blob: 1d0fafda0f761abb3630bd180f628dd7dfbbbc4d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010#include <linux/ctype.h>
11#include <linux/device-mapper.h>
12#include <linux/fs.h>
13#include <linux/init.h>
14#include <linux/kdev_t.h>
15#include <linux/list.h>
16#include <linux/mempool.h>
17#include <linux/module.h>
18#include <linux/slab.h>
19#include <linux/vmalloc.h>
20
21#include "dm-snap.h"
22#include "dm-bio-list.h"
23#include "kcopyd.h"
24
Alasdair G Kergon72d94862006-06-26 00:27:35 -070025#define DM_MSG_PREFIX "snapshots"
26
Linus Torvalds1da177e2005-04-16 15:20:36 -070027/*
28 * The percentage increment we will wake up users at
29 */
30#define WAKE_UP_PERCENT 5
31
32/*
33 * kcopyd priority of snapshot operations
34 */
35#define SNAPSHOT_COPY_PRIORITY 2
36
37/*
38 * Each snapshot reserves this many pages for io
39 */
40#define SNAPSHOT_PAGES 256
41
42struct pending_exception {
43 struct exception e;
44
45 /*
46 * Origin buffers waiting for this to complete are held
47 * in a bio list
48 */
49 struct bio_list origin_bios;
50 struct bio_list snapshot_bios;
51
52 /*
Alasdair G Kergoneccf0812006-03-27 01:17:42 -080053 * Short-term queue of pending exceptions prior to submission.
54 */
55 struct list_head list;
56
57 /*
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -080058 * The primary pending_exception is the one that holds
59 * the sibling_count and the list of origin_bios for a
60 * group of pending_exceptions. It is always last to get freed.
61 * These fields get set up when writing to the origin.
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 */
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -080063 struct pending_exception *primary_pe;
64
65 /*
66 * Number of pending_exceptions processing this chunk.
67 * When this drops to zero we must complete the origin bios.
68 * If incrementing or decrementing this, hold pe->snap->lock for
69 * the sibling concerned and not pe->primary_pe->snap->lock unless
70 * they are the same.
71 */
72 atomic_t sibling_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
74 /* Pointer back to snapshot context */
75 struct dm_snapshot *snap;
76
77 /*
78 * 1 indicates the exception has already been sent to
79 * kcopyd.
80 */
81 int started;
82};
83
84/*
85 * Hash table mapping origin volumes to lists of snapshots and
86 * a lock to protect it
87 */
88static kmem_cache_t *exception_cache;
89static kmem_cache_t *pending_cache;
90static mempool_t *pending_pool;
91
92/*
93 * One of these per registered origin, held in the snapshot_origins hash
94 */
95struct origin {
96 /* The origin device */
97 struct block_device *bdev;
98
99 struct list_head hash_list;
100
101 /* List of snapshots for this origin */
102 struct list_head snapshots;
103};
104
105/*
106 * Size of the hash table for origin volumes. If we make this
107 * the size of the minors list then it should be nearly perfect
108 */
109#define ORIGIN_HASH_SIZE 256
110#define ORIGIN_MASK 0xFF
111static struct list_head *_origins;
112static struct rw_semaphore _origins_lock;
113
114static int init_origin_hash(void)
115{
116 int i;
117
118 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
119 GFP_KERNEL);
120 if (!_origins) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700121 DMERR("unable to allocate memory");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 return -ENOMEM;
123 }
124
125 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
126 INIT_LIST_HEAD(_origins + i);
127 init_rwsem(&_origins_lock);
128
129 return 0;
130}
131
132static void exit_origin_hash(void)
133{
134 kfree(_origins);
135}
136
137static inline unsigned int origin_hash(struct block_device *bdev)
138{
139 return bdev->bd_dev & ORIGIN_MASK;
140}
141
142static struct origin *__lookup_origin(struct block_device *origin)
143{
144 struct list_head *ol;
145 struct origin *o;
146
147 ol = &_origins[origin_hash(origin)];
148 list_for_each_entry (o, ol, hash_list)
149 if (bdev_equal(o->bdev, origin))
150 return o;
151
152 return NULL;
153}
154
155static void __insert_origin(struct origin *o)
156{
157 struct list_head *sl = &_origins[origin_hash(o->bdev)];
158 list_add_tail(&o->hash_list, sl);
159}
160
161/*
162 * Make a note of the snapshot and its origin so we can look it
163 * up when the origin has a write on it.
