]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | linear.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 | ||
7 | Linear mode management functions. | |
8 | ||
9 | This program is free software; you can redistribute it and/or modify | |
10 | it under the terms of the GNU General Public License as published by | |
11 | the Free Software Foundation; either version 2, or (at your option) | |
12 | any later version. | |
13 | ||
14 | You should have received a copy of the GNU General Public License | |
15 | (for example /usr/src/linux/COPYING); if not, write to the Free | |
16 | Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
17 | */ | |
18 | ||
1da177e4 LT |
19 | #include <linux/raid/linear.h> |
20 | ||
1da177e4 LT |
21 | /* |
22 | * find which device holds a particular offset | |
23 | */ | |
24 | static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector) | |
25 | { | |
26 | dev_info_t *hash; | |
27 | linear_conf_t *conf = mddev_to_conf(mddev); | |
1da177e4 LT |
28 | |
29 | /* | |
30 | * sector_div(a,b) returns the remainer and sets a to a/b | |
31 | */ | |
ab5bd5cb AN |
32 | sector >>= conf->sector_shift; |
33 | (void)sector_div(sector, conf->spacing); | |
34 | hash = conf->hash_table[sector]; | |
1da177e4 | 35 | |
6283815d | 36 | while (sector >= hash->num_sectors + hash->start_sector) |
1da177e4 LT |
37 | hash++; |
38 | return hash; | |
39 | } | |
40 | ||
41 | /** | |
15945fee | 42 | * linear_mergeable_bvec -- tell bio layer if two requests can be merged |
1da177e4 | 43 | * @q: request queue |
cc371e66 | 44 | * @bvm: properties of new bio |
1da177e4 LT |
45 | * @biovec: the request that could be merged to it. |
46 | * | |
47 | * Return amount of bytes we can take at this offset | |
48 | */ | |
cc371e66 AK |
49 | static int linear_mergeable_bvec(struct request_queue *q, |
50 | struct bvec_merge_data *bvm, | |
51 | struct bio_vec *biovec) | |
1da177e4 LT |
52 | { |
53 | mddev_t *mddev = q->queuedata; | |
54 | dev_info_t *dev0; | |
cc371e66 AK |
55 | unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9; |
56 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); | |
1da177e4 LT |
57 | |
58 | dev0 = which_dev(mddev, sector); | |
6283815d | 59 | maxsectors = dev0->num_sectors - (sector - dev0->start_sector); |
1da177e4 LT |
60 | |
61 | if (maxsectors < bio_sectors) | |
62 | maxsectors = 0; | |
63 | else | |
64 | maxsectors -= bio_sectors; | |
65 | ||
66 | if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0) | |
67 | return biovec->bv_len; | |
68 | /* The bytes available at this offset could be really big, | |
69 | * so we cap at 2^31 to avoid overflow */ | |
70 | if (maxsectors > (1 << (31-9))) | |
71 | return 1<<31; | |
72 | return maxsectors << 9; | |
73 | } | |
74 | ||
165125e1 | 75 | static void linear_unplug(struct request_queue *q) |
1da177e4 LT |
76 | { |
77 | mddev_t *mddev = q->queuedata; | |
78 | linear_conf_t *conf = mddev_to_conf(mddev); | |
79 | int i; | |
80 | ||
81 | for (i=0; i < mddev->raid_disks; i++) { | |
165125e1 | 82 | struct request_queue *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev); |
2ad8b1ef | 83 | blk_unplug(r_queue); |
1da177e4 LT |
84 | } |
85 | } | |
86 | ||
26be34dc N |
87 | static int linear_congested(void *data, int bits) |
88 | { | |
89 | mddev_t *mddev = data; | |
90 | linear_conf_t *conf = mddev_to_conf(mddev); | |
91 | int i, ret = 0; | |
92 | ||
93 | for (i = 0; i < mddev->raid_disks && !ret ; i++) { | |
165125e1 | 94 | struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev); |
26be34dc N |
95 | ret |= bdi_congested(&q->backing_dev_info, bits); |
96 | } | |
97 | return ret; | |
98 | } | |
99 | ||
7c7546cc | 100 | static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks) |
1da177e4 LT |
101 | { |
102 | linear_conf_t *conf; | |
