2 * Copyright (C) 2010-2011 Neil Brown
3 * Copyright (C) 2010-2016 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
8 #include <linux/slab.h>
9 #include <linux/module.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)
27 static bool devices_handle_discard_safely
= false;
30 * The following flags are used by dm-raid.c to set up the array state.
31 * They must be cleared before md_run is called.
33 #define FirstUse 10 /* rdev flag */
37 * Two DM devices, one to hold metadata and one to hold the
38 * actual data/parity. The reason for this is to not confuse
39 * ti->len and give more flexibility in altering size and
42 * While it is possible for this device to be associated
43 * with a different physical device than the data_dev, it
44 * is intended for it to be the same.
45 * |--------- Physical Device ---------|
46 * |- meta_dev -|------ data_dev ------|
48 struct dm_dev
*meta_dev
;
49 struct dm_dev
*data_dev
;
54 * Bits for establishing rs->ctr_flags
59 #define __CTR_FLAG_SYNC 0 /* 1 */ /* Not with raid0! */
60 #define __CTR_FLAG_NOSYNC 1 /* 1 */ /* Not with raid0! */
61 #define __CTR_FLAG_REBUILD 2 /* 2 */ /* Not with raid0! */
62 #define __CTR_FLAG_DAEMON_SLEEP 3 /* 2 */ /* Not with raid0! */
63 #define __CTR_FLAG_MIN_RECOVERY_RATE 4 /* 2 */ /* Not with raid0! */
64 #define __CTR_FLAG_MAX_RECOVERY_RATE 5 /* 2 */ /* Not with raid0! */
65 #define __CTR_FLAG_MAX_WRITE_BEHIND 6 /* 2 */ /* Only with raid1! */
66 #define __CTR_FLAG_WRITE_MOSTLY 7 /* 2 */ /* Only with raid1! */
67 #define __CTR_FLAG_STRIPE_CACHE 8 /* 2 */ /* Only with raid4/5/6! */
68 #define __CTR_FLAG_REGION_SIZE 9 /* 2 */ /* Not with raid0! */
69 #define __CTR_FLAG_RAID10_COPIES 10 /* 2 */ /* Only with raid10 */
70 #define __CTR_FLAG_RAID10_FORMAT 11 /* 2 */ /* Only with raid10 */
72 #define __CTR_FLAG_DELTA_DISKS 12 /* 2 */ /* Only with reshapable raid1/4/5/6/10! */
73 #define __CTR_FLAG_DATA_OFFSET 13 /* 2 */ /* Only with reshapable raid4/5/6/10! */
74 #define __CTR_FLAG_RAID10_USE_NEAR_SETS 14 /* 2 */ /* Only with raid10! */
77 * Flags for rs->ctr_flags field.
79 #define CTR_FLAG_SYNC (1 << __CTR_FLAG_SYNC)
80 #define CTR_FLAG_NOSYNC (1 << __CTR_FLAG_NOSYNC)
81 #define CTR_FLAG_REBUILD (1 << __CTR_FLAG_REBUILD)
82 #define CTR_FLAG_DAEMON_SLEEP (1 << __CTR_FLAG_DAEMON_SLEEP)
83 #define CTR_FLAG_MIN_RECOVERY_RATE (1 << __CTR_FLAG_MIN_RECOVERY_RATE)
84 #define CTR_FLAG_MAX_RECOVERY_RATE (1 << __CTR_FLAG_MAX_RECOVERY_RATE)
85 #define CTR_FLAG_MAX_WRITE_BEHIND (1 << __CTR_FLAG_MAX_WRITE_BEHIND)
86 #define CTR_FLAG_WRITE_MOSTLY (1 << __CTR_FLAG_WRITE_MOSTLY)
87 #define CTR_FLAG_STRIPE_CACHE (1 << __CTR_FLAG_STRIPE_CACHE)
88 #define CTR_FLAG_REGION_SIZE (1 << __CTR_FLAG_REGION_SIZE)
89 #define CTR_FLAG_RAID10_COPIES (1 << __CTR_FLAG_RAID10_COPIES)
90 #define CTR_FLAG_RAID10_FORMAT (1 << __CTR_FLAG_RAID10_FORMAT)
91 #define CTR_FLAG_DELTA_DISKS (1 << __CTR_FLAG_DELTA_DISKS)
92 #define CTR_FLAG_DATA_OFFSET (1 << __CTR_FLAG_DATA_OFFSET)
93 #define CTR_FLAG_RAID10_USE_NEAR_SETS (1 << __CTR_FLAG_RAID10_USE_NEAR_SETS)
96 * Definitions of various constructor flags to
97 * be used in checks of valid / invalid flags
100 /* Define all any sync flags */
101 #define CTR_FLAGS_ANY_SYNC (CTR_FLAG_SYNC | CTR_FLAG_NOSYNC)
103 /* Define flags for options without argument (e.g. 'nosync') */
104 #define CTR_FLAG_OPTIONS_NO_ARGS (CTR_FLAGS_ANY_SYNC | \
105 CTR_FLAG_RAID10_USE_NEAR_SETS)
107 /* Define flags for options with one argument (e.g. 'delta_disks +2') */
108 #define CTR_FLAG_OPTIONS_ONE_ARG (CTR_FLAG_REBUILD | \
109 CTR_FLAG_WRITE_MOSTLY | \
110 CTR_FLAG_DAEMON_SLEEP | \
111 CTR_FLAG_MIN_RECOVERY_RATE | \
112 CTR_FLAG_MAX_RECOVERY_RATE | \
113 CTR_FLAG_MAX_WRITE_BEHIND | \
114 CTR_FLAG_STRIPE_CACHE | \
115 CTR_FLAG_REGION_SIZE | \
116 CTR_FLAG_RAID10_COPIES | \
117 CTR_FLAG_RAID10_FORMAT | \
118 CTR_FLAG_DELTA_DISKS | \
119 CTR_FLAG_DATA_OFFSET)
121 /* Valid options definitions per raid level... */
123 /* "raid0" does only accept data offset */
124 #define RAID0_VALID_FLAGS (CTR_FLAG_DATA_OFFSET)
126 /* "raid1" does not accept stripe cache, data offset, delta_disks or any raid10 options */
127 #define RAID1_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
129 CTR_FLAG_WRITE_MOSTLY | \
130 CTR_FLAG_DAEMON_SLEEP | \
131 CTR_FLAG_MIN_RECOVERY_RATE | \
132 CTR_FLAG_MAX_RECOVERY_RATE | \
133 CTR_FLAG_MAX_WRITE_BEHIND | \
134 CTR_FLAG_REGION_SIZE | \
135 CTR_FLAG_DELTA_DISKS | \
136 CTR_FLAG_DATA_OFFSET)
138 /* "raid10" does not accept any raid1 or stripe cache options */
139 #define RAID10_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
141 CTR_FLAG_DAEMON_SLEEP | \
142 CTR_FLAG_MIN_RECOVERY_RATE | \
143 CTR_FLAG_MAX_RECOVERY_RATE | \
144 CTR_FLAG_REGION_SIZE | \
145 CTR_FLAG_RAID10_COPIES | \
146 CTR_FLAG_RAID10_FORMAT | \
147 CTR_FLAG_DELTA_DISKS | \
148 CTR_FLAG_DATA_OFFSET | \
149 CTR_FLAG_RAID10_USE_NEAR_SETS)
152 * "raid4/5/6" do not accept any raid1 or raid10 specific options
154 * "raid6" does not accept "nosync", because it is not guaranteed
155 * that both parity and q-syndrome are being written properly with
158 #define RAID45_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
160 CTR_FLAG_DAEMON_SLEEP | \
161 CTR_FLAG_MIN_RECOVERY_RATE | \
162 CTR_FLAG_MAX_RECOVERY_RATE | \
163 CTR_FLAG_MAX_WRITE_BEHIND | \
164 CTR_FLAG_STRIPE_CACHE | \
165 CTR_FLAG_REGION_SIZE | \
166 CTR_FLAG_DELTA_DISKS | \
167 CTR_FLAG_DATA_OFFSET)
169 #define RAID6_VALID_FLAGS (CTR_FLAG_SYNC | \
171 CTR_FLAG_DAEMON_SLEEP | \
172 CTR_FLAG_MIN_RECOVERY_RATE | \
173 CTR_FLAG_MAX_RECOVERY_RATE | \
174 CTR_FLAG_MAX_WRITE_BEHIND | \
175 CTR_FLAG_STRIPE_CACHE | \
176 CTR_FLAG_REGION_SIZE | \
177 CTR_FLAG_DELTA_DISKS | \
178 CTR_FLAG_DATA_OFFSET)
179 /* ...valid options definitions per raid level */
182 * Flags for rs->runtime_flags field
183 * (RT_FLAG prefix meaning "runtime flag")
185 * These are all internal and used to define runtime state,
186 * e.g. to prevent another resume from preresume processing
187 * the raid set all over again.
189 #define RT_FLAG_RS_PRERESUMED 0
190 #define RT_FLAG_RS_RESUMED 1
191 #define RT_FLAG_RS_BITMAP_LOADED 2
192 #define RT_FLAG_UPDATE_SBS 3
193 #define RT_FLAG_RESHAPE_RS 4
194 #define RT_FLAG_KEEP_RS_FROZEN 5
196 /* Array elements of 64 bit needed for rebuild/failed disk bits */
197 #define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
200 * raid set level, layout and chunk sectors backup/restore
205 int new_chunk_sectors
;
209 struct dm_target
*ti
;
211 uint32_t bitmap_loaded
;
212 uint32_t stripe_cache_entries
;
213 unsigned long ctr_flags
;
214 unsigned long runtime_flags
;
216 uint64_t rebuild_disks
[DISKS_ARRAY_ELEMS
];
222 int requested_bitmap_chunk_sectors
;
225 struct raid_type
*raid_type
;
226 struct dm_target_callbacks callbacks
;
228 struct raid_dev dev
[0];
231 static void rs_config_backup(struct raid_set
*rs
, struct rs_layout
*l
)
233 struct mddev
*mddev
= &rs
->md
;
235 l
->new_level
= mddev
->new_level
;
236 l
->new_layout
= mddev
->new_layout
;
237 l
->new_chunk_sectors
= mddev
->new_chunk_sectors
;
240 static void rs_config_restore(struct raid_set
*rs
, struct rs_layout
*l
)
242 struct mddev
*mddev
= &rs
->md
;
244 mddev
->new_level
= l
->new_level
;
245 mddev
->new_layout
= l
->new_layout
;
246 mddev
->new_chunk_sectors
= l
->new_chunk_sectors
;
249 /* raid10 algorithms (i.e. formats) */
250 #define ALGORITHM_RAID10_DEFAULT 0
251 #define ALGORITHM_RAID10_NEAR 1
252 #define ALGORITHM_RAID10_OFFSET 2
253 #define ALGORITHM_RAID10_FAR 3
255 /* Supported raid types and properties. */
256 static struct raid_type
{
257 const char *name
; /* RAID algorithm. */
258 const char *descr
; /* Descriptor text for logging. */
259 const unsigned int parity_devs
; /* # of parity devices. */
260 const unsigned int minimal_devs
;/* minimal # of devices in set. */
261 const unsigned int level
; /* RAID level. */
262 const unsigned int algorithm
; /* RAID algorithm. */
264 {"raid0", "raid0 (striping)", 0, 2, 0, 0 /* NONE */},
265 {"raid1", "raid1 (mirroring)", 0, 2, 1, 0 /* NONE */},
266 {"raid10_far", "raid10 far (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_FAR
},
267 {"raid10_offset", "raid10 offset (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_OFFSET
},
268 {"raid10_near", "raid10 near (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_NEAR
},
269 {"raid10", "raid10 (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_DEFAULT
},
270 {"raid4", "raid4 (dedicated last parity disk)", 1, 2, 4, ALGORITHM_PARITY_N
}, /* raid4 layout = raid5_n */
271 {"raid5_n", "raid5 (dedicated last parity disk)", 1, 2, 5, ALGORITHM_PARITY_N
},
272 {"raid5_ls", "raid5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC
},
273 {"raid5_rs", "raid5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC
},
274 {"raid5_la", "raid5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC
},
275 {"raid5_ra", "raid5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC
},
276 {"raid6_zr", "raid6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART
},
277 {"raid6_nr", "raid6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART
},
278 {"raid6_nc", "raid6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE
},
279 {"raid6_n_6", "raid6 (dedicated parity/Q n/6)", 2, 4, 6, ALGORITHM_PARITY_N_6
},
280 {"raid6_ls_6", "raid6 (left symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_SYMMETRIC_6
},
281 {"raid6_rs_6", "raid6 (right symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_SYMMETRIC_6
},
282 {"raid6_la_6", "raid6 (left asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_ASYMMETRIC_6
},
283 {"raid6_ra_6", "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_ASYMMETRIC_6
}
286 /* True, if @v is in inclusive range [@min, @max] */
287 static bool __within_range(long v
, long min
, long max
)
289 return v
>= min
&& v
<= max
;
292 /* All table line arguments are defined here */
293 static struct arg_name_flag
{
294 const unsigned long flag
;
296 } __arg_name_flags
[] = {
297 { CTR_FLAG_SYNC
, "sync"},
298 { CTR_FLAG_NOSYNC
, "nosync"},
299 { CTR_FLAG_REBUILD
, "rebuild"},
300 { CTR_FLAG_DAEMON_SLEEP
, "daemon_sleep"},
301 { CTR_FLAG_MIN_RECOVERY_RATE
, "min_recovery_rate"},
302 { CTR_FLAG_MAX_RECOVERY_RATE
, "max_recovery_rate"},
303 { CTR_FLAG_MAX_WRITE_BEHIND
, "max_write_behind"},
304 { CTR_FLAG_WRITE_MOSTLY
, "write_mostly"},
305 { CTR_FLAG_STRIPE_CACHE
, "stripe_cache"},
306 { CTR_FLAG_REGION_SIZE
, "region_size"},
307 { CTR_FLAG_RAID10_COPIES
, "raid10_copies"},
308 { CTR_FLAG_RAID10_FORMAT
, "raid10_format"},
309 { CTR_FLAG_DATA_OFFSET
, "data_offset"},
310 { CTR_FLAG_DELTA_DISKS
, "delta_disks"},
311 { CTR_FLAG_RAID10_USE_NEAR_SETS
, "raid10_use_near_sets"},
314 /* Return argument name string for given @flag */
315 static const char *dm_raid_arg_name_by_flag(const uint32_t flag
)
317 if (hweight32(flag
) == 1) {
318 struct arg_name_flag
*anf
= __arg_name_flags
+ ARRAY_SIZE(__arg_name_flags
);
320 while (anf
-- > __arg_name_flags
)
321 if (flag
& anf
->flag
)
325 DMERR("%s called with more than one flag!", __func__
);
331 * Bool helpers to test for various raid levels of a raid set.
332 * It's level as reported by the superblock rather than
333 * the requested raid_type passed to the constructor.
