]>
Commit | Line | Data |
---|---|---|
9d09e663 N |
1 | /* |
2 | * Copyright (C) 2010-2011 Neil Brown | |
3 | * Copyright (C) 2010-2011 Red Hat, Inc. All rights reserved. | |
4 | * | |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include <linux/slab.h> | |
9 | ||
10 | #include "md.h" | |
11 | #include "raid5.h" | |
9d09e663 N |
12 | #include "bitmap.h" |
13 | ||
3e8dbb7f AK |
14 | #include <linux/device-mapper.h> |
15 | ||
9d09e663 N |
16 | #define DM_MSG_PREFIX "raid" |
17 | ||
18 | /* | |
19 | * If the MD doesn't support MD_SYNC_STATE_FORCED yet, then | |
20 | * make it so the flag doesn't set anything. | |
21 | */ | |
22 | #ifndef MD_SYNC_STATE_FORCED | |
23 | #define MD_SYNC_STATE_FORCED 0 | |
24 | #endif | |
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; | |
41 | struct mdk_rdev_s rdev; | |
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 | ||
59 | uint64_t print_flags; | |
60 | ||
61 | struct mddev_s md; | |
62 | struct raid_type *raid_type; | |
63 | struct dm_target_callbacks callbacks; | |
64 | ||
65 | struct raid_dev dev[0]; | |
66 | }; | |
67 | ||
68 | /* Supported raid types and properties. */ | |
69 | static struct raid_type { | |
70 | const char *name; /* RAID algorithm. */ | |
71 | const char *descr; /* Descriptor text for logging. */ | |
72 | const unsigned parity_devs; /* # of parity devices. */ | |
73 | const unsigned minimal_devs; /* minimal # of devices in set. */ | |
74 | const unsigned level; /* RAID level. */ | |
75 | const unsigned algorithm; /* RAID algorithm. */ | |
76 | } raid_types[] = { | |
77 | {"raid4", "RAID4 (dedicated parity disk)", 1, 2, 5, ALGORITHM_PARITY_0}, | |
78 | {"raid5_la", "RAID5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC}, | |
79 | {"raid5_ra", "RAID5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC}, | |
80 | {"raid5_ls", "RAID5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC}, | |
81 | {"raid5_rs", "RAID5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC}, | |
82 | {"raid6_zr", "RAID6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART}, | |
83 | {"raid6_nr", "RAID6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART}, | |
84 | {"raid6_nc", "RAID6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE} | |
85 | }; | |
86 | ||
87 | static struct raid_type *get_raid_type(char *name) | |
88 | { | |
89 | int i; | |
90 | ||
91 | for (i = 0; i < ARRAY_SIZE(raid_types); i++) | |
92 | if (!strcmp(raid_types[i].name, name)) | |
93 | return &raid_types[i]; | |
94 | ||
95 | return NULL; | |
96 | } | |
97 | ||
98 | static struct raid_set *context_alloc(struct dm_target *ti, struct raid_type *raid_type, unsigned raid_devs) | |
99 | { | |
100 | unsigned i; | |
101 | struct raid_set *rs; | |
102 | sector_t sectors_per_dev; | |
103 | ||
104 | if (raid_devs <= raid_type->parity_devs) { | |
105 | ti->error = "Insufficient number of devices"; | |
106 | return ERR_PTR(-EINVAL); | |
107 | } | |
108 | ||
109 | sectors_per_dev = ti->len; | |
110 | if (sector_div(sectors_per_dev, (raid_devs - raid_type->parity_devs))) { | |
111 | ti->error = "Target length not divisible by number of data devices"; | |
112 | return ERR_PTR(-EINVAL); | |
113 | } | |
114 | ||
115 | rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL); | |
116 | if (!rs) { | |
117 | ti->error = "Cannot allocate raid context"; | |
