1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2010-2011 Neil Brown
4 * Copyright (C) 2010-2018 Red Hat, Inc. All rights reserved.
6 * This file is released under the GPL.
9 #include <linux/slab.h>
10 #include <linux/module.h>
16 #include "md-bitmap.h"
18 #include <linux/device-mapper.h>
20 #define DM_MSG_PREFIX "raid"
21 #define MAX_RAID_DEVICES 253 /* md-raid kernel limit */
24 * Minimum sectors of free reshape space per raid device
26 #define MIN_FREE_RESHAPE_SPACE to_sector(4*4096)
29 * Minimum journal space 4 MiB in sectors.
31 #define MIN_RAID456_JOURNAL_SPACE (4*2048)
33 static bool devices_handle_discard_safely;
36 * The following flags are used by dm-raid to set up the array state.
37 * They must be cleared before md_run is called.
39 #define FirstUse 10 /* rdev flag */
43 * Two DM devices, one to hold metadata and one to hold the
44 * actual data/parity. The reason for this is to not confuse
45 * ti->len and give more flexibility in altering size and
48 * While it is possible for this device to be associated
49 * with a different physical device than the data_dev, it
50 * is intended for it to be the same.
51 * |--------- Physical Device ---------|
52 * |- meta_dev -|------ data_dev ------|
54 struct dm_dev *meta_dev;
55 struct dm_dev *data_dev;
60 * Bits for establishing rs->ctr_flags
65 #define __CTR_FLAG_SYNC 0 /* 1 */ /* Not with raid0! */
66 #define __CTR_FLAG_NOSYNC 1 /* 1 */ /* Not with raid0! */
67 #define __CTR_FLAG_REBUILD 2 /* 2 */ /* Not with raid0! */
68 #define __CTR_FLAG_DAEMON_SLEEP 3 /* 2 */ /* Not with raid0! */
69 #define __CTR_FLAG_MIN_RECOVERY_RATE 4 /* 2 */ /* Not with raid0! */
70 #define __CTR_FLAG_MAX_RECOVERY_RATE 5 /* 2 */ /* Not with raid0! */
71 #define __CTR_FLAG_MAX_WRITE_BEHIND 6 /* 2 */ /* Only with raid1! */
72 #define __CTR_FLAG_WRITE_MOSTLY 7 /* 2 */ /* Only with raid1! */
73 #define __CTR_FLAG_STRIPE_CACHE 8 /* 2 */ /* Only with raid4/5/6! */
74 #define __CTR_FLAG_REGION_SIZE 9 /* 2 */ /* Not with raid0! */
75 #define __CTR_FLAG_RAID10_COPIES 10 /* 2 */ /* Only with raid10 */
76 #define __CTR_FLAG_RAID10_FORMAT 11 /* 2 */ /* Only with raid10 */
78 #define __CTR_FLAG_DELTA_DISKS 12 /* 2 */ /* Only with reshapable raid1/4/5/6/10! */
79 #define __CTR_FLAG_DATA_OFFSET 13 /* 2 */ /* Only with reshapable raid4/5/6/10! */
80 #define __CTR_FLAG_RAID10_USE_NEAR_SETS 14 /* 2 */ /* Only with raid10! */
83 #define __CTR_FLAG_JOURNAL_DEV 15 /* 2 */ /* Only with raid4/5/6 (journal device)! */
86 #define __CTR_FLAG_JOURNAL_MODE 16 /* 2 */ /* Only with raid4/5/6 (journal mode)! */
89 * Flags for rs->ctr_flags field.
91 #define CTR_FLAG_SYNC (1 << __CTR_FLAG_SYNC)
92 #define CTR_FLAG_NOSYNC (1 << __CTR_FLAG_NOSYNC)
93 #define CTR_FLAG_REBUILD (1 << __CTR_FLAG_REBUILD)
94 #define CTR_FLAG_DAEMON_SLEEP (1 << __CTR_FLAG_DAEMON_SLEEP)
95 #define CTR_FLAG_MIN_RECOVERY_RATE (1 << __CTR_FLAG_MIN_RECOVERY_RATE)
96 #define CTR_FLAG_MAX_RECOVERY_RATE (1 << __CTR_FLAG_MAX_RECOVERY_RATE)
97 #define CTR_FLAG_MAX_WRITE_BEHIND (1 << __CTR_FLAG_MAX_WRITE_BEHIND)
98 #define CTR_FLAG_WRITE_MOSTLY (1 << __CTR_FLAG_WRITE_MOSTLY)
99 #define CTR_FLAG_STRIPE_CACHE (1 << __CTR_FLAG_STRIPE_CACHE)
100 #define CTR_FLAG_REGION_SIZE (1 << __CTR_FLAG_REGION_SIZE)
101 #define CTR_FLAG_RAID10_COPIES (1 << __CTR_FLAG_RAID10_COPIES)
102 #define CTR_FLAG_RAID10_FORMAT (1 << __CTR_FLAG_RAID10_FORMAT)
103 #define CTR_FLAG_DELTA_DISKS (1 << __CTR_FLAG_DELTA_DISKS)
104 #define CTR_FLAG_DATA_OFFSET (1 << __CTR_FLAG_DATA_OFFSET)
105 #define CTR_FLAG_RAID10_USE_NEAR_SETS (1 << __CTR_FLAG_RAID10_USE_NEAR_SETS)
106 #define CTR_FLAG_JOURNAL_DEV (1 << __CTR_FLAG_JOURNAL_DEV)
107 #define CTR_FLAG_JOURNAL_MODE (1 << __CTR_FLAG_JOURNAL_MODE)
110 * Definitions of various constructor flags to
111 * be used in checks of valid / invalid flags
114 /* Define all any sync flags */
115 #define CTR_FLAGS_ANY_SYNC (CTR_FLAG_SYNC | CTR_FLAG_NOSYNC)
117 /* Define flags for options without argument (e.g. 'nosync') */
118 #define CTR_FLAG_OPTIONS_NO_ARGS (CTR_FLAGS_ANY_SYNC | \
119 CTR_FLAG_RAID10_USE_NEAR_SETS)
121 /* Define flags for options with one argument (e.g. 'delta_disks +2') */
122 #define CTR_FLAG_OPTIONS_ONE_ARG (CTR_FLAG_REBUILD | \
123 CTR_FLAG_WRITE_MOSTLY | \
124 CTR_FLAG_DAEMON_SLEEP | \
125 CTR_FLAG_MIN_RECOVERY_RATE | \
126 CTR_FLAG_MAX_RECOVERY_RATE | \
127 CTR_FLAG_MAX_WRITE_BEHIND | \
128 CTR_FLAG_STRIPE_CACHE | \
129 CTR_FLAG_REGION_SIZE | \
130 CTR_FLAG_RAID10_COPIES | \
131 CTR_FLAG_RAID10_FORMAT | \
132 CTR_FLAG_DELTA_DISKS | \
133 CTR_FLAG_DATA_OFFSET | \
134 CTR_FLAG_JOURNAL_DEV | \
135 CTR_FLAG_JOURNAL_MODE)
137 /* Valid options definitions per raid level... */
139 /* "raid0" does only accept data offset */
140 #define RAID0_VALID_FLAGS (CTR_FLAG_DATA_OFFSET)
142 /* "raid1" does not accept stripe cache, data offset, delta_disks or any raid10 options */
143 #define RAID1_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
145 CTR_FLAG_WRITE_MOSTLY | \
146 CTR_FLAG_DAEMON_SLEEP | \
147 CTR_FLAG_MIN_RECOVERY_RATE | \
148 CTR_FLAG_MAX_RECOVERY_RATE | \
149 CTR_FLAG_MAX_WRITE_BEHIND | \
150 CTR_FLAG_REGION_SIZE | \
151 CTR_FLAG_DELTA_DISKS | \
152 CTR_FLAG_DATA_OFFSET)
154 /* "raid10" does not accept any raid1 or stripe cache options */
155 #define RAID10_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
157 CTR_FLAG_DAEMON_SLEEP | \
158 CTR_FLAG_MIN_RECOVERY_RATE | \
159 CTR_FLAG_MAX_RECOVERY_RATE | \
160 CTR_FLAG_REGION_SIZE | \
161 CTR_FLAG_RAID10_COPIES | \
162 CTR_FLAG_RAID10_FORMAT | \
163 CTR_FLAG_DELTA_DISKS | \
164 CTR_FLAG_DATA_OFFSET | \
165 CTR_FLAG_RAID10_USE_NEAR_SETS)
168 * "raid4/5/6" do not accept any raid1 or raid10 specific options
170 * "raid6" does not accept "nosync", because it is not guaranteed
171 * that both parity and q-syndrome are being written properly with
174 #define RAID45_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
176 CTR_FLAG_DAEMON_SLEEP | \
177 CTR_FLAG_MIN_RECOVERY_RATE | \
178 CTR_FLAG_MAX_RECOVERY_RATE | \
179 CTR_FLAG_STRIPE_CACHE | \
180 CTR_FLAG_REGION_SIZE | \
181 CTR_FLAG_DELTA_DISKS | \
182 CTR_FLAG_DATA_OFFSET | \
183 CTR_FLAG_JOURNAL_DEV | \
184 CTR_FLAG_JOURNAL_MODE)
186 #define RAID6_VALID_FLAGS (CTR_FLAG_SYNC | \
188 CTR_FLAG_DAEMON_SLEEP | \
189 CTR_FLAG_MIN_RECOVERY_RATE | \
190 CTR_FLAG_MAX_RECOVERY_RATE | \
191 CTR_FLAG_STRIPE_CACHE | \
192 CTR_FLAG_REGION_SIZE | \
193 CTR_FLAG_DELTA_DISKS | \
194 CTR_FLAG_DATA_OFFSET | \
195 CTR_FLAG_JOURNAL_DEV | \
196 CTR_FLAG_JOURNAL_MODE)
197 /* ...valid options definitions per raid level */
200 * Flags for rs->runtime_flags field
201 * (RT_FLAG prefix meaning "runtime flag")
203 * These are all internal and used to define runtime state,
204 * e.g. to prevent another resume from preresume processing
205 * the raid set all over again.
207 #define RT_FLAG_RS_PRERESUMED 0
208 #define RT_FLAG_RS_RESUMED 1
209 #define RT_FLAG_RS_BITMAP_LOADED 2
210 #define RT_FLAG_UPDATE_SBS 3
211 #define RT_FLAG_RESHAPE_RS 4
212 #define RT_FLAG_RS_SUSPENDED 5
213 #define RT_FLAG_RS_IN_SYNC 6
214 #define RT_FLAG_RS_RESYNCING 7
215 #define RT_FLAG_RS_GROW 8
216 #define RT_FLAG_RS_FROZEN 9
218 /* Array elements of 64 bit needed for rebuild/failed disk bits */
219 #define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
222 * raid set level, layout and chunk sectors backup/restore
227 int new_chunk_sectors;
231 struct dm_target *ti;
233 uint32_t stripe_cache_entries;
234 unsigned long ctr_flags;
235 unsigned long runtime_flags;
237 uint64_t rebuild_disks[DISKS_ARRAY_ELEMS];
243 int requested_bitmap_chunk_sectors;
246 struct raid_type *raid_type;
248 sector_t array_sectors;
249 sector_t dev_sectors;
251 /* Optional raid4/5/6 journal device */
258 struct raid_dev dev[] __counted_by(raid_disks);
261 static void rs_config_backup(struct raid_set *rs, struct rs_layout *l)
263 struct mddev *mddev = &rs->md;
265 l->new_level = mddev->new_level;
266 l->new_layout = mddev->new_layout;
267 l->new_chunk_sectors = mddev->new_chunk_sectors;
270 static void rs_config_restore(struct raid_set *rs, struct rs_layout *l)
272 struct mddev *mddev = &rs->md;
274 mddev->new_level = l->new_level;
275 mddev->new_layout = l->new_layout;
276 mddev->new_chunk_sectors = l->new_chunk_sectors;
279 /* raid10 algorithms (i.e. formats) */
280 #define ALGORITHM_RAID10_DEFAULT 0
281 #define ALGORITHM_RAID10_NEAR 1
282 #define ALGORITHM_RAID10_OFFSET 2
283 #define ALGORITHM_RAID10_FAR 3
285 /* Supported raid types and properties. */
286 static struct raid_type {
287 const char *name; /* RAID algorithm. */
288 const char *descr; /* Descriptor text for logging. */
289 const unsigned int parity_devs; /* # of parity devices. */
290 const unsigned int minimal_devs;/* minimal # of devices in set. */
291 const unsigned int level; /* RAID level. */
292 const unsigned int algorithm; /* RAID algorithm. */
294 {"raid0", "raid0 (striping)", 0, 2, 0, 0 /* NONE */},
295 {"raid1", "raid1 (mirroring)", 0, 2, 1, 0 /* NONE */},
296 {"raid10_far", "raid10 far (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_FAR},
297 {"raid10_offset", "raid10 offset (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_OFFSET},
298 {"raid10_near", "raid10 near (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_NEAR},
299 {"raid10", "raid10 (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_DEFAULT},
300 {"raid4", "raid4 (dedicated first parity disk)", 1, 2, 5, ALGORITHM_PARITY_0}, /* raid4 layout = raid5_0 */
301 {"raid5_n", "raid5 (dedicated last parity disk)", 1, 2, 5, ALGORITHM_PARITY_N},
302 {"raid5_ls", "raid5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
303 {"raid5_rs", "raid5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
304 {"raid5_la", "raid5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
305 {"raid5_ra", "raid5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
306 {"raid6_zr", "raid6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
307 {"raid6_nr", "raid6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
308 {"raid6_nc", "raid6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE},
309 {"raid6_n_6", "raid6 (dedicated parity/Q n/6)", 2, 4, 6, ALGORITHM_PARITY_N_6},
310 {"raid6_ls_6", "raid6 (left symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_SYMMETRIC_6},
311 {"raid6_rs_6", "raid6 (right symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_SYMMETRIC_6},
312 {"raid6_la_6", "raid6 (left asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_ASYMMETRIC_6},
313 {"raid6_ra_6", "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_ASYMMETRIC_6}
316 /* True, if @v is in inclusive range [@min, @max] */
317 static bool __within_range(long v, long min, long max)
319 return v >= min && v <= max;
322 /* All table line arguments are defined here */
323 static struct arg_name_flag {
324 const unsigned long flag;
326 } __arg_name_flags[] = {
327 { CTR_FLAG_SYNC, "sync"},
328 { CTR_FLAG_NOSYNC, "nosync"},
329 { CTR_FLAG_REBUILD, "rebuild"},
330 { CTR_FLAG_DAEMON_SLEEP, "daemon_sleep"},
331 { CTR_FLAG_MIN_RECOVERY_RATE, "min_recovery_rate"},
332 { CTR_FLAG_MAX_RECOVERY_RATE, "max_recovery_rate"},
333 { CTR_FLAG_MAX_WRITE_BEHIND, "max_write_behind"},
334 { CTR_FLAG_WRITE_MOSTLY, "write_mostly"},
335 { CTR_FLAG_STRIPE_CACHE, "stripe_cache"},
336 { CTR_FLAG_REGION_SIZE, "region_size"},
337 { CTR_FLAG_RAID10_COPIES, "raid10_copies"},
338 { CTR_FLAG_RAID10_FORMAT, "raid10_format"},
339 { CTR_FLAG_DATA_OFFSET, "data_offset"},
340 { CTR_FLAG_DELTA_DISKS, "delta_disks"},
341 { CTR_FLAG_RAID10_USE_NEAR_SETS, "raid10_use_near_sets"},
342 { CTR_FLAG_JOURNAL_DEV, "journal_dev" },
343 { CTR_FLAG_JOURNAL_MODE, "journal_mode" },
346 /* Return argument name string for given @flag */
347 static const char *dm_raid_arg_name_by_flag(const uint32_t flag)
349 if (hweight32(flag) == 1) {
350 struct arg_name_flag *anf = __arg_name_flags + ARRAY_SIZE(__arg_name_flags);
352 while (anf-- > __arg_name_flags)
353 if (flag & anf->flag)
357 DMERR("%s called with more than one flag!", __func__);
362 /* Define correlation of raid456 journal cache modes and dm-raid target line parameters */
366 } _raid456_journal_mode[] = {
367 { R5C_JOURNAL_MODE_WRITE_THROUGH, "writethrough" },
368 { R5C_JOURNAL_MODE_WRITE_BACK, "writeback" }
371 /* Return MD raid4/5/6 journal mode for dm @journal_mode one */
372 static int dm_raid_journal_mode_to_md(const char *mode)
374 int m = ARRAY_SIZE(_raid456_journal_mode);
377 if (!strcasecmp(mode, _raid456_journal_mode[m].param))
378 return _raid456_journal_mode[m].mode;
383 /* Return dm-raid raid4/5/6 journal mode string for @mode */
384 static const char *md_journal_mode_to_dm_raid(const int mode)
386 int m = ARRAY_SIZE(_raid456_journal_mode);
389 if (mode == _raid456_journal_mode[m].mode)
390 return _raid456_journal_mode[m].param;
396 * Bool helpers to test for various raid levels of a raid set.
