blob: 94dee05dd28e4ee76fe64bc9665c4a2d630ae414 [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/device-mapper.h>
Mikulas Patocka90fa1522009-01-06 03:04:54 +000011#include <linux/delay.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#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>
vignesh babu6f3c3f02007-10-19 22:38:44 +010020#include <linux/log2.h>
Alasdair G Kergona765e202008-04-24 22:02:01 +010021#include <linux/dm-kcopyd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022
Jonathan Brassowaea53d92009-01-06 03:05:15 +000023#include "dm-exception-store.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024
Alasdair G Kergon72d94862006-06-26 00:27:35 -070025#define DM_MSG_PREFIX "snapshots"
26
Mikulas Patockad698aa42009-12-10 23:52:30 +000027static const char dm_snapshot_merge_target_name[] = "snapshot-merge";
28
29#define dm_target_is_snapshot_merge(ti) \
30 ((ti)->type->name == dm_snapshot_merge_target_name)
31
Linus Torvalds1da177e2005-04-16 15:20:36 -070032/*
Mikulas Patockacd45daf2008-07-21 12:00:32 +010033 * The size of the mempool used to track chunks in use.
34 */
35#define MIN_IOS 256
36
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010037#define DM_TRACKED_CHUNK_HASH_SIZE 16
38#define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
39 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
40
Jon Brassow191437a2009-12-10 23:52:10 +000041struct dm_exception_table {
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010042 uint32_t hash_mask;
43 unsigned hash_shift;
44 struct list_head *table;
45};
46
47struct dm_snapshot {
48 struct rw_semaphore lock;
49
50 struct dm_dev *origin;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +000051 struct dm_dev *cow;
52
53 struct dm_target *ti;
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010054
55 /* List of snapshots per Origin */
56 struct list_head list;
57
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +000058 /*
59 * You can't use a snapshot if this is 0 (e.g. if full).
60 * A snapshot-merge target never clears this.
61 */
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010062 int valid;
63
64 /* Origin writes don't trigger exceptions until this is set */
65 int active;
66
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010067 atomic_t pending_exceptions_count;
68
Mike Snitzer924e6002010-03-06 02:32:33 +000069 mempool_t *pending_pool;
70
Jon Brassow191437a2009-12-10 23:52:10 +000071 struct dm_exception_table pending;
72 struct dm_exception_table complete;
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010073
74 /*
75 * pe_lock protects all pending_exception operations and access
76 * as well as the snapshot_bios list.
77 */
78 spinlock_t pe_lock;
79
Mike Snitzer924e6002010-03-06 02:32:33 +000080 /* Chunks with outstanding reads */
81 spinlock_t tracked_chunk_lock;
82 mempool_t *tracked_chunk_pool;
83 struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
84
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010085 /* The on disk metadata handler */
86 struct dm_exception_store *store;
87
88 struct dm_kcopyd_client *kcopyd_client;
89
Mike Snitzer924e6002010-03-06 02:32:33 +000090 /* Wait for events based on state_bits */
91 unsigned long state_bits;
92
93 /* Range of chunks currently being merged. */
94 chunk_t first_merging_chunk;
95 int num_merging_chunks;
Mikulas Patocka1e03f972009-12-10 23:52:32 +000096
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +000097 /*
98 * The merge operation failed if this flag is set.
99 * Failure modes are handled as follows:
100 * - I/O error reading the header
101 * => don't load the target; abort.
102 * - Header does not have "valid" flag set
103 * => use the origin; forget about the snapshot.
104 * - I/O error when reading exceptions
105 * => don't load the target; abort.
106 * (We can't use the intermediate origin state.)
107 * - I/O error while merging
108 * => stop merging; set merge_failed; process I/O normally.
109 */
110 int merge_failed;
111
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000112 /*
113 * Incoming bios that overlap with chunks being merged must wait
114 * for them to be committed.
115 */
116 struct bio_list bios_queued_during_merge;
Jonathan Brassowccc45ea2009-04-02 19:55:34 +0100117};
118
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000119/*
120 * state_bits:
121 * RUNNING_MERGE - Merge operation is in progress.
122 * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
123 * cleared afterwards.
124 */
125#define RUNNING_MERGE 0
126#define SHUTDOWN_MERGE 1
127
Mikulas Patockac2411042010-08-12 04:13:51 +0100128struct dm_dev *dm_snap_origin(struct dm_snapshot *s)
129{
130 return s->origin;
131}
132EXPORT_SYMBOL(dm_snap_origin);
133
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000134struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
135{
136 return s->cow;
137}
138EXPORT_SYMBOL(dm_snap_cow);
139
Jonathan Brassowccc45ea2009-04-02 19:55:34 +0100140static sector_t chunk_to_sector(struct dm_exception_store *store,
141 chunk_t chunk)
142{
143 return chunk << store->chunk_shift;
144}
145
146static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
147{
148 /*
149 * There is only ever one instance of a particular block
150 * device so we can compare pointers safely.
151 */
152 return lhs == rhs;
153}
154
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100155struct dm_snap_pending_exception {
Jon Brassow1d4989c2009-12-10 23:52:10 +0000156 struct dm_exception e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
158 /*
159 * Origin buffers waiting for this to complete are held
160 * in a bio list
161 */
162 struct bio_list origin_bios;
163 struct bio_list snapshot_bios;
164
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 /* Pointer back to snapshot context */
166 struct dm_snapshot *snap;
167
168 /*
169 * 1 indicates the exception has already been sent to
170 * kcopyd.
171 */
172 int started;
173};
174
175/*
176 * Hash table mapping origin volumes to lists of snapshots and
177 * a lock to protect it
178 */
Christoph Lametere18b8902006-12-06 20:33:20 -0800179static struct kmem_cache *exception_cache;
180static struct kmem_cache *pending_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100182struct dm_snap_tracked_chunk {
183 struct hlist_node node;
184 chunk_t chunk;
185};
186
187static struct kmem_cache *tracked_chunk_cache;
188
189static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s,
190 chunk_t chunk)
191{
192 struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool,
193 GFP_NOIO);
194 unsigned long flags;
195
196 c->chunk = chunk;
197
198 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
199 hlist_add_head(&c->node,
200 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
201 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
202
203 return c;
204}
205
206static void stop_tracking_chunk(struct dm_snapshot *s,
207 struct dm_snap_tracked_chunk *c)
208{
209 unsigned long flags;
210
211 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
212 hlist_del(&c->node);
213 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
214
215 mempool_free(c, s->tracked_chunk_pool);
216}
217
Mikulas Patockaa8d41b52008-07-21 12:00:34 +0100218static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
219{
220 struct dm_snap_tracked_chunk *c;
221 struct hlist_node *hn;
222 int found = 0;
223
224 spin_lock_irq(&s->tracked_chunk_lock);
225
226 hlist_for_each_entry(c, hn,
227 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
228 if (c->chunk == chunk) {
229 found = 1;
230 break;
231 }
232 }
233
234 spin_unlock_irq(&s->tracked_chunk_lock);
235
236 return found;
237}
238
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239/*
Mike Snitzer615d1eb2009-12-10 23:52:29 +0000240 * This conflicting I/O is extremely improbable in the caller,
241 * so msleep(1) is sufficient and there is no need for a wait queue.
242 */
243static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
244{
245 while (__chunk_is_tracked(s, chunk))
246 msleep(1);
247}
248
249/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 * One of these per registered origin, held in the snapshot_origins hash
251 */
252struct origin {
253 /* The origin device */
254 struct block_device *bdev;
255
256 struct list_head hash_list;
257
258 /* List of snapshots for this origin */
259 struct list_head snapshots;
260};
261
262/*
263 * Size of the hash table for origin volumes. If we make this
264 * the size of the minors list then it should be nearly perfect
265 */
266#define ORIGIN_HASH_SIZE 256
267#define ORIGIN_MASK 0xFF
268static struct list_head *_origins;
269static struct rw_semaphore _origins_lock;
270
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000271static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
272static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
273static uint64_t _pending_exceptions_done_count;
274
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275static int init_origin_hash(void)
276{
277 int i;
278
279 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
280 GFP_KERNEL);
281 if (!_origins) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700282 DMERR("unable to allocate memory");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 return -ENOMEM;
284 }
285
286 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
287 INIT_LIST_HEAD(_origins + i);
288 init_rwsem(&_origins_lock);
289
290 return 0;
291}
292
293static void exit_origin_hash(void)
294{
295 kfree(_origins);
296}
297
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100298static unsigned origin_hash(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
300 return bdev->bd_dev & ORIGIN_MASK;
301}
302
303static struct origin *__lookup_origin(struct block_device *origin)
304{
305 struct list_head *ol;
306 struct origin *o;
307
308 ol = &_origins[origin_hash(origin)];
309 list_for_each_entry (o, ol, hash_list)
310 if (bdev_equal(o->bdev, origin))
311 return o;
312
313 return NULL;
314}
315
316static void __insert_origin(struct origin *o)
317{
318 struct list_head *sl = &_origins[origin_hash(o->bdev)];
319 list_add_tail(&o->hash_list, sl);
320}
321
322/*
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000323 * _origins_lock must be held when calling this function.
324 * Returns number of snapshots registered using the supplied cow device, plus:
325 * snap_src - a snapshot suitable for use as a source of exception handover
326 * snap_dest - a snapshot capable of receiving exception handover.
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000327 * snap_merge - an existing snapshot-merge target linked to the same origin.
