blob: d5ecd2d53046ab5510b520df48813debdc45d3c8 [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 */
58static int linear_mergeable_bvec(request_queue_t *q, struct bio *bio, struct bio_vec *biovec)
59{
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
82static void linear_unplug(request_queue_t *q)
83{
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++) {
89 request_queue_t *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
90 if (r_queue->unplug_fn)
91 r_queue->unplug_fn(r_queue);
92 }
93}
94
95static int linear_issue_flush(request_queue_t *q, struct gendisk *disk,
96 sector_t *error_sector)
97{
98 mddev_t *mddev = q->queuedata;
99 linear_conf_t *conf = mddev_to_conf(mddev);
100 int i, ret = 0;
101
102 for (i=0; i < mddev->raid_disks && ret == 0; i++) {
103 struct block_device *bdev = conf->disks[i].rdev->bdev;
104 request_queue_t *r_queue = bdev_get_queue(bdev);
105
106 if (!r_queue->issue_flush_fn)
107 ret = -EOPNOTSUPP;
108 else
109 ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, error_sector);
110 }
111 return ret;
112}
113
NeilBrown26be34d2006-10-03 01:15:53 -0700114static int linear_congested(void *data, int bits)
115{
116 mddev_t *mddev = data;
117 linear_conf_t *conf = mddev_to_conf(mddev);
118 int i, ret = 0;
119
120 for (i = 0; i < mddev->raid_disks && !ret ; i++) {
121 request_queue_t *q = bdev_get_queue(conf->disks[i].rdev->bdev);
122 ret |= bdi_congested(&q->backing_dev_info, bits);
123 }
124 return ret;
125}
126
NeilBrown7c7546c2006-06-26 00:27:41 -0700127static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128{
129 linear_conf_t *conf;
130 dev_info_t **table;
131 mdk_rdev_t *rdev;
132 int i, nb_zone, cnt;
NeilBrown15945fe2005-09-09 16:23:47 -0700133 sector_t min_spacing;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 sector_t curr_offset;
135 struct list_head *tmp;
136
NeilBrown7c7546c2006-06-26 00:27:41 -0700137 conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 GFP_KERNEL);
139 if (!conf)
NeilBrown7c7546c2006-06-26 00:27:41 -0700140 return NULL;
141
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 mddev->private = conf;
143
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 cnt = 0;
NeilBrown7c7546c2006-06-26 00:27:41 -0700145 conf->array_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
147 ITERATE_RDEV(mddev,rdev,tmp) {
148 int j = rdev->raid_disk;
149 dev_info_t *disk = conf->disks + j;
150
NeilBrown7c7546c2006-06-26 00:27:41 -0700151 if (j < 0 || j > raid_disks || disk->rdev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 printk("linear: disk numbering problem. Aborting!\n");
153 goto out;
154 }
155
156 disk->rdev = rdev;
157
158 blk_queue_stack_limits(mddev->queue,
159 rdev->bdev->bd_disk->queue);
160 /* as we don't honour merge_bvec_fn, we must never risk
161 * violating it, so limit ->max_sector to one PAGE, as
162 * a one page request is never in violation.
163 */
164 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
165 mddev->queue->max_sectors > (PAGE_SIZE>>9))
166 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
167
168 disk->size = rdev->size;
NeilBrown7c7546c2006-06-26 00:27:41 -0700169 conf->array_size += rdev->size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 cnt++;
172 }
NeilBrown7c7546c2006-06-26 00:27:41 -0700173 if (cnt != raid_disks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 printk("linear: not enough drives present. Aborting!\n");
175 goto out;
176 }
177
NeilBrownf9abd1a2006-08-05 12:14:07 -0700178 min_spacing = conf->array_size;
NeilBrown15945fe2005-09-09 16:23:47 -0700179 sector_div(min_spacing, PAGE_SIZE/sizeof(struct dev_info *));
180
181 /* min_spacing is the minimum spacing that will fit the hash
182 * table in one PAGE. This may be much smaller than needed.
183 * We find the smallest non-terminal set of consecutive devices
184 * that is larger than min_spacing as use the size of that as
185 * the actual spacing
186 */
NeilBrownf9abd1a2006-08-05 12:14:07 -0700187 conf->hash_spacing = conf->array_size;
NeilBrown15945fe2005-09-09 16:23:47 -0700188 for (i=0; i < cnt-1 ; i++) {
189 sector_t sz = 0;
190 int j;
Andy Isaacsonbed31ed2007-03-16 13:38:04 -0800191 for (j = i; j < cnt - 1 && sz < min_spacing; j++)
NeilBrown15945fe2005-09-09 16:23:47 -0700192 sz += conf->disks[j].size;
193 if (sz >= min_spacing && sz < conf->hash_spacing)
194 conf->hash_spacing = sz;
195 }
196
197 /* hash_spacing may be too large for sector_div to work with,
198 * so we might need to pre-shift
199 */
200 conf->preshift = 0;
201 if (sizeof(sector_t) > sizeof(u32)) {
202 sector_t space = conf->hash_spacing;
203 while (space > (sector_t)(~(u32)0)) {
204 space >>= 1;
205 conf->preshift++;
206 }
207 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 /*
209 * This code was restructured to work around a gcc-2.95.3 internal
210 * compiler error. Alter it with care.
