dm-thin.c 99.9 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/log2.h>
15
#include <linux/list.h>
16
#include <linux/rculist.h>
17
18
19
#include <linux/init.h>
#include <linux/module.h>
#include <linux/slab.h>
20
#include <linux/sort.h>
21
#include <linux/rbtree.h>
22
23
24
25
26
27

#define	DM_MSG_PREFIX	"thin"

/*
 * Tunable constants
 */
28
#define ENDIO_HOOK_POOL_SIZE 1024
29
#define MAPPING_POOL_SIZE 1024
30
#define COMMIT_PERIOD HZ
31
32
33
#define NO_SPACE_TIMEOUT_SECS 60

static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS;
34

35
36
37
DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
		"A percentage of time allocated for copy on write");

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

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

/*
 * Key building.
 */
static void build_data_key(struct dm_thin_device *td,
114
			   dm_block_t b, struct dm_cell_key *key)
115
116
117
{
	key->virtual = 0;
	key->dev = dm_thin_dev_id(td);
118
119
	key->block_begin = b;
	key->block_end = b + 1ULL;
120
121
122
}

static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
123
			      struct dm_cell_key *key)
124
125
126
{
	key->virtual = 1;
	key->dev = dm_thin_dev_id(td);
127
128
	key->block_begin = b;
	key->block_end = b + 1ULL;
129
130
131
132
}

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

Joe Thornber's avatar
Joe Thornber committed
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
#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);
}

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

180
181
182
183
184
/*
 * 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
185
struct dm_thin_new_mapping;
186

187
/*
188
 * The pool runs in 4 modes.  Ordered in degraded order for comparisons.
189
190
191
 */
enum pool_mode {
	PM_WRITE,		/* metadata may be changed */
192
	PM_OUT_OF_DATA_SPACE,	/* metadata may be changed, though data may not be allocated */
193
194
195
196
	PM_READ_ONLY,		/* metadata may not be changed */
	PM_FAIL,		/* all I/O fails */
};

197
struct pool_features {
198
199
	enum pool_mode mode;

Mike Snitzer's avatar
Mike Snitzer committed
200
201
202
	bool zero_new_blocks:1;
	bool discard_enabled:1;
	bool discard_passdown:1;
203
	bool error_if_no_space:1;
204
205
};

206
207
struct thin_c;
typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
208
typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell);
209
210
typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);

211
212
#define CELL_SORT_ARRAY_SIZE 8192

213
214
215
216
217
218
219
220
221
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;
222
	uint32_t sectors_per_block;
223
	int sectors_per_block_shift;
224

225
	struct pool_features pf;
226
	bool low_water_triggered:1;	/* A dm event has been sent */
227

228
	struct dm_bio_prison *prison;
229
230
231
	struct dm_kcopyd_client *copier;

	struct workqueue_struct *wq;
Joe Thornber's avatar
Joe Thornber committed
232
	struct throttle throttle;
233
	struct work_struct worker;
234
	struct delayed_work waker;
235
	struct delayed_work no_space_timeout;
236

237
	unsigned long last_commit_jiffies;
238
	unsigned ref_count;
239
240
241
242

	spinlock_t lock;
	struct bio_list deferred_flush_bios;
	struct list_head prepared_mappings;
Joe Thornber's avatar
Joe Thornber committed
243
	struct list_head prepared_discards;
244
	struct list_head active_thins;
245

246
247
	struct dm_deferred_set *shared_read_ds;
	struct dm_deferred_set *all_io_ds;
248

Mike Snitzer's avatar
Mike Snitzer committed
249
	struct dm_thin_new_mapping *next_mapping;
250
	mempool_t *mapping_pool;
251
252
253
254

	process_bio_fn process_bio;
	process_bio_fn process_discard;

255
256
257
	process_cell_fn process_cell;
	process_cell_fn process_discard_cell;

258
259
	process_mapping_fn process_prepared_mapping;
	process_mapping_fn process_prepared_discard;
260
261

	struct dm_bio_prison_cell *cell_sort_array[CELL_SORT_ARRAY_SIZE];
262
263
};

264
static enum pool_mode get_pool_mode(struct pool *pool);
265
static void metadata_operation_failed(struct pool *pool, const char *op, int r);
266

