xref: /openbmc/linux/drivers/md/raid0.c (revision 384740dc)
1 /*
2    raid0.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              Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
7 
8 
9    RAID-0 management functions.
10 
11    This program is free software; you can redistribute it and/or modify
12    it under the terms of the GNU General Public License as published by
13    the Free Software Foundation; either version 2, or (at your option)
14    any later version.
15 
16    You should have received a copy of the GNU General Public License
17    (for example /usr/src/linux/COPYING); if not, write to the Free
18    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
19 */
20 
21 #include <linux/module.h>
22 #include <linux/raid/raid0.h>
23 
24 #define MAJOR_NR MD_MAJOR
25 #define MD_DRIVER
26 #define MD_PERSONALITY
27 
28 static void raid0_unplug(struct request_queue *q)
29 {
30 	mddev_t *mddev = q->queuedata;
31 	raid0_conf_t *conf = mddev_to_conf(mddev);
32 	mdk_rdev_t **devlist = conf->strip_zone[0].dev;
33 	int i;
34 
35 	for (i=0; i<mddev->raid_disks; i++) {
36 		struct request_queue *r_queue = bdev_get_queue(devlist[i]->bdev);
37 
38 		blk_unplug(r_queue);
39 	}
40 }
41 
42 static int raid0_congested(void *data, int bits)
43 {
44 	mddev_t *mddev = data;
45 	raid0_conf_t *conf = mddev_to_conf(mddev);
46 	mdk_rdev_t **devlist = conf->strip_zone[0].dev;
47 	int i, ret = 0;
48 
49 	for (i = 0; i < mddev->raid_disks && !ret ; i++) {
50 		struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
51 
52 		ret |= bdi_congested(&q->backing_dev_info, bits);
53 	}
54 	return ret;
55 }
56 
57 
58 static int create_strip_zones (mddev_t *mddev)
59 {
60 	int i, c, j;
61 	sector_t current_offset, curr_zone_offset;
62 	sector_t min_spacing;
63 	raid0_conf_t *conf = mddev_to_conf(mddev);
64 	mdk_rdev_t *smallest, *rdev1, *rdev2, *rdev;
65 	struct list_head *tmp1, *tmp2;
66 	struct strip_zone *zone;
67 	int cnt;
68 	char b[BDEVNAME_SIZE];
69 
70 	/*
71 	 * The number of 'same size groups'
72 	 */
73 	conf->nr_strip_zones = 0;
74 
75 	rdev_for_each(rdev1, tmp1, mddev) {
76 		printk("raid0: looking at %s\n",
77 			bdevname(rdev1->bdev,b));
78 		c = 0;
79 		rdev_for_each(rdev2, tmp2, mddev) {
80 			printk("raid0:   comparing %s(%llu)",
81 			       bdevname(rdev1->bdev,b),
82 			       (unsigned long long)rdev1->size);
83 			printk(" with %s(%llu)\n",
84 			       bdevname(rdev2->bdev,b),
85 			       (unsigned long long)rdev2->size);
86 			if (rdev2 == rdev1) {
87 				printk("raid0:   END\n");
88 				break;
89 			}
90 			if (rdev2->size == rdev1->size)
91 			{
92 				/*
93 				 * Not unique, don't count it as a new
94 				 * group
95 				 */
96 				printk("raid0:   EQUAL\n");
97 				c = 1;
98 				break;
99 			}
100 			printk("raid0:   NOT EQUAL\n");
101 		}
102 		if (!c) {
103 			printk("raid0:   ==> UNIQUE\n");
104 			conf->nr_strip_zones++;
105 			printk("raid0: %d zones\n", conf->nr_strip_zones);
106 		}
107 	}
108 	printk("raid0: FINAL %d zones\n", conf->nr_strip_zones);
109 
110 	conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
111 				conf->nr_strip_zones, GFP_KERNEL);
112 	if (!conf->strip_zone)
113 		return 1;
114 	conf->devlist = kzalloc(sizeof(mdk_rdev_t*)*
115 				conf->nr_strip_zones*mddev->raid_disks,
116 				GFP_KERNEL);
117 	if (!conf->devlist)
118 		return 1;
119 
120 	/* The first zone must contain all devices, so here we check that
121 	 * there is a proper alignment of slots to devices and find them all
122 	 */
123 	zone = &conf->strip_zone[0];
124 	cnt = 0;
125 	smallest = NULL;
126 	zone->dev = conf->devlist;
127 	rdev_for_each(rdev1, tmp1, mddev) {
128 		int j = rdev1->raid_disk;
129 
130 		if (j < 0 || j >= mddev->raid_disks) {
131 			printk("raid0: bad disk number %d - aborting!\n", j);
132 			goto abort;
133 		}
134 		if (zone->dev[j]) {
135 			printk("raid0: multiple devices for %d - aborting!\n",
136 				j);
137 			goto abort;
138 		}
139 		zone->dev[j] = rdev1;
140 
141 		blk_queue_stack_limits(mddev->queue,
142 				       rdev1->bdev->bd_disk->queue);
143 		/* as we don't honour merge_bvec_fn, we must never risk
144 		 * violating it, so limit ->max_sector to one PAGE, as
145 		 * a one page request is never in violation.
