xref: /openbmc/linux/drivers/md/dm-raid.c (revision 6dfcd296)
1 /*
2  * Copyright (C) 2010-2011 Neil Brown
3  * Copyright (C) 2010-2016 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 #define	MAX_RAID_DEVICES	253 /* md-raid kernel limit */
21 
22 /*
23  * Minimum sectors of free reshape space per raid device
24  */
25 #define	MIN_FREE_RESHAPE_SPACE to_sector(4*4096)
26 
27 static bool devices_handle_discard_safely = false;
28 
29 /*
30  * The following flags are used by dm-raid.c to set up the array state.
31  * They must be cleared before md_run is called.
32  */
33 #define FirstUse 10		/* rdev flag */
34 
35 struct raid_dev {
36 	/*
37 	 * Two DM devices, one to hold metadata and one to hold the
38 	 * actual data/parity.	The reason for this is to not confuse
39 	 * ti->len and give more flexibility in altering size and
40 	 * characteristics.
41 	 *
42 	 * While it is possible for this device to be associated
43 	 * with a different physical device than the data_dev, it
44 	 * is intended for it to be the same.
45 	 *    |--------- Physical Device ---------|
46 	 *    |- meta_dev -|------ data_dev ------|
47 	 */
48 	struct dm_dev *meta_dev;
49 	struct dm_dev *data_dev;
50 	struct md_rdev rdev;
51 };
52 
53 /*
54  * Bits for establishing rs->ctr_flags
55  *
56  * 1 = no flag value
57  * 2 = flag with value
58  */
59 #define __CTR_FLAG_SYNC			0  /* 1 */ /* Not with raid0! */
60 #define __CTR_FLAG_NOSYNC		1  /* 1 */ /* Not with raid0! */
61 #define __CTR_FLAG_REBUILD		2  /* 2 */ /* Not with raid0! */
62 #define __CTR_FLAG_DAEMON_SLEEP		3  /* 2 */ /* Not with raid0! */
63 #define __CTR_FLAG_MIN_RECOVERY_RATE	4  /* 2 */ /* Not with raid0! */
64 #define __CTR_FLAG_MAX_RECOVERY_RATE	5  /* 2 */ /* Not with raid0! */
65 #define __CTR_FLAG_MAX_WRITE_BEHIND	6  /* 2 */ /* Only with raid1! */
66 #define __CTR_FLAG_WRITE_MOSTLY		7  /* 2 */ /* Only with raid1! */
67 #define __CTR_FLAG_STRIPE_CACHE		8  /* 2 */ /* Only with raid4/5/6! */
68 #define __CTR_FLAG_REGION_SIZE		9  /* 2 */ /* Not with raid0! */
69 #define __CTR_FLAG_RAID10_COPIES	10 /* 2 */ /* Only with raid10 */
70 #define __CTR_FLAG_RAID10_FORMAT	11 /* 2 */ /* Only with raid10 */
71 /* New for v1.9.0 */
72 #define __CTR_FLAG_DELTA_DISKS		12 /* 2 */ /* Only with reshapable raid1/4/5/6/10! */
73 #define __CTR_FLAG_DATA_OFFSET		13 /* 2 */ /* Only with reshapable raid4/5/6/10! */
74 #define __CTR_FLAG_RAID10_USE_NEAR_SETS 14 /* 2 */ /* Only with raid10! */
75 
76 /*
77  * Flags for rs->ctr_flags field.
78  */
79 #define CTR_FLAG_SYNC			(1 << __CTR_FLAG_SYNC)
80 #define CTR_FLAG_NOSYNC			(1 << __CTR_FLAG_NOSYNC)
81 #define CTR_FLAG_REBUILD		(1 << __CTR_FLAG_REBUILD)
82 #define CTR_FLAG_DAEMON_SLEEP		(1 << __CTR_FLAG_DAEMON_SLEEP)
83 #define CTR_FLAG_MIN_RECOVERY_RATE	(1 << __CTR_FLAG_MIN_RECOVERY_RATE)
84 #define CTR_FLAG_MAX_RECOVERY_RATE	(1 << __CTR_FLAG_MAX_RECOVERY_RATE)
85 #define CTR_FLAG_MAX_WRITE_BEHIND	(1 << __CTR_FLAG_MAX_WRITE_BEHIND)
86 #define CTR_FLAG_WRITE_MOSTLY		(1 << __CTR_FLAG_WRITE_MOSTLY)
87 #define CTR_FLAG_STRIPE_CACHE		(1 << __CTR_FLAG_STRIPE_CACHE)
88 #define CTR_FLAG_REGION_SIZE		(1 << __CTR_FLAG_REGION_SIZE)
89 #define CTR_FLAG_RAID10_COPIES		(1 << __CTR_FLAG_RAID10_COPIES)
90 #define CTR_FLAG_RAID10_FORMAT		(1 << __CTR_FLAG_RAID10_FORMAT)
91 #define CTR_FLAG_DELTA_DISKS		(1 << __CTR_FLAG_DELTA_DISKS)
92 #define CTR_FLAG_DATA_OFFSET		(1 << __CTR_FLAG_DATA_OFFSET)
93 #define CTR_FLAG_RAID10_USE_NEAR_SETS	(1 << __CTR_FLAG_RAID10_USE_NEAR_SETS)
94 
95 /*
96  * Definitions of various constructor flags to
97  * be used in checks of valid / invalid flags
98  * per raid level.
99  */
100 /* Define all any sync flags */
101 #define	CTR_FLAGS_ANY_SYNC		(CTR_FLAG_SYNC | CTR_FLAG_NOSYNC)
102 
103 /* Define flags for options without argument (e.g. 'nosync') */
104 #define	CTR_FLAG_OPTIONS_NO_ARGS	(CTR_FLAGS_ANY_SYNC | \
105 					 CTR_FLAG_RAID10_USE_NEAR_SETS)
106 
107 /* Define flags for options with one argument (e.g. 'delta_disks +2') */
108 #define CTR_FLAG_OPTIONS_ONE_ARG (CTR_FLAG_REBUILD | \
109 				  CTR_FLAG_WRITE_MOSTLY | \
110 				  CTR_FLAG_DAEMON_SLEEP | \
111 				  CTR_FLAG_MIN_RECOVERY_RATE | \
112 				  CTR_FLAG_MAX_RECOVERY_RATE | \
113 				  CTR_FLAG_MAX_WRITE_BEHIND | \
114 				  CTR_FLAG_STRIPE_CACHE | \
115 				  CTR_FLAG_REGION_SIZE | \
116 				  CTR_FLAG_RAID10_COPIES | \
117 				  CTR_FLAG_RAID10_FORMAT | \
118 				  CTR_FLAG_DELTA_DISKS | \
119 				  CTR_FLAG_DATA_OFFSET)
120 
121 /* Valid options definitions per raid level... */
122 
123 /* "raid0" does only accept data offset */
124 #define RAID0_VALID_FLAGS	(CTR_FLAG_DATA_OFFSET)
125 
126 /* "raid1" does not accept stripe cache, data offset, delta_disks or any raid10 options */
127 #define RAID1_VALID_FLAGS	(CTR_FLAGS_ANY_SYNC | \
128 				 CTR_FLAG_REBUILD | \
129 				 CTR_FLAG_WRITE_MOSTLY | \
130 				 CTR_FLAG_DAEMON_SLEEP | \
131 				 CTR_FLAG_MIN_RECOVERY_RATE | \
132 				 CTR_FLAG_MAX_RECOVERY_RATE | \
133 				 CTR_FLAG_MAX_WRITE_BEHIND | \
134 				 CTR_FLAG_REGION_SIZE | \
135 				 CTR_FLAG_DELTA_DISKS | \
136 				 CTR_FLAG_DATA_OFFSET)
137 
138 /* "raid10" does not accept any raid1 or stripe cache options */
139 #define RAID10_VALID_FLAGS	(CTR_FLAGS_ANY_SYNC | \
140 				 CTR_FLAG_REBUILD | \
141 				 CTR_FLAG_DAEMON_SLEEP | \
142 				 CTR_FLAG_MIN_RECOVERY_RATE | \
143 				 CTR_FLAG_MAX_RECOVERY_RATE | \
144 				 CTR_FLAG_REGION_SIZE | \
145 				 CTR_FLAG_RAID10_COPIES | \
146 				 CTR_FLAG_RAID10_FORMAT | \
147 				 CTR_FLAG_DELTA_DISKS | \
148 				 CTR_FLAG_DATA_OFFSET | \
149 				 CTR_FLAG_RAID10_USE_NEAR_SETS)
150 
151 /*
152  * "raid4/5/6" do not accept any raid1 or raid10 specific options
153  *
154  * "raid6" does not accept "nosync", because it is not guaranteed
155  * that both parity and q-syndrome are being written properly with
156  * any writes
157  */
158 #define RAID45_VALID_FLAGS	(CTR_FLAGS_ANY_SYNC | \
159 				 CTR_FLAG_REBUILD | \
160 				 CTR_FLAG_DAEMON_SLEEP | \
161 				 CTR_FLAG_MIN_RECOVERY_RATE | \
162 				 CTR_FLAG_MAX_RECOVERY_RATE | \
163 				 CTR_FLAG_MAX_WRITE_BEHIND | \
164 				 CTR_FLAG_STRIPE_CACHE | \
165 				 CTR_FLAG_REGION_SIZE | \
166 				 CTR_FLAG_DELTA_DISKS | \
167 				 CTR_FLAG_DATA_OFFSET)
168 
169 #define RAID6_VALID_FLAGS	(CTR_FLAG_SYNC | \
170 				 CTR_FLAG_REBUILD | \
171 				 CTR_FLAG_DAEMON_SLEEP | \
172 				 CTR_FLAG_MIN_RECOVERY_RATE | \
173 				 CTR_FLAG_MAX_RECOVERY_RATE | \
174 				 CTR_FLAG_MAX_WRITE_BEHIND | \
175 				 CTR_FLAG_STRIPE_CACHE | \
176 				 CTR_FLAG_REGION_SIZE | \
177 				 CTR_FLAG_DELTA_DISKS | \
178 				 CTR_FLAG_DATA_OFFSET)
179 /* ...valid options definitions per raid level */
180 
181 /*
182  * Flags for rs->runtime_flags field
183  * (RT_FLAG prefix meaning "runtime flag")
184  *
185  * These are all internal and used to define runtime state,
186  * e.g. to prevent another resume from preresume processing
187  * the raid set all over again.
188  */
189 #define RT_FLAG_RS_PRERESUMED		0
190 #define RT_FLAG_RS_RESUMED		1
191 #define RT_FLAG_RS_BITMAP_LOADED	2
192 #define RT_FLAG_UPDATE_SBS		3
193 #define RT_FLAG_RESHAPE_RS		4
194 
195 /* Array elements of 64 bit needed for rebuild/failed disk bits */
196 #define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
197 
198 /*
199  * raid set level, layout and chunk sectors backup/restore
200  */
201 struct rs_layout {
202 	int new_level;
203 	int new_layout;
204 	int new_chunk_sectors;
205 };
206 
207 struct raid_set {
208 	struct dm_target *ti;
209 
210 	uint32_t bitmap_loaded;
211 	uint32_t stripe_cache_entries;
212 	unsigned long ctr_flags;
213 	unsigned long runtime_flags;
214 
215 	uint64_t rebuild_disks[DISKS_ARRAY_ELEMS];
216 
217 	int raid_disks;
218 	int delta_disks;
219 	int data_offset;
220 	int raid10_copies;
221 	int requested_bitmap_chunk_sectors;
222 
223 	struct mddev md;
224 	struct raid_type *raid_type;
225 	struct dm_target_callbacks callbacks;
226 
227 	struct raid_dev dev[0];
228 };
229 
230 static void rs_config_backup(struct raid_set *rs, struct rs_layout *l)
231 {
232 	struct mddev *mddev = &rs->md;
233 
234 	l->new_level = mddev->new_level;
235 	l->new_layout = mddev->new_layout;
236 	l->new_chunk_sectors = mddev->new_chunk_sectors;
237 }
238 
239 static void rs_config_restore(struct raid_set *rs, struct rs_layout *l)
240 {
241 	struct mddev *mddev = &rs->md;
242 
243 	mddev->new_level = l->new_level;
244 	mddev->new_layout = l->new_layout;
245 	mddev->new_chunk_sectors = l->new_chunk_sectors;
246 }
247 
248 /* raid10 algorithms (i.e. formats) */
249 #define	ALGORITHM_RAID10_DEFAULT	0
250 #define	ALGORITHM_RAID10_NEAR		1
251 #define	ALGORITHM_RAID10_OFFSET		2
252 #define	ALGORITHM_RAID10_FAR		3
253 
254 /* Supported raid types and properties. */
255 static struct raid_type {
256 	const char *name;		/* RAID algorithm. */
257 	const char *descr;		/* Descriptor text for logging. */
258 	const unsigned int parity_devs;	/* # of parity devices. */
259 	const unsigned int minimal_devs;/* minimal # of devices in set. */
260 	const unsigned int level;	/* RAID level. */
261 	const unsigned int algorithm;	/* RAID algorithm. */
262 } raid_types[] = {
263 	{"raid0",	  "raid0 (striping)",			    0, 2, 0,  0 /* NONE */},
264 	{"raid1",	  "raid1 (mirroring)",			    0, 2, 1,  0 /* NONE */},
265 	{"raid10_far",	  "raid10 far (striped mirrors)",	    0, 2, 10, ALGORITHM_RAID10_FAR},
266 	{"raid10_offset", "raid10 offset (striped mirrors)",	    0, 2, 10, ALGORITHM_RAID10_OFFSET},
267 	{"raid10_near",	  "raid10 near (striped mirrors)",	    0, 2, 10, ALGORITHM_RAID10_NEAR},
268 	{"raid10",	  "raid10 (striped mirrors)",		    0, 2, 10, ALGORITHM_RAID10_DEFAULT},
269 	{"raid4",	  "raid4 (dedicated last parity disk)",	    1, 2, 4,  ALGORITHM_PARITY_N}, /* raid4 layout = raid5_n */
270 	{"raid5_n",	  "raid5 (dedicated last parity disk)",	    1, 2, 5,  ALGORITHM_PARITY_N},
271 	{"raid5_ls",	  "raid5 (left symmetric)",		    1, 2, 5,  ALGORITHM_LEFT_SYMMETRIC},
272 	{"raid5_rs",	  "raid5 (right symmetric)",		    1, 2, 5,  ALGORITHM_RIGHT_SYMMETRIC},
273 	{"raid5_la",	  "raid5 (left asymmetric)",		    1, 2, 5,  ALGORITHM_LEFT_ASYMMETRIC},
274 	{"raid5_ra",	  "raid5 (right asymmetric)",		    1, 2, 5,  ALGORITHM_RIGHT_ASYMMETRIC},
275 	{"raid6_zr",	  "raid6 (zero restart)",		    2, 4, 6,  ALGORITHM_ROTATING_ZERO_RESTART},
276 	{"raid6_nr",	  "raid6 (N restart)",			    2, 4, 6,  ALGORITHM_ROTATING_N_RESTART},
277 	{"raid6_nc",	  "raid6 (N continue)",			    2, 4, 6,  ALGORITHM_ROTATING_N_CONTINUE},
278 	{"raid6_n_6",	  "raid6 (dedicated parity/Q n/6)",	    2, 4, 6,  ALGORITHM_PARITY_N_6},
279 	{"raid6_ls_6",	  "raid6 (left symmetric dedicated Q 6)",   2, 4, 6,  ALGORITHM_LEFT_SYMMETRIC_6},
280 	{"raid6_rs_6",	  "raid6 (right symmetric dedicated Q 6)",  2, 4, 6,  ALGORITHM_RIGHT_SYMMETRIC_6},
281 	{"raid6_la_6",	  "raid6 (left asymmetric dedicated Q 6)",  2, 4, 6,  ALGORITHM_LEFT_ASYMMETRIC_6},
282 	{"raid6_ra_6",	  "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6,  ALGORITHM_RIGHT_ASYMMETRIC_6}
283 };
284 
285 /* True, if @v is in inclusive range [@min, @max] */
286 static bool __within_range(long v, long min, long max)
287 {
288 	return v >= min && v <= max;
289 }
290 
291 /* All table line arguments are defined here */
292 static struct arg_name_flag {
293 	const unsigned long flag;
294 	const char *name;
295 } __arg_name_flags[] = {
296 	{ CTR_FLAG_SYNC, "sync"},
297 	{ CTR_FLAG_NOSYNC, "nosync"},
298 	{ CTR_FLAG_REBUILD, "rebuild"},
299 	{ CTR_FLAG_DAEMON_SLEEP, "daemon_sleep"},
300 	{ CTR_FLAG_MIN_RECOVERY_RATE, "min_recovery_rate"},
301 	{ CTR_FLAG_MAX_RECOVERY_RATE, "max_recovery_rate"},
302 	{ CTR_FLAG_MAX_WRITE_BEHIND, "max_write_behind"},
303 	{ CTR_FLAG_WRITE_MOSTLY, "write_mostly"},
304 	{ CTR_FLAG_STRIPE_CACHE, "stripe_cache"},
305 	{ CTR_FLAG_REGION_SIZE, "region_size"},
306 	{ CTR_FLAG_RAID10_COPIES, "raid10_copies"},
307 	{ CTR_FLAG_RAID10_FORMAT, "raid10_format"},
308 	{ CTR_FLAG_DATA_OFFSET, "data_offset"},
309 	{ CTR_FLAG_DELTA_DISKS, "delta_disks"},
310 	{ CTR_FLAG_RAID10_USE_NEAR_SETS, "raid10_use_near_sets"},
311 };
312 
313 /* Return argument name string for given @flag */
314 static const char *dm_raid_arg_name_by_flag(const uint32_t flag)
315 {
316 	if (hweight32(flag) == 1) {
317 		struct arg_name_flag *anf = __arg_name_flags + ARRAY_SIZE(__arg_name_flags);
318 
319 		while (anf-- > __arg_name_flags)
320 			if (flag & anf->flag)
321 				return anf->name;
322 
323 	} else
324 		DMERR("%s called with more than one flag!", __func__);
325 
326 	return NULL;
327 }
328 
329 /*
330  * Bool helpers to test for various raid levels of a raid set.
