blob: 1e3aea9eecf18009f4ff0bc2e4c474914e1bcf74 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/raid/linear.h>
20
Linus Torvalds1da177e2005-04-16 15:20:36 -070021/*
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);
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
29 /*
30 * sector_div(a,b) returns the remainer and sets a to a/b
31 */
Andre Nollab5bd5c2008-10-13 11:55:12 +110032 sector >>= conf->sector_shift;
33 (void)sector_div(sector, conf->spacing);
34 hash = conf->hash_table[sector];
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Andre Noll62838152008-10-13 11:55:12 +110036 while (sector >= hash->num_sectors + hash->start_sector)
Linus Torvalds1da177e2005-04-16 15:20:36 -070037 hash++;
38 return hash;
39}
40
41/**
NeilBrown15945fe2005-09-09 16:23:47 -070042 * linear_mergeable_bvec -- tell bio layer if two requests can be merged
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 * @q: request queue
Alasdair G Kergoncc371e62008-07-03 09:53:43 +020044 * @bvm: properties of new bio
Linus Torvalds1da177e2005-04-16 15:20:36 -070045 * @biovec: the request that could be merged to it.
46 *
47 * Return amount of bytes we can take at this offset
48 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +020049static int linear_mergeable_bvec(struct request_queue *q,
50 struct bvec_merge_data *bvm,
51 struct bio_vec *biovec)
Linus Torvalds1da177e2005-04-16 15:20:36 -070052{
53 mddev_t *mddev = q->queuedata;
54 dev_info_t *dev0;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +020055 unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9;
56 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
58 dev0 = which_dev(mddev, sector);
Andre Noll62838152008-10-13 11:55:12 +110059 maxsectors = dev0->num_sectors - (sector - dev0->start_sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
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
Jens Axboe165125e2007-07-24 09:28:11 +020075static void linear_unplug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070076{
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++) {
Jens Axboe165125e2007-07-24 09:28:11 +020082 struct request_queue *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -050083 blk_unplug(r_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 }
85}
86
NeilBrown26be34d2006-10-03 01:15:53 -070087static 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++) {
Jens Axboe165125e2007-07-24 09:28:11 +020094 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
NeilBrown26be34d2006-10-03 01:15:53 -070095 ret |= bdi_congested(&q->backing_dev_info, bits);
96 }
97 return ret;
98}
99
NeilBrown7c7546c2006-06-26 00:27:41 -0700100static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101{
102 linear_conf_t *conf;
103 dev_info_t **table;
104 mdk_rdev_t *rdev;
105 int i, nb_zone, cnt;
Andre Noll23242fb2008-10-13 11:55:12 +1100106 sector_t min_sectors;
Andre Noll62838152008-10-13 11:55:12 +1100107 sector_t curr_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108
NeilBrown7c7546c2006-06-26 00:27:41 -0700109 conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 GFP_KERNEL);
111 if (!conf)
NeilBrown7c7546c2006-06-26 00:27:41 -0700112 return NULL;
113
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 cnt = 0;
Andre Nolld6e22152008-07-21 17:05:25 +1000115 conf->array_sectors = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
Cheng Renquan159ec1f2009-01-09 08:31:08 +1100117 list_for_each_entry(rdev, &mddev->disks, same_set) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 int j = rdev->raid_disk;
119 dev_info_t *disk = conf->disks + j;
120
Nikanth Karthikesan13864512008-06-28 08:31:19 +1000121 if (j < 0 || j >= raid_disks || disk->rdev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 printk("linear: disk numbering problem. Aborting!\n");
123 goto out;
124 }
125
126 disk->rdev = rdev;
127
128 blk_queue_stack_limits(mddev->queue,
129 rdev->bdev->bd_disk->queue);
130 /* as we don't honour merge_bvec_fn, we must never risk
131 * violating it, so limit ->max_sector to one PAGE, as
132 * a one page request is never in violation.
