dm-thin.c 76.4 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
14
15
16
17
18
19
20
21
22
23

#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
 */
24
#define ENDIO_HOOK_POOL_SIZE 1024
25
26
#define MAPPING_POOL_SIZE 1024
#define PRISON_CELLS 1024
27
#define COMMIT_PERIOD HZ
28

29
30
31
DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
		"A percentage of time allocated for copy on write");

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
/*
 * 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
64
 * including all devices that share this block.  (see dm_deferred_set code)
65
66
67
68
69
 *
 * 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
70
 * (process_prepared_mapping).  This act of inserting breaks some
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
 * 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,
108
			   dm_block_t b, struct dm_cell_key *key)
109
110
111
112
113
114
115
{
	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,
116
			      struct dm_cell_key *key)
117
118
119
120
121
122
123
124
125
126
127
128
129
{
	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
130
struct dm_thin_new_mapping;
131

132
133
134
135
136
137
138
139
140
/*
 * The pool runs in 3 modes.  Ordered in degraded order for comparisons.
 */
enum pool_mode {
	PM_WRITE,		/* metadata may be changed */
	PM_READ_ONLY,		/* metadata may not be changed */
	PM_FAIL,		/* all I/O fails */
};

141
struct pool_features {
142
143
	enum pool_mode mode;

Mike Snitzer's avatar
Mike Snitzer committed
144
145
146
	bool zero_new_blocks:1;
	bool discard_enabled:1;
	bool discard_passdown:1;
147
148
};

149
150
151
152
struct thin_c;
typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);

153
154
155
156
157
158
159
160
161
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;
162
	uint32_t sectors_per_block;
163
	int sectors_per_block_shift;
164

165
	struct pool_features pf;
166
	bool low_water_triggered:1;	/* A dm event has been sent */
167
	bool no_free_space:1;		/* bios will be requeued if set */
168

169
	struct dm_bio_prison *prison;
170
171
172
173
	struct dm_kcopyd_client *copier;

	struct workqueue_struct *wq;
	struct work_struct worker;
174
	struct delayed_work waker;
175

176
	unsigned long last_commit_jiffies;
177
	unsigned ref_count;
178
179
180
181
182

	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
183
	struct list_head prepared_discards;
184
185
186

	struct bio_list retry_on_resume_list;

187
188
	struct dm_deferred_set *shared_read_ds;
	struct dm_deferred_set *all_io_ds;
189

Mike Snitzer's avatar
Mike Snitzer committed
190
	struct dm_thin_new_mapping *next_mapping;
191
	mempool_t *mapping_pool;
192
193
194
195
196
197

	process_bio_fn process_bio;
	process_bio_fn process_discard;

	process_mapping_fn process_prepared_mapping;
	process_mapping_fn process_prepared_discard;
198
199
};

200
static enum pool_mode get_pool_mode(struct pool *pool);
201
static void out_of_data_space(struct pool *pool);
202
static void metadata_operation_failed(struct pool *pool, const char *op, int r);
203

204
205
206
207
208
209
210
211
212
213
214
/*
 * 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;
215
216
	struct pool_features requested_pf; /* Features requested during table load */
	struct pool_features adjusted_pf;  /* Features used after adjusting for constituent devices */
217
218
219
220
221
222
223
};

/*
 * Target context for a thin.
 */
struct thin_c {
	struct dm_dev *pool_dev;
224
	struct dm_dev *origin_dev;
225
226
227
228
229
230
231
232
	dm_thin_id dev_id;

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

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

233
234
235
236
237
238
239
240
241
242
243
/*
 * 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);
}

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

244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
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);
}

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);
}

283
284
285
286
287
288
289
290
291
292
293
294
295
static void cell_defer_no_holder_no_free(struct thin_c *tc,
					 struct dm_bio_prison_cell *cell)
{
	struct pool *pool = tc->pool;
	unsigned long flags;

	spin_lock_irqsave(&pool->lock, flags);
	dm_cell_release_no_holder(pool->prison, cell, &pool->deferred_bios);
	spin_unlock_irqrestore(&pool->lock, flags);

	wake_worker(pool);
}

296
297
298
299
300
301
302
303
304
static void cell_error(struct pool *pool,
		       struct dm_bio_prison_cell *cell)
{
	dm_cell_error(pool->prison, cell);
	dm_bio_prison_free_cell(pool->prison, cell);
}

