blob: 0789c22ff0d4993d9f625febc7cc8cc5c878bc52 [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
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +000074 /*
75 * You can't use a snapshot if this is 0 (e.g. if full).
76 * A snapshot-merge target never clears this.
77 */
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010078 int valid;
79
80 /* Origin writes don't trigger exceptions until this is set */
81 int active;
82
Mike Snitzerc26655c2009-12-10 23:52:12 +000083 /* Whether or not owning mapped_device is suspended */
84 int suspended;
85
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010086 mempool_t *pending_pool;
87
88 atomic_t pending_exceptions_count;
89
Jon Brassow191437a2009-12-10 23:52:10 +000090 struct dm_exception_table pending;
91 struct dm_exception_table complete;
Jonathan Brassowccc45ea2009-04-02 19:55:34 +010092
93 /*
94 * pe_lock protects all pending_exception operations and access
95 * as well as the snapshot_bios list.
96 */
97 spinlock_t pe_lock;
98
99 /* The on disk metadata handler */
100 struct dm_exception_store *store;
101
102 struct dm_kcopyd_client *kcopyd_client;
103
104 /* Queue of snapshot writes for ksnapd to flush */
105 struct bio_list queued_bios;
106 struct work_struct queued_bios_work;
107
108 /* Chunks with outstanding reads */
109 mempool_t *tracked_chunk_pool;
110 spinlock_t tracked_chunk_lock;
111 struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000112
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +0000113 /*
114 * The merge operation failed if this flag is set.
115 * Failure modes are handled as follows:
116 * - I/O error reading the header
117 * => don't load the target; abort.
118 * - Header does not have "valid" flag set
119 * => use the origin; forget about the snapshot.
120 * - I/O error when reading exceptions
121 * => don't load the target; abort.
122 * (We can't use the intermediate origin state.)
123 * - I/O error while merging
124 * => stop merging; set merge_failed; process I/O normally.
125 */
126 int merge_failed;
127
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000128 /* Wait for events based on state_bits */
129 unsigned long state_bits;
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000130
131 /* Range of chunks currently being merged. */
132 chunk_t first_merging_chunk;
133 int num_merging_chunks;
134
135 /*
136 * Incoming bios that overlap with chunks being merged must wait
137 * for them to be committed.
138 */
139 struct bio_list bios_queued_during_merge;
Jonathan Brassowccc45ea2009-04-02 19:55:34 +0100140};
141
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000142/*
143 * state_bits:
144 * RUNNING_MERGE - Merge operation is in progress.
145 * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
146 * cleared afterwards.
147 */
148#define RUNNING_MERGE 0
149#define SHUTDOWN_MERGE 1
150
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000151struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
152{
153 return s->cow;
154}
155EXPORT_SYMBOL(dm_snap_cow);
156
Adrian Bunkc642f9e2006-12-08 02:41:13 -0800157static struct workqueue_struct *ksnapd;
David Howellsc4028952006-11-22 14:57:56 +0000158static void flush_queued_bios(struct work_struct *work);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -0700159
Jonathan Brassowccc45ea2009-04-02 19:55:34 +0100160static sector_t chunk_to_sector(struct dm_exception_store *store,
161 chunk_t chunk)
162{
163 return chunk << store->chunk_shift;
164}
165
166static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
167{
168 /*
169 * There is only ever one instance of a particular block
170 * device so we can compare pointers safely.
171 */
172 return lhs == rhs;
173}
174
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100175struct dm_snap_pending_exception {
Jon Brassow1d4989c2009-12-10 23:52:10 +0000176 struct dm_exception e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
178 /*
179 * Origin buffers waiting for this to complete are held
180 * in a bio list
181 */
182 struct bio_list origin_bios;
183 struct bio_list snapshot_bios;
184
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 /* Pointer back to snapshot context */
186 struct dm_snapshot *snap;
187
188 /*
189 * 1 indicates the exception has already been sent to
190 * kcopyd.
191 */
192 int started;
193};
194
195/*
196 * Hash table mapping origin volumes to lists of snapshots and
197 * a lock to protect it
198 */
Christoph Lametere18b8902006-12-06 20:33:20 -0800199static struct kmem_cache *exception_cache;
200static struct kmem_cache *pending_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
Mikulas Patockacd45daf2008-07-21 12:00:32 +0100202struct dm_snap_tracked_chunk {
203 struct hlist_node node;
204 chunk_t chunk;
205};
206
207static struct kmem_cache *tracked_chunk_cache;
208
209static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s,
210 chunk_t chunk)
211{
212 struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool,
213 GFP_NOIO);
214 unsigned long flags;
215
216 c->chunk = chunk;
217
218 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
219 hlist_add_head(&c->node,
220 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
221 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
222
223 return c;
224}
225
226static void stop_tracking_chunk(struct dm_snapshot *s,
227 struct dm_snap_tracked_chunk *c)
228{
229 unsigned long flags;
230
231 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
232 hlist_del(&c->node);
233 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
234
235 mempool_free(c, s->tracked_chunk_pool);
236}
237
Mikulas Patockaa8d41b52008-07-21 12:00:34 +0100238static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
239{
240 struct dm_snap_tracked_chunk *c;
241 struct hlist_node *hn;
242 int found = 0;
243
244 spin_lock_irq(&s->tracked_chunk_lock);
245
246 hlist_for_each_entry(c, hn,
247 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
248 if (c->chunk == chunk) {
249 found = 1;
250 break;
251 }
252 }
253
254 spin_unlock_irq(&s->tracked_chunk_lock);
255
256 return found;
257}
258
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259/*
Mike Snitzer615d1eb2009-12-10 23:52:29 +0000260 * This conflicting I/O is extremely improbable in the caller,
261 * so msleep(1) is sufficient and there is no need for a wait queue.
262 */
263static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
264{
265 while (__chunk_is_tracked(s, chunk))
266 msleep(1);
267}
268
269/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 * One of these per registered origin, held in the snapshot_origins hash
271 */
272struct origin {
273 /* The origin device */
274 struct block_device *bdev;
275
276 struct list_head hash_list;
277
278 /* List of snapshots for this origin */
279 struct list_head snapshots;
280};
281
282/*
283 * Size of the hash table for origin volumes. If we make this
284 * the size of the minors list then it should be nearly perfect
285 */
286#define ORIGIN_HASH_SIZE 256
287#define ORIGIN_MASK 0xFF
288static struct list_head *_origins;
289static struct rw_semaphore _origins_lock;
290
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000291static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
292static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
293static uint64_t _pending_exceptions_done_count;
294
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295static int init_origin_hash(void)
296{
297 int i;
298
299 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
300 GFP_KERNEL);
301 if (!_origins) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700302 DMERR("unable to allocate memory");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 return -ENOMEM;
304 }
305
306 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
307 INIT_LIST_HEAD(_origins + i);
308 init_rwsem(&_origins_lock);
309
310 return 0;
311}
312
313static void exit_origin_hash(void)
314{
315 kfree(_origins);
316}
317
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100318static unsigned origin_hash(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
320 return bdev->bd_dev & ORIGIN_MASK;
321}
322
323static struct origin *__lookup_origin(struct block_device *origin)
324{
325 struct list_head *ol;
326 struct origin *o;
327
328 ol = &_origins[origin_hash(origin)];
329 list_for_each_entry (o, ol, hash_list)
330 if (bdev_equal(o->bdev, origin))
331 return o;
332
333 return NULL;
334}
335
336static void __insert_origin(struct origin *o)
337{
338 struct list_head *sl = &_origins[origin_hash(o->bdev)];
339 list_add_tail(&o->hash_list, sl);
340}
341
342/*
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000343 * _origins_lock must be held when calling this function.
344 * Returns number of snapshots registered using the supplied cow device, plus:
345 * snap_src - a snapshot suitable for use as a source of exception handover
346 * snap_dest - a snapshot capable of receiving exception handover.
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000347 * snap_merge - an existing snapshot-merge target linked to the same origin.
348 * There can be at most one snapshot-merge target. The parameter is optional.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000349 *
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000350 * Possible return values and states of snap_src and snap_dest.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000351 * 0: NULL, NULL - first new snapshot
352 * 1: snap_src, NULL - normal snapshot
353 * 2: snap_src, snap_dest - waiting for handover
354 * 2: snap_src, NULL - handed over, waiting for old to be deleted
355 * 1: NULL, snap_dest - source got destroyed without handover
356 */
357static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
358 struct dm_snapshot **snap_src,
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000359 struct dm_snapshot **snap_dest,
360 struct dm_snapshot **snap_merge)
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000361{
362 struct dm_snapshot *s;
363 struct origin *o;
364 int count = 0;
365 int active;
366
367 o = __lookup_origin(snap->origin->bdev);
368 if (!o)
369 goto out;
370
371 list_for_each_entry(s, &o->snapshots, list) {
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000372 if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
373 *snap_merge = s;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000374 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
375 continue;
376
377 down_read(&s->lock);
378 active = s->active;
379 up_read(&s->lock);
380
381 if (active) {
382 if (snap_src)
383 *snap_src = s;
384 } else if (snap_dest)
385 *snap_dest = s;
386
387 count++;
388 }
389
390out:
391 return count;
392}
393
394/*
395 * On success, returns 1 if this snapshot is a handover destination,
396 * otherwise returns 0.
397 */
398static int __validate_exception_handover(struct dm_snapshot *snap)
399{
400 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000401 struct dm_snapshot *snap_merge = NULL;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000402
403 /* Does snapshot need exceptions handed over to it? */
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000404 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
405 &snap_merge) == 2) ||
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000406 snap_dest) {
407 snap->ti->error = "Snapshot cow pairing for exception "
408 "table handover failed";
409 return -EINVAL;
410 }
411
412 /*
413 * If no snap_src was found, snap cannot become a handover
414 * destination.
