2 * Copyright (C) 2010-2011 Neil Brown
3 * Copyright (C) 2010-2017 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
8 #include <linux/slab.h>
9 #include <linux/module.h>
15 #include "md-bitmap.h"
17 #include <linux/device-mapper.h>
19 #define DM_MSG_PREFIX "raid"
20 #define MAX_RAID_DEVICES 253 /* md-raid kernel limit */
23 * Minimum sectors of free reshape space per raid device
25 #define MIN_FREE_RESHAPE_SPACE to_sector(4*4096)
28 * Minimum journal space 4 MiB in sectors.
30 #define MIN_RAID456_JOURNAL_SPACE (4*2048)
32 static bool devices_handle_discard_safely = false;
35 * The following flags are used by dm-raid.c to set up the array state.
36 * They must be cleared before md_run is called.
38 #define FirstUse 10 /* rdev flag */
42 * Two DM devices, one to hold metadata and one to hold the
43 * actual data/parity. The reason for this is to not confuse
44 * ti->len and give more flexibility in altering size and
47 * While it is possible for this device to be associated
48 * with a different physical device than the data_dev, it
49 * is intended for it to be the same.
50 * |--------- Physical Device ---------|
51 * |- meta_dev -|------ data_dev ------|
53 struct dm_dev *meta_dev;
54 struct dm_dev *data_dev;
59 * Bits for establishing rs->ctr_flags
64 #define __CTR_FLAG_SYNC 0 /* 1 */ /* Not with raid0! */
65 #define __CTR_FLAG_NOSYNC 1 /* 1 */ /* Not with raid0! */
66 #define __CTR_FLAG_REBUILD 2 /* 2 */ /* Not with raid0! */
67 #define __CTR_FLAG_DAEMON_SLEEP 3 /* 2 */ /* Not with raid0! */
68 #define __CTR_FLAG_MIN_RECOVERY_RATE 4 /* 2 */ /* Not with raid0! */
69 #define __CTR_FLAG_MAX_RECOVERY_RATE 5 /* 2 */ /* Not with raid0! */
70 #define __CTR_FLAG_MAX_WRITE_BEHIND 6 /* 2 */ /* Only with raid1! */
71 #define __CTR_FLAG_WRITE_MOSTLY 7 /* 2 */ /* Only with raid1! */
72 #define __CTR_FLAG_STRIPE_CACHE 8 /* 2 */ /* Only with raid4/5/6! */
73 #define __CTR_FLAG_REGION_SIZE 9 /* 2 */ /* Not with raid0! */
74 #define __CTR_FLAG_RAID10_COPIES 10 /* 2 */ /* Only with raid10 */
75 #define __CTR_FLAG_RAID10_FORMAT 11 /* 2 */ /* Only with raid10 */
77 #define __CTR_FLAG_DELTA_DISKS 12 /* 2 */ /* Only with reshapable raid1/4/5/6/10! */
78 #define __CTR_FLAG_DATA_OFFSET 13 /* 2 */ /* Only with reshapable raid4/5/6/10! */
79 #define __CTR_FLAG_RAID10_USE_NEAR_SETS 14 /* 2 */ /* Only with raid10! */
82 #define __CTR_FLAG_JOURNAL_DEV 15 /* 2 */ /* Only with raid4/5/6 (journal device)! */
85 #define __CTR_FLAG_JOURNAL_MODE 16 /* 2 */ /* Only with raid4/5/6 (journal mode)! */
88 * Flags for rs->ctr_flags field.
90 #define CTR_FLAG_SYNC (1 << __CTR_FLAG_SYNC)
91 #define CTR_FLAG_NOSYNC (1 << __CTR_FLAG_NOSYNC)
92 #define CTR_FLAG_REBUILD (1 << __CTR_FLAG_REBUILD)
93 #define CTR_FLAG_DAEMON_SLEEP (1 << __CTR_FLAG_DAEMON_SLEEP)
94 #define CTR_FLAG_MIN_RECOVERY_RATE (1 << __CTR_FLAG_MIN_RECOVERY_RATE)
95 #define CTR_FLAG_MAX_RECOVERY_RATE (1 << __CTR_FLAG_MAX_RECOVERY_RATE)
96 #define CTR_FLAG_MAX_WRITE_BEHIND (1 << __CTR_FLAG_MAX_WRITE_BEHIND)
97 #define CTR_FLAG_WRITE_MOSTLY (1 << __CTR_FLAG_WRITE_MOSTLY)
98 #define CTR_FLAG_STRIPE_CACHE (1 << __CTR_FLAG_STRIPE_CACHE)
99 #define CTR_FLAG_REGION_SIZE (1 << __CTR_FLAG_REGION_SIZE)
100 #define CTR_FLAG_RAID10_COPIES (1 << __CTR_FLAG_RAID10_COPIES)
101 #define CTR_FLAG_RAID10_FORMAT (1 << __CTR_FLAG_RAID10_FORMAT)
102 #define CTR_FLAG_DELTA_DISKS (1 << __CTR_FLAG_DELTA_DISKS)
103 #define CTR_FLAG_DATA_OFFSET (1 << __CTR_FLAG_DATA_OFFSET)
104 #define CTR_FLAG_RAID10_USE_NEAR_SETS (1 << __CTR_FLAG_RAID10_USE_NEAR_SETS)
105 #define CTR_FLAG_JOURNAL_DEV (1 << __CTR_FLAG_JOURNAL_DEV)
106 #define CTR_FLAG_JOURNAL_MODE (1 << __CTR_FLAG_JOURNAL_MODE)
108 #define RESUME_STAY_FROZEN_FLAGS (CTR_FLAG_DELTA_DISKS | CTR_FLAG_DATA_OFFSET)
111 * Definitions of various constructor flags to
112 * be used in checks of valid / invalid flags
115 /* Define all any sync flags */
116 #define CTR_FLAGS_ANY_SYNC (CTR_FLAG_SYNC | CTR_FLAG_NOSYNC)
118 /* Define flags for options without argument (e.g. 'nosync') */
119 #define CTR_FLAG_OPTIONS_NO_ARGS (CTR_FLAGS_ANY_SYNC | \
120 CTR_FLAG_RAID10_USE_NEAR_SETS)
122 /* Define flags for options with one argument (e.g. 'delta_disks +2') */
123 #define CTR_FLAG_OPTIONS_ONE_ARG (CTR_FLAG_REBUILD | \
124 CTR_FLAG_WRITE_MOSTLY | \
125 CTR_FLAG_DAEMON_SLEEP | \
126 CTR_FLAG_MIN_RECOVERY_RATE | \
127 CTR_FLAG_MAX_RECOVERY_RATE | \
128 CTR_FLAG_MAX_WRITE_BEHIND | \
129 CTR_FLAG_STRIPE_CACHE | \
130 CTR_FLAG_REGION_SIZE | \
131 CTR_FLAG_RAID10_COPIES | \
132 CTR_FLAG_RAID10_FORMAT | \
133 CTR_FLAG_DELTA_DISKS | \
134 CTR_FLAG_DATA_OFFSET)
136 /* Valid options definitions per raid level... */
138 /* "raid0" does only accept data offset */
139 #define RAID0_VALID_FLAGS (CTR_FLAG_DATA_OFFSET)
141 /* "raid1" does not accept stripe cache, data offset, delta_disks or any raid10 options */
142 #define RAID1_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
144 CTR_FLAG_WRITE_MOSTLY | \
145 CTR_FLAG_DAEMON_SLEEP | \
146 CTR_FLAG_MIN_RECOVERY_RATE | \
147 CTR_FLAG_MAX_RECOVERY_RATE | \
148 CTR_FLAG_MAX_WRITE_BEHIND | \
149 CTR_FLAG_REGION_SIZE | \
150 CTR_FLAG_DELTA_DISKS | \
151 CTR_FLAG_DATA_OFFSET)
153 /* "raid10" does not accept any raid1 or stripe cache options */
154 #define RAID10_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
156 CTR_FLAG_DAEMON_SLEEP | \
157 CTR_FLAG_MIN_RECOVERY_RATE | \
158 CTR_FLAG_MAX_RECOVERY_RATE | \
159 CTR_FLAG_REGION_SIZE | \
160 CTR_FLAG_RAID10_COPIES | \
161 CTR_FLAG_RAID10_FORMAT | \
162 CTR_FLAG_DELTA_DISKS | \
163 CTR_FLAG_DATA_OFFSET | \
164 CTR_FLAG_RAID10_USE_NEAR_SETS)
167 * "raid4/5/6" do not accept any raid1 or raid10 specific options
169 * "raid6" does not accept "nosync", because it is not guaranteed
170 * that both parity and q-syndrome are being written properly with
173 #define RAID45_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
175 CTR_FLAG_DAEMON_SLEEP | \
176 CTR_FLAG_MIN_RECOVERY_RATE | \
177 CTR_FLAG_MAX_RECOVERY_RATE | \
178 CTR_FLAG_STRIPE_CACHE | \
179 CTR_FLAG_REGION_SIZE | \
180 CTR_FLAG_DELTA_DISKS | \
181 CTR_FLAG_DATA_OFFSET | \
182 CTR_FLAG_JOURNAL_DEV | \
183 CTR_FLAG_JOURNAL_MODE)
185 #define RAID6_VALID_FLAGS (CTR_FLAG_SYNC | \
187 CTR_FLAG_DAEMON_SLEEP | \
188 CTR_FLAG_MIN_RECOVERY_RATE | \
189 CTR_FLAG_MAX_RECOVERY_RATE | \
190 CTR_FLAG_STRIPE_CACHE | \
191 CTR_FLAG_REGION_SIZE | \
192 CTR_FLAG_DELTA_DISKS | \
193 CTR_FLAG_DATA_OFFSET | \
194 CTR_FLAG_JOURNAL_DEV | \
195 CTR_FLAG_JOURNAL_MODE)
196 /* ...valid options definitions per raid level */
199 * Flags for rs->runtime_flags field
200 * (RT_FLAG prefix meaning "runtime flag")
202 * These are all internal and used to define runtime state,
203 * e.g. to prevent another resume from preresume processing
204 * the raid set all over again.
206 #define RT_FLAG_RS_PRERESUMED 0
207 #define RT_FLAG_RS_RESUMED 1
208 #define RT_FLAG_RS_BITMAP_LOADED 2
209 #define RT_FLAG_UPDATE_SBS 3
210 #define RT_FLAG_RESHAPE_RS 4
211 #define RT_FLAG_RS_SUSPENDED 5
213 /* Array elements of 64 bit needed for rebuild/failed disk bits */
214 #define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
217 * raid set level, layout and chunk sectors backup/restore
222 int new_chunk_sectors;
226 struct dm_target *ti;
228 uint32_t bitmap_loaded;
229 uint32_t stripe_cache_entries;
230 unsigned long ctr_flags;
231 unsigned long runtime_flags;
233 uint64_t rebuild_disks[DISKS_ARRAY_ELEMS];
239 int requested_bitmap_chunk_sectors;
242 struct raid_type *raid_type;
243 struct dm_target_callbacks callbacks;
245 /* Optional raid4/5/6 journal device */
252 struct raid_dev dev[0];
255 static void rs_config_backup(struct raid_set *rs, struct rs_layout *l)
257 struct mddev *mddev = &rs->md;
259 l->new_level = mddev->new_level;
260 l->new_layout = mddev->new_layout;
261 l->new_chunk_sectors = mddev->new_chunk_sectors;
264 static void rs_config_restore(struct raid_set *rs, struct rs_layout *l)
266 struct mddev *mddev = &rs->md;
268 mddev->new_level = l->new_level;
269 mddev->new_layout = l->new_layout;
270 mddev->new_chunk_sectors = l->new_chunk_sectors;
273 /* raid10 algorithms (i.e. formats) */
274 #define ALGORITHM_RAID10_DEFAULT 0
275 #define ALGORITHM_RAID10_NEAR 1
276 #define ALGORITHM_RAID10_OFFSET 2
277 #define ALGORITHM_RAID10_FAR 3
279 /* Supported raid types and properties. */
280 static struct raid_type {
281 const char *name; /* RAID algorithm. */
282 const char *descr; /* Descriptor text for logging. */
283 const unsigned int parity_devs; /* # of parity devices. */
284 const unsigned int minimal_devs;/* minimal # of devices in set. */
285 const unsigned int level; /* RAID level. */
286 const unsigned int algorithm; /* RAID algorithm. */
288 {"raid0", "raid0 (striping)", 0, 2, 0, 0 /* NONE */},
289 {"raid1", "raid1 (mirroring)", 0, 2, 1, 0 /* NONE */},
290 {"raid10_far", "raid10 far (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_FAR},
291 {"raid10_offset", "raid10 offset (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_OFFSET},
292 {"raid10_near", "raid10 near (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_NEAR},
293 {"raid10", "raid10 (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_DEFAULT},
294 {"raid4", "raid4 (dedicated first parity disk)", 1, 2, 5, ALGORITHM_PARITY_0}, /* raid4 layout = raid5_0 */
295 {"raid5_n", "raid5 (dedicated last parity disk)", 1, 2, 5, ALGORITHM_PARITY_N},
296 {"raid5_ls", "raid5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
297 {"raid5_rs", "raid5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
298 {"raid5_la", "raid5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
299 {"raid5_ra", "raid5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
300 {"raid6_zr", "raid6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
301 {"raid6_nr", "raid6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
302 {"raid6_nc", "raid6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE},
303 {"raid6_n_6", "raid6 (dedicated parity/Q n/6)", 2, 4, 6, ALGORITHM_PARITY_N_6},
304 {"raid6_ls_6", "raid6 (left symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_SYMMETRIC_6},
305 {"raid6_rs_6", "raid6 (right symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_SYMMETRIC_6},
306 {"raid6_la_6", "raid6 (left asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_ASYMMETRIC_6},
307 {"raid6_ra_6", "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_ASYMMETRIC_6}
310 /* True, if @v is in inclusive range [@min, @max] */
311 static bool __within_range(long v, long min, long max)
313 return v >= min && v <= max;
316 /* All table line arguments are defined here */
317 static struct arg_name_flag {
318 const unsigned long flag;
320 } __arg_name_flags[] = {
321 { CTR_FLAG_SYNC, "sync"},
322 { CTR_FLAG_NOSYNC, "nosync"},
323 { CTR_FLAG_REBUILD, "rebuild"},
324 { CTR_FLAG_DAEMON_SLEEP, "daemon_sleep"},
325 { CTR_FLAG_MIN_RECOVERY_RATE, "min_recovery_rate"},
326 { CTR_FLAG_MAX_RECOVERY_RATE, "max_recovery_rate"},
327 { CTR_FLAG_MAX_WRITE_BEHIND, "max_write_behind"},
328 { CTR_FLAG_WRITE_MOSTLY, "write_mostly"},
329 { CTR_FLAG_STRIPE_CACHE, "stripe_cache"},
330 { CTR_FLAG_REGION_SIZE, "region_size"},
331 { CTR_FLAG_RAID10_COPIES, "raid10_copies"},
332 { CTR_FLAG_RAID10_FORMAT, "raid10_format"},
333 { CTR_FLAG_DATA_OFFSET, "data_offset"},
334 { CTR_FLAG_DELTA_DISKS, "delta_disks"},
335 { CTR_FLAG_RAID10_USE_NEAR_SETS, "raid10_use_near_sets"},
336 { CTR_FLAG_JOURNAL_DEV, "journal_dev" },
337 { CTR_FLAG_JOURNAL_MODE, "journal_mode" },
340 /* Return argument name string for given @flag */
341 static const char *dm_raid_arg_name_by_flag(const uint32_t flag)
343 if (hweight32(flag) == 1) {
344 struct arg_name_flag *anf = __arg_name_flags + ARRAY_SIZE(__arg_name_flags);
346 while (anf-- > __arg_name_flags)
347 if (flag & anf->flag)
351 DMERR("%s called with more than one flag!", __func__);
356 /* Define correlation of raid456 journal cache modes and dm-raid target line parameters */
360 } _raid456_journal_mode[] = {
361 { R5C_JOURNAL_MODE_WRITE_THROUGH , "writethrough" },
362 { R5C_JOURNAL_MODE_WRITE_BACK , "writeback" }
365 /* Return MD raid4/5/6 journal mode for dm @journal_mode one */
366 static int dm_raid_journal_mode_to_md(const char *mode)
368 int m = ARRAY_SIZE(_raid456_journal_mode);
371 if (!strcasecmp(mode, _raid456_journal_mode[m].param))
372 return _raid456_journal_mode[m].mode;
377 /* Return dm-raid raid4/5/6 journal mode string for @mode */
378 static const char *md_journal_mode_to_dm_raid(const int mode)
380 int m = ARRAY_SIZE(_raid456_journal_mode);
383 if (mode == _raid456_journal_mode[m].mode)
384 return _raid456_journal_mode[m].param;
390 * Bool helpers to test for various raid levels of a raid set.
