le32(DM_RAID_MAGIC); | |
602 | sb->features = cpu_to_le32(0); /* No features yet */ | |
603 | ||
604 | sb->num_devices = cpu_to_le32(mddev->raid_disks); | |
605 | sb->array_position = cpu_to_le32(rdev->raid_disk); | |
606 | ||
607 | sb->events = cpu_to_le64(mddev->events); | |
608 | sb->failed_devices = cpu_to_le64(failed_devices); | |
609 | ||
610 | sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset); | |
611 | sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp); | |
612 | ||
613 | sb->level = cpu_to_le32(mddev->level); | |
614 | sb->layout = cpu_to_le32(mddev->layout); | |
615 | sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors); | |
616 | } | |
617 | ||
618 | /* | |
619 | * super_load | |
620 | * | |
621 | * This function creates a superblock if one is not found on the device | |
622 | * and will decide which superblock to use if there's a choice. | |
623 | * | |
624 | * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise | |
625 | */ | |
626 | static int super_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev) | |
627 | { | |
628 | int ret; | |
629 | struct dm_raid_superblock *sb; | |
630 | struct dm_raid_superblock *refsb; | |
631 | uint64_t events_sb, events_refsb; | |
632 | ||
633 | rdev->sb_start = 0; | |
634 | rdev->sb_size = sizeof(*sb); | |
635 | ||
636 | ret = read_disk_sb(rdev, rdev->sb_size); | |
637 | if (ret) | |
638 | return ret; | |
639 | ||
640 | sb = page_address(rdev->sb_page); | |
641 | if (sb->magic != cpu_to_le32(DM_RAID_MAGIC)) { | |
642 | super_sync(rdev->mddev, rdev); | |
643 | ||
644 | set_bit(FirstUse, &rdev->flags); | |
645 | ||
646 | /* Force writing of superblocks to disk */ | |
647 | set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags); | |
648 | ||
649 | /* Any superblock is better than none, choose that if given */ | |
650 | return refdev ? 0 : 1; | |
651 | } | |
652 | ||
653 | if (!refdev) | |
654 | return 1; | |
655 | ||
656 | events_sb = le64_to_cpu(sb->events); | |
657 | ||
658 | refsb = page_address(refdev->sb_page); | |
659 | events_refsb = le64_to_cpu(refsb->events); | |
660 | ||
661 | return (events_sb > events_refsb) ? 1 : 0; | |
662 | } | |
663 | ||
664 | static int super_init_validation(mddev_t *mddev, mdk_rdev_t *rdev) | |
665 | { | |
666 | int role; | |
667 | struct raid_set *rs = container_of(mddev, struct raid_set, md); | |
668 | uint64_t events_sb; | |
669 | uint64_t failed_devices; | |
670 | struct dm_raid_superblock *sb; | |
671 | uint32_t new_devs = 0; | |
672 | uint32_t rebuilds = 0; | |
673 | mdk_rdev_t *r, *t; | |
674 | struct dm_raid_superblock *sb2; | |
675 | ||
676 | sb = page_address(rdev->sb_page); | |
677 | events_sb = le64_to_cpu(sb->events); | |
678 | failed_devices = le64_to_cpu(sb->failed_devices); | |
679 | ||
680 | /* | |
681 | * Initialise to 1 if this is a new superblock. | |
682 | */ | |
683 | mddev->events = events_sb ? : 1; | |
684 | ||
685 | /* | |
686 | * Reshaping is not currently allowed | |
687 | */ | |
688 | if ((le32_to_cpu(sb->level) != mddev->level) || | |
689 | (le32_to_cpu(sb->layout) != mddev->layout) || | |
690 | (le32_to_cpu(sb->stripe_sectors) != mddev->chunk_sectors)) { | |
691 | DMERR("Reshaping arrays not yet supported."); | |
692 | return -EINVAL; | |
693 | } | |
694 | ||
