dm-thin.c 101 KB
Newer Older
1
/*
2
 * Copyright (C) 2011-2012 Red Hat UK.
3
4
5
6
7
 *
 * This file is released under the GPL.
 */

#include "dm-thin-metadata.h"
8
#include "dm-bio-prison.h"
9
#include "dm.h"
10
11
12
13

#include <linux/device-mapper.h>
#include <linux/dm-io.h>
#include <linux/dm-kcopyd.h>
14
#include <linux/jiffies.h>
15
#include <linux/log2.h>
16
#include <linux/list.h>
17
#include <linux/rculist.h>
18
19
20
#include <linux/init.h>
#include <linux/module.h>
#include <linux/slab.h>
21
#include <linux/sort.h>
22
#include <linux/rbtree.h>
23
24
25
26
27
28

#define	DM_MSG_PREFIX	"thin"

/*
 * Tunable constants
 */
29
#define ENDIO_HOOK_POOL_SIZE 1024
30
#define MAPPING_POOL_SIZE 1024
31
#define COMMIT_PERIOD HZ
32
33
34
#define NO_SPACE_TIMEOUT_SECS 60

static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS;
35

36
37
38
DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
		"A percentage of time allocated for copy on write");

39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
/*
 * The block size of the device holding pool data must be
 * between 64KB and 1GB.
 */
#define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
#define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)

/*
 * Device id is restricted to 24 bits.
 */
#define MAX_DEV_ID ((1 << 24) - 1)

/*
 * How do we handle breaking sharing of data blocks?
 * =================================================
 *
 * We use a standard copy-on-write btree to store the mappings for the
 * devices (note I'm talking about copy-on-write of the metadata here, not
 * the data).  When you take an internal snapshot you clone the root node
 * of the origin btree.  After this there is no concept of an origin or a
 * snapshot.  They are just two device trees that happen to point to the
 * same data blocks.
 *
 * When we get a write in we decide if it's to a shared data block using
 * some timestamp magic.  If it is, we have to break sharing.
 *
 * Let's say we write to a shared block in what was the origin.  The
 * steps are:
 *
 * i) plug io further to this physical block. (see bio_prison code).
 *
 * ii) quiesce any read io to that shared data block.  Obviously
71
 * including all devices that share this block.  (see dm_deferred_set code)
72
73
74
75
76
 *
 * iii) copy the data block to a newly allocate block.  This step can be
 * missed out if the io covers the block. (schedule_copy).
 *
 * iv) insert the new mapping into the origin's btree
Joe Thornber's avatar
Joe Thornber committed
77
 * (process_prepared_mapping).  This act of inserting breaks some
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
 * sharing of btree nodes between the two devices.  Breaking sharing only
 * effects the btree of that specific device.  Btrees for the other
 * devices that share the block never change.  The btree for the origin
 * device as it was after the last commit is untouched, ie. we're using
 * persistent data structures in the functional programming sense.
 *
 * v) unplug io to this physical block, including the io that triggered
 * the breaking of sharing.
 *
 * Steps (ii) and (iii) occur in parallel.
 *
 * The metadata _doesn't_ need to be committed before the io continues.  We
 * get away with this because the io is always written to a _new_ block.
 * If there's a crash, then:
 *
 * - The origin mapping will point to the old origin block (the shared
 * one).  This will contain the data as it was before the io that triggered
 * the breaking of sharing came in.
 *
 * - The snap mapping still points to the old block.  As it would after
 * the commit.
 *
 * The downside of this scheme is the timestamp magic isn't perfect, and
 * will continue to think that data block in the snapshot device is shared
 * even after the write to the origin has broken sharing.  I suspect data
 * blocks will typically be shared by many different devices, so we're
 * breaking sharing n + 1 times, rather than n, where n is the number of
 * devices that reference this data block.  At the moment I think the
 * benefits far, far outweigh the disadvantages.
 */

/*----------------------------------------------------------------*/

/*
 * Key building.
 */
static void build_data_key(struct dm_thin_device *td,
115
			   dm_block_t b, struct dm_cell_key *key)
116
117
118
{
	key->virtual = 0;
	key->dev = dm_thin_dev_id(td);
119
120
	key->block_begin = b;
	key->block_end = b + 1ULL;
121
122
123
}

static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
124
			      struct dm_cell_key *key)
125
126
127
{
	key->virtual = 1;
	key->dev = dm_thin_dev_id(td);
128
129
	key->block_begin = b;
	key->block_end = b + 1ULL;
130
131
132
133
}