391 * It's level as reported by the superblock rather than
392 * the requested raid_type passed to the constructor.
394 /* Return true, if raid set in @rs is raid0 */
395 static bool rs_is_raid0(struct raid_set *rs)
397 return !rs->md.level;
400 /* Return true, if raid set in @rs is raid1 */
401 static bool rs_is_raid1(struct raid_set *rs)
403 return rs->md.level == 1;
406 /* Return true, if raid set in @rs is raid10 */
407 static bool rs_is_raid10(struct raid_set *rs)
409 return rs->md.level == 10;
412 /* Return true, if raid set in @rs is level 6 */
413 static bool rs_is_raid6(struct raid_set *rs)
415 return rs->md.level == 6;
418 /* Return true, if raid set in @rs is level 4, 5 or 6 */
419 static bool rs_is_raid456(struct raid_set *rs)
421 return __within_range(rs->md.level, 4, 6);
424 /* Return true, if raid set in @rs is reshapable */
425 static bool __is_raid10_far(int layout);
426 static bool rs_is_reshapable(struct raid_set *rs)
428 return rs_is_raid456(rs) ||
429 (rs_is_raid10(rs) && !__is_raid10_far(rs->md.new_layout));
432 /* Return true, if raid set in @rs is recovering */
433 static bool rs_is_recovering(struct raid_set *rs)
435 return rs->md.recovery_cp < rs->md.dev_sectors;
438 /* Return true, if raid set in @rs is reshaping */
439 static bool rs_is_reshaping(struct raid_set *rs)
441 return rs->md.reshape_position != MaxSector;
445 * bool helpers to test for various raid levels of a raid type @rt
448 /* Return true, if raid type in @rt is raid0 */
449 static bool rt_is_raid0(struct raid_type *rt)
454 /* Return true, if raid type in @rt is raid1 */
455 static bool rt_is_raid1(struct raid_type *rt)
457 return rt->level == 1;
460 /* Return true, if raid type in @rt is raid10 */
461 static bool rt_is_raid10(struct raid_type *rt)
463 return rt->level == 10;
466 /* Return true, if raid type in @rt is raid4/5 */
467 static bool rt_is_raid45(struct raid_type *rt)
469 return __within_range(rt->level, 4, 5);
472 /* Return true, if raid type in @rt is raid6 */
473 static bool rt_is_raid6(struct raid_type *rt)
475 return rt->level == 6;
478 /* Return true, if raid type in @rt is raid4/5/6 */
479 static bool rt_is_raid456(struct raid_type *rt)
481 return __within_range(rt->level, 4, 6);
483 /* END: raid level bools */
485 /* Return valid ctr flags for the raid level of @rs */
486 static unsigned long __valid_flags(struct raid_set *rs)
488 if (rt_is_raid0(rs->raid_type))
489 return RAID0_VALID_FLAGS;
490 else if (rt_is_raid1(rs->raid_type))
491 return RAID1_VALID_FLAGS;
492 else if (rt_is_raid10(rs->raid_type))
493 return RAID10_VALID_FLAGS;
494 else if (rt_is_raid45(rs->raid_type))
495 return RAID45_VALID_FLAGS;
496 else if (rt_is_raid6(rs->raid_type))
497 return RAID6_VALID_FLAGS;
503 * Check for valid flags set on @rs
505 * Has to be called after parsing of the ctr flags!
507 static int rs_check_for_valid_flags(struct raid_set *rs)
509 if (rs->ctr_flags & ~__valid_flags(rs)) {
510 rs->ti->error = "Invalid flags combination";
517 /* MD raid10 bit definitions and helpers */
518 #define RAID10_OFFSET (1 << 16) /* stripes with data copies area adjacent on devices */
519 #define RAID10_BROCKEN_USE_FAR_SETS (1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
520 #define RAID10_USE_FAR_SETS (1 << 18) /* Use sets instead of whole stripe rotation */
521 #define RAID10_FAR_COPIES_SHIFT 8 /* raid10 # far copies shift (2nd byte of layout) */
523 /* Return md raid10 near copies for @layout */
524 static unsigned int __raid10_near_copies(int layout)
526 return layout & 0xFF;
529 /* Return md raid10 far copies for @layout */
530 static unsigned int __raid10_far_copies(int layout)
532 return __raid10_near_copies(layout >> RAID10_FAR_COPIES_SHIFT);
535 /* Return true if md raid10 offset for @layout */
536 static bool __is_raid10_offset(int layout)
538 return !!(layout & RAID10_OFFSET);
541 /* Return true if md raid10 near for @layout */
542 static bool __is_raid10_near(int layout)
544 return !__is_raid10_offset(layout) && __raid10_near_copies(layout) > 1;
547 /* Return true if md raid10 far for @layout */
548 static bool __is_raid10_far(int layout)
550 return !__is_raid10_offset(layout) && __raid10_far_copies(layout) > 1;
553 /* Return md raid10 layout string for @layout */
554 static const char *raid10_md_layout_to_format(int layout)
557 * Bit 16 stands for "offset"
558 * (i.e. adjacent stripes hold copies)
560 * Refer to MD's raid10.c for details
562 if (__is_raid10_offset(layout))
565 if (__raid10_near_copies(layout) > 1)
568 if (__raid10_far_copies(layout) > 1)
574 /* Return md raid10 algorithm for @name */
575 static int raid10_name_to_format(const char *name)
577 if (!strcasecmp(name, "near"))
578 return ALGORITHM_RAID10_NEAR;
579 else if (!strcasecmp(name, "offset"))
580 return ALGORITHM_RAID10_OFFSET;
581 else if (!strcasecmp(name, "far"))
582 return ALGORITHM_RAID10_FAR;
587 /* Return md raid10 copies for @layout */
588 static unsigned int raid10_md_layout_to_copies(int layout)
590 return max(__raid10_near_copies(layout), __raid10_far_copies(layout));
593 /* Return md raid10 format id for @format string */
594 static int raid10_format_to_md_layout(struct raid_set *rs,
595 unsigned int algorithm,
598 unsigned int n = 1, f = 1, r = 0;
601 * MD resilienece flaw:
603 * enabling use_far_sets for far/offset formats causes copies
604 * to be colocated on the same devs together with their origins!
606 * -> disable it for now in the definition above
608 if (algorithm == ALGORITHM_RAID10_DEFAULT ||
609 algorithm == ALGORITHM_RAID10_NEAR)
612 else if (algorithm == ALGORITHM_RAID10_OFFSET) {
615 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
616 r |= RAID10_USE_FAR_SETS;
618 } else if (algorithm == ALGORITHM_RAID10_FAR) {
621 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
622 r |= RAID10_USE_FAR_SETS;
627 return r | (f << RAID10_FAR_COPIES_SHIFT) | n;
629 /* END: MD raid10 bit definitions and helpers */
631 /* Check for any of the raid10 algorithms */
632 static bool __got_raid10(struct raid_type *rtp, const int layout)
634 if (rtp->level == 10) {
635 switch (rtp->algorithm) {
636 case ALGORITHM_RAID10_DEFAULT:
637 case ALGORITHM_RAID10_NEAR:
638 return __is_raid10_near(layout);
639 case ALGORITHM_RAID10_OFFSET:
640 return __is_raid10_offset(layout);
641 case ALGORITHM_RAID10_FAR:
642 return __is_raid10_far(layout);
651 /* Return raid_type for @name */
652 static struct raid_type *get_raid_type(const char *name)
654 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
656 while (rtp-- > raid_types)
657 if (!strcasecmp(rtp->name, name))
663 /* Return raid_type for @name based derived from @level and @layout */
664 static struct raid_type *get_raid_type_by_ll(const int level, const int layout)
666 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
668 while (rtp-- > raid_types) {
669 /* RAID10 special checks based on @layout flags/properties */
670 if (rtp->level == level &&
671 (__got_raid10(rtp, layout) || rtp->algorithm == layout))
679 * Conditionally change bdev capacity of @rs
680 * in case of a disk add/remove reshape
682 static void rs_set_capacity(struct raid_set *rs)
684 struct mddev *mddev = &rs->md;
685 struct md_rdev *rdev;
686 struct gendisk *gendisk = dm_disk(dm_table_get_md(rs->ti->table));
689 * raid10 sets rdev->sector to the device size, which
690 * is unintended in case of out-of-place reshaping
692 rdev_for_each(rdev, mddev)
693 if (!test_bit(Journal, &rdev->flags))
694 rdev->sectors = mddev->dev_sectors;
696 set_capacity(gendisk, mddev->array_sectors);
697 revalidate_disk(gendisk);
701 * Set the mddev properties in @rs to the current
702 * ones retrieved from the freshest superblock
704 static void rs_set_cur(struct raid_set *rs)
706 struct mddev *mddev = &rs->md;
708 mddev->new_level = mddev->level;
709 mddev->new_layout = mddev->layout;
710 mddev->new_chunk_sectors = mddev->chunk_sectors;
714 * Set the mddev properties in @rs to the new
715 * ones requested by the ctr
717 static void rs_set_new(struct raid_set *rs)
719 struct mddev *mddev = &rs->md;
721 mddev->level = mddev->new_level;
722 mddev->layout = mddev->new_layout;
723 mddev->chunk_sectors = mddev->new_chunk_sectors;
724 mddev->raid_disks = rs->raid_disks;
725 mddev->delta_disks = 0;
728 static struct raid_set *raid_set_alloc(struct dm_target *ti, struct raid_type *raid_type,
729 unsigned int raid_devs)
734 if (raid_devs <= raid_type->parity_devs) {
735 ti->error = "Insufficient number of devices";
736 return ERR_PTR(-EINVAL);
739 rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
741 ti->error = "Cannot allocate raid context";
742 return ERR_PTR(-ENOMEM);
747 rs->raid_disks = raid_devs;
751 rs->raid_type = raid_type;
752 rs->stripe_cache_entries = 256;
753 rs->md.raid_disks = raid_devs;
754 rs->md.level = raid_type->level;
755 rs->md.new_level = rs->md.level;
756 rs->md.layout = raid_type->algorithm;
757 rs->md.new_layout = rs->md.layout;
758 rs->md.delta_disks = 0;
759 rs->md.recovery_cp = MaxSector;
761 for (i = 0; i < raid_devs; i++)
762 md_rdev_init(&rs->dev[i].rdev);
765 * Remaining items to be initialized by further RAID params:
768 * rs->md.chunk_sectors
769 * rs->md.new_chunk_sectors
776 static void raid_set_free(struct raid_set *rs)
780 if (rs->journal_dev.dev) {
781 md_rdev_clear(&rs->journal_dev.rdev);
782 dm_put_device(rs->ti, rs->journal_dev.dev);
785 for (i = 0; i < rs->raid_disks; i++) {
786 if (rs->dev[i].meta_dev)
787 dm_put_device(rs->ti, rs->dev[i].meta_dev);
788 md_rdev_clear(&rs->dev[i].rdev);
789 if (rs->dev[i].data_dev)
790 dm_put_device(rs->ti, rs->dev[i].data_dev);
797 * For every device we have two words
798 * <meta_dev>: meta device name or '-' if missing
799 * <data_dev>: data device name or '-' if missing
801 * The following are permitted:
804 * <meta_dev> <data_dev>
806 * The following is not allowed:
809 * This code parses those words. If there is a failure,
810 * the caller must use raid_set_free() to unwind the operations.