695 | /* We can only change the number of devices in RAID1 right now */ | |
696 | if ((rs->raid_type->level != 1) && | |
697 | (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) { | |
698 | DMERR("Reshaping arrays not yet supported."); | |
699 | return -EINVAL; | |
700 | } | |
701 | ||
702 | if (!(rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC))) | |
703 | mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset); | |
704 | ||
705 | /* | |
706 | * During load, we set FirstUse if a new superblock was written. | |
707 | * There are two reasons we might not have a superblock: | |
708 | * 1) The array is brand new - in which case, all of the | |
709 | * devices must have their In_sync bit set. Also, | |
710 | * recovery_cp must be 0, unless forced. | |
711 | * 2) This is a new device being added to an old array | |
712 | * and the new device needs to be rebuilt - in which | |
713 | * case the In_sync bit will /not/ be set and | |
714 | * recovery_cp must be MaxSector. | |
715 | */ | |
716 | rdev_for_each(r, t, mddev) { | |
717 | if (!test_bit(In_sync, &r->flags)) { | |
718 | if (!test_bit(FirstUse, &r->flags)) | |
719 | DMERR("Superblock area of " | |
720 | "rebuild device %d should have been " | |
721 | "cleared.", r->raid_disk); | |
722 | set_bit(FirstUse, &r->flags); | |
723 | rebuilds++; | |
724 | } else if (test_bit(FirstUse, &r->flags)) | |
725 | new_devs++; | |
726 | } | |
727 | ||
728 | if (!rebuilds) { | |
729 | if (new_devs == mddev->raid_disks) { | |
730 | DMINFO("Superblocks created for new array"); | |
731 | set_bit(MD_ARRAY_FIRST_USE, &mddev->flags); | |
732 | } else if (new_devs) { | |
733 | DMERR("New device injected " | |
734 | "into existing array without 'rebuild' " | |
735 | "parameter specified"); | |
736 | return -EINVAL; | |
737 | } | |
738 | } else if (new_devs) { | |
739 | DMERR("'rebuild' devices cannot be " | |
740 | "injected into an array with other first-time devices"); | |
741 | return -EINVAL; | |
742 | } else if (mddev->recovery_cp != MaxSector) { | |
743 | DMERR("'rebuild' specified while array is not in-sync"); | |
744 | return -EINVAL; | |
745 | } | |
746 | ||
747 | /* | |
748 | * Now we set the Faulty bit for those devices that are | |
749 | * recorded in the superblock as failed. | |
750 | */ | |
751 | rdev_for_each(r, t, mddev) { | |
752 | if (!r->sb_page) | |
753 | continue; | |
754 | sb2 = page_address(r->sb_page); | |
755 | sb2->failed_devices = 0; | |
756 | ||
757 | /* | |
758 | * Check for any device re-ordering. | |
759 | */ | |
760 | if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) { | |
761 | role = le32_to_cpu(sb2->array_position); | |
762 | if (role != r->raid_disk) { | |
763 | if (rs->raid_type->level != 1) { | |
764 | rs->ti->error = "Cannot change device " | |
765 | "positions in RAID array"; | |
766 | return -EINVAL; | |
767 | } | |
768 | DMINFO("RAID1 device #%d now at position #%d", | |
769 | role, r->raid_disk); | |
770 | } | |
771 | ||
772 | /* | |
773 | * Partial recovery is performed on | |
774 | * returning failed devices. | |
775 | */ | |
776 | if (failed_devices & (1 << role)) | |
777 | set_bit(Faulty, &r->flags); | |
778 | } | |
779 | } | |
780 | ||
781 | return 0; | |
782 | } | |
783 | ||
784 | static int super_validate(mddev_t *mddev, mdk_rdev_t *rdev) | |
