?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
x86/mm/pat: Don't report PAT on CPUs that don't support it
[linux/fpc-iii.git] / drivers / md / dm-raid.c
blob1e217ba84d090d1c51453c419c4cba62ba82892b
1 /*
2 * Copyright (C) 2010-2011 Neil Brown
3 * Copyright (C) 2010-2016 Red Hat, Inc. All rights reserved.
4 *
5 * This file is released under the GPL.
6 */
7
8 #include <linux/slab.h>
9 #include <linux/module.h>
10
11 #include "md.h"
12 #include "raid1.h"
13 #include "raid5.h"
14 #include "raid10.h"
15 #include "bitmap.h"
16
17 #include <linux/device-mapper.h>
18
19 #define DM_MSG_PREFIX "raid"
20 #define MAX_RAID_DEVICES 253 /* md-raid kernel limit */
21
22 /*
23 * Minimum sectors of free reshape space per raid device
24 */
25 #define MIN_FREE_RESHAPE_SPACE to_sector(4*4096)
26
27 /*
28 * Minimum journal space 4 MiB in sectors.
29 */
30 #define MIN_RAID456_JOURNAL_SPACE (4*2048)
31
32 static bool devices_handle_discard_safely = false;
33
34 /*
35 * The following flags are used by dm-raid.c to set up the array state.
36 * They must be cleared before md_run is called.
37 */
38 #define FirstUse 10 /* rdev flag */
39
40 struct raid_dev {
41 /*
42 * Two DM devices, one to hold metadata and one to hold the
43 * actual data/parity. The reason for this is to not confuse
44 * ti->len and give more flexibility in altering size and
45 * characteristics.
46 *
47 * While it is possible for this device to be associated
48 * with a different physical device than the data_dev, it
49 * is intended for it to be the same.
50 * |--------- Physical Device ---------|
51 * |- meta_dev -|------ data_dev ------|
52 */
53 struct dm_dev *meta_dev;
54 struct dm_dev *data_dev;
55 struct md_rdev rdev;
56 };
57
58 /*
59 * Bits for establishing rs->ctr_flags
60 *
61 * 1 = no flag value
62 * 2 = flag with value
63 */
64 #define __CTR_FLAG_SYNC 0 /* 1 */ /* Not with raid0! */
65 #define __CTR_FLAG_NOSYNC 1 /* 1 */ /* Not with raid0! */
66 #define __CTR_FLAG_REBUILD 2 /* 2 */ /* Not with raid0! */
67 #define __CTR_FLAG_DAEMON_SLEEP 3 /* 2 */ /* Not with raid0! */
68 #define __CTR_FLAG_MIN_RECOVERY_RATE 4 /* 2 */ /* Not with raid0! */
69 #define __CTR_FLAG_MAX_RECOVERY_RATE 5 /* 2 */ /* Not with raid0! */
70 #define __CTR_FLAG_MAX_WRITE_BEHIND 6 /* 2 */ /* Only with raid1! */
71 #define __CTR_FLAG_WRITE_MOSTLY 7 /* 2 */ /* Only with raid1! */
72 #define __CTR_FLAG_STRIPE_CACHE 8 /* 2 */ /* Only with raid4/5/6! */
73 #define __CTR_FLAG_REGION_SIZE 9 /* 2 */ /* Not with raid0! */
74 #define __CTR_FLAG_RAID10_COPIES 10 /* 2 */ /* Only with raid10 */
75 #define __CTR_FLAG_RAID10_FORMAT 11 /* 2 */ /* Only with raid10 */
76 /* New for v1.9.0 */
77 #define __CTR_FLAG_DELTA_DISKS 12 /* 2 */ /* Only with reshapable raid1/4/5/6/10! */
78 #define __CTR_FLAG_DATA_OFFSET 13 /* 2 */ /* Only with reshapable raid4/5/6/10! */
79 #define __CTR_FLAG_RAID10_USE_NEAR_SETS 14 /* 2 */ /* Only with raid10! */
80
81 /* New for v1.10.0 */
82 #define __CTR_FLAG_JOURNAL_DEV 15 /* 2 */ /* Only with raid4/5/6! */
83
84 /*
85 * Flags for rs->ctr_flags field.
86 */
87 #define CTR_FLAG_SYNC (1 << __CTR_FLAG_SYNC)
88 #define CTR_FLAG_NOSYNC (1 << __CTR_FLAG_NOSYNC)
89 #define CTR_FLAG_REBUILD (1 << __CTR_FLAG_REBUILD)
90 #define CTR_FLAG_DAEMON_SLEEP (1 << __CTR_FLAG_DAEMON_SLEEP)
91 #define CTR_FLAG_MIN_RECOVERY_RATE (1 << __CTR_FLAG_MIN_RECOVERY_RATE)
92 #define CTR_FLAG_MAX_RECOVERY_RATE (1 << __CTR_FLAG_MAX_RECOVERY_RATE)
93 #define CTR_FLAG_MAX_WRITE_BEHIND (1 << __CTR_FLAG_MAX_WRITE_BEHIND)
94 #define CTR_FLAG_WRITE_MOSTLY (1 << __CTR_FLAG_WRITE_MOSTLY)
95 #define CTR_FLAG_STRIPE_CACHE (1 << __CTR_FLAG_STRIPE_CACHE)
96 #define CTR_FLAG_REGION_SIZE (1 << __CTR_FLAG_REGION_SIZE)
97 #define CTR_FLAG_RAID10_COPIES (1 << __CTR_FLAG_RAID10_COPIES)
98 #define CTR_FLAG_RAID10_FORMAT (1 << __CTR_FLAG_RAID10_FORMAT)
99 #define CTR_FLAG_DELTA_DISKS (1 << __CTR_FLAG_DELTA_DISKS)
100 #define CTR_FLAG_DATA_OFFSET (1 << __CTR_FLAG_DATA_OFFSET)
101 #define CTR_FLAG_RAID10_USE_NEAR_SETS (1 << __CTR_FLAG_RAID10_USE_NEAR_SETS)
102 #define CTR_FLAG_JOURNAL_DEV (1 << __CTR_FLAG_JOURNAL_DEV)
103
104 #define RESUME_STAY_FROZEN_FLAGS (CTR_FLAG_DELTA_DISKS | CTR_FLAG_DATA_OFFSET)
105
106 /*
107 * Definitions of various constructor flags to
108 * be used in checks of valid / invalid flags
109 * per raid level.
110 */
111 /* Define all any sync flags */
112 #define CTR_FLAGS_ANY_SYNC (CTR_FLAG_SYNC | CTR_FLAG_NOSYNC)
113
114 /* Define flags for options without argument (e.g. 'nosync') */
115 #define CTR_FLAG_OPTIONS_NO_ARGS (CTR_FLAGS_ANY_SYNC | \
116 CTR_FLAG_RAID10_USE_NEAR_SETS)
117
118 /* Define flags for options with one argument (e.g. 'delta_disks +2') */
119 #define CTR_FLAG_OPTIONS_ONE_ARG (CTR_FLAG_REBUILD | \
120 CTR_FLAG_WRITE_MOSTLY | \
121 CTR_FLAG_DAEMON_SLEEP | \
122 CTR_FLAG_MIN_RECOVERY_RATE | \
123 CTR_FLAG_MAX_RECOVERY_RATE | \
124 CTR_FLAG_MAX_WRITE_BEHIND | \
125 CTR_FLAG_STRIPE_CACHE | \
126 CTR_FLAG_REGION_SIZE | \
127 CTR_FLAG_RAID10_COPIES | \
128 CTR_FLAG_RAID10_FORMAT | \
129 CTR_FLAG_DELTA_DISKS | \
130 CTR_FLAG_DATA_OFFSET)
131
132 /* Valid options definitions per raid level... */
133
134 /* "raid0" does only accept data offset */
135 #define RAID0_VALID_FLAGS (CTR_FLAG_DATA_OFFSET)
136
137 /* "raid1" does not accept stripe cache, data offset, delta_disks or any raid10 options */
138 #define RAID1_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
139 CTR_FLAG_REBUILD | \
140 CTR_FLAG_WRITE_MOSTLY | \
141 CTR_FLAG_DAEMON_SLEEP | \
142 CTR_FLAG_MIN_RECOVERY_RATE | \
143 CTR_FLAG_MAX_RECOVERY_RATE | \
144 CTR_FLAG_MAX_WRITE_BEHIND | \
145 CTR_FLAG_REGION_SIZE | \
146 CTR_FLAG_DELTA_DISKS | \
147 CTR_FLAG_DATA_OFFSET)
148
149 /* "raid10" does not accept any raid1 or stripe cache options */
150 #define RAID10_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
151 CTR_FLAG_REBUILD | \
152 CTR_FLAG_DAEMON_SLEEP | \
153 CTR_FLAG_MIN_RECOVERY_RATE | \
154 CTR_FLAG_MAX_RECOVERY_RATE | \
155 CTR_FLAG_REGION_SIZE | \
156 CTR_FLAG_RAID10_COPIES | \
157 CTR_FLAG_RAID10_FORMAT | \
158 CTR_FLAG_DELTA_DISKS | \
159 CTR_FLAG_DATA_OFFSET | \
160 CTR_FLAG_RAID10_USE_NEAR_SETS)
161
162 /*
163 * "raid4/5/6" do not accept any raid1 or raid10 specific options
164 *
165 * "raid6" does not accept "nosync", because it is not guaranteed
166 * that both parity and q-syndrome are being written properly with
167 * any writes
168 */
169 #define RAID45_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
170 CTR_FLAG_REBUILD | \
171 CTR_FLAG_DAEMON_SLEEP | \
172 CTR_FLAG_MIN_RECOVERY_RATE | \
173 CTR_FLAG_MAX_RECOVERY_RATE | \
174 CTR_FLAG_STRIPE_CACHE | \
175 CTR_FLAG_REGION_SIZE | \
176 CTR_FLAG_DELTA_DISKS | \
177 CTR_FLAG_DATA_OFFSET | \
178 CTR_FLAG_JOURNAL_DEV)
179
180 #define RAID6_VALID_FLAGS (CTR_FLAG_SYNC | \
181 CTR_FLAG_REBUILD | \
182 CTR_FLAG_DAEMON_SLEEP | \
183 CTR_FLAG_MIN_RECOVERY_RATE | \
184 CTR_FLAG_MAX_RECOVERY_RATE | \
185 CTR_FLAG_STRIPE_CACHE | \
186 CTR_FLAG_REGION_SIZE | \
187 CTR_FLAG_DELTA_DISKS | \
188 CTR_FLAG_DATA_OFFSET | \
189 CTR_FLAG_JOURNAL_DEV)
190 /* ...valid options definitions per raid level */
191
192 /*
193 * Flags for rs->runtime_flags field
194 * (RT_FLAG prefix meaning "runtime flag")
195 *
196 * These are all internal and used to define runtime state,
197 * e.g. to prevent another resume from preresume processing
198 * the raid set all over again.
199 */
200 #define RT_FLAG_RS_PRERESUMED 0
201 #define RT_FLAG_RS_RESUMED 1
202 #define RT_FLAG_RS_BITMAP_LOADED 2
203 #define RT_FLAG_UPDATE_SBS 3
204 #define RT_FLAG_RESHAPE_RS 4
205
206 /* Array elements of 64 bit needed for rebuild/failed disk bits */
207 #define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
208
209 /*
210 * raid set level, layout and chunk sectors backup/restore
211 */
212 struct rs_layout {
213 int new_level;
214 int new_layout;
215 int new_chunk_sectors;
216 };
217
218 struct raid_set {
219 struct dm_target *ti;
220
221 uint32_t bitmap_loaded;
222 uint32_t stripe_cache_entries;
223 unsigned long ctr_flags;
224 unsigned long runtime_flags;
225
226 uint64_t rebuild_disks[DISKS_ARRAY_ELEMS];
227
228 int raid_disks;
229 int delta_disks;
230 int data_offset;
231 int raid10_copies;
232 int requested_bitmap_chunk_sectors;
233
234 struct mddev md;
235 struct raid_type *raid_type;
236 struct dm_target_callbacks callbacks;
237
238 /* Optional raid4/5/6 journal device */
239 struct journal_dev {
240 struct dm_dev *dev;
241 struct md_rdev rdev;
242 } journal_dev;
243
244 struct raid_dev dev[0];
245 };
246
247 static void rs_config_backup(struct raid_set *rs, struct rs_layout *l)
248 {
249 struct mddev *mddev = &rs->md;
250
251 l->new_level = mddev->new_level;
252 l->new_layout = mddev->new_layout;
253 l->new_chunk_sectors = mddev->new_chunk_sectors;
254 }
255
256 static void rs_config_restore(struct raid_set *rs, struct rs_layout *l)
257 {
258 struct mddev *mddev = &rs->md;
259
260 mddev->new_level = l->new_level;
261 mddev->new_layout = l->new_layout;
262 mddev->new_chunk_sectors = l->new_chunk_sectors;
263 }
264
265 /* raid10 algorithms (i.e. formats) */
266 #define ALGORITHM_RAID10_DEFAULT 0
267 #define ALGORITHM_RAID10_NEAR 1
268 #define ALGORITHM_RAID10_OFFSET 2
269 #define ALGORITHM_RAID10_FAR 3
270
271 /* Supported raid types and properties. */
272 static struct raid_type {
273 const char *name; /* RAID algorithm. */
274 const char *descr; /* Descriptor text for logging. */
275 const unsigned int parity_devs; /* # of parity devices. */
276 const unsigned int minimal_devs;/* minimal # of devices in set. */
277 const unsigned int level; /* RAID level. */
278 const unsigned int algorithm; /* RAID algorithm. */
279 } raid_types[] = {
280 {"raid0", "raid0 (striping)", 0, 2, 0, 0 /* NONE */},
281 {"raid1", "raid1 (mirroring)", 0, 2, 1, 0 /* NONE */},
282 {"raid10_far", "raid10 far (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_FAR},
283 {"raid10_offset", "raid10 offset (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_OFFSET},
284 {"raid10_near", "raid10 near (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_NEAR},
285 {"raid10", "raid10 (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_DEFAULT},
286 {"raid4", "raid4 (dedicated first parity disk)", 1, 2, 5, ALGORITHM_PARITY_0}, /* raid4 layout = raid5_0 */
287 {"raid5_n", "raid5 (dedicated last parity disk)", 1, 2, 5, ALGORITHM_PARITY_N},
288 {"raid5_ls", "raid5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
289 {"raid5_rs", "raid5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
290 {"raid5_la", "raid5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
291 {"raid5_ra", "raid5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
292 {"raid6_zr", "raid6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
293 {"raid6_nr", "raid6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
294 {"raid6_nc", "raid6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE},
295 {"raid6_n_6", "raid6 (dedicated parity/Q n/6)", 2, 4, 6, ALGORITHM_PARITY_N_6},
296 {"raid6_ls_6", "raid6 (left symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_SYMMETRIC_6},
297 {"raid6_rs_6", "raid6 (right symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_SYMMETRIC_6},
298 {"raid6_la_6", "raid6 (left asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_ASYMMETRIC_6},
299 {"raid6_ra_6", "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_ASYMMETRIC_6}
300 };
301
302 /* True, if @v is in inclusive range [@min, @max] */
303 static bool __within_range(long v, long min, long max)
304 {
305 return v >= min && v <= max;
306 }
307
308 /* All table line arguments are defined here */
309 static struct arg_name_flag {
310 const unsigned long flag;
311 const char *name;
312 } __arg_name_flags[] = {
313 { CTR_FLAG_SYNC, "sync"},
314 { CTR_FLAG_NOSYNC, "nosync"},
315 { CTR_FLAG_REBUILD, "rebuild"},
316 { CTR_FLAG_DAEMON_SLEEP, "daemon_sleep"},
317 { CTR_FLAG_MIN_RECOVERY_RATE, "min_recovery_rate"},
318 { CTR_FLAG_MAX_RECOVERY_RATE, "max_recovery_rate"},
319 { CTR_FLAG_MAX_WRITE_BEHIND, "max_write_behind"},
320 { CTR_FLAG_WRITE_MOSTLY, "write_mostly"},
321 { CTR_FLAG_STRIPE_CACHE, "stripe_cache"},
322 { CTR_FLAG_REGION_SIZE, "region_size"},
323 { CTR_FLAG_RAID10_COPIES, "raid10_copies"},
324 { CTR_FLAG_RAID10_FORMAT, "raid10_format"},
325 { CTR_FLAG_DATA_OFFSET, "data_offset"},
326 { CTR_FLAG_DELTA_DISKS, "delta_disks"},
327 { CTR_FLAG_RAID10_USE_NEAR_SETS, "raid10_use_near_sets"},
328 { CTR_FLAG_JOURNAL_DEV, "journal_dev" },
329 };
330
331 /* Return argument name string for given @flag */
332 static const char *dm_raid_arg_name_by_flag(const uint32_t flag)
333 {
334 if (hweight32(flag) == 1) {
335 struct arg_name_flag *anf = __arg_name_flags + ARRAY_SIZE(__arg_name_flags);
336
337 while (anf-- > __arg_name_flags)
338 if (flag & anf->flag)
339 return anf->name;
340
341 } else
342 DMERR("%s called with more than one flag!", __func__);
343
344 return NULL;
345 }
346
347 /*
348 * Bool helpers to test for various raid levels of a raid set.
