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> |
056075c7 | 23 | #include <linux/module.h> |
5a0e3ad6 | 24 | #include <linux/slab.h> |
43b2e5d8 | 25 | #include "md.h" |
ef740c37 | 26 | #include "raid0.h" |
9af204cf | 27 | #include "raid5.h" |
1da177e4 | 28 | |
26be34dc N |
29 | static int raid0_congested(void *data, int bits) |
30 | { | |
fd01b88c | 31 | struct mddev *mddev = data; |
e373ab10 | 32 | struct r0conf *conf = mddev->private; |
3cb03002 | 33 | struct md_rdev **devlist = conf->devlist; |
84707f38 | 34 | int raid_disks = conf->strip_zone[0].nb_dev; |
26be34dc N |
35 | int i, ret = 0; |
36 | ||
3fa841d7 N |
37 | if (mddev_congested(mddev, bits)) |
38 | return 1; | |
39 | ||
84707f38 | 40 | for (i = 0; i < raid_disks && !ret ; i++) { |
165125e1 | 41 | struct request_queue *q = bdev_get_queue(devlist[i]->bdev); |
26be34dc N |
42 | |
43 | ret |= bdi_congested(&q->backing_dev_info, bits); | |
44 | } | |
45 | return ret; | |
46 | } | |
47 | ||
46994191 | 48 | /* |
49 | * inform the user of the raid configuration | |
50 | */ | |
fd01b88c | 51 | static void dump_zones(struct mddev *mddev) |
46994191 | 52 | { |
50de8df4 | 53 | int j, k; |
46994191 | 54 | sector_t zone_size = 0; |
55 | sector_t zone_start = 0; | |
56 | char b[BDEVNAME_SIZE]; | |
e373ab10 | 57 | struct r0conf *conf = mddev->private; |
84707f38 | 58 | int raid_disks = conf->strip_zone[0].nb_dev; |
50de8df4 N |
59 | printk(KERN_INFO "md: RAID0 configuration for %s - %d zone%s\n", |
60 | mdname(mddev), | |
61 | conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s"); | |
46994191 | 62 | for (j = 0; j < conf->nr_strip_zones; j++) { |
50de8df4 | 63 | printk(KERN_INFO "md: zone%d=[", j); |
46994191 | 64 | for (k = 0; k < conf->strip_zone[j].nb_dev; k++) |
50de8df4 | 65 | printk(KERN_CONT "%s%s", k?"/":"", |
84707f38 | 66 | bdevname(conf->devlist[j*raid_disks |
46994191 | 67 | + k]->bdev, b)); |
b5a20961 | 68 | printk(KERN_CONT "]\n"); |
46994191 | 69 | |
70 | zone_size = conf->strip_zone[j].zone_end - zone_start; | |
50de8df4 N |
71 | printk(KERN_INFO " zone-offset=%10lluKB, " |
72 | "device-offset=%10lluKB, size=%10lluKB\n", | |
46994191 | 73 | (unsigned long long)zone_start>>1, |
74 | (unsigned long long)conf->strip_zone[j].dev_start>>1, | |
75 | (unsigned long long)zone_size>>1); | |
76 | zone_start = conf->strip_zone[j].zone_end; | |
77 | } | |
50de8df4 | 78 | printk(KERN_INFO "\n"); |
46994191 | 79 | } |
80 | ||
e373ab10 | 81 | static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf) |
1da177e4 | 82 | { |
a9f326eb | 83 | int i, c, err; |
49f357a2 | 84 | sector_t curr_zone_end, sectors; |
3cb03002 | 85 | struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev; |
1da177e4 LT |
86 | struct strip_zone *zone; |
87 | int cnt; | |
88 | char b[BDEVNAME_SIZE]; | |
50de8df4 | 89 | char b2[BDEVNAME_SIZE]; |
e373ab10 | 90 | struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL); |
ed7b0038 AN |
91 | |
92 | if (!conf) | |
93 | return -ENOMEM; | |
dafb20fa | 94 | rdev_for_each(rdev1, mddev) { |
50de8df4 N |
95 | pr_debug("md/raid0:%s: looking at %s\n", |
96 | mdname(mddev), | |
97 | bdevname(rdev1->bdev, b)); | |
1da177e4 | 98 | c = 0; |
13f2682b N |
99 | |
100 | /* round size to chunk_size */ | |
101 | sectors = rdev1->sectors; | |
102 | sector_div(sectors, mddev->chunk_sectors); | |
103 | rdev1->sectors = sectors * mddev->chunk_sectors; | |
104 | ||
