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, raid_disks;
1002 unsigned int group_size, last_group_start;
1004 for (i = 0; i < rs->raid_disks; i++)
1005 if (!test_bit(FirstUse, &rs->dev[i].rdev.flags) &&
1006 ((!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
1007 !rs->dev[i].rdev.sb_page)))
1010 switch (rs->md.level) {
1014 if (rebuild_cnt >= rs->md.raid_disks)
1020 if (rebuild_cnt > rs->raid_type->parity_devs)
1024 copies = raid10_md_layout_to_copies(rs->md.new_layout);
1026 DMERR("Bogus raid10 data copies < 2!");
1030 if (rebuild_cnt < copies)
1034 * It is possible to have a higher rebuild count for RAID10,
1035 * as long as the failed devices occur in different mirror
1036 * groups (i.e. different stripes).
1038 * When checking "near" format, make sure no adjacent devices
1039 * have failed beyond what can be handled. In addition to the
1040 * simple case where the number of devices is a multiple of the
1041 * number of copies, we must also handle cases where the number
1042 * of devices is not a multiple of the number of copies.
1043 * E.g. dev1 dev2 dev3 dev4 dev5
1047 raid_disks = min(rs->raid_disks, rs->md.raid_disks);
1048 if (__is_raid10_near(rs->md.new_layout)) {
1049 for (i = 0; i < raid_disks; i++) {
1051 rebuilds_per_group = 0;
1052 if ((!rs->dev[i].rdev.sb_page ||
1053 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1054 (++rebuilds_per_group >= copies))
1061 * When checking "far" and "offset" formats, we need to ensure
1062 * that the device that holds its copy is not also dead or
1063 * being rebuilt. (Note that "far" and "offset" formats only
1064 * support two copies right now. These formats also only ever
1065 * use the 'use_far_sets' variant.)
1067 * This check is somewhat complicated by the need to account
1068 * for arrays that are not a multiple of (far) copies. This
1069 * results in the need to treat the last (potentially larger)
1072 group_size = (raid_disks / copies);
1073 last_group_start = (raid_disks / group_size) - 1;
1074 last_group_start *= group_size;
1075 for (i = 0; i < raid_disks; i++) {
1076 if (!(i % copies) && !(i > last_group_start))
1077 rebuilds_per_group = 0;
1078 if ((!rs->dev[i].rdev.sb_page ||
1079 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1080 (++rebuilds_per_group >= copies))
1096 * Possible arguments are...
1097 * <chunk_size> [optional_args]
1099 * Argument definitions
1100 * <chunk_size> The number of sectors per disk that
1101 * will form the "stripe"
1102 * [[no]sync] Force or prevent recovery of the
1104 * [rebuild <idx>] Rebuild the drive indicated by the index
1105 * [daemon_sleep <ms>] Time between bitmap daemon work to
1107 * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1108 * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1109 * [write_mostly <idx>] Indicate a write mostly drive via index
1110 * [max_write_behind <sectors>] See '-write-behind=' (man mdadm)
1111 * [stripe_cache <sectors>] Stripe cache size for higher RAIDs
1112 * [region_size <sectors>] Defines granularity of bitmap
1113 * [journal_dev <dev>] raid4/5/6 journaling deviice
1114 * (i.e. write hole closing log)
1116 * RAID10-only options:
1117 * [raid10_copies <# copies>] Number of copies. (Default: 2)
1118 * [raid10_format <near|far|offset>] Layout algorithm. (Default: near)
1120 static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
1121 unsigned int num_raid_params)
1123 int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
1124 unsigned int raid10_copies = 2;
1125 unsigned int i, write_mostly = 0;
1126 unsigned int region_size = 0;
1127 sector_t max_io_len;
1128 const char *arg, *key;
1129 struct raid_dev *rd;
1130 struct raid_type *rt = rs->raid_type;
1132 arg = dm_shift_arg(as);
1133 num_raid_params--; /* Account for chunk_size argument */
1135 if (kstrtoint(arg, 10, &value) < 0) {
1136 rs->ti->error = "Bad numerical argument given for chunk_size";
1141 * First, parse the in-order required arguments
1142 * "chunk_size" is the only argument of this type.
1144 if (rt_is_raid1(rt)) {
1146 DMERR("Ignoring chunk size parameter for RAID 1");
1148 } else if (!is_power_of_2(value)) {
1149 rs->ti->error = "Chunk size must be a power of 2";
1151 } else if (value < 8) {
1152 rs->ti->error = "Chunk size value is too small";
1156 rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
1159 * We set each individual device as In_sync with a completed
1160 * 'recovery_offset'. If there has been a device failure or
1161 * replacement then one of the following cases applies:
1163 * 1) User specifies 'rebuild'.
1164 * - Device is reset when param is read.
1165 * 2) A new device is supplied.
1166 * - No matching superblock found, resets device.
1167 * 3) Device failure was transient and returns on reload.
1168 * - Failure noticed, resets device for bitmap replay.
1169 * 4) Device hadn't completed recovery after previous failure.
1170 * - Superblock is read and overrides recovery_offset.
1172 * What is found in the superblocks of the devices is always
1173 * authoritative, unless 'rebuild' or '[no]sync' was specified.
1175 for (i = 0; i < rs->raid_disks; i++) {
1176 set_bit(In_sync, &rs->dev[i].rdev.flags);
1177 rs->dev[i].rdev.recovery_offset = MaxSector;
1181 * Second, parse the unordered optional arguments
1183 for (i = 0; i < num_raid_params; i++) {
1184 key = dm_shift_arg(as);
1186 rs->ti->error = "Not enough raid parameters given";
1190 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
1191 if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1192 rs->ti->error = "Only one 'nosync' argument allowed";
1197 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
1198 if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
1199 rs->ti->error = "Only one 'sync' argument allowed";
1204 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
1205 if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1206 rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1212 arg = dm_shift_arg(as);
1213 i++; /* Account for the argument pairs */
1215 rs->ti->error = "Wrong number of raid parameters given";
1220 * Parameters that take a string value are checked here.
1222 /* "raid10_format {near|offset|far} */
1223 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
1224 if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
1225 rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1228 if (!rt_is_raid10(rt)) {
1229 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1232 raid10_format = raid10_name_to_format(arg);
1233 if (raid10_format < 0) {
1234 rs->ti->error = "Invalid 'raid10_format' value given";
1235 return raid10_format;
1240 /* "journal_dev <dev>" */
1241 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV))) {
1243 struct md_rdev *jdev;
1245 if (test_and_set_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1246 rs->ti->error = "Only one raid4/5/6 set journaling device allowed";
1249 if (!rt_is_raid456(rt)) {
1250 rs->ti->error = "'journal_dev' is an invalid parameter for this RAID type";
1253 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
1254 &rs->journal_dev.dev);
1256 rs->ti->error = "raid4/5/6 journal device lookup failure";
1259 jdev = &rs->journal_dev.rdev;
1261 jdev->mddev = &rs->md;
1262 jdev->bdev = rs->journal_dev.dev->bdev;
1263 jdev->sectors = to_sector(i_size_read(jdev->bdev->bd_inode));
1264 if (jdev->sectors < MIN_RAID456_JOURNAL_SPACE) {
1265 rs->ti->error = "No space for raid4/5/6 journal";
1268 rs->journal_dev.mode = R5C_JOURNAL_MODE_WRITE_THROUGH;
1269 set_bit(Journal, &jdev->flags);
1273 /* "journal_mode <mode>" ("journal_dev" mandatory!) */
1274 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE))) {
1277 if (!test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1278 rs->ti->error = "raid4/5/6 'journal_mode' is invalid without 'journal_dev'";
1281 if (test_and_set_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
1282 rs->ti->error = "Only one raid4/5/6 'journal_mode' argument allowed";
1285 r = dm_raid_journal_mode_to_md(arg);
1287 rs->ti->error = "Invalid 'journal_mode' argument";
1290 rs->journal_dev.mode = r;
1295 * Parameters with number values from here on.
1297 if (kstrtoint(arg, 10, &value) < 0) {
1298 rs->ti->error = "Bad numerical argument given in raid params";
1302 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
1304 * "rebuild" is being passed in by userspace to provide
1305 * indexes of replaced devices and to set up additional
1306 * devices on raid level takeover.
