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			region_size = roundup_pow_of_two(min_region_size);
331			DMINFO("Choosing default region size of %lu sectors",
332			       region_size);
333		} else {
334			DMINFO("Choosing default region size of 4MiB");
335			region_size = 1 << 13; /* sectors */
336		}
337	} else {
338		/*
339		 * Validate user-supplied value.
340		 */
341		if (region_size > rs->ti->len) {
342			rs->ti->error = "Supplied region size is too large";
343			return -EINVAL;
344		}
345
346		if (region_size < min_region_size) {
347			DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
348			      region_size, min_region_size);
349			rs->ti->error = "Supplied region size is too small";
350			return -EINVAL;
351		}
352
353		if (!is_power_of_2(region_size)) {
354			rs->ti->error = "Region size is not a power of 2";
355			return -EINVAL;
356		}
357
358		if (region_size < rs->md.chunk_sectors) {
359			rs->ti->error = "Region size is smaller than the chunk size";
360			return -EINVAL;
361		}
362	}
363
364	/*
365	 * Convert sectors to bytes.
366	 */
367	rs->md.bitmap_info.chunksize = (region_size << 9);
368
369	return 0;
370}
371
372/*
373 * validate_raid_redundancy
374 * @rs
375 *
376 * Determine if there are enough devices in the array that haven't
377 * failed (or are being rebuilt) to form a usable array.
378 *
379 * Returns: 0 on success, -EINVAL on failure.
380 */
381static int validate_raid_redundancy(struct raid_set *rs)
382{
383	unsigned i, rebuild_cnt = 0;
384	unsigned rebuilds_per_group = 0, copies, d;
385	unsigned group_size, last_group_start;
386
387	for (i = 0; i < rs->md.raid_disks; i++)
388		if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
389		    !rs->dev[i].rdev.sb_page)
390			rebuild_cnt++;
391
392	switch (rs->raid_type->level) {
393	case 1:
394		if (rebuild_cnt >= rs->md.raid_disks)
395			goto too_many;
396		break;
397	case 4:
398	case 5:
399	case 6:
400		if (rebuild_cnt > rs->raid_type->parity_devs)
401			goto too_many;
402		break;
403	case 10:
404		copies = raid10_md_layout_to_copies(rs->md.layout);
405		if (rebuild_cnt < copies)
406			break;
407
408		/*
409		 * It is possible to have a higher rebuild count for RAID10,
410		 * as long as the failed devices occur in different mirror
411		 * groups (i.e. different stripes).
412		 *
413		 * When checking "near" format, make sure no adjacent devices
414		 * have failed beyond what can be handled.  In addition to the
415		 * simple case where the number of devices is a multiple of the
416		 * number of copies, we must also handle cases where the number
417		 * of devices is not a multiple of the number of copies.
418		 * E.g.    dev1 dev2 dev3 dev4 dev5
419		 *          A    A    B    B    C
420		 *          C    D    D    E    E
421		 */
422		if (!strcmp("near", raid10_md_layout_to_format(rs->md.layout))) {
423			for (i = 0; i < rs->md.raid_disks * copies; i++) {
424				if (!(i % copies))
425					rebuilds_per_group = 0;
426				d = i % rs->md.raid_disks;
427				if ((!rs->dev[d].rdev.sb_page ||
428				     !test_bit(In_sync, &rs->dev[d].rdev.flags)) &&
429				    (++rebuilds_per_group >= copies))
430					goto too_many;
431			}
432			break;
433		}
434
435		/*
436		 * When checking "far" and "offset" formats, we need to ensure
437		 * that the device that holds its copy is not also dead or
438		 * being rebuilt.  (Note that "far" and "offset" formats only
439		 * support two copies right now.  These formats also only ever
440		 * use the 'use_far_sets' variant.)
441		 *
442		 * This check is somewhat complicated by the need to account
443		 * for arrays that are not a multiple of (far) copies.  This
444		 * results in the need to treat the last (potentially larger)
445		 * set differently.
446		 */
447		group_size = (rs->md.raid_disks / copies);
448		last_group_start = (rs->md.raid_disks / group_size) - 1;
449		last_group_start *= group_size;
450		for (i = 0; i < rs->md.raid_disks; i++) {
451			if (!(i % copies) && !(i > last_group_start))
452				rebuilds_per_group = 0;
453			if ((!rs->dev[i].rdev.sb_page ||
454			     !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
455			    (++rebuilds_per_group >= copies))
456					goto too_many;
457		}
458		break;
459	default:
460		if (rebuild_cnt)
461			return -EINVAL;
462	}
463
464	return 0;
465
466too_many:
467	return -EINVAL;
468}
469
470/*
471 * Possible arguments are...
