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/slab.h>
28 static int raid0_congested(void *data
, int bits
)
30 mddev_t
*mddev
= data
;
31 raid0_conf_t
*conf
= mddev
->private;
32 mdk_rdev_t
**devlist
= conf
->devlist
;
33 int raid_disks
= conf
->strip_zone
[0].nb_dev
;
36 if (mddev_congested(mddev
, bits
))
39 for (i
= 0; i
< raid_disks
&& !ret
; i
++) {
40 struct request_queue
*q
= bdev_get_queue(devlist
[i
]->bdev
);
42 ret
|= bdi_congested(&q
->backing_dev_info
, bits
);
48 * inform the user of the raid configuration
50 static void dump_zones(mddev_t
*mddev
)
53 sector_t zone_size
= 0;
54 sector_t zone_start
= 0;
55 char b
[BDEVNAME_SIZE
];
56 raid0_conf_t
*conf
= mddev
->private;
57 int raid_disks
= conf
->strip_zone
[0].nb_dev
;
58 printk(KERN_INFO
"******* %s configuration *********\n",
61 for (j
= 0; j
< conf
->nr_strip_zones
; j
++) {
62 printk(KERN_INFO
"zone%d=[", j
);
63 for (k
= 0; k
< conf
->strip_zone
[j
].nb_dev
; k
++)
64 printk(KERN_CONT
"%s/",
65 bdevname(conf
->devlist
[j
*raid_disks
67 printk(KERN_CONT
"]\n");
69 zone_size
= conf
->strip_zone
[j
].zone_end
- zone_start
;
70 printk(KERN_INFO
" zone offset=%llukb "
71 "device offset=%llukb size=%llukb\n",
72 (unsigned long long)zone_start
>>1,
73 (unsigned long long)conf
->strip_zone
[j
].dev_start
>>1,
74 (unsigned long long)zone_size
>>1);
75 zone_start
= conf
->strip_zone
[j
].zone_end
;
77 printk(KERN_INFO
"**********************************\n\n");
80 static int create_strip_zones(mddev_t
*mddev
, raid0_conf_t
**private_conf
)
83 sector_t curr_zone_end
, sectors
;
84 mdk_rdev_t
*smallest
, *rdev1
, *rdev2
, *rdev
, **dev
;
85 struct strip_zone
*zone
;
87 char b
[BDEVNAME_SIZE
];
88 raid0_conf_t
*conf
= kzalloc(sizeof(*conf
), GFP_KERNEL
);
92 list_for_each_entry(rdev1
, &mddev
->disks
, same_set
) {
93 printk(KERN_INFO
"md/raid0:%s: looking at %s\n",
95 bdevname(rdev1
->bdev
, b
));
98 /* round size to chunk_size */
99 sectors
= rdev1
->sectors
;
100 sector_div(sectors
, mddev
->chunk_sectors
);
101 rdev1
->sectors
= sectors
* mddev
->chunk_sectors
;
103 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
104 printk(KERN_INFO
"md/raid0:%s: comparing %s(%llu)",
106 bdevname(rdev1
->bdev
,b
),
107 (unsigned long long)rdev1
->sectors
);
108 printk(KERN_CONT
" with %s(%llu)\n",
109 bdevname(rdev2
->bdev
,b
),
110 (unsigned long long)rdev2
->sectors
);
111 if (rdev2
== rdev1
) {
112 printk(KERN_INFO
"md/raid0:%s: END\n",
116 if (rdev2
->sectors
== rdev1
->sectors
) {
118 * Not unique, don't count it as a new
121 printk(KERN_INFO
"md/raid0:%s: EQUAL\n",
126 printk(KERN_INFO
"md/raid0:%s: NOT EQUAL\n",
130 printk(KERN_INFO
"md/raid0:%s: ==> UNIQUE\n",
132 conf
->nr_strip_zones
++;
133 printk(KERN_INFO
"md/raid0:%s: %d zones\n",
134 mdname(mddev
), conf
->nr_strip_zones
);
137 printk(KERN_INFO
"md/raid0:%s: FINAL %d zones\n",
138 mdname(mddev
), conf
->nr_strip_zones
);
140 conf
->strip_zone
= kzalloc(sizeof(struct strip_zone
)*
141 conf
->nr_strip_zones
, GFP_KERNEL
);
142 if (!conf
->strip_zone
)
144 conf
->devlist
= kzalloc(sizeof(mdk_rdev_t
*)*
145 conf
->nr_strip_zones
*mddev
->raid_disks
,
150 /* The first zone must contain all devices, so here we check that
151 * there is a proper alignment of slots to devices and find them all
153 zone
= &conf
