[go: nahoru, domu]

1/*
2 * Copyright (C) 2010-2011 Neil Brown
3 * Copyright (C) 2010-2014 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
21static bool devices_handle_discard_safely = false;
22
23/*
24 * The following flags are used by dm-raid.c to set up the array state.
25 * They must be cleared before md_run is called.
26 */
27#define FirstUse 10             /* rdev flag */
28
29struct raid_dev {
30	/*
31	 * Two DM devices, one to hold metadata and one to hold the
32	 * actual data/parity.  The reason for this is to not confuse
33	 * ti->len and give more flexibility in altering size and
34	 * characteristics.
35	 *
36	 * While it is possible for this device to be associated
37	 * with a different physical device than the data_dev, it
38	 * is intended for it to be the same.
39	 *    |--------- Physical Device ---------|
40	 *    |- meta_dev -|------ data_dev ------|
41	 */
42	struct dm_dev *meta_dev;
43	struct dm_dev *data_dev;
44	struct md_rdev rdev;
45};
46
47/*
48 * Flags for rs->print_flags field.
49 */
50#define DMPF_SYNC              0x1
51#define DMPF_NOSYNC            0x2
52#define DMPF_REBUILD           0x4
53#define DMPF_DAEMON_SLEEP      0x8
54#define DMPF_MIN_RECOVERY_RATE 0x10
55#define DMPF_MAX_RECOVERY_RATE 0x20
56#define DMPF_MAX_WRITE_BEHIND  0x40
57#define DMPF_STRIPE_CACHE      0x80
58#define DMPF_REGION_SIZE       0x100
59#define DMPF_RAID10_COPIES     0x200
60#define DMPF_RAID10_FORMAT     0x400
61
62struct raid_set {
63	struct dm_target *ti;
64
65	uint32_t bitmap_loaded;
66	uint32_t print_flags;
67
68	struct mddev md;
69	struct raid_type *raid_type;
70	struct dm_target_callbacks callbacks;
71
72	struct raid_dev dev[0];
73};
74
75/* Supported raid types and properties. */
76static struct raid_type {
77	const char *name;		/* RAID algorithm. */
78	const char *descr;		/* Descriptor text for logging. */
79	const unsigned parity_devs;	/* # of parity devices. */
80	const unsigned minimal_devs;	/* minimal # of devices in set. */
81	const unsigned level;		/* RAID level. */
82	const unsigned algorithm;	/* RAID algorithm. */
83} raid_types[] = {
84	{"raid1",    "RAID1 (mirroring)",               0, 2, 1, 0 /* NONE */},
85	{"raid10",   "RAID10 (striped mirrors)",        0, 2, 10, UINT_MAX /* Varies */},
86	{"raid4",    "RAID4 (dedicated parity disk)",	1, 2, 5, ALGORITHM_PARITY_0},
87	{"raid5_la", "RAID5 (left asymmetric)",		1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
88	{"raid5_ra", "RAID5 (right asymmetric)",	1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
89	{"raid5_ls", "RAID5 (left symmetric)",		1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
90	{"raid5_rs", "RAID5 (right symmetric)",		1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
91	{"raid6_zr", "RAID6 (zero restart)",		2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
92	{"raid6_nr", "RAID6 (N restart)",		2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
93	{"raid6_nc", "RAID6 (N continue)",		2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE}
94};
95
96static char *raid10_md_layout_to_format(int layout)
97{
98	/*
99	 * Bit 16 and 17 stand for "offset" and "use_far_sets"
100	 * Refer to MD's raid10.c for details
101	 */
102	if ((layout & 0x10000) && (layout & 0x20000))
103		return "offset";
104
105	if ((layout & 0xFF) > 1)
106		return "near";
107
108	return "far";
109}
110
111static unsigned raid10_md_layout_to_copies(int layout)
112{
113	if ((layout & 0xFF) > 1)
114		return layout & 0xFF;
115	return (layout >> 8) & 0xFF;
116}
117
118static int raid10_format_to_md_layout(char *format, unsigned copies)
119{
120	unsigned n = 1, f = 1;
121
122	if (!strcmp("near", format))
123		n = copies;
124	else
125		f = copies;
126
127	if (!strcmp("offset", format))
128		return 0x30000 | (f << 8) | n;
129
130	if (!strcmp("far", format))
131		return 0x20000 | (f << 8) | n;
132
133	return (f << 8) | n;
134}
135
136static struct raid_type *get_raid_type(char *name)
137{
138	int i;
139
140	for (i = 0; i < ARRAY_SIZE(raid_types); i++)
141		if (!strcmp(raid_types[i].name, name))
142			return &raid_types[i];
143
144	return NULL;
145}
146
147static struct raid_set *context_alloc(struct dm_target *ti, struct raid_type *raid_type, unsigned raid_devs)
148{
149	unsigned i;
150	struct raid_set *rs;
151
152	if (raid_devs <= raid_type->parity_devs) {
153		ti->error = "Insufficient number of devices";
154		return ERR_PTR(-EINVAL);
155	}
156
157	rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
158	if (!rs) {
159		ti->error = "Cannot allocate raid context";
160		return ERR_PTR(-ENOMEM);
161	}
162
163	mddev_init(&rs->md);
164
165	rs->ti = ti;
166	rs->raid_type = raid_type;
167	rs->md.raid_disks = raid_devs;
168	rs->md.level = raid_type->level;
169	rs->md.new_level = rs->md.level;
170	rs->md.layout = raid_type->algorithm;
171	rs->md.new_layout = rs->md.layout;
172	rs->md.delta_disks = 0;
173	rs->md.recovery_cp = 0;
174
175	for (i = 0; i < raid_devs; i++)
176		md_rdev_init(&rs->dev[i].rdev);
177
178	/*
179	 * Remaining items to be initialized by further RAID params:
180	 *  rs->md.persistent
181	 *  rs->md.external
182	 *  rs->md.chunk_sectors
183	 *  rs->md.new_chunk_sectors
184	 *  rs->md.dev_sectors
185	 */
186
187	return rs;
188}
189
190static void context_free(struct raid_set *rs)
191{
192	int i;
193
194	for (i = 0; i < rs->md.raid_disks; i++) {
195		if (rs->dev[i].meta_dev)
196			dm_put_device(rs->ti, rs->dev[i].meta_dev);
197		md_rdev_clear(&rs->dev[i].rdev);
198		if (rs->dev[i].data_dev)
199			dm_put_device(rs->ti, rs->dev[i].data_dev);
200	}
201
202	kfree(rs);
203}
204
205/*
206 * For every device we have two words
207 *  <meta_dev>: meta device name or '-' if missing
208 *  <data_dev>: data device name or '-' if missing
209 *
210 * The following are permitted:
211 *    - -
212 *    - <data_dev>
213 *    <meta_dev> <data_dev>
214 *
215 * The following is not allowed:
216 *    <meta_dev> -
217 *
218 * This code parses those words.  If there is a failure,
219 * the caller must use context_free to unwind the operations.
220 */
221static int dev_parms(struct raid_set *rs, char **argv)
222{
223	int i;
224	int rebuild = 0;
225	int metadata_available = 0;
226	int ret = 0;
227
228	for (i = 0; i < rs->md.raid_disks; i++, argv += 2) {
229		rs->dev[i].rdev.raid_disk = i;
230
231		rs->dev[i].meta_dev = NULL;
232		rs->dev[i].data_dev = NULL;
233
234		/*
235		 * There are no offsets, since there is a separate device
236		 * for data and metadata.