397 * It's level as reported by the superblock rather than
398 * the requested raid_type passed to the constructor.
400 /* Return true, if raid set in @rs is raid0 */
401 static bool rs_is_raid0(struct raid_set *rs)
403 return !rs->md.level;
406 /* Return true, if raid set in @rs is raid1 */
407 static bool rs_is_raid1(struct raid_set *rs)
409 return rs->md.level == 1;
412 /* Return true, if raid set in @rs is raid10 */
413 static bool rs_is_raid10(struct raid_set *rs)
415 return rs->md.level == 10;
418 /* Return true, if raid set in @rs is level 6 */
419 static bool rs_is_raid6(struct raid_set *rs)
421 return rs->md.level == 6;
424 /* Return true, if raid set in @rs is level 4, 5 or 6 */
425 static bool rs_is_raid456(struct raid_set *rs)
427 return __within_range(rs->md.level, 4, 6);
430 /* Return true, if raid set in @rs is reshapable */
431 static bool __is_raid10_far(int layout);
432 static bool rs_is_reshapable(struct raid_set *rs)
434 return rs_is_raid456(rs) ||
435 (rs_is_raid10(rs) && !__is_raid10_far(rs->md.new_layout));
438 /* Return true, if raid set in @rs is recovering */
439 static bool rs_is_recovering(struct raid_set *rs)
441 return rs->md.recovery_cp < rs->md.dev_sectors;
444 /* Return true, if raid set in @rs is reshaping */
445 static bool rs_is_reshaping(struct raid_set *rs)
447 return rs->md.reshape_position != MaxSector;
451 * bool helpers to test for various raid levels of a raid type @rt
454 /* Return true, if raid type in @rt is raid0 */
455 static bool rt_is_raid0(struct raid_type *rt)
460 /* Return true, if raid type in @rt is raid1 */
461 static bool rt_is_raid1(struct raid_type *rt)
463 return rt->level == 1;
466 /* Return true, if raid type in @rt is raid10 */
467 static bool rt_is_raid10(struct raid_type *rt)
469 return rt->level == 10;
472 /* Return true, if raid type in @rt is raid4/5 */
473 static bool rt_is_raid45(struct raid_type *rt)
475 return __within_range(rt->level, 4, 5);
478 /* Return true, if raid type in @rt is raid6 */
479 static bool rt_is_raid6(struct raid_type *rt)
481 return rt->level == 6;
484 /* Return true, if raid type in @rt is raid4/5/6 */
485 static bool rt_is_raid456(struct raid_type *rt)
487 return __within_range(rt->level, 4, 6);
489 /* END: raid level bools */
491 /* Return valid ctr flags for the raid level of @rs */
492 static unsigned long __valid_flags(struct raid_set *rs)
494 if (rt_is_raid0(rs->raid_type))
495 return RAID0_VALID_FLAGS;
496 else if (rt_is_raid1(rs->raid_type))
497 return RAID1_VALID_FLAGS;
498 else if (rt_is_raid10(rs->raid_type))
499 return RAID10_VALID_FLAGS;
500 else if (rt_is_raid45(rs->raid_type))
501 return RAID45_VALID_FLAGS;
502 else if (rt_is_raid6(rs->raid_type))
503 return RAID6_VALID_FLAGS;
509 * Check for valid flags set on @rs
511 * Has to be called after parsing of the ctr flags!
513 static int rs_check_for_valid_flags(struct raid_set *rs)
515 if (rs->ctr_flags & ~__valid_flags(rs)) {
516 rs->ti->error = "Invalid flags combination";
523 /* MD raid10 bit definitions and helpers */
524 #define RAID10_OFFSET (1 << 16) /* stripes with data copies area adjacent on devices */
525 #define RAID10_BROCKEN_USE_FAR_SETS (1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
526 #define RAID10_USE_FAR_SETS (1 << 18) /* Use sets instead of whole stripe rotation */
527 #define RAID10_FAR_COPIES_SHIFT 8 /* raid10 # far copies shift (2nd byte of layout) */
529 /* Return md raid10 near copies for @layout */
530 static unsigned int __raid10_near_copies(int layout)
532 return layout & 0xFF;
535 /* Return md raid10 far copies for @layout */
536 static unsigned int __raid10_far_copies(int layout)
538 return __raid10_near_copies(layout >> RAID10_FAR_COPIES_SHIFT);
541 /* Return true if md raid10 offset for @layout */
542 static bool __is_raid10_offset(int layout)
544 return !!(layout & RAID10_OFFSET);
547 /* Return true if md raid10 near for @layout */
548 static bool __is_raid10_near(int layout)
550 return !__is_raid10_offset(layout) && __raid10_near_copies(layout) > 1;
553 /* Return true if md raid10 far for @layout */
554 static bool __is_raid10_far(int layout)
556 return !__is_raid10_offset(layout) && __raid10_far_copies(layout) > 1;
559 /* Return md raid10 layout string for @layout */
560 static const char *raid10_md_layout_to_format(int layout)
563 * Bit 16 stands for "offset"
564 * (i.e. adjacent stripes hold copies)
566 * Refer to MD's raid10.c for details
568 if (__is_raid10_offset(layout))
571 if (__raid10_near_copies(layout) > 1)
574 if (__raid10_far_copies(layout) > 1)
580 /* Return md raid10 algorithm for @name */
581 static int raid10_name_to_format(const char *name)
583 if (!strcasecmp(name, "near"))
584 return ALGORITHM_RAID10_NEAR;
585 else if (!strcasecmp(name, "offset"))
586 return ALGORITHM_RAID10_OFFSET;
587 else if (!strcasecmp(name, "far"))
588 return ALGORITHM_RAID10_FAR;
593 /* Return md raid10 copies for @layout */
594 static unsigned int raid10_md_layout_to_copies(int layout)
596 return max(__raid10_near_copies(layout), __raid10_far_copies(layout));
599 /* Return md raid10 format id for @format string */
600 static int raid10_format_to_md_layout(struct raid_set *rs,
601 unsigned int algorithm,
604 unsigned int n = 1, f = 1, r = 0;
607 * MD resilienece flaw:
609 * enabling use_far_sets for far/offset formats causes copies
610 * to be colocated on the same devs together with their origins!
612 * -> disable it for now in the definition above
614 if (algorithm == ALGORITHM_RAID10_DEFAULT ||
615 algorithm == ALGORITHM_RAID10_NEAR)
618 else if (algorithm == ALGORITHM_RAID10_OFFSET) {
621 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
622 r |= RAID10_USE_FAR_SETS;
624 } else if (algorithm == ALGORITHM_RAID10_FAR) {
626 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
627 r |= RAID10_USE_FAR_SETS;
632 return r | (f << RAID10_FAR_COPIES_SHIFT) | n;
634 /* END: MD raid10 bit definitions and helpers */
636 /* Check for any of the raid10 algorithms */
637 static bool __got_raid10(struct raid_type *rtp, const int layout)
639 if (rtp->level == 10) {
640 switch (rtp->algorithm) {
641 case ALGORITHM_RAID10_DEFAULT:
642 case ALGORITHM_RAID10_NEAR:
643 return __is_raid10_near(layout);
644 case ALGORITHM_RAID10_OFFSET:
645 return __is_raid10_offset(layout);
646 case ALGORITHM_RAID10_FAR:
647 return __is_raid10_far(layout);
656 /* Return raid_type for @name */
657 static struct raid_type *get_raid_type(const char *name)
659 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
661 while (rtp-- > raid_types)
662 if (!strcasecmp(rtp->name, name))
668 /* Return raid_type for @name based derived from @level and @layout */
669 static struct raid_type *get_raid_type_by_ll(const int level, const int layout)
671 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
673 while (rtp-- > raid_types) {
674 /* RAID10 special checks based on @layout flags/properties */
675 if (rtp->level == level &&
676 (__got_raid10(rtp, layout) || rtp->algorithm == layout))
683 /* Adjust rdev sectors */
684 static void rs_set_rdev_sectors(struct raid_set *rs)
686 struct mddev *mddev = &rs->md;
687 struct md_rdev *rdev;
690 * raid10 sets rdev->sector to the device size, which
691 * is unintended in case of out-of-place reshaping
693 rdev_for_each(rdev, mddev)
694 if (!test_bit(Journal, &rdev->flags))
695 rdev->sectors = mddev->dev_sectors;
699 * Change bdev capacity of @rs in case of a disk add/remove reshape
701 static void rs_set_capacity(struct raid_set *rs)
703 struct gendisk *gendisk = dm_disk(dm_table_get_md(rs->ti->table));
705 set_capacity_and_notify(gendisk, rs->md.array_sectors);
709 * Set the mddev properties in @rs to the current
710 * ones retrieved from the freshest superblock
712 static void rs_set_cur(struct raid_set *rs)
714 struct mddev *mddev = &rs->md;
716 mddev->new_level = mddev->level;
717 mddev->new_layout = mddev->layout;
718 mddev->new_chunk_sectors = mddev->chunk_sectors;
722 * Set the mddev properties in @rs to the new
723 * ones requested by the ctr
725 static void rs_set_new(struct raid_set *rs)
727 struct mddev *mddev = &rs->md;
729 mddev->level = mddev->new_level;
730 mddev->layout = mddev->new_layout;
731 mddev->chunk_sectors = mddev->new_chunk_sectors;
732 mddev->raid_disks = rs->raid_disks;
733 mddev->delta_disks = 0;
736 static struct raid_set *raid_set_alloc(struct dm_target *ti, struct raid_type *raid_type,
737 unsigned int raid_devs)
742 if (raid_devs <= raid_type->parity_devs) {
743 ti->error = "Insufficient number of devices";
744 return ERR_PTR(-EINVAL);
747 rs = kzalloc(struct_size(rs, dev, raid_devs), GFP_KERNEL);
749 ti->error = "Cannot allocate raid context";
750 return ERR_PTR(-ENOMEM);
753 if (mddev_init(&rs->md)) {
755 ti->error = "Cannot initialize raid context";
756 return ERR_PTR(-ENOMEM);
759 rs->raid_disks = raid_devs;
763 rs->raid_type = raid_type;
764 rs->stripe_cache_entries = 256;
765 rs->md.raid_disks = raid_devs;
766 rs->md.level = raid_type->level;
767 rs->md.new_level = rs->md.level;
768 rs->md.layout = raid_type->algorithm;
769 rs->md.new_layout = rs->md.layout;
770 rs->md.delta_disks = 0;
771 rs->md.recovery_cp = MaxSector;
773 for (i = 0; i < raid_devs; i++)
774 md_rdev_init(&rs->dev[i].rdev);
777 * Remaining items to be initialized by further RAID params:
780 * rs->md.chunk_sectors
781 * rs->md.new_chunk_sectors
788 /* Free all @rs allocations */
789 static void raid_set_free(struct raid_set *rs)
793 if (rs->journal_dev.dev) {
794 md_rdev_clear(&rs->journal_dev.rdev);
795 dm_put_device(rs->ti, rs->journal_dev.dev);
798 for (i = 0; i < rs->raid_disks; i++) {
799 if (rs->dev[i].meta_dev)
800 dm_put_device(rs->ti, rs->dev[i].meta_dev);
801 md_rdev_clear(&rs->dev[i].rdev);
802 if (rs->dev[i].data_dev)
803 dm_put_device(rs->ti, rs->dev[i].data_dev);
806 mddev_destroy(&rs->md);
811 * For every device we have two words
812 * <meta_dev>: meta device name or '-' if missing
813 * <data_dev>: data device name or '-' if missing
815 * The following are permitted:
818 * <meta_dev> <data_dev>
820 * The following is not allowed:
823 * This code parses those words. If there is a failure,
824 * the caller must use raid_set_free() to unwind the operations.
