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
9 RAID-0 management functions.
11 This program is free software; you can redistribute it and/or modify
12 it under the terms of the GNU General Public License as published by
13 the Free Software Foundation; either version 2, or (at your option)
16 You should have received a copy of the GNU General Public License
17 (for example /usr/src/linux/COPYING); if not, write to the Free
18 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 #include <linux/blkdev.h>
22 #include <linux/seq_file.h>
23 #include <linux/module.h>
24 #include <linux/slab.h>
29 static int raid0_congested(void *data
, int bits
)
31 struct mddev
*mddev
= data
;
32 struct r0conf
*conf
= mddev
->private;
33 struct md_rdev
**devlist
= conf
->devlist
;
34 int raid_disks
= conf
->strip_zone
[0].nb_dev
;
37 if (mddev_congested(mddev
, bits
))
40 for (i
= 0; i
< raid_disks
&& !ret
; i
++) {
41 struct request_queue
*q
= bdev_get_queue(devlist
[i
]->bdev
);
43 ret
|= bdi_congested(&q
->backing_dev_info
, bits
);
49 * inform the user of the raid configuration
51 static void dump_zones(struct mddev
*mddev
)
54 sector_t zone_size
= 0;
55 sector_t zone_start
= 0;
56 char b
[BDEVNAME_SIZE
];
57 struct r0conf
*conf
= mddev
->private;
58 int raid_disks
= conf
->strip_zone
[0].nb_dev
;
59 printk(KERN_INFO
"md: RAID0 configuration for %s - %d zone%s\n",
61 conf
->nr_strip_zones
, conf
->nr_strip_zones
==1?"":"s");
62 for (j
= 0; j
< conf
->nr_strip_zones
; j
++) {
63 printk(KERN_INFO
"md: zone%d=[", j
);
64 for (k
= 0; k
< conf
->strip_zone
[j
].nb_dev
; k
++)
65 printk(KERN_CONT
"%s%s", k
?"/":"",
66 bdevname(conf
->devlist
[j
*raid_disks
68 printk(KERN_CONT
"]\n");
70 zone_size
= conf
->strip_zone
[j
].zone_end
- zone_start
;
71 printk(KERN_INFO
" zone-offset=%10lluKB, "
72 "device-offset=%10lluKB, size=%10lluKB\n",
73 (unsigned long long)zone_start
>>1,
74 (unsigned long long)conf
->strip_zone
[j
].dev_start
>>1,
75 (unsigned long long)zone_size
>>1);
76 zone_start
= conf
->strip_zone
[j
].zone_end
;
78 printk(KERN_INFO
"\n");
81 static int create_strip_zones(struct mddev
*mddev
, struct r0conf
**private_conf
)
84 sector_t curr_zone_end
, sectors
;
85 struct md_rdev
*smallest
, *rdev1
, *rdev2
, *rdev
, **dev
;
86 struct strip_zone
*zone
;
88 char b
[BDEVNAME_SIZE
];
89 char b2
[BDEVNAME_SIZE
];
90 struct r0conf
*conf
= kzalloc(sizeof(*conf
), GFP_KERNEL
);
91 bool discard_supported
= false;
95 rdev_for_each(rdev1
, mddev
) {
96 pr_debug("md/raid0:%s: looking at %s\n",
98 bdevname(rdev1
->bdev
, b
));
101 /* round size to chunk_size */
102 sectors
= rdev1
->sectors
;
103 sector_div(sectors
, mddev
->chunk_sectors
);
104 rdev1
->sectors
= sectors
* mddev
->chunk_sectors
;
106 rdev_for_each(rdev2
, mddev
) {
