xref: /openbmc/linux/drivers/md/dm-snap.c (revision d5cb9783536a41df9f9cba5b0a1d78047ed787f7)
1 /*
2  * dm-snapshot.c
3  *
4  * Copyright (C) 2001-2002 Sistina Software (UK) Limited.
5  *
6  * This file is released under the GPL.
7  */
8 
9 #include <linux/blkdev.h>
10 #include <linux/config.h>
11 #include <linux/ctype.h>
12 #include <linux/device-mapper.h>
13 #include <linux/fs.h>
14 #include <linux/init.h>
15 #include <linux/kdev_t.h>
16 #include <linux/list.h>
17 #include <linux/mempool.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
20 #include <linux/vmalloc.h>
21 
22 #include "dm-snap.h"
23 #include "dm-bio-list.h"
24 #include "kcopyd.h"
25 
26 /*
27  * The percentage increment we will wake up users at
28  */
29 #define WAKE_UP_PERCENT 5
30 
31 /*
32  * kcopyd priority of snapshot operations
33  */
34 #define SNAPSHOT_COPY_PRIORITY 2
35 
36 /*
37  * Each snapshot reserves this many pages for io
38  */
39 #define SNAPSHOT_PAGES 256
40 
41 struct pending_exception {
42 	struct exception e;
43 
44 	/*
45 	 * Origin buffers waiting for this to complete are held
46 	 * in a bio list
47 	 */
48 	struct bio_list origin_bios;
49 	struct bio_list snapshot_bios;
50 
51 	/*
52 	 * Other pending_exceptions that are processing this
53 	 * chunk.  When this list is empty, we know we can
54 	 * complete the origins.
55 	 */
56 	struct list_head siblings;
57 
58 	/* Pointer back to snapshot context */
59 	struct dm_snapshot *snap;
60 
61 	/*
62 	 * 1 indicates the exception has already been sent to
63 	 * kcopyd.
64 	 */
65 	int started;
66 };
67 
68 /*
69  * Hash table mapping origin volumes to lists of snapshots and
70  * a lock to protect it
71  */
72 static kmem_cache_t *exception_cache;
73 static kmem_cache_t *pending_cache;
74 static mempool_t *pending_pool;
75 
76 /*
77  * One of these per registered origin, held in the snapshot_origins hash
78  */
79 struct origin {
80 	/* The origin device */
81 	struct block_device *bdev;
82 
83 	struct list_head hash_list;
84 
85 	/* List of snapshots for this origin */
86 	struct list_head snapshots;
87 };
88 
89 /*
90  * Size of the hash table for origin volumes. If we make this
91  * the size of the minors list then it should be nearly perfect
92  */
93 #define ORIGIN_HASH_SIZE 256
94 #define ORIGIN_MASK      0xFF
95 static struct list_head *_origins;
96 static struct rw_semaphore _origins_lock;
97 
98 static int init_origin_hash(void)
99 {
100 	int i;
101 
102 	_origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
103 			   GFP_KERNEL);
104 	if (!_origins) {
105 		DMERR("Device mapper: Snapshot: unable to allocate memory");
106 		return -ENOMEM;
107 	}
108 
109 	for (i = 0; i < ORIGIN_HASH_SIZE; i++)
110 		INIT_LIST_HEAD(_origins + i);
111 	init_rwsem(&_origins_lock);
112 
113 	return 0;
114 }
115 
116 static void exit_origin_hash(void)
117 {
118 	kfree(_origins);
119 }
120 
121 static inline unsigned int origin_hash(struct block_device *bdev)
122 {
123 	return bdev->bd_dev & ORIGIN_MASK;
124 }
125 
126 static struct origin *__lookup_origin(struct block_device *origin)
127 {
128 	struct list_head *ol;
129 	struct origin *o;
130 
131 	ol = &_origins[origin_hash(origin)];
132 	list_for_each_entry (o, ol, hash_list)
133 		if (bdev_equal(o->bdev, origin))
134 			return o;
135 
136 	return NULL;
137 }
138 
139 static void __insert_origin(struct origin *o)
140 {
141 	struct list_head *sl = &_origins[origin_hash(o->bdev)];
142 	list_add_tail(&o->hash_list, sl);
143 }
144 
145 /*
146  * Make a note of the snapshot and its origin so we can look it
147  * up when the origin has a write on it.