328 * There can be at most one snapshot-merge target. The parameter is optional.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000329 *
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000330 * Possible return values and states of snap_src and snap_dest.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000331 * 0: NULL, NULL - first new snapshot
332 * 1: snap_src, NULL - normal snapshot
333 * 2: snap_src, snap_dest - waiting for handover
334 * 2: snap_src, NULL - handed over, waiting for old to be deleted
335 * 1: NULL, snap_dest - source got destroyed without handover
336 */
337static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
338 struct dm_snapshot **snap_src,
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000339 struct dm_snapshot **snap_dest,
340 struct dm_snapshot **snap_merge)
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000341{
342 struct dm_snapshot *s;
343 struct origin *o;
344 int count = 0;
345 int active;
346
347 o = __lookup_origin(snap->origin->bdev);
348 if (!o)
349 goto out;
350
351 list_for_each_entry(s, &o->snapshots, list) {
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000352 if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
353 *snap_merge = s;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000354 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
355 continue;
356
357 down_read(&s->lock);
358 active = s->active;
359 up_read(&s->lock);
360
361 if (active) {
362 if (snap_src)
363 *snap_src = s;
364 } else if (snap_dest)
365 *snap_dest = s;
366
367 count++;
368 }
369
370out:
371 return count;
372}
373
374/*
375 * On success, returns 1 if this snapshot is a handover destination,
376 * otherwise returns 0.
377 */
378static int __validate_exception_handover(struct dm_snapshot *snap)
379{
380 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000381 struct dm_snapshot *snap_merge = NULL;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000382
383 /* Does snapshot need exceptions handed over to it? */
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000384 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
385 &snap_merge) == 2) ||
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000386 snap_dest) {
387 snap->ti->error = "Snapshot cow pairing for exception "
388 "table handover failed";
389 return -EINVAL;
390 }
391
392 /*
393 * If no snap_src was found, snap cannot become a handover
394 * destination.
395 */
396 if (!snap_src)
397 return 0;
398
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000399 /*
400 * Non-snapshot-merge handover?
401 */
402 if (!dm_target_is_snapshot_merge(snap->ti))
403 return 1;
404
405 /*
406 * Do not allow more than one merging snapshot.
407 */
408 if (snap_merge) {
409 snap->ti->error = "A snapshot is already merging.";
410 return -EINVAL;
411 }
412
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000413 if (!snap_src->store->type->prepare_merge ||
414 !snap_src->store->type->commit_merge) {
415 snap->ti->error = "Snapshot exception store does not "
416 "support snapshot-merge.";
417 return -EINVAL;
418 }
419
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000420 return 1;
421}
422
423static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
424{
425 struct dm_snapshot *l;
426
427 /* Sort the list according to chunk size, largest-first smallest-last */
428 list_for_each_entry(l, &o->snapshots, list)
429 if (l->store->chunk_size < s->store->chunk_size)
430 break;
431 list_add_tail(&s->list, &l->list);
432}
433
434/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 * Make a note of the snapshot and its origin so we can look it
436 * up when the origin has a write on it.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000437 *
438 * Also validate snapshot exception store handovers.
439 * On success, returns 1 if this registration is a handover destination,
440 * otherwise returns 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 */
442static int register_snapshot(struct dm_snapshot *snap)
443{
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000444 struct origin *o, *new_o = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 struct block_device *bdev = snap->origin->bdev;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000446 int r = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000448 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
449 if (!new_o)
450 return -ENOMEM;
451
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 down_write(&_origins_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000454 r = __validate_exception_handover(snap);
455 if (r < 0) {
456 kfree(new_o);
457 goto out;
458 }
459
460 o = __lookup_origin(bdev);
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000461 if (o)
462 kfree(new_o);
463 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 /* New origin */
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000465 o = new_o;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
467 /* Initialise the struct */
468 INIT_LIST_HEAD(&o->snapshots);
469 o->bdev = bdev;
470
471 __insert_origin(o);
472 }
473
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000474 __insert_snapshot(o, snap);
475
476out:
477 up_write(&_origins_lock);
478
479 return r;
480}
481
482/*
483 * Move snapshot to correct place in list according to chunk size.
484 */
485static void reregister_snapshot(struct dm_snapshot *s)
486{
487 struct block_device *bdev = s->origin->bdev;
488
489 down_write(&_origins_lock);
490
491 list_del(&s->list);
492 __insert_snapshot(__lookup_origin(bdev), s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
494 up_write(&_origins_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495}
496
497static void unregister_snapshot(struct dm_snapshot *s)
498{
499 struct origin *o;
500
501 down_write(&_origins_lock);
502 o = __lookup_origin(s->origin->bdev);
503
504 list_del(&s->list);
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000505 if (o && list_empty(&o->snapshots)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 list_del(&o->hash_list);
507 kfree(o);
508 }
509
510 up_write(&_origins_lock);
511}
512
513/*
514 * Implementation of the exception hash tables.
Milan Brozd74f81f2008-02-08 02:11:27 +0000515 * The lowest hash_shift bits of the chunk number are ignored, allowing
516 * some consecutive chunks to be grouped together.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 */
Jon Brassow3510cb92009-12-10 23:52:11 +0000518static int dm_exception_table_init(struct dm_exception_table *et,
519 uint32_t size, unsigned hash_shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520{
521 unsigned int i;
522
Milan Brozd74f81f2008-02-08 02:11:27 +0000523 et->hash_shift = hash_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 et->hash_mask = size - 1;
525 et->table = dm_vcalloc(size, sizeof(struct list_head));
526 if (!et->table)
527 return -ENOMEM;
528
529 for (i = 0; i < size; i++)
530 INIT_LIST_HEAD(et->table + i);
531
532 return 0;
533}
534
Jon Brassow3510cb92009-12-10 23:52:11 +0000535static void dm_exception_table_exit(struct dm_exception_table *et,
536 struct kmem_cache *mem)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537{
538 struct list_head *slot;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000539 struct dm_exception *ex, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 int i, size;
541
542 size = et->hash_mask + 1;
543 for (i = 0; i < size; i++) {
544 slot = et->table + i;
545
546 list_for_each_entry_safe (ex, next, slot, hash_list)
547 kmem_cache_free(mem, ex);
548 }
549
550 vfree(et->table);
551}
552
Jon Brassow191437a2009-12-10 23:52:10 +0000553static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554{
Milan Brozd74f81f2008-02-08 02:11:27 +0000555 return (chunk >> et->hash_shift) & et->hash_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556}
557
Jon Brassow3510cb92009-12-10 23:52:11 +0000558static void dm_remove_exception(struct dm_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559{
560 list_del(&e->hash_list);
561}
562
563/*
564 * Return the exception data for a sector, or NULL if not
565 * remapped.
566 */
Jon Brassow3510cb92009-12-10 23:52:11 +0000567static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
568 chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569{
570 struct list_head *slot;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000571 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
573 slot = &et->table[exception_hash(et, chunk)];
574 list_for_each_entry (e, slot, hash_list)
Milan Brozd74f81f2008-02-08 02:11:27 +0000575 if (chunk >= e->old_chunk &&
576 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 return e;
578
579 return NULL;
580}
581
Jon Brassow3510cb92009-12-10 23:52:11 +0000582static struct dm_exception *alloc_completed_exception(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583{
Jon Brassow1d4989c2009-12-10 23:52:10 +0000584 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
586 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
587 if (!e)
588 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
589
590 return e;
591}
592
Jon Brassow3510cb92009-12-10 23:52:11 +0000593static void free_completed_exception(struct dm_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594{
595 kmem_cache_free(exception_cache, e);
596}
597
Mikulas Patocka92e86812008-07-21 12:00:35 +0100598static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599{
Mikulas Patocka92e86812008-07-21 12:00:35 +0100600 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
601 GFP_NOIO);
602
Mikulas Patocka879129d22008-10-30 13:33:16 +0000603 atomic_inc(&s->pending_exceptions_count);
Mikulas Patocka92e86812008-07-21 12:00:35 +0100604 pe->snap = s;
605
606 return pe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607}
608
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100609static void free_pending_exception(struct dm_snap_pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610{
Mikulas Patocka879129d22008-10-30 13:33:16 +0000611 struct dm_snapshot *s = pe->snap;
612
613 mempool_free(pe, s->pending_pool);
614 smp_mb__before_atomic_dec();
615 atomic_dec(&s->pending_exceptions_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616}
617
Jon Brassow3510cb92009-12-10 23:52:11 +0000618static void dm_insert_exception(struct dm_exception_table *eh,
619 struct dm_exception *new_e)
Milan Brozd74f81f2008-02-08 02:11:27 +0000620{
Milan Brozd74f81f2008-02-08 02:11:27 +0000621 struct list_head *l;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000622 struct dm_exception *e = NULL;
Milan Brozd74f81f2008-02-08 02:11:27 +0000623
624 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
625
626 /* Add immediately if this table doesn't support consecutive chunks */
627 if (!eh->hash_shift)
628 goto out;
629
630 /* List is ordered by old_chunk */
631 list_for_each_entry_reverse(e, l, hash_list) {
632 /* Insert after an existing chunk? */
633 if (new_e->old_chunk == (e->old_chunk +
634 dm_consecutive_chunk_count(e) + 1) &&
635 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
636 dm_consecutive_chunk_count(e) + 1)) {
637 dm_consecutive_chunk_count_inc(e);
Jon Brassow3510cb92009-12-10 23:52:11 +0000638 free_completed_exception(new_e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000639 return;
640 }
641
642 /* Insert before an existing chunk? */
643 if (new_e->old_chunk == (e->old_chunk - 1) &&
644 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
645 dm_consecutive_chunk_count_inc(e);
646 e->old_chunk--;
647 e->new_chunk--;
Jon Brassow3510cb92009-12-10 23:52:11 +0000648 free_completed_exception(new_e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000649 return;
650 }
651
652 if (new_e->old_chunk > e->old_chunk)
653 break;
654 }
655
656out:
657 list_add(&new_e->hash_list, e ? &e->hash_list : l);
658}
659
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000660/*
661 * Callback used by the exception stores to load exceptions when
662 * initialising.