812 static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
816 int metadata_available = 0;
820 /* Put off the number of raid devices argument to get to dev pairs */
821 arg = dm_shift_arg(as);
825 for (i = 0; i < rs->raid_disks; i++) {
826 rs->dev[i].rdev.raid_disk = i;
828 rs->dev[i].meta_dev = NULL;
829 rs->dev[i].data_dev = NULL;
832 * There are no offsets initially.
833 * Out of place reshape will set them accordingly.
835 rs->dev[i].rdev.data_offset = 0;
836 rs->dev[i].rdev.new_data_offset = 0;
837 rs->dev[i].rdev.mddev = &rs->md;
839 arg = dm_shift_arg(as);
843 if (strcmp(arg, "-")) {
844 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
845 &rs->dev[i].meta_dev);
847 rs->ti->error = "RAID metadata device lookup failure";
851 rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
852 if (!rs->dev[i].rdev.sb_page) {
853 rs->ti->error = "Failed to allocate superblock page";
858 arg = dm_shift_arg(as);
862 if (!strcmp(arg, "-")) {
863 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
864 (!rs->dev[i].rdev.recovery_offset)) {
865 rs->ti->error = "Drive designated for rebuild not specified";
869 if (rs->dev[i].meta_dev) {
870 rs->ti->error = "No data device supplied with metadata device";
877 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
878 &rs->dev[i].data_dev);
880 rs->ti->error = "RAID device lookup failure";
884 if (rs->dev[i].meta_dev) {
885 metadata_available = 1;
886 rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
888 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
889 list_add_tail(&rs->dev[i].rdev.same_set, &rs->md.disks);
890 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
894 if (rs->journal_dev.dev)
895 list_add_tail(&rs->journal_dev.rdev.same_set, &rs->md.disks);
897 if (metadata_available) {
899 rs->md.persistent = 1;
900 rs->md.major_version = 2;
901 } else if (rebuild && !rs->md.recovery_cp) {
903 * Without metadata, we will not be able to tell if the array
904 * is in-sync or not - we must assume it is not. Therefore,
905 * it is impossible to rebuild a drive.
907 * Even if there is metadata, the on-disk information may
908 * indicate that the array is not in-sync and it will then
911 * User could specify 'nosync' option if desperate.
913 rs->ti->error = "Unable to rebuild drive while array is not in-sync";
921 * validate_region_size
923 * @region_size: region size in sectors. If 0, pick a size (4MiB default).
925 * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
926 * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
928 * Returns: 0 on success, -EINVAL on failure.
930 static int validate_region_size(struct raid_set *rs, unsigned long region_size)
932 unsigned long min_region_size = rs->ti->len / (1 << 21);
939 * Choose a reasonable default. All figures in sectors.
941 if (min_region_size > (1 << 13)) {
942 /* If not a power of 2, make it the next power of 2 */
943 region_size = roundup_pow_of_two(min_region_size);
944 DMINFO("Choosing default region size of %lu sectors",
947 DMINFO("Choosing default region size of 4MiB");
948 region_size = 1 << 13; /* sectors */
952 * Validate user-supplied value.
954 if (region_size > rs->ti->len) {
955 rs->ti->error = "Supplied region size is too large";
959 if (region_size < min_region_size) {
960 DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
961 region_size, min_region_size);
962 rs->ti->error = "Supplied region size is too small";
966 if (!is_power_of_2(region_size)) {
967 rs->ti->error = "Region size is not a power of 2";
971 if (region_size < rs->md.chunk_sectors) {
972 rs->ti->error = "Region size is smaller than the chunk size";
978 * Convert sectors to bytes.
980 rs->md.bitmap_info.chunksize = to_bytes(region_size);
986 * validate_raid_redundancy
989 * Determine if there are enough devices in the array that haven't
990 * failed (or are being rebuilt) to form a usable array.
992 * Returns: 0 on success, -EINVAL on failure.
994 static int validate_raid_redundancy(struct raid_set *rs)
996 unsigned int i, rebuild_cnt = 0;
997 unsigned int rebuilds_per_group = 0, copies;
998 unsigned int group_size, last_group_start;
1000 for (i = 0; i < rs->md.raid_disks; i++)
1001 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
1002 !rs->dev[i].rdev.sb_page)
1005 switch (rs->raid_type->level) {
1009 if (rebuild_cnt >= rs->md.raid_disks)
1015 if (rebuild_cnt > rs->raid_type->parity_devs)
1019 copies = raid10_md_layout_to_copies(rs->md.new_layout);
1020 if (rebuild_cnt < copies)
1024 * It is possible to have a higher rebuild count for RAID10,
1025 * as long as the failed devices occur in different mirror
1026 * groups (i.e. different stripes).
1028 * When checking "near" format, make sure no adjacent devices
1029 * have failed beyond what can be handled. In addition to the
1030 * simple case where the number of devices is a multiple of the
1031 * number of copies, we must also handle cases where the number
1032 * of devices is not a multiple of the number of copies.
1033 * E.g. dev1 dev2 dev3 dev4 dev5
1037 if (__is_raid10_near(rs->md.new_layout)) {
1038 for (i = 0; i < rs->md.raid_disks; i++) {
1040 rebuilds_per_group = 0;
1041 if ((!rs->dev[i].rdev.sb_page ||
1042 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1043 (++rebuilds_per_group >= copies))
1050 * When checking "far" and "offset" formats, we need to ensure
1051 * that the device that holds its copy is not also dead or
1052 * being rebuilt. (Note that "far" and "offset" formats only
1053 * support two copies right now. These formats also only ever
1054 * use the 'use_far_sets' variant.)
1056 * This check is somewhat complicated by the need to account
1057 * for arrays that are not a multiple of (far) copies. This
1058 * results in the need to treat the last (potentially larger)
1061 group_size = (rs->md.raid_disks / copies);
1062 last_group_start = (rs->md.raid_disks / group_size) - 1;
1063 last_group_start *= group_size;
1064 for (i = 0; i < rs->md.raid_disks; i++) {
1065 if (!(i % copies) && !(i > last_group_start))
1066 rebuilds_per_group = 0;
1067 if ((!rs->dev[i].rdev.sb_page ||
1068 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1069 (++rebuilds_per_group >= copies))
1085 * Possible arguments are...
1086 * <chunk_size> [optional_args]
1088 * Argument definitions
1089 * <chunk_size> The number of sectors per disk that
1090 * will form the "stripe"
1091 * [[no]sync] Force or prevent recovery of the
1093 * [rebuild <idx>] Rebuild the drive indicated by the index
1094 * [daemon_sleep <ms>] Time between bitmap daemon work to
1096 * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1097 * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1098 * [write_mostly <idx>] Indicate a write mostly drive via index
1099 * [max_write_behind <sectors>] See '-write-behind=' (man mdadm)
1100 * [stripe_cache <sectors>] Stripe cache size for higher RAIDs
1101 * [region_size <sectors>] Defines granularity of bitmap
1102 * [journal_dev <dev>] raid4/5/6 journaling deviice
1103 * (i.e. write hole closing log)
1105 * RAID10-only options:
1106 * [raid10_copies <# copies>] Number of copies. (Default: 2)
1107 * [raid10_format <near|far|offset>] Layout algorithm. (Default: near)
1109 static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
1110 unsigned int num_raid_params)
1112 int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
1113 unsigned int raid10_copies = 2;
1114 unsigned int i, write_mostly = 0;
1115 unsigned int region_size = 0;
1116 sector_t max_io_len;
1117 const char *arg, *key;
1118 struct raid_dev *rd;
1119 struct raid_type *rt = rs->raid_type;
1121 arg = dm_shift_arg(as);
1122 num_raid_params--; /* Account for chunk_size argument */
1124 if (kstrtoint(arg, 10, &value) < 0) {
1125 rs->ti->error = "Bad numerical argument given for chunk_size";
1130 * First, parse the in-order required arguments
1131 * "chunk_size" is the only argument of this type.
1133 if (rt_is_raid1(rt)) {
1135 DMERR("Ignoring chunk size parameter for RAID 1");
1137 } else if (!is_power_of_2(value)) {
1138 rs->ti->error = "Chunk size must be a power of 2";
1140 } else if (value < 8) {
1141 rs->ti->error = "Chunk size value is too small";
1145 rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
1148 * We set each individual device as In_sync with a completed
1149 * 'recovery_offset'. If there has been a device failure or
1150 * replacement then one of the following cases applies:
1152 * 1) User specifies 'rebuild'.
1153 * - Device is reset when param is read.
1154 * 2) A new device is supplied.
1155 * - No matching superblock found, resets device.
1156 * 3) Device failure was transient and returns on reload.
1157 * - Failure noticed, resets device for bitmap replay.
1158 * 4) Device hadn't completed recovery after previous failure.
1159 * - Superblock is read and overrides recovery_offset.
1161 * What is found in the superblocks of the devices is always
1162 * authoritative, unless 'rebuild' or '[no]sync' was specified.
1164 for (i = 0; i < rs->raid_disks; i++) {
1165 set_bit(In_sync, &rs->dev[i].rdev.flags);
1166 rs->dev[i].rdev.recovery_offset = MaxSector;
1170 * Second, parse the unordered optional arguments
1172 for (i = 0; i < num_raid_params; i++) {
1173 key = dm_shift_arg(as);
1175 rs->ti->error = "Not enough raid parameters given";
1179 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
1180 if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1181 rs->ti->error = "Only one 'nosync' argument allowed";
1186 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
1187 if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
1188 rs->ti->error = "Only one 'sync' argument allowed";
1193 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
1194 if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1195 rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1201 arg = dm_shift_arg(as);
1202 i++; /* Account for the argument pairs */
1204 rs->ti->error = "Wrong number of raid parameters given";
1209 * Parameters that take a string value are checked here.
1211 /* "raid10_format {near|offset|far} */
1212 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
1213 if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
1214 rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1217 if (!rt_is_raid10(rt)) {
1218 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1221 raid10_format = raid10_name_to_format(arg);
1222 if (raid10_format < 0) {
1223 rs->ti->error = "Invalid 'raid10_format' value given";
1224 return raid10_format;
1229 /* "journal_dev <dev>" */
1230 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV))) {
1232 struct md_rdev *jdev;
1234 if (test_and_set_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1235 rs->ti->error = "Only one raid4/5/6 set journaling device allowed";
1238 if (!rt_is_raid456(rt)) {
1239 rs->ti->error = "'journal_dev' is an invalid parameter for this RAID type";
1242 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
1243 &rs->journal_dev.dev);
1245 rs->ti->error = "raid4/5/6 journal device lookup failure";
1248 jdev = &rs->journal_dev.rdev;
1250 jdev->mddev = &rs->md;
1251 jdev->bdev = rs->journal_dev.dev->bdev;
1252 jdev->sectors = to_sector(i_size_read(jdev->bdev->bd_inode));
1253 if (jdev->sectors < MIN_RAID456_JOURNAL_SPACE) {
1254 rs->ti->error = "No space for raid4/5/6 journal";
1257 rs->journal_dev.mode = R5C_JOURNAL_MODE_WRITE_THROUGH;
1258 set_bit(Journal, &jdev->flags);
1262 /* "journal_mode <mode>" ("journal_dev" mandatory!) */
1263 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE))) {
1266 if (!test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1267 rs->ti->error = "raid4/5/6 'journal_mode' is invalid without 'journal_dev'";
1270 if (test_and_set_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
1271 rs->ti->error = "Only one raid4/5/6 'journal_mode' argument allowed";
1274 r = dm_raid_journal_mode_to_md(arg);
1276 rs->ti->error = "Invalid 'journal_mode' argument";
1279 rs->journal_dev.mode = r;
1284 * Parameters with number values from here on.
1286 if (kstrtoint(arg, 10, &value) < 0) {
1287 rs->ti->error = "Bad numerical argument given in raid params";
1291 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
1293 * "rebuild" is being passed in by userspace to provide
1294 * indexes of replaced devices and to set up additional
1295 * devices on raid level takeover.