331  * It's level as reported by the superblock rather than
332  * the requested raid_type passed to the constructor.
333  */
334 /* Return true, if raid set in @rs is raid0 */
335 static bool rs_is_raid0(struct raid_set *rs)
336 {
337 	return !rs->md.level;
338 }
339 
340 /* Return true, if raid set in @rs is raid1 */
341 static bool rs_is_raid1(struct raid_set *rs)
342 {
343 	return rs->md.level == 1;
344 }
345 
346 /* Return true, if raid set in @rs is raid10 */
347 static bool rs_is_raid10(struct raid_set *rs)
348 {
349 	return rs->md.level == 10;
350 }
351 
352 /* Return true, if raid set in @rs is level 6 */
353 static bool rs_is_raid6(struct raid_set *rs)
354 {
355 	return rs->md.level == 6;
356 }
357 
358 /* Return true, if raid set in @rs is level 4, 5 or 6 */
359 static bool rs_is_raid456(struct raid_set *rs)
360 {
361 	return __within_range(rs->md.level, 4, 6);
362 }
363 
364 /* Return true, if raid set in @rs is reshapable */
365 static bool __is_raid10_far(int layout);
366 static bool rs_is_reshapable(struct raid_set *rs)
367 {
368 	return rs_is_raid456(rs) ||
369 	       (rs_is_raid10(rs) && !__is_raid10_far(rs->md.new_layout));
370 }
371 
372 /* Return true, if raid set in @rs is recovering */
373 static bool rs_is_recovering(struct raid_set *rs)
374 {
375 	return rs->md.recovery_cp < rs->dev[0].rdev.sectors;
376 }
377 
378 /* Return true, if raid set in @rs is reshaping */
379 static bool rs_is_reshaping(struct raid_set *rs)
380 {
381 	return rs->md.reshape_position != MaxSector;
382 }
383 
384 /*
385  * bool helpers to test for various raid levels of a raid type @rt
386  */
387 
388 /* Return true, if raid type in @rt is raid0 */
389 static bool rt_is_raid0(struct raid_type *rt)
390 {
391 	return !rt->level;
392 }
393 
394 /* Return true, if raid type in @rt is raid1 */
395 static bool rt_is_raid1(struct raid_type *rt)
396 {
397 	return rt->level == 1;
398 }
399 
400 /* Return true, if raid type in @rt is raid10 */
401 static bool rt_is_raid10(struct raid_type *rt)
402 {
403 	return rt->level == 10;
404 }
405 
406 /* Return true, if raid type in @rt is raid4/5 */
407 static bool rt_is_raid45(struct raid_type *rt)
408 {
409 	return __within_range(rt->level, 4, 5);
410 }
411 
412 /* Return true, if raid type in @rt is raid6 */
413 static bool rt_is_raid6(struct raid_type *rt)
414 {
415 	return rt->level == 6;
416 }
417 
418 /* Return true, if raid type in @rt is raid4/5/6 */
419 static bool rt_is_raid456(struct raid_type *rt)
420 {
421 	return __within_range(rt->level, 4, 6);
422 }
423 /* END: raid level bools */
424 
425 /* Return valid ctr flags for the raid level of @rs */
426 static unsigned long __valid_flags(struct raid_set *rs)
427 {
428 	if (rt_is_raid0(rs->raid_type))
429 		return RAID0_VALID_FLAGS;
430 	else if (rt_is_raid1(rs->raid_type))
431 		return RAID1_VALID_FLAGS;
432 	else if (rt_is_raid10(rs->raid_type))
433 		return RAID10_VALID_FLAGS;
434 	else if (rt_is_raid45(rs->raid_type))
435 		return RAID45_VALID_FLAGS;
436 	else if (rt_is_raid6(rs->raid_type))
437 		return RAID6_VALID_FLAGS;
438 
439 	return 0;
440 }
441 
442 /*
443  * Check for valid flags set on @rs
444  *
445  * Has to be called after parsing of the ctr flags!
446  */
447 static int rs_check_for_valid_flags(struct raid_set *rs)
448 {
449 	if (rs->ctr_flags & ~__valid_flags(rs)) {
450 		rs->ti->error = "Invalid flags combination";
451 		return -EINVAL;
452 	}
453 
454 	return 0;
455 }
456 
457 /* MD raid10 bit definitions and helpers */
458 #define RAID10_OFFSET			(1 << 16) /* stripes with data copies area adjacent on devices */
459 #define RAID10_BROCKEN_USE_FAR_SETS	(1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
460 #define RAID10_USE_FAR_SETS		(1 << 18) /* Use sets instead of whole stripe rotation */
461 #define RAID10_FAR_COPIES_SHIFT		8	  /* raid10 # far copies shift (2nd byte of layout) */
462 
463 /* Return md raid10 near copies for @layout */
464 static unsigned int __raid10_near_copies(int layout)
465 {
466 	return layout & 0xFF;
467 }
468 
469 /* Return md raid10 far copies for @layout */
470 static unsigned int __raid10_far_copies(int layout)
471 {
472 	return __raid10_near_copies(layout >> RAID10_FAR_COPIES_SHIFT);
473 }
474 
475 /* Return true if md raid10 offset for @layout */
476 static bool __is_raid10_offset(int layout)
477 {
478 	return !!(layout & RAID10_OFFSET);
479 }
480 
481 /* Return true if md raid10 near for @layout */
482 static bool __is_raid10_near(int layout)
483 {
484 	return !__is_raid10_offset(layout) && __raid10_near_copies(layout) > 1;
485 }
486 
487 /* Return true if md raid10 far for @layout */
488 static bool __is_raid10_far(int layout)
489 {
490 	return !__is_raid10_offset(layout) && __raid10_far_copies(layout) > 1;
491 }
492 
493 /* Return md raid10 layout string for @layout */
494 static const char *raid10_md_layout_to_format(int layout)
495 {
496 	/*
497 	 * Bit 16 stands for "offset"
498 	 * (i.e. adjacent stripes hold copies)
499 	 *
500 	 * Refer to MD's raid10.c for details
501 	 */
502 	if (__is_raid10_offset(layout))
503 		return "offset";
504 
505 	if (__raid10_near_copies(layout) > 1)
506 		return "near";
507 
508 	WARN_ON(__raid10_far_copies(layout) < 2);
509 
510 	return "far";
511 }
512 
513 /* Return md raid10 algorithm for @name */
514 static int raid10_name_to_format(const char *name)
515 {
516 	if (!strcasecmp(name, "near"))
517 		return ALGORITHM_RAID10_NEAR;
518 	else if (!strcasecmp(name, "offset"))
519 		return ALGORITHM_RAID10_OFFSET;
520 	else if (!strcasecmp(name, "far"))
521 		return ALGORITHM_RAID10_FAR;
522 
523 	return -EINVAL;
524 }
525 
526 /* Return md raid10 copies for @layout */
527 static unsigned int raid10_md_layout_to_copies(int layout)
528 {
529 	return max(__raid10_near_copies(layout), __raid10_far_copies(layout));
530 }
531 
532 /* Return md raid10 format id for @format string */
533 static int raid10_format_to_md_layout(struct raid_set *rs,
534 				      unsigned int algorithm,
535 				      unsigned int copies)
536 {
537 	unsigned int n = 1, f = 1, r = 0;
538 
539 	/*
540 	 * MD resilienece flaw:
541 	 *
542 	 * enabling use_far_sets for far/offset formats causes copies
543 	 * to be colocated on the same devs together with their origins!
544 	 *
545 	 * -> disable it for now in the definition above
546 	 */
547 	if (algorithm == ALGORITHM_RAID10_DEFAULT ||
548 	    algorithm == ALGORITHM_RAID10_NEAR)
549 		n = copies;
550 
551 	else if (algorithm == ALGORITHM_RAID10_OFFSET) {
552 		f = copies;
553 		r = RAID10_OFFSET;
554 		if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
555 			r |= RAID10_USE_FAR_SETS;
556 
557 	} else if (algorithm == ALGORITHM_RAID10_FAR) {
558 		f = copies;
559 		r = !RAID10_OFFSET;
560 		if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
561 			r |= RAID10_USE_FAR_SETS;
562 
563 	} else
564 		return -EINVAL;
565 
566 	return r | (f << RAID10_FAR_COPIES_SHIFT) | n;
567 }
568 /* END: MD raid10 bit definitions and helpers */
569 
570 /* Check for any of the raid10 algorithms */
571 static bool __got_raid10(struct raid_type *rtp, const int layout)
572 {
573 	if (rtp->level == 10) {
574 		switch (rtp->algorithm) {
575 		case ALGORITHM_RAID10_DEFAULT:
576 		case ALGORITHM_RAID10_NEAR:
577 			return __is_raid10_near(layout);
578 		case ALGORITHM_RAID10_OFFSET:
579 			return __is_raid10_offset(layout);
580 		case ALGORITHM_RAID10_FAR:
581 			return __is_raid10_far(layout);
582 		default:
583 			break;
584 		}
585 	}
586 
587 	return false;
588 }
589 
590 /* Return raid_type for @name */
591 static struct raid_type *get_raid_type(const char *name)
592 {
593 	struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
594 
595 	while (rtp-- > raid_types)
596 		if (!strcasecmp(rtp->name, name))
597 			return rtp;
598 
599 	return NULL;
600 }
601 
602 /* Return raid_type for @name based derived from @level and @layout */
603 static struct raid_type *get_raid_type_by_ll(const int level, const int layout)
604 {
605 	struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
606 
607 	while (rtp-- > raid_types) {
608 		/* RAID10 special checks based on @layout flags/properties */
609 		if (rtp->level == level &&
610 		    (__got_raid10(rtp, layout) || rtp->algorithm == layout))
611 			return rtp;
612 	}
613 
614 	return NULL;
615 }
616 
617 /*
618  * Conditionally change bdev capacity of @rs
619  * in case of a disk add/remove reshape
620  */
621 static void rs_set_capacity(struct raid_set *rs)
622 {
623 	struct mddev *mddev = &rs->md;
624 	struct md_rdev *rdev;
625 	struct gendisk *gendisk = dm_disk(dm_table_get_md(rs->ti->table));
626 
627 	/*
628 	 * raid10 sets rdev->sector to the device size, which
629 	 * is unintended in case of out-of-place reshaping
630 	 */
631 	rdev_for_each(rdev, mddev)
632 		rdev->sectors = mddev->dev_sectors;
633 
634 	set_capacity(gendisk, mddev->array_sectors);
635 	revalidate_disk(gendisk);
636 }
637 
638 /*
639  * Set the mddev properties in @rs to the current
640  * ones retrieved from the freshest superblock
641  */
642 static void rs_set_cur(struct raid_set *rs)
643 {
644 	struct mddev *mddev = &rs->md;
645 
646 	mddev->new_level = mddev->level;
647 	mddev->new_layout = mddev->layout;
648 	mddev->new_chunk_sectors = mddev->chunk_sectors;
649 }
650 
651 /*
652  * Set the mddev properties in @rs to the new
653  * ones requested by the ctr
654  */
655 static void rs_set_new(struct raid_set *rs)
656 {
657 	struct mddev *mddev = &rs->md;
658 
659 	mddev->level = mddev->new_level;
660 	mddev->layout = mddev->new_layout;
661 	mddev->chunk_sectors = mddev->new_chunk_sectors;
662 	mddev->raid_disks = rs->raid_disks;
663 	mddev->delta_disks = 0;
664 }
665 
666 static struct raid_set *raid_set_alloc(struct dm_target *ti, struct raid_type *raid_type,
667 				       unsigned int raid_devs)
668 {
669 	unsigned int i;
670 	struct raid_set *rs;
671 
672 	if (raid_devs <= raid_type->parity_devs) {
673 		ti->error = "Insufficient number of devices";
674 		return ERR_PTR(-EINVAL);
675 	}
676 
677 	rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
678 	if (!rs) {
679 		ti->error = "Cannot allocate raid context";
680 		return ERR_PTR(-ENOMEM);
681 	}
682 
683 	mddev_init(&rs->md);
684 
685 	rs->raid_disks = raid_devs;
686 	rs->delta_disks = 0;
687 
688 	rs->ti = ti;
689 	rs->raid_type = raid_type;
690 	rs->stripe_cache_entries = 256;
691 	rs->md.raid_disks = raid_devs;
692 	rs->md.level = raid_type->level;
693 	rs->md.new_level = rs->md.level;
694 	rs->md.layout = raid_type->algorithm;
695 	rs->md.new_layout = rs->md.layout;
696 	rs->md.delta_disks = 0;
697 	rs->md.recovery_cp = MaxSector;
698 
699 	for (i = 0; i < raid_devs; i++)
700 		md_rdev_init(&rs->dev[i].rdev);
701 
702 	/*
703 	 * Remaining items to be initialized by further RAID params:
704 	 *  rs->md.persistent
705 	 *  rs->md.external
706 	 *  rs->md.chunk_sectors
707 	 *  rs->md.new_chunk_sectors
708 	 *  rs->md.dev_sectors
709 	 */
710 
711 	return rs;
712 }
713 
714 static void raid_set_free(struct raid_set *rs)
715 {
716 	int i;
717 
718 	for (i = 0; i < rs->raid_disks; i++) {
719 		if (rs->dev[i].meta_dev)
720 			dm_put_device(rs->ti, rs->dev[i].meta_dev);
721 		md_rdev_clear(&rs->dev[i].rdev);
722 		if (rs->dev[i].data_dev)
723 			dm_put_device(rs->ti, rs->dev[i].data_dev);
724 	}
725 
726 	kfree(rs);
727 }
728 
729 /*
730  * For every device we have two words
731  *  <meta_dev>: meta device name or '-' if missing
732  *  <data_dev>: data device name or '-' if missing
733  *
734  * The following are permitted:
735  *    - -
736  *    - <data_dev>
737  *    <meta_dev> <data_dev>
738  *
739  * The following is not allowed:
740  *    <meta_dev> -
741  *
742  * This code parses those words.  If there is a failure,
743  * the caller must use raid_set_free() to unwind the operations.
744  */
745 static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
746 {
747 	int i;
748 	int rebuild = 0;
749 	int metadata_available = 0;
750 	int r = 0;
751 	const char *arg;
752 
753 	/* Put off the number of raid devices argument to get to dev pairs */
754 	arg = dm_shift_arg(as);
755 	if (!arg)
756 		return -EINVAL;
757 
758 	for (i = 0; i < rs->raid_disks; i++) {
759 		rs->dev[i].rdev.raid_disk = i;
760 
761 		rs->dev[i].meta_dev = NULL;
762 		rs->dev[i].data_dev = NULL;
763 
764 		/*
765 		 * There are no offsets, since there is a separate device
766 		 * for data and metadata.
