/drivers/md/linear.c

  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/blkdev.h>
 20#include <linux/raid/md_u.h>
 21#include <linux/seq_file.h>
 22#include <linux/slab.h>
 23#include "md.h"
 24#include "linear.h"
 25
 26/*
 27 * find which device holds a particular offset 
 28 */
 29static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector)
 30{
 31	int lo, mid, hi;
 32	linear_conf_t *conf;
 33
 34	lo = 0;
 35	hi = mddev->raid_disks - 1;
 36	conf = rcu_dereference(mddev->private);
 37
 38	/*
 39	 * Binary Search
 40	 */
 41
 42	while (hi > lo) {
 43
 44		mid = (hi + lo) / 2;
 45		if (sector < conf->disks[mid].end_sector)
 46			hi = mid;
 47		else
 48			lo = mid + 1;
 49	}
 50
 51	return conf->disks + lo;
 52}
 53
 54/**
 55 *	linear_mergeable_bvec -- tell bio layer if two requests can be merged
 56 *	@q: request queue
 57 *	@bvm: properties of new bio
 58 *	@biovec: the request that could be merged to it.
 59 *
 60 *	Return amount of bytes we can take at this offset
 61 */
 62static int linear_mergeable_bvec(struct request_queue *q,
 63				 struct bvec_merge_data *bvm,
 64				 struct bio_vec *biovec)
 65{
 66	mddev_t *mddev = q->queuedata;
 67	dev_info_t *dev0;
 68	unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9;
 69	sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
 70
 71	rcu_read_lock();
 72	dev0 = which_dev(mddev, sector);
 73	maxsectors = dev0->end_sector - sector;
 74	rcu_read_unlock();
 75
 76	if (maxsectors < bio_sectors)
 77		maxsectors = 0;
 78	else
 79		maxsectors -= bio_sectors;
 80
 81	if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
 82		return biovec->bv_len;
 83	/* The bytes available at this offset could be really big,
 84	 * so we cap at 2^31 to avoid overflow */
 85	if (maxsectors > (1 << (31-9)))
 86		return 1<<31;
 87	return maxsectors << 9;
 88}
 89
 90static int linear_congested(void *data, int bits)
 91{
 92	mddev_t *mddev = data;
 93	linear_conf_t *conf;
 94	int i, ret = 0;
 95
 96	if (mddev_congested(mddev, bits))
 97		return 1;
 98
 99	rcu_read_lock();
100	conf = rcu_dereference(mddev->private);
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
107	rcu_read_unlock();
108	return ret;
109}
110
111static sector_t linear_size(mddev_t *mddev, sector_t sectors, int raid_disks)
112{
113	linear_conf_t *conf;
114	sector_t array_sectors;
115
116	rcu_read_lock();
117	conf = rcu_dereference(mddev->private);
118	WARN_ONCE(sectors || raid_disks,
119		  "%s does not support generic reshape\n", __func__);
120	array_sectors = conf->array_sectors;
121	rcu_read_unlock();
122
123	return array_sectors;
124}
125
126static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
127{
128	linear_conf_t *conf;
129	mdk_rdev_t *rdev;
130	int i, cnt;
131
132	conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
133			GFP_KERNEL);
134	if (!conf)
135		return NULL;
136
137	cnt = 0;
138	conf->array_sectors = 0;
139
140	list_for_each_entry(rdev, &mddev->disks, same_set) {
141		int j = rdev->raid_disk;
142		dev_info_t *disk = conf->disks + j;
143		sector_t sectors;
144
145		if (j < 0 || j >= raid_disks || disk->rdev) {
146			printk(KERN_ERR "md/linear:%s: disk numbering problem. Aborting!\n",
147			       mdname(mddev));
148			goto out;
149		}
150
151		disk->rdev = rdev;
152		if (mddev->chunk_sectors) {
153			sectors = rdev->sectors;
154			sector_div(sectors, mddev->chunk_sectors);
155			rdev->sectors = sectors * mddev->chunk_sectors;
156		}
157
158		disk_stack_limits(mddev->gendisk, rdev->bdev,
159				  rdev->data_offset << 9);
160		/* as we don't honour merge_bvec_fn, we must never risk
161		 * violating it, so limit max_segments to 1 lying within
162		 * a single page.
163		 */
164		if (rdev->bdev->bd_disk->queue->merge_bvec_fn) {
165			blk_queue_max_segments(mddev->queue, 1);
166			blk_queue_segment_boundary(mddev->queue,
167						   PAGE_CACHE_SIZE - 1);
168		}
169
170		conf->array_sectors += rdev->sectors;
171		cnt++;
172
173	}
174	if (cnt != raid_disks) {
175		printk(KERN_ERR "md/linear:%s: not enough drives present. Aborting!\n",
176		       mdname(mddev));
177		goto out;
178	}
179
180	/*
181	 * Here we calculate the device offsets.