349 * It's level as reported by the superblock rather than
350 * the requested raid_type passed to the constructor.
351 */
352 /* Return true, if raid set in @rs is raid0 */
353 static bool rs_is_raid0(struct raid_set *rs)
354 {
355 return !rs->md.level;
356 }
357
358 /* Return true, if raid set in @rs is raid1 */
359 static bool rs_is_raid1(struct raid_set *rs)
360 {
361 return rs->md.level == 1;
362 }
363
364 /* Return true, if raid set in @rs is raid10 */
365 static bool rs_is_raid10(struct raid_set *rs)
366 {
367 return rs->md.level == 10;
368 }
369
370 /* Return true, if raid set in @rs is level 6 */
371 static bool rs_is_raid6(struct raid_set *rs)
372 {
373 return rs->md.level == 6;
374 }
375
376 /* Return true, if raid set in @rs is level 4, 5 or 6 */
377 static bool rs_is_raid456(struct raid_set *rs)
378 {
379 return __within_range(rs->md.level, 4, 6);
380 }
381
382 /* Return true, if raid set in @rs is reshapable */
383 static bool __is_raid10_far(int layout);
384 static bool rs_is_reshapable(struct raid_set *rs)
385 {
386 return rs_is_raid456(rs) ||
387 (rs_is_raid10(rs) && !__is_raid10_far(rs->md.new_layout));
388 }
389
390 /* Return true, if raid set in @rs is recovering */
391 static bool rs_is_recovering(struct raid_set *rs)
392 {
393 return rs->md.recovery_cp < rs->md.dev_sectors;
394 }
395
396 /* Return true, if raid set in @rs is reshaping */
397 static bool rs_is_reshaping(struct raid_set *rs)
398 {
399 return rs->md.reshape_position != MaxSector;
400 }
401
402 /*
403 * bool helpers to test for various raid levels of a raid type @rt
404 */
405
406 /* Return true, if raid type in @rt is raid0 */
407 static bool rt_is_raid0(struct raid_type *rt)
408 {
409 return !rt->level;
410 }
411
412 /* Return true, if raid type in @rt is raid1 */
413 static bool rt_is_raid1(struct raid_type *rt)
414 {
415 return rt->level == 1;
416 }
417
418 /* Return true, if raid type in @rt is raid10 */
419 static bool rt_is_raid10(struct raid_type *rt)
420 {
421 return rt->level == 10;
422 }
423
424 /* Return true, if raid type in @rt is raid4/5 */
425 static bool rt_is_raid45(struct raid_type *rt)
426 {
427 return __within_range(rt->level, 4, 5);
428 }
429
430 /* Return true, if raid type in @rt is raid6 */
431 static bool rt_is_raid6(struct raid_type *rt)
432 {
433 return rt->level == 6;
434 }
435
436 /* Return true, if raid type in @rt is raid4/5/6 */
437 static bool rt_is_raid456(struct raid_type *rt)
438 {
439 return __within_range(rt->level, 4, 6);
440 }
441 /* END: raid level bools */
442
443 /* Return valid ctr flags for the raid level of @rs */
444 static unsigned long __valid_flags(struct raid_set *rs)
445 {
446 if (rt_is_raid0(rs->raid_type))
447 return RAID0_VALID_FLAGS;
448 else if (rt_is_raid1(rs->raid_type))
449 return RAID1_VALID_FLAGS;
450 else if (rt_is_raid10(rs->raid_type))
451 return RAID10_VALID_FLAGS;
452 else if (rt_is_raid45(rs->raid_type))
453 return RAID45_VALID_FLAGS;
454 else if (rt_is_raid6(rs->raid_type))
455 return RAID6_VALID_FLAGS;
456
457 return 0;
458 }
459
460 /*
461 * Check for valid flags set on @rs
462 *
463 * Has to be called after parsing of the ctr flags!
464 */
465 static int rs_check_for_valid_flags(struct raid_set *rs)
466 {
467 if (rs->ctr_flags & ~__valid_flags(rs)) {
468 rs->ti->error = "Invalid flags combination";
469 return -EINVAL;
470 }
471
472 return 0;
473 }
474
475 /* MD raid10 bit definitions and helpers */
476 #define RAID10_OFFSET (1 << 16) /* stripes with data copies area adjacent on devices */
477 #define RAID10_BROCKEN_USE_FAR_SETS (1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
478 #define RAID10_USE_FAR_SETS (1 << 18) /* Use sets instead of whole stripe rotation */
479 #define RAID10_FAR_COPIES_SHIFT 8 /* raid10 # far copies shift (2nd byte of layout) */
480
481 /* Return md raid10 near copies for @layout */
482 static unsigned int __raid10_near_copies(int layout)
483 {
484 return layout & 0xFF;
485 }
486
487 /* Return md raid10 far copies for @layout */
488 static unsigned int __raid10_far_copies(int layout)
489 {
490 return __raid10_near_copies(layout >> RAID10_FAR_COPIES_SHIFT);
491 }
492
493 /* Return true if md raid10 offset for @layout */
494 static bool __is_raid10_offset(int layout)
495 {
496 return !!(layout & RAID10_OFFSET);
497 }
498
499 /* Return true if md raid10 near for @layout */
500 static bool __is_raid10_near(int layout)
501 {
502 return !__is_raid10_offset(layout) && __raid10_near_copies(layout) > 1;
503 }
504
505 /* Return true if md raid10 far for @layout */
506 static bool __is_raid10_far(int layout)
507 {
508 return !__is_raid10_offset(layout) && __raid10_far_copies(layout) > 1;
509 }
510
511 /* Return md raid10 layout string for @layout */
512 static const char *raid10_md_layout_to_format(int layout)
513 {
514 /*
515 * Bit 16 stands for "offset"
516 * (i.e. adjacent stripes hold copies)
517 *
518 * Refer to MD's raid10.c for details
519 */
520 if (__is_raid10_offset(layout))
521 return "offset";
522
523 if (__raid10_near_copies(layout) > 1)
524 return "near";
525
526 WARN_ON(__raid10_far_copies(layout) < 2);
527
528 return "far";
529 }
530
531 /* Return md raid10 algorithm for @name */
532 static int raid10_name_to_format(const char *name)
533 {
534 if (!strcasecmp(name, "near"))
535 return ALGORITHM_RAID10_NEAR;
536 else if (!strcasecmp(name, "offset"))
537 return ALGORITHM_RAID10_OFFSET;
538 else if (!strcasecmp(name, "far"))
539 return ALGORITHM_RAID10_FAR;
540
541 return -EINVAL;
542 }
543
544 /* Return md raid10 copies for @layout */
545 static unsigned int raid10_md_layout_to_copies(int layout)
546 {
547 return max(__raid10_near_copies(layout), __raid10_far_copies(layout));
548 }
549
550 /* Return md raid10 format id for @format string */
551 static int raid10_format_to_md_layout(struct raid_set *rs,
552 unsigned int algorithm,
553 unsigned int copies)
554 {
555 unsigned int n = 1, f = 1, r = 0;
556
557 /*
558 * MD resilienece flaw:
559 *
560 * enabling use_far_sets for far/offset formats causes copies
561 * to be colocated on the same devs together with their origins!
562 *
563 * -> disable it for now in the definition above
564 */
565 if (algorithm == ALGORITHM_RAID10_DEFAULT ||
566 algorithm == ALGORITHM_RAID10_NEAR)
567 n = copies;
568
569 else if (algorithm == ALGORITHM_RAID10_OFFSET) {
570 f = copies;
571 r = RAID10_OFFSET;
572 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
573 r |= RAID10_USE_FAR_SETS;
574
575 } else if (algorithm == ALGORITHM_RAID10_FAR) {
576 f = copies;
577 r = !RAID10_OFFSET;
578 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
579 r |= RAID10_USE_FAR_SETS;
580
581 } else
582 return -EINVAL;
583
584 return r | (f << RAID10_FAR_COPIES_SHIFT) | n;
585 }
586 /* END: MD raid10 bit definitions and helpers */
587
588 /* Check for any of the raid10 algorithms */
589 static bool __got_raid10(struct raid_type *rtp, const int layout)
590 {
591 if (rtp->level == 10) {
592 switch (rtp->algorithm) {
593 case ALGORITHM_RAID10_DEFAULT:
594 case ALGORITHM_RAID10_NEAR:
595 return __is_raid10_near(layout);
596 case ALGORITHM_RAID10_OFFSET:
597 return __is_raid10_offset(layout);
598 case ALGORITHM_RAID10_FAR:
599 return __is_raid10_far(layout);
600 default:
601 break;
602 }
603 }
604
605 return false;
606 }
607
608 /* Return raid_type for @name */
609 static struct raid_type *get_raid_type(const char *name)
610 {
611 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
612
613 while (rtp-- > raid_types)
614 if (!strcasecmp(rtp->name, name))
615 return rtp;
616
617 return NULL;
618 }
619
620 /* Return raid_type for @name based derived from @level and @layout */
621 static struct raid_type *get_raid_type_by_ll(const int level, const int layout)
622 {
623 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
624
625 while (rtp-- > raid_types) {
626 /* RAID10 special checks based on @layout flags/properties */
627 if (rtp->level == level &&
628 (__got_raid10(rtp, layout) || rtp->algorithm == layout))
629 return rtp;
630 }
631
632 return NULL;
633 }
634
635 /*
636 * Conditionally change bdev capacity of @rs
637 * in case of a disk add/remove reshape
638 */
639 static void rs_set_capacity(struct raid_set *rs)
640 {
641 struct mddev *mddev = &rs->md;
642 struct md_rdev *rdev;
643 struct gendisk *gendisk = dm_disk(dm_table_get_md(rs->ti->table));
644
645 /*
646 * raid10 sets rdev->sector to the device size, which
647 * is unintended in case of out-of-place reshaping
648 */
649 rdev_for_each(rdev, mddev)
650 if (!test_bit(Journal, &rdev->flags))
651 rdev->sectors = mddev->dev_sectors;
652
653 set_capacity(gendisk, mddev->array_sectors);
654 revalidate_disk(gendisk);
655 }
656
657 /*
658 * Set the mddev properties in @rs to the current
659 * ones retrieved from the freshest superblock
660 */
661 static void rs_set_cur(struct raid_set *rs)
662 {
663 struct mddev *mddev = &rs->md;
664
665 mddev->new_level = mddev->level;
666 mddev->new_layout = mddev->layout;
667 mddev->new_chunk_sectors = mddev->chunk_sectors;
668 }
669
670 /*
671 * Set the mddev properties in @rs to the new
672 * ones requested by the ctr
673 */
674 static void rs_set_new(struct raid_set *rs)
675 {
676 struct mddev *mddev = &rs->md;
677
678 mddev->level = mddev->new_level;
679 mddev->layout = mddev->new_layout;
680 mddev->chunk_sectors = mddev->new_chunk_sectors;
681 mddev->raid_disks = rs->raid_disks;
682 mddev->delta_disks = 0;
683 }
684
685 static struct raid_set *raid_set_alloc(struct dm_target *ti, struct raid_type *raid_type,
686 unsigned int raid_devs)
687 {
688 unsigned int i;
689 struct raid_set *rs;
690
691 if (raid_devs <= raid_type->parity_devs) {
692 ti->error = "Insufficient number of devices";
693 return ERR_PTR(-EINVAL);
694 }
695
696 rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
697 if (!rs) {
698 ti->error = "Cannot allocate raid context";
699 return ERR_PTR(-ENOMEM);
700 }
701
702 mddev_init(&rs->md);
703
704 rs->raid_disks = raid_devs;
705 rs->delta_disks = 0;
706
707 rs->ti = ti;
708 rs->raid_type = raid_type;
709 rs->stripe_cache_entries = 256;
710 rs->md.raid_disks = raid_devs;
711 rs->md.level = raid_type->level;
712 rs->md.new_level = rs->md.level;
713 rs->md.layout = raid_type->algorithm;
714 rs->md.new_layout = rs->md.layout;
715 rs->md.delta_disks = 0;
716 rs->md.recovery_cp = MaxSector;
717
718 for (i = 0; i < raid_devs; i++)
719 md_rdev_init(&rs->dev[i].rdev);
720
721 /*
722 * Remaining items to be initialized by further RAID params:
723 * rs->md.persistent
724 * rs->md.external
725 * rs->md.chunk_sectors
726 * rs->md.new_chunk_sectors
727 * rs->md.dev_sectors
728 */
729
730 return rs;
731 }
732
733 static void raid_set_free(struct raid_set *rs)
734 {
735 int i;
736
737 if (rs->journal_dev.dev) {
738 md_rdev_clear(&rs->journal_dev.rdev);
739 dm_put_device(rs->ti, rs->journal_dev.dev);
740 }
741
742 for (i = 0; i < rs->raid_disks; i++) {
743 if (rs->dev[i].meta_dev)
744 dm_put_device(rs->ti, rs->dev[i].meta_dev);
745 md_rdev_clear(&rs->dev[i].rdev);
746 if (rs->dev[i].data_dev)
747 dm_put_device(rs->ti, rs->dev[i].data_dev);
748 }
749
750 kfree(rs);
751 }
752
753 /*
754 * For every device we have two words
755 * <meta_dev>: meta device name or '-' if missing
756 * <data_dev>: data device name or '-' if missing
757 *
758 * The following are permitted:
759 * - -
760 * - <data_dev>
761 * <meta_dev> <data_dev>
762 *
763 * The following is not allowed:
764 * <meta_dev> -
765 *
766 * This code parses those words. If there is a failure,
767 * the caller must use raid_set_free() to unwind the operations.
768 */
769 static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
770 {
771 int i;
772 int rebuild = 0;
773 int metadata_available = 0;
774 int r = 0;
775 const char *arg;
776
777 /* Put off the number of raid devices argument to get to dev pairs */
778 arg = dm_shift_arg(as);
779 if (!arg)
780 return -EINVAL;
781
782 for (i = 0; i < rs->raid_disks; i++) {
783 rs->dev[i].rdev.raid_disk = i;
784
785 rs->dev[i].meta_dev = NULL;
786 rs->dev[i].data_dev = NULL;
787
788 /*
789 * There are no offsets initially.
790 * Out of place reshape will set them accordingly.