415 */
416 if (!snap_src)
417 return 0;
418
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +0000419 /*
420 * Non-snapshot-merge handover?
421 */
422 if (!dm_target_is_snapshot_merge(snap->ti))
423 return 1;
424
425 /*
426 * Do not allow more than one merging snapshot.
427 */
428 if (snap_merge) {
429 snap->ti->error = "A snapshot is already merging.";
430 return -EINVAL;
431 }
432
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000433 if (!snap_src->store->type->prepare_merge ||
434 !snap_src->store->type->commit_merge) {
435 snap->ti->error = "Snapshot exception store does not "
436 "support snapshot-merge.";
437 return -EINVAL;
438 }
439
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000440 return 1;
441}
442
443static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
444{
445 struct dm_snapshot *l;
446
447 /* Sort the list according to chunk size, largest-first smallest-last */
448 list_for_each_entry(l, &o->snapshots, list)
449 if (l->store->chunk_size < s->store->chunk_size)
450 break;
451 list_add_tail(&s->list, &l->list);
452}
453
454/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 * Make a note of the snapshot and its origin so we can look it
456 * up when the origin has a write on it.
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000457 *
458 * Also validate snapshot exception store handovers.
459 * On success, returns 1 if this registration is a handover destination,
460 * otherwise returns 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 */
462static int register_snapshot(struct dm_snapshot *snap)
463{
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000464 struct origin *o, *new_o = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 struct block_device *bdev = snap->origin->bdev;
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000466 int r = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000468 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
469 if (!new_o)
470 return -ENOMEM;
471
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 down_write(&_origins_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000474 r = __validate_exception_handover(snap);
475 if (r < 0) {
476 kfree(new_o);
477 goto out;
478 }
479
480 o = __lookup_origin(bdev);
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000481 if (o)
482 kfree(new_o);
483 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 /* New origin */
Mikulas Patocka60c856c82008-10-30 13:33:12 +0000485 o = new_o;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
487 /* Initialise the struct */
488 INIT_LIST_HEAD(&o->snapshots);
489 o->bdev = bdev;
490
491 __insert_origin(o);
492 }
493
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000494 __insert_snapshot(o, snap);
495
496out:
497 up_write(&_origins_lock);
498
499 return r;
500}
501
502/*
503 * Move snapshot to correct place in list according to chunk size.
504 */
505static void reregister_snapshot(struct dm_snapshot *s)
506{
507 struct block_device *bdev = s->origin->bdev;
508
509 down_write(&_origins_lock);
510
511 list_del(&s->list);
512 __insert_snapshot(__lookup_origin(bdev), s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
514 up_write(&_origins_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515}
516
517static void unregister_snapshot(struct dm_snapshot *s)
518{
519 struct origin *o;
520
521 down_write(&_origins_lock);
522 o = __lookup_origin(s->origin->bdev);
523
524 list_del(&s->list);
Mike Snitzerc1f0c182009-12-10 23:52:24 +0000525 if (o && list_empty(&o->snapshots)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 list_del(&o->hash_list);
527 kfree(o);
528 }
529
530 up_write(&_origins_lock);
531}
532
533/*
534 * Implementation of the exception hash tables.
Milan Brozd74f81f2008-02-08 02:11:27 +0000535 * The lowest hash_shift bits of the chunk number are ignored, allowing
536 * some consecutive chunks to be grouped together.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 */
Jon Brassow3510cb92009-12-10 23:52:11 +0000538static int dm_exception_table_init(struct dm_exception_table *et,
539 uint32_t size, unsigned hash_shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540{
541 unsigned int i;
542
Milan Brozd74f81f2008-02-08 02:11:27 +0000543 et->hash_shift = hash_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 et->hash_mask = size - 1;
545 et->table = dm_vcalloc(size, sizeof(struct list_head));
546 if (!et->table)
547 return -ENOMEM;
548
549 for (i = 0; i < size; i++)
550 INIT_LIST_HEAD(et->table + i);
551
552 return 0;
553}
554
Jon Brassow3510cb92009-12-10 23:52:11 +0000555static void dm_exception_table_exit(struct dm_exception_table *et,
556 struct kmem_cache *mem)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557{
558 struct list_head *slot;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000559 struct dm_exception *ex, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 int i, size;
561
562 size = et->hash_mask + 1;
563 for (i = 0; i < size; i++) {
564 slot = et->table + i;
565
566 list_for_each_entry_safe (ex, next, slot, hash_list)
567 kmem_cache_free(mem, ex);
568 }
569
570 vfree(et->table);
571}
572
Jon Brassow191437a2009-12-10 23:52:10 +0000573static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574{
Milan Brozd74f81f2008-02-08 02:11:27 +0000575 return (chunk >> et->hash_shift) & et->hash_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576}
577
Jon Brassow3510cb92009-12-10 23:52:11 +0000578static void dm_remove_exception(struct dm_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579{
580 list_del(&e->hash_list);
581}
582
583/*
584 * Return the exception data for a sector, or NULL if not
585 * remapped.
586 */
Jon Brassow3510cb92009-12-10 23:52:11 +0000587static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
588 chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589{
590 struct list_head *slot;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000591 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
593 slot = &et->table[exception_hash(et, chunk)];
594 list_for_each_entry (e, slot, hash_list)
Milan Brozd74f81f2008-02-08 02:11:27 +0000595 if (chunk >= e->old_chunk &&
596 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 return e;
598
599 return NULL;
600}
601
Jon Brassow3510cb92009-12-10 23:52:11 +0000602static struct dm_exception *alloc_completed_exception(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603{
Jon Brassow1d4989c2009-12-10 23:52:10 +0000604 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605
606 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
607 if (!e)
608 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
609
610 return e;
611}
612
Jon Brassow3510cb92009-12-10 23:52:11 +0000613static void free_completed_exception(struct dm_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614{
615 kmem_cache_free(exception_cache, e);
616}
617
Mikulas Patocka92e86812008-07-21 12:00:35 +0100618static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619{
Mikulas Patocka92e86812008-07-21 12:00:35 +0100620 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
621 GFP_NOIO);
622
Mikulas Patocka879129d22008-10-30 13:33:16 +0000623 atomic_inc(&s->pending_exceptions_count);
Mikulas Patocka92e86812008-07-21 12:00:35 +0100624 pe->snap = s;
625
626 return pe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627}
628
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100629static void free_pending_exception(struct dm_snap_pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630{
Mikulas Patocka879129d22008-10-30 13:33:16 +0000631 struct dm_snapshot *s = pe->snap;
632
633 mempool_free(pe, s->pending_pool);
634 smp_mb__before_atomic_dec();
635 atomic_dec(&s->pending_exceptions_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636}
637
Jon Brassow3510cb92009-12-10 23:52:11 +0000638static void dm_insert_exception(struct dm_exception_table *eh,
639 struct dm_exception *new_e)
Milan Brozd74f81f2008-02-08 02:11:27 +0000640{
Milan Brozd74f81f2008-02-08 02:11:27 +0000641 struct list_head *l;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000642 struct dm_exception *e = NULL;
Milan Brozd74f81f2008-02-08 02:11:27 +0000643
644 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
645
646 /* Add immediately if this table doesn't support consecutive chunks */
647 if (!eh->hash_shift)
648 goto out;
649
650 /* List is ordered by old_chunk */
651 list_for_each_entry_reverse(e, l, hash_list) {
652 /* Insert after an existing chunk? */
653 if (new_e->old_chunk == (e->old_chunk +
654 dm_consecutive_chunk_count(e) + 1) &&
655 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
656 dm_consecutive_chunk_count(e) + 1)) {
657 dm_consecutive_chunk_count_inc(e);
Jon Brassow3510cb92009-12-10 23:52:11 +0000658 free_completed_exception(new_e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000659 return;
660 }
661
662 /* Insert before an existing chunk? */
663 if (new_e->old_chunk == (e->old_chunk - 1) &&
664 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
665 dm_consecutive_chunk_count_inc(e);
666 e->old_chunk--;
667 e->new_chunk--;
Jon Brassow3510cb92009-12-10 23:52:11 +0000668 free_completed_exception(new_e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000669 return;
670 }
671
672 if (new_e->old_chunk > e->old_chunk)
673 break;
674 }
675
676out:
677 list_add(&new_e->hash_list, e ? &e->hash_list : l);
678}
679
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000680/*
681 * Callback used by the exception stores to load exceptions when
682 * initialising.
683 */
684static int dm_add_exception(void *context, chunk_t old, chunk_t new)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685{
Jonathan Brassowa159c1a2009-01-06 03:05:19 +0000686 struct dm_snapshot *s = context;
Jon Brassow1d4989c2009-12-10 23:52:10 +0000687 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688
Jon Brassow3510cb92009-12-10 23:52:11 +0000689 e = alloc_completed_exception();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 if (!e)
691 return -ENOMEM;
692
693 e->old_chunk = old;
Milan Brozd74f81f2008-02-08 02:11:27 +0000694
695 /* Consecutive_count is implicitly initialised to zero */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 e->new_chunk = new;
Milan Brozd74f81f2008-02-08 02:11:27 +0000697
Jon Brassow3510cb92009-12-10 23:52:11 +0000698 dm_insert_exception(&s->complete, e);
Milan Brozd74f81f2008-02-08 02:11:27 +0000699
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 return 0;
701}
702
Mikulas Patocka7e201b32009-12-10 23:52:08 +0000703#define min_not_zero(l, r) (((l) == 0) ? (r) : (((r) == 0) ? (l) : min(l, r)))
704
705/*
706 * Return a minimum chunk size of all snapshots that have the specified origin.
707 * Return zero if the origin has no snapshots.
