2 * Copyright (C) 2010-2011 Neil Brown
3 * Copyright (C) 2010-2018 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
8 #include <linux/slab.h>
9 #include <linux/module.h>
15 #include "md-bitmap.h"
17 #include <linux/device-mapper.h>
19 #define DM_MSG_PREFIX "raid"
20 #define MAX_RAID_DEVICES 253 /* md-raid kernel limit */
23 * Minimum sectors of free reshape space per raid device
25 #define MIN_FREE_RESHAPE_SPACE to_sector(4*4096)
28 * Minimum journal space 4 MiB in sectors.
30 #define MIN_RAID456_JOURNAL_SPACE (4*2048)
32 static bool devices_handle_discard_safely = false;
35 * The following flags are used by dm-raid.c to set up the array state.
36 * They must be cleared before md_run is called.
38 #define FirstUse 10 /* rdev flag */
42 * Two DM devices, one to hold metadata and one to hold the
43 * actual data/parity. The reason for this is to not confuse
44 * ti->len and give more flexibility in altering size and
47 * While it is possible for this device to be associated
48 * with a different physical device than the data_dev, it
49 * is intended for it to be the same.
50 * |--------- Physical Device ---------|
51 * |- meta_dev -|------ data_dev ------|
53 struct dm_dev *meta_dev;
54 struct dm_dev *data_dev;
59 * Bits for establishing rs->ctr_flags
64 #define __CTR_FLAG_SYNC 0 /* 1 */ /* Not with raid0! */
65 #define __CTR_FLAG_NOSYNC 1 /* 1 */ /* Not with raid0! */
66 #define __CTR_FLAG_REBUILD 2 /* 2 */ /* Not with raid0! */
67 #define __CTR_FLAG_DAEMON_SLEEP 3 /* 2 */ /* Not with raid0! */
68 #define __CTR_FLAG_MIN_RECOVERY_RATE 4 /* 2 */ /* Not with raid0! */
69 #define __CTR_FLAG_MAX_RECOVERY_RATE 5 /* 2 */ /* Not with raid0! */
70 #define __CTR_FLAG_MAX_WRITE_BEHIND 6 /* 2 */ /* Only with raid1! */
71 #define __CTR_FLAG_WRITE_MOSTLY 7 /* 2 */ /* Only with raid1! */
72 #define __CTR_FLAG_STRIPE_CACHE 8 /* 2 */ /* Only with raid4/5/6! */
73 #define __CTR_FLAG_REGION_SIZE 9 /* 2 */ /* Not with raid0! */
74 #define __CTR_FLAG_RAID10_COPIES 10 /* 2 */ /* Only with raid10 */
75 #define __CTR_FLAG_RAID10_FORMAT 11 /* 2 */ /* Only with raid10 */
77 #define __CTR_FLAG_DELTA_DISKS 12 /* 2 */ /* Only with reshapable raid1/4/5/6/10! */
78 #define __CTR_FLAG_DATA_OFFSET 13 /* 2 */ /* Only with reshapable raid4/5/6/10! */
79 #define __CTR_FLAG_RAID10_USE_NEAR_SETS 14 /* 2 */ /* Only with raid10! */
82 #define __CTR_FLAG_JOURNAL_DEV 15 /* 2 */ /* Only with raid4/5/6 (journal device)! */
85 #define __CTR_FLAG_JOURNAL_MODE 16 /* 2 */ /* Only with raid4/5/6 (journal mode)! */
88 * Flags for rs->ctr_flags field.
90 #define CTR_FLAG_SYNC (1 << __CTR_FLAG_SYNC)
91 #define CTR_FLAG_NOSYNC (1 << __CTR_FLAG_NOSYNC)
92 #define CTR_FLAG_REBUILD (1 << __CTR_FLAG_REBUILD)
93 #define CTR_FLAG_DAEMON_SLEEP (1 << __CTR_FLAG_DAEMON_SLEEP)
94 #define CTR_FLAG_MIN_RECOVERY_RATE (1 << __CTR_FLAG_MIN_RECOVERY_RATE)
95 #define CTR_FLAG_MAX_RECOVERY_RATE (1 << __CTR_FLAG_MAX_RECOVERY_RATE)
96 #define CTR_FLAG_MAX_WRITE_BEHIND (1 << __CTR_FLAG_MAX_WRITE_BEHIND)
97 #define CTR_FLAG_WRITE_MOSTLY (1 << __CTR_FLAG_WRITE_MOSTLY)
98 #define CTR_FLAG_STRIPE_CACHE (1 << __CTR_FLAG_STRIPE_CACHE)
99 #define CTR_FLAG_REGION_SIZE (1 << __CTR_FLAG_REGION_SIZE)
100 #define CTR_FLAG_RAID10_COPIES (1 << __CTR_FLAG_RAID10_COPIES)
101 #define CTR_FLAG_RAID10_FORMAT (1 << __CTR_FLAG_RAID10_FORMAT)
102 #define CTR_FLAG_DELTA_DISKS (1 << __CTR_FLAG_DELTA_DISKS)
103 #define CTR_FLAG_DATA_OFFSET (1 << __CTR_FLAG_DATA_OFFSET)
104 #define CTR_FLAG_RAID10_USE_NEAR_SETS (1 << __CTR_FLAG_RAID10_USE_NEAR_SETS)
105 #define CTR_FLAG_JOURNAL_DEV (1 << __CTR_FLAG_JOURNAL_DEV)
106 #define CTR_FLAG_JOURNAL_MODE (1 << __CTR_FLAG_JOURNAL_MODE)
109 * Definitions of various constructor flags to
110 * be used in checks of valid / invalid flags
113 /* Define all any sync flags */
114 #define CTR_FLAGS_ANY_SYNC (CTR_FLAG_SYNC | CTR_FLAG_NOSYNC)
116 /* Define flags for options without argument (e.g. 'nosync') */
117 #define CTR_FLAG_OPTIONS_NO_ARGS (CTR_FLAGS_ANY_SYNC | \
118 CTR_FLAG_RAID10_USE_NEAR_SETS)
120 /* Define flags for options with one argument (e.g. 'delta_disks +2') */
121 #define CTR_FLAG_OPTIONS_ONE_ARG (CTR_FLAG_REBUILD | \
122 CTR_FLAG_WRITE_MOSTLY | \
123 CTR_FLAG_DAEMON_SLEEP | \
124 CTR_FLAG_MIN_RECOVERY_RATE | \
125 CTR_FLAG_MAX_RECOVERY_RATE | \
126 CTR_FLAG_MAX_WRITE_BEHIND | \
127 CTR_FLAG_STRIPE_CACHE | \
128 CTR_FLAG_REGION_SIZE | \
129 CTR_FLAG_RAID10_COPIES | \
130 CTR_FLAG_RAID10_FORMAT | \
131 CTR_FLAG_DELTA_DISKS | \
132 CTR_FLAG_DATA_OFFSET)
134 /* Valid options definitions per raid level... */
136 /* "raid0" does only accept data offset */
137 #define RAID0_VALID_FLAGS (CTR_FLAG_DATA_OFFSET)
139 /* "raid1" does not accept stripe cache, data offset, delta_disks or any raid10 options */
140 #define RAID1_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
142 CTR_FLAG_WRITE_MOSTLY | \
143 CTR_FLAG_DAEMON_SLEEP | \
144 CTR_FLAG_MIN_RECOVERY_RATE | \
145 CTR_FLAG_MAX_RECOVERY_RATE | \
146 CTR_FLAG_MAX_WRITE_BEHIND | \
147 CTR_FLAG_REGION_SIZE | \
148 CTR_FLAG_DELTA_DISKS | \
149 CTR_FLAG_DATA_OFFSET)
151 /* "raid10" does not accept any raid1 or stripe cache options */
152 #define RAID10_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
154 CTR_FLAG_DAEMON_SLEEP | \
155 CTR_FLAG_MIN_RECOVERY_RATE | \
156 CTR_FLAG_MAX_RECOVERY_RATE | \
157 CTR_FLAG_REGION_SIZE | \
158 CTR_FLAG_RAID10_COPIES | \
159 CTR_FLAG_RAID10_FORMAT | \
160 CTR_FLAG_DELTA_DISKS | \
161 CTR_FLAG_DATA_OFFSET | \
162 CTR_FLAG_RAID10_USE_NEAR_SETS)
165 * "raid4/5/6" do not accept any raid1 or raid10 specific options
167 * "raid6" does not accept "nosync", because it is not guaranteed
168 * that both parity and q-syndrome are being written properly with
171 #define RAID45_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
173 CTR_FLAG_DAEMON_SLEEP | \
174 CTR_FLAG_MIN_RECOVERY_RATE | \
175 CTR_FLAG_MAX_RECOVERY_RATE | \
176 CTR_FLAG_STRIPE_CACHE | \
177 CTR_FLAG_REGION_SIZE | \
178 CTR_FLAG_DELTA_DISKS | \
179 CTR_FLAG_DATA_OFFSET | \
180 CTR_FLAG_JOURNAL_DEV | \
181 CTR_FLAG_JOURNAL_MODE)
183 #define RAID6_VALID_FLAGS (CTR_FLAG_SYNC | \
185 CTR_FLAG_DAEMON_SLEEP | \
186 CTR_FLAG_MIN_RECOVERY_RATE | \
187 CTR_FLAG_MAX_RECOVERY_RATE | \
188 CTR_FLAG_STRIPE_CACHE | \
189 CTR_FLAG_REGION_SIZE | \
190 CTR_FLAG_DELTA_DISKS | \
191 CTR_FLAG_DATA_OFFSET | \
192 CTR_FLAG_JOURNAL_DEV | \
193 CTR_FLAG_JOURNAL_MODE)
194 /* ...valid options definitions per raid level */
197 * Flags for rs->runtime_flags field
198 * (RT_FLAG prefix meaning "runtime flag")
200 * These are all internal and used to define runtime state,
201 * e.g. to prevent another resume from preresume processing
202 * the raid set all over again.
204 #define RT_FLAG_RS_PRERESUMED 0
205 #define RT_FLAG_RS_RESUMED 1
206 #define RT_FLAG_RS_BITMAP_LOADED 2
207 #define RT_FLAG_UPDATE_SBS 3
208 #define RT_FLAG_RESHAPE_RS 4
209 #define RT_FLAG_RS_SUSPENDED 5
210 #define RT_FLAG_RS_IN_SYNC 6
211 #define RT_FLAG_RS_RESYNCING 7
213 /* Array elements of 64 bit needed for rebuild/failed disk bits */
214 #define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
217 * raid set level, layout and chunk sectors backup/restore
222 int new_chunk_sectors;
226 struct dm_target *ti;
228 uint32_t stripe_cache_entries;
229 unsigned long ctr_flags;
230 unsigned long runtime_flags;
232 uint64_t rebuild_disks[DISKS_ARRAY_ELEMS];
238 int requested_bitmap_chunk_sectors;
241 struct raid_type *raid_type;
242 struct dm_target_callbacks callbacks;
244 /* Optional raid4/5/6 journal device */
251 struct raid_dev dev[0];
254 static void rs_config_backup(struct raid_set *rs, struct rs_layout *l)
256 struct mddev *mddev = &rs->md;
258 l->new_level = mddev->new_level;
259 l->new_layout = mddev->new_layout;
260 l->new_chunk_sectors = mddev->new_chunk_sectors;
263 static void rs_config_restore(struct raid_set *rs, struct rs_layout *l)
265 struct mddev *mddev = &rs->md;
267 mddev->new_level = l->new_level;
268 mddev->new_layout = l->new_layout;
269 mddev->new_chunk_sectors = l->new_chunk_sectors;
272 /* raid10 algorithms (i.e. formats) */
273 #define ALGORITHM_RAID10_DEFAULT 0
274 #define ALGORITHM_RAID10_NEAR 1
275 #define ALGORITHM_RAID10_OFFSET 2
276 #define ALGORITHM_RAID10_FAR 3
278 /* Supported raid types and properties. */
279 static struct raid_type {
280 const char *name; /* RAID algorithm. */
281 const char *descr; /* Descriptor text for logging. */
282 const unsigned int parity_devs; /* # of parity devices. */
283 const unsigned int minimal_devs;/* minimal # of devices in set. */
284 const unsigned int level; /* RAID level. */
285 const unsigned int algorithm; /* RAID algorithm. */
287 {"raid0", "raid0 (striping)", 0, 2, 0, 0 /* NONE */},
288 {"raid1", "raid1 (mirroring)", 0, 2, 1, 0 /* NONE */},
289 {"raid10_far", "raid10 far (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_FAR},
290 {"raid10_offset", "raid10 offset (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_OFFSET},
291 {"raid10_near", "raid10 near (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_NEAR},
292 {"raid10", "raid10 (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_DEFAULT},
293 {"raid4", "raid4 (dedicated first parity disk)", 1, 2, 5, ALGORITHM_PARITY_0}, /* raid4 layout = raid5_0 */
294 {"raid5_n", "raid5 (dedicated last parity disk)", 1, 2, 5, ALGORITHM_PARITY_N},
295 {"raid5_ls", "raid5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
296 {"raid5_rs", "raid5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
297 {"raid5_la", "raid5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
298 {"raid5_ra", "raid5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
299 {"raid6_zr", "raid6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
300 {"raid6_nr", "raid6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
301 {"raid6_nc", "raid6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE},
302 {"raid6_n_6", "raid6 (dedicated parity/Q n/6)", 2, 4, 6, ALGORITHM_PARITY_N_6},
303 {"raid6_ls_6", "raid6 (left symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_SYMMETRIC_6},
304 {"raid6_rs_6", "raid6 (right symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_SYMMETRIC_6},
305 {"raid6_la_6", "raid6 (left asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_ASYMMETRIC_6},
306 {"raid6_ra_6", "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_ASYMMETRIC_6}
309 /* True, if @v is in inclusive range [@min, @max] */
310 static bool __within_range(long v, long min, long max)
312 return v >= min && v <= max;
315 /* All table line arguments are defined here */
316 static struct arg_name_flag {
317 const unsigned long flag;
319 } __arg_name_flags[] = {
320 { CTR_FLAG_SYNC, "sync"},
321 { CTR_FLAG_NOSYNC, "nosync"},
322 { CTR_FLAG_REBUILD, "rebuild"},
323 { CTR_FLAG_DAEMON_SLEEP, "daemon_sleep"},
324 { CTR_FLAG_MIN_RECOVERY_RATE, "min_recovery_rate"},
325 { CTR_FLAG_MAX_RECOVERY_RATE, "max_recovery_rate"},
326 { CTR_FLAG_MAX_WRITE_BEHIND, "max_write_behind"},
327 { CTR_FLAG_WRITE_MOSTLY, "write_mostly"},
328 { CTR_FLAG_STRIPE_CACHE, "stripe_cache"},
329 { CTR_FLAG_REGION_SIZE, "region_size"},
330 { CTR_FLAG_RAID10_COPIES, "raid10_copies"},
331 { CTR_FLAG_RAID10_FORMAT, "raid10_format"},
332 { CTR_FLAG_DATA_OFFSET, "data_offset"},
333 { CTR_FLAG_DELTA_DISKS, "delta_disks"},
334 { CTR_FLAG_RAID10_USE_NEAR_SETS, "raid10_use_near_sets"},
335 { CTR_FLAG_JOURNAL_DEV, "journal_dev" },
336 { CTR_FLAG_JOURNAL_MODE, "journal_mode" },
339 /* Return argument name string for given @flag */
340 static const char *dm_raid_arg_name_by_flag(const uint32_t flag)
342 if (hweight32(flag) == 1) {
343 struct arg_name_flag *anf = __arg_name_flags + ARRAY_SIZE(__arg_name_flags);
345 while (anf-- > __arg_name_flags)
346 if (flag & anf->flag)
350 DMERR("%s called with more than one flag!", __func__);
355 /* Define correlation of raid456 journal cache modes and dm-raid target line parameters */
359 } _raid456_journal_mode[] = {
360 { R5C_JOURNAL_MODE_WRITE_THROUGH , "writethrough" },
361 { R5C_JOURNAL_MODE_WRITE_BACK , "writeback" }
364 /* Return MD raid4/5/6 journal mode for dm @journal_mode one */
365 static int dm_raid_journal_mode_to_md(const char *mode)
367 int m = ARRAY_SIZE(_raid456_journal_mode);
370 if (!strcasecmp(mode, _raid456_journal_mode[m].param))
371 return _raid456_journal_mode[m].mode;
376 /* Return dm-raid raid4/5/6 journal mode string for @mode */
377 static const char *md_journal_mode_to_dm_raid(const int mode)
379 int m = ARRAY_SIZE(_raid456_journal_mode);
382 if (mode == _raid456_journal_mode[m].mode)
383 return _raid456_journal_mode[m].param;
389 * Bool helpers to test for various raid levels of a raid set.
