2 * Copyright (C) 2010-2011 Neil Brown
3 * Copyright (C) 2010-2018 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
8 #include <linux/slab.h>
9 #include <linux/module.h>
15 #include "md-bitmap.h"
17 #include <linux/device-mapper.h>
19 #define DM_MSG_PREFIX "raid"
20 #define MAX_RAID_DEVICES 253 /* md-raid kernel limit */
23 * Minimum sectors of free reshape space per raid device
25 #define MIN_FREE_RESHAPE_SPACE to_sector(4*4096)
28 * Minimum journal space 4 MiB in sectors.
30 #define MIN_RAID456_JOURNAL_SPACE (4*2048)
32 static bool devices_handle_discard_safely = false;
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.
38 #define FirstUse 10 /* rdev flag */
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
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 ------|
53 struct dm_dev *meta_dev;
54 struct dm_dev *data_dev;
59 * Bits for establishing rs->ctr_flags
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 */
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! */
82 #define __CTR_FLAG_JOURNAL_DEV 15 /* 2 */ /* Only with raid4/5/6 (journal device)! */
85 #define __CTR_FLAG_JOURNAL_MODE 16 /* 2 */ /* Only with raid4/5/6 (journal mode)! */
88 * Flags for rs->ctr_flags field.
90 #define CTR_FLAG_SYNC (1 << __CTR_FLAG_SYNC)
91 #define CTR_FLAG_NOSYNC (1 << __CTR_FLAG_NOSYNC)
92 #define CTR_FLAG_REBUILD (1 << __CTR_FLAG_REBUILD)
93 #define CTR_FLAG_DAEMON_SLEEP (1 << __CTR_FLAG_DAEMON_SLEEP)
94 #define CTR_FLAG_MIN_RECOVERY_RATE (1 << __CTR_FLAG_MIN_RECOVERY_RATE)
95 #define CTR_FLAG_MAX_RECOVERY_RATE (1 << __CTR_FLAG_MAX_RECOVERY_RATE)
96 #define CTR_FLAG_MAX_WRITE_BEHIND (1 << __CTR_FLAG_MAX_WRITE_BEHIND)
97 #define CTR_FLAG_WRITE_MOSTLY (1 << __CTR_FLAG_WRITE_MOSTLY)
98 #define CTR_FLAG_STRIPE_CACHE (1 << __CTR_FLAG_STRIPE_CACHE)
99 #define CTR_FLAG_REGION_SIZE (1 << __CTR_FLAG_REGION_SIZE)
100 #define CTR_FLAG_RAID10_COPIES (1 << __CTR_FLAG_RAID10_COPIES)
101 #define CTR_FLAG_RAID10_FORMAT (1 << __CTR_FLAG_RAID10_FORMAT)
102 #define CTR_FLAG_DELTA_DISKS (1 << __CTR_FLAG_DELTA_DISKS)
103 #define CTR_FLAG_DATA_OFFSET (1 << __CTR_FLAG_DATA_OFFSET)
104 #define CTR_FLAG_RAID10_USE_NEAR_SETS (1 << __CTR_FLAG_RAID10_USE_NEAR_SETS)
105 #define CTR_FLAG_JOURNAL_DEV (1 << __CTR_FLAG_JOURNAL_DEV)
106 #define CTR_FLAG_JOURNAL_MODE (1 << __CTR_FLAG_JOURNAL_MODE)
109 * Definitions of various constructor flags to
110 * be used in checks of valid / invalid flags
113 /* Define all any sync flags */
114 #define CTR_FLAGS_ANY_SYNC (CTR_FLAG_SYNC | CTR_FLAG_NOSYNC)
116 /* Define flags for options without argument (e.g. 'nosync') */
117 #define CTR_FLAG_OPTIONS_NO_ARGS (CTR_FLAGS_ANY_SYNC | \
118 CTR_FLAG_RAID10_USE_NEAR_SETS)
120 /* Define flags for options with one argument (e.g. 'delta_disks +2') */
121 #define CTR_FLAG_OPTIONS_ONE_ARG (CTR_FLAG_REBUILD | \
122 CTR_FLAG_WRITE_MOSTLY | \
123 CTR_FLAG_DAEMON_SLEEP | \
124 CTR_FLAG_MIN_RECOVERY_RATE | \
125 CTR_FLAG_MAX_RECOVERY_RATE | \
126 CTR_FLAG_MAX_WRITE_BEHIND | \
127 CTR_FLAG_STRIPE_CACHE | \
128 CTR_FLAG_REGION_SIZE | \
129 CTR_FLAG_RAID10_COPIES | \
130 CTR_FLAG_RAID10_FORMAT | \
131 CTR_FLAG_DELTA_DISKS | \
132 CTR_FLAG_DATA_OFFSET)
134 /* Valid options definitions per raid level... */
136 /* "raid0" does only accept data offset */
137 #define RAID0_VALID_FLAGS (CTR_FLAG_DATA_OFFSET)
139 /* "raid1" does not accept stripe cache, data offset, delta_disks or any raid10 options */
140 #define RAID1_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
142 CTR_FLAG_WRITE_MOSTLY | \
143 CTR_FLAG_DAEMON_SLEEP | \
144 CTR_FLAG_MIN_RECOVERY_RATE | \
145 CTR_FLAG_MAX_RECOVERY_RATE | \
146 CTR_FLAG_MAX_WRITE_BEHIND | \
147 CTR_FLAG_REGION_SIZE | \
148 CTR_FLAG_DELTA_DISKS | \
149 CTR_FLAG_DATA_OFFSET)
151 /* "raid10" does not accept any raid1 or stripe cache options */
152 #define RAID10_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
154 CTR_FLAG_DAEMON_SLEEP | \
155 CTR_FLAG_MIN_RECOVERY_RATE | \
156 CTR_FLAG_MAX_RECOVERY_RATE | \
157 CTR_FLAG_REGION_SIZE | \
158 CTR_FLAG_RAID10_COPIES | \
159 CTR_FLAG_RAID10_FORMAT | \
160 CTR_FLAG_DELTA_DISKS | \
161 CTR_FLAG_DATA_OFFSET | \
162 CTR_FLAG_RAID10_USE_NEAR_SETS)
165 * "raid4/5/6" do not accept any raid1 or raid10 specific options
167 * "raid6" does not accept "nosync", because it is not guaranteed
168 * that both parity and q-syndrome are being written properly with
171 #define RAID45_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
173 CTR_FLAG_DAEMON_SLEEP | \
174 CTR_FLAG_MIN_RECOVERY_RATE | \
175 CTR_FLAG_MAX_RECOVERY_RATE | \
176 CTR_FLAG_STRIPE_CACHE | \
177 CTR_FLAG_REGION_SIZE | \
178 CTR_FLAG_DELTA_DISKS | \
179 CTR_FLAG_DATA_OFFSET | \
180 CTR_FLAG_JOURNAL_DEV | \
181 CTR_FLAG_JOURNAL_MODE)
183 #define RAID6_VALID_FLAGS (CTR_FLAG_SYNC | \
185 CTR_FLAG_DAEMON_SLEEP | \
186 CTR_FLAG_MIN_RECOVERY_RATE | \
187 CTR_FLAG_MAX_RECOVERY_RATE | \
188 CTR_FLAG_STRIPE_CACHE | \
189 CTR_FLAG_REGION_SIZE | \
190 CTR_FLAG_DELTA_DISKS | \
191 CTR_FLAG_DATA_OFFSET | \
192 CTR_FLAG_JOURNAL_DEV | \
193 CTR_FLAG_JOURNAL_MODE)
194 /* ...valid options definitions per raid level */
197 * Flags for rs->runtime_flags field
198 * (RT_FLAG prefix meaning "runtime flag")
200 * These are all internal and used to define runtime state,
201 * e.g. to prevent another resume from preresume processing
202 * the raid set all over again.
204 #define RT_FLAG_RS_PRERESUMED 0
205 #define RT_FLAG_RS_RESUMED 1
206 #define RT_FLAG_RS_BITMAP_LOADED 2
207 #define RT_FLAG_UPDATE_SBS 3
208 #define RT_FLAG_RESHAPE_RS 4
209 #define RT_FLAG_RS_SUSPENDED 5
210 #define RT_FLAG_RS_IN_SYNC 6
211 #define RT_FLAG_RS_RESYNCING 7
212 #define RT_FLAG_RS_GROW 8
214 /* Array elements of 64 bit needed for rebuild/failed disk bits */
215 #define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
218 * raid set level, layout and chunk sectors backup/restore
223 int new_chunk_sectors;
227 struct dm_target *ti;
229 uint32_t stripe_cache_entries;
230 unsigned long ctr_flags;
231 unsigned long runtime_flags;
233 uint64_t rebuild_disks[DISKS_ARRAY_ELEMS];
239 int requested_bitmap_chunk_sectors;
242 struct raid_type *raid_type;
243 struct dm_target_callbacks callbacks;
245 sector_t array_sectors;
246 sector_t dev_sectors;
248 /* Optional raid4/5/6 journal device */
255 struct raid_dev dev[0];
258 static void rs_config_backup(struct raid_set *rs, struct rs_layout *l)
260 struct mddev *mddev = &rs->md;
262 l->new_level = mddev->new_level;
263 l->new_layout = mddev->new_layout;
264 l->new_chunk_sectors = mddev->new_chunk_sectors;
267 static void rs_config_restore(struct raid_set *rs, struct rs_layout *l)
269 struct mddev *mddev = &rs->md;
271 mddev->new_level = l->new_level;
272 mddev->new_layout = l->new_layout;
273 mddev->new_chunk_sectors = l->new_chunk_sectors;
276 /* raid10 algorithms (i.e. formats) */
277 #define ALGORITHM_RAID10_DEFAULT 0
278 #define ALGORITHM_RAID10_NEAR 1
279 #define ALGORITHM_RAID10_OFFSET 2
280 #define ALGORITHM_RAID10_FAR 3
282 /* Supported raid types and properties. */
283 static struct raid_type {
284 const char *name; /* RAID algorithm. */
285 const char *descr; /* Descriptor text for logging. */
286 const unsigned int parity_devs; /* # of parity devices. */
287 const unsigned int minimal_devs;/* minimal # of devices in set. */
288 const unsigned int level; /* RAID level. */
289 const unsigned int algorithm; /* RAID algorithm. */
291 {"raid0", "raid0 (striping)", 0, 2, 0, 0 /* NONE */},
292 {"raid1", "raid1 (mirroring)", 0, 2, 1, 0 /* NONE */},
293 {"raid10_far", "raid10 far (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_FAR},
294 {"raid10_offset", "raid10 offset (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_OFFSET},
295 {"raid10_near", "raid10 near (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_NEAR},
296 {"raid10", "raid10 (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_DEFAULT},
297 {"raid4", "raid4 (dedicated first parity disk)", 1, 2, 5, ALGORITHM_PARITY_0}, /* raid4 layout = raid5_0 */
298 {"raid5_n", "raid5 (dedicated last parity disk)", 1, 2, 5, ALGORITHM_PARITY_N},
299 {"raid5_ls", "raid5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
300 {"raid5_rs", "raid5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
301 {"raid5_la", "raid5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
302 {"raid5_ra", "raid5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
303 {"raid6_zr", "raid6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
304 {"raid6_nr", "raid6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
305 {"raid6_nc", "raid6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE},
306 {"raid6_n_6", "raid6 (dedicated parity/Q n/6)", 2, 4, 6, ALGORITHM_PARITY_N_6},
307 {"raid6_ls_6", "raid6 (left symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_SYMMETRIC_6},
308 {"raid6_rs_6", "raid6 (right symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_SYMMETRIC_6},
309 {"raid6_la_6", "raid6 (left asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_ASYMMETRIC_6},
310 {"raid6_ra_6", "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_ASYMMETRIC_6}
313 /* True, if @v is in inclusive range [@min, @max] */
314 static bool __within_range(long v, long min, long max)
316 return v >= min && v <= max;
319 /* All table line arguments are defined here */
320 static struct arg_name_flag {
321 const unsigned long flag;
323 } __arg_name_flags[] = {
324 { CTR_FLAG_SYNC, "sync"},
325 { CTR_FLAG_NOSYNC, "nosync"},
326 { CTR_FLAG_REBUILD, "rebuild"},
327 { CTR_FLAG_DAEMON_SLEEP, "daemon_sleep"},
328 { CTR_FLAG_MIN_RECOVERY_RATE, "min_recovery_rate"},
329 { CTR_FLAG_MAX_RECOVERY_RATE, "max_recovery_rate"},
330 { CTR_FLAG_MAX_WRITE_BEHIND, "max_write_behind"},
331 { CTR_FLAG_WRITE_MOSTLY, "write_mostly"},
332 { CTR_FLAG_STRIPE_CACHE, "stripe_cache"},
333 { CTR_FLAG_REGION_SIZE, "region_size"},
334 { CTR_FLAG_RAID10_COPIES, "raid10_copies"},
335 { CTR_FLAG_RAID10_FORMAT, "raid10_format"},
336 { CTR_FLAG_DATA_OFFSET, "data_offset"},
337 { CTR_FLAG_DELTA_DISKS, "delta_disks"},
338 { CTR_FLAG_RAID10_USE_NEAR_SETS, "raid10_use_near_sets"},
339 { CTR_FLAG_JOURNAL_DEV, "journal_dev" },
340 { CTR_FLAG_JOURNAL_MODE, "journal_mode" },
343 /* Return argument name string for given @flag */
344 static const char *dm_raid_arg_name_by_flag(const uint32_t flag)
346 if (hweight32(flag) == 1) {
347 struct arg_name_flag *anf = __arg_name_flags + ARRAY_SIZE(__arg_name_flags);
349 while (anf-- > __arg_name_flags)
350 if (flag & anf->flag)
354 DMERR("%s called with more than one flag!", __func__);
359 /* Define correlation of raid456 journal cache modes and dm-raid target line parameters */
363 } _raid456_journal_mode[] = {
364 { R5C_JOURNAL_MODE_WRITE_THROUGH , "writethrough" },
365 { R5C_JOURNAL_MODE_WRITE_BACK , "writeback" }
368 /* Return MD raid4/5/6 journal mode for dm @journal_mode one */
369 static int dm_raid_journal_mode_to_md(const char *mode)
371 int m = ARRAY_SIZE(_raid456_journal_mode);
374 if (!strcasecmp(mode, _raid456_journal_mode[m].param))
375 return _raid456_journal_mode[m].mode;
380 /* Return dm-raid raid4/5/6 journal mode string for @mode */
381 static const char *md_journal_mode_to_dm_raid(const int mode)
383 int m = ARRAY_SIZE(_raid456_journal_mode);
386 if (mode == _raid456_journal_mode[m].mode)
387 return _raid456_journal_mode[m].param;
393 * Bool helpers to test for various raid levels of a raid set.