826 static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
830 int metadata_available = 0;
834 /* Put off the number of raid devices argument to get to dev pairs */
835 arg = dm_shift_arg(as);
839 for (i = 0; i < rs->raid_disks; i++) {
840 rs->dev[i].rdev.raid_disk = i;
842 rs->dev[i].meta_dev = NULL;
843 rs->dev[i].data_dev = NULL;
846 * There are no offsets initially.
847 * Out of place reshape will set them accordingly.
849 rs->dev[i].rdev.data_offset = 0;
850 rs->dev[i].rdev.new_data_offset = 0;
851 rs->dev[i].rdev.mddev = &rs->md;
853 arg = dm_shift_arg(as);
857 if (strcmp(arg, "-")) {
858 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
859 &rs->dev[i].meta_dev);
861 rs->ti->error = "RAID metadata device lookup failure";
865 rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
866 if (!rs->dev[i].rdev.sb_page) {
867 rs->ti->error = "Failed to allocate superblock page";
872 arg = dm_shift_arg(as);
876 if (!strcmp(arg, "-")) {
877 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
878 (!rs->dev[i].rdev.recovery_offset)) {
879 rs->ti->error = "Drive designated for rebuild not specified";
883 if (rs->dev[i].meta_dev) {
884 rs->ti->error = "No data device supplied with metadata device";
891 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
892 &rs->dev[i].data_dev);
894 rs->ti->error = "RAID device lookup failure";
898 if (rs->dev[i].meta_dev) {
899 metadata_available = 1;
900 rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
902 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
903 list_add_tail(&rs->dev[i].rdev.same_set, &rs->md.disks);
904 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
908 if (rs->journal_dev.dev)
909 list_add_tail(&rs->journal_dev.rdev.same_set, &rs->md.disks);
911 if (metadata_available) {
913 rs->md.persistent = 1;
914 rs->md.major_version = 2;
915 } else if (rebuild && !rs->md.recovery_cp) {
917 * Without metadata, we will not be able to tell if the array
918 * is in-sync or not - we must assume it is not. Therefore,
919 * it is impossible to rebuild a drive.
921 * Even if there is metadata, the on-disk information may
922 * indicate that the array is not in-sync and it will then
925 * User could specify 'nosync' option if desperate.
927 rs->ti->error = "Unable to rebuild drive while array is not in-sync";
935 * validate_region_size
937 * @region_size: region size in sectors. If 0, pick a size (4MiB default).
939 * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
940 * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
942 * Returns: 0 on success, -EINVAL on failure.
944 static int validate_region_size(struct raid_set *rs, unsigned long region_size)
946 unsigned long min_region_size = rs->ti->len / (1 << 21);
953 * Choose a reasonable default. All figures in sectors.
955 if (min_region_size > (1 << 13)) {
956 /* If not a power of 2, make it the next power of 2 */
957 region_size = roundup_pow_of_two(min_region_size);
958 DMINFO("Choosing default region size of %lu sectors",
961 DMINFO("Choosing default region size of 4MiB");
962 region_size = 1 << 13; /* sectors */
966 * Validate user-supplied value.
968 if (region_size > rs->ti->len) {
969 rs->ti->error = "Supplied region size is too large";
973 if (region_size < min_region_size) {
974 DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
975 region_size, min_region_size);
976 rs->ti->error = "Supplied region size is too small";
980 if (!is_power_of_2(region_size)) {
981 rs->ti->error = "Region size is not a power of 2";
985 if (region_size < rs->md.chunk_sectors) {
986 rs->ti->error = "Region size is smaller than the chunk size";
992 * Convert sectors to bytes.
994 rs->md.bitmap_info.chunksize = to_bytes(region_size);
1000 * validate_raid_redundancy
1003 * Determine if there are enough devices in the array that haven't
1004 * failed (or are being rebuilt) to form a usable array.
1006 * Returns: 0 on success, -EINVAL on failure.
1008 static int validate_raid_redundancy(struct raid_set *rs)
1010 unsigned int i, rebuild_cnt = 0;
1011 unsigned int rebuilds_per_group = 0, copies, raid_disks;
1012 unsigned int group_size, last_group_start;
1014 for (i = 0; i < rs->raid_disks; i++)
1015 if (!test_bit(FirstUse, &rs->dev[i].rdev.flags) &&
1016 ((!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
1017 !rs->dev[i].rdev.sb_page)))
1020 switch (rs->md.level) {
1024 if (rebuild_cnt >= rs->md.raid_disks)
1030 if (rebuild_cnt > rs->raid_type->parity_devs)
1034 copies = raid10_md_layout_to_copies(rs->md.new_layout);
1036 DMERR("Bogus raid10 data copies < 2!");
1040 if (rebuild_cnt < copies)
1044 * It is possible to have a higher rebuild count for RAID10,
1045 * as long as the failed devices occur in different mirror
1046 * groups (i.e. different stripes).
1048 * When checking "near" format, make sure no adjacent devices
1049 * have failed beyond what can be handled. In addition to the
1050 * simple case where the number of devices is a multiple of the
1051 * number of copies, we must also handle cases where the number
1052 * of devices is not a multiple of the number of copies.
1053 * E.g. dev1 dev2 dev3 dev4 dev5
1057 raid_disks = min(rs->raid_disks, rs->md.raid_disks);
1058 if (__is_raid10_near(rs->md.new_layout)) {
1059 for (i = 0; i < raid_disks; i++) {
1061 rebuilds_per_group = 0;
1062 if ((!rs->dev[i].rdev.sb_page ||
1063 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1064 (++rebuilds_per_group >= copies))
1071 * When checking "far" and "offset" formats, we need to ensure
1072 * that the device that holds its copy is not also dead or
1073 * being rebuilt. (Note that "far" and "offset" formats only
1074 * support two copies right now. These formats also only ever
1075 * use the 'use_far_sets' variant.)
1077 * This check is somewhat complicated by the need to account
1078 * for arrays that are not a multiple of (far) copies. This
1079 * results in the need to treat the last (potentially larger)
1082 group_size = (raid_disks / copies);
1083 last_group_start = (raid_disks / group_size) - 1;
1084 last_group_start *= group_size;
1085 for (i = 0; i < raid_disks; i++) {
1086 if (!(i % copies) && !(i > last_group_start))
1087 rebuilds_per_group = 0;
1088 if ((!rs->dev[i].rdev.sb_page ||
1089 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1090 (++rebuilds_per_group >= copies))
1106 * Possible arguments are...
1107 * <chunk_size> [optional_args]
1109 * Argument definitions
1110 * <chunk_size> The number of sectors per disk that
1111 * will form the "stripe"
1112 * [[no]sync] Force or prevent recovery of the
1114 * [rebuild <idx>] Rebuild the drive indicated by the index
1115 * [daemon_sleep <ms>] Time between bitmap daemon work to
1117 * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1118 * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1119 * [write_mostly <idx>] Indicate a write mostly drive via index
1120 * [max_write_behind <sectors>] See '-write-behind=' (man mdadm)
1121 * [stripe_cache <sectors>] Stripe cache size for higher RAIDs
1122 * [region_size <sectors>] Defines granularity of bitmap
1123 * [journal_dev <dev>] raid4/5/6 journaling deviice
1124 * (i.e. write hole closing log)
1126 * RAID10-only options:
1127 * [raid10_copies <# copies>] Number of copies. (Default: 2)
1128 * [raid10_format <near|far|offset>] Layout algorithm. (Default: near)
1130 static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
1131 unsigned int num_raid_params)
1133 int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
1134 unsigned int raid10_copies = 2;
1135 unsigned int i, write_mostly = 0;
1136 unsigned int region_size = 0;
1137 sector_t max_io_len;
1138 const char *arg, *key;
1139 struct raid_dev *rd;
1140 struct raid_type *rt = rs->raid_type;
1142 arg = dm_shift_arg(as);
1143 num_raid_params--; /* Account for chunk_size argument */
1145 if (kstrtoint(arg, 10, &value) < 0) {
1146 rs->ti->error = "Bad numerical argument given for chunk_size";
1151 * First, parse the in-order required arguments
1152 * "chunk_size" is the only argument of this type.
1154 if (rt_is_raid1(rt)) {
1156 DMERR("Ignoring chunk size parameter for RAID 1");
1158 } else if (!is_power_of_2(value)) {
1159 rs->ti->error = "Chunk size must be a power of 2";
1161 } else if (value < 8) {
1162 rs->ti->error = "Chunk size value is too small";
1166 rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
1169 * We set each individual device as In_sync with a completed
1170 * 'recovery_offset'. If there has been a device failure or
1171 * replacement then one of the following cases applies:
1173 * 1) User specifies 'rebuild'.
1174 * - Device is reset when param is read.
1175 * 2) A new device is supplied.
1176 * - No matching superblock found, resets device.
1177 * 3) Device failure was transient and returns on reload.
1178 * - Failure noticed, resets device for bitmap replay.
1179 * 4) Device hadn't completed recovery after previous failure.
1180 * - Superblock is read and overrides recovery_offset.
1182 * What is found in the superblocks of the devices is always
1183 * authoritative, unless 'rebuild' or '[no]sync' was specified.
1185 for (i = 0; i < rs->raid_disks; i++) {
1186 set_bit(In_sync, &rs->dev[i].rdev.flags);
1187 rs->dev[i].rdev.recovery_offset = MaxSector;
1191 * Second, parse the unordered optional arguments
1193 for (i = 0; i < num_raid_params; i++) {
1194 key = dm_shift_arg(as);
1196 rs->ti->error = "Not enough raid parameters given";
1200 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
1201 if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1202 rs->ti->error = "Only one 'nosync' argument allowed";
1207 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
1208 if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
1209 rs->ti->error = "Only one 'sync' argument allowed";
1214 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
1215 if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1216 rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1222 arg = dm_shift_arg(as);
1223 i++; /* Account for the argument pairs */
1225 rs->ti->error = "Wrong number of raid parameters given";
1230 * Parameters that take a string value are checked here.
1232 /* "raid10_format {near|offset|far} */
1233 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
1234 if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
1235 rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1238 if (!rt_is_raid10(rt)) {
1239 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1242 raid10_format = raid10_name_to_format(arg);
1243 if (raid10_format < 0) {
1244 rs->ti->error = "Invalid 'raid10_format' value given";
1245 return raid10_format;
1250 /* "journal_dev <dev>" */
1251 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV))) {
1253 struct md_rdev *jdev;
1255 if (test_and_set_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1256 rs->ti->error = "Only one raid4/5/6 set journaling device allowed";
1259 if (!rt_is_raid456(rt)) {
1260 rs->ti->error = "'journal_dev' is an invalid parameter for this RAID type";
1263 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
1264 &rs->journal_dev.dev);
1266 rs->ti->error = "raid4/5/6 journal device lookup failure";
1269 jdev = &rs->journal_dev.rdev;
1271 jdev->mddev = &rs->md;
1272 jdev->bdev = rs->journal_dev.dev->bdev;
1273 jdev->sectors = bdev_nr_sectors(jdev->bdev);
1274 if (jdev->sectors < MIN_RAID456_JOURNAL_SPACE) {
1275 rs->ti->error = "No space for raid4/5/6 journal";
1278 rs->journal_dev.mode = R5C_JOURNAL_MODE_WRITE_THROUGH;
1279 set_bit(Journal, &jdev->flags);
1283 /* "journal_mode <mode>" ("journal_dev" mandatory!) */
1284 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE))) {
1287 if (!test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1288 rs->ti->error = "raid4/5/6 'journal_mode' is invalid without 'journal_dev'";
1291 if (test_and_set_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
1292 rs->ti->error = "Only one raid4/5/6 'journal_mode' argument allowed";
1295 r = dm_raid_journal_mode_to_md(arg);
1297 rs->ti->error = "Invalid 'journal_mode' argument";
1300 rs->journal_dev.mode = r;
1305 * Parameters with number values from here on.
1307 if (kstrtoint(arg, 10, &value) < 0) {
1308 rs->ti->error = "Bad numerical argument given in raid params";
1312 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
1314 * "rebuild" is being passed in by userspace to provide
1315 * indexes of replaced devices and to set up additional
1316 * devices on raid level takeover.