103 | dev_info_t **table; | |
104 | mdk_rdev_t *rdev; | |
105 | int i, nb_zone, cnt; | |
23242fbb | 106 | sector_t min_sectors; |
6283815d | 107 | sector_t curr_sector; |
1da177e4 | 108 | |
7c7546cc | 109 | conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t), |
1da177e4 LT |
110 | GFP_KERNEL); |
111 | if (!conf) | |
7c7546cc N |
112 | return NULL; |
113 | ||
1da177e4 | 114 | cnt = 0; |
d6e22150 | 115 | conf->array_sectors = 0; |
1da177e4 | 116 | |
159ec1fc | 117 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
118 | int j = rdev->raid_disk; |
119 | dev_info_t *disk = conf->disks + j; | |
120 | ||
13864515 | 121 | if (j < 0 || j >= raid_disks || disk->rdev) { |
1da177e4 LT |
122 | printk("linear: disk numbering problem. Aborting!\n"); |
123 | goto out; | |
124 | } | |
125 | ||
126 | disk->rdev = rdev; | |
127 | ||
128 | blk_queue_stack_limits(mddev->queue, | |
129 | rdev->bdev->bd_disk->queue); | |
130 | /* as we don't honour merge_bvec_fn, we must never risk | |
131 | * violating it, so limit ->max_sector to one PAGE, as | |
132 | * a one page request is never in violation. | |
133 | */ | |
134 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn && | |
135 | mddev->queue->max_sectors > (PAGE_SIZE>>9)) | |
136 | blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9); | |
137 | ||
6283815d | 138 | disk->num_sectors = rdev->size * 2; |
d6e22150 | 139 | conf->array_sectors += rdev->size * 2; |
1da177e4 | 140 | |
1da177e4 LT |
141 | cnt++; |
142 | } | |
7c7546cc | 143 | if (cnt != raid_disks) { |
1da177e4 LT |
144 | printk("linear: not enough drives present. Aborting!\n"); |
145 | goto out; | |
146 | } | |
147 | ||
23242fbb AN |
148 | min_sectors = conf->array_sectors; |
149 | sector_div(min_sectors, PAGE_SIZE/sizeof(struct dev_info *)); | |
f1cd14ae AN |
150 | if (min_sectors == 0) |
151 | min_sectors = 1; | |
15945fee | 152 | |
23242fbb | 153 | /* min_sectors is the minimum spacing that will fit the hash |
15945fee N |
154 | * table in one PAGE. This may be much smaller than needed. |
155 | * We find the smallest non-terminal set of consecutive devices | |
23242fbb | 156 | * that is larger than min_sectors and use the size of that as |
15945fee N |
157 | * the actual spacing |
158 | */ | |
ab5bd5cb | 159 | conf->spacing = conf->array_sectors; |
15945fee | 160 | for (i=0; i < cnt-1 ; i++) { |
23242fbb | 161 | sector_t tmp = 0; |
15945fee | 162 | int j; |
23242fbb AN |
163 | for (j = i; j < cnt - 1 && tmp < min_sectors; j++) |
164 | tmp += conf->disks[j].num_sectors; | |
ab5bd5cb AN |
165 | if (tmp >= min_sectors && tmp < conf->spacing) |
166 | conf->spacing = tmp; | |
15945fee N |
167 | } |
168 | ||
ab5bd5cb | 169 | /* spacing may be too large for sector_div to work with, |
15945fee N |
170 | * so we might need to pre-shift |
171 | */ | |
ab5bd5cb | 172 | conf->sector_shift = 0; |
15945fee | 173 | if (sizeof(sector_t) > sizeof(u32)) { |
ab5bd5cb | 174 | sector_t space = conf->spacing; |
15945fee N |
175 | while (space > (sector_t)(~(u32)0)) { |
176 | space >>= 1; | |
ab5bd5cb | 177 | conf->sector_shift++; |
15945fee N |
178 | } |
179 | } | |
1da177e4 LT |
180 | /* |
181 | * This code was restructured to work around a gcc-2.95.3 internal | |
182 | * compiler error. Alter it with care. | |
183 | */ | |
184 | { | |
185 | sector_t sz; | |
186 | unsigned round; | |
187 | unsigned long base; | |
188 | ||
ab5bd5cb | 189 | sz = conf->array_sectors >> conf->sector_shift; |
15945fee | 190 | sz += 1; /* force round-up */ |
ab5bd5cb | 191 | base = conf->spacing >> conf->sector_shift; |
1da177e4 | 192 | round = sector_div(sz, base); |
15945fee | 193 | nb_zone = sz + (round ? 1 : 0); |
1da177e4 | 194 | } |
15945fee N |
195 | BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *)); |
196 | ||