1297 if (!__within_range(value, 0, rs->raid_disks - 1)) {
1298 rs->ti->error = "Invalid rebuild index given";
1302 if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1303 rs->ti->error = "rebuild for this index already given";
1307 rd = rs->dev + value;
1308 clear_bit(In_sync, &rd->rdev.flags);
1309 clear_bit(Faulty, &rd->rdev.flags);
1310 rd->rdev.recovery_offset = 0;
1311 set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
1312 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
1313 if (!rt_is_raid1(rt)) {
1314 rs->ti->error = "write_mostly option is only valid for RAID1";
1318 if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
1319 rs->ti->error = "Invalid write_mostly index given";
1324 set_bit(WriteMostly, &rs->dev[value].rdev.flags);
1325 set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
1326 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
1327 if (!rt_is_raid1(rt)) {
1328 rs->ti->error = "max_write_behind option is only valid for RAID1";
1332 if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
1333 rs->ti->error = "Only one max_write_behind argument pair allowed";
1338 * In device-mapper, we specify things in sectors, but
1339 * MD records this value in kB
1342 if (value > COUNTER_MAX) {
1343 rs->ti->error = "Max write-behind limit out of range";
1347 rs->md.bitmap_info.max_write_behind = value;
1348 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
1349 if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
1350 rs->ti->error = "Only one daemon_sleep argument pair allowed";
1353 if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
1354 rs->ti->error = "daemon sleep period out of range";
1357 rs->md.bitmap_info.daemon_sleep = value;
1358 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
1359 /* Userspace passes new data_offset after having extended the the data image LV */
1360 if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
1361 rs->ti->error = "Only one data_offset argument pair allowed";
1364 /* Ensure sensible data offset */
1366 (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
1367 rs->ti->error = "Bogus data_offset value";
1370 rs->data_offset = value;
1371 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
1372 /* Define the +/-# of disks to add to/remove from the given raid set */
1373 if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1374 rs->ti->error = "Only one delta_disks argument pair allowed";
1377 /* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1378 if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
1379 rs->ti->error = "Too many delta_disk requested";
1383 rs->delta_disks = value;
1384 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
1385 if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
1386 rs->ti->error = "Only one stripe_cache argument pair allowed";
1390 if (!rt_is_raid456(rt)) {
1391 rs->ti->error = "Inappropriate argument: stripe_cache";
1395 rs->stripe_cache_entries = value;
1396 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
1397 if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1398 rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1401 if (value > INT_MAX) {
1402 rs->ti->error = "min_recovery_rate out of range";
1405 rs->md.sync_speed_min = (int)value;
1406 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
1407 if (test_and_set_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags)) {
1408 rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1411 if (value > INT_MAX) {
1412 rs->ti->error = "max_recovery_rate out of range";
1415 rs->md.sync_speed_max = (int)value;
1416 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
1417 if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
1418 rs->ti->error = "Only one region_size argument pair allowed";
1422 region_size = value;
1423 rs->requested_bitmap_chunk_sectors = value;
1424 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
1425 if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
1426 rs->ti->error = "Only one raid10_copies argument pair allowed";
1430 if (!__within_range(value, 2, rs->md.raid_disks)) {
1431 rs->ti->error = "Bad value for 'raid10_copies'";
1435 raid10_copies = value;
1437 DMERR("Unable to parse RAID parameter: %s", key);
1438 rs->ti->error = "Unable to parse RAID parameter";
1443 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1444 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1445 rs->ti->error = "sync and nosync are mutually exclusive";
1449 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) &&
1450 (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ||
1451 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))) {
1452 rs->ti->error = "sync/nosync and rebuild are mutually exclusive";
1456 if (write_mostly >= rs->md.raid_disks) {
1457 rs->ti->error = "Can't set all raid1 devices to write_mostly";
1461 if (validate_region_size(rs, region_size))
1464 if (rs->md.chunk_sectors)
1465 max_io_len = rs->md.chunk_sectors;
1467 max_io_len = region_size;
1469 if (dm_set_target_max_io_len(rs->ti, max_io_len))
1472 if (rt_is_raid10(rt)) {
1473 if (raid10_copies > rs->md.raid_disks) {
1474 rs->ti->error = "Not enough devices to satisfy specification";
1478 rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
1479 if (rs->md.new_layout < 0) {
1480 rs->ti->error = "Error getting raid10 format";
1481 return rs->md.new_layout;
1484 rt = get_raid_type_by_ll(10, rs->md.new_layout);
1486 rs->ti->error = "Failed to recognize new raid10 layout";
1490 if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1491 rt->algorithm == ALGORITHM_RAID10_NEAR) &&
1492 test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1493 rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1498 rs->raid10_copies = raid10_copies;
1500 /* Assume there are no metadata devices until the drives are parsed */
1501 rs->md.persistent = 0;
1502 rs->md.external = 1;
1504 /* Check, if any invalid ctr arguments have been passed in for the raid level */
1505 return rs_check_for_valid_flags(rs);
1508 /* Set raid4/5/6 cache size */
1509 static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1512 struct r5conf *conf;
1513 struct mddev *mddev = &rs->md;
1514 uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1515 uint32_t nr_stripes = rs->stripe_cache_entries;
1517 if (!rt_is_raid456(rs->raid_type)) {
1518 rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1522 if (nr_stripes < min_stripes) {
1523 DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1524 nr_stripes, min_stripes);
1525 nr_stripes = min_stripes;
1528 conf = mddev->private;
1530 rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1534 /* Try setting number of stripes in raid456 stripe cache */
1535 if (conf->min_nr_stripes != nr_stripes) {
1536 r = raid5_set_cache_size(mddev, nr_stripes);
1538 rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1542 DMINFO("%u stripe cache entries", nr_stripes);
1548 /* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1549 static unsigned int mddev_data_stripes(struct raid_set *rs)
1551 return rs->md.raid_disks - rs->raid_type->parity_devs;
1554 /* Return # of data stripes of @rs (i.e. as of ctr) */
1555 static unsigned int rs_data_stripes(struct raid_set *rs)
1557 return rs->raid_disks - rs->raid_type->parity_devs;
1561 * Retrieve rdev->sectors from any valid raid device of @rs
1562 * to allow userpace to pass in arbitray "- -" device tupples.
1564 static sector_t __rdev_sectors(struct raid_set *rs)
1568 for (i = 0; i < rs->md.raid_disks; i++) {
1569 struct md_rdev *rdev = &rs->dev[i].rdev;
1571 if (!test_bit(Journal, &rdev->flags) &&
1572 rdev->bdev && rdev->sectors)
1573 return rdev->sectors;
1579 /* Calculate the sectors per device and per array used for @rs */
1580 static int rs_set_dev_and_array_sectors(struct raid_set *rs, bool use_mddev)
1583 unsigned int data_stripes;
1584 struct mddev *mddev = &rs->md;
1585 struct md_rdev *rdev;
1586 sector_t array_sectors = rs->ti->len, dev_sectors = rs->ti->len;
1589 delta_disks = mddev->delta_disks;
1590 data_stripes = mddev_data_stripes(rs);
1592 delta_disks = rs->delta_disks;
1593 data_stripes = rs_data_stripes(rs);
1596 /* Special raid1 case w/o delta_disks support (yet) */
1597 if (rt_is_raid1(rs->raid_type))
1599 else if (rt_is_raid10(rs->raid_type)) {
1600 if (rs->raid10_copies < 2 ||
1602 rs->ti->error = "Bogus raid10 data copies or delta disks";
1606 dev_sectors *= rs->raid10_copies;
1607 if (sector_div(dev_sectors, data_stripes))
1610 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1611 if (sector_div(array_sectors, rs->raid10_copies))
1614 } else if (sector_div(dev_sectors, data_stripes))
1618 /* Striped layouts */
1619 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1621 rdev_for_each(rdev, mddev)
1622 if (!test_bit(Journal, &rdev->flags))
1623 rdev->sectors = dev_sectors;
1625 mddev->array_sectors = array_sectors;
1626 mddev->dev_sectors = dev_sectors;
1630 rs->ti->error = "Target length not divisible by number of data devices";
1634 /* Setup recovery on @rs */
1635 static void __rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1637 /* raid0 does not recover */
1638 if (rs_is_raid0(rs))
1639 rs->md.recovery_cp = MaxSector;
1641 * A raid6 set has to be recovered either
1642 * completely or for the grown part to
1643 * ensure proper parity and Q-Syndrome
1645 else if (rs_is_raid6(rs))
1646 rs->md.recovery_cp = dev_sectors;
1648 * Other raid set types may skip recovery
1649 * depending on the 'nosync' flag.
1652 rs->md.recovery_cp = test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)
1653 ? MaxSector : dev_sectors;
1656 /* Setup recovery on @rs based on raid type, device size and 'nosync' flag */
1657 static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1660 /* New raid set or 'sync' flag provided */
1661 __rs_setup_recovery(rs, 0);
1662 else if (dev_sectors == MaxSector)
1663 /* Prevent recovery */
1664 __rs_setup_recovery(rs, MaxSector);
1665 else if (__rdev_sectors(rs) < dev_sectors)
1666 /* Grown raid set */
1667 __rs_setup_recovery(rs, __rdev_sectors(rs));
1669 __rs_setup_recovery(rs, MaxSector);
1672 static void do_table_event(struct work_struct *ws)
1674 struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1676 smp_rmb(); /* Make sure we access most actual mddev properties */
1677 if (!rs_is_reshaping(rs))
1678 rs_set_capacity(rs);
1679 dm_table_event(rs->ti->table);
1682 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
1684 struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
1686 return mddev_congested(&rs->md, bits);
1690 * Make sure a valid takover (level switch) is being requested on @rs
1692 * Conversions of raid sets from one MD personality to another
1693 * have to conform to restrictions which are enforced here.
1695 static int rs_check_takeover(struct raid_set *rs)
1697 struct mddev *mddev = &rs->md;
1698 unsigned int near_copies;
1700 if (rs->md.degraded) {
1701 rs->ti->error = "Can't takeover degraded raid set";
1705 if (rs_is_reshaping(rs)) {
1706 rs->ti->error = "Can't takeover reshaping raid set";
1710 switch (mddev->level) {
1712 /* raid0 -> raid1/5 with one disk */
1713 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1714 mddev->raid_disks == 1)
1717 /* raid0 -> raid10 */
1718 if (mddev->new_level == 10 &&
1719 !(rs->raid_disks % mddev->raid_disks))
1722 /* raid0 with multiple disks -> raid4/5/6 */
1723 if (__within_range(mddev->new_level, 4, 6) &&
1724 mddev->new_layout == ALGORITHM_PARITY_N &&
1725 mddev->raid_disks > 1)
1731 /* Can't takeover raid10_offset! */
1732 if (__is_raid10_offset(mddev->layout))
1735 near_copies = __raid10_near_copies(mddev->layout);
1737 /* raid10* -> raid0 */
1738 if (mddev->new_level == 0) {
1739 /* Can takeover raid10_near with raid disks divisable by data copies! */
1740 if (near_copies > 1 &&
1741 !(mddev->raid_disks % near_copies)) {
1742 mddev->raid_disks /= near_copies;
1743 mddev->delta_disks = mddev->raid_disks;
1747 /* Can takeover raid10_far */
1748 if (near_copies == 1 &&
1749 __raid10_far_copies(mddev->layout) > 1)
1755 /* raid10_{near,far} -> raid1 */
1756 if (mddev->new_level == 1 &&
1757 max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
1760 /* raid10_{near,far} with 2 disks -> raid4/5 */
1761 if (__within_range(mddev->new_level, 4, 5) &&
1762 mddev->raid_disks == 2)
1767 /* raid1 with 2 disks -> raid4/5 */
1768 if (__within_range(mddev->new_level, 4, 5) &&
1769 mddev->raid_disks == 2) {
1770 mddev->degraded = 1;
1774 /* raid1 -> raid0 */
1775 if (mddev->new_level == 0 &&
1776 mddev->raid_disks == 1)
1779 /* raid1 -> raid10 */
1780 if (mddev->new_level == 10)
1785 /* raid4 -> raid0 */
1786 if (mddev->new_level == 0)
1789 /* raid4 -> raid1/5 with 2 disks */
1790 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1791 mddev->raid_disks == 2)
1794 /* raid4 -> raid5/6 with parity N */
1795 if (__within_range(mddev->new_level, 5, 6) &&
1796 mddev->layout == ALGORITHM_PARITY_N)
1801 /* raid5 with parity N -> raid0 */
1802 if (mddev->new_level == 0 &&
1803 mddev->layout == ALGORITHM_PARITY_N)
1806 /* raid5 with parity N -> raid4 */
1807 if (mddev->new_level == 4 &&
1808 mddev->layout == ALGORITHM_PARITY_N)
1811 /* raid5 with 2 disks -> raid1/4/10 */
1812 if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1813 mddev->raid_disks == 2)
1816 /* raid5_* -> raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
1817 if (mddev->new_level == 6 &&
1818 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1819 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
1824 /* raid6 with parity N -> raid0 */
1825 if (mddev->new_level == 0 &&
1826 mddev->layout == ALGORITHM_PARITY_N)
1829 /* raid6 with parity N -> raid4 */
1830 if (mddev->new_level == 4 &&
1831 mddev->layout == ALGORITHM_PARITY_N)
1834 /* raid6_*_n with Q-Syndrome N -> raid5_* */
1835 if (mddev->new_level == 5 &&
1836 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1837 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
1844 rs->ti->error = "takeover not possible";
1848 /* True if @rs requested to be taken over */
1849 static bool rs_takeover_requested(struct raid_set *rs)
1851 return rs->md.new_level != rs->md.level;
1854 /* True if @rs is requested to reshape by ctr */
1855 static bool rs_reshape_requested(struct raid_set *rs)
1858 struct mddev *mddev = &rs->md;
1860 if (rs_takeover_requested(rs))
1866 change = mddev->new_layout != mddev->layout ||
1867 mddev->new_chunk_sectors != mddev->chunk_sectors ||
1870 /* Historical case to support raid1 reshape without delta disks */
1871 if (mddev->level == 1) {
1872 if (rs->delta_disks)
1873 return !!rs->delta_disks;
1876 mddev->raid_disks != rs->raid_disks;
1879 if (mddev->level == 10)
1881 !__is_raid10_far(mddev->new_layout) &&
1882 rs->delta_disks >= 0;
1888 #define FEATURE_FLAG_SUPPORTS_V190 0x1 /* Supports extended superblock */
1890 /* State flags for sb->flags */
1891 #define SB_FLAG_RESHAPE_ACTIVE 0x1
1892 #define SB_FLAG_RESHAPE_BACKWARDS 0x2
1895 * This structure is never routinely used by userspace, unlike md superblocks.
