dm-thin.c 106 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/vmalloc.h>
22
#include <linux/sort.h>
23
#include <linux/rbtree.h>
24
25
26
27
28
29

#define	DM_MSG_PREFIX	"thin"

/*
 * Tunable constants
 */
30
#define ENDIO_HOOK_POOL_SIZE 1024
31
#define MAPPING_POOL_SIZE 1024
32
#define COMMIT_PERIOD HZ
33
34
35
#define NO_SPACE_TIMEOUT_SECS 60

static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS;
36

37
38
39
DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
		"A percentage of time allocated for copy on write");

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
71
/*
 * 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
72
 * including all devices that share this block.  (see dm_deferred_set code)
73
74
75
76
77
 *
 * 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
78
 * (process_prepared_mapping).  This act of inserting breaks some
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.
 */
Joe Thornber's avatar
Joe Thornber committed
115
116
117
118
119
120
121
enum lock_space {
	VIRTUAL,
	PHYSICAL
};

static void build_key(struct dm_thin_device *td, enum lock_space ls,
		      dm_block_t b, dm_block_t e, struct dm_cell_key *key)
122
{
Joe Thornber's avatar
Joe Thornber committed
123
	key->virtual = (ls == VIRTUAL);
124
	key->dev = dm_thin_dev_id(td);
125
	key->block_begin = b;
Joe Thornber's avatar
Joe Thornber committed
126
127
128
129
130
131
132
	key->block_end = e;
}

static void build_data_key(struct dm_thin_device *td, dm_block_t b,
			   struct dm_cell_key *key)
{
	build_key(td, PHYSICAL, b, b + 1llu, key);
133
134
135
}

static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
136
			      struct dm_cell_key *key)
137
{
Joe Thornber's avatar
Joe Thornber committed
138
	build_key(td, VIRTUAL, b, b + 1llu, key);
139
140
141
142
}

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

Joe Thornber's avatar
Joe Thornber committed
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
181
182
183
184
185
186
187
188
189
#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);
}

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

190
191
192
193
194
/*
 * 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
195
struct dm_thin_new_mapping;
196

197
/*
198
 * The pool runs in 4 modes.  Ordered in degraded order for comparisons.
199
200
201
 */
enum pool_mode {
	PM_WRITE,		/* metadata may be changed */
202
	PM_OUT_OF_DATA_SPACE,	/* metadata may be changed, though data may not be allocated */
203
204
205
206
	PM_READ_ONLY,		/* metadata may not be changed */
	PM_FAIL,		/* all I/O fails */
};

207
struct pool_features {
208
209
	enum pool_mode mode;

Mike Snitzer's avatar
Mike Snitzer committed
210
211
212
	bool zero_new_blocks:1;
	bool discard_enabled:1;
	bool discard_passdown:1;
213
	bool error_if_no_space:1;
214
215
};

216
217
struct thin_c;
typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
218
typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell);
219
220
typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);

221
222
#define CELL_SORT_ARRAY_SIZE 8192

223
224
225
226
227
228
229
230
231
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;
232
	uint32_t sectors_per_block;
233
	int sectors_per_block_shift;
234

235
	struct pool_features pf;
236
	bool low_water_triggered:1;	/* A dm event has been sent */
237
	bool suspended:1;
238
	bool out_of_data_space:1;
239

240
	struct dm_bio_prison *prison;
241
242
243
	struct dm_kcopyd_client *copier;

	struct workqueue_struct *wq;
Joe Thornber's avatar
Joe Thornber committed
244
	struct throttle throttle;
245
	struct work_struct worker;
246
	struct delayed_work waker;
247
	struct delayed_work no_space_timeout;
248

249
	unsigned long last_commit_jiffies;
250
	unsigned ref_count;
251
252
253
254

	spinlock_t lock;
	struct bio_list deferred_flush_bios;
	struct list_head prepared_mappings;
Joe Thornber's avatar
Joe Thornber committed
255
	struct list_head prepared_discards;
256
	struct list_head active_thins;
257

258
259
	struct dm_deferred_set *shared_read_ds;
	struct dm_deferred_set *all_io_ds;
260

