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>
26 static void raid0_unplug(struct request_queue
*q
)
28 mddev_t
*mddev
= q
->queuedata
;
29 raid0_conf_t
*conf
= mddev
->private;
30 mdk_rdev_t
**devlist
= conf
->devlist
;
33 for (i
=0; i
<mddev
->raid_disks
; i
++) {
34 struct request_queue
*r_queue
= bdev_get_queue(devlist
[i
]->bdev
);
40 static int raid0_congested(void *data
, int bits
)
42 mddev_t
*mddev
= data
;
43 raid0_conf_t
*conf
= mddev
->private;
44 mdk_rdev_t
**devlist
= conf
->devlist
;
47 for (i
= 0; i
< mddev
->raid_disks
&& !ret
; i
++) {
48 struct request_queue
*q
= bdev_get_queue(devlist
[i
]->bdev
);
50 ret
|= bdi_congested(&q
->backing_dev_info
, bits
);
56 * inform the user of the raid configuration
58 static void dump_zones(mddev_t
*mddev
)
61 sector_t zone_size
= 0;
62 sector_t zone_start
= 0;
63 char b
[BDEVNAME_SIZE
];
64 raid0_conf_t
*conf
= mddev
->private;
65 printk(KERN_INFO
"******* %s configuration *********\n",
68 for (j
= 0; j
< conf
->nr_strip_zones
; j
++) {
69 printk(KERN_INFO
"zone%d=[", j
);
70 for (k
= 0; k
< conf
->strip_zone
[j
].nb_dev
; k
++)
72 bdevname(conf
->devlist
[j
*mddev
->raid_disks
76 zone_size
= conf
->strip_zone
[j
].zone_end
- zone_start
;
77 printk(KERN_INFO
" zone offset=%llukb "
78 "device offset=%llukb size=%llukb\n",
79 (unsigned long long)zone_start
>>1,
80 (unsigned long long)conf
->strip_zone
[j
].dev_start
>>1,
81 (unsigned long long)zone_size
>>1);
82 zone_start
= conf
->strip_zone
[j
].zone_end
;
84 printk(KERN_INFO
"**********************************\n\n");
87 static int create_strip_zones(mddev_t
*mddev
)
90 sector_t curr_zone_end
, sectors
;
91 mdk_rdev_t
*smallest
, *rdev1
, *rdev2
, *rdev
, **dev
;
92 struct strip_zone
*zone
;
94 char b
[BDEVNAME_SIZE
];
95 raid0_conf_t
*conf
= kzalloc(sizeof(*conf
), GFP_KERNEL
);
99 list_for_each_entry(rdev1
, &mddev
->disks
, same_set
) {
100 printk(KERN_INFO
"raid0: 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 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
110 printk(KERN_INFO
"raid0: comparing %s(%llu)",
111 bdevname(rdev1
->bdev
,b
),
112 (unsigned long long)rdev1
->sectors
);
113 printk(KERN_INFO
" with %s(%llu)\n",
114 bdevname(rdev2
->bdev
,b
),
115 (unsigned long long)rdev2
->sectors
);
116 if (rdev2
== rdev1
) {
117 printk(KERN_INFO
"raid0: END\n");
120 if (rdev2
->sectors
== rdev1
->sectors
) {
122 * Not unique, don't count it as a new
125 printk(KERN_INFO
"raid0: EQUAL\n");
129 printk(KERN_INFO
"raid0: NOT EQUAL\n");
132 printk(KERN_INFO
"raid0: ==> UNIQUE\n");
133 conf
->nr_strip_zones
++;
134 printk(KERN_INFO
"raid0: %d zones\n",
135 conf
->nr_strip_zones
);
138 printk(KERN_INFO
"raid0: FINAL %d zones\n", 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 (j
< 0 || j
>= mddev
->raid_disks
) {
162 printk(KERN_ERR
"raid0: bad disk number %d - "
167 printk(KERN_ERR
"raid0: multiple devices for %d - "
173 blk_queue_stack_limits(mddev
->queue
,
174 rdev1
->bdev
->bd_disk
->queue
);
175 /* as we don't honour merge_bvec_fn, we must never risk
176 * violating it, so limit ->max_sector to one PAGE, as
177 * a one page request is never in violation.