394 * It's level as reported by the superblock rather than
395 * the requested raid_type passed to the constructor.
397 /* Return true, if raid set in @rs is raid0 */
398 static bool rs_is_raid0(struct raid_set *rs)
400 return !rs->md.level;
403 /* Return true, if raid set in @rs is raid1 */
404 static bool rs_is_raid1(struct raid_set *rs)
406 return rs->md.level == 1;
409 /* Return true, if raid set in @rs is raid10 */
410 static bool rs_is_raid10(struct raid_set *rs)
412 return rs->md.level == 10;
415 /* Return true, if raid set in @rs is level 6 */
416 static bool rs_is_raid6(struct raid_set *rs)
418 return rs->md.level == 6;
421 /* Return true, if raid set in @rs is level 4, 5 or 6 */
422 static bool rs_is_raid456(struct raid_set *rs)
424 return __within_range(rs->md.level, 4, 6);
427 /* Return true, if raid set in @rs is reshapable */
428 static bool __is_raid10_far(int layout);
429 static bool rs_is_reshapable(struct raid_set *rs)
431 return rs_is_raid456(rs) ||
432 (rs_is_raid10(rs) && !__is_raid10_far(rs->md.new_layout));
435 /* Return true, if raid set in @rs is recovering */
436 static bool rs_is_recovering(struct raid_set *rs)
438 return rs->md.recovery_cp < rs->md.dev_sectors;
441 /* Return true, if raid set in @rs is reshaping */
442 static bool rs_is_reshaping(struct raid_set *rs)
444 return rs->md.reshape_position != MaxSector;
448 * bool helpers to test for various raid levels of a raid type @rt
451 /* Return true, if raid type in @rt is raid0 */
452 static bool rt_is_raid0(struct raid_type *rt)
457 /* Return true, if raid type in @rt is raid1 */
458 static bool rt_is_raid1(struct raid_type *rt)
460 return rt->level == 1;
463 /* Return true, if raid type in @rt is raid10 */
464 static bool rt_is_raid10(struct raid_type *rt)
466 return rt->level == 10;
469 /* Return true, if raid type in @rt is raid4/5 */
470 static bool rt_is_raid45(struct raid_type *rt)
472 return __within_range(rt->level, 4, 5);
475 /* Return true, if raid type in @rt is raid6 */
476 static bool rt_is_raid6(struct raid_type *rt)
478 return rt->level == 6;
481 /* Return true, if raid type in @rt is raid4/5/6 */
482 static bool rt_is_raid456(struct raid_type *rt)
484 return __within_range(rt->level, 4, 6);
486 /* END: raid level bools */
488 /* Return valid ctr flags for the raid level of @rs */
489 static unsigned long __valid_flags(struct raid_set *rs)
491 if (rt_is_raid0(rs->raid_type))
492 return RAID0_VALID_FLAGS;
493 else if (rt_is_raid1(rs->raid_type))
494 return RAID1_VALID_FLAGS;
495 else if (rt_is_raid10(rs->raid_type))
496 return RAID10_VALID_FLAGS;
497 else if (rt_is_raid45(rs->raid_type))
498 return RAID45_VALID_FLAGS;
499 else if (rt_is_raid6(rs->raid_type))
500 return RAID6_VALID_FLAGS;
506 * Check for valid flags set on @rs
508 * Has to be called after parsing of the ctr flags!
510 static int rs_check_for_valid_flags(struct raid_set *rs)
512 if (rs->ctr_flags & ~__valid_flags(rs)) {
513 rs->ti->error = "Invalid flags combination";
520 /* MD raid10 bit definitions and helpers */
521 #define RAID10_OFFSET (1 << 16) /* stripes with data copies area adjacent on devices */
522 #define RAID10_BROCKEN_USE_FAR_SETS (1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
523 #define RAID10_USE_FAR_SETS (1 << 18) /* Use sets instead of whole stripe rotation */
524 #define RAID10_FAR_COPIES_SHIFT 8 /* raid10 # far copies shift (2nd byte of layout) */
526 /* Return md raid10 near copies for @layout */
527 static unsigned int __raid10_near_copies(int layout)
529 return layout & 0xFF;
532 /* Return md raid10 far copies for @layout */
533 static unsigned int __raid10_far_copies(int layout)
535 return __raid10_near_copies(layout >> RAID10_FAR_COPIES_SHIFT);
538 /* Return true if md raid10 offset for @layout */
539 static bool __is_raid10_offset(int layout)
541 return !!(layout & RAID10_OFFSET);
544 /* Return true if md raid10 near for @layout */
545 static bool __is_raid10_near(int layout)
547 return !__is_raid10_offset(layout) && __raid10_near_copies(layout) > 1;
550 /* Return true if md raid10 far for @layout */
551 static bool __is_raid10_far(int layout)
553 return !__is_raid10_offset(layout) && __raid10_far_copies(layout) > 1;
556 /* Return md raid10 layout string for @layout */
557 static const char *raid10_md_layout_to_format(int layout)
560 * Bit 16 stands for "offset"
561 * (i.e. adjacent stripes hold copies)
563 * Refer to MD's raid10.c for details
565 if (__is_raid10_offset(layout))
568 if (__raid10_near_copies(layout) > 1)
571 if (__raid10_far_copies(layout) > 1)
577 /* Return md raid10 algorithm for @name */
578 static int raid10_name_to_format(const char *name)
580 if (!strcasecmp(name, "near"))
581 return ALGORITHM_RAID10_NEAR;
582 else if (!strcasecmp(name, "offset"))
583 return ALGORITHM_RAID10_OFFSET;
584 else if (!strcasecmp(name, "far"))
585 return ALGORITHM_RAID10_FAR;
590 /* Return md raid10 copies for @layout */
591 static unsigned int raid10_md_layout_to_copies(int layout)
593 return max(__raid10_near_copies(layout), __raid10_far_copies(layout));
596 /* Return md raid10 format id for @format string */
597 static int raid10_format_to_md_layout(struct raid_set *rs,
598 unsigned int algorithm,
601 unsigned int n = 1, f = 1, r = 0;
604 * MD resilienece flaw:
606 * enabling use_far_sets for far/offset formats causes copies
607 * to be colocated on the same devs together with their origins!
609 * -> disable it for now in the definition above
611 if (algorithm == ALGORITHM_RAID10_DEFAULT ||
612 algorithm == ALGORITHM_RAID10_NEAR)
615 else if (algorithm == ALGORITHM_RAID10_OFFSET) {
618 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
619 r |= RAID10_USE_FAR_SETS;
621 } else if (algorithm == ALGORITHM_RAID10_FAR) {
624 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
625 r |= RAID10_USE_FAR_SETS;
630 return r | (f << RAID10_FAR_COPIES_SHIFT) | n;
632 /* END: MD raid10 bit definitions and helpers */
634 /* Check for any of the raid10 algorithms */
635 static bool __got_raid10(struct raid_type *rtp, const int layout)
637 if (rtp->level == 10) {
638 switch (rtp->algorithm) {
639 case ALGORITHM_RAID10_DEFAULT:
640 case ALGORITHM_RAID10_NEAR:
641 return __is_raid10_near(layout);
642 case ALGORITHM_RAID10_OFFSET:
643 return __is_raid10_offset(layout);
644 case ALGORITHM_RAID10_FAR:
645 return __is_raid10_far(layout);
654 /* Return raid_type for @name */
655 static struct raid_type *get_raid_type(const char *name)
657 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
659 while (rtp-- > raid_types)
660 if (!strcasecmp(rtp->name, name))
666 /* Return raid_type for @name based derived from @level and @layout */
667 static struct raid_type *get_raid_type_by_ll(const int level, const int layout)
669 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
671 while (rtp-- > raid_types) {
672 /* RAID10 special checks based on @layout flags/properties */
673 if (rtp->level == level &&
674 (__got_raid10(rtp, layout) || rtp->algorithm == layout))
681 /* Adjust rdev sectors */
682 static void rs_set_rdev_sectors(struct raid_set *rs)
684 struct mddev *mddev = &rs->md;
685 struct md_rdev *rdev;
688 * raid10 sets rdev->sector to the device size, which
689 * is unintended in case of out-of-place reshaping
691 rdev_for_each(rdev, mddev)
692 if (!test_bit(Journal, &rdev->flags))
693 rdev->sectors = mddev->dev_sectors;
697 * Change bdev capacity of @rs in case of a disk add/remove reshape
699 static void rs_set_capacity(struct raid_set *rs)
701 struct gendisk *gendisk = dm_disk(dm_table_get_md(rs->ti->table));
703 set_capacity(gendisk, rs->md.array_sectors);
704 revalidate_disk(gendisk);
708 * Set the mddev properties in @rs to the current
709 * ones retrieved from the freshest superblock
711 static void rs_set_cur(struct raid_set *rs)
713 struct mddev *mddev = &rs->md;
715 mddev->new_level = mddev->level;
716 mddev->new_layout = mddev->layout;
717 mddev->new_chunk_sectors = mddev->chunk_sectors;
721 * Set the mddev properties in @rs to the new
722 * ones requested by the ctr
724 static void rs_set_new(struct raid_set *rs)
726 struct mddev *mddev = &rs->md;
728 mddev->level = mddev->new_level;
729 mddev->layout = mddev->new_layout;
730 mddev->chunk_sectors = mddev->new_chunk_sectors;
731 mddev->raid_disks = rs->raid_disks;
732 mddev->delta_disks = 0;
735 static struct raid_set *raid_set_alloc(struct dm_target *ti, struct raid_type *raid_type,
736 unsigned int raid_devs)
741 if (raid_devs <= raid_type->parity_devs) {
742 ti->error = "Insufficient number of devices";
743 return ERR_PTR(-EINVAL);
746 rs = kzalloc(struct_size(rs, dev, raid_devs), GFP_KERNEL);
748 ti->error = "Cannot allocate raid context";
749 return ERR_PTR(-ENOMEM);
754 rs->raid_disks = raid_devs;
758 rs->raid_type = raid_type;
759 rs->stripe_cache_entries = 256;
760 rs->md.raid_disks = raid_devs;
761 rs->md.level = raid_type->level;
762 rs->md.new_level = rs->md.level;
763 rs->md.layout = raid_type->algorithm;
764 rs->md.new_layout = rs->md.layout;
765 rs->md.delta_disks = 0;
766 rs->md.recovery_cp = MaxSector;
768 for (i = 0; i < raid_devs; i++)
769 md_rdev_init(&rs->dev[i].rdev);
772 * Remaining items to be initialized by further RAID params:
775 * rs->md.chunk_sectors
776 * rs->md.new_chunk_sectors
783 /* Free all @rs allocations */
784 static void raid_set_free(struct raid_set *rs)
788 if (rs->journal_dev.dev) {
789 md_rdev_clear(&rs->journal_dev.rdev);
790 dm_put_device(rs->ti, rs->journal_dev.dev);
793 for (i = 0; i < rs->raid_disks; i++) {
794 if (rs->dev[i].meta_dev)
795 dm_put_device(rs->ti, rs->dev[i].meta_dev);
796 md_rdev_clear(&rs->dev[i].rdev);
797 if (rs->dev[i].data_dev)
798 dm_put_device(rs->ti, rs->dev[i].data_dev);
805 * For every device we have two words
806 * <meta_dev>: meta device name or '-' if missing
807 * <data_dev>: data device name or '-' if missing
809 * The following are permitted:
812 * <meta_dev> <data_dev>
814 * The following is not allowed:
817 * This code parses those words. If there is a failure,
818 * the caller must use raid_set_free() to unwind the operations.
820 static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
824 int metadata_available = 0;
828 /* Put off the number of raid devices argument to get to dev pairs */
829 arg = dm_shift_arg(as);
833 for (i = 0; i < rs->raid_disks; i++) {
834 rs->dev[i].rdev.raid_disk = i;
836 rs->dev[i].meta_dev = NULL;
837 rs->dev[i].data_dev = NULL;
840 * There are no offsets initially.
841 * Out of place reshape will set them accordingly.
843 rs->dev[i].rdev.data_offset = 0;
844 rs->dev[i].rdev.new_data_offset = 0;
845 rs->dev[i].rdev.mddev = &rs->md;
847 arg = dm_shift_arg(as);
851 if (strcmp(arg, "-")) {
852 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
853 &rs->dev[i].meta_dev);
855 rs->ti->error = "RAID metadata device lookup failure";
859 rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
860 if (!rs->dev[i].rdev.sb_page) {
861 rs->ti->error = "Failed to allocate superblock page";
866 arg = dm_shift_arg(as);
870 if (!strcmp(arg, "-")) {
871 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
872 (!rs->dev[i].rdev.recovery_offset)) {
873 rs->ti->error = "Drive designated for rebuild not specified";
877 if (rs->dev[i].meta_dev) {
878 rs->ti->error = "No data device supplied with metadata device";
885 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
886 &rs->dev[i].data_dev);
888 rs->ti->error = "RAID device lookup failure";
892 if (rs->dev[i].meta_dev) {
893 metadata_available = 1;
894 rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
896 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
897 list_add_tail(&rs->dev[i].rdev.same_set, &rs->md.disks);
898 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
902 if (rs->journal_dev.dev)
903 list_add_tail(&rs->journal_dev.rdev.same_set, &rs->md.disks);
905 if (metadata_available) {
907 rs->md.persistent = 1;
908 rs->md.major_version = 2;
909 } else if (rebuild && !rs->md.recovery_cp) {
911 * Without metadata, we will not be able to tell if the array
912 * is in-sync or not - we must assume it is not. Therefore,
913 * it is impossible to rebuild a drive.