148  */
149 static int register_snapshot(struct dm_snapshot *snap)
150 {
151 	struct origin *o;
152 	struct block_device *bdev = snap->origin->bdev;
153 
154 	down_write(&_origins_lock);
155 	o = __lookup_origin(bdev);
156 
157 	if (!o) {
158 		/* New origin */
159 		o = kmalloc(sizeof(*o), GFP_KERNEL);
160 		if (!o) {
161 			up_write(&_origins_lock);
162 			return -ENOMEM;
163 		}
164 
165 		/* Initialise the struct */
166 		INIT_LIST_HEAD(&o->snapshots);
167 		o->bdev = bdev;
168 
169 		__insert_origin(o);
170 	}
171 
172 	list_add_tail(&snap->list, &o->snapshots);
173 
174 	up_write(&_origins_lock);
175 	return 0;
176 }
177 
178 static void unregister_snapshot(struct dm_snapshot *s)
179 {
180 	struct origin *o;
181 
182 	down_write(&_origins_lock);
183 	o = __lookup_origin(s->origin->bdev);
184 
185 	list_del(&s->list);
186 	if (list_empty(&o->snapshots)) {
187 		list_del(&o->hash_list);
188 		kfree(o);
189 	}
190 
191 	up_write(&_origins_lock);
192 }
193 
194 /*
195  * Implementation of the exception hash tables.
196  */
197 static int init_exception_table(struct exception_table *et, uint32_t size)
198 {
199 	unsigned int i;
200 
201 	et->hash_mask = size - 1;
202 	et->table = dm_vcalloc(size, sizeof(struct list_head));
203 	if (!et->table)
204 		return -ENOMEM;
205 
206 	for (i = 0; i < size; i++)
207 		INIT_LIST_HEAD(et->table + i);
208 
209 	return 0;
210 }
211 
212 static void exit_exception_table(struct exception_table *et, kmem_cache_t *mem)
213 {
214 	struct list_head *slot;
215 	struct exception *ex, *next;
216 	int i, size;
217 
218 	size = et->hash_mask + 1;
219 	for (i = 0; i < size; i++) {
220 		slot = et->table + i;
221 
222 		list_for_each_entry_safe (ex, next, slot, hash_list)
223 			kmem_cache_free(mem, ex);
224 	}
225 
226 	vfree(et->table);
227 }
228 
229 static inline uint32_t exception_hash(struct exception_table *et, chunk_t chunk)
230 {
231 	return chunk & et->hash_mask;
232 }
233 
234 static void insert_exception(struct exception_table *eh, struct exception *e)
235 {
236 	struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)];
237 	list_add(&e->hash_list, l);
238 }
239 
240 static inline void remove_exception(struct exception *e)
241 {
242 	list_del(&e->hash_list);
243 }
244 
245 /*
246  * Return the exception data for a sector, or NULL if not
247  * remapped.
248  */
249 static struct exception *lookup_exception(struct exception_table *et,
250 					  chunk_t chunk)
251 {
252 	struct list_head *slot;
253 	struct exception *e;
254 
255 	slot = &et->table[exception_hash(et, chunk)];
256 	list_for_each_entry (e, slot, hash_list)
257 		if (e->old_chunk == chunk)
258 			return e;
259 
260 	return NULL;
261 }
262 
263 static inline struct exception *alloc_exception(void)
264 {
265 	struct exception *e;
266 
267 	e = kmem_cache_alloc(exception_cache, GFP_NOIO);
268 	if (!e)
269 		e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
270 
271 	return e;
272 }
273 
274 static inline void free_exception(struct exception *e)
275 {
276 	kmem_cache_free(exception_cache, e);
277 }
278 
279 static inline struct pending_exception *alloc_pending_exception(void)
280 {
281 	return mempool_alloc(pending_pool, GFP_NOIO);
282 }
283 
284 static inline void free_pending_exception(struct pending_exception *pe)
285 {
286 	mempool_free(pe, pending_pool);
287 }
288 
289 int dm_add_exception(struct dm_snapshot *s, chunk_t old, chunk_t new)
290 {
291 	struct exception *e;
292 
293 	e = alloc_exception();
294 	if (!e)
295 		return -ENOMEM;
296 
297 	e->old_chunk = old;
298 	e->new_chunk = new;
299 	insert_exception(&s->complete, e);
300 	return 0;
301 }
302 
303 /*
304  * Hard coded magic.
305  */
306 static int calc_max_buckets(void)
307 {
308 	/* use a fixed size of 2MB */
309 	unsigned long mem = 2 * 1024 * 1024;
310 	mem /= sizeof(struct list_head);
311 
312 	return mem;
313 }
314 
315 /*
316  * Rounds a number down to a power of 2.
317  */
318 static inline uint32_t round_down(uint32_t n)
319 {
320 	while (n & (n - 1))
321 		n &= (n - 1);
322 	return n;
323 }
324 
325 /*
326  * Allocate room for a suitable hash table.
