]>
Commit | Line | Data |
---|---|---|
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 <linux/slab.h> | |
24 | #include "md.h" | |
25 | #include "raid0.h" | |
26 | #include "raid5.h" | |
27 | ||
28 | static int raid0_congested(void *data, int bits) | |
29 | { | |
30 | mddev_t *mddev = data; | |
31 | raid0_conf_t *conf = mddev->private; | |
32 | struct md_rdev **devlist = conf->devlist; | |
33 | int raid_disks = conf->strip_zone[0].nb_dev; | |
34 | int i, ret = 0; | |
35 | ||
36 | if (mddev_congested(mddev, bits)) | |
37 | return 1; | |
38 | ||
39 | for (i = 0; i < raid_disks && !ret ; i++) { | |
40 | struct request_queue *q = bdev_get_queue(devlist[i]->bdev); | |
41 | ||
42 | ret |= bdi_congested(&q->backing_dev_info, bits); | |
43 | } | |
44 | return ret; | |
45 | } | |
46 | ||
47 | /* | |
48 | * inform the user of the raid configuration | |
49 | */ | |
50 | static void dump_zones(mddev_t *mddev) | |
51 | { | |
52 | int j, k; | |
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 "md: RAID0 configuration for %s - %d zone%s\n", | |
59 | mdname(mddev), | |
60 | conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s"); | |
61 | for (j = 0; j < conf->nr_strip_zones; j++) { | |
62 | printk(KERN_INFO "md: zone%d=[", j); | |
63 | for (k = 0; k < conf->strip_zone[j].nb_dev; k++) | |
64 | printk(KERN_CONT "%s%s", k?"/":"", | |
65 | bdevname(conf->devlist[j*raid_disks | |
66 | + k]->bdev, b)); | |
67 | printk(KERN_CONT "]\n"); | |
68 | ||
69 | zone_size = conf->strip_zone[j].zone_end - zone_start; | |
70 | printk(KERN_INFO " zone-offset=%10lluKB, " | |
71 | "device-offset=%10lluKB, size=%10lluKB\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; | |
76 | } | |
77 | printk(KERN_INFO "\n"); | |
78 | } | |
79 | ||
80 | static int create_strip_zones(mddev_t *mddev, raid0_conf_t **private_conf) | |
81 | { | |
82 | int i, c, err; | |
83 | sector_t curr_zone_end, sectors; | |
84 | struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev; | |
85 | struct strip_zone *zone; | |
86 | int cnt; | |
87 | char b[BDEVNAME_SIZE]; | |
88 | char b2[BDEVNAME_SIZE]; | |
89 | raid0_conf_t *conf = kzalloc(sizeof(*conf), GFP_KERNEL); | |
90 | ||
91 | if (!conf) | |
92 | return -ENOMEM; | |
93 | list_for_each_entry(rdev1, &mddev->disks, same_set) { | |
94 | pr_debug("md/raid0:%s: looking at %s\n", | |
95 | mdname(mddev), | |
96 | bdevname(rdev1->bdev, b)); | |
97 | c = 0; | |
98 | ||
99 | /* round size to chunk_size */ | |
100 | sectors = rdev1->sectors; | |
101 | sector_div(sectors, mddev->chunk_sectors); | |
102 | rdev1->sectors = sectors * mddev->chunk_sectors; | |
103 | ||
104 | list_for_each_entry(rdev2, &mddev->disks, same_set) { | |
105 | pr_debug("md/raid0:%s: comparing %s(%llu)" | |
106 | " with %s(%llu)\n", | |
107 | mdname(mddev), | |
108 | bdevname(rdev1->bdev,b), | |
109 | (unsigned long long)rdev1->sectors, | |
110 | bdevname(rdev2->bdev,b2), | |
111 | (unsigned long long)rdev2->sectors); | |
112 | if (rdev2 == rdev1) { | |
113 | pr_debug("md/raid0:%s: END\n", | |
114 | mdname(mddev)); | |
115 | break; | |
116 | } | |
117 | if (rdev2->sectors == rdev1->sectors) { | |
118 | /* | |
119 | * Not unique, don't count it as a new | |
120 | * group | |
121 | */ | |
122 | pr_debug("md/raid0:%s: EQUAL\n", | |
123 | mdname(mddev)); | |
124 | c = 1; | |
125 | break; | |
126 | } | |
