Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | ||
bff61975 | 21 | #include <linux/blkdev.h> |
bff61975 | 22 | #include <linux/seq_file.h> |
5a0e3ad6 | 23 | #include <linux/slab.h> |
43b2e5d8 | 24 | #include "md.h" |
ef740c37 | 25 | #include "raid0.h" |
9af204cf | 26 | #include "raid5.h" |
1da177e4 | 27 | |
165125e1 | 28 | static void raid0_unplug(struct request_queue *q) |
1da177e4 LT |
29 | { |
30 | mddev_t *mddev = q->queuedata; | |
070ec55d | 31 | raid0_conf_t *conf = mddev->private; |
b414579f | 32 | mdk_rdev_t **devlist = conf->devlist; |
84707f38 | 33 | int raid_disks = conf->strip_zone[0].nb_dev; |
1da177e4 LT |
34 | int i; |
35 | ||
84707f38 | 36 | for (i=0; i < raid_disks; i++) { |
165125e1 | 37 | struct request_queue *r_queue = bdev_get_queue(devlist[i]->bdev); |
1da177e4 | 38 | |
2ad8b1ef | 39 | blk_unplug(r_queue); |
1da177e4 LT |
40 | } |
41 | } | |
42 | ||
26be34dc N |
43 | static int raid0_congested(void *data, int bits) |
44 | { | |
45 | mddev_t *mddev = data; | |
070ec55d | 46 | raid0_conf_t *conf = mddev->private; |
b414579f | 47 | mdk_rdev_t **devlist = conf->devlist; |
84707f38 | 48 | int raid_disks = conf->strip_zone[0].nb_dev; |
26be34dc N |
49 | int i, ret = 0; |
50 | ||
3fa841d7 N |
51 | if (mddev_congested(mddev, bits)) |
52 | return 1; | |
53 | ||
84707f38 | 54 | for (i = 0; i < raid_disks && !ret ; i++) { |
165125e1 | 55 | struct request_queue *q = bdev_get_queue(devlist[i]->bdev); |
26be34dc N |
56 | |
57 | ret |= bdi_congested(&q->backing_dev_info, bits); | |
58 | } | |
59 | return ret; | |
60 | } | |
61 | ||
46994191 | 62 | /* |
63 | * inform the user of the raid configuration | |
64 | */ | |
65 | static void dump_zones(mddev_t *mddev) | |
66 | { | |
67 | int j, k, h; | |
68 | sector_t zone_size = 0; | |
69 | sector_t zone_start = 0; | |
70 | char b[BDEVNAME_SIZE]; | |
71 | raid0_conf_t *conf = mddev->private; | |
84707f38 | 72 | int raid_disks = conf->strip_zone[0].nb_dev; |
46994191 | 73 | printk(KERN_INFO "******* %s configuration *********\n", |
74 | mdname(mddev)); | |
75 | h = 0; | |
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++) | |
b5a20961 | 79 | printk(KERN_CONT "%s/", |
84707f38 | 80 | bdevname(conf->devlist[j*raid_disks |
46994191 | 81 | + k]->bdev, b)); |
b5a20961 | 82 | printk(KERN_CONT "]\n"); |
46994191 | 83 | |
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; | |
91 | } | |
92 | printk(KERN_INFO "**********************************\n\n"); | |
93 | } | |
94 | ||
9af204cf | 95 | static int create_strip_zones(mddev_t *mddev, raid0_conf_t **private_conf) |
1da177e4 | 96 | { |
a9f326eb | 97 | int i, c, err; |
49f357a2 | 98 | sector_t curr_zone_end, sectors; |
b414579f | 99 | mdk_rdev_t *smallest, *rdev1, *rdev2, *rdev, **dev; |
1da177e4 LT |
100 | struct strip_zone *zone; |
101 | int cnt; | |
102 | char b[BDEVNAME_SIZE]; | |
ed7b0038 AN |
103 | raid0_conf_t *conf = kzalloc(sizeof(*conf), GFP_KERNEL); |
104 | ||
105 | if (!conf) | |
106 | return -ENOMEM; | |
159ec1fc | 107 | list_for_each_entry(rdev1, &mddev->disks, same_set) { |
b5a20961 N |
108 | printk(KERN_INFO "md/raid0:%s: looking at %s\n", |
109 | mdname(mddev), | |
110 | bdevname(rdev1->bdev, b)); | |
1da177e4 | 111 | c = 0; |
13f2682b N |
112 | |
113 | /* round size to chunk_size */ | |
114 | sectors = rdev1->sectors; | |
