blob: 23e3396e43b93b8bb2313d6d3f10e913447006c8 [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>
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010022#include <linux/workqueue.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
Jonathan Brassowaea53d92009-01-06 03:05:15 +000024#include "dm-exception-store.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
Alasdair G Kergon72d94862006-06-26 00:27:35 -070026#define DM_MSG_PREFIX "snapshots"
27
Mikulas Patockad698aa42009-12-10 23:52:30 +000028static const char dm_snapshot_merge_target_name[] = "snapshot-merge";
29
30#define dm_target_is_snapshot_merge(ti) \
31 ((ti)->type->name == dm_snapshot_merge_target_name)
32
Linus Torvalds1da177e2005-04-16 15:20:36 -070033/*
34 * The percentage increment we will wake up users at
35 */
36#define WAKE_UP_PERCENT 5
37
38/*
39 * kcopyd priority of snapshot operations
40 */
41#define SNAPSHOT_COPY_PRIORITY 2
42
43/*
Milan Broz8ee27672008-04-24 21:42:36 +010044 * Reserve 1MB for each snapshot initially (with minimum of 1 page).
Linus Torvalds1da177e2005-04-16 15:20:36 -070045 */
Milan Broz8ee27672008-04-24 21:42:36 +010046#define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
Mikulas Patockacd45daf2008-07-21 12:00:32 +010048/*
49 * The size of the mempool used to track chunks in use.
50 */
51#define MIN_IOS 256
52
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010053#define DM_TRACKED_CHUNK_HASH_SIZE 16
54#define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
55 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
56
Jon Brassow191437a2009-12-10 23:52:10 +000057struct dm_exception_table {
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010058 uint32_t hash_mask;
59 unsigned hash_shift;
60 struct list_head *table;
61};
62
63struct dm_snapshot {
64 struct rw_semaphore lock;
65
66 struct dm_dev *origin;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +000067 struct dm_dev *cow;
68
69 struct dm_target *ti;
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010070
71 /* List of snapshots per Origin */
72 struct list_head list;
73
74 /* You can't use a snapshot if this is 0 (e.g. if full) */
75 int valid;
76
77 /* Origin writes don't trigger exceptions until this is set */
78 int active;
79
Mike Snitzerc26655c2009-12-10 23:52:12 +000080 /* Whether or not owning mapped_device is suspended */
81 int suspended;
82
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010083 mempool_t *pending_pool;
84
85 atomic_t pending_exceptions_count;
86
Jon Brassow191437a2009-12-10 23:52:10 +000087 struct dm_exception_table pending;
88 struct dm_exception_table complete;
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010089
90 /*
91 * pe_lock protects all pending_exception operations and access
92 * as well as the snapshot_bios list.
93 */
94 spinlock_t pe_lock;
95
96 /* The on disk metadata handler */
97 struct dm_exception_store *store;
98
99 struct dm_kcopyd_client *kcopyd_client;
100
101 /* Queue of snapshot writes for ksnapd to flush */
102 struct bio_list queued_bios;
103 struct work_struct queued_bios_work;
104
105 /* Chunks with outstanding reads */
106 mempool_t *tracked_chunk_pool;
107 spinlock_t tracked_chunk_lock;
108 struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
109};
110
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000111struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
112{
113 return s->cow;
114}
115EXPORT_SYMBOL(dm_snap_cow);
116
Adrian Bunkc642f9e2006-12-08 02:41:13 -0800117static struct workqueue_struct *ksnapd;
David Howellsc4028952006-11-22 14:57:56 +0000118static void flush_queued_bios(struct work_struct *work);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -0700119
Jonathan Brassowccc45ea2009-04-02 19:55:34 +0100120static sector_t chunk_to_sector(struct dm_exception_store *store,
121 chunk_t chunk)
122{
123 return chunk << store->chunk_shift;
124}
125
126static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
127{
128 /*
129 * There is only ever one instance of a particular block
130 * device so we can compare pointers safely.
131 */
132 return lhs == rhs;
133}
134
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100135struct dm_snap_pending_exception {
Jon Brassow1d4989c2009-12-10 23:52:10 +0000136 struct dm_exception e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137
138 /*
139 * Origin buffers waiting for this to complete are held
140 * in a bio list
141 */
142 struct bio_list origin_bios;
143 struct bio_list snapshot_bios;
144
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 /* Pointer back to snapshot context */
146 struct dm_snapshot *snap;
147
148 /*
149 * 1 indicates the exception has already been sent to
150 * kcopyd.
151 */
152 int started;
153};
154
155/*
156 * Hash table mapping origin volumes to lists of snapshots and
157 * a lock to protect it
158 */
Christoph Lametere18b8902006-12-06 20:33:20 -0800159static struct kmem_cache *exception_cache;
160static struct kmem_cache *pending_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100162struct dm_snap_tracked_chunk {
163 struct hlist_node node;
164 chunk_t chunk;
165};
166
167static struct kmem_cache *tracked_chunk_cache;
168
169static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s,
170 chunk_t chunk)
171{
172 struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool,
173 GFP_NOIO);
174 unsigned long flags;
175
176 c->chunk = chunk;
177
178 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
179 hlist_add_head(&c->node,
180 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
181 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
182
183 return c;
184}
185
186static void stop_tracking_chunk(struct dm_snapshot *s,
187 struct dm_snap_tracked_chunk *c)
188{
189 unsigned long flags;
190
191 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
192 hlist_del(&c->node);
193 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
194
195 mempool_free(c, s->tracked_chunk_pool);
196}
197
Mikulas Patockaa8d41b52008-07-21 12:00:34 +0100198static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
199{
200 struct dm_snap_tracked_chunk *c;
201 struct hlist_node *hn;
202 int found = 0;
203
204 spin_lock_irq(&s->tracked_chunk_lock);
205
206 hlist_for_each_entry(c, hn,
207 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
208 if (c->chunk == chunk) {
209 found = 1;
210 break;
211 }
212 }
213
214 spin_unlock_irq(&s->tracked_chunk_lock);
215
216 return found;
217}
218
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219/*
Mike Snitzer615d1eb2009-12-10 23:52:29 +0000220 * This conflicting I/O is extremely improbable in the caller,
221 * so msleep(1) is sufficient and there is no need for a wait queue.
222 */
223static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
224{
225 while (__chunk_is_tracked(s, chunk))
226 msleep(1);
227}
228
229/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 * One of these per registered origin, held in the snapshot_origins hash
231 */
232struct origin {
233 /* The origin device */
234 struct block_device *bdev;
235
236 struct list_head hash_list;
237
238 /* List of snapshots for this origin */
239 struct list_head snapshots;
240};
241
242/*
243 * Size of the hash table for origin volumes. If we make this
244 * the size of the minors list then it should be nearly perfect
245 */
246#define ORIGIN_HASH_SIZE 256
247#define ORIGIN_MASK 0xFF
248static struct list_head *_origins;
249static struct rw_semaphore _origins_lock;
250
251static int init_origin_hash(void)
252{
253 int i;
254
255 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
256 GFP_KERNEL);
257 if (!_origins) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700258 DMERR("unable to allocate memory");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 return -ENOMEM;
260 }
261
262 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
263 INIT_LIST_HEAD(_origins + i);
264 init_rwsem(&_origins_lock);
265
266 return 0;
267}
268
269static void exit_origin_hash(void)
270{
271 kfree(_origins);
272}
273
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100274static unsigned origin_hash(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275{
276 return bdev->bd_dev & ORIGIN_MASK;
277}
278
279static struct origin *__lookup_origin(struct block_device *origin)
280{
281 struct list_head *ol;
282 struct origin *o;
283
284 ol = &_origins[origin_hash(origin)];
285 list_for_each_entry (o, ol, hash_list)
286 if (bdev_equal(o->bdev, origin))
287 return o;
288
289 return NULL;
290}
291
292static void __insert_origin(struct origin *o)
293{
294 struct list_head *sl = &_origins[origin_hash(o->bdev)];
295 list_add_tail(&o->hash_list, sl);
296}
297
298/*
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000299 * _origins_lock must be held when calling this function.