1318 if (!__within_range(value, 0, rs->raid_disks - 1)) {
1319 rs->ti->error = "Invalid rebuild index given";
1323 if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1324 rs->ti->error = "rebuild for this index already given";
1328 rd = rs->dev + value;
1329 clear_bit(In_sync, &rd->rdev.flags);
1330 clear_bit(Faulty, &rd->rdev.flags);
1331 rd->rdev.recovery_offset = 0;
1332 set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
1333 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
1334 if (!rt_is_raid1(rt)) {
1335 rs->ti->error = "write_mostly option is only valid for RAID1";
1339 if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
1340 rs->ti->error = "Invalid write_mostly index given";
1345 set_bit(WriteMostly, &rs->dev[value].rdev.flags);
1346 set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
1347 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
1348 if (!rt_is_raid1(rt)) {
1349 rs->ti->error = "max_write_behind option is only valid for RAID1";
1353 if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
1354 rs->ti->error = "Only one max_write_behind argument pair allowed";
1359 * In device-mapper, we specify things in sectors, but
1360 * MD records this value in kB
1362 if (value < 0 || value / 2 > COUNTER_MAX) {
1363 rs->ti->error = "Max write-behind limit out of range";
1367 rs->md.bitmap_info.max_write_behind = value / 2;
1368 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
1369 if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
1370 rs->ti->error = "Only one daemon_sleep argument pair allowed";
1374 rs->ti->error = "daemon sleep period out of range";
1377 rs->md.bitmap_info.daemon_sleep = value;
1378 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
1379 /* Userspace passes new data_offset after having extended the data image LV */
1380 if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
1381 rs->ti->error = "Only one data_offset argument pair allowed";
1384 /* Ensure sensible data offset */
1386 (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
1387 rs->ti->error = "Bogus data_offset value";
1390 rs->data_offset = value;
1391 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
1392 /* Define the +/-# of disks to add to/remove from the given raid set */
1393 if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1394 rs->ti->error = "Only one delta_disks argument pair allowed";
1397 /* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1398 if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
1399 rs->ti->error = "Too many delta_disk requested";
1403 rs->delta_disks = value;
1404 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
1405 if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
1406 rs->ti->error = "Only one stripe_cache argument pair allowed";
1410 if (!rt_is_raid456(rt)) {
1411 rs->ti->error = "Inappropriate argument: stripe_cache";
1416 rs->ti->error = "Bogus stripe cache entries value";
1419 rs->stripe_cache_entries = value;
1420 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
1421 if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1422 rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1427 rs->ti->error = "min_recovery_rate out of range";
1430 rs->md.sync_speed_min = value;
1431 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
1432 if (test_and_set_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags)) {
1433 rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1438 rs->ti->error = "max_recovery_rate out of range";
1441 rs->md.sync_speed_max = value;
1442 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
1443 if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
1444 rs->ti->error = "Only one region_size argument pair allowed";
1448 region_size = value;
1449 rs->requested_bitmap_chunk_sectors = value;
1450 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
1451 if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
1452 rs->ti->error = "Only one raid10_copies argument pair allowed";
1456 if (!__within_range(value, 2, rs->md.raid_disks)) {
1457 rs->ti->error = "Bad value for 'raid10_copies'";
1461 raid10_copies = value;
1463 DMERR("Unable to parse RAID parameter: %s", key);
1464 rs->ti->error = "Unable to parse RAID parameter";
1469 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1470 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1471 rs->ti->error = "sync and nosync are mutually exclusive";
1475 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) &&
1476 (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ||
1477 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))) {
1478 rs->ti->error = "sync/nosync and rebuild are mutually exclusive";
1482 if (write_mostly >= rs->md.raid_disks) {
1483 rs->ti->error = "Can't set all raid1 devices to write_mostly";
1487 if (rs->md.sync_speed_max &&
1488 rs->md.sync_speed_min > rs->md.sync_speed_max) {
1489 rs->ti->error = "Bogus recovery rates";
1493 if (validate_region_size(rs, region_size))
1496 if (rs->md.chunk_sectors)
1497 max_io_len = rs->md.chunk_sectors;
1499 max_io_len = region_size;
1501 if (dm_set_target_max_io_len(rs->ti, max_io_len))
1504 if (rt_is_raid10(rt)) {
1505 if (raid10_copies > rs->md.raid_disks) {
1506 rs->ti->error = "Not enough devices to satisfy specification";
1510 rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
1511 if (rs->md.new_layout < 0) {
1512 rs->ti->error = "Error getting raid10 format";
1513 return rs->md.new_layout;
1516 rt = get_raid_type_by_ll(10, rs->md.new_layout);
1518 rs->ti->error = "Failed to recognize new raid10 layout";
1522 if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1523 rt->algorithm == ALGORITHM_RAID10_NEAR) &&
1524 test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1525 rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1530 rs->raid10_copies = raid10_copies;
1532 /* Assume there are no metadata devices until the drives are parsed */
1533 rs->md.persistent = 0;
1534 rs->md.external = 1;
1536 /* Check, if any invalid ctr arguments have been passed in for the raid level */
1537 return rs_check_for_valid_flags(rs);
1540 /* Set raid4/5/6 cache size */
1541 static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1544 struct r5conf *conf;
1545 struct mddev *mddev = &rs->md;
1546 uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1547 uint32_t nr_stripes = rs->stripe_cache_entries;
1549 if (!rt_is_raid456(rs->raid_type)) {
1550 rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1554 if (nr_stripes < min_stripes) {
1555 DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1556 nr_stripes, min_stripes);
1557 nr_stripes = min_stripes;
1560 conf = mddev->private;
1562 rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1566 /* Try setting number of stripes in raid456 stripe cache */
1567 if (conf->min_nr_stripes != nr_stripes) {
1568 r = raid5_set_cache_size(mddev, nr_stripes);
1570 rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1574 DMINFO("%u stripe cache entries", nr_stripes);
1580 /* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1581 static unsigned int mddev_data_stripes(struct raid_set *rs)
1583 return rs->md.raid_disks - rs->raid_type->parity_devs;
1586 /* Return # of data stripes of @rs (i.e. as of ctr) */
1587 static unsigned int rs_data_stripes(struct raid_set *rs)
1589 return rs->raid_disks - rs->raid_type->parity_devs;
1593 * Retrieve rdev->sectors from any valid raid device of @rs
1594 * to allow userpace to pass in arbitray "- -" device tupples.
1596 static sector_t __rdev_sectors(struct raid_set *rs)
1600 for (i = 0; i < rs->raid_disks; i++) {
1601 struct md_rdev *rdev = &rs->dev[i].rdev;
1603 if (!test_bit(Journal, &rdev->flags) &&
1604 rdev->bdev && rdev->sectors)
1605 return rdev->sectors;
1611 /* Check that calculated dev_sectors fits all component devices. */
1612 static int _check_data_dev_sectors(struct raid_set *rs)
1615 struct md_rdev *rdev;
1617 rdev_for_each(rdev, &rs->md)
1618 if (!test_bit(Journal, &rdev->flags) && rdev->bdev) {
1619 ds = min(ds, bdev_nr_sectors(rdev->bdev));
1620 if (ds < rs->md.dev_sectors) {
1621 rs->ti->error = "Component device(s) too small";
1629 /* Calculate the sectors per device and per array used for @rs */
1630 static int rs_set_dev_and_array_sectors(struct raid_set *rs, sector_t sectors, bool use_mddev)
1633 unsigned int data_stripes;
1634 sector_t array_sectors = sectors, dev_sectors = sectors;
1635 struct mddev *mddev = &rs->md;
1638 delta_disks = mddev->delta_disks;
1639 data_stripes = mddev_data_stripes(rs);
1641 delta_disks = rs->delta_disks;
1642 data_stripes = rs_data_stripes(rs);
1645 /* Special raid1 case w/o delta_disks support (yet) */
1646 if (rt_is_raid1(rs->raid_type))
1648 else if (rt_is_raid10(rs->raid_type)) {
1649 if (rs->raid10_copies < 2 ||
1651 rs->ti->error = "Bogus raid10 data copies or delta disks";
1655 dev_sectors *= rs->raid10_copies;
1656 if (sector_div(dev_sectors, data_stripes))
1659 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1660 if (sector_div(array_sectors, rs->raid10_copies))
1663 } else if (sector_div(dev_sectors, data_stripes))
1667 /* Striped layouts */
1668 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1670 mddev->array_sectors = array_sectors;
1671 mddev->dev_sectors = dev_sectors;
1672 rs_set_rdev_sectors(rs);
1674 return _check_data_dev_sectors(rs);
1676 rs->ti->error = "Target length not divisible by number of data devices";
1680 /* Setup recovery on @rs */
1681 static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1683 /* raid0 does not recover */
1684 if (rs_is_raid0(rs))
1685 rs->md.recovery_cp = MaxSector;
1687 * A raid6 set has to be recovered either
1688 * completely or for the grown part to
1689 * ensure proper parity and Q-Syndrome
1691 else if (rs_is_raid6(rs))
1692 rs->md.recovery_cp = dev_sectors;
1694 * Other raid set types may skip recovery
1695 * depending on the 'nosync' flag.
1698 rs->md.recovery_cp = test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)
1699 ? MaxSector : dev_sectors;
1702 static void do_table_event(struct work_struct *ws)
1704 struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1706 smp_rmb(); /* Make sure we access most actual mddev properties */
1707 if (!rs_is_reshaping(rs)) {
1708 if (rs_is_raid10(rs))
1709 rs_set_rdev_sectors(rs);
1710 rs_set_capacity(rs);
1712 dm_table_event(rs->ti->table);
1716 * Make sure a valid takover (level switch) is being requested on @rs
1718 * Conversions of raid sets from one MD personality to another
1719 * have to conform to restrictions which are enforced here.
1721 static int rs_check_takeover(struct raid_set *rs)
1723 struct mddev *mddev = &rs->md;
1724 unsigned int near_copies;
1726 if (rs->md.degraded) {
1727 rs->ti->error = "Can't takeover degraded raid set";
1731 if (rs_is_reshaping(rs)) {
1732 rs->ti->error = "Can't takeover reshaping raid set";
1736 switch (mddev->level) {
1738 /* raid0 -> raid1/5 with one disk */
1739 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1740 mddev->raid_disks == 1)
1743 /* raid0 -> raid10 */
1744 if (mddev->new_level == 10 &&
1745 !(rs->raid_disks % mddev->raid_disks))
1748 /* raid0 with multiple disks -> raid4/5/6 */
1749 if (__within_range(mddev->new_level, 4, 6) &&
1750 mddev->new_layout == ALGORITHM_PARITY_N &&
1751 mddev->raid_disks > 1)
1757 /* Can't takeover raid10_offset! */
1758 if (__is_raid10_offset(mddev->layout))
1761 near_copies = __raid10_near_copies(mddev->layout);
1763 /* raid10* -> raid0 */
1764 if (mddev->new_level == 0) {
1765 /* Can takeover raid10_near with raid disks divisable by data copies! */
1766 if (near_copies > 1 &&
1767 !(mddev->raid_disks % near_copies)) {
1768 mddev->raid_disks /= near_copies;
1769 mddev->delta_disks = mddev->raid_disks;
1773 /* Can takeover raid10_far */
1774 if (near_copies == 1 &&
1775 __raid10_far_copies(mddev->layout) > 1)
1781 /* raid10_{near,far} -> raid1 */
1782 if (mddev->new_level == 1 &&
1783 max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
1786 /* raid10_{near,far} with 2 disks -> raid4/5 */
1787 if (__within_range(mddev->new_level, 4, 5) &&
1788 mddev->raid_disks == 2)
1793 /* raid1 with 2 disks -> raid4/5 */
1794 if (__within_range(mddev->new_level, 4, 5) &&
1795 mddev->raid_disks == 2) {
1796 mddev->degraded = 1;
1800 /* raid1 -> raid0 */
1801 if (mddev->new_level == 0 &&
1802 mddev->raid_disks == 1)
1805 /* raid1 -> raid10 */
1806 if (mddev->new_level == 10)
1811 /* raid4 -> raid0 */
1812 if (mddev->new_level == 0)
1815 /* raid4 -> raid1/5 with 2 disks */
1816 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1817 mddev->raid_disks == 2)
1820 /* raid4 -> raid5/6 with parity N */
1821 if (__within_range(mddev->new_level, 5, 6) &&
1822 mddev->layout == ALGORITHM_PARITY_N)
1827 /* raid5 with parity N -> raid0 */
1828 if (mddev->new_level == 0 &&
1829 mddev->layout == ALGORITHM_PARITY_N)
1832 /* raid5 with parity N -> raid4 */
1833 if (mddev->new_level == 4 &&
1834 mddev->layout == ALGORITHM_PARITY_N)
1837 /* raid5 with 2 disks -> raid1/4/10 */
1838 if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1839 mddev->raid_disks == 2)
1842 /* raid5_* -> raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
1843 if (mddev->new_level == 6 &&
1844 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1845 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
1850 /* raid6 with parity N -> raid0 */
1851 if (mddev->new_level == 0 &&
1852 mddev->layout == ALGORITHM_PARITY_N)
1855 /* raid6 with parity N -> raid4 */
1856 if (mddev->new_level == 4 &&
1857 mddev->layout == ALGORITHM_PARITY_N)
1860 /* raid6_*_n with Q-Syndrome N -> raid5_* */
1861 if (mddev->new_level == 5 &&
1862 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1863 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
1871 rs->ti->error = "takeover not possible";
1875 /* True if @rs requested to be taken over */
1876 static bool rs_takeover_requested(struct raid_set *rs)
1878 return rs->md.new_level != rs->md.level;
1881 /* True if layout is set to reshape. */
1882 static bool rs_is_layout_change(struct raid_set *rs, bool use_mddev)
1884 return (use_mddev ? rs->md.delta_disks : rs->delta_disks) ||
1885 rs->md.new_layout != rs->md.layout ||
1886 rs->md.new_chunk_sectors != rs->md.chunk_sectors;
1889 /* True if @rs is requested to reshape by ctr */
1890 static bool rs_reshape_requested(struct raid_set *rs)
1893 struct mddev *mddev = &rs->md;
1895 if (rs_takeover_requested(rs))
1898 if (rs_is_raid0(rs))
1901 change = rs_is_layout_change(rs, false);
1903 /* Historical case to support raid1 reshape without delta disks */
1904 if (rs_is_raid1(rs)) {
1905 if (rs->delta_disks)
1906 return !!rs->delta_disks;
1909 mddev->raid_disks != rs->raid_disks;
1912 if (rs_is_raid10(rs))
1914 !__is_raid10_far(mddev->new_layout) &&
1915 rs->delta_disks >= 0;
1921 #define FEATURE_FLAG_SUPPORTS_V190 0x1 /* Supports extended superblock */
1923 /* State flags for sb->flags */
1924 #define SB_FLAG_RESHAPE_ACTIVE 0x1
1925 #define SB_FLAG_RESHAPE_BACKWARDS 0x2
1928 * This structure is never routinely used by userspace, unlike md superblocks.
1929 * Devices with this superblock should only ever be accessed via device-mapper.
1931 #define DM_RAID_MAGIC 0x64526D44
1932 struct dm_raid_superblock {
1933 __le32 magic; /* "DmRd" */
1934 __le32 compat_features; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1936 __le32 num_devices; /* Number of devices in this raid set. (Max 64) */
1937 __le32 array_position; /* The position of this drive in the raid set */
1939 __le64 events; /* Incremented by md when superblock updated */
1940 __le64 failed_devices; /* Pre 1.9.0 part of bit field of devices to */
1941 /* indicate failures (see extension below) */
1944 * This offset tracks the progress of the repair or replacement of
1945 * an individual drive.
1947 __le64 disk_recovery_offset;
1950 * This offset tracks the progress of the initial raid set
1951 * synchronisation/parity calculation.
1953 __le64 array_resync_offset;
1956 * raid characteristics
1960 __le32 stripe_sectors;
1962 /********************************************************************
1963 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1965 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
1968 __le32 flags; /* Flags defining array states for reshaping */
1971 * This offset tracks the progress of a raid
1972 * set reshape in order to be able to restart it
1974 __le64 reshape_position;
1977 * These define the properties of the array in case of an interrupted reshape
1981 __le32 new_stripe_sectors;
1984 __le64 array_sectors; /* Array size in sectors */
1987 * Sector offsets to data on devices (reshaping).
