blob: 65ff82ff124e29a3c7cfedf276df89646fcbc2ac [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>
Mikulas Patocka90fa1522009-01-06 03:04:54 +000012#include <linux/delay.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#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>
vignesh babu6f3c3f02007-10-19 22:38:44 +010021#include <linux/log2.h>
Alasdair G Kergona765e202008-04-24 22:02:01 +010022#include <linux/dm-kcopyd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
Jonathan Brassowaea53d92009-01-06 03:05:15 +000024#include "dm-exception-store.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "dm-snap.h"
26#include "dm-bio-list.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
Alasdair G Kergon72d94862006-06-26 00:27:35 -070028#define DM_MSG_PREFIX "snapshots"
29
Linus Torvalds1da177e2005-04-16 15:20:36 -070030/*
31 * The percentage increment we will wake up users at
32 */
33#define WAKE_UP_PERCENT 5
34
35/*
36 * kcopyd priority of snapshot operations
37 */
38#define SNAPSHOT_COPY_PRIORITY 2
39
40/*
Milan Broz8ee27672008-04-24 21:42:36 +010041 * Reserve 1MB for each snapshot initially (with minimum of 1 page).
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 */
Milan Broz8ee27672008-04-24 21:42:36 +010043#define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
Mikulas Patockacd45daf2008-07-21 12:00:32 +010045/*
46 * The size of the mempool used to track chunks in use.
47 */
48#define MIN_IOS 256
49
Adrian Bunkc642f9e2006-12-08 02:41:13 -080050static struct workqueue_struct *ksnapd;
David Howellsc4028952006-11-22 14:57:56 +000051static void flush_queued_bios(struct work_struct *work);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -070052
Alasdair G Kergon028867a2007-07-12 17:26:32 +010053struct dm_snap_pending_exception {
54 struct dm_snap_exception e;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
56 /*
57 * Origin buffers waiting for this to complete are held
58 * in a bio list
59 */
60 struct bio_list origin_bios;
61 struct bio_list snapshot_bios;
62
63 /*
Alasdair G Kergoneccf0812006-03-27 01:17:42 -080064 * Short-term queue of pending exceptions prior to submission.
65 */
66 struct list_head list;
67
68 /*
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -080069 * The primary pending_exception is the one that holds
Alasdair G Kergon4b832e82006-10-03 01:15:30 -070070 * the ref_count and the list of origin_bios for a
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -080071 * group of pending_exceptions. It is always last to get freed.
72 * These fields get set up when writing to the origin.
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +010074 struct dm_snap_pending_exception *primary_pe;
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -080075
76 /*
77 * Number of pending_exceptions processing this chunk.
78 * When this drops to zero we must complete the origin bios.
79 * If incrementing or decrementing this, hold pe->snap->lock for
80 * the sibling concerned and not pe->primary_pe->snap->lock unless
81 * they are the same.
82 */
Alasdair G Kergon4b832e82006-10-03 01:15:30 -070083 atomic_t ref_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
85 /* Pointer back to snapshot context */
86 struct dm_snapshot *snap;
87
88 /*
89 * 1 indicates the exception has already been sent to
90 * kcopyd.
91 */
92 int started;
93};
94
95/*
96 * Hash table mapping origin volumes to lists of snapshots and
97 * a lock to protect it
98 */
Christoph Lametere18b8902006-12-06 20:33:20 -080099static struct kmem_cache *exception_cache;
100static struct kmem_cache *pending_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100102struct dm_snap_tracked_chunk {
103 struct hlist_node node;
104 chunk_t chunk;
105};
106
107static struct kmem_cache *tracked_chunk_cache;
108
109static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s,
110 chunk_t chunk)
111{
112 struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool,
113 GFP_NOIO);
114 unsigned long flags;
115
116 c->chunk = chunk;
117
118 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
119 hlist_add_head(&c->node,
120 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
121 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
122
123 return c;
124}
125
126static void stop_tracking_chunk(struct dm_snapshot *s,
127 struct dm_snap_tracked_chunk *c)
128{
129 unsigned long flags;
130
131 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
132 hlist_del(&c->node);
133 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
134
135 mempool_free(c, s->tracked_chunk_pool);
136}
137
Mikulas Patockaa8d41b52008-07-21 12:00:34 +0100138static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
139{
140 struct dm_snap_tracked_chunk *c;
141 struct hlist_node *hn;
142 int found = 0;
143
144 spin_lock_irq(&s->tracked_chunk_lock);
145
146 hlist_for_each_entry(c, hn,
147 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
148 if (c->chunk == chunk) {
149 found = 1;
150 break;
151 }
152 }
153
154 spin_unlock_irq(&s->tracked_chunk_lock);
155
156 return found;
157}
158
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159/*
160 * One of these per registered origin, held in the snapshot_origins hash
161 */
162struct origin {
163 /* The origin device */
164 struct block_device *bdev;
165
166 struct list_head hash_list;
167
168 /* List of snapshots for this origin */
169 struct list_head snapshots;
170};
171
172/*
173 * Size of the hash table for origin volumes. If we make this
174 * the size of the minors list then it should be nearly perfect
175 */
176#define ORIGIN_HASH_SIZE 256
177#define ORIGIN_MASK 0xFF
178static struct list_head *_origins;
179static struct rw_semaphore _origins_lock;
180
181static int init_origin_hash(void)
182{
183 int i;
184
185 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
186 GFP_KERNEL);
187 if (!_origins) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700188 DMERR("unable to allocate memory");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 return -ENOMEM;
190 }
191
192 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
193 INIT_LIST_HEAD(_origins + i);
194 init_rwsem(&_origins_lock);
195
196 return 0;
197}
198
199static void exit_origin_hash(void)
200{
201 kfree(_origins);
202}
203
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100204static unsigned origin_hash(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205{
206 return bdev->bd_dev & ORIGIN_MASK;
207}
208
209static struct origin *__lookup_origin(struct block_device *origin)
210{
211 struct list_head *ol;
212 struct origin *o;
213
214 ol = &_origins[origin_hash(origin)];
215 list_for_each_entry (o, ol, hash_list)
216 if (bdev_equal(o->bdev, origin))
217 return o;
218
219 return NULL;
220}
221
222static void __insert_origin(struct origin *o)
223{
224 struct list_head *sl = &_origins[origin_hash(o->bdev)];
225 list_add_tail(&o->hash_list, sl);
226}
227
228/*
229 * Make a note of the snapshot and its origin so we can look it
230 * up when the origin has a write on it.