785 | { | |
786 | struct dm_raid_superblock *sb = page_address(rdev->sb_page); | |
787 | ||
788 | /* | |
789 | * If mddev->events is not set, we know we have not yet initialized | |
790 | * the array. | |
791 | */ | |
792 | if (!mddev->events && super_init_validation(mddev, rdev)) | |
793 | return -EINVAL; | |
794 | ||
795 | mddev->bitmap_info.offset = 4096 >> 9; /* Enable bitmap creation */ | |
796 | rdev->mddev->bitmap_info.default_offset = 4096 >> 9; | |
797 | if (!test_bit(FirstUse, &rdev->flags)) { | |
798 | rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset); | |
799 | if (rdev->recovery_offset != MaxSector) | |
800 | clear_bit(In_sync, &rdev->flags); | |
801 | } | |
802 | ||
803 | /* | |
804 | * If a device comes back, set it as not In_sync and no longer faulty. | |
805 | */ | |
806 | if (test_bit(Faulty, &rdev->flags)) { | |
807 | clear_bit(Faulty, &rdev->flags); | |
808 | clear_bit(In_sync, &rdev->flags); | |
809 | rdev->saved_raid_disk = rdev->raid_disk; | |
810 | rdev->recovery_offset = 0; | |
811 | } | |
812 | ||
813 | clear_bit(FirstUse, &rdev->flags); | |
814 | ||
815 | return 0; | |
816 | } | |
817 | ||
818 | /* | |
819 | * Analyse superblocks and select the freshest. | |
820 | */ | |
821 | static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs) | |
822 | { | |
823 | int ret; | |
824 | mdk_rdev_t *rdev, *freshest, *tmp; | |
825 | mddev_t *mddev = &rs->md; | |
826 | ||
827 | freshest = NULL; | |
828 | rdev_for_each(rdev, tmp, mddev) { | |
829 | if (!rdev->meta_bdev) | |
830 | continue; | |
831 | ||
832 | ret = super_load(rdev, freshest); | |
833 | ||
834 | switch (ret) { | |
835 | case 1: | |
836 | freshest = rdev; | |
837 | break; | |
838 | case 0: | |
839 | break; | |
840 | default: | |
841 | ti->error = "Failed to load superblock"; | |
842 | return ret; | |
843 | } | |
844 | } | |
845 | ||
846 | if (!freshest) | |
847 | return 0; | |
848 | ||
849 | /* | |
850 | * Validation of the freshest device provides the source of | |
851 | * validation for the remaining devices. | |
852 | */ | |
853 | ti->error = "Unable to assemble array: Invalid superblocks"; | |
854 | if (super_validate(mddev, freshest)) | |
855 | return -EINVAL; | |
856 | ||
857 | rdev_for_each(rdev, tmp, mddev) | |
858 | if ((rdev != freshest) && super_validate(mddev, rdev)) | |
859 | return -EINVAL; | |
860 | ||
861 | return 0; | |
862 | } | |
863 | ||
9d09e663 N |
864 | /* |
865 | * Construct a RAID4/5/6 mapping: | |
866 | * Args: | |
867 | * <raid_type> <#raid_params> <raid_params> \ | |
868 | * <#raid_devs> { <meta_dev1> <dev1> .. <meta_devN> <devN> } | |
869 | * | |
9d09e663 N |
870 | * <raid_params> varies by <raid_type>. See 'parse_raid_params' for |
871 | * details on possible <raid_params>. | |
872 | */ | |
873 | static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
874 | { | |
875 | int ret; | |
876 | struct raid_type *rt; | |
877 | unsigned long num_raid_params, num_raid_devs; | |
878 | struct raid_set *rs = NULL; | |
879 | ||
880 | /* Must have at least <raid_type> <#raid_params> */ | |
881 | if (argc < 2) { | |
882 | ti->error = "Too few arguments"; | |
883 | return -EINVAL; | |
884 | } | |
885 | ||
886 | /* raid type */ | |
887 | rt = get_raid_type(argv[0]); | |
888 | if (!rt) { | |