Mike Snitzer's avatar
Mike Snitzer committed
261
	struct dm_thin_new_mapping *next_mapping;
262
	mempool_t *mapping_pool;
263
264
265
266

	process_bio_fn process_bio;
	process_bio_fn process_discard;

267
268
269
	process_cell_fn process_cell;
	process_cell_fn process_discard_cell;

270
271
	process_mapping_fn process_prepared_mapping;
	process_mapping_fn process_prepared_discard;
272

273
	struct dm_bio_prison_cell **cell_sort_array;
274
275
};

276
static enum pool_mode get_pool_mode(struct pool *pool);
277
static void metadata_operation_failed(struct pool *pool, const char *op, int r);
278

279
280
281
282
283
284
285
286
287
288
289
/*
 * 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;
290
291
	struct pool_features requested_pf; /* Features requested during table load */
	struct pool_features adjusted_pf;  /* Features used after adjusting for constituent devices */
292
293
294
295
296
297
};

/*
 * Target context for a thin.
 */
struct thin_c {
298
	struct list_head list;
299
	struct dm_dev *pool_dev;
300
	struct dm_dev *origin_dev;
301
	sector_t origin_size;
302
303
304
305
	dm_thin_id dev_id;

	struct pool *pool;
	struct dm_thin_device *td;
306
307
	struct mapped_device *thin_md;

308
	bool requeue_mode:1;
309
	spinlock_t lock;
310
	struct list_head deferred_cells;
311
312
	struct bio_list deferred_bio_list;
	struct bio_list retry_on_resume_list;
313
	struct rb_root sort_bio_list; /* sorted list of deferred bios */
314
315
316
317
318
319
320

	/*
	 * Ensures the thin is not destroyed until the worker has finished
	 * iterating the active_thins list.
	 */
	atomic_t refcount;
	struct completion can_destroy;
321
322
323
324
};

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

Joe Thornber's avatar
Joe Thornber committed
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
static bool block_size_is_power_of_two(struct pool *pool)
{
	return pool->sectors_per_block_shift >= 0;
}

static sector_t block_to_sectors(struct pool *pool, dm_block_t b)
{
	return block_size_is_power_of_two(pool) ?
		(b << pool->sectors_per_block_shift) :
		(b * pool->sectors_per_block);
}

static int issue_discard(struct thin_c *tc, dm_block_t data_b, dm_block_t data_e,
			 struct bio *parent_bio)
{
340
	int type = REQ_WRITE | REQ_DISCARD;
Joe Thornber's avatar
Joe Thornber committed
341
342
	sector_t s = block_to_sectors(tc->pool, data_b);
	sector_t len = block_to_sectors(tc->pool, data_e - data_b);
343
344
345
346
347
348
349
350
351
352
353
354
	struct bio *bio = NULL;
	struct blk_plug plug;
	int ret;

	blk_start_plug(&plug);
	ret = __blkdev_issue_discard(tc->pool_dev->bdev, s, len,
				     GFP_NOWAIT, type, &bio);
	if (!ret && bio) {
		bio_chain(bio, parent_bio);
		submit_bio(type, bio);
	}
	blk_finish_plug(&plug);
Joe Thornber's avatar
Joe Thornber committed
355

356
	return ret;
Joe Thornber's avatar
Joe Thornber committed
357
358
359
360
}

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

361
362
363
364
365
366
367
368
369
370
371
/*
 * 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);
}

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

372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
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);
}

403
404
405
406
407
408
409
410
411
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);
}

412
413
414
415
416
417
418
419
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);
}

420
421
static void cell_error_with_code(struct pool *pool,
				 struct dm_bio_prison_cell *cell, int error_code)
422
{
423
	dm_cell_error(pool->prison, cell, error_code);
424
425
426
	dm_bio_prison_free_cell(pool->prison, cell);
}

427
428
429
430
431
static int get_pool_io_error_code(struct pool *pool)
{
	return pool->out_of_data_space ? -ENOSPC : -EIO;
}