180 if (rdev1
->bdev
->bd_disk
->queue
->merge_bvec_fn
&&
181 queue_max_sectors(mddev
->queue
) > (PAGE_SIZE
>>9))
182 blk_queue_max_sectors(mddev
->queue
, PAGE_SIZE
>>9);
184 if (!smallest
|| (rdev1
->sectors
< smallest
->sectors
))
188 if (cnt
!= mddev
->raid_disks
) {
189 printk(KERN_ERR
"raid0: too few disks (%d of %d) - "
190 "aborting!\n", cnt
, mddev
->raid_disks
);
194 zone
->zone_end
= smallest
->sectors
* cnt
;
196 curr_zone_end
= zone
->zone_end
;
198 /* now do the other zones */
199 for (i
= 1; i
< conf
->nr_strip_zones
; i
++)
201 zone
= conf
->strip_zone
+ i
;
202 dev
= conf
->devlist
+ i
* mddev
->raid_disks
;
204 printk(KERN_INFO
"raid0: zone %d\n", i
);
205 zone
->dev_start
= smallest
->sectors
;
209 for (j
=0; j
<cnt
; j
++) {
210 char b
[BDEVNAME_SIZE
];
211 rdev
= conf
->devlist
[j
];
212 printk(KERN_INFO
"raid0: checking %s ...",
213 bdevname(rdev
->bdev
, b
));
214 if (rdev
->sectors
<= zone
->dev_start
) {
215 printk(KERN_INFO
" nope.\n");
218 printk(KERN_INFO
" contained as device %d\n", c
);
221 if (!smallest
|| rdev
->sectors
< smallest
->sectors
) {
223 printk(KERN_INFO
" (%llu) is smallest!.\n",
224 (unsigned long long)rdev
->sectors
);
229 sectors
= (smallest
->sectors
- zone
->dev_start
) * c
;
230 printk(KERN_INFO
"raid0: zone->nb_dev: %d, sectors: %llu\n",
231 zone
->nb_dev
, (unsigned long long)sectors
);
233 curr_zone_end
+= sectors
;
234 zone
->zone_end
= curr_zone_end
;
236 printk(KERN_INFO
"raid0: current zone start: %llu\n",
237 (unsigned long long)smallest
->sectors
);
239 mddev
->queue
->unplug_fn
= raid0_unplug
;
240 mddev
->queue
->backing_dev_info
.congested_fn
= raid0_congested
;
241 mddev
->queue
->backing_dev_info
.congested_data
= mddev
;
244 * now since we have the hard sector sizes, we can make sure
245 * chunk size is a multiple of that sector size
247 if ((mddev
->chunk_sectors
<< 9) % queue_logical_block_size(mddev
->queue
)) {
248 printk(KERN_ERR
"%s chunk_size of %d not valid\n",
250 mddev
->chunk_sectors
<< 9);
253 printk(KERN_INFO
"raid0: done.\n");
254 mddev
->private = conf
;
257 kfree(conf
->strip_zone
);
258 kfree(conf
->devlist
);
260 mddev
->private = NULL
;
265 * raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
267 * @bvm: properties of new bio
268 * @biovec: the request that could be merged to it.
270 * Return amount of bytes we can accept at this offset
272 static int raid0_mergeable_bvec(struct request_queue
*q
,
273 struct bvec_merge_data
*bvm
,
274 struct bio_vec
*biovec
)
276 mddev_t
*mddev
= q
->queuedata
;
277 sector_t sector
= bvm
->bi_sector
+ get_start_sect(bvm
->bi_bdev
);
279 unsigned int chunk_sectors
= mddev
->chunk_sectors
;
280 unsigned int bio_sectors
= bvm
->bi_size
>> 9;
282 if (is_power_of_2(chunk_sectors
))
283 max
= (chunk_sectors
- ((sector
& (chunk_sectors
-1))
284 + bio_sectors
)) << 9;
286 max
= (chunk_sectors
- (sector_div(sector
, chunk_sectors
)
287 + bio_sectors
)) << 9;
288 if (max
< 0) max
= 0; /* bio_add cannot handle a negative return */
289 if (max
<= biovec
->bv_len
&& bio_sectors
== 0)
290 return biovec
->bv_len
;
295 static sector_t
raid0_size(mddev_t
*mddev
, sector_t sectors
, int raid_disks
)
297 sector_t array_sectors
= 0;
300 WARN_ONCE(sectors
|| raid_disks
,
301 "%s does not support generic reshape\n", __func__
);
303 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
304 array_sectors
+= rdev
->sectors
;
306 return array_sectors
;
309 static int raid0_run(mddev_t
*mddev
)
313 if (mddev
->chunk_sectors
== 0) {
314 printk(KERN_ERR
"md/raid0: chunk size must be set.\n");
317 if (md_check_no_bitmap(mddev
))
319 blk_queue_max_sectors(mddev
->queue
, mddev
->chunk_sectors
);
320 mddev
->queue
->queue_lock
= &mddev
->queue
->__queue_lock
;
322 ret
= create_strip_zones(mddev
);
326 /* calculate array device size */
327 md_set_array_sectors(mddev
, raid0_size(mddev
, 0, 0));
329 printk(KERN_INFO
"raid0 : md_size is %llu sectors.\n",
330 (unsigned long long)mddev
->array_sectors
);
331 /* calculate the max read-ahead size.