->strip_zone
[0];
158 list_for_each_entry(rdev1
, &mddev
->disks
, same_set
) {
159 int j
= rdev1
->raid_disk
;
161 if (mddev
->level
== 10) {
162 /* taking over a raid10-n2 array */
164 rdev1
->new_raid_disk
= j
;
167 if (mddev
->level
== 1) {
168 /* taiking over a raid1 array-
169 * we have only one active disk
172 rdev1
->new_raid_disk
= j
;
175 if (j
< 0 || j
>= mddev
->raid_disks
) {
176 printk(KERN_ERR
"md/raid0:%s: bad disk number %d - "
177 "aborting!\n", mdname(mddev
), j
);
181 printk(KERN_ERR
"md/raid0:%s: multiple devices for %d - "
182 "aborting!\n", mdname(mddev
), j
);
187 disk_stack_limits(mddev
->gendisk
, rdev1
->bdev
,
188 rdev1
->data_offset
<< 9);
189 /* as we don't honour merge_bvec_fn, we must never risk
190 * violating it, so limit ->max_segments to 1, lying within
194 if (rdev1
->bdev
->bd_disk
->queue
->merge_bvec_fn
) {
195 blk_queue_max_segments(mddev
->queue
, 1);
196 blk_queue_segment_boundary(mddev
->queue
,
197 PAGE_CACHE_SIZE
- 1);
199 if (!smallest
|| (rdev1
->sectors
< smallest
->sectors
))
203 if (cnt
!= mddev
->raid_disks
) {
204 printk(KERN_ERR
"md/raid0:%s: too few disks (%d of %d) - "
205 "aborting!\n", mdname(mddev
), cnt
, mddev
->raid_disks
);
209 zone
->zone_end
= smallest
->sectors
* cnt
;
211 curr_zone_end
= zone
->zone_end
;
213 /* now do the other zones */
214 for (i
= 1; i
< conf
->nr_strip_zones
; i
++)
218 zone
= conf
->strip_zone
+ i
;
219 dev
= conf
->devlist
+ i
* mddev
->raid_disks
;
221 printk(KERN_INFO
"md/raid0:%s: zone %d\n",
223 zone
->dev_start
= smallest
->sectors
;
227 for (j
=0; j
<cnt
; j
++) {
228 rdev
= conf
->devlist
[j
];
229 printk(KERN_INFO
"md/raid0:%s: checking %s ...",
231 bdevname(rdev
->bdev
, b
));
232 if (rdev
->sectors
<= zone
->dev_start
) {
233 printk(KERN_CONT
" nope.\n");
236 printk(KERN_CONT
" contained as device %d\n", c
);
239 if (!smallest
|| rdev
->sectors
< smallest
->sectors
) {
241 printk(KERN_INFO
"md/raid0:%s: (%llu) is smallest!.\n",
243 (unsigned long long)rdev
->sectors
);
248 sectors
= (smallest
->sectors
- zone
->dev_start
) * c
;
249 printk(KERN_INFO
"md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
251 zone
->nb_dev
, (unsigned long long)sectors
);
253 curr_zone_end
+= sectors
;
254 zone
->zone_end
= curr_zone_end
;
256 printk(KERN_INFO
"md/raid0:%s: current zone start: %llu\n",
258 (unsigned long long)smallest
->sectors
);
260 mddev
->queue
->backing_dev_info
.congested_fn
= raid0_congested
;
261 mddev
->queue
->backing_dev_info
.congested_data
= mddev
;
264 * now since we have the hard sector sizes, we can make sure
265 * chunk size is a multiple of that sector size
267 if ((mddev
->chunk_sectors
<< 9) % queue_logical_block_size(mddev
->queue
)) {
268 printk(KERN_ERR
"md/raid0:%s: chunk_size of %d not valid\n",
270 mddev
->chunk_sectors
<< 9);
274 blk_queue_io_min(mddev
->queue
, mddev
->chunk_sectors
<< 9);
275 blk_queue_io_opt(mddev
->queue
,
276 (mddev
->chunk_sectors
<< 9) * mddev
->raid_disks
);
278 printk(KERN_INFO
"md/raid0:%s: done.\n", mdname(mddev
));
279 *private_conf
= conf
;
283 kfree(conf
->strip_zone
);
284 kfree(conf
->devlist
);
286 *private_conf
= NULL
;
291 * raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
293 * @bvm: properties of new bio
294 * @biovec: the request that could be merged to it.
