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 LT |
108 | struct list_head *tmp; |
109 | ||
7c7546cc | 110 | conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t), |
1da177e4 LT |
111 | GFP_KERNEL); |
112 | if (!conf) | |
7c7546cc N |
113 | return NULL; |
114 | ||
1da177e4 | 115 | cnt = 0; |
d6e22150 | 116 | conf->array_sectors = 0; |
1da177e4 | 117 | |
d089c6af | 118 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
119 | int j = rdev->raid_disk; |
120 | dev_info_t *disk = conf->disks + j; | |
121 | ||
13864515 | 122 | if (j < 0 || j >= raid_disks || disk->rdev) { |
1da177e4 LT |
123 | printk("linear: disk numbering problem. Aborting!\n"); |
124 | goto out; | |
125 | } | |
126 | ||
127 | disk->rdev = rdev; | |
128 | ||
129 | blk_queue_stack_limits(mddev->queue, | |
130 | rdev->bdev->bd_disk->queue); | |
131 | /* as we don't honour merge_bvec_fn, we must never risk | |
132 | * violating it, so limit ->max_sector to one PAGE, as | |
133 | * a one page request is never in violation. | |
134 | */ | |
135 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn && | |
136 | mddev->queue->max_sectors > (PAGE_SIZE>>9)) | |
137 | blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9); | |
138 | ||
6283815d | 139 | disk->num_sectors = rdev->size * 2; |
d6e22150 | 140 | conf->array_sectors += rdev->size * 2; |
1da177e4 | 141 | |
1da177e4 LT |
142 | cnt++; |
143 | } | |
7c7546cc | 144 | if (cnt != raid_disks) { |
1da177e4 LT |
145 | printk("linear: not enough drives present. Aborting!\n"); |
146 | goto out; | |
147 | } | |
148 | ||
23242fbb AN |
149 | min_sectors = conf->array_sectors; |
150 | sector_div(min_sectors, PAGE_SIZE/sizeof(struct dev_info *)); | |
15945fee | 151 | |
23242fbb | 152 | /* min_sectors is the minimum spacing that will fit the hash |
15945fee N |
153 | * table in one PAGE. This may be much smaller than needed. |
154 | * We find the smallest non-terminal set of consecutive devices | |
23242fbb | 155 | * that is larger than min_sectors and use the size of that as |
15945fee N |
156 | * the actual spacing |
157 | */ | |
ab5bd5cb | 158 | conf->spacing = conf->array_sectors; |
15945fee | 159 | for (i=0; i < cnt-1 ; i++) { |
23242fbb | 160 | sector_t tmp = 0; |
15945fee | 161 | int j; |
23242fbb AN |
162 | for (j = i; j < cnt - 1 && tmp < min_sectors; j++) |
163 | tmp += conf->disks[j].num_sectors; | |
ab5bd5cb AN |
164 | if (tmp >= min_sectors && tmp < conf->spacing) |
165 | conf->spacing = tmp; | |
15945fee N |
166 | } |
167 | ||
ab5bd5cb | 168 | /* spacing may be too large for sector_div to work with, |
15945fee N |
169 | * so we might need to pre-shift |
170 | */ | |
ab5bd5cb | 171 | conf->sector_shift = 0; |
15945fee | 172 | if (sizeof(sector_t) > sizeof(u32)) { |
ab5bd5cb | 173 | sector_t space = conf->spacing; |
15945fee N |
174 | while (space > (sector_t)(~(u32)0)) { |
175 | space >>= 1; | |
ab5bd5cb | 176 | conf->sector_shift++; |
15945fee N |
177 | } |
178 | } | |
1da177e4 LT |
179 | /* |
180 | * This code was restructured to work around a gcc-2.95.3 internal | |
181 | * compiler error. Alter it with care. | |
182 | */ | |
183 | { | |
184 | sector_t sz; | |
185 | unsigned round; | |
186 | unsigned long base; | |
187 | ||
ab5bd5cb | 188 | sz = conf->array_sectors >> conf->sector_shift; |
15945fee | 189 | sz += 1; /* force round-up */ |
ab5bd5cb | 190 | base = conf->spacing >> conf->sector_shift; |
1da177e4 | 191 | round = sector_div(sz, base); |
15945fee | 192 | nb_zone = sz + (round ? 1 : 0); |
1da177e4 | 193 | } |
15945fee N |
194 | BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *)); |
195 | ||
196 | conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone, | |
1da177e4 LT |
197 | GFP_KERNEL); |