/*----------------------------------------------------------------*/

Joe Thornber's avatar
Joe Thornber committed
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
#define THROTTLE_THRESHOLD (1 * HZ)

struct throttle {
	struct rw_semaphore lock;
	unsigned long threshold;
	bool throttle_applied;
};

static void throttle_init(struct throttle *t)
{
	init_rwsem(&t->lock);
	t->throttle_applied = false;
}

static void throttle_work_start(struct throttle *t)
{
	t->threshold = jiffies + THROTTLE_THRESHOLD;
}

static void throttle_work_update(struct throttle *t)
{
	if (!t->throttle_applied && jiffies > t->threshold) {
		down_write(&t->lock);
		t->throttle_applied = true;
	}
}

static void throttle_work_complete(struct throttle *t)
{
	if (t->throttle_applied) {
		t->throttle_applied = false;
		up_write(&t->lock);
	}
}

static void throttle_lock(struct throttle *t)
{
	down_read(&t->lock);
}

static void throttle_unlock(struct throttle *t)
{
	up_read(&t->lock);
}

/*----------------------------------------------------------------*/

181
182
183
184
185
/*
 * A pool device ties together a metadata device and a data device.  It
 * also provides the interface for creating and destroying internal
 * devices.
 */
Mike Snitzer's avatar
Mike Snitzer committed
186
struct dm_thin_new_mapping;
187

188
/*
189
 * The pool runs in 4 modes.  Ordered in degraded order for comparisons.
190
191
192
 */
enum pool_mode {
	PM_WRITE,		/* metadata may be changed */
193
	PM_OUT_OF_DATA_SPACE,	/* metadata may be changed, though data may not be allocated */
194
195
196
197
	PM_READ_ONLY,		/* metadata may not be changed */
	PM_FAIL,		/* all I/O fails */
};

198
struct pool_features {
199
200
	enum pool_mode mode;

Mike Snitzer's avatar
Mike Snitzer committed
201
202
203
	bool zero_new_blocks:1;
	bool discard_enabled:1;
	bool discard_passdown:1;
204
	bool error_if_no_space:1;
205
206
};

207
208
struct thin_c;
typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
209
typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell);
210
211
typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);

212
213
#define CELL_SORT_ARRAY_SIZE 8192

214
215
216
217
218
219
220
221
222
struct pool {
	struct list_head list;
	struct dm_target *ti;	/* Only set if a pool target is bound */

	struct mapped_device *pool_md;
	struct block_device *md_dev;
	struct dm_pool_metadata *pmd;

	dm_block_t low_water_blocks;
223
	uint32_t sectors_per_block;
224
	int sectors_per_block_shift;
225

226
	struct pool_features pf;
227
	bool low_water_triggered:1;	/* A dm event has been sent */
228
	bool suspended:1;
229

230
	struct dm_bio_prison *prison;
231
232
233
	struct dm_kcopyd_client *copier;

	struct workqueue_struct *wq;
Joe Thornber's avatar
Joe Thornber committed
234
	struct throttle throttle;
235
	struct work_struct worker;
236
	struct delayed_work waker;
237
	struct delayed_work no_space_timeout;
238

239
	unsigned long last_commit_jiffies;
240
	unsigned ref_count;
241
242
243
244

	spinlock_t lock;
	struct bio_list deferred_flush_bios;
	struct list_head prepared_mappings;
Joe Thornber's avatar
Joe Thornber committed
245
	struct list_head prepared_discards;
246
	struct list_head active_thins;
247

248
249
	struct dm_deferred_set *shared_read_ds;
	struct dm_deferred_set *all_io_ds;
250

Mike Snitzer's avatar
Mike Snitzer committed
251
	struct dm_thin_new_mapping *next_mapping;
252
	mempool_t *mapping_pool;
253
254
255
256

	process_bio_fn process_bio;
	process_bio_fn process_discard;

257
258
259
	process_cell_fn process_cell;
	process_cell_fn process_discard_cell;

260
261
	process_mapping_fn process_prepared_mapping;
	process_mapping_fn process_prepared_discard;
262
263

	struct dm_bio_prison_cell *cell_sort_array[CELL_SORT_ARRAY_SIZE];
264
265
};

266
static enum pool_mode get_pool_mode(struct pool *pool);
267
static void metadata_operation_failed(struct pool *pool, const char *op, int r);
268

269
270
271
272
273
274
275
276
277
278
279
/*
 * Target context for a pool.
 */
struct pool_c {
	struct dm_target *ti;
	struct pool *pool;
	struct dm_dev *data_dev;
	struct dm_dev *metadata_dev;
	struct dm_target_callbacks callbacks;

	dm_block_t low_water_blocks;
280
281
	struct pool_features requested_pf; /* Features requested during table load */
	struct pool_features adjusted_pf;  /* Features used after adjusting for constituent devices */
282
283
284
285
286
287
};

/*
 * Target context for a thin.
 */
struct thin_c {
288
	struct list_head list;
289
	struct dm_dev *pool_dev;
290
	struct dm_dev *origin_dev;
291
	sector_t origin_size;
292
293
294
295
	dm_thin_id dev_id;

	struct pool *pool;
	struct dm_thin_device *td;
296
297
	struct mapped_device *thin_md;

298
	bool requeue_mode:1;
299
	spinlock_t lock;
300
	struct list_head deferred_cells;
301
302
	struct bio_list deferred_bio_list;
	struct bio_list retry_on_resume_list;
303
	struct rb_root sort_bio_list; /* sorted list of deferred bios */
304
305
306
307
308
309
310

	/*
	 * Ensures the thin is not destroyed until the worker has finished
	 * iterating the active_thins list.
	 */
	atomic_t refcount;
	struct completion can_destroy;
311
312
313
314
};