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

305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
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
353
354
355
356
357
358
359
360
361
362
363
364
/*
 * 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
365
struct dm_thin_endio_hook {
366
	struct thin_c *tc;
367
368
	struct dm_deferred_entry *shared_read_entry;
	struct dm_deferred_entry *all_io_entry;
Mike Snitzer's avatar
Mike Snitzer committed
369
	struct dm_thin_new_mapping *overwrite_mapping;
370
371
};

372
373
374
375
376
377
378
379
380
381
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))) {
382
		struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Mike Snitzer's avatar
Mike Snitzer committed
383

384
		if (h->tc == tc)
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
			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.
 */

409
410
411
412
413
static bool block_size_is_power_of_two(struct pool *pool)
{
	return pool->sectors_per_block_shift >= 0;
}

414
415
static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
{
416
	struct pool *pool = tc->pool;
417
418
	sector_t block_nr = bio->bi_sector;

419
420
	if (block_size_is_power_of_two(pool))
		block_nr >>= pool->sectors_per_block_shift;
421
	else
422
		(void) sector_div(block_nr, pool->sectors_per_block);
423
424

	return block_nr;
425
426
427
428
429
}

static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
{
	struct pool *pool = tc->pool;
430
	sector_t bi_sector = bio->bi_sector;
431
432

	bio->bi_bdev = tc->pool_dev->bdev;
433
	if (block_size_is_power_of_two(pool))
434
435
		bio->bi_sector = (block << pool->sectors_per_block_shift) |
				(bi_sector & (pool->sectors_per_block - 1));
436
437
438
	else
		bio->bi_sector = (block * pool->sectors_per_block) +
				 sector_div(bi_sector, pool->sectors_per_block);
439
440
}

441
442
443
444
445
static void remap_to_origin(struct thin_c *tc, struct bio *bio)
{
	bio->bi_bdev = tc->origin_dev->bdev;
}

446
447
448
449
450
451
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);
}

452
453
454
455
456
457
458
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;

459
	h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
460
461
462
	h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
}

463
static void issue(struct thin_c *tc, struct bio *bio)
464
465
466
467
{
	struct pool *pool = tc->pool;
	unsigned long flags;

468
469
470
471
472
	if (!bio_triggers_commit(tc, bio)) {
		generic_make_request(bio);
		return;
	}

473
	/*
474
475
476
	 * 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.
477
	 */
478
479
480
481
482
483
484
485
486
487
488
489
	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);
490
491
}

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

505
506
507
508
509
/*----------------------------------------------------------------*/

/*
 * Bio endio functions.
 */
Mike Snitzer's avatar
Mike Snitzer committed
510
struct dm_thin_new_mapping {
511
512
	struct list_head list;

513
514
515
516
	bool quiesced:1;
	bool prepared:1;
	bool pass_discard:1;
	bool definitely_not_shared:1;
517

518
	int err;
519
520
521
	struct thin_c *tc;
	dm_block_t virt_block;
	dm_block_t data_block;
Mike Snitzer's avatar
Mike Snitzer committed
522
	struct dm_bio_prison_cell *cell, *cell2;
523
524
525
526
527
528
529
530
531
532
533

	/*
	 * 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
534
static void __maybe_add_mapping(struct dm_thin_new_mapping *m)
535
536
537
{
	struct pool *pool = m->tc->pool;

538
	if (m->quiesced && m->prepared) {
539
		list_add_tail(&m->list, &pool->prepared_mappings);
540
541
542
543
544
545
546
		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
547
	struct dm_thin_new_mapping *m = context;
548
549
550
551
552
	struct pool *pool = m->tc->pool;

