xref: /openbmc/linux/drivers/md/dm-raid.c (revision 6e2055a9)
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 #include <linux/module.h>
10 
11 #include "md.h"
12 #include "raid1.h"
13 #include "raid5.h"
14 #include "raid10.h"
15 #include "bitmap.h"
16 
17 #include <linux/device-mapper.h>
18 
19 #define DM_MSG_PREFIX "raid"
20 
21 /*
22  * The following flags are used by dm-raid.c to set up the array state.
23  * They must be cleared before md_run is called.
24  */
25 #define FirstUse 10             /* rdev flag */
26 
27 struct raid_dev {
28 	/*
29 	 * Two DM devices, one to hold metadata and one to hold the
30 	 * actual data/parity.  The reason for this is to not confuse
31 	 * ti->len and give more flexibility in altering size and
32 	 * characteristics.
33 	 *
34 	 * While it is possible for this device to be associated
35 	 * with a different physical device than the data_dev, it
36 	 * is intended for it to be the same.
37 	 *    |--------- Physical Device ---------|
38 	 *    |- meta_dev -|------ data_dev ------|
39 	 */
40 	struct dm_dev *meta_dev;
41 	struct dm_dev *data_dev;
42 	struct md_rdev rdev;
43 };
44 
45 /*
46  * Flags for rs->print_flags field.
47  */
48 #define DMPF_SYNC              0x1
49 #define DMPF_NOSYNC            0x2
50 #define DMPF_REBUILD           0x4
51 #define DMPF_DAEMON_SLEEP      0x8
52 #define DMPF_MIN_RECOVERY_RATE 0x10
53 #define DMPF_MAX_RECOVERY_RATE 0x20
54 #define DMPF_MAX_WRITE_BEHIND  0x40
55 #define DMPF_STRIPE_CACHE      0x80
56 #define DMPF_REGION_SIZE       0x100
57 #define DMPF_RAID10_COPIES     0x200
58 #define DMPF_RAID10_FORMAT     0x400
59 
60 struct raid_set {
61 	struct dm_target *ti;
62 
63 	uint32_t bitmap_loaded;
64 	uint32_t print_flags;
65 
66 	struct mddev md;
67 	struct raid_type *raid_type;
68 	struct dm_target_callbacks callbacks;
69 
70 	struct raid_dev dev[0];
71 };
72 
73 /* Supported raid types and properties. */
74 static struct raid_type {
75 	const char *name;		/* RAID algorithm. */
76 	const char *descr;		/* Descriptor text for logging. */
77 	const unsigned parity_devs;	/* # of parity devices. */
78 	const unsigned minimal_devs;	/* minimal # of devices in set. */
79 	const unsigned level;		/* RAID level. */
80 	const unsigned algorithm;	/* RAID algorithm. */
81 } raid_types[] = {
82 	{"raid1",    "RAID1 (mirroring)",               0, 2, 1, 0 /* NONE */},
83 	{"raid10",   "RAID10 (striped mirrors)",        0, 2, 10, UINT_MAX /* Varies */},
84 	{"raid4",    "RAID4 (dedicated parity disk)",	1, 2, 5, ALGORITHM_PARITY_0},
85 	{"raid5_la", "RAID5 (left asymmetric)",		1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
86 	{"raid5_ra", "RAID5 (right asymmetric)",	1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
87 	{"raid5_ls", "RAID5 (left symmetric)",		1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
88 	{"raid5_rs", "RAID5 (right symmetric)",		1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
89 	{"raid6_zr", "RAID6 (zero restart)",		2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
90 	{"raid6_nr", "RAID6 (N restart)",		2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
91 	{"raid6_nc", "RAID6 (N continue)",		2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE}
92 };
93 
94 static char *raid10_md_layout_to_format(int layout)
95 {
96 	/*
97 	 * Bit 16 and 17 stand for "offset" and "use_far_sets"
98 	 * Refer to MD's raid10.c for details
99 	 */
100 	if ((layout & 0x10000) && (layout & 0x20000))
101 		return "offset";
102 
103 	if ((layout & 0xFF) > 1)
104 		return "near";
105 
106 	return "far";
107 }
108 
109 static unsigned raid10_md_layout_to_copies(int layout)
110 {
111 	if ((layout & 0xFF) > 1)
112 		return layout & 0xFF;
113 	return (layout >> 8) & 0xFF;
114 }
115 
116 static int raid10_format_to_md_layout(char *format, unsigned copies)
117 {
118 	unsigned n = 1, f = 1;
119 
120 	if (!strcmp("near", format))
121 		n = copies;
122 	else
123 		f = copies;
124 
125 	if (!strcmp("offset", format))
126 		return 0x30000 | (f << 8) | n;
127 
128 	if (!strcmp("far", format))
129 		return 0x20000 | (f << 8) | n;
130 
131 	return (f << 8) | n;
132 }
133 
134 static struct raid_type *get_raid_type(char *name)
135 {
136 	int i;
137 
138 	for (i = 0; i < ARRAY_SIZE(raid_types); i++)
139 		if (!strcmp(raid_types[i].name, name))
140 			return &raid_types[i];
141 
142 	return NULL;
143 }
144 
145 static struct raid_set *context_alloc(struct dm_target *ti, struct raid_type *raid_type, unsigned raid_devs)
146 {
147 	unsigned i;
148 	struct raid_set *rs;
149 
150 	if (raid_devs <= raid_type->parity_devs) {
151 		ti->error = "Insufficient number of devices";
152 		return ERR_PTR(-EINVAL);
153 	}
154 
155 	rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
156 	if (!rs) {
157 		ti->error = "Cannot allocate raid context";
158 		return ERR_PTR(-ENOMEM);
159 	}
160 
161 	mddev_init(&rs->md);
162 
163 	rs->ti = ti;
164 	rs->raid_type = raid_type;
165 	rs->md.raid_disks = raid_devs;
166 	rs->md.level = raid_type->level;
167 	rs->md.new_level = rs->md.level;
168 	rs->md.layout = raid_type->algorithm;
169 	rs->md.new_layout = rs->md.layout;
170 	rs->md.delta_disks = 0;
171 	rs->md.recovery_cp = 0;
172 
173 	for (i = 0; i < raid_devs; i++)
174 		md_rdev_init(&rs->dev[i].rdev);
175 
176 	/*
177 	 * Remaining items to be initialized by further RAID params:
178 	 *  rs->md.persistent
179 	 *  rs->md.external
180 	 *  rs->md.chunk_sectors
181 	 *  rs->md.new_chunk_sectors
182 	 *  rs->md.dev_sectors
183 	 */
184 
185 	return rs;
186 }
187 
188 static void context_free(struct raid_set *rs)
189 {
190 	int i;
191 
192 	for (i = 0; i < rs->md.raid_disks; i++) {
193 		if (rs->dev[i].meta_dev)
194 			dm_put_device(rs->ti, rs->dev[i].meta_dev);
195 		md_rdev_clear(&rs->dev[i].rdev);
196 		if (rs->dev[i].data_dev)
197 			dm_put_device(rs->ti, rs->dev[i].data_dev);
198 	}
199 
200 	kfree(rs);
201 }
202 
203 /*
204  * For every device we have two words
205  *  <meta_dev>: meta device name or '-' if missing
206  *  <data_dev>: data device name or '-' if missing
207  *
208  * The following are permitted:
209  *    - -
210  *    - <data_dev>
211  *    <meta_dev> <data_dev>
212  *
213  * The following is not allowed:
214  *    <meta_dev> -
215  *
216  * This code parses those words.  If there is a failure,
217  * the caller must use context_free to unwind the operations.