1896 * Devices with this superblock should only ever be accessed via device-mapper.
1898 #define DM_RAID_MAGIC 0x64526D44
1899 struct dm_raid_superblock {
1900 __le32 magic; /* "DmRd" */
1901 __le32 compat_features; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1903 __le32 num_devices; /* Number of devices in this raid set. (Max 64) */
1904 __le32 array_position; /* The position of this drive in the raid set */
1906 __le64 events; /* Incremented by md when superblock updated */
1907 __le64 failed_devices; /* Pre 1.9.0 part of bit field of devices to */
1908 /* indicate failures (see extension below) */
1911 * This offset tracks the progress of the repair or replacement of
1912 * an individual drive.
1914 __le64 disk_recovery_offset;
1917 * This offset tracks the progress of the initial raid set
1918 * synchronisation/parity calculation.
1920 __le64 array_resync_offset;
1923 * raid characteristics
1927 __le32 stripe_sectors;
1929 /********************************************************************
1930 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1932 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
1935 __le32 flags; /* Flags defining array states for reshaping */
1938 * This offset tracks the progress of a raid
1939 * set reshape in order to be able to restart it
1941 __le64 reshape_position;
1944 * These define the properties of the array in case of an interrupted reshape
1948 __le32 new_stripe_sectors;
1951 __le64 array_sectors; /* Array size in sectors */
1954 * Sector offsets to data on devices (reshaping).
1955 * Needed to support out of place reshaping, thus
1956 * not writing over any stripes whilst converting
1957 * them from old to new layout
1960 __le64 new_data_offset;
1962 __le64 sectors; /* Used device size in sectors */
1965 * Additonal Bit field of devices indicating failures to support
1966 * up to 256 devices with the 1.9.0 on-disk metadata format
1968 __le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
1970 __le32 incompat_features; /* Used to indicate any incompatible features */
1972 /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
1976 * Check for reshape constraints on raid set @rs:
1978 * - reshape function non-existent
1980 * - ongoing recovery
1983 * Returns 0 if none or -EPERM if given constraint
1984 * and error message reference in @errmsg
1986 static int rs_check_reshape(struct raid_set *rs)
1988 struct mddev *mddev = &rs->md;
1990 if (!mddev->pers || !mddev->pers->check_reshape)
1991 rs->ti->error = "Reshape not supported";
1992 else if (mddev->degraded)
1993 rs->ti->error = "Can't reshape degraded raid set";
1994 else if (rs_is_recovering(rs))
1995 rs->ti->error = "Convert request on recovering raid set prohibited";
1996 else if (rs_is_reshaping(rs))
1997 rs->ti->error = "raid set already reshaping!";
1998 else if (!(rs_is_raid1(rs) || rs_is_raid10(rs) || rs_is_raid456(rs)))
1999 rs->ti->error = "Reshaping only supported for raid1/4/5/6/10";
2006 static int read_disk_sb(struct md_rdev *rdev, int size, bool force_reload)
2008 BUG_ON(!rdev->sb_page);
2010 if (rdev->sb_loaded && !force_reload)
2013 rdev->sb_loaded = 0;
2015 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) {
2016 DMERR("Failed to read superblock of device at position %d",
2018 md_error(rdev->mddev, rdev);
2019 set_bit(Faulty, &rdev->flags);
2023 rdev->sb_loaded = 1;
2028 static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2030 failed_devices[0] = le64_to_cpu(sb->failed_devices);
2031 memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
2033 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2034 int i = ARRAY_SIZE(sb->extended_failed_devices);
2037 failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
2041 static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2043 int i = ARRAY_SIZE(sb->extended_failed_devices);
2045 sb->failed_devices = cpu_to_le64(failed_devices[0]);
2047 sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
2051 * Synchronize the superblock members with the raid set properties
2053 * All superblock data is little endian.
2055 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
2057 bool update_failed_devices = false;
2059 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2060 struct dm_raid_superblock *sb;
2061 struct raid_set *rs = container_of(mddev, struct raid_set, md);
2063 /* No metadata device, no superblock */
2064 if (!rdev->meta_bdev)
2067 BUG_ON(!rdev->sb_page);
2069 sb = page_address(rdev->sb_page);
2071 sb_retrieve_failed_devices(sb, failed_devices);
2073 for (i = 0; i < rs->raid_disks; i++)
2074 if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
2075 update_failed_devices = true;
2076 set_bit(i, (void *) failed_devices);
2079 if (update_failed_devices)
2080 sb_update_failed_devices(sb, failed_devices);
2082 sb->magic = cpu_to_le32(DM_RAID_MAGIC);
2083 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2085 sb->num_devices = cpu_to_le32(mddev->raid_disks);
2086 sb->array_position = cpu_to_le32(rdev->raid_disk);
2088 sb->events = cpu_to_le64(mddev->events);
2090 sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
2091 sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
2093 sb->level = cpu_to_le32(mddev->level);
2094 sb->layout = cpu_to_le32(mddev->layout);
2095 sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
2097 /********************************************************************
2098 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
2100 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
2102 sb->new_level = cpu_to_le32(mddev->new_level);
2103 sb->new_layout = cpu_to_le32(mddev->new_layout);
2104 sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
2106 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2108 smp_rmb(); /* Make sure we access most recent reshape position */
2109 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2110 if (le64_to_cpu(sb->reshape_position) != MaxSector) {
2111 /* Flag ongoing reshape */
2112 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
2114 if (mddev->delta_disks < 0 || mddev->reshape_backwards)
2115 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
2117 /* Clear reshape flags */
2118 sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
2121 sb->array_sectors = cpu_to_le64(mddev->array_sectors);
2122 sb->data_offset = cpu_to_le64(rdev->data_offset);
2123 sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
2124 sb->sectors = cpu_to_le64(rdev->sectors);
2125 sb->incompat_features = cpu_to_le32(0);
2127 /* Zero out the rest of the payload after the size of the superblock */
2128 memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
2134 * This function creates a superblock if one is not found on the device
2135 * and will decide which superblock to use if there's a choice.
2137 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
2139 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
2142 struct dm_raid_superblock *sb;
2143 struct dm_raid_superblock *refsb;
2144 uint64_t events_sb, events_refsb;
2146 r = read_disk_sb(rdev, rdev->sb_size, false);
2150 sb = page_address(rdev->sb_page);
2153 * Two cases that we want to write new superblocks and rebuild:
2154 * 1) New device (no matching magic number)
2155 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2157 if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
2158 (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
2159 super_sync(rdev->mddev, rdev);
2161 set_bit(FirstUse, &rdev->flags);
2162 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2164 /* Force writing of superblocks to disk */
2165 set_bit(MD_SB_CHANGE_DEVS, &rdev->mddev->sb_flags);
2167 /* Any superblock is better than none, choose that if given */
2168 return refdev ? 0 : 1;
2174 events_sb = le64_to_cpu(sb->events);
2176 refsb = page_address(refdev->sb_page);
2177 events_refsb = le64_to_cpu(refsb->events);
2179 return (events_sb > events_refsb) ? 1 : 0;
2182 static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
2186 struct mddev *mddev = &rs->md;
2188 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2189 struct dm_raid_superblock *sb;
2190 uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
2192 struct dm_raid_superblock *sb2;
2194 sb = page_address(rdev->sb_page);
2195 events_sb = le64_to_cpu(sb->events);
2198 * Initialise to 1 if this is a new superblock.
2200 mddev->events = events_sb ? : 1;
2202 mddev->reshape_position = MaxSector;
2204 mddev->raid_disks = le32_to_cpu(sb->num_devices);
2205 mddev->level = le32_to_cpu(sb->level);
2206 mddev->layout = le32_to_cpu(sb->layout);
2207 mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
2210 * Reshaping is supported, e.g. reshape_position is valid
2211 * in superblock and superblock content is authoritative.
2213 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2214 /* Superblock is authoritative wrt given raid set layout! */
2215 mddev->new_level = le32_to_cpu(sb->new_level);
2216 mddev->new_layout = le32_to_cpu(sb->new_layout);
2217 mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
2218 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
2219 mddev->array_sectors = le64_to_cpu(sb->array_sectors);
2221 /* raid was reshaping and got interrupted */
2222 if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
2223 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
2224 DMERR("Reshape requested but raid set is still reshaping");
2228 if (mddev->delta_disks < 0 ||
2229 (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
2230 mddev->reshape_backwards = 1;
2232 mddev->reshape_backwards = 0;
2234 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2235 rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2240 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2242 struct raid_type *rt_cur = get_raid_type_by_ll(mddev->level, mddev->layout);
2243 struct raid_type *rt_new = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
2245 if (rs_takeover_requested(rs)) {
2246 if (rt_cur && rt_new)
2247 DMERR("Takeover raid sets from %s to %s not yet supported by metadata. (raid level change)",
2248 rt_cur->name, rt_new->name);
2250 DMERR("Takeover raid sets not yet supported by metadata. (raid level change)");
2252 } else if (rs_reshape_requested(rs)) {
2253 DMERR("Reshaping raid sets not yet supported by metadata. (raid layout change keeping level)");
2254 if (mddev->layout != mddev->new_layout) {
2255 if (rt_cur && rt_new)
2256 DMERR(" current layout %s vs new layout %s",
2257 rt_cur->name, rt_new->name);
2259 DMERR(" current layout 0x%X vs new layout 0x%X",
2260 le32_to_cpu(sb->layout), mddev->new_layout);
2262 if (mddev->chunk_sectors != mddev->new_chunk_sectors)
2263 DMERR(" current stripe sectors %u vs new stripe sectors %u",
2264 mddev->chunk_sectors, mddev->new_chunk_sectors);
2265 if (rs->delta_disks)
2266 DMERR(" current %u disks vs new %u disks",
2267 mddev->raid_disks, mddev->raid_disks + rs->delta_disks);
2268 if (rs_is_raid10(rs)) {
2269 DMERR(" Old layout: %s w/ %u copies",
2270 raid10_md_layout_to_format(mddev->layout),
2271 raid10_md_layout_to_copies(mddev->layout));
2272 DMERR(" New layout: %s w/ %u copies",
2273 raid10_md_layout_to_format(mddev->new_layout),
2274 raid10_md_layout_to_copies(mddev->new_layout));
2279 DMINFO("Discovered old metadata format; upgrading to extended metadata format");
2282 if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
2283 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2286 * During load, we set FirstUse if a new superblock was written.
2287 * There are two reasons we might not have a superblock:
2288 * 1) The raid set is brand new - in which case, all of the
2289 * devices must have their In_sync bit set. Also,
2290 * recovery_cp must be 0, unless forced.
2291 * 2) This is a new device being added to an old raid set
2292 * and the new device needs to be rebuilt - in which
2293 * case the In_sync bit will /not/ be set and
2294 * recovery_cp must be MaxSector.
2295 * 3) This is/are a new device(s) being added to an old
2296 * raid set during takeover to a higher raid level
2297 * to provide capacity for redundancy or during reshape
2298 * to add capacity to grow the raid set.
2301 rdev_for_each(r, mddev) {
2302 if (test_bit(Journal, &rdev->flags))
2305 if (test_bit(FirstUse, &r->flags))
2308 if (!test_bit(In_sync, &r->flags)) {
2309 DMINFO("Device %d specified for rebuild; clearing superblock",
2313 if (test_bit(FirstUse, &r->flags))
2320 if (new_devs == rs->raid_disks || !rebuilds) {
2321 /* Replace a broken device */
2322 if (new_devs == 1 && !rs->delta_disks)
2324 if (new_devs == rs->raid_disks) {
2325 DMINFO("Superblocks created for new raid set");
2326 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2327 } else if (new_devs != rebuilds &&
2328 new_devs != rs->delta_disks) {
2329 DMERR("New device injected into existing raid set without "
2330 "'delta_disks' or 'rebuild' parameter specified");
2333 } else if (new_devs && new_devs != rebuilds) {
2334 DMERR("%u 'rebuild' devices cannot be injected into"
2335 " a raid set with %u other first-time devices",
2336 rebuilds, new_devs);
2338 } else if (rebuilds) {
2339 if (rebuild_and_new && rebuilds != rebuild_and_new) {
2340 DMERR("new device%s provided without 'rebuild'",
2341 new_devs > 1 ? "s" : "");
2343 } else if (rs_is_recovering(rs)) {
2344 DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2345 (unsigned long long) mddev->recovery_cp);
2347 } else if (rs_is_reshaping(rs)) {
2348 DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2349 (unsigned long long) mddev->reshape_position);
2355 * Now we set the Faulty bit for those devices that are
2356 * recorded in the superblock as failed.
2358 sb_retrieve_failed_devices(sb, failed_devices);
2359 rdev_for_each(r, mddev) {
2360 if (test_bit(Journal, &rdev->flags) ||
2363 sb2 = page_address(r->sb_page);
2364 sb2->failed_devices = 0;
2365 memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
2368 * Check for any device re-ordering.
2370 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2371 role = le32_to_cpu(sb2->array_position);
2375 if (role != r->raid_disk) {
2376 if (rs_is_raid10(rs) && __is_raid10_near(mddev->layout)) {
2377 if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
2378 rs->raid_disks % rs->raid10_copies) {
2380 "Cannot change raid10 near set to odd # of devices!";
2384 sb2->array_position = cpu_to_le32(r->raid_disk);
2386 } else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
2387 !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2388 !rt_is_raid1(rs->raid_type)) {
2389 rs->ti->error = "Cannot change device positions in raid set";
2393 DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
2397 * Partial recovery is performed on
2398 * returning failed devices.