327  */
328 static int init_hash_tables(struct dm_snapshot *s)
329 {
330 	sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
331 
332 	/*
333 	 * Calculate based on the size of the original volume or
334 	 * the COW volume...
335 	 */
336 	cow_dev_size = get_dev_size(s->cow->bdev);
337 	origin_dev_size = get_dev_size(s->origin->bdev);
338 	max_buckets = calc_max_buckets();
339 
340 	hash_size = min(origin_dev_size, cow_dev_size) >> s->chunk_shift;
341 	hash_size = min(hash_size, max_buckets);
342 
343 	/* Round it down to a power of 2 */
344 	hash_size = round_down(hash_size);
345 	if (init_exception_table(&s->complete, hash_size))
346 		return -ENOMEM;
347 
348 	/*
349 	 * Allocate hash table for in-flight exceptions
350 	 * Make this smaller than the real hash table
351 	 */
352 	hash_size >>= 3;
353 	if (hash_size < 64)
354 		hash_size = 64;
355 
356 	if (init_exception_table(&s->pending, hash_size)) {
357 		exit_exception_table(&s->complete, exception_cache);
358 		return -ENOMEM;
359 	}
360 
361 	return 0;
362 }
363 
364 /*
365  * Round a number up to the nearest 'size' boundary.  size must
366  * be a power of 2.
367  */
368 static inline ulong round_up(ulong n, ulong size)
369 {
370 	size--;
371 	return (n + size) & ~size;
372 }
373 
374 /*
375  * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
376  */
377 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
378 {
379 	struct dm_snapshot *s;
380 	unsigned long chunk_size;
381 	int r = -EINVAL;
382 	char persistent;
383 	char *origin_path;
384 	char *cow_path;
385 	char *value;
386 	int blocksize;
387 
388 	if (argc < 4) {
389 		ti->error = "dm-snapshot: requires exactly 4 arguments";
390 		r = -EINVAL;
391 		goto bad1;
392 	}
393 
394 	origin_path = argv[0];
395 	cow_path = argv[1];
396 	persistent = toupper(*argv[2]);
397 
398 	if (persistent != 'P' && persistent != 'N') {
399 		ti->error = "Persistent flag is not P or N";
400 		r = -EINVAL;
401 		goto bad1;
402 	}
403 
404 	chunk_size = simple_strtoul(argv[3], &value, 10);
405 	if (chunk_size == 0 || value == NULL) {
406 		ti->error = "Invalid chunk size";
407 		r = -EINVAL;
408 		goto bad1;
409 	}
410 
411 	s = kmalloc(sizeof(*s), GFP_KERNEL);
412 	if (s == NULL) {
413 		ti->error = "Cannot allocate snapshot context private "
414 		    "structure";
415 		r = -ENOMEM;
416 		goto bad1;
417 	}
418 
419 	r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin);
420 	if (r) {
421 		ti->error = "Cannot get origin device";
422 		goto bad2;
423 	}
424 
425 	r = dm_get_device(ti, cow_path, 0, 0,
426 			  FMODE_READ | FMODE_WRITE, &s->cow);
427 	if (r) {
428 		dm_put_device(ti, s->origin);
429 		ti->error = "Cannot get COW device";
430 		goto bad2;
431 	}
432 
433 	/*
434 	 * Chunk size must be multiple of page size.  Silently
435 	 * round up if it's not.