335 /* Return true, if raid set in @rs is raid0 */
336 static bool rs_is_raid0(struct raid_set
*rs
)
338 return !rs
->md
.level
;
341 /* Return true, if raid set in @rs is raid1 */
342 static bool rs_is_raid1(struct raid_set
*rs
)
344 return rs
->md
.level
== 1;
347 /* Return true, if raid set in @rs is raid10 */
348 static bool rs_is_raid10(struct raid_set
*rs
)
350 return rs
->md
.level
== 10;
353 /* Return true, if raid set in @rs is level 6 */
354 static bool rs_is_raid6(struct raid_set
*rs
)
356 return rs
->md
.level
== 6;
359 /* Return true, if raid set in @rs is level 4, 5 or 6 */
360 static bool rs_is_raid456(struct raid_set
*rs
)
362 return __within_range(rs
->md
.level
, 4, 6);
365 /* Return true, if raid set in @rs is reshapable */
366 static bool __is_raid10_far(int layout
);
367 static bool rs_is_reshapable(struct raid_set
*rs
)
369 return rs_is_raid456(rs
) ||
370 (rs_is_raid10(rs
) && !__is_raid10_far(rs
->md
.new_layout
));
373 /* Return true, if raid set in @rs is recovering */
374 static bool rs_is_recovering(struct raid_set
*rs
)
376 return rs
->md
.recovery_cp
< rs
->dev
[0].rdev
.sectors
;
379 /* Return true, if raid set in @rs is reshaping */
380 static bool rs_is_reshaping(struct raid_set
*rs
)
382 return rs
->md
.reshape_position
!= MaxSector
;
386 * bool helpers to test for various raid levels of a raid type @rt
389 /* Return true, if raid type in @rt is raid0 */
390 static bool rt_is_raid0(struct raid_type
*rt
)
395 /* Return true, if raid type in @rt is raid1 */
396 static bool rt_is_raid1(struct raid_type
*rt
)
398 return rt
->level
== 1;
401 /* Return true, if raid type in @rt is raid10 */
402 static bool rt_is_raid10(struct raid_type
*rt
)
404 return rt
->level
== 10;
407 /* Return true, if raid type in @rt is raid4/5 */
408 static bool rt_is_raid45(struct raid_type
*rt
)
410 return __within_range(rt
->level
, 4, 5);
413 /* Return true, if raid type in @rt is raid6 */
414 static bool rt_is_raid6(struct raid_type
*rt
)
416 return rt
->level
== 6;
419 /* Return true, if raid type in @rt is raid4/5/6 */
420 static bool rt_is_raid456(struct raid_type
*rt
)
422 return __within_range(rt
->level
, 4, 6);
424 /* END: raid level bools */
426 /* Return valid ctr flags for the raid level of @rs */
427 static unsigned long __valid_flags(struct raid_set
*rs
)
429 if (rt_is_raid0(rs
->raid_type
))
430 return RAID0_VALID_FLAGS
;
431 else if (rt_is_raid1(rs
->raid_type
))
432 return RAID1_VALID_FLAGS
;
433 else if (rt_is_raid10(rs
->raid_type
))
434 return RAID10_VALID_FLAGS
;
435 else if (rt_is_raid45(rs
->raid_type
))
436 return RAID45_VALID_FLAGS
;
437 else if (rt_is_raid6(rs
->raid_type
))
438 return RAID6_VALID_FLAGS
;
444 * Check for valid flags set on @rs
446 * Has to be called after parsing of the ctr flags!
448 static int rs_check_for_valid_flags(struct raid_set
*rs
)
450 if (rs
->ctr_flags
& ~__valid_flags(rs
)) {
451 rs
->ti
->error
= "Invalid flags combination";
458 /* MD raid10 bit definitions and helpers */
459 #define RAID10_OFFSET (1 << 16) /* stripes with data copies area adjacent on devices */
460 #define RAID10_BROCKEN_USE_FAR_SETS (1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
461 #define RAID10_USE_FAR_SETS (1 << 18) /* Use sets instead of whole stripe rotation */
462 #define RAID10_FAR_COPIES_SHIFT 8 /* raid10 # far copies shift (2nd byte of layout) */
464 /* Return md raid10 near copies for @layout */
465 static unsigned int __raid10_near_copies(int layout
)
467 return layout
& 0xFF;
470 /* Return md raid10 far copies for @layout */
471 static unsigned int __raid10_far_copies(int layout
)
473 return __raid10_near_copies(layout
>> RAID10_FAR_COPIES_SHIFT
);
476 /* Return true if md raid10 offset for @layout */
477 static bool __is_raid10_offset(int layout
)
479 return !!(layout
& RAID10_OFFSET
);
482 /* Return true if md raid10 near for @layout */
483 static bool __is_raid10_near(int layout
)
485 return !__is_raid10_offset(layout
) && __raid10_near_copies(layout
) > 1;
488 /* Return true if md raid10 far for @layout */
489 static bool __is_raid10_far(int layout
)
491 return !__is_raid10_offset(layout
) && __raid10_far_copies(layout
) > 1;
494 /* Return md raid10 layout string for @layout */
495 static const char *raid10_md_layout_to_format(int layout
)
498 * Bit 16 stands for "offset"
499 * (i.e. adjacent stripes hold copies)
501 * Refer to MD's raid10.c for details
503 if (__is_raid10_offset(layout
))
506 if (__raid10_near_copies(layout
) > 1)
509 WARN_ON(__raid10_far_copies(layout
) < 2);
514 /* Return md raid10 algorithm for @name */
515 static int raid10_name_to_format(const char *name
)
517 if (!strcasecmp(name
, "near"))
518 return ALGORITHM_RAID10_NEAR
;
519 else if (!strcasecmp(name
, "offset"))
520 return ALGORITHM_RAID10_OFFSET
;
521 else if (!strcasecmp(name
, "far"))
522 return ALGORITHM_RAID10_FAR
;
527 /* Return md raid10 copies for @layout */
528 static unsigned int raid10_md_layout_to_copies(int layout
)
530 return max(__raid10_near_copies(layout
), __raid10_far_copies(layout
));
533 /* Return md raid10 format id for @format string */
534 static int raid10_format_to_md_layout(struct raid_set
*rs
,
535 unsigned int algorithm
,
538 unsigned int n
= 1, f
= 1, r
= 0;
541 * MD resilienece flaw:
543 * enabling use_far_sets for far/offset formats causes copies
544 * to be colocated on the same devs together with their origins!
546 * -> disable it for now in the definition above
548 if (algorithm
== ALGORITHM_RAID10_DEFAULT
||
549 algorithm
== ALGORITHM_RAID10_NEAR
)
552 else if (algorithm
== ALGORITHM_RAID10_OFFSET
) {
555 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS
, &rs
->ctr_flags
))
556 r
|= RAID10_USE_FAR_SETS
;
558 } else if (algorithm
== ALGORITHM_RAID10_FAR
) {
561 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS
, &rs
->ctr_flags
))
562 r
|= RAID10_USE_FAR_SETS
;
567 return r
| (f
<< RAID10_FAR_COPIES_SHIFT
) | n
;
569 /* END: MD raid10 bit definitions and helpers */
571 /* Check for any of the raid10 algorithms */
572 static bool __got_raid10(struct raid_type
*rtp
, const int layout
)
574 if (rtp
->level
== 10) {
575 switch (rtp
->algorithm
) {
576 case ALGORITHM_RAID10_DEFAULT
:
577 case ALGORITHM_RAID10_NEAR
:
578 return __is_raid10_near(layout
);
579 case ALGORITHM_RAID10_OFFSET
:
580 return __is_raid10_offset(layout
);
581 case ALGORITHM_RAID10_FAR
:
582 return __is_raid10_far(layout
);
591 /* Return raid_type for @name */
592 static struct raid_type
*get_raid_type(const char *name
)
594 struct raid_type
*rtp
= raid_types
+ ARRAY_SIZE(raid_types
);
596 while (rtp
-- > raid_types
)
597 if (!strcasecmp(rtp
->name
, name
))
603 /* Return raid_type for @name based derived from @level and @layout */
604 static struct raid_type
*get_raid_type_by_ll(const int level
, const int layout
)
606 struct raid_type
*rtp
= raid_types
+ ARRAY_SIZE(raid_types
);
608 while (rtp
-- > raid_types
) {
609 /* RAID10 special checks based on @layout flags/properties */
610 if (rtp
->level
== level
&&
611 (__got_raid10(rtp
, layout
) || rtp
->algorithm
== layout
))
619 * Conditionally change bdev capacity of @rs
620 * in case of a disk add/remove reshape
622 static void rs_set_capacity(struct raid_set
*rs
)
624 struct mddev
*mddev
= &rs
->md
;
625 struct md_rdev
*rdev
;
626 struct gendisk
*gendisk
= dm_disk(dm_table_get_md(rs
->ti
->table
));
629 * raid10 sets rdev->sector to the device size, which
630 * is unintended in case of out-of-place reshaping
632 rdev_for_each(rdev
, mddev
)
633 rdev
->sectors
= mddev
->dev_sectors
;
635 set_capacity(gendisk
, mddev
->array_sectors
);
636 revalidate_disk(gendisk
);
640 * Set the mddev properties in @rs to the current
641 * ones retrieved from the freshest superblock
643 static void rs_set_cur(struct raid_set
*rs
)
645 struct mddev
*mddev
= &rs
->md
;
647 mddev
->new_level
= mddev
->level
;
648 mddev
->new_layout
= mddev
->layout
;
649 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
653 * Set the mddev properties in @rs to the new
654 * ones requested by the ctr
656 static void rs_set_new(struct raid_set
*rs
)
658 struct mddev
*mddev
= &rs
->md
;
660 mddev
->level
= mddev
->new_level
;
661 mddev
->layout
= mddev
->new_layout
;
662 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
663 mddev
->raid_disks
= rs
->raid_disks
;
664 mddev
->delta_disks
= 0;
667 static struct raid_set
*raid_set_alloc(struct dm_target
*ti
, struct raid_type
*raid_type
,
668 unsigned int raid_devs
)
673 if (raid_devs
<= raid_type
->parity_devs
) {
674 ti
->error
= "Insufficient number of devices";
675 return ERR_PTR(-EINVAL
);
678 rs
= kzalloc(sizeof(*rs
) + raid_devs
* sizeof(rs
->dev
[0]), GFP_KERNEL
);
680 ti
->error
= "Cannot allocate raid context";
681 return ERR_PTR(-ENOMEM
);
686 rs
->raid_disks
= raid_devs
;
690 rs
->raid_type
= raid_type
;
691 rs
->stripe_cache_entries
= 256;
692 rs
->md
.raid_disks
= raid_devs
;
693 rs
->md
.level
= raid_type
->level
;
694 rs
->md
.new_level
= rs
->md
.level
;
695 rs
->md
.layout
= raid_type
->algorithm
;
696 rs
->md
.new_layout
= rs
->md
.layout
;
697 rs
->md
.delta_disks
= 0;
698 rs
->md
.recovery_cp
= MaxSector
;
700 for (i
= 0; i
< raid_devs
; i
++)
701 md_rdev_init(&rs
->dev
[i
].rdev
);
704 * Remaining items to be initialized by further RAID params:
707 * rs->md.chunk_sectors
708 * rs->md.new_chunk_sectors
715 static void raid_set_free(struct raid_set
*rs
)
719 for (i
= 0; i
< rs
->raid_disks
; i
++) {
720 if (rs
->dev
[i
].meta_dev
)
721 dm_put_device(rs
->ti
, rs
->dev
[i
].meta_dev
);
722 md_rdev_clear(&rs
->dev
[i
].rdev
);
723 if (rs
->dev
[i
].data_dev
)
724 dm_put_device(rs
->ti
, rs
->dev
[i
].data_dev
);
731 * For every device we have two words
732 * <meta_dev>: meta device name or '-' if missing
733 * <data_dev>: data device name or '-' if missing
735 * The following are permitted:
738 * <meta_dev> <data_dev>
740 * The following is not allowed:
743 * This code parses those words. If there is a failure,
744 * the caller must use raid_set_free() to unwind the operations.
746 static int parse_dev_params(struct raid_set
*rs
, struct dm_arg_set
*as
)
750 int metadata_available
= 0;
754 /* Put off the number of raid devices argument to get to dev pairs */
755 arg
= dm_shift_arg(as
);
759 for (i
= 0; i
< rs
->raid_disks
; i
++) {
760 rs
->dev
[i
].rdev
.raid_disk
= i
;
762 rs
->dev
[i
].meta_dev
= NULL
;
763 rs
->dev
[i
].data_dev
= NULL
;
766 * There are no offsets, since there is a separate device
767 * for data and metadata.
769 rs
->dev
[i
].rdev
.data_offset
= 0;
770 rs
->dev
[i
].rdev
.mddev
= &rs
->md
;
772 arg
= dm_shift_arg(as
);
776 if (strcmp(arg
, "-")) {
777 r
= dm_get_device(rs
->ti
, arg
, dm_table_get_mode(rs
->ti
->table
),
778 &rs
->dev
[i
].meta_dev
);
780 rs
->ti
->error
= "RAID metadata device lookup failure";
784 rs
->dev
[i
].rdev
.sb_page
= alloc_page(GFP_KERNEL
);
785 if (!rs
->dev
[i
].rdev
.sb_page
) {
786 rs
->ti
->error
= "Failed to allocate superblock page";
791 arg
= dm_shift_arg(as
);
795 if (!strcmp(arg
, "-")) {
796 if (!test_bit(In_sync
, &rs
->dev
[i
].rdev
.flags
) &&
797 (!rs
->dev
[i
].rdev
.recovery_offset
)) {
798 rs
->ti
->error
= "Drive designated for rebuild not specified";
802 if (rs
->dev
[i
].meta_dev
) {
803 rs
->ti
->error
= "No data device supplied with metadata device";
810 r
= dm_get_device(rs
->ti
, arg
, dm_table_get_mode(rs
->ti
->table
),
811 &rs
->dev
[i
].data_dev
);
813 rs
->ti
->error
= "RAID device lookup failure";
817 if (rs
->dev
[i
].meta_dev
) {
818 metadata_available
= 1;
819 rs
->dev
[i
].rdev
.meta_bdev
= rs
->dev
[i
].meta_dev
->bdev
;
821 rs
->dev
[i
].rdev
.bdev
= rs
->dev
[i
].data_dev
->bdev
;
822 list_add_tail(&rs
->dev
[i
].rdev
.same_set
, &rs
->md
.disks
);
823 if (!test_bit(In_sync
, &rs
->dev
[i
].rdev
.flags
))
827 if (metadata_available
) {
829 rs
->md
.persistent
= 1;
830 rs
->md
.major_version
= 2;
831 } else if (rebuild
&& !rs
->md
.recovery_cp
) {
833 * Without metadata, we will not be able to tell if the array
834 * is in-sync or not - we must assume it is not. Therefore,
835 * it is impossible to rebuild a drive.
837 * Even if there is metadata, the on-disk information may
838 * indicate that the array is not in-sync and it will then
841 * User could specify 'nosync' option if desperate.
843 rs
->ti
->error
= "Unable to rebuild drive while array is not in-sync";
851 * validate_region_size
853 * @region_size: region size in sectors. If 0, pick a size (4MiB default).
855 * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
856 * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
858 * Returns: 0 on success, -EINVAL on failure.
860 static int validate_region_size(struct raid_set
*rs
, unsigned long region_size
)
862 unsigned long min_region_size
= rs
->ti
->len
/ (1 << 21);
866 * Choose a reasonable default. All figures in sectors.