237		 */
238		rs->dev[i].rdev.data_offset = 0;
239		rs->dev[i].rdev.mddev = &rs->md;
240
241		if (strcmp(argv[0], "-")) {
242			ret = dm_get_device(rs->ti, argv[0],
243					    dm_table_get_mode(rs->ti->table),
244					    &rs->dev[i].meta_dev);
245			rs->ti->error = "RAID metadata device lookup failure";
246			if (ret)
247				return ret;
248
249			rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
250			if (!rs->dev[i].rdev.sb_page)
251				return -ENOMEM;
252		}
253
254		if (!strcmp(argv[1], "-")) {
255			if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
256			    (!rs->dev[i].rdev.recovery_offset)) {
257				rs->ti->error = "Drive designated for rebuild not specified";
258				return -EINVAL;
259			}
260
261			rs->ti->error = "No data device supplied with metadata device";
262			if (rs->dev[i].meta_dev)
263				return -EINVAL;
264
265			continue;
266		}
267
268		ret = dm_get_device(rs->ti, argv[1],
269				    dm_table_get_mode(rs->ti->table),
270				    &rs->dev[i].data_dev);
271		if (ret) {
272			rs->ti->error = "RAID device lookup failure";
273			return ret;
274		}
275
276		if (rs->dev[i].meta_dev) {
277			metadata_available = 1;
278			rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
279		}
280		rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
281		list_add(&rs->dev[i].rdev.same_set, &rs->md.disks);
282		if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
283			rebuild++;
284	}
285
286	if (metadata_available) {
287		rs->md.external = 0;
288		rs->md.persistent = 1;
289		rs->md.major_version = 2;
290	} else if (rebuild && !rs->md.recovery_cp) {
291		/*
292		 * Without metadata, we will not be able to tell if the array
293		 * is in-sync or not - we must assume it is not.  Therefore,
294		 * it is impossible to rebuild a drive.
295		 *
296		 * Even if there is metadata, the on-disk information may
297		 * indicate that the array is not in-sync and it will then
298		 * fail at that time.
299		 *
300		 * User could specify 'nosync' option if desperate.
301		 */
302		DMERR("Unable to rebuild drive while array is not in-sync");
303		rs->ti->error = "RAID device lookup failure";
304		return -EINVAL;
305	}
306
307	return 0;
308}
309
310/*
311 * validate_region_size
312 * @rs
313 * @region_size:  region size in sectors.  If 0, pick a size (4MiB default).
314 *
315 * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
316 * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
317 *
318 * Returns: 0 on success, -EINVAL on failure.
319 */
320static int validate_region_size(struct raid_set *rs, unsigned long region_size)
321{
322	unsigned long min_region_size = rs->ti->len / (1 << 21);
323
324	if (!region_size) {
325		/*
326		 * Choose a reasonable default.  All figures in sectors.
327		 */
328		if (min_region_size > (1 << 13)) {
329			/* If not a power of 2, make it the next power of 2 */
330			if (min_region_size & (min_region_size - 1))
331				region_size = 1 << fls(region_size);
332			DMINFO("Choosing default region size of %lu sectors",
333			       region_size);
334		} else {
335			DMINFO("Choosing default region size of 4MiB");
336			region_size = 1 << 13; /* sectors */
337		}
338	} else {
339		/*
340		 * Validate user-supplied value.
341		 */
342		if (region_size > rs->ti->len) {
343			rs->ti->error = "Supplied region size is too large";
344			return -EINVAL;
345		}
346
347		if (region_size < min_region_size) {
348			DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
349			      region_size, min_region_size);
350			rs->ti->error = "Supplied region size is too small";
351			return -EINVAL;
352		}
353
354		if (!is_power_of_2(region_size)) {
355			rs->ti->error = "Region size is not a power of 2";
356			return -EINVAL;
357		}
358
359		if (region_size < rs->md.chunk_sectors) {
360			rs->ti->error = "Region size is smaller than the chunk size";
361			return -EINVAL;
362		}
363	}
364
365	/*
366	 * Convert sectors to bytes.
367	 */
368	rs->md.bitmap_info.chunksize = (region_size << 9);
369
370	return 0;
371}
372
373/*
374 * validate_raid_redundancy
375 * @rs
376 *
377 * Determine if there are enough devices in the array that haven't
378 * failed (or are being rebuilt) to form a usable array.
379 *
380 * Returns: 0 on success, -EINVAL on failure.
381 */
382static int validate_raid_redundancy(struct raid_set *rs)
383{
384	unsigned i, rebuild_cnt = 0;
385	unsigned rebuilds_per_group = 0, copies, d;
386	unsigned group_size, last_group_start;
387
388	for (i = 0; i < rs->md.raid_disks; i++)
389		if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
390		    !rs->dev[i].rdev.sb_page)
391			rebuild_cnt++;
392
393	switch (rs->raid_type->level) {
394	case 1:
395		if (rebuild_cnt >= rs->md.raid_disks)
396			goto too_many;
397		break;
398	case 4:
399	case 5:
400	case 6:
401		if (rebuild_cnt > rs->raid_type->parity_devs)
402			goto too_many;
403		break;
404	case 10:
405		copies = raid10_md_layout_to_copies(rs->md.layout);
406		if (rebuild_cnt < copies)
407			break;
408
409		/*
410		 * It is possible to have a higher rebuild count for RAID10,
411		 * as long as the failed devices occur in different mirror
412		 * groups (i.e. different stripes).
413		 *
414		 * When checking "near" format, make sure no adjacent devices
415		 * have failed beyond what can be handled.  In addition to the
416		 * simple case where the number of devices is a multiple of the
417		 * number of copies, we must also handle cases where the number
418		 * of devices is not a multiple of the number of copies.
419		 * E.g.    dev1 dev2 dev3 dev4 dev5
420		 *          A    A    B    B    C
421		 *          C    D    D    E    E
422		 */
423		if (!strcmp("near", raid10_md_layout_to_format(rs->md.layout))) {
424			for (i = 0; i < rs->md.raid_disks * copies; i++) {
425				if (!(i % copies))
426					rebuilds_per_group = 0;
427				d = i % rs->md.raid_disks;
428				if ((!rs->dev[d].rdev.sb_page ||
429				     !test_bit(In_sync, &rs->dev[d].rdev.flags)) &&
430				    (++rebuilds_per_group >= copies))
431					goto too_many;
432			}
433			break;
434		}
435
436		/*
437		 * When checking "far" and "offset" formats, we need to ensure
438		 * that the device that holds its copy is not also dead or
439		 * being rebuilt.  (Note that "far" and "offset" formats only
440		 * support two copies right now.  These formats also only ever
441		 * use the 'use_far_sets' variant.)
442		 *
443		 * This check is somewhat complicated by the need to account
444		 * for arrays that are not a multiple of (far) copies.  This
445		 * results in the need to treat the last (potentially larger)
446		 * set differently.
447		 */
448		group_size = (rs->md.raid_disks / copies);
449		last_group_start = (rs->md.raid_disks / group_size) - 1;
450		last_group_start *= group_size;
451		for (i = 0; i < rs->md.raid_disks; i++) {
452			if (!(i % copies) && !(i > last_group_start))
453				rebuilds_per_group = 0;
454			if ((!rs->dev[i].rdev.sb_page ||
455			     !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
456			    (++rebuilds_per_group >= copies))
457					goto too_many;
458		}
459		break;
460	default:
461		if (rebuild_cnt)
462			return -EINVAL;
463	}
464
465	return 0;
466
467too_many:
468	return -EINVAL;
469}
470
471/*
472 * Possible arguments are...