1308 if (!__within_range(value, 0, rs->raid_disks - 1)) {
1309 rs->ti->error = "Invalid rebuild index given";
1313 if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1314 rs->ti->error = "rebuild for this index already given";
1318 rd = rs->dev + value;
1319 clear_bit(In_sync, &rd->rdev.flags);
1320 clear_bit(Faulty, &rd->rdev.flags);
1321 rd->rdev.recovery_offset = 0;
1322 set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
1323 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
1324 if (!rt_is_raid1(rt)) {
1325 rs->ti->error = "write_mostly option is only valid for RAID1";
1329 if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
1330 rs->ti->error = "Invalid write_mostly index given";
1335 set_bit(WriteMostly, &rs->dev[value].rdev.flags);
1336 set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
1337 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
1338 if (!rt_is_raid1(rt)) {
1339 rs->ti->error = "max_write_behind option is only valid for RAID1";
1343 if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
1344 rs->ti->error = "Only one max_write_behind argument pair allowed";
1349 * In device-mapper, we specify things in sectors, but
1350 * MD records this value in kB
1352 if (value < 0 || value / 2 > COUNTER_MAX) {
1353 rs->ti->error = "Max write-behind limit out of range";
1357 rs->md.bitmap_info.max_write_behind = value / 2;
1358 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
1359 if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
1360 rs->ti->error = "Only one daemon_sleep argument pair allowed";
1364 rs->ti->error = "daemon sleep period out of range";
1367 rs->md.bitmap_info.daemon_sleep = value;
1368 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
1369 /* Userspace passes new data_offset after having extended the the data image LV */
1370 if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
1371 rs->ti->error = "Only one data_offset argument pair allowed";
1374 /* Ensure sensible data offset */
1376 (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
1377 rs->ti->error = "Bogus data_offset value";
1380 rs->data_offset = value;
1381 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
1382 /* Define the +/-# of disks to add to/remove from the given raid set */
1383 if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1384 rs->ti->error = "Only one delta_disks argument pair allowed";
1387 /* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1388 if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
1389 rs->ti->error = "Too many delta_disk requested";
1393 rs->delta_disks = value;
1394 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
1395 if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
1396 rs->ti->error = "Only one stripe_cache argument pair allowed";
1400 if (!rt_is_raid456(rt)) {
1401 rs->ti->error = "Inappropriate argument: stripe_cache";
1406 rs->ti->error = "Bogus stripe cache entries value";
1409 rs->stripe_cache_entries = value;
1410 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
1411 if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1412 rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1417 rs->ti->error = "min_recovery_rate out of range";
1420 rs->md.sync_speed_min = value;
1421 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
1422 if (test_and_set_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags)) {
1423 rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1428 rs->ti->error = "max_recovery_rate out of range";
1431 rs->md.sync_speed_max = value;
1432 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
1433 if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
1434 rs->ti->error = "Only one region_size argument pair allowed";
1438 region_size = value;
1439 rs->requested_bitmap_chunk_sectors = value;
1440 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
1441 if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
1442 rs->ti->error = "Only one raid10_copies argument pair allowed";
1446 if (!__within_range(value, 2, rs->md.raid_disks)) {
1447 rs->ti->error = "Bad value for 'raid10_copies'";
1451 raid10_copies = value;
1453 DMERR("Unable to parse RAID parameter: %s", key);
1454 rs->ti->error = "Unable to parse RAID parameter";
1459 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1460 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1461 rs->ti->error = "sync and nosync are mutually exclusive";
1465 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) &&
1466 (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ||
1467 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))) {
1468 rs->ti->error = "sync/nosync and rebuild are mutually exclusive";
1472 if (write_mostly >= rs->md.raid_disks) {
1473 rs->ti->error = "Can't set all raid1 devices to write_mostly";
1477 if (rs->md.sync_speed_max &&
1478 rs->md.sync_speed_min > rs->md.sync_speed_max) {
1479 rs->ti->error = "Bogus recovery rates";
1483 if (validate_region_size(rs, region_size))
1486 if (rs->md.chunk_sectors)
1487 max_io_len = rs->md.chunk_sectors;
1489 max_io_len = region_size;
1491 if (dm_set_target_max_io_len(rs->ti, max_io_len))
1494 if (rt_is_raid10(rt)) {
1495 if (raid10_copies > rs->md.raid_disks) {
1496 rs->ti->error = "Not enough devices to satisfy specification";
1500 rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
1501 if (rs->md.new_layout < 0) {
1502 rs->ti->error = "Error getting raid10 format";
1503 return rs->md.new_layout;
1506 rt = get_raid_type_by_ll(10, rs->md.new_layout);
1508 rs->ti->error = "Failed to recognize new raid10 layout";
1512 if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1513 rt->algorithm == ALGORITHM_RAID10_NEAR) &&
1514 test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1515 rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1520 rs->raid10_copies = raid10_copies;
1522 /* Assume there are no metadata devices until the drives are parsed */
1523 rs->md.persistent = 0;
1524 rs->md.external = 1;
1526 /* Check, if any invalid ctr arguments have been passed in for the raid level */
1527 return rs_check_for_valid_flags(rs);
1530 /* Set raid4/5/6 cache size */
1531 static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1534 struct r5conf *conf;
1535 struct mddev *mddev = &rs->md;
1536 uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1537 uint32_t nr_stripes = rs->stripe_cache_entries;
1539 if (!rt_is_raid456(rs->raid_type)) {
1540 rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1544 if (nr_stripes < min_stripes) {
1545 DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1546 nr_stripes, min_stripes);
1547 nr_stripes = min_stripes;
1550 conf = mddev->private;
1552 rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1556 /* Try setting number of stripes in raid456 stripe cache */
1557 if (conf->min_nr_stripes != nr_stripes) {
1558 r = raid5_set_cache_size(mddev, nr_stripes);
1560 rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1564 DMINFO("%u stripe cache entries", nr_stripes);
1570 /* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1571 static unsigned int mddev_data_stripes(struct raid_set *rs)
1573 return rs->md.raid_disks - rs->raid_type->parity_devs;
1576 /* Return # of data stripes of @rs (i.e. as of ctr) */
1577 static unsigned int rs_data_stripes(struct raid_set *rs)
1579 return rs->raid_disks - rs->raid_type->parity_devs;
1583 * Retrieve rdev->sectors from any valid raid device of @rs
1584 * to allow userpace to pass in arbitray "- -" device tupples.
1586 static sector_t __rdev_sectors(struct raid_set *rs)
1590 for (i = 0; i < rs->raid_disks; i++) {
1591 struct md_rdev *rdev = &rs->dev[i].rdev;
1593 if (!test_bit(Journal, &rdev->flags) &&
1594 rdev->bdev && rdev->sectors)
1595 return rdev->sectors;
1601 /* Check that calculated dev_sectors fits all component devices. */
1602 static int _check_data_dev_sectors(struct raid_set *rs)
1605 struct md_rdev *rdev;
1607 rdev_for_each(rdev, &rs->md)
1608 if (!test_bit(Journal, &rdev->flags) && rdev->bdev) {
1609 ds = min(ds, to_sector(i_size_read(rdev->bdev->bd_inode)));
1610 if (ds < rs->md.dev_sectors) {
1611 rs->ti->error = "Component device(s) too small";
1619 /* Calculate the sectors per device and per array used for @rs */
1620 static int rs_set_dev_and_array_sectors(struct raid_set *rs, bool use_mddev)
1623 unsigned int data_stripes;
1624 struct mddev *mddev = &rs->md;
1625 struct md_rdev *rdev;
1626 sector_t array_sectors = rs->ti->len, dev_sectors = rs->ti->len;
1629 delta_disks = mddev->delta_disks;
1630 data_stripes = mddev_data_stripes(rs);
1632 delta_disks = rs->delta_disks;
1633 data_stripes = rs_data_stripes(rs);
1636 /* Special raid1 case w/o delta_disks support (yet) */
1637 if (rt_is_raid1(rs->raid_type))
1639 else if (rt_is_raid10(rs->raid_type)) {
1640 if (rs->raid10_copies < 2 ||
1642 rs->ti->error = "Bogus raid10 data copies or delta disks";
1646 dev_sectors *= rs->raid10_copies;
1647 if (sector_div(dev_sectors, data_stripes))
1650 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1651 if (sector_div(array_sectors, rs->raid10_copies))
1654 } else if (sector_div(dev_sectors, data_stripes))
1658 /* Striped layouts */
1659 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1661 rdev_for_each(rdev, mddev)
1662 if (!test_bit(Journal, &rdev->flags))
1663 rdev->sectors = dev_sectors;
1665 mddev->array_sectors = array_sectors;
1666 mddev->dev_sectors = dev_sectors;
1668 return _check_data_dev_sectors(rs);
1670 rs->ti->error = "Target length not divisible by number of data devices";
1674 /* Setup recovery on @rs */
1675 static void __rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1677 /* raid0 does not recover */
1678 if (rs_is_raid0(rs))
1679 rs->md.recovery_cp = MaxSector;
1681 * A raid6 set has to be recovered either
1682 * completely or for the grown part to
1683 * ensure proper parity and Q-Syndrome
1685 else if (rs_is_raid6(rs))
1686 rs->md.recovery_cp = dev_sectors;
1688 * Other raid set types may skip recovery
1689 * depending on the 'nosync' flag.
1692 rs->md.recovery_cp = test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)
1693 ? MaxSector : dev_sectors;
1696 /* Setup recovery on @rs based on raid type, device size and 'nosync' flag */
1697 static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1700 /* New raid set or 'sync' flag provided */
1701 __rs_setup_recovery(rs, 0);
1702 else if (dev_sectors == MaxSector)
1703 /* Prevent recovery */
1704 __rs_setup_recovery(rs, MaxSector);
1705 else if (__rdev_sectors(rs) < dev_sectors)
1706 /* Grown raid set */
1707 __rs_setup_recovery(rs, __rdev_sectors(rs));
1709 __rs_setup_recovery(rs, MaxSector);
1712 static void do_table_event(struct work_struct *ws)
1714 struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1716 smp_rmb(); /* Make sure we access most actual mddev properties */
1717 if (!rs_is_reshaping(rs)) {
1718 if (rs_is_raid10(rs))
1719 rs_set_rdev_sectors(rs);
1720 rs_set_capacity(rs);
1722 dm_table_event(rs->ti->table);
1725 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
1727 struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
1729 return mddev_congested(&rs->md, bits);
1733 * Make sure a valid takover (level switch) is being requested on @rs
1735 * Conversions of raid sets from one MD personality to another
1736 * have to conform to restrictions which are enforced here.