218  */
219 static int dev_parms(struct raid_set *rs, char **argv)
220 {
221 	int i;
222 	int rebuild = 0;
223 	int metadata_available = 0;
224 	int ret = 0;
225 
226 	for (i = 0; i < rs->md.raid_disks; i++, argv += 2) {
227 		rs->dev[i].rdev.raid_disk = i;
228 
229 		rs->dev[i].meta_dev = NULL;
230 		rs->dev[i].data_dev = NULL;
231 
232 		/*
233 		 * There are no offsets, since there is a separate device
234 		 * for data and metadata.
235 		 */
236 		rs->dev[i].rdev.data_offset = 0;
237 		rs->dev[i].rdev.mddev = &rs->md;
238 
239 		if (strcmp(argv[0], "-")) {
240 			ret = dm_get_device(rs->ti, argv[0],
241 					    dm_table_get_mode(rs->ti->table),
242 					    &rs->dev[i].meta_dev);
243 			rs->ti->error = "RAID metadata device lookup failure";
244 			if (ret)
245 				return ret;
246 
247 			rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
248 			if (!rs->dev[i].rdev.sb_page)
249 				return -ENOMEM;
250 		}
251 
252 		if (!strcmp(argv[1], "-")) {
253 			if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
254 			    (!rs->dev[i].rdev.recovery_offset)) {
255 				rs->ti->error = "Drive designated for rebuild not specified";
256 				return -EINVAL;
257 			}
258 
259 			rs->ti->error = "No data device supplied with metadata device";
260 			if (rs->dev[i].meta_dev)
261 				return -EINVAL;
262 
263 			continue;
264 		}
265 
266 		ret = dm_get_device(rs->ti, argv[1],
267 				    dm_table_get_mode(rs->ti->table),
268 				    &rs->dev[i].data_dev);
269 		if (ret) {
270 			rs->ti->error = "RAID device lookup failure";
271 			return ret;
272 		}
273 
274 		if (rs->dev[i].meta_dev) {
275 			metadata_available = 1;
276 			rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
277 		}
278 		rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
279 		list_add(&rs->dev[i].rdev.same_set, &rs->md.disks);
280 		if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
281 			rebuild++;
282 	}
283 
284 	if (metadata_available) {
285 		rs->md.external = 0;
286 		rs->md.persistent = 1;
287 		rs->md.major_version = 2;
288 	} else if (rebuild && !rs->md.recovery_cp) {
289 		/*
290 		 * Without metadata, we will not be able to tell if the array
291 		 * is in-sync or not - we must assume it is not.  Therefore,
292 		 * it is impossible to rebuild a drive.
293 		 *
294 		 * Even if there is metadata, the on-disk information may
295 		 * indicate that the array is not in-sync and it will then
296 		 * fail at that time.
297 		 *
298 		 * User could specify 'nosync' option if desperate.
299 		 */
300 		DMERR("Unable to rebuild drive while array is not in-sync");
301 		rs->ti->error = "RAID device lookup failure";
302 		return -EINVAL;
303 	}
304 
305 	return 0;
306 }
307 
308 /*
309  * validate_region_size
310  * @rs
311  * @region_size:  region size in sectors.  If 0, pick a size (4MiB default).
312  *
313  * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
314  * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
315  *
316  * Returns: 0 on success, -EINVAL on failure.
317  */
318 static int validate_region_size(struct raid_set *rs, unsigned long region_size)
319 {
320 	unsigned long min_region_size = rs->ti->len / (1 << 21);
321 
322 	if (!region_size) {
323 		/*
324 		 * Choose a reasonable default.  All figures in sectors.
325 		 */
326 		if (min_region_size > (1 << 13)) {
327 			/* If not a power of 2, make it the next power of 2 */
328 			if (min_region_size & (min_region_size - 1))
329 				region_size = 1 << fls(region_size);
330 			DMINFO("Choosing default region size of %lu sectors",
331 			       region_size);
332 		} else {
333 			DMINFO("Choosing default region size of 4MiB");
334 			region_size = 1 << 13; /* sectors */
335 		}
336 	} else {
337 		/*
338 		 * Validate user-supplied value.
339 		 */
340 		if (region_size > rs->ti->len) {
341 			rs->ti->error = "Supplied region size is too large";
342 			return -EINVAL;
343 		}
344 
345 		if (region_size < min_region_size) {
346 			DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
347 			      region_size, min_region_size);
348 			rs->ti->error = "Supplied region size is too small";
349 			return -EINVAL;
350 		}
351 
352 		if (!is_power_of_2(region_size)) {
353 			rs->ti->error = "Region size is not a power of 2";
354 			return -EINVAL;
355 		}
356 
357 		if (region_size < rs->md.chunk_sectors) {
358 			rs->ti->error = "Region size is smaller than the chunk size";
359 			return -EINVAL;
360 		}
361 	}
362 
363 	/*
364 	 * Convert sectors to bytes.
365 	 */
366 	rs->md.bitmap_info.chunksize = (region_size << 9);
367 
368 	return 0;
369 }
370 
371 /*
372  * validate_raid_redundancy
373  * @rs
374  *
375  * Determine if there are enough devices in the array that haven't
376  * failed (or are being rebuilt) to form a usable array.
377  *
378  * Returns: 0 on success, -EINVAL on failure.
379  */
380 static int validate_raid_redundancy(struct raid_set *rs)
381 {
382 	unsigned i, rebuild_cnt = 0;
383 	unsigned rebuilds_per_group = 0, copies, d;
384 	unsigned group_size, last_group_start;
385 
386 	for (i = 0; i < rs->md.raid_disks; i++)
387 		if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
388 		    !rs->dev[i].rdev.sb_page)
389 			rebuild_cnt++;
390 
391 	switch (rs->raid_type->level) {
392 	case 1:
393 		if (rebuild_cnt >= rs->md.raid_disks)
394 			goto too_many;
395 		break;
396 	case 4:
397 	case 5:
398 	case 6:
399 		if (rebuild_cnt > rs->raid_type->parity_devs)
400 			goto too_many;
401 		break;
402 	case 10:
403 		copies = raid10_md_layout_to_copies(rs->md.layout);
404 		if (rebuild_cnt < copies)
405 			break;
406 
407 		/*
408 		 * It is possible to have a higher rebuild count for RAID10,
409 		 * as long as the failed devices occur in different mirror
410 		 * groups (i.e. different stripes).
411 		 *
412 		 * When checking "near" format, make sure no adjacent devices
413 		 * have failed beyond what can be handled.  In addition to the
414 		 * simple case where the number of devices is a multiple of the
415 		 * number of copies, we must also handle cases where the number
416 		 * of devices is not a multiple of the number of copies.
417 		 * E.g.    dev1 dev2 dev3 dev4 dev5
418 		 *          A    A    B    B    C
419 		 *          C    D    D    E    E
420 		 */
421 		if (!strcmp("near", raid10_md_layout_to_format(rs->md.layout))) {
422 			for (i = 0; i < rs->md.raid_disks * copies; i++) {
423 				if (!(i % copies))
424 					rebuilds_per_group = 0;
425 				d = i % rs->md.raid_disks;
426 				if ((!rs->dev[d].rdev.sb_page ||
427 				     !test_bit(In_sync, &rs->dev[d].rdev.flags)) &&
428 				    (++rebuilds_per_group >= copies))
429 					goto too_many;
430 			}
431 			break;
432 		}
433 
434 		/*
435 		 * When checking "far" and "offset" formats, we need to ensure
436 		 * that the device that holds its copy is not also dead or
437 		 * being rebuilt.  (Note that "far" and "offset" formats only
438 		 * support two copies right now.  These formats also only ever
439 		 * use the 'use_far_sets' variant.)
440 		 *
441 		 * This check is somewhat complicated by the need to account
442 		 * for arrays that are not a multiple of (far) copies.  This
443 		 * results in the need to treat the last (potentially larger)
444 		 * set differently.
