?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Add linux-next specific files for 20110421
[linux-2.6/next.git] / drivers / md / dm-raid.c
blobe5d8904fc8f647162d4a7d79150491797cfda6d2
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"
12 #include "dm.h"
13 #include "bitmap.h"
14
15 #define DM_MSG_PREFIX "raid"
16
17 /*
18 * If the MD doesn't support MD_SYNC_STATE_FORCED yet, then
19 * make it so the flag doesn't set anything.
20 */
21 #ifndef MD_SYNC_STATE_FORCED
22 #define MD_SYNC_STATE_FORCED 0
23 #endif
24
25 struct raid_dev {
26 /*
27 * Two DM devices, one to hold metadata and one to hold the
28 * actual data/parity. The reason for this is to not confuse
29 * ti->len and give more flexibility in altering size and
30 * characteristics.
31 *
32 * While it is possible for this device to be associated
33 * with a different physical device than the data_dev, it
34 * is intended for it to be the same.
35 * |--------- Physical Device ---------|
36 * |- meta_dev -|------ data_dev ------|
37 */
38 struct dm_dev *meta_dev;
39 struct dm_dev *data_dev;
40 struct mdk_rdev_s rdev;
41 };
42
43 /*
44 * Flags for rs->print_flags field.
45 */
46 #define DMPF_DAEMON_SLEEP 0x1
47 #define DMPF_MAX_WRITE_BEHIND 0x2
48 #define DMPF_SYNC 0x4
49 #define DMPF_NOSYNC 0x8
50 #define DMPF_STRIPE_CACHE 0x10
51 #define DMPF_MIN_RECOVERY_RATE 0x20
52 #define DMPF_MAX_RECOVERY_RATE 0x40
53
54 struct raid_set {
55 struct dm_target *ti;
56
57 uint64_t print_flags;
58
59 struct mddev_s md;
60 struct raid_type *raid_type;
61 struct dm_target_callbacks callbacks;
62
63 struct raid_dev dev[0];
64 };
65
66 /* Supported raid types and properties. */
67 static struct raid_type {
68 const char *name; /* RAID algorithm. */
69 const char *descr; /* Descriptor text for logging. */
70 const unsigned parity_devs; /* # of parity devices. */
71 const unsigned minimal_devs; /* minimal # of devices in set. */
72 const unsigned level; /* RAID level. */
73 const unsigned algorithm; /* RAID algorithm. */
74 } raid_types[] = {
75 {"raid4", "RAID4 (dedicated parity disk)", 1, 2, 5, ALGORITHM_PARITY_0},
76 {"raid5_la", "RAID5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
77 {"raid5_ra", "RAID5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
78 {"raid5_ls", "RAID5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
79 {"raid5_rs", "RAID5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
80 {"raid6_zr", "RAID6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
81 {"raid6_nr", "RAID6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
82 {"raid6_nc", "RAID6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE}
83 };
84
85 static struct raid_type *get_raid_type(char *name)
86 {
87 int i;
88
89 for (i = 0; i < ARRAY_SIZE(raid_types); i++)
90 if (!strcmp(raid_types[i].name, name))
91 return &raid_types[i];
92
93 return NULL;
94 }
95
96 static struct raid_set *context_alloc(struct dm_target *ti, struct raid_type *raid_type, unsigned raid_devs)
97 {
98 unsigned i;
99 struct raid_set *rs;
100 sector_t sectors_per_dev;
101
102 if (raid_devs <= raid_type->parity_devs) {
103 ti->error = "Insufficient number of devices";
104 return ERR_PTR(-EINVAL);
105 }
106
107 sectors_per_dev = ti->len;
108 if (sector_div(sectors_per_dev, (raid_devs - raid_type->parity_devs))) {
109 ti->error = "Target length not divisible by number of data devices";