889 | ti->error = "Unrecognised raid_type"; | |
890 | return -EINVAL; | |
891 | } | |
892 | argc--; | |
893 | argv++; | |
894 | ||
895 | /* number of RAID parameters */ | |
896 | if (strict_strtoul(argv[0], 10, &num_raid_params) < 0) { | |
897 | ti->error = "Cannot understand number of RAID parameters"; | |
898 | return -EINVAL; | |
899 | } | |
900 | argc--; | |
901 | argv++; | |
902 | ||
903 | /* Skip over RAID params for now and find out # of devices */ | |
904 | if (num_raid_params + 1 > argc) { | |
905 | ti->error = "Arguments do not agree with counts given"; | |
906 | return -EINVAL; | |
907 | } | |
908 | ||
909 | if ((strict_strtoul(argv[num_raid_params], 10, &num_raid_devs) < 0) || | |
910 | (num_raid_devs >= INT_MAX)) { | |
911 | ti->error = "Cannot understand number of raid devices"; | |
912 | return -EINVAL; | |
913 | } | |
914 | ||
915 | rs = context_alloc(ti, rt, (unsigned)num_raid_devs); | |
916 | if (IS_ERR(rs)) | |
917 | return PTR_ERR(rs); | |
918 | ||
919 | ret = parse_raid_params(rs, argv, (unsigned)num_raid_params); | |
920 | if (ret) | |
921 | goto bad; | |
922 | ||
923 | ret = -EINVAL; | |
924 | ||
925 | argc -= num_raid_params + 1; /* +1: we already have num_raid_devs */ | |
926 | argv += num_raid_params + 1; | |
927 | ||
928 | if (argc != (num_raid_devs * 2)) { | |
929 | ti->error = "Supplied RAID devices does not match the count given"; | |
930 | goto bad; | |
931 | } | |
932 | ||
933 | ret = dev_parms(rs, argv); | |
934 | if (ret) | |
935 | goto bad; | |
936 | ||
b12d437b JB |
937 | rs->md.sync_super = super_sync; |
938 | ret = analyse_superblocks(ti, rs); | |
939 | if (ret) | |
940 | goto bad; | |
941 | ||
9d09e663 | 942 | INIT_WORK(&rs->md.event_work, do_table_event); |
9d09e663 N |
943 | ti->private = rs; |
944 | ||
945 | mutex_lock(&rs->md.reconfig_mutex); | |
946 | ret = md_run(&rs->md); | |
947 | rs->md.in_sync = 0; /* Assume already marked dirty */ | |
948 | mutex_unlock(&rs->md.reconfig_mutex); | |
949 | ||
950 | if (ret) { | |
951 | ti->error = "Fail to run raid array"; | |
952 | goto bad; | |
953 | } | |
954 | ||
955 | rs->callbacks.congested_fn = raid_is_congested; | |
9d09e663 N |
956 | dm_table_add_target_callbacks(ti->table, &rs->callbacks); |
957 | ||
958 | return 0; | |
959 | ||
960 | bad: | |
961 | context_free(rs); | |
962 | ||
963 | return ret; | |
964 | } | |
965 | ||
966 | static void raid_dtr(struct dm_target *ti) | |
967 | { | |
968 | struct raid_set *rs = ti->private; | |
969 | ||
970 | list_del_init(&rs->callbacks.list); | |
971 | md_stop(&rs->md); | |
972 | context_free(rs); | |
973 | } | |
974 | ||
975 | static int raid_map(struct dm_target *ti, struct bio *bio, union map_info *map_context) | |
976 | { | |
977 | struct raid_set *rs = ti->private; | |
978 | mddev_t *mddev = &rs->md; | |
979 | ||
980 | mddev->pers->make_request(mddev, bio); | |
981 | ||
982 | return DM_MAPIO_SUBMITTED; | |
983 | } | |
984 | ||
985 | static int raid_status(struct dm_target *ti, status_type_t type, | |
986 | char *result, unsigned maxlen) | |
987 | { | |
988 | struct raid_set *rs = ti->private; | |
989 | unsigned raid_param_cnt = 1; /* at least 1 for chunksize */ | |
990 | unsigned sz = 0; | |
991 | int i; | |
992 | sector_t sync; | |
993 | ||
994 | switch (type) { | |
