blob: aec57d76db5d616c8e692fa95cee58a8f62a0573 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * dm-snapshot.c
3 *
4 * Copyright (C) 2001-2002 Sistina Software (UK) Limited.
5 *
6 * This file is released under the GPL.
7 */
8
9#include <linux/blkdev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010#include <linux/device-mapper.h>
Mikulas Patocka90fa1522009-01-06 03:04:54 +000011#include <linux/delay.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/fs.h>
13#include <linux/init.h>
14#include <linux/kdev_t.h>
15#include <linux/list.h>
16#include <linux/mempool.h>
17#include <linux/module.h>
18#include <linux/slab.h>
19#include <linux/vmalloc.h>
vignesh babu6f3c3f02007-10-19 22:38:44 +010020#include <linux/log2.h>
Alasdair G Kergona765e202008-04-24 22:02:01 +010021#include <linux/dm-kcopyd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022
Jonathan Brassowaea53d92009-01-06 03:05:15 +000023#include "dm-exception-store.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024
Alasdair G Kergon72d94862006-06-26 00:27:35 -070025#define DM_MSG_PREFIX "snapshots"
26
Mikulas Patockad698aa42009-12-10 23:52:30 +000027static const char dm_snapshot_merge_target_name[] = "snapshot-merge";
28
29#define dm_target_is_snapshot_merge(ti) \
30 ((ti)->type->name == dm_snapshot_merge_target_name)
31
Linus Torvalds1da177e2005-04-16 15:20:36 -070032/*
Mikulas Patockacd45daf2008-07-21 12:00:32 +010033 * The size of the mempool used to track chunks in use.
34 */
35#define MIN_IOS 256
36
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010037#define DM_TRACKED_CHUNK_HASH_SIZE 16
38#define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
39 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
40
Jon Brassow191437a2009-12-10 23:52:10 +000041struct dm_exception_table {
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010042 uint32_t hash_mask;
43 unsigned hash_shift;
44 struct list_head *table;
45};
46
47struct dm_snapshot {
48 struct rw_semaphore lock;
49
50 struct dm_dev *origin;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +000051 struct dm_dev *cow;
52
53 struct dm_target *ti;
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010054
55 /* List of snapshots per Origin */
56 struct list_head list;
57
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +000058 /*
59 * You can't use a snapshot if this is 0 (e.g. if full).
60 * A snapshot-merge target never clears this.
61 */
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010062 int valid;
63
64 /* Origin writes don't trigger exceptions until this is set */
65 int active;
66
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010067 atomic_t pending_exceptions_count;
68
Mike Snitzer924e6002010-03-06 02:32:33 +000069 mempool_t *pending_pool;
70
Jon Brassow191437a2009-12-10 23:52:10 +000071 struct dm_exception_table pending;
72 struct dm_exception_table complete;
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010073
74 /*
75 * pe_lock protects all pending_exception operations and access
76 * as well as the snapshot_bios list.
77 */
78 spinlock_t pe_lock;
79
Mike Snitzer924e6002010-03-06 02:32:33 +000080 /* Chunks with outstanding reads */
81 spinlock_t tracked_chunk_lock;
Mike Snitzer924e6002010-03-06 02:32:33 +000082 struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
83
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010084 /* The on disk metadata handler */
85 struct dm_exception_store *store;
86
87 struct dm_kcopyd_client *kcopyd_client;
88
Mike Snitzer924e6002010-03-06 02:32:33 +000089 /* Wait for events based on state_bits */
90 unsigned long state_bits;
91
92 /* Range of chunks currently being merged. */
93 chunk_t first_merging_chunk;
94 int num_merging_chunks;
Mikulas Patocka1e03f972009-12-10 23:52:32 +000095
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +000096 /*
97 * The merge operation failed if this flag is set.
98 * Failure modes are handled as follows:
99 * - I/O error reading the header
100 * => don't load the target; abort.
101 * - Header does not have "valid" flag set
102 * => use the origin; forget about the snapshot.
103 * - I/O error when reading exceptions
104 * => don't load the target; abort.
105 * (We can't use the intermediate origin state.)
106 * - I/O error while merging
107 * => stop merging; set merge_failed; process I/O normally.
108 */
109 int merge_failed;
110
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000111 /*
112 * Incoming bios that overlap with chunks being merged must wait
113 * for them to be committed.
114 */
115 struct bio_list bios_queued_during_merge;
Jonathan Brassowccc45ea2009-04-02 19:55:34 +0100116};
117
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000118/*
119 * state_bits:
120 * RUNNING_MERGE - Merge operation is in progress.
121 * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
122 * cleared afterwards.
123 */
124#define RUNNING_MERGE 0
125#define SHUTDOWN_MERGE 1
126
Mikulas Patockadf5d2e92013-03-01 22:45:49 +0000127DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
128 "A percentage of time allocated for copy on write");
129
Mikulas Patockac2411042010-08-12 04:13:51 +0100130struct dm_dev *dm_snap_origin(struct dm_snapshot *s)
131{
132 return s->origin;
133}
134EXPORT_SYMBOL(dm_snap_origin);
135
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000136struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
137{
138 return s->cow;
139}
140EXPORT_SYMBOL(dm_snap_cow);
141
Jonathan Brassowccc45ea2009-04-02 19:55:34 +0100142static sector_t chunk_to_sector(struct dm_exception_store *store,
143 chunk_t chunk)
144{
145 return chunk << store->chunk_shift;
146}
147
148static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
149{
150 /*
151 * There is only ever one instance of a particular block
152 * device so we can compare pointers safely.
153 */
154 return lhs == rhs;
155}
156
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100157struct dm_snap_pending_exception {
Jon Brassow1d4989c2009-12-10 23:52:10 +0000158 struct dm_exception e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
160 /*
161 * Origin buffers waiting for this to complete are held
162 * in a bio list
163 */
164 struct bio_list origin_bios;
165 struct bio_list snapshot_bios;
166
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 /* Pointer back to snapshot context */
168 struct dm_snapshot *snap;
169
170 /*
171 * 1 indicates the exception has already been sent to
172 * kcopyd.
173 */
174 int started;
Mikulas Patockaa6e50b42011-08-02 12:32:04 +0100175
176 /*
177 * For writing a complete chunk, bypassing the copy.
178 */
179 struct bio *full_bio;
180 bio_end_io_t *full_bio_end_io;
181 void *full_bio_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182};
183
184/*
185 * Hash table mapping origin volumes to lists of snapshots and
186 * a lock to protect it
187 */
Christoph Lametere18b8902006-12-06 20:33:20 -0800188static struct kmem_cache *exception_cache;
189static struct kmem_cache *pending_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100191struct dm_snap_tracked_chunk {
192 struct hlist_node node;
193 chunk_t chunk;
194};
195
Mikulas Patockaee180262012-12-21 20:23:41 +0000196static void init_tracked_chunk(struct bio *bio)
197{
198 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
199 INIT_HLIST_NODE(&c->node);
200}
201
202static bool is_bio_tracked(struct bio *bio)
203{
204 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
205 return !hlist_unhashed(&c->node);
206}
207
208static void track_chunk(struct dm_snapshot *s, struct bio *bio, chunk_t chunk)
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100209{
Mikulas Patocka42bc9542012-12-21 20:23:38 +0000210 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100211
212 c->chunk = chunk;
213
Mikulas Patocka9aa0c0e2012-12-21 20:23:33 +0000214 spin_lock_irq(&s->tracked_chunk_lock);
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100215 hlist_add_head(&c->node,
216 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
Mikulas Patocka9aa0c0e2012-12-21 20:23:33 +0000217 spin_unlock_irq(&s->tracked_chunk_lock);
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100218}
219
Mikulas Patockaee180262012-12-21 20:23:41 +0000220static void stop_tracking_chunk(struct dm_snapshot *s, struct bio *bio)
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100221{
Mikulas Patockaee180262012-12-21 20:23:41 +0000222 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100223 unsigned long flags;
224
225 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
226 hlist_del(&c->node);
227 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100228}
229
Mikulas Patockaa8d41b52008-07-21 12:00:34 +0100230static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
231{
232 struct dm_snap_tracked_chunk *c;
Mikulas Patockaa8d41b52008-07-21 12:00:34 +0100233 int found = 0;
234
235 spin_lock_irq(&s->tracked_chunk_lock);
236
Sasha Levinb67bfe02013-02-27 17:06:00 -0800237 hlist_for_each_entry(c,
Mikulas Patockaa8d41b52008-07-21 12:00:34 +0100238 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
239 if (c->chunk == chunk) {
240 found = 1;
241 break;
242 }
243 }
244
245 spin_unlock_irq(&s->tracked_chunk_lock);
246
247 return found;
248}
249
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250/*
Mike Snitzer615d1eb2009-12-10 23:52:29 +0000251 * This conflicting I/O is extremely improbable in the caller,
252 * so msleep(1) is sufficient and there is no need for a wait queue.
253 */
254static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
255{
256 while (__chunk_is_tracked(s, chunk))
257 msleep(1);
258}
259
260/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 * One of these per registered origin, held in the snapshot_origins hash
262 */
263struct origin {
264 /* The origin device */
265 struct block_device *bdev;
266
267 struct list_head hash_list;
268
269 /* List of snapshots for this origin */
270 struct list_head snapshots;
271};
272
273/*
274 * Size of the hash table for origin volumes. If we make this
275 * the size of the minors list then it should be nearly perfect
276 */
277#define ORIGIN_HASH_SIZE 256
278#define ORIGIN_MASK 0xFF
279static struct list_head *_origins;
280static struct rw_semaphore _origins_lock;
281
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000282static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
283static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
284static uint64_t _pending_exceptions_done_count;
285
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286static int init_origin_hash(void)
287{
288 int i;
289
290 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
291 GFP_KERNEL);
292 if (!_origins) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700293 DMERR("unable to allocate memory");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 return -ENOMEM;
295 }
296
297 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
298 INIT_LIST_HEAD(_origins + i);
299 init_rwsem(&_origins_lock);
300
301 return 0;
302}
303
304static void exit_origin_hash(void)
305{
306 kfree(_origins);
307}
308
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100309static unsigned origin_hash(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310{
311 return bdev->bd_dev & ORIGIN_MASK;
312}
313
314static struct origin *__lookup_origin(struct block_device *origin)
315{
316 struct list_head *ol;
317 struct origin *o;
318
319 ol = &_origins[origin_hash(origin)];
320 list_for_each_entry (o, ol, hash_list)
321 if (bdev_equal(o->bdev, origin))
322 return o;
323
324 return NULL;
325}
326
327static void __insert_origin(struct origin *o)
328{
329 struct list_head *sl = &_origins[origin_hash(o->bdev)];
330 list_add_tail(&o->hash_list, sl);
331}
332
333/*
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000334 * _origins_lock must be held when calling this function.