868 if (min_region_size
> (1 << 13)) {
869 /* If not a power of 2, make it the next power of 2 */
870 region_size
= roundup_pow_of_two(min_region_size
);
871 DMINFO("Choosing default region size of %lu sectors",
874 DMINFO("Choosing default region size of 4MiB");
875 region_size
= 1 << 13; /* sectors */
879 * Validate user-supplied value.
881 if (region_size
> rs
->ti
->len
) {
882 rs
->ti
->error
= "Supplied region size is too large";
886 if (region_size
< min_region_size
) {
887 DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
888 region_size
, min_region_size
);
889 rs
->ti
->error
= "Supplied region size is too small";
893 if (!is_power_of_2(region_size
)) {
894 rs
->ti
->error
= "Region size is not a power of 2";
898 if (region_size
< rs
->md
.chunk_sectors
) {
899 rs
->ti
->error
= "Region size is smaller than the chunk size";
905 * Convert sectors to bytes.
907 rs
->md
.bitmap_info
.chunksize
= to_bytes(region_size
);
913 * validate_raid_redundancy
916 * Determine if there are enough devices in the array that haven't
917 * failed (or are being rebuilt) to form a usable array.
919 * Returns: 0 on success, -EINVAL on failure.
921 static int validate_raid_redundancy(struct raid_set
*rs
)
923 unsigned int i
, rebuild_cnt
= 0;
924 unsigned int rebuilds_per_group
= 0, copies
;
925 unsigned int group_size
, last_group_start
;
927 for (i
= 0; i
< rs
->md
.raid_disks
; i
++)
928 if (!test_bit(In_sync
, &rs
->dev
[i
].rdev
.flags
) ||
929 !rs
->dev
[i
].rdev
.sb_page
)
932 switch (rs
->raid_type
->level
) {
934 if (rebuild_cnt
>= rs
->md
.raid_disks
)
940 if (rebuild_cnt
> rs
->raid_type
->parity_devs
)
944 copies
= raid10_md_layout_to_copies(rs
->md
.new_layout
);
945 if (rebuild_cnt
< copies
)
949 * It is possible to have a higher rebuild count for RAID10,
950 * as long as the failed devices occur in different mirror
951 * groups (i.e. different stripes).
953 * When checking "near" format, make sure no adjacent devices
954 * have failed beyond what can be handled. In addition to the
955 * simple case where the number of devices is a multiple of the
956 * number of copies, we must also handle cases where the number
957 * of devices is not a multiple of the number of copies.
958 * E.g. dev1 dev2 dev3 dev4 dev5
962 if (__is_raid10_near(rs
->md
.new_layout
)) {
963 for (i
= 0; i
< rs
->md
.raid_disks
; i
++) {
965 rebuilds_per_group
= 0;
966 if ((!rs
->dev
[i
].rdev
.sb_page
||
967 !test_bit(In_sync
, &rs
->dev
[i
].rdev
.flags
)) &&
968 (++rebuilds_per_group
>= copies
))
975 * When checking "far" and "offset" formats, we need to ensure
976 * that the device that holds its copy is not also dead or
977 * being rebuilt. (Note that "far" and "offset" formats only
978 * support two copies right now. These formats also only ever
979 * use the 'use_far_sets' variant.)
981 * This check is somewhat complicated by the need to account
982 * for arrays that are not a multiple of (far) copies. This
983 * results in the need to treat the last (potentially larger)
986 group_size
= (rs
->md
.raid_disks
/ copies
);
987 last_group_start
= (rs
->md
.raid_disks
/ group_size
) - 1;
988 last_group_start
*= group_size
;
989 for (i
= 0; i
< rs
->md
.raid_disks
; i
++) {
990 if (!(i
% copies
) && !(i
> last_group_start
))
991 rebuilds_per_group
= 0;
992 if ((!rs
->dev
[i
].rdev
.sb_page
||
993 !test_bit(In_sync
, &rs
->dev
[i
].rdev
.flags
)) &&
994 (++rebuilds_per_group
>= copies
))
1010 * Possible arguments are...
1011 * <chunk_size> [optional_args]
1013 * Argument definitions
1014 * <chunk_size> The number of sectors per disk that
1015 * will form the "stripe"
1016 * [[no]sync] Force or prevent recovery of the
1018 * [rebuild <idx>] Rebuild the drive indicated by the index
1019 * [daemon_sleep <ms>] Time between bitmap daemon work to
1021 * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1022 * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1023 * [write_mostly <idx>] Indicate a write mostly drive via index
1024 * [max_write_behind <sectors>] See '-write-behind=' (man mdadm)
1025 * [stripe_cache <sectors>] Stripe cache size for higher RAIDs
1026 * [region_size <sectors>] Defines granularity of bitmap
1028 * RAID10-only options:
1029 * [raid10_copies <# copies>] Number of copies. (Default: 2)
1030 * [raid10_format <near|far|offset>] Layout algorithm. (Default: near)
1032 static int parse_raid_params(struct raid_set
*rs
, struct dm_arg_set
*as
,
1033 unsigned int num_raid_params
)
1035 int value
, raid10_format
= ALGORITHM_RAID10_DEFAULT
;
1036 unsigned int raid10_copies
= 2;
1037 unsigned int i
, write_mostly
= 0;
1038 unsigned int region_size
= 0;
1039 sector_t max_io_len
;
1040 const char *arg
, *key
;
1041 struct raid_dev
*rd
;
1042 struct raid_type
*rt
= rs
->raid_type
;
1044 arg
= dm_shift_arg(as
);
1045 num_raid_params
--; /* Account for chunk_size argument */
1047 if (kstrtoint(arg
, 10, &value
) < 0) {
1048 rs
->ti
->error
= "Bad numerical argument given for chunk_size";
1053 * First, parse the in-order required arguments
1054 * "chunk_size" is the only argument of this type.
1056 if (rt_is_raid1(rt
)) {
1058 DMERR("Ignoring chunk size parameter for RAID 1");
1060 } else if (!is_power_of_2(value
)) {
1061 rs
->ti
->error
= "Chunk size must be a power of 2";
1063 } else if (value
< 8) {
1064 rs
->ti
->error
= "Chunk size value is too small";
1068 rs
->md
.new_chunk_sectors
= rs
->md
.chunk_sectors
= value
;
1071 * We set each individual device as In_sync with a completed
1072 * 'recovery_offset'. If there has been a device failure or
1073 * replacement then one of the following cases applies:
1075 * 1) User specifies 'rebuild'.
1076 * - Device is reset when param is read.
1077 * 2) A new device is supplied.
1078 * - No matching superblock found, resets device.
1079 * 3) Device failure was transient and returns on reload.
1080 * - Failure noticed, resets device for bitmap replay.
1081 * 4) Device hadn't completed recovery after previous failure.
1082 * - Superblock is read and overrides recovery_offset.
1084 * What is found in the superblocks of the devices is always
1085 * authoritative, unless 'rebuild' or '[no]sync' was specified.
1087 for (i
= 0; i
< rs
->raid_disks
; i
++) {
1088 set_bit(In_sync
, &rs
->dev
[i
].rdev
.flags
);
1089 rs
->dev
[i
].rdev
.recovery_offset
= MaxSector
;
1093 * Second, parse the unordered optional arguments
1095 for (i
= 0; i
< num_raid_params
; i
++) {
1096 key
= dm_shift_arg(as
);
1098 rs
->ti
->error
= "Not enough raid parameters given";
1102 if (!strcasecmp(key
, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC
))) {
1103 if (test_and_set_bit(__CTR_FLAG_NOSYNC
, &rs
->ctr_flags
)) {
1104 rs
->ti
->error
= "Only one 'nosync' argument allowed";
1109 if (!strcasecmp(key
, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC
))) {
1110 if (test_and_set_bit(__CTR_FLAG_SYNC
, &rs
->ctr_flags
)) {
1111 rs
->ti
->error
= "Only one 'sync' argument allowed";
1116 if (!strcasecmp(key
, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS
))) {
1117 if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS
, &rs
->ctr_flags
)) {
1118 rs
->ti
->error
= "Only one 'raid10_use_new_sets' argument allowed";
1124 arg
= dm_shift_arg(as
);
1125 i
++; /* Account for the argument pairs */
1127 rs
->ti
->error
= "Wrong number of raid parameters given";
1132 * Parameters that take a string value are checked here.
1135 if (!strcasecmp(key
, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT
))) {
1136 if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT
, &rs
->ctr_flags
)) {
1137 rs
->ti
->error
= "Only one 'raid10_format' argument pair allowed";
1140 if (!rt_is_raid10(rt
)) {
1141 rs
->ti
->error
= "'raid10_format' is an invalid parameter for this RAID type";
1144 raid10_format
= raid10_name_to_format(arg
);
1145 if (raid10_format
< 0) {
1146 rs
->ti
->error
= "Invalid 'raid10_format' value given";
1147 return raid10_format
;
1152 if (kstrtoint(arg
, 10, &value
) < 0) {
1153 rs
->ti
->error
= "Bad numerical argument given in raid params";
1157 if (!strcasecmp(key
, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD
))) {
1159 * "rebuild" is being passed in by userspace to provide
1160 * indexes of replaced devices and to set up additional
1161 * devices on raid level takeover.
1163 if (!__within_range(value
, 0, rs
->raid_disks
- 1)) {
1164 rs
->ti
->error
= "Invalid rebuild index given";
1168 if (test_and_set_bit(value
, (void *) rs
->rebuild_disks
)) {
1169 rs
->ti
->error
= "rebuild for this index already given";
1173 rd
= rs
->dev
+ value
;
1174 clear_bit(In_sync
, &rd
->rdev
.flags
);
1175 clear_bit(Faulty
, &rd
->rdev
.flags
);
1176 rd
->rdev
.recovery_offset
= 0;
1177 set_bit(__CTR_FLAG_REBUILD
, &rs
->ctr_flags
);
1178 } else if (!strcasecmp(key
, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY
))) {
1179 if (!rt_is_raid1(rt
)) {
1180 rs
->ti
->error
= "write_mostly option is only valid for RAID1";
1184 if (!__within_range(value
, 0, rs
->md
.raid_disks
- 1)) {
1185 rs
->ti
->error
= "Invalid write_mostly index given";
1190 set_bit(WriteMostly
, &rs
->dev
[value
].rdev
.flags
);
1191 set_bit(__CTR_FLAG_WRITE_MOSTLY
, &rs
->ctr_flags
);
1192 } else if (!strcasecmp(key
, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND
))) {
1193 if (!rt_is_raid1(rt
)) {
1194 rs
->ti
->error
= "max_write_behind option is only valid for RAID1";
1198 if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND
, &rs
->ctr_flags
)) {
1199 rs
->ti
->error
= "Only one max_write_behind argument pair allowed";
1204 * In device-mapper, we specify things in sectors, but
1205 * MD records this value in kB
1208 if (value
> COUNTER_MAX
) {
1209 rs
->ti
->error
= "Max write-behind limit out of range";
1213 rs
->md
.bitmap_info
.max_write_behind
= value
;
1214 } else if (!strcasecmp(key
, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP
))) {
1215 if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP
, &rs
->ctr_flags
)) {
1216 rs
->ti
->error
= "Only one daemon_sleep argument pair allowed";
1219 if (!value
|| (value
> MAX_SCHEDULE_TIMEOUT
)) {
1220 rs
->ti
->error
= "daemon sleep period out of range";
1223 rs
->md
.bitmap_info
.daemon_sleep
= value
;
1224 } else if (!strcasecmp(key
, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET
))) {
1225 /* Userspace passes new data_offset after having extended the the data image LV */
1226 if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET
, &rs
->ctr_flags
)) {
1227 rs
->ti
->error
= "Only one data_offset argument pair allowed";
1230 /* Ensure sensible data offset */
1232 (value
&& (value
< MIN_FREE_RESHAPE_SPACE
|| value
% to_sector(PAGE_SIZE
)))) {
1233 rs
->ti
->error
= "Bogus data_offset value";
1236 rs
->data_offset
= value
;
1237 } else if (!strcasecmp(key
, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS
))) {
1238 /* Define the +/-# of disks to add to/remove from the given raid set */
1239 if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS
, &rs
->ctr_flags
)) {
1240 rs
->ti
->error
= "Only one delta_disks argument pair allowed";
1243 /* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1244 if (!__within_range(abs(value
), 1, MAX_RAID_DEVICES
- rt
->minimal_devs
)) {
1245 rs
->ti
->error
= "Too many delta_disk requested";
1249 rs
->delta_disks
= value
;
1250 } else if (!strcasecmp(key
, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE
))) {
1251 if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE
, &rs
->ctr_flags
)) {
1252 rs
->ti
->error
= "Only one stripe_cache argument pair allowed";
1256 if (!rt_is_raid456(rt
)) {
1257 rs
->ti
->error
= "Inappropriate argument: stripe_cache";
1261 rs
->stripe_cache_entries
= value
;
1262 } else if (!strcasecmp(key
, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE
))) {
1263 if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE
, &rs
->ctr_flags
)) {
1264 rs
->ti
->error
= "Only one min_recovery_rate argument pair allowed";
1267 if (value
> INT_MAX
) {
1268 rs
->ti
->error
= "min_recovery_rate out of range";
1271 rs
->md
.sync_speed_min
= (int)value
;
1272 } else if (!strcasecmp(key
, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE
))) {
1273 if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE
, &rs
->ctr_flags
)) {
1274 rs
->ti
->error
= "Only one max_recovery_rate argument pair allowed";
1277 if (value
> INT_MAX
) {
1278 rs
->ti
->error
= "max_recovery_rate out of range";
1281 rs
->md
.sync_speed_max
= (int)value
;
1282 } else if (!strcasecmp(key
, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE
))) {
1283 if (test_and_set_bit(__CTR_FLAG_REGION_SIZE
, &rs
->ctr_flags
)) {
1284 rs
->ti
->error
= "Only one region_size argument pair allowed";
1288 region_size
= value
;
1289 rs
->requested_bitmap_chunk_sectors
= value
;
1290 } else if (!strcasecmp(key
, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES
))) {
1291 if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES
, &rs
->ctr_flags
)) {
1292 rs
