?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
md: raid0: chunk_sectors cleanups.
[linux-2.6/kvm.git] / drivers / md / raid0.c
blob11e384253718487b3fb99f3804825281094f72f1
1 /*
2 raid0.c : Multiple Devices driver for Linux
3 Copyright (C) 1994-96 Marc ZYNGIER
4 <zyngier@ufr-info-p7.ibp.fr> or
5 <maz@gloups.fdn.fr>
6 Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
7
8
9 RAID-0 management functions.
10
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)
14 any later version.
15
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.
19 */
20
21 #include <linux/blkdev.h>
22 #include <linux/seq_file.h>
23 #include "md.h"
24 #include "raid0.h"
25
26 static void raid0_unplug(struct request_queue *q)
27 {
28 mddev_t *mddev = q->queuedata;
29 raid0_conf_t *conf = mddev->private;
30 mdk_rdev_t **devlist = conf->devlist;
31 int i;
32
33 for (i=0; i<mddev->raid_disks; i++) {
34 struct request_queue *r_queue = bdev_get_queue(devlist[i]->bdev);
35
36 blk_unplug(r_queue);
37 }
38 }
39
40 static int raid0_congested(void *data, int bits)
41 {
42 mddev_t *mddev = data;
43 raid0_conf_t *conf = mddev->private;
44 mdk_rdev_t **devlist = conf->devlist;
45 int i, ret = 0;
46
47 for (i = 0; i < mddev->raid_disks && !ret ; i++) {
48 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
49
50 ret |= bdi_congested(&q->backing_dev_info, bits);
51 }
52 return ret;
53 }
54
55 /*
56 * inform the user of the raid configuration
57 */
58 static void dump_zones(mddev_t *mddev)
59 {
60 int j, k, h;
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",
66 mdname(mddev));
67 h = 0;
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++)
71 printk("%s/",
72 bdevname(conf->devlist[j*mddev->raid_disks
73 + k]->bdev, b));
74 printk("]\n");
75
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;
83 }
84 printk(KERN_INFO "**********************************\n\n");
85 }
86
87 static int create_strip_zones(mddev_t *mddev)
88 {
89 int i, c, j, err;
90 sector_t curr_zone_end, sectors;
91 mdk_rdev_t *smallest, *rdev1, *rdev2, *rdev, **dev;
92 struct strip_zone *zone;
93 int cnt;
94 char b[BDEVNAME_SIZE];
95 raid0_conf_t *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
96
97 if (!conf)
98 return -ENOMEM;
99 list_for_each_entry(rdev1, &mddev->disks, same_set) {
100 printk(KERN_INFO "raid0: looking at %s\n",
101 bdevname(rdev1->bdev,b));
102 c = 0;
103 list_for_each_entry(rdev2, &mddev->disks, same_set) {
104 printk(KERN_INFO "raid0: comparing %s(%llu)",
105 bdevname(rdev1->bdev,b),
106 (unsigned long long)rdev1->sectors);
107 printk(KERN_INFO " with %s(%llu)\n",
108 bdevname(rdev2->bdev,b),
109 (unsigned long long)rdev2->sectors);
110 if (rdev2 == rdev1) {
111 printk(KERN_INFO "raid0: END\n");
112 break;
113 }
114 if (rdev2->sectors == rdev1->sectors) {
115 /*
116 * Not unique, don't count it as a new
117 * group
118 */
119 printk(KERN_INFO "raid0: EQUAL\n");
120 c = 1;
121 break;
122 }
123 printk(KERN_INFO "raid0: NOT EQUAL\n");
124 }
125 if (!c) {
126 printk(KERN_INFO "raid0: ==> UNIQUE\n");
127 conf->nr_strip_zones++;
128 printk(KERN_INFO "raid0: %d zones\n",
129 conf->nr_strip_zones);
130 }
131 }
132 printk(KERN_INFO "raid0: FINAL %d zones\n", conf->nr_strip_zones);
133 err = -ENOMEM;
134 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
135 conf->nr_strip_zones, GFP_KERNEL);
136 if (!conf->strip_zone)
137 goto abort;
138 conf->devlist = kzalloc(sizeof(mdk_rdev_t*)*
139 conf->nr_strip_zones*mddev->raid_disks,
140 GFP_KERNEL);
141 if (!conf->devlist)
142 goto abort;
143
144 /* The first zone must contain all devices, so here we check that
145 * there is a proper alignment of slots to devices and find them all
146 */
147 zone = &conf->strip_zone[0];
148 cnt = 0;
149 smallest = NULL;
150 dev = conf->devlist;
151 err = -EINVAL;
152 list_for_each_entry(rdev1, &mddev->disks, same_set) {
153 int j = rdev1->raid_disk;
154
155 if (j < 0 || j >= mddev->raid_disks) {
156 printk(KERN_ERR "raid0: bad disk number %d - "
157 "aborting!\n", j);
158 goto abort;
159 }
160 if (dev[j]) {
161 printk(KERN_ERR "raid0: multiple devices for %d - "
162 "aborting!\n", j);
163 goto abort;
164 }
165 dev[j] = rdev1;
166
167 blk_queue_stack_limits(mddev->queue,
168 rdev1->bdev->bd_disk->queue);
169 /* as we don't honour merge_bvec_fn, we must never risk
170 * violating it, so limit ->max_sector to one PAGE, as
171 * a one page request is never in violation.
