ocfs2: Add metaecc for ocfs2_refcount_block.
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / md / linear.c
blob5810fa906af08f21075fe903118725093e576c82
1 /*
2 linear.c : Multiple Devices driver for Linux
3 Copyright (C) 1994-96 Marc ZYNGIER
4 <zyngier@ufr-info-p7.ibp.fr> or
5 <maz@gloups.fdn.fr>
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)
12 any later version.
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>
22 #include "md.h"
23 #include "linear.h"
26 * find which device holds a particular offset
28 static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector)
30 int lo, mid, hi;
31 linear_conf_t *conf;
33 lo = 0;
34 hi = mddev->raid_disks - 1;
35 conf = rcu_dereference(mddev->private);
38 * Binary Search
41 while (hi > lo) {
43 mid = (hi + lo) / 2;
44 if (sector < conf->disks[mid].end_sector)
45 hi = mid;
46 else
47 lo = mid + 1;
50 return conf->disks + lo;
53 /**
54 * linear_mergeable_bvec -- tell bio layer if two requests can be merged
55 * @q: request queue
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;
66 dev_info_t *dev0;
67 unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9;
68 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
70 rcu_read_lock();
71 dev0 = which_dev(mddev, sector);
72 maxsectors = dev0->end_sector - sector;
73 rcu_read_unlock();
75 if (maxsectors < bio_sectors)
76 maxsectors = 0;
77 else
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)))
85 return 1<<31;
86 return maxsectors << 9;
89 static void linear_unplug(struct request_queue *q)
91 mddev_t *mddev = q->queuedata;
92 linear_conf_t *conf;
93 int i;
95 rcu_read_lock();
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);
100 blk_unplug(r_queue);
102 rcu_read_unlock();
105 static int linear_congested(void *data, int bits)
107 mddev_t *mddev = data;
108 linear_conf_t *conf;
109 int i, ret = 0;
111 rcu_read_lock();
112 conf = rcu_dereference(mddev->private);
114 for (i = 0; i < mddev->raid_disks && !ret ; i++) {
115 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
116 ret |= bdi_congested(&q->backing_dev_info, bits);
119 rcu_read_unlock();
120 return ret;
123 static sector_t linear_size(mddev_t *mddev, sector_t sectors, int raid_disks)
125 linear_conf_t *conf;
126 sector_t array_sectors;
128 rcu_read_lock();
129 conf = rcu_dereference(mddev->private);
130 WARN_ONCE(sectors || raid_disks,
131 "%s does not support generic reshape\n", __func__);
132 array_sectors = conf->array_sectors;
133 rcu_read_unlock();
135 return array_sectors;
138 static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
140 linear_conf_t *conf;
141 mdk_rdev_t *rdev;
142 int i, cnt;
144 conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
145 GFP_KERNEL);
146 if (!conf)
147 return NULL;
149 cnt = 0;
150 conf->array_sectors = 0;
152 list_for_each_entry(rdev, &mddev->disks, same_set) {
153 int j = rdev->raid_disk;
154 dev_info_t *disk = conf->disks + j;
155 sector_t sectors;
157 if (j < 0 || j >= raid_disks || disk->rdev) {
158 printk("linear: disk numbering problem. Aborting!\n");
159 goto out;
162 disk->rdev = rdev;
163 if (mddev->chunk_sectors) {
164 sectors = rdev->sectors;
165 sector_div(sectors, mddev->chunk_sectors);
166 rdev->sectors = sectors * mddev->chunk_sectors;
169 disk_stack_limits(mddev->gendisk, rdev->bdev,
170 rdev->data_offset << 9);
171 /* as we don't honour merge_bvec_fn, we must never risk
172 * violating it, so limit ->max_sector to one PAGE, as
173 * a one page request is never in violation.
175 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
176 queue_max_sectors(mddev->queue) > (PAGE_SIZE>>9))
177 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
179 conf->array_sectors += rdev->sectors;
180 cnt++;
183 if (cnt != raid_disks) {
184 printk("linear: not enough drives present. Aborting!\n");
185 goto out;
189 * Here we calculate the device offsets.
191 conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
193 for (i = 1; i < raid_disks; i++)
194 conf->disks[i].end_sector =
195 conf->disks[i-1].end_sector +
196 conf->disks[i].rdev->sectors;
198 return conf;
200 out:
201 kfree(conf);
202 return NULL;
205 static int linear_run (mddev_t *mddev)
207 linear_conf_t *conf;
209 if (md_check_no_bitmap(mddev))
210 return -EINVAL;
211 mddev->queue->queue_lock = &mddev->queue->__queue_lock;
212 conf = linear_conf(mddev, mddev->raid_disks);
214 if (!conf)
215 return 1;
216 mddev->private = conf;
217 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
219 blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
220 mddev->queue->unplug_fn = linear_unplug;
221 mddev->queue->backing_dev_info.congested_fn = linear_congested;
222 mddev->queue->backing_dev_info.congested_data = mddev;
223 return 0;
226 static void free_conf(struct rcu_head *head)
228 linear_conf_t *conf = container_of(head, linear_conf_t, rcu);
229 kfree(conf);
232 static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
234 /* Adding a drive to a linear array allows the array to grow.
