xref: /linux/drivers/md/dm-raid.c (revision 26bb0d3f38a764b743a3ad5c8b6e5b5044d7ceb4)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2010-2011 Neil Brown
4  * Copyright (C) 2010-2018 Red Hat, Inc. All rights reserved.
5  *
6  * This file is released under the GPL.
7  */
8 
9 #include <linux/slab.h>
10 #include <linux/module.h>
11 
12 #include "md.h"
13 #include "raid1.h"
14 #include "raid5.h"
15 #include "raid10.h"
16 #include "md-bitmap.h"
17 
18 #include <linux/device-mapper.h>
19 
20 #define DM_MSG_PREFIX "raid"
21 #define	MAX_RAID_DEVICES	253 /* md-raid kernel limit */
22 
23 /*
24  * Minimum sectors of free reshape space per raid device
25  */
26 #define	MIN_FREE_RESHAPE_SPACE to_sector(4*4096)
27 
28 /*
29  * Minimum journal space 4 MiB in sectors.
30  */
31 #define	MIN_RAID456_JOURNAL_SPACE (4*2048)
32 
33 static bool devices_handle_discard_safely;
34 
35 /*
36  * The following flags are used by dm-raid to set up the array state.
37  * They must be cleared before md_run is called.
38  */
39 #define FirstUse 10		/* rdev flag */
40 
41 struct raid_dev {
42 	/*
43 	 * Two DM devices, one to hold metadata and one to hold the
44 	 * actual data/parity.	The reason for this is to not confuse
45 	 * ti->len and give more flexibility in altering size and
46 	 * characteristics.
47 	 *
48 	 * While it is possible for this device to be associated
49 	 * with a different physical device than the data_dev, it
50 	 * is intended for it to be the same.
51 	 *    |--------- Physical Device ---------|
52 	 *    |- meta_dev -|------ data_dev ------|
53 	 */
54 	struct dm_dev *meta_dev;
55 	struct dm_dev *data_dev;
56 	struct md_rdev rdev;
57 };
58 
59 /*
60  * Bits for establishing rs->ctr_flags
61  *
62  * 1 = no flag value
63  * 2 = flag with value
64  */
65 #define __CTR_FLAG_SYNC			0  /* 1 */ /* Not with raid0! */
66 #define __CTR_FLAG_NOSYNC		1  /* 1 */ /* Not with raid0! */
67 #define __CTR_FLAG_REBUILD		2  /* 2 */ /* Not with raid0! */
68 #define __CTR_FLAG_DAEMON_SLEEP		3  /* 2 */ /* Not with raid0! */
69 #define __CTR_FLAG_MIN_RECOVERY_RATE	4  /* 2 */ /* Not with raid0! */
70 #define __CTR_FLAG_MAX_RECOVERY_RATE	5  /* 2 */ /* Not with raid0! */
71 #define __CTR_FLAG_MAX_WRITE_BEHIND	6  /* 2 */ /* Only with raid1! */
72 #define __CTR_FLAG_WRITE_MOSTLY		7  /* 2 */ /* Only with raid1! */
73 #define __CTR_FLAG_STRIPE_CACHE		8  /* 2 */ /* Only with raid4/5/6! */
74 #define __CTR_FLAG_REGION_SIZE		9  /* 2 */ /* Not with raid0! */
75 #define __CTR_FLAG_RAID10_COPIES	10 /* 2 */ /* Only with raid10 */
76 #define __CTR_FLAG_RAID10_FORMAT	11 /* 2 */ /* Only with raid10 */
77 /* New for v1.9.0 */
78 #define __CTR_FLAG_DELTA_DISKS		12 /* 2 */ /* Only with reshapable raid1/4/5/6/10! */
79 #define __CTR_FLAG_DATA_OFFSET		13 /* 2 */ /* Only with reshapable raid4/5/6/10! */
80 #define __CTR_FLAG_RAID10_USE_NEAR_SETS 14 /* 2 */ /* Only with raid10! */
81 
82 /* New for v1.10.0 */
83 #define __CTR_FLAG_JOURNAL_DEV		15 /* 2 */ /* Only with raid4/5/6 (journal device)! */
84 
85 /* New for v1.11.1 */
86 #define __CTR_FLAG_JOURNAL_MODE		16 /* 2 */ /* Only with raid4/5/6 (journal mode)! */
87 
88 /*
89  * Flags for rs->ctr_flags field.
90  */
91 #define CTR_FLAG_SYNC			(1 << __CTR_FLAG_SYNC)
92 #define CTR_FLAG_NOSYNC			(1 << __CTR_FLAG_NOSYNC)
93 #define CTR_FLAG_REBUILD		(1 << __CTR_FLAG_REBUILD)
94 #define CTR_FLAG_DAEMON_SLEEP		(1 << __CTR_FLAG_DAEMON_SLEEP)
95 #define CTR_FLAG_MIN_RECOVERY_RATE	(1 << __CTR_FLAG_MIN_RECOVERY_RATE)
96 #define CTR_FLAG_MAX_RECOVERY_RATE	(1 << __CTR_FLAG_MAX_RECOVERY_RATE)
97 #define CTR_FLAG_MAX_WRITE_BEHIND	(1 << __CTR_FLAG_MAX_WRITE_BEHIND)
98 #define CTR_FLAG_WRITE_MOSTLY		(1 << __CTR_FLAG_WRITE_MOSTLY)
99 #define CTR_FLAG_STRIPE_CACHE		(1 << __CTR_FLAG_STRIPE_CACHE)
100 #define CTR_FLAG_REGION_SIZE		(1 << __CTR_FLAG_REGION_SIZE)
101 #define CTR_FLAG_RAID10_COPIES		(1 << __CTR_FLAG_RAID10_COPIES)
102 #define CTR_FLAG_RAID10_FORMAT		(1 << __CTR_FLAG_RAID10_FORMAT)
103 #define CTR_FLAG_DELTA_DISKS		(1 << __CTR_FLAG_DELTA_DISKS)
104 #define CTR_FLAG_DATA_OFFSET		(1 << __CTR_FLAG_DATA_OFFSET)
105 #define CTR_FLAG_RAID10_USE_NEAR_SETS	(1 << __CTR_FLAG_RAID10_USE_NEAR_SETS)
106 #define CTR_FLAG_JOURNAL_DEV		(1 << __CTR_FLAG_JOURNAL_DEV)
107 #define CTR_FLAG_JOURNAL_MODE		(1 << __CTR_FLAG_JOURNAL_MODE)
108 
109 /*
110  * Definitions of various constructor flags to
111  * be used in checks of valid / invalid flags
112  * per raid level.
113  */
114 /* Define all any sync flags */
115 #define	CTR_FLAGS_ANY_SYNC		(CTR_FLAG_SYNC | CTR_FLAG_NOSYNC)
116 
117 /* Define flags for options without argument (e.g. 'nosync') */
118 #define	CTR_FLAG_OPTIONS_NO_ARGS	(CTR_FLAGS_ANY_SYNC | \
119 					 CTR_FLAG_RAID10_USE_NEAR_SETS)
120 
121 /* Define flags for options with one argument (e.g. 'delta_disks +2') */
122 #define CTR_FLAG_OPTIONS_ONE_ARG (CTR_FLAG_REBUILD | \
123 				  CTR_FLAG_WRITE_MOSTLY | \
124 				  CTR_FLAG_DAEMON_SLEEP | \
125 				  CTR_FLAG_MIN_RECOVERY_RATE | \
126 				  CTR_FLAG_MAX_RECOVERY_RATE | \
127 				  CTR_FLAG_MAX_WRITE_BEHIND | \
128 				  CTR_FLAG_STRIPE_CACHE | \
129 				  CTR_FLAG_REGION_SIZE | \
130 				  CTR_FLAG_RAID10_COPIES | \
131 				  CTR_FLAG_RAID10_FORMAT | \
132 				  CTR_FLAG_DELTA_DISKS | \
133 				  CTR_FLAG_DATA_OFFSET | \
134 				  CTR_FLAG_JOURNAL_DEV | \
135 				  CTR_FLAG_JOURNAL_MODE)
136 
137 /* Valid options definitions per raid level... */
138 
139 /* "raid0" does only accept data offset */
140 #define RAID0_VALID_FLAGS	(CTR_FLAG_DATA_OFFSET)
141 
142 /* "raid1" does not accept stripe cache, data offset, delta_disks or any raid10 options */
143 #define RAID1_VALID_FLAGS	(CTR_FLAGS_ANY_SYNC | \
144 				 CTR_FLAG_REBUILD | \
145 				 CTR_FLAG_WRITE_MOSTLY | \
146 				 CTR_FLAG_DAEMON_SLEEP | \
147 				 CTR_FLAG_MIN_RECOVERY_RATE | \
148 				 CTR_FLAG_MAX_RECOVERY_RATE | \
149 				 CTR_FLAG_MAX_WRITE_BEHIND | \
150 				 CTR_FLAG_REGION_SIZE | \
151 				 CTR_FLAG_DELTA_DISKS | \
152 				 CTR_FLAG_DATA_OFFSET)
153 
154 /* "raid10" does not accept any raid1 or stripe cache options */
155 #define RAID10_VALID_FLAGS	(CTR_FLAGS_ANY_SYNC | \
156 				 CTR_FLAG_REBUILD | \
157 				 CTR_FLAG_DAEMON_SLEEP | \
158 				 CTR_FLAG_MIN_RECOVERY_RATE | \
159 				 CTR_FLAG_MAX_RECOVERY_RATE | \
160 				 CTR_FLAG_REGION_SIZE | \
161 				 CTR_FLAG_RAID10_COPIES | \
162 				 CTR_FLAG_RAID10_FORMAT | \
163 				 CTR_FLAG_DELTA_DISKS | \
164 				 CTR_FLAG_DATA_OFFSET | \
165 				 CTR_FLAG_RAID10_USE_NEAR_SETS)
166 
167 /*
168  * "raid4/5/6" do not accept any raid1 or raid10 specific options
169  *
170  * "raid6" does not accept "nosync", because it is not guaranteed
171  * that both parity and q-syndrome are being written properly with
172  * any writes
173  */
174 #define RAID45_VALID_FLAGS	(CTR_FLAGS_ANY_SYNC | \
175 				 CTR_FLAG_REBUILD | \
176 				 CTR_FLAG_DAEMON_SLEEP | \
177 				 CTR_FLAG_MIN_RECOVERY_RATE | \
178 				 CTR_FLAG_MAX_RECOVERY_RATE | \
179 				 CTR_FLAG_STRIPE_CACHE | \
180 				 CTR_FLAG_REGION_SIZE | \
181 				 CTR_FLAG_DELTA_DISKS | \
182 				 CTR_FLAG_DATA_OFFSET | \
183 				 CTR_FLAG_JOURNAL_DEV | \
184 				 CTR_FLAG_JOURNAL_MODE)
185 
186 #define RAID6_VALID_FLAGS	(CTR_FLAG_SYNC | \
187 				 CTR_FLAG_REBUILD | \
188 				 CTR_FLAG_DAEMON_SLEEP | \
189 				 CTR_FLAG_MIN_RECOVERY_RATE | \
190 				 CTR_FLAG_MAX_RECOVERY_RATE | \
191 				 CTR_FLAG_STRIPE_CACHE | \
192 				 CTR_FLAG_REGION_SIZE | \
193 				 CTR_FLAG_DELTA_DISKS | \
194 				 CTR_FLAG_DATA_OFFSET | \
195 				 CTR_FLAG_JOURNAL_DEV | \
196 				 CTR_FLAG_JOURNAL_MODE)
197 /* ...valid options definitions per raid level */
198 
199 /*
200  * Flags for rs->runtime_flags field
201  * (RT_FLAG prefix meaning "runtime flag")
202  *
203  * These are all internal and used to define runtime state,
204  * e.g. to prevent another resume from preresume processing
205  * the raid set all over again.
206  */
207 #define RT_FLAG_RS_PRERESUMED		0
208 #define RT_FLAG_RS_RESUMED		1
209 #define RT_FLAG_RS_BITMAP_LOADED	2
210 #define RT_FLAG_UPDATE_SBS		3
211 #define RT_FLAG_RESHAPE_RS		4
212 #define RT_FLAG_RS_SUSPENDED		5
213 #define RT_FLAG_RS_IN_SYNC		6
214 #define RT_FLAG_RS_RESYNCING		7
215 #define RT_FLAG_RS_GROW			8
216 #define RT_FLAG_RS_FROZEN		9
217 
218 /* Array elements of 64 bit needed for rebuild/failed disk bits */
219 #define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
220 
221 /*
222  * raid set level, layout and chunk sectors backup/restore
223  */
224 struct rs_layout {
225 	int new_level;
226 	int new_layout;
227 	int new_chunk_sectors;
228 };
229 
230 struct raid_set {
231 	struct dm_target *ti;
232 
233 	uint32_t stripe_cache_entries;
234 	unsigned long ctr_flags;
235 	unsigned long runtime_flags;
236 
237 	uint64_t rebuild_disks[DISKS_ARRAY_ELEMS];
238 
239 	int raid_disks;
240 	int delta_disks;
241 	int data_offset;
242 	int raid10_copies;
243 	int requested_bitmap_chunk_sectors;
244 
245 	struct mddev md;
246 	struct raid_type *raid_type;
247 
248 	sector_t array_sectors;
249 	sector_t dev_sectors;
250 
251 	/* Optional raid4/5/6 journal device */
252 	struct journal_dev {
253 		struct dm_dev *dev;
254 		struct md_rdev rdev;
255 		int mode;
256 	} journal_dev;
257 
258 	struct raid_dev dev[] __counted_by(raid_disks);
259 };
260 
rs_config_backup(struct raid_set * rs,struct rs_layout * l)261 static void rs_config_backup(struct raid_set *rs, struct rs_layout *l)
262 {
263 	struct mddev *mddev = &rs->md;
264 
265 	l->new_level = mddev->new_level;
266 	l->new_layout = mddev->new_layout;
267 	l->new_chunk_sectors = mddev->new_chunk_sectors;
268 }
269 
rs_config_restore(struct raid_set * rs,struct rs_layout * l)270 static void rs_config_restore(struct raid_set *rs, struct rs_layout *l)
271 {
272 	struct mddev *mddev = &rs->md;
273 
274 	mddev->new_level = l->new_level;
275 	mddev->new_layout = l->new_layout;
276 	mddev->new_chunk_sectors = l->new_chunk_sectors;
277 }
278 
279 /* raid10 algorithms (i.e. formats) */
280 #define	ALGORITHM_RAID10_DEFAULT	0
281 #define	ALGORITHM_RAID10_NEAR		1
282 #define	ALGORITHM_RAID10_OFFSET		2
283 #define	ALGORITHM_RAID10_FAR		3
284 
285 /* Supported raid types and properties. */
286 static struct raid_type {
287 	const char *name;		/* RAID algorithm. */
288 	const char *descr;		/* Descriptor text for logging. */
289 	const unsigned int parity_devs;	/* # of parity devices. */
290 	const unsigned int minimal_devs;/* minimal # of devices in set. */
291 	const unsigned int level;	/* RAID level. */
292 	const unsigned int algorithm;	/* RAID algorithm. */
293 } raid_types[] = {
294 	{"raid0",	  "raid0 (striping)",			    0, 2, 0,  0 /* NONE */},
295 	{"raid1",	  "raid1 (mirroring)",			    0, 2, 1,  0 /* NONE */},
296 	{"raid10_far",	  "raid10 far (striped mirrors)",	    0, 2, 10, ALGORITHM_RAID10_FAR},
297 	{"raid10_offset", "raid10 offset (striped mirrors)",	    0, 2, 10, ALGORITHM_RAID10_OFFSET},
298 	{"raid10_near",	  "raid10 near (striped mirrors)",	    0, 2, 10, ALGORITHM_RAID10_NEAR},
299 	{"raid10",	  "raid10 (striped mirrors)",		    0, 2, 10, ALGORITHM_RAID10_DEFAULT},
300 	{"raid4",	  "raid4 (dedicated first parity disk)",    1, 2, 5,  ALGORITHM_PARITY_0}, /* raid4 layout = raid5_0 */
301 	{"raid5_n",	  "raid5 (dedicated last parity disk)",	    1, 2, 5,  ALGORITHM_PARITY_N},
302 	{"raid5_ls",	  "raid5 (left symmetric)",		    1, 2, 5,  ALGORITHM_LEFT_SYMMETRIC},
303 	{"raid5_rs",	  "raid5 (right symmetric)",		    1, 2, 5,  ALGORITHM_RIGHT_SYMMETRIC},
304 	{"raid5_la",	  "raid5 (left asymmetric)",		    1, 2, 5,  ALGORITHM_LEFT_ASYMMETRIC},
305 	{"raid5_ra",	  "raid5 (right asymmetric)",		    1, 2, 5,  ALGORITHM_RIGHT_ASYMMETRIC},
306 	{"raid6_zr",	  "raid6 (zero restart)",		    2, 4, 6,  ALGORITHM_ROTATING_ZERO_RESTART},
307 	{"raid6_nr",	  "raid6 (N restart)",			    2, 4, 6,  ALGORITHM_ROTATING_N_RESTART},
308 	{"raid6_nc",	  "raid6 (N continue)",			    2, 4, 6,  ALGORITHM_ROTATING_N_CONTINUE},
309 	{"raid6_n_6",	  "raid6 (dedicated parity/Q n/6)",	    2, 4, 6,  ALGORITHM_PARITY_N_6},
310 	{"raid6_ls_6",	  "raid6 (left symmetric dedicated Q 6)",   2, 4, 6,  ALGORITHM_LEFT_SYMMETRIC_6},
311 	{"raid6_rs_6",	  "raid6 (right symmetric dedicated Q 6)",  2, 4, 6,  ALGORITHM_RIGHT_SYMMETRIC_6},
312 	{"raid6_la_6",	  "raid6 (left asymmetric dedicated Q 6)",  2, 4, 6,  ALGORITHM_LEFT_ASYMMETRIC_6},
313 	{"raid6_ra_6",	  "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6,  ALGORITHM_RIGHT_ASYMMETRIC_6}
314 };
315 
316 /* True, if @v is in inclusive range [@min, @max] */
__within_range(long v,long min,long max)317 static bool __within_range(long v, long min, long max)
318 {
319 	return v >= min && v <= max;
320 }
321 
322 /* All table line arguments are defined here */
323 static struct arg_name_flag {
324 	const unsigned long flag;
325 	const char *name;
326 } __arg_name_flags[] = {
327 	{ CTR_FLAG_SYNC, "sync"},
328 	{ CTR_FLAG_NOSYNC, "nosync"},
329 	{ CTR_FLAG_REBUILD, "rebuild"},
330 	{ CTR_FLAG_DAEMON_SLEEP, "daemon_sleep"},
331 	{ CTR_FLAG_MIN_RECOVERY_RATE, "min_recovery_rate"},
332 	{ CTR_FLAG_MAX_RECOVERY_RATE, "max_recovery_rate"},
333 	{ CTR_FLAG_MAX_WRITE_BEHIND, "max_write_behind"},
334 	{ CTR_FLAG_WRITE_MOSTLY, "write_mostly"},
335 	{ CTR_FLAG_STRIPE_CACHE, "stripe_cache"},
336 	{ CTR_FLAG_REGION_SIZE, "region_size"},
337 	{ CTR_FLAG_RAID10_COPIES, "raid10_copies"},
338 	{ CTR_FLAG_RAID10_FORMAT, "raid10_format"},
339 	{ CTR_FLAG_DATA_OFFSET, "data_offset"},
340 	{ CTR_FLAG_DELTA_DISKS, "delta_disks"},
341 	{ CTR_FLAG_RAID10_USE_NEAR_SETS, "raid10_use_near_sets"},
342 	{ CTR_FLAG_JOURNAL_DEV, "journal_dev" },
343 	{ CTR_FLAG_JOURNAL_MODE, "journal_mode" },
344 };
345 
346 /* Return argument name string for given @flag */
dm_raid_arg_name_by_flag(const uint32_t flag)347 static const char *dm_raid_arg_name_by_flag(const uint32_t flag)
348 {
349 	if (hweight32(flag) == 1) {
350 		struct arg_name_flag *anf = __arg_name_flags + ARRAY_SIZE(__arg_name_flags);
351 
352 		while (anf-- > __arg_name_flags)
353 			if (flag & anf->flag)
354 				return anf->name;
355 
356 	} else
357 		DMERR("%s called with more than one flag!", __func__);
358 
359 	return NULL;
360 }
361 
362 /* Define correlation of raid456 journal cache modes and dm-raid target line parameters */
363 static struct {
364 	const int mode;
365 	const char *param;
366 } _raid456_journal_mode[] = {
367 	{ R5C_JOURNAL_MODE_WRITE_THROUGH, "writethrough" },
368 	{ R5C_JOURNAL_MODE_WRITE_BACK,    "writeback" }
369 };
370 
371 /* Return MD raid4/5/6 journal mode for dm @journal_mode one */
dm_raid_journal_mode_to_md(const char * mode)372 static int dm_raid_journal_mode_to_md(const char *mode)
373 {
374 	int m = ARRAY_SIZE(_raid456_journal_mode);
375 
376 	while (m--)
377 		if (!strcasecmp(mode, _raid456_journal_mode[m].param))
378 			return _raid456_journal_mode[m].mode;
379 
380 	return -EINVAL;
381 }
382 
383 /* Return dm-raid raid4/5/6 journal mode string for @mode */
md_journal_mode_to_dm_raid(const int mode)384 static const char *md_journal_mode_to_dm_raid(const int mode)
385 {
386 	int m = ARRAY_SIZE(_raid456_journal_mode);
387 
388 	while (m--)
389 		if (mode == _raid456_journal_mode[m].mode)
390 			return _raid456_journal_mode[m].param;
391 
392 	return "unknown";
393 }
394 
395 /*
396  * Bool helpers to test for various raid levels of a raid set.