231 */
232static int register_snapshot(struct dm_snapshot *snap)
233{
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000234 struct origin *o, *new_o;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 struct block_device *bdev = snap->origin->bdev;
236
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000237 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
238 if (!new_o)
239 return -ENOMEM;
240
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 down_write(&_origins_lock);
242 o = __lookup_origin(bdev);
243
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000244 if (o)
245 kfree(new_o);
246 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 /* New origin */
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000248 o = new_o;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
250 /* Initialise the struct */
251 INIT_LIST_HEAD(&o->snapshots);
252 o->bdev = bdev;
253
254 __insert_origin(o);
255 }
256
257 list_add_tail(&snap->list, &o->snapshots);
258
259 up_write(&_origins_lock);
260 return 0;
261}
262
263static void unregister_snapshot(struct dm_snapshot *s)
264{
265 struct origin *o;
266
267 down_write(&_origins_lock);
268 o = __lookup_origin(s->origin->bdev);
269
270 list_del(&s->list);
271 if (list_empty(&o->snapshots)) {
272 list_del(&o->hash_list);
273 kfree(o);
274 }
275
276 up_write(&_origins_lock);
277}
278
279/*
280 * Implementation of the exception hash tables.
Milan Brozd74f81f2008-02-08 02:11:27 +0000281 * The lowest hash_shift bits of the chunk number are ignored, allowing
282 * some consecutive chunks to be grouped together.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 */
Milan Brozd74f81f2008-02-08 02:11:27 +0000284static int init_exception_table(struct exception_table *et, uint32_t size,
285 unsigned hash_shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286{
287 unsigned int i;
288
Milan Brozd74f81f2008-02-08 02:11:27 +0000289 et->hash_shift = hash_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 et->hash_mask = size - 1;
291 et->table = dm_vcalloc(size, sizeof(struct list_head));
292 if (!et->table)
293 return -ENOMEM;
294
295 for (i = 0; i < size; i++)
296 INIT_LIST_HEAD(et->table + i);
297
298 return 0;
299}
300
Christoph Lametere18b8902006-12-06 20:33:20 -0800301static void exit_exception_table(struct exception_table *et, struct kmem_cache *mem)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
303 struct list_head *slot;
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100304 struct dm_snap_exception *ex, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 int i, size;
306
307 size = et->hash_mask + 1;
308 for (i = 0; i < size; i++) {
309 slot = et->table + i;
310
311 list_for_each_entry_safe (ex, next, slot, hash_list)
312 kmem_cache_free(mem, ex);
313 }
314
315 vfree(et->table);
316}
317
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100318static uint32_t exception_hash(struct exception_table *et, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
Milan Brozd74f81f2008-02-08 02:11:27 +0000320 return (chunk >> et->hash_shift) & et->hash_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321}
322
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100323static void insert_exception(struct exception_table *eh,
324 struct dm_snap_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
326 struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)];
327 list_add(&e->hash_list, l);
328}
329
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100330static void remove_exception(struct dm_snap_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331{
332 list_del(&e->hash_list);
333}
334
335/*
336 * Return the exception data for a sector, or NULL if not
337 * remapped.
338 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100339static struct dm_snap_exception *lookup_exception(struct exception_table *et,
340 chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341{
342 struct list_head *slot;
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100343 struct dm_snap_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344
345 slot = &et->table[exception_hash(et, chunk)];
346 list_for_each_entry (e, slot, hash_list)
Milan Brozd74f81f2008-02-08 02:11:27 +0000347 if (chunk >= e->old_chunk &&
348 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 return e;
350
351 return NULL;
352}
353
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100354static struct dm_snap_exception *alloc_exception(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355{
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100356 struct dm_snap_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357
358 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
359 if (!e)
360 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
361
362 return e;
363}
364
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100365static void free_exception(struct dm_snap_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366{
367 kmem_cache_free(exception_cache, e);
368}
369
Mikulas Patocka92e86812008-07-21 12:00:35 +0100370static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371{
Mikulas Patocka92e86812008-07-21 12:00:35 +0100372 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
373 GFP_NOIO);
374
Mikulas Patocka879129d2008-10-30 13:33:16 +0000375 atomic_inc(&s->pending_exceptions_count);
Mikulas Patocka92e86812008-07-21 12:00:35 +0100376 pe->snap = s;
377
378 return pe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379}
380
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100381static void free_pending_exception(struct dm_snap_pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382{
Mikulas Patocka879129d2008-10-30 13:33:16 +0000383 struct dm_snapshot *s = pe->snap;
384
385 mempool_free(pe, s->pending_pool);
386 smp_mb__before_atomic_dec();
387 atomic_dec(&s->pending_exceptions_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388}
389
Milan Brozd74f81f2008-02-08 02:11:27 +0000390static void insert_completed_exception(struct dm_snapshot *s,
391 struct dm_snap_exception *new_e)
392{
393 struct exception_table *eh = &s->complete;
394 struct list_head *l;
395 struct dm_snap_exception *e = NULL;
396
397 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
398
399 /* Add immediately if this table doesn't support consecutive chunks */
400 if (!eh->hash_shift)
401 goto out;
402
403 /* List is ordered by old_chunk */
404 list_for_each_entry_reverse(e, l, hash_list) {
405 /* Insert after an existing chunk? */
406 if (new_e->old_chunk == (e->old_chunk +
407 dm_consecutive_chunk_count(e) + 1) &&
408 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
409 dm_consecutive_chunk_count(e) + 1)) {
410 dm_consecutive_chunk_count_inc(e);
411 free_exception(new_e);
412 return;
413 }
414
415 /* Insert before an existing chunk? */
416 if (new_e->old_chunk == (e->old_chunk - 1) &&
417 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
418 dm_consecutive_chunk_count_inc(e);
419 e->old_chunk--;
420 e->new_chunk--;
421 free_exception(new_e);
422 return;
423 }
424
425 if (new_e->old_chunk > e->old_chunk)
426 break;
427 }
428
429out:
430 list_add(&new_e->hash_list, e ? &e->hash_list : l);
431}
432
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000433/*
434 * Callback used by the exception stores to load exceptions when
435 * initialising.