1988 * Needed to support out of place reshaping, thus
1989 * not writing over any stripes whilst converting
1990 * them from old to new layout
1993 __le64 new_data_offset;
1995 __le64 sectors; /* Used device size in sectors */
1998 * Additional Bit field of devices indicating failures to support
1999 * up to 256 devices with the 1.9.0 on-disk metadata format
2001 __le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
2003 __le32 incompat_features; /* Used to indicate any incompatible features */
2005 /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
2009 * Check for reshape constraints on raid set @rs:
2011 * - reshape function non-existent
2013 * - ongoing recovery
2016 * Returns 0 if none or -EPERM if given constraint
2017 * and error message reference in @errmsg
2019 static int rs_check_reshape(struct raid_set *rs)
2021 struct mddev *mddev = &rs->md;
2023 if (!mddev->pers || !mddev->pers->check_reshape)
2024 rs->ti->error = "Reshape not supported";
2025 else if (mddev->degraded)
2026 rs->ti->error = "Can't reshape degraded raid set";
2027 else if (rs_is_recovering(rs))
2028 rs->ti->error = "Convert request on recovering raid set prohibited";
2029 else if (rs_is_reshaping(rs))
2030 rs->ti->error = "raid set already reshaping!";
2031 else if (!(rs_is_raid1(rs) || rs_is_raid10(rs) || rs_is_raid456(rs)))
2032 rs->ti->error = "Reshaping only supported for raid1/4/5/6/10";
2039 static int read_disk_sb(struct md_rdev *rdev, int size, bool force_reload)
2041 BUG_ON(!rdev->sb_page);
2043 if (rdev->sb_loaded && !force_reload)
2046 rdev->sb_loaded = 0;
2048 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, true)) {
2049 DMERR("Failed to read superblock of device at position %d",
2051 md_error(rdev->mddev, rdev);
2052 set_bit(Faulty, &rdev->flags);
2056 rdev->sb_loaded = 1;
2061 static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2063 failed_devices[0] = le64_to_cpu(sb->failed_devices);
2064 memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
2066 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2067 int i = ARRAY_SIZE(sb->extended_failed_devices);
2070 failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
2074 static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2076 int i = ARRAY_SIZE(sb->extended_failed_devices);
2078 sb->failed_devices = cpu_to_le64(failed_devices[0]);
2080 sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
2084 * Synchronize the superblock members with the raid set properties
2086 * All superblock data is little endian.
2088 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
2090 bool update_failed_devices = false;
2092 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2093 struct dm_raid_superblock *sb;
2094 struct raid_set *rs = container_of(mddev, struct raid_set, md);
2096 /* No metadata device, no superblock */
2097 if (!rdev->meta_bdev)
2100 BUG_ON(!rdev->sb_page);
2102 sb = page_address(rdev->sb_page);
2104 sb_retrieve_failed_devices(sb, failed_devices);
2106 for (i = 0; i < rs->raid_disks; i++)
2107 if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
2108 update_failed_devices = true;
2109 set_bit(i, (void *) failed_devices);
2112 if (update_failed_devices)
2113 sb_update_failed_devices(sb, failed_devices);
2115 sb->magic = cpu_to_le32(DM_RAID_MAGIC);
2116 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2118 sb->num_devices = cpu_to_le32(mddev->raid_disks);
2119 sb->array_position = cpu_to_le32(rdev->raid_disk);
2121 sb->events = cpu_to_le64(mddev->events);
2123 sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
2124 sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
2126 sb->level = cpu_to_le32(mddev->level);
2127 sb->layout = cpu_to_le32(mddev->layout);
2128 sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
2130 /********************************************************************
2131 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
2133 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
2135 sb->new_level = cpu_to_le32(mddev->new_level);
2136 sb->new_layout = cpu_to_le32(mddev->new_layout);
2137 sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
2139 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2141 smp_rmb(); /* Make sure we access most recent reshape position */
2142 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2143 if (le64_to_cpu(sb->reshape_position) != MaxSector) {
2144 /* Flag ongoing reshape */
2145 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
2147 if (mddev->delta_disks < 0 || mddev->reshape_backwards)
2148 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
2150 /* Clear reshape flags */
2151 sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
2154 sb->array_sectors = cpu_to_le64(mddev->array_sectors);
2155 sb->data_offset = cpu_to_le64(rdev->data_offset);
2156 sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
2157 sb->sectors = cpu_to_le64(rdev->sectors);
2158 sb->incompat_features = cpu_to_le32(0);
2160 /* Zero out the rest of the payload after the size of the superblock */
2161 memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
2167 * This function creates a superblock if one is not found on the device
2168 * and will decide which superblock to use if there's a choice.
2170 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
2172 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
2175 struct dm_raid_superblock *sb;
2176 struct dm_raid_superblock *refsb;
2177 uint64_t events_sb, events_refsb;
2179 r = read_disk_sb(rdev, rdev->sb_size, false);
2183 sb = page_address(rdev->sb_page);
2186 * Two cases that we want to write new superblocks and rebuild:
2187 * 1) New device (no matching magic number)
2188 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2190 if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
2191 (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
2192 super_sync(rdev->mddev, rdev);
2194 set_bit(FirstUse, &rdev->flags);
2195 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2197 /* Force writing of superblocks to disk */
2198 set_bit(MD_SB_CHANGE_DEVS, &rdev->mddev->sb_flags);
2200 /* Any superblock is better than none, choose that if given */
2201 return refdev ? 0 : 1;
2207 events_sb = le64_to_cpu(sb->events);
2209 refsb = page_address(refdev->sb_page);
2210 events_refsb = le64_to_cpu(refsb->events);
2212 return (events_sb > events_refsb) ? 1 : 0;
2215 static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
2218 struct mddev *mddev = &rs->md;
2220 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2221 struct dm_raid_superblock *sb;
2222 uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
2224 struct dm_raid_superblock *sb2;
2226 sb = page_address(rdev->sb_page);
2227 events_sb = le64_to_cpu(sb->events);
2230 * Initialise to 1 if this is a new superblock.
2232 mddev->events = events_sb ? : 1;
2234 mddev->reshape_position = MaxSector;
2236 mddev->raid_disks = le32_to_cpu(sb->num_devices);
2237 mddev->level = le32_to_cpu(sb->level);
2238 mddev->layout = le32_to_cpu(sb->layout);
2239 mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
2242 * Reshaping is supported, e.g. reshape_position is valid
2243 * in superblock and superblock content is authoritative.
2245 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2246 /* Superblock is authoritative wrt given raid set layout! */
2247 mddev->new_level = le32_to_cpu(sb->new_level);
2248 mddev->new_layout = le32_to_cpu(sb->new_layout);
2249 mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
2250 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
2251 mddev->array_sectors = le64_to_cpu(sb->array_sectors);
2253 /* raid was reshaping and got interrupted */
2254 if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
2255 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
2256 DMERR("Reshape requested but raid set is still reshaping");
2260 if (mddev->delta_disks < 0 ||
2261 (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
2262 mddev->reshape_backwards = 1;
2264 mddev->reshape_backwards = 0;
2266 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2267 rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2272 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2274 struct raid_type *rt_cur = get_raid_type_by_ll(mddev->level, mddev->layout);
2275 struct raid_type *rt_new = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
2277 if (rs_takeover_requested(rs)) {
2278 if (rt_cur && rt_new)
2279 DMERR("Takeover raid sets from %s to %s not yet supported by metadata. (raid level change)",
2280 rt_cur->name, rt_new->name);
2282 DMERR("Takeover raid sets not yet supported by metadata. (raid level change)");
2284 } else if (rs_reshape_requested(rs)) {
2285 DMERR("Reshaping raid sets not yet supported by metadata. (raid layout change keeping level)");
2286 if (mddev->layout != mddev->new_layout) {
2287 if (rt_cur && rt_new)
2288 DMERR(" current layout %s vs new layout %s",
2289 rt_cur->name, rt_new->name);
2291 DMERR(" current layout 0x%X vs new layout 0x%X",
2292 le32_to_cpu(sb->layout), mddev->new_layout);
2294 if (mddev->chunk_sectors != mddev->new_chunk_sectors)
2295 DMERR(" current stripe sectors %u vs new stripe sectors %u",
2296 mddev->chunk_sectors, mddev->new_chunk_sectors);
2297 if (rs->delta_disks)
2298 DMERR(" current %u disks vs new %u disks",
2299 mddev->raid_disks, mddev->raid_disks + rs->delta_disks);
2300 if (rs_is_raid10(rs)) {
2301 DMERR(" Old layout: %s w/ %u copies",
2302 raid10_md_layout_to_format(mddev->layout),
2303 raid10_md_layout_to_copies(mddev->layout));
2304 DMERR(" New layout: %s w/ %u copies",
2305 raid10_md_layout_to_format(mddev->new_layout),
2306 raid10_md_layout_to_copies(mddev->new_layout));
2311 DMINFO("Discovered old metadata format; upgrading to extended metadata format");
2314 if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
2315 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2318 * During load, we set FirstUse if a new superblock was written.
2319 * There are two reasons we might not have a superblock:
2320 * 1) The raid set is brand new - in which case, all of the
2321 * devices must have their In_sync bit set. Also,
2322 * recovery_cp must be 0, unless forced.
2323 * 2) This is a new device being added to an old raid set
2324 * and the new device needs to be rebuilt - in which
2325 * case the In_sync bit will /not/ be set and
2326 * recovery_cp must be MaxSector.
2327 * 3) This is/are a new device(s) being added to an old
2328 * raid set during takeover to a higher raid level
2329 * to provide capacity for redundancy or during reshape
2330 * to add capacity to grow the raid set.
2332 rdev_for_each(r, mddev) {
2333 if (test_bit(Journal, &rdev->flags))
2336 if (test_bit(FirstUse, &r->flags))
2339 if (!test_bit(In_sync, &r->flags)) {
2340 DMINFO("Device %d specified for rebuild; clearing superblock",
2344 if (test_bit(FirstUse, &r->flags))
2349 if (new_devs == rs->raid_disks || !rebuilds) {
2350 /* Replace a broken device */
2351 if (new_devs == rs->raid_disks) {
2352 DMINFO("Superblocks created for new raid set");
2353 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2354 } else if (new_devs != rebuilds &&
2355 new_devs != rs->delta_disks) {
2356 DMERR("New device injected into existing raid set without "
2357 "'delta_disks' or 'rebuild' parameter specified");
2360 } else if (new_devs && new_devs != rebuilds) {
2361 DMERR("%u 'rebuild' devices cannot be injected into"
2362 " a raid set with %u other first-time devices",
2363 rebuilds, new_devs);
2365 } else if (rebuilds) {
2366 if (rebuild_and_new && rebuilds != rebuild_and_new) {
2367 DMERR("new device%s provided without 'rebuild'",
2368 new_devs > 1 ? "s" : "");
2370 } else if (!test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) && rs_is_recovering(rs)) {
2371 DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2372 (unsigned long long) mddev->recovery_cp);
2374 } else if (rs_is_reshaping(rs)) {
2375 DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2376 (unsigned long long) mddev->reshape_position);
2382 * Now we set the Faulty bit for those devices that are
2383 * recorded in the superblock as failed.
2385 sb_retrieve_failed_devices(sb, failed_devices);
2386 rdev_for_each(r, mddev) {
2387 if (test_bit(Journal, &rdev->flags) ||
2390 sb2 = page_address(r->sb_page);
2391 sb2->failed_devices = 0;
2392 memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
2395 * Check for any device re-ordering.
2397 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2398 role = le32_to_cpu(sb2->array_position);
2402 if (role != r->raid_disk) {
2403 if (rs_is_raid10(rs) && __is_raid10_near(mddev->layout)) {
2404 if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
2405 rs->raid_disks % rs->raid10_copies) {
2407 "Cannot change raid10 near set to odd # of devices!";
2411 sb2->array_position = cpu_to_le32(r->raid_disk);
2413 } else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
2414 !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2415 !rt_is_raid1(rs->raid_type)) {
2416 rs->ti->error = "Cannot change device positions in raid set";
2420 DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
2424 * Partial recovery is performed on
2425 * returning failed devices.
2427 if (test_bit(role, (void *) failed_devices))
2428 set_bit(Faulty, &r->flags);
2435 static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
2437 struct mddev *mddev = &rs->md;
2438 struct dm_raid_superblock *sb;
2440 if (rs_is_raid0(rs) || !rdev->sb_page || rdev->raid_disk < 0)
2443 sb = page_address(rdev->sb_page);
2446 * If mddev->events is not set, we know we have not yet initialized
2449 if (!mddev->events && super_init_validation(rs, rdev))
2452 if (le32_to_cpu(sb->compat_features) &&
2453 le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
2454 rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2458 if (sb->incompat_features) {
2459 rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
2463 /* Enable bitmap creation on @rs unless no metadevs or raid0 or journaled raid4/5/6 set. */
2464 mddev->bitmap_info.offset = (rt_is_raid0(rs->raid_type) || rs->journal_dev.dev) ? 0 : to_sector(4096);
2465 mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2467 if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2469 * Retrieve rdev size stored in superblock to be prepared for shrink.
2470 * Check extended superblock members are present otherwise the size
2473 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190)
2474 rdev->sectors = le64_to_cpu(sb->sectors);
2476 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
2477 if (rdev->recovery_offset == MaxSector)
2478 set_bit(In_sync, &rdev->flags);
2480 * If no reshape in progress -> we're recovering single
2481 * disk(s) and have to set the device(s) to out-of-sync
2483 else if (!rs_is_reshaping(rs))
2484 clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
2488 * If a device comes back, set it as not In_sync and no longer faulty.
2490 if (test_and_clear_bit(Faulty, &rdev->flags)) {
2491 rdev->recovery_offset = 0;
2492 clear_bit(In_sync, &rdev->flags);
2493 rdev->saved_raid_disk = rdev->raid_disk;
2496 /* Reshape support -> restore repective data offsets */
2497 rdev->data_offset = le64_to_cpu(sb->data_offset);
2498 rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
2504 * Analyse superblocks and select the freshest.