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

	spin_lock_irqsave(&pool->lock, flags);
553
	m->prepared = true;
554
555
556
557
558
559
560
	__maybe_add_mapping(m);
	spin_unlock_irqrestore(&pool->lock, flags);
}

static void overwrite_endio(struct bio *bio, int err)
{
	unsigned long flags;
561
	struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Mike Snitzer's avatar
Mike Snitzer committed
562
	struct dm_thin_new_mapping *m = h->overwrite_mapping;
563
564
565
566
567
	struct pool *pool = m->tc->pool;

	m->err = err;

	spin_lock_irqsave(&pool->lock, flags);
568
	m->prepared = true;
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
	__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.
 */
Joe Thornber's avatar
Joe Thornber committed
586
static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell)
587
588
589
590
591
{
	struct pool *pool = tc->pool;
	unsigned long flags;

	spin_lock_irqsave(&pool->lock, flags);
592
	cell_release(pool, cell, &pool->deferred_bios);
593
594
595
596
597
598
	spin_unlock_irqrestore(&tc->pool->lock, flags);

	wake_worker(pool);
}

/*
599
 * Same as cell_defer above, except it omits the original holder of the cell.
600
 */
601
static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
602
603
604
605
606
{
	struct pool *pool = tc->pool;
	unsigned long flags;

	spin_lock_irqsave(&pool->lock, flags);
607
	cell_release_no_holder(pool, cell, &pool->deferred_bios);
608
609
610
611
612
	spin_unlock_irqrestore(&pool->lock, flags);

	wake_worker(pool);
}

613
614
615
616
static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
{
	if (m->bio)
		m->bio->bi_end_io = m->saved_bi_end_io;
617
	cell_error(m->tc->pool, m->cell);
618
619
620
	list_del(&m->list);
	mempool_free(m, m->tc->pool->mapping_pool);
}
621

Mike Snitzer's avatar
Mike Snitzer committed
622
static void process_prepared_mapping(struct dm_thin_new_mapping *m)
623
624
{
	struct thin_c *tc = m->tc;
625
	struct pool *pool = tc->pool;
626
627
628
629
630
631
632
633
	struct bio *bio;
	int r;

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

	if (m->err) {
634
		cell_error(pool, m->cell);
635
		goto out;
636
637
638
639
640
641
642
643
644
	}

	/*
	 * 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) {
645
		metadata_operation_failed(pool, "dm_thin_insert_block", r);
646
		cell_error(pool, m->cell);
647
		goto out;
648
649
650
651
652
653
654
655
656
	}

	/*
	 * 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) {
657
		cell_defer_no_holder(tc, m->cell);
658
659
		bio_endio(bio, 0);
	} else
Joe Thornber's avatar
Joe Thornber committed
660
		cell_defer(tc, m->cell);
661

662
out:
663
	list_del(&m->list);
664
	mempool_free(m, pool->mapping_pool);
665
666
}

667
static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
Joe Thornber's avatar
Joe Thornber committed
668
669
670
{
	struct thin_c *tc = m->tc;

671
	bio_io_error(m->bio);
672
673
	cell_defer_no_holder(tc, m->cell);
	cell_defer_no_holder(tc, m->cell2);
674
675
676
677
678
679
	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
680

681
	inc_all_io_entry(tc->pool, m->bio);
682
683
	cell_defer_no_holder(tc, m->cell);
	cell_defer_no_holder(tc, m->cell2);
684

Joe Thornber's avatar
Joe Thornber committed
685
	if (m->pass_discard)
686
687
688
689
690
691
692
693
694
		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
695
696
697
698
699
700
	else
		bio_endio(m->bio, 0);

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

701
702
703
704
705
706
707
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)
708
		DMERR_LIMIT("dm_thin_remove_block() failed");
709
710
711
712

	process_prepared_discard_passdown(m);
}

Joe Thornber's avatar
Joe Thornber committed
713
static void process_prepared(struct pool *pool, struct list_head *head,
714
			     process_mapping_fn *fn)
715
716
717
{
	unsigned long flags;
	struct list_head maps;
Mike Snitzer's avatar
Mike Snitzer committed
718
	struct dm_thin_new_mapping *m, *tmp;
719
720
721

	INIT_LIST_HEAD(&maps);
	spin_lock_irqsave(&pool->lock, flags);
Joe Thornber's avatar
Joe Thornber committed
722
	list_splice_init(head, &maps);
723
724
725
	spin_unlock_irqrestore(&pool->lock, flags);

	list_for_each_entry_safe(m, tmp, &maps, list)
726
		(*fn)(m);
727
728
729
730
731
}

/*
 * Deferred bio jobs.
 */
Joe Thornber's avatar
Joe Thornber committed
732
static int io_overlaps_block(struct pool *pool, struct bio *bio)
733
{
734
	return bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT);
Joe Thornber's avatar
Joe Thornber committed
735
736
737
738
739
740
}

static int io_overwrites_block(struct pool *pool, struct bio *bio)
{
	return (bio_data_dir(bio) == WRITE) &&
		io_overlaps_block(pool, bio);
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
}