332 * For read-ahead of large files to be effective, we need to
333 * readahead at least twice a whole stripe. i.e. number of devices
334 * multiplied by chunk size times 2.
335 * If an individual device has an ra_pages greater than the
336 * chunk size, then we will not drive that device as hard as it
337 * wants. We consider this a configuration error: a larger
338 * chunksize should be used in that case.
341 int stripe
= mddev
->raid_disks
*
342 (mddev
->chunk_sectors
<< 9) / PAGE_SIZE
;
343 if (mddev
->queue
->backing_dev_info
.ra_pages
< 2* stripe
)
344 mddev
->queue
->backing_dev_info
.ra_pages
= 2* stripe
;
347 blk_queue_merge_bvec(mddev
->queue
, raid0_mergeable_bvec
);
352 static int raid0_stop(mddev_t
*mddev
)
354 raid0_conf_t
*conf
= mddev
->private;
356 blk_sync_queue(mddev
->queue
); /* the unplug fn references 'conf'*/
357 kfree(conf
->strip_zone
);
358 kfree(conf
->devlist
);
360 mddev
->private = NULL
;
364 /* Find the zone which holds a particular offset
365 * Update *sectorp to be an offset in that zone
367 static struct strip_zone
*find_zone(struct raid0_private_data
*conf
,
371 struct strip_zone
*z
= conf
->strip_zone
;
372 sector_t sector
= *sectorp
;
374 for (i
= 0; i
< conf
->nr_strip_zones
; i
++)
375 if (sector
< z
[i
].zone_end
) {
377 *sectorp
= sector
- z
[i
-1].zone_end
;
384 * remaps the bio to the target device. we separate two flows.
385 * power 2 flow and a general flow for the sake of perfromance
387 static mdk_rdev_t
*map_sector(mddev_t
*mddev
, struct strip_zone
*zone
,
388 sector_t sector
, sector_t
*sector_offset
)
390 unsigned int sect_in_chunk
;
392 raid0_conf_t
*conf
= mddev
->private;
393 unsigned int chunk_sects
= mddev
->chunk_sectors
;
395 if (is_power_of_2(chunk_sects
)) {
396 int chunksect_bits
= ffz(~chunk_sects
);
397 /* find the sector offset inside the chunk */
398 sect_in_chunk
= sector
& (chunk_sects
- 1);
399 sector
>>= chunksect_bits
;
401 chunk
= *sector_offset
;
402 /* quotient is the chunk in real device*/
403 sector_div(chunk
, zone
->nb_dev
<< chunksect_bits
);
405 sect_in_chunk
= sector_div(sector
, chunk_sects
);
406 chunk
= *sector_offset
;
407 sector_div(chunk
, chunk_sects
* zone
->nb_dev
);
410 * position the bio over the real device
411 * real sector = chunk in device + starting of zone
412 * + the position in the chunk
414 *sector_offset
= (chunk
* chunk_sects
) + sect_in_chunk
;
415 return conf
->devlist
[(zone
- conf
->strip_zone
)*mddev
->raid_disks
416 + sector_div(sector
, zone
->nb_dev
)];
420 * Is io distribute over 1 or more chunks ?