267
268
269
270
271
272
273
274
275
276
277
/*
 * 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;
278
279
	struct pool_features requested_pf; /* Features requested during table load */
	struct pool_features adjusted_pf;  /* Features used after adjusting for constituent devices */
280
281
282
283
284
285
};

/*
 * Target context for a thin.
 */
struct thin_c {
286
	struct list_head list;
287
	struct dm_dev *pool_dev;
288
	struct dm_dev *origin_dev;
289
	sector_t origin_size;
290
291
292
293
	dm_thin_id dev_id;

	struct pool *pool;
	struct dm_thin_device *td;
294
	bool requeue_mode:1;
295
	spinlock_t lock;
296
	struct list_head deferred_cells;
297
298
	struct bio_list deferred_bio_list;
	struct bio_list retry_on_resume_list;
299
	struct rb_root sort_bio_list; /* sorted list of deferred bios */
300
301
302
303
304
305
306

	/*
	 * Ensures the thin is not destroyed until the worker has finished
	 * iterating the active_thins list.
	 */
	atomic_t refcount;
	struct completion can_destroy;
307
308
309
310
};

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

311
312
313
314
315
316
317
318
319
320
321
/*
 * 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);
}

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

322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
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);
}

353
354
355
356
357
358
359
360
361
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);
}

362
363
364
365
366
367
368
369
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);
}

370
371
static void cell_error_with_code(struct pool *pool,
				 struct dm_bio_prison_cell *cell, int error_code)
372
{
373
	dm_cell_error(pool->prison, cell, error_code);
374
375
376
	dm_bio_prison_free_cell(pool->prison, cell);
}

377
378
379
380
381
static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell)
{
	cell_error_with_code(pool, cell, -EIO);
}

382
383
384
385
386
387
388
389
390
391
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);
}

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
/*
 * 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
454
struct dm_thin_endio_hook {
455
	struct thin_c *tc;
456
457
	struct dm_deferred_entry *shared_read_entry;
	struct dm_deferred_entry *all_io_entry;
Mike Snitzer's avatar
Mike Snitzer committed
458
	struct dm_thin_new_mapping *overwrite_mapping;
459
	struct rb_node rb_node;
460
461
};

462
463
464
465
466
467
468
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)
469
470
{
	struct bio *bio;
471
472
473
474
475
476
477

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

static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master, int error)
{
478
	struct bio_list bios;
479
	unsigned long flags;
480
481

	bio_list_init(&bios);
482

483
	spin_lock_irqsave(&tc->lock, flags);
484
	__merge_bio_list(&bios, master);
485
	spin_unlock_irqrestore(&tc->lock, flags);
486

487
	error_bio_list(&bios, error);
488
489
}

490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
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);
}

507
508
static void requeue_io(struct thin_c *tc)
{
509
	struct bio_list bios;
510
	unsigned long flags;
511
512
513

	bio_list_init(&bios);

514
	spin_lock_irqsave(&tc->lock, flags);
515
516
	__merge_bio_list(&bios, &tc->deferred_bio_list);
	__merge_bio_list(&bios, &tc->retry_on_resume_list);
517
	spin_unlock_irqrestore(&tc->lock, flags);
518

519
520
	error_bio_list(&bios, DM_ENDIO_REQUEUE);
	requeue_deferred_cells(tc);
521
522
}

523
524
525
526
527
528
static void error_retry_list(struct pool *pool)
{
	struct thin_c *tc;

	rcu_read_lock();
	list_for_each_entry_rcu(tc, &pool->active_thins, list)
529
		error_thin_bio_list(tc, &tc->retry_on_resume_list, -EIO);
530
531
532
	rcu_read_unlock();
}

533
534
535
536
537
538
539
/*
 * 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.
 */

540
541
542
543
544
static bool block_size_is_power_of_two(struct pool *pool)
{
	return pool->sectors_per_block_shift >= 0;
}

545
546
static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
{
547
	struct pool *pool = tc->pool;
548
	sector_t block_nr = bio->bi_iter.bi_sector;
549