473 *	<chunk_size> [optional_args]
474 *
475 * Argument definitions
476 *    <chunk_size>			The number of sectors per disk that
477 *                                      will form the "stripe"
478 *    [[no]sync]			Force or prevent recovery of the
479 *                                      entire array
480 *    [devices_handle_discard_safely]	Allow discards on RAID4/5/6; useful if RAID
481 *					member device(s) properly support TRIM/UNMAP
482 *    [rebuild <idx>]			Rebuild the drive indicated by the index
483 *    [daemon_sleep <ms>]		Time between bitmap daemon work to
484 *                                      clear bits
485 *    [min_recovery_rate <kB/sec/disk>]	Throttle RAID initialization
486 *    [max_recovery_rate <kB/sec/disk>]	Throttle RAID initialization
487 *    [write_mostly <idx>]		Indicate a write mostly drive via index
488 *    [max_write_behind <sectors>]	See '-write-behind=' (man mdadm)
489 *    [stripe_cache <sectors>]		Stripe cache size for higher RAIDs
490 *    [region_size <sectors>]           Defines granularity of bitmap
491 *
492 * RAID10-only options:
493 *    [raid10_copies <# copies>]        Number of copies.  (Default: 2)
494 *    [raid10_format <near|far|offset>] Layout algorithm.  (Default: near)
495 */
496static int parse_raid_params(struct raid_set *rs, char **argv,
497			     unsigned num_raid_params)
498{
499	char *raid10_format = "near";
500	unsigned raid10_copies = 2;
501	unsigned i;
502	unsigned long value, region_size = 0;
503	sector_t sectors_per_dev = rs->ti->len;
504	sector_t max_io_len;
505	char *key;
506
507	/*
508	 * First, parse the in-order required arguments
509	 * "chunk_size" is the only argument of this type.
510	 */
511	if ((kstrtoul(argv[0], 10, &value) < 0)) {
512		rs->ti->error = "Bad chunk size";
513		return -EINVAL;
514	} else if (rs->raid_type->level == 1) {
515		if (value)
516			DMERR("Ignoring chunk size parameter for RAID 1");
517		value = 0;
518	} else if (!is_power_of_2(value)) {
519		rs->ti->error = "Chunk size must be a power of 2";
520		return -EINVAL;
521	} else if (value < 8) {
522		rs->ti->error = "Chunk size value is too small";
523		return -EINVAL;
524	}
525
526	rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
527	argv++;
528	num_raid_params--;
529
530	/*
531	 * We set each individual device as In_sync with a completed
532	 * 'recovery_offset'.  If there has been a device failure or
533	 * replacement then one of the following cases applies:
534	 *
535	 *   1) User specifies 'rebuild'.
536	 *      - Device is reset when param is read.
537	 *   2) A new device is supplied.
538	 *      - No matching superblock found, resets device.
539	 *   3) Device failure was transient and returns on reload.
540	 *      - Failure noticed, resets device for bitmap replay.
541	 *   4) Device hadn't completed recovery after previous failure.
542	 *      - Superblock is read and overrides recovery_offset.
543	 *
544	 * What is found in the superblocks of the devices is always
545	 * authoritative, unless 'rebuild' or '[no]sync' was specified.
546	 */
547	for (i = 0; i < rs->md.raid_disks; i++) {
548		set_bit(In_sync, &rs->dev[i].rdev.flags);
549		rs->dev[i].rdev.recovery_offset = MaxSector;
550	}
551
552	/*
553	 * Second, parse the unordered optional arguments
554	 */
555	for (i = 0; i < num_raid_params; i++) {
556		if (!strcasecmp(argv[i], "nosync")) {
557			rs->md.recovery_cp = MaxSector;
558			rs->print_flags |= DMPF_NOSYNC;
559			continue;
560		}
561		if (!strcasecmp(argv[i], "sync")) {
562			rs->md.recovery_cp = 0;
563			rs->print_flags |= DMPF_SYNC;
564			continue;
565		}
566
567		/* The rest of the optional arguments come in key/value pairs */
568		if ((i + 1) >= num_raid_params) {
569			rs->ti->error = "Wrong number of raid parameters given";
570			return -EINVAL;
571		}
572
573		key = argv[i++];
574
575		/* Parameters that take a string value are checked here. */
576		if (!strcasecmp(key, "raid10_format")) {
577			if (rs->raid_type->level != 10) {
578				rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
579				return -EINVAL;
580			}
581			if (strcmp("near", argv[i]) &&
582			    strcmp("far", argv[i]) &&
583			    strcmp("offset", argv[i])) {
584				rs->ti->error = "Invalid 'raid10_format' value given";
585				return -EINVAL;
586			}
587			raid10_format = argv[i];
588			rs->print_flags |= DMPF_RAID10_FORMAT;
589			continue;
590		}
591
592		if (kstrtoul(argv[i], 10, &value) < 0) {
593			rs->ti->error = "Bad numerical argument given in raid params";
594			return -EINVAL;
595		}
596
597		/* Parameters that take a numeric value are checked here */
598		if (!strcasecmp(key, "rebuild")) {
599			if (value >= rs->md.raid_disks) {
600				rs->ti->error = "Invalid rebuild index given";
601				return -EINVAL;
602			}
603			clear_bit(In_sync, &rs->dev[value].rdev.flags);
604			rs->dev[value].rdev.recovery_offset = 0;
605			rs->print_flags |= DMPF_REBUILD;
606		} else if (!strcasecmp(key, "write_mostly")) {
607			if (rs->raid_type->level != 1) {
608				rs->ti->error = "write_mostly option is only valid for RAID1";
609				return -EINVAL;
610			}
611			if (value >= rs->md.raid_disks) {
612				rs->ti->error = "Invalid write_mostly drive index given";
613				return -EINVAL;
614			}
615			set_bit(WriteMostly, &rs->dev[value].rdev.flags);
616		} else if (!strcasecmp(key, "max_write_behind")) {
617			if (rs->raid_type->level != 1) {
618				rs->ti->error = "max_write_behind option is only valid for RAID1";
619				return -EINVAL;
620			}
621			rs->print_flags |= DMPF_MAX_WRITE_BEHIND;
622
623			/*
624			 * In device-mapper, we specify things in sectors, but
625			 * MD records this value in kB
626			 */
627			value /= 2;
628			if (value > COUNTER_MAX) {
629				rs->ti->error = "Max write-behind limit out of range";
630				return -EINVAL;
631			}
632			rs->md.bitmap_info.max_write_behind = value;
633		} else if (!strcasecmp(key, "daemon_sleep")) {
634			rs->print_flags |= DMPF_DAEMON_SLEEP;
635			if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
636				rs->ti->error = "daemon sleep period out of range";
637				return -EINVAL;
638			}
639			rs->md.bitmap_info.daemon_sleep = value;
640		} else if (!strcasecmp(key, "stripe_cache")) {
641			rs->print_flags |= DMPF_STRIPE_CACHE;
642
643			/*
644			 * In device-mapper, we specify things in sectors, but
645			 * MD records this value in kB
646			 */
647			value /= 2;
648
649			if ((rs->raid_type->level != 5) &&
650			    (rs->raid_type->level != 6)) {
651				rs->ti->error = "Inappropriate argument: stripe_cache";
652				return -EINVAL;
653			}
654			if (raid5_set_cache_size(&rs->md, (int)value)) {
655				rs->ti->error = "Bad stripe_cache size";
656				return -EINVAL;
657			}
658		} else if (!strcasecmp(key, "min_recovery_rate")) {
659			rs->print_flags |= DMPF_MIN_RECOVERY_RATE;
660			if (value > INT_MAX) {
661				rs->ti->error = "min_recovery_rate out of range";
662				return -EINVAL;
663			}
664			rs->md.sync_speed_min = (int)value;
665		} else if (!strcasecmp(key, "max_recovery_rate")) {
666			rs->print_flags |= DMPF_MAX_RECOVERY_RATE;
667			if (value > INT_MAX) {
668				rs->ti->error = "max_recovery_rate out of range";
669				return -EINVAL;
670			}
671			rs->md.sync_speed_max = (int)value;
672		} else if (!strcasecmp(key, "region_size")) {
673			rs->print_flags |= DMPF_REGION_SIZE;
674			region_size = value;
675		} else if (!strcasecmp(key, "raid10_copies") &&
676			   (rs->raid_type->level == 10)) {
677			if ((value < 2) || (value > 0xFF)) {
678				rs->ti->error = "Bad value for 'raid10_copies'";
679				return -EINVAL;
680			}
681			rs->print_flags |= DMPF_RAID10_COPIES;
682			raid10_copies = value;
683		} else {
684			DMERR("Unable to parse RAID parameter: %s", key);
685			rs->ti->error = "Unable to parse RAID parameters";
686			return -EINVAL;
687		}
688	}
689
690	if (validate_region_size(rs, region_size))
691		return -EINVAL;
692
693	if (rs->md.chunk_sectors)
694		max_io_len = rs->md.chunk_sectors;
695	else
696		max_io_len = region_size;
697
698	if (dm_set_target_max_io_len(rs->ti, max_io_len))
699		return -EINVAL;
700
701	if (rs->raid_type->level == 10) {
702		if (raid10_copies > rs->md.raid_disks) {
703			rs->ti->error = "Not enough devices to satisfy specification";
704			return -EINVAL;
705		}
706
707		/*
708		 * If the format is not "near", we only support
709		 * two copies at the moment.
