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" |
9d09e663 N |
14 | #include "bitmap.h" |
15 | ||
3e8dbb7f AK |
16 | #include <linux/device-mapper.h> |
17 | ||
9d09e663 N |
18 | #define DM_MSG_PREFIX "raid" |
19 | ||
20 | /* | |
b12d437b JB |
21 | * The following flags are used by dm-raid.c to set up the array state. |
22 | * They must be cleared before md_run is called. | |
9d09e663 | 23 | */ |
b12d437b | 24 | #define FirstUse 10 /* rdev flag */ |
9d09e663 N |
25 | |
26 | struct raid_dev { | |
27 | /* | |
28 | * Two DM devices, one to hold metadata and one to hold the | |
29 | * actual data/parity. The reason for this is to not confuse | |
30 | * ti->len and give more flexibility in altering size and | |
31 | * characteristics. | |
32 | * | |
33 | * While it is possible for this device to be associated | |
34 | * with a different physical device than the data_dev, it | |
35 | * is intended for it to be the same. | |
36 | * |--------- Physical Device ---------| | |
37 | * |- meta_dev -|------ data_dev ------| | |
38 | */ | |
39 | struct dm_dev *meta_dev; | |
40 | struct dm_dev *data_dev; | |
3cb03002 | 41 | struct md_rdev rdev; |
9d09e663 N |
42 | }; |
43 | ||
44 | /* | |
45 | * Flags for rs->print_flags field. | |
46 | */ | |
13c87583 JB |
47 | #define DMPF_SYNC 0x1 |
48 | #define DMPF_NOSYNC 0x2 | |
49 | #define DMPF_REBUILD 0x4 | |
50 | #define DMPF_DAEMON_SLEEP 0x8 | |
51 | #define DMPF_MIN_RECOVERY_RATE 0x10 | |
52 | #define DMPF_MAX_RECOVERY_RATE 0x20 | |
53 | #define DMPF_MAX_WRITE_BEHIND 0x40 | |
54 | #define DMPF_STRIPE_CACHE 0x80 | |
c1084561 | 55 | #define DMPF_REGION_SIZE 0X100 |
9d09e663 N |
56 | struct raid_set { |
57 | struct dm_target *ti; | |
58 | ||
34f8ac6d JB |
59 | uint32_t bitmap_loaded; |
60 | uint32_t print_flags; | |
9d09e663 | 61 | |
fd01b88c | 62 | struct mddev md; |
9d09e663 N |
63 | struct raid_type *raid_type; |
64 | struct dm_target_callbacks callbacks; | |
65 | ||
66 | struct raid_dev dev[0]; | |
67 | }; | |
68 | ||
69 | /* Supported raid types and properties. */ | |
70 | static struct raid_type { | |
71 | const char *name; /* RAID algorithm. */ | |
72 | const char *descr; /* Descriptor text for logging. */ | |
73 | const unsigned parity_devs; /* # of parity devices. */ | |
74 | const unsigned minimal_devs; /* minimal # of devices in set. */ | |
75 | const unsigned level; /* RAID level. */ | |
76 | const unsigned algorithm; /* RAID algorithm. */ | |
77 | } raid_types[] = { | |
32737279 | 78 | {"raid1", "RAID1 (mirroring)", 0, 2, 1, 0 /* NONE */}, |
9d09e663 N |
79 | {"raid4", "RAID4 (dedicated parity disk)", 1, 2, 5, ALGORITHM_PARITY_0}, |
80 | {"raid5_la", "RAID5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC}, | |
81 | {"raid5_ra", "RAID5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC}, | |
82 | {"raid5_ls", "RAID5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC}, | |
83 | {"raid5_rs", "RAID5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC}, | |
84 | {"raid6_zr", "RAID6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART}, | |
85 | {"raid6_nr", "RAID6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART}, | |
86 | {"raid6_nc", "RAID6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE} | |
87 | }; | |
88 | ||
89 | static struct raid_type *get_raid_type(char *name) | |
90 | { | |
91 | int i; | |
92 | ||
93 | for (i = 0; i < ARRAY_SIZE(raid_types); i++) | |
94 | if (!strcmp(raid_types[i].name, name)) | |
95 | return &raid_types[i]; | |
96 | ||
97 | return NULL; | |
98 | } | |
99 | ||
100 | static struct raid_set *context_alloc(struct dm_target *ti, struct raid_type *raid_type, unsigned raid_devs) | |
101 | { | |
102 | unsigned i; | |
103 | struct raid_set *rs; | |
104 | sector_t sectors_per_dev; | |
105 | ||
106 | if (raid_devs <= raid_type->parity_devs) { | |
107 | ti->error = "Insufficient number of devices"; | |
108 | return ERR_PTR(-EINVAL); | |
109 | } | |
110 | ||
111 | sectors_per_dev = ti->len; | |
32737279 JB |
112 | if ((raid_type->level > 1) && |
113 | sector_div(sectors_per_dev, (raid_devs - raid_type->parity_devs))) { | |
9d09e663 N |
114 | ti->error = "Target length not divisible by number of data devices"; |
115 | return ERR_PTR(-EINVAL); | |
116 | } | |
117 | ||
118 | rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL); | |
119 | if (!rs) { | |
120 | ti->error = "Cannot allocate raid context"; | |
121 | return ERR_PTR(-ENOMEM); | |
122 | } | |
123 | ||
124 | mddev_init(&rs->md); | |
125 | ||
126 | rs->ti = ti; | |
127 | rs->raid_type = raid_type; | |
128 | rs->md.raid_disks = raid_devs; | |
129 | rs->md.level = raid_type->level; | |
130 | rs->md.new_level = rs->md.level; | |
131 | rs->md.dev_sectors = sectors_per_dev; | |
132 | rs->md.layout = raid_type->algorithm; | |
133 | rs->md.new_layout = rs->md.layout; | |
134 | rs->md.delta_disks = 0; | |
135 | rs->md.recovery_cp = 0; | |
136 | ||
137 | for (i = 0; i < raid_devs; i++) | |
138 | md_rdev_init(&rs->dev[i].rdev); | |
139 | ||
140 | /* | |
141 | * Remaining items to be initialized by further RAID params: | |
142 | * rs->md.persistent | |
143 | * rs->md.external | |
144 | * rs->md.chunk_sectors | |
145 | * rs->md.new_chunk_sectors | |
146 | */ | |
147 | ||
148 | return rs; | |
149 | } | |
150 | ||
151 | static void context_free(struct raid_set *rs) | |
152 | { | |
153 | int i; | |
154 | ||
b12d437b JB |
155 | for (i = 0; i < rs->md.raid_disks; i++) { |
156 | if (rs->dev[i].meta_dev) | |
157 | dm_put_device(rs->ti, rs->dev[i].meta_dev); | |
545c8795 | 158 | md_rdev_clear(&rs->dev[i].rdev); |
9d09e663 N |
159 | if (rs->dev[i].data_dev) |
160 | dm_put_device(rs->ti, rs->dev[i].data_dev); | |
b12d437b | 161 | } |
9d09e663 N |
162 | |
163 | kfree(rs); | |
164 | } | |
165 | ||
166 | /* | |
167 | * For every device we have two words | |
