2 raid0.c : Multiple Devices driver for Linux
3 Copyright (C) 1994-96 Marc ZYNGIER
4 <zyngier@ufr-info-p7.ibp.fr> or
6 Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
8 RAID-0 management functions.
10 This program is free software; you can redistribute it and/or modify
11 it under the terms of the GNU General Public License as published by
12 the Free Software Foundation; either version 2, or (at your option)
15 You should have received a copy of the GNU General Public License
16 (for example /usr/src/linux/COPYING); if not, write to the Free
17 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 #include <linux/blkdev.h>
21 #include <linux/seq_file.h>
22 #include <linux/module.h>
23 #include <linux/slab.h>
24 #include <trace/events/block.h>
29 #define UNSUPPORTED_MDDEV_FLAGS \
30 ((1L << MD_HAS_JOURNAL) | \
31 (1L << MD_JOURNAL_CLEAN) | \
32 (1L << MD_FAILFAST_SUPPORTED) |\
35 static int raid0_congested(struct mddev
*mddev
, int bits
)
37 struct r0conf
*conf
= mddev
->private;
38 struct md_rdev
**devlist
= conf
->devlist
;
39 int raid_disks
= conf
->strip_zone
[0].nb_dev
;
42 for (i
= 0; i
< raid_disks
&& !ret
; i
++) {
43 struct request_queue
*q
= bdev_get_queue(devlist
[i
]->bdev
);
45 ret
|= bdi_congested(q
->backing_dev_info
, bits
);
51 * inform the user of the raid configuration
53 static void dump_zones(struct mddev
*mddev
)
56 sector_t zone_size
= 0;
57 sector_t zone_start
= 0;
58 char b
[BDEVNAME_SIZE
];
59 struct r0conf
*conf
= mddev
->private;
60 int raid_disks
= conf
->strip_zone
[0].nb_dev
;
61 pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
63 conf
->nr_strip_zones
, conf
->nr_strip_zones
==1?"":"s");
64 for (j
= 0; j
< conf
->nr_strip_zones
; j
++) {
68 for (k
= 0; k
< conf
->strip_zone
[j
].nb_dev
; k
++)
69 len
+= snprintf(line
+len
, 200-len
, "%s%s", k
?"/":"",
70 bdevname(conf
->devlist
[j
*raid_disks
72 pr_debug("md: zone%d=[%s]\n", j
, line
);
74 zone_size
= conf
->strip_zone
[j
].zone_end
- zone_start
;
75 pr_debug(" zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
76 (unsigned long long)zone_start
>>1,
77 (unsigned long long)conf
->strip_zone
[j
].dev_start
>>1,
78 (unsigned long long)zone_size
>>1);
79 zone_start
= conf
->strip_zone
[j
].zone_end
;
83 static int create_strip_zones(struct mddev
*mddev
, struct r0conf
**private_conf
)
86 sector_t curr_zone_end
, sectors
;
87 struct md_rdev
*smallest
, *rdev1
, *rdev2
, *rdev
, **dev
;
88 struct strip_zone
*zone
;
90 char b
[BDEVNAME_SIZE
];
91 char b2
[BDEVNAME_SIZE
];
92 struct r0conf
*conf
= kzalloc(sizeof(*conf
), GFP_KERNEL
);
93 unsigned short blksize
= 512;
95 *private_conf
= ERR_PTR(-ENOMEM
);
98 rdev_for_each(rdev1
, mddev
) {
99 pr_debug("md/raid0:%s: looking at %s\n",
101 bdevname(rdev1
->bdev
, b
));
104 /* round size to chunk_size */
105 sectors
= rdev1
->sectors
;
106 sector_div(sectors
, mddev
->chunk_sectors
);
107 rdev1
->sectors
= sectors
* mddev
->chunk_sectors
;
109 blksize
= max(blksize
, queue_logical_block_size(
110 rdev1
->bdev
->bd_disk
->queue
));
112 rdev_for_each(rdev2
, mddev
) {
113 pr_debug("md/raid0:%s: comparing %s(%llu)"
116 bdevname(rdev1
->bdev
,b
),
117 (unsigned long long)rdev1
->sectors
,
118 bdevname(rdev2
->bdev
,b2
),
119 (unsigned long long)rdev2
->sectors
);
120 if (rdev2
== rdev1
) {
121 pr_debug("md/raid0:%s: END\n",
125 if (rdev2
->sectors
== rdev1
->sectors
) {
127 * Not unique, don't count it as a new
130 pr_debug("md/raid0:%s: EQUAL\n",
135 pr_debug("md/raid0:%s: NOT EQUAL\n",
