]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | linear.c : Multiple Devices driver for Linux | |
3 | Copyright (C) 1994-96 Marc ZYNGIER | |
4 | <[email protected]> or | |
5 | <[email protected]> | |
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 | ||
19 | #include <linux/module.h> | |
20 | ||
21 | #include <linux/raid/md.h> | |
22 | #include <linux/slab.h> | |
23 | #include <linux/raid/linear.h> | |
24 | ||
25 | #define MAJOR_NR MD_MAJOR | |
26 | #define MD_DRIVER | |
27 | #define MD_PERSONALITY | |
28 | ||
29 | /* | |
30 | * find which device holds a particular offset | |
31 | */ | |
32 | static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector) | |
33 | { | |
34 | dev_info_t *hash; | |
35 | linear_conf_t *conf = mddev_to_conf(mddev); | |
36 | sector_t block = sector >> 1; | |
37 | ||
38 | /* | |
39 | * sector_div(a,b) returns the remainer and sets a to a/b | |
40 | */ | |
15945fee N |
41 | block >>= conf->preshift; |
42 | (void)sector_div(block, conf->hash_spacing); | |
1da177e4 LT |
43 | hash = conf->hash_table[block]; |
44 | ||
45 | while ((sector>>1) >= (hash->size + hash->offset)) | |
46 | hash++; | |
47 | return hash; | |
48 | } | |
49 | ||
50 | /** | |
15945fee | 51 | * linear_mergeable_bvec -- tell bio layer if two requests can be merged |
1da177e4 LT |
52 | * @q: request queue |
53 | * @bio: the buffer head that's been built up so far | |
54 | * @biovec: the request that could be merged to it. | |
55 | * | |
56 | * Return amount of bytes we can take at this offset | |
57 | */ | |
165125e1 | 58 | static int linear_mergeable_bvec(struct request_queue *q, struct bio *bio, struct bio_vec *biovec) |
1da177e4 LT |
59 | { |
60 | mddev_t *mddev = q->queuedata; | |
61 | dev_info_t *dev0; | |
62 | unsigned long maxsectors, bio_sectors = bio->bi_size >> 9; | |
63 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
64 | ||
65 | dev0 = which_dev(mddev, sector); | |
66 | maxsectors = (dev0->size << 1) - (sector - (dev0->offset<<1)); | |
67 | ||
68 | if (maxsectors < bio_sectors) | |
69 | maxsectors = 0; | |
70 | else | |
71 | maxsectors -= bio_sectors; | |
72 | ||
73 | if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0) | |
74 | return biovec->bv_len; | |
75 | /* The bytes available at this offset could be really big, | |
76 | * so we cap at 2^31 to avoid overflow */ | |
77 | if (maxsectors > (1 << (31-9))) | |
78 | return 1<<31; | |
79 | return maxsectors << 9; | |
80 | } | |
81 | ||
165125e1 | 82 | static void linear_unplug(struct request_queue *q) |
1da177e4 LT |
83 | { |
84 | mddev_t *mddev = q->queuedata; | |
85 | linear_conf_t *conf = mddev_to_conf(mddev); | |
86 | int i; | |
87 | ||
88 | for (i=0; i < mddev->raid_disks; i++) { | |
165125e1 | 89 | struct request_queue *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev); |
2ad8b1ef | 90 | blk_unplug(r_queue); |
1da177e4 LT |
91 | } |
92 | } | |
93 | ||
26be34dc N |
94 | static int linear_congested(void *data, int bits) |
95 | { | |
96 | mddev_t *mddev = data; | |
97 | linear_conf_t *conf = mddev_to_conf(mddev); | |
98 | int i, ret = 0; | |
99 | ||
100 | for (i = 0; i < mddev->raid_disks && !ret ; i++) { | |
165125e1 | 101 | struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev); |
26be34dc N |
102 | ret |= bdi_congested(&q->backing_dev_info, bits); |
103 | } | |
104 | return ret; | |
105 | } | |
106 | ||
7c7546cc | 107 | static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks) |
1da177e4 LT |
108 | { |
109 | linear_conf_t *conf; | |
110 | dev_info_t **table; | |
111 | mdk_rdev_t *rdev; | |
112 | int i, nb_zone, cnt; | |
15945fee | 113 | sector_t min_spacing; |
1da177e4 LT |
114 | sector_t curr_offset; |
