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 | ||
bff61975 N |
19 | #include <linux/blkdev.h> |
20 | #include <linux/raid/md_u.h> | |
21 | #include <linux/raid/md_k.h> | |
22 | #include <linux/seq_file.h> | |
ef740c37 | 23 | #include "linear.h" |
1da177e4 | 24 | |
1da177e4 LT |
25 | /* |
26 | * find which device holds a particular offset | |
27 | */ | |
28 | static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector) | |
29 | { | |
30 | dev_info_t *hash; | |
31 | linear_conf_t *conf = mddev_to_conf(mddev); | |
852c8bf4 | 32 | sector_t idx = sector >> conf->sector_shift; |
1da177e4 LT |
33 | |
34 | /* | |
35 | * sector_div(a,b) returns the remainer and sets a to a/b | |
36 | */ | |
852c8bf4 AN |
37 | (void)sector_div(idx, conf->spacing); |
38 | hash = conf->hash_table[idx]; | |
1da177e4 | 39 | |
6283815d | 40 | while (sector >= hash->num_sectors + hash->start_sector) |
1da177e4 LT |
41 | hash++; |
42 | return hash; | |
43 | } | |
44 | ||
45 | /** | |
15945fee | 46 | * linear_mergeable_bvec -- tell bio layer if two requests can be merged |
1da177e4 | 47 | * @q: request queue |
cc371e66 | 48 | * @bvm: properties of new bio |
1da177e4 LT |
49 | * @biovec: the request that could be merged to it. |
50 | * | |
51 | * Return amount of bytes we can take at this offset | |
52 | */ | |
cc371e66 AK |
53 | static int linear_mergeable_bvec(struct request_queue *q, |
54 | struct bvec_merge_data *bvm, | |
55 | struct bio_vec *biovec) | |
1da177e4 LT |
56 | { |
57 | mddev_t *mddev = q->queuedata; | |
58 | dev_info_t *dev0; | |
cc371e66 AK |
59 | unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9; |
60 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); | |
1da177e4 LT |
61 | |
62 | dev0 = which_dev(mddev, sector); | |
6283815d | 63 | maxsectors = dev0->num_sectors - (sector - dev0->start_sector); |
1da177e4 LT |
64 | |
65 | if (maxsectors < bio_sectors) | |
66 | maxsectors = 0; | |
67 | else | |
68 | maxsectors -= bio_sectors; | |
69 | ||
70 | if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0) | |
71 | return biovec->bv_len; | |
72 | /* The bytes available at this offset could be really big, | |
73 | * so we cap at 2^31 to avoid overflow */ | |
74 | if (maxsectors > (1 << (31-9))) | |
75 | return 1<<31; | |
76 | return maxsectors << 9; | |
77 | } | |
78 | ||
165125e1 | 79 | static void linear_unplug(struct request_queue *q) |
1da177e4 LT |
80 | { |
81 | mddev_t *mddev = q->queuedata; | |
82 | linear_conf_t *conf = mddev_to_conf(mddev); | |
83 | int i; | |
84 | ||
85 | for (i=0; i < mddev->raid_disks; i++) { | |
165125e1 | 86 | struct request_queue *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev); |
2ad8b1ef | 87 | blk_unplug(r_queue); |
1da177e4 LT |
88 | } |
89 | } | |
90 | ||
26be34dc N |
91 | static int linear_congested(void *data, int bits) |
92 | { | |
93 | mddev_t *mddev = data; | |
94 | linear_conf_t *conf = mddev_to_conf(mddev); | |
95 | int i, ret = 0; | |
96 | ||
97 | for (i = 0; i < mddev->raid_disks && !ret ; i++) { | |
165125e1 | 98 | struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev); |
26be34dc N |
99 | ret |= bdi_congested(&q->backing_dev_info, bits); |
100 | } | |
101 | return ret; | |
102 | } | |
103 | ||
7c7546cc | 104 | static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks) |
1da177e4 LT |
105 | { |
106 | linear_conf_t *conf; | |
107 | dev_info_t **table; | |
108 | mdk_rdev_t *rdev; | |
109 | int i, nb_zone, cnt; | |
23242fbb | 110 | sector_t min_sectors; |
6283815d | 111 | sector_t curr_sector; |
1da177e4 | 112 | |
7c7546cc | 113 | conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t), |
1da177e4 LT |
114 | GFP_KERNEL); |
115 | if (!conf) | |