dafb20fa | 105 | rdev_for_each(rdev2, mddev) { |
50de8df4 N |
106 | pr_debug("md/raid0:%s: comparing %s(%llu)" |
107 | " with %s(%llu)\n", | |
108 | mdname(mddev), | |
109 | bdevname(rdev1->bdev,b), | |
110 | (unsigned long long)rdev1->sectors, | |
111 | bdevname(rdev2->bdev,b2), | |
112 | (unsigned long long)rdev2->sectors); | |
1da177e4 | 113 | if (rdev2 == rdev1) { |
50de8df4 N |
114 | pr_debug("md/raid0:%s: END\n", |
115 | mdname(mddev)); | |
1da177e4 LT |
116 | break; |
117 | } | |
dd8ac336 | 118 | if (rdev2->sectors == rdev1->sectors) { |
1da177e4 LT |
119 | /* |
120 | * Not unique, don't count it as a new | |
121 | * group | |
122 | */ | |
50de8df4 N |
123 | pr_debug("md/raid0:%s: EQUAL\n", |
124 | mdname(mddev)); | |
1da177e4 LT |
125 | c = 1; |
126 | break; | |
127 | } | |
50de8df4 N |
128 | pr_debug("md/raid0:%s: NOT EQUAL\n", |
129 | mdname(mddev)); | |
1da177e4 LT |
130 | } |
131 | if (!c) { | |
50de8df4 N |
132 | pr_debug("md/raid0:%s: ==> UNIQUE\n", |
133 | mdname(mddev)); | |
1da177e4 | 134 | conf->nr_strip_zones++; |
50de8df4 N |
135 | pr_debug("md/raid0:%s: %d zones\n", |
136 | mdname(mddev), conf->nr_strip_zones); | |
1da177e4 LT |
137 | } |
138 | } | |
50de8df4 N |
139 | pr_debug("md/raid0:%s: FINAL %d zones\n", |
140 | mdname(mddev), conf->nr_strip_zones); | |
ed7b0038 | 141 | err = -ENOMEM; |
9ffae0cf | 142 | conf->strip_zone = kzalloc(sizeof(struct strip_zone)* |
1da177e4 LT |
143 | conf->nr_strip_zones, GFP_KERNEL); |
144 | if (!conf->strip_zone) | |
ed7b0038 | 145 | goto abort; |
3cb03002 | 146 | conf->devlist = kzalloc(sizeof(struct md_rdev*)* |
1da177e4 LT |
147 | conf->nr_strip_zones*mddev->raid_disks, |
148 | GFP_KERNEL); | |
149 | if (!conf->devlist) | |
ed7b0038 | 150 | goto abort; |
1da177e4 | 151 | |
1da177e4 LT |
152 | /* The first zone must contain all devices, so here we check that |
153 | * there is a proper alignment of slots to devices and find them all | |
154 | */ | |
155 | zone = &conf->strip_zone[0]; | |
156 | cnt = 0; | |
157 | smallest = NULL; | |
b414579f | 158 | dev = conf->devlist; |
ed7b0038 | 159 | err = -EINVAL; |
dafb20fa | 160 | rdev_for_each(rdev1, mddev) { |
1da177e4 LT |
161 | int j = rdev1->raid_disk; |
162 | ||
e93f68a1 | 163 | if (mddev->level == 10) { |
9af204cf TM |
164 | /* taking over a raid10-n2 array */ |
165 | j /= 2; | |
e93f68a1 N |
166 | rdev1->new_raid_disk = j; |
167 | } | |
9af204cf | 168 | |
fc3a08b8 KW |
169 | if (mddev->level == 1) { |
170 | /* taiking over a raid1 array- | |
171 | * we have only one active disk | |
172 | */ | |
173 | j = 0; | |
174 | rdev1->new_raid_disk = j; | |
175 | } | |
176 | ||
1da177e4 | 177 | if (j < 0 || j >= mddev->raid_disks) { |
b5a20961 N |
178 | printk(KERN_ERR "md/raid0:%s: bad disk number %d - " |
179 | "aborting!\n", mdname(mddev), j); | |
1da177e4 LT |
180 | goto abort; |
181 | } | |
b414579f | 182 | if (dev[j]) { |
b5a20961 N |
183 | printk(KERN_ERR "md/raid0:%s: multiple devices for %d - " |
184 | "aborting!\n", mdname(mddev), j); | |
1da177e4 LT |
185 | goto abort; |
186 | } | |
b414579f | 187 | dev[j] = rdev1; |
1da177e4 | 188 | |
8f6c2e4b MP |
189 | disk_stack_limits(mddev->gendisk, rdev1->bdev, |
190 | rdev1->data_offset << 9); | |
1da177e4 | 191 | |
ba13da47 N |
192 | if (rdev1->bdev->bd_disk->queue->merge_bvec_fn) |
193 | conf->has_merge_bvec = 1; | |
194 | ||
dd8ac336 | 195 | if (!smallest || (rdev1->sectors < smallest->sectors)) |
1da177e4 LT |
