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 raid4/5/6/10! */
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_DATA_OFFSET)
137 /* "raid10" does not accept any raid1 or stripe cache options */
138 #define RAID10_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
140 CTR_FLAG_DAEMON_SLEEP | \
141 CTR_FLAG_MIN_RECOVERY_RATE | \
142 CTR_FLAG_MAX_RECOVERY_RATE | \
143 CTR_FLAG_REGION_SIZE | \
144 CTR_FLAG_RAID10_COPIES | \
145 CTR_FLAG_RAID10_FORMAT | \
146 CTR_FLAG_DELTA_DISKS | \
147 CTR_FLAG_DATA_OFFSET | \
148 CTR_FLAG_RAID10_USE_NEAR_SETS)
151 * "raid4/5/6" do not accept any raid1 or raid10 specific options
153 * "raid6" does not accept "nosync", because it is not guaranteed
154 * that both parity and q-syndrome are being written properly with
157 #define RAID45_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
159 CTR_FLAG_DAEMON_SLEEP | \
160 CTR_FLAG_MIN_RECOVERY_RATE | \
161 CTR_FLAG_MAX_RECOVERY_RATE | \
162 CTR_FLAG_MAX_WRITE_BEHIND | \
163 CTR_FLAG_STRIPE_CACHE | \
164 CTR_FLAG_REGION_SIZE | \
165 CTR_FLAG_DELTA_DISKS | \
166 CTR_FLAG_DATA_OFFSET)
168 #define RAID6_VALID_FLAGS (CTR_FLAG_SYNC | \
170 CTR_FLAG_DAEMON_SLEEP | \
171 CTR_FLAG_MIN_RECOVERY_RATE | \
172 CTR_FLAG_MAX_RECOVERY_RATE | \
173 CTR_FLAG_MAX_WRITE_BEHIND | \
174 CTR_FLAG_STRIPE_CACHE | \
175 CTR_FLAG_REGION_SIZE | \
176 CTR_FLAG_DELTA_DISKS | \
177 CTR_FLAG_DATA_OFFSET)
178 /* ...valid options definitions per raid level */
181 * Flags for rs->runtime_flags field
182 * (RT_FLAG prefix meaning "runtime flag")
184 * These are all internal and used to define runtime state,
185 * e.g. to prevent another resume from preresume processing
186 * the raid set all over again.
188 #define RT_FLAG_RS_PRERESUMED 0
189 #define RT_FLAG_RS_RESUMED 1
190 #define RT_FLAG_RS_BITMAP_LOADED 2
191 #define RT_FLAG_UPDATE_SBS 3
192 #define RT_FLAG_RESHAPE_RS 4
193 #define RT_FLAG_KEEP_RS_FROZEN 5
195 /* Array elements of 64 bit needed for rebuild/write_mostly bits */
196 #define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
199 * raid set level, layout and chunk sectors backup/restore
204 int new_chunk_sectors
;
208 struct dm_target
*ti
;
210 uint32_t bitmap_loaded
;
211 uint32_t stripe_cache_entries
;
212 unsigned long ctr_flags
;
213 unsigned long runtime_flags
;
215 uint64_t rebuild_disks
[DISKS_ARRAY_ELEMS
];
221 int requested_bitmap_chunk_sectors
;
224 struct raid_type
*raid_type
;
225 struct dm_target_callbacks callbacks
;
227 struct raid_dev dev
[0];
230 static void rs_config_backup(struct raid_set
*rs
, struct rs_layout
*l
)
232 struct mddev
*mddev
= &rs
->md
;
234 l
->new_level
= mddev
->new_level
;
235 l
->new_layout
= mddev
->new_layout
;
236 l
->new_chunk_sectors
= mddev
->new_chunk_sectors
;
239 static void rs_config_restore(struct raid_set
*rs
, struct rs_layout
*l
)
241 struct mddev
*mddev
= &rs
->md
;
243 mddev
->new_level
= l
->new_level
;
244 mddev
->new_layout
= l
->new_layout
;
245 mddev
->new_chunk_sectors
= l
->new_chunk_sectors
;
248 /* raid10 algorithms (i.e. formats) */
249 #define ALGORITHM_RAID10_DEFAULT 0
250 #define ALGORITHM_RAID10_NEAR 1
251 #define ALGORITHM_RAID10_OFFSET 2
252 #define ALGORITHM_RAID10_FAR 3
254 /* Supported raid types and properties. */
255 static struct raid_type
{
256 const char *name
; /* RAID algorithm. */
257 const char *descr
; /* Descriptor text for logging. */
258 const unsigned parity_devs
; /* # of parity devices. */
259 const unsigned minimal_devs
; /* minimal # of devices in set. */
260 const unsigned level
; /* RAID level. */
261 const unsigned algorithm
; /* RAID algorithm. */
263 {"raid0", "raid0 (striping)", 0, 2, 0, 0 /* NONE */},
264 {"raid1", "raid1 (mirroring)", 0, 2, 1, 0 /* NONE */},
265 {"raid10_far", "raid10 far (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_FAR
},
266 {"raid10_offset", "raid10 offset (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_OFFSET
},
267 {"raid10_near", "raid10 near (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_NEAR
},
268 {"raid10", "raid10 (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_DEFAULT
},
269 {"raid4", "raid4 (dedicated last parity disk)", 1, 2, 4, ALGORITHM_PARITY_N
}, /* raid4 layout = raid5_n */
270 {"raid5_n", "raid5 (dedicated last parity disk)", 1, 2, 5, ALGORITHM_PARITY_N
},
271 {"raid5_ls", "raid5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC
},
272 {"raid5_rs", "raid5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC
},
273 {"raid5_la", "raid5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC
},
274 {"raid5_ra", "raid5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC
},
275 {"raid6_zr", "raid6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART
},
276 {"raid6_nr", "raid6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART
},
277 {"raid6_nc", "raid6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE
},
278 {"raid6_n_6", "raid6 (dedicated parity/Q n/6)", 2, 4, 6, ALGORITHM_PARITY_N_6
},
279 {"raid6_ls_6", "raid6 (left symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_SYMMETRIC_6
},
280 {"raid6_rs_6", "raid6 (right symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_SYMMETRIC_6
},
281 {"raid6_la_6", "raid6 (left asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_ASYMMETRIC_6
},
282 {"raid6_ra_6", "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_ASYMMETRIC_6
}
285 /* True, if @v is in inclusive range [@min, @max] */
286 static bool __within_range(long v
, long min
, long max
)
288 return v
>= min
&& v
<= max
;
291 /* All table line arguments are defined here */
292 static struct arg_name_flag
{
293 const unsigned long flag
;
295 } __arg_name_flags
[] = {
296 { CTR_FLAG_SYNC
, "sync"},
297 { CTR_FLAG_NOSYNC
, "nosync"},
298 { CTR_FLAG_REBUILD
, "rebuild"},
299 { CTR_FLAG_DAEMON_SLEEP
, "daemon_sleep"},
300 { CTR_FLAG_MIN_RECOVERY_RATE
, "min_recovery_rate"},
301 { CTR_FLAG_MAX_RECOVERY_RATE
, "max_recovery_rate"},
302 { CTR_FLAG_MAX_WRITE_BEHIND
, "max_write_behind"},
303 { CTR_FLAG_WRITE_MOSTLY
, "write_mostly"},
304 { CTR_FLAG_STRIPE_CACHE
, "stripe_cache"},
305 { CTR_FLAG_REGION_SIZE
, "region_size"},
306 { CTR_FLAG_RAID10_COPIES
, "raid10_copies"},
307 { CTR_FLAG_RAID10_FORMAT
, "raid10_format"},
308 { CTR_FLAG_DATA_OFFSET
, "data_offset"},
309 { CTR_FLAG_DELTA_DISKS
, "delta_disks"},
310 { CTR_FLAG_RAID10_USE_NEAR_SETS
, "raid10_use_near_sets"},
313 /* Return argument name string for given @flag */
314 static const char *dm_raid_arg_name_by_flag(const uint32_t flag
)
316 if (hweight32(flag
) == 1) {
317 struct arg_name_flag
*anf
= __arg_name_flags
+ ARRAY_SIZE(__arg_name_flags
);
319 while (anf
-- > __arg_name_flags
)
320 if (flag
& anf
->flag
)
324 DMERR("%s called with more than one flag!", __func__
);
330 * bool helpers to test for various raid levels of a raid set,
331 * is. it's level as reported by the superblock rather than
332 * the requested raid_type passed to the constructor.
