md linear: Protecting mddev with rcu locks to avoid races
[GitHub/mt8127/android_kernel_alcatel_ttab.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
af11c397
S
111 rcu_read_lock();
112 conf = rcu_dereference(mddev->private);
113
26be34dc 114 for (i = 0; i < mddev->raid_disks && !ret ; i++) {
165125e1 115 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
26be34dc
N
116 ret |= bdi_congested(&q->backing_dev_info, bits);
117 }
af11c397
S
118
119 rcu_read_unlock();
26be34dc
N
120 return ret;
121}
122
80c3a6ce
DW
123static sector_t linear_size(mddev_t *mddev, sector_t sectors, int raid_disks)
124{
af11c397
S
125 linear_conf_t *conf;
126 sector_t array_sectors;
80c3a6ce 127
af11c397
S
128 rcu_read_lock();
129 conf = rcu_dereference(mddev->private);
80c3a6ce
DW
130 WARN_ONCE(sectors || raid_disks,
131 "%s does not support generic reshape\n", __func__);
af11c397
S
132 array_sectors = conf->array_sectors;
133 rcu_read_unlock();
80c3a6ce 134
af11c397 135 return array_sectors;
80c3a6ce
DW
136}
137
7c7546cc 138static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
1da177e4
LT
139{
140 linear_conf_t *conf;
1da177e4 141 mdk_rdev_t *rdev;
45d4582f 142 int i, cnt;
1da177e4 143
7c7546cc 144 conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
1da177e4
LT
145 GFP_KERNEL);
146 if (!conf)
7c7546cc
N
147 return NULL;
148
1da177e4 149 cnt = 0;
d6e22150 150 conf->array_sectors = 0;
1da177e4 151
159ec1fc 152 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT
153 int j = rdev->raid_disk;
154 dev_info_t *disk = conf->disks + j;
13f2682b 155 sector_t sectors;
1da177e4 156
13864515 157 if (j < 0 || j >= raid_disks || disk->rdev) {
1da177e4
LT
158 printk("linear: disk numbering problem. Aborting!\n");
159 goto out;
160 }
161
162 disk->rdev = rdev;
13f2682b
N
163 if (mddev->chunk_sectors) {
164 sectors = rdev->sectors;
165 sector_div(sectors, mddev->chunk_sectors);
166 rdev->sectors = sectors * mddev->chunk_sectors;
167 }
1da177e4
LT
168
169 blk_queue_stack_limits(mddev->queue,
170 rdev->bdev->bd_disk->queue);
171 /* as we don't honour merge_bvec_fn, we must never risk
172 * violating it, so limit ->max_sector to one PAGE, as
173 * a one page request is never in violation.
174 */
175 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
ae03bf63 176 queue_max_sectors(mddev->queue) > (PAGE_SIZE>>9))
1da177e4
LT
177 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
178
dd8ac336 179 conf->array_sectors += rdev->sectors;
1da177e4 180 cnt++;
4db7cdc8 181
1da177e4 182 }
7c7546cc 183 if (cnt != raid_disks) {
1da177e4
LT
184 printk("linear: not enough drives present. Aborting!\n");
185 goto out;
186 }
187
188 /*
45d4582f 189 * Here we calculate the device offsets.
1da177e4 190 */
4db7cdc8
SS
191 conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
192
a778b73f 193 for (i = 1; i < raid_disks; i++)
4db7cdc8
SS
194 conf->disks[i].end_sector =
195 conf->disks[i-1].end_sector +
196 conf->disks[i].rdev->sectors;
15945fee 197
7c7546cc
N
198 return conf;
199
200out:
201 kfree(conf);
202 return NULL;
203}
204
205static int linear_run (mddev_t *mddev)
206{
207 linear_conf_t *conf;
208
0894cc30
AN
209 if (md_check_no_bitmap(mddev))
210 return -EINVAL;
e7e72bf6 211 mddev->queue->queue_lock = &mddev->queue->__queue_lock;
7c7546cc
N
212 conf = linear_conf(mddev, mddev->raid_disks);
213
214 if (!conf)
215 return 1;
216 mddev->private = conf;
1f403624 217 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
7c7546cc 218
1da177e4
LT
219 blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
220 mddev->queue->unplug_fn = linear_unplug;
26be34dc
N
221 mddev->queue->backing_dev_info.congested_fn = linear_congested;
222 mddev->queue->backing_dev_info.congested_data = mddev;
1da177e4 223 return 0;
7c7546cc 224}
1da177e4 225
7c7546cc
N
226static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
227{
228 /* Adding a drive to a linear array allows the array to grow.
229 * It is permitted if the new drive has a matching superblock
230 * already on it, with raid_disk equal to raid_disks.
231 * It is achieved by creating a new linear_private_data structure
232 * and swapping it in in-place of the current one.
233 * The current one is never freed until the array is stopped.
234 * This avoids races.
