dm-thin.c 68.8 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
/*
 * Copyright (C) 2011 Red Hat UK.
 *
 * This file is released under the GPL.
 */

#include "dm-thin-metadata.h"

#include <linux/device-mapper.h>
#include <linux/dm-io.h>
#include <linux/dm-kcopyd.h>
#include <linux/list.h>
#include <linux/init.h>
#include <linux/module.h>
#include <linux/slab.h>

#define	DM_MSG_PREFIX	"thin"

/*
 * Tunable constants
 */
22
#define ENDIO_HOOK_POOL_SIZE 1024
23
24
25
#define DEFERRED_SET_SIZE 64
#define MAPPING_POOL_SIZE 1024
#define PRISON_CELLS 1024
26
#define COMMIT_PERIOD HZ
27
28
29
30
31
32
33
34
35
36
37
38
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

/*
 * 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
 * including all devices that share this block.  (see deferred_set code)
 *
 * 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
66
 * (process_prepared_mapping).  This act of inserting breaks some
67
68
69
70
71
72
73
74
75
76
77
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
 * 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.
 */

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

/*
 * Sometimes we can't deal with a bio straight away.  We put them in prison
 * where they can't cause any mischief.  Bios are put in a cell identified
 * by a key, multiple bios can be in the same cell.  When the cell is
 * subsequently unlocked the bios become available.
 */
struct bio_prison;

struct cell_key {
	int virtual;
	dm_thin_id dev;
	dm_block_t block;
};

Mike Snitzer's avatar
Mike Snitzer committed
114
struct dm_bio_prison_cell {
115
116
117
	struct hlist_node list;
	struct bio_prison *prison;
	struct cell_key key;
118
	struct bio *holder;
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
	struct bio_list bios;
};

struct bio_prison {
	spinlock_t lock;
	mempool_t *cell_pool;

	unsigned nr_buckets;
	unsigned hash_mask;
	struct hlist_head *cells;
};

static uint32_t calc_nr_buckets(unsigned nr_cells)
{
	uint32_t n = 128;

	nr_cells /= 4;
	nr_cells = min(nr_cells, 8192u);

	while (n < nr_cells)
		n <<= 1;

	return n;
}

Mike Snitzer's avatar
Mike Snitzer committed
144
145
static struct kmem_cache *_cell_cache;

146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
/*
 * @nr_cells should be the number of cells you want in use _concurrently_.
 * Don't confuse it with the number of distinct keys.
 */
static struct bio_prison *prison_create(unsigned nr_cells)
{
	unsigned i;
	uint32_t nr_buckets = calc_nr_buckets(nr_cells);
	size_t len = sizeof(struct bio_prison) +
		(sizeof(struct hlist_head) * nr_buckets);
	struct bio_prison *prison = kmalloc(len, GFP_KERNEL);

	if (!prison)
		return NULL;

	spin_lock_init(&prison->lock);
Mike Snitzer's avatar
Mike Snitzer committed
162
	prison->cell_pool = mempool_create_slab_pool(nr_cells, _cell_cache);
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
190
191
192
193
194
195
196
197
	if (!prison->cell_pool) {
		kfree(prison);
		return NULL;
	}

	prison->nr_buckets = nr_buckets;
	prison->hash_mask = nr_buckets - 1;
	prison->cells = (struct hlist_head *) (prison + 1);
	for (i = 0; i < nr_buckets; i++)
		INIT_HLIST_HEAD(prison->cells + i);

	return prison;
}

static void prison_destroy(struct bio_prison *prison)
{
	mempool_destroy(prison->cell_pool);
	kfree(prison);
}

static uint32_t hash_key(struct bio_prison *prison, struct cell_key *key)
{
	const unsigned long BIG_PRIME = 4294967291UL;
	uint64_t hash = key->block * BIG_PRIME;

	return (uint32_t) (hash & prison->hash_mask);
}

static int keys_equal(struct cell_key *lhs, struct cell_key *rhs)
{
	       return (lhs->virtual == rhs->virtual) &&
		       (lhs->dev == rhs->dev) &&
		       (lhs->block == rhs->block);
}

Mike Snitzer's avatar
Mike Snitzer committed
198
199
static struct dm_bio_prison_cell *__search_bucket(struct hlist_head *bucket,
						  struct cell_key *key)
200
{
Mike Snitzer's avatar
Mike Snitzer committed
201
	struct dm_bio_prison_cell *cell;
202
203
204
205
206
207
208
209
210
211
212
213
214
	struct hlist_node *tmp;

	hlist_for_each_entry(cell, tmp, bucket, list)
		if (keys_equal(&cell->key, key))
			return cell;

	return NULL;
}

/*
 * This may block if a new cell needs allocating.  You must ensure that
 * cells will be unlocked even if the calling thread is blocked.
 *
215
 * Returns 1 if the cell was already held, 0 if @inmate is the new holder.
216
217
 */
static int bio_detain(struct bio_prison *prison, struct cell_key *key,
Mike Snitzer's avatar
Mike Snitzer committed
218
		      struct bio *inmate, struct dm_bio_prison_cell **ref)
219
{
220
	int r = 1;
221
222
	unsigned long flags;
	uint32_t hash = hash_key(prison, key);
Mike Snitzer's avatar
Mike Snitzer committed
223
	struct dm_bio_prison_cell *cell, *cell2;
224
225
226
227
228

	BUG_ON(hash > prison->nr_buckets);

	spin_lock_irqsave(&prison->lock, flags);

229
230
231
232
	cell = __search_bucket(prison->cells + hash, key);
	if (cell) {
		bio_list_add(&cell->bios, inmate);
		goto out;
233
234
	}

235
236
237
	/*
	 * Allocate a new cell
	 */
238
	spin_unlock_irqrestore(&prison->lock, flags);
239
240
	cell2 = mempool_alloc(prison->cell_pool, GFP_NOIO);
	spin_lock_irqsave(&prison->lock, flags);
241

242
243
244
245
246
247
	/*
	 * We've been unlocked, so we have to double check that
	 * nobody else has inserted this cell in the meantime.
	 */
	cell = __search_bucket(prison->cells + hash, key);
	if (cell) {
248
		mempool_free(cell2, prison->cell_pool);
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
		bio_list_add(&cell->bios, inmate);
		goto out;
	}

	/*
	 * Use new cell.
	 */
	cell = cell2;

	cell->prison = prison;
	memcpy(&cell->key, key, sizeof(cell->key));
	cell->holder = inmate;
	bio_list_init(&cell->bios);
	hlist_add_head(&cell->list, prison->cells + hash);

	r = 0;

out:
	spin_unlock_irqrestore(&prison->lock, flags);
268
269
270
271
272
273
274
275
276