196 | smallest = rdev1; |
197 | cnt++; | |
198 | } | |
199 | if (cnt != mddev->raid_disks) { | |
b5a20961 N |
200 | printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - " |
201 | "aborting!\n", mdname(mddev), cnt, mddev->raid_disks); | |
1da177e4 LT |
202 | goto abort; |
203 | } | |
204 | zone->nb_dev = cnt; | |
49f357a2 | 205 | zone->zone_end = smallest->sectors * cnt; |
1da177e4 | 206 | |
49f357a2 | 207 | curr_zone_end = zone->zone_end; |
1da177e4 LT |
208 | |
209 | /* now do the other zones */ | |
210 | for (i = 1; i < conf->nr_strip_zones; i++) | |
211 | { | |
a9f326eb N |
212 | int j; |
213 | ||
1da177e4 | 214 | zone = conf->strip_zone + i; |
b414579f | 215 | dev = conf->devlist + i * mddev->raid_disks; |
1da177e4 | 216 | |
50de8df4 | 217 | pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i); |
d27a43ab | 218 | zone->dev_start = smallest->sectors; |
1da177e4 LT |
219 | smallest = NULL; |
220 | c = 0; | |
221 | ||
222 | for (j=0; j<cnt; j++) { | |
b414579f | 223 | rdev = conf->devlist[j]; |
d27a43ab | 224 | if (rdev->sectors <= zone->dev_start) { |
50de8df4 N |
225 | pr_debug("md/raid0:%s: checking %s ... nope\n", |
226 | mdname(mddev), | |
227 | bdevname(rdev->bdev, b)); | |
dd8ac336 AN |
228 | continue; |
229 | } | |
50de8df4 N |
230 | pr_debug("md/raid0:%s: checking %s ..." |
231 | " contained as device %d\n", | |
232 | mdname(mddev), | |
233 | bdevname(rdev->bdev, b), c); | |
b414579f | 234 | dev[c] = rdev; |
dd8ac336 AN |
235 | c++; |
236 | if (!smallest || rdev->sectors < smallest->sectors) { | |
237 | smallest = rdev; | |
50de8df4 N |
238 | pr_debug("md/raid0:%s: (%llu) is smallest!.\n", |
239 | mdname(mddev), | |
240 | (unsigned long long)rdev->sectors); | |
dd8ac336 | 241 | } |
1da177e4 LT |
242 | } |
243 | ||
244 | zone->nb_dev = c; | |
49f357a2 | 245 | sectors = (smallest->sectors - zone->dev_start) * c; |
50de8df4 N |
246 | pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n", |
247 | mdname(mddev), | |
248 | zone->nb_dev, (unsigned long long)sectors); | |
1da177e4 | 249 | |
49f357a2 | 250 | curr_zone_end += sectors; |
d27a43ab | 251 | zone->zone_end = curr_zone_end; |
1da177e4 | 252 | |
50de8df4 N |
253 | pr_debug("md/raid0:%s: current zone start: %llu\n", |
254 | mdname(mddev), | |
255 | (unsigned long long)smallest->sectors); | |
1da177e4 | 256 | } |
26be34dc N |
257 | mddev->queue->backing_dev_info.congested_fn = raid0_congested; |
258 | mddev->queue->backing_dev_info.congested_data = mddev; | |
1da177e4 | 259 | |
92e59b6b | 260 | /* |
261 | * now since we have the hard sector sizes, we can make sure | |
262 | * chunk size is a multiple of that sector size | |
263 | */ | |
9d8f0363 | 264 | if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) { |
b5a20961 | 265 | printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n", |
92e59b6b | 266 | mdname(mddev), |
9d8f0363 | 267 | mddev->chunk_sectors << 9); |
92e59b6b | 268 | goto abort; |
269 | } | |
8f6c2e4b MP |
270 | |
271 | blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9); | |
272 | blk_queue_io_opt(mddev->queue, | |
273 | (mddev->chunk_sectors << 9) * mddev->raid_disks); | |
274 | ||
50de8df4 | 275 | pr_debug("md/raid0:%s: done.\n", mdname(mddev)); |
9af204cf TM |
276 | *private_conf = conf; |
277 | ||
1da177e4 | 278 | return 0; |
5568a603 | 279 | abort: |
ed7b0038 AN |
280 | kfree(conf->strip_zone); |
281 | kfree(conf->devlist); | |
