Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | raid0.c : Multiple Devices driver for Linux | |
f72ffdd6 | 3 | Copyright (C) 1994-96 Marc ZYNGIER |
1da177e4 LT |
4 | <zyngier@ufr-info-p7.ibp.fr> or |
5 | <maz@gloups.fdn.fr> | |
f72ffdd6 | 6 | Copyright (C) 1999, 2000 Ingo Molnar, Red Hat |
1da177e4 LT |
7 | |
8 | RAID-0 management functions. | |
9 | ||
10 | This program is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation; either version 2, or (at your option) | |
13 | any later version. | |
f72ffdd6 | 14 | |
1da177e4 LT |
15 | You should have received a copy of the GNU General Public License |
16 | (for example /usr/src/linux/COPYING); if not, write to the Free | |
f72ffdd6 | 17 | Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
1da177e4 LT |
18 | */ |
19 | ||
bff61975 | 20 | #include <linux/blkdev.h> |
bff61975 | 21 | #include <linux/seq_file.h> |
056075c7 | 22 | #include <linux/module.h> |
5a0e3ad6 | 23 | #include <linux/slab.h> |
43b2e5d8 | 24 | #include "md.h" |
ef740c37 | 25 | #include "raid0.h" |
9af204cf | 26 | #include "raid5.h" |
1da177e4 | 27 | |
5c675f83 | 28 | static int raid0_congested(struct mddev *mddev, int bits) |
26be34dc | 29 | { |
e373ab10 | 30 | struct r0conf *conf = mddev->private; |
3cb03002 | 31 | struct md_rdev **devlist = conf->devlist; |
84707f38 | 32 | int raid_disks = conf->strip_zone[0].nb_dev; |
26be34dc N |
33 | int i, ret = 0; |
34 | ||
84707f38 | 35 | for (i = 0; i < raid_disks && !ret ; i++) { |
165125e1 | 36 | struct request_queue *q = bdev_get_queue(devlist[i]->bdev); |
26be34dc N |
37 | |
38 | ret |= bdi_congested(&q->backing_dev_info, bits); | |
39 | } | |
40 | return ret; | |
41 | } | |
42 | ||
46994191 | 43 | /* |
44 | * inform the user of the raid configuration | |
45 | */ | |
fd01b88c | 46 | static void dump_zones(struct mddev *mddev) |
46994191 | 47 | { |
50de8df4 | 48 | int j, k; |
46994191 | 49 | sector_t zone_size = 0; |
50 | sector_t zone_start = 0; | |
51 | char b[BDEVNAME_SIZE]; | |
e373ab10 | 52 | struct r0conf *conf = mddev->private; |
84707f38 | 53 | int raid_disks = conf->strip_zone[0].nb_dev; |
50de8df4 N |
54 | printk(KERN_INFO "md: RAID0 configuration for %s - %d zone%s\n", |
55 | mdname(mddev), | |
56 | conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s"); | |
46994191 | 57 | for (j = 0; j < conf->nr_strip_zones; j++) { |
50de8df4 | 58 | printk(KERN_INFO "md: zone%d=[", j); |
46994191 | 59 | for (k = 0; k < conf->strip_zone[j].nb_dev; k++) |
50de8df4 | 60 | printk(KERN_CONT "%s%s", k?"/":"", |
84707f38 | 61 | bdevname(conf->devlist[j*raid_disks |
46994191 | 62 | + k]->bdev, b)); |
b5a20961 | 63 | printk(KERN_CONT "]\n"); |
46994191 | 64 | |
65 | zone_size = conf->strip_zone[j].zone_end - zone_start; | |
50de8df4 N |
66 | printk(KERN_INFO " zone-offset=%10lluKB, " |
67 | "device-offset=%10lluKB, size=%10lluKB\n", | |
46994191 | 68 | (unsigned long long)zone_start>>1, |
69 | (unsigned long long)conf->strip_zone[j].dev_start>>1, | |
70 | (unsigned long long)zone_size>>1); | |
71 | zone_start = conf->strip_zone[j].zone_end; | |
72 | } | |
50de8df4 | 73 | printk(KERN_INFO "\n"); |
46994191 | 74 | } |
75 | ||
e373ab10 | 76 | static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf) |
1da177e4 | 77 | { |
a9f326eb | 78 | int i, c, err; |
49f357a2 | 79 | sector_t curr_zone_end, sectors; |
3cb03002 | 80 | struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev; |
1da177e4 LT |
81 | struct strip_zone *zone; |
82 | int cnt; | |
83 | char b[BDEVNAME_SIZE]; | |