432
433
static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell)
{
434
435
436
	int error = get_pool_io_error_code(pool);

	cell_error_with_code(pool, cell, error);
437
438
}

439
440
441
442
443
444
445
446
447
448
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);
}

449
450
/*----------------------------------------------------------------*/

451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
/*
 * 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
511
struct dm_thin_endio_hook {
512
	struct thin_c *tc;
513
514
	struct dm_deferred_entry *shared_read_entry;
	struct dm_deferred_entry *all_io_entry;
Mike Snitzer's avatar
Mike Snitzer committed
515
	struct dm_thin_new_mapping *overwrite_mapping;
516
	struct rb_node rb_node;
Joe Thornber's avatar
Joe Thornber committed
517
	struct dm_bio_prison_cell *cell;
518
519
};

520
521
522
523
524
525
526
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)
527
528
{
	struct bio *bio;
529

530
531
532
533
	while ((bio = bio_list_pop(bios))) {
		bio->bi_error = error;
		bio_endio(bio);
	}
534
535
536
537
}

static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master, int error)
{
538
	struct bio_list bios;
539
	unsigned long flags;
540
541

	bio_list_init(&bios);
542

543
	spin_lock_irqsave(&tc->lock, flags);
544
	__merge_bio_list(&bios, master);
545
	spin_unlock_irqrestore(&tc->lock, flags);
546

547
	error_bio_list(&bios, error);
548
549
}

550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
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);
}

567
568
static void requeue_io(struct thin_c *tc)
{
569
	struct bio_list bios;
570
	unsigned long flags;
571
572
573

	bio_list_init(&bios);

574
	spin_lock_irqsave(&tc->lock, flags);
575
576
	__merge_bio_list(&bios, &tc->deferred_bio_list);
	__merge_bio_list(&bios, &tc->retry_on_resume_list);
577
	spin_unlock_irqrestore(&tc->lock, flags);
578

579
580
	error_bio_list(&bios, DM_ENDIO_REQUEUE);
	requeue_deferred_cells(tc);
581
582
}

583
static void error_retry_list_with_code(struct pool *pool, int error)
584
585
586
587
588
{
	struct thin_c *tc;

	rcu_read_lock();
	list_for_each_entry_rcu(tc, &pool->active_thins, list)
589
		error_thin_bio_list(tc, &tc->retry_on_resume_list, error);
590
591
592
	rcu_read_unlock();
}

593
594
static void error_retry_list(struct pool *pool)
{
595
596
	int error = get_pool_io_error_code(pool);

597
	error_retry_list_with_code(pool, error);
598
599
}

600
601
602
603
604
605
606
607
608
/*
 * 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.
 */

static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
{
609
	struct pool *pool = tc->pool;
610
	sector_t block_nr = bio->bi_iter.bi_sector;
611

612
613
	if (block_size_is_power_of_two(pool))
		block_nr >>= pool->sectors_per_block_shift;
614
	else
615
		(void) sector_div(block_nr, pool->sectors_per_block);
616
617

	return block_nr;
618
619
}

Joe Thornber's avatar
Joe Thornber committed
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
/*
 * Returns the _complete_ blocks that this bio covers.
 */
static void get_bio_block_range(struct thin_c *tc, struct bio *bio,
				dm_block_t *begin, dm_block_t *end)
{
	struct pool *pool = tc->pool;
	sector_t b = bio->bi_iter.bi_sector;
	sector_t e = b + (bio->bi_iter.bi_size >> SECTOR_SHIFT);

	b += pool->sectors_per_block - 1ull; /* so we round up */

	if (block_size_is_power_of_two(pool)) {
		b >>= pool->sectors_per_block_shift;
		e >>= pool->sectors_per_block_shift;
	} else {
		(void) sector_div(b, pool->sectors_per_block);
		(void) sector_div(e, pool->sectors_per_block);
	}

	if (e < b)
		/* Can happen if the bio is within a single block. */
		e = b;