708 */
709static sector_t __minimum_chunk_size(struct origin *o)
710{
711 struct dm_snapshot *snap;
712 unsigned chunk_size = 0;
713
714 if (o)
715 list_for_each_entry(snap, &o->snapshots, list)
716 chunk_size = min_not_zero(chunk_size,
717 snap->store->chunk_size);
718
719 return chunk_size;
720}
721
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722/*
723 * Hard coded magic.
724 */
725static int calc_max_buckets(void)
726{
727 /* use a fixed size of 2MB */
728 unsigned long mem = 2 * 1024 * 1024;
729 mem /= sizeof(struct list_head);
730
731 return mem;
732}
733
734/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 * Allocate room for a suitable hash table.
736 */
Jonathan Brassowfee19982009-04-02 19:55:34 +0100737static int init_hash_tables(struct dm_snapshot *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738{
739 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
740
741 /*
742 * Calculate based on the size of the original volume or
743 * the COW volume...
744 */
Mike Snitzerfc56f6f2009-12-10 23:52:12 +0000745 cow_dev_size = get_dev_size(s->cow->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 origin_dev_size = get_dev_size(s->origin->bdev);
747 max_buckets = calc_max_buckets();
748
Jonathan Brassowfee19982009-04-02 19:55:34 +0100749 hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 hash_size = min(hash_size, max_buckets);
751
Mikulas Patocka8e87b9b2009-12-10 23:51:54 +0000752 if (hash_size < 64)
753 hash_size = 64;
Robert P. J. Day8defd832008-02-08 02:10:06 +0000754 hash_size = rounddown_pow_of_two(hash_size);
Jon Brassow3510cb92009-12-10 23:52:11 +0000755 if (dm_exception_table_init(&s->complete, hash_size,
756 DM_CHUNK_CONSECUTIVE_BITS))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 return -ENOMEM;
758
759 /*
760 * Allocate hash table for in-flight exceptions
761 * Make this smaller than the real hash table
762 */
763 hash_size >>= 3;
764 if (hash_size < 64)
765 hash_size = 64;
766
Jon Brassow3510cb92009-12-10 23:52:11 +0000767 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
768 dm_exception_table_exit(&s->complete, exception_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 return -ENOMEM;
770 }
771
772 return 0;
773}
774
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000775static void merge_shutdown(struct dm_snapshot *s)
776{
777 clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
778 smp_mb__after_clear_bit();
779 wake_up_bit(&s->state_bits, RUNNING_MERGE);
780}
781
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000782static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
783{
784 s->first_merging_chunk = 0;
785 s->num_merging_chunks = 0;
786
787 return bio_list_get(&s->bios_queued_during_merge);
788}
789
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000790/*
791 * Remove one chunk from the index of completed exceptions.
792 */
793static int __remove_single_exception_chunk(struct dm_snapshot *s,
794 chunk_t old_chunk)
795{
796 struct dm_exception *e;
797
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000798 e = dm_lookup_exception(&s->complete, old_chunk);
799 if (!e) {
800 DMERR("Corruption detected: exception for block %llu is "
801 "on disk but not in memory",
802 (unsigned long long)old_chunk);
803 return -EINVAL;
804 }
805
806 /*
807 * If this is the only chunk using this exception, remove exception.
808 */
809 if (!dm_consecutive_chunk_count(e)) {
810 dm_remove_exception(e);
811 free_completed_exception(e);
812 return 0;
813 }
814
815 /*
816 * The chunk may be either at the beginning or the end of a
817 * group of consecutive chunks - never in the middle. We are
818 * removing chunks in the opposite order to that in which they
819 * were added, so this should always be true.
820 * Decrement the consecutive chunk counter and adjust the
821 * starting point if necessary.
822 */
823 if (old_chunk == e->old_chunk) {
824 e->old_chunk++;
825 e->new_chunk++;
826 } else if (old_chunk != e->old_chunk +
827 dm_consecutive_chunk_count(e)) {
828 DMERR("Attempt to merge block %llu from the "
829 "middle of a chunk range [%llu - %llu]",
830 (unsigned long long)old_chunk,
831 (unsigned long long)e->old_chunk,
832 (unsigned long long)
833 e->old_chunk + dm_consecutive_chunk_count(e));
834 return -EINVAL;
835 }
836
837 dm_consecutive_chunk_count_dec(e);
838
839 return 0;
840}
841
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000842static void flush_bios(struct bio *bio);
843
844static int remove_single_exception_chunk(struct dm_snapshot *s)
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000845{
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000846 struct bio *b = NULL;
847 int r;
848 chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000849
850 down_write(&s->lock);
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000851
852 /*
853 * Process chunks (and associated exceptions) in reverse order
854 * so that dm_consecutive_chunk_count_dec() accounting works.
855 */
856 do {
857 r = __remove_single_exception_chunk(s, old_chunk);
858 if (r)
859 goto out;
860 } while (old_chunk-- > s->first_merging_chunk);
861
862 b = __release_queued_bios_after_merge(s);
863
864out:
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000865 up_write(&s->lock);
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000866 if (b)
867 flush_bios(b);
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000868
869 return r;
870}
871
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000872static int origin_write_extent(struct dm_snapshot *merging_snap,
873 sector_t sector, unsigned chunk_size);
874
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000875static void merge_callback(int read_err, unsigned long write_err,
876 void *context);
877
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000878static uint64_t read_pending_exceptions_done_count(void)
879{
880 uint64_t pending_exceptions_done;
881
882 spin_lock(&_pending_exceptions_done_spinlock);
883 pending_exceptions_done = _pending_exceptions_done_count;
884 spin_unlock(&_pending_exceptions_done_spinlock);
885
886 return pending_exceptions_done;
887}
888
889static void increment_pending_exceptions_done_count(void)
890{
891 spin_lock(&_pending_exceptions_done_spinlock);
892 _pending_exceptions_done_count++;
893 spin_unlock(&_pending_exceptions_done_spinlock);
894
895 wake_up_all(&_pending_exceptions_done);
896}
897
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000898static void snapshot_merge_next_chunks(struct dm_snapshot *s)
899{
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000900 int i, linear_chunks;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000901 chunk_t old_chunk, new_chunk;
902 struct dm_io_region src, dest;
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000903 sector_t io_size;
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000904 uint64_t previous_count;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000905
906 BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
907 if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
908 goto shut;
909
910 /*
911 * valid flag never changes during merge, so no lock required.
912 */
913 if (!s->valid) {
914 DMERR("Snapshot is invalid: can't merge");
915 goto shut;
916 }
917
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000918 linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
919 &new_chunk);
920 if (linear_chunks <= 0) {
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +0000921 if (linear_chunks < 0) {
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000922 DMERR("Read error in exception store: "
923 "shutting down merge");
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +0000924 down_write(&s->lock);
925 s->merge_failed = 1;
926 up_write(&s->lock);
927 }
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000928 goto shut;
929 }
930
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000931 /* Adjust old_chunk and new_chunk to reflect start of linear region */
932 old_chunk = old_chunk + 1 - linear_chunks;
933 new_chunk = new_chunk + 1 - linear_chunks;
934
935 /*
936 * Use one (potentially large) I/O to copy all 'linear_chunks'
937 * from the exception store to the origin
938 */
939 io_size = linear_chunks * s->store->chunk_size;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000940
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000941 dest.bdev = s->origin->bdev;
942 dest.sector = chunk_to_sector(s->store, old_chunk);
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000943 dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000944
945 src.bdev = s->cow->bdev;
946 src.sector = chunk_to_sector(s->store, new_chunk);
947 src.count = dest.count;
948
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000949 /*
950 * Reallocate any exceptions needed in other snapshots then
951 * wait for the pending exceptions to complete.
952 * Each time any pending exception (globally on the system)
953 * completes we are woken and repeat the process to find out
954 * if we can proceed. While this may not seem a particularly
955 * efficient algorithm, it is not expected to have any
956 * significant impact on performance.
957 */
958 previous_count = read_pending_exceptions_done_count();
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000959 while (origin_write_extent(s, dest.sector, io_size)) {
Mikulas Patocka73dfd072009-12-10 23:52:34 +0000960 wait_event(_pending_exceptions_done,
961 (read_pending_exceptions_done_count() !=
962 previous_count));
963 /* Retry after the wait, until all exceptions are done. */
964 previous_count = read_pending_exceptions_done_count();
965 }
966
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000967 down_write(&s->lock);
968 s->first_merging_chunk = old_chunk;
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000969 s->num_merging_chunks = linear_chunks;
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000970 up_write(&s->lock);
971
Mike Snitzer8a2d5282009-12-10 23:52:34 +0000972 /* Wait until writes to all 'linear_chunks' drain */
973 for (i = 0; i < linear_chunks; i++)
974 __check_for_conflicting_io(s, old_chunk + i);
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000975
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000976 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
977 return;
978
979shut:
980 merge_shutdown(s);
981}
982
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000983static void error_bios(struct bio *bio);
984
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000985static void merge_callback(int read_err, unsigned long write_err, void *context)
986{
987 struct dm_snapshot *s = context;
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000988 struct bio *b = NULL;
Mikulas Patocka1e03f972009-12-10 23:52:32 +0000989
990 if (read_err || write_err) {
991 if (read_err)
992 DMERR("Read error: shutting down merge.");
993 else
994 DMERR("Write error: shutting down merge.");
995 goto shut;
996 }
997
Mikulas Patocka9fe862542009-12-10 23:52:33 +0000998 if (s->store->type->commit_merge(s->store,
999 s->num_merging_chunks) < 0) {
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001000 DMERR("Write error in exception store: shutting down merge");
1001 goto shut;
1002 }
1003
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001004 if (remove_single_exception_chunk(s) < 0)
1005 goto shut;
1006
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001007 snapshot_merge_next_chunks(s);
1008
1009 return;
1010
1011shut:
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001012 down_write(&s->lock);
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +00001013 s->merge_failed = 1;
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001014 b = __release_queued_bios_after_merge(s);
1015 up_write(&s->lock);
1016 error_bios(b);
1017
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001018 merge_shutdown(s);
1019}
1020
1021static void start_merge(struct dm_snapshot *s)
1022{
1023 if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
1024 snapshot_merge_next_chunks(s);
1025}
1026
1027static int wait_schedule(void *ptr)
1028{
1029 schedule();
1030
1031 return 0;
1032}
1033
1034/*
1035 * Stop the merging process and wait until it finishes.