397  * It's level as reported by the superblock rather than
398  * the requested raid_type passed to the constructor.
399  */
400 /* Return true, if raid set in @rs is raid0 */
rs_is_raid0(struct raid_set * rs)401 static bool rs_is_raid0(struct raid_set *rs)
402 {
403 	return !rs->md.level;
404 }
405 
406 /* Return true, if raid set in @rs is raid1 */
rs_is_raid1(struct raid_set * rs)407 static bool rs_is_raid1(struct raid_set *rs)
408 {
409 	return rs->md.level == 1;
410 }
411 
412 /* Return true, if raid set in @rs is raid10 */
rs_is_raid10(struct raid_set * rs)413 static bool rs_is_raid10(struct raid_set *rs)
414 {
415 	return rs->md.level == 10;
416 }
417 
418 /* Return true, if raid set in @rs is level 6 */
rs_is_raid6(struct raid_set * rs)419 static bool rs_is_raid6(struct raid_set *rs)
420 {
421 	return rs->md.level == 6;
422 }
423 
424 /* Return true, if raid set in @rs is level 4, 5 or 6 */
rs_is_raid456(struct raid_set * rs)425 static bool rs_is_raid456(struct raid_set *rs)
426 {
427 	return __within_range(rs->md.level, 4, 6);
428 }
429 
430 /* Return true, if raid set in @rs is reshapable */
431 static bool __is_raid10_far(int layout);
rs_is_reshapable(struct raid_set * rs)432 static bool rs_is_reshapable(struct raid_set *rs)
433 {
434 	return rs_is_raid456(rs) ||
435 	       (rs_is_raid10(rs) && !__is_raid10_far(rs->md.new_layout));
436 }
437 
438 /* Return true, if raid set in @rs is recovering */
rs_is_recovering(struct raid_set * rs)439 static bool rs_is_recovering(struct raid_set *rs)
440 {
441 	return rs->md.recovery_cp < rs->md.dev_sectors;
442 }
443 
444 /* Return true, if raid set in @rs is reshaping */
rs_is_reshaping(struct raid_set * rs)445 static bool rs_is_reshaping(struct raid_set *rs)
446 {
447 	return rs->md.reshape_position != MaxSector;
448 }
449 
450 /*
451  * bool helpers to test for various raid levels of a raid type @rt
452  */
453 
454 /* Return true, if raid type in @rt is raid0 */
rt_is_raid0(struct raid_type * rt)455 static bool rt_is_raid0(struct raid_type *rt)
456 {
457 	return !rt->level;
458 }
459 
460 /* Return true, if raid type in @rt is raid1 */
rt_is_raid1(struct raid_type * rt)461 static bool rt_is_raid1(struct raid_type *rt)
462 {
463 	return rt->level == 1;
464 }
465 
466 /* Return true, if raid type in @rt is raid10 */
rt_is_raid10(struct raid_type * rt)467 static bool rt_is_raid10(struct raid_type *rt)
468 {
469 	return rt->level == 10;
470 }
471 
472 /* Return true, if raid type in @rt is raid4/5 */
rt_is_raid45(struct raid_type * rt)473 static bool rt_is_raid45(struct raid_type *rt)
474 {
475 	return __within_range(rt->level, 4, 5);
476 }
477 
478 /* Return true, if raid type in @rt is raid6 */
rt_is_raid6(struct raid_type * rt)479 static bool rt_is_raid6(struct raid_type *rt)
480 {
481 	return rt->level == 6;
482 }
483 
484 /* Return true, if raid type in @rt is raid4/5/6 */
rt_is_raid456(struct raid_type * rt)485 static bool rt_is_raid456(struct raid_type *rt)
486 {
487 	return __within_range(rt->level, 4, 6);
488 }
489 /* END: raid level bools */
490 
491 /* Return valid ctr flags for the raid level of @rs */
__valid_flags(struct raid_set * rs)492 static unsigned long __valid_flags(struct raid_set *rs)
493 {
494 	if (rt_is_raid0(rs->raid_type))
495 		return RAID0_VALID_FLAGS;
496 	else if (rt_is_raid1(rs->raid_type))
497 		return RAID1_VALID_FLAGS;
498 	else if (rt_is_raid10(rs->raid_type))
499 		return RAID10_VALID_FLAGS;
500 	else if (rt_is_raid45(rs->raid_type))
501 		return RAID45_VALID_FLAGS;
502 	else if (rt_is_raid6(rs->raid_type))
503 		return RAID6_VALID_FLAGS;
504 
505 	return 0;
506 }
507 
508 /*
509  * Check for valid flags set on @rs
510  *
511  * Has to be called after parsing of the ctr flags!
512  */
rs_check_for_valid_flags(struct raid_set * rs)513 static int rs_check_for_valid_flags(struct raid_set *rs)
514 {
515 	if (rs->ctr_flags & ~__valid_flags(rs)) {
516 		rs->ti->error = "Invalid flags combination";
517 		return -EINVAL;
518 	}
519 
520 	return 0;
521 }
522 
523 /* MD raid10 bit definitions and helpers */
524 #define RAID10_OFFSET			(1 << 16) /* stripes with data copies area adjacent on devices */
525 #define RAID10_BROCKEN_USE_FAR_SETS	(1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
526 #define RAID10_USE_FAR_SETS		(1 << 18) /* Use sets instead of whole stripe rotation */
527 #define RAID10_FAR_COPIES_SHIFT		8	  /* raid10 # far copies shift (2nd byte of layout) */
528 
529 /* Return md raid10 near copies for @layout */
__raid10_near_copies(int layout)530 static unsigned int __raid10_near_copies(int layout)
531 {
532 	return layout & 0xFF;
533 }
534 
535 /* Return md raid10 far copies for @layout */
__raid10_far_copies(int layout)536 static unsigned int __raid10_far_copies(int layout)
537 {
538 	return __raid10_near_copies(layout >> RAID10_FAR_COPIES_SHIFT);
539 }
540 
541 /* Return true if md raid10 offset for @layout */
__is_raid10_offset(int layout)542 static bool __is_raid10_offset(int layout)
543 {
544 	return !!(layout & RAID10_OFFSET);
545 }
546 
547 /* Return true if md raid10 near for @layout */
__is_raid10_near(int layout)548 static bool __is_raid10_near(int layout)
549 {
550 	return !__is_raid10_offset(layout) && __raid10_near_copies(layout) > 1;
551 }
552 
553 /* Return true if md raid10 far for @layout */
__is_raid10_far(int layout)554 static bool __is_raid10_far(int layout)
555 {
556 	return !__is_raid10_offset(layout) && __raid10_far_copies(layout) > 1;
557 }
558 
559 /* Return md raid10 layout string for @layout */
raid10_md_layout_to_format(int layout)560 static const char *raid10_md_layout_to_format(int layout)
561 {
562 	/*
563 	 * Bit 16 stands for "offset"
564 	 * (i.e. adjacent stripes hold copies)
565 	 *
566 	 * Refer to MD's raid10.c for details
567 	 */
568 	if (__is_raid10_offset(layout))
569 		return "offset";
570 
571 	if (__raid10_near_copies(layout) > 1)
572 		return "near";
573 
574 	if (__raid10_far_copies(layout) > 1)
575 		return "far";
576 
577 	return "unknown";
578 }
579 
580 /* Return md raid10 algorithm for @name */
raid10_name_to_format(const char * name)581 static int raid10_name_to_format(const char *name)
582 {
583 	if (!strcasecmp(name, "near"))
584 		return ALGORITHM_RAID10_NEAR;
585 	else if (!strcasecmp(name, "offset"))
586 		return ALGORITHM_RAID10_OFFSET;
587 	else if (!strcasecmp(name, "far"))
588 		return ALGORITHM_RAID10_FAR;
589 
590 	return -EINVAL;
591 }
592 
593 /* Return md raid10 copies for @layout */
raid10_md_layout_to_copies(int layout)594 static unsigned int raid10_md_layout_to_copies(int layout)
595 {
596 	return max(__raid10_near_copies(layout), __raid10_far_copies(layout));
597 }
598 
599 /* Return md raid10 format id for @format string */
raid10_format_to_md_layout(struct raid_set * rs,unsigned int algorithm,unsigned int copies)600 static int raid10_format_to_md_layout(struct raid_set *rs,
601 				      unsigned int algorithm,
602 				      unsigned int copies)
603 {
604 	unsigned int n = 1, f = 1, r = 0;
605 
606 	/*
607 	 * MD resilienece flaw:
608 	 *
609 	 * enabling use_far_sets for far/offset formats causes copies
610 	 * to be colocated on the same devs together with their origins!
611 	 *
612 	 * -> disable it for now in the definition above
613 	 */
614 	if (algorithm == ALGORITHM_RAID10_DEFAULT ||
615 	    algorithm == ALGORITHM_RAID10_NEAR)
616 		n = copies;
617 
618 	else if (algorithm == ALGORITHM_RAID10_OFFSET) {
619 		f = copies;
620 		r = RAID10_OFFSET;
621 		if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
622 			r |= RAID10_USE_FAR_SETS;
623 
624 	} else if (algorithm == ALGORITHM_RAID10_FAR) {
625 		f = copies;
626 		if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
627 			r |= RAID10_USE_FAR_SETS;
628 
629 	} else
630 		return -EINVAL;
631 
632 	return r | (f << RAID10_FAR_COPIES_SHIFT) | n;
633 }
634 /* END: MD raid10 bit definitions and helpers */
635 
636 /* Check for any of the raid10 algorithms */
__got_raid10(struct raid_type * rtp,const int layout)637 static bool __got_raid10(struct raid_type *rtp, const int layout)
638 {
639 	if (rtp->level == 10) {
640 		switch (rtp->algorithm) {
641 		case ALGORITHM_RAID10_DEFAULT:
642 		case ALGORITHM_RAID10_NEAR:
643 			return __is_raid10_near(layout);
644 		case ALGORITHM_RAID10_OFFSET:
645 			return __is_raid10_offset(layout);
646 		case ALGORITHM_RAID10_FAR:
647 			return __is_raid10_far(layout);
648 		default:
649 			break;
650 		}
651 	}
652 
653 	return false;
654 }
655 
656 /* Return raid_type for @name */
get_raid_type(const char * name)657 static struct raid_type *get_raid_type(const char *name)
658 {
659 	struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
660 
661 	while (rtp-- > raid_types)
662 		if (!strcasecmp(rtp->name, name))
663 			return rtp;
664 
665 	return NULL;
666 }
667 
668 /* Return raid_type for @name based derived from @level and @layout */
get_raid_type_by_ll(const int level,const int layout)669 static struct raid_type *get_raid_type_by_ll(const int level, const int layout)
670 {
671 	struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
672 
673 	while (rtp-- > raid_types) {
674 		/* RAID10 special checks based on @layout flags/properties */
675 		if (rtp->level == level &&
676 		    (__got_raid10(rtp, layout) || rtp->algorithm == layout))
677 			return rtp;
678 	}
679 
680 	return NULL;
681 }
682 
683 /* Adjust rdev sectors */
rs_set_rdev_sectors(struct raid_set * rs)684 static void rs_set_rdev_sectors(struct raid_set *rs)
685 {
686 	struct mddev *mddev = &rs->md;
687 	struct md_rdev *rdev;
688 
689 	/*
690 	 * raid10 sets rdev->sector to the device size, which
691 	 * is unintended in case of out-of-place reshaping
692 	 */
693 	rdev_for_each(rdev, mddev)
694 		if (!test_bit(Journal, &rdev->flags))
695 			rdev->sectors = mddev->dev_sectors;
696 }
697 
698 /*
699  * Change bdev capacity of @rs in case of a disk add/remove reshape
700  */
rs_set_capacity(struct raid_set * rs)701 static void rs_set_capacity(struct raid_set *rs)
702 {
703 	struct gendisk *gendisk = dm_disk(dm_table_get_md(rs->ti->table));
704 
705 	set_capacity_and_notify(gendisk, rs->md.array_sectors);
706 }
707 
708 /*
709  * Set the mddev properties in @rs to the current
710  * ones retrieved from the freshest superblock
711  */
rs_set_cur(struct raid_set * rs)712 static void rs_set_cur(struct raid_set *rs)
713 {
714 	struct mddev *mddev = &rs->md;
715 
716 	mddev->new_level = mddev->level;
717 	mddev->new_layout = mddev->layout;
718 	mddev->new_chunk_sectors = mddev->chunk_sectors;
719 }
720 
721 /*
722  * Set the mddev properties in @rs to the new
723  * ones requested by the ctr
724  */
rs_set_new(struct raid_set * rs)725 static void rs_set_new(struct raid_set *rs)
726 {
727 	struct mddev *mddev = &rs->md;
728 
729 	mddev->level = mddev->new_level;
730 	mddev->layout = mddev->new_layout;
731 	mddev->chunk_sectors = mddev->new_chunk_sectors;
732 	mddev->raid_disks = rs->raid_disks;
733 	mddev->delta_disks = 0;
734 }
735 
raid_set_alloc(struct dm_target * ti,struct raid_type * raid_type,unsigned int raid_devs)736 static struct raid_set *raid_set_alloc(struct dm_target *ti, struct raid_type *raid_type,
737 				       unsigned int raid_devs)
738 {
739 	unsigned int i;
740 	struct raid_set *rs;
741 
742 	if (raid_devs <= raid_type->parity_devs) {
743 		ti->error = "Insufficient number of devices";
744 		return ERR_PTR(-EINVAL);
745 	}
746 
747 	rs = kzalloc(struct_size(rs, dev, raid_devs), GFP_KERNEL);
748 	if (!rs) {
749 		ti->error = "Cannot allocate raid context";
750 		return ERR_PTR(-ENOMEM);
751 	}
752 
753 	if (mddev_init(&rs->md)) {
754 		kfree(rs);
755 		ti->error = "Cannot initialize raid context";
756 		return ERR_PTR(-ENOMEM);
757 	}
758 
759 	rs->raid_disks = raid_devs;
760 	rs->delta_disks = 0;
761 
762 	rs->ti = ti;
763 	rs->raid_type = raid_type;
764 	rs->stripe_cache_entries = 256;
765 	rs->md.raid_disks = raid_devs;
766 	rs->md.level = raid_type->level;
767 	rs->md.new_level = rs->md.level;
768 	rs->md.layout = raid_type->algorithm;
769 	rs->md.new_layout = rs->md.layout;
770 	rs->md.delta_disks = 0;
771 	rs->md.recovery_cp = MaxSector;
772 
773 	for (i = 0; i < raid_devs; i++)
774 		md_rdev_init(&rs->dev[i].rdev);
775 
776 	/*
777 	 * Remaining items to be initialized by further RAID params:
778 	 *  rs->md.persistent
779 	 *  rs->md.external
780 	 *  rs->md.chunk_sectors
781 	 *  rs->md.new_chunk_sectors
782 	 *  rs->md.dev_sectors
783 	 */
784 
785 	return rs;
786 }
787 
788 /* Free all @rs allocations */
raid_set_free(struct raid_set * rs)789 static void raid_set_free(struct raid_set *rs)
790 {
791 	int i;
792 
793 	if (rs->journal_dev.dev) {
794 		md_rdev_clear(&rs->journal_dev.rdev);
795 		dm_put_device(rs->ti, rs->journal_dev.dev);
796 	}
797 
798 	for (i = 0; i < rs->raid_disks; i++) {
799 		if (rs->dev[i].meta_dev)
800 			dm_put_device(rs->ti, rs->dev[i].meta_dev);
801 		md_rdev_clear(&rs->dev[i].rdev);
802 		if (rs->dev[i].data_dev)
803 			dm_put_device(rs->ti, rs->dev[i].data_dev);
804 	}
805 
806 	mddev_destroy(&rs->md);
807 	kfree(rs);
808 }
809 
810 /*
811  * For every device we have two words
812  *  <meta_dev>: meta device name or '-' if missing
813  *  <data_dev>: data device name or '-' if missing
814  *
815  * The following are permitted:
816  *    - -
817  *    - <data_dev>
818  *    <meta_dev> <data_dev>
819  *
820  * The following is not allowed:
821  *    <meta_dev> -
822  *
823  * This code parses those words.  If there is a failure,
824  * the caller must use raid_set_free() to unwind the operations.
825  */
parse_dev_params(struct raid_set * rs,struct dm_arg_set * as)826 static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
827 {
828 	int i;
829 	int rebuild = 0;
830 	int metadata_available = 0;
831 	int r = 0;
832 	const char *arg;
833 
834 	/* Put off the number of raid devices argument to get to dev pairs */
835 	arg = dm_shift_arg(as);
836 	if (!arg)
837 		return -EINVAL;
838 
839 	for (i = 0; i < rs->raid_disks; i++) {
840 		rs->dev[i].rdev.raid_disk = i;
841 
842 		rs->dev[i].meta_dev = NULL;
843 		rs->dev[i].data_dev = NULL;
844 
845 		/*
846 		 * There are no offsets initially.
847 		 * Out of place reshape will set them accordingly.
848 		 */
849 		rs->dev[i].rdev.data_offset = 0;
850 		rs->dev[i].rdev.new_data_offset = 0;
851 		rs->dev[i].rdev.mddev = &rs->md;
852 
853 		arg = dm_shift_arg(as);
854 		if (!arg)
855 			return -EINVAL;
856 
857 		if (strcmp(arg, "-")) {
858 			r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
859 					  &rs->dev[i].meta_dev);
860 			if (r) {
861 				rs->ti->error = "RAID metadata device lookup failure";
862 				return r;
863 			}
864 
865 			rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
866 			if (!rs->dev[i].rdev.sb_page) {
867 				rs->ti->error = "Failed to allocate superblock page";
868 				return -ENOMEM;
869 			}
870 		}
871 
872 		arg = dm_shift_arg(as);
873 		if (!arg)
874 			return -EINVAL;
875 
876 		if (!strcmp(arg, "-")) {
877 			if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
878 			    (!rs->dev[i].rdev.recovery_offset)) {
879 				rs->ti->error = "Drive designated for rebuild not specified";
880 				return -EINVAL;
881 			}
882 
883 			if (rs->dev[i].meta_dev) {
884 				rs->ti->error = "No data device supplied with metadata device";
885 				return -EINVAL;
886 			}
887 
888 			continue;
889 		}
890 
891 		r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
892 				  &rs->dev[i].data_dev);
893 		if (r) {
894 			rs->ti->error = "RAID device lookup failure";
895 			return r;
896 		}
897 
898 		if (rs->dev[i].meta_dev) {
899 			metadata_available = 1;
900 			rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
901 		}
902 		rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
903 		list_add_tail(&rs->dev[i].rdev.same_set, &rs->md.disks);
904 		if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
905 			rebuild++;
906 	}
907 
908 	if (rs->journal_dev.dev)
909 		list_add_tail(&rs->journal_dev.rdev.same_set, &rs->md.disks);
910 
911 	if (metadata_available) {
912 		rs->md.external = 0;
913 		rs->md.persistent = 1;
914 		rs->md.major_version = 2;
915 	} else if (rebuild && !rs->md.recovery_cp) {
916 		/*
917 		 * Without metadata, we will not be able to tell if the array
918 		 * is in-sync or not - we must assume it is not.  Therefore,
919 		 * it is impossible to rebuild a drive.