767 		 */
768 		rs->dev[i].rdev.data_offset = 0;
769 		rs->dev[i].rdev.mddev = &rs->md;
770 
771 		arg = dm_shift_arg(as);
772 		if (!arg)
773 			return -EINVAL;
774 
775 		if (strcmp(arg, "-")) {
776 			r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
777 					  &rs->dev[i].meta_dev);
778 			if (r) {
779 				rs->ti->error = "RAID metadata device lookup failure";
780 				return r;
781 			}
782 
783 			rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
784 			if (!rs->dev[i].rdev.sb_page) {
785 				rs->ti->error = "Failed to allocate superblock page";
786 				return -ENOMEM;
787 			}
788 		}
789 
790 		arg = dm_shift_arg(as);
791 		if (!arg)
792 			return -EINVAL;
793 
794 		if (!strcmp(arg, "-")) {
795 			if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
796 			    (!rs->dev[i].rdev.recovery_offset)) {
797 				rs->ti->error = "Drive designated for rebuild not specified";
798 				return -EINVAL;
799 			}
800 
801 			if (rs->dev[i].meta_dev) {
802 				rs->ti->error = "No data device supplied with metadata device";
803 				return -EINVAL;
804 			}
805 
806 			continue;
807 		}
808 
809 		r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
810 				  &rs->dev[i].data_dev);
811 		if (r) {
812 			rs->ti->error = "RAID device lookup failure";
813 			return r;
814 		}
815 
816 		if (rs->dev[i].meta_dev) {
817 			metadata_available = 1;
818 			rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
819 		}
820 		rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
821 		list_add_tail(&rs->dev[i].rdev.same_set, &rs->md.disks);
822 		if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
823 			rebuild++;
824 	}
825 
826 	if (metadata_available) {
827 		rs->md.external = 0;
828 		rs->md.persistent = 1;
829 		rs->md.major_version = 2;
830 	} else if (rebuild && !rs->md.recovery_cp) {
831 		/*
832 		 * Without metadata, we will not be able to tell if the array
833 		 * is in-sync or not - we must assume it is not.  Therefore,
834 		 * it is impossible to rebuild a drive.
835 		 *
836 		 * Even if there is metadata, the on-disk information may
837 		 * indicate that the array is not in-sync and it will then
838 		 * fail at that time.
839 		 *
840 		 * User could specify 'nosync' option if desperate.
841 		 */
842 		rs->ti->error = "Unable to rebuild drive while array is not in-sync";
843 		return -EINVAL;
844 	}
845 
846 	return 0;
847 }
848 
849 /*
850  * validate_region_size
851  * @rs
852  * @region_size:  region size in sectors.  If 0, pick a size (4MiB default).
853  *
854  * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
855  * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
856  *
857  * Returns: 0 on success, -EINVAL on failure.
858  */
859 static int validate_region_size(struct raid_set *rs, unsigned long region_size)
860 {
861 	unsigned long min_region_size = rs->ti->len / (1 << 21);
862 
863 	if (rs_is_raid0(rs))
864 		return 0;
865 
866 	if (!region_size) {
867 		/*
868 		 * Choose a reasonable default.	 All figures in sectors.
869 		 */
870 		if (min_region_size > (1 << 13)) {
871 			/* If not a power of 2, make it the next power of 2 */
872 			region_size = roundup_pow_of_two(min_region_size);
873 			DMINFO("Choosing default region size of %lu sectors",
874 			       region_size);
875 		} else {
876 			DMINFO("Choosing default region size of 4MiB");
877 			region_size = 1 << 13; /* sectors */
878 		}
879 	} else {
880 		/*
881 		 * Validate user-supplied value.
882 		 */
883 		if (region_size > rs->ti->len) {
884 			rs->ti->error = "Supplied region size is too large";
885 			return -EINVAL;
886 		}
887 
888 		if (region_size < min_region_size) {
889 			DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
890 			      region_size, min_region_size);
891 			rs->ti->error = "Supplied region size is too small";
892 			return -EINVAL;
893 		}
894 
895 		if (!is_power_of_2(region_size)) {
896 			rs->ti->error = "Region size is not a power of 2";
897 			return -EINVAL;
898 		}
899 
900 		if (region_size < rs->md.chunk_sectors) {
901 			rs->ti->error = "Region size is smaller than the chunk size";
902 			return -EINVAL;
903 		}
904 	}
905 
906 	/*
907 	 * Convert sectors to bytes.
908 	 */
909 	rs->md.bitmap_info.chunksize = to_bytes(region_size);
910 
911 	return 0;
912 }
913 
914 /*
915  * validate_raid_redundancy
916  * @rs
917  *
918  * Determine if there are enough devices in the array that haven't
919  * failed (or are being rebuilt) to form a usable array.
920  *
921  * Returns: 0 on success, -EINVAL on failure.
922  */
923 static int validate_raid_redundancy(struct raid_set *rs)
924 {
925 	unsigned int i, rebuild_cnt = 0;
926 	unsigned int rebuilds_per_group = 0, copies;
927 	unsigned int group_size, last_group_start;
928 
929 	for (i = 0; i < rs->md.raid_disks; i++)
930 		if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
931 		    !rs->dev[i].rdev.sb_page)
932 			rebuild_cnt++;
933 
934 	switch (rs->raid_type->level) {
935 	case 0:
936 		break;
937 	case 1:
938 		if (rebuild_cnt >= rs->md.raid_disks)
939 			goto too_many;
940 		break;
941 	case 4:
942 	case 5:
943 	case 6:
944 		if (rebuild_cnt > rs->raid_type->parity_devs)
945 			goto too_many;
946 		break;
947 	case 10:
948 		copies = raid10_md_layout_to_copies(rs->md.new_layout);
949 		if (rebuild_cnt < copies)
950 			break;
951 
952 		/*
953 		 * It is possible to have a higher rebuild count for RAID10,
954 		 * as long as the failed devices occur in different mirror
955 		 * groups (i.e. different stripes).
956 		 *
957 		 * When checking "near" format, make sure no adjacent devices
958 		 * have failed beyond what can be handled.  In addition to the
959 		 * simple case where the number of devices is a multiple of the
960 		 * number of copies, we must also handle cases where the number
961 		 * of devices is not a multiple of the number of copies.
962 		 * E.g.	   dev1 dev2 dev3 dev4 dev5
963 		 *	    A	 A    B	   B	C
964 		 *	    C	 D    D	   E	E
965 		 */
966 		if (__is_raid10_near(rs->md.new_layout)) {
967 			for (i = 0; i < rs->md.raid_disks; i++) {
968 				if (!(i % copies))
969 					rebuilds_per_group = 0;
970 				if ((!rs->dev[i].rdev.sb_page ||
971 				    !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
972 				    (++rebuilds_per_group >= copies))
973 					goto too_many;
974 			}
975 			break;
976 		}
977 
978 		/*
979 		 * When checking "far" and "offset" formats, we need to ensure
980 		 * that the device that holds its copy is not also dead or
981 		 * being rebuilt.  (Note that "far" and "offset" formats only
982 		 * support two copies right now.  These formats also only ever
983 		 * use the 'use_far_sets' variant.)
984 		 *
985 		 * This check is somewhat complicated by the need to account
986 		 * for arrays that are not a multiple of (far) copies.	This
987 		 * results in the need to treat the last (potentially larger)
988 		 * set differently.
989 		 */
990 		group_size = (rs->md.raid_disks / copies);
991 		last_group_start = (rs->md.raid_disks / group_size) - 1;
992 		last_group_start *= group_size;
993 		for (i = 0; i < rs->md.raid_disks; i++) {
994 			if (!(i % copies) && !(i > last_group_start))
995 				rebuilds_per_group = 0;
996 			if ((!rs->dev[i].rdev.sb_page ||
997 			     !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
998 			    (++rebuilds_per_group >= copies))
999 					goto too_many;
1000 		}
1001 		break;
1002 	default:
1003 		if (rebuild_cnt)
1004 			return -EINVAL;
1005 	}
1006 
1007 	return 0;
1008 
1009 too_many:
1010 	return -EINVAL;
1011 }
1012 
1013 /*
1014  * Possible arguments are...
1015  *	<chunk_size> [optional_args]
1016  *
1017  * Argument definitions
1018  *    <chunk_size>			The number of sectors per disk that
1019  *					will form the "stripe"
1020  *    [[no]sync]			Force or prevent recovery of the
1021  *					entire array
1022  *    [rebuild <idx>]			Rebuild the drive indicated by the index
1023  *    [daemon_sleep <ms>]		Time between bitmap daemon work to
1024  *					clear bits
1025  *    [min_recovery_rate <kB/sec/disk>]	Throttle RAID initialization
1026  *    [max_recovery_rate <kB/sec/disk>]	Throttle RAID initialization
1027  *    [write_mostly <idx>]		Indicate a write mostly drive via index
1028  *    [max_write_behind <sectors>]	See '-write-behind=' (man mdadm)
1029  *    [stripe_cache <sectors>]		Stripe cache size for higher RAIDs
1030  *    [region_size <sectors>]		Defines granularity of bitmap
1031  *
1032  * RAID10-only options:
1033  *    [raid10_copies <# copies>]	Number of copies.  (Default: 2)
1034  *    [raid10_format <near|far|offset>] Layout algorithm.  (Default: near)
1035  */
1036 static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
1037 			     unsigned int num_raid_params)
1038 {
1039 	int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
1040 	unsigned int raid10_copies = 2;
1041 	unsigned int i, write_mostly = 0;
1042 	unsigned int region_size = 0;
1043 	sector_t max_io_len;
1044 	const char *arg, *key;
1045 	struct raid_dev *rd;
1046 	struct raid_type *rt = rs->raid_type;
1047 
1048 	arg = dm_shift_arg(as);
1049 	num_raid_params--; /* Account for chunk_size argument */
1050 
1051 	if (kstrtoint(arg, 10, &value) < 0) {
1052 		rs->ti->error = "Bad numerical argument given for chunk_size";
1053 		return -EINVAL;
1054 	}
1055 
1056 	/*
1057 	 * First, parse the in-order required arguments
1058 	 * "chunk_size" is the only argument of this type.
1059 	 */
1060 	if (rt_is_raid1(rt)) {
1061 		if (value)
1062 			DMERR("Ignoring chunk size parameter for RAID 1");
1063 		value = 0;
1064 	} else if (!is_power_of_2(value)) {
1065 		rs->ti->error = "Chunk size must be a power of 2";
1066 		return -EINVAL;
1067 	} else if (value < 8) {
1068 		rs->ti->error = "Chunk size value is too small";
1069 		return -EINVAL;
1070 	}
1071 
1072 	rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
1073 
1074 	/*
1075 	 * We set each individual device as In_sync with a completed
1076 	 * 'recovery_offset'.  If there has been a device failure or
1077 	 * replacement then one of the following cases applies:
1078 	 *
1079 	 *   1) User specifies 'rebuild'.
1080 	 *	- Device is reset when param is read.
1081 	 *   2) A new device is supplied.
1082 	 *	- No matching superblock found, resets device.
1083 	 *   3) Device failure was transient and returns on reload.
1084 	 *	- Failure noticed, resets device for bitmap replay.
1085 	 *   4) Device hadn't completed recovery after previous failure.
1086 	 *	- Superblock is read and overrides recovery_offset.
1087 	 *
1088 	 * What is found in the superblocks of the devices is always
1089 	 * authoritative, unless 'rebuild' or '[no]sync' was specified.
1090 	 */
1091 	for (i = 0; i < rs->raid_disks; i++) {
1092 		set_bit(In_sync, &rs->dev[i].rdev.flags);
1093 		rs->dev[i].rdev.recovery_offset = MaxSector;
1094 	}
1095 
1096 	/*
1097 	 * Second, parse the unordered optional arguments
1098 	 */
1099 	for (i = 0; i < num_raid_params; i++) {
1100 		key = dm_shift_arg(as);
1101 		if (!key) {
1102 			rs->ti->error = "Not enough raid parameters given";
1103 			return -EINVAL;
1104 		}
1105 
1106 		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
1107 			if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1108 				rs->ti->error = "Only one 'nosync' argument allowed";
1109 				return -EINVAL;
1110 			}
1111 			continue;
1112 		}
1113 		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
1114 			if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
1115 				rs->ti->error = "Only one 'sync' argument allowed";
1116 				return -EINVAL;
1117 			}
1118 			continue;
1119 		}
1120 		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
1121 			if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1122 				rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1123 				return -EINVAL;
1124 			}
1125 			continue;
1126 		}
1127 
1128 		arg = dm_shift_arg(as);
1129 		i++; /* Account for the argument pairs */
1130 		if (!arg) {
1131 			rs->ti->error = "Wrong number of raid parameters given";
1132 			return -EINVAL;
1133 		}
1134 
1135 		/*
1136 		 * Parameters that take a string value are checked here.
1137 		 */
1138 
1139 		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
1140 			if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
1141 				rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1142 				return -EINVAL;
1143 			}
1144 			if (!rt_is_raid10(rt)) {
1145 				rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1146 				return -EINVAL;
1147 			}
1148 			raid10_format = raid10_name_to_format(arg);
1149 			if (raid10_format < 0) {
1150 				rs->ti->error = "Invalid 'raid10_format' value given";
1151 				return raid10_format;
1152 			}
1153 			continue;
1154 		}
1155 
1156 		if (kstrtoint(arg, 10, &value) < 0) {
1157 			rs->ti->error = "Bad numerical argument given in raid params";
1158 			return -EINVAL;
1159 		}
1160 
1161 		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
1162 			/*
1163 			 * "rebuild" is being passed in by userspace to provide
1164 			 * indexes of replaced devices and to set up additional
1165 			 * devices on raid level takeover.