335 * Returns number of snapshots registered using the supplied cow device, plus:
336 * snap_src - a snapshot suitable for use as a source of exception handover
337 * snap_dest - a snapshot capable of receiving exception handover.
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000338 * snap_merge - an existing snapshot-merge target linked to the same origin.
339 * There can be at most one snapshot-merge target. The parameter is optional.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000340 *
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000341 * Possible return values and states of snap_src and snap_dest.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000342 * 0: NULL, NULL - first new snapshot
343 * 1: snap_src, NULL - normal snapshot
344 * 2: snap_src, snap_dest - waiting for handover
345 * 2: snap_src, NULL - handed over, waiting for old to be deleted
346 * 1: NULL, snap_dest - source got destroyed without handover
347 */
348static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
349 struct dm_snapshot **snap_src,
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000350 struct dm_snapshot **snap_dest,
351 struct dm_snapshot **snap_merge)
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000352{
353 struct dm_snapshot *s;
354 struct origin *o;
355 int count = 0;
356 int active;
357
358 o = __lookup_origin(snap->origin->bdev);
359 if (!o)
360 goto out;
361
362 list_for_each_entry(s, &o->snapshots, list) {
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000363 if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
364 *snap_merge = s;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000365 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
366 continue;
367
368 down_read(&s->lock);
369 active = s->active;
370 up_read(&s->lock);
371
372 if (active) {
373 if (snap_src)
374 *snap_src = s;
375 } else if (snap_dest)
376 *snap_dest = s;
377
378 count++;
379 }
380
381out:
382 return count;
383}
384
385/*
386 * On success, returns 1 if this snapshot is a handover destination,
387 * otherwise returns 0.
388 */
389static int __validate_exception_handover(struct dm_snapshot *snap)
390{
391 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000392 struct dm_snapshot *snap_merge = NULL;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000393
394 /* Does snapshot need exceptions handed over to it? */
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000395 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
396 &snap_merge) == 2) ||
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000397 snap_dest) {
398 snap->ti->error = "Snapshot cow pairing for exception "
399 "table handover failed";
400 return -EINVAL;
401 }
402
403 /*
404 * If no snap_src was found, snap cannot become a handover
405 * destination.
406 */
407 if (!snap_src)
408 return 0;
409
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000410 /*
411 * Non-snapshot-merge handover?
412 */
413 if (!dm_target_is_snapshot_merge(snap->ti))
414 return 1;
415
416 /*
417 * Do not allow more than one merging snapshot.
418 */
419 if (snap_merge) {
420 snap->ti->error = "A snapshot is already merging.";
421 return -EINVAL;
422 }
423
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000424 if (!snap_src->store->type->prepare_merge ||
425 !snap_src->store->type->commit_merge) {
426 snap->ti->error = "Snapshot exception store does not "
427 "support snapshot-merge.";
428 return -EINVAL;
429 }
430
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000431 return 1;
432}
433
434static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
435{
436 struct dm_snapshot *l;
437
438 /* Sort the list according to chunk size, largest-first smallest-last */
439 list_for_each_entry(l, &o->snapshots, list)
440 if (l->store->chunk_size < s->store->chunk_size)
441 break;
442 list_add_tail(&s->list, &l->list);
443}
444
445/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 * Make a note of the snapshot and its origin so we can look it
447 * up when the origin has a write on it.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000448 *
449 * Also validate snapshot exception store handovers.
450 * On success, returns 1 if this registration is a handover destination,
451 * otherwise returns 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 */
453static int register_snapshot(struct dm_snapshot *snap)
454{
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000455 struct origin *o, *new_o = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 struct block_device *bdev = snap->origin->bdev;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000457 int r = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000459 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
460 if (!new_o)
461 return -ENOMEM;
462
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 down_write(&_origins_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000465 r = __validate_exception_handover(snap);
466 if (r < 0) {
467 kfree(new_o);
468 goto out;
469 }
470
471 o = __lookup_origin(bdev);
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000472 if (o)
473 kfree(new_o);
474 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 /* New origin */
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000476 o = new_o;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477
478 /* Initialise the struct */
479 INIT_LIST_HEAD(&o->snapshots);
480 o->bdev = bdev;
481
482 __insert_origin(o);
483 }
484
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000485 __insert_snapshot(o, snap);
486
487out:
488 up_write(&_origins_lock);
489
490 return r;
491}
492
493/*
494 * Move snapshot to correct place in list according to chunk size.
495 */
496static void reregister_snapshot(struct dm_snapshot *s)
497{
498 struct block_device *bdev = s->origin->bdev;
499
500 down_write(&_origins_lock);
501
502 list_del(&s->list);
503 __insert_snapshot(__lookup_origin(bdev), s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
505 up_write(&_origins_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506}
507
508static void unregister_snapshot(struct dm_snapshot *s)
509{
510 struct origin *o;
511
512 down_write(&_origins_lock);
513 o = __lookup_origin(s->origin->bdev);
514
515 list_del(&s->list);
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000516 if (o && list_empty(&o->snapshots)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 list_del(&o->hash_list);
518 kfree(o);
519 }
520
521 up_write(&_origins_lock);
522}
523
524/*
525 * Implementation of the exception hash tables.
Milan Brozd74f81f2008-02-08 02:11:27 +0000526 * The lowest hash_shift bits of the chunk number are ignored, allowing
527 * some consecutive chunks to be grouped together.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 */
Jon Brassow3510cb92009-12-10 23:52:11 +0000529static int dm_exception_table_init(struct dm_exception_table *et,
530 uint32_t size, unsigned hash_shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531{
532 unsigned int i;
533
Milan Brozd74f81f2008-02-08 02:11:27 +0000534 et->hash_shift = hash_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 et->hash_mask = size - 1;
536 et->table = dm_vcalloc(size, sizeof(struct list_head));
537 if (!et->table)
538 return -ENOMEM;
539
540 for (i = 0; i < size; i++)
541 INIT_LIST_HEAD(et->table + i);
542
543 return 0;
544}
545
Jon Brassow3510cb92009-12-10 23:52:11 +0000546static void dm_exception_table_exit(struct dm_exception_table *et,
547 struct kmem_cache *mem)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548{
549 struct list_head *slot;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000550 struct dm_exception *ex, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 int i, size;
552
553 size = et->hash_mask + 1;
554 for (i = 0; i < size; i++) {
555 slot = et->table + i;
556
557 list_for_each_entry_safe (ex, next, slot, hash_list)
558 kmem_cache_free(mem, ex);
559 }
560
561 vfree(et->table);
562}
563
Jon Brassow191437a2009-12-10 23:52:10 +0000564static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565{
Milan Brozd74f81f2008-02-08 02:11:27 +0000566 return (chunk >> et->hash_shift) & et->hash_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567}
568
Jon Brassow3510cb92009-12-10 23:52:11 +0000569static void dm_remove_exception(struct dm_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570{
571 list_del(&e->hash_list);
572}
573
574/*
575 * Return the exception data for a sector, or NULL if not
576 * remapped.
577 */
Jon Brassow3510cb92009-12-10 23:52:11 +0000578static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
579 chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580{
581 struct list_head *slot;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000582 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
584 slot = &et->table[exception_hash(et, chunk)];
585 list_for_each_entry (e, slot, hash_list)
Milan Brozd74f81f2008-02-08 02:11:27 +0000586 if (chunk >= e->old_chunk &&
587 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 return e;
589
590 return NULL;
591}
592
Jon Brassow3510cb92009-12-10 23:52:11 +0000593static struct dm_exception *alloc_completed_exception(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594{
Jon Brassow1d4989c2009-12-10 23:52:10 +0000595 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
597 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
598 if (!e)
599 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
600
601 return e;
602}
603
Jon Brassow3510cb92009-12-10 23:52:11 +0000604static void free_completed_exception(struct dm_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605{
606 kmem_cache_free(exception_cache, e);
607}
608
Mikulas Patocka92e86812008-07-21 12:00:35 +0100609static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610{
Mikulas Patocka92e86812008-07-21 12:00:35 +0100611 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
612 GFP_NOIO);
613
Mikulas Patocka879129d22008-10-30 13:33:16 +0000614 atomic_inc(&s->pending_exceptions_count);
Mikulas Patocka92e86812008-07-21 12:00:35 +0100615 pe->snap = s;
616
617 return pe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618}
619
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100620static void free_pending_exception(struct dm_snap_pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621{
Mikulas Patocka879129d22008-10-30 13:33:16 +0000622 struct dm_snapshot *s = pe->snap;
623
624 mempool_free(pe, s->pending_pool);
625 smp_mb__before_atomic_dec();
626 atomic_dec(&s->pending_exceptions_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627}
628
Jon Brassow3510cb92009-12-10 23:52:11 +0000629static void dm_insert_exception(struct dm_exception_table *eh,
630 struct dm_exception *new_e)
Milan Brozd74f81f2008-02-08 02:11:27 +0000631{
Milan Brozd74f81f2008-02-08 02:11:27 +0000632 struct list_head *l;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000633 struct dm_exception *e = NULL;
Milan Brozd74f81f2008-02-08 02:11:27 +0000634
635 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
636
637 /* Add immediately if this table doesn't support consecutive chunks */
638 if (!eh->hash_shift)
639 goto out;
640
641 /* List is ordered by old_chunk */
642 list_for_each_entry_reverse(e, l, hash_list) {
643 /* Insert after an existing chunk? */
644 if (new_e->old_chunk == (e->old_chunk +
645 dm_consecutive_chunk_count(e) + 1) &&
646 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
647 dm_consecutive_chunk_count(e) + 1)) {
648 dm_consecutive_chunk_count_inc(e);
Jon Brassow3510cb92009-12-10 23:52:11 +0000649 free_completed_exception(new_e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000650 return;
651 }
652
653 /* Insert before an existing chunk? */
654 if (new_e->old_chunk == (e->old_chunk - 1) &&
655 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
656 dm_consecutive_chunk_count_inc(e);
657 e->old_chunk--;
658 e->new_chunk--;
Jon Brassow3510cb92009-12-10 23:52:11 +0000659 free_completed_exception(new_e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000660 return;
661 }
662
663 if (new_e->old_chunk > e->old_chunk)
664 break;
665 }
666
667out:
668 list_add(&new_e->hash_list, e ? &e->hash_list : l);
669}
670
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000671/*
672 * Callback used by the exception stores to load exceptions when
673 * initialising.
