blob: e1ecacf2456f4e9bf9ab43385e10f98b1bf18db7 [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 Patockac2411042010-08-12 04:13:51 +0100127struct dm_dev *dm_snap_origin(struct dm_snapshot *s)
128{
129 return s->origin;
130}
131EXPORT_SYMBOL(dm_snap_origin);
132
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000133struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
134{
135 return s->cow;
136}
137EXPORT_SYMBOL(dm_snap_cow);
138
Jonathan Brassowccc45ea2009-04-02 19:55:34 +0100139static sector_t chunk_to_sector(struct dm_exception_store *store,
140 chunk_t chunk)
141{
142 return chunk << store->chunk_shift;
143}
144
145static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
146{
147 /*
148 * There is only ever one instance of a particular block
149 * device so we can compare pointers safely.
150 */
151 return lhs == rhs;
152}
153
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100154struct dm_snap_pending_exception {
Jon Brassow1d4989c2009-12-10 23:52:10 +0000155 struct dm_exception e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
157 /*
158 * Origin buffers waiting for this to complete are held
159 * in a bio list
160 */
161 struct bio_list origin_bios;
162 struct bio_list snapshot_bios;
163
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 /* Pointer back to snapshot context */
165 struct dm_snapshot *snap;
166
167 /*
168 * 1 indicates the exception has already been sent to
169 * kcopyd.
170 */
171 int started;
Mikulas Patockaa6e50b42011-08-02 12:32:04 +0100172
173 /*
174 * For writing a complete chunk, bypassing the copy.
175 */
176 struct bio *full_bio;
177 bio_end_io_t *full_bio_end_io;
178 void *full_bio_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179};
180
181/*
182 * Hash table mapping origin volumes to lists of snapshots and
183 * a lock to protect it
184 */
Christoph Lametere18b8902006-12-06 20:33:20 -0800185static struct kmem_cache *exception_cache;
186static struct kmem_cache *pending_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100188struct dm_snap_tracked_chunk {
189 struct hlist_node node;
190 chunk_t chunk;
191};
192
Mikulas Patockaee180262012-12-21 20:23:41 +0000193static void init_tracked_chunk(struct bio *bio)
194{
195 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
196 INIT_HLIST_NODE(&c->node);
197}
198
199static bool is_bio_tracked(struct bio *bio)
200{
201 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
202 return !hlist_unhashed(&c->node);
203}
204
205static void track_chunk(struct dm_snapshot *s, struct bio *bio, chunk_t chunk)
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100206{
Mikulas Patocka42bc9542012-12-21 20:23:38 +0000207 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100208
209 c->chunk = chunk;
210
Mikulas Patocka9aa0c0e2012-12-21 20:23:33 +0000211 spin_lock_irq(&s->tracked_chunk_lock);
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100212 hlist_add_head(&c->node,
213 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
Mikulas Patocka9aa0c0e2012-12-21 20:23:33 +0000214 spin_unlock_irq(&s->tracked_chunk_lock);
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100215}
216
Mikulas Patockaee180262012-12-21 20:23:41 +0000217static void stop_tracking_chunk(struct dm_snapshot *s, struct bio *bio)
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100218{
Mikulas Patockaee180262012-12-21 20:23:41 +0000219 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100220 unsigned long flags;
221
222 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
223 hlist_del(&c->node);
224 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100225}
226
Mikulas Patockaa8d41b52008-07-21 12:00:34 +0100227static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
228{
229 struct dm_snap_tracked_chunk *c;
230 struct hlist_node *hn;
231 int found = 0;
232
233 spin_lock_irq(&s->tracked_chunk_lock);
234
235 hlist_for_each_entry(c, hn,
236 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
237 if (c->chunk == chunk) {
238 found = 1;
239 break;
240 }
241 }
242
243 spin_unlock_irq(&s->tracked_chunk_lock);
244
245 return found;
246}
247
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248/*
Mike Snitzer615d1eb2009-12-10 23:52:29 +0000249 * This conflicting I/O is extremely improbable in the caller,
250 * so msleep(1) is sufficient and there is no need for a wait queue.
251 */
252static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
253{
254 while (__chunk_is_tracked(s, chunk))
255 msleep(1);
256}
257
258/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 * One of these per registered origin, held in the snapshot_origins hash
260 */
261struct origin {
262 /* The origin device */
263 struct block_device *bdev;
264
265 struct list_head hash_list;
266
267 /* List of snapshots for this origin */
268 struct list_head snapshots;
269};
270
271/*
272 * Size of the hash table for origin volumes. If we make this
273 * the size of the minors list then it should be nearly perfect
274 */
275#define ORIGIN_HASH_SIZE 256
276#define ORIGIN_MASK 0xFF
277static struct list_head *_origins;
278static struct rw_semaphore _origins_lock;
279
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000280static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
281static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
282static uint64_t _pending_exceptions_done_count;
283
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284static int init_origin_hash(void)
285{
286 int i;
287
288 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
289 GFP_KERNEL);
290 if (!_origins) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700291 DMERR("unable to allocate memory");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 return -ENOMEM;
293 }
294
295 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
296 INIT_LIST_HEAD(_origins + i);
297 init_rwsem(&_origins_lock);
298
299 return 0;
300}
301
302static void exit_origin_hash(void)
303{
304 kfree(_origins);
305}
306
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100307static unsigned origin_hash(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308{
309 return bdev->bd_dev & ORIGIN_MASK;
310}
311
312static struct origin *__lookup_origin(struct block_device *origin)
313{
314 struct list_head *ol;
315 struct origin *o;
316
317 ol = &_origins[origin_hash(origin)];
318 list_for_each_entry (o, ol, hash_list)
319 if (bdev_equal(o->bdev, origin))
320 return o;
321
322 return NULL;
323}
324
325static void __insert_origin(struct origin *o)
326{
327 struct list_head *sl = &_origins[origin_hash(o->bdev)];
328 list_add_tail(&o->hash_list, sl);
329}
330
331/*
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000332 * _origins_lock must be held when calling this function.
333 * Returns number of snapshots registered using the supplied cow device, plus:
334 * snap_src - a snapshot suitable for use as a source of exception handover
335 * snap_dest - a snapshot capable of receiving exception handover.
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000336 * snap_merge - an existing snapshot-merge target linked to the same origin.
337 * There can be at most one snapshot-merge target. The parameter is optional.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000338 *
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000339 * Possible return values and states of snap_src and snap_dest.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000340 * 0: NULL, NULL - first new snapshot
341 * 1: snap_src, NULL - normal snapshot
342 * 2: snap_src, snap_dest - waiting for handover
343 * 2: snap_src, NULL - handed over, waiting for old to be deleted
344 * 1: NULL, snap_dest - source got destroyed without handover
345 */
346static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
347 struct dm_snapshot **snap_src,
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000348 struct dm_snapshot **snap_dest,
349 struct dm_snapshot **snap_merge)
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000350{
351 struct dm_snapshot *s;
352 struct origin *o;
353 int count = 0;
354 int active;
355
356 o = __lookup_origin(snap->origin->bdev);
357 if (!o)
358 goto out;
359
360 list_for_each_entry(s, &o->snapshots, list) {
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000361 if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
362 *snap_merge = s;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000363 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
364 continue;
365
366 down_read(&s->lock);
367 active = s->active;
368 up_read(&s->lock);
369
370 if (active) {
371 if (snap_src)
372 *snap_src = s;
373 } else if (snap_dest)
374 *snap_dest = s;
375
376 count++;
377 }
378
379out:
380 return count;
381}
382
383/*
384 * On success, returns 1 if this snapshot is a handover destination,
385 * otherwise returns 0.
386 */
387static int __validate_exception_handover(struct dm_snapshot *snap)
388{
389 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000390 struct dm_snapshot *snap_merge = NULL;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000391
392 /* Does snapshot need exceptions handed over to it? */
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000393 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
394 &snap_merge) == 2) ||
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000395 snap_dest) {
396 snap->ti->error = "Snapshot cow pairing for exception "
397 "table handover failed";
398 return -EINVAL;
399 }
400
401 /*
402 * If no snap_src was found, snap cannot become a handover
403 * destination.
404 */
405 if (!snap_src)
406 return 0;
407
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000408 /*
409 * Non-snapshot-merge handover?
410 */
411 if (!dm_target_is_snapshot_merge(snap->ti))
412 return 1;
413
414 /*
415 * Do not allow more than one merging snapshot.
416 */
417 if (snap_merge) {
418 snap->ti->error = "A snapshot is already merging.";
419 return -EINVAL;
420 }
421
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000422 if (!snap_src->store->type->prepare_merge ||
423 !snap_src->store->type->commit_merge) {
424 snap->ti->error = "Snapshot exception store does not "
425 "support snapshot-merge.";
426 return -EINVAL;
427 }
428
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000429 return 1;
430}
431
432static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
433{
434 struct dm_snapshot *l;
435
436 /* Sort the list according to chunk size, largest-first smallest-last */
437 list_for_each_entry(l, &o->snapshots, list)
438 if (l->store->chunk_size < s->store->chunk_size)
439 break;
440 list_add_tail(&s->list, &l->list);
441}
442
443/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 * Make a note of the snapshot and its origin so we can look it
445 * up when the origin has a write on it.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000446 *
447 * Also validate snapshot exception store handovers.