445 		 */
446 		group_size = (rs->md.raid_disks / copies);
447 		last_group_start = (rs->md.raid_disks / group_size) - 1;
448 		last_group_start *= group_size;
449 		for (i = 0; i < rs->md.raid_disks; i++) {
450 			if (!(i % copies) && !(i > last_group_start))
451 				rebuilds_per_group = 0;
452 			if ((!rs->dev[i].rdev.sb_page ||
453 			     !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
454 			    (++rebuilds_per_group >= copies))
455 					goto too_many;
456 		}
457 		break;
458 	default:
459 		if (rebuild_cnt)
460 			return -EINVAL;
461 	}
462 
463 	return 0;
464 
465 too_many:
466 	return -EINVAL;
467 }
468 
469 /*
470  * Possible arguments are...
471  *	<chunk_size> [optional_args]
472  *
473  * Argument definitions
474  *    <chunk_size>			The number of sectors per disk that
475  *                                      will form the "stripe"
476  *    [[no]sync]			Force or prevent recovery of the
477  *                                      entire array
478  *    [rebuild <idx>]			Rebuild the drive indicated by the index
479  *    [daemon_sleep <ms>]		Time between bitmap daemon work to
480  *                                      clear bits
481  *    [min_recovery_rate <kB/sec/disk>]	Throttle RAID initialization
482  *    [max_recovery_rate <kB/sec/disk>]	Throttle RAID initialization
483  *    [write_mostly <idx>]		Indicate a write mostly drive via index
484  *    [max_write_behind <sectors>]	See '-write-behind=' (man mdadm)
485  *    [stripe_cache <sectors>]		Stripe cache size for higher RAIDs
486  *    [region_size <sectors>]           Defines granularity of bitmap
487  *
488  * RAID10-only options:
489  *    [raid10_copies <# copies>]        Number of copies.  (Default: 2)
490  *    [raid10_format <near|far|offset>] Layout algorithm.  (Default: near)
491  */
492 static int parse_raid_params(struct raid_set *rs, char **argv,
493 			     unsigned num_raid_params)
494 {
495 	char *raid10_format = "near";
496 	unsigned raid10_copies = 2;
497 	unsigned i;
498 	unsigned long value, region_size = 0;
499 	sector_t sectors_per_dev = rs->ti->len;
500 	sector_t max_io_len;
501 	char *key;
502 
503 	/*
504 	 * First, parse the in-order required arguments
505 	 * "chunk_size" is the only argument of this type.
506 	 */
507 	if ((kstrtoul(argv[0], 10, &value) < 0)) {
508 		rs->ti->error = "Bad chunk size";
509 		return -EINVAL;
510 	} else if (rs->raid_type->level == 1) {
511 		if (value)
512 			DMERR("Ignoring chunk size parameter for RAID 1");
513 		value = 0;
514 	} else if (!is_power_of_2(value)) {
515 		rs->ti->error = "Chunk size must be a power of 2";
516 		return -EINVAL;
517 	} else if (value < 8) {
518 		rs->ti->error = "Chunk size value is too small";
519 		return -EINVAL;
520 	}
521 
522 	rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
523 	argv++;
524 	num_raid_params--;
525 
526 	/*
527 	 * We set each individual device as In_sync with a completed
528 	 * 'recovery_offset'.  If there has been a device failure or
529 	 * replacement then one of the following cases applies:
530 	 *
531 	 *   1) User specifies 'rebuild'.
532 	 *      - Device is reset when param is read.
533 	 *   2) A new device is supplied.
534 	 *      - No matching superblock found, resets device.
535 	 *   3) Device failure was transient and returns on reload.
536 	 *      - Failure noticed, resets device for bitmap replay.
537 	 *   4) Device hadn't completed recovery after previous failure.
538 	 *      - Superblock is read and overrides recovery_offset.
539 	 *
540 	 * What is found in the superblocks of the devices is always
541 	 * authoritative, unless 'rebuild' or '[no]sync' was specified.
542 	 */
543 	for (i = 0; i < rs->md.raid_disks; i++) {
544 		set_bit(In_sync, &rs->dev[i].rdev.flags);
545 		rs->dev[i].rdev.recovery_offset = MaxSector;
546 	}
547 
548 	/*
549 	 * Second, parse the unordered optional arguments
550 	 */
551 	for (i = 0; i < num_raid_params; i++) {
552 		if (!strcasecmp(argv[i], "nosync")) {
553 			rs->md.recovery_cp = MaxSector;
554 			rs->print_flags |= DMPF_NOSYNC;
555 			continue;
556 		}
557 		if (!strcasecmp(argv[i], "sync")) {
558 			rs->md.recovery_cp = 0;
559 			rs->print_flags |= DMPF_SYNC;
560 			continue;
561 		}
562 
563 		/* The rest of the optional arguments come in key/value pairs */
564 		if ((i + 1) >= num_raid_params) {
565 			rs->ti->error = "Wrong number of raid parameters given";
566 			return -EINVAL;
567 		}
568 
569 		key = argv[i++];
570 
571 		/* Parameters that take a string value are checked here. */
572 		if (!strcasecmp(key, "raid10_format")) {
573 			if (rs->raid_type->level != 10) {
574 				rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
575 				return -EINVAL;
576 			}
577 			if (strcmp("near", argv[i]) &&
578 			    strcmp("far", argv[i]) &&
579 			    strcmp("offset", argv[i])) {
580 				rs->ti->error = "Invalid 'raid10_format' value given";
581 				return -EINVAL;
582 			}
583 			raid10_format = argv[i];
584 			rs->print_flags |= DMPF_RAID10_FORMAT;
585 			continue;
586 		}
587 
588 		if (kstrtoul(argv[i], 10, &value) < 0) {
589 			rs->ti->error = "Bad numerical argument given in raid params";
590 			return -EINVAL;
591 		}
592 
593 		/* Parameters that take a numeric value are checked here */
594 		if (!strcasecmp(key, "rebuild")) {
595 			if (value >= rs->md.raid_disks) {
596 				rs->ti->error = "Invalid rebuild index given";
597 				return -EINVAL;
598 			}
599 			clear_bit(In_sync, &rs->dev[value].rdev.flags);
600 			rs->dev[value].rdev.recovery_offset = 0;
601 			rs->print_flags |= DMPF_REBUILD;
602 		} else if (!strcasecmp(key, "write_mostly")) {
603 			if (rs->raid_type->level != 1) {
604 				rs->ti->error = "write_mostly option is only valid for RAID1";
605 				return -EINVAL;
606 			}
607 			if (value >= rs->md.raid_disks) {
608 				rs->ti->error = "Invalid write_mostly drive index given";
609 				return -EINVAL;
610 			}
611 			set_bit(WriteMostly, &rs->dev[value].rdev.flags);
612 		} else if (!strcasecmp(key, "max_write_behind")) {
613 			if (rs->raid_type->level != 1) {
614 				rs->ti->error = "max_write_behind option is only valid for RAID1";
615 				return -EINVAL;
616 			}
617 			rs->print_flags |= DMPF_MAX_WRITE_BEHIND;
618 
619 			/*
620 			 * In device-mapper, we specify things in sectors, but
621 			 * MD records this value in kB
622 			 */
623 			value /= 2;
624 			if (value > COUNTER_MAX) {
625 				rs->ti->error = "Max write-behind limit out of range";
626 				return -EINVAL;
627 			}
628 			rs->md.bitmap_info.max_write_behind = value;
629 		} else if (!strcasecmp(key, "daemon_sleep")) {
630 			rs->print_flags |= DMPF_DAEMON_SLEEP;
631 			if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
632 				rs->ti->error = "daemon sleep period out of range";
633 				return -EINVAL;