127 | pr_debug("md/raid0:%s: NOT EQUAL\n", | |
128 | mdname(mddev)); | |
129 | } | |
130 | if (!c) { | |
131 | pr_debug("md/raid0:%s: ==> UNIQUE\n", | |
132 | mdname(mddev)); | |
133 | conf->nr_strip_zones++; | |
134 | pr_debug("md/raid0:%s: %d zones\n", | |
135 | mdname(mddev), conf->nr_strip_zones); | |
136 | } | |
137 | } | |
138 | pr_debug("md/raid0:%s: FINAL %d zones\n", | |
139 | mdname(mddev), conf->nr_strip_zones); | |
140 | err = -ENOMEM; | |
141 | conf->strip_zone = kzalloc(sizeof(struct strip_zone)* | |
142 | conf->nr_strip_zones, GFP_KERNEL); | |
143 | if (!conf->strip_zone) | |
144 | goto abort; | |
145 | conf->devlist = kzalloc(sizeof(struct md_rdev*)* | |
146 | conf->nr_strip_zones*mddev->raid_disks, | |
147 | GFP_KERNEL); | |
148 | if (!conf->devlist) | |
149 | goto abort; | |
150 | ||
151 | /* The first zone must contain all devices, so here we check that | |
152 | * there is a proper alignment of slots to devices and find them all | |
153 | */ | |
154 | zone = &conf->strip_zone[0]; | |
155 | cnt = 0; | |
156 | smallest = NULL; | |
157 | dev = conf->devlist; | |
158 | err = -EINVAL; | |
159 | list_for_each_entry(rdev1, &mddev->disks, same_set) { | |
160 | int j = rdev1->raid_disk; | |
161 | ||
162 | if (mddev->level == 10) { | |
163 | /* taking over a raid10-n2 array */ | |
164 | j /= 2; | |
165 | rdev1->new_raid_disk = j; | |
166 | } | |
167 | ||
168 | if (mddev->level == 1) { | |
169 | /* taiking over a raid1 array- | |
170 | * we have only one active disk | |
171 | */ | |
172 | j = 0; | |
173 | rdev1->new_raid_disk = j; | |
174 | } | |
175 | ||
176 | if (j < 0 || j >= mddev->raid_disks) { | |
177 | printk(KERN_ERR "md/raid0:%s: bad disk number %d - " | |
178 | "aborting!\n", mdname(mddev), j); | |
179 | goto abort; | |
180 | } | |
181 | if (dev[j]) { | |
182 | printk(KERN_ERR "md/raid0:%s: multiple devices for %d - " | |
183 | "aborting!\n", mdname(mddev), j); | |
184 | goto abort; | |
185 | } | |
186 | dev[j] = rdev1; | |
187 | ||
188 | disk_stack_limits(mddev->gendisk, rdev1->bdev, | |
189 | rdev1->data_offset << 9); | |
190 | /* as we don't honour merge_bvec_fn, we must never risk | |
191 | * violating it, so limit ->max_segments to 1, lying within | |
192 | * a single page. | |
193 | */ | |
194 | ||
195 | if (rdev1->bdev->bd_disk->queue->merge_bvec_fn) { | |
196 | blk_queue_max_segments(mddev->queue, 1); | |
197 | blk_queue_segment_boundary(mddev->queue, | |
198 | PAGE_CACHE_SIZE - 1); | |
199 | } | |
200 | if (!smallest || (rdev1->sectors < smallest->sectors)) | |
201 | smallest = rdev1; | |
202 | cnt++; | |
203 | } | |
204 | if (cnt != mddev->raid_disks) { | |
205 | printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - " | |
206 | "aborting!\n", mdname(mddev), cnt, mddev->raid_disks); | |
207 | goto abort; | |
208 | } | |
209 | zone->nb_dev = cnt; | |
210 | zone->zone_end = smallest->sectors * cnt; | |
211 | ||
212 | curr_zone_end = zone->zone_end; | |
213 | ||
214 | /* now do the other zones */ | |
215 | for (i = 1; i < conf->nr_strip_zones; i++) | |
216 | { | |
217 | int j; | |
218 | ||
219 | zone = conf->strip_zone + i; | |
220 | dev = conf->devlist + i * mddev->raid_disks; | |
221 | ||
222 | pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i); | |
223 | zone->dev_start = smallest->sectors; | |
224 | smallest = NULL; | |
225 | c = 0; | |
226 | ||
227 | for (j=0; j<cnt; j++) { | |