115 | sector_div(sectors, mddev->chunk_sectors); | |
116 | rdev1->sectors = sectors * mddev->chunk_sectors; | |
117 | ||
159ec1fc | 118 | list_for_each_entry(rdev2, &mddev->disks, same_set) { |
b5a20961 N |
119 | printk(KERN_INFO "md/raid0:%s: comparing %s(%llu)", |
120 | mdname(mddev), | |
1da177e4 | 121 | bdevname(rdev1->bdev,b), |
dd8ac336 | 122 | (unsigned long long)rdev1->sectors); |
b5a20961 | 123 | printk(KERN_CONT " with %s(%llu)\n", |
1da177e4 | 124 | bdevname(rdev2->bdev,b), |
dd8ac336 | 125 | (unsigned long long)rdev2->sectors); |
1da177e4 | 126 | if (rdev2 == rdev1) { |
b5a20961 N |
127 | printk(KERN_INFO "md/raid0:%s: END\n", |
128 | mdname(mddev)); | |
1da177e4 LT |
129 | break; |
130 | } | |
dd8ac336 | 131 | if (rdev2->sectors == rdev1->sectors) { |
1da177e4 LT |
132 | /* |
133 | * Not unique, don't count it as a new | |
134 | * group | |
135 | */ | |
b5a20961 N |
136 | printk(KERN_INFO "md/raid0:%s: EQUAL\n", |
137 | mdname(mddev)); | |
1da177e4 LT |
138 | c = 1; |
139 | break; | |
140 | } | |
b5a20961 N |
141 | printk(KERN_INFO "md/raid0:%s: NOT EQUAL\n", |
142 | mdname(mddev)); | |
1da177e4 LT |
143 | } |
144 | if (!c) { | |
b5a20961 N |
145 | printk(KERN_INFO "md/raid0:%s: ==> UNIQUE\n", |
146 | mdname(mddev)); | |
1da177e4 | 147 | conf->nr_strip_zones++; |
b5a20961 N |
148 | printk(KERN_INFO "md/raid0:%s: %d zones\n", |
149 | mdname(mddev), conf->nr_strip_zones); | |
1da177e4 LT |
150 | } |
151 | } | |
b5a20961 N |
152 | printk(KERN_INFO "md/raid0:%s: FINAL %d zones\n", |
153 | mdname(mddev), conf->nr_strip_zones); | |
ed7b0038 | 154 | err = -ENOMEM; |
9ffae0cf | 155 | conf->strip_zone = kzalloc(sizeof(struct strip_zone)* |
1da177e4 LT |
156 | conf->nr_strip_zones, GFP_KERNEL); |
157 | if (!conf->strip_zone) | |
ed7b0038 | 158 | goto abort; |
9ffae0cf | 159 | conf->devlist = kzalloc(sizeof(mdk_rdev_t*)* |
1da177e4 LT |
160 | conf->nr_strip_zones*mddev->raid_disks, |
161 | GFP_KERNEL); | |
162 | if (!conf->devlist) | |
ed7b0038 | 163 | goto abort; |
1da177e4 | 164 | |
1da177e4 LT |
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 | |
167 | */ | |
168 | zone = &conf->strip_zone[0]; | |
169 | cnt = 0; | |
170 | smallest = NULL; | |
b414579f | 171 | dev = conf->devlist; |
ed7b0038 | 172 | err = -EINVAL; |
159ec1fc | 173 | list_for_each_entry(rdev1, &mddev->disks, same_set) { |
1da177e4 LT |
174 | int j = rdev1->raid_disk; |
175 | ||
e93f68a1 | 176 | if (mddev->level == 10) { |
9af204cf TM |
177 | /* taking over a raid10-n2 array */ |
178 | j /= 2; | |
e93f68a1 N |
179 | rdev1->new_raid_disk = j; |
180 | } | |
9af204cf | 181 | |
1da177e4 | 182 | if (j < 0 || j >= mddev->raid_disks) { |
b5a20961 N |
183 | printk(KERN_ERR "md/raid0:%s: bad disk number %d - " |
184 | "aborting!\n", mdname(mddev), j); | |
1da177e4 LT |
185 | goto abort; |
186 | } | |
b414579f | 187 | if (dev[j]) { |
b5a20961 N |
188 | printk(KERN_ERR "md/raid0:%s: multiple devices for %d - " |
189 | "aborting!\n", mdname(mddev), j); | |
1da177e4 LT |
190 | goto abort; |
191 | } | |
b414579f | 192 | dev[j] = rdev1; |
1da177e4 | 193 | |
8f6c2e4b MP |
194 | disk_stack_limits(mddev->gendisk, rdev1->bdev, |
195 | rdev1->data_offset << 9); | |
1da177e4 | 196 | /* as we don't honour merge_bvec_fn, we must never risk |
627a2d3c N |
197 | * violating it, so limit ->max_segments to 1, lying within |
