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)
34 hi = mddev->raid_disks - 1;
35 conf = rcu_dereference(mddev->private);
44 if (sector < conf->disks[mid].end_sector)
50 return conf->disks + lo;
54 * linear_mergeable_bvec -- tell bio layer if two requests can be merged
56 * @bvm: properties of new bio
57 * @biovec: the request that could be merged to it.
59 * Return amount of bytes we can take at this offset
61 static int linear_mergeable_bvec(struct request_queue *q,
62 struct bvec_merge_data *bvm,
63 struct bio_vec *biovec)
65 mddev_t *mddev = q->queuedata;
67 unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9;
68 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
71 dev0 = which_dev(mddev, sector);
72 maxsectors = dev0->end_sector - sector;
75 if (maxsectors < bio_sectors)
78 maxsectors -= bio_sectors;
80 if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
81 return biovec->bv_len;
82 /* The bytes available at this offset could be really big,
83 * so we cap at 2^31 to avoid overflow */
84 if (maxsectors > (1 << (31-9)))
86 return maxsectors << 9;
89 static void linear_unplug(struct request_queue *q)
91 mddev_t *mddev = q->queuedata;
96 conf = rcu_dereference(mddev->private);
98 for (i=0; i < mddev->raid_disks; i++) {
99 struct request_queue *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
105 static int linear_congested(void *data, int bits)
107 mddev_t *mddev = data;
111 if (mddev_congested(mddev, bits))
115 conf = rcu_dereference(mddev->private);
117 for (i = 0; i < mddev->raid_disks && !ret ; i++) {
118 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
119 ret |= bdi_congested(&q->backing_dev_info, bits);
126 static sector_t linear_size(mddev_t *mddev, sector_t sectors, int raid_disks)
129 sector_t array_sectors;
132 conf = rcu_dereference(mddev->private);
133 WARN_ONCE(sectors || raid_disks,
134 "%s does not support generic reshape\n", __func__);
135 array_sectors = conf->array_sectors;
138 return array_sectors;
141 static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
147 conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
153 conf->array_sectors = 0;
155 list_for_each_entry(rdev, &mddev->disks, same_set) {
156 int j = rdev->raid_disk;
157 dev_info_t *disk = conf->disks + j;
160 if (j < 0 || j >= raid_disks || disk->rdev) {
161 printk("linear: disk numbering problem. Aborting!\n");
166 if (mddev->chunk_sectors) {
167 sectors = rdev->sectors;
168 sector_div(sectors, mddev->chunk_sectors);
169 rdev->sectors = sectors * mddev->chunk_sectors;
172 disk_stack_limits(mddev->gendisk, rdev->bdev,
173 rdev->data_offset << 9);
174 /* as we don't honour merge_bvec_fn, we must never risk
175 * violating it, so limit max_phys_segments to 1 lying within
178 if (rdev->bdev->bd_disk->queue->merge_bvec_fn) {
179 blk_queue_max_phys_segments(mddev->queue, 1);
180 blk_queue_segment_boundary(mddev->queue,
181 PAGE_CACHE_SIZE - 1);
184 conf->array_sectors += rdev->sectors;
188 if (cnt != raid_disks) {
189 printk("linear: not enough drives present. Aborting!\n");
194 * Here we calculate the device offsets.
196 conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
198 for (i = 1; i < raid_disks; i++)
199 conf->disks[i].end_sector =
200 conf->disks[i-1].end_sector +
201 conf->disks[i].rdev->sectors;
210 static int linear_run (mddev_t *mddev)
214 if (md_check_no_bitmap(mddev))
216 mddev->queue->queue_lock = &mddev->queue->__queue_lock;
217 conf = linear_conf(mddev, mddev->raid_disks);
221 mddev->private = conf;
222 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
224 blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
225 mddev->queue->unplug_fn = linear_unplug;
226 mddev->queue->backing_dev_info.congested_fn = linear_congested;
227 mddev->queue->backing_dev_info.congested_data = mddev;
228 md_integrity_register(mddev);
232 static void free_conf(struct rcu_head *head)
234 linear_conf_t *conf = container_of(head, linear_conf_t, rcu);
238 static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
240 /* Adding a drive to a linear array allows the array to grow.
241 * It is permitted if the new drive has a matching superblock
242 * already on it, with raid_disk equal to raid_disks.
243 * It is achieved by creating a new linear_private_data structure
244 * and swapping it in in-place of the current one.
245 * The current one is never freed until the array is stopped.
248 linear_conf_t *newconf, *oldconf;
250 if (rdev->saved_raid_disk != mddev->raid_disks)
253 rdev->raid_disk = rdev->saved_raid_disk;
255 newconf = linear_conf(mddev,mddev->raid_disks+1);
260 oldconf = rcu_dereference(mddev->private);
262 rcu_assign_pointer(mddev->private, newconf);
263 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
264 set_capacity(mddev->gendisk, mddev->array_sectors);
265 revalidate_disk(mddev->gendisk);
266 call_rcu(&oldconf->rcu, free_conf);
270 static int linear_stop (mddev_t *mddev)
272 linear_conf_t *conf = mddev->private;
275 * We do not require rcu protection here since
276 * we hold reconfig_mutex for both linear_add and
277 * linear_stop, so they cannot race.
278 * We should make sure any old 'conf's are properly
282 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
288 static int linear_make_request (struct request_queue *q, struct bio *bio)
290 const int rw = bio_data_dir(bio);
291 mddev_t *mddev = q->queuedata;
293 sector_t start_sector;
296 if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) {
297 bio_endio(bio, -EOPNOTSUPP);
301 cpu = part_stat_lock();
302 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
303 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
308 tmp_dev = which_dev(mddev, bio->bi_sector);
309 start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
312 if (unlikely(bio->bi_sector >= (tmp_dev->end_sector)
313 || (bio->bi_sector < start_sector))) {
314 char b[BDEVNAME_SIZE];
316 printk("linear_make_request: Sector %llu out of bounds on "
317 "dev %s: %llu sectors, offset %llu\n",
318 (unsigned long long)bio->bi_sector,
319 bdevname(tmp_dev->rdev->bdev, b),
320 (unsigned long long)tmp_dev->rdev->sectors,
321 (unsigned long long)start_sector);
326 if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
327 tmp_dev->end_sector)) {
328 /* This bio crosses a device boundary, so we have to
332 sector_t end_sector = tmp_dev->end_sector;
336 bp = bio_split(bio, end_sector - bio->bi_sector);
338 if (linear_make_request(q, &bp->bio1))
339 generic_make_request(&bp->bio1);
340 if (linear_make_request(q, &bp->bio2))
341 generic_make_request(&bp->bio2);
342 bio_pair_release(bp);
346 bio->bi_bdev = tmp_dev->rdev->bdev;
347 bio->bi_sector = bio->bi_sector - start_sector
348 + tmp_dev->rdev->data_offset;
354 static void linear_status (struct seq_file *seq, mddev_t *mddev)
357 seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
361 static struct mdk_personality linear_personality =
364 .level = LEVEL_LINEAR,
365 .owner = THIS_MODULE,
366 .make_request = linear_make_request,
369 .status = linear_status,
370 .hot_add_disk = linear_add,
374 static int __init linear_init (void)
376 return register_md_personality (&linear_personality);
379 static void linear_exit (void)
381 unregister_md_personality (&linear_personality);
385 module_init(linear_init);
386 module_exit(linear_exit);
387 MODULE_LICENSE("GPL");
388 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
389 MODULE_ALIAS("md-linear");
390 MODULE_ALIAS("md-level--1");