linear.c revision cc371e66e340f35eed8dc4651c7c18e754c7fb26 
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	block >>= conf->preshift;
42	(void)sector_div(block, conf->hash_spacing);
43	hash = conf->hash_table[block];
44
45	while ((sector>>1) >= (hash->size + hash->offset))
46		hash++;
47	return hash;
48}
49
50/**
51 *	linear_mergeable_bvec -- tell bio layer if two requests can be merged
52 *	@q: request queue
53 *	@bvm: properties of new bio
54 *	@biovec: the request that could be merged to it.
55 *
56 *	Return amount of bytes we can take at this offset
57 */
58static int linear_mergeable_bvec(struct request_queue *q,
59				 struct bvec_merge_data *bvm,
60				 struct bio_vec *biovec)
61{
62	mddev_t *mddev = q->queuedata;
63	dev_info_t *dev0;
64	unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9;
65	sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
66
67	dev0 = which_dev(mddev, sector);
68	maxsectors = (dev0->size << 1) - (sector - (dev0->offset<<1));
69
70	if (maxsectors < bio_sectors)
71		maxsectors = 0;
72	else
73		maxsectors -= bio_sectors;
74
75	if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
76		return biovec->bv_len;
77	/* The bytes available at this offset could be really big,
78	 * so we cap at 2^31 to avoid overflow */
79	if (maxsectors > (1 << (31-9)))
80		return 1<<31;
81	return maxsectors << 9;
82}
83
84static void linear_unplug(struct request_queue *q)
85{
86	mddev_t *mddev = q->queuedata;
87	linear_conf_t *conf = mddev_to_conf(mddev);
88	int i;
89
90	for (i=0; i < mddev->raid_disks; i++) {
91		struct request_queue *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
92		blk_unplug(r_queue);
93	}
94}
95
96static int linear_congested(void *data, int bits)
97{
98	mddev_t *mddev = data;
99	linear_conf_t *conf = mddev_to_conf(mddev);
100	int i, ret = 0;
101
102	for (i = 0; i < mddev->raid_disks && !ret ; i++) {
103		struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
104		ret |= bdi_congested(&q->backing_dev_info, bits);
105	}
106	return ret;
107}
108
109static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
110{
111	linear_conf_t *conf;
112	dev_info_t **table;
113	mdk_rdev_t *rdev;
114	int i, nb_zone, cnt;
115	sector_t min_spacing;
116	sector_t curr_offset;
117	struct list_head *tmp;
118
119	conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
120			GFP_KERNEL);
121	if (!conf)
122		return NULL;
123
124	cnt = 0;
125	conf->array_size = 0;
126
127	rdev_for_each(rdev, tmp, mddev) {
128		int j = rdev->raid_disk;
129		dev_info_t *disk = conf->disks + j;
130
131		if (j < 0 || j > raid_disks || disk->rdev) {
132			printk("linear: disk numbering problem. Aborting!\n");
133			goto out;
134		}
135
136		disk->rdev = rdev;
137
138		blk_queue_stack_limits(mddev->queue,
139				       rdev->bdev->bd_disk->queue);
140		/* as we don't honour merge_bvec_fn, we must never risk
141		 * violating it, so limit ->max_sector to one PAGE, as
142		 * a one page request is never in violation.
143		 */
144		if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
145		    mddev->queue->max_sectors > (PAGE_SIZE>>9))
146			blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
147
148		disk->size = rdev->size;
149		conf->array_size += rdev->size;
150
151		cnt++;
152	}
153	if (cnt != raid_disks) {
154		printk("linear: not enough drives present. Aborting!\n");
155		goto out;
156	}
157
158	min_spacing = conf->array_size;
159	sector_div(min_spacing, PAGE_SIZE/sizeof(struct dev_info *));
160
161	/* min_spacing is the minimum spacing that will fit the hash
162	 * table in one PAGE.  This may be much smaller than needed.