107 pr_debug("md/raid0:%s: comparing %s(%llu)"
110 bdevname(rdev1
->bdev
,b
),
111 (unsigned long long)rdev1
->sectors
,
112 bdevname(rdev2
->bdev
,b2
),
113 (unsigned long long)rdev2
->sectors
);
114 if (rdev2
== rdev1
) {
115 pr_debug("md/raid0:%s: END\n",
119 if (rdev2
->sectors
== rdev1
->sectors
) {
121 * Not unique, don't count it as a new
124 pr_debug("md/raid0:%s: EQUAL\n",
129 pr_debug("md/raid0:%s: NOT EQUAL\n",
133 pr_debug("md/raid0:%s: ==> UNIQUE\n",
135 conf
->nr_strip_zones
++;
136 pr_debug("md/raid0:%s: %d zones\n",
137 mdname(mddev
), conf
->nr_strip_zones
);
140 pr_debug("md/raid0:%s: FINAL %d zones\n",
141 mdname(mddev
), conf
->nr_strip_zones
);
143 conf
->strip_zone
= kzalloc(sizeof(struct strip_zone
)*
144 conf
->nr_strip_zones
, GFP_KERNEL
);
145 if (!conf
->strip_zone
)
147 conf
->devlist
= kzalloc(sizeof(struct md_rdev
*)*
148 conf
->nr_strip_zones
*mddev
->raid_disks
,
153 /* The first zone must contain all devices, so here we check that
154 * there is a proper alignment of slots to devices and find them all
156 zone
= &conf
->strip_zone
[0];
161 rdev_for_each(rdev1
, mddev
) {
162 int j
= rdev1
->raid_disk
;
164 if (mddev
->level
== 10) {
165 /* taking over a raid10-n2 array */
167 rdev1
->new_raid_disk
= j
;
170 if (mddev
->level
== 1) {
171 /* taiking over a raid1 array-
172 * we have only one active disk
175 rdev1
->new_raid_disk
= j
;
180 "md/raid0:%s: remove inactive devices before converting to RAID0\n",
184 if (j
>= mddev
->raid_disks
) {
185 printk(KERN_ERR
"md/raid0:%s: bad disk number %d - "
186 "aborting!\n", mdname(mddev
), j
);
190 printk(KERN_ERR
"md/raid0:%s: multiple devices for %d - "
191 "aborting!\n", mdname(mddev
), j
);
196 disk_stack_limits(mddev
->gendisk
, rdev1
->bdev
,
197 rdev1
->data_offset
<< 9);
199 if (rdev1
->bdev
->bd_disk
->queue
->merge_bvec_fn
)
200 conf
->has_merge_bvec
= 1;
202 if (!smallest
|| (rdev1
->sectors
< smallest
->sectors
))
206 if (blk_queue_discard(bdev_get_queue(rdev1
->bdev
)))
207 discard_supported
= true;
209 if (cnt
!= mddev
->raid_disks
) {
210 printk(KERN_ERR
"md/raid0:%s: too few disks (%d of %d) - "
211 "aborting!\n", mdname(mddev
), cnt
, mddev
->raid_disks
);
215 zone
->zone_end
= smallest
->sectors
* cnt
;
217 curr_zone_end
= zone
->zone_end
;
219 /* now do the other zones */
220 for (i
= 1; i
< conf
->nr_strip_zones
; i
++)
224 zone
= conf
->strip_zone
+ i
;
225 dev
= conf
->devlist
+ i
* mddev
->raid_disks
;
227 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev
), i
);
228 zone
->dev_start
= smallest
->sectors
;
232 for (j
=0; j
<cnt
; j
++) {
233 rdev
= conf
->devlist
[j
];
234 if (rdev
->sectors
<= zone
->dev_start
) {
235 pr_debug("md/raid0:%s: checking %s ... nope\n",
237 bdevname(rdev
->bdev
, b
));
240 pr_debug("md/raid0:%s: checking %s ..."