448 * On success, returns 1 if this registration is a handover destination,
449 * otherwise returns 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 */
451static int register_snapshot(struct dm_snapshot *snap)
452{
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000453 struct origin *o, *new_o = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 struct block_device *bdev = snap->origin->bdev;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000455 int r = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000457 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
458 if (!new_o)
459 return -ENOMEM;
460
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 down_write(&_origins_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000463 r = __validate_exception_handover(snap);
464 if (r < 0) {
465 kfree(new_o);
466 goto out;
467 }
468
469 o = __lookup_origin(bdev);
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000470 if (o)
471 kfree(new_o);
472 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 /* New origin */
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000474 o = new_o;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
476 /* Initialise the struct */
477 INIT_LIST_HEAD(&o->snapshots);
478 o->bdev = bdev;
479
480 __insert_origin(o);
481 }
482
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000483 __insert_snapshot(o, snap);
484
485out:
486 up_write(&_origins_lock);
487
488 return r;
489}
490
491/*
492 * Move snapshot to correct place in list according to chunk size.
493 */
494static void reregister_snapshot(struct dm_snapshot *s)
495{
496 struct block_device *bdev = s->origin->bdev;
497
498 down_write(&_origins_lock);
499
500 list_del(&s->list);
501 __insert_snapshot(__lookup_origin(bdev), s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
503 up_write(&_origins_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504}
505
506static void unregister_snapshot(struct dm_snapshot *s)
507{
508 struct origin *o;
509
510 down_write(&_origins_lock);
511 o = __lookup_origin(s->origin->bdev);
512
513 list_del(&s->list);
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000514 if (o && list_empty(&o->snapshots)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 list_del(&o->hash_list);
516 kfree(o);
517 }
518
519 up_write(&_origins_lock);
520}
521
522/*
523 * Implementation of the exception hash tables.
Milan Brozd74f81f2008-02-08 02:11:27 +0000524 * The lowest hash_shift bits of the chunk number are ignored, allowing
525 * some consecutive chunks to be grouped together.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 */
Jon Brassow3510cb92009-12-10 23:52:11 +0000527static int dm_exception_table_init(struct dm_exception_table *et,
528 uint32_t size, unsigned hash_shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529{
530 unsigned int i;
531
Milan Brozd74f81f2008-02-08 02:11:27 +0000532 et->hash_shift = hash_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 et->hash_mask = size - 1;
534 et->table = dm_vcalloc(size, sizeof(struct list_head));
535 if (!et->table)
536 return -ENOMEM;
537
538 for (i = 0; i < size; i++)
539 INIT_LIST_HEAD(et->table + i);
540
541 return 0;
542}
543
Jon Brassow3510cb92009-12-10 23:52:11 +0000544static void dm_exception_table_exit(struct dm_exception_table *et,
545 struct kmem_cache *mem)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546{
547 struct list_head *slot;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000548 struct dm_exception *ex, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 int i, size;
550
551 size = et->hash_mask + 1;
552 for (i = 0; i < size; i++) {
553 slot = et->table + i;
554
555 list_for_each_entry_safe (ex, next, slot, hash_list)
556 kmem_cache_free(mem, ex);
557 }
558
559 vfree(et->table);
560}
561
Jon Brassow191437a2009-12-10 23:52:10 +0000562static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563{
Milan Brozd74f81f2008-02-08 02:11:27 +0000564 return (chunk >> et->hash_shift) & et->hash_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565}
566
Jon Brassow3510cb92009-12-10 23:52:11 +0000567static void dm_remove_exception(struct dm_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
569 list_del(&e->hash_list);
570}
571
572/*
573 * Return the exception data for a sector, or NULL if not
574 * remapped.
575 */
Jon Brassow3510cb92009-12-10 23:52:11 +0000576static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
577 chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578{
579 struct list_head *slot;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000580 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581
582 slot = &et->table[exception_hash(et, chunk)];
583 list_for_each_entry (e, slot, hash_list)
Milan Brozd74f81f2008-02-08 02:11:27 +0000584 if (chunk >= e->old_chunk &&
585 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 return e;
587
588 return NULL;
589}
590
Jon Brassow3510cb92009-12-10 23:52:11 +0000591static struct dm_exception *alloc_completed_exception(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592{
Jon Brassow1d4989c2009-12-10 23:52:10 +0000593 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
595 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
596 if (!e)
597 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
598
599 return e;
600}
601
Jon Brassow3510cb92009-12-10 23:52:11 +0000602static void free_completed_exception(struct dm_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603{
604 kmem_cache_free(exception_cache, e);
605}
606
Mikulas Patocka92e86812008-07-21 12:00:35 +0100607static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608{
Mikulas Patocka92e86812008-07-21 12:00:35 +0100609 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
610 GFP_NOIO);
611
Mikulas Patocka879129d22008-10-30 13:33:16 +0000612 atomic_inc(&s->pending_exceptions_count);
Mikulas Patocka92e86812008-07-21 12:00:35 +0100613 pe->snap = s;
614
615 return pe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616}
617
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100618static void free_pending_exception(struct dm_snap_pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619{
Mikulas Patocka879129d22008-10-30 13:33:16 +0000620 struct dm_snapshot *s = pe->snap;
621
622 mempool_free(pe, s->pending_pool);
623 smp_mb__before_atomic_dec();
624 atomic_dec(&s->pending_exceptions_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625}
626
Jon Brassow3510cb92009-12-10 23:52:11 +0000627static void dm_insert_exception(struct dm_exception_table *eh,
628 struct dm_exception *new_e)
Milan Brozd74f81f2008-02-08 02:11:27 +0000629{
Milan Brozd74f81f2008-02-08 02:11:27 +0000630 struct list_head *l;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000631 struct dm_exception *e = NULL;
Milan Brozd74f81f2008-02-08 02:11:27 +0000632
633 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
634
635 /* Add immediately if this table doesn't support consecutive chunks */
636 if (!eh->hash_shift)
637 goto out;
638
639 /* List is ordered by old_chunk */
640 list_for_each_entry_reverse(e, l, hash_list) {
641 /* Insert after an existing chunk? */
642 if (new_e->old_chunk == (e->old_chunk +
643 dm_consecutive_chunk_count(e) + 1) &&
644 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
645 dm_consecutive_chunk_count(e) + 1)) {
646 dm_consecutive_chunk_count_inc(e);
Jon Brassow3510cb92009-12-10 23:52:11 +0000647 free_completed_exception(new_e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000648 return;
649 }
650
651 /* Insert before an existing chunk? */
652 if (new_e->old_chunk == (e->old_chunk - 1) &&
653 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
654 dm_consecutive_chunk_count_inc(e);
655 e->old_chunk--;
656 e->new_chunk--;
Jon Brassow3510cb92009-12-10 23:52:11 +0000657 free_completed_exception(new_e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000658 return;
659 }
660
661 if (new_e->old_chunk > e->old_chunk)
662 break;
663 }
664
665out:
666 list_add(&new_e->hash_list, e ? &e->hash_list : l);
667}
668
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000669/*
670 * Callback used by the exception stores to load exceptions when
671 * initialising.
672 */
673static int dm_add_exception(void *context, chunk_t old, chunk_t new)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674{
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000675 struct dm_snapshot *s = context;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000676 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
Jon Brassow3510cb92009-12-10 23:52:11 +0000678 e = alloc_completed_exception();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 if (!e)
680 return -ENOMEM;
681
682 e->old_chunk = old;
Milan Brozd74f81f2008-02-08 02:11:27 +0000683
684 /* Consecutive_count is implicitly initialised to zero */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 e->new_chunk = new;
Milan Brozd74f81f2008-02-08 02:11:27 +0000686
Jon Brassow3510cb92009-12-10 23:52:11 +0000687 dm_insert_exception(&s->complete, e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000688
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 return 0;
690}
691
Mikulas Patocka7e201b32009-12-10 23:52:08 +0000692/*
693 * Return a minimum chunk size of all snapshots that have the specified origin.
694 * Return zero if the origin has no snapshots.
695 */
Mike Snitzer542f9032012-07-27 15:08:00 +0100696static uint32_t __minimum_chunk_size(struct origin *o)
Mikulas Patocka7e201b32009-12-10 23:52:08 +0000697{
698 struct dm_snapshot *snap;
699 unsigned chunk_size = 0;
700
701 if (o)
702 list_for_each_entry(snap, &o->snapshots, list)
703 chunk_size = min_not_zero(chunk_size,
704 snap->store->chunk_size);
705
Mike Snitzer542f9032012-07-27 15:08:00 +0100706 return (uint32_t) chunk_size;
Mikulas Patocka7e201b32009-12-10 23:52:08 +0000707}
708
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709/*
710 * Hard coded magic.
711 */
712static int calc_max_buckets(void)
713{
714 /* use a fixed size of 2MB */
715 unsigned long mem = 2 * 1024 * 1024;
716 mem /= sizeof(struct list_head);
717
718 return mem;
719}
720
721/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 * Allocate room for a suitable hash table.
723 */
Jonathan Brassowfee19982009-04-02 19:55:34 +0100724static int init_hash_tables(struct dm_snapshot *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725{
726 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
727
728 /*
729 * Calculate based on the size of the original volume or
730 * the COW volume...
