md: raid0: make hash_spacing and preshift sector-based.
[linux-2.6-block.git] / drivers / md / linear.c
CommitLineData
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 */
24static 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
49static 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 75static 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
87static 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 100static 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 *));
f1cd14ae
AN
151 if (min_sectors == 0)
152 min_sectors = 1;
15945fee 153
23242fbb 154 /* min_sectors is the minimum spacing that will fit the hash
15945fee
N
155 * table in one PAGE. This may be much smaller than needed.
156 * We find the smallest non-terminal set of consecutive devices
23242fbb 157 * that is larger than min_sectors and use the size of that as
15945fee
N
158 * the actual spacing
159 */
ab5bd5cb 160 conf->spacing = conf->array_sectors;
15945fee 161 for (i=0; i < cnt-1 ; i++) {
23242fbb 162 sector_t tmp = 0;
15945fee 163 int j;
23242fbb
AN
164 for (j = i; j < cnt - 1 && tmp < min_sectors; j++)
165 tmp += conf->disks[j].num_sectors;
ab5bd5cb
AN
166 if (tmp >= min_sectors && tmp < conf->spacing)
167 conf->spacing = tmp;
15945fee
N
168 }
169
ab5bd5cb 170 /* spacing may be too large for sector_div to work with,
15945fee
N
171 * so we might need to pre-shift
172 */
ab5bd5cb 173 conf->sector_shift = 0;
15945fee 174 if (sizeof(sector_t) > sizeof(u32)) {
ab5bd5cb 175 sector_t space = conf->spacing;
15945fee
N
176 while (space > (sector_t)(~(u32)0)) {
177 space >>= 1;
ab5bd5cb 178 conf->sector_shift++;
15945fee
N
179 }
180 }
1da177e4
LT
181 /*
182 * This code was restructured to work around a gcc-2.95.3 internal
183 * compiler error. Alter it with care.
184 */
185 {
186 sector_t sz;
187 unsigned round;
188 unsigned long base;
189
ab5bd5cb 190 sz = conf->array_sectors >> conf->sector_shift;
15945fee 191 sz += 1; /* force round-up */
ab5bd5cb 192 base = conf->spacing >> conf->sector_shift;
1da177e4 193 round = sector_div(sz, base);
15945fee 194 nb_zone = sz + (round ? 1 : 0);
1da177e4 195 }
15945fee
N
196 BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *));
197
198 conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone,
1da177e4
LT
199 GFP_KERNEL);
200 if (!conf->hash_table)
201 goto out;
202
203 /*
204 * Here we generate the linear hash table
15945fee 205 * First calculate the device offsets.
1da177e4 206 */
6283815d 207 conf->disks[0].start_sector = 0;
a778b73f 208 for (i = 1; i < raid_disks; i++)
6283815d
AN
209 conf->disks[i].start_sector =
210 conf->disks[i-1].start_sector +
211 conf->disks[i-1].num_sectors;
15945fee 212
1da177e4 213 table = conf->hash_table;
15945fee 214 i = 0;
6283815d
AN
215 for (curr_sector = 0;
216 curr_sector < conf->array_sectors;
ab5bd5cb 217 curr_sector += conf->spacing) {
1da177e4 218
a778b73f 219 while (i < raid_disks-1 &&
6283815d 220 curr_sector >= conf->disks[i+1].start_sector)
15945fee 221 i++;
1da177e4 222
15945fee
N
223 *table ++ = conf->disks + i;
224 }
225
ab5bd5cb
AN
226 if (conf->sector_shift) {
227 conf->spacing >>= conf->sector_shift;
228 /* round spacing up so that when we divide by it,
15945fee 229 * we err on the side of "too-low", which is safest.
1da177e4 230 */
ab5bd5cb 231 conf->spacing++;
1da177e4 232 }
15945fee
N
233
234 BUG_ON(table - conf->hash_table > nb_zone);
1da177e4 235
7c7546cc
N
236 return conf;
237
238out:
239 kfree(conf);
240 return NULL;
241}
242
243static int linear_run (mddev_t *mddev)
244{
245 linear_conf_t *conf;
246
e7e72bf6 247 mddev->queue->queue_lock = &mddev->queue->__queue_lock;
7c7546cc
N
248 conf = linear_conf(mddev, mddev->raid_disks);
249
250 if (!conf)
251 return 1;
252 mddev->private = conf;
d6e22150 253 mddev->array_sectors = conf->array_sectors;
7c7546cc 254
1da177e4
LT
255 blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
256 mddev->queue->unplug_fn = linear_unplug;
26be34dc
N
257 mddev->queue->backing_dev_info.congested_fn = linear_congested;
258 mddev->queue->backing_dev_info.congested_data = mddev;
1da177e4 259 return 0;
7c7546cc 260}
1da177e4 261
7c7546cc
N
262static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
263{
264 /* Adding a drive to a linear array allows the array to grow.
265 * It is permitted if the new drive has a matching superblock
266 * already on it, with raid_disk equal to raid_disks.
267 * It is achieved by creating a new linear_private_data structure
268 * and swapping it in in-place of the current one.