228 | rdev = conf->devlist[j]; | |
229 | if (rdev->sectors <= zone->dev_start) { | |
230 | pr_debug("md/raid0:%s: checking %s ... nope\n", | |
231 | mdname(mddev), | |
232 | bdevname(rdev->bdev, b)); | |
233 | continue; | |
234 | } | |
235 | pr_debug("md/raid0:%s: checking %s ..." | |
236 | " contained as device %d\n", | |
237 | mdname(mddev), | |
238 | bdevname(rdev->bdev, b), c); | |
239 | dev[c] = rdev; | |
240 | c++; | |
241 | if (!smallest || rdev->sectors < smallest->sectors) { | |
242 | smallest = rdev; | |
243 | pr_debug("md/raid0:%s: (%llu) is smallest!.\n", | |
244 | mdname(mddev), | |
245 | (unsigned long long)rdev->sectors); | |
246 | } | |
247 | } | |
248 | ||
249 | zone->nb_dev = c; | |
250 | sectors = (smallest->sectors - zone->dev_start) * c; | |
251 | pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n", | |
252 | mdname(mddev), | |
253 | zone->nb_dev, (unsigned long long)sectors); | |
254 | ||
255 | curr_zone_end += sectors; | |
256 | zone->zone_end = curr_zone_end; | |
257 | ||
258 | pr_debug("md/raid0:%s: current zone start: %llu\n", | |
259 | mdname(mddev), | |
260 | (unsigned long long)smallest->sectors); | |
261 | } | |
262 | mddev->queue->backing_dev_info.congested_fn = raid0_congested; | |
263 | mddev->queue->backing_dev_info.congested_data = mddev; | |
264 | ||
265 | /* | |
266 | * now since we have the hard sector sizes, we can make sure | |
267 | * chunk size is a multiple of that sector size | |
268 | */ | |
269 | if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) { | |
270 | printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n", | |
271 | mdname(mddev), | |
272 | mddev->chunk_sectors << 9); | |
273 | goto abort; | |
274 | } | |
275 | ||
276 | blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9); | |
277 | blk_queue_io_opt(mddev->queue, | |
278 | (mddev->chunk_sectors << 9) * mddev->raid_disks); | |
279 | ||
280 | pr_debug("md/raid0:%s: done.\n", mdname(mddev)); | |
281 | *private_conf = conf; | |
282 | ||
283 | return 0; | |
284 | abort: | |
285 | kfree(conf->strip_zone); | |
286 | kfree(conf->devlist); | |
287 | kfree(conf); | |
288 | *private_conf = NULL; | |
289 | return err; | |
290 | } | |
291 | ||
292 | /** | |
293 | * raid0_mergeable_bvec -- tell bio layer if a two requests can be merged | |
294 | * @q: request queue | |
295 | * @bvm: properties of new bio | |
296 | * @biovec: the request that could be merged to it. | |
297 | * | |
298 | * Return amount of bytes we can accept at this offset | |
299 | */ | |
300 | static int raid0_mergeable_bvec(struct request_queue *q, | |
301 | struct bvec_merge_data *bvm, | |
302 | struct bio_vec *biovec) | |
303 | { | |
304 | mddev_t *mddev = q->queuedata; | |
305 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); | |
306 | int max; | |
307 | unsigned int chunk_sectors = mddev->chunk_sectors; | |
308 | unsigned int bio_sectors = bvm->bi_size >> 9; | |
309 | ||
310 | if (is_power_of_2(chunk_sectors)) | |
311 | max = (chunk_sectors - ((sector & (chunk_sectors-1)) | |
312 | + bio_sectors)) << 9; | |
313 | else | |
314 | max = (chunk_sectors - (sector_div(sector, chunk_sectors) | |
315 | + bio_sectors)) << 9; | |
316 | if (max < 0) max = 0; /* bio_add cannot handle a negative return */ | |
317 | if (max <= biovec->bv_len && bio_sectors == 0) | |
318 | return biovec->bv_len; | |