663 */
664static int dm_add_exception(void *context, chunk_t old, chunk_t new)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665{
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000666 struct dm_snapshot *s = context;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000667 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
Jon Brassow3510cb92009-12-10 23:52:11 +0000669 e = alloc_completed_exception();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 if (!e)
671 return -ENOMEM;
672
673 e->old_chunk = old;
Milan Brozd74f81f2008-02-08 02:11:27 +0000674
675 /* Consecutive_count is implicitly initialised to zero */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 e->new_chunk = new;
Milan Brozd74f81f2008-02-08 02:11:27 +0000677
Jon Brassow3510cb92009-12-10 23:52:11 +0000678 dm_insert_exception(&s->complete, e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000679
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 return 0;
681}
682
Mikulas Patocka7e201b32009-12-10 23:52:08 +0000683/*
684 * Return a minimum chunk size of all snapshots that have the specified origin.
685 * Return zero if the origin has no snapshots.
686 */
687static sector_t __minimum_chunk_size(struct origin *o)
688{
689 struct dm_snapshot *snap;
690 unsigned chunk_size = 0;
691
692 if (o)
693 list_for_each_entry(snap, &o->snapshots, list)
694 chunk_size = min_not_zero(chunk_size,
695 snap->store->chunk_size);
696
697 return chunk_size;
698}
699
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700/*
701 * Hard coded magic.
702 */
703static int calc_max_buckets(void)
704{
705 /* use a fixed size of 2MB */
706 unsigned long mem = 2 * 1024 * 1024;
707 mem /= sizeof(struct list_head);
708
709 return mem;
710}
711
712/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 * Allocate room for a suitable hash table.
714 */
Jonathan Brassowfee19982009-04-02 19:55:34 +0100715static int init_hash_tables(struct dm_snapshot *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716{
717 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
718
719 /*
720 * Calculate based on the size of the original volume or
721 * the COW volume...
722 */
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000723 cow_dev_size = get_dev_size(s->cow->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 origin_dev_size = get_dev_size(s->origin->bdev);
725 max_buckets = calc_max_buckets();
726
Jonathan Brassowfee19982009-04-02 19:55:34 +0100727 hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 hash_size = min(hash_size, max_buckets);
729
Mikulas Patocka8e87b9b2009-12-10 23:51:54 +0000730 if (hash_size < 64)
731 hash_size = 64;
Robert P. J. Day8defd832008-02-08 02:10:06 +0000732 hash_size = rounddown_pow_of_two(hash_size);
Jon Brassow3510cb92009-12-10 23:52:11 +0000733 if (dm_exception_table_init(&s->complete, hash_size,
734 DM_CHUNK_CONSECUTIVE_BITS))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 return -ENOMEM;
736
737 /*
738 * Allocate hash table for in-flight exceptions
739 * Make this smaller than the real hash table
740 */
741 hash_size >>= 3;
742 if (hash_size < 64)
743 hash_size = 64;
744
Jon Brassow3510cb92009-12-10 23:52:11 +0000745 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
746 dm_exception_table_exit(&s->complete, exception_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 return -ENOMEM;
748 }
749
750 return 0;
751}
752
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000753static void merge_shutdown(struct dm_snapshot *s)
754{
755 clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
756 smp_mb__after_clear_bit();
757 wake_up_bit(&s->state_bits, RUNNING_MERGE);
758}
759
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000760static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
761{
762 s->first_merging_chunk = 0;
763 s->num_merging_chunks = 0;
764
765 return bio_list_get(&s->bios_queued_during_merge);
766}
767
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000768/*
769 * Remove one chunk from the index of completed exceptions.
770 */
771static int __remove_single_exception_chunk(struct dm_snapshot *s,
772 chunk_t old_chunk)
773{
774 struct dm_exception *e;
775
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000776 e = dm_lookup_exception(&s->complete, old_chunk);
777 if (!e) {
778 DMERR("Corruption detected: exception for block %llu is "
779 "on disk but not in memory",
780 (unsigned long long)old_chunk);
781 return -EINVAL;
782 }
783
784 /*
785 * If this is the only chunk using this exception, remove exception.
786 */
787 if (!dm_consecutive_chunk_count(e)) {
788 dm_remove_exception(e);
789 free_completed_exception(e);
790 return 0;
791 }
792
793 /*
794 * The chunk may be either at the beginning or the end of a
795 * group of consecutive chunks - never in the middle. We are
796 * removing chunks in the opposite order to that in which they
797 * were added, so this should always be true.
798 * Decrement the consecutive chunk counter and adjust the
799 * starting point if necessary.
800 */
801 if (old_chunk == e->old_chunk) {
802 e->old_chunk++;
803 e->new_chunk++;
804 } else if (old_chunk != e->old_chunk +
805 dm_consecutive_chunk_count(e)) {
806 DMERR("Attempt to merge block %llu from the "
807 "middle of a chunk range [%llu - %llu]",
808 (unsigned long long)old_chunk,
809 (unsigned long long)e->old_chunk,
810 (unsigned long long)
811 e->old_chunk + dm_consecutive_chunk_count(e));
812 return -EINVAL;
813 }
814
815 dm_consecutive_chunk_count_dec(e);
816
817 return 0;
818}
819
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000820static void flush_bios(struct bio *bio);
821
822static int remove_single_exception_chunk(struct dm_snapshot *s)
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000823{
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000824 struct bio *b = NULL;
825 int r;
826 chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000827
828 down_write(&s->lock);
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000829
830 /*
831 * Process chunks (and associated exceptions) in reverse order
832 * so that dm_consecutive_chunk_count_dec() accounting works.
833 */
834 do {
835 r = __remove_single_exception_chunk(s, old_chunk);
836 if (r)
837 goto out;
838 } while (old_chunk-- > s->first_merging_chunk);
839
840 b = __release_queued_bios_after_merge(s);
841
842out:
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000843 up_write(&s->lock);
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000844 if (b)
845 flush_bios(b);
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000846
847 return r;
848}
849
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000850static int origin_write_extent(struct dm_snapshot *merging_snap,
851 sector_t sector, unsigned chunk_size);
852
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000853static void merge_callback(int read_err, unsigned long write_err,
854 void *context);
855
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000856static uint64_t read_pending_exceptions_done_count(void)
857{
858 uint64_t pending_exceptions_done;
859
860 spin_lock(&_pending_exceptions_done_spinlock);
861 pending_exceptions_done = _pending_exceptions_done_count;
862 spin_unlock(&_pending_exceptions_done_spinlock);
863
864 return pending_exceptions_done;
865}
866
867static void increment_pending_exceptions_done_count(void)
868{
869 spin_lock(&_pending_exceptions_done_spinlock);
870 _pending_exceptions_done_count++;
871 spin_unlock(&_pending_exceptions_done_spinlock);
872
873 wake_up_all(&_pending_exceptions_done);
874}
875
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000876static void snapshot_merge_next_chunks(struct dm_snapshot *s)
877{
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000878 int i, linear_chunks;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000879 chunk_t old_chunk, new_chunk;
880 struct dm_io_region src, dest;
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000881 sector_t io_size;
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000882 uint64_t previous_count;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000883
884 BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
885 if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
886 goto shut;
887
888 /*
889 * valid flag never changes during merge, so no lock required.
890 */
891 if (!s->valid) {
892 DMERR("Snapshot is invalid: can't merge");
893 goto shut;
894 }
895
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000896 linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
897 &new_chunk);
898 if (linear_chunks <= 0) {
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +0000899 if (linear_chunks < 0) {
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000900 DMERR("Read error in exception store: "
901 "shutting down merge");
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +0000902 down_write(&s->lock);
903 s->merge_failed = 1;
904 up_write(&s->lock);
905 }
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000906 goto shut;
907 }
908
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000909 /* Adjust old_chunk and new_chunk to reflect start of linear region */
910 old_chunk = old_chunk + 1 - linear_chunks;
911 new_chunk = new_chunk + 1 - linear_chunks;
912
913 /*
914 * Use one (potentially large) I/O to copy all 'linear_chunks'
915 * from the exception store to the origin
916 */
917 io_size = linear_chunks * s->store->chunk_size;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000918
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000919 dest.bdev = s->origin->bdev;
920 dest.sector = chunk_to_sector(s->store, old_chunk);
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000921 dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000922
923 src.bdev = s->cow->bdev;
924 src.sector = chunk_to_sector(s->store, new_chunk);
925 src.count = dest.count;
926
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000927 /*
928 * Reallocate any exceptions needed in other snapshots then
929 * wait for the pending exceptions to complete.
930 * Each time any pending exception (globally on the system)
931 * completes we are woken and repeat the process to find out
932 * if we can proceed. While this may not seem a particularly
933 * efficient algorithm, it is not expected to have any
934 * significant impact on performance.