390 * It's level as reported by the superblock rather than
391 * the requested raid_type passed to the constructor.
393 /* Return true, if raid set in @rs is raid0 */
394 static bool rs_is_raid0(struct raid_set *rs)
396 return !rs->md.level;
399 /* Return true, if raid set in @rs is raid1 */
400 static bool rs_is_raid1(struct raid_set *rs)
402 return rs->md.level == 1;
405 /* Return true, if raid set in @rs is raid10 */
406 static bool rs_is_raid10(struct raid_set *rs)
408 return rs->md.level == 10;
411 /* Return true, if raid set in @rs is level 6 */
412 static bool rs_is_raid6(struct raid_set *rs)
414 return rs->md.level == 6;
417 /* Return true, if raid set in @rs is level 4, 5 or 6 */
418 static bool rs_is_raid456(struct raid_set *rs)
420 return __within_range(rs->md.level, 4, 6);
423 /* Return true, if raid set in @rs is reshapable */
424 static bool __is_raid10_far(int layout);
425 static bool rs_is_reshapable(struct raid_set *rs)
427 return rs_is_raid456(rs) ||
428 (rs_is_raid10(rs) && !__is_raid10_far(rs->md.new_layout));
431 /* Return true, if raid set in @rs is recovering */
432 static bool rs_is_recovering(struct raid_set *rs)
434 return rs->md.recovery_cp < rs->md.dev_sectors;
437 /* Return true, if raid set in @rs is reshaping */
438 static bool rs_is_reshaping(struct raid_set *rs)
440 return rs->md.reshape_position != MaxSector;
444 * bool helpers to test for various raid levels of a raid type @rt
447 /* Return true, if raid type in @rt is raid0 */
448 static bool rt_is_raid0(struct raid_type *rt)
453 /* Return true, if raid type in @rt is raid1 */
454 static bool rt_is_raid1(struct raid_type *rt)
456 return rt->level == 1;
459 /* Return true, if raid type in @rt is raid10 */
460 static bool rt_is_raid10(struct raid_type *rt)
462 return rt->level == 10;
465 /* Return true, if raid type in @rt is raid4/5 */
466 static bool rt_is_raid45(struct raid_type *rt)
468 return __within_range(rt->level, 4, 5);
471 /* Return true, if raid type in @rt is raid6 */
472 static bool rt_is_raid6(struct raid_type *rt)
474 return rt->level == 6;
477 /* Return true, if raid type in @rt is raid4/5/6 */
478 static bool rt_is_raid456(struct raid_type *rt)
480 return __within_range(rt->level, 4, 6);
482 /* END: raid level bools */
484 /* Return valid ctr flags for the raid level of @rs */
485 static unsigned long __valid_flags(struct raid_set *rs)
487 if (rt_is_raid0(rs->raid_type))
488 return RAID0_VALID_FLAGS;
489 else if (rt_is_raid1(rs->raid_type))
490 return RAID1_VALID_FLAGS;
491 else if (rt_is_raid10(rs->raid_type))
492 return RAID10_VALID_FLAGS;
493 else if (rt_is_raid45(rs->raid_type))
494 return RAID45_VALID_FLAGS;
495 else if (rt_is_raid6(rs->raid_type))
496 return RAID6_VALID_FLAGS;
502 * Check for valid flags set on @rs
504 * Has to be called after parsing of the ctr flags!
506 static int rs_check_for_valid_flags(struct raid_set *rs)
508 if (rs->ctr_flags & ~__valid_flags(rs)) {
509 rs->ti->error = "Invalid flags combination";
516 /* MD raid10 bit definitions and helpers */
517 #define RAID10_OFFSET (1 << 16) /* stripes with data copies area adjacent on devices */
518 #define RAID10_BROCKEN_USE_FAR_SETS (1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
519 #define RAID10_USE_FAR_SETS (1 << 18) /* Use sets instead of whole stripe rotation */
520 #define RAID10_FAR_COPIES_SHIFT 8 /* raid10 # far copies shift (2nd byte of layout) */
522 /* Return md raid10 near copies for @layout */
523 static unsigned int __raid10_near_copies(int layout)
525 return layout & 0xFF;
528 /* Return md raid10 far copies for @layout */
529 static unsigned int __raid10_far_copies(int layout)
531 return __raid10_near_copies(layout >> RAID10_FAR_COPIES_SHIFT);
534 /* Return true if md raid10 offset for @layout */
535 static bool __is_raid10_offset(int layout)
537 return !!(layout & RAID10_OFFSET);
540 /* Return true if md raid10 near for @layout */
541 static bool __is_raid10_near(int layout)
543 return !__is_raid10_offset(layout) && __raid10_near_copies(layout) > 1;
546 /* Return true if md raid10 far for @layout */
547 static bool __is_raid10_far(int layout)
549 return !__is_raid10_offset(layout) && __raid10_far_copies(layout) > 1;
552 /* Return md raid10 layout string for @layout */
553 static const char *raid10_md_layout_to_format(int layout)
556 * Bit 16 stands for "offset"
557 * (i.e. adjacent stripes hold copies)
559 * Refer to MD's raid10.c for details
561 if (__is_raid10_offset(layout))
564 if (__raid10_near_copies(layout) > 1)
567 if (__raid10_far_copies(layout) > 1)
573 /* Return md raid10 algorithm for @name */
574 static int raid10_name_to_format(const char *name)
576 if (!strcasecmp(name, "near"))
577 return ALGORITHM_RAID10_NEAR;
578 else if (!strcasecmp(name, "offset"))
579 return ALGORITHM_RAID10_OFFSET;
580 else if (!strcasecmp(name, "far"))
581 return ALGORITHM_RAID10_FAR;
586 /* Return md raid10 copies for @layout */
587 static unsigned int raid10_md_layout_to_copies(int layout)
589 return max(__raid10_near_copies(layout), __raid10_far_copies(layout));
592 /* Return md raid10 format id for @format string */
593 static int raid10_format_to_md_layout(struct raid_set *rs,
594 unsigned int algorithm,
597 unsigned int n = 1, f = 1, r = 0;
600 * MD resilienece flaw:
602 * enabling use_far_sets for far/offset formats causes copies
603 * to be colocated on the same devs together with their origins!
605 * -> disable it for now in the definition above
607 if (algorithm == ALGORITHM_RAID10_DEFAULT ||
608 algorithm == ALGORITHM_RAID10_NEAR)
611 else if (algorithm == ALGORITHM_RAID10_OFFSET) {
614 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
615 r |= RAID10_USE_FAR_SETS;
617 } else if (algorithm == ALGORITHM_RAID10_FAR) {
620 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
621 r |= RAID10_USE_FAR_SETS;
626 return r | (f << RAID10_FAR_COPIES_SHIFT) | n;
628 /* END: MD raid10 bit definitions and helpers */
630 /* Check for any of the raid10 algorithms */
631 static bool __got_raid10(struct raid_type *rtp, const int layout)
633 if (rtp->level == 10) {
634 switch (rtp->algorithm) {
635 case ALGORITHM_RAID10_DEFAULT:
636 case ALGORITHM_RAID10_NEAR:
637 return __is_raid10_near(layout);
638 case ALGORITHM_RAID10_OFFSET:
639 return __is_raid10_offset(layout);
640 case ALGORITHM_RAID10_FAR:
641 return __is_raid10_far(layout);
650 /* Return raid_type for @name */
651 static struct raid_type *get_raid_type(const char *name)
653 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
655 while (rtp-- > raid_types)
656 if (!strcasecmp(rtp->name, name))
662 /* Return raid_type for @name based derived from @level and @layout */
663 static struct raid_type *get_raid_type_by_ll(const int level, const int layout)
665 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
667 while (rtp-- > raid_types) {
668 /* RAID10 special checks based on @layout flags/properties */
669 if (rtp->level == level &&
670 (__got_raid10(rtp, layout) || rtp->algorithm == layout))
677 /* Adjust rdev sectors */
678 static void rs_set_rdev_sectors(struct raid_set *rs)
680 struct mddev *mddev = &rs->md;
681 struct md_rdev *rdev;
684 * raid10 sets rdev->sector to the device size, which
685 * is unintended in case of out-of-place reshaping
687 rdev_for_each(rdev, mddev)
688 if (!test_bit(Journal, &rdev->flags))
689 rdev->sectors = mddev->dev_sectors;
693 * Change bdev capacity of @rs in case of a disk add/remove reshape
695 static void rs_set_capacity(struct raid_set *rs)
697 struct gendisk *gendisk = dm_disk(dm_table_get_md(rs->ti->table));
699 set_capacity(gendisk, rs->md.array_sectors);
700 revalidate_disk(gendisk);
704 * Set the mddev properties in @rs to the current
705 * ones retrieved from the freshest superblock
707 static void rs_set_cur(struct raid_set *rs)
709 struct mddev *mddev = &rs->md;
711 mddev->new_level = mddev->level;
712 mddev->new_layout = mddev->layout;
713 mddev->new_chunk_sectors = mddev->chunk_sectors;
717 * Set the mddev properties in @rs to the new
718 * ones requested by the ctr
720 static void rs_set_new(struct raid_set *rs)
722 struct mddev *mddev = &rs->md;
724 mddev->level = mddev->new_level;
725 mddev->layout = mddev->new_layout;
726 mddev->chunk_sectors = mddev->new_chunk_sectors;
727 mddev->raid_disks = rs->raid_disks;
728 mddev->delta_disks = 0;
731 static struct raid_set *raid_set_alloc(struct dm_target *ti, struct raid_type *raid_type,
732 unsigned int raid_devs)
737 if (raid_devs <= raid_type->parity_devs) {
738 ti->error = "Insufficient number of devices";
739 return ERR_PTR(-EINVAL);
742 rs = kzalloc(struct_size(rs, dev, raid_devs), GFP_KERNEL);
744 ti->error = "Cannot allocate raid context";
745 return ERR_PTR(-ENOMEM);
750 rs->raid_disks = raid_devs;
754 rs->raid_type = raid_type;
755 rs->stripe_cache_entries = 256;
756 rs->md.raid_disks = raid_devs;
757 rs->md.level = raid_type->level;
758 rs->md.new_level = rs->md.level;
759 rs->md.layout = raid_type->algorithm;
760 rs->md.new_layout = rs->md.layout;
761 rs->md.delta_disks = 0;
762 rs->md.recovery_cp = MaxSector;
764 for (i = 0; i < raid_devs; i++)
765 md_rdev_init(&rs->dev[i].rdev);
768 * Remaining items to be initialized by further RAID params:
771 * rs->md.chunk_sectors
772 * rs->md.new_chunk_sectors
779 /* Free all @rs allocations */
780 static void raid_set_free(struct raid_set *rs)
784 if (rs->journal_dev.dev) {
785 md_rdev_clear(&rs->journal_dev.rdev);
786 dm_put_device(rs->ti, rs->journal_dev.dev);
789 for (i = 0; i < rs->raid_disks; i++) {
790 if (rs->dev[i].meta_dev)
791 dm_put_device(rs->ti, rs->dev[i].meta_dev);
792 md_rdev_clear(&rs->dev[i].rdev);
793 if (rs->dev[i].data_dev)
794 dm_put_device(rs->ti, rs->dev[i].data_dev);
801 * For every device we have two words
802 * <meta_dev>: meta device name or '-' if missing
803 * <data_dev>: data device name or '-' if missing
805 * The following are permitted:
808 * <meta_dev> <data_dev>
810 * The following is not allowed:
813 * This code parses those words. If there is a failure,
814 * the caller must use raid_set_free() to unwind the operations.