1036 */
1037static void stop_merge(struct dm_snapshot *s)
1038{
1039 set_bit(SHUTDOWN_MERGE, &s->state_bits);
1040 wait_on_bit(&s->state_bits, RUNNING_MERGE, wait_schedule,
1041 TASK_UNINTERRUPTIBLE);
1042 clear_bit(SHUTDOWN_MERGE, &s->state_bits);
1043}
1044
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
1047 */
1048static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1049{
1050 struct dm_snapshot *s;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001051 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 int r = -EINVAL;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001053 char *origin_path, *cow_path;
Mike Snitzer10b81062009-12-10 23:52:31 +00001054 unsigned args_used, num_flush_requests = 1;
1055 fmode_t origin_mode = FMODE_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -07001057 if (argc != 4) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001058 ti->error = "requires exactly 4 arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 r = -EINVAL;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001060 goto bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 }
1062
Mike Snitzer10b81062009-12-10 23:52:31 +00001063 if (dm_target_is_snapshot_merge(ti)) {
1064 num_flush_requests = 2;
1065 origin_mode = FMODE_WRITE;
1066 }
1067
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 origin_path = argv[0];
Jonathan Brassowfee19982009-04-02 19:55:34 +01001069 argv++;
1070 argc--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 s = kmalloc(sizeof(*s), GFP_KERNEL);
Jonathan Brassowfee19982009-04-02 19:55:34 +01001073 if (!s) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 ti->error = "Cannot allocate snapshot context private "
1075 "structure";
1076 r = -ENOMEM;
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001077 goto bad;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 }
1079
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001080 cow_path = argv[0];
1081 argv++;
1082 argc--;
1083
Nikanth Karthikesan8215d6e2010-03-06 02:32:27 +00001084 r = dm_get_device(ti, cow_path, FMODE_READ | FMODE_WRITE, &s->cow);
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001085 if (r) {
1086 ti->error = "Cannot get COW device";
1087 goto bad_cow;
1088 }
1089
1090 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
1091 if (r) {
1092 ti->error = "Couldn't create exception store";
1093 r = -EINVAL;
1094 goto bad_store;
1095 }
1096
1097 argv += args_used;
1098 argc -= args_used;
1099
Nikanth Karthikesan8215d6e2010-03-06 02:32:27 +00001100 r = dm_get_device(ti, origin_path, origin_mode, &s->origin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 if (r) {
1102 ti->error = "Cannot get origin device";
Jonathan Brassowfee19982009-04-02 19:55:34 +01001103 goto bad_origin;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 }
1105
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001106 s->ti = ti;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 s->valid = 1;
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001108 s->active = 0;
Mike Snitzerc26655c2009-12-10 23:52:12 +00001109 s->suspended = 0;
Mikulas Patocka879129d22008-10-30 13:33:16 +00001110 atomic_set(&s->pending_exceptions_count, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 init_rwsem(&s->lock);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001112 INIT_LIST_HEAD(&s->list);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07001113 spin_lock_init(&s->pe_lock);
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001114 s->state_bits = 0;
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +00001115 s->merge_failed = 0;
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001116 s->first_merging_chunk = 0;
1117 s->num_merging_chunks = 0;
1118 bio_list_init(&s->bios_queued_during_merge);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
1120 /* Allocate hash table for COW data */
Jonathan Brassowfee19982009-04-02 19:55:34 +01001121 if (init_hash_tables(s)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 ti->error = "Unable to allocate hash table space";
1123 r = -ENOMEM;
Jonathan Brassowfee19982009-04-02 19:55:34 +01001124 goto bad_hash_tables;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 }
1126
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +01001127 r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 if (r) {
1129 ti->error = "Could not create kcopyd client";
Jonathan Brassowfee19982009-04-02 19:55:34 +01001130 goto bad_kcopyd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 }
1132
Mikulas Patocka92e86812008-07-21 12:00:35 +01001133 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
1134 if (!s->pending_pool) {
1135 ti->error = "Could not allocate mempool for pending exceptions";
Jonathan Brassowfee19982009-04-02 19:55:34 +01001136 goto bad_pending_pool;
Mikulas Patocka92e86812008-07-21 12:00:35 +01001137 }
1138
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001139 s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS,
1140 tracked_chunk_cache);
1141 if (!s->tracked_chunk_pool) {
1142 ti->error = "Could not allocate tracked_chunk mempool for "
1143 "tracking reads";
Mikulas Patocka92e86812008-07-21 12:00:35 +01001144 goto bad_tracked_chunk_pool;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001145 }
1146
1147 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1148 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
1149
1150 spin_lock_init(&s->tracked_chunk_lock);
1151
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001152 bio_list_init(&s->queued_bios);
1153 INIT_WORK(&s->queued_bios_work, flush_queued_bios);
1154
1155 ti->private = s;
Mike Snitzer10b81062009-12-10 23:52:31 +00001156 ti->num_flush_requests = num_flush_requests;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001157
1158 /* Add snapshot to the list of snapshots for this origin */
1159 /* Exceptions aren't triggered till snapshot_resume() is called */
1160 r = register_snapshot(s);
1161 if (r == -ENOMEM) {
1162 ti->error = "Snapshot origin struct allocation failed";
1163 goto bad_load_and_register;
1164 } else if (r < 0) {
1165 /* invalid handover, register_snapshot has set ti->error */
1166 goto bad_load_and_register;
1167 }
1168
1169 /*
1170 * Metadata must only be loaded into one table at once, so skip this
1171 * if metadata will be handed over during resume.
1172 * Chunk size will be set during the handover - set it to zero to
1173 * ensure it's ignored.
1174 */
1175 if (r > 0) {
1176 s->store->chunk_size = 0;
1177 return 0;
1178 }
1179
Jonathan Brassow493df712009-04-02 19:55:31 +01001180 r = s->store->type->read_metadata(s->store, dm_add_exception,
1181 (void *)s);
Milan Broz07641472007-07-12 17:28:13 +01001182 if (r < 0) {
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -07001183 ti->error = "Failed to read snapshot metadata";
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001184 goto bad_read_metadata;
Milan Broz07641472007-07-12 17:28:13 +01001185 } else if (r > 0) {
1186 s->valid = 0;
1187 DMWARN("Snapshot is marked invalid.");
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -07001188 }
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001189
Mikulas Patocka3f2412d2009-10-16 23:18:16 +01001190 if (!s->store->chunk_size) {
1191 ti->error = "Chunk size not set";
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001192 goto bad_read_metadata;
Mikulas Patocka3f2412d2009-10-16 23:18:16 +01001193 }
Jonathan Brassowd0216842009-04-02 19:55:32 +01001194 ti->split_io = s->store->chunk_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195
1196 return 0;
1197
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001198bad_read_metadata:
1199 unregister_snapshot(s);
1200
Jonathan Brassowfee19982009-04-02 19:55:34 +01001201bad_load_and_register:
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001202 mempool_destroy(s->tracked_chunk_pool);
1203
Jonathan Brassowfee19982009-04-02 19:55:34 +01001204bad_tracked_chunk_pool:
Mikulas Patocka92e86812008-07-21 12:00:35 +01001205 mempool_destroy(s->pending_pool);
1206
Jonathan Brassowfee19982009-04-02 19:55:34 +01001207bad_pending_pool:
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +01001208 dm_kcopyd_client_destroy(s->kcopyd_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209
Jonathan Brassowfee19982009-04-02 19:55:34 +01001210bad_kcopyd:
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);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213
Jonathan Brassowfee19982009-04-02 19:55:34 +01001214bad_hash_tables:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 dm_put_device(ti, s->origin);
1216
Jonathan Brassowfee19982009-04-02 19:55:34 +01001217bad_origin:
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001218 dm_exception_store_destroy(s->store);
1219
1220bad_store:
1221 dm_put_device(ti, s->cow);
1222
1223bad_cow:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 kfree(s);
1225
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001226bad:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 return r;
1228}
1229
Milan Broz31c93a02006-12-08 02:41:11 -08001230static void __free_exceptions(struct dm_snapshot *s)
1231{
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +01001232 dm_kcopyd_client_destroy(s->kcopyd_client);
Milan Broz31c93a02006-12-08 02:41:11 -08001233 s->kcopyd_client = NULL;
1234
Jon Brassow3510cb92009-12-10 23:52:11 +00001235 dm_exception_table_exit(&s->pending, pending_cache);
1236 dm_exception_table_exit(&s->complete, exception_cache);
Milan Broz31c93a02006-12-08 02:41:11 -08001237}
1238
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001239static void __handover_exceptions(struct dm_snapshot *snap_src,
1240 struct dm_snapshot *snap_dest)
1241{
1242 union {
1243 struct dm_exception_table table_swap;
1244 struct dm_exception_store *store_swap;
1245 } u;
1246
1247 /*
1248 * Swap all snapshot context information between the two instances.
1249 */
1250 u.table_swap = snap_dest->complete;
1251 snap_dest->complete = snap_src->complete;
1252 snap_src->complete = u.table_swap;
1253
1254 u.store_swap = snap_dest->store;
1255 snap_dest->store = snap_src->store;
1256 snap_src->store = u.store_swap;
1257
1258 snap_dest->store->snap = snap_dest;
1259 snap_src->store->snap = snap_src;
1260
1261 snap_dest->ti->split_io = snap_dest->store->chunk_size;
1262 snap_dest->valid = snap_src->valid;
1263
1264 /*
1265 * Set source invalid to ensure it receives no further I/O.