110 return ERR_PTR(-EINVAL);
111 }
112
113 rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
114 if (!rs) {
115 ti->error = "Cannot allocate raid context";
116 return ERR_PTR(-ENOMEM);
117 }
118
119 mddev_init(&rs->md);
120
121 rs->ti = ti;
122 rs->raid_type = raid_type;
123 rs->md.raid_disks = raid_devs;
124 rs->md.level = raid_type->level;
125 rs->md.new_level = rs->md.level;
126 rs->md.dev_sectors = sectors_per_dev;
127 rs->md.layout = raid_type->algorithm;
128 rs->md.new_layout = rs->md.layout;
129 rs->md.delta_disks = 0;
130 rs->md.recovery_cp = 0;
131
132 for (i = 0; i < raid_devs; i++)
133 md_rdev_init(&rs->dev[i].rdev);
134
135 /*
136 * Remaining items to be initialized by further RAID params:
137 * rs->md.persistent
138 * rs->md.external
139 * rs->md.chunk_sectors
140 * rs->md.new_chunk_sectors
141 */
142
143 return rs;
144 }
145
146 static void context_free(struct raid_set *rs)
147 {
148 int i;
149
150 for (i = 0; i < rs->md.raid_disks; i++)
151 if (rs->dev[i].data_dev)
152 dm_put_device(rs->ti, rs->dev[i].data_dev);
153
154 kfree(rs);
155 }
156
157 /*
158 * For every device we have two words
159 * <meta_dev>: meta device name or '-' if missing
160 * <data_dev>: data device name or '-' if missing
161 *
162 * This code parses those words.
163 */
164 static int dev_parms(struct raid_set *rs, char **argv)
165 {
166 int i;
167 int rebuild = 0;
168 int metadata_available = 0;
169 int ret = 0;
170
171 for (i = 0; i < rs->md.raid_disks; i++, argv += 2) {
172 rs->dev[i].rdev.raid_disk = i;
173
174 rs->dev[i].meta_dev = NULL;
175 rs->dev[i].data_dev = NULL;
176
177 /*
178 * There are no offsets, since there is a separate device
179 * for data and metadata.
180 */
181 rs->dev[i].rdev.data_offset = 0;
182 rs->dev[i].rdev.mddev = &rs->md;
183
184 if (strcmp(argv[0], "-")) {
185 rs->ti->error = "Metadata devices not supported";
186 return -EINVAL;
187 }
188
189 if (!strcmp(argv[1], "-")) {
190 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
191 (!rs->dev[i].rdev.recovery_offset)) {
192 rs->ti->error = "Drive designated for rebuild not specified";
193 return -EINVAL;
194 }
195
196 continue;
197 }
198
199 ret = dm_get_device(rs->ti, argv[1],
200 dm_table_get_mode(rs->ti->table),
201 &rs->dev[i].data_dev);
202 if (ret) {
203 rs->ti->error = "RAID device lookup failure";
204 return ret;
205 }
206
207 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
208 list_add(&rs->dev[i].rdev.same_set, &rs->md.disks);
209 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
210 rebuild++;
211 }
212
213 if (metadata_available) {
214 rs->md.external = 0;
215 rs->md.persistent = 1;
216 rs->md.major_version = 2;
217 } else if (rebuild && !rs->md.recovery_cp) {
218 /*
219 * Without metadata, we will not be able to tell if the array
220 * is in-sync or not - we must assume it is not. Therefore,
221 * it is impossible to rebuild a drive.
222 *
223 * Even if there is metadata, the on-disk information may
224 * indicate that the array is not in-sync and it will then
225 * fail at that time.
226 *
227 * User could specify 'nosync' option if desperate.
228 */
229 DMERR("Unable to rebuild drive while array is not in-sync");
230 rs->ti->error = "RAID device lookup failure";
231 return -EINVAL;
232 }
233
234 return 0;
235 }
236
237 /*
238 * Possible arguments are...