731 */
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000732 cow_dev_size = get_dev_size(s->cow->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 origin_dev_size = get_dev_size(s->origin->bdev);
734 max_buckets = calc_max_buckets();
735
Jonathan Brassowfee19982009-04-02 19:55:34 +0100736 hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 hash_size = min(hash_size, max_buckets);
738
Mikulas Patocka8e87b9b2009-12-10 23:51:54 +0000739 if (hash_size < 64)
740 hash_size = 64;
Robert P. J. Day8defd832008-02-08 02:10:06 +0000741 hash_size = rounddown_pow_of_two(hash_size);
Jon Brassow3510cb92009-12-10 23:52:11 +0000742 if (dm_exception_table_init(&s->complete, hash_size,
743 DM_CHUNK_CONSECUTIVE_BITS))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 return -ENOMEM;
745
746 /*
747 * Allocate hash table for in-flight exceptions
748 * Make this smaller than the real hash table
749 */
750 hash_size >>= 3;
751 if (hash_size < 64)
752 hash_size = 64;
753
Jon Brassow3510cb92009-12-10 23:52:11 +0000754 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
755 dm_exception_table_exit(&s->complete, exception_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 return -ENOMEM;
757 }
758
759 return 0;
760}
761
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000762static void merge_shutdown(struct dm_snapshot *s)
763{
764 clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
765 smp_mb__after_clear_bit();
766 wake_up_bit(&s->state_bits, RUNNING_MERGE);
767}
768
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000769static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
770{
771 s->first_merging_chunk = 0;
772 s->num_merging_chunks = 0;
773
774 return bio_list_get(&s->bios_queued_during_merge);
775}
776
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000777/*
778 * Remove one chunk from the index of completed exceptions.
779 */
780static int __remove_single_exception_chunk(struct dm_snapshot *s,
781 chunk_t old_chunk)
782{
783 struct dm_exception *e;
784
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000785 e = dm_lookup_exception(&s->complete, old_chunk);
786 if (!e) {
787 DMERR("Corruption detected: exception for block %llu is "
788 "on disk but not in memory",
789 (unsigned long long)old_chunk);
790 return -EINVAL;
791 }
792
793 /*
794 * If this is the only chunk using this exception, remove exception.
795 */
796 if (!dm_consecutive_chunk_count(e)) {
797 dm_remove_exception(e);
798 free_completed_exception(e);
799 return 0;
800 }
801
802 /*
803 * The chunk may be either at the beginning or the end of a
804 * group of consecutive chunks - never in the middle. We are
805 * removing chunks in the opposite order to that in which they
806 * were added, so this should always be true.
807 * Decrement the consecutive chunk counter and adjust the
808 * starting point if necessary.
809 */
810 if (old_chunk == e->old_chunk) {
811 e->old_chunk++;
812 e->new_chunk++;
813 } else if (old_chunk != e->old_chunk +
814 dm_consecutive_chunk_count(e)) {
815 DMERR("Attempt to merge block %llu from the "
816 "middle of a chunk range [%llu - %llu]",
817 (unsigned long long)old_chunk,
818 (unsigned long long)e->old_chunk,
819 (unsigned long long)
820 e->old_chunk + dm_consecutive_chunk_count(e));
821 return -EINVAL;
822 }
823
824 dm_consecutive_chunk_count_dec(e);
825
826 return 0;
827}
828
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000829static void flush_bios(struct bio *bio);
830
831static int remove_single_exception_chunk(struct dm_snapshot *s)
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000832{
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000833 struct bio *b = NULL;
834 int r;
835 chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000836
837 down_write(&s->lock);
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000838
839 /*
840 * Process chunks (and associated exceptions) in reverse order
841 * so that dm_consecutive_chunk_count_dec() accounting works.
842 */
843 do {
844 r = __remove_single_exception_chunk(s, old_chunk);
845 if (r)
846 goto out;
847 } while (old_chunk-- > s->first_merging_chunk);
848
849 b = __release_queued_bios_after_merge(s);
850
851out:
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000852 up_write(&s->lock);
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000853 if (b)
854 flush_bios(b);
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000855
856 return r;
857}
858
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000859static int origin_write_extent(struct dm_snapshot *merging_snap,
860 sector_t sector, unsigned chunk_size);
861
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000862static void merge_callback(int read_err, unsigned long write_err,
863 void *context);
864
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000865static uint64_t read_pending_exceptions_done_count(void)
866{
867 uint64_t pending_exceptions_done;
868
869 spin_lock(&_pending_exceptions_done_spinlock);
870 pending_exceptions_done = _pending_exceptions_done_count;
871 spin_unlock(&_pending_exceptions_done_spinlock);
872
873 return pending_exceptions_done;
874}
875
876static void increment_pending_exceptions_done_count(void)
877{
878 spin_lock(&_pending_exceptions_done_spinlock);
879 _pending_exceptions_done_count++;
880 spin_unlock(&_pending_exceptions_done_spinlock);
881
882 wake_up_all(&_pending_exceptions_done);
883}
884
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000885static void snapshot_merge_next_chunks(struct dm_snapshot *s)
886{
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000887 int i, linear_chunks;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000888 chunk_t old_chunk, new_chunk;
889 struct dm_io_region src, dest;
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000890 sector_t io_size;
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000891 uint64_t previous_count;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000892
893 BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
894 if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
895 goto shut;
896
897 /*
898 * valid flag never changes during merge, so no lock required.
899 */
900 if (!s->valid) {
901 DMERR("Snapshot is invalid: can't merge");
902 goto shut;
903 }
904
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000905 linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
906 &new_chunk);
907 if (linear_chunks <= 0) {
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +0000908 if (linear_chunks < 0) {
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000909 DMERR("Read error in exception store: "
910 "shutting down merge");
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +0000911 down_write(&s->lock);
912 s->merge_failed = 1;
913 up_write(&s->lock);
914 }
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000915 goto shut;
916 }
917
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000918 /* Adjust old_chunk and new_chunk to reflect start of linear region */
919 old_chunk = old_chunk + 1 - linear_chunks;
920 new_chunk = new_chunk + 1 - linear_chunks;
921
922 /*
923 * Use one (potentially large) I/O to copy all 'linear_chunks'
924 * from the exception store to the origin
925 */
926 io_size = linear_chunks * s->store->chunk_size;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000927
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000928 dest.bdev = s->origin->bdev;
929 dest.sector = chunk_to_sector(s->store, old_chunk);
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000930 dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000931
932 src.bdev = s->cow->bdev;
933 src.sector = chunk_to_sector(s->store, new_chunk);
934 src.count = dest.count;
935
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000936 /*
937 * Reallocate any exceptions needed in other snapshots then
938 * wait for the pending exceptions to complete.
939 * Each time any pending exception (globally on the system)
940 * completes we are woken and repeat the process to find out
941 * if we can proceed. While this may not seem a particularly
942 * efficient algorithm, it is not expected to have any
943 * significant impact on performance.
944 */
945 previous_count = read_pending_exceptions_done_count();
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000946 while (origin_write_extent(s, dest.sector, io_size)) {
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000947 wait_event(_pending_exceptions_done,
948 (read_pending_exceptions_done_count() !=
949 previous_count));
950 /* Retry after the wait, until all exceptions are done. */
951 previous_count = read_pending_exceptions_done_count();
952 }
953
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000954 down_write(&s->lock);
955 s->first_merging_chunk = old_chunk;
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000956 s->num_merging_chunks = linear_chunks;
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000957 up_write(&s->lock);
958
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000959 /* Wait until writes to all 'linear_chunks' drain */
960 for (i = 0; i < linear_chunks; i++)
961 __check_for_conflicting_io(s, old_chunk + i);
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000962
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000963 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
964 return;
965
966shut:
967 merge_shutdown(s);
968}
969
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000970static void error_bios(struct bio *bio);
971
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000972static void merge_callback(int read_err, unsigned long write_err, void *context)
973{
974 struct dm_snapshot *s = context;
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000975 struct bio *b = NULL;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000976
977 if (read_err || write_err) {
978 if (read_err)
979 DMERR("Read error: shutting down merge.");
980 else
981 DMERR("Write error: shutting down merge.");
982 goto shut;
983 }
984
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000985 if (s->store->type->commit_merge(s->store,
986 s->num_merging_chunks) < 0) {
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000987 DMERR("Write error in exception store: shutting down merge");
988 goto shut;
989 }
990
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000991 if (remove_single_exception_chunk(s) < 0)
992 goto shut;
993
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000994 snapshot_merge_next_chunks(s);
995
996 return;
997
998shut:
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000999 down_write(&s->lock);
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +00001000 s->merge_failed = 1;
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001001 b = __release_queued_bios_after_merge(s);
1002 up_write(&s->lock);
1003 error_bios(b);
1004
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001005 merge_shutdown(s);
1006}
1007
1008static void start_merge(struct dm_snapshot *s)
1009{
1010 if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
1011 snapshot_merge_next_chunks(s);
1012}
1013
1014static int wait_schedule(void *ptr)
1015{
1016 schedule();
1017
1018 return 0;
1019}
1020
1021/*
1022 * Stop the merging process and wait until it finishes.