935 */
936 previous_count = read_pending_exceptions_done_count();
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000937 while (origin_write_extent(s, dest.sector, io_size)) {
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000938 wait_event(_pending_exceptions_done,
939 (read_pending_exceptions_done_count() !=
940 previous_count));
941 /* Retry after the wait, until all exceptions are done. */
942 previous_count = read_pending_exceptions_done_count();
943 }
944
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000945 down_write(&s->lock);
946 s->first_merging_chunk = old_chunk;
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000947 s->num_merging_chunks = linear_chunks;
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000948 up_write(&s->lock);
949
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000950 /* Wait until writes to all 'linear_chunks' drain */
951 for (i = 0; i < linear_chunks; i++)
952 __check_for_conflicting_io(s, old_chunk + i);
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000953
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000954 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
955 return;
956
957shut:
958 merge_shutdown(s);
959}
960
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000961static void error_bios(struct bio *bio);
962
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000963static void merge_callback(int read_err, unsigned long write_err, void *context)
964{
965 struct dm_snapshot *s = context;
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000966 struct bio *b = NULL;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000967
968 if (read_err || write_err) {
969 if (read_err)
970 DMERR("Read error: shutting down merge.");
971 else
972 DMERR("Write error: shutting down merge.");
973 goto shut;
974 }
975
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000976 if (s->store->type->commit_merge(s->store,
977 s->num_merging_chunks) < 0) {
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000978 DMERR("Write error in exception store: shutting down merge");
979 goto shut;
980 }
981
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000982 if (remove_single_exception_chunk(s) < 0)
983 goto shut;
984
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000985 snapshot_merge_next_chunks(s);
986
987 return;
988
989shut:
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000990 down_write(&s->lock);
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +0000991 s->merge_failed = 1;
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000992 b = __release_queued_bios_after_merge(s);
993 up_write(&s->lock);
994 error_bios(b);
995
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000996 merge_shutdown(s);
997}
998
999static void start_merge(struct dm_snapshot *s)
1000{
1001 if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
1002 snapshot_merge_next_chunks(s);
1003}
1004
1005static int wait_schedule(void *ptr)
1006{
1007 schedule();
1008
1009 return 0;
1010}
1011
1012/*
1013 * Stop the merging process and wait until it finishes.
1014 */
1015static void stop_merge(struct dm_snapshot *s)
1016{
1017 set_bit(SHUTDOWN_MERGE, &s->state_bits);
1018 wait_on_bit(&s->state_bits, RUNNING_MERGE, wait_schedule,
1019 TASK_UNINTERRUPTIBLE);
1020 clear_bit(SHUTDOWN_MERGE, &s->state_bits);
1021}
1022
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
1025 */
1026static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1027{
1028 struct dm_snapshot *s;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001029 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 int r = -EINVAL;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001031 char *origin_path, *cow_path;
Mike Snitzer10b81062009-12-10 23:52:31 +00001032 unsigned args_used, num_flush_requests = 1;
1033 fmode_t origin_mode = FMODE_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -07001035 if (argc != 4) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001036 ti->error = "requires exactly 4 arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 r = -EINVAL;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001038 goto bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 }
1040
Mike Snitzer10b81062009-12-10 23:52:31 +00001041 if (dm_target_is_snapshot_merge(ti)) {
1042 num_flush_requests = 2;
1043 origin_mode = FMODE_WRITE;
1044 }
1045
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 s = kmalloc(sizeof(*s), GFP_KERNEL);
Jonathan Brassowfee19982009-04-02 19:55:34 +01001047 if (!s) {
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001048 ti->error = "Cannot allocate private snapshot structure";
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 r = -ENOMEM;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001050 goto bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 }
1052
Mikulas Patockac2411042010-08-12 04:13:51 +01001053 origin_path = argv[0];
1054 argv++;
1055 argc--;
1056
1057 r = dm_get_device(ti, origin_path, origin_mode, &s->origin);
1058 if (r) {
1059 ti->error = "Cannot get origin device";
1060 goto bad_origin;
1061 }
1062
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001063 cow_path = argv[0];
1064 argv++;
1065 argc--;
1066
Milan Broz024d37e2011-03-24 13:52:14 +00001067 r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow);
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001068 if (r) {
1069 ti->error = "Cannot get COW device";
1070 goto bad_cow;
1071 }
1072
1073 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
1074 if (r) {
1075 ti->error = "Couldn't create exception store";
1076 r = -EINVAL;
1077 goto bad_store;
1078 }
1079
1080 argv += args_used;
1081 argc -= args_used;
1082
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001083 s->ti = ti;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 s->valid = 1;
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001085 s->active = 0;
Mikulas Patocka879129d22008-10-30 13:33:16 +00001086 atomic_set(&s->pending_exceptions_count, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 init_rwsem(&s->lock);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001088 INIT_LIST_HEAD(&s->list);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07001089 spin_lock_init(&s->pe_lock);
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001090 s->state_bits = 0;
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +00001091 s->merge_failed = 0;
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001092 s->first_merging_chunk = 0;
1093 s->num_merging_chunks = 0;
1094 bio_list_init(&s->bios_queued_during_merge);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
1096 /* Allocate hash table for COW data */
Jonathan Brassowfee19982009-04-02 19:55:34 +01001097 if (init_hash_tables(s)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 ti->error = "Unable to allocate hash table space";
1099 r = -ENOMEM;
Jonathan Brassowfee19982009-04-02 19:55:34 +01001100 goto bad_hash_tables;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 }
1102
Mikulas Patockafa34ce72011-05-29 13:03:13 +01001103 s->kcopyd_client = dm_kcopyd_client_create();
1104 if (IS_ERR(s->kcopyd_client)) {
1105 r = PTR_ERR(s->kcopyd_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 ti->error = "Could not create kcopyd client";
Jonathan Brassowfee19982009-04-02 19:55:34 +01001107 goto bad_kcopyd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 }
1109
Mikulas Patocka92e86812008-07-21 12:00:35 +01001110 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
1111 if (!s->pending_pool) {
1112 ti->error = "Could not allocate mempool for pending exceptions";
Jonathan Brassowfee19982009-04-02 19:55:34 +01001113 goto bad_pending_pool;
Mikulas Patocka92e86812008-07-21 12:00:35 +01001114 }
1115
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001116 s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS,
1117 tracked_chunk_cache);
1118 if (!s->tracked_chunk_pool) {
1119 ti->error = "Could not allocate tracked_chunk mempool for "
1120 "tracking reads";
Mikulas Patocka92e86812008-07-21 12:00:35 +01001121 goto bad_tracked_chunk_pool;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001122 }
1123
1124 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1125 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
1126
1127 spin_lock_init(&s->tracked_chunk_lock);
1128
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001129 ti->private = s;
Mike Snitzer10b81062009-12-10 23:52:31 +00001130 ti->num_flush_requests = num_flush_requests;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001131
1132 /* Add snapshot to the list of snapshots for this origin */
1133 /* Exceptions aren't triggered till snapshot_resume() is called */
1134 r = register_snapshot(s);
1135 if (r == -ENOMEM) {
1136 ti->error = "Snapshot origin struct allocation failed";
1137 goto bad_load_and_register;
1138 } else if (r < 0) {
1139 /* invalid handover, register_snapshot has set ti->error */
1140 goto bad_load_and_register;
1141 }
1142
1143 /*
1144 * Metadata must only be loaded into one table at once, so skip this
1145 * if metadata will be handed over during resume.
1146 * Chunk size will be set during the handover - set it to zero to
1147 * ensure it's ignored.
1148 */
1149 if (r > 0) {
1150 s->store->chunk_size = 0;
1151 return 0;
1152 }
1153
Jonathan Brassow493df712009-04-02 19:55:31 +01001154 r = s->store->type->read_metadata(s->store, dm_add_exception,
1155 (void *)s);
Milan Broz07641472007-07-12 17:28:13 +01001156 if (r < 0) {
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -07001157 ti->error = "Failed to read snapshot metadata";
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001158 goto bad_read_metadata;
Milan Broz07641472007-07-12 17:28:13 +01001159 } else if (r > 0) {
1160 s->valid = 0;
1161 DMWARN("Snapshot is marked invalid.");
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -07001162 }
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001163
Mikulas Patocka3f2412d2009-10-16 23:18:16 +01001164 if (!s->store->chunk_size) {
1165 ti->error = "Chunk size not set";
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001166 goto bad_read_metadata;
Mikulas Patocka3f2412d2009-10-16 23:18:16 +01001167 }
Jonathan Brassowd0216842009-04-02 19:55:32 +01001168 ti->split_io = s->store->chunk_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
1170 return 0;
1171
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001172bad_read_metadata:
1173 unregister_snapshot(s);
1174
Jonathan Brassowfee19982009-04-02 19:55:34 +01001175bad_load_and_register:
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001176 mempool_destroy(s->tracked_chunk_pool);
1177
Jonathan Brassowfee19982009-04-02 19:55:34 +01001178bad_tracked_chunk_pool:
Mikulas Patocka92e86812008-07-21 12:00:35 +01001179 mempool_destroy(s->pending_pool);
1180
Jonathan Brassowfee19982009-04-02 19:55:34 +01001181bad_pending_pool:
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +01001182 dm_kcopyd_client_destroy(s->kcopyd_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
Jonathan Brassowfee19982009-04-02 19:55:34 +01001184bad_kcopyd:
Jon Brassow3510cb92009-12-10 23:52:11 +00001185 dm_exception_table_exit(&s->pending, pending_cache);
1186 dm_exception_table_exit(&s->complete, exception_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187
Jonathan Brassowfee19982009-04-02 19:55:34 +01001188bad_hash_tables:
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001189 dm_exception_store_destroy(s->store);
1190
1191bad_store:
1192 dm_put_device(ti, s->cow);
1193
1194bad_cow:
Mikulas Patockac2411042010-08-12 04:13:51 +01001195 dm_put_device(ti, s->origin);
1196
1197bad_origin:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 kfree(s);
1199
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001200bad:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 return r;
1202}
1203
Milan Broz31c93a02006-12-08 02:41:11 -08001204static void __free_exceptions(struct dm_snapshot *s)
1205{
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +01001206 dm_kcopyd_client_destroy(s->kcopyd_client);
Milan Broz31c93a02006-12-08 02:41:11 -08001207 s->kcopyd_client = NULL;
1208
Jon Brassow3510cb92009-12-10 23:52:11 +00001209 dm_exception_table_exit(&s->pending, pending_cache);
1210 dm_exception_table_exit(&s->complete, exception_cache);
Milan Broz31c93a02006-12-08 02:41:11 -08001211}
1212
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001213static void __handover_exceptions(struct dm_snapshot *snap_src,
1214 struct dm_snapshot *snap_dest)
1215{
1216 union {
1217 struct dm_exception_table table_swap;
1218 struct dm_exception_store *store_swap;
1219 } u;
1220
1221 /*
1222 * Swap all snapshot context information between the two instances.
