xref: /openbmc/linux/drivers/md/dm-raid.c (revision 8730046c)
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_STRIPE_CACHE | \
164 				 CTR_FLAG_REGION_SIZE | \
165 				 CTR_FLAG_DELTA_DISKS | \
166 				 CTR_FLAG_DATA_OFFSET)
167 
168 #define RAID6_VALID_FLAGS	(CTR_FLAG_SYNC | \
169 				 CTR_FLAG_REBUILD | \
170 				 CTR_FLAG_DAEMON_SLEEP | \
171 				 CTR_FLAG_MIN_RECOVERY_RATE | \
172 				 CTR_FLAG_MAX_RECOVERY_RATE | \
173 				 CTR_FLAG_STRIPE_CACHE | \
174 				 CTR_FLAG_REGION_SIZE | \
175 				 CTR_FLAG_DELTA_DISKS | \
176 				 CTR_FLAG_DATA_OFFSET)
177 /* ...valid options definitions per raid level */
178 
179 /*
180  * Flags for rs->runtime_flags field
181  * (RT_FLAG prefix meaning "runtime flag")
182  *
183  * These are all internal and used to define runtime state,
184  * e.g. to prevent another resume from preresume processing
185  * the raid set all over again.
186  */
187 #define RT_FLAG_RS_PRERESUMED		0
188 #define RT_FLAG_RS_RESUMED		1
189 #define RT_FLAG_RS_BITMAP_LOADED	2
190 #define RT_FLAG_UPDATE_SBS		3
191 #define RT_FLAG_RESHAPE_RS		4
192 
193 /* Array elements of 64 bit needed for rebuild/failed disk bits */
194 #define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
195 
196 /*
197  * raid set level, layout and chunk sectors backup/restore
198  */
199 struct rs_layout {
200 	int new_level;
201 	int new_layout;
202 	int new_chunk_sectors;
203 };
204 
205 struct raid_set {
206 	struct dm_target *ti;
207 
208 	uint32_t bitmap_loaded;
209 	uint32_t stripe_cache_entries;
210 	unsigned long ctr_flags;
211 	unsigned long runtime_flags;
212 
213 	uint64_t rebuild_disks[DISKS_ARRAY_ELEMS];
214 
215 	int raid_disks;
216 	int delta_disks;
217 	int data_offset;
218 	int raid10_copies;
219 	int requested_bitmap_chunk_sectors;
220 
221 	struct mddev md;
222 	struct raid_type *raid_type;
223 	struct dm_target_callbacks callbacks;
224 
225 	struct raid_dev dev[0];
226 };
227 
228 static void rs_config_backup(struct raid_set *rs, struct rs_layout *l)
229 {
230 	struct mddev *mddev = &rs->md;
231 
232 	l->new_level = mddev->new_level;
233 	l->new_layout = mddev->new_layout;
234 	l->new_chunk_sectors = mddev->new_chunk_sectors;
235 }
236 
237 static void rs_config_restore(struct raid_set *rs, struct rs_layout *l)
238 {
239 	struct mddev *mddev = &rs->md;
240 
241 	mddev->new_level = l->new_level;
242 	mddev->new_layout = l->new_layout;
243 	mddev->new_chunk_sectors = l->new_chunk_sectors;
244 }
245 
246 /* raid10 algorithms (i.e. formats) */
247 #define	ALGORITHM_RAID10_DEFAULT	0
248 #define	ALGORITHM_RAID10_NEAR		1
249 #define	ALGORITHM_RAID10_OFFSET		2
250 #define	ALGORITHM_RAID10_FAR		3
251 
252 /* Supported raid types and properties. */
253 static struct raid_type {
254 	const char *name;		/* RAID algorithm. */
255 	const char *descr;		/* Descriptor text for logging. */
256 	const unsigned int parity_devs;	/* # of parity devices. */
257 	const unsigned int minimal_devs;/* minimal # of devices in set. */
258 	const unsigned int level;	/* RAID level. */
259 	const unsigned int algorithm;	/* RAID algorithm. */
260 } raid_types[] = {
261 	{"raid0",	  "raid0 (striping)",			    0, 2, 0,  0 /* NONE */},
262 	{"raid1",	  "raid1 (mirroring)",			    0, 2, 1,  0 /* NONE */},
263 	{"raid10_far",	  "raid10 far (striped mirrors)",	    0, 2, 10, ALGORITHM_RAID10_FAR},
264 	{"raid10_offset", "raid10 offset (striped mirrors)",	    0, 2, 10, ALGORITHM_RAID10_OFFSET},
265 	{"raid10_near",	  "raid10 near (striped mirrors)",	    0, 2, 10, ALGORITHM_RAID10_NEAR},
266 	{"raid10",	  "raid10 (striped mirrors)",		    0, 2, 10, ALGORITHM_RAID10_DEFAULT},
267 	{"raid4",	  "raid4 (dedicated first parity disk)",    1, 2, 5,  ALGORITHM_PARITY_0}, /* raid4 layout = raid5_0 */
268 	{"raid5_n",	  "raid5 (dedicated last parity disk)",	    1, 2, 5,  ALGORITHM_PARITY_N},
269 	{"raid5_ls",	  "raid5 (left symmetric)",		    1, 2, 5,  ALGORITHM_LEFT_SYMMETRIC},
270 	{"raid5_rs",	  "raid5 (right symmetric)",		    1, 2, 5,  ALGORITHM_RIGHT_SYMMETRIC},
271 	{"raid5_la",	  "raid5 (left asymmetric)",		    1, 2, 5,  ALGORITHM_LEFT_ASYMMETRIC},
272 	{"raid5_ra",	  "raid5 (right asymmetric)",		    1, 2, 5,  ALGORITHM_RIGHT_ASYMMETRIC},
273 	{"raid6_zr",	  "raid6 (zero restart)",		    2, 4, 6,  ALGORITHM_ROTATING_ZERO_RESTART},
274 	{"raid6_nr",	  "raid6 (N restart)",			    2, 4, 6,  ALGORITHM_ROTATING_N_RESTART},
275 	{"raid6_nc",	  "raid6 (N continue)",			    2, 4, 6,  ALGORITHM_ROTATING_N_CONTINUE},
276 	{"raid6_n_6",	  "raid6 (dedicated parity/Q n/6)",	    2, 4, 6,  ALGORITHM_PARITY_N_6},
277 	{"raid6_ls_6",	  "raid6 (left symmetric dedicated Q 6)",   2, 4, 6,  ALGORITHM_LEFT_SYMMETRIC_6},
278 	{"raid6_rs_6",	  "raid6 (right symmetric dedicated Q 6)",  2, 4, 6,  ALGORITHM_RIGHT_SYMMETRIC_6},
279 	{"raid6_la_6",	  "raid6 (left asymmetric dedicated Q 6)",  2, 4, 6,  ALGORITHM_LEFT_ASYMMETRIC_6},
280 	{"raid6_ra_6",	  "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6,  ALGORITHM_RIGHT_ASYMMETRIC_6}
281 };
282 
283 /* True, if @v is in inclusive range [@min, @max] */
284 static bool __within_range(long v, long min, long max)
285 {
286 	return v >= min && v <= max;
287 }
288 
289 /* All table line arguments are defined here */
290 static struct arg_name_flag {
291 	const unsigned long flag;
292 	const char *name;
293 } __arg_name_flags[] = {
294 	{ CTR_FLAG_SYNC, "sync"},
295 	{ CTR_FLAG_NOSYNC, "nosync"},
296 	{ CTR_FLAG_REBUILD, "rebuild"},
297 	{ CTR_FLAG_DAEMON_SLEEP, "daemon_sleep"},
298 	{ CTR_FLAG_MIN_RECOVERY_RATE, "min_recovery_rate"},
299 	{ CTR_FLAG_MAX_RECOVERY_RATE, "max_recovery_rate"},
300 	{ CTR_FLAG_MAX_WRITE_BEHIND, "max_write_behind"},
301 	{ CTR_FLAG_WRITE_MOSTLY, "write_mostly"},
302 	{ CTR_FLAG_STRIPE_CACHE, "stripe_cache"},
303 	{ CTR_FLAG_REGION_SIZE, "region_size"},
304 	{ CTR_FLAG_RAID10_COPIES, "raid10_copies"},
305 	{ CTR_FLAG_RAID10_FORMAT, "raid10_format"},
306 	{ CTR_FLAG_DATA_OFFSET, "data_offset"},
307 	{ CTR_FLAG_DELTA_DISKS, "delta_disks"},
308 	{ CTR_FLAG_RAID10_USE_NEAR_SETS, "raid10_use_near_sets"},
309 };
310 
311 /* Return argument name string for given @flag */
312 static const char *dm_raid_arg_name_by_flag(const uint32_t flag)
313 {
314 	if (hweight32(flag) == 1) {
315 		struct arg_name_flag *anf = __arg_name_flags + ARRAY_SIZE(__arg_name_flags);
316 
317 		while (anf-- > __arg_name_flags)
318 			if (flag & anf->flag)
319 				return anf->name;
320 
321 	} else
322 		DMERR("%s called with more than one flag!", __func__);
323 
324 	return NULL;
325 }
326 
327 /*
328  * Bool helpers to test for various raid levels of a raid set.
329  * It's level as reported by the superblock rather than
330  * the requested raid_type passed to the constructor.
331  */
332 /* Return true, if raid set in @rs is raid0 */
333 static bool rs_is_raid0(struct raid_set *rs)
334 {
335 	return !rs->md.level;
336 }
337 
338 /* Return true, if raid set in @rs is raid1 */
339 static bool rs_is_raid1(struct raid_set *rs)
340 {
341 	return rs->md.level == 1;
342 }
343 
344 /* Return true, if raid set in @rs is raid10 */
345 static bool rs_is_raid10(struct raid_set *rs)
346 {
347 	return rs->md.level == 10;
348 }
349 
350 /* Return true, if raid set in @rs is level 6 */
351 static bool rs_is_raid6(struct raid_set *rs)
352 {
353 	return rs->md.level == 6;
354 }
355 
356 /* Return true, if raid set in @rs is level 4, 5 or 6 */
357 static bool rs_is_raid456(struct raid_set *rs)
358 {
359 	return __within_range(rs->md.level, 4, 6);
360 }
361 
362 /* Return true, if raid set in @rs is reshapable */
363 static bool __is_raid10_far(int layout);
364 static bool rs_is_reshapable(struct raid_set *rs)
365 {
366 	return rs_is_raid456(rs) ||
367 	       (rs_is_raid10(rs) && !__is_raid10_far(rs->md.new_layout));
368 }
369 
370 /* Return true, if raid set in @rs is recovering */
371 static bool rs_is_recovering(struct raid_set *rs)
372 {
373 	return rs->md.recovery_cp < rs->dev[0].rdev.sectors;
374 }
375 
376 /* Return true, if raid set in @rs is reshaping */
377 static bool rs_is_reshaping(struct raid_set *rs)
378 {
379 	return rs->md.reshape_position != MaxSector;
380 }
381 
382 /*
383  * bool helpers to test for various raid levels of a raid type @rt
384  */
385 
386 /* Return true, if raid type in @rt is raid0 */
387 static bool rt_is_raid0(struct raid_type *rt)
388 {
389 	return !rt->level;
390 }
391 
392 /* Return true, if raid type in @rt is raid1 */
393 static bool rt_is_raid1(struct raid_type *rt)
394 {
395 	return rt->level == 1;
396 }
397 
398 /* Return true, if raid type in @rt is raid10 */
399 static bool rt_is_raid10(struct raid_type *rt)
400 {
401 	return rt->level == 10;
402 }
403 
404 /* Return true, if raid type in @rt is raid4/5 */
405 static bool rt_is_raid45(struct raid_type *rt)
406 {
407 	return __within_range(rt->level, 4, 5);
408 }
409 
410 /* Return true, if raid type in @rt is raid6 */
411 static bool rt_is_raid6(struct raid_type *rt)
412 {
413 	return rt->level == 6;
414 }
415 
416 /* Return true, if raid type in @rt is raid4/5/6 */
417 static bool rt_is_raid456(struct raid_type *rt)
418 {
419 	return __within_range(rt->level, 4, 6);
420 }
421 /* END: raid level bools */
422 
423 /* Return valid ctr flags for the raid level of @rs */
424 static unsigned long __valid_flags(struct raid_set *rs)
425 {
426 	if (rt_is_raid0(rs->raid_type))
427 		return RAID0_VALID_FLAGS;
428 	else if (rt_is_raid1(rs->raid_type))
429 		return RAID1_VALID_FLAGS;
430 	else if (rt_is_raid10(rs->raid_type))
431 		return RAID10_VALID_FLAGS;
432 	else if (rt_is_raid45(rs->raid_type))
433 		return RAID45_VALID_FLAGS;
434 	else if (rt_is_raid6(rs->raid_type))
435 		return RAID6_VALID_FLAGS;
436 
437 	return 0;
438 }
439 
440 /*
441  * Check for valid flags set on @rs
442  *
443  * Has to be called after parsing of the ctr flags!
444  */
445 static int rs_check_for_valid_flags(struct raid_set *rs)
446 {
447 	if (rs->ctr_flags & ~__valid_flags(rs)) {
448 		rs->ti->error = "Invalid flags combination";
449 		return -EINVAL;
450 	}
451 
452 	return 0;
453 }
454 
455 /* MD raid10 bit definitions and helpers */
456 #define RAID10_OFFSET			(1 << 16) /* stripes with data copies area adjacent on devices */
457 #define RAID10_BROCKEN_USE_FAR_SETS	(1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
458 #define RAID10_USE_FAR_SETS		(1 << 18) /* Use sets instead of whole stripe rotation */
459 #define RAID10_FAR_COPIES_SHIFT		8	  /* raid10 # far copies shift (2nd byte of layout) */
460 
461 /* Return md raid10 near copies for @layout */
462 static unsigned int __raid10_near_copies(int layout)
463 {
464 	return layout & 0xFF;
465 }
466 
467 /* Return md raid10 far copies for @layout */
468 static unsigned int __raid10_far_copies(int layout)
469 {
470 	return __raid10_near_copies(layout >> RAID10_FAR_COPIES_SHIFT);
471 }
472 
473 /* Return true if md raid10 offset for @layout */
474 static bool __is_raid10_offset(int layout)
475 {
476 	return !!(layout & RAID10_OFFSET);
477 }
478 
479 /* Return true if md raid10 near for @layout */
480 static bool __is_raid10_near(int layout)
481 {
482 	return !__is_raid10_offset(layout) && __raid10_near_copies(layout) > 1;
483 }
484 
485 /* Return true if md raid10 far for @layout */
486 static bool __is_raid10_far(int layout)
487 {
488 	return !__is_raid10_offset(layout) && __raid10_far_copies(layout) > 1;
489 }
490 
491 /* Return md raid10 layout string for @layout */
492 static const char *raid10_md_layout_to_format(int layout)
493 {
494 	/*
495 	 * Bit 16 stands for "offset"
496 	 * (i.e. adjacent stripes hold copies)
497 	 *
498 	 * Refer to MD's raid10.c for details
499 	 */
500 	if (__is_raid10_offset(layout))
501 		return "offset";
502 
503 	if (__raid10_near_copies(layout) > 1)
504 		return "near";
505 
506 	WARN_ON(__raid10_far_copies(layout) < 2);
507 
508 	return "far";
509 }
510 
511 /* Return md raid10 algorithm for @name */
512 static int raid10_name_to_format(const char *name)
513 {
514 	if (!strcasecmp(name, "near"))
515 		return ALGORITHM_RAID10_NEAR;
516 	else if (!strcasecmp(name, "offset"))
517 		return ALGORITHM_RAID10_OFFSET;
518 	else if (!strcasecmp(name, "far"))
519 		return ALGORITHM_RAID10_FAR;
520 
521 	return -EINVAL;
522 }
523 
524 /* Return md raid10 copies for @layout */
525 static unsigned int raid10_md_layout_to_copies(int layout)
526 {
527 	return max(__raid10_near_copies(layout), __raid10_far_copies(layout));
528 }
529 
530 /* Return md raid10 format id for @format string */
531 static int raid10_format_to_md_layout(struct raid_set *rs,
532 				      unsigned int algorithm,
533 				      unsigned int copies)
534 {
535 	unsigned int n = 1, f = 1, r = 0;
536 
537 	/*
538 	 * MD resilienece flaw:
539 	 *
540 	 * enabling use_far_sets for far/offset formats causes copies
541 	 * to be colocated on the same devs together with their origins!