472 *	<chunk_size> [optional_args]
473 *
474 * Argument definitions
475 *    <chunk_size>			The number of sectors per disk that
476 *                                      will form the "stripe"
477 *    [[no]sync]			Force or prevent recovery of the
478 *                                      entire array
479 *    [devices_handle_discard_safely]	Allow discards on RAID4/5/6; useful if RAID
480 *					member device(s) properly support TRIM/UNMAP
481 *    [rebuild <idx>]			Rebuild the drive indicated by the index
482 *    [daemon_sleep <ms>]		Time between bitmap daemon work to
483 *                                      clear bits
484 *    [min_recovery_rate <kB/sec/disk>]	Throttle RAID initialization
485 *    [max_recovery_rate <kB/sec/disk>]	Throttle RAID initialization
486 *    [write_mostly <idx>]		Indicate a write mostly drive via index
487 *    [max_write_behind <sectors>]	See '-write-behind=' (man mdadm)
488 *    [stripe_cache <sectors>]		Stripe cache size for higher RAIDs
489 *    [region_size <sectors>]           Defines granularity of bitmap
490 *
491 * RAID10-only options:
492 *    [raid10_copies <# copies>]        Number of copies.  (Default: 2)
493 *    [raid10_format <near|far|offset>] Layout algorithm.  (Default: near)
494 */
495static int parse_raid_params(struct raid_set *rs, char **argv,
496			     unsigned num_raid_params)
497{
498	char *raid10_format = "near";
499	unsigned raid10_copies = 2;
500	unsigned i;
501	unsigned long value, region_size = 0;
502	sector_t sectors_per_dev = rs->ti->len;
503	sector_t max_io_len;
504	char *key;
505
506	/*
507	 * First, parse the in-order required arguments
508	 * "chunk_size" is the only argument of this type.
509	 */
510	if ((kstrtoul(argv[0], 10, &value) < 0)) {
511		rs->ti->error = "Bad chunk size";
512		return -EINVAL;
513	} else if (rs->raid_type->level == 1) {
514		if (value)
515			DMERR("Ignoring chunk size parameter for RAID 1");
516		value = 0;
517	} else if (!is_power_of_2(value)) {
518		rs->ti->error = "Chunk size must be a power of 2";
519		return -EINVAL;
520	} else if (value < 8) {
521		rs->ti->error = "Chunk size value is too small";
522		return -EINVAL;
523	}
524
525	rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
526	argv++;
527	num_raid_params--;
528
529	/*
530	 * We set each individual device as In_sync with a completed
531	 * 'recovery_offset'.  If there has been a device failure or
532	 * replacement then one of the following cases applies:
533	 *
534	 *   1) User specifies 'rebuild'.
535	 *      - Device is reset when param is read.
536	 *   2) A new device is supplied.
537	 *      - No matching superblock found, resets device.
538	 *   3) Device failure was transient and returns on reload.
539	 *      - Failure noticed, resets device for bitmap replay.
540	 *   4) Device hadn't completed recovery after previous failure.
541	 *      - Superblock is read and overrides recovery_offset.
542	 *
543	 * What is found in the superblocks of the devices is always
544	 * authoritative, unless 'rebuild' or '[no]sync' was specified.
545	 */
546	for (i = 0; i < rs->md.raid_disks; i++) {
547		set_bit(In_sync, &rs->dev[i].rdev.flags);
548		rs->dev[i].rdev.recovery_offset = MaxSector;
549	}
550
551	/*
552	 * Second, parse the unordered optional arguments
553	 */
554	for (i = 0; i < num_raid_params; i++) {
555		if (!strcasecmp(argv[i], "nosync")) {
556			rs->md.recovery_cp = MaxSector;
557			rs->print_flags |= DMPF_NOSYNC;
558			continue;
559		}
560		if (!strcasecmp(argv[i], "sync")) {
561			rs->md.recovery_cp = 0;
562			rs->print_flags |= DMPF_SYNC;
563			continue;
564		}
565
566		/* The rest of the optional arguments come in key/value pairs */
567		if ((i + 1) >= num_raid_params) {
568			rs->ti->error = "Wrong number of raid parameters given";
569			return -EINVAL;
570		}
571
572		key = argv[i++];
573
574		/* Parameters that take a string value are checked here. */
575		if (!strcasecmp(key, "raid10_format")) {
576			if (rs->raid_type->level != 10) {
577				rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
578				return -EINVAL;
579			}
580			if (strcmp("near", argv[i]) &&
581			    strcmp("far", argv[i]) &&
582			    strcmp("offset", argv[i])) {
583				rs->ti->error = "Invalid 'raid10_format' value given";
584				return -EINVAL;
585			}
586			raid10_format = argv[i];
587			rs->print_flags |= DMPF_RAID10_FORMAT;
588			continue;
589		}
590
591		if (kstrtoul(argv[i], 10, &value) < 0) {
592			rs->ti->error = "Bad numerical argument given in raid params";
593			return -EINVAL;
594		}
595
596		/* Parameters that take a numeric value are checked here */
597		if (!strcasecmp(key, "rebuild")) {
598			if (value >= rs->md.raid_disks) {
599				rs->ti->error = "Invalid rebuild index given";
600				return -EINVAL;
601			}
602			clear_bit(In_sync, &rs->dev[value].rdev.flags);
603			rs->dev[value].rdev.recovery_offset = 0;