1223 */
1224 u.table_swap = snap_dest->complete;
1225 snap_dest->complete = snap_src->complete;
1226 snap_src->complete = u.table_swap;
1227
1228 u.store_swap = snap_dest->store;
1229 snap_dest->store = snap_src->store;
1230 snap_src->store = u.store_swap;
1231
1232 snap_dest->store->snap = snap_dest;
1233 snap_src->store->snap = snap_src;
1234
1235 snap_dest->ti->split_io = snap_dest->store->chunk_size;
1236 snap_dest->valid = snap_src->valid;
1237
1238 /*
1239 * Set source invalid to ensure it receives no further I/O.
1240 */
1241 snap_src->valid = 0;
1242}
1243
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244static void snapshot_dtr(struct dm_target *ti)
1245{
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001246#ifdef CONFIG_DM_DEBUG
1247 int i;
1248#endif
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001249 struct dm_snapshot *s = ti->private;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001250 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001252 down_read(&_origins_lock);
1253 /* Check whether exception handover must be cancelled */
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +00001254 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001255 if (snap_src && snap_dest && (s == snap_src)) {
1256 down_write(&snap_dest->lock);
1257 snap_dest->valid = 0;
1258 up_write(&snap_dest->lock);
1259 DMERR("Cancelling snapshot handover.");
1260 }
1261 up_read(&_origins_lock);
1262
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001263 if (dm_target_is_snapshot_merge(ti))
1264 stop_merge(s);
1265
Alasdair G Kergon138728dc2006-03-27 01:17:50 -08001266 /* Prevent further origin writes from using this snapshot. */
1267 /* After this returns there can be no new kcopyd jobs. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 unregister_snapshot(s);
1269
Mikulas Patocka879129d22008-10-30 13:33:16 +00001270 while (atomic_read(&s->pending_exceptions_count))
Mikulas Patocka90fa1522009-01-06 03:04:54 +00001271 msleep(1);
Mikulas Patocka879129d22008-10-30 13:33:16 +00001272 /*
1273 * Ensure instructions in mempool_destroy aren't reordered
1274 * before atomic_read.
1275 */
1276 smp_mb();
1277
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001278#ifdef CONFIG_DM_DEBUG
1279 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1280 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
1281#endif
1282
1283 mempool_destroy(s->tracked_chunk_pool);
1284
Milan Broz31c93a02006-12-08 02:41:11 -08001285 __free_exceptions(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286
Mikulas Patocka92e86812008-07-21 12:00:35 +01001287 mempool_destroy(s->pending_pool);
1288
Jonathan Brassowfee19982009-04-02 19:55:34 +01001289 dm_exception_store_destroy(s->store);
Alasdair G Kergon138728dc2006-03-27 01:17:50 -08001290
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001291 dm_put_device(ti, s->cow);
1292
Mikulas Patockac2411042010-08-12 04:13:51 +01001293 dm_put_device(ti, s->origin);
1294
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 kfree(s);
1296}
1297
1298/*
1299 * Flush a list of buffers.
1300 */
1301static void flush_bios(struct bio *bio)
1302{
1303 struct bio *n;
1304
1305 while (bio) {
1306 n = bio->bi_next;
1307 bio->bi_next = NULL;
1308 generic_make_request(bio);
1309 bio = n;
1310 }
1311}
1312
Mikulas Patocka515ad662009-12-10 23:52:30 +00001313static int do_origin(struct dm_dev *origin, struct bio *bio);
1314
1315/*
1316 * Flush a list of buffers.
1317 */
1318static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1319{
1320 struct bio *n;
1321 int r;
1322
1323 while (bio) {
1324 n = bio->bi_next;
1325 bio->bi_next = NULL;
1326 r = do_origin(s->origin, bio);
1327 if (r == DM_MAPIO_REMAPPED)
1328 generic_make_request(bio);
1329 bio = n;
1330 }
1331}
1332
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333/*
1334 * Error a list of buffers.
1335 */
1336static void error_bios(struct bio *bio)
1337{
1338 struct bio *n;
1339
1340 while (bio) {
1341 n = bio->bi_next;
1342 bio->bi_next = NULL;
NeilBrown6712ecf2007-09-27 12:47:43 +02001343 bio_io_error(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 bio = n;
1345 }
1346}
1347
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001348static void __invalidate_snapshot(struct dm_snapshot *s, int err)
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001349{
1350 if (!s->valid)
1351 return;
1352
1353 if (err == -EIO)
1354 DMERR("Invalidating snapshot: Error reading/writing.");
1355 else if (err == -ENOMEM)
1356 DMERR("Invalidating snapshot: Unable to allocate exception.");
1357
Jonathan Brassow493df712009-04-02 19:55:31 +01001358 if (s->store->type->drop_snapshot)
1359 s->store->type->drop_snapshot(s->store);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001360
1361 s->valid = 0;
1362
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001363 dm_table_event(s->ti->table);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001364}
1365
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001366static void pending_complete(struct dm_snap_pending_exception *pe, int success)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367{
Jon Brassow1d4989c2009-12-10 23:52:10 +00001368 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 struct dm_snapshot *s = pe->snap;
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001370 struct bio *origin_bios = NULL;
1371 struct bio *snapshot_bios = NULL;
1372 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001374 if (!success) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 /* Read/write error - snapshot is unusable */
1376 down_write(&s->lock);
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001377 __invalidate_snapshot(s, -EIO);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001378 error = 1;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001379 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 }
1381
Jon Brassow3510cb92009-12-10 23:52:11 +00001382 e = alloc_completed_exception();
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001383 if (!e) {
1384 down_write(&s->lock);
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001385 __invalidate_snapshot(s, -ENOMEM);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001386 error = 1;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001387 goto out;
1388 }
1389 *e = pe->e;
1390
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001391 down_write(&s->lock);
1392 if (!s->valid) {
Jon Brassow3510cb92009-12-10 23:52:11 +00001393 free_completed_exception(e);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001394 error = 1;
1395 goto out;
1396 }
1397
Mike Snitzer615d1eb2009-12-10 23:52:29 +00001398 /* Check for conflicting reads */
1399 __check_for_conflicting_io(s, pe->e.old_chunk);
Mikulas Patockaa8d41b52008-07-21 12:00:34 +01001400
1401 /*
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001402 * Add a proper exception, and remove the
1403 * in-flight exception from the list.
1404 */
Jon Brassow3510cb92009-12-10 23:52:11 +00001405 dm_insert_exception(&s->complete, e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001406
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001407out:
Jon Brassow3510cb92009-12-10 23:52:11 +00001408 dm_remove_exception(&pe->e);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001409 snapshot_bios = bio_list_get(&pe->snapshot_bios);
Mikulas Patocka515ad662009-12-10 23:52:30 +00001410 origin_bios = bio_list_get(&pe->origin_bios);
1411 free_pending_exception(pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412
Mikulas Patocka73dfd072009-12-10 23:52:34 +00001413 increment_pending_exceptions_done_count();
1414
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001415 up_write(&s->lock);
1416
1417 /* Submit any pending write bios */
1418 if (error)
1419 error_bios(snapshot_bios);
1420 else
1421 flush_bios(snapshot_bios);
1422
Mikulas Patocka515ad662009-12-10 23:52:30 +00001423 retry_origin_bios(s, origin_bios);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424}
1425
1426static void commit_callback(void *context, int success)
1427{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001428 struct dm_snap_pending_exception *pe = context;
1429
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 pending_complete(pe, success);
1431}
1432
1433/*
1434 * Called when the copy I/O has finished. kcopyd actually runs
1435 * this code so don't block.
1436 */
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -07001437static void copy_callback(int read_err, unsigned long write_err, void *context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001439 struct dm_snap_pending_exception *pe = context;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 struct dm_snapshot *s = pe->snap;
1441
1442 if (read_err || write_err)
1443 pending_complete(pe, 0);
1444
1445 else
1446 /* Update the metadata if we are persistent */
Jonathan Brassow493df712009-04-02 19:55:31 +01001447 s->store->type->commit_exception(s->store, &pe->e,
1448 commit_callback, pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449}
1450
1451/*
1452 * Dispatches the copy operation to kcopyd.
1453 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001454static void start_copy(struct dm_snap_pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455{
1456 struct dm_snapshot *s = pe->snap;
Heinz Mauelshagen22a1ceb2008-04-24 21:43:17 +01001457 struct dm_io_region src, dest;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 struct block_device *bdev = s->origin->bdev;
1459 sector_t dev_size;
1460
1461 dev_size = get_dev_size(bdev);
1462
1463 src.bdev = bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001464 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
Mikulas Patockadf96eee2009-10-16 23:18:17 +01001465 src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001467 dest.bdev = s->cow->bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001468 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 dest.count = src.count;
1470
1471 /* Hand over to kcopyd */
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001472 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473}
1474
Mikulas Patocka29138082009-04-02 19:55:25 +01001475static struct dm_snap_pending_exception *
1476__lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1477{
Jon Brassow3510cb92009-12-10 23:52:11 +00001478 struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01001479
1480 if (!e)
1481 return NULL;
1482
1483 return container_of(e, struct dm_snap_pending_exception, e);
1484}
1485
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486/*
1487 * Looks to see if this snapshot already has a pending exception
1488 * for this chunk, otherwise it allocates a new one and inserts
1489 * it into the pending table.