1166 			 */
1167 			if (!__within_range(value, 0, rs->raid_disks - 1)) {
1168 				rs->ti->error = "Invalid rebuild index given";
1169 				return -EINVAL;
1170 			}
1171 
1172 			if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1173 				rs->ti->error = "rebuild for this index already given";
1174 				return -EINVAL;
1175 			}
1176 
1177 			rd = rs->dev + value;
1178 			clear_bit(In_sync, &rd->rdev.flags);
1179 			clear_bit(Faulty, &rd->rdev.flags);
1180 			rd->rdev.recovery_offset = 0;
1181 			set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
1182 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
1183 			if (!rt_is_raid1(rt)) {
1184 				rs->ti->error = "write_mostly option is only valid for RAID1";
1185 				return -EINVAL;
1186 			}
1187 
1188 			if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
1189 				rs->ti->error = "Invalid write_mostly index given";
1190 				return -EINVAL;
1191 			}
1192 
1193 			write_mostly++;
1194 			set_bit(WriteMostly, &rs->dev[value].rdev.flags);
1195 			set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
1196 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
1197 			if (!rt_is_raid1(rt)) {
1198 				rs->ti->error = "max_write_behind option is only valid for RAID1";
1199 				return -EINVAL;
1200 			}
1201 
1202 			if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
1203 				rs->ti->error = "Only one max_write_behind argument pair allowed";
1204 				return -EINVAL;
1205 			}
1206 
1207 			/*
1208 			 * In device-mapper, we specify things in sectors, but
1209 			 * MD records this value in kB
1210 			 */
1211 			value /= 2;
1212 			if (value > COUNTER_MAX) {
1213 				rs->ti->error = "Max write-behind limit out of range";
1214 				return -EINVAL;
1215 			}
1216 
1217 			rs->md.bitmap_info.max_write_behind = value;
1218 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
1219 			if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
1220 				rs->ti->error = "Only one daemon_sleep argument pair allowed";
1221 				return -EINVAL;
1222 			}
1223 			if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
1224 				rs->ti->error = "daemon sleep period out of range";
1225 				return -EINVAL;
1226 			}
1227 			rs->md.bitmap_info.daemon_sleep = value;
1228 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
1229 			/* Userspace passes new data_offset after having extended the the data image LV */
1230 			if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
1231 				rs->ti->error = "Only one data_offset argument pair allowed";
1232 				return -EINVAL;
1233 			}
1234 			/* Ensure sensible data offset */
1235 			if (value < 0 ||
1236 			    (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
1237 				rs->ti->error = "Bogus data_offset value";
1238 				return -EINVAL;
1239 			}
1240 			rs->data_offset = value;
1241 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
1242 			/* Define the +/-# of disks to add to/remove from the given raid set */
1243 			if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1244 				rs->ti->error = "Only one delta_disks argument pair allowed";
1245 				return -EINVAL;
1246 			}
1247 			/* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1248 			if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
1249 				rs->ti->error = "Too many delta_disk requested";
1250 				return -EINVAL;
1251 			}
1252 
1253 			rs->delta_disks = value;
1254 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
1255 			if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
1256 				rs->ti->error = "Only one stripe_cache argument pair allowed";
1257 				return -EINVAL;
1258 			}
1259 
1260 			if (!rt_is_raid456(rt)) {
1261 				rs->ti->error = "Inappropriate argument: stripe_cache";
1262 				return -EINVAL;
1263 			}
1264 
1265 			rs->stripe_cache_entries = value;
1266 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
1267 			if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1268 				rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1269 				return -EINVAL;
1270 			}
1271 			if (value > INT_MAX) {
1272 				rs->ti->error = "min_recovery_rate out of range";
1273 				return -EINVAL;
1274 			}
1275 			rs->md.sync_speed_min = (int)value;
1276 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
1277 			if (test_and_set_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags)) {
1278 				rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1279 				return -EINVAL;
1280 			}
1281 			if (value > INT_MAX) {
1282 				rs->ti->error = "max_recovery_rate out of range";
1283 				return -EINVAL;
1284 			}
1285 			rs->md.sync_speed_max = (int)value;
1286 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
1287 			if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
1288 				rs->ti->error = "Only one region_size argument pair allowed";
1289 				return -EINVAL;
1290 			}
1291 
1292 			region_size = value;
1293 			rs->requested_bitmap_chunk_sectors = value;
1294 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
1295 			if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
1296 				rs->ti->error = "Only one raid10_copies argument pair allowed";
1297 				return -EINVAL;
1298 			}
1299 
1300 			if (!__within_range(value, 2, rs->md.raid_disks)) {
1301 				rs->ti->error = "Bad value for 'raid10_copies'";
1302 				return -EINVAL;
1303 			}
1304 
1305 			raid10_copies = value;
1306 		} else {
1307 			DMERR("Unable to parse RAID parameter: %s", key);
1308 			rs->ti->error = "Unable to parse RAID parameter";
1309 			return -EINVAL;
1310 		}
1311 	}
1312 
1313 	if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1314 	    test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1315 		rs->ti->error = "sync and nosync are mutually exclusive";
1316 		return -EINVAL;
1317 	}
1318 
1319 	if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) &&
1320 	    (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ||
1321 	     test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))) {
1322 		rs->ti->error = "sync/nosync and rebuild are mutually exclusive";
1323 		return -EINVAL;
1324 	}
1325 
1326 	if (write_mostly >= rs->md.raid_disks) {
1327 		rs->ti->error = "Can't set all raid1 devices to write_mostly";
1328 		return -EINVAL;
1329 	}
1330 
1331 	if (validate_region_size(rs, region_size))
1332 		return -EINVAL;
1333 
1334 	if (rs->md.chunk_sectors)
1335 		max_io_len = rs->md.chunk_sectors;
1336 	else
1337 		max_io_len = region_size;
1338 
1339 	if (dm_set_target_max_io_len(rs->ti, max_io_len))
1340 		return -EINVAL;
1341 
1342 	if (rt_is_raid10(rt)) {
1343 		if (raid10_copies > rs->md.raid_disks) {
1344 			rs->ti->error = "Not enough devices to satisfy specification";
1345 			return -EINVAL;
1346 		}
1347 
1348 		rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
1349 		if (rs->md.new_layout < 0) {
1350 			rs->ti->error = "Error getting raid10 format";
1351 			return rs->md.new_layout;
1352 		}
1353 
1354 		rt = get_raid_type_by_ll(10, rs->md.new_layout);
1355 		if (!rt) {
1356 			rs->ti->error = "Failed to recognize new raid10 layout";
1357 			return -EINVAL;
1358 		}
1359 
1360 		if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1361 		     rt->algorithm == ALGORITHM_RAID10_NEAR) &&
1362 		    test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1363 			rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1364 			return -EINVAL;
1365 		}
1366 	}
1367 
1368 	rs->raid10_copies = raid10_copies;
1369 
1370 	/* Assume there are no metadata devices until the drives are parsed */
1371 	rs->md.persistent = 0;
1372 	rs->md.external = 1;
1373 
1374 	/* Check, if any invalid ctr arguments have been passed in for the raid level */
1375 	return rs_check_for_valid_flags(rs);
1376 }
1377 
1378 /* Set raid4/5/6 cache size */
1379 static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1380 {
1381 	int r;
1382 	struct r5conf *conf;
1383 	struct mddev *mddev = &rs->md;
1384 	uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1385 	uint32_t nr_stripes = rs->stripe_cache_entries;
1386 
1387 	if (!rt_is_raid456(rs->raid_type)) {
1388 		rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1389 		return -EINVAL;
1390 	}
1391 
1392 	if (nr_stripes < min_stripes) {
1393 		DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1394 		       nr_stripes, min_stripes);
1395 		nr_stripes = min_stripes;
1396 	}
1397 
1398 	conf = mddev->private;
1399 	if (!conf) {
1400 		rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1401 		return -EINVAL;
1402 	}
1403 
1404 	/* Try setting number of stripes in raid456 stripe cache */
1405 	if (conf->min_nr_stripes != nr_stripes) {
1406 		r = raid5_set_cache_size(mddev, nr_stripes);
1407 		if (r) {
1408 			rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1409 			return r;
1410 		}
1411 
1412 		DMINFO("%u stripe cache entries", nr_stripes);
1413 	}
1414 
1415 	return 0;
1416 }
1417 
1418 /* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1419 static unsigned int mddev_data_stripes(struct raid_set *rs)
1420 {
1421 	return rs->md.raid_disks - rs->raid_type->parity_devs;
1422 }
1423 
1424 /* Return # of data stripes of @rs (i.e. as of ctr) */
1425 static unsigned int rs_data_stripes(struct raid_set *rs)
1426 {
1427 	return rs->raid_disks - rs->raid_type->parity_devs;
1428 }
1429 
1430 /* Calculate the sectors per device and per array used for @rs */
1431 static int rs_set_dev_and_array_sectors(struct raid_set *rs, bool use_mddev)
1432 {
1433 	int delta_disks;
1434 	unsigned int data_stripes;
1435 	struct mddev *mddev = &rs->md;
1436 	struct md_rdev *rdev;
1437 	sector_t array_sectors = rs->ti->len, dev_sectors = rs->ti->len;
1438 
1439 	if (use_mddev) {
1440 		delta_disks = mddev->delta_disks;
1441 		data_stripes = mddev_data_stripes(rs);
1442 	} else {
1443 		delta_disks = rs->delta_disks;
1444 		data_stripes = rs_data_stripes(rs);
1445 	}
1446 
1447 	/* Special raid1 case w/o delta_disks support (yet) */
1448 	if (rt_is_raid1(rs->raid_type))
1449 		;
1450 	else if (rt_is_raid10(rs->raid_type)) {
1451 		if (rs->raid10_copies < 2 ||
1452 		    delta_disks < 0) {
1453 			rs->ti->error = "Bogus raid10 data copies or delta disks";
1454 			return -EINVAL;
1455 		}
1456 
1457 		dev_sectors *= rs->raid10_copies;
1458 		if (sector_div(dev_sectors, data_stripes))
1459 			goto bad;
1460 
1461 		array_sectors = (data_stripes + delta_disks) * dev_sectors;
1462 		if (sector_div(array_sectors, rs->raid10_copies))
1463 			goto bad;
1464 
1465 	} else if (sector_div(dev_sectors, data_stripes))
1466 		goto bad;
1467 
1468 	else
1469 		/* Striped layouts */
1470 		array_sectors = (data_stripes + delta_disks) * dev_sectors;
1471 
1472 	rdev_for_each(rdev, mddev)
1473 		rdev->sectors = dev_sectors;
1474 
1475 	mddev->array_sectors = array_sectors;
1476 	mddev->dev_sectors = dev_sectors;
1477 
1478 	return 0;
1479 bad:
1480 	rs->ti->error = "Target length not divisible by number of data devices";
1481 	return -EINVAL;
1482 }
1483 
1484 /* Setup recovery on @rs */
1485 static void __rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1486 {
1487 	/* raid0 does not recover */
1488 	if (rs_is_raid0(rs))
1489 		rs->md.recovery_cp = MaxSector;
1490 	/*
1491 	 * A raid6 set has to be recovered either
1492 	 * completely or for the grown part to
1493 	 * ensure proper parity and Q-Syndrome
1494 	 */
1495 	else if (rs_is_raid6(rs))
1496 		rs->md.recovery_cp = dev_sectors;
1497 	/*
1498 	 * Other raid set types may skip recovery
1499 	 * depending on the 'nosync' flag.
1500 	 */
1501 	else
1502 		rs->md.recovery_cp = test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)
1503 				     ? MaxSector : dev_sectors;
1504 }
1505 
1506 /* Setup recovery on @rs based on raid type, device size and 'nosync' flag */
1507 static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1508 {
1509 	if (!dev_sectors)
1510 		/* New raid set or 'sync' flag provided */
1511 		__rs_setup_recovery(rs, 0);
1512 	else if (dev_sectors == MaxSector)
1513 		/* Prevent recovery */
1514 		__rs_setup_recovery(rs, MaxSector);
1515 	else if (rs->dev[0].rdev.sectors < dev_sectors)
1516 		/* Grown raid set */
1517 		__rs_setup_recovery(rs, rs->dev[0].rdev.sectors);
1518 	else
1519 		__rs_setup_recovery(rs, MaxSector);
1520 }
1521 
1522 static void do_table_event(struct work_struct *ws)
1523 {
1524 	struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1525 
1526 	smp_rmb(); /* Make sure we access most actual mddev properties */
1527 	if (!rs_is_reshaping(rs))
1528 		rs_set_capacity(rs);
1529 	dm_table_event(rs->ti->table);
1530 }
1531 
1532 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
1533 {
1534 	struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
1535 
1536 	return mddev_congested(&rs->md, bits);
1537 }
1538 
1539 /*
1540  * Make sure a valid takover (level switch) is being requested on @rs
1541  *
1542  * Conversions of raid sets from one MD personality to another
1543  * have to conform to restrictions which are enforced here.
1544  */
1545 static int rs_check_takeover(struct raid_set *rs)
1546 {
1547 	struct mddev *mddev = &rs->md;
1548 	unsigned int near_copies;
1549 
1550 	if (rs->md.degraded) {
1551 		rs->ti->error = "Can't takeover degraded raid set";
1552 		return -EPERM;
1553 	}
1554 
1555 	if (rs_is_reshaping(rs)) {
1556 		rs->ti->error = "Can't takeover reshaping raid set";
1557 		return -EPERM;
1558 	}
1559 
1560 	switch (mddev->level) {
1561 	case 0:
1562 		/* raid0 -> raid1/5 with one disk */
1563 		if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1564 		    mddev->raid_disks == 1)
1565 			return 0;
1566 
1567 		/* raid0 -> raid10 */
1568 		if (mddev->new_level == 10 &&
1569 		    !(rs->raid_disks % mddev->raid_disks))
1570 			return 0;
1571 
1572 		/* raid0 with multiple disks -> raid4/5/6 */
1573 		if (__within_range(mddev->new_level, 4, 6) &&
1574 		    mddev->new_layout == ALGORITHM_PARITY_N &&
1575 		    mddev->raid_disks > 1)
1576 			return 0;
1577 
1578 		break;
1579 
1580 	case 10:
1581 		/* Can't takeover raid10_offset! */
1582 		if (__is_raid10_offset(mddev->layout))
1583 			break;
1584 
1585 		near_copies = __raid10_near_copies(mddev->layout);
1586 
1587 		/* raid10* -> raid0 */
1588 		if (mddev->new_level == 0) {
1589 			/* Can takeover raid10_near with raid disks divisable by data copies! */
1590 			if (near_copies > 1 &&
1591 			    !(mddev->raid_disks % near_copies)) {
1592 				mddev->raid_disks /= near_copies;
1593 				mddev->delta_disks = mddev->raid_disks;
1594 				return 0;
1595 			}
1596 
1597 			/* Can takeover raid10_far */
1598 			if (near_copies == 1 &&
1599 			    __raid10_far_copies(mddev->layout) > 1)
1600 				return 0;
1601 
1602 			break;
1603 		}
1604 
1605 		/* raid10_{near,far} -> raid1 */
1606 		if (mddev->new_level == 1 &&
1607 		    max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
1608 			return 0;
1609 
1610 		/* raid10_{near,far} with 2 disks -> raid4/5 */
1611 		if (__within_range(mddev->new_level, 4, 5) &&
1612 		    mddev->raid_disks == 2)
1613 			return 0;
1614 		break;
1615 
1616 	case 1:
1617 		/* raid1 with 2 disks -> raid4/5 */
1618 		if (__within_range(mddev->new_level, 4, 5) &&
1619 		    mddev->raid_disks == 2) {
1620 			mddev->degraded = 1;
1621 			return 0;
1622 		}
1623 
1624 		/* raid1 -> raid0 */
1625 		if (mddev->new_level == 0 &&
1626 		    mddev->raid_disks == 1)
1627 			return 0;
1628 
1629 		/* raid1 -> raid10 */
1630 		if (mddev->new_level == 10)
1631 			return 0;
1632 		break;
1633 
1634 	case 4:
1635 		/* raid4 -> raid0 */
1636 		if (mddev->new_level == 0)
1637 			return 0;
1638 
1639 		/* raid4 -> raid1/5 with 2 disks */
1640 		if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1641 		    mddev->raid_disks == 2)
1642 			return 0;
1643 
1644 		/* raid4 -> raid5/6 with parity N */
1645 		if (__within_range(mddev->new_level, 5, 6) &&
1646 		    mddev->layout == ALGORITHM_PARITY_N)
1647 			return 0;
1648 		break;
1649 
1650 	case 5:
1651 		/* raid5 with parity N -> raid0 */
1652 		if (mddev->new_level == 0 &&
1653 		    mddev->layout == ALGORITHM_PARITY_N)
1654 			return 0;
1655 
1656 		/* raid5 with parity N -> raid4 */
1657 		if (mddev->new_level == 4 &&
1658 		    mddev->layout == ALGORITHM_PARITY_N)
1659 			return 0;
1660 
1661 		/* raid5 with 2 disks -> raid1/4/10 */
1662 		if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1663 		    mddev->raid_disks == 2)
1664 			return 0;
1665 
1666 		/* raid5_* ->  raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
1667 		if (mddev->new_level == 6 &&
1668 		    ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1669 		      __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
1670 			return 0;
1671 		break;
1672 
1673 	case 6:
1674 		/* raid6 with parity N -> raid0 */
1675 		if (mddev->new_level == 0 &&
1676 		    mddev->layout == ALGORITHM_PARITY_N)
1677 			return 0;
1678 
1679 		/* raid6 with parity N -> raid4 */
1680 		if (mddev->new_level == 4 &&
1681 		    mddev->layout == ALGORITHM_PARITY_N)
1682 			return 0;
1683 
1684 		/* raid6_*_n with Q-Syndrome N -> raid5_* */
1685 		if (mddev->new_level == 5 &&
1686 		    ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1687 		     __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
1688 			return 0;
1689 
1690 	default:
1691 		break;
1692 	}
1693 
1694 	rs->ti->error = "takeover not possible";
1695 	return -EINVAL;
1696 }
1697 
1698 /* True if @rs requested to be taken over */
1699 static bool rs_takeover_requested(struct raid_set *rs)
1700 {
1701 	return rs->md.new_level != rs->md.level;
1702 }
1703 
1704 /* True if @rs is requested to reshape by ctr */
1705 static bool rs_reshape_requested(struct raid_set *rs)
1706 {
1707 	bool change;
1708 	struct mddev *mddev = &rs->md;
1709 
1710 	if (rs_takeover_requested(rs))
1711 		return false;
1712 
1713 	if (!mddev->level)
1714 		return false;
1715 
1716 	change = mddev->new_layout != mddev->layout ||
1717 		 mddev->new_chunk_sectors != mddev->chunk_sectors ||
1718 		 rs->delta_disks;
1719 
1720 	/* Historical case to support raid1 reshape without delta disks */
1721 	if (mddev->level == 1) {
1722 		if (rs->delta_disks)
1723 			return !!rs->delta_disks;
1724 
1725 		return !change &&
1726 		       mddev->raid_disks != rs->raid_disks;
1727 	}
1728 
1729 	if (mddev->level == 10)
1730 		return change &&
1731 		       !__is_raid10_far(mddev->new_layout) &&
1732 		       rs->delta_disks >= 0;
1733 
1734 	return change;
1735 }
1736 
1737 /*  Features */
1738 #define	FEATURE_FLAG_SUPPORTS_V190	0x1 /* Supports extended superblock */
1739 
1740 /* State flags for sb->flags */
1741 #define	SB_FLAG_RESHAPE_ACTIVE		0x1
1742 #define	SB_FLAG_RESHAPE_BACKWARDS	0x2
1743 
1744 /*
1745  * This structure is never routinely used by userspace, unlike md superblocks.
1746  * Devices with this superblock should only ever be accessed via device-mapper.
1747  */
1748 #define DM_RAID_MAGIC 0x64526D44
1749 struct dm_raid_superblock {
1750 	__le32 magic;		/* "DmRd" */
1751 	__le32 compat_features;	/* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1752 
1753 	__le32 num_devices;	/* Number of devices in this raid set. (Max 64) */
1754 	__le32 array_position;	/* The position of this drive in the raid set */
1755 
1756 	__le64 events;		/* Incremented by md when superblock updated */
1757 	__le64 failed_devices;	/* Pre 1.9.0 part of bit field of devices to */
1758 				/* indicate failures (see extension below) */
1759 
1760 	/*
1761 	 * This offset tracks the progress of the repair or replacement of
1762 	 * an individual drive.