/*----------------------------------------------------------------*/

315
316
317
318
319
320
321
322
323
324
325
/*
 * wake_worker() is used when new work is queued and when pool_resume is
 * ready to continue deferred IO processing.
 */
static void wake_worker(struct pool *pool)
{
	queue_work(pool->wq, &pool->worker);
}

/*----------------------------------------------------------------*/

326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio,
		      struct dm_bio_prison_cell **cell_result)
{
	int r;
	struct dm_bio_prison_cell *cell_prealloc;

	/*
	 * Allocate a cell from the prison's mempool.
	 * This might block but it can't fail.
	 */
	cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO);

	r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result);
	if (r)
		/*
		 * We reused an old cell; we can get rid of
		 * the new one.
		 */
		dm_bio_prison_free_cell(pool->prison, cell_prealloc);

	return r;
}

static void cell_release(struct pool *pool,
			 struct dm_bio_prison_cell *cell,
			 struct bio_list *bios)
{
	dm_cell_release(pool->prison, cell, bios);
	dm_bio_prison_free_cell(pool->prison, cell);
}

357
358
359
360
361
362
363
364
365
static void cell_visit_release(struct pool *pool,
			       void (*fn)(void *, struct dm_bio_prison_cell *),
			       void *context,
			       struct dm_bio_prison_cell *cell)
{
	dm_cell_visit_release(pool->prison, fn, context, cell);
	dm_bio_prison_free_cell(pool->prison, cell);
}

366
367
368
369
370
371
372
373
static void cell_release_no_holder(struct pool *pool,
				   struct dm_bio_prison_cell *cell,
				   struct bio_list *bios)
{
	dm_cell_release_no_holder(pool->prison, cell, bios);
	dm_bio_prison_free_cell(pool->prison, cell);
}

374
375
static void cell_error_with_code(struct pool *pool,
				 struct dm_bio_prison_cell *cell, int error_code)
376
{
377
	dm_cell_error(pool->prison, cell, error_code);
378
379
380
	dm_bio_prison_free_cell(pool->prison, cell);
}

381
382
383
384
385
static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell)
{
	cell_error_with_code(pool, cell, -EIO);
}

386
387
388
389
390
391
392
393
394
395
static void cell_success(struct pool *pool, struct dm_bio_prison_cell *cell)
{
	cell_error_with_code(pool, cell, 0);
}

static void cell_requeue(struct pool *pool, struct dm_bio_prison_cell *cell)
{
	cell_error_with_code(pool, cell, DM_ENDIO_REQUEUE);
}

396
397
/*----------------------------------------------------------------*/

398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
/*
 * A global list of pools that uses a struct mapped_device as a key.
 */
static struct dm_thin_pool_table {
	struct mutex mutex;
	struct list_head pools;
} dm_thin_pool_table;

static void pool_table_init(void)
{
	mutex_init(&dm_thin_pool_table.mutex);
	INIT_LIST_HEAD(&dm_thin_pool_table.pools);
}

static void __pool_table_insert(struct pool *pool)
{
	BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
	list_add(&pool->list, &dm_thin_pool_table.pools);
}

static void __pool_table_remove(struct pool *pool)
{
	BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
	list_del(&pool->list);
}

static struct pool *__pool_table_lookup(struct mapped_device *md)
{
	struct pool *pool = NULL, *tmp;

	BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));

	list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
		if (tmp->pool_md == md) {
			pool = tmp;
			break;
		}
	}

	return pool;
}

static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
{
	struct pool *pool = NULL, *tmp;

	BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));

	list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
		if (tmp->md_dev == md_dev) {
			pool = tmp;
			break;
		}
	}

	return pool;
}

/*----------------------------------------------------------------*/

Mike Snitzer's avatar
Mike Snitzer committed
458
struct dm_thin_endio_hook {
459
	struct thin_c *tc;
460
461
	struct dm_deferred_entry *shared_read_entry;
	struct dm_deferred_entry *all_io_entry;
Mike Snitzer's avatar
Mike Snitzer committed
462
	struct dm_thin_new_mapping *overwrite_mapping;
463
	struct rb_node rb_node;
464
465
};

466
467
468
469
470
471
472
static void __merge_bio_list(struct bio_list *bios, struct bio_list *master)
{
	bio_list_merge(bios, master);
	bio_list_init(master);
}

static void error_bio_list(struct bio_list *bios, int error)
473
474
{
	struct bio *bio;
475
476
477
478
479
480
481

	while ((bio = bio_list_pop(bios)))
		bio_endio(bio, error);
}

static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master, int error)
{
482
	struct bio_list bios;
483
	unsigned long flags;
484
485

	bio_list_init(&bios);
486

487
	spin_lock_irqsave(&tc->lock, flags);
488
	__merge_bio_list(&bios, master);
489
	spin_unlock_irqrestore(&tc->lock, flags);
490

491
	error_bio_list(&bios, error);
492
493
}

494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
static void requeue_deferred_cells(struct thin_c *tc)
{
	struct pool *pool = tc->pool;
	unsigned long flags;
	struct list_head cells;
	struct dm_bio_prison_cell *cell, *tmp;