1490 *
1491 * NOTE: a write lock must be held on snap->lock before calling
1492 * this.
1493 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001494static struct dm_snap_pending_exception *
Mikulas Patockac6621392009-04-02 19:55:25 +01001495__find_pending_exception(struct dm_snapshot *s,
1496 struct dm_snap_pending_exception *pe, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497{
Mikulas Patockac6621392009-04-02 19:55:25 +01001498 struct dm_snap_pending_exception *pe2;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001499
Mikulas Patocka29138082009-04-02 19:55:25 +01001500 pe2 = __lookup_pending_exception(s, chunk);
1501 if (pe2) {
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001502 free_pending_exception(pe);
Mikulas Patocka29138082009-04-02 19:55:25 +01001503 return pe2;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001504 }
1505
1506 pe->e.old_chunk = chunk;
1507 bio_list_init(&pe->origin_bios);
1508 bio_list_init(&pe->snapshot_bios);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001509 pe->started = 0;
1510
Jonathan Brassow493df712009-04-02 19:55:31 +01001511 if (s->store->type->prepare_exception(s->store, &pe->e)) {
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001512 free_pending_exception(pe);
1513 return NULL;
1514 }
1515
Jon Brassow3510cb92009-12-10 23:52:11 +00001516 dm_insert_exception(&s->pending, &pe->e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001517
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 return pe;
1519}
1520
Jon Brassow1d4989c2009-12-10 23:52:10 +00001521static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
Milan Brozd74f81f2008-02-08 02:11:27 +00001522 struct bio *bio, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523{
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001524 bio->bi_bdev = s->cow->bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001525 bio->bi_sector = chunk_to_sector(s->store,
1526 dm_chunk_number(e->new_chunk) +
1527 (chunk - e->old_chunk)) +
1528 (bio->bi_sector &
1529 s->store->chunk_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530}
1531
1532static int snapshot_map(struct dm_target *ti, struct bio *bio,
1533 union map_info *map_context)
1534{
Jon Brassow1d4989c2009-12-10 23:52:10 +00001535 struct dm_exception *e;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001536 struct dm_snapshot *s = ti->private;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001537 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 chunk_t chunk;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001539 struct dm_snap_pending_exception *pe = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
Tejun Heod87f4c12010-09-03 11:56:19 +02001541 if (bio->bi_rw & REQ_FLUSH) {
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001542 bio->bi_bdev = s->cow->bdev;
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01001543 return DM_MAPIO_REMAPPED;
1544 }
1545
Jonathan Brassow71fab002009-04-02 19:55:33 +01001546 chunk = sector_to_chunk(s->store, bio->bi_sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
1548 /* Full snapshots are not usable */
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001549 /* To get here the table must be live so s->active is always set. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 if (!s->valid)
Alasdair G Kergonf6a80ea2005-07-12 15:53:01 -07001551 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001553 /* FIXME: should only take write lock if we need
1554 * to copy an exception */
1555 down_write(&s->lock);
1556
1557 if (!s->valid) {
1558 r = -EIO;
1559 goto out_unlock;
1560 }
1561
1562 /* If the block is already remapped - use that, else remap it */
Jon Brassow3510cb92009-12-10 23:52:11 +00001563 e = dm_lookup_exception(&s->complete, chunk);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001564 if (e) {
Milan Brozd74f81f2008-02-08 02:11:27 +00001565 remap_exception(s, e, bio, chunk);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001566 goto out_unlock;
1567 }
1568
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 /*
1570 * Write to snapshot - higher level takes care of RW/RO
1571 * flags so we should only get this if we are
1572 * writeable.
1573 */
1574 if (bio_rw(bio) == WRITE) {
Mikulas Patocka29138082009-04-02 19:55:25 +01001575 pe = __lookup_pending_exception(s, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001576 if (!pe) {
Mikulas Patockac6621392009-04-02 19:55:25 +01001577 up_write(&s->lock);
1578 pe = alloc_pending_exception(s);
1579 down_write(&s->lock);
1580
1581 if (!s->valid) {
1582 free_pending_exception(pe);
1583 r = -EIO;
1584 goto out_unlock;
1585 }
1586
Jon Brassow3510cb92009-12-10 23:52:11 +00001587 e = dm_lookup_exception(&s->complete, chunk);
Mikulas Patocka35bf6592009-04-02 19:55:26 +01001588 if (e) {
1589 free_pending_exception(pe);
1590 remap_exception(s, e, bio, chunk);
1591 goto out_unlock;
1592 }
1593
Mikulas Patockac6621392009-04-02 19:55:25 +01001594 pe = __find_pending_exception(s, pe, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01001595 if (!pe) {
1596 __invalidate_snapshot(s, -ENOMEM);
1597 r = -EIO;
1598 goto out_unlock;
1599 }
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001600 }
1601
Milan Brozd74f81f2008-02-08 02:11:27 +00001602 remap_exception(s, &pe->e, bio, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001603 bio_list_add(&pe->snapshot_bios, bio);
1604
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001605 r = DM_MAPIO_SUBMITTED;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001606
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001607 if (!pe->started) {
1608 /* this is protected by snap->lock */
1609 pe->started = 1;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001610 up_write(&s->lock);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001611 start_copy(pe);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001612 goto out;
1613 }
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001614 } else {
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001615 bio->bi_bdev = s->origin->bdev;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001616 map_context->ptr = track_chunk(s, chunk);
1617 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001619out_unlock:
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001620 up_write(&s->lock);
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001621out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 return r;
1623}
1624
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001625/*
1626 * A snapshot-merge target behaves like a combination of a snapshot
1627 * target and a snapshot-origin target. It only generates new
1628 * exceptions in other snapshots and not in the one that is being
1629 * merged.
1630 *
1631 * For each chunk, if there is an existing exception, it is used to
1632 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
1633 * which in turn might generate exceptions in other snapshots.
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001634 * If merging is currently taking place on the chunk in question, the
1635 * I/O is deferred by adding it to s->bios_queued_during_merge.
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001636 */
1637static int snapshot_merge_map(struct dm_target *ti, struct bio *bio,
1638 union map_info *map_context)
1639{
1640 struct dm_exception *e;
1641 struct dm_snapshot *s = ti->private;
1642 int r = DM_MAPIO_REMAPPED;
1643 chunk_t chunk;
1644
Tejun Heod87f4c12010-09-03 11:56:19 +02001645 if (bio->bi_rw & REQ_FLUSH) {
Mike Snitzer57cba5d2010-08-12 04:14:04 +01001646 if (!map_context->target_request_nr)
Mike Snitzer10b81062009-12-10 23:52:31 +00001647 bio->bi_bdev = s->origin->bdev;
1648 else
1649 bio->bi_bdev = s->cow->bdev;
1650 map_context->ptr = NULL;
1651 return DM_MAPIO_REMAPPED;
1652 }
1653
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001654 chunk = sector_to_chunk(s->store, bio->bi_sector);
1655
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001656 down_write(&s->lock);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001657
Mikulas Patockad2fdb772009-12-10 23:52:36 +00001658 /* Full merging snapshots are redirected to the origin */
1659 if (!s->valid)
1660 goto redirect_to_origin;
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001661
1662 /* If the block is already remapped - use that */
1663 e = dm_lookup_exception(&s->complete, chunk);
1664 if (e) {
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001665 /* Queue writes overlapping with chunks being merged */
1666 if (bio_rw(bio) == WRITE &&
1667 chunk >= s->first_merging_chunk &&
1668 chunk < (s->first_merging_chunk +
1669 s->num_merging_chunks)) {
1670 bio->bi_bdev = s->origin->bdev;
1671 bio_list_add(&s->bios_queued_during_merge, bio);
1672 r = DM_MAPIO_SUBMITTED;
1673 goto out_unlock;
1674 }
Mikulas Patocka17aa0332009-12-10 23:52:33 +00001675
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001676 remap_exception(s, e, bio, chunk);
Mikulas Patocka17aa0332009-12-10 23:52:33 +00001677
1678 if (bio_rw(bio) == WRITE)
1679 map_context->ptr = track_chunk(s, chunk);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001680 goto out_unlock;
1681 }
1682
Mikulas Patockad2fdb772009-12-10 23:52:36 +00001683redirect_to_origin:
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001684 bio->bi_bdev = s->origin->bdev;
1685
1686 if (bio_rw(bio) == WRITE) {
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001687 up_write(&s->lock);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001688 return do_origin(s->origin, bio);
1689 }
1690
1691out_unlock:
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001692 up_write(&s->lock);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001693
1694 return r;
1695}
1696
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001697static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1698 int error, union map_info *map_context)
1699{
1700 struct dm_snapshot *s = ti->private;
1701 struct dm_snap_tracked_chunk *c = map_context->ptr;
1702
1703 if (c)
1704 stop_tracking_chunk(s, c);
1705
1706 return 0;
1707}
1708
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001709static void snapshot_merge_presuspend(struct dm_target *ti)
1710{
1711 struct dm_snapshot *s = ti->private;
1712
1713 stop_merge(s);
1714}
1715
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001716static int snapshot_preresume(struct dm_target *ti)
1717{
1718 int r = 0;
1719 struct dm_snapshot *s = ti->private;
1720 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1721
1722 down_read(&_origins_lock);
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +00001723 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001724 if (snap_src && snap_dest) {
1725 down_read(&snap_src->lock);
1726 if (s == snap_src) {
1727 DMERR("Unable to resume snapshot source until "
1728 "handover completes.");
1729 r = -EINVAL;
Mike Snitzerb83b2f22011-01-13 19:59:59 +00001730 } else if (!dm_suspended(snap_src->ti)) {
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001731 DMERR("Unable to perform snapshot handover until "
1732 "source is suspended.");
1733 r = -EINVAL;
1734 }
1735 up_read(&snap_src->lock);
1736 }
1737 up_read(&_origins_lock);
1738
1739 return r;
1740}
1741
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742static void snapshot_resume(struct dm_target *ti)
1743{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001744 struct dm_snapshot *s = ti->private;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001745 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1746
1747 down_read(&_origins_lock);
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +00001748 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001749 if (snap_src && snap_dest) {
1750 down_write(&snap_src->lock);
1751 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1752 __handover_exceptions(snap_src, snap_dest);
1753 up_write(&snap_dest->lock);
1754 up_write(&snap_src->lock);
1755 }
1756 up_read(&_origins_lock);
1757
1758 /* Now we have correct chunk size, reregister */
1759 reregister_snapshot(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001761 down_write(&s->lock);
1762 s->active = 1;
1763 up_write(&s->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764}
1765
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001766static sector_t get_origin_minimum_chunksize(struct block_device *bdev)
1767{
1768 sector_t min_chunksize;
1769
1770 down_read(&_origins_lock);
1771 min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
1772 up_read(&_origins_lock);
1773
1774 return min_chunksize;
1775}
1776
1777static void snapshot_merge_resume(struct dm_target *ti)
1778{
1779 struct dm_snapshot *s = ti->private;
1780
1781 /*
1782 * Handover exceptions from existing snapshot.
