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 void raid0_unplug(struct request_queue
*q
)
30 mddev_t
*mddev
= q
->queuedata
;
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 for (i
=0; i
< raid_disks
; i
++) {
37 struct request_queue
*r_queue
= bdev_get_queue(devlist
[i
]->bdev
);
43 static int raid0_congested(void *data
, int bits
)
45 mddev_t
*mddev
= data
;
46 raid0_conf_t
*conf
= mddev
->private;
47 mdk_rdev_t
**devlist
= conf
->devlist
;
48 int raid_disks
= conf
->strip_zone
[0].nb_dev
;
51 if (mddev_congested(mddev
, bits
))
54 for (i
= 0; i
< raid_disks
&& !ret
; i
++) {
55 struct request_queue
*q
= bdev_get_queue(devlist
[i
]->bdev
);
57 ret
|= bdi_congested(&q
->backing_dev_info
, bits
);
63 * inform the user of the raid configuration
65 static void dump_zones(mddev_t
*mddev
)
68 sector_t zone_size
= 0;
69 sector_t zone_start
= 0;
70 char b
[BDEVNAME_SIZE
];
71 raid0_conf_t
*conf
= mddev
->private;
72 int raid_disks
= conf
->strip_zone
[0].nb_dev
;
73 printk(KERN_INFO
"******* %s configuration *********\n",
76 for (j
= 0; j
< conf
->nr_strip_zones
; j
++) {
77 printk(KERN_INFO
"zone%d=[", j
);
78 for (k
= 0; k
< conf
->strip_zone
[j
].nb_dev
; k
++)
79 printk(KERN_CONT
"%s/",
80 bdevname(conf
->devlist
[j
*raid_disks
82 printk(KERN_CONT
"]\n");
84 zone_size
= conf
->strip_zone
[j
].zone_end
- zone_start
;
85 printk(KERN_INFO
" zone offset=%llukb "
86 "device offset=%llukb size=%llukb\n",
87 (unsigned long long)zone_start
>>1,
88 (unsigned long long)conf
->strip_zone
[j
].dev_start
>>1,
89 (unsigned long long)zone_size
>>1);
90 zone_start
= conf
->strip_zone
[j
].zone_end
;
92 printk(KERN_INFO
"**********************************\n\n");
95 static int create_strip_zones(mddev_t
*mddev
, raid0_conf_t
**private_conf
)
98 sector_t curr_zone_end
, sectors
;
99 mdk_rdev_t
*smallest
, *rdev1
, *rdev2
, *rdev
, **dev
;
100 struct strip_zone
*zone
;
102 char b
[BDEVNAME_SIZE
];
103 raid0_conf_t
*conf
= kzalloc(sizeof(*conf
), GFP_KERNEL
);
107 list_for_each_entry(rdev1
, &mddev
->disks
, same_set
) {
108 printk(KERN_INFO
"md/raid0:%s: looking at %s\n",
110 bdevname(rdev1
->bdev
, b
));
113 /* round size to chunk_size */
114 sectors
= rdev1
->sectors
;
115 sector_div(sectors
, mddev
->chunk_sectors
);
116 rdev1
->sectors
= sectors
* mddev
->chunk_sectors
;
118 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
119 printk(KERN_INFO
"md/raid0:%s: comparing %s(%llu)",
121 bdevname(rdev1
->bdev
,b
),
122 (unsigned long long)rdev1
->sectors
);
123 printk(KERN_CONT
" with %s(%llu)\n",
124 bdevname(rdev2
->bdev
,b
),
125 (unsigned long long)rdev2
->sectors
);
126 if (rdev2
== rdev1
) {
127 printk(KERN_INFO
"md/raid0:%s: END\n",
131 if (rdev2
->sectors
== rdev1
->sectors
) {
133 * Not unique, don't count it as a new
136 printk(KERN_INFO
"md/raid0:%s: EQUAL\n",
141 printk(KERN_INFO
"md/raid0:%s: NOT EQUAL\n",
145 printk(KERN_INFO
"md/raid0:%s: ==> UNIQUE\n",
147 conf
->nr_strip_zones
++;
148 printk(KERN_INFO
"md/raid0:%s: %d zones\n",
149 mdname(mddev
), conf
->nr_strip_zones
);
152 printk(KERN_INFO
"md/raid0:%s: FINAL %d zones\n",
153 mdname(mddev
), conf
->nr_strip_zones
);
155 conf
->strip_zone
= kzalloc(sizeof(struct strip_zone
)*
156 conf
->nr_strip_zones
, GFP_KERNEL
);
157 if (!conf
->strip_zone
)
159 conf
->devlist
= kzalloc(sizeof(mdk_rdev_t
*)*
160 conf
->nr_strip_zones