920 		 *
921 		 * Even if there is metadata, the on-disk information may
922 		 * indicate that the array is not in-sync and it will then
923 		 * fail at that time.
924 		 *
925 		 * User could specify 'nosync' option if desperate.
926 		 */
927 		rs->ti->error = "Unable to rebuild drive while array is not in-sync";
928 		return -EINVAL;
929 	}
930 
931 	return 0;
932 }
933 
934 /*
935  * validate_region_size
936  * @rs
937  * @region_size:  region size in sectors.  If 0, pick a size (4MiB default).
938  *
939  * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
940  * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
941  *
942  * Returns: 0 on success, -EINVAL on failure.
943  */
validate_region_size(struct raid_set * rs,unsigned long region_size)944 static int validate_region_size(struct raid_set *rs, unsigned long region_size)
945 {
946 	unsigned long min_region_size = rs->ti->len / (1 << 21);
947 
948 	if (rs_is_raid0(rs))
949 		return 0;
950 
951 	if (!region_size) {
952 		/*
953 		 * Choose a reasonable default.	 All figures in sectors.
954 		 */
955 		if (min_region_size > (1 << 13)) {
956 			/* If not a power of 2, make it the next power of 2 */
957 			region_size = roundup_pow_of_two(min_region_size);
958 			DMINFO("Choosing default region size of %lu sectors",
959 			       region_size);
960 		} else {
961 			DMINFO("Choosing default region size of 4MiB");
962 			region_size = 1 << 13; /* sectors */
963 		}
964 	} else {
965 		/*
966 		 * Validate user-supplied value.
967 		 */
968 		if (region_size > rs->ti->len) {
969 			rs->ti->error = "Supplied region size is too large";
970 			return -EINVAL;
971 		}
972 
973 		if (region_size < min_region_size) {
974 			DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
975 			      region_size, min_region_size);
976 			rs->ti->error = "Supplied region size is too small";
977 			return -EINVAL;
978 		}
979 
980 		if (!is_power_of_2(region_size)) {
981 			rs->ti->error = "Region size is not a power of 2";
982 			return -EINVAL;
983 		}
984 
985 		if (region_size < rs->md.chunk_sectors) {
986 			rs->ti->error = "Region size is smaller than the chunk size";
987 			return -EINVAL;
988 		}
989 	}
990 
991 	/*
992 	 * Convert sectors to bytes.
993 	 */
994 	rs->md.bitmap_info.chunksize = to_bytes(region_size);
995 
996 	return 0;
997 }
998 
999 /*
1000  * validate_raid_redundancy
1001  * @rs
1002  *
1003  * Determine if there are enough devices in the array that haven't
1004  * failed (or are being rebuilt) to form a usable array.
1005  *
1006  * Returns: 0 on success, -EINVAL on failure.
1007  */
validate_raid_redundancy(struct raid_set * rs)1008 static int validate_raid_redundancy(struct raid_set *rs)
1009 {
1010 	unsigned int i, rebuild_cnt = 0;
1011 	unsigned int rebuilds_per_group = 0, copies, raid_disks;
1012 	unsigned int group_size, last_group_start;
1013 
1014 	for (i = 0; i < rs->raid_disks; i++)
1015 		if (!test_bit(FirstUse, &rs->dev[i].rdev.flags) &&
1016 		    ((!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
1017 		      !rs->dev[i].rdev.sb_page)))
1018 			rebuild_cnt++;
1019 
1020 	switch (rs->md.level) {
1021 	case 0:
1022 		break;
1023 	case 1:
1024 		if (rebuild_cnt >= rs->md.raid_disks)
1025 			goto too_many;
1026 		break;
1027 	case 4:
1028 	case 5:
1029 	case 6:
1030 		if (rebuild_cnt > rs->raid_type->parity_devs)
1031 			goto too_many;
1032 		break;
1033 	case 10:
1034 		copies = raid10_md_layout_to_copies(rs->md.new_layout);
1035 		if (copies < 2) {
1036 			DMERR("Bogus raid10 data copies < 2!");
1037 			return -EINVAL;
1038 		}
1039 
1040 		if (rebuild_cnt < copies)
1041 			break;
1042 
1043 		/*
1044 		 * It is possible to have a higher rebuild count for RAID10,
1045 		 * as long as the failed devices occur in different mirror
1046 		 * groups (i.e. different stripes).
1047 		 *
1048 		 * When checking "near" format, make sure no adjacent devices
1049 		 * have failed beyond what can be handled.  In addition to the
1050 		 * simple case where the number of devices is a multiple of the
1051 		 * number of copies, we must also handle cases where the number
1052 		 * of devices is not a multiple of the number of copies.
1053 		 * E.g.	   dev1 dev2 dev3 dev4 dev5
1054 		 *	    A	 A    B	   B	C
1055 		 *	    C	 D    D	   E	E
1056 		 */
1057 		raid_disks = min(rs->raid_disks, rs->md.raid_disks);
1058 		if (__is_raid10_near(rs->md.new_layout)) {
1059 			for (i = 0; i < raid_disks; i++) {
1060 				if (!(i % copies))
1061 					rebuilds_per_group = 0;
1062 				if ((!rs->dev[i].rdev.sb_page ||
1063 				    !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1064 				    (++rebuilds_per_group >= copies))
1065 					goto too_many;
1066 			}
1067 			break;
1068 		}
1069 
1070 		/*
1071 		 * When checking "far" and "offset" formats, we need to ensure
1072 		 * that the device that holds its copy is not also dead or
1073 		 * being rebuilt.  (Note that "far" and "offset" formats only
1074 		 * support two copies right now.  These formats also only ever
1075 		 * use the 'use_far_sets' variant.)
1076 		 *
1077 		 * This check is somewhat complicated by the need to account
1078 		 * for arrays that are not a multiple of (far) copies.	This
1079 		 * results in the need to treat the last (potentially larger)
1080 		 * set differently.
1081 		 */
1082 		group_size = (raid_disks / copies);
1083 		last_group_start = (raid_disks / group_size) - 1;
1084 		last_group_start *= group_size;
1085 		for (i = 0; i < raid_disks; i++) {
1086 			if (!(i % copies) && !(i > last_group_start))
1087 				rebuilds_per_group = 0;
1088 			if ((!rs->dev[i].rdev.sb_page ||
1089 			     !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1090 			    (++rebuilds_per_group >= copies))
1091 				goto too_many;
1092 		}
1093 		break;
1094 	default:
1095 		if (rebuild_cnt)
1096 			return -EINVAL;
1097 	}
1098 
1099 	return 0;
1100 
1101 too_many:
1102 	return -EINVAL;
1103 }
1104 
1105 /*
1106  * Possible arguments are...
1107  *	<chunk_size> [optional_args]
1108  *
1109  * Argument definitions
1110  *    <chunk_size>			The number of sectors per disk that
1111  *					will form the "stripe"
1112  *    [[no]sync]			Force or prevent recovery of the
1113  *					entire array
1114  *    [rebuild <idx>]			Rebuild the drive indicated by the index
1115  *    [daemon_sleep <ms>]		Time between bitmap daemon work to
1116  *					clear bits
1117  *    [min_recovery_rate <kB/sec/disk>]	Throttle RAID initialization
1118  *    [max_recovery_rate <kB/sec/disk>]	Throttle RAID initialization
1119  *    [write_mostly <idx>]		Indicate a write mostly drive via index
1120  *    [max_write_behind <sectors>]	See '-write-behind=' (man mdadm)
1121  *    [stripe_cache <sectors>]		Stripe cache size for higher RAIDs
1122  *    [region_size <sectors>]		Defines granularity of bitmap
1123  *    [journal_dev <dev>]		raid4/5/6 journaling deviice
1124  *					(i.e. write hole closing log)
1125  *
1126  * RAID10-only options:
1127  *    [raid10_copies <# copies>]	Number of copies.  (Default: 2)
1128  *    [raid10_format <near|far|offset>] Layout algorithm.  (Default: near)
1129  */
parse_raid_params(struct raid_set * rs,struct dm_arg_set * as,unsigned int num_raid_params)1130 static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
1131 			     unsigned int num_raid_params)
1132 {
1133 	int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
1134 	unsigned int raid10_copies = 2;
1135 	unsigned int i, write_mostly = 0;
1136 	unsigned int region_size = 0;
1137 	sector_t max_io_len;
1138 	const char *arg, *key;
1139 	struct raid_dev *rd;
1140 	struct raid_type *rt = rs->raid_type;
1141 
1142 	arg = dm_shift_arg(as);
1143 	num_raid_params--; /* Account for chunk_size argument */
1144 
1145 	if (kstrtoint(arg, 10, &value) < 0) {
1146 		rs->ti->error = "Bad numerical argument given for chunk_size";
1147 		return -EINVAL;
1148 	}
1149 
1150 	/*
1151 	 * First, parse the in-order required arguments
1152 	 * "chunk_size" is the only argument of this type.
1153 	 */
1154 	if (rt_is_raid1(rt)) {
1155 		if (value)
1156 			DMERR("Ignoring chunk size parameter for RAID 1");
1157 		value = 0;
1158 	} else if (!is_power_of_2(value)) {
1159 		rs->ti->error = "Chunk size must be a power of 2";
1160 		return -EINVAL;
1161 	} else if (value < 8) {
1162 		rs->ti->error = "Chunk size value is too small";
1163 		return -EINVAL;
1164 	}
1165 
1166 	rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
1167 
1168 	/*
1169 	 * We set each individual device as In_sync with a completed
1170 	 * 'recovery_offset'.  If there has been a device failure or
1171 	 * replacement then one of the following cases applies:
1172 	 *
1173 	 *   1) User specifies 'rebuild'.
1174 	 *	- Device is reset when param is read.
1175 	 *   2) A new device is supplied.
1176 	 *	- No matching superblock found, resets device.
1177 	 *   3) Device failure was transient and returns on reload.
1178 	 *	- Failure noticed, resets device for bitmap replay.
1179 	 *   4) Device hadn't completed recovery after previous failure.
1180 	 *	- Superblock is read and overrides recovery_offset.
1181 	 *
1182 	 * What is found in the superblocks of the devices is always
1183 	 * authoritative, unless 'rebuild' or '[no]sync' was specified.
1184 	 */
1185 	for (i = 0; i < rs->raid_disks; i++) {
1186 		set_bit(In_sync, &rs->dev[i].rdev.flags);
1187 		rs->dev[i].rdev.recovery_offset = MaxSector;
1188 	}
1189 
1190 	/*
1191 	 * Second, parse the unordered optional arguments
1192 	 */
1193 	for (i = 0; i < num_raid_params; i++) {
1194 		key = dm_shift_arg(as);
1195 		if (!key) {
1196 			rs->ti->error = "Not enough raid parameters given";
1197 			return -EINVAL;
1198 		}
1199 
1200 		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
1201 			if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1202 				rs->ti->error = "Only one 'nosync' argument allowed";
1203 				return -EINVAL;
1204 			}
1205 			continue;
1206 		}
1207 		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
1208 			if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
1209 				rs->ti->error = "Only one 'sync' argument allowed";
1210 				return -EINVAL;
1211 			}
1212 			continue;
1213 		}
1214 		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
1215 			if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1216 				rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1217 				return -EINVAL;
1218 			}
1219 			continue;
1220 		}
1221 
1222 		arg = dm_shift_arg(as);
1223 		i++; /* Account for the argument pairs */
1224 		if (!arg) {
1225 			rs->ti->error = "Wrong number of raid parameters given";
1226 			return -EINVAL;
1227 		}
1228 
1229 		/*
1230 		 * Parameters that take a string value are checked here.
1231 		 */
1232 		/* "raid10_format {near|offset|far} */
1233 		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
1234 			if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
1235 				rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1236 				return -EINVAL;
1237 			}
1238 			if (!rt_is_raid10(rt)) {
1239 				rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1240 				return -EINVAL;
1241 			}
1242 			raid10_format = raid10_name_to_format(arg);
1243 			if (raid10_format < 0) {
1244 				rs->ti->error = "Invalid 'raid10_format' value given";
1245 				return raid10_format;
1246 			}
1247 			continue;
1248 		}
1249 
1250 		/* "journal_dev <dev>" */
1251 		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV))) {
1252 			int r;
1253 			struct md_rdev *jdev;
1254 
1255 			if (test_and_set_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1256 				rs->ti->error = "Only one raid4/5/6 set journaling device allowed";
1257 				return -EINVAL;
1258 			}
1259 			if (!rt_is_raid456(rt)) {
1260 				rs->ti->error = "'journal_dev' is an invalid parameter for this RAID type";
1261 				return -EINVAL;
1262 			}
1263 			r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
1264 					  &rs->journal_dev.dev);
1265 			if (r) {
1266 				rs->ti->error = "raid4/5/6 journal device lookup failure";
1267 				return r;
1268 			}
1269 			jdev = &rs->journal_dev.rdev;
1270 			md_rdev_init(jdev);
1271 			jdev->mddev = &rs->md;
1272 			jdev->bdev = rs->journal_dev.dev->bdev;
1273 			jdev->sectors = bdev_nr_sectors(jdev->bdev);
1274 			if (jdev->sectors < MIN_RAID456_JOURNAL_SPACE) {
1275 				rs->ti->error = "No space for raid4/5/6 journal";
1276 				return -ENOSPC;
1277 			}
1278 			rs->journal_dev.mode = R5C_JOURNAL_MODE_WRITE_THROUGH;
1279 			set_bit(Journal, &jdev->flags);
1280 			continue;
1281 		}
1282 
1283 		/* "journal_mode <mode>" ("journal_dev" mandatory!) */
1284 		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE))) {
1285 			int r;
1286 
1287 			if (!test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1288 				rs->ti->error = "raid4/5/6 'journal_mode' is invalid without 'journal_dev'";
1289 				return -EINVAL;
1290 			}
1291 			if (test_and_set_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
1292 				rs->ti->error = "Only one raid4/5/6 'journal_mode' argument allowed";
1293 				return -EINVAL;
1294 			}
1295 			r = dm_raid_journal_mode_to_md(arg);
1296 			if (r < 0) {
1297 				rs->ti->error = "Invalid 'journal_mode' argument";
1298 				return r;
1299 			}
1300 			rs->journal_dev.mode = r;
1301 			continue;
1302 		}
1303 
1304 		/*
1305 		 * Parameters with number values from here on.
1306 		 */
1307 		if (kstrtoint(arg, 10, &value) < 0) {
1308 			rs->ti->error = "Bad numerical argument given in raid params";
1309 			return -EINVAL;
1310 		}
1311 
1312 		if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
1313 			/*
1314 			 * "rebuild" is being passed in by userspace to provide
1315 			 * indexes of replaced devices and to set up additional
1316 			 * devices on raid level takeover.