211 */
212 {
213 sector_t sz;
214 unsigned round;
215 unsigned long base;
216
NeilBrown7c7546c2006-06-26 00:27:41 -0700217 sz = conf->array_size >> conf->preshift;
NeilBrown15945fe2005-09-09 16:23:47 -0700218 sz += 1; /* force round-up */
219 base = conf->hash_spacing >> conf->preshift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 round = sector_div(sz, base);
NeilBrown15945fe2005-09-09 16:23:47 -0700221 nb_zone = sz + (round ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 }
NeilBrown15945fe2005-09-09 16:23:47 -0700223 BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *));
224
225 conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 GFP_KERNEL);
227 if (!conf->hash_table)
228 goto out;
229
230 /*
231 * Here we generate the linear hash table
NeilBrown15945fe2005-09-09 16:23:47 -0700232 * First calculate the device offsets.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 */
NeilBrown15945fe2005-09-09 16:23:47 -0700234 conf->disks[0].offset = 0;
235 for (i=1; i<mddev->raid_disks; i++)
236 conf->disks[i].offset =
237 conf->disks[i-1].offset +
238 conf->disks[i-1].size;
239
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 table = conf->hash_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 curr_offset = 0;
NeilBrown15945fe2005-09-09 16:23:47 -0700242 i = 0;
243 for (curr_offset = 0;
NeilBrownf9abd1a2006-08-05 12:14:07 -0700244 curr_offset < conf->array_size;
NeilBrown15945fe2005-09-09 16:23:47 -0700245 curr_offset += conf->hash_spacing) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
NeilBrown15945fe2005-09-09 16:23:47 -0700247 while (i < mddev->raid_disks-1 &&
248 curr_offset >= conf->disks[i+1].offset)
249 i++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250
NeilBrown15945fe2005-09-09 16:23:47 -0700251 *table ++ = conf->disks + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 }
NeilBrown15945fe2005-09-09 16:23:47 -0700253
254 if (conf->preshift) {
255 conf->hash_spacing >>= conf->preshift;
256 /* round hash_spacing up so that when we divide by it,
257 * we err on the side of "too-low", which is safest.
258 */
259 conf->hash_spacing++;
260 }
261
262 BUG_ON(table - conf->hash_table > nb_zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263
NeilBrown7c7546c2006-06-26 00:27:41 -0700264 return conf;
265
266out:
267 kfree(conf);
268 return NULL;
269}
270
271static int linear_run (mddev_t *mddev)
272{
273 linear_conf_t *conf;
274
275 conf = linear_conf(mddev, mddev->raid_disks);
276
277 if (!conf)
278 return 1;
279 mddev->private = conf;
280 mddev->array_size = conf->array_size;
281
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
283 mddev->queue->unplug_fn = linear_unplug;
284 mddev->queue->issue_flush_fn = linear_issue_flush;
NeilBrown26be34d2006-10-03 01:15:53 -0700285 mddev->queue->backing_dev_info.congested_fn = linear_congested;
286 mddev->queue->backing_dev_info.congested_data = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 return 0;
NeilBrown7c7546c2006-06-26 00:27:41 -0700288}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289
NeilBrown7c7546c2006-06-26 00:27:41 -0700290static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
291{
292 /* Adding a drive to a linear array allows the array to grow.
293 * It is permitted if the new drive has a matching superblock
294 * already on it, with raid_disk equal to raid_disks.
295 * It is achieved by creating a new linear_private_data structure
296 * and swapping it in in-place of the current one.
297 * The current one is never freed until the array is stopped.
298 * This avoids races.
