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> | |
056075c7 | 9 | #include <linux/module.h> |
9d09e663 N |
10 | |
11 | #include "md.h" | |
32737279 | 12 | #include "raid1.h" |
9d09e663 | 13 | #include "raid5.h" |
63f33b8d | 14 | #include "raid10.h" |
9d09e663 N |
15 | #include "bitmap.h" |
16 | ||
3e8dbb7f AK |
17 | #include <linux/device-mapper.h> |
18 | ||
9d09e663 N |
19 | #define DM_MSG_PREFIX "raid" |
20 | ||
21 | /* | |
b12d437b JB |
22 | * 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. | |
9d09e663 | 24 | */ |
b12d437b | 25 | #define FirstUse 10 /* rdev flag */ |
9d09e663 N |
26 | |
27 | struct 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; | |
3cb03002 | 42 | struct md_rdev rdev; |
9d09e663 N |
43 | }; |
44 | ||
45 | /* | |
46 | * Flags for rs->print_flags field. | |
47 | */ | |
13c87583 JB |
48 | #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 | |
63f33b8d JB |
56 | #define DMPF_REGION_SIZE 0x100 |
57 | #define DMPF_RAID10_COPIES 0x200 | |
58 | #define DMPF_RAID10_FORMAT 0x400 | |
59 | ||
9d09e663 N |
60 | struct raid_set { |
61 | struct dm_target *ti; | |
62 | ||
34f8ac6d JB |
63 | uint32_t bitmap_loaded; |
64 | uint32_t print_flags; | |
9d09e663 | 65 | |
fd01b88c | 66 | struct mddev md; |
9d09e663 N |
67 | 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. */ | |
74 | static 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[] = { | |
32737279 | 82 | {"raid1", "RAID1 (mirroring)", 0, 2, 1, 0 /* NONE */}, |
63f33b8d | 83 | {"raid10", "RAID10 (striped mirrors)", 0, 2, 10, UINT_MAX /* Varies */}, |
9d09e663 N |
84 | {"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 | ||
63f33b8d JB |
94 | static unsigned raid10_md_layout_to_copies(int layout) |
95 | { | |
96 | return layout & 0xFF; | |
97 | } | |
98 | ||
99 | static 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 | ||
9d09e663 N |
105 | static 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 | ||
116 | static 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; | |
9d09e663 N |
120 | |
121 | if (raid_devs <= raid_type->parity_devs) { | |
122 | ti->error = "Insufficient number of devices"; | |
123 | return ERR_PTR(-EINVAL); | |
124 | } | |
125 | ||
9d09e663 N |
126 | 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; | |
9d09e663 N |
139 | 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 | |
c039c332 | 153 | * rs->md.dev_sectors |
9d09e663 N |
154 | */ |
155 | ||
156 | return rs; | |
157 | } | |
158 | ||
159 | static void context_free(struct raid_set *rs) | |
160 | { | |
161 | int i; | |
162 | ||
b12d437b JB |
163 | 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); | |
545c8795 | 166 | md_rdev_clear(&rs->dev[i].rdev); |
9d09e663 N |
167 | if (rs->dev[i].data_dev) |
168 | dm_put_device(rs->ti, rs->dev[i].data_dev); | |
b12d437b | 169 | } |
9d09e663 N |
170 | |
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 | * | |
b12d437b JB |
179 | * 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. | |
9d09e663 N |
189 | */ |
190 | static 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], "-")) { | |
b12d437b JB |
211 | 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; | |
9d09e663 N |
221 | } |
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 | ||
b12d437b JB |
230 | rs->ti->error = "No data device supplied with metadata device"; |
231 | if (rs->dev[i].meta_dev) | |
232 | return -EINVAL; | |
233 | ||
9d09e663 N |
234 | 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 | ||
b12d437b JB |
245 | if (rs->dev[i].meta_dev) { |
246 | metadata_available = 1; | |
247 | rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev; | |
248 | } | |
9d09e663 N |
249 | 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 | ||
c1084561 JB |
279 | /* |
280 | * 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 | */ | |
289 | static 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)) { | |
298 | DMINFO("Choosing default region size of %lu sectors", | |
299 | region_size); | |
300 | region_size = min_region_size; | |
301 | } else { | |
302 | DMINFO("Choosing default region size of 4MiB"); | |
303 | region_size = 1 << 13; /* sectors */ | |
304 | } | |
305 | } else { | |
306 | /* | |
307 | * Validate user-supplied value. | |
308 | */ | |
309 | if (region_size > rs->ti->len) { | |
310 | rs->ti->error = "Supplied region size is too large"; | |
311 | return -EINVAL; | |
312 | } | |
313 | ||
314 | if (region_size < min_region_size) { | |
315 | DMERR("Supplied region_size (%lu sectors) below minimum (%lu)", | |
316 | region_size, min_region_size); | |
317 | rs->ti->error = "Supplied region size is too small"; | |
318 | return -EINVAL; | |
319 | } | |
320 | ||
321 | if (!is_power_of_2(region_size)) { | |
322 | rs->ti->error = "Region size is not a power of 2"; | |
323 | return -EINVAL; | |
324 | } | |
325 | ||
326 | if (region_size < rs->md.chunk_sectors) { | |
327 | rs->ti->error = "Region size is smaller than the chunk size"; | |
328 | return -EINVAL; | |
329 | } | |
330 | } | |
331 | ||
332 | /* | |
333 | * Convert sectors to bytes. | |
334 | */ | |
335 | rs->md.bitmap_info.chunksize = (region_size << 9); | |
336 | ||
337 | return 0; | |
338 | } | |
339 | ||
eb649123 JB |
340 | /* |
341 | * validate_rebuild_devices | |
342 | * @rs | |
343 | * | |
344 | * Determine if the devices specified for rebuild can result in a valid | |
345 | * usable array that is capable of rebuilding the given devices. | |
346 | * | |
347 | * Returns: 0 on success, -EINVAL on failure. | |
348 | */ | |
349 | static int validate_rebuild_devices(struct raid_set *rs) | |
350 | { | |
351 | unsigned i, rebuild_cnt = 0; | |