1317 			 */
1318 			if (!__within_range(value, 0, rs->raid_disks - 1)) {
1319 				rs->ti->error = "Invalid rebuild index given";
1320 				return -EINVAL;
1321 			}
1322 
1323 			if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1324 				rs->ti->error = "rebuild for this index already given";
1325 				return -EINVAL;
1326 			}
1327 
1328 			rd = rs->dev + value;
1329 			clear_bit(In_sync, &rd->rdev.flags);
1330 			clear_bit(Faulty, &rd->rdev.flags);
1331 			rd->rdev.recovery_offset = 0;
1332 			set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
1333 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
1334 			if (!rt_is_raid1(rt)) {
1335 				rs->ti->error = "write_mostly option is only valid for RAID1";
1336 				return -EINVAL;
1337 			}
1338 
1339 			if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
1340 				rs->ti->error = "Invalid write_mostly index given";
1341 				return -EINVAL;
1342 			}
1343 
1344 			write_mostly++;
1345 			set_bit(WriteMostly, &rs->dev[value].rdev.flags);
1346 			set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
1347 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
1348 			if (!rt_is_raid1(rt)) {
1349 				rs->ti->error = "max_write_behind option is only valid for RAID1";
1350 				return -EINVAL;
1351 			}
1352 
1353 			if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
1354 				rs->ti->error = "Only one max_write_behind argument pair allowed";
1355 				return -EINVAL;
1356 			}
1357 
1358 			/*
1359 			 * In device-mapper, we specify things in sectors, but
1360 			 * MD records this value in kB
1361 			 */
1362 			if (value < 0 || value / 2 > COUNTER_MAX) {
1363 				rs->ti->error = "Max write-behind limit out of range";
1364 				return -EINVAL;
1365 			}
1366 
1367 			rs->md.bitmap_info.max_write_behind = value / 2;
1368 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
1369 			if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
1370 				rs->ti->error = "Only one daemon_sleep argument pair allowed";
1371 				return -EINVAL;
1372 			}
1373 			if (value < 0) {
1374 				rs->ti->error = "daemon sleep period out of range";
1375 				return -EINVAL;
1376 			}
1377 			rs->md.bitmap_info.daemon_sleep = value;
1378 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
1379 			/* Userspace passes new data_offset after having extended the data image LV */
1380 			if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
1381 				rs->ti->error = "Only one data_offset argument pair allowed";
1382 				return -EINVAL;
1383 			}
1384 			/* Ensure sensible data offset */
1385 			if (value < 0 ||
1386 			    (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
1387 				rs->ti->error = "Bogus data_offset value";
1388 				return -EINVAL;
1389 			}
1390 			rs->data_offset = value;
1391 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
1392 			/* Define the +/-# of disks to add to/remove from the given raid set */
1393 			if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1394 				rs->ti->error = "Only one delta_disks argument pair allowed";
1395 				return -EINVAL;
1396 			}
1397 			/* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1398 			if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
1399 				rs->ti->error = "Too many delta_disk requested";
1400 				return -EINVAL;
1401 			}
1402 
1403 			rs->delta_disks = value;
1404 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
1405 			if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
1406 				rs->ti->error = "Only one stripe_cache argument pair allowed";
1407 				return -EINVAL;
1408 			}
1409 
1410 			if (!rt_is_raid456(rt)) {
1411 				rs->ti->error = "Inappropriate argument: stripe_cache";
1412 				return -EINVAL;
1413 			}
1414 
1415 			if (value < 0) {
1416 				rs->ti->error = "Bogus stripe cache entries value";
1417 				return -EINVAL;
1418 			}
1419 			rs->stripe_cache_entries = value;
1420 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
1421 			if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1422 				rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1423 				return -EINVAL;
1424 			}
1425 
1426 			if (value < 0) {
1427 				rs->ti->error = "min_recovery_rate out of range";
1428 				return -EINVAL;
1429 			}
1430 			rs->md.sync_speed_min = value;
1431 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
1432 			if (test_and_set_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags)) {
1433 				rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1434 				return -EINVAL;
1435 			}
1436 
1437 			if (value < 0) {
1438 				rs->ti->error = "max_recovery_rate out of range";
1439 				return -EINVAL;
1440 			}
1441 			rs->md.sync_speed_max = value;
1442 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
1443 			if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
1444 				rs->ti->error = "Only one region_size argument pair allowed";
1445 				return -EINVAL;
1446 			}
1447 
1448 			region_size = value;
1449 			rs->requested_bitmap_chunk_sectors = value;
1450 		} else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
1451 			if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
1452 				rs->ti->error = "Only one raid10_copies argument pair allowed";
1453 				return -EINVAL;
1454 			}
1455 
1456 			if (!__within_range(value, 2, rs->md.raid_disks)) {
1457 				rs->ti->error = "Bad value for 'raid10_copies'";
1458 				return -EINVAL;
1459 			}
1460 
1461 			raid10_copies = value;
1462 		} else {
1463 			DMERR("Unable to parse RAID parameter: %s", key);
1464 			rs->ti->error = "Unable to parse RAID parameter";
1465 			return -EINVAL;
1466 		}
1467 	}
1468 
1469 	if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1470 	    test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1471 		rs->ti->error = "sync and nosync are mutually exclusive";
1472 		return -EINVAL;
1473 	}
1474 
1475 	if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) &&
1476 	    (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ||
1477 	     test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))) {
1478 		rs->ti->error = "sync/nosync and rebuild are mutually exclusive";
1479 		return -EINVAL;
1480 	}
1481 
1482 	if (write_mostly >= rs->md.raid_disks) {
1483 		rs->ti->error = "Can't set all raid1 devices to write_mostly";
1484 		return -EINVAL;
1485 	}
1486 
1487 	if (rs->md.sync_speed_max &&
1488 	    rs->md.sync_speed_min > rs->md.sync_speed_max) {
1489 		rs->ti->error = "Bogus recovery rates";
1490 		return -EINVAL;
1491 	}
1492 
1493 	if (validate_region_size(rs, region_size))
1494 		return -EINVAL;
1495 
1496 	if (rs->md.chunk_sectors)
1497 		max_io_len = rs->md.chunk_sectors;
1498 	else
1499 		max_io_len = region_size;
1500 
1501 	if (dm_set_target_max_io_len(rs->ti, max_io_len))
1502 		return -EINVAL;
1503 
1504 	if (rt_is_raid10(rt)) {
1505 		if (raid10_copies > rs->md.raid_disks) {
1506 			rs->ti->error = "Not enough devices to satisfy specification";
1507 			return -EINVAL;
1508 		}
1509 
1510 		rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
1511 		if (rs->md.new_layout < 0) {
1512 			rs->ti->error = "Error getting raid10 format";
1513 			return rs->md.new_layout;
1514 		}
1515 
1516 		rt = get_raid_type_by_ll(10, rs->md.new_layout);
1517 		if (!rt) {
1518 			rs->ti->error = "Failed to recognize new raid10 layout";
1519 			return -EINVAL;
1520 		}
1521 
1522 		if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1523 		     rt->algorithm == ALGORITHM_RAID10_NEAR) &&
1524 		    test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1525 			rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1526 			return -EINVAL;
1527 		}
1528 	}
1529 
1530 	rs->raid10_copies = raid10_copies;
1531 
1532 	/* Assume there are no metadata devices until the drives are parsed */
1533 	rs->md.persistent = 0;
1534 	rs->md.external = 1;
1535 
1536 	/* Check, if any invalid ctr arguments have been passed in for the raid level */
1537 	return rs_check_for_valid_flags(rs);
1538 }
1539 
1540 /* Set raid4/5/6 cache size */
rs_set_raid456_stripe_cache(struct raid_set * rs)1541 static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1542 {
1543 	int r;
1544 	struct r5conf *conf;
1545 	struct mddev *mddev = &rs->md;
1546 	uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1547 	uint32_t nr_stripes = rs->stripe_cache_entries;
1548 
1549 	if (!rt_is_raid456(rs->raid_type)) {
1550 		rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1551 		return -EINVAL;
1552 	}
1553 
1554 	if (nr_stripes < min_stripes) {
1555 		DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1556 		       nr_stripes, min_stripes);
1557 		nr_stripes = min_stripes;
1558 	}
1559 
1560 	conf = mddev->private;
1561 	if (!conf) {
1562 		rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1563 		return -EINVAL;
1564 	}
1565 
1566 	/* Try setting number of stripes in raid456 stripe cache */
1567 	if (conf->min_nr_stripes != nr_stripes) {
1568 		r = raid5_set_cache_size(mddev, nr_stripes);
1569 		if (r) {
1570 			rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1571 			return r;
1572 		}
1573 
1574 		DMINFO("%u stripe cache entries", nr_stripes);
1575 	}
1576 
1577 	return 0;
1578 }
1579 
1580 /* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
mddev_data_stripes(struct raid_set * rs)1581 static unsigned int mddev_data_stripes(struct raid_set *rs)
1582 {
1583 	return rs->md.raid_disks - rs->raid_type->parity_devs;
1584 }
1585 
1586 /* Return # of data stripes of @rs (i.e. as of ctr) */
rs_data_stripes(struct raid_set * rs)1587 static unsigned int rs_data_stripes(struct raid_set *rs)
1588 {
1589 	return rs->raid_disks - rs->raid_type->parity_devs;
1590 }
1591 
1592 /*
1593  * Retrieve rdev->sectors from any valid raid device of @rs
1594  * to allow userpace to pass in arbitray "- -" device tupples.
1595  */
__rdev_sectors(struct raid_set * rs)1596 static sector_t __rdev_sectors(struct raid_set *rs)
1597 {
1598 	int i;
1599 
1600 	for (i = 0; i < rs->raid_disks; i++) {
1601 		struct md_rdev *rdev = &rs->dev[i].rdev;
1602 
1603 		if (!test_bit(Journal, &rdev->flags) &&
1604 		    rdev->bdev && rdev->sectors)
1605 			return rdev->sectors;
1606 	}
1607 
1608 	return 0;
1609 }
1610 
1611 /* Check that calculated dev_sectors fits all component devices. */
_check_data_dev_sectors(struct raid_set * rs)1612 static int _check_data_dev_sectors(struct raid_set *rs)
1613 {
1614 	sector_t ds = ~0;
1615 	struct md_rdev *rdev;
1616 
1617 	rdev_for_each(rdev, &rs->md)
1618 		if (!test_bit(Journal, &rdev->flags) && rdev->bdev) {
1619 			ds = min(ds, bdev_nr_sectors(rdev->bdev));
1620 			if (ds < rs->md.dev_sectors) {
1621 				rs->ti->error = "Component device(s) too small";
1622 				return -EINVAL;
1623 			}
1624 		}
1625 
1626 	return 0;
1627 }
1628 
1629 /* Get reshape sectors from data_offsets or raid set */
_get_reshape_sectors(struct raid_set * rs)1630 static sector_t _get_reshape_sectors(struct raid_set *rs)
1631 {
1632 	struct md_rdev *rdev;
1633 	sector_t reshape_sectors = 0;
1634 
1635 	rdev_for_each(rdev, &rs->md)
1636 		if (!test_bit(Journal, &rdev->flags)) {
1637 			reshape_sectors = (rdev->data_offset > rdev->new_data_offset) ?
1638 					rdev->data_offset - rdev->new_data_offset :
1639 					rdev->new_data_offset - rdev->data_offset;
1640 			break;
1641 		}
1642 
1643 	return max(reshape_sectors, (sector_t) rs->data_offset);
1644 }
1645 
1646 /* Calculate the sectors per device and per array used for @rs */
rs_set_dev_and_array_sectors(struct raid_set * rs,sector_t sectors,bool use_mddev)1647 static int rs_set_dev_and_array_sectors(struct raid_set *rs, sector_t sectors, bool use_mddev)
1648 {
1649 	int delta_disks;
1650 	unsigned int data_stripes;
1651 	sector_t array_sectors = sectors, dev_sectors = sectors;
1652 	struct mddev *mddev = &rs->md;
1653 
1654 	if (use_mddev) {
1655 		delta_disks = mddev->delta_disks;
1656 		data_stripes = mddev_data_stripes(rs);
1657 	} else {
1658 		delta_disks = rs->delta_disks;
1659 		data_stripes = rs_data_stripes(rs);
1660 	}
1661 
1662 	/* Special raid1 case w/o delta_disks support (yet) */
1663 	if (rt_is_raid1(rs->raid_type))
1664 		;
1665 	else if (rt_is_raid10(rs->raid_type)) {
1666 		if (rs->raid10_copies < 2 ||
1667 		    delta_disks < 0) {
1668 			rs->ti->error = "Bogus raid10 data copies or delta disks";
1669 			return -EINVAL;
1670 		}
1671 
1672 		dev_sectors *= rs->raid10_copies;
1673 		if (sector_div(dev_sectors, data_stripes))
1674 			goto bad;
1675 
1676 		array_sectors = (data_stripes + delta_disks) * (dev_sectors - _get_reshape_sectors(rs));
1677 		if (sector_div(array_sectors, rs->raid10_copies))
1678 			goto bad;
1679 
1680 	} else if (sector_div(dev_sectors, data_stripes))
1681 		goto bad;
1682 
1683 	else
1684 		/* Striped layouts */
1685 		array_sectors = (data_stripes + delta_disks) * (dev_sectors - _get_reshape_sectors(rs));
1686 
1687 	mddev->array_sectors = array_sectors;
1688 	mddev->dev_sectors = dev_sectors;
1689 	rs_set_rdev_sectors(rs);
1690 
1691 	return _check_data_dev_sectors(rs);
1692 bad:
1693 	rs->ti->error = "Target length not divisible by number of data devices";
1694 	return -EINVAL;
1695 }
1696 
1697 /* Setup recovery on @rs */
rs_setup_recovery(struct raid_set * rs,sector_t dev_sectors)1698 static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1699 {
1700 	/* raid0 does not recover */
1701 	if (rs_is_raid0(rs))
1702 		rs->md.recovery_cp = MaxSector;
1703 	/*
1704 	 * A raid6 set has to be recovered either
1705 	 * completely or for the grown part to
1706 	 * ensure proper parity and Q-Syndrome
1707 	 */
1708 	else if (rs_is_raid6(rs))
1709 		rs->md.recovery_cp = dev_sectors;
1710 	/*
1711 	 * Other raid set types may skip recovery
1712 	 * depending on the 'nosync' flag.
1713 	 */
1714 	else
1715 		rs->md.recovery_cp = test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)
1716 				     ? MaxSector : dev_sectors;
1717 }
1718 
do_table_event(struct work_struct * ws)1719 static void do_table_event(struct work_struct *ws)
1720 {
1721 	struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1722 
1723 	smp_rmb(); /* Make sure we access most actual mddev properties */
1724 
1725 	/* Only grow size resulting from added stripe(s) after reshape ended. */
1726 	if (!rs_is_reshaping(rs) &&
1727 	    rs->array_sectors > rs->md.array_sectors &&
1728 	    !rs->md.delta_disks &&
1729 	    rs->md.raid_disks == rs->raid_disks) {
1730 		/* The raid10 personality doesn't provide proper device sizes -> correct. */
1731 		if (rs_is_raid10(rs))
1732 			rs_set_rdev_sectors(rs);
1733 
1734 		rs->md.array_sectors = rs->array_sectors;
1735 		rs_set_capacity(rs);
1736 	}
1737 
1738 	dm_table_event(rs->ti->table);
1739 }
1740 
1741 /*
1742  * Make sure a valid takover (level switch) is being requested on @rs
1743  *
1744  * Conversions of raid sets from one MD personality to another
1745  * have to conform to restrictions which are enforced here.
1746  */
rs_check_takeover(struct raid_set * rs)1747 static int rs_check_takeover(struct raid_set *rs)
1748 {
1749 	struct mddev *mddev = &rs->md;
1750 	unsigned int near_copies;
1751 
1752 	if (rs->md.degraded) {
1753 		rs->ti->error = "Can't takeover degraded raid set";
1754 		return -EPERM;
1755 	}
1756 
1757 	if (rs_is_reshaping(rs)) {
1758 		rs->ti->error = "Can't takeover reshaping raid set";
1759 		return -EPERM;
1760 	}
1761 
1762 	switch (mddev->level) {
1763 	case 0:
1764 		/* raid0 -> raid1/5 with one disk */
1765 		if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1766 		    mddev->raid_disks == 1)
1767 			return 0;
1768 
1769 		/* raid0 -> raid10 */
1770 		if (mddev->new_level == 10 &&
1771 		    !(rs->raid_disks % mddev->raid_disks))
1772 			return 0;
1773 
1774 		/* raid0 with multiple disks -> raid4/5/6 */
1775 		if (__within_range(mddev->new_level, 4, 6) &&
1776 		    mddev->new_layout == ALGORITHM_PARITY_N &&
1777 		    mddev->raid_disks > 1)
1778 			return 0;
1779 
1780 		break;
1781 
1782 	case 10:
1783 		/* Can't takeover raid10_offset! */
1784 		if (__is_raid10_offset(mddev->layout))
1785 			break;
1786 
1787 		near_copies = __raid10_near_copies(mddev->layout);
1788 
1789 		/* raid10* -> raid0 */
1790 		if (mddev->new_level == 0) {
1791 			/* Can takeover raid10_near with raid disks divisable by data copies! */
1792 			if (near_copies > 1 &&
1793 			    !(mddev->raid_disks % near_copies)) {
1794 				mddev->raid_disks /= near_copies;
1795 				mddev->delta_disks = mddev->raid_disks;
1796 				return 0;
1797 			}
1798 
1799 			/* Can takeover raid10_far */
1800 			if (near_copies == 1 &&
1801 			    __raid10_far_copies(mddev->layout) > 1)
1802 				return 0;
1803 
1804 			break;
1805 		}
1806 
1807 		/* raid10_{near,far} -> raid1 */
1808 		if (mddev->new_level == 1 &&
1809 		    max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
1810 			return 0;
1811 
1812 		/* raid10_{near,far} with 2 disks -> raid4/5 */
1813 		if (__within_range(mddev->new_level, 4, 5) &&
1814 		    mddev->raid_disks == 2)
1815 			return 0;
1816 		break;
1817 
1818 	case 1:
1819 		/* raid1 with 2 disks -> raid4/5 */
1820 		if (__within_range(mddev->new_level, 4, 5) &&
1821 		    mddev->raid_disks == 2) {
1822 			mddev->degraded = 1;
1823 			return 0;
1824 		}
1825 
1826 		/* raid1 -> raid0 */
1827 		if (mddev->new_level == 0 &&
1828 		    mddev->raid_disks == 1)
1829 			return 0;
1830 
1831 		/* raid1 -> raid10 */
1832 		if (mddev->new_level == 10)
1833 			return 0;
1834 		break;
1835 
1836 	case 4:
1837 		/* raid4 -> raid0 */
1838 		if (mddev->new_level == 0)
1839 			return 0;
1840 
1841 		/* raid4 -> raid1/5 with 2 disks */
1842 		if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1843 		    mddev->raid_disks == 2)
1844 			return 0;
1845 
1846 		/* raid4 -> raid5/6 with parity N */
1847 		if (__within_range(mddev->new_level, 5, 6) &&
1848 		    mddev->layout == ALGORITHM_PARITY_N)
1849 			return 0;
1850 		break;
1851 
1852 	case 5:
1853 		/* raid5 with parity N -> raid0 */
1854 		if (mddev->new_level == 0 &&
1855 		    mddev->layout == ALGORITHM_PARITY_N)
1856 			return 0;
1857 
1858 		/* raid5 with parity N -> raid4 */
1859 		if (mddev->new_level == 4 &&
1860 		    mddev->layout == ALGORITHM_PARITY_N)
1861 			return 0;
1862 
1863 		/* raid5 with 2 disks -> raid1/4/10 */
1864 		if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1865 		    mddev->raid_disks == 2)
1866 			return 0;
1867 
1868 		/* raid5_* ->  raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
1869 		if (mddev->new_level == 6 &&
1870 		    ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1871 		      __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
1872 			return 0;
1873 		break;
1874 
1875 	case 6:
1876 		/* raid6 with parity N -> raid0 */
1877 		if (mddev->new_level == 0 &&
1878 		    mddev->layout == ALGORITHM_PARITY_N)
1879 			return 0;
1880 
1881 		/* raid6 with parity N -> raid4 */
1882 		if (mddev->new_level == 4 &&
1883 		    mddev->layout == ALGORITHM_PARITY_N)
1884 			return 0;
1885 
1886 		/* raid6_*_n with Q-Syndrome N -> raid5_* */
1887 		if (mddev->new_level == 5 &&
1888 		    ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1889 		     __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
1890 			return 0;
1891 		break;
1892 
1893 	default:
1894 		break;
1895 	}
1896 
1897 	rs->ti->error = "takeover not possible";
1898 	return -EINVAL;
1899 }
1900 
1901 /* True if @rs requested to be taken over */
rs_takeover_requested(struct raid_set * rs)1902 static bool rs_takeover_requested(struct raid_set *rs)
1903 {
1904 	return rs->md.new_level != rs->md.level;
1905 }
1906 
1907 /* True if layout is set to reshape. */
rs_is_layout_change(struct raid_set * rs,bool use_mddev)1908 static bool rs_is_layout_change(struct raid_set *rs, bool use_mddev)
1909 {
1910 	return (use_mddev ? rs->md.delta_disks : rs->delta_disks) ||
1911 	       rs->md.new_layout != rs->md.layout ||
1912 	       rs->md.new_chunk_sectors != rs->md.chunk_sectors;
1913 }
1914 
1915 /* True if @rs is requested to reshape by ctr */
rs_reshape_requested(struct raid_set * rs)1916 static bool rs_reshape_requested(struct raid_set *rs)
1917 {
1918 	bool change;
1919 	struct mddev *mddev = &rs->md;
1920 
1921 	if (rs_takeover_requested(rs))
1922 		return false;
1923 
1924 	if (rs_is_raid0(rs))
1925 		return false;
1926 
1927 	change = rs_is_layout_change(rs, false);
1928 
1929 	/* Historical case to support raid1 reshape without delta disks */
1930 	if (rs_is_raid1(rs)) {
1931 		if (rs->delta_disks)
1932 			return !!rs->delta_disks;
1933 
1934 		return !change &&
1935 		       mddev->raid_disks != rs->raid_disks;
1936 	}
1937 
1938 	if (rs_is_raid10(rs))
1939 		return change &&
1940 		       !__is_raid10_far(mddev->new_layout) &&
1941 		       rs->delta_disks >= 0;
1942 
1943 	return change;
1944 }
1945 
1946 /*  Features */
1947 #define	FEATURE_FLAG_SUPPORTS_V190	0x1 /* Supports extended superblock */
1948 
1949 /* State flags for sb->flags */
1950 #define	SB_FLAG_RESHAPE_ACTIVE		0x1
1951 #define	SB_FLAG_RESHAPE_BACKWARDS	0x2
1952 
1953 /*
1954  * This structure is never routinely used by userspace, unlike md superblocks.
1955  * Devices with this superblock should only ever be accessed via device-mapper.
1956  */
1957 #define DM_RAID_MAGIC 0x64526D44
1958 struct dm_raid_superblock {
1959 	__le32 magic;		/* "DmRd" */
1960 	__le32 compat_features;	/* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1961 
1962 	__le32 num_devices;	/* Number of devices in this raid set. (Max 64) */
1963 	__le32 array_position;	/* The position of this drive in the raid set */
1964 
1965 	__le64 events;		/* Incremented by md when superblock updated */
1966 	__le64 failed_devices;	/* Pre 1.9.0 part of bit field of devices to */
1967 				/* indicate failures (see extension below) */
1968 
1969 	/*
1970 	 * This offset tracks the progress of the repair or replacement of
1971 	 * an individual drive.
1972 	 */
1973 	__le64 disk_recovery_offset;
1974 
1975 	/*
1976 	 * This offset tracks the progress of the initial raid set
1977 	 * synchronisation/parity calculation.
1978 	 */
1979 	__le64 array_resync_offset;
1980 
1981 	/*
1982 	 * raid characteristics
1983 	 */
1984 	__le32 level;
1985 	__le32 layout;
1986 	__le32 stripe_sectors;
1987 
1988 	/********************************************************************
1989 	 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1990 	 *
1991 	 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
1992 	 */
1993 
1994 	__le32 flags; /* Flags defining array states for reshaping */
1995 
1996 	/*
1997 	 * This offset tracks the progress of a raid
1998 	 * set reshape in order to be able to restart it
1999 	 */
2000 	__le64 reshape_position;
2001 
2002 	/*
2003 	 * These define the properties of the array in case of an interrupted reshape
2004 	 */
2005 	__le32 new_level;
2006 	__le32 new_layout;
2007 	__le32 new_stripe_sectors;
2008 	__le32 delta_disks;
2009 
2010 	__le64 array_sectors; /* Array size in sectors */
2011 
2012 	/*
2013 	 * Sector offsets to data on devices (reshaping).
