blob: 6cd2c313e800521d033493fb65d98544566d1211 [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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090022#include <linux/slab.h>
NeilBrown43b2e5d2009-03-31 14:33:13 +110023#include "md.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110024#include "linear.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
Linus Torvalds1da177e2005-04-16 15:20:36 -070026/*
27 * find which device holds a particular offset
28 */
29static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector)
30{
Sandeep K Sinhaaece3d12009-06-16 16:57:08 +100031 int lo, mid, hi;
SandeepKsinhaaf11c392009-06-18 08:49:35 +100032 linear_conf_t *conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
Sandeep K Sinhaaece3d12009-06-16 16:57:08 +100034 lo = 0;
35 hi = mddev->raid_disks - 1;
SandeepKsinhaaf11c392009-06-18 08:49:35 +100036 conf = rcu_dereference(mddev->private);
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Sandeep K Sinhaaece3d12009-06-16 16:57:08 +100038 /*
39 * Binary Search
40 */
41
42 while (hi > lo) {
43
44 mid = (hi + lo) / 2;
45 if (sector < conf->disks[mid].end_sector)
46 hi = mid;
47 else
48 lo = mid + 1;
49 }
50
51 return conf->disks + lo;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052}
53
54/**
NeilBrown15945fe2005-09-09 16:23:47 -070055 * linear_mergeable_bvec -- tell bio layer if two requests can be merged
Linus Torvalds1da177e2005-04-16 15:20:36 -070056 * @q: request queue
Alasdair G Kergoncc371e62008-07-03 09:53:43 +020057 * @bvm: properties of new bio
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 * @biovec: the request that could be merged to it.
59 *
60 * Return amount of bytes we can take at this offset
61 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +020062static int linear_mergeable_bvec(struct request_queue *q,
63 struct bvec_merge_data *bvm,
64 struct bio_vec *biovec)
Linus Torvalds1da177e2005-04-16 15:20:36 -070065{
66 mddev_t *mddev = q->queuedata;
67 dev_info_t *dev0;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +020068 unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9;
69 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
SandeepKsinhaaf11c392009-06-18 08:49:35 +100071 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 dev0 = which_dev(mddev, sector);
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +100073 maxsectors = dev0->end_sector - sector;
SandeepKsinhaaf11c392009-06-18 08:49:35 +100074 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
76 if (maxsectors < bio_sectors)
77 maxsectors = 0;
78 else
79 maxsectors -= bio_sectors;
80
81 if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
82 return biovec->bv_len;
83 /* The bytes available at this offset could be really big,
84 * so we cap at 2^31 to avoid overflow */
85 if (maxsectors > (1 << (31-9)))
86 return 1<<31;
87 return maxsectors << 9;
88}
89
NeilBrown26be34d2006-10-03 01:15:53 -070090static int linear_congested(void *data, int bits)
91{
92 mddev_t *mddev = data;
SandeepKsinhaaf11c392009-06-18 08:49:35 +100093 linear_conf_t *conf;
NeilBrown26be34d2006-10-03 01:15:53 -070094 int i, ret = 0;
95
NeilBrown3fa841d2009-09-23 18:10:29 +100096 if (mddev_congested(mddev, bits))
97 return 1;
98
SandeepKsinhaaf11c392009-06-18 08:49:35 +100099 rcu_read_lock();
100 conf = rcu_dereference(mddev->private);
101
NeilBrown26be34d2006-10-03 01:15:53 -0700102 for (i = 0; i < mddev->raid_disks && !ret ; i++) {
Jens Axboe165125e2007-07-24 09:28:11 +0200103 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
NeilBrown26be34d2006-10-03 01:15:53 -0700104 ret |= bdi_congested(&q->backing_dev_info, bits);
105 }
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000106