319 | else | |
320 | return max; | |
321 | } | |
322 | ||
323 | static sector_t raid0_size(mddev_t *mddev, sector_t sectors, int raid_disks) | |
324 | { | |
325 | sector_t array_sectors = 0; | |
326 | struct md_rdev *rdev; | |
327 | ||
328 | WARN_ONCE(sectors || raid_disks, | |
329 | "%s does not support generic reshape\n", __func__); | |
330 | ||
331 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
332 | array_sectors += rdev->sectors; | |
333 | ||
334 | return array_sectors; | |
335 | } | |
336 | ||
337 | static int raid0_run(mddev_t *mddev) | |
338 | { | |
339 | raid0_conf_t *conf; | |
340 | int ret; | |
341 | ||
342 | if (mddev->chunk_sectors == 0) { | |
343 | printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n", | |
344 | mdname(mddev)); | |
345 | return -EINVAL; | |
346 | } | |
347 | if (md_check_no_bitmap(mddev)) | |
348 | return -EINVAL; | |
349 | blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors); | |
350 | ||
351 | /* if private is not null, we are here after takeover */ | |
352 | if (mddev->private == NULL) { | |
353 | ret = create_strip_zones(mddev, &conf); | |
354 | if (ret < 0) | |
355 | return ret; | |
356 | mddev->private = conf; | |
357 | } | |
358 | conf = mddev->private; | |
359 | ||
360 | /* calculate array device size */ | |
361 | md_set_array_sectors(mddev, raid0_size(mddev, 0, 0)); | |
362 | ||
363 | printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n", | |
364 | mdname(mddev), | |
365 | (unsigned long long)mddev->array_sectors); | |
366 | /* calculate the max read-ahead size. | |
367 | * For read-ahead of large files to be effective, we need to | |
368 | * readahead at least twice a whole stripe. i.e. number of devices | |
369 | * multiplied by chunk size times 2. | |
370 | * If an individual device has an ra_pages greater than the | |
371 | * chunk size, then we will not drive that device as hard as it | |
372 | * wants. We consider this a configuration error: a larger | |
373 | * chunksize should be used in that case. | |
374 | */ | |
375 | { | |
376 | int stripe = mddev->raid_disks * | |
377 | (mddev->chunk_sectors << 9) / PAGE_SIZE; | |
378 | if (mddev->queue->backing_dev_info.ra_pages < 2* stripe) | |
379 | mddev->queue->backing_dev_info.ra_pages = 2* stripe; | |
380 | } | |
381 | ||
382 | blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec); | |
383 | dump_zones(mddev); | |
384 | return md_integrity_register(mddev); | |
385 | } | |
386 | ||
387 | static int raid0_stop(mddev_t *mddev) | |
388 | { | |
389 | raid0_conf_t *conf = mddev->private; | |
390 | ||
391 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
392 | kfree(conf->strip_zone); | |
393 | kfree(conf->devlist); | |
394 | kfree(conf); | |
395 | mddev->private = NULL; | |
396 | return 0; | |
397 | } | |
398 | ||
399 | /* Find the zone which holds a particular offset | |
400 | * Update *sectorp to be an offset in that zone | |
401 | */ | |
402 | static struct strip_zone *find_zone(struct raid0_private_data *conf, | |
403 | sector_t *sectorp) | |
404 | { | |
405 | int i; | |
406 | struct strip_zone *z = conf->strip_zone; | |
407 | sector_t sector = *sectorp; | |
408 | ||
409 | for (i = 0; i < conf->nr_strip_zones; i++) | |
410 | if (sector < z[i].zone_end) { | |
411 | if (i) | |
412 | *sectorp = sector - z[i-1].zone_end; | |
413 | return z + i; | |
414 | } | |
415 | BUG(); | |