139 pr_debug("md/raid0:%s: ==> UNIQUE\n",
141 conf
->nr_strip_zones
++;
142 pr_debug("md/raid0:%s: %d zones\n",
143 mdname(mddev
), conf
->nr_strip_zones
);
146 pr_debug("md/raid0:%s: FINAL %d zones\n",
147 mdname(mddev
), conf
->nr_strip_zones
);
149 * now since we have the hard sector sizes, we can make sure
150 * chunk size is a multiple of that sector size
152 if ((mddev
->chunk_sectors
<< 9) % blksize
) {
153 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
155 mddev
->chunk_sectors
<< 9, blksize
);
161 conf
->strip_zone
= kzalloc(sizeof(struct strip_zone
)*
162 conf
->nr_strip_zones
, GFP_KERNEL
);
163 if (!conf
->strip_zone
)
165 conf
->devlist
= kzalloc(sizeof(struct md_rdev
*)*
166 conf
->nr_strip_zones
*mddev
->raid_disks
,
171 /* The first zone must contain all devices, so here we check that
172 * there is a proper alignment of slots to devices and find them all
174 zone
= &conf
->strip_zone
[0];
179 rdev_for_each(rdev1
, mddev
) {
180 int j
= rdev1
->raid_disk
;
182 if (mddev
->level
== 10) {
183 /* taking over a raid10-n2 array */
185 rdev1
->new_raid_disk
= j
;
188 if (mddev
->level
== 1) {
189 /* taiking over a raid1 array-
190 * we have only one active disk
193 rdev1
->new_raid_disk
= j
;
197 pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
201 if (j
>= mddev
->raid_disks
) {
202 pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
207 pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
213 if (!smallest
|| (rdev1
->sectors
< smallest
->sectors
))
217 if (cnt
!= mddev
->raid_disks
) {
218 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
219 mdname(mddev
), cnt
, mddev
->raid_disks
);
223 zone
->zone_end
= smallest
->sectors
* cnt
;
225 curr_zone_end
= zone
->zone_end
;
227 /* now do the other zones */
228 for (i
= 1; i
< conf
->nr_strip_zones
; i
++)
232 zone
= conf
->strip_zone
+ i
;
233 dev
= conf
->devlist
+ i
* mddev
->raid_disks
;
235 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev
), i
);
236 zone
->dev_start
= smallest
->sectors
;
240 for (j
=0; j
<cnt
; j
++) {
241 rdev
= conf
->devlist
[j
];
242 if (rdev
->sectors
<= zone
->dev_start
) {
243 pr_debug("md/raid0:%s: checking %s ... nope\n",
245 bdevname(rdev
->bdev
, b
));
248 pr_debug("md/raid0:%s: checking %s ..."
249 " contained as device %d\n",
251 bdevname(rdev
->bdev
, b
), c
);
254 if (!smallest
|| rdev
->sectors
< smallest
->sectors
) {
256 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
258 (unsigned long long)rdev
->sectors
);
263 sectors
= (smallest
->sectors
- zone
->dev_start
) * c
;
264 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
266 zone
->nb_dev
, (unsigned long long)sectors
);
268 curr_zone_end
+= sectors
;
269 zone
->zone_end
= curr_zone_end
;
271 pr_debug("md/raid0:%s: current zone start: %llu\n",
273 (unsigned long long)smallest
->sectors
);
276 pr_debug("md/raid0:%s: done.\n", mdname(mddev
));
277 *private_conf
= conf
;
281 kfree(conf
->strip_zone
);
282 kfree(conf
->devlist
);
284 *private_conf
= ERR_PTR(err
);
288 /* Find the zone which holds a particular offset
289 * Update *sectorp to be an offset in that zone
291 static struct strip_zone
*find_zone(struct r0conf
*conf
,
295 struct strip_zone
*z
= conf
->strip_zone
;
296 sector_t sector
= *sectorp
;
298 for (i
= 0; i
< conf
->nr_strip_zones
; i
++)
299 if (sector
< z
[i
].zone_end
) {
301 *sectorp
= sector
- z
[i
-1].zone_end
;
308 * remaps the bio to the target device. we separate two flows.