1738 static int rs_check_takeover(struct raid_set *rs)
1740 struct mddev *mddev = &rs->md;
1741 unsigned int near_copies;
1743 if (rs->md.degraded) {
1744 rs->ti->error = "Can't takeover degraded raid set";
1748 if (rs_is_reshaping(rs)) {
1749 rs->ti->error = "Can't takeover reshaping raid set";
1753 switch (mddev->level) {
1755 /* raid0 -> raid1/5 with one disk */
1756 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1757 mddev->raid_disks == 1)
1760 /* raid0 -> raid10 */
1761 if (mddev->new_level == 10 &&
1762 !(rs->raid_disks % mddev->raid_disks))
1765 /* raid0 with multiple disks -> raid4/5/6 */
1766 if (__within_range(mddev->new_level, 4, 6) &&
1767 mddev->new_layout == ALGORITHM_PARITY_N &&
1768 mddev->raid_disks > 1)
1774 /* Can't takeover raid10_offset! */
1775 if (__is_raid10_offset(mddev->layout))
1778 near_copies = __raid10_near_copies(mddev->layout);
1780 /* raid10* -> raid0 */
1781 if (mddev->new_level == 0) {
1782 /* Can takeover raid10_near with raid disks divisable by data copies! */
1783 if (near_copies > 1 &&
1784 !(mddev->raid_disks % near_copies)) {
1785 mddev->raid_disks /= near_copies;
1786 mddev->delta_disks = mddev->raid_disks;
1790 /* Can takeover raid10_far */
1791 if (near_copies == 1 &&
1792 __raid10_far_copies(mddev->layout) > 1)
1798 /* raid10_{near,far} -> raid1 */
1799 if (mddev->new_level == 1 &&
1800 max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
1803 /* raid10_{near,far} with 2 disks -> raid4/5 */
1804 if (__within_range(mddev->new_level, 4, 5) &&
1805 mddev->raid_disks == 2)
1810 /* raid1 with 2 disks -> raid4/5 */
1811 if (__within_range(mddev->new_level, 4, 5) &&
1812 mddev->raid_disks == 2) {
1813 mddev->degraded = 1;
1817 /* raid1 -> raid0 */
1818 if (mddev->new_level == 0 &&
1819 mddev->raid_disks == 1)
1822 /* raid1 -> raid10 */
1823 if (mddev->new_level == 10)
1828 /* raid4 -> raid0 */
1829 if (mddev->new_level == 0)
1832 /* raid4 -> raid1/5 with 2 disks */
1833 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1834 mddev->raid_disks == 2)
1837 /* raid4 -> raid5/6 with parity N */
1838 if (__within_range(mddev->new_level, 5, 6) &&
1839 mddev->layout == ALGORITHM_PARITY_N)
1844 /* raid5 with parity N -> raid0 */
1845 if (mddev->new_level == 0 &&
1846 mddev->layout == ALGORITHM_PARITY_N)
1849 /* raid5 with parity N -> raid4 */
1850 if (mddev->new_level == 4 &&
1851 mddev->layout == ALGORITHM_PARITY_N)
1854 /* raid5 with 2 disks -> raid1/4/10 */
1855 if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1856 mddev->raid_disks == 2)
1859 /* raid5_* -> raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
1860 if (mddev->new_level == 6 &&
1861 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1862 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
1867 /* raid6 with parity N -> raid0 */
1868 if (mddev->new_level == 0 &&
1869 mddev->layout == ALGORITHM_PARITY_N)
1872 /* raid6 with parity N -> raid4 */
1873 if (mddev->new_level == 4 &&
1874 mddev->layout == ALGORITHM_PARITY_N)
1877 /* raid6_*_n with Q-Syndrome N -> raid5_* */
1878 if (mddev->new_level == 5 &&
1879 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1880 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
1887 rs->ti->error = "takeover not possible";
1891 /* True if @rs requested to be taken over */
1892 static bool rs_takeover_requested(struct raid_set *rs)
1894 return rs->md.new_level != rs->md.level;
1897 /* True if layout is set to reshape. */
1898 static bool rs_is_layout_change(struct raid_set *rs, bool use_mddev)
1900 return (use_mddev ? rs->md.delta_disks : rs->delta_disks) ||
1901 rs->md.new_layout != rs->md.layout ||
1902 rs->md.new_chunk_sectors != rs->md.chunk_sectors;
1905 /* True if @rs is requested to reshape by ctr */
1906 static bool rs_reshape_requested(struct raid_set *rs)
1909 struct mddev *mddev = &rs->md;
1911 if (rs_takeover_requested(rs))
1914 if (rs_is_raid0(rs))
1917 change = rs_is_layout_change(rs, false);
1919 /* Historical case to support raid1 reshape without delta disks */
1920 if (rs_is_raid1(rs)) {
1921 if (rs->delta_disks)
1922 return !!rs->delta_disks;
1925 mddev->raid_disks != rs->raid_disks;
1928 if (rs_is_raid10(rs))
1930 !__is_raid10_far(mddev->new_layout) &&
1931 rs->delta_disks >= 0;
1937 #define FEATURE_FLAG_SUPPORTS_V190 0x1 /* Supports extended superblock */
1939 /* State flags for sb->flags */
1940 #define SB_FLAG_RESHAPE_ACTIVE 0x1
1941 #define SB_FLAG_RESHAPE_BACKWARDS 0x2
1944 * This structure is never routinely used by userspace, unlike md superblocks.
1945 * Devices with this superblock should only ever be accessed via device-mapper.
1947 #define DM_RAID_MAGIC 0x64526D44
1948 struct dm_raid_superblock {
1949 __le32 magic; /* "DmRd" */
1950 __le32 compat_features; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1952 __le32 num_devices; /* Number of devices in this raid set. (Max 64) */
1953 __le32 array_position; /* The position of this drive in the raid set */
1955 __le64 events; /* Incremented by md when superblock updated */
1956 __le64 failed_devices; /* Pre 1.9.0 part of bit field of devices to */
1957 /* indicate failures (see extension below) */
1960 * This offset tracks the progress of the repair or replacement of
1961 * an individual drive.
1963 __le64 disk_recovery_offset;
1966 * This offset tracks the progress of the initial raid set
1967 * synchronisation/parity calculation.
1969 __le64 array_resync_offset;
1972 * raid characteristics
1976 __le32 stripe_sectors;
1978 /********************************************************************
1979 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1981 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
1984 __le32 flags; /* Flags defining array states for reshaping */
1987 * This offset tracks the progress of a raid
1988 * set reshape in order to be able to restart it
1990 __le64 reshape_position;
1993 * These define the properties of the array in case of an interrupted reshape
1997 __le32 new_stripe_sectors;
2000 __le64 array_sectors; /* Array size in sectors */
2003 * Sector offsets to data on devices (reshaping).
2004 * Needed to support out of place reshaping, thus
2005 * not writing over any stripes whilst converting
2006 * them from old to new layout
2009 __le64 new_data_offset;
2011 __le64 sectors; /* Used device size in sectors */
2014 * Additonal Bit field of devices indicating failures to support
2015 * up to 256 devices with the 1.9.0 on-disk metadata format
2017 __le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
2019 __le32 incompat_features; /* Used to indicate any incompatible features */
2021 /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
2025 * Check for reshape constraints on raid set @rs:
2027 * - reshape function non-existent
2029 * - ongoing recovery
2032 * Returns 0 if none or -EPERM if given constraint
2033 * and error message reference in @errmsg
2035 static int rs_check_reshape(struct raid_set *rs)
2037 struct mddev *mddev = &rs->md;
2039 if (!mddev->pers || !mddev->pers->check_reshape)
2040 rs->ti->error = "Reshape not supported";
2041 else if (mddev->degraded)
2042 rs->ti->error = "Can't reshape degraded raid set";
2043 else if (rs_is_recovering(rs))
2044 rs->ti->error = "Convert request on recovering raid set prohibited";
2045 else if (rs_is_reshaping(rs))
2046 rs->ti->error = "raid set already reshaping!";
2047 else if (!(rs_is_raid1(rs) || rs_is_raid10(rs) || rs_is_raid456(rs)))
2048 rs->ti->error = "Reshaping only supported for raid1/4/5/6/10";
2055 static int read_disk_sb(struct md_rdev *rdev, int size, bool force_reload)
2057 BUG_ON(!rdev->sb_page);
2059 if (rdev->sb_loaded && !force_reload)
2062 rdev->sb_loaded = 0;
2064 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) {
2065 DMERR("Failed to read superblock of device at position %d",
2067 md_error(rdev->mddev, rdev);
2068 set_bit(Faulty, &rdev->flags);
2072 rdev->sb_loaded = 1;
2077 static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2079 failed_devices[0] = le64_to_cpu(sb->failed_devices);
2080 memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
2082 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2083 int i = ARRAY_SIZE(sb->extended_failed_devices);
2086 failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
2090 static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2092 int i = ARRAY_SIZE(sb->extended_failed_devices);
2094 sb->failed_devices = cpu_to_le64(failed_devices[0]);
2096 sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
2100 * Synchronize the superblock members with the raid set properties
2102 * All superblock data is little endian.
2104 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
2106 bool update_failed_devices = false;
2108 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2109 struct dm_raid_superblock *sb;
2110 struct raid_set *rs = container_of(mddev, struct raid_set, md);
2112 /* No metadata device, no superblock */
2113 if (!rdev->meta_bdev)
2116 BUG_ON(!rdev->sb_page);
2118 sb = page_address(rdev->sb_page);
2120 sb_retrieve_failed_devices(sb, failed_devices);
2122 for (i = 0; i < rs->raid_disks; i++)
2123 if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
2124 update_failed_devices = true;
2125 set_bit(i, (void *) failed_devices);
2128 if (update_failed_devices)
2129 sb_update_failed_devices(sb, failed_devices);
2131 sb->magic = cpu_to_le32(DM_RAID_MAGIC);
2132 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2134 sb->num_devices = cpu_to_le32(mddev->raid_disks);
2135 sb->array_position = cpu_to_le32(rdev->raid_disk);
2137 sb->events = cpu_to_le64(mddev->events);
2139 sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
2140 sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
2142 sb->level = cpu_to_le32(mddev->level);
2143 sb->layout = cpu_to_le32(mddev->layout);
2144 sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
2146 /********************************************************************
2147 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
2149 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
2151 sb->new_level = cpu_to_le32(mddev->new_level);
2152 sb->new_layout = cpu_to_le32(mddev->new_layout);
2153 sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
2155 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2157 smp_rmb(); /* Make sure we access most recent reshape position */
2158 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2159 if (le64_to_cpu(sb->reshape_position) != MaxSector) {
2160 /* Flag ongoing reshape */
2161 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
2163 if (mddev->delta_disks < 0 || mddev->reshape_backwards)
2164 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
2166 /* Clear reshape flags */
2167 sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
2170 sb->array_sectors = cpu_to_le64(mddev->array_sectors);
2171 sb->data_offset = cpu_to_le64(rdev->data_offset);
2172 sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
2173 sb->sectors = cpu_to_le64(rdev->sectors);
2174 sb->incompat_features = cpu_to_le32(0);
2176 /* Zero out the rest of the payload after the size of the superblock */
2177 memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
2183 * This function creates a superblock if one is not found on the device
2184 * and will decide which superblock to use if there's a choice.