	*ref = cell;

	return r;
}

/*
 * @inmates must have been initialised prior to this call
 */
Mike Snitzer's avatar
Mike Snitzer committed
277
static void __cell_release(struct dm_bio_prison_cell *cell, struct bio_list *inmates)
278
279
280
281
282
{
	struct bio_prison *prison = cell->prison;

	hlist_del(&cell->list);

283
284
285
286
	if (inmates) {
		bio_list_add(inmates, cell->holder);
		bio_list_merge(inmates, &cell->bios);
	}
287
288
289
290

	mempool_free(cell, prison->cell_pool);
}

Mike Snitzer's avatar
Mike Snitzer committed
291
static void cell_release(struct dm_bio_prison_cell *cell, struct bio_list *bios)
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
{
	unsigned long flags;
	struct bio_prison *prison = cell->prison;

	spin_lock_irqsave(&prison->lock, flags);
	__cell_release(cell, bios);
	spin_unlock_irqrestore(&prison->lock, flags);
}

/*
 * There are a couple of places where we put a bio into a cell briefly
 * before taking it out again.  In these situations we know that no other
 * bio may be in the cell.  This function releases the cell, and also does
 * a sanity check.
 */
Mike Snitzer's avatar
Mike Snitzer committed
307
static void __cell_release_singleton(struct dm_bio_prison_cell *cell, struct bio *bio)
308
309
310
{
	BUG_ON(cell->holder != bio);
	BUG_ON(!bio_list_empty(&cell->bios));
311
312

	__cell_release(cell, NULL);
313
314
}

Mike Snitzer's avatar
Mike Snitzer committed
315
static void cell_release_singleton(struct dm_bio_prison_cell *cell, struct bio *bio)
316
317
{
	unsigned long flags;
318
	struct bio_prison *prison = cell->prison;
319
320

	spin_lock_irqsave(&prison->lock, flags);
321
	__cell_release_singleton(cell, bio);
322
	spin_unlock_irqrestore(&prison->lock, flags);
323
324
325
326
327
}

/*
 * Sometimes we don't want the holder, just the additional bios.
 */
Mike Snitzer's avatar
Mike Snitzer committed
328
329
static void __cell_release_no_holder(struct dm_bio_prison_cell *cell,
				     struct bio_list *inmates)
330
331
332
333
334
335
336
337
338
{
	struct bio_prison *prison = cell->prison;

	hlist_del(&cell->list);
	bio_list_merge(inmates, &cell->bios);

	mempool_free(cell, prison->cell_pool);
}

Mike Snitzer's avatar
Mike Snitzer committed
339
340
static void cell_release_no_holder(struct dm_bio_prison_cell *cell,
				   struct bio_list *inmates)
341
342
343
{
	unsigned long flags;
	struct bio_prison *prison = cell->prison;
344

345
346
347
	spin_lock_irqsave(&prison->lock, flags);
	__cell_release_no_holder(cell, inmates);
	spin_unlock_irqrestore(&prison->lock, flags);
348
349
}

Mike Snitzer's avatar
Mike Snitzer committed
350
static void cell_error(struct dm_bio_prison_cell *cell)
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
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
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
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
{
	struct bio_prison *prison = cell->prison;
	struct bio_list bios;
	struct bio *bio;
	unsigned long flags;

	bio_list_init(&bios);

	spin_lock_irqsave(&prison->lock, flags);
	__cell_release(cell, &bios);
	spin_unlock_irqrestore(&prison->lock, flags);

	while ((bio = bio_list_pop(&bios)))
		bio_io_error(bio);
}

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

/*
 * We use the deferred set to keep track of pending reads to shared blocks.
 * We do this to ensure the new mapping caused by a write isn't performed
 * until these prior reads have completed.  Otherwise the insertion of the
 * new mapping could free the old block that the read bios are mapped to.
 */

struct deferred_set;
struct deferred_entry {
	struct deferred_set *ds;
	unsigned count;
	struct list_head work_items;
};

struct deferred_set {
	spinlock_t lock;
	unsigned current_entry;
	unsigned sweeper;
	struct deferred_entry entries[DEFERRED_SET_SIZE];
};

static void ds_init(struct deferred_set *ds)
{
	int i;

	spin_lock_init(&ds->lock);
	ds->current_entry = 0;
	ds->sweeper = 0;
	for (i = 0; i < DEFERRED_SET_SIZE; i++) {
		ds->entries[i].ds = ds;
		ds->entries[i].count = 0;
		INIT_LIST_HEAD(&ds->entries[i].work_items);
	}
}

static struct deferred_entry *ds_inc(struct deferred_set *ds)
{
	unsigned long flags;
	struct deferred_entry *entry;

	spin_lock_irqsave(&ds->lock, flags);
	entry = ds->entries + ds->current_entry;
	entry->count++;
	spin_unlock_irqrestore(&ds->lock, flags);

	return entry;
}

static unsigned ds_next(unsigned index)
{
	return (index + 1) % DEFERRED_SET_SIZE;
}

static void __sweep(struct deferred_set *ds, struct list_head *head)
{
	while ((ds->sweeper != ds->current_entry) &&
	       !ds->entries[ds->sweeper].count) {
		list_splice_init(&ds->entries[ds->sweeper].work_items, head);
		ds->sweeper = ds_next(ds->sweeper);
	}

	if ((ds->sweeper == ds->current_entry) && !ds->entries[ds->sweeper].count)
		list_splice_init(&ds->entries[ds->sweeper].work_items, head);
}

static void ds_dec(struct deferred_entry *entry, struct list_head *head)
{
	unsigned long flags;

	spin_lock_irqsave(&entry->ds->lock, flags);
	BUG_ON(!entry->count);
	--entry->count;
	__sweep(entry->ds, head);
	spin_unlock_irqrestore(&entry->ds->lock, flags);
}

/*
 * Returns 1 if deferred or 0 if no pending items to delay job.
 */
static int ds_add_work(struct deferred_set *ds, struct list_head *work)
{
	int r = 1;
	unsigned long flags;
	unsigned next_entry;

	spin_lock_irqsave(&ds->lock, flags);
	if ((ds->sweeper == ds->current_entry) &&
	    !ds->entries[ds->current_entry].count)
		r = 0;
	else {
		list_add(work, &ds->entries[ds->current_entry].work_items);
		next_entry = ds_next(ds->current_entry);
		if (!ds->entries[next_entry].count)
			ds->current_entry = next_entry;
	}
	spin_unlock_irqrestore(&ds->lock, flags);

	return r;
}

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

/*
 * Key building.
 */
static void build_data_key(struct dm_thin_device *td,
			   dm_block_t b, struct cell_key *key)
{
	key->virtual = 0;
	key->dev = dm_thin_dev_id(td);
	key->block = b;
}

static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
			      struct cell_key *key)
{
	key->virtual = 1;
	key->dev = dm_thin_dev_id(td);
	key->block = b;
}