241 " contained as device %d\n",
243 bdevname(rdev
->bdev
, b
), c
);
246 if (!smallest
|| rdev
->sectors
< smallest
->sectors
) {
248 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
250 (unsigned long long)rdev
->sectors
);
255 sectors
= (smallest
->sectors
- zone
->dev_start
) * c
;
256 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
258 zone
->nb_dev
, (unsigned long long)sectors
);
260 curr_zone_end
+= sectors
;
261 zone
->zone_end
= curr_zone_end
;
263 pr_debug("md/raid0:%s: current zone start: %llu\n",
265 (unsigned long long)smallest
->sectors
);
267 mddev
->queue
->backing_dev_info
.congested_fn
= raid0_congested
;
268 mddev
->queue
->backing_dev_info
.congested_data
= mddev
;
271 * now since we have the hard sector sizes, we can make sure
272 * chunk size is a multiple of that sector size
274 if ((mddev
->chunk_sectors
<< 9) % queue_logical_block_size(mddev
->queue
)) {
275 printk(KERN_ERR
"md/raid0:%s: chunk_size of %d not valid\n",
277 mddev
->chunk_sectors
<< 9);
281 blk_queue_io_min(mddev
->queue
, mddev
->chunk_sectors
<< 9);
282 blk_queue_io_opt(mddev
->queue
,
283 (mddev
->chunk_sectors
<< 9) * mddev
->raid_disks
);
285 if (!discard_supported
)
286 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, mddev
->queue
);
288 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, mddev
->queue
);
290 pr_debug("md/raid0:%s: done.\n", mdname(mddev
));
291 *private_conf
= conf
;
295 kfree(conf
->strip_zone
);
296 kfree(conf
->devlist
);
298 *private_conf
= ERR_PTR(err
);
302 /* Find the zone which holds a particular offset
303 * Update *sectorp to be an offset in that zone
305 static struct strip_zone
*find_zone(struct r0conf
*conf
,
309 struct strip_zone
*z
= conf
->strip_zone
;
310 sector_t sector
= *sectorp
;
312 for (i
= 0; i
< conf
->nr_strip_zones
; i
++)
313 if (sector
< z
[i
].zone_end
) {
315 *sectorp
= sector
- z
[i
-1].zone_end
;
322 * remaps the bio to the target device. we separate two flows.
323 * power 2 flow and a general flow for the sake of perfromance
325 static struct md_rdev
*map_sector(struct mddev
*mddev
, struct strip_zone
*zone
,
326 sector_t sector
, sector_t
*sector_offset
)
328 unsigned int sect_in_chunk
;
330 struct r0conf
*conf
= mddev
->private;
331 int raid_disks
= conf
->strip_zone
[0].nb_dev
;
332 unsigned int chunk_sects
= mddev
->chunk_sectors
;
334 if (is_power_of_2(chunk_sects
)) {
335 int chunksect_bits
= ffz(~chunk_sects
);
336 /* find the sector offset inside the chunk */
337 sect_in_chunk
= sector
& (chunk_sects
- 1);
338 sector
>>= chunksect_bits
;
340 chunk
= *sector_offset
;
341 /* quotient is the chunk in real device*/
342 sector_div(chunk
, zone
->nb_dev
<< chunksect_bits
);
344 sect_in_chunk
= sector_div(sector
, chunk_sects
);
345 chunk
= *sector_offset
;
346 sector_div(chunk
, chunk_sects
* zone
->nb_dev
);
349 * position the bio over the real device
350 * real sector = chunk in device + starting of zone
351 * + the position in the chunk
353 *sector_offset
= (chunk
* chunk_sects
) + sect_in_chunk
;
354 return conf
->devlist
[(zone
- conf
->strip_zone
)*raid_disks
355 + sector_div(sector
, zone
->nb_dev
)];
359 * raid0_mergeable_bvec -- tell bio layer if two requests can be merged
361 * @bvm: properties of new bio
362 * @biovec: the request that could be merged to it.