309 * power 2 flow and a general flow for the sake of performance
311 static struct md_rdev
*map_sector(struct mddev
*mddev
, struct strip_zone
*zone
,
312 sector_t sector
, sector_t
*sector_offset
)
314 unsigned int sect_in_chunk
;
316 struct r0conf
*conf
= mddev
->private;
317 int raid_disks
= conf
->strip_zone
[0].nb_dev
;
318 unsigned int chunk_sects
= mddev
->chunk_sectors
;
320 if (is_power_of_2(chunk_sects
)) {
321 int chunksect_bits
= ffz(~chunk_sects
);
322 /* find the sector offset inside the chunk */
323 sect_in_chunk
= sector
& (chunk_sects
- 1);
324 sector
>>= chunksect_bits
;
326 chunk
= *sector_offset
;
327 /* quotient is the chunk in real device*/
328 sector_div(chunk
, zone
->nb_dev
<< chunksect_bits
);
330 sect_in_chunk
= sector_div(sector
, chunk_sects
);
331 chunk
= *sector_offset
;
332 sector_div(chunk
, chunk_sects
* zone
->nb_dev
);
335 * position the bio over the real device
336 * real sector = chunk in device + starting of zone
337 * + the position in the chunk
339 *sector_offset
= (chunk
* chunk_sects
) + sect_in_chunk
;
340 return conf
->devlist
[(zone
- conf
->strip_zone
)*raid_disks
341 + sector_div(sector
, zone
->nb_dev
)];
344 static sector_t
raid0_size(struct mddev
*mddev
, sector_t sectors
, int raid_disks
)
346 sector_t array_sectors
= 0;
347 struct md_rdev
*rdev
;
349 WARN_ONCE(sectors
|| raid_disks
,
350 "%s does not support generic reshape\n", __func__
);
352 rdev_for_each(rdev
, mddev
)
353 array_sectors
+= (rdev
->sectors
&
354 ~(sector_t
)(mddev
->chunk_sectors
-1));
356 return array_sectors
;
359 static void raid0_free(struct mddev
*mddev
, void *priv
);
361 static int raid0_run(struct mddev
*mddev
)
366 if (mddev
->chunk_sectors
== 0) {
367 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev
));
370 if (md_check_no_bitmap(mddev
))
373 /* if private is not null, we are here after takeover */
374 if (mddev
->private == NULL
) {
375 ret
= create_strip_zones(mddev
, &conf
);
378 mddev
->private = conf
;
380 conf
= mddev
->private;
382 struct md_rdev
*rdev
;
383 bool discard_supported
= false;
385 blk_queue_max_hw_sectors(mddev
->queue
, mddev
->chunk_sectors
);
386 blk_queue_max_write_same_sectors(mddev
->queue
, mddev
->chunk_sectors
);
387 blk_queue_max_write_zeroes_sectors(mddev
->queue
, mddev
->chunk_sectors
);
388 blk_queue_max_discard_sectors(mddev
->queue
, mddev
->chunk_sectors
);
390 blk_queue_io_min(mddev
->queue
, mddev
->chunk_sectors
<< 9);
391 blk_queue_io_opt(mddev
->queue
,
392 (mddev
->chunk_sectors
<< 9) * mddev
->raid_disks
);
394 rdev_for_each(rdev
, mddev
) {
395 disk_stack_limits(mddev
->gendisk
, rdev
->bdev
,
396 rdev
->data_offset
<< 9);
397 if (blk_queue_discard(bdev_get_queue(rdev
->bdev
)))
398 discard_supported
= true;
400 if (!discard_supported
)
401 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, mddev
->queue
);
403 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, mddev
->queue
);
406 /* calculate array device size */
407 md_set_array_sectors(mddev
, raid0_size(mddev
, 0, 0));
409 pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
411 (unsigned long long)mddev
->array_sectors
);
414 /* calculate the max read-ahead size.
415 * For read-ahead of large files to be effective, we need to
416 * readahead at least twice a whole stripe. i.e. number of devices
417 * multiplied by chunk size times 2.
418 * If an individual device has an ra_pages greater than the
419 * chunk size, then we will not drive that device as hard as it
420 * wants. We consider this a configuration error: a larger
421 * chunksize should be used in that case.
