blob: 0b8511776b3e75276e9a472ea759e0aa304fff48 [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
19#include <linux/module.h>
20
21#include <linux/raid/md.h>
22#include <linux/slab.h>
23#include <linux/raid/linear.h>
24
25#define MAJOR_NR MD_MAJOR
26#define MD_DRIVER
27#define MD_PERSONALITY
28
29/*
30 * find which device holds a particular offset
31 */
32static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector)
33{
34 dev_info_t *hash;
35 linear_conf_t *conf = mddev_to_conf(mddev);
36 sector_t block = sector >> 1;
37
38 /*
39 * sector_div(a,b) returns the remainer and sets a to a/b
40 */
NeilBrown15945fe2005-09-09 16:23:47 -070041 block >>= conf->preshift;
42 (void)sector_div(block, conf->hash_spacing);
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 hash = conf->hash_table[block];
44
45 while ((sector>>1) >= (hash->size + hash->offset))
46 hash++;
47 return hash;
48}
49
50/**
NeilBrown15945fe2005-09-09 16:23:47 -070051 * linear_mergeable_bvec -- tell bio layer if two requests can be merged
Linus Torvalds1da177e2005-04-16 15:20:36 -070052 * @q: request queue
53 * @bio: the buffer head that's been built up so far
54 * @biovec: the request that could be merged to it.
55 *
56 * Return amount of bytes we can take at this offset
57 */
Jens Axboe165125e2007-07-24 09:28:11 +020058static int linear_mergeable_bvec(struct request_queue *q, struct bio *bio, struct bio_vec *biovec)
Linus Torvalds1da177e2005-04-16 15:20:36 -070059{
60 mddev_t *mddev = q->queuedata;
61 dev_info_t *dev0;
62 unsigned long maxsectors, bio_sectors = bio->bi_size >> 9;
63 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
64
65 dev0 = which_dev(mddev, sector);
66 maxsectors = (dev0->size << 1) - (sector - (dev0->offset<<1));
67
68 if (maxsectors < bio_sectors)
69 maxsectors = 0;
70 else
71 maxsectors -= bio_sectors;
72
73 if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
74 return biovec->bv_len;
75 /* The bytes available at this offset could be really big,
76 * so we cap at 2^31 to avoid overflow */
77 if (maxsectors > (1 << (31-9)))
78 return 1<<31;
79 return maxsectors << 9;
80}
81
Jens Axboe165125e2007-07-24 09:28:11 +020082static void linear_unplug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070083{
84 mddev_t *mddev = q->queuedata;
85 linear_conf_t *conf = mddev_to_conf(mddev);
86 int i;
87
88 for (i=0; i < mddev->raid_disks; i++) {
Jens Axboe165125e2007-07-24 09:28:11 +020089 struct request_queue *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -050090 blk_unplug(r_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 }
92}
93
NeilBrown26be34d2006-10-03 01:15:53 -070094static int linear_congested(void *data, int bits)
95{
96 mddev_t *mddev = data;
97 linear_conf_t *conf = mddev_to_conf(mddev);
98 int i, ret = 0;
99
100 for (i = 0; i < mddev->raid_disks && !ret ; i++) {
Jens Axboe165125e2007-07-24 09:28:11 +0200101 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
NeilBrown26be34d2006-10-03 01:15:53 -0700102 ret |= bdi_congested(&q->backing_dev_info, bits);
103 }
104 return ret;
105}
106
NeilBrown7c7546c2006-06-26 00:27:41 -0700107static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108{
109 linear_conf_t *conf;
110 dev_info_t **table;
111 mdk_rdev_t *rdev;
112 int i, nb_zone, cnt;
NeilBrown15945fe2005-09-09 16:23:47 -0700113 sector_t min_spacing;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 sector_t curr_offset;
115 struct list_head *tmp;
116
NeilBrown7c7546c2006-06-26 00:27:41 -0700117 conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 GFP_KERNEL);
119 if (!conf)
NeilBrown7c7546c2006-06-26 00:27:41 -0700120 return NULL;
121
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 cnt = 0;
NeilBrown7c7546c2006-06-26 00:27:41 -0700123 conf->array_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
NeilBrownd089c6a2008-02-06 01:39:59 -0800125 rdev_for_each(rdev, tmp, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 int j = rdev->raid_disk;
127 dev_info_t *disk = conf->disks + j;
128
NeilBrown7c7546c2006-06-26 00:27:41 -0700129 if (j < 0 || j > raid_disks || disk->rdev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 printk("linear: disk numbering problem. Aborting!\n");
131 goto out;
132 }
133
134 disk->rdev = rdev;
135
136 blk_queue_stack_limits(mddev->queue,
137 rdev->bdev->bd_disk->queue);
138 /* as we don't honour merge_bvec_fn, we must never risk
139 * violating it, so limit ->max_sector to one PAGE, as
140 * a one page request is never in violation.