107 rcu_read_unlock();
NeilBrown26be34d2006-10-03 01:15:53 -0700108 return ret;
109}
110
Dan Williams80c3a6c2009-03-17 18:10:40 -0700111static sector_t linear_size(mddev_t *mddev, sector_t sectors, int raid_disks)
112{
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000113 linear_conf_t *conf;
114 sector_t array_sectors;
Dan Williams80c3a6c2009-03-17 18:10:40 -0700115
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000116 rcu_read_lock();
117 conf = rcu_dereference(mddev->private);
Dan Williams80c3a6c2009-03-17 18:10:40 -0700118 WARN_ONCE(sectors || raid_disks,
119 "%s does not support generic reshape\n", __func__);
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000120 array_sectors = conf->array_sectors;
121 rcu_read_unlock();
Dan Williams80c3a6c2009-03-17 18:10:40 -0700122
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000123 return array_sectors;
Dan Williams80c3a6c2009-03-17 18:10:40 -0700124}
125
NeilBrown7c7546c2006-06-26 00:27:41 -0700126static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127{
128 linear_conf_t *conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 mdk_rdev_t *rdev;
Sandeep K Sinha45d45822009-06-16 16:55:26 +1000130 int i, cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
NeilBrown7c7546c2006-06-26 00:27:41 -0700132 conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 GFP_KERNEL);
134 if (!conf)
NeilBrown7c7546c2006-06-26 00:27:41 -0700135 return NULL;
136
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 cnt = 0;
Andre Nolld6e22152008-07-21 17:05:25 +1000138 conf->array_sectors = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139
Cheng Renquan159ec1f2009-01-09 08:31:08 +1100140 list_for_each_entry(rdev, &mddev->disks, same_set) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 int j = rdev->raid_disk;
142 dev_info_t *disk = conf->disks + j;
NeilBrown13f26822009-06-18 08:48:55 +1000143 sector_t sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Nikanth Karthikesan13864512008-06-28 08:31:19 +1000145 if (j < 0 || j >= raid_disks || disk->rdev) {
NeilBrown2dc40f82010-05-03 15:12:04 +1000146 printk(KERN_ERR "md/linear:%s: disk numbering problem. Aborting!\n",
147 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 goto out;
149 }
150
151 disk->rdev = rdev;
NeilBrown13f26822009-06-18 08:48:55 +1000152 if (mddev->chunk_sectors) {
153 sectors = rdev->sectors;
154 sector_div(sectors, mddev->chunk_sectors);
155 rdev->sectors = sectors * mddev->chunk_sectors;
156 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +1000158 disk_stack_limits(mddev->gendisk, rdev->bdev,
159 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 /* as we don't honour merge_bvec_fn, we must never risk
NeilBrown627a2d32010-03-08 16:44:38 +1100161 * violating it, so limit max_segments to 1 lying within
162 * a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 */
NeilBrown627a2d32010-03-08 16:44:38 +1100164 if (rdev->bdev->bd_disk->queue->merge_bvec_fn) {
165 blk_queue_max_segments(mddev->queue, 1);
166 blk_queue_segment_boundary(mddev->queue,
167 PAGE_CACHE_SIZE - 1);
168 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169
Andre Nolldd8ac332009-03-31 14:33:13 +1100170 conf->array_sectors += rdev->sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 cnt++;
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +1000172
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 }
NeilBrown7c7546c2006-06-26 00:27:41 -0700174 if (cnt != raid_disks) {
NeilBrown2dc40f82010-05-03 15:12:04 +1000175 printk(KERN_ERR "md/linear:%s: not enough drives present. Aborting!\n",
176 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 goto out;
178 }
179
180 /*
