linear.c revision e5dcdd80a60627371f40797426273048630dc8ca 
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>
6
7   Linear mode management functions.
8
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.
13
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.
17*/
18
19#include <linux/module.h>
20
21#include <linux/raid/md.h>
22#include <linux/slab.h>
23#include <linux/raid/linear.h>
24
25#define MAJOR_NR MD_MAJOR
26#define MD_DRIVER
27#define MD_PERSONALITY
28
29/*
30 * find which device holds a particular offset
31 */
32static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector)
33{
34	dev_info_t *hash;
35	linear_conf_t *conf = mddev_to_conf(mddev);
36	sector_t block = sector >> 1;
37
38	/*
39	 * sector_div(a,b) returns the remainer and sets a to a/b
40	 */
41	(void)sector_div(block, conf->smallest->size);
42	hash = conf->hash_table[block];
43
44	while ((sector>>1) >= (hash->size + hash->offset))
45		hash++;
46	return hash;
47}
48
49/**
50 *	linear_mergeable_bvec -- tell bio layer if a two requests can be merged
51 *	@q: request queue
52 *	@bio: the buffer head that's been built up so far
53 *	@biovec: the request that could be merged to it.
54 *
55 *	Return amount of bytes we can take at this offset
56 */
57static int linear_mergeable_bvec(request_queue_t *q, struct bio *bio, struct bio_vec *biovec)
58{
59	mddev_t *mddev = q->queuedata;
60	dev_info_t *dev0;
61	unsigned long maxsectors, bio_sectors = bio->bi_size >> 9;
62	sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
63
64	dev0 = which_dev(mddev, sector);
65	maxsectors = (dev0->size << 1) - (sector - (dev0->offset<<1));
66
67	if (maxsectors < bio_sectors)
68		maxsectors = 0;
69	else
70		maxsectors -= bio_sectors;
71
72	if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
73		return biovec->bv_len;
74	/* The bytes available at this offset could be really big,
75	 * so we cap at 2^31 to avoid overflow */
76	if (maxsectors > (1 << (31-9)))
77		return 1<<31;
78	return maxsectors << 9;
79}
80
81static void linear_unplug(request_queue_t *q)
82{
83	mddev_t *mddev = q->queuedata;
84	linear_conf_t *conf = mddev_to_conf(mddev);
85	int i;
86
87	for (i=0; i < mddev->raid_disks; i++) {
88		request_queue_t *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
89		if (r_queue->unplug_fn)
90			r_queue->unplug_fn(r_queue);
91	}
92}
93
94static int linear_issue_flush(request_queue_t *q, struct gendisk *disk,
95			      sector_t *error_sector)
96{
97	mddev_t *mddev = q->queuedata;
98	linear_conf_t *conf = mddev_to_conf(mddev);
99	int i, ret = 0;
100
101	for (i=0; i < mddev->raid_disks && ret == 0; i++) {
102		struct block_device *bdev = conf->disks[i].rdev->bdev;
103		request_queue_t *r_queue = bdev_get_queue(bdev);
104
105		if (!r_queue->issue_flush_fn)
106			ret = -EOPNOTSUPP;
107		else
108			ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, error_sector);
109	}
110	return ret;
111}
112
113static int linear_run (mddev_t *mddev)
114{
115	linear_conf_t *conf;
116	dev_info_t **table;
117	mdk_rdev_t *rdev;
118	int i, nb_zone, cnt;
119	sector_t start;
120	sector_t curr_offset;
121	struct list_head *tmp;
122
123	conf = kmalloc (sizeof (*conf) + mddev->raid_disks*sizeof(dev_info_t),
124			GFP_KERNEL);
125	if (!conf)
126		goto out;
127	memset(conf, 0, sizeof(*conf) + mddev->raid_disks*sizeof(dev_info_t));
128	mddev->private = conf;
129
130	/*
131	 * Find the smallest device.
132	 */
133
134	conf->smallest = NULL;
135	cnt = 0;
136	mddev->array_size = 0;
137
138	ITERATE_RDEV(mddev,rdev,tmp) {
139		int j = rdev->raid_disk;
140		dev_info_t *disk = conf->disks + j;
141
142		if (j < 0 || j > mddev->raid_disks || disk->rdev) {
143			printk("linear: disk numbering problem. Aborting!\n");
144			goto out;
145		}
146
147		disk->rdev = rdev;
148
149		blk_queue_stack_limits(mddev->queue,
150				       rdev->bdev->bd_disk->queue);
151		/* as we don't honour merge_bvec_fn, we must never risk
152		 * violating it, so limit ->max_sector to one PAGE, as
153		 * a one page request is never in violation.
154		 */
155		if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
156		    mddev->queue->max_sectors > (PAGE_SIZE>>9))
157			blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
158
159		disk->size = rdev->size;
160		mddev->array_size += rdev->size;
161
162		if (!conf->smallest || (disk->size < conf->smallest->size))
163			conf->smallest = disk;
164		cnt++;
165	}
166	if (cnt != mddev->raid_disks) {
167		printk("linear: not enough drives present. Aborting!\n");
168		goto out;
169	}
170
171	/*
172	 * This code was restructured to work around a gcc-2.95.3 internal
173	 * compiler error.  Alter it with care.