2506 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2509 struct md_rdev *rdev, *freshest;
2510 struct mddev *mddev = &rs->md;
2513 rdev_for_each(rdev, mddev) {
2514 if (test_bit(Journal, &rdev->flags))
2517 if (!rdev->meta_bdev)
2520 /* Set superblock offset/size for metadata device. */
2522 rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
2523 if (rdev->sb_size < sizeof(struct dm_raid_superblock) || rdev->sb_size > PAGE_SIZE) {
2524 DMERR("superblock size of a logical block is no longer valid");
2529 * Skipping super_load due to CTR_FLAG_SYNC will cause
2530 * the array to undergo initialization again as
2531 * though it were new. This is the intended effect
2532 * of the "sync" directive.
2534 * With reshaping capability added, we must ensure that
2535 * the "sync" directive is disallowed during the reshape.
2537 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2540 r = super_load(rdev, freshest);
2549 /* This is a failure to read the superblock from the metadata device. */
2551 * We have to keep any raid0 data/metadata device pairs or
2552 * the MD raid0 personality will fail to start the array.
2554 if (rs_is_raid0(rs))
2558 * We keep the dm_devs to be able to emit the device tuple
2559 * properly on the table line in raid_status() (rather than
2560 * mistakenly acting as if '- -' got passed into the constructor).
2562 * The rdev has to stay on the same_set list to allow for
2563 * the attempt to restore faulty devices on second resume.
2565 rdev->raid_disk = rdev->saved_raid_disk = -1;
2574 * Validation of the freshest device provides the source of
2575 * validation for the remaining devices.
2577 rs->ti->error = "Unable to assemble array: Invalid superblocks";
2578 if (super_validate(rs, freshest))
2581 if (validate_raid_redundancy(rs)) {
2582 rs->ti->error = "Insufficient redundancy to activate array";
2586 rdev_for_each(rdev, mddev)
2587 if (!test_bit(Journal, &rdev->flags) &&
2589 super_validate(rs, rdev))
2595 * Adjust data_offset and new_data_offset on all disk members of @rs
2596 * for out of place reshaping if requested by constructor
2598 * We need free space at the beginning of each raid disk for forward
2599 * and at the end for backward reshapes which userspace has to provide
2600 * via remapping/reordering of space.
2602 static int rs_adjust_data_offsets(struct raid_set *rs)
2604 sector_t data_offset = 0, new_data_offset = 0;
2605 struct md_rdev *rdev;
2607 /* Constructor did not request data offset change */
2608 if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2609 if (!rs_is_reshapable(rs))
2615 /* HM FIXME: get In_Sync raid_dev? */
2616 rdev = &rs->dev[0].rdev;
2618 if (rs->delta_disks < 0) {
2620 * Removing disks (reshaping backwards):
2622 * - before reshape: data is at offset 0 and free space
2623 * is at end of each component LV
2625 * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2628 new_data_offset = rs->data_offset;
2630 } else if (rs->delta_disks > 0) {
2632 * Adding disks (reshaping forwards):
2634 * - before reshape: data is at offset rs->data_offset != 0 and
2635 * free space is at begin of each component LV
2637 * - after reshape: data is at offset 0 on each component LV
2639 data_offset = rs->data_offset;
2640 new_data_offset = 0;
2644 * User space passes in 0 for data offset after having removed reshape space
2646 * - or - (data offset != 0)
2648 * Changing RAID layout or chunk size -> toggle offsets
2650 * - before reshape: data is at offset rs->data_offset 0 and
2651 * free space is at end of each component LV
2653 * data is at offset rs->data_offset != 0 and
2654 * free space is at begin of each component LV
2656 * - after reshape: data is at offset 0 if it was at offset != 0
2657 * or at offset != 0 if it was at offset 0
2658 * on each component LV
2661 data_offset = rs->data_offset ? rdev->data_offset : 0;
2662 new_data_offset = data_offset ? 0 : rs->data_offset;
2663 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2667 * Make sure we got a minimum amount of free sectors per device
2669 if (rs->data_offset &&
2670 bdev_nr_sectors(rdev->bdev) - rs->md.dev_sectors < MIN_FREE_RESHAPE_SPACE) {
2671 rs->ti->error = data_offset ? "No space for forward reshape" :
2672 "No space for backward reshape";
2677 * Raise recovery_cp in case data_offset != 0 to
2678 * avoid false recovery positives in the constructor.
2680 if (rs->md.recovery_cp < rs->md.dev_sectors)
2681 rs->md.recovery_cp += rs->dev[0].rdev.data_offset;
2683 /* Adjust data offsets on all rdevs but on any raid4/5/6 journal device */
2684 rdev_for_each(rdev, &rs->md) {
2685 if (!test_bit(Journal, &rdev->flags)) {
2686 rdev->data_offset = data_offset;
2687 rdev->new_data_offset = new_data_offset;
2694 /* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2695 static void __reorder_raid_disk_indexes(struct raid_set *rs)
2698 struct md_rdev *rdev;
2700 rdev_for_each(rdev, &rs->md) {
2701 if (!test_bit(Journal, &rdev->flags)) {
2702 rdev->raid_disk = i++;
2703 rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2709 * Setup @rs for takeover by a different raid level
2711 static int rs_setup_takeover(struct raid_set *rs)
2713 struct mddev *mddev = &rs->md;
2714 struct md_rdev *rdev;
2715 unsigned int d = mddev->raid_disks = rs->raid_disks;
2716 sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2718 if (rt_is_raid10(rs->raid_type)) {
2719 if (rs_is_raid0(rs)) {
2720 /* Userpace reordered disks -> adjust raid_disk indexes */
2721 __reorder_raid_disk_indexes(rs);
2723 /* raid0 -> raid10_far layout */
2724 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2726 } else if (rs_is_raid1(rs))
2727 /* raid1 -> raid10_near layout */
2728 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2735 clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2736 mddev->recovery_cp = MaxSector;
2739 rdev = &rs->dev[d].rdev;
2741 if (test_bit(d, (void *) rs->rebuild_disks)) {
2742 clear_bit(In_sync, &rdev->flags);
2743 clear_bit(Faulty, &rdev->flags);
2744 mddev->recovery_cp = rdev->recovery_offset = 0;
2745 /* Bitmap has to be created when we do an "up" takeover */
2746 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2749 rdev->new_data_offset = new_data_offset;
2755 /* Prepare @rs for reshape */
2756 static int rs_prepare_reshape(struct raid_set *rs)
2759 struct mddev *mddev = &rs->md;
2761 if (rs_is_raid10(rs)) {
2762 if (rs->raid_disks != mddev->raid_disks &&
2763 __is_raid10_near(mddev->layout) &&
2764 rs->raid10_copies &&
2765 rs->raid10_copies != __raid10_near_copies(mddev->layout)) {
2767 * raid disk have to be multiple of data copies to allow this conversion,
2769 * This is actually not a reshape it is a
2770 * rebuild of any additional mirrors per group
2772 if (rs->raid_disks % rs->raid10_copies) {
2773 rs->ti->error = "Can't reshape raid10 mirror groups";
2777 /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2778 __reorder_raid_disk_indexes(rs);
2779 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2781 mddev->new_layout = mddev->layout;
2786 } else if (rs_is_raid456(rs))
2789 else if (rs_is_raid1(rs)) {
2790 if (rs->delta_disks) {
2791 /* Process raid1 via delta_disks */
2792 mddev->degraded = rs->delta_disks < 0 ? -rs->delta_disks : rs->delta_disks;
2795 /* Process raid1 without delta_disks */
2796 mddev->raid_disks = rs->raid_disks;
2800 rs->ti->error = "Called with bogus raid type";
2805 set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2806 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2807 } else if (mddev->raid_disks < rs->raid_disks)
2808 /* Create new superblocks and bitmaps, if any new disks */
2809 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2814 /* Get reshape sectors from data_offsets or raid set */
2815 static sector_t _get_reshape_sectors(struct raid_set *rs)
2817 struct md_rdev *rdev;
2818 sector_t reshape_sectors = 0;
2820 rdev_for_each(rdev, &rs->md)
2821 if (!test_bit(Journal, &rdev->flags)) {
2822 reshape_sectors = (rdev->data_offset > rdev->new_data_offset) ?
2823 rdev->data_offset - rdev->new_data_offset :
2824 rdev->new_data_offset - rdev->data_offset;
2828 return max(reshape_sectors, (sector_t) rs->data_offset);
2833 * - change raid layout
2834 * - change chunk size
2838 static int rs_setup_reshape(struct raid_set *rs)
2841 unsigned int cur_raid_devs, d;
2842 sector_t reshape_sectors = _get_reshape_sectors(rs);
2843 struct mddev *mddev = &rs->md;
2844 struct md_rdev *rdev;
2846 mddev->delta_disks = rs->delta_disks;
2847 cur_raid_devs = mddev->raid_disks;
2849 /* Ignore impossible layout change whilst adding/removing disks */
2850 if (mddev->delta_disks &&
2851 mddev->layout != mddev->new_layout) {
2852 DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2853 mddev->new_layout = mddev->layout;
2857 * Adjust array size:
2859 * - in case of adding disk(s), array size has
2860 * to grow after the disk adding reshape,
2861 * which'll happen in the event handler;
2862 * reshape will happen forward, so space has to
2863 * be available at the beginning of each disk
2865 * - in case of removing disk(s), array size
2866 * has to shrink before starting the reshape,
2867 * which'll happen here;
2868 * reshape will happen backward, so space has to
2869 * be available at the end of each disk
2871 * - data_offset and new_data_offset are
2872 * adjusted for aforementioned out of place
2873 * reshaping based on userspace passing in
2874 * the "data_offset <sectors>" key/value
2875 * pair via the constructor
2879 if (rs->delta_disks > 0) {
2880 /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2881 for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2882 rdev = &rs->dev[d].rdev;
2883 clear_bit(In_sync, &rdev->flags);
2886 * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2887 * by md, which'll store that erroneously in the superblock on reshape
2889 rdev->saved_raid_disk = -1;
2890 rdev->raid_disk = d;
2892 rdev->sectors = mddev->dev_sectors;
2893 rdev->recovery_offset = rs_is_raid1(rs) ? 0 : MaxSector;
2896 mddev->reshape_backwards = 0; /* adding disk(s) -> forward reshape */
2898 /* Remove disk(s) */
2899 } else if (rs->delta_disks < 0) {
2900 r = rs_set_dev_and_array_sectors(rs, rs->ti->len, true);
2901 mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2903 /* Change layout and/or chunk size */
2906 * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2908 * keeping number of disks and do layout change ->
2910 * toggle reshape_backward depending on data_offset:
2912 * - free space upfront -> reshape forward
2914 * - free space at the end -> reshape backward
2917 * This utilizes free reshape space avoiding the need
2918 * for userspace to move (parts of) LV segments in
2919 * case of layout/chunksize change (for disk
2920 * adding/removing reshape space has to be at
2921 * the proper address (see above with delta_disks):
2923 * add disk(s) -> begin
2924 * remove disk(s)-> end
2926 mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2930 * Adjust device size for forward reshape
2931 * because md_finish_reshape() reduces it.
2933 if (!mddev->reshape_backwards)
2934 rdev_for_each(rdev, &rs->md)
2935 if (!test_bit(Journal, &rdev->flags))
2936 rdev->sectors += reshape_sectors;
2942 * If the md resync thread has updated superblock with max reshape position
2943 * at the end of a reshape but not (yet) reset the layout configuration
2944 * changes -> reset the latter.
2946 static void rs_reset_inconclusive_reshape(struct raid_set *rs)
2948 if (!rs_is_reshaping(rs) && rs_is_layout_change(rs, true)) {
2950 rs->md.delta_disks = 0;
2951 rs->md.reshape_backwards = 0;
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 if (!rs->dev[i].rdev.bdev ||
2972 !bdev_max_discard_sectors(rs->dev[i].rdev.bdev))
2976 if (!devices_handle_discard_safely) {
2977 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2978 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2984 ti->num_discard_bios = 1;
2988 * Construct a RAID0/1/10/4/5/6 mapping:
2990 * <raid_type> <#raid_params> <raid_params>{0,} \
2991 * <#raid_devs> [<meta_dev1> <dev1>]{1,}
2993 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
2994 * details on possible <raid_params>.
2996 * Userspace is free to initialize the metadata devices, hence the superblocks to
2997 * enforce recreation based on the passed in table parameters.
3000 static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3003 bool resize = false;
3004 struct raid_type *rt;
3005 unsigned int num_raid_params, num_raid_devs;
3006 sector_t sb_array_sectors, rdev_sectors, reshape_sectors;
3007 struct raid_set *rs = NULL;
3009 struct rs_layout rs_layout;
3010 struct dm_arg_set as = { argc, argv }, as_nrd;
3011 struct dm_arg _args[] = {
3012 { 0, as.argc, "Cannot understand number of raid parameters" },
3013 { 1, 254, "Cannot understand number of raid devices parameters" }
3016 arg = dm_shift_arg(&as);
3018 ti->error = "No arguments";
3022 rt = get_raid_type(arg);
3024 ti->error = "Unrecognised raid_type";
3028 /* Must have <#raid_params> */
3029 if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
3032 /* number of raid device tupples <meta_dev data_dev> */
3034 dm_consume_args(&as_nrd, num_raid_params);
3035 _args[1].max = (as_nrd.argc - 1) / 2;
3036 if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
3039 if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
3040 ti->error = "Invalid number of supplied raid devices";
3044 rs = raid_set_alloc(ti, rt, num_raid_devs);
3048 r = parse_raid_params(rs, &as, num_raid_params);
3052 r = parse_dev_params(rs, &as);
3056 rs->md.sync_super = super_sync;
3059 * Calculate ctr requested array and device sizes to allow
3060 * for superblock analysis needing device sizes defined.
3062 * Any existing superblock will overwrite the array and device sizes
3064 r = rs_set_dev_and_array_sectors(rs, rs->ti->len, false);
3068 /* Memorize just calculated, potentially larger sizes to grow the raid set in preresume */
3069 rs->array_sectors = rs->md.array_sectors;
3070 rs->dev_sectors = rs->md.dev_sectors;
3073 * Backup any new raid set level, layout, ...
3074 * requested to be able to compare to superblock
3075 * members for conversion decisions.