995 | case STATUSTYPE_INFO: | |
996 | DMEMIT("%s %d ", rs->raid_type->name, rs->md.raid_disks); | |
997 | ||
998 | for (i = 0; i < rs->md.raid_disks; i++) { | |
999 | if (test_bit(Faulty, &rs->dev[i].rdev.flags)) | |
1000 | DMEMIT("D"); | |
1001 | else if (test_bit(In_sync, &rs->dev[i].rdev.flags)) | |
1002 | DMEMIT("A"); | |
1003 | else | |
1004 | DMEMIT("a"); | |
1005 | } | |
1006 | ||
1007 | if (test_bit(MD_RECOVERY_RUNNING, &rs->md.recovery)) | |
1008 | sync = rs->md.curr_resync_completed; | |
1009 | else | |
1010 | sync = rs->md.recovery_cp; | |
1011 | ||
1012 | if (sync > rs->md.resync_max_sectors) | |
1013 | sync = rs->md.resync_max_sectors; | |
1014 | ||
1015 | DMEMIT(" %llu/%llu", | |
1016 | (unsigned long long) sync, | |
1017 | (unsigned long long) rs->md.resync_max_sectors); | |
1018 | ||
1019 | break; | |
1020 | case STATUSTYPE_TABLE: | |
1021 | /* The string you would use to construct this array */ | |
46bed2b5 | 1022 | for (i = 0; i < rs->md.raid_disks; i++) { |
13c87583 JB |
1023 | if ((rs->print_flags & DMPF_REBUILD) && |
1024 | rs->dev[i].data_dev && | |
9d09e663 | 1025 | !test_bit(In_sync, &rs->dev[i].rdev.flags)) |
13c87583 | 1026 | raid_param_cnt += 2; /* for rebuilds */ |
46bed2b5 JB |
1027 | if (rs->dev[i].data_dev && |
1028 | test_bit(WriteMostly, &rs->dev[i].rdev.flags)) | |
1029 | raid_param_cnt += 2; | |
1030 | } | |
9d09e663 | 1031 | |
13c87583 | 1032 | raid_param_cnt += (hweight64(rs->print_flags & ~DMPF_REBUILD) * 2); |
9d09e663 N |
1033 | if (rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC)) |
1034 | raid_param_cnt--; | |
1035 | ||
1036 | DMEMIT("%s %u %u", rs->raid_type->name, | |
1037 | raid_param_cnt, rs->md.chunk_sectors); | |
1038 | ||
1039 | if ((rs->print_flags & DMPF_SYNC) && | |
1040 | (rs->md.recovery_cp == MaxSector)) | |
1041 | DMEMIT(" sync"); | |
1042 | if (rs->print_flags & DMPF_NOSYNC) | |
1043 | DMEMIT(" nosync"); | |
1044 | ||
1045 | for (i = 0; i < rs->md.raid_disks; i++) | |
13c87583 JB |
1046 | if ((rs->print_flags & DMPF_REBUILD) && |
1047 | rs->dev[i].data_dev && | |
9d09e663 N |
1048 | !test_bit(In_sync, &rs->dev[i].rdev.flags)) |
1049 | DMEMIT(" rebuild %u", i); | |
1050 | ||
1051 | if (rs->print_flags & DMPF_DAEMON_SLEEP) | |
1052 | DMEMIT(" daemon_sleep %lu", | |
1053 | rs->md.bitmap_info.daemon_sleep); | |
1054 | ||
1055 | if (rs->print_flags & DMPF_MIN_RECOVERY_RATE) | |
1056 | DMEMIT(" min_recovery_rate %d", rs->md.sync_speed_min); | |
1057 | ||
1058 | if (rs->print_flags & DMPF_MAX_RECOVERY_RATE) | |
1059 | DMEMIT(" max_recovery_rate %d", rs->md.sync_speed_max); | |
1060 | ||
46bed2b5 JB |
1061 | for (i = 0; i < rs->md.raid_disks; i++) |
1062 | if (rs->dev[i].data_dev && | |
1063 | test_bit(WriteMostly, &rs->dev[i].rdev.flags)) | |
1064 | DMEMIT(" write_mostly %u", i); | |
1065 | ||
9d09e663 N |
1066 | if (rs->print_flags & DMPF_MAX_WRITE_BEHIND) |
1067 | DMEMIT(" max_write_behind %lu", | |
1068 | rs->md.bitmap_info.max_write_behind); | |
1069 | ||
1070 | if (rs->print_flags & DMPF_STRIPE_CACHE) { | |
1071 | raid5_conf_t *conf = rs->md.private; | |
1072 | ||
1073 | /* convert from kiB to sectors */ | |
1074 | DMEMIT(" stripe_cache %d", | |
1075 | conf ? conf->max_nr_stripes * 2 : 0); | |
1076 | } | |
1077 | ||
c1084561 JB |