299 */
300 linear_conf_t *newconf;
301
302 if (rdev->raid_disk != mddev->raid_disks)
303 return -EINVAL;
304
305 newconf = linear_conf(mddev,mddev->raid_disks+1);
306
307 if (!newconf)
308 return -ENOMEM;
309
310 newconf->prev = mddev_to_conf(mddev);
311 mddev->private = newconf;
312 mddev->raid_disks++;
313 mddev->array_size = newconf->array_size;
314 set_capacity(mddev->gendisk, mddev->array_size << 1);
315 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316}
317
318static int linear_stop (mddev_t *mddev)
319{
320 linear_conf_t *conf = mddev_to_conf(mddev);
321
322 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
NeilBrown7c7546c2006-06-26 00:27:41 -0700323 do {
324 linear_conf_t *t = conf->prev;
325 kfree(conf->hash_table);
326 kfree(conf);
327 conf = t;
328 } while (conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329
330 return 0;
331}
332
333static int linear_make_request (request_queue_t *q, struct bio *bio)
334{
Jens Axboea3623572005-11-01 09:26:16 +0100335 const int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 mddev_t *mddev = q->queuedata;
337 dev_info_t *tmp_dev;
338 sector_t block;
339
NeilBrowne5dcdd82005-09-09 16:23:41 -0700340 if (unlikely(bio_barrier(bio))) {
341 bio_endio(bio, bio->bi_size, -EOPNOTSUPP);
342 return 0;
343 }
344
Jens Axboea3623572005-11-01 09:26:16 +0100345 disk_stat_inc(mddev->gendisk, ios[rw]);
346 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
348 tmp_dev = which_dev(mddev, bio->bi_sector);
349 block = bio->bi_sector >> 1;
350
351 if (unlikely(block >= (tmp_dev->size + tmp_dev->offset)
352 || block < tmp_dev->offset)) {
353 char b[BDEVNAME_SIZE];
354
355 printk("linear_make_request: Block %llu out of bounds on "
356 "dev %s size %llu offset %llu\n",
357 (unsigned long long)block,
358 bdevname(tmp_dev->rdev->bdev, b),
359 (unsigned long long)tmp_dev->size,
360 (unsigned long long)tmp_dev->offset);
361 bio_io_error(bio, bio->bi_size);
362 return 0;
363 }
364 if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
365 (tmp_dev->offset + tmp_dev->size)<<1)) {
366 /* This bio crosses a device boundary, so we have to
367 * split it.
368 */
369 struct bio_pair *bp;
NeilBrown29ac8e02005-05-16 21:53:15 -0700370 bp = bio_split(bio, bio_split_pool,
371 ((tmp_dev->offset + tmp_dev->size)<<1) - bio->bi_sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 if (linear_make_request(q, &bp->bio1))
373 generic_make_request(&bp->bio1);
374 if (linear_make_request(q, &bp->bio2))
375 generic_make_request(&bp->bio2);
376 bio_pair_release(bp);
377 return 0;
378 }
379
380 bio->bi_bdev = tmp_dev->rdev->bdev;
381 bio->bi_sector = bio->bi_sector - (tmp_dev->offset << 1) + tmp_dev->rdev->data_offset;
382
383 return 1;
384}
385
386static void linear_status (struct seq_file *seq, mddev_t *mddev)
387{
388
389#undef MD_DEBUG
390#ifdef MD_DEBUG
391 int j;
392 linear_conf_t *conf = mddev_to_conf(mddev);
393 sector_t s = 0;
394
395 seq_printf(seq, " ");
NeilBrown15945fe2005-09-09 16:23:47 -0700396 for (j = 0; j < mddev->raid_disks; j++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 {
398 char b[BDEVNAME_SIZE];
399 s += conf->smallest_size;
400 seq_printf(seq, "[%s",
401 bdevname(conf->hash_table[j][0].rdev->bdev,b));
402
403 while (s > conf->hash_table[j][0].offset +
404 conf->hash_table[j][0].size)
405 seq_printf(seq, "/%s] ",
406 bdevname(conf->hash_table[j][1].rdev->bdev,b));
407 else
408 seq_printf(seq, "] ");
409 }
410 seq_printf(seq, "\n");
411#endif
412 seq_printf(seq, " %dk rounding", mddev->chunk_size/1024);
413}
414
415
NeilBrown2604b702006-01-06 00:20:36 -0800416static struct mdk_personality linear_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417{
418 .name = "linear",
NeilBrown2604b702006-01-06 00:20:36 -0800419 .level = LEVEL_LINEAR,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 .owner = THIS_MODULE,
421 .make_request = linear_make_request,
422 .run = linear_run,
423 .stop = linear_stop,
424 .status = linear_status,
NeilBrown7c7546c2006-06-26 00:27:41 -0700425 .hot_add_disk = linear_add,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426};
427
428static int __init linear_init (void)
429{
NeilBrown2604b702006-01-06 00:20:36 -0800430 return register_md_personality (&linear_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431}
432
433static void linear_exit (void)
434{
NeilBrown2604b702006-01-06 00:20:36 -0800435 unregister_md_personality (&linear_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436}
437
438
439module_init(linear_init);
440module_exit(linear_exit);
441MODULE_LICENSE("GPL");
NeilBrownd9d166c2006-01-06 00:20:51 -0800442MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
443MODULE_ALIAS("md-linear");
NeilBrown2604b702006-01-06 00:20:36 -0800444MODULE_ALIAS("md-level--1");