198 | * a single page. | |
1da177e4 LT |
199 | */ |
200 | ||
627a2d3c N |
201 | if (rdev1->bdev->bd_disk->queue->merge_bvec_fn) { |
202 | blk_queue_max_segments(mddev->queue, 1); | |
203 | blk_queue_segment_boundary(mddev->queue, | |
204 | PAGE_CACHE_SIZE - 1); | |
205 | } | |
dd8ac336 | 206 | if (!smallest || (rdev1->sectors < smallest->sectors)) |
1da177e4 LT |
207 | smallest = rdev1; |
208 | cnt++; | |
209 | } | |
210 | if (cnt != mddev->raid_disks) { | |
b5a20961 N |
211 | printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - " |
212 | "aborting!\n", mdname(mddev), cnt, mddev->raid_disks); | |
1da177e4 LT |
213 | goto abort; |
214 | } | |
215 | zone->nb_dev = cnt; | |
49f357a2 | 216 | zone->zone_end = smallest->sectors * cnt; |
1da177e4 | 217 | |
49f357a2 | 218 | curr_zone_end = zone->zone_end; |
1da177e4 LT |
219 | |
220 | /* now do the other zones */ | |
221 | for (i = 1; i < conf->nr_strip_zones; i++) | |
222 | { | |
a9f326eb N |
223 | int j; |
224 | ||
1da177e4 | 225 | zone = conf->strip_zone + i; |
b414579f | 226 | dev = conf->devlist + i * mddev->raid_disks; |
1da177e4 | 227 | |
b5a20961 N |
228 | printk(KERN_INFO "md/raid0:%s: zone %d\n", |
229 | mdname(mddev), i); | |
d27a43ab | 230 | zone->dev_start = smallest->sectors; |
1da177e4 LT |
231 | smallest = NULL; |
232 | c = 0; | |
233 | ||
234 | for (j=0; j<cnt; j++) { | |
b414579f | 235 | rdev = conf->devlist[j]; |
b5a20961 N |
236 | printk(KERN_INFO "md/raid0:%s: checking %s ...", |
237 | mdname(mddev), | |
238 | bdevname(rdev->bdev, b)); | |
d27a43ab | 239 | if (rdev->sectors <= zone->dev_start) { |
b5a20961 | 240 | printk(KERN_CONT " nope.\n"); |
dd8ac336 AN |
241 | continue; |
242 | } | |
b5a20961 | 243 | printk(KERN_CONT " contained as device %d\n", c); |
b414579f | 244 | dev[c] = rdev; |
dd8ac336 AN |
245 | c++; |
246 | if (!smallest || rdev->sectors < smallest->sectors) { | |
247 | smallest = rdev; | |
b5a20961 N |
248 | printk(KERN_INFO "md/raid0:%s: (%llu) is smallest!.\n", |
249 | mdname(mddev), | |
250 | (unsigned long long)rdev->sectors); | |
dd8ac336 | 251 | } |
1da177e4 LT |
252 | } |
253 | ||
254 | zone->nb_dev = c; | |
49f357a2 | 255 | sectors = (smallest->sectors - zone->dev_start) * c; |
b5a20961 N |
256 | printk(KERN_INFO "md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n", |
257 | mdname(mddev), | |
258 | zone->nb_dev, (unsigned long long)sectors); | |
1da177e4 | 259 | |
49f357a2 | 260 | curr_zone_end += sectors; |
d27a43ab | 261 | zone->zone_end = curr_zone_end; |
1da177e4 | 262 | |
b5a20961 N |
263 | printk(KERN_INFO "md/raid0:%s: current zone start: %llu\n", |
264 | mdname(mddev), | |
265 | (unsigned long long)smallest->sectors); | |
1da177e4 | 266 | } |
1da177e4 | 267 | mddev->queue->unplug_fn = raid0_unplug; |
26be34dc N |
268 | mddev->queue->backing_dev_info.congested_fn = raid0_congested; |
269 | mddev->queue->backing_dev_info.congested_data = mddev; | |
1da177e4 | 270 | |
92e59b6b | 271 | /* |
272 | * now since we have the hard sector sizes, we can make sure | |
273 | * chunk size is a multiple of that sector size | |
274 | */ | |
9d8f0363 | 275 | if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) { |
b5a20961 | 276 | printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n", |
92e59b6b | 277 | mdname(mddev), |
9d8f0363 | 278 | mddev->chunk_sectors << 9); |