50de8df4 | 84 | char b2[BDEVNAME_SIZE]; |
e373ab10 | 85 | struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL); |
c83057a1 | 86 | bool discard_supported = false; |
ed7b0038 AN |
87 | |
88 | if (!conf) | |
89 | return -ENOMEM; | |
dafb20fa | 90 | rdev_for_each(rdev1, mddev) { |
50de8df4 N |
91 | pr_debug("md/raid0:%s: looking at %s\n", |
92 | mdname(mddev), | |
93 | bdevname(rdev1->bdev, b)); | |
1da177e4 | 94 | c = 0; |
13f2682b N |
95 | |
96 | /* round size to chunk_size */ | |
97 | sectors = rdev1->sectors; | |
98 | sector_div(sectors, mddev->chunk_sectors); | |
99 | rdev1->sectors = sectors * mddev->chunk_sectors; | |
100 | ||
dafb20fa | 101 | rdev_for_each(rdev2, mddev) { |
50de8df4 N |
102 | pr_debug("md/raid0:%s: comparing %s(%llu)" |
103 | " with %s(%llu)\n", | |
104 | mdname(mddev), | |
105 | bdevname(rdev1->bdev,b), | |
106 | (unsigned long long)rdev1->sectors, | |
107 | bdevname(rdev2->bdev,b2), | |
108 | (unsigned long long)rdev2->sectors); | |
1da177e4 | 109 | if (rdev2 == rdev1) { |
50de8df4 N |
110 | pr_debug("md/raid0:%s: END\n", |
111 | mdname(mddev)); | |
1da177e4 LT |
112 | break; |
113 | } | |
dd8ac336 | 114 | if (rdev2->sectors == rdev1->sectors) { |
1da177e4 LT |
115 | /* |
116 | * Not unique, don't count it as a new | |
117 | * group | |
118 | */ | |
50de8df4 N |
119 | pr_debug("md/raid0:%s: EQUAL\n", |
120 | mdname(mddev)); | |
1da177e4 LT |
121 | c = 1; |
122 | break; | |
123 | } | |
50de8df4 N |
124 | pr_debug("md/raid0:%s: NOT EQUAL\n", |
125 | mdname(mddev)); | |
1da177e4 LT |
126 | } |
127 | if (!c) { | |
50de8df4 N |
128 | pr_debug("md/raid0:%s: ==> UNIQUE\n", |
129 | mdname(mddev)); | |
1da177e4 | 130 | conf->nr_strip_zones++; |
50de8df4 N |
131 | pr_debug("md/raid0:%s: %d zones\n", |
132 | mdname(mddev), conf->nr_strip_zones); | |
1da177e4 LT |
133 | } |
134 | } | |
50de8df4 N |
135 | pr_debug("md/raid0:%s: FINAL %d zones\n", |
136 | mdname(mddev), conf->nr_strip_zones); | |
ed7b0038 | 137 | err = -ENOMEM; |
9ffae0cf | 138 | conf->strip_zone = kzalloc(sizeof(struct strip_zone)* |
1da177e4 LT |
139 | conf->nr_strip_zones, GFP_KERNEL); |
140 | if (!conf->strip_zone) | |
ed7b0038 | 141 | goto abort; |
3cb03002 | 142 | conf->devlist = kzalloc(sizeof(struct md_rdev*)* |
1da177e4 LT |
143 | conf->nr_strip_zones*mddev->raid_disks, |
144 | GFP_KERNEL); | |
145 | if (!conf->devlist) | |
ed7b0038 | 146 | goto abort; |
1da177e4 | 147 | |
1da177e4 LT |
148 | /* The first zone must contain all devices, so here we check that |
149 | * there is a proper alignment of slots to devices and find them all | |
150 | */ | |
151 | zone = &conf->strip_zone[0]; | |
152 | cnt = 0; | |
153 | smallest = NULL; | |
b414579f | 154 | dev = conf->devlist; |
ed7b0038 | 155 | err = -EINVAL; |
dafb20fa | 156 | rdev_for_each(rdev1, mddev) { |
1da177e4 LT |
157 | int j = rdev1->raid_disk; |
158 | ||
e93f68a1 | 159 | if (mddev->level == 10) { |
9af204cf TM |
160 | /* taking over a raid10-n2 array */ |
161 | j /= 2; | |
e93f68a1 N |
162 | rdev1->new_raid_disk = j; |
163 | } | |
9af204cf | 164 | |
fc3a08b8 KW |
165 | if (mddev->level == 1) { |
166 | /* taiking over a raid1 array- | |
167 | * we have only one active disk | |
168 | */ | |
169 | j = 0; | |
170 | rdev1->new_raid_disk = j; | |
171 | } | |
172 | ||
f96c9f30 N |
173 | if (j < 0) { |
174 | printk(KERN_ERR | |
175 | "md/raid0:%s: remove inactive devices before converting to RAID0\n", | |
176 | mdname(mddev)); | |
177 | goto abort; | |
178 | } | |