168 | * <meta_dev>: meta device name or '-' if missing | |
169 | * <data_dev>: data device name or '-' if missing | |
170 | * | |
b12d437b JB |
171 | * The following are permitted: |
172 | * - - | |
173 | * - <data_dev> | |
174 | * <meta_dev> <data_dev> | |
175 | * | |
176 | * The following is not allowed: | |
177 | * <meta_dev> - | |
178 | * | |
179 | * This code parses those words. If there is a failure, | |
180 | * the caller must use context_free to unwind the operations. | |
9d09e663 N |
181 | */ |
182 | static int dev_parms(struct raid_set *rs, char **argv) | |
183 | { | |
184 | int i; | |
185 | int rebuild = 0; | |
186 | int metadata_available = 0; | |
187 | int ret = 0; | |
188 | ||
189 | for (i = 0; i < rs->md.raid_disks; i++, argv += 2) { | |
190 | rs->dev[i].rdev.raid_disk = i; | |
191 | ||
192 | rs->dev[i].meta_dev = NULL; | |
193 | rs->dev[i].data_dev = NULL; | |
194 | ||
195 | /* | |
196 | * There are no offsets, since there is a separate device | |
197 | * for data and metadata. | |
198 | */ | |
199 | rs->dev[i].rdev.data_offset = 0; | |
200 | rs->dev[i].rdev.mddev = &rs->md; | |
201 | ||
202 | if (strcmp(argv[0], "-")) { | |
b12d437b JB |
203 | ret = dm_get_device(rs->ti, argv[0], |
204 | dm_table_get_mode(rs->ti->table), | |
205 | &rs->dev[i].meta_dev); | |
206 | rs->ti->error = "RAID metadata device lookup failure"; | |
207 | if (ret) | |
208 | return ret; | |
209 | ||
210 | rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL); | |
211 | if (!rs->dev[i].rdev.sb_page) | |
212 | return -ENOMEM; | |
9d09e663 N |
213 | } |
214 | ||
215 | if (!strcmp(argv[1], "-")) { | |
216 | if (!test_bit(In_sync, &rs->dev[i].rdev.flags) && | |
217 | (!rs->dev[i].rdev.recovery_offset)) { | |
218 | rs->ti->error = "Drive designated for rebuild not specified"; | |
219 | return -EINVAL; | |
220 | } | |
221 | ||
b12d437b JB |
222 | rs->ti->error = "No data device supplied with metadata device"; |
223 | if (rs->dev[i].meta_dev) | |
224 | return -EINVAL; | |
225 | ||
9d09e663 N |
226 | continue; |
227 | } | |
228 | ||
229 | ret = dm_get_device(rs->ti, argv[1], | |
230 | dm_table_get_mode(rs->ti->table), | |
231 | &rs->dev[i].data_dev); | |
232 | if (ret) { | |
233 | rs->ti->error = "RAID device lookup failure"; | |
234 | return ret; | |
235 | } | |
236 | ||
b12d437b JB |
237 | if (rs->dev[i].meta_dev) { |
238 | metadata_available = 1; | |
239 | rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev; | |
240 | } | |
9d09e663 N |
241 | rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev; |
242 | list_add(&rs->dev[i].rdev.same_set, &rs->md.disks); | |
243 | if (!test_bit(In_sync, &rs->dev[i].rdev.flags)) | |
244 | rebuild++; | |
245 | } | |
246 | ||
247 | if (metadata_available) { | |
248 | rs->md.external = 0; | |
249 | rs->md.persistent = 1; | |
250 | rs->md.major_version = 2; | |
251 | } else if (rebuild && !rs->md.recovery_cp) { | |
252 | /* | |
253 | * Without metadata, we will not be able to tell if the array | |
254 | * is in-sync or not - we must assume it is not. Therefore, | |
255 | * it is impossible to rebuild a drive. | |
256 | * | |
257 | * Even if there is metadata, the on-disk information may | |
258 | * indicate that the array is not in-sync and it will then | |
259 | * fail at that time. | |
260 | * | |
261 | * User could specify 'nosync' option if desperate. | |
262 | */ | |
263 | DMERR("Unable to rebuild drive while array is not in-sync"); | |
264 | rs->ti->error = "RAID device lookup failure"; | |
265 | return -EINVAL; | |
266 | } | |
267 | ||
268 | return 0; | |
269 | } | |
270 | ||
c1084561 JB |
271 | /* |
272 | * validate_region_size | |
273 | * @rs | |
274 | * @region_size: region size in sectors. If 0, pick a size (4MiB default). | |
275 | * | |
276 | * Set rs->md.bitmap_info.chunksize (which really refers to 'region size'). | |
277 | * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap. | |
278 | * | |
279 | * Returns: 0 on success, -EINVAL on failure. | |
280 | */ | |
281 | static int validate_region_size(struct raid_set *rs, unsigned long region_size) | |
282 | { | |
283 | unsigned long min_region_size = rs->ti->len / (1 << 21); | |
284 | ||
285 | if (!region_size) { | |
286 | /* | |
287 | * Choose a reasonable default. All figures in sectors. | |
288 | */ | |
289 | if (min_region_size > (1 << 13)) { | |
290 | DMINFO("Choosing default region size of %lu sectors", | |
291 | region_size); | |
292 | region_size = min_region_size; | |
293 | } else { | |
294 | DMINFO("Choosing default region size of 4MiB"); | |
295 | region_size = 1 << 13; /* sectors */ | |
296 | } | |
297 | } else { | |
298 | /* | |
299 | * Validate user-supplied value. | |
300 | */ | |
301 | if (region_size > rs->ti->len) { | |
302 | rs->ti->error = "Supplied region size is too large"; | |
303 | return -EINVAL; | |
304 | } | |
305 | ||
306 | if (region_size < min_region_size) { | |
307 | DMERR("Supplied region_size (%lu sectors) below minimum (%lu)", | |
308 | region_size, min_region_size); | |
309 | rs->ti->error = "Supplied region size is too small"; | |
310 | return -EINVAL; | |
311 | } | |
312 | ||
313 | if (!is_power_of_2(region_size)) { | |
314 | rs->ti->error = "Region size is not a power of 2"; | |
315 | return -EINVAL; | |
316 | } | |
317 | ||
318 | if (region_size < rs->md.chunk_sectors) { | |
319 | rs->ti->error = "Region size is smaller than the chunk size"; | |
320 | return -EINVAL; | |
321 | } | |
322 | } | |
323 | ||
324 | /* | |
325 | * Convert sectors to bytes. | |
326 | */ | |
327 | rs->md.bitmap_info.chunksize = (region_size << 9); | |
328 | ||
329 | return 0; | |
330 | } | |
331 | ||
9d09e663 N |
332 | /* |
333 | * Possible arguments are... | |
9d09e663 N |
334 | * <chunk_size> [optional_args] |
335 | * | |
