2 linear.c : Multiple Devices driver for Linux
3 Copyright (C) 1994-96 Marc ZYNGIER
4 <zyngier@ufr-info-p7.ibp.fr> or
7 Linear mode management functions.
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)
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.
19 #include <linux/module.h>
21 #include <linux/raid/md.h>
22 #include <linux/slab.h>
23 #include <linux/raid/linear.h>
25 #define MAJOR_NR MD_MAJOR
27 #define MD_PERSONALITY
30 * find which device holds a particular offset
32 static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector)
35 linear_conf_t *conf = mddev_to_conf(mddev);
36 sector_t block = sector >> 1;
39 * sector_div(a,b) returns the remainer and sets a to a/b
41 block >>= conf->preshift;
42 (void)sector_div(block, conf->hash_spacing);
43 hash = conf->hash_table[block];
45 while ((sector>>1) >= (hash->size + hash->offset))
51 * linear_mergeable_bvec -- tell bio layer if two requests can be merged
53 * @bio: the buffer head that's been built up so far
54 * @biovec: the request that could be merged to it.
56 * Return amount of bytes we can take at this offset
58 static int linear_mergeable_bvec(struct request_queue *q, struct bio *bio, struct bio_vec *biovec)
60 mddev_t *mddev = q->queuedata;
62 unsigned long maxsectors, bio_sectors = bio->bi_size >> 9;
63 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
65 dev0 = which_dev(mddev, sector);
66 maxsectors = (dev0->size << 1) - (sector - (dev0->offset<<1));
68 if (maxsectors < bio_sectors)
71 maxsectors -= bio_sectors;
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)))
79 return maxsectors << 9;
82 static void linear_unplug(struct request_queue *q)
84 mddev_t *mddev = q->queuedata;
85 linear_conf_t *conf = mddev_to_conf(mddev);
88 for (i=0; i < mddev->raid_disks; i++) {
89 struct request_queue *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
94 static int linear_congested(void *data, int bits)
96 mddev_t *mddev = data;
97 linear_conf_t *conf = mddev_to_conf(mddev);
100 for (i = 0; i < mddev->raid_disks && !ret ; i++) {
101 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
102 ret |= bdi_congested(&q->backing_dev_info, bits);
107 static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
113 sector_t min_spacing;
114 sector_t curr_offset;
115 struct list_head *tmp;
117 conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
123 conf->array_size = 0;
125 ITERATE_RDEV(mddev,rdev,tmp) {
126 int j = rdev->raid_disk;
127 dev_info_t *disk = conf->disks + j;
129 if (j < 0 || j > raid_disks || disk->rdev) {
130 printk("linear: disk numbering problem. Aborting!\n");
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.
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);
146 disk->size = rdev->size;
147 conf->array_size += rdev->size;
151 if (cnt != raid_disks) {
152 printk("linear: not enough drives present. Aborting!\n");
156 min_spacing = conf->array_size;
157 sector_div(min_spacing, PAGE_SIZE/sizeof(struct dev_info *));
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
165 conf->hash_spacing = conf->array_size;
166 for (i=0; i < cnt-1 ; i++) {
169 for (j = i; j < cnt - 1 && sz < min_spacing; j++)
170 sz += conf->disks[j].size;
171 if (sz >= min_spacing && sz < conf->hash_spacing)
172 conf->hash_spacing = sz;
175 /* hash_spacing may be too large for sector_div to work with,
176 * so we might need to pre-shift
179 if (sizeof(sector_t) > sizeof(u32)) {
180 sector_t space = conf->hash_spacing;
181 while (space > (sector_t)(~(u32)0)) {
187 * This code was restructured to work around a gcc-2.95.3 internal
188 * compiler error. Alter it with care.
195 sz = conf->array_size >> conf->preshift;
196 sz += 1; /* force round-up */
197 base = conf->hash_spacing >> conf->preshift;
198 round = sector_div(sz, base);
199 nb_zone = sz + (round ? 1 : 0);
201 BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *));
203 conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone,
205 if (!conf->hash_table)
209 * Here we generate the linear hash table
210 * First calculate the device offsets.