164 */
165static int register_snapshot(struct dm_snapshot *snap)
166{
167 struct origin *o;
168 struct block_device *bdev = snap->origin->bdev;
169
170 down_write(&_origins_lock);
171 o = __lookup_origin(bdev);
172
173 if (!o) {
174 /* New origin */
175 o = kmalloc(sizeof(*o), GFP_KERNEL);
176 if (!o) {
177 up_write(&_origins_lock);
178 return -ENOMEM;
179 }
180
181 /* Initialise the struct */
182 INIT_LIST_HEAD(&o->snapshots);
183 o->bdev = bdev;
184
185 __insert_origin(o);
186 }
187
188 list_add_tail(&snap->list, &o->snapshots);
189
190 up_write(&_origins_lock);
191 return 0;
192}
193
194static void unregister_snapshot(struct dm_snapshot *s)
195{
196 struct origin *o;
197
198 down_write(&_origins_lock);
199 o = __lookup_origin(s->origin->bdev);
200
201 list_del(&s->list);
202 if (list_empty(&o->snapshots)) {
203 list_del(&o->hash_list);
204 kfree(o);
205 }
206
207 up_write(&_origins_lock);
208}
209
210/*
211 * Implementation of the exception hash tables.
212 */
213static int init_exception_table(struct exception_table *et, uint32_t size)
214{
215 unsigned int i;
216
217 et->hash_mask = size - 1;
218 et->table = dm_vcalloc(size, sizeof(struct list_head));
219 if (!et->table)
220 return -ENOMEM;
221
222 for (i = 0; i < size; i++)
223 INIT_LIST_HEAD(et->table + i);
224
225 return 0;
226}
227
228static void exit_exception_table(struct exception_table *et, kmem_cache_t *mem)
229{
230 struct list_head *slot;
231 struct exception *ex, *next;
232 int i, size;
233
234 size = et->hash_mask + 1;
235 for (i = 0; i < size; i++) {
236 slot = et->table + i;
237
238 list_for_each_entry_safe (ex, next, slot, hash_list)
239 kmem_cache_free(mem, ex);
240 }
241
242 vfree(et->table);
243}
244
245static inline uint32_t exception_hash(struct exception_table *et, chunk_t chunk)
246{
247 return chunk & et->hash_mask;
248}
249
250static void insert_exception(struct exception_table *eh, struct exception *e)
251{
252 struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)];
253 list_add(&e->hash_list, l);
254}
255
256static inline void remove_exception(struct exception *e)
257{
258 list_del(&e->hash_list);
259}
260
261/*
262 * Return the exception data for a sector, or NULL if not
263 * remapped.
264 */
265static struct exception *lookup_exception(struct exception_table *et,
266 chunk_t chunk)
267{
268 struct list_head *slot;
269 struct exception *e;
270
271 slot = &et->table[exception_hash(et, chunk)];
272 list_for_each_entry (e, slot, hash_list)
273 if (e->old_chunk == chunk)
274 return e;
275
276 return NULL;
277}
278
279static inline struct exception *alloc_exception(void)
280{
281 struct exception *e;
282
283 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
284 if (!e)
285 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
286
287 return e;
288}
289
290static inline void free_exception(struct exception *e)
291{
292 kmem_cache_free(exception_cache, e);
293}
294
295static inline struct pending_exception *alloc_pending_exception(void)
296{
297 return mempool_alloc(pending_pool, GFP_NOIO);
298}
299
300static inline void free_pending_exception(struct pending_exception *pe)
301{
302 mempool_free(pe, pending_pool);
303}
304
305int dm_add_exception(struct dm_snapshot *s, chunk_t old, chunk_t new)
306{
307 struct exception *e;
308
309 e = alloc_exception();
310 if (!e)
311 return -ENOMEM;
312
313 e->old_chunk = old;
314 e->new_chunk = new;
315 insert_exception(&s->complete, e);
316 return 0;
317}
318
319/*
320 * Hard coded magic.
321 */
322static int calc_max_buckets(void)
323{
324 /* use a fixed size of 2MB */
325 unsigned long mem = 2 * 1024 * 1024;
326 mem /= sizeof(struct list_head);
327
328 return mem;
329}
330
331/*
332 * Rounds a number down to a power of 2.
333 */
334static inline uint32_t round_down(uint32_t n)
335{
336 while (n & (n - 1))
337 n &= (n - 1);
338 return n;
339}
340
341/*
342 * Allocate room for a suitable hash table.
343 */
344static int init_hash_tables(struct dm_snapshot *s)
345{
346 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
347
348 /*
349 * Calculate based on the size of the original volume or
350 * the COW volume...
351 */
352 cow_dev_size = get_dev_size(s->cow->bdev);
353 origin_dev_size = get_dev_size(s->origin->bdev);
354 max_buckets = calc_max_buckets();
355
356 hash_size = min(origin_dev_size, cow_dev_size) >> s->chunk_shift;
357 hash_size = min(hash_size, max_buckets);
358
359 /* Round it down to a power of 2 */
360 hash_size = round_down(hash_size);
361 if (init_exception_table(&s->complete, hash_size))
362 return -ENOMEM;
363
364 /*
365 * Allocate hash table for in-flight exceptions
366 * Make this smaller than the real hash table
367 */
368 hash_size >>= 3;
369 if (hash_size < 64)
370 hash_size = 64;
371
372 if (init_exception_table(&s->pending, hash_size)) {
373 exit_exception_table(&s->complete, exception_cache);
374 return -ENOMEM;
375 }
376
377 return 0;
378}
379
380/*
381 * Round a number up to the nearest 'size' boundary. size must
382 * be a power of 2.
