Merge branch 'for-33' of git://repo.or.cz/linux-kbuild
[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
bff61975
N
19#include <linux/blkdev.h>
20#include <linux/raid/md_u.h>
bff61975 21#include <linux/seq_file.h>
43b2e5d8 22#include "md.h"
ef740c37 23#include "linear.h"
1da177e4 24
1da177e4
LT
25/*
26 * find which device holds a particular offset
27 */
28static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector)
29{
aece3d1f 30 int lo, mid, hi;
af11c397 31 linear_conf_t *conf;
1da177e4 32
aece3d1f
SS
33 lo = 0;
34 hi = mddev->raid_disks - 1;
af11c397 35 conf = rcu_dereference(mddev->private);
1da177e4 36
aece3d1f
SS
37 /*
38 * Binary Search
39 */
40
41 while (hi > lo) {
42
43 mid = (hi + lo) / 2;
44 if (sector < conf->disks[mid].end_sector)
45 hi = mid;
46 else
47 lo = mid + 1;
48 }
49
50 return conf->disks + lo;
1da177e4
LT
51}
52
53/**
15945fee 54 * linear_mergeable_bvec -- tell bio layer if two requests can be merged
1da177e4 55 * @q: request queue
cc371e66 56 * @bvm: properties of new bio
1da177e4
LT
57 * @biovec: the request that could be merged to it.
58 *
59 * Return amount of bytes we can take at this offset
60 */
cc371e66
AK
61static int linear_mergeable_bvec(struct request_queue *q,
62 struct bvec_merge_data *bvm,
63 struct bio_vec *biovec)
1da177e4
LT
64{
65 mddev_t *mddev = q->queuedata;
66 dev_info_t *dev0;
cc371e66
AK
67 unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9;
68 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
1da177e4 69
af11c397 70 rcu_read_lock();
1da177e4 71 dev0 = which_dev(mddev, sector);
4db7cdc8 72 maxsectors = dev0->end_sector - sector;
af11c397 73 rcu_read_unlock();
1da177e4
LT
74
75 if (maxsectors < bio_sectors)
76 maxsectors = 0;
77 else
78 maxsectors -= bio_sectors;
79
80 if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
81 return biovec->bv_len;
82 /* The bytes available at this offset could be really big,
83 * so we cap at 2^31 to avoid overflow */
84 if (maxsectors > (1 << (31-9)))
85 return 1<<31;
86 return maxsectors << 9;
87}
88
165125e1 89static void linear_unplug(struct request_queue *q)
1da177e4
LT
90{
91 mddev_t *mddev = q->queuedata;
af11c397 92 linear_conf_t *conf;
1da177e4
LT
93 int i;
94
af11c397
S
95 rcu_read_lock();
96 conf = rcu_dereference(mddev->private);
97
1da177e4 98 for (i=0; i < mddev->raid_disks; i++) {
165125e1 99 struct request_queue *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
2ad8b1ef 100 blk_unplug(r_queue);
1da177e4 101 }
af11c397 102 rcu_read_unlock();
1da177e4
LT
103}
104
26be34dc
N
105static int linear_congested(void *data, int bits)
106{
107 mddev_t *mddev = data;
af11c397 108 linear_conf_t *conf;
26be34dc
N
109 int i, ret = 0;
110
3fa841d7
N
111 if (mddev_congested(mddev, bits))
112 return 1;
113
af11c397
S
114 rcu_read_lock();
115 conf = rcu_dereference(mddev->private);
116
26be34dc 117 for (i = 0; i < mddev->raid_disks && !ret ; i++) {
165125e1 118 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
26be34dc
N
119 ret |= bdi_congested(&q->backing_dev_info, bits);
120 }
af11c397
S
121
122 rcu_read_unlock();
26be34dc
N
123 return ret;
124}
125
80c3a6ce
DW
126static sector_t linear_size(mddev_t *mddev, sector_t sectors, int raid_disks)
127{
af11c397
S
128 linear_conf_t *conf;
129 sector_t array_sectors;
80c3a6ce 130
af11c397
S
131 rcu_read_lock();
132 conf = rcu_dereference(mddev->private);
80c3a6ce
DW
133 WARN_ONCE(sectors || raid_disks,
134 "%s does not support generic reshape\n", __func__);
af11c397
S
135 array_sectors = conf->array_sectors;
136 rcu_read_unlock();
80c3a6ce 137
af11c397 138 return array_sectors;
80c3a6ce
DW
139}
140
7c7546cc 141static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
1da177e4
LT
142{
143 linear_conf_t *conf;
1da177e4 144 mdk_rdev_t *rdev;
45d4582f 145 int i, cnt;
1da177e4 146
7c7546cc 147 conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
1da177e4
LT
148 GFP_KERNEL);
149 if (!conf)
7c7546cc
N
150 return NULL;
151
1da177e4 152 cnt = 0;
d6e22150 153 conf->array_sectors = 0;
1da177e4 154
159ec1fc 155 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT
156 int j = rdev->raid_disk;
157 dev_info_t *disk = conf->disks + j;
13f2682b 158 sector_t sectors;
1da177e4 159
13864515 160 if (j < 0 || j >= raid_disks || disk->rdev) {
1da177e4
LT
161 printk("linear: disk numbering problem. Aborting!\n");
162 goto out;
163 }
164
165 disk->rdev = rdev;
13f2682b
N
166 if (mddev->chunk_sectors) {
167 sectors = rdev->sectors;
168 sector_div(sectors, mddev->chunk_sectors);
169 rdev->sectors = sectors * mddev->chunk_sectors;
170 }
1da177e4 171
8f6c2e4b
MP
172 disk_stack_limits(mddev->gendisk, rdev->bdev,
173 rdev->data_offset << 9);
1da177e4
LT
174 /* as we don't honour merge_bvec_fn, we must never risk
175 * violating it, so limit ->max_sector to one PAGE, as
176 * a one page request is never in violation.