791 */
792 rs->dev[i].rdev.data_offset = 0;
793 rs->dev[i].rdev.new_data_offset = 0;
794 rs->dev[i].rdev.mddev = &rs->md;
795
796 arg = dm_shift_arg(as);
797 if (!arg)
798 return -EINVAL;
799
800 if (strcmp(arg, "-")) {
801 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
802 &rs->dev[i].meta_dev);
803 if (r) {
804 rs->ti->error = "RAID metadata device lookup failure";
805 return r;
806 }
807
808 rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
809 if (!rs->dev[i].rdev.sb_page) {
810 rs->ti->error = "Failed to allocate superblock page";
811 return -ENOMEM;
812 }
813 }
814
815 arg = dm_shift_arg(as);
816 if (!arg)
817 return -EINVAL;
818
819 if (!strcmp(arg, "-")) {
820 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
821 (!rs->dev[i].rdev.recovery_offset)) {
822 rs->ti->error = "Drive designated for rebuild not specified";
823 return -EINVAL;
824 }
825
826 if (rs->dev[i].meta_dev) {
827 rs->ti->error = "No data device supplied with metadata device";
828 return -EINVAL;
829 }
830
831 continue;
832 }
833
834 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
835 &rs->dev[i].data_dev);
836 if (r) {
837 rs->ti->error = "RAID device lookup failure";
838 return r;
839 }
840
841 if (rs->dev[i].meta_dev) {
842 metadata_available = 1;
843 rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
844 }
845 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
846 list_add_tail(&rs->dev[i].rdev.same_set, &rs->md.disks);
847 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
848 rebuild++;
849 }
850
851 if (rs->journal_dev.dev)
852 list_add_tail(&rs->journal_dev.rdev.same_set, &rs->md.disks);
853
854 if (metadata_available) {
855 rs->md.external = 0;
856 rs->md.persistent = 1;
857 rs->md.major_version = 2;
858 } else if (rebuild && !rs->md.recovery_cp) {
859 /*
860 * Without metadata, we will not be able to tell if the array
861 * is in-sync or not - we must assume it is not. Therefore,
862 * it is impossible to rebuild a drive.
863 *
864 * Even if there is metadata, the on-disk information may
865 * indicate that the array is not in-sync and it will then
866 * fail at that time.
867 *
868 * User could specify 'nosync' option if desperate.
869 */
870 rs->ti->error = "Unable to rebuild drive while array is not in-sync";
871 return -EINVAL;
872 }
873
874 return 0;
875 }
876
877 /*
878 * validate_region_size
879 * @rs
880 * @region_size: region size in sectors. If 0, pick a size (4MiB default).
881 *
882 * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
883 * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
884 *
885 * Returns: 0 on success, -EINVAL on failure.
886 */
887 static int validate_region_size(struct raid_set *rs, unsigned long region_size)
888 {
889 unsigned long min_region_size = rs->ti->len / (1 << 21);
890
891 if (rs_is_raid0(rs))
892 return 0;
893
894 if (!region_size) {
895 /*
896 * Choose a reasonable default. All figures in sectors.
897 */
898 if (min_region_size > (1 << 13)) {
899 /* If not a power of 2, make it the next power of 2 */
900 region_size = roundup_pow_of_two(min_region_size);
901 DMINFO("Choosing default region size of %lu sectors",
902 region_size);
903 } else {
904 DMINFO("Choosing default region size of 4MiB");
905 region_size = 1 << 13; /* sectors */
906 }
907 } else {
908 /*
909 * Validate user-supplied value.
910 */
911 if (region_size > rs->ti->len) {
912 rs->ti->error = "Supplied region size is too large";
913 return -EINVAL;
914 }
915
916 if (region_size < min_region_size) {
917 DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
918 region_size, min_region_size);
919 rs->ti->error = "Supplied region size is too small";
920 return -EINVAL;
921 }
922
923 if (!is_power_of_2(region_size)) {
924 rs->ti->error = "Region size is not a power of 2";
925 return -EINVAL;
926 }
927
928 if (region_size < rs->md.chunk_sectors) {
929 rs->ti->error = "Region size is smaller than the chunk size";
930 return -EINVAL;
931 }
932 }
933
934 /*
935 * Convert sectors to bytes.
936 */
937 rs->md.bitmap_info.chunksize = to_bytes(region_size);
938
939 return 0;
940 }
941
942 /*
943 * validate_raid_redundancy
944 * @rs
945 *
946 * Determine if there are enough devices in the array that haven't
947 * failed (or are being rebuilt) to form a usable array.
948 *
949 * Returns: 0 on success, -EINVAL on failure.
950 */
951 static int validate_raid_redundancy(struct raid_set *rs)
952 {
953 unsigned int i, rebuild_cnt = 0;
954 unsigned int rebuilds_per_group = 0, copies;
955 unsigned int group_size, last_group_start;
956
957 for (i = 0; i < rs->md.raid_disks; i++)
958 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
959 !rs->dev[i].rdev.sb_page)
960 rebuild_cnt++;
961
962 switch (rs->raid_type->level) {
963 case 0:
964 break;
965 case 1:
966 if (rebuild_cnt >= rs->md.raid_disks)
967 goto too_many;
968 break;
969 case 4:
970 case 5:
971 case 6:
972 if (rebuild_cnt > rs->raid_type->parity_devs)
973 goto too_many;
974 break;
975 case 10:
976 copies = raid10_md_layout_to_copies(rs->md.new_layout);
977 if (rebuild_cnt < copies)
978 break;
979
980 /*
981 * It is possible to have a higher rebuild count for RAID10,
982 * as long as the failed devices occur in different mirror
983 * groups (i.e. different stripes).
984 *
985 * When checking "near" format, make sure no adjacent devices
986 * have failed beyond what can be handled. In addition to the
987 * simple case where the number of devices is a multiple of the
988 * number of copies, we must also handle cases where the number
989 * of devices is not a multiple of the number of copies.
990 * E.g. dev1 dev2 dev3 dev4 dev5
991 * A A B B C
992 * C D D E E
993 */
994 if (__is_raid10_near(rs->md.new_layout)) {
995 for (i = 0; i < rs->md.raid_disks; i++) {
996 if (!(i % copies))
997 rebuilds_per_group = 0;
998 if ((!rs->dev[i].rdev.sb_page ||
999 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1000 (++rebuilds_per_group >= copies))
1001 goto too_many;
1002 }
1003 break;
1004 }
1005
1006 /*
1007 * When checking "far" and "offset" formats, we need to ensure
1008 * that the device that holds its copy is not also dead or
1009 * being rebuilt. (Note that "far" and "offset" formats only
1010 * support two copies right now. These formats also only ever
1011 * use the 'use_far_sets' variant.)
1012 *
1013 * This check is somewhat complicated by the need to account
1014 * for arrays that are not a multiple of (far) copies. This
1015 * results in the need to treat the last (potentially larger)
1016 * set differently.
1017 */
1018 group_size = (rs->md.raid_disks / copies);
1019 last_group_start = (rs->md.raid_disks / group_size) - 1;
1020 last_group_start *= group_size;
1021 for (i = 0; i < rs->md.raid_disks; i++) {
1022 if (!(i % copies) && !(i > last_group_start))
1023 rebuilds_per_group = 0;
1024 if ((!rs->dev[i].rdev.sb_page ||
1025 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1026 (++rebuilds_per_group >= copies))
1027 goto too_many;
1028 }
1029 break;
1030 default:
1031 if (rebuild_cnt)
1032 return -EINVAL;
1033 }
1034
1035 return 0;
1036
1037 too_many:
1038 return -EINVAL;
1039 }
1040
1041 /*
1042 * Possible arguments are...
1043 * <chunk_size> [optional_args]
1044 *
1045 * Argument definitions
1046 * <chunk_size> The number of sectors per disk that
1047 * will form the "stripe"
1048 * [[no]sync] Force or prevent recovery of the
1049 * entire array
1050 * [rebuild <idx>] Rebuild the drive indicated by the index
1051 * [daemon_sleep <ms>] Time between bitmap daemon work to
1052 * clear bits
1053 * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1054 * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1055 * [write_mostly <idx>] Indicate a write mostly drive via index
1056 * [max_write_behind <sectors>] See '-write-behind=' (man mdadm)
1057 * [stripe_cache <sectors>] Stripe cache size for higher RAIDs
1058 * [region_size <sectors>] Defines granularity of bitmap
1059 * [journal_dev <dev>] raid4/5/6 journaling deviice
1060 * (i.e. write hole closing log)
1061 *
1062 * RAID10-only options:
1063 * [raid10_copies <# copies>] Number of copies. (Default: 2)
1064 * [raid10_format <near|far|offset>] Layout algorithm. (Default: near)
1065 */
1066 static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
1067 unsigned int num_raid_params)
1068 {
1069 int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
1070 unsigned int raid10_copies = 2;
1071 unsigned int i, write_mostly = 0;
1072 unsigned int region_size = 0;
1073 sector_t max_io_len;
1074 const char *arg, *key;
1075 struct raid_dev *rd;
1076 struct raid_type *rt = rs->raid_type;
1077
1078 arg = dm_shift_arg(as);
1079 num_raid_params--; /* Account for chunk_size argument */
1080
1081 if (kstrtoint(arg, 10, &value) < 0) {
1082 rs->ti->error = "Bad numerical argument given for chunk_size";
1083 return -EINVAL;
1084 }
1085
1086 /*
1087 * First, parse the in-order required arguments
1088 * "chunk_size" is the only argument of this type.
1089 */
1090 if (rt_is_raid1(rt)) {
1091 if (value)
1092 DMERR("Ignoring chunk size parameter for RAID 1");
1093 value = 0;
1094 } else if (!is_power_of_2(value)) {
1095 rs->ti->error = "Chunk size must be a power of 2";
1096 return -EINVAL;
1097 } else if (value < 8) {
1098 rs->ti->error = "Chunk size value is too small";
1099 return -EINVAL;
1100 }
1101
1102 rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
1103
1104 /*
1105 * We set each individual device as In_sync with a completed
1106 * 'recovery_offset'. If there has been a device failure or
1107 * replacement then one of the following cases applies:
1108 *
1109 * 1) User specifies 'rebuild'.
1110 * - Device is reset when param is read.
1111 * 2) A new device is supplied.
1112 * - No matching superblock found, resets device.
1113 * 3) Device failure was transient and returns on reload.
1114 * - Failure noticed, resets device for bitmap replay.
1115 * 4) Device hadn't completed recovery after previous failure.
1116 * - Superblock is read and overrides recovery_offset.
1117 *
1118 * What is found in the superblocks of the devices is always
1119 * authoritative, unless 'rebuild' or '[no]sync' was specified.
1120 */
1121 for (i = 0; i < rs->raid_disks; i++) {
1122 set_bit(In_sync, &rs->dev[i].rdev.flags);
1123 rs->dev[i].rdev.recovery_offset = MaxSector;
1124 }
1125
1126 /*
1127 * Second, parse the unordered optional arguments
1128 */
1129 for (i = 0; i < num_raid_params; i++) {
1130 key = dm_shift_arg(as);
1131 if (!key) {
1132 rs->ti->error = "Not enough raid parameters given";
1133 return -EINVAL;
1134 }
1135
1136 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
1137 if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1138 rs->ti->error = "Only one 'nosync' argument allowed";
1139 return -EINVAL;
1140 }
1141 continue;
1142 }
1143 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
1144 if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
1145 rs->ti->error = "Only one 'sync' argument allowed";
1146 return -EINVAL;
1147 }
1148 continue;
1149 }
1150 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
1151 if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1152 rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1153 return -EINVAL;
1154 }
1155 continue;
1156 }
1157
1158 arg = dm_shift_arg(as);
1159 i++; /* Account for the argument pairs */
1160 if (!arg) {
1161 rs->ti->error = "Wrong number of raid parameters given";
1162 return -EINVAL;
1163 }
1164
1165 /*
1166 * Parameters that take a string value are checked here.
1167 */
1168 /* "raid10_format {near|offset|far} */
1169 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
1170 if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
1171 rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1172 return -EINVAL;
1173 }
1174 if (!rt_is_raid10(rt)) {
1175 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1176 return -EINVAL;
1177 }
1178 raid10_format = raid10_name_to_format(arg);
1179 if (raid10_format < 0) {
1180 rs->ti->error = "Invalid 'raid10_format' value given";
1181 return raid10_format;
1182 }
1183 continue;
1184 }
1185
1186 /* "journal_dev dev" */
1187 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV))) {
1188 int r;
1189 struct md_rdev *jdev;
1190
1191 if (test_and_set_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1192 rs->ti->error = "Only one raid4/5/6 set journaling device allowed";
1193 return -EINVAL;
1194 }
1195 if (!rt_is_raid456(rt)) {
1196 rs->ti->error = "'journal_dev' is an invalid parameter for this RAID type";
1197 return -EINVAL;
1198 }
1199 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
1200 &rs->journal_dev.dev);
1201 if (r) {
1202 rs->ti->error = "raid4/5/6 journal device lookup failure";
1203 return r;
1204 }
1205 jdev = &rs->journal_dev.rdev;
1206 md_rdev_init(jdev);
1207 jdev->mddev = &rs->md;
1208 jdev->bdev = rs->journal_dev.dev->bdev;
1209 jdev->sectors = to_sector(i_size_read(jdev->bdev->bd_inode));
1210 if (jdev->sectors < MIN_RAID456_JOURNAL_SPACE) {
1211 rs->ti->error = "No space for raid4/5/6 journal";
1212 return -ENOSPC;
1213 }
1214 set_bit(Journal, &jdev->flags);
1215 continue;
1216 }
1217
1218 /*
1219 * Parameters with number values from here on.
1220 */
1221 if (kstrtoint(arg, 10, &value) < 0) {
1222 rs->ti->error = "Bad numerical argument given in raid params";
1223 return -EINVAL;
1224 }
1225
1226 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
1227 /*
1228 * "rebuild" is being passed in by userspace to provide
1229 * indexes of replaced devices and to set up additional
1230 * devices on raid level takeover.