915 * Even if there is metadata, the on-disk information may
916 * indicate that the array is not in-sync and it will then
919 * User could specify 'nosync' option if desperate.
921 rs->ti->error = "Unable to rebuild drive while array is not in-sync";
929 * validate_region_size
931 * @region_size: region size in sectors. If 0, pick a size (4MiB default).
933 * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
934 * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
936 * Returns: 0 on success, -EINVAL on failure.
938 static int validate_region_size(struct raid_set *rs, unsigned long region_size)
940 unsigned long min_region_size = rs->ti->len / (1 << 21);
947 * Choose a reasonable default. All figures in sectors.
949 if (min_region_size > (1 << 13)) {
950 /* If not a power of 2, make it the next power of 2 */
951 region_size = roundup_pow_of_two(min_region_size);
952 DMINFO("Choosing default region size of %lu sectors",
955 DMINFO("Choosing default region size of 4MiB");
956 region_size = 1 << 13; /* sectors */
960 * Validate user-supplied value.
962 if (region_size > rs->ti->len) {
963 rs->ti->error = "Supplied region size is too large";
967 if (region_size < min_region_size) {
968 DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
969 region_size, min_region_size);
970 rs->ti->error = "Supplied region size is too small";
974 if (!is_power_of_2(region_size)) {
975 rs->ti->error = "Region size is not a power of 2";
979 if (region_size < rs->md.chunk_sectors) {
980 rs->ti->error = "Region size is smaller than the chunk size";
986 * Convert sectors to bytes.
988 rs->md.bitmap_info.chunksize = to_bytes(region_size);
994 * validate_raid_redundancy
997 * Determine if there are enough devices in the array that haven't
998 * failed (or are being rebuilt) to form a usable array.
1000 * Returns: 0 on success, -EINVAL on failure.
1002 static int validate_raid_redundancy(struct raid_set *rs)
1004 unsigned int i, rebuild_cnt = 0;
1005 unsigned int rebuilds_per_group = 0, copies;
1006 unsigned int group_size, last_group_start;
1008 for (i = 0; i < rs->md.raid_disks; i++)
1009 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
1010 !rs->dev[i].rdev.sb_page)
1013 switch (rs->md.level) {
1017 if (rebuild_cnt >= rs->md.raid_disks)
1023 if (rebuild_cnt > rs->raid_type->parity_devs)
1027 copies = raid10_md_layout_to_copies(rs->md.new_layout);
1029 DMERR("Bogus raid10 data copies < 2!");
1033 if (rebuild_cnt < copies)
1037 * It is possible to have a higher rebuild count for RAID10,
1038 * as long as the failed devices occur in different mirror
1039 * groups (i.e. different stripes).
1041 * When checking "near" format, make sure no adjacent devices
1042 * have failed beyond what can be handled. In addition to the
1043 * simple case where the number of devices is a multiple of the
1044 * number of copies, we must also handle cases where the number
1045 * of devices is not a multiple of the number of copies.
1046 * E.g. dev1 dev2 dev3 dev4 dev5
1050 if (__is_raid10_near(rs->md.new_layout)) {
1051 for (i = 0; i < rs->md.raid_disks; i++) {
1053 rebuilds_per_group = 0;
1054 if ((!rs->dev[i].rdev.sb_page ||
1055 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1056 (++rebuilds_per_group >= copies))
1063 * When checking "far" and "offset" formats, we need to ensure
1064 * that the device that holds its copy is not also dead or
1065 * being rebuilt. (Note that "far" and "offset" formats only
1066 * support two copies right now. These formats also only ever
1067 * use the 'use_far_sets' variant.)
1069 * This check is somewhat complicated by the need to account
1070 * for arrays that are not a multiple of (far) copies. This
1071 * results in the need to treat the last (potentially larger)
1074 group_size = (rs->md.raid_disks / copies);
1075 last_group_start = (rs->md.raid_disks / group_size) - 1;
1076 last_group_start *= group_size;
1077 for (i = 0; i < rs->md.raid_disks; i++) {
1078 if (!(i % copies) && !(i > last_group_start))
1079 rebuilds_per_group = 0;
1080 if ((!rs->dev[i].rdev.sb_page ||
1081 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1082 (++rebuilds_per_group >= copies))
1098 * Possible arguments are...
1099 * <chunk_size> [optional_args]
1101 * Argument definitions
1102 * <chunk_size> The number of sectors per disk that
1103 * will form the "stripe"
1104 * [[no]sync] Force or prevent recovery of the
1106 * [rebuild <idx>] Rebuild the drive indicated by the index
1107 * [daemon_sleep <ms>] Time between bitmap daemon work to
1109 * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1110 * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1111 * [write_mostly <idx>] Indicate a write mostly drive via index
1112 * [max_write_behind <sectors>] See '-write-behind=' (man mdadm)
1113 * [stripe_cache <sectors>] Stripe cache size for higher RAIDs
1114 * [region_size <sectors>] Defines granularity of bitmap
1115 * [journal_dev <dev>] raid4/5/6 journaling deviice
1116 * (i.e. write hole closing log)
1118 * RAID10-only options:
1119 * [raid10_copies <# copies>] Number of copies. (Default: 2)
1120 * [raid10_format <near|far|offset>] Layout algorithm. (Default: near)
1122 static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
1123 unsigned int num_raid_params)
1125 int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
1126 unsigned int raid10_copies = 2;
1127 unsigned int i, write_mostly = 0;
1128 unsigned int region_size = 0;
1129 sector_t max_io_len;
1130 const char *arg, *key;
1131 struct raid_dev *rd;
1132 struct raid_type *rt = rs->raid_type;
1134 arg = dm_shift_arg(as);
1135 num_raid_params--; /* Account for chunk_size argument */
1137 if (kstrtoint(arg, 10, &value) < 0) {
1138 rs->ti->error = "Bad numerical argument given for chunk_size";
1143 * First, parse the in-order required arguments
1144 * "chunk_size" is the only argument of this type.
1146 if (rt_is_raid1(rt)) {
1148 DMERR("Ignoring chunk size parameter for RAID 1");
1150 } else if (!is_power_of_2(value)) {
1151 rs->ti->error = "Chunk size must be a power of 2";
1153 } else if (value < 8) {
1154 rs->ti->error = "Chunk size value is too small";
1158 rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
1161 * We set each individual device as In_sync with a completed
1162 * 'recovery_offset'. If there has been a device failure or
1163 * replacement then one of the following cases applies:
1165 * 1) User specifies 'rebuild'.
1166 * - Device is reset when param is read.
1167 * 2) A new device is supplied.
1168 * - No matching superblock found, resets device.
1169 * 3) Device failure was transient and returns on reload.
1170 * - Failure noticed, resets device for bitmap replay.
1171 * 4) Device hadn't completed recovery after previous failure.
1172 * - Superblock is read and overrides recovery_offset.
1174 * What is found in the superblocks of the devices is always
1175 * authoritative, unless 'rebuild' or '[no]sync' was specified.
1177 for (i = 0; i < rs->raid_disks; i++) {
1178 set_bit(In_sync, &rs->dev[i].rdev.flags);
1179 rs->dev[i].rdev.recovery_offset = MaxSector;
1183 * Second, parse the unordered optional arguments
1185 for (i = 0; i < num_raid_params; i++) {
1186 key = dm_shift_arg(as);
1188 rs->ti->error = "Not enough raid parameters given";
1192 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
1193 if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1194 rs->ti->error = "Only one 'nosync' argument allowed";
1199 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
1200 if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
1201 rs->ti->error = "Only one 'sync' argument allowed";
1206 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
1207 if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1208 rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1214 arg = dm_shift_arg(as);
1215 i++; /* Account for the argument pairs */
1217 rs->ti->error = "Wrong number of raid parameters given";
1222 * Parameters that take a string value are checked here.
1224 /* "raid10_format {near|offset|far} */
1225 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
1226 if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
1227 rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1230 if (!rt_is_raid10(rt)) {
1231 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1234 raid10_format = raid10_name_to_format(arg);
1235 if (raid10_format < 0) {
1236 rs->ti->error = "Invalid 'raid10_format' value given";
1237 return raid10_format;
1242 /* "journal_dev <dev>" */
1243 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV))) {
1245 struct md_rdev *jdev;
1247 if (test_and_set_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1248 rs->ti->error = "Only one raid4/5/6 set journaling device allowed";
1251 if (!rt_is_raid456(rt)) {
1252 rs->ti->error = "'journal_dev' is an invalid parameter for this RAID type";
1255 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
1256 &rs->journal_dev.dev);
1258 rs->ti->error = "raid4/5/6 journal device lookup failure";
1261 jdev = &rs->journal_dev.rdev;
1263 jdev->mddev = &rs->md;
1264 jdev->bdev = rs->journal_dev.dev->bdev;
1265 jdev->sectors = to_sector(i_size_read(jdev->bdev->bd_inode));
1266 if (jdev->sectors < MIN_RAID456_JOURNAL_SPACE) {
1267 rs->ti->error = "No space for raid4/5/6 journal";
1270 rs->journal_dev.mode = R5C_JOURNAL_MODE_WRITE_THROUGH;
1271 set_bit(Journal, &jdev->flags);
1275 /* "journal_mode <mode>" ("journal_dev" mandatory!) */
1276 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE))) {
1279 if (!test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1280 rs->ti->error = "raid4/5/6 'journal_mode' is invalid without 'journal_dev'";
1283 if (test_and_set_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
1284 rs->ti->error = "Only one raid4/5/6 'journal_mode' argument allowed";
1287 r = dm_raid_journal_mode_to_md(arg);
1289 rs->ti->error = "Invalid 'journal_mode' argument";
1292 rs->journal_dev.mode = r;
1297 * Parameters with number values from here on.
1299 if (kstrtoint(arg, 10, &value) < 0) {
1300 rs->ti->error = "Bad numerical argument given in raid params";
1304 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
1306 * "rebuild" is being passed in by userspace to provide
1307 * indexes of replaced devices and to set up additional
1308 * devices on raid level takeover.