179 | if (j >= mddev->raid_disks) { | |
b5a20961 N |
180 | printk(KERN_ERR "md/raid0:%s: bad disk number %d - " |
181 | "aborting!\n", mdname(mddev), j); | |
1da177e4 LT |
182 | goto abort; |
183 | } | |
b414579f | 184 | if (dev[j]) { |
b5a20961 N |
185 | printk(KERN_ERR "md/raid0:%s: multiple devices for %d - " |
186 | "aborting!\n", mdname(mddev), j); | |
1da177e4 LT |
187 | goto abort; |
188 | } | |
b414579f | 189 | dev[j] = rdev1; |
1da177e4 | 190 | |
8f6c2e4b MP |
191 | disk_stack_limits(mddev->gendisk, rdev1->bdev, |
192 | rdev1->data_offset << 9); | |
1da177e4 | 193 | |
ba13da47 N |
194 | if (rdev1->bdev->bd_disk->queue->merge_bvec_fn) |
195 | conf->has_merge_bvec = 1; | |
196 | ||
dd8ac336 | 197 | if (!smallest || (rdev1->sectors < smallest->sectors)) |
1da177e4 LT |
198 | smallest = rdev1; |
199 | cnt++; | |
c83057a1 SL |
200 | |
201 | if (blk_queue_discard(bdev_get_queue(rdev1->bdev))) | |
202 | discard_supported = true; | |
1da177e4 LT |
203 | } |
204 | if (cnt != mddev->raid_disks) { | |
b5a20961 N |
205 | printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - " |
206 | "aborting!\n", mdname(mddev), cnt, mddev->raid_disks); | |
1da177e4 LT |
207 | goto abort; |
208 | } | |
209 | zone->nb_dev = cnt; | |
49f357a2 | 210 | zone->zone_end = smallest->sectors * cnt; |
1da177e4 | 211 | |
49f357a2 | 212 | curr_zone_end = zone->zone_end; |
1da177e4 LT |
213 | |
214 | /* now do the other zones */ | |
215 | for (i = 1; i < conf->nr_strip_zones; i++) | |
216 | { | |
a9f326eb N |
217 | int j; |
218 | ||
1da177e4 | 219 | zone = conf->strip_zone + i; |
b414579f | 220 | dev = conf->devlist + i * mddev->raid_disks; |
1da177e4 | 221 | |
50de8df4 | 222 | pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i); |
d27a43ab | 223 | zone->dev_start = smallest->sectors; |
1da177e4 LT |
224 | smallest = NULL; |
225 | c = 0; | |
226 | ||
227 | for (j=0; j<cnt; j++) { | |
b414579f | 228 | rdev = conf->devlist[j]; |
d27a43ab | 229 | if (rdev->sectors <= zone->dev_start) { |
50de8df4 N |
230 | pr_debug("md/raid0:%s: checking %s ... nope\n", |
231 | mdname(mddev), | |
232 | bdevname(rdev->bdev, b)); | |
dd8ac336 AN |
233 | continue; |
234 | } | |
50de8df4 N |
235 | pr_debug("md/raid0:%s: checking %s ..." |
236 | " contained as device %d\n", | |
237 | mdname(mddev), | |
238 | bdevname(rdev->bdev, b), c); | |
b414579f | 239 | dev[c] = rdev; |
dd8ac336 AN |
240 | c++; |
241 | if (!smallest || rdev->sectors < smallest->sectors) { | |
242 | smallest = rdev; | |
50de8df4 N |
243 | pr_debug("md/raid0:%s: (%llu) is smallest!.\n", |
244 | mdname(mddev), | |
245 | (unsigned long long)rdev->sectors); | |
dd8ac336 | 246 | } |
1da177e4 LT |
247 | } |
248 | ||
249 | zone->nb_dev = c; | |
49f357a2 | 250 | sectors = (smallest->sectors - zone->dev_start) * c; |
50de8df4 N |
251 | pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n", |
252 | mdname(mddev), | |
253 | zone->nb_dev, (unsigned long long)sectors); | |
1da177e4 | 254 | |
49f357a2 | 255 | curr_zone_end += sectors; |
d27a43ab | 256 | zone->zone_end = curr_zone_end; |
1da177e4 | 257 | |
50de8df4 N |
258 | pr_debug("md/raid0:%s: current zone start: %llu\n", |
259 | mdname(mddev), | |
260 | (unsigned long long)smallest->sectors); | |
1da177e4 | 261 | } |
1da177e4 | 262 | |
92e59b6b | 263 | /* |
264 | * now since we have the hard sector sizes, we can make sure | |
265 | * chunk size is a multiple of that sector size | |
266 | */ | |
9d8f0363 | 267 | if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) { |