4ec1e369 | 352 | unsigned rebuilds_per_group, copies, d; |
eb649123 JB |
353 | |
354 | if (!(rs->print_flags & DMPF_REBUILD)) | |
355 | return 0; | |
356 | ||
357 | for (i = 0; i < rs->md.raid_disks; i++) | |
358 | if (!test_bit(In_sync, &rs->dev[i].rdev.flags)) | |
359 | 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: | |
4ec1e369 JB |
373 | 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 | */ | |
394 | rebuilds_per_group = 0; | |
395 | for (i = 0; i < rs->md.raid_disks * copies; i++) { | |
396 | d = i % rs->md.raid_disks; | |
397 | if (!test_bit(In_sync, &rs->dev[d].rdev.flags) && | |
398 | (++rebuilds_per_group >= copies)) | |
399 | goto too_many; | |
400 | if (!((i + 1) % copies)) | |
401 | rebuilds_per_group = 0; | |
402 | } | |
403 | break; | |
eb649123 JB |
404 | default: |
405 | DMERR("The rebuild parameter is not supported for %s", | |
406 | rs->raid_type->name); | |
407 | rs->ti->error = "Rebuild not supported for this RAID type"; | |
408 | return -EINVAL; | |
409 | } | |
410 | ||
411 | return 0; | |
412 | ||
413 | too_many: | |
414 | rs->ti->error = "Too many rebuild devices specified"; | |
415 | return -EINVAL; | |
416 | } | |
417 | ||
9d09e663 N |
418 | /* |
419 | * Possible arguments are... | |
9d09e663 N |
420 | * <chunk_size> [optional_args] |
421 | * | |
32737279 JB |
422 | * Argument definitions |
423 | * <chunk_size> The number of sectors per disk that | |
424 | * will form the "stripe" | |
425 | * [[no]sync] Force or prevent recovery of the | |
426 | * entire array | |
9d09e663 | 427 | * [rebuild <idx>] Rebuild the drive indicated by the index |
32737279 JB |
428 | * [daemon_sleep <ms>] Time between bitmap daemon work to |
429 | * clear bits | |
9d09e663 N |
430 | * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization |
431 | * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization | |
46bed2b5 | 432 | * [write_mostly <idx>] Indicate a write mostly drive via index |
9d09e663 N |
433 | * [max_write_behind <sectors>] See '-write-behind=' (man mdadm) |
434 | * [stripe_cache <sectors>] Stripe cache size for higher RAIDs | |
c1084561 | 435 | * [region_size <sectors>] Defines granularity of bitmap |
63f33b8d JB |
436 | * |
437 | * RAID10-only options: | |
438 | * [raid10_copies <# copies>] Number of copies. (Default: 2) | |
439 | * [raid10_format <near>] Layout algorithm. (Default: near) | |
9d09e663 N |
440 | */ |
441 | static int parse_raid_params(struct raid_set *rs, char **argv, | |
442 | unsigned num_raid_params) | |
443 | { | |
63f33b8d JB |
444 | char *raid10_format = "near"; |
445 | unsigned raid10_copies = 2; | |
eb649123 | 446 | unsigned i; |
c1084561 | 447 | unsigned long value, region_size = 0; |
c039c332 | 448 | sector_t sectors_per_dev = rs->ti->len; |
542f9038 | 449 | sector_t max_io_len; |
9d09e663 N |
450 | char *key; |
451 | ||
452 | /* | |
453 | * First, parse the in-order required arguments | |
32737279 | 454 | * "chunk_size" is the only argument of this type. |
9d09e663 | 455 | */ |
32737279 | 456 | if ((strict_strtoul(argv[0], 10, &value) < 0)) { |
9d09e663 N |
457 | rs->ti->error = "Bad chunk size"; |
458 | return -EINVAL; | |
32737279 JB |
459 | } else if (rs->raid_type->level == 1) { |
460 | if (value) | |
461 | DMERR("Ignoring chunk size parameter for RAID 1"); | |
462 | value = 0; | |
463 | } else if (!is_power_of_2(value)) { | |
464 | rs->ti->error = "Chunk size must be a power of 2"; | |
465 | return -EINVAL; | |
466 | } else if (value < 8) { | |
467 | rs->ti->error = "Chunk size value is too small"; | |
468 | return -EINVAL; | |
9d09e663 N |
469 | } |
470 | ||
471 | rs->md.new_chunk_sectors = rs->md.chunk_sectors = value; | |
472 | argv++; | |
473 | num_raid_params--; | |
474 | ||
475 | /* | |
b12d437b JB |
476 | * We set each individual device as In_sync with a completed |
477 | * 'recovery_offset'. If there has been a device failure or | |
478 | * replacement then one of the following cases applies: | |
479 | * | |
480 | * 1) User specifies 'rebuild'. | |
481 | * - Device is reset when param is read. | |
482 | * 2) A new device is supplied. | |
483 | * - No matching superblock found, resets device. | |
484 | * 3) Device failure was transient and returns on reload. | |
485 | * - Failure noticed, resets device for bitmap replay. | |
486 | * 4) Device hadn't completed recovery after previous failure. | |
487 | * - Superblock is read and overrides recovery_offset. | |
488 | * | |
489 | * What is found in the superblocks of the devices is always | |
490 | * authoritative, unless 'rebuild' or '[no]sync' was specified. | |
9d09e663 | 491 | */ |
b12d437b | 492 | for (i = 0; i < rs->md.raid_disks; i++) { |
9d09e663 | 493 | set_bit(In_sync, &rs->dev[i].rdev.flags); |
b12d437b JB |
494 | rs->dev[i].rdev.recovery_offset = MaxSector; |
495 | } | |
9d09e663 | 496 | |
b12d437b JB |
497 | /* |
498 | * Second, parse the unordered optional arguments | |
499 | */ | |
9d09e663 | 500 | for (i = 0; i < num_raid_params; i++) { |
13c87583 | 501 | if (!strcasecmp(argv[i], "nosync")) { |
9d09e663 N |
502 | rs->md.recovery_cp = MaxSector; |
503 | rs->print_flags |= DMPF_NOSYNC; | |
9d09e663 N |
504 | continue; |
505 | } | |
13c87583 | 506 | if (!strcasecmp(argv[i], "sync")) { |
9d09e663 N |
507 | rs->md.recovery_cp = 0; |
508 | rs->print_flags |= DMPF_SYNC; | |
9d09e663 N |
509 | continue; |
510 | } | |
511 | ||
512 | /* The rest of the optional arguments come in key/value pairs */ | |
513 | if ((i + 1) >= num_raid_params) { | |
514 | rs->ti->error = "Wrong number of raid parameters given"; | |
515 | return -EINVAL; | |
516 | } | |
517 | ||