1023 */
1024static void stop_merge(struct dm_snapshot *s)
1025{
1026 set_bit(SHUTDOWN_MERGE, &s->state_bits);
1027 wait_on_bit(&s->state_bits, RUNNING_MERGE, wait_schedule,
1028 TASK_UNINTERRUPTIBLE);
1029 clear_bit(SHUTDOWN_MERGE, &s->state_bits);
1030}
1031
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
1034 */
1035static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1036{
1037 struct dm_snapshot *s;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001038 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 int r = -EINVAL;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001040 char *origin_path, *cow_path;
Mike Snitzer10b81062009-12-10 23:52:31 +00001041 unsigned args_used, num_flush_requests = 1;
1042 fmode_t origin_mode = FMODE_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -07001044 if (argc != 4) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001045 ti->error = "requires exactly 4 arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 r = -EINVAL;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001047 goto bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 }
1049
Mike Snitzer10b81062009-12-10 23:52:31 +00001050 if (dm_target_is_snapshot_merge(ti)) {
1051 num_flush_requests = 2;
1052 origin_mode = FMODE_WRITE;
1053 }
1054
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 s = kmalloc(sizeof(*s), GFP_KERNEL);
Jonathan Brassowfee19982009-04-02 19:55:34 +01001056 if (!s) {
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001057 ti->error = "Cannot allocate private snapshot structure";
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 r = -ENOMEM;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001059 goto bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 }
1061
Mikulas Patockac2411042010-08-12 04:13:51 +01001062 origin_path = argv[0];
1063 argv++;
1064 argc--;
1065
1066 r = dm_get_device(ti, origin_path, origin_mode, &s->origin);
1067 if (r) {
1068 ti->error = "Cannot get origin device";
1069 goto bad_origin;
1070 }
1071
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001072 cow_path = argv[0];
1073 argv++;
1074 argc--;
1075
Milan Broz024d37e2011-03-24 13:52:14 +00001076 r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow);
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001077 if (r) {
1078 ti->error = "Cannot get COW device";
1079 goto bad_cow;
1080 }
1081
1082 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
1083 if (r) {
1084 ti->error = "Couldn't create exception store";
1085 r = -EINVAL;
1086 goto bad_store;
1087 }
1088
1089 argv += args_used;
1090 argc -= args_used;
1091
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001092 s->ti = ti;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 s->valid = 1;
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001094 s->active = 0;
Mikulas Patocka879129d22008-10-30 13:33:16 +00001095 atomic_set(&s->pending_exceptions_count, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 init_rwsem(&s->lock);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001097 INIT_LIST_HEAD(&s->list);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07001098 spin_lock_init(&s->pe_lock);
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001099 s->state_bits = 0;
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +00001100 s->merge_failed = 0;
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001101 s->first_merging_chunk = 0;
1102 s->num_merging_chunks = 0;
1103 bio_list_init(&s->bios_queued_during_merge);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104
1105 /* Allocate hash table for COW data */
Jonathan Brassowfee19982009-04-02 19:55:34 +01001106 if (init_hash_tables(s)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 ti->error = "Unable to allocate hash table space";
1108 r = -ENOMEM;
Jonathan Brassowfee19982009-04-02 19:55:34 +01001109 goto bad_hash_tables;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 }
1111
Mikulas Patockafa34ce72011-05-29 13:03:13 +01001112 s->kcopyd_client = dm_kcopyd_client_create();
1113 if (IS_ERR(s->kcopyd_client)) {
1114 r = PTR_ERR(s->kcopyd_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 ti->error = "Could not create kcopyd client";
Jonathan Brassowfee19982009-04-02 19:55:34 +01001116 goto bad_kcopyd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 }
1118
Mikulas Patocka92e86812008-07-21 12:00:35 +01001119 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
1120 if (!s->pending_pool) {
1121 ti->error = "Could not allocate mempool for pending exceptions";
Jonathan Brassowfee19982009-04-02 19:55:34 +01001122 goto bad_pending_pool;
Mikulas Patocka92e86812008-07-21 12:00:35 +01001123 }
1124
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001125 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1126 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
1127
1128 spin_lock_init(&s->tracked_chunk_lock);
1129
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001130 ti->private = s;
Mike Snitzer10b81062009-12-10 23:52:31 +00001131 ti->num_flush_requests = num_flush_requests;
Mikulas Patocka42bc9542012-12-21 20:23:38 +00001132 ti->per_bio_data_size = sizeof(struct dm_snap_tracked_chunk);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001133
1134 /* Add snapshot to the list of snapshots for this origin */
1135 /* Exceptions aren't triggered till snapshot_resume() is called */
1136 r = register_snapshot(s);
1137 if (r == -ENOMEM) {
1138 ti->error = "Snapshot origin struct allocation failed";
1139 goto bad_load_and_register;
1140 } else if (r < 0) {
1141 /* invalid handover, register_snapshot has set ti->error */
1142 goto bad_load_and_register;
1143 }
1144
1145 /*
1146 * Metadata must only be loaded into one table at once, so skip this
1147 * if metadata will be handed over during resume.
1148 * Chunk size will be set during the handover - set it to zero to
1149 * ensure it's ignored.
1150 */
1151 if (r > 0) {
1152 s->store->chunk_size = 0;
1153 return 0;
1154 }
1155
Jonathan Brassow493df712009-04-02 19:55:31 +01001156 r = s->store->type->read_metadata(s->store, dm_add_exception,
1157 (void *)s);
Milan Broz07641472007-07-12 17:28:13 +01001158 if (r < 0) {
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -07001159 ti->error = "Failed to read snapshot metadata";
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001160 goto bad_read_metadata;
Milan Broz07641472007-07-12 17:28:13 +01001161 } else if (r > 0) {
1162 s->valid = 0;
1163 DMWARN("Snapshot is marked invalid.");
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -07001164 }
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001165
Mikulas Patocka3f2412d2009-10-16 23:18:16 +01001166 if (!s->store->chunk_size) {
1167 ti->error = "Chunk size not set";
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001168 goto bad_read_metadata;
Mikulas Patocka3f2412d2009-10-16 23:18:16 +01001169 }
Mike Snitzer542f9032012-07-27 15:08:00 +01001170
1171 r = dm_set_target_max_io_len(ti, s->store->chunk_size);
1172 if (r)
1173 goto bad_read_metadata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174
1175 return 0;
1176
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001177bad_read_metadata:
1178 unregister_snapshot(s);
1179
Jonathan Brassowfee19982009-04-02 19:55:34 +01001180bad_load_and_register:
Mikulas Patocka92e86812008-07-21 12:00:35 +01001181 mempool_destroy(s->pending_pool);
1182
Jonathan Brassowfee19982009-04-02 19:55:34 +01001183bad_pending_pool:
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +01001184 dm_kcopyd_client_destroy(s->kcopyd_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
Jonathan Brassowfee19982009-04-02 19:55:34 +01001186bad_kcopyd:
Jon Brassow3510cb92009-12-10 23:52:11 +00001187 dm_exception_table_exit(&s->pending, pending_cache);
1188 dm_exception_table_exit(&s->complete, exception_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189
Jonathan Brassowfee19982009-04-02 19:55:34 +01001190bad_hash_tables:
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001191 dm_exception_store_destroy(s->store);
1192
1193bad_store:
1194 dm_put_device(ti, s->cow);
1195
1196bad_cow:
Mikulas Patockac2411042010-08-12 04:13:51 +01001197 dm_put_device(ti, s->origin);
1198
1199bad_origin:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 kfree(s);
1201
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001202bad:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 return r;
1204}
1205
Milan Broz31c93a02006-12-08 02:41:11 -08001206static void __free_exceptions(struct dm_snapshot *s)
1207{
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +01001208 dm_kcopyd_client_destroy(s->kcopyd_client);
Milan Broz31c93a02006-12-08 02:41:11 -08001209 s->kcopyd_client = NULL;
1210
Jon Brassow3510cb92009-12-10 23:52:11 +00001211 dm_exception_table_exit(&s->pending, pending_cache);
1212 dm_exception_table_exit(&s->complete, exception_cache);
Milan Broz31c93a02006-12-08 02:41:11 -08001213}
1214
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001215static void __handover_exceptions(struct dm_snapshot *snap_src,
1216 struct dm_snapshot *snap_dest)
1217{
1218 union {
1219 struct dm_exception_table table_swap;
1220 struct dm_exception_store *store_swap;
1221 } u;
1222
1223 /*
1224 * Swap all snapshot context information between the two instances.
1225 */
1226 u.table_swap = snap_dest->complete;
1227 snap_dest->complete = snap_src->complete;
1228 snap_src->complete = u.table_swap;
1229
1230 u.store_swap = snap_dest->store;
1231 snap_dest->store = snap_src->store;
1232 snap_src->store = u.store_swap;
1233
1234 snap_dest->store->snap = snap_dest;
1235 snap_src->store->snap = snap_src;
1236
Mike Snitzer542f9032012-07-27 15:08:00 +01001237 snap_dest->ti->max_io_len = snap_dest->store->chunk_size;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001238 snap_dest->valid = snap_src->valid;
1239
1240 /*
1241 * Set source invalid to ensure it receives no further I/O.
