2 * Copyright (C) 2010-2011 Neil Brown
3 * Copyright (C) 2010-2018 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
8 #include <linux/slab.h>
9 #include <linux/module.h>
15 #include "md-bitmap.h"
17 #include <linux/device-mapper.h>
19 #define DM_MSG_PREFIX "raid"
20 #define MAX_RAID_DEVICES 253 /* md-raid kernel limit */
23 * Minimum sectors of free reshape space per raid device
25 #define MIN_FREE_RESHAPE_SPACE to_sector(4*4096)
28 * Minimum journal space 4 MiB in sectors.
30 #define MIN_RAID456_JOURNAL_SPACE (4*2048)
32 static bool devices_handle_discard_safely = false;
35 * The following flags are used by dm-raid.c to set up the array state.
36 * They must be cleared before md_run is called.
38 #define FirstUse 10 /* rdev flag */
42 * Two DM devices, one to hold metadata and one to hold the
43 * actual data/parity. The reason for this is to not confuse
44 * ti->len and give more flexibility in altering size and
47 * While it is possible for this device to be associated
48 * with a different physical device than the data_dev, it
49 * is intended for it to be the same.
50 * |--------- Physical Device ---------|
51 * |- meta_dev -|------ data_dev ------|
53 struct dm_dev *meta_dev;
54 struct dm_dev *data_dev;
59 * Bits for establishing rs->ctr_flags
64 #define __CTR_FLAG_SYNC 0 /* 1 */ /* Not with raid0! */
65 #define __CTR_FLAG_NOSYNC 1 /* 1 */ /* Not with raid0! */
66 #define __CTR_FLAG_REBUILD 2 /* 2 */ /* Not with raid0! */
67 #define __CTR_FLAG_DAEMON_SLEEP 3 /* 2 */ /* Not with raid0! */
68 #define __CTR_FLAG_MIN_RECOVERY_RATE 4 /* 2 */ /* Not with raid0! */
69 #define __CTR_FLAG_MAX_RECOVERY_RATE 5 /* 2 */ /* Not with raid0! */
70 #define __CTR_FLAG_MAX_WRITE_BEHIND 6 /* 2 */ /* Only with raid1! */
71 #define __CTR_FLAG_WRITE_MOSTLY 7 /* 2 */ /* Only with raid1! */
72 #define __CTR_FLAG_STRIPE_CACHE 8 /* 2 */ /* Only with raid4/5/6! */
73 #define __CTR_FLAG_REGION_SIZE 9 /* 2 */ /* Not with raid0! */
74 #define __CTR_FLAG_RAID10_COPIES 10 /* 2 */ /* Only with raid10 */
75 #define __CTR_FLAG_RAID10_FORMAT 11 /* 2 */ /* Only with raid10 */
77 #define __CTR_FLAG_DELTA_DISKS 12 /* 2 */ /* Only with reshapable raid1/4/5/6/10! */
78 #define __CTR_FLAG_DATA_OFFSET 13 /* 2 */ /* Only with reshapable raid4/5/6/10! */
79 #define __CTR_FLAG_RAID10_USE_NEAR_SETS 14 /* 2 */ /* Only with raid10! */
82 #define __CTR_FLAG_JOURNAL_DEV 15 /* 2 */ /* Only with raid4/5/6 (journal device)! */
85 #define __CTR_FLAG_JOURNAL_MODE 16 /* 2 */ /* Only with raid4/5/6 (journal mode)! */
88 * Flags for rs->ctr_flags field.
90 #define CTR_FLAG_SYNC (1 << __CTR_FLAG_SYNC)
91 #define CTR_FLAG_NOSYNC (1 << __CTR_FLAG_NOSYNC)
92 #define CTR_FLAG_REBUILD (1 << __CTR_FLAG_REBUILD)
93 #define CTR_FLAG_DAEMON_SLEEP (1 << __CTR_FLAG_DAEMON_SLEEP)
94 #define CTR_FLAG_MIN_RECOVERY_RATE (1 << __CTR_FLAG_MIN_RECOVERY_RATE)
95 #define CTR_FLAG_MAX_RECOVERY_RATE (1 << __CTR_FLAG_MAX_RECOVERY_RATE)
96 #define CTR_FLAG_MAX_WRITE_BEHIND (1 << __CTR_FLAG_MAX_WRITE_BEHIND)
97 #define CTR_FLAG_WRITE_MOSTLY (1 << __CTR_FLAG_WRITE_MOSTLY)
98 #define CTR_FLAG_STRIPE_CACHE (1 << __CTR_FLAG_STRIPE_CACHE)
99 #define CTR_FLAG_REGION_SIZE (1 << __CTR_FLAG_REGION_SIZE)
100 #define CTR_FLAG_RAID10_COPIES (1 << __CTR_FLAG_RAID10_COPIES)
101 #define CTR_FLAG_RAID10_FORMAT (1 << __CTR_FLAG_RAID10_FORMAT)
102 #define CTR_FLAG_DELTA_DISKS (1 << __CTR_FLAG_DELTA_DISKS)
103 #define CTR_FLAG_DATA_OFFSET (1 << __CTR_FLAG_DATA_OFFSET)
104 #define CTR_FLAG_RAID10_USE_NEAR_SETS (1 << __CTR_FLAG_RAID10_USE_NEAR_SETS)
105 #define CTR_FLAG_JOURNAL_DEV (1 << __CTR_FLAG_JOURNAL_DEV)
106 #define CTR_FLAG_JOURNAL_MODE (1 << __CTR_FLAG_JOURNAL_MODE)
109 * Definitions of various constructor flags to
110 * be used in checks of valid / invalid flags
113 /* Define all any sync flags */
114 #define CTR_FLAGS_ANY_SYNC (CTR_FLAG_SYNC | CTR_FLAG_NOSYNC)
116 /* Define flags for options without argument (e.g. 'nosync') */
117 #define CTR_FLAG_OPTIONS_NO_ARGS (CTR_FLAGS_ANY_SYNC | \
118 CTR_FLAG_RAID10_USE_NEAR_SETS)
120 /* Define flags for options with one argument (e.g. 'delta_disks +2') */
121 #define CTR_FLAG_OPTIONS_ONE_ARG (CTR_FLAG_REBUILD | \
122 CTR_FLAG_WRITE_MOSTLY | \
123 CTR_FLAG_DAEMON_SLEEP | \
124 CTR_FLAG_MIN_RECOVERY_RATE | \
125 CTR_FLAG_MAX_RECOVERY_RATE | \
126 CTR_FLAG_MAX_WRITE_BEHIND | \
127 CTR_FLAG_STRIPE_CACHE | \
128 CTR_FLAG_REGION_SIZE | \
129 CTR_FLAG_RAID10_COPIES | \
130 CTR_FLAG_RAID10_FORMAT | \
131 CTR_FLAG_DELTA_DISKS | \
132 CTR_FLAG_DATA_OFFSET | \
133 CTR_FLAG_JOURNAL_DEV | \
134 CTR_FLAG_JOURNAL_MODE)
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
212 #define RT_FLAG_RS_IN_SYNC 6
213 #define RT_FLAG_RS_RESYNCING 7
214 #define RT_FLAG_RS_GROW 8
216 /* Array elements of 64 bit needed for rebuild/failed disk bits */
217 #define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
220 * raid set level, layout and chunk sectors backup/restore
225 int new_chunk_sectors;
229 struct dm_target *ti;
231 uint32_t stripe_cache_entries;
232 unsigned long ctr_flags;
233 unsigned long runtime_flags;
235 uint64_t rebuild_disks[DISKS_ARRAY_ELEMS];
241 int requested_bitmap_chunk_sectors;
244 struct raid_type *raid_type;
246 sector_t array_sectors;
247 sector_t dev_sectors;
249 /* Optional raid4/5/6 journal device */
256 struct raid_dev dev[];
259 static void rs_config_backup(struct raid_set *rs, struct rs_layout *l)
261 struct mddev *mddev = &rs->md;
263 l->new_level = mddev->new_level;
264 l->new_layout = mddev->new_layout;
265 l->new_chunk_sectors = mddev->new_chunk_sectors;
268 static void rs_config_restore(struct raid_set *rs, struct rs_layout *l)
270 struct mddev *mddev = &rs->md;
272 mddev->new_level = l->new_level;
273 mddev->new_layout = l->new_layout;
274 mddev->new_chunk_sectors = l->new_chunk_sectors;
277 /* raid10 algorithms (i.e. formats) */
278 #define ALGORITHM_RAID10_DEFAULT 0
279 #define ALGORITHM_RAID10_NEAR 1
280 #define ALGORITHM_RAID10_OFFSET 2
281 #define ALGORITHM_RAID10_FAR 3
283 /* Supported raid types and properties. */
284 static struct raid_type {
285 const char *name; /* RAID algorithm. */
286 const char *descr; /* Descriptor text for logging. */
287 const unsigned int parity_devs; /* # of parity devices. */
288 const unsigned int minimal_devs;/* minimal # of devices in set. */
289 const unsigned int level; /* RAID level. */
290 const unsigned int algorithm; /* RAID algorithm. */
292 {"raid0", "raid0 (striping)", 0, 2, 0, 0 /* NONE */},
293 {"raid1", "raid1 (mirroring)", 0, 2, 1, 0 /* NONE */},
294 {"raid10_far", "raid10 far (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_FAR},
295 {"raid10_offset", "raid10 offset (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_OFFSET},
296 {"raid10_near", "raid10 near (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_NEAR},
297 {"raid10", "raid10 (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_DEFAULT},
298 {"raid4", "raid4 (dedicated first parity disk)", 1, 2, 5, ALGORITHM_PARITY_0}, /* raid4 layout = raid5_0 */
299 {"raid5_n", "raid5 (dedicated last parity disk)", 1, 2, 5, ALGORITHM_PARITY_N},
300 {"raid5_ls", "raid5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
301 {"raid5_rs", "raid5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
302 {"raid5_la", "raid5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
303 {"raid5_ra", "raid5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
304 {"raid6_zr", "raid6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
305 {"raid6_nr", "raid6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
306 {"raid6_nc", "raid6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE},
307 {"raid6_n_6", "raid6 (dedicated parity/Q n/6)", 2, 4, 6, ALGORITHM_PARITY_N_6},
308 {"raid6_ls_6", "raid6 (left symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_SYMMETRIC_6},
309 {"raid6_rs_6", "raid6 (right symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_SYMMETRIC_6},
310 {"raid6_la_6", "raid6 (left asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_ASYMMETRIC_6},
311 {"raid6_ra_6", "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_ASYMMETRIC_6}
314 /* True, if @v is in inclusive range [@min, @max] */
315 static bool __within_range(long v, long min, long max)
317 return v >= min && v <= max;
320 /* All table line arguments are defined here */
321 static struct arg_name_flag {
322 const unsigned long flag;
324 } __arg_name_flags[] = {
325 { CTR_FLAG_SYNC, "sync"},
326 { CTR_FLAG_NOSYNC, "nosync"},
327 { CTR_FLAG_REBUILD, "rebuild"},
328 { CTR_FLAG_DAEMON_SLEEP, "daemon_sleep"},
329 { CTR_FLAG_MIN_RECOVERY_RATE, "min_recovery_rate"},
330 { CTR_FLAG_MAX_RECOVERY_RATE, "max_recovery_rate"},
331 { CTR_FLAG_MAX_WRITE_BEHIND, "max_write_behind"},
332 { CTR_FLAG_WRITE_MOSTLY, "write_mostly"},
333 { CTR_FLAG_STRIPE_CACHE, "stripe_cache"},
334 { CTR_FLAG_REGION_SIZE, "region_size"},
335 { CTR_FLAG_RAID10_COPIES, "raid10_copies"},
336 { CTR_FLAG_RAID10_FORMAT, "raid10_format"},
337 { CTR_FLAG_DATA_OFFSET, "data_offset"},
338 { CTR_FLAG_DELTA_DISKS, "delta_disks"},
339 { CTR_FLAG_RAID10_USE_NEAR_SETS, "raid10_use_near_sets"},
340 { CTR_FLAG_JOURNAL_DEV, "journal_dev" },
341 { CTR_FLAG_JOURNAL_MODE, "journal_mode" },
344 /* Return argument name string for given @flag */
345 static const char *dm_raid_arg_name_by_flag(const uint32_t flag)
347 if (hweight32(flag) == 1) {
348 struct arg_name_flag *anf = __arg_name_flags + ARRAY_SIZE(__arg_name_flags);
350 while (anf-- > __arg_name_flags)
351 if (flag & anf->flag)
355 DMERR("%s called with more than one flag!", __func__);
360 /* Define correlation of raid456 journal cache modes and dm-raid target line parameters */
364 } _raid456_journal_mode[] = {
365 { R5C_JOURNAL_MODE_WRITE_THROUGH , "writethrough" },
366 { R5C_JOURNAL_MODE_WRITE_BACK , "writeback" }
369 /* Return MD raid4/5/6 journal mode for dm @journal_mode one */
370 static int dm_raid_journal_mode_to_md(const char *mode)
372 int m = ARRAY_SIZE(_raid456_journal_mode);
375 if (!strcasecmp(mode, _raid456_journal_mode[m].param))
376 return _raid456_journal_mode[m].mode;
381 /* Return dm-raid raid4/5/6 journal mode string for @mode */
382 static const char *md_journal_mode_to_dm_raid(const int mode)
384 int m = ARRAY_SIZE(_raid456_journal_mode);
387 if (mode == _raid456_journal_mode[m].mode)
388 return _raid456_journal_mode[m].param;
394 * Bool helpers to test for various raid levels of a raid set.
395 * It's level as reported by the superblock rather than
396 * the requested raid_type passed to the constructor.