518 | key = argv[i++]; | |
63f33b8d JB |
519 | |
520 | /* Parameters that take a string value are checked here. */ | |
521 | if (!strcasecmp(key, "raid10_format")) { | |
522 | if (rs->raid_type->level != 10) { | |
523 | rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type"; | |
524 | return -EINVAL; | |
525 | } | |
526 | if (strcmp("near", argv[i])) { | |
527 | rs->ti->error = "Invalid 'raid10_format' value given"; | |
528 | return -EINVAL; | |
529 | } | |
530 | raid10_format = argv[i]; | |
531 | rs->print_flags |= DMPF_RAID10_FORMAT; | |
532 | continue; | |
533 | } | |
534 | ||
9d09e663 N |
535 | if (strict_strtoul(argv[i], 10, &value) < 0) { |
536 | rs->ti->error = "Bad numerical argument given in raid params"; | |
537 | return -EINVAL; | |
538 | } | |
539 | ||
63f33b8d | 540 | /* Parameters that take a numeric value are checked here */ |
13c87583 | 541 | if (!strcasecmp(key, "rebuild")) { |
7386199c | 542 | if (value >= rs->md.raid_disks) { |
9d09e663 N |
543 | rs->ti->error = "Invalid rebuild index given"; |
544 | return -EINVAL; | |
545 | } | |
546 | clear_bit(In_sync, &rs->dev[value].rdev.flags); | |
547 | rs->dev[value].rdev.recovery_offset = 0; | |
13c87583 | 548 | rs->print_flags |= DMPF_REBUILD; |
46bed2b5 JB |
549 | } else if (!strcasecmp(key, "write_mostly")) { |
550 | if (rs->raid_type->level != 1) { | |
551 | rs->ti->error = "write_mostly option is only valid for RAID1"; | |
552 | return -EINVAL; | |
553 | } | |
82324809 | 554 | if (value >= rs->md.raid_disks) { |
46bed2b5 JB |
555 | rs->ti->error = "Invalid write_mostly drive index given"; |
556 | return -EINVAL; | |
557 | } | |
558 | set_bit(WriteMostly, &rs->dev[value].rdev.flags); | |
13c87583 | 559 | } else if (!strcasecmp(key, "max_write_behind")) { |
46bed2b5 JB |
560 | if (rs->raid_type->level != 1) { |
561 | rs->ti->error = "max_write_behind option is only valid for RAID1"; | |
562 | return -EINVAL; | |
563 | } | |
9d09e663 N |
564 | rs->print_flags |= DMPF_MAX_WRITE_BEHIND; |
565 | ||
566 | /* | |
567 | * In device-mapper, we specify things in sectors, but | |
568 | * MD records this value in kB | |
569 | */ | |
570 | value /= 2; | |
571 | if (value > COUNTER_MAX) { | |
572 | rs->ti->error = "Max write-behind limit out of range"; | |
573 | return -EINVAL; | |
574 | } | |
575 | rs->md.bitmap_info.max_write_behind = value; | |
13c87583 | 576 | } else if (!strcasecmp(key, "daemon_sleep")) { |
9d09e663 N |
577 | rs->print_flags |= DMPF_DAEMON_SLEEP; |
578 | if (!value || (value > MAX_SCHEDULE_TIMEOUT)) { | |
579 | rs->ti->error = "daemon sleep period out of range"; | |
580 | return -EINVAL; | |
581 | } | |
582 | rs->md.bitmap_info.daemon_sleep = value; | |
13c87583 | 583 | } else if (!strcasecmp(key, "stripe_cache")) { |
9d09e663 N |
584 | rs->print_flags |= DMPF_STRIPE_CACHE; |
585 | ||
586 | /* | |
587 | * In device-mapper, we specify things in sectors, but | |
588 | * MD records this value in kB | |
589 | */ | |
590 | value /= 2; | |
591 | ||
63f33b8d JB |
592 | if ((rs->raid_type->level != 5) && |
593 | (rs->raid_type->level != 6)) { | |
9d09e663 N |
594 | rs->ti->error = "Inappropriate argument: stripe_cache"; |
595 | return -EINVAL; | |
596 | } | |
597 | if (raid5_set_cache_size(&rs->md, (int)value)) { | |
598 | rs->ti->error = "Bad stripe_cache size"; | |
599 | return -EINVAL; | |
600 | } | |
13c87583 | 601 | } else if (!strcasecmp(key, "min_recovery_rate")) { |
9d09e663 N |
602 | rs->print_flags |= DMPF_MIN_RECOVERY_RATE; |
603 | if (value > INT_MAX) { | |
604 | rs->ti->error = "min_recovery_rate out of range"; | |
605 | return -EINVAL; | |
606 | } | |
607 | rs->md.sync_speed_min = (int)value; | |
13c87583 | 608 | } else if (!strcasecmp(key, "max_recovery_rate")) { |
9d09e663 N |
609 | rs->print_flags |= DMPF_MAX_RECOVERY_RATE; |
610 | if (value > INT_MAX) { | |
611 | rs->ti->error = "max_recovery_rate out of range"; | |
612 | return -EINVAL; | |
613 | } | |
614 | rs->md.sync_speed_max = (int)value; | |
c1084561 JB |
615 | } else if (!strcasecmp(key, "region_size")) { |
616 | rs->print_flags |= DMPF_REGION_SIZE; | |
617 | region_size = value; | |
63f33b8d JB |
618 | } else if (!strcasecmp(key, "raid10_copies") && |
619 | (rs->raid_type->level == 10)) { | |
620 | if ((value < 2) || (value > 0xFF)) { | |
621 | rs->ti->error = "Bad value for 'raid10_copies'"; | |
622 | return -EINVAL; | |
623 | } | |
624 | rs->print_flags |= DMPF_RAID10_COPIES; | |
625 | raid10_copies = value; | |
9d09e663 N |
626 | } else { |
627 | DMERR("Unable to parse RAID parameter: %s", key); | |
628 | rs->ti->error = "Unable to parse RAID parameters"; | |
629 | return -EINVAL; | |
630 | } | |
631 | } | |
632 | ||
c1084561 JB |
633 | if (validate_region_size(rs, region_size)) |
634 | return -EINVAL; | |
635 | ||
636 | if (rs->md.chunk_sectors) | |
542f9038 | 637 | max_io_len = rs->md.chunk_sectors; |
c1084561 | 638 | else |
542f9038 | 639 | max_io_len = region_size; |
c1084561 | 640 | |
542f9038 MS |
641 | if (dm_set_target_max_io_len(rs->ti, max_io_len)) |
642 | return -EINVAL; | |
32737279 | 643 | |
63f33b8d JB |
644 | if (rs->raid_type->level == 10) { |
645 | if (raid10_copies > rs->md.raid_disks) { | |
646 | rs->ti->error = "Not enough devices to satisfy specification"; | |
647 | return -EINVAL; | |
648 | } | |
649 | ||
650 | /* (Len * #mirrors) / #devices */ | |
651 | sectors_per_dev = rs->ti->len * raid10_copies; | |
652 | sector_div(sectors_per_dev, rs->md.raid_disks); | |
653 | ||
654 | rs->md.layout = raid10_format_to_md_layout(raid10_format, | |
655 | raid10_copies); | |
656 | rs->md.new_layout = rs->md.layout; | |