172 */
173
174 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn &&
175 queue_max_sectors(mddev->queue) > (PAGE_SIZE>>9))
176 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
177
178 if (!smallest || (rdev1->sectors < smallest->sectors))
179 smallest = rdev1;
180 cnt++;
181 }
182 if (cnt != mddev->raid_disks) {
183 printk(KERN_ERR "raid0: too few disks (%d of %d) - "
184 "aborting!\n", cnt, mddev->raid_disks);
185 goto abort;
186 }
187 zone->nb_dev = cnt;
188 zone->zone_end = smallest->sectors * cnt;
189
190 curr_zone_end = zone->zone_end;
191
192 /* now do the other zones */
193 for (i = 1; i < conf->nr_strip_zones; i++)
194 {
195 zone = conf->strip_zone + i;
196 dev = conf->devlist + i * mddev->raid_disks;
197
198 printk(KERN_INFO "raid0: zone %d\n", i);
199 zone->dev_start = smallest->sectors;
200 smallest = NULL;
201 c = 0;
202
203 for (j=0; j<cnt; j++) {
204 char b[BDEVNAME_SIZE];
205 rdev = conf->devlist[j];
206 printk(KERN_INFO "raid0: checking %s ...",
207 bdevname(rdev->bdev, b));
208 if (rdev->sectors <= zone->dev_start) {
209 printk(KERN_INFO " nope.\n");
210 continue;
211 }
212 printk(KERN_INFO " contained as device %d\n", c);
213 dev[c] = rdev;
214 c++;
215 if (!smallest || rdev->sectors < smallest->sectors) {
216 smallest = rdev;
217 printk(KERN_INFO " (%llu) is smallest!.\n",
218 (unsigned long long)rdev->sectors);
219 }
220 }
221
222 zone->nb_dev = c;
223 sectors = (smallest->sectors - zone->dev_start) * c;
224 printk(KERN_INFO "raid0: zone->nb_dev: %d, sectors: %llu\n",
225 zone->nb_dev, (unsigned long long)sectors);
226
227 curr_zone_end += sectors;
228 zone->zone_end = curr_zone_end;
229
230 printk(KERN_INFO "raid0: current zone start: %llu\n",
231 (unsigned long long)smallest->sectors);
232 }
233 mddev->queue->unplug_fn = raid0_unplug;
234 mddev->queue->backing_dev_info.congested_fn = raid0_congested;
235 mddev->queue->backing_dev_info.congested_data = mddev;
236
237 /*
238 * now since we have the hard sector sizes, we can make sure
239 * chunk size is a multiple of that sector size
240 */
241 if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
242 printk(KERN_ERR "%s chunk_size of %d not valid\n",
243 mdname(mddev),
244 mddev->chunk_sectors << 9);
245 goto abort;
246 }
247 printk(KERN_INFO "raid0: done.\n");
248 mddev->private = conf;
249 return 0;
250 abort:
251 kfree(conf->strip_zone);
252 kfree(conf->devlist);
253 kfree(conf);
254 mddev->private = NULL;
255 return err;
256 }
257
258 /**
259 * raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
260 * @q: request queue
261 * @bvm: properties of new bio
262 * @biovec: the request that could be merged to it.