634 			}
635 			rs->md.bitmap_info.daemon_sleep = value;
636 		} else if (!strcasecmp(key, "stripe_cache")) {
637 			rs->print_flags |= DMPF_STRIPE_CACHE;
638 
639 			/*
640 			 * In device-mapper, we specify things in sectors, but
641 			 * MD records this value in kB
642 			 */
643 			value /= 2;
644 
645 			if ((rs->raid_type->level != 5) &&
646 			    (rs->raid_type->level != 6)) {
647 				rs->ti->error = "Inappropriate argument: stripe_cache";
648 				return -EINVAL;
649 			}
650 			if (raid5_set_cache_size(&rs->md, (int)value)) {
651 				rs->ti->error = "Bad stripe_cache size";
652 				return -EINVAL;
653 			}
654 		} else if (!strcasecmp(key, "min_recovery_rate")) {
655 			rs->print_flags |= DMPF_MIN_RECOVERY_RATE;
656 			if (value > INT_MAX) {
657 				rs->ti->error = "min_recovery_rate out of range";
658 				return -EINVAL;
659 			}
660 			rs->md.sync_speed_min = (int)value;
661 		} else if (!strcasecmp(key, "max_recovery_rate")) {
662 			rs->print_flags |= DMPF_MAX_RECOVERY_RATE;
663 			if (value > INT_MAX) {
664 				rs->ti->error = "max_recovery_rate out of range";
665 				return -EINVAL;
666 			}
667 			rs->md.sync_speed_max = (int)value;
668 		} else if (!strcasecmp(key, "region_size")) {
669 			rs->print_flags |= DMPF_REGION_SIZE;
670 			region_size = value;
671 		} else if (!strcasecmp(key, "raid10_copies") &&
672 			   (rs->raid_type->level == 10)) {
673 			if ((value < 2) || (value > 0xFF)) {
674 				rs->ti->error = "Bad value for 'raid10_copies'";
675 				return -EINVAL;
676 			}
677 			rs->print_flags |= DMPF_RAID10_COPIES;
678 			raid10_copies = value;
679 		} else {
680 			DMERR("Unable to parse RAID parameter: %s", key);
681 			rs->ti->error = "Unable to parse RAID parameters";
682 			return -EINVAL;
683 		}
684 	}
685 
686 	if (validate_region_size(rs, region_size))
687 		return -EINVAL;
688 
689 	if (rs->md.chunk_sectors)
690 		max_io_len = rs->md.chunk_sectors;
691 	else
692 		max_io_len = region_size;
693 
694 	if (dm_set_target_max_io_len(rs->ti, max_io_len))
695 		return -EINVAL;
696 
697 	if (rs->raid_type->level == 10) {
698 		if (raid10_copies > rs->md.raid_disks) {
699 			rs->ti->error = "Not enough devices to satisfy specification";
700 			return -EINVAL;
701 		}
702 
703 		/*
704 		 * If the format is not "near", we only support
705 		 * two copies at the moment.
706 		 */
707 		if (strcmp("near", raid10_format) && (raid10_copies > 2)) {
708 			rs->ti->error = "Too many copies for given RAID10 format.";
709 			return -EINVAL;
710 		}
711 
712 		/* (Len * #mirrors) / #devices */
713 		sectors_per_dev = rs->ti->len * raid10_copies;
714 		sector_div(sectors_per_dev, rs->md.raid_disks);
715 
716 		rs->md.layout = raid10_format_to_md_layout(raid10_format,
717 							   raid10_copies);
718 		rs->md.new_layout = rs->md.layout;
719 	} else if ((rs->raid_type->level > 1) &&
720 		   sector_div(sectors_per_dev,
721 			      (rs->md.raid_disks - rs->raid_type->parity_devs))) {
722 		rs->ti->error = "Target length not divisible by number of data devices";
723 		return -EINVAL;
724 	}
725 	rs->md.dev_sectors = sectors_per_dev;
726 
727 	/* Assume there are no metadata devices until the drives are parsed */
728 	rs->md.persistent = 0;
729 	rs->md.external = 1;
730 
731 	return 0;
732 }
733 
734 static void do_table_event(struct work_struct *ws)
735 {
736 	struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
737 
738 	dm_table_event(rs->ti->table);
739 }
740 
741 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
742 {
743 	struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
744 
745 	if (rs->raid_type->level == 1)
746 		return md_raid1_congested(&rs->md, bits);
747 
748 	if (rs->raid_type->level == 10)
749 		return md_raid10_congested(&rs->md, bits);
750 
751 	return md_raid5_congested(&rs->md, bits);
752 }
753 
754 /*
755  * This structure is never routinely used by userspace, unlike md superblocks.
756  * Devices with this superblock should only ever be accessed via device-mapper.
757  */
758 #define DM_RAID_MAGIC 0x64526D44
759 struct dm_raid_superblock {
760 	__le32 magic;		/* "DmRd" */
761 	__le32 features;	/* Used to indicate possible future changes */
762 
763 	__le32 num_devices;	/* Number of devices in this array. (Max 64) */
764 	__le32 array_position;	/* The position of this drive in the array */
765 
766 	__le64 events;		/* Incremented by md when superblock updated */
767 	__le64 failed_devices;	/* Bit field of devices to indicate failures */
768 
769 	/*
770 	 * This offset tracks the progress of the repair or replacement of
771 	 * an individual drive.
772 	 */
773 	__le64 disk_recovery_offset;
774 
775 	/*
776 	 * This offset tracks the progress of the initial array
777 	 * synchronisation/parity calculation.
778 	 */
779 	__le64 array_resync_offset;
780 
781 	/*
782 	 * RAID characteristics
783 	 */
784 	__le32 level;
785 	__le32 layout;
786 	__le32 stripe_sectors;
787 
788 	__u8 pad[452];		/* Round struct to 512 bytes. */
789 				/* Always set to 0 when writing. */
790 } __packed;
791 
792 static int read_disk_sb(struct md_rdev *rdev, int size)
793 {
794 	BUG_ON(!rdev->sb_page);
795 
796 	if (rdev->sb_loaded)
797 		return 0;
798 
799 	if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, 1)) {
800 		DMERR("Failed to read superblock of device at position %d",
801 		      rdev->raid_disk);
802 		md_error(rdev->mddev, rdev);
803 		return -EINVAL;
804 	}
805 
806 	rdev->sb_loaded = 1;
807 
808 	return 0;
809 }
810 
811 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
812 {
813 	int i;
814 	uint64_t failed_devices;
815 	struct dm_raid_superblock *sb;
816 	struct raid_set *rs = container_of(mddev, struct raid_set, md);
817 
818 	sb = page_address(rdev->sb_page);
819 	failed_devices = le64_to_cpu(sb->failed_devices);
820 
821 	for (i = 0; i < mddev->raid_disks; i++)
822 		if (!rs->dev[i].data_dev ||
823 		    test_bit(Faulty, &(rs->dev[i].rdev.flags)))
824 			failed_devices |= (1ULL << i);
825 
826 	memset(sb, 0, sizeof(*sb));
827 
828 	sb->magic = cpu_to_le32(DM_RAID_MAGIC);
829 	sb->features = cpu_to_le32(0);	/* No features yet */
830 
831 	sb->num_devices = cpu_to_le32(mddev->raid_disks);
832 	sb->array_position = cpu_to_le32(rdev->raid_disk);
833 
834 	sb->events = cpu_to_le64(mddev->events);
835 	sb->failed_devices = cpu_to_le64(failed_devices);
836 
837 	sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
838 	sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
839 
840 	sb->level = cpu_to_le32(mddev->level);
841 	sb->layout = cpu_to_le32(mddev->layout);
842 	sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
843 }
844 
845 /*
846  * super_load
847  *
848  * This function creates a superblock if one is not found on the device
849  * and will decide which superblock to use if there's a choice.