657 | } else if ((rs->raid_type->level > 1) && | |
658 | sector_div(sectors_per_dev, | |
659 | (rs->md.raid_disks - rs->raid_type->parity_devs))) { | |
c039c332 JB |
660 | rs->ti->error = "Target length not divisible by number of data devices"; |
661 | return -EINVAL; | |
662 | } | |
663 | rs->md.dev_sectors = sectors_per_dev; | |
664 | ||
eb649123 JB |
665 | if (validate_rebuild_devices(rs)) |
666 | return -EINVAL; | |
667 | ||
9d09e663 N |
668 | /* Assume there are no metadata devices until the drives are parsed */ |
669 | rs->md.persistent = 0; | |
670 | rs->md.external = 1; | |
671 | ||
672 | return 0; | |
673 | } | |
674 | ||
675 | static void do_table_event(struct work_struct *ws) | |
676 | { | |
677 | struct raid_set *rs = container_of(ws, struct raid_set, md.event_work); | |
678 | ||
679 | dm_table_event(rs->ti->table); | |
680 | } | |
681 | ||
682 | static int raid_is_congested(struct dm_target_callbacks *cb, int bits) | |
683 | { | |
684 | struct raid_set *rs = container_of(cb, struct raid_set, callbacks); | |
685 | ||
32737279 JB |
686 | if (rs->raid_type->level == 1) |
687 | return md_raid1_congested(&rs->md, bits); | |
688 | ||
63f33b8d JB |
689 | if (rs->raid_type->level == 10) |
690 | return md_raid10_congested(&rs->md, bits); | |
691 | ||
9d09e663 N |
692 | return md_raid5_congested(&rs->md, bits); |
693 | } | |
694 | ||
b12d437b JB |
695 | /* |
696 | * This structure is never routinely used by userspace, unlike md superblocks. | |
697 | * Devices with this superblock should only ever be accessed via device-mapper. | |
698 | */ | |
699 | #define DM_RAID_MAGIC 0x64526D44 | |
700 | struct dm_raid_superblock { | |
701 | __le32 magic; /* "DmRd" */ | |
702 | __le32 features; /* Used to indicate possible future changes */ | |
703 | ||
704 | __le32 num_devices; /* Number of devices in this array. (Max 64) */ | |
705 | __le32 array_position; /* The position of this drive in the array */ | |
706 | ||
707 | __le64 events; /* Incremented by md when superblock updated */ | |
708 | __le64 failed_devices; /* Bit field of devices to indicate failures */ | |
709 | ||
710 | /* | |
711 | * This offset tracks the progress of the repair or replacement of | |
712 | * an individual drive. | |
713 | */ | |
714 | __le64 disk_recovery_offset; | |
715 | ||
716 | /* | |
717 | * This offset tracks the progress of the initial array | |
718 | * synchronisation/parity calculation. | |
719 | */ | |
720 | __le64 array_resync_offset; | |
721 | ||
722 | /* | |
723 | * RAID characteristics | |
724 | */ | |
725 | __le32 level; | |
726 | __le32 layout; | |
727 | __le32 stripe_sectors; | |
728 | ||
729 | __u8 pad[452]; /* Round struct to 512 bytes. */ | |
730 | /* Always set to 0 when writing. */ | |
731 | } __packed; | |
732 | ||
3cb03002 | 733 | static int read_disk_sb(struct md_rdev *rdev, int size) |
b12d437b JB |
734 | { |
735 | BUG_ON(!rdev->sb_page); | |
736 | ||
737 | if (rdev->sb_loaded) | |
738 | return 0; | |
739 | ||
740 | if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, 1)) { | |
0447568f JB |
741 | DMERR("Failed to read superblock of device at position %d", |
742 | rdev->raid_disk); | |
c32fb9e7 | 743 | md_error(rdev->mddev, rdev); |
b12d437b JB |
744 | return -EINVAL; |
745 | } | |
746 | ||
747 | rdev->sb_loaded = 1; | |
748 | ||
749 | return 0; | |
750 | } | |
751 | ||
fd01b88c | 752 | static void super_sync(struct mddev *mddev, struct md_rdev *rdev) |
b12d437b | 753 | { |
81f382f9 | 754 | int i; |
b12d437b JB |
755 | uint64_t failed_devices; |
756 | struct dm_raid_superblock *sb; | |
81f382f9 | 757 | struct raid_set *rs = container_of(mddev, struct raid_set, md); |
b12d437b JB |
758 | |
759 | sb = page_address(rdev->sb_page); | |
760 | failed_devices = le64_to_cpu(sb->failed_devices); | |
761 | ||
81f382f9 JB |
762 | for (i = 0; i < mddev->raid_disks; i++) |
763 | if (!rs->dev[i].data_dev || | |
764 | test_bit(Faulty, &(rs->dev[i].rdev.flags))) | |
765 | failed_devices |= (1ULL << i); | |
b12d437b JB |
766 | |
767 | memset(sb, 0, sizeof(*sb)); | |
768 | ||
769 | sb->magic = cpu_to_le32(DM_RAID_MAGIC); | |
770 | sb->features = cpu_to_le32(0); /* No features yet */ | |
771 | ||
772 | sb->num_devices = cpu_to_le32(mddev->raid_disks); | |
773 | sb->array_position = cpu_to_le32(rdev->raid_disk); | |
774 | ||
775 | sb->events = cpu_to_le64(mddev->events); | |
776 | sb->failed_devices = cpu_to_le64(failed_devices); | |
777 | ||
778 | sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset); | |
779 | sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp); | |
780 | ||
781 | sb->level = cpu_to_le32(mddev->level); | |
782 | sb->layout = cpu_to_le32(mddev->layout); | |
783 | sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors); | |
784 | } | |
785 | ||
786 | /* | |
787 | * super_load | |
788 | * | |
789 | * This function creates a superblock if one is not found on the device | |
790 | * and will decide which superblock to use if there's a choice. | |
791 | * | |
792 | * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise | |
793 | */ | |
3cb03002 | 794 | static int super_load(struct md_rdev *rdev, struct md_rdev *refdev) |
b12d437b JB |
795 | { |
796 | int ret; | |
797 | struct dm_raid_superblock *sb; | |
798 | struct dm_raid_superblock *refsb; | |
799 | uint64_t events_sb, events_refsb; | |
800 | ||
801 | rdev->sb_start = 0; | |
802 | rdev->sb_size = sizeof(*sb); | |
803 | ||
804 | ret = read_disk_sb(rdev, rdev->sb_size); | |
805 | if (ret) | |
806 | return ret; | |
807 | ||
808 | sb = page_address(rdev->sb_page); | |
3aa3b2b2 JB |
809 | |
810 | /* | |
811 | * Two cases that we want to write new superblocks and rebuild: | |