92e59b6b | 279 | goto abort; |
280 | } | |
8f6c2e4b MP |
281 | |
282 | blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9); | |
283 | blk_queue_io_opt(mddev->queue, | |
284 | (mddev->chunk_sectors << 9) * mddev->raid_disks); | |
285 | ||
b5a20961 | 286 | printk(KERN_INFO "md/raid0:%s: done.\n", mdname(mddev)); |
9af204cf TM |
287 | *private_conf = conf; |
288 | ||
1da177e4 | 289 | return 0; |
5568a603 | 290 | abort: |
ed7b0038 AN |
291 | kfree(conf->strip_zone); |
292 | kfree(conf->devlist); | |
293 | kfree(conf); | |
9af204cf | 294 | *private_conf = NULL; |
ed7b0038 | 295 | return err; |
1da177e4 LT |
296 | } |
297 | ||
298 | /** | |
299 | * raid0_mergeable_bvec -- tell bio layer if a two requests can be merged | |
300 | * @q: request queue | |
cc371e66 | 301 | * @bvm: properties of new bio |
1da177e4 LT |
302 | * @biovec: the request that could be merged to it. |
303 | * | |
304 | * Return amount of bytes we can accept at this offset | |
305 | */ | |
cc371e66 AK |
306 | static int raid0_mergeable_bvec(struct request_queue *q, |
307 | struct bvec_merge_data *bvm, | |
308 | struct bio_vec *biovec) | |
1da177e4 LT |
309 | { |
310 | mddev_t *mddev = q->queuedata; | |
cc371e66 | 311 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
1da177e4 | 312 | int max; |
9d8f0363 | 313 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 314 | unsigned int bio_sectors = bvm->bi_size >> 9; |
1da177e4 | 315 | |
d6e412ea | 316 | if (is_power_of_2(chunk_sectors)) |
fbb704ef | 317 | max = (chunk_sectors - ((sector & (chunk_sectors-1)) |
318 | + bio_sectors)) << 9; | |
319 | else | |
320 | max = (chunk_sectors - (sector_div(sector, chunk_sectors) | |
321 | + bio_sectors)) << 9; | |
1da177e4 LT |
322 | if (max < 0) max = 0; /* bio_add cannot handle a negative return */ |
323 | if (max <= biovec->bv_len && bio_sectors == 0) | |
324 | return biovec->bv_len; | |
325 | else | |
326 | return max; | |
327 | } | |
328 | ||
80c3a6ce DW |
329 | static sector_t raid0_size(mddev_t *mddev, sector_t sectors, int raid_disks) |
330 | { | |
331 | sector_t array_sectors = 0; | |
332 | mdk_rdev_t *rdev; | |
333 | ||
334 | WARN_ONCE(sectors || raid_disks, | |
335 | "%s does not support generic reshape\n", __func__); | |
336 | ||
337 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
338 | array_sectors += rdev->sectors; | |
339 | ||
340 | return array_sectors; | |
341 | } | |
342 | ||
8f79cfcd | 343 | static int raid0_run(mddev_t *mddev) |
1da177e4 | 344 | { |
9af204cf | 345 | raid0_conf_t *conf; |
5568a603 | 346 | int ret; |
1da177e4 | 347 | |
9d8f0363 | 348 | if (mddev->chunk_sectors == 0) { |
b5a20961 N |
349 | printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n", |
350 | mdname(mddev)); | |
2604b703 N |
351 | return -EINVAL; |
352 | } | |
0894cc30 AN |
353 | if (md_check_no_bitmap(mddev)) |
354 | return -EINVAL; | |
086fa5ff | 355 | blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors); |
e7e72bf6 | 356 | mddev->queue->queue_lock = &mddev->queue->__queue_lock; |
1da177e4 | 357 | |
9af204cf TM |
358 | /* if private is not null, we are here after takeover */ |
359 | if (mddev->private == NULL) { | |
360 | ret = create_strip_zones(mddev, &conf); | |
361 | if (ret < 0) | |
362 | return ret; | |
363 | mddev->private = conf; | |
364 | } | |
365 | conf = mddev->private; | |
1da177e4 LT |