816 static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
820 int metadata_available = 0;
824 /* Put off the number of raid devices argument to get to dev pairs */
825 arg = dm_shift_arg(as);
829 for (i = 0; i < rs->raid_disks; i++) {
830 rs->dev[i].rdev.raid_disk = i;
832 rs->dev[i].meta_dev = NULL;
833 rs->dev[i].data_dev = NULL;
836 * There are no offsets initially.
837 * Out of place reshape will set them accordingly.
839 rs->dev[i].rdev.data_offset = 0;
840 rs->dev[i].rdev.new_data_offset = 0;
841 rs->dev[i].rdev.mddev = &rs->md;
843 arg = dm_shift_arg(as);
847 if (strcmp(arg, "-")) {
848 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
849 &rs->dev[i].meta_dev);
851 rs->ti->error = "RAID metadata device lookup failure";
855 rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
856 if (!rs->dev[i].rdev.sb_page) {
857 rs->ti->error = "Failed to allocate superblock page";
862 arg = dm_shift_arg(as);
866 if (!strcmp(arg, "-")) {
867 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
868 (!rs->dev[i].rdev.recovery_offset)) {
869 rs->ti->error = "Drive designated for rebuild not specified";
873 if (rs->dev[i].meta_dev) {
874 rs->ti->error = "No data device supplied with metadata device";
881 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
882 &rs->dev[i].data_dev);
884 rs->ti->error = "RAID device lookup failure";
888 if (rs->dev[i].meta_dev) {
889 metadata_available = 1;
890 rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
892 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
893 list_add_tail(&rs->dev[i].rdev.same_set, &rs->md.disks);
894 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
898 if (rs->journal_dev.dev)
899 list_add_tail(&rs->journal_dev.rdev.same_set, &rs->md.disks);
901 if (metadata_available) {
903 rs->md.persistent = 1;
904 rs->md.major_version = 2;
905 } else if (rebuild && !rs->md.recovery_cp) {
907 * Without metadata, we will not be able to tell if the array
908 * is in-sync or not - we must assume it is not. Therefore,
909 * it is impossible to rebuild a drive.
911 * Even if there is metadata, the on-disk information may
912 * indicate that the array is not in-sync and it will then
915 * User could specify 'nosync' option if desperate.
917 rs->ti->error = "Unable to rebuild drive while array is not in-sync";
925 * validate_region_size
927 * @region_size: region size in sectors. If 0, pick a size (4MiB default).
929 * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
930 * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
932 * Returns: 0 on success, -EINVAL on failure.
934 static int validate_region_size(struct raid_set *rs, unsigned long region_size)
936 unsigned long min_region_size = rs->ti->len / (1 << 21);
943 * Choose a reasonable default. All figures in sectors.
945 if (min_region_size > (1 << 13)) {
946 /* If not a power of 2, make it the next power of 2 */
947 region_size = roundup_pow_of_two(min_region_size);
948 DMINFO("Choosing default region size of %lu sectors",
951 DMINFO("Choosing default region size of 4MiB");
952 region_size = 1 << 13; /* sectors */
956 * Validate user-supplied value.
958 if (region_size > rs->ti->len) {
959 rs->ti->error = "Supplied region size is too large";
963 if (region_size < min_region_size) {
964 DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
965 region_size, min_region_size);
966 rs->ti->error = "Supplied region size is too small";
970 if (!is_power_of_2(region_size)) {
971 rs->ti->error = "Region size is not a power of 2";
975 if (region_size < rs->md.chunk_sectors) {
976 rs->ti->error = "Region size is smaller than the chunk size";
982 * Convert sectors to bytes.
984 rs->md.bitmap_info.chunksize = to_bytes(region_size);
990 * validate_raid_redundancy
993 * Determine if there are enough devices in the array that haven't
994 * failed (or are being rebuilt) to form a usable array.
996 * Returns: 0 on success, -EINVAL on failure.
998 static int validate_raid_redundancy(struct raid_set *rs)
1000 unsigned int i, rebuild_cnt = 0;
1001 unsigned int rebuilds_per_group = 0, copies;
1002 unsigned int group_size, last_group_start;
1004 for (i = 0; i < rs->md.raid_disks; i++)
1005 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
1006 !rs->dev[i].rdev.sb_page)
1009 switch (rs->md.level) {
1013 if (rebuild_cnt >= rs->md.raid_disks)
1019 if (rebuild_cnt > rs->raid_type->parity_devs)
1023 copies = raid10_md_layout_to_copies(rs->md.new_layout);
1025 DMERR("Bogus raid10 data copies < 2!");
1029 if (rebuild_cnt < copies)
1033 * It is possible to have a higher rebuild count for RAID10,
1034 * as long as the failed devices occur in different mirror
1035 * groups (i.e. different stripes).
1037 * When checking "near" format, make sure no adjacent devices
1038 * have failed beyond what can be handled. In addition to the
1039 * simple case where the number of devices is a multiple of the
1040 * number of copies, we must also handle cases where the number
1041 * of devices is not a multiple of the number of copies.
1042 * E.g. dev1 dev2 dev3 dev4 dev5
1046 if (__is_raid10_near(rs->md.new_layout)) {
1047 for (i = 0; i < rs->md.raid_disks; i++) {
1049 rebuilds_per_group = 0;
1050 if ((!rs->dev[i].rdev.sb_page ||
1051 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1052 (++rebuilds_per_group >= copies))
1059 * When checking "far" and "offset" formats, we need to ensure
1060 * that the device that holds its copy is not also dead or
1061 * being rebuilt. (Note that "far" and "offset" formats only
1062 * support two copies right now. These formats also only ever
1063 * use the 'use_far_sets' variant.)
1065 * This check is somewhat complicated by the need to account
1066 * for arrays that are not a multiple of (far) copies. This
1067 * results in the need to treat the last (potentially larger)
1070 group_size = (rs->md.raid_disks / copies);
1071 last_group_start = (rs->md.raid_disks / group_size) - 1;
1072 last_group_start *= group_size;
1073 for (i = 0; i < rs->md.raid_disks; i++) {
1074 if (!(i % copies) && !(i > last_group_start))
1075 rebuilds_per_group = 0;
1076 if ((!rs->dev[i].rdev.sb_page ||
1077 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1078 (++rebuilds_per_group >= copies))
1094 * Possible arguments are...
1095 * <chunk_size> [optional_args]
1097 * Argument definitions
1098 * <chunk_size> The number of sectors per disk that
1099 * will form the "stripe"
1100 * [[no]sync] Force or prevent recovery of the
1102 * [rebuild <idx>] Rebuild the drive indicated by the index
1103 * [daemon_sleep <ms>] Time between bitmap daemon work to
1105 * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1106 * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1107 * [write_mostly <idx>] Indicate a write mostly drive via index
1108 * [max_write_behind <sectors>] See '-write-behind=' (man mdadm)
1109 * [stripe_cache <sectors>] Stripe cache size for higher RAIDs
1110 * [region_size <sectors>] Defines granularity of bitmap
1111 * [journal_dev <dev>] raid4/5/6 journaling deviice
1112 * (i.e. write hole closing log)
1114 * RAID10-only options:
1115 * [raid10_copies <# copies>] Number of copies. (Default: 2)
1116 * [raid10_format <near|far|offset>] Layout algorithm. (Default: near)
1118 static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
1119 unsigned int num_raid_params)
1121 int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
1122 unsigned int raid10_copies = 2;
1123 unsigned int i, write_mostly = 0;
1124 unsigned int region_size = 0;
1125 sector_t max_io_len;
1126 const char *arg, *key;
1127 struct raid_dev *rd;
1128 struct raid_type *rt = rs->raid_type;
1130 arg = dm_shift_arg(as);
1131 num_raid_params--; /* Account for chunk_size argument */
1133 if (kstrtoint(arg, 10, &value) < 0) {
1134 rs->ti->error = "Bad numerical argument given for chunk_size";
1139 * First, parse the in-order required arguments
1140 * "chunk_size" is the only argument of this type.
1142 if (rt_is_raid1(rt)) {
1144 DMERR("Ignoring chunk size parameter for RAID 1");
1146 } else if (!is_power_of_2(value)) {
1147 rs->ti->error = "Chunk size must be a power of 2";
1149 } else if (value < 8) {
1150 rs->ti->error = "Chunk size value is too small";
1154 rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
1157 * We set each individual device as In_sync with a completed
1158 * 'recovery_offset'. If there has been a device failure or
1159 * replacement then one of the following cases applies:
1161 * 1) User specifies 'rebuild'.
1162 * - Device is reset when param is read.
1163 * 2) A new device is supplied.
1164 * - No matching superblock found, resets device.
1165 * 3) Device failure was transient and returns on reload.
1166 * - Failure noticed, resets device for bitmap replay.
1167 * 4) Device hadn't completed recovery after previous failure.
1168 * - Superblock is read and overrides recovery_offset.
1170 * What is found in the superblocks of the devices is always
1171 * authoritative, unless 'rebuild' or '[no]sync' was specified.
1173 for (i = 0; i < rs->raid_disks; i++) {
1174 set_bit(In_sync, &rs->dev[i].rdev.flags);
1175 rs->dev[i].rdev.recovery_offset = MaxSector;
1179 * Second, parse the unordered optional arguments
1181 for (i = 0; i < num_raid_params; i++) {
1182 key = dm_shift_arg(as);
1184 rs->ti->error = "Not enough raid parameters given";
1188 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
1189 if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1190 rs->ti->error = "Only one 'nosync' argument allowed";
1195 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
1196 if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
1197 rs->ti->error = "Only one 'sync' argument allowed";
1202 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
1203 if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1204 rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1210 arg = dm_shift_arg(as);
1211 i++; /* Account for the argument pairs */
1213 rs->ti->error = "Wrong number of raid parameters given";
1218 * Parameters that take a string value are checked here.
1220 /* "raid10_format {near|offset|far} */
1221 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
1222 if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
1223 rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1226 if (!rt_is_raid10(rt)) {
1227 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1230 raid10_format = raid10_name_to_format(arg);
1231 if (raid10_format < 0) {
1232 rs->ti->error = "Invalid 'raid10_format' value given";
1233 return raid10_format;
1238 /* "journal_dev <dev>" */
1239 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV))) {
1241 struct md_rdev *jdev;
1243 if (test_and_set_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1244 rs->ti->error = "Only one raid4/5/6 set journaling device allowed";
1247 if (!rt_is_raid456(rt)) {
1248 rs->ti->error = "'journal_dev' is an invalid parameter for this RAID type";
1251 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
1252 &rs->journal_dev.dev);
1254 rs->ti->error = "raid4/5/6 journal device lookup failure";
1257 jdev = &rs->journal_dev.rdev;
1259 jdev->mddev = &rs->md;
1260 jdev->bdev = rs->journal_dev.dev->bdev;
1261 jdev->sectors = to_sector(i_size_read(jdev->bdev->bd_inode));
1262 if (jdev->sectors < MIN_RAID456_JOURNAL_SPACE) {
1263 rs->ti->error = "No space for raid4/5/6 journal";
1266 rs->journal_dev.mode = R5C_JOURNAL_MODE_WRITE_THROUGH;
1267 set_bit(Journal, &jdev->flags);
1271 /* "journal_mode <mode>" ("journal_dev" mandatory!) */
1272 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE))) {
1275 if (!test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1276 rs->ti->error = "raid4/5/6 'journal_mode' is invalid without 'journal_dev'";
1279 if (test_and_set_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
1280 rs->ti->error = "Only one raid4/5/6 'journal_mode' argument allowed";
1283 r = dm_raid_journal_mode_to_md(arg);
1285 rs->ti->error = "Invalid 'journal_mode' argument";
1288 rs->journal_dev.mode = r;
1293 * Parameters with number values from here on.
1295 if (kstrtoint(arg, 10, &value) < 0) {
1296 rs->ti->error = "Bad numerical argument given in raid params";
1300 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
1302 * "rebuild" is being passed in by userspace to provide
1303 * indexes of replaced devices and to set up additional
1304 * devices on raid level takeover.