2400 if (test_bit(role, (void *) failed_devices))
2401 set_bit(Faulty, &r->flags);
2408 static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
2410 struct mddev *mddev = &rs->md;
2411 struct dm_raid_superblock *sb;
2413 if (rs_is_raid0(rs) || !rdev->sb_page || rdev->raid_disk < 0)
2416 sb = page_address(rdev->sb_page);
2419 * If mddev->events is not set, we know we have not yet initialized
2422 if (!mddev->events && super_init_validation(rs, rdev))
2425 if (le32_to_cpu(sb->compat_features) &&
2426 le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
2427 rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2431 if (sb->incompat_features) {
2432 rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
2436 /* Enable bitmap creation for RAID levels != 0 */
2437 mddev->bitmap_info.offset = rt_is_raid0(rs->raid_type) ? 0 : to_sector(4096);
2438 mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2440 if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2442 * Retrieve rdev size stored in superblock to be prepared for shrink.
2443 * Check extended superblock members are present otherwise the size
2446 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190)
2447 rdev->sectors = le64_to_cpu(sb->sectors);
2449 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
2450 if (rdev->recovery_offset == MaxSector)
2451 set_bit(In_sync, &rdev->flags);
2453 * If no reshape in progress -> we're recovering single
2454 * disk(s) and have to set the device(s) to out-of-sync
2456 else if (!rs_is_reshaping(rs))
2457 clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
2461 * If a device comes back, set it as not In_sync and no longer faulty.
2463 if (test_and_clear_bit(Faulty, &rdev->flags)) {
2464 rdev->recovery_offset = 0;
2465 clear_bit(In_sync, &rdev->flags);
2466 rdev->saved_raid_disk = rdev->raid_disk;
2469 /* Reshape support -> restore repective data offsets */
2470 rdev->data_offset = le64_to_cpu(sb->data_offset);
2471 rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
2477 * Analyse superblocks and select the freshest.
2479 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2482 struct md_rdev *rdev, *freshest;
2483 struct mddev *mddev = &rs->md;
2486 rdev_for_each(rdev, mddev) {
2487 if (test_bit(Journal, &rdev->flags))
2490 if (!rdev->meta_bdev)
2493 /* Set superblock offset/size for metadata device. */
2495 rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
2496 if (rdev->sb_size < sizeof(struct dm_raid_superblock) || rdev->sb_size > PAGE_SIZE) {
2497 DMERR("superblock size of a logical block is no longer valid");
2502 * Skipping super_load due to CTR_FLAG_SYNC will cause
2503 * the array to undergo initialization again as
2504 * though it were new. This is the intended effect
2505 * of the "sync" directive.
2507 * With reshaping capability added, we must ensure that
2508 * that the "sync" directive is disallowed during the reshape.
2510 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2513 r = super_load(rdev, freshest);
2522 /* This is a failure to read the superblock from the metadata device. */
2524 * We have to keep any raid0 data/metadata device pairs or
2525 * the MD raid0 personality will fail to start the array.
2527 if (rs_is_raid0(rs))
2531 * We keep the dm_devs to be able to emit the device tuple
2532 * properly on the table line in raid_status() (rather than
2533 * mistakenly acting as if '- -' got passed into the constructor).
2535 * The rdev has to stay on the same_set list to allow for
2536 * the attempt to restore faulty devices on second resume.
2538 rdev->raid_disk = rdev->saved_raid_disk = -1;
2547 * Validation of the freshest device provides the source of
2548 * validation for the remaining devices.
2550 rs->ti->error = "Unable to assemble array: Invalid superblocks";
2551 if (super_validate(rs, freshest))
2554 if (validate_raid_redundancy(rs)) {
2555 rs->ti->error = "Insufficient redundancy to activate array";
2559 rdev_for_each(rdev, mddev)
2560 if (!test_bit(Journal, &rdev->flags) &&
2562 super_validate(rs, rdev))
2568 * Adjust data_offset and new_data_offset on all disk members of @rs
2569 * for out of place reshaping if requested by contructor
2571 * We need free space at the beginning of each raid disk for forward
2572 * and at the end for backward reshapes which userspace has to provide
2573 * via remapping/reordering of space.
2575 static int rs_adjust_data_offsets(struct raid_set *rs)
2577 sector_t data_offset = 0, new_data_offset = 0;
2578 struct md_rdev *rdev;
2580 /* Constructor did not request data offset change */
2581 if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2582 if (!rs_is_reshapable(rs))
2588 /* HM FIXME: get InSync raid_dev? */
2589 rdev = &rs->dev[0].rdev;
2591 if (rs->delta_disks < 0) {
2593 * Removing disks (reshaping backwards):
2595 * - before reshape: data is at offset 0 and free space
2596 * is at end of each component LV
2598 * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2601 new_data_offset = rs->data_offset;
2603 } else if (rs->delta_disks > 0) {
2605 * Adding disks (reshaping forwards):
2607 * - before reshape: data is at offset rs->data_offset != 0 and
2608 * free space is at begin of each component LV
2610 * - after reshape: data is at offset 0 on each component LV
2612 data_offset = rs->data_offset;
2613 new_data_offset = 0;
2617 * User space passes in 0 for data offset after having removed reshape space
2619 * - or - (data offset != 0)
2621 * Changing RAID layout or chunk size -> toggle offsets
2623 * - before reshape: data is at offset rs->data_offset 0 and
2624 * free space is at end of each component LV
2626 * data is at offset rs->data_offset != 0 and
2627 * free space is at begin of each component LV
2629 * - after reshape: data is at offset 0 if it was at offset != 0
2630 * or at offset != 0 if it was at offset 0
2631 * on each component LV
2634 data_offset = rs->data_offset ? rdev->data_offset : 0;
2635 new_data_offset = data_offset ? 0 : rs->data_offset;
2636 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2640 * Make sure we got a minimum amount of free sectors per device
2642 if (rs->data_offset &&
2643 to_sector(i_size_read(rdev->bdev->bd_inode)) - rs->md.dev_sectors < MIN_FREE_RESHAPE_SPACE) {
2644 rs->ti->error = data_offset ? "No space for forward reshape" :
2645 "No space for backward reshape";
2650 * Raise recovery_cp in case data_offset != 0 to
2651 * avoid false recovery positives in the constructor.
2653 if (rs->md.recovery_cp < rs->md.dev_sectors)
2654 rs->md.recovery_cp += rs->dev[0].rdev.data_offset;
2656 /* Adjust data offsets on all rdevs but on any raid4/5/6 journal device */
2657 rdev_for_each(rdev, &rs->md) {
2658 if (!test_bit(Journal, &rdev->flags)) {
2659 rdev->data_offset = data_offset;
2660 rdev->new_data_offset = new_data_offset;
2667 /* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2668 static void __reorder_raid_disk_indexes(struct raid_set *rs)
2671 struct md_rdev *rdev;
2673 rdev_for_each(rdev, &rs->md) {
2674 if (!test_bit(Journal, &rdev->flags)) {
2675 rdev->raid_disk = i++;
2676 rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2682 * Setup @rs for takeover by a different raid level
2684 static int rs_setup_takeover(struct raid_set *rs)
2686 struct mddev *mddev = &rs->md;
2687 struct md_rdev *rdev;
2688 unsigned int d = mddev->raid_disks = rs->raid_disks;
2689 sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2691 if (rt_is_raid10(rs->raid_type)) {
2692 if (mddev->level == 0) {
2693 /* Userpace reordered disks -> adjust raid_disk indexes */
2694 __reorder_raid_disk_indexes(rs);
2696 /* raid0 -> raid10_far layout */
2697 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2699 } else if (mddev->level == 1)
2700 /* raid1 -> raid10_near layout */
2701 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2708 clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2709 mddev->recovery_cp = MaxSector;
2712 rdev = &rs->dev[d].rdev;
2714 if (test_bit(d, (void *) rs->rebuild_disks)) {
2715 clear_bit(In_sync, &rdev->flags);
2716 clear_bit(Faulty, &rdev->flags);
2717 mddev->recovery_cp = rdev->recovery_offset = 0;
2718 /* Bitmap has to be created when we do an "up" takeover */
2719 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2722 rdev->new_data_offset = new_data_offset;
2728 /* Prepare @rs for reshape */
2729 static int rs_prepare_reshape(struct raid_set *rs)
2732 struct mddev *mddev = &rs->md;
2734 if (rs_is_raid10(rs)) {
2735 if (rs->raid_disks != mddev->raid_disks &&
2736 __is_raid10_near(mddev->layout) &&
2737 rs->raid10_copies &&
2738 rs->raid10_copies != __raid10_near_copies(mddev->layout)) {
2740 * raid disk have to be multiple of data copies to allow this conversion,
2742 * This is actually not a reshape it is a
2743 * rebuild of any additional mirrors per group
2745 if (rs->raid_disks % rs->raid10_copies) {
2746 rs->ti->error = "Can't reshape raid10 mirror groups";
2750 /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2751 __reorder_raid_disk_indexes(rs);
2752 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2754 mddev->new_layout = mddev->layout;
2759 } else if (rs_is_raid456(rs))
2762 else if (rs_is_raid1(rs)) {
2763 if (rs->delta_disks) {
2764 /* Process raid1 via delta_disks */
2765 mddev->degraded = rs->delta_disks < 0 ? -rs->delta_disks : rs->delta_disks;
2768 /* Process raid1 without delta_disks */
2769 mddev->raid_disks = rs->raid_disks;
2773 rs->ti->error = "Called with bogus raid type";
2778 set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2779 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2780 } else if (mddev->raid_disks < rs->raid_disks)
2781 /* Create new superblocks and bitmaps, if any new disks */
2782 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2787 /* Get reshape sectors from data_offsets or raid set */
2788 static sector_t _get_reshape_sectors(struct raid_set *rs)
2790 struct md_rdev *rdev;
2791 sector_t reshape_sectors = 0;
2793 rdev_for_each(rdev, &rs->md)
2794 if (!test_bit(Journal, &rdev->flags)) {
2795 reshape_sectors = (rdev->data_offset > rdev->new_data_offset) ?
2796 rdev->data_offset - rdev->new_data_offset :
2797 rdev->new_data_offset - rdev->data_offset;
2801 return max(reshape_sectors, (sector_t) rs->data_offset);
2806 * - change raid layout
2807 * - change chunk size
2811 static int rs_setup_reshape(struct raid_set *rs)
2814 unsigned int cur_raid_devs, d;
2815 sector_t reshape_sectors = _get_reshape_sectors(rs);
2816 struct mddev *mddev = &rs->md;
2817 struct md_rdev *rdev;
2819 mddev->delta_disks = rs->delta_disks;
2820 cur_raid_devs = mddev->raid_disks;
2822 /* Ignore impossible layout change whilst adding/removing disks */
2823 if (mddev->delta_disks &&
2824 mddev->layout != mddev->new_layout) {
2825 DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2826 mddev->new_layout = mddev->layout;
2830 * Adjust array size:
2832 * - in case of adding disk(s), array size has
2833 * to grow after the disk adding reshape,
2834 * which'll hapen in the event handler;
2835 * reshape will happen forward, so space has to
2836 * be available at the beginning of each disk
2838 * - in case of removing disk(s), array size
2839 * has to shrink before starting the reshape,
2840 * which'll happen here;
2841 * reshape will happen backward, so space has to
2842 * be available at the end of each disk
2844 * - data_offset and new_data_offset are
2845 * adjusted for aforementioned out of place
2846 * reshaping based on userspace passing in
2847 * the "data_offset <sectors>" key/value
2848 * pair via the constructor
2852 if (rs->delta_disks > 0) {
2853 /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2854 for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2855 rdev = &rs->dev[d].rdev;
2856 clear_bit(In_sync, &rdev->flags);
2859 * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2860 * by md, which'll store that erroneously in the superblock on reshape
2862 rdev->saved_raid_disk = -1;
2863 rdev->raid_disk = d;
2865 rdev->sectors = mddev->dev_sectors;
2866 rdev->recovery_offset = rs_is_raid1(rs) ? 0 : MaxSector;
2869 mddev->reshape_backwards = 0; /* adding disk(s) -> forward reshape */
2871 /* Remove disk(s) */
2872 } else if (rs->delta_disks < 0) {
2873 r = rs_set_dev_and_array_sectors(rs, true);
2874 mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2876 /* Change layout and/or chunk size */
2879 * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2881 * keeping number of disks and do layout change ->
2883 * toggle reshape_backward depending on data_offset:
2885 * - free space upfront -> reshape forward
2887 * - free space at the end -> reshape backward
2890 * This utilizes free reshape space avoiding the need
2891 * for userspace to move (parts of) LV segments in
2892 * case of layout/chunksize change (for disk
2893 * adding/removing reshape space has to be at
2894 * the proper address (see above with delta_disks):
2896 * add disk(s) -> begin
2897 * remove disk(s)-> end
2899 mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2903 * Adjust device size for forward reshape
2904 * because md_finish_reshape() reduces it.
2906 if (!mddev->reshape_backwards)
2907 rdev_for_each(rdev, &rs->md)
2908 if (!test_bit(Journal, &rdev->flags))
2909 rdev->sectors += reshape_sectors;
2915 * Enable/disable discard support on RAID set depending on
2916 * RAID level and discard properties of underlying RAID members.
2918 static void configure_discard_support(struct raid_set *rs)
2922 struct dm_target *ti = rs->ti;
2925 * XXX: RAID level 4,5,6 require zeroing for safety.