436 	 */
437 	chunk_size = round_up(chunk_size, PAGE_SIZE >> 9);
438 
439 	/* Validate the chunk size against the device block size */
440 	blocksize = s->cow->bdev->bd_disk->queue->hardsect_size;
441 	if (chunk_size % (blocksize >> 9)) {
442 		ti->error = "Chunk size is not a multiple of device blocksize";
443 		r = -EINVAL;
444 		goto bad3;
445 	}
446 
447 	/* Check chunk_size is a power of 2 */
448 	if (chunk_size & (chunk_size - 1)) {
449 		ti->error = "Chunk size is not a power of 2";
450 		r = -EINVAL;
451 		goto bad3;
452 	}
453 
454 	s->chunk_size = chunk_size;
455 	s->chunk_mask = chunk_size - 1;
456 	s->type = persistent;
457 	s->chunk_shift = ffs(chunk_size) - 1;
458 
459 	s->valid = 1;
460 	s->have_metadata = 0;
461 	s->last_percent = 0;
462 	init_rwsem(&s->lock);
463 	s->table = ti->table;
464 
465 	/* Allocate hash table for COW data */
466 	if (init_hash_tables(s)) {
467 		ti->error = "Unable to allocate hash table space";
468 		r = -ENOMEM;
469 		goto bad3;
470 	}
471 
472 	/*
473 	 * Check the persistent flag - done here because we need the iobuf
474 	 * to check the LV header
475 	 */
476 	s->store.snap = s;
477 
478 	if (persistent == 'P')
479 		r = dm_create_persistent(&s->store, chunk_size);
480 	else
481 		r = dm_create_transient(&s->store, s, blocksize);
482 
483 	if (r) {
484 		ti->error = "Couldn't create exception store";
485 		r = -EINVAL;
486 		goto bad4;
487 	}
488 
489 	r = kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
490 	if (r) {
491 		ti->error = "Could not create kcopyd client";
492 		goto bad5;
493 	}
494 
495 	/* Add snapshot to the list of snapshots for this origin */
496 	if (register_snapshot(s)) {
497 		r = -EINVAL;
498 		ti->error = "Cannot register snapshot origin";
499 		goto bad6;
500 	}
501 
502 	ti->private = s;
503 	ti->split_io = chunk_size;
504 
505 	return 0;
506 
507  bad6:
508 	kcopyd_client_destroy(s->kcopyd_client);
509 
510  bad5:
511 	s->store.destroy(&s->store);
512 
513  bad4:
514 	exit_exception_table(&s->pending, pending_cache);
515 	exit_exception_table(&s->complete, exception_cache);
516 
517  bad3:
518 	dm_put_device(ti, s->cow);
519 	dm_put_device(ti, s->origin);
520 
521  bad2:
522 	kfree(s);
523 
524  bad1:
525 	return r;
526 }
527 
528 static void snapshot_dtr(struct dm_target *ti)
529 {
530 	struct dm_snapshot *s = (struct dm_snapshot *) ti->private;
531 
532 	unregister_snapshot(s);
533 
534 	exit_exception_table(&s->pending, pending_cache);
535 	exit_exception_table(&s->complete, exception_cache);
536 
537 	/* Deallocate memory used */
538 	s->store.destroy(&s->store);
539 
540 	dm_put_device(ti, s->origin);
541 	dm_put_device(ti, s->cow);
542 	kcopyd_client_destroy(s->kcopyd_client);
543 	kfree(s);
544 }
545 
546 /*
547  * Flush a list of buffers.
548  */
549 static void flush_bios(struct bio *bio)
550 {
551 	struct bio *n;
552 
553 	while (bio) {
554 		n = bio->bi_next;
555 		bio->bi_next = NULL;
556 		generic_make_request(bio);
557 		bio = n;
558 	}
559 }
560 
561 /*
562  * Error a list of buffers.
563  */
564 static void error_bios(struct bio *bio)
565 {
566 	struct bio *n;
567 
568 	while (bio) {
569 		n = bio->bi_next;
570 		bio->bi_next = NULL;
571 		bio_io_error(bio, bio->bi_size);
572 		bio = n;
573 	}
574 }
575 
576 static struct bio *__flush_bios(struct pending_exception *pe)
577 {
578 	struct pending_exception *sibling;
579 
580 	if (list_empty(&pe->siblings))
581 		return bio_list_get(&pe->origin_bios);
582 
583 	sibling = list_entry(pe->siblings.next,
584 			     struct pending_exception, siblings);
585 
586 	list_del(&pe->siblings);
587 
588 	/* This is fine as long as kcopyd is single-threaded. If kcopyd
589 	 * becomes multi-threaded, we'll need some locking here.
590 	 */
591 	bio_list_merge(&sibling->origin_bios, &pe->origin_bios);
592 
593 	return NULL;
594 }
595 
596 static void pending_complete(struct pending_exception *pe, int success)
597 {
598 	struct exception *e;
599 	struct dm_snapshot *s = pe->snap;
600 	struct bio *flush = NULL;
601 
602 	if (success) {
603 		e = alloc_exception();
604 		if (!e) {
605 			DMWARN("Unable to allocate exception.");
606 			down_write(&s->lock);
607 			s->store.drop_snapshot(&s->store);
608 			s->valid = 0;
609 			flush = __flush_bios(pe);
610 			up_write(&s->lock);
611 
612 			error_bios(bio_list_get(&pe->snapshot_bios));
613 			goto out;
614 		}
615 		*e = pe->e;
616 
617 		/*
618 		 * Add a proper exception, and remove the
619 		 * in-flight exception from the list.