1763 	 */
1764 	__le64 disk_recovery_offset;
1765 
1766 	/*
1767 	 * This offset tracks the progress of the initial raid set
1768 	 * synchronisation/parity calculation.
1769 	 */
1770 	__le64 array_resync_offset;
1771 
1772 	/*
1773 	 * raid characteristics
1774 	 */
1775 	__le32 level;
1776 	__le32 layout;
1777 	__le32 stripe_sectors;
1778 
1779 	/********************************************************************
1780 	 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1781 	 *
1782 	 * FEATURE_FLAG_SUPPORTS_V190 in the features member indicates that those exist
1783 	 */
1784 
1785 	__le32 flags; /* Flags defining array states for reshaping */
1786 
1787 	/*
1788 	 * This offset tracks the progress of a raid
1789 	 * set reshape in order to be able to restart it
1790 	 */
1791 	__le64 reshape_position;
1792 
1793 	/*
1794 	 * These define the properties of the array in case of an interrupted reshape
1795 	 */
1796 	__le32 new_level;
1797 	__le32 new_layout;
1798 	__le32 new_stripe_sectors;
1799 	__le32 delta_disks;
1800 
1801 	__le64 array_sectors; /* Array size in sectors */
1802 
1803 	/*
1804 	 * Sector offsets to data on devices (reshaping).
1805 	 * Needed to support out of place reshaping, thus
1806 	 * not writing over any stripes whilst converting
1807 	 * them from old to new layout
1808 	 */
1809 	__le64 data_offset;
1810 	__le64 new_data_offset;
1811 
1812 	__le64 sectors; /* Used device size in sectors */
1813 
1814 	/*
1815 	 * Additonal Bit field of devices indicating failures to support
1816 	 * up to 256 devices with the 1.9.0 on-disk metadata format
1817 	 */
1818 	__le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
1819 
1820 	__le32 incompat_features;	/* Used to indicate any incompatible features */
1821 
1822 	/* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
1823 } __packed;
1824 
1825 /*
1826  * Check for reshape constraints on raid set @rs:
1827  *
1828  * - reshape function non-existent
1829  * - degraded set
1830  * - ongoing recovery
1831  * - ongoing reshape
1832  *
1833  * Returns 0 if none or -EPERM if given constraint
1834  * and error message reference in @errmsg
1835  */
1836 static int rs_check_reshape(struct raid_set *rs)
1837 {
1838 	struct mddev *mddev = &rs->md;
1839 
1840 	if (!mddev->pers || !mddev->pers->check_reshape)
1841 		rs->ti->error = "Reshape not supported";
1842 	else if (mddev->degraded)
1843 		rs->ti->error = "Can't reshape degraded raid set";
1844 	else if (rs_is_recovering(rs))
1845 		rs->ti->error = "Convert request on recovering raid set prohibited";
1846 	else if (rs_is_reshaping(rs))
1847 		rs->ti->error = "raid set already reshaping!";
1848 	else if (!(rs_is_raid1(rs) || rs_is_raid10(rs) || rs_is_raid456(rs)))
1849 		rs->ti->error = "Reshaping only supported for raid1/4/5/6/10";
1850 	else
1851 		return 0;
1852 
1853 	return -EPERM;
1854 }
1855 
1856 static int read_disk_sb(struct md_rdev *rdev, int size)
1857 {
1858 	BUG_ON(!rdev->sb_page);
1859 
1860 	if (rdev->sb_loaded)
1861 		return 0;
1862 
1863 	if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) {
1864 		DMERR("Failed to read superblock of device at position %d",
1865 		      rdev->raid_disk);
1866 		md_error(rdev->mddev, rdev);
1867 		return -EINVAL;
1868 	}
1869 
1870 	rdev->sb_loaded = 1;
1871 
1872 	return 0;
1873 }
1874 
1875 static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
1876 {
1877 	failed_devices[0] = le64_to_cpu(sb->failed_devices);
1878 	memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
1879 
1880 	if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
1881 		int i = ARRAY_SIZE(sb->extended_failed_devices);
1882 
1883 		while (i--)
1884 			failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
1885 	}
1886 }
1887 
1888 static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
1889 {
1890 	int i = ARRAY_SIZE(sb->extended_failed_devices);
1891 
1892 	sb->failed_devices = cpu_to_le64(failed_devices[0]);
1893 	while (i--)
1894 		sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
1895 }
1896 
1897 /*
1898  * Synchronize the superblock members with the raid set properties
1899  *
1900  * All superblock data is little endian.
1901  */
1902 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
1903 {
1904 	bool update_failed_devices = false;
1905 	unsigned int i;
1906 	uint64_t failed_devices[DISKS_ARRAY_ELEMS];
1907 	struct dm_raid_superblock *sb;
1908 	struct raid_set *rs = container_of(mddev, struct raid_set, md);
1909 
1910 	/* No metadata device, no superblock */
1911 	if (!rdev->meta_bdev)
1912 		return;
1913 
1914 	BUG_ON(!rdev->sb_page);
1915 
1916 	sb = page_address(rdev->sb_page);
1917 
1918 	sb_retrieve_failed_devices(sb, failed_devices);
1919 
1920 	for (i = 0; i < rs->raid_disks; i++)
1921 		if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
1922 			update_failed_devices = true;
1923 			set_bit(i, (void *) failed_devices);
1924 		}
1925 
1926 	if (update_failed_devices)
1927 		sb_update_failed_devices(sb, failed_devices);
1928 
1929 	sb->magic = cpu_to_le32(DM_RAID_MAGIC);
1930 	sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
1931 
1932 	sb->num_devices = cpu_to_le32(mddev->raid_disks);
1933 	sb->array_position = cpu_to_le32(rdev->raid_disk);
1934 
1935 	sb->events = cpu_to_le64(mddev->events);
1936 
1937 	sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
1938 	sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
1939 
1940 	sb->level = cpu_to_le32(mddev->level);
1941 	sb->layout = cpu_to_le32(mddev->layout);
1942 	sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
1943 
1944 	sb->new_level = cpu_to_le32(mddev->new_level);
1945 	sb->new_layout = cpu_to_le32(mddev->new_layout);
1946 	sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
1947 
1948 	sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1949 
1950 	smp_rmb(); /* Make sure we access most recent reshape position */
1951 	sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1952 	if (le64_to_cpu(sb->reshape_position) != MaxSector) {
1953 		/* Flag ongoing reshape */
1954 		sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
1955 
1956 		if (mddev->delta_disks < 0 || mddev->reshape_backwards)
1957 			sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
1958 	} else {
1959 		/* Clear reshape flags */
1960 		sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
1961 	}
1962 
1963 	sb->array_sectors = cpu_to_le64(mddev->array_sectors);
1964 	sb->data_offset = cpu_to_le64(rdev->data_offset);
1965 	sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
1966 	sb->sectors = cpu_to_le64(rdev->sectors);
1967 	sb->incompat_features = cpu_to_le32(0);
1968 
1969 	/* Zero out the rest of the payload after the size of the superblock */
1970 	memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
1971 }
1972 
1973 /*
1974  * super_load
1975  *
1976  * This function creates a superblock if one is not found on the device
1977  * and will decide which superblock to use if there's a choice.
1978  *
1979  * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
1980  */
1981 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
1982 {
1983 	int r;
1984 	struct dm_raid_superblock *sb;
1985 	struct dm_raid_superblock *refsb;
1986 	uint64_t events_sb, events_refsb;
1987 
1988 	rdev->sb_start = 0;
1989 	rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
1990 	if (rdev->sb_size < sizeof(*sb) || rdev->sb_size > PAGE_SIZE) {
1991 		DMERR("superblock size of a logical block is no longer valid");
1992 		return -EINVAL;
1993 	}
1994 
1995 	r = read_disk_sb(rdev, rdev->sb_size);
1996 	if (r)
1997 		return r;
1998 
1999 	sb = page_address(rdev->sb_page);
2000 
2001 	/*
2002 	 * Two cases that we want to write new superblocks and rebuild:
2003 	 * 1) New device (no matching magic number)
2004 	 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2005 	 */
2006 	if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
2007 	    (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
2008 		super_sync(rdev->mddev, rdev);
2009 
2010 		set_bit(FirstUse, &rdev->flags);
2011 		sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2012 
2013 		/* Force writing of superblocks to disk */
2014 		set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags);
2015 
2016 		/* Any superblock is better than none, choose that if given */
2017 		return refdev ? 0 : 1;
2018 	}
2019 
2020 	if (!refdev)
2021 		return 1;
2022 
2023 	events_sb = le64_to_cpu(sb->events);
2024 
2025 	refsb = page_address(refdev->sb_page);
2026 	events_refsb = le64_to_cpu(refsb->events);
2027 
2028 	return (events_sb > events_refsb) ? 1 : 0;
2029 }
2030 
2031 static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
2032 {
2033 	int role;
2034 	unsigned int d;
2035 	struct mddev *mddev = &rs->md;
2036 	uint64_t events_sb;
2037 	uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2038 	struct dm_raid_superblock *sb;
2039 	uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
2040 	struct md_rdev *r;
2041 	struct dm_raid_superblock *sb2;
2042 
2043 	sb = page_address(rdev->sb_page);
2044 	events_sb = le64_to_cpu(sb->events);
2045 
2046 	/*
2047 	 * Initialise to 1 if this is a new superblock.
2048 	 */
2049 	mddev->events = events_sb ? : 1;
2050 
2051 	mddev->reshape_position = MaxSector;
2052 
2053 	/*
2054 	 * Reshaping is supported, e.g. reshape_position is valid
2055 	 * in superblock and superblock content is authoritative.
2056 	 */
2057 	if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2058 		/* Superblock is authoritative wrt given raid set layout! */
2059 		mddev->raid_disks = le32_to_cpu(sb->num_devices);
2060 		mddev->level = le32_to_cpu(sb->level);
2061 		mddev->layout = le32_to_cpu(sb->layout);
2062 		mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
2063 		mddev->new_level = le32_to_cpu(sb->new_level);
2064 		mddev->new_layout = le32_to_cpu(sb->new_layout);
2065 		mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
2066 		mddev->delta_disks = le32_to_cpu(sb->delta_disks);
2067 		mddev->array_sectors = le64_to_cpu(sb->array_sectors);
2068 
2069 		/* raid was reshaping and got interrupted */
2070 		if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
2071 			if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
2072 				DMERR("Reshape requested but raid set is still reshaping");
2073 				return -EINVAL;
2074 			}
2075 
2076 			if (mddev->delta_disks < 0 ||
2077 			    (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
2078 				mddev->reshape_backwards = 1;
2079 			else
2080 				mddev->reshape_backwards = 0;
2081 
2082 			mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2083 			rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2084 		}
2085 
2086 	} else {
2087 		/*
2088 		 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2089 		 */
2090 		if (le32_to_cpu(sb->level) != mddev->level) {
2091 			DMERR("Reshaping/takeover raid sets not yet supported. (raid level/stripes/size change)");
2092 			return -EINVAL;
2093 		}
2094 		if (le32_to_cpu(sb->layout) != mddev->layout) {
2095 			DMERR("Reshaping raid sets not yet supported. (raid layout change)");
2096 			DMERR("	 0x%X vs 0x%X", le32_to_cpu(sb->layout), mddev->layout);
2097 			DMERR("	 Old layout: %s w/ %d copies",
2098 			      raid10_md_layout_to_format(le32_to_cpu(sb->layout)),
2099 			      raid10_md_layout_to_copies(le32_to_cpu(sb->layout)));
2100 			DMERR("	 New layout: %s w/ %d copies",
2101 			      raid10_md_layout_to_format(mddev->layout),
2102 			      raid10_md_layout_to_copies(mddev->layout));
2103 			return -EINVAL;
2104 		}
2105 		if (le32_to_cpu(sb->stripe_sectors) != mddev->chunk_sectors) {
2106 			DMERR("Reshaping raid sets not yet supported. (stripe sectors change)");
2107 			return -EINVAL;
2108 		}
2109 
2110 		/* We can only change the number of devices in raid1 with old (i.e. pre 1.0.7) metadata */
2111 		if (!rt_is_raid1(rs->raid_type) &&
2112 		    (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) {
2113 			DMERR("Reshaping raid sets not yet supported. (device count change from %u to %u)",
2114 			      sb->num_devices, mddev->raid_disks);
2115 			return -EINVAL;
2116 		}
2117 
2118 		/* Table line is checked vs. authoritative superblock */
2119 		rs_set_new(rs);
2120 	}
2121 
2122 	if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
2123 		mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2124 
2125 	/*
2126 	 * During load, we set FirstUse if a new superblock was written.
2127 	 * There are two reasons we might not have a superblock:
2128 	 * 1) The raid set is brand new - in which case, all of the
2129 	 *    devices must have their In_sync bit set.	Also,
2130 	 *    recovery_cp must be 0, unless forced.
2131 	 * 2) This is a new device being added to an old raid set
2132 	 *    and the new device needs to be rebuilt - in which
2133 	 *    case the In_sync bit will /not/ be set and
2134 	 *    recovery_cp must be MaxSector.
2135 	 * 3) This is/are a new device(s) being added to an old
2136 	 *    raid set during takeover to a higher raid level
2137 	 *    to provide capacity for redundancy or during reshape
2138 	 *    to add capacity to grow the raid set.
2139 	 */
2140 	d = 0;
2141 	rdev_for_each(r, mddev) {
2142 		if (test_bit(FirstUse, &r->flags))
2143 			new_devs++;
2144 
2145 		if (!test_bit(In_sync, &r->flags)) {
2146 			DMINFO("Device %d specified for rebuild; clearing superblock",
2147 				r->raid_disk);
2148 			rebuilds++;
2149 
2150 			if (test_bit(FirstUse, &r->flags))
2151 				rebuild_and_new++;
2152 		}
2153 
2154 		d++;
2155 	}
2156 
2157 	if (new_devs == rs->raid_disks || !rebuilds) {
2158 		/* Replace a broken device */
2159 		if (new_devs == 1 && !rs->delta_disks)
2160 			;
2161 		if (new_devs == rs->raid_disks) {
2162 			DMINFO("Superblocks created for new raid set");
2163 			set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2164 		} else if (new_devs != rebuilds &&
2165 			   new_devs != rs->delta_disks) {
2166 			DMERR("New device injected into existing raid set without "
2167 			      "'delta_disks' or 'rebuild' parameter specified");
2168 			return -EINVAL;
2169 		}
2170 	} else if (new_devs && new_devs != rebuilds) {
2171 		DMERR("%u 'rebuild' devices cannot be injected into"
2172 		      " a raid set with %u other first-time devices",
2173 		      rebuilds, new_devs);
2174 		return -EINVAL;
2175 	} else if (rebuilds) {
2176 		if (rebuild_and_new && rebuilds != rebuild_and_new) {
2177 			DMERR("new device%s provided without 'rebuild'",
2178 			      new_devs > 1 ? "s" : "");
2179 			return -EINVAL;
2180 		} else if (rs_is_recovering(rs)) {
2181 			DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2182 			      (unsigned long long) mddev->recovery_cp);
2183 			return -EINVAL;
2184 		} else if (rs_is_reshaping(rs)) {
2185 			DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2186 			      (unsigned long long) mddev->reshape_position);
2187 			return -EINVAL;
2188 		}
2189 	}
2190 
2191 	/*
2192 	 * Now we set the Faulty bit for those devices that are
2193 	 * recorded in the superblock as failed.
2194 	 */
2195 	sb_retrieve_failed_devices(sb, failed_devices);
2196 	rdev_for_each(r, mddev) {
2197 		if (!r->sb_page)
2198 			continue;
2199 		sb2 = page_address(r->sb_page);
2200 		sb2->failed_devices = 0;
2201 		memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
2202 
2203 		/*
2204 		 * Check for any device re-ordering.
2205 		 */
2206 		if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2207 			role = le32_to_cpu(sb2->array_position);
2208 			if (role < 0)
2209 				continue;
2210 
2211 			if (role != r->raid_disk) {
2212 				if (__is_raid10_near(mddev->layout)) {
2213 					if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
2214 					    rs->raid_disks % rs->raid10_copies) {
2215 						rs->ti->error =
2216 							"Cannot change raid10 near set to odd # of devices!";
2217 						return -EINVAL;
2218 					}
2219 
2220 					sb2->array_position = cpu_to_le32(r->raid_disk);
2221 
2222 				} else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
2223 					   !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2224 					   !rt_is_raid1(rs->raid_type)) {
2225 					rs->ti->error = "Cannot change device positions in raid set";
2226 					return -EINVAL;
2227 				}
2228 
2229 				DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
2230 			}
2231 
2232 			/*
2233 			 * Partial recovery is performed on
2234 			 * returning failed devices.