239 * RAID456:
240 * <chunk_size> [optional_args]
241 *
242 * Optional args:
243 * [[no]sync] Force or prevent recovery of the entire array
244 * [rebuild <idx>] Rebuild the drive indicated by the index
245 * [daemon_sleep <ms>] Time between bitmap daemon work to clear bits
246 * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
247 * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
248 * [max_write_behind <sectors>] See '-write-behind=' (man mdadm)
249 * [stripe_cache <sectors>] Stripe cache size for higher RAIDs
250 */
251 static int parse_raid_params(struct raid_set *rs, char **argv,
252 unsigned num_raid_params)
253 {
254 unsigned i, rebuild_cnt = 0;
255 unsigned long value;
256 char *key;
257
258 /*
259 * First, parse the in-order required arguments
260 */
261 if ((strict_strtoul(argv[0], 10, &value) < 0) ||
262 !is_power_of_2(value) || (value < 8)) {
263 rs->ti->error = "Bad chunk size";
264 return -EINVAL;
265 }
266
267 rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
268 argv++;
269 num_raid_params--;
270
271 /*
272 * Second, parse the unordered optional arguments
273 */
274 for (i = 0; i < rs->md.raid_disks; i++)
275 set_bit(In_sync, &rs->dev[i].rdev.flags);
276
277 for (i = 0; i < num_raid_params; i++) {
278 if (!strcmp(argv[i], "nosync")) {
279 rs->md.recovery_cp = MaxSector;
280 rs->print_flags |= DMPF_NOSYNC;
281 rs->md.flags |= MD_SYNC_STATE_FORCED;
282 continue;
283 }
284 if (!strcmp(argv[i], "sync")) {
285 rs->md.recovery_cp = 0;
286 rs->print_flags |= DMPF_SYNC;
287 rs->md.flags |= MD_SYNC_STATE_FORCED;
288 continue;
289 }
290
291 /* The rest of the optional arguments come in key/value pairs */
292 if ((i + 1) >= num_raid_params) {
293 rs->ti->error = "Wrong number of raid parameters given";
294 return -EINVAL;
295 }
296
297 key = argv[i++];
298 if (strict_strtoul(argv[i], 10, &value) < 0) {
299 rs->ti->error = "Bad numerical argument given in raid params";
300 return -EINVAL;
301 }
302
303 if (!strcmp(key, "rebuild")) {
304 if (++rebuild_cnt > rs->raid_type->parity_devs) {
305 rs->ti->error = "Too many rebuild drives given";
306 return -EINVAL;
307 }
308 if (value > rs->md.raid_disks) {
309 rs->ti->error = "Invalid rebuild index given";
310 return -EINVAL;
311 }
312 clear_bit(In_sync, &rs->dev[value].rdev.flags);
313 rs->dev[value].rdev.recovery_offset = 0;
314 } else if (!strcmp(key, "max_write_behind")) {
315 rs->print_flags |= DMPF_MAX_WRITE_BEHIND;
316
317 /*
318 * In device-mapper, we specify things in sectors, but
319 * MD records this value in kB
320 */
321 value /= 2;
322 if (value > COUNTER_MAX) {
323 rs->ti->error = "Max write-behind limit out of range";
324 return -EINVAL;
325 }
326 rs->md.bitmap_info.max_write_behind = value;
327 } else if (!strcmp(key, "daemon_sleep")) {
328 rs->print_flags |= DMPF_DAEMON_SLEEP;
329 if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
330 rs->ti->error = "daemon sleep period out of range";
331 return -EINVAL;
332 }
333 rs->md.bitmap_info.daemon_sleep = value;
334 } else if (!strcmp(key, "stripe_cache")) {
335 rs->print_flags |= DMPF_STRIPE_CACHE;
336
337 /*