1266 */
1267 snap_src->valid = 0;
1268}
1269
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270static void snapshot_dtr(struct dm_target *ti)
1271{
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001272#ifdef CONFIG_DM_DEBUG
1273 int i;
1274#endif
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001275 struct dm_snapshot *s = ti->private;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001276 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07001278 flush_workqueue(ksnapd);
1279
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001280 down_read(&_origins_lock);
1281 /* Check whether exception handover must be cancelled */
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +00001282 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001283 if (snap_src && snap_dest && (s == snap_src)) {
1284 down_write(&snap_dest->lock);
1285 snap_dest->valid = 0;
1286 up_write(&snap_dest->lock);
1287 DMERR("Cancelling snapshot handover.");
1288 }
1289 up_read(&_origins_lock);
1290
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001291 if (dm_target_is_snapshot_merge(ti))
1292 stop_merge(s);
1293
Alasdair G Kergon138728dc2006-03-27 01:17:50 -08001294 /* Prevent further origin writes from using this snapshot. */
1295 /* After this returns there can be no new kcopyd jobs. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 unregister_snapshot(s);
1297
Mikulas Patocka879129d22008-10-30 13:33:16 +00001298 while (atomic_read(&s->pending_exceptions_count))
Mikulas Patocka90fa1522009-01-06 03:04:54 +00001299 msleep(1);
Mikulas Patocka879129d22008-10-30 13:33:16 +00001300 /*
1301 * Ensure instructions in mempool_destroy aren't reordered
1302 * before atomic_read.
1303 */
1304 smp_mb();
1305
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001306#ifdef CONFIG_DM_DEBUG
1307 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1308 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
1309#endif
1310
1311 mempool_destroy(s->tracked_chunk_pool);
1312
Milan Broz31c93a02006-12-08 02:41:11 -08001313 __free_exceptions(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
Mikulas Patocka92e86812008-07-21 12:00:35 +01001315 mempool_destroy(s->pending_pool);
1316
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 dm_put_device(ti, s->origin);
Jonathan Brassowfee19982009-04-02 19:55:34 +01001318
1319 dm_exception_store_destroy(s->store);
Alasdair G Kergon138728dc2006-03-27 01:17:50 -08001320
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001321 dm_put_device(ti, s->cow);
1322
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 kfree(s);
1324}
1325
1326/*
1327 * Flush a list of buffers.
1328 */
1329static void flush_bios(struct bio *bio)
1330{
1331 struct bio *n;
1332
1333 while (bio) {
1334 n = bio->bi_next;
1335 bio->bi_next = NULL;
1336 generic_make_request(bio);
1337 bio = n;
1338 }
1339}
1340
David Howellsc4028952006-11-22 14:57:56 +00001341static void flush_queued_bios(struct work_struct *work)
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07001342{
David Howellsc4028952006-11-22 14:57:56 +00001343 struct dm_snapshot *s =
1344 container_of(work, struct dm_snapshot, queued_bios_work);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07001345 struct bio *queued_bios;
1346 unsigned long flags;
1347
1348 spin_lock_irqsave(&s->pe_lock, flags);
1349 queued_bios = bio_list_get(&s->queued_bios);
1350 spin_unlock_irqrestore(&s->pe_lock, flags);
1351
1352 flush_bios(queued_bios);
1353}
1354
Mikulas Patocka515ad662009-12-10 23:52:30 +00001355static int do_origin(struct dm_dev *origin, struct bio *bio);
1356
1357/*
1358 * Flush a list of buffers.
1359 */
1360static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1361{
1362 struct bio *n;
1363 int r;
1364
1365 while (bio) {
1366 n = bio->bi_next;
1367 bio->bi_next = NULL;
1368 r = do_origin(s->origin, bio);
1369 if (r == DM_MAPIO_REMAPPED)
1370 generic_make_request(bio);
1371 bio = n;
1372 }
1373}
1374
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375/*
1376 * Error a list of buffers.
1377 */
1378static void error_bios(struct bio *bio)
1379{
1380 struct bio *n;
1381
1382 while (bio) {
1383 n = bio->bi_next;
1384 bio->bi_next = NULL;
NeilBrown6712ecf2007-09-27 12:47:43 +02001385 bio_io_error(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 bio = n;
1387 }
1388}
1389
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001390static void __invalidate_snapshot(struct dm_snapshot *s, int err)
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001391{
1392 if (!s->valid)
1393 return;
1394
1395 if (err == -EIO)
1396 DMERR("Invalidating snapshot: Error reading/writing.");
1397 else if (err == -ENOMEM)
1398 DMERR("Invalidating snapshot: Unable to allocate exception.");
1399
Jonathan Brassow493df712009-04-02 19:55:31 +01001400 if (s->store->type->drop_snapshot)
1401 s->store->type->drop_snapshot(s->store);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001402
1403 s->valid = 0;
1404
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001405 dm_table_event(s->ti->table);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001406}
1407
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001408static void pending_complete(struct dm_snap_pending_exception *pe, int success)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409{
Jon Brassow1d4989c2009-12-10 23:52:10 +00001410 struct dm_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 struct dm_snapshot *s = pe->snap;
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001412 struct bio *origin_bios = NULL;
1413 struct bio *snapshot_bios = NULL;
1414 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001416 if (!success) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 /* Read/write error - snapshot is unusable */
1418 down_write(&s->lock);
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001419 __invalidate_snapshot(s, -EIO);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001420 error = 1;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001421 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 }
1423
Jon Brassow3510cb92009-12-10 23:52:11 +00001424 e = alloc_completed_exception();
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001425 if (!e) {
1426 down_write(&s->lock);
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001427 __invalidate_snapshot(s, -ENOMEM);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001428 error = 1;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001429 goto out;
1430 }
1431 *e = pe->e;
1432
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001433 down_write(&s->lock);
1434 if (!s->valid) {
Jon Brassow3510cb92009-12-10 23:52:11 +00001435 free_completed_exception(e);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001436 error = 1;
1437 goto out;
1438 }
1439
Mike Snitzer615d1eb2009-12-10 23:52:29 +00001440 /* Check for conflicting reads */
1441 __check_for_conflicting_io(s, pe->e.old_chunk);
Mikulas Patockaa8d41b52008-07-21 12:00:34 +01001442
1443 /*
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001444 * Add a proper exception, and remove the
1445 * in-flight exception from the list.
1446 */
Jon Brassow3510cb92009-12-10 23:52:11 +00001447 dm_insert_exception(&s->complete, e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001448
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 out:
Jon Brassow3510cb92009-12-10 23:52:11 +00001450 dm_remove_exception(&pe->e);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001451 snapshot_bios = bio_list_get(&pe->snapshot_bios);
Mikulas Patocka515ad662009-12-10 23:52:30 +00001452 origin_bios = bio_list_get(&pe->origin_bios);
1453 free_pending_exception(pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454
Mikulas Patocka73dfd072009-12-10 23:52:34 +00001455 increment_pending_exceptions_done_count();
1456
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -07001457 up_write(&s->lock);
1458
1459 /* Submit any pending write bios */
1460 if (error)
1461 error_bios(snapshot_bios);
1462 else
1463 flush_bios(snapshot_bios);
1464
Mikulas Patocka515ad662009-12-10 23:52:30 +00001465 retry_origin_bios(s, origin_bios);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466}
1467
1468static void commit_callback(void *context, int success)
1469{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001470 struct dm_snap_pending_exception *pe = context;
1471
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 pending_complete(pe, success);
1473}
1474
1475/*
1476 * Called when the copy I/O has finished. kcopyd actually runs
1477 * this code so don't block.
1478 */
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -07001479static void copy_callback(int read_err, unsigned long write_err, void *context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001481 struct dm_snap_pending_exception *pe = context;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 struct dm_snapshot *s = pe->snap;
1483
1484 if (read_err || write_err)
1485 pending_complete(pe, 0);
1486
1487 else
1488 /* Update the metadata if we are persistent */
Jonathan Brassow493df712009-04-02 19:55:31 +01001489 s->store->type->commit_exception(s->store, &pe->e,
1490 commit_callback, pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491}
1492
1493/*
1494 * Dispatches the copy operation to kcopyd.
1495 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001496static void start_copy(struct dm_snap_pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497{
1498 struct dm_snapshot *s = pe->snap;
Heinz Mauelshagen22a1ceb2008-04-24 21:43:17 +01001499 struct dm_io_region src, dest;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 struct block_device *bdev = s->origin->bdev;
1501 sector_t dev_size;
1502
1503 dev_size = get_dev_size(bdev);
1504
1505 src.bdev = bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001506 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
Mikulas Patockadf96eee2009-10-16 23:18:17 +01001507 src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001509 dest.bdev = s->cow->bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001510 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 dest.count = src.count;
1512
1513 /* Hand over to kcopyd */
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +01001514 dm_kcopyd_copy(s->kcopyd_client,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 &src, 1, &dest, 0, copy_callback, pe);
1516}
1517
Mikulas Patocka29138082009-04-02 19:55:25 +01001518static struct dm_snap_pending_exception *
1519__lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1520{
Jon Brassow3510cb92009-12-10 23:52:11 +00001521 struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01001522
1523 if (!e)
1524 return NULL;
1525
1526 return container_of(e, struct dm_snap_pending_exception, e);
1527}
1528
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529/*
1530 * Looks to see if this snapshot already has a pending exception
1531 * for this chunk, otherwise it allocates a new one and inserts
1532 * it into the pending table.
1533 *
1534 * NOTE: a write lock must be held on snap->lock before calling
1535 * this.