436 */
437static int dm_add_exception(void *context, chunk_t old, chunk_t new)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438{
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000439 struct dm_snapshot *s = context;
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100440 struct dm_snap_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441
442 e = alloc_exception();
443 if (!e)
444 return -ENOMEM;
445
446 e->old_chunk = old;
Milan Brozd74f81f2008-02-08 02:11:27 +0000447
448 /* Consecutive_count is implicitly initialised to zero */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 e->new_chunk = new;
Milan Brozd74f81f2008-02-08 02:11:27 +0000450
451 insert_completed_exception(s, e);
452
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 return 0;
454}
455
456/*
457 * Hard coded magic.
458 */
459static int calc_max_buckets(void)
460{
461 /* use a fixed size of 2MB */
462 unsigned long mem = 2 * 1024 * 1024;
463 mem /= sizeof(struct list_head);
464
465 return mem;
466}
467
468/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 * Allocate room for a suitable hash table.
470 */
471static int init_hash_tables(struct dm_snapshot *s)
472{
473 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
474
475 /*
476 * Calculate based on the size of the original volume or
477 * the COW volume...
478 */
479 cow_dev_size = get_dev_size(s->cow->bdev);
480 origin_dev_size = get_dev_size(s->origin->bdev);
481 max_buckets = calc_max_buckets();
482
483 hash_size = min(origin_dev_size, cow_dev_size) >> s->chunk_shift;
484 hash_size = min(hash_size, max_buckets);
485
Robert P. J. Day8defd832008-02-08 02:10:06 +0000486 hash_size = rounddown_pow_of_two(hash_size);
Milan Brozd74f81f2008-02-08 02:11:27 +0000487 if (init_exception_table(&s->complete, hash_size,
488 DM_CHUNK_CONSECUTIVE_BITS))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 return -ENOMEM;
490
491 /*
492 * Allocate hash table for in-flight exceptions
493 * Make this smaller than the real hash table
494 */
495 hash_size >>= 3;
496 if (hash_size < 64)
497 hash_size = 64;
498
Milan Brozd74f81f2008-02-08 02:11:27 +0000499 if (init_exception_table(&s->pending, hash_size, 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 exit_exception_table(&s->complete, exception_cache);
501 return -ENOMEM;
502 }
503
504 return 0;
505}
506
507/*
508 * Round a number up to the nearest 'size' boundary. size must
509 * be a power of 2.
510 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100511static ulong round_up(ulong n, ulong size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512{
513 size--;
514 return (n + size) & ~size;
515}
516
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -0700517static int set_chunk_size(struct dm_snapshot *s, const char *chunk_size_arg,
518 char **error)
519{
520 unsigned long chunk_size;
521 char *value;
522
523 chunk_size = simple_strtoul(chunk_size_arg, &value, 10);
524 if (*chunk_size_arg == '\0' || *value != '\0') {
525 *error = "Invalid chunk size";
526 return -EINVAL;
527 }
528
529 if (!chunk_size) {
530 s->chunk_size = s->chunk_mask = s->chunk_shift = 0;
531 return 0;
532 }
533
534 /*
535 * Chunk size must be multiple of page size. Silently
536 * round up if it's not.
537 */
538 chunk_size = round_up(chunk_size, PAGE_SIZE >> 9);
539
540 /* Check chunk_size is a power of 2 */
vignesh babu6f3c3f02007-10-19 22:38:44 +0100541 if (!is_power_of_2(chunk_size)) {
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -0700542 *error = "Chunk size is not a power of 2";
543 return -EINVAL;
544 }
545
546 /* Validate the chunk size against the device block size */
547 if (chunk_size % (bdev_hardsect_size(s->cow->bdev) >> 9)) {
548 *error = "Chunk size is not a multiple of device blocksize";
549 return -EINVAL;
550 }
551
552 s->chunk_size = chunk_size;
553 s->chunk_mask = chunk_size - 1;
554 s->chunk_shift = ffs(chunk_size) - 1;
555
556 return 0;
557}
558
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559/*
560 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
561 */
562static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
563{
564 struct dm_snapshot *s;
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100565 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 int r = -EINVAL;
567 char persistent;
568 char *origin_path;
569 char *cow_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -0700571 if (argc != 4) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700572 ti->error = "requires exactly 4 arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 r = -EINVAL;
574 goto bad1;
575 }
576
577 origin_path = argv[0];
578 cow_path = argv[1];
579 persistent = toupper(*argv[2]);
580
581 if (persistent != 'P' && persistent != 'N') {
582 ti->error = "Persistent flag is not P or N";
583 r = -EINVAL;
584 goto bad1;
585 }
586
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 s = kmalloc(sizeof(*s), GFP_KERNEL);
588 if (s == NULL) {
589 ti->error = "Cannot allocate snapshot context private "
590 "structure";
591 r = -ENOMEM;
592 goto bad1;
593 }
594
595 r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin);
596 if (r) {
597 ti->error = "Cannot get origin device";
598 goto bad2;
599 }
600
601 r = dm_get_device(ti, cow_path, 0, 0,
602 FMODE_READ | FMODE_WRITE, &s->cow);
603 if (r) {
604 dm_put_device(ti, s->origin);
605 ti->error = "Cannot get COW device";
606 goto bad2;
607 }
608
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -0700609 r = set_chunk_size(s, argv[3], &ti->error);
610 if (r)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 goto bad3;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 s->type = persistent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
615 s->valid = 1;
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -0800616 s->active = 0;
Mikulas Patocka879129d2008-10-30 13:33:16 +0000617 atomic_set(&s->pending_exceptions_count, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 init_rwsem(&s->lock);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -0700619 spin_lock_init(&s->pe_lock);
Mikulas Patocka72727ba2008-04-24 21:43:11 +0100620 s->ti = ti;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
622 /* Allocate hash table for COW data */
623 if (init_hash_tables(s)) {
624 ti->error = "Unable to allocate hash table space";
625 r = -ENOMEM;
626 goto bad3;
627 }
628
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 s->store.snap = s;
630
631 if (persistent == 'P')
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -0700632 r = dm_create_persistent(&s->store);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 else
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -0700634 r = dm_create_transient(&s->store);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635