32737279 JB |
336 | * Argument definitions |
337 | * <chunk_size> The number of sectors per disk that | |
338 | * will form the "stripe" | |
339 | * [[no]sync] Force or prevent recovery of the | |
340 | * entire array | |
9d09e663 | 341 | * [rebuild <idx>] Rebuild the drive indicated by the index |
32737279 JB |
342 | * [daemon_sleep <ms>] Time between bitmap daemon work to |
343 | * clear bits | |
9d09e663 N |
344 | * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization |
345 | * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization | |
46bed2b5 | 346 | * [write_mostly <idx>] Indicate a write mostly drive via index |
9d09e663 N |
347 | * [max_write_behind <sectors>] See '-write-behind=' (man mdadm) |
348 | * [stripe_cache <sectors>] Stripe cache size for higher RAIDs | |
c1084561 | 349 | * [region_size <sectors>] Defines granularity of bitmap |
9d09e663 N |
350 | */ |
351 | static int parse_raid_params(struct raid_set *rs, char **argv, | |
352 | unsigned num_raid_params) | |
353 | { | |
354 | unsigned i, rebuild_cnt = 0; | |
c1084561 | 355 | unsigned long value, region_size = 0; |
9d09e663 N |
356 | char *key; |
357 | ||
358 | /* | |
359 | * First, parse the in-order required arguments | |
32737279 | 360 | * "chunk_size" is the only argument of this type. |
9d09e663 | 361 | */ |
32737279 | 362 | if ((strict_strtoul(argv[0], 10, &value) < 0)) { |
9d09e663 N |
363 | rs->ti->error = "Bad chunk size"; |
364 | return -EINVAL; | |
32737279 JB |
365 | } else if (rs->raid_type->level == 1) { |
366 | if (value) | |
367 | DMERR("Ignoring chunk size parameter for RAID 1"); | |
368 | value = 0; | |
369 | } else if (!is_power_of_2(value)) { | |
370 | rs->ti->error = "Chunk size must be a power of 2"; | |
371 | return -EINVAL; | |
372 | } else if (value < 8) { | |
373 | rs->ti->error = "Chunk size value is too small"; | |
374 | return -EINVAL; | |
9d09e663 N |
375 | } |
376 | ||
377 | rs->md.new_chunk_sectors = rs->md.chunk_sectors = value; | |
378 | argv++; | |
379 | num_raid_params--; | |
380 | ||
381 | /* | |
b12d437b JB |
382 | * We set each individual device as In_sync with a completed |
383 | * 'recovery_offset'. If there has been a device failure or | |
384 | * replacement then one of the following cases applies: | |
385 | * | |
386 | * 1) User specifies 'rebuild'. | |
387 | * - Device is reset when param is read. | |
388 | * 2) A new device is supplied. | |
389 | * - No matching superblock found, resets device. | |
390 | * 3) Device failure was transient and returns on reload. | |
391 | * - Failure noticed, resets device for bitmap replay. | |
392 | * 4) Device hadn't completed recovery after previous failure. | |
393 | * - Superblock is read and overrides recovery_offset. | |
394 | * | |
395 | * What is found in the superblocks of the devices is always | |
396 | * authoritative, unless 'rebuild' or '[no]sync' was specified. | |
9d09e663 | 397 | */ |
b12d437b | 398 | for (i = 0; i < rs->md.raid_disks; i++) { |
9d09e663 | 399 | set_bit(In_sync, &rs->dev[i].rdev.flags); |
b12d437b JB |
400 | rs->dev[i].rdev.recovery_offset = MaxSector; |
401 | } | |
9d09e663 | 402 | |
b12d437b JB |
403 | /* |
404 | * Second, parse the unordered optional arguments | |
405 | */ | |
9d09e663 | 406 | for (i = 0; i < num_raid_params; i++) { |
13c87583 | 407 | if (!strcasecmp(argv[i], "nosync")) { |
9d09e663 N |
408 | rs->md.recovery_cp = MaxSector; |
409 | rs->print_flags |= DMPF_NOSYNC; | |
9d09e663 N |
410 | continue; |
411 | } | |
13c87583 | 412 | if (!strcasecmp(argv[i], "sync")) { |
9d09e663 N |
413 | rs->md.recovery_cp = 0; |
414 | rs->print_flags |= DMPF_SYNC; | |
9d09e663 N |
415 | continue; |
416 | } | |
417 | ||
418 | /* The rest of the optional arguments come in key/value pairs */ | |
419 | if ((i + 1) >= num_raid_params) { | |
420 | rs->ti->error = "Wrong number of raid parameters given"; | |
421 | return -EINVAL; | |
422 | } | |
423 | ||
424 | key = argv[i++]; | |
425 | if (strict_strtoul(argv[i], 10, &value) < 0) { | |
426 | rs->ti->error = "Bad numerical argument given in raid params"; | |
427 | return -EINVAL; | |
428 | } | |
429 | ||
13c87583 | 430 | if (!strcasecmp(key, "rebuild")) { |
32737279 JB |
431 | rebuild_cnt++; |
432 | if (((rs->raid_type->level != 1) && | |
433 | (rebuild_cnt > rs->raid_type->parity_devs)) || | |
434 | ((rs->raid_type->level == 1) && | |
435 | (rebuild_cnt > (rs->md.raid_disks - 1)))) { | |
436 | rs->ti->error = "Too many rebuild devices specified for given RAID type"; | |
9d09e663 N |
437 | return -EINVAL; |
438 | } | |
439 | if (value > rs->md.raid_disks) { | |
440 | rs->ti->error = "Invalid rebuild index given"; | |
441 | return -EINVAL; | |
442 | } | |
443 | clear_bit(In_sync, &rs->dev[value].rdev.flags); | |
444 | rs->dev[value].rdev.recovery_offset = 0; | |
13c87583 | 445 | rs->print_flags |= DMPF_REBUILD; |
46bed2b5 JB |
446 | } else if (!strcasecmp(key, "write_mostly")) { |
447 | if (rs->raid_type->level != 1) { | |
448 | rs->ti->error = "write_mostly option is only valid for RAID1"; | |
449 | return -EINVAL; | |
450 | } | |
82324809 | 451 | if (value >= rs->md.raid_disks) { |
46bed2b5 JB |
452 | rs->ti->error = "Invalid write_mostly drive index given"; |
453 | return -EINVAL; | |
454 | } | |
455 | set_bit(WriteMostly, &rs->dev[value].rdev.flags); | |
13c87583 | 456 | } else if (!strcasecmp(key, "max_write_behind")) { |
46bed2b5 JB |
457 | if (rs->raid_type->level != 1) { |
458 | rs->ti->error = "max_write_behind option is only valid for RAID1"; | |
459 | return -EINVAL; | |
460 | } | |
9d09e663 N |
461 | rs->print_flags |= DMPF_MAX_WRITE_BEHIND; |
462 | ||
463 | /* | |
464 | * In device-mapper, we specify things in sectors, but | |
465 | * MD records this value in kB | |
466 | */ | |
467 | value /= 2; | |