1310 if (!__within_range(value, 0, rs->raid_disks - 1)) {
1311 rs->ti->error = "Invalid rebuild index given";
1315 if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1316 rs->ti->error = "rebuild for this index already given";
1320 rd = rs->dev + value;
1321 clear_bit(In_sync, &rd->rdev.flags);
1322 clear_bit(Faulty, &rd->rdev.flags);
1323 rd->rdev.recovery_offset = 0;
1324 set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
1325 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
1326 if (!rt_is_raid1(rt)) {
1327 rs->ti->error = "write_mostly option is only valid for RAID1";
1331 if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
1332 rs->ti->error = "Invalid write_mostly index given";
1337 set_bit(WriteMostly, &rs->dev[value].rdev.flags);
1338 set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
1339 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
1340 if (!rt_is_raid1(rt)) {
1341 rs->ti->error = "max_write_behind option is only valid for RAID1";
1345 if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
1346 rs->ti->error = "Only one max_write_behind argument pair allowed";
1351 * In device-mapper, we specify things in sectors, but
1352 * MD records this value in kB
1354 if (value < 0 || value / 2 > COUNTER_MAX) {
1355 rs->ti->error = "Max write-behind limit out of range";
1359 rs->md.bitmap_info.max_write_behind = value / 2;
1360 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
1361 if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
1362 rs->ti->error = "Only one daemon_sleep argument pair allowed";
1366 rs->ti->error = "daemon sleep period out of range";
1369 rs->md.bitmap_info.daemon_sleep = value;
1370 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
1371 /* Userspace passes new data_offset after having extended the the data image LV */
1372 if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
1373 rs->ti->error = "Only one data_offset argument pair allowed";
1376 /* Ensure sensible data offset */
1378 (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
1379 rs->ti->error = "Bogus data_offset value";
1382 rs->data_offset = value;
1383 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
1384 /* Define the +/-# of disks to add to/remove from the given raid set */
1385 if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1386 rs->ti->error = "Only one delta_disks argument pair allowed";
1389 /* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1390 if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
1391 rs->ti->error = "Too many delta_disk requested";
1395 rs->delta_disks = value;
1396 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
1397 if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
1398 rs->ti->error = "Only one stripe_cache argument pair allowed";
1402 if (!rt_is_raid456(rt)) {
1403 rs->ti->error = "Inappropriate argument: stripe_cache";
1408 rs->ti->error = "Bogus stripe cache entries value";
1411 rs->stripe_cache_entries = value;
1412 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
1413 if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1414 rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1419 rs->ti->error = "min_recovery_rate out of range";
1422 rs->md.sync_speed_min = value;
1423 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
1424 if (test_and_set_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags)) {
1425 rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1430 rs->ti->error = "max_recovery_rate out of range";
1433 rs->md.sync_speed_max = value;
1434 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
1435 if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
1436 rs->ti->error = "Only one region_size argument pair allowed";
1440 region_size = value;
1441 rs->requested_bitmap_chunk_sectors = value;
1442 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
1443 if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
1444 rs->ti->error = "Only one raid10_copies argument pair allowed";
1448 if (!__within_range(value, 2, rs->md.raid_disks)) {
1449 rs->ti->error = "Bad value for 'raid10_copies'";
1453 raid10_copies = value;
1455 DMERR("Unable to parse RAID parameter: %s", key);
1456 rs->ti->error = "Unable to parse RAID parameter";
1461 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1462 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1463 rs->ti->error = "sync and nosync are mutually exclusive";
1467 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) &&
1468 (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ||
1469 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))) {
1470 rs->ti->error = "sync/nosync and rebuild are mutually exclusive";
1474 if (write_mostly >= rs->md.raid_disks) {
1475 rs->ti->error = "Can't set all raid1 devices to write_mostly";
1479 if (rs->md.sync_speed_max &&
1480 rs->md.sync_speed_min > rs->md.sync_speed_max) {
1481 rs->ti->error = "Bogus recovery rates";
1485 if (validate_region_size(rs, region_size))
1488 if (rs->md.chunk_sectors)
1489 max_io_len = rs->md.chunk_sectors;
1491 max_io_len = region_size;
1493 if (dm_set_target_max_io_len(rs->ti, max_io_len))
1496 if (rt_is_raid10(rt)) {
1497 if (raid10_copies > rs->md.raid_disks) {
1498 rs->ti->error = "Not enough devices to satisfy specification";
1502 rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
1503 if (rs->md.new_layout < 0) {
1504 rs->ti->error = "Error getting raid10 format";
1505 return rs->md.new_layout;
1508 rt = get_raid_type_by_ll(10, rs->md.new_layout);
1510 rs->ti->error = "Failed to recognize new raid10 layout";
1514 if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1515 rt->algorithm == ALGORITHM_RAID10_NEAR) &&
1516 test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1517 rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1522 rs->raid10_copies = raid10_copies;
1524 /* Assume there are no metadata devices until the drives are parsed */
1525 rs->md.persistent = 0;
1526 rs->md.external = 1;
1528 /* Check, if any invalid ctr arguments have been passed in for the raid level */
1529 return rs_check_for_valid_flags(rs);
1532 /* Set raid4/5/6 cache size */
1533 static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1536 struct r5conf *conf;
1537 struct mddev *mddev = &rs->md;
1538 uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1539 uint32_t nr_stripes = rs->stripe_cache_entries;
1541 if (!rt_is_raid456(rs->raid_type)) {
1542 rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1546 if (nr_stripes < min_stripes) {
1547 DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1548 nr_stripes, min_stripes);
1549 nr_stripes = min_stripes;
1552 conf = mddev->private;
1554 rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1558 /* Try setting number of stripes in raid456 stripe cache */
1559 if (conf->min_nr_stripes != nr_stripes) {
1560 r = raid5_set_cache_size(mddev, nr_stripes);
1562 rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1566 DMINFO("%u stripe cache entries", nr_stripes);
1572 /* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1573 static unsigned int mddev_data_stripes(struct raid_set *rs)
1575 return rs->md.raid_disks - rs->raid_type->parity_devs;
1578 /* Return # of data stripes of @rs (i.e. as of ctr) */
1579 static unsigned int rs_data_stripes(struct raid_set *rs)
1581 return rs->raid_disks - rs->raid_type->parity_devs;
1585 * Retrieve rdev->sectors from any valid raid device of @rs
1586 * to allow userpace to pass in arbitray "- -" device tupples.
1588 static sector_t __rdev_sectors(struct raid_set *rs)
1592 for (i = 0; i < rs->md.raid_disks; i++) {
1593 struct md_rdev *rdev = &rs->dev[i].rdev;
1595 if (!test_bit(Journal, &rdev->flags) &&
1596 rdev->bdev && rdev->sectors)
1597 return rdev->sectors;
1603 /* Check that calculated dev_sectors fits all component devices. */
1604 static int _check_data_dev_sectors(struct raid_set *rs)
1607 struct md_rdev *rdev;
1609 rdev_for_each(rdev, &rs->md)
1610 if (!test_bit(Journal, &rdev->flags) && rdev->bdev) {
1611 ds = min(ds, to_sector(i_size_read(rdev->bdev->bd_inode)));
1612 if (ds < rs->md.dev_sectors) {
1613 rs->ti->error = "Component device(s) too small";
1621 /* Calculate the sectors per device and per array used for @rs */
1622 static int rs_set_dev_and_array_sectors(struct raid_set *rs, sector_t sectors, bool use_mddev)
1625 unsigned int data_stripes;
1626 sector_t array_sectors = sectors, dev_sectors = sectors;
1627 struct mddev *mddev = &rs->md;
1630 delta_disks = mddev->delta_disks;
1631 data_stripes = mddev_data_stripes(rs);
1633 delta_disks = rs->delta_disks;
1634 data_stripes = rs_data_stripes(rs);
1637 /* Special raid1 case w/o delta_disks support (yet) */
1638 if (rt_is_raid1(rs->raid_type))
1640 else if (rt_is_raid10(rs->raid_type)) {
1641 if (rs->raid10_copies < 2 ||
1643 rs->ti->error = "Bogus raid10 data copies or delta disks";
1647 dev_sectors *= rs->raid10_copies;
1648 if (sector_div(dev_sectors, data_stripes))
1651 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1652 if (sector_div(array_sectors, rs->raid10_copies))
1655 } else if (sector_div(dev_sectors, data_stripes))
1659 /* Striped layouts */
1660 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1662 mddev->array_sectors = array_sectors;
1663 mddev->dev_sectors = dev_sectors;
1664 rs_set_rdev_sectors(rs);
1666 return _check_data_dev_sectors(rs);
1668 rs->ti->error = "Target length not divisible by number of data devices";
1672 /* Setup recovery on @rs */
1673 static void __rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1675 /* raid0 does not recover */
1676 if (rs_is_raid0(rs))
1677 rs->md.recovery_cp = MaxSector;
1679 * A raid6 set has to be recovered either
1680 * completely or for the grown part to
1681 * ensure proper parity and Q-Syndrome
1683 else if (rs_is_raid6(rs))
1684 rs->md.recovery_cp = dev_sectors;
1686 * Other raid set types may skip recovery
1687 * depending on the 'nosync' flag.
1690 rs->md.recovery_cp = test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)
1691 ? MaxSector : dev_sectors;
1694 /* Setup recovery on @rs based on raid type, device size and 'nosync' flag */
1695 static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1698 /* New raid set or 'sync' flag provided */
1699 __rs_setup_recovery(rs, 0);
1700 else if (dev_sectors == MaxSector)
1701 /* Prevent recovery */
1702 __rs_setup_recovery(rs, MaxSector);
1703 else if (__rdev_sectors(rs) < dev_sectors)
1704 /* Grown raid set */
1705 __rs_setup_recovery(rs, __rdev_sectors(rs));
1707 __rs_setup_recovery(rs, MaxSector);
1710 static void do_table_event(struct work_struct *ws)
1712 struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1714 smp_rmb(); /* Make sure we access most actual mddev properties */
1715 if (!rs_is_reshaping(rs)) {
1716 if (rs_is_raid10(rs))
1717 rs_set_rdev_sectors(rs);
1718 rs_set_capacity(rs);
1720 dm_table_event(rs->ti->table);
1723 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
1725 struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
1727 return mddev_congested(&rs->md, bits);
1731 * Make sure a valid takover (level switch) is being requested on @rs
1733 * Conversions of raid sets from one MD personality to another
1734 * have to conform to restrictions which are enforced here.
1736 static int rs_check_takeover(struct raid_set *rs)
1738 struct mddev *mddev = &rs->md;
1739 unsigned int near_copies;
1741 if (rs->md.degraded) {
1742 rs->ti->error = "Can't takeover degraded raid set";
1746 if (rs_is_reshaping(rs)) {
1747 rs->ti->error = "Can't takeover reshaping raid set";
1751 switch (mddev->level) {
1753 /* raid0 -> raid1/5 with one disk */
1754 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1755 mddev->raid_disks == 1)
1758 /* raid0 -> raid10 */
1759 if (mddev->new_level == 10 &&
1760 !(rs->raid_disks % mddev->raid_disks))
1763 /* raid0 with multiple disks -> raid4/5/6 */
1764 if (__within_range(mddev->new_level, 4, 6) &&
1765 mddev->new_layout == ALGORITHM_PARITY_N &&
1766 mddev->raid_disks > 1)
1772 /* Can't takeover raid10_offset! */
1773 if (__is_raid10_offset(mddev->layout))
1776 near_copies = __raid10_near_copies(mddev->layout);
1778 /* raid10* -> raid0 */
1779 if (mddev->new_level == 0) {
1780 /* Can takeover raid10_near with raid disks divisable by data copies! */
1781 if (near_copies > 1 &&
1782 !(mddev->raid_disks % near_copies)) {
1783 mddev->raid_disks /= near_copies;
1784 mddev->delta_disks = mddev->raid_disks;
1788 /* Can takeover raid10_far */
1789 if (near_copies == 1 &&
1790 __raid10_far_copies(mddev->layout) > 1)
1796 /* raid10_{near,far} -> raid1 */
1797 if (mddev->new_level == 1 &&
1798 max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
1801 /* raid10_{near,far} with 2 disks -> raid4/5 */
1802 if (__within_range(mddev->new_level, 4, 5) &&
1803 mddev->raid_disks == 2)
1808 /* raid1 with 2 disks -> raid4/5 */
1809 if (__within_range(mddev->new_level, 4, 5) &&
1810 mddev->raid_disks == 2) {
1811 mddev->degraded = 1;
1815 /* raid1 -> raid0 */
1816 if (mddev->new_level == 0 &&
1817 mddev->raid_disks == 1)
1820 /* raid1 -> raid10 */
1821 if (mddev->new_level == 10)
1826 /* raid4 -> raid0 */
1827 if (mddev->new_level == 0)
1830 /* raid4 -> raid1/5 with 2 disks */
1831 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1832 mddev->raid_disks == 2)
1835 /* raid4 -> raid5/6 with parity N */
1836 if (__within_range(mddev->new_level, 5, 6) &&
1837 mddev->layout == ALGORITHM_PARITY_N)
1842 /* raid5 with parity N -> raid0 */
1843 if (mddev->new_level == 0 &&
1844 mddev->layout == ALGORITHM_PARITY_N)
1847 /* raid5 with parity N -> raid4 */
1848 if (mddev->new_level == 4 &&
1849 mddev->layout == ALGORITHM_PARITY_N)
1852 /* raid5 with 2 disks -> raid1/4/10 */
1853 if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1854 mddev->raid_disks == 2)
1857 /* raid5_* -> raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
1858 if (mddev->new_level == 6 &&
1859 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1860 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
1865 /* raid6 with parity N -> raid0 */
1866 if (mddev->new_level == 0 &&
1867 mddev->layout == ALGORITHM_PARITY_N)
1870 /* raid6 with parity N -> raid4 */
1871 if (mddev->new_level == 4 &&
1872 mddev->layout == ALGORITHM_PARITY_N)
1875 /* raid6_*_n with Q-Syndrome N -> raid5_* */
1876 if (mddev->new_level == 5 &&
1877 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1878 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
1885 rs->ti->error = "takeover not possible";
1889 /* True if @rs requested to be taken over */
1890 static bool rs_takeover_requested(struct raid_set *rs)
1892 return rs->md.new_level != rs->md.level;
1895 /* True if @rs is requested to reshape by ctr */
1896 static bool rs_reshape_requested(struct raid_set *rs)
1899 struct mddev *mddev = &rs->md;
1901 if (rs_takeover_requested(rs))
1904 if (rs_is_raid0(rs))
1907 change = mddev->new_layout != mddev->layout ||
1908 mddev->new_chunk_sectors != mddev->chunk_sectors ||
1911 /* Historical case to support raid1 reshape without delta disks */
1912 if (rs_is_raid1(rs)) {
1913 if (rs->delta_disks)
1914 return !!rs->delta_disks;
1917 mddev->raid_disks != rs->raid_disks;
1920 if (rs_is_raid10(rs))
1922 !__is_raid10_far(mddev->new_layout) &&
1923 rs->delta_disks >= 0;
1929 #define FEATURE_FLAG_SUPPORTS_V190 0x1 /* Supports extended superblock */
1931 /* State flags for sb->flags */
1932 #define SB_FLAG_RESHAPE_ACTIVE 0x1
1933 #define SB_FLAG_RESHAPE_BACKWARDS 0x2
1936 * This structure is never routinely used by userspace, unlike md superblocks.
1937 * Devices with this superblock should only ever be accessed via device-mapper.
1939 #define DM_RAID_MAGIC 0x64526D44
1940 struct dm_raid_superblock {
1941 __le32 magic; /* "DmRd" */
1942 __le32 compat_features; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1944 __le32 num_devices; /* Number of devices in this raid set. (Max 64) */
1945 __le32 array_position; /* The position of this drive in the raid set */
1947 __le64 events; /* Incremented by md when superblock updated */
1948 __le64 failed_devices; /* Pre 1.9.0 part of bit field of devices to */
1949 /* indicate failures (see extension below) */
1952 * This offset tracks the progress of the repair or replacement of
1953 * an individual drive.
1955 __le64 disk_recovery_offset;
1958 * This offset tracks the progress of the initial raid set
1959 * synchronisation/parity calculation.
1961 __le64 array_resync_offset;
1964 * raid characteristics
1968 __le32 stripe_sectors;
1970 /********************************************************************
1971 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1973 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
1976 __le32 flags; /* Flags defining array states for reshaping */
1979 * This offset tracks the progress of a raid
1980 * set reshape in order to be able to restart it
1982 __le64 reshape_position;
1985 * These define the properties of the array in case of an interrupted reshape
1989 __le32 new_stripe_sectors;
1992 __le64 array_sectors; /* Array size in sectors */
1995 * Sector offsets to data on devices (reshaping).