282 | kfree(conf); | |
9af204cf | 283 | *private_conf = NULL; |
ed7b0038 | 284 | return err; |
1da177e4 LT |
285 | } |
286 | ||
ba13da47 N |
287 | /* Find the zone which holds a particular offset |
288 | * Update *sectorp to be an offset in that zone | |
289 | */ | |
290 | static struct strip_zone *find_zone(struct r0conf *conf, | |
291 | sector_t *sectorp) | |
292 | { | |
293 | int i; | |
294 | struct strip_zone *z = conf->strip_zone; | |
295 | sector_t sector = *sectorp; | |
296 | ||
297 | for (i = 0; i < conf->nr_strip_zones; i++) | |
298 | if (sector < z[i].zone_end) { | |
299 | if (i) | |
300 | *sectorp = sector - z[i-1].zone_end; | |
301 | return z + i; | |
302 | } | |
303 | BUG(); | |
304 | } | |
305 | ||
306 | /* | |
307 | * remaps the bio to the target device. we separate two flows. | |
308 | * power 2 flow and a general flow for the sake of perfromance | |
309 | */ | |
310 | static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone, | |
311 | sector_t sector, sector_t *sector_offset) | |
312 | { | |
313 | unsigned int sect_in_chunk; | |
314 | sector_t chunk; | |
315 | struct r0conf *conf = mddev->private; | |
316 | int raid_disks = conf->strip_zone[0].nb_dev; | |
317 | unsigned int chunk_sects = mddev->chunk_sectors; | |
318 | ||
319 | if (is_power_of_2(chunk_sects)) { | |
320 | int chunksect_bits = ffz(~chunk_sects); | |
321 | /* find the sector offset inside the chunk */ | |
322 | sect_in_chunk = sector & (chunk_sects - 1); | |
323 | sector >>= chunksect_bits; | |
324 | /* chunk in zone */ | |
325 | chunk = *sector_offset; | |
326 | /* quotient is the chunk in real device*/ | |
327 | sector_div(chunk, zone->nb_dev << chunksect_bits); | |
328 | } else{ | |
329 | sect_in_chunk = sector_div(sector, chunk_sects); | |
330 | chunk = *sector_offset; | |
331 | sector_div(chunk, chunk_sects * zone->nb_dev); | |
332 | } | |
333 | /* | |
334 | * position the bio over the real device | |
335 | * real sector = chunk in device + starting of zone | |
336 | * + the position in the chunk | |
337 | */ | |
338 | *sector_offset = (chunk * chunk_sects) + sect_in_chunk; | |
339 | return conf->devlist[(zone - conf->strip_zone)*raid_disks | |
340 | + sector_div(sector, zone->nb_dev)]; | |
341 | } | |
342 | ||
1da177e4 | 343 | /** |
ba13da47 | 344 | * raid0_mergeable_bvec -- tell bio layer if two requests can be merged |
1da177e4 | 345 | * @q: request queue |
cc371e66 | 346 | * @bvm: properties of new bio |
1da177e4 LT |
347 | * @biovec: the request that could be merged to it. |
348 | * | |
349 | * Return amount of bytes we can accept at this offset | |
350 | */ | |
cc371e66 AK |
351 | static int raid0_mergeable_bvec(struct request_queue *q, |
352 | struct bvec_merge_data *bvm, | |
353 | struct bio_vec *biovec) | |
1da177e4 | 354 | { |
fd01b88c | 355 | struct mddev *mddev = q->queuedata; |
ba13da47 | 356 | struct r0conf *conf = mddev->private; |
cc371e66 | 357 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
ba13da47 | 358 | sector_t sector_offset = sector; |
1da177e4 | 359 | int max; |
9d8f0363 | 360 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 361 | unsigned int bio_sectors = bvm->bi_size >> 9; |
ba13da47 N |
362 | struct strip_zone *zone; |
363 | struct md_rdev *rdev; | |
364 | struct request_queue *subq; | |
1da177e4 | 365 | |
d6e412ea | 366 | if (is_power_of_2(chunk_sectors)) |
fbb704ef | 367 | max = (chunk_sectors - ((sector & (chunk_sectors-1)) |