7c7546cc N |
116 | return NULL; |
117 | ||
1da177e4 | 118 | cnt = 0; |
d6e22150 | 119 | conf->array_sectors = 0; |
1da177e4 | 120 | |
159ec1fc | 121 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 LT |
122 | int j = rdev->raid_disk; |
123 | dev_info_t *disk = conf->disks + j; | |
124 | ||
13864515 | 125 | if (j < 0 || j >= raid_disks || disk->rdev) { |
1da177e4 LT |
126 | printk("linear: disk numbering problem. Aborting!\n"); |
127 | goto out; | |
128 | } | |
129 | ||
130 | disk->rdev = rdev; | |
131 | ||
132 | blk_queue_stack_limits(mddev->queue, | |
133 | rdev->bdev->bd_disk->queue); | |
134 | /* as we don't honour merge_bvec_fn, we must never risk | |
135 | * violating it, so limit ->max_sector to one PAGE, as | |
136 | * a one page request is never in violation. | |
137 | */ | |
138 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn && | |
139 | mddev->queue->max_sectors > (PAGE_SIZE>>9)) | |
140 | blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9); | |
141 | ||
6283815d | 142 | disk->num_sectors = rdev->size * 2; |
d6e22150 | 143 | conf->array_sectors += rdev->size * 2; |
1da177e4 | 144 | |
1da177e4 LT |
145 | cnt++; |
146 | } | |
7c7546cc | 147 | if (cnt != raid_disks) { |
1da177e4 LT |
148 | printk("linear: not enough drives present. Aborting!\n"); |
149 | goto out; | |
150 | } | |
151 | ||
23242fbb AN |
152 | min_sectors = conf->array_sectors; |
153 | sector_div(min_sectors, PAGE_SIZE/sizeof(struct dev_info *)); | |
f1cd14ae AN |
154 | if (min_sectors == 0) |
155 | min_sectors = 1; | |
15945fee | 156 | |
23242fbb | 157 | /* min_sectors is the minimum spacing that will fit the hash |
15945fee N |
158 | * table in one PAGE. This may be much smaller than needed. |
159 | * We find the smallest non-terminal set of consecutive devices | |
23242fbb | 160 | * that is larger than min_sectors and use the size of that as |
15945fee N |
161 | * the actual spacing |
162 | */ | |
ab5bd5cb | 163 | conf->spacing = conf->array_sectors; |
15945fee | 164 | for (i=0; i < cnt-1 ; i++) { |
23242fbb | 165 | sector_t tmp = 0; |
15945fee | 166 | int j; |
23242fbb AN |
167 | for (j = i; j < cnt - 1 && tmp < min_sectors; j++) |
168 | tmp += conf->disks[j].num_sectors; | |
ab5bd5cb AN |
169 | if (tmp >= min_sectors && tmp < conf->spacing) |
170 | conf->spacing = tmp; | |
15945fee N |
171 | } |
172 | ||
ab5bd5cb | 173 | /* spacing may be too large for sector_div to work with, |
15945fee N |
174 | * so we might need to pre-shift |
175 | */ | |
ab5bd5cb | 176 | conf->sector_shift = 0; |
15945fee | 177 | if (sizeof(sector_t) > sizeof(u32)) { |
ab5bd5cb | 178 | sector_t space = conf->spacing; |
15945fee N |
179 | while (space > (sector_t)(~(u32)0)) { |
180 | space >>= 1; | |
ab5bd5cb | 181 | conf->sector_shift++; |
15945fee N |
182 | } |
183 | } | |
1da177e4 LT |
184 | /* |
185 | * This code was restructured to work around a gcc-2.95.3 internal | |
186 | * compiler error. Alter it with care. | |
187 | */ | |
188 | { | |
189 | sector_t sz; | |
190 | unsigned round; | |
191 | unsigned long base; | |
192 | ||
ab5bd5cb | 193 | sz = conf->array_sectors >> conf->sector_shift; |
15945fee | 194 | sz += 1; /* force round-up */ |
ab5bd5cb | 195 | base = conf->spacing >> conf->sector_shift; |
1da177e4 | 196 | round = sector_div(sz, base); |
15945fee | 197 | nb_zone = sz + (round ? 1 : 0); |
1da177e4 | 198 | } |
15945fee N |
199 | BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *)); |
200 | ||
201 | conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone, | |
1da177e4 LT |
202 | GFP_KERNEL); |
203 | if (!conf->hash_table) | |
204 | goto out; | |
205 | ||
206 | /* | |
207 | * Here we generate the linear hash table | |