364 * Return amount of bytes we can accept at this offset
366 static int raid0_mergeable_bvec(struct request_queue
*q
,
367 struct bvec_merge_data
*bvm
,
368 struct bio_vec
*biovec
)
370 struct mddev
*mddev
= q
->queuedata
;
371 struct r0conf
*conf
= mddev
->private;
372 sector_t sector
= bvm
->bi_sector
+ get_start_sect(bvm
->bi_bdev
);
373 sector_t sector_offset
= sector
;
375 unsigned int chunk_sectors
= mddev
->chunk_sectors
;
376 unsigned int bio_sectors
= bvm
->bi_size
>> 9;
377 struct strip_zone
*zone
;
378 struct md_rdev
*rdev
;
379 struct request_queue
*subq
;
381 if (is_power_of_2(chunk_sectors
))
382 max
= (chunk_sectors
- ((sector
& (chunk_sectors
-1))
383 + bio_sectors
)) << 9;
385 max
= (chunk_sectors
- (sector_div(sector
, chunk_sectors
)
386 + bio_sectors
)) << 9;
388 max
= 0; /* bio_add cannot handle a negative return */
389 if (max
<= biovec
->bv_len
&& bio_sectors
== 0)
390 return biovec
->bv_len
;
391 if (max
< biovec
->bv_len
)
392 /* too small already, no need to check further */
394 if (!conf
->has_merge_bvec
)
397 /* May need to check subordinate device */
398 sector
= sector_offset
;
399 zone
= find_zone(mddev
->private, §or_offset
);
400 rdev
= map_sector(mddev
, zone
, sector
, §or_offset
);
401 subq
= bdev_get_queue(rdev
->bdev
);
402 if (subq
->merge_bvec_fn
) {
403 bvm
->bi_bdev
= rdev
->bdev
;
404 bvm
->bi_sector
= sector_offset
+ zone
->dev_start
+
406 return min(max
, subq
->merge_bvec_fn(subq
, bvm
, biovec
));
411 static sector_t
raid0_size(struct mddev
*mddev
, sector_t sectors
, int raid_disks
)
413 sector_t array_sectors
= 0;
414 struct md_rdev
*rdev
;
416 WARN_ONCE(sectors
|| raid_disks
,
417 "%s does not support generic reshape\n", __func__
);
419 rdev_for_each(rdev
, mddev
)
420 array_sectors
+= (rdev
->sectors
&
421 ~(sector_t
)(mddev
->chunk_sectors
-1));
423 return array_sectors
;
426 static int raid0_stop(struct mddev
*mddev
);
428 static int raid0_run(struct mddev
*mddev
)
433 if (mddev
->chunk_sectors
== 0) {
434 printk(KERN_ERR
"md/raid0:%s: chunk size must be set.\n",
438 if (md_check_no_bitmap(mddev
))
440 blk_queue_max_hw_sectors(mddev
->queue
, mddev
->chunk_sectors
);
441 blk_queue_max_write_same_sectors(mddev
->queue
, mddev
->chunk_sectors
);
442 blk_queue_max_discard_sectors(mddev
->queue
, mddev
->chunk_sectors
);
444 /* if private is not null, we are here after takeover */
445 if (mddev
->private == NULL
) {
446 ret
= create_strip_zones(mddev
, &conf
);
449 mddev
->private = conf
;
451 conf
= mddev
->private;
453 /* calculate array device size */
454 md_set_array_sectors(mddev
, raid0_size(mddev
, 0, 0));
456 printk(KERN_INFO
"md/raid0:%s: md_size is %llu sectors.\n",
458 (unsigned long long)mddev
->array_sectors
);
459 /* calculate the max read-ahead size.
460 * For read-ahead of large files to be effective, we need to
461 * readahead at least twice a whole stripe. i.e. number of devices
462 * multiplied by chunk size times 2.
463 * If an individual device has an ra_pages greater than the
464 * chunk size, then we will not drive that device as hard as it
465 * wants. We consider this a configuration error: a larger
466 * chunksize should be used in that case.
469 int stripe
= mddev
->raid_disks
*
470 (mddev
->chunk_sectors
<< 9) / PAGE_SIZE
;
471 if (mddev
->queue
->backing_dev_info
.ra_pages
< 2* stripe
)
472 mddev
->queue
->backing_dev_info
.ra_pages
= 2* stripe
;
475 blk_queue_merge_bvec(mddev
->queue
, raid0_mergeable_bvec
);
478 ret
= md_integrity_register(mddev
);
485 static int raid0_stop(struct mddev
*mddev
)
487 struct r0conf
*conf
= mddev
->private;
489 blk_sync_queue(mddev
->queue
); /* the unplug fn references 'conf'*/
490 kfree(conf
->strip_zone
);
491 kfree(conf
->devlist
);
493 mddev
->private = NULL
;
498 * Is io distribute over 1 or more chunks ?