383 */
384static inline ulong round_up(ulong n, ulong size)
385{
386 size--;
387 return (n + size) & ~size;
388}
389
Alasdair G Kergon2d38fe22006-01-06 00:20:02 -0800390static void read_snapshot_metadata(struct dm_snapshot *s)
391{
Alasdair G Kergon2d38fe22006-01-06 00:20:02 -0800392 if (s->store.read_metadata(&s->store)) {
393 down_write(&s->lock);
394 s->valid = 0;
395 up_write(&s->lock);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800396
397 dm_table_event(s->table);
Alasdair G Kergon2d38fe22006-01-06 00:20:02 -0800398 }
Alasdair G Kergon2d38fe22006-01-06 00:20:02 -0800399}
400
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401/*
402 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
403 */
404static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
405{
406 struct dm_snapshot *s;
407 unsigned long chunk_size;
408 int r = -EINVAL;
409 char persistent;
410 char *origin_path;
411 char *cow_path;
412 char *value;
413 int blocksize;
414
415 if (argc < 4) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700416 ti->error = "requires exactly 4 arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 r = -EINVAL;
418 goto bad1;
419 }
420
421 origin_path = argv[0];
422 cow_path = argv[1];
423 persistent = toupper(*argv[2]);
424
425 if (persistent != 'P' && persistent != 'N') {
426 ti->error = "Persistent flag is not P or N";
427 r = -EINVAL;
428 goto bad1;
429 }
430
431 chunk_size = simple_strtoul(argv[3], &value, 10);
432 if (chunk_size == 0 || value == NULL) {
433 ti->error = "Invalid chunk size";
434 r = -EINVAL;
435 goto bad1;
436 }
437
438 s = kmalloc(sizeof(*s), GFP_KERNEL);
439 if (s == NULL) {
440 ti->error = "Cannot allocate snapshot context private "
441 "structure";
442 r = -ENOMEM;
443 goto bad1;
444 }
445
446 r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin);
447 if (r) {
448 ti->error = "Cannot get origin device";
449 goto bad2;
450 }
451
452 r = dm_get_device(ti, cow_path, 0, 0,
453 FMODE_READ | FMODE_WRITE, &s->cow);
454 if (r) {
455 dm_put_device(ti, s->origin);
456 ti->error = "Cannot get COW device";
457 goto bad2;
458 }
459
460 /*
461 * Chunk size must be multiple of page size. Silently
462 * round up if it's not.
463 */
464 chunk_size = round_up(chunk_size, PAGE_SIZE >> 9);
465
466 /* Validate the chunk size against the device block size */
467 blocksize = s->cow->bdev->bd_disk->queue->hardsect_size;
468 if (chunk_size % (blocksize >> 9)) {
469 ti->error = "Chunk size is not a multiple of device blocksize";
470 r = -EINVAL;
471 goto bad3;
472 }
473
474 /* Check chunk_size is a power of 2 */
475 if (chunk_size & (chunk_size - 1)) {
476 ti->error = "Chunk size is not a power of 2";
477 r = -EINVAL;
478 goto bad3;
479 }
480
481 s->chunk_size = chunk_size;
482 s->chunk_mask = chunk_size - 1;
483 s->type = persistent;
484 s->chunk_shift = ffs(chunk_size) - 1;
485
486 s->valid = 1;
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -0800487 s->active = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 s->last_percent = 0;
489 init_rwsem(&s->lock);
490 s->table = ti->table;
491
492 /* Allocate hash table for COW data */
493 if (init_hash_tables(s)) {
494 ti->error = "Unable to allocate hash table space";
495 r = -ENOMEM;
496 goto bad3;
497 }
498
499 /*
500 * Check the persistent flag - done here because we need the iobuf
501 * to check the LV header
502 */
503 s->store.snap = s;
504
505 if (persistent == 'P')
506 r = dm_create_persistent(&s->store, chunk_size);
507 else
508 r = dm_create_transient(&s->store, s, blocksize);
509
510 if (r) {
511 ti->error = "Couldn't create exception store";
512 r = -EINVAL;
513 goto bad4;
514 }
515
516 r = kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
517 if (r) {
518 ti->error = "Could not create kcopyd client";
519 goto bad5;
520 }
521
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -0800522 /* Metadata must only be loaded into one table at once */
523 read_snapshot_metadata(s);
524
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 /* Add snapshot to the list of snapshots for this origin */
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -0800526 /* Exceptions aren't triggered till snapshot_resume() is called */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 if (register_snapshot(s)) {
528 r = -EINVAL;
529 ti->error = "Cannot register snapshot origin";
530 goto bad6;
531 }
532
533 ti->private = s;
Alasdair G Kergonc51c2752006-06-26 00:27:18 -0700534 ti->split_io = s->chunk_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535