177 */
178 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
ae03bf63 179 queue_max_sectors(mddev->queue) > (PAGE_SIZE>>9))
1da177e4
LT
180 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
181
dd8ac336 182 conf->array_sectors += rdev->sectors;
1da177e4 183 cnt++;
4db7cdc8 184
1da177e4 185 }
7c7546cc 186 if (cnt != raid_disks) {
1da177e4
LT
187 printk("linear: not enough drives present. Aborting!\n");
188 goto out;
189 }
190
191 /*
45d4582f 192 * Here we calculate the device offsets.
1da177e4 193 */
4db7cdc8
SS
194 conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
195
a778b73f 196 for (i = 1; i < raid_disks; i++)
4db7cdc8
SS
197 conf->disks[i].end_sector =
198 conf->disks[i-1].end_sector +
199 conf->disks[i].rdev->sectors;
15945fee 200
7c7546cc
N
201 return conf;
202
203out:
204 kfree(conf);
205 return NULL;
206}
207
208static int linear_run (mddev_t *mddev)
209{
210 linear_conf_t *conf;
211
0894cc30
AN
212 if (md_check_no_bitmap(mddev))
213 return -EINVAL;
e7e72bf6 214 mddev->queue->queue_lock = &mddev->queue->__queue_lock;
7c7546cc
N
215 conf = linear_conf(mddev, mddev->raid_disks);
216
217 if (!conf)
218 return 1;
219 mddev->private = conf;
1f403624 220 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
7c7546cc 221
1da177e4
LT
222 blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
223 mddev->queue->unplug_fn = linear_unplug;
26be34dc
N
224 mddev->queue->backing_dev_info.congested_fn = linear_congested;
225 mddev->queue->backing_dev_info.congested_data = mddev;
ac5e7113 226 md_integrity_register(mddev);
1da177e4 227 return 0;
7c7546cc 228}
1da177e4 229
495d3573
N
230static void free_conf(struct rcu_head *head)
231{
232 linear_conf_t *conf = container_of(head, linear_conf_t, rcu);
233 kfree(conf);
234}
235
7c7546cc
N
236static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
237{
238 /* Adding a drive to a linear array allows the array to grow.
239 * It is permitted if the new drive has a matching superblock
240 * already on it, with raid_disk equal to raid_disks.
241 * It is achieved by creating a new linear_private_data structure
242 * and swapping it in in-place of the current one.
243 * The current one is never freed until the array is stopped.
244 * This avoids races.
245 */
495d3573 246 linear_conf_t *newconf, *oldconf;
7c7546cc 247
a778b73f 248 if (rdev->saved_raid_disk != mddev->raid_disks)
7c7546cc
N
249 return -EINVAL;
250
a778b73f
N
251 rdev->raid_disk = rdev->saved_raid_disk;
252
7c7546cc
N
253 newconf = linear_conf(mddev,mddev->raid_disks+1);
254
255 if (!newconf)
256 return -ENOMEM;
257
495d3573 258 oldconf = rcu_dereference(mddev->private);
7c7546cc 259 mddev->raid_disks++;
af11c397 260 rcu_assign_pointer(mddev->private, newconf);
1f403624 261 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
f233ea5c 262 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 263 revalidate_disk(mddev->gendisk);
495d3573 264 call_rcu(&oldconf->rcu, free_conf);
7c7546cc 265 return 0;
1da177e4
LT
266}
267
268static int linear_stop (mddev_t *mddev)
269{
070ec55d 270 linear_conf_t *conf = mddev->private;
af11c397
S
271
272 /*
273 * We do not require rcu protection here since
274 * we hold reconfig_mutex for both linear_add and
275 * linear_stop, so they cannot race.