542 	 *
543 	 * -> disable it for now in the definition above
544 	 */
545 	if (algorithm == ALGORITHM_RAID10_DEFAULT ||
546 	    algorithm == ALGORITHM_RAID10_NEAR)
547 		n = copies;
548 
549 	else if (algorithm == ALGORITHM_RAID10_OFFSET) {
550 		f = copies;
551 		r = RAID10_OFFSET;
552 		if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
553 			r |= RAID10_USE_FAR_SETS;
554 
555 	} else if (algorithm == ALGORITHM_RAID10_FAR) {
556 		f = copies;
557 		r = !RAID10_OFFSET;
558 		if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
559 			r |= RAID10_USE_FAR_SETS;
560 
561 	} else
562 		return -EINVAL;
563 
564 	return r | (f << RAID10_FAR_COPIES_SHIFT) | n;
565 }
566 /* END: MD raid10 bit definitions and helpers */
567 
568 /* Check for any of the raid10 algorithms */
569 static bool __got_raid10(struct raid_type *rtp, const int layout)
570 {
571 	if (rtp->level == 10) {
572 		switch (rtp->algorithm) {
573 		case ALGORITHM_RAID10_DEFAULT:
574 		case ALGORITHM_RAID10_NEAR:
575 			return __is_raid10_near(layout);
576 		case ALGORITHM_RAID10_OFFSET:
577 			return __is_raid10_offset(layout);
578 		case ALGORITHM_RAID10_FAR:
579 			return __is_raid10_far(layout);
580 		default:
581 			break;
582 		}
583 	}
584 
585 	return false;
586 }
587 
588 /* Return raid_type for @name */
589 static struct raid_type *get_raid_type(const char *name)
590 {
591 	struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
592 
593 	while (rtp-- > raid_types)
594 		if (!strcasecmp(rtp->name, name))
595 			return rtp;
596 
597 	return NULL;
598 }
599 
600 /* Return raid_type for @name based derived from @level and @layout */
601 static struct raid_type *get_raid_type_by_ll(const int level, const int layout)
602 {
603 	struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
604 
605 	while (rtp-- > raid_types) {
606 		/* RAID10 special checks based on @layout flags/properties */
607 		if (rtp->level == level &&
608 		    (__got_raid10(rtp, layout) || rtp->algorithm == layout))
609 			return rtp;
610 	}
611 
612 	return NULL;
613 }
614 
615 /*
616  * Conditionally change bdev capacity of @rs
617  * in case of a disk add/remove reshape
618  */
619 static void rs_set_capacity(struct raid_set *rs)
620 {
621 	struct mddev *mddev = &rs->md;
622 	struct md_rdev *rdev;
623 	struct gendisk *gendisk = dm_disk(dm_table_get_md(rs->ti->table));
624 
625 	/*
626 	 * raid10 sets rdev->sector to the device size, which
627 	 * is unintended in case of out-of-place reshaping
628 	 */
629 	rdev_for_each(rdev, mddev)
630 		rdev->sectors = mddev->dev_sectors;
631 
632 	set_capacity(gendisk, mddev->array_sectors);
633 	revalidate_disk(gendisk);
634 }
635 
636 /*
637  * Set the mddev properties in @rs to the current
638  * ones retrieved from the freshest superblock
639  */
640 static void rs_set_cur(struct raid_set *rs)
641 {
642 	struct mddev *mddev = &rs->md;
643 
644 	mddev->new_level = mddev->level;
645 	mddev->new_layout = mddev->layout;
646 	mddev->new_chunk_sectors = mddev->chunk_sectors;
647 }
648 
649 /*
650  * Set the mddev properties in @rs to the new
651  * ones requested by the ctr
652  */
653 static void rs_set_new(struct raid_set *rs)
654 {
655 	struct mddev *mddev = &rs->md;
656 
657 	mddev->level = mddev->new_level;
658 	mddev->layout = mddev->new_layout;
659 	mddev->chunk_sectors = mddev->new_chunk_sectors;
660 	mddev->raid_disks = rs->raid_disks;
661 	mddev->delta_disks = 0;
662 }
663 
664 static struct raid_set *raid_set_alloc(struct dm_target *ti, struct raid_type *raid_type,
665 				       unsigned int raid_devs)
666 {
667 	unsigned int i;
668 	struct raid_set *rs;
669 
670 	if (raid_devs <= raid_type->parity_devs) {
671 		ti->error = "Insufficient number of devices";
672 		return ERR_PTR(-EINVAL);
673 	}
674 
675 	rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
676 	if (!rs) {
677 		ti->error = "Cannot allocate raid context";
678 		return ERR_PTR(-ENOMEM);
679 	}
680 
681 	mddev_init(&rs->md);
682 
683 	rs->raid_disks = raid_devs;
684 	rs->delta_disks = 0;
685 
686 	rs->ti = ti;
687 	rs->raid_type = raid_type;
688 	rs->stripe_cache_entries = 256;
689 	rs->md.raid_disks = raid_devs;
690 	rs->md.level = raid_type->level;
691 	rs->md.new_level = rs->md.level;
692 	rs->md.layout = raid_type->algorithm;
693 	rs->md.new_layout = rs->md.layout;
694 	rs->md.delta_disks = 0;
695 	rs->md.recovery_cp = MaxSector;
696 
697 	for (i = 0; i < raid_devs; i++)
698 		md_rdev_init(&rs->dev[i].rdev);
699 
700 	/*
701 	 * Remaining items to be initialized by further RAID params:
702 	 *  rs->md.persistent
703 	 *  rs->md.external
704 	 *  rs->md.chunk_sectors
705 	 *  rs->md.new_chunk_sectors
706 	 *  rs->md.dev_sectors
707 	 */
708 
709 	return rs;
710 }
711 
712 static void raid_set_free(struct raid_set *rs)
713 {
714 	int i;
715 
716 	for (i = 0; i < rs->raid_disks; i++) {
717 		if (rs->dev[i].meta_dev)
718 			dm_put_device(rs->ti, rs->dev[i].meta_dev);
719 		md_rdev_clear(&rs->dev[i].rdev);
720 		if (rs->dev[i].data_dev)
721 			dm_put_device(rs->ti, rs->dev[i].data_dev);
722 	}
723 
724 	kfree(rs);
725 }
726 
727 /*
728  * For every device we have two words
729  *  <meta_dev>: meta device name or '-' if missing
730  *  <data_dev>: data device name or '-' if missing
731  *
732  * The following are permitted:
733  *    - -
734  *    - <data_dev>
735  *    <meta_dev> <data_dev>
736  *
737  * The following is not allowed:
738  *    <meta_dev> -
739  *
740  * This code parses those words.  If there is a failure,
741  * the caller must use raid_set_free() to unwind the operations.
742  */
743 static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
744 {
745 	int i;
746 	int rebuild = 0;
747 	int metadata_available = 0;
748 	int r = 0;
749 	const char *arg;
750 
751 	/* Put off the number of raid devices argument to get to dev pairs */
752 	arg = dm_shift_arg(as);
753 	if (!arg)
754 		return -EINVAL;
755 
756 	for (i = 0; i < rs->raid_disks; i++) {
757 		rs->dev[i].rdev.raid_disk = i;
758 
759 		rs->dev[i].meta_dev = NULL;
760 		rs->dev[i].data_dev = NULL;
761 
762 		/*
763 		 * There are no offsets, since there is a separate device
764 		 * for data and metadata.
765 		 */
766 		rs->dev[i].rdev.data_offset = 0;
767 		rs->dev[i].rdev.mddev = &rs->md;
768 
769 		arg = dm_shift_arg(as);
770 		if (!arg)
771 			return -EINVAL;
772 
773 		if (strcmp(arg, "-")) {
774 			r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
775 					  &rs->dev[i].meta_dev);
776 			if (r) {
777 				rs->ti->error = "RAID metadata device lookup failure";
778 				return r;
779 			}
780 
781 			rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
782 			if (!rs->dev[i].rdev.sb_page) {
783 				rs->ti->error = "Failed to allocate superblock page";
784 				return -ENOMEM;
785 			}
786 		}
787 
788 		arg = dm_shift_arg(as);
789 		if (!arg)
790 			return -EINVAL;
791 
792 		if (!strcmp(arg, "-")) {
793 			if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
794 			    (!rs->dev[i].rdev.recovery_offset)) {
795 				rs->ti->error = "Drive designated for rebuild not specified";
796 				return -EINVAL;
797 			}
798 
799 			if (rs->dev[i].meta_dev) {
800 				rs->ti->error = "No data device supplied with metadata device";
801 				return -EINVAL;
802 			}
803 
804 			continue;
805 		}
806 
807 		r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
808 				  &rs->dev[i].data_dev);
809 		if (r) {
810 			rs->ti->error = "RAID device lookup failure";
811 			return r;
812 		}
813 
814 		if (rs->dev[i].meta_dev) {
815 			metadata_available = 1;
816 			rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
817 		}
818 		rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
819 		list_add_tail(&rs->dev[i].rdev.same_set, &rs->md.disks);
820 		if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
821 			rebuild++;
822 	}
823 
824 	if (metadata_available) {
825 		rs->md.external = 0;
826 		rs->md.persistent = 1;
827 		rs->md.major_version = 2;
828 	} else if (rebuild && !rs->md.recovery_cp) {
829 		/*
830 		 * Without metadata, we will not be able to tell if the array
831 		 * is in-sync or not - we must assume it is not.  Therefore,
832 		 * it is impossible to rebuild a drive.
833 		 *
834 		 * Even if there is metadata, the on-disk information may
835 		 * indicate that the array is not in-sync and it will then
836 		 * fail at that time.
837 		 *
838 		 * User could specify 'nosync' option if desperate.
839 		 */
840 		rs->ti->error = "Unable to rebuild drive while array is not in-sync";
841 		return -EINVAL;
842 	}
843 
844 	return 0;
845 }
846 
847 /*
848  * validate_region_size
849  * @rs
850  * @region_size:  region size in sectors.  If 0, pick a size (4MiB default).
851  *
852  * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
853  * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
854  *
855  * Returns: 0 on success, -EINVAL on failure.
856  */
857 static int validate_region_size(struct raid_set *rs, unsigned long region_size)
858 {
859 	unsigned long min_region_size = rs->ti->len / (1 << 21);
860 
861 	if (rs_is_raid0(rs))
862 		return 0;
863 
864 	if (!region_size) {
865 		/*
866 		 * Choose a reasonable default.	 All figures in sectors.
867 		 */
868 		if (min_region_size > (1 << 13)) {
869 			/* If not a power of 2, make it the next power of 2 */
870 			region_size = roundup_pow_of_two(min_region_size);
871 			DMINFO("Choosing default region size of %lu sectors",
872 			       region_size);
873 		} else {
874 			DMINFO("Choosing default region size of 4MiB");
875 			region_size = 1 << 13; /* sectors */
876 		}
877 	} else {
878 		/*
879 		 * Validate user-supplied value.
880 		 */
881 		if (region_size > rs->ti->len) {
882 			rs->ti->error = "Supplied region size is too large";
883 			return -EINVAL;
884 		}
885 
886 		if (region_size < min_region_size) {
887 			DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
888 			      region_size, min_region_size);
889 			rs->ti->error = "Supplied region size is too small";
890 			return -EINVAL;
891 		}
892 
893 		if (!is_power_of_2(region_size)) {
894 			rs->ti->error = "Region size is not a power of 2";
895 			return -EINVAL;
896 		}
897 
898 		if (region_size < rs->md.chunk_sectors) {
899 			rs->ti->error = "Region size is smaller than the chunk size";
900 			return -EINVAL;
901 		}
902 	}
903 
904 	/*
905 	 * Convert sectors to bytes.
906 	 */
907 	rs->md.bitmap_info.chunksize = to_bytes(region_size);
908 
909 	return 0;
910 }
911 
912 /*
913  * validate_raid_redundancy
914  * @rs
915  *
916  * Determine if there are enough devices in the array that haven't
917  * failed (or are being rebuilt) to form a usable array.
918  *
919  * Returns: 0 on success, -EINVAL on failure.
920  */
921 static int validate_raid_redundancy(struct raid_set *rs)
922 {
923 	unsigned int i, rebuild_cnt = 0;
924 	unsigned int rebuilds_per_group = 0, copies;
925 	unsigned int group_size, last_group_start;
926 
927 	for (i = 0; i < rs->md.raid_disks; i++)
928 		if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
929 		    !rs->dev[i].rdev.sb_page)
930 			rebuild_cnt++;
931 
932 	switch (rs->raid_type->level) {
933 	case 0:
934 		break;
935 	case 1:
936 		if (rebuild_cnt >= rs->md.raid_disks)
937 			goto too_many;
938 		break;
939 	case 4:
940 	case 5:
941 	case 6:
942 		if (rebuild_cnt > rs->raid_type->parity_devs)
943 			goto too_many;
944 		break;
945 	case 10:
946 		copies = raid10_md_layout_to_copies(rs->md.new_layout);
947 		if (rebuild_cnt < copies)
948 			break;
949 
950 		/*
951 		 * It is possible to have a higher rebuild count for RAID10,
952 		 * as long as the failed devices occur in different mirror
953 		 * groups (i.e. different stripes).
954 		 *
955 		 * When checking "near" format, make sure no adjacent devices
956 		 * have failed beyond what can be handled.  In addition to the
957 		 * simple case where the number of devices is a multiple of the
958 		 * number of copies, we must also handle cases where the number
959 		 * of devices is not a multiple of the number of copies.
960 		 * E.g.	   dev1 dev2 dev3 dev4 dev5
961 		 *	    A	 A    B	   B	C
962 		 *	    C	 D    D	   E	E
963 		 */
964 		if (__is_raid10_near(rs->md.new_layout)) {
965 			for (i = 0; i < rs->md.raid_disks; i++) {
966 				if (!(i % copies))
967 					rebuilds_per_group = 0;
968 				if ((!rs->dev[i].rdev.sb_page ||
969 				    !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
970 				    (++rebuilds_per_group >= copies))
971 					goto too_many;
972 			}
973 			break;
974 		}
975 
976 		/*
977 		 * When checking "far" and "offset" formats, we need to ensure
978 		 * that the device that holds its copy is not also dead or
979 		 * being rebuilt.  (Note that "far" and "offset" formats only
980 		 * support two copies right now.  These formats also only ever
981 		 * use the 'use_far_sets' variant.)
982 		 *
983 		 * This check is somewhat complicated by the need to account
984 		 * for arrays that are not a multiple of (far) copies.	This
985 		 * results in the need to treat the last (potentially larger)
986 		 * set differently.