1536 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001537static struct dm_snap_pending_exception *
Mikulas Patockac6621392009-04-02 19:55:25 +01001538__find_pending_exception(struct dm_snapshot *s,
1539 struct dm_snap_pending_exception *pe, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540{
Mikulas Patockac6621392009-04-02 19:55:25 +01001541 struct dm_snap_pending_exception *pe2;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001542
Mikulas Patocka29138082009-04-02 19:55:25 +01001543 pe2 = __lookup_pending_exception(s, chunk);
1544 if (pe2) {
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001545 free_pending_exception(pe);
Mikulas Patocka29138082009-04-02 19:55:25 +01001546 return pe2;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001547 }
1548
1549 pe->e.old_chunk = chunk;
1550 bio_list_init(&pe->origin_bios);
1551 bio_list_init(&pe->snapshot_bios);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001552 pe->started = 0;
1553
Jonathan Brassow493df712009-04-02 19:55:31 +01001554 if (s->store->type->prepare_exception(s->store, &pe->e)) {
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001555 free_pending_exception(pe);
1556 return NULL;
1557 }
1558
Jon Brassow3510cb92009-12-10 23:52:11 +00001559 dm_insert_exception(&s->pending, &pe->e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001560
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 return pe;
1562}
1563
Jon Brassow1d4989c2009-12-10 23:52:10 +00001564static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
Milan Brozd74f81f2008-02-08 02:11:27 +00001565 struct bio *bio, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566{
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001567 bio->bi_bdev = s->cow->bdev;
Jonathan Brassow71fab002009-04-02 19:55:33 +01001568 bio->bi_sector = chunk_to_sector(s->store,
1569 dm_chunk_number(e->new_chunk) +
1570 (chunk - e->old_chunk)) +
1571 (bio->bi_sector &
1572 s->store->chunk_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573}
1574
1575static int snapshot_map(struct dm_target *ti, struct bio *bio,
1576 union map_info *map_context)
1577{
Jon Brassow1d4989c2009-12-10 23:52:10 +00001578 struct dm_exception *e;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001579 struct dm_snapshot *s = ti->private;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001580 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 chunk_t chunk;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001582 struct dm_snap_pending_exception *pe = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01001584 if (unlikely(bio_empty_barrier(bio))) {
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001585 bio->bi_bdev = s->cow->bdev;
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01001586 return DM_MAPIO_REMAPPED;
1587 }
1588
Jonathan Brassow71fab002009-04-02 19:55:33 +01001589 chunk = sector_to_chunk(s->store, bio->bi_sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590
1591 /* Full snapshots are not usable */
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001592 /* To get here the table must be live so s->active is always set. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 if (!s->valid)
Alasdair G Kergonf6a80ea2005-07-12 15:53:01 -07001594 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001596 /* FIXME: should only take write lock if we need
1597 * to copy an exception */
1598 down_write(&s->lock);
1599
1600 if (!s->valid) {
1601 r = -EIO;
1602 goto out_unlock;
1603 }
1604
1605 /* If the block is already remapped - use that, else remap it */
Jon Brassow3510cb92009-12-10 23:52:11 +00001606 e = dm_lookup_exception(&s->complete, chunk);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001607 if (e) {
Milan Brozd74f81f2008-02-08 02:11:27 +00001608 remap_exception(s, e, bio, chunk);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001609 goto out_unlock;
1610 }
1611
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 /*
1613 * Write to snapshot - higher level takes care of RW/RO
1614 * flags so we should only get this if we are
1615 * writeable.
1616 */
1617 if (bio_rw(bio) == WRITE) {
Mikulas Patocka29138082009-04-02 19:55:25 +01001618 pe = __lookup_pending_exception(s, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001619 if (!pe) {
Mikulas Patockac6621392009-04-02 19:55:25 +01001620 up_write(&s->lock);
1621 pe = alloc_pending_exception(s);
1622 down_write(&s->lock);
1623
1624 if (!s->valid) {
1625 free_pending_exception(pe);
1626 r = -EIO;
1627 goto out_unlock;
1628 }
1629
Jon Brassow3510cb92009-12-10 23:52:11 +00001630 e = dm_lookup_exception(&s->complete, chunk);
Mikulas Patocka35bf6592009-04-02 19:55:26 +01001631 if (e) {
1632 free_pending_exception(pe);
1633 remap_exception(s, e, bio, chunk);
1634 goto out_unlock;
1635 }
1636
Mikulas Patockac6621392009-04-02 19:55:25 +01001637 pe = __find_pending_exception(s, pe, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01001638 if (!pe) {
1639 __invalidate_snapshot(s, -ENOMEM);
1640 r = -EIO;
1641 goto out_unlock;
1642 }
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001643 }
1644
Milan Brozd74f81f2008-02-08 02:11:27 +00001645 remap_exception(s, &pe->e, bio, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001646 bio_list_add(&pe->snapshot_bios, bio);
1647
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001648 r = DM_MAPIO_SUBMITTED;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001649
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001650 if (!pe->started) {
1651 /* this is protected by snap->lock */
1652 pe->started = 1;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001653 up_write(&s->lock);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001654 start_copy(pe);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001655 goto out;
1656 }
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001657 } else {
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001658 bio->bi_bdev = s->origin->bdev;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001659 map_context->ptr = track_chunk(s, chunk);
1660 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -07001662 out_unlock:
1663 up_write(&s->lock);
1664 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 return r;
1666}
1667
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001668/*
1669 * A snapshot-merge target behaves like a combination of a snapshot
1670 * target and a snapshot-origin target. It only generates new
1671 * exceptions in other snapshots and not in the one that is being
1672 * merged.
1673 *
1674 * For each chunk, if there is an existing exception, it is used to
1675 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
1676 * which in turn might generate exceptions in other snapshots.
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001677 * If merging is currently taking place on the chunk in question, the
1678 * I/O is deferred by adding it to s->bios_queued_during_merge.
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001679 */
1680static int snapshot_merge_map(struct dm_target *ti, struct bio *bio,
1681 union map_info *map_context)
1682{
1683 struct dm_exception *e;
1684 struct dm_snapshot *s = ti->private;
1685 int r = DM_MAPIO_REMAPPED;
1686 chunk_t chunk;
1687
Mike Snitzer10b81062009-12-10 23:52:31 +00001688 if (unlikely(bio_empty_barrier(bio))) {
1689 if (!map_context->flush_request)
1690 bio->bi_bdev = s->origin->bdev;
1691 else
1692 bio->bi_bdev = s->cow->bdev;
1693 map_context->ptr = NULL;
1694 return DM_MAPIO_REMAPPED;
1695 }
1696
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001697 chunk = sector_to_chunk(s->store, bio->bi_sector);
1698
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001699 down_write(&s->lock);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001700
Mikulas Patockad2fdb772009-12-10 23:52:36 +00001701 /* Full merging snapshots are redirected to the origin */
1702 if (!s->valid)
1703 goto redirect_to_origin;
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001704
1705 /* If the block is already remapped - use that */
1706 e = dm_lookup_exception(&s->complete, chunk);
1707 if (e) {
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001708 /* Queue writes overlapping with chunks being merged */
1709 if (bio_rw(bio) == WRITE &&
1710 chunk >= s->first_merging_chunk &&
1711 chunk < (s->first_merging_chunk +
1712 s->num_merging_chunks)) {
1713 bio->bi_bdev = s->origin->bdev;
1714 bio_list_add(&s->bios_queued_during_merge, bio);
1715 r = DM_MAPIO_SUBMITTED;
1716 goto out_unlock;
1717 }
Mikulas Patocka17aa0332009-12-10 23:52:33 +00001718
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001719 remap_exception(s, e, bio, chunk);
Mikulas Patocka17aa0332009-12-10 23:52:33 +00001720
1721 if (bio_rw(bio) == WRITE)
1722 map_context->ptr = track_chunk(s, chunk);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001723 goto out_unlock;
1724 }
1725
Mikulas Patockad2fdb772009-12-10 23:52:36 +00001726redirect_to_origin:
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001727 bio->bi_bdev = s->origin->bdev;
1728
1729 if (bio_rw(bio) == WRITE) {
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001730 up_write(&s->lock);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001731 return do_origin(s->origin, bio);
1732 }
1733
1734out_unlock:
Mikulas Patocka9fe862542009-12-10 23:52:33 +00001735 up_write(&s->lock);
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001736
1737 return r;
1738}
1739
Mikulas Patockacd45daf2008-07-21 12:00:32 +01001740static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1741 int error, union map_info *map_context)
1742{
1743 struct dm_snapshot *s = ti->private;
1744 struct dm_snap_tracked_chunk *c = map_context->ptr;
1745
1746 if (c)
1747 stop_tracking_chunk(s, c);
1748
1749 return 0;
1750}
1751
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001752static void snapshot_merge_presuspend(struct dm_target *ti)
1753{
1754 struct dm_snapshot *s = ti->private;
1755
1756 stop_merge(s);
1757}
1758
Mike Snitzerc26655c2009-12-10 23:52:12 +00001759static void snapshot_postsuspend(struct dm_target *ti)
1760{
1761 struct dm_snapshot *s = ti->private;
1762
1763 down_write(&s->lock);
1764 s->suspended = 1;
1765 up_write(&s->lock);
1766}
1767
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001768static int snapshot_preresume(struct dm_target *ti)
1769{
1770 int r = 0;
1771 struct dm_snapshot *s = ti->private;
1772 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1773
1774 down_read(&_origins_lock);
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +00001775 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001776 if (snap_src && snap_dest) {
1777 down_read(&snap_src->lock);
1778 if (s == snap_src) {
1779 DMERR("Unable to resume snapshot source until "
1780 "handover completes.");
1781 r = -EINVAL;
1782 } else if (!snap_src->suspended) {
1783 DMERR("Unable to perform snapshot handover until "
1784 "source is suspended.");
1785 r = -EINVAL;
1786 }
1787 up_read(&snap_src->lock);
1788 }
1789 up_read(&_origins_lock);
1790
1791 return r;
1792}
1793
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794static void snapshot_resume(struct dm_target *ti)
1795{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001796 struct dm_snapshot *s = ti->private;
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001797 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1798
1799 down_read(&_origins_lock);
Mikulas Patocka9d3b15c2009-12-10 23:52:32 +00001800 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
Mike Snitzerc1f0c182009-12-10 23:52:24 +00001801 if (snap_src && snap_dest) {
1802 down_write(&snap_src->lock);
1803 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1804 __handover_exceptions(snap_src, snap_dest);
1805 up_write(&snap_dest->lock);
1806 up_write(&snap_src->lock);
1807 }
1808 up_read(&_origins_lock);
1809
1810 /* Now we have correct chunk size, reregister */
1811 reregister_snapshot(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001813 down_write(&s->lock);
1814 s->active = 1;
Mike Snitzerc26655c2009-12-10 23:52:12 +00001815 s->suspended = 0;
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001816 up_write(&s->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817}
1818
Mikulas Patocka1e03f972009-12-10 23:52:32 +00001819static sector_t get_origin_minimum_chunksize(struct block_device *bdev)
1820{
1821 sector_t min_chunksize;
1822
1823 down_read(&_origins_lock);
1824 min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
1825 up_read(&_origins_lock);
1826
1827 return min_chunksize;
1828}
1829
1830static void snapshot_merge_resume(struct dm_target *ti)
1831{
1832 struct dm_snapshot *s = ti->private;
1833
1834 /*
1835 * Handover exceptions from existing snapshot.