300 * Returns number of snapshots registered using the supplied cow device, plus:
301 * snap_src - a snapshot suitable for use as a source of exception handover
302 * snap_dest - a snapshot capable of receiving exception handover.
303 *
304 * Possible return values and states:
305 * 0: NULL, NULL - first new snapshot
306 * 1: snap_src, NULL - normal snapshot
307 * 2: snap_src, snap_dest - waiting for handover
308 * 2: snap_src, NULL - handed over, waiting for old to be deleted
309 * 1: NULL, snap_dest - source got destroyed without handover
310 */
311static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
312 struct dm_snapshot **snap_src,
313 struct dm_snapshot **snap_dest)
314{
315 struct dm_snapshot *s;
316 struct origin *o;
317 int count = 0;
318 int active;
319
320 o = __lookup_origin(snap->origin->bdev);
321 if (!o)
322 goto out;
323
324 list_for_each_entry(s, &o->snapshots, list) {
325 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
326 continue;
327
328 down_read(&s->lock);
329 active = s->active;
330 up_read(&s->lock);
331
332 if (active) {
333 if (snap_src)
334 *snap_src = s;
335 } else if (snap_dest)
336 *snap_dest = s;
337
338 count++;
339 }
340
341out:
342 return count;
343}
344
345/*
346 * On success, returns 1 if this snapshot is a handover destination,
347 * otherwise returns 0.
348 */
349static int __validate_exception_handover(struct dm_snapshot *snap)
350{
351 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
352
353 /* Does snapshot need exceptions handed over to it? */
354 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest) == 2) ||
355 snap_dest) {
356 snap->ti->error = "Snapshot cow pairing for exception "
357 "table handover failed";
358 return -EINVAL;
359 }
360
361 /*
362 * If no snap_src was found, snap cannot become a handover
363 * destination.
364 */
365 if (!snap_src)
366 return 0;
367
368 return 1;
369}
370
371static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
372{
373 struct dm_snapshot *l;
374
375 /* Sort the list according to chunk size, largest-first smallest-last */
376 list_for_each_entry(l, &o->snapshots, list)
377 if (l->store->chunk_size < s->store->chunk_size)
378 break;
379 list_add_tail(&s->list, &l->list);
380}
381
382/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 * Make a note of the snapshot and its origin so we can look it
384 * up when the origin has a write on it.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000385 *
386 * Also validate snapshot exception store handovers.
387 * On success, returns 1 if this registration is a handover destination,
388 * otherwise returns 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 */
390static int register_snapshot(struct dm_snapshot *snap)
391{
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000392 struct origin *o, *new_o = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 struct block_device *bdev = snap->origin->bdev;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000394 int r = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000396 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
397 if (!new_o)
398 return -ENOMEM;
399
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 down_write(&_origins_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000402 r = __validate_exception_handover(snap);
403 if (r < 0) {
404 kfree(new_o);
405 goto out;
406 }
407
408 o = __lookup_origin(bdev);
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000409 if (o)
410 kfree(new_o);
411 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 /* New origin */
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000413 o = new_o;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
415 /* Initialise the struct */
416 INIT_LIST_HEAD(&o->snapshots);
417 o->bdev = bdev;
418
419 __insert_origin(o);
420 }
421
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000422 __insert_snapshot(o, snap);
423
424out:
425 up_write(&_origins_lock);
426
427 return r;
428}
429
430/*
431 * Move snapshot to correct place in list according to chunk size.
432 */
433static void reregister_snapshot(struct dm_snapshot *s)
434{
435 struct block_device *bdev = s->origin->bdev;
436
437 down_write(&_origins_lock);
438
439 list_del(&s->list);
440 __insert_snapshot(__lookup_origin(bdev), s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441
442 up_write(&_origins_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443}
444
445static void unregister_snapshot(struct dm_snapshot *s)
446{
447 struct origin *o;
448
449 down_write(&_origins_lock);
450 o = __lookup_origin(s->origin->bdev);
451
452 list_del(&s->list);
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000453 if (o && list_empty(&o->snapshots)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 list_del(&o->hash_list);
455 kfree(o);
456 }
457
458 up_write(&_origins_lock);
459}
460
461/*
462 * Implementation of the exception hash tables.
Milan Brozd74f81f2008-02-08 02:11:27 +0000463 * The lowest hash_shift bits of the chunk number are ignored, allowing
464 * some consecutive chunks to be grouped together.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 */
Jon Brassow3510cb92009-12-10 23:52:11 +0000466static int dm_exception_table_init(struct dm_exception_table *et,
467 uint32_t size, unsigned hash_shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468{
469 unsigned int i;
470
Milan Brozd74f81f2008-02-08 02:11:27 +0000471 et->hash_shift = hash_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 et->hash_mask = size - 1;
473 et->table = dm_vcalloc(size, sizeof(struct list_head));
474 if (!et->table)
475 return -ENOMEM;
476
477 for (i = 0; i < size; i++)
478 INIT_LIST_HEAD(et->table + i);
479
480 return 0;
481}
482
Jon Brassow3510cb92009-12-10 23:52:11 +0000483static void dm_exception_table_exit(struct dm_exception_table *et,
484 struct kmem_cache *mem)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485{
486 struct list_head *slot;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000487 struct dm_exception *ex, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 int i, size;
489
490 size = et->hash_mask + 1;
491 for (i = 0; i < size; i++) {
492 slot = et->table + i;
493
494 list_for_each_entry_safe (ex, next, slot, hash_list)
495 kmem_cache_free(mem, ex);
496 }
497
498 vfree(et->table);
499}
500
Jon Brassow191437a2009-12-10 23:52:10 +0000501static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502{
Milan Brozd74f81f2008-02-08 02:11:27 +0000503 return (chunk >> et->hash_shift) & et->hash_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504}
505
Jon Brassow3510cb92009-12-10 23:52:11 +0000506static void dm_remove_exception(struct dm_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507{
508 list_del(&e->hash_list);
509}
510
511/*
512 * Return the exception data for a sector, or NULL if not
513 * remapped.
514 */
Jon Brassow3510cb92009-12-10 23:52:11 +0000515static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
516 chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517{
518 struct list_head *slot;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000519 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
521 slot = &et->table[exception_hash(et, chunk)];
522 list_for_each_entry (e, slot, hash_list)
Milan Brozd74f81f2008-02-08 02:11:27 +0000523 if (chunk >= e->old_chunk &&
524 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 return e;
526
527 return NULL;
528}
529
Jon Brassow3510cb92009-12-10 23:52:11 +0000530static struct dm_exception *alloc_completed_exception(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531{
Jon Brassow1d4989c2009-12-10 23:52:10 +0000532 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533
534 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
535 if (!e)
536 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
537
538 return e;
539}
540
Jon Brassow3510cb92009-12-10 23:52:11 +0000541static void free_completed_exception(struct dm_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542{
543 kmem_cache_free(exception_cache, e);
544}
545
Mikulas Patocka92e86812008-07-21 12:00:35 +0100546static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547{
Mikulas Patocka92e86812008-07-21 12:00:35 +0100548 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
549 GFP_NOIO);
550
Mikulas Patocka879129d22008-10-30 13:33:16 +0000551 atomic_inc(&s->pending_exceptions_count);
Mikulas Patocka92e86812008-07-21 12:00:35 +0100552 pe->snap = s;
553
554 return pe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555}
556
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100557static void free_pending_exception(struct dm_snap_pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558{
Mikulas Patocka879129d22008-10-30 13:33:16 +0000559 struct dm_snapshot *s = pe->snap;
560
561 mempool_free(pe, s->pending_pool);
562 smp_mb__before_atomic_dec();
563 atomic_dec(&s->pending_exceptions_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564}
565
Jon Brassow3510cb92009-12-10 23:52:11 +0000566static void dm_insert_exception(struct dm_exception_table *eh,
567 struct dm_exception *new_e)
Milan Brozd74f81f2008-02-08 02:11:27 +0000568{
Milan Brozd74f81f2008-02-08 02:11:27 +0000569 struct list_head *l;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000570 struct dm_exception *e = NULL;
Milan Brozd74f81f2008-02-08 02:11:27 +0000571
572 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
573
574 /* Add immediately if this table doesn't support consecutive chunks */
575 if (!eh->hash_shift)
576 goto out;
577
578 /* List is ordered by old_chunk */
579 list_for_each_entry_reverse(e, l, hash_list) {
580 /* Insert after an existing chunk? */
581 if (new_e->old_chunk == (e->old_chunk +
582 dm_consecutive_chunk_count(e) + 1) &&
583 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
584 dm_consecutive_chunk_count(e) + 1)) {
585 dm_consecutive_chunk_count_inc(e);
Jon Brassow3510cb92009-12-10 23:52:11 +0000586 free_completed_exception(new_e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000587 return;
588 }
589
590 /* Insert before an existing chunk? */
591 if (new_e->old_chunk == (e->old_chunk - 1) &&
592 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
593 dm_consecutive_chunk_count_inc(e);
594 e->old_chunk--;
595 e->new_chunk--;
Jon Brassow3510cb92009-12-10 23:52:11 +0000596 free_completed_exception(new_e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000597 return;
598 }
599
600 if (new_e->old_chunk > e->old_chunk)
601 break;
602 }
603
604out:
605 list_add(&new_e->hash_list, e ? &e->hash_list : l);
606}
607
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000608/*
609 * Callback used by the exception stores to load exceptions when
610 * initialising.