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
760
static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
761
{
762
	struct dm_thin_new_mapping *m = pool->next_mapping;
763
764
765

	BUG_ON(!pool->next_mapping);

766
767
768
769
	memset(m, 0, sizeof(struct dm_thin_new_mapping));
	INIT_LIST_HEAD(&m->list);
	m->bio = NULL;

770
771
	pool->next_mapping = NULL;

772
	return m;
773
774
775
}

static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
776
777
			  struct dm_dev *origin, dm_block_t data_origin,
			  dm_block_t data_dest,
Mike Snitzer's avatar
Mike Snitzer committed
778
			  struct dm_bio_prison_cell *cell, struct bio *bio)
779
780
781
{
	int r;
	struct pool *pool = tc->pool;
Mike Snitzer's avatar
Mike Snitzer committed
782
	struct dm_thin_new_mapping *m = get_next_mapping(pool);
783
784
785
786
787
788

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

789
	if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
790
		m->quiesced = true;
791
792
793
794
795
796
797
798

	/*
	 * 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)) {
799
		struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Mike Snitzer's avatar
Mike Snitzer committed
800

801
		h->overwrite_mapping = m;
802
803
		m->bio = bio;
		save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
804
		inc_all_io_entry(pool, bio);
805
806
807
808
		remap_and_issue(tc, bio, data_dest);
	} else {
		struct dm_io_region from, to;

809
		from.bdev = origin->bdev;
810
811
812
813
814
815
816
817
818
819
820
		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);
821
			DMERR_LIMIT("dm_kcopyd_copy() failed");
822
			cell_error(pool, cell);
823
824
825
826
		}
	}
}

827
828
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
829
				   struct dm_bio_prison_cell *cell, struct bio *bio)
830
831
832
833
834
835
836
{
	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
837
				   struct dm_bio_prison_cell *cell, struct bio *bio)
838
839
840
841
842
{
	schedule_copy(tc, virt_block, tc->origin_dev,
		      virt_block, data_dest, cell, bio);
}

843
static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
Mike Snitzer's avatar
Mike Snitzer committed
844
			  dm_block_t data_block, struct dm_bio_prison_cell *cell,
845
846
847
			  struct bio *bio)
{
	struct pool *pool = tc->pool;
Mike Snitzer's avatar
Mike Snitzer committed
848
	struct dm_thin_new_mapping *m = get_next_mapping(pool);
849

850
851
	m->quiesced = true;
	m->prepared = false;
852
853
854
855
856
857
858
859
860
861
	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.
	 */
862
	if (!pool->pf.zero_new_blocks)
863
864
865
		process_prepared_mapping(m);

	else if (io_overwrites_block(pool, bio)) {
866
		struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
Mike Snitzer's avatar
Mike Snitzer committed
867

868
		h->overwrite_mapping = m;
869
870
		m->bio = bio;
		save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
871
		inc_all_io_entry(pool, bio);
872
873
874
875
876
877
878
879
880
881
882
883
		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);
884
			DMERR_LIMIT("dm_kcopyd_zero() failed");
885
			cell_error(pool, cell);
886
887
888
889
		}
	}
}

890
891
892
893
/*
 * A non-zero return indicates read_only or fail_io mode.
 * Many callers don't care about the return value.
 */
894
static int commit(struct pool *pool)
895
896
897
898
899
900
{
	int r;

	if (get_pool_mode(pool) != PM_WRITE)
		return -EINVAL;

901
	r = dm_pool_commit_metadata(pool->pmd);
902
903
	if (r)
		metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
904
905
906
907

	return r;
}

908
909
910
911
912
913
914
915
916
917
918
919
920
921
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);
	}
}

922
923
924
925
926
927
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;

928
929
930
	if (get_pool_mode(pool) != PM_WRITE)
		return -EINVAL;

931
	r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
932
933
	if (r) {
		metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
934
		return r;
935
	}
936

937
	check_low_water_mark(pool, free_blocks);
938
939

	if (!free_blocks) {
940
941
942
943
		/*
		 * Try to commit to see if that will free up some
		 * more space.
		 */
944
945
946
		r = commit(pool);
		if (r)
			return r;
947