366 | |
367 | /* calculate array device size */ | |
1f403624 | 368 | md_set_array_sectors(mddev, raid0_size(mddev, 0, 0)); |
1da177e4 | 369 | |
b5a20961 N |
370 | printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n", |
371 | mdname(mddev), | |
372 | (unsigned long long)mddev->array_sectors); | |
1da177e4 LT |
373 | /* calculate the max read-ahead size. |
374 | * For read-ahead of large files to be effective, we need to | |
375 | * readahead at least twice a whole stripe. i.e. number of devices | |
376 | * multiplied by chunk size times 2. | |
377 | * If an individual device has an ra_pages greater than the | |
378 | * chunk size, then we will not drive that device as hard as it | |
379 | * wants. We consider this a configuration error: a larger | |
380 | * chunksize should be used in that case. | |
381 | */ | |
382 | { | |
9d8f0363 AN |
383 | int stripe = mddev->raid_disks * |
384 | (mddev->chunk_sectors << 9) / PAGE_SIZE; | |
1da177e4 LT |
385 | if (mddev->queue->backing_dev_info.ra_pages < 2* stripe) |
386 | mddev->queue->backing_dev_info.ra_pages = 2* stripe; | |
387 | } | |
388 | ||
1da177e4 | 389 | blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec); |
46994191 | 390 | dump_zones(mddev); |
ac5e7113 | 391 | md_integrity_register(mddev); |
1da177e4 | 392 | return 0; |
1da177e4 LT |
393 | } |
394 | ||
fb5ab4b5 | 395 | static int raid0_stop(mddev_t *mddev) |
1da177e4 | 396 | { |
070ec55d | 397 | raid0_conf_t *conf = mddev->private; |
1da177e4 LT |
398 | |
399 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
990a8baf | 400 | kfree(conf->strip_zone); |
fb5ab4b5 | 401 | kfree(conf->devlist); |
990a8baf | 402 | kfree(conf); |
1da177e4 | 403 | mddev->private = NULL; |
1da177e4 LT |
404 | return 0; |
405 | } | |
406 | ||
49f357a2 N |
407 | /* Find the zone which holds a particular offset |
408 | * Update *sectorp to be an offset in that zone | |
409 | */ | |
dc582663 | 410 | static struct strip_zone *find_zone(struct raid0_private_data *conf, |
49f357a2 | 411 | sector_t *sectorp) |
dc582663 AN |
412 | { |
413 | int i; | |
414 | struct strip_zone *z = conf->strip_zone; | |
49f357a2 | 415 | sector_t sector = *sectorp; |
dc582663 AN |
416 | |
417 | for (i = 0; i < conf->nr_strip_zones; i++) | |
49f357a2 N |
418 | if (sector < z[i].zone_end) { |
419 | if (i) | |
420 | *sectorp = sector - z[i-1].zone_end; | |
dc582663 | 421 | return z + i; |
49f357a2 | 422 | } |
dc582663 AN |
423 | BUG(); |
424 | } | |
425 | ||
fbb704ef | 426 | /* |
427 | * remaps the bio to the target device. we separate two flows. | |
428 | * power 2 flow and a general flow for the sake of perfromance | |
429 | */ | |
430 | static mdk_rdev_t *map_sector(mddev_t *mddev, struct strip_zone *zone, | |
431 | sector_t sector, sector_t *sector_offset) | |
1da177e4 | 432 | { |
fbb704ef | 433 | unsigned int sect_in_chunk; |
434 | sector_t chunk; | |
070ec55d | 435 | raid0_conf_t *conf = mddev->private; |
84707f38 | 436 | int raid_disks = conf->strip_zone[0].nb_dev; |
9d8f0363 | 437 | unsigned int chunk_sects = mddev->chunk_sectors; |
fbb704ef | 438 | |
d6e412ea | 439 | if (is_power_of_2(chunk_sects)) { |
fbb704ef | 440 | int chunksect_bits = ffz(~chunk_sects); |
441 | /* find the sector offset inside the chunk */ | |
442 | sect_in_chunk = sector & (chunk_sects - 1); | |
443 | sector >>= chunksect_bits; | |