611 */
612static int dm_add_exception(void *context, chunk_t old, chunk_t new)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613{
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000614 struct dm_snapshot *s = context;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000615 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
Jon Brassow3510cb92009-12-10 23:52:11 +0000617 e = alloc_completed_exception();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 if (!e)
619 return -ENOMEM;
620
621 e->old_chunk = old;
Milan Brozd74f81f2008-02-08 02:11:27 +0000622
623 /* Consecutive_count is implicitly initialised to zero */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 e->new_chunk = new;
Milan Brozd74f81f2008-02-08 02:11:27 +0000625
Jon Brassow3510cb92009-12-10 23:52:11 +0000626 dm_insert_exception(&s->complete, e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000627
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 return 0;
629}
630
Mikulas Patocka7e201b32009-12-10 23:52:08 +0000631#define min_not_zero(l, r) (((l) == 0) ? (r) : (((r) == 0) ? (l) : min(l, r)))
632
633/*
634 * Return a minimum chunk size of all snapshots that have the specified origin.
635 * Return zero if the origin has no snapshots.
636 */
637static sector_t __minimum_chunk_size(struct origin *o)
638{
639 struct dm_snapshot *snap;
640 unsigned chunk_size = 0;
641
642 if (o)
643 list_for_each_entry(snap, &o->snapshots, list)
644 chunk_size = min_not_zero(chunk_size,
645 snap->store->chunk_size);
646
647 return chunk_size;
648}
649
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650/*
651 * Hard coded magic.
652 */
653static int calc_max_buckets(void)
654{
655 /* use a fixed size of 2MB */
656 unsigned long mem = 2 * 1024 * 1024;
657 mem /= sizeof(struct list_head);
658
659 return mem;
660}
661
662/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 * Allocate room for a suitable hash table.
664 */
Jonathan Brassowfee19982009-04-02 19:55:34 +0100665static int init_hash_tables(struct dm_snapshot *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666{
667 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
668
669 /*
670 * Calculate based on the size of the original volume or
671 * the COW volume...
672 */
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000673 cow_dev_size = get_dev_size(s->cow->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 origin_dev_size = get_dev_size(s->origin->bdev);
675 max_buckets = calc_max_buckets();
676
Jonathan Brassowfee19982009-04-02 19:55:34 +0100677 hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 hash_size = min(hash_size, max_buckets);
679
Mikulas Patocka8e87b9b2009-12-10 23:51:54 +0000680 if (hash_size < 64)
681 hash_size = 64;
Robert P. J. Day8defd832008-02-08 02:10:06 +0000682 hash_size = rounddown_pow_of_two(hash_size);
Jon Brassow3510cb92009-12-10 23:52:11 +0000683 if (dm_exception_table_init(&s->complete, hash_size,
684 DM_CHUNK_CONSECUTIVE_BITS))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 return -ENOMEM;
686
687 /*
688 * Allocate hash table for in-flight exceptions
689 * Make this smaller than the real hash table
690 */
691 hash_size >>= 3;
692 if (hash_size < 64)
693 hash_size = 64;
694
Jon Brassow3510cb92009-12-10 23:52:11 +0000695 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
696 dm_exception_table_exit(&s->complete, exception_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 return -ENOMEM;
698 }
699
700 return 0;
701}
702
703/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
705 */
706static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
707{
708 struct dm_snapshot *s;
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100709 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 int r = -EINVAL;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000711 char *origin_path, *cow_path;
Mike Snitzer10b81062009-12-10 23:52:31 +0000712 unsigned args_used, num_flush_requests = 1;
713 fmode_t origin_mode = FMODE_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -0700715 if (argc != 4) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700716 ti->error = "requires exactly 4 arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 r = -EINVAL;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000718 goto bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 }
720
Mike Snitzer10b81062009-12-10 23:52:31 +0000721 if (dm_target_is_snapshot_merge(ti)) {
722 num_flush_requests = 2;
723 origin_mode = FMODE_WRITE;
724 }
725
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 origin_path = argv[0];
Jonathan Brassowfee19982009-04-02 19:55:34 +0100727 argv++;
728 argc--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 s = kmalloc(sizeof(*s), GFP_KERNEL);
Jonathan Brassowfee19982009-04-02 19:55:34 +0100731 if (!s) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 ti->error = "Cannot allocate snapshot context private "
733 "structure";
734 r = -ENOMEM;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000735 goto bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 }
737
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000738 cow_path = argv[0];
739 argv++;
740 argc--;
741
742 r = dm_get_device(ti, cow_path, 0, 0,
743 FMODE_READ | FMODE_WRITE, &s->cow);
744 if (r) {
745 ti->error = "Cannot get COW device";
746 goto bad_cow;
747 }
748
749 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
750 if (r) {
751 ti->error = "Couldn't create exception store";
752 r = -EINVAL;
753 goto bad_store;
754 }
755
756 argv += args_used;
757 argc -= args_used;
758
Mike Snitzer10b81062009-12-10 23:52:31 +0000759 r = dm_get_device(ti, origin_path, 0, ti->len, origin_mode, &s->origin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 if (r) {
761 ti->error = "Cannot get origin device";
Jonathan Brassowfee19982009-04-02 19:55:34 +0100762 goto bad_origin;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 }
764
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000765 s->ti = ti;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 s->valid = 1;
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -0800767 s->active = 0;
Mike Snitzerc26655c2009-12-10 23:52:12 +0000768 s->suspended = 0;
Mikulas Patocka879129d22008-10-30 13:33:16 +0000769 atomic_set(&s->pending_exceptions_count, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 init_rwsem(&s->lock);
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000771 INIT_LIST_HEAD(&s->list);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -0700772 spin_lock_init(&s->pe_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
774 /* Allocate hash table for COW data */
Jonathan Brassowfee19982009-04-02 19:55:34 +0100775 if (init_hash_tables(s)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 ti->error = "Unable to allocate hash table space";
777 r = -ENOMEM;
Jonathan Brassowfee19982009-04-02 19:55:34 +0100778 goto bad_hash_tables;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 }
780
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100781 r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 if (r) {
783 ti->error = "Could not create kcopyd client";
Jonathan Brassowfee19982009-04-02 19:55:34 +0100784 goto bad_kcopyd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 }
786
Mikulas Patocka92e86812008-07-21 12:00:35 +0100787 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
788 if (!s->pending_pool) {
789 ti->error = "Could not allocate mempool for pending exceptions";
Jonathan Brassowfee19982009-04-02 19:55:34 +0100790 goto bad_pending_pool;
Mikulas Patocka92e86812008-07-21 12:00:35 +0100791 }
792
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100793 s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS,
794 tracked_chunk_cache);
795 if (!s->tracked_chunk_pool) {
796 ti->error = "Could not allocate tracked_chunk mempool for "
797 "tracking reads";
Mikulas Patocka92e86812008-07-21 12:00:35 +0100798 goto bad_tracked_chunk_pool;
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100799 }
800
801 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
802 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
803
804 spin_lock_init(&s->tracked_chunk_lock);
805
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000806 bio_list_init(&s->queued_bios);
807 INIT_WORK(&s->queued_bios_work, flush_queued_bios);
808
809 ti->private = s;
Mike Snitzer10b81062009-12-10 23:52:31 +0000810 ti->num_flush_requests = num_flush_requests;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000811
812 /* Add snapshot to the list of snapshots for this origin */
813 /* Exceptions aren't triggered till snapshot_resume() is called */
814 r = register_snapshot(s);
815 if (r == -ENOMEM) {
816 ti->error = "Snapshot origin struct allocation failed";
817 goto bad_load_and_register;
818 } else if (r < 0) {
819 /* invalid handover, register_snapshot has set ti->error */
820 goto bad_load_and_register;
821 }
822
823 /*
824 * Metadata must only be loaded into one table at once, so skip this
825 * if metadata will be handed over during resume.