115 | struct list_head *tmp; | |
116 | ||
7c7546cc | 117 | conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t), |
1da177e4 LT |
118 | GFP_KERNEL); |
119 | if (!conf) | |
7c7546cc N |
120 | return NULL; |
121 | ||
1da177e4 | 122 | cnt = 0; |
7c7546cc | 123 | conf->array_size = 0; |
1da177e4 | 124 | |
d089c6af | 125 | rdev_for_each(rdev, tmp, mddev) { |
1da177e4 LT |
126 | int j = rdev->raid_disk; |
127 | dev_info_t *disk = conf->disks + j; | |
128 | ||
7c7546cc | 129 | if (j < 0 || j > raid_disks || disk->rdev) { |
1da177e4 LT |
130 | printk("linear: disk numbering problem. Aborting!\n"); |
131 | goto out; | |
132 | } | |
133 | ||
134 | disk->rdev = rdev; | |
135 | ||
136 | blk_queue_stack_limits(mddev->queue, | |
137 | rdev->bdev->bd_disk->queue); | |
138 | /* as we don't honour merge_bvec_fn, we must never risk | |
139 | * violating it, so limit ->max_sector to one PAGE, as | |
140 | * a one page request is never in violation. | |
141 | */ | |
142 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn && | |
143 | mddev->queue->max_sectors > (PAGE_SIZE>>9)) | |
144 | blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9); | |
145 | ||
146 | disk->size = rdev->size; | |
7c7546cc | 147 | conf->array_size += rdev->size; |
1da177e4 | 148 | |
1da177e4 LT |
149 | cnt++; |
150 | } | |
7c7546cc | 151 | if (cnt != raid_disks) { |
1da177e4 LT |
152 | printk("linear: not enough drives present. Aborting!\n"); |
153 | goto out; | |
154 | } | |
155 | ||
f9abd1ac | 156 | min_spacing = conf->array_size; |
15945fee N |
157 | sector_div(min_spacing, PAGE_SIZE/sizeof(struct dev_info *)); |
158 | ||
159 | /* min_spacing is the minimum spacing that will fit the hash | |
160 | * table in one PAGE. This may be much smaller than needed. | |
161 | * We find the smallest non-terminal set of consecutive devices | |
162 | * that is larger than min_spacing as use the size of that as | |
163 | * the actual spacing | |
164 | */ | |
f9abd1ac | 165 | conf->hash_spacing = conf->array_size; |
15945fee N |
166 | for (i=0; i < cnt-1 ; i++) { |
167 | sector_t sz = 0; | |
168 | int j; | |
bed31ed9 | 169 | for (j = i; j < cnt - 1 && sz < min_spacing; j++) |
15945fee N |
170 | sz += conf->disks[j].size; |
171 | if (sz >= min_spacing && sz < conf->hash_spacing) | |
172 | conf->hash_spacing = sz; | |
173 | } | |
174 | ||
175 | /* hash_spacing may be too large for sector_div to work with, | |
176 | * so we might need to pre-shift | |
177 | */ | |
178 | conf->preshift = 0; | |
179 | if (sizeof(sector_t) > sizeof(u32)) { | |
180 | sector_t space = conf->hash_spacing; | |
181 | while (space > (sector_t)(~(u32)0)) { | |
182 | space >>= 1; | |
183 | conf->preshift++; | |
184 | } | |
185 | } | |
1da177e4 LT |
186 | /* |
187 | * This code was restructured to work around a gcc-2.95.3 internal | |
188 | * compiler error. Alter it with care. | |
189 | */ | |
190 | { | |
191 | sector_t sz; | |
192 | unsigned round; | |
193 | unsigned long base; | |
194 | ||
7c7546cc | 195 | sz = conf->array_size >> conf->preshift; |
15945fee N |
196 | sz += 1; /* force round-up */ |
197 | base = conf->hash_spacing >> conf->preshift; | |
1da177e4 | 198 | round = sector_div(sz, base); |
15945fee | 199 | nb_zone = sz + (round ? 1 : 0); |
1da177e4 | 200 | } |
15945fee N |
201 | BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *)); |
202 | ||
203 | conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone, | |
1da177e4 LT |
204 | GFP_KERNEL); |
205 | if (!conf->hash_table) | |
206 | goto out; | |
207 | ||
208 | /* | |
209 | * Here we generate the linear hash table | |
15945fee | 210 | * First calculate the device offsets. |
1da177e4 | 211 | */ |