338 * In device-mapper, we specify things in sectors, but
339 * MD records this value in kB
340 */
341 value /= 2;
342
343 if (rs->raid_type->level < 5) {
344 rs->ti->error = "Inappropriate argument: stripe_cache";
345 return -EINVAL;
346 }
347 if (raid5_set_cache_size(&rs->md, (int)value)) {
348 rs->ti->error = "Bad stripe_cache size";
349 return -EINVAL;
350 }
351 } else if (!strcmp(key, "min_recovery_rate")) {
352 rs->print_flags |= DMPF_MIN_RECOVERY_RATE;
353 if (value > INT_MAX) {
354 rs->ti->error = "min_recovery_rate out of range";
355 return -EINVAL;
356 }
357 rs->md.sync_speed_min = (int)value;
358 } else if (!strcmp(key, "max_recovery_rate")) {
359 rs->print_flags |= DMPF_MAX_RECOVERY_RATE;
360 if (value > INT_MAX) {
361 rs->ti->error = "max_recovery_rate out of range";
362 return -EINVAL;
363 }
364 rs->md.sync_speed_max = (int)value;
365 } else {
366 DMERR("Unable to parse RAID parameter: %s", key);
367 rs->ti->error = "Unable to parse RAID parameters";
368 return -EINVAL;
369 }
370 }
371
372 /* Assume there are no metadata devices until the drives are parsed */
373 rs->md.persistent = 0;
374 rs->md.external = 1;
375
376 return 0;
377 }
378
379 static void do_table_event(struct work_struct *ws)
380 {
381 struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
382
383 dm_table_event(rs->ti->table);
384 }
385
386 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
387 {
388 struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
389
390 return md_raid5_congested(&rs->md, bits);
391 }
392
393 /*
394 * Construct a RAID4/5/6 mapping:
395 * Args:
396 * <raid_type> <#raid_params> <raid_params> \
397 * <#raid_devs> { <meta_dev1> <dev1> .. <meta_devN> <devN> }
398 *
399 * ** metadata devices are not supported yet, use '-' instead **
400 *
401 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
402 * details on possible <raid_params>.
403 */
404 static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv)
405 {
406 int ret;
407 struct raid_type *rt;
408 unsigned long num_raid_params, num_raid_devs;
409 struct raid_set *rs = NULL;
410
411 /* Must have at least <raid_type> <#raid_params> */
412 if (argc < 2) {
413 ti->error = "Too few arguments";
414 return -EINVAL;
415 }
416
417 /* raid type */
418 rt = get_raid_type(argv[0]);
419 if (!rt) {
420 ti->error = "Unrecognised raid_type";
421 return -EINVAL;
422 }
423 argc--;
424 argv++;
425
426 /* number of RAID parameters */
427 if (strict_strtoul(argv[0], 10, &num_raid_params) < 0) {
428 ti->error = "Cannot understand number of RAID parameters";
429 return -EINVAL;
430 }
431 argc--;
432 argv++;
433
434 /* Skip over RAID params for now and find out # of devices */
435 if (num_raid_params + 1 > argc) {
436 ti->error = "Arguments do not agree with counts given";
437 return -EINVAL;
438 }
439
440 if ((strict_strtoul(argv[num_raid_params], 10, &num_raid_devs) < 0) ||
441 (num_raid_devs >= INT_MAX)) {
442 ti->error = "Cannot understand number of raid devices";
443 return -EINVAL;
444 }
445
446 rs = context_alloc(ti, rt, (unsigned)num_raid_devs);
447 if (IS_ERR(rs))
448 return PTR_ERR(rs);
449
