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(request_queue_t *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(request_queue_t *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 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);
95 static int linear_issue_flush(request_queue_t *q, struct gendisk *disk,
96 sector_t *error_sector)
98 mddev_t *mddev = q->queuedata;
99 linear_conf_t *conf = mddev_to_conf(mddev);
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);
106 if (!r_queue->issue_flush_fn)
109 ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, error_sector);
114 static int linear_congested(void *data, int bits)
116 mddev_t *mddev = data;
117 linear_conf_t *conf = mddev_to_conf(mddev);
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);
127 static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
133 sector_t min_spacing;
134 sector_t curr_offset;
135 struct list_head *tmp;
137 conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
142 mddev->private = conf;
145 conf->array_size = 0;
147 ITERATE_RDEV(mddev,rdev,tmp) {
148 int j = rdev->raid_disk;
149 dev_info_t *disk = conf->disks + j;
151 if (j < 0 || j > raid_disks || disk->rdev) {
152 printk("linear: disk numbering problem. Aborting!\n");
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.
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);
168 disk->size = rdev->size;
169 conf->array_size += rdev->size;
173 if (cnt != raid_disks) {
174 printk("linear: not enough drives present. Aborting!\n");
178 min_spacing = conf->array_size;
179 sector_div(min_spacing, PAGE_SIZE/sizeof(struct dev_info *));
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
187 conf->hash_spacing = conf->array_size;
188 for (i=0; i < cnt-1 ; i++) {
191 for (j=i; i<cnt-1 && sz < min_spacing ; j++)
192 sz += conf->disks[j].size;
193 if (sz >= min_spacing && sz < conf->hash_spacing)
194 conf->hash_spacing = sz;
197 /* hash_spacing may be too large for sector_div to work with,
198 * so we might need to pre-shift
201 if (sizeof(sector_t) > sizeof(u32)) {
202 sector_t space = conf->hash_spacing;
203 while (space > (sector_t)(~(u32)0)) {
209 * This code was restructured to work around a gcc-2.95.3 internal
210 * compiler error. Alter it with care.
217 sz = conf->array_size >> conf->preshift;
218 sz += 1; /* force round-up */
219 base = conf->hash_spacing >> conf->preshift;
220 round = sector_div(sz, base);
221 nb_zone = sz + (round ? 1 : 0);
223 BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *));
225 conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone,
227 if (!conf->hash_table)
231 * Here we generate the linear hash table
232 * First calculate the device offsets.
234 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;
240 table = conf->hash_table;
243 for (curr_offset = 0;
244 curr_offset < conf->array_size;
245 curr_offset += conf->hash_spacing) {
247 while (i < mddev->raid_disks-1 &&
248 curr_offset >= conf->disks[i+1].offset)
251 *table ++ = conf->disks + i;
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.
259 conf->hash_spacing++;
262 BUG_ON(table - conf->hash_table > nb_zone);
271 static int linear_run (mddev_t *mddev)
275 conf = linear_conf(mddev, mddev->raid_disks);
279 mddev->private = conf;
280 mddev->array_size = conf->array_size;
282 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;
285 mddev->queue->backing_dev_info.congested_fn = linear_congested;
286 mddev->queue->backing_dev_info.congested_data = mddev;
290 static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
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.
300 linear_conf_t *newconf;
302 if (rdev->raid_disk != mddev->raid_disks)
305 newconf = linear_conf(mddev,mddev->raid_disks+1);
310 newconf->prev = mddev_to_conf(mddev);
311 mddev->private = newconf;
313 mddev->array_size = newconf->array_size;
314 set_capacity(mddev->gendisk, mddev->array_size << 1);
318 static int linear_stop (mddev_t *mddev)
320 linear_conf_t *conf = mddev_to_conf(mddev);
322 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
324 linear_conf_t *t = conf->prev;
325 kfree(conf->hash_table);
333 static int linear_make_request (request_queue_t *q, struct bio *bio)
335 const int rw = bio_data_dir(bio);
336 mddev_t *mddev = q->queuedata;
340 if (unlikely(bio_barrier(bio))) {
341 bio_endio(bio, bio->bi_size, -EOPNOTSUPP);
345 disk_stat_inc(mddev->gendisk, ios[rw]);
346 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
348 tmp_dev = which_dev(mddev, bio->bi_sector);
349 block = bio->bi_sector >> 1;
351 if (unlikely(block >= (tmp_dev->size + tmp_dev->offset)
352 || block < tmp_dev->offset)) {
353 char b[BDEVNAME_SIZE];
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);
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
370 bp = bio_split(bio, bio_split_pool,
371 ((tmp_dev->offset + tmp_dev->size)<<1) - bio->bi_sector);
372 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);
380 bio->bi_bdev = tmp_dev->rdev->bdev;
381 bio->bi_sector = bio->bi_sector - (tmp_dev->offset << 1) + tmp_dev->rdev->data_offset;
386 static void linear_status (struct seq_file *seq, mddev_t *mddev)
392 linear_conf_t *conf = mddev_to_conf(mddev);
395 seq_printf(seq, " ");
396 for (j = 0; j < mddev->raid_disks; j++)
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));
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));
408 seq_printf(seq, "] ");
410 seq_printf(seq, "\n");
412 seq_printf(seq, " %dk rounding", mddev->chunk_size/1024);
416 static struct mdk_personality linear_personality =
419 .level = LEVEL_LINEAR,
420 .owner = THIS_MODULE,
421 .make_request = linear_make_request,
424 .status = linear_status,
425 .hot_add_disk = linear_add,
428 static int __init linear_init (void)
430 return register_md_personality (&linear_personality);
433 static void linear_exit (void)
435 unregister_md_personality (&linear_personality);
439 module_init(linear_init);
440 module_exit(linear_exit);
441 MODULE_LICENSE("GPL");
442 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
443 MODULE_ALIAS("md-linear");
444 MODULE_ALIAS("md-level--1");