1836 */
1837 snapshot_resume(ti);
1838
1839 /*
1840 * snapshot-merge acts as an origin, so set ti->split_io
1841 */
1842 ti->split_io = get_origin_minimum_chunksize(s->origin->bdev);
1843
1844 start_merge(s);
1845}
1846
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847static int snapshot_status(struct dm_target *ti, status_type_t type,
1848 char *result, unsigned int maxlen)
1849{
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01001850 unsigned sz = 0;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001851 struct dm_snapshot *snap = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852
1853 switch (type) {
1854 case STATUSTYPE_INFO:
Mikulas Patocka94e765722009-12-10 23:51:53 +00001855
1856 down_write(&snap->lock);
1857
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 if (!snap->valid)
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01001859 DMEMIT("Invalid");
Mike Snitzerd8ddb1c2009-12-10 23:52:35 +00001860 else if (snap->merge_failed)
1861 DMEMIT("Merge failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 else {
Mike Snitzer985903b2009-12-10 23:52:11 +00001863 if (snap->store->type->usage) {
1864 sector_t total_sectors, sectors_allocated,
1865 metadata_sectors;
1866 snap->store->type->usage(snap->store,
1867 &total_sectors,
1868 &sectors_allocated,
1869 &metadata_sectors);
1870 DMEMIT("%llu/%llu %llu",
1871 (unsigned long long)sectors_allocated,
1872 (unsigned long long)total_sectors,
1873 (unsigned long long)metadata_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 }
1875 else
Jonathan Brassow2e4a31d2009-04-02 19:55:34 +01001876 DMEMIT("Unknown");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 }
Mikulas Patocka94e765722009-12-10 23:51:53 +00001878
1879 up_write(&snap->lock);
1880
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 break;
1882
1883 case STATUSTYPE_TABLE:
1884 /*
1885 * kdevname returns a static pointer so we need
1886 * to make private copies if the output is to
1887 * make sense.
1888 */
Mike Snitzerfc56f6f2009-12-10 23:52:12 +00001889 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
Jonathan Brassow1e302a92009-04-02 19:55:35 +01001890 snap->store->type->status(snap->store, type, result + sz,
1891 maxlen - sz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 break;
1893 }
1894
1895 return 0;
1896}
1897
Mike Snitzer8811f462009-09-04 20:40:19 +01001898static int snapshot_iterate_devices(struct dm_target *ti,
1899 iterate_devices_callout_fn fn, void *data)
1900{
1901 struct dm_snapshot *snap = ti->private;
1902
1903 return fn(ti, snap->origin, 0, ti->len, data);
1904}
1905
1906
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907/*-----------------------------------------------------------------
1908 * Origin methods
1909 *---------------------------------------------------------------*/
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001910
1911/*
1912 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1913 * supplied bio was ignored. The caller may submit it immediately.
1914 * (No remapping actually occurs as the origin is always a direct linear
1915 * map.)
1916 *
1917 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1918 * and any supplied bio is added to a list to be submitted once all
1919 * the necessary exceptions exist.
1920 */
1921static int __origin_write(struct list_head *snapshots, sector_t sector,
1922 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923{
Mikulas Patocka515ad662009-12-10 23:52:30 +00001924 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 struct dm_snapshot *snap;
Jon Brassow1d4989c2009-12-10 23:52:10 +00001926 struct dm_exception *e;
Mikulas Patocka515ad662009-12-10 23:52:30 +00001927 struct dm_snap_pending_exception *pe;
1928 struct dm_snap_pending_exception *pe_to_start_now = NULL;
1929 struct dm_snap_pending_exception *pe_to_start_last = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 chunk_t chunk;
1931
1932 /* Do all the snapshots on this origin */
1933 list_for_each_entry (snap, snapshots, list) {
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00001934 /*
1935 * Don't make new exceptions in a merging snapshot
1936 * because it has effectively been deleted
1937 */
1938 if (dm_target_is_snapshot_merge(snap->ti))
1939 continue;
1940
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001941 down_write(&snap->lock);
1942
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001943 /* Only deal with valid and active snapshots */
1944 if (!snap->valid || !snap->active)
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001945 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946
Alasdair G Kergond5e404c2005-07-12 15:53:05 -07001947 /* Nothing to do if writing beyond end of snapshot */
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001948 if (sector >= dm_table_get_size(snap->ti->table))
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001949 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950
1951 /*
1952 * Remember, different snapshots can have
1953 * different chunk sizes.
1954 */
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00001955 chunk = sector_to_chunk(snap->store, sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956
1957 /*
1958 * Check exception table to see if block
1959 * is already remapped in this snapshot
1960 * and trigger an exception if not.
1961 */
Jon Brassow3510cb92009-12-10 23:52:11 +00001962 e = dm_lookup_exception(&snap->complete, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001963 if (e)
1964 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965
Mikulas Patocka29138082009-04-02 19:55:25 +01001966 pe = __lookup_pending_exception(snap, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001967 if (!pe) {
Mikulas Patockac6621392009-04-02 19:55:25 +01001968 up_write(&snap->lock);
1969 pe = alloc_pending_exception(snap);
1970 down_write(&snap->lock);
1971
1972 if (!snap->valid) {
1973 free_pending_exception(pe);
1974 goto next_snapshot;
1975 }
1976
Jon Brassow3510cb92009-12-10 23:52:11 +00001977 e = dm_lookup_exception(&snap->complete, chunk);
Mikulas Patocka35bf6592009-04-02 19:55:26 +01001978 if (e) {
1979 free_pending_exception(pe);
1980 goto next_snapshot;
1981 }
1982
Mikulas Patockac6621392009-04-02 19:55:25 +01001983 pe = __find_pending_exception(snap, pe, chunk);
Mikulas Patocka29138082009-04-02 19:55:25 +01001984 if (!pe) {
1985 __invalidate_snapshot(snap, -ENOMEM);
1986 goto next_snapshot;
1987 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 }
1989
Mikulas Patocka515ad662009-12-10 23:52:30 +00001990 r = DM_MAPIO_SUBMITTED;
1991
1992 /*
1993 * If an origin bio was supplied, queue it to wait for the
1994 * completion of this exception, and start this one last,
1995 * at the end of the function.
1996 */
1997 if (bio) {
1998 bio_list_add(&pe->origin_bios, bio);
1999 bio = NULL;
2000
2001 if (!pe->started) {
2002 pe->started = 1;
2003 pe_to_start_last = pe;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002004 }
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002005 }
2006
2007 if (!pe->started) {
2008 pe->started = 1;
Mikulas Patocka515ad662009-12-10 23:52:30 +00002009 pe_to_start_now = pe;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08002010 }
2011
2012 next_snapshot:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 up_write(&snap->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014
Mikulas Patocka515ad662009-12-10 23:52:30 +00002015 if (pe_to_start_now) {
2016 start_copy(pe_to_start_now);
2017 pe_to_start_now = NULL;
2018 }
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08002019 }
2020
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 /*
Mikulas Patocka515ad662009-12-10 23:52:30 +00002022 * Submit the exception against which the bio is queued last,
2023 * to give the other exceptions a head start.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 */
Mikulas Patocka515ad662009-12-10 23:52:30 +00002025 if (pe_to_start_last)
2026 start_copy(pe_to_start_last);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027
2028 return r;
2029}
2030
2031/*
2032 * Called on a write from the origin driver.
2033 */
2034static int do_origin(struct dm_dev *origin, struct bio *bio)
2035{
2036 struct origin *o;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08002037 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038
2039 down_read(&_origins_lock);
2040 o = __lookup_origin(origin->bdev);
2041 if (o)
Mikulas Patocka9eaae8f2009-12-10 23:52:28 +00002042 r = __origin_write(&o->snapshots, bio->bi_sector, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 up_read(&_origins_lock);
2044
2045 return r;
2046}
2047
2048/*
Mikulas Patocka73dfd072009-12-10 23:52:34 +00002049 * Trigger exceptions in all non-merging snapshots.
2050 *
2051 * The chunk size of the merging snapshot may be larger than the chunk
2052 * size of some other snapshot so we may need to reallocate multiple
2053 * chunks in other snapshots.
2054 *
2055 * We scan all the overlapping exceptions in the other snapshots.
2056 * Returns 1 if anything was reallocated and must be waited for,
2057 * otherwise returns 0.
2058 *
2059 * size must be a multiple of merging_snap's chunk_size.