2186 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
2188 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
2191 struct dm_raid_superblock *sb;
2192 struct dm_raid_superblock *refsb;
2193 uint64_t events_sb, events_refsb;
2195 r = read_disk_sb(rdev, rdev->sb_size, false);
2199 sb = page_address(rdev->sb_page);
2202 * Two cases that we want to write new superblocks and rebuild:
2203 * 1) New device (no matching magic number)
2204 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2206 if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
2207 (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
2208 super_sync(rdev->mddev, rdev);
2210 set_bit(FirstUse, &rdev->flags);
2211 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2213 /* Force writing of superblocks to disk */
2214 set_bit(MD_SB_CHANGE_DEVS, &rdev->mddev->sb_flags);
2216 /* Any superblock is better than none, choose that if given */
2217 return refdev ? 0 : 1;
2223 events_sb = le64_to_cpu(sb->events);
2225 refsb = page_address(refdev->sb_page);
2226 events_refsb = le64_to_cpu(refsb->events);
2228 return (events_sb > events_refsb) ? 1 : 0;
2231 static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
2235 struct mddev *mddev = &rs->md;
2237 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2238 struct dm_raid_superblock *sb;
2239 uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
2241 struct dm_raid_superblock *sb2;
2243 sb = page_address(rdev->sb_page);
2244 events_sb = le64_to_cpu(sb->events);
2247 * Initialise to 1 if this is a new superblock.
2249 mddev->events = events_sb ? : 1;
2251 mddev->reshape_position = MaxSector;
2253 mddev->raid_disks = le32_to_cpu(sb->num_devices);
2254 mddev->level = le32_to_cpu(sb->level);
2255 mddev->layout = le32_to_cpu(sb->layout);
2256 mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
2259 * Reshaping is supported, e.g. reshape_position is valid
2260 * in superblock and superblock content is authoritative.
2262 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2263 /* Superblock is authoritative wrt given raid set layout! */
2264 mddev->new_level = le32_to_cpu(sb->new_level);
2265 mddev->new_layout = le32_to_cpu(sb->new_layout);
2266 mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
2267 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
2268 mddev->array_sectors = le64_to_cpu(sb->array_sectors);
2270 /* raid was reshaping and got interrupted */
2271 if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
2272 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
2273 DMERR("Reshape requested but raid set is still reshaping");
2277 if (mddev->delta_disks < 0 ||
2278 (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
2279 mddev->reshape_backwards = 1;
2281 mddev->reshape_backwards = 0;
2283 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2284 rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2289 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2291 struct raid_type *rt_cur = get_raid_type_by_ll(mddev->level, mddev->layout);
2292 struct raid_type *rt_new = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
2294 if (rs_takeover_requested(rs)) {
2295 if (rt_cur && rt_new)
2296 DMERR("Takeover raid sets from %s to %s not yet supported by metadata. (raid level change)",
2297 rt_cur->name, rt_new->name);
2299 DMERR("Takeover raid sets not yet supported by metadata. (raid level change)");
2301 } else if (rs_reshape_requested(rs)) {
2302 DMERR("Reshaping raid sets not yet supported by metadata. (raid layout change keeping level)");
2303 if (mddev->layout != mddev->new_layout) {
2304 if (rt_cur && rt_new)
2305 DMERR(" current layout %s vs new layout %s",
2306 rt_cur->name, rt_new->name);
2308 DMERR(" current layout 0x%X vs new layout 0x%X",
2309 le32_to_cpu(sb->layout), mddev->new_layout);
2311 if (mddev->chunk_sectors != mddev->new_chunk_sectors)
2312 DMERR(" current stripe sectors %u vs new stripe sectors %u",
2313 mddev->chunk_sectors, mddev->new_chunk_sectors);
2314 if (rs->delta_disks)
2315 DMERR(" current %u disks vs new %u disks",
2316 mddev->raid_disks, mddev->raid_disks + rs->delta_disks);
2317 if (rs_is_raid10(rs)) {
2318 DMERR(" Old layout: %s w/ %u copies",
2319 raid10_md_layout_to_format(mddev->layout),
2320 raid10_md_layout_to_copies(mddev->layout));
2321 DMERR(" New layout: %s w/ %u copies",
2322 raid10_md_layout_to_format(mddev->new_layout),
2323 raid10_md_layout_to_copies(mddev->new_layout));
2328 DMINFO("Discovered old metadata format; upgrading to extended metadata format");
2331 if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
2332 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2335 * During load, we set FirstUse if a new superblock was written.
2336 * There are two reasons we might not have a superblock:
2337 * 1) The raid set is brand new - in which case, all of the
2338 * devices must have their In_sync bit set. Also,
2339 * recovery_cp must be 0, unless forced.
2340 * 2) This is a new device being added to an old raid set
2341 * and the new device needs to be rebuilt - in which
2342 * case the In_sync bit will /not/ be set and
2343 * recovery_cp must be MaxSector.
2344 * 3) This is/are a new device(s) being added to an old
2345 * raid set during takeover to a higher raid level
2346 * to provide capacity for redundancy or during reshape
2347 * to add capacity to grow the raid set.
2350 rdev_for_each(r, mddev) {
2351 if (test_bit(Journal, &rdev->flags))
2354 if (test_bit(FirstUse, &r->flags))
2357 if (!test_bit(In_sync, &r->flags)) {
2358 DMINFO("Device %d specified for rebuild; clearing superblock",
2362 if (test_bit(FirstUse, &r->flags))
2369 if (new_devs == rs->raid_disks || !rebuilds) {
2370 /* Replace a broken device */
2371 if (new_devs == 1 && !rs->delta_disks)
2373 if (new_devs == rs->raid_disks) {
2374 DMINFO("Superblocks created for new raid set");
2375 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2376 } else if (new_devs != rebuilds &&
2377 new_devs != rs->delta_disks) {
2378 DMERR("New device injected into existing raid set without "
2379 "'delta_disks' or 'rebuild' parameter specified");
2382 } else if (new_devs && new_devs != rebuilds) {
2383 DMERR("%u 'rebuild' devices cannot be injected into"
2384 " a raid set with %u other first-time devices",
2385 rebuilds, new_devs);
2387 } else if (rebuilds) {
2388 if (rebuild_and_new && rebuilds != rebuild_and_new) {
2389 DMERR("new device%s provided without 'rebuild'",
2390 new_devs > 1 ? "s" : "");
2392 } else if (!test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) && rs_is_recovering(rs)) {
2393 DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2394 (unsigned long long) mddev->recovery_cp);
2396 } else if (rs_is_reshaping(rs)) {
2397 DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2398 (unsigned long long) mddev->reshape_position);
2404 * Now we set the Faulty bit for those devices that are
2405 * recorded in the superblock as failed.
2407 sb_retrieve_failed_devices(sb, failed_devices);
2408 rdev_for_each(r, mddev) {
2409 if (test_bit(Journal, &rdev->flags) ||
2412 sb2 = page_address(r->sb_page);
2413 sb2->failed_devices = 0;
2414 memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
2417 * Check for any device re-ordering.
2419 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2420 role = le32_to_cpu(sb2->array_position);
2424 if (role != r->raid_disk) {
2425 if (rs_is_raid10(rs) && __is_raid10_near(mddev->layout)) {
2426 if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
2427 rs->raid_disks % rs->raid10_copies) {
2429 "Cannot change raid10 near set to odd # of devices!";
2433 sb2->array_position = cpu_to_le32(r->raid_disk);
2435 } else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
2436 !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2437 !rt_is_raid1(rs->raid_type)) {
2438 rs->ti->error = "Cannot change device positions in raid set";
2442 DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
2446 * Partial recovery is performed on
2447 * returning failed devices.
2449 if (test_bit(role, (void *) failed_devices))
2450 set_bit(Faulty, &r->flags);
2457 static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
2459 struct mddev *mddev = &rs->md;
2460 struct dm_raid_superblock *sb;
2462 if (rs_is_raid0(rs) || !rdev->sb_page || rdev->raid_disk < 0)
2465 sb = page_address(rdev->sb_page);
2468 * If mddev->events is not set, we know we have not yet initialized
2471 if (!mddev->events && super_init_validation(rs, rdev))
2474 if (le32_to_cpu(sb->compat_features) &&
2475 le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
2476 rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2480 if (sb->incompat_features) {
2481 rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
2485 /* Enable bitmap creation for RAID levels != 0 */
2486 mddev->bitmap_info.offset = (rt_is_raid0(rs->raid_type) || rs->journal_dev.dev) ? 0 : to_sector(4096);
2487 mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2489 if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2491 * Retrieve rdev size stored in superblock to be prepared for shrink.
2492 * Check extended superblock members are present otherwise the size
2495 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190)
2496 rdev->sectors = le64_to_cpu(sb->sectors);
2498 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
2499 if (rdev->recovery_offset == MaxSector)
2500 set_bit(In_sync, &rdev->flags);
2502 * If no reshape in progress -> we're recovering single
2503 * disk(s) and have to set the device(s) to out-of-sync
2505 else if (!rs_is_reshaping(rs))
2506 clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
2510 * If a device comes back, set it as not In_sync and no longer faulty.
2512 if (test_and_clear_bit(Faulty, &rdev->flags)) {
2513 rdev->recovery_offset = 0;
2514 clear_bit(In_sync, &rdev->flags);
2515 rdev->saved_raid_disk = rdev->raid_disk;
2518 /* Reshape support -> restore repective data offsets */
2519 rdev->data_offset = le64_to_cpu(sb->data_offset);
2520 rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
2526 * Analyse superblocks and select the freshest.
2528 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2531 struct md_rdev *rdev, *freshest;
2532 struct mddev *mddev = &rs->md;
2535 rdev_for_each(rdev, mddev) {
2536 if (test_bit(Journal, &rdev->flags))
2539 if (!rdev->meta_bdev)
2542 /* Set superblock offset/size for metadata device. */
2544 rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
2545 if (rdev->sb_size < sizeof(struct dm_raid_superblock) || rdev->sb_size > PAGE_SIZE) {
2546 DMERR("superblock size of a logical block is no longer valid");
2551 * Skipping super_load due to CTR_FLAG_SYNC will cause
2552 * the array to undergo initialization again as
2553 * though it were new. This is the intended effect
2554 * of the "sync" directive.