334 /* Return true, if raid set in @rs is raid0 */
335 static bool rs_is_raid0(struct raid_set
*rs
)
337 return !rs
->md
.level
;
340 /* Return true, if raid set in @rs is raid1 */
341 static bool rs_is_raid1(struct raid_set
*rs
)
343 return rs
->md
.level
== 1;
346 /* Return true, if raid set in @rs is raid10 */
347 static bool rs_is_raid10(struct raid_set
*rs
)
349 return rs
->md
.level
== 10;
352 /* Return true, if raid set in @rs is level 6 */
353 static bool rs_is_raid6(struct raid_set
*rs
)
355 return rs
->md
.level
== 6;
358 /* Return true, if raid set in @rs is level 4, 5 or 6 */
359 static bool rs_is_raid456(struct raid_set
*rs
)
361 return __within_range(rs
->md
.level
, 4, 6);
364 /* Return true, if raid set in @rs is reshapable */
365 static unsigned int __is_raid10_far(int layout
);
366 static bool rs_is_reshapable(struct raid_set
*rs
)
368 return rs_is_raid456(rs
) ||
369 (rs_is_raid10(rs
) && !__is_raid10_far(rs
->md
.new_layout
));
372 /* Return true, if raid set in @rs is recovering */
373 static bool rs_is_recovering(struct raid_set
*rs
)
375 return rs
->md
.recovery_cp
< rs
->dev
[0].rdev
.sectors
;
378 /* Return true, if raid set in @rs is reshaping */
379 static bool rs_is_reshaping(struct raid_set
*rs
)
381 return rs
->md
.reshape_position
!= MaxSector
;
385 * bool helpers to test for various raid levels of a raid type
388 /* Return true, if raid type in @rt is raid0 */
389 static bool rt_is_raid0(struct raid_type
*rt
)
394 /* Return true, if raid type in @rt is raid1 */
395 static bool rt_is_raid1(struct raid_type
*rt
)
397 return rt
->level
== 1;
400 /* Return true, if raid type in @rt is raid10 */
401 static bool rt_is_raid10(struct raid_type
*rt
)
403 return rt
->level
== 10;
406 /* Return true, if raid type in @rt is raid4/5 */
407 static bool rt_is_raid45(struct raid_type
*rt
)
409 return __within_range(rt
->level
, 4, 5);
412 /* Return true, if raid type in @rt is raid6 */
413 static bool rt_is_raid6(struct raid_type
*rt
)
415 return rt
->level
== 6;
418 /* Return true, if raid type in @rt is raid4/5/6 */
419 static bool rt_is_raid456(struct raid_type
*rt
)
421 return __within_range(rt
->level
, 4, 6);
423 /* END: raid level bools */
425 /* Return valid ctr flags for the raid level of @rs */
426 static unsigned long __valid_flags(struct raid_set
*rs
)
428 if (rt_is_raid0(rs
->raid_type
))
429 return RAID0_VALID_FLAGS
;
430 else if (rt_is_raid1(rs
->raid_type
))
431 return RAID1_VALID_FLAGS
;
432 else if (rt_is_raid10(rs
->raid_type
))
433 return RAID10_VALID_FLAGS
;
434 else if (rt_is_raid45(rs
->raid_type
))
435 return RAID45_VALID_FLAGS
;
436 else if (rt_is_raid6(rs
->raid_type
))
437 return RAID6_VALID_FLAGS
;
443 * Check for valid flags set on @rs
445 * Has to be called after parsing of the ctr flags!
447 static int rs_check_for_valid_flags(struct raid_set
*rs
)
449 if (rs
->ctr_flags
& ~__valid_flags(rs
)) {
450 rs
->ti
->error
= "Invalid flags combination";
457 /* MD raid10 bit definitions and helpers */
458 #define RAID10_OFFSET (1 << 16) /* stripes with data copies area adjacent on devices */
459 #define RAID10_BROCKEN_USE_FAR_SETS (1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
460 #define RAID10_USE_FAR_SETS (1 << 18) /* Use sets instead of whole stripe rotation */
461 #define RAID10_FAR_COPIES_SHIFT 8 /* raid10 # far copies shift (2nd byte of layout) */
463 /* Return md raid10 near copies for @layout */
464 static unsigned int __raid10_near_copies(int layout
)
466 return layout
& 0xFF;
469 /* Return md raid10 far copies for @layout */
470 static unsigned int __raid10_far_copies(int layout
)
472 return __raid10_near_copies(layout
>> RAID10_FAR_COPIES_SHIFT
);
475 /* Return true if md raid10 offset for @layout */
476 static unsigned int __is_raid10_offset(int layout
)
478 return layout
& RAID10_OFFSET
;
481 /* Return true if md raid10 near for @layout */
482 static unsigned int __is_raid10_near(int layout
)
484 return !__is_raid10_offset(layout
) && __raid10_near_copies(layout
) > 1;
487 /* Return true if md raid10 far for @layout */
488 static unsigned int __is_raid10_far(int layout
)
490 return !__is_raid10_offset(layout
) && __raid10_far_copies(layout
) > 1;
493 /* Return md raid10 layout string for @layout */
494 static const char *raid10_md_layout_to_format(int layout
)
497 * Bit 16 stands for "offset"
498 * (i.e. adjacent stripes hold copies)
500 * Refer to MD's raid10.c for details
502 if (__is_raid10_offset(layout
))
505 if (__raid10_near_copies(layout
) > 1)
508 WARN_ON(__raid10_far_copies(layout
) < 2);
513 /* Return md raid10 algorithm for @name */
514 static int raid10_name_to_format(const char *name
)
516 if (!strcasecmp(name
, "near"))
517 return ALGORITHM_RAID10_NEAR
;
518 else if (!strcasecmp(name
, "offset"))
519 return ALGORITHM_RAID10_OFFSET
;
520 else if (!strcasecmp(name
, "far"))
521 return ALGORITHM_RAID10_FAR
;
526 /* Return md raid10 copies for @layout */
527 static unsigned int raid10_md_layout_to_copies(int layout
)
529 return __raid10_near_copies(layout
) > 1 ?
530 __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 int __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
,
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
->md
.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
->md
.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
= (region_size
<< 9);
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 i
, rebuild_cnt
= 0;
924 unsigned rebuilds_per_group
= 0, copies
;
925 unsigned 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
->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 num_raid_params
)
1035 int value
, raid10_format
= ALGORITHM_RAID10_DEFAULT
;
1036 unsigned raid10_copies
= 2;
1037 unsigned i
, write_mostly
= 0;
1038 unsigned 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
->md
.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
)
1703 struct mddev
*mddev
= &rs
->md
;
1708 return !__is_raid10_far(mddev
->new_layout
) &&
1709 mddev
->new_level
== mddev
->level
&&
1710 (mddev
->new_layout
!= mddev
->layout
||
1711 mddev
->new_chunk_sectors
!= mddev
->chunk_sectors
||
1712 rs
->raid_disks
+ rs
->delta_disks
!= mddev
->raid_disks
);
1716 #define FEATURE_FLAG_SUPPORTS_V190 0x1 /* Supports extended superblock */
1718 /* State flags for sb->flags */
1719 #define SB_FLAG_RESHAPE_ACTIVE 0x1
1720 #define SB_FLAG_RESHAPE_BACKWARDS 0x2
1723 * This structure is never routinely used by userspace, unlike md superblocks.
1724 * Devices with this superblock should only ever be accessed via device-mapper.
1726 #define DM_RAID_MAGIC 0x64526D44
1727 struct dm_raid_superblock
{
1728 __le32 magic
; /* "DmRd" */
1729 __le32 compat_features
; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1731 __le32 num_devices
; /* Number of devices in this raid set. (Max 64) */
1732 __le32 array_position
; /* The position of this drive in the raid set */
1734 __le64 events
; /* Incremented by md when superblock updated */
1735 __le64 failed_devices
; /* Pre 1.9.0 part of bit field of devices to */
1736 /* indicate failures (see extension below) */
1739 * This offset tracks the progress of the repair or replacement of
1740 * an individual drive.
1742 __le64 disk_recovery_offset
;
1745 * This offset tracks the progress of the initial raid set
1746 * synchronisation/parity calculation.
1748 __le64 array_resync_offset
;
1751 * raid characteristics
1755 __le32 stripe_sectors
;
1757 /********************************************************************
1758 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1760 * FEATURE_FLAG_SUPPORTS_V190 in the features member indicates that those exist
1763 __le32 flags
; /* Flags defining array states for reshaping */
1766 * This offset tracks the progress of a raid
1767 * set reshape in order to be able to restart it
1769 __le64 reshape_position
;
1772 * These define the properties of the array in case of an interrupted reshape
1776 __le32 new_stripe_sectors
;
1779 __le64 array_sectors
; /* Array size in sectors */
1782 * Sector offsets to data on devices (reshaping).
1783 * Needed to support out of place reshaping, thus
1784 * not writing over any stripes whilst converting
1785 * them from old to new layout
1788 __le64 new_data_offset
;
1790 __le64 sectors
; /* Used device size in sectors */
1793 * Additonal Bit field of devices indicating failures to support
1794 * up to 256 devices with the 1.9.0 on-disk metadata format
1796 __le64 extended_failed_devices
[DISKS_ARRAY_ELEMS
- 1];
1798 __le32 incompat_features
; /* Used to indicate any incompatible features */
1800 /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
1804 * Check for reshape constraints on raid set @rs:
1806 * - reshape function non-existent
1808 * - ongoing recovery
1811 * Returns 0 if none or -EPERM if given constraint
1812 * and error message reference in @errmsg
1814 static int rs_check_reshape(struct raid_set
*rs
)
1816 struct mddev
*mddev
= &rs
->md
;
1818 if (!mddev
->pers
|| !mddev
->pers
->check_reshape
)
1819 rs
->ti
->error
= "Reshape not supported";
1820 else if (mddev
->degraded
)
1821 rs
->ti
->error
= "Can't reshape degraded raid set";
1822 else if (rs_is_recovering(rs
))
1823 rs
->ti
->error
= "Convert request on recovering raid set prohibited";
1824 else if (mddev
->reshape_position
&& rs_is_reshaping(rs
))
1825 rs
->ti
->error
= "raid set already reshaping!";
1826 else if (!(rs_is_raid10(rs
) || rs_is_raid456(rs
)))
1827 rs
->ti
->error
= "Reshaping only supported for raid4/5/6/10";
1834 static int read_disk_sb(struct md_rdev
*rdev
, int size
)
1836 BUG_ON(!rdev
->sb_page
);
1838 if (rdev
->sb_loaded
)
1841 if (!sync_page_io(rdev
, 0, size
, rdev
->sb_page
, REQ_OP_READ
, 0, true)) {
1842 DMERR("Failed to read superblock of device at position %d",
1844 md_error(rdev
->mddev
, rdev
);
1848 rdev
->sb_loaded
= 1;
1853 static void sb_retrieve_failed_devices(struct dm_raid_superblock
*sb
, uint64_t *failed_devices
)
1855 failed_devices
[0] = le64_to_cpu(sb
->failed_devices
);
1856 memset(failed_devices
+ 1, 0, sizeof(sb
->extended_failed_devices
));
1858 if (le32_to_cpu(sb
->compat_features
) & FEATURE_FLAG_SUPPORTS_V190
) {
1859 int i
= ARRAY_SIZE(sb
->extended_failed_devices
);
1862 failed_devices
[i
+1] = le64_to_cpu(sb
->extended_failed_devices
[i
]);
1866 static void sb_update_failed_devices(struct dm_raid_superblock
*sb
, uint64_t *failed_devices
)
1868 int i
= ARRAY_SIZE(sb
->extended_failed_devices
);
1870 sb
->failed_devices
= cpu_to_le64(failed_devices
[0]);
1872 sb
->extended_failed_devices
[i
] = cpu_to_le64(failed_devices
[i
+1]);
1876 * Synchronize the superblock members with the raid set properties
1878 * All superblock data is little endian.
