blob: 864b03f477276f9c01bee8fd34a25e2303af1116 [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
Mikulas Patocka230c83a2013-11-29 18:13:37 -050069 /* Protected by "lock" */
70 sector_t exception_start_sequence;
71
72 /* Protected by kcopyd single-threaded callback */
73 sector_t exception_complete_sequence;
74
75 /*
76 * A list of pending exceptions that completed out of order.
77 * Protected by kcopyd single-threaded callback.
78 */
79 struct list_head out_of_order_list;
80
Mike Snitzer924e6002010-03-06 02:32:33 +000081 mempool_t *pending_pool;
82
Jon Brassow191437a2009-12-10 23:52:10 +000083 struct dm_exception_table pending;
84 struct dm_exception_table complete;
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010085
86 /*
87 * pe_lock protects all pending_exception operations and access
88 * as well as the snapshot_bios list.
89 */
90 spinlock_t pe_lock;
91
Mike Snitzer924e6002010-03-06 02:32:33 +000092 /* Chunks with outstanding reads */
93 spinlock_t tracked_chunk_lock;
Mike Snitzer924e6002010-03-06 02:32:33 +000094 struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
95
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010096 /* The on disk metadata handler */
97 struct dm_exception_store *store;
98
99 struct dm_kcopyd_client *kcopyd_client;
100
Mike Snitzer924e6002010-03-06 02:32:33 +0000101 /* Wait for events based on state_bits */
102 unsigned long state_bits;
103
104 /* Range of chunks currently being merged. */
105 chunk_t first_merging_chunk;
106 int num_merging_chunks;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000107
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +0000108 /*
109 * The merge operation failed if this flag is set.
110 * Failure modes are handled as follows:
111 * - I/O error reading the header
112 * => don't load the target; abort.
113 * - Header does not have "valid" flag set
114 * => use the origin; forget about the snapshot.
115 * - I/O error when reading exceptions
116 * => don't load the target; abort.
117 * (We can't use the intermediate origin state.)
118 * - I/O error while merging
119 * => stop merging; set merge_failed; process I/O normally.
120 */
121 int merge_failed;
122
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000123 /*
124 * Incoming bios that overlap with chunks being merged must wait
125 * for them to be committed.
126 */
127 struct bio_list bios_queued_during_merge;
Jonathan Brassowccc45ea2009-04-02 19:55:34 +0100128};
129
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000130/*
131 * state_bits:
132 * RUNNING_MERGE - Merge operation is in progress.
133 * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
134 * cleared afterwards.
135 */
136#define RUNNING_MERGE 0
137#define SHUTDOWN_MERGE 1
138
Mikulas Patockadf5d2e92013-03-01 22:45:49 +0000139DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
140 "A percentage of time allocated for copy on write");
141
Mikulas Patockac2411042010-08-12 04:13:51 +0100142struct dm_dev *dm_snap_origin(struct dm_snapshot *s)
143{
144 return s->origin;
145}
146EXPORT_SYMBOL(dm_snap_origin);
147
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000148struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
149{
150 return s->cow;
151}
152EXPORT_SYMBOL(dm_snap_cow);
153
Jonathan Brassowccc45ea2009-04-02 19:55:34 +0100154static sector_t chunk_to_sector(struct dm_exception_store *store,
155 chunk_t chunk)
156{
157 return chunk << store->chunk_shift;
158}
159
160static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
161{
162 /*
163 * There is only ever one instance of a particular block
164 * device so we can compare pointers safely.
165 */
166 return lhs == rhs;
167}
168
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100169struct dm_snap_pending_exception {
Jon Brassow1d4989c2009-12-10 23:52:10 +0000170 struct dm_exception e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171
172 /*
173 * Origin buffers waiting for this to complete are held
174 * in a bio list
175 */
176 struct bio_list origin_bios;
177 struct bio_list snapshot_bios;
178
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 /* Pointer back to snapshot context */
180 struct dm_snapshot *snap;
181
182 /*
183 * 1 indicates the exception has already been sent to
184 * kcopyd.
185 */
186 int started;
Mikulas Patockaa6e50b42011-08-02 12:32:04 +0100187
Mikulas Patocka230c83a2013-11-29 18:13:37 -0500188 /* There was copying error. */
189 int copy_error;
190
191 /* A sequence number, it is used for in-order completion. */
192 sector_t exception_sequence;
193
194 struct list_head out_of_order_entry;
195
Mikulas Patockaa6e50b42011-08-02 12:32:04 +0100196 /*
197 * For writing a complete chunk, bypassing the copy.
198 */
199 struct bio *full_bio;
200 bio_end_io_t *full_bio_end_io;
201 void *full_bio_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202};
203
204/*
205 * Hash table mapping origin volumes to lists of snapshots and
206 * a lock to protect it
207 */
Christoph Lametere18b8902006-12-06 20:33:20 -0800208static struct kmem_cache *exception_cache;
209static struct kmem_cache *pending_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100211struct dm_snap_tracked_chunk {
212 struct hlist_node node;
213 chunk_t chunk;
214};
215
Mikulas Patockaee180262012-12-21 20:23:41 +0000216static void init_tracked_chunk(struct bio *bio)
217{
218 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
219 INIT_HLIST_NODE(&c->node);
220}
221
222static bool is_bio_tracked(struct bio *bio)
223{
224 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
225 return !hlist_unhashed(&c->node);
226}
227
228static void track_chunk(struct dm_snapshot *s, struct bio *bio, chunk_t chunk)
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100229{
Mikulas Patocka42bc9542012-12-21 20:23:38 +0000230 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100231
232 c->chunk = chunk;
233
Mikulas Patocka9aa0c0e2012-12-21 20:23:33 +0000234 spin_lock_irq(&s->tracked_chunk_lock);
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100235 hlist_add_head(&c->node,
236 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
Mikulas Patocka9aa0c0e2012-12-21 20:23:33 +0000237 spin_unlock_irq(&s->tracked_chunk_lock);
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100238}
239
Mikulas Patockaee180262012-12-21 20:23:41 +0000240static void stop_tracking_chunk(struct dm_snapshot *s, struct bio *bio)
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100241{
Mikulas Patockaee180262012-12-21 20:23:41 +0000242 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100243 unsigned long flags;
244
245 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
246 hlist_del(&c->node);
247 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100248}
249
Mikulas Patockaa8d41b52008-07-21 12:00:34 +0100250static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
251{
252 struct dm_snap_tracked_chunk *c;
Mikulas Patockaa8d41b52008-07-21 12:00:34 +0100253 int found = 0;
254
255 spin_lock_irq(&s->tracked_chunk_lock);
256
Sasha Levinb67bfe02013-02-27 17:06:00 -0800257 hlist_for_each_entry(c,
Mikulas Patockaa8d41b52008-07-21 12:00:34 +0100258 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
259 if (c->chunk == chunk) {
260 found = 1;
261 break;
262 }
263 }
264
265 spin_unlock_irq(&s->tracked_chunk_lock);
266
267 return found;
268}
269
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270/*
Mike Snitzer615d1eb2009-12-10 23:52:29 +0000271 * This conflicting I/O is extremely improbable in the caller,
272 * so msleep(1) is sufficient and there is no need for a wait queue.
273 */
274static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
275{
276 while (__chunk_is_tracked(s, chunk))
277 msleep(1);
278}
279
280/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 * One of these per registered origin, held in the snapshot_origins hash
282 */
283struct origin {
284 /* The origin device */
285 struct block_device *bdev;
286
287 struct list_head hash_list;
288
289 /* List of snapshots for this origin */
290 struct list_head snapshots;
291};
292
293/*
294 * Size of the hash table for origin volumes. If we make this
295 * the size of the minors list then it should be nearly perfect
296 */
297#define ORIGIN_HASH_SIZE 256
298#define ORIGIN_MASK 0xFF
299static struct list_head *_origins;
300static struct rw_semaphore _origins_lock;
301
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000302static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
303static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
304static uint64_t _pending_exceptions_done_count;
305
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306static int init_origin_hash(void)
307{
308 int i;
309
310 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
311 GFP_KERNEL);
312 if (!_origins) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700313 DMERR("unable to allocate memory");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 return -ENOMEM;
315 }
316
317 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
318 INIT_LIST_HEAD(_origins + i);
319 init_rwsem(&_origins_lock);
320
321 return 0;
322}
323
324static void exit_origin_hash(void)
325{
326 kfree(_origins);
327}
328
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100329static unsigned origin_hash(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330{
331 return bdev->bd_dev & ORIGIN_MASK;
332}
333
334static struct origin *__lookup_origin(struct block_device *origin)
335{
336 struct list_head *ol;
337 struct origin *o;
338
339 ol = &_origins[origin_hash(origin)];
340 list_for_each_entry (o, ol, hash_list)
341 if (bdev_equal(o->bdev, origin))
342 return o;
343
344 return NULL;
345}
346
347static void __insert_origin(struct origin *o)
348{
349 struct list_head *sl = &_origins[origin_hash(o->bdev)];
350 list_add_tail(&o->hash_list, sl);
351}
352
353/*
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000354 * _origins_lock must be held when calling this function.
355 * Returns number of snapshots registered using the supplied cow device, plus:
356 * snap_src - a snapshot suitable for use as a source of exception handover
357 * snap_dest - a snapshot capable of receiving exception handover.
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000358 * snap_merge - an existing snapshot-merge target linked to the same origin.
359 * There can be at most one snapshot-merge target. The parameter is optional.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000360 *
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000361 * Possible return values and states of snap_src and snap_dest.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000362 * 0: NULL, NULL - first new snapshot
363 * 1: snap_src, NULL - normal snapshot
364 * 2: snap_src, snap_dest - waiting for handover
365 * 2: snap_src, NULL - handed over, waiting for old to be deleted
366 * 1: NULL, snap_dest - source got destroyed without handover
367 */
368static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
369 struct dm_snapshot **snap_src,
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000370 struct dm_snapshot **snap_dest,
371 struct dm_snapshot **snap_merge)
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000372{
373 struct dm_snapshot *s;
374 struct origin *o;
375 int count = 0;
376 int active;
377
378 o = __lookup_origin(snap->origin->bdev);
379 if (!o)
380 goto out;
381
382 list_for_each_entry(s, &o->snapshots, list) {
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000383 if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
384 *snap_merge = s;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000385 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
386 continue;
387
388 down_read(&s->lock);
389 active = s->active;
390 up_read(&s->lock);
391
392 if (active) {
393 if (snap_src)
394 *snap_src = s;
395 } else if (snap_dest)
396 *snap_dest = s;
397
398 count++;
399 }
400
401out:
402 return count;
403}
404
405/*
406 * On success, returns 1 if this snapshot is a handover destination,
407 * otherwise returns 0.