987 		 */
988 		group_size = (rs->md.raid_disks / copies);
989 		last_group_start = (rs->md.raid_disks / group_size) - 1;
990 		last_group_start *= group_size;
991 		for (i = 0; i < rs->md.raid_disks; i++) {
992 			if (!(i % copies) && !(i > last_group_start))
993 				rebuilds_per_group = 0;
994 			if ((!rs->dev[i].rdev.sb_page ||
995 			     !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
996 			    (++rebuilds_per_group >= copies))
997 					goto too_many;
998 		}
999 		break;
1000 	default:
1001 		if (rebuild_cnt)
1002 			return -EINVAL;
1003 	}
1004 
1005 	return 0;
1006 
1007 too_many:
1008 	return -EINVAL;
1009 }
1010 
1011 /*
1012  * Possible arguments are...
1013  *	<chunk_size> [optional_args]
1014  *
1015  * Argument definitions
1016  *    <chunk_size>			The number of sectors per disk that
1017  *					will form the "stripe"
1018  *    [[no]sync]			Force or prevent recovery of the
1019  *					entire array
1020  *    [rebuild <idx>]			Rebuild the drive indicated by the index
1021  *    [daemon_sleep <ms>]		Time between bitmap daemon work to
1022  *					clear bits
1023  *    [min_recovery_rate <kB/sec/disk>]	Throttle RAID initialization
1024  *    [max_recovery_rate <kB/sec/disk>]	Throttle RAID initialization
1025  *    [write_mostly <idx>]		Indicate a write mostly drive via index
1026  *    [max_write_behind <sectors>]	See '-write-behind=' (man mdadm)
1027  *    [stripe_cache <sectors>]		Stripe cache size for higher RAIDs
1028  *    [region_size <sectors>]		Defines granularity of bitmap
1029  *
1030  * RAID10-only options:
1031  *    [raid10_copies <# copies>]	Number of copies.  (Default: 2)
1032  *    [raid10_format <near|far|offset>] Layout algorithm.  (Default: near)
1033  */
1034 static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
1035 			     unsigned int num_raid_params)
1036 {
1037 	int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
1038 	unsigned int raid10_copies = 2;
1039 	unsigned int i, write_mostly = 0;
1040 	unsigned int region_size = 0;
1041 	sector_t max_io_len;
1042 	const char *arg, *key;
1043 	struct raid_dev *rd;
1044 	struct raid_type *rt = rs->raid_type;
1045 
1046 	arg = dm_shift_arg(as);
1047 	num_raid_params--; /* Account for chunk_size argument */
1048 
1049 	if (kstrtoint(arg, 10, &value) < 0) {
1050 		rs->ti->error = "Bad numerical argument given for chunk_size";
1051 		return -EINVAL;
1052 	}
1053 
1054 	/*
1055 	 * First, parse the in-order required arguments
1056 	 * "chunk_size" is the only argument of this type.
1057 	 */
1058 	if (rt_is_raid1(rt)) {
1059 		if (value)
1060 			DMERR("Ignoring chunk size parameter for RAID 1");
1061 		value = 0;
1062 	} else if (!is_power_of_2(value)) {
1063 		rs->ti->error = "Chunk size must be a power of 2";
1064 		return -EINVAL;
1065 	} else if (value < 8) {
1066 		rs->ti->error = "Chunk size value is too small";
1067 		return -EINVAL;
1068 	}
1069 
1070 	rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
1071 
1072 	/*
1073 	 * We set each individual device as In_sync with a completed
1074 	 * 'recovery_offset'.  If there has been a device failure or
1075 	 * replacement then one of the following cases applies:
1076 	 *
1077 	 *   1) User specifies 'rebuild'.
1078 	 *	- Device is reset when param is read.
1079 	 *   2) A new device is supplied.
1080 	 *	- No matching superblock found, resets device.
1081 	 *   3) Device failure was transient and returns on reload.
1082 	 *	- Failure noticed, resets device for bitmap replay.
1083 	 *   4) Device hadn't completed recovery after previous failure.
1084 	 *	- Superblock is read and overrides recovery_offset.
1085 	 *
1086 	 * What is found in the superblocks of the devices is always
1087 	 * authoritative, unless 'rebuild' or '[no]sync' was specified.
1088 	 */
1089 	for (i = 0; i < rs->raid_disks; i++) {
1090 		set_bit(In_sync, &rs->dev[i].rdev.flags);
1091 		rs->dev[i].rdev.recovery_offset = MaxSector;
1092 	}
1093 
1094 	/*
1095 	 * Second, parse the unordered optional arguments
1096 	 */
1097 	for (i = 0; i < num_raid_params; i++) {
1098 		key = dm_shift_arg(as);
1099 		if (!key) {
1100 			rs->ti->error = "Not enough raid parameters given";
1101 			return -EINVAL;
1102 		}
1103 
1104 		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
1105 			if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1106 				rs->ti->error = "Only one 'nosync' argument allowed";
1107 				return -EINVAL;
1108 			}
1109 			continue;
1110 		}
1111 		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
1112 			if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
1113 				rs->ti->error = "Only one 'sync' argument allowed";
1114 				return -EINVAL;
1115 			}
1116 			continue;
1117 		}
1118 		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
1119 			if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1120 				rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1121 				return -EINVAL;
1122 			}
1123 			continue;
1124 		}
1125 
1126 		arg = dm_shift_arg(as);
1127 		i++; /* Account for the argument pairs */
1128 		if (!arg) {
1129 			rs->ti->error = "Wrong number of raid parameters given";
1130 			return -EINVAL;
1131 		}
1132 
1133 		/*
1134 		 * Parameters that take a string value are checked here.
1135 		 */
1136 
1137 		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
1138 			if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
1139 				rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1140 				return -EINVAL;
1141 			}
1142 			if (!rt_is_raid10(rt)) {
1143 				rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1144 				return -EINVAL;
1145 			}
1146 			raid10_format = raid10_name_to_format(arg);
1147 			if (raid10_format < 0) {
1148 				rs->ti->error = "Invalid 'raid10_format' value given";
1149 				return raid10_format;
1150 			}
1151 			continue;
1152 		}
1153 
1154 		if (kstrtoint(arg, 10, &value) < 0) {
1155 			rs->ti->error = "Bad numerical argument given in raid params";
1156 			return -EINVAL;
1157 		}
1158 
1159 		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
1160 			/*
1161 			 * "rebuild" is being passed in by userspace to provide
1162 			 * indexes of replaced devices and to set up additional
1163 			 * devices on raid level takeover.
1164 			 */
1165 			if (!__within_range(value, 0, rs->raid_disks - 1)) {
1166 				rs->ti->error = "Invalid rebuild index given";
1167 				return -EINVAL;
1168 			}
1169 
1170 			if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1171 				rs->ti->error = "rebuild for this index already given";
1172 				return -EINVAL;
1173 			}
1174 
1175 			rd = rs->dev + value;
1176 			clear_bit(In_sync, &rd->rdev.flags);
1177 			clear_bit(Faulty, &rd->rdev.flags);
1178 			rd->rdev.recovery_offset = 0;
1179 			set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
1180 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
1181 			if (!rt_is_raid1(rt)) {
1182 				rs->ti->error = "write_mostly option is only valid for RAID1";
1183 				return -EINVAL;
1184 			}
1185 
1186 			if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
1187 				rs->ti->error = "Invalid write_mostly index given";
1188 				return -EINVAL;
1189 			}
1190 
1191 			write_mostly++;
1192 			set_bit(WriteMostly, &rs->dev[value].rdev.flags);
1193 			set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
1194 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
1195 			if (!rt_is_raid1(rt)) {
1196 				rs->ti->error = "max_write_behind option is only valid for RAID1";
1197 				return -EINVAL;
1198 			}
1199 
1200 			if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
1201 				rs->ti->error = "Only one max_write_behind argument pair allowed";
1202 				return -EINVAL;
1203 			}
1204 
1205 			/*
1206 			 * In device-mapper, we specify things in sectors, but
1207 			 * MD records this value in kB
1208 			 */
1209 			value /= 2;
1210 			if (value > COUNTER_MAX) {
1211 				rs->ti->error = "Max write-behind limit out of range";
1212 				return -EINVAL;
1213 			}
1214 
1215 			rs->md.bitmap_info.max_write_behind = value;
1216 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
1217 			if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
1218 				rs->ti->error = "Only one daemon_sleep argument pair allowed";
1219 				return -EINVAL;
1220 			}
1221 			if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
1222 				rs->ti->error = "daemon sleep period out of range";
1223 				return -EINVAL;
1224 			}
1225 			rs->md.bitmap_info.daemon_sleep = value;
1226 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
1227 			/* Userspace passes new data_offset after having extended the the data image LV */
1228 			if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
1229 				rs->ti->error = "Only one data_offset argument pair allowed";
1230 				return -EINVAL;
1231 			}
1232 			/* Ensure sensible data offset */
1233 			if (value < 0 ||
1234 			    (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
1235 				rs->ti->error = "Bogus data_offset value";
1236 				return -EINVAL;
1237 			}
1238 			rs->data_offset = value;
1239 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
1240 			/* Define the +/-# of disks to add to/remove from the given raid set */
1241 			if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1242 				rs->ti->error = "Only one delta_disks argument pair allowed";
1243 				return -EINVAL;
1244 			}
1245 			/* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1246 			if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
1247 				rs->ti->error = "Too many delta_disk requested";
1248 				return -EINVAL;
1249 			}
1250 
1251 			rs->delta_disks = value;
1252 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
1253 			if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
1254 				rs->ti->error = "Only one stripe_cache argument pair allowed";
1255 				return -EINVAL;
1256 			}
1257 
1258 			if (!rt_is_raid456(rt)) {
1259 				rs->ti->error = "Inappropriate argument: stripe_cache";
1260 				return -EINVAL;
1261 			}
1262 
1263 			rs->stripe_cache_entries = value;
1264 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
1265 			if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1266 				rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1267 				return -EINVAL;
1268 			}
1269 			if (value > INT_MAX) {
1270 				rs->ti->error = "min_recovery_rate out of range";
1271 				return -EINVAL;
1272 			}
1273 			rs->md.sync_speed_min = (int)value;
1274 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
1275 			if (test_and_set_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags)) {
1276 				rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1277 				return -EINVAL;
1278 			}
1279 			if (value > INT_MAX) {
1280 				rs->ti->error = "max_recovery_rate out of range";
1281 				return -EINVAL;
1282 			}
1283 			rs->md.sync_speed_max = (int)value;
1284 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
1285 			if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
1286 				rs->ti->error = "Only one region_size argument pair allowed";
1287 				return -EINVAL;
1288 			}
1289 
1290 			region_size = value;
1291 			rs->requested_bitmap_chunk_sectors = value;
1292 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
1293 			if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
1294 				rs->ti->error = "Only one raid10_copies argument pair allowed";
1295 				return -EINVAL;
1296 			}
1297 
1298 			if (!__within_range(value, 2, rs->md.raid_disks)) {
1299 				rs->ti->error = "Bad value for 'raid10_copies'";
1300 				return -EINVAL;
1301 			}
1302 
1303 			raid10_copies = value;
1304 		} else {
1305 			DMERR("Unable to parse RAID parameter: %s", key);
1306 			rs->ti->error = "Unable to parse RAID parameter";
1307 			return -EINVAL;
1308 		}
1309 	}
1310 
1311 	if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1312 	    test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1313 		rs->ti->error = "sync and nosync are mutually exclusive";
1314 		return -EINVAL;
1315 	}
1316 
1317 	if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) &&
1318 	    (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ||
1319 	     test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))) {
1320 		rs->ti->error = "sync/nosync and rebuild are mutually exclusive";
1321 		return -EINVAL;
1322 	}
1323 
1324 	if (write_mostly >= rs->md.raid_disks) {
1325 		rs->ti->error = "Can't set all raid1 devices to write_mostly";
1326 		return -EINVAL;
1327 	}
1328 
1329 	if (validate_region_size(rs, region_size))
1330 		return -EINVAL;
1331 
1332 	if (rs->md.chunk_sectors)
1333 		max_io_len = rs->md.chunk_sectors;
1334 	else
1335 		max_io_len = region_size;
1336 
1337 	if (dm_set_target_max_io_len(rs->ti, max_io_len))
1338 		return -EINVAL;
1339 
1340 	if (rt_is_raid10(rt)) {
1341 		if (raid10_copies > rs->md.raid_disks) {
1342 			rs->ti->error = "Not enough devices to satisfy specification";
1343 			return -EINVAL;
1344 		}
1345 
1346 		rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
1347 		if (rs->md.new_layout < 0) {
1348 			rs->ti->error = "Error getting raid10 format";
1349 			return rs->md.new_layout;
1350 		}
1351 
1352 		rt = get_raid_type_by_ll(10, rs->md.new_layout);
1353 		if (!rt) {
1354 			rs->ti->error = "Failed to recognize new raid10 layout";
1355 			return -EINVAL;
1356 		}
1357 
1358 		if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1359 		     rt->algorithm == ALGORITHM_RAID10_NEAR) &&
1360 		    test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1361 			rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1362 			return -EINVAL;
1363 		}
1364 	}
1365 
1366 	rs->raid10_copies = raid10_copies;
1367 
1368 	/* Assume there are no metadata devices until the drives are parsed */
1369 	rs->md.persistent = 0;
1370 	rs->md.external = 1;
1371 
1372 	/* Check, if any invalid ctr arguments have been passed in for the raid level */
1373 	return rs_check_for_valid_flags(rs);
1374 }
1375 
1376 /* Set raid4/5/6 cache size */
1377 static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1378 {
1379 	int r;
1380 	struct r5conf *conf;
1381 	struct mddev *mddev = &rs->md;
1382 	uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1383 	uint32_t nr_stripes = rs->stripe_cache_entries;
1384 
1385 	if (!rt_is_raid456(rs->raid_type)) {
1386 		rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1387 		return -EINVAL;
1388 	}
1389 
1390 	if (nr_stripes < min_stripes) {
1391 		DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1392 		       nr_stripes, min_stripes);
1393 		nr_stripes = min_stripes;
1394 	}
1395 
1396 	conf = mddev->private;
1397 	if (!conf) {
1398 		rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1399 		return -EINVAL;
1400 	}
1401 
1402 	/* Try setting number of stripes in raid456 stripe cache */
1403 	if (conf->min_nr_stripes != nr_stripes) {
1404 		r = raid5_set_cache_size(mddev, nr_stripes);
1405 		if (r) {
1406 			rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1407 			return r;
1408 		}
1409 
1410 		DMINFO("%u stripe cache entries", nr_stripes);
1411 	}
1412 
1413 	return 0;
1414 }
1415 
1416 /* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1417 static unsigned int mddev_data_stripes(struct raid_set *rs)
1418 {
1419 	return rs->md.raid_disks - rs->raid_type->parity_devs;
1420 }
1421 
1422 /* Return # of data stripes of @rs (i.e. as of ctr) */
1423 static unsigned int rs_data_stripes(struct raid_set *rs)
1424 {
1425 	return rs->raid_disks - rs->raid_type->parity_devs;
1426 }
1427 
1428 /* Calculate the sectors per device and per array used for @rs */
1429 static int rs_set_dev_and_array_sectors(struct raid_set *rs, bool use_mddev)
1430 {
1431 	int delta_disks;
1432 	unsigned int data_stripes;
1433 	struct mddev *mddev = &rs->md;
1434 	struct md_rdev *rdev;
1435 	sector_t array_sectors = rs->ti->len, dev_sectors = rs->ti->len;
1436 
1437 	if (use_mddev) {
1438 		delta_disks = mddev->delta_disks;
1439 		data_stripes = mddev_data_stripes(rs);
1440 	} else {
1441 		delta_disks = rs->delta_disks;
1442 		data_stripes = rs_data_stripes(rs);
1443 	}
1444 
1445 	/* Special raid1 case w/o delta_disks support (yet) */
1446 	if (rt_is_raid1(rs->raid_type))
1447 		;
1448 	else if (rt_is_raid10(rs->raid_type)) {
1449 		if (rs->raid10_copies < 2 ||
1450 		    delta_disks < 0) {
1451 			rs->ti->error = "Bogus raid10 data copies or delta disks";
1452 			return -EINVAL;
1453 		}
1454 
1455 		dev_sectors *= rs->raid10_copies;
1456 		if (sector_div(dev_sectors, data_stripes))
1457 			goto bad;
1458 
1459 		array_sectors = (data_stripes + delta_disks) * dev_sectors;
1460 		if (sector_div(array_sectors, rs->raid10_copies))
1461 			goto bad;
1462 
1463 	} else if (sector_div(dev_sectors, data_stripes))
1464 		goto bad;
1465 
1466 	else
1467 		/* Striped layouts */
1468 		array_sectors = (data_stripes + delta_disks) * dev_sectors;
1469 
1470 	rdev_for_each(rdev, mddev)
1471 		rdev->sectors = dev_sectors;
1472 
1473 	mddev->array_sectors = array_sectors;
1474 	mddev->dev_sectors = dev_sectors;
1475 
1476 	return 0;
1477 bad:
1478 	rs->ti->error = "Target length not divisible by number of data devices";
1479 	return -EINVAL;
1480 }
1481 
1482 /* Setup recovery on @rs */
1483 static void __rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1484 {
1485 	/* raid0 does not recover */
1486 	if (rs_is_raid0(rs))
1487 		rs->md.recovery_cp = MaxSector;
1488 	/*
1489 	 * A raid6 set has to be recovered either
1490 	 * completely or for the grown part to
1491 	 * ensure proper parity and Q-Syndrome
1492 	 */
1493 	else if (rs_is_raid6(rs))
1494 		rs->md.recovery_cp = dev_sectors;
1495 	/*
1496 	 * Other raid set types may skip recovery
1497 	 * depending on the 'nosync' flag.