398 /* Return true, if raid set in @rs is raid0 */
399 static bool rs_is_raid0(struct raid_set *rs)
401 return !rs->md.level;
404 /* Return true, if raid set in @rs is raid1 */
405 static bool rs_is_raid1(struct raid_set *rs)
407 return rs->md.level == 1;
410 /* Return true, if raid set in @rs is raid10 */
411 static bool rs_is_raid10(struct raid_set *rs)
413 return rs->md.level == 10;
416 /* Return true, if raid set in @rs is level 6 */
417 static bool rs_is_raid6(struct raid_set *rs)
419 return rs->md.level == 6;
422 /* Return true, if raid set in @rs is level 4, 5 or 6 */
423 static bool rs_is_raid456(struct raid_set *rs)
425 return __within_range(rs->md.level, 4, 6);
428 /* Return true, if raid set in @rs is reshapable */
429 static bool __is_raid10_far(int layout);
430 static bool rs_is_reshapable(struct raid_set *rs)
432 return rs_is_raid456(rs) ||
433 (rs_is_raid10(rs) && !__is_raid10_far(rs->md.new_layout));
436 /* Return true, if raid set in @rs is recovering */
437 static bool rs_is_recovering(struct raid_set *rs)
439 return rs->md.recovery_cp < rs->md.dev_sectors;
442 /* Return true, if raid set in @rs is reshaping */
443 static bool rs_is_reshaping(struct raid_set *rs)
445 return rs->md.reshape_position != MaxSector;
449 * bool helpers to test for various raid levels of a raid type @rt
452 /* Return true, if raid type in @rt is raid0 */
453 static bool rt_is_raid0(struct raid_type *rt)
458 /* Return true, if raid type in @rt is raid1 */
459 static bool rt_is_raid1(struct raid_type *rt)
461 return rt->level == 1;
464 /* Return true, if raid type in @rt is raid10 */
465 static bool rt_is_raid10(struct raid_type *rt)
467 return rt->level == 10;
470 /* Return true, if raid type in @rt is raid4/5 */
471 static bool rt_is_raid45(struct raid_type *rt)
473 return __within_range(rt->level, 4, 5);
476 /* Return true, if raid type in @rt is raid6 */
477 static bool rt_is_raid6(struct raid_type *rt)
479 return rt->level == 6;
482 /* Return true, if raid type in @rt is raid4/5/6 */
483 static bool rt_is_raid456(struct raid_type *rt)
485 return __within_range(rt->level, 4, 6);
487 /* END: raid level bools */
489 /* Return valid ctr flags for the raid level of @rs */
490 static unsigned long __valid_flags(struct raid_set *rs)
492 if (rt_is_raid0(rs->raid_type))
493 return RAID0_VALID_FLAGS;
494 else if (rt_is_raid1(rs->raid_type))
495 return RAID1_VALID_FLAGS;
496 else if (rt_is_raid10(rs->raid_type))
497 return RAID10_VALID_FLAGS;
498 else if (rt_is_raid45(rs->raid_type))
499 return RAID45_VALID_FLAGS;
500 else if (rt_is_raid6(rs->raid_type))
501 return RAID6_VALID_FLAGS;
507 * Check for valid flags set on @rs
509 * Has to be called after parsing of the ctr flags!
511 static int rs_check_for_valid_flags(struct raid_set *rs)
513 if (rs->ctr_flags & ~__valid_flags(rs)) {
514 rs->ti->error = "Invalid flags combination";
521 /* MD raid10 bit definitions and helpers */
522 #define RAID10_OFFSET (1 << 16) /* stripes with data copies area adjacent on devices */
523 #define RAID10_BROCKEN_USE_FAR_SETS (1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
524 #define RAID10_USE_FAR_SETS (1 << 18) /* Use sets instead of whole stripe rotation */
525 #define RAID10_FAR_COPIES_SHIFT 8 /* raid10 # far copies shift (2nd byte of layout) */
527 /* Return md raid10 near copies for @layout */
528 static unsigned int __raid10_near_copies(int layout)
530 return layout & 0xFF;
533 /* Return md raid10 far copies for @layout */
534 static unsigned int __raid10_far_copies(int layout)
536 return __raid10_near_copies(layout >> RAID10_FAR_COPIES_SHIFT);
539 /* Return true if md raid10 offset for @layout */
540 static bool __is_raid10_offset(int layout)
542 return !!(layout & RAID10_OFFSET);
545 /* Return true if md raid10 near for @layout */
546 static bool __is_raid10_near(int layout)
548 return !__is_raid10_offset(layout) && __raid10_near_copies(layout) > 1;
551 /* Return true if md raid10 far for @layout */
552 static bool __is_raid10_far(int layout)
554 return !__is_raid10_offset(layout) && __raid10_far_copies(layout) > 1;
557 /* Return md raid10 layout string for @layout */
558 static const char *raid10_md_layout_to_format(int layout)
561 * Bit 16 stands for "offset"
562 * (i.e. adjacent stripes hold copies)
564 * Refer to MD's raid10.c for details
566 if (__is_raid10_offset(layout))
569 if (__raid10_near_copies(layout) > 1)
572 if (__raid10_far_copies(layout) > 1)
578 /* Return md raid10 algorithm for @name */
579 static int raid10_name_to_format(const char *name)
581 if (!strcasecmp(name, "near"))
582 return ALGORITHM_RAID10_NEAR;
583 else if (!strcasecmp(name, "offset"))
584 return ALGORITHM_RAID10_OFFSET;
585 else if (!strcasecmp(name, "far"))
586 return ALGORITHM_RAID10_FAR;
591 /* Return md raid10 copies for @layout */
592 static unsigned int raid10_md_layout_to_copies(int layout)
594 return max(__raid10_near_copies(layout), __raid10_far_copies(layout));
597 /* Return md raid10 format id for @format string */
598 static int raid10_format_to_md_layout(struct raid_set *rs,
599 unsigned int algorithm,
602 unsigned int n = 1, f = 1, r = 0;
605 * MD resilienece flaw:
607 * enabling use_far_sets for far/offset formats causes copies
608 * to be colocated on the same devs together with their origins!
610 * -> disable it for now in the definition above
612 if (algorithm == ALGORITHM_RAID10_DEFAULT ||
613 algorithm == ALGORITHM_RAID10_NEAR)
616 else if (algorithm == ALGORITHM_RAID10_OFFSET) {
619 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
620 r |= RAID10_USE_FAR_SETS;
622 } else if (algorithm == ALGORITHM_RAID10_FAR) {
624 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
625 r |= RAID10_USE_FAR_SETS;
630 return r | (f << RAID10_FAR_COPIES_SHIFT) | n;
632 /* END: MD raid10 bit definitions and helpers */
634 /* Check for any of the raid10 algorithms */
635 static bool __got_raid10(struct raid_type *rtp, const int layout)
637 if (rtp->level == 10) {
638 switch (rtp->algorithm) {
639 case ALGORITHM_RAID10_DEFAULT:
640 case ALGORITHM_RAID10_NEAR:
641 return __is_raid10_near(layout);
642 case ALGORITHM_RAID10_OFFSET:
643 return __is_raid10_offset(layout);
644 case ALGORITHM_RAID10_FAR:
645 return __is_raid10_far(layout);
654 /* Return raid_type for @name */
655 static struct raid_type *get_raid_type(const char *name)
657 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
659 while (rtp-- > raid_types)
660 if (!strcasecmp(rtp->name, name))
666 /* Return raid_type for @name based derived from @level and @layout */
667 static struct raid_type *get_raid_type_by_ll(const int level, const int layout)
669 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
671 while (rtp-- > raid_types) {
672 /* RAID10 special checks based on @layout flags/properties */
673 if (rtp->level == level &&
674 (__got_raid10(rtp, layout) || rtp->algorithm == layout))
681 /* Adjust rdev sectors */
682 static void rs_set_rdev_sectors(struct raid_set *rs)
684 struct mddev *mddev = &rs->md;
685 struct md_rdev *rdev;
688 * raid10 sets rdev->sector to the device size, which
689 * is unintended in case of out-of-place reshaping
691 rdev_for_each(rdev, mddev)
692 if (!test_bit(Journal, &rdev->flags))
693 rdev->sectors = mddev->dev_sectors;
697 * Change bdev capacity of @rs in case of a disk add/remove reshape
699 static void rs_set_capacity(struct raid_set *rs)
701 struct gendisk *gendisk = dm_disk(dm_table_get_md(rs->ti->table));
703 set_capacity_and_notify(gendisk, rs->md.array_sectors);
707 * Set the mddev properties in @rs to the current
708 * ones retrieved from the freshest superblock
710 static void rs_set_cur(struct raid_set *rs)
712 struct mddev *mddev = &rs->md;
714 mddev->new_level = mddev->level;
715 mddev->new_layout = mddev->layout;
716 mddev->new_chunk_sectors = mddev->chunk_sectors;
720 * Set the mddev properties in @rs to the new
721 * ones requested by the ctr
723 static void rs_set_new(struct raid_set *rs)
725 struct mddev *mddev = &rs->md;
727 mddev->level = mddev->new_level;
728 mddev->layout = mddev->new_layout;
729 mddev->chunk_sectors = mddev->new_chunk_sectors;
730 mddev->raid_disks = rs->raid_disks;
731 mddev->delta_disks = 0;
734 static struct raid_set *raid_set_alloc(struct dm_target *ti, struct raid_type *raid_type,
735 unsigned int raid_devs)
740 if (raid_devs <= raid_type->parity_devs) {
741 ti->error = "Insufficient number of devices";
742 return ERR_PTR(-EINVAL);
745 rs = kzalloc(struct_size(rs, dev, raid_devs), GFP_KERNEL);
747 ti->error = "Cannot allocate raid context";
748 return ERR_PTR(-ENOMEM);
753 rs->raid_disks = raid_devs;
757 rs->raid_type = raid_type;
758 rs->stripe_cache_entries = 256;
759 rs->md.raid_disks = raid_devs;
760 rs->md.level = raid_type->level;
761 rs->md.new_level = rs->md.level;
762 rs->md.layout = raid_type->algorithm;
763 rs->md.new_layout = rs->md.layout;
764 rs->md.delta_disks = 0;
765 rs->md.recovery_cp = MaxSector;
767 for (i = 0; i < raid_devs; i++)
768 md_rdev_init(&rs->dev[i].rdev);
771 * Remaining items to be initialized by further RAID params:
774 * rs->md.chunk_sectors
775 * rs->md.new_chunk_sectors
782 /* Free all @rs allocations */
783 static void raid_set_free(struct raid_set *rs)
787 if (rs->journal_dev.dev) {
788 md_rdev_clear(&rs->journal_dev.rdev);
789 dm_put_device(rs->ti, rs->journal_dev.dev);
792 for (i = 0; i < rs->raid_disks; i++) {
793 if (rs->dev[i].meta_dev)
794 dm_put_device(rs->ti, rs->dev[i].meta_dev);
795 md_rdev_clear(&rs->dev[i].rdev);
796 if (rs->dev[i].data_dev)
797 dm_put_device(rs->ti, rs->dev[i].data_dev);
804 * For every device we have two words
805 * <meta_dev>: meta device name or '-' if missing
806 * <data_dev>: data device name or '-' if missing
808 * The following are permitted:
811 * <meta_dev> <data_dev>
813 * The following is not allowed:
816 * This code parses those words. If there is a failure,
817 * the caller must use raid_set_free() to unwind the operations.
819 static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
823 int metadata_available = 0;
827 /* Put off the number of raid devices argument to get to dev pairs */
828 arg = dm_shift_arg(as);
832 for (i = 0; i < rs->raid_disks; i++) {
833 rs->dev[i].rdev.raid_disk = i;
835 rs->dev[i].meta_dev = NULL;
836 rs->dev[i].data_dev = NULL;
839 * There are no offsets initially.
840 * Out of place reshape will set them accordingly.
842 rs->dev[i].rdev.data_offset = 0;
843 rs->dev[i].rdev.new_data_offset = 0;
844 rs->dev[i].rdev.mddev = &rs->md;
846 arg = dm_shift_arg(as);
850 if (strcmp(arg, "-")) {
851 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
852 &rs->dev[i].meta_dev);
854 rs->ti->error = "RAID metadata device lookup failure";
858 rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
859 if (!rs->dev[i].rdev.sb_page) {
860 rs->ti->error = "Failed to allocate superblock page";
865 arg = dm_shift_arg(as);
869 if (!strcmp(arg, "-")) {
870 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
871 (!rs->dev[i].rdev.recovery_offset)) {
872 rs->ti->error = "Drive designated for rebuild not specified";
876 if (rs->dev[i].meta_dev) {
877 rs->ti->error = "No data device supplied with metadata device";
884 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
885 &rs->dev[i].data_dev);
887 rs->ti->error = "RAID device lookup failure";
891 if (rs->dev[i].meta_dev) {
892 metadata_available = 1;
893 rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
895 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
896 list_add_tail(&rs->dev[i].rdev.same_set, &rs->md.disks);
897 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
901 if (rs->journal_dev.dev)
902 list_add_tail(&rs->journal_dev.rdev.same_set, &rs->md.disks);
904 if (metadata_available) {
906 rs->md.persistent = 1;
907 rs->md.major_version = 2;
908 } else if (rebuild && !rs->md.recovery_cp) {
910 * Without metadata, we will not be able to tell if the array
911 * is in-sync or not - we must assume it is not. Therefore,
912 * it is impossible to rebuild a drive.
914 * Even if there is metadata, the on-disk information may
915 * indicate that the array is not in-sync and it will then
918 * User could specify 'nosync' option if desperate.
920 rs->ti->error = "Unable to rebuild drive while array is not in-sync";
928 * validate_region_size
930 * @region_size: region size in sectors. If 0, pick a size (4MiB default).
932 * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
933 * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
935 * Returns: 0 on success, -EINVAL on failure.