->ti
->error
= "Only one raid10_copies argument pair allowed";
1296 if (!__within_range(value
, 2, rs
->md
.raid_disks
)) {
1297 rs
->ti
->error
= "Bad value for 'raid10_copies'";
1301 raid10_copies
= value
;
1303 DMERR("Unable to parse RAID parameter: %s", key
);
1304 rs
->ti
->error
= "Unable to parse RAID parameter";
1309 if (test_bit(__CTR_FLAG_SYNC
, &rs
->ctr_flags
) &&
1310 test_bit(__CTR_FLAG_NOSYNC
, &rs
->ctr_flags
)) {
1311 rs
->ti
->error
= "sync and nosync are mutually exclusive";
1315 if (test_bit(__CTR_FLAG_REBUILD
, &rs
->ctr_flags
) &&
1316 (test_bit(__CTR_FLAG_SYNC
, &rs
->ctr_flags
) ||
1317 test_bit(__CTR_FLAG_NOSYNC
, &rs
->ctr_flags
))) {
1318 rs
->ti
->error
= "sync/nosync and rebuild are mutually exclusive";
1322 if (write_mostly
>= rs
->md
.raid_disks
) {
1323 rs
->ti
->error
= "Can't set all raid1 devices to write_mostly";
1327 if (validate_region_size(rs
, region_size
))
1330 if (rs
->md
.chunk_sectors
)
1331 max_io_len
= rs
->md
.chunk_sectors
;
1333 max_io_len
= region_size
;
1335 if (dm_set_target_max_io_len(rs
->ti
, max_io_len
))
1338 if (rt_is_raid10(rt
)) {
1339 if (raid10_copies
> rs
->md
.raid_disks
) {
1340 rs
->ti
->error
= "Not enough devices to satisfy specification";
1344 rs
->md
.new_layout
= raid10_format_to_md_layout(rs
, raid10_format
, raid10_copies
);
1345 if (rs
->md
.new_layout
< 0) {
1346 rs
->ti
->error
= "Error getting raid10 format";
1347 return rs
->md
.new_layout
;
1350 rt
= get_raid_type_by_ll(10, rs
->md
.new_layout
);
1352 rs
->ti
->error
= "Failed to recognize new raid10 layout";
1356 if ((rt
->algorithm
== ALGORITHM_RAID10_DEFAULT
||
1357 rt
->algorithm
== ALGORITHM_RAID10_NEAR
) &&
1358 test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS
, &rs
->ctr_flags
)) {
1359 rs
->ti
->error
= "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1364 rs
->raid10_copies
= raid10_copies
;
1366 /* Assume there are no metadata devices until the drives are parsed */
1367 rs
->md
.persistent
= 0;
1368 rs
->md
.external
= 1;
1370 /* Check, if any invalid ctr arguments have been passed in for the raid level */
1371 return rs_check_for_valid_flags(rs
);
1374 /* Set raid4/5/6 cache size */
1375 static int rs_set_raid456_stripe_cache(struct raid_set
*rs
)
1378 struct r5conf
*conf
;
1379 struct mddev
*mddev
= &rs
->md
;
1380 uint32_t min_stripes
= max(mddev
->chunk_sectors
, mddev
->new_chunk_sectors
) / 2;
1381 uint32_t nr_stripes
= rs
->stripe_cache_entries
;
1383 if (!rt_is_raid456(rs
->raid_type
)) {
1384 rs
->ti
->error
= "Inappropriate raid level; cannot change stripe_cache size";
1388 if (nr_stripes
< min_stripes
) {
1389 DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1390 nr_stripes
, min_stripes
);
1391 nr_stripes
= min_stripes
;
1394 conf
= mddev
->private;
1396 rs
->ti
->error
= "Cannot change stripe_cache size on inactive RAID set";
1400 /* Try setting number of stripes in raid456 stripe cache */
1401 if (conf
->min_nr_stripes
!= nr_stripes
) {
1402 r
= raid5_set_cache_size(mddev
, nr_stripes
);
1404 rs
->ti
->error
= "Failed to set raid4/5/6 stripe cache size";
1408 DMINFO("%u stripe cache entries", nr_stripes
);
1414 /* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1415 static unsigned int mddev_data_stripes(struct raid_set
*rs
)
1417 return rs
->md
.raid_disks
- rs
->raid_type
->parity_devs
;
1420 /* Return # of data stripes of @rs (i.e. as of ctr) */
1421 static unsigned int rs_data_stripes(struct raid_set
*rs
)
1423 return rs
->raid_disks
- rs
->raid_type
->parity_devs
;
1426 /* Calculate the sectors per device and per array used for @rs */
1427 static int rs_set_dev_and_array_sectors(struct raid_set
*rs
, bool use_mddev
)
1430 unsigned int data_stripes
;
1431 struct mddev
*mddev
= &rs
->md
;
1432 struct md_rdev
*rdev
;
1433 sector_t array_sectors
= rs
->ti
->len
, dev_sectors
= rs
->ti
->len
;
1436 delta_disks
= mddev
->delta_disks
;
1437 data_stripes
= mddev_data_stripes(rs
);
1439 delta_disks
= rs
->delta_disks
;
1440 data_stripes
= rs_data_stripes(rs
);
1443 /* Special raid1 case w/o delta_disks support (yet) */
1444 if (rt_is_raid1(rs
->raid_type
))
1446 else if (rt_is_raid10(rs
->raid_type
)) {
1447 if (rs
->raid10_copies
< 2 ||
1449 rs
->ti
->error
= "Bogus raid10 data copies or delta disks";
1453 dev_sectors
*= rs
->raid10_copies
;
1454 if (sector_div(dev_sectors
, data_stripes
))
1457 array_sectors
= (data_stripes
+ delta_disks
) * dev_sectors
;
1458 if (sector_div(array_sectors
, rs
->raid10_copies
))
1461 } else if (sector_div(dev_sectors
, data_stripes
))
1465 /* Striped layouts */
1466 array_sectors
= (data_stripes
+ delta_disks
) * dev_sectors
;
1468 rdev_for_each(rdev
, mddev
)
1469 rdev
->sectors
= dev_sectors
;
1471 mddev
->array_sectors
= array_sectors
;
1472 mddev
->dev_sectors
= dev_sectors
;
1476 rs
->ti
->error
= "Target length not divisible by number of data devices";
1480 /* Setup recovery on @rs */
1481 static void __rs_setup_recovery(struct raid_set
*rs
, sector_t dev_sectors
)
1483 /* raid0 does not recover */
1484 if (rs_is_raid0(rs
))
1485 rs
->md
.recovery_cp
= MaxSector
;
1487 * A raid6 set has to be recovered either
1488 * completely or for the grown part to
1489 * ensure proper parity and Q-Syndrome
1491 else if (rs_is_raid6(rs
))
1492 rs
->md
.recovery_cp
= dev_sectors
;
1494 * Other raid set types may skip recovery
1495 * depending on the 'nosync' flag.
1498 rs
->md
.recovery_cp
= test_bit(__CTR_FLAG_NOSYNC
, &rs
->ctr_flags
)
1499 ? MaxSector
: dev_sectors
;
1502 /* Setup recovery on @rs based on raid type, device size and 'nosync' flag */
1503 static void rs_setup_recovery(struct raid_set
*rs
, sector_t dev_sectors
)
1506 /* New raid set or 'sync' flag provided */
1507 __rs_setup_recovery(rs
, 0);
1508 else if (dev_sectors
== MaxSector
)
1509 /* Prevent recovery */
1510 __rs_setup_recovery(rs
, MaxSector
);
1511 else if (rs
->dev
[0].rdev
.sectors
< dev_sectors
)
1512 /* Grown raid set */
1513 __rs_setup_recovery(rs
, rs
->dev
[0].rdev
.sectors
);
1515 __rs_setup_recovery(rs
, MaxSector
);
1518 static void do_table_event(struct work_struct
*ws
)
1520 struct raid_set
*rs
= container_of(ws
, struct raid_set
, md
.event_work
);
1522 smp_rmb(); /* Make sure we access most actual mddev properties */
1523 if (!rs_is_reshaping(rs
))
1524 rs_set_capacity(rs
);
1525 dm_table_event(rs
->ti
->table
);
1528 static int raid_is_congested(struct dm_target_callbacks
*cb
, int bits
)
1530 struct raid_set
*rs
= container_of(cb
, struct raid_set
, callbacks
);
1532 return mddev_congested(&rs
->md
, bits
);
1536 * Make sure a valid takover (level switch) is being requested on @rs
1538 * Conversions of raid sets from one MD personality to another
1539 * have to conform to restrictions which are enforced here.
1541 static int rs_check_takeover(struct raid_set
*rs
)
1543 struct mddev
*mddev
= &rs
->md
;
1544 unsigned int near_copies
;
1546 if (rs
->md
.degraded
) {
1547 rs
->ti
->error
= "Can't takeover degraded raid set";
1551 if (rs_is_reshaping(rs
)) {
1552 rs
->ti
->error
= "Can't takeover reshaping raid set";
1556 switch (mddev
->level
) {
1558 /* raid0 -> raid1/5 with one disk */
1559 if ((mddev
->new_level
== 1 || mddev
->new_level
== 5) &&
1560 mddev
->raid_disks
== 1)
1563 /* raid0 -> raid10 */
1564 if (mddev
->new_level
== 10 &&
1565 !(rs
->raid_disks
% mddev
->raid_disks
))
1568 /* raid0 with multiple disks -> raid4/5/6 */
1569 if (__within_range(mddev
->new_level
, 4, 6) &&
1570 mddev
->new_layout
== ALGORITHM_PARITY_N
&&
1571 mddev
->raid_disks
> 1)
1577 /* Can't takeover raid10_offset! */
1578 if (__is_raid10_offset(mddev
->layout
))
1581 near_copies
= __raid10_near_copies(mddev
->layout
);
1583 /* raid10* -> raid0 */
1584 if (mddev
->new_level
== 0) {
1585 /* Can takeover raid10_near with raid disks divisable by data copies! */
1586 if (near_copies
> 1 &&
1587 !(mddev
->raid_disks
% near_copies
)) {
1588 mddev
->raid_disks
/= near_copies
;
1589 mddev
->delta_disks
= mddev
->raid_disks
;
1593 /* Can takeover raid10_far */
1594 if (near_copies
== 1 &&
1595 __raid10_far_copies(mddev
->layout
) > 1)
1601 /* raid10_{near,far} -> raid1 */
1602 if (mddev
->new_level
== 1 &&
1603 max(near_copies
, __raid10_far_copies(mddev
->layout
)) == mddev
->raid_disks
)
1606 /* raid10_{near,far} with 2 disks -> raid4/5 */
1607 if (__within_range(mddev
->new_level
, 4, 5) &&
1608 mddev
->raid_disks
== 2)
1613 /* raid1 with 2 disks -> raid4/5 */
1614 if (__within_range(mddev
->new_level
, 4, 5) &&
1615 mddev
->raid_disks
== 2) {
1616 mddev
->degraded
= 1;
1620 /* raid1 -> raid0 */
1621 if (mddev
->new_level
== 0 &&
1622 mddev
->raid_disks
== 1)
1625 /* raid1 -> raid10 */
1626 if (mddev
->new_level
== 10)
1631 /* raid4 -> raid0 */
1632 if (mddev
->new_level
== 0)
1635 /* raid4 -> raid1/5 with 2 disks */
1636 if ((mddev
->new_level
== 1 || mddev
->new_level
== 5) &&
1637 mddev
->raid_disks
== 2)
1640 /* raid4 -> raid5/6 with parity N */
1641 if (__within_range(mddev
->new_level
, 5, 6) &&
1642 mddev
->layout
== ALGORITHM_PARITY_N
)
1647 /* raid5 with parity N -> raid0 */
1648 if (mddev
->new_level
== 0 &&
1649 mddev
->layout
== ALGORITHM_PARITY_N
)
1652 /* raid5 with parity N -> raid4 */
1653 if (mddev
->new_level
== 4 &&
1654 mddev
->layout
== ALGORITHM_PARITY_N
)
1657 /* raid5 with 2 disks -> raid1/4/10 */
1658 if ((mddev
->new_level
== 1 || mddev
->new_level
== 4 || mddev
->new_level
== 10) &&
1659 mddev
->raid_disks
== 2)
1662 /* raid5_* -> raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
1663 if (mddev
->new_level
== 6 &&
1664 ((mddev
->layout
== ALGORITHM_PARITY_N
&& mddev
->new_layout
== ALGORITHM_PARITY_N
) ||
1665 __within_range(mddev
->new_layout
, ALGORITHM_LEFT_ASYMMETRIC_6
, ALGORITHM_RIGHT_SYMMETRIC_6
)))
1670 /* raid6 with parity N -> raid0 */
1671 if (mddev
->new_level
== 0 &&
1672 mddev
->layout
== ALGORITHM_PARITY_N
)
1675 /* raid6 with parity N -> raid4 */
1676 if (mddev
->new_level
== 4 &&
1677 mddev
->layout
== ALGORITHM_PARITY_N
)
1680 /* raid6_*_n with Q-Syndrome N -> raid5_* */
1681 if (mddev
->new_level
== 5 &&
1682 ((mddev
->layout
== ALGORITHM_PARITY_N
&& mddev
->new_layout
== ALGORITHM_PARITY_N
) ||
1683 __within_range(mddev
->new_layout
, ALGORITHM_LEFT_ASYMMETRIC
, ALGORITHM_RIGHT_SYMMETRIC
)))
1690 rs
->ti
->error
= "takeover not possible";
1694 /* True if @rs requested to be taken over */
1695 static bool rs_takeover_requested(struct raid_set
*rs
)
1697 return rs
->md
.new_level
!= rs
->md
.level
;
1700 /* True if @rs is requested to reshape by ctr */
1701 static bool rs_reshape_requested(struct raid_set
*rs
)
1704 struct mddev
*mddev
= &rs
->md
;
1706 if (rs_takeover_requested(rs
))
1712 change
= mddev
->new_layout
!= mddev
->layout
||
1713 mddev
->new_chunk_sectors
!= mddev
->chunk_sectors
||
1716 /* Historical case to support raid1 reshape without delta disks */
1717 if (mddev
->level
== 1) {
1718 if (rs
->delta_disks
)
1719 return !!rs
->delta_disks
;
1722 mddev
->raid_disks
!= rs
->raid_disks
;
1725 if (mddev
->level
== 10)
1727 !__is_raid10_far(mddev
->new_layout
) &&
1728 rs
->delta_disks
>= 0;
1734 #define FEATURE_FLAG_SUPPORTS_V190 0x1 /* Supports extended superblock */
1736 /* State flags for sb->flags */
1737 #define SB_FLAG_RESHAPE_ACTIVE 0x1
1738 #define SB_FLAG_RESHAPE_BACKWARDS 0x2
1741 * This structure is never routinely used by userspace, unlike md superblocks.
1742 * Devices with this superblock should only ever be accessed via device-mapper.
1744 #define DM_RAID_MAGIC 0x64526D44
1745 struct dm_raid_superblock
{
1746 __le32 magic
; /* "DmRd" */
1747 __le32 compat_features
; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1749 __le32 num_devices
; /* Number of devices in this raid set. (Max 64) */
1750 __le32 array_position
; /* The position of this drive in the raid set */
1752 __le64 events
; /* Incremented by md when superblock updated */
1753 __le64 failed_devices
; /* Pre 1.9.0 part of bit field of devices to */
1754 /* indicate failures (see extension below) */
1757 * This offset tracks the progress of the repair or replacement of
1758 * an individual drive.
1760 __le64 disk_recovery_offset
;
1763 * This offset tracks the progress of the initial raid set
1764 * synchronisation/parity calculation.
1766 __le64 array_resync_offset
;
1769 * raid characteristics
1773 __le32 stripe_sectors
;
1775 /********************************************************************
1776 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1778 * FEATURE_FLAG_SUPPORTS_V190 in the features member indicates that those exist
1781 __le32 flags
; /* Flags defining array states for reshaping */
1784 * This offset tracks the progress of a raid
1785 * set reshape in order to be able to restart it
1787 __le64 reshape_position
;
1790 * These define the properties of the array in case of an interrupted reshape
1794 __le32 new_stripe_sectors
;
1797 __le64 array_sectors
; /* Array size in sectors */
1800 * Sector offsets to data on devices (reshaping).