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

/*
 * 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
497
struct dm_thin_new_mapping;
498
499
500
501
502
503
504

struct pool_features {
	unsigned zero_new_blocks:1;
	unsigned discard_enabled:1;
	unsigned discard_passdown:1;
};

505
506
507
508
509
510
511
512
513
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;
514
	uint32_t sectors_per_block;
515

516
	struct pool_features pf;
517
518
519
520
521
522
523
524
	unsigned low_water_triggered:1;	/* A dm event has been sent */
	unsigned no_free_space:1;	/* A -ENOSPC warning has been issued */

	struct bio_prison *prison;
	struct dm_kcopyd_client *copier;

	struct workqueue_struct *wq;
	struct work_struct worker;
525
	struct delayed_work waker;
526

527
	unsigned long last_commit_jiffies;
528
	unsigned ref_count;
529
530
531
532
533

	spinlock_t lock;
	struct bio_list deferred_bios;
	struct bio_list deferred_flush_bios;
	struct list_head prepared_mappings;
Joe Thornber's avatar
Joe Thornber committed
534
	struct list_head prepared_discards;
535
536
537

	struct bio_list retry_on_resume_list;

538
	struct deferred_set shared_read_ds;
Joe Thornber's avatar
Joe Thornber committed
539
	struct deferred_set all_io_ds;
540

Mike Snitzer's avatar
Mike Snitzer committed
541
	struct dm_thin_new_mapping *next_mapping;
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
	mempool_t *mapping_pool;
	mempool_t *endio_hook_pool;
};

/*
 * 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;
557
	struct pool_features pf;
558
559
560
561
562
563
564
};

/*
 * Target context for a thin.
 */
struct thin_c {
	struct dm_dev *pool_dev;
565
	struct dm_dev *origin_dev;
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
	dm_thin_id dev_id;

	struct pool *pool;
	struct dm_thin_device *td;
};

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

/*
 * 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
634
struct dm_thin_endio_hook {
635
636
	struct thin_c *tc;
	struct deferred_entry *shared_read_entry;
Joe Thornber's avatar
Joe Thornber committed
637
	struct deferred_entry *all_io_entry;
Mike Snitzer's avatar
Mike Snitzer committed
638
	struct dm_thin_new_mapping *overwrite_mapping;
639
640
};

641
642
643
644
645
646
647
648
649
650
static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master)
{
	struct bio *bio;
	struct bio_list bios;

	bio_list_init(&bios);
	bio_list_merge(&bios, master);
	bio_list_init(master);

	while ((bio = bio_list_pop(&bios))) {
Mike Snitzer's avatar
Mike Snitzer committed
651
652
		struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;

653
		if (h->tc == tc)
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
			bio_endio(bio, DM_ENDIO_REQUEUE);
		else
			bio_list_add(master, bio);
	}
}

static void requeue_io(struct thin_c *tc)
{
	struct pool *pool = tc->pool;
	unsigned long flags;

	spin_lock_irqsave(&pool->lock, flags);
	__requeue_bio_list(tc, &pool->deferred_bios);
	__requeue_bio_list(tc, &pool->retry_on_resume_list);
	spin_unlock_irqrestore(&pool->lock, flags);
}

/*
 * 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)
{
680
681
682
683
684
	sector_t block_nr = bio->bi_sector;

	(void) sector_div(block_nr, tc->pool->sectors_per_block);

	return block_nr;
685
686
687
688
689
}

static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
{
	struct pool *pool = tc->pool;
690
	sector_t bi_sector = bio->bi_sector;
691
692

	bio->bi_bdev = tc->pool_dev->bdev;
693
694
	bio->bi_sector = (block * pool->sectors_per_block) +
			 sector_div(bi_sector, pool->sectors_per_block);
695
696
}

697
698
699
700
701
702
static void remap_to_origin(struct thin_c *tc, struct bio *bio)
{
	bio->bi_bdev = tc->origin_dev->bdev;
}

static void issue(struct thin_c *tc, struct bio *bio)
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
{
	struct pool *pool = tc->pool;
	unsigned long flags;

	/*
	 * Batch together any FUA/FLUSH bios we find and then issue
	 * a single commit for them in process_deferred_bios().
	 */
	if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
		spin_lock_irqsave(&pool->lock, flags);
		bio_list_add(&pool->deferred_flush_bios, bio);
		spin_unlock_irqrestore(&pool->lock, flags);
	} else
		generic_make_request(bio);
}

719
720
721
722
723
724
725
726
727
728
729
730
731
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);
}

732
733
734
735
736
737
738
739
740
741
742
743
744
745
/*
 * 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);
}

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

/*
 * Bio endio functions.
 */
Mike Snitzer's avatar
Mike Snitzer committed
746
struct dm_thin_new_mapping {
747
748
	struct list_head list;

749
750
	unsigned quiesced:1;
	unsigned prepared:1;
Joe Thornber's avatar
Joe Thornber committed
751
	unsigned pass_discard:1;
752
753
754
755

	struct thin_c *tc;
	dm_block_t virt_block;
	dm_block_t data_block;
Mike Snitzer's avatar
Mike Snitzer committed
756
	struct dm_bio_prison_cell *cell, *cell2;
757
758
759
760
761
762
763
764
765
766
767
768
	int err;

	/*
	 * 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;
};

Mike Snitzer's avatar
Mike Snitzer committed
769
static void __maybe_add_mapping(struct dm_thin_new_mapping *m)
770
771
772
{
	struct pool *pool = m->tc->pool;

773
	if (m->quiesced && m->prepared) {
774
775
776
777
778
779
780
781
		list_add(&m->list, &pool->prepared_mappings);
		wake_worker(pool);
	}
}

static void copy_complete(int read_err, unsigned long write_err, void *context)
{
	unsigned long flags;
Mike Snitzer's avatar
Mike Snitzer committed
782
	struct dm_thin_new_mapping *m = context;
783
784
785
786
787
788
789
790
791
792
793
794
795
	struct pool *pool = m->tc->pool;

	m->err = read_err || write_err ? -EIO : 0;

	spin_lock_irqsave(&pool->lock, flags);
	m->prepared = 1;
	__maybe_add_mapping(m);
	spin_unlock_irqrestore(&pool->lock, flags);
}

static void overwrite_endio(struct bio *bio, int err)
{
	unsigned long flags;
Mike Snitzer's avatar
Mike Snitzer committed
796
797
	struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
	struct dm_thin_new_mapping *m = h->overwrite_mapping;
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
	struct pool *pool = m->tc->pool;

	m->err = err;

	spin_lock_irqsave(&pool->lock, flags);
	m->prepared = 1;
	__maybe_add_mapping(m);
	spin_unlock_irqrestore(&pool->lock, flags);
}

