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 | 273 | |
753f2856 HM |
274 | if (mddev->queue) { |
275 | blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9); | |
276 | blk_queue_io_opt(mddev->queue, | |
277 | (mddev->chunk_sectors << 9) * mddev->raid_disks); | |
278 | ||
279 | if (!discard_supported) | |
280 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); | |
281 | else | |
282 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); | |
283 | } | |
c83057a1 | 284 | |
50de8df4 | 285 | pr_debug("md/raid0:%s: done.\n", mdname(mddev)); |
9af204cf TM |
286 | *private_conf = conf; |
287 | ||
1da177e4 | 288 | return 0; |
5568a603 | 289 | abort: |
ed7b0038 AN |
290 | kfree(conf->strip_zone); |
291 | kfree(conf->devlist); | |
292 | kfree(conf); | |
58ebb34c | 293 | *private_conf = ERR_PTR(err); |
ed7b0038 | 294 | return err; |
1da177e4 LT |
295 | } |
296 | ||
ba13da47 N |
297 | /* Find the zone which holds a particular offset |
298 | * Update *sectorp to be an offset in that zone | |
299 | */ | |
300 | static struct strip_zone *find_zone(struct r0conf *conf, | |
301 | sector_t *sectorp) | |
302 | { | |
303 | int i; | |
304 | struct strip_zone *z = conf->strip_zone; | |
305 | sector_t sector = *sectorp; | |
306 | ||
307 | for (i = 0; i < conf->nr_strip_zones; i++) | |
308 | if (sector < z[i].zone_end) { | |
309 | if (i) | |
310 | *sectorp = sector - z[i-1].zone_end; | |
311 | return z + i; | |
312 | } | |
313 | BUG(); | |
314 | } | |
315 | ||
316 | /* | |
317 | * remaps the bio to the target device. we separate two flows. | |
47d68979 | 318 | * power 2 flow and a general flow for the sake of performance |
ba13da47 N |
319 | */ |
320 | static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone, | |
321 | sector_t sector, sector_t *sector_offset) | |
322 | { | |
323 | unsigned int sect_in_chunk; | |
324 | sector_t chunk; | |
325 | struct r0conf *conf = mddev->private; | |
326 | int raid_disks = conf->strip_zone[0].nb_dev; | |
327 | unsigned int chunk_sects = mddev->chunk_sectors; | |
328 | ||
329 | if (is_power_of_2(chunk_sects)) { | |
330 | int chunksect_bits = ffz(~chunk_sects); | |
331 | /* find the sector offset inside the chunk */ | |
332 | sect_in_chunk = sector & (chunk_sects - 1); | |
333 | sector >>= chunksect_bits; | |
334 | /* chunk in zone */ | |
335 | chunk = *sector_offset; | |
336 | /* quotient is the chunk in real device*/ | |
337 | sector_div(chunk, zone->nb_dev << chunksect_bits); | |
338 | } else{ | |
339 | sect_in_chunk = sector_div(sector, chunk_sects); | |
340 | chunk = *sector_offset; | |
341 | sector_div(chunk, chunk_sects * zone->nb_dev); | |
342 | } | |
343 | /* | |
344 | * position the bio over the real device | |
345 | * real sector = chunk in device + starting of zone | |
346 | * + the position in the chunk | |
347 | */ | |
348 | *sector_offset = (chunk * chunk_sects) + sect_in_chunk; | |
349 | return conf->devlist[(zone - conf->strip_zone)*raid_disks | |
350 | + sector_div(sector, zone->nb_dev)]; | |
351 | } | |
352 | ||
1da177e4 | 353 | /** |
ba13da47 | 354 | * raid0_mergeable_bvec -- tell bio layer if two requests can be merged |
64590f45 | 355 | * @mddev: the md device |
cc371e66 | 356 | * @bvm: properties of new bio |
1da177e4 LT |
357 | * @biovec: the request that could be merged to it. |
358 | * | |
359 | * Return amount of bytes we can accept at this offset | |
360 | */ | |
64590f45 | 361 | static int raid0_mergeable_bvec(struct mddev *mddev, |
cc371e66 AK |
362 | struct bvec_merge_data *bvm, |
363 | struct bio_vec *biovec) | |
1da177e4 | 364 | { |
ba13da47 | 365 | struct r0conf *conf = mddev->private; |
cc371e66 | 366 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