1242 */
1243 snap_src->valid = 0;
1244}
1245
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246static void snapshot_dtr(struct dm_target *ti)
1247{
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001248#ifdef CONFIG_DM_DEBUG
1249 int i;
1250#endif
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001251 struct dm_snapshot *s = ti->private;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001252 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001254 down_read(&_origins_lock);
1255 /* Check whether exception handover must be cancelled */
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +00001256 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001257 if (snap_src && snap_dest && (s == snap_src)) {
1258 down_write(&snap_dest->lock);
1259 snap_dest->valid = 0;
1260 up_write(&snap_dest->lock);
1261 DMERR("Cancelling snapshot handover.");
1262 }
1263 up_read(&_origins_lock);
1264
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001265 if (dm_target_is_snapshot_merge(ti))
1266 stop_merge(s);
1267
Alasdair G Kergon138728dc2006-03-27 01:17:50 -08001268 /* Prevent further origin writes from using this snapshot. */
1269 /* After this returns there can be no new kcopyd jobs. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 unregister_snapshot(s);
1271
Mikulas Patocka879129d22008-10-30 13:33:16 +00001272 while (atomic_read(&s->pending_exceptions_count))
Mikulas Patocka90fa1522009-01-06 03:04:54 +00001273 msleep(1);
Mikulas Patocka879129d22008-10-30 13:33:16 +00001274 /*
1275 * Ensure instructions in mempool_destroy aren't reordered
1276 * before atomic_read.
1277 */
1278 smp_mb();
1279
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001280#ifdef CONFIG_DM_DEBUG
1281 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1282 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
1283#endif
1284
Milan Broz31c93a02006-12-08 02:41:11 -08001285 __free_exceptions(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286
Mikulas Patocka92e86812008-07-21 12:00:35 +01001287 mempool_destroy(s->pending_pool);
1288
Jonathan Brassowfee19982009-04-02 19:55:34 +01001289 dm_exception_store_destroy(s->store);
Alasdair G Kergon138728dc2006-03-27 01:17:50 -08001290
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001291 dm_put_device(ti, s->cow);
1292
Mikulas Patockac2411042010-08-12 04:13:51 +01001293 dm_put_device(ti, s->origin);
1294
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 kfree(s);
1296}
1297
1298/*
1299 * Flush a list of buffers.
1300 */
1301static void flush_bios(struct bio *bio)
1302{
1303 struct bio *n;
1304
1305 while (bio) {
1306 n = bio->bi_next;
1307 bio->bi_next = NULL;
1308 generic_make_request(bio);
1309 bio = n;
1310 }
1311}
1312
Mikulas Patocka515ad662009-12-10 23:52:30 +00001313static int do_origin(struct dm_dev *origin, struct bio *bio);
1314
1315/*
1316 * Flush a list of buffers.
1317 */
1318static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1319{
1320 struct bio *n;
1321 int r;
1322
1323 while (bio) {
1324 n = bio->bi_next;
1325 bio->bi_next = NULL;
1326 r = do_origin(s->origin, bio);
1327 if (r == DM_MAPIO_REMAPPED)
1328 generic_make_request(bio);
1329 bio = n;
1330 }
1331}
1332
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333/*
1334 * Error a list of buffers.
1335 */
1336static void error_bios(struct bio *bio)
1337{
1338 struct bio *n;
1339
1340 while (bio) {
1341 n = bio->bi_next;
1342 bio->bi_next = NULL;
NeilBrown6712ecf2007-09-27 12:47:43 +02001343 bio_io_error(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 bio = n;
1345 }
1346}
1347
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001348static void __invalidate_snapshot(struct dm_snapshot *s, int err)
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001349{
1350 if (!s->valid)
1351 return;
1352
1353 if (err == -EIO)
1354 DMERR("Invalidating snapshot: Error reading/writing.");
1355 else if (err == -ENOMEM)
1356 DMERR("Invalidating snapshot: Unable to allocate exception.");
1357
Jonathan Brassow493df712009-04-02 19:55:31 +01001358 if (s->store->type->drop_snapshot)
1359 s->store->type->drop_snapshot(s->store);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001360
1361 s->valid = 0;
1362
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001363 dm_table_event(s->ti->table);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001364}
1365
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001366static void pending_complete(struct dm_snap_pending_exception *pe, int success)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367{
Jon Brassow1d4989c2009-12-10 23:52:10 +00001368 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 struct dm_snapshot *s = pe->snap;
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001370 struct bio *origin_bios = NULL;
1371 struct bio *snapshot_bios = NULL;
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001372 struct bio *full_bio = NULL;
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001373 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001375 if (!success) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 /* Read/write error - snapshot is unusable */
1377 down_write(&s->lock);
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001378 __invalidate_snapshot(s, -EIO);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001379 error = 1;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001380 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 }
1382
Jon Brassow3510cb92009-12-10 23:52:11 +00001383 e = alloc_completed_exception();
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001384 if (!e) {
1385 down_write(&s->lock);
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001386 __invalidate_snapshot(s, -ENOMEM);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001387 error = 1;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001388 goto out;
1389 }
1390 *e = pe->e;
1391
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001392 down_write(&s->lock);
1393 if (!s->valid) {
Jon Brassow3510cb92009-12-10 23:52:11 +00001394 free_completed_exception(e);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001395 error = 1;
1396 goto out;
1397 }
1398
Mike Snitzer615d1eb2009-12-10 23:52:29 +00001399 /* Check for conflicting reads */
1400 __check_for_conflicting_io(s, pe->e.old_chunk);
Mikulas Patockaa8d41b52008-07-21 12:00:34 +01001401
1402 /*
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001403 * Add a proper exception, and remove the
1404 * in-flight exception from the list.
1405 */
Jon Brassow3510cb92009-12-10 23:52:11 +00001406 dm_insert_exception(&s->complete, e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001407
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001408out:
Jon Brassow3510cb92009-12-10 23:52:11 +00001409 dm_remove_exception(&pe->e);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001410 snapshot_bios = bio_list_get(&pe->snapshot_bios);
Mikulas Patocka515ad662009-12-10 23:52:30 +00001411 origin_bios = bio_list_get(&pe->origin_bios);
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001412 full_bio = pe->full_bio;
1413 if (full_bio) {
1414 full_bio->bi_end_io = pe->full_bio_end_io;
1415 full_bio->bi_private = pe->full_bio_private;
1416 }
Mikulas Patocka515ad662009-12-10 23:52:30 +00001417 free_pending_exception(pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
Mikulas Patocka73dfd072009-12-10 23:52:34 +00001419 increment_pending_exceptions_done_count();
1420
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001421 up_write(&s->lock);
1422
1423 /* Submit any pending write bios */
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001424 if (error) {
1425 if (full_bio)
1426 bio_io_error(full_bio);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001427 error_bios(snapshot_bios);
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001428 } else {
1429 if (full_bio)
1430 bio_endio(full_bio, 0);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001431 flush_bios(snapshot_bios);
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001432 }
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001433
Mikulas Patocka515ad662009-12-10 23:52:30 +00001434 retry_origin_bios(s, origin_bios);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435}
1436
1437static void commit_callback(void *context, int success)
1438{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001439 struct dm_snap_pending_exception *pe = context;
1440
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 pending_complete(pe, success);
1442}
1443
1444/*
1445 * Called when the copy I/O has finished. kcopyd actually runs
1446 * this code so don't block.
1447 */
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -07001448static void copy_callback(int read_err, unsigned long write_err, void *context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001450 struct dm_snap_pending_exception *pe = context;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 struct dm_snapshot *s = pe->snap;
1452
1453 if (read_err || write_err)
1454 pending_complete(pe, 0);
1455
1456 else
1457 /* Update the metadata if we are persistent */
Jonathan Brassow493df712009-04-02 19:55:31 +01001458 s->store->type->commit_exception(s->store, &pe->e,
1459 commit_callback, pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460}
1461
1462/*
1463 * Dispatches the copy operation to kcopyd.
1464 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001465static void start_copy(struct dm_snap_pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466{
1467 struct dm_snapshot *s = pe->snap;
Heinz Mauelshagen22a1ceb2008-04-24 21:43:17 +01001468 struct dm_io_region src, dest;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 struct block_device *bdev = s->origin->bdev;
1470 sector_t dev_size;
1471
1472 dev_size = get_dev_size(bdev);
1473
1474 src.bdev = bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001475 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
Mikulas Patockadf96eee2009-10-16 23:18:17 +01001476 src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001478 dest.bdev = s->cow->bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001479 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 dest.count = src.count;
1481
1482 /* Hand over to kcopyd */
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001483 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484}
1485
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001486static void full_bio_end_io(struct bio *bio, int error)
1487{
1488 void *callback_data = bio->bi_private;
1489
1490 dm_kcopyd_do_callback(callback_data, 0, error ? 1 : 0);
1491}
1492
1493static void start_full_bio(struct dm_snap_pending_exception *pe,
1494 struct bio *bio)
1495{
1496 struct dm_snapshot *s = pe->snap;
1497 void *callback_data;
1498
1499 pe->full_bio = bio;
1500 pe->full_bio_end_io = bio->bi_end_io;
1501 pe->full_bio_private = bio->bi_private;
1502
1503 callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client,
1504 copy_callback, pe);
1505
1506 bio->bi_end_io = full_bio_end_io;
1507 bio->bi_private = callback_data;
1508
1509 generic_make_request(bio);
1510}
1511
Mikulas Patocka29138082009-04-02 19:55:25 +01001512static struct dm_snap_pending_exception *
1513__lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1514{
Jon Brassow3510cb92009-12-10 23:52:11 +00001515 struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01001516
1517 if (!e)
1518 return NULL;
1519
1520 return container_of(e, struct dm_snap_pending_exception, e);
1521}
1522
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523/*
1524 * Looks to see if this snapshot already has a pending exception
1525 * for this chunk, otherwise it allocates a new one and inserts
1526 * it into the pending table.