2014 	 * Needed to support out of place reshaping, thus
2015 	 * not writing over any stripes whilst converting
2016 	 * them from old to new layout
2017 	 */
2018 	__le64 data_offset;
2019 	__le64 new_data_offset;
2020 
2021 	__le64 sectors; /* Used device size in sectors */
2022 
2023 	/*
2024 	 * Additional Bit field of devices indicating failures to support
2025 	 * up to 256 devices with the 1.9.0 on-disk metadata format
2026 	 */
2027 	__le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
2028 
2029 	__le32 incompat_features;	/* Used to indicate any incompatible features */
2030 
2031 	/* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
2032 } __packed;
2033 
2034 /*
2035  * Check for reshape constraints on raid set @rs:
2036  *
2037  * - reshape function non-existent
2038  * - degraded set
2039  * - ongoing recovery
2040  * - ongoing reshape
2041  *
2042  * Returns 0 if none or -EPERM if given constraint
2043  * and error message reference in @errmsg
2044  */
rs_check_reshape(struct raid_set * rs)2045 static int rs_check_reshape(struct raid_set *rs)
2046 {
2047 	struct mddev *mddev = &rs->md;
2048 
2049 	if (!mddev->pers || !mddev->pers->check_reshape)
2050 		rs->ti->error = "Reshape not supported";
2051 	else if (mddev->degraded)
2052 		rs->ti->error = "Can't reshape degraded raid set";
2053 	else if (rs_is_recovering(rs))
2054 		rs->ti->error = "Convert request on recovering raid set prohibited";
2055 	else if (rs_is_reshaping(rs))
2056 		rs->ti->error = "raid set already reshaping!";
2057 	else if (!(rs_is_raid1(rs) || rs_is_raid10(rs) || rs_is_raid456(rs)))
2058 		rs->ti->error = "Reshaping only supported for raid1/4/5/6/10";
2059 	else
2060 		return 0;
2061 
2062 	return -EPERM;
2063 }
2064 
read_disk_sb(struct md_rdev * rdev,int size,bool force_reload)2065 static int read_disk_sb(struct md_rdev *rdev, int size, bool force_reload)
2066 {
2067 	BUG_ON(!rdev->sb_page);
2068 
2069 	if (rdev->sb_loaded && !force_reload)
2070 		return 0;
2071 
2072 	rdev->sb_loaded = 0;
2073 
2074 	if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, true)) {
2075 		DMERR("Failed to read superblock of device at position %d",
2076 		      rdev->raid_disk);
2077 		md_error(rdev->mddev, rdev);
2078 		set_bit(Faulty, &rdev->flags);
2079 		return -EIO;
2080 	}
2081 
2082 	rdev->sb_loaded = 1;
2083 
2084 	return 0;
2085 }
2086 
sb_retrieve_failed_devices(struct dm_raid_superblock * sb,uint64_t * failed_devices)2087 static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2088 {
2089 	failed_devices[0] = le64_to_cpu(sb->failed_devices);
2090 	memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
2091 
2092 	if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2093 		int i = ARRAY_SIZE(sb->extended_failed_devices);
2094 
2095 		while (i--)
2096 			failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
2097 	}
2098 }
2099 
sb_update_failed_devices(struct dm_raid_superblock * sb,uint64_t * failed_devices)2100 static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2101 {
2102 	int i = ARRAY_SIZE(sb->extended_failed_devices);
2103 
2104 	sb->failed_devices = cpu_to_le64(failed_devices[0]);
2105 	while (i--)
2106 		sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
2107 }
2108 
2109 /*
2110  * Synchronize the superblock members with the raid set properties
2111  *
2112  * All superblock data is little endian.
2113  */
super_sync(struct mddev * mddev,struct md_rdev * rdev)2114 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
2115 {
2116 	bool update_failed_devices = false;
2117 	unsigned int i;
2118 	uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2119 	struct dm_raid_superblock *sb;
2120 	struct raid_set *rs = container_of(mddev, struct raid_set, md);
2121 
2122 	/* No metadata device, no superblock */
2123 	if (!rdev->meta_bdev)
2124 		return;
2125 
2126 	BUG_ON(!rdev->sb_page);
2127 
2128 	sb = page_address(rdev->sb_page);
2129 
2130 	sb_retrieve_failed_devices(sb, failed_devices);
2131 
2132 	for (i = 0; i < rs->raid_disks; i++)
2133 		if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
2134 			update_failed_devices = true;
2135 			set_bit(i, (void *) failed_devices);
2136 		}
2137 
2138 	if (update_failed_devices)
2139 		sb_update_failed_devices(sb, failed_devices);
2140 
2141 	sb->magic = cpu_to_le32(DM_RAID_MAGIC);
2142 	sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2143 
2144 	sb->num_devices = cpu_to_le32(mddev->raid_disks);
2145 	sb->array_position = cpu_to_le32(rdev->raid_disk);
2146 
2147 	sb->events = cpu_to_le64(mddev->events);
2148 
2149 	sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
2150 	sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
2151 
2152 	sb->level = cpu_to_le32(mddev->level);
2153 	sb->layout = cpu_to_le32(mddev->layout);
2154 	sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
2155 
2156 	/********************************************************************
2157 	 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
2158 	 *
2159 	 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
2160 	 */
2161 	sb->new_level = cpu_to_le32(mddev->new_level);
2162 	sb->new_layout = cpu_to_le32(mddev->new_layout);
2163 	sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
2164 
2165 	sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2166 
2167 	smp_rmb(); /* Make sure we access most recent reshape position */
2168 	sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2169 	if (le64_to_cpu(sb->reshape_position) != MaxSector) {
2170 		/* Flag ongoing reshape */
2171 		sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
2172 
2173 		if (mddev->delta_disks < 0 || mddev->reshape_backwards)
2174 			sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
2175 	} else {
2176 		/* Clear reshape flags */
2177 		sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
2178 	}
2179 
2180 	sb->array_sectors = cpu_to_le64(mddev->array_sectors);
2181 	sb->data_offset = cpu_to_le64(rdev->data_offset);
2182 	sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
2183 	sb->sectors = cpu_to_le64(rdev->sectors);
2184 	sb->incompat_features = cpu_to_le32(0);
2185 
2186 	/* Zero out the rest of the payload after the size of the superblock */
2187 	memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
2188 }
2189 
2190 /*
2191  * super_load
2192  *
2193  * This function creates a superblock if one is not found on the device
2194  * and will decide which superblock to use if there's a choice.
2195  *
2196  * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
2197  */
super_load(struct md_rdev * rdev,struct md_rdev * refdev)2198 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
2199 {
2200 	int r;
2201 	struct dm_raid_superblock *sb;
2202 	struct dm_raid_superblock *refsb;
2203 	uint64_t events_sb, events_refsb;
2204 
2205 	r = read_disk_sb(rdev, rdev->sb_size, false);
2206 	if (r)
2207 		return r;
2208 
2209 	sb = page_address(rdev->sb_page);
2210 
2211 	/*
2212 	 * Two cases that we want to write new superblocks and rebuild:
2213 	 * 1) New device (no matching magic number)
2214 	 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2215 	 */
2216 	if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
2217 	    (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
2218 		super_sync(rdev->mddev, rdev);
2219 
2220 		set_bit(FirstUse, &rdev->flags);
2221 		sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2222 
2223 		/* Force writing of superblocks to disk */
2224 		set_bit(MD_SB_CHANGE_DEVS, &rdev->mddev->sb_flags);
2225 
2226 		/* Any superblock is better than none, choose that if given */
2227 		return refdev ? 0 : 1;
2228 	}
2229 
2230 	if (!refdev)
2231 		return 1;
2232 
2233 	events_sb = le64_to_cpu(sb->events);
2234 
2235 	refsb = page_address(refdev->sb_page);
2236 	events_refsb = le64_to_cpu(refsb->events);
2237 
2238 	return (events_sb > events_refsb) ? 1 : 0;
2239 }
2240 
super_init_validation(struct raid_set * rs,struct md_rdev * rdev)2241 static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
2242 {
2243 	int role;
2244 	struct mddev *mddev = &rs->md;
2245 	uint64_t events_sb;
2246 	uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2247 	struct dm_raid_superblock *sb;
2248 	uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
2249 	struct md_rdev *r;
2250 	struct dm_raid_superblock *sb2;
2251 
2252 	sb = page_address(rdev->sb_page);
2253 	events_sb = le64_to_cpu(sb->events);
2254 
2255 	/*
2256 	 * Initialise to 1 if this is a new superblock.
2257 	 */
2258 	mddev->events = events_sb ? : 1;
2259 
2260 	mddev->reshape_position = MaxSector;
2261 
2262 	mddev->raid_disks = le32_to_cpu(sb->num_devices);
2263 	mddev->level = le32_to_cpu(sb->level);
2264 	mddev->layout = le32_to_cpu(sb->layout);
2265 	mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
2266 
2267 	/*
2268 	 * Reshaping is supported, e.g. reshape_position is valid
2269 	 * in superblock and superblock content is authoritative.
2270 	 */
2271 	if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2272 		/* Superblock is authoritative wrt given raid set layout! */
2273 		mddev->new_level = le32_to_cpu(sb->new_level);
2274 		mddev->new_layout = le32_to_cpu(sb->new_layout);
2275 		mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
2276 		mddev->delta_disks = le32_to_cpu(sb->delta_disks);
2277 		mddev->array_sectors = le64_to_cpu(sb->array_sectors);
2278 
2279 		/* raid was reshaping and got interrupted */
2280 		if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
2281 			if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
2282 				DMERR("Reshape requested but raid set is still reshaping");
2283 				return -EINVAL;
2284 			}
2285 
2286 			if (mddev->delta_disks < 0 ||
2287 			    (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
2288 				mddev->reshape_backwards = 1;
2289 			else
2290 				mddev->reshape_backwards = 0;
2291 
2292 			mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2293 			rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2294 		}
2295 
2296 	} else {
2297 		/*
2298 		 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2299 		 */
2300 		struct raid_type *rt_cur = get_raid_type_by_ll(mddev->level, mddev->layout);
2301 		struct raid_type *rt_new = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
2302 
2303 		if (rs_takeover_requested(rs)) {
2304 			if (rt_cur && rt_new)
2305 				DMERR("Takeover raid sets from %s to %s not yet supported by metadata. (raid level change)",
2306 				      rt_cur->name, rt_new->name);
2307 			else
2308 				DMERR("Takeover raid sets not yet supported by metadata. (raid level change)");
2309 			return -EINVAL;
2310 		} else if (rs_reshape_requested(rs)) {
2311 			DMERR("Reshaping raid sets not yet supported by metadata. (raid layout change keeping level)");
2312 			if (mddev->layout != mddev->new_layout) {
2313 				if (rt_cur && rt_new)
2314 					DMERR("	 current layout %s vs new layout %s",
2315 					      rt_cur->name, rt_new->name);
2316 				else
2317 					DMERR("	 current layout 0x%X vs new layout 0x%X",
2318 					      le32_to_cpu(sb->layout), mddev->new_layout);
2319 			}
2320 			if (mddev->chunk_sectors != mddev->new_chunk_sectors)
2321 				DMERR("	 current stripe sectors %u vs new stripe sectors %u",
2322 				      mddev->chunk_sectors, mddev->new_chunk_sectors);
2323 			if (rs->delta_disks)
2324 				DMERR("	 current %u disks vs new %u disks",
2325 				      mddev->raid_disks, mddev->raid_disks + rs->delta_disks);
2326 			if (rs_is_raid10(rs)) {
2327 				DMERR("	 Old layout: %s w/ %u copies",
2328 				      raid10_md_layout_to_format(mddev->layout),
2329 				      raid10_md_layout_to_copies(mddev->layout));
2330 				DMERR("	 New layout: %s w/ %u copies",
2331 				      raid10_md_layout_to_format(mddev->new_layout),
2332 				      raid10_md_layout_to_copies(mddev->new_layout));
2333 			}
2334 			return -EINVAL;
2335 		}
2336 
2337 		DMINFO("Discovered old metadata format; upgrading to extended metadata format");
2338 	}
2339 
2340 	if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
2341 		mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2342 
2343 	/*
2344 	 * During load, we set FirstUse if a new superblock was written.
2345 	 * There are two reasons we might not have a superblock:
2346 	 * 1) The raid set is brand new - in which case, all of the
2347 	 *    devices must have their In_sync bit set.	Also,
2348 	 *    recovery_cp must be 0, unless forced.
2349 	 * 2) This is a new device being added to an old raid set
2350 	 *    and the new device needs to be rebuilt - in which
2351 	 *    case the In_sync bit will /not/ be set and
2352 	 *    recovery_cp must be MaxSector.
2353 	 * 3) This is/are a new device(s) being added to an old
2354 	 *    raid set during takeover to a higher raid level
2355 	 *    to provide capacity for redundancy or during reshape
2356 	 *    to add capacity to grow the raid set.
2357 	 */
2358 	rdev_for_each(r, mddev) {
2359 		if (test_bit(Journal, &rdev->flags))
2360 			continue;
2361 
2362 		if (test_bit(FirstUse, &r->flags))
2363 			new_devs++;
2364 
2365 		if (!test_bit(In_sync, &r->flags)) {
2366 			DMINFO("Device %d specified for rebuild; clearing superblock",
2367 				r->raid_disk);
2368 			rebuilds++;
2369 
2370 			if (test_bit(FirstUse, &r->flags))
2371 				rebuild_and_new++;
2372 		}
2373 	}
2374 
2375 	if (new_devs == rs->raid_disks || !rebuilds) {
2376 		/* Replace a broken device */
2377 		if (new_devs == rs->raid_disks) {
2378 			DMINFO("Superblocks created for new raid set");
2379 			set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2380 		} else if (new_devs != rebuilds &&
2381 			   new_devs != rs->delta_disks) {
2382 			DMERR("New device injected into existing raid set without "
2383 			      "'delta_disks' or 'rebuild' parameter specified");
2384 			return -EINVAL;
2385 		}
2386 	} else if (new_devs && new_devs != rebuilds) {
2387 		DMERR("%u 'rebuild' devices cannot be injected into"
2388 		      " a raid set with %u other first-time devices",
2389 		      rebuilds, new_devs);
2390 		return -EINVAL;
2391 	} else if (rebuilds) {
2392 		if (rebuild_and_new && rebuilds != rebuild_and_new) {
2393 			DMERR("new device%s provided without 'rebuild'",
2394 			      new_devs > 1 ? "s" : "");
2395 			return -EINVAL;
2396 		} else if (!test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) && rs_is_recovering(rs)) {
2397 			DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2398 			      (unsigned long long) mddev->recovery_cp);
2399 			return -EINVAL;
2400 		} else if (rs_is_reshaping(rs)) {
2401 			DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2402 			      (unsigned long long) mddev->reshape_position);
2403 			return -EINVAL;
2404 		}
2405 	}
2406 
2407 	/*
2408 	 * Now we set the Faulty bit for those devices that are
2409 	 * recorded in the superblock as failed.
2410 	 */
2411 	sb_retrieve_failed_devices(sb, failed_devices);
2412 	rdev_for_each(r, mddev) {
2413 		if (test_bit(Journal, &rdev->flags) ||
2414 		    !r->sb_page)
2415 			continue;
2416 		sb2 = page_address(r->sb_page);
2417 		sb2->failed_devices = 0;
2418 		memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
2419 
2420 		/*
2421 		 * Check for any device re-ordering.
2422 		 */
2423 		if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2424 			role = le32_to_cpu(sb2->array_position);
2425 			if (role < 0)
2426 				continue;
2427 
2428 			if (role != r->raid_disk) {
2429 				if (rs_is_raid10(rs) && __is_raid10_near(mddev->layout)) {
2430 					if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
2431 					    rs->raid_disks % rs->raid10_copies) {
2432 						rs->ti->error =
2433 							"Cannot change raid10 near set to odd # of devices!";
2434 						return -EINVAL;
2435 					}
2436 
2437 					sb2->array_position = cpu_to_le32(r->raid_disk);
2438 
2439 				} else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
2440 					   !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2441 					   !rt_is_raid1(rs->raid_type)) {
2442 					rs->ti->error = "Cannot change device positions in raid set";
2443 					return -EINVAL;
2444 				}
2445 
2446 				DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
2447 			}
2448 
2449 			/*
2450 			 * Partial recovery is performed on
2451 			 * returning failed devices.
2452 			 */
2453 			if (test_bit(role, (void *) failed_devices))
2454 				set_bit(Faulty, &r->flags);
2455 		}
2456 	}
2457 
2458 	return 0;
2459 }
2460 
super_validate(struct raid_set * rs,struct md_rdev * rdev)2461 static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
2462 {
2463 	struct mddev *mddev = &rs->md;
2464 	struct dm_raid_superblock *sb;
2465 
2466 	if (rs_is_raid0(rs) || !rdev->sb_page || rdev->raid_disk < 0)
2467 		return 0;
2468 
2469 	sb = page_address(rdev->sb_page);
2470 
2471 	/*
2472 	 * If mddev->events is not set, we know we have not yet initialized
2473 	 * the array.
2474 	 */
2475 	if (!mddev->events && super_init_validation(rs, rdev))
2476 		return -EINVAL;
2477 
2478 	if (le32_to_cpu(sb->compat_features) &&
2479 	    le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
2480 		rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2481 		return -EINVAL;
2482 	}
2483 
2484 	if (sb->incompat_features) {
2485 		rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
2486 		return -EINVAL;
2487 	}
2488 
2489 	/* Enable bitmap creation on @rs unless no metadevs or raid0 or journaled raid4/5/6 set. */
2490 	mddev->bitmap_info.offset = (rt_is_raid0(rs->raid_type) || rs->journal_dev.dev) ? 0 : to_sector(4096);
2491 	mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2492 
2493 	if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2494 		/*
2495 		 * Retrieve rdev size stored in superblock to be prepared for shrink.
2496 		 * Check extended superblock members are present otherwise the size
2497 		 * will not be set!
2498 		 */
2499 		if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190)
2500 			rdev->sectors = le64_to_cpu(sb->sectors);
2501 
2502 		rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
2503 		if (rdev->recovery_offset == MaxSector)
2504 			set_bit(In_sync, &rdev->flags);
2505 		/*
2506 		 * If no reshape in progress -> we're recovering single
2507 		 * disk(s) and have to set the device(s) to out-of-sync
2508 		 */
2509 		else if (!rs_is_reshaping(rs))
2510 			clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
2511 	}
2512 
2513 	/*
2514 	 * If a device comes back, set it as not In_sync and no longer faulty.
2515 	 */
2516 	if (test_and_clear_bit(Faulty, &rdev->flags)) {
2517 		rdev->recovery_offset = 0;
2518 		clear_bit(In_sync, &rdev->flags);
2519 		rdev->saved_raid_disk = rdev->raid_disk;
2520 	}
2521 
2522 	/* Reshape support -> restore repective data offsets */
2523 	rdev->data_offset = le64_to_cpu(sb->data_offset);
2524 	rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
2525 
2526 	return 0;
2527 }
2528 
2529 /*
2530  * Analyse superblocks and select the freshest.
2531  */
analyse_superblocks(struct dm_target * ti,struct raid_set * rs)2532 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2533 {
2534 	int r;
2535 	struct md_rdev *rdev, *freshest;
2536 	struct mddev *mddev = &rs->md;
2537 
2538 	freshest = NULL;
2539 	rdev_for_each(rdev, mddev) {
2540 		if (test_bit(Journal, &rdev->flags))
2541 			continue;
2542 
2543 		if (!rdev->meta_bdev)
2544 			continue;
2545 
2546 		/* Set superblock offset/size for metadata device. */
2547 		rdev->sb_start = 0;
2548 		rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
2549 		if (rdev->sb_size < sizeof(struct dm_raid_superblock) || rdev->sb_size > PAGE_SIZE) {
2550 			DMERR("superblock size of a logical block is no longer valid");
2551 			return -EINVAL;
2552 		}
2553 
2554 		/*
2555 		 * Skipping super_load due to CTR_FLAG_SYNC will cause
2556 		 * the array to undergo initialization again as
2557 		 * though it were new.	This is the intended effect
2558 		 * of the "sync" directive.