636 if (r) {
637 ti->error = "Couldn't create exception store";
638 r = -EINVAL;
639 goto bad4;
640 }
641
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100642 r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 if (r) {
644 ti->error = "Could not create kcopyd client";
645 goto bad5;
646 }
647
Mikulas Patocka92e86812008-07-21 12:00:35 +0100648 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
649 if (!s->pending_pool) {
650 ti->error = "Could not allocate mempool for pending exceptions";
651 goto bad6;
652 }
653
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100654 s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS,
655 tracked_chunk_cache);
656 if (!s->tracked_chunk_pool) {
657 ti->error = "Could not allocate tracked_chunk mempool for "
658 "tracking reads";
Mikulas Patocka92e86812008-07-21 12:00:35 +0100659 goto bad_tracked_chunk_pool;
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100660 }
661
662 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
663 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
664
665 spin_lock_init(&s->tracked_chunk_lock);
666
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -0800667 /* Metadata must only be loaded into one table at once */
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000668 r = s->store.read_metadata(&s->store, dm_add_exception, (void *)s);
Milan Broz07641472007-07-12 17:28:13 +0100669 if (r < 0) {
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -0700670 ti->error = "Failed to read snapshot metadata";
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100671 goto bad_load_and_register;
Milan Broz07641472007-07-12 17:28:13 +0100672 } else if (r > 0) {
673 s->valid = 0;
674 DMWARN("Snapshot is marked invalid.");
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -0700675 }
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -0800676
Alasdair G Kergonca3a9312006-10-03 01:15:30 -0700677 bio_list_init(&s->queued_bios);
David Howellsc4028952006-11-22 14:57:56 +0000678 INIT_WORK(&s->queued_bios_work, flush_queued_bios);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -0700679
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 /* Add snapshot to the list of snapshots for this origin */
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -0800681 /* Exceptions aren't triggered till snapshot_resume() is called */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 if (register_snapshot(s)) {
683 r = -EINVAL;
684 ti->error = "Cannot register snapshot origin";
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100685 goto bad_load_and_register;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 }
687
688 ti->private = s;
Alasdair G Kergonc51c2752006-06-26 00:27:18 -0700689 ti->split_io = s->chunk_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690
691 return 0;
692
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100693 bad_load_and_register:
694 mempool_destroy(s->tracked_chunk_pool);
695
Mikulas Patocka92e86812008-07-21 12:00:35 +0100696 bad_tracked_chunk_pool:
697 mempool_destroy(s->pending_pool);
698
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 bad6:
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100700 dm_kcopyd_client_destroy(s->kcopyd_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701
702 bad5:
703 s->store.destroy(&s->store);
704
705 bad4:
706 exit_exception_table(&s->pending, pending_cache);
707 exit_exception_table(&s->complete, exception_cache);
708
709 bad3:
710 dm_put_device(ti, s->cow);
711 dm_put_device(ti, s->origin);
712
713 bad2:
714 kfree(s);
715
716 bad1:
717 return r;
718}
719
Milan Broz31c93a02006-12-08 02:41:11 -0800720static void __free_exceptions(struct dm_snapshot *s)
721{
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100722 dm_kcopyd_client_destroy(s->kcopyd_client);
Milan Broz31c93a02006-12-08 02:41:11 -0800723 s->kcopyd_client = NULL;
724
725 exit_exception_table(&s->pending, pending_cache);
726 exit_exception_table(&s->complete, exception_cache);
727
728 s->store.destroy(&s->store);
729}
730
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731static void snapshot_dtr(struct dm_target *ti)
732{
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100733#ifdef CONFIG_DM_DEBUG
734 int i;
735#endif
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100736 struct dm_snapshot *s = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737
Alasdair G Kergonca3a9312006-10-03 01:15:30 -0700738 flush_workqueue(ksnapd);
739
Alasdair G Kergon138728d2006-03-27 01:17:50 -0800740 /* Prevent further origin writes from using this snapshot. */
741 /* After this returns there can be no new kcopyd jobs. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 unregister_snapshot(s);
743
Mikulas Patocka879129d2008-10-30 13:33:16 +0000744 while (atomic_read(&s->pending_exceptions_count))
Mikulas Patocka90fa1522009-01-06 03:04:54 +0000745 msleep(1);
Mikulas Patocka879129d2008-10-30 13:33:16 +0000746 /*
747 * Ensure instructions in mempool_destroy aren't reordered
748 * before atomic_read.
749 */
750 smp_mb();
751
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100752#ifdef CONFIG_DM_DEBUG
753 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
754 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
755#endif
756
757 mempool_destroy(s->tracked_chunk_pool);
758
Milan Broz31c93a02006-12-08 02:41:11 -0800759 __free_exceptions(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
Mikulas Patocka92e86812008-07-21 12:00:35 +0100761 mempool_destroy(s->pending_pool);
762
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 dm_put_device(ti, s->origin);
764 dm_put_device(ti, s->cow);
Alasdair G Kergon138728d2006-03-27 01:17:50 -0800765
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 kfree(s);
767}
768
769/*
770 * Flush a list of buffers.
771 */
772static void flush_bios(struct bio *bio)
773{
774 struct bio *n;
775
776 while (bio) {
777 n = bio->bi_next;
778 bio->bi_next = NULL;
779 generic_make_request(bio);
780 bio = n;
781 }
782}
783
David Howellsc4028952006-11-22 14:57:56 +0000784static void flush_queued_bios(struct work_struct *work)
Alasdair G Kergonca3a9312006-10-03 01:15:30 -0700785{
David Howellsc4028952006-11-22 14:57:56 +0000786 struct dm_snapshot *s =
787 container_of(work, struct dm_snapshot, queued_bios_work);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -0700788 struct bio *queued_bios;
789 unsigned long flags;
790
791 spin_lock_irqsave(&s->pe_lock, flags);
792 queued_bios = bio_list_get(&s->queued_bios);
793 spin_unlock_irqrestore(&s->pe_lock, flags);
794
795 flush_bios(queued_bios);
796}
797
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798/*
799 * Error a list of buffers.