1801 * Needed to support out of place reshaping, thus
1802 * not writing over any stripes whilst converting
1803 * them from old to new layout
1806 __le64 new_data_offset
;
1808 __le64 sectors
; /* Used device size in sectors */
1811 * Additonal Bit field of devices indicating failures to support
1812 * up to 256 devices with the 1.9.0 on-disk metadata format
1814 __le64 extended_failed_devices
[DISKS_ARRAY_ELEMS
- 1];
1816 __le32 incompat_features
; /* Used to indicate any incompatible features */
1818 /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
1822 * Check for reshape constraints on raid set @rs:
1824 * - reshape function non-existent
1826 * - ongoing recovery
1829 * Returns 0 if none or -EPERM if given constraint
1830 * and error message reference in @errmsg
1832 static int rs_check_reshape(struct raid_set
*rs
)
1834 struct mddev
*mddev
= &rs
->md
;
1836 if (!mddev
->pers
|| !mddev
->pers
->check_reshape
)
1837 rs
->ti
->error
= "Reshape not supported";
1838 else if (mddev
->degraded
)
1839 rs
->ti
->error
= "Can't reshape degraded raid set";
1840 else if (rs_is_recovering(rs
))
1841 rs
->ti
->error
= "Convert request on recovering raid set prohibited";
1842 else if (rs_is_reshaping(rs
))
1843 rs
->ti
->error
= "raid set already reshaping!";
1844 else if (!(rs_is_raid1(rs
) || rs_is_raid10(rs
) || rs_is_raid456(rs
)))
1845 rs
->ti
->error
= "Reshaping only supported for raid1/4/5/6/10";
1852 static int read_disk_sb(struct md_rdev
*rdev
, int size
)
1854 BUG_ON(!rdev
->sb_page
);
1856 if (rdev
->sb_loaded
)
1859 if (!sync_page_io(rdev
, 0, size
, rdev
->sb_page
, REQ_OP_READ
, 0, true)) {
1860 DMERR("Failed to read superblock of device at position %d",
1862 md_error(rdev
->mddev
, rdev
);
1866 rdev
->sb_loaded
= 1;
1871 static void sb_retrieve_failed_devices(struct dm_raid_superblock
*sb
, uint64_t *failed_devices
)
1873 failed_devices
[0] = le64_to_cpu(sb
->failed_devices
);
1874 memset(failed_devices
+ 1, 0, sizeof(sb
->extended_failed_devices
));
1876 if (le32_to_cpu(sb
->compat_features
) & FEATURE_FLAG_SUPPORTS_V190
) {
1877 int i
= ARRAY_SIZE(sb
->extended_failed_devices
);
1880 failed_devices
[i
+1] = le64_to_cpu(sb
->extended_failed_devices
[i
]);
1884 static void sb_update_failed_devices(struct dm_raid_superblock
*sb
, uint64_t *failed_devices
)
1886 int i
= ARRAY_SIZE(sb
->extended_failed_devices
);
1888 sb
->failed_devices
= cpu_to_le64(failed_devices
[0]);
1890 sb
->extended_failed_devices
[i
] = cpu_to_le64(failed_devices
[i
+1]);
1894 * Synchronize the superblock members with the raid set properties
1896 * All superblock data is little endian.
1898 static void super_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1900 bool update_failed_devices
= false;
1902 uint64_t failed_devices
[DISKS_ARRAY_ELEMS
];
1903 struct dm_raid_superblock
*sb
;
1904 struct raid_set
*rs
= container_of(mddev
, struct raid_set
, md
);
1906 /* No metadata device, no superblock */
1907 if (!rdev
->meta_bdev
)
1910 BUG_ON(!rdev
->sb_page
);
1912 sb
= page_address(rdev
->sb_page
);
1914 sb_retrieve_failed_devices(sb
, failed_devices
);
1916 for (i
= 0; i
< rs
->raid_disks
; i
++)
1917 if (!rs
->dev
[i
].data_dev
|| test_bit(Faulty
, &rs
->dev
[i
].rdev
.flags
)) {
1918 update_failed_devices
= true;
1919 set_bit(i
, (void *) failed_devices
);
1922 if (update_failed_devices
)
1923 sb_update_failed_devices(sb
, failed_devices
);
1925 sb
->magic
= cpu_to_le32(DM_RAID_MAGIC
);
1926 sb
->compat_features
= cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190
);
1928 sb
->num_devices
= cpu_to_le32(mddev
->raid_disks
);
1929 sb
->array_position
= cpu_to_le32(rdev
->raid_disk
);
1931 sb
->events
= cpu_to_le64(mddev
->events
);
1933 sb
->disk_recovery_offset
= cpu_to_le64(rdev
->recovery_offset
);
1934 sb
->array_resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1936 sb
->level
= cpu_to_le32(mddev
->level
);
1937 sb
->layout
= cpu_to_le32(mddev
->layout
);
1938 sb
->stripe_sectors
= cpu_to_le32(mddev
->chunk_sectors
);
1940 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1941 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1942 sb
->new_stripe_sectors
= cpu_to_le32(mddev
->new_chunk_sectors
);
1944 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1946 smp_rmb(); /* Make sure we access most recent reshape position */
1947 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1948 if (le64_to_cpu(sb
->reshape_position
) != MaxSector
) {
1949 /* Flag ongoing reshape */
1950 sb
->flags
|= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE
);
1952 if (mddev
->delta_disks
< 0 || mddev
->reshape_backwards
)
1953 sb
->flags
|= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS
);
1955 /* Clear reshape flags */
1956 sb
->flags
&= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE
|SB_FLAG_RESHAPE_BACKWARDS
));
1959 sb
->array_sectors
= cpu_to_le64(mddev
->array_sectors
);
1960 sb
->data_offset
= cpu_to_le64(rdev
->data_offset
);
1961 sb
->new_data_offset
= cpu_to_le64(rdev
->new_data_offset
);
1962 sb
->sectors
= cpu_to_le64(rdev
->sectors
);
1964 /* Zero out the rest of the payload after the size of the superblock */
1965 memset(sb
+ 1, 0, rdev
->sb_size
- sizeof(*sb
));
1971 * This function creates a superblock if one is not found on the device
1972 * and will decide which superblock to use if there's a choice.
1974 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
1976 static int super_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
)
1979 struct dm_raid_superblock
*sb
;
1980 struct dm_raid_superblock
*refsb
;
1981 uint64_t events_sb
, events_refsb
;
1984 rdev
->sb_size
= bdev_logical_block_size(rdev
->meta_bdev
);
1985 if (rdev
->sb_size
< sizeof(*sb
) || rdev
->sb_size
> PAGE_SIZE
) {
1986 DMERR("superblock size of a logical block is no longer valid");
1990 r
= read_disk_sb(rdev
, rdev
->sb_size
);
1994 sb
= page_address(rdev
->sb_page
);
1997 * Two cases that we want to write new superblocks and rebuild:
1998 * 1) New device (no matching magic number)
1999 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2001 if ((sb
->magic
!= cpu_to_le32(DM_RAID_MAGIC
)) ||
2002 (!test_bit(In_sync
, &rdev
->flags
) && !rdev
->recovery_offset
)) {
2003 super_sync(rdev
->mddev
, rdev
);
2005 set_bit(FirstUse
, &rdev
->flags
);
2006 sb
->compat_features
= cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190
);
2008 /* Force writing of superblocks to disk */
2009 set_bit(MD_CHANGE_DEVS
, &rdev
->mddev
->flags
);
2011 /* Any superblock is better than none, choose that if given */
2012 return refdev
? 0 : 1;
2018 events_sb
= le64_to_cpu(sb
->events
);
2020 refsb
= page_address(refdev
->sb_page
);
2021 events_refsb
= le64_to_cpu(refsb
->events
);
2023 return (events_sb
> events_refsb
) ? 1 : 0;
2026 static int super_init_validation(struct raid_set
*rs
, struct md_rdev
*rdev
)
2030 struct mddev
*mddev
= &rs
->md
;
2032 uint64_t failed_devices
[DISKS_ARRAY_ELEMS
];
2033 struct dm_raid_superblock
*sb
;
2034 uint32_t new_devs
= 0, rebuild_and_new
= 0, rebuilds
= 0;
2036 struct dm_raid_superblock
*sb2
;
2038 sb
= page_address(rdev
->sb_page
);
2039 events_sb
= le64_to_cpu(sb
->events
);
2042 * Initialise to 1 if this is a new superblock.
2044 mddev
->events
= events_sb
? : 1;
2046 mddev
->reshape_position
= MaxSector
;
2049 * Reshaping is supported, e.g. reshape_position is valid
2050 * in superblock and superblock content is authoritative.
2052 if (le32_to_cpu(sb
->compat_features
) & FEATURE_FLAG_SUPPORTS_V190
) {
2053 /* Superblock is authoritative wrt given raid set layout! */
2054 mddev
->raid_disks
= le32_to_cpu(sb
->num_devices
);
2055 mddev
->level
= le32_to_cpu(sb
->level
);
2056 mddev
->layout
= le32_to_cpu(sb
->layout
);
2057 mddev
->chunk_sectors
= le32_to_cpu(sb
->stripe_sectors
);
2058 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
2059 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
2060 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_stripe_sectors
);
2061 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
2062 mddev
->array_sectors
= le64_to_cpu(sb
->array_sectors
);
2064 /* raid was reshaping and got interrupted */
2065 if (le32_to_cpu(sb
->flags
) & SB_FLAG_RESHAPE_ACTIVE
) {
2066 if (test_bit(__CTR_FLAG_DELTA_DISKS
, &rs
->ctr_flags
)) {
2067 DMERR("Reshape requested but raid set is still reshaping");
2071 if (mddev
->delta_disks
< 0 ||
2072 (!mddev
->delta_disks
&& (le32_to_cpu(sb
->flags
) & SB_FLAG_RESHAPE_BACKWARDS
)))
2073 mddev
->reshape_backwards
= 1;
2075 mddev
->reshape_backwards
= 0;
2077 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
2078 rs
->raid_type
= get_raid_type_by_ll(mddev
->level
, mddev
->layout
);
2083 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2085 if (le32_to_cpu(sb
->level
) != mddev
->level
) {
2086 DMERR("Reshaping/takeover raid sets not yet supported. (raid level/stripes/size change)");
2089 if (le32_to_cpu(sb
->layout
) != mddev
->layout
) {
2090 DMERR("Reshaping raid sets not yet supported. (raid layout change)");
2091 DMERR(" 0x%X vs 0x%X", le32_to_cpu(sb
->layout
), mddev
->layout
);
2092 DMERR(" Old layout: %s w/ %d copies",
2093 raid10_md_layout_to_format(le32_to_cpu(sb
->layout
)),
2094 raid10_md_layout_to_copies(le32_to_cpu(sb
->layout
)));
2095 DMERR(" New layout: %s w/ %d copies",
2096 raid10_md_layout_to_format(mddev
->layout
),
2097 raid10_md_layout_to_copies(mddev
->layout
));
2100 if (le32_to_cpu(sb
->stripe_sectors
) != mddev
->chunk_sectors
) {
2101 DMERR("Reshaping raid sets not yet supported. (stripe sectors change)");
2105 /* We can only change the number of devices in raid1 with old (i.e. pre 1.0.7) metadata */
2106 if (!rt_is_raid1(rs
->raid_type
) &&
2107 (le32_to_cpu(sb
->num_devices
) != mddev
->raid_disks
)) {
2108 DMERR("Reshaping raid sets not yet supported. (device count change from %u to %u)",
2109 sb
->num_devices
, mddev
->raid_disks
);
2113 /* Table line is checked vs. authoritative superblock */
2117 if (!test_bit(__CTR_FLAG_NOSYNC
, &rs
->ctr_flags
))
2118 mddev
->recovery_cp
= le64_to_cpu(sb
->array_resync_offset
);
2121 * During load, we set FirstUse if a new superblock was written.
2122 * There are two reasons we might not have a superblock:
2123 * 1) The raid set is brand new - in which case, all of the
2124 * devices must have their In_sync bit set. Also,
2125 * recovery_cp must be 0, unless forced.
2126 * 2) This is a new device being added to an old raid set
2127 * and the new device needs to be rebuilt - in which
2128 * case the In_sync bit will /not/ be set and
2129 * recovery_cp must be MaxSector.
2130 * 3) This is/are a new device(s) being added to an old
2131 * raid set during takeover to a higher raid level
2132 * to provide capacity for redundancy or during reshape
2133 * to add capacity to grow the raid set.
2136 rdev_for_each(r
, mddev
) {
2137 if (test_bit(FirstUse
, &r
->flags
))
2140 if (!test_bit(In_sync
, &r
->flags
)) {
2141 DMINFO("Device %d specified for rebuild; clearing superblock",
2145 if (test_bit(FirstUse
, &r
->flags
))
2152 if (new_devs
== rs
->raid_disks
|| !rebuilds
) {
2153 /* Replace a broken device */
2154 if (new_devs
== 1 && !rs
->delta_disks
)
2156 if (new_devs
== rs
->raid_disks
) {
2157 DMINFO("Superblocks created for new raid set");
2158 set_bit(MD_ARRAY_FIRST_USE
, &mddev
->flags
);
2159 } else if (new_devs
!= rebuilds
&&
2160 new_devs
!= rs
->delta_disks
) {
2161 DMERR("New device injected into existing raid set without "
2162 "'delta_disks' or 'rebuild' parameter specified");
2165 } else if (new_devs
&& new_devs
!= rebuilds
) {
2166 DMERR("%u 'rebuild' devices cannot be injected into"
2167 " a raid set with %u other first-time devices",
2168 rebuilds
, new_devs
);
2170 } else if (rebuilds
) {
2171 if (rebuild_and_new
&& rebuilds
!= rebuild_and_new
) {
2172 DMERR("new device%s provided without 'rebuild'",
2173 new_devs
> 1 ? "s" : "");
2175 } else if (rs_is_recovering(rs
)) {
2176 DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2177 (unsigned long long) mddev
->recovery_cp
);
2179 } else if (rs_is_reshaping(rs
)) {
2180 DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2181 (unsigned long long) mddev
->reshape_position
);
2187 * Now we set the Faulty bit for those devices that are
2188 * recorded in the superblock as failed.
2190 sb_retrieve_failed_devices(sb
, failed_devices
);
2191 rdev_for_each(r
, mddev
) {
2194 sb2
= page_address(r
->sb_page
);
2195 sb2
->failed_devices
= 0;
2196 memset(sb2
->extended_failed_devices
, 0, sizeof(sb2
->extended_failed_devices
));
2199 * Check for any device re-ordering.
2201 if (!test_bit(FirstUse
, &r
->flags
) && (r
->raid_disk
>= 0)) {
2202 role
= le32_to_cpu(sb2
->array_position
);
2206 if (role
!= r
->raid_disk
) {
2207 if (__is_raid10_near(mddev
->layout
)) {
2208 if (mddev
->raid_disks
% __raid10_near_copies(mddev
->layout
) ||
2209 rs
->raid_disks
% rs
->raid10_copies
) {
2211 "Cannot change raid10 near set to odd # of devices!";
2215 sb2
->array_position
= cpu_to_le32(r
->raid_disk
);
2217 } else if (!(rs_is_raid10(rs
) && rt_is_raid0(rs
->raid_type
)) &&
2218 !(rs_is_raid0(rs
) && rt_is_raid10(rs
->raid_type
)) &&
2219 !rt_is_raid1(rs
->raid_type
)) {
2220 rs
->ti
->error
= "Cannot change device positions in raid set";
2224 DMINFO("raid device #%d now at position #%d", role
, r
->raid_disk
);
2228 * Partial recovery is performed on
2229 * returning failed devices.