1880 static void super_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1882 bool update_failed_devices
= false;
1884 uint64_t failed_devices
[DISKS_ARRAY_ELEMS
];
1885 struct dm_raid_superblock
*sb
;
1886 struct raid_set
*rs
= container_of(mddev
, struct raid_set
, md
);
1888 /* No metadata device, no superblock */
1889 if (!rdev
->meta_bdev
)
1892 BUG_ON(!rdev
->sb_page
);
1894 sb
= page_address(rdev
->sb_page
);
1896 sb_retrieve_failed_devices(sb
, failed_devices
);
1898 for (i
= 0; i
< rs
->raid_disks
; i
++)
1899 if (!rs
->dev
[i
].data_dev
|| test_bit(Faulty
, &rs
->dev
[i
].rdev
.flags
)) {
1900 update_failed_devices
= true;
1901 set_bit(i
, (void *) failed_devices
);
1904 if (update_failed_devices
)
1905 sb_update_failed_devices(sb
, failed_devices
);
1907 sb
->magic
= cpu_to_le32(DM_RAID_MAGIC
);
1908 sb
->compat_features
= cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190
);
1910 sb
->num_devices
= cpu_to_le32(mddev
->raid_disks
);
1911 sb
->array_position
= cpu_to_le32(rdev
->raid_disk
);
1913 sb
->events
= cpu_to_le64(mddev
->events
);
1915 sb
->disk_recovery_offset
= cpu_to_le64(rdev
->recovery_offset
);
1916 sb
->array_resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1918 sb
->level
= cpu_to_le32(mddev
->level
);
1919 sb
->layout
= cpu_to_le32(mddev
->layout
);
1920 sb
->stripe_sectors
= cpu_to_le32(mddev
->chunk_sectors
);
1922 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1923 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1924 sb
->new_stripe_sectors
= cpu_to_le32(mddev
->new_chunk_sectors
);
1926 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1928 smp_rmb(); /* Make sure we access most recent reshape position */
1929 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1930 if (le64_to_cpu(sb
->reshape_position
) != MaxSector
) {
1931 /* Flag ongoing reshape */
1932 sb
->flags
|= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE
);
1934 if (mddev
->delta_disks
< 0 || mddev
->reshape_backwards
)
1935 sb
->flags
|= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS
);
1937 /* Clear reshape flags */
1938 sb
->flags
&= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE
|SB_FLAG_RESHAPE_BACKWARDS
));
1941 sb
->array_sectors
= cpu_to_le64(mddev
->array_sectors
);
1942 sb
->data_offset
= cpu_to_le64(rdev
->data_offset
);
1943 sb
->new_data_offset
= cpu_to_le64(rdev
->new_data_offset
);
1944 sb
->sectors
= cpu_to_le64(rdev
->sectors
);
1946 /* Zero out the rest of the payload after the size of the superblock */
1947 memset(sb
+ 1, 0, rdev
->sb_size
- sizeof(*sb
));
1953 * This function creates a superblock if one is not found on the device
1954 * and will decide which superblock to use if there's a choice.
1956 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
1958 static int super_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
)
1961 struct dm_raid_superblock
*sb
;
1962 struct dm_raid_superblock
*refsb
;
1963 uint64_t events_sb
, events_refsb
;
1966 rdev
->sb_size
= bdev_logical_block_size(rdev
->meta_bdev
);
1967 if (rdev
->sb_size
< sizeof(*sb
) || rdev
->sb_size
> PAGE_SIZE
) {
1968 DMERR("superblock size of a logical block is no longer valid");
1972 r
= read_disk_sb(rdev
, rdev
->sb_size
);
1976 sb
= page_address(rdev
->sb_page
);
1979 * Two cases that we want to write new superblocks and rebuild:
1980 * 1) New device (no matching magic number)
1981 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
1983 if ((sb
->magic
!= cpu_to_le32(DM_RAID_MAGIC
)) ||
1984 (!test_bit(In_sync
, &rdev
->flags
) && !rdev
->recovery_offset
)) {
1985 super_sync(rdev
->mddev
, rdev
);
1987 set_bit(FirstUse
, &rdev
->flags
);
1988 sb
->compat_features
= cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190
);
1990 /* Force writing of superblocks to disk */
1991 set_bit(MD_CHANGE_DEVS
, &rdev
->mddev
->flags
);
1993 /* Any superblock is better than none, choose that if given */
1994 return refdev
? 0 : 1;
2000 events_sb
= le64_to_cpu(sb
->events
);
2002 refsb
= page_address(refdev
->sb_page
);
2003 events_refsb
= le64_to_cpu(refsb
->events
);
2005 return (events_sb
> events_refsb
) ? 1 : 0;
2008 static int super_init_validation(struct raid_set
*rs
, struct md_rdev
*rdev
)
2012 struct mddev
*mddev
= &rs
->md
;
2014 uint64_t failed_devices
[DISKS_ARRAY_ELEMS
];
2015 struct dm_raid_superblock
*sb
;
2016 uint32_t new_devs
= 0, rebuild_and_new
= 0, rebuilds
= 0;
2018 struct dm_raid_superblock
*sb2
;
2020 sb
= page_address(rdev
->sb_page
);
2021 events_sb
= le64_to_cpu(sb
->events
);
2024 * Initialise to 1 if this is a new superblock.
2026 mddev
->events
= events_sb
? : 1;
2028 mddev
->reshape_position
= MaxSector
;
2031 * Reshaping is supported, e.g. reshape_position is valid
2032 * in superblock and superblock content is authoritative.
2034 if (le32_to_cpu(sb
->compat_features
) & FEATURE_FLAG_SUPPORTS_V190
) {
2035 /* Superblock is authoritative wrt given raid set layout! */
2036 mddev
->raid_disks
= le32_to_cpu(sb
->num_devices
);
2037 mddev
->level
= le32_to_cpu(sb
->level
);
2038 mddev
->layout
= le32_to_cpu(sb
->layout
);
2039 mddev
->chunk_sectors
= le32_to_cpu(sb
->stripe_sectors
);
2040 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
2041 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
2042 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_stripe_sectors
);
2043 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
2044 mddev
->array_sectors
= le64_to_cpu(sb
->array_sectors
);
2046 /* raid was reshaping and got interrupted */
2047 if (le32_to_cpu(sb
->flags
) & SB_FLAG_RESHAPE_ACTIVE
) {
2048 if (test_bit(__CTR_FLAG_DELTA_DISKS
, &rs
->ctr_flags
)) {
2049 DMERR("Reshape requested but raid set is still reshaping");
2053 if (mddev
->delta_disks
< 0 ||
2054 (!mddev
->delta_disks
&& (le32_to_cpu(sb
->flags
) & SB_FLAG_RESHAPE_BACKWARDS
)))
2055 mddev
->reshape_backwards
= 1;
2057 mddev
->reshape_backwards
= 0;
2059 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
2060 rs
->raid_type
= get_raid_type_by_ll(mddev
->level
, mddev
->layout
);
2065 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2067 if (le32_to_cpu(sb
->level
) != mddev
->level
) {
2068 DMERR("Reshaping/takeover raid sets not yet supported. (raid level/stripes/size change)");
2071 if (le32_to_cpu(sb
->layout
) != mddev
->layout
) {
2072 DMERR("Reshaping raid sets not yet supported. (raid layout change)");
2073 DMERR(" 0x%X vs 0x%X", le32_to_cpu(sb
->layout
), mddev
->layout
);
2074 DMERR(" Old layout: %s w/ %d copies",
2075 raid10_md_layout_to_format(le32_to_cpu(sb
->layout
)),
2076 raid10_md_layout_to_copies(le32_to_cpu(sb
->layout
)));
2077 DMERR(" New layout: %s w/ %d copies",
2078 raid10_md_layout_to_format(mddev
->layout
),
2079 raid10_md_layout_to_copies(mddev
->layout
));
2082 if (le32_to_cpu(sb
->stripe_sectors
) != mddev
->chunk_sectors
) {
2083 DMERR("Reshaping raid sets not yet supported. (stripe sectors change)");
2087 /* We can only change the number of devices in raid1 with old (i.e. pre 1.0.7) metadata */
2088 if (!rt_is_raid1(rs
->raid_type
) &&
2089 (le32_to_cpu(sb
->num_devices
) != mddev
->raid_disks
)) {
2090 DMERR("Reshaping raid sets not yet supported. (device count change from %u to %u)",
2091 sb
->num_devices
, mddev
->raid_disks
);
2095 /* Table line is checked vs. authoritative superblock */
2099 if (!test_bit(__CTR_FLAG_NOSYNC
, &rs
->ctr_flags
))
2100 mddev
->recovery_cp
= le64_to_cpu(sb
->array_resync_offset
);
2103 * During load, we set FirstUse if a new superblock was written.
2104 * There are two reasons we might not have a superblock:
2105 * 1) The raid set is brand new - in which case, all of the
2106 * devices must have their In_sync bit set. Also,
2107 * recovery_cp must be 0, unless forced.