Sandeep K Sinha45d45822009-06-16 16:55:26 +1000181 * Here we calculate the device offsets.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 */
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +1000183 conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
184
NeilBrowna778b732007-05-23 13:58:10 -0700185 for (i = 1; i < raid_disks; i++)
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +1000186 conf->disks[i].end_sector =
187 conf->disks[i-1].end_sector +
188 conf->disks[i].rdev->sectors;
NeilBrown15945fe2005-09-09 16:23:47 -0700189
NeilBrown7c7546c2006-06-26 00:27:41 -0700190 return conf;
191
192out:
193 kfree(conf);
194 return NULL;
195}
196
197static int linear_run (mddev_t *mddev)
198{
199 linear_conf_t *conf;
200
Andre Noll0894cc32009-06-18 08:49:23 +1000201 if (md_check_no_bitmap(mddev))
202 return -EINVAL;
NeilBrown7c7546c2006-06-26 00:27:41 -0700203 conf = linear_conf(mddev, mddev->raid_disks);
204
205 if (!conf)
206 return 1;
207 mddev->private = conf;
Dan Williams1f403622009-03-31 14:59:03 +1100208 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
NeilBrown7c7546c2006-06-26 00:27:41 -0700209
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
NeilBrown26be34d2006-10-03 01:15:53 -0700211 mddev->queue->backing_dev_info.congested_fn = linear_congested;
212 mddev->queue->backing_dev_info.congested_data = mddev;
Martin K. Petersena91a2782011-03-17 11:11:05 +0100213 return md_integrity_register(mddev);
NeilBrown7c7546c2006-06-26 00:27:41 -0700214}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
NeilBrown7c7546c2006-06-26 00:27:41 -0700216static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
217{
218 /* Adding a drive to a linear array allows the array to grow.
219 * It is permitted if the new drive has a matching superblock
220 * already on it, with raid_disk equal to raid_disks.
221 * It is achieved by creating a new linear_private_data structure
222 * and swapping it in in-place of the current one.
223 * The current one is never freed until the array is stopped.
224 * This avoids races.
225 */
NeilBrown495d3572009-06-18 08:49:42 +1000226 linear_conf_t *newconf, *oldconf;
NeilBrown7c7546c2006-06-26 00:27:41 -0700227
NeilBrowna778b732007-05-23 13:58:10 -0700228 if (rdev->saved_raid_disk != mddev->raid_disks)
NeilBrown7c7546c2006-06-26 00:27:41 -0700229 return -EINVAL;
230
NeilBrowna778b732007-05-23 13:58:10 -0700231 rdev->raid_disk = rdev->saved_raid_disk;
232
NeilBrown7c7546c2006-06-26 00:27:41 -0700233 newconf = linear_conf(mddev,mddev->raid_disks+1);
234
235 if (!newconf)
236 return -ENOMEM;
237
NeilBrown495d3572009-06-18 08:49:42 +1000238 oldconf = rcu_dereference(mddev->private);
NeilBrown7c7546c2006-06-26 00:27:41 -0700239 mddev->raid_disks++;
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000240 rcu_assign_pointer(mddev->private, newconf);
Dan Williams1f403622009-03-31 14:59:03 +1100241 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
Andre Nollf233ea52008-07-21 17:05:22 +1000242 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +1000243 revalidate_disk(mddev->gendisk);
Lai Jiangshanb119cba2011-03-18 11:59:14 +0800244 kfree_rcu(oldconf, rcu);
NeilBrown7c7546c2006-06-26 00:27:41 -0700245 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246}
247
248static int linear_stop (mddev_t *mddev)
249{
NeilBrown070ec552009-06-16 16:54:21 +1000250 linear_conf_t *conf = mddev->private;
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000251
252 /*
253 * We do not require rcu protection here since
254 * we hold reconfig_mutex for both linear_add and
255 * linear_stop, so they cannot race.
NeilBrown495d3572009-06-18 08:49:42 +1000256 * We should make sure any old 'conf's are properly
257 * freed though.