118 | return ERR_PTR(-ENOMEM); | |
119 | } | |
120 | ||
121 | mddev_init(&rs->md); | |
122 | ||
123 | rs->ti = ti; | |
124 | rs->raid_type = raid_type; | |
125 | rs->md.raid_disks = raid_devs; | |
126 | rs->md.level = raid_type->level; | |
127 | rs->md.new_level = rs->md.level; | |
128 | rs->md.dev_sectors = sectors_per_dev; | |
129 | rs->md.layout = raid_type->algorithm; | |
130 | rs->md.new_layout = rs->md.layout; | |
131 | rs->md.delta_disks = 0; | |
132 | rs->md.recovery_cp = 0; | |
133 | ||
134 | for (i = 0; i < raid_devs; i++) | |
135 | md_rdev_init(&rs->dev[i].rdev); | |
136 | ||
137 | /* | |
138 | * Remaining items to be initialized by further RAID params: | |
139 | * rs->md.persistent | |
140 | * rs->md.external | |
141 | * rs->md.chunk_sectors | |
142 | * rs->md.new_chunk_sectors | |
143 | */ | |
144 | ||
145 | return rs; | |
146 | } | |
147 | ||
148 | static void context_free(struct raid_set *rs) | |
149 | { | |
150 | int i; | |
151 | ||
152 | for (i = 0; i < rs->md.raid_disks; i++) | |
153 | if (rs->dev[i].data_dev) | |
154 | dm_put_device(rs->ti, rs->dev[i].data_dev); | |
155 | ||
156 | kfree(rs); | |
157 | } | |
158 | ||
159 | /* | |
160 | * For every device we have two words | |
161 | * <meta_dev>: meta device name or '-' if missing | |
162 | * <data_dev>: data device name or '-' if missing | |
163 | * | |
164 | * This code parses those words. | |
165 | */ | |
166 | static int dev_parms(struct raid_set *rs, char **argv) | |
167 | { | |
168 | int i; | |
169 | int rebuild = 0; | |
170 | int metadata_available = 0; | |
171 | int ret = 0; | |
172 | ||
173 | for (i = 0; i < rs->md.raid_disks; i++, argv += 2) { | |
174 | rs->dev[i].rdev.raid_disk = i; | |
175 | ||
176 | rs->dev[i].meta_dev = NULL; | |
177 | rs->dev[i].data_dev = NULL; | |
178 | ||
179 | /* | |
180 | * There are no offsets, since there is a separate device | |
181 | * for data and metadata. | |
182 | */ | |
183 | rs->dev[i].rdev.data_offset = 0; | |
184 | rs->dev[i].rdev.mddev = &rs->md; | |
185 | ||
186 | if (strcmp(argv[0], "-")) { | |
187 | rs->ti->error = "Metadata devices not supported"; | |
188 | return -EINVAL; | |
189 | } | |
190 | ||
191 | if (!strcmp(argv[1], "-")) { | |
192 | if (!test_bit(In_sync, &rs->dev[i].rdev.flags) && | |
193 | (!rs->dev[i].rdev.recovery_offset)) { | |
194 | rs->ti->error = "Drive designated for rebuild not specified"; | |
195 | return -EINVAL; | |
196 | } | |
197 | ||
198 | continue; | |
199 | } | |
200 | ||
201 | ret = dm_get_device(rs->ti, argv[1], | |
202 | dm_table_get_mode(rs->ti->table), | |
203 | &rs->dev[i].data_dev); | |
204 | if (ret) { | |
205 | rs->ti->error = "RAID device lookup failure"; | |
206 | return ret; | |
207 | } | |
208 | ||
209 | rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev; | |
210 | list_add(&rs->dev[i].rdev.same_set, &rs->md.disks); | |
211 | if (!test_bit(In_sync, &rs->dev[i].rdev.flags)) | |
212 | rebuild++; | |
213 | } | |
214 | ||
215 | if (metadata_available) { | |
216 | rs->md.external = 0; | |
217 | rs->md.persistent = 1; | |
218 | rs->md.major_version = 2; | |
219 | } else if (rebuild && !rs->md.recovery_cp) { | |
220 | /* | |
221 | * Without metadata, we will not be able to tell if the array | |
222 | * is in-sync or not - we must assume it is not. Therefore, | |
223 | * it is impossible to rebuild a drive. | |
