blob: 15c8b7b25a9b9a053a7296e0ab7abc91efa88e05 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
NeilBrownbff61972009-03-31 14:33:13 +110019#include <linux/blkdev.h>
20#include <linux/raid/md_u.h>
NeilBrownbff61972009-03-31 14:33:13 +110021#include <linux/seq_file.h>
NeilBrown43b2e5d2009-03-31 14:33:13 +110022#include "md.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110023#include "linear.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024
Linus Torvalds1da177e2005-04-16 15:20:36 -070025/*
26 * find which device holds a particular offset
27 */
28static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector)
29{
Sandeep K Sinhaaece3d12009-06-16 16:57:08 +100030 int lo, mid, hi;
SandeepKsinhaaf11c392009-06-18 08:49:35 +100031 linear_conf_t *conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
Sandeep K Sinhaaece3d12009-06-16 16:57:08 +100033 lo = 0;
34 hi = mddev->raid_disks - 1;
SandeepKsinhaaf11c392009-06-18 08:49:35 +100035 conf = rcu_dereference(mddev->private);
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Sandeep K Sinhaaece3d12009-06-16 16:57:08 +100037 /*
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051}
52
53/**
NeilBrown15945fe2005-09-09 16:23:47 -070054 * linear_mergeable_bvec -- tell bio layer if two requests can be merged
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 * @q: request queue
Alasdair G Kergoncc371e62008-07-03 09:53:43 +020056 * @bvm: properties of new bio
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 * @biovec: the request that could be merged to it.
58 *
59 * Return amount of bytes we can take at this offset
60 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +020061static int linear_mergeable_bvec(struct request_queue *q,
62 struct bvec_merge_data *bvm,
63 struct bio_vec *biovec)
Linus Torvalds1da177e2005-04-16 15:20:36 -070064{
65 mddev_t *mddev = q->queuedata;
66 dev_info_t *dev0;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +020067 unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9;
68 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -070069
SandeepKsinhaaf11c392009-06-18 08:49:35 +100070 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 dev0 = which_dev(mddev, sector);
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +100072 maxsectors = dev0->end_sector - sector;
SandeepKsinhaaf11c392009-06-18 08:49:35 +100073 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
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
Jens Axboe165125e2007-07-24 09:28:11 +020089static void linear_unplug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070090{
91 mddev_t *mddev = q->queuedata;
SandeepKsinhaaf11c392009-06-18 08:49:35 +100092 linear_conf_t *conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 int i;
94
SandeepKsinhaaf11c392009-06-18 08:49:35 +100095 rcu_read_lock();
96 conf = rcu_dereference(mddev->private);
97
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 for (i=0; i < mddev->raid_disks; i++) {
Jens Axboe165125e2007-07-24 09:28:11 +020099 struct request_queue *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500100 blk_unplug(r_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 }
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000102 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103}
104
NeilBrown26be34d2006-10-03 01:15:53 -0700105static int linear_congested(void *data, int bits)
106{
107 mddev_t *mddev = data;
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000108 linear_conf_t *conf;
NeilBrown26be34d2006-10-03 01:15:53 -0700109 int i, ret = 0;
110
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000111 rcu_read_lock();
112 conf = rcu_dereference(mddev->private);
113
NeilBrown26be34d2006-10-03 01:15:53 -0700114 for (i = 0; i < mddev->raid_disks && !ret ; i++) {
Jens Axboe165125e2007-07-24 09:28:11 +0200115 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
NeilBrown26be34d2006-10-03 01:15:53 -0700116 ret |= bdi_congested(&q->backing_dev_info, bits);
117 }
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000118
119 rcu_read_unlock();
NeilBrown26be34d2006-10-03 01:15:53 -0700120 return ret;
121}
122
Dan Williams80c3a6c2009-03-17 18:10:40 -0700123static sector_t linear_size(mddev_t *mddev, sector_t sectors, int raid_disks)
124{
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000125 linear_conf_t *conf;
126 sector_t array_sectors;
Dan Williams80c3a6c2009-03-17 18:10:40 -0700127
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000128 rcu_read_lock();
129 conf = rcu_dereference(mddev->private);
Dan Williams80c3a6c2009-03-17 18:10:40 -0700130 WARN_ONCE(sectors || raid_disks,
131 "%s does not support generic reshape\n", __func__);
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000132 array_sectors = conf->array_sectors;
133 rcu_read_unlock();
Dan Williams80c3a6c2009-03-17 18:10:40 -0700134
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000135 return array_sectors;
Dan Williams80c3a6c2009-03-17 18:10:40 -0700136}
137
NeilBrown7c7546c2006-06-26 00:27:41 -0700138static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139{
140 linear_conf_t *conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 mdk_rdev_t *rdev;
Sandeep K Sinha45d45822009-06-16 16:55:26 +1000142 int i, cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
NeilBrown7c7546c2006-06-26 00:27:41 -0700144 conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 GFP_KERNEL);
146 if (!conf)
NeilBrown7c7546c2006-06-26 00:27:41 -0700147 return NULL;