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000258 */
NeilBrown495d3572009-06-18 08:49:42 +1000259 rcu_barrier();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
NeilBrown495d3572009-06-18 08:49:42 +1000261 kfree(conf);
NeilBrownef2f80f2010-05-17 11:27:00 +1000262 mddev->private = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263
264 return 0;
265}
266
NeilBrown21a52c62010-04-01 15:02:13 +1100267static int linear_make_request (mddev_t *mddev, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 dev_info_t *tmp_dev;
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +1000270 sector_t start_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271
Tejun Heoe9c74692010-09-03 11:56:18 +0200272 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
273 md_flush_request(mddev, bio);
NeilBrowne5dcdd82005-09-09 16:23:41 -0700274 return 0;
275 }
276
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000277 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 tmp_dev = which_dev(mddev, bio->bi_sector);
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +1000279 start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
280
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000281
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +1000282 if (unlikely(bio->bi_sector >= (tmp_dev->end_sector)
283 || (bio->bi_sector < start_sector))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 char b[BDEVNAME_SIZE];
285
NeilBrown2dc40f82010-05-03 15:12:04 +1000286 printk(KERN_ERR
287 "md/linear:%s: make_request: Sector %llu out of bounds on "
288 "dev %s: %llu sectors, offset %llu\n",
289 mdname(mddev),
290 (unsigned long long)bio->bi_sector,
291 bdevname(tmp_dev->rdev->bdev, b),
292 (unsigned long long)tmp_dev->rdev->sectors,
293 (unsigned long long)start_sector);
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000294 rcu_read_unlock();
NeilBrown6712ecf2007-09-27 12:47:43 +0200295 bio_io_error(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 return 0;
297 }
298 if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +1000299 tmp_dev->end_sector)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 /* This bio crosses a device boundary, so we have to
301 * split it.
302 */
303 struct bio_pair *bp;
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000304 sector_t end_sector = tmp_dev->end_sector;
Andre Noll62838152008-10-13 11:55:12 +1100305
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000306 rcu_read_unlock();
307
308 bp = bio_split(bio, end_sector - bio->bi_sector);
Andre Noll62838152008-10-13 11:55:12 +1100309
NeilBrown21a52c62010-04-01 15:02:13 +1100310 if (linear_make_request(mddev, &bp->bio1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 generic_make_request(&bp->bio1);
NeilBrown21a52c62010-04-01 15:02:13 +1100312 if (linear_make_request(mddev, &bp->bio2))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 generic_make_request(&bp->bio2);
314 bio_pair_release(bp);
315 return 0;
316 }
317
318 bio->bi_bdev = tmp_dev->rdev->bdev;
Sandeep K Sinha4db7cdc2009-06-16 16:56:13 +1000319 bio->bi_sector = bio->bi_sector - start_sector
Andre Noll62838152008-10-13 11:55:12 +1100320 + tmp_dev->rdev->data_offset;
SandeepKsinhaaf11c392009-06-18 08:49:35 +1000321 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
323 return 1;
324}
325
326static void linear_status (struct seq_file *seq, mddev_t *mddev)
327{
328
Andre Noll9d8f0362009-06-18 08:45:01 +1000329 seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330}
331
332
NeilBrown2604b702006-01-06 00:20:36 -0800333static struct mdk_personality linear_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334{
335 .name = "linear",
NeilBrown2604b702006-01-06 00:20:36 -0800336 .level = LEVEL_LINEAR,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 .owner = THIS_MODULE,
338 .make_request = linear_make_request,
339 .run = linear_run,
340 .stop = linear_stop,
341 .status = linear_status,
NeilBrown7c7546c2006-06-26 00:27:41 -0700342 .hot_add_disk = linear_add,
Dan Williams80c3a6c2009-03-17 18:10:40 -0700343 .size = linear_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344};
345
346static int __init linear_init (void)
347{
NeilBrown2604b702006-01-06 00:20:36 -0800348 return register_md_personality (&linear_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349}
350
351static void linear_exit (void)
352{
NeilBrown2604b702006-01-06 00:20:36 -0800353 unregister_md_personality (&linear_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354}
355
356
357module_init(linear_init);
358module_exit(linear_exit);
359MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +1100360MODULE_DESCRIPTION("Linear device concatenation personality for MD");
NeilBrownd9d166c2006-01-06 00:20:51 -0800361MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
362MODULE_ALIAS("md-linear");
NeilBrown2604b702006-01-06 00:20:36 -0800363MODULE_ALIAS("md-level--1");