b5a20961 | 268 | printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n", |
92e59b6b | 269 | mdname(mddev), |
9d8f0363 | 270 | mddev->chunk_sectors << 9); |
92e59b6b | 271 | goto abort; |
272 | } | |
8f6c2e4b MP |
273 | |
274 | blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9); | |
275 | blk_queue_io_opt(mddev->queue, | |
276 | (mddev->chunk_sectors << 9) * mddev->raid_disks); | |
277 | ||
c83057a1 SL |
278 | if (!discard_supported) |
279 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); | |
280 | else | |
281 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); | |
282 | ||
50de8df4 | 283 | pr_debug("md/raid0:%s: done.\n", mdname(mddev)); |
9af204cf TM |
284 | *private_conf = conf; |
285 | ||
1da177e4 | 286 | return 0; |
5568a603 | 287 | abort: |
ed7b0038 AN |
288 | kfree(conf->strip_zone); |
289 | kfree(conf->devlist); | |
290 | kfree(conf); | |
58ebb34c | 291 | *private_conf = ERR_PTR(err); |
ed7b0038 | 292 | return err; |
1da177e4 LT |
293 | } |
294 | ||
ba13da47 N |
295 | /* Find the zone which holds a particular offset |
296 | * Update *sectorp to be an offset in that zone | |
297 | */ | |
298 | static struct strip_zone *find_zone(struct r0conf *conf, | |
299 | sector_t *sectorp) | |
300 | { | |
301 | int i; | |
302 | struct strip_zone *z = conf->strip_zone; | |
303 | sector_t sector = *sectorp; | |
304 | ||
305 | for (i = 0; i < conf->nr_strip_zones; i++) | |
306 | if (sector < z[i].zone_end) { | |
307 | if (i) | |
308 | *sectorp = sector - z[i-1].zone_end; | |
309 | return z + i; | |
310 | } | |
311 | BUG(); | |
312 | } | |
313 | ||
314 | /* | |
315 | * remaps the bio to the target device. we separate two flows. | |
316 | * power 2 flow and a general flow for the sake of perfromance | |
317 | */ | |
318 | static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone, | |
319 | sector_t sector, sector_t *sector_offset) | |
320 | { | |
321 | unsigned int sect_in_chunk; | |
322 | sector_t chunk; | |
323 | struct r0conf *conf = mddev->private; | |
324 | int raid_disks = conf->strip_zone[0].nb_dev; | |
325 | unsigned int chunk_sects = mddev->chunk_sectors; | |
326 | ||
327 | if (is_power_of_2(chunk_sects)) { | |
328 | int chunksect_bits = ffz(~chunk_sects); | |
329 | /* find the sector offset inside the chunk */ | |
330 | sect_in_chunk = sector & (chunk_sects - 1); | |
331 | sector >>= chunksect_bits; | |
332 | /* chunk in zone */ | |
333 | chunk = *sector_offset; | |
334 | /* quotient is the chunk in real device*/ | |
335 | sector_div(chunk, zone->nb_dev << chunksect_bits); | |
336 | } else{ | |
337 | sect_in_chunk = sector_div(sector, chunk_sects); | |
338 | chunk = *sector_offset; | |
339 | sector_div(chunk, chunk_sects * zone->nb_dev); | |
340 | } | |
341 | /* | |
342 | * position the bio over the real device | |
343 | * real sector = chunk in device + starting of zone | |
344 | * + the position in the chunk | |
345 | */ | |
346 | *sector_offset = (chunk * chunk_sects) + sect_in_chunk; | |
347 | return conf->devlist[(zone - conf->strip_zone)*raid_disks | |
348 | + sector_div(sector, zone->nb_dev)]; | |
349 | } | |
350 | ||
1da177e4 | 351 | /** |
ba13da47 | 352 | * raid0_mergeable_bvec -- tell bio layer if two requests can be merged |
64590f45 | 353 | * @mddev: the md device |
cc371e66 | 354 | * @bvm: properties of new bio |
1da177e4 LT |
355 | * @biovec: the request that could be merged to it. |
356 | * | |
357 | * Return amount of bytes we can accept at this offset | |
358 | */ | |