1231 */
1232 if (!__within_range(value, 0, rs->raid_disks - 1)) {
1233 rs->ti->error = "Invalid rebuild index given";
1234 return -EINVAL;
1235 }
1236
1237 if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1238 rs->ti->error = "rebuild for this index already given";
1239 return -EINVAL;
1240 }
1241
1242 rd = rs->dev + value;
1243 clear_bit(In_sync, &rd->rdev.flags);
1244 clear_bit(Faulty, &rd->rdev.flags);
1245 rd->rdev.recovery_offset = 0;
1246 set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
1247 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
1248 if (!rt_is_raid1(rt)) {
1249 rs->ti->error = "write_mostly option is only valid for RAID1";
1250 return -EINVAL;
1251 }
1252
1253 if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
1254 rs->ti->error = "Invalid write_mostly index given";
1255 return -EINVAL;
1256 }
1257
1258 write_mostly++;
1259 set_bit(WriteMostly, &rs->dev[value].rdev.flags);
1260 set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
1261 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
1262 if (!rt_is_raid1(rt)) {
1263 rs->ti->error = "max_write_behind option is only valid for RAID1";
1264 return -EINVAL;
1265 }
1266
1267 if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
1268 rs->ti->error = "Only one max_write_behind argument pair allowed";
1269 return -EINVAL;
1270 }
1271
1272 /*
1273 * In device-mapper, we specify things in sectors, but
1274 * MD records this value in kB
1275 */
1276 value /= 2;
1277 if (value > COUNTER_MAX) {
1278 rs->ti->error = "Max write-behind limit out of range";
1279 return -EINVAL;
1280 }
1281
1282 rs->md.bitmap_info.max_write_behind = value;
1283 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
1284 if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
1285 rs->ti->error = "Only one daemon_sleep argument pair allowed";
1286 return -EINVAL;
1287 }
1288 if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
1289 rs->ti->error = "daemon sleep period out of range";
1290 return -EINVAL;
1291 }
1292 rs->md.bitmap_info.daemon_sleep = value;
1293 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
1294 /* Userspace passes new data_offset after having extended the the data image LV */
1295 if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
1296 rs->ti->error = "Only one data_offset argument pair allowed";
1297 return -EINVAL;
1298 }
1299 /* Ensure sensible data offset */
1300 if (value < 0 ||
1301 (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
1302 rs->ti->error = "Bogus data_offset value";
1303 return -EINVAL;
1304 }
1305 rs->data_offset = value;
1306 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
1307 /* Define the +/-# of disks to add to/remove from the given raid set */
1308 if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1309 rs->ti->error = "Only one delta_disks argument pair allowed";
1310 return -EINVAL;
1311 }
1312 /* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1313 if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
1314 rs->ti->error = "Too many delta_disk requested";
1315 return -EINVAL;
1316 }
1317
1318 rs->delta_disks = value;
1319 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
1320 if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
1321 rs->ti->error = "Only one stripe_cache argument pair allowed";
1322 return -EINVAL;
1323 }
1324
1325 if (!rt_is_raid456(rt)) {
1326 rs->ti->error = "Inappropriate argument: stripe_cache";
1327 return -EINVAL;
1328 }
1329
1330 rs->stripe_cache_entries = value;
1331 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
1332 if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1333 rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1334 return -EINVAL;
1335 }
1336 if (value > INT_MAX) {
1337 rs->ti->error = "min_recovery_rate out of range";
1338 return -EINVAL;
1339 }
1340 rs->md.sync_speed_min = (int)value;
1341 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
1342 if (test_and_set_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags)) {
1343 rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1344 return -EINVAL;
1345 }
1346 if (value > INT_MAX) {
1347 rs->ti->error = "max_recovery_rate out of range";
1348 return -EINVAL;
1349 }
1350 rs->md.sync_speed_max = (int)value;
1351 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
1352 if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
1353 rs->ti->error = "Only one region_size argument pair allowed";
1354 return -EINVAL;
1355 }
1356
1357 region_size = value;
1358 rs->requested_bitmap_chunk_sectors = value;
1359 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
1360 if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
1361 rs->ti->error = "Only one raid10_copies argument pair allowed";
1362 return -EINVAL;
1363 }
1364
1365 if (!__within_range(value, 2, rs->md.raid_disks)) {
1366 rs->ti->error = "Bad value for 'raid10_copies'";
1367 return -EINVAL;
1368 }
1369
1370 raid10_copies = value;
1371 } else {
1372 DMERR("Unable to parse RAID parameter: %s", key);
1373 rs->ti->error = "Unable to parse RAID parameter";
1374 return -EINVAL;
1375 }
1376 }
1377
1378 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1379 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1380 rs->ti->error = "sync and nosync are mutually exclusive";
1381 return -EINVAL;
1382 }
1383
1384 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) &&
1385 (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ||
1386 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))) {
1387 rs->ti->error = "sync/nosync and rebuild are mutually exclusive";
1388 return -EINVAL;
1389 }
1390
1391 if (write_mostly >= rs->md.raid_disks) {
1392 rs->ti->error = "Can't set all raid1 devices to write_mostly";
1393 return -EINVAL;
1394 }
1395
1396 if (validate_region_size(rs, region_size))
1397 return -EINVAL;
1398
1399 if (rs->md.chunk_sectors)
1400 max_io_len = rs->md.chunk_sectors;
1401 else
1402 max_io_len = region_size;
1403
1404 if (dm_set_target_max_io_len(rs->ti, max_io_len))
1405 return -EINVAL;
1406
1407 if (rt_is_raid10(rt)) {
1408 if (raid10_copies > rs->md.raid_disks) {
1409 rs->ti->error = "Not enough devices to satisfy specification";
1410 return -EINVAL;
1411 }
1412
1413 rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
1414 if (rs->md.new_layout < 0) {
1415 rs->ti->error = "Error getting raid10 format";
1416 return rs->md.new_layout;
1417 }
1418
1419 rt = get_raid_type_by_ll(10, rs->md.new_layout);
1420 if (!rt) {
1421 rs->ti->error = "Failed to recognize new raid10 layout";
1422 return -EINVAL;
1423 }
1424
1425 if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1426 rt->algorithm == ALGORITHM_RAID10_NEAR) &&
1427 test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1428 rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1429 return -EINVAL;
1430 }
1431 }
1432
1433 rs->raid10_copies = raid10_copies;
1434
1435 /* Assume there are no metadata devices until the drives are parsed */
1436 rs->md.persistent = 0;
1437 rs->md.external = 1;
1438
1439 /* Check, if any invalid ctr arguments have been passed in for the raid level */
1440 return rs_check_for_valid_flags(rs);
1441 }
1442
1443 /* Set raid4/5/6 cache size */
1444 static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1445 {
1446 int r;
1447 struct r5conf *conf;
1448 struct mddev *mddev = &rs->md;
1449 uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1450 uint32_t nr_stripes = rs->stripe_cache_entries;
1451
1452 if (!rt_is_raid456(rs->raid_type)) {
1453 rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1454 return -EINVAL;
1455 }
1456
1457 if (nr_stripes < min_stripes) {
1458 DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1459 nr_stripes, min_stripes);
1460 nr_stripes = min_stripes;
1461 }
1462
1463 conf = mddev->private;
1464 if (!conf) {
1465 rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1466 return -EINVAL;
1467 }
1468
1469 /* Try setting number of stripes in raid456 stripe cache */
1470 if (conf->min_nr_stripes != nr_stripes) {
1471 r = raid5_set_cache_size(mddev, nr_stripes);
1472 if (r) {
1473 rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1474 return r;
1475 }
1476
1477 DMINFO("%u stripe cache entries", nr_stripes);
1478 }
1479
1480 return 0;
1481 }
1482
1483 /* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1484 static unsigned int mddev_data_stripes(struct raid_set *rs)
1485 {
1486 return rs->md.raid_disks - rs->raid_type->parity_devs;
1487 }
1488
1489 /* Return # of data stripes of @rs (i.e. as of ctr) */
1490 static unsigned int rs_data_stripes(struct raid_set *rs)
1491 {
1492 return rs->raid_disks - rs->raid_type->parity_devs;
1493 }
1494
1495 /*
1496 * Retrieve rdev->sectors from any valid raid device of @rs
1497 * to allow userpace to pass in arbitray "- -" device tupples.
1498 */
1499 static sector_t __rdev_sectors(struct raid_set *rs)
1500 {
1501 int i;
1502
1503 for (i = 0; i < rs->md.raid_disks; i++) {
1504 struct md_rdev *rdev = &rs->dev[i].rdev;
1505
1506 if (!test_bit(Journal, &rdev->flags) &&
1507 rdev->bdev && rdev->sectors)
1508 return rdev->sectors;
1509 }
1510
1511 BUG(); /* Constructor ensures we got some. */
1512 }
1513
1514 /* Calculate the sectors per device and per array used for @rs */
1515 static int rs_set_dev_and_array_sectors(struct raid_set *rs, bool use_mddev)
1516 {
1517 int delta_disks;
1518 unsigned int data_stripes;
1519 struct mddev *mddev = &rs->md;
1520 struct md_rdev *rdev;
1521 sector_t array_sectors = rs->ti->len, dev_sectors = rs->ti->len;
1522
1523 if (use_mddev) {
1524 delta_disks = mddev->delta_disks;
1525 data_stripes = mddev_data_stripes(rs);
1526 } else {
1527 delta_disks = rs->delta_disks;
1528 data_stripes = rs_data_stripes(rs);
1529 }
1530
1531 /* Special raid1 case w/o delta_disks support (yet) */
1532 if (rt_is_raid1(rs->raid_type))
1533 ;
1534 else if (rt_is_raid10(rs->raid_type)) {
1535 if (rs->raid10_copies < 2 ||
1536 delta_disks < 0) {
1537 rs->ti->error = "Bogus raid10 data copies or delta disks";
1538 return -EINVAL;
1539 }
1540
1541 dev_sectors *= rs->raid10_copies;
1542 if (sector_div(dev_sectors, data_stripes))
1543 goto bad;
1544
1545 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1546 if (sector_div(array_sectors, rs->raid10_copies))
1547 goto bad;
1548
1549 } else if (sector_div(dev_sectors, data_stripes))
1550 goto bad;
1551
1552 else
1553 /* Striped layouts */
1554 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1555
1556 rdev_for_each(rdev, mddev)
1557 if (!test_bit(Journal, &rdev->flags))
1558 rdev->sectors = dev_sectors;
1559
1560 mddev->array_sectors = array_sectors;
1561 mddev->dev_sectors = dev_sectors;
1562
1563 return 0;
1564 bad:
1565 rs->ti->error = "Target length not divisible by number of data devices";
1566 return -EINVAL;
1567 }
1568
1569 /* Setup recovery on @rs */
1570 static void __rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1571 {
1572 /* raid0 does not recover */
1573 if (rs_is_raid0(rs))
1574 rs->md.recovery_cp = MaxSector;
1575 /*
1576 * A raid6 set has to be recovered either
1577 * completely or for the grown part to
1578 * ensure proper parity and Q-Syndrome
1579 */
1580 else if (rs_is_raid6(rs))
1581 rs->md.recovery_cp = dev_sectors;
1582 /*
1583 * Other raid set types may skip recovery
1584 * depending on the 'nosync' flag.
1585 */
1586 else
1587 rs->md.recovery_cp = test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)
1588 ? MaxSector : dev_sectors;
1589 }
1590
1591 /* Setup recovery on @rs based on raid type, device size and 'nosync' flag */
1592 static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1593 {
1594 if (!dev_sectors)
1595 /* New raid set or 'sync' flag provided */
1596 __rs_setup_recovery(rs, 0);
1597 else if (dev_sectors == MaxSector)
1598 /* Prevent recovery */
1599 __rs_setup_recovery(rs, MaxSector);
1600 else if (__rdev_sectors(rs) < dev_sectors)
1601 /* Grown raid set */
1602 __rs_setup_recovery(rs, __rdev_sectors(rs));
1603 else
1604 __rs_setup_recovery(rs, MaxSector);
1605 }
1606
1607 static void do_table_event(struct work_struct *ws)
1608 {
1609 struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1610
1611 smp_rmb(); /* Make sure we access most actual mddev properties */
1612 if (!rs_is_reshaping(rs))
1613 rs_set_capacity(rs);
1614 dm_table_event(rs->ti->table);
1615 }
1616
1617 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
1618 {
1619 struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
1620
1621 return mddev_congested(&rs->md, bits);
1622 }
1623
1624 /*
1625 * Make sure a valid takover (level switch) is being requested on @rs
1626 *
1627 * Conversions of raid sets from one MD personality to another
1628 * have to conform to restrictions which are enforced here.
1629 */
1630 static int rs_check_takeover(struct raid_set *rs)
1631 {
1632 struct mddev *mddev = &rs->md;
1633 unsigned int near_copies;
1634
1635 if (rs->md.degraded) {
1636 rs->ti->error = "Can't takeover degraded raid set";
1637 return -EPERM;
1638 }
1639
1640 if (rs_is_reshaping(rs)) {
1641 rs->ti->error = "Can't takeover reshaping raid set";
1642 return -EPERM;
1643 }
1644
1645 switch (mddev->level) {
1646 case 0:
1647 /* raid0 -> raid1/5 with one disk */
1648 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1649 mddev->raid_disks == 1)
1650 return 0;
1651
1652 /* raid0 -> raid10 */
1653 if (mddev->new_level == 10 &&
1654 !(rs->raid_disks % mddev->raid_disks))
1655 return 0;
1656
1657 /* raid0 with multiple disks -> raid4/5/6 */
1658 if (__within_range(mddev->new_level, 4, 6) &&
1659 mddev->new_layout == ALGORITHM_PARITY_N &&
1660 mddev->raid_disks > 1)
1661 return 0;
1662
1663 break;
1664
1665 case 10:
1666 /* Can't takeover raid10_offset! */
1667 if (__is_raid10_offset(mddev->layout))
1668 break;
1669
1670 near_copies = __raid10_near_copies(mddev->layout);
1671
1672 /* raid10* -> raid0 */
1673 if (mddev->new_level == 0) {
1674 /* Can takeover raid10_near with raid disks divisable by data copies! */
1675 if (near_copies > 1 &&
1676 !(mddev->raid_disks % near_copies)) {
1677 mddev->raid_disks /= near_copies;
1678 mddev->delta_disks = mddev->raid_disks;
1679 return 0;
1680 }
1681
1682 /* Can takeover raid10_far */
1683 if (near_copies == 1 &&
1684 __raid10_far_copies(mddev->layout) > 1)
1685 return 0;
1686
1687 break;
1688 }
1689
1690 /* raid10_{near,far} -> raid1 */
1691 if (mddev->new_level == 1 &&
1692 max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
1693 return 0;
1694
1695 /* raid10_{near,far} with 2 disks -> raid4/5 */
1696 if (__within_range(mddev->new_level, 4, 5) &&
1697 mddev->raid_disks == 2)
1698 return 0;
1699 break;
1700
1701 case 1:
1702 /* raid1 with 2 disks -> raid4/5 */
1703 if (__within_range(mddev->new_level, 4, 5) &&
1704 mddev->raid_disks == 2) {
1705 mddev->degraded = 1;
1706 return 0;
1707 }
1708
1709 /* raid1 -> raid0 */
1710 if (mddev->new_level == 0 &&
1711 mddev->raid_disks == 1)
1712 return 0;
1713
1714 /* raid1 -> raid10 */
1715 if (mddev->new_level == 10)
1716 return 0;
1717 break;
1718
1719 case 4:
1720 /* raid4 -> raid0 */
1721 if (mddev->new_level == 0)
1722 return 0;
1723
1724 /* raid4 -> raid1/5 with 2 disks */
1725 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1726 mddev->raid_disks == 2)
1727 return 0;
1728
1729 /* raid4 -> raid5/6 with parity N */
1730 if (__within_range(mddev->new_level, 5, 6) &&
1731 mddev->layout == ALGORITHM_PARITY_N)
1732 return 0;
1733 break;
1734
1735 case 5:
1736 /* raid5 with parity N -> raid0 */
1737 if (mddev->new_level == 0 &&
1738 mddev->layout == ALGORITHM_PARITY_N)
1739 return 0;
1740
1741 /* raid5 with parity N -> raid4 */
1742 if (mddev->new_level == 4 &&
1743 mddev->layout == ALGORITHM_PARITY_N)
1744 return 0;
1745
1746 /* raid5 with 2 disks -> raid1/4/10 */
1747 if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1748 mddev->raid_disks == 2)
1749 return 0;
1750
1751 /* raid5_* -> raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
1752 if (mddev->new_level == 6 &&
1753 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1754 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
1755 return 0;
1756 break;
1757
1758 case 6:
1759 /* raid6 with parity N -> raid0 */
1760 if (mddev->new_level == 0 &&
1761 mddev->layout == ALGORITHM_PARITY_N)
1762 return 0;
1763
1764 /* raid6 with parity N -> raid4 */
1765 if (mddev->new_level == 4 &&
1766 mddev->layout == ALGORITHM_PARITY_N)
1767 return 0;
1768
1769 /* raid6_*_n with Q-Syndrome N -> raid5_* */
1770 if (mddev->new_level == 5 &&
1771 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1772 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
1773 return 0;
1774
1775 default:
1776 break;
1777 }
1778
1779 rs->ti->error = "takeover not possible";
1780 return -EINVAL;
1781 }
1782
1783 /* True if @rs requested to be taken over */
1784 static bool rs_takeover_requested(struct raid_set *rs)
1785 {
1786 return rs->md.new_level != rs->md.level;
1787 }
1788
1789 /* True if @rs is requested to reshape by ctr */
1790 static bool rs_reshape_requested(struct raid_set *rs)
1791 {
1792 bool change;
1793 struct mddev *mddev = &rs->md;
1794
1795 if (rs_takeover_requested(rs))
1796 return false;
1797
1798 if (!mddev->level)
1799 return false;
1800
1801 change = mddev->new_layout != mddev->layout ||
1802 mddev->new_chunk_sectors != mddev->chunk_sectors ||
1803 rs->delta_disks;
1804
1805 /* Historical case to support raid1 reshape without delta disks */
1806 if (mddev->level == 1) {
1807 if (rs->delta_disks)
1808 return !!rs->delta_disks;
1809
1810 return !change &&
1811 mddev->raid_disks != rs->raid_disks;
1812 }
1813
1814 if (mddev->level == 10)
1815 return change &&
1816 !__is_raid10_far(mddev->new_layout) &&
1817 rs->delta_disks >= 0;
1818
1819 return change;
1820 }
1821
1822 /* Features */
1823 #define FEATURE_FLAG_SUPPORTS_V190 0x1 /* Supports extended superblock */
1824
1825 /* State flags for sb->flags */
1826 #define SB_FLAG_RESHAPE_ACTIVE 0x1
1827 #define SB_FLAG_RESHAPE_BACKWARDS 0x2
1828
1829 /*
1830 * This structure is never routinely used by userspace, unlike md superblocks.
1831 * Devices with this superblock should only ever be accessed via device-mapper.