224 | * | |
225 | * Even if there is metadata, the on-disk information may | |
226 | * indicate that the array is not in-sync and it will then | |
227 | * fail at that time. | |
228 | * | |
229 | * User could specify 'nosync' option if desperate. | |
230 | */ | |
231 | DMERR("Unable to rebuild drive while array is not in-sync"); | |
232 | rs->ti->error = "RAID device lookup failure"; | |
233 | return -EINVAL; | |
234 | } | |
235 | ||
236 | return 0; | |
237 | } | |
238 | ||
c1084561 JB |
239 | /* |
240 | * validate_region_size | |
241 | * @rs | |
242 | * @region_size: region size in sectors. If 0, pick a size (4MiB default). | |
243 | * | |
244 | * Set rs->md.bitmap_info.chunksize (which really refers to 'region size'). | |
245 | * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap. | |
246 | * | |
247 | * Returns: 0 on success, -EINVAL on failure. | |
248 | */ | |
249 | static int validate_region_size(struct raid_set *rs, unsigned long region_size) | |
250 | { | |
251 | unsigned long min_region_size = rs->ti->len / (1 << 21); | |
252 | ||
253 | if (!region_size) { | |
254 | /* | |
255 | * Choose a reasonable default. All figures in sectors. | |
256 | */ | |
257 | if (min_region_size > (1 << 13)) { | |
258 | DMINFO("Choosing default region size of %lu sectors", | |
259 | region_size); | |
260 | region_size = min_region_size; | |
261 | } else { | |
262 | DMINFO("Choosing default region size of 4MiB"); | |
263 | region_size = 1 << 13; /* sectors */ | |
264 | } | |
265 | } else { | |
266 | /* | |
267 | * Validate user-supplied value. | |
268 | */ | |
269 | if (region_size > rs->ti->len) { | |
270 | rs->ti->error = "Supplied region size is too large"; | |
271 | return -EINVAL; | |
272 | } | |
273 | ||
274 | if (region_size < min_region_size) { | |
275 | DMERR("Supplied region_size (%lu sectors) below minimum (%lu)", | |
276 | region_size, min_region_size); | |
277 | rs->ti->error = "Supplied region size is too small"; | |
278 | return -EINVAL; | |
279 | } | |
280 | ||
281 | if (!is_power_of_2(region_size)) { | |
282 | rs->ti->error = "Region size is not a power of 2"; | |
283 | return -EINVAL; | |
284 | } | |
285 | ||
286 | if (region_size < rs->md.chunk_sectors) { | |
287 | rs->ti->error = "Region size is smaller than the chunk size"; | |
288 | return -EINVAL; | |
289 | } | |
290 | } | |
291 | ||
292 | /* | |
293 | * Convert sectors to bytes. | |
294 | */ | |
295 | rs->md.bitmap_info.chunksize = (region_size << 9); | |
296 | ||
297 | return 0; | |
298 | } | |
299 | ||
9d09e663 N |
300 | /* |
301 | * Possible arguments are... | |
302 | * RAID456: | |
303 | * <chunk_size> [optional_args] | |
304 | * | |
305 | * Optional args: | |
306 | * [[no]sync] Force or prevent recovery of the entire array | |
307 | * [rebuild <idx>] Rebuild the drive indicated by the index | |
308 | * [daemon_sleep <ms>] Time between bitmap daemon work to clear bits | |
309 | * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization | |
310 | * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization | |
46bed2b5 | 311 | * [write_mostly <idx>] Indicate a write mostly drive via index |
9d09e663 N |
312 | * [max_write_behind <sectors>] See '-write-behind=' (man mdadm) |
313 | * [stripe_cache <sectors>] Stripe cache size for higher RAIDs | |