1498 	 */
1499 	else
1500 		rs->md.recovery_cp = test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)
1501 				     ? MaxSector : dev_sectors;
1502 }
1503 
1504 /* Setup recovery on @rs based on raid type, device size and 'nosync' flag */
1505 static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1506 {
1507 	if (!dev_sectors)
1508 		/* New raid set or 'sync' flag provided */
1509 		__rs_setup_recovery(rs, 0);
1510 	else if (dev_sectors == MaxSector)
1511 		/* Prevent recovery */
1512 		__rs_setup_recovery(rs, MaxSector);
1513 	else if (rs->dev[0].rdev.sectors < dev_sectors)
1514 		/* Grown raid set */
1515 		__rs_setup_recovery(rs, rs->dev[0].rdev.sectors);
1516 	else
1517 		__rs_setup_recovery(rs, MaxSector);
1518 }
1519 
1520 static void do_table_event(struct work_struct *ws)
1521 {
1522 	struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1523 
1524 	smp_rmb(); /* Make sure we access most actual mddev properties */
1525 	if (!rs_is_reshaping(rs))
1526 		rs_set_capacity(rs);
1527 	dm_table_event(rs->ti->table);
1528 }
1529 
1530 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
1531 {
1532 	struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
1533 
1534 	return mddev_congested(&rs->md, bits);
1535 }
1536 
1537 /*
1538  * Make sure a valid takover (level switch) is being requested on @rs
1539  *
1540  * Conversions of raid sets from one MD personality to another
1541  * have to conform to restrictions which are enforced here.
1542  */
1543 static int rs_check_takeover(struct raid_set *rs)
1544 {
1545 	struct mddev *mddev = &rs->md;
1546 	unsigned int near_copies;
1547 
1548 	if (rs->md.degraded) {
1549 		rs->ti->error = "Can't takeover degraded raid set";
1550 		return -EPERM;
1551 	}
1552 
1553 	if (rs_is_reshaping(rs)) {
1554 		rs->ti->error = "Can't takeover reshaping raid set";
1555 		return -EPERM;
1556 	}
1557 
1558 	switch (mddev->level) {
1559 	case 0:
1560 		/* raid0 -> raid1/5 with one disk */
1561 		if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1562 		    mddev->raid_disks == 1)
1563 			return 0;
1564 
1565 		/* raid0 -> raid10 */
1566 		if (mddev->new_level == 10 &&
1567 		    !(rs->raid_disks % mddev->raid_disks))
1568 			return 0;
1569 
1570 		/* raid0 with multiple disks -> raid4/5/6 */
1571 		if (__within_range(mddev->new_level, 4, 6) &&
1572 		    mddev->new_layout == ALGORITHM_PARITY_N &&
1573 		    mddev->raid_disks > 1)
1574 			return 0;
1575 
1576 		break;
1577 
1578 	case 10:
1579 		/* Can't takeover raid10_offset! */
1580 		if (__is_raid10_offset(mddev->layout))
1581 			break;
1582 
1583 		near_copies = __raid10_near_copies(mddev->layout);
1584 
1585 		/* raid10* -> raid0 */
1586 		if (mddev->new_level == 0) {
1587 			/* Can takeover raid10_near with raid disks divisable by data copies! */
1588 			if (near_copies > 1 &&
1589 			    !(mddev->raid_disks % near_copies)) {
1590 				mddev->raid_disks /= near_copies;
1591 				mddev->delta_disks = mddev->raid_disks;
1592 				return 0;
1593 			}
1594 
1595 			/* Can takeover raid10_far */
1596 			if (near_copies == 1 &&
1597 			    __raid10_far_copies(mddev->layout) > 1)
1598 				return 0;
1599 
1600 			break;
1601 		}
1602 
1603 		/* raid10_{near,far} -> raid1 */
1604 		if (mddev->new_level == 1 &&
1605 		    max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
1606 			return 0;
1607 
1608 		/* raid10_{near,far} with 2 disks -> raid4/5 */
1609 		if (__within_range(mddev->new_level, 4, 5) &&
1610 		    mddev->raid_disks == 2)
1611 			return 0;
1612 		break;
1613 
1614 	case 1:
1615 		/* raid1 with 2 disks -> raid4/5 */
1616 		if (__within_range(mddev->new_level, 4, 5) &&
1617 		    mddev->raid_disks == 2) {
1618 			mddev->degraded = 1;
1619 			return 0;
1620 		}
1621 
1622 		/* raid1 -> raid0 */
1623 		if (mddev->new_level == 0 &&
1624 		    mddev->raid_disks == 1)
1625 			return 0;
1626 
1627 		/* raid1 -> raid10 */
1628 		if (mddev->new_level == 10)
1629 			return 0;
1630 		break;
1631 
1632 	case 4:
1633 		/* raid4 -> raid0 */
1634 		if (mddev->new_level == 0)
1635 			return 0;
1636 
1637 		/* raid4 -> raid1/5 with 2 disks */
1638 		if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1639 		    mddev->raid_disks == 2)
1640 			return 0;
1641 
1642 		/* raid4 -> raid5/6 with parity N */
1643 		if (__within_range(mddev->new_level, 5, 6) &&
1644 		    mddev->layout == ALGORITHM_PARITY_N)
1645 			return 0;
1646 		break;
1647 
1648 	case 5:
1649 		/* raid5 with parity N -> raid0 */
1650 		if (mddev->new_level == 0 &&
1651 		    mddev->layout == ALGORITHM_PARITY_N)
1652 			return 0;
1653 
1654 		/* raid5 with parity N -> raid4 */
1655 		if (mddev->new_level == 4 &&
1656 		    mddev->layout == ALGORITHM_PARITY_N)
1657 			return 0;
1658 
1659 		/* raid5 with 2 disks -> raid1/4/10 */
1660 		if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1661 		    mddev->raid_disks == 2)
1662 			return 0;
1663 
1664 		/* raid5_* ->  raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
1665 		if (mddev->new_level == 6 &&
1666 		    ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1667 		      __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
1668 			return 0;
1669 		break;
1670 
1671 	case 6:
1672 		/* raid6 with parity N -> raid0 */
1673 		if (mddev->new_level == 0 &&
1674 		    mddev->layout == ALGORITHM_PARITY_N)
1675 			return 0;
1676 
1677 		/* raid6 with parity N -> raid4 */
1678 		if (mddev->new_level == 4 &&
1679 		    mddev->layout == ALGORITHM_PARITY_N)
1680 			return 0;
1681 
1682 		/* raid6_*_n with Q-Syndrome N -> raid5_* */
1683 		if (mddev->new_level == 5 &&
1684 		    ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1685 		     __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
1686 			return 0;
1687 
1688 	default:
1689 		break;
1690 	}
1691 
1692 	rs->ti->error = "takeover not possible";
1693 	return -EINVAL;
1694 }
1695 
1696 /* True if @rs requested to be taken over */
1697 static bool rs_takeover_requested(struct raid_set *rs)
1698 {
1699 	return rs->md.new_level != rs->md.level;
1700 }
1701 
1702 /* True if @rs is requested to reshape by ctr */
1703 static bool rs_reshape_requested(struct raid_set *rs)
1704 {
1705 	bool change;
1706 	struct mddev *mddev = &rs->md;
1707 
1708 	if (rs_takeover_requested(rs))
1709 		return false;
1710 
1711 	if (!mddev->level)
1712 		return false;
1713 
1714 	change = mddev->new_layout != mddev->layout ||
1715 		 mddev->new_chunk_sectors != mddev->chunk_sectors ||
1716 		 rs->delta_disks;
1717 
1718 	/* Historical case to support raid1 reshape without delta disks */
1719 	if (mddev->level == 1) {
1720 		if (rs->delta_disks)
1721 			return !!rs->delta_disks;
1722 
1723 		return !change &&
1724 		       mddev->raid_disks != rs->raid_disks;
1725 	}
1726 
1727 	if (mddev->level == 10)
1728 		return change &&
1729 		       !__is_raid10_far(mddev->new_layout) &&
1730 		       rs->delta_disks >= 0;
1731 
1732 	return change;
1733 }
1734 
1735 /*  Features */
1736 #define	FEATURE_FLAG_SUPPORTS_V190	0x1 /* Supports extended superblock */
1737 
1738 /* State flags for sb->flags */
1739 #define	SB_FLAG_RESHAPE_ACTIVE		0x1
1740 #define	SB_FLAG_RESHAPE_BACKWARDS	0x2
1741 
1742 /*
1743  * This structure is never routinely used by userspace, unlike md superblocks.
1744  * Devices with this superblock should only ever be accessed via device-mapper.
1745  */
1746 #define DM_RAID_MAGIC 0x64526D44
1747 struct dm_raid_superblock {
1748 	__le32 magic;		/* "DmRd" */
1749 	__le32 compat_features;	/* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1750 
1751 	__le32 num_devices;	/* Number of devices in this raid set. (Max 64) */
1752 	__le32 array_position;	/* The position of this drive in the raid set */
1753 
1754 	__le64 events;		/* Incremented by md when superblock updated */
1755 	__le64 failed_devices;	/* Pre 1.9.0 part of bit field of devices to */
1756 				/* indicate failures (see extension below) */
1757 
1758 	/*
1759 	 * This offset tracks the progress of the repair or replacement of
1760 	 * an individual drive.
1761 	 */
1762 	__le64 disk_recovery_offset;
1763 
1764 	/*
1765 	 * This offset tracks the progress of the initial raid set
1766 	 * synchronisation/parity calculation.
1767 	 */
1768 	__le64 array_resync_offset;
1769 
1770 	/*
1771 	 * raid characteristics
1772 	 */
1773 	__le32 level;
1774 	__le32 layout;
1775 	__le32 stripe_sectors;
1776 
1777 	/********************************************************************
1778 	 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1779 	 *
1780 	 * FEATURE_FLAG_SUPPORTS_V190 in the features member indicates that those exist
1781 	 */
1782 
1783 	__le32 flags; /* Flags defining array states for reshaping */
1784 
1785 	/*
1786 	 * This offset tracks the progress of a raid
1787 	 * set reshape in order to be able to restart it
1788 	 */
1789 	__le64 reshape_position;
1790 
1791 	/*
1792 	 * These define the properties of the array in case of an interrupted reshape
1793 	 */
1794 	__le32 new_level;
1795 	__le32 new_layout;
1796 	__le32 new_stripe_sectors;
1797 	__le32 delta_disks;
1798 
1799 	__le64 array_sectors; /* Array size in sectors */
1800 
1801 	/*
1802 	 * Sector offsets to data on devices (reshaping).
1803 	 * Needed to support out of place reshaping, thus
1804 	 * not writing over any stripes whilst converting
1805 	 * them from old to new layout
1806 	 */
1807 	__le64 data_offset;
1808 	__le64 new_data_offset;
1809 
1810 	__le64 sectors; /* Used device size in sectors */
1811 
1812 	/*
1813 	 * Additonal Bit field of devices indicating failures to support
1814 	 * up to 256 devices with the 1.9.0 on-disk metadata format
1815 	 */
1816 	__le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
1817 
1818 	__le32 incompat_features;	/* Used to indicate any incompatible features */
1819 
1820 	/* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
1821 } __packed;
1822 
1823 /*
1824  * Check for reshape constraints on raid set @rs:
1825  *
1826  * - reshape function non-existent
1827  * - degraded set
1828  * - ongoing recovery
1829  * - ongoing reshape
1830  *
1831  * Returns 0 if none or -EPERM if given constraint
1832  * and error message reference in @errmsg
1833  */
1834 static int rs_check_reshape(struct raid_set *rs)
1835 {
1836 	struct mddev *mddev = &rs->md;
1837 
1838 	if (!mddev->pers || !mddev->pers->check_reshape)
1839 		rs->ti->error = "Reshape not supported";
1840 	else if (mddev->degraded)
1841 		rs->ti->error = "Can't reshape degraded raid set";
1842 	else if (rs_is_recovering(rs))
1843 		rs->ti->error = "Convert request on recovering raid set prohibited";
1844 	else if (rs_is_reshaping(rs))
1845 		rs->ti->error = "raid set already reshaping!";
1846 	else if (!(rs_is_raid1(rs) || rs_is_raid10(rs) || rs_is_raid456(rs)))
1847 		rs->ti->error = "Reshaping only supported for raid1/4/5/6/10";
1848 	else
1849 		return 0;
1850 
1851 	return -EPERM;
1852 }
1853 
1854 static int read_disk_sb(struct md_rdev *rdev, int size)
1855 {
1856 	BUG_ON(!rdev->sb_page);
1857 
1858 	if (rdev->sb_loaded)
1859 		return 0;
1860 
1861 	if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) {
1862 		DMERR("Failed to read superblock of device at position %d",
1863 		      rdev->raid_disk);
1864 		md_error(rdev->mddev, rdev);
1865 		return -EINVAL;
1866 	}
1867 
1868 	rdev->sb_loaded = 1;
1869 
1870 	return 0;
1871 }
1872 
1873 static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
1874 {
1875 	failed_devices[0] = le64_to_cpu(sb->failed_devices);
1876 	memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
1877 
1878 	if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
1879 		int i = ARRAY_SIZE(sb->extended_failed_devices);
1880 
1881 		while (i--)
1882 			failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
1883 	}
1884 }
1885 
1886 static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
1887 {
1888 	int i = ARRAY_SIZE(sb->extended_failed_devices);
1889 
1890 	sb->failed_devices = cpu_to_le64(failed_devices[0]);
1891 	while (i--)
1892 		sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
1893 }
1894 
1895 /*
1896  * Synchronize the superblock members with the raid set properties
1897  *
1898  * All superblock data is little endian.