468 | if (value > COUNTER_MAX) { | |
469 | rs->ti->error = "Max write-behind limit out of range"; | |
470 | return -EINVAL; | |
471 | } | |
472 | rs->md.bitmap_info.max_write_behind = value; | |
13c87583 | 473 | } else if (!strcasecmp(key, "daemon_sleep")) { |
9d09e663 N |
474 | rs->print_flags |= DMPF_DAEMON_SLEEP; |
475 | if (!value || (value > MAX_SCHEDULE_TIMEOUT)) { | |
476 | rs->ti->error = "daemon sleep period out of range"; | |
477 | return -EINVAL; | |
478 | } | |
479 | rs->md.bitmap_info.daemon_sleep = value; | |
13c87583 | 480 | } else if (!strcasecmp(key, "stripe_cache")) { |
9d09e663 N |
481 | rs->print_flags |= DMPF_STRIPE_CACHE; |
482 | ||
483 | /* | |
484 | * In device-mapper, we specify things in sectors, but | |
485 | * MD records this value in kB | |
486 | */ | |
487 | value /= 2; | |
488 | ||
489 | if (rs->raid_type->level < 5) { | |
490 | rs->ti->error = "Inappropriate argument: stripe_cache"; | |
491 | return -EINVAL; | |
492 | } | |
493 | if (raid5_set_cache_size(&rs->md, (int)value)) { | |
494 | rs->ti->error = "Bad stripe_cache size"; | |
495 | return -EINVAL; | |
496 | } | |
13c87583 | 497 | } else if (!strcasecmp(key, "min_recovery_rate")) { |
9d09e663 N |
498 | rs->print_flags |= DMPF_MIN_RECOVERY_RATE; |
499 | if (value > INT_MAX) { | |
500 | rs->ti->error = "min_recovery_rate out of range"; | |
501 | return -EINVAL; | |
502 | } | |
503 | rs->md.sync_speed_min = (int)value; | |
13c87583 | 504 | } else if (!strcasecmp(key, "max_recovery_rate")) { |
9d09e663 N |
505 | rs->print_flags |= DMPF_MAX_RECOVERY_RATE; |
506 | if (value > INT_MAX) { | |
507 | rs->ti->error = "max_recovery_rate out of range"; | |
508 | return -EINVAL; | |
509 | } | |
510 | rs->md.sync_speed_max = (int)value; | |
c1084561 JB |
511 | } else if (!strcasecmp(key, "region_size")) { |
512 | rs->print_flags |= DMPF_REGION_SIZE; | |
513 | region_size = value; | |
9d09e663 N |
514 | } else { |
515 | DMERR("Unable to parse RAID parameter: %s", key); | |
516 | rs->ti->error = "Unable to parse RAID parameters"; | |
517 | return -EINVAL; | |
518 | } | |
519 | } | |
520 | ||
c1084561 JB |
521 | if (validate_region_size(rs, region_size)) |
522 | return -EINVAL; | |
523 | ||
524 | if (rs->md.chunk_sectors) | |
525 | rs->ti->split_io = rs->md.chunk_sectors; | |
526 | else | |
527 | rs->ti->split_io = region_size; | |
528 | ||
32737279 JB |
529 | if (rs->md.chunk_sectors) |
530 | rs->ti->split_io = rs->md.chunk_sectors; | |
531 | else | |
532 | rs->ti->split_io = region_size; | |
533 | ||
9d09e663 N |
534 | /* Assume there are no metadata devices until the drives are parsed */ |
535 | rs->md.persistent = 0; | |
536 | rs->md.external = 1; | |
537 | ||
538 | return 0; | |
539 | } | |
540 | ||
541 | static void do_table_event(struct work_struct *ws) | |
542 | { | |
543 | struct raid_set *rs = container_of(ws, struct raid_set, md.event_work); | |
544 | ||
545 | dm_table_event(rs->ti->table); | |
546 | } | |
547 | ||
548 | static int raid_is_congested(struct dm_target_callbacks *cb, int bits) | |
549 | { | |
550 | struct raid_set *rs = container_of(cb, struct raid_set, callbacks); | |
551 | ||
32737279 JB |
552 | if (rs->raid_type->level == 1) |
553 | return md_raid1_congested(&rs->md, bits); | |
554 | ||
9d09e663 N |
555 | return md_raid5_congested(&rs->md, bits); |
556 | } | |
557 | ||
b12d437b JB |
558 | /* |
559 | * This structure is never routinely used by userspace, unlike md superblocks. | |
560 | * Devices with this superblock should only ever be accessed via device-mapper. | |
561 | */ | |
562 | #define DM_RAID_MAGIC 0x64526D44 | |
563 | struct dm_raid_superblock { | |
564 | __le32 magic; /* "DmRd" */ | |
565 | __le32 features; /* Used to indicate possible future changes */ | |
566 | ||
567 | __le32 num_devices; /* Number of devices in this array. (Max 64) */ | |
568 | __le32 array_position; /* The position of this drive in the array */ | |
569 | ||
570 | __le64 events; /* Incremented by md when superblock updated */ | |
571 | __le64 failed_devices; /* Bit field of devices to indicate failures */ | |
572 | ||
573 | /* | |
574 | * This offset tracks the progress of the repair or replacement of | |
575 | * an individual drive. | |
576 | */ | |
577 | __le64 disk_recovery_offset; | |
578 | ||
579 | /* | |
580 | * This offset tracks the progress of the initial array | |
581 | * synchronisation/parity calculation. | |
582 | */ | |
583 | __le64 array_resync_offset; | |
584 | ||
585 | /* | |
586 | * RAID characteristics | |
587 | */ | |
588 | __le32 level; | |
589 | __le32 layout; | |
590 | __le32 stripe_sectors; | |
591 | ||
592 | __u8 pad[452]; /* Round struct to 512 bytes. */ | |
593 | /* Always set to 0 when writing. */ | |
594 | } __packed; | |
595 | ||
3cb03002 | 596 | static int read_disk_sb(struct md_rdev *rdev, int size) |
b12d437b JB |
597 | { |
598 | BUG_ON(!rdev->sb_page); | |
599 | ||
600 | if (rdev->sb_loaded) | |
601 | return 0; | |
602 | ||
603 | if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, 1)) { | |
0447568f JB |
604 | DMERR("Failed to read superblock of device at position %d", |
605 | rdev->raid_disk); | |
606 | set_bit(Faulty, &rdev->flags); | |
b12d437b JB |
607 | return -EINVAL; |
608 | } | |
609 | ||
610 | rdev->sb_loaded = 1; | |
611 | ||
612 | return 0; | |
613 | } | |
614 | ||
fd01b88c | 615 | static void super_sync(struct mddev *mddev, struct md_rdev *rdev) |
b12d437b | 616 | { |
dafb20fa | 617 | struct md_rdev *r; |
b12d437b JB |
618 | uint64_t failed_devices; |
619 | struct dm_raid_superblock *sb; | |
620 | ||
621 | sb = page_address(rdev->sb_page); | |
622 | failed_devices = le64_to_cpu(sb->failed_devices); | |
623 | ||
dafb20fa | 624 | rdev_for_each(r, mddev) |
b12d437b JB |
625 | if ((r->raid_disk >= 0) && test_bit(Faulty, &r->flags)) |