416 | } | |
417 | ||
418 | /* | |
419 | * remaps the bio to the target device. we separate two flows. | |
420 | * power 2 flow and a general flow for the sake of perfromance | |
421 | */ | |
422 | static struct md_rdev *map_sector(mddev_t *mddev, struct strip_zone *zone, | |
423 | sector_t sector, sector_t *sector_offset) | |
424 | { | |
425 | unsigned int sect_in_chunk; | |
426 | sector_t chunk; | |
427 | raid0_conf_t *conf = mddev->private; | |
428 | int raid_disks = conf->strip_zone[0].nb_dev; | |
429 | unsigned int chunk_sects = mddev->chunk_sectors; | |
430 | ||
431 | if (is_power_of_2(chunk_sects)) { | |
432 | int chunksect_bits = ffz(~chunk_sects); | |
433 | /* find the sector offset inside the chunk */ | |
434 | sect_in_chunk = sector & (chunk_sects - 1); | |
435 | sector >>= chunksect_bits; | |
436 | /* chunk in zone */ | |
437 | chunk = *sector_offset; | |
438 | /* quotient is the chunk in real device*/ | |
439 | sector_div(chunk, zone->nb_dev << chunksect_bits); | |
440 | } else{ | |
441 | sect_in_chunk = sector_div(sector, chunk_sects); | |
442 | chunk = *sector_offset; | |
443 | sector_div(chunk, chunk_sects * zone->nb_dev); | |
444 | } | |
445 | /* | |
446 | * position the bio over the real device | |
447 | * real sector = chunk in device + starting of zone | |
448 | * + the position in the chunk | |
449 | */ | |
450 | *sector_offset = (chunk * chunk_sects) + sect_in_chunk; | |
451 | return conf->devlist[(zone - conf->strip_zone)*raid_disks | |
452 | + sector_div(sector, zone->nb_dev)]; | |
453 | } | |
454 | ||
455 | /* | |
456 | * Is io distribute over 1 or more chunks ? | |
457 | */ | |
458 | static inline int is_io_in_chunk_boundary(mddev_t *mddev, | |
459 | unsigned int chunk_sects, struct bio *bio) | |
460 | { | |
461 | if (likely(is_power_of_2(chunk_sects))) { | |
462 | return chunk_sects >= ((bio->bi_sector & (chunk_sects-1)) | |
463 | + (bio->bi_size >> 9)); | |
464 | } else{ | |
465 | sector_t sector = bio->bi_sector; | |
466 | return chunk_sects >= (sector_div(sector, chunk_sects) | |
467 | + (bio->bi_size >> 9)); | |
468 | } | |
469 | } | |
470 | ||
471 | static int raid0_make_request(mddev_t *mddev, struct bio *bio) | |
472 | { | |
473 | unsigned int chunk_sects; | |
474 | sector_t sector_offset; | |
475 | struct strip_zone *zone; | |
476 | struct md_rdev *tmp_dev; | |
477 | ||
478 | if (unlikely(bio->bi_rw & REQ_FLUSH)) { | |
479 | md_flush_request(mddev, bio); | |
480 | return 0; | |
481 | } | |
482 | ||
483 | chunk_sects = mddev->chunk_sectors; | |
484 | if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) { | |
485 | sector_t sector = bio->bi_sector; | |
486 | struct bio_pair *bp; | |
487 | /* Sanity check -- queue functions should prevent this happening */ | |
488 | if (bio->bi_vcnt != 1 || | |
489 | bio->bi_idx != 0) | |
490 | goto bad_map; | |
491 | /* This is a one page bio that upper layers | |
492 | * refuse to split for us, so we need to split it. | |
493 | */ | |
494 | if (likely(is_power_of_2(chunk_sects))) | |
495 | bp = bio_split(bio, chunk_sects - (sector & | |
496 | (chunk_sects-1))); | |
497 | else | |
498 | bp = bio_split(bio, chunk_sects - | |
499 | sector_div(sector, chunk_sects)); | |
500 | if (raid0_make_request(mddev, &bp->bio1)) | |
501 | generic_make_request(&bp->bio1); | |
502 | if (raid0_make_request(mddev, &bp->bio2)) | |