15945fee | 208 | * First calculate the device offsets. |
1da177e4 | 209 | */ |
6283815d | 210 | conf->disks[0].start_sector = 0; |
a778b73f | 211 | for (i = 1; i < raid_disks; i++) |
6283815d AN |
212 | conf->disks[i].start_sector = |
213 | conf->disks[i-1].start_sector + | |
214 | conf->disks[i-1].num_sectors; | |
15945fee | 215 | |
1da177e4 | 216 | table = conf->hash_table; |
15945fee | 217 | i = 0; |
6283815d AN |
218 | for (curr_sector = 0; |
219 | curr_sector < conf->array_sectors; | |
ab5bd5cb | 220 | curr_sector += conf->spacing) { |
1da177e4 | 221 | |
a778b73f | 222 | while (i < raid_disks-1 && |
6283815d | 223 | curr_sector >= conf->disks[i+1].start_sector) |
15945fee | 224 | i++; |
1da177e4 | 225 | |
15945fee N |
226 | *table ++ = conf->disks + i; |
227 | } | |
228 | ||
ab5bd5cb AN |
229 | if (conf->sector_shift) { |
230 | conf->spacing >>= conf->sector_shift; | |
231 | /* round spacing up so that when we divide by it, | |
15945fee | 232 | * we err on the side of "too-low", which is safest. |
1da177e4 | 233 | */ |
ab5bd5cb | 234 | conf->spacing++; |
1da177e4 | 235 | } |
15945fee N |
236 | |
237 | BUG_ON(table - conf->hash_table > nb_zone); | |
1da177e4 | 238 | |
7c7546cc N |
239 | return conf; |
240 | ||
241 | out: | |
242 | kfree(conf); | |
243 | return NULL; | |
244 | } | |
245 | ||
246 | static int linear_run (mddev_t *mddev) | |
247 | { | |
248 | linear_conf_t *conf; | |
249 | ||
e7e72bf6 | 250 | mddev->queue->queue_lock = &mddev->queue->__queue_lock; |
7c7546cc N |
251 | conf = linear_conf(mddev, mddev->raid_disks); |
252 | ||
253 | if (!conf) | |
254 | return 1; | |
255 | mddev->private = conf; | |
d6e22150 | 256 | mddev->array_sectors = conf->array_sectors; |
7c7546cc | 257 | |
1da177e4 LT |
258 | blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec); |
259 | mddev->queue->unplug_fn = linear_unplug; | |
26be34dc N |
260 | mddev->queue->backing_dev_info.congested_fn = linear_congested; |
261 | mddev->queue->backing_dev_info.congested_data = mddev; | |
1da177e4 | 262 | return 0; |
7c7546cc | 263 | } |
1da177e4 | 264 | |
7c7546cc N |
265 | static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev) |
266 | { | |
267 | /* Adding a drive to a linear array allows the array to grow. | |
268 | * It is permitted if the new drive has a matching superblock | |
269 | * already on it, with raid_disk equal to raid_disks. | |
270 | * It is achieved by creating a new linear_private_data structure | |
271 | * and swapping it in in-place of the current one. | |
272 | * The current one is never freed until the array is stopped. | |
273 | * This avoids races. | |
274 | */ | |
275 | linear_conf_t *newconf; | |
276 | ||
a778b73f | 277 | if (rdev->saved_raid_disk != mddev->raid_disks) |
7c7546cc N |
278 | return -EINVAL; |
279 | ||
a778b73f N |
280 | rdev->raid_disk = rdev->saved_raid_disk; |
281 | ||
7c7546cc N |
282 | newconf = linear_conf(mddev,mddev->raid_disks+1); |
283 | ||
284 | if (!newconf) | |
285 | return -ENOMEM; | |
286 | ||
287 | newconf->prev = mddev_to_conf(mddev); | |
288 | mddev->private = newconf; | |
289 | mddev->raid_disks++; | |
d6e22150 | 290 | mddev->array_sectors = newconf->array_sectors; |
f233ea5c | 291 | set_capacity(mddev->gendisk, mddev->array_sectors); |
7c7546cc | 292 | return 0; |
1da177e4 LT |
293 | } |
294 | ||
295 | static int linear_stop (mddev_t *mddev) | |
296 | { | |
297 | linear_conf_t *conf = mddev_to_conf(mddev); | |
298 | ||
299 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
7c7546cc N |
300 | do { |
301 | linear_conf_t *t = conf->prev; | |
302 | kfree(conf->hash_table); | |
303 | kfree(conf); | |
304 | conf = t; | |
305 | } while (conf); | |