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

/*
 * Workqueue.
 */

/*
 * Prepared mapping jobs.
 */

/*
 * This sends the bios in the cell back to the deferred_bios list.
 */
Mike Snitzer's avatar
Mike Snitzer committed
821
static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell,
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
		       dm_block_t data_block)
{
	struct pool *pool = tc->pool;
	unsigned long flags;

	spin_lock_irqsave(&pool->lock, flags);
	cell_release(cell, &pool->deferred_bios);
	spin_unlock_irqrestore(&tc->pool->lock, flags);

	wake_worker(pool);
}

/*
 * Same as cell_defer above, except it omits one particular detainee,
 * a write bio that covers the block and has already been processed.
 */
Mike Snitzer's avatar
Mike Snitzer committed
838
static void cell_defer_except(struct thin_c *tc, struct dm_bio_prison_cell *cell)
839
840
841
842
843
844
845
846
{
	struct bio_list bios;
	struct pool *pool = tc->pool;
	unsigned long flags;

	bio_list_init(&bios);

	spin_lock_irqsave(&pool->lock, flags);
847
	cell_release_no_holder(cell, &pool->deferred_bios);
848
849
850
851
852
	spin_unlock_irqrestore(&pool->lock, flags);

	wake_worker(pool);
}

Mike Snitzer's avatar
Mike Snitzer committed
853
static void process_prepared_mapping(struct dm_thin_new_mapping *m)
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
882
883
884
885
886
{
	struct thin_c *tc = m->tc;
	struct bio *bio;
	int r;

	bio = m->bio;
	if (bio)
		bio->bi_end_io = m->saved_bi_end_io;

	if (m->err) {
		cell_error(m->cell);
		return;
	}

	/*
	 * 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) {
		DMERR("dm_thin_insert_block() failed");
		cell_error(m->cell);
		return;
	}

	/*
	 * 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) {
887
		cell_defer_except(tc, m->cell);
888
889
890
891
892
893
894
895
		bio_endio(bio, 0);
	} else
		cell_defer(tc, m->cell, m->data_block);

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

Mike Snitzer's avatar
Mike Snitzer committed
896
static void process_prepared_discard(struct dm_thin_new_mapping *m)
Joe Thornber's avatar
Joe Thornber committed
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
{
	int r;
	struct thin_c *tc = m->tc;

	r = dm_thin_remove_block(tc->td, m->virt_block);
	if (r)
		DMERR("dm_thin_remove_block() failed");

	/*
	 * Pass the discard down to the underlying device?
	 */
	if (m->pass_discard)
		remap_and_issue(tc, m->bio, m->data_block);
	else
		bio_endio(m->bio, 0);

	cell_defer_except(tc, m->cell);
	cell_defer_except(tc, m->cell2);
	mempool_free(m, tc->pool->mapping_pool);
}

static void process_prepared(struct pool *pool, struct list_head *head,
Mike Snitzer's avatar
Mike Snitzer committed
919
			     void (*fn)(struct dm_thin_new_mapping *))
920
921
922
{
	unsigned long flags;
	struct list_head maps;
Mike Snitzer's avatar
Mike Snitzer committed
923
	struct dm_thin_new_mapping *m, *tmp;
924
925
926

	INIT_LIST_HEAD(&maps);
	spin_lock_irqsave(&pool->lock, flags);
Joe Thornber's avatar
Joe Thornber committed
927
	list_splice_init(head, &maps);
928
929
930
	spin_unlock_irqrestore(&pool->lock, flags);

	list_for_each_entry_safe(m, tmp, &maps, list)
Joe Thornber's avatar
Joe Thornber committed
931
		fn(m);
932
933
934
935
936
}

/*
 * Deferred bio jobs.
 */
Joe Thornber's avatar
Joe Thornber committed
937
static int io_overlaps_block(struct pool *pool, struct bio *bio)
938
{
939
	sector_t bi_sector = bio->bi_sector;
Joe Thornber's avatar
Joe Thornber committed
940

941
942
	return !sector_div(bi_sector, pool->sectors_per_block) &&
		(bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT));
Joe Thornber's avatar
Joe Thornber committed
943
944
945
946
947
948
}

static int io_overwrites_block(struct pool *pool, struct bio *bio)
{
	return (bio_data_dir(bio) == WRITE) &&
		io_overlaps_block(pool, bio);
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
}

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
968
static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
969
{
Mike Snitzer's avatar
Mike Snitzer committed
970
	struct dm_thin_new_mapping *r = pool->next_mapping;
971
972
973
974
975
976
977
978
979

	BUG_ON(!pool->next_mapping);

	pool->next_mapping = NULL;

	return r;
}

static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
980
981
			  struct dm_dev *origin, dm_block_t data_origin,
			  dm_block_t data_dest,
Mike Snitzer's avatar
Mike Snitzer committed
982
			  struct dm_bio_prison_cell *cell, struct bio *bio)
983
984
985
{
	int r;
	struct pool *pool = tc->pool;
Mike Snitzer's avatar
Mike Snitzer committed
986
	struct dm_thin_new_mapping *m = get_next_mapping(pool);
987
988

	INIT_LIST_HEAD(&m->list);
989
	m->quiesced = 0;
990
991
992
993
994
995
996
997
	m->prepared = 0;
	m->tc = tc;
	m->virt_block = virt_block;
	m->data_block = data_dest;
	m->cell = cell;
	m->err = 0;
	m->bio = NULL;

998
999
	if (!ds_add_work(&pool->shared_read_ds, &m->list))
		m->quiesced = 1;
1000
1001
1002
1003
1004
1005
1006
1007

	/*
	 * 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.
	 */
	if (io_overwrites_block(pool, bio)) {
Mike Snitzer's avatar
Mike Snitzer committed
1008
1009
		struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;

1010
		h->overwrite_mapping = m;
1011
1012
1013
1014
1015
1016
		m->bio = bio;
		save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
		remap_and_issue(tc, bio, data_dest);
	} else {
		struct dm_io_region from, to;