	INIT_LIST_HEAD(&cells);

	spin_lock_irqsave(&tc->lock, flags);
	list_splice_init(&tc->deferred_cells, &cells);
	spin_unlock_irqrestore(&tc->lock, flags);

	list_for_each_entry_safe(cell, tmp, &cells, user_list)
		cell_requeue(pool, cell);
}

511
512
static void requeue_io(struct thin_c *tc)
{
513
	struct bio_list bios;
514
	unsigned long flags;
515
516
517

	bio_list_init(&bios);

518
	spin_lock_irqsave(&tc->lock, flags);
519
520
	__merge_bio_list(&bios, &tc->deferred_bio_list);
	__merge_bio_list(&bios, &tc->retry_on_resume_list);
521
	spin_unlock_irqrestore(&tc->lock, flags);
522

523
524
	error_bio_list(&bios, DM_ENDIO_REQUEUE);
	requeue_deferred_cells(tc);
525
526
}

527
528
529
530
531
532
static void error_retry_list(struct pool *pool)
{
	struct thin_c *tc;

	rcu_read_lock();
	list_for_each_entry_rcu(tc, &pool->active_thins, list)
533
		error_thin_bio_list(tc, &tc->retry_on_resume_list, -EIO);
534
535
536
	rcu_read_unlock();
}

537
538
539
540
541
542
543
/*
 * This section of code contains the logic for processing a thin device's IO.
 * Much of the code depends on pool object resources (lists, workqueues, etc)
 * but most is exclusively called from the thin target rather than the thin-pool
 * target.
 */

544
545
546
547
548
static bool block_size_is_power_of_two(struct pool *pool)
{
	return pool->sectors_per_block_shift >= 0;
}

549
550
static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
{
551
	struct pool *pool = tc->pool;
552
	sector_t block_nr = bio->bi_iter.bi_sector;
553

554
555
	if (block_size_is_power_of_two(pool))
		block_nr >>= pool->sectors_per_block_shift;
556
	else
557
		(void) sector_div(block_nr, pool->sectors_per_block);
558
559

	return block_nr;
560
561
562
563
564
}

static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
{
	struct pool *pool = tc->pool;
565
	sector_t bi_sector = bio->bi_iter.bi_sector;
566
567

	bio->bi_bdev = tc->pool_dev->bdev;
568
	if (block_size_is_power_of_two(pool))
569
570
571
		bio->bi_iter.bi_sector =
			(block << pool->sectors_per_block_shift) |
			(bi_sector & (pool->sectors_per_block - 1));
572
	else
573
		bio->bi_iter.bi_sector = (block * pool->sectors_per_block) +
574
				 sector_div(bi_sector, pool->sectors_per_block);
575
576
}

577
578
579
580
581
static void remap_to_origin(struct thin_c *tc, struct bio *bio)
{
	bio->bi_bdev = tc->origin_dev->bdev;
}

582
583
584
585
586
587
static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
{
	return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) &&
		dm_thin_changed_this_transaction(tc->td);
}

588
589
590
591
592
593
594
static void inc_all_io_entry(struct pool *pool, struct bio *bio)
{
	struct dm_thin_endio_hook *h;

	if (bio->bi_rw & REQ_DISCARD)
		return;

595
	h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
596
597
598
	h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
}

599
static void issue(struct thin_c *tc, struct bio *bio)
600
601
602
603
{
	struct pool *pool = tc->pool;
	unsigned long flags;

604
605
606
607
608
	if (!bio_triggers_commit(tc, bio)) {
		generic_make_request(bio);
		return;
	}

609
	/*
610
611
612
	 * Complete bio with an error if earlier I/O caused changes to
	 * the metadata that can't be committed e.g, due to I/O errors
	 * on the metadata device.
613
	 */
614
615
616
617
618
619
620
621
622
623
624
625
	if (dm_thin_aborted_changes(tc->td)) {
		bio_io_error(bio);
		return;
	}

	/*
	 * Batch together any bios that trigger commits and then issue a
	 * single commit for them in process_deferred_bios().
	 */
	spin_lock_irqsave(&pool->lock, flags);
	bio_list_add(&pool->deferred_flush_bios, bio);
	spin_unlock_irqrestore(&pool->lock, flags);
626
627
}

628
629
630
631
632
633
634
635
636
637
638
639
640
static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
{
	remap_to_origin(tc, bio);
	issue(tc, bio);
}

static void remap_and_issue(struct thin_c *tc, struct bio *bio,
			    dm_block_t block)
{
	remap(tc, bio, block);
	issue(tc, bio);
}

641
642
643
644
645
/*----------------------------------------------------------------*/

/*
 * Bio endio functions.
 */
Mike Snitzer's avatar
Mike Snitzer committed
646
struct dm_thin_new_mapping {
647
648
	struct list_head list;

649
650
	bool pass_discard:1;
	bool definitely_not_shared:1;
651

652
653
654
655
656
657
658
	/*
	 * Track quiescing, copying and zeroing preparation actions.  When this
	 * counter hits zero the block is prepared and can be inserted into the
	 * btree.
	 */
	atomic_t prepare_actions;