1da177e4 LT |
306 | |
307 | return 0; | |
308 | } | |
309 | ||
165125e1 | 310 | static int linear_make_request (struct request_queue *q, struct bio *bio) |
1da177e4 | 311 | { |
a362357b | 312 | const int rw = bio_data_dir(bio); |
1da177e4 LT |
313 | mddev_t *mddev = q->queuedata; |
314 | dev_info_t *tmp_dev; | |
c9959059 | 315 | int cpu; |
1da177e4 | 316 | |
e5dcdd80 | 317 | if (unlikely(bio_barrier(bio))) { |
6712ecf8 | 318 | bio_endio(bio, -EOPNOTSUPP); |
e5dcdd80 N |
319 | return 0; |
320 | } | |
321 | ||
074a7aca TH |
322 | cpu = part_stat_lock(); |
323 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
324 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], | |
325 | bio_sectors(bio)); | |
326 | part_stat_unlock(); | |
1da177e4 LT |
327 | |
328 | tmp_dev = which_dev(mddev, bio->bi_sector); | |
1da177e4 | 329 | |
6283815d AN |
330 | if (unlikely(bio->bi_sector >= (tmp_dev->num_sectors + |
331 | tmp_dev->start_sector) | |
332 | || (bio->bi_sector < | |
333 | tmp_dev->start_sector))) { | |
1da177e4 LT |
334 | char b[BDEVNAME_SIZE]; |
335 | ||
6283815d AN |
336 | printk("linear_make_request: Sector %llu out of bounds on " |
337 | "dev %s: %llu sectors, offset %llu\n", | |
338 | (unsigned long long)bio->bi_sector, | |
1da177e4 | 339 | bdevname(tmp_dev->rdev->bdev, b), |
6283815d AN |
340 | (unsigned long long)tmp_dev->num_sectors, |
341 | (unsigned long long)tmp_dev->start_sector); | |
6712ecf8 | 342 | bio_io_error(bio); |
1da177e4 LT |
343 | return 0; |
344 | } | |
345 | if (unlikely(bio->bi_sector + (bio->bi_size >> 9) > | |
6283815d | 346 | tmp_dev->start_sector + tmp_dev->num_sectors)) { |
1da177e4 LT |
347 | /* This bio crosses a device boundary, so we have to |
348 | * split it. | |
349 | */ | |
350 | struct bio_pair *bp; | |
6283815d | 351 | |
6feef531 | 352 | bp = bio_split(bio, |
6283815d AN |
353 | tmp_dev->start_sector + tmp_dev->num_sectors |
354 | - bio->bi_sector); | |
355 | ||
1da177e4 LT |
356 | if (linear_make_request(q, &bp->bio1)) |
357 | generic_make_request(&bp->bio1); | |
358 | if (linear_make_request(q, &bp->bio2)) | |
359 | generic_make_request(&bp->bio2); | |
360 | bio_pair_release(bp); | |
361 | return 0; | |
362 | } | |
363 | ||
364 | bio->bi_bdev = tmp_dev->rdev->bdev; | |
6283815d AN |
365 | bio->bi_sector = bio->bi_sector - tmp_dev->start_sector |
366 | + tmp_dev->rdev->data_offset; | |
1da177e4 LT |
367 | |
368 | return 1; | |
369 | } | |
370 | ||
371 | static void linear_status (struct seq_file *seq, mddev_t *mddev) | |
372 | { | |
373 | ||
1da177e4 LT |
374 | seq_printf(seq, " %dk rounding", mddev->chunk_size/1024); |
375 | } | |
376 | ||
377 | ||
2604b703 | 378 | static struct mdk_personality linear_personality = |
1da177e4 LT |
379 | { |
380 | .name = "linear", | |
2604b703 | 381 | .level = LEVEL_LINEAR, |
1da177e4 LT |
382 | .owner = THIS_MODULE, |
383 | .make_request = linear_make_request, | |
384 | .run = linear_run, | |
385 | .stop = linear_stop, | |
386 | .status = linear_status, | |
7c7546cc | 387 | .hot_add_disk = linear_add, |
1da177e4 LT |
388 | }; |
389 | ||
390 | static int __init linear_init (void) | |
391 | { | |
2604b703 | 392 | return register_md_personality (&linear_personality); |
1da177e4 LT |
393 | } |
394 | ||
395 | static void linear_exit (void) | |
396 | { | |
2604b703 | 397 | unregister_md_personality (&linear_personality); |
1da177e4 LT |
398 | } |
399 | ||
400 | ||
401 | module_init(linear_init); | |
402 | module_exit(linear_exit); | |
403 | MODULE_LICENSE("GPL"); | |
d9d166c2 N |
404 | MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/ |
405 | MODULE_ALIAS("md-linear"); | |
2604b703 | 406 | MODULE_ALIAS("md-level--1"); |