408 */
409static int __validate_exception_handover(struct dm_snapshot *snap)
410{
411 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000412 struct dm_snapshot *snap_merge = NULL;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000413
414 /* Does snapshot need exceptions handed over to it? */
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000415 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
416 &snap_merge) == 2) ||
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000417 snap_dest) {
418 snap->ti->error = "Snapshot cow pairing for exception "
419 "table handover failed";
420 return -EINVAL;
421 }
422
423 /*
424 * If no snap_src was found, snap cannot become a handover
425 * destination.
426 */
427 if (!snap_src)
428 return 0;
429
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000430 /*
431 * Non-snapshot-merge handover?
432 */
433 if (!dm_target_is_snapshot_merge(snap->ti))
434 return 1;
435
436 /*
437 * Do not allow more than one merging snapshot.
438 */
439 if (snap_merge) {
440 snap->ti->error = "A snapshot is already merging.";
441 return -EINVAL;
442 }
443
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000444 if (!snap_src->store->type->prepare_merge ||
445 !snap_src->store->type->commit_merge) {
446 snap->ti->error = "Snapshot exception store does not "
447 "support snapshot-merge.";
448 return -EINVAL;
449 }
450
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000451 return 1;
452}
453
454static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
455{
456 struct dm_snapshot *l;
457
458 /* Sort the list according to chunk size, largest-first smallest-last */
459 list_for_each_entry(l, &o->snapshots, list)
460 if (l->store->chunk_size < s->store->chunk_size)
461 break;
462 list_add_tail(&s->list, &l->list);
463}
464
465/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 * Make a note of the snapshot and its origin so we can look it
467 * up when the origin has a write on it.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000468 *
469 * Also validate snapshot exception store handovers.
470 * On success, returns 1 if this registration is a handover destination,
471 * otherwise returns 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 */
473static int register_snapshot(struct dm_snapshot *snap)
474{
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000475 struct origin *o, *new_o = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 struct block_device *bdev = snap->origin->bdev;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000477 int r = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000479 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
480 if (!new_o)
481 return -ENOMEM;
482
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 down_write(&_origins_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000485 r = __validate_exception_handover(snap);
486 if (r < 0) {
487 kfree(new_o);
488 goto out;
489 }
490
491 o = __lookup_origin(bdev);
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000492 if (o)
493 kfree(new_o);
494 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 /* New origin */
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000496 o = new_o;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
498 /* Initialise the struct */
499 INIT_LIST_HEAD(&o->snapshots);
500 o->bdev = bdev;
501
502 __insert_origin(o);
503 }
504
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000505 __insert_snapshot(o, snap);
506
507out:
508 up_write(&_origins_lock);
509
510 return r;
511}
512
513/*
514 * Move snapshot to correct place in list according to chunk size.
515 */
516static void reregister_snapshot(struct dm_snapshot *s)
517{
518 struct block_device *bdev = s->origin->bdev;
519
520 down_write(&_origins_lock);
521
522 list_del(&s->list);
523 __insert_snapshot(__lookup_origin(bdev), s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
525 up_write(&_origins_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526}
527
528static void unregister_snapshot(struct dm_snapshot *s)
529{
530 struct origin *o;
531
532 down_write(&_origins_lock);
533 o = __lookup_origin(s->origin->bdev);
534
535 list_del(&s->list);
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000536 if (o && list_empty(&o->snapshots)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 list_del(&o->hash_list);
538 kfree(o);
539 }
540
541 up_write(&_origins_lock);
542}
543
544/*
545 * Implementation of the exception hash tables.
Milan Brozd74f81f2008-02-08 02:11:27 +0000546 * The lowest hash_shift bits of the chunk number are ignored, allowing
547 * some consecutive chunks to be grouped together.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 */
Jon Brassow3510cb92009-12-10 23:52:11 +0000549static int dm_exception_table_init(struct dm_exception_table *et,
550 uint32_t size, unsigned hash_shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551{
552 unsigned int i;
553
Milan Brozd74f81f2008-02-08 02:11:27 +0000554 et->hash_shift = hash_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 et->hash_mask = size - 1;
556 et->table = dm_vcalloc(size, sizeof(struct list_head));
557 if (!et->table)
558 return -ENOMEM;
559
560 for (i = 0; i < size; i++)
561 INIT_LIST_HEAD(et->table + i);
562
563 return 0;
564}
565
Jon Brassow3510cb92009-12-10 23:52:11 +0000566static void dm_exception_table_exit(struct dm_exception_table *et,
567 struct kmem_cache *mem)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
569 struct list_head *slot;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000570 struct dm_exception *ex, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 int i, size;
572
573 size = et->hash_mask + 1;
574 for (i = 0; i < size; i++) {
575 slot = et->table + i;
576
577 list_for_each_entry_safe (ex, next, slot, hash_list)
578 kmem_cache_free(mem, ex);
579 }
580
581 vfree(et->table);
582}
583
Jon Brassow191437a2009-12-10 23:52:10 +0000584static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585{
Milan Brozd74f81f2008-02-08 02:11:27 +0000586 return (chunk >> et->hash_shift) & et->hash_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587}
588
Jon Brassow3510cb92009-12-10 23:52:11 +0000589static void dm_remove_exception(struct dm_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590{
591 list_del(&e->hash_list);
592}
593
594/*
595 * Return the exception data for a sector, or NULL if not
596 * remapped.
597 */
Jon Brassow3510cb92009-12-10 23:52:11 +0000598static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
599 chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
601 struct list_head *slot;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000602 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603
604 slot = &et->table[exception_hash(et, chunk)];
605 list_for_each_entry (e, slot, hash_list)
Milan Brozd74f81f2008-02-08 02:11:27 +0000606 if (chunk >= e->old_chunk &&
607 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 return e;
609
610 return NULL;
611}
612
Mikulas Patocka119bc542014-01-13 19:13:36 -0500613static struct dm_exception *alloc_completed_exception(gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614{
Jon Brassow1d4989c2009-12-10 23:52:10 +0000615 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
Mikulas Patocka119bc542014-01-13 19:13:36 -0500617 e = kmem_cache_alloc(exception_cache, gfp);
618 if (!e && gfp == GFP_NOIO)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
620
621 return e;
622}
623
Jon Brassow3510cb92009-12-10 23:52:11 +0000624static void free_completed_exception(struct dm_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625{
626 kmem_cache_free(exception_cache, e);
627}
628
Mikulas Patocka92e86812008-07-21 12:00:35 +0100629static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630{
Mikulas Patocka92e86812008-07-21 12:00:35 +0100631 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
632 GFP_NOIO);
633
Mikulas Patocka879129d22008-10-30 13:33:16 +0000634 atomic_inc(&s->pending_exceptions_count);
Mikulas Patocka92e86812008-07-21 12:00:35 +0100635 pe->snap = s;
636
637 return pe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638}
639
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100640static void free_pending_exception(struct dm_snap_pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641{
Mikulas Patocka879129d22008-10-30 13:33:16 +0000642 struct dm_snapshot *s = pe->snap;
643
644 mempool_free(pe, s->pending_pool);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100645 smp_mb__before_atomic();
Mikulas Patocka879129d22008-10-30 13:33:16 +0000646 atomic_dec(&s->pending_exceptions_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647}
648
Jon Brassow3510cb92009-12-10 23:52:11 +0000649static void dm_insert_exception(struct dm_exception_table *eh,
650 struct dm_exception *new_e)
Milan Brozd74f81f2008-02-08 02:11:27 +0000651{
Milan Brozd74f81f2008-02-08 02:11:27 +0000652 struct list_head *l;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000653 struct dm_exception *e = NULL;
Milan Brozd74f81f2008-02-08 02:11:27 +0000654
655 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
656
657 /* Add immediately if this table doesn't support consecutive chunks */
658 if (!eh->hash_shift)
659 goto out;
660
661 /* List is ordered by old_chunk */
662 list_for_each_entry_reverse(e, l, hash_list) {
663 /* Insert after an existing chunk? */
664 if (new_e->old_chunk == (e->old_chunk +
665 dm_consecutive_chunk_count(e) + 1) &&
666 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
667 dm_consecutive_chunk_count(e) + 1)) {
668 dm_consecutive_chunk_count_inc(e);
Jon Brassow3510cb92009-12-10 23:52:11 +0000669 free_completed_exception(new_e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000670 return;
671 }
672
673 /* Insert before an existing chunk? */
674 if (new_e->old_chunk == (e->old_chunk - 1) &&
675 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
676 dm_consecutive_chunk_count_inc(e);
677 e->old_chunk--;
678 e->new_chunk--;
Jon Brassow3510cb92009-12-10 23:52:11 +0000679 free_completed_exception(new_e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000680 return;
681 }
682
683 if (new_e->old_chunk > e->old_chunk)
684 break;
685 }
686
687out:
688 list_add(&new_e->hash_list, e ? &e->hash_list : l);
689}
690
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000691/*
692 * Callback used by the exception stores to load exceptions when
693 * initialising.
694 */
695static int dm_add_exception(void *context, chunk_t old, chunk_t new)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696{
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000697 struct dm_snapshot *s = context;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000698 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
Mikulas Patocka119bc542014-01-13 19:13:36 -0500700 e = alloc_completed_exception(GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 if (!e)
702 return -ENOMEM;
703
704 e->old_chunk = old;
Milan Brozd74f81f2008-02-08 02:11:27 +0000705
706 /* Consecutive_count is implicitly initialised to zero */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 e->new_chunk = new;
Milan Brozd74f81f2008-02-08 02:11:27 +0000708
Jon Brassow3510cb92009-12-10 23:52:11 +0000709 dm_insert_exception(&s->complete, e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000710
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 return 0;
712}
713
Mikulas Patocka7e201b32009-12-10 23:52:08 +0000714/*
715 * Return a minimum chunk size of all snapshots that have the specified origin.
716 * Return zero if the origin has no snapshots.