826 * Chunk size will be set during the handover - set it to zero to
827 * ensure it's ignored.
828 */
829 if (r > 0) {
830 s->store->chunk_size = 0;
831 return 0;
832 }
833
Jonathan Brassow493df712009-04-02 19:55:31 +0100834 r = s->store->type->read_metadata(s->store, dm_add_exception,
835 (void *)s);
Milan Broz07641472007-07-12 17:28:13 +0100836 if (r < 0) {
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -0700837 ti->error = "Failed to read snapshot metadata";
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000838 goto bad_read_metadata;
Milan Broz07641472007-07-12 17:28:13 +0100839 } else if (r > 0) {
840 s->valid = 0;
841 DMWARN("Snapshot is marked invalid.");
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -0700842 }
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -0800843
Mikulas Patocka3f2412d2009-10-16 23:18:16 +0100844 if (!s->store->chunk_size) {
845 ti->error = "Chunk size not set";
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000846 goto bad_read_metadata;
Mikulas Patocka3f2412d2009-10-16 23:18:16 +0100847 }
Jonathan Brassowd0216842009-04-02 19:55:32 +0100848 ti->split_io = s->store->chunk_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
850 return 0;
851
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000852bad_read_metadata:
853 unregister_snapshot(s);
854
Jonathan Brassowfee19982009-04-02 19:55:34 +0100855bad_load_and_register:
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100856 mempool_destroy(s->tracked_chunk_pool);
857
Jonathan Brassowfee19982009-04-02 19:55:34 +0100858bad_tracked_chunk_pool:
Mikulas Patocka92e86812008-07-21 12:00:35 +0100859 mempool_destroy(s->pending_pool);
860
Jonathan Brassowfee19982009-04-02 19:55:34 +0100861bad_pending_pool:
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100862 dm_kcopyd_client_destroy(s->kcopyd_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
Jonathan Brassowfee19982009-04-02 19:55:34 +0100864bad_kcopyd:
Jon Brassow3510cb92009-12-10 23:52:11 +0000865 dm_exception_table_exit(&s->pending, pending_cache);
866 dm_exception_table_exit(&s->complete, exception_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
Jonathan Brassowfee19982009-04-02 19:55:34 +0100868bad_hash_tables:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 dm_put_device(ti, s->origin);
870
Jonathan Brassowfee19982009-04-02 19:55:34 +0100871bad_origin:
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000872 dm_exception_store_destroy(s->store);
873
874bad_store:
875 dm_put_device(ti, s->cow);
876
877bad_cow:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 kfree(s);
879
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000880bad:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 return r;
882}
883
Milan Broz31c93a02006-12-08 02:41:11 -0800884static void __free_exceptions(struct dm_snapshot *s)
885{
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100886 dm_kcopyd_client_destroy(s->kcopyd_client);
Milan Broz31c93a02006-12-08 02:41:11 -0800887 s->kcopyd_client = NULL;
888
Jon Brassow3510cb92009-12-10 23:52:11 +0000889 dm_exception_table_exit(&s->pending, pending_cache);
890 dm_exception_table_exit(&s->complete, exception_cache);
Milan Broz31c93a02006-12-08 02:41:11 -0800891}
892
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000893static void __handover_exceptions(struct dm_snapshot *snap_src,
894 struct dm_snapshot *snap_dest)
895{
896 union {
897 struct dm_exception_table table_swap;
898 struct dm_exception_store *store_swap;
899 } u;
900
901 /*
902 * Swap all snapshot context information between the two instances.
903 */
904 u.table_swap = snap_dest->complete;
905 snap_dest->complete = snap_src->complete;
906 snap_src->complete = u.table_swap;
907
908 u.store_swap = snap_dest->store;
909 snap_dest->store = snap_src->store;
910 snap_src->store = u.store_swap;
911
912 snap_dest->store->snap = snap_dest;
913 snap_src->store->snap = snap_src;
914
915 snap_dest->ti->split_io = snap_dest->store->chunk_size;
916 snap_dest->valid = snap_src->valid;
917
918 /*
919 * Set source invalid to ensure it receives no further I/O.
920 */
921 snap_src->valid = 0;
922}
923
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924static void snapshot_dtr(struct dm_target *ti)
925{
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100926#ifdef CONFIG_DM_DEBUG
927 int i;
928#endif
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100929 struct dm_snapshot *s = ti->private;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000930 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931
Alasdair G Kergonca3a9312006-10-03 01:15:30 -0700932 flush_workqueue(ksnapd);
933
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000934 down_read(&_origins_lock);
935 /* Check whether exception handover must be cancelled */
936 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest);
937 if (snap_src && snap_dest && (s == snap_src)) {
938 down_write(&snap_dest->lock);
939 snap_dest->valid = 0;
940 up_write(&snap_dest->lock);
941 DMERR("Cancelling snapshot handover.");
942 }
943 up_read(&_origins_lock);
944
Alasdair G Kergon138728dc2006-03-27 01:17:50 -0800945 /* Prevent further origin writes from using this snapshot. */
946 /* After this returns there can be no new kcopyd jobs. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 unregister_snapshot(s);
948
Mikulas Patocka879129d22008-10-30 13:33:16 +0000949 while (atomic_read(&s->pending_exceptions_count))
Mikulas Patocka90fa1522009-01-06 03:04:54 +0000950 msleep(1);
Mikulas Patocka879129d22008-10-30 13:33:16 +0000951 /*
952 * Ensure instructions in mempool_destroy aren't reordered
953 * before atomic_read.
954 */
955 smp_mb();
956
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100957#ifdef CONFIG_DM_DEBUG
958 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
959 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
960#endif
961
962 mempool_destroy(s->tracked_chunk_pool);
963
Milan Broz31c93a02006-12-08 02:41:11 -0800964 __free_exceptions(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
Mikulas Patocka92e86812008-07-21 12:00:35 +0100966 mempool_destroy(s->pending_pool);
967
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 dm_put_device(ti, s->origin);
Jonathan Brassowfee19982009-04-02 19:55:34 +0100969
970 dm_exception_store_destroy(s->store);
Alasdair G Kergon138728dc2006-03-27 01:17:50 -0800971
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000972 dm_put_device(ti, s->cow);
973
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 kfree(s);
975}
976
977/*
978 * Flush a list of buffers.