444 | /* chunk in zone */ | |
445 | chunk = *sector_offset; | |
446 | /* quotient is the chunk in real device*/ | |
447 | sector_div(chunk, zone->nb_dev << chunksect_bits); | |
448 | } else{ | |
449 | sect_in_chunk = sector_div(sector, chunk_sects); | |
450 | chunk = *sector_offset; | |
451 | sector_div(chunk, chunk_sects * zone->nb_dev); | |
452 | } | |
453 | /* | |
454 | * position the bio over the real device | |
455 | * real sector = chunk in device + starting of zone | |
456 | * + the position in the chunk | |
457 | */ | |
458 | *sector_offset = (chunk * chunk_sects) + sect_in_chunk; | |
84707f38 | 459 | return conf->devlist[(zone - conf->strip_zone)*raid_disks |
fbb704ef | 460 | + sector_div(sector, zone->nb_dev)]; |
461 | } | |
462 | ||
463 | /* | |
464 | * Is io distribute over 1 or more chunks ? | |
465 | */ | |
466 | static inline int is_io_in_chunk_boundary(mddev_t *mddev, | |
467 | unsigned int chunk_sects, struct bio *bio) | |
468 | { | |
d6e412ea | 469 | if (likely(is_power_of_2(chunk_sects))) { |
fbb704ef | 470 | return chunk_sects >= ((bio->bi_sector & (chunk_sects-1)) |
471 | + (bio->bi_size >> 9)); | |
472 | } else{ | |
473 | sector_t sector = bio->bi_sector; | |
474 | return chunk_sects >= (sector_div(sector, chunk_sects) | |
475 | + (bio->bi_size >> 9)); | |
476 | } | |
477 | } | |
478 | ||
21a52c6d | 479 | static int raid0_make_request(mddev_t *mddev, struct bio *bio) |
fbb704ef | 480 | { |
fbb704ef | 481 | unsigned int chunk_sects; |
482 | sector_t sector_offset; | |
1da177e4 LT |
483 | struct strip_zone *zone; |
484 | mdk_rdev_t *tmp_dev; | |
1da177e4 | 485 | |
7b6d91da | 486 | if (unlikely(bio->bi_rw & REQ_HARDBARRIER)) { |
a2826aa9 | 487 | md_barrier_request(mddev, bio); |
e5dcdd80 N |
488 | return 0; |
489 | } | |
490 | ||
9d8f0363 | 491 | chunk_sects = mddev->chunk_sectors; |
fbb704ef | 492 | if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) { |
493 | sector_t sector = bio->bi_sector; | |
1da177e4 LT |
494 | struct bio_pair *bp; |
495 | /* Sanity check -- queue functions should prevent this happening */ | |
496 | if (bio->bi_vcnt != 1 || | |
497 | bio->bi_idx != 0) | |
498 | goto bad_map; | |
499 | /* This is a one page bio that upper layers | |
500 | * refuse to split for us, so we need to split it. | |
501 | */ | |
d6e412ea | 502 | if (likely(is_power_of_2(chunk_sects))) |
fbb704ef | 503 | bp = bio_split(bio, chunk_sects - (sector & |
504 | (chunk_sects-1))); | |
505 | else | |
506 | bp = bio_split(bio, chunk_sects - | |
507 | sector_div(sector, chunk_sects)); | |
21a52c6d | 508 | if (raid0_make_request(mddev, &bp->bio1)) |
1da177e4 | 509 | generic_make_request(&bp->bio1); |
21a52c6d | 510 | if (raid0_make_request(mddev, &bp->bio2)) |
1da177e4 LT |
511 | generic_make_request(&bp->bio2); |
512 | ||
513 | bio_pair_release(bp); | |
514 | return 0; | |
515 | } | |
1da177e4 | 516 | |
fbb704ef | 517 | sector_offset = bio->bi_sector; |
518 | zone = find_zone(mddev->private, §or_offset); | |
519 | tmp_dev = map_sector(mddev, zone, bio->bi_sector, | |
520 | §or_offset); | |
1da177e4 | 521 | bio->bi_bdev = tmp_dev->bdev; |
fbb704ef | 522 | bio->bi_sector = sector_offset + zone->dev_start + |
523 | tmp_dev->data_offset; | |
1da177e4 LT |
524 | /* |
525 | * Let the main block layer submit the IO and resolve recursion: | |
526 | */ | |