674 */
675static int dm_add_exception(void *context, chunk_t old, chunk_t new)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676{
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000677 struct dm_snapshot *s = context;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000678 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
Jon Brassow3510cb92009-12-10 23:52:11 +0000680 e = alloc_completed_exception();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 if (!e)
682 return -ENOMEM;
683
684 e->old_chunk = old;
Milan Brozd74f81f2008-02-08 02:11:27 +0000685
686 /* Consecutive_count is implicitly initialised to zero */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 e->new_chunk = new;
Milan Brozd74f81f2008-02-08 02:11:27 +0000688
Jon Brassow3510cb92009-12-10 23:52:11 +0000689 dm_insert_exception(&s->complete, e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000690
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 return 0;
692}
693
Mikulas Patocka7e201b32009-12-10 23:52:08 +0000694/*
695 * Return a minimum chunk size of all snapshots that have the specified origin.
696 * Return zero if the origin has no snapshots.
697 */
Mike Snitzer542f9032012-07-27 15:08:00 +0100698static uint32_t __minimum_chunk_size(struct origin *o)
Mikulas Patocka7e201b32009-12-10 23:52:08 +0000699{
700 struct dm_snapshot *snap;
701 unsigned chunk_size = 0;
702
703 if (o)
704 list_for_each_entry(snap, &o->snapshots, list)
705 chunk_size = min_not_zero(chunk_size,
706 snap->store->chunk_size);
707
Mike Snitzer542f9032012-07-27 15:08:00 +0100708 return (uint32_t) chunk_size;
Mikulas Patocka7e201b32009-12-10 23:52:08 +0000709}
710
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711/*
712 * Hard coded magic.
713 */
714static int calc_max_buckets(void)
715{
716 /* use a fixed size of 2MB */
717 unsigned long mem = 2 * 1024 * 1024;
718 mem /= sizeof(struct list_head);
719
720 return mem;
721}
722
723/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 * Allocate room for a suitable hash table.
725 */
Jonathan Brassowfee19982009-04-02 19:55:34 +0100726static int init_hash_tables(struct dm_snapshot *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727{
Mikulas Patocka60e356f2013-09-18 19:40:42 -0400728 sector_t hash_size, cow_dev_size, max_buckets;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
730 /*
731 * Calculate based on the size of the original volume or
732 * the COW volume...
733 */
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000734 cow_dev_size = get_dev_size(s->cow->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 max_buckets = calc_max_buckets();
736
Mikulas Patocka60e356f2013-09-18 19:40:42 -0400737 hash_size = cow_dev_size >> s->store->chunk_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 hash_size = min(hash_size, max_buckets);
739
Mikulas Patocka8e87b9b2009-12-10 23:51:54 +0000740 if (hash_size < 64)
741 hash_size = 64;
Robert P. J. Day8defd832008-02-08 02:10:06 +0000742 hash_size = rounddown_pow_of_two(hash_size);
Jon Brassow3510cb92009-12-10 23:52:11 +0000743 if (dm_exception_table_init(&s->complete, hash_size,
744 DM_CHUNK_CONSECUTIVE_BITS))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 return -ENOMEM;
746
747 /*
748 * Allocate hash table for in-flight exceptions
749 * Make this smaller than the real hash table
750 */
751 hash_size >>= 3;
752 if (hash_size < 64)
753 hash_size = 64;
754
Jon Brassow3510cb92009-12-10 23:52:11 +0000755 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
756 dm_exception_table_exit(&s->complete, exception_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 return -ENOMEM;
758 }
759
760 return 0;
761}
762
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000763static void merge_shutdown(struct dm_snapshot *s)
764{
765 clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
766 smp_mb__after_clear_bit();
767 wake_up_bit(&s->state_bits, RUNNING_MERGE);
768}
769
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000770static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
771{
772 s->first_merging_chunk = 0;
773 s->num_merging_chunks = 0;
774
775 return bio_list_get(&s->bios_queued_during_merge);
776}
777
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000778/*
779 * Remove one chunk from the index of completed exceptions.
780 */
781static int __remove_single_exception_chunk(struct dm_snapshot *s,
782 chunk_t old_chunk)
783{
784 struct dm_exception *e;
785
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000786 e = dm_lookup_exception(&s->complete, old_chunk);
787 if (!e) {
788 DMERR("Corruption detected: exception for block %llu is "
789 "on disk but not in memory",
790 (unsigned long long)old_chunk);
791 return -EINVAL;
792 }
793
794 /*
795 * If this is the only chunk using this exception, remove exception.
796 */
797 if (!dm_consecutive_chunk_count(e)) {
798 dm_remove_exception(e);
799 free_completed_exception(e);
800 return 0;
801 }
802
803 /*
804 * The chunk may be either at the beginning or the end of a
805 * group of consecutive chunks - never in the middle. We are
806 * removing chunks in the opposite order to that in which they
807 * were added, so this should always be true.
808 * Decrement the consecutive chunk counter and adjust the
809 * starting point if necessary.
810 */
811 if (old_chunk == e->old_chunk) {
812 e->old_chunk++;
813 e->new_chunk++;
814 } else if (old_chunk != e->old_chunk +
815 dm_consecutive_chunk_count(e)) {
816 DMERR("Attempt to merge block %llu from the "
817 "middle of a chunk range [%llu - %llu]",
818 (unsigned long long)old_chunk,
819 (unsigned long long)e->old_chunk,
820 (unsigned long long)
821 e->old_chunk + dm_consecutive_chunk_count(e));
822 return -EINVAL;
823 }
824
825 dm_consecutive_chunk_count_dec(e);
826
827 return 0;
828}
829
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000830static void flush_bios(struct bio *bio);
831
832static int remove_single_exception_chunk(struct dm_snapshot *s)
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000833{
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000834 struct bio *b = NULL;
835 int r;
836 chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000837
838 down_write(&s->lock);
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000839
840 /*
841 * Process chunks (and associated exceptions) in reverse order
842 * so that dm_consecutive_chunk_count_dec() accounting works.
843 */
844 do {
845 r = __remove_single_exception_chunk(s, old_chunk);
846 if (r)
847 goto out;
848 } while (old_chunk-- > s->first_merging_chunk);
849
850 b = __release_queued_bios_after_merge(s);
851
852out:
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000853 up_write(&s->lock);
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000854 if (b)
855 flush_bios(b);
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000856
857 return r;
858}
859
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000860static int origin_write_extent(struct dm_snapshot *merging_snap,
861 sector_t sector, unsigned chunk_size);
862
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000863static void merge_callback(int read_err, unsigned long write_err,
864 void *context);
865
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000866static uint64_t read_pending_exceptions_done_count(void)
867{
868 uint64_t pending_exceptions_done;
869
870 spin_lock(&_pending_exceptions_done_spinlock);
871 pending_exceptions_done = _pending_exceptions_done_count;
872 spin_unlock(&_pending_exceptions_done_spinlock);
873
874 return pending_exceptions_done;
875}
876
877static void increment_pending_exceptions_done_count(void)
878{
879 spin_lock(&_pending_exceptions_done_spinlock);
880 _pending_exceptions_done_count++;
881 spin_unlock(&_pending_exceptions_done_spinlock);
882
883 wake_up_all(&_pending_exceptions_done);
884}
885
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000886static void snapshot_merge_next_chunks(struct dm_snapshot *s)
887{
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000888 int i, linear_chunks;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000889 chunk_t old_chunk, new_chunk;
890 struct dm_io_region src, dest;
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000891 sector_t io_size;
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000892 uint64_t previous_count;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000893
894 BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
895 if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
896 goto shut;
897
898 /*
899 * valid flag never changes during merge, so no lock required.
900 */
901 if (!s->valid) {
902 DMERR("Snapshot is invalid: can't merge");
903 goto shut;
904 }
905
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000906 linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
907 &new_chunk);
908 if (linear_chunks <= 0) {
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +0000909 if (linear_chunks < 0) {
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000910 DMERR("Read error in exception store: "
911 "shutting down merge");
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +0000912 down_write(&s->lock);
913 s->merge_failed = 1;
914 up_write(&s->lock);
915 }
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000916 goto shut;
917 }
918
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000919 /* Adjust old_chunk and new_chunk to reflect start of linear region */
920 old_chunk = old_chunk + 1 - linear_chunks;
921 new_chunk = new_chunk + 1 - linear_chunks;
922
923 /*
924 * Use one (potentially large) I/O to copy all 'linear_chunks'
925 * from the exception store to the origin
926 */
927 io_size = linear_chunks * s->store->chunk_size;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000928
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000929 dest.bdev = s->origin->bdev;
930 dest.sector = chunk_to_sector(s->store, old_chunk);
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000931 dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000932
933 src.bdev = s->cow->bdev;
934 src.sector = chunk_to_sector(s->store, new_chunk);
935 src.count = dest.count;
936
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000937 /*
938 * Reallocate any exceptions needed in other snapshots then
939 * wait for the pending exceptions to complete.
940 * Each time any pending exception (globally on the system)
941 * completes we are woken and repeat the process to find out
942 * if we can proceed. While this may not seem a particularly
943 * efficient algorithm, it is not expected to have any
944 * significant impact on performance.