ba13da47 | 367 | sector_t sector_offset = sector; |
1da177e4 | 368 | int max; |
9d8f0363 | 369 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 370 | unsigned int bio_sectors = bvm->bi_size >> 9; |
ba13da47 N |
371 | struct strip_zone *zone; |
372 | struct md_rdev *rdev; | |
373 | struct request_queue *subq; | |
1da177e4 | 374 | |
d6e412ea | 375 | if (is_power_of_2(chunk_sectors)) |
fbb704ef | 376 | max = (chunk_sectors - ((sector & (chunk_sectors-1)) |
377 | + bio_sectors)) << 9; | |
378 | else | |
379 | max = (chunk_sectors - (sector_div(sector, chunk_sectors) | |
380 | + bio_sectors)) << 9; | |
ba13da47 N |
381 | if (max < 0) |
382 | max = 0; /* bio_add cannot handle a negative return */ | |
1da177e4 LT |
383 | if (max <= biovec->bv_len && bio_sectors == 0) |
384 | return biovec->bv_len; | |
ba13da47 N |
385 | if (max < biovec->bv_len) |
386 | /* too small already, no need to check further */ | |
387 | return max; | |
388 | if (!conf->has_merge_bvec) | |
389 | return max; | |
390 | ||
391 | /* May need to check subordinate device */ | |
392 | sector = sector_offset; | |
393 | zone = find_zone(mddev->private, §or_offset); | |
394 | rdev = map_sector(mddev, zone, sector, §or_offset); | |
395 | subq = bdev_get_queue(rdev->bdev); | |
396 | if (subq->merge_bvec_fn) { | |
397 | bvm->bi_bdev = rdev->bdev; | |
398 | bvm->bi_sector = sector_offset + zone->dev_start + | |
399 | rdev->data_offset; | |
400 | return min(max, subq->merge_bvec_fn(subq, bvm, biovec)); | |
401 | } else | |
1da177e4 LT |
402 | return max; |
403 | } | |
404 | ||
fd01b88c | 405 | static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
406 | { |
407 | sector_t array_sectors = 0; | |
3cb03002 | 408 | struct md_rdev *rdev; |
80c3a6ce DW |
409 | |
410 | WARN_ONCE(sectors || raid_disks, | |
411 | "%s does not support generic reshape\n", __func__); | |
412 | ||
dafb20fa | 413 | rdev_for_each(rdev, mddev) |
a6468539 N |
414 | array_sectors += (rdev->sectors & |
415 | ~(sector_t)(mddev->chunk_sectors-1)); | |
80c3a6ce DW |
416 | |
417 | return array_sectors; | |
418 | } | |
419 | ||
afa0f557 | 420 | static void raid0_free(struct mddev *mddev, void *priv); |
0366ef84 | 421 | |
fd01b88c | 422 | static int raid0_run(struct mddev *mddev) |
1da177e4 | 423 | { |
e373ab10 | 424 | struct r0conf *conf; |
5568a603 | 425 | int ret; |
1da177e4 | 426 | |
9d8f0363 | 427 | if (mddev->chunk_sectors == 0) { |
b5a20961 N |
428 | printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n", |
429 | mdname(mddev)); | |
2604b703 N |
430 | return -EINVAL; |
431 | } | |
0894cc30 AN |
432 | if (md_check_no_bitmap(mddev)) |
433 | return -EINVAL; | |
753f2856 HM |
434 | |
435 | if (mddev->queue) { | |
436 | blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors); | |
437 | blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors); | |
438 | blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors); | |
439 | } | |
1da177e4 | 440 | |
9af204cf TM |
441 | /* if private is not null, we are here after takeover */ |
442 | if (mddev->private == NULL) { | |
443 | ret = create_strip_zones(mddev, &conf); | |
444 | if (ret < 0) | |
445 | return ret; | |
446 | mddev->private = conf; | |
447 | } | |
448 | conf = mddev->private; | |
1da177e4 LT |
449 | |
450 | /* calculate array device size */ | |
1f403624 | 451 | md_set_array_sectors(mddev, raid0_size(mddev, 0, 0)); |
1da177e4 | 452 | |
b5a20961 N |
453 | printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n", |
454 | mdname(mddev), | |
455 | (unsigned long long)mddev->array_sectors); | |
753f2856 HM |
456 | |
457 | if (mddev->queue) { | |
458 | /* calculate the max read-ahead size. | |