368 | + bio_sectors)) << 9; | |
369 | else | |
370 | max = (chunk_sectors - (sector_div(sector, chunk_sectors) | |
371 | + bio_sectors)) << 9; | |
ba13da47 N |
372 | if (max < 0) |
373 | max = 0; /* bio_add cannot handle a negative return */ | |
1da177e4 LT |
374 | if (max <= biovec->bv_len && bio_sectors == 0) |
375 | return biovec->bv_len; | |
ba13da47 N |
376 | if (max < biovec->bv_len) |
377 | /* too small already, no need to check further */ | |
378 | return max; | |
379 | if (!conf->has_merge_bvec) | |
380 | return max; | |
381 | ||
382 | /* May need to check subordinate device */ | |
383 | sector = sector_offset; | |
384 | zone = find_zone(mddev->private, §or_offset); | |
385 | rdev = map_sector(mddev, zone, sector, §or_offset); | |
386 | subq = bdev_get_queue(rdev->bdev); | |
387 | if (subq->merge_bvec_fn) { | |
388 | bvm->bi_bdev = rdev->bdev; | |
389 | bvm->bi_sector = sector_offset + zone->dev_start + | |
390 | rdev->data_offset; | |
391 | return min(max, subq->merge_bvec_fn(subq, bvm, biovec)); | |
392 | } else | |
1da177e4 LT |
393 | return max; |
394 | } | |
395 | ||
fd01b88c | 396 | static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
397 | { |
398 | sector_t array_sectors = 0; | |
3cb03002 | 399 | struct md_rdev *rdev; |
80c3a6ce DW |
400 | |
401 | WARN_ONCE(sectors || raid_disks, | |
402 | "%s does not support generic reshape\n", __func__); | |
403 | ||
dafb20fa | 404 | rdev_for_each(rdev, mddev) |
80c3a6ce DW |
405 | array_sectors += rdev->sectors; |
406 | ||
407 | return array_sectors; | |
408 | } | |
409 | ||
fd01b88c | 410 | static int raid0_run(struct mddev *mddev) |
1da177e4 | 411 | { |
e373ab10 | 412 | struct r0conf *conf; |
5568a603 | 413 | int ret; |
1da177e4 | 414 | |
9d8f0363 | 415 | if (mddev->chunk_sectors == 0) { |
b5a20961 N |
416 | printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n", |
417 | mdname(mddev)); | |
2604b703 N |
418 | return -EINVAL; |
419 | } | |
0894cc30 AN |
420 | if (md_check_no_bitmap(mddev)) |
421 | return -EINVAL; | |
086fa5ff | 422 | blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors); |
1da177e4 | 423 | |
9af204cf TM |
424 | /* if private is not null, we are here after takeover */ |
425 | if (mddev->private == NULL) { | |
426 | ret = create_strip_zones(mddev, &conf); | |
427 | if (ret < 0) | |
428 | return ret; | |
429 | mddev->private = conf; | |
430 | } | |
431 | conf = mddev->private; | |
1da177e4 LT |
432 | |
433 | /* calculate array device size */ | |
1f403624 | 434 | md_set_array_sectors(mddev, raid0_size(mddev, 0, 0)); |
1da177e4 | 435 | |
b5a20961 N |
436 | printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n", |
437 | mdname(mddev), | |
438 | (unsigned long long)mddev->array_sectors); | |
1da177e4 LT |
439 | /* calculate the max read-ahead size. |
440 | * For read-ahead of large files to be effective, we need to | |
441 | * readahead at least twice a whole stripe. i.e. number of devices | |
442 | * multiplied by chunk size times 2. | |
443 | * If an individual device has an ra_pages greater than the | |
444 | * chunk size, then we will not drive that device as hard as it | |
445 | * wants. We consider this a configuration error: a larger | |
446 | * chunksize should be used in that case. | |
447 | */ | |
448 | { | |
9d8f0363 AN |
449 | int stripe = mddev->raid_disks * |
450 | (mddev->chunk_sectors << 9) / PAGE_SIZE; | |
1da177e4 LT |