163	 * We find the smallest non-terminal set of consecutive devices
164	 * that is larger than min_spacing as use the size of that as
165	 * the actual spacing
166	 */
167	conf->hash_spacing = conf->array_size;
168	for (i=0; i < cnt-1 ; i++) {
169		sector_t sz = 0;
170		int j;
171		for (j = i; j < cnt - 1 && sz < min_spacing; j++)
172			sz += conf->disks[j].size;
173		if (sz >= min_spacing && sz < conf->hash_spacing)
174			conf->hash_spacing = sz;
175	}
176
177	/* hash_spacing may be too large for sector_div to work with,
178	 * so we might need to pre-shift
179	 */
180	conf->preshift = 0;
181	if (sizeof(sector_t) > sizeof(u32)) {
182		sector_t space = conf->hash_spacing;
183		while (space > (sector_t)(~(u32)0)) {
184			space >>= 1;
185			conf->preshift++;
186		}
187	}
188	/*
189	 * This code was restructured to work around a gcc-2.95.3 internal
190	 * compiler error.  Alter it with care.
191	 */
192	{
193		sector_t sz;
194		unsigned round;
195		unsigned long base;
196
197		sz = conf->array_size >> conf->preshift;
198		sz += 1; /* force round-up */
199		base = conf->hash_spacing >> conf->preshift;
200		round = sector_div(sz, base);
201		nb_zone = sz + (round ? 1 : 0);
202	}
203	BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *));
204
205	conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone,
206					GFP_KERNEL);
207	if (!conf->hash_table)
208		goto out;
209
210	/*
211	 * Here we generate the linear hash table
212	 * First calculate the device offsets.
213	 */
214	conf->disks[0].offset = 0;
215	for (i = 1; i < raid_disks; i++)
216		conf->disks[i].offset =
217			conf->disks[i-1].offset +
218			conf->disks[i-1].size;
219
220	table = conf->hash_table;
221	curr_offset = 0;
222	i = 0;
223	for (curr_offset = 0;
224	     curr_offset < conf->array_size;
225	     curr_offset += conf->hash_spacing) {
226
227		while (i < raid_disks-1 &&
228		       curr_offset >= conf->disks[i+1].offset)
229			i++;
230
231		*table ++ = conf->disks + i;
232	}
233
234	if (conf->preshift) {
235		conf->hash_spacing >>= conf->preshift;
236		/* round hash_spacing up so that when we divide by it,
237		 * we err on the side of "too-low", which is safest.
238		 */
239		conf->hash_spacing++;
240	}
241
242	BUG_ON(table - conf->hash_table > nb_zone);
243
244	return conf;
245
246out:
247	kfree(conf);
248	return NULL;
249}
250
251static int linear_run (mddev_t *mddev)
252{
253	linear_conf_t *conf;
254
255	mddev->queue->queue_lock = &mddev->queue->__queue_lock;
256	conf = linear_conf(mddev, mddev->raid_disks);
257
258	if (!conf)
259		return 1;
260	mddev->private = conf;
261	mddev->array_size = conf->array_size;
262
263	blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
264	mddev->queue->unplug_fn = linear_unplug;
265	mddev->queue->backing_dev_info.congested_fn = linear_congested;
266	mddev->queue->backing_dev_info.congested_data = mddev;
267	return 0;
268}
269
270static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
271{
272	/* Adding a drive to a linear array allows the array to grow.
273	 * It is permitted if the new drive has a matching superblock
274	 * already on it, with raid_disk equal to raid_disks.
275	 * It is achieved by creating a new linear_private_data structure
276	 * and swapping it in in-place of the current one.
277	 * The current one is never freed until the array is stopped.
278	 * This avoids races.
