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);
38 * sector_div(a,b) returns the remainer and sets a to a/b
40 sector >>= conf->sector_shift;
41 (void)sector_div(sector, conf->spacing);
42 hash = conf->hash_table[sector];
44 while (sector >= hash->num_sectors + hash->start_sector)
50 * linear_mergeable_bvec -- tell bio layer if two requests can be merged
52 * @bvm: properties of new bio
53 * @biovec: the request that could be merged to it.
55 * Return amount of bytes we can take at this offset
57 static int linear_mergeable_bvec(struct request_queue *q,
58 struct bvec_merge_data *bvm,
59 struct bio_vec *biovec)
61 mddev_t *mddev = q->queuedata;
63 unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9;
64 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
66 dev0 = which_dev(mddev, sector);
67 maxsectors = dev0->num_sectors - (sector - dev0->start_sector);
69 if (maxsectors < bio_sectors)
72 maxsectors -= bio_sectors;
74 if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
75 return biovec->bv_len;
76 /* The bytes available at this offset could be really big,
77 * so we cap at 2^31 to avoid overflow */
78 if (maxsectors > (1 << (31-9)))
80 return maxsectors << 9;
83 static void linear_unplug(struct request_queue *q)
85 mddev_t *mddev = q->queuedata;
86 linear_conf_t *conf = mddev_to_conf(mddev);
89 for (i=0; i < mddev->raid_disks; i++) {
90 struct request_queue *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
95 static int linear_congested(void *data, int bits)
97 mddev_t *mddev = data;
98 linear_conf_t *conf = mddev_to_conf(mddev);
101 for (i = 0; i < mddev->raid_disks && !ret ; i++) {
102 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
103 ret |= bdi_congested(&q->backing_dev_info, bits);
108 static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
114 sector_t min_sectors;
115 sector_t curr_sector;
116 struct list_head *tmp;
118 conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
124 conf->array_sectors = 0;
126 rdev_for_each(rdev, tmp, mddev) {
127 int j = rdev->raid_disk;
128 dev_info_t *disk = conf->disks + j;
130 if (j < 0 || j >= raid_disks || disk->rdev) {
131 printk("linear: disk numbering problem. Aborting!\n");
137 blk_queue_stack_limits(mddev->queue,
138 rdev->bdev->bd_disk->queue);
139 /* as we don't honour merge_bvec_fn, we must never risk
140 * violating it, so limit ->max_sector to one PAGE, as
141 * a one page request is never in violation.
143 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
144 mddev->queue->max_sectors > (PAGE_SIZE>>9))
145 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
147 disk->num_sectors = rdev->size * 2;
148 conf->array_sectors += rdev->size * 2;
152 if (cnt != raid_disks) {
153 printk("linear: not enough drives present. Aborting!\n");
157 min_sectors = conf->array_sectors;
158 sector_div(min_sectors, PAGE_SIZE/sizeof(struct dev_info *));
160 /* min_sectors is the minimum spacing that will fit the hash
161 * table in one PAGE. This may be much smaller than needed.
162 * We find the smallest non-terminal set of consecutive devices
163 * that is larger than min_sectors and use the size of that as
166 conf->spacing = conf->array_sectors;
167 for (i=0; i < cnt-1 ; i++) {
170 for (j = i; j < cnt - 1 && tmp < min_sectors; j++)
171 tmp += conf->disks[j].num_sectors;
172 if (tmp >= min_sectors && tmp < conf->spacing)
176 /* spacing may be too large for sector_div to work with,
177 * so we might need to pre-shift
179 conf->sector_shift = 0;
180 if (sizeof(sector_t) > sizeof(u32)) {
181 sector_t space = conf->spacing;
182 while (space > (sector_t)(~(u32)0)) {
184 conf->sector_shift++;
188 * This code was restructured to work around a gcc-2.95.3 internal
189 * compiler error. Alter it with care.
196 sz = conf->array_sectors >> conf->sector_shift;
197 sz += 1; /* force round-up */
198 base = conf->spacing >> conf->sector_shift;
199 round = sector_div(sz, base);
200 nb_zone = sz + (round ? 1 : 0);
202 BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *));
204 conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone,
206 if (!conf->hash_table)
210 * Here we generate the linear hash table
211 * First calculate the device offsets.