604			rs->print_flags |= DMPF_REBUILD;
605		} else if (!strcasecmp(key, "write_mostly")) {
606			if (rs->raid_type->level != 1) {
607				rs->ti->error = "write_mostly option is only valid for RAID1";
608				return -EINVAL;
609			}
610			if (value >= rs->md.raid_disks) {
611				rs->ti->error = "Invalid write_mostly drive index given";
612				return -EINVAL;
613			}
614			set_bit(WriteMostly, &rs->dev[value].rdev.flags);
615		} else if (!strcasecmp(key, "max_write_behind")) {
616			if (rs->raid_type->level != 1) {
617				rs->ti->error = "max_write_behind option is only valid for RAID1";
618				return -EINVAL;
619			}
620			rs->print_flags |= DMPF_MAX_WRITE_BEHIND;
621
622			/*
623			 * In device-mapper, we specify things in sectors, but
624			 * MD records this value in kB
625			 */
626			value /= 2;
627			if (value > COUNTER_MAX) {
628				rs->ti->error = "Max write-behind limit out of range";
629				return -EINVAL;
630			}
631			rs->md.bitmap_info.max_write_behind = value;
632		} else if (!strcasecmp(key, "daemon_sleep")) {
633			rs->print_flags |= DMPF_DAEMON_SLEEP;
634			if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
635				rs->ti->error = "daemon sleep period out of range";
636				return -EINVAL;
637			}
638			rs->md.bitmap_info.daemon_sleep = value;
639		} else if (!strcasecmp(key, "stripe_cache")) {
640			rs->print_flags |= DMPF_STRIPE_CACHE;
641
642			/*
643			 * In device-mapper, we specify things in sectors, but
644			 * MD records this value in kB
645			 */
646			value /= 2;
647
648			if ((rs->raid_type->level != 5) &&
649			    (rs->raid_type->level != 6)) {
650				rs->ti->error = "Inappropriate argument: stripe_cache";
651				return -EINVAL;
652			}
653			if (raid5_set_cache_size(&rs->md, (int)value)) {
654				rs->ti->error = "Bad stripe_cache size";
655				return -EINVAL;
656			}
657		} else if (!strcasecmp(key, "min_recovery_rate")) {
658			rs->print_flags |= DMPF_MIN_RECOVERY_RATE;
659			if (value > INT_MAX) {
660				rs->ti->error = "min_recovery_rate out of range";
661				return -EINVAL;
662			}
663			rs->md.sync_speed_min = (int)value;
664		} else if (!strcasecmp(key, "max_recovery_rate")) {
665			rs->print_flags |= DMPF_MAX_RECOVERY_RATE;
666			if (value > INT_MAX) {
667				rs->ti->error = "max_recovery_rate out of range";
668				return -EINVAL;
669			}
670			rs->md.sync_speed_max = (int)value;
671		} else if (!strcasecmp(key, "region_size")) {
672			rs->print_flags |= DMPF_REGION_SIZE;
673			region_size = value;
674		} else if (!strcasecmp(key, "raid10_copies") &&
675			   (rs->raid_type->level == 10)) {
676			if ((value < 2) || (value > 0xFF)) {
677				rs->ti->error = "Bad value for 'raid10_copies'";
678				return -EINVAL;
679			}
680			rs->print_flags |= DMPF_RAID10_COPIES;
681			raid10_copies = value;
682		} else {
683			DMERR("Unable to parse RAID parameter: %s", key);
684			rs->ti->error = "Unable to parse RAID parameters";
685			return -EINVAL;
686		}
687	}
688
689	if (validate_region_size(rs, region_size))
690		return -EINVAL;
691
692	if (rs->md.chunk_sectors)
693		max_io_len = rs->md.chunk_sectors;
694	else
695		max_io_len = region_size;
696
697	if (dm_set_target_max_io_len(rs->ti, max_io_len))
698		return -EINVAL;
699
700	if (rs->raid_type->level == 10) {
701		if (raid10_copies > rs->md.raid_disks) {
702			rs->ti->error = "Not enough devices to satisfy specification";
703			return -EINVAL;
704		}
705
706		/*
707		 * If the format is not "near", we only support
708		 * two copies at the moment.
709		 */
710		if (strcmp("near", raid10_format) && (raid10_copies > 2)) {
711			rs->ti->error = "Too many copies for given RAID10 format.";
712			return -EINVAL;
713		}
714
715		/* (Len * #mirrors) / #devices */
716		sectors_per_dev = rs->ti->len * raid10_copies;
717		sector_div(sectors_per_dev, rs->md.raid_disks);
718
719		rs->md.layout = raid10_format_to_md_layout(raid10_format,
720							   raid10_copies);
721		rs->md.new_layout = rs->md.layout;
722	} else if ((rs->raid_type->level > 1) &&
723		   sector_div(sectors_per_dev,
724			      (rs->md.raid_disks - rs->raid_type->parity_devs))) {
725		rs->ti->error = "Target length not divisible by number of data devices";
726		return -EINVAL;
727	}
728	rs->md.dev_sectors = sectors_per_dev;
729
730	/* Assume there are no metadata devices until the drives are parsed */
731	rs->md.persistent = 0;
732	rs->md.external = 1;
733
734	return 0;
735}
736
737static void do_table_event(struct work_struct *ws)
738{
739	struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
740
741	dm_table_event(rs->ti->table);
742}
743
744static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
745{
746	struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
747
748	return mddev_congested(&rs->md, bits);
749}
750
751/*
752 * This structure is never routinely used by userspace, unlike md superblocks.