459 | * For read-ahead of large files to be effective, we need to | |
460 | * readahead at least twice a whole stripe. i.e. number of devices | |
461 | * multiplied by chunk size times 2. | |
462 | * If an individual device has an ra_pages greater than the | |
463 | * chunk size, then we will not drive that device as hard as it | |
464 | * wants. We consider this a configuration error: a larger | |
465 | * chunksize should be used in that case. | |
466 | */ | |
9d8f0363 AN |
467 | int stripe = mddev->raid_disks * |
468 | (mddev->chunk_sectors << 9) / PAGE_SIZE; | |
1da177e4 LT |
469 | if (mddev->queue->backing_dev_info.ra_pages < 2* stripe) |
470 | mddev->queue->backing_dev_info.ra_pages = 2* stripe; | |
471 | } | |
472 | ||
46994191 | 473 | dump_zones(mddev); |
0366ef84 | 474 | |
475 | ret = md_integrity_register(mddev); | |
0366ef84 | 476 | |
477 | return ret; | |
1da177e4 LT |
478 | } |
479 | ||
afa0f557 | 480 | static void raid0_free(struct mddev *mddev, void *priv) |
1da177e4 | 481 | { |
afa0f557 | 482 | struct r0conf *conf = priv; |
1da177e4 | 483 | |
990a8baf | 484 | kfree(conf->strip_zone); |
fb5ab4b5 | 485 | kfree(conf->devlist); |
990a8baf | 486 | kfree(conf); |
1da177e4 LT |
487 | } |
488 | ||
fbb704ef | 489 | /* |
490 | * Is io distribute over 1 or more chunks ? | |
491 | */ | |
fd01b88c | 492 | static inline int is_io_in_chunk_boundary(struct mddev *mddev, |
fbb704ef | 493 | unsigned int chunk_sects, struct bio *bio) |
494 | { | |
d6e412ea | 495 | if (likely(is_power_of_2(chunk_sects))) { |
4f024f37 KO |
496 | return chunk_sects >= |
497 | ((bio->bi_iter.bi_sector & (chunk_sects-1)) | |
aa8b57aa | 498 | + bio_sectors(bio)); |
fbb704ef | 499 | } else{ |
4f024f37 | 500 | sector_t sector = bio->bi_iter.bi_sector; |
fbb704ef | 501 | return chunk_sects >= (sector_div(sector, chunk_sects) |
aa8b57aa | 502 | + bio_sectors(bio)); |
fbb704ef | 503 | } |
504 | } | |
505 | ||
b4fdcb02 | 506 | static void raid0_make_request(struct mddev *mddev, struct bio *bio) |
fbb704ef | 507 | { |
1da177e4 | 508 | struct strip_zone *zone; |
3cb03002 | 509 | struct md_rdev *tmp_dev; |
20d0189b | 510 | struct bio *split; |
1da177e4 | 511 | |
e9c7469b TH |
512 | if (unlikely(bio->bi_rw & REQ_FLUSH)) { |
513 | md_flush_request(mddev, bio); | |
5a7bbad2 | 514 | return; |
e5dcdd80 N |
515 | } |
516 | ||
20d0189b | 517 | do { |
4f024f37 | 518 | sector_t sector = bio->bi_iter.bi_sector; |
20d0189b KO |
519 | unsigned chunk_sects = mddev->chunk_sectors; |
520 | ||
521 | unsigned sectors = chunk_sects - | |
522 | (likely(is_power_of_2(chunk_sects)) | |
523 | ? (sector & (chunk_sects-1)) | |
524 | : sector_div(sector, chunk_sects)); | |
525 | ||
526 | if (sectors < bio_sectors(bio)) { | |
527 | split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set); | |
528 | bio_chain(split, bio); | |
529 | } else { | |
530 | split = bio; | |
531 | } | |
1da177e4 | 532 | |
47d68979 | 533 | sector = bio->bi_iter.bi_sector; |
20d0189b KO |
534 | zone = find_zone(mddev->private, §or); |
535 | tmp_dev = map_sector(mddev, zone, sector, §or); | |
536 | split->bi_bdev = tmp_dev->bdev; | |
537 | split->bi_iter.bi_sector = sector + zone->dev_start + | |
538 | tmp_dev->data_offset; | |
539 | ||
540 | if (unlikely((split->bi_rw & REQ_DISCARD) && | |
541 | !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) { | |
542 | /* Just ignore it */ | |
543 | bio_endio(split, 0); | |
544 | } else | |
545 | generic_make_request(split); | |
546 | } while (split != bio); | |
1da177e4 | 547 | } |
8299d7f7 | 548 | |
fd01b88c | 549 | static void raid0_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 550 | { |
9d8f0363 | 551 | seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2); |
1da177e4 LT |