2927 raid456 = (rs->md.level == 4 || rs->md.level == 5 || rs->md.level == 6);
2929 for (i = 0; i < rs->raid_disks; i++) {
2930 struct request_queue *q;
2932 if (!rs->dev[i].rdev.bdev)
2935 q = bdev_get_queue(rs->dev[i].rdev.bdev);
2936 if (!q || !blk_queue_discard(q))
2940 if (!devices_handle_discard_safely) {
2941 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2942 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2949 * RAID1 and RAID10 personalities require bio splitting,
2950 * RAID0/4/5/6 don't and process large discard bios properly.
2952 ti->split_discard_bios = !!(rs->md.level == 1 || rs->md.level == 10);
2953 ti->num_discard_bios = 1;
2957 * Construct a RAID0/1/10/4/5/6 mapping:
2959 * <raid_type> <#raid_params> <raid_params>{0,} \
2960 * <#raid_devs> [<meta_dev1> <dev1>]{1,}
2962 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
2963 * details on possible <raid_params>.
2965 * Userspace is free to initialize the metadata devices, hence the superblocks to
2966 * enforce recreation based on the passed in table parameters.
2969 static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2972 bool resize = false;
2973 struct raid_type *rt;
2974 unsigned int num_raid_params, num_raid_devs;
2975 sector_t calculated_dev_sectors, rdev_sectors, reshape_sectors;
2976 struct raid_set *rs = NULL;
2978 struct rs_layout rs_layout;
2979 struct dm_arg_set as = { argc, argv }, as_nrd;
2980 struct dm_arg _args[] = {
2981 { 0, as.argc, "Cannot understand number of raid parameters" },
2982 { 1, 254, "Cannot understand number of raid devices parameters" }
2985 /* Must have <raid_type> */
2986 arg = dm_shift_arg(&as);
2988 ti->error = "No arguments";
2992 rt = get_raid_type(arg);
2994 ti->error = "Unrecognised raid_type";
2998 /* Must have <#raid_params> */
2999 if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
3002 /* number of raid device tupples <meta_dev data_dev> */
3004 dm_consume_args(&as_nrd, num_raid_params);
3005 _args[1].max = (as_nrd.argc - 1) / 2;
3006 if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
3009 if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
3010 ti->error = "Invalid number of supplied raid devices";
3014 rs = raid_set_alloc(ti, rt, num_raid_devs);
3018 r = parse_raid_params(rs, &as, num_raid_params);
3022 r = parse_dev_params(rs, &as);
3026 rs->md.sync_super = super_sync;
3029 * Calculate ctr requested array and device sizes to allow
3030 * for superblock analysis needing device sizes defined.
3032 * Any existing superblock will overwrite the array and device sizes
3034 r = rs_set_dev_and_array_sectors(rs, false);
3038 calculated_dev_sectors = rs->md.dev_sectors;
3041 * Backup any new raid set level, layout, ...
3042 * requested to be able to compare to superblock
3043 * members for conversion decisions.
3045 rs_config_backup(rs, &rs_layout);
3047 r = analyse_superblocks(ti, rs);
3051 rdev_sectors = __rdev_sectors(rs);
3052 if (!rdev_sectors) {
3053 ti->error = "Invalid rdev size";
3059 reshape_sectors = _get_reshape_sectors(rs);
3060 if (calculated_dev_sectors != rdev_sectors)
3061 resize = calculated_dev_sectors != (reshape_sectors ? rdev_sectors - reshape_sectors : rdev_sectors);
3063 INIT_WORK(&rs->md.event_work, do_table_event);
3065 ti->num_flush_bios = 1;
3067 /* Restore any requested new layout for conversion decision */
3068 rs_config_restore(rs, &rs_layout);
3071 * Now that we have any superblock metadata available,
3072 * check for new, recovering, reshaping, to be taken over,
3073 * to be reshaped or an existing, unchanged raid set to
3076 if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
3077 /* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
3078 if (rs_is_raid6(rs) &&
3079 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
3080 ti->error = "'nosync' not allowed for new raid6 set";
3084 rs_setup_recovery(rs, 0);
3085 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3087 } else if (rs_is_recovering(rs)) {
3088 /* A recovering raid set may be resized */
3089 ; /* skip setup rs */
3090 } else if (rs_is_reshaping(rs)) {
3091 /* Have to reject size change request during reshape */
3093 ti->error = "Can't resize a reshaping raid set";
3098 } else if (rs_takeover_requested(rs)) {
3099 if (rs_is_reshaping(rs)) {
3100 ti->error = "Can't takeover a reshaping raid set";
3105 /* We can't takeover a journaled raid4/5/6 */
3106 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3107 ti->error = "Can't takeover a journaled raid4/5/6 set";
3113 * If a takeover is needed, userspace sets any additional
3114 * devices to rebuild and we can check for a valid request here.
3116 * If acceptible, set the level to the new requested
3117 * one, prohibit requesting recovery, allow the raid
3118 * set to run and store superblocks during resume.
3120 r = rs_check_takeover(rs);
3124 r = rs_setup_takeover(rs);
3128 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3129 /* Takeover ain't recovery, so disable recovery */
3130 rs_setup_recovery(rs, MaxSector);
3132 } else if (rs_reshape_requested(rs)) {
3134 * No need to check for 'ongoing' takeover here, because takeover
3135 * is an instant operation as oposed to an ongoing reshape.
3138 /* We can't reshape a journaled raid4/5/6 */
3139 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3140 ti->error = "Can't reshape a journaled raid4/5/6 set";
3146 * We can only prepare for a reshape here, because the
3147 * raid set needs to run to provide the repective reshape
3148 * check functions via its MD personality instance.
3150 * So do the reshape check after md_run() succeeded.
3152 r = rs_prepare_reshape(rs);
3156 /* Reshaping ain't recovery, so disable recovery */
3157 rs_setup_recovery(rs, MaxSector);
3160 /* May not set recovery when a device rebuild is requested */
3161 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3162 rs_setup_recovery(rs, MaxSector);
3163 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3165 rs_setup_recovery(rs, test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ?
3166 0 : (resize ? calculated_dev_sectors : MaxSector));
3170 /* If constructor requested it, change data and new_data offsets */
3171 r = rs_adjust_data_offsets(rs);
3175 /* Start raid set read-only and assumed clean to change in raid_resume() */
3178 set_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
3180 /* Has to be held on running the array */
3181 mddev_lock_nointr(&rs->md);
3182 r = md_run(&rs->md);
3183 rs->md.in_sync = 0; /* Assume already marked dirty */
3185 ti->error = "Failed to run raid array";
3186 mddev_unlock(&rs->md);
3190 rs->callbacks.congested_fn = raid_is_congested;
3191 dm_table_add_target_callbacks(ti->table, &rs->callbacks);
3193 /* If raid4/5/6 journal mode explictely requested (only possible with journal dev) -> set it */
3194 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
3195 r = r5c_journal_mode_set(&rs->md, rs->journal_dev.mode);
3197 ti->error = "Failed to set raid4/5/6 journal mode";
3198 mddev_unlock(&rs->md);
3199 goto bad_journal_mode_set;
3203 mddev_suspend(&rs->md);
3204 set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags);
3206 /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
3207 if (rs_is_raid456(rs)) {
3208 r = rs_set_raid456_stripe_cache(rs);
3210 goto bad_stripe_cache;
3213 /* Now do an early reshape check */
3214 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3215 r = rs_check_reshape(rs);
3217 goto bad_check_reshape;
3219 /* Restore new, ctr requested layout to perform check */
3220 rs_config_restore(rs, &rs_layout);
3222 if (rs->md.pers->start_reshape) {
3223 r = rs->md.pers->check_reshape(&rs->md);
3225 ti->error = "Reshape check failed";
3226 goto bad_check_reshape;
3231 /* Disable/enable discard support on raid set. */
3232 configure_discard_support(rs);
3234 mddev_unlock(&rs->md);
3237 bad_journal_mode_set:
3247 static void raid_dtr(struct dm_target *ti)
3249 struct raid_set *rs = ti->private;
3251 list_del_init(&rs->callbacks.list);
3256 static int raid_map(struct dm_target *ti, struct bio *bio)
3258 struct raid_set *rs = ti->private;
3259 struct mddev *mddev = &rs->md;
3262 * If we're reshaping to add disk(s)), ti->len and
3263 * mddev->array_sectors will differ during the process
3264 * (ti->len > mddev->array_sectors), so we have to requeue
3265 * bios with addresses > mddev->array_sectors here or
3266 * there will occur accesses past EOD of the component
3267 * data images thus erroring the raid set.
3269 if (unlikely(bio_end_sector(bio) > mddev->array_sectors))
3270 return DM_MAPIO_REQUEUE;
3272 md_handle_request(mddev, bio);
3274 return DM_MAPIO_SUBMITTED;
3277 /* Return string describing the current sync action of @mddev */
3278 static const char *decipher_sync_action(struct mddev *mddev)
3280 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3283 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3284 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3285 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3288 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3289 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3291 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3296 if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3304 * Return status string for @rdev
3306 * Status characters:
3308 * 'D' = Dead/Failed raid set component or raid4/5/6 journal device
3309 * 'a' = Alive but not in-sync raid set component _or_ alive raid4/5/6 'write_back' journal device
3310 * 'A' = Alive and in-sync raid set component _or_ alive raid4/5/6 'write_through' journal device
3311 * '-' = Non-existing device (i.e. uspace passed '- -' into the ctr)
3313 static const char *__raid_dev_status(struct raid_set *rs, struct md_rdev *rdev, bool array_in_sync)
3317 else if (test_bit(Faulty, &rdev->flags))
3319 else if (test_bit(Journal, &rdev->flags))
3320 return (rs->journal_dev.mode == R5C_JOURNAL_MODE_WRITE_THROUGH) ? "A" : "a";
3321 else if (!array_in_sync || !test_bit(In_sync, &rdev->flags))
3327 /* Helper to return resync/reshape progress for @rs and @array_in_sync */
3328 static sector_t rs_get_progress(struct raid_set *rs,
3329 sector_t resync_max_sectors, bool *array_in_sync)
3331 sector_t r, curr_resync_completed;
3332 struct mddev *mddev = &rs->md;
3334 curr_resync_completed = mddev->curr_resync_completed ?: mddev->recovery_cp;
3335 *array_in_sync = false;
3337 if (rs_is_raid0(rs)) {
3338 r = resync_max_sectors;
3339 *array_in_sync = true;
3342 r = mddev->reshape_position;
3344 /* Reshape is relative to the array size */
3345 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
3347 if (r == MaxSector) {
3348 *array_in_sync = true;
3349 r = resync_max_sectors;
3351 /* Got to reverse on backward reshape */
3352 if (mddev->reshape_backwards)
3353 r = mddev->array_sectors - r;
3355 /* Devide by # of data stripes */
3356 sector_div(r, mddev_data_stripes(rs));
3359 /* Sync is relative to the component device size */
3360 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3361 r = curr_resync_completed;
3363 r = mddev->recovery_cp;
3365 if ((r == MaxSector) ||
3366 (test_bit(MD_RECOVERY_DONE, &mddev->recovery) &&
3367 (mddev->curr_resync_completed == resync_max_sectors))) {
3371 *array_in_sync = true;
3372 r = resync_max_sectors;
3373 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
3375 * If "check" or "repair" is occurring, the raid set has
3376 * undergone an initial sync and the health characters
3377 * should not be 'a' anymore.
3379 *array_in_sync = true;
3381 struct md_rdev *rdev;
3384 * The raid set may be doing an initial sync, or it may
3385 * be rebuilding individual components. If all the
3386 * devices are In_sync, then it is the raid set that is
3387 * being initialized.
3389 rdev_for_each(rdev, mddev)
3390 if (!test_bit(Journal, &rdev->flags) &&
3391 !test_bit(In_sync, &rdev->flags))
3392 *array_in_sync = true;
3394 r = 0; /* HM FIXME: TESTME: https://bugzilla.redhat.com/show_bug.cgi?id=1210637 ? */
3402 /* Helper to return @dev name or "-" if !@dev */
3403 static const char *__get_dev_name(struct dm_dev *dev)
3405 return dev ? dev->name : "-";
3408 static void raid_status(struct dm_target *ti, status_type_t type,
3409 unsigned int status_flags, char *result, unsigned int maxlen)
3411 struct raid_set *rs = ti->private;
3412 struct mddev *mddev = &rs->md;
3413 struct r5conf *conf = mddev->private;
3414 int i, max_nr_stripes = conf ? conf->max_nr_stripes : 0;
3416 unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3417 unsigned int sz = 0;
3418 unsigned int rebuild_disks;
3419 unsigned int write_mostly_params = 0;
3420 sector_t progress, resync_max_sectors, resync_mismatches;
3421 const char *sync_action;
3422 struct raid_type *rt;
3425 case STATUSTYPE_INFO:
3426 /* *Should* always succeed */
3427 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3431 DMEMIT("%s %d ", rt->name, mddev->raid_disks);
3433 /* Access most recent mddev properties for status output */
3435 /* Get sensible max sectors even if raid set not yet started */
3436 resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
3437 mddev->resync_max_sectors : mddev->dev_sectors;
3438 progress = rs_get_progress(rs, resync_max_sectors, &array_in_sync);
3439 resync_mismatches = (mddev->last_sync_action && !strcasecmp(mddev->last_sync_action, "check")) ?