620 		 */
621 		down_write(&s->lock);
622 		insert_exception(&s->complete, e);
623 		remove_exception(&pe->e);
624 		flush = __flush_bios(pe);
625 
626 		/* Submit any pending write bios */
627 		up_write(&s->lock);
628 
629 		flush_bios(bio_list_get(&pe->snapshot_bios));
630 	} else {
631 		/* Read/write error - snapshot is unusable */
632 		down_write(&s->lock);
633 		if (s->valid)
634 			DMERR("Error reading/writing snapshot");
635 		s->store.drop_snapshot(&s->store);
636 		s->valid = 0;
637 		remove_exception(&pe->e);
638 		flush = __flush_bios(pe);
639 		up_write(&s->lock);
640 
641 		error_bios(bio_list_get(&pe->snapshot_bios));
642 
643 		dm_table_event(s->table);
644 	}
645 
646  out:
647 	free_pending_exception(pe);
648 
649 	if (flush)
650 		flush_bios(flush);
651 }
652 
653 static void commit_callback(void *context, int success)
654 {
655 	struct pending_exception *pe = (struct pending_exception *) context;
656 	pending_complete(pe, success);
657 }
658 
659 /*
660  * Called when the copy I/O has finished.  kcopyd actually runs
661  * this code so don't block.
662  */
663 static void copy_callback(int read_err, unsigned int write_err, void *context)
664 {
665 	struct pending_exception *pe = (struct pending_exception *) context;
666 	struct dm_snapshot *s = pe->snap;
667 
668 	if (read_err || write_err)
669 		pending_complete(pe, 0);
670 
671 	else
672 		/* Update the metadata if we are persistent */
673 		s->store.commit_exception(&s->store, &pe->e, commit_callback,
674 					  pe);
675 }
676 
677 /*
678  * Dispatches the copy operation to kcopyd.
679  */
680 static inline void start_copy(struct pending_exception *pe)
681 {
682 	struct dm_snapshot *s = pe->snap;
683 	struct io_region src, dest;
684 	struct block_device *bdev = s->origin->bdev;
685 	sector_t dev_size;
686 
687 	dev_size = get_dev_size(bdev);
688 
689 	src.bdev = bdev;
690 	src.sector = chunk_to_sector(s, pe->e.old_chunk);
691 	src.count = min(s->chunk_size, dev_size - src.sector);
692 
693 	dest.bdev = s->cow->bdev;
694 	dest.sector = chunk_to_sector(s, pe->e.new_chunk);
695 	dest.count = src.count;
696 
697 	/* Hand over to kcopyd */
698 	kcopyd_copy(s->kcopyd_client,
699 		    &src, 1, &dest, 0, copy_callback, pe);
700 }
701 
702 /*
703  * Looks to see if this snapshot already has a pending exception
704  * for this chunk, otherwise it allocates a new one and inserts
705  * it into the pending table.
706  *
707  * NOTE: a write lock must be held on snap->lock before calling
708  * this.
709  */
710 static struct pending_exception *
711 __find_pending_exception(struct dm_snapshot *s, struct bio *bio)
712 {
713 	struct exception *e;
714 	struct pending_exception *pe;
715 	chunk_t chunk = sector_to_chunk(s, bio->bi_sector);
716 
717 	/*
718 	 * Is there a pending exception for this already ?
719 	 */
720 	e = lookup_exception(&s->pending, chunk);
721 	if (e) {
722 		/* cast the exception to a pending exception */
723 		pe = container_of(e, struct pending_exception, e);
724 
725 	} else {
726 		/*
727 		 * Create a new pending exception, we don't want
728 		 * to hold the lock while we do this.
729 		 */
730 		up_write(&s->lock);
731 		pe = alloc_pending_exception();
732 		down_write(&s->lock);
733 
734 		e = lookup_exception(&s->pending, chunk);
735 		if (e) {
736 			free_pending_exception(pe);
737 			pe = container_of(e, struct pending_exception, e);
738 		} else {
739 			pe->e.old_chunk = chunk;
740 			bio_list_init(&pe->origin_bios);
741 			bio_list_init(&pe->snapshot_bios);
742 			INIT_LIST_HEAD(&pe->siblings);
743 			pe->snap = s;
744 			pe->started = 0;
745 
746 			if (s->store.prepare_exception(&s->store, &pe->e)) {
747 				free_pending_exception(pe);
748 				s->valid = 0;
749 				return NULL;
750 			}
751 
752 			insert_exception(&s->pending, &pe->e);
753 		}
754 	}
755 
756 	return pe;
757 }
758 
759 static inline void remap_exception(struct dm_snapshot *s, struct exception *e,
760 				   struct bio *bio)
761 {
762 	bio->bi_bdev = s->cow->bdev;
763 	bio->bi_sector = chunk_to_sector(s, e->new_chunk) +
764 		(bio->bi_sector & s->chunk_mask);
765 }
766 
767 static int snapshot_map(struct dm_target *ti, struct bio *bio,
768 			union map_info *map_context)
769 {
770 	struct exception *e;
771 	struct dm_snapshot *s = (struct dm_snapshot *) ti->private;
772 	int r = 1;
773 	chunk_t chunk;
774 	struct pending_exception *pe;
775 
776 	chunk = sector_to_chunk(s, bio->bi_sector);
777 
778 	/* Full snapshots are not usable */
779 	if (!s->valid)
780 		return -EIO;
781 
782 	/*
783 	 * Write to snapshot - higher level takes care of RW/RO
784 	 * flags so we should only get this if we are
785 	 * writeable.