2108 * 2) This is a new device being added to an old raid set
2109 * and the new device needs to be rebuilt - in which
2110 * case the In_sync bit will /not/ be set and
2111 * recovery_cp must be MaxSector.
2112 * 3) This is/are a new device(s) being added to an old
2113 * raid set during takeover to a higher raid level
2114 * to provide capacity for redundancy or during reshape
2115 * to add capacity to grow the raid set.
2118 rdev_for_each(r
, mddev
) {
2119 if (test_bit(FirstUse
, &r
->flags
))
2122 if (!test_bit(In_sync
, &r
->flags
)) {
2123 DMINFO("Device %d specified for rebuild; clearing superblock",
2127 if (test_bit(FirstUse
, &r
->flags
))
2134 if (new_devs
== rs
->raid_disks
|| !rebuilds
) {
2135 /* Replace a broken device */
2136 if (new_devs
== 1 && !rs
->delta_disks
)
2138 if (new_devs
== rs
->raid_disks
) {
2139 DMINFO("Superblocks created for new raid set");
2140 set_bit(MD_ARRAY_FIRST_USE
, &mddev
->flags
);
2141 } else if (new_devs
!= rebuilds
&&
2142 new_devs
!= rs
->delta_disks
) {
2143 DMERR("New device injected into existing raid set without "
2144 "'delta_disks' or 'rebuild' parameter specified");
2147 } else if (new_devs
&& new_devs
!= rebuilds
) {
2148 DMERR("%u 'rebuild' devices cannot be injected into"
2149 " a raid set with %u other first-time devices",
2150 rebuilds
, new_devs
);
2152 } else if (rebuilds
) {
2153 if (rebuild_and_new
&& rebuilds
!= rebuild_and_new
) {
2154 DMERR("new device%s provided without 'rebuild'",
2155 new_devs
> 1 ? "s" : "");
2157 } else if (rs_is_recovering(rs
)) {
2158 DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2159 (unsigned long long) mddev
->recovery_cp
);
2161 } else if (rs_is_reshaping(rs
)) {
2162 DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2163 (unsigned long long) mddev
->reshape_position
);
2169 * Now we set the Faulty bit for those devices that are
2170 * recorded in the superblock as failed.
2172 sb_retrieve_failed_devices(sb
, failed_devices
);
2173 rdev_for_each(r
, mddev
) {
2176 sb2
= page_address(r
->sb_page
);
2177 sb2
->failed_devices
= 0;
2178 memset(sb2
->extended_failed_devices
, 0, sizeof(sb2
->extended_failed_devices
));
2181 * Check for any device re-ordering.
2183 if (!test_bit(FirstUse
, &r
->flags
) && (r
->raid_disk
>= 0)) {
2184 role
= le32_to_cpu(sb2
->array_position
);
2188 if (role
!= r
->raid_disk
) {
2189 if (__is_raid10_near(mddev
->layout
)) {
2190 if (mddev
->raid_disks
% __raid10_near_copies(mddev
->layout
) ||
2191 rs
->raid_disks
% rs
->raid10_copies
) {
2193 "Cannot change raid10 near set to odd # of devices!";
2197 sb2
->array_position
= cpu_to_le32(r
->raid_disk
);
2199 } else if (!(rs_is_raid10(rs
) && rt_is_raid0(rs
->raid_type
)) &&
2200 !(rs_is_raid0(rs
) && rt_is_raid10(rs
->raid_type
)) &&
2201 !rt_is_raid1(rs
->raid_type
)) {
2202 rs
->ti
->error
= "Cannot change device positions in raid set";
2206 DMINFO("raid device #%d now at position #%d", role
, r
->raid_disk
);
2210 * Partial recovery is performed on
2211 * returning failed devices.
2213 if (test_bit(role
, (void *) failed_devices
))
2214 set_bit(Faulty
, &r
->flags
);
2221 static int super_validate(struct raid_set
*rs
, struct md_rdev
*rdev
)
2223 struct mddev
*mddev
= &rs
->md
;
2224 struct dm_raid_superblock
*sb
;
2226 if (rs_is_raid0(rs
) || !rdev
->sb_page
)
2229 sb
= page_address(rdev
->sb_page
);
2232 * If mddev->events is not set, we know we have not yet initialized
2235 if (!mddev
->events
&& super_init_validation(rs
, rdev
))
2238 if (le32_to_cpu(sb
->compat_features
) != FEATURE_FLAG_SUPPORTS_V190
) {
2239 rs
->ti
->error
= "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2243 if (sb
->incompat_features
) {
2244 rs
->ti
->error
= "Unable to assemble array: No incompatible feature flags supported yet";
2248 /* Enable bitmap creation for RAID levels != 0 */
2249 mddev
->bitmap_info
.offset
= rt_is_raid0(rs
->raid_type
) ? 0 : to_sector(4096);
2250 rdev
->mddev
->bitmap_info
.default_offset
= mddev
->bitmap_info
.offset
;
2252 if (!test_and_clear_bit(FirstUse
, &rdev
->flags
)) {
2253 /* Retrieve device size stored in superblock to be prepared for shrink */
2254 rdev
->sectors
= le64_to_cpu(sb
->sectors
);
2255 rdev
->recovery_offset
= le64_to_cpu(sb
->disk_recovery_offset
);
2256 if (rdev
->recovery_offset
== MaxSector
)
2257 set_bit(In_sync
, &rdev
->flags
);
2259 * If no reshape in progress -> we're recovering single
2260 * disk(s) and have to set the device(s) to out-of-sync
2262 else if (!rs_is_reshaping(rs
))
2263 clear_bit(In_sync
, &rdev
->flags
); /* Mandatory for recovery */
2267 * If a device comes back, set it as not In_sync and no longer faulty.
2269 if (test_and_clear_bit(Faulty
, &rdev
->flags
)) {
2270 rdev
->recovery_offset
= 0;
2271 clear_bit(In_sync
, &rdev
->flags
);
2272 rdev
->saved_raid_disk
= rdev
->raid_disk
;
2275 /* Reshape support -> restore repective data offsets */
2276 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
2277 rdev
->new_data_offset
= le64_to_cpu(sb
->new_data_offset
);
2283 * Analyse superblocks and select the freshest.
2285 static int analyse_superblocks(struct dm_target
*ti
, struct raid_set
*rs
)
2288 struct raid_dev
*dev
;
2289 struct md_rdev
*rdev
, *tmp
, *freshest
;
2290 struct mddev
*mddev
= &rs
->md
;
2293 rdev_for_each_safe(rdev
, tmp
, mddev
) {
2295 * Skipping super_load due to CTR_FLAG_SYNC will cause
2296 * the array to undergo initialization again as
2297 * though it were new. This is the intended effect
2298 * of the "sync" directive.
2300 * When reshaping capability is added, we must ensure
2301 * that the "sync" directive is disallowed during the
2304 if (test_bit(__CTR_FLAG_SYNC
, &rs
->ctr_flags
))
2307 if (!rdev
->meta_bdev
)
2310 r
= super_load(rdev
, freshest
);
2319 dev
= container_of(rdev
, struct raid_dev
, rdev
);
2321 dm_put_device(ti
, dev
->meta_dev
);
2323 dev
->meta_dev
= NULL
;
2324 rdev
->meta_bdev
= NULL
;
2327 put_page(rdev
->sb_page
);
2329 rdev
->sb_page
= NULL
;
2331 rdev
->sb_loaded
= 0;
2334 * We might be able to salvage the data device
2335 * even though the meta device has failed. For
2336 * now, we behave as though '- -' had been
2337 * set for this device in the table.
2340 dm_put_device(ti
, dev
->data_dev
);
2342 dev
->data_dev
= NULL
;
2345 list_del(&rdev
->same_set
);
2352 if (validate_raid_redundancy(rs
)) {
2353 rs
->ti
->error
= "Insufficient redundancy to activate array";
2358 * Validation of the freshest device provides the source of
2359 * validation for the remaining devices.
2361 rs
->ti
->error
= "Unable to assemble array: Invalid superblocks";
2362 if (super_validate(rs
, freshest
))
2365 rdev_for_each(rdev
, mddev
)
2366 if ((rdev
!= freshest
) && super_validate(rs
, rdev
))
2372 * Adjust data_offset and new_data_offset on all disk members of @rs
2373 * for out of place reshaping if requested by contructor
2375 * We need free space at the beginning of each raid disk for forward
2376 * and at the end for backward reshapes which userspace has to provide
2377 * via remapping/reordering of space.