235 */
236 linear_conf_t *newconf;
237
a778b73f 238 if (rdev->saved_raid_disk != mddev->raid_disks)
7c7546cc
N
239 return -EINVAL;
240
a778b73f
N
241 rdev->raid_disk = rdev->saved_raid_disk;
242
7c7546cc
N
243 newconf = linear_conf(mddev,mddev->raid_disks+1);
244
245 if (!newconf)
246 return -ENOMEM;
247
070ec55d 248 newconf->prev = mddev->private;
7c7546cc 249 mddev->raid_disks++;
af11c397 250 rcu_assign_pointer(mddev->private, newconf);
1f403624 251 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
f233ea5c 252 set_capacity(mddev->gendisk, mddev->array_sectors);
7c7546cc 253 return 0;
1da177e4
LT
254}
255
256static int linear_stop (mddev_t *mddev)
257{
070ec55d 258 linear_conf_t *conf = mddev->private;
af11c397
S
259
260 /*
261 * We do not require rcu protection here since
262 * we hold reconfig_mutex for both linear_add and
263 * linear_stop, so they cannot race.
264 */
265
1da177e4 266 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
7c7546cc
N
267 do {
268 linear_conf_t *t = conf->prev;
7c7546cc
N
269 kfree(conf);
270 conf = t;
271 } while (conf);
1da177e4
LT
272
273 return 0;
274}
275
165125e1 276static int linear_make_request (struct request_queue *q, struct bio *bio)
1da177e4 277{
a362357b 278 const int rw = bio_data_dir(bio);
1da177e4
LT
279 mddev_t *mddev = q->queuedata;
280 dev_info_t *tmp_dev;
4db7cdc8 281 sector_t start_sector;
c9959059 282 int cpu;
1da177e4 283
e5dcdd80 284 if (unlikely(bio_barrier(bio))) {
6712ecf8 285 bio_endio(bio, -EOPNOTSUPP);
e5dcdd80
N
286 return 0;
287 }
288
074a7aca
TH
289 cpu = part_stat_lock();
290 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
291 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
292 bio_sectors(bio));
293 part_stat_unlock();
1da177e4 294
af11c397 295 rcu_read_lock();
1da177e4 296 tmp_dev = which_dev(mddev, bio->bi_sector);
4db7cdc8
SS
297 start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
298
af11c397 299
4db7cdc8
SS
300 if (unlikely(bio->bi_sector >= (tmp_dev->end_sector)
301 || (bio->bi_sector < start_sector))) {
1da177e4
LT
302 char b[BDEVNAME_SIZE];
303
6283815d
AN
304 printk("linear_make_request: Sector %llu out of bounds on "
305 "dev %s: %llu sectors, offset %llu\n",
306 (unsigned long long)bio->bi_sector,
1da177e4 307 bdevname(tmp_dev->rdev->bdev, b),
4db7cdc8
SS
308 (unsigned long long)tmp_dev->rdev->sectors,
309 (unsigned long long)start_sector);
af11c397 310 rcu_read_unlock();
6712ecf8 311 bio_io_error(bio);
1da177e4
LT
312 return 0;
313 }
314 if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
4db7cdc8 315 tmp_dev->end_sector)) {
1da177e4
LT
316 /* This bio crosses a device boundary, so we have to
317 * split it.
318 */
319 struct bio_pair *bp;
af11c397
S
320 sector_t end_sector = tmp_dev->end_sector;
321
322 rcu_read_unlock();
6283815d 323
af11c397 324 bp = bio_split(bio, end_sector - bio->bi_sector);
6283815d 325
1da177e4
LT
326 if (linear_make_request(q, &bp->bio1))
327 generic_make_request(&bp->bio1);
328 if (linear_make_request(q, &bp->bio2))
329 generic_make_request(&bp->bio2);
330 bio_pair_release(bp);
331 return 0;
332 }
333
334 bio->bi_bdev = tmp_dev->rdev->bdev;
4db7cdc8 335 bio->bi_sector = bio->bi_sector - start_sector
6283815d 336 + tmp_dev->rdev->data_offset;
af11c397 337 rcu_read_unlock();
1da177e4
LT
338
339 return 1;
340}
341
342static void linear_status (struct seq_file *seq, mddev_t *mddev)
343{
344
9d8f0363 345 seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
1da177e4
LT
346}
347
348
2604b703 349static struct mdk_personality linear_personality =
1da177e4
LT
350{
351 .name = "linear",
2604b703 352 .level = LEVEL_LINEAR,
1da177e4
LT
353 .owner = THIS_MODULE,
354 .make_request = linear_make_request,
355 .run = linear_run,
356 .stop = linear_stop,
357 .status = linear_status,
7c7546cc 358 .hot_add_disk = linear_add,
80c3a6ce 359 .size = linear_size,
1da177e4
LT
360};
361
362static int __init linear_init (void)
363{
2604b703 364 return register_md_personality (&linear_personality);
1da177e4
LT
365}
366
367static void linear_exit (void)
368{
2604b703 369 unregister_md_personality (&linear_personality);
1da177e4
LT
370}
371
372
373module_init(linear_init);
374module_exit(linear_exit);
375MODULE_LICENSE("GPL");
d9d166c2
N
376MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
377MODULE_ALIAS("md-linear");
2604b703 378MODULE_ALIAS("md-level--1");