423 int stripe
= mddev
->raid_disks
*
424 (mddev
->chunk_sectors
<< 9) / PAGE_SIZE
;
425 if (mddev
->queue
->backing_dev_info
->ra_pages
< 2* stripe
)
426 mddev
->queue
->backing_dev_info
->ra_pages
= 2* stripe
;
431 ret
= md_integrity_register(mddev
);
436 static void raid0_free(struct mddev
*mddev
, void *priv
)
438 struct r0conf
*conf
= priv
;
440 kfree(conf
->strip_zone
);
441 kfree(conf
->devlist
);
446 * Is io distribute over 1 or more chunks ?
448 static inline int is_io_in_chunk_boundary(struct mddev
*mddev
,
449 unsigned int chunk_sects
, struct bio
*bio
)
451 if (likely(is_power_of_2(chunk_sects
))) {
452 return chunk_sects
>=
453 ((bio
->bi_iter
.bi_sector
& (chunk_sects
-1))
456 sector_t sector
= bio
->bi_iter
.bi_sector
;
457 return chunk_sects
>= (sector_div(sector
, chunk_sects
)
462 static void raid0_make_request(struct mddev
*mddev
, struct bio
*bio
)
464 struct strip_zone
*zone
;
465 struct md_rdev
*tmp_dev
;
468 unsigned chunk_sects
;
471 if (unlikely(bio
->bi_opf
& REQ_PREFLUSH
)) {
472 md_flush_request(mddev
, bio
);
476 bio_sector
= bio
->bi_iter
.bi_sector
;
478 chunk_sects
= mddev
->chunk_sectors
;
480 sectors
= chunk_sects
-
481 (likely(is_power_of_2(chunk_sects
))
482 ? (sector
& (chunk_sects
-1))
483 : sector_div(sector
, chunk_sects
));
485 /* Restore due to sector_div */
488 if (sectors
< bio_sectors(bio
)) {
489 struct bio
*split
= bio_split(bio
, sectors
, GFP_NOIO
, mddev
->bio_set
);
490 bio_chain(split
, bio
);
491 generic_make_request(bio
);
495 zone
= find_zone(mddev
->private, §or
);
496 tmp_dev
= map_sector(mddev
, zone
, sector
, §or
);
497 bio
->bi_bdev
= tmp_dev
->bdev
;
498 bio
->bi_iter
.bi_sector
= sector
+ zone
->dev_start
+
499 tmp_dev
->data_offset
;
501 if (unlikely((bio_op(bio
) == REQ_OP_DISCARD
) &&
502 !blk_queue_discard(bdev_get_queue(bio
->bi_bdev
)))) {
507 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
),
508 bio
, disk_devt(mddev
->gendisk
),
510 mddev_check_writesame(mddev
, bio
);
511 mddev_check_write_zeroes(mddev
, bio
);
512 generic_make_request(bio
);
516 static void raid0_status(struct seq_file
*seq
, struct mddev
*mddev
)
518 seq_printf(seq
, " %dk chunks", mddev
->chunk_sectors
/ 2);
522 static void *raid0_takeover_raid45(struct mddev
*mddev
)
524 struct md_rdev
*rdev
;
525 struct r0conf
*priv_conf
;
527 if (mddev
->degraded
!= 1) {
528 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
531 return ERR_PTR(-EINVAL
);
534 rdev_for_each(rdev
, mddev
) {
535 /* check slot number for a disk */
536 if (rdev
->raid_disk
== mddev
->raid_disks
-1) {
537 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
539 return ERR_PTR(-EINVAL
);
541 rdev
->sectors
= mddev
->dev_sectors
;
544 /* Set new parameters */
545 mddev
->new_level
= 0;
546 mddev
->new_layout
= 0;
547 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
549 mddev
->delta_disks
= -1;
550 /* make sure it will be not marked as dirty */
551 mddev
->recovery_cp
= MaxSector
;
552 mddev_clear_unsupported_flags(mddev
, UNSUPPORTED_MDDEV_FLAGS
);
554 create_strip_zones(mddev
, &priv_conf
);
559 static void *raid0_takeover_raid10(struct mddev
*mddev
)
561 struct r0conf
*priv_conf
;
564 * - far_copies must be 1
565 * - near_copies must be 2
566 * - disks number must be even
567 * - all mirrors must be already degraded
569 if (mddev
->layout
!= ((1 << 8) + 2)) {
570 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
573 return ERR_PTR(-EINVAL
);
575 if (mddev
->raid_disks
& 1) {