659
	int err;
660
661
662
	struct thin_c *tc;
	dm_block_t virt_block;
	dm_block_t data_block;
Mike Snitzer's avatar
Mike Snitzer committed
663
	struct dm_bio_prison_cell *cell, *cell2;
664
665
666
667
668
669
670
671
672
673
674

	/*
	 * If the bio covers the whole area of a block then we can avoid
	 * zeroing or copying.  Instead this bio is hooked.  The bio will
	 * still be in the cell, so care has to be taken to avoid issuing
	 * the bio twice.
	 */
	struct bio *bio;
	bio_end_io_t *saved_bi_end_io;
};

675
static void __complete_mapping_preparation(struct dm_thin_new_mapping *m)
676
677
678
{
	struct pool *pool = m->tc->pool;

679
	if (atomic_dec_and_test(&m->prepare_actions)) {
680
		list_add_tail(&m->list, &pool->prepared_mappings);
681
682
683
684
		wake_worker(pool);
	}
}

685
static void complete_mapping_preparation(struct dm_thin_new_mapping *m)
686
687
688
689
690
{
	unsigned long flags;
	struct pool *pool = m->tc->pool;

	spin_lock_irqsave(&pool->lock, flags);
691
	__complete_mapping_preparation(m);
692
693
694
	spin_unlock_irqrestore(&pool->lock, flags);
}

695
696
697
698
699
700
701
702
static void copy_complete(int read_err, unsigned long write_err, void *context)
{
	struct dm_thin_new_mapping *m = context;

	m->err = read_err || write_err ? -EIO : 0;
	complete_mapping_preparation(m);
}

703
704
static void overwrite_endio(struct bio *bio, int err)
{
705
	struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Mike Snitzer's avatar
Mike Snitzer committed
706
	struct dm_thin_new_mapping *m = h->overwrite_mapping;
707

708
709
	bio->bi_end_io = m->saved_bi_end_io;

710
	m->err = err;
711
	complete_mapping_preparation(m);
712
713
714
715
716
717
718
719
720
721
722
723
724
}

/*----------------------------------------------------------------*/

/*
 * Workqueue.
 */

/*
 * Prepared mapping jobs.
 */

/*
725
726
 * This sends the bios in the cell, except the original holder, back
 * to the deferred_bios list.
727
 */
728
static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
729
730
731
732
{
	struct pool *pool = tc->pool;
	unsigned long flags;

733
734
735
	spin_lock_irqsave(&tc->lock, flags);
	cell_release_no_holder(pool, cell, &tc->deferred_bio_list);
	spin_unlock_irqrestore(&tc->lock, flags);
736
737
738
739

	wake_worker(pool);
}

740
741
static void thin_defer_bio(struct thin_c *tc, struct bio *bio);

742
743
744
745
746
747
748
749
struct remap_info {
	struct thin_c *tc;
	struct bio_list defer_bios;
	struct bio_list issue_bios;
};

static void __inc_remap_and_issue_cell(void *context,
				       struct dm_bio_prison_cell *cell)
750
{
751
	struct remap_info *info = context;
752
753
	struct bio *bio;

754
	while ((bio = bio_list_pop(&cell->bios))) {
755
		if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA))
756
			bio_list_add(&info->defer_bios, bio);
757
		else {
758
759
760
761
762
763
764
765
			inc_all_io_entry(info->tc->pool, bio);

			/*
			 * We can't issue the bios with the bio prison lock
			 * held, so we add them to a list to issue on
			 * return from this function.
			 */
			bio_list_add(&info->issue_bios, bio);
766
767
768
769
		}
	}
}

770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
static void inc_remap_and_issue_cell(struct thin_c *tc,
				     struct dm_bio_prison_cell *cell,
				     dm_block_t block)
{
	struct bio *bio;
	struct remap_info info;

	info.tc = tc;
	bio_list_init(&info.defer_bios);
	bio_list_init(&info.issue_bios);

	/*
	 * We have to be careful to inc any bios we're about to issue
	 * before the cell is released, and avoid a race with new bios
	 * being added to the cell.
	 */
	cell_visit_release(tc->pool, __inc_remap_and_issue_cell,
			   &info, cell);

	while ((bio = bio_list_pop(&info.defer_bios)))
		thin_defer_bio(tc, bio);

	while ((bio = bio_list_pop(&info.issue_bios)))
		remap_and_issue(info.tc, bio, block);
}

796
797
static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
{
798
	cell_error(m->tc->pool, m->cell);
799
800
801
	list_del(&m->list);
	mempool_free(m, m->tc->pool->mapping_pool);
}
802

Mike Snitzer's avatar
Mike Snitzer committed
803
static void process_prepared_mapping(struct dm_thin_new_mapping *m)
804
805
{
	struct thin_c *tc = m->tc;
806
	struct pool *pool = tc->pool;
807
	struct bio *bio = m->bio;
808
809
810
	int r;

	if (m->err) {
811
		cell_error(pool, m->cell);
812
		goto out;
813
814
815
816
817
818
819
820
821
	}