850  *
851  * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
852  */
853 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
854 {
855 	int ret;
856 	struct dm_raid_superblock *sb;
857 	struct dm_raid_superblock *refsb;
858 	uint64_t events_sb, events_refsb;
859 
860 	rdev->sb_start = 0;
861 	rdev->sb_size = sizeof(*sb);
862 
863 	ret = read_disk_sb(rdev, rdev->sb_size);
864 	if (ret)
865 		return ret;
866 
867 	sb = page_address(rdev->sb_page);
868 
869 	/*
870 	 * Two cases that we want to write new superblocks and rebuild:
871 	 * 1) New device (no matching magic number)
872 	 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
873 	 */
874 	if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
875 	    (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
876 		super_sync(rdev->mddev, rdev);
877 
878 		set_bit(FirstUse, &rdev->flags);
879 
880 		/* Force writing of superblocks to disk */
881 		set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags);
882 
883 		/* Any superblock is better than none, choose that if given */
884 		return refdev ? 0 : 1;
885 	}
886 
887 	if (!refdev)
888 		return 1;
889 
890 	events_sb = le64_to_cpu(sb->events);
891 
892 	refsb = page_address(refdev->sb_page);
893 	events_refsb = le64_to_cpu(refsb->events);
894 
895 	return (events_sb > events_refsb) ? 1 : 0;
896 }
897 
898 static int super_init_validation(struct mddev *mddev, struct md_rdev *rdev)
899 {
900 	int role;
901 	struct raid_set *rs = container_of(mddev, struct raid_set, md);
902 	uint64_t events_sb;
903 	uint64_t failed_devices;
904 	struct dm_raid_superblock *sb;
905 	uint32_t new_devs = 0;
906 	uint32_t rebuilds = 0;
907 	struct md_rdev *r;
908 	struct dm_raid_superblock *sb2;
909 
910 	sb = page_address(rdev->sb_page);
911 	events_sb = le64_to_cpu(sb->events);
912 	failed_devices = le64_to_cpu(sb->failed_devices);
913 
914 	/*
915 	 * Initialise to 1 if this is a new superblock.
916 	 */
917 	mddev->events = events_sb ? : 1;
918 
919 	/*
920 	 * Reshaping is not currently allowed
921 	 */
922 	if (le32_to_cpu(sb->level) != mddev->level) {
923 		DMERR("Reshaping arrays not yet supported. (RAID level change)");
924 		return -EINVAL;
925 	}
926 	if (le32_to_cpu(sb->layout) != mddev->layout) {
927 		DMERR("Reshaping arrays not yet supported. (RAID layout change)");
928 		DMERR("  0x%X vs 0x%X", le32_to_cpu(sb->layout), mddev->layout);
929 		DMERR("  Old layout: %s w/ %d copies",
930 		      raid10_md_layout_to_format(le32_to_cpu(sb->layout)),
931 		      raid10_md_layout_to_copies(le32_to_cpu(sb->layout)));
932 		DMERR("  New layout: %s w/ %d copies",
933 		      raid10_md_layout_to_format(mddev->layout),
934 		      raid10_md_layout_to_copies(mddev->layout));
935 		return -EINVAL;
936 	}
937 	if (le32_to_cpu(sb->stripe_sectors) != mddev->chunk_sectors) {
938 		DMERR("Reshaping arrays not yet supported. (stripe sectors change)");
939 		return -EINVAL;
940 	}
941 
942 	/* We can only change the number of devices in RAID1 right now */
943 	if ((rs->raid_type->level != 1) &&
944 	    (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) {
945 		DMERR("Reshaping arrays not yet supported. (device count change)");
946 		return -EINVAL;
947 	}
948 
949 	if (!(rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC)))
950 		mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
951 
952 	/*
953 	 * During load, we set FirstUse if a new superblock was written.
954 	 * There are two reasons we might not have a superblock:
955 	 * 1) The array is brand new - in which case, all of the
956 	 *    devices must have their In_sync bit set.  Also,
957 	 *    recovery_cp must be 0, unless forced.
958 	 * 2) This is a new device being added to an old array
959 	 *    and the new device needs to be rebuilt - in which
960 	 *    case the In_sync bit will /not/ be set and
961 	 *    recovery_cp must be MaxSector.
962 	 */
963 	rdev_for_each(r, mddev) {
964 		if (!test_bit(In_sync, &r->flags)) {
965 			DMINFO("Device %d specified for rebuild: "
966 			       "Clearing superblock", r->raid_disk);
967 			rebuilds++;
968 		} else if (test_bit(FirstUse, &r->flags))
969 			new_devs++;
970 	}
971 
972 	if (!rebuilds) {
973 		if (new_devs == mddev->raid_disks) {
974 			DMINFO("Superblocks created for new array");
975 			set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
976 		} else if (new_devs) {
977 			DMERR("New device injected "
978 			      "into existing array without 'rebuild' "
979 			      "parameter specified");
980 			return -EINVAL;
981 		}
982 	} else if (new_devs) {
983 		DMERR("'rebuild' devices cannot be "
984 		      "injected into an array with other first-time devices");
985 		return -EINVAL;
986 	} else if (mddev->recovery_cp != MaxSector) {
987 		DMERR("'rebuild' specified while array is not in-sync");
988 		return -EINVAL;
989 	}
990 
991 	/*
992 	 * Now we set the Faulty bit for those devices that are
993 	 * recorded in the superblock as failed.
994 	 */
995 	rdev_for_each(r, mddev) {
996 		if (!r->sb_page)
997 			continue;
998 		sb2 = page_address(r->sb_page);
999 		sb2->failed_devices = 0;
1000 
1001 		/*
1002 		 * Check for any device re-ordering.
1003 		 */
1004 		if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
1005 			role = le32_to_cpu(sb2->array_position);
1006 			if (role != r->raid_disk) {
1007 				if (rs->raid_type->level != 1) {
1008 					rs->ti->error = "Cannot change device "
1009 						"positions in RAID array";
1010 					return -EINVAL;
1011 				}
1012 				DMINFO("RAID1 device #%d now at position #%d",
1013 				       role, r->raid_disk);
1014 			}
1015 
1016 			/*
1017 			 * Partial recovery is performed on
1018 			 * returning failed devices.
1019 			 */
1020 			if (failed_devices & (1 << role))
1021 				set_bit(Faulty, &r->flags);
1022 		}
1023 	}
1024 
1025 	return 0;
1026 }
1027 
1028 static int super_validate(struct mddev *mddev, struct md_rdev *rdev)
1029 {
1030 	struct dm_raid_superblock *sb = page_address(rdev->sb_page);
1031 
1032 	/*
1033 	 * If mddev->events is not set, we know we have not yet initialized
1034 	 * the array.
1035 	 */
1036 	if (!mddev->events && super_init_validation(mddev, rdev))
1037 		return -EINVAL;
1038 
1039 	mddev->bitmap_info.offset = 4096 >> 9; /* Enable bitmap creation */
1040 	rdev->mddev->bitmap_info.default_offset = 4096 >> 9;
1041 	if (!test_bit(FirstUse, &rdev->flags)) {
1042 		rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
1043 		if (rdev->recovery_offset != MaxSector)
1044 			clear_bit(In_sync, &rdev->flags);
1045 	}
1046 
1047 	/*
1048 	 * If a device comes back, set it as not In_sync and no longer faulty.
1049 	 */
1050 	if (test_bit(Faulty, &rdev->flags)) {
1051 		clear_bit(Faulty, &rdev->flags);
1052 		clear_bit(In_sync, &rdev->flags);
1053 		rdev->saved_raid_disk = rdev->raid_disk;
1054 		rdev->recovery_offset = 0;
1055 	}
1056 
1057 	clear_bit(FirstUse, &rdev->flags);
1058 
1059 	return 0;
1060 }
1061 
1062 /*
1063  * Analyse superblocks and select the freshest.