812 | * 1) New device (no matching magic number) | |
813 | * 2) Device specified for rebuild (!In_sync w/ offset == 0) | |
814 | */ | |
815 | if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) || | |
816 | (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) { | |
b12d437b JB |
817 | super_sync(rdev->mddev, rdev); |
818 | ||
819 | set_bit(FirstUse, &rdev->flags); | |
820 | ||
821 | /* Force writing of superblocks to disk */ | |
822 | set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags); | |
823 | ||
824 | /* Any superblock is better than none, choose that if given */ | |
825 | return refdev ? 0 : 1; | |
826 | } | |
827 | ||
828 | if (!refdev) | |
829 | return 1; | |
830 | ||
831 | events_sb = le64_to_cpu(sb->events); | |
832 | ||
833 | refsb = page_address(refdev->sb_page); | |
834 | events_refsb = le64_to_cpu(refsb->events); | |
835 | ||
836 | return (events_sb > events_refsb) ? 1 : 0; | |
837 | } | |
838 | ||
fd01b88c | 839 | static int super_init_validation(struct mddev *mddev, struct md_rdev *rdev) |
b12d437b JB |
840 | { |
841 | int role; | |
842 | struct raid_set *rs = container_of(mddev, struct raid_set, md); | |
843 | uint64_t events_sb; | |
844 | uint64_t failed_devices; | |
845 | struct dm_raid_superblock *sb; | |
846 | uint32_t new_devs = 0; | |
847 | uint32_t rebuilds = 0; | |
dafb20fa | 848 | struct md_rdev *r; |
b12d437b JB |
849 | struct dm_raid_superblock *sb2; |
850 | ||
851 | sb = page_address(rdev->sb_page); | |
852 | events_sb = le64_to_cpu(sb->events); | |
853 | failed_devices = le64_to_cpu(sb->failed_devices); | |
854 | ||
855 | /* | |
856 | * Initialise to 1 if this is a new superblock. | |
857 | */ | |
858 | mddev->events = events_sb ? : 1; | |
859 | ||
860 | /* | |
861 | * Reshaping is not currently allowed | |
862 | */ | |
863 | if ((le32_to_cpu(sb->level) != mddev->level) || | |
864 | (le32_to_cpu(sb->layout) != mddev->layout) || | |
865 | (le32_to_cpu(sb->stripe_sectors) != mddev->chunk_sectors)) { | |
866 | DMERR("Reshaping arrays not yet supported."); | |
867 | return -EINVAL; | |
868 | } | |
869 | ||
870 | /* We can only change the number of devices in RAID1 right now */ | |
871 | if ((rs->raid_type->level != 1) && | |
872 | (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) { | |
873 | DMERR("Reshaping arrays not yet supported."); | |
874 | return -EINVAL; | |
875 | } | |
876 | ||
877 | if (!(rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC))) | |
878 | mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset); | |
879 | ||
880 | /* | |
881 | * During load, we set FirstUse if a new superblock was written. | |
882 | * There are two reasons we might not have a superblock: | |
883 | * 1) The array is brand new - in which case, all of the | |
884 | * devices must have their In_sync bit set. Also, | |
885 | * recovery_cp must be 0, unless forced. | |
886 | * 2) This is a new device being added to an old array | |
887 | * and the new device needs to be rebuilt - in which | |
888 | * case the In_sync bit will /not/ be set and | |
889 | * recovery_cp must be MaxSector. | |
890 | */ | |
dafb20fa | 891 | rdev_for_each(r, mddev) { |
b12d437b | 892 | if (!test_bit(In_sync, &r->flags)) { |
3aa3b2b2 JB |
893 | DMINFO("Device %d specified for rebuild: " |
894 | "Clearing superblock", r->raid_disk); | |
b12d437b JB |
895 | rebuilds++; |
896 | } else if (test_bit(FirstUse, &r->flags)) | |
897 | new_devs++; | |
898 | } | |
899 | ||
900 | if (!rebuilds) { | |
901 | if (new_devs == mddev->raid_disks) { | |
902 | DMINFO("Superblocks created for new array"); | |
903 | set_bit(MD_ARRAY_FIRST_USE, &mddev->flags); | |
904 | } else if (new_devs) { | |
905 | DMERR("New device injected " | |
906 | "into existing array without 'rebuild' " | |
907 | "parameter specified"); | |
908 | return -EINVAL; | |
909 | } | |
910 | } else if (new_devs) { | |
911 | DMERR("'rebuild' devices cannot be " | |
912 | "injected into an array with other first-time devices"); | |
913 | return -EINVAL; | |
914 | } else if (mddev->recovery_cp != MaxSector) { | |
915 | DMERR("'rebuild' specified while array is not in-sync"); | |
916 | return -EINVAL; | |
917 | } | |
918 | ||
919 | /* | |
920 | * Now we set the Faulty bit for those devices that are | |
921 | * recorded in the superblock as failed. | |
922 | */ | |
dafb20fa | 923 | rdev_for_each(r, mddev) { |
b12d437b JB |
924 | if (!r->sb_page) |
925 | continue; | |
926 | sb2 = page_address(r->sb_page); | |
927 | sb2->failed_devices = 0; | |
928 | ||
929 | /* | |
930 | * Check for any device re-ordering. | |
931 | */ | |
932 | if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) { | |
933 | role = le32_to_cpu(sb2->array_position); | |
934 | if (role != r->raid_disk) { | |
935 | if (rs->raid_type->level != 1) { | |
936 | rs->ti->error = "Cannot change device " | |
937 | "positions in RAID array"; | |
938 | return -EINVAL; | |
939 | } | |
940 | DMINFO("RAID1 device #%d now at position #%d", | |
941 | role, r->raid_disk); | |
942 | } | |
943 | ||
944 | /* | |
945 | * Partial recovery is performed on | |
946 | * returning failed devices. | |
947 | */ | |
948 | if (failed_devices & (1 << role)) | |
949 | set_bit(Faulty, &r->flags); | |
950 | } | |
951 | } | |
952 | ||
953 | return 0; | |
954 | } | |
955 | ||
fd01b88c | 956 | static int super_validate(struct mddev *mddev, struct md_rdev *rdev) |
b12d437b JB |
957 | { |
958 | struct dm_raid_superblock *sb = page_address(rdev->sb_page); | |
959 | ||
960 | /* | |
961 | * If mddev->events is not set, we know we have not yet initialized | |
962 | * the array. | |
963 | */ | |