800 */
801static void error_bios(struct bio *bio)
802{
803 struct bio *n;
804
805 while (bio) {
806 n = bio->bi_next;
807 bio->bi_next = NULL;
NeilBrown6712ecf2007-09-27 12:47:43 +0200808 bio_io_error(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 bio = n;
810 }
811}
812
Alasdair G Kergon695368a2006-10-03 01:15:31 -0700813static void __invalidate_snapshot(struct dm_snapshot *s, int err)
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800814{
815 if (!s->valid)
816 return;
817
818 if (err == -EIO)
819 DMERR("Invalidating snapshot: Error reading/writing.");
820 else if (err == -ENOMEM)
821 DMERR("Invalidating snapshot: Unable to allocate exception.");
822
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800823 if (s->store.drop_snapshot)
824 s->store.drop_snapshot(&s->store);
825
826 s->valid = 0;
827
Mikulas Patocka72727ba2008-04-24 21:43:11 +0100828 dm_table_event(s->ti->table);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800829}
830
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100831static void get_pending_exception(struct dm_snap_pending_exception *pe)
Alasdair G Kergon4b832e82006-10-03 01:15:30 -0700832{
833 atomic_inc(&pe->ref_count);
834}
835
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100836static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe)
Alasdair G Kergon4b832e82006-10-03 01:15:30 -0700837{
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100838 struct dm_snap_pending_exception *primary_pe;
Alasdair G Kergon4b832e82006-10-03 01:15:30 -0700839 struct bio *origin_bios = NULL;
840
841 primary_pe = pe->primary_pe;
842
843 /*
844 * If this pe is involved in a write to the origin and
845 * it is the last sibling to complete then release
846 * the bios for the original write to the origin.
847 */
848 if (primary_pe &&
Mikulas Patocka7c5f78b2008-10-21 17:44:51 +0100849 atomic_dec_and_test(&primary_pe->ref_count)) {
Alasdair G Kergon4b832e82006-10-03 01:15:30 -0700850 origin_bios = bio_list_get(&primary_pe->origin_bios);
Mikulas Patocka7c5f78b2008-10-21 17:44:51 +0100851 free_pending_exception(primary_pe);
852 }
Alasdair G Kergon4b832e82006-10-03 01:15:30 -0700853
854 /*
855 * Free the pe if it's not linked to an origin write or if
856 * it's not itself a primary pe.
857 */
858 if (!primary_pe || primary_pe != pe)
859 free_pending_exception(pe);
860
Alasdair G Kergon4b832e82006-10-03 01:15:30 -0700861 return origin_bios;
862}
863
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100864static void pending_complete(struct dm_snap_pending_exception *pe, int success)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865{
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100866 struct dm_snap_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 struct dm_snapshot *s = pe->snap;
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -0700868 struct bio *origin_bios = NULL;
869 struct bio *snapshot_bios = NULL;
870 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800872 if (!success) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 /* Read/write error - snapshot is unusable */
874 down_write(&s->lock);
Alasdair G Kergon695368a2006-10-03 01:15:31 -0700875 __invalidate_snapshot(s, -EIO);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -0700876 error = 1;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800877 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 }
879
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800880 e = alloc_exception();
881 if (!e) {
882 down_write(&s->lock);
Alasdair G Kergon695368a2006-10-03 01:15:31 -0700883 __invalidate_snapshot(s, -ENOMEM);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -0700884 error = 1;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800885 goto out;
886 }
887 *e = pe->e;
888
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -0700889 down_write(&s->lock);
890 if (!s->valid) {
891 free_exception(e);
892 error = 1;
893 goto out;
894 }
895
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800896 /*
Mikulas Patockaa8d41b52008-07-21 12:00:34 +0100897 * Check for conflicting reads. This is extremely improbable,
Mikulas Patocka90fa1522009-01-06 03:04:54 +0000898 * so msleep(1) is sufficient and there is no need for a wait queue.
Mikulas Patockaa8d41b52008-07-21 12:00:34 +0100899 */
900 while (__chunk_is_tracked(s, pe->e.old_chunk))
Mikulas Patocka90fa1522009-01-06 03:04:54 +0000901 msleep(1);
Mikulas Patockaa8d41b52008-07-21 12:00:34 +0100902
903 /*
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800904 * Add a proper exception, and remove the
905 * in-flight exception from the list.
906 */
Milan Brozd74f81f2008-02-08 02:11:27 +0000907 insert_completed_exception(s, e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800908
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 out:
Alasdair G Kergon695368a2006-10-03 01:15:31 -0700910 remove_exception(&pe->e);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -0700911 snapshot_bios = bio_list_get(&pe->snapshot_bios);
Alasdair G Kergon4b832e82006-10-03 01:15:30 -0700912 origin_bios = put_pending_exception(pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -0700914 up_write(&s->lock);
915
916 /* Submit any pending write bios */
917 if (error)
918 error_bios(snapshot_bios);
919 else
920 flush_bios(snapshot_bios);
921
922 flush_bios(origin_bios);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923}
924
925static void commit_callback(void *context, int success)
926{
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100927 struct dm_snap_pending_exception *pe = context;
928
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 pending_complete(pe, success);
930}
931
932/*
933 * Called when the copy I/O has finished. kcopyd actually runs
934 * this code so don't block.
935 */
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -0700936static void copy_callback(int read_err, unsigned long write_err, void *context)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937{
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100938 struct dm_snap_pending_exception *pe = context;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 struct dm_snapshot *s = pe->snap;
940
941 if (read_err || write_err)
942 pending_complete(pe, 0);
943
944 else
945 /* Update the metadata if we are persistent */
946 s->store.commit_exception(&s->store, &pe->e, commit_callback,
947 pe);
948}
949
950/*
951 * Dispatches the copy operation to kcopyd.