1064  */
1065 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
1066 {
1067 	int ret;
1068 	struct raid_dev *dev;
1069 	struct md_rdev *rdev, *tmp, *freshest;
1070 	struct mddev *mddev = &rs->md;
1071 
1072 	freshest = NULL;
1073 	rdev_for_each_safe(rdev, tmp, mddev) {
1074 		/*
1075 		 * Skipping super_load due to DMPF_SYNC will cause
1076 		 * the array to undergo initialization again as
1077 		 * though it were new.  This is the intended effect
1078 		 * of the "sync" directive.
1079 		 *
1080 		 * When reshaping capability is added, we must ensure
1081 		 * that the "sync" directive is disallowed during the
1082 		 * reshape.
1083 		 */
1084 		if (rs->print_flags & DMPF_SYNC)
1085 			continue;
1086 
1087 		if (!rdev->meta_bdev)
1088 			continue;
1089 
1090 		ret = super_load(rdev, freshest);
1091 
1092 		switch (ret) {
1093 		case 1:
1094 			freshest = rdev;
1095 			break;
1096 		case 0:
1097 			break;
1098 		default:
1099 			dev = container_of(rdev, struct raid_dev, rdev);
1100 			if (dev->meta_dev)
1101 				dm_put_device(ti, dev->meta_dev);
1102 
1103 			dev->meta_dev = NULL;
1104 			rdev->meta_bdev = NULL;
1105 
1106 			if (rdev->sb_page)
1107 				put_page(rdev->sb_page);
1108 
1109 			rdev->sb_page = NULL;
1110 
1111 			rdev->sb_loaded = 0;
1112 
1113 			/*
1114 			 * We might be able to salvage the data device
1115 			 * even though the meta device has failed.  For
1116 			 * now, we behave as though '- -' had been
1117 			 * set for this device in the table.
1118 			 */
1119 			if (dev->data_dev)
1120 				dm_put_device(ti, dev->data_dev);
1121 
1122 			dev->data_dev = NULL;
1123 			rdev->bdev = NULL;
1124 
1125 			list_del(&rdev->same_set);
1126 		}
1127 	}
1128 
1129 	if (!freshest)
1130 		return 0;
1131 
1132 	if (validate_raid_redundancy(rs)) {
1133 		rs->ti->error = "Insufficient redundancy to activate array";
1134 		return -EINVAL;
1135 	}
1136 
1137 	/*
1138 	 * Validation of the freshest device provides the source of
1139 	 * validation for the remaining devices.
1140 	 */
1141 	ti->error = "Unable to assemble array: Invalid superblocks";
1142 	if (super_validate(mddev, freshest))
1143 		return -EINVAL;
1144 
1145 	rdev_for_each(rdev, mddev)
1146 		if ((rdev != freshest) && super_validate(mddev, rdev))
1147 			return -EINVAL;
1148 
1149 	return 0;
1150 }
1151 
1152 /*
1153  * Construct a RAID4/5/6 mapping:
1154  * Args:
1155  *	<raid_type> <#raid_params> <raid_params>		\
1156  *	<#raid_devs> { <meta_dev1> <dev1> .. <meta_devN> <devN> }
1157  *
1158  * <raid_params> varies by <raid_type>.  See 'parse_raid_params' for
1159  * details on possible <raid_params>.
1160  */
1161 static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv)
1162 {
1163 	int ret;
1164 	struct raid_type *rt;
1165 	unsigned long num_raid_params, num_raid_devs;
1166 	struct raid_set *rs = NULL;
1167 
1168 	/* Must have at least <raid_type> <#raid_params> */
1169 	if (argc < 2) {
1170 		ti->error = "Too few arguments";
1171 		return -EINVAL;
1172 	}
1173 
1174 	/* raid type */
1175 	rt = get_raid_type(argv[0]);
1176 	if (!rt) {
1177 		ti->error = "Unrecognised raid_type";
1178 		return -EINVAL;
1179 	}
1180 	argc--;
1181 	argv++;
1182 
1183 	/* number of RAID parameters */
1184 	if (kstrtoul(argv[0], 10, &num_raid_params) < 0) {
1185 		ti->error = "Cannot understand number of RAID parameters";
1186 		return -EINVAL;
1187 	}
1188 	argc--;
1189 	argv++;
1190 
1191 	/* Skip over RAID params for now and find out # of devices */
1192 	if (num_raid_params + 1 > argc) {
1193 		ti->error = "Arguments do not agree with counts given";
1194 		return -EINVAL;
1195 	}
1196 
1197 	if ((kstrtoul(argv[num_raid_params], 10, &num_raid_devs) < 0) ||
1198 	    (num_raid_devs >= INT_MAX)) {
1199 		ti->error = "Cannot understand number of raid devices";
1200 		return -EINVAL;
1201 	}
1202 
1203 	rs = context_alloc(ti, rt, (unsigned)num_raid_devs);
1204 	if (IS_ERR(rs))
1205 		return PTR_ERR(rs);
1206 
1207 	ret = parse_raid_params(rs, argv, (unsigned)num_raid_params);
1208 	if (ret)
1209 		goto bad;
1210 
1211 	ret = -EINVAL;
1212 
1213 	argc -= num_raid_params + 1; /* +1: we already have num_raid_devs */
1214 	argv += num_raid_params + 1;
1215 
1216 	if (argc != (num_raid_devs * 2)) {
1217 		ti->error = "Supplied RAID devices does not match the count given";
1218 		goto bad;
1219 	}
1220 
1221 	ret = dev_parms(rs, argv);
1222 	if (ret)
1223 		goto bad;
1224 
1225 	rs->md.sync_super = super_sync;
1226 	ret = analyse_superblocks(ti, rs);
1227 	if (ret)
1228 		goto bad;
1229 
1230 	INIT_WORK(&rs->md.event_work, do_table_event);
1231 	ti->private = rs;
1232 	ti->num_flush_bios = 1;
1233 
1234 	mutex_lock(&rs->md.reconfig_mutex);
1235 	ret = md_run(&rs->md);
1236 	rs->md.in_sync = 0; /* Assume already marked dirty */
1237 	mutex_unlock(&rs->md.reconfig_mutex);
1238 
1239 	if (ret) {
1240 		ti->error = "Fail to run raid array";
1241 		goto bad;
1242 	}
1243 
1244 	if (ti->len != rs->md.array_sectors) {
1245 		ti->error = "Array size does not match requested target length";
1246 		ret = -EINVAL;
1247 		goto size_mismatch;
1248 	}
1249 	rs->callbacks.congested_fn = raid_is_congested;
1250 	dm_table_add_target_callbacks(ti->table, &rs->callbacks);
1251 
1252 	mddev_suspend(&rs->md);
1253 	return 0;
1254 
1255 size_mismatch:
1256 	md_stop(&rs->md);
1257 bad:
1258 	context_free(rs);
1259 
1260 	return ret;
1261 }
1262 
1263 static void raid_dtr(struct dm_target *ti)
1264 {
1265 	struct raid_set *rs = ti->private;
1266 
1267 	list_del_init(&rs->callbacks.list);
1268 	md_stop(&rs->md);
1269 	context_free(rs);
1270 }
1271 
1272 static int raid_map(struct dm_target *ti, struct bio *bio)
1273 {
1274 	struct raid_set *rs = ti->private;
1275 	struct mddev *mddev = &rs->md;
1276 
1277 	mddev->pers->make_request(mddev, bio);
1278 
1279 	return DM_MAPIO_SUBMITTED;
1280 }
1281 
1282 static const char *decipher_sync_action(struct mddev *mddev)
1283 {
1284 	if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
1285 		return "frozen";
1286 
1287 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1288 	    (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
1289 		if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
1290 			return "reshape";
1291 
1292 		if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1293 			if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1294 				return "resync";
1295 			else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1296 				return "check";
1297 			return "repair";
1298 		}
1299 
1300 		if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
1301 			return "recover";
1302 	}
1303 
1304 	return "idle";
1305 }
1306 
1307 static void raid_status(struct dm_target *ti, status_type_t type,
1308 			unsigned status_flags, char *result, unsigned maxlen)
1309 {
1310 	struct raid_set *rs = ti->private;
1311 	unsigned raid_param_cnt = 1; /* at least 1 for chunksize */
1312 	unsigned sz = 0;
1313 	int i, array_in_sync = 0;
1314 	sector_t sync;
1315 
1316 	switch (type) {
1317 	case STATUSTYPE_INFO:
1318 		DMEMIT("%s %d ", rs->raid_type->name, rs->md.raid_disks);
1319 
1320 		if (test_bit(MD_RECOVERY_RUNNING, &rs->md.recovery))
1321 			sync = rs->md.curr_resync_completed;
1322 		else
1323 			sync = rs->md.recovery_cp;
1324 
1325 		if (sync >= rs->md.resync_max_sectors) {
1326 			/*
1327 			 * Sync complete.