2235 			 */
2236 			if (test_bit(role, (void *) failed_devices))
2237 				set_bit(Faulty, &r->flags);
2238 		}
2239 	}
2240 
2241 	return 0;
2242 }
2243 
2244 static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
2245 {
2246 	struct mddev *mddev = &rs->md;
2247 	struct dm_raid_superblock *sb;
2248 
2249 	if (rs_is_raid0(rs) || !rdev->sb_page)
2250 		return 0;
2251 
2252 	sb = page_address(rdev->sb_page);
2253 
2254 	/*
2255 	 * If mddev->events is not set, we know we have not yet initialized
2256 	 * the array.
2257 	 */
2258 	if (!mddev->events && super_init_validation(rs, rdev))
2259 		return -EINVAL;
2260 
2261 	if (le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
2262 		rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2263 		return -EINVAL;
2264 	}
2265 
2266 	if (sb->incompat_features) {
2267 		rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
2268 		return -EINVAL;
2269 	}
2270 
2271 	/* Enable bitmap creation for RAID levels != 0 */
2272 	mddev->bitmap_info.offset = rt_is_raid0(rs->raid_type) ? 0 : to_sector(4096);
2273 	rdev->mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2274 
2275 	if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2276 		/* Retrieve device size stored in superblock to be prepared for shrink */
2277 		rdev->sectors = le64_to_cpu(sb->sectors);
2278 		rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
2279 		if (rdev->recovery_offset == MaxSector)
2280 			set_bit(In_sync, &rdev->flags);
2281 		/*
2282 		 * If no reshape in progress -> we're recovering single
2283 		 * disk(s) and have to set the device(s) to out-of-sync
2284 		 */
2285 		else if (!rs_is_reshaping(rs))
2286 			clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
2287 	}
2288 
2289 	/*
2290 	 * If a device comes back, set it as not In_sync and no longer faulty.
2291 	 */
2292 	if (test_and_clear_bit(Faulty, &rdev->flags)) {
2293 		rdev->recovery_offset = 0;
2294 		clear_bit(In_sync, &rdev->flags);
2295 		rdev->saved_raid_disk = rdev->raid_disk;
2296 	}
2297 
2298 	/* Reshape support -> restore repective data offsets */
2299 	rdev->data_offset = le64_to_cpu(sb->data_offset);
2300 	rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
2301 
2302 	return 0;
2303 }
2304 
2305 /*
2306  * Analyse superblocks and select the freshest.
2307  */
2308 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2309 {
2310 	int r;
2311 	struct raid_dev *dev;
2312 	struct md_rdev *rdev, *tmp, *freshest;
2313 	struct mddev *mddev = &rs->md;
2314 
2315 	freshest = NULL;
2316 	rdev_for_each_safe(rdev, tmp, mddev) {
2317 		/*
2318 		 * Skipping super_load due to CTR_FLAG_SYNC will cause
2319 		 * the array to undergo initialization again as
2320 		 * though it were new.	This is the intended effect
2321 		 * of the "sync" directive.
2322 		 *
2323 		 * When reshaping capability is added, we must ensure
2324 		 * that the "sync" directive is disallowed during the
2325 		 * reshape.
2326 		 */
2327 		if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2328 			continue;
2329 
2330 		if (!rdev->meta_bdev)
2331 			continue;
2332 
2333 		r = super_load(rdev, freshest);
2334 
2335 		switch (r) {
2336 		case 1:
2337 			freshest = rdev;
2338 			break;
2339 		case 0:
2340 			break;
2341 		default:
2342 			/*
2343 			 * We have to keep any raid0 data/metadata device pairs or
2344 			 * the MD raid0 personality will fail to start the array.
2345 			 */
2346 			if (rs_is_raid0(rs))
2347 				continue;
2348 
2349 			dev = container_of(rdev, struct raid_dev, rdev);
2350 			if (dev->meta_dev)
2351 				dm_put_device(ti, dev->meta_dev);
2352 
2353 			dev->meta_dev = NULL;
2354 			rdev->meta_bdev = NULL;
2355 
2356 			if (rdev->sb_page)
2357 				put_page(rdev->sb_page);
2358 
2359 			rdev->sb_page = NULL;
2360 
2361 			rdev->sb_loaded = 0;
2362 
2363 			/*
2364 			 * We might be able to salvage the data device
2365 			 * even though the meta device has failed.  For
2366 			 * now, we behave as though '- -' had been
2367 			 * set for this device in the table.
2368 			 */
2369 			if (dev->data_dev)
2370 				dm_put_device(ti, dev->data_dev);
2371 
2372 			dev->data_dev = NULL;
2373 			rdev->bdev = NULL;
2374 
2375 			list_del(&rdev->same_set);
2376 		}
2377 	}
2378 
2379 	if (!freshest)
2380 		return 0;
2381 
2382 	if (validate_raid_redundancy(rs)) {
2383 		rs->ti->error = "Insufficient redundancy to activate array";
2384 		return -EINVAL;
2385 	}
2386 
2387 	/*
2388 	 * Validation of the freshest device provides the source of
2389 	 * validation for the remaining devices.
2390 	 */
2391 	rs->ti->error = "Unable to assemble array: Invalid superblocks";
2392 	if (super_validate(rs, freshest))
2393 		return -EINVAL;
2394 
2395 	rdev_for_each(rdev, mddev)
2396 		if ((rdev != freshest) && super_validate(rs, rdev))
2397 			return -EINVAL;
2398 	return 0;
2399 }
2400 
2401 /*
2402  * Adjust data_offset and new_data_offset on all disk members of @rs
2403  * for out of place reshaping if requested by contructor
2404  *
2405  * We need free space at the beginning of each raid disk for forward
2406  * and at the end for backward reshapes which userspace has to provide
2407  * via remapping/reordering of space.
2408  */
2409 static int rs_adjust_data_offsets(struct raid_set *rs)
2410 {
2411 	sector_t data_offset = 0, new_data_offset = 0;
2412 	struct md_rdev *rdev;
2413 
2414 	/* Constructor did not request data offset change */
2415 	if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2416 		if (!rs_is_reshapable(rs))
2417 			goto out;
2418 
2419 		return 0;
2420 	}
2421 
2422 	/* HM FIXME: get InSync raid_dev? */
2423 	rdev = &rs->dev[0].rdev;
2424 
2425 	if (rs->delta_disks < 0) {
2426 		/*
2427 		 * Removing disks (reshaping backwards):
2428 		 *
2429 		 * - before reshape: data is at offset 0 and free space
2430 		 *		     is at end of each component LV
2431 		 *
2432 		 * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2433 		 */
2434 		data_offset = 0;
2435 		new_data_offset = rs->data_offset;
2436 
2437 	} else if (rs->delta_disks > 0) {
2438 		/*
2439 		 * Adding disks (reshaping forwards):
2440 		 *
2441 		 * - before reshape: data is at offset rs->data_offset != 0 and
2442 		 *		     free space is at begin of each component LV
2443 		 *
2444 		 * - after reshape: data is at offset 0 on each component LV
2445 		 */
2446 		data_offset = rs->data_offset;
2447 		new_data_offset = 0;
2448 
2449 	} else {
2450 		/*
2451 		 * User space passes in 0 for data offset after having removed reshape space
2452 		 *
2453 		 * - or - (data offset != 0)
2454 		 *
2455 		 * Changing RAID layout or chunk size -> toggle offsets
2456 		 *
2457 		 * - before reshape: data is at offset rs->data_offset 0 and
2458 		 *		     free space is at end of each component LV
2459 		 *		     -or-
2460 		 *                   data is at offset rs->data_offset != 0 and
2461 		 *		     free space is at begin of each component LV
2462 		 *
2463 		 * - after reshape: data is at offset 0 if it was at offset != 0
2464 		 *                  or at offset != 0 if it was at offset 0
2465 		 *                  on each component LV
2466 		 *
2467 		 */
2468 		data_offset = rs->data_offset ? rdev->data_offset : 0;
2469 		new_data_offset = data_offset ? 0 : rs->data_offset;
2470 		set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2471 	}
2472 
2473 	/*
2474 	 * Make sure we got a minimum amount of free sectors per device
2475 	 */
2476 	if (rs->data_offset &&
2477 	    to_sector(i_size_read(rdev->bdev->bd_inode)) - rdev->sectors < MIN_FREE_RESHAPE_SPACE) {
2478 		rs->ti->error = data_offset ? "No space for forward reshape" :
2479 					      "No space for backward reshape";
2480 		return -ENOSPC;
2481 	}
2482 out:
2483 	/* Adjust data offsets on all rdevs */
2484 	rdev_for_each(rdev, &rs->md) {
2485 		rdev->data_offset = data_offset;
2486 		rdev->new_data_offset = new_data_offset;
2487 	}
2488 
2489 	return 0;
2490 }
2491 
2492 /* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2493 static void __reorder_raid_disk_indexes(struct raid_set *rs)
2494 {
2495 	int i = 0;
2496 	struct md_rdev *rdev;
2497 
2498 	rdev_for_each(rdev, &rs->md) {
2499 		rdev->raid_disk = i++;
2500 		rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2501 	}
2502 }
2503 
2504 /*
2505  * Setup @rs for takeover by a different raid level
2506  */
2507 static int rs_setup_takeover(struct raid_set *rs)
2508 {
2509 	struct mddev *mddev = &rs->md;
2510 	struct md_rdev *rdev;
2511 	unsigned int d = mddev->raid_disks = rs->raid_disks;
2512 	sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2513 
2514 	if (rt_is_raid10(rs->raid_type)) {
2515 		if (mddev->level == 0) {
2516 			/* Userpace reordered disks -> adjust raid_disk indexes */
2517 			__reorder_raid_disk_indexes(rs);
2518 
2519 			/* raid0 -> raid10_far layout */
2520 			mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2521 								   rs->raid10_copies);
2522 		} else if (mddev->level == 1)
2523 			/* raid1 -> raid10_near layout */
2524 			mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2525 								   rs->raid_disks);
2526 		else
2527 			return -EINVAL;
2528 
2529 	}
2530 
2531 	clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2532 	mddev->recovery_cp = MaxSector;
2533 
2534 	while (d--) {
2535 		rdev = &rs->dev[d].rdev;
2536 
2537 		if (test_bit(d, (void *) rs->rebuild_disks)) {
2538 			clear_bit(In_sync, &rdev->flags);
2539 			clear_bit(Faulty, &rdev->flags);
2540 			mddev->recovery_cp = rdev->recovery_offset = 0;
2541 			/* Bitmap has to be created when we do an "up" takeover */
2542 			set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2543 		}
2544 
2545 		rdev->new_data_offset = new_data_offset;
2546 	}
2547 
2548 	return 0;
2549 }
2550 
2551 /* Prepare @rs for reshape */
2552 static int rs_prepare_reshape(struct raid_set *rs)
2553 {
2554 	bool reshape;
2555 	struct mddev *mddev = &rs->md;
2556 
2557 	if (rs_is_raid10(rs)) {
2558 		if (rs->raid_disks != mddev->raid_disks &&
2559 		    __is_raid10_near(mddev->layout) &&
2560 		    rs->raid10_copies &&
2561 		    rs->raid10_copies != __raid10_near_copies(mddev->layout)) {
2562 			/*
2563 			 * raid disk have to be multiple of data copies to allow this conversion,
2564 			 *
2565 			 * This is actually not a reshape it is a
2566 			 * rebuild of any additional mirrors per group
2567 			 */
2568 			if (rs->raid_disks % rs->raid10_copies) {
2569 				rs->ti->error = "Can't reshape raid10 mirror groups";
2570 				return -EINVAL;
2571 			}
2572 
2573 			/* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2574 			__reorder_raid_disk_indexes(rs);
2575 			mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2576 								   rs->raid10_copies);
2577 			mddev->new_layout = mddev->layout;
2578 			reshape = false;
2579 		} else
2580 			reshape = true;
2581 
2582 	} else if (rs_is_raid456(rs))
2583 		reshape = true;
2584 
2585 	else if (rs_is_raid1(rs)) {
2586 		if (rs->delta_disks) {
2587 			/* Process raid1 via delta_disks */
2588 			mddev->degraded = rs->delta_disks < 0 ? -rs->delta_disks : rs->delta_disks;
2589 			reshape = true;
2590 		} else {
2591 			/* Process raid1 without delta_disks */
2592 			mddev->raid_disks = rs->raid_disks;
2593 			reshape = false;
2594 		}
2595 	} else {
2596 		rs->ti->error = "Called with bogus raid type";
2597 		return -EINVAL;
2598 	}
2599 
2600 	if (reshape) {
2601 		set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2602 		set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2603 	} else if (mddev->raid_disks < rs->raid_disks)
2604 		/* Create new superblocks and bitmaps, if any new disks */
2605 		set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2606 
2607 	return 0;
2608 }
2609 
2610 /*
2611  *
2612  * - change raid layout
2613  * - change chunk size
2614  * - add disks
2615  * - remove disks
2616  */
2617 static int rs_setup_reshape(struct raid_set *rs)
2618 {
2619 	int r = 0;
2620 	unsigned int cur_raid_devs, d;
2621 	struct mddev *mddev = &rs->md;
2622 	struct md_rdev *rdev;
2623 
2624 	mddev->delta_disks = rs->delta_disks;
2625 	cur_raid_devs = mddev->raid_disks;
2626 
2627 	/* Ignore impossible layout change whilst adding/removing disks */
2628 	if (mddev->delta_disks &&
2629 	    mddev->layout != mddev->new_layout) {
2630 		DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2631 		mddev->new_layout = mddev->layout;
2632 	}
2633 
2634 	/*
2635 	 * Adjust array size:
2636 	 *
2637 	 * - in case of adding disks, array size has
2638 	 *   to grow after the disk adding reshape,
2639 	 *   which'll hapen in the event handler;
2640 	 *   reshape will happen forward, so space has to
2641 	 *   be available at the beginning of each disk
2642 	 *
2643 	 * - in case of removing disks, array size
2644 	 *   has to shrink before starting the reshape,
2645 	 *   which'll happen here;
2646 	 *   reshape will happen backward, so space has to
2647 	 *   be available at the end of each disk
2648 	 *
2649 	 * - data_offset and new_data_offset are
2650 	 *   adjusted for aforementioned out of place
2651 	 *   reshaping based on userspace passing in
2652 	 *   the "data_offset <sectors>" key/value
2653 	 *   pair via the constructor
2654 	 */
2655 
2656 	/* Add disk(s) */
2657 	if (rs->delta_disks > 0) {
2658 		/* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2659 		for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2660 			rdev = &rs->dev[d].rdev;
2661 			clear_bit(In_sync, &rdev->flags);
2662 
2663 			/*
2664 			 * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2665 			 * by md, which'll store that erroneously in the superblock on reshape
2666 			 */
2667 			rdev->saved_raid_disk = -1;
2668 			rdev->raid_disk = d;
2669 
2670 			rdev->sectors = mddev->dev_sectors;
2671 			rdev->recovery_offset = rs_is_raid1(rs) ? 0 : MaxSector;
2672 		}
2673 
2674 		mddev->reshape_backwards = 0; /* adding disks -> forward reshape */
2675 
2676 	/* Remove disk(s) */
2677 	} else if (rs->delta_disks < 0) {
2678 		r = rs_set_dev_and_array_sectors(rs, true);
2679 		mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2680 
2681 	/* Change layout and/or chunk size */
2682 	} else {
2683 		/*
2684 		 * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2685 		 *
2686 		 * keeping number of disks and do layout change ->
2687 		 *
2688 		 * toggle reshape_backward depending on data_offset:
2689 		 *
2690 		 * - free space upfront -> reshape forward
2691 		 *
2692 		 * - free space at the end -> reshape backward
2693 		 *
2694 		 *
2695 		 * This utilizes free reshape space avoiding the need
2696 		 * for userspace to move (parts of) LV segments in
2697 		 * case of layout/chunksize change  (for disk
2698 		 * adding/removing reshape space has to be at
2699 		 * the proper address (see above with delta_disks):
2700 		 *
2701 		 * add disk(s)   -> begin
2702 		 * remove disk(s)-> end
2703 		 */
2704 		mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2705 	}
2706 
2707 	return r;
2708 }
2709 
2710 /*
2711  * Enable/disable discard support on RAID set depending on
2712  * RAID level and discard properties of underlying RAID members.