952 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100953static void start_copy(struct dm_snap_pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954{
955 struct dm_snapshot *s = pe->snap;
Heinz Mauelshagen22a1ceb2008-04-24 21:43:17 +0100956 struct dm_io_region src, dest;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 struct block_device *bdev = s->origin->bdev;
958 sector_t dev_size;
959
960 dev_size = get_dev_size(bdev);
961
962 src.bdev = bdev;
963 src.sector = chunk_to_sector(s, pe->e.old_chunk);
964 src.count = min(s->chunk_size, dev_size - src.sector);
965
966 dest.bdev = s->cow->bdev;
967 dest.sector = chunk_to_sector(s, pe->e.new_chunk);
968 dest.count = src.count;
969
970 /* Hand over to kcopyd */
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100971 dm_kcopyd_copy(s->kcopyd_client,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 &src, 1, &dest, 0, copy_callback, pe);
973}
974
975/*
976 * Looks to see if this snapshot already has a pending exception
977 * for this chunk, otherwise it allocates a new one and inserts
978 * it into the pending table.
979 *
980 * NOTE: a write lock must be held on snap->lock before calling
981 * this.
982 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100983static struct dm_snap_pending_exception *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984__find_pending_exception(struct dm_snapshot *s, struct bio *bio)
985{
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100986 struct dm_snap_exception *e;
987 struct dm_snap_pending_exception *pe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 chunk_t chunk = sector_to_chunk(s, bio->bi_sector);
989
990 /*
991 * Is there a pending exception for this already ?
992 */
993 e = lookup_exception(&s->pending, chunk);
994 if (e) {
995 /* cast the exception to a pending exception */
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100996 pe = container_of(e, struct dm_snap_pending_exception, e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800997 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 }
999
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001000 /*
1001 * Create a new pending exception, we don't want
1002 * to hold the lock while we do this.
1003 */
1004 up_write(&s->lock);
Mikulas Patocka92e86812008-07-21 12:00:35 +01001005 pe = alloc_pending_exception(s);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001006 down_write(&s->lock);
1007
1008 if (!s->valid) {
1009 free_pending_exception(pe);
1010 return NULL;
1011 }
1012
1013 e = lookup_exception(&s->pending, chunk);
1014 if (e) {
1015 free_pending_exception(pe);
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001016 pe = container_of(e, struct dm_snap_pending_exception, e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001017 goto out;
1018 }
1019
1020 pe->e.old_chunk = chunk;
1021 bio_list_init(&pe->origin_bios);
1022 bio_list_init(&pe->snapshot_bios);
1023 pe->primary_pe = NULL;
Alasdair G Kergon4b832e82006-10-03 01:15:30 -07001024 atomic_set(&pe->ref_count, 0);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001025 pe->started = 0;
1026
1027 if (s->store.prepare_exception(&s->store, &pe->e)) {
1028 free_pending_exception(pe);
1029 return NULL;
1030 }
1031
Alasdair G Kergon4b832e82006-10-03 01:15:30 -07001032 get_pending_exception(pe);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001033 insert_exception(&s->pending, &pe->e);
1034
1035 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 return pe;
1037}
1038
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001039static void remap_exception(struct dm_snapshot *s, struct dm_snap_exception *e,
Milan Brozd74f81f2008-02-08 02:11:27 +00001040 struct bio *bio, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041{
1042 bio->bi_bdev = s->cow->bdev;
Milan Brozd74f81f2008-02-08 02:11:27 +00001043 bio->bi_sector = chunk_to_sector(s, dm_chunk_number(e->new_chunk) +
1044 (chunk - e->old_chunk)) +
1045 (bio->bi_sector & s->chunk_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046}
1047
1048static int snapshot_map(struct dm_target *ti, struct bio *bio,
1049 union map_info *map_context)
1050{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001051 struct dm_snap_exception *e;
1052 struct dm_snapshot *s = ti->private;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001053 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 chunk_t chunk;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001055 struct dm_snap_pending_exception *pe = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056
1057 chunk = sector_to_chunk(s, bio->bi_sector);
1058
1059 /* Full snapshots are not usable */
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001060 /* To get here the table must be live so s->active is always set. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 if (!s->valid)
Alasdair G Kergonf6a80ea2005-07-12 15:53:01 -07001062 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001064 /* FIXME: should only take write lock if we need
1065 * to copy an exception */
1066 down_write(&s->lock);
1067
1068 if (!s->valid) {
1069 r = -EIO;
1070 goto out_unlock;
1071 }
1072
1073 /* If the block is already remapped - use that, else remap it */
1074 e = lookup_exception(&s->complete, chunk);
1075 if (e) {
Milan Brozd74f81f2008-02-08 02:11:27 +00001076 remap_exception(s, e, bio, chunk);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001077 goto out_unlock;
1078 }
1079
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 /*
1081 * Write to snapshot - higher level takes care of RW/RO
1082 * flags so we should only get this if we are
1083 * writeable.
1084 */
1085 if (bio_rw(bio) == WRITE) {
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001086 pe = __find_pending_exception(s, bio);
1087 if (!pe) {
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001088 __invalidate_snapshot(s, -ENOMEM);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001089 r = -EIO;
1090 goto out_unlock;
1091 }
1092
Milan Brozd74f81f2008-02-08 02:11:27 +00001093 remap_exception(s, &pe->e, bio, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001094 bio_list_add(&pe->snapshot_bios, bio);
1095
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001096 r = DM_MAPIO_SUBMITTED;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001097
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001098 if (!pe->started) {
1099 /* this is protected by snap->lock */
1100 pe->started = 1;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001101 up_write(&s->lock);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001102 start_copy(pe);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001103 goto out;
1104 }
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001105 } else {
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001106 bio->bi_bdev = s->origin->bdev;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001107 map_context->ptr = track_chunk(s, chunk);
1108 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001110 out_unlock:
1111 up_write(&s->lock);
1112 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 return r;
1114}
1115
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001116static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1117 int error, union map_info *map_context)
1118{
1119 struct dm_snapshot *s = ti->private;
1120 struct dm_snap_tracked_chunk *c = map_context->ptr;
1121
1122 if (c)
1123 stop_tracking_chunk(s, c);
1124
1125 return 0;
1126}
1127
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128static void snapshot_resume(struct dm_target *ti)
1129{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001130 struct dm_snapshot *s = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001132 down_write(&s->lock);
1133 s->active = 1;
1134 up_write(&s->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135}
1136
1137static int snapshot_status(struct dm_target *ti, status_type_t type,
1138 char *result, unsigned int maxlen)
1139{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001140 struct dm_snapshot *snap = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
1142 switch (type) {
1143 case STATUSTYPE_INFO:
1144 if (!snap->valid)
1145 snprintf(result, maxlen, "Invalid");
1146 else {
1147 if (snap->store.fraction_full) {
1148 sector_t numerator, denominator;
1149 snap->store.fraction_full(&snap->store,
1150 &numerator,
1151 &denominator);
Andrew Morton4ee218c2006-03-27 01:17:48 -08001152 snprintf(result, maxlen, "%llu/%llu",
1153 (unsigned long long)numerator,
1154 (unsigned long long)denominator);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 }
1156 else
1157 snprintf(result, maxlen, "Unknown");
1158 }
1159 break;
1160
1161 case STATUSTYPE_TABLE:
1162 /*
1163 * kdevname returns a static pointer so we need
1164 * to make private copies if the output is to
1165 * make sense.