945 */
946 previous_count = read_pending_exceptions_done_count();
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000947 while (origin_write_extent(s, dest.sector, io_size)) {
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000948 wait_event(_pending_exceptions_done,
949 (read_pending_exceptions_done_count() !=
950 previous_count));
951 /* Retry after the wait, until all exceptions are done. */
952 previous_count = read_pending_exceptions_done_count();
953 }
954
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000955 down_write(&s->lock);
956 s->first_merging_chunk = old_chunk;
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000957 s->num_merging_chunks = linear_chunks;
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000958 up_write(&s->lock);
959
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000960 /* Wait until writes to all 'linear_chunks' drain */
961 for (i = 0; i < linear_chunks; i++)
962 __check_for_conflicting_io(s, old_chunk + i);
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000963
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000964 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
965 return;
966
967shut:
968 merge_shutdown(s);
969}
970
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000971static void error_bios(struct bio *bio);
972
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000973static void merge_callback(int read_err, unsigned long write_err, void *context)
974{
975 struct dm_snapshot *s = context;
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000976 struct bio *b = NULL;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000977
978 if (read_err || write_err) {
979 if (read_err)
980 DMERR("Read error: shutting down merge.");
981 else
982 DMERR("Write error: shutting down merge.");
983 goto shut;
984 }
985
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000986 if (s->store->type->commit_merge(s->store,
987 s->num_merging_chunks) < 0) {
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000988 DMERR("Write error in exception store: shutting down merge");
989 goto shut;
990 }
991
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000992 if (remove_single_exception_chunk(s) < 0)
993 goto shut;
994
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000995 snapshot_merge_next_chunks(s);
996
997 return;
998
999shut:
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001000 down_write(&s->lock);
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +00001001 s->merge_failed = 1;
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001002 b = __release_queued_bios_after_merge(s);
1003 up_write(&s->lock);
1004 error_bios(b);
1005
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001006 merge_shutdown(s);
1007}
1008
1009static void start_merge(struct dm_snapshot *s)
1010{
1011 if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
1012 snapshot_merge_next_chunks(s);
1013}
1014
1015static int wait_schedule(void *ptr)
1016{
1017 schedule();
1018
1019 return 0;
1020}
1021
1022/*
1023 * Stop the merging process and wait until it finishes.
1024 */
1025static void stop_merge(struct dm_snapshot *s)
1026{
1027 set_bit(SHUTDOWN_MERGE, &s->state_bits);
1028 wait_on_bit(&s->state_bits, RUNNING_MERGE, wait_schedule,
1029 TASK_UNINTERRUPTIBLE);
1030 clear_bit(SHUTDOWN_MERGE, &s->state_bits);
1031}
1032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
1035 */
1036static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1037{
1038 struct dm_snapshot *s;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001039 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 int r = -EINVAL;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001041 char *origin_path, *cow_path;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00001042 unsigned args_used, num_flush_bios = 1;
Mike Snitzer10b81062009-12-10 23:52:31 +00001043 fmode_t origin_mode = FMODE_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -07001045 if (argc != 4) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001046 ti->error = "requires exactly 4 arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 r = -EINVAL;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001048 goto bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 }
1050
Mike Snitzer10b81062009-12-10 23:52:31 +00001051 if (dm_target_is_snapshot_merge(ti)) {
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00001052 num_flush_bios = 2;
Mike Snitzer10b81062009-12-10 23:52:31 +00001053 origin_mode = FMODE_WRITE;
1054 }
1055
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 s = kmalloc(sizeof(*s), GFP_KERNEL);
Jonathan Brassowfee19982009-04-02 19:55:34 +01001057 if (!s) {
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001058 ti->error = "Cannot allocate private snapshot structure";
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 r = -ENOMEM;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001060 goto bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 }
1062
Mikulas Patockac2411042010-08-12 04:13:51 +01001063 origin_path = argv[0];
1064 argv++;
1065 argc--;
1066
1067 r = dm_get_device(ti, origin_path, origin_mode, &s->origin);
1068 if (r) {
1069 ti->error = "Cannot get origin device";
1070 goto bad_origin;
1071 }
1072
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001073 cow_path = argv[0];
1074 argv++;
1075 argc--;
1076
Milan Broz024d37e2011-03-24 13:52:14 +00001077 r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow);
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001078 if (r) {
1079 ti->error = "Cannot get COW device";
1080 goto bad_cow;
1081 }
1082
1083 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
1084 if (r) {
1085 ti->error = "Couldn't create exception store";
1086 r = -EINVAL;
1087 goto bad_store;
1088 }
1089
1090 argv += args_used;
1091 argc -= args_used;
1092
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001093 s->ti = ti;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 s->valid = 1;
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001095 s->active = 0;
Mikulas Patocka879129d22008-10-30 13:33:16 +00001096 atomic_set(&s->pending_exceptions_count, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 init_rwsem(&s->lock);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001098 INIT_LIST_HEAD(&s->list);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07001099 spin_lock_init(&s->pe_lock);
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001100 s->state_bits = 0;
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +00001101 s->merge_failed = 0;
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001102 s->first_merging_chunk = 0;
1103 s->num_merging_chunks = 0;
1104 bio_list_init(&s->bios_queued_during_merge);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
1106 /* Allocate hash table for COW data */
Jonathan Brassowfee19982009-04-02 19:55:34 +01001107 if (init_hash_tables(s)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 ti->error = "Unable to allocate hash table space";
1109 r = -ENOMEM;
Jonathan Brassowfee19982009-04-02 19:55:34 +01001110 goto bad_hash_tables;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 }
1112
Mikulas Patockadf5d2e92013-03-01 22:45:49 +00001113 s->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle);
Mikulas Patockafa34ce72011-05-29 13:03:13 +01001114 if (IS_ERR(s->kcopyd_client)) {
1115 r = PTR_ERR(s->kcopyd_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 ti->error = "Could not create kcopyd client";
Jonathan Brassowfee19982009-04-02 19:55:34 +01001117 goto bad_kcopyd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 }
1119
Mikulas Patocka92e86812008-07-21 12:00:35 +01001120 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
1121 if (!s->pending_pool) {
1122 ti->error = "Could not allocate mempool for pending exceptions";
Wei Yongjun09e8b812013-05-10 14:37:15 +01001123 r = -ENOMEM;
Jonathan Brassowfee19982009-04-02 19:55:34 +01001124 goto bad_pending_pool;
Mikulas Patocka92e86812008-07-21 12:00:35 +01001125 }
1126
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001127 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1128 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
1129
1130 spin_lock_init(&s->tracked_chunk_lock);
1131
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001132 ti->private = s;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00001133 ti->num_flush_bios = num_flush_bios;
Mikulas Patocka42bc9542012-12-21 20:23:38 +00001134 ti->per_bio_data_size = sizeof(struct dm_snap_tracked_chunk);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001135
1136 /* Add snapshot to the list of snapshots for this origin */
1137 /* Exceptions aren't triggered till snapshot_resume() is called */
1138 r = register_snapshot(s);
1139 if (r == -ENOMEM) {
1140 ti->error = "Snapshot origin struct allocation failed";
1141 goto bad_load_and_register;
1142 } else if (r < 0) {
1143 /* invalid handover, register_snapshot has set ti->error */
1144 goto bad_load_and_register;
1145 }
1146
1147 /*
1148 * Metadata must only be loaded into one table at once, so skip this
1149 * if metadata will be handed over during resume.
1150 * Chunk size will be set during the handover - set it to zero to
1151 * ensure it's ignored.
1152 */
1153 if (r > 0) {
1154 s->store->chunk_size = 0;
1155 return 0;
1156 }
1157
Jonathan Brassow493df712009-04-02 19:55:31 +01001158 r = s->store->type->read_metadata(s->store, dm_add_exception,
1159 (void *)s);
Milan Broz07641472007-07-12 17:28:13 +01001160 if (r < 0) {
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -07001161 ti->error = "Failed to read snapshot metadata";
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001162 goto bad_read_metadata;
Milan Broz07641472007-07-12 17:28:13 +01001163 } else if (r > 0) {
1164 s->valid = 0;
1165 DMWARN("Snapshot is marked invalid.");
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -07001166 }
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001167
Mikulas Patocka3f2412d2009-10-16 23:18:16 +01001168 if (!s->store->chunk_size) {
1169 ti->error = "Chunk size not set";
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001170 goto bad_read_metadata;
Mikulas Patocka3f2412d2009-10-16 23:18:16 +01001171 }
Mike Snitzer542f9032012-07-27 15:08:00 +01001172
1173 r = dm_set_target_max_io_len(ti, s->store->chunk_size);
1174 if (r)
1175 goto bad_read_metadata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
1177 return 0;
1178
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001179bad_read_metadata:
1180 unregister_snapshot(s);
1181
Jonathan Brassowfee19982009-04-02 19:55:34 +01001182bad_load_and_register:
Mikulas Patocka92e86812008-07-21 12:00:35 +01001183 mempool_destroy(s->pending_pool);
1184
Jonathan Brassowfee19982009-04-02 19:55:34 +01001185bad_pending_pool:
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +01001186 dm_kcopyd_client_destroy(s->kcopyd_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187
Jonathan Brassowfee19982009-04-02 19:55:34 +01001188bad_kcopyd:
Jon Brassow3510cb92009-12-10 23:52:11 +00001189 dm_exception_table_exit(&s->pending, pending_cache);
1190 dm_exception_table_exit(&s->complete, exception_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
Jonathan Brassowfee19982009-04-02 19:55:34 +01001192bad_hash_tables:
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001193 dm_exception_store_destroy(s->store);
1194
1195bad_store:
1196 dm_put_device(ti, s->cow);
1197
1198bad_cow:
Mikulas Patockac2411042010-08-12 04:13:51 +01001199 dm_put_device(ti, s->origin);
1200
1201bad_origin:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 kfree(s);
1203
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001204bad:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 return r;
1206}
1207
Milan Broz31c93a02006-12-08 02:41:11 -08001208static void __free_exceptions(struct dm_snapshot *s)
1209{
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +01001210 dm_kcopyd_client_destroy(s->kcopyd_client);
Milan Broz31c93a02006-12-08 02:41:11 -08001211 s->kcopyd_client = NULL;
1212
Jon Brassow3510cb92009-12-10 23:52:11 +00001213 dm_exception_table_exit(&s->pending, pending_cache);
1214 dm_exception_table_exit(&s->complete, exception_cache);
Milan Broz31c93a02006-12-08 02:41:11 -08001215}
1216
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001217static void __handover_exceptions(struct dm_snapshot *snap_src,
1218 struct dm_snapshot *snap_dest)
1219{
1220 union {
1221 struct dm_exception_table table_swap;
1222 struct dm_exception_store *store_swap;
1223 } u;
1224
1225 /*
1226 * Swap all snapshot context information between the two instances.