2379 static int rs_adjust_data_offsets(struct raid_set
*rs
)
2381 sector_t data_offset
= 0, new_data_offset
= 0;
2382 struct md_rdev
*rdev
;
2384 /* Constructor did not request data offset change */
2385 if (!test_bit(__CTR_FLAG_DATA_OFFSET
, &rs
->ctr_flags
)) {
2386 if (!rs_is_reshapable(rs
))
2392 /* HM FIXME: get InSync raid_dev? */
2393 rdev
= &rs
->dev
[0].rdev
;
2395 if (rs
->delta_disks
< 0) {
2397 * Removing disks (reshaping backwards):
2399 * - before reshape: data is at offset 0 and free space
2400 * is at end of each component LV
2402 * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2405 new_data_offset
= rs
->data_offset
;
2407 } else if (rs
->delta_disks
> 0) {
2409 * Adding disks (reshaping forwards):
2411 * - before reshape: data is at offset rs->data_offset != 0 and
2412 * free space is at begin of each component LV
2414 * - after reshape: data is at offset 0 on each component LV
2416 data_offset
= rs
->data_offset
;
2417 new_data_offset
= 0;
2421 * User space passes in 0 for data offset after having removed reshape space
2423 * - or - (data offset != 0)
2425 * Changing RAID layout or chunk size -> toggle offsets
2427 * - before reshape: data is at offset rs->data_offset 0 and
2428 * free space is at end of each component LV
2430 * data is at offset rs->data_offset != 0 and
2431 * free space is at begin of each component LV
2433 * - after reshape: data is at offset 0 if it was at offset != 0
2434 * or at offset != 0 if it was at offset 0
2435 * on each component LV
2438 data_offset
= rs
->data_offset
? rdev
->data_offset
: 0;
2439 new_data_offset
= data_offset
? 0 : rs
->data_offset
;
2440 set_bit(RT_FLAG_UPDATE_SBS
, &rs
->runtime_flags
);
2444 * Make sure we got a minimum amount of free sectors per device
2446 if (rs
->data_offset
&&
2447 to_sector(i_size_read(rdev
->bdev
->bd_inode
)) - rdev
->sectors
< MIN_FREE_RESHAPE_SPACE
) {
2448 rs
->ti
->error
= data_offset
? "No space for forward reshape" :
2449 "No space for backward reshape";
2453 /* Adjust data offsets on all rdevs */
2454 rdev_for_each(rdev
, &rs
->md
) {
2455 rdev
->data_offset
= data_offset
;
2456 rdev
->new_data_offset
= new_data_offset
;
2462 /* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2463 static void __reorder_raid_disk_indexes(struct raid_set
*rs
)
2466 struct md_rdev
*rdev
;
2468 rdev_for_each(rdev
, &rs
->md
) {
2469 rdev
->raid_disk
= i
++;
2470 rdev
->saved_raid_disk
= rdev
->new_raid_disk
= -1;
2475 * Setup @rs for takeover by a different raid level
2477 static int rs_setup_takeover(struct raid_set
*rs
)
2479 struct mddev
*mddev
= &rs
->md
;
2480 struct md_rdev
*rdev
;
2481 unsigned int d
= mddev
->raid_disks
= rs
->raid_disks
;
2482 sector_t new_data_offset
= rs
->dev
[0].rdev
.data_offset
? 0 : rs
->data_offset
;
2484 if (rt_is_raid10(rs
->raid_type
)) {
2485 if (mddev
->level
== 0) {
2486 /* Userpace reordered disks -> adjust raid_disk indexes */
2487 __reorder_raid_disk_indexes(rs
);
2489 /* raid0 -> raid10_far layout */
2490 mddev
->layout
= raid10_format_to_md_layout(rs
, ALGORITHM_RAID10_FAR
,
2492 } else if (mddev
->level
== 1)
2493 /* raid1 -> raid10_near layout */
2494 mddev
->layout
= raid10_format_to_md_layout(rs
, ALGORITHM_RAID10_NEAR
,
2501 clear_bit(MD_ARRAY_FIRST_USE
, &mddev
->flags
);
2502 mddev
->recovery_cp
= MaxSector
;
2505 rdev
= &rs
->dev
[d
].rdev
;
2507 if (test_bit(d
, (void *) rs
->rebuild_disks
)) {
2508 clear_bit(In_sync
, &rdev
->flags
);
2509 clear_bit(Faulty
, &rdev
->flags
);
2510 mddev
->recovery_cp
= rdev
->recovery_offset
= 0;
2511 /* Bitmap has to be created when we do an "up" takeover */
2512 set_bit(MD_ARRAY_FIRST_USE
, &mddev
->flags
);
2515 rdev
->new_data_offset
= new_data_offset
;
2523 * - change raid layout
2524 * - change chunk size
2528 static int rs_setup_reshape(struct raid_set
*rs
)
2531 unsigned int cur_raid_devs
, d
;
2532 struct mddev
*mddev
= &rs
->md
;
2533 struct md_rdev
*rdev
;
2535 mddev
->delta_disks
= rs
->delta_disks
;
2536 cur_raid_devs
= mddev
->raid_disks
;
2538 /* Ignore impossible layout change whilst adding/removing disks */
2539 if (mddev
->delta_disks
&&
2540 mddev
->layout
!= mddev
->new_layout
) {
2541 DMINFO("Ignoring invalid layout change with delta_disks=%d", rs
->delta_disks
);
2542 mddev
->new_layout
= mddev
->layout
;
2546 * Adjust array size:
2548 * - in case of adding disks, array size has
2549 * to grow after the disk adding reshape,
2550 * which'll hapen in the event handler;
2551 * reshape will happen forward, so space has to
2552 * be available at the beginning of each disk
2554 * - in case of removing disks, array size
2555 * has to shrink before starting the reshape,
2556 * which'll happen here;
2557 * reshape will happen backward, so space has to
2558 * be available at the end of each disk
2560 * - data_offset and new_data_offset are
2561 * adjusted for aforementioned out of place
2562 * reshaping based on userspace passing in
2563 * the "data_offset <sectors>" key/value
2564 * pair via the constructor
2568 if (rs
->delta_disks
> 0) {
2569 /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2570 for (d
= cur_raid_devs
; d
< rs
->raid_disks
; d
++) {
2571 rdev
= &rs
->dev
[d
].rdev
;
2572 clear_bit(In_sync
, &rdev
->flags
);
2575 * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2576 * by md, which'll store that erroneously in the superblock on reshape
2578 rdev
->saved_raid_disk
= -1;
2579 rdev
->raid_disk
= d
;
2581 rdev
->sectors
= mddev
->dev_sectors
;
2582 rdev
->recovery_offset
= MaxSector
;
2585 mddev
->reshape_backwards
= 0; /* adding disks -> forward reshape */
2587 /* Remove disk(s) */
2588 } else if (rs
->delta_disks
< 0) {
2589 r
= rs_set_dev_and_array_sectors(rs
, true);
2590 mddev
->reshape_backwards
= 1; /* removing disk(s) -> backward reshape */
2592 /* Change layout and/or chunk size */
2595 * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2597 * keeping number of disks and do layout change ->
2599 * toggle reshape_backward depending on data_offset:
2601 * - free space upfront -> reshape forward
2603 * - free space at the end -> reshape backward
2606 * This utilizes free reshape space avoiding the need
2607 * for userspace to move (parts of) LV segments in
2608 * case of layout/chunksize change (for disk
2609 * adding/removing reshape space has to be at
2610 * the proper address (see above with delta_disks):
2612 * add disk(s) -> begin
2613 * remove disk(s)-> end
2615 mddev
->reshape_backwards
= rs
->dev
[0].rdev
.data_offset
? 0 : 1;
2622 * Enable/disable discard support on RAID set depending on
2623 * RAID level and discard properties of underlying RAID members.
2625 static void configure_discard_support(struct raid_set
*rs
)
2629 struct dm_target
*ti
= rs
->ti
;
2631 /* Assume discards not supported until after checks below. */
2632 ti
->discards_supported
= false;
2634 /* RAID level 4,5,6 require discard_zeroes_data for data integrity! */
2635 raid456
= (rs
->md
.level
== 4 || rs
->md
.level
== 5 || rs
->md
.level
== 6);
2637 for (i
= 0; i
< rs
->md
.raid_disks
; i
++) {
2638 struct request_queue
*q
;
2640 if (!rs
->dev
[i
].rdev
.bdev
)
2643 q
= bdev_get_queue(rs
->dev
[i
].rdev
.bdev
);
2644 if (!q
|| !blk_queue_discard(q
))
2648 if (!q
->limits
.discard_zeroes_data
)
2650 if (!devices_handle_discard_safely
) {
2651 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2652 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2658 /* All RAID members properly support discards */
2659 ti
->discards_supported
= true;
2662 * RAID1 and RAID10 personalities require bio splitting,
2663 * RAID0/4/5/6 don't and process large discard bios properly.
2665 ti
->split_discard_bios
= !!(rs
->md
.level
== 1 || rs
->md
.level
== 10);
2666 ti
->num_discard_bios
= 1;
2670 * Construct a RAID0/1/10/4/5/6 mapping:
2672 * <raid_type> <#raid_params> <raid_params>{0,} \
2673 * <#raid_devs> [<meta_dev1> <dev1>]{1,}
2675 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
2676 * details on possible <raid_params>.
2678 * Userspace is free to initialize the metadata devices, hence the superblocks to
2679 * enforce recreation based on the passed in table parameters.
2682 static int raid_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2685 bool resize
= false;
2686 struct raid_type
*rt
;
2687 unsigned num_raid_params
, num_raid_devs
;
2688 sector_t calculated_dev_sectors
;
2689 struct raid_set
*rs
= NULL
;
2691 struct rs_layout rs_layout
;
2692 struct dm_arg_set as
= { argc
, argv
}, as_nrd
;
2693 struct dm_arg _args
[] = {
2694 { 0, as
.argc
, "Cannot understand number of raid parameters" },
2695 { 1, 254, "Cannot understand number of raid devices parameters" }
2698 /* Must have <raid_type> */
2699 arg
= dm_shift_arg(&as
);
2701 ti
->error
= "No arguments";
2705 rt
= get_raid_type(arg
);
2707 ti
->error
= "Unrecognised raid_type";
2711 /* Must have <#raid_params> */
2712 if (dm_read_arg_group(_args
, &as
, &num_raid_params
, &ti
->error
))
2715 /* number of raid device tupples <meta_dev data_dev> */
2717 dm_consume_args(&as_nrd
, num_raid_params
);
2718 _args
[1].max
= (as_nrd
.argc
- 1) / 2;
2719 if (dm_read_arg(_args
+ 1, &as_nrd
, &num_raid_devs
, &ti
->error
))
2722 if (!__within_range(num_raid_devs
, 1, MAX_RAID_DEVICES
)) {
2723 ti
->error
= "Invalid number of supplied raid devices";
2727 rs
= raid_set_alloc(ti
, rt
, num_raid_devs
);
2731 r
= parse_raid_params(rs
, &as
, num_raid_params
);
2735 r
= parse_dev_params(rs
, &as
);
2739 rs
->md
.sync_super
= super_sync
;
2742 * Calculate ctr requested array and device sizes to allow
2743 * for superblock analysis needing device sizes defined.