450 ret = parse_raid_params(rs, argv, (unsigned)num_raid_params);
451 if (ret)
452 goto bad;
453
454 ret = -EINVAL;
455
456 argc -= num_raid_params + 1; /* +1: we already have num_raid_devs */
457 argv += num_raid_params + 1;
458
459 if (argc != (num_raid_devs * 2)) {
460 ti->error = "Supplied RAID devices does not match the count given";
461 goto bad;
462 }
463
464 ret = dev_parms(rs, argv);
465 if (ret)
466 goto bad;
467
468 INIT_WORK(&rs->md.event_work, do_table_event);
469 ti->split_io = rs->md.chunk_sectors;
470 ti->private = rs;
471
472 mutex_lock(&rs->md.reconfig_mutex);
473 ret = md_run(&rs->md);
474 rs->md.in_sync = 0; /* Assume already marked dirty */
475 mutex_unlock(&rs->md.reconfig_mutex);
476
477 if (ret) {
478 ti->error = "Fail to run raid array";
479 goto bad;
480 }
481
482 rs->callbacks.congested_fn = raid_is_congested;
483 dm_table_add_target_callbacks(ti->table, &rs->callbacks);
484
485 return 0;
486
487 bad:
488 context_free(rs);
489
490 return ret;
491 }
492
493 static void raid_dtr(struct dm_target *ti)
494 {
495 struct raid_set *rs = ti->private;
496
497 list_del_init(&rs->callbacks.list);
498 md_stop(&rs->md);
499 context_free(rs);
500 }
501
502 static int raid_map(struct dm_target *ti, struct bio *bio, union map_info *map_context)
503 {
504 struct raid_set *rs = ti->private;
505 mddev_t *mddev = &rs->md;
506
507 mddev->pers->make_request(mddev, bio);
508
509 return DM_MAPIO_SUBMITTED;
510 }
511
512 static int raid_status(struct dm_target *ti, status_type_t type,
513 char *result, unsigned maxlen)
514 {
515 struct raid_set *rs = ti->private;
516 unsigned raid_param_cnt = 1; /* at least 1 for chunksize */
517 unsigned sz = 0;
518 int i;
519 sector_t sync;
520
521 switch (type) {
522 case STATUSTYPE_INFO:
523 DMEMIT("%s %d ", rs->raid_type->name, rs->md.raid_disks);
524
525 for (i = 0; i < rs->md.raid_disks; i++) {
526 if (test_bit(Faulty, &rs->dev[i].rdev.flags))
527 DMEMIT("D");
528 else if (test_bit(In_sync, &rs->dev[i].rdev.flags))
529 DMEMIT("A");
530 else
531 DMEMIT("a");
532 }
533
534 if (test_bit(MD_RECOVERY_RUNNING, &rs->md.recovery))
535 sync = rs->md.curr_resync_completed;
536 else
537 sync = rs->md.recovery_cp;
538
539 if (sync > rs->md.resync_max_sectors)
540 sync = rs->md.resync_max_sectors;
541
542 DMEMIT(" %llu/%llu",
543 (unsigned long long) sync,
544 (unsigned long long) rs->md.resync_max_sectors);
545
546 break;
547 case STATUSTYPE_TABLE:
548 /* The string you would use to construct this array */
549 for (i = 0; i < rs->md.raid_disks; i++)
550 if (rs->dev[i].data_dev &&
551 !test_bit(In_sync, &rs->dev[i].rdev.flags))
552 raid_param_cnt++; /* for rebuilds */
553
554 raid_param_cnt += (hweight64(rs->print_flags) * 2);
555 if (rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC))
556 raid_param_cnt--;
557
558 DMEMIT("%s %u %u", rs->raid_type->name,
559 raid_param_cnt, rs->md.chunk_sectors);
560
561 if ((rs->print_flags & DMPF_SYNC) &&
562 (rs->md.recovery_cp == MaxSector))
563 DMEMIT(" sync");
564 if (rs->print_flags & DMPF_NOSYNC)
565 DMEMIT(" nosync");
566
567 for (i = 0; i < rs->md.raid_disks; i++)
568 if (rs->dev[i].data_dev &&