2060 */
2061static int origin_write_extent(struct dm_snapshot *merging_snap,
2062 sector_t sector, unsigned size)
2063{
2064 int must_wait = 0;
2065 sector_t n;
2066 struct origin *o;
2067
2068 /*
2069 * The origin's __minimum_chunk_size() got stored in split_io
2070 * by snapshot_merge_resume().
2071 */
2072 down_read(&_origins_lock);
2073 o = __lookup_origin(merging_snap->origin->bdev);
2074 for (n = 0; n < size; n += merging_snap->ti->split_io)
2075 if (__origin_write(&o->snapshots, sector + n, NULL) ==
2076 DM_MAPIO_SUBMITTED)
2077 must_wait = 1;
2078 up_read(&_origins_lock);
2079
2080 return must_wait;
2081}
2082
2083/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 * Origin: maps a linear range of a device, with hooks for snapshotting.
2085 */
2086
2087/*
2088 * Construct an origin mapping: <dev_path>
2089 * The context for an origin is merely a 'struct dm_dev *'
2090 * pointing to the real device.
2091 */
2092static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2093{
2094 int r;
2095 struct dm_dev *dev;
2096
2097 if (argc != 1) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07002098 ti->error = "origin: incorrect number of arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 return -EINVAL;
2100 }
2101
Nikanth Karthikesan8215d6e2010-03-06 02:32:27 +00002102 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 if (r) {
2104 ti->error = "Cannot get target device";
2105 return r;
2106 }
2107
2108 ti->private = dev;
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01002109 ti->num_flush_requests = 1;
2110
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 return 0;
2112}
2113
2114static void origin_dtr(struct dm_target *ti)
2115{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002116 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 dm_put_device(ti, dev);
2118}
2119
2120static int origin_map(struct dm_target *ti, struct bio *bio,
2121 union map_info *map_context)
2122{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002123 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 bio->bi_bdev = dev->bdev;
2125
Mikulas Patocka494b3ee2009-06-22 10:12:25 +01002126 if (unlikely(bio_empty_barrier(bio)))
2127 return DM_MAPIO_REMAPPED;
2128
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 /* Only tell snapshots if this is a write */
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08002130 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131}
2132
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133/*
2134 * Set the target "split_io" field to the minimum of all the snapshots'
2135 * chunk sizes.
2136 */
2137static void origin_resume(struct dm_target *ti)
2138{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002139 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002141 ti->split_io = get_origin_minimum_chunksize(dev->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142}
2143
2144static int origin_status(struct dm_target *ti, status_type_t type, char *result,
2145 unsigned int maxlen)
2146{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002147 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148
2149 switch (type) {
2150 case STATUSTYPE_INFO:
2151 result[0] = '\0';
2152 break;
2153
2154 case STATUSTYPE_TABLE:
2155 snprintf(result, maxlen, "%s", dev->name);
2156 break;
2157 }
2158
2159 return 0;
2160}
2161
Mike Snitzer8811f462009-09-04 20:40:19 +01002162static int origin_iterate_devices(struct dm_target *ti,
2163 iterate_devices_callout_fn fn, void *data)
2164{
2165 struct dm_dev *dev = ti->private;
2166
2167 return fn(ti, dev, 0, ti->len, data);
2168}
2169
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170static struct target_type origin_target = {
2171 .name = "snapshot-origin",
Mike Snitzer8811f462009-09-04 20:40:19 +01002172 .version = {1, 7, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 .module = THIS_MODULE,
2174 .ctr = origin_ctr,
2175 .dtr = origin_dtr,
2176 .map = origin_map,
2177 .resume = origin_resume,
2178 .status = origin_status,
Mike Snitzer8811f462009-09-04 20:40:19 +01002179 .iterate_devices = origin_iterate_devices,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180};
2181
2182static struct target_type snapshot_target = {
2183 .name = "snapshot",
Mike Snitzerc26655c2009-12-10 23:52:12 +00002184 .version = {1, 9, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 .module = THIS_MODULE,
2186 .ctr = snapshot_ctr,
2187 .dtr = snapshot_dtr,
2188 .map = snapshot_map,
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002189 .end_io = snapshot_end_io,
Mike Snitzerc26655c2009-12-10 23:52:12 +00002190 .postsuspend = snapshot_postsuspend,
Mike Snitzerc1f0c182009-12-10 23:52:24 +00002191 .preresume = snapshot_preresume,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 .resume = snapshot_resume,
2193 .status = snapshot_status,
Mike Snitzer8811f462009-09-04 20:40:19 +01002194 .iterate_devices = snapshot_iterate_devices,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195};
2196
Mikulas Patockad698aa42009-12-10 23:52:30 +00002197static struct target_type merge_target = {
2198 .name = dm_snapshot_merge_target_name,
2199 .version = {1, 0, 0},
2200 .module = THIS_MODULE,
2201 .ctr = snapshot_ctr,
2202 .dtr = snapshot_dtr,
Mikulas Patocka3452c2a2009-12-10 23:52:31 +00002203 .map = snapshot_merge_map,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002204 .end_io = snapshot_end_io,
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002205 .presuspend = snapshot_merge_presuspend,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002206 .postsuspend = snapshot_postsuspend,
2207 .preresume = snapshot_preresume,
Mikulas Patocka1e03f972009-12-10 23:52:32 +00002208 .resume = snapshot_merge_resume,
Mikulas Patockad698aa42009-12-10 23:52:30 +00002209 .status = snapshot_status,
2210 .iterate_devices = snapshot_iterate_devices,
2211};
2212
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213static int __init dm_snapshot_init(void)
2214{
2215 int r;
2216
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00002217 r = dm_exception_store_init();
2218 if (r) {
2219 DMERR("Failed to initialize exception stores");
2220 return r;
2221 }
2222
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 r = dm_register_target(&snapshot_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002224 if (r < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 DMERR("snapshot target register failed %d", r);
Jonathan Brassow034a1862009-10-16 23:18:14 +01002226 goto bad_register_snapshot_target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 }
2228
2229 r = dm_register_target(&origin_target);
2230 if (r < 0) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07002231 DMERR("Origin target register failed %d", r);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002232 goto bad_register_origin_target;
2233 }
2234
2235 r = dm_register_target(&merge_target);
2236 if (r < 0) {
2237 DMERR("Merge target register failed %d", r);
2238 goto bad_register_merge_target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 }
2240
2241 r = init_origin_hash();
2242 if (r) {
2243 DMERR("init_origin_hash failed.");
Mikulas Patockad698aa42009-12-10 23:52:30 +00002244 goto bad_origin_hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 }
2246
Jon Brassow1d4989c2009-12-10 23:52:10 +00002247 exception_cache = KMEM_CACHE(dm_exception, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 if (!exception_cache) {
2249 DMERR("Couldn't create exception cache.");
2250 r = -ENOMEM;
Mikulas Patockad698aa42009-12-10 23:52:30 +00002251 goto bad_exception_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 }
2253
Alasdair G Kergon028867a2007-07-12 17:26:32 +01002254 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 if (!pending_cache) {
2256 DMERR("Couldn't create pending cache.");
2257 r = -ENOMEM;
Mikulas Patockad698aa42009-12-10 23:52:30 +00002258 goto bad_pending_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 }
2260
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002261 tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
2262 if (!tracked_chunk_cache) {
2263 DMERR("Couldn't create cache to track chunks in use.");
2264 r = -ENOMEM;
Mikulas Patockad698aa42009-12-10 23:52:30 +00002265 goto bad_tracked_chunk_cache;
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002266 }
2267
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07002268 ksnapd = create_singlethread_workqueue("ksnapd");
2269 if (!ksnapd) {
2270 DMERR("Failed to create ksnapd workqueue.");
2271 r = -ENOMEM;
Mikulas Patocka92e86812008-07-21 12:00:35 +01002272 goto bad_pending_pool;
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07002273 }
2274
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 return 0;
2276
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00002277bad_pending_pool:
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002278 kmem_cache_destroy(tracked_chunk_cache);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002279bad_tracked_chunk_cache:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 kmem_cache_destroy(pending_cache);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002281bad_pending_cache:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 kmem_cache_destroy(exception_cache);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002283bad_exception_cache:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 exit_origin_hash();
Mikulas Patockad698aa42009-12-10 23:52:30 +00002285bad_origin_hash:
2286 dm_unregister_target(&merge_target);
2287bad_register_merge_target:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 dm_unregister_target(&origin_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002289bad_register_origin_target:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 dm_unregister_target(&snapshot_target);
Jonathan Brassow034a1862009-10-16 23:18:14 +01002291bad_register_snapshot_target:
2292 dm_exception_store_exit();
Mikulas Patockad698aa42009-12-10 23:52:30 +00002293
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 return r;
2295}
2296
2297static void __exit dm_snapshot_exit(void)
2298{
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07002299 destroy_workqueue(ksnapd);
2300
Mikulas Patocka10d3bd02009-01-06 03:04:58 +00002301 dm_unregister_target(&snapshot_target);
2302 dm_unregister_target(&origin_target);
Mikulas Patockad698aa42009-12-10 23:52:30 +00002303 dm_unregister_target(&merge_target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304
2305 exit_origin_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 kmem_cache_destroy(pending_cache);
2307 kmem_cache_destroy(exception_cache);
Mikulas Patockacd45daf2008-07-21 12:00:32 +01002308 kmem_cache_destroy(tracked_chunk_cache);
Alasdair G Kergon4db6bfe2009-01-06 03:05:17 +00002309
2310 dm_exception_store_exit();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311}
2312
2313/* Module hooks */
2314module_init(dm_snapshot_init);
2315module_exit(dm_snapshot_exit);
2316
2317MODULE_DESCRIPTION(DM_NAME " snapshot target");
2318MODULE_AUTHOR("Joe Thornber");
2319MODULE_LICENSE("GPL");