3077 rs_config_backup(rs, &rs_layout);
3079 r = analyse_superblocks(ti, rs);
3083 /* All in-core metadata now as of current superblocks after calling analyse_superblocks() */
3084 sb_array_sectors = rs->md.array_sectors;
3085 rdev_sectors = __rdev_sectors(rs);
3086 if (!rdev_sectors) {
3087 ti->error = "Invalid rdev size";
3093 reshape_sectors = _get_reshape_sectors(rs);
3094 if (rs->dev_sectors != rdev_sectors) {
3095 resize = (rs->dev_sectors != rdev_sectors - reshape_sectors);
3096 if (rs->dev_sectors > rdev_sectors - reshape_sectors)
3097 set_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3100 INIT_WORK(&rs->md.event_work, do_table_event);
3102 ti->num_flush_bios = 1;
3103 ti->needs_bio_set_dev = true;
3105 /* Restore any requested new layout for conversion decision */
3106 rs_config_restore(rs, &rs_layout);
3109 * Now that we have any superblock metadata available,
3110 * check for new, recovering, reshaping, to be taken over,
3111 * to be reshaped or an existing, unchanged raid set to
3114 if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
3115 /* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
3116 if (rs_is_raid6(rs) &&
3117 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
3118 ti->error = "'nosync' not allowed for new raid6 set";
3122 rs_setup_recovery(rs, 0);
3123 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3125 } else if (rs_is_recovering(rs)) {
3126 /* A recovering raid set may be resized */
3128 } else if (rs_is_reshaping(rs)) {
3129 /* Have to reject size change request during reshape */
3131 ti->error = "Can't resize a reshaping raid set";
3136 } else if (rs_takeover_requested(rs)) {
3137 if (rs_is_reshaping(rs)) {
3138 ti->error = "Can't takeover a reshaping raid set";
3143 /* We can't takeover a journaled raid4/5/6 */
3144 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3145 ti->error = "Can't takeover a journaled raid4/5/6 set";
3151 * If a takeover is needed, userspace sets any additional
3152 * devices to rebuild and we can check for a valid request here.
3154 * If acceptable, set the level to the new requested
3155 * one, prohibit requesting recovery, allow the raid
3156 * set to run and store superblocks during resume.
3158 r = rs_check_takeover(rs);
3162 r = rs_setup_takeover(rs);
3166 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3167 /* Takeover ain't recovery, so disable recovery */
3168 rs_setup_recovery(rs, MaxSector);
3170 } else if (rs_reshape_requested(rs)) {
3171 /* Only request grow on raid set size extensions, not on reshapes. */
3172 clear_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3175 * No need to check for 'ongoing' takeover here, because takeover
3176 * is an instant operation as oposed to an ongoing reshape.
3179 /* We can't reshape a journaled raid4/5/6 */
3180 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3181 ti->error = "Can't reshape a journaled raid4/5/6 set";
3186 /* Out-of-place space has to be available to allow for a reshape unless raid1! */
3187 if (reshape_sectors || rs_is_raid1(rs)) {
3189 * We can only prepare for a reshape here, because the
3190 * raid set needs to run to provide the repective reshape
3191 * check functions via its MD personality instance.
3193 * So do the reshape check after md_run() succeeded.
3195 r = rs_prepare_reshape(rs);
3199 /* Reshaping ain't recovery, so disable recovery */
3200 rs_setup_recovery(rs, MaxSector);
3205 /* May not set recovery when a device rebuild is requested */
3206 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3207 clear_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3208 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3209 rs_setup_recovery(rs, MaxSector);
3210 } else if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags)) {
3212 * Set raid set to current size, i.e. size as of
3213 * superblocks to grow to larger size in preresume.
3215 r = rs_set_dev_and_array_sectors(rs, sb_array_sectors, false);
3219 rs_setup_recovery(rs, rs->md.recovery_cp < rs->md.dev_sectors ? rs->md.recovery_cp : rs->md.dev_sectors);
3221 /* This is no size change or it is shrinking, update size and record in superblocks */
3222 r = rs_set_dev_and_array_sectors(rs, rs->ti->len, false);
3226 if (sb_array_sectors > rs->array_sectors)
3227 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3232 /* If constructor requested it, change data and new_data offsets */
3233 r = rs_adjust_data_offsets(rs);
3237 /* Catch any inconclusive reshape superblock content. */
3238 rs_reset_inconclusive_reshape(rs);
3240 /* Start raid set read-only and assumed clean to change in raid_resume() */
3244 /* Has to be held on running the array */
3245 mddev_suspend_and_lock_nointr(&rs->md);
3247 /* Keep array frozen until resume. */
3248 md_frozen_sync_thread(&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);
3260 ti->error = "Failed to start raid array";
3264 /* If raid4/5/6 journal mode explicitly requested (only possible with journal dev) -> set it */
3265 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
3266 r = r5c_journal_mode_set(&rs->md, rs->journal_dev.mode);
3268 ti->error = "Failed to set raid4/5/6 journal mode";
3273 set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags);
3275 /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
3276 if (rs_is_raid456(rs)) {
3277 r = rs_set_raid456_stripe_cache(rs);
3282 /* Now do an early reshape check */
3283 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3284 r = rs_check_reshape(rs);
3288 /* Restore new, ctr requested layout to perform check */
3289 rs_config_restore(rs, &rs_layout);
3291 if (rs->md.pers->start_reshape) {
3292 r = rs->md.pers->check_reshape(&rs->md);
3294 ti->error = "Reshape check failed";
3300 /* Disable/enable discard support on raid set. */
3301 configure_discard_support(rs);
3303 mddev_unlock(&rs->md);
3308 mddev_unlock(&rs->md);
3315 static void raid_dtr(struct dm_target *ti)
3317 struct raid_set *rs = ti->private;
3319 mddev_lock_nointr(&rs->md);
3321 mddev_unlock(&rs->md);
3323 if (work_pending(&rs->md.event_work))
3324 flush_work(&rs->md.event_work);
3328 static int raid_map(struct dm_target *ti, struct bio *bio)
3330 struct raid_set *rs = ti->private;
3331 struct mddev *mddev = &rs->md;
3334 * If we're reshaping to add disk(s), ti->len and
3335 * mddev->array_sectors will differ during the process
3336 * (ti->len > mddev->array_sectors), so we have to requeue
3337 * bios with addresses > mddev->array_sectors here or
3338 * there will occur accesses past EOD of the component
3339 * data images thus erroring the raid set.
3341 if (unlikely(bio_has_data(bio) && bio_end_sector(bio) > mddev->array_sectors))
3342 return DM_MAPIO_REQUEUE;
3344 if (unlikely(!md_handle_request(mddev, bio)))
3345 return DM_MAPIO_REQUEUE;
3347 return DM_MAPIO_SUBMITTED;
3350 /* Return sync state string for @state */
3351 enum sync_state { st_frozen, st_reshape, st_resync, st_check, st_repair, st_recover, st_idle };
3352 static const char *sync_str(enum sync_state state)
3354 /* Has to be in above sync_state order! */
3355 static const char *sync_strs[] = {
3365 return __within_range(state, 0, ARRAY_SIZE(sync_strs) - 1) ? sync_strs[state] : "undef";
3368 /* Return enum sync_state for @mddev derived from @recovery flags */
3369 static enum sync_state decipher_sync_action(struct mddev *mddev, unsigned long recovery)
3371 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
3374 /* The MD sync thread can be done with io or be interrupted but still be running */
3375 if (!test_bit(MD_RECOVERY_DONE, &recovery) &&
3376 (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
3377 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery)))) {
3378 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
3381 if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
3382 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
3384 if (test_bit(MD_RECOVERY_CHECK, &recovery))
3389 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3392 if (mddev->reshape_position != MaxSector)
3400 * Return status string for @rdev
3402 * Status characters:
3404 * 'D' = Dead/Failed raid set component or raid4/5/6 journal device
3405 * 'a' = Alive but not in-sync raid set component _or_ alive raid4/5/6 'write_back' journal device
3406 * 'A' = Alive and in-sync raid set component _or_ alive raid4/5/6 'write_through' journal device
3407 * '-' = Non-existing device (i.e. uspace passed '- -' into the ctr)
3409 static const char *__raid_dev_status(struct raid_set *rs, struct md_rdev *rdev)
3413 else if (test_bit(Faulty, &rdev->flags))
3415 else if (test_bit(Journal, &rdev->flags))
3416 return (rs->journal_dev.mode == R5C_JOURNAL_MODE_WRITE_THROUGH) ? "A" : "a";
3417 else if (test_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags) ||
3418 (!test_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags) &&
3419 !test_bit(In_sync, &rdev->flags)))
3425 /* Helper to return resync/reshape progress for @rs and runtime flags for raid set in sync / resynching */
3426 static sector_t rs_get_progress(struct raid_set *rs, unsigned long recovery,
3427 enum sync_state state, sector_t resync_max_sectors)
3430 struct mddev *mddev = &rs->md;
3432 clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3433 clear_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3435 if (rs_is_raid0(rs)) {
3436 r = resync_max_sectors;
3437 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3440 if (state == st_idle && !test_bit(MD_RECOVERY_INTR, &recovery))
3441 r = mddev->recovery_cp;
3443 r = mddev->curr_resync_completed;
3445 if (state == st_idle && r >= resync_max_sectors) {
3449 /* In case we have finished recovering, the array is in sync. */
3450 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3451 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3453 } else if (state == st_recover)
3455 * In case we are recovering, the array is not in sync
3456 * and health chars should show the recovering legs.
3458 * Already retrieved recovery offset from curr_resync_completed above.
3462 else if (state == st_resync || state == st_reshape)
3464 * If "resync/reshape" is occurring, the raid set
3465 * is or may be out of sync hence the health
3466 * characters shall be 'a'.
3468 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3470 else if (state == st_check || state == st_repair)
3472 * If "check" or "repair" is occurring, the raid set has
3473 * undergone an initial sync and the health characters
3474 * should not be 'a' anymore.
3476 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3478 else if (test_bit(MD_RECOVERY_NEEDED, &recovery))
3480 * We are idle and recovery is needed, prevent 'A' chars race
3481 * caused by components still set to in-sync by constructor.
3483 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3487 * We are idle and the raid set may be doing an initial
3488 * sync, or it may be rebuilding individual components.
3489 * If all the devices are In_sync, then it is the raid set
3490 * that is being initialized.
3492 struct md_rdev *rdev;
3494 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3495 rdev_for_each(rdev, mddev)
3496 if (!test_bit(Journal, &rdev->flags) &&
3497 !test_bit(In_sync, &rdev->flags)) {
3498 clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3504 return min(r, resync_max_sectors);
3507 /* Helper to return @dev name or "-" if !@dev */
3508 static const char *__get_dev_name(struct dm_dev *dev)
3510 return dev ? dev->name : "-";
3513 static void raid_status(struct dm_target *ti, status_type_t type,
3514 unsigned int status_flags, char *result, unsigned int maxlen)
3516 struct raid_set *rs = ti->private;
3517 struct mddev *mddev = &rs->md;
3518 struct r5conf *conf = rs_is_raid456(rs) ? mddev->private : NULL;
3519 int i, max_nr_stripes = conf ? conf->max_nr_stripes : 0;
3520 unsigned long recovery;
3521 unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3522 unsigned int sz = 0;
3523 unsigned int rebuild_writemostly_count = 0;
3524 sector_t progress, resync_max_sectors, resync_mismatches;
3525 enum sync_state state;
3526 struct raid_type *rt;
3529 case STATUSTYPE_INFO:
3530 /* *Should* always succeed */
3531 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3535 DMEMIT("%s %d ", rt->name, mddev->raid_disks);
3537 /* Access most recent mddev properties for status output */
3539 /* Get sensible max sectors even if raid set not yet started */
3540 resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
3541 mddev->resync_max_sectors : mddev->dev_sectors;
3542 recovery = rs->md.recovery;
3543 state = decipher_sync_action(mddev, recovery);
3544 progress = rs_get_progress(rs, recovery, state, resync_max_sectors);
3545 resync_mismatches = (mddev->last_sync_action && !strcasecmp(mddev->last_sync_action, "check")) ?
3546 atomic64_read(&mddev->resync_mismatches) : 0;
3548 /* HM FIXME: do we want another state char for raid0? It shows 'D'/'A'/'-' now */
3549 for (i = 0; i < rs->raid_disks; i++)
3550 DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev));
3553 * In-sync/Reshape ratio:
3554 * The in-sync ratio shows the progress of:
3555 * - Initializing the raid set
3556 * - Rebuilding a subset of devices of the raid set
3557 * The user can distinguish between the two by referring
3558 * to the status characters.
3560 * The reshape ratio shows the progress of
3561 * changing the raid layout or the number of
3562 * disks of a raid set
3564 DMEMIT(" %llu/%llu", (unsigned long long) progress,
3565 (unsigned long long) resync_max_sectors);
3571 * See Documentation/admin-guide/device-mapper/dm-raid.rst for
3572 * information on each of these states.
3574 DMEMIT(" %s", sync_str(state));
3579 * resync_mismatches/mismatch_cnt
3580 * This field shows the number of discrepancies found when
3581 * performing a "check" of the raid set.
3583 DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3588 * data_offset (needed for out of space reshaping)
3589 * This field shows the data offset into the data
3590 * image LV where the first stripes data starts.
3592 * We keep data_offset equal on all raid disks of the set,
3593 * so retrieving it from the first raid disk is sufficient.
3595 DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
3600 DMEMIT(" %s", test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ?
3601 __raid_dev_status(rs, &rs->journal_dev.rdev) : "-");
3604 case STATUSTYPE_TABLE:
3605 /* Report the table line string you would use to construct this raid set */
3608 * Count any rebuild or writemostly argument pairs and subtract the
3609 * hweight count being added below of any rebuild and writemostly ctr flags.