500 static inline int is_io_in_chunk_boundary(struct mddev
*mddev
,
501 unsigned int chunk_sects
, struct bio
*bio
)
503 if (likely(is_power_of_2(chunk_sects
))) {
504 return chunk_sects
>=
505 ((bio
->bi_iter
.bi_sector
& (chunk_sects
-1))
508 sector_t sector
= bio
->bi_iter
.bi_sector
;
509 return chunk_sects
>= (sector_div(sector
, chunk_sects
)
514 static void raid0_make_request(struct mddev
*mddev
, struct bio
*bio
)
516 struct strip_zone
*zone
;
517 struct md_rdev
*tmp_dev
;
520 if (unlikely(bio
->bi_rw
& REQ_FLUSH
)) {
521 md_flush_request(mddev
, bio
);
526 sector_t sector
= bio
->bi_iter
.bi_sector
;
527 unsigned chunk_sects
= mddev
->chunk_sectors
;
529 unsigned sectors
= chunk_sects
-
530 (likely(is_power_of_2(chunk_sects
))
531 ? (sector
& (chunk_sects
-1))
532 : sector_div(sector
, chunk_sects
));
534 if (sectors
< bio_sectors(bio
)) {
535 split
= bio_split(bio
, sectors
, GFP_NOIO
, fs_bio_set
);
536 bio_chain(split
, bio
);
541 zone
= find_zone(mddev
->private, §or
);
542 tmp_dev
= map_sector(mddev
, zone
, sector
, §or
);
543 split
->bi_bdev
= tmp_dev
->bdev
;
544 split
->bi_iter
.bi_sector
= sector
+ zone
->dev_start
+
545 tmp_dev
->data_offset
;
547 if (unlikely((split
->bi_rw
& REQ_DISCARD
) &&
548 !blk_queue_discard(bdev_get_queue(split
->bi_bdev
)))) {
552 generic_make_request(split
);
553 } while (split
!= bio
);
556 static void raid0_status(struct seq_file
*seq
, struct mddev
*mddev
)
558 seq_printf(seq
, " %dk chunks", mddev
->chunk_sectors
/ 2);
562 static void *raid0_takeover_raid45(struct mddev
*mddev
)
564 struct md_rdev
*rdev
;
565 struct r0conf
*priv_conf
;
567 if (mddev
->degraded
!= 1) {
568 printk(KERN_ERR
"md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
571 return ERR_PTR(-EINVAL
);
574 rdev_for_each(rdev
, mddev
) {
575 /* check slot number for a disk */
576 if (rdev
->raid_disk
== mddev
->raid_disks
-1) {
577 printk(KERN_ERR
"md/raid0:%s: raid5 must have missing parity disk!\n",
579 return ERR_PTR(-EINVAL
);
581 rdev
->sectors
= mddev
->dev_sectors
;
584 /* Set new parameters */
585 mddev
->new_level
= 0;
586 mddev
->new_layout
= 0;
587 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
589 mddev
->delta_disks
= -1;
590 /* make sure it will be not marked as dirty */
591 mddev
->recovery_cp
= MaxSector
;
593 create_strip_zones(mddev
, &priv_conf
);
597 static void *raid0_takeover_raid10(struct mddev
*mddev
)
599 struct r0conf
*priv_conf
;
602 * - far_copies must be 1
603 * - near_copies must be 2
604 * - disks number must be even
605 * - all mirrors must be already degraded
607 if (mddev
->layout
!= ((1 << 8) + 2)) {
608 printk(KERN_ERR
"md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
611 return ERR_PTR(-EINVAL
);
613 if (mddev
->raid_disks
& 1) {
614 printk(KERN_ERR
"md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
616 return ERR_PTR(-EINVAL
);
618 if (mddev
->degraded
!= (mddev
->raid_disks
>>1)) {
619 printk(KERN_ERR
"md/raid0:%s: All mirrors must be already degraded!\n",
621 return ERR_PTR(-EINVAL
);