*mddev
->raid_disks
,
165 /* The first zone must contain all devices, so here we check that
166 * there is a proper alignment of slots to devices and find them all
168 zone
= &conf
->strip_zone
[0];
173 list_for_each_entry(rdev1
, &mddev
->disks
, same_set
) {
174 int j
= rdev1
->raid_disk
;
176 if (mddev
->level
== 10)
177 /* taking over a raid10-n2 array */
180 if (j
< 0 || j
>= mddev
->raid_disks
) {
181 printk(KERN_ERR
"md/raid0:%s: bad disk number %d - "
182 "aborting!\n", mdname(mddev
), j
);
186 printk(KERN_ERR
"md/raid0:%s: multiple devices for %d - "
187 "aborting!\n", mdname(mddev
), j
);
192 disk_stack_limits(mddev
->gendisk
, rdev1
->bdev
,
193 rdev1
->data_offset
<< 9);
194 /* as we don't honour merge_bvec_fn, we must never risk
195 * violating it, so limit ->max_segments to 1, lying within
199 if (rdev1
->bdev
->bd_disk
->queue
->merge_bvec_fn
) {
200 blk_queue_max_segments(mddev
->queue
, 1);
201 blk_queue_segment_boundary(mddev
->queue
,
202 PAGE_CACHE_SIZE
- 1);
204 if (!smallest
|| (rdev1
->sectors
< smallest
->sectors
))
208 if (cnt
!= mddev
->raid_disks
) {
209 printk(KERN_ERR
"md/raid0:%s: too few disks (%d of %d) - "
210 "aborting!\n", mdname(mddev
), cnt
, mddev
->raid_disks
);
214 zone
->zone_end
= smallest
->sectors
* cnt
;
216 curr_zone_end
= zone
->zone_end
;
218 /* now do the other zones */
219 for (i
= 1; i
< conf
->nr_strip_zones
; i
++)
223 zone
= conf
->strip_zone
+ i
;
224 dev
= conf
->devlist
+ i
* mddev
->raid_disks
;
226 printk(KERN_INFO
"md/raid0:%s: zone %d\n",
228 zone
->dev_start
= smallest
->sectors
;
232 for (j
=0; j
<cnt
; j
++) {
233 rdev
= conf
->devlist
[j
];
234 printk(KERN_INFO
"md/raid0:%s: checking %s ...",
236 bdevname(rdev
->bdev
, b
));
237 if (rdev
->sectors
<= zone
->dev_start
) {
238 printk(KERN_CONT
" nope.\n");
241 printk(KERN_CONT
" contained as device %d\n", c
);
244 if (!smallest
|| rdev
->sectors
< smallest
->sectors
) {
246 printk(KERN_INFO
"md/raid0:%s: (%llu) is smallest!.\n",
248 (unsigned long long)rdev
->sectors
);
253 sectors
= (smallest
->sectors
- zone
->dev_start
) * c
;
254 printk(KERN_INFO
"md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
256 zone
->nb_dev
, (unsigned long long)sectors
);
258 curr_zone_end
+= sectors
;
259 zone
->zone_end
= curr_zone_end
;
261 printk(KERN_INFO
"md/raid0:%s: current zone start: %llu\n",
263 (unsigned long long)smallest
->sectors
);
265 mddev
->queue
->unplug_fn
= raid0_unplug
;
266 mddev
->queue
->backing_dev_info
.congested_fn
= raid0_congested
;
267 mddev
->queue
->backing_dev_info
.congested_data
= mddev
;
270 * now since we have the hard sector sizes, we can make sure
271 * chunk size is a multiple of that sector size
273 if ((mddev
->chunk_sectors
<< 9) % queue_logical_block_size(mddev
->queue
)) {
274 printk(KERN_ERR
"md/raid0:%s: chunk_size of %d not valid\n",
276 mddev
->chunk_sectors
<< 9);
280 blk_queue_io_min(mddev
->queue
, mddev
->chunk_sectors
<< 9);
281 blk_queue_io_opt(mddev
->queue
,
282 (mddev
->chunk_sectors
<< 9) * mddev
->raid_disks
);
284 printk(KERN_INFO
"md/raid0:%s: done.\n", mdname(mddev
));
285 *private_conf
= conf
;
289 kfree(conf
->strip_zone
);
290 kfree(conf
->devlist
);
292 *private_conf
= NULL
;
297 * raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
299 * @bvm: properties of new bio
300 * @biovec: the request that could be merged to it.