197 | conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone, | |
1da177e4 LT |
198 | GFP_KERNEL); |
199 | if (!conf->hash_table) | |
200 | goto out; | |
201 | ||
202 | /* | |
203 | * Here we generate the linear hash table | |
15945fee | 204 | * First calculate the device offsets. |
1da177e4 | 205 | */ |
6283815d | 206 | conf->disks[0].start_sector = 0; |
a778b73f | 207 | for (i = 1; i < raid_disks; i++) |
6283815d AN |
208 | conf->disks[i].start_sector = |
209 | conf->disks[i-1].start_sector + | |
210 | conf->disks[i-1].num_sectors; | |
15945fee | 211 | |
1da177e4 | 212 | table = conf->hash_table; |
15945fee | 213 | i = 0; |
6283815d AN |
214 | for (curr_sector = 0; |
215 | curr_sector < conf->array_sectors; | |
ab5bd5cb | 216 | curr_sector += conf->spacing) { |
1da177e4 | 217 | |
a778b73f | 218 | while (i < raid_disks-1 && |
6283815d | 219 | curr_sector >= conf->disks[i+1].start_sector) |
15945fee | 220 | i++; |
1da177e4 | 221 | |
15945fee N |
222 | *table ++ = conf->disks + i; |
223 | } | |
224 | ||
ab5bd5cb AN |
225 | if (conf->sector_shift) { |
226 | conf->spacing >>= conf->sector_shift; | |
227 | /* round spacing up so that when we divide by it, | |
15945fee | 228 | * we err on the side of "too-low", which is safest. |
1da177e4 | 229 | */ |
ab5bd5cb | 230 | conf->spacing++; |
1da177e4 | 231 | } |
15945fee N |
232 | |
233 | BUG_ON(table - conf->hash_table > nb_zone); | |
1da177e4 | 234 | |
7c7546cc N |
235 | return conf; |
236 | ||
237 | out: | |
238 | kfree(conf); | |
239 | return NULL; | |
240 | } | |
241 | ||
242 | static int linear_run (mddev_t *mddev) | |
243 | { | |
244 | linear_conf_t *conf; | |
245 | ||
e7e72bf6 | 246 | mddev->queue->queue_lock = &mddev->queue->__queue_lock; |
7c7546cc N |
247 | conf = linear_conf(mddev, mddev->raid_disks); |
248 | ||
249 | if (!conf) | |
250 | return 1; | |
251 | mddev->private = conf; | |
d6e22150 | 252 | mddev->array_sectors = conf->array_sectors; |
7c7546cc | 253 | |
1da177e4 LT |
254 | blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec); |
255 | mddev->queue->unplug_fn = linear_unplug; | |
26be34dc N |
256 | mddev->queue->backing_dev_info.congested_fn = linear_congested; |
257 | mddev->queue->backing_dev_info.congested_data = mddev; | |
1da177e4 | 258 | return 0; |
7c7546cc | 259 | } |
1da177e4 | 260 | |
7c7546cc N |
261 | static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev) |
262 | { | |
263 | /* Adding a drive to a linear array allows the array to grow. | |
264 | * It is permitted if the new drive has a matching superblock | |
265 | * already on it, with raid_disk equal to raid_disks. | |
266 | * It is achieved by creating a new linear_private_data structure | |
267 | * and swapping it in in-place of the current one. | |
268 | * The current one is never freed until the array is stopped. | |
269 | * This avoids races. | |
270 | */ | |
271 | linear_conf_t *newconf; | |
272 | ||
a778b73f | 273 | if (rdev->saved_raid_disk != mddev->raid_disks) |
7c7546cc N |
274 | return -EINVAL; |
275 | ||
a778b73f N |
276 | rdev->raid_disk = rdev->saved_raid_disk; |
277 | ||
7c7546cc N |
278 | newconf = linear_conf(mddev,mddev->raid_disks+1); |
279 | ||
280 | if (!newconf) | |
281 | return -ENOMEM; | |
282 | ||
283 | newconf->prev = mddev_to_conf(mddev); | |
284 | mddev->private = newconf; | |
285 | mddev->raid_disks++; | |
d6e22150 | 286 | mddev->array_sectors = newconf->array_sectors; |
f233ea5c | 287 | set_capacity(mddev->gendisk, mddev->array_sectors); |
7c7546cc | 288 | return 0; |
1da177e4 LT |
289 | } |
290 | ||
291 | static int linear_stop (mddev_t *mddev) | |
292 | { | |
293 | linear_conf_t *conf = mddev_to_conf(mddev); | |
294 | ||
295 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
7c7546cc N |
296 | do { |
297 | linear_conf_t *t = conf->prev; | |