2745 * Any existing superblock will overwrite the array and device sizes
2747 r
= rs_set_dev_and_array_sectors(rs
, false);
2751 calculated_dev_sectors
= rs
->dev
[0].rdev
.sectors
;
2754 * Backup any new raid set level, layout, ...
2755 * requested to be able to compare to superblock
2756 * members for conversion decisions.
2758 rs_config_backup(rs
, &rs_layout
);
2760 r
= analyse_superblocks(ti
, rs
);
2764 resize
= calculated_dev_sectors
!= rs
->dev
[0].rdev
.sectors
;
2766 INIT_WORK(&rs
->md
.event_work
, do_table_event
);
2768 ti
->num_flush_bios
= 1;
2770 /* Restore any requested new layout for conversion decision */
2771 rs_config_restore(rs
, &rs_layout
);
2773 if (test_bit(MD_ARRAY_FIRST_USE
, &rs
->md
.flags
)) {
2774 /* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
2775 if (rs_is_raid6(rs
) &&
2776 test_bit(__CTR_FLAG_NOSYNC
, &rs
->ctr_flags
)) {
2777 ti
->error
= "'nosync' not allowed for new raid6 set";
2781 rs_setup_recovery(rs
, 0);
2782 set_bit(RT_FLAG_UPDATE_SBS
, &rs
->runtime_flags
);
2784 } else if (rs_is_recovering(rs
)) {
2785 ; /* skip setup rs */
2786 } else if (rs_is_reshaping(rs
)) {
2787 /* Have to reject size change request during reshape */
2789 ti
->error
= "Can't resize a reshaping raid set";
2793 ; /* skip setup rs */
2794 } else if (rs_takeover_requested(rs
)) {
2795 if (rs_is_reshaping(rs
)) {
2796 ti
->error
= "Can't takeover a reshaping raid set";
2802 * If a takeover is needed, userspace sets any additional
2803 * devices to rebuild, so set the level to the new requested
2804 * one, prohibit requesting recovery, allow the raid
2805 * set to run and store superblocks during resume.
2807 r
= rs_check_takeover(rs
);
2811 r
= rs_setup_takeover(rs
);
2815 set_bit(RT_FLAG_UPDATE_SBS
, &rs
->runtime_flags
);
2816 set_bit(RT_FLAG_KEEP_RS_FROZEN
, &rs
->runtime_flags
);
2817 rs_setup_recovery(rs
, MaxSector
);
2819 } else if (rs_reshape_requested(rs
)) {
2820 if (rs_is_reshaping(rs
)) {
2821 ti
->error
= "raid set already reshaping!";
2826 if (rs_is_raid10(rs
)) {
2827 if (rs
->raid_disks
!= rs
->md
.raid_disks
&&
2828 __is_raid10_near(rs
->md
.layout
) &&
2829 rs
->raid10_copies
&&
2830 rs
->raid10_copies
!= __raid10_near_copies(rs
->md
.layout
)) {
2832 * raid disk have to be multiple of data copies to allow this conversion,
2834 * This is actually not a reshape it is a
2835 * rebuild of any additional mirrors per group
2837 if (rs
->raid_disks
% rs
->raid10_copies
) {
2838 ti
->error
= "Can't reshape raid10 mirror groups";
2843 /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2844 __reorder_raid_disk_indexes(rs
);
2845 rs
->md
.layout
= raid10_format_to_md_layout(rs
, ALGORITHM_RAID10_NEAR
,
2847 rs
->md
.new_layout
= rs
->md
.layout
;
2850 set_bit(RT_FLAG_RESHAPE_RS
, &rs
->runtime_flags
);
2852 } else if (rs_is_raid456(rs
))
2853 set_bit(RT_FLAG_RESHAPE_RS
, &rs
->runtime_flags
);
2856 * HM FIXME: process raid1 via delta_disks as well?
2857 * Would cause allocations in raid1->check_reshape
2858 * though, thus more issues with potential failures
2860 else if (rs_is_raid1(rs
)) {
2861 set_bit(RT_FLAG_KEEP_RS_FROZEN
, &rs
->runtime_flags
);
2862 rs
->md
.raid_disks
= rs
->raid_disks
;
2865 if (test_bit(RT_FLAG_RESHAPE_RS
, &rs
->runtime_flags
)) {
2866 set_bit(RT_FLAG_UPDATE_SBS
, &rs
->runtime_flags
);
2867 set_bit(RT_FLAG_KEEP_RS_FROZEN
, &rs
->runtime_flags
);
2870 /* Create new superblocks and bitmaps, if any */
2871 if (rs
->md
.raid_disks
< rs
->raid_disks
)
2872 set_bit(RT_FLAG_UPDATE_SBS
, &rs
->runtime_flags
);
2874 rs_setup_recovery(rs
, MaxSector
);
2877 /* May not set recovery when a device rebuild is requested */
2878 if (test_bit(__CTR_FLAG_REBUILD
, &rs
->ctr_flags
)) {
2879 rs_setup_recovery(rs
, MaxSector
);
2880 set_bit(RT_FLAG_UPDATE_SBS
, &rs
->runtime_flags
);
2882 rs_setup_recovery(rs
, test_bit(__CTR_FLAG_SYNC
, &rs
->ctr_flags
) ?
2883 0 : (resize
? calculated_dev_sectors
: MaxSector
));
2887 /* If constructor requested it, change data and new_data offsets */
2888 r
= rs_adjust_data_offsets(rs
);
2892 /* Start raid set read-only and assumed clean to change in raid_resume() */
2895 set_bit(MD_RECOVERY_FROZEN
, &rs
->md
.recovery
);
2897 /* Has to be held on running the array */
2898 mddev_lock_nointr(&rs
->md
);
2899 r
= md_run(&rs
->md
);
2900 rs
->md
.in_sync
= 0; /* Assume already marked dirty */
2903 ti
->error
= "Failed to run raid array";
2904 mddev_unlock(&rs
->md
);
2908 rs
->callbacks
.congested_fn
= raid_is_congested
;
2909 dm_table_add_target_callbacks(ti
->table
, &rs
->callbacks
);
2911 mddev_suspend(&rs
->md
);
2913 /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
2914 if (rs_is_raid456(rs
)) {
2915 r
= rs_set_raid456_stripe_cache(rs
);
2917 goto bad_stripe_cache
;
2920 /* Now do an early reshape check */
2921 if (test_bit(RT_FLAG_RESHAPE_RS
, &rs
->runtime_flags
)) {
2922 r
= rs_check_reshape(rs
);
2924 goto bad_check_reshape
;
2926 /* Restore new, ctr requested layout to perform check */
2927 rs_config_restore(rs
, &rs_layout
);
2929 r
= rs
->md
.pers
->check_reshape(&rs
->md
);
2931 ti
->error
= "Reshape check failed";
2932 goto bad_check_reshape
;
2936 mddev_unlock(&rs
->md
);
2948 static void raid_dtr(struct dm_target
*ti
)
2950 struct raid_set
*rs
= ti
->private;
2952 list_del_init(&rs
->callbacks
.list
);
2957 static int raid_map(struct dm_target
*ti
, struct bio
*bio
)
2959 struct raid_set
*rs
= ti
->private;
2960 struct mddev
*mddev
= &rs
->md
;
2963 * If we're reshaping to add disk(s)), ti->len and
2964 * mddev->array_sectors will differ during the process
2965 * (ti->len > mddev->array_sectors), so we have to requeue
2966 * bios with addresses > mddev->array_sectors here or
2967 * there will occur accesses past EOD of the component
2968 * data images thus erroring the raid set.
2970 if (unlikely(bio_end_sector(bio
) > mddev
->array_sectors
))
2971 return DM_MAPIO_REQUEUE
;
2973 mddev
->pers
->make_request(mddev
, bio
);
2975 return DM_MAPIO_SUBMITTED
;
2978 /* Return string describing the current sync action of @mddev */
2979 static const char *decipher_sync_action(struct mddev
*mddev
)
2981 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
2984 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2985 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
2986 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
2989 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
2990 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
2992 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
2997 if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
3005 * Return status string @rdev
3007 * Status characters:
3009 * 'D' = Dead/Failed device
3010 * 'a' = Alive but not in-sync
3011 * 'A' = Alive and in-sync
3013 static const char *__raid_dev_status(struct md_rdev
*rdev
, bool array_in_sync
)
3015 if (test_bit(Faulty
, &rdev
->flags
))
3017 else if (!array_in_sync
|| !test_bit(In_sync
, &rdev
->flags
))
3023 /* Helper to return resync/reshape progress for @rs and @array_in_sync */
3024 static sector_t
rs_get_progress(struct raid_set
*rs
,
3025 sector_t resync_max_sectors
, bool *array_in_sync
)
3027 sector_t r
, recovery_cp
, curr_resync_completed
;
3028 struct mddev
*mddev
= &rs
->md
;
3030 curr_resync_completed
= mddev
->curr_resync_completed
?: mddev
->recovery_cp
;
3031 recovery_cp
= mddev
->recovery_cp
;
3032 *array_in_sync
= false;
3034 if (rs_is_raid0(rs
)) {
3035 r
= resync_max_sectors
;
3036 *array_in_sync
= true;
3039 r
= mddev
->reshape_position
;
3041 /* Reshape is relative to the array size */
3042 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) ||
3044 if (r
== MaxSector
) {
3045 *array_in_sync
= true;
3046 r
= resync_max_sectors
;
3048 /* Got to reverse on backward reshape */
3049 if (mddev
->reshape_backwards
)
3050 r
= mddev
->array_sectors
- r
;
3052 /* Devide by # of data stripes */
3053 sector_div(r
, mddev_data_stripes(rs
));
3056 /* Sync is relative to the component device size */
3057 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3058 r
= curr_resync_completed
;
3062 if (r
== MaxSector
) {
3066 *array_in_sync
= true;
3067 r
= resync_max_sectors
;
3068 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
3070 * If "check" or "repair" is occurring, the raid set has
3071 * undergone an initial sync and the health characters
3072 * should not be 'a' anymore.