141 */
142 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
143 mddev->queue->max_sectors > (PAGE_SIZE>>9))
144 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
145
146 disk->size = rdev->size;
NeilBrown7c7546c2006-06-26 00:27:41 -0700147 conf->array_size += rdev->size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 cnt++;
150 }
NeilBrown7c7546c2006-06-26 00:27:41 -0700151 if (cnt != raid_disks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 printk("linear: not enough drives present. Aborting!\n");
153 goto out;
154 }
155
NeilBrownf9abd1a2006-08-05 12:14:07 -0700156 min_spacing = conf->array_size;
NeilBrown15945fe2005-09-09 16:23:47 -0700157 sector_div(min_spacing, PAGE_SIZE/sizeof(struct dev_info *));
158
159 /* min_spacing is the minimum spacing that will fit the hash
160 * table in one PAGE. This may be much smaller than needed.
161 * We find the smallest non-terminal set of consecutive devices
162 * that is larger than min_spacing as use the size of that as
163 * the actual spacing
164 */
NeilBrownf9abd1a2006-08-05 12:14:07 -0700165 conf->hash_spacing = conf->array_size;
NeilBrown15945fe2005-09-09 16:23:47 -0700166 for (i=0; i < cnt-1 ; i++) {
167 sector_t sz = 0;
168 int j;
Andy Isaacsonbed31ed2007-03-16 13:38:04 -0800169 for (j = i; j < cnt - 1 && sz < min_spacing; j++)
NeilBrown15945fe2005-09-09 16:23:47 -0700170 sz += conf->disks[j].size;
171 if (sz >= min_spacing && sz < conf->hash_spacing)
172 conf->hash_spacing = sz;
173 }
174
175 /* hash_spacing may be too large for sector_div to work with,
176 * so we might need to pre-shift
177 */
178 conf->preshift = 0;
179 if (sizeof(sector_t) > sizeof(u32)) {
180 sector_t space = conf->hash_spacing;
181 while (space > (sector_t)(~(u32)0)) {
182 space >>= 1;
183 conf->preshift++;
184 }
185 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 /*
187 * This code was restructured to work around a gcc-2.95.3 internal
188 * compiler error. Alter it with care.
189 */
190 {
191 sector_t sz;
192 unsigned round;
193 unsigned long base;
194
NeilBrown7c7546c2006-06-26 00:27:41 -0700195 sz = conf->array_size >> conf->preshift;
NeilBrown15945fe2005-09-09 16:23:47 -0700196 sz += 1; /* force round-up */
197 base = conf->hash_spacing >> conf->preshift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 round = sector_div(sz, base);
NeilBrown15945fe2005-09-09 16:23:47 -0700199 nb_zone = sz + (round ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 }
NeilBrown15945fe2005-09-09 16:23:47 -0700201 BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *));
202
203 conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 GFP_KERNEL);
205 if (!conf->hash_table)
206 goto out;
207
208 /*
209 * Here we generate the linear hash table
NeilBrown15945fe2005-09-09 16:23:47 -0700210 * First calculate the device offsets.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 */
NeilBrown15945fe2005-09-09 16:23:47 -0700212 conf->disks[0].offset = 0;
NeilBrowna778b732007-05-23 13:58:10 -0700213 for (i = 1; i < raid_disks; i++)
NeilBrown15945fe2005-09-09 16:23:47 -0700214 conf->disks[i].offset =
215 conf->disks[i-1].offset +
216 conf->disks[i-1].size;
217
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 table = conf->hash_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 curr_offset = 0;
NeilBrown15945fe2005-09-09 16:23:47 -0700220 i = 0;
221 for (curr_offset = 0;
NeilBrownf9abd1a2006-08-05 12:14:07 -0700222 curr_offset < conf->array_size;
NeilBrown15945fe2005-09-09 16:23:47 -0700223 curr_offset += conf->hash_spacing) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224
NeilBrowna778b732007-05-23 13:58:10 -0700225 while (i < raid_disks-1 &&
NeilBrown15945fe2005-09-09 16:23:47 -0700226 curr_offset >= conf->disks[i+1].offset)
227 i++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
NeilBrown15945fe2005-09-09 16:23:47 -0700229 *table ++ = conf->disks + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 }
NeilBrown15945fe2005-09-09 16:23:47 -0700231
232 if (conf->preshift) {
233 conf->hash_spacing >>= conf->preshift;
234 /* round hash_spacing up so that when we divide by it,
235 * we err on the side of "too-low", which is safest.