182	 */
183	conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
184
185	for (i = 1; i < raid_disks; i++)
186		conf->disks[i].end_sector =
187			conf->disks[i-1].end_sector +
188			conf->disks[i].rdev->sectors;
189
190	return conf;
191
192out:
193	kfree(conf);
194	return NULL;
195}
196
197static int linear_run (mddev_t *mddev)
198{
199	linear_conf_t *conf;
200
201	if (md_check_no_bitmap(mddev))
202		return -EINVAL;
203	conf = linear_conf(mddev, mddev->raid_disks);
204
205	if (!conf)
206		return 1;
207	mddev->private = conf;
208	md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
209
210	blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
211	mddev->queue->backing_dev_info.congested_fn = linear_congested;
212	mddev->queue->backing_dev_info.congested_data = mddev;
213	return md_integrity_register(mddev);
214}
215
216static void free_conf(struct rcu_head *head)
217{
218	linear_conf_t *conf = container_of(head, linear_conf_t, rcu);
219	kfree(conf);
220}
221
222static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
223{
224	/* Adding a drive to a linear array allows the array to grow.
225	 * It is permitted if the new drive has a matching superblock
226	 * already on it, with raid_disk equal to raid_disks.
227	 * It is achieved by creating a new linear_private_data structure
228	 * and swapping it in in-place of the current one.
229	 * The current one is never freed until the array is stopped.
230	 * This avoids races.
231	 */
232	linear_conf_t *newconf, *oldconf;
233
234	if (rdev->saved_raid_disk != mddev->raid_disks)
235		return -EINVAL;
236
237	rdev->raid_disk = rdev->saved_raid_disk;
238
239	newconf = linear_conf(mddev,mddev->raid_disks+1);
240
241	if (!newconf)
242		return -ENOMEM;
243
244	oldconf = rcu_dereference(mddev->private);
245	mddev->raid_disks++;
246	rcu_assign_pointer(mddev->private, newconf);
247	md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
248	set_capacity(mddev->gendisk, mddev->array_sectors);
249	revalidate_disk(mddev->gendisk);
250	call_rcu(&oldconf->rcu, free_conf);
251	return 0;
252}
253
254static int linear_stop (mddev_t *mddev)
255{
256	linear_conf_t *conf = mddev->private;
257
258	/*
259	 * We do not require rcu protection here since
260	 * we hold reconfig_mutex for both linear_add and
261	 * linear_stop, so they cannot race.
262	 * We should make sure any old 'conf's are properly
263	 * freed though.
264	 */
265	rcu_barrier();
266	blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
267	kfree(conf);
268	mddev->private = NULL;
269
270	return 0;
271}
272
273static int linear_make_request (mddev_t *mddev, struct bio *bio)
274{
275	dev_info_t *tmp_dev;
276	sector_t start_sector;
277
278	if (unlikely(bio->bi_rw & REQ_FLUSH)) {
279		md_flush_request(mddev, bio);
280		return 0;
281	}
282
283	rcu_read_lock();
284	tmp_dev = which_dev(mddev, bio->bi_sector);
285	start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
286
287
288	if (unlikely(bio->bi_sector >= (tmp_dev->end_sector)
289		     || (bio->bi_sector < start_sector))) {
290		char b[BDEVNAME_SIZE];
291
292		printk(KERN_ERR
293		       "md/linear:%s: make_request: Sector %llu out of bounds on "
294		       "dev %s: %llu sectors, offset %llu\n",
295		       mdname(mddev),
296		       (unsigned long long)bio->bi_sector,
297		       bdevname(tmp_dev->rdev->bdev, b),
298		       (unsigned long long)tmp_dev->rdev->sectors,
299		       (unsigned long long)start_sector);
300		rcu_read_unlock();
301		bio_io_error(bio);
302		return 0;
303	}
304	if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
305		     tmp_dev->end_sector)) {
306		/* This bio crosses a device boundary, so we have to
307		 * split it.
308		 */
309		struct bio_pair *bp;
310		sector_t end_sector = tmp_dev->end_sector;
311
312		rcu_read_unlock();
313
314		bp = bio_split(bio, end_sector - bio->bi_sector);
315
316		if (linear_make_request(mddev, &bp->bio1))
317			generic_make_request(&bp->bio1);
318		if (linear_make_request(mddev, &bp->bio2))
319			generic_make_request(&bp->bio2);
320		bio_pair_release(bp);
321		return 0;
322	}
323		    
324	bio->bi_bdev = tmp_dev->rdev->bdev;
325	bio->bi_sector = bio->bi_sector - start_sector
326		+ tmp_dev->rdev->data_offset;
327	rcu_read_unlock();
328
329	return 1;
330}
331
332static void linear_status (struct seq_file *seq, mddev_t *mddev)
333{
334
335	seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
336}
337
338
339static struct mdk_personality linear_personality =
340{
341	.name		= "linear",
342	.level		= LEVEL_LINEAR,
343	.owner		= THIS_MODULE,
344	.make_request	= linear_make_request,
345	.run		= linear_run,
346	.stop		= linear_stop,
347	.status		= linear_status,
348	.hot_add_disk	= linear_add,
349	.size		= linear_size,
350};
351
352static int __init linear_init (void)
353{
354	return register_md_personality (&linear_personality);
355}
356
357static void linear_exit (void)
358{
359	unregister_md_personality (&linear_personality);
360}
361
362
363module_init(linear_init);
364module_exit(linear_exit);
365MODULE_LICENSE("GPL");
366MODULE_DESCRIPTION("Linear device concatenation personality for MD");
367MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
368MODULE_ALIAS("md-linear");
369MODULE_ALIAS("md-level--1");