536 return 0;
537
538 bad6:
539 kcopyd_client_destroy(s->kcopyd_client);
540
541 bad5:
542 s->store.destroy(&s->store);
543
544 bad4:
545 exit_exception_table(&s->pending, pending_cache);
546 exit_exception_table(&s->complete, exception_cache);
547
548 bad3:
549 dm_put_device(ti, s->cow);
550 dm_put_device(ti, s->origin);
551
552 bad2:
553 kfree(s);
554
555 bad1:
556 return r;
557}
558
559static void snapshot_dtr(struct dm_target *ti)
560{
561 struct dm_snapshot *s = (struct dm_snapshot *) ti->private;
562
Alasdair G Kergon138728dc2006-03-27 01:17:50 -0800563 /* Prevent further origin writes from using this snapshot. */
564 /* After this returns there can be no new kcopyd jobs. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 unregister_snapshot(s);
566
Alasdair G Kergon138728dc2006-03-27 01:17:50 -0800567 kcopyd_client_destroy(s->kcopyd_client);
568
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 exit_exception_table(&s->pending, pending_cache);
570 exit_exception_table(&s->complete, exception_cache);
571
572 /* Deallocate memory used */
573 s->store.destroy(&s->store);
574
575 dm_put_device(ti, s->origin);
576 dm_put_device(ti, s->cow);
Alasdair G Kergon138728dc2006-03-27 01:17:50 -0800577
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 kfree(s);
579}
580
581/*
582 * Flush a list of buffers.
583 */
584static void flush_bios(struct bio *bio)
585{
586 struct bio *n;
587
588 while (bio) {
589 n = bio->bi_next;
590 bio->bi_next = NULL;
591 generic_make_request(bio);
592 bio = n;
593 }
594}
595
596/*
597 * Error a list of buffers.
598 */
599static void error_bios(struct bio *bio)
600{
601 struct bio *n;
602
603 while (bio) {
604 n = bio->bi_next;
605 bio->bi_next = NULL;
606 bio_io_error(bio, bio->bi_size);
607 bio = n;
608 }
609}
610
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800611static inline void error_snapshot_bios(struct pending_exception *pe)
612{
613 error_bios(bio_list_get(&pe->snapshot_bios));
614}
615
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616static struct bio *__flush_bios(struct pending_exception *pe)
617{
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -0800618 /*
619 * If this pe is involved in a write to the origin and
620 * it is the last sibling to complete then release
621 * the bios for the original write to the origin.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 */
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -0800623
624 if (pe->primary_pe &&
625 atomic_dec_and_test(&pe->primary_pe->sibling_count))
626 return bio_list_get(&pe->primary_pe->origin_bios);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
628 return NULL;
629}
630
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800631static void __invalidate_snapshot(struct dm_snapshot *s,
632 struct pending_exception *pe, int err)
633{
634 if (!s->valid)
635 return;
636
637 if (err == -EIO)
638 DMERR("Invalidating snapshot: Error reading/writing.");
639 else if (err == -ENOMEM)
640 DMERR("Invalidating snapshot: Unable to allocate exception.");
641
642 if (pe)
643 remove_exception(&pe->e);
644
645 if (s->store.drop_snapshot)
646 s->store.drop_snapshot(&s->store);
647
648 s->valid = 0;
649
650 dm_table_event(s->table);
651}
652
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653static void pending_complete(struct pending_exception *pe, int success)
654{
655 struct exception *e;
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -0800656 struct pending_exception *primary_pe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 struct dm_snapshot *s = pe->snap;
658 struct bio *flush = NULL;
659
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800660 if (!success) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 /* Read/write error - snapshot is unusable */
662 down_write(&s->lock);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800663 __invalidate_snapshot(s, pe, -EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 flush = __flush_bios(pe);
665 up_write(&s->lock);
666
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800667 error_snapshot_bios(pe);
668 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 }
670
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800671 e = alloc_exception();
672 if (!e) {
673 down_write(&s->lock);
674 __invalidate_snapshot(s, pe, -ENOMEM);
675 flush = __flush_bios(pe);
676 up_write(&s->lock);
677
678 error_snapshot_bios(pe);
679 goto out;
680 }
681 *e = pe->e;
682
683 /*
684 * Add a proper exception, and remove the
685 * in-flight exception from the list.