1783 */
1784 snapshot_resume(ti);
1785
1786 /*
1787 * snapshot-merge acts as an origin, so set ti->split_io
1788 */
1789 ti->split_io = get_origin_minimum_chunksize(s->origin->bdev);
1790
1791 start_merge(s);
1792}
1793
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794static int snapshot_status(struct dm_target *ti, status_type_t type,
1795 char *result, unsigned int maxlen)
1796{
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01001797 unsigned sz = 0;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001798 struct dm_snapshot *snap = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799
1800 switch (type) {
1801 case STATUSTYPE_INFO:
Mikulas Patocka94e765722009-12-10 23:51:53 +00001802
1803 down_write(&snap->lock);
1804
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 if (!snap->valid)
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01001806 DMEMIT("Invalid");
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +00001807 else if (snap->merge_failed)
1808 DMEMIT("Merge failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 else {
Mike Snitzer985903b2009-12-10 23:52:11 +00001810 if (snap->store->type->usage) {
1811 sector_t total_sectors, sectors_allocated,
1812 metadata_sectors;
1813 snap->store->type->usage(snap->store,
1814 &total_sectors,
1815 &sectors_allocated,
1816 &metadata_sectors);
1817 DMEMIT("%llu/%llu %llu",
1818 (unsigned long long)sectors_allocated,
1819 (unsigned long long)total_sectors,
1820 (unsigned long long)metadata_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 }
1822 else
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01001823 DMEMIT("Unknown");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 }
Mikulas Patocka94e765722009-12-10 23:51:53 +00001825
1826 up_write(&snap->lock);
1827
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 break;
1829
1830 case STATUSTYPE_TABLE:
1831 /*
1832 * kdevname returns a static pointer so we need
1833 * to make private copies if the output is to
1834 * make sense.
1835 */
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001836 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
Jonathan Brassow1e302a92009-04-02 19:55:35 +01001837 snap->store->type->status(snap->store, type, result + sz,
1838 maxlen - sz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 break;
1840 }
1841
1842 return 0;
1843}
1844
Mike Snitzer8811f462009-09-04 20:40:19 +01001845static int snapshot_iterate_devices(struct dm_target *ti,
1846 iterate_devices_callout_fn fn, void *data)
1847{
1848 struct dm_snapshot *snap = ti->private;
Mikulas Patocka1e5554c2010-08-12 04:13:50 +01001849 int r;
Mike Snitzer8811f462009-09-04 20:40:19 +01001850
Mikulas Patocka1e5554c2010-08-12 04:13:50 +01001851 r = fn(ti, snap->origin, 0, ti->len, data);
1852
1853 if (!r)
1854 r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data);
1855
1856 return r;
Mike Snitzer8811f462009-09-04 20:40:19 +01001857}
1858
1859
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860/*-----------------------------------------------------------------
1861 * Origin methods
1862 *---------------------------------------------------------------*/
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001863
1864/*
1865 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1866 * supplied bio was ignored. The caller may submit it immediately.
1867 * (No remapping actually occurs as the origin is always a direct linear
1868 * map.)
1869 *
1870 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1871 * and any supplied bio is added to a list to be submitted once all
1872 * the necessary exceptions exist.
1873 */
1874static int __origin_write(struct list_head *snapshots, sector_t sector,
1875 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876{
Mikulas Patocka515ad662009-12-10 23:52:30 +00001877 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 struct dm_snapshot *snap;
Jon Brassow1d4989c2009-12-10 23:52:10 +00001879 struct dm_exception *e;
Mikulas Patocka515ad662009-12-10 23:52:30 +00001880 struct dm_snap_pending_exception *pe;
1881 struct dm_snap_pending_exception *pe_to_start_now = NULL;
1882 struct dm_snap_pending_exception *pe_to_start_last = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 chunk_t chunk;
1884
1885 /* Do all the snapshots on this origin */
1886 list_for_each_entry (snap, snapshots, list) {
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001887 /*
1888 * Don't make new exceptions in a merging snapshot
1889 * because it has effectively been deleted
1890 */
1891 if (dm_target_is_snapshot_merge(snap->ti))
1892 continue;
1893
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001894 down_write(&snap->lock);
1895
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001896 /* Only deal with valid and active snapshots */
1897 if (!snap->valid || !snap->active)
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001898 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899
Alasdair G Kergond5e404c2005-07-12 15:53:05 -07001900 /* Nothing to do if writing beyond end of snapshot */
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001901 if (sector >= dm_table_get_size(snap->ti->table))
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001902 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903
1904 /*
1905 * Remember, different snapshots can have
1906 * different chunk sizes.
1907 */
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001908 chunk = sector_to_chunk(snap->store, sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909
1910 /*
1911 * Check exception table to see if block
1912 * is already remapped in this snapshot
1913 * and trigger an exception if not.
1914 */
Jon Brassow3510cb92009-12-10 23:52:11 +00001915 e = dm_lookup_exception(&snap->complete, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001916 if (e)
1917 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918
Mikulas Patocka29138082009-04-02 19:55:25 +01001919 pe = __lookup_pending_exception(snap, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001920 if (!pe) {
Mikulas Patockac6621392009-04-02 19:55:25 +01001921 up_write(&snap->lock);
1922 pe = alloc_pending_exception(snap);
1923 down_write(&snap->lock);
1924
1925 if (!snap->valid) {
1926 free_pending_exception(pe);
1927 goto next_snapshot;
1928 }
1929
Jon Brassow3510cb92009-12-10 23:52:11 +00001930 e = dm_lookup_exception(&snap->complete, chunk);
Mikulas Patocka35bf6592009-04-02 19:55:26 +01001931 if (e) {
1932 free_pending_exception(pe);
1933 goto next_snapshot;
1934 }
1935
Mikulas Patockac6621392009-04-02 19:55:25 +01001936 pe = __find_pending_exception(snap, pe, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01001937 if (!pe) {
1938 __invalidate_snapshot(snap, -ENOMEM);
1939 goto next_snapshot;
1940 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 }
1942
Mikulas Patocka515ad662009-12-10 23:52:30 +00001943 r = DM_MAPIO_SUBMITTED;
1944
1945 /*
1946 * If an origin bio was supplied, queue it to wait for the
1947 * completion of this exception, and start this one last,
1948 * at the end of the function.
1949 */
1950 if (bio) {
1951 bio_list_add(&pe->origin_bios, bio);
1952 bio = NULL;
1953
1954 if (!pe->started) {
1955 pe->started = 1;
1956 pe_to_start_last = pe;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001957 }
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001958 }
1959
1960 if (!pe->started) {
1961 pe->started = 1;
Mikulas Patocka515ad662009-12-10 23:52:30 +00001962 pe_to_start_now = pe;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001963 }
1964
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001965next_snapshot:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 up_write(&snap->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967
Mikulas Patocka515ad662009-12-10 23:52:30 +00001968 if (pe_to_start_now) {
1969 start_copy(pe_to_start_now);
1970 pe_to_start_now = NULL;
1971 }
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001972 }
1973
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 /*
Mikulas Patocka515ad662009-12-10 23:52:30 +00001975 * Submit the exception against which the bio is queued last,
1976 * to give the other exceptions a head start.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 */
Mikulas Patocka515ad662009-12-10 23:52:30 +00001978 if (pe_to_start_last)
1979 start_copy(pe_to_start_last);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980
1981 return r;
1982}
1983
1984/*
1985 * Called on a write from the origin driver.
1986 */
1987static int do_origin(struct dm_dev *origin, struct bio *bio)
1988{
1989 struct origin *o;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001990 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991
1992 down_read(&_origins_lock);
1993 o = __lookup_origin(origin->bdev);
1994 if (o)
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001995 r = __origin_write(&o->snapshots, bio->bi_sector, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 up_read(&_origins_lock);
1997
1998 return r;
1999}
2000
2001/*
Mikulas Patocka73dfd072009-12-10 23:52:34 +00002002 * Trigger exceptions in all non-merging snapshots.
2003 *
2004 * The chunk size of the merging snapshot may be larger than the chunk
2005 * size of some other snapshot so we may need to reallocate multiple
2006 * chunks in other snapshots.
2007 *
2008 * We scan all the overlapping exceptions in the other snapshots.