451 | if (mddev->queue->backing_dev_info.ra_pages < 2* stripe) |
452 | mddev->queue->backing_dev_info.ra_pages = 2* stripe; | |
453 | } | |
454 | ||
1da177e4 | 455 | blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec); |
46994191 | 456 | dump_zones(mddev); |
a91a2785 | 457 | return md_integrity_register(mddev); |
1da177e4 LT |
458 | } |
459 | ||
fd01b88c | 460 | static int raid0_stop(struct mddev *mddev) |
1da177e4 | 461 | { |
e373ab10 | 462 | struct r0conf *conf = mddev->private; |
1da177e4 LT |
463 | |
464 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
990a8baf | 465 | kfree(conf->strip_zone); |
fb5ab4b5 | 466 | kfree(conf->devlist); |
990a8baf | 467 | kfree(conf); |
1da177e4 | 468 | mddev->private = NULL; |
1da177e4 LT |
469 | return 0; |
470 | } | |
471 | ||
fbb704ef | 472 | /* |
473 | * Is io distribute over 1 or more chunks ? | |
474 | */ | |
fd01b88c | 475 | static inline int is_io_in_chunk_boundary(struct mddev *mddev, |
fbb704ef | 476 | unsigned int chunk_sects, struct bio *bio) |
477 | { | |
d6e412ea | 478 | if (likely(is_power_of_2(chunk_sects))) { |
fbb704ef | 479 | return chunk_sects >= ((bio->bi_sector & (chunk_sects-1)) |
480 | + (bio->bi_size >> 9)); | |
481 | } else{ | |
482 | sector_t sector = bio->bi_sector; | |
483 | return chunk_sects >= (sector_div(sector, chunk_sects) | |
484 | + (bio->bi_size >> 9)); | |
485 | } | |
486 | } | |
487 | ||
b4fdcb02 | 488 | static void raid0_make_request(struct mddev *mddev, struct bio *bio) |
fbb704ef | 489 | { |
fbb704ef | 490 | unsigned int chunk_sects; |
491 | sector_t sector_offset; | |
1da177e4 | 492 | struct strip_zone *zone; |
3cb03002 | 493 | struct md_rdev *tmp_dev; |
1da177e4 | 494 | |
e9c7469b TH |
495 | if (unlikely(bio->bi_rw & REQ_FLUSH)) { |
496 | md_flush_request(mddev, bio); | |
5a7bbad2 | 497 | return; |
e5dcdd80 N |
498 | } |
499 | ||
9d8f0363 | 500 | chunk_sects = mddev->chunk_sectors; |
fbb704ef | 501 | if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) { |
502 | sector_t sector = bio->bi_sector; | |
1da177e4 LT |
503 | struct bio_pair *bp; |
504 | /* Sanity check -- queue functions should prevent this happening */ | |
505 | if (bio->bi_vcnt != 1 || | |
506 | bio->bi_idx != 0) | |
507 | goto bad_map; | |
508 | /* This is a one page bio that upper layers | |
509 | * refuse to split for us, so we need to split it. | |
510 | */ | |
d6e412ea | 511 | if (likely(is_power_of_2(chunk_sects))) |
fbb704ef | 512 | bp = bio_split(bio, chunk_sects - (sector & |
513 | (chunk_sects-1))); | |
514 | else | |
515 | bp = bio_split(bio, chunk_sects - | |
516 | sector_div(sector, chunk_sects)); | |
5a7bbad2 CH |
517 | raid0_make_request(mddev, &bp->bio1); |
518 | raid0_make_request(mddev, &bp->bio2); | |
1da177e4 | 519 | bio_pair_release(bp); |
5a7bbad2 | 520 | return; |
1da177e4 | 521 | } |
1da177e4 | 522 | |
fbb704ef | 523 | sector_offset = bio->bi_sector; |
ba13da47 | 524 | zone = find_zone(mddev->private, §or_offset); |
fbb704ef | 525 | tmp_dev = map_sector(mddev, zone, bio->bi_sector, |
526 | §or_offset); | |
1da177e4 | 527 | bio->bi_bdev = tmp_dev->bdev; |
fbb704ef | 528 | bio->bi_sector = sector_offset + zone->dev_start + |
529 | tmp_dev->data_offset; | |
5a7bbad2 CH |
530 | |
531 | generic_make_request(bio); | |
532 | return; | |
1da177e4 LT |
533 | |
534 | bad_map: | |
b5a20961 N |
535 | printk("md/raid0:%s: make_request bug: can't convert block across chunks" |