235 * It is permitted if the new drive has a matching superblock
236 * already on it, with raid_disk equal to raid_disks.
237 * It is achieved by creating a new linear_private_data structure
238 * and swapping it in in-place of the current one.
239 * The current one is never freed until the array is stopped.
240 * This avoids races.
242 linear_conf_t *newconf, *oldconf;
244 if (rdev->saved_raid_disk != mddev->raid_disks)
245 return -EINVAL;
247 rdev->raid_disk = rdev->saved_raid_disk;
249 newconf = linear_conf(mddev,mddev->raid_disks+1);
251 if (!newconf)
252 return -ENOMEM;
254 oldconf = rcu_dereference(mddev->private);
255 mddev->raid_disks++;
256 rcu_assign_pointer(mddev->private, newconf);
257 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
258 set_capacity(mddev->gendisk, mddev->array_sectors);
259 call_rcu(&oldconf->rcu, free_conf);
260 return 0;
263 static int linear_stop (mddev_t *mddev)
265 linear_conf_t *conf = mddev->private;
268 * We do not require rcu protection here since
269 * we hold reconfig_mutex for both linear_add and
270 * linear_stop, so they cannot race.
271 * We should make sure any old 'conf's are properly
272 * freed though.
274 rcu_barrier();
275 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
276 kfree(conf);
278 return 0;
281 static int linear_make_request (struct request_queue *q, struct bio *bio)
283 const int rw = bio_data_dir(bio);
284 mddev_t *mddev = q->queuedata;
285 dev_info_t *tmp_dev;
286 sector_t start_sector;
287 int cpu;
289 if (unlikely(bio_barrier(bio))) {
290 bio_endio(bio, -EOPNOTSUPP);
291 return 0;
294 cpu = part_stat_lock();
295 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
296 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
297 bio_sectors(bio));
298 part_stat_unlock();
300 rcu_read_lock();
301 tmp_dev = which_dev(mddev, bio->bi_sector);
302 start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
305 if (unlikely(bio->bi_sector >= (tmp_dev->end_sector)
306 || (bio->bi_sector < start_sector))) {
307 char b[BDEVNAME_SIZE];
309 printk("linear_make_request: Sector %llu out of bounds on "
310 "dev %s: %llu sectors, offset %llu\n",
311 (unsigned long long)bio->bi_sector,
312 bdevname(tmp_dev->rdev->bdev, b),
313 (unsigned long long)tmp_dev->rdev->sectors,
314 (unsigned long long)start_sector);
315 rcu_read_unlock();
316 bio_io_error(bio);
317 return 0;
319 if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
320 tmp_dev->end_sector)) {
321 /* This bio crosses a device boundary, so we have to
322 * split it.
324 struct bio_pair *bp;
325 sector_t end_sector = tmp_dev->end_sector;
327 rcu_read_unlock();
329 bp = bio_split(bio, end_sector - bio->bi_sector);
331 if (linear_make_request(q, &bp->bio1))
332 generic_make_request(&bp->bio1);
333 if (linear_make_request(q, &bp->bio2))
334 generic_make_request(&bp->bio2);
335 bio_pair_release(bp);
336 return 0;
339 bio->bi_bdev = tmp_dev->rdev->bdev;
340 bio->bi_sector = bio->bi_sector - start_sector
341 + tmp_dev->rdev->data_offset;
342 rcu_read_unlock();
344 return 1;
347 static void linear_status (struct seq_file *seq, mddev_t *mddev)
350 seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
354 static struct mdk_personality linear_personality =
356 .name = "linear",
357 .level = LEVEL_LINEAR,
358 .owner = THIS_MODULE,
359 .make_request = linear_make_request,
360 .run = linear_run,
361 .stop = linear_stop,
362 .status = linear_status,
363 .hot_add_disk = linear_add,
364 .size = linear_size,
367 static int __init linear_init (void)
369 return register_md_personality (&linear_personality);
372 static void linear_exit (void)
374 unregister_md_personality (&linear_personality);
378 module_init(linear_init);
379 module_exit(linear_exit);
380 MODULE_LICENSE("GPL");
381 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
382 MODULE_ALIAS("md-linear");
383 MODULE_ALIAS("md-level--1");