2559 		 *
2560 		 * With reshaping capability added, we must ensure that
2561 		 * the "sync" directive is disallowed during the reshape.
2562 		 */
2563 		if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2564 			continue;
2565 
2566 		r = super_load(rdev, freshest);
2567 
2568 		switch (r) {
2569 		case 1:
2570 			freshest = rdev;
2571 			break;
2572 		case 0:
2573 			break;
2574 		default:
2575 			/* This is a failure to read the superblock from the metadata device. */
2576 			/*
2577 			 * We have to keep any raid0 data/metadata device pairs or
2578 			 * the MD raid0 personality will fail to start the array.
2579 			 */
2580 			if (rs_is_raid0(rs))
2581 				continue;
2582 
2583 			/*
2584 			 * We keep the dm_devs to be able to emit the device tuple
2585 			 * properly on the table line in raid_status() (rather than
2586 			 * mistakenly acting as if '- -' got passed into the constructor).
2587 			 *
2588 			 * The rdev has to stay on the same_set list to allow for
2589 			 * the attempt to restore faulty devices on second resume.
2590 			 */
2591 			rdev->raid_disk = rdev->saved_raid_disk = -1;
2592 			break;
2593 		}
2594 	}
2595 
2596 	if (!freshest)
2597 		return 0;
2598 
2599 	/*
2600 	 * Validation of the freshest device provides the source of
2601 	 * validation for the remaining devices.
2602 	 */
2603 	rs->ti->error = "Unable to assemble array: Invalid superblocks";
2604 	if (super_validate(rs, freshest))
2605 		return -EINVAL;
2606 
2607 	if (validate_raid_redundancy(rs)) {
2608 		rs->ti->error = "Insufficient redundancy to activate array";
2609 		return -EINVAL;
2610 	}
2611 
2612 	rdev_for_each(rdev, mddev)
2613 		if (!test_bit(Journal, &rdev->flags) &&
2614 		    rdev != freshest &&
2615 		    super_validate(rs, rdev))
2616 			return -EINVAL;
2617 	return 0;
2618 }
2619 
2620 /*
2621  * Adjust data_offset and new_data_offset on all disk members of @rs
2622  * for out of place reshaping if requested by constructor
2623  *
2624  * We need free space at the beginning of each raid disk for forward
2625  * and at the end for backward reshapes which userspace has to provide
2626  * via remapping/reordering of space.
2627  */
rs_adjust_data_offsets(struct raid_set * rs)2628 static int rs_adjust_data_offsets(struct raid_set *rs)
2629 {
2630 	sector_t data_offset = 0, new_data_offset = 0;
2631 	struct md_rdev *rdev;
2632 
2633 	/* Constructor did not request data offset change */
2634 	if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2635 		if (!rs_is_reshapable(rs))
2636 			goto out;
2637 
2638 		return 0;
2639 	}
2640 
2641 	/* HM FIXME: get In_Sync raid_dev? */
2642 	rdev = &rs->dev[0].rdev;
2643 
2644 	if (rs->delta_disks < 0) {
2645 		/*
2646 		 * Removing disks (reshaping backwards):
2647 		 *
2648 		 * - before reshape: data is at offset 0 and free space
2649 		 *		     is at end of each component LV
2650 		 *
2651 		 * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2652 		 */
2653 		data_offset = 0;
2654 		new_data_offset = rs->data_offset;
2655 
2656 	} else if (rs->delta_disks > 0) {
2657 		/*
2658 		 * Adding disks (reshaping forwards):
2659 		 *
2660 		 * - before reshape: data is at offset rs->data_offset != 0 and
2661 		 *		     free space is at begin of each component LV
2662 		 *
2663 		 * - after reshape: data is at offset 0 on each component LV
2664 		 */
2665 		data_offset = rs->data_offset;
2666 		new_data_offset = 0;
2667 
2668 	} else {
2669 		/*
2670 		 * User space passes in 0 for data offset after having removed reshape space
2671 		 *
2672 		 * - or - (data offset != 0)
2673 		 *
2674 		 * Changing RAID layout or chunk size -> toggle offsets
2675 		 *
2676 		 * - before reshape: data is at offset rs->data_offset 0 and
2677 		 *		     free space is at end of each component LV
2678 		 *		     -or-
2679 		 *                   data is at offset rs->data_offset != 0 and
2680 		 *		     free space is at begin of each component LV
2681 		 *
2682 		 * - after reshape: data is at offset 0 if it was at offset != 0
2683 		 *                  or at offset != 0 if it was at offset 0
2684 		 *                  on each component LV
2685 		 *
2686 		 */
2687 		data_offset = rs->data_offset ? rdev->data_offset : 0;
2688 		new_data_offset = data_offset ? 0 : rs->data_offset;
2689 		set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2690 	}
2691 
2692 	/*
2693 	 * Make sure we got a minimum amount of free sectors per device
2694 	 */
2695 	if (rs->data_offset &&
2696 	    bdev_nr_sectors(rdev->bdev) - rs->md.dev_sectors < MIN_FREE_RESHAPE_SPACE) {
2697 		rs->ti->error = data_offset ? "No space for forward reshape" :
2698 					      "No space for backward reshape";
2699 		return -ENOSPC;
2700 	}
2701 out:
2702 	/*
2703 	 * Raise recovery_cp in case data_offset != 0 to
2704 	 * avoid false recovery positives in the constructor.
2705 	 */
2706 	if (rs->md.recovery_cp < rs->md.dev_sectors)
2707 		rs->md.recovery_cp += rs->dev[0].rdev.data_offset;
2708 
2709 	/* Adjust data offsets on all rdevs but on any raid4/5/6 journal device */
2710 	rdev_for_each(rdev, &rs->md) {
2711 		if (!test_bit(Journal, &rdev->flags)) {
2712 			rdev->data_offset = data_offset;
2713 			rdev->new_data_offset = new_data_offset;
2714 		}
2715 	}
2716 
2717 	return 0;
2718 }
2719 
2720 /* Userpace reordered disks -> adjust raid_disk indexes in @rs */
__reorder_raid_disk_indexes(struct raid_set * rs)2721 static void __reorder_raid_disk_indexes(struct raid_set *rs)
2722 {
2723 	int i = 0;
2724 	struct md_rdev *rdev;
2725 
2726 	rdev_for_each(rdev, &rs->md) {
2727 		if (!test_bit(Journal, &rdev->flags)) {
2728 			rdev->raid_disk = i++;
2729 			rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2730 		}
2731 	}
2732 }
2733 
2734 /*
2735  * Setup @rs for takeover by a different raid level
2736  */
rs_setup_takeover(struct raid_set * rs)2737 static int rs_setup_takeover(struct raid_set *rs)
2738 {
2739 	struct mddev *mddev = &rs->md;
2740 	struct md_rdev *rdev;
2741 	unsigned int d = mddev->raid_disks = rs->raid_disks;
2742 	sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2743 
2744 	if (rt_is_raid10(rs->raid_type)) {
2745 		if (rs_is_raid0(rs)) {
2746 			/* Userpace reordered disks -> adjust raid_disk indexes */
2747 			__reorder_raid_disk_indexes(rs);
2748 
2749 			/* raid0 -> raid10_far layout */
2750 			mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2751 								   rs->raid10_copies);
2752 		} else if (rs_is_raid1(rs))
2753 			/* raid1 -> raid10_near layout */
2754 			mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2755 								   rs->raid_disks);
2756 		else
2757 			return -EINVAL;
2758 
2759 	}
2760 
2761 	clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2762 	mddev->recovery_cp = MaxSector;
2763 
2764 	while (d--) {
2765 		rdev = &rs->dev[d].rdev;
2766 
2767 		if (test_bit(d, (void *) rs->rebuild_disks)) {
2768 			clear_bit(In_sync, &rdev->flags);
2769 			clear_bit(Faulty, &rdev->flags);
2770 			mddev->recovery_cp = rdev->recovery_offset = 0;
2771 			/* Bitmap has to be created when we do an "up" takeover */
2772 			set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2773 		}
2774 
2775 		rdev->new_data_offset = new_data_offset;
2776 	}
2777 
2778 	return 0;
2779 }
2780 
2781 /* Prepare @rs for reshape */
rs_prepare_reshape(struct raid_set * rs)2782 static int rs_prepare_reshape(struct raid_set *rs)
2783 {
2784 	bool reshape;
2785 	struct mddev *mddev = &rs->md;
2786 
2787 	if (rs_is_raid10(rs)) {
2788 		if (rs->raid_disks != mddev->raid_disks &&
2789 		    __is_raid10_near(mddev->layout) &&
2790 		    rs->raid10_copies &&
2791 		    rs->raid10_copies != __raid10_near_copies(mddev->layout)) {
2792 			/*
2793 			 * raid disk have to be multiple of data copies to allow this conversion,
2794 			 *
2795 			 * This is actually not a reshape it is a
2796 			 * rebuild of any additional mirrors per group
2797 			 */
2798 			if (rs->raid_disks % rs->raid10_copies) {
2799 				rs->ti->error = "Can't reshape raid10 mirror groups";
2800 				return -EINVAL;
2801 			}
2802 
2803 			/* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2804 			__reorder_raid_disk_indexes(rs);
2805 			mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2806 								   rs->raid10_copies);
2807 			mddev->new_layout = mddev->layout;
2808 			reshape = false;
2809 		} else
2810 			reshape = true;
2811 
2812 	} else if (rs_is_raid456(rs))
2813 		reshape = true;
2814 
2815 	else if (rs_is_raid1(rs)) {
2816 		if (rs->delta_disks) {
2817 			/* Process raid1 via delta_disks */
2818 			mddev->degraded = rs->delta_disks < 0 ? -rs->delta_disks : rs->delta_disks;
2819 			reshape = true;
2820 		} else {
2821 			/* Process raid1 without delta_disks */
2822 			mddev->raid_disks = rs->raid_disks;
2823 			reshape = false;
2824 		}
2825 	} else {
2826 		rs->ti->error = "Called with bogus raid type";
2827 		return -EINVAL;
2828 	}
2829 
2830 	if (reshape) {
2831 		set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2832 		set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2833 	} else if (mddev->raid_disks < rs->raid_disks)
2834 		/* Create new superblocks and bitmaps, if any new disks */
2835 		set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2836 
2837 	return 0;
2838 }
2839 
2840 /*
2841  * Reshape:
2842  * - change raid layout
2843  * - change chunk size
2844  * - add disks
2845  * - remove disks
2846  */
rs_setup_reshape(struct raid_set * rs)2847 static int rs_setup_reshape(struct raid_set *rs)
2848 {
2849 	int r = 0;
2850 	unsigned int cur_raid_devs, d;
2851 	sector_t reshape_sectors = _get_reshape_sectors(rs);
2852 	struct mddev *mddev = &rs->md;
2853 	struct md_rdev *rdev;
2854 
2855 	mddev->delta_disks = rs->delta_disks;
2856 	cur_raid_devs = mddev->raid_disks;
2857 
2858 	/* Ignore impossible layout change whilst adding/removing disks */
2859 	if (mddev->delta_disks &&
2860 	    mddev->layout != mddev->new_layout) {
2861 		DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2862 		mddev->new_layout = mddev->layout;
2863 	}
2864 
2865 	/*
2866 	 * Adjust array size:
2867 	 *
2868 	 * - in case of adding disk(s), array size has
2869 	 *   to grow after the disk adding reshape,
2870 	 *   which'll happen in the event handler;
2871 	 *   reshape will happen forward, so space has to
2872 	 *   be available at the beginning of each disk
2873 	 *
2874 	 * - in case of removing disk(s), array size
2875 	 *   has to shrink before starting the reshape,
2876 	 *   which'll happen here;
2877 	 *   reshape will happen backward, so space has to
2878 	 *   be available at the end of each disk
2879 	 *
2880 	 * - data_offset and new_data_offset are
2881 	 *   adjusted for aforementioned out of place
2882 	 *   reshaping based on userspace passing in
2883 	 *   the "data_offset <sectors>" key/value
2884 	 *   pair via the constructor
2885 	 */
2886 
2887 	/* Add disk(s) */
2888 	if (rs->delta_disks > 0) {
2889 		/* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2890 		for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2891 			rdev = &rs->dev[d].rdev;
2892 			clear_bit(In_sync, &rdev->flags);
2893 
2894 			/*
2895 			 * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2896 			 * by md, which'll store that erroneously in the superblock on reshape
2897 			 */
2898 			rdev->saved_raid_disk = -1;
2899 			rdev->raid_disk = d;
2900 
2901 			rdev->sectors = mddev->dev_sectors;
2902 			rdev->recovery_offset = rs_is_raid1(rs) ? 0 : MaxSector;
2903 		}
2904 
2905 		mddev->reshape_backwards = 0; /* adding disk(s) -> forward reshape */
2906 
2907 	/* Remove disk(s) */
2908 	} else if (rs->delta_disks < 0) {
2909 		r = rs_set_dev_and_array_sectors(rs, rs->ti->len, true);
2910 		mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2911 
2912 	/* Change layout and/or chunk size */
2913 	} else {
2914 		/*
2915 		 * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2916 		 *
2917 		 * keeping number of disks and do layout change ->
2918 		 *
2919 		 * toggle reshape_backward depending on data_offset:
2920 		 *
2921 		 * - free space upfront -> reshape forward
2922 		 *
2923 		 * - free space at the end -> reshape backward
2924 		 *
2925 		 *
2926 		 * This utilizes free reshape space avoiding the need
2927 		 * for userspace to move (parts of) LV segments in
2928 		 * case of layout/chunksize change  (for disk
2929 		 * adding/removing reshape space has to be at
2930 		 * the proper address (see above with delta_disks):
2931 		 *
2932 		 * add disk(s)   -> begin
2933 		 * remove disk(s)-> end
2934 		 */
2935 		mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2936 	}
2937 
2938 	/*
2939 	 * Adjust device size for forward reshape
2940 	 * because md_finish_reshape() reduces it.
2941 	 */
2942 	if (!mddev->reshape_backwards)
2943 		rdev_for_each(rdev, &rs->md)
2944 			if (!test_bit(Journal, &rdev->flags))
2945 				rdev->sectors += reshape_sectors;
2946 
2947 	return r;
2948 }
2949 
2950 /*
2951  * If the md resync thread has updated superblock with max reshape position
2952  * at the end of a reshape but not (yet) reset the layout configuration
2953  * changes -> reset the latter.
2954  */
rs_reset_inconclusive_reshape(struct raid_set * rs)2955 static void rs_reset_inconclusive_reshape(struct raid_set *rs)
2956 {
2957 	if (!rs_is_reshaping(rs) && rs_is_layout_change(rs, true)) {
2958 		rs_set_cur(rs);
2959 		rs->md.delta_disks = 0;
2960 		rs->md.reshape_backwards = 0;
2961 	}
2962 }
2963 
2964 /*
2965  * Enable/disable discard support on RAID set depending on
2966  * RAID level and discard properties of underlying RAID members.
2967  */
configure_discard_support(struct raid_set * rs)2968 static void configure_discard_support(struct raid_set *rs)
2969 {
2970 	int i;
2971 	bool raid456;
2972 	struct dm_target *ti = rs->ti;
2973 
2974 	/*
2975 	 * XXX: RAID level 4,5,6 require zeroing for safety.
2976 	 */
2977 	raid456 = rs_is_raid456(rs);
2978 
2979 	for (i = 0; i < rs->raid_disks; i++) {
2980 		if (!rs->dev[i].rdev.bdev ||
2981 		    !bdev_max_discard_sectors(rs->dev[i].rdev.bdev))
2982 			return;
2983 
2984 		if (raid456) {
2985 			if (!devices_handle_discard_safely) {
2986 				DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2987 				DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2988 				return;
2989 			}
2990 		}
2991 	}
2992 
2993 	ti->num_discard_bios = 1;
2994 }
2995 
2996 /*
2997  * Construct a RAID0/1/10/4/5/6 mapping:
2998  * Args:
2999  *	<raid_type> <#raid_params> <raid_params>{0,}	\
3000  *	<#raid_devs> [<meta_dev1> <dev1>]{1,}
3001  *
3002  * <raid_params> varies by <raid_type>.	 See 'parse_raid_params' for
3003  * details on possible <raid_params>.
3004  *
3005  * Userspace is free to initialize the metadata devices, hence the superblocks to
3006  * enforce recreation based on the passed in table parameters.
3007  *
3008  */
raid_ctr(struct dm_target * ti,unsigned int argc,char ** argv)3009 static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3010 {
3011 	int r;
3012 	bool resize = false;
3013 	struct raid_type *rt;
3014 	unsigned int num_raid_params, num_raid_devs;
3015 	sector_t sb_array_sectors, rdev_sectors, reshape_sectors;
3016 	struct raid_set *rs = NULL;
3017 	const char *arg;
3018 	struct rs_layout rs_layout;
3019 	struct dm_arg_set as = { argc, argv }, as_nrd;
3020 	struct dm_arg _args[] = {
3021 		{ 0, as.argc, "Cannot understand number of raid parameters" },
3022 		{ 1, 254, "Cannot understand number of raid devices parameters" }
3023 	};
3024 
3025 	arg = dm_shift_arg(&as);
3026 	if (!arg) {
3027 		ti->error = "No arguments";
3028 		return -EINVAL;
3029 	}
3030 
3031 	rt = get_raid_type(arg);
3032 	if (!rt) {
3033 		ti->error = "Unrecognised raid_type";
3034 		return -EINVAL;
3035 	}
3036 
3037 	/* Must have <#raid_params> */
3038 	if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
3039 		return -EINVAL;
3040 
3041 	/* number of raid device tupples <meta_dev data_dev> */
3042 	as_nrd = as;
3043 	dm_consume_args(&as_nrd, num_raid_params);
3044 	_args[1].max = (as_nrd.argc - 1) / 2;
3045 	if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
3046 		return -EINVAL;
3047 
3048 	if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
3049 		ti->error = "Invalid number of supplied raid devices";
3050 		return -EINVAL;
3051 	}
3052 
3053 	rs = raid_set_alloc(ti, rt, num_raid_devs);
3054 	if (IS_ERR(rs))
3055 		return PTR_ERR(rs);
3056 
3057 	r = parse_raid_params(rs, &as, num_raid_params);
3058 	if (r)
3059 		goto bad;
3060 
3061 	r = parse_dev_params(rs, &as);
3062 	if (r)
3063 		goto bad;
3064 
3065 	rs->md.sync_super = super_sync;
3066 
3067 	/*
3068 	 * Calculate ctr requested array and device sizes to allow
3069 	 * for superblock analysis needing device sizes defined.
3070 	 *
3071 	 * Any existing superblock will overwrite the array and device sizes
3072 	 */
3073 	r = rs_set_dev_and_array_sectors(rs, rs->ti->len, false);
3074 	if (r)
3075 		goto bad;
3076 
3077 	/* Memorize just calculated, potentially larger sizes to grow the raid set in preresume */
3078 	rs->array_sectors = rs->md.array_sectors;
3079 	rs->dev_sectors = rs->md.dev_sectors;
3080 
3081 	/*
3082 	 * Backup any new raid set level, layout, ...
3083 	 * requested to be able to compare to superblock
3084 	 * members for conversion decisions.