717 */
Mike Snitzer542f9032012-07-27 15:08:00 +0100718static uint32_t __minimum_chunk_size(struct origin *o)
Mikulas Patocka7e201b32009-12-10 23:52:08 +0000719{
720 struct dm_snapshot *snap;
721 unsigned chunk_size = 0;
722
723 if (o)
724 list_for_each_entry(snap, &o->snapshots, list)
725 chunk_size = min_not_zero(chunk_size,
726 snap->store->chunk_size);
727
Mike Snitzer542f9032012-07-27 15:08:00 +0100728 return (uint32_t) chunk_size;
Mikulas Patocka7e201b32009-12-10 23:52:08 +0000729}
730
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731/*
732 * Hard coded magic.
733 */
734static int calc_max_buckets(void)
735{
736 /* use a fixed size of 2MB */
737 unsigned long mem = 2 * 1024 * 1024;
738 mem /= sizeof(struct list_head);
739
740 return mem;
741}
742
743/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 * Allocate room for a suitable hash table.
745 */
Jonathan Brassowfee19982009-04-02 19:55:34 +0100746static int init_hash_tables(struct dm_snapshot *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747{
Mikulas Patocka60e356f2013-09-18 19:40:42 -0400748 sector_t hash_size, cow_dev_size, max_buckets;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
750 /*
751 * Calculate based on the size of the original volume or
752 * the COW volume...
753 */
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000754 cow_dev_size = get_dev_size(s->cow->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 max_buckets = calc_max_buckets();
756
Mikulas Patocka60e356f2013-09-18 19:40:42 -0400757 hash_size = cow_dev_size >> s->store->chunk_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 hash_size = min(hash_size, max_buckets);
759
Mikulas Patocka8e87b9b2009-12-10 23:51:54 +0000760 if (hash_size < 64)
761 hash_size = 64;
Robert P. J. Day8defd832008-02-08 02:10:06 +0000762 hash_size = rounddown_pow_of_two(hash_size);
Jon Brassow3510cb92009-12-10 23:52:11 +0000763 if (dm_exception_table_init(&s->complete, hash_size,
764 DM_CHUNK_CONSECUTIVE_BITS))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 return -ENOMEM;
766
767 /*
768 * Allocate hash table for in-flight exceptions
769 * Make this smaller than the real hash table
770 */
771 hash_size >>= 3;
772 if (hash_size < 64)
773 hash_size = 64;
774
Jon Brassow3510cb92009-12-10 23:52:11 +0000775 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
776 dm_exception_table_exit(&s->complete, exception_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 return -ENOMEM;
778 }
779
780 return 0;
781}
782
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000783static void merge_shutdown(struct dm_snapshot *s)
784{
785 clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100786 smp_mb__after_atomic();
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000787 wake_up_bit(&s->state_bits, RUNNING_MERGE);
788}
789
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000790static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
791{
792 s->first_merging_chunk = 0;
793 s->num_merging_chunks = 0;
794
795 return bio_list_get(&s->bios_queued_during_merge);
796}
797
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000798/*
799 * Remove one chunk from the index of completed exceptions.
800 */
801static int __remove_single_exception_chunk(struct dm_snapshot *s,
802 chunk_t old_chunk)
803{
804 struct dm_exception *e;
805
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000806 e = dm_lookup_exception(&s->complete, old_chunk);
807 if (!e) {
808 DMERR("Corruption detected: exception for block %llu is "
809 "on disk but not in memory",
810 (unsigned long long)old_chunk);
811 return -EINVAL;
812 }
813
814 /*
815 * If this is the only chunk using this exception, remove exception.
816 */
817 if (!dm_consecutive_chunk_count(e)) {
818 dm_remove_exception(e);
819 free_completed_exception(e);
820 return 0;
821 }
822
823 /*
824 * The chunk may be either at the beginning or the end of a
825 * group of consecutive chunks - never in the middle. We are
826 * removing chunks in the opposite order to that in which they
827 * were added, so this should always be true.
828 * Decrement the consecutive chunk counter and adjust the
829 * starting point if necessary.
830 */
831 if (old_chunk == e->old_chunk) {
832 e->old_chunk++;
833 e->new_chunk++;
834 } else if (old_chunk != e->old_chunk +
835 dm_consecutive_chunk_count(e)) {
836 DMERR("Attempt to merge block %llu from the "
837 "middle of a chunk range [%llu - %llu]",
838 (unsigned long long)old_chunk,
839 (unsigned long long)e->old_chunk,
840 (unsigned long long)
841 e->old_chunk + dm_consecutive_chunk_count(e));
842 return -EINVAL;
843 }
844
845 dm_consecutive_chunk_count_dec(e);
846
847 return 0;
848}
849
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000850static void flush_bios(struct bio *bio);
851
852static int remove_single_exception_chunk(struct dm_snapshot *s)
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000853{
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000854 struct bio *b = NULL;
855 int r;
856 chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000857
858 down_write(&s->lock);
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000859
860 /*
861 * Process chunks (and associated exceptions) in reverse order
862 * so that dm_consecutive_chunk_count_dec() accounting works.
863 */
864 do {
865 r = __remove_single_exception_chunk(s, old_chunk);
866 if (r)
867 goto out;
868 } while (old_chunk-- > s->first_merging_chunk);
869
870 b = __release_queued_bios_after_merge(s);
871
872out:
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000873 up_write(&s->lock);
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000874 if (b)
875 flush_bios(b);
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000876
877 return r;
878}
879
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000880static int origin_write_extent(struct dm_snapshot *merging_snap,
881 sector_t sector, unsigned chunk_size);
882
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000883static void merge_callback(int read_err, unsigned long write_err,
884 void *context);
885
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000886static uint64_t read_pending_exceptions_done_count(void)
887{
888 uint64_t pending_exceptions_done;
889
890 spin_lock(&_pending_exceptions_done_spinlock);
891 pending_exceptions_done = _pending_exceptions_done_count;
892 spin_unlock(&_pending_exceptions_done_spinlock);
893
894 return pending_exceptions_done;
895}
896
897static void increment_pending_exceptions_done_count(void)
898{
899 spin_lock(&_pending_exceptions_done_spinlock);
900 _pending_exceptions_done_count++;
901 spin_unlock(&_pending_exceptions_done_spinlock);
902
903 wake_up_all(&_pending_exceptions_done);
904}
905
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000906static void snapshot_merge_next_chunks(struct dm_snapshot *s)
907{
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000908 int i, linear_chunks;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000909 chunk_t old_chunk, new_chunk;
910 struct dm_io_region src, dest;
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000911 sector_t io_size;
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000912 uint64_t previous_count;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000913
914 BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
915 if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
916 goto shut;
917
918 /*
919 * valid flag never changes during merge, so no lock required.
920 */
921 if (!s->valid) {
922 DMERR("Snapshot is invalid: can't merge");
923 goto shut;
924 }
925
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000926 linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
927 &new_chunk);
928 if (linear_chunks <= 0) {
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +0000929 if (linear_chunks < 0) {
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000930 DMERR("Read error in exception store: "
931 "shutting down merge");
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +0000932 down_write(&s->lock);
933 s->merge_failed = 1;
934 up_write(&s->lock);
935 }
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000936 goto shut;
937 }
938
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000939 /* Adjust old_chunk and new_chunk to reflect start of linear region */
940 old_chunk = old_chunk + 1 - linear_chunks;
941 new_chunk = new_chunk + 1 - linear_chunks;
942
943 /*
944 * Use one (potentially large) I/O to copy all 'linear_chunks'
945 * from the exception store to the origin
946 */
947 io_size = linear_chunks * s->store->chunk_size;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000948
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000949 dest.bdev = s->origin->bdev;
950 dest.sector = chunk_to_sector(s->store, old_chunk);
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000951 dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000952
953 src.bdev = s->cow->bdev;
954 src.sector = chunk_to_sector(s->store, new_chunk);
955 src.count = dest.count;
956
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000957 /*
958 * Reallocate any exceptions needed in other snapshots then
959 * wait for the pending exceptions to complete.
960 * Each time any pending exception (globally on the system)
961 * completes we are woken and repeat the process to find out
962 * if we can proceed. While this may not seem a particularly
963 * efficient algorithm, it is not expected to have any
964 * significant impact on performance.
965 */
966 previous_count = read_pending_exceptions_done_count();
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000967 while (origin_write_extent(s, dest.sector, io_size)) {
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000968 wait_event(_pending_exceptions_done,
969 (read_pending_exceptions_done_count() !=
970 previous_count));
971 /* Retry after the wait, until all exceptions are done. */
972 previous_count = read_pending_exceptions_done_count();
973 }
974
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000975 down_write(&s->lock);
976 s->first_merging_chunk = old_chunk;
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000977 s->num_merging_chunks = linear_chunks;
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000978 up_write(&s->lock);
979
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000980 /* Wait until writes to all 'linear_chunks' drain */
981 for (i = 0; i < linear_chunks; i++)
982 __check_for_conflicting_io(s, old_chunk + i);
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000983
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000984 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
985 return;
986
987shut:
988 merge_shutdown(s);
989}
990
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000991static void error_bios(struct bio *bio);
992
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000993static void merge_callback(int read_err, unsigned long write_err, void *context)
994{
995 struct dm_snapshot *s = context;
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000996 struct bio *b = NULL;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000997
998 if (read_err || write_err) {
999 if (read_err)
1000 DMERR("Read error: shutting down merge.");
1001 else
1002 DMERR("Write error: shutting down merge.");
1003 goto shut;
1004 }
1005
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001006 if (s->store->type->commit_merge(s->store,
1007 s->num_merging_chunks) < 0) {
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001008 DMERR("Write error in exception store: shutting down merge");
1009 goto shut;
1010 }
1011
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001012 if (remove_single_exception_chunk(s) < 0)
1013 goto shut;
1014
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001015 snapshot_merge_next_chunks(s);
1016
1017 return;
1018
1019shut:
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001020 down_write(&s->lock);
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +00001021 s->merge_failed = 1;
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001022 b = __release_queued_bios_after_merge(s);
1023 up_write(&s->lock);
1024 error_bios(b);
1025
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001026 merge_shutdown(s);
1027}
1028
1029static void start_merge(struct dm_snapshot *s)
1030{
1031 if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
1032 snapshot_merge_next_chunks(s);
1033}
1034
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001035/*
1036 * Stop the merging process and wait until it finishes.