979 */
980static void flush_bios(struct bio *bio)
981{
982 struct bio *n;
983
984 while (bio) {
985 n = bio->bi_next;
986 bio->bi_next = NULL;
987 generic_make_request(bio);
988 bio = n;
989 }
990}
991
David Howellsc4028952006-11-22 14:57:56 +0000992static void flush_queued_bios(struct work_struct *work)
Alasdair G Kergonca3a9312006-10-03 01:15:30 -0700993{
David Howellsc4028952006-11-22 14:57:56 +0000994 struct dm_snapshot *s =
995 container_of(work, struct dm_snapshot, queued_bios_work);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -0700996 struct bio *queued_bios;
997 unsigned long flags;
998
999 spin_lock_irqsave(&s->pe_lock, flags);
1000 queued_bios = bio_list_get(&s->queued_bios);
1001 spin_unlock_irqrestore(&s->pe_lock, flags);
1002
1003 flush_bios(queued_bios);
1004}
1005
Mikulas Patocka515ad662009-12-10 23:52:30 +00001006static int do_origin(struct dm_dev *origin, struct bio *bio);
1007
1008/*
1009 * Flush a list of buffers.
1010 */
1011static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1012{
1013 struct bio *n;
1014 int r;
1015
1016 while (bio) {
1017 n = bio->bi_next;
1018 bio->bi_next = NULL;
1019 r = do_origin(s->origin, bio);
1020 if (r == DM_MAPIO_REMAPPED)
1021 generic_make_request(bio);
1022 bio = n;
1023 }
1024}
1025
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026/*
1027 * Error a list of buffers.
1028 */
1029static void error_bios(struct bio *bio)
1030{
1031 struct bio *n;
1032
1033 while (bio) {
1034 n = bio->bi_next;
1035 bio->bi_next = NULL;
NeilBrown6712ecf2007-09-27 12:47:43 +02001036 bio_io_error(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 bio = n;
1038 }
1039}
1040
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001041static void __invalidate_snapshot(struct dm_snapshot *s, int err)
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001042{
1043 if (!s->valid)
1044 return;
1045
1046 if (err == -EIO)
1047 DMERR("Invalidating snapshot: Error reading/writing.");
1048 else if (err == -ENOMEM)
1049 DMERR("Invalidating snapshot: Unable to allocate exception.");
1050
Jonathan Brassow493df712009-04-02 19:55:31 +01001051 if (s->store->type->drop_snapshot)
1052 s->store->type->drop_snapshot(s->store);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001053
1054 s->valid = 0;
1055
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001056 dm_table_event(s->ti->table);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001057}
1058
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001059static void pending_complete(struct dm_snap_pending_exception *pe, int success)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060{
Jon Brassow1d4989c2009-12-10 23:52:10 +00001061 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 struct dm_snapshot *s = pe->snap;
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001063 struct bio *origin_bios = NULL;
1064 struct bio *snapshot_bios = NULL;
1065 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001067 if (!success) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 /* Read/write error - snapshot is unusable */
1069 down_write(&s->lock);
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001070 __invalidate_snapshot(s, -EIO);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001071 error = 1;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001072 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 }
1074
Jon Brassow3510cb92009-12-10 23:52:11 +00001075 e = alloc_completed_exception();
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001076 if (!e) {
1077 down_write(&s->lock);
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001078 __invalidate_snapshot(s, -ENOMEM);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001079 error = 1;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001080 goto out;
1081 }
1082 *e = pe->e;
1083
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001084 down_write(&s->lock);
1085 if (!s->valid) {
Jon Brassow3510cb92009-12-10 23:52:11 +00001086 free_completed_exception(e);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001087 error = 1;
1088 goto out;
1089 }
1090
Mike Snitzer615d1eb2009-12-10 23:52:29 +00001091 /* Check for conflicting reads */
1092 __check_for_conflicting_io(s, pe->e.old_chunk);
Mikulas Patockaa8d41b52008-07-21 12:00:34 +01001093
1094 /*
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001095 * Add a proper exception, and remove the
1096 * in-flight exception from the list.
1097 */
Jon Brassow3510cb92009-12-10 23:52:11 +00001098 dm_insert_exception(&s->complete, e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001099
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 out:
Jon Brassow3510cb92009-12-10 23:52:11 +00001101 dm_remove_exception(&pe->e);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001102 snapshot_bios = bio_list_get(&pe->snapshot_bios);
Mikulas Patocka515ad662009-12-10 23:52:30 +00001103 origin_bios = bio_list_get(&pe->origin_bios);
1104 free_pending_exception(pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001106 up_write(&s->lock);
1107
1108 /* Submit any pending write bios */
1109 if (error)
1110 error_bios(snapshot_bios);
1111 else
1112 flush_bios(snapshot_bios);
1113
Mikulas Patocka515ad662009-12-10 23:52:30 +00001114 retry_origin_bios(s, origin_bios);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115}
1116
1117static void commit_callback(void *context, int success)
1118{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001119 struct dm_snap_pending_exception *pe = context;
1120
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 pending_complete(pe, success);
1122}
1123
1124/*
1125 * Called when the copy I/O has finished. kcopyd actually runs
1126 * this code so don't block.
1127 */
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -07001128static void copy_callback(int read_err, unsigned long write_err, void *context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001130 struct dm_snap_pending_exception *pe = context;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 struct dm_snapshot *s = pe->snap;
1132
1133 if (read_err || write_err)
1134 pending_complete(pe, 0);
1135
1136 else
1137 /* Update the metadata if we are persistent */
Jonathan Brassow493df712009-04-02 19:55:31 +01001138 s->store->type->commit_exception(s->store, &pe->e,
1139 commit_callback, pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140}
1141
1142/*
1143 * Dispatches the copy operation to kcopyd.
1144 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001145static void start_copy(struct dm_snap_pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146{
1147 struct dm_snapshot *s = pe->snap;
Heinz Mauelshagen22a1ceb2008-04-24 21:43:17 +01001148 struct dm_io_region src, dest;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 struct block_device *bdev = s->origin->bdev;
1150 sector_t dev_size;
1151
1152 dev_size = get_dev_size(bdev);
1153
1154 src.bdev = bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001155 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
Mikulas Patockadf96eee2009-10-16 23:18:17 +01001156 src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001158 dest.bdev = s->cow->bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001159 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 dest.count = src.count;
1161
1162 /* Hand over to kcopyd */
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +01001163 dm_kcopyd_copy(s->kcopyd_client,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 &src, 1, &dest, 0, copy_callback, pe);
1165}
1166
Mikulas Patocka29138082009-04-02 19:55:25 +01001167static struct dm_snap_pending_exception *
1168__lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1169{
Jon Brassow3510cb92009-12-10 23:52:11 +00001170 struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01001171
1172 if (!e)
1173 return NULL;
1174
1175 return container_of(e, struct dm_snap_pending_exception, e);
1176}
1177
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178/*
1179 * Looks to see if this snapshot already has a pending exception
1180 * for this chunk, otherwise it allocates a new one and inserts
1181 * it into the pending table.
1182 *
1183 * NOTE: a write lock must be held on snap->lock before calling
1184 * this.