550
551
	if (block_size_is_power_of_two(pool))
		block_nr >>= pool->sectors_per_block_shift;
552
	else
553
		(void) sector_div(block_nr, pool->sectors_per_block);
554
555

	return block_nr;
556
557
558
559
560
}

static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
{
	struct pool *pool = tc->pool;
561
	sector_t bi_sector = bio->bi_iter.bi_sector;
562
563

	bio->bi_bdev = tc->pool_dev->bdev;
564
	if (block_size_is_power_of_two(pool))
565
566
567
		bio->bi_iter.bi_sector =
			(block << pool->sectors_per_block_shift) |
			(bi_sector & (pool->sectors_per_block - 1));
568
	else
569
		bio->bi_iter.bi_sector = (block * pool->sectors_per_block) +
570
				 sector_div(bi_sector, pool->sectors_per_block);
571
572
}

573
574
575
576
577
static void remap_to_origin(struct thin_c *tc, struct bio *bio)
{
	bio->bi_bdev = tc->origin_dev->bdev;
}

578
579
580
581
582
583
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);
}

584
585
586
587
588
589
590
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;

591
	h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
592
593
594
	h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
}

595
static void issue(struct thin_c *tc, struct bio *bio)
596
597
598
599
{
	struct pool *pool = tc->pool;
	unsigned long flags;

600
601
602
603
604
	if (!bio_triggers_commit(tc, bio)) {
		generic_make_request(bio);
		return;
	}

605
	/*
606
607
608
	 * 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.
609
	 */
610
611
612
613
614
615
616
617
618
619
620
621
	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);
622
623
}

624
625
626
627
628
629
630
631
632
633
634
635
636
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);
}

637
638
639
640
641
/*----------------------------------------------------------------*/

/*
 * Bio endio functions.
 */
Mike Snitzer's avatar
Mike Snitzer committed
642
struct dm_thin_new_mapping {
643
644
	struct list_head list;

645
646
	bool pass_discard:1;
	bool definitely_not_shared:1;
647

648
649
650
651
652
653
654
	/*
	 * 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;

655
	int err;
656
657
658
	struct thin_c *tc;
	dm_block_t virt_block;
	dm_block_t data_block;
Mike Snitzer's avatar
Mike Snitzer committed
659
	struct dm_bio_prison_cell *cell, *cell2;
660
661
662
663
664
665
666
667
668
669
670

	/*
	 * 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;
};

671
static void __complete_mapping_preparation(struct dm_thin_new_mapping *m)
672
673
674
{
	struct pool *pool = m->tc->pool;

675
	if (atomic_dec_and_test(&m->prepare_actions)) {
676
		list_add_tail(&m->list, &pool->prepared_mappings);
677
678
679
680
		wake_worker(pool);
	}
}

681
static void complete_mapping_preparation(struct dm_thin_new_mapping *m)
682
683
684
685
686
{
	unsigned long flags;
	struct pool *pool = m->tc->pool;

	spin_lock_irqsave(&pool->lock, flags);
687
	__complete_mapping_preparation(m);
688
689
690
	spin_unlock_irqrestore(&pool->lock, flags);
}

691
692
693
694
695
696
697
698
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);
}

699
700
static void overwrite_endio(struct bio *bio, int err)
{
701
	struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Mike Snitzer's avatar
Mike Snitzer committed
702
	struct dm_thin_new_mapping *m = h->overwrite_mapping;
703
704

	m->err = err;
705
	complete_mapping_preparation(m);
706
707
708
709
710
711
712
713
714
715
716
717
718
}

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

/*
 * Workqueue.
 */

/*
 * Prepared mapping jobs.
 */

/*
719
720
 * This sends the bios in the cell, except the original holder, back
 * to the deferred_bios list.
721
 */
722
static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
723
724
725
726
{
	struct pool *pool = tc->pool;
	unsigned long flags;

727
728
729
	spin_lock_irqsave(&tc->lock, flags);
	cell_release_no_holder(pool, cell, &tc->deferred_bio_list);
	spin_unlock_irqrestore(&tc->lock, flags);
730
731
732
733

	wake_worker(pool);
}

734
735
static void thin_defer_bio(struct thin_c *tc, struct bio *bio);

736
737
738
739
740
741
742
743
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)
744
{
745
	struct remap_info *info = context;
746
747
	struct bio *bio;