1227 */
1228 u.table_swap = snap_dest->complete;
1229 snap_dest->complete = snap_src->complete;
1230 snap_src->complete = u.table_swap;
1231
1232 u.store_swap = snap_dest->store;
1233 snap_dest->store = snap_src->store;
1234 snap_src->store = u.store_swap;
1235
1236 snap_dest->store->snap = snap_dest;
1237 snap_src->store->snap = snap_src;
1238
Mike Snitzer542f9032012-07-27 15:08:00 +01001239 snap_dest->ti->max_io_len = snap_dest->store->chunk_size;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001240 snap_dest->valid = snap_src->valid;
1241
1242 /*
1243 * Set source invalid to ensure it receives no further I/O.
1244 */
1245 snap_src->valid = 0;
1246}
1247
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248static void snapshot_dtr(struct dm_target *ti)
1249{
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001250#ifdef CONFIG_DM_DEBUG
1251 int i;
1252#endif
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001253 struct dm_snapshot *s = ti->private;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001254 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001256 down_read(&_origins_lock);
1257 /* Check whether exception handover must be cancelled */
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +00001258 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001259 if (snap_src && snap_dest && (s == snap_src)) {
1260 down_write(&snap_dest->lock);
1261 snap_dest->valid = 0;
1262 up_write(&snap_dest->lock);
1263 DMERR("Cancelling snapshot handover.");
1264 }
1265 up_read(&_origins_lock);
1266
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001267 if (dm_target_is_snapshot_merge(ti))
1268 stop_merge(s);
1269
Alasdair G Kergon138728dc2006-03-27 01:17:50 -08001270 /* Prevent further origin writes from using this snapshot. */
1271 /* After this returns there can be no new kcopyd jobs. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 unregister_snapshot(s);
1273
Mikulas Patocka879129d22008-10-30 13:33:16 +00001274 while (atomic_read(&s->pending_exceptions_count))
Mikulas Patocka90fa1522009-01-06 03:04:54 +00001275 msleep(1);
Mikulas Patocka879129d22008-10-30 13:33:16 +00001276 /*
1277 * Ensure instructions in mempool_destroy aren't reordered
1278 * before atomic_read.
1279 */
1280 smp_mb();
1281
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001282#ifdef CONFIG_DM_DEBUG
1283 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1284 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
1285#endif
1286
Milan Broz31c93a02006-12-08 02:41:11 -08001287 __free_exceptions(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288
Mikulas Patocka92e86812008-07-21 12:00:35 +01001289 mempool_destroy(s->pending_pool);
1290
Jonathan Brassowfee19982009-04-02 19:55:34 +01001291 dm_exception_store_destroy(s->store);
Alasdair G Kergon138728dc2006-03-27 01:17:50 -08001292
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001293 dm_put_device(ti, s->cow);
1294
Mikulas Patockac2411042010-08-12 04:13:51 +01001295 dm_put_device(ti, s->origin);
1296
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 kfree(s);
1298}
1299
1300/*
1301 * Flush a list of buffers.
1302 */
1303static void flush_bios(struct bio *bio)
1304{
1305 struct bio *n;
1306
1307 while (bio) {
1308 n = bio->bi_next;
1309 bio->bi_next = NULL;
1310 generic_make_request(bio);
1311 bio = n;
1312 }
1313}
1314
Mikulas Patocka515ad662009-12-10 23:52:30 +00001315static int do_origin(struct dm_dev *origin, struct bio *bio);
1316
1317/*
1318 * Flush a list of buffers.
1319 */
1320static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1321{
1322 struct bio *n;
1323 int r;
1324
1325 while (bio) {
1326 n = bio->bi_next;
1327 bio->bi_next = NULL;
1328 r = do_origin(s->origin, bio);
1329 if (r == DM_MAPIO_REMAPPED)
1330 generic_make_request(bio);
1331 bio = n;
1332 }
1333}
1334
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335/*
1336 * Error a list of buffers.
1337 */
1338static void error_bios(struct bio *bio)
1339{
1340 struct bio *n;
1341
1342 while (bio) {
1343 n = bio->bi_next;
1344 bio->bi_next = NULL;
NeilBrown6712ecf2007-09-27 12:47:43 +02001345 bio_io_error(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 bio = n;
1347 }
1348}
1349
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001350static void __invalidate_snapshot(struct dm_snapshot *s, int err)
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001351{
1352 if (!s->valid)
1353 return;
1354
1355 if (err == -EIO)
1356 DMERR("Invalidating snapshot: Error reading/writing.");
1357 else if (err == -ENOMEM)
1358 DMERR("Invalidating snapshot: Unable to allocate exception.");
1359
Jonathan Brassow493df712009-04-02 19:55:31 +01001360 if (s->store->type->drop_snapshot)
1361 s->store->type->drop_snapshot(s->store);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001362
1363 s->valid = 0;
1364
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001365 dm_table_event(s->ti->table);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001366}
1367
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001368static void pending_complete(struct dm_snap_pending_exception *pe, int success)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369{
Jon Brassow1d4989c2009-12-10 23:52:10 +00001370 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 struct dm_snapshot *s = pe->snap;
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001372 struct bio *origin_bios = NULL;
1373 struct bio *snapshot_bios = NULL;
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001374 struct bio *full_bio = NULL;
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001375 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001377 if (!success) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 /* Read/write error - snapshot is unusable */
1379 down_write(&s->lock);
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001380 __invalidate_snapshot(s, -EIO);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001381 error = 1;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001382 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 }
1384
Jon Brassow3510cb92009-12-10 23:52:11 +00001385 e = alloc_completed_exception();
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001386 if (!e) {
1387 down_write(&s->lock);
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001388 __invalidate_snapshot(s, -ENOMEM);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001389 error = 1;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001390 goto out;
1391 }
1392 *e = pe->e;
1393
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001394 down_write(&s->lock);
1395 if (!s->valid) {
Jon Brassow3510cb92009-12-10 23:52:11 +00001396 free_completed_exception(e);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001397 error = 1;
1398 goto out;
1399 }
1400
Mike Snitzer615d1eb2009-12-10 23:52:29 +00001401 /* Check for conflicting reads */
1402 __check_for_conflicting_io(s, pe->e.old_chunk);
Mikulas Patockaa8d41b52008-07-21 12:00:34 +01001403
1404 /*
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001405 * Add a proper exception, and remove the
1406 * in-flight exception from the list.
1407 */
Jon Brassow3510cb92009-12-10 23:52:11 +00001408 dm_insert_exception(&s->complete, e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001409
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001410out:
Jon Brassow3510cb92009-12-10 23:52:11 +00001411 dm_remove_exception(&pe->e);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001412 snapshot_bios = bio_list_get(&pe->snapshot_bios);
Mikulas Patocka515ad662009-12-10 23:52:30 +00001413 origin_bios = bio_list_get(&pe->origin_bios);
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001414 full_bio = pe->full_bio;
1415 if (full_bio) {
1416 full_bio->bi_end_io = pe->full_bio_end_io;
1417 full_bio->bi_private = pe->full_bio_private;
1418 }
Mikulas Patocka515ad662009-12-10 23:52:30 +00001419 free_pending_exception(pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420
Mikulas Patocka73dfd072009-12-10 23:52:34 +00001421 increment_pending_exceptions_done_count();
1422
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001423 up_write(&s->lock);
1424
1425 /* Submit any pending write bios */
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001426 if (error) {
1427 if (full_bio)
1428 bio_io_error(full_bio);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001429 error_bios(snapshot_bios);
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001430 } else {
1431 if (full_bio)
1432 bio_endio(full_bio, 0);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001433 flush_bios(snapshot_bios);
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001434 }
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001435
Mikulas Patocka515ad662009-12-10 23:52:30 +00001436 retry_origin_bios(s, origin_bios);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437}
1438
1439static void commit_callback(void *context, int success)
1440{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001441 struct dm_snap_pending_exception *pe = context;
1442
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 pending_complete(pe, success);
1444}
1445
1446/*
1447 * Called when the copy I/O has finished. kcopyd actually runs
1448 * this code so don't block.
1449 */
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -07001450static void copy_callback(int read_err, unsigned long write_err, void *context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001452 struct dm_snap_pending_exception *pe = context;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 struct dm_snapshot *s = pe->snap;
1454
1455 if (read_err || write_err)
1456 pending_complete(pe, 0);
1457
1458 else
1459 /* Update the metadata if we are persistent */
Jonathan Brassow493df712009-04-02 19:55:31 +01001460 s->store->type->commit_exception(s->store, &pe->e,
1461 commit_callback, pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462}
1463
1464/*
1465 * Dispatches the copy operation to kcopyd.
1466 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001467static void start_copy(struct dm_snap_pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468{
1469 struct dm_snapshot *s = pe->snap;
Heinz Mauelshagen22a1ceb2008-04-24 21:43:17 +01001470 struct dm_io_region src, dest;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 struct block_device *bdev = s->origin->bdev;
1472 sector_t dev_size;
1473
1474 dev_size = get_dev_size(bdev);
1475
1476 src.bdev = bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001477 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
Mikulas Patockadf96eee2009-10-16 23:18:17 +01001478 src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001480 dest.bdev = s->cow->bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001481 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 dest.count = src.count;
1483
1484 /* Hand over to kcopyd */
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001485 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486}
1487
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001488static void full_bio_end_io(struct bio *bio, int error)
1489{
1490 void *callback_data = bio->bi_private;
1491
1492 dm_kcopyd_do_callback(callback_data, 0, error ? 1 : 0);
1493}
1494
1495static void start_full_bio(struct dm_snap_pending_exception *pe,
1496 struct bio *bio)
1497{
1498 struct dm_snapshot *s = pe->snap;
1499 void *callback_data;
1500
1501 pe->full_bio = bio;
1502 pe->full_bio_end_io = bio->bi_end_io;
1503 pe->full_bio_private = bio->bi_private;
1504
1505 callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client,
1506 copy_callback, pe);
1507
1508 bio->bi_end_io = full_bio_end_io;
1509 bio->bi_private = callback_data;
1510
1511 generic_make_request(bio);
1512}
1513
Mikulas Patocka29138082009-04-02 19:55:25 +01001514static struct dm_snap_pending_exception *
1515__lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1516{
Jon Brassow3510cb92009-12-10 23:52:11 +00001517 struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01001518
1519 if (!e)
1520 return NULL;
1521
1522 return container_of(e, struct dm_snap_pending_exception, e);
1523}
1524
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525/*
1526 * Looks to see if this snapshot already has a pending exception
1527 * for this chunk, otherwise it allocates a new one and inserts
1528 * it into the pending table.