	/*
	 * Commit the prepared block into the mapping btree.
	 * Any I/O for this block arriving after this point will get
	 * remapped to it directly.
	 */
	r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block);
	if (r) {
822
		metadata_operation_failed(pool, "dm_thin_insert_block", r);
823
		cell_error(pool, m->cell);
824
		goto out;
825
826
827
828
829
830
831
832
833
	}

	/*
	 * Release any bios held while the block was being provisioned.
	 * If we are processing a write bio that completely covers the block,
	 * we already processed it so can ignore it now when processing
	 * the bios in the cell.
	 */
	if (bio) {
834
		inc_remap_and_issue_cell(tc, m->cell, m->data_block);
835
		bio_endio(bio, 0);
836
837
838
839
840
	} else {
		inc_all_io_entry(tc->pool, m->cell->holder);
		remap_and_issue(tc, m->cell->holder, m->data_block);
		inc_remap_and_issue_cell(tc, m->cell, m->data_block);
	}
841

842
out:
843
	list_del(&m->list);
844
	mempool_free(m, pool->mapping_pool);
845
846
}

847
static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
Joe Thornber's avatar
Joe Thornber committed
848
849
850
{
	struct thin_c *tc = m->tc;

851
	bio_io_error(m->bio);
852
853
	cell_defer_no_holder(tc, m->cell);
	cell_defer_no_holder(tc, m->cell2);
854
855
856
857
858
859
	mempool_free(m, tc->pool->mapping_pool);
}

static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m)
{
	struct thin_c *tc = m->tc;
Joe Thornber's avatar
Joe Thornber committed
860

861
	inc_all_io_entry(tc->pool, m->bio);
862
863
	cell_defer_no_holder(tc, m->cell);
	cell_defer_no_holder(tc, m->cell2);
864

Joe Thornber's avatar
Joe Thornber committed
865
	if (m->pass_discard)
866
867
868
869
870
871
872
873
874
		if (m->definitely_not_shared)
			remap_and_issue(tc, m->bio, m->data_block);
		else {
			bool used = false;
			if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used)
				bio_endio(m->bio, 0);
			else
				remap_and_issue(tc, m->bio, m->data_block);
		}
Joe Thornber's avatar
Joe Thornber committed
875
876
877
878
879
880
	else
		bio_endio(m->bio, 0);

	mempool_free(m, tc->pool->mapping_pool);
}

881
882
883
884
885
886
887
static void process_prepared_discard(struct dm_thin_new_mapping *m)
{
	int r;
	struct thin_c *tc = m->tc;

	r = dm_thin_remove_block(tc->td, m->virt_block);
	if (r)
888
		DMERR_LIMIT("dm_thin_remove_block() failed");
889
890
891
892

	process_prepared_discard_passdown(m);
}

Joe Thornber's avatar
Joe Thornber committed
893
static void process_prepared(struct pool *pool, struct list_head *head,
894
			     process_mapping_fn *fn)
895
896
897
{
	unsigned long flags;
	struct list_head maps;
Mike Snitzer's avatar
Mike Snitzer committed
898
	struct dm_thin_new_mapping *m, *tmp;
899
900
901

	INIT_LIST_HEAD(&maps);
	spin_lock_irqsave(&pool->lock, flags);
Joe Thornber's avatar
Joe Thornber committed
902
	list_splice_init(head, &maps);
903
904
905
	spin_unlock_irqrestore(&pool->lock, flags);

	list_for_each_entry_safe(m, tmp, &maps, list)
906
		(*fn)(m);
907
908
909
910
911
}

/*
 * Deferred bio jobs.
 */
Joe Thornber's avatar
Joe Thornber committed
912
static int io_overlaps_block(struct pool *pool, struct bio *bio)
913
{
914
915
	return bio->bi_iter.bi_size ==
		(pool->sectors_per_block << SECTOR_SHIFT);
Joe Thornber's avatar
Joe Thornber committed
916
917
918
919
920
921
}

static int io_overwrites_block(struct pool *pool, struct bio *bio)
{
	return (bio_data_dir(bio) == WRITE) &&
		io_overlaps_block(pool, bio);
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
}

static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
			       bio_end_io_t *fn)
{
	*save = bio->bi_end_io;
	bio->bi_end_io = fn;
}

static int ensure_next_mapping(struct pool *pool)
{
	if (pool->next_mapping)
		return 0;

	pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);

	return pool->next_mapping ? 0 : -ENOMEM;
}

Mike Snitzer's avatar
Mike Snitzer committed
941
static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
942
{
943
	struct dm_thin_new_mapping *m = pool->next_mapping;
944
945
946

	BUG_ON(!pool->next_mapping);

947
948
949
950
	memset(m, 0, sizeof(struct dm_thin_new_mapping));
	INIT_LIST_HEAD(&m->list);
	m->bio = NULL;

951
952
	pool->next_mapping = NULL;

953
	return m;
954
955
}

956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m,
		    sector_t begin, sector_t end)
{
	int r;
	struct dm_io_region to;

	to.bdev = tc->pool_dev->bdev;
	to.sector = begin;
	to.count = end - begin;

	r = dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m);
	if (r < 0) {
		DMERR_LIMIT("dm_kcopyd_zero() failed");
		copy_complete(1, 1, m);
	}
}