146 		 */
147 
148 		if (rdev1->bdev->bd_disk->queue->merge_bvec_fn &&
149 		    mddev->queue->max_sectors > (PAGE_SIZE>>9))
150 			blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
151 
152 		if (!smallest || (rdev1->size <smallest->size))
153 			smallest = rdev1;
154 		cnt++;
155 	}
156 	if (cnt != mddev->raid_disks) {
157 		printk("raid0: too few disks (%d of %d) - aborting!\n",
158 			cnt, mddev->raid_disks);
159 		goto abort;
160 	}
161 	zone->nb_dev = cnt;
162 	zone->size = smallest->size * cnt;
163 	zone->zone_offset = 0;
164 
165 	current_offset = smallest->size;
166 	curr_zone_offset = zone->size;
167 
168 	/* now do the other zones */
169 	for (i = 1; i < conf->nr_strip_zones; i++)
170 	{
171 		zone = conf->strip_zone + i;
172 		zone->dev = conf->strip_zone[i-1].dev + mddev->raid_disks;
173 
174 		printk("raid0: zone %d\n", i);
175 		zone->dev_offset = current_offset;
176 		smallest = NULL;
177 		c = 0;
178 
179 		for (j=0; j<cnt; j++) {
180 			char b[BDEVNAME_SIZE];
181 			rdev = conf->strip_zone[0].dev[j];
182 			printk("raid0: checking %s ...", bdevname(rdev->bdev,b));
183 			if (rdev->size > current_offset)
184 			{
185 				printk(" contained as device %d\n", c);
186 				zone->dev[c] = rdev;
187 				c++;
188 				if (!smallest || (rdev->size <smallest->size)) {
189 					smallest = rdev;
190 					printk("  (%llu) is smallest!.\n",
191 						(unsigned long long)rdev->size);
192 				}
193 			} else
194 				printk(" nope.\n");
195 		}
196 
197 		zone->nb_dev = c;
198 		zone->size = (smallest->size - current_offset) * c;
199 		printk("raid0: zone->nb_dev: %d, size: %llu\n",
200 			zone->nb_dev, (unsigned long long)zone->size);
201 
202 		zone->zone_offset = curr_zone_offset;
203 		curr_zone_offset += zone->size;
204 
205 		current_offset = smallest->size;
206 		printk("raid0: current zone offset: %llu\n",
207 			(unsigned long long)current_offset);
208 	}
209 
210 	/* Now find appropriate hash spacing.
211 	 * We want a number which causes most hash entries to cover
212 	 * at most two strips, but the hash table must be at most
213 	 * 1 PAGE.  We choose the smallest strip, or contiguous collection
214 	 * of strips, that has big enough size.  We never consider the last
215 	 * strip though as it's size has no bearing on the efficacy of the hash
216 	 * table.