1529 *
1530 * NOTE: a write lock must be held on snap->lock before calling
1531 * this.
1532 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001533static struct dm_snap_pending_exception *
Mikulas Patockac6621392009-04-02 19:55:25 +01001534__find_pending_exception(struct dm_snapshot *s,
1535 struct dm_snap_pending_exception *pe, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536{
Mikulas Patockac6621392009-04-02 19:55:25 +01001537 struct dm_snap_pending_exception *pe2;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001538
Mikulas Patocka29138082009-04-02 19:55:25 +01001539 pe2 = __lookup_pending_exception(s, chunk);
1540 if (pe2) {
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001541 free_pending_exception(pe);
Mikulas Patocka29138082009-04-02 19:55:25 +01001542 return pe2;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001543 }
1544
1545 pe->e.old_chunk = chunk;
1546 bio_list_init(&pe->origin_bios);
1547 bio_list_init(&pe->snapshot_bios);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001548 pe->started = 0;
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001549 pe->full_bio = NULL;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001550
Jonathan Brassow493df712009-04-02 19:55:31 +01001551 if (s->store->type->prepare_exception(s->store, &pe->e)) {
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001552 free_pending_exception(pe);
1553 return NULL;
1554 }
1555
Jon Brassow3510cb92009-12-10 23:52:11 +00001556 dm_insert_exception(&s->pending, &pe->e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001557
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 return pe;
1559}
1560
Jon Brassow1d4989c2009-12-10 23:52:10 +00001561static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
Milan Brozd74f81f2008-02-08 02:11:27 +00001562 struct bio *bio, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563{
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001564 bio->bi_bdev = s->cow->bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001565 bio->bi_sector = chunk_to_sector(s->store,
1566 dm_chunk_number(e->new_chunk) +
1567 (chunk - e->old_chunk)) +
1568 (bio->bi_sector &
1569 s->store->chunk_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570}
1571
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00001572static int snapshot_map(struct dm_target *ti, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573{
Jon Brassow1d4989c2009-12-10 23:52:10 +00001574 struct dm_exception *e;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001575 struct dm_snapshot *s = ti->private;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001576 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 chunk_t chunk;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001578 struct dm_snap_pending_exception *pe = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
Mikulas Patockaee180262012-12-21 20:23:41 +00001580 init_tracked_chunk(bio);
1581
Tejun Heod87f4c12010-09-03 11:56:19 +02001582 if (bio->bi_rw & REQ_FLUSH) {
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001583 bio->bi_bdev = s->cow->bdev;
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01001584 return DM_MAPIO_REMAPPED;
1585 }
1586
Jonathan Brassow71fab002009-04-02 19:55:33 +01001587 chunk = sector_to_chunk(s->store, bio->bi_sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588
1589 /* Full snapshots are not usable */
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001590 /* To get here the table must be live so s->active is always set. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 if (!s->valid)
Alasdair G Kergonf6a80ea2005-07-12 15:53:01 -07001592 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001594 /* FIXME: should only take write lock if we need
1595 * to copy an exception */
1596 down_write(&s->lock);
1597
1598 if (!s->valid) {
1599 r = -EIO;
1600 goto out_unlock;
1601 }
1602
1603 /* If the block is already remapped - use that, else remap it */
Jon Brassow3510cb92009-12-10 23:52:11 +00001604 e = dm_lookup_exception(&s->complete, chunk);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001605 if (e) {
Milan Brozd74f81f2008-02-08 02:11:27 +00001606 remap_exception(s, e, bio, chunk);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001607 goto out_unlock;
1608 }
1609
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 /*
1611 * Write to snapshot - higher level takes care of RW/RO
1612 * flags so we should only get this if we are
1613 * writeable.
1614 */
1615 if (bio_rw(bio) == WRITE) {
Mikulas Patocka29138082009-04-02 19:55:25 +01001616 pe = __lookup_pending_exception(s, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001617 if (!pe) {
Mikulas Patockac6621392009-04-02 19:55:25 +01001618 up_write(&s->lock);
1619 pe = alloc_pending_exception(s);
1620 down_write(&s->lock);
1621
1622 if (!s->valid) {
1623 free_pending_exception(pe);
1624 r = -EIO;
1625 goto out_unlock;
1626 }
1627
Jon Brassow3510cb92009-12-10 23:52:11 +00001628 e = dm_lookup_exception(&s->complete, chunk);
Mikulas Patocka35bf6592009-04-02 19:55:26 +01001629 if (e) {
1630 free_pending_exception(pe);
1631 remap_exception(s, e, bio, chunk);
1632 goto out_unlock;
1633 }
1634
Mikulas Patockac6621392009-04-02 19:55:25 +01001635 pe = __find_pending_exception(s, pe, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01001636 if (!pe) {
1637 __invalidate_snapshot(s, -ENOMEM);
1638 r = -EIO;
1639 goto out_unlock;
1640 }
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001641 }
1642
Milan Brozd74f81f2008-02-08 02:11:27 +00001643 remap_exception(s, &pe->e, bio, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001644
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001645 r = DM_MAPIO_SUBMITTED;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001646
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001647 if (!pe->started &&
1648 bio->bi_size == (s->store->chunk_size << SECTOR_SHIFT)) {
1649 pe->started = 1;
1650 up_write(&s->lock);
1651 start_full_bio(pe, bio);
1652 goto out;
1653 }
1654
1655 bio_list_add(&pe->snapshot_bios, bio);
1656
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001657 if (!pe->started) {
1658 /* this is protected by snap->lock */
1659 pe->started = 1;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001660 up_write(&s->lock);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001661 start_copy(pe);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001662 goto out;
1663 }
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001664 } else {
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001665 bio->bi_bdev = s->origin->bdev;
Mikulas Patockaee180262012-12-21 20:23:41 +00001666 track_chunk(s, bio, chunk);
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001667 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001669out_unlock:
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001670 up_write(&s->lock);
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001671out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 return r;
1673}
1674
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001675/*
1676 * A snapshot-merge target behaves like a combination of a snapshot
1677 * target and a snapshot-origin target. It only generates new
1678 * exceptions in other snapshots and not in the one that is being
1679 * merged.
1680 *
1681 * For each chunk, if there is an existing exception, it is used to
1682 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
1683 * which in turn might generate exceptions in other snapshots.
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001684 * If merging is currently taking place on the chunk in question, the
1685 * I/O is deferred by adding it to s->bios_queued_during_merge.
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001686 */
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00001687static int snapshot_merge_map(struct dm_target *ti, struct bio *bio)
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001688{
1689 struct dm_exception *e;
1690 struct dm_snapshot *s = ti->private;
1691 int r = DM_MAPIO_REMAPPED;
1692 chunk_t chunk;
1693
Mikulas Patockaee180262012-12-21 20:23:41 +00001694 init_tracked_chunk(bio);
1695
Tejun Heod87f4c12010-09-03 11:56:19 +02001696 if (bio->bi_rw & REQ_FLUSH) {
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00001697 if (!dm_bio_get_target_bio_nr(bio))
Mike Snitzer10b81062009-12-10 23:52:31 +00001698 bio->bi_bdev = s->origin->bdev;
1699 else
1700 bio->bi_bdev = s->cow->bdev;
Mike Snitzer10b81062009-12-10 23:52:31 +00001701 return DM_MAPIO_REMAPPED;
1702 }
1703
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001704 chunk = sector_to_chunk(s->store, bio->bi_sector);
1705
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001706 down_write(&s->lock);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001707
Mikulas Patockad2fdb772009-12-10 23:52:36 +00001708 /* Full merging snapshots are redirected to the origin */
1709 if (!s->valid)
1710 goto redirect_to_origin;
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001711
1712 /* If the block is already remapped - use that */
1713 e = dm_lookup_exception(&s->complete, chunk);
1714 if (e) {
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001715 /* Queue writes overlapping with chunks being merged */
1716 if (bio_rw(bio) == WRITE &&
1717 chunk >= s->first_merging_chunk &&
1718 chunk < (s->first_merging_chunk +
1719 s->num_merging_chunks)) {
1720 bio->bi_bdev = s->origin->bdev;
1721 bio_list_add(&s->bios_queued_during_merge, bio);
1722 r = DM_MAPIO_SUBMITTED;
1723 goto out_unlock;
1724 }
Mikulas Patocka17aa0332009-12-10 23:52:33 +00001725
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001726 remap_exception(s, e, bio, chunk);
Mikulas Patocka17aa0332009-12-10 23:52:33 +00001727
1728 if (bio_rw(bio) == WRITE)
Mikulas Patockaee180262012-12-21 20:23:41 +00001729 track_chunk(s, bio, chunk);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001730 goto out_unlock;
1731 }
1732
Mikulas Patockad2fdb772009-12-10 23:52:36 +00001733redirect_to_origin:
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001734 bio->bi_bdev = s->origin->bdev;
1735
1736 if (bio_rw(bio) == WRITE) {
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001737 up_write(&s->lock);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001738 return do_origin(s->origin, bio);
1739 }
1740
1741out_unlock:
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001742 up_write(&s->lock);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001743
1744 return r;
1745}
1746
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00001747static int snapshot_end_io(struct dm_target *ti, struct bio *bio, int error)
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001748{
1749 struct dm_snapshot *s = ti->private;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001750
Mikulas Patockaee180262012-12-21 20:23:41 +00001751 if (is_bio_tracked(bio))
1752 stop_tracking_chunk(s, bio);
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001753
1754 return 0;
1755}
1756
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001757static void snapshot_merge_presuspend(struct dm_target *ti)
1758{
1759 struct dm_snapshot *s = ti->private;
1760
1761 stop_merge(s);
1762}
1763
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001764static int snapshot_preresume(struct dm_target *ti)
1765{
1766 int r = 0;
1767 struct dm_snapshot *s = ti->private;
1768 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1769
1770 down_read(&_origins_lock);
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +00001771 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001772 if (snap_src && snap_dest) {
1773 down_read(&snap_src->lock);
1774 if (s == snap_src) {
1775 DMERR("Unable to resume snapshot source until "
1776 "handover completes.");
1777 r = -EINVAL;
Mike Snitzerb83b2f22011-01-13 19:59:59 +00001778 } else if (!dm_suspended(snap_src->ti)) {
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001779 DMERR("Unable to perform snapshot handover until "
1780 "source is suspended.");
1781 r = -EINVAL;
1782 }
1783 up_read(&snap_src->lock);
1784 }
1785 up_read(&_origins_lock);
1786
1787 return r;
1788}
1789
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790static void snapshot_resume(struct dm_target *ti)
1791{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001792 struct dm_snapshot *s = ti->private;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001793 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1794
1795 down_read(&_origins_lock);
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +00001796 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001797 if (snap_src && snap_dest) {
1798 down_write(&snap_src->lock);
1799 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1800 __handover_exceptions(snap_src, snap_dest);
1801 up_write(&snap_dest->lock);
1802 up_write(&snap_src->lock);
1803 }
1804 up_read(&_origins_lock);
1805
1806 /* Now we have correct chunk size, reregister */
1807 reregister_snapshot(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001809 down_write(&s->lock);
1810 s->active = 1;
1811 up_write(&s->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812}
1813
Mike Snitzer542f9032012-07-27 15:08:00 +01001814static uint32_t get_origin_minimum_chunksize(struct block_device *bdev)
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001815{
Mike Snitzer542f9032012-07-27 15:08:00 +01001816 uint32_t min_chunksize;
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001817
1818 down_read(&_origins_lock);
1819 min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
1820 up_read(&_origins_lock);
1821
1822 return min_chunksize;
1823}
1824
1825static void snapshot_merge_resume(struct dm_target *ti)
1826{
1827 struct dm_snapshot *s = ti->private;
1828
1829 /*
1830 * Handover exceptions from existing snapshot.