	*begin = b;
	*end = e;
}

648
649
650
static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
{
	struct pool *pool = tc->pool;
651
	sector_t bi_sector = bio->bi_iter.bi_sector;
652
653

	bio->bi_bdev = tc->pool_dev->bdev;
654
	if (block_size_is_power_of_two(pool))
655
656
657
		bio->bi_iter.bi_sector =
			(block << pool->sectors_per_block_shift) |
			(bi_sector & (pool->sectors_per_block - 1));
658
	else
659
		bio->bi_iter.bi_sector = (block * pool->sectors_per_block) +
660
				 sector_div(bi_sector, pool->sectors_per_block);
661
662
}

663
664
665
666
667
static void remap_to_origin(struct thin_c *tc, struct bio *bio)
{
	bio->bi_bdev = tc->origin_dev->bdev;
}

668
669
670
671
672
673
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);
}

674
675
676
677
678
679
680
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;

681
	h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
682
683
684
	h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
}

685
static void issue(struct thin_c *tc, struct bio *bio)
686
687
688
689
{
	struct pool *pool = tc->pool;
	unsigned long flags;

690
691
692
693
694
	if (!bio_triggers_commit(tc, bio)) {
		generic_make_request(bio);
		return;
	}

695
	/*
696
697
698
	 * 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.
699
	 */
700
701
702
703
704
705
706
707
708
709
710
711
	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);
712
713
}

714
715
716
717
718
719
720
721
722
723
724
725
726
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);
}

727
728
729
730
731
/*----------------------------------------------------------------*/

/*
 * Bio endio functions.
 */
Mike Snitzer's avatar
Mike Snitzer committed
732
struct dm_thin_new_mapping {
733
734
	struct list_head list;

735
	bool pass_discard:1;
Joe Thornber's avatar
Joe Thornber committed
736
	bool maybe_shared:1;
737

738
739
740
741
742
743
744
	/*
	 * 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;

745
	int err;
746
	struct thin_c *tc;
Joe Thornber's avatar
Joe Thornber committed
747
	dm_block_t virt_begin, virt_end;
748
	dm_block_t data_block;
Joe Thornber's avatar
Joe Thornber committed
749
	struct dm_bio_prison_cell *cell;
750
751
752
753
754
755
756
757
758
759
760

	/*
	 * 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;
};

761
static void __complete_mapping_preparation(struct dm_thin_new_mapping *m)
762
763
764
{
	struct pool *pool = m->tc->pool;

765
	if (atomic_dec_and_test(&m->prepare_actions)) {
766
		list_add_tail(&m->list, &pool->prepared_mappings);
767
768
769
770
		wake_worker(pool);
	}
}

771
static void complete_mapping_preparation(struct dm_thin_new_mapping *m)
772
773
774
775
776
{
	unsigned long flags;
	struct pool *pool = m->tc->pool;

	spin_lock_irqsave(&pool->lock, flags);
777
	__complete_mapping_preparation(m);
778
779
780
	spin_unlock_irqrestore(&pool->lock, flags);
}

781
782
783
784
785
786
787
788
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);
}

789
static void overwrite_endio(struct bio *bio)
790
{
791
	struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Mike Snitzer's avatar
Mike Snitzer committed
792
	struct dm_thin_new_mapping *m = h->overwrite_mapping;
793

794
795
	bio->bi_end_io = m->saved_bi_end_io;

796
	m->err = bio->bi_error;
797
	complete_mapping_preparation(m);
798
799
800
801
802
803
804
805
806
807
808
809
810
}

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

/*
 * Workqueue.
 */

/*
 * Prepared mapping jobs.
 */

/*
811
812
 * This sends the bios in the cell, except the original holder, back
 * to the deferred_bios list.
813
 */
814
static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
815
816
817
818
{
	struct pool *pool = tc->pool;
	unsigned long flags;

819
820
821
	spin_lock_irqsave(&tc->lock, flags);
	cell_release_no_holder(pool, cell, &tc->deferred_bio_list);
	spin_unlock_irqrestore(&tc->lock, flags);
822
823
824
825

	wake_worker(pool);
}

826
827
static void thin_defer_bio(struct thin_c *tc, struct bio *bio);