753 * Devices with this superblock should only ever be accessed via device-mapper.
754 */
755#define DM_RAID_MAGIC 0x64526D44
756struct dm_raid_superblock {
757	__le32 magic;		/* "DmRd" */
758	__le32 features;	/* Used to indicate possible future changes */
759
760	__le32 num_devices;	/* Number of devices in this array. (Max 64) */
761	__le32 array_position;	/* The position of this drive in the array */
762
763	__le64 events;		/* Incremented by md when superblock updated */
764	__le64 failed_devices;	/* Bit field of devices to indicate failures */
765
766	/*
767	 * This offset tracks the progress of the repair or replacement of
768	 * an individual drive.
769	 */
770	__le64 disk_recovery_offset;
771
772	/*
773	 * This offset tracks the progress of the initial array
774	 * synchronisation/parity calculation.
775	 */
776	__le64 array_resync_offset;
777
778	/*
779	 * RAID characteristics
780	 */
781	__le32 level;
782	__le32 layout;
783	__le32 stripe_sectors;
784
785	/* Remainder of a logical block is zero-filled when writing (see super_sync()). */
786} __packed;
787
788static int read_disk_sb(struct md_rdev *rdev, int size)
789{
790	BUG_ON(!rdev->sb_page);
791
792	if (rdev->sb_loaded)
793		return 0;
794
795	if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, 1)) {
796		DMERR("Failed to read superblock of device at position %d",
797		      rdev->raid_disk);
798		md_error(rdev->mddev, rdev);
799		return -EINVAL;
800	}
801
802	rdev->sb_loaded = 1;
803
804	return 0;
805}
806
807static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
808{
809	int i;
810	uint64_t failed_devices;
811	struct dm_raid_superblock *sb;
812	struct raid_set *rs = container_of(mddev, struct raid_set, md);
813
814	sb = page_address(rdev->sb_page);
815	failed_devices = le64_to_cpu(sb->failed_devices);
816
817	for (i = 0; i < mddev->raid_disks; i++)
818		if (!rs->dev[i].data_dev ||
819		    test_bit(Faulty, &(rs->dev[i].rdev.flags)))
820			failed_devices |= (1ULL << i);
821
822	memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
823
824	sb->magic = cpu_to_le32(DM_RAID_MAGIC);
825	sb->features = cpu_to_le32(0);	/* No features yet */
826
827	sb->num_devices = cpu_to_le32(mddev->raid_disks);
828	sb->array_position = cpu_to_le32(rdev->raid_disk);
829
830	sb->events = cpu_to_le64(mddev->events);
831	sb->failed_devices = cpu_to_le64(failed_devices);
832
833	sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
834	sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
835
836	sb->level = cpu_to_le32(mddev->level);
837	sb->layout = cpu_to_le32(mddev->layout);
838	sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
839}
840
841/*
842 * super_load
843 *
844 * This function creates a superblock if one is not found on the device
845 * and will decide which superblock to use if there's a choice.
846 *
847 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
848 */
849static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
850{
851	int ret;
852	struct dm_raid_superblock *sb;
853	struct dm_raid_superblock *refsb;
854	uint64_t events_sb, events_refsb;
855
856	rdev->sb_start = 0;
857	rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
858	if (rdev->sb_size < sizeof(*sb) || rdev->sb_size > PAGE_SIZE) {
859		DMERR("superblock size of a logical block is no longer valid");
860		return -EINVAL;
861	}
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
898static 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
1028static 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 */
1065static 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 * Enable/disable discard support on RAID set depending on
1154 * RAID level and discard properties of underlying RAID members.
1155 */
1156static void configure_discard_support(struct dm_target *ti, struct raid_set *rs)
1157{
1158	int i;
1159	bool raid456;
1160
1161	/* Assume discards not supported until after checks below. */
1162	ti->discards_supported = false;
1163
1164	/* RAID level 4,5,6 require discard_zeroes_data for data integrity! */
1165	raid456 = (rs->md.level == 4 || rs->md.level == 5 || rs->md.level == 6);
1166
1167	for (i = 0; i < rs->md.raid_disks; i++) {
1168		struct request_queue *q;
1169
1170		if (!rs->dev[i].rdev.bdev)
1171			continue;
1172
1173		q = bdev_get_queue(rs->dev[i].rdev.bdev);
1174		if (!q || !blk_queue_discard(q))
1175			return;
1176
1177		if (raid456) {
1178			if (!q->limits.discard_zeroes_data)
1179				return;
1180			if (!devices_handle_discard_safely) {
1181				DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
1182				DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
1183				return;
1184			}
1185		}
1186	}
1187
1188	/* All RAID members properly support discards */
1189	ti->discards_supported = true;
1190
1191	/*
1192	 * RAID1 and RAID10 personalities require bio splitting,
1193	 * RAID0/4/5/6 don't and process large discard bios properly.