1306 if (!__within_range(value, 0, rs->raid_disks - 1)) {
1307 rs->ti->error = "Invalid rebuild index given";
1311 if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1312 rs->ti->error = "rebuild for this index already given";
1316 rd = rs->dev + value;
1317 clear_bit(In_sync, &rd->rdev.flags);
1318 clear_bit(Faulty, &rd->rdev.flags);
1319 rd->rdev.recovery_offset = 0;
1320 set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
1321 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
1322 if (!rt_is_raid1(rt)) {
1323 rs->ti->error = "write_mostly option is only valid for RAID1";
1327 if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
1328 rs->ti->error = "Invalid write_mostly index given";
1333 set_bit(WriteMostly, &rs->dev[value].rdev.flags);
1334 set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
1335 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
1336 if (!rt_is_raid1(rt)) {
1337 rs->ti->error = "max_write_behind option is only valid for RAID1";
1341 if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
1342 rs->ti->error = "Only one max_write_behind argument pair allowed";
1347 * In device-mapper, we specify things in sectors, but
1348 * MD records this value in kB
1350 if (value < 0 || value / 2 > COUNTER_MAX) {
1351 rs->ti->error = "Max write-behind limit out of range";
1355 rs->md.bitmap_info.max_write_behind = value / 2;
1356 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
1357 if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
1358 rs->ti->error = "Only one daemon_sleep argument pair allowed";
1362 rs->ti->error = "daemon sleep period out of range";
1365 rs->md.bitmap_info.daemon_sleep = value;
1366 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
1367 /* Userspace passes new data_offset after having extended the the data image LV */
1368 if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
1369 rs->ti->error = "Only one data_offset argument pair allowed";
1372 /* Ensure sensible data offset */
1374 (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
1375 rs->ti->error = "Bogus data_offset value";
1378 rs->data_offset = value;
1379 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
1380 /* Define the +/-# of disks to add to/remove from the given raid set */
1381 if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1382 rs->ti->error = "Only one delta_disks argument pair allowed";
1385 /* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1386 if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
1387 rs->ti->error = "Too many delta_disk requested";
1391 rs->delta_disks = value;
1392 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
1393 if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
1394 rs->ti->error = "Only one stripe_cache argument pair allowed";
1398 if (!rt_is_raid456(rt)) {
1399 rs->ti->error = "Inappropriate argument: stripe_cache";
1404 rs->ti->error = "Bogus stripe cache entries value";
1407 rs->stripe_cache_entries = value;
1408 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
1409 if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1410 rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1415 rs->ti->error = "min_recovery_rate out of range";
1418 rs->md.sync_speed_min = value;
1419 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
1420 if (test_and_set_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags)) {
1421 rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1426 rs->ti->error = "max_recovery_rate out of range";
1429 rs->md.sync_speed_max = value;
1430 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
1431 if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
1432 rs->ti->error = "Only one region_size argument pair allowed";
1436 region_size = value;
1437 rs->requested_bitmap_chunk_sectors = value;
1438 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
1439 if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
1440 rs->ti->error = "Only one raid10_copies argument pair allowed";
1444 if (!__within_range(value, 2, rs->md.raid_disks)) {
1445 rs->ti->error = "Bad value for 'raid10_copies'";
1449 raid10_copies = value;
1451 DMERR("Unable to parse RAID parameter: %s", key);
1452 rs->ti->error = "Unable to parse RAID parameter";
1457 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1458 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1459 rs->ti->error = "sync and nosync are mutually exclusive";
1463 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) &&
1464 (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ||
1465 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))) {
1466 rs->ti->error = "sync/nosync and rebuild are mutually exclusive";
1470 if (write_mostly >= rs->md.raid_disks) {
1471 rs->ti->error = "Can't set all raid1 devices to write_mostly";
1475 if (rs->md.sync_speed_max &&
1476 rs->md.sync_speed_min > rs->md.sync_speed_max) {
1477 rs->ti->error = "Bogus recovery rates";
1481 if (validate_region_size(rs, region_size))
1484 if (rs->md.chunk_sectors)
1485 max_io_len = rs->md.chunk_sectors;
1487 max_io_len = region_size;
1489 if (dm_set_target_max_io_len(rs->ti, max_io_len))
1492 if (rt_is_raid10(rt)) {
1493 if (raid10_copies > rs->md.raid_disks) {
1494 rs->ti->error = "Not enough devices to satisfy specification";
1498 rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
1499 if (rs->md.new_layout < 0) {
1500 rs->ti->error = "Error getting raid10 format";
1501 return rs->md.new_layout;
1504 rt = get_raid_type_by_ll(10, rs->md.new_layout);
1506 rs->ti->error = "Failed to recognize new raid10 layout";
1510 if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1511 rt->algorithm == ALGORITHM_RAID10_NEAR) &&
1512 test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1513 rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1518 rs->raid10_copies = raid10_copies;
1520 /* Assume there are no metadata devices until the drives are parsed */
1521 rs->md.persistent = 0;
1522 rs->md.external = 1;
1524 /* Check, if any invalid ctr arguments have been passed in for the raid level */
1525 return rs_check_for_valid_flags(rs);
1528 /* Set raid4/5/6 cache size */
1529 static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1532 struct r5conf *conf;
1533 struct mddev *mddev = &rs->md;
1534 uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1535 uint32_t nr_stripes = rs->stripe_cache_entries;
1537 if (!rt_is_raid456(rs->raid_type)) {
1538 rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1542 if (nr_stripes < min_stripes) {
1543 DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1544 nr_stripes, min_stripes);
1545 nr_stripes = min_stripes;
1548 conf = mddev->private;
1550 rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1554 /* Try setting number of stripes in raid456 stripe cache */
1555 if (conf->min_nr_stripes != nr_stripes) {
1556 r = raid5_set_cache_size(mddev, nr_stripes);
1558 rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1562 DMINFO("%u stripe cache entries", nr_stripes);
1568 /* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1569 static unsigned int mddev_data_stripes(struct raid_set *rs)
1571 return rs->md.raid_disks - rs->raid_type->parity_devs;
1574 /* Return # of data stripes of @rs (i.e. as of ctr) */
1575 static unsigned int rs_data_stripes(struct raid_set *rs)
1577 return rs->raid_disks - rs->raid_type->parity_devs;
1581 * Retrieve rdev->sectors from any valid raid device of @rs
1582 * to allow userpace to pass in arbitray "- -" device tupples.
1584 static sector_t __rdev_sectors(struct raid_set *rs)
1588 for (i = 0; i < rs->md.raid_disks; i++) {
1589 struct md_rdev *rdev = &rs->dev[i].rdev;
1591 if (!test_bit(Journal, &rdev->flags) &&
1592 rdev->bdev && rdev->sectors)
1593 return rdev->sectors;
1599 /* Check that calculated dev_sectors fits all component devices. */
1600 static int _check_data_dev_sectors(struct raid_set *rs)
1603 struct md_rdev *rdev;
1605 rdev_for_each(rdev, &rs->md)
1606 if (!test_bit(Journal, &rdev->flags) && rdev->bdev) {
1607 ds = min(ds, to_sector(i_size_read(rdev->bdev->bd_inode)));
1608 if (ds < rs->md.dev_sectors) {
1609 rs->ti->error = "Component device(s) too small";
1617 /* Calculate the sectors per device and per array used for @rs */
1618 static int rs_set_dev_and_array_sectors(struct raid_set *rs, bool use_mddev)
1621 unsigned int data_stripes;
1622 struct mddev *mddev = &rs->md;
1623 struct md_rdev *rdev;
1624 sector_t array_sectors = rs->ti->len, dev_sectors = rs->ti->len;
1627 delta_disks = mddev->delta_disks;
1628 data_stripes = mddev_data_stripes(rs);
1630 delta_disks = rs->delta_disks;
1631 data_stripes = rs_data_stripes(rs);
1634 /* Special raid1 case w/o delta_disks support (yet) */
1635 if (rt_is_raid1(rs->raid_type))
1637 else if (rt_is_raid10(rs->raid_type)) {
1638 if (rs->raid10_copies < 2 ||
1640 rs->ti->error = "Bogus raid10 data copies or delta disks";
1644 dev_sectors *= rs->raid10_copies;
1645 if (sector_div(dev_sectors, data_stripes))
1648 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1649 if (sector_div(array_sectors, rs->raid10_copies))
1652 } else if (sector_div(dev_sectors, data_stripes))
1656 /* Striped layouts */
1657 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1659 rdev_for_each(rdev, mddev)
1660 if (!test_bit(Journal, &rdev->flags))
1661 rdev->sectors = dev_sectors;
1663 mddev->array_sectors = array_sectors;
1664 mddev->dev_sectors = dev_sectors;
1666 return _check_data_dev_sectors(rs);
1668 rs->ti->error = "Target length not divisible by number of data devices";
1672 /* Setup recovery on @rs */
1673 static void __rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1675 /* raid0 does not recover */
1676 if (rs_is_raid0(rs))
1677 rs->md.recovery_cp = MaxSector;
1679 * A raid6 set has to be recovered either
1680 * completely or for the grown part to
1681 * ensure proper parity and Q-Syndrome
1683 else if (rs_is_raid6(rs))
1684 rs->md.recovery_cp = dev_sectors;
1686 * Other raid set types may skip recovery
1687 * depending on the 'nosync' flag.
1690 rs->md.recovery_cp = test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)
1691 ? MaxSector : dev_sectors;
1694 /* Setup recovery on @rs based on raid type, device size and 'nosync' flag */
1695 static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1698 /* New raid set or 'sync' flag provided */
1699 __rs_setup_recovery(rs, 0);
1700 else if (dev_sectors == MaxSector)
1701 /* Prevent recovery */
1702 __rs_setup_recovery(rs, MaxSector);
1703 else if (__rdev_sectors(rs) < dev_sectors)
1704 /* Grown raid set */
1705 __rs_setup_recovery(rs, __rdev_sectors(rs));
1707 __rs_setup_recovery(rs, MaxSector);
1710 static void do_table_event(struct work_struct *ws)
1712 struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1714 smp_rmb(); /* Make sure we access most actual mddev properties */
1715 if (!rs_is_reshaping(rs)) {
1716 if (rs_is_raid10(rs))
1717 rs_set_rdev_sectors(rs);
1718 rs_set_capacity(rs);
1720 dm_table_event(rs->ti->table);
1723 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
1725 struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
1727 return mddev_congested(&rs->md, bits);
1731 * Make sure a valid takover (level switch) is being requested on @rs
1733 * Conversions of raid sets from one MD personality to another
1734 * have to conform to restrictions which are enforced here.
1736 static int rs_check_takeover(struct raid_set *rs)
1738 struct mddev *mddev = &rs->md;
1739 unsigned int near_copies;
1741 if (rs->md.degraded) {
1742 rs->ti->error = "Can't takeover degraded raid set";
1746 if (rs_is_reshaping(rs)) {
1747 rs->ti->error = "Can't takeover reshaping raid set";
1751 switch (mddev->level) {
1753 /* raid0 -> raid1/5 with one disk */
1754 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1755 mddev->raid_disks == 1)
1758 /* raid0 -> raid10 */
1759 if (mddev->new_level == 10 &&
1760 !(rs->raid_disks % mddev->raid_disks))
1763 /* raid0 with multiple disks -> raid4/5/6 */
1764 if (__within_range(mddev->new_level, 4, 6) &&
1765 mddev->new_layout == ALGORITHM_PARITY_N &&
1766 mddev->raid_disks > 1)
1772 /* Can't takeover raid10_offset! */
1773 if (__is_raid10_offset(mddev->layout))
1776 near_copies = __raid10_near_copies(mddev->layout);
1778 /* raid10* -> raid0 */
1779 if (mddev->new_level == 0) {
1780 /* Can takeover raid10_near with raid disks divisable by data copies! */
1781 if (near_copies > 1 &&
1782 !(mddev->raid_disks % near_copies)) {
1783 mddev->raid_disks /= near_copies;
1784 mddev->delta_disks = mddev->raid_disks;
1788 /* Can takeover raid10_far */
1789 if (near_copies == 1 &&
1790 __raid10_far_copies(mddev->layout) > 1)
1796 /* raid10_{near,far} -> raid1 */
1797 if (mddev->new_level == 1 &&
1798 max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
1801 /* raid10_{near,far} with 2 disks -> raid4/5 */
1802 if (__within_range(mddev->new_level, 4, 5) &&
1803 mddev->raid_disks == 2)
1808 /* raid1 with 2 disks -> raid4/5 */
1809 if (__within_range(mddev->new_level, 4, 5) &&
1810 mddev->raid_disks == 2) {
1811 mddev->degraded = 1;
1815 /* raid1 -> raid0 */
1816 if (mddev->new_level == 0 &&
1817 mddev->raid_disks == 1)
1820 /* raid1 -> raid10 */
1821 if (mddev->new_level == 10)
1826 /* raid4 -> raid0 */
1827 if (mddev->new_level == 0)
1830 /* raid4 -> raid1/5 with 2 disks */
1831 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1832 mddev->raid_disks == 2)
1835 /* raid4 -> raid5/6 with parity N */
1836 if (__within_range(mddev->new_level, 5, 6) &&
1837 mddev->layout == ALGORITHM_PARITY_N)
1842 /* raid5 with parity N -> raid0 */
1843 if (mddev->new_level == 0 &&
1844 mddev->layout == ALGORITHM_PARITY_N)
1847 /* raid5 with parity N -> raid4 */
1848 if (mddev->new_level == 4 &&
1849 mddev->layout == ALGORITHM_PARITY_N)
1852 /* raid5 with 2 disks -> raid1/4/10 */
1853 if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1854 mddev->raid_disks == 2)
1857 /* raid5_* -> raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
1858 if (mddev->new_level == 6 &&
1859 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1860 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
1865 /* raid6 with parity N -> raid0 */
1866 if (mddev->new_level == 0 &&
1867 mddev->layout == ALGORITHM_PARITY_N)
1870 /* raid6 with parity N -> raid4 */
1871 if (mddev->new_level == 4 &&
1872 mddev->layout == ALGORITHM_PARITY_N)
1875 /* raid6_*_n with Q-Syndrome N -> raid5_* */
1876 if (mddev->new_level == 5 &&
1877 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1878 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
1885 rs->ti->error = "takeover not possible";
1889 /* True if @rs requested to be taken over */
1890 static bool rs_takeover_requested(struct raid_set *rs)
1892 return rs->md.new_level != rs->md.level;
1895 /* True if layout is set to reshape. */
1896 static bool rs_is_layout_change(struct raid_set *rs, bool use_mddev)
1898 return (use_mddev ? rs->md.delta_disks : rs->delta_disks) ||
1899 rs->md.new_layout != rs->md.layout ||
1900 rs->md.new_chunk_sectors != rs->md.chunk_sectors;
1903 /* True if @rs is requested to reshape by ctr */
1904 static bool rs_reshape_requested(struct raid_set *rs)
1907 struct mddev *mddev = &rs->md;
1909 if (rs_takeover_requested(rs))
1912 if (rs_is_raid0(rs))
1915 change = rs_is_layout_change(rs, false);
1917 /* Historical case to support raid1 reshape without delta disks */
1918 if (rs_is_raid1(rs)) {
1919 if (rs->delta_disks)
1920 return !!rs->delta_disks;
1923 mddev->raid_disks != rs->raid_disks;
1926 if (rs_is_raid10(rs))
1928 !__is_raid10_far(mddev->new_layout) &&
1929 rs->delta_disks >= 0;
1935 #define FEATURE_FLAG_SUPPORTS_V190 0x1 /* Supports extended superblock */
1937 /* State flags for sb->flags */
1938 #define SB_FLAG_RESHAPE_ACTIVE 0x1
1939 #define SB_FLAG_RESHAPE_BACKWARDS 0x2
1942 * This structure is never routinely used by userspace, unlike md superblocks.