1831 */
1832 snapshot_resume(ti);
1833
1834 /*
Mike Snitzer542f9032012-07-27 15:08:00 +01001835 * snapshot-merge acts as an origin, so set ti->max_io_len
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001836 */
Mike Snitzer542f9032012-07-27 15:08:00 +01001837 ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev);
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001838
1839 start_merge(s);
1840}
1841
Mikulas Patockafd7c0922013-03-01 22:45:44 +00001842static void snapshot_status(struct dm_target *ti, status_type_t type,
1843 unsigned status_flags, char *result, unsigned maxlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844{
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01001845 unsigned sz = 0;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001846 struct dm_snapshot *snap = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847
1848 switch (type) {
1849 case STATUSTYPE_INFO:
Mikulas Patocka94e765722009-12-10 23:51:53 +00001850
1851 down_write(&snap->lock);
1852
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 if (!snap->valid)
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01001854 DMEMIT("Invalid");
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +00001855 else if (snap->merge_failed)
1856 DMEMIT("Merge failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 else {
Mike Snitzer985903b2009-12-10 23:52:11 +00001858 if (snap->store->type->usage) {
1859 sector_t total_sectors, sectors_allocated,
1860 metadata_sectors;
1861 snap->store->type->usage(snap->store,
1862 &total_sectors,
1863 &sectors_allocated,
1864 &metadata_sectors);
1865 DMEMIT("%llu/%llu %llu",
1866 (unsigned long long)sectors_allocated,
1867 (unsigned long long)total_sectors,
1868 (unsigned long long)metadata_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 }
1870 else
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01001871 DMEMIT("Unknown");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 }
Mikulas Patocka94e765722009-12-10 23:51:53 +00001873
1874 up_write(&snap->lock);
1875
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 break;
1877
1878 case STATUSTYPE_TABLE:
1879 /*
1880 * kdevname returns a static pointer so we need
1881 * to make private copies if the output is to
1882 * make sense.
1883 */
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001884 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
Jonathan Brassow1e302a92009-04-02 19:55:35 +01001885 snap->store->type->status(snap->store, type, result + sz,
1886 maxlen - sz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 break;
1888 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889}
1890
Mike Snitzer8811f462009-09-04 20:40:19 +01001891static int snapshot_iterate_devices(struct dm_target *ti,
1892 iterate_devices_callout_fn fn, void *data)
1893{
1894 struct dm_snapshot *snap = ti->private;
Mikulas Patocka1e5554c2010-08-12 04:13:50 +01001895 int r;
Mike Snitzer8811f462009-09-04 20:40:19 +01001896
Mikulas Patocka1e5554c2010-08-12 04:13:50 +01001897 r = fn(ti, snap->origin, 0, ti->len, data);
1898
1899 if (!r)
1900 r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data);
1901
1902 return r;
Mike Snitzer8811f462009-09-04 20:40:19 +01001903}
1904
1905
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906/*-----------------------------------------------------------------
1907 * Origin methods
1908 *---------------------------------------------------------------*/
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001909
1910/*
1911 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1912 * supplied bio was ignored. The caller may submit it immediately.
1913 * (No remapping actually occurs as the origin is always a direct linear
1914 * map.)
1915 *
1916 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1917 * and any supplied bio is added to a list to be submitted once all
1918 * the necessary exceptions exist.
1919 */
1920static int __origin_write(struct list_head *snapshots, sector_t sector,
1921 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922{
Mikulas Patocka515ad662009-12-10 23:52:30 +00001923 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 struct dm_snapshot *snap;
Jon Brassow1d4989c2009-12-10 23:52:10 +00001925 struct dm_exception *e;
Mikulas Patocka515ad662009-12-10 23:52:30 +00001926 struct dm_snap_pending_exception *pe;
1927 struct dm_snap_pending_exception *pe_to_start_now = NULL;
1928 struct dm_snap_pending_exception *pe_to_start_last = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 chunk_t chunk;
1930
1931 /* Do all the snapshots on this origin */
1932 list_for_each_entry (snap, snapshots, list) {
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001933 /*
1934 * Don't make new exceptions in a merging snapshot
1935 * because it has effectively been deleted
1936 */
1937 if (dm_target_is_snapshot_merge(snap->ti))
1938 continue;
1939
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001940 down_write(&snap->lock);
1941
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001942 /* Only deal with valid and active snapshots */
1943 if (!snap->valid || !snap->active)
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001944 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945
Alasdair G Kergond5e404c2005-07-12 15:53:05 -07001946 /* Nothing to do if writing beyond end of snapshot */
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001947 if (sector >= dm_table_get_size(snap->ti->table))
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001948 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949
1950 /*
1951 * Remember, different snapshots can have
1952 * different chunk sizes.
1953 */
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001954 chunk = sector_to_chunk(snap->store, sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955
1956 /*
1957 * Check exception table to see if block
1958 * is already remapped in this snapshot
1959 * and trigger an exception if not.
1960 */
Jon Brassow3510cb92009-12-10 23:52:11 +00001961 e = dm_lookup_exception(&snap->complete, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001962 if (e)
1963 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964
Mikulas Patocka29138082009-04-02 19:55:25 +01001965 pe = __lookup_pending_exception(snap, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001966 if (!pe) {
Mikulas Patockac6621392009-04-02 19:55:25 +01001967 up_write(&snap->lock);
1968 pe = alloc_pending_exception(snap);
1969 down_write(&snap->lock);
1970
1971 if (!snap->valid) {
1972 free_pending_exception(pe);
1973 goto next_snapshot;
1974 }
1975
Jon Brassow3510cb92009-12-10 23:52:11 +00001976 e = dm_lookup_exception(&snap->complete, chunk);
Mikulas Patocka35bf6592009-04-02 19:55:26 +01001977 if (e) {
1978 free_pending_exception(pe);
1979 goto next_snapshot;
1980 }
1981
Mikulas Patockac6621392009-04-02 19:55:25 +01001982 pe = __find_pending_exception(snap, pe, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01001983 if (!pe) {
1984 __invalidate_snapshot(snap, -ENOMEM);
1985 goto next_snapshot;
1986 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 }
1988
Mikulas Patocka515ad662009-12-10 23:52:30 +00001989 r = DM_MAPIO_SUBMITTED;
1990
1991 /*
1992 * If an origin bio was supplied, queue it to wait for the
1993 * completion of this exception, and start this one last,
1994 * at the end of the function.
1995 */
1996 if (bio) {
1997 bio_list_add(&pe->origin_bios, bio);
1998 bio = NULL;
1999
2000 if (!pe->started) {
2001 pe->started = 1;
2002 pe_to_start_last = pe;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002003 }
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002004 }
2005
2006 if (!pe->started) {
2007 pe->started = 1;
Mikulas Patocka515ad662009-12-10 23:52:30 +00002008 pe_to_start_now = pe;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002009 }
2010
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01002011next_snapshot:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 up_write(&snap->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013
Mikulas Patocka515ad662009-12-10 23:52:30 +00002014 if (pe_to_start_now) {
2015 start_copy(pe_to_start_now);
2016 pe_to_start_now = NULL;
2017 }
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08002018 }
2019
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 /*
Mikulas Patocka515ad662009-12-10 23:52:30 +00002021 * Submit the exception against which the bio is queued last,
2022 * to give the other exceptions a head start.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 */
Mikulas Patocka515ad662009-12-10 23:52:30 +00002024 if (pe_to_start_last)
2025 start_copy(pe_to_start_last);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026
2027 return r;
2028}
2029
2030/*
2031 * Called on a write from the origin driver.
2032 */
2033static int do_origin(struct dm_dev *origin, struct bio *bio)
2034{
2035 struct origin *o;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08002036 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037
2038 down_read(&_origins_lock);
2039 o = __lookup_origin(origin->bdev);
2040 if (o)
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00002041 r = __origin_write(&o->snapshots, bio->bi_sector, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 up_read(&_origins_lock);
2043
2044 return r;
2045}
2046
2047/*
Mikulas Patocka73dfd072009-12-10 23:52:34 +00002048 * Trigger exceptions in all non-merging snapshots.