15945fee | 212 | conf->disks[0].offset = 0; |
a778b73f | 213 | for (i = 1; i < raid_disks; i++) |
15945fee N |
214 | conf->disks[i].offset = |
215 | conf->disks[i-1].offset + | |
216 | conf->disks[i-1].size; | |
217 | ||
1da177e4 | 218 | table = conf->hash_table; |
1da177e4 | 219 | curr_offset = 0; |
15945fee N |
220 | i = 0; |
221 | for (curr_offset = 0; | |
f9abd1ac | 222 | curr_offset < conf->array_size; |
15945fee | 223 | curr_offset += conf->hash_spacing) { |
1da177e4 | 224 | |
a778b73f | 225 | while (i < raid_disks-1 && |
15945fee N |
226 | curr_offset >= conf->disks[i+1].offset) |
227 | i++; | |
1da177e4 | 228 | |
15945fee N |
229 | *table ++ = conf->disks + i; |
230 | } | |
231 | ||
232 | if (conf->preshift) { | |
233 | conf->hash_spacing >>= conf->preshift; | |
234 | /* round hash_spacing up so that when we divide by it, | |
235 | * we err on the side of "too-low", which is safest. | |
1da177e4 | 236 | */ |
15945fee | 237 | conf->hash_spacing++; |
1da177e4 | 238 | } |
15945fee N |
239 | |
240 | BUG_ON(table - conf->hash_table > nb_zone); | |
1da177e4 | 241 | |
7c7546cc N |
242 | return conf; |
243 | ||
244 | out: | |
245 | kfree(conf); | |
246 | return NULL; | |
247 | } | |
248 | ||
249 | static int linear_run (mddev_t *mddev) | |
250 | { | |
251 | linear_conf_t *conf; | |
252 | ||
e7e72bf6 | 253 | mddev->queue->queue_lock = &mddev->queue->__queue_lock; |
7c7546cc N |
254 | conf = linear_conf(mddev, mddev->raid_disks); |
255 | ||
256 | if (!conf) | |
257 | return 1; | |
258 | mddev->private = conf; | |
259 | mddev->array_size = conf->array_size; | |
260 | ||
1da177e4 LT |
261 | blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec); |
262 | mddev->queue->unplug_fn = linear_unplug; | |
26be34dc N |
263 | mddev->queue->backing_dev_info.congested_fn = linear_congested; |
264 | mddev->queue->backing_dev_info.congested_data = mddev; | |
1da177e4 | 265 | return 0; |
7c7546cc | 266 | } |
1da177e4 | 267 | |
7c7546cc N |
268 | static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev) |
269 | { | |
270 | /* Adding a drive to a linear array allows the array to grow. | |
271 | * It is permitted if the new drive has a matching superblock | |
272 | * already on it, with raid_disk equal to raid_disks. | |
273 | * It is achieved by creating a new linear_private_data structure | |
274 | * and swapping it in in-place of the current one. | |
275 | * The current one is never freed until the array is stopped. | |
276 | * This avoids races. | |
277 | */ | |
278 | linear_conf_t *newconf; | |
279 | ||
a778b73f | 280 | if (rdev->saved_raid_disk != mddev->raid_disks) |
7c7546cc N |
281 | return -EINVAL; |
282 | ||
a778b73f N |
283 | rdev->raid_disk = rdev->saved_raid_disk; |
284 | ||
7c7546cc N |
285 | newconf = linear_conf(mddev,mddev->raid_disks+1); |
286 | ||
287 | if (!newconf) | |
288 | return -ENOMEM; | |
289 | ||
290 | newconf->prev = mddev_to_conf(mddev); | |
291 | mddev->private = newconf; | |
292 | mddev->raid_disks++; | |
293 | mddev->array_size = newconf->array_size; | |
294 | set_capacity(mddev->gendisk, mddev->array_size << 1); | |
295 | return 0; | |
1da177e4 LT |
296 | } |
297 | ||
298 | static int linear_stop (mddev_t *mddev) | |
299 | { | |
300 | linear_conf_t *conf = mddev_to_conf(mddev); | |
301 | ||
302 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
7c7546cc N |
303 | do { |
304 | linear_conf_t *t = conf->prev; | |
305 | kfree(conf->hash_table); | |
306 | kfree(conf); | |
307 | conf = t; | |
308 | } while (conf); | |
1da177e4 LT |
309 | |
310 | return 0; | |
311 | } | |
312 | ||
165125e1 | 313 | static int linear_make_request (struct request_queue *q, struct bio *bio) |