302 * Return amount of bytes we can accept at this offset
304 static int raid0_mergeable_bvec(struct request_queue
*q
,
305 struct bvec_merge_data
*bvm
,
306 struct bio_vec
*biovec
)
308 mddev_t
*mddev
= q
->queuedata
;
309 sector_t sector
= bvm
->bi_sector
+ get_start_sect(bvm
->bi_bdev
);
311 unsigned int chunk_sectors
= mddev
->chunk_sectors
;
312 unsigned int bio_sectors
= bvm
->bi_size
>> 9;
314 if (is_power_of_2(chunk_sectors
))
315 max
= (chunk_sectors
- ((sector
& (chunk_sectors
-1))
316 + bio_sectors
)) << 9;
318 max
= (chunk_sectors
- (sector_div(sector
, chunk_sectors
)
319 + bio_sectors
)) << 9;
320 if (max
< 0) max
= 0; /* bio_add cannot handle a negative return */
321 if (max
<= biovec
->bv_len
&& bio_sectors
== 0)
322 return biovec
->bv_len
;
327 static sector_t
raid0_size(mddev_t
*mddev
, sector_t sectors
, int raid_disks
)
329 sector_t array_sectors
= 0;
332 WARN_ONCE(sectors
|| raid_disks
,
333 "%s does not support generic reshape\n", __func__
);
335 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
336 array_sectors
+= rdev
->sectors
;
338 return array_sectors
;
341 static int raid0_run(mddev_t
*mddev
)
346 if (mddev
->chunk_sectors
== 0) {
347 printk(KERN_ERR
"md/raid0:%s: chunk size must be set.\n",
351 if (md_check_no_bitmap(mddev
))
353 blk_queue_max_hw_sectors(mddev
->queue
, mddev
->chunk_sectors
);
354 mddev
->queue
->queue_lock
= &mddev
->queue
->__queue_lock
;
356 /* if private is not null, we are here after takeover */
357 if (mddev
->private == NULL
) {
358 ret
= create_strip_zones(mddev
, &conf
);
361 mddev
->private = conf
;
363 conf
= mddev
->private;
364 if (conf
->scale_raid_disks
) {
366 for (i
=0; i
< conf
->strip_zone
[0].nb_dev
; i
++)
367 conf
->devlist
[i
]->raid_disk
/= conf
->scale_raid_disks
;
368 /* FIXME update sysfs rd links */
371 /* calculate array device size */
372 md_set_array_sectors(mddev
, raid0_size(mddev
, 0, 0));
374 printk(KERN_INFO
"md/raid0:%s: md_size is %llu sectors.\n",
376 (unsigned long long)mddev
->array_sectors
);
377 /* calculate the max read-ahead size.
378 * For read-ahead of large files to be effective, we need to
379 * readahead at least twice a whole stripe. i.e. number of devices
380 * multiplied by chunk size times 2.
381 * If an individual device has an ra_pages greater than the
382 * chunk size, then we will not drive that device as hard as it
383 * wants. We consider this a configuration error: a larger
384 * chunksize should be used in that case.
387 int stripe
= mddev
->raid_disks
*
388 (mddev
->chunk_sectors
<< 9) / PAGE_SIZE
;
389 if (mddev
->queue
->backing_dev_info
.ra_pages
< 2* stripe
)
390 mddev
->queue
->backing_dev_info
.ra_pages
= 2* stripe
;
393 blk_queue_merge_bvec(mddev
->queue
, raid0_mergeable_bvec
);
395 md_integrity_register(mddev
);
399 static int raid0_stop(mddev_t
*mddev
)
401 raid0_conf_t
*conf
= mddev
->private;
403 blk_sync_queue(mddev
->queue
); /* the unplug fn references 'conf'*/
404 kfree(conf
->strip_zone
);
405 kfree(conf
->devlist
);
407 mddev
->private = NULL
;
411 /* Find the zone which holds a particular offset
412 * Update *sectorp to be an offset in that zone
414 static struct strip_zone
*find_zone(struct raid0_private_data
*conf
,
418 struct strip_zone
*z
= conf
->strip_zone
;
419 sector_t sector
= *sectorp
;
421 for (i
= 0; i
< conf
->nr_strip_zones
; i
++)
422 if (sector
< z
[i
].zone_end
) {
424 *sectorp
= sector
- z
[i
-1].zone_end
;
431 * remaps the bio to the target device. we separate two flows.