1899  */
1900 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
1901 {
1902 	bool update_failed_devices = false;
1903 	unsigned int i;
1904 	uint64_t failed_devices[DISKS_ARRAY_ELEMS];
1905 	struct dm_raid_superblock *sb;
1906 	struct raid_set *rs = container_of(mddev, struct raid_set, md);
1907 
1908 	/* No metadata device, no superblock */
1909 	if (!rdev->meta_bdev)
1910 		return;
1911 
1912 	BUG_ON(!rdev->sb_page);
1913 
1914 	sb = page_address(rdev->sb_page);
1915 
1916 	sb_retrieve_failed_devices(sb, failed_devices);
1917 
1918 	for (i = 0; i < rs->raid_disks; i++)
1919 		if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
1920 			update_failed_devices = true;
1921 			set_bit(i, (void *) failed_devices);
1922 		}
1923 
1924 	if (update_failed_devices)
1925 		sb_update_failed_devices(sb, failed_devices);
1926 
1927 	sb->magic = cpu_to_le32(DM_RAID_MAGIC);
1928 	sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
1929 
1930 	sb->num_devices = cpu_to_le32(mddev->raid_disks);
1931 	sb->array_position = cpu_to_le32(rdev->raid_disk);
1932 
1933 	sb->events = cpu_to_le64(mddev->events);
1934 
1935 	sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
1936 	sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
1937 
1938 	sb->level = cpu_to_le32(mddev->level);
1939 	sb->layout = cpu_to_le32(mddev->layout);
1940 	sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
1941 
1942 	sb->new_level = cpu_to_le32(mddev->new_level);
1943 	sb->new_layout = cpu_to_le32(mddev->new_layout);
1944 	sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
1945 
1946 	sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1947 
1948 	smp_rmb(); /* Make sure we access most recent reshape position */
1949 	sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1950 	if (le64_to_cpu(sb->reshape_position) != MaxSector) {
1951 		/* Flag ongoing reshape */
1952 		sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
1953 
1954 		if (mddev->delta_disks < 0 || mddev->reshape_backwards)
1955 			sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
1956 	} else {
1957 		/* Clear reshape flags */
1958 		sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
1959 	}
1960 
1961 	sb->array_sectors = cpu_to_le64(mddev->array_sectors);
1962 	sb->data_offset = cpu_to_le64(rdev->data_offset);
1963 	sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
1964 	sb->sectors = cpu_to_le64(rdev->sectors);
1965 	sb->incompat_features = cpu_to_le32(0);
1966 
1967 	/* Zero out the rest of the payload after the size of the superblock */
1968 	memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
1969 }
1970 
1971 /*
1972  * super_load
1973  *
1974  * This function creates a superblock if one is not found on the device
1975  * and will decide which superblock to use if there's a choice.
1976  *
1977  * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
1978  */
1979 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
1980 {
1981 	int r;
1982 	struct dm_raid_superblock *sb;
1983 	struct dm_raid_superblock *refsb;
1984 	uint64_t events_sb, events_refsb;
1985 
1986 	rdev->sb_start = 0;
1987 	rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
1988 	if (rdev->sb_size < sizeof(*sb) || rdev->sb_size > PAGE_SIZE) {
1989 		DMERR("superblock size of a logical block is no longer valid");
1990 		return -EINVAL;
1991 	}
1992 
1993 	r = read_disk_sb(rdev, rdev->sb_size);
1994 	if (r)
1995 		return r;
1996 
1997 	sb = page_address(rdev->sb_page);
1998 
1999 	/*
2000 	 * Two cases that we want to write new superblocks and rebuild:
2001 	 * 1) New device (no matching magic number)
2002 	 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2003 	 */
2004 	if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
2005 	    (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
2006 		super_sync(rdev->mddev, rdev);
2007 
2008 		set_bit(FirstUse, &rdev->flags);
2009 		sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2010 
2011 		/* Force writing of superblocks to disk */
2012 		set_bit(MD_SB_CHANGE_DEVS, &rdev->mddev->sb_flags);
2013 
2014 		/* Any superblock is better than none, choose that if given */
2015 		return refdev ? 0 : 1;
2016 	}
2017 
2018 	if (!refdev)
2019 		return 1;
2020 
2021 	events_sb = le64_to_cpu(sb->events);
2022 
2023 	refsb = page_address(refdev->sb_page);
2024 	events_refsb = le64_to_cpu(refsb->events);
2025 
2026 	return (events_sb > events_refsb) ? 1 : 0;
2027 }
2028 
2029 static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
2030 {
2031 	int role;
2032 	unsigned int d;
2033 	struct mddev *mddev = &rs->md;
2034 	uint64_t events_sb;
2035 	uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2036 	struct dm_raid_superblock *sb;
2037 	uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
2038 	struct md_rdev *r;
2039 	struct dm_raid_superblock *sb2;
2040 
2041 	sb = page_address(rdev->sb_page);
2042 	events_sb = le64_to_cpu(sb->events);
2043 
2044 	/*
2045 	 * Initialise to 1 if this is a new superblock.
2046 	 */
2047 	mddev->events = events_sb ? : 1;
2048 
2049 	mddev->reshape_position = MaxSector;
2050 
2051 	mddev->raid_disks = le32_to_cpu(sb->num_devices);
2052 	mddev->level = le32_to_cpu(sb->level);
2053 	mddev->layout = le32_to_cpu(sb->layout);
2054 	mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
2055 
2056 	/*
2057 	 * Reshaping is supported, e.g. reshape_position is valid
2058 	 * in superblock and superblock content is authoritative.
2059 	 */
2060 	if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2061 		/* Superblock is authoritative wrt given raid set layout! */
2062 		mddev->new_level = le32_to_cpu(sb->new_level);
2063 		mddev->new_layout = le32_to_cpu(sb->new_layout);
2064 		mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
2065 		mddev->delta_disks = le32_to_cpu(sb->delta_disks);
2066 		mddev->array_sectors = le64_to_cpu(sb->array_sectors);
2067 
2068 		/* raid was reshaping and got interrupted */
2069 		if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
2070 			if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
2071 				DMERR("Reshape requested but raid set is still reshaping");
2072 				return -EINVAL;
2073 			}
2074 
2075 			if (mddev->delta_disks < 0 ||
2076 			    (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
2077 				mddev->reshape_backwards = 1;
2078 			else
2079 				mddev->reshape_backwards = 0;
2080 
2081 			mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2082 			rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2083 		}
2084 
2085 	} else {
2086 		/*
2087 		 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2088 		 */
2089 		struct raid_type *rt_cur = get_raid_type_by_ll(mddev->level, mddev->layout);
2090 		struct raid_type *rt_new = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
2091 
2092 		if (rs_takeover_requested(rs)) {
2093 			if (rt_cur && rt_new)
2094 				DMERR("Takeover raid sets from %s to %s not yet supported by metadata. (raid level change)",
2095 				      rt_cur->name, rt_new->name);
2096 			else
2097 				DMERR("Takeover raid sets not yet supported by metadata. (raid level change)");
2098 			return -EINVAL;
2099 		} else if (rs_reshape_requested(rs)) {
2100 			DMERR("Reshaping raid sets not yet supported by metadata. (raid layout change keeping level)");
2101 			if (mddev->layout != mddev->new_layout) {
2102 				if (rt_cur && rt_new)
2103 					DMERR("	 current layout %s vs new layout %s",
2104 					      rt_cur->name, rt_new->name);
2105 				else
2106 					DMERR("	 current layout 0x%X vs new layout 0x%X",
2107 					      le32_to_cpu(sb->layout), mddev->new_layout);
2108 			}
2109 			if (mddev->chunk_sectors != mddev->new_chunk_sectors)
2110 				DMERR("	 current stripe sectors %u vs new stripe sectors %u",
2111 				      mddev->chunk_sectors, mddev->new_chunk_sectors);
2112 			if (rs->delta_disks)
2113 				DMERR("	 current %u disks vs new %u disks",
2114 				      mddev->raid_disks, mddev->raid_disks + rs->delta_disks);
2115 			if (rs_is_raid10(rs)) {
2116 				DMERR("	 Old layout: %s w/ %u copies",
2117 				      raid10_md_layout_to_format(mddev->layout),
2118 				      raid10_md_layout_to_copies(mddev->layout));
2119 				DMERR("	 New layout: %s w/ %u copies",
2120 				      raid10_md_layout_to_format(mddev->new_layout),
2121 				      raid10_md_layout_to_copies(mddev->new_layout));
2122 			}
2123 			return -EINVAL;
2124 		}
2125 
2126 		DMINFO("Discovered old metadata format; upgrading to extended metadata format");
2127 	}
2128 
2129 	if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
2130 		mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2131 
2132 	/*
2133 	 * During load, we set FirstUse if a new superblock was written.
2134 	 * There are two reasons we might not have a superblock:
2135 	 * 1) The raid set is brand new - in which case, all of the
2136 	 *    devices must have their In_sync bit set.	Also,
2137 	 *    recovery_cp must be 0, unless forced.
2138 	 * 2) This is a new device being added to an old raid set
2139 	 *    and the new device needs to be rebuilt - in which
2140 	 *    case the In_sync bit will /not/ be set and
2141 	 *    recovery_cp must be MaxSector.
2142 	 * 3) This is/are a new device(s) being added to an old
2143 	 *    raid set during takeover to a higher raid level
2144 	 *    to provide capacity for redundancy or during reshape
2145 	 *    to add capacity to grow the raid set.
2146 	 */
2147 	d = 0;
2148 	rdev_for_each(r, mddev) {
2149 		if (test_bit(FirstUse, &r->flags))
2150 			new_devs++;
2151 
2152 		if (!test_bit(In_sync, &r->flags)) {
2153 			DMINFO("Device %d specified for rebuild; clearing superblock",
2154 				r->raid_disk);
2155 			rebuilds++;
2156 
2157 			if (test_bit(FirstUse, &r->flags))
2158 				rebuild_and_new++;
2159 		}
2160 
2161 		d++;
2162 	}
2163 
2164 	if (new_devs == rs->raid_disks || !rebuilds) {
2165 		/* Replace a broken device */
2166 		if (new_devs == 1 && !rs->delta_disks)
2167 			;
2168 		if (new_devs == rs->raid_disks) {
2169 			DMINFO("Superblocks created for new raid set");
2170 			set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2171 		} else if (new_devs != rebuilds &&
2172 			   new_devs != rs->delta_disks) {
2173 			DMERR("New device injected into existing raid set without "
2174 			      "'delta_disks' or 'rebuild' parameter specified");
2175 			return -EINVAL;
2176 		}
2177 	} else if (new_devs && new_devs != rebuilds) {
2178 		DMERR("%u 'rebuild' devices cannot be injected into"
2179 		      " a raid set with %u other first-time devices",
2180 		      rebuilds, new_devs);
2181 		return -EINVAL;
2182 	} else if (rebuilds) {
2183 		if (rebuild_and_new && rebuilds != rebuild_and_new) {
2184 			DMERR("new device%s provided without 'rebuild'",
2185 			      new_devs > 1 ? "s" : "");
2186 			return -EINVAL;
2187 		} else if (rs_is_recovering(rs)) {
2188 			DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2189 			      (unsigned long long) mddev->recovery_cp);
2190 			return -EINVAL;
2191 		} else if (rs_is_reshaping(rs)) {
2192 			DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2193 			      (unsigned long long) mddev->reshape_position);
2194 			return -EINVAL;
2195 		}
2196 	}
2197 
2198 	/*
2199 	 * Now we set the Faulty bit for those devices that are
2200 	 * recorded in the superblock as failed.
2201 	 */
2202 	sb_retrieve_failed_devices(sb, failed_devices);
2203 	rdev_for_each(r, mddev) {
2204 		if (!r->sb_page)
2205 			continue;
2206 		sb2 = page_address(r->sb_page);
2207 		sb2->failed_devices = 0;
2208 		memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
2209 
2210 		/*
2211 		 * Check for any device re-ordering.
2212 		 */
2213 		if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2214 			role = le32_to_cpu(sb2->array_position);
2215 			if (role < 0)
2216 				continue;
2217 
2218 			if (role != r->raid_disk) {
2219 				if (rs_is_raid10(rs) && __is_raid10_near(mddev->layout)) {
2220 					if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
2221 					    rs->raid_disks % rs->raid10_copies) {
2222 						rs->ti->error =
2223 							"Cannot change raid10 near set to odd # of devices!";
2224 						return -EINVAL;
2225 					}
2226 
2227 					sb2->array_position = cpu_to_le32(r->raid_disk);
2228 
2229 				} else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
2230 					   !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2231 					   !rt_is_raid1(rs->raid_type)) {
2232 					rs->ti->error = "Cannot change device positions in raid set";
2233 					return -EINVAL;
2234 				}
2235 
2236 				DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
2237 			}
2238 
2239 			/*
2240 			 * Partial recovery is performed on
2241 			 * returning failed devices.
2242 			 */
2243 			if (test_bit(role, (void *) failed_devices))
2244 				set_bit(Faulty, &r->flags);
2245 		}
2246 	}
2247 
2248 	return 0;
2249 }
2250 
2251 static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
2252 {
2253 	struct mddev *mddev = &rs->md;
2254 	struct dm_raid_superblock *sb;
2255 
2256 	if (rs_is_raid0(rs) || !rdev->sb_page)
2257 		return 0;
2258 
2259 	sb = page_address(rdev->sb_page);
2260 
2261 	/*
2262 	 * If mddev->events is not set, we know we have not yet initialized
2263 	 * the array.
2264 	 */
2265 	if (!mddev->events && super_init_validation(rs, rdev))
2266 		return -EINVAL;
2267 
2268 	if (le32_to_cpu(sb->compat_features) &&
2269 	    le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
2270 		rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2271 		return -EINVAL;
2272 	}
2273 
2274 	if (sb->incompat_features) {
2275 		rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
2276 		return -EINVAL;
2277 	}
2278 
2279 	/* Enable bitmap creation for RAID levels != 0 */
2280 	mddev->bitmap_info.offset = rt_is_raid0(rs->raid_type) ? 0 : to_sector(4096);
2281 	rdev->mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2282 
2283 	if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2284 		/* Retrieve device size stored in superblock to be prepared for shrink */
2285 		rdev->sectors = le64_to_cpu(sb->sectors);
2286 		rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
2287 		if (rdev->recovery_offset == MaxSector)
2288 			set_bit(In_sync, &rdev->flags);
2289 		/*
2290 		 * If no reshape in progress -> we're recovering single
2291 		 * disk(s) and have to set the device(s) to out-of-sync
2292 		 */
2293 		else if (!rs_is_reshaping(rs))
2294 			clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
2295 	}
2296 
2297 	/*
2298 	 * If a device comes back, set it as not In_sync and no longer faulty.
2299 	 */
2300 	if (test_and_clear_bit(Faulty, &rdev->flags)) {
2301 		rdev->recovery_offset = 0;
2302 		clear_bit(In_sync, &rdev->flags);
2303 		rdev->saved_raid_disk = rdev->raid_disk;
2304 	}
2305 
2306 	/* Reshape support -> restore repective data offsets */
2307 	rdev->data_offset = le64_to_cpu(sb->data_offset);
2308 	rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
2309 
2310 	return 0;
2311 }
2312 
2313 /*
2314  * Analyse superblocks and select the freshest.
2315  */
2316 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2317 {
2318 	int r;
2319 	struct raid_dev *dev;
2320 	struct md_rdev *rdev, *tmp, *freshest;
2321 	struct mddev *mddev = &rs->md;
2322 
2323 	freshest = NULL;
2324 	rdev_for_each_safe(rdev, tmp, mddev) {
2325 		/*
2326 		 * Skipping super_load due to CTR_FLAG_SYNC will cause
2327 		 * the array to undergo initialization again as
2328 		 * though it were new.	This is the intended effect
2329 		 * of the "sync" directive.
2330 		 *
2331 		 * When reshaping capability is added, we must ensure
2332 		 * that the "sync" directive is disallowed during the
2333 		 * reshape.