1017
		from.bdev = origin->bdev;
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
		from.sector = data_origin * pool->sectors_per_block;
		from.count = pool->sectors_per_block;

		to.bdev = tc->pool_dev->bdev;
		to.sector = data_dest * pool->sectors_per_block;
		to.count = pool->sectors_per_block;

		r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
				   0, copy_complete, m);
		if (r < 0) {
			mempool_free(m, pool->mapping_pool);
			DMERR("dm_kcopyd_copy() failed");
			cell_error(cell);
		}
	}
}

1035
1036
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
1037
				   struct dm_bio_prison_cell *cell, struct bio *bio)
1038
1039
1040
1041
1042
1043
1044
{
	schedule_copy(tc, virt_block, tc->pool_dev,
		      data_origin, data_dest, cell, bio);
}

static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
				   dm_block_t data_dest,
Mike Snitzer's avatar
Mike Snitzer committed
1045
				   struct dm_bio_prison_cell *cell, struct bio *bio)
1046
1047
1048
1049
1050
{
	schedule_copy(tc, virt_block, tc->origin_dev,
		      virt_block, data_dest, cell, bio);
}

1051
static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
Mike Snitzer's avatar
Mike Snitzer committed
1052
			  dm_block_t data_block, struct dm_bio_prison_cell *cell,
1053
1054
1055
			  struct bio *bio)
{
	struct pool *pool = tc->pool;
Mike Snitzer's avatar
Mike Snitzer committed
1056
	struct dm_thin_new_mapping *m = get_next_mapping(pool);
1057
1058

	INIT_LIST_HEAD(&m->list);
1059
	m->quiesced = 1;
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
	m->prepared = 0;
	m->tc = tc;
	m->virt_block = virt_block;
	m->data_block = data_block;
	m->cell = cell;
	m->err = 0;
	m->bio = NULL;

	/*
	 * 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.
	 */
1073
	if (!pool->pf.zero_new_blocks)
1074
1075
1076
		process_prepared_mapping(m);

	else if (io_overwrites_block(pool, bio)) {
Mike Snitzer's avatar
Mike Snitzer committed
1077
1078
		struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;

1079
		h->overwrite_mapping = m;
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
		m->bio = bio;
		save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
		remap_and_issue(tc, bio, data_block);
	} else {
		int r;
		struct dm_io_region to;

		to.bdev = tc->pool_dev->bdev;
		to.sector = data_block * pool->sectors_per_block;
		to.count = pool->sectors_per_block;

		r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m);
		if (r < 0) {
			mempool_free(m, pool->mapping_pool);
			DMERR("dm_kcopyd_zero() failed");
			cell_error(cell);
		}
	}
}

static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
{
	int r;
	dm_block_t free_blocks;
	unsigned long flags;
	struct pool *pool = tc->pool;

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

	if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
		DMWARN("%s: reached low water mark, sending event.",
		       dm_device_name(pool->pool_md));
		spin_lock_irqsave(&pool->lock, flags);
		pool->low_water_triggered = 1;
		spin_unlock_irqrestore(&pool->lock, flags);
		dm_table_event(pool->ti->table);
	}

	if (!free_blocks) {
		if (pool->no_free_space)
			return -ENOSPC;
		else {
			/*
			 * Try to commit to see if that will free up some
			 * more space.
			 */
			r = dm_pool_commit_metadata(pool->pmd);
			if (r) {
				DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
				      __func__, r);
				return r;
			}

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

			/*
			 * If we still have no space we set a flag to avoid
			 * doing all this checking and return -ENOSPC.
			 */
			if (!free_blocks) {
				DMWARN("%s: no free space available.",
				       dm_device_name(pool->pool_md));
				spin_lock_irqsave(&pool->lock, flags);
				pool->no_free_space = 1;
				spin_unlock_irqrestore(&pool->lock, flags);
				return -ENOSPC;
			}
		}
	}

	r = dm_pool_alloc_data_block(pool->pmd, result);
	if (r)
		return r;

	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)
{
Mike Snitzer's avatar
Mike Snitzer committed
1167
	struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
1168
	struct thin_c *tc = h->tc;
1169
1170
1171
1172
1173
1174
1175
1176
	struct pool *pool = tc->pool;
	unsigned long flags;

	spin_lock_irqsave(&pool->lock, flags);
	bio_list_add(&pool->retry_on_resume_list, bio);
	spin_unlock_irqrestore(&pool->lock, flags);
}

Mike Snitzer's avatar
Mike Snitzer committed
1177
static void no_space(struct dm_bio_prison_cell *cell)
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
{
	struct bio *bio;
	struct bio_list bios;

	bio_list_init(&bios);
	cell_release(cell, &bios);

	while ((bio = bio_list_pop(&bios)))
		retry_on_resume(bio);
}

Joe Thornber's avatar
Joe Thornber committed
1189
1190
1191
static void process_discard(struct thin_c *tc, struct bio *bio)
{
	int r;
1192
	unsigned long flags;
Joe Thornber's avatar
Joe Thornber committed
1193
	struct pool *pool = tc->pool;
Mike Snitzer's avatar
Mike Snitzer committed
1194
	struct dm_bio_prison_cell *cell, *cell2;
Joe Thornber's avatar
Joe Thornber committed
1195
1196
1197
	struct cell_key key, key2;
	dm_block_t block = get_bio_block(tc, bio);
	struct dm_thin_lookup_result lookup_result;
Mike Snitzer's avatar
Mike Snitzer committed
1198
	struct dm_thin_new_mapping *m;
Joe Thornber's avatar
Joe Thornber committed
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224

	build_virtual_key(tc->td, block, &key);
	if (bio_detain(tc->pool->prison, &key, bio, &cell))
		return;

	r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
	switch (r) {
	case 0:
		/*
		 * Check nobody is fiddling with this pool block.  This can
		 * happen if someone's in the process of breaking sharing
		 * on this block.
		 */
		build_data_key(tc->td, lookup_result.block, &key2);
		if (bio_detain(tc->pool->prison, &key2, bio, &cell2)) {
			cell_release_singleton(cell, bio);
			break;
		}

		if (io_overlaps_block(pool, bio)) {
			/*
			 * IO may still be going to the destination block.  We must
			 * quiesce before we can do the removal.
			 */
			m = get_next_mapping(pool);
			m->tc = tc;
1225
			m->pass_discard = (!lookup_result.shared) && pool->pf.discard_passdown;
Joe Thornber's avatar
Joe Thornber committed
1226
1227
1228
1229
1230
1231
1232
1233
			m->virt_block = block;
			m->data_block = lookup_result.block;
			m->cell = cell;
			m->cell2 = cell2;
			m->err = 0;
			m->bio = bio;