298 | kfree(conf->hash_table); | |
299 | kfree(conf); | |
300 | conf = t; | |
301 | } while (conf); | |
1da177e4 LT |
302 | |
303 | return 0; | |
304 | } | |
305 | ||
165125e1 | 306 | static int linear_make_request (struct request_queue *q, struct bio *bio) |
1da177e4 | 307 | { |
a362357b | 308 | const int rw = bio_data_dir(bio); |
1da177e4 LT |
309 | mddev_t *mddev = q->queuedata; |
310 | dev_info_t *tmp_dev; | |
c9959059 | 311 | int cpu; |
1da177e4 | 312 | |
e5dcdd80 | 313 | if (unlikely(bio_barrier(bio))) { |
6712ecf8 | 314 | bio_endio(bio, -EOPNOTSUPP); |
e5dcdd80 N |
315 | return 0; |
316 | } | |
317 | ||
074a7aca TH |
318 | cpu = part_stat_lock(); |
319 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
320 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], | |
321 | bio_sectors(bio)); | |
322 | part_stat_unlock(); | |
1da177e4 LT |
323 | |
324 | tmp_dev = which_dev(mddev, bio->bi_sector); | |
1da177e4 | 325 | |
6283815d AN |
326 | if (unlikely(bio->bi_sector >= (tmp_dev->num_sectors + |
327 | tmp_dev->start_sector) | |
328 | || (bio->bi_sector < | |
329 | tmp_dev->start_sector))) { | |
1da177e4 LT |
330 | char b[BDEVNAME_SIZE]; |
331 | ||
6283815d AN |
332 | printk("linear_make_request: Sector %llu out of bounds on " |
333 | "dev %s: %llu sectors, offset %llu\n", | |
334 | (unsigned long long)bio->bi_sector, | |
1da177e4 | 335 | bdevname(tmp_dev->rdev->bdev, b), |
6283815d AN |
336 | (unsigned long long)tmp_dev->num_sectors, |
337 | (unsigned long long)tmp_dev->start_sector); | |
6712ecf8 | 338 | bio_io_error(bio); |
1da177e4 LT |
339 | return 0; |
340 | } | |
341 | if (unlikely(bio->bi_sector + (bio->bi_size >> 9) > | |
6283815d | 342 | tmp_dev->start_sector + tmp_dev->num_sectors)) { |
1da177e4 LT |
343 | /* This bio crosses a device boundary, so we have to |
344 | * split it. | |
345 | */ | |
346 | struct bio_pair *bp; | |
6283815d | 347 | |
6feef531 | 348 | bp = bio_split(bio, |
6283815d AN |
349 | tmp_dev->start_sector + tmp_dev->num_sectors |
350 | - bio->bi_sector); | |
351 | ||
1da177e4 LT |
352 | if (linear_make_request(q, &bp->bio1)) |
353 | generic_make_request(&bp->bio1); | |
354 | if (linear_make_request(q, &bp->bio2)) | |
355 | generic_make_request(&bp->bio2); | |
356 | bio_pair_release(bp); | |
357 | return 0; | |
358 | } | |
359 | ||
360 | bio->bi_bdev = tmp_dev->rdev->bdev; | |
6283815d AN |
361 | bio->bi_sector = bio->bi_sector - tmp_dev->start_sector |
362 | + tmp_dev->rdev->data_offset; | |
1da177e4 LT |
363 | |
364 | return 1; | |
365 | } | |
366 | ||
367 | static void linear_status (struct seq_file *seq, mddev_t *mddev) | |
368 | { | |
369 | ||
1da177e4 LT |
370 | seq_printf(seq, " %dk rounding", mddev->chunk_size/1024); |
371 | } | |
372 | ||
373 | ||
2604b703 | 374 | static struct mdk_personality linear_personality = |
1da177e4 LT |
375 | { |
376 | .name = "linear", | |
2604b703 | 377 | .level = LEVEL_LINEAR, |
1da177e4 LT |
378 | .owner = THIS_MODULE, |
379 | .make_request = linear_make_request, | |
380 | .run = linear_run, | |
381 | .stop = linear_stop, | |
382 | .status = linear_status, | |
7c7546cc | 383 | .hot_add_disk = linear_add, |
1da177e4 LT |
384 | }; |
385 | ||
386 | static int __init linear_init (void) | |
387 | { | |
2604b703 | 388 | return register_md_personality (&linear_personality); |
1da177e4 LT |
389 | } |
390 | ||
391 | static void linear_exit (void) | |
392 | { | |
2604b703 | 393 | unregister_md_personality (&linear_personality); |
1da177e4 LT |
394 | } |
395 | ||
396 | ||
397 | module_init(linear_init); | |
398 | module_exit(linear_exit); | |
399 | MODULE_LICENSE("GPL"); | |
d9d166c2 N |
400 | MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/ |
401 | MODULE_ALIAS("md-linear"); | |
2604b703 | 402 | MODULE_ALIAS("md-level--1"); |