624 /* Set new parameters */
625 mddev
->new_level
= 0;
626 mddev
->new_layout
= 0;
627 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
628 mddev
->delta_disks
= - mddev
->raid_disks
/ 2;
629 mddev
->raid_disks
+= mddev
->delta_disks
;
631 /* make sure it will be not marked as dirty */
632 mddev
->recovery_cp
= MaxSector
;
634 create_strip_zones(mddev
, &priv_conf
);
638 static void *raid0_takeover_raid1(struct mddev
*mddev
)
640 struct r0conf
*priv_conf
;
644 * - (N - 1) mirror drives must be already faulty
646 if ((mddev
->raid_disks
- 1) != mddev
->degraded
) {
647 printk(KERN_ERR
"md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
649 return ERR_PTR(-EINVAL
);
653 * a raid1 doesn't have the notion of chunk size, so
654 * figure out the largest suitable size we can use.
656 chunksect
= 64 * 2; /* 64K by default */
658 /* The array must be an exact multiple of chunksize */
659 while (chunksect
&& (mddev
->array_sectors
& (chunksect
- 1)))
662 if ((chunksect
<< 9) < PAGE_SIZE
)
663 /* array size does not allow a suitable chunk size */
664 return ERR_PTR(-EINVAL
);
666 /* Set new parameters */
667 mddev
->new_level
= 0;
668 mddev
->new_layout
= 0;
669 mddev
->new_chunk_sectors
= chunksect
;
670 mddev
->chunk_sectors
= chunksect
;
671 mddev
->delta_disks
= 1 - mddev
->raid_disks
;
672 mddev
->raid_disks
= 1;
673 /* make sure it will be not marked as dirty */
674 mddev
->recovery_cp
= MaxSector
;
676 create_strip_zones(mddev
, &priv_conf
);
680 static void *raid0_takeover(struct mddev
*mddev
)
682 /* raid0 can take over:
683 * raid4 - if all data disks are active.
684 * raid5 - providing it is Raid4 layout and one disk is faulty
685 * raid10 - assuming we have all necessary active disks
686 * raid1 - with (N -1) mirror drives faulty
688 if (mddev
->level
== 4)
689 return raid0_takeover_raid45(mddev
);
691 if (mddev
->level
== 5) {
692 if (mddev
->layout
== ALGORITHM_PARITY_N
)
693 return raid0_takeover_raid45(mddev
);
695 printk(KERN_ERR
"md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
696 mdname(mddev
), ALGORITHM_PARITY_N
);
699 if (mddev
->level
== 10)
700 return raid0_takeover_raid10(mddev
);
702 if (mddev
->level
== 1)
703 return raid0_takeover_raid1(mddev
);
705 printk(KERN_ERR
"Takeover from raid%i to raid0 not supported\n",
708 return ERR_PTR(-EINVAL
);
711 static void raid0_quiesce(struct mddev
*mddev
, int state
)
715 static struct md_personality raid0_personality
=
719 .owner
= THIS_MODULE
,
720 .make_request
= raid0_make_request
,
723 .status
= raid0_status
,
725 .takeover
= raid0_takeover
,
726 .quiesce
= raid0_quiesce
,
729 static int __init
raid0_init (void)
731 return register_md_personality (&raid0_personality
);
734 static void raid0_exit (void)
736 unregister_md_personality (&raid0_personality
);
739 module_init(raid0_init
);
740 module_exit(raid0_exit
);
741 MODULE_LICENSE("GPL");
742 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
743 MODULE_ALIAS("md-personality-2"); /* RAID0 */
744 MODULE_ALIAS("md-raid0");
745 MODULE_ALIAS("md-level-0");