2231 if (test_bit(role
, (void *) failed_devices
))
2232 set_bit(Faulty
, &r
->flags
);
2239 static int super_validate(struct raid_set
*rs
, struct md_rdev
*rdev
)
2241 struct mddev
*mddev
= &rs
->md
;
2242 struct dm_raid_superblock
*sb
;
2244 if (rs_is_raid0(rs
) || !rdev
->sb_page
)
2247 sb
= page_address(rdev
->sb_page
);
2250 * If mddev->events is not set, we know we have not yet initialized
2253 if (!mddev
->events
&& super_init_validation(rs
, rdev
))
2256 if (le32_to_cpu(sb
->compat_features
) != FEATURE_FLAG_SUPPORTS_V190
) {
2257 rs
->ti
->error
= "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2261 if (sb
->incompat_features
) {
2262 rs
->ti
->error
= "Unable to assemble array: No incompatible feature flags supported yet";
2266 /* Enable bitmap creation for RAID levels != 0 */
2267 mddev
->bitmap_info
.offset
= rt_is_raid0(rs
->raid_type
) ? 0 : to_sector(4096);
2268 rdev
->mddev
->bitmap_info
.default_offset
= mddev
->bitmap_info
.offset
;
2270 if (!test_and_clear_bit(FirstUse
, &rdev
->flags
)) {
2271 /* Retrieve device size stored in superblock to be prepared for shrink */
2272 rdev
->sectors
= le64_to_cpu(sb
->sectors
);
2273 rdev
->recovery_offset
= le64_to_cpu(sb
->disk_recovery_offset
);
2274 if (rdev
->recovery_offset
== MaxSector
)
2275 set_bit(In_sync
, &rdev
->flags
);
2277 * If no reshape in progress -> we're recovering single
2278 * disk(s) and have to set the device(s) to out-of-sync
2280 else if (!rs_is_reshaping(rs
))
2281 clear_bit(In_sync
, &rdev
->flags
); /* Mandatory for recovery */
2285 * If a device comes back, set it as not In_sync and no longer faulty.
2287 if (test_and_clear_bit(Faulty
, &rdev
->flags
)) {
2288 rdev
->recovery_offset
= 0;
2289 clear_bit(In_sync
, &rdev
->flags
);
2290 rdev
->saved_raid_disk
= rdev
->raid_disk
;
2293 /* Reshape support -> restore repective data offsets */
2294 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
2295 rdev
->new_data_offset
= le64_to_cpu(sb
->new_data_offset
);
2301 * Analyse superblocks and select the freshest.
2303 static int analyse_superblocks(struct dm_target
*ti
, struct raid_set
*rs
)
2306 struct raid_dev
*dev
;
2307 struct md_rdev
*rdev
, *tmp
, *freshest
;
2308 struct mddev
*mddev
= &rs
->md
;
2311 rdev_for_each_safe(rdev
, tmp
, mddev
) {
2313 * Skipping super_load due to CTR_FLAG_SYNC will cause
2314 * the array to undergo initialization again as
2315 * though it were new. This is the intended effect
2316 * of the "sync" directive.
2318 * When reshaping capability is added, we must ensure
2319 * that the "sync" directive is disallowed during the
2322 if (test_bit(__CTR_FLAG_SYNC
, &rs
->ctr_flags
))
2325 if (!rdev
->meta_bdev
)
2328 r
= super_load(rdev
, freshest
);
2337 dev
= container_of(rdev
, struct raid_dev
, rdev
);
2339 dm_put_device(ti
, dev
->meta_dev
);
2341 dev
->meta_dev
= NULL
;
2342 rdev
->meta_bdev
= NULL
;
2345 put_page(rdev
->sb_page
);
2347 rdev
->sb_page
= NULL
;
2349 rdev
->sb_loaded
= 0;
2352 * We might be able to salvage the data device
2353 * even though the meta device has failed. For
2354 * now, we behave as though '- -' had been
2355 * set for this device in the table.
2358 dm_put_device(ti
, dev
->data_dev
);
2360 dev
->data_dev
= NULL
;
2363 list_del(&rdev
->same_set
);
2370 if (validate_raid_redundancy(rs
)) {
2371 rs
->ti
->error
= "Insufficient redundancy to activate array";
2376 * Validation of the freshest device provides the source of
2377 * validation for the remaining devices.
2379 rs
->ti
->error
= "Unable to assemble array: Invalid superblocks";
2380 if (super_validate(rs
, freshest
))
2383 rdev_for_each(rdev
, mddev
)
2384 if ((rdev
!= freshest
) && super_validate(rs
, rdev
))
2390 * Adjust data_offset and new_data_offset on all disk members of @rs
2391 * for out of place reshaping if requested by contructor
2393 * We need free space at the beginning of each raid disk for forward
2394 * and at the end for backward reshapes which userspace has to provide
2395 * via remapping/reordering of space.
2397 static int rs_adjust_data_offsets(struct raid_set
*rs
)
2399 sector_t data_offset
= 0, new_data_offset
= 0;
2400 struct md_rdev
*rdev
;
2402 /* Constructor did not request data offset change */
2403 if (!test_bit(__CTR_FLAG_DATA_OFFSET
, &rs
->ctr_flags
)) {
2404 if (!rs_is_reshapable(rs
))
2410 /* HM FIXME: get InSync raid_dev? */
2411 rdev
= &rs
->dev
[0].rdev
;
2413 if (rs
->delta_disks
< 0) {
2415 * Removing disks (reshaping backwards):
2417 * - before reshape: data is at offset 0 and free space
2418 * is at end of each component LV
2420 * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2423 new_data_offset
= rs
->data_offset
;
2425 } else if (rs
->delta_disks
> 0) {
2427 * Adding disks (reshaping forwards):
2429 * - before reshape: data is at offset rs->data_offset != 0 and
2430 * free space is at begin of each component LV
2432 * - after reshape: data is at offset 0 on each component LV
2434 data_offset
= rs
->data_offset
;
2435 new_data_offset
= 0;
2439 * User space passes in 0 for data offset after having removed reshape space
2441 * - or - (data offset != 0)
2443 * Changing RAID layout or chunk size -> toggle offsets
2445 * - before reshape: data is at offset rs->data_offset 0 and
2446 * free space is at end of each component LV
2448 * data is at offset rs->data_offset != 0 and
2449 * free space is at begin of each component LV
2451 * - after reshape: data is at offset 0 if it was at offset != 0
2452 * or at offset != 0 if it was at offset 0
2453 * on each component LV
2456 data_offset
= rs
->data_offset
? rdev
->data_offset
: 0;
2457 new_data_offset
= data_offset
? 0 : rs
->data_offset
;
2458 set_bit(RT_FLAG_UPDATE_SBS
, &rs
->runtime_flags
);
2462 * Make sure we got a minimum amount of free sectors per device
2464 if (rs
->data_offset
&&
2465 to_sector(i_size_read(rdev
->bdev
->bd_inode
)) - rdev
->sectors
< MIN_FREE_RESHAPE_SPACE
) {
2466 rs
->ti
->error
= data_offset
? "No space for forward reshape" :
2467 "No space for backward reshape";
2471 /* Adjust data offsets on all rdevs */
2472 rdev_for_each(rdev
, &rs
->md
) {
2473 rdev
->data_offset
= data_offset
;
2474 rdev
->new_data_offset
= new_data_offset
;
2480 /* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2481 static void __reorder_raid_disk_indexes(struct raid_set
*rs
)
2484 struct md_rdev
*rdev
;
2486 rdev_for_each(rdev
, &rs
->md
) {
2487 rdev
->raid_disk
= i
++;
2488 rdev
->saved_raid_disk
= rdev
->new_raid_disk
= -1;
2493 * Setup @rs for takeover by a different raid level
2495 static int rs_setup_takeover(struct raid_set
*rs
)
2497 struct mddev
*mddev
= &rs
->md
;
2498 struct md_rdev
*rdev
;
2499 unsigned int d
= mddev
->raid_disks
= rs
->raid_disks
;
2500 sector_t new_data_offset
= rs
->dev
[0].rdev
.data_offset
? 0 : rs
->data_offset
;
2502 if (rt_is_raid10(rs
->raid_type
)) {
2503 if (mddev
->level
== 0) {
2504 /* Userpace reordered disks -> adjust raid_disk indexes */
2505 __reorder_raid_disk_indexes(rs
);
2507 /* raid0 -> raid10_far layout */
2508 mddev
->layout
= raid10_format_to_md_layout(rs
, ALGORITHM_RAID10_FAR
,
2510 } else if (mddev
->level
== 1)
2511 /* raid1 -> raid10_near layout */
2512 mddev
->layout
= raid10_format_to_md_layout(rs
, ALGORITHM_RAID10_NEAR
,
2519 clear_bit(MD_ARRAY_FIRST_USE
, &mddev
->flags
);
2520 mddev
->recovery_cp
= MaxSector
;
2523 rdev
= &rs
->dev
[d
].rdev
;
2525 if (test_bit(d
, (void *) rs
->rebuild_disks
)) {
2526 clear_bit(In_sync
, &rdev
->flags
);
2527 clear_bit(Faulty
, &rdev
->flags
);
2528 mddev
->recovery_cp
= rdev
->recovery_offset
= 0;
2529 /* Bitmap has to be created when we do an "up" takeover */
2530 set_bit(MD_ARRAY_FIRST_USE
, &mddev
->flags
);
2533 rdev
->new_data_offset
= new_data_offset
;
2539 /* Prepare @rs for reshape */
2540 static int rs_prepare_reshape(struct raid_set
*rs
)
2543 struct mddev
*mddev
= &rs
->md
;
2545 if (rs_is_raid10(rs
)) {
2546 if (rs
->raid_disks
!= mddev
->raid_disks
&&
2547 __is_raid10_near(mddev
->layout
) &&
2548 rs
->raid10_copies
&&
2549 rs
->raid10_copies
!= __raid10_near_copies(mddev
->layout
)) {
2551 * raid disk have to be multiple of data copies to allow this conversion,
2553 * This is actually not a reshape it is a
2554 * rebuild of any additional mirrors per group
2556 if (rs
->raid_disks
% rs
->raid10_copies
) {
2557 rs
->ti
->error
= "Can't reshape raid10 mirror groups";
2561 /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2562 __reorder_raid_disk_indexes(rs
);
2563 mddev
->layout
= raid10_format_to_md_layout(rs
, ALGORITHM_RAID10_NEAR
,
2565 mddev
->new_layout
= mddev
->layout
;
2570 } else if (rs_is_raid456(rs
))
2573 else if (rs_is_raid1(rs
)) {
2574 if (rs
->delta_disks
) {
2575 /* Process raid1 via delta_disks */
2576 mddev
->degraded
= rs
->delta_disks
< 0 ? -rs
->delta_disks
: rs
->delta_disks
;
2579 /* Process raid1 without delta_disks */
2580 mddev
->raid_disks
= rs
->raid_disks
;
2581 set_bit(RT_FLAG_KEEP_RS_FROZEN
, &rs
->runtime_flags
);
2585 rs
->ti
->error
= "Called with bogus raid type";
2590 set_bit(RT_FLAG_RESHAPE_RS
, &rs
->runtime_flags
);
2591 set_bit(RT_FLAG_UPDATE_SBS
, &rs
->runtime_flags
);
2592 set_bit(RT_FLAG_KEEP_RS_FROZEN
, &rs
->runtime_flags
);
2593 } else if (mddev
->raid_disks
< rs
->raid_disks
)
2594 /* Create new superblocks and bitmaps, if any new disks */
2595 set_bit(RT_FLAG_UPDATE_SBS
, &rs
->runtime_flags
);
2602 * - change raid layout
2603 * - change chunk size
2607 static int rs_setup_reshape(struct raid_set
*rs
)
2610 unsigned int cur_raid_devs
, d
;
2611 struct mddev
*mddev
= &rs
->md
;
2612 struct md_rdev
*rdev
;
2614 mddev
->delta_disks
= rs
->delta_disks
;
2615 cur_raid_devs
= mddev
->raid_disks
;
2617 /* Ignore impossible layout change whilst adding/removing disks */
2618 if (mddev
->delta_disks
&&
2619 mddev
->layout
!= mddev
->new_layout
) {
2620 DMINFO("Ignoring invalid layout change with delta_disks=%d", rs
->delta_disks
);
2621 mddev
->new_layout
= mddev
->layout
;
2625 * Adjust array size:
2627 * - in case of adding disks, array size has
2628 * to grow after the disk adding reshape,
2629 * which'll hapen in the event handler;
2630 * reshape will happen forward, so space has to
2631 * be available at the beginning of each disk
2633 * - in case of removing disks, array size
2634 * has to shrink before starting the reshape,
2635 * which'll happen here;
2636 * reshape will happen backward, so space has to
2637 * be available at the end of each disk
2639 * - data_offset and new_data_offset are
2640 * adjusted for aforementioned out of place
2641 * reshaping based on userspace passing in
2642 * the "data_offset <sectors>" key/value
2643 * pair via the constructor
2647 if (rs
->delta_disks
> 0) {
2648 /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2649 for (d
= cur_raid_devs
; d
< rs
->raid_disks
; d
++) {
2650 rdev
= &rs
->dev
[d
].rdev
;
2651 clear_bit(In_sync
, &rdev
->flags
);
2654 * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2655 * by md, which'll store that erroneously in the superblock on reshape
2657 rdev
->saved_raid_disk
= -1;
2658 rdev
->raid_disk
= d
;
2660 rdev
->sectors
= mddev
->dev_sectors
;
2661 rdev
->recovery_offset
= rs_is_raid1(rs
) ? 0 : MaxSector
;
2664 mddev
->reshape_backwards
= 0; /* adding disks -> forward reshape */
2666 /* Remove disk(s) */
2667 } else if (rs
->delta_disks
< 0) {
2668 r
= rs_set_dev_and_array_sectors(rs
, true);
2669 mddev
->reshape_backwards
= 1; /* removing disk(s) -> backward reshape */
2671 /* Change layout and/or chunk size */
2674 * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2676 * keeping number of disks and do layout change ->
2678 * toggle reshape_backward depending on data_offset:
2680 * - free space upfront -> reshape forward
2682 * - free space at the end -> reshape backward
2685 * This utilizes free reshape space avoiding the need
2686 * for userspace to move (parts of) LV segments in
2687 * case of layout/chunksize change (for disk
2688 * adding/removing reshape space has to be at
2689 * the proper address (see above with delta_disks):
2691 * add disk(s) -> begin
2692 * remove disk(s)-> end
2694 mddev
->reshape_backwards
= rs
->dev
[0].rdev
.data_offset
? 0 : 1;
2701 * Enable/disable discard support on RAID set depending on
2702 * RAID level and discard properties of underlying RAID members.
2704 static void configure_discard_support(struct raid_set
*rs
)
2708 struct dm_target
*ti
= rs
->ti
;
2710 /* Assume discards not supported until after checks below. */
2711 ti
->discards_supported
= false;
2713 /* RAID level 4,5,6 require discard_zeroes_data for data integrity! */
2714 raid456
= (rs
->md
.level
== 4 || rs
->md
.level
== 5 || rs
->md
.level
== 6);
2716 for (i
= 0; i
< rs
->raid_disks
; i
++) {
2717 struct request_queue
*q
;
2719 if (!rs
->dev
[i
].rdev
.bdev
)
2722 q
= bdev_get_queue(rs
->dev
[i
].rdev
.bdev
);
2723 if (!q
|| !blk_queue_discard(q
))
2727 if (!q
->limits
.discard_zeroes_data
)
2729 if (!devices_handle_discard_safely
) {
2730 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2731 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2737 /* All RAID members properly support discards */
2738 ti
->discards_supported
= true;
2741 * RAID1 and RAID10 personalities require bio splitting,
2742 * RAID0/4/5/6 don't and process large discard bios properly.