710		 */
711		if (strcmp("near", raid10_format) && (raid10_copies > 2)) {
712			rs->ti->error = "Too many copies for given RAID10 format.";
713			return -EINVAL;
714		}
715
716		/* (Len * #mirrors) / #devices */
717		sectors_per_dev = rs->ti->len * raid10_copies;
718		sector_div(sectors_per_dev, rs->md.raid_disks);
719
720		rs->md.layout = raid10_format_to_md_layout(raid10_format,
721							   raid10_copies);
722		rs->md.new_layout = rs->md.layout;
723	} else if ((rs->raid_type->level > 1) &&
724		   sector_div(sectors_per_dev,
725			      (rs->md.raid_disks - rs->raid_type->parity_devs))) {
726		rs->ti->error = "Target length not divisible by number of data devices";
727		return -EINVAL;
728	}
729	rs->md.dev_sectors = sectors_per_dev;
730
731	/* Assume there are no metadata devices until the drives are parsed */
732	rs->md.persistent = 0;
733	rs->md.external = 1;
734
735	return 0;
736}
737
738static void do_table_event(struct work_struct *ws)
739{
740	struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
741
742	dm_table_event(rs->ti->table);
743}
744
745static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
746{
747	struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
748
749	if (rs->raid_type->level == 1)
750		return md_raid1_congested(&rs->md, bits);
751
752	if (rs->raid_type->level == 10)
753		return md_raid10_congested(&rs->md, bits);
754
755	return md_raid5_congested(&rs->md, bits);
756}
757
758/*
759 * This structure is never routinely used by userspace, unlike md superblocks.
760 * Devices with this superblock should only ever be accessed via device-mapper.
761 */
762#define DM_RAID_MAGIC 0x64526D44
763struct dm_raid_superblock {
764	__le32 magic;		/* "DmRd" */
765	__le32 features;	/* Used to indicate possible future changes */
766
767	__le32 num_devices;	/* Number of devices in this array. (Max 64) */
768	__le32 array_position;	/* The position of this drive in the array */
769
770	__le64 events;		/* Incremented by md when superblock updated */
771	__le64 failed_devices;	/* Bit field of devices to indicate failures */
772
773	/*
774	 * This offset tracks the progress of the repair or replacement of
775	 * an individual drive.
776	 */
777	__le64 disk_recovery_offset;
778
779	/*
780	 * This offset tracks the progress of the initial array
781	 * synchronisation/parity calculation.
782	 */
783	__le64 array_resync_offset;
784
785	/*
786	 * RAID characteristics
787	 */
788	__le32 level;
789	__le32 layout;
790	__le32 stripe_sectors;
791
792	/* Remainder of a logical block is zero-filled when writing (see super_sync()). */
793} __packed;
794
795static int read_disk_sb(struct md_rdev *rdev, int size)
796{
797	BUG_ON(!rdev->sb_page);
798
799	if (rdev->sb_loaded)
800		return 0;
801
802	if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, 1)) {
803		DMERR("Failed to read superblock of device at position %d",
804		      rdev->raid_disk);
805		md_error(rdev->mddev, rdev);
806		return -EINVAL;
807	}
808
809	rdev->sb_loaded = 1;
810
811	return 0;
812}
813
814static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
815{
816	int i;
817	uint64_t failed_devices;
818	struct dm_raid_superblock *sb;
819	struct raid_set *rs = container_of(mddev, struct raid_set, md);
820
821	sb = page_address(rdev->sb_page);
822	failed_devices = le64_to_cpu(sb->failed_devices);
823
824	for (i = 0; i < mddev->raid_disks; i++)
825		if (!rs->dev[i].data_dev ||
826		    test_bit(Faulty, &(rs->dev[i].rdev.flags)))
827			failed_devices |= (1ULL << i);
828
829	memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
830
831	sb->magic = cpu_to_le32(DM_RAID_MAGIC);
832	sb->features = cpu_to_le32(0);	/* No features yet */
833
834	sb->num_devices = cpu_to_le32(mddev->raid_disks);
835	sb->array_position = cpu_to_le32(rdev->raid_disk);
836
837	sb->events = cpu_to_le64(mddev->events);
838	sb->failed_devices = cpu_to_le64(failed_devices);
839
840	sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
841	sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
842
843	sb->level = cpu_to_le32(mddev->level);
844	sb->layout = cpu_to_le32(mddev->layout);
845	sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
846}
847
848/*
849 * super_load
850 *
851 * This function creates a superblock if one is not found on the device
852 * and will decide which superblock to use if there's a choice.
853 *
854 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
855 */
856static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
857{
858	int ret;
859	struct dm_raid_superblock *sb;
860	struct dm_raid_superblock *refsb;
861	uint64_t events_sb, events_refsb;
862
863	rdev->sb_start = 0;
864	rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
865	if (rdev->sb_size < sizeof(*sb) || rdev->sb_size > PAGE_SIZE) {
866		DMERR("superblock size of a logical block is no longer valid");
867		return -EINVAL;
868	}
869
870	ret = read_disk_sb(rdev, rdev->sb_size);
871	if (ret)
872		return ret;
873
874	sb = page_address(rdev->sb_page);
875
876	/*
877	 * Two cases that we want to write new superblocks and rebuild:
878	 * 1) New device (no matching magic number)
879	 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
880	 */
881	if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
882	    (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
883		super_sync(rdev->mddev, rdev);
884
885		set_bit(FirstUse, &rdev->flags);
886
887		/* Force writing of superblocks to disk */
888		set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags);
889
890		/* Any superblock is better than none, choose that if given */
891		return refdev ? 0 : 1;
892	}
893
894	if (!refdev)
895		return 1;
896
897	events_sb = le64_to_cpu(sb->events);
898
899	refsb = page_address(refdev->sb_page);
900	events_refsb = le64_to_cpu(refsb->events);
901
902	return (events_sb > events_refsb) ? 1 : 0;
903}
904
905static int super_init_validation(struct mddev *mddev, struct md_rdev *rdev)
906{
907	int role;
908	struct raid_set *rs = container_of(mddev, struct raid_set, md);
909	uint64_t events_sb;
910	uint64_t failed_devices;
911	struct dm_raid_superblock *sb;
912	uint32_t new_devs = 0;
913	uint32_t rebuilds = 0;
914	struct md_rdev *r;
915	struct dm_raid_superblock *sb2;
916
917	sb = page_address(rdev->sb_page);
918	events_sb = le64_to_cpu(sb->events);
919	failed_devices = le64_to_cpu(sb->failed_devices);
920
921	/*
922	 * Initialise to 1 if this is a new superblock.