3074 *array_in_sync
= true;
3076 struct md_rdev
*rdev
;
3079 * The raid set may be doing an initial sync, or it may
3080 * be rebuilding individual components. If all the
3081 * devices are In_sync, then it is the raid set that is
3082 * being initialized.
3084 rdev_for_each(rdev
, mddev
)
3085 if (!test_bit(In_sync
, &rdev
->flags
))
3086 *array_in_sync
= true;
3088 r
= 0; /* HM FIXME: TESTME: https://bugzilla.redhat.com/show_bug.cgi?id=1210637 ? */
3096 /* Helper to return @dev name or "-" if !@dev */
3097 static const char *__get_dev_name(struct dm_dev
*dev
)
3099 return dev
? dev
->name
: "-";
3102 static void raid_status(struct dm_target
*ti
, status_type_t type
,
3103 unsigned int status_flags
, char *result
, unsigned int maxlen
)
3105 struct raid_set
*rs
= ti
->private;
3106 struct mddev
*mddev
= &rs
->md
;
3107 struct r5conf
*conf
= mddev
->private;
3108 int max_nr_stripes
= conf
? conf
->max_nr_stripes
: 0;
3110 unsigned int raid_param_cnt
= 1; /* at least 1 for chunksize */
3111 unsigned int sz
= 0;
3112 unsigned int write_mostly_params
= 0;
3113 sector_t progress
, resync_max_sectors
, resync_mismatches
;
3114 const char *sync_action
;
3115 struct raid_type
*rt
;
3116 struct md_rdev
*rdev
;
3119 case STATUSTYPE_INFO
:
3120 /* *Should* always succeed */
3121 rt
= get_raid_type_by_ll(mddev
->new_level
, mddev
->new_layout
);
3125 DMEMIT("%s %d ", rt
->name
, mddev
->raid_disks
);
3127 /* Access most recent mddev properties for status output */
3129 /* Get sensible max sectors even if raid set not yet started */
3130 resync_max_sectors
= test_bit(RT_FLAG_RS_PRERESUMED
, &rs
->runtime_flags
) ?
3131 mddev
->resync_max_sectors
: mddev
->dev_sectors
;
3132 progress
= rs_get_progress(rs
, resync_max_sectors
, &array_in_sync
);
3133 resync_mismatches
= (mddev
->last_sync_action
&& !strcasecmp(mddev
->last_sync_action
, "check")) ?
3134 atomic64_read(&mddev
->resync_mismatches
) : 0;
3135 sync_action
= decipher_sync_action(&rs
->md
);
3137 /* HM FIXME: do we want another state char for raid0? It shows 'D' or 'A' now */
3138 rdev_for_each(rdev
, mddev
)
3139 DMEMIT(__raid_dev_status(rdev
, array_in_sync
));
3142 * In-sync/Reshape ratio:
3143 * The in-sync ratio shows the progress of:
3144 * - Initializing the raid set
3145 * - Rebuilding a subset of devices of the raid set
3146 * The user can distinguish between the two by referring
3147 * to the status characters.
3149 * The reshape ratio shows the progress of
3150 * changing the raid layout or the number of
3151 * disks of a raid set
3153 DMEMIT(" %llu/%llu", (unsigned long long) progress
,
3154 (unsigned long long) resync_max_sectors
);
3160 * See Documentation/device-mapper/dm-raid.txt for
3161 * information on each of these states.
3163 DMEMIT(" %s", sync_action
);
3168 * resync_mismatches/mismatch_cnt
3169 * This field shows the number of discrepancies found when
3170 * performing a "check" of the raid set.
3172 DMEMIT(" %llu", (unsigned long long) resync_mismatches
);
3177 * data_offset (needed for out of space reshaping)
3178 * This field shows the data offset into the data
3179 * image LV where the first stripes data starts.
3181 * We keep data_offset equal on all raid disks of the set,
3182 * so retrieving it from the first raid disk is sufficient.
3184 DMEMIT(" %llu", (unsigned long long) rs
->dev
[0].rdev
.data_offset
);
3187 case STATUSTYPE_TABLE
:
3188 /* Report the table line string you would use to construct this raid set */
3190 /* Calculate raid parameter count */
3191 rdev_for_each(rdev
, mddev
)
3192 if (test_bit(WriteMostly
, &rdev
->flags
))
3193 write_mostly_params
+= 2;
3194 raid_param_cnt
+= memweight(rs
->rebuild_disks
,
3195 DISKS_ARRAY_ELEMS
* sizeof(*rs
->rebuild_disks
)) * 2 +
3196 write_mostly_params
+
3197 hweight32(rs
->ctr_flags
& CTR_FLAG_OPTIONS_NO_ARGS
) +
3198 hweight32(rs
->ctr_flags
& CTR_FLAG_OPTIONS_ONE_ARG
) * 2;
3199 /* Emit table line */
3200 DMEMIT("%s %u %u", rs
->raid_type
->name
, raid_param_cnt
, mddev
->new_chunk_sectors
);
3201 if (test_bit(__CTR_FLAG_RAID10_FORMAT
, &rs
->ctr_flags
))
3202 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT
),
3203 raid10_md_layout_to_format(mddev
->layout
));
3204 if (test_bit(__CTR_FLAG_RAID10_COPIES
, &rs
->ctr_flags
))
3205 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES
),
3206 raid10_md_layout_to_copies(mddev
->layout
));
3207 if (test_bit(__CTR_FLAG_NOSYNC
, &rs
->ctr_flags
))
3208 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC
));
3209 if (test_bit(__CTR_FLAG_SYNC
, &rs
->ctr_flags
))
3210 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC
));
3211 if (test_bit(__CTR_FLAG_REGION_SIZE
, &rs
->ctr_flags
))
3212 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE
),
3213 (unsigned long long) to_sector(mddev
->bitmap_info
.chunksize
));
3214 if (test_bit(__CTR_FLAG_DATA_OFFSET
, &rs
->ctr_flags
))
3215 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET
),
3216 (unsigned long long) rs
->data_offset
);
3217 if (test_bit(__CTR_FLAG_DAEMON_SLEEP
, &rs
->ctr_flags
))
3218 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP
),
3219 mddev
->bitmap_info
.daemon_sleep
);
3220 if (test_bit(__CTR_FLAG_DELTA_DISKS
, &rs
->ctr_flags
))
3221 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS
),
3222 mddev
->delta_disks
);
3223 if (test_bit(__CTR_FLAG_STRIPE_CACHE
, &rs
->ctr_flags
))
3224 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE
),
3226 rdev_for_each(rdev
, mddev
)
3227 if (test_bit(rdev
->raid_disk
, (void *) rs
->rebuild_disks
))
3228 DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD
),
3230 rdev_for_each(rdev
, mddev
)
3231 if (test_bit(WriteMostly
, &rdev
->flags
))
3232 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY
),
3234 if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND
, &rs
->ctr_flags
))
3235 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND
),
3236 mddev
->bitmap_info
.max_write_behind
);
3237 if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE
, &rs
->ctr_flags
))
3238 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE
),
3239 mddev
->sync_speed_max
);
3240 if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE
, &rs
->ctr_flags
))
3241 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE
),
3242 mddev
->sync_speed_min
);
3243 DMEMIT(" %d", rs
->raid_disks
);
3244 rdev_for_each(rdev
, mddev
) {
3245 struct raid_dev
*rd
= container_of(rdev
, struct raid_dev
, rdev
);
3247 DMEMIT(" %s %s", __get_dev_name(rd
->meta_dev
),
3248 __get_dev_name(rd
->data_dev
));
3253 static int raid_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
3255 struct raid_set
*rs
= ti
->private;
3256 struct mddev
*mddev
= &rs
->md
;
3258 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
3261 if (!strcasecmp(argv
[0], "frozen"))
3262 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3264 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3266 if (!strcasecmp(argv
[0], "idle") || !strcasecmp(argv
[0], "frozen")) {
3267 if (mddev
->sync_thread
) {
3268 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3269 md_reap_sync_thread(mddev
);
3271 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3272 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
3274 else if (!strcasecmp(argv
[0], "resync"))
3275 ; /* MD_RECOVERY_NEEDED set below */
3276 else if (!strcasecmp(argv
[0], "recover"))
3277 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
3279 if (!strcasecmp(argv
[0], "check"))
3280 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
3281 else if (!!strcasecmp(argv
[0], "repair"))
3283 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
3284 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
3286 if (mddev
->ro
== 2) {
3287 /* A write to sync_action is enough to justify
3288 * canceling read-auto mode
3291 if (!mddev
->suspended
&& mddev
->sync_thread
)
3292 md_wakeup_thread(mddev
->sync_thread
);
3294 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3295 if (!mddev
->suspended
&& mddev
->thread
)
3296 md_wakeup_thread(mddev
->thread
);
3301 static int raid_iterate_devices(struct dm_target
*ti
,
3302 iterate_devices_callout_fn fn
, void *data
)
3304 struct raid_set
*rs
= ti
->private;
3308 for (i
= 0; !r
&& i
< rs
->md
.raid_disks
; i
++)
3309 if (rs
->dev
[i
].data_dev
)
3311 rs
->dev
[i
].data_dev
,
3312 0, /* No offset on data devs */
3319 static void raid_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
3321 struct raid_set
*rs
= ti
->private;
3322 unsigned chunk_size
= rs
->md
.chunk_sectors
<< 9;
3323 struct r5conf
*conf
= rs
->md
.private;
3325 blk_limits_io_min(limits
, chunk_size
);
3326 blk_limits_io_opt(limits
, chunk_size
* (conf
->raid_disks
- conf
->max_degraded
));
3329 static void raid_presuspend(struct dm_target
*ti
)
3331 struct raid_set
*rs
= ti
->private;
3333 md_stop_writes(&rs
->md
);
3336 static void raid_postsuspend(struct dm_target
*ti
)
3338 struct raid_set
*rs
= ti
->private;
3340 if (test_and_clear_bit(RT_FLAG_RS_RESUMED
, &rs
->runtime_flags
)) {
3341 if (!rs
->md
.suspended
)
3342 mddev_suspend(&rs
->md
);
3347 static void attempt_restore_of_faulty_devices(struct raid_set
*rs
)
3350 uint64_t failed_devices
, cleared_failed_devices
= 0;
3351 unsigned long flags
;
3352 struct dm_raid_superblock
*sb
;
3355 for (i
= 0; i
< rs
->md
.raid_disks
; i
++) {
3356 r
= &rs
->dev
[i
].rdev
;