2556 * With reshaping capability added, we must ensure that
2557 * that the "sync" directive is disallowed during the reshape.
2559 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2562 r = super_load(rdev, freshest);
2571 /* This is a failure to read the superblock from the metadata device. */
2573 * We have to keep any raid0 data/metadata device pairs or
2574 * the MD raid0 personality will fail to start the array.
2576 if (rs_is_raid0(rs))
2580 * We keep the dm_devs to be able to emit the device tuple
2581 * properly on the table line in raid_status() (rather than
2582 * mistakenly acting as if '- -' got passed into the constructor).
2584 * The rdev has to stay on the same_set list to allow for
2585 * the attempt to restore faulty devices on second resume.
2587 rdev->raid_disk = rdev->saved_raid_disk = -1;
2596 * Validation of the freshest device provides the source of
2597 * validation for the remaining devices.
2599 rs->ti->error = "Unable to assemble array: Invalid superblocks";
2600 if (super_validate(rs, freshest))
2603 if (validate_raid_redundancy(rs)) {
2604 rs->ti->error = "Insufficient redundancy to activate array";
2608 rdev_for_each(rdev, mddev)
2609 if (!test_bit(Journal, &rdev->flags) &&
2611 super_validate(rs, rdev))
2617 * Adjust data_offset and new_data_offset on all disk members of @rs
2618 * for out of place reshaping if requested by contructor
2620 * We need free space at the beginning of each raid disk for forward
2621 * and at the end for backward reshapes which userspace has to provide
2622 * via remapping/reordering of space.
2624 static int rs_adjust_data_offsets(struct raid_set *rs)
2626 sector_t data_offset = 0, new_data_offset = 0;
2627 struct md_rdev *rdev;
2629 /* Constructor did not request data offset change */
2630 if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2631 if (!rs_is_reshapable(rs))
2637 /* HM FIXME: get In_Sync raid_dev? */
2638 rdev = &rs->dev[0].rdev;
2640 if (rs->delta_disks < 0) {
2642 * Removing disks (reshaping backwards):
2644 * - before reshape: data is at offset 0 and free space
2645 * is at end of each component LV
2647 * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2650 new_data_offset = rs->data_offset;
2652 } else if (rs->delta_disks > 0) {
2654 * Adding disks (reshaping forwards):
2656 * - before reshape: data is at offset rs->data_offset != 0 and
2657 * free space is at begin of each component LV
2659 * - after reshape: data is at offset 0 on each component LV
2661 data_offset = rs->data_offset;
2662 new_data_offset = 0;
2666 * User space passes in 0 for data offset after having removed reshape space
2668 * - or - (data offset != 0)
2670 * Changing RAID layout or chunk size -> toggle offsets
2672 * - before reshape: data is at offset rs->data_offset 0 and
2673 * free space is at end of each component LV
2675 * data is at offset rs->data_offset != 0 and
2676 * free space is at begin of each component LV
2678 * - after reshape: data is at offset 0 if it was at offset != 0
2679 * or at offset != 0 if it was at offset 0
2680 * on each component LV
2683 data_offset = rs->data_offset ? rdev->data_offset : 0;
2684 new_data_offset = data_offset ? 0 : rs->data_offset;
2685 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2689 * Make sure we got a minimum amount of free sectors per device
2691 if (rs->data_offset &&
2692 to_sector(i_size_read(rdev->bdev->bd_inode)) - rs->md.dev_sectors < MIN_FREE_RESHAPE_SPACE) {
2693 rs->ti->error = data_offset ? "No space for forward reshape" :
2694 "No space for backward reshape";
2699 * Raise recovery_cp in case data_offset != 0 to
2700 * avoid false recovery positives in the constructor.
2702 if (rs->md.recovery_cp < rs->md.dev_sectors)
2703 rs->md.recovery_cp += rs->dev[0].rdev.data_offset;
2705 /* Adjust data offsets on all rdevs but on any raid4/5/6 journal device */
2706 rdev_for_each(rdev, &rs->md) {
2707 if (!test_bit(Journal, &rdev->flags)) {
2708 rdev->data_offset = data_offset;
2709 rdev->new_data_offset = new_data_offset;
2716 /* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2717 static void __reorder_raid_disk_indexes(struct raid_set *rs)
2720 struct md_rdev *rdev;
2722 rdev_for_each(rdev, &rs->md) {
2723 if (!test_bit(Journal, &rdev->flags)) {
2724 rdev->raid_disk = i++;
2725 rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2731 * Setup @rs for takeover by a different raid level
2733 static int rs_setup_takeover(struct raid_set *rs)
2735 struct mddev *mddev = &rs->md;
2736 struct md_rdev *rdev;
2737 unsigned int d = mddev->raid_disks = rs->raid_disks;
2738 sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2740 if (rt_is_raid10(rs->raid_type)) {
2741 if (rs_is_raid0(rs)) {
2742 /* Userpace reordered disks -> adjust raid_disk indexes */
2743 __reorder_raid_disk_indexes(rs);
2745 /* raid0 -> raid10_far layout */
2746 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2748 } else if (rs_is_raid1(rs))
2749 /* raid1 -> raid10_near layout */
2750 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2757 clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2758 mddev->recovery_cp = MaxSector;
2761 rdev = &rs->dev[d].rdev;
2763 if (test_bit(d, (void *) rs->rebuild_disks)) {
2764 clear_bit(In_sync, &rdev->flags);
2765 clear_bit(Faulty, &rdev->flags);
2766 mddev->recovery_cp = rdev->recovery_offset = 0;
2767 /* Bitmap has to be created when we do an "up" takeover */
2768 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2771 rdev->new_data_offset = new_data_offset;
2777 /* Prepare @rs for reshape */
2778 static int rs_prepare_reshape(struct raid_set *rs)
2781 struct mddev *mddev = &rs->md;
2783 if (rs_is_raid10(rs)) {
2784 if (rs->raid_disks != mddev->raid_disks &&
2785 __is_raid10_near(mddev->layout) &&
2786 rs->raid10_copies &&
2787 rs->raid10_copies != __raid10_near_copies(mddev->layout)) {
2789 * raid disk have to be multiple of data copies to allow this conversion,
2791 * This is actually not a reshape it is a
2792 * rebuild of any additional mirrors per group
2794 if (rs->raid_disks % rs->raid10_copies) {
2795 rs->ti->error = "Can't reshape raid10 mirror groups";
2799 /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2800 __reorder_raid_disk_indexes(rs);
2801 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2803 mddev->new_layout = mddev->layout;
2808 } else if (rs_is_raid456(rs))
2811 else if (rs_is_raid1(rs)) {
2812 if (rs->delta_disks) {
2813 /* Process raid1 via delta_disks */
2814 mddev->degraded = rs->delta_disks < 0 ? -rs->delta_disks : rs->delta_disks;
2817 /* Process raid1 without delta_disks */
2818 mddev->raid_disks = rs->raid_disks;
2822 rs->ti->error = "Called with bogus raid type";
2827 set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2828 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2829 } else if (mddev->raid_disks < rs->raid_disks)
2830 /* Create new superblocks and bitmaps, if any new disks */
2831 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2836 /* Get reshape sectors from data_offsets or raid set */
2837 static sector_t _get_reshape_sectors(struct raid_set *rs)
2839 struct md_rdev *rdev;
2840 sector_t reshape_sectors = 0;
2842 rdev_for_each(rdev, &rs->md)
2843 if (!test_bit(Journal, &rdev->flags)) {
2844 reshape_sectors = (rdev->data_offset > rdev->new_data_offset) ?
2845 rdev->data_offset - rdev->new_data_offset :
2846 rdev->new_data_offset - rdev->data_offset;
2850 return max(reshape_sectors, (sector_t) rs->data_offset);
2855 * - change raid layout
2856 * - change chunk size
2860 static int rs_setup_reshape(struct raid_set *rs)
2863 unsigned int cur_raid_devs, d;
2864 sector_t reshape_sectors = _get_reshape_sectors(rs);
2865 struct mddev *mddev = &rs->md;
2866 struct md_rdev *rdev;
2868 mddev->delta_disks = rs->delta_disks;
2869 cur_raid_devs = mddev->raid_disks;
2871 /* Ignore impossible layout change whilst adding/removing disks */
2872 if (mddev->delta_disks &&
2873 mddev->layout != mddev->new_layout) {
2874 DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2875 mddev->new_layout = mddev->layout;
2879 * Adjust array size:
2881 * - in case of adding disk(s), array size has
2882 * to grow after the disk adding reshape,
2883 * which'll hapen in the event handler;
2884 * reshape will happen forward, so space has to
2885 * be available at the beginning of each disk
2887 * - in case of removing disk(s), array size
2888 * has to shrink before starting the reshape,
2889 * which'll happen here;
2890 * reshape will happen backward, so space has to
2891 * be available at the end of each disk
2893 * - data_offset and new_data_offset are
2894 * adjusted for aforementioned out of place
2895 * reshaping based on userspace passing in
2896 * the "data_offset <sectors>" key/value
2897 * pair via the constructor
2901 if (rs->delta_disks > 0) {
2902 /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2903 for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2904 rdev = &rs->dev[d].rdev;
2905 clear_bit(In_sync, &rdev->flags);
2908 * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2909 * by md, which'll store that erroneously in the superblock on reshape
2911 rdev->saved_raid_disk = -1;
2912 rdev->raid_disk = d;
2914 rdev->sectors = mddev->dev_sectors;
2915 rdev->recovery_offset = rs_is_raid1(rs) ? 0 : MaxSector;
2918 mddev->reshape_backwards = 0; /* adding disk(s) -> forward reshape */
2920 /* Remove disk(s) */
2921 } else if (rs->delta_disks < 0) {
2922 r = rs_set_dev_and_array_sectors(rs, true);
2923 mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2925 /* Change layout and/or chunk size */
2928 * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2930 * keeping number of disks and do layout change ->
2932 * toggle reshape_backward depending on data_offset:
2934 * - free space upfront -> reshape forward
2936 * - free space at the end -> reshape backward
2939 * This utilizes free reshape space avoiding the need
2940 * for userspace to move (parts of) LV segments in
2941 * case of layout/chunksize change (for disk
2942 * adding/removing reshape space has to be at
2943 * the proper address (see above with delta_disks):
2945 * add disk(s) -> begin
2946 * remove disk(s)-> end
2948 mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2952 * Adjust device size for forward reshape
2953 * because md_finish_reshape() reduces it.