1328 			 */
1329 			array_in_sync = 1;
1330 			sync = rs->md.resync_max_sectors;
1331 		} else if (test_bit(MD_RECOVERY_REQUESTED, &rs->md.recovery)) {
1332 			/*
1333 			 * If "check" or "repair" is occurring, the array has
1334 			 * undergone and initial sync and the health characters
1335 			 * should not be 'a' anymore.
1336 			 */
1337 			array_in_sync = 1;
1338 		} else {
1339 			/*
1340 			 * The array may be doing an initial sync, or it may
1341 			 * be rebuilding individual components.  If all the
1342 			 * devices are In_sync, then it is the array that is
1343 			 * being initialized.
1344 			 */
1345 			for (i = 0; i < rs->md.raid_disks; i++)
1346 				if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
1347 					array_in_sync = 1;
1348 		}
1349 
1350 		/*
1351 		 * Status characters:
1352 		 *  'D' = Dead/Failed device
1353 		 *  'a' = Alive but not in-sync
1354 		 *  'A' = Alive and in-sync
1355 		 */
1356 		for (i = 0; i < rs->md.raid_disks; i++) {
1357 			if (test_bit(Faulty, &rs->dev[i].rdev.flags))
1358 				DMEMIT("D");
1359 			else if (!array_in_sync ||
1360 				 !test_bit(In_sync, &rs->dev[i].rdev.flags))
1361 				DMEMIT("a");
1362 			else
1363 				DMEMIT("A");
1364 		}
1365 
1366 		/*
1367 		 * In-sync ratio:
1368 		 *  The in-sync ratio shows the progress of:
1369 		 *   - Initializing the array
1370 		 *   - Rebuilding a subset of devices of the array
1371 		 *  The user can distinguish between the two by referring
1372 		 *  to the status characters.
1373 		 */
1374 		DMEMIT(" %llu/%llu",
1375 		       (unsigned long long) sync,
1376 		       (unsigned long long) rs->md.resync_max_sectors);
1377 
1378 		/*
1379 		 * Sync action:
1380 		 *   See Documentation/device-mapper/dm-raid.c for
1381 		 *   information on each of these states.
1382 		 */
1383 		DMEMIT(" %s", decipher_sync_action(&rs->md));
1384 
1385 		/*
1386 		 * resync_mismatches/mismatch_cnt
1387 		 *   This field shows the number of discrepancies found when
1388 		 *   performing a "check" of the array.
1389 		 */
1390 		DMEMIT(" %llu",
1391 		       (strcmp(rs->md.last_sync_action, "check")) ? 0 :
1392 		       (unsigned long long)
1393 		       atomic64_read(&rs->md.resync_mismatches));
1394 		break;
1395 	case STATUSTYPE_TABLE:
1396 		/* The string you would use to construct this array */
1397 		for (i = 0; i < rs->md.raid_disks; i++) {
1398 			if ((rs->print_flags & DMPF_REBUILD) &&
1399 			    rs->dev[i].data_dev &&
1400 			    !test_bit(In_sync, &rs->dev[i].rdev.flags))
1401 				raid_param_cnt += 2; /* for rebuilds */
1402 			if (rs->dev[i].data_dev &&
1403 			    test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1404 				raid_param_cnt += 2;
1405 		}
1406 
1407 		raid_param_cnt += (hweight32(rs->print_flags & ~DMPF_REBUILD) * 2);
1408 		if (rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC))
1409 			raid_param_cnt--;
1410 
1411 		DMEMIT("%s %u %u", rs->raid_type->name,
1412 		       raid_param_cnt, rs->md.chunk_sectors);
1413 
1414 		if ((rs->print_flags & DMPF_SYNC) &&
1415 		    (rs->md.recovery_cp == MaxSector))
1416 			DMEMIT(" sync");
1417 		if (rs->print_flags & DMPF_NOSYNC)
1418 			DMEMIT(" nosync");
1419 
1420 		for (i = 0; i < rs->md.raid_disks; i++)
1421 			if ((rs->print_flags & DMPF_REBUILD) &&
1422 			    rs->dev[i].data_dev &&
1423 			    !test_bit(In_sync, &rs->dev[i].rdev.flags))
1424 				DMEMIT(" rebuild %u", i);
1425 
1426 		if (rs->print_flags & DMPF_DAEMON_SLEEP)
1427 			DMEMIT(" daemon_sleep %lu",
1428 			       rs->md.bitmap_info.daemon_sleep);
1429 
1430 		if (rs->print_flags & DMPF_MIN_RECOVERY_RATE)
1431 			DMEMIT(" min_recovery_rate %d", rs->md.sync_speed_min);
1432 
1433 		if (rs->print_flags & DMPF_MAX_RECOVERY_RATE)
1434 			DMEMIT(" max_recovery_rate %d", rs->md.sync_speed_max);
1435 
1436 		for (i = 0; i < rs->md.raid_disks; i++)
1437 			if (rs->dev[i].data_dev &&
1438 			    test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1439 				DMEMIT(" write_mostly %u", i);
1440 
1441 		if (rs->print_flags & DMPF_MAX_WRITE_BEHIND)
1442 			DMEMIT(" max_write_behind %lu",
1443 			       rs->md.bitmap_info.max_write_behind);
1444 
1445 		if (rs->print_flags & DMPF_STRIPE_CACHE) {
1446 			struct r5conf *conf = rs->md.private;
1447 
1448 			/* convert from kiB to sectors */
1449 			DMEMIT(" stripe_cache %d",
1450 			       conf ? conf->max_nr_stripes * 2 : 0);
1451 		}
1452 
1453 		if (rs->print_flags & DMPF_REGION_SIZE)
1454 			DMEMIT(" region_size %lu",
1455 			       rs->md.bitmap_info.chunksize >> 9);
1456 
1457 		if (rs->print_flags & DMPF_RAID10_COPIES)
1458 			DMEMIT(" raid10_copies %u",
1459 			       raid10_md_layout_to_copies(rs->md.layout));
1460 
1461 		if (rs->print_flags & DMPF_RAID10_FORMAT)
1462 			DMEMIT(" raid10_format %s",
1463 			       raid10_md_layout_to_format(rs->md.layout));
1464 
1465 		DMEMIT(" %d", rs->md.raid_disks);
1466 		for (i = 0; i < rs->md.raid_disks; i++) {
1467 			if (rs->dev[i].meta_dev)
1468 				DMEMIT(" %s", rs->dev[i].meta_dev->name);
1469 			else
1470 				DMEMIT(" -");
1471 
1472 			if (rs->dev[i].data_dev)
1473 				DMEMIT(" %s", rs->dev[i].data_dev->name);
1474 			else
1475 				DMEMIT(" -");
1476 		}
1477 	}
1478 }
1479 
1480 static int raid_message(struct dm_target *ti, unsigned argc, char **argv)
1481 {
1482 	struct raid_set *rs = ti->private;
1483 	struct mddev *mddev = &rs->md;
1484 
1485 	if (!strcasecmp(argv[0], "reshape")) {
1486 		DMERR("Reshape not supported.");
1487 		return -EINVAL;
1488 	}
1489 
1490 	if (!mddev->pers || !mddev->pers->sync_request)
1491 		return -EINVAL;
1492 
1493 	if (!strcasecmp(argv[0], "frozen"))
1494 		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1495 	else
1496 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1497 
1498 	if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
1499 		if (mddev->sync_thread) {
1500 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1501 			md_reap_sync_thread(mddev);
1502 		}