296 * Return amount of bytes we can accept at this offset
298 static int raid0_mergeable_bvec(struct request_queue
*q
,
299 struct bvec_merge_data
*bvm
,
300 struct bio_vec
*biovec
)
302 mddev_t
*mddev
= q
->queuedata
;
303 sector_t sector
= bvm
->bi_sector
+ get_start_sect(bvm
->bi_bdev
);
305 unsigned int chunk_sectors
= mddev
->chunk_sectors
;
306 unsigned int bio_sectors
= bvm
->bi_size
>> 9;
308 if (is_power_of_2(chunk_sectors
))
309 max
= (chunk_sectors
- ((sector
& (chunk_sectors
-1))
310 + bio_sectors
)) << 9;
312 max
= (chunk_sectors
- (sector_div(sector
, chunk_sectors
)
313 + bio_sectors
)) << 9;
314 if (max
< 0) max
= 0; /* bio_add cannot handle a negative return */
315 if (max
<= biovec
->bv_len
&& bio_sectors
== 0)
316 return biovec
->bv_len
;
321 static sector_t
raid0_size(mddev_t
*mddev
, sector_t sectors
, int raid_disks
)
323 sector_t array_sectors
= 0;
326 WARN_ONCE(sectors
|| raid_disks
,
327 "%s does not support generic reshape\n", __func__
);
329 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
330 array_sectors
+= rdev
->sectors
;
332 return array_sectors
;
335 static int raid0_run(mddev_t
*mddev
)
340 if (mddev
->chunk_sectors
== 0) {
341 printk(KERN_ERR
"md/raid0:%s: chunk size must be set.\n",
345 if (md_check_no_bitmap(mddev
))
347 blk_queue_max_hw_sectors(mddev
->queue
, mddev
->chunk_sectors
);
349 /* if private is not null, we are here after takeover */
350 if (mddev
->private == NULL
) {
351 ret
= create_strip_zones(mddev
, &conf
);
354 mddev
->private = conf
;
356 conf
= mddev
->private;
358 /* calculate array device size */
359 md_set_array_sectors(mddev
, raid0_size(mddev
, 0, 0));
361 printk(KERN_INFO
"md/raid0:%s: md_size is %llu sectors.\n",
363 (unsigned long long)mddev
->array_sectors
);
364 /* calculate the max read-ahead size.
365 * For read-ahead of large files to be effective, we need to
366 * readahead at least twice a whole stripe. i.e. number of devices
367 * multiplied by chunk size times 2.
368 * If an individual device has an ra_pages greater than the
369 * chunk size, then we will not drive that device as hard as it
370 * wants. We consider this a configuration error: a larger
371 * chunksize should be used in that case.
374 int stripe
= mddev
->raid_disks
*
375 (mddev
->chunk_sectors
<< 9) / PAGE_SIZE
;
376 if (mddev
->queue
->backing_dev_info
.ra_pages
< 2* stripe
)
377 mddev
->queue
->backing_dev_info
.ra_pages
= 2* stripe
;
380 blk_queue_merge_bvec(mddev
->queue
, raid0_mergeable_bvec
);
382 return md_integrity_register(mddev
);
385 static int raid0_stop(mddev_t
*mddev
)
387 raid0_conf_t
*conf
= mddev
->private;
389 blk_sync_queue(mddev
->queue
); /* the unplug fn references 'conf'*/
390 kfree(conf
->strip_zone
);
391 kfree(conf
->devlist
);
393 mddev
->private = NULL
;
397 /* Find the zone which holds a particular offset
398 * Update *sectorp to be an offset in that zone
400 static struct strip_zone
*find_zone(struct raid0_private_data
*conf
,
404 struct strip_zone
*z
= conf
->strip_zone
;
405 sector_t sector
= *sectorp
;
407 for (i
= 0; i
< conf
->nr_strip_zones
; i
++)
408 if (sector
< z
[i
].zone_end
) {
410 *sectorp
= sector
- z
[i
-1].zone_end
;
417 * remaps the bio to the target device. we separate two flows.