1996 * Needed to support out of place reshaping, thus
1997 * not writing over any stripes whilst converting
1998 * them from old to new layout
2001 __le64 new_data_offset;
2003 __le64 sectors; /* Used device size in sectors */
2006 * Additonal Bit field of devices indicating failures to support
2007 * up to 256 devices with the 1.9.0 on-disk metadata format
2009 __le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
2011 __le32 incompat_features; /* Used to indicate any incompatible features */
2013 /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
2017 * Check for reshape constraints on raid set @rs:
2019 * - reshape function non-existent
2021 * - ongoing recovery
2024 * Returns 0 if none or -EPERM if given constraint
2025 * and error message reference in @errmsg
2027 static int rs_check_reshape(struct raid_set *rs)
2029 struct mddev *mddev = &rs->md;
2031 if (!mddev->pers || !mddev->pers->check_reshape)
2032 rs->ti->error = "Reshape not supported";
2033 else if (mddev->degraded)
2034 rs->ti->error = "Can't reshape degraded raid set";
2035 else if (rs_is_recovering(rs))
2036 rs->ti->error = "Convert request on recovering raid set prohibited";
2037 else if (rs_is_reshaping(rs))
2038 rs->ti->error = "raid set already reshaping!";
2039 else if (!(rs_is_raid1(rs) || rs_is_raid10(rs) || rs_is_raid456(rs)))
2040 rs->ti->error = "Reshaping only supported for raid1/4/5/6/10";
2047 static int read_disk_sb(struct md_rdev *rdev, int size, bool force_reload)
2049 BUG_ON(!rdev->sb_page);
2051 if (rdev->sb_loaded && !force_reload)
2054 rdev->sb_loaded = 0;
2056 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) {
2057 DMERR("Failed to read superblock of device at position %d",
2059 md_error(rdev->mddev, rdev);
2060 set_bit(Faulty, &rdev->flags);
2064 rdev->sb_loaded = 1;
2069 static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2071 failed_devices[0] = le64_to_cpu(sb->failed_devices);
2072 memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
2074 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2075 int i = ARRAY_SIZE(sb->extended_failed_devices);
2078 failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
2082 static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2084 int i = ARRAY_SIZE(sb->extended_failed_devices);
2086 sb->failed_devices = cpu_to_le64(failed_devices[0]);
2088 sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
2092 * Synchronize the superblock members with the raid set properties
2094 * All superblock data is little endian.
2096 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
2098 bool update_failed_devices = false;
2100 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2101 struct dm_raid_superblock *sb;
2102 struct raid_set *rs = container_of(mddev, struct raid_set, md);
2104 /* No metadata device, no superblock */
2105 if (!rdev->meta_bdev)
2108 BUG_ON(!rdev->sb_page);
2110 sb = page_address(rdev->sb_page);
2112 sb_retrieve_failed_devices(sb, failed_devices);
2114 for (i = 0; i < rs->raid_disks; i++)
2115 if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
2116 update_failed_devices = true;
2117 set_bit(i, (void *) failed_devices);
2120 if (update_failed_devices)
2121 sb_update_failed_devices(sb, failed_devices);
2123 sb->magic = cpu_to_le32(DM_RAID_MAGIC);
2124 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2126 sb->num_devices = cpu_to_le32(mddev->raid_disks);
2127 sb->array_position = cpu_to_le32(rdev->raid_disk);
2129 sb->events = cpu_to_le64(mddev->events);
2131 sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
2132 sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
2134 sb->level = cpu_to_le32(mddev->level);
2135 sb->layout = cpu_to_le32(mddev->layout);
2136 sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
2138 /********************************************************************
2139 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
2141 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
2143 sb->new_level = cpu_to_le32(mddev->new_level);
2144 sb->new_layout = cpu_to_le32(mddev->new_layout);
2145 sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
2147 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2149 smp_rmb(); /* Make sure we access most recent reshape position */
2150 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2151 if (le64_to_cpu(sb->reshape_position) != MaxSector) {
2152 /* Flag ongoing reshape */
2153 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
2155 if (mddev->delta_disks < 0 || mddev->reshape_backwards)
2156 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
2158 /* Clear reshape flags */
2159 sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
2162 sb->array_sectors = cpu_to_le64(mddev->array_sectors);
2163 sb->data_offset = cpu_to_le64(rdev->data_offset);
2164 sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
2165 sb->sectors = cpu_to_le64(rdev->sectors);
2166 sb->incompat_features = cpu_to_le32(0);
2168 /* Zero out the rest of the payload after the size of the superblock */
2169 memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
2175 * This function creates a superblock if one is not found on the device
2176 * and will decide which superblock to use if there's a choice.
2178 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
2180 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
2183 struct dm_raid_superblock *sb;
2184 struct dm_raid_superblock *refsb;
2185 uint64_t events_sb, events_refsb;
2187 r = read_disk_sb(rdev, rdev->sb_size, false);
2191 sb = page_address(rdev->sb_page);
2194 * Two cases that we want to write new superblocks and rebuild:
2195 * 1) New device (no matching magic number)
2196 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2198 if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
2199 (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
2200 super_sync(rdev->mddev, rdev);
2202 set_bit(FirstUse, &rdev->flags);
2203 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2205 /* Force writing of superblocks to disk */
2206 set_bit(MD_SB_CHANGE_DEVS, &rdev->mddev->sb_flags);
2208 /* Any superblock is better than none, choose that if given */
2209 return refdev ? 0 : 1;
2215 events_sb = le64_to_cpu(sb->events);
2217 refsb = page_address(refdev->sb_page);
2218 events_refsb = le64_to_cpu(refsb->events);
2220 return (events_sb > events_refsb) ? 1 : 0;
2223 static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
2227 struct mddev *mddev = &rs->md;
2229 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2230 struct dm_raid_superblock *sb;
2231 uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
2233 struct dm_raid_superblock *sb2;
2235 sb = page_address(rdev->sb_page);
2236 events_sb = le64_to_cpu(sb->events);
2239 * Initialise to 1 if this is a new superblock.
2241 mddev->events = events_sb ? : 1;
2243 mddev->reshape_position = MaxSector;
2245 mddev->raid_disks = le32_to_cpu(sb->num_devices);
2246 mddev->level = le32_to_cpu(sb->level);
2247 mddev->layout = le32_to_cpu(sb->layout);
2248 mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
2251 * Reshaping is supported, e.g. reshape_position is valid
2252 * in superblock and superblock content is authoritative.
2254 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2255 /* Superblock is authoritative wrt given raid set layout! */
2256 mddev->new_level = le32_to_cpu(sb->new_level);
2257 mddev->new_layout = le32_to_cpu(sb->new_layout);
2258 mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
2259 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
2260 mddev->array_sectors = le64_to_cpu(sb->array_sectors);
2262 /* raid was reshaping and got interrupted */
2263 if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
2264 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
2265 DMERR("Reshape requested but raid set is still reshaping");
2269 if (mddev->delta_disks < 0 ||
2270 (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
2271 mddev->reshape_backwards = 1;
2273 mddev->reshape_backwards = 0;
2275 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2276 rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2281 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2283 struct raid_type *rt_cur = get_raid_type_by_ll(mddev->level, mddev->layout);
2284 struct raid_type *rt_new = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
2286 if (rs_takeover_requested(rs)) {
2287 if (rt_cur && rt_new)
2288 DMERR("Takeover raid sets from %s to %s not yet supported by metadata. (raid level change)",
2289 rt_cur->name, rt_new->name);
2291 DMERR("Takeover raid sets not yet supported by metadata. (raid level change)");
2293 } else if (rs_reshape_requested(rs)) {
2294 DMERR("Reshaping raid sets not yet supported by metadata. (raid layout change keeping level)");
2295 if (mddev->layout != mddev->new_layout) {
2296 if (rt_cur && rt_new)
2297 DMERR(" current layout %s vs new layout %s",
2298 rt_cur->name, rt_new->name);
2300 DMERR(" current layout 0x%X vs new layout 0x%X",
2301 le32_to_cpu(sb->layout), mddev->new_layout);
2303 if (mddev->chunk_sectors != mddev->new_chunk_sectors)
2304 DMERR(" current stripe sectors %u vs new stripe sectors %u",
2305 mddev->chunk_sectors, mddev->new_chunk_sectors);
2306 if (rs->delta_disks)
2307 DMERR(" current %u disks vs new %u disks",
2308 mddev->raid_disks, mddev->raid_disks + rs->delta_disks);
2309 if (rs_is_raid10(rs)) {
2310 DMERR(" Old layout: %s w/ %u copies",
2311 raid10_md_layout_to_format(mddev->layout),
2312 raid10_md_layout_to_copies(mddev->layout));
2313 DMERR(" New layout: %s w/ %u copies",
2314 raid10_md_layout_to_format(mddev->new_layout),
2315 raid10_md_layout_to_copies(mddev->new_layout));
2320 DMINFO("Discovered old metadata format; upgrading to extended metadata format");
2323 if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
2324 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2327 * During load, we set FirstUse if a new superblock was written.
2328 * There are two reasons we might not have a superblock:
2329 * 1) The raid set is brand new - in which case, all of the
2330 * devices must have their In_sync bit set. Also,
2331 * recovery_cp must be 0, unless forced.
2332 * 2) This is a new device being added to an old raid set
2333 * and the new device needs to be rebuilt - in which
2334 * case the In_sync bit will /not/ be set and
2335 * recovery_cp must be MaxSector.
2336 * 3) This is/are a new device(s) being added to an old
2337 * raid set during takeover to a higher raid level
2338 * to provide capacity for redundancy or during reshape
2339 * to add capacity to grow the raid set.
2342 rdev_for_each(r, mddev) {
2343 if (test_bit(Journal, &rdev->flags))
2346 if (test_bit(FirstUse, &r->flags))
2349 if (!test_bit(In_sync, &r->flags)) {
2350 DMINFO("Device %d specified for rebuild; clearing superblock",
2354 if (test_bit(FirstUse, &r->flags))
2361 if (new_devs == rs->raid_disks || !rebuilds) {
2362 /* Replace a broken device */
2363 if (new_devs == 1 && !rs->delta_disks)
2365 if (new_devs == rs->raid_disks) {
2366 DMINFO("Superblocks created for new raid set");
2367 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2368 } else if (new_devs != rebuilds &&
2369 new_devs != rs->delta_disks) {
2370 DMERR("New device injected into existing raid set without "
2371 "'delta_disks' or 'rebuild' parameter specified");
2374 } else if (new_devs && new_devs != rebuilds) {
2375 DMERR("%u 'rebuild' devices cannot be injected into"
2376 " a raid set with %u other first-time devices",
2377 rebuilds, new_devs);
2379 } else if (rebuilds) {
2380 if (rebuild_and_new && rebuilds != rebuild_and_new) {
2381 DMERR("new device%s provided without 'rebuild'",
2382 new_devs > 1 ? "s" : "");
2384 } else if (!test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) && rs_is_recovering(rs)) {
2385 DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2386 (unsigned long long) mddev->recovery_cp);
2388 } else if (rs_is_reshaping(rs)) {
2389 DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2390 (unsigned long long) mddev->reshape_position);
2396 * Now we set the Faulty bit for those devices that are
2397 * recorded in the superblock as failed.
2399 sb_retrieve_failed_devices(sb, failed_devices);
2400 rdev_for_each(r, mddev) {
2401 if (test_bit(Journal, &rdev->flags) ||
2404 sb2 = page_address(r->sb_page);
2405 sb2->failed_devices = 0;
2406 memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
2409 * Check for any device re-ordering.
2411 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2412 role = le32_to_cpu(sb2->array_position);
2416 if (role != r->raid_disk) {
2417 if (rs_is_raid10(rs) && __is_raid10_near(mddev->layout)) {
2418 if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
2419 rs->raid_disks % rs->raid10_copies) {
2421 "Cannot change raid10 near set to odd # of devices!";
2425 sb2->array_position = cpu_to_le32(r->raid_disk);
2427 } else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
2428 !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2429 !rt_is_raid1(rs->raid_type)) {
2430 rs->ti->error = "Cannot change device positions in raid set";
2434 DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
2438 * Partial recovery is performed on
2439 * returning failed devices.
2441 if (test_bit(role, (void *) failed_devices))
2442 set_bit(Faulty, &r->flags);
2449 static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
2451 struct mddev *mddev = &rs->md;
2452 struct dm_raid_superblock *sb;
2454 if (rs_is_raid0(rs) || !rdev->sb_page || rdev->raid_disk < 0)
2457 sb = page_address(rdev->sb_page);
2460 * If mddev->events is not set, we know we have not yet initialized
2463 if (!mddev->events && super_init_validation(rs, rdev))
2466 if (le32_to_cpu(sb->compat_features) &&
2467 le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
2468 rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2472 if (sb->incompat_features) {
2473 rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
2477 /* Enable bitmap creation for RAID levels != 0 */
2478 mddev->bitmap_info.offset = (rt_is_raid0(rs->raid_type) || rs->journal_dev.dev) ? 0 : to_sector(4096);
2479 mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2481 if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2483 * Retrieve rdev size stored in superblock to be prepared for shrink.
2484 * Check extended superblock members are present otherwise the size
2487 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190)
2488 rdev->sectors = le64_to_cpu(sb->sectors);
2490 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
2491 if (rdev->recovery_offset == MaxSector)
2492 set_bit(In_sync, &rdev->flags);
2494 * If no reshape in progress -> we're recovering single
2495 * disk(s) and have to set the device(s) to out-of-sync
2497 else if (!rs_is_reshaping(rs))
2498 clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
2502 * If a device comes back, set it as not In_sync and no longer faulty.
2504 if (test_and_clear_bit(Faulty, &rdev->flags)) {
2505 rdev->recovery_offset = 0;
2506 clear_bit(In_sync, &rdev->flags);
2507 rdev->saved_raid_disk = rdev->raid_disk;
2510 /* Reshape support -> restore repective data offsets */
2511 rdev->data_offset = le64_to_cpu(sb->data_offset);
2512 rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
2518 * Analyse superblocks and select the freshest.