2713  */
2714 static void configure_discard_support(struct raid_set *rs)
2715 {
2716 	int i;
2717 	bool raid456;
2718 	struct dm_target *ti = rs->ti;
2719 
2720 	/* Assume discards not supported until after checks below. */
2721 	ti->discards_supported = false;
2722 
2723 	/* RAID level 4,5,6 require discard_zeroes_data for data integrity! */
2724 	raid456 = (rs->md.level == 4 || rs->md.level == 5 || rs->md.level == 6);
2725 
2726 	for (i = 0; i < rs->raid_disks; i++) {
2727 		struct request_queue *q;
2728 
2729 		if (!rs->dev[i].rdev.bdev)
2730 			continue;
2731 
2732 		q = bdev_get_queue(rs->dev[i].rdev.bdev);
2733 		if (!q || !blk_queue_discard(q))
2734 			return;
2735 
2736 		if (raid456) {
2737 			if (!q->limits.discard_zeroes_data)
2738 				return;
2739 			if (!devices_handle_discard_safely) {
2740 				DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2741 				DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2742 				return;
2743 			}
2744 		}
2745 	}
2746 
2747 	/* All RAID members properly support discards */
2748 	ti->discards_supported = true;
2749 
2750 	/*
2751 	 * RAID1 and RAID10 personalities require bio splitting,
2752 	 * RAID0/4/5/6 don't and process large discard bios properly.
2753 	 */
2754 	ti->split_discard_bios = !!(rs->md.level == 1 || rs->md.level == 10);
2755 	ti->num_discard_bios = 1;
2756 }
2757 
2758 /*
2759  * Construct a RAID0/1/10/4/5/6 mapping:
2760  * Args:
2761  *	<raid_type> <#raid_params> <raid_params>{0,}	\
2762  *	<#raid_devs> [<meta_dev1> <dev1>]{1,}
2763  *
2764  * <raid_params> varies by <raid_type>.	 See 'parse_raid_params' for
2765  * details on possible <raid_params>.
2766  *
2767  * Userspace is free to initialize the metadata devices, hence the superblocks to
2768  * enforce recreation based on the passed in table parameters.
2769  *
2770  */
2771 static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2772 {
2773 	int r;
2774 	bool resize;
2775 	struct raid_type *rt;
2776 	unsigned int num_raid_params, num_raid_devs;
2777 	sector_t calculated_dev_sectors;
2778 	struct raid_set *rs = NULL;
2779 	const char *arg;
2780 	struct rs_layout rs_layout;
2781 	struct dm_arg_set as = { argc, argv }, as_nrd;
2782 	struct dm_arg _args[] = {
2783 		{ 0, as.argc, "Cannot understand number of raid parameters" },
2784 		{ 1, 254, "Cannot understand number of raid devices parameters" }
2785 	};
2786 
2787 	/* Must have <raid_type> */
2788 	arg = dm_shift_arg(&as);
2789 	if (!arg) {
2790 		ti->error = "No arguments";
2791 		return -EINVAL;
2792 	}
2793 
2794 	rt = get_raid_type(arg);
2795 	if (!rt) {
2796 		ti->error = "Unrecognised raid_type";
2797 		return -EINVAL;
2798 	}
2799 
2800 	/* Must have <#raid_params> */
2801 	if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
2802 		return -EINVAL;
2803 
2804 	/* number of raid device tupples <meta_dev data_dev> */
2805 	as_nrd = as;
2806 	dm_consume_args(&as_nrd, num_raid_params);
2807 	_args[1].max = (as_nrd.argc - 1) / 2;
2808 	if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
2809 		return -EINVAL;
2810 
2811 	if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
2812 		ti->error = "Invalid number of supplied raid devices";
2813 		return -EINVAL;
2814 	}
2815 
2816 	rs = raid_set_alloc(ti, rt, num_raid_devs);
2817 	if (IS_ERR(rs))
2818 		return PTR_ERR(rs);
2819 
2820 	r = parse_raid_params(rs, &as, num_raid_params);
2821 	if (r)
2822 		goto bad;
2823 
2824 	r = parse_dev_params(rs, &as);
2825 	if (r)
2826 		goto bad;
2827 
2828 	rs->md.sync_super = super_sync;
2829 
2830 	/*
2831 	 * Calculate ctr requested array and device sizes to allow
2832 	 * for superblock analysis needing device sizes defined.
2833 	 *
2834 	 * Any existing superblock will overwrite the array and device sizes
2835 	 */
2836 	r = rs_set_dev_and_array_sectors(rs, false);
2837 	if (r)
2838 		goto bad;
2839 
2840 	calculated_dev_sectors = rs->dev[0].rdev.sectors;
2841 
2842 	/*
2843 	 * Backup any new raid set level, layout, ...
2844 	 * requested to be able to compare to superblock
2845 	 * members for conversion decisions.
2846 	 */
2847 	rs_config_backup(rs, &rs_layout);
2848 
2849 	r = analyse_superblocks(ti, rs);
2850 	if (r)
2851 		goto bad;
2852 
2853 	resize = calculated_dev_sectors != rs->dev[0].rdev.sectors;
2854 
2855 	INIT_WORK(&rs->md.event_work, do_table_event);
2856 	ti->private = rs;
2857 	ti->num_flush_bios = 1;
2858 
2859 	/* Restore any requested new layout for conversion decision */
2860 	rs_config_restore(rs, &rs_layout);
2861 
2862 	/*
2863 	 * Now that we have any superblock metadata available,
2864 	 * check for new, recovering, reshaping, to be taken over,
2865 	 * to be reshaped or an existing, unchanged raid set to
2866 	 * run in sequence.
2867 	 */
2868 	if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
2869 		/* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
2870 		if (rs_is_raid6(rs) &&
2871 		    test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
2872 			ti->error = "'nosync' not allowed for new raid6 set";
2873 			r = -EINVAL;
2874 			goto bad;
2875 		}
2876 		rs_setup_recovery(rs, 0);
2877 		set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2878 		rs_set_new(rs);
2879 	} else if (rs_is_recovering(rs)) {
2880 		/* A recovering raid set may be resized */
2881 		; /* skip setup rs */
2882 	} else if (rs_is_reshaping(rs)) {
2883 		/* Have to reject size change request during reshape */
2884 		if (resize) {
2885 			ti->error = "Can't resize a reshaping raid set";
2886 			r = -EPERM;
2887 			goto bad;
2888 		}
2889 		/* skip setup rs */
2890 	} else if (rs_takeover_requested(rs)) {
2891 		if (rs_is_reshaping(rs)) {
2892 			ti->error = "Can't takeover a reshaping raid set";
2893 			r = -EPERM;
2894 			goto bad;
2895 		}
2896 
2897 		/*
2898 		 * If a takeover is needed, userspace sets any additional
2899 		 * devices to rebuild and we can check for a valid request here.
2900 		 *
2901 		 * If acceptible, set the level to the new requested
2902 		 * one, prohibit requesting recovery, allow the raid
2903 		 * set to run and store superblocks during resume.
2904 		 */
2905 		r = rs_check_takeover(rs);
2906 		if (r)
2907 			goto bad;
2908 
2909 		r = rs_setup_takeover(rs);
2910 		if (r)
2911 			goto bad;
2912 
2913 		set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2914 		/* Takeover ain't recovery, so disable recovery */
2915 		rs_setup_recovery(rs, MaxSector);
2916 		rs_set_new(rs);
2917 	} else if (rs_reshape_requested(rs)) {
2918 		/*
2919 		  * We can only prepare for a reshape here, because the
2920 		  * raid set needs to run to provide the repective reshape
2921 		  * check functions via its MD personality instance.
2922 		  *
2923 		  * So do the reshape check after md_run() succeeded.
2924 		  */
2925 		r = rs_prepare_reshape(rs);
2926 		if (r)
2927 			return r;
2928 
2929 		/* Reshaping ain't recovery, so disable recovery */
2930 		rs_setup_recovery(rs, MaxSector);
2931 		rs_set_cur(rs);
2932 	} else {
2933 		/* May not set recovery when a device rebuild is requested */
2934 		if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
2935 			rs_setup_recovery(rs, MaxSector);
2936 			set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2937 		} else
2938 			rs_setup_recovery(rs, test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ?
2939 					      0 : (resize ? calculated_dev_sectors : MaxSector));
2940 		rs_set_cur(rs);
2941 	}
2942 
2943 	/* If constructor requested it, change data and new_data offsets */
2944 	r = rs_adjust_data_offsets(rs);
2945 	if (r)
2946 		goto bad;
2947 
2948 	/* Start raid set read-only and assumed clean to change in raid_resume() */
2949 	rs->md.ro = 1;
2950 	rs->md.in_sync = 1;
2951 	set_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
2952 
2953 	/* Has to be held on running the array */
2954 	mddev_lock_nointr(&rs->md);
2955 	r = md_run(&rs->md);
2956 	rs->md.in_sync = 0; /* Assume already marked dirty */
2957 
2958 	if (r) {
2959 		ti->error = "Failed to run raid array";
2960 		mddev_unlock(&rs->md);
2961 		goto bad;
2962 	}
2963 
2964 	rs->callbacks.congested_fn = raid_is_congested;
2965 	dm_table_add_target_callbacks(ti->table, &rs->callbacks);
2966 
2967 	mddev_suspend(&rs->md);
2968 
2969 	/* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
2970 	if (rs_is_raid456(rs)) {
2971 		r = rs_set_raid456_stripe_cache(rs);
2972 		if (r)
2973 			goto bad_stripe_cache;
2974 	}
2975 
2976 	/* Now do an early reshape check */
2977 	if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
2978 		r = rs_check_reshape(rs);
2979 		if (r)
2980 			goto bad_check_reshape;
2981 
2982 		/* Restore new, ctr requested layout to perform check */
2983 		rs_config_restore(rs, &rs_layout);
2984 
2985 		if (rs->md.pers->start_reshape) {
2986 			r = rs->md.pers->check_reshape(&rs->md);
2987 			if (r) {
2988 				ti->error = "Reshape check failed";
2989 				goto bad_check_reshape;
2990 			}
2991 		}
2992 	}
2993 
2994 	mddev_unlock(&rs->md);
2995 	return 0;
2996 
2997 bad_stripe_cache:
2998 bad_check_reshape:
2999 	md_stop(&rs->md);
3000 bad:
3001 	raid_set_free(rs);
3002 
3003 	return r;
3004 }
3005 
3006 static void raid_dtr(struct dm_target *ti)
3007 {
3008 	struct raid_set *rs = ti->private;
3009 
3010 	list_del_init(&rs->callbacks.list);
3011 	md_stop(&rs->md);
3012 	raid_set_free(rs);
3013 }
3014 
3015 static int raid_map(struct dm_target *ti, struct bio *bio)
3016 {
3017 	struct raid_set *rs = ti->private;
3018 	struct mddev *mddev = &rs->md;
3019 
3020 	/*
3021 	 * If we're reshaping to add disk(s)), ti->len and
3022 	 * mddev->array_sectors will differ during the process
3023 	 * (ti->len > mddev->array_sectors), so we have to requeue
3024 	 * bios with addresses > mddev->array_sectors here or
3025 	 * there will occur accesses past EOD of the component
3026 	 * data images thus erroring the raid set.
3027 	 */
3028 	if (unlikely(bio_end_sector(bio) > mddev->array_sectors))
3029 		return DM_MAPIO_REQUEUE;
3030 
3031 	mddev->pers->make_request(mddev, bio);
3032 
3033 	return DM_MAPIO_SUBMITTED;
3034 }
3035 
3036 /* Return string describing the current sync action of @mddev */
3037 static const char *decipher_sync_action(struct mddev *mddev)
3038 {
3039 	if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3040 		return "frozen";
3041 
3042 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3043 	    (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3044 		if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3045 			return "reshape";
3046 
3047 		if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3048 			if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3049 				return "resync";
3050 			else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3051 				return "check";
3052 			return "repair";
3053 		}
3054 
3055 		if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3056 			return "recover";
3057 	}
3058 
3059 	return "idle";
3060 }
3061 
3062 /*
3063  * Return status string @rdev
3064  *
3065  * Status characters:
3066  *
3067  *  'D' = Dead/Failed device
3068  *  'a' = Alive but not in-sync
3069  *  'A' = Alive and in-sync
3070  */
3071 static const char *__raid_dev_status(struct md_rdev *rdev, bool array_in_sync)
3072 {
3073 	if (test_bit(Faulty, &rdev->flags))
3074 		return "D";
3075 	else if (!array_in_sync || !test_bit(In_sync, &rdev->flags))
3076 		return "a";
3077 	else
3078 		return "A";
3079 }
3080 
3081 /* Helper to return resync/reshape progress for @rs and @array_in_sync */
3082 static sector_t rs_get_progress(struct raid_set *rs,
3083 				sector_t resync_max_sectors, bool *array_in_sync)
3084 {
3085 	sector_t r, recovery_cp, curr_resync_completed;
3086 	struct mddev *mddev = &rs->md;
3087 
3088 	curr_resync_completed = mddev->curr_resync_completed ?: mddev->recovery_cp;
3089 	recovery_cp = mddev->recovery_cp;
3090 	*array_in_sync = false;
3091 
3092 	if (rs_is_raid0(rs)) {
3093 		r = resync_max_sectors;
3094 		*array_in_sync = true;
3095 
3096 	} else {
3097 		r = mddev->reshape_position;
3098 
3099 		/* Reshape is relative to the array size */
3100 		if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
3101 		    r != MaxSector) {
3102 			if (r == MaxSector) {
3103 				*array_in_sync = true;
3104 				r = resync_max_sectors;
3105 			} else {
3106 				/* Got to reverse on backward reshape */
3107 				if (mddev->reshape_backwards)
3108 					r = mddev->array_sectors - r;
3109 
3110 				/* Devide by # of data stripes */
3111 				sector_div(r, mddev_data_stripes(rs));
3112 			}
3113 
3114 		/* Sync is relative to the component device size */
3115 		} else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3116 			r = curr_resync_completed;
3117 		else
3118 			r = recovery_cp;
3119 
3120 		if (r == MaxSector) {
3121 			/*
3122 			 * Sync complete.
3123 			 */
3124 			*array_in_sync = true;
3125 			r = resync_max_sectors;
3126 		} else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
3127 			/*
3128 			 * If "check" or "repair" is occurring, the raid set has
3129 			 * undergone an initial sync and the health characters
3130 			 * should not be 'a' anymore.
3131 			 */
3132 			*array_in_sync = true;
3133 		} else {
3134 			struct md_rdev *rdev;
3135 
3136 			/*
3137 			 * The raid set may be doing an initial sync, or it may
3138 			 * be rebuilding individual components.	 If all the
3139 			 * devices are In_sync, then it is the raid set that is
3140 			 * being initialized.
3141 			 */
3142 			rdev_for_each(rdev, mddev)
3143 				if (!test_bit(In_sync, &rdev->flags))
3144 					*array_in_sync = true;
3145 #if 0
3146 			r = 0; /* HM FIXME: TESTME: https://bugzilla.redhat.com/show_bug.cgi?id=1210637 ? */
3147 #endif
3148 		}
3149 	}
3150 
3151 	return r;
3152 }
3153 
3154 /* Helper to return @dev name or "-" if !@dev */
3155 static const char *__get_dev_name(struct dm_dev *dev)
3156 {
3157 	return dev ? dev->name : "-";
3158 }
3159 
3160 static void raid_status(struct dm_target *ti, status_type_t type,
3161 			unsigned int status_flags, char *result, unsigned int maxlen)
3162 {
3163 	struct raid_set *rs = ti->private;
3164 	struct mddev *mddev = &rs->md;
3165 	struct r5conf *conf = mddev->private;
3166 	int i, max_nr_stripes = conf ? conf->max_nr_stripes : 0;
3167 	bool array_in_sync;
3168 	unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3169 	unsigned int sz = 0;
3170 	unsigned int rebuild_disks;
3171 	unsigned int write_mostly_params = 0;
3172 	sector_t progress, resync_max_sectors, resync_mismatches;
3173 	const char *sync_action;
3174 	struct raid_type *rt;
3175 	struct md_rdev *rdev;
3176 
3177 	switch (type) {
3178 	case STATUSTYPE_INFO:
3179 		/* *Should* always succeed */
3180 		rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3181 		if (!rt)
3182 			return;
3183 
3184 		DMEMIT("%s %d ", rt->name, mddev->raid_disks);
3185 
3186 		/* Access most recent mddev properties for status output */
3187 		smp_rmb();
3188 		/* Get sensible max sectors even if raid set not yet started */
3189 		resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
3190 				      mddev->resync_max_sectors : mddev->dev_sectors;
3191 		progress = rs_get_progress(rs, resync_max_sectors, &array_in_sync);
3192 		resync_mismatches = (mddev->last_sync_action && !strcasecmp(mddev->last_sync_action, "check")) ?
3193 				    atomic64_read(&mddev->resync_mismatches) : 0;
3194 		sync_action = decipher_sync_action(&rs->md);
3195 
3196 		/* HM FIXME: do we want another state char for raid0? It shows 'D' or 'A' now */
3197 		rdev_for_each(rdev, mddev)
3198 			DMEMIT(__raid_dev_status(rdev, array_in_sync));
3199 
3200 		/*
3201 		 * In-sync/Reshape ratio:
3202 		 *  The in-sync ratio shows the progress of:
3203 		 *   - Initializing the raid set
3204 		 *   - Rebuilding a subset of devices of the raid set
3205 		 *  The user can distinguish between the two by referring
3206 		 *  to the status characters.