418 * power 2 flow and a general flow for the sake of perfromance
420 static mdk_rdev_t
*map_sector(mddev_t
*mddev
, struct strip_zone
*zone
,
421 sector_t sector
, sector_t
*sector_offset
)
423 unsigned int sect_in_chunk
;
425 raid0_conf_t
*conf
= mddev
->private;
426 int raid_disks
= conf
->strip_zone
[0].nb_dev
;
427 unsigned int chunk_sects
= mddev
->chunk_sectors
;
429 if (is_power_of_2(chunk_sects
)) {
430 int chunksect_bits
= ffz(~chunk_sects
);
431 /* find the sector offset inside the chunk */
432 sect_in_chunk
= sector
& (chunk_sects
- 1);
433 sector
>>= chunksect_bits
;
435 chunk
= *sector_offset
;
436 /* quotient is the chunk in real device*/
437 sector_div(chunk
, zone
->nb_dev
<< chunksect_bits
);
439 sect_in_chunk
= sector_div(sector
, chunk_sects
);
440 chunk
= *sector_offset
;
441 sector_div(chunk
, chunk_sects
* zone
->nb_dev
);
444 * position the bio over the real device
445 * real sector = chunk in device + starting of zone
446 * + the position in the chunk
448 *sector_offset
= (chunk
* chunk_sects
) + sect_in_chunk
;
449 return conf
->devlist
[(zone
- conf
->strip_zone
)*raid_disks
450 + sector_div(sector
, zone
->nb_dev
)];
454 * Is io distribute over 1 or more chunks ?
456 static inline int is_io_in_chunk_boundary(mddev_t
*mddev
,
457 unsigned int chunk_sects
, struct bio
*bio
)
459 if (likely(is_power_of_2(chunk_sects
))) {
460 return chunk_sects
>= ((bio
->bi_sector
& (chunk_sects
-1))
461 + (bio
->bi_size
>> 9));
463 sector_t sector
= bio
->bi_sector
;
464 return chunk_sects
>= (sector_div(sector
, chunk_sects
)
465 + (bio
->bi_size
>> 9));
469 static int raid0_make_request(mddev_t
*mddev
, struct bio
*bio
)
471 unsigned int chunk_sects
;
472 sector_t sector_offset
;
473 struct strip_zone
*zone
;
476 if (unlikely(bio
->bi_rw
& REQ_FLUSH
)) {
477 md_flush_request(mddev
, bio
);
481 chunk_sects
= mddev
->chunk_sectors
;
482 if (unlikely(!is_io_in_chunk_boundary(mddev
, chunk_sects
, bio
))) {
483 sector_t sector
= bio
->bi_sector
;
485 /* Sanity check -- queue functions should prevent this happening */
486 if (bio
->bi_vcnt
!= 1 ||
489 /* This is a one page bio that upper layers
490 * refuse to split for us, so we need to split it.
492 if (likely(is_power_of_2(chunk_sects
)))
493 bp
= bio_split(bio
, chunk_sects
- (sector
&
496 bp
= bio_split(bio
, chunk_sects
-
497 sector_div(sector
, chunk_sects
));
498 if (raid0_make_request(mddev
, &bp
->bio1
))
499 generic_make_request(&bp
->bio1
);
500 if (raid0_make_request(mddev
, &bp
->bio2
))
501 generic_make_request(&bp
->bio2
);
503 bio_pair_release(bp
);
507 sector_offset
= bio
->bi_sector
;
508 zone
= find_zone(mddev
->private, §or_offset
);
509 tmp_dev
= map_sector(mddev
, zone
, bio
->bi_sector
,
511 bio
->bi_bdev
= tmp_dev
->bdev
;
512 bio
->bi_sector
= sector_offset
+ zone
->dev_start
+
513 tmp_dev
->data_offset
;
515 * Let the main block layer submit the IO and resolve recursion:
520 printk("md/raid0:%s: make_request bug: can't convert block across chunks"
521 " or bigger than %dk %llu %d\n",
522 mdname(mddev
), chunk_sects
/ 2,
523 (unsigned long long)bio
->bi_sector
, bio
->bi_size
>> 10);
529 static void raid0_status(struct seq_file
*seq
, mddev_t
*mddev
)
534 char b
[BDEVNAME_SIZE
];
535 raid0_conf_t
*conf
= mddev
->private;
536 int raid_disks
= conf
->strip_zone
[0].nb_dev
;
539 sector_t zone_start
= 0;
542 for (j
= 0; j
< conf
->nr_strip_zones
; j
++) {
543 seq_printf(seq
, " z%d", j
);
544 seq_printf(seq
, "=[");
545 for (k
= 0; k
< conf
->strip_zone
[j
].nb_dev
; k
++)
546 seq_printf(seq
, "%s/", bdevname(
547 conf
->devlist
[j
*raid_disks
+ k
]
550 zone_size
= conf
->strip_zone
[j
].zone_end
- zone_start
;
551 seq_printf(seq
, "] ze=%lld ds=%lld s=%lld\n",
552 (unsigned long long)zone_start
>>1,
553 (unsigned long long)conf
->strip_zone
[j
].dev_start
>>1,
554 (unsigned long long)zone_size
>>1);
555 zone_start
= conf
->strip_zone
[j
].zone_end