495d3573
N
276 * We should make sure any old 'conf's are properly
277 * freed though.
af11c397 278 */
495d3573 279 rcu_barrier();
1da177e4 280 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
495d3573 281 kfree(conf);
1da177e4
LT
282
283 return 0;
284}
285
165125e1 286static int linear_make_request (struct request_queue *q, struct bio *bio)
1da177e4 287{
a362357b 288 const int rw = bio_data_dir(bio);
1da177e4
LT
289 mddev_t *mddev = q->queuedata;
290 dev_info_t *tmp_dev;
4db7cdc8 291 sector_t start_sector;
c9959059 292 int cpu;
1da177e4 293
1f98a13f 294 if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) {
a2826aa9 295 md_barrier_request(mddev, bio);
e5dcdd80
N
296 return 0;
297 }
298
074a7aca
TH
299 cpu = part_stat_lock();
300 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
301 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
302 bio_sectors(bio));
303 part_stat_unlock();
1da177e4 304
af11c397 305 rcu_read_lock();
1da177e4 306 tmp_dev = which_dev(mddev, bio->bi_sector);
4db7cdc8
SS
307 start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
308
af11c397 309
4db7cdc8
SS
310 if (unlikely(bio->bi_sector >= (tmp_dev->end_sector)
311 || (bio->bi_sector < start_sector))) {
1da177e4
LT
312 char b[BDEVNAME_SIZE];
313
6283815d
AN
314 printk("linear_make_request: Sector %llu out of bounds on "
315 "dev %s: %llu sectors, offset %llu\n",
316 (unsigned long long)bio->bi_sector,
1da177e4 317 bdevname(tmp_dev->rdev->bdev, b),
4db7cdc8
SS
318 (unsigned long long)tmp_dev->rdev->sectors,
319 (unsigned long long)start_sector);
af11c397 320 rcu_read_unlock();
6712ecf8 321 bio_io_error(bio);
1da177e4
LT
322 return 0;
323 }
324 if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
4db7cdc8 325 tmp_dev->end_sector)) {
1da177e4
LT
326 /* This bio crosses a device boundary, so we have to
327 * split it.
328 */
329 struct bio_pair *bp;
af11c397
S
330 sector_t end_sector = tmp_dev->end_sector;
331
332 rcu_read_unlock();
6283815d 333
af11c397 334 bp = bio_split(bio, end_sector - bio->bi_sector);
6283815d 335
1da177e4
LT
336 if (linear_make_request(q, &bp->bio1))
337 generic_make_request(&bp->bio1);
338 if (linear_make_request(q, &bp->bio2))
339 generic_make_request(&bp->bio2);
340 bio_pair_release(bp);
341 return 0;
342 }
343
344 bio->bi_bdev = tmp_dev->rdev->bdev;
4db7cdc8 345 bio->bi_sector = bio->bi_sector - start_sector
6283815d 346 + tmp_dev->rdev->data_offset;
af11c397 347 rcu_read_unlock();
1da177e4
LT
348
349 return 1;
350}
351
352static void linear_status (struct seq_file *seq, mddev_t *mddev)
353{
354
9d8f0363 355 seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
1da177e4
LT
356}
357
358
2604b703 359static struct mdk_personality linear_personality =
1da177e4
LT
360{
361 .name = "linear",
2604b703 362 .level = LEVEL_LINEAR,
1da177e4
LT
363 .owner = THIS_MODULE,
364 .make_request = linear_make_request,
365 .run = linear_run,
366 .stop = linear_stop,
367 .status = linear_status,
7c7546cc 368 .hot_add_disk = linear_add,
80c3a6ce 369 .size = linear_size,
1da177e4
LT
370};
371
372static int __init linear_init (void)
373{
2604b703 374 return register_md_personality (&linear_personality);
1da177e4
LT
375}
376
377static void linear_exit (void)
378{
2604b703 379 unregister_md_personality (&linear_personality);
1da177e4
LT
380}
381
382
383module_init(linear_init);
384module_exit(linear_exit);
385MODULE_LICENSE("GPL");
0efb9e61 386MODULE_DESCRIPTION("Linear device concatenation personality for MD");
d9d166c2
N
387MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
388MODULE_ALIAS("md-linear");
2604b703 389MODULE_ALIAS("md-level--1");