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/blkdev.h>
20 #include <linux/raid/md_u.h>
21 #include <linux/seq_file.h>
26 * find which device holds a particular offset
28 static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector)
31 linear_conf_t *conf = mddev_to_conf(mddev);
32 sector_t idx = sector >> conf->sector_shift;
35 * sector_div(a,b) returns the remainer and sets a to a/b
37 (void)sector_div(idx, conf->spacing);
38 hash = conf->hash_table[idx];
40 while (sector >= hash->num_sectors + hash->start_sector)
46 * linear_mergeable_bvec -- tell bio layer if two requests can be merged
48 * @bvm: properties of new bio
49 * @biovec: the request that could be merged to it.
51 * Return amount of bytes we can take at this offset
53 static int linear_mergeable_bvec(struct request_queue *q,
54 struct bvec_merge_data *bvm,
55 struct bio_vec *biovec)
57 mddev_t *mddev = q->queuedata;
59 unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9;
60 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
62 dev0 = which_dev(mddev, sector);
63 maxsectors = dev0->num_sectors - (sector - dev0->start_sector);
65 if (maxsectors < bio_sectors)
68 maxsectors -= bio_sectors;
70 if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
71 return biovec->bv_len;
72 /* The bytes available at this offset could be really big,
73 * so we cap at 2^31 to avoid overflow */
74 if (maxsectors > (1 << (31-9)))
76 return maxsectors << 9;
79 static void linear_unplug(struct request_queue *q)
81 mddev_t *mddev = q->queuedata;
82 linear_conf_t *conf = mddev_to_conf(mddev);
85 for (i=0; i < mddev->raid_disks; i++) {
86 struct request_queue *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
91 static int linear_congested(void *data, int bits)
93 mddev_t *mddev = data;
94 linear_conf_t *conf = mddev_to_conf(mddev);
97 for (i = 0; i < mddev->raid_disks && !ret ; i++) {
98 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
99 ret |= bdi_congested(&q->backing_dev_info, bits);
104 static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
110 sector_t min_sectors;
111 sector_t curr_sector;
113 conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
119 conf->array_sectors = 0;
121 list_for_each_entry(rdev, &mddev->disks, same_set) {
122 int j = rdev->raid_disk;
123 dev_info_t *disk = conf->disks + j;
125 if (j < 0 || j >= raid_disks || disk->rdev) {
126 printk("linear: disk numbering problem. Aborting!\n");
132 blk_queue_stack_limits(mddev->queue,
133 rdev->bdev->bd_disk->queue);
134 /* as we don't honour merge_bvec_fn, we must never risk
135 * violating it, so limit ->max_sector to one PAGE, as
136 * a one page request is never in violation.
138 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
139 mddev->queue->max_sectors > (PAGE_SIZE>>9))
140 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
142 disk->num_sectors = rdev->sectors;
143 conf->array_sectors += rdev->sectors;
147 if (cnt != raid_disks) {
148 printk("linear: not enough drives present. Aborting!\n");
152 min_sectors = conf->array_sectors;
153 sector_div(min_sectors, PAGE_SIZE/sizeof(struct dev_info *));
154 if (min_sectors == 0)
157 /* min_sectors is the minimum spacing that will fit the hash
158 * table in one PAGE. This may be much smaller than needed.
159 * We find the smallest non-terminal set of consecutive devices
160 * that is larger than min_sectors and use the size of that as
163 conf->spacing = conf->array_sectors;
164 for (i=0; i < cnt-1 ; i++) {
167 for (j = i; j < cnt - 1 && tmp < min_sectors; j++)
168 tmp += conf->disks[j].num_sectors;
169 if (tmp >= min_sectors && tmp < conf->spacing)
173 /* spacing may be too large for sector_div to work with,
174 * so we might need to pre-shift
176 conf->sector_shift = 0;
177 if (sizeof(sector_t) > sizeof(u32)) {
178 sector_t space = conf->spacing;
179 while (space > (sector_t)(~(u32)0)) {
181 conf->sector_shift++;
185 * This code was restructured to work around a gcc-2.95.3 internal
186 * compiler error. Alter it with care.
193 sz = conf->array_sectors >> conf->sector_shift;
194 sz += 1; /* force round-up */
195 base = conf->spacing >> conf->sector_shift;
196 round = sector_div(sz, base);
197 nb_zone = sz + (round ? 1 : 0);
199 BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *));
201 conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone,
203 if (!conf->hash_table)
207 * Here we generate the linear hash table
208 * First calculate the device offsets.