217 	 */
218 	conf->hash_spacing = curr_zone_offset;
219 	min_spacing = curr_zone_offset;
220 	sector_div(min_spacing, PAGE_SIZE/sizeof(struct strip_zone*));
221 	for (i=0; i < conf->nr_strip_zones-1; i++) {
222 		sector_t sz = 0;
223 		for (j=i; j<conf->nr_strip_zones-1 &&
224 			     sz < min_spacing ; j++)
225 			sz += conf->strip_zone[j].size;
226 		if (sz >= min_spacing && sz < conf->hash_spacing)
227 			conf->hash_spacing = sz;
228 	}
229 
230 	mddev->queue->unplug_fn = raid0_unplug;
231 
232 	mddev->queue->backing_dev_info.congested_fn = raid0_congested;
233 	mddev->queue->backing_dev_info.congested_data = mddev;
234 
235 	printk("raid0: done.\n");
236 	return 0;
237  abort:
238 	return 1;
239 }
240 
241 /**
242  *	raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
243  *	@q: request queue
244  *	@bvm: properties of new bio
245  *	@biovec: the request that could be merged to it.
246  *
247  *	Return amount of bytes we can accept at this offset
248  */
249 static int raid0_mergeable_bvec(struct request_queue *q,
250 				struct bvec_merge_data *bvm,
251 				struct bio_vec *biovec)
252 {
253 	mddev_t *mddev = q->queuedata;
254 	sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
255 	int max;
256 	unsigned int chunk_sectors = mddev->chunk_size >> 9;
257 	unsigned int bio_sectors = bvm->bi_size >> 9;
258 
259 	max =  (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
260 	if (max < 0) max = 0; /* bio_add cannot handle a negative return */
261 	if (max <= biovec->bv_len && bio_sectors == 0)
262 		return biovec->bv_len;
263 	else
264 		return max;
265 }
266 
267 static int raid0_run (mddev_t *mddev)
268 {
269 	unsigned  cur=0, i=0, nb_zone;
270 	s64 size;
271 	raid0_conf_t *conf;
272 	mdk_rdev_t *rdev;
273 	struct list_head *tmp;
274 
275 	if (mddev->chunk_size == 0) {
276 		printk(KERN_ERR "md/raid0: non-zero chunk size required.\n");
277 		return -EINVAL;
278 	}
279 	printk(KERN_INFO "%s: setting max_sectors to %d, segment boundary to %d\n",
280 	       mdname(mddev),
281 	       mddev->chunk_size >> 9,
282 	       (mddev->chunk_size>>1)-1);
283 	blk_queue_max_sectors(mddev->queue, mddev->chunk_size >> 9);
284 	blk_queue_segment_boundary(mddev->queue, (mddev->chunk_size>>1) - 1);
285 	mddev->queue->queue_lock = &mddev->queue->__queue_lock;
286 
287 	conf = kmalloc(sizeof (raid0_conf_t), GFP_KERNEL);
288 	if (!conf)
289 		goto out;
290 	mddev->private = (void *)conf;
291 
292 	conf->strip_zone = NULL;
293 	conf->devlist = NULL;
294 	if (create_strip_zones (mddev))
295 		goto out_free_conf;
296 
297 	/* calculate array device size */
298 	mddev->array_sectors = 0;
299 	rdev_for_each(rdev, tmp, mddev)
300 		mddev->array_sectors += rdev->size * 2;
301 
302 	printk("raid0 : md_size is %llu blocks.\n",
303 		(unsigned long long)mddev->array_sectors / 2);
304 	printk("raid0 : conf->hash_spacing is %llu blocks.\n",
305 		(unsigned long long)conf->hash_spacing);
306 	{
307 		sector_t s = mddev->array_sectors / 2;
308 		sector_t space = conf->hash_spacing;
309 		int round;
310 		conf->preshift = 0;
311 		if (sizeof(sector_t) > sizeof(u32)) {
312 			/*shift down space and s so that sector_div will work */
313 			while (space > (sector_t) (~(u32)0)) {
314 				s >>= 1;
315 				space >>= 1;
316 				s += 1; /* force round-up */
317 				conf->preshift++;
318 			}
319 		}
320 		round = sector_div(s, (u32)space) ? 1 : 0;
321 		nb_zone = s + round;
322 	}