828
829
830
831
832
833
834
835
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)
836
{
837
	struct remap_info *info = context;
838
839
	struct bio *bio;

840
	while ((bio = bio_list_pop(&cell->bios))) {
841
		if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA))
842
			bio_list_add(&info->defer_bios, bio);
843
		else {
844
845
846
847
848
849
850
851
			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);
852
853
854
855
		}
	}
}

856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
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);
}

882
883
static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
{
884
	cell_error(m->tc->pool, m->cell);
885
886
887
	list_del(&m->list);
	mempool_free(m, m->tc->pool->mapping_pool);
}
888

Mike Snitzer's avatar
Mike Snitzer committed
889
static void process_prepared_mapping(struct dm_thin_new_mapping *m)
890
891
{
	struct thin_c *tc = m->tc;
892
	struct pool *pool = tc->pool;
893
	struct bio *bio = m->bio;
894
895
896
	int r;

	if (m->err) {
897
		cell_error(pool, m->cell);
898
		goto out;
899
900
901
902
903
904
905
	}

	/*
	 * Commit the prepared block into the mapping btree.
	 * Any I/O for this block arriving after this point will get
	 * remapped to it directly.
	 */
Joe Thornber's avatar
Joe Thornber committed
906
	r = dm_thin_insert_block(tc->td, m->virt_begin, m->data_block);
907
	if (r) {
908
		metadata_operation_failed(pool, "dm_thin_insert_block", r);
909
		cell_error(pool, m->cell);
910
		goto out;
911
912
913
914
915
916
917
918
919
	}

	/*
	 * 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) {
920
		inc_remap_and_issue_cell(tc, m->cell, m->data_block);
921
		bio_endio(bio);
922
923
924
925
926
	} 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);
	}
927

928
out:
929
	list_del(&m->list);
930
	mempool_free(m, pool->mapping_pool);
931
932
}

Joe Thornber's avatar
Joe Thornber committed
933
934
935
/*----------------------------------------------------------------*/

static void free_discard_mapping(struct dm_thin_new_mapping *m)
Joe Thornber's avatar
Joe Thornber committed
936
937
{
	struct thin_c *tc = m->tc;
Joe Thornber's avatar
Joe Thornber committed
938
939
940
941
	if (m->cell)
		cell_defer_no_holder(tc, m->cell);
	mempool_free(m, tc->pool->mapping_pool);
}
Joe Thornber's avatar
Joe Thornber committed
942

Joe Thornber's avatar
Joe Thornber committed
943
944
static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
{
945
	bio_io_error(m->bio);
Joe Thornber's avatar
Joe Thornber committed
946
947
948
949
950
	free_discard_mapping(m);
}

static void process_prepared_discard_success(struct dm_thin_new_mapping *m)
{
951
	bio_endio(m->bio);
Joe Thornber's avatar
Joe Thornber committed
952
953
954
955
956
957
958
959
960
961
962
963
964
	free_discard_mapping(m);
}

static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m)
{
	int r;
	struct thin_c *tc = m->tc;

	r = dm_thin_remove_range(tc->td, m->cell->key.block_begin, m->cell->key.block_end);
	if (r) {
		metadata_operation_failed(tc->pool, "dm_thin_remove_range", r);
		bio_io_error(m->bio);
	} else
965
		bio_endio(m->bio);
Joe Thornber's avatar
Joe Thornber committed
966

967
	cell_defer_no_holder(tc, m->cell);
968
969
970
	mempool_free(m, tc->pool->mapping_pool);
}

Joe Thornber's avatar
Joe Thornber committed
971
static int passdown_double_checking_shared_status(struct dm_thin_new_mapping *m)
972
{
Joe Thornber's avatar
Joe Thornber committed
973
974
975
976
977
978
	/*
	 * We've already unmapped this range of blocks, but before we
	 * passdown we have to check that these blocks are now unused.
	 */
	int r;
	bool used = true;
979
	struct thin_c *tc = m->tc;
Joe Thornber's avatar
Joe Thornber committed
980
981
	struct pool *pool = tc->pool;
	dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin;
Joe Thornber's avatar
Joe Thornber committed
982