133 */
134 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
135 mddev->queue->max_sectors > (PAGE_SIZE>>9))
136 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
137
Andre Noll62838152008-10-13 11:55:12 +1100138 disk->num_sectors = rdev->size * 2;
Andre Nolld6e22152008-07-21 17:05:25 +1000139 conf->array_sectors += rdev->size * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 cnt++;
142 }
NeilBrown7c7546c2006-06-26 00:27:41 -0700143 if (cnt != raid_disks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 printk("linear: not enough drives present. Aborting!\n");
145 goto out;
146 }
147
Andre Noll23242fb2008-10-13 11:55:12 +1100148 min_sectors = conf->array_sectors;
149 sector_div(min_sectors, PAGE_SIZE/sizeof(struct dev_info *));
Andre Nollf1cd14a2008-11-06 19:41:24 +1100150 if (min_sectors == 0)
151 min_sectors = 1;
NeilBrown15945fe2005-09-09 16:23:47 -0700152
Andre Noll23242fb2008-10-13 11:55:12 +1100153 /* min_sectors is the minimum spacing that will fit the hash
NeilBrown15945fe2005-09-09 16:23:47 -0700154 * table in one PAGE. This may be much smaller than needed.
155 * We find the smallest non-terminal set of consecutive devices
Andre Noll23242fb2008-10-13 11:55:12 +1100156 * that is larger than min_sectors and use the size of that as
NeilBrown15945fe2005-09-09 16:23:47 -0700157 * the actual spacing
158 */
Andre Nollab5bd5c2008-10-13 11:55:12 +1100159 conf->spacing = conf->array_sectors;
NeilBrown15945fe2005-09-09 16:23:47 -0700160 for (i=0; i < cnt-1 ; i++) {
Andre Noll23242fb2008-10-13 11:55:12 +1100161 sector_t tmp = 0;
NeilBrown15945fe2005-09-09 16:23:47 -0700162 int j;
Andre Noll23242fb2008-10-13 11:55:12 +1100163 for (j = i; j < cnt - 1 && tmp < min_sectors; j++)
164 tmp += conf->disks[j].num_sectors;
Andre Nollab5bd5c2008-10-13 11:55:12 +1100165 if (tmp >= min_sectors && tmp < conf->spacing)
166 conf->spacing = tmp;
NeilBrown15945fe2005-09-09 16:23:47 -0700167 }
168
Andre Nollab5bd5c2008-10-13 11:55:12 +1100169 /* spacing may be too large for sector_div to work with,
NeilBrown15945fe2005-09-09 16:23:47 -0700170 * so we might need to pre-shift
171 */
Andre Nollab5bd5c2008-10-13 11:55:12 +1100172 conf->sector_shift = 0;
NeilBrown15945fe2005-09-09 16:23:47 -0700173 if (sizeof(sector_t) > sizeof(u32)) {
Andre Nollab5bd5c2008-10-13 11:55:12 +1100174 sector_t space = conf->spacing;
NeilBrown15945fe2005-09-09 16:23:47 -0700175 while (space > (sector_t)(~(u32)0)) {
176 space >>= 1;
Andre Nollab5bd5c2008-10-13 11:55:12 +1100177 conf->sector_shift++;
NeilBrown15945fe2005-09-09 16:23:47 -0700178 }
179 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 /*
181 * This code was restructured to work around a gcc-2.95.3 internal
182 * compiler error. Alter it with care.
183 */
184 {
185 sector_t sz;
186 unsigned round;
187 unsigned long base;
188
Andre Nollab5bd5c2008-10-13 11:55:12 +1100189 sz = conf->array_sectors >> conf->sector_shift;
NeilBrown15945fe2005-09-09 16:23:47 -0700190 sz += 1; /* force round-up */
Andre Nollab5bd5c2008-10-13 11:55:12 +1100191 base = conf->spacing >> conf->sector_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 round = sector_div(sz, base);
NeilBrown15945fe2005-09-09 16:23:47 -0700193 nb_zone = sz + (round ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 }
NeilBrown15945fe2005-09-09 16:23:47 -0700195 BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *));
196
197 conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 GFP_KERNEL);
199 if (!conf->hash_table)
200 goto out;
201
202 /*
203 * Here we generate the linear hash table
NeilBrown15945fe2005-09-09 16:23:47 -0700204 * First calculate the device offsets.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 */
Andre Noll62838152008-10-13 11:55:12 +1100206 conf->disks[0].start_sector = 0;
NeilBrowna778b732007-05-23 13:58:10 -0700207 for (i = 1; i < raid_disks; i++)
Andre Noll62838152008-10-13 11:55:12 +1100208 conf->disks[i].start_sector =
209 conf->disks[i-1].start_sector +
210 conf->disks[i-1].num_sectors;
NeilBrown15945fe2005-09-09 16:23:47 -0700211
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 table = conf->hash_table;
NeilBrown15945fe2005-09-09 16:23:47 -0700213 i = 0;
Andre Noll62838152008-10-13 11:55:12 +1100214 for (curr_sector = 0;
215 curr_sector < conf->array_sectors;
Andre Nollab5bd5c2008-10-13 11:55:12 +1100216 curr_sector += conf->spacing) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
NeilBrowna778b732007-05-23 13:58:10 -0700218 while (i < raid_disks-1 &&
Andre Noll62838152008-10-13 11:55:12 +1100219 curr_sector >= conf->disks[i+1].start_sector)
NeilBrown15945fe2005-09-09 16:23:47 -0700220 i++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221
NeilBrown15945fe2005-09-09 16:23:47 -0700222 *table ++ = conf->disks + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 }
NeilBrown15945fe2005-09-09 16:23:47 -0700224
Andre Nollab5bd5c2008-10-13 11:55:12 +1100225 if (conf->sector_shift) {
226 conf->spacing >>= conf->sector_shift;
227 /* round spacing up so that when we divide by it,
NeilBrown15945fe2005-09-09 16:23:47 -0700228 * we err on the side of "too-low", which is safest.