2744 ti
->split_discard_bios
= !!(rs
->md
.level
== 1 || rs
->md
.level
== 10);
2745 ti
->num_discard_bios
= 1;
2749 * Construct a RAID0/1/10/4/5/6 mapping:
2751 * <raid_type> <#raid_params> <raid_params>{0,} \
2752 * <#raid_devs> [<meta_dev1> <dev1>]{1,}
2754 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
2755 * details on possible <raid_params>.
2757 * Userspace is free to initialize the metadata devices, hence the superblocks to
2758 * enforce recreation based on the passed in table parameters.
2761 static int raid_ctr(struct dm_target
*ti
, unsigned int argc
, char **argv
)
2765 struct raid_type
*rt
;
2766 unsigned int num_raid_params
, num_raid_devs
;
2767 sector_t calculated_dev_sectors
;
2768 struct raid_set
*rs
= NULL
;
2770 struct rs_layout rs_layout
;
2771 struct dm_arg_set as
= { argc
, argv
}, as_nrd
;
2772 struct dm_arg _args
[] = {
2773 { 0, as
.argc
, "Cannot understand number of raid parameters" },
2774 { 1, 254, "Cannot understand number of raid devices parameters" }
2777 /* Must have <raid_type> */
2778 arg
= dm_shift_arg(&as
);
2780 ti
->error
= "No arguments";
2784 rt
= get_raid_type(arg
);
2786 ti
->error
= "Unrecognised raid_type";
2790 /* Must have <#raid_params> */
2791 if (dm_read_arg_group(_args
, &as
, &num_raid_params
, &ti
->error
))
2794 /* number of raid device tupples <meta_dev data_dev> */
2796 dm_consume_args(&as_nrd
, num_raid_params
);
2797 _args
[1].max
= (as_nrd
.argc
- 1) / 2;
2798 if (dm_read_arg(_args
+ 1, &as_nrd
, &num_raid_devs
, &ti
->error
))
2801 if (!__within_range(num_raid_devs
, 1, MAX_RAID_DEVICES
)) {
2802 ti
->error
= "Invalid number of supplied raid devices";
2806 rs
= raid_set_alloc(ti
, rt
, num_raid_devs
);
2810 r
= parse_raid_params(rs
, &as
, num_raid_params
);
2814 r
= parse_dev_params(rs
, &as
);
2818 rs
->md
.sync_super
= super_sync
;
2821 * Calculate ctr requested array and device sizes to allow
2822 * for superblock analysis needing device sizes defined.
2824 * Any existing superblock will overwrite the array and device sizes
2826 r
= rs_set_dev_and_array_sectors(rs
, false);
2830 calculated_dev_sectors
= rs
->dev
[0].rdev
.sectors
;
2833 * Backup any new raid set level, layout, ...
2834 * requested to be able to compare to superblock
2835 * members for conversion decisions.
2837 rs_config_backup(rs
, &rs_layout
);
2839 r
= analyse_superblocks(ti
, rs
);
2843 resize
= calculated_dev_sectors
!= rs
->dev
[0].rdev
.sectors
;
2845 INIT_WORK(&rs
->md
.event_work
, do_table_event
);
2847 ti
->num_flush_bios
= 1;
2849 /* Restore any requested new layout for conversion decision */
2850 rs_config_restore(rs
, &rs_layout
);
2853 * Now that we have any superblock metadata available,
2854 * check for new, recovering, reshaping, to be taken over,
2855 * to be reshaped or an existing, unchanged raid set to
2858 if (test_bit(MD_ARRAY_FIRST_USE
, &rs
->md
.flags
)) {
2859 /* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
2860 if (rs_is_raid6(rs
) &&
2861 test_bit(__CTR_FLAG_NOSYNC
, &rs
->ctr_flags
)) {
2862 ti
->error
= "'nosync' not allowed for new raid6 set";
2866 rs_setup_recovery(rs
, 0);
2867 set_bit(RT_FLAG_UPDATE_SBS
, &rs
->runtime_flags
);
2869 } else if (rs_is_recovering(rs
)) {
2870 /* A recovering raid set may be resized */
2871 ; /* skip setup rs */
2872 } else if (rs_is_reshaping(rs
)) {
2873 /* Have to reject size change request during reshape */
2875 ti
->error
= "Can't resize a reshaping raid set";
2880 } else if (rs_takeover_requested(rs
)) {
2881 if (rs_is_reshaping(rs
)) {
2882 ti
->error
= "Can't takeover a reshaping raid set";
2888 * If a takeover is needed, userspace sets any additional
2889 * devices to rebuild and we can check for a valid request here.
2891 * If acceptible, set the level to the new requested
2892 * one, prohibit requesting recovery, allow the raid
2893 * set to run and store superblocks during resume.
2895 r
= rs_check_takeover(rs
);
2899 r
= rs_setup_takeover(rs
);
2903 set_bit(RT_FLAG_UPDATE_SBS
, &rs
->runtime_flags
);
2904 set_bit(RT_FLAG_KEEP_RS_FROZEN
, &rs
->runtime_flags
);
2905 /* Takeover ain't recovery, so disable recovery */
2906 rs_setup_recovery(rs
, MaxSector
);
2908 } else if (rs_reshape_requested(rs
)) {
2910 * We can only prepare for a reshape here, because the
2911 * raid set needs to run to provide the repective reshape
2912 * check functions via its MD personality instance.
2914 * So do the reshape check after md_run() succeeded.
2916 r
= rs_prepare_reshape(rs
);
2920 /* Reshaping ain't recovery, so disable recovery */
2921 rs_setup_recovery(rs
, MaxSector
);
2924 /* May not set recovery when a device rebuild is requested */
2925 if (test_bit(__CTR_FLAG_REBUILD
, &rs
->ctr_flags
)) {
2926 rs_setup_recovery(rs
, MaxSector
);
2927 set_bit(RT_FLAG_UPDATE_SBS
, &rs
->runtime_flags
);
2929 rs_setup_recovery(rs
, test_bit(__CTR_FLAG_SYNC
, &rs
->ctr_flags
) ?
2930 0 : (resize
? calculated_dev_sectors
: MaxSector
));
2934 /* If constructor requested it, change data and new_data offsets */
2935 r
= rs_adjust_data_offsets(rs
);
2939 /* Start raid set read-only and assumed clean to change in raid_resume() */
2942 set_bit(MD_RECOVERY_FROZEN
, &rs
->md
.recovery
);
2944 /* Has to be held on running the array */
2945 mddev_lock_nointr(&rs
->md
);
2946 r
= md_run(&rs
->md
);
2947 rs
->md
.in_sync
= 0; /* Assume already marked dirty */
2950 ti
->error
= "Failed to run raid array";
2951 mddev_unlock(&rs
->md
);
2955 rs
->callbacks
.congested_fn
= raid_is_congested
;
2956 dm_table_add_target_callbacks(ti
->table
, &rs
->callbacks
);
2958 mddev_suspend(&rs
->md
);
2960 /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
2961 if (rs_is_raid456(rs
)) {
2962 r
= rs_set_raid456_stripe_cache(rs
);
2964 goto bad_stripe_cache
;
2967 /* Now do an early reshape check */
2968 if (test_bit(RT_FLAG_RESHAPE_RS
, &rs
->runtime_flags
)) {
2969 r
= rs_check_reshape(rs
);
2971 goto bad_check_reshape
;
2973 /* Restore new, ctr requested layout to perform check */
2974 rs_config_restore(rs
, &rs_layout
);
2976 if (rs
->md
.pers
->start_reshape
) {
2977 r
= rs
->md
.pers
->check_reshape(&rs
->md
);
2979 ti
->error
= "Reshape check failed";
2980 goto bad_check_reshape
;
2985 mddev_unlock(&rs
->md
);
2997 static void raid_dtr(struct dm_target
*ti
)
2999 struct raid_set
*rs
= ti
->private;
3001 list_del_init(&rs
->callbacks
.list
);
3006 static int raid_map(struct dm_target
*ti
, struct bio
*bio
)
3008 struct raid_set
*rs
= ti
->private;
3009 struct mddev
*mddev
= &rs
->md
;
3012 * If we're reshaping to add disk(s)), ti->len and
3013 * mddev->array_sectors will differ during the process
3014 * (ti->len > mddev->array_sectors), so we have to requeue
3015 * bios with addresses > mddev->array_sectors here or
3016 * there will occur accesses past EOD of the component
3017 * data images thus erroring the raid set.
3019 if (unlikely(bio_end_sector(bio
) > mddev
->array_sectors
))
3020 return DM_MAPIO_REQUEUE
;
3022 mddev
->pers
->make_request(mddev
, bio
);
3024 return DM_MAPIO_SUBMITTED
;
3027 /* Return string describing the current sync action of @mddev */
3028 static const char *decipher_sync_action(struct mddev
*mddev
)
3030 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3033 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3034 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
3035 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
3038 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
3039 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
3041 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
3046 if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
3054 * Return status string @rdev
3056 * Status characters:
3058 * 'D' = Dead/Failed device
3059 * 'a' = Alive but not in-sync
3060 * 'A' = Alive and in-sync
3062 static const char *__raid_dev_status(struct md_rdev
*rdev
, bool array_in_sync
)
3064 if (test_bit(Faulty
, &rdev
->flags
))
3066 else if (!array_in_sync
|| !test_bit(In_sync
, &rdev
->flags
))
3072 /* Helper to return resync/reshape progress for @rs and @array_in_sync */
3073 static sector_t
rs_get_progress(struct raid_set
*rs
,
3074 sector_t resync_max_sectors
, bool *array_in_sync
)
3076 sector_t r
, recovery_cp
, curr_resync_completed
;
3077 struct mddev
*mddev
= &rs
->md
;
3079 curr_resync_completed
= mddev
->curr_resync_completed
?: mddev
->recovery_cp
;
3080 recovery_cp
= mddev
->recovery_cp
;
3081 *array_in_sync
= false;
3083 if (rs_is_raid0(rs
)) {
3084 r
= resync_max_sectors
;
3085 *array_in_sync
= true;
3088 r
= mddev
->reshape_position
;
3090 /* Reshape is relative to the array size */
3091 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) ||
3093 if (r
== MaxSector
) {
3094 *array_in_sync
= true;
3095 r
= resync_max_sectors
;
3097 /* Got to reverse on backward reshape */
3098 if (mddev
->reshape_backwards
)
3099 r
= mddev
->array_sectors
- r
;
3101 /* Devide by # of data stripes */
3102 sector_div(r
, mddev_data_stripes(rs
));
3105 /* Sync is relative to the component device size */
3106 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3107 r
= curr_resync_completed
;
3111 if (r
== MaxSector
) {
3115 *array_in_sync
= true;
3116 r
= resync_max_sectors
;
3117 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
3119 * If "check" or "repair" is occurring, the raid set has
3120 * undergone an initial sync and the health characters
3121 * should not be 'a' anymore.
3123 *array_in_sync
= true;
3125 struct md_rdev
*rdev
;
3128 * The raid set may be doing an initial sync, or it may
3129 * be rebuilding individual components. If all the
3130 * devices are In_sync, then it is the raid set that is
3131 * being initialized.
3133 rdev_for_each(rdev
, mddev
)
3134 if (!test_bit(In_sync
, &rdev
->flags
))
3135 *array_in_sync
= true;
3137 r
= 0; /* HM FIXME: TESTME: https://bugzilla.redhat.com/show_bug.cgi?id=1210637 ? */
3145 /* Helper to return @dev name or "-" if !@dev */
3146 static const char *__get_dev_name(struct dm_dev
*dev
)
3148 return dev
? dev
->name
: "-";
3151 static void raid_status(struct dm_target
*ti
, status_type_t type
,
3152 unsigned int status_flags
, char *result
, unsigned int maxlen
)
3154 struct raid_set
*rs
= ti
->private;
3155 struct mddev
*mddev
= &rs
->md
;
3156 struct r5conf
*conf
= mddev
->private;
3157 int i
, max_nr_stripes
= conf
? conf
->max_nr_stripes
: 0;
3159 unsigned int raid_param_cnt
= 1; /* at least 1 for chunksize */
3160 unsigned int sz
= 0;
3161 unsigned int rebuild_disks
;
3162 unsigned int write_mostly_params
= 0;
3163 sector_t progress
, resync_max_sectors
, resync_mismatches
;
3164 const char *sync_action
;
3165 struct raid_type
*rt
;
3166 struct md_rdev
*rdev
;
3169 case STATUSTYPE_INFO
:
3170 /* *Should* always succeed */
3171 rt
= get_raid_type_by_ll(mddev
->new_level
, mddev
->new_layout
);
3175 DMEMIT("%s %d ", rt
->name
, mddev
->raid_disks
);
3177 /* Access most recent mddev properties for status output */
3179 /* Get sensible max sectors even if raid set not yet started */
3180 resync_max_sectors
= test_bit(RT_FLAG_RS_PRERESUMED
, &rs
->runtime_flags
) ?
3181 mddev
->resync_max_sectors
: mddev
->dev_sectors
;
3182 progress
= rs_get_progress(rs
, resync_max_sectors
, &array_in_sync
);
3183 resync_mismatches
= (mddev
->last_sync_action
&& !strcasecmp(mddev
->last_sync_action
, "check")) ?
3184 atomic64_read(&mddev
->resync_mismatches
) : 0;
3185 sync_action
= decipher_sync_action(&rs
->md
);
3187 /* HM FIXME: do we want another state char for raid0? It shows 'D' or 'A' now */
3188 rdev_for_each(rdev
, mddev
)
3189 DMEMIT(__raid_dev_status(rdev
, array_in_sync
));
3192 * In-sync/Reshape ratio:
3193 * The in-sync ratio shows the progress of:
3194 * - Initializing the raid set
3195 * - Rebuilding a subset of devices of the raid set
3196 * The user can distinguish between the two by referring
3197 * to the status characters.
3199 * The reshape ratio shows the progress of
3200 * changing the raid layout or the number of
3201 * disks of a raid set
3203 DMEMIT(" %llu/%llu", (unsigned long long) progress
,
3204 (unsigned long long) resync_max_sectors
);
3210 * See Documentation/device-mapper/dm-raid.txt for
3211 * information on each of these states.
3213 DMEMIT(" %s", sync_action
);
3218 * resync_mismatches/mismatch_cnt
3219 * This field shows the number of discrepancies found when
3220 * performing a "check" of the raid set.
3222 DMEMIT(" %llu", (unsigned long long) resync_mismatches
);
3227 * data_offset (needed for out of space reshaping)
3228 * This field shows the data offset into the data
3229 * image LV where the first stripes data starts.
3231 * We keep data_offset equal on all raid disks of the set,
3232 * so retrieving it from the first raid disk is sufficient.