937 static int validate_region_size(struct raid_set *rs, unsigned long region_size)
939 unsigned long min_region_size = rs->ti->len / (1 << 21);
946 * Choose a reasonable default. All figures in sectors.
948 if (min_region_size > (1 << 13)) {
949 /* If not a power of 2, make it the next power of 2 */
950 region_size = roundup_pow_of_two(min_region_size);
951 DMINFO("Choosing default region size of %lu sectors",
954 DMINFO("Choosing default region size of 4MiB");
955 region_size = 1 << 13; /* sectors */
959 * Validate user-supplied value.
961 if (region_size > rs->ti->len) {
962 rs->ti->error = "Supplied region size is too large";
966 if (region_size < min_region_size) {
967 DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
968 region_size, min_region_size);
969 rs->ti->error = "Supplied region size is too small";
973 if (!is_power_of_2(region_size)) {
974 rs->ti->error = "Region size is not a power of 2";
978 if (region_size < rs->md.chunk_sectors) {
979 rs->ti->error = "Region size is smaller than the chunk size";
985 * Convert sectors to bytes.
987 rs->md.bitmap_info.chunksize = to_bytes(region_size);
993 * validate_raid_redundancy
996 * Determine if there are enough devices in the array that haven't
997 * failed (or are being rebuilt) to form a usable array.
999 * Returns: 0 on success, -EINVAL on failure.
1001 static int validate_raid_redundancy(struct raid_set *rs)
1003 unsigned int i, rebuild_cnt = 0;
1004 unsigned int rebuilds_per_group = 0, copies;
1005 unsigned int group_size, last_group_start;
1007 for (i = 0; i < rs->md.raid_disks; i++)
1008 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
1009 !rs->dev[i].rdev.sb_page)
1012 switch (rs->md.level) {
1016 if (rebuild_cnt >= rs->md.raid_disks)
1022 if (rebuild_cnt > rs->raid_type->parity_devs)
1026 copies = raid10_md_layout_to_copies(rs->md.new_layout);
1028 DMERR("Bogus raid10 data copies < 2!");
1032 if (rebuild_cnt < copies)
1036 * It is possible to have a higher rebuild count for RAID10,
1037 * as long as the failed devices occur in different mirror
1038 * groups (i.e. different stripes).
1040 * When checking "near" format, make sure no adjacent devices
1041 * have failed beyond what can be handled. In addition to the
1042 * simple case where the number of devices is a multiple of the
1043 * number of copies, we must also handle cases where the number
1044 * of devices is not a multiple of the number of copies.
1045 * E.g. dev1 dev2 dev3 dev4 dev5
1049 if (__is_raid10_near(rs->md.new_layout)) {
1050 for (i = 0; i < rs->md.raid_disks; i++) {
1052 rebuilds_per_group = 0;
1053 if ((!rs->dev[i].rdev.sb_page ||
1054 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1055 (++rebuilds_per_group >= copies))
1062 * When checking "far" and "offset" formats, we need to ensure
1063 * that the device that holds its copy is not also dead or
1064 * being rebuilt. (Note that "far" and "offset" formats only
1065 * support two copies right now. These formats also only ever
1066 * use the 'use_far_sets' variant.)
1068 * This check is somewhat complicated by the need to account
1069 * for arrays that are not a multiple of (far) copies. This
1070 * results in the need to treat the last (potentially larger)
1073 group_size = (rs->md.raid_disks / copies);
1074 last_group_start = (rs->md.raid_disks / group_size) - 1;
1075 last_group_start *= group_size;
1076 for (i = 0; i < rs->md.raid_disks; i++) {
1077 if (!(i % copies) && !(i > last_group_start))
1078 rebuilds_per_group = 0;
1079 if ((!rs->dev[i].rdev.sb_page ||
1080 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1081 (++rebuilds_per_group >= copies))
1097 * Possible arguments are...
1098 * <chunk_size> [optional_args]
1100 * Argument definitions
1101 * <chunk_size> The number of sectors per disk that
1102 * will form the "stripe"
1103 * [[no]sync] Force or prevent recovery of the
1105 * [rebuild <idx>] Rebuild the drive indicated by the index
1106 * [daemon_sleep <ms>] Time between bitmap daemon work to
1108 * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1109 * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1110 * [write_mostly <idx>] Indicate a write mostly drive via index
1111 * [max_write_behind <sectors>] See '-write-behind=' (man mdadm)
1112 * [stripe_cache <sectors>] Stripe cache size for higher RAIDs
1113 * [region_size <sectors>] Defines granularity of bitmap
1114 * [journal_dev <dev>] raid4/5/6 journaling deviice
1115 * (i.e. write hole closing log)
1117 * RAID10-only options:
1118 * [raid10_copies <# copies>] Number of copies. (Default: 2)
1119 * [raid10_format <near|far|offset>] Layout algorithm. (Default: near)
1121 static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
1122 unsigned int num_raid_params)
1124 int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
1125 unsigned int raid10_copies = 2;
1126 unsigned int i, write_mostly = 0;
1127 unsigned int region_size = 0;
1128 sector_t max_io_len;
1129 const char *arg, *key;
1130 struct raid_dev *rd;
1131 struct raid_type *rt = rs->raid_type;
1133 arg = dm_shift_arg(as);
1134 num_raid_params--; /* Account for chunk_size argument */
1136 if (kstrtoint(arg, 10, &value) < 0) {
1137 rs->ti->error = "Bad numerical argument given for chunk_size";
1142 * First, parse the in-order required arguments
1143 * "chunk_size" is the only argument of this type.
1145 if (rt_is_raid1(rt)) {
1147 DMERR("Ignoring chunk size parameter for RAID 1");
1149 } else if (!is_power_of_2(value)) {
1150 rs->ti->error = "Chunk size must be a power of 2";
1152 } else if (value < 8) {
1153 rs->ti->error = "Chunk size value is too small";
1157 rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
1160 * We set each individual device as In_sync with a completed
1161 * 'recovery_offset'. If there has been a device failure or
1162 * replacement then one of the following cases applies:
1164 * 1) User specifies 'rebuild'.
1165 * - Device is reset when param is read.
1166 * 2) A new device is supplied.
1167 * - No matching superblock found, resets device.
1168 * 3) Device failure was transient and returns on reload.
1169 * - Failure noticed, resets device for bitmap replay.
1170 * 4) Device hadn't completed recovery after previous failure.
1171 * - Superblock is read and overrides recovery_offset.
1173 * What is found in the superblocks of the devices is always
1174 * authoritative, unless 'rebuild' or '[no]sync' was specified.
1176 for (i = 0; i < rs->raid_disks; i++) {
1177 set_bit(In_sync, &rs->dev[i].rdev.flags);
1178 rs->dev[i].rdev.recovery_offset = MaxSector;
1182 * Second, parse the unordered optional arguments
1184 for (i = 0; i < num_raid_params; i++) {
1185 key = dm_shift_arg(as);
1187 rs->ti->error = "Not enough raid parameters given";
1191 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
1192 if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1193 rs->ti->error = "Only one 'nosync' argument allowed";
1198 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
1199 if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
1200 rs->ti->error = "Only one 'sync' argument allowed";
1205 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
1206 if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1207 rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1213 arg = dm_shift_arg(as);
1214 i++; /* Account for the argument pairs */
1216 rs->ti->error = "Wrong number of raid parameters given";
1221 * Parameters that take a string value are checked here.
1223 /* "raid10_format {near|offset|far} */
1224 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
1225 if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
1226 rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1229 if (!rt_is_raid10(rt)) {
1230 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1233 raid10_format = raid10_name_to_format(arg);
1234 if (raid10_format < 0) {
1235 rs->ti->error = "Invalid 'raid10_format' value given";
1236 return raid10_format;
1241 /* "journal_dev <dev>" */
1242 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV))) {
1244 struct md_rdev *jdev;
1246 if (test_and_set_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1247 rs->ti->error = "Only one raid4/5/6 set journaling device allowed";
1250 if (!rt_is_raid456(rt)) {
1251 rs->ti->error = "'journal_dev' is an invalid parameter for this RAID type";
1254 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
1255 &rs->journal_dev.dev);
1257 rs->ti->error = "raid4/5/6 journal device lookup failure";
1260 jdev = &rs->journal_dev.rdev;
1262 jdev->mddev = &rs->md;
1263 jdev->bdev = rs->journal_dev.dev->bdev;
1264 jdev->sectors = to_sector(i_size_read(jdev->bdev->bd_inode));
1265 if (jdev->sectors < MIN_RAID456_JOURNAL_SPACE) {
1266 rs->ti->error = "No space for raid4/5/6 journal";
1269 rs->journal_dev.mode = R5C_JOURNAL_MODE_WRITE_THROUGH;
1270 set_bit(Journal, &jdev->flags);
1274 /* "journal_mode <mode>" ("journal_dev" mandatory!) */
1275 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE))) {
1278 if (!test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1279 rs->ti->error = "raid4/5/6 'journal_mode' is invalid without 'journal_dev'";
1282 if (test_and_set_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
1283 rs->ti->error = "Only one raid4/5/6 'journal_mode' argument allowed";
1286 r = dm_raid_journal_mode_to_md(arg);
1288 rs->ti->error = "Invalid 'journal_mode' argument";
1291 rs->journal_dev.mode = r;
1296 * Parameters with number values from here on.
1298 if (kstrtoint(arg, 10, &value) < 0) {
1299 rs->ti->error = "Bad numerical argument given in raid params";
1303 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
1305 * "rebuild" is being passed in by userspace to provide
1306 * indexes of replaced devices and to set up additional
1307 * devices on raid level takeover.
1309 if (!__within_range(value, 0, rs->raid_disks - 1)) {
1310 rs->ti->error = "Invalid rebuild index given";
1314 if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1315 rs->ti->error = "rebuild for this index already given";
1319 rd = rs->dev + value;
1320 clear_bit(In_sync, &rd->rdev.flags);
1321 clear_bit(Faulty, &rd->rdev.flags);
1322 rd->rdev.recovery_offset = 0;
1323 set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
1324 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
1325 if (!rt_is_raid1(rt)) {
1326 rs->ti->error = "write_mostly option is only valid for RAID1";
1330 if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
1331 rs->ti->error = "Invalid write_mostly index given";
1336 set_bit(WriteMostly, &rs->dev[value].rdev.flags);
1337 set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
1338 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
1339 if (!rt_is_raid1(rt)) {
1340 rs->ti->error = "max_write_behind option is only valid for RAID1";
1344 if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
1345 rs->ti->error = "Only one max_write_behind argument pair allowed";
1350 * In device-mapper, we specify things in sectors, but
1351 * MD records this value in kB
1353 if (value < 0 || value / 2 > COUNTER_MAX) {
1354 rs->ti->error = "Max write-behind limit out of range";
1358 rs->md.bitmap_info.max_write_behind = value / 2;
1359 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
1360 if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
1361 rs->ti->error = "Only one daemon_sleep argument pair allowed";
1365 rs->ti->error = "daemon sleep period out of range";
1368 rs->md.bitmap_info.daemon_sleep = value;
1369 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
1370 /* Userspace passes new data_offset after having extended the the data image LV */
1371 if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
1372 rs->ti->error = "Only one data_offset argument pair allowed";
1375 /* Ensure sensible data offset */
1377 (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
1378 rs->ti->error = "Bogus data_offset value";
1381 rs->data_offset = value;
1382 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
1383 /* Define the +/-# of disks to add to/remove from the given raid set */
1384 if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1385 rs->ti->error = "Only one delta_disks argument pair allowed";
1388 /* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1389 if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
1390 rs->ti->error = "Too many delta_disk requested";
1394 rs->delta_disks = value;
1395 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
1396 if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
1397 rs->ti->error = "Only one stripe_cache argument pair allowed";
1401 if (!rt_is_raid456(rt)) {
1402 rs->ti->error = "Inappropriate argument: stripe_cache";
1407 rs->ti->error = "Bogus stripe cache entries value";
1410 rs->stripe_cache_entries = value;
1411 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
1412 if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1413 rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1418 rs->ti->error = "min_recovery_rate out of range";
1421 rs->md.sync_speed_min = value;
1422 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
1423 if (test_and_set_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags)) {
1424 rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1429 rs->ti->error = "max_recovery_rate out of range";
1432 rs->md.sync_speed_max = value;
1433 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
1434 if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
1435 rs->ti->error = "Only one region_size argument pair allowed";
1439 region_size = value;
1440 rs->requested_bitmap_chunk_sectors = value;
1441 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
1442 if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
1443 rs->ti->error = "Only one raid10_copies argument pair allowed";
1447 if (!__within_range(value, 2, rs->md.raid_disks)) {
1448 rs->ti->error = "Bad value for 'raid10_copies'";
1452 raid10_copies = value;
1454 DMERR("Unable to parse RAID parameter: %s", key);
1455 rs->ti->error = "Unable to parse RAID parameter";
1460 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1461 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1462 rs->ti->error = "sync and nosync are mutually exclusive";
1466 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) &&
1467 (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ||
1468 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))) {
1469 rs->ti->error = "sync/nosync and rebuild are mutually exclusive";
1473 if (write_mostly >= rs->md.raid_disks) {
1474 rs->ti->error = "Can't set all raid1 devices to write_mostly";
1478 if (rs->md.sync_speed_max &&
1479 rs->md.sync_speed_min > rs->md.sync_speed_max) {
1480 rs->ti->error = "Bogus recovery rates";
1484 if (validate_region_size(rs, region_size))
1487 if (rs->md.chunk_sectors)
1488 max_io_len = rs->md.chunk_sectors;
1490 max_io_len = region_size;
1492 if (dm_set_target_max_io_len(rs->ti, max_io_len))
1495 if (rt_is_raid10(rt)) {
1496 if (raid10_copies > rs->md.raid_disks) {
1497 rs->ti->error = "Not enough devices to satisfy specification";
1501 rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
1502 if (rs->md.new_layout < 0) {
1503 rs->ti->error = "Error getting raid10 format";
1504 return rs->md.new_layout;
1507 rt = get_raid_type_by_ll(10, rs->md.new_layout);
1509 rs->ti->error = "Failed to recognize new raid10 layout";
1513 if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1514 rt->algorithm == ALGORITHM_RAID10_NEAR) &&
1515 test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1516 rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1521 rs->raid10_copies = raid10_copies;
1523 /* Assume there are no metadata devices until the drives are parsed */
1524 rs->md.persistent = 0;
1525 rs->md.external = 1;
1527 /* Check, if any invalid ctr arguments have been passed in for the raid level */
1528 return rs_check_for_valid_flags(rs);
1531 /* Set raid4/5/6 cache size */
1532 static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1535 struct r5conf *conf;
1536 struct mddev *mddev = &rs->md;
1537 uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1538 uint32_t nr_stripes = rs->stripe_cache_entries;
1540 if (!rt_is_raid456(rs->raid_type)) {
1541 rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1545 if (nr_stripes < min_stripes) {
1546 DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1547 nr_stripes, min_stripes);
1548 nr_stripes = min_stripes;
1551 conf = mddev->private;
1553 rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1557 /* Try setting number of stripes in raid456 stripe cache */
1558 if (conf->min_nr_stripes != nr_stripes) {
1559 r = raid5_set_cache_size(mddev, nr_stripes);
1561 rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1565 DMINFO("%u stripe cache entries", nr_stripes);
1571 /* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1572 static unsigned int mddev_data_stripes(struct raid_set *rs)
1574 return rs->md.raid_disks - rs->raid_type->parity_devs;
1577 /* Return # of data stripes of @rs (i.e. as of ctr) */
1578 static unsigned int rs_data_stripes(struct raid_set *rs)
1580 return rs->raid_disks - rs->raid_type->parity_devs;
1584 * Retrieve rdev->sectors from any valid raid device of @rs
1585 * to allow userpace to pass in arbitray "- -" device tupples.