229 */
Andre Nollab5bd5c2008-10-13 11:55:12 +1100230 conf->spacing++;
NeilBrown15945fe2005-09-09 16:23:47 -0700231 }
232
233 BUG_ON(table - conf->hash_table > nb_zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
NeilBrown7c7546c2006-06-26 00:27:41 -0700235 return conf;
236
237out:
238 kfree(conf);
239 return NULL;
240}
241
242static int linear_run (mddev_t *mddev)
243{
244 linear_conf_t *conf;
245
Neil Browne7e72bf2008-05-14 16:05:54 -0700246 mddev->queue->queue_lock = &mddev->queue->__queue_lock;
NeilBrown7c7546c2006-06-26 00:27:41 -0700247 conf = linear_conf(mddev, mddev->raid_disks);
248
249 if (!conf)
250 return 1;
251 mddev->private = conf;
Andre Nolld6e22152008-07-21 17:05:25 +1000252 mddev->array_sectors = conf->array_sectors;
NeilBrown7c7546c2006-06-26 00:27:41 -0700253
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
255 mddev->queue->unplug_fn = linear_unplug;
NeilBrown26be34d2006-10-03 01:15:53 -0700256 mddev->queue->backing_dev_info.congested_fn = linear_congested;
257 mddev->queue->backing_dev_info.congested_data = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 return 0;
NeilBrown7c7546c2006-06-26 00:27:41 -0700259}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
NeilBrown7c7546c2006-06-26 00:27:41 -0700261static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
262{
263 /* Adding a drive to a linear array allows the array to grow.
264 * It is permitted if the new drive has a matching superblock
265 * already on it, with raid_disk equal to raid_disks.
266 * It is achieved by creating a new linear_private_data structure
267 * and swapping it in in-place of the current one.
268 * The current one is never freed until the array is stopped.
269 * This avoids races.
270 */
271 linear_conf_t *newconf;
272
NeilBrowna778b732007-05-23 13:58:10 -0700273 if (rdev->saved_raid_disk != mddev->raid_disks)
NeilBrown7c7546c2006-06-26 00:27:41 -0700274 return -EINVAL;
275
NeilBrowna778b732007-05-23 13:58:10 -0700276 rdev->raid_disk = rdev->saved_raid_disk;
277
NeilBrown7c7546c2006-06-26 00:27:41 -0700278 newconf = linear_conf(mddev,mddev->raid_disks+1);
279
280 if (!newconf)
281 return -ENOMEM;
282
283 newconf->prev = mddev_to_conf(mddev);
284 mddev->private = newconf;
285 mddev->raid_disks++;
Andre Nolld6e22152008-07-21 17:05:25 +1000286 mddev->array_sectors = newconf->array_sectors;
Andre Nollf233ea52008-07-21 17:05:22 +1000287 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown7c7546c2006-06-26 00:27:41 -0700288 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289}
290
291static int linear_stop (mddev_t *mddev)
292{
293 linear_conf_t *conf = mddev_to_conf(mddev);
294
295 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
NeilBrown7c7546c2006-06-26 00:27:41 -0700296 do {
297 linear_conf_t *t = conf->prev;
298 kfree(conf->hash_table);
299 kfree(conf);
300 conf = t;
301 } while (conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
303 return 0;
304}
305
Jens Axboe165125e2007-07-24 09:28:11 +0200306static int linear_make_request (struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307{
Jens Axboea3623572005-11-01 09:26:16 +0100308 const int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 mddev_t *mddev = q->queuedata;
310 dev_info_t *tmp_dev;
Tejun Heoc9959052008-08-25 19:47:21 +0900311 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
NeilBrowne5dcdd82005-09-09 16:23:41 -0700313 if (unlikely(bio_barrier(bio))) {
NeilBrown6712ecf2007-09-27 12:47:43 +0200314 bio_endio(bio, -EOPNOTSUPP);
NeilBrowne5dcdd82005-09-09 16:23:41 -0700315 return 0;
316 }
317
Tejun Heo074a7ac2008-08-25 19:56:14 +0900318 cpu = part_stat_lock();