1194	 */
1195	ti->split_discard_bios = !!(rs->md.level == 1 || rs->md.level == 10);
1196	ti->num_discard_bios = 1;
1197}
1198
1199/*
1200 * Construct a RAID4/5/6 mapping:
1201 * Args:
1202 *	<raid_type> <#raid_params> <raid_params>		\
1203 *	<#raid_devs> { <meta_dev1> <dev1> .. <meta_devN> <devN> }
1204 *
1205 * <raid_params> varies by <raid_type>.  See 'parse_raid_params' for
1206 * details on possible <raid_params>.
1207 */
1208static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv)
1209{
1210	int ret;
1211	struct raid_type *rt;
1212	unsigned long num_raid_params, num_raid_devs;
1213	struct raid_set *rs = NULL;
1214
1215	/* Must have at least <raid_type> <#raid_params> */
1216	if (argc < 2) {
1217		ti->error = "Too few arguments";
1218		return -EINVAL;
1219	}
1220
1221	/* raid type */
1222	rt = get_raid_type(argv[0]);
1223	if (!rt) {
1224		ti->error = "Unrecognised raid_type";
1225		return -EINVAL;
1226	}
1227	argc--;
1228	argv++;
1229
1230	/* number of RAID parameters */
1231	if (kstrtoul(argv[0], 10, &num_raid_params) < 0) {
1232		ti->error = "Cannot understand number of RAID parameters";
1233		return -EINVAL;
1234	}
1235	argc--;
1236	argv++;
1237
1238	/* Skip over RAID params for now and find out # of devices */
1239	if (num_raid_params >= argc) {
1240		ti->error = "Arguments do not agree with counts given";
1241		return -EINVAL;
1242	}
1243
1244	if ((kstrtoul(argv[num_raid_params], 10, &num_raid_devs) < 0) ||
1245	    (num_raid_devs >= INT_MAX)) {
1246		ti->error = "Cannot understand number of raid devices";
1247		return -EINVAL;
1248	}
1249
1250	argc -= num_raid_params + 1; /* +1: we already have num_raid_devs */
1251	if (argc != (num_raid_devs * 2)) {
1252		ti->error = "Supplied RAID devices does not match the count given";
1253		return -EINVAL;
1254	}
1255
1256	rs = context_alloc(ti, rt, (unsigned)num_raid_devs);
1257	if (IS_ERR(rs))
1258		return PTR_ERR(rs);
1259
1260	ret = parse_raid_params(rs, argv, (unsigned)num_raid_params);
1261	if (ret)
1262		goto bad;
1263
1264	argv += num_raid_params + 1;
1265
1266	ret = dev_parms(rs, argv);
1267	if (ret)
1268		goto bad;
1269
1270	rs->md.sync_super = super_sync;
1271	ret = analyse_superblocks(ti, rs);
1272	if (ret)
1273		goto bad;
1274
1275	INIT_WORK(&rs->md.event_work, do_table_event);
1276	ti->private = rs;
1277	ti->num_flush_bios = 1;
1278
1279	/*
1280	 * Disable/enable discard support on RAID set.
1281	 */
1282	configure_discard_support(ti, rs);
1283
1284	mutex_lock(&rs->md.reconfig_mutex);
1285	ret = md_run(&rs->md);
1286	rs->md.in_sync = 0; /* Assume already marked dirty */
1287	mutex_unlock(&rs->md.reconfig_mutex);
1288
1289	if (ret) {
1290		ti->error = "Fail to run raid array";
1291		goto bad;
1292	}
1293
1294	if (ti->len != rs->md.array_sectors) {
1295		ti->error = "Array size does not match requested target length";
1296		ret = -EINVAL;
1297		goto size_mismatch;
1298	}
1299	rs->callbacks.congested_fn = raid_is_congested;
1300	dm_table_add_target_callbacks(ti->table, &rs->callbacks);
1301
1302	mddev_suspend(&rs->md);
1303	return 0;
1304
1305size_mismatch:
1306	md_stop(&rs->md);
1307bad:
1308	context_free(rs);
1309
1310	return ret;
1311}
1312
1313static void raid_dtr(struct dm_target *ti)
1314{
1315	struct raid_set *rs = ti->private;
1316
1317	list_del_init(&rs->callbacks.list);
1318	md_stop(&rs->md);
1319	context_free(rs);
1320}
1321
1322static int raid_map(struct dm_target *ti, struct bio *bio)
1323{
1324	struct raid_set *rs = ti->private;
1325	struct mddev *mddev = &rs->md;
1326
1327	mddev->pers->make_request(mddev, bio);
1328
1329	return DM_MAPIO_SUBMITTED;
1330}
1331
1332static const char *decipher_sync_action(struct mddev *mddev)
1333{
1334	if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
1335		return "frozen";
1336
1337	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1338	    (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
1339		if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
1340			return "reshape";
1341
1342		if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1343			if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1344				return "resync";
1345			else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1346				return "check";
1347			return "repair";
1348		}
1349
1350		if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
1351			return "recover";
1352	}
1353
1354	return "idle";
1355}
1356
1357static void raid_status(struct dm_target *ti, status_type_t type,
1358			unsigned status_flags, char *result, unsigned maxlen)
1359{
1360	struct raid_set *rs = ti->private;
1361	unsigned raid_param_cnt = 1; /* at least 1 for chunksize */
1362	unsigned sz = 0;
1363	int i, array_in_sync = 0;
1364	sector_t sync;
1365
1366	switch (type) {
1367	case STATUSTYPE_INFO:
1368		DMEMIT("%s %d ", rs->raid_type->name, rs->md.raid_disks);
1369
1370		if (test_bit(MD_RECOVERY_RUNNING, &rs->md.recovery))
1371			sync = rs->md.curr_resync_completed;
1372		else
1373			sync = rs->md.recovery_cp;
1374
1375		if (sync >= rs->md.resync_max_sectors) {
1376			/*
1377			 * Sync complete.