552 | return; |
553 | } | |
554 | ||
fd01b88c | 555 | static void *raid0_takeover_raid45(struct mddev *mddev) |
9af204cf | 556 | { |
3cb03002 | 557 | struct md_rdev *rdev; |
e373ab10 | 558 | struct r0conf *priv_conf; |
9af204cf TM |
559 | |
560 | if (mddev->degraded != 1) { | |
b5a20961 N |
561 | printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n", |
562 | mdname(mddev), | |
9af204cf TM |
563 | mddev->degraded); |
564 | return ERR_PTR(-EINVAL); | |
565 | } | |
566 | ||
dafb20fa | 567 | rdev_for_each(rdev, mddev) { |
9af204cf TM |
568 | /* check slot number for a disk */ |
569 | if (rdev->raid_disk == mddev->raid_disks-1) { | |
b5a20961 N |
570 | printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n", |
571 | mdname(mddev)); | |
9af204cf TM |
572 | return ERR_PTR(-EINVAL); |
573 | } | |
eea136d6 | 574 | rdev->sectors = mddev->dev_sectors; |
9af204cf TM |
575 | } |
576 | ||
577 | /* Set new parameters */ | |
578 | mddev->new_level = 0; | |
001048a3 | 579 | mddev->new_layout = 0; |
9af204cf TM |
580 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
581 | mddev->raid_disks--; | |
582 | mddev->delta_disks = -1; | |
583 | /* make sure it will be not marked as dirty */ | |
584 | mddev->recovery_cp = MaxSector; | |
585 | ||
586 | create_strip_zones(mddev, &priv_conf); | |
587 | return priv_conf; | |
588 | } | |
589 | ||
fd01b88c | 590 | static void *raid0_takeover_raid10(struct mddev *mddev) |
9af204cf | 591 | { |
e373ab10 | 592 | struct r0conf *priv_conf; |
9af204cf TM |
593 | |
594 | /* Check layout: | |
595 | * - far_copies must be 1 | |
596 | * - near_copies must be 2 | |
597 | * - disks number must be even | |
598 | * - all mirrors must be already degraded | |
599 | */ | |
600 | if (mddev->layout != ((1 << 8) + 2)) { | |
b5a20961 N |
601 | printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n", |
602 | mdname(mddev), | |
9af204cf TM |
603 | mddev->layout); |
604 | return ERR_PTR(-EINVAL); | |
605 | } | |
606 | if (mddev->raid_disks & 1) { | |
b5a20961 N |
607 | printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n", |
608 | mdname(mddev)); | |
9af204cf TM |
609 | return ERR_PTR(-EINVAL); |
610 | } | |
611 | if (mddev->degraded != (mddev->raid_disks>>1)) { | |
b5a20961 N |
612 | printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n", |
613 | mdname(mddev)); | |
9af204cf TM |
614 | return ERR_PTR(-EINVAL); |
615 | } | |
616 | ||
617 | /* Set new parameters */ | |
618 | mddev->new_level = 0; | |
001048a3 | 619 | mddev->new_layout = 0; |
9af204cf TM |
620 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
621 | mddev->delta_disks = - mddev->raid_disks / 2; | |
622 | mddev->raid_disks += mddev->delta_disks; | |
623 | mddev->degraded = 0; | |
624 | /* make sure it will be not marked as dirty */ | |
625 | mddev->recovery_cp = MaxSector; | |
626 | ||
627 | create_strip_zones(mddev, &priv_conf); | |
9af204cf TM |
628 | return priv_conf; |
629 | } | |
630 | ||
fd01b88c | 631 | static void *raid0_takeover_raid1(struct mddev *mddev) |
fc3a08b8 | 632 | { |
e373ab10 | 633 | struct r0conf *priv_conf; |
24b961f8 | 634 | int chunksect; |
fc3a08b8 KW |
635 | |
636 | /* Check layout: | |
637 | * - (N - 1) mirror drives must be already faulty | |
638 | */ | |
639 | if ((mddev->raid_disks - 1) != mddev->degraded) { | |
640 | printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n", | |
641 | mdname(mddev)); | |
642 | return ERR_PTR(-EINVAL); | |
643 | } | |
644 | ||
24b961f8 JS |
645 | /* |
646 | * a raid1 doesn't have the notion of chunk size, so | |
647 | * figure out the largest suitable size we can use. | |
648 | */ | |
649 | chunksect = 64 * 2; /* 64K by default */ | |
650 | ||