198 | if (!conf->hash_table) | |
199 | goto out; | |
200 | ||
201 | /* | |
202 | * Here we generate the linear hash table | |
15945fee | 203 | * First calculate the device offsets. |
1da177e4 | 204 | */ |
6283815d | 205 | conf->disks[0].start_sector = 0; |
a778b73f | 206 | for (i = 1; i < raid_disks; i++) |
6283815d AN |
207 | conf->disks[i].start_sector = |
208 | conf->disks[i-1].start_sector + | |
209 | conf->disks[i-1].num_sectors; | |
15945fee | 210 | |
1da177e4 | 211 | table = conf->hash_table; |
15945fee | 212 | i = 0; |
6283815d AN |
213 | for (curr_sector = 0; |
214 | curr_sector < conf->array_sectors; | |
ab5bd5cb | 215 | curr_sector += conf->spacing) { |
1da177e4 | 216 | |
a778b73f | 217 | while (i < raid_disks-1 && |
6283815d | 218 | curr_sector >= conf->disks[i+1].start_sector) |
15945fee | 219 | i++; |
1da177e4 | 220 | |
15945fee N |
221 | *table ++ = conf->disks + i; |
222 | } | |
223 | ||
ab5bd5cb AN |
224 | if (conf->sector_shift) { |
225 | conf->spacing >>= conf->sector_shift; | |
226 | /* round spacing up so that when we divide by it, | |
15945fee | 227 | * we err on the side of "too-low", which is safest. |
1da177e4 | 228 | */ |
ab5bd5cb | 229 | conf->spacing++; |
1da177e4 | 230 | } |
15945fee N |
231 | |
232 | BUG_ON(table - conf->hash_table > nb_zone); | |
1da177e4 | 233 | |
7c7546cc N |
234 | return conf; |
235 | ||
236 | out: | |
237 | kfree(conf); | |
238 | return NULL; | |
239 | } | |
240 | ||
241 | static int linear_run (mddev_t *mddev) | |
242 | { | |
243 | linear_conf_t *conf; | |
244 | ||
e7e72bf6 | 245 | mddev->queue->queue_lock = &mddev->queue->__queue_lock; |
7c7546cc N |
246 | conf = linear_conf(mddev, mddev->raid_disks); |
247 | ||
248 | if (!conf) | |
249 | return 1; | |
250 | mddev->private = conf; | |
d6e22150 | 251 | mddev->array_sectors = conf->array_sectors; |
7c7546cc | 252 | |
1da177e4 LT |
253 | blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec); |
254 | mddev->queue->unplug_fn = linear_unplug; | |
26be34dc N |
255 | mddev->queue->backing_dev_info.congested_fn = linear_congested; |
256 | mddev->queue->backing_dev_info.congested_data = mddev; | |
1da177e4 | 257 | return 0; |
7c7546cc | 258 | } |
1da177e4 | 259 | |
7c7546cc N |
260 | static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev) |
261 | { | |
262 | /* Adding a drive to a linear array allows the array to grow. | |
263 | * It is permitted if the new drive has a matching superblock | |
264 | * already on it, with raid_disk equal to raid_disks. | |
265 | * It is achieved by creating a new linear_private_data structure | |
266 | * and swapping it in in-place of the current one. | |
267 | * The current one is never freed until the array is stopped. | |
268 | * This avoids races. | |
269 | */ | |
270 | linear_conf_t *newconf; | |
271 | ||
a778b73f | 272 | if (rdev->saved_raid_disk != mddev->raid_disks) |
7c7546cc N |
273 | return -EINVAL; |
274 | ||
a778b73f N |
275 | rdev->raid_disk = rdev->saved_raid_disk; |
276 | ||
7c7546cc N |
277 | newconf = linear_conf(mddev,mddev->raid_disks+1); |
278 | ||
279 | if (!newconf) | |
280 | return -ENOMEM; | |
281 | ||
282 | newconf->prev = mddev_to_conf(mddev); | |
283 | mddev->private = newconf; | |
284 | mddev->raid_disks++; | |
d6e22150 | 285 | mddev->array_sectors = newconf->array_sectors; |
f233ea5c | 286 | set_capacity(mddev->gendisk, mddev->array_sectors); |
7c7546cc | 287 | return 0; |
1da177e4 LT |
288 | } |
289 | ||
290 | static int linear_stop (mddev_t *mddev) | |
291 | { | |
292 | linear_conf_t *conf = mddev_to_conf(mddev); | |
293 | ||
294 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
7c7546cc N |
295 | do { |
296 | linear_conf_t *t = conf->prev; | |
297 | kfree(conf->hash_table); | |