319 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
320 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
321 bio_sectors(bio));
322 part_stat_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
324 tmp_dev = which_dev(mddev, bio->bi_sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325
Andre Noll62838152008-10-13 11:55:12 +1100326 if (unlikely(bio->bi_sector >= (tmp_dev->num_sectors +
327 tmp_dev->start_sector)
328 || (bio->bi_sector <
329 tmp_dev->start_sector))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 char b[BDEVNAME_SIZE];
331
Andre Noll62838152008-10-13 11:55:12 +1100332 printk("linear_make_request: Sector %llu out of bounds on "
333 "dev %s: %llu sectors, offset %llu\n",
334 (unsigned long long)bio->bi_sector,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 bdevname(tmp_dev->rdev->bdev, b),
Andre Noll62838152008-10-13 11:55:12 +1100336 (unsigned long long)tmp_dev->num_sectors,
337 (unsigned long long)tmp_dev->start_sector);
NeilBrown6712ecf2007-09-27 12:47:43 +0200338 bio_io_error(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 return 0;
340 }
341 if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
Andre Noll62838152008-10-13 11:55:12 +1100342 tmp_dev->start_sector + tmp_dev->num_sectors)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 /* This bio crosses a device boundary, so we have to
344 * split it.
345 */
346 struct bio_pair *bp;
Andre Noll62838152008-10-13 11:55:12 +1100347
Denis ChengRq6feef532008-10-09 08:57:05 +0200348 bp = bio_split(bio,
Andre Noll62838152008-10-13 11:55:12 +1100349 tmp_dev->start_sector + tmp_dev->num_sectors
350 - bio->bi_sector);
351
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 if (linear_make_request(q, &bp->bio1))
353 generic_make_request(&bp->bio1);
354 if (linear_make_request(q, &bp->bio2))
355 generic_make_request(&bp->bio2);
356 bio_pair_release(bp);
357 return 0;
358 }
359
360 bio->bi_bdev = tmp_dev->rdev->bdev;
Andre Noll62838152008-10-13 11:55:12 +1100361 bio->bi_sector = bio->bi_sector - tmp_dev->start_sector
362 + tmp_dev->rdev->data_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363
364 return 1;
365}
366
367static void linear_status (struct seq_file *seq, mddev_t *mddev)
368{
369
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 seq_printf(seq, " %dk rounding", mddev->chunk_size/1024);
371}
372
373
NeilBrown2604b702006-01-06 00:20:36 -0800374static struct mdk_personality linear_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375{
376 .name = "linear",
NeilBrown2604b702006-01-06 00:20:36 -0800377 .level = LEVEL_LINEAR,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 .owner = THIS_MODULE,
379 .make_request = linear_make_request,
380 .run = linear_run,
381 .stop = linear_stop,
382 .status = linear_status,
NeilBrown7c7546c2006-06-26 00:27:41 -0700383 .hot_add_disk = linear_add,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384};
385
386static int __init linear_init (void)
387{
NeilBrown2604b702006-01-06 00:20:36 -0800388 return register_md_personality (&linear_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389}
390
391static void linear_exit (void)
392{
NeilBrown2604b702006-01-06 00:20:36 -0800393 unregister_md_personality (&linear_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394}
395
396
397module_init(linear_init);
398module_exit(linear_exit);
399MODULE_LICENSE("GPL");
NeilBrownd9d166c2006-01-06 00:20:51 -0800400MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
401MODULE_ALIAS("md-linear");
NeilBrown2604b702006-01-06 00:20:36 -0800402MODULE_ALIAS("md-level--1");