236 */
237 conf->hash_spacing++;
238 }
239
240 BUG_ON(table - conf->hash_table > nb_zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
NeilBrown7c7546c2006-06-26 00:27:41 -0700242 return conf;
243
244out:
245 kfree(conf);
246 return NULL;
247}
248
249static int linear_run (mddev_t *mddev)
250{
251 linear_conf_t *conf;
252
253 conf = linear_conf(mddev, mddev->raid_disks);
254
255 if (!conf)
256 return 1;
257 mddev->private = conf;
258 mddev->array_size = conf->array_size;
259
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
261 mddev->queue->unplug_fn = linear_unplug;
NeilBrown26be34d2006-10-03 01:15:53 -0700262 mddev->queue->backing_dev_info.congested_fn = linear_congested;
263 mddev->queue->backing_dev_info.congested_data = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 return 0;
NeilBrown7c7546c2006-06-26 00:27:41 -0700265}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266
NeilBrown7c7546c2006-06-26 00:27:41 -0700267static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
268{
269 /* Adding a drive to a linear array allows the array to grow.
270 * It is permitted if the new drive has a matching superblock
271 * already on it, with raid_disk equal to raid_disks.
272 * It is achieved by creating a new linear_private_data structure
273 * and swapping it in in-place of the current one.
274 * The current one is never freed until the array is stopped.
275 * This avoids races.
276 */
277 linear_conf_t *newconf;
278
NeilBrowna778b732007-05-23 13:58:10 -0700279 if (rdev->saved_raid_disk != mddev->raid_disks)
NeilBrown7c7546c2006-06-26 00:27:41 -0700280 return -EINVAL;
281
NeilBrowna778b732007-05-23 13:58:10 -0700282 rdev->raid_disk = rdev->saved_raid_disk;
283
NeilBrown7c7546c2006-06-26 00:27:41 -0700284 newconf = linear_conf(mddev,mddev->raid_disks+1);
285
286 if (!newconf)
287 return -ENOMEM;
288
289 newconf->prev = mddev_to_conf(mddev);
290 mddev->private = newconf;
291 mddev->raid_disks++;
292 mddev->array_size = newconf->array_size;
293 set_capacity(mddev->gendisk, mddev->array_size << 1);
294 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295}
296
297static int linear_stop (mddev_t *mddev)
298{
299 linear_conf_t *conf = mddev_to_conf(mddev);
300
301 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
NeilBrown7c7546c2006-06-26 00:27:41 -0700302 do {
303 linear_conf_t *t = conf->prev;
304 kfree(conf->hash_table);
305 kfree(conf);
306 conf = t;
307 } while (conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
309 return 0;
310}
311
Jens Axboe165125e2007-07-24 09:28:11 +0200312static int linear_make_request (struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313{
Jens Axboea3623572005-11-01 09:26:16 +0100314 const int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 mddev_t *mddev = q->queuedata;
316 dev_info_t *tmp_dev;
317 sector_t block;
318
NeilBrowne5dcdd82005-09-09 16:23:41 -0700319 if (unlikely(bio_barrier(bio))) {
NeilBrown6712ecf2007-09-27 12:47:43 +0200320 bio_endio(bio, -EOPNOTSUPP);
NeilBrowne5dcdd82005-09-09 16:23:41 -0700321 return 0;
322 }
323
Jens Axboea3623572005-11-01 09:26:16 +0100324 disk_stat_inc(mddev->gendisk, ios[rw]);
325 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
327 tmp_dev = which_dev(mddev, bio->bi_sector);
328 block = bio->bi_sector >> 1;
329
330 if (unlikely(block >= (tmp_dev->size + tmp_dev->offset)
331 || block < tmp_dev->offset)) {
332 char b[BDEVNAME_SIZE];
333
334 printk("linear_make_request: Block %llu out of bounds on "
335 "dev %s size %llu offset %llu\n",
336 (unsigned long long)block,
337 bdevname(tmp_dev->rdev->bdev, b),
338 (unsigned long long)tmp_dev->size,
339 (unsigned long long)tmp_dev->offset);
NeilBrown6712ecf2007-09-27 12:47:43 +0200340 bio_io_error(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 return 0;
342 }
343 if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
344 (tmp_dev->offset + tmp_dev->size)<<1)) {
345 /* This bio crosses a device boundary, so we have to
346 * split it.