786 	 */
787 	if (bio_rw(bio) == WRITE) {
788 
789 		/* FIXME: should only take write lock if we need
790 		 * to copy an exception */
791 		down_write(&s->lock);
792 
793 		/* If the block is already remapped - use that, else remap it */
794 		e = lookup_exception(&s->complete, chunk);
795 		if (e) {
796 			remap_exception(s, e, bio);
797 			up_write(&s->lock);
798 
799 		} else {
800 			pe = __find_pending_exception(s, bio);
801 
802 			if (!pe) {
803 				if (s->store.drop_snapshot)
804 					s->store.drop_snapshot(&s->store);
805 				s->valid = 0;
806 				r = -EIO;
807 				up_write(&s->lock);
808 			} else {
809 				remap_exception(s, &pe->e, bio);
810 				bio_list_add(&pe->snapshot_bios, bio);
811 
812 				if (!pe->started) {
813 					/* this is protected by snap->lock */
814 					pe->started = 1;
815 					up_write(&s->lock);
816 					start_copy(pe);
817 				} else
818 					up_write(&s->lock);
819 				r = 0;
820 			}
821 		}
822 
823 	} else {
824 		/*
825 		 * FIXME: this read path scares me because we
826 		 * always use the origin when we have a pending
827 		 * exception.  However I can't think of a
828 		 * situation where this is wrong - ejt.
829 		 */
830 
831 		/* Do reads */
832 		down_read(&s->lock);
833 
834 		/* See if it it has been remapped */
835 		e = lookup_exception(&s->complete, chunk);
836 		if (e)
837 			remap_exception(s, e, bio);
838 		else
839 			bio->bi_bdev = s->origin->bdev;
840 
841 		up_read(&s->lock);
842 	}
843 
844 	return r;
845 }
846 
847 static void snapshot_resume(struct dm_target *ti)
848 {
849 	struct dm_snapshot *s = (struct dm_snapshot *) ti->private;
850 
851 	if (s->have_metadata)
852 		return;
853 
854 	if (s->store.read_metadata(&s->store)) {
855 		down_write(&s->lock);
856 		s->valid = 0;
857 		up_write(&s->lock);
858 	}
859 
860 	s->have_metadata = 1;
861 }
862 
863 static int snapshot_status(struct dm_target *ti, status_type_t type,
864 			   char *result, unsigned int maxlen)
865 {
866 	struct dm_snapshot *snap = (struct dm_snapshot *) ti->private;
867 
868 	switch (type) {
869 	case STATUSTYPE_INFO:
870 		if (!snap->valid)
871 			snprintf(result, maxlen, "Invalid");
872 		else {
873 			if (snap->store.fraction_full) {
874 				sector_t numerator, denominator;
875 				snap->store.fraction_full(&snap->store,
876 							  &numerator,
877 							  &denominator);
878 				snprintf(result, maxlen,
879 					 SECTOR_FORMAT "/" SECTOR_FORMAT,
880 					 numerator, denominator);
881 			}
882 			else
883 				snprintf(result, maxlen, "Unknown");
884 		}
885 		break;
886 
887 	case STATUSTYPE_TABLE:
888 		/*
889 		 * kdevname returns a static pointer so we need
890 		 * to make private copies if the output is to
891 		 * make sense.
892 		 */
893 		snprintf(result, maxlen, "%s %s %c " SECTOR_FORMAT,
894 			 snap->origin->name, snap->cow->name,
895 			 snap->type, snap->chunk_size);
896 		break;
897 	}
898 
899 	return 0;
900 }
901 
902 /*-----------------------------------------------------------------
903  * Origin methods
904  *---------------------------------------------------------------*/
905 static void list_merge(struct list_head *l1, struct list_head *l2)
906 {
907 	struct list_head *l1_n, *l2_p;
908 
909 	l1_n = l1->next;
910 	l2_p = l2->prev;
911 
912 	l1->next = l2;
913 	l2->prev = l1;
914 
915 	l2_p->next = l1_n;
916 	l1_n->prev = l2_p;
917 }
918 
919 static int __origin_write(struct list_head *snapshots, struct bio *bio)
920 {
921 	int r = 1, first = 1;
922 	struct dm_snapshot *snap;
923 	struct exception *e;
924 	struct pending_exception *pe, *last = NULL;
925 	chunk_t chunk;
926 
927 	/* Do all the snapshots on this origin */
928 	list_for_each_entry (snap, snapshots, list) {
929 
930 		/* Only deal with valid snapshots */
931 		if (!snap->valid)
932 			continue;
933 
934 		/* Nothing to do if writing beyond end of snapshot */
935 		if (bio->bi_sector >= dm_table_get_size(snap->table))
936 			continue;
937 
938 		down_write(&snap->lock);
939 
940 		/*
941 		 * Remember, different snapshots can have
942 		 * different chunk sizes.