527 | return 1; | |
528 | ||
529 | bad_map: | |
b5a20961 N |
530 | printk("md/raid0:%s: make_request bug: can't convert block across chunks" |
531 | " or bigger than %dk %llu %d\n", | |
532 | mdname(mddev), chunk_sects / 2, | |
533 | (unsigned long long)bio->bi_sector, bio->bi_size >> 10); | |
1da177e4 | 534 | |
6712ecf8 | 535 | bio_io_error(bio); |
1da177e4 LT |
536 | return 0; |
537 | } | |
8299d7f7 | 538 | |
1b961429 | 539 | static void raid0_status(struct seq_file *seq, mddev_t *mddev) |
1da177e4 LT |
540 | { |
541 | #undef MD_DEBUG | |
542 | #ifdef MD_DEBUG | |
543 | int j, k, h; | |
544 | char b[BDEVNAME_SIZE]; | |
070ec55d | 545 | raid0_conf_t *conf = mddev->private; |
84707f38 | 546 | int raid_disks = conf->strip_zone[0].nb_dev; |
8299d7f7 | 547 | |
1b961429 | 548 | sector_t zone_size; |
549 | sector_t zone_start = 0; | |
1da177e4 | 550 | h = 0; |
1b961429 | 551 | |
1da177e4 LT |
552 | for (j = 0; j < conf->nr_strip_zones; j++) { |
553 | seq_printf(seq, " z%d", j); | |
1da177e4 LT |
554 | seq_printf(seq, "=["); |
555 | for (k = 0; k < conf->strip_zone[j].nb_dev; k++) | |
8299d7f7 | 556 | seq_printf(seq, "%s/", bdevname( |
84707f38 | 557 | conf->devlist[j*raid_disks + k] |
1b961429 | 558 | ->bdev, b)); |
559 | ||
560 | zone_size = conf->strip_zone[j].zone_end - zone_start; | |
561 | seq_printf(seq, "] ze=%lld ds=%lld s=%lld\n", | |
562 | (unsigned long long)zone_start>>1, | |
563 | (unsigned long long)conf->strip_zone[j].dev_start>>1, | |
564 | (unsigned long long)zone_size>>1); | |
565 | zone_start = conf->strip_zone[j].zone_end; | |
1da177e4 LT |
566 | } |
567 | #endif | |
9d8f0363 | 568 | seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2); |
1da177e4 LT |
569 | return; |
570 | } | |
571 | ||
049d6c1e | 572 | static void *raid0_takeover_raid45(mddev_t *mddev) |
9af204cf TM |
573 | { |
574 | mdk_rdev_t *rdev; | |
575 | raid0_conf_t *priv_conf; | |
576 | ||
577 | if (mddev->degraded != 1) { | |
b5a20961 N |
578 | printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n", |
579 | mdname(mddev), | |
9af204cf TM |
580 | mddev->degraded); |
581 | return ERR_PTR(-EINVAL); | |
582 | } | |
583 | ||
584 | list_for_each_entry(rdev, &mddev->disks, same_set) { | |
585 | /* check slot number for a disk */ | |
586 | if (rdev->raid_disk == mddev->raid_disks-1) { | |
b5a20961 N |
587 | printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n", |
588 | mdname(mddev)); | |
9af204cf TM |
589 | return ERR_PTR(-EINVAL); |
590 | } | |
591 | } | |
592 | ||
593 | /* Set new parameters */ | |
594 | mddev->new_level = 0; | |
001048a3 | 595 | mddev->new_layout = 0; |
9af204cf TM |
596 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
597 | mddev->raid_disks--; | |
598 | mddev->delta_disks = -1; | |
599 | /* make sure it will be not marked as dirty */ | |
600 | mddev->recovery_cp = MaxSector; | |
601 | ||
602 | create_strip_zones(mddev, &priv_conf); | |
603 | return priv_conf; | |
604 | } | |
605 | ||
606 | static void *raid0_takeover_raid10(mddev_t *mddev) | |
607 | { | |
608 | raid0_conf_t *priv_conf; | |
609 | ||
610 | /* Check layout: | |
611 | * - far_copies must be 1 | |
612 | * - near_copies must be 2 | |
613 | * - disks number must be even | |
614 | * - all mirrors must be already degraded | |
615 | */ | |
616 | if (mddev->layout != ((1 << 8) + 2)) { | |