2334 		 */
2335 		if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2336 			continue;
2337 
2338 		if (!rdev->meta_bdev)
2339 			continue;
2340 
2341 		r = super_load(rdev, freshest);
2342 
2343 		switch (r) {
2344 		case 1:
2345 			freshest = rdev;
2346 			break;
2347 		case 0:
2348 			break;
2349 		default:
2350 			/*
2351 			 * We have to keep any raid0 data/metadata device pairs or
2352 			 * the MD raid0 personality will fail to start the array.
2353 			 */
2354 			if (rs_is_raid0(rs))
2355 				continue;
2356 
2357 			dev = container_of(rdev, struct raid_dev, rdev);
2358 			if (dev->meta_dev)
2359 				dm_put_device(ti, dev->meta_dev);
2360 
2361 			dev->meta_dev = NULL;
2362 			rdev->meta_bdev = NULL;
2363 
2364 			if (rdev->sb_page)
2365 				put_page(rdev->sb_page);
2366 
2367 			rdev->sb_page = NULL;
2368 
2369 			rdev->sb_loaded = 0;
2370 
2371 			/*
2372 			 * We might be able to salvage the data device
2373 			 * even though the meta device has failed.  For
2374 			 * now, we behave as though '- -' had been
2375 			 * set for this device in the table.
2376 			 */
2377 			if (dev->data_dev)
2378 				dm_put_device(ti, dev->data_dev);
2379 
2380 			dev->data_dev = NULL;
2381 			rdev->bdev = NULL;
2382 
2383 			list_del(&rdev->same_set);
2384 		}
2385 	}
2386 
2387 	if (!freshest)
2388 		return 0;
2389 
2390 	if (validate_raid_redundancy(rs)) {
2391 		rs->ti->error = "Insufficient redundancy to activate array";
2392 		return -EINVAL;
2393 	}
2394 
2395 	/*
2396 	 * Validation of the freshest device provides the source of
2397 	 * validation for the remaining devices.
2398 	 */
2399 	rs->ti->error = "Unable to assemble array: Invalid superblocks";
2400 	if (super_validate(rs, freshest))
2401 		return -EINVAL;
2402 
2403 	rdev_for_each(rdev, mddev)
2404 		if ((rdev != freshest) && super_validate(rs, rdev))
2405 			return -EINVAL;
2406 	return 0;
2407 }
2408 
2409 /*
2410  * Adjust data_offset and new_data_offset on all disk members of @rs
2411  * for out of place reshaping if requested by contructor
2412  *
2413  * We need free space at the beginning of each raid disk for forward
2414  * and at the end for backward reshapes which userspace has to provide
2415  * via remapping/reordering of space.
2416  */
2417 static int rs_adjust_data_offsets(struct raid_set *rs)
2418 {
2419 	sector_t data_offset = 0, new_data_offset = 0;
2420 	struct md_rdev *rdev;
2421 
2422 	/* Constructor did not request data offset change */
2423 	if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2424 		if (!rs_is_reshapable(rs))
2425 			goto out;
2426 
2427 		return 0;
2428 	}
2429 
2430 	/* HM FIXME: get InSync raid_dev? */
2431 	rdev = &rs->dev[0].rdev;
2432 
2433 	if (rs->delta_disks < 0) {
2434 		/*
2435 		 * Removing disks (reshaping backwards):
2436 		 *
2437 		 * - before reshape: data is at offset 0 and free space
2438 		 *		     is at end of each component LV
2439 		 *
2440 		 * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2441 		 */
2442 		data_offset = 0;
2443 		new_data_offset = rs->data_offset;
2444 
2445 	} else if (rs->delta_disks > 0) {
2446 		/*
2447 		 * Adding disks (reshaping forwards):
2448 		 *
2449 		 * - before reshape: data is at offset rs->data_offset != 0 and
2450 		 *		     free space is at begin of each component LV
2451 		 *
2452 		 * - after reshape: data is at offset 0 on each component LV
2453 		 */
2454 		data_offset = rs->data_offset;
2455 		new_data_offset = 0;
2456 
2457 	} else {
2458 		/*
2459 		 * User space passes in 0 for data offset after having removed reshape space
2460 		 *
2461 		 * - or - (data offset != 0)
2462 		 *
2463 		 * Changing RAID layout or chunk size -> toggle offsets
2464 		 *
2465 		 * - before reshape: data is at offset rs->data_offset 0 and
2466 		 *		     free space is at end of each component LV
2467 		 *		     -or-
2468 		 *                   data is at offset rs->data_offset != 0 and
2469 		 *		     free space is at begin of each component LV
2470 		 *
2471 		 * - after reshape: data is at offset 0 if it was at offset != 0
2472 		 *                  or at offset != 0 if it was at offset 0
2473 		 *                  on each component LV
2474 		 *
2475 		 */
2476 		data_offset = rs->data_offset ? rdev->data_offset : 0;
2477 		new_data_offset = data_offset ? 0 : rs->data_offset;
2478 		set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2479 	}
2480 
2481 	/*
2482 	 * Make sure we got a minimum amount of free sectors per device
2483 	 */
2484 	if (rs->data_offset &&
2485 	    to_sector(i_size_read(rdev->bdev->bd_inode)) - rdev->sectors < MIN_FREE_RESHAPE_SPACE) {
2486 		rs->ti->error = data_offset ? "No space for forward reshape" :
2487 					      "No space for backward reshape";
2488 		return -ENOSPC;
2489 	}
2490 out:
2491 	/* Adjust data offsets on all rdevs */
2492 	rdev_for_each(rdev, &rs->md) {
2493 		rdev->data_offset = data_offset;
2494 		rdev->new_data_offset = new_data_offset;
2495 	}
2496 
2497 	return 0;
2498 }
2499 
2500 /* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2501 static void __reorder_raid_disk_indexes(struct raid_set *rs)
2502 {
2503 	int i = 0;
2504 	struct md_rdev *rdev;
2505 
2506 	rdev_for_each(rdev, &rs->md) {
2507 		rdev->raid_disk = i++;
2508 		rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2509 	}
2510 }
2511 
2512 /*
2513  * Setup @rs for takeover by a different raid level
2514  */
2515 static int rs_setup_takeover(struct raid_set *rs)
2516 {
2517 	struct mddev *mddev = &rs->md;
2518 	struct md_rdev *rdev;
2519 	unsigned int d = mddev->raid_disks = rs->raid_disks;
2520 	sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2521 
2522 	if (rt_is_raid10(rs->raid_type)) {
2523 		if (mddev->level == 0) {
2524 			/* Userpace reordered disks -> adjust raid_disk indexes */
2525 			__reorder_raid_disk_indexes(rs);
2526 
2527 			/* raid0 -> raid10_far layout */
2528 			mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2529 								   rs->raid10_copies);
2530 		} else if (mddev->level == 1)
2531 			/* raid1 -> raid10_near layout */
2532 			mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2533 								   rs->raid_disks);
2534 		else
2535 			return -EINVAL;
2536 
2537 	}
2538 
2539 	clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2540 	mddev->recovery_cp = MaxSector;
2541 
2542 	while (d--) {
2543 		rdev = &rs->dev[d].rdev;
2544 
2545 		if (test_bit(d, (void *) rs->rebuild_disks)) {
2546 			clear_bit(In_sync, &rdev->flags);
2547 			clear_bit(Faulty, &rdev->flags);
2548 			mddev->recovery_cp = rdev->recovery_offset = 0;
2549 			/* Bitmap has to be created when we do an "up" takeover */
2550 			set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2551 		}
2552 
2553 		rdev->new_data_offset = new_data_offset;
2554 	}
2555 
2556 	return 0;
2557 }
2558 
2559 /* Prepare @rs for reshape */
2560 static int rs_prepare_reshape(struct raid_set *rs)
2561 {
2562 	bool reshape;
2563 	struct mddev *mddev = &rs->md;
2564 
2565 	if (rs_is_raid10(rs)) {
2566 		if (rs->raid_disks != mddev->raid_disks &&
2567 		    __is_raid10_near(mddev->layout) &&
2568 		    rs->raid10_copies &&
2569 		    rs->raid10_copies != __raid10_near_copies(mddev->layout)) {
2570 			/*
2571 			 * raid disk have to be multiple of data copies to allow this conversion,
2572 			 *
2573 			 * This is actually not a reshape it is a
2574 			 * rebuild of any additional mirrors per group
2575 			 */
2576 			if (rs->raid_disks % rs->raid10_copies) {
2577 				rs->ti->error = "Can't reshape raid10 mirror groups";
2578 				return -EINVAL;
2579 			}
2580 
2581 			/* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2582 			__reorder_raid_disk_indexes(rs);
2583 			mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2584 								   rs->raid10_copies);
2585 			mddev->new_layout = mddev->layout;
2586 			reshape = false;
2587 		} else
2588 			reshape = true;
2589 
2590 	} else if (rs_is_raid456(rs))
2591 		reshape = true;
2592 
2593 	else if (rs_is_raid1(rs)) {
2594 		if (rs->delta_disks) {
2595 			/* Process raid1 via delta_disks */
2596 			mddev->degraded = rs->delta_disks < 0 ? -rs->delta_disks : rs->delta_disks;
2597 			reshape = true;
2598 		} else {
2599 			/* Process raid1 without delta_disks */
2600 			mddev->raid_disks = rs->raid_disks;
2601 			reshape = false;
2602 		}
2603 	} else {
2604 		rs->ti->error = "Called with bogus raid type";
2605 		return -EINVAL;
2606 	}
2607 
2608 	if (reshape) {
2609 		set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2610 		set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2611 	} else if (mddev->raid_disks < rs->raid_disks)
2612 		/* Create new superblocks and bitmaps, if any new disks */
2613 		set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2614 
2615 	return 0;
2616 }
2617 
2618 /*
2619  *
2620  * - change raid layout
2621  * - change chunk size
2622  * - add disks
2623  * - remove disks
2624  */
2625 static int rs_setup_reshape(struct raid_set *rs)
2626 {
2627 	int r = 0;
2628 	unsigned int cur_raid_devs, d;
2629 	struct mddev *mddev = &rs->md;
2630 	struct md_rdev *rdev;
2631 
2632 	mddev->delta_disks = rs->delta_disks;
2633 	cur_raid_devs = mddev->raid_disks;
2634 
2635 	/* Ignore impossible layout change whilst adding/removing disks */
2636 	if (mddev->delta_disks &&
2637 	    mddev->layout != mddev->new_layout) {
2638 		DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2639 		mddev->new_layout = mddev->layout;
2640 	}
2641 
2642 	/*
2643 	 * Adjust array size:
2644 	 *
2645 	 * - in case of adding disks, array size has
2646 	 *   to grow after the disk adding reshape,
2647 	 *   which'll hapen in the event handler;
2648 	 *   reshape will happen forward, so space has to
2649 	 *   be available at the beginning of each disk
2650 	 *
2651 	 * - in case of removing disks, array size
2652 	 *   has to shrink before starting the reshape,
2653 	 *   which'll happen here;
2654 	 *   reshape will happen backward, so space has to
2655 	 *   be available at the end of each disk
2656 	 *
2657 	 * - data_offset and new_data_offset are
2658 	 *   adjusted for aforementioned out of place
2659 	 *   reshaping based on userspace passing in
2660 	 *   the "data_offset <sectors>" key/value
2661 	 *   pair via the constructor
2662 	 */
2663 
2664 	/* Add disk(s) */
2665 	if (rs->delta_disks > 0) {
2666 		/* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2667 		for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2668 			rdev = &rs->dev[d].rdev;
2669 			clear_bit(In_sync, &rdev->flags);
2670 
2671 			/*
2672 			 * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2673 			 * by md, which'll store that erroneously in the superblock on reshape
2674 			 */
2675 			rdev->saved_raid_disk = -1;
2676 			rdev->raid_disk = d;
2677 
2678 			rdev->sectors = mddev->dev_sectors;
2679 			rdev->recovery_offset = rs_is_raid1(rs) ? 0 : MaxSector;
2680 		}
2681 
2682 		mddev->reshape_backwards = 0; /* adding disks -> forward reshape */
2683 
2684 	/* Remove disk(s) */
2685 	} else if (rs->delta_disks < 0) {
2686 		r = rs_set_dev_and_array_sectors(rs, true);
2687 		mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2688 
2689 	/* Change layout and/or chunk size */
2690 	} else {
2691 		/*
2692 		 * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2693 		 *
2694 		 * keeping number of disks and do layout change ->
2695 		 *
2696 		 * toggle reshape_backward depending on data_offset:
2697 		 *
2698 		 * - free space upfront -> reshape forward
2699 		 *
2700 		 * - free space at the end -> reshape backward
2701 		 *
2702 		 *
2703 		 * This utilizes free reshape space avoiding the need
2704 		 * for userspace to move (parts of) LV segments in
2705 		 * case of layout/chunksize change  (for disk
2706 		 * adding/removing reshape space has to be at
2707 		 * the proper address (see above with delta_disks):
2708 		 *
2709 		 * add disk(s)   -> begin
2710 		 * remove disk(s)-> end
2711 		 */
2712 		mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2713 	}
2714 
2715 	return r;
2716 }
2717 
2718 /*
2719  * Enable/disable discard support on RAID set depending on
2720  * RAID level and discard properties of underlying RAID members.
2721  */
2722 static void configure_discard_support(struct raid_set *rs)
2723 {
2724 	int i;
2725 	bool raid456;
2726 	struct dm_target *ti = rs->ti;
2727 
2728 	/* Assume discards not supported until after checks below. */
2729 	ti->discards_supported = false;
2730 
2731 	/* RAID level 4,5,6 require discard_zeroes_data for data integrity! */
2732 	raid456 = (rs->md.level == 4 || rs->md.level == 5 || rs->md.level == 6);
2733 
2734 	for (i = 0; i < rs->raid_disks; i++) {
2735 		struct request_queue *q;
2736 
2737 		if (!rs->dev[i].rdev.bdev)
2738 			continue;
2739 
2740 		q = bdev_get_queue(rs->dev[i].rdev.bdev);
2741 		if (!q || !blk_queue_discard(q))
2742 			return;
2743 
2744 		if (raid456) {
2745 			if (!q->limits.discard_zeroes_data)
2746 				return;
2747 			if (!devices_handle_discard_safely) {
2748 				DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2749 				DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2750 				return;
2751 			}
2752 		}
2753 	}
2754 
2755 	/* All RAID members properly support discards */
2756 	ti->discards_supported = true;
2757 
2758 	/*
2759 	 * RAID1 and RAID10 personalities require bio splitting,
2760 	 * RAID0/4/5/6 don't and process large discard bios properly.
2761 	 */
2762 	ti->split_discard_bios = !!(rs->md.level == 1 || rs->md.level == 10);
2763 	ti->num_discard_bios = 1;
2764 }
2765 
2766 /*
2767  * Construct a RAID0/1/10/4/5/6 mapping:
2768  * Args:
2769  *	<raid_type> <#raid_params> <raid_params>{0,}	\
2770  *	<#raid_devs> [<meta_dev1> <dev1>]{1,}
2771  *
2772  * <raid_params> varies by <raid_type>.	 See 'parse_raid_params' for
2773  * details on possible <raid_params>.
2774  *
2775  * Userspace is free to initialize the metadata devices, hence the superblocks to
2776  * enforce recreation based on the passed in table parameters.