1078 | if (rs->print_flags & DMPF_REGION_SIZE) |
1079 | DMEMIT(" region_size %lu", | |
1080 | rs->md.bitmap_info.chunksize >> 9); | |
1081 | ||
9d09e663 N |
1082 | DMEMIT(" %d", rs->md.raid_disks); |
1083 | for (i = 0; i < rs->md.raid_disks; i++) { | |
b12d437b JB |
1084 | if (rs->dev[i].meta_dev) |
1085 | DMEMIT(" %s", rs->dev[i].meta_dev->name); | |
1086 | else | |
1087 | DMEMIT(" -"); | |
9d09e663 N |
1088 | |
1089 | if (rs->dev[i].data_dev) | |
1090 | DMEMIT(" %s", rs->dev[i].data_dev->name); | |
1091 | else | |
1092 | DMEMIT(" -"); | |
1093 | } | |
1094 | } | |
1095 | ||
1096 | return 0; | |
1097 | } | |
1098 | ||
1099 | static int raid_iterate_devices(struct dm_target *ti, iterate_devices_callout_fn fn, void *data) | |
1100 | { | |
1101 | struct raid_set *rs = ti->private; | |
1102 | unsigned i; | |
1103 | int ret = 0; | |
1104 | ||
1105 | for (i = 0; !ret && i < rs->md.raid_disks; i++) | |
1106 | if (rs->dev[i].data_dev) | |
1107 | ret = fn(ti, | |
1108 | rs->dev[i].data_dev, | |
1109 | 0, /* No offset on data devs */ | |
1110 | rs->md.dev_sectors, | |
1111 | data); | |
1112 | ||
1113 | return ret; | |
1114 | } | |
1115 | ||
1116 | static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits) | |
1117 | { | |
1118 | struct raid_set *rs = ti->private; | |
1119 | unsigned chunk_size = rs->md.chunk_sectors << 9; | |
1120 | raid5_conf_t *conf = rs->md.private; | |
1121 | ||
1122 | blk_limits_io_min(limits, chunk_size); | |
1123 | blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded)); | |
1124 | } | |
1125 | ||
1126 | static void raid_presuspend(struct dm_target *ti) | |
1127 | { | |
1128 | struct raid_set *rs = ti->private; | |
1129 | ||
1130 | md_stop_writes(&rs->md); | |
1131 | } | |
1132 | ||
1133 | static void raid_postsuspend(struct dm_target *ti) | |
1134 | { | |
1135 | struct raid_set *rs = ti->private; | |
1136 | ||
1137 | mddev_suspend(&rs->md); | |
1138 | } | |
1139 | ||
1140 | static void raid_resume(struct dm_target *ti) | |
1141 | { | |
1142 | struct raid_set *rs = ti->private; | |
1143 | ||
b12d437b | 1144 | bitmap_load(&rs->md); |
9d09e663 N |
1145 | mddev_resume(&rs->md); |
1146 | } | |
1147 | ||
1148 | static struct target_type raid_target = { | |
1149 | .name = "raid", | |
1150 | .version = {1, 0, 0}, | |
1151 | .module = THIS_MODULE, | |
1152 | .ctr = raid_ctr, | |
1153 | .dtr = raid_dtr, | |
1154 | .map = raid_map, | |
1155 | .status = raid_status, | |
1156 | .iterate_devices = raid_iterate_devices, | |
1157 | .io_hints = raid_io_hints, | |
1158 | .presuspend = raid_presuspend, | |
1159 | .postsuspend = raid_postsuspend, | |
1160 | .resume = raid_resume, | |
1161 | }; | |
1162 | ||
1163 | static int __init dm_raid_init(void) | |
1164 | { | |
1165 | return dm_register_target(&raid_target); | |
1166 | } | |
1167 | ||
1168 | static void __exit dm_raid_exit(void) | |
1169 | { | |
1170 | dm_unregister_target(&raid_target); | |
1171 | } | |
1172 | ||
1173 | module_init(dm_raid_init); | |
1174 | module_exit(dm_raid_exit); | |
1175 | ||
1176 | MODULE_DESCRIPTION(DM_NAME " raid4/5/6 target"); | |
1177 | MODULE_ALIAS("dm-raid4"); | |
1178 | MODULE_ALIAS("dm-raid5"); | |
1179 | MODULE_ALIAS("dm-raid6"); | |
1180 | MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>"); | |
1181 | MODULE_LICENSE("GPL"); |