3611 for (i = 0; i < rs->raid_disks; i++) {
3612 rebuild_writemostly_count += (test_bit(i, (void *) rs->rebuild_disks) ? 2 : 0) +
3613 (test_bit(WriteMostly, &rs->dev[i].rdev.flags) ? 2 : 0);
3615 rebuild_writemostly_count -= (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) ? 2 : 0) +
3616 (test_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags) ? 2 : 0);
3617 /* Calculate raid parameter count based on ^ rebuild/writemostly argument counts and ctr flags set. */
3618 raid_param_cnt += rebuild_writemostly_count +
3619 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3620 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2;
3621 /* Emit table line */
3622 /* This has to be in the documented order for userspace! */
3623 DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
3624 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
3625 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
3626 if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3627 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
3628 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags))
3629 for (i = 0; i < rs->raid_disks; i++)
3630 if (test_bit(i, (void *) rs->rebuild_disks))
3631 DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD), i);
3632 if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3633 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3634 mddev->bitmap_info.daemon_sleep);
3635 if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3636 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3637 mddev->sync_speed_min);
3638 if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3639 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3640 mddev->sync_speed_max);
3641 if (test_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags))
3642 for (i = 0; i < rs->raid_disks; i++)
3643 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3644 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3645 rs->dev[i].rdev.raid_disk);
3646 if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
3647 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
3648 mddev->bitmap_info.max_write_behind);
3649 if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3650 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3652 if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3653 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3654 (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3655 if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3656 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3657 raid10_md_layout_to_copies(mddev->layout));
3658 if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3659 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3660 raid10_md_layout_to_format(mddev->layout));
3661 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3662 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3663 max(rs->delta_disks, mddev->delta_disks));
3664 if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3665 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3666 (unsigned long long) rs->data_offset);
3667 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags))
3668 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV),
3669 __get_dev_name(rs->journal_dev.dev));
3670 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags))
3671 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE),
3672 md_journal_mode_to_dm_raid(rs->journal_dev.mode));
3673 DMEMIT(" %d", rs->raid_disks);
3674 for (i = 0; i < rs->raid_disks; i++)
3675 DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
3676 __get_dev_name(rs->dev[i].data_dev));
3679 case STATUSTYPE_IMA:
3680 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3684 DMEMIT_TARGET_NAME_VERSION(ti->type);
3685 DMEMIT(",raid_type=%s,raid_disks=%d", rt->name, mddev->raid_disks);
3687 /* Access most recent mddev properties for status output */
3689 recovery = rs->md.recovery;
3690 state = decipher_sync_action(mddev, recovery);
3691 DMEMIT(",raid_state=%s", sync_str(state));
3693 for (i = 0; i < rs->raid_disks; i++) {
3694 DMEMIT(",raid_device_%d_status=", i);
3695 DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev));
3698 if (rt_is_raid456(rt)) {
3699 DMEMIT(",journal_dev_mode=");
3700 switch (rs->journal_dev.mode) {
3701 case R5C_JOURNAL_MODE_WRITE_THROUGH:
3703 _raid456_journal_mode[R5C_JOURNAL_MODE_WRITE_THROUGH].param);
3705 case R5C_JOURNAL_MODE_WRITE_BACK:
3707 _raid456_journal_mode[R5C_JOURNAL_MODE_WRITE_BACK].param);
3719 static int raid_message(struct dm_target *ti, unsigned int argc, char **argv,
3720 char *result, unsigned int maxlen)
3722 struct raid_set *rs = ti->private;
3723 struct mddev *mddev = &rs->md;
3726 if (!mddev->pers || !mddev->pers->sync_request)
3729 if (test_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags) ||
3730 test_bit(RT_FLAG_RS_FROZEN, &rs->runtime_flags))
3733 if (!strcasecmp(argv[0], "frozen")) {
3734 ret = mddev_lock(mddev);
3738 md_frozen_sync_thread(mddev);
3739 mddev_unlock(mddev);
3740 } else if (!strcasecmp(argv[0], "idle")) {
3741 ret = mddev_lock(mddev);
3745 md_idle_sync_thread(mddev);
3746 mddev_unlock(mddev);
3749 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3750 if (decipher_sync_action(mddev, mddev->recovery) != st_idle)
3752 else if (!strcasecmp(argv[0], "resync"))
3753 ; /* MD_RECOVERY_NEEDED set below */
3754 else if (!strcasecmp(argv[0], "recover"))
3755 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3757 if (!strcasecmp(argv[0], "check")) {
3758 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3759 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3760 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3761 } else if (!strcasecmp(argv[0], "repair")) {
3762 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3763 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3767 if (mddev->ro == 2) {
3768 /* A write to sync_action is enough to justify
3769 * canceling read-auto mode
3772 if (!mddev->suspended)
3773 md_wakeup_thread(mddev->sync_thread);
3775 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3776 if (!mddev->suspended)
3777 md_wakeup_thread(mddev->thread);
3782 static int raid_iterate_devices(struct dm_target *ti,
3783 iterate_devices_callout_fn fn, void *data)
3785 struct raid_set *rs = ti->private;
3789 for (i = 0; !r && i < rs->raid_disks; i++) {
3790 if (rs->dev[i].data_dev) {
3791 r = fn(ti, rs->dev[i].data_dev,
3792 0, /* No offset on data devs */
3793 rs->md.dev_sectors, data);
3800 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3802 struct raid_set *rs = ti->private;
3803 unsigned int chunk_size_bytes = to_bytes(rs->md.chunk_sectors);
3805 blk_limits_io_min(limits, chunk_size_bytes);
3806 blk_limits_io_opt(limits, chunk_size_bytes * mddev_data_stripes(rs));
3809 static void raid_presuspend(struct dm_target *ti)
3811 struct raid_set *rs = ti->private;
3812 struct mddev *mddev = &rs->md;
3815 * From now on, disallow raid_message() to change sync_thread until
3816 * resume, raid_postsuspend() is too late.
3818 set_bit(RT_FLAG_RS_FROZEN, &rs->runtime_flags);
3820 if (!reshape_interrupted(mddev))
3824 * For raid456, if reshape is interrupted, IO across reshape position
3825 * will never make progress, while caller will wait for IO to be done.
3826 * Inform raid456 to handle those IO to prevent deadlock.
3828 if (mddev->pers && mddev->pers->prepare_suspend)
3829 mddev->pers->prepare_suspend(mddev);
3832 static void raid_presuspend_undo(struct dm_target *ti)
3834 struct raid_set *rs = ti->private;
3836 clear_bit(RT_FLAG_RS_FROZEN, &rs->runtime_flags);
3839 static void raid_postsuspend(struct dm_target *ti)
3841 struct raid_set *rs = ti->private;
3843 if (!test_and_set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
3845 * sync_thread must be stopped during suspend, and writes have
3846 * to be stopped before suspending to avoid deadlocks.
3848 md_stop_writes(&rs->md);
3849 mddev_suspend(&rs->md, false);
3853 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
3856 uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
3857 unsigned long flags;
3858 bool cleared = false;
3859 struct dm_raid_superblock *sb;
3860 struct mddev *mddev = &rs->md;
3863 /* RAID personalities have to provide hot add/remove methods or we need to bail out. */
3864 if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
3867 memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
3869 for (i = 0; i < rs->raid_disks; i++) {
3870 r = &rs->dev[i].rdev;
3871 /* HM FIXME: enhance journal device recovery processing */
3872 if (test_bit(Journal, &r->flags))
3875 if (test_bit(Faulty, &r->flags) &&
3876 r->meta_bdev && !read_disk_sb(r, r->sb_size, true)) {
3877 DMINFO("Faulty %s device #%d has readable super block."
3878 " Attempting to revive it.",
3879 rs->raid_type->name, i);
3882 * Faulty bit may be set, but sometimes the array can
3883 * be suspended before the personalities can respond
3884 * by removing the device from the array (i.e. calling
3885 * 'hot_remove_disk'). If they haven't yet removed
3886 * the failed device, its 'raid_disk' number will be
3887 * '>= 0' - meaning we must call this function
3891 clear_bit(In_sync, &r->flags); /* Mandatory for hot remove. */
3892 if (r->raid_disk >= 0) {
3893 if (mddev->pers->hot_remove_disk(mddev, r)) {
3894 /* Failed to revive this device, try next */
3899 r->raid_disk = r->saved_raid_disk = i;
3901 clear_bit(Faulty, &r->flags);
3902 clear_bit(WriteErrorSeen, &r->flags);
3904 if (mddev->pers->hot_add_disk(mddev, r)) {
3905 /* Failed to revive this device, try next */
3906 r->raid_disk = r->saved_raid_disk = -1;
3909 clear_bit(In_sync, &r->flags);
3910 r->recovery_offset = 0;
3911 set_bit(i, (void *) cleared_failed_devices);
3917 /* If any failed devices could be cleared, update all sbs failed_devices bits */
3919 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
3921 rdev_for_each(r, &rs->md) {
3922 if (test_bit(Journal, &r->flags))
3925 sb = page_address(r->sb_page);
3926 sb_retrieve_failed_devices(sb, failed_devices);
3928 for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
3929 failed_devices[i] &= ~cleared_failed_devices[i];
3931 sb_update_failed_devices(sb, failed_devices);
3936 static int __load_dirty_region_bitmap(struct raid_set *rs)
3940 /* Try loading the bitmap unless "raid0", which does not have one */
3941 if (!rs_is_raid0(rs) &&
3942 !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
3943 r = md_bitmap_load(&rs->md);
3945 DMERR("Failed to load bitmap");
3951 /* Enforce updating all superblocks */
3952 static void rs_update_sbs(struct raid_set *rs)
3954 struct mddev *mddev = &rs->md;
3957 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3959 md_update_sb(mddev, 1);
3964 * Reshape changes raid algorithm of @rs to new one within personality
3965 * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3966 * disks from a raid set thus growing/shrinking it or resizes the set
3968 * Call mddev_lock_nointr() before!
3970 static int rs_start_reshape(struct raid_set *rs)
3973 struct mddev *mddev = &rs->md;
3974 struct md_personality *pers = mddev->pers;
3976 /* Don't allow the sync thread to work until the table gets reloaded. */
3977 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
3979 r = rs_setup_reshape(rs);
3984 * Check any reshape constraints enforced by the personalility
3986 * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3988 r = pers->check_reshape(mddev);
3990 rs->ti->error = "pers->check_reshape() failed";
3995 * Personality may not provide start reshape method in which
3996 * case check_reshape above has already covered everything
3998 if (pers->start_reshape) {
3999 r = pers->start_reshape(mddev);
4001 rs->ti->error = "pers->start_reshape() failed";
4007 * Now reshape got set up, update superblocks to
4008 * reflect the fact so that a table reload will
4009 * access proper superblock content in the ctr.
4016 static int raid_preresume(struct dm_target *ti)
4019 struct raid_set *rs = ti->private;
4020 struct mddev *mddev = &rs->md;
4022 /* This is a resume after a suspend of the set -> it's already started. */
4023 if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
4027 * The superblocks need to be updated on disk if the
4028 * array is new or new devices got added (thus zeroed
4029 * out by userspace) or __load_dirty_region_bitmap
4030 * will overwrite them in core with old data or fail.
4032 if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
4035 /* Load the bitmap from disk unless raid0 */
4036 r = __load_dirty_region_bitmap(rs);
4040 /* We are extending the raid set size, adjust mddev/md_rdev sizes and set capacity. */
4041 if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags)) {
4042 mddev->array_sectors = rs->array_sectors;
4043 mddev->dev_sectors = rs->dev_sectors;
4044 rs_set_rdev_sectors(rs);
4045 rs_set_capacity(rs);
4048 /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) or grown device size */
4049 if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) && mddev->bitmap &&
4050 (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags) ||
4051 (rs->requested_bitmap_chunk_sectors &&
4052 mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)))) {
4053 int chunksize = to_bytes(rs->requested_bitmap_chunk_sectors) ?: mddev->bitmap_info.chunksize;
4055 r = md_bitmap_resize(mddev->bitmap, mddev->dev_sectors, chunksize, 0);
4057 DMERR("Failed to resize bitmap");
4060 /* Check for any resize/reshape on @rs and adjust/initiate */
4061 if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
4062 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4063 mddev->resync_min = mddev->recovery_cp;
4064 if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags))
4065 mddev->resync_max_sectors = mddev->dev_sectors;
4068 /* Check for any reshape request unless new raid set */
4069 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
4070 /* Initiate a reshape. */
4071 rs_set_rdev_sectors(rs);
4072 mddev_lock_nointr(mddev);
4073 r = rs_start_reshape(rs);
4074 mddev_unlock(mddev);
4076 DMWARN("Failed to check/start reshape, continuing without change");
4083 static void raid_resume(struct dm_target *ti)
4085 struct raid_set *rs = ti->private;
4086 struct mddev *mddev = &rs->md;
4088 if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
4090 * A secondary resume while the device is active.
4091 * Take this opportunity to check whether any failed
4092 * devices are reachable again.
4094 mddev_lock_nointr(mddev);
4095 attempt_restore_of_faulty_devices(rs);
4096 mddev_unlock(mddev);
4099 if (test_and_clear_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
4100 /* Only reduce raid set size before running a disk removing reshape. */
4101 if (mddev->delta_disks < 0)
4102 rs_set_capacity(rs);
4104 WARN_ON_ONCE(!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery));
4105 WARN_ON_ONCE(test_bit(MD_RECOVERY_RUNNING, &mddev->recovery));
4106 clear_bit(RT_FLAG_RS_FROZEN, &rs->runtime_flags);
4107 mddev_lock_nointr(mddev);
4110 md_unfrozen_sync_thread(mddev);
4111 mddev_unlock_and_resume(mddev);
4115 static struct target_type raid_target = {
4117 .version = {1, 15, 1},
4118 .module = THIS_MODULE,
4122 .status = raid_status,
4123 .message = raid_message,
4124 .iterate_devices = raid_iterate_devices,
4125 .io_hints = raid_io_hints,
4126 .presuspend = raid_presuspend,
4127 .presuspend_undo = raid_presuspend_undo,
4128 .postsuspend = raid_postsuspend,
4129 .preresume = raid_preresume,
4130 .resume = raid_resume,
4134 module_param(devices_handle_discard_safely, bool, 0644);
4135 MODULE_PARM_DESC(devices_handle_discard_safely,
4136 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
4138 MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
4139 MODULE_ALIAS("dm-raid0");
4140 MODULE_ALIAS("dm-raid1");
4141 MODULE_ALIAS("dm-raid10");
4142 MODULE_ALIAS("dm-raid4");
4143 MODULE_ALIAS("dm-raid5");
4144 MODULE_ALIAS("dm-raid6");
4145 MODULE_AUTHOR("Neil Brown <dm-devel@lists.linux.dev>");
4146 MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@lists.linux.dev>");
4147 MODULE_LICENSE("GPL");