2777  *
2778  */
2779 static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2780 {
2781 	int r;
2782 	bool resize;
2783 	struct raid_type *rt;
2784 	unsigned int num_raid_params, num_raid_devs;
2785 	sector_t calculated_dev_sectors;
2786 	struct raid_set *rs = NULL;
2787 	const char *arg;
2788 	struct rs_layout rs_layout;
2789 	struct dm_arg_set as = { argc, argv }, as_nrd;
2790 	struct dm_arg _args[] = {
2791 		{ 0, as.argc, "Cannot understand number of raid parameters" },
2792 		{ 1, 254, "Cannot understand number of raid devices parameters" }
2793 	};
2794 
2795 	/* Must have <raid_type> */
2796 	arg = dm_shift_arg(&as);
2797 	if (!arg) {
2798 		ti->error = "No arguments";
2799 		return -EINVAL;
2800 	}
2801 
2802 	rt = get_raid_type(arg);
2803 	if (!rt) {
2804 		ti->error = "Unrecognised raid_type";
2805 		return -EINVAL;
2806 	}
2807 
2808 	/* Must have <#raid_params> */
2809 	if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
2810 		return -EINVAL;
2811 
2812 	/* number of raid device tupples <meta_dev data_dev> */
2813 	as_nrd = as;
2814 	dm_consume_args(&as_nrd, num_raid_params);
2815 	_args[1].max = (as_nrd.argc - 1) / 2;
2816 	if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
2817 		return -EINVAL;
2818 
2819 	if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
2820 		ti->error = "Invalid number of supplied raid devices";
2821 		return -EINVAL;
2822 	}
2823 
2824 	rs = raid_set_alloc(ti, rt, num_raid_devs);
2825 	if (IS_ERR(rs))
2826 		return PTR_ERR(rs);
2827 
2828 	r = parse_raid_params(rs, &as, num_raid_params);
2829 	if (r)
2830 		goto bad;
2831 
2832 	r = parse_dev_params(rs, &as);
2833 	if (r)
2834 		goto bad;
2835 
2836 	rs->md.sync_super = super_sync;
2837 
2838 	/*
2839 	 * Calculate ctr requested array and device sizes to allow
2840 	 * for superblock analysis needing device sizes defined.
2841 	 *
2842 	 * Any existing superblock will overwrite the array and device sizes
2843 	 */
2844 	r = rs_set_dev_and_array_sectors(rs, false);
2845 	if (r)
2846 		goto bad;
2847 
2848 	calculated_dev_sectors = rs->dev[0].rdev.sectors;
2849 
2850 	/*
2851 	 * Backup any new raid set level, layout, ...
2852 	 * requested to be able to compare to superblock
2853 	 * members for conversion decisions.
2854 	 */
2855 	rs_config_backup(rs, &rs_layout);
2856 
2857 	r = analyse_superblocks(ti, rs);
2858 	if (r)
2859 		goto bad;
2860 
2861 	resize = calculated_dev_sectors != rs->dev[0].rdev.sectors;
2862 
2863 	INIT_WORK(&rs->md.event_work, do_table_event);
2864 	ti->private = rs;
2865 	ti->num_flush_bios = 1;
2866 
2867 	/* Restore any requested new layout for conversion decision */
2868 	rs_config_restore(rs, &rs_layout);
2869 
2870 	/*
2871 	 * Now that we have any superblock metadata available,
2872 	 * check for new, recovering, reshaping, to be taken over,
2873 	 * to be reshaped or an existing, unchanged raid set to
2874 	 * run in sequence.
2875 	 */
2876 	if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
2877 		/* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
2878 		if (rs_is_raid6(rs) &&
2879 		    test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
2880 			ti->error = "'nosync' not allowed for new raid6 set";
2881 			r = -EINVAL;
2882 			goto bad;
2883 		}
2884 		rs_setup_recovery(rs, 0);
2885 		set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2886 		rs_set_new(rs);
2887 	} else if (rs_is_recovering(rs)) {
2888 		/* A recovering raid set may be resized */
2889 		; /* skip setup rs */
2890 	} else if (rs_is_reshaping(rs)) {
2891 		/* Have to reject size change request during reshape */
2892 		if (resize) {
2893 			ti->error = "Can't resize a reshaping raid set";
2894 			r = -EPERM;
2895 			goto bad;
2896 		}
2897 		/* skip setup rs */
2898 	} else if (rs_takeover_requested(rs)) {
2899 		if (rs_is_reshaping(rs)) {
2900 			ti->error = "Can't takeover a reshaping raid set";
2901 			r = -EPERM;
2902 			goto bad;
2903 		}
2904 
2905 		/*
2906 		 * If a takeover is needed, userspace sets any additional
2907 		 * devices to rebuild and we can check for a valid request here.
2908 		 *
2909 		 * If acceptible, set the level to the new requested
2910 		 * one, prohibit requesting recovery, allow the raid
2911 		 * set to run and store superblocks during resume.
2912 		 */
2913 		r = rs_check_takeover(rs);
2914 		if (r)
2915 			goto bad;
2916 
2917 		r = rs_setup_takeover(rs);
2918 		if (r)
2919 			goto bad;
2920 
2921 		set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2922 		/* Takeover ain't recovery, so disable recovery */
2923 		rs_setup_recovery(rs, MaxSector);
2924 		rs_set_new(rs);
2925 	} else if (rs_reshape_requested(rs)) {
2926 		/*
2927 		  * We can only prepare for a reshape here, because the
2928 		  * raid set needs to run to provide the repective reshape
2929 		  * check functions via its MD personality instance.
2930 		  *
2931 		  * So do the reshape check after md_run() succeeded.
2932 		  */
2933 		r = rs_prepare_reshape(rs);
2934 		if (r)
2935 			return r;
2936 
2937 		/* Reshaping ain't recovery, so disable recovery */
2938 		rs_setup_recovery(rs, MaxSector);
2939 		rs_set_cur(rs);
2940 	} else {
2941 		/* May not set recovery when a device rebuild is requested */
2942 		if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
2943 			rs_setup_recovery(rs, MaxSector);
2944 			set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2945 		} else
2946 			rs_setup_recovery(rs, test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ?
2947 					      0 : (resize ? calculated_dev_sectors : MaxSector));
2948 		rs_set_cur(rs);
2949 	}
2950 
2951 	/* If constructor requested it, change data and new_data offsets */
2952 	r = rs_adjust_data_offsets(rs);
2953 	if (r)
2954 		goto bad;
2955 
2956 	/* Start raid set read-only and assumed clean to change in raid_resume() */
2957 	rs->md.ro = 1;
2958 	rs->md.in_sync = 1;
2959 	set_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
2960 
2961 	/* Has to be held on running the array */
2962 	mddev_lock_nointr(&rs->md);
2963 	r = md_run(&rs->md);
2964 	rs->md.in_sync = 0; /* Assume already marked dirty */
2965 
2966 	if (r) {
2967 		ti->error = "Failed to run raid array";
2968 		mddev_unlock(&rs->md);
2969 		goto bad;
2970 	}
2971 
2972 	rs->callbacks.congested_fn = raid_is_congested;
2973 	dm_table_add_target_callbacks(ti->table, &rs->callbacks);
2974 
2975 	mddev_suspend(&rs->md);
2976 
2977 	/* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
2978 	if (rs_is_raid456(rs)) {
2979 		r = rs_set_raid456_stripe_cache(rs);
2980 		if (r)
2981 			goto bad_stripe_cache;
2982 	}
2983 
2984 	/* Now do an early reshape check */
2985 	if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
2986 		r = rs_check_reshape(rs);
2987 		if (r)
2988 			goto bad_check_reshape;
2989 
2990 		/* Restore new, ctr requested layout to perform check */
2991 		rs_config_restore(rs, &rs_layout);
2992 
2993 		if (rs->md.pers->start_reshape) {
2994 			r = rs->md.pers->check_reshape(&rs->md);
2995 			if (r) {
2996 				ti->error = "Reshape check failed";
2997 				goto bad_check_reshape;
2998 			}
2999 		}
3000 	}
3001 
3002 	/* Disable/enable discard support on raid set. */
3003 	configure_discard_support(rs);
3004 
3005 	mddev_unlock(&rs->md);
3006 	return 0;
3007 
3008 bad_stripe_cache:
3009 bad_check_reshape:
3010 	md_stop(&rs->md);
3011 bad:
3012 	raid_set_free(rs);
3013 
3014 	return r;
3015 }
3016 
3017 static void raid_dtr(struct dm_target *ti)
3018 {
3019 	struct raid_set *rs = ti->private;
3020 
3021 	list_del_init(&rs->callbacks.list);
3022 	md_stop(&rs->md);
3023 	raid_set_free(rs);
3024 }
3025 
3026 static int raid_map(struct dm_target *ti, struct bio *bio)
3027 {
3028 	struct raid_set *rs = ti->private;
3029 	struct mddev *mddev = &rs->md;
3030 
3031 	/*
3032 	 * If we're reshaping to add disk(s)), ti->len and
3033 	 * mddev->array_sectors will differ during the process
3034 	 * (ti->len > mddev->array_sectors), so we have to requeue
3035 	 * bios with addresses > mddev->array_sectors here or
3036 	 * there will occur accesses past EOD of the component
3037 	 * data images thus erroring the raid set.
3038 	 */
3039 	if (unlikely(bio_end_sector(bio) > mddev->array_sectors))
3040 		return DM_MAPIO_REQUEUE;
3041 
3042 	mddev->pers->make_request(mddev, bio);
3043 
3044 	return DM_MAPIO_SUBMITTED;
3045 }
3046 
3047 /* Return string describing the current sync action of @mddev */
3048 static const char *decipher_sync_action(struct mddev *mddev)
3049 {
3050 	if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3051 		return "frozen";
3052 
3053 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3054 	    (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3055 		if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3056 			return "reshape";
3057 
3058 		if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3059 			if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3060 				return "resync";
3061 			else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3062 				return "check";
3063 			return "repair";
3064 		}
3065 
3066 		if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3067 			return "recover";
3068 	}
3069 
3070 	return "idle";
3071 }
3072 
3073 /*
3074  * Return status string @rdev
3075  *
3076  * Status characters:
3077  *
3078  *  'D' = Dead/Failed device
3079  *  'a' = Alive but not in-sync
3080  *  'A' = Alive and in-sync
3081  */
3082 static const char *__raid_dev_status(struct md_rdev *rdev, bool array_in_sync)
3083 {
3084 	if (test_bit(Faulty, &rdev->flags))
3085 		return "D";
3086 	else if (!array_in_sync || !test_bit(In_sync, &rdev->flags))
3087 		return "a";
3088 	else
3089 		return "A";
3090 }
3091 
3092 /* Helper to return resync/reshape progress for @rs and @array_in_sync */
3093 static sector_t rs_get_progress(struct raid_set *rs,
3094 				sector_t resync_max_sectors, bool *array_in_sync)
3095 {
3096 	sector_t r, recovery_cp, curr_resync_completed;
3097 	struct mddev *mddev = &rs->md;
3098 
3099 	curr_resync_completed = mddev->curr_resync_completed ?: mddev->recovery_cp;
3100 	recovery_cp = mddev->recovery_cp;
3101 	*array_in_sync = false;
3102 
3103 	if (rs_is_raid0(rs)) {
3104 		r = resync_max_sectors;
3105 		*array_in_sync = true;
3106 
3107 	} else {
3108 		r = mddev->reshape_position;
3109 
3110 		/* Reshape is relative to the array size */
3111 		if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
3112 		    r != MaxSector) {
3113 			if (r == MaxSector) {
3114 				*array_in_sync = true;
3115 				r = resync_max_sectors;
3116 			} else {
3117 				/* Got to reverse on backward reshape */
3118 				if (mddev->reshape_backwards)
3119 					r = mddev->array_sectors - r;
3120 
3121 				/* Devide by # of data stripes */
3122 				sector_div(r, mddev_data_stripes(rs));
3123 			}
3124 
3125 		/* Sync is relative to the component device size */
3126 		} else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3127 			r = curr_resync_completed;
3128 		else
3129 			r = recovery_cp;
3130 
3131 		if (r == MaxSector) {
3132 			/*
3133 			 * Sync complete.
3134 			 */
3135 			*array_in_sync = true;
3136 			r = resync_max_sectors;
3137 		} else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
3138 			/*
3139 			 * If "check" or "repair" is occurring, the raid set has
3140 			 * undergone an initial sync and the health characters
3141 			 * should not be 'a' anymore.
3142 			 */
3143 			*array_in_sync = true;
3144 		} else {
3145 			struct md_rdev *rdev;
3146 
3147 			/*
3148 			 * The raid set may be doing an initial sync, or it may
3149 			 * be rebuilding individual components.	 If all the
3150 			 * devices are In_sync, then it is the raid set that is
3151 			 * being initialized.
3152 			 */
3153 			rdev_for_each(rdev, mddev)
3154 				if (!test_bit(In_sync, &rdev->flags))
3155 					*array_in_sync = true;
3156 #if 0
3157 			r = 0; /* HM FIXME: TESTME: https://bugzilla.redhat.com/show_bug.cgi?id=1210637 ? */
3158 #endif
3159 		}
3160 	}
3161 
3162 	return r;
3163 }
3164 
3165 /* Helper to return @dev name or "-" if !@dev */
3166 static const char *__get_dev_name(struct dm_dev *dev)
3167 {
3168 	return dev ? dev->name : "-";
3169 }
3170 
3171 static void raid_status(struct dm_target *ti, status_type_t type,
3172 			unsigned int status_flags, char *result, unsigned int maxlen)
3173 {
3174 	struct raid_set *rs = ti->private;
3175 	struct mddev *mddev = &rs->md;
3176 	struct r5conf *conf = mddev->private;
3177 	int i, max_nr_stripes = conf ? conf->max_nr_stripes : 0;
3178 	bool array_in_sync;
3179 	unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3180 	unsigned int sz = 0;
3181 	unsigned int rebuild_disks;
3182 	unsigned int write_mostly_params = 0;
3183 	sector_t progress, resync_max_sectors, resync_mismatches;
3184 	const char *sync_action;
3185 	struct raid_type *rt;
3186 	struct md_rdev *rdev;
3187 
3188 	switch (type) {
3189 	case STATUSTYPE_INFO:
3190 		/* *Should* always succeed */
3191 		rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3192 		if (!rt)
3193 			return;
3194 
3195 		DMEMIT("%s %d ", rt->name, mddev->raid_disks);
3196 
3197 		/* Access most recent mddev properties for status output */
3198 		smp_rmb();
3199 		/* Get sensible max sectors even if raid set not yet started */
3200 		resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
3201 				      mddev->resync_max_sectors : mddev->dev_sectors;
3202 		progress = rs_get_progress(rs, resync_max_sectors, &array_in_sync);
3203 		resync_mismatches = (mddev->last_sync_action && !strcasecmp(mddev->last_sync_action, "check")) ?
3204 				    atomic64_read(&mddev->resync_mismatches) : 0;
3205 		sync_action = decipher_sync_action(&rs->md);
3206 
3207 		/* HM FIXME: do we want another state char for raid0? It shows 'D' or 'A' now */
3208 		rdev_for_each(rdev, mddev)
3209 			DMEMIT(__raid_dev_status(rdev, array_in_sync));
3210 
3211 		/*
3212 		 * In-sync/Reshape ratio:
3213 		 *  The in-sync ratio shows the progress of:
3214 		 *   - Initializing the raid set
3215 		 *   - Rebuilding a subset of devices of the raid set
3216 		 *  The user can distinguish between the two by referring
3217 		 *  to the status characters.
3218 		 *
3219 		 *  The reshape ratio shows the progress of
3220 		 *  changing the raid layout or the number of
3221 		 *  disks of a raid set
3222 		 */
3223 		DMEMIT(" %llu/%llu", (unsigned long long) progress,
3224 				     (unsigned long long) resync_max_sectors);
3225 
3226 		/*
3227 		 * v1.5.0+:
3228 		 *
3229 		 * Sync action:
3230 		 *   See Documentation/device-mapper/dm-raid.txt for
3231 		 *   information on each of these states.