1832 */
1833 #define DM_RAID_MAGIC 0x64526D44
1834 struct dm_raid_superblock {
1835 __le32 magic; /* "DmRd" */
1836 __le32 compat_features; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1837
1838 __le32 num_devices; /* Number of devices in this raid set. (Max 64) */
1839 __le32 array_position; /* The position of this drive in the raid set */
1840
1841 __le64 events; /* Incremented by md when superblock updated */
1842 __le64 failed_devices; /* Pre 1.9.0 part of bit field of devices to */
1843 /* indicate failures (see extension below) */
1844
1845 /*
1846 * This offset tracks the progress of the repair or replacement of
1847 * an individual drive.
1848 */
1849 __le64 disk_recovery_offset;
1850
1851 /*
1852 * This offset tracks the progress of the initial raid set
1853 * synchronisation/parity calculation.
1854 */
1855 __le64 array_resync_offset;
1856
1857 /*
1858 * raid characteristics
1859 */
1860 __le32 level;
1861 __le32 layout;
1862 __le32 stripe_sectors;
1863
1864 /********************************************************************
1865 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1866 *
1867 * FEATURE_FLAG_SUPPORTS_V190 in the features member indicates that those exist
1868 */
1869
1870 __le32 flags; /* Flags defining array states for reshaping */
1871
1872 /*
1873 * This offset tracks the progress of a raid
1874 * set reshape in order to be able to restart it
1875 */
1876 __le64 reshape_position;
1877
1878 /*
1879 * These define the properties of the array in case of an interrupted reshape
1880 */
1881 __le32 new_level;
1882 __le32 new_layout;
1883 __le32 new_stripe_sectors;
1884 __le32 delta_disks;
1885
1886 __le64 array_sectors; /* Array size in sectors */
1887
1888 /*
1889 * Sector offsets to data on devices (reshaping).
1890 * Needed to support out of place reshaping, thus
1891 * not writing over any stripes whilst converting
1892 * them from old to new layout
1893 */
1894 __le64 data_offset;
1895 __le64 new_data_offset;
1896
1897 __le64 sectors; /* Used device size in sectors */
1898
1899 /*
1900 * Additonal Bit field of devices indicating failures to support
1901 * up to 256 devices with the 1.9.0 on-disk metadata format
1902 */
1903 __le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
1904
1905 __le32 incompat_features; /* Used to indicate any incompatible features */
1906
1907 /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
1908 } __packed;
1909
1910 /*
1911 * Check for reshape constraints on raid set @rs:
1912 *
1913 * - reshape function non-existent
1914 * - degraded set
1915 * - ongoing recovery
1916 * - ongoing reshape
1917 *
1918 * Returns 0 if none or -EPERM if given constraint
1919 * and error message reference in @errmsg
1920 */
1921 static int rs_check_reshape(struct raid_set *rs)
1922 {
1923 struct mddev *mddev = &rs->md;
1924
1925 if (!mddev->pers || !mddev->pers->check_reshape)
1926 rs->ti->error = "Reshape not supported";
1927 else if (mddev->degraded)
1928 rs->ti->error = "Can't reshape degraded raid set";
1929 else if (rs_is_recovering(rs))
1930 rs->ti->error = "Convert request on recovering raid set prohibited";
1931 else if (rs_is_reshaping(rs))
1932 rs->ti->error = "raid set already reshaping!";
1933 else if (!(rs_is_raid1(rs) || rs_is_raid10(rs) || rs_is_raid456(rs)))
1934 rs->ti->error = "Reshaping only supported for raid1/4/5/6/10";
1935 else
1936 return 0;
1937
1938 return -EPERM;
1939 }
1940
1941 static int read_disk_sb(struct md_rdev *rdev, int size, bool force_reload)
1942 {
1943 BUG_ON(!rdev->sb_page);
1944
1945 if (rdev->sb_loaded && !force_reload)
1946 return 0;
1947
1948 rdev->sb_loaded = 0;
1949
1950 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) {
1951 DMERR("Failed to read superblock of device at position %d",
1952 rdev->raid_disk);
1953 md_error(rdev->mddev, rdev);
1954 set_bit(Faulty, &rdev->flags);
1955 return -EIO;
1956 }
1957
1958 rdev->sb_loaded = 1;
1959
1960 return 0;
1961 }
1962
1963 static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
1964 {
1965 failed_devices[0] = le64_to_cpu(sb->failed_devices);
1966 memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
1967
1968 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
1969 int i = ARRAY_SIZE(sb->extended_failed_devices);
1970
1971 while (i--)
1972 failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
1973 }
1974 }
1975
1976 static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
1977 {
1978 int i = ARRAY_SIZE(sb->extended_failed_devices);
1979
1980 sb->failed_devices = cpu_to_le64(failed_devices[0]);
1981 while (i--)
1982 sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
1983 }
1984
1985 /*
1986 * Synchronize the superblock members with the raid set properties
1987 *
1988 * All superblock data is little endian.
1989 */
1990 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
1991 {
1992 bool update_failed_devices = false;
1993 unsigned int i;
1994 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
1995 struct dm_raid_superblock *sb;
1996 struct raid_set *rs = container_of(mddev, struct raid_set, md);
1997
1998 /* No metadata device, no superblock */
1999 if (!rdev->meta_bdev)
2000 return;
2001
2002 BUG_ON(!rdev->sb_page);
2003
2004 sb = page_address(rdev->sb_page);
2005
2006 sb_retrieve_failed_devices(sb, failed_devices);
2007
2008 for (i = 0; i < rs->raid_disks; i++)
2009 if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
2010 update_failed_devices = true;
2011 set_bit(i, (void *) failed_devices);
2012 }
2013
2014 if (update_failed_devices)
2015 sb_update_failed_devices(sb, failed_devices);
2016
2017 sb->magic = cpu_to_le32(DM_RAID_MAGIC);
2018 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2019
2020 sb->num_devices = cpu_to_le32(mddev->raid_disks);
2021 sb->array_position = cpu_to_le32(rdev->raid_disk);
2022
2023 sb->events = cpu_to_le64(mddev->events);
2024
2025 sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
2026 sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
2027
2028 sb->level = cpu_to_le32(mddev->level);
2029 sb->layout = cpu_to_le32(mddev->layout);
2030 sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
2031
2032 sb->new_level = cpu_to_le32(mddev->new_level);
2033 sb->new_layout = cpu_to_le32(mddev->new_layout);
2034 sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
2035
2036 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2037
2038 smp_rmb(); /* Make sure we access most recent reshape position */
2039 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2040 if (le64_to_cpu(sb->reshape_position) != MaxSector) {
2041 /* Flag ongoing reshape */
2042 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
2043
2044 if (mddev->delta_disks < 0 || mddev->reshape_backwards)
2045 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
2046 } else {
2047 /* Clear reshape flags */
2048 sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
2049 }
2050
2051 sb->array_sectors = cpu_to_le64(mddev->array_sectors);
2052 sb->data_offset = cpu_to_le64(rdev->data_offset);
2053 sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
2054 sb->sectors = cpu_to_le64(rdev->sectors);
2055 sb->incompat_features = cpu_to_le32(0);
2056
2057 /* Zero out the rest of the payload after the size of the superblock */
2058 memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
2059 }
2060
2061 /*
2062 * super_load
2063 *
2064 * This function creates a superblock if one is not found on the device
2065 * and will decide which superblock to use if there's a choice.
2066 *
2067 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
2068 */
2069 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
2070 {
2071 int r;
2072 struct dm_raid_superblock *sb;
2073 struct dm_raid_superblock *refsb;
2074 uint64_t events_sb, events_refsb;
2075
2076 rdev->sb_start = 0;
2077 rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
2078 if (rdev->sb_size < sizeof(*sb) || rdev->sb_size > PAGE_SIZE) {
2079 DMERR("superblock size of a logical block is no longer valid");
2080 return -EINVAL;
2081 }
2082
2083 r = read_disk_sb(rdev, rdev->sb_size, false);
2084 if (r)
2085 return r;
2086
2087 sb = page_address(rdev->sb_page);
2088
2089 /*
2090 * Two cases that we want to write new superblocks and rebuild:
2091 * 1) New device (no matching magic number)
2092 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2093 */
2094 if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
2095 (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
2096 super_sync(rdev->mddev, rdev);
2097
2098 set_bit(FirstUse, &rdev->flags);
2099 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2100
2101 /* Force writing of superblocks to disk */
2102 set_bit(MD_SB_CHANGE_DEVS, &rdev->mddev->sb_flags);
2103
2104 /* Any superblock is better than none, choose that if given */
2105 return refdev ? 0 : 1;
2106 }
2107
2108 if (!refdev)
2109 return 1;
2110
2111 events_sb = le64_to_cpu(sb->events);
2112
2113 refsb = page_address(refdev->sb_page);
2114 events_refsb = le64_to_cpu(refsb->events);
2115
2116 return (events_sb > events_refsb) ? 1 : 0;
2117 }
2118
2119 static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
2120 {
2121 int role;
2122 unsigned int d;
2123 struct mddev *mddev = &rs->md;
2124 uint64_t events_sb;
2125 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2126 struct dm_raid_superblock *sb;
2127 uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
2128 struct md_rdev *r;
2129 struct dm_raid_superblock *sb2;
2130
2131 sb = page_address(rdev->sb_page);
2132 events_sb = le64_to_cpu(sb->events);
2133
2134 /*
2135 * Initialise to 1 if this is a new superblock.
2136 */
2137 mddev->events = events_sb ? : 1;
2138
2139 mddev->reshape_position = MaxSector;
2140
2141 mddev->raid_disks = le32_to_cpu(sb->num_devices);
2142 mddev->level = le32_to_cpu(sb->level);
2143 mddev->layout = le32_to_cpu(sb->layout);
2144 mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
2145
2146 /*
2147 * Reshaping is supported, e.g. reshape_position is valid
2148 * in superblock and superblock content is authoritative.
2149 */
2150 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2151 /* Superblock is authoritative wrt given raid set layout! */
2152 mddev->new_level = le32_to_cpu(sb->new_level);
2153 mddev->new_layout = le32_to_cpu(sb->new_layout);
2154 mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
2155 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
2156 mddev->array_sectors = le64_to_cpu(sb->array_sectors);
2157
2158 /* raid was reshaping and got interrupted */
2159 if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
2160 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
2161 DMERR("Reshape requested but raid set is still reshaping");
2162 return -EINVAL;
2163 }
2164
2165 if (mddev->delta_disks < 0 ||
2166 (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
2167 mddev->reshape_backwards = 1;
2168 else
2169 mddev->reshape_backwards = 0;
2170
2171 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2172 rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2173 }
2174
2175 } else {
2176 /*
2177 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2178 */
2179 struct raid_type *rt_cur = get_raid_type_by_ll(mddev->level, mddev->layout);
2180 struct raid_type *rt_new = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
2181
2182 if (rs_takeover_requested(rs)) {
2183 if (rt_cur && rt_new)
2184 DMERR("Takeover raid sets from %s to %s not yet supported by metadata. (raid level change)",
2185 rt_cur->name, rt_new->name);
2186 else
2187 DMERR("Takeover raid sets not yet supported by metadata. (raid level change)");
2188 return -EINVAL;
2189 } else if (rs_reshape_requested(rs)) {
2190 DMERR("Reshaping raid sets not yet supported by metadata. (raid layout change keeping level)");
2191 if (mddev->layout != mddev->new_layout) {
2192 if (rt_cur && rt_new)
2193 DMERR(" current layout %s vs new layout %s",
2194 rt_cur->name, rt_new->name);
2195 else
2196 DMERR(" current layout 0x%X vs new layout 0x%X",
2197 le32_to_cpu(sb->layout), mddev->new_layout);
2198 }
2199 if (mddev->chunk_sectors != mddev->new_chunk_sectors)
2200 DMERR(" current stripe sectors %u vs new stripe sectors %u",
2201 mddev->chunk_sectors, mddev->new_chunk_sectors);
2202 if (rs->delta_disks)
2203 DMERR(" current %u disks vs new %u disks",
2204 mddev->raid_disks, mddev->raid_disks + rs->delta_disks);
2205 if (rs_is_raid10(rs)) {
2206 DMERR(" Old layout: %s w/ %u copies",
2207 raid10_md_layout_to_format(mddev->layout),
2208 raid10_md_layout_to_copies(mddev->layout));
2209 DMERR(" New layout: %s w/ %u copies",
2210 raid10_md_layout_to_format(mddev->new_layout),
2211 raid10_md_layout_to_copies(mddev->new_layout));
2212 }
2213 return -EINVAL;
2214 }
2215
2216 DMINFO("Discovered old metadata format; upgrading to extended metadata format");
2217 }
2218
2219 if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
2220 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2221
2222 /*
2223 * During load, we set FirstUse if a new superblock was written.
2224 * There are two reasons we might not have a superblock:
2225 * 1) The raid set is brand new - in which case, all of the
2226 * devices must have their In_sync bit set. Also,
2227 * recovery_cp must be 0, unless forced.
2228 * 2) This is a new device being added to an old raid set
2229 * and the new device needs to be rebuilt - in which
2230 * case the In_sync bit will /not/ be set and
2231 * recovery_cp must be MaxSector.
2232 * 3) This is/are a new device(s) being added to an old
2233 * raid set during takeover to a higher raid level
2234 * to provide capacity for redundancy or during reshape
2235 * to add capacity to grow the raid set.
2236 */
2237 d = 0;
2238 rdev_for_each(r, mddev) {
2239 if (test_bit(Journal, &rdev->flags))
2240 continue;
2241
2242 if (test_bit(FirstUse, &r->flags))
2243 new_devs++;
2244
2245 if (!test_bit(In_sync, &r->flags)) {
2246 DMINFO("Device %d specified for rebuild; clearing superblock",
2247 r->raid_disk);
2248 rebuilds++;
2249
2250 if (test_bit(FirstUse, &r->flags))
2251 rebuild_and_new++;
2252 }
2253
2254 d++;
2255 }
2256
2257 if (new_devs == rs->raid_disks || !rebuilds) {
2258 /* Replace a broken device */
2259 if (new_devs == 1 && !rs->delta_disks)
2260 ;
2261 if (new_devs == rs->raid_disks) {
2262 DMINFO("Superblocks created for new raid set");
2263 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2264 } else if (new_devs != rebuilds &&
2265 new_devs != rs->delta_disks) {
2266 DMERR("New device injected into existing raid set without "
2267 "'delta_disks' or 'rebuild' parameter specified");
2268 return -EINVAL;
2269 }
2270 } else if (new_devs && new_devs != rebuilds) {
2271 DMERR("%u 'rebuild' devices cannot be injected into"
2272 " a raid set with %u other first-time devices",
2273 rebuilds, new_devs);
2274 return -EINVAL;
2275 } else if (rebuilds) {
2276 if (rebuild_and_new && rebuilds != rebuild_and_new) {
2277 DMERR("new device%s provided without 'rebuild'",
2278 new_devs > 1 ? "s" : "");
2279 return -EINVAL;
2280 } else if (rs_is_recovering(rs)) {
2281 DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2282 (unsigned long long) mddev->recovery_cp);
2283 return -EINVAL;
2284 } else if (rs_is_reshaping(rs)) {
2285 DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2286 (unsigned long long) mddev->reshape_position);
2287 return -EINVAL;
2288 }
2289 }
2290
2291 /*
2292 * Now we set the Faulty bit for those devices that are
2293 * recorded in the superblock as failed.
2294 */
2295 sb_retrieve_failed_devices(sb, failed_devices);
2296 rdev_for_each(r, mddev) {
2297 if (test_bit(Journal, &rdev->flags) ||
2298 !r->sb_page)
2299 continue;
2300 sb2 = page_address(r->sb_page);
2301 sb2->failed_devices = 0;
2302 memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
2303
2304 /*
2305 * Check for any device re-ordering.
2306 */
2307 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2308 role = le32_to_cpu(sb2->array_position);
2309 if (role < 0)
2310 continue;
2311
2312 if (role != r->raid_disk) {
2313 if (rs_is_raid10(rs) && __is_raid10_near(mddev->layout)) {
2314 if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
2315 rs->raid_disks % rs->raid10_copies) {
2316 rs->ti->error =
2317 "Cannot change raid10 near set to odd # of devices!";
2318 return -EINVAL;
2319 }
2320
2321 sb2->array_position = cpu_to_le32(r->raid_disk);
2322
2323 } else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
2324 !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2325 !rt_is_raid1(rs->raid_type)) {
2326 rs->ti->error = "Cannot change device positions in raid set";
2327 return -EINVAL;
2328 }
2329
2330 DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
2331 }
2332
2333 /*
2334 * Partial recovery is performed on
2335 * returning failed devices.