c1084561 | 314 | * [region_size <sectors>] Defines granularity of bitmap |
9d09e663 N |
315 | */ |
316 | static int parse_raid_params(struct raid_set *rs, char **argv, | |
317 | unsigned num_raid_params) | |
318 | { | |
319 | unsigned i, rebuild_cnt = 0; | |
c1084561 | 320 | unsigned long value, region_size = 0; |
9d09e663 N |
321 | char *key; |
322 | ||
323 | /* | |
324 | * First, parse the in-order required arguments | |
325 | */ | |
326 | if ((strict_strtoul(argv[0], 10, &value) < 0) || | |
327 | !is_power_of_2(value) || (value < 8)) { | |
328 | rs->ti->error = "Bad chunk size"; | |
329 | return -EINVAL; | |
330 | } | |
331 | ||
332 | rs->md.new_chunk_sectors = rs->md.chunk_sectors = value; | |
333 | argv++; | |
334 | num_raid_params--; | |
335 | ||
336 | /* | |
337 | * Second, parse the unordered optional arguments | |
338 | */ | |
339 | for (i = 0; i < rs->md.raid_disks; i++) | |
340 | set_bit(In_sync, &rs->dev[i].rdev.flags); | |
341 | ||
342 | for (i = 0; i < num_raid_params; i++) { | |
13c87583 | 343 | if (!strcasecmp(argv[i], "nosync")) { |
9d09e663 N |
344 | rs->md.recovery_cp = MaxSector; |
345 | rs->print_flags |= DMPF_NOSYNC; | |
346 | rs->md.flags |= MD_SYNC_STATE_FORCED; | |
347 | continue; | |
348 | } | |
13c87583 | 349 | if (!strcasecmp(argv[i], "sync")) { |
9d09e663 N |
350 | rs->md.recovery_cp = 0; |
351 | rs->print_flags |= DMPF_SYNC; | |
352 | rs->md.flags |= MD_SYNC_STATE_FORCED; | |
353 | continue; | |
354 | } | |
355 | ||
356 | /* The rest of the optional arguments come in key/value pairs */ | |
357 | if ((i + 1) >= num_raid_params) { | |
358 | rs->ti->error = "Wrong number of raid parameters given"; | |
359 | return -EINVAL; | |
360 | } | |
361 | ||
362 | key = argv[i++]; | |
363 | if (strict_strtoul(argv[i], 10, &value) < 0) { | |
364 | rs->ti->error = "Bad numerical argument given in raid params"; | |
365 | return -EINVAL; | |
366 | } | |
367 | ||
13c87583 | 368 | if (!strcasecmp(key, "rebuild")) { |
9d09e663 N |
369 | if (++rebuild_cnt > rs->raid_type->parity_devs) { |
370 | rs->ti->error = "Too many rebuild drives given"; | |
371 | return -EINVAL; | |
372 | } | |
373 | if (value > rs->md.raid_disks) { | |
374 | rs->ti->error = "Invalid rebuild index given"; | |
375 | return -EINVAL; | |
376 | } | |
377 | clear_bit(In_sync, &rs->dev[value].rdev.flags); | |
378 | rs->dev[value].rdev.recovery_offset = 0; | |
13c87583 | 379 | rs->print_flags |= DMPF_REBUILD; |
46bed2b5 JB |
380 | } else if (!strcasecmp(key, "write_mostly")) { |
381 | if (rs->raid_type->level != 1) { | |
382 | rs->ti->error = "write_mostly option is only valid for RAID1"; | |
383 | return -EINVAL; | |
384 | } | |
385 | if (value > rs->md.raid_disks) { | |
386 | rs->ti->error = "Invalid write_mostly drive index given"; | |
387 | return -EINVAL; | |
388 | } | |
389 | set_bit(WriteMostly, &rs->dev[value].rdev.flags); | |
13c87583 | 390 | } else if (!strcasecmp(key, "max_write_behind")) { |
46bed2b5 JB |
391 | if (rs->raid_type->level != 1) { |
392 | rs->ti->error = "max_write_behind option is only valid for RAID1"; | |
393 | return -EINVAL; | |
394 | } | |
9d09e663 N |
395 | rs->print_flags |= DMPF_MAX_WRITE_BEHIND; |
396 | ||
397 | /* | |
398 | * In device-mapper, we specify things in sectors, but | |
399 | * MD records this value in kB | |
400 | */ | |
401 | value /= 2; | |