422 static inline int is_io_in_chunk_boundary(mddev_t
*mddev
,
423 unsigned int chunk_sects
, struct bio
*bio
)
425 if (likely(is_power_of_2(chunk_sects
))) {
426 return chunk_sects
>= ((bio
->bi_sector
& (chunk_sects
-1))
427 + (bio
->bi_size
>> 9));
429 sector_t sector
= bio
->bi_sector
;
430 return chunk_sects
>= (sector_div(sector
, chunk_sects
)
431 + (bio
->bi_size
>> 9));
435 static int raid0_make_request(struct request_queue
*q
, struct bio
*bio
)
437 mddev_t
*mddev
= q
->queuedata
;
438 unsigned int chunk_sects
;
439 sector_t sector_offset
;
440 struct strip_zone
*zone
;
442 const int rw
= bio_data_dir(bio
);
445 if (unlikely(bio_barrier(bio
))) {
446 bio_endio(bio
, -EOPNOTSUPP
);
450 cpu
= part_stat_lock();
451 part_stat_inc(cpu
, &mddev
->gendisk
->part0
, ios
[rw
]);
452 part_stat_add(cpu
, &mddev
->gendisk
->part0
, sectors
[rw
],
456 chunk_sects
= mddev
->chunk_sectors
;
457 if (unlikely(!is_io_in_chunk_boundary(mddev
, chunk_sects
, bio
))) {
458 sector_t sector
= bio
->bi_sector
;
460 /* Sanity check -- queue functions should prevent this happening */
461 if (bio
->bi_vcnt
!= 1 ||
464 /* This is a one page bio that upper layers
465 * refuse to split for us, so we need to split it.
467 if (likely(is_power_of_2(chunk_sects
)))
468 bp
= bio_split(bio
, chunk_sects
- (sector
&
471 bp
= bio_split(bio
, chunk_sects
-
472 sector_div(sector
, chunk_sects
));
473 if (raid0_make_request(q
, &bp
->bio1
))
474 generic_make_request(&bp
->bio1
);
475 if (raid0_make_request(q
, &bp
->bio2
))
476 generic_make_request(&bp
->bio2
);
478 bio_pair_release(bp
);
482 sector_offset
= bio
->bi_sector
;
483 zone
= find_zone(mddev
->private, §or_offset
);
484 tmp_dev
= map_sector(mddev
, zone
, bio
->bi_sector
,
486 bio
->bi_bdev
= tmp_dev
->bdev
;
487 bio
->bi_sector
= sector_offset
+ zone
->dev_start
+
488 tmp_dev
->data_offset
;
490 * Let the main block layer submit the IO and resolve recursion:
495 printk("raid0_make_request bug: can't convert block across chunks"
496 " or bigger than %dk %llu %d\n", chunk_sects
/ 2,
497 (unsigned long long)bio
->bi_sector
, bio
->bi_size
>> 10);
503 static void raid0_status(struct seq_file
*seq
, mddev_t
*mddev
)
508 char b
[BDEVNAME_SIZE
];
509 raid0_conf_t
*conf
= mddev
->private;
512 sector_t zone_start
= 0;
515 for (j
= 0; j
< conf
->nr_strip_zones
; j
++) {
516 seq_printf(seq
, " z%d", j
);
517 seq_printf(seq
, "=[");
518 for (k
= 0; k
< conf
->strip_zone
[j
].nb_dev
; k
++)
519 seq_printf(seq
, "%s/", bdevname(
520 conf
->devlist
[j
*mddev
->raid_disks
+ k
]
523 zone_size
= conf
->strip_zone
[j
].zone_end
- zone_start
;
524 seq_printf(seq
, "] ze=%lld ds=%lld s=%lld\n",
525 (unsigned long long)zone_start
>>1,
526 (unsigned long long)conf
->strip_zone
[j
].dev_start
>>1,
527 (unsigned long long)zone_size
>>1);
528 zone_start
= conf
->strip_zone
[j
].zone_end
;
531 seq_printf(seq
, " %dk chunks", mddev
->chunk_sectors
/ 2);
535 static struct mdk_personality raid0_personality
=
539 .owner
= THIS_MODULE
,
540 .make_request
= raid0_make_request
,
543 .status
= raid0_status
,
547 static int __init
raid0_init (void)
549 return register_md_personality (&raid0_personality
);
552 static void raid0_exit (void)
554 unregister_md_personality (&raid0_personality
);
557 module_init(raid0_init
);
558 module_exit(raid0_exit
);
559 MODULE_LICENSE("GPL");
560 MODULE_ALIAS("md-personality-2"); /* RAID0 */
561 MODULE_ALIAS("md-raid0");
562 MODULE_ALIAS("md-level-0");