2955 if (!mddev->reshape_backwards)
2956 rdev_for_each(rdev, &rs->md)
2957 if (!test_bit(Journal, &rdev->flags))
2958 rdev->sectors += reshape_sectors;
2964 * If the md resync thread has updated superblock with max reshape position
2965 * at the end of a reshape but not (yet) reset the layout configuration
2966 * changes -> reset the latter.
2968 static void rs_reset_inconclusive_reshape(struct raid_set *rs)
2970 if (!rs_is_reshaping(rs) && rs_is_layout_change(rs, true)) {
2972 rs->md.delta_disks = 0;
2973 rs->md.reshape_backwards = 0;
2978 * Enable/disable discard support on RAID set depending on
2979 * RAID level and discard properties of underlying RAID members.
2981 static void configure_discard_support(struct raid_set *rs)
2985 struct dm_target *ti = rs->ti;
2988 * XXX: RAID level 4,5,6 require zeroing for safety.
2990 raid456 = rs_is_raid456(rs);
2992 for (i = 0; i < rs->raid_disks; i++) {
2993 struct request_queue *q;
2995 if (!rs->dev[i].rdev.bdev)
2998 q = bdev_get_queue(rs->dev[i].rdev.bdev);
2999 if (!q || !blk_queue_discard(q))
3003 if (!devices_handle_discard_safely) {
3004 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
3005 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
3012 * RAID1 and RAID10 personalities require bio splitting,
3013 * RAID0/4/5/6 don't and process large discard bios properly.
3015 ti->split_discard_bios = !!(rs_is_raid1(rs) || rs_is_raid10(rs));
3016 ti->num_discard_bios = 1;
3020 * Construct a RAID0/1/10/4/5/6 mapping:
3022 * <raid_type> <#raid_params> <raid_params>{0,} \
3023 * <#raid_devs> [<meta_dev1> <dev1>]{1,}
3025 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
3026 * details on possible <raid_params>.
3028 * Userspace is free to initialize the metadata devices, hence the superblocks to
3029 * enforce recreation based on the passed in table parameters.
3032 static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3035 bool resize = false;
3036 struct raid_type *rt;
3037 unsigned int num_raid_params, num_raid_devs;
3038 sector_t calculated_dev_sectors, rdev_sectors, reshape_sectors;
3039 struct raid_set *rs = NULL;
3041 struct rs_layout rs_layout;
3042 struct dm_arg_set as = { argc, argv }, as_nrd;
3043 struct dm_arg _args[] = {
3044 { 0, as.argc, "Cannot understand number of raid parameters" },
3045 { 1, 254, "Cannot understand number of raid devices parameters" }
3048 /* Must have <raid_type> */
3049 arg = dm_shift_arg(&as);
3051 ti->error = "No arguments";
3055 rt = get_raid_type(arg);
3057 ti->error = "Unrecognised raid_type";
3061 /* Must have <#raid_params> */
3062 if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
3065 /* number of raid device tupples <meta_dev data_dev> */
3067 dm_consume_args(&as_nrd, num_raid_params);
3068 _args[1].max = (as_nrd.argc - 1) / 2;
3069 if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
3072 if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
3073 ti->error = "Invalid number of supplied raid devices";
3077 rs = raid_set_alloc(ti, rt, num_raid_devs);
3081 r = parse_raid_params(rs, &as, num_raid_params);
3085 r = parse_dev_params(rs, &as);
3089 rs->md.sync_super = super_sync;
3092 * Calculate ctr requested array and device sizes to allow
3093 * for superblock analysis needing device sizes defined.
3095 * Any existing superblock will overwrite the array and device sizes
3097 r = rs_set_dev_and_array_sectors(rs, false);
3101 calculated_dev_sectors = rs->md.dev_sectors;
3104 * Backup any new raid set level, layout, ...
3105 * requested to be able to compare to superblock
3106 * members for conversion decisions.
3108 rs_config_backup(rs, &rs_layout);
3110 r = analyse_superblocks(ti, rs);
3114 rdev_sectors = __rdev_sectors(rs);
3115 if (!rdev_sectors) {
3116 ti->error = "Invalid rdev size";
3122 reshape_sectors = _get_reshape_sectors(rs);
3123 if (calculated_dev_sectors != rdev_sectors)
3124 resize = calculated_dev_sectors != (reshape_sectors ? rdev_sectors - reshape_sectors : rdev_sectors);
3126 INIT_WORK(&rs->md.event_work, do_table_event);
3128 ti->num_flush_bios = 1;
3130 /* Restore any requested new layout for conversion decision */
3131 rs_config_restore(rs, &rs_layout);
3134 * Now that we have any superblock metadata available,
3135 * check for new, recovering, reshaping, to be taken over,
3136 * to be reshaped or an existing, unchanged raid set to
3139 if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
3140 /* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
3141 if (rs_is_raid6(rs) &&
3142 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
3143 ti->error = "'nosync' not allowed for new raid6 set";
3147 rs_setup_recovery(rs, 0);
3148 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3150 } else if (rs_is_recovering(rs)) {
3151 /* Rebuild particular devices */
3152 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3153 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3154 rs_setup_recovery(rs, MaxSector);
3156 /* A recovering raid set may be resized */
3157 ; /* skip setup rs */
3158 } else if (rs_is_reshaping(rs)) {
3159 /* Have to reject size change request during reshape */
3161 ti->error = "Can't resize a reshaping raid set";
3166 } else if (rs_takeover_requested(rs)) {
3167 if (rs_is_reshaping(rs)) {
3168 ti->error = "Can't takeover a reshaping raid set";
3173 /* We can't takeover a journaled raid4/5/6 */
3174 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3175 ti->error = "Can't takeover a journaled raid4/5/6 set";
3181 * If a takeover is needed, userspace sets any additional
3182 * devices to rebuild and we can check for a valid request here.
3184 * If acceptible, set the level to the new requested
3185 * one, prohibit requesting recovery, allow the raid
3186 * set to run and store superblocks during resume.
3188 r = rs_check_takeover(rs);
3192 r = rs_setup_takeover(rs);
3196 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3197 /* Takeover ain't recovery, so disable recovery */
3198 rs_setup_recovery(rs, MaxSector);
3200 } else if (rs_reshape_requested(rs)) {
3202 * No need to check for 'ongoing' takeover here, because takeover
3203 * is an instant operation as oposed to an ongoing reshape.
3206 /* We can't reshape a journaled raid4/5/6 */
3207 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3208 ti->error = "Can't reshape a journaled raid4/5/6 set";
3213 /* Out-of-place space has to be available to allow for a reshape unless raid1! */
3214 if (reshape_sectors || rs_is_raid1(rs)) {
3216 * We can only prepare for a reshape here, because the
3217 * raid set needs to run to provide the repective reshape
3218 * check functions via its MD personality instance.
3220 * So do the reshape check after md_run() succeeded.
3222 r = rs_prepare_reshape(rs);
3226 /* Reshaping ain't recovery, so disable recovery */
3227 rs_setup_recovery(rs, MaxSector);
3231 /* May not set recovery when a device rebuild is requested */
3232 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3233 rs_setup_recovery(rs, MaxSector);
3234 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3236 rs_setup_recovery(rs, test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ?
3237 0 : (resize ? calculated_dev_sectors : MaxSector));
3241 /* If constructor requested it, change data and new_data offsets */
3242 r = rs_adjust_data_offsets(rs);
3246 /* Catch any inconclusive reshape superblock content. */
3247 rs_reset_inconclusive_reshape(rs);
3249 /* Start raid set read-only and assumed clean to change in raid_resume() */
3253 /* Keep array frozen until resume. */
3254 set_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
3256 /* Has to be held on running the array */
3257 mddev_lock_nointr(&rs->md);
3258 r = md_run(&rs->md);
3259 rs->md.in_sync = 0; /* Assume already marked dirty */
3261 ti->error = "Failed to run raid array";
3262 mddev_unlock(&rs->md);
3266 r = md_start(&rs->md);
3268 ti->error = "Failed to start raid array";
3269 mddev_unlock(&rs->md);
3273 rs->callbacks.congested_fn = raid_is_congested;
3274 dm_table_add_target_callbacks(ti->table, &rs->callbacks);
3276 /* If raid4/5/6 journal mode explicitly requested (only possible with journal dev) -> set it */
3277 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
3278 r = r5c_journal_mode_set(&rs->md, rs->journal_dev.mode);
3280 ti->error = "Failed to set raid4/5/6 journal mode";
3281 mddev_unlock(&rs->md);
3282 goto bad_journal_mode_set;
3286 mddev_suspend(&rs->md);
3287 set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags);
3289 /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
3290 if (rs_is_raid456(rs)) {
3291 r = rs_set_raid456_stripe_cache(rs);
3293 mddev_unlock(&rs->md);
3294 goto bad_stripe_cache;
3298 /* Now do an early reshape check */
3299 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3300 r = rs_check_reshape(rs);
3302 mddev_unlock(&rs->md);
3303 goto bad_check_reshape;
3306 /* Restore new, ctr requested layout to perform check */
3307 rs_config_restore(rs, &rs_layout);
3309 if (rs->md.pers->start_reshape) {
3310 r = rs->md.pers->check_reshape(&rs->md);
3312 ti->error = "Reshape check failed";
3313 mddev_unlock(&rs->md);
3314 goto bad_check_reshape;
3319 /* Disable/enable discard support on raid set. */
3320 configure_discard_support(rs);
3322 mddev_unlock(&rs->md);
3326 bad_journal_mode_set:
3336 static void raid_dtr(struct dm_target *ti)
3338 struct raid_set *rs = ti->private;
3340 list_del_init(&rs->callbacks.list);
3345 static int raid_map(struct dm_target *ti, struct bio *bio)
3347 struct raid_set *rs = ti->private;
3348 struct mddev *mddev = &rs->md;
3351 * If we're reshaping to add disk(s), ti->len and
3352 * mddev->array_sectors will differ during the process
3353 * (ti->len > mddev->array_sectors), so we have to requeue
3354 * bios with addresses > mddev->array_sectors here or
3355 * there will occur accesses past EOD of the component
3356 * data images thus erroring the raid set.