748
	while ((bio = bio_list_pop(&cell->bios))) {
749
		if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA))
750
			bio_list_add(&info->defer_bios, bio);
751
		else {
752
753
754
755
756
757
758
759
			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);
760
761
762
763
		}
	}
}

764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
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);
}

790
791
static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
{
Kent Overstreet's avatar
Kent Overstreet committed
792
	if (m->bio) {
793
		m->bio->bi_end_io = m->saved_bi_end_io;
Kent Overstreet's avatar
Kent Overstreet committed
794
795
		atomic_inc(&m->bio->bi_remaining);
	}
796
	cell_error(m->tc->pool, m->cell);
797
798
799
	list_del(&m->list);
	mempool_free(m, m->tc->pool->mapping_pool);
}
800

Mike Snitzer's avatar
Mike Snitzer committed
801
static void process_prepared_mapping(struct dm_thin_new_mapping *m)
802
803
{
	struct thin_c *tc = m->tc;
804
	struct pool *pool = tc->pool;
805
806
807
808
	struct bio *bio;
	int r;

	bio = m->bio;
Kent Overstreet's avatar
Kent Overstreet committed
809
	if (bio) {
810
		bio->bi_end_io = m->saved_bi_end_io;
Kent Overstreet's avatar
Kent Overstreet committed
811
812
		atomic_inc(&bio->bi_remaining);
	}
813
814

	if (m->err) {
815
		cell_error(pool, m->cell);
816
		goto out;
817
818
819
820
821
822
823
824
825
	}

	/*
	 * 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) {
826
		metadata_operation_failed(pool, "dm_thin_insert_block", r);
827
		cell_error(pool, m->cell);
828
		goto out;
829
830
831
832
833
834
835
836
837
	}

	/*
	 * 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) {
838
		inc_remap_and_issue_cell(tc, m->cell, m->data_block);
839
		bio_endio(bio, 0);
840
841
842
843
844
	} 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);
	}
845

846
out:
847
	list_del(&m->list);
848
	mempool_free(m, pool->mapping_pool);
849
850
}

851
static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
Joe Thornber's avatar
Joe Thornber committed
852
853
854
{
	struct thin_c *tc = m->tc;

855
	bio_io_error(m->bio);
856
857
	cell_defer_no_holder(tc, m->cell);
	cell_defer_no_holder(tc, m->cell2);
858
859
860
861
862
863
	mempool_free(m, tc->pool->mapping_pool);
}

static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m)
{
	struct thin_c *tc = m->tc;
Joe Thornber's avatar
Joe Thornber committed
864

865
	inc_all_io_entry(tc->pool, m->bio);
866
867
	cell_defer_no_holder(tc, m->cell);
	cell_defer_no_holder(tc, m->cell2);
868

Joe Thornber's avatar
Joe Thornber committed
869
	if (m->pass_discard)
870
871
872
873
874
875
876
877
878
		if (m->definitely_not_shared)
			remap_and_issue(tc, m->bio, m->data_block);
		else {
			bool used = false;
			if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used)
				bio_endio(m->bio, 0);
			else
				remap_and_issue(tc, m->bio, m->data_block);
		}
Joe Thornber's avatar
Joe Thornber committed
879
880
881
882
883
884
	else
		bio_endio(m->bio, 0);

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

885
886
887
888
889
890
891
static void process_prepared_discard(struct dm_thin_new_mapping *m)
{
	int r;
	struct thin_c *tc = m->tc;

	r = dm_thin_remove_block(tc->td, m->virt_block);
	if (r)
892
		DMERR_LIMIT("dm_thin_remove_block() failed");
893
894
895
896

	process_prepared_discard_passdown(m);
}

Joe Thornber's avatar
Joe Thornber committed
897
static void process_prepared(struct pool *pool, struct list_head *head,
898
			     process_mapping_fn *fn)
899
900
901
{
	unsigned long flags;
	struct list_head maps;
Mike Snitzer's avatar
Mike Snitzer committed
902
	struct dm_thin_new_mapping *m, *tmp;
903
904
905

	INIT_LIST_HEAD(&maps);
	spin_lock_irqsave(&pool->lock, flags);
Joe Thornber's avatar
Joe Thornber committed
906
	list_splice_init(head, &maps);
907
908
909
	spin_unlock_irqrestore(&pool->lock, flags);

	list_for_each_entry_safe(m, tmp, &maps, list)
910
		(*fn)(m);
911
912
913
914
915
}