503 | generic_make_request(&bp->bio2); | |
504 | ||
505 | bio_pair_release(bp); | |
506 | return 0; | |
507 | } | |
508 | ||
509 | sector_offset = bio->bi_sector; | |
510 | zone = find_zone(mddev->private, §or_offset); | |
511 | tmp_dev = map_sector(mddev, zone, bio->bi_sector, | |
512 | §or_offset); | |
513 | bio->bi_bdev = tmp_dev->bdev; | |
514 | bio->bi_sector = sector_offset + zone->dev_start + | |
515 | tmp_dev->data_offset; | |
516 | /* | |
517 | * Let the main block layer submit the IO and resolve recursion: | |
518 | */ | |
519 | return 1; | |
520 | ||
521 | bad_map: | |
522 | printk("md/raid0:%s: make_request bug: can't convert block across chunks" | |
523 | " or bigger than %dk %llu %d\n", | |
524 | mdname(mddev), chunk_sects / 2, | |
525 | (unsigned long long)bio->bi_sector, bio->bi_size >> 10); | |
526 | ||
527 | bio_io_error(bio); | |
528 | return 0; | |
529 | } | |
530 | ||
531 | static void raid0_status(struct seq_file *seq, mddev_t *mddev) | |
532 | { | |
533 | seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2); | |
534 | return; | |
535 | } | |
536 | ||
537 | static void *raid0_takeover_raid45(mddev_t *mddev) | |
538 | { | |
539 | struct md_rdev *rdev; | |
540 | raid0_conf_t *priv_conf; | |
541 | ||
542 | if (mddev->degraded != 1) { | |
543 | printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n", | |
544 | mdname(mddev), | |
545 | mddev->degraded); | |
546 | return ERR_PTR(-EINVAL); | |
547 | } | |
548 | ||
549 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
550 | /* check slot number for a disk */ | |
551 | if (rdev->raid_disk == mddev->raid_disks-1) { | |
552 | printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n", | |
553 | mdname(mddev)); | |
554 | return ERR_PTR(-EINVAL); | |
555 | } | |
556 | } | |
557 | ||
558 | /* Set new parameters */ | |
559 | mddev->new_level = 0; | |
560 | mddev->new_layout = 0; | |
561 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
562 | mddev->raid_disks--; | |
563 | mddev->delta_disks = -1; | |
564 | /* make sure it will be not marked as dirty */ | |
565 | mddev->recovery_cp = MaxSector; | |
566 | ||
567 | create_strip_zones(mddev, &priv_conf); | |
568 | return priv_conf; | |
569 | } | |
570 | ||
571 | static void *raid0_takeover_raid10(mddev_t *mddev) | |
572 | { | |
573 | raid0_conf_t *priv_conf; | |
574 | ||
575 | /* Check layout: | |
576 | * - far_copies must be 1 | |
577 | * - near_copies must be 2 | |
578 | * - disks number must be even | |
579 | * - all mirrors must be already degraded | |
580 | */ | |
581 | if (mddev->layout != ((1 << 8) + 2)) { | |
582 | printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n", | |
583 | mdname(mddev), | |
584 | mddev->layout); | |
585 | return ERR_PTR(-EINVAL); | |
586 | } | |
587 | if (mddev->raid_disks & 1) { | |
588 | printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n", | |
589 | mdname(mddev)); | |
590 | return ERR_PTR(-EINVAL); | |
591 | } | |
592 | if (mddev->degraded != (mddev->raid_disks>>1)) { | |
593 | printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n", | |
594 | mdname(mddev)); | |
595 | return ERR_PTR(-EINVAL); | |
596 | } | |
597 | ||
598 | /* Set new parameters */ | |
599 | mddev->new_level = 0; | |
600 | mddev->new_layout = 0; | |
601 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
602 | mddev->delta_disks = - mddev->raid_disks / 2; | |