1da177e4 | 314 | { |
a362357b | 315 | const int rw = bio_data_dir(bio); |
1da177e4 LT |
316 | mddev_t *mddev = q->queuedata; |
317 | dev_info_t *tmp_dev; | |
318 | sector_t block; | |
319 | ||
e5dcdd80 | 320 | if (unlikely(bio_barrier(bio))) { |
6712ecf8 | 321 | bio_endio(bio, -EOPNOTSUPP); |
e5dcdd80 N |
322 | return 0; |
323 | } | |
324 | ||
a362357b JA |
325 | disk_stat_inc(mddev->gendisk, ios[rw]); |
326 | disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio)); | |
1da177e4 LT |
327 | |
328 | tmp_dev = which_dev(mddev, bio->bi_sector); | |
329 | block = bio->bi_sector >> 1; | |
330 | ||
331 | if (unlikely(block >= (tmp_dev->size + tmp_dev->offset) | |
332 | || block < tmp_dev->offset)) { | |
333 | char b[BDEVNAME_SIZE]; | |
334 | ||
335 | printk("linear_make_request: Block %llu out of bounds on " | |
336 | "dev %s size %llu offset %llu\n", | |
337 | (unsigned long long)block, | |
338 | bdevname(tmp_dev->rdev->bdev, b), | |
339 | (unsigned long long)tmp_dev->size, | |
340 | (unsigned long long)tmp_dev->offset); | |
6712ecf8 | 341 | bio_io_error(bio); |
1da177e4 LT |
342 | return 0; |
343 | } | |
344 | if (unlikely(bio->bi_sector + (bio->bi_size >> 9) > | |
345 | (tmp_dev->offset + tmp_dev->size)<<1)) { | |
346 | /* This bio crosses a device boundary, so we have to | |
347 | * split it. | |
348 | */ | |
349 | struct bio_pair *bp; | |
29ac8e05 N |
350 | bp = bio_split(bio, bio_split_pool, |
351 | ((tmp_dev->offset + tmp_dev->size)<<1) - bio->bi_sector); | |
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; | |
361 | bio->bi_sector = bio->bi_sector - (tmp_dev->offset << 1) + tmp_dev->rdev->data_offset; | |
362 | ||
363 | return 1; | |
364 | } | |
365 | ||
366 | static void linear_status (struct seq_file *seq, mddev_t *mddev) | |
367 | { | |
368 | ||
369 | #undef MD_DEBUG | |
370 | #ifdef MD_DEBUG | |
371 | int j; | |
372 | linear_conf_t *conf = mddev_to_conf(mddev); | |
373 | sector_t s = 0; | |
374 | ||
375 | seq_printf(seq, " "); | |
15945fee | 376 | for (j = 0; j < mddev->raid_disks; j++) |
1da177e4 LT |
377 | { |
378 | char b[BDEVNAME_SIZE]; | |
379 | s += conf->smallest_size; | |
380 | seq_printf(seq, "[%s", | |
381 | bdevname(conf->hash_table[j][0].rdev->bdev,b)); | |
382 | ||
383 | while (s > conf->hash_table[j][0].offset + | |
384 | conf->hash_table[j][0].size) | |
385 | seq_printf(seq, "/%s] ", | |
386 | bdevname(conf->hash_table[j][1].rdev->bdev,b)); | |
387 | else | |
388 | seq_printf(seq, "] "); | |
389 | } | |
390 | seq_printf(seq, "\n"); | |
391 | #endif | |
392 | seq_printf(seq, " %dk rounding", mddev->chunk_size/1024); | |
393 | } | |
394 | ||
395 | ||
2604b703 | 396 | static struct mdk_personality linear_personality = |
1da177e4 LT |
397 | { |
398 | .name = "linear", | |
2604b703 | 399 | .level = LEVEL_LINEAR, |
1da177e4 LT |
400 | .owner = THIS_MODULE, |
401 | .make_request = linear_make_request, | |
402 | .run = linear_run, | |
403 | .stop = linear_stop, | |
404 | .status = linear_status, | |
7c7546cc | 405 | .hot_add_disk = linear_add, |
1da177e4 LT |
406 | }; |
407 | ||
408 | static int __init linear_init (void) | |
409 | { | |
2604b703 | 410 | return register_md_personality (&linear_personality); |
1da177e4 LT |
411 | } |
412 | ||
413 | static void linear_exit (void) | |
414 | { | |
2604b703 | 415 | unregister_md_personality (&linear_personality); |
1da177e4 LT |
416 | } |
417 | ||
418 | ||
419 | module_init(linear_init); | |
420 | module_exit(linear_exit); | |
421 | MODULE_LICENSE("GPL"); | |
d9d166c2 N |
422 | MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/ |
423 | MODULE_ALIAS("md-linear"); | |
2604b703 | 424 | MODULE_ALIAS("md-level--1"); |