b5a20961 N |
617 | printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n", |
618 | mdname(mddev), | |
9af204cf TM |
619 | mddev->layout); |
620 | return ERR_PTR(-EINVAL); | |
621 | } | |
622 | if (mddev->raid_disks & 1) { | |
b5a20961 N |
623 | printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n", |
624 | mdname(mddev)); | |
9af204cf TM |
625 | return ERR_PTR(-EINVAL); |
626 | } | |
627 | if (mddev->degraded != (mddev->raid_disks>>1)) { | |
b5a20961 N |
628 | printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n", |
629 | mdname(mddev)); | |
9af204cf TM |
630 | return ERR_PTR(-EINVAL); |
631 | } | |
632 | ||
633 | /* Set new parameters */ | |
634 | mddev->new_level = 0; | |
001048a3 | 635 | mddev->new_layout = 0; |
9af204cf TM |
636 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
637 | mddev->delta_disks = - mddev->raid_disks / 2; | |
638 | mddev->raid_disks += mddev->delta_disks; | |
639 | mddev->degraded = 0; | |
640 | /* make sure it will be not marked as dirty */ | |
641 | mddev->recovery_cp = MaxSector; | |
642 | ||
643 | create_strip_zones(mddev, &priv_conf); | |
9af204cf TM |
644 | return priv_conf; |
645 | } | |
646 | ||
647 | static void *raid0_takeover(mddev_t *mddev) | |
648 | { | |
649 | /* raid0 can take over: | |
049d6c1e | 650 | * raid4 - if all data disks are active. |
9af204cf TM |
651 | * raid5 - providing it is Raid4 layout and one disk is faulty |
652 | * raid10 - assuming we have all necessary active disks | |
653 | */ | |
049d6c1e MT |
654 | if (mddev->level == 4) |
655 | return raid0_takeover_raid45(mddev); | |
656 | ||
9af204cf TM |
657 | if (mddev->level == 5) { |
658 | if (mddev->layout == ALGORITHM_PARITY_N) | |
049d6c1e | 659 | return raid0_takeover_raid45(mddev); |
9af204cf | 660 | |
b5a20961 N |
661 | printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n", |
662 | mdname(mddev), ALGORITHM_PARITY_N); | |
9af204cf TM |
663 | } |
664 | ||
665 | if (mddev->level == 10) | |
666 | return raid0_takeover_raid10(mddev); | |
667 | ||
668 | return ERR_PTR(-EINVAL); | |
669 | } | |
670 | ||
671 | static void raid0_quiesce(mddev_t *mddev, int state) | |
672 | { | |
673 | } | |
674 | ||
2604b703 | 675 | static struct mdk_personality raid0_personality= |
1da177e4 LT |
676 | { |
677 | .name = "raid0", | |
2604b703 | 678 | .level = 0, |
1da177e4 LT |
679 | .owner = THIS_MODULE, |
680 | .make_request = raid0_make_request, | |
681 | .run = raid0_run, | |
682 | .stop = raid0_stop, | |
683 | .status = raid0_status, | |
80c3a6ce | 684 | .size = raid0_size, |
9af204cf TM |
685 | .takeover = raid0_takeover, |
686 | .quiesce = raid0_quiesce, | |
1da177e4 LT |
687 | }; |
688 | ||
689 | static int __init raid0_init (void) | |
690 | { | |
2604b703 | 691 | return register_md_personality (&raid0_personality); |
1da177e4 LT |
692 | } |
693 | ||
694 | static void raid0_exit (void) | |
695 | { | |
2604b703 | 696 | unregister_md_personality (&raid0_personality); |
1da177e4 LT |
697 | } |
698 | ||
699 | module_init(raid0_init); | |
700 | module_exit(raid0_exit); | |
701 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 702 | MODULE_DESCRIPTION("RAID0 (striping) personality for MD"); |
1da177e4 | 703 | MODULE_ALIAS("md-personality-2"); /* RAID0 */ |
d9d166c2 | 704 | MODULE_ALIAS("md-raid0"); |
2604b703 | 705 | MODULE_ALIAS("md-level-0"); |