347 */
348 struct bio_pair *bp;
NeilBrown29ac8e02005-05-16 21:53:15 -0700349 bp = bio_split(bio, bio_split_pool,
350 ((tmp_dev->offset + tmp_dev->size)<<1) - bio->bi_sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 if (linear_make_request(q, &bp->bio1))
352 generic_make_request(&bp->bio1);
353 if (linear_make_request(q, &bp->bio2))
354 generic_make_request(&bp->bio2);
355 bio_pair_release(bp);
356 return 0;
357 }
358
359 bio->bi_bdev = tmp_dev->rdev->bdev;
360 bio->bi_sector = bio->bi_sector - (tmp_dev->offset << 1) + tmp_dev->rdev->data_offset;
361
362 return 1;
363}
364
365static void linear_status (struct seq_file *seq, mddev_t *mddev)
366{
367
368#undef MD_DEBUG
369#ifdef MD_DEBUG
370 int j;
371 linear_conf_t *conf = mddev_to_conf(mddev);
372 sector_t s = 0;
373
374 seq_printf(seq, " ");
NeilBrown15945fe2005-09-09 16:23:47 -0700375 for (j = 0; j < mddev->raid_disks; j++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 {
377 char b[BDEVNAME_SIZE];
378 s += conf->smallest_size;
379 seq_printf(seq, "[%s",
380 bdevname(conf->hash_table[j][0].rdev->bdev,b));
381
382 while (s > conf->hash_table[j][0].offset +
383 conf->hash_table[j][0].size)
384 seq_printf(seq, "/%s] ",
385 bdevname(conf->hash_table[j][1].rdev->bdev,b));
386 else
387 seq_printf(seq, "] ");
388 }
389 seq_printf(seq, "\n");
390#endif
391 seq_printf(seq, " %dk rounding", mddev->chunk_size/1024);
392}
393
394
NeilBrown2604b702006-01-06 00:20:36 -0800395static struct mdk_personality linear_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396{
397 .name = "linear",
NeilBrown2604b702006-01-06 00:20:36 -0800398 .level = LEVEL_LINEAR,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 .owner = THIS_MODULE,
400 .make_request = linear_make_request,
401 .run = linear_run,
402 .stop = linear_stop,
403 .status = linear_status,
NeilBrown7c7546c2006-06-26 00:27:41 -0700404 .hot_add_disk = linear_add,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405};
406
407static int __init linear_init (void)
408{
NeilBrown2604b702006-01-06 00:20:36 -0800409 return register_md_personality (&linear_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410}
411
412static void linear_exit (void)
413{
NeilBrown2604b702006-01-06 00:20:36 -0800414 unregister_md_personality (&linear_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415}
416
417
418module_init(linear_init);
419module_exit(linear_exit);
420MODULE_LICENSE("GPL");
NeilBrownd9d166c2006-01-06 00:20:51 -0800421MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
422MODULE_ALIAS("md-linear");
NeilBrown2604b702006-01-06 00:20:36 -0800423MODULE_ALIAS("md-level--1");