1943 * Devices with this superblock should only ever be accessed via device-mapper.
1945 #define DM_RAID_MAGIC 0x64526D44
1946 struct dm_raid_superblock {
1947 __le32 magic; /* "DmRd" */
1948 __le32 compat_features; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1950 __le32 num_devices; /* Number of devices in this raid set. (Max 64) */
1951 __le32 array_position; /* The position of this drive in the raid set */
1953 __le64 events; /* Incremented by md when superblock updated */
1954 __le64 failed_devices; /* Pre 1.9.0 part of bit field of devices to */
1955 /* indicate failures (see extension below) */
1958 * This offset tracks the progress of the repair or replacement of
1959 * an individual drive.
1961 __le64 disk_recovery_offset;
1964 * This offset tracks the progress of the initial raid set
1965 * synchronisation/parity calculation.
1967 __le64 array_resync_offset;
1970 * raid characteristics
1974 __le32 stripe_sectors;
1976 /********************************************************************
1977 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1979 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
1982 __le32 flags; /* Flags defining array states for reshaping */
1985 * This offset tracks the progress of a raid
1986 * set reshape in order to be able to restart it
1988 __le64 reshape_position;
1991 * These define the properties of the array in case of an interrupted reshape
1995 __le32 new_stripe_sectors;
1998 __le64 array_sectors; /* Array size in sectors */
2001 * Sector offsets to data on devices (reshaping).
2002 * Needed to support out of place reshaping, thus
2003 * not writing over any stripes whilst converting
2004 * them from old to new layout
2007 __le64 new_data_offset;
2009 __le64 sectors; /* Used device size in sectors */
2012 * Additonal Bit field of devices indicating failures to support
2013 * up to 256 devices with the 1.9.0 on-disk metadata format
2015 __le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
2017 __le32 incompat_features; /* Used to indicate any incompatible features */
2019 /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
2023 * Check for reshape constraints on raid set @rs:
2025 * - reshape function non-existent
2027 * - ongoing recovery
2030 * Returns 0 if none or -EPERM if given constraint
2031 * and error message reference in @errmsg
2033 static int rs_check_reshape(struct raid_set *rs)
2035 struct mddev *mddev = &rs->md;
2037 if (!mddev->pers || !mddev->pers->check_reshape)
2038 rs->ti->error = "Reshape not supported";
2039 else if (mddev->degraded)
2040 rs->ti->error = "Can't reshape degraded raid set";
2041 else if (rs_is_recovering(rs))
2042 rs->ti->error = "Convert request on recovering raid set prohibited";
2043 else if (rs_is_reshaping(rs))
2044 rs->ti->error = "raid set already reshaping!";
2045 else if (!(rs_is_raid1(rs) || rs_is_raid10(rs) || rs_is_raid456(rs)))
2046 rs->ti->error = "Reshaping only supported for raid1/4/5/6/10";
2053 static int read_disk_sb(struct md_rdev *rdev, int size, bool force_reload)
2055 BUG_ON(!rdev->sb_page);
2057 if (rdev->sb_loaded && !force_reload)
2060 rdev->sb_loaded = 0;
2062 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) {
2063 DMERR("Failed to read superblock of device at position %d",
2065 md_error(rdev->mddev, rdev);
2066 set_bit(Faulty, &rdev->flags);
2070 rdev->sb_loaded = 1;
2075 static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2077 failed_devices[0] = le64_to_cpu(sb->failed_devices);
2078 memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
2080 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2081 int i = ARRAY_SIZE(sb->extended_failed_devices);
2084 failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
2088 static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2090 int i = ARRAY_SIZE(sb->extended_failed_devices);
2092 sb->failed_devices = cpu_to_le64(failed_devices[0]);
2094 sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
2098 * Synchronize the superblock members with the raid set properties
2100 * All superblock data is little endian.
2102 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
2104 bool update_failed_devices = false;
2106 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2107 struct dm_raid_superblock *sb;
2108 struct raid_set *rs = container_of(mddev, struct raid_set, md);
2110 /* No metadata device, no superblock */
2111 if (!rdev->meta_bdev)
2114 BUG_ON(!rdev->sb_page);
2116 sb = page_address(rdev->sb_page);
2118 sb_retrieve_failed_devices(sb, failed_devices);
2120 for (i = 0; i < rs->raid_disks; i++)
2121 if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
2122 update_failed_devices = true;
2123 set_bit(i, (void *) failed_devices);
2126 if (update_failed_devices)
2127 sb_update_failed_devices(sb, failed_devices);
2129 sb->magic = cpu_to_le32(DM_RAID_MAGIC);
2130 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2132 sb->num_devices = cpu_to_le32(mddev->raid_disks);
2133 sb->array_position = cpu_to_le32(rdev->raid_disk);
2135 sb->events = cpu_to_le64(mddev->events);
2137 sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
2138 sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
2140 sb->level = cpu_to_le32(mddev->level);
2141 sb->layout = cpu_to_le32(mddev->layout);
2142 sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
2144 /********************************************************************
2145 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
2147 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
2149 sb->new_level = cpu_to_le32(mddev->new_level);
2150 sb->new_layout = cpu_to_le32(mddev->new_layout);
2151 sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
2153 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2155 smp_rmb(); /* Make sure we access most recent reshape position */
2156 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2157 if (le64_to_cpu(sb->reshape_position) != MaxSector) {
2158 /* Flag ongoing reshape */
2159 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
2161 if (mddev->delta_disks < 0 || mddev->reshape_backwards)
2162 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
2164 /* Clear reshape flags */
2165 sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
2168 sb->array_sectors = cpu_to_le64(mddev->array_sectors);
2169 sb->data_offset = cpu_to_le64(rdev->data_offset);
2170 sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
2171 sb->sectors = cpu_to_le64(rdev->sectors);
2172 sb->incompat_features = cpu_to_le32(0);
2174 /* Zero out the rest of the payload after the size of the superblock */
2175 memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
2181 * This function creates a superblock if one is not found on the device
2182 * and will decide which superblock to use if there's a choice.
2184 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
2186 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
2189 struct dm_raid_superblock *sb;
2190 struct dm_raid_superblock *refsb;
2191 uint64_t events_sb, events_refsb;
2193 r = read_disk_sb(rdev, rdev->sb_size, false);
2197 sb = page_address(rdev->sb_page);
2200 * Two cases that we want to write new superblocks and rebuild:
2201 * 1) New device (no matching magic number)
2202 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2204 if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
2205 (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
2206 super_sync(rdev->mddev, rdev);
2208 set_bit(FirstUse, &rdev->flags);
2209 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2211 /* Force writing of superblocks to disk */
2212 set_bit(MD_SB_CHANGE_DEVS, &rdev->mddev->sb_flags);
2214 /* Any superblock is better than none, choose that if given */
2215 return refdev ? 0 : 1;
2221 events_sb = le64_to_cpu(sb->events);
2223 refsb = page_address(refdev->sb_page);
2224 events_refsb = le64_to_cpu(refsb->events);
2226 return (events_sb > events_refsb) ? 1 : 0;
2229 static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
2233 struct mddev *mddev = &rs->md;
2235 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2236 struct dm_raid_superblock *sb;
2237 uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
2239 struct dm_raid_superblock *sb2;
2241 sb = page_address(rdev->sb_page);
2242 events_sb = le64_to_cpu(sb->events);
2245 * Initialise to 1 if this is a new superblock.
2247 mddev->events = events_sb ? : 1;
2249 mddev->reshape_position = MaxSector;
2251 mddev->raid_disks = le32_to_cpu(sb->num_devices);
2252 mddev->level = le32_to_cpu(sb->level);
2253 mddev->layout = le32_to_cpu(sb->layout);
2254 mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
2257 * Reshaping is supported, e.g. reshape_position is valid
2258 * in superblock and superblock content is authoritative.
2260 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2261 /* Superblock is authoritative wrt given raid set layout! */
2262 mddev->new_level = le32_to_cpu(sb->new_level);
2263 mddev->new_layout = le32_to_cpu(sb->new_layout);
2264 mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
2265 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
2266 mddev->array_sectors = le64_to_cpu(sb->array_sectors);
2268 /* raid was reshaping and got interrupted */
2269 if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
2270 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
2271 DMERR("Reshape requested but raid set is still reshaping");
2275 if (mddev->delta_disks < 0 ||
2276 (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
2277 mddev->reshape_backwards = 1;
2279 mddev->reshape_backwards = 0;
2281 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2282 rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2287 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2289 struct raid_type *rt_cur = get_raid_type_by_ll(mddev->level, mddev->layout);
2290 struct raid_type *rt_new = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
2292 if (rs_takeover_requested(rs)) {
2293 if (rt_cur && rt_new)
2294 DMERR("Takeover raid sets from %s to %s not yet supported by metadata. (raid level change)",
2295 rt_cur->name, rt_new->name);
2297 DMERR("Takeover raid sets not yet supported by metadata. (raid level change)");
2299 } else if (rs_reshape_requested(rs)) {
2300 DMERR("Reshaping raid sets not yet supported by metadata. (raid layout change keeping level)");
2301 if (mddev->layout != mddev->new_layout) {
2302 if (rt_cur && rt_new)
2303 DMERR(" current layout %s vs new layout %s",
2304 rt_cur->name, rt_new->name);
2306 DMERR(" current layout 0x%X vs new layout 0x%X",
2307 le32_to_cpu(sb->layout), mddev->new_layout);
2309 if (mddev->chunk_sectors != mddev->new_chunk_sectors)
2310 DMERR(" current stripe sectors %u vs new stripe sectors %u",
2311 mddev->chunk_sectors, mddev->new_chunk_sectors);
2312 if (rs->delta_disks)
2313 DMERR(" current %u disks vs new %u disks",
2314 mddev->raid_disks, mddev->raid_disks + rs->delta_disks);
2315 if (rs_is_raid10(rs)) {
2316 DMERR(" Old layout: %s w/ %u copies",
2317 raid10_md_layout_to_format(mddev->layout),
2318 raid10_md_layout_to_copies(mddev->layout));
2319 DMERR(" New layout: %s w/ %u copies",
2320 raid10_md_layout_to_format(mddev->new_layout),
2321 raid10_md_layout_to_copies(mddev->new_layout));
2326 DMINFO("Discovered old metadata format; upgrading to extended metadata format");
2329 if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
2330 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2333 * During load, we set FirstUse if a new superblock was written.
2334 * There are two reasons we might not have a superblock:
2335 * 1) The raid set is brand new - in which case, all of the
2336 * devices must have their In_sync bit set. Also,
2337 * recovery_cp must be 0, unless forced.
2338 * 2) This is a new device being added to an old raid set
2339 * and the new device needs to be rebuilt - in which
2340 * case the In_sync bit will /not/ be set and
2341 * recovery_cp must be MaxSector.
2342 * 3) This is/are a new device(s) being added to an old
2343 * raid set during takeover to a higher raid level
2344 * to provide capacity for redundancy or during reshape
2345 * to add capacity to grow the raid set.
2348 rdev_for_each(r, mddev) {
2349 if (test_bit(Journal, &rdev->flags))
2352 if (test_bit(FirstUse, &r->flags))
2355 if (!test_bit(In_sync, &r->flags)) {
2356 DMINFO("Device %d specified for rebuild; clearing superblock",
2360 if (test_bit(FirstUse, &r->flags))
2367 if (new_devs == rs->raid_disks || !rebuilds) {
2368 /* Replace a broken device */
2369 if (new_devs == 1 && !rs->delta_disks)
2371 if (new_devs == rs->raid_disks) {
2372 DMINFO("Superblocks created for new raid set");
2373 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2374 } else if (new_devs != rebuilds &&
2375 new_devs != rs->delta_disks) {
2376 DMERR("New device injected into existing raid set without "
2377 "'delta_disks' or 'rebuild' parameter specified");
2380 } else if (new_devs && new_devs != rebuilds) {
2381 DMERR("%u 'rebuild' devices cannot be injected into"
2382 " a raid set with %u other first-time devices",
2383 rebuilds, new_devs);
2385 } else if (rebuilds) {
2386 if (rebuild_and_new && rebuilds != rebuild_and_new) {
2387 DMERR("new device%s provided without 'rebuild'",
2388 new_devs > 1 ? "s" : "");
2390 } else if (!test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) && rs_is_recovering(rs)) {
2391 DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2392 (unsigned long long) mddev->recovery_cp);
2394 } else if (rs_is_reshaping(rs)) {
2395 DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2396 (unsigned long long) mddev->reshape_position);
2402 * Now we set the Faulty bit for those devices that are
2403 * recorded in the superblock as failed.
2405 sb_retrieve_failed_devices(sb, failed_devices);
2406 rdev_for_each(r, mddev) {
2407 if (test_bit(Journal, &rdev->flags) ||
2410 sb2 = page_address(r->sb_page);
2411 sb2->failed_devices = 0;
2412 memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
2415 * Check for any device re-ordering.
2417 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2418 role = le32_to_cpu(sb2->array_position);
2422 if (role != r->raid_disk) {
2423 if (rs_is_raid10(rs) && __is_raid10_near(mddev->layout)) {
2424 if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
2425 rs->raid_disks % rs->raid10_copies) {
2427 "Cannot change raid10 near set to odd # of devices!";
2431 sb2->array_position = cpu_to_le32(r->raid_disk);
2433 } else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
2434 !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2435 !rt_is_raid1(rs->raid_type)) {
2436 rs->ti->error = "Cannot change device positions in raid set";
2440 DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
2444 * Partial recovery is performed on
2445 * returning failed devices.