2009 * Returns 1 if anything was reallocated and must be waited for,
2010 * otherwise returns 0.
2011 *
2012 * size must be a multiple of merging_snap's chunk_size.
2013 */
2014static int origin_write_extent(struct dm_snapshot *merging_snap,
2015 sector_t sector, unsigned size)
2016{
2017 int must_wait = 0;
2018 sector_t n;
2019 struct origin *o;
2020
2021 /*
2022 * The origin's __minimum_chunk_size() got stored in split_io
2023 * by snapshot_merge_resume().
2024 */
2025 down_read(&_origins_lock);
2026 o = __lookup_origin(merging_snap->origin->bdev);
2027 for (n = 0; n < size; n += merging_snap->ti->split_io)
2028 if (__origin_write(&o->snapshots, sector + n, NULL) ==
2029 DM_MAPIO_SUBMITTED)
2030 must_wait = 1;
2031 up_read(&_origins_lock);
2032
2033 return must_wait;
2034}
2035
2036/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 * Origin: maps a linear range of a device, with hooks for snapshotting.
2038 */
2039
2040/*
2041 * Construct an origin mapping: <dev_path>
2042 * The context for an origin is merely a 'struct dm_dev *'
2043 * pointing to the real device.
2044 */
2045static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2046{
2047 int r;
2048 struct dm_dev *dev;
2049
2050 if (argc != 1) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07002051 ti->error = "origin: incorrect number of arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 return -EINVAL;
2053 }
2054
Nikanth Karthikesan8215d6e2010-03-06 02:32:27 +00002055 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 if (r) {
2057 ti->error = "Cannot get target device";
2058 return r;
2059 }
2060
2061 ti->private = dev;
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01002062 ti->num_flush_requests = 1;
2063
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 return 0;
2065}
2066
2067static void origin_dtr(struct dm_target *ti)
2068{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002069 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 dm_put_device(ti, dev);
2071}
2072
2073static int origin_map(struct dm_target *ti, struct bio *bio,
2074 union map_info *map_context)
2075{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002076 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 bio->bi_bdev = dev->bdev;
2078
Tejun Heod87f4c12010-09-03 11:56:19 +02002079 if (bio->bi_rw & REQ_FLUSH)
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01002080 return DM_MAPIO_REMAPPED;
2081
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 /* Only tell snapshots if this is a write */
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08002083 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084}
2085
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086/*
2087 * Set the target "split_io" field to the minimum of all the snapshots'
2088 * chunk sizes.
2089 */
2090static void origin_resume(struct dm_target *ti)
2091{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002092 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002094 ti->split_io = get_origin_minimum_chunksize(dev->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095}
2096
2097static int origin_status(struct dm_target *ti, status_type_t type, char *result,
2098 unsigned int maxlen)
2099{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002100 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101
2102 switch (type) {
2103 case STATUSTYPE_INFO:
2104 result[0] = '\0';
2105 break;
2106
2107 case STATUSTYPE_TABLE:
2108 snprintf(result, maxlen, "%s", dev->name);
2109 break;
2110 }
2111
2112 return 0;
2113}
2114
Mikulas Patockab1d55522010-08-12 04:14:02 +01002115static int origin_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2116 struct bio_vec *biovec, int max_size)
2117{
2118 struct dm_dev *dev = ti->private;
2119 struct request_queue *q = bdev_get_queue(dev->bdev);
2120
2121 if (!q->merge_bvec_fn)
2122 return max_size;
2123
2124 bvm->bi_bdev = dev->bdev;
2125 bvm->bi_sector = bvm->bi_sector;
2126
2127 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2128}
2129
Mike Snitzer8811f462009-09-04 20:40:19 +01002130static int origin_iterate_devices(struct dm_target *ti,
2131 iterate_devices_callout_fn fn, void *data)
2132{
2133 struct dm_dev *dev = ti->private;
2134
2135 return fn(ti, dev, 0, ti->len, data);
2136}
2137
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138static struct target_type origin_target = {
2139 .name = "snapshot-origin",
Mike Snitzerb83b2f22011-01-13 19:59:59 +00002140 .version = {1, 7, 1},
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 .module = THIS_MODULE,
2142 .ctr = origin_ctr,
2143 .dtr = origin_dtr,
2144 .map = origin_map,
2145 .resume = origin_resume,
2146 .status = origin_status,
Mikulas Patockab1d55522010-08-12 04:14:02 +01002147 .merge = origin_merge,
Mike Snitzer8811f462009-09-04 20:40:19 +01002148 .iterate_devices = origin_iterate_devices,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149};
2150
2151static struct target_type snapshot_target = {
2152 .name = "snapshot",
Mike Snitzerb83b2f22011-01-13 19:59:59 +00002153 .version = {1, 10, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 .module = THIS_MODULE,
2155 .ctr = snapshot_ctr,
2156 .dtr = snapshot_dtr,
2157 .map = snapshot_map,
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002158 .end_io = snapshot_end_io,
Mike Snitzerc1f0c182009-12-10 23:52:24 +00002159 .preresume = snapshot_preresume,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 .resume = snapshot_resume,
2161 .status = snapshot_status,
Mike Snitzer8811f462009-09-04 20:40:19 +01002162 .iterate_devices = snapshot_iterate_devices,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163};
2164
Mikulas Patockad698aa42009-12-10 23:52:30 +00002165static struct target_type merge_target = {
2166 .name = dm_snapshot_merge_target_name,
Mike Snitzerb83b2f22011-01-13 19:59:59 +00002167 .version = {1, 1, 0},
Mikulas Patockad698aa42009-12-10 23:52:30 +00002168 .module = THIS_MODULE,
2169 .ctr = snapshot_ctr,
2170 .dtr = snapshot_dtr,
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00002171 .map = snapshot_merge_map,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002172 .end_io = snapshot_end_io,
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002173 .presuspend = snapshot_merge_presuspend,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002174 .preresume = snapshot_preresume,
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002175 .resume = snapshot_merge_resume,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002176 .status = snapshot_status,
2177 .iterate_devices = snapshot_iterate_devices,
2178};
2179
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180static int __init dm_snapshot_init(void)
2181{
2182 int r;
2183
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00002184 r = dm_exception_store_init();
2185 if (r) {
2186 DMERR("Failed to initialize exception stores");
2187 return r;
2188 }
2189
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 r = dm_register_target(&snapshot_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002191 if (r < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 DMERR("snapshot target register failed %d", r);
Jonathan Brassow034a1862009-10-16 23:18:14 +01002193 goto bad_register_snapshot_target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 }
2195
2196 r = dm_register_target(&origin_target);
2197 if (r < 0) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07002198 DMERR("Origin target register failed %d", r);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002199 goto bad_register_origin_target;
2200 }
2201
2202 r = dm_register_target(&merge_target);
2203 if (r < 0) {
2204 DMERR("Merge target register failed %d", r);
2205 goto bad_register_merge_target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 }
2207
2208 r = init_origin_hash();
2209 if (r) {
2210 DMERR("init_origin_hash failed.");
Mikulas Patockad698aa42009-12-10 23:52:30 +00002211 goto bad_origin_hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 }
2213
Jon Brassow1d4989c2009-12-10 23:52:10 +00002214 exception_cache = KMEM_CACHE(dm_exception, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 if (!exception_cache) {
2216 DMERR("Couldn't create exception cache.");
2217 r = -ENOMEM;
Mikulas Patockad698aa42009-12-10 23:52:30 +00002218 goto bad_exception_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 }
2220
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002221 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 if (!pending_cache) {
2223 DMERR("Couldn't create pending cache.");
2224 r = -ENOMEM;
Mikulas Patockad698aa42009-12-10 23:52:30 +00002225 goto bad_pending_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 }
2227
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002228 tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
2229 if (!tracked_chunk_cache) {
2230 DMERR("Couldn't create cache to track chunks in use.");
2231 r = -ENOMEM;
Mikulas Patockad698aa42009-12-10 23:52:30 +00002232 goto bad_tracked_chunk_cache;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002233 }
2234
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 return 0;
2236
Mikulas Patockad698aa42009-12-10 23:52:30 +00002237bad_tracked_chunk_cache:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 kmem_cache_destroy(pending_cache);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002239bad_pending_cache:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 kmem_cache_destroy(exception_cache);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002241bad_exception_cache:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 exit_origin_hash();
Mikulas Patockad698aa42009-12-10 23:52:30 +00002243bad_origin_hash:
2244 dm_unregister_target(&merge_target);
2245bad_register_merge_target:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 dm_unregister_target(&origin_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002247bad_register_origin_target:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 dm_unregister_target(&snapshot_target);
Jonathan Brassow034a1862009-10-16 23:18:14 +01002249bad_register_snapshot_target:
2250 dm_exception_store_exit();
Mikulas Patockad698aa42009-12-10 23:52:30 +00002251
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 return r;
2253}
2254
2255static void __exit dm_snapshot_exit(void)
2256{
Mikulas Patocka10d3bd02009-01-06 03:04:58 +00002257 dm_unregister_target(&snapshot_target);
2258 dm_unregister_target(&origin_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002259 dm_unregister_target(&merge_target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260
2261 exit_origin_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 kmem_cache_destroy(pending_cache);
2263 kmem_cache_destroy(exception_cache);
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002264 kmem_cache_destroy(tracked_chunk_cache);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00002265
2266 dm_exception_store_exit();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267}
2268
2269/* Module hooks */
2270module_init(dm_snapshot_init);
2271module_exit(dm_snapshot_exit);
2272
2273MODULE_DESCRIPTION(DM_NAME " snapshot target");
2274MODULE_AUTHOR("Joe Thornber");
2275MODULE_LICENSE("GPL");