402 | if (value > COUNTER_MAX) { | |
403 | rs->ti->error = "Max write-behind limit out of range"; | |
404 | return -EINVAL; | |
405 | } | |
406 | rs->md.bitmap_info.max_write_behind = value; | |
13c87583 | 407 | } else if (!strcasecmp(key, "daemon_sleep")) { |
9d09e663 N |
408 | rs->print_flags |= DMPF_DAEMON_SLEEP; |
409 | if (!value || (value > MAX_SCHEDULE_TIMEOUT)) { | |
410 | rs->ti->error = "daemon sleep period out of range"; | |
411 | return -EINVAL; | |
412 | } | |
413 | rs->md.bitmap_info.daemon_sleep = value; | |
13c87583 | 414 | } else if (!strcasecmp(key, "stripe_cache")) { |
9d09e663 N |
415 | rs->print_flags |= DMPF_STRIPE_CACHE; |
416 | ||
417 | /* | |
418 | * In device-mapper, we specify things in sectors, but | |
419 | * MD records this value in kB | |
420 | */ | |
421 | value /= 2; | |
422 | ||
423 | if (rs->raid_type->level < 5) { | |
424 | rs->ti->error = "Inappropriate argument: stripe_cache"; | |
425 | return -EINVAL; | |
426 | } | |
427 | if (raid5_set_cache_size(&rs->md, (int)value)) { | |
428 | rs->ti->error = "Bad stripe_cache size"; | |
429 | return -EINVAL; | |
430 | } | |
13c87583 | 431 | } else if (!strcasecmp(key, "min_recovery_rate")) { |
9d09e663 N |
432 | rs->print_flags |= DMPF_MIN_RECOVERY_RATE; |
433 | if (value > INT_MAX) { | |
434 | rs->ti->error = "min_recovery_rate out of range"; | |
435 | return -EINVAL; | |
436 | } | |
437 | rs->md.sync_speed_min = (int)value; | |
13c87583 | 438 | } else if (!strcasecmp(key, "max_recovery_rate")) { |
9d09e663 N |
439 | rs->print_flags |= DMPF_MAX_RECOVERY_RATE; |
440 | if (value > INT_MAX) { | |
441 | rs->ti->error = "max_recovery_rate out of range"; | |
442 | return -EINVAL; | |
443 | } | |
444 | rs->md.sync_speed_max = (int)value; | |
c1084561 JB |
445 | } else if (!strcasecmp(key, "region_size")) { |
446 | rs->print_flags |= DMPF_REGION_SIZE; | |
447 | region_size = value; | |
9d09e663 N |
448 | } else { |
449 | DMERR("Unable to parse RAID parameter: %s", key); | |
450 | rs->ti->error = "Unable to parse RAID parameters"; | |
451 | return -EINVAL; | |
452 | } | |
453 | } | |
454 | ||
c1084561 JB |
455 | if (validate_region_size(rs, region_size)) |
456 | return -EINVAL; | |
457 | ||
458 | if (rs->md.chunk_sectors) | |
459 | rs->ti->split_io = rs->md.chunk_sectors; | |
460 | else | |
461 | rs->ti->split_io = region_size; | |
462 | ||
9d09e663 N |
463 | /* Assume there are no metadata devices until the drives are parsed */ |
464 | rs->md.persistent = 0; | |
465 | rs->md.external = 1; | |
466 | ||
467 | return 0; | |
468 | } | |
469 | ||
470 | static void do_table_event(struct work_struct *ws) | |
471 | { | |
472 | struct raid_set *rs = container_of(ws, struct raid_set, md.event_work); | |
473 | ||
474 | dm_table_event(rs->ti->table); | |
475 | } | |
476 | ||
477 | static int raid_is_congested(struct dm_target_callbacks *cb, int bits) | |
478 | { | |
479 | struct raid_set *rs = container_of(cb, struct raid_set, callbacks); | |
480 | ||
481 | return md_raid5_congested(&rs->md, bits); | |
482 | } | |
483 | ||
9d09e663 N |
484 | /* |
485 | * Construct a RAID4/5/6 mapping: | |
486 | * Args: | |
487 | * <raid_type> <#raid_params> <raid_params> \ | |
488 | * <#raid_devs> { <meta_dev1> <dev1> .. <meta_devN> <devN> } | |
489 | * | |
490 | * ** metadata devices are not supported yet, use '-' instead ** | |