923	 */
924	mddev->events = events_sb ? : 1;
925
926	/*
927	 * Reshaping is not currently allowed
928	 */
929	if (le32_to_cpu(sb->level) != mddev->level) {
930		DMERR("Reshaping arrays not yet supported. (RAID level change)");
931		return -EINVAL;
932	}
933	if (le32_to_cpu(sb->layout) != mddev->layout) {
934		DMERR("Reshaping arrays not yet supported. (RAID layout change)");
935		DMERR("  0x%X vs 0x%X", le32_to_cpu(sb->layout), mddev->layout);
936		DMERR("  Old layout: %s w/ %d copies",
937		      raid10_md_layout_to_format(le32_to_cpu(sb->layout)),
938		      raid10_md_layout_to_copies(le32_to_cpu(sb->layout)));
939		DMERR("  New layout: %s w/ %d copies",
940		      raid10_md_layout_to_format(mddev->layout),
941		      raid10_md_layout_to_copies(mddev->layout));
942		return -EINVAL;
943	}
944	if (le32_to_cpu(sb->stripe_sectors) != mddev->chunk_sectors) {
945		DMERR("Reshaping arrays not yet supported. (stripe sectors change)");
946		return -EINVAL;
947	}
948
949	/* We can only change the number of devices in RAID1 right now */
950	if ((rs->raid_type->level != 1) &&
951	    (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) {
952		DMERR("Reshaping arrays not yet supported. (device count change)");
953		return -EINVAL;
954	}
955
956	if (!(rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC)))
957		mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
958
959	/*
960	 * During load, we set FirstUse if a new superblock was written.
961	 * There are two reasons we might not have a superblock:
962	 * 1) The array is brand new - in which case, all of the
963	 *    devices must have their In_sync bit set.  Also,
964	 *    recovery_cp must be 0, unless forced.
965	 * 2) This is a new device being added to an old array
966	 *    and the new device needs to be rebuilt - in which
967	 *    case the In_sync bit will /not/ be set and
968	 *    recovery_cp must be MaxSector.
969	 */
970	rdev_for_each(r, mddev) {
971		if (!test_bit(In_sync, &r->flags)) {
972			DMINFO("Device %d specified for rebuild: "
973			       "Clearing superblock", r->raid_disk);
974			rebuilds++;
975		} else if (test_bit(FirstUse, &r->flags))
976			new_devs++;
977	}
978
979	if (!rebuilds) {
980		if (new_devs == mddev->raid_disks) {
981			DMINFO("Superblocks created for new array");
982			set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
983		} else if (new_devs) {
984			DMERR("New device injected "
985			      "into existing array without 'rebuild' "
986			      "parameter specified");
987			return -EINVAL;
988		}
989	} else if (new_devs) {
990		DMERR("'rebuild' devices cannot be "
991		      "injected into an array with other first-time devices");
992		return -EINVAL;
993	} else if (mddev->recovery_cp != MaxSector) {
994		DMERR("'rebuild' specified while array is not in-sync");
995		return -EINVAL;
996	}
997
998	/*
999	 * Now we set the Faulty bit for those devices that are
1000	 * recorded in the superblock as failed.
1001	 */
1002	rdev_for_each(r, mddev) {
1003		if (!r->sb_page)
1004			continue;
1005		sb2 = page_address(r->sb_page);
1006		sb2->failed_devices = 0;
1007
1008		/*
1009		 * Check for any device re-ordering.
1010		 */
1011		if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
1012			role = le32_to_cpu(sb2->array_position);
1013			if (role != r->raid_disk) {
1014				if (rs->raid_type->level != 1) {
1015					rs->ti->error = "Cannot change device "
1016						"positions in RAID array";
1017					return -EINVAL;
1018				}
1019				DMINFO("RAID1 device #%d now at position #%d",
1020				       role, r->raid_disk);
1021			}
1022
1023			/*
1024			 * Partial recovery is performed on
1025			 * returning failed devices.
1026			 */
1027			if (failed_devices & (1 << role))
1028				set_bit(Faulty, &r->flags);
1029		}
1030	}
1031
1032	return 0;
1033}
1034
1035static int super_validate(struct mddev *mddev, struct md_rdev *rdev)
1036{
1037	struct dm_raid_superblock *sb = page_address(rdev->sb_page);
1038
1039	/*
1040	 * If mddev->events is not set, we know we have not yet initialized
1041	 * the array.
1042	 */
1043	if (!mddev->events && super_init_validation(mddev, rdev))
1044		return -EINVAL;
1045
1046	mddev->bitmap_info.offset = 4096 >> 9; /* Enable bitmap creation */
1047	rdev->mddev->bitmap_info.default_offset = 4096 >> 9;
1048	if (!test_bit(FirstUse, &rdev->flags)) {
1049		rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
1050		if (rdev->recovery_offset != MaxSector)
1051			clear_bit(In_sync, &rdev->flags);
1052	}
1053
1054	/*
1055	 * If a device comes back, set it as not In_sync and no longer faulty.
1056	 */
1057	if (test_bit(Faulty, &rdev->flags)) {
1058		clear_bit(Faulty, &rdev->flags);
1059		clear_bit(In_sync, &rdev->flags);
1060		rdev->saved_raid_disk = rdev->raid_disk;
1061		rdev->recovery_offset = 0;
1062	}
1063
1064	clear_bit(FirstUse, &rdev->flags);
1065
1066	return 0;
1067}
1068
1069/*
1070 * Analyse superblocks and select the freshest.
1071 */
1072static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
1073{
1074	int ret;
1075	struct raid_dev *dev;
1076	struct md_rdev *rdev, *tmp, *freshest;
1077	struct mddev *mddev = &rs->md;
1078
1079	freshest = NULL;
1080	rdev_for_each_safe(rdev, tmp, mddev) {
1081		/*
1082		 * Skipping super_load due to DMPF_SYNC will cause
1083		 * the array to undergo initialization again as
1084		 * though it were new.  This is the intended effect
1085		 * of the "sync" directive.
1086		 *
1087		 * When reshaping capability is added, we must ensure
1088		 * that the "sync" directive is disallowed during the
1089		 * reshape.