1037 */
1038static void stop_merge(struct dm_snapshot *s)
1039{
1040 set_bit(SHUTDOWN_MERGE, &s->state_bits);
NeilBrown74316202014-07-07 15:16:04 +10001041 wait_on_bit(&s->state_bits, RUNNING_MERGE, TASK_UNINTERRUPTIBLE);
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001042 clear_bit(SHUTDOWN_MERGE, &s->state_bits);
1043}
1044
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
1047 */
1048static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1049{
1050 struct dm_snapshot *s;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001051 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 int r = -EINVAL;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001053 char *origin_path, *cow_path;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00001054 unsigned args_used, num_flush_bios = 1;
Mike Snitzer10b81062009-12-10 23:52:31 +00001055 fmode_t origin_mode = FMODE_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -07001057 if (argc != 4) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001058 ti->error = "requires exactly 4 arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 r = -EINVAL;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001060 goto bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 }
1062
Mike Snitzer10b81062009-12-10 23:52:31 +00001063 if (dm_target_is_snapshot_merge(ti)) {
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00001064 num_flush_bios = 2;
Mike Snitzer10b81062009-12-10 23:52:31 +00001065 origin_mode = FMODE_WRITE;
1066 }
1067
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 s = kmalloc(sizeof(*s), GFP_KERNEL);
Jonathan Brassowfee19982009-04-02 19:55:34 +01001069 if (!s) {
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001070 ti->error = "Cannot allocate private snapshot structure";
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 r = -ENOMEM;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001072 goto bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 }
1074
Mikulas Patockac2411042010-08-12 04:13:51 +01001075 origin_path = argv[0];
1076 argv++;
1077 argc--;
1078
1079 r = dm_get_device(ti, origin_path, origin_mode, &s->origin);
1080 if (r) {
1081 ti->error = "Cannot get origin device";
1082 goto bad_origin;
1083 }
1084
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001085 cow_path = argv[0];
1086 argv++;
1087 argc--;
1088
Milan Broz024d37e2011-03-24 13:52:14 +00001089 r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow);
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001090 if (r) {
1091 ti->error = "Cannot get COW device";
1092 goto bad_cow;
1093 }
1094
1095 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
1096 if (r) {
1097 ti->error = "Couldn't create exception store";
1098 r = -EINVAL;
1099 goto bad_store;
1100 }
1101
1102 argv += args_used;
1103 argc -= args_used;
1104
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001105 s->ti = ti;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 s->valid = 1;
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001107 s->active = 0;
Mikulas Patocka879129d22008-10-30 13:33:16 +00001108 atomic_set(&s->pending_exceptions_count, 0);
Mikulas Patocka230c83a2013-11-29 18:13:37 -05001109 s->exception_start_sequence = 0;
1110 s->exception_complete_sequence = 0;
1111 INIT_LIST_HEAD(&s->out_of_order_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 init_rwsem(&s->lock);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001113 INIT_LIST_HEAD(&s->list);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07001114 spin_lock_init(&s->pe_lock);
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001115 s->state_bits = 0;
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +00001116 s->merge_failed = 0;
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001117 s->first_merging_chunk = 0;
1118 s->num_merging_chunks = 0;
1119 bio_list_init(&s->bios_queued_during_merge);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
1121 /* Allocate hash table for COW data */
Jonathan Brassowfee19982009-04-02 19:55:34 +01001122 if (init_hash_tables(s)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 ti->error = "Unable to allocate hash table space";
1124 r = -ENOMEM;
Jonathan Brassowfee19982009-04-02 19:55:34 +01001125 goto bad_hash_tables;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 }
1127
Mikulas Patockadf5d2e92013-03-01 22:45:49 +00001128 s->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle);
Mikulas Patockafa34ce72011-05-29 13:03:13 +01001129 if (IS_ERR(s->kcopyd_client)) {
1130 r = PTR_ERR(s->kcopyd_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 ti->error = "Could not create kcopyd client";
Jonathan Brassowfee19982009-04-02 19:55:34 +01001132 goto bad_kcopyd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 }
1134
Mikulas Patocka92e86812008-07-21 12:00:35 +01001135 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
1136 if (!s->pending_pool) {
1137 ti->error = "Could not allocate mempool for pending exceptions";
Wei Yongjun09e8b812013-05-10 14:37:15 +01001138 r = -ENOMEM;
Jonathan Brassowfee19982009-04-02 19:55:34 +01001139 goto bad_pending_pool;
Mikulas Patocka92e86812008-07-21 12:00:35 +01001140 }
1141
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001142 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1143 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
1144
1145 spin_lock_init(&s->tracked_chunk_lock);
1146
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001147 ti->private = s;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00001148 ti->num_flush_bios = num_flush_bios;
Mikulas Patocka42bc9542012-12-21 20:23:38 +00001149 ti->per_bio_data_size = sizeof(struct dm_snap_tracked_chunk);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001150
1151 /* Add snapshot to the list of snapshots for this origin */
1152 /* Exceptions aren't triggered till snapshot_resume() is called */
1153 r = register_snapshot(s);
1154 if (r == -ENOMEM) {
1155 ti->error = "Snapshot origin struct allocation failed";
1156 goto bad_load_and_register;
1157 } else if (r < 0) {
1158 /* invalid handover, register_snapshot has set ti->error */
1159 goto bad_load_and_register;
1160 }
1161
1162 /*
1163 * Metadata must only be loaded into one table at once, so skip this
1164 * if metadata will be handed over during resume.
1165 * Chunk size will be set during the handover - set it to zero to
1166 * ensure it's ignored.
1167 */
1168 if (r > 0) {
1169 s->store->chunk_size = 0;
1170 return 0;
1171 }
1172
Jonathan Brassow493df712009-04-02 19:55:31 +01001173 r = s->store->type->read_metadata(s->store, dm_add_exception,
1174 (void *)s);
Milan Broz07641472007-07-12 17:28:13 +01001175 if (r < 0) {
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -07001176 ti->error = "Failed to read snapshot metadata";
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001177 goto bad_read_metadata;
Milan Broz07641472007-07-12 17:28:13 +01001178 } else if (r > 0) {
1179 s->valid = 0;
1180 DMWARN("Snapshot is marked invalid.");
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -07001181 }
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001182
Mikulas Patocka3f2412d2009-10-16 23:18:16 +01001183 if (!s->store->chunk_size) {
1184 ti->error = "Chunk size not set";
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001185 goto bad_read_metadata;
Mikulas Patocka3f2412d2009-10-16 23:18:16 +01001186 }
Mike Snitzer542f9032012-07-27 15:08:00 +01001187
1188 r = dm_set_target_max_io_len(ti, s->store->chunk_size);
1189 if (r)
1190 goto bad_read_metadata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
1192 return 0;
1193
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001194bad_read_metadata:
1195 unregister_snapshot(s);
1196
Jonathan Brassowfee19982009-04-02 19:55:34 +01001197bad_load_and_register:
Mikulas Patocka92e86812008-07-21 12:00:35 +01001198 mempool_destroy(s->pending_pool);
1199
Jonathan Brassowfee19982009-04-02 19:55:34 +01001200bad_pending_pool:
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +01001201 dm_kcopyd_client_destroy(s->kcopyd_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202
Jonathan Brassowfee19982009-04-02 19:55:34 +01001203bad_kcopyd:
Jon Brassow3510cb92009-12-10 23:52:11 +00001204 dm_exception_table_exit(&s->pending, pending_cache);
1205 dm_exception_table_exit(&s->complete, exception_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
Jonathan Brassowfee19982009-04-02 19:55:34 +01001207bad_hash_tables:
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001208 dm_exception_store_destroy(s->store);
1209
1210bad_store:
1211 dm_put_device(ti, s->cow);
1212
1213bad_cow:
Mikulas Patockac2411042010-08-12 04:13:51 +01001214 dm_put_device(ti, s->origin);
1215
1216bad_origin:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 kfree(s);
1218
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001219bad:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 return r;
1221}
1222
Milan Broz31c93a02006-12-08 02:41:11 -08001223static void __free_exceptions(struct dm_snapshot *s)
1224{
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +01001225 dm_kcopyd_client_destroy(s->kcopyd_client);
Milan Broz31c93a02006-12-08 02:41:11 -08001226 s->kcopyd_client = NULL;
1227
Jon Brassow3510cb92009-12-10 23:52:11 +00001228 dm_exception_table_exit(&s->pending, pending_cache);
1229 dm_exception_table_exit(&s->complete, exception_cache);
Milan Broz31c93a02006-12-08 02:41:11 -08001230}
1231
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001232static void __handover_exceptions(struct dm_snapshot *snap_src,
1233 struct dm_snapshot *snap_dest)
1234{
1235 union {
1236 struct dm_exception_table table_swap;
1237 struct dm_exception_store *store_swap;
1238 } u;
1239
1240 /*
1241 * Swap all snapshot context information between the two instances.
1242 */
1243 u.table_swap = snap_dest->complete;
1244 snap_dest->complete = snap_src->complete;
1245 snap_src->complete = u.table_swap;
1246
1247 u.store_swap = snap_dest->store;
1248 snap_dest->store = snap_src->store;
1249 snap_src->store = u.store_swap;
1250
1251 snap_dest->store->snap = snap_dest;
1252 snap_src->store->snap = snap_src;
1253
Mike Snitzer542f9032012-07-27 15:08:00 +01001254 snap_dest->ti->max_io_len = snap_dest->store->chunk_size;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001255 snap_dest->valid = snap_src->valid;
1256
1257 /*
1258 * Set source invalid to ensure it receives no further I/O.
1259 */
1260 snap_src->valid = 0;
1261}
1262
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263static void snapshot_dtr(struct dm_target *ti)
1264{
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001265#ifdef CONFIG_DM_DEBUG
1266 int i;
1267#endif
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001268 struct dm_snapshot *s = ti->private;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001269 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001271 down_read(&_origins_lock);
1272 /* Check whether exception handover must be cancelled */
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +00001273 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001274 if (snap_src && snap_dest && (s == snap_src)) {
1275 down_write(&snap_dest->lock);
1276 snap_dest->valid = 0;
1277 up_write(&snap_dest->lock);
1278 DMERR("Cancelling snapshot handover.");
1279 }
1280 up_read(&_origins_lock);
1281
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001282 if (dm_target_is_snapshot_merge(ti))
1283 stop_merge(s);
1284
Alasdair G Kergon138728dc2006-03-27 01:17:50 -08001285 /* Prevent further origin writes from using this snapshot. */
1286 /* After this returns there can be no new kcopyd jobs. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 unregister_snapshot(s);
1288
Mikulas Patocka879129d22008-10-30 13:33:16 +00001289 while (atomic_read(&s->pending_exceptions_count))
Mikulas Patocka90fa1522009-01-06 03:04:54 +00001290 msleep(1);
Mikulas Patocka879129d22008-10-30 13:33:16 +00001291 /*
1292 * Ensure instructions in mempool_destroy aren't reordered
1293 * before atomic_read.