686 */
687 down_write(&s->lock);
688 if (!s->valid) {
689 flush = __flush_bios(pe);
690 up_write(&s->lock);
691
692 free_exception(e);
693
694 error_snapshot_bios(pe);
695 goto out;
696 }
697
698 insert_exception(&s->complete, e);
699 remove_exception(&pe->e);
700 flush = __flush_bios(pe);
701
702 up_write(&s->lock);
703
704 /* Submit any pending write bios */
705 flush_bios(bio_list_get(&pe->snapshot_bios));
706
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 out:
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -0800708 primary_pe = pe->primary_pe;
709
710 /*
711 * Free the pe if it's not linked to an origin write or if
712 * it's not itself a primary pe.
713 */
714 if (!primary_pe || primary_pe != pe)
715 free_pending_exception(pe);
716
717 /*
718 * Free the primary pe if nothing references it.
719 */
720 if (primary_pe && !atomic_read(&primary_pe->sibling_count))
721 free_pending_exception(primary_pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
723 if (flush)
724 flush_bios(flush);
725}
726
727static void commit_callback(void *context, int success)
728{
729 struct pending_exception *pe = (struct pending_exception *) context;
730 pending_complete(pe, success);
731}
732
733/*
734 * Called when the copy I/O has finished. kcopyd actually runs
735 * this code so don't block.
736 */
737static void copy_callback(int read_err, unsigned int write_err, void *context)
738{
739 struct pending_exception *pe = (struct pending_exception *) context;
740 struct dm_snapshot *s = pe->snap;
741
742 if (read_err || write_err)
743 pending_complete(pe, 0);
744
745 else
746 /* Update the metadata if we are persistent */
747 s->store.commit_exception(&s->store, &pe->e, commit_callback,
748 pe);
749}
750
751/*
752 * Dispatches the copy operation to kcopyd.
753 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800754static void start_copy(struct pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755{
756 struct dm_snapshot *s = pe->snap;
757 struct io_region src, dest;
758 struct block_device *bdev = s->origin->bdev;
759 sector_t dev_size;
760
761 dev_size = get_dev_size(bdev);
762
763 src.bdev = bdev;
764 src.sector = chunk_to_sector(s, pe->e.old_chunk);
765 src.count = min(s->chunk_size, dev_size - src.sector);
766
767 dest.bdev = s->cow->bdev;
768 dest.sector = chunk_to_sector(s, pe->e.new_chunk);
769 dest.count = src.count;
770
771 /* Hand over to kcopyd */
772 kcopyd_copy(s->kcopyd_client,
773 &src, 1, &dest, 0, copy_callback, pe);
774}
775
776/*
777 * Looks to see if this snapshot already has a pending exception
778 * for this chunk, otherwise it allocates a new one and inserts
779 * it into the pending table.
780 *
781 * NOTE: a write lock must be held on snap->lock before calling
782 * this.
783 */
784static struct pending_exception *
785__find_pending_exception(struct dm_snapshot *s, struct bio *bio)
786{
787 struct exception *e;
788 struct pending_exception *pe;
789 chunk_t chunk = sector_to_chunk(s, bio->bi_sector);
790
791 /*
792 * Is there a pending exception for this already ?
793 */
794 e = lookup_exception(&s->pending, chunk);
795 if (e) {
796 /* cast the exception to a pending exception */
797 pe = container_of(e, struct pending_exception, e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800798 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 }
800
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800801 /*
802 * Create a new pending exception, we don't want
803 * to hold the lock while we do this.
804 */
805 up_write(&s->lock);
806 pe = alloc_pending_exception();
807 down_write(&s->lock);
808
809 if (!s->valid) {
810 free_pending_exception(pe);
811 return NULL;
812 }
813
814 e = lookup_exception(&s->pending, chunk);
815 if (e) {
816 free_pending_exception(pe);
817 pe = container_of(e, struct pending_exception, e);
818 goto out;
819 }
820
821 pe->e.old_chunk = chunk;
822 bio_list_init(&pe->origin_bios);
823 bio_list_init(&pe->snapshot_bios);
824 pe->primary_pe = NULL;
825 atomic_set(&pe->sibling_count, 1);
826 pe->snap = s;
827 pe->started = 0;
828
829 if (s->store.prepare_exception(&s->store, &pe->e)) {
830 free_pending_exception(pe);
831 return NULL;
832 }
833
834 insert_exception(&s->pending, &pe->e);
835
836 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 return pe;
838}
839
840static inline void remap_exception(struct dm_snapshot *s, struct exception *e,
841 struct bio *bio)
842{
843 bio->bi_bdev = s->cow->bdev;
844 bio->bi_sector = chunk_to_sector(s, e->new_chunk) +
845 (bio->bi_sector & s->chunk_mask);
846}
847
848static int snapshot_map(struct dm_target *ti, struct bio *bio,
849 union map_info *map_context)
850{
851 struct exception *e;
852 struct dm_snapshot *s = (struct dm_snapshot *) ti->private;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800853 int copy_needed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 int r = 1;
855 chunk_t chunk;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800856 struct pending_exception *pe = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
858 chunk = sector_to_chunk(s, bio->bi_sector);
859
860 /* Full snapshots are not usable */
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800861 /* To get here the table must be live so s->active is always set. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 if (!s->valid)
Alasdair G Kergonf6a80ea2005-07-12 15:53:01 -0700863 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
Alasdair G Kergon4aac0a62006-02-01 03:04:55 -0800865 if (unlikely(bio_barrier(bio)))
866 return -EOPNOTSUPP;
867
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 /*
869 * Write to snapshot - higher level takes care of RW/RO
870 * flags so we should only get this if we are
871 * writeable.