1090		 */
1091		if (rs->print_flags & DMPF_SYNC)
1092			continue;
1093
1094		if (!rdev->meta_bdev)
1095			continue;
1096
1097		ret = super_load(rdev, freshest);
1098
1099		switch (ret) {
1100		case 1:
1101			freshest = rdev;
1102			break;
1103		case 0:
1104			break;
1105		default:
1106			dev = container_of(rdev, struct raid_dev, rdev);
1107			if (dev->meta_dev)
1108				dm_put_device(ti, dev->meta_dev);
1109
1110			dev->meta_dev = NULL;
1111			rdev->meta_bdev = NULL;
1112
1113			if (rdev->sb_page)
1114				put_page(rdev->sb_page);
1115
1116			rdev->sb_page = NULL;
1117
1118			rdev->sb_loaded = 0;
1119
1120			/*
1121			 * We might be able to salvage the data device
1122			 * even though the meta device has failed.  For
1123			 * now, we behave as though '- -' had been
1124			 * set for this device in the table.
1125			 */
1126			if (dev->data_dev)
1127				dm_put_device(ti, dev->data_dev);
1128
1129			dev->data_dev = NULL;
1130			rdev->bdev = NULL;
1131
1132			list_del(&rdev->same_set);
1133		}
1134	}
1135
1136	if (!freshest)
1137		return 0;
1138
1139	if (validate_raid_redundancy(rs)) {
1140		rs->ti->error = "Insufficient redundancy to activate array";
1141		return -EINVAL;
1142	}
1143
1144	/*
1145	 * Validation of the freshest device provides the source of
1146	 * validation for the remaining devices.
1147	 */
1148	ti->error = "Unable to assemble array: Invalid superblocks";
1149	if (super_validate(mddev, freshest))
1150		return -EINVAL;
1151
1152	rdev_for_each(rdev, mddev)
1153		if ((rdev != freshest) && super_validate(mddev, rdev))
1154			return -EINVAL;
1155
1156	return 0;
1157}
1158
1159/*
1160 * Enable/disable discard support on RAID set depending on
1161 * RAID level and discard properties of underlying RAID members.
1162 */
1163static void configure_discard_support(struct dm_target *ti, struct raid_set *rs)
1164{
1165	int i;
1166	bool raid456;
1167
1168	/* Assume discards not supported until after checks below. */
1169	ti->discards_supported = false;
1170
1171	/* RAID level 4,5,6 require discard_zeroes_data for data integrity! */
1172	raid456 = (rs->md.level == 4 || rs->md.level == 5 || rs->md.level == 6);
1173
1174	for (i = 0; i < rs->md.raid_disks; i++) {
1175		struct request_queue *q;
1176
1177		if (!rs->dev[i].rdev.bdev)
1178			continue;
1179
1180		q = bdev_get_queue(rs->dev[i].rdev.bdev);
1181		if (!q || !blk_queue_discard(q))
1182			return;
1183
1184		if (raid456) {
1185			if (!q->limits.discard_zeroes_data)
1186				return;
1187			if (!devices_handle_discard_safely) {
1188				DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
1189				DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
1190				return;
1191			}
1192		}
1193	}
1194
1195	/* All RAID members properly support discards */
1196	ti->discards_supported = true;
1197
1198	/*
1199	 * RAID1 and RAID10 personalities require bio splitting,
1200	 * RAID0/4/5/6 don't and process large discard bios properly.
1201	 */
1202	ti->split_discard_bios = !!(rs->md.level == 1 || rs->md.level == 10);
1203	ti->num_discard_bios = 1;
1204}
1205
1206/*
1207 * Construct a RAID4/5/6 mapping:
1208 * Args:
1209 *	<raid_type> <#raid_params> <raid_params>		\
1210 *	<#raid_devs> { <meta_dev1> <dev1> .. <meta_devN> <devN> }
1211 *
1212 * <raid_params> varies by <raid_type>.  See 'parse_raid_params' for
1213 * details on possible <raid_params>.
1214 */
1215static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv)
1216{
1217	int ret;
1218	struct raid_type *rt;
1219	unsigned long num_raid_params, num_raid_devs;
1220	struct raid_set *rs = NULL;
1221
1222	/* Must have at least <raid_type> <#raid_params> */
1223	if (argc < 2) {
1224		ti->error = "Too few arguments";
1225		return -EINVAL;
1226	}
1227
1228	/* raid type */
1229	rt = get_raid_type(argv[0]);
1230	if (!rt) {
1231		ti->error = "Unrecognised raid_type";
1232		return -EINVAL;
1233	}
1234	argc--;
1235	argv++;
1236
1237	/* number of RAID parameters */
1238	if (kstrtoul(argv[0], 10, &num_raid_params) < 0) {
1239		ti->error = "Cannot understand number of RAID parameters";
1240		return -EINVAL;
1241	}
1242	argc--;
1243	argv++;
1244
1245	/* Skip over RAID params for now and find out # of devices */
1246	if (num_raid_params + 1 > argc) {
1247		ti->error = "Arguments do not agree with counts given";
1248		return -EINVAL;
1249	}
1250
1251	if ((kstrtoul(argv[num_raid_params], 10, &num_raid_devs) < 0) ||
1252	    (num_raid_devs >= INT_MAX)) {
1253		ti->error = "Cannot understand number of raid devices";
1254		return -EINVAL;
1255	}
1256
1257	rs = context_alloc(ti, rt, (unsigned)num_raid_devs);
1258	if (IS_ERR(rs))
1259		return PTR_ERR(rs);
1260
1261	ret = parse_raid_params(rs, argv, (unsigned)num_raid_params);
1262	if (ret)
1263		goto bad;
1264
1265	ret = -EINVAL;
1266
1267	argc -= num_raid_params + 1; /* +1: we already have num_raid_devs */
1268	argv += num_raid_params + 1;
1269
1270	if (argc != (num_raid_devs * 2)) {
1271		ti->error = "Supplied RAID devices does not match the count given";
1272		goto bad;
1273	}
1274
1275	ret = dev_parms(rs, argv);
1276	if (ret)
1277		goto bad;
1278
1279	rs->md.sync_super = super_sync;
1280	ret = analyse_superblocks(ti, rs);
1281	if (ret)
1282		goto bad;
1283
1284	INIT_WORK(&rs->md.event_work, do_table_event);
1285	ti->private = rs;
1286	ti->num_flush_bios = 1;
1287
1288	/*
1289	 * Disable/enable discard support on RAID set.
1290	 */
1291	configure_discard_support(ti, rs);
1292
1293	mutex_lock(&rs->md.reconfig_mutex);
1294	ret = md_run(&rs->md);
1295	rs->md.in_sync = 0; /* Assume already marked dirty */
1296	mutex_unlock(&rs->md.reconfig_mutex);
1297
1298	if (ret) {
1299		ti->error = "Fail to run raid array";
1300		goto bad;
1301	}
1302
1303	if (ti->len != rs->md.array_sectors) {
1304		ti->error = "Array size does not match requested target length";
1305		ret = -EINVAL;
1306		goto size_mismatch;
1307	}
1308	rs->callbacks.congested_fn = raid_is_congested;
1309	dm_table_add_target_callbacks(ti->table, &rs->callbacks);
1310
1311	mddev_suspend(&rs->md);
1312	return 0;
1313
1314size_mismatch:
1315	md_stop(&rs->md);
1316bad:
1317	context_free(rs);
1318
1319	return ret;
1320}
1321
1322static void raid_dtr(struct dm_target *ti)
1323{
1324	struct raid_set *rs = ti->private;
1325
1326	list_del_init(&rs->callbacks.list);
1327	md_stop(&rs->md);
1328	context_free(rs);
1329}
1330
1331static int raid_map(struct dm_target *ti, struct bio *bio)
1332{
1333	struct raid_set *rs = ti->private;
1334	struct mddev *mddev = &rs->md;
1335
1336	mddev->pers->make_request(mddev, bio);
1337
1338	return DM_MAPIO_SUBMITTED;
1339}
1340
1341static const char *decipher_sync_action(struct mddev *mddev)
1342{
1343	if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
1344		return "frozen";
1345
1346	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1347	    (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
1348		if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
1349			return "reshape";
1350
1351		if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1352			if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1353				return "resync";
1354			else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1355				return "check";
1356			return "repair";
1357		}
1358
1359		if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
1360			return "recover";
1361	}
1362
1363	return "idle";
1364}
1365
1366static void raid_status(struct dm_target *ti, status_type_t type,
1367			unsigned status_flags, char *result, unsigned maxlen)
1368{
1369	struct raid_set *rs = ti->private;
1370	unsigned raid_param_cnt = 1; /* at least 1 for chunksize */
1371	unsigned sz = 0;
1372	int i, array_in_sync = 0;
1373	sector_t sync;
1374
1375	switch (type) {
1376	case STATUSTYPE_INFO:
1377		DMEMIT("%s %d ", rs->raid_type->name, rs->md.raid_disks);
1378
1379		if (test_bit(MD_RECOVERY_RUNNING, &rs->md.recovery))
1380			sync = rs->md.curr_resync_completed;
1381		else
1382			sync = rs->md.recovery_cp;
1383
1384		if (sync >= rs->md.resync_max_sectors) {
1385			/*
1386			 * Sync complete.