626 | failed_devices |= (1ULL << r->raid_disk); | |
627 | ||
628 | memset(sb, 0, sizeof(*sb)); | |
629 | ||
630 | sb->magic = cpu_to_le32(DM_RAID_MAGIC); | |
631 | sb->features = cpu_to_le32(0); /* No features yet */ | |
632 | ||
633 | sb->num_devices = cpu_to_le32(mddev->raid_disks); | |
634 | sb->array_position = cpu_to_le32(rdev->raid_disk); | |
635 | ||
636 | sb->events = cpu_to_le64(mddev->events); | |
637 | sb->failed_devices = cpu_to_le64(failed_devices); | |
638 | ||
639 | sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset); | |
640 | sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp); | |
641 | ||
642 | sb->level = cpu_to_le32(mddev->level); | |
643 | sb->layout = cpu_to_le32(mddev->layout); | |
644 | sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors); | |
645 | } | |
646 | ||
647 | /* | |
648 | * super_load | |
649 | * | |
650 | * This function creates a superblock if one is not found on the device | |
651 | * and will decide which superblock to use if there's a choice. | |
652 | * | |
653 | * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise | |
654 | */ | |
3cb03002 | 655 | static int super_load(struct md_rdev *rdev, struct md_rdev *refdev) |
b12d437b JB |
656 | { |
657 | int ret; | |
658 | struct dm_raid_superblock *sb; | |
659 | struct dm_raid_superblock *refsb; | |
660 | uint64_t events_sb, events_refsb; | |
661 | ||
662 | rdev->sb_start = 0; | |
663 | rdev->sb_size = sizeof(*sb); | |
664 | ||
665 | ret = read_disk_sb(rdev, rdev->sb_size); | |
666 | if (ret) | |
667 | return ret; | |
668 | ||
669 | sb = page_address(rdev->sb_page); | |
3aa3b2b2 JB |
670 | |
671 | /* | |
672 | * Two cases that we want to write new superblocks and rebuild: | |
673 | * 1) New device (no matching magic number) | |
674 | * 2) Device specified for rebuild (!In_sync w/ offset == 0) | |
675 | */ | |
676 | if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) || | |
677 | (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) { | |
b12d437b JB |
678 | super_sync(rdev->mddev, rdev); |
679 | ||
680 | set_bit(FirstUse, &rdev->flags); | |
681 | ||
682 | /* Force writing of superblocks to disk */ | |
683 | set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags); | |
684 | ||
685 | /* Any superblock is better than none, choose that if given */ | |
686 | return refdev ? 0 : 1; | |
687 | } | |
688 | ||
689 | if (!refdev) | |
690 | return 1; | |
691 | ||
692 | events_sb = le64_to_cpu(sb->events); | |
693 | ||
694 | refsb = page_address(refdev->sb_page); | |
695 | events_refsb = le64_to_cpu(refsb->events); | |
696 | ||
697 | return (events_sb > events_refsb) ? 1 : 0; | |
698 | } | |
699 | ||
fd01b88c | 700 | static int super_init_validation(struct mddev *mddev, struct md_rdev *rdev) |
b12d437b JB |
701 | { |
702 | int role; | |
703 | struct raid_set *rs = container_of(mddev, struct raid_set, md); | |
704 | uint64_t events_sb; | |
705 | uint64_t failed_devices; | |
706 | struct dm_raid_superblock *sb; | |
707 | uint32_t new_devs = 0; | |
708 | uint32_t rebuilds = 0; | |
dafb20fa | 709 | struct md_rdev *r; |
b12d437b JB |
710 | struct dm_raid_superblock *sb2; |
711 | ||
712 | sb = page_address(rdev->sb_page); | |
713 | events_sb = le64_to_cpu(sb->events); | |
714 | failed_devices = le64_to_cpu(sb->failed_devices); | |
715 | ||
716 | /* | |
717 | * Initialise to 1 if this is a new superblock. | |
718 | */ | |
719 | mddev->events = events_sb ? : 1; | |
720 | ||
721 | /* | |
722 | * Reshaping is not currently allowed | |
723 | */ | |
724 | if ((le32_to_cpu(sb->level) != mddev->level) || | |
725 | (le32_to_cpu(sb->layout) != mddev->layout) || | |
726 | (le32_to_cpu(sb->stripe_sectors) != mddev->chunk_sectors)) { | |
727 | DMERR("Reshaping arrays not yet supported."); | |
728 | return -EINVAL; | |
729 | } | |
730 | ||
731 | /* We can only change the number of devices in RAID1 right now */ | |
732 | if ((rs->raid_type->level != 1) && | |
733 | (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) { | |
734 | DMERR("Reshaping arrays not yet supported."); | |
735 | return -EINVAL; | |
736 | } | |
737 | ||
738 | if (!(rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC))) | |
739 | mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset); | |
740 | ||
741 | /* | |
742 | * During load, we set FirstUse if a new superblock was written. | |
743 | * There are two reasons we might not have a superblock: | |
744 | * 1) The array is brand new - in which case, all of the | |
745 | * devices must have their In_sync bit set. Also, | |
746 | * recovery_cp must be 0, unless forced. | |
747 | * 2) This is a new device being added to an old array | |
748 | * and the new device needs to be rebuilt - in which | |
749 | * case the In_sync bit will /not/ be set and | |
750 | * recovery_cp must be MaxSector. | |
751 | */ | |
dafb20fa | 752 | rdev_for_each(r, mddev) { |
b12d437b | 753 | if (!test_bit(In_sync, &r->flags)) { |
3aa3b2b2 JB |
754 | DMINFO("Device %d specified for rebuild: " |
755 | "Clearing superblock", r->raid_disk); | |
b12d437b JB |
756 | rebuilds++; |
757 | } else if (test_bit(FirstUse, &r->flags)) | |
758 | new_devs++; | |
759 | } | |
760 | ||
761 | if (!rebuilds) { | |
762 | if (new_devs == mddev->raid_disks) { | |
763 | DMINFO("Superblocks created for new array"); | |
764 | set_bit(MD_ARRAY_FIRST_USE, &mddev->flags); | |
765 | } else if (new_devs) { | |
766 | DMERR("New device injected " | |
767 | "into existing array without 'rebuild' " | |
768 | "parameter specified"); | |
769 | return -EINVAL; | |
770 | } | |
771 | } else if (new_devs) { | |
772 | DMERR("'rebuild' devices cannot be " | |
773 | "injected into an array with other first-time devices"); | |
774 | return -EINVAL; | |
775 | } else if (mddev->recovery_cp != MaxSector) { | |