1185 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001186static struct dm_snap_pending_exception *
Mikulas Patockac6621392009-04-02 19:55:25 +01001187__find_pending_exception(struct dm_snapshot *s,
1188 struct dm_snap_pending_exception *pe, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189{
Mikulas Patockac6621392009-04-02 19:55:25 +01001190 struct dm_snap_pending_exception *pe2;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001191
Mikulas Patocka29138082009-04-02 19:55:25 +01001192 pe2 = __lookup_pending_exception(s, chunk);
1193 if (pe2) {
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001194 free_pending_exception(pe);
Mikulas Patocka29138082009-04-02 19:55:25 +01001195 return pe2;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001196 }
1197
1198 pe->e.old_chunk = chunk;
1199 bio_list_init(&pe->origin_bios);
1200 bio_list_init(&pe->snapshot_bios);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001201 pe->started = 0;
1202
Jonathan Brassow493df712009-04-02 19:55:31 +01001203 if (s->store->type->prepare_exception(s->store, &pe->e)) {
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001204 free_pending_exception(pe);
1205 return NULL;
1206 }
1207
Jon Brassow3510cb92009-12-10 23:52:11 +00001208 dm_insert_exception(&s->pending, &pe->e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001209
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 return pe;
1211}
1212
Jon Brassow1d4989c2009-12-10 23:52:10 +00001213static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
Milan Brozd74f81f2008-02-08 02:11:27 +00001214 struct bio *bio, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215{
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001216 bio->bi_bdev = s->cow->bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001217 bio->bi_sector = chunk_to_sector(s->store,
1218 dm_chunk_number(e->new_chunk) +
1219 (chunk - e->old_chunk)) +
1220 (bio->bi_sector &
1221 s->store->chunk_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222}
1223
1224static int snapshot_map(struct dm_target *ti, struct bio *bio,
1225 union map_info *map_context)
1226{
Jon Brassow1d4989c2009-12-10 23:52:10 +00001227 struct dm_exception *e;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001228 struct dm_snapshot *s = ti->private;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001229 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 chunk_t chunk;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001231 struct dm_snap_pending_exception *pe = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01001233 if (unlikely(bio_empty_barrier(bio))) {
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001234 bio->bi_bdev = s->cow->bdev;
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01001235 return DM_MAPIO_REMAPPED;
1236 }
1237
Jonathan Brassow71fab002009-04-02 19:55:33 +01001238 chunk = sector_to_chunk(s->store, bio->bi_sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
1240 /* Full snapshots are not usable */
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001241 /* To get here the table must be live so s->active is always set. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 if (!s->valid)
Alasdair G Kergonf6a80ea2005-07-12 15:53:01 -07001243 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001245 /* FIXME: should only take write lock if we need
1246 * to copy an exception */
1247 down_write(&s->lock);
1248
1249 if (!s->valid) {
1250 r = -EIO;
1251 goto out_unlock;
1252 }
1253
1254 /* If the block is already remapped - use that, else remap it */
Jon Brassow3510cb92009-12-10 23:52:11 +00001255 e = dm_lookup_exception(&s->complete, chunk);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001256 if (e) {
Milan Brozd74f81f2008-02-08 02:11:27 +00001257 remap_exception(s, e, bio, chunk);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001258 goto out_unlock;
1259 }
1260
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 /*
1262 * Write to snapshot - higher level takes care of RW/RO
1263 * flags so we should only get this if we are
1264 * writeable.
1265 */
1266 if (bio_rw(bio) == WRITE) {
Mikulas Patocka29138082009-04-02 19:55:25 +01001267 pe = __lookup_pending_exception(s, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001268 if (!pe) {
Mikulas Patockac6621392009-04-02 19:55:25 +01001269 up_write(&s->lock);
1270 pe = alloc_pending_exception(s);
1271 down_write(&s->lock);
1272
1273 if (!s->valid) {
1274 free_pending_exception(pe);
1275 r = -EIO;
1276 goto out_unlock;
1277 }
1278
Jon Brassow3510cb92009-12-10 23:52:11 +00001279 e = dm_lookup_exception(&s->complete, chunk);
Mikulas Patocka35bf6592009-04-02 19:55:26 +01001280 if (e) {
1281 free_pending_exception(pe);
1282 remap_exception(s, e, bio, chunk);
1283 goto out_unlock;
1284 }
1285
Mikulas Patockac6621392009-04-02 19:55:25 +01001286 pe = __find_pending_exception(s, pe, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01001287 if (!pe) {
1288 __invalidate_snapshot(s, -ENOMEM);
1289 r = -EIO;
1290 goto out_unlock;
1291 }
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001292 }
1293
Milan Brozd74f81f2008-02-08 02:11:27 +00001294 remap_exception(s, &pe->e, bio, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001295 bio_list_add(&pe->snapshot_bios, bio);
1296
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001297 r = DM_MAPIO_SUBMITTED;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001298
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001299 if (!pe->started) {
1300 /* this is protected by snap->lock */
1301 pe->started = 1;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001302 up_write(&s->lock);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001303 start_copy(pe);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001304 goto out;
1305 }
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001306 } else {
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001307 bio->bi_bdev = s->origin->bdev;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001308 map_context->ptr = track_chunk(s, chunk);
1309 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001311 out_unlock:
1312 up_write(&s->lock);
1313 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 return r;
1315}
1316
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001317/*
1318 * A snapshot-merge target behaves like a combination of a snapshot
1319 * target and a snapshot-origin target. It only generates new
1320 * exceptions in other snapshots and not in the one that is being
1321 * merged.
1322 *
1323 * For each chunk, if there is an existing exception, it is used to
1324 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
1325 * which in turn might generate exceptions in other snapshots.
1326 */
1327static int snapshot_merge_map(struct dm_target *ti, struct bio *bio,
1328 union map_info *map_context)
1329{
1330 struct dm_exception *e;
1331 struct dm_snapshot *s = ti->private;
1332 int r = DM_MAPIO_REMAPPED;
1333 chunk_t chunk;
1334
Mike Snitzer10b81062009-12-10 23:52:31 +00001335 if (unlikely(bio_empty_barrier(bio))) {
1336 if (!map_context->flush_request)
1337 bio->bi_bdev = s->origin->bdev;
1338 else
1339 bio->bi_bdev = s->cow->bdev;
1340 map_context->ptr = NULL;
1341 return DM_MAPIO_REMAPPED;
1342 }
1343
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001344 chunk = sector_to_chunk(s->store, bio->bi_sector);
1345
1346 down_read(&s->lock);
1347
1348 /* Full snapshots are not usable */
1349 if (!s->valid) {
1350 r = -EIO;
1351 goto out_unlock;
1352 }
1353
1354 /* If the block is already remapped - use that */
1355 e = dm_lookup_exception(&s->complete, chunk);
1356 if (e) {
1357 remap_exception(s, e, bio, chunk);
1358 goto out_unlock;
1359 }
1360
1361 bio->bi_bdev = s->origin->bdev;
1362
1363 if (bio_rw(bio) == WRITE) {
1364 up_read(&s->lock);
1365 return do_origin(s->origin, bio);
1366 }
1367
1368out_unlock:
1369 up_read(&s->lock);
1370
1371 return r;
1372}
1373
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001374static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1375 int error, union map_info *map_context)
1376{
1377 struct dm_snapshot *s = ti->private;
1378 struct dm_snap_tracked_chunk *c = map_context->ptr;
1379
1380 if (c)
1381 stop_tracking_chunk(s, c);
1382
1383 return 0;
1384}
1385
Mike Snitzerc26655c2009-12-10 23:52:12 +00001386static void snapshot_postsuspend(struct dm_target *ti)
1387{
1388 struct dm_snapshot *s = ti->private;
1389
1390 down_write(&s->lock);
1391 s->suspended = 1;
1392 up_write(&s->lock);
1393}
1394
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001395static int snapshot_preresume(struct dm_target *ti)
1396{
1397 int r = 0;
1398 struct dm_snapshot *s = ti->private;
1399 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1400
1401 down_read(&_origins_lock);
1402 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest);
1403 if (snap_src && snap_dest) {
1404 down_read(&snap_src->lock);
1405 if (s == snap_src) {
1406 DMERR("Unable to resume snapshot source until "
1407 "handover completes.");
1408 r = -EINVAL;
1409 } else if (!snap_src->suspended) {
1410 DMERR("Unable to perform snapshot handover until "
1411 "source is suspended.");
1412 r = -EINVAL;
1413 }
1414 up_read(&snap_src->lock);
1415 }
1416 up_read(&_origins_lock);
1417
1418 return r;
1419}
1420
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421static void snapshot_resume(struct dm_target *ti)
1422{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001423 struct dm_snapshot *s = ti->private;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001424 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1425
1426 down_read(&_origins_lock);
1427 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest);
1428 if (snap_src && snap_dest) {
1429 down_write(&snap_src->lock);
1430 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1431 __handover_exceptions(snap_src, snap_dest);
1432 up_write(&snap_dest->lock);
1433 up_write(&snap_src->lock);
1434 }
1435 up_read(&_origins_lock);
1436
1437 /* Now we have correct chunk size, reregister */
1438 reregister_snapshot(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001440 down_write(&s->lock);
1441 s->active = 1;
Mike Snitzerc26655c2009-12-10 23:52:12 +00001442 s->suspended = 0;
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001443 up_write(&s->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444}
1445
1446static int snapshot_status(struct dm_target *ti, status_type_t type,
1447 char *result, unsigned int maxlen)
1448{
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01001449 unsigned sz = 0;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001450 struct dm_snapshot *snap = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451
1452 switch (type) {
1453 case STATUSTYPE_INFO:
Mikulas Patocka94e765722009-12-10 23:51:53 +00001454
1455 down_write(&snap->lock);
1456
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 if (!snap->valid)
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01001458 DMEMIT("Invalid");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 else {
Mike Snitzer985903b2009-12-10 23:52:11 +00001460 if (snap->store->type->usage) {
1461 sector_t total_sectors, sectors_allocated,
1462 metadata_sectors;
1463 snap->store->type->usage(snap->store,
1464 &total_sectors,
1465 &sectors_allocated,
1466 &metadata_sectors);
1467 DMEMIT("%llu/%llu %llu",
1468 (unsigned long long)sectors_allocated,
1469 (unsigned long long)total_sectors,
1470 (unsigned long long)metadata_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 }
1472 else
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01001473 DMEMIT("Unknown");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 }
Mikulas Patocka94e765722009-12-10 23:51:53 +00001475
1476 up_write(&snap->lock);
1477
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 break;
1479
1480 case STATUSTYPE_TABLE:
1481 /*
1482 * kdevname returns a static pointer so we need
1483 * to make private copies if the output is to
1484 * make sense.