279	 */
280	linear_conf_t *newconf;
281
282	if (rdev->saved_raid_disk != mddev->raid_disks)
283		return -EINVAL;
284
285	rdev->raid_disk = rdev->saved_raid_disk;
286
287	newconf = linear_conf(mddev,mddev->raid_disks+1);
288
289	if (!newconf)
290		return -ENOMEM;
291
292	newconf->prev = mddev_to_conf(mddev);
293	mddev->private = newconf;
294	mddev->raid_disks++;
295	mddev->array_size = newconf->array_size;
296	set_capacity(mddev->gendisk, mddev->array_size << 1);
297	return 0;
298}
299
300static int linear_stop (mddev_t *mddev)
301{
302	linear_conf_t *conf = mddev_to_conf(mddev);
303
304	blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
305	do {
306		linear_conf_t *t = conf->prev;
307		kfree(conf->hash_table);
308		kfree(conf);
309		conf = t;
310	} while (conf);
311
312	return 0;
313}
314
315static int linear_make_request (struct request_queue *q, struct bio *bio)
316{
317	const int rw = bio_data_dir(bio);
318	mddev_t *mddev = q->queuedata;
319	dev_info_t *tmp_dev;
320	sector_t block;
321
322	if (unlikely(bio_barrier(bio))) {
323		bio_endio(bio, -EOPNOTSUPP);
324		return 0;
325	}
326
327	disk_stat_inc(mddev->gendisk, ios[rw]);
328	disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
329
330	tmp_dev = which_dev(mddev, bio->bi_sector);
331	block = bio->bi_sector >> 1;
332
333	if (unlikely(block >= (tmp_dev->size + tmp_dev->offset)
334		     || block < tmp_dev->offset)) {
335		char b[BDEVNAME_SIZE];
336
337		printk("linear_make_request: Block %llu out of bounds on "
338			"dev %s size %llu offset %llu\n",
339			(unsigned long long)block,
340			bdevname(tmp_dev->rdev->bdev, b),
341			(unsigned long long)tmp_dev->size,
342		        (unsigned long long)tmp_dev->offset);
343		bio_io_error(bio);
344		return 0;
345	}
346	if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
347		     (tmp_dev->offset + tmp_dev->size)<<1)) {
348		/* This bio crosses a device boundary, so we have to
349		 * split it.
350		 */
351		struct bio_pair *bp;
352		bp = bio_split(bio, bio_split_pool,
353			       ((tmp_dev->offset + tmp_dev->size)<<1) - bio->bi_sector);
354		if (linear_make_request(q, &bp->bio1))
355			generic_make_request(&bp->bio1);
356		if (linear_make_request(q, &bp->bio2))
357			generic_make_request(&bp->bio2);
358		bio_pair_release(bp);
359		return 0;
360	}
361
362	bio->bi_bdev = tmp_dev->rdev->bdev;
363	bio->bi_sector = bio->bi_sector - (tmp_dev->offset << 1) + tmp_dev->rdev->data_offset;
364
365	return 1;
366}
367
368static void linear_status (struct seq_file *seq, mddev_t *mddev)
369{
370
371#undef MD_DEBUG
372#ifdef MD_DEBUG
373	int j;
374	linear_conf_t *conf = mddev_to_conf(mddev);
375	sector_t s = 0;
376
377	seq_printf(seq, "      ");
378	for (j = 0; j < mddev->raid_disks; j++)
379	{
380		char b[BDEVNAME_SIZE];
381		s += conf->smallest_size;
382		seq_printf(seq, "[%s",
383			   bdevname(conf->hash_table[j][0].rdev->bdev,b));
384
385		while (s > conf->hash_table[j][0].offset +
386		           conf->hash_table[j][0].size)
387			seq_printf(seq, "/%s] ",
388				   bdevname(conf->hash_table[j][1].rdev->bdev,b));
389		else
390			seq_printf(seq, "] ");
391	}
392	seq_printf(seq, "\n");
393#endif
394	seq_printf(seq, " %dk rounding", mddev->chunk_size/1024);
395}
396
397
398static struct mdk_personality linear_personality =
399{
400	.name		= "linear",
401	.level		= LEVEL_LINEAR,
402	.owner		= THIS_MODULE,
403	.make_request	= linear_make_request,
404	.run		= linear_run,
405	.stop		= linear_stop,
406	.status		= linear_status,
407	.hot_add_disk	= linear_add,
408};
409
410static int __init linear_init (void)
411{
412	return register_md_personality (&linear_personality);
413}
414
415static void linear_exit (void)
416{
417	unregister_md_personality (&linear_personality);
418}
419
420
421module_init(linear_init);
422module_exit(linear_exit);
423MODULE_LICENSE("GPL");
424MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
425MODULE_ALIAS("md-linear");
426MODULE_ALIAS("md-level--1");
427