213 conf->disks[0].start_sector = 0;
214 for (i = 1; i < raid_disks; i++)
215 conf->disks[i].start_sector =
216 conf->disks[i-1].start_sector +
217 conf->disks[i-1].num_sectors;
219 table = conf->hash_table;
221 for (curr_sector = 0;
222 curr_sector < conf->array_sectors;
223 curr_sector += conf->spacing) {
225 while (i < raid_disks-1 &&
226 curr_sector >= conf->disks[i+1].start_sector)
229 *table ++ = conf->disks + i;
232 if (conf->sector_shift) {
233 conf->spacing >>= conf->sector_shift;
234 /* round spacing up so that when we divide by it,
235 * we err on the side of "too-low", which is safest.
240 BUG_ON(table - conf->hash_table > nb_zone);
249 static int linear_run (mddev_t *mddev)
253 mddev->queue->queue_lock = &mddev->queue->__queue_lock;
254 conf = linear_conf(mddev, mddev->raid_disks);
258 mddev->private = conf;
259 mddev->array_sectors = conf->array_sectors;
261 blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
262 mddev->queue->unplug_fn = linear_unplug;
263 mddev->queue->backing_dev_info.congested_fn = linear_congested;
264 mddev->queue->backing_dev_info.congested_data = mddev;
268 static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
270 /* Adding a drive to a linear array allows the array to grow.
271 * It is permitted if the new drive has a matching superblock
272 * already on it, with raid_disk equal to raid_disks.
273 * It is achieved by creating a new linear_private_data structure
274 * and swapping it in in-place of the current one.
275 * The current one is never freed until the array is stopped.
278 linear_conf_t *newconf;
280 if (rdev->saved_raid_disk != mddev->raid_disks)
283 rdev->raid_disk = rdev->saved_raid_disk;
285 newconf = linear_conf(mddev,mddev->raid_disks+1);
290 newconf->prev = mddev_to_conf(mddev);
291 mddev->private = newconf;
293 mddev->array_sectors = newconf->array_sectors;
294 set_capacity(mddev->gendisk, mddev->array_sectors);
298 static int linear_stop (mddev_t *mddev)
300 linear_conf_t *conf = mddev_to_conf(mddev);
302 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
304 linear_conf_t *t = conf->prev;
305 kfree(conf->hash_table);
313 static int linear_make_request (struct request_queue *q, struct bio *bio)
315 const int rw = bio_data_dir(bio);
316 mddev_t *mddev = q->queuedata;
320 if (unlikely(bio_barrier(bio))) {
321 bio_endio(bio, -EOPNOTSUPP);
325 cpu = part_stat_lock();
326 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
327 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
331 tmp_dev = which_dev(mddev, bio->bi_sector);
333 if (unlikely(bio->bi_sector >= (tmp_dev->num_sectors +
334 tmp_dev->start_sector)
336 tmp_dev->start_sector))) {
337 char b[BDEVNAME_SIZE];
339 printk("linear_make_request: Sector %llu out of bounds on "
340 "dev %s: %llu sectors, offset %llu\n",
341 (unsigned long long)bio->bi_sector,
342 bdevname(tmp_dev->rdev->bdev, b),
343 (unsigned long long)tmp_dev->num_sectors,
344 (unsigned long long)tmp_dev->start_sector);
348 if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
349 tmp_dev->start_sector + tmp_dev->num_sectors)) {
350 /* This bio crosses a device boundary, so we have to
356 tmp_dev->start_sector + tmp_dev->num_sectors
359 if (linear_make_request(q, &bp->bio1))
360 generic_make_request(&bp->bio1);
361 if (linear_make_request(q, &bp->bio2))
362 generic_make_request(&bp->bio2);
363 bio_pair_release(bp);
367 bio->bi_bdev = tmp_dev->rdev->bdev;
368 bio->bi_sector = bio->bi_sector - tmp_dev->start_sector
369 + tmp_dev->rdev->data_offset;
374 static void linear_status (struct seq_file *seq, mddev_t *mddev)
377 seq_printf(seq, " %dk rounding", mddev->chunk_size/1024);
381 static struct mdk_personality linear_personality =
384 .level = LEVEL_LINEAR,
385 .owner = THIS_MODULE,
386 .make_request = linear_make_request,
389 .status = linear_status,
390 .hot_add_disk = linear_add,
393 static int __init linear_init (void)
395 return register_md_personality (&linear_personality);
398 static void linear_exit (void)
400 unregister_md_personality (&linear_personality);
404 module_init(linear_init);
405 module_exit(linear_exit);
406 MODULE_LICENSE("GPL");
407 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
408 MODULE_ALIAS("md-linear");
409 MODULE_ALIAS("md-level--1");