323 	printk("raid0 : nb_zone is %d.\n", nb_zone);
324 
325 	printk("raid0 : Allocating %Zd bytes for hash.\n",
326 				nb_zone*sizeof(struct strip_zone*));
327 	conf->hash_table = kmalloc (sizeof (struct strip_zone *)*nb_zone, GFP_KERNEL);
328 	if (!conf->hash_table)
329 		goto out_free_conf;
330 	size = conf->strip_zone[cur].size;
331 
332 	conf->hash_table[0] = conf->strip_zone + cur;
333 	for (i=1; i< nb_zone; i++) {
334 		while (size <= conf->hash_spacing) {
335 			cur++;
336 			size += conf->strip_zone[cur].size;
337 		}
338 		size -= conf->hash_spacing;
339 		conf->hash_table[i] = conf->strip_zone + cur;
340 	}
341 	if (conf->preshift) {
342 		conf->hash_spacing >>= conf->preshift;
343 		/* round hash_spacing up so when we divide by it, we
344 		 * err on the side of too-low, which is safest
345 		 */
346 		conf->hash_spacing++;
347 	}
348 
349 	/* calculate the max read-ahead size.
350 	 * For read-ahead of large files to be effective, we need to
351 	 * readahead at least twice a whole stripe. i.e. number of devices
352 	 * multiplied by chunk size times 2.
353 	 * If an individual device has an ra_pages greater than the
354 	 * chunk size, then we will not drive that device as hard as it
355 	 * wants.  We consider this a configuration error: a larger
356 	 * chunksize should be used in that case.
357 	 */
358 	{
359 		int stripe = mddev->raid_disks * mddev->chunk_size / PAGE_SIZE;
360 		if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
361 			mddev->queue->backing_dev_info.ra_pages = 2* stripe;
362 	}
363 
364 
365 	blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
366 	return 0;
367 
368 out_free_conf:
369 	kfree(conf->strip_zone);
370 	kfree(conf->devlist);
371 	kfree(conf);
372 	mddev->private = NULL;
373 out:
374 	return -ENOMEM;
375 }
376 
377 static int raid0_stop (mddev_t *mddev)
378 {
379 	raid0_conf_t *conf = mddev_to_conf(mddev);
380 
381 	blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
382 	kfree(conf->hash_table);
383 	conf->hash_table = NULL;
384 	kfree(conf->strip_zone);
385 	conf->strip_zone = NULL;
386 	kfree(conf);
387 	mddev->private = NULL;
388 
389 	return 0;
390 }
391 
392 static int raid0_make_request (struct request_queue *q, struct bio *bio)
393 {
394 	mddev_t *mddev = q->queuedata;
395 	unsigned int sect_in_chunk, chunksize_bits,  chunk_size, chunk_sects;
396 	raid0_conf_t *conf = mddev_to_conf(mddev);
397 	struct strip_zone *zone;
398 	mdk_rdev_t *tmp_dev;
399 	sector_t chunk;
400 	sector_t block, rsect;
401 	const int rw = bio_data_dir(bio);
402 	int cpu;
403 
404 	if (unlikely(bio_barrier(bio))) {
405 		bio_endio(bio, -EOPNOTSUPP);
406 		return 0;
407 	}
408 
409 	cpu = part_stat_lock();
410 	part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
411 	part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
412 		      bio_sectors(bio));
413 	part_stat_unlock();
414 
415 	chunk_size = mddev->chunk_size >> 10;
416 	chunk_sects = mddev->chunk_size >> 9;
417 	chunksize_bits = ffz(~chunk_size);
418 	block = bio->bi_sector >> 1;
419 
420 
421 	if (unlikely(chunk_sects < (bio->bi_sector & (chunk_sects - 1)) + (bio->bi_size >> 9))) {
422 		struct bio_pair *bp;
423 		/* Sanity check -- queue functions should prevent this happening */
424 		if (bio->bi_vcnt != 1 ||
425 		    bio->bi_idx != 0)
426 			goto bad_map;
427 		/* This is a one page bio that upper layers
428 		 * refuse to split for us, so we need to split it.