1378			 */
1379			array_in_sync = 1;
1380			sync = rs->md.resync_max_sectors;
1381		} else if (test_bit(MD_RECOVERY_REQUESTED, &rs->md.recovery)) {
1382			/*
1383			 * If "check" or "repair" is occurring, the array has
1384			 * undergone and initial sync and the health characters
1385			 * should not be 'a' anymore.
1386			 */
1387			array_in_sync = 1;
1388		} else {
1389			/*
1390			 * The array may be doing an initial sync, or it may
1391			 * be rebuilding individual components.  If all the
1392			 * devices are In_sync, then it is the array that is
1393			 * being initialized.
1394			 */
1395			for (i = 0; i < rs->md.raid_disks; i++)
1396				if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
1397					array_in_sync = 1;
1398		}
1399
1400		/*
1401		 * Status characters:
1402		 *  'D' = Dead/Failed device
1403		 *  'a' = Alive but not in-sync
1404		 *  'A' = Alive and in-sync
1405		 */
1406		for (i = 0; i < rs->md.raid_disks; i++) {
1407			if (test_bit(Faulty, &rs->dev[i].rdev.flags))
1408				DMEMIT("D");
1409			else if (!array_in_sync ||
1410				 !test_bit(In_sync, &rs->dev[i].rdev.flags))
1411				DMEMIT("a");
1412			else
1413				DMEMIT("A");
1414		}
1415
1416		/*
1417		 * In-sync ratio:
1418		 *  The in-sync ratio shows the progress of:
1419		 *   - Initializing the array
1420		 *   - Rebuilding a subset of devices of the array
1421		 *  The user can distinguish between the two by referring
1422		 *  to the status characters.
1423		 */
1424		DMEMIT(" %llu/%llu",
1425		       (unsigned long long) sync,
1426		       (unsigned long long) rs->md.resync_max_sectors);
1427
1428		/*
1429		 * Sync action:
1430		 *   See Documentation/device-mapper/dm-raid.c for
1431		 *   information on each of these states.
1432		 */
1433		DMEMIT(" %s", decipher_sync_action(&rs->md));
1434
1435		/*
1436		 * resync_mismatches/mismatch_cnt
1437		 *   This field shows the number of discrepancies found when
1438		 *   performing a "check" of the array.
1439		 */
1440		DMEMIT(" %llu",
1441		       (strcmp(rs->md.last_sync_action, "check")) ? 0 :
1442		       (unsigned long long)
1443		       atomic64_read(&rs->md.resync_mismatches));
1444		break;
1445	case STATUSTYPE_TABLE:
1446		/* The string you would use to construct this array */
1447		for (i = 0; i < rs->md.raid_disks; i++) {
1448			if ((rs->print_flags & DMPF_REBUILD) &&
1449			    rs->dev[i].data_dev &&
1450			    !test_bit(In_sync, &rs->dev[i].rdev.flags))
1451				raid_param_cnt += 2; /* for rebuilds */
1452			if (rs->dev[i].data_dev &&
1453			    test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1454				raid_param_cnt += 2;
1455		}
1456
1457		raid_param_cnt += (hweight32(rs->print_flags & ~DMPF_REBUILD) * 2);
1458		if (rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC))
1459			raid_param_cnt--;
1460
1461		DMEMIT("%s %u %u", rs->raid_type->name,
1462		       raid_param_cnt, rs->md.chunk_sectors);
1463
1464		if ((rs->print_flags & DMPF_SYNC) &&
1465		    (rs->md.recovery_cp == MaxSector))
1466			DMEMIT(" sync");
1467		if (rs->print_flags & DMPF_NOSYNC)
1468			DMEMIT(" nosync");
1469
1470		for (i = 0; i < rs->md.raid_disks; i++)
1471			if ((rs->print_flags & DMPF_REBUILD) &&
1472			    rs->dev[i].data_dev &&
1473			    !test_bit(In_sync, &rs->dev[i].rdev.flags))
1474				DMEMIT(" rebuild %u", i);
1475
1476		if (rs->print_flags & DMPF_DAEMON_SLEEP)
1477			DMEMIT(" daemon_sleep %lu",
1478			       rs->md.bitmap_info.daemon_sleep);
1479
1480		if (rs->print_flags & DMPF_MIN_RECOVERY_RATE)
1481			DMEMIT(" min_recovery_rate %d", rs->md.sync_speed_min);
1482
1483		if (rs->print_flags & DMPF_MAX_RECOVERY_RATE)
1484			DMEMIT(" max_recovery_rate %d", rs->md.sync_speed_max);
1485
1486		for (i = 0; i < rs->md.raid_disks; i++)
1487			if (rs->dev[i].data_dev &&
1488			    test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1489				DMEMIT(" write_mostly %u", i);
1490
1491		if (rs->print_flags & DMPF_MAX_WRITE_BEHIND)
1492			DMEMIT(" max_write_behind %lu",
1493			       rs->md.bitmap_info.max_write_behind);
1494
1495		if (rs->print_flags & DMPF_STRIPE_CACHE) {
1496			struct r5conf *conf = rs->md.private;
1497