Joe Thornber's avatar
Joe Thornber committed
983
984
985
986
987
988
	while (b != end) {
		/* find start of unmapped run */
		for (; b < end; b++) {
			r = dm_pool_block_is_used(pool->pmd, b, &used);
			if (r)
				return r;
989

Joe Thornber's avatar
Joe Thornber committed
990
991
			if (!used)
				break;
992
		}
Joe Thornber's avatar
Joe Thornber committed
993

Joe Thornber's avatar
Joe Thornber committed
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
		if (b == end)
			break;

		/* find end of run */
		for (e = b + 1; e != end; e++) {
			r = dm_pool_block_is_used(pool->pmd, e, &used);
			if (r)
				return r;

			if (used)
				break;
		}

		r = issue_discard(tc, b, e, m->bio);
		if (r)
			return r;

		b = e;
	}

	return 0;
Joe Thornber's avatar
Joe Thornber committed
1015
1016
}

Joe Thornber's avatar
Joe Thornber committed
1017
static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m)
1018
1019
1020
{
	int r;
	struct thin_c *tc = m->tc;
Joe Thornber's avatar
Joe Thornber committed
1021
	struct pool *pool = tc->pool;
1022

Joe Thornber's avatar
Joe Thornber committed
1023
	r = dm_thin_remove_range(tc->td, m->virt_begin, m->virt_end);
1024
	if (r)
Joe Thornber's avatar
Joe Thornber committed
1025
1026
1027
1028
1029
1030
		metadata_operation_failed(pool, "dm_thin_remove_range", r);

	else if (m->maybe_shared)
		r = passdown_double_checking_shared_status(m);
	else
		r = issue_discard(tc, m->data_block, m->data_block + (m->virt_end - m->virt_begin), m->bio);
1031

Joe Thornber's avatar
Joe Thornber committed
1032
1033
1034
1035
	/*
	 * Even if r is set, there could be sub discards in flight that we
	 * need to wait for.
	 */
1036
1037
	m->bio->bi_error = r;
	bio_endio(m->bio);
Joe Thornber's avatar
Joe Thornber committed
1038
1039
	cell_defer_no_holder(tc, m->cell);
	mempool_free(m, pool->mapping_pool);
1040
1041
}

Joe Thornber's avatar
Joe Thornber committed
1042
static void process_prepared(struct pool *pool, struct list_head *head,
1043
			     process_mapping_fn *fn)
1044
1045
1046
{
	unsigned long flags;
	struct list_head maps;
Mike Snitzer's avatar
Mike Snitzer committed
1047
	struct dm_thin_new_mapping *m, *tmp;
1048
1049
1050

	INIT_LIST_HEAD(&maps);
	spin_lock_irqsave(&pool->lock, flags);
Joe Thornber's avatar
Joe Thornber committed
1051
	list_splice_init(head, &maps);
1052
1053
1054
	spin_unlock_irqrestore(&pool->lock, flags);

	list_for_each_entry_safe(m, tmp, &maps, list)
1055
		(*fn)(m);
1056
1057
1058
1059
1060
}

/*
 * Deferred bio jobs.
 */
Joe Thornber's avatar
Joe Thornber committed
1061
static int io_overlaps_block(struct pool *pool, struct bio *bio)
1062
{
1063
1064
	return bio->bi_iter.bi_size ==
		(pool->sectors_per_block << SECTOR_SHIFT);
Joe Thornber's avatar
Joe Thornber committed
1065
1066
1067
1068
1069
1070
}

static int io_overwrites_block(struct pool *pool, struct bio *bio)
{
	return (bio_data_dir(bio) == WRITE) &&
		io_overlaps_block(pool, bio);
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
}

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
1090
static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
1091
{
1092
	struct dm_thin_new_mapping *m = pool->next_mapping;
1093
1094
1095

	BUG_ON(!pool->next_mapping);

1096
1097
1098
1099
	memset(m, 0, sizeof(struct dm_thin_new_mapping));
	INIT_LIST_HEAD(&m->list);
	m->bio = NULL;

1100
1101
	pool->next_mapping = NULL;