576 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
578 return ERR_PTR(-EINVAL
);
580 if (mddev
->degraded
!= (mddev
->raid_disks
>>1)) {
581 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
583 return ERR_PTR(-EINVAL
);
586 /* Set new parameters */
587 mddev
->new_level
= 0;
588 mddev
->new_layout
= 0;
589 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
590 mddev
->delta_disks
= - mddev
->raid_disks
/ 2;
591 mddev
->raid_disks
+= mddev
->delta_disks
;
593 /* make sure it will be not marked as dirty */
594 mddev
->recovery_cp
= MaxSector
;
595 mddev_clear_unsupported_flags(mddev
, UNSUPPORTED_MDDEV_FLAGS
);
597 create_strip_zones(mddev
, &priv_conf
);
601 static void *raid0_takeover_raid1(struct mddev
*mddev
)
603 struct r0conf
*priv_conf
;
607 * - (N - 1) mirror drives must be already faulty
609 if ((mddev
->raid_disks
- 1) != mddev
->degraded
) {
610 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
612 return ERR_PTR(-EINVAL
);
616 * a raid1 doesn't have the notion of chunk size, so
617 * figure out the largest suitable size we can use.
619 chunksect
= 64 * 2; /* 64K by default */
621 /* The array must be an exact multiple of chunksize */
622 while (chunksect
&& (mddev
->array_sectors
& (chunksect
- 1)))
625 if ((chunksect
<< 9) < PAGE_SIZE
)
626 /* array size does not allow a suitable chunk size */
627 return ERR_PTR(-EINVAL
);
629 /* Set new parameters */
630 mddev
->new_level
= 0;
631 mddev
->new_layout
= 0;
632 mddev
->new_chunk_sectors
= chunksect
;
633 mddev
->chunk_sectors
= chunksect
;
634 mddev
->delta_disks
= 1 - mddev
->raid_disks
;
635 mddev
->raid_disks
= 1;
636 /* make sure it will be not marked as dirty */
637 mddev
->recovery_cp
= MaxSector
;
638 mddev_clear_unsupported_flags(mddev
, UNSUPPORTED_MDDEV_FLAGS
);
640 create_strip_zones(mddev
, &priv_conf
);
644 static void *raid0_takeover(struct mddev
*mddev
)
646 /* raid0 can take over:
647 * raid4 - if all data disks are active.
648 * raid5 - providing it is Raid4 layout and one disk is faulty
649 * raid10 - assuming we have all necessary active disks
650 * raid1 - with (N -1) mirror drives faulty
654 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
656 return ERR_PTR(-EBUSY
);
658 if (mddev
->level
== 4)
659 return raid0_takeover_raid45(mddev
);
661 if (mddev
->level
== 5) {
662 if (mddev
->layout
== ALGORITHM_PARITY_N
)
663 return raid0_takeover_raid45(mddev
);
665 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
666 mdname(mddev
), ALGORITHM_PARITY_N
);
669 if (mddev
->level
== 10)
670 return raid0_takeover_raid10(mddev
);
672 if (mddev
->level
== 1)
673 return raid0_takeover_raid1(mddev
);
675 pr_warn("Takeover from raid%i to raid0 not supported\n",
678 return ERR_PTR(-EINVAL
);
681 static void raid0_quiesce(struct mddev
*mddev
, int state
)
685 static struct md_personality raid0_personality
=
689 .owner
= THIS_MODULE
,
690 .make_request
= raid0_make_request
,
693 .status
= raid0_status
,
695 .takeover
= raid0_takeover
,
696 .quiesce
= raid0_quiesce
,
697 .congested
= raid0_congested
,
700 static int __init
raid0_init (void)
702 return register_md_personality (&raid0_personality
);
705 static void raid0_exit (void)
707 unregister_md_personality (&raid0_personality
);
710 module_init(raid0_init
);
711 module_exit(raid0_exit
);
712 MODULE_LICENSE("GPL");
713 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
714 MODULE_ALIAS("md-personality-2"); /* RAID0 */
715 MODULE_ALIAS("md-raid0");
716 MODULE_ALIAS("md-level-0");