1587 static sector_t __rdev_sectors(struct raid_set *rs)
1591 for (i = 0; i < rs->md.raid_disks; i++) {
1592 struct md_rdev *rdev = &rs->dev[i].rdev;
1594 if (!test_bit(Journal, &rdev->flags) &&
1595 rdev->bdev && rdev->sectors)
1596 return rdev->sectors;
1602 /* Check that calculated dev_sectors fits all component devices. */
1603 static int _check_data_dev_sectors(struct raid_set *rs)
1606 struct md_rdev *rdev;
1608 rdev_for_each(rdev, &rs->md)
1609 if (!test_bit(Journal, &rdev->flags) && rdev->bdev) {
1610 ds = min(ds, to_sector(i_size_read(rdev->bdev->bd_inode)));
1611 if (ds < rs->md.dev_sectors) {
1612 rs->ti->error = "Component device(s) too small";
1620 /* Calculate the sectors per device and per array used for @rs */
1621 static int rs_set_dev_and_array_sectors(struct raid_set *rs, sector_t sectors, bool use_mddev)
1624 unsigned int data_stripes;
1625 sector_t array_sectors = sectors, dev_sectors = sectors;
1626 struct mddev *mddev = &rs->md;
1629 delta_disks = mddev->delta_disks;
1630 data_stripes = mddev_data_stripes(rs);
1632 delta_disks = rs->delta_disks;
1633 data_stripes = rs_data_stripes(rs);
1636 /* Special raid1 case w/o delta_disks support (yet) */
1637 if (rt_is_raid1(rs->raid_type))
1639 else if (rt_is_raid10(rs->raid_type)) {
1640 if (rs->raid10_copies < 2 ||
1642 rs->ti->error = "Bogus raid10 data copies or delta disks";
1646 dev_sectors *= rs->raid10_copies;
1647 if (sector_div(dev_sectors, data_stripes))
1650 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1651 if (sector_div(array_sectors, rs->raid10_copies))
1654 } else if (sector_div(dev_sectors, data_stripes))
1658 /* Striped layouts */
1659 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1661 mddev->array_sectors = array_sectors;
1662 mddev->dev_sectors = dev_sectors;
1663 rs_set_rdev_sectors(rs);
1665 return _check_data_dev_sectors(rs);
1667 rs->ti->error = "Target length not divisible by number of data devices";
1671 /* Setup recovery on @rs */
1672 static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1674 /* raid0 does not recover */
1675 if (rs_is_raid0(rs))
1676 rs->md.recovery_cp = MaxSector;
1678 * A raid6 set has to be recovered either
1679 * completely or for the grown part to
1680 * ensure proper parity and Q-Syndrome
1682 else if (rs_is_raid6(rs))
1683 rs->md.recovery_cp = dev_sectors;
1685 * Other raid set types may skip recovery
1686 * depending on the 'nosync' flag.
1689 rs->md.recovery_cp = test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)
1690 ? MaxSector : dev_sectors;
1693 static void do_table_event(struct work_struct *ws)
1695 struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1697 smp_rmb(); /* Make sure we access most actual mddev properties */
1698 if (!rs_is_reshaping(rs)) {
1699 if (rs_is_raid10(rs))
1700 rs_set_rdev_sectors(rs);
1701 rs_set_capacity(rs);
1703 dm_table_event(rs->ti->table);
1707 * Make sure a valid takover (level switch) is being requested on @rs
1709 * Conversions of raid sets from one MD personality to another
1710 * have to conform to restrictions which are enforced here.
1712 static int rs_check_takeover(struct raid_set *rs)
1714 struct mddev *mddev = &rs->md;
1715 unsigned int near_copies;
1717 if (rs->md.degraded) {
1718 rs->ti->error = "Can't takeover degraded raid set";
1722 if (rs_is_reshaping(rs)) {
1723 rs->ti->error = "Can't takeover reshaping raid set";
1727 switch (mddev->level) {
1729 /* raid0 -> raid1/5 with one disk */
1730 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1731 mddev->raid_disks == 1)
1734 /* raid0 -> raid10 */
1735 if (mddev->new_level == 10 &&
1736 !(rs->raid_disks % mddev->raid_disks))
1739 /* raid0 with multiple disks -> raid4/5/6 */
1740 if (__within_range(mddev->new_level, 4, 6) &&
1741 mddev->new_layout == ALGORITHM_PARITY_N &&
1742 mddev->raid_disks > 1)
1748 /* Can't takeover raid10_offset! */
1749 if (__is_raid10_offset(mddev->layout))
1752 near_copies = __raid10_near_copies(mddev->layout);
1754 /* raid10* -> raid0 */
1755 if (mddev->new_level == 0) {
1756 /* Can takeover raid10_near with raid disks divisable by data copies! */
1757 if (near_copies > 1 &&
1758 !(mddev->raid_disks % near_copies)) {
1759 mddev->raid_disks /= near_copies;
1760 mddev->delta_disks = mddev->raid_disks;
1764 /* Can takeover raid10_far */
1765 if (near_copies == 1 &&
1766 __raid10_far_copies(mddev->layout) > 1)
1772 /* raid10_{near,far} -> raid1 */
1773 if (mddev->new_level == 1 &&
1774 max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
1777 /* raid10_{near,far} with 2 disks -> raid4/5 */
1778 if (__within_range(mddev->new_level, 4, 5) &&
1779 mddev->raid_disks == 2)
1784 /* raid1 with 2 disks -> raid4/5 */
1785 if (__within_range(mddev->new_level, 4, 5) &&
1786 mddev->raid_disks == 2) {
1787 mddev->degraded = 1;
1791 /* raid1 -> raid0 */
1792 if (mddev->new_level == 0 &&
1793 mddev->raid_disks == 1)
1796 /* raid1 -> raid10 */
1797 if (mddev->new_level == 10)
1802 /* raid4 -> raid0 */
1803 if (mddev->new_level == 0)
1806 /* raid4 -> raid1/5 with 2 disks */
1807 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1808 mddev->raid_disks == 2)
1811 /* raid4 -> raid5/6 with parity N */
1812 if (__within_range(mddev->new_level, 5, 6) &&
1813 mddev->layout == ALGORITHM_PARITY_N)
1818 /* raid5 with parity N -> raid0 */
1819 if (mddev->new_level == 0 &&
1820 mddev->layout == ALGORITHM_PARITY_N)
1823 /* raid5 with parity N -> raid4 */
1824 if (mddev->new_level == 4 &&
1825 mddev->layout == ALGORITHM_PARITY_N)
1828 /* raid5 with 2 disks -> raid1/4/10 */
1829 if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1830 mddev->raid_disks == 2)
1833 /* raid5_* -> raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
1834 if (mddev->new_level == 6 &&
1835 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1836 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
1841 /* raid6 with parity N -> raid0 */
1842 if (mddev->new_level == 0 &&
1843 mddev->layout == ALGORITHM_PARITY_N)
1846 /* raid6 with parity N -> raid4 */
1847 if (mddev->new_level == 4 &&
1848 mddev->layout == ALGORITHM_PARITY_N)
1851 /* raid6_*_n with Q-Syndrome N -> raid5_* */
1852 if (mddev->new_level == 5 &&
1853 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1854 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
1862 rs->ti->error = "takeover not possible";
1866 /* True if @rs requested to be taken over */
1867 static bool rs_takeover_requested(struct raid_set *rs)
1869 return rs->md.new_level != rs->md.level;
1872 /* True if layout is set to reshape. */
1873 static bool rs_is_layout_change(struct raid_set *rs, bool use_mddev)
1875 return (use_mddev ? rs->md.delta_disks : rs->delta_disks) ||
1876 rs->md.new_layout != rs->md.layout ||
1877 rs->md.new_chunk_sectors != rs->md.chunk_sectors;
1880 /* True if @rs is requested to reshape by ctr */
1881 static bool rs_reshape_requested(struct raid_set *rs)
1884 struct mddev *mddev = &rs->md;
1886 if (rs_takeover_requested(rs))
1889 if (rs_is_raid0(rs))
1892 change = rs_is_layout_change(rs, false);
1894 /* Historical case to support raid1 reshape without delta disks */
1895 if (rs_is_raid1(rs)) {
1896 if (rs->delta_disks)
1897 return !!rs->delta_disks;
1900 mddev->raid_disks != rs->raid_disks;
1903 if (rs_is_raid10(rs))
1905 !__is_raid10_far(mddev->new_layout) &&
1906 rs->delta_disks >= 0;
1912 #define FEATURE_FLAG_SUPPORTS_V190 0x1 /* Supports extended superblock */
1914 /* State flags for sb->flags */
1915 #define SB_FLAG_RESHAPE_ACTIVE 0x1
1916 #define SB_FLAG_RESHAPE_BACKWARDS 0x2
1919 * This structure is never routinely used by userspace, unlike md superblocks.
1920 * Devices with this superblock should only ever be accessed via device-mapper.
1922 #define DM_RAID_MAGIC 0x64526D44
1923 struct dm_raid_superblock {
1924 __le32 magic; /* "DmRd" */
1925 __le32 compat_features; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1927 __le32 num_devices; /* Number of devices in this raid set. (Max 64) */
1928 __le32 array_position; /* The position of this drive in the raid set */
1930 __le64 events; /* Incremented by md when superblock updated */
1931 __le64 failed_devices; /* Pre 1.9.0 part of bit field of devices to */
1932 /* indicate failures (see extension below) */
1935 * This offset tracks the progress of the repair or replacement of
1936 * an individual drive.
1938 __le64 disk_recovery_offset;
1941 * This offset tracks the progress of the initial raid set
1942 * synchronisation/parity calculation.
1944 __le64 array_resync_offset;
1947 * raid characteristics
1951 __le32 stripe_sectors;
1953 /********************************************************************
1954 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1956 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
1959 __le32 flags; /* Flags defining array states for reshaping */
1962 * This offset tracks the progress of a raid
1963 * set reshape in order to be able to restart it
1965 __le64 reshape_position;
1968 * These define the properties of the array in case of an interrupted reshape
1972 __le32 new_stripe_sectors;
1975 __le64 array_sectors; /* Array size in sectors */
1978 * Sector offsets to data on devices (reshaping).
1979 * Needed to support out of place reshaping, thus
1980 * not writing over any stripes whilst converting
1981 * them from old to new layout
1984 __le64 new_data_offset;
1986 __le64 sectors; /* Used device size in sectors */
1989 * Additonal Bit field of devices indicating failures to support
1990 * up to 256 devices with the 1.9.0 on-disk metadata format
1992 __le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
1994 __le32 incompat_features; /* Used to indicate any incompatible features */
1996 /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
2000 * Check for reshape constraints on raid set @rs:
2002 * - reshape function non-existent
2004 * - ongoing recovery
2007 * Returns 0 if none or -EPERM if given constraint
2008 * and error message reference in @errmsg
2010 static int rs_check_reshape(struct raid_set *rs)
2012 struct mddev *mddev = &rs->md;
2014 if (!mddev->pers || !mddev->pers->check_reshape)
2015 rs->ti->error = "Reshape not supported";
2016 else if (mddev->degraded)
2017 rs->ti->error = "Can't reshape degraded raid set";
2018 else if (rs_is_recovering(rs))
2019 rs->ti->error = "Convert request on recovering raid set prohibited";
2020 else if (rs_is_reshaping(rs))
2021 rs->ti->error = "raid set already reshaping!";
2022 else if (!(rs_is_raid1(rs) || rs_is_raid10(rs) || rs_is_raid456(rs)))
2023 rs->ti->error = "Reshaping only supported for raid1/4/5/6/10";
2030 static int read_disk_sb(struct md_rdev *rdev, int size, bool force_reload)
2032 BUG_ON(!rdev->sb_page);
2034 if (rdev->sb_loaded && !force_reload)
2037 rdev->sb_loaded = 0;
2039 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) {
2040 DMERR("Failed to read superblock of device at position %d",
2042 md_error(rdev->mddev, rdev);
2043 set_bit(Faulty, &rdev->flags);
2047 rdev->sb_loaded = 1;
2052 static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2054 failed_devices[0] = le64_to_cpu(sb->failed_devices);
2055 memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
2057 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2058 int i = ARRAY_SIZE(sb->extended_failed_devices);
2061 failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
2065 static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2067 int i = ARRAY_SIZE(sb->extended_failed_devices);
2069 sb->failed_devices = cpu_to_le64(failed_devices[0]);
2071 sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
2075 * Synchronize the superblock members with the raid set properties
2077 * All superblock data is little endian.