2520 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2523 struct md_rdev *rdev, *freshest;
2524 struct mddev *mddev = &rs->md;
2527 rdev_for_each(rdev, mddev) {
2528 if (test_bit(Journal, &rdev->flags))
2531 if (!rdev->meta_bdev)
2534 /* Set superblock offset/size for metadata device. */
2536 rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
2537 if (rdev->sb_size < sizeof(struct dm_raid_superblock) || rdev->sb_size > PAGE_SIZE) {
2538 DMERR("superblock size of a logical block is no longer valid");
2543 * Skipping super_load due to CTR_FLAG_SYNC will cause
2544 * the array to undergo initialization again as
2545 * though it were new. This is the intended effect
2546 * of the "sync" directive.
2548 * With reshaping capability added, we must ensure that
2549 * that the "sync" directive is disallowed during the reshape.
2551 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2554 r = super_load(rdev, freshest);
2563 /* This is a failure to read the superblock from the metadata device. */
2565 * We have to keep any raid0 data/metadata device pairs or
2566 * the MD raid0 personality will fail to start the array.
2568 if (rs_is_raid0(rs))
2572 * We keep the dm_devs to be able to emit the device tuple
2573 * properly on the table line in raid_status() (rather than
2574 * mistakenly acting as if '- -' got passed into the constructor).
2576 * The rdev has to stay on the same_set list to allow for
2577 * the attempt to restore faulty devices on second resume.
2579 rdev->raid_disk = rdev->saved_raid_disk = -1;
2588 * Validation of the freshest device provides the source of
2589 * validation for the remaining devices.
2591 rs->ti->error = "Unable to assemble array: Invalid superblocks";
2592 if (super_validate(rs, freshest))
2595 if (validate_raid_redundancy(rs)) {
2596 rs->ti->error = "Insufficient redundancy to activate array";
2600 rdev_for_each(rdev, mddev)
2601 if (!test_bit(Journal, &rdev->flags) &&
2603 super_validate(rs, rdev))
2609 * Adjust data_offset and new_data_offset on all disk members of @rs
2610 * for out of place reshaping if requested by contructor
2612 * We need free space at the beginning of each raid disk for forward
2613 * and at the end for backward reshapes which userspace has to provide
2614 * via remapping/reordering of space.
2616 static int rs_adjust_data_offsets(struct raid_set *rs)
2618 sector_t data_offset = 0, new_data_offset = 0;
2619 struct md_rdev *rdev;
2621 /* Constructor did not request data offset change */
2622 if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2623 if (!rs_is_reshapable(rs))
2629 /* HM FIXME: get In_Sync raid_dev? */
2630 rdev = &rs->dev[0].rdev;
2632 if (rs->delta_disks < 0) {
2634 * Removing disks (reshaping backwards):
2636 * - before reshape: data is at offset 0 and free space
2637 * is at end of each component LV
2639 * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2642 new_data_offset = rs->data_offset;
2644 } else if (rs->delta_disks > 0) {
2646 * Adding disks (reshaping forwards):
2648 * - before reshape: data is at offset rs->data_offset != 0 and
2649 * free space is at begin of each component LV
2651 * - after reshape: data is at offset 0 on each component LV
2653 data_offset = rs->data_offset;
2654 new_data_offset = 0;
2658 * User space passes in 0 for data offset after having removed reshape space
2660 * - or - (data offset != 0)
2662 * Changing RAID layout or chunk size -> toggle offsets
2664 * - before reshape: data is at offset rs->data_offset 0 and
2665 * free space is at end of each component LV
2667 * data is at offset rs->data_offset != 0 and
2668 * free space is at begin of each component LV
2670 * - after reshape: data is at offset 0 if it was at offset != 0
2671 * or at offset != 0 if it was at offset 0
2672 * on each component LV
2675 data_offset = rs->data_offset ? rdev->data_offset : 0;
2676 new_data_offset = data_offset ? 0 : rs->data_offset;
2677 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2681 * Make sure we got a minimum amount of free sectors per device
2683 if (rs->data_offset &&
2684 to_sector(i_size_read(rdev->bdev->bd_inode)) - rs->md.dev_sectors < MIN_FREE_RESHAPE_SPACE) {
2685 rs->ti->error = data_offset ? "No space for forward reshape" :
2686 "No space for backward reshape";
2691 * Raise recovery_cp in case data_offset != 0 to
2692 * avoid false recovery positives in the constructor.
2694 if (rs->md.recovery_cp < rs->md.dev_sectors)
2695 rs->md.recovery_cp += rs->dev[0].rdev.data_offset;
2697 /* Adjust data offsets on all rdevs but on any raid4/5/6 journal device */
2698 rdev_for_each(rdev, &rs->md) {
2699 if (!test_bit(Journal, &rdev->flags)) {
2700 rdev->data_offset = data_offset;
2701 rdev->new_data_offset = new_data_offset;
2708 /* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2709 static void __reorder_raid_disk_indexes(struct raid_set *rs)
2712 struct md_rdev *rdev;
2714 rdev_for_each(rdev, &rs->md) {
2715 if (!test_bit(Journal, &rdev->flags)) {
2716 rdev->raid_disk = i++;
2717 rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2723 * Setup @rs for takeover by a different raid level
2725 static int rs_setup_takeover(struct raid_set *rs)
2727 struct mddev *mddev = &rs->md;
2728 struct md_rdev *rdev;
2729 unsigned int d = mddev->raid_disks = rs->raid_disks;
2730 sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2732 if (rt_is_raid10(rs->raid_type)) {
2733 if (rs_is_raid0(rs)) {
2734 /* Userpace reordered disks -> adjust raid_disk indexes */
2735 __reorder_raid_disk_indexes(rs);
2737 /* raid0 -> raid10_far layout */
2738 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2740 } else if (rs_is_raid1(rs))
2741 /* raid1 -> raid10_near layout */
2742 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2749 clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2750 mddev->recovery_cp = MaxSector;
2753 rdev = &rs->dev[d].rdev;
2755 if (test_bit(d, (void *) rs->rebuild_disks)) {
2756 clear_bit(In_sync, &rdev->flags);
2757 clear_bit(Faulty, &rdev->flags);
2758 mddev->recovery_cp = rdev->recovery_offset = 0;
2759 /* Bitmap has to be created when we do an "up" takeover */
2760 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2763 rdev->new_data_offset = new_data_offset;
2769 /* Prepare @rs for reshape */
2770 static int rs_prepare_reshape(struct raid_set *rs)
2773 struct mddev *mddev = &rs->md;
2775 if (rs_is_raid10(rs)) {
2776 if (rs->raid_disks != mddev->raid_disks &&
2777 __is_raid10_near(mddev->layout) &&
2778 rs->raid10_copies &&
2779 rs->raid10_copies != __raid10_near_copies(mddev->layout)) {
2781 * raid disk have to be multiple of data copies to allow this conversion,
2783 * This is actually not a reshape it is a
2784 * rebuild of any additional mirrors per group
2786 if (rs->raid_disks % rs->raid10_copies) {
2787 rs->ti->error = "Can't reshape raid10 mirror groups";
2791 /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2792 __reorder_raid_disk_indexes(rs);
2793 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2795 mddev->new_layout = mddev->layout;
2800 } else if (rs_is_raid456(rs))
2803 else if (rs_is_raid1(rs)) {
2804 if (rs->delta_disks) {
2805 /* Process raid1 via delta_disks */
2806 mddev->degraded = rs->delta_disks < 0 ? -rs->delta_disks : rs->delta_disks;
2809 /* Process raid1 without delta_disks */
2810 mddev->raid_disks = rs->raid_disks;
2814 rs->ti->error = "Called with bogus raid type";
2819 set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2820 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2821 } else if (mddev->raid_disks < rs->raid_disks)
2822 /* Create new superblocks and bitmaps, if any new disks */
2823 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2828 /* Get reshape sectors from data_offsets or raid set */
2829 static sector_t _get_reshape_sectors(struct raid_set *rs)
2831 struct md_rdev *rdev;
2832 sector_t reshape_sectors = 0;
2834 rdev_for_each(rdev, &rs->md)
2835 if (!test_bit(Journal, &rdev->flags)) {
2836 reshape_sectors = (rdev->data_offset > rdev->new_data_offset) ?
2837 rdev->data_offset - rdev->new_data_offset :
2838 rdev->new_data_offset - rdev->data_offset;
2842 return max(reshape_sectors, (sector_t) rs->data_offset);
2847 * - change raid layout
2848 * - change chunk size
2852 static int rs_setup_reshape(struct raid_set *rs)
2855 unsigned int cur_raid_devs, d;
2856 sector_t reshape_sectors = _get_reshape_sectors(rs);
2857 struct mddev *mddev = &rs->md;
2858 struct md_rdev *rdev;
2860 mddev->delta_disks = rs->delta_disks;
2861 cur_raid_devs = mddev->raid_disks;
2863 /* Ignore impossible layout change whilst adding/removing disks */
2864 if (mddev->delta_disks &&
2865 mddev->layout != mddev->new_layout) {
2866 DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2867 mddev->new_layout = mddev->layout;
2871 * Adjust array size:
2873 * - in case of adding disk(s), array size has
2874 * to grow after the disk adding reshape,
2875 * which'll hapen in the event handler;
2876 * reshape will happen forward, so space has to
2877 * be available at the beginning of each disk
2879 * - in case of removing disk(s), array size
2880 * has to shrink before starting the reshape,
2881 * which'll happen here;
2882 * reshape will happen backward, so space has to
2883 * be available at the end of each disk
2885 * - data_offset and new_data_offset are
2886 * adjusted for aforementioned out of place
2887 * reshaping based on userspace passing in
2888 * the "data_offset <sectors>" key/value
2889 * pair via the constructor
2893 if (rs->delta_disks > 0) {
2894 /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2895 for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2896 rdev = &rs->dev[d].rdev;
2897 clear_bit(In_sync, &rdev->flags);
2900 * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2901 * by md, which'll store that erroneously in the superblock on reshape
2903 rdev->saved_raid_disk = -1;
2904 rdev->raid_disk = d;
2906 rdev->sectors = mddev->dev_sectors;
2907 rdev->recovery_offset = rs_is_raid1(rs) ? 0 : MaxSector;
2910 mddev->reshape_backwards = 0; /* adding disk(s) -> forward reshape */
2912 /* Remove disk(s) */
2913 } else if (rs->delta_disks < 0) {
2914 r = rs_set_dev_and_array_sectors(rs, rs->ti->len, true);
2915 mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2917 /* Change layout and/or chunk size */
2920 * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2922 * keeping number of disks and do layout change ->
2924 * toggle reshape_backward depending on data_offset:
2926 * - free space upfront -> reshape forward
2928 * - free space at the end -> reshape backward
2931 * This utilizes free reshape space avoiding the need
2932 * for userspace to move (parts of) LV segments in
2933 * case of layout/chunksize change (for disk
2934 * adding/removing reshape space has to be at
2935 * the proper address (see above with delta_disks):
2937 * add disk(s) -> begin
2938 * remove disk(s)-> end
2940 mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2944 * Adjust device size for forward reshape
2945 * because md_finish_reshape() reduces it.
2947 if (!mddev->reshape_backwards)
2948 rdev_for_each(rdev, &rs->md)
2949 if (!test_bit(Journal, &rdev->flags))
2950 rdev->sectors += reshape_sectors;
2956 * Enable/disable discard support on RAID set depending on
2957 * RAID level and discard properties of underlying RAID members.
2959 static void configure_discard_support(struct raid_set *rs)
2963 struct dm_target *ti = rs->ti;
2966 * XXX: RAID level 4,5,6 require zeroing for safety.
2968 raid456 = rs_is_raid456(rs);
2970 for (i = 0; i < rs->raid_disks; i++) {
2971 struct request_queue *q;
2973 if (!rs->dev[i].rdev.bdev)
2976 q = bdev_get_queue(rs->dev[i].rdev.bdev);
2977 if (!q || !blk_queue_discard(q))
2981 if (!devices_handle_discard_safely) {
2982 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2983 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2989 ti->num_discard_bios = 1;
2993 * Construct a RAID0/1/10/4/5/6 mapping:
2995 * <raid_type> <#raid_params> <raid_params>{0,} \
2996 * <#raid_devs> [<meta_dev1> <dev1>]{1,}
2998 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
2999 * details on possible <raid_params>.
3001 * Userspace is free to initialize the metadata devices, hence the superblocks to
3002 * enforce recreation based on the passed in table parameters.
3005 static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3008 bool resize = false;
3009 struct raid_type *rt;
3010 unsigned int num_raid_params, num_raid_devs;
3011 sector_t sb_array_sectors, rdev_sectors, reshape_sectors;
3012 struct raid_set *rs = NULL;
3014 struct rs_layout rs_layout;
3015 struct dm_arg_set as = { argc, argv }, as_nrd;
3016 struct dm_arg _args[] = {
3017 { 0, as.argc, "Cannot understand number of raid parameters" },
3018 { 1, 254, "Cannot understand number of raid devices parameters" }
3021 /* Must have <raid_type> */
3022 arg = dm_shift_arg(&as);
3024 ti->error = "No arguments";
3028 rt = get_raid_type(arg);
3030 ti->error = "Unrecognised raid_type";
3034 /* Must have <#raid_params> */
3035 if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
3038 /* number of raid device tupples <meta_dev data_dev> */
3040 dm_consume_args(&as_nrd, num_raid_params);
3041 _args[1].max = (as_nrd.argc - 1) / 2;
3042 if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
3045 if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
3046 ti->error = "Invalid number of supplied raid devices";
3050 rs = raid_set_alloc(ti, rt, num_raid_devs);
3054 r = parse_raid_params(rs, &as, num_raid_params);
3058 r = parse_dev_params(rs, &as);
3062 rs->md.sync_super = super_sync;
3065 * Calculate ctr requested array and device sizes to allow
3066 * for superblock analysis needing device sizes defined.