			if (!ds_add_work(&pool->all_io_ds, &m->list)) {
1234
				spin_lock_irqsave(&pool->lock, flags);
Joe Thornber's avatar
Joe Thornber committed
1235
				list_add(&m->list, &pool->prepared_discards);
1236
				spin_unlock_irqrestore(&pool->lock, flags);
Joe Thornber's avatar
Joe Thornber committed
1237
1238
1239
1240
				wake_worker(pool);
			}
		} else {
			/*
1241
1242
1243
			 * The DM core makes sure that the discard doesn't span
			 * a block boundary.  So we submit the discard of a
			 * partial block appropriately.
Joe Thornber's avatar
Joe Thornber committed
1244
1245
1246
			 */
			cell_release_singleton(cell, bio);
			cell_release_singleton(cell2, bio);
1247
1248
1249
1250
			if ((!lookup_result.shared) && pool->pf.discard_passdown)
				remap_and_issue(tc, bio, lookup_result.block);
			else
				bio_endio(bio, 0);
Joe Thornber's avatar
Joe Thornber committed
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
		}
		break;

	case -ENODATA:
		/*
		 * It isn't provisioned, just forget it.
		 */
		cell_release_singleton(cell, bio);
		bio_endio(bio, 0);
		break;

	default:
		DMERR("discard: find block unexpectedly returned %d", r);
		cell_release_singleton(cell, bio);
		bio_io_error(bio);
		break;
	}
}

1270
1271
1272
static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
			  struct cell_key *key,
			  struct dm_thin_lookup_result *lookup_result,
Mike Snitzer's avatar
Mike Snitzer committed
1273
			  struct dm_bio_prison_cell *cell)
1274
1275
1276
1277
1278
1279
1280
{
	int r;
	dm_block_t data_block;

	r = alloc_data_block(tc, &data_block);
	switch (r) {
	case 0:
1281
1282
		schedule_internal_copy(tc, block, lookup_result->block,
				       data_block, cell, bio);
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
		break;

	case -ENOSPC:
		no_space(cell);
		break;

	default:
		DMERR("%s: alloc_data_block() failed, error = %d", __func__, r);
		cell_error(cell);
		break;
	}
}

static void process_shared_bio(struct thin_c *tc, struct bio *bio,
			       dm_block_t block,
			       struct dm_thin_lookup_result *lookup_result)
{
Mike Snitzer's avatar
Mike Snitzer committed
1300
	struct dm_bio_prison_cell *cell;
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
	struct pool *pool = tc->pool;
	struct cell_key key;

	/*
	 * If cell is already occupied, then sharing is already in the process
	 * of being broken so we have nothing further to do here.
	 */
	build_data_key(tc->td, lookup_result->block, &key);
	if (bio_detain(pool->prison, &key, bio, &cell))
		return;

	if (bio_data_dir(bio) == WRITE)
		break_sharing(tc, bio, block, &key, lookup_result, cell);
	else {
Mike Snitzer's avatar
Mike Snitzer committed
1315
		struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
1316

1317
		h->shared_read_entry = ds_inc(&pool->shared_read_ds);
1318
1319
1320
1321
1322
1323
1324

		cell_release_singleton(cell, bio);
		remap_and_issue(tc, bio, lookup_result->block);
	}
}

static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
Mike Snitzer's avatar
Mike Snitzer committed
1325
			    struct dm_bio_prison_cell *cell)
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
{
	int r;
	dm_block_t data_block;

	/*
	 * Remap empty bios (flushes) immediately, without provisioning.
	 */
	if (!bio->bi_size) {
		cell_release_singleton(cell, bio);
		remap_and_issue(tc, bio, 0);
		return;
	}

	/*
	 * Fill read bios with zeroes and complete them immediately.
	 */
	if (bio_data_dir(bio) == READ) {
		zero_fill_bio(bio);
		cell_release_singleton(cell, bio);
		bio_endio(bio, 0);
		return;
	}

	r = alloc_data_block(tc, &data_block);
	switch (r) {
	case 0:
1352
1353
1354
1355
		if (tc->origin_dev)
			schedule_external_copy(tc, block, data_block, cell, bio);
		else
			schedule_zero(tc, block, data_block, cell, bio);
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
		break;

	case -ENOSPC:
		no_space(cell);
		break;

	default:
		DMERR("%s: alloc_data_block() failed, error = %d", __func__, r);
		cell_error(cell);
		break;
	}
}

static void process_bio(struct thin_c *tc, struct bio *bio)
{
	int r;
	dm_block_t block = get_bio_block(tc, bio);
Mike Snitzer's avatar
Mike Snitzer committed
1373
	struct dm_bio_prison_cell *cell;
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
	struct cell_key key;
	struct dm_thin_lookup_result lookup_result;

	/*
	 * If cell is already occupied, then the block is already
	 * being provisioned so we have nothing further to do here.
	 */
	build_virtual_key(tc->td, block, &key);
	if (bio_detain(tc->pool->prison, &key, bio, &cell))
		return;

	r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
	switch (r) {
	case 0:
		/*
		 * We can release this cell now.  This thread is the only
		 * one that puts bios into a cell, and we know there were
		 * no preceding bios.
		 */
		/*
		 * TODO: this will probably have to change when discard goes
		 * back in.
		 */
		cell_release_singleton(cell, bio);

		if (lookup_result.shared)
			process_shared_bio(tc, bio, block, &lookup_result);
		else
			remap_and_issue(tc, bio, lookup_result.block);
		break;

	case -ENODATA:
1406
1407
1408
1409
1410
		if (bio_data_dir(bio) == READ && tc->origin_dev) {
			cell_release_singleton(cell, bio);
			remap_to_origin_and_issue(tc, bio);
		} else
			provision_block(tc, bio, block, cell);
1411
1412
1413
1414
		break;

	default:
		DMERR("dm_thin_find_block() failed, error = %d", r);
Joe Thornber's avatar
Joe Thornber committed
1415
		cell_release_singleton(cell, bio);
1416
1417
1418
1419
1420
		bio_io_error(bio);
		break;
	}
}

1421
1422
1423
1424
1425
1426
static int need_commit_due_to_time(struct pool *pool)
{
	return jiffies < pool->last_commit_jiffies ||
	       jiffies > pool->last_commit_jiffies + COMMIT_PERIOD;
}