;
558 seq_printf(seq
, " %dk chunks", mddev
->chunk_sectors
/ 2);
562 static void *raid0_takeover_raid45(mddev_t
*mddev
)
565 raid0_conf_t
*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 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
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
);
583 /* Set new parameters */
584 mddev
->new_level
= 0;
585 mddev
->new_layout
= 0;
586 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
588 mddev
->delta_disks
= -1;
589 /* make sure it will be not marked as dirty */
590 mddev
->recovery_cp
= MaxSector
;
592 create_strip_zones(mddev
, &priv_conf
);
596 static void *raid0_takeover_raid10(mddev_t
*mddev
)
598 raid0_conf_t
*priv_conf
;
601 * - far_copies must be 1
602 * - near_copies must be 2
603 * - disks number must be even
604 * - all mirrors must be already degraded
606 if (mddev
->layout
!= ((1 << 8) + 2)) {
607 printk(KERN_ERR
"md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
610 return ERR_PTR(-EINVAL
);
612 if (mddev
->raid_disks
& 1) {
613 printk(KERN_ERR
"md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
615 return ERR_PTR(-EINVAL
);
617 if (mddev
->degraded
!= (mddev
->raid_disks
>>1)) {
618 printk(KERN_ERR
"md/raid0:%s: All mirrors must be already degraded!\n",
620 return ERR_PTR(-EINVAL
);
623 /* Set new parameters */
624 mddev
->new_level
= 0;
625 mddev
->new_layout
= 0;
626 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
627 mddev
->delta_disks
= - mddev
->raid_disks
/ 2;
628 mddev
->raid_disks
+= mddev
->delta_disks
;
630 /* make sure it will be not marked as dirty */
631 mddev
->recovery_cp
= MaxSector
;
633 create_strip_zones(mddev
, &priv_conf
);
637 static void *raid0_takeover_raid1(mddev_t
*mddev
)
639 raid0_conf_t
*priv_conf
;
642 * - (N - 1) mirror drives must be already faulty
644 if ((mddev
->raid_disks
- 1) != mddev
->degraded
) {
645 printk(KERN_ERR
"md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
647 return ERR_PTR(-EINVAL
);
650 /* Set new parameters */
651 mddev
->new_level
= 0;
652 mddev
->new_layout
= 0;
653 mddev
->new_chunk_sectors
= 128; /* by default set chunk size to 64k */
654 mddev
->delta_disks
= 1 - mddev
->raid_disks
;
655 mddev
->raid_disks
= 1;
656 /* make sure it will be not marked as dirty */
657 mddev
->recovery_cp
= MaxSector
;
659 create_strip_zones(mddev
, &priv_conf
);
663 static void *raid0_takeover(mddev_t
*mddev
)
665 /* raid0 can take over:
666 * raid4 - if all data disks are active.
667 * raid5 - providing it is Raid4 layout and one disk is faulty
668 * raid10 - assuming we have all necessary active disks
669 * raid1 - with (N -1) mirror drives faulty
671 if (mddev
->level
== 4)
672 return raid0_takeover_raid45(mddev
);
674 if (mddev
->level
== 5) {
675 if (mddev
->layout
== ALGORITHM_PARITY_N
)
676 return raid0_takeover_raid45(mddev
);
678 printk(KERN_ERR
"md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
679 mdname(mddev
), ALGORITHM_PARITY_N
);
682 if (mddev
->level
== 10)
683 return raid0_takeover_raid10(mddev
);
685 if (mddev
->level
== 1)
686 return raid0_takeover_raid1(mddev
);
688 printk(KERN_ERR
"Takeover from raid%i to raid0 not supported\n",
691 return ERR_PTR(-EINVAL
);
694 static void raid0_quiesce(mddev_t
*mddev
, int state
)
698 static struct mdk_personality raid0_personality
=
702 .owner
= THIS_MODULE
,
703 .make_request
= raid0_make_request
,
706 .status
= raid0_status
,
708 .takeover
= raid0_takeover
,
709 .quiesce
= raid0_quiesce
,
712 static int __init
raid0_init (void)
714 return register_md_personality (&raid0_personality
);
717 static void raid0_exit (void)
719 unregister_md_personality (&raid0_personality
);
722 module_init(raid0_init
);
723 module_exit(raid0_exit
);
724 MODULE_LICENSE("GPL");
725 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
726 MODULE_ALIAS("md-personality-2"); /* RAID0 */
727 MODULE_ALIAS("md-raid0");
728 MODULE_ALIAS("md-level-0");