872 */
873 if (bio_rw(bio) == WRITE) {
874
875 /* FIXME: should only take write lock if we need
876 * to copy an exception */
877 down_write(&s->lock);
878
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800879 if (!s->valid) {
880 r = -EIO;
881 goto out_unlock;
882 }
883
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 /* If the block is already remapped - use that, else remap it */
885 e = lookup_exception(&s->complete, chunk);
886 if (e) {
887 remap_exception(s, e, bio);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800888 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 }
890
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800891 pe = __find_pending_exception(s, bio);
892 if (!pe) {
893 __invalidate_snapshot(s, pe, -ENOMEM);
894 r = -EIO;
895 goto out_unlock;
896 }
897
898 remap_exception(s, &pe->e, bio);
899 bio_list_add(&pe->snapshot_bios, bio);
900
901 if (!pe->started) {
902 /* this is protected by snap->lock */
903 pe->started = 1;
904 copy_needed = 1;
905 }
906
907 r = 0;
908
909 out_unlock:
910 up_write(&s->lock);
911
912 if (copy_needed)
913 start_copy(pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 } else {
915 /*
916 * FIXME: this read path scares me because we
917 * always use the origin when we have a pending
918 * exception. However I can't think of a
919 * situation where this is wrong - ejt.
920 */
921
922 /* Do reads */
923 down_read(&s->lock);
924
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800925 if (!s->valid) {
926 up_read(&s->lock);
927 return -EIO;
928 }
929
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 /* See if it it has been remapped */
931 e = lookup_exception(&s->complete, chunk);
932 if (e)
933 remap_exception(s, e, bio);
934 else
935 bio->bi_bdev = s->origin->bdev;
936
937 up_read(&s->lock);
938 }
939
940 return r;
941}
942
943static void snapshot_resume(struct dm_target *ti)
944{
945 struct dm_snapshot *s = (struct dm_snapshot *) ti->private;
946
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -0800947 down_write(&s->lock);
948 s->active = 1;
949 up_write(&s->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950}
951
952static int snapshot_status(struct dm_target *ti, status_type_t type,
953 char *result, unsigned int maxlen)
954{
955 struct dm_snapshot *snap = (struct dm_snapshot *) ti->private;
956
957 switch (type) {
958 case STATUSTYPE_INFO:
959 if (!snap->valid)
960 snprintf(result, maxlen, "Invalid");
961 else {
962 if (snap->store.fraction_full) {
963 sector_t numerator, denominator;
964 snap->store.fraction_full(&snap->store,
965 &numerator,
966 &denominator);
Andrew Morton4ee218c2006-03-27 01:17:48 -0800967 snprintf(result, maxlen, "%llu/%llu",
968 (unsigned long long)numerator,
969 (unsigned long long)denominator);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 }
971 else
972 snprintf(result, maxlen, "Unknown");
973 }
974 break;
975
976 case STATUSTYPE_TABLE:
977 /*
978 * kdevname returns a static pointer so we need
979 * to make private copies if the output is to
980 * make sense.
981 */
Andrew Morton4ee218c2006-03-27 01:17:48 -0800982 snprintf(result, maxlen, "%s %s %c %llu",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 snap->origin->name, snap->cow->name,
Andrew Morton4ee218c2006-03-27 01:17:48 -0800984 snap->type,
985 (unsigned long long)snap->chunk_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 break;
987 }
988
989 return 0;
990}
991
992/*-----------------------------------------------------------------
993 * Origin methods
994 *---------------------------------------------------------------*/
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995static int __origin_write(struct list_head *snapshots, struct bio *bio)
996{
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -0800997 int r = 1, first = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 struct dm_snapshot *snap;
999 struct exception *e;
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001000 struct pending_exception *pe, *next_pe, *primary_pe = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 chunk_t chunk;
Alasdair G Kergoneccf0812006-03-27 01:17:42 -08001002 LIST_HEAD(pe_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003
1004 /* Do all the snapshots on this origin */
1005 list_for_each_entry (snap, snapshots, list) {
1006
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001007 down_write(&snap->lock);
1008
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001009 /* Only deal with valid and active snapshots */
1010 if (!snap->valid || !snap->active)
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001011 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012
Alasdair G Kergond5e404c2005-07-12 15:53:05 -07001013 /* Nothing to do if writing beyond end of snapshot */
1014 if (bio->bi_sector >= dm_table_get_size(snap->table))
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001015 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
1017 /*
1018 * Remember, different snapshots can have
1019 * different chunk sizes.