943 		 */
944 		chunk = sector_to_chunk(snap, bio->bi_sector);
945 
946 		/*
947 		 * Check exception table to see if block
948 		 * is already remapped in this snapshot
949 		 * and trigger an exception if not.
950 		 */
951 		e = lookup_exception(&snap->complete, chunk);
952 		if (!e) {
953 			pe = __find_pending_exception(snap, bio);
954 			if (!pe) {
955 				snap->store.drop_snapshot(&snap->store);
956 				snap->valid = 0;
957 
958 			} else {
959 				if (last)
960 					list_merge(&pe->siblings,
961 						   &last->siblings);
962 
963 				last = pe;
964 				r = 0;
965 			}
966 		}
967 
968 		up_write(&snap->lock);
969 	}
970 
971 	/*
972 	 * Now that we have a complete pe list we can start the copying.
973 	 */
974 	if (last) {
975 		pe = last;
976 		do {
977 			down_write(&pe->snap->lock);
978 			if (first)
979 				bio_list_add(&pe->origin_bios, bio);
980 			if (!pe->started) {
981 				pe->started = 1;
982 				up_write(&pe->snap->lock);
983 				start_copy(pe);
984 			} else
985 				up_write(&pe->snap->lock);
986 			first = 0;
987 			pe = list_entry(pe->siblings.next,
988 					struct pending_exception, siblings);
989 
990 		} while (pe != last);
991 	}
992 
993 	return r;
994 }
995 
996 /*
997  * Called on a write from the origin driver.
998  */
999 static int do_origin(struct dm_dev *origin, struct bio *bio)
1000 {
1001 	struct origin *o;
1002 	int r = 1;
1003 
1004 	down_read(&_origins_lock);
1005 	o = __lookup_origin(origin->bdev);
1006 	if (o)
1007 		r = __origin_write(&o->snapshots, bio);
1008 	up_read(&_origins_lock);
1009 
1010 	return r;
1011 }
1012 
1013 /*
1014  * Origin: maps a linear range of a device, with hooks for snapshotting.
1015  */
1016 
1017 /*
1018  * Construct an origin mapping: <dev_path>
1019  * The context for an origin is merely a 'struct dm_dev *'
1020  * pointing to the real device.
1021  */
1022 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1023 {
1024 	int r;
1025 	struct dm_dev *dev;
1026 
1027 	if (argc != 1) {
1028 		ti->error = "dm-origin: incorrect number of arguments";
1029 		return -EINVAL;
1030 	}
1031 
1032 	r = dm_get_device(ti, argv[0], 0, ti->len,
1033 			  dm_table_get_mode(ti->table), &dev);
1034 	if (r) {
1035 		ti->error = "Cannot get target device";
1036 		return r;
1037 	}
1038 
1039 	ti->private = dev;
1040 	return 0;
1041 }
1042 
1043 static void origin_dtr(struct dm_target *ti)
1044 {
1045 	struct dm_dev *dev = (struct dm_dev *) ti->private;
1046 	dm_put_device(ti, dev);
1047 }
1048 
1049 static int origin_map(struct dm_target *ti, struct bio *bio,
1050 		      union map_info *map_context)
1051 {
1052 	struct dm_dev *dev = (struct dm_dev *) ti->private;
1053 	bio->bi_bdev = dev->bdev;
1054 
1055 	/* Only tell snapshots if this is a write */
1056 	return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : 1;
1057 }
1058 
1059 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
1060 
1061 /*
1062  * Set the target "split_io" field to the minimum of all the snapshots'
1063  * chunk sizes.