/*
 * Deferred bio jobs.
 */
Joe Thornber's avatar
Joe Thornber committed
916
static int io_overlaps_block(struct pool *pool, struct bio *bio)
917
{
918
919
	return bio->bi_iter.bi_size ==
		(pool->sectors_per_block << SECTOR_SHIFT);
Joe Thornber's avatar
Joe Thornber committed
920
921
922
923
924
925
}

static int io_overwrites_block(struct pool *pool, struct bio *bio)
{
	return (bio_data_dir(bio) == WRITE) &&
		io_overlaps_block(pool, bio);
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
}

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
945
static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
946
{
947
	struct dm_thin_new_mapping *m = pool->next_mapping;
948
949
950

	BUG_ON(!pool->next_mapping);

951
952
953
954
	memset(m, 0, sizeof(struct dm_thin_new_mapping));
	INIT_LIST_HEAD(&m->list);
	m->bio = NULL;

955
956
	pool->next_mapping = NULL;

957
	return m;
958
959
}

960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
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);
	}
}

977
978
979
980
981
982
983
984
985
986
987
988
989
990
static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio,
				      dm_block_t data_block,
				      struct dm_thin_new_mapping *m)
{
	struct pool *pool = tc->pool;
	struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));

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

991
992
993
/*
 * A partial copy also needs to zero the uncopied region.
 */
994
static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
995
996
			  struct dm_dev *origin, dm_block_t data_origin,
			  dm_block_t data_dest,
997
998
			  struct dm_bio_prison_cell *cell, struct bio *bio,
			  sector_t len)
999
1000
1001
{
	int r;
	struct pool *pool = tc->pool;
Mike Snitzer's avatar
Mike Snitzer committed
1002
	struct dm_thin_new_mapping *m = get_next_mapping(pool);
1003
1004
1005
1006
1007
1008

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

1009
1010
1011
1012
1013
1014
1015
	/*
	 * 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);

1016
	if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
1017
		complete_mapping_preparation(m); /* already quiesced */
1018
1019
1020
1021
1022
1023
1024

	/*
	 * 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.
	 */
1025
1026
1027
	if (io_overwrites_block(pool, bio))
		remap_and_issue_overwrite(tc, bio, data_dest, m);
	else {
1028
1029
		struct dm_io_region from, to;

1030
		from.bdev = origin->bdev;
1031
		from.sector = data_origin * pool->sectors_per_block;
1032
		from.count = len;
1033
1034
1035

		to.bdev = tc->pool_dev->bdev;
		to.sector = data_dest * pool->sectors_per_block;
1036
		to.count = len;
1037
1038
1039
1040

		r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
				   0, copy_complete, m);
		if (r < 0) {
1041
			DMERR_LIMIT("dm_kcopyd_copy() failed");
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
			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);
1060
1061
		}
	}
1062
1063

	complete_mapping_preparation(m); /* drop our ref */
1064
1065
}

1066
1067
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
1068
				   struct dm_bio_prison_cell *cell, struct bio *bio)
1069
1070
{
	schedule_copy(tc, virt_block, tc->pool_dev,
1071
1072
		      data_origin, data_dest, cell, bio,
		      tc->pool->sectors_per_block);
1073
1074
}

1075
static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
Mike Snitzer's avatar
Mike Snitzer committed
1076
			  dm_block_t data_block, struct dm_bio_prison_cell *cell,
1077
1078
1079
			  struct bio *bio)
{
	struct pool *pool = tc->pool;
Mike Snitzer's avatar
Mike Snitzer committed
1080
	struct dm_thin_new_mapping *m = get_next_mapping(pool);
1081

1082
	atomic_set(&m->prepare_actions, 1); /* no need to quiesce */
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
	m->tc = tc;
	m->virt_block = virt_block;
	m->data_block = data_block;
	m->cell = cell;

	/*
	 * If the whole block of data is being overwritten or we are not
	 * zeroing pre-existing data, we can issue the bio immediately.
	 * Otherwise we use kcopyd to zero the data first.
	 */
1093
	if (!pool->pf.zero_new_blocks)
1094
1095
		process_prepared_mapping(m);