536 | " or bigger than %dk %llu %d\n", | |
537 | mdname(mddev), chunk_sects / 2, | |
538 | (unsigned long long)bio->bi_sector, bio->bi_size >> 10); | |
1da177e4 | 539 | |
6712ecf8 | 540 | bio_io_error(bio); |
5a7bbad2 | 541 | return; |
1da177e4 | 542 | } |
8299d7f7 | 543 | |
fd01b88c | 544 | static void raid0_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 545 | { |
9d8f0363 | 546 | seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2); |
1da177e4 LT |
547 | return; |
548 | } | |
549 | ||
fd01b88c | 550 | static void *raid0_takeover_raid45(struct mddev *mddev) |
9af204cf | 551 | { |
3cb03002 | 552 | struct md_rdev *rdev; |
e373ab10 | 553 | struct r0conf *priv_conf; |
9af204cf TM |
554 | |
555 | if (mddev->degraded != 1) { | |
b5a20961 N |
556 | printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n", |
557 | mdname(mddev), | |
9af204cf TM |
558 | mddev->degraded); |
559 | return ERR_PTR(-EINVAL); | |
560 | } | |
561 | ||
dafb20fa | 562 | rdev_for_each(rdev, mddev) { |
9af204cf TM |
563 | /* check slot number for a disk */ |
564 | if (rdev->raid_disk == mddev->raid_disks-1) { | |
b5a20961 N |
565 | printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n", |
566 | mdname(mddev)); | |
9af204cf TM |
567 | return ERR_PTR(-EINVAL); |
568 | } | |
569 | } | |
570 | ||
571 | /* Set new parameters */ | |
572 | mddev->new_level = 0; | |
001048a3 | 573 | mddev->new_layout = 0; |
9af204cf TM |
574 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
575 | mddev->raid_disks--; | |
576 | mddev->delta_disks = -1; | |
577 | /* make sure it will be not marked as dirty */ | |
578 | mddev->recovery_cp = MaxSector; | |
579 | ||
580 | create_strip_zones(mddev, &priv_conf); | |
581 | return priv_conf; | |
582 | } | |
583 | ||
fd01b88c | 584 | static void *raid0_takeover_raid10(struct mddev *mddev) |
9af204cf | 585 | { |
e373ab10 | 586 | struct r0conf *priv_conf; |
9af204cf TM |
587 | |
588 | /* Check layout: | |
589 | * - far_copies must be 1 | |
590 | * - near_copies must be 2 | |
591 | * - disks number must be even | |
592 | * - all mirrors must be already degraded | |
593 | */ | |
594 | if (mddev->layout != ((1 << 8) + 2)) { | |
b5a20961 N |
595 | printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n", |
596 | mdname(mddev), | |
9af204cf TM |
597 | mddev->layout); |
598 | return ERR_PTR(-EINVAL); | |
599 | } | |
600 | if (mddev->raid_disks & 1) { | |
b5a20961 N |
601 | printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n", |
602 | mdname(mddev)); | |
9af204cf TM |
603 | return ERR_PTR(-EINVAL); |
604 | } | |
605 | if (mddev->degraded != (mddev->raid_disks>>1)) { | |
b5a20961 N |
606 | printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n", |
607 | mdname(mddev)); | |
9af204cf TM |
608 | return ERR_PTR(-EINVAL); |
609 | } | |
610 | ||
611 | /* Set new parameters */ | |
612 | mddev->new_level = 0; | |
001048a3 | 613 | mddev->new_layout = 0; |
9af204cf TM |
614 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
615 | mddev->delta_disks = - mddev->raid_disks / 2; | |
616 | mddev->raid_disks += mddev->delta_disks; | |
617 | mddev->degraded = 0; | |
618 | /* make sure it will be not marked as dirty */ | |
619 | mddev->recovery_cp = MaxSector; | |
620 | ||
621 | create_strip_zones(mddev, &priv_conf); | |
9af204cf TM |
622 | return priv_conf; |
623 | } | |
624 | ||
fd01b88c | 625 | static void *raid0_takeover_raid1(struct mddev *mddev) |