776 | DMERR("'rebuild' specified while array is not in-sync"); | |
777 | return -EINVAL; | |
778 | } | |
779 | ||
780 | /* | |
781 | * Now we set the Faulty bit for those devices that are | |
782 | * recorded in the superblock as failed. | |
783 | */ | |
dafb20fa | 784 | rdev_for_each(r, mddev) { |
b12d437b JB |
785 | if (!r->sb_page) |
786 | continue; | |
787 | sb2 = page_address(r->sb_page); | |
788 | sb2->failed_devices = 0; | |
789 | ||
790 | /* | |
791 | * Check for any device re-ordering. | |
792 | */ | |
793 | if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) { | |
794 | role = le32_to_cpu(sb2->array_position); | |
795 | if (role != r->raid_disk) { | |
796 | if (rs->raid_type->level != 1) { | |
797 | rs->ti->error = "Cannot change device " | |
798 | "positions in RAID array"; | |
799 | return -EINVAL; | |
800 | } | |
801 | DMINFO("RAID1 device #%d now at position #%d", | |
802 | role, r->raid_disk); | |
803 | } | |
804 | ||
805 | /* | |
806 | * Partial recovery is performed on | |
807 | * returning failed devices. | |
808 | */ | |
809 | if (failed_devices & (1 << role)) | |
810 | set_bit(Faulty, &r->flags); | |
811 | } | |
812 | } | |
813 | ||
814 | return 0; | |
815 | } | |
816 | ||
fd01b88c | 817 | static int super_validate(struct mddev *mddev, struct md_rdev *rdev) |
b12d437b JB |
818 | { |
819 | struct dm_raid_superblock *sb = page_address(rdev->sb_page); | |
820 | ||
821 | /* | |
822 | * If mddev->events is not set, we know we have not yet initialized | |
823 | * the array. | |
824 | */ | |
825 | if (!mddev->events && super_init_validation(mddev, rdev)) | |
826 | return -EINVAL; | |
827 | ||
828 | mddev->bitmap_info.offset = 4096 >> 9; /* Enable bitmap creation */ | |
829 | rdev->mddev->bitmap_info.default_offset = 4096 >> 9; | |
830 | if (!test_bit(FirstUse, &rdev->flags)) { | |
831 | rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset); | |
832 | if (rdev->recovery_offset != MaxSector) | |
833 | clear_bit(In_sync, &rdev->flags); | |
834 | } | |
835 | ||
836 | /* | |
837 | * If a device comes back, set it as not In_sync and no longer faulty. | |
838 | */ | |
839 | if (test_bit(Faulty, &rdev->flags)) { | |
840 | clear_bit(Faulty, &rdev->flags); | |
841 | clear_bit(In_sync, &rdev->flags); | |
842 | rdev->saved_raid_disk = rdev->raid_disk; | |
843 | rdev->recovery_offset = 0; | |
844 | } | |
845 | ||
846 | clear_bit(FirstUse, &rdev->flags); | |
847 | ||
848 | return 0; | |
849 | } | |
850 | ||
851 | /* | |
852 | * Analyse superblocks and select the freshest. | |
853 | */ | |
854 | static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs) | |
855 | { | |
856 | int ret; | |
0447568f JB |
857 | unsigned redundancy = 0; |
858 | struct raid_dev *dev; | |
a9ad8526 | 859 | struct md_rdev *rdev, *tmp, *freshest; |
fd01b88c | 860 | struct mddev *mddev = &rs->md; |
b12d437b | 861 | |
0447568f JB |
862 | switch (rs->raid_type->level) { |
863 | case 1: | |
864 | redundancy = rs->md.raid_disks - 1; | |
865 | break; | |
866 | case 4: | |
867 | case 5: | |
868 | case 6: | |
869 | redundancy = rs->raid_type->parity_devs; | |
870 | break; | |
871 | default: | |
872 | ti->error = "Unknown RAID type"; | |
873 | return -EINVAL; | |
874 | } | |
875 | ||
b12d437b | 876 | freshest = NULL; |
a9ad8526 | 877 | rdev_for_each_safe(rdev, tmp, mddev) { |
b12d437b JB |
878 | if (!rdev->meta_bdev) |
879 | continue; | |
880 | ||
881 | ret = super_load(rdev, freshest); | |
882 | ||
883 | switch (ret) { | |
884 | case 1: | |
885 | freshest = rdev; | |
886 | break; | |
887 | case 0: | |
888 | break; | |
889 | default: | |
0447568f JB |
890 | dev = container_of(rdev, struct raid_dev, rdev); |
891 | if (redundancy--) { | |
892 | if (dev->meta_dev) | |
893 | dm_put_device(ti, dev->meta_dev); | |
894 | ||
895 | dev->meta_dev = NULL; | |
896 | rdev->meta_bdev = NULL; | |
897 | ||
898 | if (rdev->sb_page) | |
899 | put_page(rdev->sb_page); | |
900 | ||
901 | rdev->sb_page = NULL; | |
902 | ||
903 | rdev->sb_loaded = 0; | |
904 | ||
905 | /* | |
906 | * We might be able to salvage the data device | |
907 | * even though the meta device has failed. For | |
908 | * now, we behave as though '- -' had been | |
909 | * set for this device in the table. | |
910 | */ | |
911 | if (dev->data_dev) | |
912 | dm_put_device(ti, dev->data_dev); | |
913 | ||
914 | dev->data_dev = NULL; | |
915 | rdev->bdev = NULL; | |
916 | ||
917 | list_del(&rdev->same_set); | |
918 | ||
919 | continue; | |
920 | } | |
b12d437b JB |
921 | ti->error = "Failed to load superblock"; |
922 | return ret; | |
923 | } | |
924 | } | |
925 | ||
926 | if (!freshest) | |
927 | return 0; | |
928 | ||
929 | /* | |
930 | * Validation of the freshest device provides the source of | |
931 | * validation for the remaining devices. | |
932 | */ | |
933 | ti->error = "Unable to assemble array: Invalid superblocks"; | |
934 | if (super_validate(mddev, freshest)) | |
935 | return -EINVAL; | |
936 | ||
dafb20fa | 937 | rdev_for_each(rdev, mddev) |
b12d437b JB |
938 | if ((rdev != freshest) && super_validate(mddev, rdev)) |
939 | return -EINVAL; | |
940 | ||
941 | return 0; | |
942 | } | |
943 | ||
9d09e663 N |
944 | /* |
945 | * Construct a RAID4/5/6 mapping: | |
946 | * Args: | |
947 | * <raid_type> <#raid_params> <raid_params> \ | |
948 | * <#raid_devs> { <meta_dev1> <dev1> .. <meta_devN> <devN> } | |
949 | * | |
9d09e663 N |
950 | * <raid_params> varies by <raid_type>. See 'parse_raid_params' for |
951 | * details on possible <raid_params>. | |
952 | */ | |
953 | static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
954 | { | |
955 | int ret; | |
956 | struct raid_type *rt; | |
957 | unsigned long num_raid_params, num_raid_devs; | |
958 | struct raid_set *rs = NULL; | |
959 | ||
960 | /* Must have at least <raid_type> <#raid_params> */ | |