964 | if (!mddev->events && super_init_validation(mddev, rdev)) | |
965 | return -EINVAL; | |
966 | ||
967 | mddev->bitmap_info.offset = 4096 >> 9; /* Enable bitmap creation */ | |
968 | rdev->mddev->bitmap_info.default_offset = 4096 >> 9; | |
969 | if (!test_bit(FirstUse, &rdev->flags)) { | |
970 | rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset); | |
971 | if (rdev->recovery_offset != MaxSector) | |
972 | clear_bit(In_sync, &rdev->flags); | |
973 | } | |
974 | ||
975 | /* | |
976 | * If a device comes back, set it as not In_sync and no longer faulty. | |
977 | */ | |
978 | if (test_bit(Faulty, &rdev->flags)) { | |
979 | clear_bit(Faulty, &rdev->flags); | |
980 | clear_bit(In_sync, &rdev->flags); | |
981 | rdev->saved_raid_disk = rdev->raid_disk; | |
982 | rdev->recovery_offset = 0; | |
983 | } | |
984 | ||
985 | clear_bit(FirstUse, &rdev->flags); | |
986 | ||
987 | return 0; | |
988 | } | |
989 | ||
990 | /* | |
991 | * Analyse superblocks and select the freshest. | |
992 | */ | |
993 | static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs) | |
994 | { | |
995 | int ret; | |
0447568f JB |
996 | unsigned redundancy = 0; |
997 | struct raid_dev *dev; | |
a9ad8526 | 998 | struct md_rdev *rdev, *tmp, *freshest; |
fd01b88c | 999 | struct mddev *mddev = &rs->md; |
b12d437b | 1000 | |
0447568f JB |
1001 | switch (rs->raid_type->level) { |
1002 | case 1: | |
1003 | redundancy = rs->md.raid_disks - 1; | |
1004 | break; | |
1005 | case 4: | |
1006 | case 5: | |
1007 | case 6: | |
1008 | redundancy = rs->raid_type->parity_devs; | |
1009 | break; | |
63f33b8d JB |
1010 | case 10: |
1011 | redundancy = raid10_md_layout_to_copies(mddev->layout) - 1; | |
1012 | break; | |
0447568f JB |
1013 | default: |
1014 | ti->error = "Unknown RAID type"; | |
1015 | return -EINVAL; | |
1016 | } | |
1017 | ||
b12d437b | 1018 | freshest = NULL; |
a9ad8526 | 1019 | rdev_for_each_safe(rdev, tmp, mddev) { |
b12d437b JB |
1020 | if (!rdev->meta_bdev) |
1021 | continue; | |
1022 | ||
1023 | ret = super_load(rdev, freshest); | |
1024 | ||
1025 | switch (ret) { | |
1026 | case 1: | |
1027 | freshest = rdev; | |
1028 | break; | |
1029 | case 0: | |
1030 | break; | |
1031 | default: | |
0447568f JB |
1032 | dev = container_of(rdev, struct raid_dev, rdev); |
1033 | if (redundancy--) { | |
1034 | if (dev->meta_dev) | |
1035 | dm_put_device(ti, dev->meta_dev); | |
1036 | ||
1037 | dev->meta_dev = NULL; | |
1038 | rdev->meta_bdev = NULL; | |
1039 | ||
1040 | if (rdev->sb_page) | |
1041 | put_page(rdev->sb_page); | |
1042 | ||
1043 | rdev->sb_page = NULL; | |
1044 | ||
1045 | rdev->sb_loaded = 0; | |
1046 | ||
1047 | /* | |
1048 | * We might be able to salvage the data device | |
1049 | * even though the meta device has failed. For | |
1050 | * now, we behave as though '- -' had been | |
1051 | * set for this device in the table. | |
1052 | */ | |
1053 | if (dev->data_dev) | |
1054 | dm_put_device(ti, dev->data_dev); | |
1055 | ||
1056 | dev->data_dev = NULL; | |
1057 | rdev->bdev = NULL; | |
1058 | ||
1059 | list_del(&rdev->same_set); | |
1060 | ||
1061 | continue; | |
1062 | } | |
b12d437b JB |
1063 | ti->error = "Failed to load superblock"; |
1064 | return ret; | |
1065 | } | |
1066 | } | |
1067 | ||
1068 | if (!freshest) | |
1069 | return 0; | |
1070 | ||
1071 | /* | |
1072 | * Validation of the freshest device provides the source of | |
1073 | * validation for the remaining devices. | |
1074 | */ | |
1075 | ti->error = "Unable to assemble array: Invalid superblocks"; | |
1076 | if (super_validate(mddev, freshest)) | |
1077 | return -EINVAL; | |
1078 | ||
dafb20fa | 1079 | rdev_for_each(rdev, mddev) |
b12d437b JB |
1080 | if ((rdev != freshest) && super_validate(mddev, rdev)) |
1081 | return -EINVAL; | |
1082 | ||
1083 | return 0; | |
1084 | } | |
1085 | ||
9d09e663 N |
1086 | /* |
1087 | * Construct a RAID4/5/6 mapping: | |
1088 | * Args: | |
1089 | * <raid_type> <#raid_params> <raid_params> \ | |
1090 | * <#raid_devs> { <meta_dev1> <dev1> .. <meta_devN> <devN> } | |
1091 | * | |
9d09e663 N |
1092 | * <raid_params> varies by <raid_type>. See 'parse_raid_params' for |
1093 | * details on possible <raid_params>. | |
1094 | */ | |
1095 | static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
1096 | { | |
1097 | int ret; | |
1098 | struct raid_type *rt; | |
1099 | unsigned long num_raid_params, num_raid_devs; | |
1100 | struct raid_set *rs = NULL; | |
1101 | ||
1102 | /* Must have at least <raid_type> <#raid_params> */ | |
1103 | if (argc < 2) { | |
1104 | ti->error = "Too few arguments"; | |
1105 | return -EINVAL; | |
1106 | } | |
1107 | ||
1108 | /* raid type */ | |
1109 | rt = get_raid_type(argv[0]); | |
1110 | if (!rt) { | |
1111 | ti->error = "Unrecognised raid_type"; | |
1112 | return -EINVAL; | |
1113 | } | |
1114 | argc--; | |
1115 | argv++; | |
1116 | ||
1117 | /* number of RAID parameters */ | |
1118 | if (strict_strtoul(argv[0], 10, &num_raid_params) < 0) { | |
1119 | ti->error = "Cannot understand number of RAID parameters"; | |
1120 | return -EINVAL; | |
1121 | } | |
1122 | argc--; | |
1123 | argv++; | |
1124 | ||
1125 | /* Skip over RAID params for now and find out # of devices */ | |
1126 | if (num_raid_params + 1 > argc) { | |
1127 | ti->error = "Arguments do not agree with counts given"; | |
1128 | return -EINVAL; | |
1129 | } | |
1130 | ||
1131 | if ((strict_strtoul(argv[num_raid_params], 10, &num_raid_devs) < 0) || | |
1132 | (num_raid_devs >= INT_MAX)) { | |
1133 | ti->error = "Cannot understand number of raid devices"; | |
1134 | return -EINVAL; | |
1135 | } | |
1136 | ||
1137 | rs = context_alloc(ti, rt, (unsigned)num_raid_devs); | |