1294 */
1295 smp_mb();
1296
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001297#ifdef CONFIG_DM_DEBUG
1298 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1299 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
1300#endif
1301
Milan Broz31c93a02006-12-08 02:41:11 -08001302 __free_exceptions(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
Mikulas Patocka92e86812008-07-21 12:00:35 +01001304 mempool_destroy(s->pending_pool);
1305
Jonathan Brassowfee19982009-04-02 19:55:34 +01001306 dm_exception_store_destroy(s->store);
Alasdair G Kergon138728dc2006-03-27 01:17:50 -08001307
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001308 dm_put_device(ti, s->cow);
1309
Mikulas Patockac2411042010-08-12 04:13:51 +01001310 dm_put_device(ti, s->origin);
1311
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 kfree(s);
1313}
1314
1315/*
1316 * Flush a list of buffers.
1317 */
1318static void flush_bios(struct bio *bio)
1319{
1320 struct bio *n;
1321
1322 while (bio) {
1323 n = bio->bi_next;
1324 bio->bi_next = NULL;
1325 generic_make_request(bio);
1326 bio = n;
1327 }
1328}
1329
Mikulas Patocka515ad662009-12-10 23:52:30 +00001330static int do_origin(struct dm_dev *origin, struct bio *bio);
1331
1332/*
1333 * Flush a list of buffers.
1334 */
1335static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1336{
1337 struct bio *n;
1338 int r;
1339
1340 while (bio) {
1341 n = bio->bi_next;
1342 bio->bi_next = NULL;
1343 r = do_origin(s->origin, bio);
1344 if (r == DM_MAPIO_REMAPPED)
1345 generic_make_request(bio);
1346 bio = n;
1347 }
1348}
1349
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350/*
1351 * Error a list of buffers.
1352 */
1353static void error_bios(struct bio *bio)
1354{
1355 struct bio *n;
1356
1357 while (bio) {
1358 n = bio->bi_next;
1359 bio->bi_next = NULL;
NeilBrown6712ecf2007-09-27 12:47:43 +02001360 bio_io_error(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 bio = n;
1362 }
1363}
1364
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001365static void __invalidate_snapshot(struct dm_snapshot *s, int err)
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001366{
1367 if (!s->valid)
1368 return;
1369
1370 if (err == -EIO)
1371 DMERR("Invalidating snapshot: Error reading/writing.");
1372 else if (err == -ENOMEM)
1373 DMERR("Invalidating snapshot: Unable to allocate exception.");
1374
Jonathan Brassow493df712009-04-02 19:55:31 +01001375 if (s->store->type->drop_snapshot)
1376 s->store->type->drop_snapshot(s->store);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001377
1378 s->valid = 0;
1379
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001380 dm_table_event(s->ti->table);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001381}
1382
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001383static void pending_complete(struct dm_snap_pending_exception *pe, int success)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384{
Jon Brassow1d4989c2009-12-10 23:52:10 +00001385 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 struct dm_snapshot *s = pe->snap;
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001387 struct bio *origin_bios = NULL;
1388 struct bio *snapshot_bios = NULL;
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001389 struct bio *full_bio = NULL;
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001390 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001392 if (!success) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 /* Read/write error - snapshot is unusable */
1394 down_write(&s->lock);
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001395 __invalidate_snapshot(s, -EIO);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001396 error = 1;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001397 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 }
1399
Mikulas Patocka119bc542014-01-13 19:13:36 -05001400 e = alloc_completed_exception(GFP_NOIO);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001401 if (!e) {
1402 down_write(&s->lock);
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001403 __invalidate_snapshot(s, -ENOMEM);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001404 error = 1;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001405 goto out;
1406 }
1407 *e = pe->e;
1408
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001409 down_write(&s->lock);
1410 if (!s->valid) {
Jon Brassow3510cb92009-12-10 23:52:11 +00001411 free_completed_exception(e);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001412 error = 1;
1413 goto out;
1414 }
1415
Mike Snitzer615d1eb2009-12-10 23:52:29 +00001416 /* Check for conflicting reads */
1417 __check_for_conflicting_io(s, pe->e.old_chunk);
Mikulas Patockaa8d41b52008-07-21 12:00:34 +01001418
1419 /*
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001420 * Add a proper exception, and remove the
1421 * in-flight exception from the list.
1422 */
Jon Brassow3510cb92009-12-10 23:52:11 +00001423 dm_insert_exception(&s->complete, e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001424
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001425out:
Jon Brassow3510cb92009-12-10 23:52:11 +00001426 dm_remove_exception(&pe->e);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001427 snapshot_bios = bio_list_get(&pe->snapshot_bios);
Mikulas Patocka515ad662009-12-10 23:52:30 +00001428 origin_bios = bio_list_get(&pe->origin_bios);
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001429 full_bio = pe->full_bio;
1430 if (full_bio) {
1431 full_bio->bi_end_io = pe->full_bio_end_io;
1432 full_bio->bi_private = pe->full_bio_private;
Kent Overstreet196d38b2013-11-23 18:34:15 -08001433 atomic_inc(&full_bio->bi_remaining);
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001434 }
Mikulas Patocka515ad662009-12-10 23:52:30 +00001435 free_pending_exception(pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436
Mikulas Patocka73dfd072009-12-10 23:52:34 +00001437 increment_pending_exceptions_done_count();
1438
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001439 up_write(&s->lock);
1440
1441 /* Submit any pending write bios */
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001442 if (error) {
1443 if (full_bio)
1444 bio_io_error(full_bio);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001445 error_bios(snapshot_bios);
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001446 } else {
1447 if (full_bio)
1448 bio_endio(full_bio, 0);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001449 flush_bios(snapshot_bios);
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001450 }
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001451
Mikulas Patocka515ad662009-12-10 23:52:30 +00001452 retry_origin_bios(s, origin_bios);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453}
1454
1455static void commit_callback(void *context, int success)
1456{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001457 struct dm_snap_pending_exception *pe = context;
1458
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 pending_complete(pe, success);
1460}
1461
Mikulas Patocka230c83a2013-11-29 18:13:37 -05001462static void complete_exception(struct dm_snap_pending_exception *pe)
1463{
1464 struct dm_snapshot *s = pe->snap;
1465
1466 if (unlikely(pe->copy_error))
1467 pending_complete(pe, 0);
1468
1469 else
1470 /* Update the metadata if we are persistent */
1471 s->store->type->commit_exception(s->store, &pe->e,
1472 commit_callback, pe);
1473}
1474
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475/*
1476 * Called when the copy I/O has finished. kcopyd actually runs
1477 * this code so don't block.
1478 */
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -07001479static void copy_callback(int read_err, unsigned long write_err, void *context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001481 struct dm_snap_pending_exception *pe = context;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 struct dm_snapshot *s = pe->snap;
1483
Mikulas Patocka230c83a2013-11-29 18:13:37 -05001484 pe->copy_error = read_err || write_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485
Mikulas Patocka230c83a2013-11-29 18:13:37 -05001486 if (pe->exception_sequence == s->exception_complete_sequence) {
1487 s->exception_complete_sequence++;
1488 complete_exception(pe);
1489
1490 while (!list_empty(&s->out_of_order_list)) {
1491 pe = list_entry(s->out_of_order_list.next,
1492 struct dm_snap_pending_exception, out_of_order_entry);
1493 if (pe->exception_sequence != s->exception_complete_sequence)
1494 break;
1495 s->exception_complete_sequence++;
1496 list_del(&pe->out_of_order_entry);
1497 complete_exception(pe);
1498 }
1499 } else {
1500 struct list_head *lh;
1501 struct dm_snap_pending_exception *pe2;
1502
1503 list_for_each_prev(lh, &s->out_of_order_list) {
1504 pe2 = list_entry(lh, struct dm_snap_pending_exception, out_of_order_entry);
1505 if (pe2->exception_sequence < pe->exception_sequence)
1506 break;
1507 }
1508 list_add(&pe->out_of_order_entry, lh);
1509 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510}
1511
1512/*
1513 * Dispatches the copy operation to kcopyd.
1514 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001515static void start_copy(struct dm_snap_pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516{
1517 struct dm_snapshot *s = pe->snap;
Heinz Mauelshagen22a1ceb2008-04-24 21:43:17 +01001518 struct dm_io_region src, dest;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 struct block_device *bdev = s->origin->bdev;
1520 sector_t dev_size;
1521
1522 dev_size = get_dev_size(bdev);
1523
1524 src.bdev = bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001525 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
Mikulas Patockadf96eee2009-10-16 23:18:17 +01001526 src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001528 dest.bdev = s->cow->bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001529 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 dest.count = src.count;
1531
1532 /* Hand over to kcopyd */
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001533 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534}
1535
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001536static void full_bio_end_io(struct bio *bio, int error)
1537{
1538 void *callback_data = bio->bi_private;
1539
1540 dm_kcopyd_do_callback(callback_data, 0, error ? 1 : 0);
1541}
1542
1543static void start_full_bio(struct dm_snap_pending_exception *pe,
1544 struct bio *bio)
1545{
1546 struct dm_snapshot *s = pe->snap;
1547 void *callback_data;
1548
1549 pe->full_bio = bio;
1550 pe->full_bio_end_io = bio->bi_end_io;
1551 pe->full_bio_private = bio->bi_private;
1552
1553 callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client,
1554 copy_callback, pe);
1555
1556 bio->bi_end_io = full_bio_end_io;
1557 bio->bi_private = callback_data;
1558
1559 generic_make_request(bio);
1560}
1561
Mikulas Patocka29138082009-04-02 19:55:25 +01001562static struct dm_snap_pending_exception *
1563__lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1564{
Jon Brassow3510cb92009-12-10 23:52:11 +00001565 struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01001566
1567 if (!e)
1568 return NULL;
1569
1570 return container_of(e, struct dm_snap_pending_exception, e);
1571}
1572
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573/*
1574 * Looks to see if this snapshot already has a pending exception
1575 * for this chunk, otherwise it allocates a new one and inserts
1576 * it into the pending table.
1577 *
1578 * NOTE: a write lock must be held on snap->lock before calling
1579 * this.