603 | mddev->raid_disks += mddev->delta_disks; | |
604 | mddev->degraded = 0; | |
605 | /* make sure it will be not marked as dirty */ | |
606 | mddev->recovery_cp = MaxSector; | |
607 | ||
608 | create_strip_zones(mddev, &priv_conf); | |
609 | return priv_conf; | |
610 | } | |
611 | ||
612 | static void *raid0_takeover_raid1(mddev_t *mddev) | |
613 | { | |
614 | raid0_conf_t *priv_conf; | |
615 | ||
616 | /* Check layout: | |
617 | * - (N - 1) mirror drives must be already faulty | |
618 | */ | |
619 | if ((mddev->raid_disks - 1) != mddev->degraded) { | |
620 | printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n", | |
621 | mdname(mddev)); | |
622 | return ERR_PTR(-EINVAL); | |
623 | } | |
624 | ||
625 | /* Set new parameters */ | |
626 | mddev->new_level = 0; | |
627 | mddev->new_layout = 0; | |
628 | mddev->new_chunk_sectors = 128; /* by default set chunk size to 64k */ | |
629 | mddev->delta_disks = 1 - mddev->raid_disks; | |
630 | mddev->raid_disks = 1; | |
631 | /* make sure it will be not marked as dirty */ | |
632 | mddev->recovery_cp = MaxSector; | |
633 | ||
634 | create_strip_zones(mddev, &priv_conf); | |
635 | return priv_conf; | |
636 | } | |
637 | ||
638 | static void *raid0_takeover(mddev_t *mddev) | |
639 | { | |
640 | /* raid0 can take over: | |
641 | * raid4 - if all data disks are active. | |
642 | * raid5 - providing it is Raid4 layout and one disk is faulty | |
643 | * raid10 - assuming we have all necessary active disks | |
644 | * raid1 - with (N -1) mirror drives faulty | |
645 | */ | |
646 | if (mddev->level == 4) | |
647 | return raid0_takeover_raid45(mddev); | |
648 | ||
649 | if (mddev->level == 5) { | |
650 | if (mddev->layout == ALGORITHM_PARITY_N) | |
651 | return raid0_takeover_raid45(mddev); | |
652 | ||
653 | printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n", | |
654 | mdname(mddev), ALGORITHM_PARITY_N); | |
655 | } | |
656 | ||
657 | if (mddev->level == 10) | |
658 | return raid0_takeover_raid10(mddev); | |
659 | ||
660 | if (mddev->level == 1) | |
661 | return raid0_takeover_raid1(mddev); | |
662 | ||
663 | printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n", | |
664 | mddev->level); | |
665 | ||
666 | return ERR_PTR(-EINVAL); | |
667 | } | |
668 | ||
669 | static void raid0_quiesce(mddev_t *mddev, int state) | |
670 | { | |
671 | } | |
672 | ||
673 | static struct mdk_personality raid0_personality= | |
674 | { | |
675 | .name = "raid0", | |
676 | .level = 0, | |
677 | .owner = THIS_MODULE, | |
678 | .make_request = raid0_make_request, | |
679 | .run = raid0_run, | |
680 | .stop = raid0_stop, | |
681 | .status = raid0_status, | |
682 | .size = raid0_size, | |
683 | .takeover = raid0_takeover, | |
684 | .quiesce = raid0_quiesce, | |
685 | }; | |
686 | ||
687 | static int __init raid0_init (void) | |
688 | { | |
689 | return register_md_personality (&raid0_personality); | |
690 | } | |
691 | ||
692 | static void raid0_exit (void) | |
693 | { | |
694 | unregister_md_personality (&raid0_personality); | |
695 | } | |
696 | ||
697 | module_init(raid0_init); | |
698 | module_exit(raid0_exit); | |
699 | MODULE_LICENSE("GPL"); | |
700 | MODULE_DESCRIPTION("RAID0 (striping) personality for MD"); | |
701 | MODULE_ALIAS("md-personality-2"); /* RAID0 */ | |
702 | MODULE_ALIAS("md-raid0"); | |
703 | MODULE_ALIAS("md-level-0"); |