3068 * Any existing superblock will overwrite the array and device sizes
3070 r = rs_set_dev_and_array_sectors(rs, rs->ti->len, false);
3074 /* Memorize just calculated, potentially larger sizes to grow the raid set in preresume */
3075 rs->array_sectors = rs->md.array_sectors;
3076 rs->dev_sectors = rs->md.dev_sectors;
3079 * Backup any new raid set level, layout, ...
3080 * requested to be able to compare to superblock
3081 * members for conversion decisions.
3083 rs_config_backup(rs, &rs_layout);
3085 r = analyse_superblocks(ti, rs);
3089 /* All in-core metadata now as of current superblocks after calling analyse_superblocks() */
3090 sb_array_sectors = rs->md.array_sectors;
3091 rdev_sectors = __rdev_sectors(rs);
3092 if (!rdev_sectors) {
3093 ti->error = "Invalid rdev size";
3099 reshape_sectors = _get_reshape_sectors(rs);
3100 if (rs->dev_sectors != rdev_sectors) {
3101 resize = (rs->dev_sectors != rdev_sectors - reshape_sectors);
3102 if (rs->dev_sectors > rdev_sectors - reshape_sectors)
3103 set_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3106 INIT_WORK(&rs->md.event_work, do_table_event);
3108 ti->num_flush_bios = 1;
3110 /* Restore any requested new layout for conversion decision */
3111 rs_config_restore(rs, &rs_layout);
3114 * Now that we have any superblock metadata available,
3115 * check for new, recovering, reshaping, to be taken over,
3116 * to be reshaped or an existing, unchanged raid set to
3119 if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
3120 /* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
3121 if (rs_is_raid6(rs) &&
3122 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
3123 ti->error = "'nosync' not allowed for new raid6 set";
3127 rs_setup_recovery(rs, 0);
3128 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3130 } else if (rs_is_recovering(rs)) {
3131 /* A recovering raid set may be resized */
3133 } else if (rs_is_reshaping(rs)) {
3134 /* Have to reject size change request during reshape */
3136 ti->error = "Can't resize a reshaping raid set";
3141 } else if (rs_takeover_requested(rs)) {
3142 if (rs_is_reshaping(rs)) {
3143 ti->error = "Can't takeover a reshaping raid set";
3148 /* We can't takeover a journaled raid4/5/6 */
3149 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3150 ti->error = "Can't takeover a journaled raid4/5/6 set";
3156 * If a takeover is needed, userspace sets any additional
3157 * devices to rebuild and we can check for a valid request here.
3159 * If acceptible, set the level to the new requested
3160 * one, prohibit requesting recovery, allow the raid
3161 * set to run and store superblocks during resume.
3163 r = rs_check_takeover(rs);
3167 r = rs_setup_takeover(rs);
3171 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3172 /* Takeover ain't recovery, so disable recovery */
3173 rs_setup_recovery(rs, MaxSector);
3175 } else if (rs_reshape_requested(rs)) {
3176 /* Only on size extensions, not on reshapes. */
3177 clear_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3180 * No need to check for 'ongoing' takeover here, because takeover
3181 * is an instant operation as oposed to an ongoing reshape.
3184 /* We can't reshape a journaled raid4/5/6 */
3185 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3186 ti->error = "Can't reshape a journaled raid4/5/6 set";
3191 /* Out-of-place space has to be available to allow for a reshape unless raid1! */
3192 if (reshape_sectors || rs_is_raid1(rs)) {
3194 * We can only prepare for a reshape here, because the
3195 * raid set needs to run to provide the repective reshape
3196 * check functions via its MD personality instance.
3198 * So do the reshape check after md_run() succeeded.
3200 r = rs_prepare_reshape(rs);
3204 /* Reshaping ain't recovery, so disable recovery */
3205 rs_setup_recovery(rs, MaxSector);
3210 /* May not set recovery when a device rebuild is requested */
3211 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3212 clear_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3213 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3214 } else if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags)) {
3216 * Set raid set to current size, i.e. non-grown size
3217 * as of superblocks to grow to new size in preresume.
3219 r = rs_set_dev_and_array_sectors(rs, sb_array_sectors, false);
3223 rs_setup_recovery(rs, rs->md.recovery_cp < rs->md.dev_sectors ? rs->md.recovery_cp : rs->md.dev_sectors);
3225 /* This is no size change or it is shrinking, update size and record in superblocks */
3226 r = rs_set_dev_and_array_sectors(rs, rs->ti->len, false);
3230 if (sb_array_sectors > rs->array_sectors)
3231 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3236 /* If constructor requested it, change data and new_data offsets */
3237 r = rs_adjust_data_offsets(rs);
3241 /* Start raid set read-only and assumed clean to change in raid_resume() */
3245 /* Keep array frozen */
3246 set_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
3248 /* Has to be held on running the array */
3249 mddev_lock_nointr(&rs->md);
3250 r = md_run(&rs->md);
3251 rs->md.in_sync = 0; /* Assume already marked dirty */
3253 ti->error = "Failed to run raid array";
3254 mddev_unlock(&rs->md);
3258 r = md_start(&rs->md);
3261 ti->error = "Failed to start raid array";
3262 mddev_unlock(&rs->md);
3266 rs->callbacks.congested_fn = raid_is_congested;
3267 dm_table_add_target_callbacks(ti->table, &rs->callbacks);
3269 /* If raid4/5/6 journal mode explicitly requested (only possible with journal dev) -> set it */
3270 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
3271 r = r5c_journal_mode_set(&rs->md, rs->journal_dev.mode);
3273 ti->error = "Failed to set raid4/5/6 journal mode";
3274 mddev_unlock(&rs->md);
3275 goto bad_journal_mode_set;
3279 mddev_suspend(&rs->md);
3280 set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags);
3282 /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
3283 if (rs_is_raid456(rs)) {
3284 r = rs_set_raid456_stripe_cache(rs);
3286 goto bad_stripe_cache;
3289 /* Now do an early reshape check */
3290 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3291 r = rs_check_reshape(rs);
3293 goto bad_check_reshape;
3295 /* Restore new, ctr requested layout to perform check */
3296 rs_config_restore(rs, &rs_layout);
3298 if (rs->md.pers->start_reshape) {
3299 r = rs->md.pers->check_reshape(&rs->md);
3301 ti->error = "Reshape check failed";
3302 goto bad_check_reshape;
3307 /* Disable/enable discard support on raid set. */
3308 configure_discard_support(rs);
3310 mddev_unlock(&rs->md);
3314 bad_journal_mode_set:
3324 static void raid_dtr(struct dm_target *ti)
3326 struct raid_set *rs = ti->private;
3328 list_del_init(&rs->callbacks.list);
3333 static int raid_map(struct dm_target *ti, struct bio *bio)
3335 struct raid_set *rs = ti->private;
3336 struct mddev *mddev = &rs->md;
3339 * If we're reshaping to add disk(s)), ti->len and
3340 * mddev->array_sectors will differ during the process
3341 * (ti->len > mddev->array_sectors), so we have to requeue
3342 * bios with addresses > mddev->array_sectors here or
3343 * there will occur accesses past EOD of the component
3344 * data images thus erroring the raid set.
3346 if (unlikely(bio_end_sector(bio) > mddev->array_sectors))
3347 return DM_MAPIO_REQUEUE;
3349 md_handle_request(mddev, bio);
3351 return DM_MAPIO_SUBMITTED;
3354 /* Return sync state string for @state */
3355 enum sync_state { st_frozen, st_reshape, st_resync, st_check, st_repair, st_recover, st_idle };
3356 static const char *sync_str(enum sync_state state)
3358 /* Has to be in above sync_state order! */
3359 static const char *sync_strs[] = {
3369 return __within_range(state, 0, ARRAY_SIZE(sync_strs) - 1) ? sync_strs[state] : "undef";
3372 /* Return enum sync_state for @mddev derived from @recovery flags */
3373 static enum sync_state decipher_sync_action(struct mddev *mddev, unsigned long recovery)
3375 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
3378 /* The MD sync thread can be done with io or be interrupted but still be running */
3379 if (!test_bit(MD_RECOVERY_DONE, &recovery) &&
3380 (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
3381 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery)))) {
3382 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
3385 if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
3386 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
3388 if (test_bit(MD_RECOVERY_CHECK, &recovery))
3393 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3396 if (mddev->reshape_position != MaxSector)
3404 * Return status string for @rdev
3406 * Status characters:
3408 * 'D' = Dead/Failed raid set component or raid4/5/6 journal device
3409 * 'a' = Alive but not in-sync raid set component _or_ alive raid4/5/6 'write_back' journal device
3410 * 'A' = Alive and in-sync raid set component _or_ alive raid4/5/6 'write_through' journal device
3411 * '-' = Non-existing device (i.e. uspace passed '- -' into the ctr)
3413 static const char *__raid_dev_status(struct raid_set *rs, struct md_rdev *rdev)
3417 else if (test_bit(Faulty, &rdev->flags))
3419 else if (test_bit(Journal, &rdev->flags))
3420 return (rs->journal_dev.mode == R5C_JOURNAL_MODE_WRITE_THROUGH) ? "A" : "a";
3421 else if (test_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags) ||
3422 (!test_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags) &&
3423 !test_bit(In_sync, &rdev->flags)))
3429 /* Helper to return resync/reshape progress for @rs and runtime flags for raid set in sync / resynching */
3430 static sector_t rs_get_progress(struct raid_set *rs, unsigned long recovery,
3431 sector_t resync_max_sectors)
3434 enum sync_state state;
3435 struct mddev *mddev = &rs->md;
3437 clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3438 clear_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3440 if (rs_is_raid0(rs)) {
3441 r = resync_max_sectors;
3442 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3445 state = decipher_sync_action(mddev, recovery);
3447 if (state == st_idle && !test_bit(MD_RECOVERY_INTR, &recovery))
3448 r = mddev->recovery_cp;
3450 r = mddev->curr_resync_completed;
3452 if (state == st_idle && r >= resync_max_sectors) {
3456 /* In case we have finished recovering, the array is in sync. */
3457 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3458 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3460 } else if (state == st_recover)
3462 * In case we are recovering, the array is not in sync
3463 * and health chars should show the recovering legs.
3466 else if (state == st_resync)
3468 * If "resync" is occurring, the raid set
3469 * is or may be out of sync hence the health
3470 * characters shall be 'a'.
3472 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3473 else if (state == st_reshape)
3475 * If "reshape" is occurring, the raid set
3476 * is or may be out of sync hence the health
3477 * characters shall be 'a'.
3479 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3481 else if (state == st_check || state == st_repair)
3483 * If "check" or "repair" is occurring, the raid set has
3484 * undergone an initial sync and the health characters
3485 * should not be 'a' anymore.
3487 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3490 struct md_rdev *rdev;
3493 * We are idle and recovery is needed, prevent 'A' chars race
3494 * caused by components still set to in-sync by constructor.
3496 if (test_bit(MD_RECOVERY_NEEDED, &recovery))
3497 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3500 * The raid set may be doing an initial sync, or it may
3501 * be rebuilding individual components. If all the
3502 * devices are In_sync, then it is the raid set that is
3503 * being initialized.
3505 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3506 rdev_for_each(rdev, mddev)
3507 if (!test_bit(Journal, &rdev->flags) &&
3508 !test_bit(In_sync, &rdev->flags)) {
3509 clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3515 return min(r, resync_max_sectors);
3518 /* Helper to return @dev name or "-" if !@dev */
3519 static const char *__get_dev_name(struct dm_dev *dev)
3521 return dev ? dev->name : "-";
3524 static void raid_status(struct dm_target *ti, status_type_t type,
3525 unsigned int status_flags, char *result, unsigned int maxlen)
3527 struct raid_set *rs = ti->private;
3528 struct mddev *mddev = &rs->md;
3529 struct r5conf *conf = mddev->private;
3530 int i, max_nr_stripes = conf ? conf->max_nr_stripes : 0;
3531 unsigned long recovery;
3532 unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3533 unsigned int sz = 0;
3534 unsigned int rebuild_disks;
3535 unsigned int write_mostly_params = 0;
3536 sector_t progress, resync_max_sectors, resync_mismatches;
3537 const char *sync_action;
3538 struct raid_type *rt;
3541 case STATUSTYPE_INFO:
3542 /* *Should* always succeed */
3543 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3547 DMEMIT("%s %d ", rt->name, mddev->raid_disks);
3549 /* Access most recent mddev properties for status output */
3551 recovery = rs->md.recovery;
3552 /* Get sensible max sectors even if raid set not yet started */
3553 resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
3554 mddev->resync_max_sectors : mddev->dev_sectors;
3555 progress = rs_get_progress(rs, recovery, resync_max_sectors);
3556 resync_mismatches = (mddev->last_sync_action && !strcasecmp(mddev->last_sync_action, "check")) ?
3557 atomic64_read(&mddev->resync_mismatches) : 0;
3558 sync_action = sync_str(decipher_sync_action(&rs->md, recovery));
3560 /* HM FIXME: do we want another state char for raid0? It shows 'D'/'A'/'-' now */
3561 for (i = 0; i < rs->raid_disks; i++)
3562 DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev));
3565 * In-sync/Reshape ratio:
3566 * The in-sync ratio shows the progress of:
3567 * - Initializing the raid set
3568 * - Rebuilding a subset of devices of the raid set
3569 * The user can distinguish between the two by referring
3570 * to the status characters.