1527 *
1528 * NOTE: a write lock must be held on snap->lock before calling
1529 * this.
1530 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001531static struct dm_snap_pending_exception *
Mikulas Patockac6621392009-04-02 19:55:25 +01001532__find_pending_exception(struct dm_snapshot *s,
1533 struct dm_snap_pending_exception *pe, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534{
Mikulas Patockac6621392009-04-02 19:55:25 +01001535 struct dm_snap_pending_exception *pe2;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001536
Mikulas Patocka29138082009-04-02 19:55:25 +01001537 pe2 = __lookup_pending_exception(s, chunk);
1538 if (pe2) {
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001539 free_pending_exception(pe);
Mikulas Patocka29138082009-04-02 19:55:25 +01001540 return pe2;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001541 }
1542
1543 pe->e.old_chunk = chunk;
1544 bio_list_init(&pe->origin_bios);
1545 bio_list_init(&pe->snapshot_bios);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001546 pe->started = 0;
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001547 pe->full_bio = NULL;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001548
Jonathan Brassow493df712009-04-02 19:55:31 +01001549 if (s->store->type->prepare_exception(s->store, &pe->e)) {
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001550 free_pending_exception(pe);
1551 return NULL;
1552 }
1553
Jon Brassow3510cb92009-12-10 23:52:11 +00001554 dm_insert_exception(&s->pending, &pe->e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001555
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 return pe;
1557}
1558
Jon Brassow1d4989c2009-12-10 23:52:10 +00001559static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
Milan Brozd74f81f2008-02-08 02:11:27 +00001560 struct bio *bio, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561{
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001562 bio->bi_bdev = s->cow->bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001563 bio->bi_sector = chunk_to_sector(s->store,
1564 dm_chunk_number(e->new_chunk) +
1565 (chunk - e->old_chunk)) +
1566 (bio->bi_sector &
1567 s->store->chunk_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568}
1569
1570static int snapshot_map(struct dm_target *ti, struct bio *bio,
1571 union map_info *map_context)
1572{
Jon Brassow1d4989c2009-12-10 23:52:10 +00001573 struct dm_exception *e;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001574 struct dm_snapshot *s = ti->private;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001575 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 chunk_t chunk;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001577 struct dm_snap_pending_exception *pe = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578
Mikulas Patockaee180262012-12-21 20:23:41 +00001579 init_tracked_chunk(bio);
1580
Tejun Heod87f4c12010-09-03 11:56:19 +02001581 if (bio->bi_rw & REQ_FLUSH) {
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001582 bio->bi_bdev = s->cow->bdev;
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01001583 return DM_MAPIO_REMAPPED;
1584 }
1585
Jonathan Brassow71fab002009-04-02 19:55:33 +01001586 chunk = sector_to_chunk(s->store, bio->bi_sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587
1588 /* Full snapshots are not usable */
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001589 /* To get here the table must be live so s->active is always set. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 if (!s->valid)
Alasdair G Kergonf6a80ea2005-07-12 15:53:01 -07001591 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001593 /* FIXME: should only take write lock if we need
1594 * to copy an exception */
1595 down_write(&s->lock);
1596
1597 if (!s->valid) {
1598 r = -EIO;
1599 goto out_unlock;
1600 }
1601
1602 /* If the block is already remapped - use that, else remap it */
Jon Brassow3510cb92009-12-10 23:52:11 +00001603 e = dm_lookup_exception(&s->complete, chunk);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001604 if (e) {
Milan Brozd74f81f2008-02-08 02:11:27 +00001605 remap_exception(s, e, bio, chunk);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001606 goto out_unlock;
1607 }
1608
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 /*
1610 * Write to snapshot - higher level takes care of RW/RO
1611 * flags so we should only get this if we are
1612 * writeable.
1613 */
1614 if (bio_rw(bio) == WRITE) {
Mikulas Patocka29138082009-04-02 19:55:25 +01001615 pe = __lookup_pending_exception(s, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001616 if (!pe) {
Mikulas Patockac6621392009-04-02 19:55:25 +01001617 up_write(&s->lock);
1618 pe = alloc_pending_exception(s);
1619 down_write(&s->lock);
1620
1621 if (!s->valid) {
1622 free_pending_exception(pe);
1623 r = -EIO;
1624 goto out_unlock;
1625 }
1626
Jon Brassow3510cb92009-12-10 23:52:11 +00001627 e = dm_lookup_exception(&s->complete, chunk);
Mikulas Patocka35bf6592009-04-02 19:55:26 +01001628 if (e) {
1629 free_pending_exception(pe);
1630 remap_exception(s, e, bio, chunk);
1631 goto out_unlock;
1632 }
1633
Mikulas Patockac6621392009-04-02 19:55:25 +01001634 pe = __find_pending_exception(s, pe, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01001635 if (!pe) {
1636 __invalidate_snapshot(s, -ENOMEM);
1637 r = -EIO;
1638 goto out_unlock;
1639 }
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001640 }
1641
Milan Brozd74f81f2008-02-08 02:11:27 +00001642 remap_exception(s, &pe->e, bio, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001643
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001644 r = DM_MAPIO_SUBMITTED;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001645
Mikulas Patockaa6e50b42011-08-02 12:32:04 +01001646 if (!pe->started &&
1647 bio->bi_size == (s->store->chunk_size << SECTOR_SHIFT)) {
1648 pe->started = 1;
1649 up_write(&s->lock);
1650 start_full_bio(pe, bio);
1651 goto out;
1652 }
1653
1654 bio_list_add(&pe->snapshot_bios, bio);
1655
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001656 if (!pe->started) {
1657 /* this is protected by snap->lock */
1658 pe->started = 1;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001659 up_write(&s->lock);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001660 start_copy(pe);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001661 goto out;
1662 }
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001663 } else {
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001664 bio->bi_bdev = s->origin->bdev;
Mikulas Patockaee180262012-12-21 20:23:41 +00001665 track_chunk(s, bio, chunk);
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001666 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001668out_unlock:
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001669 up_write(&s->lock);
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01001670out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 return r;
1672}
1673
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001674/*
1675 * A snapshot-merge target behaves like a combination of a snapshot
1676 * target and a snapshot-origin target. It only generates new
1677 * exceptions in other snapshots and not in the one that is being
1678 * merged.
1679 *
1680 * For each chunk, if there is an existing exception, it is used to
1681 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
1682 * which in turn might generate exceptions in other snapshots.
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001683 * If merging is currently taking place on the chunk in question, the
1684 * I/O is deferred by adding it to s->bios_queued_during_merge.
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001685 */
1686static int snapshot_merge_map(struct dm_target *ti, struct bio *bio,
1687 union map_info *map_context)
1688{
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) {
Mikulas Patockaddbd6582012-12-21 20:23:39 +00001697 if (!dm_bio_get_target_request_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 Patockacd45daf2008-07-21 12:00:32 +01001747static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1748 int error, union map_info *map_context)
1749{
1750 struct dm_snapshot *s = ti->private;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001751
Mikulas Patockaee180262012-12-21 20:23:41 +00001752 if (is_bio_tracked(bio))
1753 stop_tracking_chunk(s, bio);
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001754
1755 return 0;
1756}
1757
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001758static void snapshot_merge_presuspend(struct dm_target *ti)
1759{
1760 struct dm_snapshot *s = ti->private;
1761
1762 stop_merge(s);
1763}
1764
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001765static int snapshot_preresume(struct dm_target *ti)
1766{
1767 int r = 0;
1768 struct dm_snapshot *s = ti->private;
1769 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1770
1771 down_read(&_origins_lock);
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +00001772 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001773 if (snap_src && snap_dest) {
1774 down_read(&snap_src->lock);
1775 if (s == snap_src) {
1776 DMERR("Unable to resume snapshot source until "
1777 "handover completes.");
1778 r = -EINVAL;
Mike Snitzerb83b2f22011-01-13 19:59:59 +00001779 } else if (!dm_suspended(snap_src->ti)) {
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001780 DMERR("Unable to perform snapshot handover until "
1781 "source is suspended.");
1782 r = -EINVAL;
1783 }
1784 up_read(&snap_src->lock);
1785 }
1786 up_read(&_origins_lock);
1787
1788 return r;
1789}
1790
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791static void snapshot_resume(struct dm_target *ti)
1792{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001793 struct dm_snapshot *s = ti->private;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001794 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1795
1796 down_read(&_origins_lock);
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +00001797 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001798 if (snap_src && snap_dest) {
1799 down_write(&snap_src->lock);
1800 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1801 __handover_exceptions(snap_src, snap_dest);
1802 up_write(&snap_dest->lock);
1803 up_write(&snap_src->lock);
1804 }
1805 up_read(&_origins_lock);
1806
1807 /* Now we have correct chunk size, reregister */
1808 reregister_snapshot(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001810 down_write(&s->lock);
1811 s->active = 1;
1812 up_write(&s->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813}
1814
Mike Snitzer542f9032012-07-27 15:08:00 +01001815static uint32_t get_origin_minimum_chunksize(struct block_device *bdev)
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001816{
Mike Snitzer542f9032012-07-27 15:08:00 +01001817 uint32_t min_chunksize;
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001818
1819 down_read(&_origins_lock);
1820 min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
1821 up_read(&_origins_lock);
1822
1823 return min_chunksize;
1824}
1825
1826static void snapshot_merge_resume(struct dm_target *ti)
1827{
1828 struct dm_snapshot *s = ti->private;
1829
1830 /*
1831 * Handover exceptions from existing snapshot.