2336 */
2337 if (test_bit(role, (void *) failed_devices))
2338 set_bit(Faulty, &r->flags);
2339 }
2340 }
2341
2342 return 0;
2343 }
2344
2345 static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
2346 {
2347 struct mddev *mddev = &rs->md;
2348 struct dm_raid_superblock *sb;
2349
2350 if (rs_is_raid0(rs) || !rdev->sb_page || rdev->raid_disk < 0)
2351 return 0;
2352
2353 sb = page_address(rdev->sb_page);
2354
2355 /*
2356 * If mddev->events is not set, we know we have not yet initialized
2357 * the array.
2358 */
2359 if (!mddev->events && super_init_validation(rs, rdev))
2360 return -EINVAL;
2361
2362 if (le32_to_cpu(sb->compat_features) &&
2363 le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
2364 rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2365 return -EINVAL;
2366 }
2367
2368 if (sb->incompat_features) {
2369 rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
2370 return -EINVAL;
2371 }
2372
2373 /* Enable bitmap creation for RAID levels != 0 */
2374 mddev->bitmap_info.offset = rt_is_raid0(rs->raid_type) ? 0 : to_sector(4096);
2375 mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2376
2377 if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2378 /* Retrieve device size stored in superblock to be prepared for shrink */
2379 rdev->sectors = le64_to_cpu(sb->sectors);
2380 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
2381 if (rdev->recovery_offset == MaxSector)
2382 set_bit(In_sync, &rdev->flags);
2383 /*
2384 * If no reshape in progress -> we're recovering single
2385 * disk(s) and have to set the device(s) to out-of-sync
2386 */
2387 else if (!rs_is_reshaping(rs))
2388 clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
2389 }
2390
2391 /*
2392 * If a device comes back, set it as not In_sync and no longer faulty.
2393 */
2394 if (test_and_clear_bit(Faulty, &rdev->flags)) {
2395 rdev->recovery_offset = 0;
2396 clear_bit(In_sync, &rdev->flags);
2397 rdev->saved_raid_disk = rdev->raid_disk;
2398 }
2399
2400 /* Reshape support -> restore repective data offsets */
2401 rdev->data_offset = le64_to_cpu(sb->data_offset);
2402 rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
2403
2404 return 0;
2405 }
2406
2407 /*
2408 * Analyse superblocks and select the freshest.
2409 */
2410 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2411 {
2412 int r;
2413 struct md_rdev *rdev, *freshest;
2414 struct mddev *mddev = &rs->md;
2415
2416 freshest = NULL;
2417 rdev_for_each(rdev, mddev) {
2418 if (test_bit(Journal, &rdev->flags))
2419 continue;
2420
2421 /*
2422 * Skipping super_load due to CTR_FLAG_SYNC will cause
2423 * the array to undergo initialization again as
2424 * though it were new. This is the intended effect
2425 * of the "sync" directive.
2426 *
2427 * With reshaping capability added, we must ensure that
2428 * that the "sync" directive is disallowed during the reshape.
2429 */
2430 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2431 continue;
2432
2433 if (!rdev->meta_bdev)
2434 continue;
2435
2436 r = super_load(rdev, freshest);
2437
2438 switch (r) {
2439 case 1:
2440 freshest = rdev;
2441 break;
2442 case 0:
2443 break;
2444 default:
2445 /* This is a failure to read the superblock from the metadata device. */
2446 /*
2447 * We have to keep any raid0 data/metadata device pairs or
2448 * the MD raid0 personality will fail to start the array.
2449 */
2450 if (rs_is_raid0(rs))
2451 continue;
2452
2453 /*
2454 * We keep the dm_devs to be able to emit the device tuple
2455 * properly on the table line in raid_status() (rather than
2456 * mistakenly acting as if '- -' got passed into the constructor).
2457 *
2458 * The rdev has to stay on the same_set list to allow for
2459 * the attempt to restore faulty devices on second resume.
2460 */
2461 rdev->raid_disk = rdev->saved_raid_disk = -1;
2462 break;
2463 }
2464 }
2465
2466 if (!freshest)
2467 return 0;
2468
2469 if (validate_raid_redundancy(rs)) {
2470 rs->ti->error = "Insufficient redundancy to activate array";
2471 return -EINVAL;
2472 }
2473
2474 /*
2475 * Validation of the freshest device provides the source of
2476 * validation for the remaining devices.
2477 */
2478 rs->ti->error = "Unable to assemble array: Invalid superblocks";
2479 if (super_validate(rs, freshest))
2480 return -EINVAL;
2481
2482 rdev_for_each(rdev, mddev)
2483 if (!test_bit(Journal, &rdev->flags) &&
2484 rdev != freshest &&
2485 super_validate(rs, rdev))
2486 return -EINVAL;
2487 return 0;
2488 }
2489
2490 /*
2491 * Adjust data_offset and new_data_offset on all disk members of @rs
2492 * for out of place reshaping if requested by contructor
2493 *
2494 * We need free space at the beginning of each raid disk for forward
2495 * and at the end for backward reshapes which userspace has to provide
2496 * via remapping/reordering of space.
2497 */
2498 static int rs_adjust_data_offsets(struct raid_set *rs)
2499 {
2500 sector_t data_offset = 0, new_data_offset = 0;
2501 struct md_rdev *rdev;
2502
2503 /* Constructor did not request data offset change */
2504 if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2505 if (!rs_is_reshapable(rs))
2506 goto out;
2507
2508 return 0;
2509 }
2510
2511 /* HM FIXME: get InSync raid_dev? */
2512 rdev = &rs->dev[0].rdev;
2513
2514 if (rs->delta_disks < 0) {
2515 /*
2516 * Removing disks (reshaping backwards):
2517 *
2518 * - before reshape: data is at offset 0 and free space
2519 * is at end of each component LV
2520 *
2521 * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2522 */
2523 data_offset = 0;
2524 new_data_offset = rs->data_offset;
2525
2526 } else if (rs->delta_disks > 0) {
2527 /*
2528 * Adding disks (reshaping forwards):
2529 *
2530 * - before reshape: data is at offset rs->data_offset != 0 and
2531 * free space is at begin of each component LV
2532 *
2533 * - after reshape: data is at offset 0 on each component LV
2534 */
2535 data_offset = rs->data_offset;
2536 new_data_offset = 0;
2537
2538 } else {
2539 /*
2540 * User space passes in 0 for data offset after having removed reshape space
2541 *
2542 * - or - (data offset != 0)
2543 *
2544 * Changing RAID layout or chunk size -> toggle offsets
2545 *
2546 * - before reshape: data is at offset rs->data_offset 0 and
2547 * free space is at end of each component LV
2548 * -or-
2549 * data is at offset rs->data_offset != 0 and
2550 * free space is at begin of each component LV
2551 *
2552 * - after reshape: data is at offset 0 if it was at offset != 0
2553 * or at offset != 0 if it was at offset 0
2554 * on each component LV
2555 *
2556 */
2557 data_offset = rs->data_offset ? rdev->data_offset : 0;
2558 new_data_offset = data_offset ? 0 : rs->data_offset;
2559 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2560 }
2561
2562 /*
2563 * Make sure we got a minimum amount of free sectors per device
2564 */
2565 if (rs->data_offset &&
2566 to_sector(i_size_read(rdev->bdev->bd_inode)) - rdev->sectors < MIN_FREE_RESHAPE_SPACE) {
2567 rs->ti->error = data_offset ? "No space for forward reshape" :
2568 "No space for backward reshape";
2569 return -ENOSPC;
2570 }
2571 out:
2572 /* Adjust data offsets on all rdevs but on any raid4/5/6 journal device */
2573 rdev_for_each(rdev, &rs->md) {
2574 if (!test_bit(Journal, &rdev->flags)) {
2575 rdev->data_offset = data_offset;
2576 rdev->new_data_offset = new_data_offset;
2577 }
2578 }
2579
2580 return 0;
2581 }
2582
2583 /* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2584 static void __reorder_raid_disk_indexes(struct raid_set *rs)
2585 {
2586 int i = 0;
2587 struct md_rdev *rdev;
2588
2589 rdev_for_each(rdev, &rs->md) {
2590 if (!test_bit(Journal, &rdev->flags)) {
2591 rdev->raid_disk = i++;
2592 rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2593 }
2594 }
2595 }
2596
2597 /*
2598 * Setup @rs for takeover by a different raid level
2599 */
2600 static int rs_setup_takeover(struct raid_set *rs)
2601 {
2602 struct mddev *mddev = &rs->md;
2603 struct md_rdev *rdev;
2604 unsigned int d = mddev->raid_disks = rs->raid_disks;
2605 sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2606
2607 if (rt_is_raid10(rs->raid_type)) {
2608 if (mddev->level == 0) {
2609 /* Userpace reordered disks -> adjust raid_disk indexes */
2610 __reorder_raid_disk_indexes(rs);
2611
2612 /* raid0 -> raid10_far layout */
2613 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2614 rs->raid10_copies);
2615 } else if (mddev->level == 1)
2616 /* raid1 -> raid10_near layout */
2617 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2618 rs->raid_disks);
2619 else
2620 return -EINVAL;
2621
2622 }
2623
2624 clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2625 mddev->recovery_cp = MaxSector;
2626
2627 while (d--) {
2628 rdev = &rs->dev[d].rdev;
2629
2630 if (test_bit(d, (void *) rs->rebuild_disks)) {
2631 clear_bit(In_sync, &rdev->flags);
2632 clear_bit(Faulty, &rdev->flags);
2633 mddev->recovery_cp = rdev->recovery_offset = 0;
2634 /* Bitmap has to be created when we do an "up" takeover */
2635 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2636 }
2637
2638 rdev->new_data_offset = new_data_offset;
2639 }
2640
2641 return 0;
2642 }
2643
2644 /* Prepare @rs for reshape */
2645 static int rs_prepare_reshape(struct raid_set *rs)
2646 {
2647 bool reshape;
2648 struct mddev *mddev = &rs->md;
2649
2650 if (rs_is_raid10(rs)) {
2651 if (rs->raid_disks != mddev->raid_disks &&
2652 __is_raid10_near(mddev->layout) &&
2653 rs->raid10_copies &&
2654 rs->raid10_copies != __raid10_near_copies(mddev->layout)) {
2655 /*
2656 * raid disk have to be multiple of data copies to allow this conversion,
2657 *
2658 * This is actually not a reshape it is a
2659 * rebuild of any additional mirrors per group
2660 */
2661 if (rs->raid_disks % rs->raid10_copies) {
2662 rs->ti->error = "Can't reshape raid10 mirror groups";
2663 return -EINVAL;
2664 }
2665
2666 /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2667 __reorder_raid_disk_indexes(rs);
2668 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2669 rs->raid10_copies);
2670 mddev->new_layout = mddev->layout;
2671 reshape = false;
2672 } else
2673 reshape = true;
2674
2675 } else if (rs_is_raid456(rs))
2676 reshape = true;
2677
2678 else if (rs_is_raid1(rs)) {
2679 if (rs->delta_disks) {
2680 /* Process raid1 via delta_disks */
2681 mddev->degraded = rs->delta_disks < 0 ? -rs->delta_disks : rs->delta_disks;
2682 reshape = true;
2683 } else {
2684 /* Process raid1 without delta_disks */
2685 mddev->raid_disks = rs->raid_disks;
2686 reshape = false;
2687 }
2688 } else {
2689 rs->ti->error = "Called with bogus raid type";
2690 return -EINVAL;
2691 }
2692
2693 if (reshape) {
2694 set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2695 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2696 } else if (mddev->raid_disks < rs->raid_disks)
2697 /* Create new superblocks and bitmaps, if any new disks */
2698 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2699
2700 return 0;
2701 }
2702
2703 /*
2704 *
2705 * - change raid layout
2706 * - change chunk size
2707 * - add disks
2708 * - remove disks
2709 */
2710 static int rs_setup_reshape(struct raid_set *rs)
2711 {
2712 int r = 0;
2713 unsigned int cur_raid_devs, d;
2714 struct mddev *mddev = &rs->md;
2715 struct md_rdev *rdev;
2716
2717 mddev->delta_disks = rs->delta_disks;
2718 cur_raid_devs = mddev->raid_disks;
2719
2720 /* Ignore impossible layout change whilst adding/removing disks */
2721 if (mddev->delta_disks &&
2722 mddev->layout != mddev->new_layout) {
2723 DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2724 mddev->new_layout = mddev->layout;
2725 }
2726
2727 /*
2728 * Adjust array size:
2729 *
2730 * - in case of adding disks, array size has
2731 * to grow after the disk adding reshape,
2732 * which'll hapen in the event handler;
2733 * reshape will happen forward, so space has to
2734 * be available at the beginning of each disk
2735 *
2736 * - in case of removing disks, array size
2737 * has to shrink before starting the reshape,
2738 * which'll happen here;
2739 * reshape will happen backward, so space has to
2740 * be available at the end of each disk
2741 *
2742 * - data_offset and new_data_offset are
2743 * adjusted for aforementioned out of place
2744 * reshaping based on userspace passing in
2745 * the "data_offset <sectors>" key/value
2746 * pair via the constructor
2747 */
2748
2749 /* Add disk(s) */
2750 if (rs->delta_disks > 0) {
2751 /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2752 for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2753 rdev = &rs->dev[d].rdev;
2754 clear_bit(In_sync, &rdev->flags);
2755
2756 /*
2757 * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2758 * by md, which'll store that erroneously in the superblock on reshape
2759 */
2760 rdev->saved_raid_disk = -1;
2761 rdev->raid_disk = d;
2762
2763 rdev->sectors = mddev->dev_sectors;
2764 rdev->recovery_offset = rs_is_raid1(rs) ? 0 : MaxSector;
2765 }
2766
2767 mddev->reshape_backwards = 0; /* adding disks -> forward reshape */
2768
2769 /* Remove disk(s) */
2770 } else if (rs->delta_disks < 0) {
2771 r = rs_set_dev_and_array_sectors(rs, true);
2772 mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2773
2774 /* Change layout and/or chunk size */
2775 } else {
2776 /*
2777 * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2778 *
2779 * keeping number of disks and do layout change ->
2780 *
2781 * toggle reshape_backward depending on data_offset:
2782 *
2783 * - free space upfront -> reshape forward
2784 *
2785 * - free space at the end -> reshape backward
2786 *
2787 *
2788 * This utilizes free reshape space avoiding the need
2789 * for userspace to move (parts of) LV segments in
2790 * case of layout/chunksize change (for disk
2791 * adding/removing reshape space has to be at
2792 * the proper address (see above with delta_disks):
2793 *
2794 * add disk(s) -> begin
2795 * remove disk(s)-> end
2796 */
2797 mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2798 }
2799
2800 return r;
2801 }
2802
2803 /*
2804 * Enable/disable discard support on RAID set depending on
2805 * RAID level and discard properties of underlying RAID members.
2806 */
2807 static void configure_discard_support(struct raid_set *rs)
2808 {
2809 int i;
2810 bool raid456;
2811 struct dm_target *ti = rs->ti;
2812
2813 /* Assume discards not supported until after checks below. */
2814 ti->discards_supported = false;
2815
2816 /* RAID level 4,5,6 require discard_zeroes_data for data integrity! */
2817 raid456 = (rs->md.level == 4 || rs->md.level == 5 || rs->md.level == 6);
2818
2819 for (i = 0; i < rs->raid_disks; i++) {
2820 struct request_queue *q;
2821
2822 if (!rs->dev[i].rdev.bdev)
2823 continue;
2824
2825 q = bdev_get_queue(rs->dev[i].rdev.bdev);
2826 if (!q || !blk_queue_discard(q))
2827 return;
2828
2829 if (raid456) {
2830 if (!q->limits.discard_zeroes_data)
2831 return;
2832 if (!devices_handle_discard_safely) {
2833 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2834 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2835 return;
2836 }
2837 }
2838 }
2839
2840 /* All RAID members properly support discards */
2841 ti->discards_supported = true;
2842
2843 /*
2844 * RAID1 and RAID10 personalities require bio splitting,
2845 * RAID0/4/5/6 don't and process large discard bios properly.