1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
static void process_deferred_bios(struct pool *pool)
{
	unsigned long flags;
	struct bio *bio;
	struct bio_list bios;
	int r;

	bio_list_init(&bios);

	spin_lock_irqsave(&pool->lock, flags);
	bio_list_merge(&bios, &pool->deferred_bios);
	bio_list_init(&pool->deferred_bios);
	spin_unlock_irqrestore(&pool->lock, flags);

	while ((bio = bio_list_pop(&bios))) {
Mike Snitzer's avatar
Mike Snitzer committed
1442
		struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
1443
1444
		struct thin_c *tc = h->tc;

1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
		/*
		 * If we've got no free new_mapping structs, and processing
		 * this bio might require one, we pause until there are some
		 * prepared mappings to process.
		 */
		if (ensure_next_mapping(pool)) {
			spin_lock_irqsave(&pool->lock, flags);
			bio_list_merge(&pool->deferred_bios, &bios);
			spin_unlock_irqrestore(&pool->lock, flags);

			break;
		}
Joe Thornber's avatar
Joe Thornber committed
1457
1458
1459
1460
1461

		if (bio->bi_rw & REQ_DISCARD)
			process_discard(tc, bio);
		else
			process_bio(tc, bio);
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
	}

	/*
	 * If there are any deferred flush bios, we must commit
	 * the metadata before issuing them.
	 */
	bio_list_init(&bios);
	spin_lock_irqsave(&pool->lock, flags);
	bio_list_merge(&bios, &pool->deferred_flush_bios);
	bio_list_init(&pool->deferred_flush_bios);
	spin_unlock_irqrestore(&pool->lock, flags);

1474
	if (bio_list_empty(&bios) && !need_commit_due_to_time(pool))
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
		return;

	r = dm_pool_commit_metadata(pool->pmd);
	if (r) {
		DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
		      __func__, r);
		while ((bio = bio_list_pop(&bios)))
			bio_io_error(bio);
		return;
	}
1485
	pool->last_commit_jiffies = jiffies;
1486
1487
1488
1489
1490
1491
1492
1493
1494

	while ((bio = bio_list_pop(&bios)))
		generic_make_request(bio);
}

static void do_worker(struct work_struct *ws)
{
	struct pool *pool = container_of(ws, struct pool, worker);

Joe Thornber's avatar
Joe Thornber committed
1495
1496
	process_prepared(pool, &pool->prepared_mappings, process_prepared_mapping);
	process_prepared(pool, &pool->prepared_discards, process_prepared_discard);
1497
1498
1499
	process_deferred_bios(pool);
}

1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
/*
 * We want to commit periodically so that not too much
 * unwritten data builds up.
 */
static void do_waker(struct work_struct *ws)
{
	struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
	wake_worker(pool);
	queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
}

1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
/*----------------------------------------------------------------*/

/*
 * Mapping functions.
 */

/*
 * Called only while mapping a thin bio to hand it over to the workqueue.
 */
static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
{
	unsigned long flags;
	struct pool *pool = tc->pool;

	spin_lock_irqsave(&pool->lock, flags);
	bio_list_add(&pool->deferred_bios, bio);
	spin_unlock_irqrestore(&pool->lock, flags);

	wake_worker(pool);
}

Mike Snitzer's avatar
Mike Snitzer committed
1532
static struct dm_thin_endio_hook *thin_hook_bio(struct thin_c *tc, struct bio *bio)
1533
1534
{
	struct pool *pool = tc->pool;
Mike Snitzer's avatar
Mike Snitzer committed
1535
	struct dm_thin_endio_hook *h = mempool_alloc(pool->endio_hook_pool, GFP_NOIO);
1536
1537
1538

	h->tc = tc;
	h->shared_read_entry = NULL;
Joe Thornber's avatar
Joe Thornber committed
1539
	h->all_io_entry = bio->bi_rw & REQ_DISCARD ? NULL : ds_inc(&pool->all_io_ds);
1540
1541
1542
1543
1544
	h->overwrite_mapping = NULL;

	return h;
}

1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
/*
 * Non-blocking function called from the thin target's map function.
 */
static int thin_bio_map(struct dm_target *ti, struct bio *bio,
			union map_info *map_context)
{
	int r;
	struct thin_c *tc = ti->private;
	dm_block_t block = get_bio_block(tc, bio);
	struct dm_thin_device *td = tc->td;
	struct dm_thin_lookup_result result;

1557
	map_context->ptr = thin_hook_bio(tc, bio);
Joe Thornber's avatar
Joe Thornber committed
1558
	if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) {
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
		thin_defer_bio(tc, bio);
		return DM_MAPIO_SUBMITTED;
	}

	r = dm_thin_find_block(td, block, 0, &result);

	/*
	 * Note that we defer readahead too.
	 */
	switch (r) {
	case 0:
		if (unlikely(result.shared)) {
			/*
			 * We have a race condition here between the
			 * result.shared value returned by the lookup and
			 * snapshot creation, which may cause new
			 * sharing.
			 *
			 * To avoid this always quiesce the origin before
			 * taking the snap.  You want to do this anyway to
			 * ensure a consistent application view
			 * (i.e. lockfs).
			 *
			 * More distant ancestors are irrelevant. The
			 * shared flag will be set in their case.
			 */
			thin_defer_bio(tc, bio);
			r = DM_MAPIO_SUBMITTED;
		} else {
			remap(tc, bio, result.block);
			r = DM_MAPIO_REMAPPED;
		}
		break;

	case -ENODATA:
		/*
		 * In future, the failed dm_thin_find_block above could
		 * provide the hint to load the metadata into cache.
		 */
	case -EWOULDBLOCK:
		thin_defer_bio(tc, bio);
		r = DM_MAPIO_SUBMITTED;
		break;
	}

	return r;
}

static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
{
	int r;
	unsigned long flags;
	struct pool_c *pt = container_of(cb, struct pool_c, callbacks);

	spin_lock_irqsave(&pt->pool->lock, flags);
	r = !bio_list_empty(&pt->pool->retry_on_resume_list);
	spin_unlock_irqrestore(&pt->pool->lock, flags);

	if (!r) {
		struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
		r = bdi_congested(&q->backing_dev_info, bdi_bits);
	}

	return r;
}

static void __requeue_bios(struct pool *pool)
{
	bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list);
	bio_list_init(&pool->retry_on_resume_list);
}

/*----------------------------------------------------------------
 * Binding of control targets to a pool object
 *--------------------------------------------------------------*/
static int bind_control_target(struct pool *pool, struct dm_target *ti)
{
	struct pool_c *pt = ti->private;

	pool->ti = ti;
	pool->low_water_blocks = pt->low_water_blocks;
1640
	pool->pf = pt->pf;
1641