1503 	} else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1504 		   test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1505 		return -EBUSY;
1506 	else if (!strcasecmp(argv[0], "resync"))
1507 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1508 	else if (!strcasecmp(argv[0], "recover")) {
1509 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
1510 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1511 	} else {
1512 		if (!strcasecmp(argv[0], "check"))
1513 			set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1514 		else if (!!strcasecmp(argv[0], "repair"))
1515 			return -EINVAL;
1516 		set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1517 		set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1518 	}
1519 	if (mddev->ro == 2) {
1520 		/* A write to sync_action is enough to justify
1521 		 * canceling read-auto mode
1522 		 */
1523 		mddev->ro = 0;
1524 		if (!mddev->suspended)
1525 			md_wakeup_thread(mddev->sync_thread);
1526 	}
1527 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1528 	if (!mddev->suspended)
1529 		md_wakeup_thread(mddev->thread);
1530 
1531 	return 0;
1532 }
1533 
1534 static int raid_iterate_devices(struct dm_target *ti,
1535 				iterate_devices_callout_fn fn, void *data)
1536 {
1537 	struct raid_set *rs = ti->private;
1538 	unsigned i;
1539 	int ret = 0;
1540 
1541 	for (i = 0; !ret && i < rs->md.raid_disks; i++)
1542 		if (rs->dev[i].data_dev)
1543 			ret = fn(ti,
1544 				 rs->dev[i].data_dev,
1545 				 0, /* No offset on data devs */
1546 				 rs->md.dev_sectors,
1547 				 data);
1548 
1549 	return ret;
1550 }
1551 
1552 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
1553 {
1554 	struct raid_set *rs = ti->private;
1555 	unsigned chunk_size = rs->md.chunk_sectors << 9;
1556 	struct r5conf *conf = rs->md.private;
1557 
1558 	blk_limits_io_min(limits, chunk_size);
1559 	blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded));
1560 }
1561 
1562 static void raid_presuspend(struct dm_target *ti)
1563 {
1564 	struct raid_set *rs = ti->private;
1565 
1566 	md_stop_writes(&rs->md);
1567 }
1568 
1569 static void raid_postsuspend(struct dm_target *ti)
1570 {
1571 	struct raid_set *rs = ti->private;
1572 
1573 	mddev_suspend(&rs->md);
1574 }
1575 
1576 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
1577 {
1578 	int i;
1579 	uint64_t failed_devices, cleared_failed_devices = 0;
1580 	unsigned long flags;
1581 	struct dm_raid_superblock *sb;
1582 	struct md_rdev *r;
1583 
1584 	for (i = 0; i < rs->md.raid_disks; i++) {
1585 		r = &rs->dev[i].rdev;
1586 		if (test_bit(Faulty, &r->flags) && r->sb_page &&
1587 		    sync_page_io(r, 0, r->sb_size, r->sb_page, READ, 1)) {
1588 			DMINFO("Faulty %s device #%d has readable super block."
1589 			       "  Attempting to revive it.",
1590 			       rs->raid_type->name, i);
1591 
1592 			/*
1593 			 * Faulty bit may be set, but sometimes the array can
1594 			 * be suspended before the personalities can respond
1595 			 * by removing the device from the array (i.e. calling
1596 			 * 'hot_remove_disk').  If they haven't yet removed
1597 			 * the failed device, its 'raid_disk' number will be
1598 			 * '>= 0' - meaning we must call this function
1599 			 * ourselves.
1600 			 */
1601 			if ((r->raid_disk >= 0) &&
1602 			    (r->mddev->pers->hot_remove_disk(r->mddev, r) != 0))
1603 				/* Failed to revive this device, try next */
1604 				continue;
1605 
1606 			r->raid_disk = i;
1607 			r->saved_raid_disk = i;
1608 			flags = r->flags;
1609 			clear_bit(Faulty, &r->flags);
1610 			clear_bit(WriteErrorSeen, &r->flags);
1611 			clear_bit(In_sync, &r->flags);
1612 			if (r->mddev->pers->hot_add_disk(r->mddev, r)) {
1613 				r->raid_disk = -1;
1614 				r->saved_raid_disk = -1;
1615 				r->flags = flags;
1616 			} else {
1617 				r->recovery_offset = 0;
1618 				cleared_failed_devices |= 1 << i;
1619 			}
1620 		}
1621 	}
1622 	if (cleared_failed_devices) {
1623 		rdev_for_each(r, &rs->md) {
1624 			sb = page_address(r->sb_page);
1625 			failed_devices = le64_to_cpu(sb->failed_devices);
1626 			failed_devices &= ~cleared_failed_devices;
1627 			sb->failed_devices = cpu_to_le64(failed_devices);
1628 		}
1629 	}
1630 }
1631 
1632 static void raid_resume(struct dm_target *ti)
1633 {
1634 	struct raid_set *rs = ti->private;
1635 
1636 	set_bit(MD_CHANGE_DEVS, &rs->md.flags);
1637 	if (!rs->bitmap_loaded) {
1638 		bitmap_load(&rs->md);
1639 		rs->bitmap_loaded = 1;
1640 	} else {
1641 		/*
1642 		 * A secondary resume while the device is active.
1643 		 * Take this opportunity to check whether any failed
1644 		 * devices are reachable again.
1645 		 */
1646 		attempt_restore_of_faulty_devices(rs);
1647 	}
1648 
1649 	clear_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
1650 	mddev_resume(&rs->md);
1651 }
1652 
1653 static struct target_type raid_target = {
1654 	.name = "raid",
1655 	.version = {1, 5, 2},
1656 	.module = THIS_MODULE,
1657 	.ctr = raid_ctr,
1658 	.dtr = raid_dtr,
1659 	.map = raid_map,
1660 	.status = raid_status,
1661 	.message = raid_message,
1662 	.iterate_devices = raid_iterate_devices,
1663 	.io_hints = raid_io_hints,
1664 	.presuspend = raid_presuspend,
1665 	.postsuspend = raid_postsuspend,
1666 	.resume = raid_resume,
1667 };
1668 
1669 static int __init dm_raid_init(void)
1670 {
1671 	DMINFO("Loading target version %u.%u.%u",
1672 	       raid_target.version[0],
1673 	       raid_target.version[1],
1674 	       raid_target.version[2]);
1675 	return dm_register_target(&raid_target);
1676 }
1677 
1678 static void __exit dm_raid_exit(void)
1679 {
1680 	dm_unregister_target(&raid_target);
1681 }
1682 
1683 module_init(dm_raid_init);
1684 module_exit(dm_raid_exit);
1685 
1686 MODULE_DESCRIPTION(DM_NAME " raid4/5/6 target");
1687 MODULE_ALIAS("dm-raid1");
1688 MODULE_ALIAS("dm-raid10");
1689 MODULE_ALIAS("dm-raid4");
1690 MODULE_ALIAS("dm-raid5");
1691 MODULE_ALIAS("dm-raid6");
1692 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
1693 MODULE_LICENSE("GPL");
1694