2846 */
2847 ti->split_discard_bios = !!(rs->md.level == 1 || rs->md.level == 10);
2848 ti->num_discard_bios = 1;
2849 }
2850
2851 /*
2852 * Construct a RAID0/1/10/4/5/6 mapping:
2853 * Args:
2854 * <raid_type> <#raid_params> <raid_params>{0,} \
2855 * <#raid_devs> [<meta_dev1> <dev1>]{1,}
2856 *
2857 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
2858 * details on possible <raid_params>.
2859 *
2860 * Userspace is free to initialize the metadata devices, hence the superblocks to
2861 * enforce recreation based on the passed in table parameters.
2862 *
2863 */
2864 static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2865 {
2866 int r;
2867 bool resize;
2868 struct raid_type *rt;
2869 unsigned int num_raid_params, num_raid_devs;
2870 sector_t calculated_dev_sectors;
2871 struct raid_set *rs = NULL;
2872 const char *arg;
2873 struct rs_layout rs_layout;
2874 struct dm_arg_set as = { argc, argv }, as_nrd;
2875 struct dm_arg _args[] = {
2876 { 0, as.argc, "Cannot understand number of raid parameters" },
2877 { 1, 254, "Cannot understand number of raid devices parameters" }
2878 };
2879
2880 /* Must have <raid_type> */
2881 arg = dm_shift_arg(&as);
2882 if (!arg) {
2883 ti->error = "No arguments";
2884 return -EINVAL;
2885 }
2886
2887 rt = get_raid_type(arg);
2888 if (!rt) {
2889 ti->error = "Unrecognised raid_type";
2890 return -EINVAL;
2891 }
2892
2893 /* Must have <#raid_params> */
2894 if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
2895 return -EINVAL;
2896
2897 /* number of raid device tupples <meta_dev data_dev> */
2898 as_nrd = as;
2899 dm_consume_args(&as_nrd, num_raid_params);
2900 _args[1].max = (as_nrd.argc - 1) / 2;
2901 if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
2902 return -EINVAL;
2903
2904 if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
2905 ti->error = "Invalid number of supplied raid devices";
2906 return -EINVAL;
2907 }
2908
2909 rs = raid_set_alloc(ti, rt, num_raid_devs);
2910 if (IS_ERR(rs))
2911 return PTR_ERR(rs);
2912
2913 r = parse_raid_params(rs, &as, num_raid_params);
2914 if (r)
2915 goto bad;
2916
2917 r = parse_dev_params(rs, &as);
2918 if (r)
2919 goto bad;
2920
2921 rs->md.sync_super = super_sync;
2922
2923 /*
2924 * Calculate ctr requested array and device sizes to allow
2925 * for superblock analysis needing device sizes defined.
2926 *
2927 * Any existing superblock will overwrite the array and device sizes
2928 */
2929 r = rs_set_dev_and_array_sectors(rs, false);
2930 if (r)
2931 goto bad;
2932
2933 calculated_dev_sectors = rs->md.dev_sectors;
2934
2935 /*
2936 * Backup any new raid set level, layout, ...
2937 * requested to be able to compare to superblock
2938 * members for conversion decisions.
2939 */
2940 rs_config_backup(rs, &rs_layout);
2941
2942 r = analyse_superblocks(ti, rs);
2943 if (r)
2944 goto bad;
2945
2946 resize = calculated_dev_sectors != __rdev_sectors(rs);
2947
2948 INIT_WORK(&rs->md.event_work, do_table_event);
2949 ti->private = rs;
2950 ti->num_flush_bios = 1;
2951
2952 /* Restore any requested new layout for conversion decision */
2953 rs_config_restore(rs, &rs_layout);
2954
2955 /*
2956 * Now that we have any superblock metadata available,
2957 * check for new, recovering, reshaping, to be taken over,
2958 * to be reshaped or an existing, unchanged raid set to
2959 * run in sequence.
2960 */
2961 if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
2962 /* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
2963 if (rs_is_raid6(rs) &&
2964 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
2965 ti->error = "'nosync' not allowed for new raid6 set";
2966 r = -EINVAL;
2967 goto bad;
2968 }
2969 rs_setup_recovery(rs, 0);
2970 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2971 rs_set_new(rs);
2972 } else if (rs_is_recovering(rs)) {
2973 /* A recovering raid set may be resized */
2974 ; /* skip setup rs */
2975 } else if (rs_is_reshaping(rs)) {
2976 /* Have to reject size change request during reshape */
2977 if (resize) {
2978 ti->error = "Can't resize a reshaping raid set";
2979 r = -EPERM;
2980 goto bad;
2981 }
2982 /* skip setup rs */
2983 } else if (rs_takeover_requested(rs)) {
2984 if (rs_is_reshaping(rs)) {
2985 ti->error = "Can't takeover a reshaping raid set";
2986 r = -EPERM;
2987 goto bad;
2988 }
2989
2990 /* We can't takeover a journaled raid4/5/6 */
2991 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
2992 ti->error = "Can't takeover a journaled raid4/5/6 set";
2993 r = -EPERM;
2994 goto bad;
2995 }
2996
2997 /*
2998 * If a takeover is needed, userspace sets any additional
2999 * devices to rebuild and we can check for a valid request here.
3000 *
3001 * If acceptible, set the level to the new requested
3002 * one, prohibit requesting recovery, allow the raid
3003 * set to run and store superblocks during resume.
3004 */
3005 r = rs_check_takeover(rs);
3006 if (r)
3007 goto bad;
3008
3009 r = rs_setup_takeover(rs);
3010 if (r)
3011 goto bad;
3012
3013 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3014 /* Takeover ain't recovery, so disable recovery */
3015 rs_setup_recovery(rs, MaxSector);
3016 rs_set_new(rs);
3017 } else if (rs_reshape_requested(rs)) {
3018 /*
3019 * No need to check for 'ongoing' takeover here, because takeover
3020 * is an instant operation as oposed to an ongoing reshape.
3021 */
3022
3023 /* We can't reshape a journaled raid4/5/6 */
3024 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3025 ti->error = "Can't reshape a journaled raid4/5/6 set";
3026 r = -EPERM;
3027 goto bad;
3028 }
3029
3030 /*
3031 * We can only prepare for a reshape here, because the
3032 * raid set needs to run to provide the repective reshape
3033 * check functions via its MD personality instance.
3034 *
3035 * So do the reshape check after md_run() succeeded.
3036 */
3037 r = rs_prepare_reshape(rs);
3038 if (r)
3039 return r;
3040
3041 /* Reshaping ain't recovery, so disable recovery */
3042 rs_setup_recovery(rs, MaxSector);
3043 rs_set_cur(rs);
3044 } else {
3045 /* May not set recovery when a device rebuild is requested */
3046 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3047 rs_setup_recovery(rs, MaxSector);
3048 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3049 } else
3050 rs_setup_recovery(rs, test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ?
3051 0 : (resize ? calculated_dev_sectors : MaxSector));
3052 rs_set_cur(rs);
3053 }
3054
3055 /* If constructor requested it, change data and new_data offsets */
3056 r = rs_adjust_data_offsets(rs);
3057 if (r)
3058 goto bad;
3059
3060 /* Start raid set read-only and assumed clean to change in raid_resume() */
3061 rs->md.ro = 1;
3062 rs->md.in_sync = 1;
3063 set_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
3064
3065 /* Has to be held on running the array */
3066 mddev_lock_nointr(&rs->md);
3067 r = md_run(&rs->md);
3068 rs->md.in_sync = 0; /* Assume already marked dirty */
3069
3070 if (r) {
3071 ti->error = "Failed to run raid array";
3072 mddev_unlock(&rs->md);
3073 goto bad;
3074 }
3075
3076 rs->callbacks.congested_fn = raid_is_congested;
3077 dm_table_add_target_callbacks(ti->table, &rs->callbacks);
3078
3079 mddev_suspend(&rs->md);
3080
3081 /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
3082 if (rs_is_raid456(rs)) {
3083 r = rs_set_raid456_stripe_cache(rs);
3084 if (r)
3085 goto bad_stripe_cache;
3086 }
3087
3088 /* Now do an early reshape check */
3089 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3090 r = rs_check_reshape(rs);
3091 if (r)
3092 goto bad_check_reshape;
3093
3094 /* Restore new, ctr requested layout to perform check */
3095 rs_config_restore(rs, &rs_layout);
3096
3097 if (rs->md.pers->start_reshape) {
3098 r = rs->md.pers->check_reshape(&rs->md);
3099 if (r) {
3100 ti->error = "Reshape check failed";
3101 goto bad_check_reshape;
3102 }
3103 }
3104 }
3105
3106 /* Disable/enable discard support on raid set. */
3107 configure_discard_support(rs);
3108
3109 mddev_unlock(&rs->md);
3110 return 0;
3111
3112 bad_stripe_cache:
3113 bad_check_reshape:
3114 md_stop(&rs->md);
3115 bad:
3116 raid_set_free(rs);
3117
3118 return r;
3119 }
3120
3121 static void raid_dtr(struct dm_target *ti)
3122 {
3123 struct raid_set *rs = ti->private;
3124
3125 list_del_init(&rs->callbacks.list);
3126 md_stop(&rs->md);
3127 raid_set_free(rs);
3128 }
3129
3130 static int raid_map(struct dm_target *ti, struct bio *bio)
3131 {
3132 struct raid_set *rs = ti->private;
3133 struct mddev *mddev = &rs->md;
3134
3135 /*
3136 * If we're reshaping to add disk(s)), ti->len and
3137 * mddev->array_sectors will differ during the process
3138 * (ti->len > mddev->array_sectors), so we have to requeue
3139 * bios with addresses > mddev->array_sectors here or
3140 * there will occur accesses past EOD of the component
3141 * data images thus erroring the raid set.
3142 */
3143 if (unlikely(bio_end_sector(bio) > mddev->array_sectors))
3144 return DM_MAPIO_REQUEUE;
3145
3146 mddev->pers->make_request(mddev, bio);
3147
3148 return DM_MAPIO_SUBMITTED;
3149 }
3150
3151 /* Return string describing the current sync action of @mddev */
3152 static const char *decipher_sync_action(struct mddev *mddev)
3153 {
3154 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3155 return "frozen";
3156
3157 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3158 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3159 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3160 return "reshape";
3161
3162 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3163 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3164 return "resync";
3165 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3166 return "check";
3167 return "repair";
3168 }
3169
3170 if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3171 return "recover";
3172 }
3173
3174 return "idle";
3175 }
3176
3177 /*
3178 * Return status string for @rdev
3179 *
3180 * Status characters:
3181 *
3182 * 'D' = Dead/Failed raid set component or raid4/5/6 journal device
3183 * 'a' = Alive but not in-sync
3184 * 'A' = Alive and in-sync raid set component or alive raid4/5/6 journal device
3185 * '-' = Non-existing device (i.e. uspace passed '- -' into the ctr)
3186 */
3187 static const char *__raid_dev_status(struct md_rdev *rdev, bool array_in_sync)
3188 {
3189 if (!rdev->bdev)
3190 return "-";
3191 else if (test_bit(Faulty, &rdev->flags))
3192 return "D";
3193 else if (test_bit(Journal, &rdev->flags))
3194 return "A";
3195 else if (!array_in_sync || !test_bit(In_sync, &rdev->flags))
3196 return "a";
3197 else
3198 return "A";
3199 }
3200
3201 /* Helper to return resync/reshape progress for @rs and @array_in_sync */
3202 static sector_t rs_get_progress(struct raid_set *rs,
3203 sector_t resync_max_sectors, bool *array_in_sync)
3204 {
3205 sector_t r, recovery_cp, curr_resync_completed;
3206 struct mddev *mddev = &rs->md;
3207
3208 curr_resync_completed = mddev->curr_resync_completed ?: mddev->recovery_cp;
3209 recovery_cp = mddev->recovery_cp;
3210 *array_in_sync = false;
3211
3212 if (rs_is_raid0(rs)) {
3213 r = resync_max_sectors;
3214 *array_in_sync = true;
3215
3216 } else {
3217 r = mddev->reshape_position;
3218
3219 /* Reshape is relative to the array size */
3220 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
3221 r != MaxSector) {
3222 if (r == MaxSector) {
3223 *array_in_sync = true;
3224 r = resync_max_sectors;
3225 } else {
3226 /* Got to reverse on backward reshape */
3227 if (mddev->reshape_backwards)
3228 r = mddev->array_sectors - r;
3229
3230 /* Devide by # of data stripes */
3231 sector_div(r, mddev_data_stripes(rs));
3232 }
3233
3234 /* Sync is relative to the component device size */
3235 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3236 r = curr_resync_completed;
3237 else
3238 r = recovery_cp;
3239
3240 if (r == MaxSector) {
3241 /*
3242 * Sync complete.
3243 */
3244 *array_in_sync = true;
3245 r = resync_max_sectors;
3246 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
3247 /*
3248 * If "check" or "repair" is occurring, the raid set has
3249 * undergone an initial sync and the health characters
3250 * should not be 'a' anymore.
3251 */
3252 *array_in_sync = true;
3253 } else {
3254 struct md_rdev *rdev;
3255
3256 /*
3257 * The raid set may be doing an initial sync, or it may
3258 * be rebuilding individual components. If all the
3259 * devices are In_sync, then it is the raid set that is
3260 * being initialized.
3261 */
3262 rdev_for_each(rdev, mddev)
3263 if (!test_bit(Journal, &rdev->flags) &&
3264 !test_bit(In_sync, &rdev->flags))
3265 *array_in_sync = true;
3266 #if 0
3267 r = 0; /* HM FIXME: TESTME: https://bugzilla.redhat.com/show_bug.cgi?id=1210637 ? */
3268 #endif
3269 }
3270 }
3271
3272 return r;
3273 }
3274
3275 /* Helper to return @dev name or "-" if !@dev */
3276 static const char *__get_dev_name(struct dm_dev *dev)
3277 {
3278 return dev ? dev->name : "-";
3279 }
3280
3281 static void raid_status(struct dm_target *ti, status_type_t type,
3282 unsigned int status_flags, char *result, unsigned int maxlen)
3283 {
3284 struct raid_set *rs = ti->private;
3285 struct mddev *mddev = &rs->md;
3286 struct r5conf *conf = mddev->private;
3287 int i, max_nr_stripes = conf ? conf->max_nr_stripes : 0;
3288 bool array_in_sync;
3289 unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3290 unsigned int sz = 0;
3291 unsigned int rebuild_disks;
3292 unsigned int write_mostly_params = 0;
3293 sector_t progress, resync_max_sectors, resync_mismatches;
3294 const char *sync_action;
3295 struct raid_type *rt;
3296
3297 switch (type) {
3298 case STATUSTYPE_INFO:
3299 /* *Should* always succeed */
3300 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3301 if (!rt)
3302 return;
3303
3304 DMEMIT("%s %d ", rt->name, mddev->raid_disks);
3305
3306 /* Access most recent mddev properties for status output */
3307 smp_rmb();
3308 /* Get sensible max sectors even if raid set not yet started */
3309 resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
3310 mddev->resync_max_sectors : mddev->dev_sectors;
3311 progress = rs_get_progress(rs, resync_max_sectors, &array_in_sync);
3312 resync_mismatches = (mddev->last_sync_action && !strcasecmp(mddev->last_sync_action, "check")) ?
3313 atomic64_read(&mddev->resync_mismatches) : 0;
3314 sync_action = decipher_sync_action(&rs->md);
3315
3316 /* HM FIXME: do we want another state char for raid0? It shows 'D'/'A'/'-' now */
3317 for (i = 0; i < rs->raid_disks; i++)
3318 DMEMIT(__raid_dev_status(&rs->dev[i].rdev, array_in_sync));
3319
3320 /*
3321 * In-sync/Reshape ratio:
3322 * The in-sync ratio shows the progress of:
3323 * - Initializing the raid set
3324 * - Rebuilding a subset of devices of the raid set
3325 * The user can distinguish between the two by referring
3326 * to the status characters.