432 * power 2 flow and a general flow for the sake of perfromance
434 static mdk_rdev_t
*map_sector(mddev_t
*mddev
, struct strip_zone
*zone
,
435 sector_t sector
, sector_t
*sector_offset
)
437 unsigned int sect_in_chunk
;
439 raid0_conf_t
*conf
= mddev
->private;
440 int raid_disks
= conf
->strip_zone
[0].nb_dev
;
441 unsigned int chunk_sects
= mddev
->chunk_sectors
;
443 if (is_power_of_2(chunk_sects
)) {
444 int chunksect_bits
= ffz(~chunk_sects
);
445 /* find the sector offset inside the chunk */
446 sect_in_chunk
= sector
& (chunk_sects
- 1);
447 sector
>>= chunksect_bits
;
449 chunk
= *sector_offset
;
450 /* quotient is the chunk in real device*/
451 sector_div(chunk
, zone
->nb_dev
<< chunksect_bits
);
453 sect_in_chunk
= sector_div(sector
, chunk_sects
);
454 chunk
= *sector_offset
;
455 sector_div(chunk
, chunk_sects
* zone
->nb_dev
);
458 * position the bio over the real device
459 * real sector = chunk in device + starting of zone
460 * + the position in the chunk
462 *sector_offset
= (chunk
* chunk_sects
) + sect_in_chunk
;
463 return conf
->devlist
[(zone
- conf
->strip_zone
)*raid_disks
464 + sector_div(sector
, zone
->nb_dev
)];
468 * Is io distribute over 1 or more chunks ?
470 static inline int is_io_in_chunk_boundary(mddev_t
*mddev
,
471 unsigned int chunk_sects
, struct bio
*bio
)
473 if (likely(is_power_of_2(chunk_sects
))) {
474 return chunk_sects
>= ((bio
->bi_sector
& (chunk_sects
-1))
475 + (bio
->bi_size
>> 9));
477 sector_t sector
= bio
->bi_sector
;
478 return chunk_sects
>= (sector_div(sector
, chunk_sects
)
479 + (bio
->bi_size
>> 9));
483 static int raid0_make_request(mddev_t
*mddev
, struct bio
*bio
)
485 unsigned int chunk_sects
;
486 sector_t sector_offset
;
487 struct strip_zone
*zone
;
490 if (unlikely(bio_rw_flagged(bio
, BIO_RW_BARRIER
))) {
491 md_barrier_request(mddev
, bio
);
495 chunk_sects
= mddev
->chunk_sectors
;
496 if (unlikely(!is_io_in_chunk_boundary(mddev
, chunk_sects
, bio
))) {
497 sector_t sector
= bio
->bi_sector
;
499 /* Sanity check -- queue functions should prevent this happening */
500 if (bio
->bi_vcnt
!= 1 ||
503 /* This is a one page bio that upper layers
504 * refuse to split for us, so we need to split it.
506 if (likely(is_power_of_2(chunk_sects
)))
507 bp
= bio_split(bio
, chunk_sects
- (sector
&
510 bp
= bio_split(bio
, chunk_sects
-
511 sector_div(sector
, chunk_sects
));
512 if (raid0_make_request(mddev
, &bp
->bio1
))
513 generic_make_request(&bp
->bio1
);
514 if (raid0_make_request(mddev
, &bp
->bio2
))
515 generic_make_request(&bp
->bio2
);
517 bio_pair_release(bp
);
521 sector_offset
= bio
->bi_sector
;
522 zone
= find_zone(mddev
->private, §or_offset
);
523 tmp_dev
= map_sector(mddev
, zone
, bio
->bi_sector
,
525 bio
->bi_bdev
= tmp_dev
->bdev
;
526 bio
->bi_sector
= sector_offset
+ zone
->dev_start
+
527 tmp_dev
->data_offset
;
529 * Let the main block layer submit the IO and resolve recursion:
534 printk("md/raid0:%s: make_request bug: can't convert block across chunks"
535 " or bigger than %dk %llu %d\n",
536 mdname(mddev
), chunk_sects
/ 2,
537 (unsigned long long)bio
->bi_sector
, bio
->bi_size
>> 10);
543 static void raid0_status(struct seq_file
*seq
, mddev_t
*mddev
)
548 char b
[BDEVNAME_SIZE
];
549 raid0_conf_t
*conf
= mddev
->private;
550 int raid_disks
= conf
->strip_zone
[0].nb_dev
;
553 sector_t zone_start
= 0;
556 for (j
= 0; j
< conf
->nr_strip_zones
; j
++) {
557 seq_printf(seq
, " z%d", j
);
558 seq_printf(seq
, "=[");
559 for (k
= 0; k
< conf
->strip_zone
[j
].nb_dev
; k