3234 DMEMIT(" %llu", (unsigned long long) rs
->dev
[0].rdev
.data_offset
);
3237 case STATUSTYPE_TABLE
:
3238 /* Report the table line string you would use to construct this raid set */
3240 /* Calculate raid parameter count */
3241 for (i
= 0; i
< rs
->raid_disks
; i
++)
3242 if (test_bit(WriteMostly
, &rs
->dev
[i
].rdev
.flags
))
3243 write_mostly_params
+= 2;
3244 rebuild_disks
= memweight(rs
->rebuild_disks
, DISKS_ARRAY_ELEMS
* sizeof(*rs
->rebuild_disks
));
3245 raid_param_cnt
+= rebuild_disks
* 2 +
3246 write_mostly_params
+
3247 hweight32(rs
->ctr_flags
& CTR_FLAG_OPTIONS_NO_ARGS
) +
3248 hweight32(rs
->ctr_flags
& CTR_FLAG_OPTIONS_ONE_ARG
) * 2;
3249 /* Emit table line */
3250 DMEMIT("%s %u %u", rs
->raid_type
->name
, raid_param_cnt
, mddev
->new_chunk_sectors
);
3251 if (test_bit(__CTR_FLAG_RAID10_FORMAT
, &rs
->ctr_flags
))
3252 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT
),
3253 raid10_md_layout_to_format(mddev
->layout
));
3254 if (test_bit(__CTR_FLAG_RAID10_COPIES
, &rs
->ctr_flags
))
3255 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES
),
3256 raid10_md_layout_to_copies(mddev
->layout
));
3257 if (test_bit(__CTR_FLAG_NOSYNC
, &rs
->ctr_flags
))
3258 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC
));
3259 if (test_bit(__CTR_FLAG_SYNC
, &rs
->ctr_flags
))
3260 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC
));
3261 if (test_bit(__CTR_FLAG_REGION_SIZE
, &rs
->ctr_flags
))
3262 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE
),
3263 (unsigned long long) to_sector(mddev
->bitmap_info
.chunksize
));
3264 if (test_bit(__CTR_FLAG_DATA_OFFSET
, &rs
->ctr_flags
))
3265 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET
),
3266 (unsigned long long) rs
->data_offset
);
3267 if (test_bit(__CTR_FLAG_DAEMON_SLEEP
, &rs
->ctr_flags
))
3268 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP
),
3269 mddev
->bitmap_info
.daemon_sleep
);
3270 if (test_bit(__CTR_FLAG_DELTA_DISKS
, &rs
->ctr_flags
))
3271 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS
),
3272 max(rs
->delta_disks
, mddev
->delta_disks
));
3273 if (test_bit(__CTR_FLAG_STRIPE_CACHE
, &rs
->ctr_flags
))
3274 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE
),
3277 for (i
= 0; i
< rs
->raid_disks
; i
++)
3278 if (test_bit(rs
->dev
[i
].rdev
.raid_disk
, (void *) rs
->rebuild_disks
))
3279 DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD
),
3280 rs
->dev
[i
].rdev
.raid_disk
);
3281 if (write_mostly_params
)
3282 for (i
= 0; i
< rs
->raid_disks
; i
++)
3283 if (test_bit(WriteMostly
, &rs
->dev
[i
].rdev
.flags
))
3284 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY
),
3285 rs
->dev
[i
].rdev
.raid_disk
);
3286 if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND
, &rs
->ctr_flags
))
3287 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND
),
3288 mddev
->bitmap_info
.max_write_behind
);
3289 if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE
, &rs
->ctr_flags
))
3290 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE
),
3291 mddev
->sync_speed_max
);
3292 if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE
, &rs
->ctr_flags
))
3293 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE
),
3294 mddev
->sync_speed_min
);
3295 DMEMIT(" %d", rs
->raid_disks
);
3296 for (i
= 0; i
< rs
->raid_disks
; i
++)
3297 DMEMIT(" %s %s", __get_dev_name(rs
->dev
[i
].meta_dev
),
3298 __get_dev_name(rs
->dev
[i
].data_dev
));
3302 static int raid_message(struct dm_target
*ti
, unsigned int argc
, char **argv
)
3304 struct raid_set
*rs
= ti
->private;
3305 struct mddev
*mddev
= &rs
->md
;
3307 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
3310 if (!strcasecmp(argv
[0], "frozen"))
3311 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3313 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3315 if (!strcasecmp(argv
[0], "idle") || !strcasecmp(argv
[0], "frozen")) {
3316 if (mddev
->sync_thread
) {
3317 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3318 md_reap_sync_thread(mddev
);
3320 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3321 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
3323 else if (!strcasecmp(argv
[0], "resync"))
3324 ; /* MD_RECOVERY_NEEDED set below */
3325 else if (!strcasecmp(argv
[0], "recover"))
3326 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
3328 if (!strcasecmp(argv
[0], "check"))
3329 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
3330 else if (!!strcasecmp(argv
[0], "repair"))
3332 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
3333 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
3335 if (mddev
->ro
== 2) {
3336 /* A write to sync_action is enough to justify
3337 * canceling read-auto mode
3340 if (!mddev
->suspended
&& mddev
->sync_thread
)
3341 md_wakeup_thread(mddev
->sync_thread
);
3343 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3344 if (!mddev
->suspended
&& mddev
->thread
)
3345 md_wakeup_thread(mddev
->thread
);
3350 static int raid_iterate_devices(struct dm_target
*ti
,
3351 iterate_devices_callout_fn fn
, void *data
)
3353 struct raid_set
*rs
= ti
->private;
3357 for (i
= 0; !r
&& i
< rs
->md
.raid_disks
; i
++)
3358 if (rs
->dev
[i
].data_dev
)
3360 rs
->dev
[i
].data_dev
,
3361 0, /* No offset on data devs */
3368 static void raid_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
3370 struct raid_set
*rs
= ti
->private;
3371 unsigned int chunk_size
= to_bytes(rs
->md
.chunk_sectors
);
3373 blk_limits_io_min(limits
, chunk_size
);
3374 blk_limits_io_opt(limits
, chunk_size
* mddev_data_stripes(rs
));
3377 static void raid_presuspend(struct dm_target
*ti
)
3379 struct raid_set
*rs
= ti
->private;
3381 md_stop_writes(&rs
->md
);
3384 static void raid_postsuspend(struct dm_target
*ti
)
3386 struct raid_set
*rs
= ti
->private;
3388 if (test_and_clear_bit(RT_FLAG_RS_RESUMED
, &rs
->runtime_flags
)) {
3389 if (!rs
->md
.suspended
)
3390 mddev_suspend(&rs
->md
);
3395 static void attempt_restore_of_faulty_devices(struct raid_set
*rs
)
3398 uint64_t failed_devices
, cleared_failed_devices
= 0;
3399 unsigned long flags
;
3400 struct dm_raid_superblock
*sb
;
3403 for (i
= 0; i
< rs
->md
.raid_disks
; i
++) {
3404 r
= &rs
->dev
[i
].rdev
;
3405 if (test_bit(Faulty
, &r
->flags
) && r
->sb_page
&&
3406 sync_page_io(r
, 0, r
->sb_size
, r
->sb_page
,
3407 REQ_OP_READ
, 0, true)) {
3408 DMINFO("Faulty %s device #%d has readable super block."
3409 " Attempting to revive it.",
3410 rs
->raid_type
->name
, i
);
3413 * Faulty bit may be set, but sometimes the array can
3414 * be suspended before the personalities can respond
3415 * by removing the device from the array (i.e. calling
3416 * 'hot_remove_disk'). If they haven't yet removed
3417 * the failed device, its 'raid_disk' number will be
3418 * '>= 0' - meaning we must call this function
3421 if ((r
->raid_disk
>= 0) &&
3422 (r
->mddev
->pers
->hot_remove_disk(r
->mddev
, r
) != 0))
3423 /* Failed to revive this device, try next */
3427 r
->saved_raid_disk
= i
;
3429 clear_bit(Faulty
, &r
->flags
);
3430 clear_bit(WriteErrorSeen
, &r
->flags
);
3431 clear_bit(In_sync
, &r
->flags
);
3432 if (r
->mddev
->pers
->hot_add_disk(r
->mddev
, r
)) {
3434 r
->saved_raid_disk
= -1;
3437 r
->recovery_offset
= 0;
3438 cleared_failed_devices
|= 1 << i
;
3442 if (cleared_failed_devices
) {
3443 rdev_for_each(r
, &rs
->md
) {
3444 sb
= page_address(r
->sb_page
);
3445 failed_devices
= le64_to_cpu(sb
->failed_devices
);
3446 failed_devices
&= ~cleared_failed_devices
;
3447 sb
->failed_devices
= cpu_to_le64(failed_devices
);
3452 static int __load_dirty_region_bitmap(struct raid_set
*rs
)
3456 /* Try loading the bitmap unless "raid0", which does not have one */
3457 if (!rs_is_raid0(rs
) &&
3458 !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED
, &rs
->runtime_flags
)) {
3459 r
= bitmap_load(&rs
->md
);
3461 DMERR("Failed to load bitmap");
3467 /* Enforce updating all superblocks */
3468 static void rs_update_sbs(struct raid_set
*rs
)
3470 struct mddev
*mddev
= &rs
->md
;
3473 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3475 md_update_sb(mddev
, 1);
3480 * Reshape changes raid algorithm of @rs to new one within personality
3481 * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3482 * disks from a raid set thus growing/shrinking it or resizes the set
3484 * Call mddev_lock_nointr() before!
3486 static int rs_start_reshape(struct raid_set
*rs
)
3489 struct mddev
*mddev
= &rs
->md
;
3490 struct md_personality
*pers
= mddev
->pers
;
3492 r
= rs_setup_reshape(rs
);
3496 /* Need to be resumed to be able to start reshape, recovery is frozen until raid_resume() though */
3497 if (mddev
->suspended
)
3498 mddev_resume(mddev
);
3501 * Check any reshape constraints enforced by the personalility
3503 * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3505 r
= pers
->check_reshape(mddev
);
3507 rs
->ti
->error
= "pers->check_reshape() failed";
3512 * Personality may not provide start reshape method in which
3513 * case check_reshape above has already covered everything
3515 if (pers
->start_reshape
) {
3516 r
= pers
->start_reshape(mddev
);
3518 rs
->ti
->error
= "pers->start_reshape() failed";
3523 /* Suspend because a resume will happen in raid_resume() */
3524 if (!mddev
->suspended
)
3525 mddev_suspend(mddev
);
3528 * Now reshape got set up, update superblocks to
3529 * reflect the fact so that a table reload will
3530 * access proper superblock content in the ctr.
3537 static int raid_preresume(struct dm_target
*ti
)
3540 struct raid_set
*rs
= ti
->private;
3541 struct mddev
*mddev
= &rs
->md
;
3543 /* This is a resume after a suspend of the set -> it's already started */
3544 if (test_and_set_bit(RT_FLAG_RS_PRERESUMED
, &rs
->runtime_flags
))
3548 * The superblocks need to be updated on disk if the
3549 * array is new or new devices got added (thus zeroed
3550 * out by userspace) or __load_dirty_region_bitmap
3551 * will overwrite them in core with old data or fail.
3553 if (test_bit(RT_FLAG_UPDATE_SBS
, &rs
->runtime_flags
))
3557 * Disable/enable discard support on raid set after any
3558 * conversion, because devices can have been added
3560 configure_discard_support(rs
);
3562 /* Load the bitmap from disk unless raid0 */
3563 r
= __load_dirty_region_bitmap(rs
);
3567 /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) */
3568 if (test_bit(RT_FLAG_RS_BITMAP_LOADED
, &rs
->runtime_flags
) &&
3569 mddev
->bitmap_info
.chunksize
!= to_bytes(rs
->requested_bitmap_chunk_sectors
)) {
3570 r
= bitmap_resize(mddev
->bitmap
, mddev
->dev_sectors
,
3571 to_bytes(rs
->requested_bitmap_chunk_sectors
), 0);
3573 DMERR("Failed to resize bitmap");
3576 /* Check for any resize/reshape on @rs and adjust/initiate */
3577 /* Be prepared for mddev_resume() in raid_resume() */
3578 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3579 if (mddev
->recovery_cp
&& mddev
->recovery_cp
< MaxSector
) {
3580 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
3581 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
3582 mddev
->resync_min
= mddev
->recovery_cp
;
3585 rs_set_capacity(rs
);
3587 /* Check for any reshape request unless new raid set */
3588 if (test_and_clear_bit(RT_FLAG_RESHAPE_RS
, &rs
->runtime_flags
)) {
3589 /* Initiate a reshape. */
3590 mddev_lock_nointr(mddev
);
3591 r
= rs_start_reshape(rs
);
3592 mddev_unlock(mddev
);
3594 DMWARN("Failed to check/start reshape, continuing without change");
3601 static void raid_resume(struct dm_target
*ti
)
3603 struct raid_set
*rs
= ti
->private;
3604 struct mddev
*mddev
= &rs
->md
;
3606 if (test_and_set_bit(RT_FLAG_RS_RESUMED
, &rs
->runtime_flags
)) {
3608 * A secondary resume while the device is active.
3609 * Take this opportunity to check whether any failed
3610 * devices are reachable again.
3612 attempt_restore_of_faulty_devices(rs
);
3618 * When passing in flags to the ctr, we expect userspace
3619 * to reset them because they made it to the superblocks
3620 * and reload the mapping anyway.
3622 * -> only unfreeze recovery in case of a table reload or
3623 * we'll have a bogus recovery/reshape position
3624 * retrieved from the superblock by the ctr because
3625 * the ongoing recovery/reshape will change it after read.
3627 if (!test_bit(RT_FLAG_KEEP_RS_FROZEN
, &rs
->runtime_flags
))
3628 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3630 if (mddev
->suspended
)
3631 mddev_resume(mddev
);
3635 static struct target_type raid_target
= {
3637 .version
= {1, 9, 0},
3638 .module
= THIS_MODULE
,
3642 .status
= raid_status
,
3643 .message
= raid_message
,
3644 .iterate_devices
= raid_iterate_devices
,
3645 .io_hints
= raid_io_hints
,
3646 .presuspend
= raid_presuspend
,
3647 .postsuspend
= raid_postsuspend
,
3648 .preresume
= raid_preresume
,
3649 .resume
= raid_resume
,
3652 static int __init
dm_raid_init(void)
3654 DMINFO("Loading target version %u.%u.%u",
3655 raid_target
.version
[0],
3656 raid_target
.version
[1],
3657 raid_target
.version
[2]);
3658 return dm_register_target(&raid_target
);
3661 static void __exit
dm_raid_exit(void)
3663 dm_unregister_target(&raid_target
);
3666 module_init(dm_raid_init
);
3667 module_exit(dm_raid_exit
);
3669 module_param(devices_handle_discard_safely
, bool, 0644);
3670 MODULE_PARM_DESC(devices_handle_discard_safely
,
3671 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
3673 MODULE_DESCRIPTION(DM_NAME
" raid0/1/10/4/5/6 target");
3674 MODULE_ALIAS("dm-raid0");
3675 MODULE_ALIAS("dm-raid1");
3676 MODULE_ALIAS("dm-raid10");
3677 MODULE_ALIAS("dm-raid4");
3678 MODULE_ALIAS("dm-raid5");
3679 MODULE_ALIAS("dm-raid6");
3680 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
3681 MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
3682 MODULE_LICENSE("GPL");