3440 atomic64_read(&mddev->resync_mismatches) : 0;
3441 sync_action = decipher_sync_action(&rs->md);
3443 /* HM FIXME: do we want another state char for raid0? It shows 'D'/'A'/'-' now */
3444 for (i = 0; i < rs->raid_disks; i++)
3445 DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev, array_in_sync));
3448 * In-sync/Reshape ratio:
3449 * The in-sync ratio shows the progress of:
3450 * - Initializing the raid set
3451 * - Rebuilding a subset of devices of the raid set
3452 * The user can distinguish between the two by referring
3453 * to the status characters.
3455 * The reshape ratio shows the progress of
3456 * changing the raid layout or the number of
3457 * disks of a raid set
3459 DMEMIT(" %llu/%llu", (unsigned long long) progress,
3460 (unsigned long long) resync_max_sectors);
3466 * See Documentation/device-mapper/dm-raid.txt for
3467 * information on each of these states.
3469 DMEMIT(" %s", sync_action);
3474 * resync_mismatches/mismatch_cnt
3475 * This field shows the number of discrepancies found when
3476 * performing a "check" of the raid set.
3478 DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3483 * data_offset (needed for out of space reshaping)
3484 * This field shows the data offset into the data
3485 * image LV where the first stripes data starts.
3487 * We keep data_offset equal on all raid disks of the set,
3488 * so retrieving it from the first raid disk is sufficient.
3490 DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
3495 DMEMIT(" %s", test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ?
3496 __raid_dev_status(rs, &rs->journal_dev.rdev, 0) : "-");
3499 case STATUSTYPE_TABLE:
3500 /* Report the table line string you would use to construct this raid set */
3502 /* Calculate raid parameter count */
3503 for (i = 0; i < rs->raid_disks; i++)
3504 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3505 write_mostly_params += 2;
3506 rebuild_disks = memweight(rs->rebuild_disks, DISKS_ARRAY_ELEMS * sizeof(*rs->rebuild_disks));
3507 raid_param_cnt += rebuild_disks * 2 +
3508 write_mostly_params +
3509 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3510 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2 +
3511 (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ? 2 : 0) +
3512 (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags) ? 2 : 0);
3514 /* Emit table line */
3515 /* This has to be in the documented order for userspace! */
3516 DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
3517 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
3518 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
3519 if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3520 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
3522 for (i = 0; i < rs->raid_disks; i++)
3523 if (test_bit(rs->dev[i].rdev.raid_disk, (void *) rs->rebuild_disks))
3524 DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD),
3525 rs->dev[i].rdev.raid_disk);
3526 if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3527 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3528 mddev->bitmap_info.daemon_sleep);
3529 if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3530 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3531 mddev->sync_speed_min);
3532 if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3533 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3534 mddev->sync_speed_max);
3535 if (write_mostly_params)
3536 for (i = 0; i < rs->raid_disks; i++)
3537 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3538 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3539 rs->dev[i].rdev.raid_disk);
3540 if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
3541 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
3542 mddev->bitmap_info.max_write_behind);
3543 if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3544 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3546 if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3547 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3548 (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3549 if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3550 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3551 raid10_md_layout_to_copies(mddev->layout));
3552 if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3553 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3554 raid10_md_layout_to_format(mddev->layout));
3555 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3556 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3557 max(rs->delta_disks, mddev->delta_disks));
3558 if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3559 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3560 (unsigned long long) rs->data_offset);
3561 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags))
3562 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV),
3563 __get_dev_name(rs->journal_dev.dev));
3564 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags))
3565 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE),
3566 md_journal_mode_to_dm_raid(rs->journal_dev.mode));
3567 DMEMIT(" %d", rs->raid_disks);
3568 for (i = 0; i < rs->raid_disks; i++)
3569 DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
3570 __get_dev_name(rs->dev[i].data_dev));
3574 static int raid_message(struct dm_target *ti, unsigned int argc, char **argv)
3576 struct raid_set *rs = ti->private;
3577 struct mddev *mddev = &rs->md;
3579 if (!mddev->pers || !mddev->pers->sync_request)
3582 if (!strcasecmp(argv[0], "frozen"))
3583 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3585 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3587 if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
3588 if (mddev->sync_thread) {
3589 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3590 md_reap_sync_thread(mddev);
3592 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3593 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3595 else if (!strcasecmp(argv[0], "resync"))
3596 ; /* MD_RECOVERY_NEEDED set below */
3597 else if (!strcasecmp(argv[0], "recover"))
3598 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3600 if (!strcasecmp(argv[0], "check")) {
3601 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3602 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3603 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3604 } else if (!strcasecmp(argv[0], "repair")) {
3605 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3606 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3610 if (mddev->ro == 2) {
3611 /* A write to sync_action is enough to justify
3612 * canceling read-auto mode
3615 if (!mddev->suspended && mddev->sync_thread)
3616 md_wakeup_thread(mddev->sync_thread);
3618 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3619 if (!mddev->suspended && mddev->thread)
3620 md_wakeup_thread(mddev->thread);
3625 static int raid_iterate_devices(struct dm_target *ti,
3626 iterate_devices_callout_fn fn, void *data)
3628 struct raid_set *rs = ti->private;
3632 for (i = 0; !r && i < rs->md.raid_disks; i++)
3633 if (rs->dev[i].data_dev)
3635 rs->dev[i].data_dev,
3636 0, /* No offset on data devs */
3643 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3645 struct raid_set *rs = ti->private;
3646 unsigned int chunk_size = to_bytes(rs->md.chunk_sectors);
3648 blk_limits_io_min(limits, chunk_size);
3649 blk_limits_io_opt(limits, chunk_size * mddev_data_stripes(rs));
3652 static void raid_postsuspend(struct dm_target *ti)
3654 struct raid_set *rs = ti->private;
3656 if (!test_and_set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
3657 /* Writes have to be stopped before suspending to avoid deadlocks. */
3658 if (!test_bit(MD_RECOVERY_FROZEN, &rs->md.recovery))
3659 md_stop_writes(&rs->md);
3661 mddev_lock_nointr(&rs->md);
3662 mddev_suspend(&rs->md);
3663 mddev_unlock(&rs->md);
3667 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
3670 uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
3671 unsigned long flags;
3672 bool cleared = false;
3673 struct dm_raid_superblock *sb;
3674 struct mddev *mddev = &rs->md;
3677 /* RAID personalities have to provide hot add/remove methods or we need to bail out. */
3678 if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
3681 memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
3683 for (i = 0; i < mddev->raid_disks; i++) {
3684 r = &rs->dev[i].rdev;
3685 /* HM FIXME: enhance journal device recovery processing */
3686 if (test_bit(Journal, &r->flags))
3689 if (test_bit(Faulty, &r->flags) &&
3690 r->meta_bdev && !read_disk_sb(r, r->sb_size, true)) {
3691 DMINFO("Faulty %s device #%d has readable super block."
3692 " Attempting to revive it.",
3693 rs->raid_type->name, i);
3696 * Faulty bit may be set, but sometimes the array can
3697 * be suspended before the personalities can respond
3698 * by removing the device from the array (i.e. calling
3699 * 'hot_remove_disk'). If they haven't yet removed
3700 * the failed device, its 'raid_disk' number will be
3701 * '>= 0' - meaning we must call this function
3705 clear_bit(In_sync, &r->flags); /* Mandatory for hot remove. */
3706 if (r->raid_disk >= 0) {
3707 if (mddev->pers->hot_remove_disk(mddev, r)) {
3708 /* Failed to revive this device, try next */
3713 r->raid_disk = r->saved_raid_disk = i;
3715 clear_bit(Faulty, &r->flags);
3716 clear_bit(WriteErrorSeen, &r->flags);
3718 if (mddev->pers->hot_add_disk(mddev, r)) {
3719 /* Failed to revive this device, try next */
3720 r->raid_disk = r->saved_raid_disk = -1;
3723 clear_bit(In_sync, &r->flags);
3724 r->recovery_offset = 0;
3725 set_bit(i, (void *) cleared_failed_devices);
3731 /* If any failed devices could be cleared, update all sbs failed_devices bits */
3733 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
3735 rdev_for_each(r, &rs->md) {
3736 if (test_bit(Journal, &r->flags))
3739 sb = page_address(r->sb_page);
3740 sb_retrieve_failed_devices(sb, failed_devices);
3742 for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
3743 failed_devices[i] &= ~cleared_failed_devices[i];
3745 sb_update_failed_devices(sb, failed_devices);
3750 static int __load_dirty_region_bitmap(struct raid_set *rs)
3754 /* Try loading the bitmap unless "raid0", which does not have one */
3755 if (!rs_is_raid0(rs) &&
3756 !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
3757 r = bitmap_load(&rs->md);
3759 DMERR("Failed to load bitmap");
3765 /* Enforce updating all superblocks */
3766 static void rs_update_sbs(struct raid_set *rs)
3768 struct mddev *mddev = &rs->md;
3771 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3773 md_update_sb(mddev, 1);
3778 * Reshape changes raid algorithm of @rs to new one within personality
3779 * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3780 * disks from a raid set thus growing/shrinking it or resizes the set
3782 * Call mddev_lock_nointr() before!
3784 static int rs_start_reshape(struct raid_set *rs)
3787 struct mddev *mddev = &rs->md;
3788 struct md_personality *pers = mddev->pers;
3790 r = rs_setup_reshape(rs);
3794 /* Need to be resumed to be able to start reshape, recovery is frozen until raid_resume() though */
3795 if (test_and_clear_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags))
3796 mddev_resume(mddev);
3799 * Check any reshape constraints enforced by the personalility
3801 * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3803 r = pers->check_reshape(mddev);
3805 rs->ti->error = "pers->check_reshape() failed";
3810 * Personality may not provide start reshape method in which
3811 * case check_reshape above has already covered everything
3813 if (pers->start_reshape) {
3814 r = pers->start_reshape(mddev);
3816 rs->ti->error = "pers->start_reshape() failed";
3821 /* Suspend because a resume will happen in raid_resume() */
3822 set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags);
3823 mddev_suspend(mddev);
3826 * Now reshape got set up, update superblocks to
3827 * reflect the fact so that a table reload will
3828 * access proper superblock content in the ctr.
3835 static int raid_preresume(struct dm_target *ti)
3838 struct raid_set *rs = ti->private;
3839 struct mddev *mddev = &rs->md;
3841 /* This is a resume after a suspend of the set -> it's already started */
3842 if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
3846 * The superblocks need to be updated on disk if the
3847 * array is new or new devices got added (thus zeroed
3848 * out by userspace) or __load_dirty_region_bitmap
3849 * will overwrite them in core with old data or fail.
3851 if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
3854 /* Load the bitmap from disk unless raid0 */
3855 r = __load_dirty_region_bitmap(rs);
3859 /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) */
3860 if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) && mddev->bitmap &&
3861 mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)) {
3862 r = bitmap_resize(mddev->bitmap, mddev->dev_sectors,
3863 to_bytes(rs->requested_bitmap_chunk_sectors), 0);
3865 DMERR("Failed to resize bitmap");
3868 /* Check for any resize/reshape on @rs and adjust/initiate */
3869 /* Be prepared for mddev_resume() in raid_resume() */
3870 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3871 if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
3872 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3873 mddev->resync_min = mddev->recovery_cp;
3876 rs_set_capacity(rs);
3878 /* Check for any reshape request unless new raid set */
3879 if (test_and_clear_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3880 /* Initiate a reshape. */
3881 mddev_lock_nointr(mddev);
3882 r = rs_start_reshape(rs);
3883 mddev_unlock(mddev);
3885 DMWARN("Failed to check/start reshape, continuing without change");
3892 static void raid_resume(struct dm_target *ti)
3894 struct raid_set *rs = ti->private;
3895 struct mddev *mddev = &rs->md;
3897 if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
3899 * A secondary resume while the device is active.
3900 * Take this opportunity to check whether any failed
3901 * devices are reachable again.
3903 attempt_restore_of_faulty_devices(rs);
3910 * Keep the RAID set frozen if reshape/rebuild flags are set.
3911 * The RAID set is unfrozen once the next table load/resume,
3912 * which clears the reshape/rebuild flags, occurs.
3913 * This ensures that the constructor for the inactive table
3914 * retrieves an up-to-date reshape_position.
3916 if (!(rs->ctr_flags & RESUME_STAY_FROZEN_FLAGS))
3917 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3919 if (test_and_clear_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
3920 mddev_lock_nointr(mddev);
3921 mddev_resume(mddev);
3922 mddev_unlock(mddev);
3926 static struct target_type raid_target = {
3928 .version = {1, 13, 0},
3929 .module = THIS_MODULE,
3933 .status = raid_status,
3934 .message = raid_message,
3935 .iterate_devices = raid_iterate_devices,
3936 .io_hints = raid_io_hints,
3937 .postsuspend = raid_postsuspend,
3938 .preresume = raid_preresume,
3939 .resume = raid_resume,
3942 static int __init dm_raid_init(void)
3944 DMINFO("Loading target version %u.%u.%u",
3945 raid_target.version[0],
3946 raid_target.version[1],
3947 raid_target.version[2]);
3948 return dm_register_target(&raid_target);
3951 static void __exit dm_raid_exit(void)
3953 dm_unregister_target(&raid_target);
3956 module_init(dm_raid_init);
3957 module_exit(dm_raid_exit);
3959 module_param(devices_handle_discard_safely, bool, 0644);
3960 MODULE_PARM_DESC(devices_handle_discard_safely,
3961 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
3963 MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
3964 MODULE_ALIAS("dm-raid0");
3965 MODULE_ALIAS("dm-raid1");
3966 MODULE_ALIAS("dm-raid10");
3967 MODULE_ALIAS("dm-raid4");
3968 MODULE_ALIAS("dm-raid5");
3969 MODULE_ALIAS("dm-raid6");
3970 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
3971 MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
3972 MODULE_LICENSE("GPL");