2447 if (test_bit(role, (void *) failed_devices))
2448 set_bit(Faulty, &r->flags);
2455 static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
2457 struct mddev *mddev = &rs->md;
2458 struct dm_raid_superblock *sb;
2460 if (rs_is_raid0(rs) || !rdev->sb_page || rdev->raid_disk < 0)
2463 sb = page_address(rdev->sb_page);
2466 * If mddev->events is not set, we know we have not yet initialized
2469 if (!mddev->events && super_init_validation(rs, rdev))
2472 if (le32_to_cpu(sb->compat_features) &&
2473 le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
2474 rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2478 if (sb->incompat_features) {
2479 rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
2483 /* Enable bitmap creation for RAID levels != 0 */
2484 mddev->bitmap_info.offset = (rt_is_raid0(rs->raid_type) || rs->journal_dev.dev) ? 0 : to_sector(4096);
2485 mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2487 if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2489 * Retrieve rdev size stored in superblock to be prepared for shrink.
2490 * Check extended superblock members are present otherwise the size
2493 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190)
2494 rdev->sectors = le64_to_cpu(sb->sectors);
2496 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
2497 if (rdev->recovery_offset == MaxSector)
2498 set_bit(In_sync, &rdev->flags);
2500 * If no reshape in progress -> we're recovering single
2501 * disk(s) and have to set the device(s) to out-of-sync
2503 else if (!rs_is_reshaping(rs))
2504 clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
2508 * If a device comes back, set it as not In_sync and no longer faulty.
2510 if (test_and_clear_bit(Faulty, &rdev->flags)) {
2511 rdev->recovery_offset = 0;
2512 clear_bit(In_sync, &rdev->flags);
2513 rdev->saved_raid_disk = rdev->raid_disk;
2516 /* Reshape support -> restore repective data offsets */
2517 rdev->data_offset = le64_to_cpu(sb->data_offset);
2518 rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
2524 * Analyse superblocks and select the freshest.
2526 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2529 struct md_rdev *rdev, *freshest;
2530 struct mddev *mddev = &rs->md;
2533 rdev_for_each(rdev, mddev) {
2534 if (test_bit(Journal, &rdev->flags))
2537 if (!rdev->meta_bdev)
2540 /* Set superblock offset/size for metadata device. */
2542 rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
2543 if (rdev->sb_size < sizeof(struct dm_raid_superblock) || rdev->sb_size > PAGE_SIZE) {
2544 DMERR("superblock size of a logical block is no longer valid");
2549 * Skipping super_load due to CTR_FLAG_SYNC will cause
2550 * the array to undergo initialization again as
2551 * though it were new. This is the intended effect
2552 * of the "sync" directive.
2554 * With reshaping capability added, we must ensure that
2555 * that the "sync" directive is disallowed during the reshape.
2557 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2560 r = super_load(rdev, freshest);
2569 /* This is a failure to read the superblock from the metadata device. */
2571 * We have to keep any raid0 data/metadata device pairs or
2572 * the MD raid0 personality will fail to start the array.
2574 if (rs_is_raid0(rs))
2578 * We keep the dm_devs to be able to emit the device tuple
2579 * properly on the table line in raid_status() (rather than
2580 * mistakenly acting as if '- -' got passed into the constructor).
2582 * The rdev has to stay on the same_set list to allow for
2583 * the attempt to restore faulty devices on second resume.
2585 rdev->raid_disk = rdev->saved_raid_disk = -1;
2594 * Validation of the freshest device provides the source of
2595 * validation for the remaining devices.
2597 rs->ti->error = "Unable to assemble array: Invalid superblocks";
2598 if (super_validate(rs, freshest))
2601 if (validate_raid_redundancy(rs)) {
2602 rs->ti->error = "Insufficient redundancy to activate array";
2606 rdev_for_each(rdev, mddev)
2607 if (!test_bit(Journal, &rdev->flags) &&
2609 super_validate(rs, rdev))
2615 * Adjust data_offset and new_data_offset on all disk members of @rs
2616 * for out of place reshaping if requested by contructor
2618 * We need free space at the beginning of each raid disk for forward
2619 * and at the end for backward reshapes which userspace has to provide
2620 * via remapping/reordering of space.
2622 static int rs_adjust_data_offsets(struct raid_set *rs)
2624 sector_t data_offset = 0, new_data_offset = 0;
2625 struct md_rdev *rdev;
2627 /* Constructor did not request data offset change */
2628 if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2629 if (!rs_is_reshapable(rs))
2635 /* HM FIXME: get In_Sync raid_dev? */
2636 rdev = &rs->dev[0].rdev;
2638 if (rs->delta_disks < 0) {
2640 * Removing disks (reshaping backwards):
2642 * - before reshape: data is at offset 0 and free space
2643 * is at end of each component LV
2645 * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2648 new_data_offset = rs->data_offset;
2650 } else if (rs->delta_disks > 0) {
2652 * Adding disks (reshaping forwards):
2654 * - before reshape: data is at offset rs->data_offset != 0 and
2655 * free space is at begin of each component LV
2657 * - after reshape: data is at offset 0 on each component LV
2659 data_offset = rs->data_offset;
2660 new_data_offset = 0;
2664 * User space passes in 0 for data offset after having removed reshape space
2666 * - or - (data offset != 0)
2668 * Changing RAID layout or chunk size -> toggle offsets
2670 * - before reshape: data is at offset rs->data_offset 0 and
2671 * free space is at end of each component LV
2673 * data is at offset rs->data_offset != 0 and
2674 * free space is at begin of each component LV
2676 * - after reshape: data is at offset 0 if it was at offset != 0
2677 * or at offset != 0 if it was at offset 0
2678 * on each component LV
2681 data_offset = rs->data_offset ? rdev->data_offset : 0;
2682 new_data_offset = data_offset ? 0 : rs->data_offset;
2683 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2687 * Make sure we got a minimum amount of free sectors per device
2689 if (rs->data_offset &&
2690 to_sector(i_size_read(rdev->bdev->bd_inode)) - rs->md.dev_sectors < MIN_FREE_RESHAPE_SPACE) {
2691 rs->ti->error = data_offset ? "No space for forward reshape" :
2692 "No space for backward reshape";
2697 * Raise recovery_cp in case data_offset != 0 to
2698 * avoid false recovery positives in the constructor.
2700 if (rs->md.recovery_cp < rs->md.dev_sectors)
2701 rs->md.recovery_cp += rs->dev[0].rdev.data_offset;
2703 /* Adjust data offsets on all rdevs but on any raid4/5/6 journal device */
2704 rdev_for_each(rdev, &rs->md) {
2705 if (!test_bit(Journal, &rdev->flags)) {
2706 rdev->data_offset = data_offset;
2707 rdev->new_data_offset = new_data_offset;
2714 /* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2715 static void __reorder_raid_disk_indexes(struct raid_set *rs)
2718 struct md_rdev *rdev;
2720 rdev_for_each(rdev, &rs->md) {
2721 if (!test_bit(Journal, &rdev->flags)) {
2722 rdev->raid_disk = i++;
2723 rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2729 * Setup @rs for takeover by a different raid level
2731 static int rs_setup_takeover(struct raid_set *rs)
2733 struct mddev *mddev = &rs->md;
2734 struct md_rdev *rdev;
2735 unsigned int d = mddev->raid_disks = rs->raid_disks;
2736 sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2738 if (rt_is_raid10(rs->raid_type)) {
2739 if (rs_is_raid0(rs)) {
2740 /* Userpace reordered disks -> adjust raid_disk indexes */
2741 __reorder_raid_disk_indexes(rs);
2743 /* raid0 -> raid10_far layout */
2744 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2746 } else if (rs_is_raid1(rs))
2747 /* raid1 -> raid10_near layout */
2748 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2755 clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2756 mddev->recovery_cp = MaxSector;
2759 rdev = &rs->dev[d].rdev;
2761 if (test_bit(d, (void *) rs->rebuild_disks)) {
2762 clear_bit(In_sync, &rdev->flags);
2763 clear_bit(Faulty, &rdev->flags);
2764 mddev->recovery_cp = rdev->recovery_offset = 0;
2765 /* Bitmap has to be created when we do an "up" takeover */
2766 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2769 rdev->new_data_offset = new_data_offset;
2775 /* Prepare @rs for reshape */
2776 static int rs_prepare_reshape(struct raid_set *rs)
2779 struct mddev *mddev = &rs->md;
2781 if (rs_is_raid10(rs)) {
2782 if (rs->raid_disks != mddev->raid_disks &&
2783 __is_raid10_near(mddev->layout) &&
2784 rs->raid10_copies &&
2785 rs->raid10_copies != __raid10_near_copies(mddev->layout)) {
2787 * raid disk have to be multiple of data copies to allow this conversion,
2789 * This is actually not a reshape it is a
2790 * rebuild of any additional mirrors per group
2792 if (rs->raid_disks % rs->raid10_copies) {
2793 rs->ti->error = "Can't reshape raid10 mirror groups";
2797 /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2798 __reorder_raid_disk_indexes(rs);
2799 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2801 mddev->new_layout = mddev->layout;
2806 } else if (rs_is_raid456(rs))
2809 else if (rs_is_raid1(rs)) {
2810 if (rs->delta_disks) {
2811 /* Process raid1 via delta_disks */
2812 mddev->degraded = rs->delta_disks < 0 ? -rs->delta_disks : rs->delta_disks;
2815 /* Process raid1 without delta_disks */
2816 mddev->raid_disks = rs->raid_disks;
2820 rs->ti->error = "Called with bogus raid type";
2825 set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2826 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2827 } else if (mddev->raid_disks < rs->raid_disks)
2828 /* Create new superblocks and bitmaps, if any new disks */
2829 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2834 /* Get reshape sectors from data_offsets or raid set */
2835 static sector_t _get_reshape_sectors(struct raid_set *rs)
2837 struct md_rdev *rdev;
2838 sector_t reshape_sectors = 0;
2840 rdev_for_each(rdev, &rs->md)
2841 if (!test_bit(Journal, &rdev->flags)) {
2842 reshape_sectors = (rdev->data_offset > rdev->new_data_offset) ?
2843 rdev->data_offset - rdev->new_data_offset :
2844 rdev->new_data_offset - rdev->data_offset;
2848 return max(reshape_sectors, (sector_t) rs->data_offset);
2853 * - change raid layout
2854 * - change chunk size
2858 static int rs_setup_reshape(struct raid_set *rs)
2861 unsigned int cur_raid_devs, d;
2862 sector_t reshape_sectors = _get_reshape_sectors(rs);
2863 struct mddev *mddev = &rs->md;
2864 struct md_rdev *rdev;
2866 mddev->delta_disks = rs->delta_disks;
2867 cur_raid_devs = mddev->raid_disks;
2869 /* Ignore impossible layout change whilst adding/removing disks */
2870 if (mddev->delta_disks &&
2871 mddev->layout != mddev->new_layout) {
2872 DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2873 mddev->new_layout = mddev->layout;
2877 * Adjust array size:
2879 * - in case of adding disk(s), array size has
2880 * to grow after the disk adding reshape,
2881 * which'll hapen in the event handler;
2882 * reshape will happen forward, so space has to
2883 * be available at the beginning of each disk
2885 * - in case of removing disk(s), array size
2886 * has to shrink before starting the reshape,
2887 * which'll happen here;
2888 * reshape will happen backward, so space has to
2889 * be available at the end of each disk
2891 * - data_offset and new_data_offset are
2892 * adjusted for aforementioned out of place
2893 * reshaping based on userspace passing in
2894 * the "data_offset <sectors>" key/value
2895 * pair via the constructor
2899 if (rs->delta_disks > 0) {
2900 /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2901 for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2902 rdev = &rs->dev[d].rdev;
2903 clear_bit(In_sync, &rdev->flags);
2906 * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2907 * by md, which'll store that erroneously in the superblock on reshape
2909 rdev->saved_raid_disk = -1;
2910 rdev->raid_disk = d;
2912 rdev->sectors = mddev->dev_sectors;
2913 rdev->recovery_offset = rs_is_raid1(rs) ? 0 : MaxSector;
2916 mddev->reshape_backwards = 0; /* adding disk(s) -> forward reshape */
2918 /* Remove disk(s) */
2919 } else if (rs->delta_disks < 0) {
2920 r = rs_set_dev_and_array_sectors(rs, true);
2921 mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2923 /* Change layout and/or chunk size */
2926 * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2928 * keeping number of disks and do layout change ->
2930 * toggle reshape_backward depending on data_offset:
2932 * - free space upfront -> reshape forward
2934 * - free space at the end -> reshape backward
2937 * This utilizes free reshape space avoiding the need
2938 * for userspace to move (parts of) LV segments in
2939 * case of layout/chunksize change (for disk
2940 * adding/removing reshape space has to be at
2941 * the proper address (see above with delta_disks):
2943 * add disk(s) -> begin
2944 * remove disk(s)-> end
2946 mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2950 * Adjust device size for forward reshape
2951 * because md_finish_reshape() reduces it.
2953 if (!mddev->reshape_backwards)
2954 rdev_for_each(rdev, &rs->md)
2955 if (!test_bit(Journal, &rdev->flags))
2956 rdev->sectors += reshape_sectors;
2962 * If the md resync thread has updated superblock with max reshape position
2963 * at the end of a reshape but not (yet) reset the layout configuration
2964 * changes -> reset the latter.
2966 static void rs_reset_inconclusive_reshape(struct raid_set *rs)
2968 if (!rs_is_reshaping(rs) && rs_is_layout_change(rs, true)) {
2970 rs->md.delta_disks = 0;
2971 rs->md.reshape_backwards = 0;
2976 * Enable/disable discard support on RAID set depending on
2977 * RAID level and discard properties of underlying RAID members.
2979 static void configure_discard_support(struct raid_set *rs)
2983 struct dm_target *ti = rs->ti;
2986 * XXX: RAID level 4,5,6 require zeroing for safety.
2988 raid456 = rs_is_raid456(rs);
2990 for (i = 0; i < rs->raid_disks; i++) {
2991 struct request_queue *q;
2993 if (!rs->dev[i].rdev.bdev)
2996 q = bdev_get_queue(rs->dev[i].rdev.bdev);
2997 if (!q || !blk_queue_discard(q))
3001 if (!devices_handle_discard_safely) {
3002 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
3003 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
3010 * RAID1 and RAID10 personalities require bio splitting,
3011 * RAID0/4/5/6 don't and process large discard bios properly.