1102
	return m;
1103
1104
}

1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
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);
	}
}

1122
static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio,
Joe Thornber's avatar
Joe Thornber committed
1123
				      dm_block_t data_begin,
1124
1125
1126
1127
1128
1129
1130
1131
1132
				      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);
Joe Thornber's avatar
Joe Thornber committed
1133
	remap_and_issue(tc, bio, data_begin);
1134
1135
}

1136
1137
1138
/*
 * A partial copy also needs to zero the uncopied region.
 */
1139
static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
1140
1141
			  struct dm_dev *origin, dm_block_t data_origin,
			  dm_block_t data_dest,
1142
1143
			  struct dm_bio_prison_cell *cell, struct bio *bio,
			  sector_t len)
1144
1145
1146
{
	int r;
	struct pool *pool = tc->pool;
Mike Snitzer's avatar
Mike Snitzer committed
1147
	struct dm_thin_new_mapping *m = get_next_mapping(pool);
1148
1149

	m->tc = tc;
Joe Thornber's avatar
Joe Thornber committed
1150
1151
	m->virt_begin = virt_block;
	m->virt_end = virt_block + 1u;
1152
1153
1154
	m->data_block = data_dest;
	m->cell = cell;

1155
1156
1157
1158
1159
1160
1161
	/*
	 * 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);

1162
	if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
1163
		complete_mapping_preparation(m); /* already quiesced */
1164
1165
1166
1167
1168
1169
1170

	/*
	 * 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.
	 */
1171
1172
1173
	if (io_overwrites_block(pool, bio))
		remap_and_issue_overwrite(tc, bio, data_dest, m);
	else {
1174
1175
		struct dm_io_region from, to;

1176
		from.bdev = origin->bdev;
1177
		from.sector = data_origin * pool->sectors_per_block;
1178
		from.count = len;
1179
1180
1181

		to.bdev = tc->pool_dev->bdev;
		to.sector = data_dest * pool->sectors_per_block;
1182
		to.count = len;
1183
1184
1185
1186

		r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
				   0, copy_complete, m);
		if (r < 0) {
1187
			DMERR_LIMIT("dm_kcopyd_copy() failed");
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
			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);
1206
1207
		}
	}
1208
1209

	complete_mapping_preparation(m); /* drop our ref */
1210
1211
}

1212
1213
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
1214
				   struct dm_bio_prison_cell *cell, struct bio *bio)
1215
1216
{
	schedule_copy(tc, virt_block, tc->pool_dev,
1217
1218
		      data_origin, data_dest, cell, bio,
		      tc->pool->sectors_per_block);
1219
1220
}

1221
static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
Mike Snitzer's avatar
Mike Snitzer committed
1222
			  dm_block_t data_block, struct dm_bio_prison_cell *cell,
1223
1224
1225
			  struct bio *bio)
{
	struct pool *pool = tc->pool;
Mike Snitzer's avatar
Mike Snitzer committed
1226
	struct dm_thin_new_mapping *m = get_next_mapping(pool);
1227

1228
	atomic_set(&m->prepare_actions, 1); /* no need to quiesce */
1229
	m->tc = tc;
Joe Thornber's avatar
Joe Thornber committed
1230
1231
	m->virt_begin = virt_block;
	m->virt_end = virt_block + 1u;
1232
1233
1234
1235
1236
1237
1238
1239
	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.
	 */
1240
1241
1242
1243
1244
1245
1246
	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
1247
		process_prepared_mapping(m);
1248
}
1249

1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
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);
1270
1271
}

1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
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);
}

1290
1291
1292
1293
/*
 * A non-zero return indicates read_only or fail_io mode.
 * Many callers don't care about the return value.
 */
1294
static int commit(struct pool *pool)
1295
1296
1297
{
	int r;

1298
	if (get_pool_mode(pool) >= PM_READ_ONLY)
1299
1300
		return -EINVAL;

1301
	r = dm_pool_commit_metadata(pool->pmd);
1302
1303
	if (r)
		metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
1304
1305
	else
		check_for_space(pool);
1306