298 | kfree(conf); | |
299 | conf = t; | |
300 | } while (conf); | |
1da177e4 LT |
301 | |
302 | return 0; | |
303 | } | |
304 | ||
165125e1 | 305 | static int linear_make_request (struct request_queue *q, struct bio *bio) |
1da177e4 | 306 | { |
a362357b | 307 | const int rw = bio_data_dir(bio); |
1da177e4 LT |
308 | mddev_t *mddev = q->queuedata; |
309 | dev_info_t *tmp_dev; | |
c9959059 | 310 | int cpu; |
1da177e4 | 311 | |
e5dcdd80 | 312 | if (unlikely(bio_barrier(bio))) { |
6712ecf8 | 313 | bio_endio(bio, -EOPNOTSUPP); |
e5dcdd80 N |
314 | return 0; |
315 | } | |
316 | ||
074a7aca TH |
317 | cpu = part_stat_lock(); |
318 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
319 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], | |
320 | bio_sectors(bio)); | |
321 | part_stat_unlock(); | |
1da177e4 LT |
322 | |
323 | tmp_dev = which_dev(mddev, bio->bi_sector); | |
1da177e4 | 324 | |
6283815d AN |
325 | if (unlikely(bio->bi_sector >= (tmp_dev->num_sectors + |
326 | tmp_dev->start_sector) | |
327 | || (bio->bi_sector < | |
328 | tmp_dev->start_sector))) { | |
1da177e4 LT |
329 | char b[BDEVNAME_SIZE]; |
330 | ||
6283815d AN |
331 | printk("linear_make_request: Sector %llu out of bounds on " |
332 | "dev %s: %llu sectors, offset %llu\n", | |
333 | (unsigned long long)bio->bi_sector, | |
1da177e4 | 334 | bdevname(tmp_dev->rdev->bdev, b), |
6283815d AN |
335 | (unsigned long long)tmp_dev->num_sectors, |
336 | (unsigned long long)tmp_dev->start_sector); | |
6712ecf8 | 337 | bio_io_error(bio); |
1da177e4 LT |
338 | return 0; |
339 | } | |
340 | if (unlikely(bio->bi_sector + (bio->bi_size >> 9) > | |
6283815d | 341 | tmp_dev->start_sector + tmp_dev->num_sectors)) { |
1da177e4 LT |
342 | /* This bio crosses a device boundary, so we have to |
343 | * split it. | |
344 | */ | |
345 | struct bio_pair *bp; | |
6283815d | 346 | |
6feef531 | 347 | bp = bio_split(bio, |
6283815d AN |
348 | tmp_dev->start_sector + tmp_dev->num_sectors |
349 | - bio->bi_sector); | |
350 | ||
1da177e4 LT |
351 | if (linear_make_request(q, &bp->bio1)) |
352 | generic_make_request(&bp->bio1); | |
353 | if (linear_make_request(q, &bp->bio2)) | |
354 | generic_make_request(&bp->bio2); | |
355 | bio_pair_release(bp); | |
356 | return 0; | |
357 | } | |
358 | ||
359 | bio->bi_bdev = tmp_dev->rdev->bdev; | |
6283815d AN |
360 | bio->bi_sector = bio->bi_sector - tmp_dev->start_sector |
361 | + tmp_dev->rdev->data_offset; | |
1da177e4 LT |
362 | |
363 | return 1; | |
364 | } | |
365 | ||
366 | static void linear_status (struct seq_file *seq, mddev_t *mddev) | |
367 | { | |
368 | ||
1da177e4 LT |
369 | seq_printf(seq, " %dk rounding", mddev->chunk_size/1024); |
370 | } | |
371 | ||
372 | ||
2604b703 | 373 | static struct mdk_personality linear_personality = |
1da177e4 LT |
374 | { |
375 | .name = "linear", | |
2604b703 | 376 | .level = LEVEL_LINEAR, |
1da177e4 LT |
377 | .owner = THIS_MODULE, |
378 | .make_request = linear_make_request, | |
379 | .run = linear_run, | |
380 | .stop = linear_stop, | |
381 | .status = linear_status, | |
7c7546cc | 382 | .hot_add_disk = linear_add, |
1da177e4 LT |
383 | }; |
384 | ||
385 | static int __init linear_init (void) | |
386 | { | |
2604b703 | 387 | return register_md_personality (&linear_personality); |
1da177e4 LT |
388 | } |
389 | ||
390 | static void linear_exit (void) | |
391 | { | |
2604b703 | 392 | unregister_md_personality (&linear_personality); |
1da177e4 LT |
393 | } |
394 | ||
395 | ||
396 | module_init(linear_init); | |
397 | module_exit(linear_exit); | |
398 | MODULE_LICENSE("GPL"); | |
d9d166c2 N |
399 | MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/ |
400 | MODULE_ALIAS("md-linear"); | |
2604b703 | 401 | MODULE_ALIAS("md-level--1"); |