961 | if (argc < 2) { | |
962 | ti->error = "Too few arguments"; | |
963 | return -EINVAL; | |
964 | } | |
965 | ||
966 | /* raid type */ | |
967 | rt = get_raid_type(argv[0]); | |
968 | if (!rt) { | |
969 | ti->error = "Unrecognised raid_type"; | |
970 | return -EINVAL; | |
971 | } | |
972 | argc--; | |
973 | argv++; | |
974 | ||
975 | /* number of RAID parameters */ | |
976 | if (strict_strtoul(argv[0], 10, &num_raid_params) < 0) { | |
977 | ti->error = "Cannot understand number of RAID parameters"; | |
978 | return -EINVAL; | |
979 | } | |
980 | argc--; | |
981 | argv++; | |
982 | ||
983 | /* Skip over RAID params for now and find out # of devices */ | |
984 | if (num_raid_params + 1 > argc) { | |
985 | ti->error = "Arguments do not agree with counts given"; | |
986 | return -EINVAL; | |
987 | } | |
988 | ||
989 | if ((strict_strtoul(argv[num_raid_params], 10, &num_raid_devs) < 0) || | |
990 | (num_raid_devs >= INT_MAX)) { | |
991 | ti->error = "Cannot understand number of raid devices"; | |
992 | return -EINVAL; | |
993 | } | |
994 | ||
995 | rs = context_alloc(ti, rt, (unsigned)num_raid_devs); | |
996 | if (IS_ERR(rs)) | |
997 | return PTR_ERR(rs); | |
998 | ||
999 | ret = parse_raid_params(rs, argv, (unsigned)num_raid_params); | |
1000 | if (ret) | |
1001 | goto bad; | |
1002 | ||
1003 | ret = -EINVAL; | |
1004 | ||
1005 | argc -= num_raid_params + 1; /* +1: we already have num_raid_devs */ | |
1006 | argv += num_raid_params + 1; | |
1007 | ||
1008 | if (argc != (num_raid_devs * 2)) { | |
1009 | ti->error = "Supplied RAID devices does not match the count given"; | |
1010 | goto bad; | |
1011 | } | |
1012 | ||
1013 | ret = dev_parms(rs, argv); | |
1014 | if (ret) | |
1015 | goto bad; | |
1016 | ||
b12d437b JB |
1017 | rs->md.sync_super = super_sync; |
1018 | ret = analyse_superblocks(ti, rs); | |
1019 | if (ret) | |
1020 | goto bad; | |
1021 | ||
9d09e663 | 1022 | INIT_WORK(&rs->md.event_work, do_table_event); |
9d09e663 | 1023 | ti->private = rs; |
0ca93de9 | 1024 | ti->num_flush_requests = 1; |
9d09e663 N |
1025 | |
1026 | mutex_lock(&rs->md.reconfig_mutex); | |
1027 | ret = md_run(&rs->md); | |
1028 | rs->md.in_sync = 0; /* Assume already marked dirty */ | |
1029 | mutex_unlock(&rs->md.reconfig_mutex); | |
1030 | ||
1031 | if (ret) { | |
1032 | ti->error = "Fail to run raid array"; | |
1033 | goto bad; | |
1034 | } | |
1035 | ||
1036 | rs->callbacks.congested_fn = raid_is_congested; | |
9d09e663 N |
1037 | dm_table_add_target_callbacks(ti->table, &rs->callbacks); |
1038 | ||
32737279 | 1039 | mddev_suspend(&rs->md); |
9d09e663 N |
1040 | return 0; |
1041 | ||
1042 | bad: | |
1043 | context_free(rs); | |
1044 | ||
1045 | return ret; | |
1046 | } | |
1047 | ||
1048 | static void raid_dtr(struct dm_target *ti) | |
1049 | { | |
1050 | struct raid_set *rs = ti->private; | |
1051 | ||
1052 | list_del_init(&rs->callbacks.list); | |
1053 | md_stop(&rs->md); | |
1054 | context_free(rs); | |
1055 | } | |
1056 | ||
1057 | static int raid_map(struct dm_target *ti, struct bio *bio, union map_info *map_context) | |
1058 | { | |
1059 | struct raid_set *rs = ti->private; | |
fd01b88c | 1060 | struct mddev *mddev = &rs->md; |
9d09e663 N |
1061 | |
1062 | mddev->pers->make_request(mddev, bio); | |
1063 | ||
1064 | return DM_MAPIO_SUBMITTED; | |
1065 | } | |
1066 | ||
1067 | static int raid_status(struct dm_target *ti, status_type_t type, | |
1068 | char *result, unsigned maxlen) | |
1069 | { | |
1070 | struct raid_set *rs = ti->private; | |
1071 | unsigned raid_param_cnt = 1; /* at least 1 for chunksize */ | |
1072 | unsigned sz = 0; | |
2e727c3c | 1073 | int i, array_in_sync = 0; |
9d09e663 N |
1074 | sector_t sync; |
1075 | ||
1076 | switch (type) { | |
1077 | case STATUSTYPE_INFO: | |
1078 | DMEMIT("%s %d ", rs->raid_type->name, rs->md.raid_disks); | |
1079 | ||
9d09e663 N |
1080 | if (test_bit(MD_RECOVERY_RUNNING, &rs->md.recovery)) |
1081 | sync = rs->md.curr_resync_completed; | |
1082 | else | |
1083 | sync = rs->md.recovery_cp; | |
1084 | ||
2e727c3c JB |
1085 | if (sync >= rs->md.resync_max_sectors) { |
1086 | array_in_sync = 1; | |
9d09e663 | 1087 | sync = rs->md.resync_max_sectors; |
2e727c3c JB |
1088 | } else { |
1089 | /* | |
1090 | * The array may be doing an initial sync, or it may | |
1091 | * be rebuilding individual components. If all the | |
1092 | * devices are In_sync, then it is the array that is | |
1093 | * being initialized. | |
1094 | */ | |
1095 | for (i = 0; i < rs->md.raid_disks; i++) | |
1096 | if (!test_bit(In_sync, &rs->dev[i].rdev.flags)) | |
1097 | array_in_sync = 1; | |
1098 | } | |
1099 | /* | |
1100 | * Status characters: | |
1101 | * 'D' = Dead/Failed device | |
1102 | * 'a' = Alive but not in-sync | |
1103 | * 'A' = Alive and in-sync | |
1104 | */ | |
1105 | for (i = 0; i < rs->md.raid_disks; i++) { | |
1106 | if (test_bit(Faulty, &rs->dev[i].rdev.flags)) | |
1107 | DMEMIT("D"); | |
1108 | else if (!array_in_sync || | |
1109 | !test_bit(In_sync, &rs->dev[i].rdev.flags)) | |
1110 | DMEMIT("a"); | |
1111 | else | |
1112 | DMEMIT("A"); | |
1113 | } | |
9d09e663 | 1114 | |
2e727c3c JB |
1115 | /* |
1116 | * In-sync ratio: | |
1117 | * The in-sync ratio shows the progress of: | |
1118 | * - Initializing the array | |
1119 | * - Rebuilding a subset of devices of the array | |
1120 | * The user can distinguish between the two by referring | |
1121 | * to the status characters. | |
1122 | */ | |
9d09e663 N |
1123 | DMEMIT(" %llu/%llu", |
1124 | (unsigned long long) sync, | |
1125 | (unsigned long long) rs->md.resync_max_sectors); | |
1126 | ||
1127 | break; | |
1128 | case STATUSTYPE_TABLE: | |
1129 | /* The string you would use to construct this array */ | |
46bed2b5 | 1130 | for (i = 0; i < rs->md.raid_disks; i++) { |