1485 */
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001486 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
Jonathan Brassow1e302a92009-04-02 19:55:35 +01001487 snap->store->type->status(snap->store, type, result + sz,
1488 maxlen - sz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 break;
1490 }
1491
1492 return 0;
1493}
1494
Mike Snitzer8811f462009-09-04 20:40:19 +01001495static int snapshot_iterate_devices(struct dm_target *ti,
1496 iterate_devices_callout_fn fn, void *data)
1497{
1498 struct dm_snapshot *snap = ti->private;
1499
1500 return fn(ti, snap->origin, 0, ti->len, data);
1501}
1502
1503
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504/*-----------------------------------------------------------------
1505 * Origin methods
1506 *---------------------------------------------------------------*/
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001507
1508/*
1509 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1510 * supplied bio was ignored. The caller may submit it immediately.
1511 * (No remapping actually occurs as the origin is always a direct linear
1512 * map.)
1513 *
1514 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1515 * and any supplied bio is added to a list to be submitted once all
1516 * the necessary exceptions exist.
1517 */
1518static int __origin_write(struct list_head *snapshots, sector_t sector,
1519 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520{
Mikulas Patocka515ad662009-12-10 23:52:30 +00001521 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 struct dm_snapshot *snap;
Jon Brassow1d4989c2009-12-10 23:52:10 +00001523 struct dm_exception *e;
Mikulas Patocka515ad662009-12-10 23:52:30 +00001524 struct dm_snap_pending_exception *pe;
1525 struct dm_snap_pending_exception *pe_to_start_now = NULL;
1526 struct dm_snap_pending_exception *pe_to_start_last = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 chunk_t chunk;
1528
1529 /* Do all the snapshots on this origin */
1530 list_for_each_entry (snap, snapshots, list) {
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001531 /*
1532 * Don't make new exceptions in a merging snapshot
1533 * because it has effectively been deleted
1534 */
1535 if (dm_target_is_snapshot_merge(snap->ti))
1536 continue;
1537
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001538 down_write(&snap->lock);
1539
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001540 /* Only deal with valid and active snapshots */
1541 if (!snap->valid || !snap->active)
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001542 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543
Alasdair G Kergond5e404c2005-07-12 15:53:05 -07001544 /* Nothing to do if writing beyond end of snapshot */
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001545 if (sector >= dm_table_get_size(snap->ti->table))
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001546 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
1548 /*
1549 * Remember, different snapshots can have
1550 * different chunk sizes.
1551 */
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001552 chunk = sector_to_chunk(snap->store, sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553
1554 /*
1555 * Check exception table to see if block
1556 * is already remapped in this snapshot
1557 * and trigger an exception if not.
1558 */
Jon Brassow3510cb92009-12-10 23:52:11 +00001559 e = dm_lookup_exception(&snap->complete, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001560 if (e)
1561 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562
Mikulas Patocka29138082009-04-02 19:55:25 +01001563 pe = __lookup_pending_exception(snap, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001564 if (!pe) {
Mikulas Patockac6621392009-04-02 19:55:25 +01001565 up_write(&snap->lock);
1566 pe = alloc_pending_exception(snap);
1567 down_write(&snap->lock);
1568
1569 if (!snap->valid) {
1570 free_pending_exception(pe);
1571 goto next_snapshot;
1572 }
1573
Jon Brassow3510cb92009-12-10 23:52:11 +00001574 e = dm_lookup_exception(&snap->complete, chunk);
Mikulas Patocka35bf6592009-04-02 19:55:26 +01001575 if (e) {
1576 free_pending_exception(pe);
1577 goto next_snapshot;
1578 }
1579
Mikulas Patockac6621392009-04-02 19:55:25 +01001580 pe = __find_pending_exception(snap, pe, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01001581 if (!pe) {
1582 __invalidate_snapshot(snap, -ENOMEM);
1583 goto next_snapshot;
1584 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 }
1586
Mikulas Patocka515ad662009-12-10 23:52:30 +00001587 r = DM_MAPIO_SUBMITTED;
1588
1589 /*
1590 * If an origin bio was supplied, queue it to wait for the
1591 * completion of this exception, and start this one last,
1592 * at the end of the function.
1593 */
1594 if (bio) {
1595 bio_list_add(&pe->origin_bios, bio);
1596 bio = NULL;
1597
1598 if (!pe->started) {
1599 pe->started = 1;
1600 pe_to_start_last = pe;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001601 }
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001602 }
1603
1604 if (!pe->started) {
1605 pe->started = 1;
Mikulas Patocka515ad662009-12-10 23:52:30 +00001606 pe_to_start_now = pe;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001607 }
1608
1609 next_snapshot:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 up_write(&snap->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
Mikulas Patocka515ad662009-12-10 23:52:30 +00001612 if (pe_to_start_now) {
1613 start_copy(pe_to_start_now);
1614 pe_to_start_now = NULL;
1615 }
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001616 }
1617
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 /*
Mikulas Patocka515ad662009-12-10 23:52:30 +00001619 * Submit the exception against which the bio is queued last,
1620 * to give the other exceptions a head start.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 */
Mikulas Patocka515ad662009-12-10 23:52:30 +00001622 if (pe_to_start_last)
1623 start_copy(pe_to_start_last);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624
1625 return r;
1626}
1627
1628/*
1629 * Called on a write from the origin driver.
1630 */
1631static int do_origin(struct dm_dev *origin, struct bio *bio)
1632{
1633 struct origin *o;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001634 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635
1636 down_read(&_origins_lock);
1637 o = __lookup_origin(origin->bdev);
1638 if (o)
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001639 r = __origin_write(&o->snapshots, bio->bi_sector, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 up_read(&_origins_lock);
1641
1642 return r;
1643}
1644
1645/*
1646 * Origin: maps a linear range of a device, with hooks for snapshotting.
1647 */
1648
1649/*
1650 * Construct an origin mapping: <dev_path>
1651 * The context for an origin is merely a 'struct dm_dev *'
1652 * pointing to the real device.
1653 */
1654static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1655{
1656 int r;
1657 struct dm_dev *dev;
1658
1659 if (argc != 1) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001660 ti->error = "origin: incorrect number of arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 return -EINVAL;
1662 }
1663
1664 r = dm_get_device(ti, argv[0], 0, ti->len,
1665 dm_table_get_mode(ti->table), &dev);
1666 if (r) {
1667 ti->error = "Cannot get target device";
1668 return r;
1669 }
1670
1671 ti->private = dev;
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01001672 ti->num_flush_requests = 1;
1673
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 return 0;
1675}
1676
1677static void origin_dtr(struct dm_target *ti)
1678{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001679 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 dm_put_device(ti, dev);
1681}
1682
1683static int origin_map(struct dm_target *ti, struct bio *bio,
1684 union map_info *map_context)
1685{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001686 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 bio->bi_bdev = dev->bdev;
1688
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01001689 if (unlikely(bio_empty_barrier(bio)))
1690 return DM_MAPIO_REMAPPED;
1691
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 /* Only tell snapshots if this is a write */
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001693 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694}
1695
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696/*
1697 * Set the target "split_io" field to the minimum of all the snapshots'
1698 * chunk sizes.