1096
1097
	else if (io_overwrites_block(pool, bio))
		remap_and_issue_overwrite(tc, bio, data_block, m);
1098

1099
	else
1100
1101
1102
1103
		ll_zero(tc, m,
			data_block * pool->sectors_per_block,
			(data_block + 1) * pool->sectors_per_block);
}
1104

1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
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);
1125
1126
}

1127
1128
1129
1130
/*
 * A non-zero return indicates read_only or fail_io mode.
 * Many callers don't care about the return value.
 */
1131
static int commit(struct pool *pool)
1132
1133
1134
{
	int r;

1135
	if (get_pool_mode(pool) >= PM_READ_ONLY)
1136
1137
		return -EINVAL;

1138
	r = dm_pool_commit_metadata(pool->pmd);
1139
1140
	if (r)
		metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
1141
1142
1143
1144

	return r;
}

1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks)
{
	unsigned long flags;

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

1159
1160
static void set_pool_mode(struct pool *pool, enum pool_mode new_mode);

1161
1162
1163
1164
1165
1166
static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
{
	int r;
	dm_block_t free_blocks;
	struct pool *pool = tc->pool;

1167
	if (WARN_ON(get_pool_mode(pool) != PM_WRITE))
1168
1169
		return -EINVAL;

1170
	r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1171
1172
	if (r) {
		metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1173
		return r;
1174
	}
1175

1176
	check_low_water_mark(pool, free_blocks);
1177
1178

	if (!free_blocks) {
1179
1180
1181
1182
		/*
		 * Try to commit to see if that will free up some
		 * more space.
		 */
1183
1184
1185
		r = commit(pool);
		if (r)
			return r;
1186

1187
		r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1188
1189
		if (r) {
			metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1190
			return r;
1191
		}
1192

1193
		if (!free_blocks) {
1194
			set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1195
			return -ENOSPC;
1196
1197
1198
1199
		}
	}

	r = dm_pool_alloc_data_block(pool->pmd, result);
1200
	if (r) {
1201
		metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
1202
		return r;
1203
	}
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213

	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)
{
1214
	struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1215
	struct thin_c *tc = h->tc;
1216
1217
	unsigned long flags;

1218
1219
1220
	spin_lock_irqsave(&tc->lock, flags);
	bio_list_add(&tc->retry_on_resume_list, bio);
	spin_unlock_irqrestore(&tc->lock, flags);
1221
1222
}

1223
static int should_error_unserviceable_bio(struct pool *pool)
1224
{
1225
1226
1227
1228
1229
1230
	enum pool_mode m = get_pool_mode(pool);

	switch (m) {
	case PM_WRITE:
		/* Shouldn't get here */
		DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
1231
		return -EIO;
1232
1233

	case PM_OUT_OF_DATA_SPACE:
1234
		return pool->pf.error_if_no_space ? -ENOSPC : 0;
1235
1236
1237

	case PM_READ_ONLY:
	case PM_FAIL:
1238
		return -EIO;
1239
1240
1241
	default:
		/* Shouldn't get here */
		DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1242
		return -EIO;
1243
1244
	}
}
1245

1246
1247
static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
{
1248
1249
1250
1251
	int error = should_error_unserviceable_bio(pool);

	if (error)
		bio_endio(bio, error);
1252
1253
	else
		retry_on_resume(bio);
1254
1255
}

1256
static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell)
1257
1258
1259
{
	struct bio *bio;
	struct bio_list bios;
1260
	int error;
1261

1262
1263
1264
	error = should_error_unserviceable_bio(pool);
	if (error) {
		cell_error_with_code(pool, cell, error);
1265
1266
1267
		return;
	}

1268
	bio_list_init(&bios);
1269
	cell_release(pool, cell, &bios);
1270

1271
1272
	while ((bio = bio_list_pop(&bios)))
		retry_on_resume(bio);
1273
1274
}

1275
static void process_discard_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
Joe Thornber's avatar
Joe Thornber committed
1276
1277
{
	int r;
1278
	struct bio *bio = cell->holder;
Joe Thornber's avatar
Joe Thornber committed
1279
	struct pool *pool = tc->pool;