2079 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
2081 bool update_failed_devices = false;
2083 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2084 struct dm_raid_superblock *sb;
2085 struct raid_set *rs = container_of(mddev, struct raid_set, md);
2087 /* No metadata device, no superblock */
2088 if (!rdev->meta_bdev)
2091 BUG_ON(!rdev->sb_page);
2093 sb = page_address(rdev->sb_page);
2095 sb_retrieve_failed_devices(sb, failed_devices);
2097 for (i = 0; i < rs->raid_disks; i++)
2098 if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
2099 update_failed_devices = true;
2100 set_bit(i, (void *) failed_devices);
2103 if (update_failed_devices)
2104 sb_update_failed_devices(sb, failed_devices);
2106 sb->magic = cpu_to_le32(DM_RAID_MAGIC);
2107 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2109 sb->num_devices = cpu_to_le32(mddev->raid_disks);
2110 sb->array_position = cpu_to_le32(rdev->raid_disk);
2112 sb->events = cpu_to_le64(mddev->events);
2114 sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
2115 sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
2117 sb->level = cpu_to_le32(mddev->level);
2118 sb->layout = cpu_to_le32(mddev->layout);
2119 sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
2121 /********************************************************************
2122 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
2124 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
2126 sb->new_level = cpu_to_le32(mddev->new_level);
2127 sb->new_layout = cpu_to_le32(mddev->new_layout);
2128 sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
2130 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2132 smp_rmb(); /* Make sure we access most recent reshape position */
2133 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2134 if (le64_to_cpu(sb->reshape_position) != MaxSector) {
2135 /* Flag ongoing reshape */
2136 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
2138 if (mddev->delta_disks < 0 || mddev->reshape_backwards)
2139 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
2141 /* Clear reshape flags */
2142 sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
2145 sb->array_sectors = cpu_to_le64(mddev->array_sectors);
2146 sb->data_offset = cpu_to_le64(rdev->data_offset);
2147 sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
2148 sb->sectors = cpu_to_le64(rdev->sectors);
2149 sb->incompat_features = cpu_to_le32(0);
2151 /* Zero out the rest of the payload after the size of the superblock */
2152 memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
2158 * This function creates a superblock if one is not found on the device
2159 * and will decide which superblock to use if there's a choice.
2161 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
2163 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
2166 struct dm_raid_superblock *sb;
2167 struct dm_raid_superblock *refsb;
2168 uint64_t events_sb, events_refsb;
2170 r = read_disk_sb(rdev, rdev->sb_size, false);
2174 sb = page_address(rdev->sb_page);
2177 * Two cases that we want to write new superblocks and rebuild:
2178 * 1) New device (no matching magic number)
2179 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2181 if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
2182 (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
2183 super_sync(rdev->mddev, rdev);
2185 set_bit(FirstUse, &rdev->flags);
2186 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2188 /* Force writing of superblocks to disk */
2189 set_bit(MD_SB_CHANGE_DEVS, &rdev->mddev->sb_flags);
2191 /* Any superblock is better than none, choose that if given */
2192 return refdev ? 0 : 1;
2198 events_sb = le64_to_cpu(sb->events);
2200 refsb = page_address(refdev->sb_page);
2201 events_refsb = le64_to_cpu(refsb->events);
2203 return (events_sb > events_refsb) ? 1 : 0;
2206 static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
2210 struct mddev *mddev = &rs->md;
2212 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2213 struct dm_raid_superblock *sb;
2214 uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
2216 struct dm_raid_superblock *sb2;
2218 sb = page_address(rdev->sb_page);
2219 events_sb = le64_to_cpu(sb->events);
2222 * Initialise to 1 if this is a new superblock.
2224 mddev->events = events_sb ? : 1;
2226 mddev->reshape_position = MaxSector;
2228 mddev->raid_disks = le32_to_cpu(sb->num_devices);
2229 mddev->level = le32_to_cpu(sb->level);
2230 mddev->layout = le32_to_cpu(sb->layout);
2231 mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
2234 * Reshaping is supported, e.g. reshape_position is valid
2235 * in superblock and superblock content is authoritative.
2237 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2238 /* Superblock is authoritative wrt given raid set layout! */
2239 mddev->new_level = le32_to_cpu(sb->new_level);
2240 mddev->new_layout = le32_to_cpu(sb->new_layout);
2241 mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
2242 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
2243 mddev->array_sectors = le64_to_cpu(sb->array_sectors);
2245 /* raid was reshaping and got interrupted */
2246 if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
2247 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
2248 DMERR("Reshape requested but raid set is still reshaping");
2252 if (mddev->delta_disks < 0 ||
2253 (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
2254 mddev->reshape_backwards = 1;
2256 mddev->reshape_backwards = 0;
2258 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2259 rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2264 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2266 struct raid_type *rt_cur = get_raid_type_by_ll(mddev->level, mddev->layout);
2267 struct raid_type *rt_new = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
2269 if (rs_takeover_requested(rs)) {
2270 if (rt_cur && rt_new)
2271 DMERR("Takeover raid sets from %s to %s not yet supported by metadata. (raid level change)",
2272 rt_cur->name, rt_new->name);
2274 DMERR("Takeover raid sets not yet supported by metadata. (raid level change)");
2276 } else if (rs_reshape_requested(rs)) {
2277 DMERR("Reshaping raid sets not yet supported by metadata. (raid layout change keeping level)");
2278 if (mddev->layout != mddev->new_layout) {
2279 if (rt_cur && rt_new)
2280 DMERR(" current layout %s vs new layout %s",
2281 rt_cur->name, rt_new->name);
2283 DMERR(" current layout 0x%X vs new layout 0x%X",
2284 le32_to_cpu(sb->layout), mddev->new_layout);
2286 if (mddev->chunk_sectors != mddev->new_chunk_sectors)
2287 DMERR(" current stripe sectors %u vs new stripe sectors %u",
2288 mddev->chunk_sectors, mddev->new_chunk_sectors);
2289 if (rs->delta_disks)
2290 DMERR(" current %u disks vs new %u disks",
2291 mddev->raid_disks, mddev->raid_disks + rs->delta_disks);
2292 if (rs_is_raid10(rs)) {
2293 DMERR(" Old layout: %s w/ %u copies",
2294 raid10_md_layout_to_format(mddev->layout),
2295 raid10_md_layout_to_copies(mddev->layout));
2296 DMERR(" New layout: %s w/ %u copies",
2297 raid10_md_layout_to_format(mddev->new_layout),
2298 raid10_md_layout_to_copies(mddev->new_layout));
2303 DMINFO("Discovered old metadata format; upgrading to extended metadata format");
2306 if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
2307 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2310 * During load, we set FirstUse if a new superblock was written.
2311 * There are two reasons we might not have a superblock:
2312 * 1) The raid set is brand new - in which case, all of the
2313 * devices must have their In_sync bit set. Also,
2314 * recovery_cp must be 0, unless forced.
2315 * 2) This is a new device being added to an old raid set
2316 * and the new device needs to be rebuilt - in which
2317 * case the In_sync bit will /not/ be set and
2318 * recovery_cp must be MaxSector.
2319 * 3) This is/are a new device(s) being added to an old
2320 * raid set during takeover to a higher raid level
2321 * to provide capacity for redundancy or during reshape
2322 * to add capacity to grow the raid set.
2325 rdev_for_each(r, mddev) {
2326 if (test_bit(Journal, &rdev->flags))
2329 if (test_bit(FirstUse, &r->flags))
2332 if (!test_bit(In_sync, &r->flags)) {
2333 DMINFO("Device %d specified for rebuild; clearing superblock",
2337 if (test_bit(FirstUse, &r->flags))
2344 if (new_devs == rs->raid_disks || !rebuilds) {
2345 /* Replace a broken device */
2346 if (new_devs == rs->raid_disks) {
2347 DMINFO("Superblocks created for new raid set");
2348 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2349 } else if (new_devs != rebuilds &&
2350 new_devs != rs->delta_disks) {
2351 DMERR("New device injected into existing raid set without "
2352 "'delta_disks' or 'rebuild' parameter specified");
2355 } else if (new_devs && new_devs != rebuilds) {
2356 DMERR("%u 'rebuild' devices cannot be injected into"
2357 " a raid set with %u other first-time devices",
2358 rebuilds, new_devs);
2360 } else if (rebuilds) {
2361 if (rebuild_and_new && rebuilds != rebuild_and_new) {
2362 DMERR("new device%s provided without 'rebuild'",
2363 new_devs > 1 ? "s" : "");
2365 } else if (!test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) && rs_is_recovering(rs)) {
2366 DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2367 (unsigned long long) mddev->recovery_cp);
2369 } else if (rs_is_reshaping(rs)) {
2370 DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2371 (unsigned long long) mddev->reshape_position);
2377 * Now we set the Faulty bit for those devices that are
2378 * recorded in the superblock as failed.
2380 sb_retrieve_failed_devices(sb, failed_devices);
2381 rdev_for_each(r, mddev) {
2382 if (test_bit(Journal, &rdev->flags) ||
2385 sb2 = page_address(r->sb_page);
2386 sb2->failed_devices = 0;
2387 memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
2390 * Check for any device re-ordering.
2392 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2393 role = le32_to_cpu(sb2->array_position);
2397 if (role != r->raid_disk) {
2398 if (rs_is_raid10(rs) && __is_raid10_near(mddev->layout)) {
2399 if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
2400 rs->raid_disks % rs->raid10_copies) {
2402 "Cannot change raid10 near set to odd # of devices!";
2406 sb2->array_position = cpu_to_le32(r->raid_disk);
2408 } else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
2409 !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2410 !rt_is_raid1(rs->raid_type)) {
2411 rs->ti->error = "Cannot change device positions in raid set";
2415 DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
2419 * Partial recovery is performed on
2420 * returning failed devices.
2422 if (test_bit(role, (void *) failed_devices))
2423 set_bit(Faulty, &r->flags);
2430 static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
2432 struct mddev *mddev = &rs->md;
2433 struct dm_raid_superblock *sb;
2435 if (rs_is_raid0(rs) || !rdev->sb_page || rdev->raid_disk < 0)
2438 sb = page_address(rdev->sb_page);
2441 * If mddev->events is not set, we know we have not yet initialized
2444 if (!mddev->events && super_init_validation(rs, rdev))
2447 if (le32_to_cpu(sb->compat_features) &&
2448 le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
2449 rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2453 if (sb->incompat_features) {
2454 rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
2458 /* Enable bitmap creation on @rs unless no metadevs or raid0 or journaled raid4/5/6 set. */
2459 mddev->bitmap_info.offset = (rt_is_raid0(rs->raid_type) || rs->journal_dev.dev) ? 0 : to_sector(4096);
2460 mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2462 if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2464 * Retrieve rdev size stored in superblock to be prepared for shrink.
2465 * Check extended superblock members are present otherwise the size
2468 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190)
2469 rdev->sectors = le64_to_cpu(sb->sectors);
2471 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
2472 if (rdev->recovery_offset == MaxSector)
2473 set_bit(In_sync, &rdev->flags);
2475 * If no reshape in progress -> we're recovering single
2476 * disk(s) and have to set the device(s) to out-of-sync
2478 else if (!rs_is_reshaping(rs))
2479 clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
2483 * If a device comes back, set it as not In_sync and no longer faulty.
2485 if (test_and_clear_bit(Faulty, &rdev->flags)) {
2486 rdev->recovery_offset = 0;
2487 clear_bit(In_sync, &rdev->flags);
2488 rdev->saved_raid_disk = rdev->raid_disk;
2491 /* Reshape support -> restore repective data offsets */
2492 rdev->data_offset = le64_to_cpu(sb->data_offset);
2493 rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
2499 * Analyse superblocks and select the freshest.
2501 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2504 struct md_rdev *rdev, *freshest;
2505 struct mddev *mddev = &rs->md;
2508 rdev_for_each(rdev, mddev) {
2509 if (test_bit(Journal, &rdev->flags))
2512 if (!rdev->meta_bdev)
2515 /* Set superblock offset/size for metadata device. */
2517 rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
2518 if (rdev->sb_size < sizeof(struct dm_raid_superblock) || rdev->sb_size > PAGE_SIZE) {
2519 DMERR("superblock size of a logical block is no longer valid");
2524 * Skipping super_load due to CTR_FLAG_SYNC will cause
2525 * the array to undergo initialization again as
2526 * though it were new. This is the intended effect
2527 * of the "sync" directive.
2529 * With reshaping capability added, we must ensure that
2530 * that the "sync" directive is disallowed during the reshape.
2532 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2535 r = super_load(rdev, freshest);
2544 /* This is a failure to read the superblock from the metadata device. */
2546 * We have to keep any raid0 data/metadata device pairs or
2547 * the MD raid0 personality will fail to start the array.
2549 if (rs_is_raid0(rs))
2553 * We keep the dm_devs to be able to emit the device tuple
2554 * properly on the table line in raid_status() (rather than
2555 * mistakenly acting as if '- -' got passed into the constructor).