64590f45 | 359 | static int raid0_mergeable_bvec(struct mddev *mddev, |
cc371e66 AK |
360 | struct bvec_merge_data *bvm, |
361 | struct bio_vec *biovec) | |
1da177e4 | 362 | { |
ba13da47 | 363 | struct r0conf *conf = mddev->private; |
cc371e66 | 364 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
ba13da47 | 365 | sector_t sector_offset = sector; |
1da177e4 | 366 | int max; |
9d8f0363 | 367 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 368 | unsigned int bio_sectors = bvm->bi_size >> 9; |
ba13da47 N |
369 | struct strip_zone *zone; |
370 | struct md_rdev *rdev; | |
371 | struct request_queue *subq; | |
1da177e4 | 372 | |
d6e412ea | 373 | if (is_power_of_2(chunk_sectors)) |
fbb704ef | 374 | max = (chunk_sectors - ((sector & (chunk_sectors-1)) |
375 | + bio_sectors)) << 9; | |
376 | else | |
377 | max = (chunk_sectors - (sector_div(sector, chunk_sectors) | |
378 | + bio_sectors)) << 9; | |
ba13da47 N |
379 | if (max < 0) |
380 | max = 0; /* bio_add cannot handle a negative return */ | |
1da177e4 LT |
381 | if (max <= biovec->bv_len && bio_sectors == 0) |
382 | return biovec->bv_len; | |
ba13da47 N |
383 | if (max < biovec->bv_len) |
384 | /* too small already, no need to check further */ | |
385 | return max; | |
386 | if (!conf->has_merge_bvec) | |
387 | return max; | |
388 | ||
389 | /* May need to check subordinate device */ | |
390 | sector = sector_offset; | |
391 | zone = find_zone(mddev->private, §or_offset); | |
392 | rdev = map_sector(mddev, zone, sector, §or_offset); | |
393 | subq = bdev_get_queue(rdev->bdev); | |
394 | if (subq->merge_bvec_fn) { | |
395 | bvm->bi_bdev = rdev->bdev; | |
396 | bvm->bi_sector = sector_offset + zone->dev_start + | |
397 | rdev->data_offset; | |
398 | return min(max, subq->merge_bvec_fn(subq, bvm, biovec)); | |
399 | } else | |
1da177e4 LT |
400 | return max; |
401 | } | |
402 | ||
fd01b88c | 403 | static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
404 | { |
405 | sector_t array_sectors = 0; | |
3cb03002 | 406 | struct md_rdev *rdev; |
80c3a6ce DW |
407 | |
408 | WARN_ONCE(sectors || raid_disks, | |
409 | "%s does not support generic reshape\n", __func__); | |
410 | ||
dafb20fa | 411 | rdev_for_each(rdev, mddev) |
a6468539 N |
412 | array_sectors += (rdev->sectors & |
413 | ~(sector_t)(mddev->chunk_sectors-1)); | |
80c3a6ce DW |
414 | |
415 | return array_sectors; | |
416 | } | |
417 | ||
afa0f557 | 418 | static void raid0_free(struct mddev *mddev, void *priv); |
0366ef84 | 419 | |
fd01b88c | 420 | static int raid0_run(struct mddev *mddev) |
1da177e4 | 421 | { |
e373ab10 | 422 | struct r0conf *conf; |
5568a603 | 423 | int ret; |
1da177e4 | 424 | |
9d8f0363 | 425 | if (mddev->chunk_sectors == 0) { |
b5a20961 N |
426 | printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n", |
427 | mdname(mddev)); | |
2604b703 N |
428 | return -EINVAL; |
429 | } | |
0894cc30 AN |
430 | if (md_check_no_bitmap(mddev)) |
431 | return -EINVAL; | |
086fa5ff | 432 | blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors); |
4363ac7c | 433 | blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors); |
c83057a1 | 434 | blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors); |
1da177e4 | 435 | |
9af204cf TM |
436 | /* if private is not null, we are here after takeover */ |
437 | if (mddev->private == NULL) { | |
438 | ret = create_strip_zones(mddev, &conf); | |
439 | if (ret < 0) | |
440 | return ret; | |
441 | mddev->private = conf; | |
442 | } | |
443 | conf = mddev->private; | |
1da177e4 LT |
444 | |