3327 *
3328 * The reshape ratio shows the progress of
3329 * changing the raid layout or the number of
3330 * disks of a raid set
3331 */
3332 DMEMIT(" %llu/%llu", (unsigned long long) progress,
3333 (unsigned long long) resync_max_sectors);
3334
3335 /*
3336 * v1.5.0+:
3337 *
3338 * Sync action:
3339 * See Documentation/device-mapper/dm-raid.txt for
3340 * information on each of these states.
3341 */
3342 DMEMIT(" %s", sync_action);
3343
3344 /*
3345 * v1.5.0+:
3346 *
3347 * resync_mismatches/mismatch_cnt
3348 * This field shows the number of discrepancies found when
3349 * performing a "check" of the raid set.
3350 */
3351 DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3352
3353 /*
3354 * v1.9.0+:
3355 *
3356 * data_offset (needed for out of space reshaping)
3357 * This field shows the data offset into the data
3358 * image LV where the first stripes data starts.
3359 *
3360 * We keep data_offset equal on all raid disks of the set,
3361 * so retrieving it from the first raid disk is sufficient.
3362 */
3363 DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
3364
3365 /*
3366 * v1.10.0+:
3367 */
3368 DMEMIT(" %s", test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ?
3369 __raid_dev_status(&rs->journal_dev.rdev, 0) : "-");
3370 break;
3371
3372 case STATUSTYPE_TABLE:
3373 /* Report the table line string you would use to construct this raid set */
3374
3375 /* Calculate raid parameter count */
3376 for (i = 0; i < rs->raid_disks; i++)
3377 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3378 write_mostly_params += 2;
3379 rebuild_disks = memweight(rs->rebuild_disks, DISKS_ARRAY_ELEMS * sizeof(*rs->rebuild_disks));
3380 raid_param_cnt += rebuild_disks * 2 +
3381 write_mostly_params +
3382 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3383 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2 +
3384 (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ? 2 : 0);
3385 /* Emit table line */
3386 DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
3387 if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3388 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3389 raid10_md_layout_to_format(mddev->layout));
3390 if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3391 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3392 raid10_md_layout_to_copies(mddev->layout));
3393 if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3394 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
3395 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
3396 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
3397 if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3398 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3399 (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3400 if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3401 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3402 (unsigned long long) rs->data_offset);
3403 if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3404 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3405 mddev->bitmap_info.daemon_sleep);
3406 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3407 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3408 max(rs->delta_disks, mddev->delta_disks));
3409 if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3410 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3411 max_nr_stripes);
3412 if (rebuild_disks)
3413 for (i = 0; i < rs->raid_disks; i++)
3414 if (test_bit(rs->dev[i].rdev.raid_disk, (void *) rs->rebuild_disks))
3415 DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD),
3416 rs->dev[i].rdev.raid_disk);
3417 if (write_mostly_params)
3418 for (i = 0; i < rs->raid_disks; i++)
3419 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3420 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3421 rs->dev[i].rdev.raid_disk);
3422 if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
3423 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
3424 mddev->bitmap_info.max_write_behind);
3425 if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3426 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3427 mddev->sync_speed_max);
3428 if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3429 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3430 mddev->sync_speed_min);
3431 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags))
3432 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV),
3433 __get_dev_name(rs->journal_dev.dev));
3434 DMEMIT(" %d", rs->raid_disks);
3435 for (i = 0; i < rs->raid_disks; i++)
3436 DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
3437 __get_dev_name(rs->dev[i].data_dev));
3438 }
3439 }
3440
3441 static int raid_message(struct dm_target *ti, unsigned int argc, char **argv)
3442 {
3443 struct raid_set *rs = ti->private;
3444 struct mddev *mddev = &rs->md;
3445
3446 if (!mddev->pers || !mddev->pers->sync_request)
3447 return -EINVAL;
3448
3449 if (!strcasecmp(argv[0], "frozen"))
3450 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3451 else
3452 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3453
3454 if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
3455 if (mddev->sync_thread) {
3456 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3457 md_reap_sync_thread(mddev);
3458 }
3459 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3460 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3461 return -EBUSY;
3462 else if (!strcasecmp(argv[0], "resync"))
3463 ; /* MD_RECOVERY_NEEDED set below */
3464 else if (!strcasecmp(argv[0], "recover"))
3465 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3466 else {
3467 if (!strcasecmp(argv[0], "check")) {
3468 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3469 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3470 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3471 } else if (!strcasecmp(argv[0], "repair")) {
3472 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3473 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3474 } else
3475 return -EINVAL;
3476 }
3477 if (mddev->ro == 2) {
3478 /* A write to sync_action is enough to justify
3479 * canceling read-auto mode
3480 */
3481 mddev->ro = 0;
3482 if (!mddev->suspended && mddev->sync_thread)
3483 md_wakeup_thread(mddev->sync_thread);
3484 }
3485 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3486 if (!mddev->suspended && mddev->thread)
3487 md_wakeup_thread(mddev->thread);
3488
3489 return 0;
3490 }
3491
3492 static int raid_iterate_devices(struct dm_target *ti,
3493 iterate_devices_callout_fn fn, void *data)
3494 {
3495 struct raid_set *rs = ti->private;
3496 unsigned int i;
3497 int r = 0;
3498
3499 for (i = 0; !r && i < rs->md.raid_disks; i++)
3500 if (rs->dev[i].data_dev)
3501 r = fn(ti,
3502 rs->dev[i].data_dev,
3503 0, /* No offset on data devs */
3504 rs->md.dev_sectors,
3505 data);
3506
3507 return r;
3508 }
3509
3510 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3511 {
3512 struct raid_set *rs = ti->private;
3513 unsigned int chunk_size = to_bytes(rs->md.chunk_sectors);
3514
3515 blk_limits_io_min(limits, chunk_size);
3516 blk_limits_io_opt(limits, chunk_size * mddev_data_stripes(rs));
3517 }
3518
3519 static void raid_presuspend(struct dm_target *ti)
3520 {
3521 struct raid_set *rs = ti->private;
3522
3523 md_stop_writes(&rs->md);
3524 }
3525
3526 static void raid_postsuspend(struct dm_target *ti)
3527 {
3528 struct raid_set *rs = ti->private;
3529
3530 if (!rs->md.suspended)
3531 mddev_suspend(&rs->md);
3532
3533 rs->md.ro = 1;
3534 }
3535
3536 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
3537 {
3538 int i;
3539 uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
3540 unsigned long flags;
3541 bool cleared = false;
3542 struct dm_raid_superblock *sb;
3543 struct mddev *mddev = &rs->md;
3544 struct md_rdev *r;
3545
3546 /* RAID personalities have to provide hot add/remove methods or we need to bail out. */
3547 if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
3548 return;
3549
3550 memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
3551
3552 for (i = 0; i < mddev->raid_disks; i++) {
3553 r = &rs->dev[i].rdev;
3554 /* HM FIXME: enhance journal device recovery processing */
3555 if (test_bit(Journal, &r->flags))
3556 continue;
3557
3558 if (test_bit(Faulty, &r->flags) &&
3559 r->meta_bdev && !read_disk_sb(r, r->sb_size, true)) {
3560 DMINFO("Faulty %s device #%d has readable super block."
3561 " Attempting to revive it.",
3562 rs->raid_type->name, i);
3563
3564 /*
3565 * Faulty bit may be set, but sometimes the array can
3566 * be suspended before the personalities can respond
3567 * by removing the device from the array (i.e. calling
3568 * 'hot_remove_disk'). If they haven't yet removed
3569 * the failed device, its 'raid_disk' number will be
3570 * '>= 0' - meaning we must call this function
3571 * ourselves.
3572 */
3573 flags = r->flags;
3574 clear_bit(In_sync, &r->flags); /* Mandatory for hot remove. */
3575 if (r->raid_disk >= 0) {
3576 if (mddev->pers->hot_remove_disk(mddev, r)) {
3577 /* Failed to revive this device, try next */
3578 r->flags = flags;
3579 continue;
3580 }
3581 } else
3582 r->raid_disk = r->saved_raid_disk = i;
3583
3584 clear_bit(Faulty, &r->flags);
3585 clear_bit(WriteErrorSeen, &r->flags);
3586
3587 if (mddev->pers->hot_add_disk(mddev, r)) {
3588 /* Failed to revive this device, try next */
3589 r->raid_disk = r->saved_raid_disk = -1;
3590 r->flags = flags;
3591 } else {
3592 clear_bit(In_sync, &r->flags);
3593 r->recovery_offset = 0;
3594 set_bit(i, (void *) cleared_failed_devices);
3595 cleared = true;
3596 }
3597 }
3598 }
3599
3600 /* If any failed devices could be cleared, update all sbs failed_devices bits */
3601 if (cleared) {
3602 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
3603
3604 rdev_for_each(r, &rs->md) {
3605 if (test_bit(Journal, &r->flags))
3606 continue;
3607
3608 sb = page_address(r->sb_page);
3609 sb_retrieve_failed_devices(sb, failed_devices);
3610
3611 for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
3612 failed_devices[i] &= ~cleared_failed_devices[i];
3613
3614 sb_update_failed_devices(sb, failed_devices);
3615 }
3616 }
3617 }
3618
3619 static int __load_dirty_region_bitmap(struct raid_set *rs)
3620 {
3621 int r = 0;
3622
3623 /* Try loading the bitmap unless "raid0", which does not have one */
3624 if (!rs_is_raid0(rs) &&
3625 !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
3626 r = bitmap_load(&rs->md);
3627 if (r)
3628 DMERR("Failed to load bitmap");
3629 }
3630
3631 return r;
3632 }
3633
3634 /* Enforce updating all superblocks */
3635 static void rs_update_sbs(struct raid_set *rs)
3636 {
3637 struct mddev *mddev = &rs->md;
3638 int ro = mddev->ro;
3639
3640 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3641 mddev->ro = 0;
3642 md_update_sb(mddev, 1);
3643 mddev->ro = ro;
3644 }
3645
3646 /*
3647 * Reshape changes raid algorithm of @rs to new one within personality
3648 * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3649 * disks from a raid set thus growing/shrinking it or resizes the set
3650 *
3651 * Call mddev_lock_nointr() before!
3652 */
3653 static int rs_start_reshape(struct raid_set *rs)
3654 {
3655 int r;
3656 struct mddev *mddev = &rs->md;
3657 struct md_personality *pers = mddev->pers;
3658
3659 r = rs_setup_reshape(rs);
3660 if (r)
3661 return r;
3662
3663 /* Need to be resumed to be able to start reshape, recovery is frozen until raid_resume() though */
3664 if (mddev->suspended)
3665 mddev_resume(mddev);
3666
3667 /*
3668 * Check any reshape constraints enforced by the personalility
3669 *
3670 * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3671 */
3672 r = pers->check_reshape(mddev);
3673 if (r) {
3674 rs->ti->error = "pers->check_reshape() failed";
3675 return r;
3676 }
3677
3678 /*
3679 * Personality may not provide start reshape method in which
3680 * case check_reshape above has already covered everything
3681 */
3682 if (pers->start_reshape) {
3683 r = pers->start_reshape(mddev);
3684 if (r) {
3685 rs->ti->error = "pers->start_reshape() failed";
3686 return r;
3687 }
3688 }
3689
3690 /* Suspend because a resume will happen in raid_resume() */
3691 if (!mddev->suspended)
3692 mddev_suspend(mddev);
3693
3694 /*
3695 * Now reshape got set up, update superblocks to
3696 * reflect the fact so that a table reload will
3697 * access proper superblock content in the ctr.
3698 */
3699 rs_update_sbs(rs);
3700
3701 return 0;
3702 }
3703
3704 static int raid_preresume(struct dm_target *ti)
3705 {
3706 int r;
3707 struct raid_set *rs = ti->private;
3708 struct mddev *mddev = &rs->md;
3709
3710 /* This is a resume after a suspend of the set -> it's already started */
3711 if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
3712 return 0;
3713
3714 /*
3715 * The superblocks need to be updated on disk if the
3716 * array is new or new devices got added (thus zeroed
3717 * out by userspace) or __load_dirty_region_bitmap
3718 * will overwrite them in core with old data or fail.
3719 */
3720 if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
3721 rs_update_sbs(rs);
3722
3723 /* Load the bitmap from disk unless raid0 */
3724 r = __load_dirty_region_bitmap(rs);
3725 if (r)
3726 return r;
3727
3728 /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) */
3729 if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) && mddev->bitmap &&
3730 mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)) {
3731 r = bitmap_resize(mddev->bitmap, mddev->dev_sectors,
3732 to_bytes(rs->requested_bitmap_chunk_sectors), 0);
3733 if (r)
3734 DMERR("Failed to resize bitmap");
3735 }
3736
3737 /* Check for any resize/reshape on @rs and adjust/initiate */
3738 /* Be prepared for mddev_resume() in raid_resume() */
3739 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3740 if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
3741 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3742 mddev->resync_min = mddev->recovery_cp;
3743 }
3744
3745 rs_set_capacity(rs);
3746
3747 /* Check for any reshape request unless new raid set */
3748 if (test_and_clear_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3749 /* Initiate a reshape. */
3750 mddev_lock_nointr(mddev);
3751 r = rs_start_reshape(rs);
3752 mddev_unlock(mddev);
3753 if (r)
3754 DMWARN("Failed to check/start reshape, continuing without change");
3755 r = 0;
3756 }
3757
3758 return r;
3759 }
3760
3761 static void raid_resume(struct dm_target *ti)
3762 {
3763 struct raid_set *rs = ti->private;
3764 struct mddev *mddev = &rs->md;
3765
3766 if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
3767 /*
3768 * A secondary resume while the device is active.
3769 * Take this opportunity to check whether any failed
3770 * devices are reachable again.
3771 */
3772 attempt_restore_of_faulty_devices(rs);
3773 }
3774
3775 mddev->ro = 0;
3776 mddev->in_sync = 0;
3777
3778 /*
3779 * Keep the RAID set frozen if reshape/rebuild flags are set.
3780 * The RAID set is unfrozen once the next table load/resume,
3781 * which clears the reshape/rebuild flags, occurs.
3782 * This ensures that the constructor for the inactive table
3783 * retrieves an up-to-date reshape_position.
3784 */
3785 if (!(rs->ctr_flags & RESUME_STAY_FROZEN_FLAGS))
3786 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3787
3788 if (mddev->suspended)
3789 mddev_resume(mddev);
3790 }
3791
3792 static struct target_type raid_target = {
3793 .name = "raid",
3794 .version = {1, 10, 1},
3795 .module = THIS_MODULE,
3796 .ctr = raid_ctr,
3797 .dtr = raid_dtr,
3798 .map = raid_map,
3799 .status = raid_status,
3800 .message = raid_message,
3801 .iterate_devices = raid_iterate_devices,
3802 .io_hints = raid_io_hints,
3803 .presuspend = raid_presuspend,
3804 .postsuspend = raid_postsuspend,
3805 .preresume = raid_preresume,
3806 .resume = raid_resume,
3807 };
3808
3809 static int __init dm_raid_init(void)
3810 {
3811 DMINFO("Loading target version %u.%u.%u",
3812 raid_target.version[0],
3813 raid_target.version[1],
3814 raid_target.version[2]);
3815 return dm_register_target(&raid_target);
3816 }
3817
3818 static void __exit dm_raid_exit(void)
3819 {
3820 dm_unregister_target(&raid_target);
3821 }
3822
3823 module_init(dm_raid_init);
3824 module_exit(dm_raid_exit);
3825
3826 module_param(devices_handle_discard_safely, bool, 0644);
3827 MODULE_PARM_DESC(devices_handle_discard_safely,
3828 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
3829
3830 MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
3831 MODULE_ALIAS("dm-raid0");
3832 MODULE_ALIAS("dm-raid1");
3833 MODULE_ALIAS("dm-raid10");
3834 MODULE_ALIAS("dm-raid4");
3835 MODULE_ALIAS("dm-raid5");
3836 MODULE_ALIAS("dm-raid6");
3837 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
3838 MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
3839 MODULE_LICENSE("GPL");
Linux with added FPC-III support