1832 */
1833 snapshot_resume(ti);
1834
1835 /*
Mike Snitzer542f9032012-07-27 15:08:00 +01001836 * snapshot-merge acts as an origin, so set ti->max_io_len
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001837 */
Mike Snitzer542f9032012-07-27 15:08:00 +01001838 ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev);
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001839
1840 start_merge(s);
1841}
1842
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843static int snapshot_status(struct dm_target *ti, status_type_t type,
Alasdair G Kergon1f4e0ff2012-07-27 15:08:16 +01001844 unsigned status_flags, char *result, unsigned maxlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845{
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01001846 unsigned sz = 0;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001847 struct dm_snapshot *snap = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848
1849 switch (type) {
1850 case STATUSTYPE_INFO:
Mikulas Patocka94e765722009-12-10 23:51:53 +00001851
1852 down_write(&snap->lock);
1853
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 if (!snap->valid)
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01001855 DMEMIT("Invalid");
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +00001856 else if (snap->merge_failed)
1857 DMEMIT("Merge failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 else {
Mike Snitzer985903b2009-12-10 23:52:11 +00001859 if (snap->store->type->usage) {
1860 sector_t total_sectors, sectors_allocated,
1861 metadata_sectors;
1862 snap->store->type->usage(snap->store,
1863 &total_sectors,
1864 &sectors_allocated,
1865 &metadata_sectors);
1866 DMEMIT("%llu/%llu %llu",
1867 (unsigned long long)sectors_allocated,
1868 (unsigned long long)total_sectors,
1869 (unsigned long long)metadata_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 }
1871 else
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01001872 DMEMIT("Unknown");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 }
Mikulas Patocka94e765722009-12-10 23:51:53 +00001874
1875 up_write(&snap->lock);
1876
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 break;
1878
1879 case STATUSTYPE_TABLE:
1880 /*
1881 * kdevname returns a static pointer so we need
1882 * to make private copies if the output is to
1883 * make sense.
1884 */
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001885 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
Jonathan Brassow1e302a92009-04-02 19:55:35 +01001886 snap->store->type->status(snap->store, type, result + sz,
1887 maxlen - sz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 break;
1889 }
1890
1891 return 0;
1892}
1893
Mike Snitzer8811f462009-09-04 20:40:19 +01001894static int snapshot_iterate_devices(struct dm_target *ti,
1895 iterate_devices_callout_fn fn, void *data)
1896{
1897 struct dm_snapshot *snap = ti->private;
Mikulas Patocka1e5554c2010-08-12 04:13:50 +01001898 int r;
Mike Snitzer8811f462009-09-04 20:40:19 +01001899
Mikulas Patocka1e5554c2010-08-12 04:13:50 +01001900 r = fn(ti, snap->origin, 0, ti->len, data);
1901
1902 if (!r)
1903 r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data);
1904
1905 return r;
Mike Snitzer8811f462009-09-04 20:40:19 +01001906}
1907
1908
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909/*-----------------------------------------------------------------
1910 * Origin methods
1911 *---------------------------------------------------------------*/
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001912
1913/*
1914 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1915 * supplied bio was ignored. The caller may submit it immediately.
1916 * (No remapping actually occurs as the origin is always a direct linear
1917 * map.)
1918 *
1919 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1920 * and any supplied bio is added to a list to be submitted once all
1921 * the necessary exceptions exist.
1922 */
1923static int __origin_write(struct list_head *snapshots, sector_t sector,
1924 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925{
Mikulas Patocka515ad662009-12-10 23:52:30 +00001926 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 struct dm_snapshot *snap;
Jon Brassow1d4989c2009-12-10 23:52:10 +00001928 struct dm_exception *e;
Mikulas Patocka515ad662009-12-10 23:52:30 +00001929 struct dm_snap_pending_exception *pe;
1930 struct dm_snap_pending_exception *pe_to_start_now = NULL;
1931 struct dm_snap_pending_exception *pe_to_start_last = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 chunk_t chunk;
1933
1934 /* Do all the snapshots on this origin */
1935 list_for_each_entry (snap, snapshots, list) {
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001936 /*
1937 * Don't make new exceptions in a merging snapshot
1938 * because it has effectively been deleted
1939 */
1940 if (dm_target_is_snapshot_merge(snap->ti))
1941 continue;
1942
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001943 down_write(&snap->lock);
1944
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001945 /* Only deal with valid and active snapshots */
1946 if (!snap->valid || !snap->active)
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001947 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948
Alasdair G Kergond5e404c2005-07-12 15:53:05 -07001949 /* Nothing to do if writing beyond end of snapshot */
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001950 if (sector >= dm_table_get_size(snap->ti->table))
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001951 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
1953 /*
1954 * Remember, different snapshots can have
1955 * different chunk sizes.
1956 */
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001957 chunk = sector_to_chunk(snap->store, sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958
1959 /*
1960 * Check exception table to see if block
1961 * is already remapped in this snapshot
1962 * and trigger an exception if not.
1963 */
Jon Brassow3510cb92009-12-10 23:52:11 +00001964 e = dm_lookup_exception(&snap->complete, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001965 if (e)
1966 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967
Mikulas Patocka29138082009-04-02 19:55:25 +01001968 pe = __lookup_pending_exception(snap, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001969 if (!pe) {
Mikulas Patockac6621392009-04-02 19:55:25 +01001970 up_write(&snap->lock);
1971 pe = alloc_pending_exception(snap);
1972 down_write(&snap->lock);
1973
1974 if (!snap->valid) {
1975 free_pending_exception(pe);
1976 goto next_snapshot;
1977 }
1978
Jon Brassow3510cb92009-12-10 23:52:11 +00001979 e = dm_lookup_exception(&snap->complete, chunk);
Mikulas Patocka35bf6592009-04-02 19:55:26 +01001980 if (e) {
1981 free_pending_exception(pe);
1982 goto next_snapshot;
1983 }
1984
Mikulas Patockac6621392009-04-02 19:55:25 +01001985 pe = __find_pending_exception(snap, pe, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01001986 if (!pe) {
1987 __invalidate_snapshot(snap, -ENOMEM);
1988 goto next_snapshot;
1989 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 }
1991
Mikulas Patocka515ad662009-12-10 23:52:30 +00001992 r = DM_MAPIO_SUBMITTED;
1993
1994 /*
1995 * If an origin bio was supplied, queue it to wait for the
1996 * completion of this exception, and start this one last,
1997 * at the end of the function.
1998 */
1999 if (bio) {
2000 bio_list_add(&pe->origin_bios, bio);
2001 bio = NULL;
2002
2003 if (!pe->started) {
2004 pe->started = 1;
2005 pe_to_start_last = pe;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002006 }
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002007 }
2008
2009 if (!pe->started) {
2010 pe->started = 1;
Mikulas Patocka515ad662009-12-10 23:52:30 +00002011 pe_to_start_now = pe;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002012 }
2013
Jonathan Brassowa2d2b032011-08-02 12:32:03 +01002014next_snapshot:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 up_write(&snap->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016
Mikulas Patocka515ad662009-12-10 23:52:30 +00002017 if (pe_to_start_now) {
2018 start_copy(pe_to_start_now);
2019 pe_to_start_now = NULL;
2020 }
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08002021 }
2022
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 /*
Mikulas Patocka515ad662009-12-10 23:52:30 +00002024 * Submit the exception against which the bio is queued last,
2025 * to give the other exceptions a head start.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 */
Mikulas Patocka515ad662009-12-10 23:52:30 +00002027 if (pe_to_start_last)
2028 start_copy(pe_to_start_last);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029
2030 return r;
2031}
2032
2033/*
2034 * Called on a write from the origin driver.
2035 */
2036static int do_origin(struct dm_dev *origin, struct bio *bio)
2037{
2038 struct origin *o;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08002039 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040
2041 down_read(&_origins_lock);
2042 o = __lookup_origin(origin->bdev);
2043 if (o)
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00002044 r = __origin_write(&o->snapshots, bio->bi_sector, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 up_read(&_origins_lock);
2046
2047 return r;
2048}
2049
2050/*
Mikulas Patocka73dfd072009-12-10 23:52:34 +00002051 * Trigger exceptions in all non-merging snapshots.
2052 *
2053 * The chunk size of the merging snapshot may be larger than the chunk
2054 * size of some other snapshot so we may need to reallocate multiple
2055 * chunks in other snapshots.
2056 *
2057 * We scan all the overlapping exceptions in the other snapshots.
2058 * Returns 1 if anything was reallocated and must be waited for,
2059 * otherwise returns 0.
2060 *
2061 * size must be a multiple of merging_snap's chunk_size.