1580 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001581static struct dm_snap_pending_exception *
Mikulas Patockac6621392009-04-02 19:55:25 +01001582__find_pending_exception(struct dm_snapshot *s,
1583 struct dm_snap_pending_exception *pe, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584{
Mikulas Patockac6621392009-04-02 19:55:25 +01001585 struct dm_snap_pending_exception *pe2;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001586
Mikulas Patocka29138082009-04-02 19:55:25 +01001587 pe2 = __lookup_pending_exception(s, chunk);
1588 if (pe2) {
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001589 free_pending_exception(pe);
Mikulas Patocka29138082009-04-02 19:55:25 +01001590 return pe2;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001591 }
1592
1593 pe->e.old_chunk = chunk;
1594 bio_list_init(&pe->origin_bios);
1595 bio_list_init(&pe->snapshot_bios);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001596 pe->started = 0;
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001597 pe->full_bio = NULL;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001598
Jonathan Brassow493df712009-04-02 19:55:31 +01001599 if (s->store->type->prepare_exception(s->store, &pe->e)) {
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001600 free_pending_exception(pe);
1601 return NULL;
1602 }
1603
Mikulas Patocka230c83a2013-11-29 18:13:37 -05001604 pe->exception_sequence = s->exception_start_sequence++;
1605
Jon Brassow3510cb92009-12-10 23:52:11 +00001606 dm_insert_exception(&s->pending, &pe->e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001607
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 return pe;
1609}
1610
Jon Brassow1d4989c2009-12-10 23:52:10 +00001611static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
Milan Brozd74f81f2008-02-08 02:11:27 +00001612 struct bio *bio, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613{
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001614 bio->bi_bdev = s->cow->bdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001615 bio->bi_iter.bi_sector =
1616 chunk_to_sector(s->store, dm_chunk_number(e->new_chunk) +
1617 (chunk - e->old_chunk)) +
1618 (bio->bi_iter.bi_sector & s->store->chunk_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619}
1620
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00001621static int snapshot_map(struct dm_target *ti, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622{
Jon Brassow1d4989c2009-12-10 23:52:10 +00001623 struct dm_exception *e;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001624 struct dm_snapshot *s = ti->private;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001625 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 chunk_t chunk;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001627 struct dm_snap_pending_exception *pe = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628
Mikulas Patockaee180262012-12-21 20:23:41 +00001629 init_tracked_chunk(bio);
1630
Tejun Heod87f4c12010-09-03 11:56:19 +02001631 if (bio->bi_rw & REQ_FLUSH) {
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001632 bio->bi_bdev = s->cow->bdev;
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01001633 return DM_MAPIO_REMAPPED;
1634 }
1635
Kent Overstreet4f024f32013-10-11 15:44:27 -07001636 chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637
1638 /* Full snapshots are not usable */
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001639 /* To get here the table must be live so s->active is always set. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 if (!s->valid)
Alasdair G Kergonf6a80ea2005-07-12 15:53:01 -07001641 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001643 /* FIXME: should only take write lock if we need
1644 * to copy an exception */
1645 down_write(&s->lock);
1646
1647 if (!s->valid) {
1648 r = -EIO;
1649 goto out_unlock;
1650 }
1651
1652 /* If the block is already remapped - use that, else remap it */
Jon Brassow3510cb92009-12-10 23:52:11 +00001653 e = dm_lookup_exception(&s->complete, chunk);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001654 if (e) {
Milan Brozd74f81f2008-02-08 02:11:27 +00001655 remap_exception(s, e, bio, chunk);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001656 goto out_unlock;
1657 }
1658
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 /*
1660 * Write to snapshot - higher level takes care of RW/RO
1661 * flags so we should only get this if we are
1662 * writeable.
1663 */
1664 if (bio_rw(bio) == WRITE) {
Mikulas Patocka29138082009-04-02 19:55:25 +01001665 pe = __lookup_pending_exception(s, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001666 if (!pe) {
Mikulas Patockac6621392009-04-02 19:55:25 +01001667 up_write(&s->lock);
1668 pe = alloc_pending_exception(s);
1669 down_write(&s->lock);
1670
1671 if (!s->valid) {
1672 free_pending_exception(pe);
1673 r = -EIO;
1674 goto out_unlock;
1675 }
1676
Jon Brassow3510cb92009-12-10 23:52:11 +00001677 e = dm_lookup_exception(&s->complete, chunk);
Mikulas Patocka35bf6592009-04-02 19:55:26 +01001678 if (e) {
1679 free_pending_exception(pe);
1680 remap_exception(s, e, bio, chunk);
1681 goto out_unlock;
1682 }
1683
Mikulas Patockac6621392009-04-02 19:55:25 +01001684 pe = __find_pending_exception(s, pe, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01001685 if (!pe) {
1686 __invalidate_snapshot(s, -ENOMEM);
1687 r = -EIO;
1688 goto out_unlock;
1689 }
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001690 }
1691
Milan Brozd74f81f2008-02-08 02:11:27 +00001692 remap_exception(s, &pe->e, bio, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001693
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001694 r = DM_MAPIO_SUBMITTED;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001695
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001696 if (!pe->started &&
Kent Overstreet4f024f32013-10-11 15:44:27 -07001697 bio->bi_iter.bi_size ==
1698 (s->store->chunk_size << SECTOR_SHIFT)) {
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001699 pe->started = 1;
1700 up_write(&s->lock);
1701 start_full_bio(pe, bio);
1702 goto out;
1703 }
1704
1705 bio_list_add(&pe->snapshot_bios, bio);
1706
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001707 if (!pe->started) {
1708 /* this is protected by snap->lock */
1709 pe->started = 1;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001710 up_write(&s->lock);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001711 start_copy(pe);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001712 goto out;
1713 }
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001714 } else {
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001715 bio->bi_bdev = s->origin->bdev;
Mikulas Patockaee180262012-12-21 20:23:41 +00001716 track_chunk(s, bio, chunk);
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001717 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001719out_unlock:
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001720 up_write(&s->lock);
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001721out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 return r;
1723}
1724
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001725/*
1726 * A snapshot-merge target behaves like a combination of a snapshot
1727 * target and a snapshot-origin target. It only generates new
1728 * exceptions in other snapshots and not in the one that is being
1729 * merged.
1730 *
1731 * For each chunk, if there is an existing exception, it is used to
1732 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
1733 * which in turn might generate exceptions in other snapshots.
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001734 * If merging is currently taking place on the chunk in question, the
1735 * I/O is deferred by adding it to s->bios_queued_during_merge.
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001736 */
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00001737static int snapshot_merge_map(struct dm_target *ti, struct bio *bio)
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001738{
1739 struct dm_exception *e;
1740 struct dm_snapshot *s = ti->private;
1741 int r = DM_MAPIO_REMAPPED;
1742 chunk_t chunk;
1743
Mikulas Patockaee180262012-12-21 20:23:41 +00001744 init_tracked_chunk(bio);
1745
Tejun Heod87f4c12010-09-03 11:56:19 +02001746 if (bio->bi_rw & REQ_FLUSH) {
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00001747 if (!dm_bio_get_target_bio_nr(bio))
Mike Snitzer10b81062009-12-10 23:52:31 +00001748 bio->bi_bdev = s->origin->bdev;
1749 else
1750 bio->bi_bdev = s->cow->bdev;
Mike Snitzer10b81062009-12-10 23:52:31 +00001751 return DM_MAPIO_REMAPPED;
1752 }
1753
Kent Overstreet4f024f32013-10-11 15:44:27 -07001754 chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001755
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001756 down_write(&s->lock);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001757
Mikulas Patockad2fdb772009-12-10 23:52:36 +00001758 /* Full merging snapshots are redirected to the origin */
1759 if (!s->valid)
1760 goto redirect_to_origin;
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001761
1762 /* If the block is already remapped - use that */
1763 e = dm_lookup_exception(&s->complete, chunk);
1764 if (e) {
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001765 /* Queue writes overlapping with chunks being merged */
1766 if (bio_rw(bio) == WRITE &&
1767 chunk >= s->first_merging_chunk &&
1768 chunk < (s->first_merging_chunk +
1769 s->num_merging_chunks)) {
1770 bio->bi_bdev = s->origin->bdev;
1771 bio_list_add(&s->bios_queued_during_merge, bio);
1772 r = DM_MAPIO_SUBMITTED;
1773 goto out_unlock;
1774 }
Mikulas Patocka17aa0332009-12-10 23:52:33 +00001775
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001776 remap_exception(s, e, bio, chunk);
Mikulas Patocka17aa0332009-12-10 23:52:33 +00001777
1778 if (bio_rw(bio) == WRITE)
Mikulas Patockaee180262012-12-21 20:23:41 +00001779 track_chunk(s, bio, chunk);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001780 goto out_unlock;
1781 }
1782
Mikulas Patockad2fdb772009-12-10 23:52:36 +00001783redirect_to_origin:
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001784 bio->bi_bdev = s->origin->bdev;
1785
1786 if (bio_rw(bio) == WRITE) {
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001787 up_write(&s->lock);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001788 return do_origin(s->origin, bio);
1789 }
1790
1791out_unlock:
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001792 up_write(&s->lock);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001793
1794 return r;
1795}
1796
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00001797static int snapshot_end_io(struct dm_target *ti, struct bio *bio, int error)
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001798{
1799 struct dm_snapshot *s = ti->private;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001800
Mikulas Patockaee180262012-12-21 20:23:41 +00001801 if (is_bio_tracked(bio))
1802 stop_tracking_chunk(s, bio);
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001803
1804 return 0;
1805}
1806
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001807static void snapshot_merge_presuspend(struct dm_target *ti)
1808{
1809 struct dm_snapshot *s = ti->private;
1810
1811 stop_merge(s);
1812}
1813
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001814static int snapshot_preresume(struct dm_target *ti)
1815{
1816 int r = 0;
1817 struct dm_snapshot *s = ti->private;
1818 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1819
1820 down_read(&_origins_lock);
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +00001821 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001822 if (snap_src && snap_dest) {
1823 down_read(&snap_src->lock);
1824 if (s == snap_src) {
1825 DMERR("Unable to resume snapshot source until "
1826 "handover completes.");
1827 r = -EINVAL;
Mike Snitzerb83b2f22011-01-13 19:59:59 +00001828 } else if (!dm_suspended(snap_src->ti)) {
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001829 DMERR("Unable to perform snapshot handover until "
1830 "source is suspended.");
1831 r = -EINVAL;
1832 }
1833 up_read(&snap_src->lock);
1834 }
1835 up_read(&_origins_lock);
1836
1837 return r;
1838}
1839
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840static void snapshot_resume(struct dm_target *ti)
1841{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001842 struct dm_snapshot *s = ti->private;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001843 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1844
1845 down_read(&_origins_lock);
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +00001846 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001847 if (snap_src && snap_dest) {
1848 down_write(&snap_src->lock);
1849 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1850 __handover_exceptions(snap_src, snap_dest);
1851 up_write(&snap_dest->lock);
1852 up_write(&snap_src->lock);
1853 }
1854 up_read(&_origins_lock);
1855
1856 /* Now we have correct chunk size, reregister */
1857 reregister_snapshot(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001859 down_write(&s->lock);
1860 s->active = 1;
1861 up_write(&s->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862}
1863
Mike Snitzer542f9032012-07-27 15:08:00 +01001864static uint32_t get_origin_minimum_chunksize(struct block_device *bdev)
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001865{
Mike Snitzer542f9032012-07-27 15:08:00 +01001866 uint32_t min_chunksize;
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001867
1868 down_read(&_origins_lock);
1869 min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
1870 up_read(&_origins_lock);
1871
1872 return min_chunksize;
1873}
1874
1875static void snapshot_merge_resume(struct dm_target *ti)
1876{
1877 struct dm_snapshot *s = ti->private;
1878
1879 /*
1880 * Handover exceptions from existing snapshot.