445 | /* calculate array device size */ | |
1f403624 | 446 | md_set_array_sectors(mddev, raid0_size(mddev, 0, 0)); |
1da177e4 | 447 | |
b5a20961 N |
448 | printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n", |
449 | mdname(mddev), | |
450 | (unsigned long long)mddev->array_sectors); | |
1da177e4 LT |
451 | /* calculate the max read-ahead size. |
452 | * For read-ahead of large files to be effective, we need to | |
453 | * readahead at least twice a whole stripe. i.e. number of devices | |
454 | * multiplied by chunk size times 2. | |
455 | * If an individual device has an ra_pages greater than the | |
456 | * chunk size, then we will not drive that device as hard as it | |
457 | * wants. We consider this a configuration error: a larger | |
458 | * chunksize should be used in that case. | |
459 | */ | |
460 | { | |
9d8f0363 AN |
461 | int stripe = mddev->raid_disks * |
462 | (mddev->chunk_sectors << 9) / PAGE_SIZE; | |
1da177e4 LT |
463 | if (mddev->queue->backing_dev_info.ra_pages < 2* stripe) |
464 | mddev->queue->backing_dev_info.ra_pages = 2* stripe; | |
465 | } | |
466 | ||
46994191 | 467 | dump_zones(mddev); |
0366ef84 | 468 | |
469 | ret = md_integrity_register(mddev); | |
470 | if (ret) | |
afa0f557 | 471 | raid0_free(mddev, conf); |
0366ef84 | 472 | |
473 | return ret; | |
1da177e4 LT |
474 | } |
475 | ||
afa0f557 | 476 | static void raid0_free(struct mddev *mddev, void *priv) |
1da177e4 | 477 | { |
afa0f557 | 478 | struct r0conf *conf = priv; |
1da177e4 | 479 | |
990a8baf | 480 | kfree(conf->strip_zone); |
fb5ab4b5 | 481 | kfree(conf->devlist); |
990a8baf | 482 | kfree(conf); |
1da177e4 LT |
483 | } |
484 | ||
fbb704ef | 485 | /* |
486 | * Is io distribute over 1 or more chunks ? | |
487 | */ | |
fd01b88c | 488 | static inline int is_io_in_chunk_boundary(struct mddev *mddev, |
fbb704ef | 489 | unsigned int chunk_sects, struct bio *bio) |
490 | { | |
d6e412ea | 491 | if (likely(is_power_of_2(chunk_sects))) { |
4f024f37 KO |
492 | return chunk_sects >= |
493 | ((bio->bi_iter.bi_sector & (chunk_sects-1)) | |
aa8b57aa | 494 | + bio_sectors(bio)); |
fbb704ef | 495 | } else{ |
4f024f37 | 496 | sector_t sector = bio->bi_iter.bi_sector; |
fbb704ef | 497 | return chunk_sects >= (sector_div(sector, chunk_sects) |
aa8b57aa | 498 | + bio_sectors(bio)); |
fbb704ef | 499 | } |
500 | } | |
501 | ||
b4fdcb02 | 502 | static void raid0_make_request(struct mddev *mddev, struct bio *bio) |
fbb704ef | 503 | { |
1da177e4 | 504 | struct strip_zone *zone; |
3cb03002 | 505 | struct md_rdev *tmp_dev; |
20d0189b | 506 | struct bio *split; |
1da177e4 | 507 | |
e9c7469b TH |
508 | if (unlikely(bio->bi_rw & REQ_FLUSH)) { |
509 | md_flush_request(mddev, bio); | |
5a7bbad2 | 510 | return; |
e5dcdd80 N |
511 | } |
512 | ||
20d0189b | 513 | do { |
4f024f37 | 514 | sector_t sector = bio->bi_iter.bi_sector; |
20d0189b KO |
515 | unsigned chunk_sects = mddev->chunk_sectors; |
516 | ||
517 | unsigned sectors = chunk_sects - | |
518 | (likely(is_power_of_2(chunk_sects)) | |
519 | ? (sector & (chunk_sects-1)) | |
520 | : sector_div(sector, chunk_sects)); | |
521 | ||
522 | if (sectors < bio_sectors(bio)) { | |
523 | split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set); | |
524 | bio_chain(split, bio); | |
525 | } else { | |
526 | split = bio; | |
527 | } | |
1da177e4 | 528 | |
20d0189b KO |
529 | zone = find_zone(mddev->private, §or); |
530 | tmp_dev = map_sector(mddev, zone, sector, §or); | |
531 | split->bi_bdev = tmp_dev->bdev; | |