210 conf->disks[0].start_sector = 0;
211 for (i = 1; i < raid_disks; i++)
212 conf->disks[i].start_sector =
213 conf->disks[i-1].start_sector +
214 conf->disks[i-1].num_sectors;
216 table = conf->hash_table;
218 for (curr_sector = 0;
219 curr_sector < conf->array_sectors;
220 curr_sector += conf->spacing) {
222 while (i < raid_disks-1 &&
223 curr_sector >= conf->disks[i+1].start_sector)
226 *table ++ = conf->disks + i;
229 if (conf->sector_shift) {
230 conf->spacing >>= conf->sector_shift;
231 /* round spacing up so that when we divide by it,
232 * we err on the side of "too-low", which is safest.
237 BUG_ON(table - conf->hash_table > nb_zone);
246 static int linear_run (mddev_t *mddev)
250 mddev->queue->queue_lock = &mddev->queue->__queue_lock;
251 conf = linear_conf(mddev, mddev->raid_disks);
255 mddev->private = conf;
256 mddev->array_sectors = conf->array_sectors;
258 blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
259 mddev->queue->unplug_fn = linear_unplug;
260 mddev->queue->backing_dev_info.congested_fn = linear_congested;
261 mddev->queue->backing_dev_info.congested_data = mddev;
265 static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
267 /* Adding a drive to a linear array allows the array to grow.
268 * It is permitted if the new drive has a matching superblock
269 * already on it, with raid_disk equal to raid_disks.
270 * It is achieved by creating a new linear_private_data structure
271 * and swapping it in in-place of the current one.
272 * The current one is never freed until the array is stopped.
275 linear_conf_t *newconf;
277 if (rdev->saved_raid_disk != mddev->raid_disks)
280 rdev->raid_disk = rdev->saved_raid_disk;
282 newconf = linear_conf(mddev,mddev->raid_disks+1);
287 newconf->prev = mddev_to_conf(mddev);
288 mddev->private = newconf;
290 mddev->array_sectors = newconf->array_sectors;
291 set_capacity(mddev->gendisk, mddev->array_sectors);
295 static int linear_stop (mddev_t *mddev)
297 linear_conf_t *conf = mddev_to_conf(mddev);
299 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
301 linear_conf_t *t = conf->prev;
302 kfree(conf->hash_table);
310 static int linear_make_request (struct request_queue *q, struct bio *bio)
312 const int rw = bio_data_dir(bio);
313 mddev_t *mddev = q->queuedata;
317 if (unlikely(bio_barrier(bio))) {
318 bio_endio(bio, -EOPNOTSUPP);
322 cpu = part_stat_lock();
323 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
324 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
328 tmp_dev = which_dev(mddev, bio->bi_sector);
330 if (unlikely(bio->bi_sector >= (tmp_dev->num_sectors +
331 tmp_dev->start_sector)
333 tmp_dev->start_sector))) {
334 char b[BDEVNAME_SIZE];
336 printk("linear_make_request: Sector %llu out of bounds on "
337 "dev %s: %llu sectors, offset %llu\n",
338 (unsigned long long)bio->bi_sector,
339 bdevname(tmp_dev->rdev->bdev, b),
340 (unsigned long long)tmp_dev->num_sectors,
341 (unsigned long long)tmp_dev->start_sector);
345 if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
346 tmp_dev->start_sector + tmp_dev->num_sectors)) {
347 /* This bio crosses a device boundary, so we have to
353 tmp_dev->start_sector + tmp_dev->num_sectors
356 if (linear_make_request(q, &bp->bio1))
357 generic_make_request(&bp->bio1);
358 if (linear_make_request(q, &bp->bio2))
359 generic_make_request(&bp->bio2);
360 bio_pair_release(bp);
364 bio->bi_bdev = tmp_dev->rdev->bdev;
365 bio->bi_sector = bio->bi_sector - tmp_dev->start_sector
366 + tmp_dev->rdev->data_offset;
371 static void linear_status (struct seq_file *seq, mddev_t *mddev)
374 seq_printf(seq, " %dk rounding", mddev->chunk_size/1024);
378 static struct mdk_personality linear_personality =
381 .level = LEVEL_LINEAR,
382 .owner = THIS_MODULE,
383 .make_request = linear_make_request,
386 .status = linear_status,
387 .hot_add_disk = linear_add,
390 static int __init linear_init (void)
392 return register_md_personality (&linear_personality);
395 static void linear_exit (void)
397 unregister_md_personality (&linear_personality);
401 module_init(linear_init);
402 module_exit(linear_exit);
403 MODULE_LICENSE("GPL");
404 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
405 MODULE_ALIAS("md-linear");
406 MODULE_ALIAS("md-level--1");