263 *
264 * Return amount of bytes we can accept at this offset
265 */
266 static int raid0_mergeable_bvec(struct request_queue *q,
267 struct bvec_merge_data *bvm,
268 struct bio_vec *biovec)
269 {
270 mddev_t *mddev = q->queuedata;
271 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
272 int max;
273 unsigned int chunk_sectors = mddev->chunk_sectors;
274 unsigned int bio_sectors = bvm->bi_size >> 9;
275
276 if (is_power_of_2(chunk_sectors))
277 max = (chunk_sectors - ((sector & (chunk_sectors-1))
278 + bio_sectors)) << 9;
279 else
280 max = (chunk_sectors - (sector_div(sector, chunk_sectors)
281 + bio_sectors)) << 9;
282 if (max < 0) max = 0; /* bio_add cannot handle a negative return */
283 if (max <= biovec->bv_len && bio_sectors == 0)
284 return biovec->bv_len;
285 else
286 return max;
287 }
288
289 static sector_t raid0_size(mddev_t *mddev, sector_t sectors, int raid_disks)
290 {
291 sector_t array_sectors = 0;
292 mdk_rdev_t *rdev;
293
294 WARN_ONCE(sectors || raid_disks,
295 "%s does not support generic reshape\n", __func__);
296
297 list_for_each_entry(rdev, &mddev->disks, same_set)
298 array_sectors += rdev->sectors;
299
300 return array_sectors;
301 }
302
303 static int raid0_run(mddev_t *mddev)
304 {
305 int ret;
306
307 if (mddev->chunk_sectors == 0) {
308 printk(KERN_ERR "md/raid0: chunk size must be set.\n");
309 return -EINVAL;
310 }
311 blk_queue_max_sectors(mddev->queue, mddev->chunk_sectors);
312 mddev->queue->queue_lock = &mddev->queue->__queue_lock;
313
314 ret = create_strip_zones(mddev);
315 if (ret < 0)
316 return ret;
317
318 /* calculate array device size */
319 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
320
321 printk(KERN_INFO "raid0 : md_size is %llu sectors.\n",
322 (unsigned long long)mddev->array_sectors);
323 /* calculate the max read-ahead size.
324 * For read-ahead of large files to be effective, we need to
325 * readahead at least twice a whole stripe. i.e. number of devices
326 * multiplied by chunk size times 2.
327 * If an individual device has an ra_pages greater than the
328 * chunk size, then we will not drive that device as hard as it
329 * wants. We consider this a configuration error: a larger
330 * chunksize should be used in that case.
331 */
332 {
333 int stripe = mddev->raid_disks *
334 (mddev->chunk_sectors << 9) / PAGE_SIZE;
335 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
336 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
337 }
338
339 blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
340 dump_zones(mddev);
341 return 0;
342 }
343
344 static int raid0_stop(mddev_t *mddev)
345 {
346 raid0_conf_t *conf = mddev->private;
347
348 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
349 kfree(conf->strip_zone);
350 kfree(conf->devlist);
351 kfree(conf);
352 mddev->private = NULL;
353 return 0;
354 }
355
356 /* Find the zone which holds a particular offset
357 * Update *sectorp to be an offset in that zone
358 */
359 static struct strip_zone *find_zone(struct raid0_private_data *conf,
360 sector_t *sectorp)
361 {
362 int i;
363 struct strip_zone *z = conf->strip_zone;
364 sector_t sector = *sectorp;
365
366 for (i = 0; i < conf->nr_strip_zones; i++)
367 if (sector < z[i].zone_end) {
368 if (i)
369 *sectorp = sector - z[i-1].zone_end;
370 return z + i;
371 }
372 BUG();
373 }
374
375 /*
376 * remaps the bio to the target device. we separate two flows.
377 * power 2 flow and a general flow for the sake of perfromance
378 */
379 static mdk_rdev_t *map_sector(mddev_t *mddev, struct strip_zone *zone,
380 sector_t sector, sector_t *sector_offset)
381 {
382 unsigned int sect_in_chunk;
383 sector_t chunk;
384 raid0_conf_t *conf = mddev->private;
385 unsigned int chunk_sects = mddev->chunk_sectors;
386
387 if (is_power_of_2(chunk_sects)) {
388 int chunksect_bits = ffz(~chunk_sects);
389 /* find the sector offset inside the chunk */
390 sect_in_chunk = sector & (chunk_sects - 1);
391 sector >>= chunksect_bits;
392 /* chunk in zone */
393 chunk = *sector_offset;
394 /* quotient is the chunk in real device*/
395 sector_div(chunk, zone->nb_dev << chunksect_bits);
396 } else{
397 sect_in_chunk = sector_div(sector, chunk_sects);
398 chunk = *sector_offset;
399 sector_div(chunk, chunk_sects * zone->nb_dev);
400 }
401 /*
402 * position the bio over the real device
403 * real sector = chunk in device + starting of zone
404 * + the position in the chunk
405 */
406 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
407 return conf->devlist[(zone - conf->strip_zone)*mddev->raid_disks
408 + sector_div(sector, zone->nb_dev)];
409 }
410
411 /*
412 * Is io distribute over 1 or more chunks ?