3358 if (unlikely(bio_has_data(bio) && bio_end_sector(bio) > mddev->array_sectors))
3359 return DM_MAPIO_REQUEUE;
3361 md_handle_request(mddev, bio);
3363 return DM_MAPIO_SUBMITTED;
3366 /* Return sync state string for @state */
3367 enum sync_state { st_frozen, st_reshape, st_resync, st_check, st_repair, st_recover, st_idle };
3368 static const char *sync_str(enum sync_state state)
3370 /* Has to be in above sync_state order! */
3371 static const char *sync_strs[] = {
3381 return __within_range(state, 0, ARRAY_SIZE(sync_strs) - 1) ? sync_strs[state] : "undef";
3384 /* Return enum sync_state for @mddev derived from @recovery flags */
3385 static enum sync_state decipher_sync_action(struct mddev *mddev, unsigned long recovery)
3387 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
3390 /* The MD sync thread can be done with io or be interrupted but still be running */
3391 if (!test_bit(MD_RECOVERY_DONE, &recovery) &&
3392 (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
3393 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery)))) {
3394 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
3397 if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
3398 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
3400 if (test_bit(MD_RECOVERY_CHECK, &recovery))
3405 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3408 if (mddev->reshape_position != MaxSector)
3416 * Return status string for @rdev
3418 * Status characters:
3420 * 'D' = Dead/Failed raid set component or raid4/5/6 journal device
3421 * 'a' = Alive but not in-sync raid set component _or_ alive raid4/5/6 'write_back' journal device
3422 * 'A' = Alive and in-sync raid set component _or_ alive raid4/5/6 'write_through' journal device
3423 * '-' = Non-existing device (i.e. uspace passed '- -' into the ctr)
3425 static const char *__raid_dev_status(struct raid_set *rs, struct md_rdev *rdev)
3429 else if (test_bit(Faulty, &rdev->flags))
3431 else if (test_bit(Journal, &rdev->flags))
3432 return (rs->journal_dev.mode == R5C_JOURNAL_MODE_WRITE_THROUGH) ? "A" : "a";
3433 else if (test_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags) ||
3434 (!test_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags) &&
3435 !test_bit(In_sync, &rdev->flags)))
3441 /* Helper to return resync/reshape progress for @rs and runtime flags for raid set in sync / resynching */
3442 static sector_t rs_get_progress(struct raid_set *rs, unsigned long recovery,
3443 sector_t resync_max_sectors)
3446 enum sync_state state;
3447 struct mddev *mddev = &rs->md;
3449 clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3450 clear_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3452 if (rs_is_raid0(rs)) {
3453 r = resync_max_sectors;
3454 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3457 state = decipher_sync_action(mddev, recovery);
3459 if (state == st_idle && !test_bit(MD_RECOVERY_INTR, &recovery))
3460 r = mddev->recovery_cp;
3462 r = mddev->curr_resync_completed;
3464 if (state == st_idle && r >= resync_max_sectors) {
3468 /* In case we have finished recovering, the array is in sync. */
3469 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3470 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3472 } else if (state == st_recover)
3474 * In case we are recovering, the array is not in sync
3475 * and health chars should show the recovering legs.
3478 else if (state == st_resync)
3480 * If "resync" 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);
3485 else if (state == st_reshape)
3487 * If "reshape" is occurring, the raid set
3488 * is or may be out of sync hence the health
3489 * characters shall be 'a'.
3491 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3493 else if (state == st_check || state == st_repair)
3495 * If "check" or "repair" is occurring, the raid set has
3496 * undergone an initial sync and the health characters
3497 * should not be 'a' anymore.
3499 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3502 struct md_rdev *rdev;
3505 * We are idle and recovery is needed, prevent 'A' chars race
3506 * caused by components still set to in-sync by constructor.
3508 if (test_bit(MD_RECOVERY_NEEDED, &recovery))
3509 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3512 * The raid set may be doing an initial sync, or it may
3513 * be rebuilding individual components. If all the
3514 * devices are In_sync, then it is the raid set that is
3515 * being initialized.
3517 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3518 rdev_for_each(rdev, mddev)
3519 if (!test_bit(Journal, &rdev->flags) &&
3520 !test_bit(In_sync, &rdev->flags)) {
3521 clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3527 return min(r, resync_max_sectors);
3530 /* Helper to return @dev name or "-" if !@dev */
3531 static const char *__get_dev_name(struct dm_dev *dev)
3533 return dev ? dev->name : "-";
3536 static void raid_status(struct dm_target *ti, status_type_t type,
3537 unsigned int status_flags, char *result, unsigned int maxlen)
3539 struct raid_set *rs = ti->private;
3540 struct mddev *mddev = &rs->md;
3541 struct r5conf *conf = rs_is_raid456(rs) ? mddev->private : NULL;
3542 int i, max_nr_stripes = conf ? conf->max_nr_stripes : 0;
3543 unsigned long recovery;
3544 unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3545 unsigned int sz = 0;
3546 unsigned int rebuild_disks;
3547 unsigned int write_mostly_params = 0;
3548 sector_t progress, resync_max_sectors, resync_mismatches;
3549 const char *sync_action;
3550 struct raid_type *rt;
3553 case STATUSTYPE_INFO:
3554 /* *Should* always succeed */
3555 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3559 DMEMIT("%s %d ", rt->name, mddev->raid_disks);
3561 /* Access most recent mddev properties for status output */
3563 recovery = rs->md.recovery;
3564 /* Get sensible max sectors even if raid set not yet started */
3565 resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
3566 mddev->resync_max_sectors : mddev->dev_sectors;
3567 progress = rs_get_progress(rs, recovery, resync_max_sectors);
3568 resync_mismatches = (mddev->last_sync_action && !strcasecmp(mddev->last_sync_action, "check")) ?
3569 atomic64_read(&mddev->resync_mismatches) : 0;
3570 sync_action = sync_str(decipher_sync_action(&rs->md, recovery));
3572 /* HM FIXME: do we want another state char for raid0? It shows 'D'/'A'/'-' now */
3573 for (i = 0; i < rs->raid_disks; i++)
3574 DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev));
3577 * In-sync/Reshape ratio:
3578 * The in-sync ratio shows the progress of:
3579 * - Initializing the raid set
3580 * - Rebuilding a subset of devices of the raid set
3581 * The user can distinguish between the two by referring
3582 * to the status characters.
3584 * The reshape ratio shows the progress of
3585 * changing the raid layout or the number of
3586 * disks of a raid set
3588 DMEMIT(" %llu/%llu", (unsigned long long) progress,
3589 (unsigned long long) resync_max_sectors);
3595 * See Documentation/device-mapper/dm-raid.txt for
3596 * information on each of these states.
3598 DMEMIT(" %s", sync_action);
3603 * resync_mismatches/mismatch_cnt
3604 * This field shows the number of discrepancies found when
3605 * performing a "check" of the raid set.
3607 DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3612 * data_offset (needed for out of space reshaping)
3613 * This field shows the data offset into the data
3614 * image LV where the first stripes data starts.
3616 * We keep data_offset equal on all raid disks of the set,
3617 * so retrieving it from the first raid disk is sufficient.