569 !test_bit(In_sync, &rs->dev[i].rdev.flags))
570 DMEMIT(" rebuild %u", i);
571
572 if (rs->print_flags & DMPF_DAEMON_SLEEP)
573 DMEMIT(" daemon_sleep %lu",
574 rs->md.bitmap_info.daemon_sleep);
575
576 if (rs->print_flags & DMPF_MIN_RECOVERY_RATE)
577 DMEMIT(" min_recovery_rate %d", rs->md.sync_speed_min);
578
579 if (rs->print_flags & DMPF_MAX_RECOVERY_RATE)
580 DMEMIT(" max_recovery_rate %d", rs->md.sync_speed_max);
581
582 if (rs->print_flags & DMPF_MAX_WRITE_BEHIND)
583 DMEMIT(" max_write_behind %lu",
584 rs->md.bitmap_info.max_write_behind);
585
586 if (rs->print_flags & DMPF_STRIPE_CACHE) {
587 raid5_conf_t *conf = rs->md.private;
588
589 /* convert from kiB to sectors */
590 DMEMIT(" stripe_cache %d",
591 conf ? conf->max_nr_stripes * 2 : 0);
592 }
593
594 DMEMIT(" %d", rs->md.raid_disks);
595 for (i = 0; i < rs->md.raid_disks; i++) {
596 DMEMIT(" -"); /* metadata device */
597
598 if (rs->dev[i].data_dev)
599 DMEMIT(" %s", rs->dev[i].data_dev->name);
600 else
601 DMEMIT(" -");
602 }
603 }
604
605 return 0;
606 }
607
608 static int raid_iterate_devices(struct dm_target *ti, iterate_devices_callout_fn fn, void *data)
609 {
610 struct raid_set *rs = ti->private;
611 unsigned i;
612 int ret = 0;
613
614 for (i = 0; !ret && i < rs->md.raid_disks; i++)
615 if (rs->dev[i].data_dev)
616 ret = fn(ti,
617 rs->dev[i].data_dev,
618 0, /* No offset on data devs */
619 rs->md.dev_sectors,
620 data);
621
622 return ret;
623 }
624
625 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
626 {
627 struct raid_set *rs = ti->private;
628 unsigned chunk_size = rs->md.chunk_sectors << 9;
629 raid5_conf_t *conf = rs->md.private;
630
631 blk_limits_io_min(limits, chunk_size);
632 blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded));
633 }
634
635 static void raid_presuspend(struct dm_target *ti)
636 {
637 struct raid_set *rs = ti->private;
638
639 md_stop_writes(&rs->md);
640 }
641
642 static void raid_postsuspend(struct dm_target *ti)
643 {
644 struct raid_set *rs = ti->private;
645
646 mddev_suspend(&rs->md);
647 }
648
649 static void raid_resume(struct dm_target *ti)
650 {
651 struct raid_set *rs = ti->private;
652
653 mddev_resume(&rs->md);
654 }
655
656 static struct target_type raid_target = {
657 .name = "raid",
658 .version = {1, 0, 0},
659 .module = THIS_MODULE,
660 .ctr = raid_ctr,
661 .dtr = raid_dtr,
662 .map = raid_map,
663 .status = raid_status,
664 .iterate_devices = raid_iterate_devices,
665 .io_hints = raid_io_hints,
666 .presuspend = raid_presuspend,
667 .postsuspend = raid_postsuspend,
668 .resume = raid_resume,
669 };
670
671 static int __init dm_raid_init(void)
672 {
673 return dm_register_target(&raid_target);
674 }
675
676 static void __exit dm_raid_exit(void)
677 {
678 dm_unregister_target(&raid_target);
679 }
680
681 module_init(dm_raid_init);
682 module_exit(dm_raid_exit);
683
684 MODULE_DESCRIPTION(DM_NAME " raid4/5/6 target");
685 MODULE_ALIAS("dm-raid4");
686 MODULE_ALIAS("dm-raid5");
687 MODULE_ALIAS("dm-raid6");
688 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
689 MODULE_LICENSE("GPL");
linux-next