3357 if (test_bit(Faulty
, &r
->flags
) && r
->sb_page
&&
3358 sync_page_io(r
, 0, r
->sb_size
, r
->sb_page
,
3359 REQ_OP_READ
, 0, true)) {
3360 DMINFO("Faulty %s device #%d has readable super block."
3361 " Attempting to revive it.",
3362 rs
->raid_type
->name
, i
);
3365 * Faulty bit may be set, but sometimes the array can
3366 * be suspended before the personalities can respond
3367 * by removing the device from the array (i.e. calling
3368 * 'hot_remove_disk'). If they haven't yet removed
3369 * the failed device, its 'raid_disk' number will be
3370 * '>= 0' - meaning we must call this function
3373 if ((r
->raid_disk
>= 0) &&
3374 (r
->mddev
->pers
->hot_remove_disk(r
->mddev
, r
) != 0))
3375 /* Failed to revive this device, try next */
3379 r
->saved_raid_disk
= i
;
3381 clear_bit(Faulty
, &r
->flags
);
3382 clear_bit(WriteErrorSeen
, &r
->flags
);
3383 clear_bit(In_sync
, &r
->flags
);
3384 if (r
->mddev
->pers
->hot_add_disk(r
->mddev
, r
)) {
3386 r
->saved_raid_disk
= -1;
3389 r
->recovery_offset
= 0;
3390 cleared_failed_devices
|= 1 << i
;
3394 if (cleared_failed_devices
) {
3395 rdev_for_each(r
, &rs
->md
) {
3396 sb
= page_address(r
->sb_page
);
3397 failed_devices
= le64_to_cpu(sb
->failed_devices
);
3398 failed_devices
&= ~cleared_failed_devices
;
3399 sb
->failed_devices
= cpu_to_le64(failed_devices
);
3404 static int __load_dirty_region_bitmap(struct raid_set
*rs
)
3408 /* Try loading the bitmap unless "raid0", which does not have one */
3409 if (!rs_is_raid0(rs
) &&
3410 !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED
, &rs
->runtime_flags
)) {
3411 r
= bitmap_load(&rs
->md
);
3413 DMERR("Failed to load bitmap");
3419 /* Enforce updating all superblocks */
3420 static void rs_update_sbs(struct raid_set
*rs
)
3422 struct mddev
*mddev
= &rs
->md
;
3425 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3427 md_update_sb(mddev
, 1);
3432 * Reshape changes raid algorithm of @rs to new one within personality
3433 * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3434 * disks from a raid set thus growing/shrinking it or resizes the set
3436 * Call mddev_lock_nointr() before!
3438 static int rs_start_reshape(struct raid_set
*rs
)
3441 struct mddev
*mddev
= &rs
->md
;
3442 struct md_personality
*pers
= mddev
->pers
;
3444 r
= rs_setup_reshape(rs
);
3448 /* Need to be resumed to be able to start reshape, recovery is frozen until raid_resume() though */
3449 if (mddev
->suspended
)
3450 mddev_resume(mddev
);
3453 * Check any reshape constraints enforced by the personalility
3455 * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3457 r
= pers
->check_reshape(mddev
);
3459 rs
->ti
->error
= "pers->check_reshape() failed";
3464 * Personality may not provide start reshape method in which
3465 * case check_reshape above has already covered everything
3467 if (pers
->start_reshape
) {
3468 r
= pers
->start_reshape(mddev
);
3470 rs
->ti
->error
= "pers->start_reshape() failed";
3475 /* Suspend because a resume will happen in raid_resume() */
3476 if (!mddev
->suspended
)
3477 mddev_suspend(mddev
);
3480 * Now reshape got set up, update superblocks to
3481 * reflect the fact so that a table reload will
3482 * access proper superblock content in the ctr.
3489 static int raid_preresume(struct dm_target
*ti
)
3492 struct raid_set
*rs
= ti
->private;
3493 struct mddev
*mddev
= &rs
->md
;
3495 /* This is a resume after a suspend of the set -> it's already started */
3496 if (test_and_set_bit(RT_FLAG_RS_PRERESUMED
, &rs
->runtime_flags
))
3500 * The superblocks need to be updated on disk if the
3501 * array is new or new devices got added (thus zeroed
3502 * out by userspace) or __load_dirty_region_bitmap
3503 * will overwrite them in core with old data or fail.
3505 if (test_bit(RT_FLAG_UPDATE_SBS
, &rs
->runtime_flags
))
3509 * Disable/enable discard support on raid set after any
3510 * conversion, because devices can have been added
3512 configure_discard_support(rs
);
3514 /* Load the bitmap from disk unless raid0 */
3515 r
= __load_dirty_region_bitmap(rs
);
3519 /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) */
3520 if (test_bit(RT_FLAG_RS_BITMAP_LOADED
, &rs
->runtime_flags
) &&
3521 mddev
->bitmap_info
.chunksize
!= to_bytes(rs
->requested_bitmap_chunk_sectors
)) {
3522 r
= bitmap_resize(mddev
->bitmap
, mddev
->dev_sectors
,
3523 to_bytes(rs
->requested_bitmap_chunk_sectors
), 0);
3525 DMERR("Failed to resize bitmap");
3528 /* Check for any resize/reshape on @rs and adjust/initiate */
3529 /* Be prepared for mddev_resume() in raid_resume() */
3530 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3531 if (mddev
->recovery_cp
&& mddev
->recovery_cp
< MaxSector
) {
3532 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
3533 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
3534 mddev
->resync_min
= mddev
->recovery_cp
;
3537 rs_set_capacity(rs
);
3539 /* Check for any reshape request unless new raid set */
3540 if (test_and_clear_bit(RT_FLAG_RESHAPE_RS
, &rs
->runtime_flags
)) {
3541 /* Initiate a reshape. */
3542 mddev_lock_nointr(mddev
);
3543 r
= rs_start_reshape(rs
);
3544 mddev_unlock(mddev
);
3546 DMWARN("Failed to check/start reshape, continuing without change");
3553 static void raid_resume(struct dm_target
*ti
)
3555 struct raid_set
*rs
= ti
->private;
3556 struct mddev
*mddev
= &rs
->md
;
3558 if (test_and_set_bit(RT_FLAG_RS_RESUMED
, &rs
->runtime_flags
)) {
3560 * A secondary resume while the device is active.
3561 * Take this opportunity to check whether any failed
3562 * devices are reachable again.
3564 attempt_restore_of_faulty_devices(rs
);
3570 * When passing in flags to the ctr, we expect userspace
3571 * to reset them because they made it to the superblocks
3572 * and reload the mapping anyway.
3574 * -> only unfreeze recovery in case of a table reload or
3575 * we'll have a bogus recovery/reshape position
3576 * retrieved from the superblock by the ctr because
3577 * the ongoing recovery/reshape will change it after read.
3579 if (!test_bit(RT_FLAG_KEEP_RS_FROZEN
, &rs
->runtime_flags
))
3580 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3582 if (mddev
->suspended
)
3583 mddev_resume(mddev
);
3587 static struct target_type raid_target
= {
3589 .version
= {1, 9, 0},
3590 .module
= THIS_MODULE
,
3594 .status
= raid_status
,
3595 .message
= raid_message
,
3596 .iterate_devices
= raid_iterate_devices
,
3597 .io_hints
= raid_io_hints
,
3598 .presuspend
= raid_presuspend
,
3599 .postsuspend
= raid_postsuspend
,
3600 .preresume
= raid_preresume
,
3601 .resume
= raid_resume
,
3604 static int __init
dm_raid_init(void)
3606 DMINFO("Loading target version %u.%u.%u",
3607 raid_target
.version
[0],
3608 raid_target
.version
[1],
3609 raid_target
.version
[2]);
3610 return dm_register_target(&raid_target
);
3613 static void __exit
dm_raid_exit(void)
3615 dm_unregister_target(&raid_target
);
3618 module_init(dm_raid_init
);
3619 module_exit(dm_raid_exit
);
3621 module_param(devices_handle_discard_safely
, bool, 0644);
3622 MODULE_PARM_DESC(devices_handle_discard_safely
,
3623 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
3625 MODULE_DESCRIPTION(DM_NAME
" raid0/1/10/4/5/6 target");
3626 MODULE_ALIAS("dm-raid0");
3627 MODULE_ALIAS("dm-raid1");
3628 MODULE_ALIAS("dm-raid10");
3629 MODULE_ALIAS("dm-raid4");
3630 MODULE_ALIAS("dm-raid5");
3631 MODULE_ALIAS("dm-raid6");
3632 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
3633 MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
3634 MODULE_LICENSE("GPL");