973
974
975
976
977
978
979
980
981
982
983
984
985
986
static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio,
				      dm_block_t data_block,
				      struct dm_thin_new_mapping *m)
{
	struct pool *pool = tc->pool;
	struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));

	h->overwrite_mapping = m;
	m->bio = bio;
	save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
	inc_all_io_entry(pool, bio);
	remap_and_issue(tc, bio, data_block);
}

987
988
989
/*
 * A partial copy also needs to zero the uncopied region.
 */
990
static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
991
992
			  struct dm_dev *origin, dm_block_t data_origin,
			  dm_block_t data_dest,
993
994
			  struct dm_bio_prison_cell *cell, struct bio *bio,
			  sector_t len)
995
996
997
{
	int r;
	struct pool *pool = tc->pool;
Mike Snitzer's avatar
Mike Snitzer committed
998
	struct dm_thin_new_mapping *m = get_next_mapping(pool);
999
1000
1001
1002
1003
1004

	m->tc = tc;
	m->virt_block = virt_block;
	m->data_block = data_dest;
	m->cell = cell;

1005
1006
1007
1008
1009
1010
1011
	/*
	 * quiesce action + copy action + an extra reference held for the
	 * duration of this function (we may need to inc later for a
	 * partial zero).
	 */
	atomic_set(&m->prepare_actions, 3);

1012
	if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
1013
		complete_mapping_preparation(m); /* already quiesced */
1014
1015
1016
1017
1018
1019
1020

	/*
	 * IO to pool_dev remaps to the pool target's data_dev.
	 *
	 * If the whole block of data is being overwritten, we can issue the
	 * bio immediately. Otherwise we use kcopyd to clone the data first.
	 */
1021
1022
1023
	if (io_overwrites_block(pool, bio))
		remap_and_issue_overwrite(tc, bio, data_dest, m);
	else {
1024
1025
		struct dm_io_region from, to;

1026
		from.bdev = origin->bdev;
1027
		from.sector = data_origin * pool->sectors_per_block;
1028
		from.count = len;
1029
1030
1031

		to.bdev = tc->pool_dev->bdev;
		to.sector = data_dest * pool->sectors_per_block;
1032
		to.count = len;
1033
1034
1035
1036

		r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
				   0, copy_complete, m);
		if (r < 0) {
1037
			DMERR_LIMIT("dm_kcopyd_copy() failed");
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
			copy_complete(1, 1, m);

			/*
			 * We allow the zero to be issued, to simplify the
			 * error path.  Otherwise we'd need to start
			 * worrying about decrementing the prepare_actions
			 * counter.
			 */
		}

		/*
		 * Do we need to zero a tail region?
		 */
		if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) {
			atomic_inc(&m->prepare_actions);
			ll_zero(tc, m,
				data_dest * pool->sectors_per_block + len,
				(data_dest + 1) * pool->sectors_per_block);
1056
1057
		}
	}
1058
1059

	complete_mapping_preparation(m); /* drop our ref */
1060
1061
}

1062
1063
static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
				   dm_block_t data_origin, dm_block_t data_dest,
Mike Snitzer's avatar
Mike Snitzer committed
1064
				   struct dm_bio_prison_cell *cell, struct bio *bio)
1065
1066
{
	schedule_copy(tc, virt_block, tc->pool_dev,
1067
1068
		      data_origin, data_dest, cell, bio,
		      tc->pool->sectors_per_block);
1069
1070
}

1071
static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
Mike Snitzer's avatar
Mike Snitzer committed
1072
			  dm_block_t data_block, struct dm_bio_prison_cell *cell,
1073
1074
1075
			  struct bio *bio)
{
	struct pool *pool = tc->pool;
Mike Snitzer's avatar
Mike Snitzer committed
1076
	struct dm_thin_new_mapping *m = get_next_mapping(pool);
1077

1078
	atomic_set(&m->prepare_actions, 1); /* no need to quiesce */
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
	m->tc = tc;
	m->virt_block = virt_block;
	m->data_block = data_block;
	m->cell = cell;

	/*
	 * If the whole block of data is being overwritten or we are not
	 * zeroing pre-existing data, we can issue the bio immediately.
	 * Otherwise we use kcopyd to zero the data first.
	 */
1089
1090
1091
1092
1093
1094
1095
	if (pool->pf.zero_new_blocks) {
		if (io_overwrites_block(pool, bio))
			remap_and_issue_overwrite(tc, bio, data_block, m);
		else
			ll_zero(tc, m, data_block * pool->sectors_per_block,
				(data_block + 1) * pool->sectors_per_block);
	} else
1096
		process_prepared_mapping(m);
1097
}
1098