148
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 cnt = 0;
Andre Nolld6e22152008-07-21 17:05:25 +1000150 conf->array_sectors = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
Cheng Renquan159ec1f2009-01-09 08:31:08 +1100152 list_for_each_entry(rdev, &mddev->disks, same_set) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 int j = rdev->raid_disk;
154 dev_info_t *disk = conf->disks + j;
NeilBrown13f26822009-06-18 08:48:55 +1000155 sector_t sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
Nikanth Karthikesan13864512008-06-28 08:31:19 +1000157 if (j < 0 || j >= raid_disks || disk->rdev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 printk("linear: disk numbering problem. Aborting!\n");
159 goto out;
160 }
161
162 disk->rdev = rdev;
NeilBrown13f26822009-06-18 08:48:55 +1000163 if (mddev->chunk_sectors) {
164 sectors = rdev->sectors;
165 sector_div(sectors, mddev->chunk_sectors);
166 rdev->sectors = sectors * mddev->chunk_sectors;
167 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168
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 &&
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400176 queue_max_sectors(mddev->queue) > (PAGE_SIZE>>9))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
178
Andre Nolldd8ac332009-03-31 14:33:13 +1100179 conf->array_sectors += rdev->sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 cnt++;
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +1000181
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 }
NeilBrown7c7546c2006-06-26 00:27:41 -0700183 if (cnt != raid_disks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 printk("linear: not enough drives present. Aborting!\n");
185 goto out;
186 }
187
188 /*
Sandeep K Sinha45d45822009-06-16 16:55:26 +1000189 * Here we calculate the device offsets.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 */
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +1000191 conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
192
NeilBrowna778b732007-05-23 13:58:10 -0700193 for (i = 1; i < raid_disks; i++)
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +1000194 conf->disks[i].end_sector =
195 conf->disks[i-1].end_sector +
196 conf->disks[i].rdev->sectors;
NeilBrown15945fe2005-09-09 16:23:47 -0700197
NeilBrown7c7546c2006-06-26 00:27:41 -0700198 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
Andre Noll0894cc32009-06-18 08:49:23 +1000209 if (md_check_no_bitmap(mddev))
210 return -EINVAL;
Neil Browne7e72bf2008-05-14 16:05:54 -0700211 mddev->queue->queue_lock = &mddev->queue->__queue_lock;
NeilBrown7c7546c2006-06-26 00:27:41 -0700212 conf = linear_conf(mddev, mddev->raid_disks);
213
214 if (!conf)
215 return 1;
216 mddev->private = conf;
Dan Williams1f403622009-03-31 14:59:03 +1100217 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
NeilBrown7c7546c2006-06-26 00:27:41 -0700218
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
220 mddev->queue->unplug_fn = linear_unplug;
NeilBrown26be34d2006-10-03 01:15:53 -0700221 mddev->queue->backing_dev_info.congested_fn = linear_congested;
222 mddev->queue->backing_dev_info.congested_data = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 return 0;
NeilBrown7c7546c2006-06-26 00:27:41 -0700224}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
NeilBrown495d3572009-06-18 08:49:42 +1000226static void free_conf(struct rcu_head *head)
227{
228 linear_conf_t *conf = container_of(head, linear_conf_t, rcu);
229 kfree(conf);
230}
231
NeilBrown7c7546c2006-06-26 00:27:41 -0700232static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
233{
234 /* Adding a drive to a linear array allows the array to grow.
235 * It is permitted if the new drive has a matching superblock
236 * already on it, with raid_disk equal to raid_disks.
237 * It is achieved by creating a new linear_private_data structure
238 * and swapping it in in-place of the current one.
239 * The current one is never freed until the array is stopped.
240 * This avoids races.
241 */
NeilBrown495d3572009-06-18 08:49:42 +1000242 linear_conf_t *newconf, *oldconf;
NeilBrown7c7546c2006-06-26 00:27:41 -0700243
NeilBrowna778b732007-05-23 13:58:10 -0700244 if (rdev->saved_raid_disk != mddev->raid_disks)
NeilBrown7c7546c2006-06-26 00:27:41 -0700245 return -EINVAL;
246
NeilBrowna778b732007-05-23 13:58:10 -0700247 rdev->raid_disk = rdev->saved_raid_disk;
248
NeilBrown7c7546c2006-06-26 00:27:41 -0700249 newconf = linear_conf(mddev,mddev->raid_disks+1);
250
251 if (!newconf)
252 return -ENOMEM;
253
NeilBrown495d3572009-06-18 08:49:42 +1000254 oldconf = rcu_dereference(mddev->private);
NeilBrown7c7546c2006-06-26 00:27:41 -0700255 mddev->raid_disks++;
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000256 rcu_assign_pointer(mddev->private, newconf);
Dan Williams1f403622009-03-31 14:59:03 +1100257 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
Andre Nollf233ea52008-07-21 17:05:22 +1000258 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown495d3572009-06-18 08:49:42 +1000259 call_rcu(&oldconf->rcu, free_conf);
NeilBrown7c7546c2006-06-26 00:27:41 -0700260 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261}
262
263static int linear_stop (mddev_t *mddev)
264{
NeilBrown070ec552009-06-16 16:54:21 +1000265 linear_conf_t *conf = mddev->private;
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000266
267 /*
268 * We do not require rcu protection here since
269 * we hold reconfig_mutex for both linear_add and
270 * linear_stop, so they cannot race.