++)
560 seq_printf(seq
, "%s/", bdevname(
561 conf
->devlist
[j
*raid_disks
+ k
]
564 zone_size
= conf
->strip_zone
[j
].zone_end
- zone_start
;
565 seq_printf(seq
, "] ze=%lld ds=%lld s=%lld\n",
566 (unsigned long long)zone_start
>>1,
567 (unsigned long long)conf
->strip_zone
[j
].dev_start
>>1,
568 (unsigned long long)zone_size
>>1);
569 zone_start
= conf
->strip_zone
[j
].zone_end
;
572 seq_printf(seq
, " %dk chunks", mddev
->chunk_sectors
/ 2);
576 static void *raid0_takeover_raid5(mddev_t
*mddev
)
579 raid0_conf_t
*priv_conf
;
581 if (mddev
->degraded
!= 1) {
582 printk(KERN_ERR
"md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
585 return ERR_PTR(-EINVAL
);
588 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
589 /* check slot number for a disk */
590 if (rdev
->raid_disk
== mddev
->raid_disks
-1) {
591 printk(KERN_ERR
"md/raid0:%s: raid5 must have missing parity disk!\n",
593 return ERR_PTR(-EINVAL
);
597 /* Set new parameters */
598 mddev
->new_level
= 0;
599 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
601 mddev
->delta_disks
= -1;
602 /* make sure it will be not marked as dirty */
603 mddev
->recovery_cp
= MaxSector
;
605 create_strip_zones(mddev
, &priv_conf
);
609 static void *raid0_takeover_raid10(mddev_t
*mddev
)
611 raid0_conf_t
*priv_conf
;
614 * - far_copies must be 1
615 * - near_copies must be 2
616 * - disks number must be even
617 * - all mirrors must be already degraded
619 if (mddev
->layout
!= ((1 << 8) + 2)) {
620 printk(KERN_ERR
"md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
623 return ERR_PTR(-EINVAL
);
625 if (mddev
->raid_disks
& 1) {
626 printk(KERN_ERR
"md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
628 return ERR_PTR(-EINVAL
);
630 if (mddev
->degraded
!= (mddev
->raid_disks
>>1)) {
631 printk(KERN_ERR
"md/raid0:%s: All mirrors must be already degraded!\n",
633 return ERR_PTR(-EINVAL
);
636 /* Set new parameters */
637 mddev
->new_level
= 0;
638 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
639 mddev
->delta_disks
= - mddev
->raid_disks
/ 2;
640 mddev
->raid_disks
+= mddev
->delta_disks
;
642 /* make sure it will be not marked as dirty */
643 mddev
->recovery_cp
= MaxSector
;
645 create_strip_zones(mddev
, &priv_conf
);
646 priv_conf
->scale_raid_disks
= 2;
650 static void *raid0_takeover(mddev_t
*mddev
)
652 /* raid0 can take over:
653 * raid5 - providing it is Raid4 layout and one disk is faulty
654 * raid10 - assuming we have all necessary active disks
656 if (mddev
->level
== 5) {
657 if (mddev
->layout
== ALGORITHM_PARITY_N
)
658 return raid0_takeover_raid5(mddev
);
660 printk(KERN_ERR
"md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
661 mdname(mddev
), ALGORITHM_PARITY_N
);
664 if (mddev
->level
== 10)
665 return raid0_takeover_raid10(mddev
);
667 return ERR_PTR(-EINVAL
);
670 static void raid0_quiesce(mddev_t
*mddev
, int state
)
674 static struct mdk_personality raid0_personality
=
678 .owner
= THIS_MODULE
,
679 .make_request
= raid0_make_request
,
682 .status
= raid0_status
,
684 .takeover
= raid0_takeover
,
685 .quiesce
= raid0_quiesce
,
688 static int __init
raid0_init (void)
690 return register_md_personality (&raid0_personality
);
693 static void raid0_exit (void)
695 unregister_md_personality (&raid0_personality
);
698 module_init(raid0_init
);
699 module_exit(raid0_exit
);
700 MODULE_LICENSE("GPL");
701 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
702 MODULE_ALIAS("md-personality-2"); /* RAID0 */
703 MODULE_ALIAS("md-raid0");
704 MODULE_ALIAS("md-level-0");