1138 | if (IS_ERR(rs)) | |
1139 | return PTR_ERR(rs); | |
1140 | ||
1141 | ret = parse_raid_params(rs, argv, (unsigned)num_raid_params); | |
1142 | if (ret) | |
1143 | goto bad; | |
1144 | ||
1145 | ret = -EINVAL; | |
1146 | ||
1147 | argc -= num_raid_params + 1; /* +1: we already have num_raid_devs */ | |
1148 | argv += num_raid_params + 1; | |
1149 | ||
1150 | if (argc != (num_raid_devs * 2)) { | |
1151 | ti->error = "Supplied RAID devices does not match the count given"; | |
1152 | goto bad; | |
1153 | } | |
1154 | ||
1155 | ret = dev_parms(rs, argv); | |
1156 | if (ret) | |
1157 | goto bad; | |
1158 | ||
b12d437b JB |
1159 | rs->md.sync_super = super_sync; |
1160 | ret = analyse_superblocks(ti, rs); | |
1161 | if (ret) | |
1162 | goto bad; | |
1163 | ||
9d09e663 | 1164 | INIT_WORK(&rs->md.event_work, do_table_event); |
9d09e663 | 1165 | ti->private = rs; |
0ca93de9 | 1166 | ti->num_flush_requests = 1; |
9d09e663 N |
1167 | |
1168 | mutex_lock(&rs->md.reconfig_mutex); | |
1169 | ret = md_run(&rs->md); | |
1170 | rs->md.in_sync = 0; /* Assume already marked dirty */ | |
1171 | mutex_unlock(&rs->md.reconfig_mutex); | |
1172 | ||
1173 | if (ret) { | |
1174 | ti->error = "Fail to run raid array"; | |
1175 | goto bad; | |
1176 | } | |
1177 | ||
63f33b8d JB |
1178 | if (ti->len != rs->md.array_sectors) { |
1179 | ti->error = "Array size does not match requested target length"; | |
1180 | ret = -EINVAL; | |
1181 | goto size_mismatch; | |
1182 | } | |
9d09e663 | 1183 | rs->callbacks.congested_fn = raid_is_congested; |
9d09e663 N |
1184 | dm_table_add_target_callbacks(ti->table, &rs->callbacks); |
1185 | ||
32737279 | 1186 | mddev_suspend(&rs->md); |
9d09e663 N |
1187 | return 0; |
1188 | ||
63f33b8d JB |
1189 | size_mismatch: |
1190 | md_stop(&rs->md); | |
9d09e663 N |
1191 | bad: |
1192 | context_free(rs); | |
1193 | ||
1194 | return ret; | |
1195 | } | |
1196 | ||
1197 | static void raid_dtr(struct dm_target *ti) | |
1198 | { | |
1199 | struct raid_set *rs = ti->private; | |
1200 | ||
1201 | list_del_init(&rs->callbacks.list); | |
1202 | md_stop(&rs->md); | |
1203 | context_free(rs); | |
1204 | } | |
1205 | ||
1206 | static int raid_map(struct dm_target *ti, struct bio *bio, union map_info *map_context) | |
1207 | { | |
1208 | struct raid_set *rs = ti->private; | |
fd01b88c | 1209 | struct mddev *mddev = &rs->md; |
9d09e663 N |
1210 | |
1211 | mddev->pers->make_request(mddev, bio); | |
1212 | ||
1213 | return DM_MAPIO_SUBMITTED; | |
1214 | } | |
1215 | ||
1216 | static int raid_status(struct dm_target *ti, status_type_t type, | |
1f4e0ff0 | 1217 | unsigned status_flags, char *result, unsigned maxlen) |
9d09e663 N |
1218 | { |
1219 | struct raid_set *rs = ti->private; | |
1220 | unsigned raid_param_cnt = 1; /* at least 1 for chunksize */ | |
1221 | unsigned sz = 0; | |
2e727c3c | 1222 | int i, array_in_sync = 0; |
9d09e663 N |
1223 | sector_t sync; |
1224 | ||
1225 | switch (type) { | |
1226 | case STATUSTYPE_INFO: | |
1227 | DMEMIT("%s %d ", rs->raid_type->name, rs->md.raid_disks); | |
1228 | ||
9d09e663 N |
1229 | if (test_bit(MD_RECOVERY_RUNNING, &rs->md.recovery)) |
1230 | sync = rs->md.curr_resync_completed; | |
1231 | else | |
1232 | sync = rs->md.recovery_cp; | |
1233 | ||
2e727c3c JB |
1234 | if (sync >= rs->md.resync_max_sectors) { |
1235 | array_in_sync = 1; | |
9d09e663 | 1236 | sync = rs->md.resync_max_sectors; |
2e727c3c JB |
1237 | } else { |
1238 | /* | |
1239 | * The array may be doing an initial sync, or it may | |
1240 | * be rebuilding individual components. If all the | |
1241 | * devices are In_sync, then it is the array that is | |
1242 | * being initialized. | |
1243 | */ | |
1244 | for (i = 0; i < rs->md.raid_disks; i++) | |
1245 | if (!test_bit(In_sync, &rs->dev[i].rdev.flags)) | |
1246 | array_in_sync = 1; | |
1247 | } | |
1248 | /* | |
1249 | * Status characters: | |
1250 | * 'D' = Dead/Failed device | |
1251 | * 'a' = Alive but not in-sync | |
1252 | * 'A' = Alive and in-sync | |
1253 | */ | |
1254 | for (i = 0; i < rs->md.raid_disks; i++) { | |
1255 | if (test_bit(Faulty, &rs->dev[i].rdev.flags)) | |
1256 | DMEMIT("D"); | |
1257 | else if (!array_in_sync || | |
1258 | !test_bit(In_sync, &rs->dev[i].rdev.flags)) | |
1259 | DMEMIT("a"); | |
1260 | else | |
1261 | DMEMIT("A"); | |
1262 | } | |
9d09e663 | 1263 | |
2e727c3c JB |
1264 | /* |
1265 | * In-sync ratio: | |
1266 | * The in-sync ratio shows the progress of: | |
1267 | * - Initializing the array | |
1268 | * - Rebuilding a subset of devices of the array | |
1269 | * The user can distinguish between the two by referring | |
1270 | * to the status characters. | |
1271 | */ | |
9d09e663 N |
1272 | DMEMIT(" %llu/%llu", |
1273 | (unsigned long long) sync, | |
1274 | (unsigned long long) rs->md.resync_max_sectors); | |
1275 | ||
1276 | break; | |
1277 | case STATUSTYPE_TABLE: | |
1278 | /* The string you would use to construct this array */ | |
46bed2b5 | 1279 | for (i = 0; i < rs->md.raid_disks; i++) { |
13c87583 JB |
1280 | if ((rs->print_flags & DMPF_REBUILD) && |
1281 | rs->dev[i].data_dev && | |
9d09e663 | 1282 | !test_bit(In_sync, &rs->dev[i].rdev.flags)) |
13c87583 | 1283 | raid_param_cnt += 2; /* for rebuilds */ |
46bed2b5 JB |
1284 | if (rs->dev[i].data_dev && |
1285 | test_bit(WriteMostly, &rs->dev[i].rdev.flags)) | |
1286 | raid_param_cnt += 2; | |
1287 | } | |
9d09e663 | 1288 | |
34f8ac6d | 1289 | raid_param_cnt += (hweight32(rs->print_flags & ~DMPF_REBUILD) * 2); |
9d09e663 N |
1290 | if (rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC)) |
1291 | raid_param_cnt--; | |
1292 | ||