174	 */
175	{
176		sector_t sz;
177		unsigned round;
178		unsigned long base;
179
180		sz = mddev->array_size;
181		base = conf->smallest->size;
182		round = sector_div(sz, base);
183		nb_zone = conf->nr_zones = sz + (round ? 1 : 0);
184	}
185
186	conf->hash_table = kmalloc (sizeof (dev_info_t*) * nb_zone,
187					GFP_KERNEL);
188	if (!conf->hash_table)
189		goto out;
190
191	/*
192	 * Here we generate the linear hash table
193	 */
194	table = conf->hash_table;
195	start = 0;
196	curr_offset = 0;
197	for (i = 0; i < cnt; i++) {
198		dev_info_t *disk = conf->disks + i;
199
200		disk->offset = curr_offset;
201		curr_offset += disk->size;
202
203		/* 'curr_offset' is the end of this disk
204		 * 'start' is the start of table
205		 */
206		while (start < curr_offset) {
207			*table++ = disk;
208			start += conf->smallest->size;
209		}
210	}
211	if (table-conf->hash_table != nb_zone)
212		BUG();
213
214	blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
215	mddev->queue->unplug_fn = linear_unplug;
216	mddev->queue->issue_flush_fn = linear_issue_flush;
217	return 0;
218
219out:
220	kfree(conf);
221	return 1;
222}
223
224static int linear_stop (mddev_t *mddev)
225{
226	linear_conf_t *conf = mddev_to_conf(mddev);
227
228	blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
229	kfree(conf->hash_table);
230	kfree(conf);
231
232	return 0;
233}
234
235static int linear_make_request (request_queue_t *q, struct bio *bio)
236{
237	mddev_t *mddev = q->queuedata;
238	dev_info_t *tmp_dev;
239	sector_t block;
240
241	if (unlikely(bio_barrier(bio))) {
242		bio_endio(bio, bio->bi_size, -EOPNOTSUPP);
243		return 0;
244	}
245
246	if (bio_data_dir(bio)==WRITE) {
247		disk_stat_inc(mddev->gendisk, writes);
248		disk_stat_add(mddev->gendisk, write_sectors, bio_sectors(bio));
249	} else {
250		disk_stat_inc(mddev->gendisk, reads);
251		disk_stat_add(mddev->gendisk, read_sectors, bio_sectors(bio));
252	}
253
254	tmp_dev = which_dev(mddev, bio->bi_sector);
255	block = bio->bi_sector >> 1;
256
257	if (unlikely(block >= (tmp_dev->size + tmp_dev->offset)
258		     || block < tmp_dev->offset)) {
259		char b[BDEVNAME_SIZE];
260
261		printk("linear_make_request: Block %llu out of bounds on "
262			"dev %s size %llu offset %llu\n",
263			(unsigned long long)block,
264			bdevname(tmp_dev->rdev->bdev, b),
265			(unsigned long long)tmp_dev->size,
266		        (unsigned long long)tmp_dev->offset);
267		bio_io_error(bio, bio->bi_size);
268		return 0;
269	}
270	if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
271		     (tmp_dev->offset + tmp_dev->size)<<1)) {
272		/* This bio crosses a device boundary, so we have to
273		 * split it.
274		 */
275		struct bio_pair *bp;
276		bp = bio_split(bio, bio_split_pool,
277			       ((tmp_dev->offset + tmp_dev->size)<<1) - bio->bi_sector);
278		if (linear_make_request(q, &bp->bio1))
279			generic_make_request(&bp->bio1);
280		if (linear_make_request(q, &bp->bio2))
281			generic_make_request(&bp->bio2);
282		bio_pair_release(bp);
283		return 0;
284	}
285
286	bio->bi_bdev = tmp_dev->rdev->bdev;
287	bio->bi_sector = bio->bi_sector - (tmp_dev->offset << 1) + tmp_dev->rdev->data_offset;
288
289	return 1;
290}
291
292static void linear_status (struct seq_file *seq, mddev_t *mddev)
293{
294
295#undef MD_DEBUG
296#ifdef MD_DEBUG
297	int j;
298	linear_conf_t *conf = mddev_to_conf(mddev);
299	sector_t s = 0;
300
301	seq_printf(seq, "      ");
302	for (j = 0; j < conf->nr_zones; j++)
303	{
304		char b[BDEVNAME_SIZE];
305		s += conf->smallest_size;
306		seq_printf(seq, "[%s",
307			   bdevname(conf->hash_table[j][0].rdev->bdev,b));
308
309		while (s > conf->hash_table[j][0].offset +
310		           conf->hash_table[j][0].size)
311			seq_printf(seq, "/%s] ",
312				   bdevname(conf->hash_table[j][1].rdev->bdev,b));
313		else
314			seq_printf(seq, "] ");
315	}
316	seq_printf(seq, "\n");
317#endif
318	seq_printf(seq, " %dk rounding", mddev->chunk_size/1024);
319}
320
321
322static mdk_personality_t linear_personality=
323{
324	.name		= "linear",
325	.owner		= THIS_MODULE,
326	.make_request	= linear_make_request,
327	.run		= linear_run,
328	.stop		= linear_stop,
329	.status		= linear_status,
330};
331
332static int __init linear_init (void)
333{
334	return register_md_personality (LINEAR, &linear_personality);
335}
336
337static void linear_exit (void)
338{
339	unregister_md_personality (LINEAR);
340}
341
342
343module_init(linear_init);
344module_exit(linear_exit);
345MODULE_LICENSE("GPL");
346MODULE_ALIAS("md-personality-1"); /* LINEAR */
347