948
		r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
949
950
		if (r) {
			metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
951
			return r;
952
		}
953

954
		if (!free_blocks) {
955
			out_of_data_space(pool);
956
			return -ENOSPC;
957
958
959
960
		}
	}

	r = dm_pool_alloc_data_block(pool->pmd, result);
961
	if (r) {
962
		metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
963
		return r;
964
	}
965
966
967
968
969
970
971
972
973
974

	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)
{
975
	struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
976
	struct thin_c *tc = h->tc;
977
978
979
980
981
982
983
984
	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);
}

985
986
987
988
989
990
991
992
993
994
995
996
997
998
static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
{
	/*
	 * When pool is read-only, no cell locking is needed because
	 * nothing is changing.
	 */
	WARN_ON_ONCE(get_pool_mode(pool) != PM_READ_ONLY);

	if (pool->no_free_space)
		retry_on_resume(bio);
	else
		bio_io_error(bio);
}

999
static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell)
1000
1001
1002
1003
1004
{
	struct bio *bio;
	struct bio_list bios;

	bio_list_init(&bios);
1005
	cell_release(pool, cell, &bios);
1006
1007

	while ((bio = bio_list_pop(&bios)))
1008
		handle_unserviceable_bio(pool, bio);
1009
1010
}

Joe Thornber's avatar
Joe Thornber committed
1011
1012
1013
static void process_discard(struct thin_c *tc, struct bio *bio)
{
	int r;
1014
	unsigned long flags;
Joe Thornber's avatar
Joe Thornber committed
1015
	struct pool *pool = tc->pool;
Mike Snitzer's avatar
Mike Snitzer committed
1016
	struct dm_bio_prison_cell *cell, *cell2;
1017
	struct dm_cell_key key, key2;
Joe Thornber's avatar
Joe Thornber committed
1018
1019
	dm_block_t block = get_bio_block(tc, bio);
	struct dm_thin_lookup_result lookup_result;
Mike Snitzer's avatar
Mike Snitzer committed
1020
	struct dm_thin_new_mapping *m;
Joe Thornber's avatar
Joe Thornber committed
1021
1022

	build_virtual_key(tc->td, block, &key);
1023
	if (bio_detain(tc->pool, &key, bio, &cell))
Joe Thornber's avatar
Joe Thornber committed
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
		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);
1035
		if (bio_detain(tc->pool, &key2, bio, &cell2)) {
1036
			cell_defer_no_holder(tc, cell);
Joe Thornber's avatar
Joe Thornber committed
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
			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;
1047
1048
			m->pass_discard = pool->pf.discard_passdown;
			m->definitely_not_shared = !lookup_result.shared;
Joe Thornber's avatar
Joe Thornber committed
1049
1050
1051
1052
1053
1054
			m->virt_block = block;
			m->data_block = lookup_result.block;
			m->cell = cell;
			m->cell2 = cell2;
			m->bio = bio;

1055
			if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) {
1056
				spin_lock_irqsave(&pool->lock, flags);
1057
				list_add_tail(&m->list, &pool->prepared_discards);
1058
				spin_unlock_irqrestore(&pool->lock, flags);
Joe Thornber's avatar
Joe Thornber committed
1059
1060
1061
				wake_worker(pool);
			}
		} else {
1062
			inc_all_io_entry(pool, bio);
1063
1064
			cell_defer_no_holder(tc, cell);
			cell_defer_no_holder(tc, cell2);
1065

Joe Thornber's avatar
Joe Thornber committed
1066
			/*
1067
1068
1069
			 * 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
1070
			 */
1071
1072
1073
1074
			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
1075
1076
1077
1078
1079
1080
1081
		}
		break;

	case -ENODATA:
		/*
		 * It isn't provisioned, just forget it.
		 */
1082
		cell_defer_no_holder(tc, cell);
Joe Thornber's avatar
Joe Thornber committed
1083
1084
1085
1086
		bio_endio(bio, 0);
		break;

	default:
1087
1088
		DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
			    __func__, r);
1089
		cell_defer_no_holder(tc, cell);
Joe Thornber's avatar
Joe Thornber committed
1090
1091
1092
1093
1094
		bio_io_error(bio);
		break;
	}
}