1166 */
Andrew Morton4ee218c2006-03-27 01:17:48 -08001167 snprintf(result, maxlen, "%s %s %c %llu",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 snap->origin->name, snap->cow->name,
Andrew Morton4ee218c2006-03-27 01:17:48 -08001169 snap->type,
1170 (unsigned long long)snap->chunk_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 break;
1172 }
1173
1174 return 0;
1175}
1176
1177/*-----------------------------------------------------------------
1178 * Origin methods
1179 *---------------------------------------------------------------*/
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180static int __origin_write(struct list_head *snapshots, struct bio *bio)
1181{
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001182 int r = DM_MAPIO_REMAPPED, first = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 struct dm_snapshot *snap;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001184 struct dm_snap_exception *e;
1185 struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 chunk_t chunk;
Alasdair G Kergoneccf0812006-03-27 01:17:42 -08001187 LIST_HEAD(pe_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188
1189 /* Do all the snapshots on this origin */
1190 list_for_each_entry (snap, snapshots, list) {
1191
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001192 down_write(&snap->lock);
1193
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001194 /* Only deal with valid and active snapshots */
1195 if (!snap->valid || !snap->active)
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001196 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197
Alasdair G Kergond5e404c2005-07-12 15:53:05 -07001198 /* Nothing to do if writing beyond end of snapshot */
Mikulas Patocka72727ba2008-04-24 21:43:11 +01001199 if (bio->bi_sector >= dm_table_get_size(snap->ti->table))
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001200 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
1202 /*
1203 * Remember, different snapshots can have
1204 * different chunk sizes.
1205 */
1206 chunk = sector_to_chunk(snap, bio->bi_sector);
1207
1208 /*
1209 * Check exception table to see if block
1210 * is already remapped in this snapshot
1211 * and trigger an exception if not.
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001212 *
Alasdair G Kergon4b832e82006-10-03 01:15:30 -07001213 * ref_count is initialised to 1 so pending_complete()
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001214 * won't destroy the primary_pe while we're inside this loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 */
1216 e = lookup_exception(&snap->complete, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001217 if (e)
1218 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001220 pe = __find_pending_exception(snap, bio);
1221 if (!pe) {
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001222 __invalidate_snapshot(snap, -ENOMEM);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001223 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 }
1225
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001226 if (!primary_pe) {
1227 /*
1228 * Either every pe here has same
1229 * primary_pe or none has one yet.
1230 */
1231 if (pe->primary_pe)
1232 primary_pe = pe->primary_pe;
1233 else {
1234 primary_pe = pe;
1235 first = 1;
1236 }
1237
1238 bio_list_add(&primary_pe->origin_bios, bio);
1239
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001240 r = DM_MAPIO_SUBMITTED;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001241 }
1242
1243 if (!pe->primary_pe) {
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001244 pe->primary_pe = primary_pe;
Alasdair G Kergon4b832e82006-10-03 01:15:30 -07001245 get_pending_exception(primary_pe);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001246 }
1247
1248 if (!pe->started) {
1249 pe->started = 1;
1250 list_add_tail(&pe->list, &pe_queue);
1251 }
1252
1253 next_snapshot:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 up_write(&snap->lock);
1255 }
1256
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001257 if (!primary_pe)
Alasdair G Kergon4b832e82006-10-03 01:15:30 -07001258 return r;
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001259
1260 /*
1261 * If this is the first time we're processing this chunk and
Alasdair G Kergon4b832e82006-10-03 01:15:30 -07001262 * ref_count is now 1 it means all the pending exceptions
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001263 * got completed while we were in the loop above, so it falls to
1264 * us here to remove the primary_pe and submit any origin_bios.
1265 */
1266
Alasdair G Kergon4b832e82006-10-03 01:15:30 -07001267 if (first && atomic_dec_and_test(&primary_pe->ref_count)) {
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001268 flush_bios(bio_list_get(&primary_pe->origin_bios));
1269 free_pending_exception(primary_pe);
1270 /* If we got here, pe_queue is necessarily empty. */
Alasdair G Kergon4b832e82006-10-03 01:15:30 -07001271 return r;
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001272 }
1273
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 /*
1275 * Now that we have a complete pe list we can start the copying.
1276 */
Alasdair G Kergoneccf0812006-03-27 01:17:42 -08001277 list_for_each_entry_safe(pe, next_pe, &pe_queue, list)
1278 start_copy(pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279
1280 return r;
1281}
1282
1283/*
1284 * Called on a write from the origin driver.
1285 */
1286static int do_origin(struct dm_dev *origin, struct bio *bio)
1287{
1288 struct origin *o;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001289 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
1291 down_read(&_origins_lock);
1292 o = __lookup_origin(origin->bdev);
1293 if (o)
1294 r = __origin_write(&o->snapshots, bio);
1295 up_read(&_origins_lock);
1296
1297 return r;
1298}
1299
1300/*
1301 * Origin: maps a linear range of a device, with hooks for snapshotting.