491 | * | |
492 | * <raid_params> varies by <raid_type>. See 'parse_raid_params' for | |
493 | * details on possible <raid_params>. | |
494 | */ | |
495 | static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv) | |
496 | { | |
497 | int ret; | |
498 | struct raid_type *rt; | |
499 | unsigned long num_raid_params, num_raid_devs; | |
500 | struct raid_set *rs = NULL; | |
501 | ||
502 | /* Must have at least <raid_type> <#raid_params> */ | |
503 | if (argc < 2) { | |
504 | ti->error = "Too few arguments"; | |
505 | return -EINVAL; | |
506 | } | |
507 | ||
508 | /* raid type */ | |
509 | rt = get_raid_type(argv[0]); | |
510 | if (!rt) { | |
511 | ti->error = "Unrecognised raid_type"; | |
512 | return -EINVAL; | |
513 | } | |
514 | argc--; | |
515 | argv++; | |
516 | ||
517 | /* number of RAID parameters */ | |
518 | if (strict_strtoul(argv[0], 10, &num_raid_params) < 0) { | |
519 | ti->error = "Cannot understand number of RAID parameters"; | |
520 | return -EINVAL; | |
521 | } | |
522 | argc--; | |
523 | argv++; | |
524 | ||
525 | /* Skip over RAID params for now and find out # of devices */ | |
526 | if (num_raid_params + 1 > argc) { | |
527 | ti->error = "Arguments do not agree with counts given"; | |
528 | return -EINVAL; | |
529 | } | |
530 | ||
531 | if ((strict_strtoul(argv[num_raid_params], 10, &num_raid_devs) < 0) || | |
532 | (num_raid_devs >= INT_MAX)) { | |
533 | ti->error = "Cannot understand number of raid devices"; | |
534 | return -EINVAL; | |
535 | } | |
536 | ||
537 | rs = context_alloc(ti, rt, (unsigned)num_raid_devs); | |
538 | if (IS_ERR(rs)) | |
539 | return PTR_ERR(rs); | |
540 | ||
541 | ret = parse_raid_params(rs, argv, (unsigned)num_raid_params); | |
542 | if (ret) | |
543 | goto bad; | |
544 | ||
545 | ret = -EINVAL; | |
546 | ||
547 | argc -= num_raid_params + 1; /* +1: we already have num_raid_devs */ | |
548 | argv += num_raid_params + 1; | |
549 | ||
550 | if (argc != (num_raid_devs * 2)) { | |
551 | ti->error = "Supplied RAID devices does not match the count given"; | |
552 | goto bad; | |
553 | } | |
554 | ||
555 | ret = dev_parms(rs, argv); | |
556 | if (ret) | |
557 | goto bad; | |
558 | ||
559 | INIT_WORK(&rs->md.event_work, do_table_event); | |
9d09e663 N |
560 | ti->private = rs; |
561 | ||
562 | mutex_lock(&rs->md.reconfig_mutex); | |
563 | ret = md_run(&rs->md); | |
564 | rs->md.in_sync = 0; /* Assume already marked dirty */ | |
565 | mutex_unlock(&rs->md.reconfig_mutex); | |
566 | ||
567 | if (ret) { | |
568 | ti->error = "Fail to run raid array"; | |
569 | goto bad; | |
570 | } | |
571 | ||
572 | rs->callbacks.congested_fn = raid_is_congested; | |
9d09e663 N |
573 | dm_table_add_target_callbacks(ti->table, &rs->callbacks); |
574 | ||
575 | return 0; | |
576 | ||
577 | bad: | |
578 | context_free(rs); | |
579 | ||
580 | return ret; | |
581 | } | |
582 | ||
583 | static void raid_dtr(struct dm_target *ti) | |
584 | { | |
585 | struct raid_set *rs = ti->private; | |
586 | ||
587 | list_del_init(&rs->callbacks.list); | |
588 | md_stop(&rs->md); | |
589 | context_free(rs); | |
590 | } | |
591 | ||
592 | static int raid_map(struct dm_target *ti, struct bio *bio, union map_info *map_context) | |
593 | { | |
594 | struct raid_set *rs = ti->private; | |
595 | mddev_t *mddev = &rs->md; | |
596 | ||
597 | mddev->pers->make_request(mddev, bio); | |