2049 *
2050 * The chunk size of the merging snapshot may be larger than the chunk
2051 * size of some other snapshot so we may need to reallocate multiple
2052 * chunks in other snapshots.
2053 *
2054 * We scan all the overlapping exceptions in the other snapshots.
2055 * Returns 1 if anything was reallocated and must be waited for,
2056 * otherwise returns 0.
2057 *
2058 * size must be a multiple of merging_snap's chunk_size.
2059 */
2060static int origin_write_extent(struct dm_snapshot *merging_snap,
2061 sector_t sector, unsigned size)
2062{
2063 int must_wait = 0;
2064 sector_t n;
2065 struct origin *o;
2066
2067 /*
Mike Snitzer542f9032012-07-27 15:08:00 +01002068 * The origin's __minimum_chunk_size() got stored in max_io_len
Mikulas Patocka73dfd072009-12-10 23:52:34 +00002069 * by snapshot_merge_resume().
2070 */
2071 down_read(&_origins_lock);
2072 o = __lookup_origin(merging_snap->origin->bdev);
Mike Snitzer542f9032012-07-27 15:08:00 +01002073 for (n = 0; n < size; n += merging_snap->ti->max_io_len)
Mikulas Patocka73dfd072009-12-10 23:52:34 +00002074 if (__origin_write(&o->snapshots, sector + n, NULL) ==
2075 DM_MAPIO_SUBMITTED)
2076 must_wait = 1;
2077 up_read(&_origins_lock);
2078
2079 return must_wait;
2080}
2081
2082/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 * Origin: maps a linear range of a device, with hooks for snapshotting.
2084 */
2085
2086/*
2087 * Construct an origin mapping: <dev_path>
2088 * The context for an origin is merely a 'struct dm_dev *'
2089 * pointing to the real device.
2090 */
2091static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2092{
2093 int r;
2094 struct dm_dev *dev;
2095
2096 if (argc != 1) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07002097 ti->error = "origin: incorrect number of arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 return -EINVAL;
2099 }
2100
Nikanth Karthikesan8215d6e2010-03-06 02:32:27 +00002101 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 if (r) {
2103 ti->error = "Cannot get target device";
2104 return r;
2105 }
2106
2107 ti->private = dev;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002108 ti->num_flush_bios = 1;
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01002109
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 return 0;
2111}
2112
2113static void origin_dtr(struct dm_target *ti)
2114{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002115 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 dm_put_device(ti, dev);
2117}
2118
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00002119static int origin_map(struct dm_target *ti, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002121 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 bio->bi_bdev = dev->bdev;
2123
Tejun Heod87f4c12010-09-03 11:56:19 +02002124 if (bio->bi_rw & REQ_FLUSH)
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01002125 return DM_MAPIO_REMAPPED;
2126
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 /* Only tell snapshots if this is a write */
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08002128 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129}
2130
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131/*
Mike Snitzer542f9032012-07-27 15:08:00 +01002132 * Set the target "max_io_len" field to the minimum of all the snapshots'
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 * chunk sizes.
2134 */
2135static void origin_resume(struct dm_target *ti)
2136{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002137 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138
Mike Snitzer542f9032012-07-27 15:08:00 +01002139 ti->max_io_len = get_origin_minimum_chunksize(dev->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140}
2141
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002142static void origin_status(struct dm_target *ti, status_type_t type,
2143 unsigned status_flags, char *result, unsigned maxlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002145 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146
2147 switch (type) {
2148 case STATUSTYPE_INFO:
2149 result[0] = '\0';
2150 break;
2151
2152 case STATUSTYPE_TABLE:
2153 snprintf(result, maxlen, "%s", dev->name);
2154 break;
2155 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156}
2157
Mikulas Patockab1d55522010-08-12 04:14:02 +01002158static int origin_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2159 struct bio_vec *biovec, int max_size)
2160{
2161 struct dm_dev *dev = ti->private;
2162 struct request_queue *q = bdev_get_queue(dev->bdev);
2163
2164 if (!q->merge_bvec_fn)
2165 return max_size;
2166
2167 bvm->bi_bdev = dev->bdev;
Mikulas Patockab1d55522010-08-12 04:14:02 +01002168
2169 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2170}
2171
Mike Snitzer8811f462009-09-04 20:40:19 +01002172static int origin_iterate_devices(struct dm_target *ti,
2173 iterate_devices_callout_fn fn, void *data)
2174{
2175 struct dm_dev *dev = ti->private;
2176
2177 return fn(ti, dev, 0, ti->len, data);
2178}
2179
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180static struct target_type origin_target = {
2181 .name = "snapshot-origin",
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002182 .version = {1, 8, 1},
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 .module = THIS_MODULE,
2184 .ctr = origin_ctr,
2185 .dtr = origin_dtr,
2186 .map = origin_map,
2187 .resume = origin_resume,
2188 .status = origin_status,
Mikulas Patockab1d55522010-08-12 04:14:02 +01002189 .merge = origin_merge,
Mike Snitzer8811f462009-09-04 20:40:19 +01002190 .iterate_devices = origin_iterate_devices,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191};
2192
2193static struct target_type snapshot_target = {
2194 .name = "snapshot",
Mikulas Patockafd7c0922013-03-01 22:45:44 +00002195 .version = {1, 11, 1},
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 .module = THIS_MODULE,
2197 .ctr = snapshot_ctr,
2198 .dtr = snapshot_dtr,
2199 .map = snapshot_map,
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002200 .end_io = snapshot_end_io,
Mike Snitzerc1f0c182009-12-10 23:52:24 +00002201 .preresume = snapshot_preresume,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 .resume = snapshot_resume,
2203 .status = snapshot_status,
Mike Snitzer8811f462009-09-04 20:40:19 +01002204 .iterate_devices = snapshot_iterate_devices,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205};
2206
Mikulas Patockad698aa42009-12-10 23:52:30 +00002207static struct target_type merge_target = {
2208 .name = dm_snapshot_merge_target_name,
Mikulas Patocka42bc9542012-12-21 20:23:38 +00002209 .version = {1, 2, 0},
Mikulas Patockad698aa42009-12-10 23:52:30 +00002210 .module = THIS_MODULE,
2211 .ctr = snapshot_ctr,
2212 .dtr = snapshot_dtr,
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00002213 .map = snapshot_merge_map,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002214 .end_io = snapshot_end_io,
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002215 .presuspend = snapshot_merge_presuspend,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002216 .preresume = snapshot_preresume,
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002217 .resume = snapshot_merge_resume,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002218 .status = snapshot_status,
2219 .iterate_devices = snapshot_iterate_devices,
2220};
2221
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222static int __init dm_snapshot_init(void)
2223{
2224 int r;
2225
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00002226 r = dm_exception_store_init();
2227 if (r) {
2228 DMERR("Failed to initialize exception stores");
2229 return r;
2230 }
2231
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 r = dm_register_target(&snapshot_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002233 if (r < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 DMERR("snapshot target register failed %d", r);
Jonathan Brassow034a1862009-10-16 23:18:14 +01002235 goto bad_register_snapshot_target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 }
2237
2238 r = dm_register_target(&origin_target);
2239 if (r < 0) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07002240 DMERR("Origin target register failed %d", r);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002241 goto bad_register_origin_target;
2242 }
2243
2244 r = dm_register_target(&merge_target);
2245 if (r < 0) {
2246 DMERR("Merge target register failed %d", r);
2247 goto bad_register_merge_target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 }
2249
2250 r = init_origin_hash();
2251 if (r) {
2252 DMERR("init_origin_hash failed.");
Mikulas Patockad698aa42009-12-10 23:52:30 +00002253 goto bad_origin_hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 }
2255
Jon Brassow1d4989c2009-12-10 23:52:10 +00002256 exception_cache = KMEM_CACHE(dm_exception, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 if (!exception_cache) {
2258 DMERR("Couldn't create exception cache.");
2259 r = -ENOMEM;
Mikulas Patockad698aa42009-12-10 23:52:30 +00002260 goto bad_exception_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 }
2262
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002263 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 if (!pending_cache) {
2265 DMERR("Couldn't create pending cache.");
2266 r = -ENOMEM;
Mikulas Patockad698aa42009-12-10 23:52:30 +00002267 goto bad_pending_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 }
2269
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 return 0;
2271
Mikulas Patockad698aa42009-12-10 23:52:30 +00002272bad_pending_cache:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 kmem_cache_destroy(exception_cache);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002274bad_exception_cache:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 exit_origin_hash();
Mikulas Patockad698aa42009-12-10 23:52:30 +00002276bad_origin_hash:
2277 dm_unregister_target(&merge_target);
2278bad_register_merge_target:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 dm_unregister_target(&origin_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002280bad_register_origin_target:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 dm_unregister_target(&snapshot_target);
Jonathan Brassow034a1862009-10-16 23:18:14 +01002282bad_register_snapshot_target:
2283 dm_exception_store_exit();
Mikulas Patockad698aa42009-12-10 23:52:30 +00002284
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 return r;
2286}
2287
2288static void __exit dm_snapshot_exit(void)
2289{
Mikulas Patocka10d3bd02009-01-06 03:04:58 +00002290 dm_unregister_target(&snapshot_target);
2291 dm_unregister_target(&origin_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002292 dm_unregister_target(&merge_target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293
2294 exit_origin_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 kmem_cache_destroy(pending_cache);
2296 kmem_cache_destroy(exception_cache);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00002297
2298 dm_exception_store_exit();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299}
2300
2301/* Module hooks */
2302module_init(dm_snapshot_init);
2303module_exit(dm_snapshot_exit);
2304
2305MODULE_DESCRIPTION(DM_NAME " snapshot target");
2306MODULE_AUTHOR("Joe Thornber");
2307MODULE_LICENSE("GPL");
Mikulas Patocka23cb2102013-03-01 22:45:47 +00002308MODULE_ALIAS("dm-snapshot-origin");
2309MODULE_ALIAS("dm-snapshot-merge");