3572 * The reshape ratio shows the progress of
3573 * changing the raid layout or the number of
3574 * disks of a raid set
3576 DMEMIT(" %llu/%llu", (unsigned long long) progress,
3577 (unsigned long long) resync_max_sectors);
3583 * See Documentation/admin-guide/device-mapper/dm-raid.rst for
3584 * information on each of these states.
3586 DMEMIT(" %s", sync_action);
3591 * resync_mismatches/mismatch_cnt
3592 * This field shows the number of discrepancies found when
3593 * performing a "check" of the raid set.
3595 DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3600 * data_offset (needed for out of space reshaping)
3601 * This field shows the data offset into the data
3602 * image LV where the first stripes data starts.
3604 * We keep data_offset equal on all raid disks of the set,
3605 * so retrieving it from the first raid disk is sufficient.
3607 DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
3612 DMEMIT(" %s", test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ?
3613 __raid_dev_status(rs, &rs->journal_dev.rdev) : "-");
3616 case STATUSTYPE_TABLE:
3617 /* Report the table line string you would use to construct this raid set */
3619 /* Calculate raid parameter count */
3620 for (i = 0; i < rs->raid_disks; i++)
3621 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3622 write_mostly_params += 2;
3623 rebuild_disks = memweight(rs->rebuild_disks, DISKS_ARRAY_ELEMS * sizeof(*rs->rebuild_disks));
3624 raid_param_cnt += rebuild_disks * 2 +
3625 write_mostly_params +
3626 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3627 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2 +
3628 (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ? 2 : 0) +
3629 (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags) ? 2 : 0);
3631 /* Emit table line */
3632 /* This has to be in the documented order for userspace! */
3633 DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
3634 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
3635 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
3636 if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3637 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
3639 for (i = 0; i < rs->raid_disks; i++)
3640 if (test_bit(rs->dev[i].rdev.raid_disk, (void *) rs->rebuild_disks))
3641 DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD),
3642 rs->dev[i].rdev.raid_disk);
3643 if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3644 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3645 mddev->bitmap_info.daemon_sleep);
3646 if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3647 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3648 mddev->sync_speed_min);
3649 if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3650 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3651 mddev->sync_speed_max);
3652 if (write_mostly_params)
3653 for (i = 0; i < rs->raid_disks; i++)
3654 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3655 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3656 rs->dev[i].rdev.raid_disk);
3657 if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
3658 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
3659 mddev->bitmap_info.max_write_behind);
3660 if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3661 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3663 if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3664 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3665 (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3666 if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3667 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3668 raid10_md_layout_to_copies(mddev->layout));
3669 if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3670 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3671 raid10_md_layout_to_format(mddev->layout));
3672 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3673 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3674 max(rs->delta_disks, mddev->delta_disks));
3675 if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3676 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3677 (unsigned long long) rs->data_offset);
3678 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags))
3679 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV),
3680 __get_dev_name(rs->journal_dev.dev));
3681 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags))
3682 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE),
3683 md_journal_mode_to_dm_raid(rs->journal_dev.mode));
3684 DMEMIT(" %d", rs->raid_disks);
3685 for (i = 0; i < rs->raid_disks; i++)
3686 DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
3687 __get_dev_name(rs->dev[i].data_dev));
3691 static int raid_message(struct dm_target *ti, unsigned int argc, char **argv,
3692 char *result, unsigned maxlen)
3694 struct raid_set *rs = ti->private;
3695 struct mddev *mddev = &rs->md;
3697 if (!mddev->pers || !mddev->pers->sync_request)
3700 if (!strcasecmp(argv[0], "frozen"))
3701 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3703 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3705 if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
3706 if (mddev->sync_thread) {
3707 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3708 md_reap_sync_thread(mddev);
3710 } else if (decipher_sync_action(mddev, mddev->recovery) != st_idle)
3712 else if (!strcasecmp(argv[0], "resync"))
3713 ; /* MD_RECOVERY_NEEDED set below */
3714 else if (!strcasecmp(argv[0], "recover"))
3715 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3717 if (!strcasecmp(argv[0], "check")) {
3718 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3719 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3720 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3721 } else if (!strcasecmp(argv[0], "repair")) {
3722 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3723 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3727 if (mddev->ro == 2) {
3728 /* A write to sync_action is enough to justify
3729 * canceling read-auto mode
3732 if (!mddev->suspended && mddev->sync_thread)
3733 md_wakeup_thread(mddev->sync_thread);
3735 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3736 if (!mddev->suspended && mddev->thread)
3737 md_wakeup_thread(mddev->thread);
3742 static int raid_iterate_devices(struct dm_target *ti,
3743 iterate_devices_callout_fn fn, void *data)
3745 struct raid_set *rs = ti->private;
3749 for (i = 0; !r && i < rs->md.raid_disks; i++)
3750 if (rs->dev[i].data_dev)
3752 rs->dev[i].data_dev,
3753 0, /* No offset on data devs */
3760 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3762 struct raid_set *rs = ti->private;
3763 unsigned int chunk_size_bytes = to_bytes(rs->md.chunk_sectors);
3765 blk_limits_io_min(limits, chunk_size_bytes);
3766 blk_limits_io_opt(limits, chunk_size_bytes * mddev_data_stripes(rs));
3769 * RAID1 and RAID10 personalities require bio splitting,
3770 * RAID0/4/5/6 don't and process large discard bios properly.
3772 if (rs_is_raid1(rs) || rs_is_raid10(rs)) {
3773 limits->discard_granularity = chunk_size_bytes;
3774 limits->max_discard_sectors = rs->md.chunk_sectors;
3778 static void raid_postsuspend(struct dm_target *ti)
3780 struct raid_set *rs = ti->private;
3782 if (!test_and_set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
3783 /* Writes have to be stopped before suspending to avoid deadlocks. */
3784 if (!test_bit(MD_RECOVERY_FROZEN, &rs->md.recovery))
3785 md_stop_writes(&rs->md);
3787 mddev_lock_nointr(&rs->md);
3788 mddev_suspend(&rs->md);
3789 mddev_unlock(&rs->md);
3793 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
3796 uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
3797 unsigned long flags;
3798 bool cleared = false;
3799 struct dm_raid_superblock *sb;
3800 struct mddev *mddev = &rs->md;
3803 /* RAID personalities have to provide hot add/remove methods or we need to bail out. */
3804 if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
3807 memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
3809 for (i = 0; i < mddev->raid_disks; i++) {
3810 r = &rs->dev[i].rdev;
3811 /* HM FIXME: enhance journal device recovery processing */
3812 if (test_bit(Journal, &r->flags))
3815 if (test_bit(Faulty, &r->flags) &&
3816 r->meta_bdev && !read_disk_sb(r, r->sb_size, true)) {
3817 DMINFO("Faulty %s device #%d has readable super block."
3818 " Attempting to revive it.",
3819 rs->raid_type->name, i);
3822 * Faulty bit may be set, but sometimes the array can
3823 * be suspended before the personalities can respond
3824 * by removing the device from the array (i.e. calling
3825 * 'hot_remove_disk'). If they haven't yet removed
3826 * the failed device, its 'raid_disk' number will be
3827 * '>= 0' - meaning we must call this function
3831 clear_bit(In_sync, &r->flags); /* Mandatory for hot remove. */
3832 if (r->raid_disk >= 0) {
3833 if (mddev->pers->hot_remove_disk(mddev, r)) {
3834 /* Failed to revive this device, try next */
3839 r->raid_disk = r->saved_raid_disk = i;
3841 clear_bit(Faulty, &r->flags);
3842 clear_bit(WriteErrorSeen, &r->flags);
3844 if (mddev->pers->hot_add_disk(mddev, r)) {
3845 /* Failed to revive this device, try next */
3846 r->raid_disk = r->saved_raid_disk = -1;
3849 clear_bit(In_sync, &r->flags);
3850 r->recovery_offset = 0;
3851 set_bit(i, (void *) cleared_failed_devices);
3857 /* If any failed devices could be cleared, update all sbs failed_devices bits */
3859 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
3861 rdev_for_each(r, &rs->md) {
3862 if (test_bit(Journal, &r->flags))
3865 sb = page_address(r->sb_page);
3866 sb_retrieve_failed_devices(sb, failed_devices);
3868 for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
3869 failed_devices[i] &= ~cleared_failed_devices[i];
3871 sb_update_failed_devices(sb, failed_devices);
3876 static int __load_dirty_region_bitmap(struct raid_set *rs)
3880 /* Try loading the bitmap unless "raid0", which does not have one */
3881 if (!rs_is_raid0(rs) &&
3882 !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
3883 r = md_bitmap_load(&rs->md);
3885 DMERR("Failed to load bitmap");
3891 /* Enforce updating all superblocks */
3892 static void rs_update_sbs(struct raid_set *rs)
3894 struct mddev *mddev = &rs->md;
3897 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3899 md_update_sb(mddev, 1);
3904 * Reshape changes raid algorithm of @rs to new one within personality
3905 * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3906 * disks from a raid set thus growing/shrinking it or resizes the set
3908 * Call mddev_lock_nointr() before!
3910 static int rs_start_reshape(struct raid_set *rs)
3913 struct mddev *mddev = &rs->md;
3914 struct md_personality *pers = mddev->pers;
3916 /* Don't allow the sync thread to work until the table gets reloaded. */
3917 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
3919 r = rs_setup_reshape(rs);
3924 * Check any reshape constraints enforced by the personalility
3926 * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3928 r = pers->check_reshape(mddev);
3930 rs->ti->error = "pers->check_reshape() failed";
3935 * Personality may not provide start reshape method in which
3936 * case check_reshape above has already covered everything
3938 if (pers->start_reshape) {
3939 r = pers->start_reshape(mddev);
3941 rs->ti->error = "pers->start_reshape() failed";
3947 * Now reshape got set up, update superblocks to
3948 * reflect the fact so that a table reload will
3949 * access proper superblock content in the ctr.
3956 static int raid_preresume(struct dm_target *ti)
3959 struct raid_set *rs = ti->private;
3960 struct mddev *mddev = &rs->md;
3962 /* This is a resume after a suspend of the set -> it's already started. */
3963 if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
3967 * The superblocks need to be updated on disk if the
3968 * array is new or new devices got added (thus zeroed
3969 * out by userspace) or __load_dirty_region_bitmap
3970 * will overwrite them in core with old data or fail.
3972 if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
3975 /* Load the bitmap from disk unless raid0 */
3976 r = __load_dirty_region_bitmap(rs);
3980 /* We are extending the raid set size, adjust mddev/md_rdev sizes and set capacity. */
3981 if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags)) {
3982 mddev->array_sectors = rs->array_sectors;
3983 mddev->dev_sectors = rs->dev_sectors;
3984 rs_set_rdev_sectors(rs);
3985 rs_set_capacity(rs);
3988 /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) or grown device size */
3989 if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) && mddev->bitmap &&
3990 (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags) ||
3991 (rs->requested_bitmap_chunk_sectors &&
3992 mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)))) {
3993 int chunksize = to_bytes(rs->requested_bitmap_chunk_sectors) ?: mddev->bitmap_info.chunksize;
3995 r = md_bitmap_resize(mddev->bitmap, mddev->dev_sectors, chunksize, 0);
3997 DMERR("Failed to resize bitmap");
4000 /* Check for any resize/reshape on @rs and adjust/initiate */
4001 /* Be prepared for mddev_resume() in raid_resume() */
4002 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4003 if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
4004 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4005 mddev->resync_min = mddev->recovery_cp;
4006 if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags))
4007 mddev->resync_max_sectors = mddev->dev_sectors;
4010 /* Check for any reshape request unless new raid set */
4011 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
4012 /* Initiate a reshape. */
4013 rs_set_rdev_sectors(rs);
4014 mddev_lock_nointr(mddev);
4015 r = rs_start_reshape(rs);
4016 mddev_unlock(mddev);
4018 DMWARN("Failed to check/start reshape, continuing without change");
4025 static void raid_resume(struct dm_target *ti)
4027 struct raid_set *rs = ti->private;
4028 struct mddev *mddev = &rs->md;
4030 if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
4032 * A secondary resume while the device is active.
4033 * Take this opportunity to check whether any failed
4034 * devices are reachable again.
4036 attempt_restore_of_faulty_devices(rs);
4039 if (test_and_clear_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
4040 /* Only reduce raid set size before running a disk removing reshape. */
4041 if (mddev->delta_disks < 0)
4042 rs_set_capacity(rs);
4044 mddev_lock_nointr(mddev);
4045 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4048 mddev_resume(mddev);
4049 mddev_unlock(mddev);
4053 static struct target_type raid_target = {
4055 .version = {1, 15, 0},
4056 .module = THIS_MODULE,
4060 .status = raid_status,
4061 .message = raid_message,
4062 .iterate_devices = raid_iterate_devices,
4063 .io_hints = raid_io_hints,
4064 .postsuspend = raid_postsuspend,
4065 .preresume = raid_preresume,
4066 .resume = raid_resume,
4069 static int __init dm_raid_init(void)
4071 DMINFO("Loading target version %u.%u.%u",
4072 raid_target.version[0],
4073 raid_target.version[1],
4074 raid_target.version[2]);
4075 return dm_register_target(&raid_target);
4078 static void __exit dm_raid_exit(void)
4080 dm_unregister_target(&raid_target);
4083 module_init(dm_raid_init);
4084 module_exit(dm_raid_exit);
4086 module_param(devices_handle_discard_safely, bool, 0644);
4087 MODULE_PARM_DESC(devices_handle_discard_safely,
4088 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
4090 MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
4091 MODULE_ALIAS("dm-raid0");
4092 MODULE_ALIAS("dm-raid1");
4093 MODULE_ALIAS("dm-raid10");
4094 MODULE_ALIAS("dm-raid4");
4095 MODULE_ALIAS("dm-raid5");
4096 MODULE_ALIAS("dm-raid6");
4097 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
4098 MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
4099 MODULE_LICENSE("GPL");