1881 */
1882 snapshot_resume(ti);
1883
1884 /*
Mike Snitzer542f9032012-07-27 15:08:00 +01001885 * snapshot-merge acts as an origin, so set ti->max_io_len
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001886 */
Mike Snitzer542f9032012-07-27 15:08:00 +01001887 ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev);
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001888
1889 start_merge(s);
1890}
1891
Mikulas Patockafd7c0922013-03-01 22:45:44 +00001892static void snapshot_status(struct dm_target *ti, status_type_t type,
1893 unsigned status_flags, char *result, unsigned maxlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894{
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01001895 unsigned sz = 0;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001896 struct dm_snapshot *snap = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897
1898 switch (type) {
1899 case STATUSTYPE_INFO:
Mikulas Patocka94e765722009-12-10 23:51:53 +00001900
1901 down_write(&snap->lock);
1902
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 if (!snap->valid)
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01001904 DMEMIT("Invalid");
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +00001905 else if (snap->merge_failed)
1906 DMEMIT("Merge failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 else {
Mike Snitzer985903b2009-12-10 23:52:11 +00001908 if (snap->store->type->usage) {
1909 sector_t total_sectors, sectors_allocated,
1910 metadata_sectors;
1911 snap->store->type->usage(snap->store,
1912 &total_sectors,
1913 &sectors_allocated,
1914 &metadata_sectors);
1915 DMEMIT("%llu/%llu %llu",
1916 (unsigned long long)sectors_allocated,
1917 (unsigned long long)total_sectors,
1918 (unsigned long long)metadata_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 }
1920 else
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01001921 DMEMIT("Unknown");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 }
Mikulas Patocka94e765722009-12-10 23:51:53 +00001923
1924 up_write(&snap->lock);
1925
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 break;
1927
1928 case STATUSTYPE_TABLE:
1929 /*
1930 * kdevname returns a static pointer so we need
1931 * to make private copies if the output is to
1932 * make sense.
1933 */
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001934 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
Jonathan Brassow1e302a92009-04-02 19:55:35 +01001935 snap->store->type->status(snap->store, type, result + sz,
1936 maxlen - sz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 break;
1938 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939}
1940
Mike Snitzer8811f462009-09-04 20:40:19 +01001941static int snapshot_iterate_devices(struct dm_target *ti,
1942 iterate_devices_callout_fn fn, void *data)
1943{
1944 struct dm_snapshot *snap = ti->private;
Mikulas Patocka1e5554c2010-08-12 04:13:50 +01001945 int r;
Mike Snitzer8811f462009-09-04 20:40:19 +01001946
Mikulas Patocka1e5554c2010-08-12 04:13:50 +01001947 r = fn(ti, snap->origin, 0, ti->len, data);
1948
1949 if (!r)
1950 r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data);
1951
1952 return r;
Mike Snitzer8811f462009-09-04 20:40:19 +01001953}
1954
1955
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956/*-----------------------------------------------------------------
1957 * Origin methods
1958 *---------------------------------------------------------------*/
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001959
1960/*
1961 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1962 * supplied bio was ignored. The caller may submit it immediately.
1963 * (No remapping actually occurs as the origin is always a direct linear
1964 * map.)
1965 *
1966 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1967 * and any supplied bio is added to a list to be submitted once all
1968 * the necessary exceptions exist.
1969 */
1970static int __origin_write(struct list_head *snapshots, sector_t sector,
1971 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972{
Mikulas Patocka515ad662009-12-10 23:52:30 +00001973 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 struct dm_snapshot *snap;
Jon Brassow1d4989c2009-12-10 23:52:10 +00001975 struct dm_exception *e;
Mikulas Patocka515ad662009-12-10 23:52:30 +00001976 struct dm_snap_pending_exception *pe;
1977 struct dm_snap_pending_exception *pe_to_start_now = NULL;
1978 struct dm_snap_pending_exception *pe_to_start_last = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 chunk_t chunk;
1980
1981 /* Do all the snapshots on this origin */
1982 list_for_each_entry (snap, snapshots, list) {
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001983 /*
1984 * Don't make new exceptions in a merging snapshot
1985 * because it has effectively been deleted
1986 */
1987 if (dm_target_is_snapshot_merge(snap->ti))
1988 continue;
1989
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001990 down_write(&snap->lock);
1991
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001992 /* Only deal with valid and active snapshots */
1993 if (!snap->valid || !snap->active)
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001994 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995
Alasdair G Kergond5e404c2005-07-12 15:53:05 -07001996 /* Nothing to do if writing beyond end of snapshot */
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001997 if (sector >= dm_table_get_size(snap->ti->table))
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001998 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999
2000 /*
2001 * Remember, different snapshots can have
2002 * different chunk sizes.
2003 */
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00002004 chunk = sector_to_chunk(snap->store, sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005
2006 /*
2007 * Check exception table to see if block
2008 * is already remapped in this snapshot
2009 * and trigger an exception if not.
2010 */
Jon Brassow3510cb92009-12-10 23:52:11 +00002011 e = dm_lookup_exception(&snap->complete, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002012 if (e)
2013 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014
Mikulas Patocka29138082009-04-02 19:55:25 +01002015 pe = __lookup_pending_exception(snap, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002016 if (!pe) {
Mikulas Patockac6621392009-04-02 19:55:25 +01002017 up_write(&snap->lock);
2018 pe = alloc_pending_exception(snap);
2019 down_write(&snap->lock);
2020
2021 if (!snap->valid) {
2022 free_pending_exception(pe);
2023 goto next_snapshot;
2024 }
2025
Jon Brassow3510cb92009-12-10 23:52:11 +00002026 e = dm_lookup_exception(&snap->complete, chunk);
Mikulas Patocka35bf6592009-04-02 19:55:26 +01002027 if (e) {
2028 free_pending_exception(pe);
2029 goto next_snapshot;
2030 }
2031
Mikulas Patockac6621392009-04-02 19:55:25 +01002032 pe = __find_pending_exception(snap, pe, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01002033 if (!pe) {
2034 __invalidate_snapshot(snap, -ENOMEM);
2035 goto next_snapshot;
2036 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 }
2038
Mikulas Patocka515ad662009-12-10 23:52:30 +00002039 r = DM_MAPIO_SUBMITTED;
2040
2041 /*
2042 * If an origin bio was supplied, queue it to wait for the
2043 * completion of this exception, and start this one last,
2044 * at the end of the function.
2045 */
2046 if (bio) {
2047 bio_list_add(&pe->origin_bios, bio);
2048 bio = NULL;
2049
2050 if (!pe->started) {
2051 pe->started = 1;
2052 pe_to_start_last = pe;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002053 }
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002054 }
2055
2056 if (!pe->started) {
2057 pe->started = 1;
Mikulas Patocka515ad662009-12-10 23:52:30 +00002058 pe_to_start_now = pe;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002059 }
2060
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01002061next_snapshot:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 up_write(&snap->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063
Mikulas Patocka515ad662009-12-10 23:52:30 +00002064 if (pe_to_start_now) {
2065 start_copy(pe_to_start_now);
2066 pe_to_start_now = NULL;
2067 }
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08002068 }
2069
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 /*
Mikulas Patocka515ad662009-12-10 23:52:30 +00002071 * Submit the exception against which the bio is queued last,
2072 * to give the other exceptions a head start.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 */
Mikulas Patocka515ad662009-12-10 23:52:30 +00002074 if (pe_to_start_last)
2075 start_copy(pe_to_start_last);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076
2077 return r;
2078}
2079
2080/*
2081 * Called on a write from the origin driver.
2082 */
2083static int do_origin(struct dm_dev *origin, struct bio *bio)
2084{
2085 struct origin *o;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08002086 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087
2088 down_read(&_origins_lock);
2089 o = __lookup_origin(origin->bdev);
2090 if (o)
Kent Overstreet4f024f32013-10-11 15:44:27 -07002091 r = __origin_write(&o->snapshots, bio->bi_iter.bi_sector, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 up_read(&_origins_lock);
2093
2094 return r;
2095}
2096
2097/*
Mikulas Patocka73dfd072009-12-10 23:52:34 +00002098 * Trigger exceptions in all non-merging snapshots.
2099 *
2100 * The chunk size of the merging snapshot may be larger than the chunk
2101 * size of some other snapshot so we may need to reallocate multiple
2102 * chunks in other snapshots.
2103 *
2104 * We scan all the overlapping exceptions in the other snapshots.
2105 * Returns 1 if anything was reallocated and must be waited for,
2106 * otherwise returns 0.
2107 *
2108 * size must be a multiple of merging_snap's chunk_size.
2109 */
2110static int origin_write_extent(struct dm_snapshot *merging_snap,
2111 sector_t sector, unsigned size)
2112{
2113 int must_wait = 0;
2114 sector_t n;
2115 struct origin *o;
2116
2117 /*
Mike Snitzer542f9032012-07-27 15:08:00 +01002118 * The origin's __minimum_chunk_size() got stored in max_io_len
Mikulas Patocka73dfd072009-12-10 23:52:34 +00002119 * by snapshot_merge_resume().