3013 ti->split_discard_bios = !!(rs_is_raid1(rs) || rs_is_raid10(rs));
3014 ti->num_discard_bios = 1;
3018 * Construct a RAID0/1/10/4/5/6 mapping:
3020 * <raid_type> <#raid_params> <raid_params>{0,} \
3021 * <#raid_devs> [<meta_dev1> <dev1>]{1,}
3023 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
3024 * details on possible <raid_params>.
3026 * Userspace is free to initialize the metadata devices, hence the superblocks to
3027 * enforce recreation based on the passed in table parameters.
3030 static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3033 bool resize = false;
3034 struct raid_type *rt;
3035 unsigned int num_raid_params, num_raid_devs;
3036 sector_t calculated_dev_sectors, rdev_sectors, reshape_sectors;
3037 struct raid_set *rs = NULL;
3039 struct rs_layout rs_layout;
3040 struct dm_arg_set as = { argc, argv }, as_nrd;
3041 struct dm_arg _args[] = {
3042 { 0, as.argc, "Cannot understand number of raid parameters" },
3043 { 1, 254, "Cannot understand number of raid devices parameters" }
3046 /* Must have <raid_type> */
3047 arg = dm_shift_arg(&as);
3049 ti->error = "No arguments";
3053 rt = get_raid_type(arg);
3055 ti->error = "Unrecognised raid_type";
3059 /* Must have <#raid_params> */
3060 if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
3063 /* number of raid device tupples <meta_dev data_dev> */
3065 dm_consume_args(&as_nrd, num_raid_params);
3066 _args[1].max = (as_nrd.argc - 1) / 2;
3067 if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
3070 if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
3071 ti->error = "Invalid number of supplied raid devices";
3075 rs = raid_set_alloc(ti, rt, num_raid_devs);
3079 r = parse_raid_params(rs, &as, num_raid_params);
3083 r = parse_dev_params(rs, &as);
3087 rs->md.sync_super = super_sync;
3090 * Calculate ctr requested array and device sizes to allow
3091 * for superblock analysis needing device sizes defined.
3093 * Any existing superblock will overwrite the array and device sizes
3095 r = rs_set_dev_and_array_sectors(rs, false);
3099 calculated_dev_sectors = rs->md.dev_sectors;
3102 * Backup any new raid set level, layout, ...
3103 * requested to be able to compare to superblock
3104 * members for conversion decisions.
3106 rs_config_backup(rs, &rs_layout);
3108 r = analyse_superblocks(ti, rs);
3112 rdev_sectors = __rdev_sectors(rs);
3113 if (!rdev_sectors) {
3114 ti->error = "Invalid rdev size";
3120 reshape_sectors = _get_reshape_sectors(rs);
3121 if (calculated_dev_sectors != rdev_sectors)
3122 resize = calculated_dev_sectors != (reshape_sectors ? rdev_sectors - reshape_sectors : rdev_sectors);
3124 INIT_WORK(&rs->md.event_work, do_table_event);
3126 ti->num_flush_bios = 1;
3128 /* Restore any requested new layout for conversion decision */
3129 rs_config_restore(rs, &rs_layout);
3132 * Now that we have any superblock metadata available,
3133 * check for new, recovering, reshaping, to be taken over,
3134 * to be reshaped or an existing, unchanged raid set to
3137 if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
3138 /* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
3139 if (rs_is_raid6(rs) &&
3140 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
3141 ti->error = "'nosync' not allowed for new raid6 set";
3145 rs_setup_recovery(rs, 0);
3146 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3148 } else if (rs_is_recovering(rs)) {
3149 /* Rebuild particular devices */
3150 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3151 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3152 rs_setup_recovery(rs, MaxSector);
3154 /* A recovering raid set may be resized */
3155 ; /* skip setup rs */
3156 } else if (rs_is_reshaping(rs)) {
3157 /* Have to reject size change request during reshape */
3159 ti->error = "Can't resize a reshaping raid set";
3164 } else if (rs_takeover_requested(rs)) {
3165 if (rs_is_reshaping(rs)) {
3166 ti->error = "Can't takeover a reshaping raid set";
3171 /* We can't takeover a journaled raid4/5/6 */
3172 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3173 ti->error = "Can't takeover a journaled raid4/5/6 set";
3179 * If a takeover is needed, userspace sets any additional
3180 * devices to rebuild and we can check for a valid request here.
3182 * If acceptible, set the level to the new requested
3183 * one, prohibit requesting recovery, allow the raid
3184 * set to run and store superblocks during resume.
3186 r = rs_check_takeover(rs);
3190 r = rs_setup_takeover(rs);
3194 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3195 /* Takeover ain't recovery, so disable recovery */
3196 rs_setup_recovery(rs, MaxSector);
3198 } else if (rs_reshape_requested(rs)) {
3200 * No need to check for 'ongoing' takeover here, because takeover
3201 * is an instant operation as oposed to an ongoing reshape.
3204 /* We can't reshape a journaled raid4/5/6 */
3205 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3206 ti->error = "Can't reshape a journaled raid4/5/6 set";
3211 /* Out-of-place space has to be available to allow for a reshape unless raid1! */
3212 if (reshape_sectors || rs_is_raid1(rs)) {
3214 * We can only prepare for a reshape here, because the
3215 * raid set needs to run to provide the repective reshape
3216 * check functions via its MD personality instance.
3218 * So do the reshape check after md_run() succeeded.
3220 r = rs_prepare_reshape(rs);
3224 /* Reshaping ain't recovery, so disable recovery */
3225 rs_setup_recovery(rs, MaxSector);
3229 /* May not set recovery when a device rebuild is requested */
3230 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3231 rs_setup_recovery(rs, MaxSector);
3232 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3234 rs_setup_recovery(rs, test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ?
3235 0 : (resize ? calculated_dev_sectors : MaxSector));
3239 /* If constructor requested it, change data and new_data offsets */
3240 r = rs_adjust_data_offsets(rs);
3244 /* Catch any inconclusive reshape superblock content. */
3245 rs_reset_inconclusive_reshape(rs);
3247 /* Start raid set read-only and assumed clean to change in raid_resume() */
3251 /* Keep array frozen until resume. */
3252 set_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
3254 /* Has to be held on running the array */
3255 mddev_lock_nointr(&rs->md);
3256 r = md_run(&rs->md);
3257 rs->md.in_sync = 0; /* Assume already marked dirty */
3259 ti->error = "Failed to run raid array";
3260 mddev_unlock(&rs->md);
3264 r = md_start(&rs->md);
3266 ti->error = "Failed to start raid array";
3267 mddev_unlock(&rs->md);
3271 rs->callbacks.congested_fn = raid_is_congested;
3272 dm_table_add_target_callbacks(ti->table, &rs->callbacks);
3274 /* If raid4/5/6 journal mode explicitly requested (only possible with journal dev) -> set it */
3275 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
3276 r = r5c_journal_mode_set(&rs->md, rs->journal_dev.mode);
3278 ti->error = "Failed to set raid4/5/6 journal mode";
3279 mddev_unlock(&rs->md);
3280 goto bad_journal_mode_set;
3284 mddev_suspend(&rs->md);
3285 set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags);
3287 /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
3288 if (rs_is_raid456(rs)) {
3289 r = rs_set_raid456_stripe_cache(rs);
3291 goto bad_stripe_cache;
3294 /* Now do an early reshape check */
3295 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3296 r = rs_check_reshape(rs);
3298 goto bad_check_reshape;
3300 /* Restore new, ctr requested layout to perform check */
3301 rs_config_restore(rs, &rs_layout);
3303 if (rs->md.pers->start_reshape) {
3304 r = rs->md.pers->check_reshape(&rs->md);
3306 ti->error = "Reshape check failed";
3307 goto bad_check_reshape;
3312 /* Disable/enable discard support on raid set. */
3313 configure_discard_support(rs);
3315 mddev_unlock(&rs->md);
3319 bad_journal_mode_set:
3329 static void raid_dtr(struct dm_target *ti)
3331 struct raid_set *rs = ti->private;
3333 list_del_init(&rs->callbacks.list);
3338 static int raid_map(struct dm_target *ti, struct bio *bio)
3340 struct raid_set *rs = ti->private;
3341 struct mddev *mddev = &rs->md;
3344 * If we're reshaping to add disk(s)), ti->len and
3345 * mddev->array_sectors will differ during the process
3346 * (ti->len > mddev->array_sectors), so we have to requeue
3347 * bios with addresses > mddev->array_sectors here or
3348 * there will occur accesses past EOD of the component
3349 * data images thus erroring the raid set.
3351 if (unlikely(bio_end_sector(bio) > mddev->array_sectors))
3352 return DM_MAPIO_REQUEUE;
3354 md_handle_request(mddev, bio);
3356 return DM_MAPIO_SUBMITTED;
3359 /* Return sync state string for @state */
3360 enum sync_state { st_frozen, st_reshape, st_resync, st_check, st_repair, st_recover, st_idle };
3361 static const char *sync_str(enum sync_state state)
3363 /* Has to be in above sync_state order! */
3364 static const char *sync_strs[] = {
3374 return __within_range(state, 0, ARRAY_SIZE(sync_strs) - 1) ? sync_strs[state] : "undef";
3377 /* Return enum sync_state for @mddev derived from @recovery flags */
3378 static enum sync_state decipher_sync_action(struct mddev *mddev, unsigned long recovery)
3380 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
3383 /* The MD sync thread can be done with io or be interrupted but still be running */
3384 if (!test_bit(MD_RECOVERY_DONE, &recovery) &&
3385 (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
3386 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery)))) {
3387 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
3390 if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
3391 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
3393 if (test_bit(MD_RECOVERY_CHECK, &recovery))
3398 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3401 if (mddev->reshape_position != MaxSector)
3409 * Return status string for @rdev
3411 * Status characters:
3413 * 'D' = Dead/Failed raid set component or raid4/5/6 journal device
3414 * 'a' = Alive but not in-sync raid set component _or_ alive raid4/5/6 'write_back' journal device
3415 * 'A' = Alive and in-sync raid set component _or_ alive raid4/5/6 'write_through' journal device
3416 * '-' = Non-existing device (i.e. uspace passed '- -' into the ctr)
3418 static const char *__raid_dev_status(struct raid_set *rs, struct md_rdev *rdev)
3422 else if (test_bit(Faulty, &rdev->flags))
3424 else if (test_bit(Journal, &rdev->flags))
3425 return (rs->journal_dev.mode == R5C_JOURNAL_MODE_WRITE_THROUGH) ? "A" : "a";
3426 else if (test_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags) ||
3427 (!test_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags) &&
3428 !test_bit(In_sync, &rdev->flags)))
3434 /* Helper to return resync/reshape progress for @rs and runtime flags for raid set in sync / resynching */
3435 static sector_t rs_get_progress(struct raid_set *rs, unsigned long recovery,
3436 sector_t resync_max_sectors)
3439 enum sync_state state;
3440 struct mddev *mddev = &rs->md;
3442 clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3443 clear_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3445 if (rs_is_raid0(rs)) {
3446 r = resync_max_sectors;
3447 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3450 state = decipher_sync_action(mddev, recovery);
3452 if (state == st_idle && !test_bit(MD_RECOVERY_INTR, &recovery))
3453 r = mddev->recovery_cp;
3455 r = mddev->curr_resync_completed;
3457 if (state == st_idle && r >= resync_max_sectors) {
3461 /* In case we have finished recovering, the array is in sync. */
3462 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3463 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3465 } else if (state == st_recover)
3467 * In case we are recovering, the array is not in sync
3468 * and health chars should show the recovering legs.
3471 else if (state == st_resync)
3473 * If "resync" is occurring, the raid set
3474 * is or may be out of sync hence the health
3475 * characters shall be 'a'.
3477 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3478 else if (state == st_reshape)
3480 * If "reshape" is occurring, the raid set
3481 * is or may be out of sync hence the health
3482 * characters shall be 'a'.
3484 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3486 else if (state == st_check || state == st_repair)
3488 * If "check" or "repair" is occurring, the raid set has
3489 * undergone an initial sync and the health characters
3490 * should not be 'a' anymore.
3492 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3495 struct md_rdev *rdev;
3498 * We are idle and recovery is needed, prevent 'A' chars race
3499 * caused by components still set to in-sync by constructor.
3501 if (test_bit(MD_RECOVERY_NEEDED, &recovery))
3502 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3505 * The raid set may be doing an initial sync, or it may
3506 * be rebuilding individual components. If all the
3507 * devices are In_sync, then it is the raid set that is
3508 * being initialized.
3510 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3511 rdev_for_each(rdev, mddev)
3512 if (!test_bit(Journal, &rdev->flags) &&
3513 !test_bit(In_sync, &rdev->flags)) {
3514 clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3520 return min(r, resync_max_sectors);
3523 /* Helper to return @dev name or "-" if !@dev */
3524 static const char *__get_dev_name(struct dm_dev *dev)
3526 return dev ? dev->name : "-";
3529 static void raid_status(struct dm_target *ti, status_type_t type,
3530 unsigned int status_flags, char *result, unsigned int maxlen)
3532 struct raid_set *rs = ti->private;
3533 struct mddev *mddev = &rs->md;
3534 struct r5conf *conf = mddev->private;
3535 int i, max_nr_stripes = conf ? conf->max_nr_stripes : 0;
3536 unsigned long recovery;
3537 unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3538 unsigned int sz = 0;
3539 unsigned int rebuild_disks;
3540 unsigned int write_mostly_params = 0;
3541 sector_t progress, resync_max_sectors, resync_mismatches;
3542 const char *sync_action;
3543 struct raid_type *rt;
3546 case STATUSTYPE_INFO:
3547 /* *Should* always succeed */
3548 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3552 DMEMIT("%s %d ", rt->name, mddev->raid_disks);
3554 /* Access most recent mddev properties for status output */
3556 recovery = rs->md.recovery;
3557 /* Get sensible max sectors even if raid set not yet started */
3558 resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
3559 mddev->resync_max_sectors : mddev->dev_sectors;
3560 progress = rs_get_progress(rs, recovery, resync_max_sectors);
3561 resync_mismatches = (mddev->last_sync_action && !strcasecmp(mddev->last_sync_action, "check")) ?