fc3a08b8 | 626 | { |
e373ab10 | 627 | struct r0conf *priv_conf; |
fc3a08b8 KW |
628 | |
629 | /* Check layout: | |
630 | * - (N - 1) mirror drives must be already faulty | |
631 | */ | |
632 | if ((mddev->raid_disks - 1) != mddev->degraded) { | |
633 | printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n", | |
634 | mdname(mddev)); | |
635 | return ERR_PTR(-EINVAL); | |
636 | } | |
637 | ||
638 | /* Set new parameters */ | |
639 | mddev->new_level = 0; | |
640 | mddev->new_layout = 0; | |
641 | mddev->new_chunk_sectors = 128; /* by default set chunk size to 64k */ | |
642 | mddev->delta_disks = 1 - mddev->raid_disks; | |
f7bee809 | 643 | mddev->raid_disks = 1; |
fc3a08b8 KW |
644 | /* make sure it will be not marked as dirty */ |
645 | mddev->recovery_cp = MaxSector; | |
646 | ||
647 | create_strip_zones(mddev, &priv_conf); | |
648 | return priv_conf; | |
649 | } | |
650 | ||
fd01b88c | 651 | static void *raid0_takeover(struct mddev *mddev) |
9af204cf TM |
652 | { |
653 | /* raid0 can take over: | |
049d6c1e | 654 | * raid4 - if all data disks are active. |
9af204cf TM |
655 | * raid5 - providing it is Raid4 layout and one disk is faulty |
656 | * raid10 - assuming we have all necessary active disks | |
fc3a08b8 | 657 | * raid1 - with (N -1) mirror drives faulty |
9af204cf | 658 | */ |
049d6c1e MT |
659 | if (mddev->level == 4) |
660 | return raid0_takeover_raid45(mddev); | |
661 | ||
9af204cf TM |
662 | if (mddev->level == 5) { |
663 | if (mddev->layout == ALGORITHM_PARITY_N) | |
049d6c1e | 664 | return raid0_takeover_raid45(mddev); |
9af204cf | 665 | |
b5a20961 N |
666 | printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n", |
667 | mdname(mddev), ALGORITHM_PARITY_N); | |
9af204cf TM |
668 | } |
669 | ||
670 | if (mddev->level == 10) | |
671 | return raid0_takeover_raid10(mddev); | |
672 | ||
fc3a08b8 KW |
673 | if (mddev->level == 1) |
674 | return raid0_takeover_raid1(mddev); | |
675 | ||
676 | printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n", | |
677 | mddev->level); | |
678 | ||
9af204cf TM |
679 | return ERR_PTR(-EINVAL); |
680 | } | |
681 | ||
fd01b88c | 682 | static void raid0_quiesce(struct mddev *mddev, int state) |
9af204cf TM |
683 | { |
684 | } | |
685 | ||
84fc4b56 | 686 | static struct md_personality raid0_personality= |
1da177e4 LT |
687 | { |
688 | .name = "raid0", | |
2604b703 | 689 | .level = 0, |
1da177e4 LT |
690 | .owner = THIS_MODULE, |
691 | .make_request = raid0_make_request, | |
692 | .run = raid0_run, | |
693 | .stop = raid0_stop, | |
694 | .status = raid0_status, | |
80c3a6ce | 695 | .size = raid0_size, |
9af204cf TM |
696 | .takeover = raid0_takeover, |
697 | .quiesce = raid0_quiesce, | |
1da177e4 LT |
698 | }; |
699 | ||
700 | static int __init raid0_init (void) | |
701 | { | |
2604b703 | 702 | return register_md_personality (&raid0_personality); |
1da177e4 LT |
703 | } |
704 | ||
705 | static void raid0_exit (void) | |
706 | { | |
2604b703 | 707 | unregister_md_personality (&raid0_personality); |
1da177e4 LT |
708 | } |
709 | ||
710 | module_init(raid0_init); | |
711 | module_exit(raid0_exit); | |
712 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 713 | MODULE_DESCRIPTION("RAID0 (striping) personality for MD"); |
1da177e4 | 714 | MODULE_ALIAS("md-personality-2"); /* RAID0 */ |
d9d166c2 | 715 | MODULE_ALIAS("md-raid0"); |
2604b703 | 716 | MODULE_ALIAS("md-level-0"); |