651 | /* The array must be an exact multiple of chunksize */ | |
652 | while (chunksect && (mddev->array_sectors & (chunksect - 1))) | |
653 | chunksect >>= 1; | |
654 | ||
655 | if ((chunksect << 9) < PAGE_SIZE) | |
656 | /* array size does not allow a suitable chunk size */ | |
657 | return ERR_PTR(-EINVAL); | |
658 | ||
fc3a08b8 KW |
659 | /* Set new parameters */ |
660 | mddev->new_level = 0; | |
661 | mddev->new_layout = 0; | |
24b961f8 JS |
662 | mddev->new_chunk_sectors = chunksect; |
663 | mddev->chunk_sectors = chunksect; | |
fc3a08b8 | 664 | mddev->delta_disks = 1 - mddev->raid_disks; |
f7bee809 | 665 | mddev->raid_disks = 1; |
fc3a08b8 KW |
666 | /* make sure it will be not marked as dirty */ |
667 | mddev->recovery_cp = MaxSector; | |
668 | ||
669 | create_strip_zones(mddev, &priv_conf); | |
670 | return priv_conf; | |
671 | } | |
672 | ||
fd01b88c | 673 | static void *raid0_takeover(struct mddev *mddev) |
9af204cf TM |
674 | { |
675 | /* raid0 can take over: | |
049d6c1e | 676 | * raid4 - if all data disks are active. |
9af204cf TM |
677 | * raid5 - providing it is Raid4 layout and one disk is faulty |
678 | * raid10 - assuming we have all necessary active disks | |
fc3a08b8 | 679 | * raid1 - with (N -1) mirror drives faulty |
9af204cf | 680 | */ |
a8461a61 N |
681 | |
682 | if (mddev->bitmap) { | |
683 | printk(KERN_ERR "md/raid0: %s: cannot takeover array with bitmap\n", | |
684 | mdname(mddev)); | |
685 | return ERR_PTR(-EBUSY); | |
686 | } | |
049d6c1e MT |
687 | if (mddev->level == 4) |
688 | return raid0_takeover_raid45(mddev); | |
689 | ||
9af204cf TM |
690 | if (mddev->level == 5) { |
691 | if (mddev->layout == ALGORITHM_PARITY_N) | |
049d6c1e | 692 | return raid0_takeover_raid45(mddev); |
9af204cf | 693 | |
b5a20961 N |
694 | printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n", |
695 | mdname(mddev), ALGORITHM_PARITY_N); | |
9af204cf TM |
696 | } |
697 | ||
698 | if (mddev->level == 10) | |
699 | return raid0_takeover_raid10(mddev); | |
700 | ||
fc3a08b8 KW |
701 | if (mddev->level == 1) |
702 | return raid0_takeover_raid1(mddev); | |
703 | ||
704 | printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n", | |
705 | mddev->level); | |
706 | ||
9af204cf TM |
707 | return ERR_PTR(-EINVAL); |
708 | } | |
709 | ||
fd01b88c | 710 | static void raid0_quiesce(struct mddev *mddev, int state) |
9af204cf TM |
711 | { |
712 | } | |
713 | ||
84fc4b56 | 714 | static struct md_personality raid0_personality= |
1da177e4 LT |
715 | { |
716 | .name = "raid0", | |
2604b703 | 717 | .level = 0, |
1da177e4 LT |
718 | .owner = THIS_MODULE, |
719 | .make_request = raid0_make_request, | |
720 | .run = raid0_run, | |
afa0f557 | 721 | .free = raid0_free, |
1da177e4 | 722 | .status = raid0_status, |
80c3a6ce | 723 | .size = raid0_size, |
9af204cf TM |
724 | .takeover = raid0_takeover, |
725 | .quiesce = raid0_quiesce, | |
5c675f83 | 726 | .congested = raid0_congested, |
64590f45 | 727 | .mergeable_bvec = raid0_mergeable_bvec, |
1da177e4 LT |
728 | }; |
729 | ||
730 | static int __init raid0_init (void) | |
731 | { | |
2604b703 | 732 | return register_md_personality (&raid0_personality); |
1da177e4 LT |
733 | } |
734 | ||
735 | static void raid0_exit (void) | |
736 | { | |
2604b703 | 737 | unregister_md_personality (&raid0_personality); |
1da177e4 LT |
738 | } |
739 | ||
740 | module_init(raid0_init); | |
741 | module_exit(raid0_exit); | |
742 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 743 | MODULE_DESCRIPTION("RAID0 (striping) personality for MD"); |
1da177e4 | 744 | MODULE_ALIAS("md-personality-2"); /* RAID0 */ |
d9d166c2 | 745 | MODULE_ALIAS("md-raid0"); |
2604b703 | 746 | MODULE_ALIAS("md-level-0"); |