3085 	 */
3086 	rs_config_backup(rs, &rs_layout);
3087 
3088 	r = analyse_superblocks(ti, rs);
3089 	if (r)
3090 		goto bad;
3091 
3092 	/* All in-core metadata now as of current superblocks after calling analyse_superblocks() */
3093 	sb_array_sectors = rs->md.array_sectors;
3094 	rdev_sectors = __rdev_sectors(rs);
3095 	if (!rdev_sectors) {
3096 		ti->error = "Invalid rdev size";
3097 		r = -EINVAL;
3098 		goto bad;
3099 	}
3100 
3101 
3102 	reshape_sectors = _get_reshape_sectors(rs);
3103 	if (rs->dev_sectors != rdev_sectors) {
3104 		resize = (rs->dev_sectors != rdev_sectors - reshape_sectors);
3105 		if (rs->dev_sectors > rdev_sectors - reshape_sectors)
3106 			set_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3107 	}
3108 
3109 	INIT_WORK(&rs->md.event_work, do_table_event);
3110 	ti->private = rs;
3111 	ti->num_flush_bios = 1;
3112 	ti->needs_bio_set_dev = true;
3113 
3114 	/* Restore any requested new layout for conversion decision */
3115 	rs_config_restore(rs, &rs_layout);
3116 
3117 	/*
3118 	 * Now that we have any superblock metadata available,
3119 	 * check for new, recovering, reshaping, to be taken over,
3120 	 * to be reshaped or an existing, unchanged raid set to
3121 	 * run in sequence.
3122 	 */
3123 	if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
3124 		/* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
3125 		if (rs_is_raid6(rs) &&
3126 		    test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
3127 			ti->error = "'nosync' not allowed for new raid6 set";
3128 			r = -EINVAL;
3129 			goto bad;
3130 		}
3131 		rs_setup_recovery(rs, 0);
3132 		set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3133 		rs_set_new(rs);
3134 	} else if (rs_is_recovering(rs)) {
3135 		/* A recovering raid set may be resized */
3136 		goto size_check;
3137 	} else if (rs_is_reshaping(rs)) {
3138 		/* Have to reject size change request during reshape */
3139 		if (resize) {
3140 			ti->error = "Can't resize a reshaping raid set";
3141 			r = -EPERM;
3142 			goto bad;
3143 		}
3144 		/* skip setup rs */
3145 	} else if (rs_takeover_requested(rs)) {
3146 		if (rs_is_reshaping(rs)) {
3147 			ti->error = "Can't takeover a reshaping raid set";
3148 			r = -EPERM;
3149 			goto bad;
3150 		}
3151 
3152 		/* We can't takeover a journaled raid4/5/6 */
3153 		if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3154 			ti->error = "Can't takeover a journaled raid4/5/6 set";
3155 			r = -EPERM;
3156 			goto bad;
3157 		}
3158 
3159 		/*
3160 		 * If a takeover is needed, userspace sets any additional
3161 		 * devices to rebuild and we can check for a valid request here.
3162 		 *
3163 		 * If acceptable, set the level to the new requested
3164 		 * one, prohibit requesting recovery, allow the raid
3165 		 * set to run and store superblocks during resume.
3166 		 */
3167 		r = rs_check_takeover(rs);
3168 		if (r)
3169 			goto bad;
3170 
3171 		r = rs_setup_takeover(rs);
3172 		if (r)
3173 			goto bad;
3174 
3175 		set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3176 		/* Takeover ain't recovery, so disable recovery */
3177 		rs_setup_recovery(rs, MaxSector);
3178 		rs_set_new(rs);
3179 	} else if (rs_reshape_requested(rs)) {
3180 		/* Only request grow on raid set size extensions, not on reshapes. */
3181 		clear_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3182 
3183 		/*
3184 		 * No need to check for 'ongoing' takeover here, because takeover
3185 		 * is an instant operation as oposed to an ongoing reshape.
3186 		 */
3187 
3188 		/* We can't reshape a journaled raid4/5/6 */
3189 		if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3190 			ti->error = "Can't reshape a journaled raid4/5/6 set";
3191 			r = -EPERM;
3192 			goto bad;
3193 		}
3194 
3195 		/* Out-of-place space has to be available to allow for a reshape unless raid1! */
3196 		if (reshape_sectors || rs_is_raid1(rs)) {
3197 			/*
3198 			 * We can only prepare for a reshape here, because the
3199 			 * raid set needs to run to provide the repective reshape
3200 			 * check functions via its MD personality instance.
3201 			 *
3202 			 * So do the reshape check after md_run() succeeded.
3203 			 */
3204 			r = rs_prepare_reshape(rs);
3205 			if (r)
3206 				goto bad;
3207 
3208 			/* Reshaping ain't recovery, so disable recovery */
3209 			rs_setup_recovery(rs, MaxSector);
3210 		}
3211 		rs_set_cur(rs);
3212 	} else {
3213 size_check:
3214 		/* May not set recovery when a device rebuild is requested */
3215 		if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3216 			clear_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3217 			set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3218 			rs_setup_recovery(rs, MaxSector);
3219 		} else if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags)) {
3220 			/*
3221 			 * Set raid set to current size, i.e. size as of
3222 			 * superblocks to grow to larger size in preresume.
3223 			 */
3224 			r = rs_set_dev_and_array_sectors(rs, sb_array_sectors, false);
3225 			if (r)
3226 				goto bad;
3227 
3228 			rs_setup_recovery(rs, rs->md.recovery_cp < rs->md.dev_sectors ? rs->md.recovery_cp : rs->md.dev_sectors);
3229 		} else {
3230 			/* This is no size change or it is shrinking, update size and record in superblocks */
3231 			r = rs_set_dev_and_array_sectors(rs, rs->ti->len, false);
3232 			if (r)
3233 				goto bad;
3234 
3235 			if (sb_array_sectors > rs->array_sectors)
3236 				set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3237 		}
3238 		rs_set_cur(rs);
3239 	}
3240 
3241 	/* If constructor requested it, change data and new_data offsets */
3242 	r = rs_adjust_data_offsets(rs);
3243 	if (r)
3244 		goto bad;
3245 
3246 	/* Catch any inconclusive reshape superblock content. */
3247 	rs_reset_inconclusive_reshape(rs);
3248 
3249 	/* Start raid set read-only and assumed clean to change in raid_resume() */
3250 	rs->md.ro = 1;
3251 	rs->md.in_sync = 1;
3252 
3253 	/* Has to be held on running the array */
3254 	mddev_suspend_and_lock_nointr(&rs->md);
3255 
3256 	/* Keep array frozen until resume. */
3257 	md_frozen_sync_thread(&rs->md);
3258 
3259 	r = md_run(&rs->md);
3260 	rs->md.in_sync = 0; /* Assume already marked dirty */
3261 	if (r) {
3262 		ti->error = "Failed to run raid array";
3263 		mddev_unlock(&rs->md);
3264 		goto bad;
3265 	}
3266 
3267 	r = md_start(&rs->md);
3268 	if (r) {
3269 		ti->error = "Failed to start raid array";
3270 		goto bad_unlock;
3271 	}
3272 
3273 	/* If raid4/5/6 journal mode explicitly requested (only possible with journal dev) -> set it */
3274 	if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
3275 		r = r5c_journal_mode_set(&rs->md, rs->journal_dev.mode);
3276 		if (r) {
3277 			ti->error = "Failed to set raid4/5/6 journal mode";
3278 			goto bad_unlock;
3279 		}
3280 	}
3281 
3282 	set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags);
3283 
3284 	/* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
3285 	if (rs_is_raid456(rs)) {
3286 		r = rs_set_raid456_stripe_cache(rs);
3287 		if (r)
3288 			goto bad_unlock;
3289 	}
3290 
3291 	/* Now do an early reshape check */
3292 	if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3293 		r = rs_check_reshape(rs);
3294 		if (r)
3295 			goto bad_unlock;
3296 
3297 		/* Restore new, ctr requested layout to perform check */
3298 		rs_config_restore(rs, &rs_layout);
3299 
3300 		if (rs->md.pers->start_reshape) {
3301 			r = rs->md.pers->check_reshape(&rs->md);
3302 			if (r) {
3303 				ti->error = "Reshape check failed";
3304 				goto bad_unlock;
3305 			}
3306 		}
3307 	}
3308 
3309 	/* Disable/enable discard support on raid set. */
3310 	configure_discard_support(rs);
3311 
3312 	mddev_unlock(&rs->md);
3313 	return 0;
3314 
3315 bad_unlock:
3316 	md_stop(&rs->md);
3317 	mddev_unlock(&rs->md);
3318 bad:
3319 	raid_set_free(rs);
3320 
3321 	return r;
3322 }
3323 
raid_dtr(struct dm_target * ti)3324 static void raid_dtr(struct dm_target *ti)
3325 {
3326 	struct raid_set *rs = ti->private;
3327 
3328 	mddev_lock_nointr(&rs->md);
3329 	md_stop(&rs->md);
3330 	mddev_unlock(&rs->md);
3331 
3332 	if (work_pending(&rs->md.event_work))
3333 		flush_work(&rs->md.event_work);
3334 	raid_set_free(rs);
3335 }
3336 
raid_map(struct dm_target * ti,struct bio * bio)3337 static int raid_map(struct dm_target *ti, struct bio *bio)
3338 {
3339 	struct raid_set *rs = ti->private;
3340 	struct mddev *mddev = &rs->md;
3341 
3342 	/*
3343 	 * If we're reshaping to add disk(s), ti->len and
3344 	 * mddev->array_sectors will differ during the process
3345 	 * (ti->len > mddev->array_sectors), so we have to requeue
3346 	 * bios with addresses > mddev->array_sectors here or
3347 	 * there will occur accesses past EOD of the component
3348 	 * data images thus erroring the raid set.
3349 	 */
3350 	if (unlikely(bio_has_data(bio) && bio_end_sector(bio) > mddev->array_sectors))
3351 		return DM_MAPIO_REQUEUE;
3352 
3353 	if (unlikely(!md_handle_request(mddev, bio)))
3354 		return DM_MAPIO_REQUEUE;
3355 
3356 	return DM_MAPIO_SUBMITTED;
3357 }
3358 
3359 /* Return sync state string for @state */
3360 enum sync_state { st_frozen, st_reshape, st_resync, st_check, st_repair, st_recover, st_idle };
sync_str(enum sync_state state)3361 static const char *sync_str(enum sync_state state)
3362 {
3363 	/* Has to be in above sync_state order! */
3364 	static const char *sync_strs[] = {
3365 		"frozen",
3366 		"reshape",
3367 		"resync",
3368 		"check",
3369 		"repair",
3370 		"recover",
3371 		"idle"
3372 	};
3373 
3374 	return __within_range(state, 0, ARRAY_SIZE(sync_strs) - 1) ? sync_strs[state] : "undef";
3375 };
3376 
3377 /* Return enum sync_state for @mddev derived from @recovery flags */
decipher_sync_action(struct mddev * mddev,unsigned long recovery)3378 static enum sync_state decipher_sync_action(struct mddev *mddev, unsigned long recovery)
3379 {
3380 	if (test_bit(MD_RECOVERY_FROZEN, &recovery))
3381 		return st_frozen;
3382 
3383 	/* The MD sync thread can be done with io or be interrupted but still be running */
3384 	if (!test_bit(MD_RECOVERY_DONE, &recovery) &&
3385 	    (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
3386 	     (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery)))) {
3387 		if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
3388 			return st_reshape;
3389 
3390 		if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
3391 			if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
3392 				return st_resync;
3393 			if (test_bit(MD_RECOVERY_CHECK, &recovery))
3394 				return st_check;
3395 			return st_repair;
3396 		}
3397 
3398 		if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3399 			return st_recover;
3400 
3401 		if (mddev->reshape_position != MaxSector)
3402 			return st_reshape;
3403 	}
3404 
3405 	return st_idle;
3406 }
3407 
3408 /*
3409  * Return status string for @rdev
3410  *
3411  * Status characters:
3412  *
3413  *  'D' = Dead/Failed raid set component or raid4/5/6 journal device
3414  *  'a' = Alive but not in-sync raid set component _or_ alive raid4/5/6 'write_back' journal device
3415  *  'A' = Alive and in-sync raid set component _or_ alive raid4/5/6 'write_through' journal device
3416  *  '-' = Non-existing device (i.e. uspace passed '- -' into the ctr)
3417  */
__raid_dev_status(struct raid_set * rs,struct md_rdev * rdev)3418 static const char *__raid_dev_status(struct raid_set *rs, struct md_rdev *rdev)
3419 {
3420 	if (!rdev->bdev)
3421 		return "-";
3422 	else if (test_bit(Faulty, &rdev->flags))
3423 		return "D";
3424 	else if (test_bit(Journal, &rdev->flags))
3425 		return (rs->journal_dev.mode == R5C_JOURNAL_MODE_WRITE_THROUGH) ? "A" : "a";
3426 	else if (test_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags) ||
3427 		 (!test_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags) &&
3428 		  !test_bit(In_sync, &rdev->flags)))
3429 		return "a";
3430 	else
3431 		return "A";
3432 }
3433 
3434 /* Helper to return resync/reshape progress for @rs and runtime flags for raid set in sync / resynching */
rs_get_progress(struct raid_set * rs,unsigned long recovery,enum sync_state state,sector_t resync_max_sectors)3435 static sector_t rs_get_progress(struct raid_set *rs, unsigned long recovery,
3436 				enum sync_state state, sector_t resync_max_sectors)
3437 {
3438 	sector_t r;
3439 	struct mddev *mddev = &rs->md;
3440 
3441 	clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3442 	clear_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3443 
3444 	if (rs_is_raid0(rs)) {
3445 		r = resync_max_sectors;
3446 		set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3447 
3448 	} else {
3449 		if (state == st_idle && !test_bit(MD_RECOVERY_INTR, &recovery))
3450 			r = mddev->recovery_cp;
3451 		else
3452 			r = mddev->curr_resync_completed;
3453 
3454 		if (state == st_idle && r >= resync_max_sectors) {
3455 			/*
3456 			 * Sync complete.
3457 			 */
3458 			/* In case we have finished recovering, the array is in sync. */
3459 			if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3460 				set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3461 
3462 		} else if (state == st_recover)
3463 			/*
3464 			 * In case we are recovering, the array is not in sync
3465 			 * and health chars should show the recovering legs.
3466 			 *
3467 			 * Already retrieved recovery offset from curr_resync_completed above.
3468 			 */
3469 			;
3470 
3471 		else if (state == st_resync || state == st_reshape)
3472 			/*
3473 			 * If "resync/reshape" is occurring, the raid set
3474 			 * is or may be out of sync hence the health
3475 			 * characters shall be 'a'.
3476 			 */
3477 			set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3478 
3479 		else if (state == st_check || state == st_repair)
3480 			/*
3481 			 * If "check" or "repair" is occurring, the raid set has
3482 			 * undergone an initial sync and the health characters
3483 			 * should not be 'a' anymore.
3484 			 */
3485 			set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3486 
3487 		else if (test_bit(MD_RECOVERY_NEEDED, &recovery))
3488 			/*
3489 			 * We are idle and recovery is needed, prevent 'A' chars race
3490 			 * caused by components still set to in-sync by constructor.
3491 			 */
3492 			set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3493 
3494 		else {
3495 			/*
3496 			 * We are idle and the raid set may be doing an initial
3497 			 * sync, or it may be rebuilding individual components.
3498 			 * If all the devices are In_sync, then it is the raid set
3499 			 * that is being initialized.
3500 			 */
3501 			struct md_rdev *rdev;
3502 
3503 			set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3504 			rdev_for_each(rdev, mddev)
3505 				if (!test_bit(Journal, &rdev->flags) &&
3506 				    !test_bit(In_sync, &rdev->flags)) {
3507 					clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3508 					break;
3509 				}
3510 		}
3511 	}
3512 
3513 	return min(r, resync_max_sectors);
3514 }
3515 
3516 /* Helper to return @dev name or "-" if !@dev */
__get_dev_name(struct dm_dev * dev)3517 static const char *__get_dev_name(struct dm_dev *dev)
3518 {
3519 	return dev ? dev->name : "-";
3520 }
3521 
raid_status(struct dm_target * ti,status_type_t type,unsigned int status_flags,char * result,unsigned int maxlen)3522 static void raid_status(struct dm_target *ti, status_type_t type,
3523 			unsigned int status_flags, char *result, unsigned int maxlen)
3524 {
3525 	struct raid_set *rs = ti->private;
3526 	struct mddev *mddev = &rs->md;
3527 	struct r5conf *conf = rs_is_raid456(rs) ? mddev->private : NULL;
3528 	int i, max_nr_stripes = conf ? conf->max_nr_stripes : 0;
3529 	unsigned long recovery;
3530 	unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3531 	unsigned int sz = 0;
3532 	unsigned int rebuild_writemostly_count = 0;
3533 	sector_t progress, resync_max_sectors, resync_mismatches;
3534 	enum sync_state state;
3535 	struct raid_type *rt;
3536 
3537 	switch (type) {
3538 	case STATUSTYPE_INFO:
3539 		/* *Should* always succeed */
3540 		rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3541 		if (!rt)
3542 			return;
3543 
3544 		DMEMIT("%s %d ", rt->name, mddev->raid_disks);
3545 
3546 		/* Access most recent mddev properties for status output */
3547 		smp_rmb();
3548 		/* Get sensible max sectors even if raid set not yet started */
3549 		resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
3550 				      mddev->resync_max_sectors : mddev->dev_sectors;
3551 		recovery = rs->md.recovery;
3552 		state = decipher_sync_action(mddev, recovery);
3553 		progress = rs_get_progress(rs, recovery, state, resync_max_sectors);
3554 		resync_mismatches = mddev->last_sync_action == ACTION_CHECK ?
3555 				    atomic64_read(&mddev->resync_mismatches) : 0;
3556 
3557 		/* HM FIXME: do we want another state char for raid0? It shows 'D'/'A'/'-' now */
3558 		for (i = 0; i < rs->raid_disks; i++)
3559 			DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev));
3560 
3561 		/*
3562 		 * In-sync/Reshape ratio:
3563 		 *  The in-sync ratio shows the progress of:
3564 		 *   - Initializing the raid set
3565 		 *   - Rebuilding a subset of devices of the raid set
3566 		 *  The user can distinguish between the two by referring
3567 		 *  to the status characters.
3568 		 *
3569 		 *  The reshape ratio shows the progress of
3570 		 *  changing the raid layout or the number of
3571 		 *  disks of a raid set
3572 		 */
3573 		DMEMIT(" %llu/%llu", (unsigned long long) progress,
3574 				     (unsigned long long) resync_max_sectors);
3575 
3576 		/*
3577 		 * v1.5.0+:
3578 		 *
3579 		 * Sync action:
3580 		 *   See Documentation/admin-guide/device-mapper/dm-raid.rst for
3581 		 *   information on each of these states.
3582 		 */
3583 		DMEMIT(" %s", sync_str(state));
3584 
3585 		/*
3586 		 * v1.5.0+:
3587 		 *
3588 		 * resync_mismatches/mismatch_cnt
3589 		 *   This field shows the number of discrepancies found when
3590 		 *   performing a "check" of the raid set.
3591 		 */
3592 		DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3593 
3594 		/*
3595 		 * v1.9.0+:
3596 		 *
3597 		 * data_offset (needed for out of space reshaping)
3598 		 *   This field shows the data offset into the data
3599 		 *   image LV where the first stripes data starts.
3600 		 *
3601 		 * We keep data_offset equal on all raid disks of the set,
3602 		 * so retrieving it from the first raid disk is sufficient.
3603 		 */
3604 		DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
3605 
3606 		/*
3607 		 * v1.10.0+:
3608 		 */
3609 		DMEMIT(" %s", test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ?
3610 			      __raid_dev_status(rs, &rs->journal_dev.rdev) : "-");
3611 		break;
3612 
3613 	case STATUSTYPE_TABLE:
3614 		/* Report the table line string you would use to construct this raid set */
3615 
3616 		/*
3617 		 * Count any rebuild or writemostly argument pairs and subtract the
3618 		 * hweight count being added below of any rebuild and writemostly ctr flags.