1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
				   dm_block_t data_dest,
				   struct dm_bio_prison_cell *cell, struct bio *bio)
{
	struct pool *pool = tc->pool;
	sector_t virt_block_begin = virt_block * pool->sectors_per_block;
	sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block;

	if (virt_block_end <= tc->origin_size)
		schedule_copy(tc, virt_block, tc->origin_dev,
			      virt_block, data_dest, cell, bio,
			      pool->sectors_per_block);

	else if (virt_block_begin < tc->origin_size)
		schedule_copy(tc, virt_block, tc->origin_dev,
			      virt_block, data_dest, cell, bio,
			      tc->origin_size - virt_block_begin);

	else
		schedule_zero(tc, virt_block, data_dest, cell, bio);
1119
1120
}

1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
static void set_pool_mode(struct pool *pool, enum pool_mode new_mode);

static void check_for_space(struct pool *pool)
{
	int r;
	dm_block_t nr_free;

	if (get_pool_mode(pool) != PM_OUT_OF_DATA_SPACE)
		return;

	r = dm_pool_get_free_block_count(pool->pmd, &nr_free);
	if (r)
		return;

	if (nr_free)
		set_pool_mode(pool, PM_WRITE);
}

1139
1140
1141
1142
/*
 * A non-zero return indicates read_only or fail_io mode.
 * Many callers don't care about the return value.
 */
1143
static int commit(struct pool *pool)
1144
1145
1146
{
	int r;

1147
	if (get_pool_mode(pool) >= PM_READ_ONLY)
1148
1149
		return -EINVAL;

1150
	r = dm_pool_commit_metadata(pool->pmd);
1151
1152
	if (r)
		metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
1153
1154
	else
		check_for_space(pool);
1155
1156
1157
1158

	return r;
}

1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks)
{
	unsigned long flags;

	if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
		DMWARN("%s: reached low water mark for data device: sending event.",
		       dm_device_name(pool->pool_md));
		spin_lock_irqsave(&pool->lock, flags);
		pool->low_water_triggered = true;
		spin_unlock_irqrestore(&pool->lock, flags);
		dm_table_event(pool->ti->table);
	}
}

1173
1174
1175
1176
1177
1178
static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
{
	int r;
	dm_block_t free_blocks;
	struct pool *pool = tc->pool;

1179
	if (WARN_ON(get_pool_mode(pool) != PM_WRITE))
1180
1181
		return -EINVAL;

1182
	r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1183
1184
	if (r) {
		metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1185
		return r;
1186
	}
1187

1188
	check_low_water_mark(pool, free_blocks);
1189
1190

	if (!free_blocks) {
1191
1192
1193
1194
		/*
		 * Try to commit to see if that will free up some
		 * more space.
		 */
1195
1196
1197
		r = commit(pool);
		if (r)
			return r;
1198

1199
		r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1200
1201
		if (r) {
			metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1202
			return r;
1203
		}
1204

1205
		if (!free_blocks) {
1206
			set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1207
			return -ENOSPC;
1208
1209
1210
1211
		}
	}

	r = dm_pool_alloc_data_block(pool->pmd, result);
1212
	if (r) {
1213
		metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
1214
		return r;
1215
	}
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225

	return 0;
}

/*
 * If we have run out of space, queue bios until the device is
 * resumed, presumably after having been reloaded with more space.
 */
static void retry_on_resume(struct bio *bio)
{
1226
	struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1227
	struct thin_c *tc = h->tc;
1228
1229
	unsigned long flags;

1230
1231
1232
	spin_lock_irqsave(&tc->lock, flags);
	bio_list_add(&tc->retry_on_resume_list, bio);
	spin_unlock_irqrestore(&tc->lock, flags);
1233
1234
}

1235
static int should_error_unserviceable_bio(struct pool *pool)
1236
{
1237
1238
1239
1240
1241
1242
	enum pool_mode m = get_pool_mode(pool);

	switch (m) {
	case PM_WRITE:
		/* Shouldn't get here */
		DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
1243
		return -EIO;
1244
1245

	case PM_OUT_OF_DATA_SPACE:
1246
		return pool->pf.error_if_no_space ? -ENOSPC : 0;
1247
1248
1249

	case PM_READ_ONLY:
	case PM_FAIL:
1250
		return -EIO;
1251
1252
1253
	default:
		/* Shouldn't get here */
		DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1254
		return -EIO;
1255
1256
	}
}
1257

1258
1259
static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
{
1260
1261
1262
1263
	int error = should_error_unserviceable_bio(pool);

	if (error)
		bio_endio(bio, error);
1264
1265
	else
		retry_on_resume(bio);
1266
1267
}

1268
static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell)
1269
1270
1271
{
	struct bio *bio;
	struct bio_list bios;
1272
	int error;
1273

1274
1275
1276
	error = should_error_unserviceable_bio(pool);
	if (error) {
		cell_error_with_code(pool, cell, error);
1277
1278
1279
		return;
	}

1280
	bio_list_init(&bios);
1281
	cell_release(pool, cell, &bios);
1282

1283
1284
	while ((bio = bio_list_pop(&bios)))
		retry_on_resume(bio);
1285
1286
}

1287
static void process_discard_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
Joe Thornber's avatar
Joe Thornber committed
1288
1289
{
	int r;
1290
	struct bio *bio = cell->holder;
Joe Thornber's avatar
Joe Thornber committed
1291
	struct pool *pool = tc->pool;