13c87583 JB |
1131 | if ((rs->print_flags & DMPF_REBUILD) && |
1132 | rs->dev[i].data_dev && | |
9d09e663 | 1133 | !test_bit(In_sync, &rs->dev[i].rdev.flags)) |
13c87583 | 1134 | raid_param_cnt += 2; /* for rebuilds */ |
46bed2b5 JB |
1135 | if (rs->dev[i].data_dev && |
1136 | test_bit(WriteMostly, &rs->dev[i].rdev.flags)) | |
1137 | raid_param_cnt += 2; | |
1138 | } | |
9d09e663 | 1139 | |
34f8ac6d | 1140 | raid_param_cnt += (hweight32(rs->print_flags & ~DMPF_REBUILD) * 2); |
9d09e663 N |
1141 | if (rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC)) |
1142 | raid_param_cnt--; | |
1143 | ||
1144 | DMEMIT("%s %u %u", rs->raid_type->name, | |
1145 | raid_param_cnt, rs->md.chunk_sectors); | |
1146 | ||
1147 | if ((rs->print_flags & DMPF_SYNC) && | |
1148 | (rs->md.recovery_cp == MaxSector)) | |
1149 | DMEMIT(" sync"); | |
1150 | if (rs->print_flags & DMPF_NOSYNC) | |
1151 | DMEMIT(" nosync"); | |
1152 | ||
1153 | for (i = 0; i < rs->md.raid_disks; i++) | |
13c87583 JB |
1154 | if ((rs->print_flags & DMPF_REBUILD) && |
1155 | rs->dev[i].data_dev && | |
9d09e663 N |
1156 | !test_bit(In_sync, &rs->dev[i].rdev.flags)) |
1157 | DMEMIT(" rebuild %u", i); | |
1158 | ||
1159 | if (rs->print_flags & DMPF_DAEMON_SLEEP) | |
1160 | DMEMIT(" daemon_sleep %lu", | |
1161 | rs->md.bitmap_info.daemon_sleep); | |
1162 | ||
1163 | if (rs->print_flags & DMPF_MIN_RECOVERY_RATE) | |
1164 | DMEMIT(" min_recovery_rate %d", rs->md.sync_speed_min); | |
1165 | ||
1166 | if (rs->print_flags & DMPF_MAX_RECOVERY_RATE) | |
1167 | DMEMIT(" max_recovery_rate %d", rs->md.sync_speed_max); | |
1168 | ||
46bed2b5 JB |
1169 | for (i = 0; i < rs->md.raid_disks; i++) |
1170 | if (rs->dev[i].data_dev && | |
1171 | test_bit(WriteMostly, &rs->dev[i].rdev.flags)) | |
1172 | DMEMIT(" write_mostly %u", i); | |
1173 | ||
9d09e663 N |
1174 | if (rs->print_flags & DMPF_MAX_WRITE_BEHIND) |
1175 | DMEMIT(" max_write_behind %lu", | |
1176 | rs->md.bitmap_info.max_write_behind); | |
1177 | ||
1178 | if (rs->print_flags & DMPF_STRIPE_CACHE) { | |
d1688a6d | 1179 | struct r5conf *conf = rs->md.private; |
9d09e663 N |
1180 | |
1181 | /* convert from kiB to sectors */ | |
1182 | DMEMIT(" stripe_cache %d", | |
1183 | conf ? conf->max_nr_stripes * 2 : 0); | |
1184 | } | |
1185 | ||
c1084561 JB |
1186 | if (rs->print_flags & DMPF_REGION_SIZE) |
1187 | DMEMIT(" region_size %lu", | |
1188 | rs->md.bitmap_info.chunksize >> 9); | |
1189 | ||
9d09e663 N |
1190 | DMEMIT(" %d", rs->md.raid_disks); |
1191 | for (i = 0; i < rs->md.raid_disks; i++) { | |
b12d437b JB |
1192 | if (rs->dev[i].meta_dev) |
1193 | DMEMIT(" %s", rs->dev[i].meta_dev->name); | |
1194 | else | |
1195 | DMEMIT(" -"); | |
9d09e663 N |
1196 | |
1197 | if (rs->dev[i].data_dev) | |
1198 | DMEMIT(" %s", rs->dev[i].data_dev->name); | |
1199 | else | |
1200 | DMEMIT(" -"); | |
1201 | } | |
1202 | } | |
1203 | ||
1204 | return 0; | |
1205 | } | |
1206 | ||
1207 | static int raid_iterate_devices(struct dm_target *ti, iterate_devices_callout_fn fn, void *data) | |
1208 | { | |
1209 | struct raid_set *rs = ti->private; | |
1210 | unsigned i; | |
1211 | int ret = 0; | |
1212 | ||
1213 | for (i = 0; !ret && i < rs->md.raid_disks; i++) | |
1214 | if (rs->dev[i].data_dev) | |
1215 | ret = fn(ti, | |
1216 | rs->dev[i].data_dev, | |
1217 | 0, /* No offset on data devs */ | |
1218 | rs->md.dev_sectors, | |
1219 | data); | |
1220 | ||
1221 | return ret; | |
1222 | } | |
1223 | ||
1224 | static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits) | |
1225 | { | |
1226 | struct raid_set *rs = ti->private; | |
1227 | unsigned chunk_size = rs->md.chunk_sectors << 9; | |
d1688a6d | 1228 | struct r5conf *conf = rs->md.private; |
9d09e663 N |
1229 | |
1230 | blk_limits_io_min(limits, chunk_size); | |
1231 | blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded)); | |
1232 | } | |
1233 | ||
1234 | static void raid_presuspend(struct dm_target *ti) | |
1235 | { | |
1236 | struct raid_set *rs = ti->private; | |
1237 | ||
1238 | md_stop_writes(&rs->md); | |
1239 | } | |
1240 | ||
1241 | static void raid_postsuspend(struct dm_target *ti) | |
1242 | { | |
1243 | struct raid_set *rs = ti->private; | |
1244 | ||
1245 | mddev_suspend(&rs->md); | |
1246 | } | |
1247 | ||
1248 | static void raid_resume(struct dm_target *ti) | |
1249 | { | |
1250 | struct raid_set *rs = ti->private; | |
1251 | ||
34f8ac6d JB |
1252 | if (!rs->bitmap_loaded) { |
1253 | bitmap_load(&rs->md); | |
1254 | rs->bitmap_loaded = 1; | |
1255 | } else | |
1256 | md_wakeup_thread(rs->md.thread); | |
1257 | ||
9d09e663 N |
1258 | mddev_resume(&rs->md); |
1259 | } | |
1260 | ||
1261 | static struct target_type raid_target = { | |
1262 | .name = "raid", | |
0447568f | 1263 | .version = {1, 2, 0}, |
9d09e663 N |
1264 | .module = THIS_MODULE, |
1265 | .ctr = raid_ctr, | |
1266 | .dtr = raid_dtr, | |
1267 | .map = raid_map, | |
1268 | .status = raid_status, | |
1269 | .iterate_devices = raid_iterate_devices, | |
1270 | .io_hints = raid_io_hints, | |
1271 | .presuspend = raid_presuspend, | |
1272 | .postsuspend = raid_postsuspend, | |
1273 | .resume = raid_resume, | |
1274 | }; | |
1275 | ||
1276 | static int __init dm_raid_init(void) | |
1277 | { | |
1278 | return dm_register_target(&raid_target); | |
1279 | } | |
1280 | ||
1281 | static void __exit dm_raid_exit(void) | |
1282 | { | |
1283 | dm_unregister_target(&raid_target); | |
1284 | } | |
1285 | ||
1286 | module_init(dm_raid_init); | |
1287 | module_exit(dm_raid_exit); | |
1288 | ||
1289 | MODULE_DESCRIPTION(DM_NAME " raid4/5/6 target"); | |
1290 | MODULE_ALIAS("dm-raid4"); | |
1291 | MODULE_ALIAS("dm-raid5"); | |
1292 | MODULE_ALIAS("dm-raid6"); | |
1293 | MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>"); | |
1294 | MODULE_LICENSE("GPL"); |