532 | split->bi_iter.bi_sector = sector + zone->dev_start + | |
533 | tmp_dev->data_offset; | |
534 | ||
535 | if (unlikely((split->bi_rw & REQ_DISCARD) && | |
536 | !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) { | |
537 | /* Just ignore it */ | |
538 | bio_endio(split, 0); | |
539 | } else | |
540 | generic_make_request(split); | |
541 | } while (split != bio); | |
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 | } | |
eea136d6 | 569 | rdev->sectors = mddev->dev_sectors; |
9af204cf TM |
570 | } |
571 | ||
572 | /* Set new parameters */ | |
573 | mddev->new_level = 0; | |
001048a3 | 574 | mddev->new_layout = 0; |
9af204cf TM |
575 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
576 | mddev->raid_disks--; | |
577 | mddev->delta_disks = -1; | |
578 | /* make sure it will be not marked as dirty */ | |
579 | mddev->recovery_cp = MaxSector; | |
580 | ||
581 | create_strip_zones(mddev, &priv_conf); | |
582 | return priv_conf; | |
583 | } | |
584 | ||
fd01b88c | 585 | static void *raid0_takeover_raid10(struct mddev *mddev) |
9af204cf | 586 | { |
e373ab10 | 587 | struct r0conf *priv_conf; |
9af204cf TM |
588 | |
589 | /* Check layout: | |
590 | * - far_copies must be 1 | |
591 | * - near_copies must be 2 | |
592 | * - disks number must be even | |
593 | * - all mirrors must be already degraded | |
594 | */ | |
595 | if (mddev->layout != ((1 << 8) + 2)) { | |
b5a20961 N |
596 | printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n", |
597 | mdname(mddev), | |
9af204cf TM |
598 | mddev->layout); |
599 | return ERR_PTR(-EINVAL); | |
600 | } | |
601 | if (mddev->raid_disks & 1) { | |
b5a20961 N |
602 | printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n", |
603 | mdname(mddev)); | |
9af204cf TM |
604 | return ERR_PTR(-EINVAL); |
605 | } | |
606 | if (mddev->degraded != (mddev->raid_disks>>1)) { | |
b5a20961 N |
607 | printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n", |
608 | mdname(mddev)); | |
9af204cf TM |
609 | return ERR_PTR(-EINVAL); |
610 | } | |
611 | ||
612 | /* Set new parameters */ | |
613 | mddev->new_level = 0; | |
001048a3 | 614 | mddev->new_layout = 0; |
9af204cf TM |
615 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
616 | mddev->delta_disks = - mddev->raid_disks / 2; | |
617 | mddev->raid_disks += mddev->delta_disks; | |
618 | mddev->degraded = 0; | |
619 | /* make sure it will be not marked as dirty */ | |
620 | mddev->recovery_cp = MaxSector; | |
621 | ||
622 | create_strip_zones(mddev, &priv_conf); | |
9af204cf TM |
623 | return priv_conf; |
624 | } | |
625 | ||
fd01b88c | 626 | static void *raid0_takeover_raid1(struct mddev *mddev) |
fc3a08b8 | 627 | { |
e373ab10 | 628 | struct r0conf *priv_conf; |
24b961f8 | 629 | int chunksect; |
fc3a08b8 KW |
630 | |
631 | /* Check layout: | |
632 | * - (N - 1) mirror drives must be already faulty | |
633 | */ | |
634 | if ((mddev->raid_disks - 1) != mddev->degraded) { | |
635 | printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n", | |
636 | mdname(mddev)); | |
637 | return ERR_PTR(-EINVAL); | |
638 | } | |
639 | ||
24b961f8 JS |
640 | /* |
641 | * a raid1 doesn't have the notion of chunk size, so | |
642 | * figure out the largest suitable size we can use. | |
643 | */ | |
644 | chunksect = 64 * 2; /* 64K by default */ | |
645 | ||
646 | /* The array must be an exact multiple of chunksize */ | |
647 | while (chunksect && (mddev->array_sectors & (chunksect - 1))) | |
648 | chunksect >>= 1; | |
649 | ||
650 | if ((chunksect << 9) < PAGE_SIZE) | |