3562 atomic64_read(&mddev->resync_mismatches) : 0;
3563 sync_action = sync_str(decipher_sync_action(&rs->md, recovery));
3565 /* HM FIXME: do we want another state char for raid0? It shows 'D'/'A'/'-' now */
3566 for (i = 0; i < rs->raid_disks; i++)
3567 DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev));
3570 * In-sync/Reshape ratio:
3571 * The in-sync ratio shows the progress of:
3572 * - Initializing the raid set
3573 * - Rebuilding a subset of devices of the raid set
3574 * The user can distinguish between the two by referring
3575 * to the status characters.
3577 * The reshape ratio shows the progress of
3578 * changing the raid layout or the number of
3579 * disks of a raid set
3581 DMEMIT(" %llu/%llu", (unsigned long long) progress,
3582 (unsigned long long) resync_max_sectors);
3588 * See Documentation/device-mapper/dm-raid.txt for
3589 * information on each of these states.
3591 DMEMIT(" %s", sync_action);
3596 * resync_mismatches/mismatch_cnt
3597 * This field shows the number of discrepancies found when
3598 * performing a "check" of the raid set.
3600 DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3605 * data_offset (needed for out of space reshaping)
3606 * This field shows the data offset into the data
3607 * image LV where the first stripes data starts.
3609 * We keep data_offset equal on all raid disks of the set,
3610 * so retrieving it from the first raid disk is sufficient.
3612 DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
3617 DMEMIT(" %s", test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ?
3618 __raid_dev_status(rs, &rs->journal_dev.rdev) : "-");
3621 case STATUSTYPE_TABLE:
3622 /* Report the table line string you would use to construct this raid set */
3624 /* Calculate raid parameter count */
3625 for (i = 0; i < rs->raid_disks; i++)
3626 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3627 write_mostly_params += 2;
3628 rebuild_disks = memweight(rs->rebuild_disks, DISKS_ARRAY_ELEMS * sizeof(*rs->rebuild_disks));
3629 raid_param_cnt += rebuild_disks * 2 +
3630 write_mostly_params +
3631 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3632 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2 +
3633 (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ? 2 : 0) +
3634 (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags) ? 2 : 0);
3636 /* Emit table line */
3637 /* This has to be in the documented order for userspace! */
3638 DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
3639 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
3640 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
3641 if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3642 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
3644 for (i = 0; i < rs->raid_disks; i++)
3645 if (test_bit(rs->dev[i].rdev.raid_disk, (void *) rs->rebuild_disks))
3646 DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD),
3647 rs->dev[i].rdev.raid_disk);
3648 if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3649 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3650 mddev->bitmap_info.daemon_sleep);
3651 if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3652 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3653 mddev->sync_speed_min);
3654 if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3655 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3656 mddev->sync_speed_max);
3657 if (write_mostly_params)
3658 for (i = 0; i < rs->raid_disks; i++)
3659 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3660 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3661 rs->dev[i].rdev.raid_disk);
3662 if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
3663 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
3664 mddev->bitmap_info.max_write_behind);
3665 if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3666 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3668 if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3669 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3670 (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3671 if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3672 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3673 raid10_md_layout_to_copies(mddev->layout));
3674 if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3675 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3676 raid10_md_layout_to_format(mddev->layout));
3677 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3678 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3679 max(rs->delta_disks, mddev->delta_disks));
3680 if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3681 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3682 (unsigned long long) rs->data_offset);
3683 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags))
3684 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV),
3685 __get_dev_name(rs->journal_dev.dev));
3686 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags))
3687 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE),
3688 md_journal_mode_to_dm_raid(rs->journal_dev.mode));
3689 DMEMIT(" %d", rs->raid_disks);
3690 for (i = 0; i < rs->raid_disks; i++)
3691 DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
3692 __get_dev_name(rs->dev[i].data_dev));
3696 static int raid_message(struct dm_target *ti, unsigned int argc, char **argv,
3697 char *result, unsigned maxlen)
3699 struct raid_set *rs = ti->private;
3700 struct mddev *mddev = &rs->md;
3702 if (!mddev->pers || !mddev->pers->sync_request)
3705 if (!strcasecmp(argv[0], "frozen"))
3706 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3708 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3710 if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
3711 if (mddev->sync_thread) {
3712 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3713 md_reap_sync_thread(mddev);
3715 } else if (decipher_sync_action(mddev, mddev->recovery) != st_idle)
3717 else if (!strcasecmp(argv[0], "resync"))
3718 ; /* MD_RECOVERY_NEEDED set below */
3719 else if (!strcasecmp(argv[0], "recover"))
3720 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3722 if (!strcasecmp(argv[0], "check")) {
3723 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3724 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3725 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3726 } else if (!strcasecmp(argv[0], "repair")) {
3727 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3728 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3732 if (mddev->ro == 2) {
3733 /* A write to sync_action is enough to justify
3734 * canceling read-auto mode
3737 if (!mddev->suspended && mddev->sync_thread)
3738 md_wakeup_thread(mddev->sync_thread);
3740 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3741 if (!mddev->suspended && mddev->thread)
3742 md_wakeup_thread(mddev->thread);
3747 static int raid_iterate_devices(struct dm_target *ti,
3748 iterate_devices_callout_fn fn, void *data)
3750 struct raid_set *rs = ti->private;
3754 for (i = 0; !r && i < rs->md.raid_disks; i++)
3755 if (rs->dev[i].data_dev)
3757 rs->dev[i].data_dev,
3758 0, /* No offset on data devs */
3765 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3767 struct raid_set *rs = ti->private;
3768 unsigned int chunk_size = to_bytes(rs->md.chunk_sectors);
3770 blk_limits_io_min(limits, chunk_size);
3771 blk_limits_io_opt(limits, chunk_size * mddev_data_stripes(rs));
3774 static void raid_postsuspend(struct dm_target *ti)
3776 struct raid_set *rs = ti->private;
3778 if (!test_and_set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
3779 /* Writes have to be stopped before suspending to avoid deadlocks. */
3780 if (!test_bit(MD_RECOVERY_FROZEN, &rs->md.recovery))
3781 md_stop_writes(&rs->md);
3783 mddev_lock_nointr(&rs->md);
3784 mddev_suspend(&rs->md);
3785 mddev_unlock(&rs->md);
3789 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
3792 uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
3793 unsigned long flags;
3794 bool cleared = false;
3795 struct dm_raid_superblock *sb;
3796 struct mddev *mddev = &rs->md;
3799 /* RAID personalities have to provide hot add/remove methods or we need to bail out. */
3800 if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
3803 memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
3805 for (i = 0; i < mddev->raid_disks; i++) {
3806 r = &rs->dev[i].rdev;
3807 /* HM FIXME: enhance journal device recovery processing */
3808 if (test_bit(Journal, &r->flags))
3811 if (test_bit(Faulty, &r->flags) &&
3812 r->meta_bdev && !read_disk_sb(r, r->sb_size, true)) {
3813 DMINFO("Faulty %s device #%d has readable super block."
3814 " Attempting to revive it.",
3815 rs->raid_type->name, i);
3818 * Faulty bit may be set, but sometimes the array can
3819 * be suspended before the personalities can respond
3820 * by removing the device from the array (i.e. calling
3821 * 'hot_remove_disk'). If they haven't yet removed
3822 * the failed device, its 'raid_disk' number will be
3823 * '>= 0' - meaning we must call this function
3827 clear_bit(In_sync, &r->flags); /* Mandatory for hot remove. */
3828 if (r->raid_disk >= 0) {
3829 if (mddev->pers->hot_remove_disk(mddev, r)) {
3830 /* Failed to revive this device, try next */
3835 r->raid_disk = r->saved_raid_disk = i;
3837 clear_bit(Faulty, &r->flags);
3838 clear_bit(WriteErrorSeen, &r->flags);
3840 if (mddev->pers->hot_add_disk(mddev, r)) {
3841 /* Failed to revive this device, try next */
3842 r->raid_disk = r->saved_raid_disk = -1;
3845 clear_bit(In_sync, &r->flags);
3846 r->recovery_offset = 0;
3847 set_bit(i, (void *) cleared_failed_devices);
3853 /* If any failed devices could be cleared, update all sbs failed_devices bits */
3855 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
3857 rdev_for_each(r, &rs->md) {
3858 if (test_bit(Journal, &r->flags))
3861 sb = page_address(r->sb_page);
3862 sb_retrieve_failed_devices(sb, failed_devices);
3864 for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
3865 failed_devices[i] &= ~cleared_failed_devices[i];
3867 sb_update_failed_devices(sb, failed_devices);
3872 static int __load_dirty_region_bitmap(struct raid_set *rs)
3876 /* Try loading the bitmap unless "raid0", which does not have one */
3877 if (!rs_is_raid0(rs) &&
3878 !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
3879 r = md_bitmap_load(&rs->md);
3881 DMERR("Failed to load bitmap");
3887 /* Enforce updating all superblocks */
3888 static void rs_update_sbs(struct raid_set *rs)
3890 struct mddev *mddev = &rs->md;
3893 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3895 md_update_sb(mddev, 1);
3900 * Reshape changes raid algorithm of @rs to new one within personality
3901 * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3902 * disks from a raid set thus growing/shrinking it or resizes the set
3904 * Call mddev_lock_nointr() before!
3906 static int rs_start_reshape(struct raid_set *rs)
3909 struct mddev *mddev = &rs->md;
3910 struct md_personality *pers = mddev->pers;
3912 /* Don't allow the sync thread to work until the table gets reloaded. */
3913 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
3915 r = rs_setup_reshape(rs);
3920 * Check any reshape constraints enforced by the personalility
3922 * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3924 r = pers->check_reshape(mddev);
3926 rs->ti->error = "pers->check_reshape() failed";
3931 * Personality may not provide start reshape method in which
3932 * case check_reshape above has already covered everything
3934 if (pers->start_reshape) {
3935 r = pers->start_reshape(mddev);
3937 rs->ti->error = "pers->start_reshape() failed";
3943 * Now reshape got set up, update superblocks to
3944 * reflect the fact so that a table reload will
3945 * access proper superblock content in the ctr.
3952 static int raid_preresume(struct dm_target *ti)
3955 struct raid_set *rs = ti->private;
3956 struct mddev *mddev = &rs->md;
3958 /* This is a resume after a suspend of the set -> it's already started. */
3959 if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
3963 * The superblocks need to be updated on disk if the
3964 * array is new or new devices got added (thus zeroed
3965 * out by userspace) or __load_dirty_region_bitmap
3966 * will overwrite them in core with old data or fail.
3968 if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
3971 /* Load the bitmap from disk unless raid0 */
3972 r = __load_dirty_region_bitmap(rs);
3976 /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) */
3977 if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) && mddev->bitmap &&
3978 mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)) {
3979 r = md_bitmap_resize(mddev->bitmap, mddev->dev_sectors,
3980 to_bytes(rs->requested_bitmap_chunk_sectors), 0);
3982 DMERR("Failed to resize bitmap");
3985 /* Check for any resize/reshape on @rs and adjust/initiate */
3986 /* Be prepared for mddev_resume() in raid_resume() */
3987 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3988 if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
3989 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3990 mddev->resync_min = mddev->recovery_cp;
3993 /* Check for any reshape request unless new raid set */
3994 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3995 /* Initiate a reshape. */
3996 rs_set_rdev_sectors(rs);
3997 mddev_lock_nointr(mddev);
3998 r = rs_start_reshape(rs);
3999 mddev_unlock(mddev);
4001 DMWARN("Failed to check/start reshape, continuing without change");
4008 static void raid_resume(struct dm_target *ti)
4010 struct raid_set *rs = ti->private;
4011 struct mddev *mddev = &rs->md;
4013 if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
4015 * A secondary resume while the device is active.
4016 * Take this opportunity to check whether any failed
4017 * devices are reachable again.
4019 attempt_restore_of_faulty_devices(rs);
4022 if (test_and_clear_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
4023 /* Only reduce raid set size before running a disk removing reshape. */
4024 if (mddev->delta_disks < 0)
4025 rs_set_capacity(rs);
4027 mddev_lock_nointr(mddev);
4028 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4031 mddev_resume(mddev);
4032 mddev_unlock(mddev);
4036 static struct target_type raid_target = {
4038 .version = {1, 14, 0},
4039 .module = THIS_MODULE,
4043 .status = raid_status,
4044 .message = raid_message,
4045 .iterate_devices = raid_iterate_devices,
4046 .io_hints = raid_io_hints,
4047 .postsuspend = raid_postsuspend,
4048 .preresume = raid_preresume,
4049 .resume = raid_resume,
4052 static int __init dm_raid_init(void)
4054 DMINFO("Loading target version %u.%u.%u",
4055 raid_target.version[0],
4056 raid_target.version[1],
4057 raid_target.version[2]);
4058 return dm_register_target(&raid_target);
4061 static void __exit dm_raid_exit(void)
4063 dm_unregister_target(&raid_target);
4066 module_init(dm_raid_init);
4067 module_exit(dm_raid_exit);
4069 module_param(devices_handle_discard_safely, bool, 0644);
4070 MODULE_PARM_DESC(devices_handle_discard_safely,
4071 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
4073 MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
4074 MODULE_ALIAS("dm-raid0");
4075 MODULE_ALIAS("dm-raid1");
4076 MODULE_ALIAS("dm-raid10");
4077 MODULE_ALIAS("dm-raid4");
4078 MODULE_ALIAS("dm-raid5");
4079 MODULE_ALIAS("dm-raid6");
4080 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
4081 MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
4082 MODULE_LICENSE("GPL");