2557 * The rdev has to stay on the same_set list to allow for
2558 * the attempt to restore faulty devices on second resume.
2560 rdev->raid_disk = rdev->saved_raid_disk = -1;
2569 * Validation of the freshest device provides the source of
2570 * validation for the remaining devices.
2572 rs->ti->error = "Unable to assemble array: Invalid superblocks";
2573 if (super_validate(rs, freshest))
2576 if (validate_raid_redundancy(rs)) {
2577 rs->ti->error = "Insufficient redundancy to activate array";
2581 rdev_for_each(rdev, mddev)
2582 if (!test_bit(Journal, &rdev->flags) &&
2584 super_validate(rs, rdev))
2590 * Adjust data_offset and new_data_offset on all disk members of @rs
2591 * for out of place reshaping if requested by contructor
2593 * We need free space at the beginning of each raid disk for forward
2594 * and at the end for backward reshapes which userspace has to provide
2595 * via remapping/reordering of space.
2597 static int rs_adjust_data_offsets(struct raid_set *rs)
2599 sector_t data_offset = 0, new_data_offset = 0;
2600 struct md_rdev *rdev;
2602 /* Constructor did not request data offset change */
2603 if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2604 if (!rs_is_reshapable(rs))
2610 /* HM FIXME: get In_Sync raid_dev? */
2611 rdev = &rs->dev[0].rdev;
2613 if (rs->delta_disks < 0) {
2615 * Removing disks (reshaping backwards):
2617 * - before reshape: data is at offset 0 and free space
2618 * is at end of each component LV
2620 * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2623 new_data_offset = rs->data_offset;
2625 } else if (rs->delta_disks > 0) {
2627 * Adding disks (reshaping forwards):
2629 * - before reshape: data is at offset rs->data_offset != 0 and
2630 * free space is at begin of each component LV
2632 * - after reshape: data is at offset 0 on each component LV
2634 data_offset = rs->data_offset;
2635 new_data_offset = 0;
2639 * User space passes in 0 for data offset after having removed reshape space
2641 * - or - (data offset != 0)
2643 * Changing RAID layout or chunk size -> toggle offsets
2645 * - before reshape: data is at offset rs->data_offset 0 and
2646 * free space is at end of each component LV
2648 * data is at offset rs->data_offset != 0 and
2649 * free space is at begin of each component LV
2651 * - after reshape: data is at offset 0 if it was at offset != 0
2652 * or at offset != 0 if it was at offset 0
2653 * on each component LV
2656 data_offset = rs->data_offset ? rdev->data_offset : 0;
2657 new_data_offset = data_offset ? 0 : rs->data_offset;
2658 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2662 * Make sure we got a minimum amount of free sectors per device
2664 if (rs->data_offset &&
2665 to_sector(i_size_read(rdev->bdev->bd_inode)) - rs->md.dev_sectors < MIN_FREE_RESHAPE_SPACE) {
2666 rs->ti->error = data_offset ? "No space for forward reshape" :
2667 "No space for backward reshape";
2672 * Raise recovery_cp in case data_offset != 0 to
2673 * avoid false recovery positives in the constructor.
2675 if (rs->md.recovery_cp < rs->md.dev_sectors)
2676 rs->md.recovery_cp += rs->dev[0].rdev.data_offset;
2678 /* Adjust data offsets on all rdevs but on any raid4/5/6 journal device */
2679 rdev_for_each(rdev, &rs->md) {
2680 if (!test_bit(Journal, &rdev->flags)) {
2681 rdev->data_offset = data_offset;
2682 rdev->new_data_offset = new_data_offset;
2689 /* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2690 static void __reorder_raid_disk_indexes(struct raid_set *rs)
2693 struct md_rdev *rdev;
2695 rdev_for_each(rdev, &rs->md) {
2696 if (!test_bit(Journal, &rdev->flags)) {
2697 rdev->raid_disk = i++;
2698 rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2704 * Setup @rs for takeover by a different raid level
2706 static int rs_setup_takeover(struct raid_set *rs)
2708 struct mddev *mddev = &rs->md;
2709 struct md_rdev *rdev;
2710 unsigned int d = mddev->raid_disks = rs->raid_disks;
2711 sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2713 if (rt_is_raid10(rs->raid_type)) {
2714 if (rs_is_raid0(rs)) {
2715 /* Userpace reordered disks -> adjust raid_disk indexes */
2716 __reorder_raid_disk_indexes(rs);
2718 /* raid0 -> raid10_far layout */
2719 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2721 } else if (rs_is_raid1(rs))
2722 /* raid1 -> raid10_near layout */
2723 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2730 clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2731 mddev->recovery_cp = MaxSector;
2734 rdev = &rs->dev[d].rdev;
2736 if (test_bit(d, (void *) rs->rebuild_disks)) {
2737 clear_bit(In_sync, &rdev->flags);
2738 clear_bit(Faulty, &rdev->flags);
2739 mddev->recovery_cp = rdev->recovery_offset = 0;
2740 /* Bitmap has to be created when we do an "up" takeover */
2741 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2744 rdev->new_data_offset = new_data_offset;
2750 /* Prepare @rs for reshape */
2751 static int rs_prepare_reshape(struct raid_set *rs)
2754 struct mddev *mddev = &rs->md;
2756 if (rs_is_raid10(rs)) {
2757 if (rs->raid_disks != mddev->raid_disks &&
2758 __is_raid10_near(mddev->layout) &&
2759 rs->raid10_copies &&
2760 rs->raid10_copies != __raid10_near_copies(mddev->layout)) {
2762 * raid disk have to be multiple of data copies to allow this conversion,
2764 * This is actually not a reshape it is a
2765 * rebuild of any additional mirrors per group
2767 if (rs->raid_disks % rs->raid10_copies) {
2768 rs->ti->error = "Can't reshape raid10 mirror groups";
2772 /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2773 __reorder_raid_disk_indexes(rs);
2774 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2776 mddev->new_layout = mddev->layout;
2781 } else if (rs_is_raid456(rs))
2784 else if (rs_is_raid1(rs)) {
2785 if (rs->delta_disks) {
2786 /* Process raid1 via delta_disks */
2787 mddev->degraded = rs->delta_disks < 0 ? -rs->delta_disks : rs->delta_disks;
2790 /* Process raid1 without delta_disks */
2791 mddev->raid_disks = rs->raid_disks;
2795 rs->ti->error = "Called with bogus raid type";
2800 set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2801 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2802 } else if (mddev->raid_disks < rs->raid_disks)
2803 /* Create new superblocks and bitmaps, if any new disks */
2804 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2809 /* Get reshape sectors from data_offsets or raid set */
2810 static sector_t _get_reshape_sectors(struct raid_set *rs)
2812 struct md_rdev *rdev;
2813 sector_t reshape_sectors = 0;
2815 rdev_for_each(rdev, &rs->md)
2816 if (!test_bit(Journal, &rdev->flags)) {
2817 reshape_sectors = (rdev->data_offset > rdev->new_data_offset) ?
2818 rdev->data_offset - rdev->new_data_offset :
2819 rdev->new_data_offset - rdev->data_offset;
2823 return max(reshape_sectors, (sector_t) rs->data_offset);
2828 * - change raid layout
2829 * - change chunk size
2833 static int rs_setup_reshape(struct raid_set *rs)
2836 unsigned int cur_raid_devs, d;
2837 sector_t reshape_sectors = _get_reshape_sectors(rs);
2838 struct mddev *mddev = &rs->md;
2839 struct md_rdev *rdev;
2841 mddev->delta_disks = rs->delta_disks;
2842 cur_raid_devs = mddev->raid_disks;
2844 /* Ignore impossible layout change whilst adding/removing disks */
2845 if (mddev->delta_disks &&
2846 mddev->layout != mddev->new_layout) {
2847 DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2848 mddev->new_layout = mddev->layout;
2852 * Adjust array size:
2854 * - in case of adding disk(s), array size has
2855 * to grow after the disk adding reshape,
2856 * which'll hapen in the event handler;
2857 * reshape will happen forward, so space has to
2858 * be available at the beginning of each disk
2860 * - in case of removing disk(s), array size
2861 * has to shrink before starting the reshape,
2862 * which'll happen here;
2863 * reshape will happen backward, so space has to
2864 * be available at the end of each disk
2866 * - data_offset and new_data_offset are
2867 * adjusted for aforementioned out of place
2868 * reshaping based on userspace passing in
2869 * the "data_offset <sectors>" key/value
2870 * pair via the constructor
2874 if (rs->delta_disks > 0) {
2875 /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2876 for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2877 rdev = &rs->dev[d].rdev;
2878 clear_bit(In_sync, &rdev->flags);
2881 * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2882 * by md, which'll store that erroneously in the superblock on reshape
2884 rdev->saved_raid_disk = -1;
2885 rdev->raid_disk = d;
2887 rdev->sectors = mddev->dev_sectors;
2888 rdev->recovery_offset = rs_is_raid1(rs) ? 0 : MaxSector;
2891 mddev->reshape_backwards = 0; /* adding disk(s) -> forward reshape */
2893 /* Remove disk(s) */
2894 } else if (rs->delta_disks < 0) {
2895 r = rs_set_dev_and_array_sectors(rs, rs->ti->len, true);
2896 mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2898 /* Change layout and/or chunk size */
2901 * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2903 * keeping number of disks and do layout change ->
2905 * toggle reshape_backward depending on data_offset:
2907 * - free space upfront -> reshape forward
2909 * - free space at the end -> reshape backward
2912 * This utilizes free reshape space avoiding the need
2913 * for userspace to move (parts of) LV segments in
2914 * case of layout/chunksize change (for disk
2915 * adding/removing reshape space has to be at
2916 * the proper address (see above with delta_disks):
2918 * add disk(s) -> begin
2919 * remove disk(s)-> end
2921 mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2925 * Adjust device size for forward reshape
2926 * because md_finish_reshape() reduces it.
2928 if (!mddev->reshape_backwards)
2929 rdev_for_each(rdev, &rs->md)
2930 if (!test_bit(Journal, &rdev->flags))
2931 rdev->sectors += reshape_sectors;
2937 * If the md resync thread has updated superblock with max reshape position
2938 * at the end of a reshape but not (yet) reset the layout configuration
2939 * changes -> reset the latter.
2941 static void rs_reset_inconclusive_reshape(struct raid_set *rs)
2943 if (!rs_is_reshaping(rs) && rs_is_layout_change(rs, true)) {
2945 rs->md.delta_disks = 0;
2946 rs->md.reshape_backwards = 0;
2951 * Enable/disable discard support on RAID set depending on
2952 * RAID level and discard properties of underlying RAID members.
2954 static void configure_discard_support(struct raid_set *rs)
2958 struct dm_target *ti = rs->ti;
2961 * XXX: RAID level 4,5,6 require zeroing for safety.
2963 raid456 = rs_is_raid456(rs);
2965 for (i = 0; i < rs->raid_disks; i++) {
2966 struct request_queue *q;
2968 if (!rs->dev[i].rdev.bdev)
2971 q = bdev_get_queue(rs->dev[i].rdev.bdev);
2972 if (!q || !blk_queue_discard(q))
2976 if (!devices_handle_discard_safely) {
2977 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2978 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2984 ti->num_discard_bios = 1;
2988 * Construct a RAID0/1/10/4/5/6 mapping:
2990 * <raid_type> <#raid_params> <raid_params>{0,} \
2991 * <#raid_devs> [<meta_dev1> <dev1>]{1,}
2993 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
2994 * details on possible <raid_params>.
2996 * Userspace is free to initialize the metadata devices, hence the superblocks to
2997 * enforce recreation based on the passed in table parameters.
3000 static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3003 bool resize = false;
3004 struct raid_type *rt;
3005 unsigned int num_raid_params, num_raid_devs;
3006 sector_t sb_array_sectors, rdev_sectors, reshape_sectors;
3007 struct raid_set *rs = NULL;
3009 struct rs_layout rs_layout;
3010 struct dm_arg_set as = { argc, argv }, as_nrd;
3011 struct dm_arg _args[] = {
3012 { 0, as.argc, "Cannot understand number of raid parameters" },
3013 { 1, 254, "Cannot understand number of raid devices parameters" }
3016 arg = dm_shift_arg(&as);
3018 ti->error = "No arguments";
3022 rt = get_raid_type(arg);
3024 ti->error = "Unrecognised raid_type";
3028 /* Must have <#raid_params> */
3029 if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
3032 /* number of raid device tupples <meta_dev data_dev> */
3034 dm_consume_args(&as_nrd, num_raid_params);
3035 _args[1].max = (as_nrd.argc - 1) / 2;
3036 if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
3039 if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
3040 ti->error = "Invalid number of supplied raid devices";
3044 rs = raid_set_alloc(ti, rt, num_raid_devs);
3048 r = parse_raid_params(rs, &as, num_raid_params);
3052 r = parse_dev_params(rs, &as);
3056 rs->md.sync_super = super_sync;
3059 * Calculate ctr requested array and device sizes to allow
3060 * for superblock analysis needing device sizes defined.
3062 * Any existing superblock will overwrite the array and device sizes
3064 r = rs_set_dev_and_array_sectors(rs, rs->ti->len, false);
3068 /* Memorize just calculated, potentially larger sizes to grow the raid set in preresume */
3069 rs->array_sectors = rs->md.array_sectors;
3070 rs->dev_sectors = rs->md.dev_sectors;
3073 * Backup any new raid set level, layout, ...
3074 * requested to be able to compare to superblock
3075 * members for conversion decisions.