212 conf->disks[0].offset = 0;
213 for (i = 1; i < raid_disks; i++)
214 conf->disks[i].offset =
215 conf->disks[i-1].offset +
216 conf->disks[i-1].size;
218 table = conf->hash_table;
221 for (curr_offset = 0;
222 curr_offset < conf->array_size;
223 curr_offset += conf->hash_spacing) {
225 while (i < raid_disks-1 &&
226 curr_offset >= conf->disks[i+1].offset)
229 *table ++ = conf->disks + i;
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.
237 conf->hash_spacing++;
240 BUG_ON(table - conf->hash_table > nb_zone);
249 static int linear_run (mddev_t *mddev)
253 conf = linear_conf(mddev, mddev->raid_disks);
257 mddev->private = conf;
258 mddev->array_size = conf->array_size;
260 blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
261 mddev->queue->unplug_fn = linear_unplug;
262 mddev->queue->backing_dev_info.congested_fn = linear_congested;
263 mddev->queue->backing_dev_info.congested_data = mddev;
267 static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
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.
277 linear_conf_t *newconf;
279 if (rdev->saved_raid_disk != mddev->raid_disks)
282 rdev->raid_disk = rdev->saved_raid_disk;
284 newconf = linear_conf(mddev,mddev->raid_disks+1);
289 newconf->prev = mddev_to_conf(mddev);
290 mddev->private = newconf;
292 mddev->array_size = newconf->array_size;
293 set_capacity(mddev->gendisk, mddev->array_size << 1);
297 static int linear_stop (mddev_t *mddev)
299 linear_conf_t *conf = mddev_to_conf(mddev);
301 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
303 linear_conf_t *t = conf->prev;
304 kfree(conf->hash_table);
312 static int linear_make_request (struct request_queue *q, struct bio *bio)
314 const int rw = bio_data_dir(bio);
315 mddev_t *mddev = q->queuedata;
319 if (unlikely(bio_barrier(bio))) {
320 bio_endio(bio, -EOPNOTSUPP);
324 disk_stat_inc(mddev->gendisk, ios[rw]);
325 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
327 tmp_dev = which_dev(mddev, bio->bi_sector);
328 block = bio->bi_sector >> 1;
330 if (unlikely(block >= (tmp_dev->size + tmp_dev->offset)
331 || block < tmp_dev->offset)) {
332 char b[BDEVNAME_SIZE];
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);
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
349 bp = bio_split(bio, bio_split_pool,
350 ((tmp_dev->offset + tmp_dev->size)<<1) - bio->bi_sector);
351 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);
359 bio->bi_bdev = tmp_dev->rdev->bdev;
360 bio->bi_sector = bio->bi_sector - (tmp_dev->offset << 1) + tmp_dev->rdev->data_offset;
365 static void linear_status (struct seq_file *seq, mddev_t *mddev)
371 linear_conf_t *conf = mddev_to_conf(mddev);
374 seq_printf(seq, " ");
375 for (j = 0; j < mddev->raid_disks; j++)
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));
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));
387 seq_printf(seq, "] ");
389 seq_printf(seq, "\n");
391 seq_printf(seq, " %dk rounding", mddev->chunk_size/1024);
395 static struct mdk_personality linear_personality =
398 .level = LEVEL_LINEAR,
399 .owner = THIS_MODULE,
400 .make_request = linear_make_request,
403 .status = linear_status,
404 .hot_add_disk = linear_add,
407 static int __init linear_init (void)
409 return register_md_personality (&linear_personality);
412 static void linear_exit (void)
414 unregister_md_personality (&linear_personality);
418 module_init(linear_init);
419 module_exit(linear_exit);
420 MODULE_LICENSE("GPL");
421 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
422 MODULE_ALIAS("md-linear");
423 MODULE_ALIAS("md-level--1");