1064  */
1065 static void origin_resume(struct dm_target *ti)
1066 {
1067 	struct dm_dev *dev = (struct dm_dev *) ti->private;
1068 	struct dm_snapshot *snap;
1069 	struct origin *o;
1070 	chunk_t chunk_size = 0;
1071 
1072 	down_read(&_origins_lock);
1073 	o = __lookup_origin(dev->bdev);
1074 	if (o)
1075 		list_for_each_entry (snap, &o->snapshots, list)
1076 			chunk_size = min_not_zero(chunk_size, snap->chunk_size);
1077 	up_read(&_origins_lock);
1078 
1079 	ti->split_io = chunk_size;
1080 }
1081 
1082 static int origin_status(struct dm_target *ti, status_type_t type, char *result,
1083 			 unsigned int maxlen)
1084 {
1085 	struct dm_dev *dev = (struct dm_dev *) ti->private;
1086 
1087 	switch (type) {
1088 	case STATUSTYPE_INFO:
1089 		result[0] = '\0';
1090 		break;
1091 
1092 	case STATUSTYPE_TABLE:
1093 		snprintf(result, maxlen, "%s", dev->name);
1094 		break;
1095 	}
1096 
1097 	return 0;
1098 }
1099 
1100 static struct target_type origin_target = {
1101 	.name    = "snapshot-origin",
1102 	.version = {1, 0, 1},
1103 	.module  = THIS_MODULE,
1104 	.ctr     = origin_ctr,
1105 	.dtr     = origin_dtr,
1106 	.map     = origin_map,
1107 	.resume  = origin_resume,
1108 	.status  = origin_status,
1109 };
1110 
1111 static struct target_type snapshot_target = {
1112 	.name    = "snapshot",
1113 	.version = {1, 0, 1},
1114 	.module  = THIS_MODULE,
1115 	.ctr     = snapshot_ctr,
1116 	.dtr     = snapshot_dtr,
1117 	.map     = snapshot_map,
1118 	.resume  = snapshot_resume,
1119 	.status  = snapshot_status,
1120 };
1121 
1122 static int __init dm_snapshot_init(void)
1123 {
1124 	int r;
1125 
1126 	r = dm_register_target(&snapshot_target);
1127 	if (r) {
1128 		DMERR("snapshot target register failed %d", r);
1129 		return r;
1130 	}
1131 
1132 	r = dm_register_target(&origin_target);
1133 	if (r < 0) {
1134 		DMERR("Device mapper: Origin: register failed %d\n", r);
1135 		goto bad1;
1136 	}
1137 
1138 	r = init_origin_hash();
1139 	if (r) {
1140 		DMERR("init_origin_hash failed.");
1141 		goto bad2;
1142 	}
1143 
1144 	exception_cache = kmem_cache_create("dm-snapshot-ex",
1145 					    sizeof(struct exception),
1146 					    __alignof__(struct exception),
1147 					    0, NULL, NULL);
1148 	if (!exception_cache) {
1149 		DMERR("Couldn't create exception cache.");
1150 		r = -ENOMEM;
1151 		goto bad3;
1152 	}
1153 
1154 	pending_cache =
1155 	    kmem_cache_create("dm-snapshot-in",
1156 			      sizeof(struct pending_exception),
1157 			      __alignof__(struct pending_exception),
1158 			      0, NULL, NULL);
1159 	if (!pending_cache) {
1160 		DMERR("Couldn't create pending cache.");
1161 		r = -ENOMEM;
1162 		goto bad4;
1163 	}
1164 
1165 	pending_pool = mempool_create(128, mempool_alloc_slab,
1166 				      mempool_free_slab, pending_cache);
1167 	if (!pending_pool) {
1168 		DMERR("Couldn't create pending pool.");
1169 		r = -ENOMEM;
1170 		goto bad5;
1171 	}
1172 
1173 	return 0;
1174 
1175       bad5:
1176 	kmem_cache_destroy(pending_cache);
1177       bad4:
1178 	kmem_cache_destroy(exception_cache);
1179       bad3:
1180 	exit_origin_hash();
1181       bad2:
1182 	dm_unregister_target(&origin_target);
1183       bad1:
1184 	dm_unregister_target(&snapshot_target);
1185 	return r;
1186 }
1187 
1188 static void __exit dm_snapshot_exit(void)
1189 {
1190 	int r;
1191 
1192 	r = dm_unregister_target(&snapshot_target);
1193 	if (r)
1194 		DMERR("snapshot unregister failed %d", r);
1195 
1196 	r = dm_unregister_target(&origin_target);
1197 	if (r)
1198 		DMERR("origin unregister failed %d", r);
1199 
1200 	exit_origin_hash();
1201 	mempool_destroy(pending_pool);
1202 	kmem_cache_destroy(pending_cache);
1203 	kmem_cache_destroy(exception_cache);
1204 }
1205 
1206 /* Module hooks */
1207 module_init(dm_snapshot_init);
1208 module_exit(dm_snapshot_exit);
1209 
1210 MODULE_DESCRIPTION(DM_NAME " snapshot target");
1211 MODULE_AUTHOR("Joe Thornber");
1212 MODULE_LICENSE("GPL");
1213