1498			/* convert from kiB to sectors */
1499			DMEMIT(" stripe_cache %d",
1500			       conf ? conf->max_nr_stripes * 2 : 0);
1501		}
1502
1503		if (rs->print_flags & DMPF_REGION_SIZE)
1504			DMEMIT(" region_size %lu",
1505			       rs->md.bitmap_info.chunksize >> 9);
1506
1507		if (rs->print_flags & DMPF_RAID10_COPIES)
1508			DMEMIT(" raid10_copies %u",
1509			       raid10_md_layout_to_copies(rs->md.layout));
1510
1511		if (rs->print_flags & DMPF_RAID10_FORMAT)
1512			DMEMIT(" raid10_format %s",
1513			       raid10_md_layout_to_format(rs->md.layout));
1514
1515		DMEMIT(" %d", rs->md.raid_disks);
1516		for (i = 0; i < rs->md.raid_disks; i++) {
1517			if (rs->dev[i].meta_dev)
1518				DMEMIT(" %s", rs->dev[i].meta_dev->name);
1519			else
1520				DMEMIT(" -");
1521
1522			if (rs->dev[i].data_dev)
1523				DMEMIT(" %s", rs->dev[i].data_dev->name);
1524			else
1525				DMEMIT(" -");
1526		}
1527	}
1528}
1529
1530static int raid_message(struct dm_target *ti, unsigned argc, char **argv)
1531{
1532	struct raid_set *rs = ti->private;
1533	struct mddev *mddev = &rs->md;
1534
1535	if (!strcasecmp(argv[0], "reshape")) {
1536		DMERR("Reshape not supported.");
1537		return -EINVAL;
1538	}
1539
1540	if (!mddev->pers || !mddev->pers->sync_request)
1541		return -EINVAL;
1542
1543	if (!strcasecmp(argv[0], "frozen"))
1544		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1545	else
1546		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1547
1548	if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
1549		if (mddev->sync_thread) {
1550			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1551			md_reap_sync_thread(mddev);
1552		}
1553	} else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1554		   test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1555		return -EBUSY;
1556	else if (!strcasecmp(argv[0], "resync"))
1557		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1558	else if (!strcasecmp(argv[0], "recover")) {
1559		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
1560		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1561	} else {
1562		if (!strcasecmp(argv[0], "check"))
1563			set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1564		else if (!!strcasecmp(argv[0], "repair"))
1565			return -EINVAL;
1566		set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1567		set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1568	}
1569	if (mddev->ro == 2) {
1570		/* A write to sync_action is enough to justify
1571		 * canceling read-auto mode
1572		 */
1573		mddev->ro = 0;
1574		if (!mddev->suspended)
1575			md_wakeup_thread(mddev->sync_thread);
1576	}
1577	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1578	if (!mddev->suspended)
1579		md_wakeup_thread(mddev->thread);
1580
1581	return 0;
1582}
1583
1584static int raid_iterate_devices(struct dm_target *ti,
1585				iterate_devices_callout_fn fn, void *data)
1586{
1587	struct raid_set *rs = ti->private;
1588	unsigned i;
1589	int ret = 0;
1590
1591	for (i = 0; !ret && i < rs->md.raid_disks; i++)
1592		if (rs->dev[i].data_dev)
1593			ret = fn(ti,
1594				 rs->dev[i].data_dev,
1595				 0, /* No offset on data devs */
1596				 rs->md.dev_sectors,
1597				 data);
1598
1599	return ret;
1600}
1601
1602static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
1603{
1604	struct raid_set *rs = ti->private;
1605	unsigned chunk_size = rs->md.chunk_sectors << 9;
1606	struct r5conf *conf = rs->md.private;
1607
1608	blk_limits_io_min(limits, chunk_size);
1609	blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded));
1610}
1611
1612static void raid_presuspend(struct dm_target *ti)
1613{
1614	struct raid_set *rs = ti->private;
1615
1616	md_stop_writes(&rs->md);
1617}
1618
1619static void raid_postsuspend(struct dm_target *ti)
1620{
1621	struct raid_set *rs = ti->private;
1622
1623	mddev_suspend(&rs->md);
1624}
1625
1626static void attempt_restore_of_faulty_devices(struct raid_set *rs)
1627{
1628	int i;
1629	uint64_t failed_devices, cleared_failed_devices = 0;
1630	unsigned long flags;
1631	struct dm_raid_superblock *sb;
1632	struct md_rdev *r;
1633
1634	for (i = 0; i < rs->md.raid_disks; i++) {
1635		r = &rs->dev[i].rdev;