1387			 */
1388			array_in_sync = 1;
1389			sync = rs->md.resync_max_sectors;
1390		} else if (test_bit(MD_RECOVERY_REQUESTED, &rs->md.recovery)) {
1391			/*
1392			 * If "check" or "repair" is occurring, the array has
1393			 * undergone and initial sync and the health characters
1394			 * should not be 'a' anymore.
1395			 */
1396			array_in_sync = 1;
1397		} else {
1398			/*
1399			 * The array may be doing an initial sync, or it may
1400			 * be rebuilding individual components.  If all the
1401			 * devices are In_sync, then it is the array that is
1402			 * being initialized.
1403			 */
1404			for (i = 0; i < rs->md.raid_disks; i++)
1405				if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
1406					array_in_sync = 1;
1407		}
1408
1409		/*
1410		 * Status characters:
1411		 *  'D' = Dead/Failed device
1412		 *  'a' = Alive but not in-sync
1413		 *  'A' = Alive and in-sync
1414		 */
1415		for (i = 0; i < rs->md.raid_disks; i++) {
1416			if (test_bit(Faulty, &rs->dev[i].rdev.flags))
1417				DMEMIT("D");
1418			else if (!array_in_sync ||
1419				 !test_bit(In_sync, &rs->dev[i].rdev.flags))
1420				DMEMIT("a");
1421			else
1422				DMEMIT("A");
1423		}
1424
1425		/*
1426		 * In-sync ratio:
1427		 *  The in-sync ratio shows the progress of:
1428		 *   - Initializing the array
1429		 *   - Rebuilding a subset of devices of the array
1430		 *  The user can distinguish between the two by referring
1431		 *  to the status characters.
1432		 */
1433		DMEMIT(" %llu/%llu",
1434		       (unsigned long long) sync,
1435		       (unsigned long long) rs->md.resync_max_sectors);
1436
1437		/*
1438		 * Sync action:
1439		 *   See Documentation/device-mapper/dm-raid.c for
1440		 *   information on each of these states.
1441		 */
1442		DMEMIT(" %s", decipher_sync_action(&rs->md));
1443
1444		/*
1445		 * resync_mismatches/mismatch_cnt
1446		 *   This field shows the number of discrepancies found when
1447		 *   performing a "check" of the array.
1448		 */
1449		DMEMIT(" %llu",
1450		       (strcmp(rs->md.last_sync_action, "check")) ? 0 :
1451		       (unsigned long long)
1452		       atomic64_read(&rs->md.resync_mismatches));
1453		break;
1454	case STATUSTYPE_TABLE:
1455		/* The string you would use to construct this array */
1456		for (i = 0; i < rs->md.raid_disks; i++) {
1457			if ((rs->print_flags & DMPF_REBUILD) &&
1458			    rs->dev[i].data_dev &&
1459			    !test_bit(In_sync, &rs->dev[i].rdev.flags))
1460				raid_param_cnt += 2; /* for rebuilds */
1461			if (rs->dev[i].data_dev &&
1462			    test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1463				raid_param_cnt += 2;
1464		}
1465
1466		raid_param_cnt += (hweight32(rs->print_flags & ~DMPF_REBUILD) * 2);
1467		if (rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC))
1468			raid_param_cnt--;
1469
1470		DMEMIT("%s %u %u", rs->raid_type->name,
1471		       raid_param_cnt, rs->md.chunk_sectors);
1472
1473		if ((rs->print_flags & DMPF_SYNC) &&
1474		    (rs->md.recovery_cp == MaxSector))
1475			DMEMIT(" sync");
1476		if (rs->print_flags & DMPF_NOSYNC)
1477			DMEMIT(" nosync");
1478
1479		for (i = 0; i < rs->md.raid_disks; i++)
1480			if ((rs->print_flags & DMPF_REBUILD) &&
1481			    rs->dev[i].data_dev &&
1482			    !test_bit(In_sync, &rs->dev[i].rdev.flags))
1483				DMEMIT(" rebuild %u", i);
1484
1485		if (rs->print_flags & DMPF_DAEMON_SLEEP)
1486			DMEMIT(" daemon_sleep %lu",
1487			       rs->md.bitmap_info.daemon_sleep);
1488
1489		if (rs->print_flags & DMPF_MIN_RECOVERY_RATE)
1490			DMEMIT(" min_recovery_rate %d", rs->md.sync_speed_min);
1491
1492		if (rs->print_flags & DMPF_MAX_RECOVERY_RATE)
1493			DMEMIT(" max_recovery_rate %d", rs->md.sync_speed_max);
1494
1495		for (i = 0; i < rs->md.raid_disks; i++)
1496			if (rs->dev[i].data_dev &&
1497			    test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1498				DMEMIT(" write_mostly %u", i);
1499
1500		if (rs->print_flags & DMPF_MAX_WRITE_BEHIND)
1501			DMEMIT(" max_write_behind %lu",
1502			       rs->md.bitmap_info.max_write_behind);
1503
1504		if (rs->print_flags & DMPF_STRIPE_CACHE) {
1505			struct r5conf *conf = rs->md.private;
1506
1507			/* convert from kiB to sectors */
1508			DMEMIT(" stripe_cache %d",
1509			       conf ? conf->max_nr_stripes * 2 : 0);
1510		}
1511
1512		if (rs->print_flags & DMPF_REGION_SIZE)
1513			DMEMIT(" region_size %lu",
1514			       rs->md.bitmap_info.chunksize >> 9);
1515
1516		if (rs->print_flags & DMPF_RAID10_COPIES)
1517			DMEMIT(" raid10_copies %u",
1518			       raid10_md_layout_to_copies(rs->md.layout));
1519
1520		if (rs->print_flags & DMPF_RAID10_FORMAT)
1521			DMEMIT(" raid10_format %s",
1522			       raid10_md_layout_to_format(rs->md.layout));
1523
1524		DMEMIT(" %d", rs->md.raid_disks);
1525		for (i = 0; i < rs->md.raid_disks; i++) {
1526			if (rs->dev[i].meta_dev)
1527				DMEMIT(" %s", rs->dev[i].meta_dev->name);
1528			else
1529				DMEMIT(" -");
1530
1531			if (rs->dev[i].data_dev)
1532				DMEMIT(" %s", rs->dev[i].data_dev->name);
1533			else
1534				DMEMIT(" -");
1535		}
1536	}
1537}
1538
1539static int raid_message(struct dm_target *ti, unsigned argc, char **argv)
1540{
1541	struct raid_set *rs = ti->private;
1542	struct mddev *mddev = &rs->md;
1543
1544	if (!strcasecmp(argv[0], "reshape")) {
1545		DMERR("Reshape not supported.");
1546		return -EINVAL;
1547	}
1548
1549	if (!mddev->pers || !mddev->pers->sync_request)
1550		return -EINVAL;
1551
1552	if (!strcasecmp(argv[0], "frozen"))
1553		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1554	else
1555		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1556
1557	if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
1558		if (mddev->sync_thread) {
1559			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1560			md_reap_sync_thread(mddev);
1561		}
1562	} else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1563		   test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1564		return -EBUSY;