3232 		 */
3233 		DMEMIT(" %s", sync_action);
3234 
3235 		/*
3236 		 * v1.5.0+:
3237 		 *
3238 		 * resync_mismatches/mismatch_cnt
3239 		 *   This field shows the number of discrepancies found when
3240 		 *   performing a "check" of the raid set.
3241 		 */
3242 		DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3243 
3244 		/*
3245 		 * v1.9.0+:
3246 		 *
3247 		 * data_offset (needed for out of space reshaping)
3248 		 *   This field shows the data offset into the data
3249 		 *   image LV where the first stripes data starts.
3250 		 *
3251 		 * We keep data_offset equal on all raid disks of the set,
3252 		 * so retrieving it from the first raid disk is sufficient.
3253 		 */
3254 		DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
3255 		break;
3256 
3257 	case STATUSTYPE_TABLE:
3258 		/* Report the table line string you would use to construct this raid set */
3259 
3260 		/* Calculate raid parameter count */
3261 		for (i = 0; i < rs->raid_disks; i++)
3262 			if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3263 				write_mostly_params += 2;
3264 		rebuild_disks = memweight(rs->rebuild_disks, DISKS_ARRAY_ELEMS * sizeof(*rs->rebuild_disks));
3265 		raid_param_cnt += rebuild_disks * 2 +
3266 				  write_mostly_params +
3267 				  hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3268 				  hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2;
3269 		/* Emit table line */
3270 		DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
3271 		if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3272 			DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3273 					 raid10_md_layout_to_format(mddev->layout));
3274 		if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3275 			DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3276 					 raid10_md_layout_to_copies(mddev->layout));
3277 		if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3278 			DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
3279 		if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
3280 			DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
3281 		if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3282 			DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3283 					   (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3284 		if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3285 			DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3286 					   (unsigned long long) rs->data_offset);
3287 		if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3288 			DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3289 					  mddev->bitmap_info.daemon_sleep);
3290 		if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3291 			DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3292 					 max(rs->delta_disks, mddev->delta_disks));
3293 		if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3294 			DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3295 					 max_nr_stripes);
3296 		if (rebuild_disks)
3297 			for (i = 0; i < rs->raid_disks; i++)
3298 				if (test_bit(rs->dev[i].rdev.raid_disk, (void *) rs->rebuild_disks))
3299 					DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD),
3300 							 rs->dev[i].rdev.raid_disk);
3301 		if (write_mostly_params)
3302 			for (i = 0; i < rs->raid_disks; i++)
3303 				if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3304 					DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3305 					       rs->dev[i].rdev.raid_disk);
3306 		if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
3307 			DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
3308 					  mddev->bitmap_info.max_write_behind);
3309 		if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3310 			DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3311 					 mddev->sync_speed_max);
3312 		if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3313 			DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3314 					 mddev->sync_speed_min);
3315 		DMEMIT(" %d", rs->raid_disks);
3316 		for (i = 0; i < rs->raid_disks; i++)
3317 			DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
3318 					 __get_dev_name(rs->dev[i].data_dev));
3319 	}
3320 }
3321 
3322 static int raid_message(struct dm_target *ti, unsigned int argc, char **argv)
3323 {
3324 	struct raid_set *rs = ti->private;
3325 	struct mddev *mddev = &rs->md;
3326 
3327 	if (!mddev->pers || !mddev->pers->sync_request)
3328 		return -EINVAL;
3329 
3330 	if (!strcasecmp(argv[0], "frozen"))
3331 		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3332 	else
3333 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3334 
3335 	if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
3336 		if (mddev->sync_thread) {
3337 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3338 			md_reap_sync_thread(mddev);
3339 		}
3340 	} else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3341 		   test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3342 		return -EBUSY;
3343 	else if (!strcasecmp(argv[0], "resync"))
3344 		; /* MD_RECOVERY_NEEDED set below */
3345 	else if (!strcasecmp(argv[0], "recover"))
3346 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3347 	else {
3348 		if (!strcasecmp(argv[0], "check"))
3349 			set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3350 		else if (!!strcasecmp(argv[0], "repair"))
3351 			return -EINVAL;
3352 		set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3353 		set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3354 	}
3355 	if (mddev->ro == 2) {
3356 		/* A write to sync_action is enough to justify
3357 		 * canceling read-auto mode
3358 		 */
3359 		mddev->ro = 0;
3360 		if (!mddev->suspended && mddev->sync_thread)
3361 			md_wakeup_thread(mddev->sync_thread);
3362 	}
3363 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3364 	if (!mddev->suspended && mddev->thread)
3365 		md_wakeup_thread(mddev->thread);
3366 
3367 	return 0;
3368 }
3369 
3370 static int raid_iterate_devices(struct dm_target *ti,
3371 				iterate_devices_callout_fn fn, void *data)
3372 {
3373 	struct raid_set *rs = ti->private;
3374 	unsigned int i;
3375 	int r = 0;
3376 
3377 	for (i = 0; !r && i < rs->md.raid_disks; i++)
3378 		if (rs->dev[i].data_dev)
3379 			r = fn(ti,
3380 				 rs->dev[i].data_dev,
3381 				 0, /* No offset on data devs */
3382 				 rs->md.dev_sectors,
3383 				 data);
3384 
3385 	return r;
3386 }
3387 
3388 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3389 {
3390 	struct raid_set *rs = ti->private;
3391 	unsigned int chunk_size = to_bytes(rs->md.chunk_sectors);
3392 
3393 	blk_limits_io_min(limits, chunk_size);
3394 	blk_limits_io_opt(limits, chunk_size * mddev_data_stripes(rs));
3395 }
3396 
3397 static void raid_presuspend(struct dm_target *ti)
3398 {
3399 	struct raid_set *rs = ti->private;
3400 
3401 	md_stop_writes(&rs->md);
3402 }
3403 
3404 static void raid_postsuspend(struct dm_target *ti)
3405 {
3406 	struct raid_set *rs = ti->private;
3407 
3408 	if (!rs->md.suspended)
3409 		mddev_suspend(&rs->md);
3410 
3411 	rs->md.ro = 1;
3412 }
3413 
3414 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
3415 {
3416 	int i;
3417 	uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
3418 	unsigned long flags;
3419 	bool cleared = false;
3420 	struct dm_raid_superblock *sb;
3421 	struct mddev *mddev = &rs->md;
3422 	struct md_rdev *r;
3423 
3424 	/* RAID personalities have to provide hot add/remove methods or we need to bail out. */
3425 	if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
3426 		return;
3427 
3428 	memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
3429 
3430 	for (i = 0; i < rs->md.raid_disks; i++) {
3431 		r = &rs->dev[i].rdev;
3432 		if (test_bit(Faulty, &r->flags) && r->sb_page &&
3433 		    sync_page_io(r, 0, r->sb_size, r->sb_page,
3434 				 REQ_OP_READ, 0, true)) {
3435 			DMINFO("Faulty %s device #%d has readable super block."
3436 			       "  Attempting to revive it.",
3437 			       rs->raid_type->name, i);
3438 
3439 			/*
3440 			 * Faulty bit may be set, but sometimes the array can
3441 			 * be suspended before the personalities can respond
3442 			 * by removing the device from the array (i.e. calling
3443 			 * 'hot_remove_disk').	If they haven't yet removed
3444 			 * the failed device, its 'raid_disk' number will be
3445 			 * '>= 0' - meaning we must call this function
3446 			 * ourselves.
3447 			 */
3448 			if ((r->raid_disk >= 0) &&
3449 			    (mddev->pers->hot_remove_disk(mddev, r) != 0))
3450 				/* Failed to revive this device, try next */
3451 				continue;
3452 
3453 			r->raid_disk = i;
3454 			r->saved_raid_disk = i;
3455 			flags = r->flags;
3456 			clear_bit(Faulty, &r->flags);
3457 			clear_bit(WriteErrorSeen, &r->flags);
3458 			clear_bit(In_sync, &r->flags);
3459 			if (mddev->pers->hot_add_disk(mddev, r)) {
3460 				r->raid_disk = -1;
3461 				r->saved_raid_disk = -1;
3462 				r->flags = flags;
3463 			} else {
3464 				r->recovery_offset = 0;
3465 				set_bit(i, (void *) cleared_failed_devices);
3466 				cleared = true;
3467 			}
3468 		}
3469 	}
3470 
3471 	/* If any failed devices could be cleared, update all sbs failed_devices bits */
3472 	if (cleared) {
3473 		uint64_t failed_devices[DISKS_ARRAY_ELEMS];
3474 
3475 		rdev_for_each(r, &rs->md) {
3476 			sb = page_address(r->sb_page);
3477 			sb_retrieve_failed_devices(sb, failed_devices);
3478 
3479 			for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
3480 				failed_devices[i] &= ~cleared_failed_devices[i];
3481 
3482 			sb_update_failed_devices(sb, failed_devices);
3483 		}
3484 	}
3485 }
3486 
3487 static int __load_dirty_region_bitmap(struct raid_set *rs)
3488 {
3489 	int r = 0;
3490 
3491 	/* Try loading the bitmap unless "raid0", which does not have one */
3492 	if (!rs_is_raid0(rs) &&
3493 	    !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
3494 		r = bitmap_load(&rs->md);
3495 		if (r)
3496 			DMERR("Failed to load bitmap");
3497 	}
3498 
3499 	return r;
3500 }
3501 
3502 /* Enforce updating all superblocks */
3503 static void rs_update_sbs(struct raid_set *rs)
3504 {
3505 	struct mddev *mddev = &rs->md;
3506 	int ro = mddev->ro;
3507 
3508 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3509 	mddev->ro = 0;
3510 	md_update_sb(mddev, 1);
3511 	mddev->ro = ro;
3512 }
3513 
3514 /*
3515  * Reshape changes raid algorithm of @rs to new one within personality
3516  * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3517  * disks from a raid set thus growing/shrinking it or resizes the set
3518  *
3519  * Call mddev_lock_nointr() before!
3520  */
3521 static int rs_start_reshape(struct raid_set *rs)
3522 {
3523 	int r;
3524 	struct mddev *mddev = &rs->md;
3525 	struct md_personality *pers = mddev->pers;
3526 
3527 	r = rs_setup_reshape(rs);
3528 	if (r)
3529 		return r;
3530 
3531 	/* Need to be resumed to be able to start reshape, recovery is frozen until raid_resume() though */
3532 	if (mddev->suspended)
3533 		mddev_resume(mddev);
3534 
3535 	/*
3536 	 * Check any reshape constraints enforced by the personalility
3537 	 *
3538 	 * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3539 	 */
3540 	r = pers->check_reshape(mddev);
3541 	if (r) {
3542 		rs->ti->error = "pers->check_reshape() failed";
3543 		return r;
3544 	}
3545 
3546 	/*
3547 	 * Personality may not provide start reshape method in which
3548 	 * case check_reshape above has already covered everything
3549 	 */
3550 	if (pers->start_reshape) {
3551 		r = pers->start_reshape(mddev);
3552 		if (r) {
3553 			rs->ti->error = "pers->start_reshape() failed";
3554 			return r;
3555 		}
3556 	}
3557 
3558 	/* Suspend because a resume will happen in raid_resume() */
3559 	if (!mddev->suspended)
3560 		mddev_suspend(mddev);
3561 
3562 	/*
3563 	 * Now reshape got set up, update superblocks to
3564 	 * reflect the fact so that a table reload will
3565 	 * access proper superblock content in the ctr.
3566 	 */
3567 	rs_update_sbs(rs);
3568 
3569 	return 0;
3570 }
3571 
3572 static int raid_preresume(struct dm_target *ti)
3573 {
3574 	int r;
3575 	struct raid_set *rs = ti->private;
3576 	struct mddev *mddev = &rs->md;
3577 
3578 	/* This is a resume after a suspend of the set -> it's already started */
3579 	if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
3580 		return 0;
3581 
3582 	/*
3583 	 * The superblocks need to be updated on disk if the
3584 	 * array is new or new devices got added (thus zeroed
3585 	 * out by userspace) or __load_dirty_region_bitmap
3586 	 * will overwrite them in core with old data or fail.
3587 	 */
3588 	if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
3589 		rs_update_sbs(rs);
3590 
3591 	/* Load the bitmap from disk unless raid0 */
3592 	r = __load_dirty_region_bitmap(rs);
3593 	if (r)
3594 		return r;
3595 
3596 	/* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) */
3597 	if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) &&
3598 	    mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)) {
3599 		r = bitmap_resize(mddev->bitmap, mddev->dev_sectors,
3600 				  to_bytes(rs->requested_bitmap_chunk_sectors), 0);
3601 		if (r)
3602 			DMERR("Failed to resize bitmap");
3603 	}
3604 
3605 	/* Check for any resize/reshape on @rs and adjust/initiate */
3606 	/* Be prepared for mddev_resume() in raid_resume() */
3607 	set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3608 	if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
3609 		set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3610 		mddev->resync_min = mddev->recovery_cp;
3611 	}
3612 
3613 	rs_set_capacity(rs);
3614 
3615 	/* Check for any reshape request unless new raid set */
3616 	if (test_and_clear_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3617 		/* Initiate a reshape. */
3618 		mddev_lock_nointr(mddev);
3619 		r = rs_start_reshape(rs);
3620 		mddev_unlock(mddev);
3621 		if (r)
3622 			DMWARN("Failed to check/start reshape, continuing without change");
3623 		r = 0;
3624 	}
3625 
3626 	return r;
3627 }
3628 
3629 static void raid_resume(struct dm_target *ti)
3630 {
3631 	struct raid_set *rs = ti->private;
3632 	struct mddev *mddev = &rs->md;
3633 
3634 	if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
3635 		/*
3636 		 * A secondary resume while the device is active.
3637 		 * Take this opportunity to check whether any failed
3638 		 * devices are reachable again.
3639 		 */
3640 		attempt_restore_of_faulty_devices(rs);
3641 	}
3642 
3643 	mddev->ro = 0;
3644 	mddev->in_sync = 0;
3645 
3646 	clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3647 
3648 	if (mddev->suspended)
3649 		mddev_resume(mddev);
3650 }
3651 
3652 static struct target_type raid_target = {
3653 	.name = "raid",
3654 	.version = {1, 9, 1},
3655 	.module = THIS_MODULE,
3656 	.ctr = raid_ctr,
3657 	.dtr = raid_dtr,
3658 	.map = raid_map,
3659 	.status = raid_status,
3660 	.message = raid_message,
3661 	.iterate_devices = raid_iterate_devices,
3662 	.io_hints = raid_io_hints,
3663 	.presuspend = raid_presuspend,
3664 	.postsuspend = raid_postsuspend,
3665 	.preresume = raid_preresume,
3666 	.resume = raid_resume,
3667 };
3668 
3669 static int __init dm_raid_init(void)
3670 {
3671 	DMINFO("Loading target version %u.%u.%u",
3672 	       raid_target.version[0],
3673 	       raid_target.version[1],
3674 	       raid_target.version[2]);
3675 	return dm_register_target(&raid_target);
3676 }
3677 
3678 static void __exit dm_raid_exit(void)
3679 {
3680 	dm_unregister_target(&raid_target);
3681 }
3682 
3683 module_init(dm_raid_init);
3684 module_exit(dm_raid_exit);
3685 
3686 module_param(devices_handle_discard_safely, bool, 0644);
3687 MODULE_PARM_DESC(devices_handle_discard_safely,
3688 		 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
3689 
3690 MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
3691 MODULE_ALIAS("dm-raid0");
3692 MODULE_ALIAS("dm-raid1");
3693 MODULE_ALIAS("dm-raid10");
3694 MODULE_ALIAS("dm-raid4");
3695 MODULE_ALIAS("dm-raid5");
3696 MODULE_ALIAS("dm-raid6");
3697 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
3698 MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
3699 MODULE_LICENSE("GPL");
3700