1636		if (test_bit(Faulty, &r->flags) && r->sb_page &&
1637		    sync_page_io(r, 0, r->sb_size, r->sb_page, READ, 1)) {
1638			DMINFO("Faulty %s device #%d has readable super block."
1639			       "  Attempting to revive it.",
1640			       rs->raid_type->name, i);
1641
1642			/*
1643			 * Faulty bit may be set, but sometimes the array can
1644			 * be suspended before the personalities can respond
1645			 * by removing the device from the array (i.e. calling
1646			 * 'hot_remove_disk').  If they haven't yet removed
1647			 * the failed device, its 'raid_disk' number will be
1648			 * '>= 0' - meaning we must call this function
1649			 * ourselves.
1650			 */
1651			if ((r->raid_disk >= 0) &&
1652			    (r->mddev->pers->hot_remove_disk(r->mddev, r) != 0))
1653				/* Failed to revive this device, try next */
1654				continue;
1655
1656			r->raid_disk = i;
1657			r->saved_raid_disk = i;
1658			flags = r->flags;
1659			clear_bit(Faulty, &r->flags);
1660			clear_bit(WriteErrorSeen, &r->flags);
1661			clear_bit(In_sync, &r->flags);
1662			if (r->mddev->pers->hot_add_disk(r->mddev, r)) {
1663				r->raid_disk = -1;
1664				r->saved_raid_disk = -1;
1665				r->flags = flags;
1666			} else {
1667				r->recovery_offset = 0;
1668				cleared_failed_devices |= 1 << i;
1669			}
1670		}
1671	}
1672	if (cleared_failed_devices) {
1673		rdev_for_each(r, &rs->md) {
1674			sb = page_address(r->sb_page);
1675			failed_devices = le64_to_cpu(sb->failed_devices);
1676			failed_devices &= ~cleared_failed_devices;
1677			sb->failed_devices = cpu_to_le64(failed_devices);
1678		}
1679	}
1680}
1681
1682static void raid_resume(struct dm_target *ti)
1683{
1684	struct raid_set *rs = ti->private;
1685
1686	set_bit(MD_CHANGE_DEVS, &rs->md.flags);
1687	if (!rs->bitmap_loaded) {
1688		bitmap_load(&rs->md);
1689		rs->bitmap_loaded = 1;
1690	} else {
1691		/*
1692		 * A secondary resume while the device is active.
1693		 * Take this opportunity to check whether any failed
1694		 * devices are reachable again.
1695		 */
1696		attempt_restore_of_faulty_devices(rs);
1697	}
1698
1699	clear_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
1700	mddev_resume(&rs->md);
1701}
1702
1703static struct target_type raid_target = {
1704	.name = "raid",
1705	.version = {1, 6, 0},
1706	.module = THIS_MODULE,
1707	.ctr = raid_ctr,
1708	.dtr = raid_dtr,
1709	.map = raid_map,
1710	.status = raid_status,
1711	.message = raid_message,
1712	.iterate_devices = raid_iterate_devices,
1713	.io_hints = raid_io_hints,
1714	.presuspend = raid_presuspend,
1715	.postsuspend = raid_postsuspend,
1716	.resume = raid_resume,
1717};
1718
1719static int __init dm_raid_init(void)
1720{
1721	DMINFO("Loading target version %u.%u.%u",
1722	       raid_target.version[0],
1723	       raid_target.version[1],
1724	       raid_target.version[2]);
1725	return dm_register_target(&raid_target);
1726}
1727
1728static void __exit dm_raid_exit(void)
1729{
1730	dm_unregister_target(&raid_target);
1731}
1732
1733module_init(dm_raid_init);
1734module_exit(dm_raid_exit);
1735
1736module_param(devices_handle_discard_safely, bool, 0644);
1737MODULE_PARM_DESC(devices_handle_discard_safely,
1738		 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
1739
1740MODULE_DESCRIPTION(DM_NAME " raid4/5/6 target");
1741MODULE_ALIAS("dm-raid1");
1742MODULE_ALIAS("dm-raid10");
1743MODULE_ALIAS("dm-raid4");
1744MODULE_ALIAS("dm-raid5");
1745MODULE_ALIAS("dm-raid6");
1746MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
1747MODULE_LICENSE("GPL");
1748