1020 */
1021 chunk = sector_to_chunk(snap, bio->bi_sector);
1022
1023 /*
1024 * Check exception table to see if block
1025 * is already remapped in this snapshot
1026 * and trigger an exception if not.
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001027 *
1028 * sibling_count is initialised to 1 so pending_complete()
1029 * won't destroy the primary_pe while we're inside this loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 */
1031 e = lookup_exception(&snap->complete, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001032 if (e)
1033 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001035 pe = __find_pending_exception(snap, bio);
1036 if (!pe) {
1037 __invalidate_snapshot(snap, pe, ENOMEM);
1038 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 }
1040
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001041 if (!primary_pe) {
1042 /*
1043 * Either every pe here has same
1044 * primary_pe or none has one yet.
1045 */
1046 if (pe->primary_pe)
1047 primary_pe = pe->primary_pe;
1048 else {
1049 primary_pe = pe;
1050 first = 1;
1051 }
1052
1053 bio_list_add(&primary_pe->origin_bios, bio);
1054
1055 r = 0;
1056 }
1057
1058 if (!pe->primary_pe) {
1059 atomic_inc(&primary_pe->sibling_count);
1060 pe->primary_pe = primary_pe;
1061 }
1062
1063 if (!pe->started) {
1064 pe->started = 1;
1065 list_add_tail(&pe->list, &pe_queue);
1066 }
1067
1068 next_snapshot:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 up_write(&snap->lock);
1070 }
1071
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001072 if (!primary_pe)
1073 goto out;
1074
1075 /*
1076 * If this is the first time we're processing this chunk and
1077 * sibling_count is now 1 it means all the pending exceptions
1078 * got completed while we were in the loop above, so it falls to
1079 * us here to remove the primary_pe and submit any origin_bios.
1080 */
1081
1082 if (first && atomic_dec_and_test(&primary_pe->sibling_count)) {
1083 flush_bios(bio_list_get(&primary_pe->origin_bios));
1084 free_pending_exception(primary_pe);
1085 /* If we got here, pe_queue is necessarily empty. */
1086 goto out;
1087 }
1088
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 /*
1090 * Now that we have a complete pe list we can start the copying.
1091 */
Alasdair G Kergoneccf0812006-03-27 01:17:42 -08001092 list_for_each_entry_safe(pe, next_pe, &pe_queue, list)
1093 start_copy(pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001095 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 return r;
1097}
1098
1099/*
1100 * Called on a write from the origin driver.
1101 */
1102static int do_origin(struct dm_dev *origin, struct bio *bio)
1103{
1104 struct origin *o;
1105 int r = 1;
1106
1107 down_read(&_origins_lock);
1108 o = __lookup_origin(origin->bdev);
1109 if (o)
1110 r = __origin_write(&o->snapshots, bio);
1111 up_read(&_origins_lock);
1112
1113 return r;
1114}
1115
1116/*
1117 * Origin: maps a linear range of a device, with hooks for snapshotting.
1118 */
1119
1120/*
1121 * Construct an origin mapping: <dev_path>
1122 * The context for an origin is merely a 'struct dm_dev *'
1123 * pointing to the real device.
1124 */
1125static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1126{
1127 int r;
1128 struct dm_dev *dev;
1129
1130 if (argc != 1) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001131 ti->error = "origin: incorrect number of arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 return -EINVAL;
1133 }
1134
1135 r = dm_get_device(ti, argv[0], 0, ti->len,
1136 dm_table_get_mode(ti->table), &dev);
1137 if (r) {
1138 ti->error = "Cannot get target device";
1139 return r;
1140 }
1141
1142 ti->private = dev;
1143 return 0;
1144}
1145
1146static void origin_dtr(struct dm_target *ti)
1147{
1148 struct dm_dev *dev = (struct dm_dev *) ti->private;
1149 dm_put_device(ti, dev);
1150}
1151
1152static int origin_map(struct dm_target *ti, struct bio *bio,
1153 union map_info *map_context)
1154{
1155 struct dm_dev *dev = (struct dm_dev *) ti->private;
1156 bio->bi_bdev = dev->bdev;
1157
Alasdair G Kergon4aac0a62006-02-01 03:04:55 -08001158 if (unlikely(bio_barrier(bio)))
1159 return -EOPNOTSUPP;
1160
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 /* Only tell snapshots if this is a write */
1162 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : 1;
1163}
1164
1165#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
1166
1167/*
1168 * Set the target "split_io" field to the minimum of all the snapshots'
1169 * chunk sizes.