2062 */
2063static int origin_write_extent(struct dm_snapshot *merging_snap,
2064 sector_t sector, unsigned size)
2065{
2066 int must_wait = 0;
2067 sector_t n;
2068 struct origin *o;
2069
2070 /*
Mike Snitzer542f9032012-07-27 15:08:00 +01002071 * The origin's __minimum_chunk_size() got stored in max_io_len
Mikulas Patocka73dfd072009-12-10 23:52:34 +00002072 * by snapshot_merge_resume().
2073 */
2074 down_read(&_origins_lock);
2075 o = __lookup_origin(merging_snap->origin->bdev);
Mike Snitzer542f9032012-07-27 15:08:00 +01002076 for (n = 0; n < size; n += merging_snap->ti->max_io_len)
Mikulas Patocka73dfd072009-12-10 23:52:34 +00002077 if (__origin_write(&o->snapshots, sector + n, NULL) ==
2078 DM_MAPIO_SUBMITTED)
2079 must_wait = 1;
2080 up_read(&_origins_lock);
2081
2082 return must_wait;
2083}
2084
2085/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 * Origin: maps a linear range of a device, with hooks for snapshotting.
2087 */
2088
2089/*
2090 * Construct an origin mapping: <dev_path>
2091 * The context for an origin is merely a 'struct dm_dev *'
2092 * pointing to the real device.
2093 */
2094static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2095{
2096 int r;
2097 struct dm_dev *dev;
2098
2099 if (argc != 1) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07002100 ti->error = "origin: incorrect number of arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 return -EINVAL;
2102 }
2103
Nikanth Karthikesan8215d6e2010-03-06 02:32:27 +00002104 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 if (r) {
2106 ti->error = "Cannot get target device";
2107 return r;
2108 }
2109
2110 ti->private = dev;
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01002111 ti->num_flush_requests = 1;
2112
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 return 0;
2114}
2115
2116static void origin_dtr(struct dm_target *ti)
2117{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002118 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 dm_put_device(ti, dev);
2120}
2121
2122static int origin_map(struct dm_target *ti, struct bio *bio,
2123 union map_info *map_context)
2124{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002125 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 bio->bi_bdev = dev->bdev;
2127
Tejun Heod87f4c12010-09-03 11:56:19 +02002128 if (bio->bi_rw & REQ_FLUSH)
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01002129 return DM_MAPIO_REMAPPED;
2130
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 /* Only tell snapshots if this is a write */
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08002132 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133}
2134
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135/*
Mike Snitzer542f9032012-07-27 15:08:00 +01002136 * Set the target "max_io_len" field to the minimum of all the snapshots'
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 * chunk sizes.
2138 */
2139static void origin_resume(struct dm_target *ti)
2140{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002141 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142
Mike Snitzer542f9032012-07-27 15:08:00 +01002143 ti->max_io_len = get_origin_minimum_chunksize(dev->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144}
2145
Alasdair G Kergon1f4e0ff2012-07-27 15:08:16 +01002146static int origin_status(struct dm_target *ti, status_type_t type,
2147 unsigned status_flags, char *result, unsigned maxlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002149 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150
2151 switch (type) {
2152 case STATUSTYPE_INFO:
2153 result[0] = '\0';
2154 break;
2155
2156 case STATUSTYPE_TABLE:
2157 snprintf(result, maxlen, "%s", dev->name);
2158 break;
2159 }
2160
2161 return 0;
2162}
2163
Mikulas Patockab1d55522010-08-12 04:14:02 +01002164static int origin_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2165 struct bio_vec *biovec, int max_size)
2166{
2167 struct dm_dev *dev = ti->private;
2168 struct request_queue *q = bdev_get_queue(dev->bdev);
2169
2170 if (!q->merge_bvec_fn)
2171 return max_size;
2172
2173 bvm->bi_bdev = dev->bdev;
Mikulas Patockab1d55522010-08-12 04:14:02 +01002174
2175 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2176}
2177
Mike Snitzer8811f462009-09-04 20:40:19 +01002178static int origin_iterate_devices(struct dm_target *ti,
2179 iterate_devices_callout_fn fn, void *data)
2180{
2181 struct dm_dev *dev = ti->private;
2182
2183 return fn(ti, dev, 0, ti->len, data);
2184}
2185
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186static struct target_type origin_target = {
2187 .name = "snapshot-origin",
Mikulas Patocka42bc9542012-12-21 20:23:38 +00002188 .version = {1, 8, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 .module = THIS_MODULE,
2190 .ctr = origin_ctr,
2191 .dtr = origin_dtr,
2192 .map = origin_map,
2193 .resume = origin_resume,
2194 .status = origin_status,
Mikulas Patockab1d55522010-08-12 04:14:02 +01002195 .merge = origin_merge,
Mike Snitzer8811f462009-09-04 20:40:19 +01002196 .iterate_devices = origin_iterate_devices,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197};
2198
2199static struct target_type snapshot_target = {
2200 .name = "snapshot",
Mikulas Patocka42bc9542012-12-21 20:23:38 +00002201 .version = {1, 11, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 .module = THIS_MODULE,
2203 .ctr = snapshot_ctr,
2204 .dtr = snapshot_dtr,
2205 .map = snapshot_map,
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002206 .end_io = snapshot_end_io,
Mike Snitzerc1f0c182009-12-10 23:52:24 +00002207 .preresume = snapshot_preresume,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 .resume = snapshot_resume,
2209 .status = snapshot_status,
Mike Snitzer8811f462009-09-04 20:40:19 +01002210 .iterate_devices = snapshot_iterate_devices,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211};
2212
Mikulas Patockad698aa42009-12-10 23:52:30 +00002213static struct target_type merge_target = {
2214 .name = dm_snapshot_merge_target_name,
Mikulas Patocka42bc9542012-12-21 20:23:38 +00002215 .version = {1, 2, 0},
Mikulas Patockad698aa42009-12-10 23:52:30 +00002216 .module = THIS_MODULE,
2217 .ctr = snapshot_ctr,
2218 .dtr = snapshot_dtr,
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00002219 .map = snapshot_merge_map,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002220 .end_io = snapshot_end_io,
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002221 .presuspend = snapshot_merge_presuspend,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002222 .preresume = snapshot_preresume,
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002223 .resume = snapshot_merge_resume,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002224 .status = snapshot_status,
2225 .iterate_devices = snapshot_iterate_devices,
2226};
2227
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228static int __init dm_snapshot_init(void)
2229{
2230 int r;
2231
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00002232 r = dm_exception_store_init();
2233 if (r) {
2234 DMERR("Failed to initialize exception stores");
2235 return r;
2236 }
2237
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 r = dm_register_target(&snapshot_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002239 if (r < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 DMERR("snapshot target register failed %d", r);
Jonathan Brassow034a1862009-10-16 23:18:14 +01002241 goto bad_register_snapshot_target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 }
2243
2244 r = dm_register_target(&origin_target);
2245 if (r < 0) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07002246 DMERR("Origin target register failed %d", r);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002247 goto bad_register_origin_target;
2248 }
2249
2250 r = dm_register_target(&merge_target);
2251 if (r < 0) {
2252 DMERR("Merge target register failed %d", r);
2253 goto bad_register_merge_target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 }
2255
2256 r = init_origin_hash();
2257 if (r) {
2258 DMERR("init_origin_hash failed.");
Mikulas Patockad698aa42009-12-10 23:52:30 +00002259 goto bad_origin_hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 }
2261
Jon Brassow1d4989c2009-12-10 23:52:10 +00002262 exception_cache = KMEM_CACHE(dm_exception, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 if (!exception_cache) {
2264 DMERR("Couldn't create exception cache.");
2265 r = -ENOMEM;
Mikulas Patockad698aa42009-12-10 23:52:30 +00002266 goto bad_exception_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 }
2268
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002269 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 if (!pending_cache) {
2271 DMERR("Couldn't create pending cache.");
2272 r = -ENOMEM;
Mikulas Patockad698aa42009-12-10 23:52:30 +00002273 goto bad_pending_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 }
2275
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 return 0;
2277
Mikulas Patockad698aa42009-12-10 23:52:30 +00002278bad_pending_cache:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 kmem_cache_destroy(exception_cache);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002280bad_exception_cache:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 exit_origin_hash();
Mikulas Patockad698aa42009-12-10 23:52:30 +00002282bad_origin_hash:
2283 dm_unregister_target(&merge_target);
2284bad_register_merge_target:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 dm_unregister_target(&origin_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002286bad_register_origin_target:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 dm_unregister_target(&snapshot_target);
Jonathan Brassow034a1862009-10-16 23:18:14 +01002288bad_register_snapshot_target:
2289 dm_exception_store_exit();
Mikulas Patockad698aa42009-12-10 23:52:30 +00002290
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 return r;
2292}
2293
2294static void __exit dm_snapshot_exit(void)
2295{
Mikulas Patocka10d3bd02009-01-06 03:04:58 +00002296 dm_unregister_target(&snapshot_target);
2297 dm_unregister_target(&origin_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002298 dm_unregister_target(&merge_target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299
2300 exit_origin_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 kmem_cache_destroy(pending_cache);
2302 kmem_cache_destroy(exception_cache);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00002303
2304 dm_exception_store_exit();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305}
2306
2307/* Module hooks */
2308module_init(dm_snapshot_init);
2309module_exit(dm_snapshot_exit);
2310
2311MODULE_DESCRIPTION(DM_NAME " snapshot target");
2312MODULE_AUTHOR("Joe Thornber");
2313MODULE_LICENSE("GPL");