269 * The current one is never freed until the array is stopped.
270 * This avoids races.
271 */
272 linear_conf_t *newconf;
273
a778b73f 274 if (rdev->saved_raid_disk != mddev->raid_disks)
7c7546cc
N
275 return -EINVAL;
276
a778b73f
N
277 rdev->raid_disk = rdev->saved_raid_disk;
278
7c7546cc
N
279 newconf = linear_conf(mddev,mddev->raid_disks+1);
280
281 if (!newconf)
282 return -ENOMEM;
283
284 newconf->prev = mddev_to_conf(mddev);
285 mddev->private = newconf;
286 mddev->raid_disks++;
d6e22150 287 mddev->array_sectors = newconf->array_sectors;
f233ea5c 288 set_capacity(mddev->gendisk, mddev->array_sectors);
7c7546cc 289 return 0;
1da177e4
LT
290}
291
292static int linear_stop (mddev_t *mddev)
293{
294 linear_conf_t *conf = mddev_to_conf(mddev);
295
296 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
7c7546cc
N
297 do {
298 linear_conf_t *t = conf->prev;
299 kfree(conf->hash_table);
300 kfree(conf);
301 conf = t;
302 } while (conf);
1da177e4
LT
303
304 return 0;
305}
306
165125e1 307static int linear_make_request (struct request_queue *q, struct bio *bio)
1da177e4 308{
a362357b 309 const int rw = bio_data_dir(bio);
1da177e4
LT
310 mddev_t *mddev = q->queuedata;
311 dev_info_t *tmp_dev;
c9959059 312 int cpu;
1da177e4 313
e5dcdd80 314 if (unlikely(bio_barrier(bio))) {
6712ecf8 315 bio_endio(bio, -EOPNOTSUPP);
e5dcdd80
N
316 return 0;
317 }
318
074a7aca
TH
319 cpu = part_stat_lock();
320 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
321 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
322 bio_sectors(bio));
323 part_stat_unlock();
1da177e4
LT
324
325 tmp_dev = which_dev(mddev, bio->bi_sector);
1da177e4 326
6283815d
AN
327 if (unlikely(bio->bi_sector >= (tmp_dev->num_sectors +
328 tmp_dev->start_sector)
329 || (bio->bi_sector <
330 tmp_dev->start_sector))) {
1da177e4
LT
331 char b[BDEVNAME_SIZE];
332
6283815d
AN
333 printk("linear_make_request: Sector %llu out of bounds on "
334 "dev %s: %llu sectors, offset %llu\n",
335 (unsigned long long)bio->bi_sector,
1da177e4 336 bdevname(tmp_dev->rdev->bdev, b),
6283815d
AN
337 (unsigned long long)tmp_dev->num_sectors,
338 (unsigned long long)tmp_dev->start_sector);
6712ecf8 339 bio_io_error(bio);
1da177e4
LT
340 return 0;
341 }
342 if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
6283815d 343 tmp_dev->start_sector + tmp_dev->num_sectors)) {
1da177e4
LT
344 /* This bio crosses a device boundary, so we have to
345 * split it.
346 */
347 struct bio_pair *bp;
6283815d 348
6feef531 349 bp = bio_split(bio,
6283815d
AN
350 tmp_dev->start_sector + tmp_dev->num_sectors
351 - bio->bi_sector);
352
1da177e4
LT
353 if (linear_make_request(q, &bp->bio1))
354 generic_make_request(&bp->bio1);
355 if (linear_make_request(q, &bp->bio2))
356 generic_make_request(&bp->bio2);
357 bio_pair_release(bp);
358 return 0;
359 }
360
361 bio->bi_bdev = tmp_dev->rdev->bdev;
6283815d
AN
362 bio->bi_sector = bio->bi_sector - tmp_dev->start_sector
363 + tmp_dev->rdev->data_offset;
1da177e4
LT
364
365 return 1;
366}
367
368static void linear_status (struct seq_file *seq, mddev_t *mddev)
369{
370
1da177e4
LT
371 seq_printf(seq, " %dk rounding", mddev->chunk_size/1024);
372}
373
374
2604b703 375static struct mdk_personality linear_personality =
1da177e4
LT
376{
377 .name = "linear",
2604b703 378 .level = LEVEL_LINEAR,
1da177e4
LT
379 .owner = THIS_MODULE,
380 .make_request = linear_make_request,
381 .run = linear_run,
382 .stop = linear_stop,
383 .status = linear_status,
7c7546cc 384 .hot_add_disk = linear_add,
1da177e4
LT
385};
386
387static int __init linear_init (void)
388{
2604b703 389 return register_md_personality (&linear_personality);
1da177e4
LT
390}
391
392static void linear_exit (void)
393{
2604b703 394 unregister_md_personality (&linear_personality);
1da177e4
LT
395}
396
397
398module_init(linear_init);
399module_exit(linear_exit);
400MODULE_LICENSE("GPL");
d9d166c2
N
401MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
402MODULE_ALIAS("md-linear");
2604b703 403MODULE_ALIAS("md-level--1");