3619 DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
3624 DMEMIT(" %s", test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ?
3625 __raid_dev_status(rs, &rs->journal_dev.rdev) : "-");
3628 case STATUSTYPE_TABLE:
3629 /* Report the table line string you would use to construct this raid set */
3631 /* Calculate raid parameter count */
3632 for (i = 0; i < rs->raid_disks; i++)
3633 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3634 write_mostly_params += 2;
3635 rebuild_disks = memweight(rs->rebuild_disks, DISKS_ARRAY_ELEMS * sizeof(*rs->rebuild_disks));
3636 raid_param_cnt += rebuild_disks * 2 +
3637 write_mostly_params +
3638 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3639 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2 +
3640 (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ? 2 : 0) +
3641 (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags) ? 2 : 0);
3643 /* Emit table line */
3644 /* This has to be in the documented order for userspace! */
3645 DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
3646 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
3647 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
3648 if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3649 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
3651 for (i = 0; i < rs->raid_disks; i++)
3652 if (test_bit(rs->dev[i].rdev.raid_disk, (void *) rs->rebuild_disks))
3653 DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD),
3654 rs->dev[i].rdev.raid_disk);
3655 if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3656 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3657 mddev->bitmap_info.daemon_sleep);
3658 if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3659 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3660 mddev->sync_speed_min);
3661 if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3662 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3663 mddev->sync_speed_max);
3664 if (write_mostly_params)
3665 for (i = 0; i < rs->raid_disks; i++)
3666 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3667 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3668 rs->dev[i].rdev.raid_disk);
3669 if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
3670 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
3671 mddev->bitmap_info.max_write_behind);
3672 if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3673 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3675 if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3676 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3677 (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3678 if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3679 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3680 raid10_md_layout_to_copies(mddev->layout));
3681 if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3682 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3683 raid10_md_layout_to_format(mddev->layout));
3684 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3685 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3686 max(rs->delta_disks, mddev->delta_disks));
3687 if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3688 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3689 (unsigned long long) rs->data_offset);
3690 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags))
3691 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV),
3692 __get_dev_name(rs->journal_dev.dev));
3693 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags))
3694 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE),
3695 md_journal_mode_to_dm_raid(rs->journal_dev.mode));
3696 DMEMIT(" %d", rs->raid_disks);
3697 for (i = 0; i < rs->raid_disks; i++)
3698 DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
3699 __get_dev_name(rs->dev[i].data_dev));
3703 static int raid_message(struct dm_target *ti, unsigned int argc, char **argv,
3704 char *result, unsigned maxlen)
3706 struct raid_set *rs = ti->private;
3707 struct mddev *mddev = &rs->md;
3709 if (!mddev->pers || !mddev->pers->sync_request)
3712 if (!strcasecmp(argv[0], "frozen"))
3713 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3715 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3717 if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
3718 if (mddev->sync_thread) {
3719 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3720 md_reap_sync_thread(mddev);
3722 } else if (decipher_sync_action(mddev, mddev->recovery) != st_idle)
3724 else if (!strcasecmp(argv[0], "resync"))
3725 ; /* MD_RECOVERY_NEEDED set below */
3726 else if (!strcasecmp(argv[0], "recover"))
3727 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3729 if (!strcasecmp(argv[0], "check")) {
3730 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3731 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3732 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3733 } else if (!strcasecmp(argv[0], "repair")) {
3734 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3735 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3739 if (mddev->ro == 2) {
3740 /* A write to sync_action is enough to justify
3741 * canceling read-auto mode
3744 if (!mddev->suspended && mddev->sync_thread)
3745 md_wakeup_thread(mddev->sync_thread);
3747 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3748 if (!mddev->suspended && mddev->thread)
3749 md_wakeup_thread(mddev->thread);
3754 static int raid_iterate_devices(struct dm_target *ti,
3755 iterate_devices_callout_fn fn, void *data)
3757 struct raid_set *rs = ti->private;
3761 for (i = 0; !r && i < rs->raid_disks; i++) {
3762 if (rs->dev[i].data_dev) {
3763 r = fn(ti, rs->dev[i].data_dev,
3764 0, /* No offset on data devs */
3765 rs->md.dev_sectors, data);
3772 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3774 struct raid_set *rs = ti->private;
3775 unsigned int chunk_size = to_bytes(rs->md.chunk_sectors);
3777 blk_limits_io_min(limits, chunk_size);
3778 blk_limits_io_opt(limits, chunk_size * mddev_data_stripes(rs));
3781 static void raid_postsuspend(struct dm_target *ti)
3783 struct raid_set *rs = ti->private;
3785 if (!test_and_set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
3786 /* Writes have to be stopped before suspending to avoid deadlocks. */
3787 if (!test_bit(MD_RECOVERY_FROZEN, &rs->md.recovery))
3788 md_stop_writes(&rs->md);
3790 mddev_lock_nointr(&rs->md);
3791 mddev_suspend(&rs->md);
3792 mddev_unlock(&rs->md);
3796 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
3799 uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
3800 unsigned long flags;
3801 bool cleared = false;
3802 struct dm_raid_superblock *sb;
3803 struct mddev *mddev = &rs->md;
3806 /* RAID personalities have to provide hot add/remove methods or we need to bail out. */
3807 if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
3810 memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
3812 for (i = 0; i < rs->raid_disks; i++) {
3813 r = &rs->dev[i].rdev;
3814 /* HM FIXME: enhance journal device recovery processing */
3815 if (test_bit(Journal, &r->flags))
3818 if (test_bit(Faulty, &r->flags) &&
3819 r->meta_bdev && !read_disk_sb(r, r->sb_size, true)) {
3820 DMINFO("Faulty %s device #%d has readable super block."
3821 " Attempting to revive it.",
3822 rs->raid_type->name, i);
3825 * Faulty bit may be set, but sometimes the array can
3826 * be suspended before the personalities can respond
3827 * by removing the device from the array (i.e. calling
3828 * 'hot_remove_disk'). If they haven't yet removed
3829 * the failed device, its 'raid_disk' number will be
3830 * '>= 0' - meaning we must call this function
3834 clear_bit(In_sync, &r->flags); /* Mandatory for hot remove. */
3835 if (r->raid_disk >= 0) {
3836 if (mddev->pers->hot_remove_disk(mddev, r)) {
3837 /* Failed to revive this device, try next */
3842 r->raid_disk = r->saved_raid_disk = i;
3844 clear_bit(Faulty, &r->flags);
3845 clear_bit(WriteErrorSeen, &r->flags);
3847 if (mddev->pers->hot_add_disk(mddev, r)) {
3848 /* Failed to revive this device, try next */
3849 r->raid_disk = r->saved_raid_disk = -1;
3852 clear_bit(In_sync, &r->flags);
3853 r->recovery_offset = 0;
3854 set_bit(i, (void *) cleared_failed_devices);
3860 /* If any failed devices could be cleared, update all sbs failed_devices bits */
3862 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
3864 rdev_for_each(r, &rs->md) {
3865 if (test_bit(Journal, &r->flags))
3868 sb = page_address(r->sb_page);
3869 sb_retrieve_failed_devices(sb, failed_devices);
3871 for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
3872 failed_devices[i] &= ~cleared_failed_devices[i];
3874 sb_update_failed_devices(sb, failed_devices);
3879 static int __load_dirty_region_bitmap(struct raid_set *rs)
3883 /* Try loading the bitmap unless "raid0", which does not have one */
3884 if (!rs_is_raid0(rs) &&
3885 !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
3886 r = md_bitmap_load(&rs->md);
3888 DMERR("Failed to load bitmap");
3894 /* Enforce updating all superblocks */
3895 static void rs_update_sbs(struct raid_set *rs)
3897 struct mddev *mddev = &rs->md;
3900 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3902 md_update_sb(mddev, 1);
3907 * Reshape changes raid algorithm of @rs to new one within personality
3908 * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3909 * disks from a raid set thus growing/shrinking it or resizes the set
3911 * Call mddev_lock_nointr() before!
3913 static int rs_start_reshape(struct raid_set *rs)
3916 struct mddev *mddev = &rs->md;
3917 struct md_personality *pers = mddev->pers;
3919 /* Don't allow the sync thread to work until the table gets reloaded. */
3920 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
3922 r = rs_setup_reshape(rs);
3927 * Check any reshape constraints enforced by the personalility
3929 * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3931 r = pers->check_reshape(mddev);
3933 rs->ti->error = "pers->check_reshape() failed";
3938 * Personality may not provide start reshape method in which
3939 * case check_reshape above has already covered everything
3941 if (pers->start_reshape) {
3942 r = pers->start_reshape(mddev);
3944 rs->ti->error = "pers->start_reshape() failed";
3950 * Now reshape got set up, update superblocks to
3951 * reflect the fact so that a table reload will
3952 * access proper superblock content in the ctr.
3959 static int raid_preresume(struct dm_target *ti)
3962 struct raid_set *rs = ti->private;
3963 struct mddev *mddev = &rs->md;
3965 /* This is a resume after a suspend of the set -> it's already started. */
3966 if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
3970 * The superblocks need to be updated on disk if the
3971 * array is new or new devices got added (thus zeroed
3972 * out by userspace) or __load_dirty_region_bitmap
3973 * will overwrite them in core with old data or fail.
3975 if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
3978 /* Load the bitmap from disk unless raid0 */
3979 r = __load_dirty_region_bitmap(rs);
3983 /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) */
3984 if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) && mddev->bitmap &&
3985 mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)) {
3986 r = md_bitmap_resize(mddev->bitmap, mddev->dev_sectors,
3987 to_bytes(rs->requested_bitmap_chunk_sectors), 0);
3989 DMERR("Failed to resize bitmap");
3992 /* Check for any resize/reshape on @rs and adjust/initiate */
3993 /* Be prepared for mddev_resume() in raid_resume() */
3994 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3995 if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
3996 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3997 mddev->resync_min = mddev->recovery_cp;
4000 /* Check for any reshape request unless new raid set */
4001 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
4002 /* Initiate a reshape. */
4003 rs_set_rdev_sectors(rs);
4004 mddev_lock_nointr(mddev);
4005 r = rs_start_reshape(rs);
4006 mddev_unlock(mddev);
4008 DMWARN("Failed to check/start reshape, continuing without change");
4015 static void raid_resume(struct dm_target *ti)
4017 struct raid_set *rs = ti->private;
4018 struct mddev *mddev = &rs->md;
4020 if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
4022 * A secondary resume while the device is active.
4023 * Take this opportunity to check whether any failed
4024 * devices are reachable again.
4026 mddev_lock_nointr(mddev);
4027 attempt_restore_of_faulty_devices(rs);
4028 mddev_unlock(mddev);
4031 if (test_and_clear_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
4032 /* Only reduce raid set size before running a disk removing reshape. */
4033 if (mddev->delta_disks < 0)
4034 rs_set_capacity(rs);
4036 mddev_lock_nointr(mddev);
4037 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4040 mddev_resume(mddev);
4041 mddev_unlock(mddev);
4045 static struct target_type raid_target = {
4047 .version = {1, 14, 0},
4048 .module = THIS_MODULE,
4052 .status = raid_status,
4053 .message = raid_message,
4054 .iterate_devices = raid_iterate_devices,
4055 .io_hints = raid_io_hints,
4056 .postsuspend = raid_postsuspend,
4057 .preresume = raid_preresume,
4058 .resume = raid_resume,
4061 static int __init dm_raid_init(void)
4063 DMINFO("Loading target version %u.%u.%u",
4064 raid_target.version[0],
4065 raid_target.version[1],
4066 raid_target.version[2]);
4067 return dm_register_target(&raid_target);
4070 static void __exit dm_raid_exit(void)
4072 dm_unregister_target(&raid_target);
4075 module_init(dm_raid_init);
4076 module_exit(dm_raid_exit);
4078 module_param(devices_handle_discard_safely, bool, 0644);
4079 MODULE_PARM_DESC(devices_handle_discard_safely,
4080 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
4082 MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
4083 MODULE_ALIAS("dm-raid0");
4084 MODULE_ALIAS("dm-raid1");
4085 MODULE_ALIAS("dm-raid10");
4086 MODULE_ALIAS("dm-raid4");
4087 MODULE_ALIAS("dm-raid5");
4088 MODULE_ALIAS("dm-raid6");
4089 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
4090 MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
4091 MODULE_LICENSE("GPL");