1699 */
1700static void origin_resume(struct dm_target *ti)
1701{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001702 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703
1704 down_read(&_origins_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705
Mikulas Patocka7e201b32009-12-10 23:52:08 +00001706 ti->split_io = __minimum_chunk_size(__lookup_origin(dev->bdev));
1707
1708 up_read(&_origins_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709}
1710
1711static int origin_status(struct dm_target *ti, status_type_t type, char *result,
1712 unsigned int maxlen)
1713{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001714 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715
1716 switch (type) {
1717 case STATUSTYPE_INFO:
1718 result[0] = '\0';
1719 break;
1720
1721 case STATUSTYPE_TABLE:
1722 snprintf(result, maxlen, "%s", dev->name);
1723 break;
1724 }
1725
1726 return 0;
1727}
1728
Mike Snitzer8811f462009-09-04 20:40:19 +01001729static int origin_iterate_devices(struct dm_target *ti,
1730 iterate_devices_callout_fn fn, void *data)
1731{
1732 struct dm_dev *dev = ti->private;
1733
1734 return fn(ti, dev, 0, ti->len, data);
1735}
1736
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737static struct target_type origin_target = {
1738 .name = "snapshot-origin",
Mike Snitzer8811f462009-09-04 20:40:19 +01001739 .version = {1, 7, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 .module = THIS_MODULE,
1741 .ctr = origin_ctr,
1742 .dtr = origin_dtr,
1743 .map = origin_map,
1744 .resume = origin_resume,
1745 .status = origin_status,
Mike Snitzer8811f462009-09-04 20:40:19 +01001746 .iterate_devices = origin_iterate_devices,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747};
1748
1749static struct target_type snapshot_target = {
1750 .name = "snapshot",
Mike Snitzerc26655c2009-12-10 23:52:12 +00001751 .version = {1, 9, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 .module = THIS_MODULE,
1753 .ctr = snapshot_ctr,
1754 .dtr = snapshot_dtr,
1755 .map = snapshot_map,
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001756 .end_io = snapshot_end_io,
Mike Snitzerc26655c2009-12-10 23:52:12 +00001757 .postsuspend = snapshot_postsuspend,
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001758 .preresume = snapshot_preresume,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 .resume = snapshot_resume,
1760 .status = snapshot_status,
Mike Snitzer8811f462009-09-04 20:40:19 +01001761 .iterate_devices = snapshot_iterate_devices,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762};
1763
Mikulas Patockad698aa42009-12-10 23:52:30 +00001764static struct target_type merge_target = {
1765 .name = dm_snapshot_merge_target_name,
1766 .version = {1, 0, 0},
1767 .module = THIS_MODULE,
1768 .ctr = snapshot_ctr,
1769 .dtr = snapshot_dtr,
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001770 .map = snapshot_merge_map,
Mikulas Patockad698aa42009-12-10 23:52:30 +00001771 .end_io = snapshot_end_io,
1772 .postsuspend = snapshot_postsuspend,
1773 .preresume = snapshot_preresume,
1774 .resume = snapshot_resume,
1775 .status = snapshot_status,
1776 .iterate_devices = snapshot_iterate_devices,
1777};
1778
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779static int __init dm_snapshot_init(void)
1780{
1781 int r;
1782
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00001783 r = dm_exception_store_init();
1784 if (r) {
1785 DMERR("Failed to initialize exception stores");
1786 return r;
1787 }
1788
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 r = dm_register_target(&snapshot_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00001790 if (r < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 DMERR("snapshot target register failed %d", r);
Jonathan Brassow034a1862009-10-16 23:18:14 +01001792 goto bad_register_snapshot_target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 }
1794
1795 r = dm_register_target(&origin_target);
1796 if (r < 0) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001797 DMERR("Origin target register failed %d", r);
Mikulas Patockad698aa42009-12-10 23:52:30 +00001798 goto bad_register_origin_target;
1799 }
1800
1801 r = dm_register_target(&merge_target);
1802 if (r < 0) {
1803 DMERR("Merge target register failed %d", r);
1804 goto bad_register_merge_target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 }
1806
1807 r = init_origin_hash();
1808 if (r) {
1809 DMERR("init_origin_hash failed.");
Mikulas Patockad698aa42009-12-10 23:52:30 +00001810 goto bad_origin_hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 }
1812
Jon Brassow1d4989c2009-12-10 23:52:10 +00001813 exception_cache = KMEM_CACHE(dm_exception, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 if (!exception_cache) {
1815 DMERR("Couldn't create exception cache.");
1816 r = -ENOMEM;
Mikulas Patockad698aa42009-12-10 23:52:30 +00001817 goto bad_exception_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 }
1819
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001820 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 if (!pending_cache) {
1822 DMERR("Couldn't create pending cache.");
1823 r = -ENOMEM;
Mikulas Patockad698aa42009-12-10 23:52:30 +00001824 goto bad_pending_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 }
1826
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001827 tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
1828 if (!tracked_chunk_cache) {
1829 DMERR("Couldn't create cache to track chunks in use.");
1830 r = -ENOMEM;
Mikulas Patockad698aa42009-12-10 23:52:30 +00001831 goto bad_tracked_chunk_cache;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001832 }
1833
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07001834 ksnapd = create_singlethread_workqueue("ksnapd");
1835 if (!ksnapd) {
1836 DMERR("Failed to create ksnapd workqueue.");
1837 r = -ENOMEM;
Mikulas Patocka92e86812008-07-21 12:00:35 +01001838 goto bad_pending_pool;
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07001839 }
1840
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 return 0;
1842
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00001843bad_pending_pool:
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001844 kmem_cache_destroy(tracked_chunk_cache);
Mikulas Patockad698aa42009-12-10 23:52:30 +00001845bad_tracked_chunk_cache:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 kmem_cache_destroy(pending_cache);
Mikulas Patockad698aa42009-12-10 23:52:30 +00001847bad_pending_cache:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 kmem_cache_destroy(exception_cache);
Mikulas Patockad698aa42009-12-10 23:52:30 +00001849bad_exception_cache:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 exit_origin_hash();
Mikulas Patockad698aa42009-12-10 23:52:30 +00001851bad_origin_hash:
1852 dm_unregister_target(&merge_target);
1853bad_register_merge_target:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 dm_unregister_target(&origin_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00001855bad_register_origin_target:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 dm_unregister_target(&snapshot_target);
Jonathan Brassow034a1862009-10-16 23:18:14 +01001857bad_register_snapshot_target:
1858 dm_exception_store_exit();
Mikulas Patockad698aa42009-12-10 23:52:30 +00001859
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 return r;
1861}
1862
1863static void __exit dm_snapshot_exit(void)
1864{
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07001865 destroy_workqueue(ksnapd);
1866
Mikulas Patocka10d3bd02009-01-06 03:04:58 +00001867 dm_unregister_target(&snapshot_target);
1868 dm_unregister_target(&origin_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00001869 dm_unregister_target(&merge_target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870
1871 exit_origin_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 kmem_cache_destroy(pending_cache);
1873 kmem_cache_destroy(exception_cache);
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001874 kmem_cache_destroy(tracked_chunk_cache);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00001875
1876 dm_exception_store_exit();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877}
1878
1879/* Module hooks */
1880module_init(dm_snapshot_init);
1881module_exit(dm_snapshot_exit);
1882
1883MODULE_DESCRIPTION(DM_NAME " snapshot target");
1884MODULE_AUTHOR("Joe Thornber");
1885MODULE_LICENSE("GPL");