429 		 */
430 		bp = bio_split(bio, chunk_sects - (bio->bi_sector & (chunk_sects - 1)));
431 		if (raid0_make_request(q, &bp->bio1))
432 			generic_make_request(&bp->bio1);
433 		if (raid0_make_request(q, &bp->bio2))
434 			generic_make_request(&bp->bio2);
435 
436 		bio_pair_release(bp);
437 		return 0;
438 	}
439 
440 
441 	{
442 		sector_t x = block >> conf->preshift;
443 		sector_div(x, (u32)conf->hash_spacing);
444 		zone = conf->hash_table[x];
445 	}
446 
447 	while (block >= (zone->zone_offset + zone->size))
448 		zone++;
449 
450 	sect_in_chunk = bio->bi_sector & ((chunk_size<<1) -1);
451 
452 
453 	{
454 		sector_t x =  (block - zone->zone_offset) >> chunksize_bits;
455 
456 		sector_div(x, zone->nb_dev);
457 		chunk = x;
458 
459 		x = block >> chunksize_bits;
460 		tmp_dev = zone->dev[sector_div(x, zone->nb_dev)];
461 	}
462 	rsect = (((chunk << chunksize_bits) + zone->dev_offset)<<1)
463 		+ sect_in_chunk;
464 
465 	bio->bi_bdev = tmp_dev->bdev;
466 	bio->bi_sector = rsect + tmp_dev->data_offset;
467 
468 	/*
469 	 * Let the main block layer submit the IO and resolve recursion:
470 	 */
471 	return 1;
472 
473 bad_map:
474 	printk("raid0_make_request bug: can't convert block across chunks"
475 		" or bigger than %dk %llu %d\n", chunk_size,
476 		(unsigned long long)bio->bi_sector, bio->bi_size >> 10);
477 
478 	bio_io_error(bio);
479 	return 0;
480 }
481 
482 static void raid0_status (struct seq_file *seq, mddev_t *mddev)
483 {
484 #undef MD_DEBUG
485 #ifdef MD_DEBUG
486 	int j, k, h;
487 	char b[BDEVNAME_SIZE];
488 	raid0_conf_t *conf = mddev_to_conf(mddev);
489 
490 	h = 0;
491 	for (j = 0; j < conf->nr_strip_zones; j++) {
492 		seq_printf(seq, "      z%d", j);
493 		if (conf->hash_table[h] == conf->strip_zone+j)
494 			seq_printf(seq, "(h%d)", h++);
495 		seq_printf(seq, "=[");
496 		for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
497 			seq_printf(seq, "%s/", bdevname(
498 				conf->strip_zone[j].dev[k]->bdev,b));
499 
500 		seq_printf(seq, "] zo=%d do=%d s=%d\n",
501 				conf->strip_zone[j].zone_offset,
502 				conf->strip_zone[j].dev_offset,
503 				conf->strip_zone[j].size);
504 	}
505 #endif
506 	seq_printf(seq, " %dk chunks", mddev->chunk_size/1024);
507 	return;
508 }
509 
510 static struct mdk_personality raid0_personality=
511 {
512 	.name		= "raid0",
513 	.level		= 0,
514 	.owner		= THIS_MODULE,
515 	.make_request	= raid0_make_request,
516 	.run		= raid0_run,
517 	.stop		= raid0_stop,
518 	.status		= raid0_status,
519 };
520 
521 static int __init raid0_init (void)
522 {
523 	return register_md_personality (&raid0_personality);
524 }
525 
526 static void raid0_exit (void)
527 {
528 	unregister_md_personality (&raid0_personality);
529 }
530 
531 module_init(raid0_init);
532 module_exit(raid0_exit);
533 MODULE_LICENSE("GPL");
534 MODULE_ALIAS("md-personality-2"); /* RAID0 */
535 MODULE_ALIAS("md-raid0");
536 MODULE_ALIAS("md-level-0");
537