3207 		 *
3208 		 *  The reshape ratio shows the progress of
3209 		 *  changing the raid layout or the number of
3210 		 *  disks of a raid set
3211 		 */
3212 		DMEMIT(" %llu/%llu", (unsigned long long) progress,
3213 				     (unsigned long long) resync_max_sectors);
3214 
3215 		/*
3216 		 * v1.5.0+:
3217 		 *
3218 		 * Sync action:
3219 		 *   See Documentation/device-mapper/dm-raid.txt for
3220 		 *   information on each of these states.
3221 		 */
3222 		DMEMIT(" %s", sync_action);
3223 
3224 		/*
3225 		 * v1.5.0+:
3226 		 *
3227 		 * resync_mismatches/mismatch_cnt
3228 		 *   This field shows the number of discrepancies found when
3229 		 *   performing a "check" of the raid set.
3230 		 */
3231 		DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3232 
3233 		/*
3234 		 * v1.9.0+:
3235 		 *
3236 		 * data_offset (needed for out of space reshaping)
3237 		 *   This field shows the data offset into the data
3238 		 *   image LV where the first stripes data starts.
3239 		 *
3240 		 * We keep data_offset equal on all raid disks of the set,
3241 		 * so retrieving it from the first raid disk is sufficient.
3242 		 */
3243 		DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
3244 		break;
3245 
3246 	case STATUSTYPE_TABLE:
3247 		/* Report the table line string you would use to construct this raid set */
3248 
3249 		/* Calculate raid parameter count */
3250 		for (i = 0; i < rs->raid_disks; i++)
3251 			if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3252 				write_mostly_params += 2;
3253 		rebuild_disks = memweight(rs->rebuild_disks, DISKS_ARRAY_ELEMS * sizeof(*rs->rebuild_disks));
3254 		raid_param_cnt += rebuild_disks * 2 +
3255 				  write_mostly_params +
3256 				  hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3257 				  hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2;
3258 		/* Emit table line */
3259 		DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
3260 		if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3261 			DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3262 					 raid10_md_layout_to_format(mddev->layout));
3263 		if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3264 			DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3265 					 raid10_md_layout_to_copies(mddev->layout));
3266 		if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3267 			DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
3268 		if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
3269 			DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
3270 		if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3271 			DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3272 					   (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3273 		if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3274 			DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3275 					   (unsigned long long) rs->data_offset);
3276 		if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3277 			DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3278 					  mddev->bitmap_info.daemon_sleep);
3279 		if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3280 			DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3281 					 max(rs->delta_disks, mddev->delta_disks));
3282 		if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3283 			DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3284 					 max_nr_stripes);
3285 		if (rebuild_disks)
3286 			for (i = 0; i < rs->raid_disks; i++)
3287 				if (test_bit(rs->dev[i].rdev.raid_disk, (void *) rs->rebuild_disks))
3288 					DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD),
3289 							 rs->dev[i].rdev.raid_disk);
3290 		if (write_mostly_params)
3291 			for (i = 0; i < rs->raid_disks; i++)
3292 				if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3293 					DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3294 					       rs->dev[i].rdev.raid_disk);
3295 		if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
3296 			DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
3297 					  mddev->bitmap_info.max_write_behind);
3298 		if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3299 			DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3300 					 mddev->sync_speed_max);
3301 		if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3302 			DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3303 					 mddev->sync_speed_min);
3304 		DMEMIT(" %d", rs->raid_disks);
3305 		for (i = 0; i < rs->raid_disks; i++)
3306 			DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
3307 					 __get_dev_name(rs->dev[i].data_dev));
3308 	}
3309 }
3310 
3311 static int raid_message(struct dm_target *ti, unsigned int argc, char **argv)
3312 {
3313 	struct raid_set *rs = ti->private;
3314 	struct mddev *mddev = &rs->md;
3315 
3316 	if (!mddev->pers || !mddev->pers->sync_request)
3317 		return -EINVAL;
3318 
3319 	if (!strcasecmp(argv[0], "frozen"))
3320 		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3321 	else
3322 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3323 
3324 	if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
3325 		if (mddev->sync_thread) {
3326 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3327 			md_reap_sync_thread(mddev);
3328 		}
3329 	} else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3330 		   test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3331 		return -EBUSY;
3332 	else if (!strcasecmp(argv[0], "resync"))
3333 		; /* MD_RECOVERY_NEEDED set below */
3334 	else if (!strcasecmp(argv[0], "recover"))
3335 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3336 	else {
3337 		if (!strcasecmp(argv[0], "check"))
3338 			set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3339 		else if (!!strcasecmp(argv[0], "repair"))
3340 			return -EINVAL;
3341 		set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3342 		set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3343 	}
3344 	if (mddev->ro == 2) {
3345 		/* A write to sync_action is enough to justify
3346 		 * canceling read-auto mode
3347 		 */
3348 		mddev->ro = 0;
3349 		if (!mddev->suspended && mddev->sync_thread)
3350 			md_wakeup_thread(mddev->sync_thread);
3351 	}
3352 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3353 	if (!mddev->suspended && mddev->thread)
3354 		md_wakeup_thread(mddev->thread);
3355 
3356 	return 0;
3357 }
3358 
3359 static int raid_iterate_devices(struct dm_target *ti,
3360 				iterate_devices_callout_fn fn, void *data)
3361 {
3362 	struct raid_set *rs = ti->private;
3363 	unsigned int i;
3364 	int r = 0;
3365 
3366 	for (i = 0; !r && i < rs->md.raid_disks; i++)
3367 		if (rs->dev[i].data_dev)
3368 			r = fn(ti,
3369 				 rs->dev[i].data_dev,
3370 				 0, /* No offset on data devs */
3371 				 rs->md.dev_sectors,
3372 				 data);
3373 
3374 	return r;
3375 }
3376 
3377 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3378 {
3379 	struct raid_set *rs = ti->private;
3380 	unsigned int chunk_size = to_bytes(rs->md.chunk_sectors);
3381 
3382 	blk_limits_io_min(limits, chunk_size);
3383 	blk_limits_io_opt(limits, chunk_size * mddev_data_stripes(rs));
3384 }
3385 
3386 static void raid_presuspend(struct dm_target *ti)
3387 {
3388 	struct raid_set *rs = ti->private;
3389 
3390 	md_stop_writes(&rs->md);
3391 }
3392 
3393 static void raid_postsuspend(struct dm_target *ti)
3394 {
3395 	struct raid_set *rs = ti->private;
3396 
3397 	if (!rs->md.suspended)
3398 		mddev_suspend(&rs->md);
3399 
3400 	rs->md.ro = 1;
3401 }
3402 
3403 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
3404 {
3405 	int i;
3406 	uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
3407 	unsigned long flags;
3408 	bool cleared = false;
3409 	struct dm_raid_superblock *sb;
3410 	struct mddev *mddev = &rs->md;
3411 	struct md_rdev *r;
3412 
3413 	/* RAID personalities have to provide hot add/remove methods or we need to bail out. */
3414 	if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
3415 		return;
3416 
3417 	memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
3418 
3419 	for (i = 0; i < rs->md.raid_disks; i++) {
3420 		r = &rs->dev[i].rdev;
3421 		if (test_bit(Faulty, &r->flags) && r->sb_page &&
3422 		    sync_page_io(r, 0, r->sb_size, r->sb_page,
3423 				 REQ_OP_READ, 0, true)) {
3424 			DMINFO("Faulty %s device #%d has readable super block."
3425 			       "  Attempting to revive it.",
3426 			       rs->raid_type->name, i);
3427 
3428 			/*
3429 			 * Faulty bit may be set, but sometimes the array can
3430 			 * be suspended before the personalities can respond
3431 			 * by removing the device from the array (i.e. calling
3432 			 * 'hot_remove_disk').	If they haven't yet removed
3433 			 * the failed device, its 'raid_disk' number will be
3434 			 * '>= 0' - meaning we must call this function
3435 			 * ourselves.
3436 			 */
3437 			if ((r->raid_disk >= 0) &&
3438 			    (mddev->pers->hot_remove_disk(mddev, r) != 0))
3439 				/* Failed to revive this device, try next */
3440 				continue;
3441 
3442 			r->raid_disk = i;
3443 			r->saved_raid_disk = i;
3444 			flags = r->flags;
3445 			clear_bit(Faulty, &r->flags);
3446 			clear_bit(WriteErrorSeen, &r->flags);
3447 			clear_bit(In_sync, &r->flags);
3448 			if (mddev->pers->hot_add_disk(mddev, r)) {
3449 				r->raid_disk = -1;
3450 				r->saved_raid_disk = -1;
3451 				r->flags = flags;
3452 			} else {
3453 				r->recovery_offset = 0;
3454 				set_bit(i, (void *) cleared_failed_devices);
3455 				cleared = true;
3456 			}
3457 		}
3458 	}
3459 
3460 	/* If any failed devices could be cleared, update all sbs failed_devices bits */
3461 	if (cleared) {
3462 		uint64_t failed_devices[DISKS_ARRAY_ELEMS];
3463 
3464 		rdev_for_each(r, &rs->md) {
3465 			sb = page_address(r->sb_page);
3466 			sb_retrieve_failed_devices(sb, failed_devices);
3467 
3468 			for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
3469 				failed_devices[i] &= ~cleared_failed_devices[i];
3470 
3471 			sb_update_failed_devices(sb, failed_devices);
3472 		}
3473 	}
3474 }
3475 
3476 static int __load_dirty_region_bitmap(struct raid_set *rs)
3477 {
3478 	int r = 0;
3479 
3480 	/* Try loading the bitmap unless "raid0", which does not have one */
3481 	if (!rs_is_raid0(rs) &&
3482 	    !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
3483 		r = bitmap_load(&rs->md);
3484 		if (r)
3485 			DMERR("Failed to load bitmap");
3486 	}
3487 
3488 	return r;
3489 }
3490 
3491 /* Enforce updating all superblocks */
3492 static void rs_update_sbs(struct raid_set *rs)
3493 {
3494 	struct mddev *mddev = &rs->md;
3495 	int ro = mddev->ro;
3496 
3497 	set_bit(MD_CHANGE_DEVS, &mddev->flags);
3498 	mddev->ro = 0;
3499 	md_update_sb(mddev, 1);
3500 	mddev->ro = ro;
3501 }
3502 
3503 /*
3504  * Reshape changes raid algorithm of @rs to new one within personality
3505  * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3506  * disks from a raid set thus growing/shrinking it or resizes the set
3507  *
3508  * Call mddev_lock_nointr() before!
3509  */
3510 static int rs_start_reshape(struct raid_set *rs)
3511 {
3512 	int r;
3513 	struct mddev *mddev = &rs->md;
3514 	struct md_personality *pers = mddev->pers;
3515 
3516 	r = rs_setup_reshape(rs);
3517 	if (r)
3518 		return r;
3519 
3520 	/* Need to be resumed to be able to start reshape, recovery is frozen until raid_resume() though */
3521 	if (mddev->suspended)
3522 		mddev_resume(mddev);
3523 
3524 	/*
3525 	 * Check any reshape constraints enforced by the personalility
3526 	 *
3527 	 * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3528 	 */
3529 	r = pers->check_reshape(mddev);
3530 	if (r) {
3531 		rs->ti->error = "pers->check_reshape() failed";
3532 		return r;
3533 	}
3534 
3535 	/*
3536 	 * Personality may not provide start reshape method in which
3537 	 * case check_reshape above has already covered everything
3538 	 */
3539 	if (pers->start_reshape) {
3540 		r = pers->start_reshape(mddev);
3541 		if (r) {
3542 			rs->ti->error = "pers->start_reshape() failed";
3543 			return r;
3544 		}
3545 	}
3546 
3547 	/* Suspend because a resume will happen in raid_resume() */
3548 	if (!mddev->suspended)
3549 		mddev_suspend(mddev);
3550 
3551 	/*
3552 	 * Now reshape got set up, update superblocks to
3553 	 * reflect the fact so that a table reload will
3554 	 * access proper superblock content in the ctr.
3555 	 */
3556 	rs_update_sbs(rs);
3557 
3558 	return 0;
3559 }
3560 
3561 static int raid_preresume(struct dm_target *ti)
3562 {
3563 	int r;
3564 	struct raid_set *rs = ti->private;
3565 	struct mddev *mddev = &rs->md;
3566 
3567 	/* This is a resume after a suspend of the set -> it's already started */
3568 	if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
3569 		return 0;
3570 
3571 	/*
3572 	 * The superblocks need to be updated on disk if the
3573 	 * array is new or new devices got added (thus zeroed
3574 	 * out by userspace) or __load_dirty_region_bitmap
3575 	 * will overwrite them in core with old data or fail.
3576 	 */
3577 	if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
3578 		rs_update_sbs(rs);
3579 
3580 	/*
3581 	 * Disable/enable discard support on raid set after any
3582 	 * conversion, because devices can have been added
3583 	 */
3584 	configure_discard_support(rs);
3585 
3586 	/* Load the bitmap from disk unless raid0 */
3587 	r = __load_dirty_region_bitmap(rs);
3588 	if (r)
3589 		return r;
3590 
3591 	/* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) */
3592 	if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) &&
3593 	    mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)) {
3594 		r = bitmap_resize(mddev->bitmap, mddev->dev_sectors,
3595 				  to_bytes(rs->requested_bitmap_chunk_sectors), 0);
3596 		if (r)
3597 			DMERR("Failed to resize bitmap");
3598 	}
3599 
3600 	/* Check for any resize/reshape on @rs and adjust/initiate */
3601 	/* Be prepared for mddev_resume() in raid_resume() */
3602 	set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3603 	if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
3604 		set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3605 		mddev->resync_min = mddev->recovery_cp;
3606 	}
3607 
3608 	rs_set_capacity(rs);
3609 
3610 	/* Check for any reshape request unless new raid set */
3611 	if (test_and_clear_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3612 		/* Initiate a reshape. */
3613 		mddev_lock_nointr(mddev);
3614 		r = rs_start_reshape(rs);
3615 		mddev_unlock(mddev);
3616 		if (r)
3617 			DMWARN("Failed to check/start reshape, continuing without change");
3618 		r = 0;
3619 	}
3620 
3621 	return r;
3622 }
3623 
3624 static void raid_resume(struct dm_target *ti)
3625 {
3626 	struct raid_set *rs = ti->private;
3627 	struct mddev *mddev = &rs->md;
3628 
3629 	if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
3630 		/*
3631 		 * A secondary resume while the device is active.
3632 		 * Take this opportunity to check whether any failed
3633 		 * devices are reachable again.
3634 		 */
3635 		attempt_restore_of_faulty_devices(rs);
3636 	}
3637 
3638 	mddev->ro = 0;
3639 	mddev->in_sync = 0;
3640 
3641 	clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3642 
3643 	if (mddev->suspended)
3644 		mddev_resume(mddev);
3645 }
3646 
3647 static struct target_type raid_target = {
3648 	.name = "raid",
3649 	.version = {1, 9, 0},
3650 	.module = THIS_MODULE,
3651 	.ctr = raid_ctr,
3652 	.dtr = raid_dtr,
3653 	.map = raid_map,
3654 	.status = raid_status,
3655 	.message = raid_message,
3656 	.iterate_devices = raid_iterate_devices,
3657 	.io_hints = raid_io_hints,
3658 	.presuspend = raid_presuspend,
3659 	.postsuspend = raid_postsuspend,
3660 	.preresume = raid_preresume,
3661 	.resume = raid_resume,
3662 };
3663 
3664 static int __init dm_raid_init(void)
3665 {
3666 	DMINFO("Loading target version %u.%u.%u",
3667 	       raid_target.version[0],
3668 	       raid_target.version[1],
3669 	       raid_target.version[2]);
3670 	return dm_register_target(&raid_target);
3671 }
3672 
3673 static void __exit dm_raid_exit(void)
3674 {
3675 	dm_unregister_target(&raid_target);
3676 }
3677 
3678 module_init(dm_raid_init);
3679 module_exit(dm_raid_exit);
3680 
3681 module_param(devices_handle_discard_safely, bool, 0644);
3682 MODULE_PARM_DESC(devices_handle_discard_safely,
3683 		 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
3684 
3685 MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
3686 MODULE_ALIAS("dm-raid0");
3687 MODULE_ALIAS("dm-raid1");
3688 MODULE_ALIAS("dm-raid10");
3689 MODULE_ALIAS("dm-raid4");
3690 MODULE_ALIAS("dm-raid5");
3691 MODULE_ALIAS("dm-raid6");
3692 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
3693 MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
3694 MODULE_LICENSE("GPL");
3695