413 */
414 static inline int is_io_in_chunk_boundary(mddev_t *mddev,
415 unsigned int chunk_sects, struct bio *bio)
416 {
417 if (likely(is_power_of_2(chunk_sects))) {
418 return chunk_sects >= ((bio->bi_sector & (chunk_sects-1))
419 + (bio->bi_size >> 9));
420 } else{
421 sector_t sector = bio->bi_sector;
422 return chunk_sects >= (sector_div(sector, chunk_sects)
423 + (bio->bi_size >> 9));
424 }
425 }
426
427 static int raid0_make_request(struct request_queue *q, struct bio *bio)
428 {
429 mddev_t *mddev = q->queuedata;
430 unsigned int chunk_sects;
431 sector_t sector_offset;
432 struct strip_zone *zone;
433 mdk_rdev_t *tmp_dev;
434 const int rw = bio_data_dir(bio);
435 int cpu;
436
437 if (unlikely(bio_barrier(bio))) {
438 bio_endio(bio, -EOPNOTSUPP);
439 return 0;
440 }
441
442 cpu = part_stat_lock();
443 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
444 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
445 bio_sectors(bio));
446 part_stat_unlock();
447
448 chunk_sects = mddev->chunk_sectors;
449 if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) {
450 sector_t sector = bio->bi_sector;
451 struct bio_pair *bp;
452 /* Sanity check -- queue functions should prevent this happening */
453 if (bio->bi_vcnt != 1 ||
454 bio->bi_idx != 0)
455 goto bad_map;
456 /* This is a one page bio that upper layers
457 * refuse to split for us, so we need to split it.
458 */
459 if (likely(is_power_of_2(chunk_sects)))
460 bp = bio_split(bio, chunk_sects - (sector &
461 (chunk_sects-1)));
462 else
463 bp = bio_split(bio, chunk_sects -
464 sector_div(sector, chunk_sects));
465 if (raid0_make_request(q, &bp->bio1))
466 generic_make_request(&bp->bio1);
467 if (raid0_make_request(q, &bp->bio2))
468 generic_make_request(&bp->bio2);
469
470 bio_pair_release(bp);
471 return 0;
472 }
473
474 sector_offset = bio->bi_sector;
475 zone = find_zone(mddev->private, &sector_offset);
476 tmp_dev = map_sector(mddev, zone, bio->bi_sector,
477 &sector_offset);
478 bio->bi_bdev = tmp_dev->bdev;
479 bio->bi_sector = sector_offset + zone->dev_start +
480 tmp_dev->data_offset;
481 /*
482 * Let the main block layer submit the IO and resolve recursion:
483 */
484 return 1;
485
486 bad_map:
487 printk("raid0_make_request bug: can't convert block across chunks"
488 " or bigger than %dk %llu %d\n", chunk_sects / 2,
489 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
490
491 bio_io_error(bio);
492 return 0;
493 }
494
495 static void raid0_status(struct seq_file *seq, mddev_t *mddev)
496 {
497 #undef MD_DEBUG
498 #ifdef MD_DEBUG
499 int j, k, h;
500 char b[BDEVNAME_SIZE];
501 raid0_conf_t *conf = mddev->private;
502
503 sector_t zone_size;
504 sector_t zone_start = 0;
505 h = 0;
506
507 for (j = 0; j < conf->nr_strip_zones; j++) {
508 seq_printf(seq, " z%d", j);
509 seq_printf(seq, "=[");
510 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
511 seq_printf(seq, "%s/", bdevname(
512 conf->devlist[j*mddev->raid_disks + k]
513 ->bdev, b));
514
515 zone_size = conf->strip_zone[j].zone_end - zone_start;
516 seq_printf(seq, "] ze=%lld ds=%lld s=%lld\n",
517 (unsigned long long)zone_start>>1,
518 (unsigned long long)conf->strip_zone[j].dev_start>>1,
519 (unsigned long long)zone_size>>1);
520 zone_start = conf->strip_zone[j].zone_end;
521 }
522 #endif
523 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
524 return;
525 }
526
527 static struct mdk_personality raid0_personality=
528 {
529 .name = "raid0",
530 .level = 0,
531 .owner = THIS_MODULE,
532 .make_request = raid0_make_request,
533 .run = raid0_run,
534 .stop = raid0_stop,
535 .status = raid0_status,
536 .size = raid0_size,
537 };
538
539 static int __init raid0_init (void)
540 {
541 return register_md_personality (&raid0_personality);
542 }
543
544 static void raid0_exit (void)
545 {
546 unregister_md_personality (&raid0_personality);
547 }
548
549 module_init(raid0_init);
550 module_exit(raid0_exit);
551 MODULE_LICENSE("GPL");
552 MODULE_ALIAS("md-personality-2"); /* RAID0 */
553 MODULE_ALIAS("md-raid0");
554 MODULE_ALIAS("md-level-0");
kernel-based virtual machine