1170 */
1171static void origin_resume(struct dm_target *ti)
1172{
1173 struct dm_dev *dev = (struct dm_dev *) ti->private;
1174 struct dm_snapshot *snap;
1175 struct origin *o;
1176 chunk_t chunk_size = 0;
1177
1178 down_read(&_origins_lock);
1179 o = __lookup_origin(dev->bdev);
1180 if (o)
1181 list_for_each_entry (snap, &o->snapshots, list)
1182 chunk_size = min_not_zero(chunk_size, snap->chunk_size);
1183 up_read(&_origins_lock);
1184
1185 ti->split_io = chunk_size;
1186}
1187
1188static int origin_status(struct dm_target *ti, status_type_t type, char *result,
1189 unsigned int maxlen)
1190{
1191 struct dm_dev *dev = (struct dm_dev *) ti->private;
1192
1193 switch (type) {
1194 case STATUSTYPE_INFO:
1195 result[0] = '\0';
1196 break;
1197
1198 case STATUSTYPE_TABLE:
1199 snprintf(result, maxlen, "%s", dev->name);
1200 break;
1201 }
1202
1203 return 0;
1204}
1205
1206static struct target_type origin_target = {
1207 .name = "snapshot-origin",
Alasdair G Kergonc51c2752006-06-26 00:27:18 -07001208 .version = {1, 4, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 .module = THIS_MODULE,
1210 .ctr = origin_ctr,
1211 .dtr = origin_dtr,
1212 .map = origin_map,
1213 .resume = origin_resume,
1214 .status = origin_status,
1215};
1216
1217static struct target_type snapshot_target = {
1218 .name = "snapshot",
Alasdair G Kergonc51c2752006-06-26 00:27:18 -07001219 .version = {1, 4, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 .module = THIS_MODULE,
1221 .ctr = snapshot_ctr,
1222 .dtr = snapshot_dtr,
1223 .map = snapshot_map,
1224 .resume = snapshot_resume,
1225 .status = snapshot_status,
1226};
1227
1228static int __init dm_snapshot_init(void)
1229{
1230 int r;
1231
1232 r = dm_register_target(&snapshot_target);
1233 if (r) {
1234 DMERR("snapshot target register failed %d", r);
1235 return r;
1236 }
1237
1238 r = dm_register_target(&origin_target);
1239 if (r < 0) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001240 DMERR("Origin target register failed %d", r);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 goto bad1;
1242 }
1243
1244 r = init_origin_hash();
1245 if (r) {
1246 DMERR("init_origin_hash failed.");
1247 goto bad2;
1248 }
1249
1250 exception_cache = kmem_cache_create("dm-snapshot-ex",
1251 sizeof(struct exception),
1252 __alignof__(struct exception),
1253 0, NULL, NULL);
1254 if (!exception_cache) {
1255 DMERR("Couldn't create exception cache.");
1256 r = -ENOMEM;
1257 goto bad3;
1258 }
1259
1260 pending_cache =
1261 kmem_cache_create("dm-snapshot-in",
1262 sizeof(struct pending_exception),
1263 __alignof__(struct pending_exception),
1264 0, NULL, NULL);
1265 if (!pending_cache) {
1266 DMERR("Couldn't create pending cache.");
1267 r = -ENOMEM;
1268 goto bad4;
1269 }
1270
Matthew Dobson93d23412006-03-26 01:37:50 -08001271 pending_pool = mempool_create_slab_pool(128, pending_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 if (!pending_pool) {
1273 DMERR("Couldn't create pending pool.");
1274 r = -ENOMEM;
1275 goto bad5;
1276 }
1277
1278 return 0;
1279
1280 bad5:
1281 kmem_cache_destroy(pending_cache);
1282 bad4:
1283 kmem_cache_destroy(exception_cache);
1284 bad3:
1285 exit_origin_hash();
1286 bad2:
1287 dm_unregister_target(&origin_target);
1288 bad1:
1289 dm_unregister_target(&snapshot_target);
1290 return r;
1291}
1292
1293static void __exit dm_snapshot_exit(void)
1294{
1295 int r;
1296
1297 r = dm_unregister_target(&snapshot_target);
1298 if (r)
1299 DMERR("snapshot unregister failed %d", r);
1300
1301 r = dm_unregister_target(&origin_target);
1302 if (r)
1303 DMERR("origin unregister failed %d", r);
1304
1305 exit_origin_hash();
1306 mempool_destroy(pending_pool);
1307 kmem_cache_destroy(pending_cache);
1308 kmem_cache_destroy(exception_cache);
1309}
1310
1311/* Module hooks */
1312module_init(dm_snapshot_init);
1313module_exit(dm_snapshot_exit);
1314
1315MODULE_DESCRIPTION(DM_NAME " snapshot target");
1316MODULE_AUTHOR("Joe Thornber");
1317MODULE_LICENSE("GPL");