3077 rs_config_backup(rs, &rs_layout);
3079 r = analyse_superblocks(ti, rs);
3083 /* All in-core metadata now as of current superblocks after calling analyse_superblocks() */
3084 sb_array_sectors = rs->md.array_sectors;
3085 rdev_sectors = __rdev_sectors(rs);
3086 if (!rdev_sectors) {
3087 ti->error = "Invalid rdev size";
3093 reshape_sectors = _get_reshape_sectors(rs);
3094 if (rs->dev_sectors != rdev_sectors) {
3095 resize = (rs->dev_sectors != rdev_sectors - reshape_sectors);
3096 if (rs->dev_sectors > rdev_sectors - reshape_sectors)
3097 set_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3100 INIT_WORK(&rs->md.event_work, do_table_event);
3102 ti->num_flush_bios = 1;
3104 /* Restore any requested new layout for conversion decision */
3105 rs_config_restore(rs, &rs_layout);
3108 * Now that we have any superblock metadata available,
3109 * check for new, recovering, reshaping, to be taken over,
3110 * to be reshaped or an existing, unchanged raid set to
3113 if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
3114 /* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
3115 if (rs_is_raid6(rs) &&
3116 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
3117 ti->error = "'nosync' not allowed for new raid6 set";
3121 rs_setup_recovery(rs, 0);
3122 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3124 } else if (rs_is_recovering(rs)) {
3125 /* A recovering raid set may be resized */
3127 } else if (rs_is_reshaping(rs)) {
3128 /* Have to reject size change request during reshape */
3130 ti->error = "Can't resize a reshaping raid set";
3135 } else if (rs_takeover_requested(rs)) {
3136 if (rs_is_reshaping(rs)) {
3137 ti->error = "Can't takeover a reshaping raid set";
3142 /* We can't takeover a journaled raid4/5/6 */
3143 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3144 ti->error = "Can't takeover a journaled raid4/5/6 set";
3150 * If a takeover is needed, userspace sets any additional
3151 * devices to rebuild and we can check for a valid request here.
3153 * If acceptible, set the level to the new requested
3154 * one, prohibit requesting recovery, allow the raid
3155 * set to run and store superblocks during resume.
3157 r = rs_check_takeover(rs);
3161 r = rs_setup_takeover(rs);
3165 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3166 /* Takeover ain't recovery, so disable recovery */
3167 rs_setup_recovery(rs, MaxSector);
3169 } else if (rs_reshape_requested(rs)) {
3170 /* Only request grow on raid set size extensions, not on reshapes. */
3171 clear_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3174 * No need to check for 'ongoing' takeover here, because takeover
3175 * is an instant operation as oposed to an ongoing reshape.
3178 /* We can't reshape a journaled raid4/5/6 */
3179 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3180 ti->error = "Can't reshape a journaled raid4/5/6 set";
3185 /* Out-of-place space has to be available to allow for a reshape unless raid1! */
3186 if (reshape_sectors || rs_is_raid1(rs)) {
3188 * We can only prepare for a reshape here, because the
3189 * raid set needs to run to provide the repective reshape
3190 * check functions via its MD personality instance.
3192 * So do the reshape check after md_run() succeeded.
3194 r = rs_prepare_reshape(rs);
3198 /* Reshaping ain't recovery, so disable recovery */
3199 rs_setup_recovery(rs, MaxSector);
3204 /* May not set recovery when a device rebuild is requested */
3205 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3206 clear_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3207 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3208 rs_setup_recovery(rs, MaxSector);
3209 } else if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags)) {
3211 * Set raid set to current size, i.e. size as of
3212 * superblocks to grow to larger size in preresume.
3214 r = rs_set_dev_and_array_sectors(rs, sb_array_sectors, false);
3218 rs_setup_recovery(rs, rs->md.recovery_cp < rs->md.dev_sectors ? rs->md.recovery_cp : rs->md.dev_sectors);
3220 /* This is no size change or it is shrinking, update size and record in superblocks */
3221 r = rs_set_dev_and_array_sectors(rs, rs->ti->len, false);
3225 if (sb_array_sectors > rs->array_sectors)
3226 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3231 /* If constructor requested it, change data and new_data offsets */
3232 r = rs_adjust_data_offsets(rs);
3236 /* Catch any inconclusive reshape superblock content. */
3237 rs_reset_inconclusive_reshape(rs);
3239 /* Start raid set read-only and assumed clean to change in raid_resume() */
3243 /* Keep array frozen until resume. */
3244 set_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
3246 /* Has to be held on running the array */
3247 mddev_lock_nointr(&rs->md);
3248 r = md_run(&rs->md);
3249 rs->md.in_sync = 0; /* Assume already marked dirty */
3251 ti->error = "Failed to run raid array";
3252 mddev_unlock(&rs->md);
3256 r = md_start(&rs->md);
3258 ti->error = "Failed to start raid array";
3259 mddev_unlock(&rs->md);
3263 /* If raid4/5/6 journal mode explicitly requested (only possible with journal dev) -> set it */
3264 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
3265 r = r5c_journal_mode_set(&rs->md, rs->journal_dev.mode);
3267 ti->error = "Failed to set raid4/5/6 journal mode";
3268 mddev_unlock(&rs->md);
3269 goto bad_journal_mode_set;
3273 mddev_suspend(&rs->md);
3274 set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags);
3276 /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
3277 if (rs_is_raid456(rs)) {
3278 r = rs_set_raid456_stripe_cache(rs);
3280 goto bad_stripe_cache;
3283 /* Now do an early reshape check */
3284 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3285 r = rs_check_reshape(rs);
3287 goto bad_check_reshape;
3289 /* Restore new, ctr requested layout to perform check */
3290 rs_config_restore(rs, &rs_layout);
3292 if (rs->md.pers->start_reshape) {
3293 r = rs->md.pers->check_reshape(&rs->md);
3295 ti->error = "Reshape check failed";
3296 goto bad_check_reshape;
3301 /* Disable/enable discard support on raid set. */
3302 configure_discard_support(rs);
3304 mddev_unlock(&rs->md);
3308 bad_journal_mode_set:
3318 static void raid_dtr(struct dm_target *ti)
3320 struct raid_set *rs = ti->private;
3326 static int raid_map(struct dm_target *ti, struct bio *bio)
3328 struct raid_set *rs = ti->private;
3329 struct mddev *mddev = &rs->md;
3332 * If we're reshaping to add disk(s)), ti->len and
3333 * mddev->array_sectors will differ during the process
3334 * (ti->len > mddev->array_sectors), so we have to requeue
3335 * bios with addresses > mddev->array_sectors here or
3336 * there will occur accesses past EOD of the component
3337 * data images thus erroring the raid set.
3339 if (unlikely(bio_end_sector(bio) > mddev->array_sectors))
3340 return DM_MAPIO_REQUEUE;
3342 md_handle_request(mddev, bio);
3344 return DM_MAPIO_SUBMITTED;
3347 /* Return sync state string for @state */
3348 enum sync_state { st_frozen, st_reshape, st_resync, st_check, st_repair, st_recover, st_idle };
3349 static const char *sync_str(enum sync_state state)
3351 /* Has to be in above sync_state order! */
3352 static const char *sync_strs[] = {
3362 return __within_range(state, 0, ARRAY_SIZE(sync_strs) - 1) ? sync_strs[state] : "undef";
3365 /* Return enum sync_state for @mddev derived from @recovery flags */
3366 static enum sync_state decipher_sync_action(struct mddev *mddev, unsigned long recovery)
3368 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
3371 /* The MD sync thread can be done with io or be interrupted but still be running */
3372 if (!test_bit(MD_RECOVERY_DONE, &recovery) &&
3373 (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
3374 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery)))) {
3375 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
3378 if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
3379 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
3381 if (test_bit(MD_RECOVERY_CHECK, &recovery))
3386 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3389 if (mddev->reshape_position != MaxSector)
3397 * Return status string for @rdev
3399 * Status characters:
3401 * 'D' = Dead/Failed raid set component or raid4/5/6 journal device
3402 * 'a' = Alive but not in-sync raid set component _or_ alive raid4/5/6 'write_back' journal device
3403 * 'A' = Alive and in-sync raid set component _or_ alive raid4/5/6 'write_through' journal device
3404 * '-' = Non-existing device (i.e. uspace passed '- -' into the ctr)
3406 static const char *__raid_dev_status(struct raid_set *rs, struct md_rdev *rdev)
3410 else if (test_bit(Faulty, &rdev->flags))
3412 else if (test_bit(Journal, &rdev->flags))
3413 return (rs->journal_dev.mode == R5C_JOURNAL_MODE_WRITE_THROUGH) ? "A" : "a";
3414 else if (test_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags) ||
3415 (!test_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags) &&
3416 !test_bit(In_sync, &rdev->flags)))
3422 /* Helper to return resync/reshape progress for @rs and runtime flags for raid set in sync / resynching */
3423 static sector_t rs_get_progress(struct raid_set *rs, unsigned long recovery,
3424 enum sync_state state, sector_t resync_max_sectors)
3427 struct mddev *mddev = &rs->md;
3429 clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3430 clear_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3432 if (rs_is_raid0(rs)) {
3433 r = resync_max_sectors;
3434 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3437 if (state == st_idle && !test_bit(MD_RECOVERY_INTR, &recovery))
3438 r = mddev->recovery_cp;
3440 r = mddev->curr_resync_completed;
3442 if (state == st_idle && r >= resync_max_sectors) {
3446 /* In case we have finished recovering, the array is in sync. */
3447 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3448 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3450 } else if (state == st_recover)
3452 * In case we are recovering, the array is not in sync
3453 * and health chars should show the recovering legs.
3455 * Already retrieved recovery offset from curr_resync_completed above.
3459 else if (state == st_resync || state == st_reshape)
3461 * If "resync/reshape" is occurring, the raid set
3462 * is or may be out of sync hence the health
3463 * characters shall be 'a'.
3465 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3467 else if (state == st_check || state == st_repair)
3469 * If "check" or "repair" is occurring, the raid set has
3470 * undergone an initial sync and the health characters
3471 * should not be 'a' anymore.
3473 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3475 else if (test_bit(MD_RECOVERY_NEEDED, &recovery))
3477 * We are idle and recovery is needed, prevent 'A' chars race
3478 * caused by components still set to in-sync by constructor.
3480 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3484 * We are idle and the raid set may be doing an initial
3485 * sync, or it may be rebuilding individual components.
3486 * If all the devices are In_sync, then it is the raid set
3487 * that is being initialized.
3489 struct md_rdev *rdev;
3491 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3492 rdev_for_each(rdev, mddev)
3493 if (!test_bit(Journal, &rdev->flags) &&
3494 !test_bit(In_sync, &rdev->flags)) {
3495 clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3501 return min(r, resync_max_sectors);
3504 /* Helper to return @dev name or "-" if !@dev */
3505 static const char *__get_dev_name(struct dm_dev *dev)
3507 return dev ? dev->name : "-";
3510 static void raid_status(struct dm_target *ti, status_type_t type,
3511 unsigned int status_flags, char *result, unsigned int maxlen)
3513 struct raid_set *rs = ti->private;
3514 struct mddev *mddev = &rs->md;
3515 struct r5conf *conf = mddev->private;
3516 int i, max_nr_stripes = conf ? conf->max_nr_stripes : 0;
3517 unsigned long recovery;
3518 unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3519 unsigned int sz = 0;
3520 unsigned int rebuild_writemostly_count = 0;
3521 sector_t progress, resync_max_sectors, resync_mismatches;
3522 enum sync_state state;
3523 struct raid_type *rt;
3526 case STATUSTYPE_INFO:
3527 /* *Should* always succeed */
3528 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3532 DMEMIT("%s %d ", rt->name, mddev->raid_disks);
3534 /* Access most recent mddev properties for status output */
3536 /* Get sensible max sectors even if raid set not yet started */
3537 resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
3538 mddev->resync_max_sectors : mddev->dev_sectors;
3539 recovery = rs->md.recovery;
3540 state = decipher_sync_action(mddev, recovery);
3541 progress = rs_get_progress(rs, recovery, state, resync_max_sectors);
3542 resync_mismatches = (mddev->last_sync_action && !strcasecmp(mddev->last_sync_action, "check")) ?
3543 atomic64_read(&mddev->resync_mismatches) : 0;
3545 /* HM FIXME: do we want another state char for raid0? It shows 'D'/'A'/'-' now */
3546 for (i = 0; i < rs->raid_disks; i++)
3547 DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev));
3550 * In-sync/Reshape ratio:
3551 * The in-sync ratio shows the progress of:
3552 * - Initializing the raid set
3553 * - Rebuilding a subset of devices of the raid set
3554 * The user can distinguish between the two by referring
3555 * to the status characters.
3557 * The reshape ratio shows the progress of
3558 * changing the raid layout or the number of
3559 * disks of a raid set
3561 DMEMIT(" %llu/%llu", (unsigned long long) progress,
3562 (unsigned long long) resync_max_sectors);
3568 * See Documentation/admin-guide/device-mapper/dm-raid.rst for
3569 * information on each of these states.
3571 DMEMIT(" %s", sync_str(state));
3576 * resync_mismatches/mismatch_cnt
3577 * This field shows the number of discrepancies found when
3578 * performing a "check" of the raid set.
3580 DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3585 * data_offset (needed for out of space reshaping)
3586 * This field shows the data offset into the data
3587 * image LV where the first stripes data starts.
3589 * We keep data_offset equal on all raid disks of the set,
3590 * so retrieving it from the first raid disk is sufficient.
3592 DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
3597 DMEMIT(" %s", test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ?
3598 __raid_dev_status(rs, &rs->journal_dev.rdev) : "-");
3601 case STATUSTYPE_TABLE:
3602 /* Report the table line string you would use to construct this raid set */
3605 * Count any rebuild or writemostly argument pairs and subtract the
3606 * hweight count being added below of any rebuild and writemostly ctr flags.