1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
	/*
	 * If discard_passdown was enabled verify that the data device
	 * supports discards.  Disable discard_passdown if not; otherwise
	 * -EOPNOTSUPP will be returned.
	 */
	if (pt->pf.discard_passdown) {
		struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
		if (!q || !blk_queue_discard(q)) {
			char buf[BDEVNAME_SIZE];
			DMWARN("Discard unsupported by data device (%s): Disabling discard passdown.",
			       bdevname(pt->data_dev->bdev, buf));
			pool->pf.discard_passdown = 0;
		}
	}

1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
	return 0;
}

static void unbind_control_target(struct pool *pool, struct dm_target *ti)
{
	if (pool->ti == ti)
		pool->ti = NULL;
}

/*----------------------------------------------------------------
 * Pool creation
 *--------------------------------------------------------------*/
1669
1670
1671
1672
1673
1674
1675
1676
/* Initialize pool features. */
static void pool_features_init(struct pool_features *pf)
{
	pf->zero_new_blocks = 1;
	pf->discard_enabled = 1;
	pf->discard_passdown = 1;
}

1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
static void __pool_destroy(struct pool *pool)
{
	__pool_table_remove(pool);

	if (dm_pool_metadata_close(pool->pmd) < 0)
		DMWARN("%s: dm_pool_metadata_close() failed.", __func__);

	prison_destroy(pool->prison);
	dm_kcopyd_client_destroy(pool->copier);

	if (pool->wq)
		destroy_workqueue(pool->wq);

	if (pool->next_mapping)
		mempool_free(pool->next_mapping, pool->mapping_pool);
	mempool_destroy(pool->mapping_pool);
	mempool_destroy(pool->endio_hook_pool);
	kfree(pool);
}

Mike Snitzer's avatar
Mike Snitzer committed
1697
1698
1699
static struct kmem_cache *_new_mapping_cache;
static struct kmem_cache *_endio_hook_cache;

1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
static struct pool *pool_create(struct mapped_device *pool_md,
				struct block_device *metadata_dev,
				unsigned long block_size, char **error)
{
	int r;
	void *err_p;
	struct pool *pool;
	struct dm_pool_metadata *pmd;

	pmd = dm_pool_metadata_open(metadata_dev, block_size);
	if (IS_ERR(pmd)) {
		*error = "Error creating metadata object";
		return (struct pool *)pmd;
	}

	pool = kmalloc(sizeof(*pool), GFP_KERNEL);
	if (!pool) {
		*error = "Error allocating memory for pool";
		err_p = ERR_PTR(-ENOMEM);
		goto bad_pool;
	}

	pool->pmd = pmd;
	pool->sectors_per_block = block_size;
	pool->low_water_blocks = 0;
1725
	pool_features_init(&pool->pf);
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
	pool->prison = prison_create(PRISON_CELLS);
	if (!pool->prison) {
		*error = "Error creating pool's bio prison";
		err_p = ERR_PTR(-ENOMEM);
		goto bad_prison;
	}

	pool->copier = dm_kcopyd_client_create();
	if (IS_ERR(pool->copier)) {
		r = PTR_ERR(pool->copier);
		*error = "Error creating pool's kcopyd client";
		err_p = ERR_PTR(r);
		goto bad_kcopyd_client;
	}

	/*
	 * Create singlethreaded workqueue that will service all devices
	 * that use this metadata.
	 */
	pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
	if (!pool->wq) {
		*error = "Error creating pool's workqueue";
		err_p = ERR_PTR(-ENOMEM);
		goto bad_wq;
	}

	INIT_WORK(&pool->worker, do_worker);
1753
	INIT_DELAYED_WORK(&pool->waker, do_waker);
1754
1755
1756
1757
	spin_lock_init(&pool->lock);
	bio_list_init(&pool->deferred_bios);
	bio_list_init(&pool->deferred_flush_bios);
	INIT_LIST_HEAD(&pool->prepared_mappings);
Joe Thornber's avatar
Joe Thornber committed
1758
	INIT_LIST_HEAD(&pool->prepared_discards);
1759
1760
1761
	pool->low_water_triggered = 0;
	pool->no_free_space = 0;
	bio_list_init(&pool->retry_on_resume_list);
1762
	ds_init(&pool->shared_read_ds);
Joe Thornber's avatar
Joe Thornber committed
1763
	ds_init(&pool->all_io_ds);
1764
1765

	pool->next_mapping = NULL;
Mike Snitzer's avatar
Mike Snitzer committed
1766
1767
	pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE,
						      _new_mapping_cache);
1768
1769
1770
1771
1772
1773
	if (!pool->mapping_pool) {
		*error = "Error creating pool's mapping mempool";
		err_p = ERR_PTR(-ENOMEM);
		goto bad_mapping_pool;
	}

Mike Snitzer's avatar
Mike Snitzer committed
1774
1775
	pool->endio_hook_pool = mempool_create_slab_pool(ENDIO_HOOK_POOL_SIZE,
							 _endio_hook_cache);
1776
1777
1778
1779
1780
1781
	if (!pool->endio_hook_pool) {
		*error = "Error creating pool's endio_hook mempool";
		err_p = ERR_PTR(-ENOMEM);
		goto bad_endio_hook_pool;
	}
	pool->ref_count = 1;
1782
	pool->last_commit_jiffies = jiffies;
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
	pool->pool_md = pool_md;
	pool->md_dev = metadata_dev;
	__pool_table_insert(pool);

	return pool;

bad_endio_hook_pool:
	mempool_destroy(pool->mapping_pool);
bad_mapping_pool:
	destroy_workqueue(pool->wq);
bad_wq:
	dm_kcopyd_client_destroy(pool->copier);
bad_kcopyd_client:
	prison_destroy(pool->prison);
bad_prison:
	kfree(pool);
bad_pool:
	if (dm_pool_metadata_close(pmd))
		DMWARN("%s: dm_pool_metadata_close() failed.", __func__);

	return err_p;
}

static void __pool_inc(struct pool *pool)
{
	BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
	pool->ref_count++;
}

static void __pool_dec(struct pool *pool)
{
	BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
	BUG_ON(!pool->ref_count);
	if (!--pool->ref_count)
		__pool_destroy(pool);
}

static struct pool *__pool_find(struct mapped_device *pool_md,
				struct block_device *metadata_dev,
1822
1823
				unsigned long block_size, char **error,
				int *created)
1824
1825
1826
1827
{
	struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);

	if (pool) {
1828
1829
		if (pool->pool_md != pool_md) {
			*error = "metadata device already in use by a pool";
1830
			return ERR_PTR(-EBUSY);