1293 | DMEMIT("%s %u %u", rs->raid_type->name, | |
1294 | raid_param_cnt, rs->md.chunk_sectors); | |
1295 | ||
1296 | if ((rs->print_flags & DMPF_SYNC) && | |
1297 | (rs->md.recovery_cp == MaxSector)) | |
1298 | DMEMIT(" sync"); | |
1299 | if (rs->print_flags & DMPF_NOSYNC) | |
1300 | DMEMIT(" nosync"); | |
1301 | ||
1302 | for (i = 0; i < rs->md.raid_disks; i++) | |
13c87583 JB |
1303 | if ((rs->print_flags & DMPF_REBUILD) && |
1304 | rs->dev[i].data_dev && | |
9d09e663 N |
1305 | !test_bit(In_sync, &rs->dev[i].rdev.flags)) |
1306 | DMEMIT(" rebuild %u", i); | |
1307 | ||
1308 | if (rs->print_flags & DMPF_DAEMON_SLEEP) | |
1309 | DMEMIT(" daemon_sleep %lu", | |
1310 | rs->md.bitmap_info.daemon_sleep); | |
1311 | ||
1312 | if (rs->print_flags & DMPF_MIN_RECOVERY_RATE) | |
1313 | DMEMIT(" min_recovery_rate %d", rs->md.sync_speed_min); | |
1314 | ||
1315 | if (rs->print_flags & DMPF_MAX_RECOVERY_RATE) | |
1316 | DMEMIT(" max_recovery_rate %d", rs->md.sync_speed_max); | |
1317 | ||
46bed2b5 JB |
1318 | for (i = 0; i < rs->md.raid_disks; i++) |
1319 | if (rs->dev[i].data_dev && | |
1320 | test_bit(WriteMostly, &rs->dev[i].rdev.flags)) | |
1321 | DMEMIT(" write_mostly %u", i); | |
1322 | ||
9d09e663 N |
1323 | if (rs->print_flags & DMPF_MAX_WRITE_BEHIND) |
1324 | DMEMIT(" max_write_behind %lu", | |
1325 | rs->md.bitmap_info.max_write_behind); | |
1326 | ||
1327 | if (rs->print_flags & DMPF_STRIPE_CACHE) { | |
d1688a6d | 1328 | struct r5conf *conf = rs->md.private; |
9d09e663 N |
1329 | |
1330 | /* convert from kiB to sectors */ | |
1331 | DMEMIT(" stripe_cache %d", | |
1332 | conf ? conf->max_nr_stripes * 2 : 0); | |
1333 | } | |
1334 | ||
c1084561 JB |
1335 | if (rs->print_flags & DMPF_REGION_SIZE) |
1336 | DMEMIT(" region_size %lu", | |
1337 | rs->md.bitmap_info.chunksize >> 9); | |
1338 | ||
63f33b8d JB |
1339 | if (rs->print_flags & DMPF_RAID10_COPIES) |
1340 | DMEMIT(" raid10_copies %u", | |
1341 | raid10_md_layout_to_copies(rs->md.layout)); | |
1342 | ||
1343 | if (rs->print_flags & DMPF_RAID10_FORMAT) | |
1344 | DMEMIT(" raid10_format near"); | |
1345 | ||
9d09e663 N |
1346 | DMEMIT(" %d", rs->md.raid_disks); |
1347 | for (i = 0; i < rs->md.raid_disks; i++) { | |
b12d437b JB |
1348 | if (rs->dev[i].meta_dev) |
1349 | DMEMIT(" %s", rs->dev[i].meta_dev->name); | |
1350 | else | |
1351 | DMEMIT(" -"); | |
9d09e663 N |
1352 | |
1353 | if (rs->dev[i].data_dev) | |
1354 | DMEMIT(" %s", rs->dev[i].data_dev->name); | |
1355 | else | |
1356 | DMEMIT(" -"); | |
1357 | } | |
1358 | } | |
1359 | ||
1360 | return 0; | |
1361 | } | |
1362 | ||
1363 | static int raid_iterate_devices(struct dm_target *ti, iterate_devices_callout_fn fn, void *data) | |
1364 | { | |
1365 | struct raid_set *rs = ti->private; | |
1366 | unsigned i; | |
1367 | int ret = 0; | |
1368 | ||
1369 | for (i = 0; !ret && i < rs->md.raid_disks; i++) | |
1370 | if (rs->dev[i].data_dev) | |
1371 | ret = fn(ti, | |
1372 | rs->dev[i].data_dev, | |
1373 | 0, /* No offset on data devs */ | |
1374 | rs->md.dev_sectors, | |
1375 | data); | |
1376 | ||
1377 | return ret; | |
1378 | } | |
1379 | ||
1380 | static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits) | |
1381 | { | |
1382 | struct raid_set *rs = ti->private; | |
1383 | unsigned chunk_size = rs->md.chunk_sectors << 9; | |
d1688a6d | 1384 | struct r5conf *conf = rs->md.private; |
9d09e663 N |
1385 | |
1386 | blk_limits_io_min(limits, chunk_size); | |
1387 | blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded)); | |
1388 | } | |
1389 | ||
1390 | static void raid_presuspend(struct dm_target *ti) | |
1391 | { | |
1392 | struct raid_set *rs = ti->private; | |
1393 | ||
1394 | md_stop_writes(&rs->md); | |
1395 | } | |
1396 | ||
1397 | static void raid_postsuspend(struct dm_target *ti) | |
1398 | { | |
1399 | struct raid_set *rs = ti->private; | |
1400 | ||
1401 | mddev_suspend(&rs->md); | |
1402 | } | |
1403 | ||
1404 | static void raid_resume(struct dm_target *ti) | |
1405 | { | |
1406 | struct raid_set *rs = ti->private; | |
1407 | ||
81f382f9 | 1408 | set_bit(MD_CHANGE_DEVS, &rs->md.flags); |
34f8ac6d JB |
1409 | if (!rs->bitmap_loaded) { |
1410 | bitmap_load(&rs->md); | |
1411 | rs->bitmap_loaded = 1; | |
47525e59 | 1412 | } |
34f8ac6d | 1413 | |
47525e59 | 1414 | clear_bit(MD_RECOVERY_FROZEN, &rs->md.recovery); |
9d09e663 N |
1415 | mddev_resume(&rs->md); |
1416 | } | |
1417 | ||
1418 | static struct target_type raid_target = { | |
1419 | .name = "raid", | |
4ec1e369 | 1420 | .version = {1, 3, 1}, |
9d09e663 N |
1421 | .module = THIS_MODULE, |
1422 | .ctr = raid_ctr, | |
1423 | .dtr = raid_dtr, | |
1424 | .map = raid_map, | |
1425 | .status = raid_status, | |
1426 | .iterate_devices = raid_iterate_devices, | |
1427 | .io_hints = raid_io_hints, | |
1428 | .presuspend = raid_presuspend, | |
1429 | .postsuspend = raid_postsuspend, | |
1430 | .resume = raid_resume, | |
1431 | }; | |
1432 | ||
1433 | static int __init dm_raid_init(void) | |
1434 | { | |
1435 | return dm_register_target(&raid_target); | |
1436 | } | |
1437 | ||
1438 | static void __exit dm_raid_exit(void) | |
1439 | { | |
1440 | dm_unregister_target(&raid_target); | |
1441 | } | |
1442 | ||
1443 | module_init(dm_raid_init); | |
1444 | module_exit(dm_raid_exit); | |
1445 | ||
1446 | MODULE_DESCRIPTION(DM_NAME " raid4/5/6 target"); | |
63f33b8d JB |
1447 | MODULE_ALIAS("dm-raid1"); |
1448 | MODULE_ALIAS("dm-raid10"); | |
9d09e663 N |
1449 | MODULE_ALIAS("dm-raid4"); |
1450 | MODULE_ALIAS("dm-raid5"); | |
1451 | MODULE_ALIAS("dm-raid6"); | |
1452 | MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>"); | |
1453 | MODULE_LICENSE("GPL"); |