2120 */
2121 down_read(&_origins_lock);
2122 o = __lookup_origin(merging_snap->origin->bdev);
Mike Snitzer542f9032012-07-27 15:08:00 +01002123 for (n = 0; n < size; n += merging_snap->ti->max_io_len)
Mikulas Patocka73dfd072009-12-10 23:52:34 +00002124 if (__origin_write(&o->snapshots, sector + n, NULL) ==
2125 DM_MAPIO_SUBMITTED)
2126 must_wait = 1;
2127 up_read(&_origins_lock);
2128
2129 return must_wait;
2130}
2131
2132/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 * Origin: maps a linear range of a device, with hooks for snapshotting.
2134 */
2135
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002136struct dm_origin {
2137 struct dm_dev *dev;
Mikulas Patocka298eaa82014-03-14 18:43:07 -04002138 unsigned split_boundary;
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002139};
2140
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141/*
2142 * Construct an origin mapping: <dev_path>
2143 * The context for an origin is merely a 'struct dm_dev *'
2144 * pointing to the real device.
2145 */
2146static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2147{
2148 int r;
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002149 struct dm_origin *o;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150
2151 if (argc != 1) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07002152 ti->error = "origin: incorrect number of arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 return -EINVAL;
2154 }
2155
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002156 o = kmalloc(sizeof(struct dm_origin), GFP_KERNEL);
2157 if (!o) {
2158 ti->error = "Cannot allocate private origin structure";
2159 r = -ENOMEM;
2160 goto bad_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 }
2162
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002163 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &o->dev);
2164 if (r) {
2165 ti->error = "Cannot get target device";
2166 goto bad_open;
2167 }
2168
2169 ti->private = o;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002170 ti->num_flush_bios = 1;
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01002171
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 return 0;
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002173
2174bad_open:
2175 kfree(o);
2176bad_alloc:
2177 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178}
2179
2180static void origin_dtr(struct dm_target *ti)
2181{
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002182 struct dm_origin *o = ti->private;
2183 dm_put_device(ti, o->dev);
2184 kfree(o);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185}
2186
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00002187static int origin_map(struct dm_target *ti, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188{
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002189 struct dm_origin *o = ti->private;
Mikulas Patocka298eaa82014-03-14 18:43:07 -04002190 unsigned available_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002192 bio->bi_bdev = o->dev->bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193
Mikulas Patocka298eaa82014-03-14 18:43:07 -04002194 if (unlikely(bio->bi_rw & REQ_FLUSH))
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01002195 return DM_MAPIO_REMAPPED;
2196
Mikulas Patocka298eaa82014-03-14 18:43:07 -04002197 if (bio_rw(bio) != WRITE)
2198 return DM_MAPIO_REMAPPED;
2199
2200 available_sectors = o->split_boundary -
2201 ((unsigned)bio->bi_iter.bi_sector & (o->split_boundary - 1));
2202
2203 if (bio_sectors(bio) > available_sectors)
2204 dm_accept_partial_bio(bio, available_sectors);
2205
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 /* Only tell snapshots if this is a write */
Mikulas Patocka298eaa82014-03-14 18:43:07 -04002207 return do_origin(o->dev, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208}
2209
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210/*
Mike Snitzer542f9032012-07-27 15:08:00 +01002211 * Set the target "max_io_len" field to the minimum of all the snapshots'
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 * chunk sizes.
2213 */
2214static void origin_resume(struct dm_target *ti)
2215{
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002216 struct dm_origin *o = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217
Mikulas Patocka298eaa82014-03-14 18:43:07 -04002218 o->split_boundary = get_origin_minimum_chunksize(o->dev->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219}
2220
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002221static void origin_status(struct dm_target *ti, status_type_t type,
2222 unsigned status_flags, char *result, unsigned maxlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223{
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002224 struct dm_origin *o = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225
2226 switch (type) {
2227 case STATUSTYPE_INFO:
2228 result[0] = '\0';
2229 break;
2230
2231 case STATUSTYPE_TABLE:
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002232 snprintf(result, maxlen, "%s", o->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 break;
2234 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235}
2236
Mikulas Patockab1d55522010-08-12 04:14:02 +01002237static int origin_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2238 struct bio_vec *biovec, int max_size)
2239{
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002240 struct dm_origin *o = ti->private;
2241 struct request_queue *q = bdev_get_queue(o->dev->bdev);
Mikulas Patockab1d55522010-08-12 04:14:02 +01002242
2243 if (!q->merge_bvec_fn)
2244 return max_size;
2245
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002246 bvm->bi_bdev = o->dev->bdev;
Mikulas Patockab1d55522010-08-12 04:14:02 +01002247
2248 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2249}
2250
Mike Snitzer8811f462009-09-04 20:40:19 +01002251static int origin_iterate_devices(struct dm_target *ti,
2252 iterate_devices_callout_fn fn, void *data)
2253{
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002254 struct dm_origin *o = ti->private;
Mike Snitzer8811f462009-09-04 20:40:19 +01002255
Mikulas Patocka599cdf32014-03-14 18:42:12 -04002256 return fn(ti, o->dev, 0, ti->len, data);
Mike Snitzer8811f462009-09-04 20:40:19 +01002257}
2258
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259static struct target_type origin_target = {
2260 .name = "snapshot-origin",
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002261 .version = {1, 8, 1},
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 .module = THIS_MODULE,
2263 .ctr = origin_ctr,
2264 .dtr = origin_dtr,
2265 .map = origin_map,
2266 .resume = origin_resume,
2267 .status = origin_status,
Mikulas Patockab1d55522010-08-12 04:14:02 +01002268 .merge = origin_merge,
Mike Snitzer8811f462009-09-04 20:40:19 +01002269 .iterate_devices = origin_iterate_devices,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270};
2271
2272static struct target_type snapshot_target = {
2273 .name = "snapshot",
Mikulas Patocka230c83a2013-11-29 18:13:37 -05002274 .version = {1, 12, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 .module = THIS_MODULE,
2276 .ctr = snapshot_ctr,
2277 .dtr = snapshot_dtr,
2278 .map = snapshot_map,
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002279 .end_io = snapshot_end_io,
Mike Snitzerc1f0c182009-12-10 23:52:24 +00002280 .preresume = snapshot_preresume,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 .resume = snapshot_resume,
2282 .status = snapshot_status,
Mike Snitzer8811f462009-09-04 20:40:19 +01002283 .iterate_devices = snapshot_iterate_devices,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284};
2285
Mikulas Patockad698aa42009-12-10 23:52:30 +00002286static struct target_type merge_target = {
2287 .name = dm_snapshot_merge_target_name,
Mikulas Patocka42bc9542012-12-21 20:23:38 +00002288 .version = {1, 2, 0},
Mikulas Patockad698aa42009-12-10 23:52:30 +00002289 .module = THIS_MODULE,
2290 .ctr = snapshot_ctr,
2291 .dtr = snapshot_dtr,
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00002292 .map = snapshot_merge_map,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002293 .end_io = snapshot_end_io,
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002294 .presuspend = snapshot_merge_presuspend,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002295 .preresume = snapshot_preresume,
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002296 .resume = snapshot_merge_resume,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002297 .status = snapshot_status,
2298 .iterate_devices = snapshot_iterate_devices,
2299};
2300
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301static int __init dm_snapshot_init(void)
2302{
2303 int r;
2304
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00002305 r = dm_exception_store_init();
2306 if (r) {
2307 DMERR("Failed to initialize exception stores");
2308 return r;
2309 }
2310
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 r = dm_register_target(&snapshot_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002312 if (r < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 DMERR("snapshot target register failed %d", r);
Jonathan Brassow034a1862009-10-16 23:18:14 +01002314 goto bad_register_snapshot_target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 }
2316
2317 r = dm_register_target(&origin_target);
2318 if (r < 0) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07002319 DMERR("Origin target register failed %d", r);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002320 goto bad_register_origin_target;
2321 }
2322
2323 r = dm_register_target(&merge_target);
2324 if (r < 0) {
2325 DMERR("Merge target register failed %d", r);
2326 goto bad_register_merge_target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 }
2328
2329 r = init_origin_hash();
2330 if (r) {
2331 DMERR("init_origin_hash failed.");
Mikulas Patockad698aa42009-12-10 23:52:30 +00002332 goto bad_origin_hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 }
2334
Jon Brassow1d4989c2009-12-10 23:52:10 +00002335 exception_cache = KMEM_CACHE(dm_exception, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 if (!exception_cache) {
2337 DMERR("Couldn't create exception cache.");
2338 r = -ENOMEM;
Mikulas Patockad698aa42009-12-10 23:52:30 +00002339 goto bad_exception_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 }
2341
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002342 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 if (!pending_cache) {
2344 DMERR("Couldn't create pending cache.");
2345 r = -ENOMEM;
Mikulas Patockad698aa42009-12-10 23:52:30 +00002346 goto bad_pending_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 }
2348
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 return 0;
2350
Mikulas Patockad698aa42009-12-10 23:52:30 +00002351bad_pending_cache:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 kmem_cache_destroy(exception_cache);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002353bad_exception_cache:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 exit_origin_hash();
Mikulas Patockad698aa42009-12-10 23:52:30 +00002355bad_origin_hash:
2356 dm_unregister_target(&merge_target);
2357bad_register_merge_target:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 dm_unregister_target(&origin_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002359bad_register_origin_target:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 dm_unregister_target(&snapshot_target);
Jonathan Brassow034a1862009-10-16 23:18:14 +01002361bad_register_snapshot_target:
2362 dm_exception_store_exit();
Mikulas Patockad698aa42009-12-10 23:52:30 +00002363
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 return r;
2365}
2366
2367static void __exit dm_snapshot_exit(void)
2368{
Mikulas Patocka10d3bd02009-01-06 03:04:58 +00002369 dm_unregister_target(&snapshot_target);
2370 dm_unregister_target(&origin_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002371 dm_unregister_target(&merge_target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372
2373 exit_origin_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 kmem_cache_destroy(pending_cache);
2375 kmem_cache_destroy(exception_cache);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00002376
2377 dm_exception_store_exit();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378}
2379
2380/* Module hooks */
2381module_init(dm_snapshot_init);
2382module_exit(dm_snapshot_exit);
2383
2384MODULE_DESCRIPTION(DM_NAME " snapshot target");
2385MODULE_AUTHOR("Joe Thornber");
2386MODULE_LICENSE("GPL");
Mikulas Patocka23cb2102013-03-01 22:45:47 +00002387MODULE_ALIAS("dm-snapshot-origin");
2388MODULE_ALIAS("dm-snapshot-merge");