598 | ||
599 | return DM_MAPIO_SUBMITTED; | |
600 | } | |
601 | ||
602 | static int raid_status(struct dm_target *ti, status_type_t type, | |
603 | char *result, unsigned maxlen) | |
604 | { | |
605 | struct raid_set *rs = ti->private; | |
606 | unsigned raid_param_cnt = 1; /* at least 1 for chunksize */ | |
607 | unsigned sz = 0; | |
608 | int i; | |
609 | sector_t sync; | |
610 | ||
611 | switch (type) { | |
612 | case STATUSTYPE_INFO: | |
613 | DMEMIT("%s %d ", rs->raid_type->name, rs->md.raid_disks); | |
614 | ||
615 | for (i = 0; i < rs->md.raid_disks; i++) { | |
616 | if (test_bit(Faulty, &rs->dev[i].rdev.flags)) | |
617 | DMEMIT("D"); | |
618 | else if (test_bit(In_sync, &rs->dev[i].rdev.flags)) | |
619 | DMEMIT("A"); | |
620 | else | |
621 | DMEMIT("a"); | |
622 | } | |
623 | ||
624 | if (test_bit(MD_RECOVERY_RUNNING, &rs->md.recovery)) | |
625 | sync = rs->md.curr_resync_completed; | |
626 | else | |
627 | sync = rs->md.recovery_cp; | |
628 | ||
629 | if (sync > rs->md.resync_max_sectors) | |
630 | sync = rs->md.resync_max_sectors; | |
631 | ||
632 | DMEMIT(" %llu/%llu", | |
633 | (unsigned long long) sync, | |
634 | (unsigned long long) rs->md.resync_max_sectors); | |
635 | ||
636 | break; | |
637 | case STATUSTYPE_TABLE: | |
638 | /* The string you would use to construct this array */ | |
46bed2b5 | 639 | for (i = 0; i < rs->md.raid_disks; i++) { |
13c87583 JB |
640 | if ((rs->print_flags & DMPF_REBUILD) && |
641 | rs->dev[i].data_dev && | |
9d09e663 | 642 | !test_bit(In_sync, &rs->dev[i].rdev.flags)) |
13c87583 | 643 | raid_param_cnt += 2; /* for rebuilds */ |
46bed2b5 JB |
644 | if (rs->dev[i].data_dev && |
645 | test_bit(WriteMostly, &rs->dev[i].rdev.flags)) | |
646 | raid_param_cnt += 2; | |
647 | } | |
9d09e663 | 648 | |
13c87583 | 649 | raid_param_cnt += (hweight64(rs->print_flags & ~DMPF_REBUILD) * 2); |
9d09e663 N |
650 | if (rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC)) |
651 | raid_param_cnt--; | |
652 | ||
653 | DMEMIT("%s %u %u", rs->raid_type->name, | |
654 | raid_param_cnt, rs->md.chunk_sectors); | |
655 | ||
656 | if ((rs->print_flags & DMPF_SYNC) && | |
657 | (rs->md.recovery_cp == MaxSector)) | |
658 | DMEMIT(" sync"); | |
659 | if (rs->print_flags & DMPF_NOSYNC) | |
660 | DMEMIT(" nosync"); | |
661 | ||
662 | for (i = 0; i < rs->md.raid_disks; i++) | |
13c87583 JB |
663 | if ((rs->print_flags & DMPF_REBUILD) && |
664 | rs->dev[i].data_dev && | |
9d09e663 N |
665 | !test_bit(In_sync, &rs->dev[i].rdev.flags)) |
666 | DMEMIT(" rebuild %u", i); | |
667 | ||
668 | if (rs->print_flags & DMPF_DAEMON_SLEEP) | |
669 | DMEMIT(" daemon_sleep %lu", | |
670 | rs->md.bitmap_info.daemon_sleep); | |
671 | ||
672 | if (rs->print_flags & DMPF_MIN_RECOVERY_RATE) | |
673 | DMEMIT(" min_recovery_rate %d", rs->md.sync_speed_min); | |
674 | ||
675 | if (rs->print_flags & DMPF_MAX_RECOVERY_RATE) | |
676 | DMEMIT(" max_recovery_rate %d", rs->md.sync_speed_max); | |
677 | ||
46bed2b5 JB |
678 | for (i = 0; i < rs->md.raid_disks; i++) |
679 | if (rs->dev[i].data_dev && | |
680 | test_bit(WriteMostly, &rs->dev[i].rdev.flags)) | |
681 | DMEMIT(" write_mostly %u", i); | |
682 | ||
9d09e663 N |
683 | if (rs->print_flags & DMPF_MAX_WRITE_BEHIND) |
684 | DMEMIT(" max_write_behind %lu", | |
685 | rs->md.bitmap_info.max_write_behind); | |