NeilBrown495d3572009-06-18 08:49:42 +1000271 * We should make sure any old 'conf's are properly
272 * freed though.
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000273 */
NeilBrown495d3572009-06-18 08:49:42 +1000274 rcu_barrier();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
NeilBrown495d3572009-06-18 08:49:42 +1000276 kfree(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277
278 return 0;
279}
280
Jens Axboe165125e2007-07-24 09:28:11 +0200281static int linear_make_request (struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282{
Jens Axboea3623572005-11-01 09:26:16 +0100283 const int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 mddev_t *mddev = q->queuedata;
285 dev_info_t *tmp_dev;
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +1000286 sector_t start_sector;
Tejun Heoc9959052008-08-25 19:47:21 +0900287 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288
NeilBrowne5dcdd82005-09-09 16:23:41 -0700289 if (unlikely(bio_barrier(bio))) {
NeilBrown6712ecf2007-09-27 12:47:43 +0200290 bio_endio(bio, -EOPNOTSUPP);
NeilBrowne5dcdd82005-09-09 16:23:41 -0700291 return 0;
292 }
293
Tejun Heo074a7ac2008-08-25 19:56:14 +0900294 cpu = part_stat_lock();
295 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
296 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
297 bio_sectors(bio));
298 part_stat_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000300 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 tmp_dev = which_dev(mddev, bio->bi_sector);
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +1000302 start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
303
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000304
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +1000305 if (unlikely(bio->bi_sector >= (tmp_dev->end_sector)
306 || (bio->bi_sector < start_sector))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 char b[BDEVNAME_SIZE];
308
Andre Noll62838152008-10-13 11:55:12 +1100309 printk("linear_make_request: Sector %llu out of bounds on "
310 "dev %s: %llu sectors, offset %llu\n",
311 (unsigned long long)bio->bi_sector,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 bdevname(tmp_dev->rdev->bdev, b),
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +1000313 (unsigned long long)tmp_dev->rdev->sectors,
314 (unsigned long long)start_sector);
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000315 rcu_read_unlock();
NeilBrown6712ecf2007-09-27 12:47:43 +0200316 bio_io_error(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 return 0;
318 }
319 if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +1000320 tmp_dev->end_sector)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 /* This bio crosses a device boundary, so we have to
322 * split it.
323 */
324 struct bio_pair *bp;
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000325 sector_t end_sector = tmp_dev->end_sector;
Andre Noll62838152008-10-13 11:55:12 +1100326
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000327 rcu_read_unlock();
328
329 bp = bio_split(bio, end_sector - bio->bi_sector);
Andre Noll62838152008-10-13 11:55:12 +1100330
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 if (linear_make_request(q, &bp->bio1))
332 generic_make_request(&bp->bio1);
333 if (linear_make_request(q, &bp->bio2))
334 generic_make_request(&bp->bio2);
335 bio_pair_release(bp);
336 return 0;
337 }
338
339 bio->bi_bdev = tmp_dev->rdev->bdev;
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +1000340 bio->bi_sector = bio->bi_sector - start_sector
Andre Noll62838152008-10-13 11:55:12 +1100341 + tmp_dev->rdev->data_offset;
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000342 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343
344 return 1;
345}
346
347static void linear_status (struct seq_file *seq, mddev_t *mddev)
348{
349
Andre Noll9d8f0362009-06-18 08:45:01 +1000350 seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351}
352
353
NeilBrown2604b702006-01-06 00:20:36 -0800354static struct mdk_personality linear_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355{
356 .name = "linear",
NeilBrown2604b702006-01-06 00:20:36 -0800357 .level = LEVEL_LINEAR,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 .owner = THIS_MODULE,
359 .make_request = linear_make_request,
360 .run = linear_run,
361 .stop = linear_stop,
362 .status = linear_status,
NeilBrown7c7546c2006-06-26 00:27:41 -0700363 .hot_add_disk = linear_add,
Dan Williams80c3a6c2009-03-17 18:10:40 -0700364 .size = linear_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365};
366
367static int __init linear_init (void)
368{
NeilBrown2604b702006-01-06 00:20:36 -0800369 return register_md_personality (&linear_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370}
371
372static void linear_exit (void)
373{
NeilBrown2604b702006-01-06 00:20:36 -0800374 unregister_md_personality (&linear_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375}
376
377
378module_init(linear_init);
379module_exit(linear_exit);
380MODULE_LICENSE("GPL");
NeilBrownd9d166c2006-01-06 00:20:51 -0800381MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
382MODULE_ALIAS("md-linear");
NeilBrown2604b702006-01-06 00:20:36 -0800383MODULE_ALIAS("md-level--1");