3619 		 */
3620 		for (i = 0; i < rs->raid_disks; i++) {
3621 			rebuild_writemostly_count += (test_bit(i, (void *) rs->rebuild_disks) ? 2 : 0) +
3622 						     (test_bit(WriteMostly, &rs->dev[i].rdev.flags) ? 2 : 0);
3623 		}
3624 		rebuild_writemostly_count -= (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) ? 2 : 0) +
3625 					     (test_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags) ? 2 : 0);
3626 		/* Calculate raid parameter count based on ^ rebuild/writemostly argument counts and ctr flags set. */
3627 		raid_param_cnt += rebuild_writemostly_count +
3628 				  hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3629 				  hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2;
3630 		/* Emit table line */
3631 		/* This has to be in the documented order for userspace! */
3632 		DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
3633 		if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
3634 			DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
3635 		if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3636 			DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
3637 		if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags))
3638 			for (i = 0; i < rs->raid_disks; i++)
3639 				if (test_bit(i, (void *) rs->rebuild_disks))
3640 					DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD), i);
3641 		if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3642 			DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3643 					  mddev->bitmap_info.daemon_sleep);
3644 		if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3645 			DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3646 					 mddev->sync_speed_min);
3647 		if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3648 			DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3649 					 mddev->sync_speed_max);
3650 		if (test_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags))
3651 			for (i = 0; i < rs->raid_disks; i++)
3652 				if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3653 					DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3654 					       rs->dev[i].rdev.raid_disk);
3655 		if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
3656 			DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
3657 					  mddev->bitmap_info.max_write_behind);
3658 		if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3659 			DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3660 					 max_nr_stripes);
3661 		if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3662 			DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3663 					   (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3664 		if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3665 			DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3666 					 raid10_md_layout_to_copies(mddev->layout));
3667 		if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3668 			DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3669 					 raid10_md_layout_to_format(mddev->layout));
3670 		if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3671 			DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3672 					 max(rs->delta_disks, mddev->delta_disks));
3673 		if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3674 			DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3675 					   (unsigned long long) rs->data_offset);
3676 		if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags))
3677 			DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV),
3678 					__get_dev_name(rs->journal_dev.dev));
3679 		if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags))
3680 			DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE),
3681 					 md_journal_mode_to_dm_raid(rs->journal_dev.mode));
3682 		DMEMIT(" %d", rs->raid_disks);
3683 		for (i = 0; i < rs->raid_disks; i++)
3684 			DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
3685 					 __get_dev_name(rs->dev[i].data_dev));
3686 		break;
3687 
3688 	case STATUSTYPE_IMA:
3689 		rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3690 		if (!rt)
3691 			return;
3692 
3693 		DMEMIT_TARGET_NAME_VERSION(ti->type);
3694 		DMEMIT(",raid_type=%s,raid_disks=%d", rt->name, mddev->raid_disks);
3695 
3696 		/* Access most recent mddev properties for status output */
3697 		smp_rmb();
3698 		recovery = rs->md.recovery;
3699 		state = decipher_sync_action(mddev, recovery);
3700 		DMEMIT(",raid_state=%s", sync_str(state));
3701 
3702 		for (i = 0; i < rs->raid_disks; i++) {
3703 			DMEMIT(",raid_device_%d_status=", i);
3704 			DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev));
3705 		}
3706 
3707 		if (rt_is_raid456(rt)) {
3708 			DMEMIT(",journal_dev_mode=");
3709 			switch (rs->journal_dev.mode) {
3710 			case R5C_JOURNAL_MODE_WRITE_THROUGH:
3711 				DMEMIT("%s",
3712 				       _raid456_journal_mode[R5C_JOURNAL_MODE_WRITE_THROUGH].param);
3713 				break;
3714 			case R5C_JOURNAL_MODE_WRITE_BACK:
3715 				DMEMIT("%s",
3716 				       _raid456_journal_mode[R5C_JOURNAL_MODE_WRITE_BACK].param);
3717 				break;
3718 			default:
3719 				DMEMIT("invalid");
3720 				break;
3721 			}
3722 		}
3723 		DMEMIT(";");
3724 		break;
3725 	}
3726 }
3727 
raid_message(struct dm_target * ti,unsigned int argc,char ** argv,char * result,unsigned int maxlen)3728 static int raid_message(struct dm_target *ti, unsigned int argc, char **argv,
3729 			char *result, unsigned int maxlen)
3730 {
3731 	struct raid_set *rs = ti->private;
3732 	struct mddev *mddev = &rs->md;
3733 	int ret = 0;
3734 
3735 	if (!mddev->pers || !mddev->pers->sync_request)
3736 		return -EINVAL;
3737 
3738 	if (test_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags) ||
3739 	    test_bit(RT_FLAG_RS_FROZEN, &rs->runtime_flags))
3740 		return -EBUSY;
3741 
3742 	if (!strcasecmp(argv[0], "frozen")) {
3743 		ret = mddev_lock(mddev);
3744 		if (ret)
3745 			return ret;
3746 
3747 		md_frozen_sync_thread(mddev);
3748 		mddev_unlock(mddev);
3749 	} else if (!strcasecmp(argv[0], "idle")) {
3750 		ret = mddev_lock(mddev);
3751 		if (ret)
3752 			return ret;
3753 
3754 		md_idle_sync_thread(mddev);
3755 		mddev_unlock(mddev);
3756 	}
3757 
3758 	clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3759 	if (decipher_sync_action(mddev, mddev->recovery) != st_idle)
3760 		return -EBUSY;
3761 	else if (!strcasecmp(argv[0], "resync"))
3762 		; /* MD_RECOVERY_NEEDED set below */
3763 	else if (!strcasecmp(argv[0], "recover"))
3764 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3765 	else {
3766 		if (!strcasecmp(argv[0], "check")) {
3767 			set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3768 			set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3769 			set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3770 		} else if (!strcasecmp(argv[0], "repair")) {
3771 			set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3772 			set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3773 		} else
3774 			return -EINVAL;
3775 	}
3776 	if (mddev->ro == 2) {
3777 		/* A write to sync_action is enough to justify
3778 		 * canceling read-auto mode
3779 		 */
3780 		mddev->ro = 0;
3781 		if (!mddev->suspended)
3782 			md_wakeup_thread(mddev->sync_thread);
3783 	}
3784 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3785 	if (!mddev->suspended)
3786 		md_wakeup_thread(mddev->thread);
3787 
3788 	return 0;
3789 }
3790 
raid_iterate_devices(struct dm_target * ti,iterate_devices_callout_fn fn,void * data)3791 static int raid_iterate_devices(struct dm_target *ti,
3792 				iterate_devices_callout_fn fn, void *data)
3793 {
3794 	struct raid_set *rs = ti->private;
3795 	unsigned int i;
3796 	int r = 0;
3797 
3798 	for (i = 0; !r && i < rs->raid_disks; i++) {
3799 		if (rs->dev[i].data_dev) {
3800 			r = fn(ti, rs->dev[i].data_dev,
3801 			       0, /* No offset on data devs */
3802 			       rs->md.dev_sectors, data);
3803 		}
3804 	}
3805 
3806 	return r;
3807 }
3808 
raid_io_hints(struct dm_target * ti,struct queue_limits * limits)3809 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3810 {
3811 	struct raid_set *rs = ti->private;
3812 	unsigned int chunk_size_bytes = to_bytes(rs->md.chunk_sectors);
3813 
3814 	limits->io_min = chunk_size_bytes;
3815 	limits->io_opt = chunk_size_bytes * mddev_data_stripes(rs);
3816 }
3817 
raid_presuspend(struct dm_target * ti)3818 static void raid_presuspend(struct dm_target *ti)
3819 {
3820 	struct raid_set *rs = ti->private;
3821 	struct mddev *mddev = &rs->md;
3822 
3823 	/*
3824 	 * From now on, disallow raid_message() to change sync_thread until
3825 	 * resume, raid_postsuspend() is too late.
3826 	 */
3827 	set_bit(RT_FLAG_RS_FROZEN, &rs->runtime_flags);
3828 
3829 	if (!reshape_interrupted(mddev))
3830 		return;
3831 
3832 	/*
3833 	 * For raid456, if reshape is interrupted, IO across reshape position
3834 	 * will never make progress, while caller will wait for IO to be done.
3835 	 * Inform raid456 to handle those IO to prevent deadlock.
3836 	 */
3837 	if (mddev->pers && mddev->pers->prepare_suspend)
3838 		mddev->pers->prepare_suspend(mddev);
3839 }
3840 
raid_presuspend_undo(struct dm_target * ti)3841 static void raid_presuspend_undo(struct dm_target *ti)
3842 {
3843 	struct raid_set *rs = ti->private;
3844 
3845 	clear_bit(RT_FLAG_RS_FROZEN, &rs->runtime_flags);
3846 }
3847 
raid_postsuspend(struct dm_target * ti)3848 static void raid_postsuspend(struct dm_target *ti)
3849 {
3850 	struct raid_set *rs = ti->private;
3851 
3852 	if (!test_and_set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
3853 		/*
3854 		 * sync_thread must be stopped during suspend, and writes have
3855 		 * to be stopped before suspending to avoid deadlocks.
3856 		 */
3857 		md_stop_writes(&rs->md);
3858 		mddev_suspend(&rs->md, false);
3859 	}
3860 }
3861 
attempt_restore_of_faulty_devices(struct raid_set * rs)3862 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
3863 {
3864 	int i;
3865 	uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
3866 	unsigned long flags;
3867 	bool cleared = false;
3868 	struct dm_raid_superblock *sb;
3869 	struct mddev *mddev = &rs->md;
3870 	struct md_rdev *r;
3871 
3872 	/* RAID personalities have to provide hot add/remove methods or we need to bail out. */
3873 	if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
3874 		return;
3875 
3876 	memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
3877 
3878 	for (i = 0; i < rs->raid_disks; i++) {
3879 		r = &rs->dev[i].rdev;
3880 		/* HM FIXME: enhance journal device recovery processing */
3881 		if (test_bit(Journal, &r->flags))
3882 			continue;
3883 
3884 		if (test_bit(Faulty, &r->flags) &&
3885 		    r->meta_bdev && !read_disk_sb(r, r->sb_size, true)) {
3886 			DMINFO("Faulty %s device #%d has readable super block."
3887 			       "  Attempting to revive it.",
3888 			       rs->raid_type->name, i);
3889 
3890 			/*
3891 			 * Faulty bit may be set, but sometimes the array can
3892 			 * be suspended before the personalities can respond
3893 			 * by removing the device from the array (i.e. calling
3894 			 * 'hot_remove_disk').	If they haven't yet removed
3895 			 * the failed device, its 'raid_disk' number will be
3896 			 * '>= 0' - meaning we must call this function
3897 			 * ourselves.
3898 			 */
3899 			flags = r->flags;
3900 			clear_bit(In_sync, &r->flags); /* Mandatory for hot remove. */
3901 			if (r->raid_disk >= 0) {
3902 				if (mddev->pers->hot_remove_disk(mddev, r)) {
3903 					/* Failed to revive this device, try next */
3904 					r->flags = flags;
3905 					continue;
3906 				}
3907 			} else
3908 				r->raid_disk = r->saved_raid_disk = i;
3909 
3910 			clear_bit(Faulty, &r->flags);
3911 			clear_bit(WriteErrorSeen, &r->flags);
3912 
3913 			if (mddev->pers->hot_add_disk(mddev, r)) {
3914 				/* Failed to revive this device, try next */
3915 				r->raid_disk = r->saved_raid_disk = -1;
3916 				r->flags = flags;
3917 			} else {
3918 				clear_bit(In_sync, &r->flags);
3919 				r->recovery_offset = 0;
3920 				set_bit(i, (void *) cleared_failed_devices);
3921 				cleared = true;
3922 			}
3923 		}
3924 	}
3925 
3926 	/* If any failed devices could be cleared, update all sbs failed_devices bits */
3927 	if (cleared) {
3928 		uint64_t failed_devices[DISKS_ARRAY_ELEMS];
3929 
3930 		rdev_for_each(r, &rs->md) {
3931 			if (test_bit(Journal, &r->flags))
3932 				continue;
3933 
3934 			sb = page_address(r->sb_page);
3935 			sb_retrieve_failed_devices(sb, failed_devices);
3936 
3937 			for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
3938 				failed_devices[i] &= ~cleared_failed_devices[i];
3939 
3940 			sb_update_failed_devices(sb, failed_devices);
3941 		}
3942 	}
3943 }
3944 
__load_dirty_region_bitmap(struct raid_set * rs)3945 static int __load_dirty_region_bitmap(struct raid_set *rs)
3946 {
3947 	int r = 0;
3948 
3949 	/* Try loading the bitmap unless "raid0", which does not have one */
3950 	if (!rs_is_raid0(rs) &&
3951 	    !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
3952 		struct mddev *mddev = &rs->md;
3953 
3954 		r = mddev->bitmap_ops->load(mddev);
3955 		if (r)
3956 			DMERR("Failed to load bitmap");
3957 	}
3958 
3959 	return r;
3960 }
3961 
3962 /* Enforce updating all superblocks */
rs_update_sbs(struct raid_set * rs)3963 static void rs_update_sbs(struct raid_set *rs)
3964 {
3965 	struct mddev *mddev = &rs->md;
3966 	int ro = mddev->ro;
3967 
3968 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3969 	mddev->ro = 0;
3970 	md_update_sb(mddev, 1);
3971 	mddev->ro = ro;
3972 }
3973 
3974 /*
3975  * Reshape changes raid algorithm of @rs to new one within personality
3976  * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3977  * disks from a raid set thus growing/shrinking it or resizes the set
3978  *
3979  * Call mddev_lock_nointr() before!
3980  */
rs_start_reshape(struct raid_set * rs)3981 static int rs_start_reshape(struct raid_set *rs)
3982 {
3983 	int r;
3984 	struct mddev *mddev = &rs->md;
3985 	struct md_personality *pers = mddev->pers;
3986 
3987 	/* Don't allow the sync thread to work until the table gets reloaded. */
3988 	set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
3989 
3990 	r = rs_setup_reshape(rs);
3991 	if (r)
3992 		return r;
3993 
3994 	/*
3995 	 * Check any reshape constraints enforced by the personalility
3996 	 *
3997 	 * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3998 	 */
3999 	r = pers->check_reshape(mddev);
4000 	if (r) {
4001 		rs->ti->error = "pers->check_reshape() failed";
4002 		return r;
4003 	}
4004 
4005 	/*
4006 	 * Personality may not provide start reshape method in which
4007 	 * case check_reshape above has already covered everything
4008 	 */
4009 	if (pers->start_reshape) {
4010 		r = pers->start_reshape(mddev);
4011 		if (r) {
4012 			rs->ti->error = "pers->start_reshape() failed";
4013 			return r;
4014 		}
4015 	}
4016 
4017 	/*
4018 	 * Now reshape got set up, update superblocks to
4019 	 * reflect the fact so that a table reload will
4020 	 * access proper superblock content in the ctr.
4021 	 */
4022 	rs_update_sbs(rs);
4023 
4024 	return 0;
4025 }
4026 
raid_preresume(struct dm_target * ti)4027 static int raid_preresume(struct dm_target *ti)
4028 {
4029 	int r;
4030 	struct raid_set *rs = ti->private;
4031 	struct mddev *mddev = &rs->md;
4032 
4033 	/* This is a resume after a suspend of the set -> it's already started. */
4034 	if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
4035 		return 0;
4036 
4037 	/* If different and no explicit grow request, expose MD array size as of superblock. */
4038 	if (!test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags) &&
4039 	    rs->array_sectors != mddev->array_sectors)
4040 		rs_set_capacity(rs);
4041 
4042 	/*
4043 	 * The superblocks need to be updated on disk if the
4044 	 * array is new or new devices got added (thus zeroed
4045 	 * out by userspace) or __load_dirty_region_bitmap
4046 	 * will overwrite them in core with old data or fail.
4047 	 */
4048 	if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
4049 		rs_update_sbs(rs);
4050 
4051 	/* Load the bitmap from disk unless raid0 */
4052 	r = __load_dirty_region_bitmap(rs);
4053 	if (r)
4054 		return r;
4055 
4056 	/* We are extending the raid set size, adjust mddev/md_rdev sizes and set capacity. */
4057 	if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags)) {
4058 		mddev->array_sectors = rs->array_sectors;
4059 		mddev->dev_sectors = rs->dev_sectors;
4060 		rs_set_rdev_sectors(rs);
4061 		rs_set_capacity(rs);
4062 	}
4063 
4064 	/* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) or grown device size */
4065 	if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) && mddev->bitmap &&
4066 	    (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags) ||
4067 	     (rs->requested_bitmap_chunk_sectors &&
4068 	       mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)))) {
4069 		int chunksize = to_bytes(rs->requested_bitmap_chunk_sectors) ?: mddev->bitmap_info.chunksize;
4070 
4071 		r = mddev->bitmap_ops->resize(mddev, mddev->dev_sectors,
4072 					      chunksize, false);
4073 		if (r)
4074 			DMERR("Failed to resize bitmap");
4075 	}
4076 
4077 	/* Check for any resize/reshape on @rs and adjust/initiate */
4078 	if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
4079 		set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4080 		mddev->resync_min = mddev->recovery_cp;
4081 		if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags))
4082 			mddev->resync_max_sectors = mddev->dev_sectors;
4083 	}
4084 
4085 	/* Check for any reshape request unless new raid set */
4086 	if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
4087 		/* Initiate a reshape. */
4088 		rs_set_rdev_sectors(rs);
4089 		mddev_lock_nointr(mddev);
4090 		r = rs_start_reshape(rs);
4091 		mddev_unlock(mddev);
4092 		if (r)
4093 			DMWARN("Failed to check/start reshape, continuing without change");
4094 		r = 0;
4095 	}
4096 
4097 	return r;
4098 }
4099 
raid_resume(struct dm_target * ti)4100 static void raid_resume(struct dm_target *ti)
4101 {
4102 	struct raid_set *rs = ti->private;
4103 	struct mddev *mddev = &rs->md;
4104 
4105 	if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
4106 		/*
4107 		 * A secondary resume while the device is active.
4108 		 * Take this opportunity to check whether any failed
4109 		 * devices are reachable again.
4110 		 */
4111 		mddev_lock_nointr(mddev);
4112 		attempt_restore_of_faulty_devices(rs);
4113 		mddev_unlock(mddev);
4114 	}
4115 
4116 	if (test_and_clear_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
4117 		/* Only reduce raid set size before running a disk removing reshape. */
4118 		if (mddev->delta_disks < 0)
4119 			rs_set_capacity(rs);
4120 
4121 		mddev_lock_nointr(mddev);
4122 		WARN_ON_ONCE(!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery));
4123 		WARN_ON_ONCE(rcu_dereference_protected(mddev->sync_thread,
4124 						       lockdep_is_held(&mddev->reconfig_mutex)));
4125 		clear_bit(RT_FLAG_RS_FROZEN, &rs->runtime_flags);
4126 		mddev->ro = 0;
4127 		mddev->in_sync = 0;
4128 		md_unfrozen_sync_thread(mddev);
4129 		mddev_unlock_and_resume(mddev);
4130 	}
4131 }
4132 
4133 static struct target_type raid_target = {
4134 	.name = "raid",
4135 	.version = {1, 15, 1},
4136 	.module = THIS_MODULE,
4137 	.ctr = raid_ctr,
4138 	.dtr = raid_dtr,
4139 	.map = raid_map,
4140 	.status = raid_status,
4141 	.message = raid_message,
4142 	.iterate_devices = raid_iterate_devices,
4143 	.io_hints = raid_io_hints,
4144 	.presuspend = raid_presuspend,
4145 	.presuspend_undo = raid_presuspend_undo,
4146 	.postsuspend = raid_postsuspend,
4147 	.preresume = raid_preresume,
4148 	.resume = raid_resume,
4149 };
4150 module_dm(raid);
4151 
4152 module_param(devices_handle_discard_safely, bool, 0644);
4153 MODULE_PARM_DESC(devices_handle_discard_safely,
4154 		 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
4155 
4156 MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
4157 MODULE_ALIAS("dm-raid0");
4158 MODULE_ALIAS("dm-raid1");
4159 MODULE_ALIAS("dm-raid10");
4160 MODULE_ALIAS("dm-raid4");
4161 MODULE_ALIAS("dm-raid5");
4162 MODULE_ALIAS("dm-raid6");
4163 MODULE_AUTHOR("Neil Brown <dm-devel@lists.linux.dev>");
4164 MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@lists.linux.dev>");
4165 MODULE_LICENSE("GPL");
4166