686 | ||
687 | if (rs->print_flags & DMPF_STRIPE_CACHE) { | |
688 | raid5_conf_t *conf = rs->md.private; | |
689 | ||
690 | /* convert from kiB to sectors */ | |
691 | DMEMIT(" stripe_cache %d", | |
692 | conf ? conf->max_nr_stripes * 2 : 0); | |
693 | } | |
694 | ||
c1084561 JB |
695 | if (rs->print_flags & DMPF_REGION_SIZE) |
696 | DMEMIT(" region_size %lu", | |
697 | rs->md.bitmap_info.chunksize >> 9); | |
698 | ||
9d09e663 N |
699 | DMEMIT(" %d", rs->md.raid_disks); |
700 | for (i = 0; i < rs->md.raid_disks; i++) { | |
701 | DMEMIT(" -"); /* metadata device */ | |
702 | ||
703 | if (rs->dev[i].data_dev) | |
704 | DMEMIT(" %s", rs->dev[i].data_dev->name); | |
705 | else | |
706 | DMEMIT(" -"); | |
707 | } | |
708 | } | |
709 | ||
710 | return 0; | |
711 | } | |
712 | ||
713 | static int raid_iterate_devices(struct dm_target *ti, iterate_devices_callout_fn fn, void *data) | |
714 | { | |
715 | struct raid_set *rs = ti->private; | |
716 | unsigned i; | |
717 | int ret = 0; | |
718 | ||
719 | for (i = 0; !ret && i < rs->md.raid_disks; i++) | |
720 | if (rs->dev[i].data_dev) | |
721 | ret = fn(ti, | |
722 | rs->dev[i].data_dev, | |
723 | 0, /* No offset on data devs */ | |
724 | rs->md.dev_sectors, | |
725 | data); | |
726 | ||
727 | return ret; | |
728 | } | |
729 | ||
730 | static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits) | |
731 | { | |
732 | struct raid_set *rs = ti->private; | |
733 | unsigned chunk_size = rs->md.chunk_sectors << 9; | |
734 | raid5_conf_t *conf = rs->md.private; | |
735 | ||
736 | blk_limits_io_min(limits, chunk_size); | |
737 | blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded)); | |
738 | } | |
739 | ||
740 | static void raid_presuspend(struct dm_target *ti) | |
741 | { | |
742 | struct raid_set *rs = ti->private; | |
743 | ||
744 | md_stop_writes(&rs->md); | |
745 | } | |
746 | ||
747 | static void raid_postsuspend(struct dm_target *ti) | |
748 | { | |
749 | struct raid_set *rs = ti->private; | |
750 | ||
751 | mddev_suspend(&rs->md); | |
752 | } | |
753 | ||
754 | static void raid_resume(struct dm_target *ti) | |
755 | { | |
756 | struct raid_set *rs = ti->private; | |
757 | ||
758 | mddev_resume(&rs->md); | |
759 | } | |
760 | ||
761 | static struct target_type raid_target = { | |
762 | .name = "raid", | |
763 | .version = {1, 0, 0}, | |
764 | .module = THIS_MODULE, | |
765 | .ctr = raid_ctr, | |
766 | .dtr = raid_dtr, | |
767 | .map = raid_map, | |
768 | .status = raid_status, | |
769 | .iterate_devices = raid_iterate_devices, | |
770 | .io_hints = raid_io_hints, | |
771 | .presuspend = raid_presuspend, | |
772 | .postsuspend = raid_postsuspend, | |
773 | .resume = raid_resume, | |
774 | }; | |
775 | ||
776 | static int __init dm_raid_init(void) | |
777 | { | |
778 | return dm_register_target(&raid_target); | |
779 | } | |
780 | ||
781 | static void __exit dm_raid_exit(void) | |
782 | { | |
783 | dm_unregister_target(&raid_target); | |
784 | } | |
785 | ||
786 | module_init(dm_raid_init); | |
787 | module_exit(dm_raid_exit); | |
788 | ||
789 | MODULE_DESCRIPTION(DM_NAME " raid4/5/6 target"); | |
790 | MODULE_ALIAS("dm-raid4"); | |
791 | MODULE_ALIAS("dm-raid5"); | |
792 | MODULE_ALIAS("dm-raid6"); | |
793 | MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>"); | |
794 | MODULE_LICENSE("GPL"); |