651 | /* array size does not allow a suitable chunk size */ | |
652 | return ERR_PTR(-EINVAL); | |
653 | ||
fc3a08b8 KW |
654 | /* Set new parameters */ |
655 | mddev->new_level = 0; | |
656 | mddev->new_layout = 0; | |
24b961f8 JS |
657 | mddev->new_chunk_sectors = chunksect; |
658 | mddev->chunk_sectors = chunksect; | |
fc3a08b8 | 659 | mddev->delta_disks = 1 - mddev->raid_disks; |
f7bee809 | 660 | mddev->raid_disks = 1; |
fc3a08b8 KW |
661 | /* make sure it will be not marked as dirty */ |
662 | mddev->recovery_cp = MaxSector; | |
663 | ||
664 | create_strip_zones(mddev, &priv_conf); | |
665 | return priv_conf; | |
666 | } | |
667 | ||
fd01b88c | 668 | static void *raid0_takeover(struct mddev *mddev) |
9af204cf TM |
669 | { |
670 | /* raid0 can take over: | |
049d6c1e | 671 | * raid4 - if all data disks are active. |
9af204cf TM |
672 | * raid5 - providing it is Raid4 layout and one disk is faulty |
673 | * raid10 - assuming we have all necessary active disks | |
fc3a08b8 | 674 | * raid1 - with (N -1) mirror drives faulty |
9af204cf | 675 | */ |
a8461a61 N |
676 | |
677 | if (mddev->bitmap) { | |
678 | printk(KERN_ERR "md/raid0: %s: cannot takeover array with bitmap\n", | |
679 | mdname(mddev)); | |
680 | return ERR_PTR(-EBUSY); | |
681 | } | |
049d6c1e MT |
682 | if (mddev->level == 4) |
683 | return raid0_takeover_raid45(mddev); | |
684 | ||
9af204cf TM |
685 | if (mddev->level == 5) { |
686 | if (mddev->layout == ALGORITHM_PARITY_N) | |
049d6c1e | 687 | return raid0_takeover_raid45(mddev); |
9af204cf | 688 | |
b5a20961 N |
689 | printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n", |
690 | mdname(mddev), ALGORITHM_PARITY_N); | |
9af204cf TM |
691 | } |
692 | ||
693 | if (mddev->level == 10) | |
694 | return raid0_takeover_raid10(mddev); | |
695 | ||
fc3a08b8 KW |
696 | if (mddev->level == 1) |
697 | return raid0_takeover_raid1(mddev); | |
698 | ||
699 | printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n", | |
700 | mddev->level); | |
701 | ||
9af204cf TM |
702 | return ERR_PTR(-EINVAL); |
703 | } | |
704 | ||
fd01b88c | 705 | static void raid0_quiesce(struct mddev *mddev, int state) |
9af204cf TM |
706 | { |
707 | } | |
708 | ||
84fc4b56 | 709 | static struct md_personality raid0_personality= |
1da177e4 LT |
710 | { |
711 | .name = "raid0", | |
2604b703 | 712 | .level = 0, |
1da177e4 LT |
713 | .owner = THIS_MODULE, |
714 | .make_request = raid0_make_request, | |
715 | .run = raid0_run, | |
afa0f557 | 716 | .free = raid0_free, |
1da177e4 | 717 | .status = raid0_status, |
80c3a6ce | 718 | .size = raid0_size, |
9af204cf TM |
719 | .takeover = raid0_takeover, |
720 | .quiesce = raid0_quiesce, | |
5c675f83 | 721 | .congested = raid0_congested, |
64590f45 | 722 | .mergeable_bvec = raid0_mergeable_bvec, |
1da177e4 LT |
723 | }; |
724 | ||
725 | static int __init raid0_init (void) | |
726 | { | |
2604b703 | 727 | return register_md_personality (&raid0_personality); |
1da177e4 LT |
728 | } |
729 | ||
730 | static void raid0_exit (void) | |
731 | { | |
2604b703 | 732 | unregister_md_personality (&raid0_personality); |
1da177e4 LT |
733 | } |
734 | ||
735 | module_init(raid0_init); | |
736 | module_exit(raid0_exit); | |
737 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 738 | MODULE_DESCRIPTION("RAID0 (striping) personality for MD"); |
1da177e4 | 739 | MODULE_ALIAS("md-personality-2"); /* RAID0 */ |
d9d166c2 | 740 | MODULE_ALIAS("md-raid0"); |
2604b703 | 741 | MODULE_ALIAS("md-level-0"); |