1302 */
1303
1304/*
1305 * Construct an origin mapping: <dev_path>
1306 * The context for an origin is merely a 'struct dm_dev *'
1307 * pointing to the real device.
1308 */
1309static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1310{
1311 int r;
1312 struct dm_dev *dev;
1313
1314 if (argc != 1) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001315 ti->error = "origin: incorrect number of arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 return -EINVAL;
1317 }
1318
1319 r = dm_get_device(ti, argv[0], 0, ti->len,
1320 dm_table_get_mode(ti->table), &dev);
1321 if (r) {
1322 ti->error = "Cannot get target device";
1323 return r;
1324 }
1325
1326 ti->private = dev;
1327 return 0;
1328}
1329
1330static void origin_dtr(struct dm_target *ti)
1331{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001332 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 dm_put_device(ti, dev);
1334}
1335
1336static int origin_map(struct dm_target *ti, struct bio *bio,
1337 union map_info *map_context)
1338{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001339 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 bio->bi_bdev = dev->bdev;
1341
1342 /* Only tell snapshots if this is a write */
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001343 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344}
1345
1346#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
1347
1348/*
1349 * Set the target "split_io" field to the minimum of all the snapshots'
1350 * chunk sizes.
1351 */
1352static void origin_resume(struct dm_target *ti)
1353{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001354 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 struct dm_snapshot *snap;
1356 struct origin *o;
1357 chunk_t chunk_size = 0;
1358
1359 down_read(&_origins_lock);
1360 o = __lookup_origin(dev->bdev);
1361 if (o)
1362 list_for_each_entry (snap, &o->snapshots, list)
1363 chunk_size = min_not_zero(chunk_size, snap->chunk_size);
1364 up_read(&_origins_lock);
1365
1366 ti->split_io = chunk_size;
1367}
1368
1369static int origin_status(struct dm_target *ti, status_type_t type, char *result,
1370 unsigned int maxlen)
1371{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001372 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373
1374 switch (type) {
1375 case STATUSTYPE_INFO:
1376 result[0] = '\0';
1377 break;
1378
1379 case STATUSTYPE_TABLE:
1380 snprintf(result, maxlen, "%s", dev->name);
1381 break;
1382 }
1383
1384 return 0;
1385}
1386
1387static struct target_type origin_target = {
1388 .name = "snapshot-origin",
Milan Brozd74f81f2008-02-08 02:11:27 +00001389 .version = {1, 6, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 .module = THIS_MODULE,
1391 .ctr = origin_ctr,
1392 .dtr = origin_dtr,
1393 .map = origin_map,
1394 .resume = origin_resume,
1395 .status = origin_status,
1396};
1397
1398static struct target_type snapshot_target = {
1399 .name = "snapshot",
Milan Brozd74f81f2008-02-08 02:11:27 +00001400 .version = {1, 6, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 .module = THIS_MODULE,
1402 .ctr = snapshot_ctr,
1403 .dtr = snapshot_dtr,
1404 .map = snapshot_map,
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001405 .end_io = snapshot_end_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 .resume = snapshot_resume,
1407 .status = snapshot_status,
1408};
1409
1410static int __init dm_snapshot_init(void)
1411{
1412 int r;
1413
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00001414 r = dm_exception_store_init();
1415 if (r) {
1416 DMERR("Failed to initialize exception stores");
1417 return r;
1418 }
1419
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 r = dm_register_target(&snapshot_target);
1421 if (r) {
1422 DMERR("snapshot target register failed %d", r);
1423 return r;
1424 }
1425
1426 r = dm_register_target(&origin_target);
1427 if (r < 0) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001428 DMERR("Origin target register failed %d", r);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 goto bad1;
1430 }
1431
1432 r = init_origin_hash();
1433 if (r) {
1434 DMERR("init_origin_hash failed.");
1435 goto bad2;
1436 }
1437
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001438 exception_cache = KMEM_CACHE(dm_snap_exception, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 if (!exception_cache) {
1440 DMERR("Couldn't create exception cache.");
1441 r = -ENOMEM;
1442 goto bad3;
1443 }
1444
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001445 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 if (!pending_cache) {
1447 DMERR("Couldn't create pending cache.");
1448 r = -ENOMEM;
1449 goto bad4;
1450 }
1451
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001452 tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
1453 if (!tracked_chunk_cache) {
1454 DMERR("Couldn't create cache to track chunks in use.");
1455 r = -ENOMEM;
1456 goto bad5;
1457 }
1458
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07001459 ksnapd = create_singlethread_workqueue("ksnapd");
1460 if (!ksnapd) {
1461 DMERR("Failed to create ksnapd workqueue.");
1462 r = -ENOMEM;
Mikulas Patocka92e86812008-07-21 12:00:35 +01001463 goto bad_pending_pool;
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07001464 }
1465
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 return 0;
1467
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00001468bad_pending_pool:
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001469 kmem_cache_destroy(tracked_chunk_cache);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00001470bad5:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 kmem_cache_destroy(pending_cache);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00001472bad4:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 kmem_cache_destroy(exception_cache);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00001474bad3:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 exit_origin_hash();
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00001476bad2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 dm_unregister_target(&origin_target);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00001478bad1:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 dm_unregister_target(&snapshot_target);
1480 return r;
1481}
1482
1483static void __exit dm_snapshot_exit(void)
1484{
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07001485 destroy_workqueue(ksnapd);
1486
Mikulas Patocka10d3bd02009-01-06 03:04:58 +00001487 dm_unregister_target(&snapshot_target);
1488 dm_unregister_target(&origin_target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489
1490 exit_origin_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 kmem_cache_destroy(pending_cache);
1492 kmem_cache_destroy(exception_cache);
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001493 kmem_cache_destroy(tracked_chunk_cache);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00001494
1495 dm_exception_store_exit();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496}
1497
1498/* Module hooks */
1499module_init(dm_snapshot_init);
1500module_exit(dm_snapshot_exit);
1501
1502MODULE_DESCRIPTION(DM_NAME " snapshot target");
1503MODULE_AUTHOR("Joe Thornber");
1504MODULE_LICENSE("GPL");