1565	else if (!strcasecmp(argv[0], "resync"))
1566		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1567	else if (!strcasecmp(argv[0], "recover")) {
1568		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
1569		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1570	} else {
1571		if (!strcasecmp(argv[0], "check"))
1572			set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1573		else if (!!strcasecmp(argv[0], "repair"))
1574			return -EINVAL;
1575		set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1576		set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1577	}
1578	if (mddev->ro == 2) {
1579		/* A write to sync_action is enough to justify
1580		 * canceling read-auto mode
1581		 */
1582		mddev->ro = 0;
1583		if (!mddev->suspended)
1584			md_wakeup_thread(mddev->sync_thread);
1585	}
1586	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1587	if (!mddev->suspended)
1588		md_wakeup_thread(mddev->thread);
1589
1590	return 0;
1591}
1592
1593static int raid_iterate_devices(struct dm_target *ti,
1594				iterate_devices_callout_fn fn, void *data)
1595{
1596	struct raid_set *rs = ti->private;
1597	unsigned i;
1598	int ret = 0;
1599
1600	for (i = 0; !ret && i < rs->md.raid_disks; i++)
1601		if (rs->dev[i].data_dev)
1602			ret = fn(ti,
1603				 rs->dev[i].data_dev,
1604				 0, /* No offset on data devs */
1605				 rs->md.dev_sectors,
1606				 data);
1607
1608	return ret;
1609}
1610
1611static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
1612{
1613	struct raid_set *rs = ti->private;
1614	unsigned chunk_size = rs->md.chunk_sectors << 9;
1615	struct r5conf *conf = rs->md.private;
1616
1617	blk_limits_io_min(limits, chunk_size);
1618	blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded));
1619}
1620
1621static void raid_presuspend(struct dm_target *ti)
1622{
1623	struct raid_set *rs = ti->private;
1624
1625	md_stop_writes(&rs->md);
1626}
1627
1628static void raid_postsuspend(struct dm_target *ti)
1629{
1630	struct raid_set *rs = ti->private;
1631
1632	mddev_suspend(&rs->md);
1633}
1634
1635static void attempt_restore_of_faulty_devices(struct raid_set *rs)
1636{
1637	int i;
1638	uint64_t failed_devices, cleared_failed_devices = 0;
1639	unsigned long flags;
1640	struct dm_raid_superblock *sb;
1641	struct md_rdev *r;
1642
1643	for (i = 0; i < rs->md.raid_disks; i++) {
1644		r = &rs->dev[i].rdev;
1645		if (test_bit(Faulty, &r->flags) && r->sb_page &&
1646		    sync_page_io(r, 0, r->sb_size, r->sb_page, READ, 1)) {
1647			DMINFO("Faulty %s device #%d has readable super block."
1648			       "  Attempting to revive it.",
1649			       rs->raid_type->name, i);
1650
1651			/*
1652			 * Faulty bit may be set, but sometimes the array can
1653			 * be suspended before the personalities can respond
1654			 * by removing the device from the array (i.e. calling
1655			 * 'hot_remove_disk').  If they haven't yet removed
1656			 * the failed device, its 'raid_disk' number will be
1657			 * '>= 0' - meaning we must call this function
1658			 * ourselves.
1659			 */
1660			if ((r->raid_disk >= 0) &&
1661			    (r->mddev->pers->hot_remove_disk(r->mddev, r) != 0))
1662				/* Failed to revive this device, try next */
1663				continue;
1664
1665			r->raid_disk = i;
1666			r->saved_raid_disk = i;
1667			flags = r->flags;
1668			clear_bit(Faulty, &r->flags);
1669			clear_bit(WriteErrorSeen, &r->flags);
1670			clear_bit(In_sync, &r->flags);
1671			if (r->mddev->pers->hot_add_disk(r->mddev, r)) {
1672				r->raid_disk = -1;
1673				r->saved_raid_disk = -1;
1674				r->flags = flags;
1675			} else {
1676				r->recovery_offset = 0;
1677				cleared_failed_devices |= 1 << i;
1678			}
1679		}
1680	}
1681	if (cleared_failed_devices) {
1682		rdev_for_each(r, &rs->md) {
1683			sb = page_address(r->sb_page);
1684			failed_devices = le64_to_cpu(sb->failed_devices);
1685			failed_devices &= ~cleared_failed_devices;
1686			sb->failed_devices = cpu_to_le64(failed_devices);
1687		}
1688	}
1689}
1690
1691static void raid_resume(struct dm_target *ti)
1692{
1693	struct raid_set *rs = ti->private;
1694
1695	set_bit(MD_CHANGE_DEVS, &rs->md.flags);
1696	if (!rs->bitmap_loaded) {
1697		bitmap_load(&rs->md);
1698		rs->bitmap_loaded = 1;
1699	} else {
1700		/*
1701		 * A secondary resume while the device is active.
1702		 * Take this opportunity to check whether any failed
1703		 * devices are reachable again.
1704		 */
1705		attempt_restore_of_faulty_devices(rs);
1706	}
1707
1708	clear_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
1709	mddev_resume(&rs->md);
1710}
1711
1712static struct target_type raid_target = {
1713	.name = "raid",
1714	.version = {1, 6, 0},
1715	.module = THIS_MODULE,
1716	.ctr = raid_ctr,
1717	.dtr = raid_dtr,
1718	.map = raid_map,
1719	.status = raid_status,
1720	.message = raid_message,
1721	.iterate_devices = raid_iterate_devices,
1722	.io_hints = raid_io_hints,
1723	.presuspend = raid_presuspend,
1724	.postsuspend = raid_postsuspend,
1725	.resume = raid_resume,
1726};
1727
1728static int __init dm_raid_init(void)
1729{
1730	DMINFO("Loading target version %u.%u.%u",
1731	       raid_target.version[0],
1732	       raid_target.version[1],
1733	       raid_target.version[2]);
1734	return dm_register_target(&raid_target);
1735}
1736
1737static void __exit dm_raid_exit(void)
1738{
1739	dm_unregister_target(&raid_target);
1740}
1741
1742module_init(dm_raid_init);
1743module_exit(dm_raid_exit);
1744
1745module_param(devices_handle_discard_safely, bool, 0644);
1746MODULE_PARM_DESC(devices_handle_discard_safely,
1747		 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
1748
1749MODULE_DESCRIPTION(DM_NAME " raid4/5/6 target");
1750MODULE_ALIAS("dm-raid1");
1751MODULE_ALIAS("dm-raid10");
1752MODULE_ALIAS("dm-raid4");
1753MODULE_ALIAS("dm-raid5");
1754MODULE_ALIAS("dm-raid6");
1755MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
1756MODULE_LICENSE("GPL");
1757