3608 for (i = 0; i < rs->raid_disks; i++) {
3609 rebuild_writemostly_count += (test_bit(i, (void *) rs->rebuild_disks) ? 2 : 0) +
3610 (test_bit(WriteMostly, &rs->dev[i].rdev.flags) ? 2 : 0);
3612 rebuild_writemostly_count -= (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) ? 2 : 0) +
3613 (test_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags) ? 2 : 0);
3614 /* Calculate raid parameter count based on ^ rebuild/writemostly argument counts and ctr flags set. */
3615 raid_param_cnt += rebuild_writemostly_count +
3616 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3617 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2;
3618 /* Emit table line */
3619 /* This has to be in the documented order for userspace! */
3620 DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
3621 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
3622 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
3623 if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3624 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
3625 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags))
3626 for (i = 0; i < rs->raid_disks; i++)
3627 if (test_bit(i, (void *) rs->rebuild_disks))
3628 DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD), i);
3629 if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3630 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3631 mddev->bitmap_info.daemon_sleep);
3632 if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3633 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3634 mddev->sync_speed_min);
3635 if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3636 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3637 mddev->sync_speed_max);
3638 if (test_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags))
3639 for (i = 0; i < rs->raid_disks; i++)
3640 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3641 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3642 rs->dev[i].rdev.raid_disk);
3643 if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
3644 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
3645 mddev->bitmap_info.max_write_behind);
3646 if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3647 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3649 if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3650 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3651 (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3652 if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3653 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3654 raid10_md_layout_to_copies(mddev->layout));
3655 if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3656 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3657 raid10_md_layout_to_format(mddev->layout));
3658 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3659 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3660 max(rs->delta_disks, mddev->delta_disks));
3661 if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3662 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3663 (unsigned long long) rs->data_offset);
3664 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags))
3665 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV),
3666 __get_dev_name(rs->journal_dev.dev));
3667 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags))
3668 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE),
3669 md_journal_mode_to_dm_raid(rs->journal_dev.mode));
3670 DMEMIT(" %d", rs->raid_disks);
3671 for (i = 0; i < rs->raid_disks; i++)
3672 DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
3673 __get_dev_name(rs->dev[i].data_dev));
3676 case STATUSTYPE_IMA:
3677 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3681 DMEMIT_TARGET_NAME_VERSION(ti->type);
3682 DMEMIT(",raid_type=%s,raid_disks=%d", rt->name, mddev->raid_disks);
3684 /* Access most recent mddev properties for status output */
3686 recovery = rs->md.recovery;
3687 state = decipher_sync_action(mddev, recovery);
3688 DMEMIT(",raid_state=%s", sync_str(state));
3690 for (i = 0; i < rs->raid_disks; i++) {
3691 DMEMIT(",raid_device_%d_status=", i);
3692 DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev));
3695 if (rt_is_raid456(rt)) {
3696 DMEMIT(",journal_dev_mode=");
3697 switch (rs->journal_dev.mode) {
3698 case R5C_JOURNAL_MODE_WRITE_THROUGH:
3700 _raid456_journal_mode[R5C_JOURNAL_MODE_WRITE_THROUGH].param);
3702 case R5C_JOURNAL_MODE_WRITE_BACK:
3704 _raid456_journal_mode[R5C_JOURNAL_MODE_WRITE_BACK].param);
3716 static int raid_message(struct dm_target *ti, unsigned int argc, char **argv,
3717 char *result, unsigned maxlen)
3719 struct raid_set *rs = ti->private;
3720 struct mddev *mddev = &rs->md;
3722 if (!mddev->pers || !mddev->pers->sync_request)
3725 if (!strcasecmp(argv[0], "frozen"))
3726 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3728 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3730 if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
3731 if (mddev->sync_thread) {
3732 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3733 md_reap_sync_thread(mddev);
3735 } else if (decipher_sync_action(mddev, mddev->recovery) != st_idle)
3737 else if (!strcasecmp(argv[0], "resync"))
3738 ; /* MD_RECOVERY_NEEDED set below */
3739 else if (!strcasecmp(argv[0], "recover"))
3740 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3742 if (!strcasecmp(argv[0], "check")) {
3743 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3744 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3745 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3746 } else if (!strcasecmp(argv[0], "repair")) {
3747 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3748 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3752 if (mddev->ro == 2) {
3753 /* A write to sync_action is enough to justify
3754 * canceling read-auto mode
3757 if (!mddev->suspended && mddev->sync_thread)
3758 md_wakeup_thread(mddev->sync_thread);
3760 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3761 if (!mddev->suspended && mddev->thread)
3762 md_wakeup_thread(mddev->thread);
3767 static int raid_iterate_devices(struct dm_target *ti,
3768 iterate_devices_callout_fn fn, void *data)
3770 struct raid_set *rs = ti->private;
3774 for (i = 0; !r && i < rs->md.raid_disks; i++)
3775 if (rs->dev[i].data_dev)
3777 rs->dev[i].data_dev,
3778 0, /* No offset on data devs */
3785 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3787 struct raid_set *rs = ti->private;
3788 unsigned int chunk_size_bytes = to_bytes(rs->md.chunk_sectors);
3790 blk_limits_io_min(limits, chunk_size_bytes);
3791 blk_limits_io_opt(limits, chunk_size_bytes * mddev_data_stripes(rs));
3794 static void raid_postsuspend(struct dm_target *ti)
3796 struct raid_set *rs = ti->private;
3798 if (!test_and_set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
3799 /* Writes have to be stopped before suspending to avoid deadlocks. */
3800 if (!test_bit(MD_RECOVERY_FROZEN, &rs->md.recovery))
3801 md_stop_writes(&rs->md);
3803 mddev_lock_nointr(&rs->md);
3804 mddev_suspend(&rs->md);
3805 mddev_unlock(&rs->md);
3809 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
3812 uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
3813 unsigned long flags;
3814 bool cleared = false;
3815 struct dm_raid_superblock *sb;
3816 struct mddev *mddev = &rs->md;
3819 /* RAID personalities have to provide hot add/remove methods or we need to bail out. */
3820 if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
3823 memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
3825 for (i = 0; i < mddev->raid_disks; i++) {
3826 r = &rs->dev[i].rdev;
3827 /* HM FIXME: enhance journal device recovery processing */
3828 if (test_bit(Journal, &r->flags))
3831 if (test_bit(Faulty, &r->flags) &&
3832 r->meta_bdev && !read_disk_sb(r, r->sb_size, true)) {
3833 DMINFO("Faulty %s device #%d has readable super block."
3834 " Attempting to revive it.",
3835 rs->raid_type->name, i);
3838 * Faulty bit may be set, but sometimes the array can
3839 * be suspended before the personalities can respond
3840 * by removing the device from the array (i.e. calling
3841 * 'hot_remove_disk'). If they haven't yet removed
3842 * the failed device, its 'raid_disk' number will be
3843 * '>= 0' - meaning we must call this function
3847 clear_bit(In_sync, &r->flags); /* Mandatory for hot remove. */
3848 if (r->raid_disk >= 0) {
3849 if (mddev->pers->hot_remove_disk(mddev, r)) {
3850 /* Failed to revive this device, try next */
3855 r->raid_disk = r->saved_raid_disk = i;
3857 clear_bit(Faulty, &r->flags);
3858 clear_bit(WriteErrorSeen, &r->flags);
3860 if (mddev->pers->hot_add_disk(mddev, r)) {
3861 /* Failed to revive this device, try next */
3862 r->raid_disk = r->saved_raid_disk = -1;
3865 clear_bit(In_sync, &r->flags);
3866 r->recovery_offset = 0;
3867 set_bit(i, (void *) cleared_failed_devices);
3873 /* If any failed devices could be cleared, update all sbs failed_devices bits */
3875 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
3877 rdev_for_each(r, &rs->md) {
3878 if (test_bit(Journal, &r->flags))
3881 sb = page_address(r->sb_page);
3882 sb_retrieve_failed_devices(sb, failed_devices);
3884 for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
3885 failed_devices[i] &= ~cleared_failed_devices[i];
3887 sb_update_failed_devices(sb, failed_devices);
3892 static int __load_dirty_region_bitmap(struct raid_set *rs)
3896 /* Try loading the bitmap unless "raid0", which does not have one */
3897 if (!rs_is_raid0(rs) &&
3898 !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
3899 r = md_bitmap_load(&rs->md);
3901 DMERR("Failed to load bitmap");
3907 /* Enforce updating all superblocks */
3908 static void rs_update_sbs(struct raid_set *rs)
3910 struct mddev *mddev = &rs->md;
3913 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3915 md_update_sb(mddev, 1);
3920 * Reshape changes raid algorithm of @rs to new one within personality
3921 * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3922 * disks from a raid set thus growing/shrinking it or resizes the set
3924 * Call mddev_lock_nointr() before!
3926 static int rs_start_reshape(struct raid_set *rs)
3929 struct mddev *mddev = &rs->md;
3930 struct md_personality *pers = mddev->pers;
3932 /* Don't allow the sync thread to work until the table gets reloaded. */
3933 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
3935 r = rs_setup_reshape(rs);
3940 * Check any reshape constraints enforced by the personalility
3942 * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3944 r = pers->check_reshape(mddev);
3946 rs->ti->error = "pers->check_reshape() failed";
3951 * Personality may not provide start reshape method in which
3952 * case check_reshape above has already covered everything
3954 if (pers->start_reshape) {
3955 r = pers->start_reshape(mddev);
3957 rs->ti->error = "pers->start_reshape() failed";
3963 * Now reshape got set up, update superblocks to
3964 * reflect the fact so that a table reload will
3965 * access proper superblock content in the ctr.
3972 static int raid_preresume(struct dm_target *ti)
3975 struct raid_set *rs = ti->private;
3976 struct mddev *mddev = &rs->md;
3978 /* This is a resume after a suspend of the set -> it's already started. */
3979 if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
3983 * The superblocks need to be updated on disk if the
3984 * array is new or new devices got added (thus zeroed
3985 * out by userspace) or __load_dirty_region_bitmap
3986 * will overwrite them in core with old data or fail.
3988 if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
3991 /* Load the bitmap from disk unless raid0 */
3992 r = __load_dirty_region_bitmap(rs);
3996 /* We are extending the raid set size, adjust mddev/md_rdev sizes and set capacity. */
3997 if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags)) {
3998 mddev->array_sectors = rs->array_sectors;
3999 mddev->dev_sectors = rs->dev_sectors;
4000 rs_set_rdev_sectors(rs);
4001 rs_set_capacity(rs);
4004 /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) or grown device size */
4005 if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) && mddev->bitmap &&
4006 (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags) ||
4007 (rs->requested_bitmap_chunk_sectors &&
4008 mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)))) {
4009 int chunksize = to_bytes(rs->requested_bitmap_chunk_sectors) ?: mddev->bitmap_info.chunksize;
4011 r = md_bitmap_resize(mddev->bitmap, mddev->dev_sectors, chunksize, 0);
4013 DMERR("Failed to resize bitmap");
4016 /* Check for any resize/reshape on @rs and adjust/initiate */
4017 /* Be prepared for mddev_resume() in raid_resume() */
4018 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4019 if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
4020 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4021 mddev->resync_min = mddev->recovery_cp;
4022 if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags))
4023 mddev->resync_max_sectors = mddev->dev_sectors;
4026 /* Check for any reshape request unless new raid set */
4027 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
4028 /* Initiate a reshape. */
4029 rs_set_rdev_sectors(rs);
4030 mddev_lock_nointr(mddev);
4031 r = rs_start_reshape(rs);
4032 mddev_unlock(mddev);
4034 DMWARN("Failed to check/start reshape, continuing without change");
4041 static void raid_resume(struct dm_target *ti)
4043 struct raid_set *rs = ti->private;
4044 struct mddev *mddev = &rs->md;
4046 if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
4048 * A secondary resume while the device is active.
4049 * Take this opportunity to check whether any failed
4050 * devices are reachable again.
4052 attempt_restore_of_faulty_devices(rs);
4055 if (test_and_clear_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
4056 /* Only reduce raid set size before running a disk removing reshape. */
4057 if (mddev->delta_disks < 0)
4058 rs_set_capacity(rs);
4060 mddev_lock_nointr(mddev);
4061 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4064 mddev_resume(mddev);
4065 mddev_unlock(mddev);
4069 static struct target_type raid_target = {
4071 .version = {1, 15, 1},
4072 .module = THIS_MODULE,
4076 .status = raid_status,
4077 .message = raid_message,
4078 .iterate_devices = raid_iterate_devices,
4079 .io_hints = raid_io_hints,
4080 .postsuspend = raid_postsuspend,
4081 .preresume = raid_preresume,
4082 .resume = raid_resume,
4085 static int __init dm_raid_init(void)
4087 DMINFO("Loading target version %u.%u.%u",
4088 raid_target.version[0],
4089 raid_target.version[1],
4090 raid_target.version[2]);
4091 return dm_register_target(&raid_target);
4094 static void __exit dm_raid_exit(void)
4096 dm_unregister_target(&raid_target);
4099 module_init(dm_raid_init);
4100 module_exit(dm_raid_exit);
4102 module_param(devices_handle_discard_safely, bool, 0644);
4103 MODULE_PARM_DESC(devices_handle_discard_safely,
4104 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
4106 MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
4107 MODULE_ALIAS("dm-raid0");
4108 MODULE_ALIAS("dm-raid1");
4109 MODULE_ALIAS("dm-raid10");
4110 MODULE_ALIAS("dm-raid4");
4111 MODULE_ALIAS("dm-raid5");
4112 MODULE_ALIAS("dm-raid6");
4113 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
4114 MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
4115 MODULE_LICENSE("GPL");
projects / linux-2.6-microblaze.git / blob