1095
static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
1096
			  struct dm_cell_key *key,
1097
			  struct dm_thin_lookup_result *lookup_result,
Mike Snitzer's avatar
Mike Snitzer committed
1098
			  struct dm_bio_prison_cell *cell)
1099
1100
1101
{
	int r;
	dm_block_t data_block;
1102
	struct pool *pool = tc->pool;
1103
1104
1105
1106

	r = alloc_data_block(tc, &data_block);
	switch (r) {
	case 0:
1107
1108
		schedule_internal_copy(tc, block, lookup_result->block,
				       data_block, cell, bio);
1109
1110
1111
		break;

	case -ENOSPC:
1112
		retry_bios_on_resume(pool, cell);
1113
1114
1115
		break;

	default:
1116
1117
		DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
			    __func__, r);
1118
		cell_error(pool, cell);
1119
1120
1121
1122
1123
1124
1125
1126
		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
1127
	struct dm_bio_prison_cell *cell;
1128
	struct pool *pool = tc->pool;
1129
	struct dm_cell_key key;
1130
1131
1132
1133
1134
1135

	/*
	 * 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);
1136
	if (bio_detain(pool, &key, bio, &cell))
1137
1138
		return;

1139
	if (bio_data_dir(bio) == WRITE && bio->bi_size)
1140
1141
		break_sharing(tc, bio, block, &key, lookup_result, cell);
	else {
1142
		struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1143

1144
		h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
1145
		inc_all_io_entry(pool, bio);
1146
		cell_defer_no_holder(tc, cell);
1147

1148
1149
1150
1151
1152
		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
1153
			    struct dm_bio_prison_cell *cell)
1154
1155
1156
{
	int r;
	dm_block_t data_block;
1157
	struct pool *pool = tc->pool;
1158
1159
1160
1161
1162

	/*
	 * Remap empty bios (flushes) immediately, without provisioning.
	 */
	if (!bio->bi_size) {
1163
		inc_all_io_entry(pool, bio);
1164
		cell_defer_no_holder(tc, cell);
1165

1166
1167
1168
1169
1170
1171
1172
1173
1174
		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);
1175
		cell_defer_no_holder(tc, cell);
1176
1177
1178
1179
1180
1181
1182
		bio_endio(bio, 0);
		return;
	}

	r = alloc_data_block(tc, &data_block);
	switch (r) {
	case 0:
1183
1184
1185
1186
		if (tc->origin_dev)
			schedule_external_copy(tc, block, data_block, cell, bio);
		else
			schedule_zero(tc, block, data_block, cell, bio);
1187
1188
1189
		break;

	case -ENOSPC:
1190
		retry_bios_on_resume(pool, cell);
1191
1192
1193
		break;

	default:
1194
1195
		DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
			    __func__, r);
1196
		cell_error(pool, cell);
1197
1198
1199
1200
1201
1202
1203
		break;
	}
}

static void process_bio(struct thin_c *tc, struct bio *bio)
{
	int r;
1204
	struct pool *pool = tc->pool;
1205
	dm_block_t block = get_bio_block(tc, bio);
Mike Snitzer's avatar
Mike Snitzer committed
1206
	struct dm_bio_prison_cell *cell;
1207
	struct dm_cell_key key;
1208
1209
1210
1211
1212
1213
1214
	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);
1215
	if (bio_detain(pool, &key, bio, &cell))
1216
1217
1218
1219
1220
		return;

	r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
	switch (r) {
	case 0:
1221
		if (lookup_result.shared) {
1222
			process_shared_bio(tc, bio, block, &lookup_result);
1223
			cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */
1224
		} else {
1225
			inc_all_io_entry(pool, bio);
1226
			cell_defer_no_holder(tc, cell);
1227

1228
			remap_and_issue(tc, bio, lookup_result.block);
1229
		}
1230
1231
1232
		break;

	case -ENODATA:
1233
		if (bio_data_dir(bio) == READ && tc->origin_dev) {
1234
			inc_all_io_entry(pool, bio);
1235
			cell_defer_no_holder(tc, cell);
1236

1237
1238
1239
			remap_to_origin_and_issue(tc, bio);
		} else
			provision_block(tc, bio, block, cell);
1240
1241
1242
		break;

	default:
1243
1244
		DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
			    __func__, r);
1245
		cell_defer_no_holder(tc, cell);
1246
1247
1248
1249
1250
		bio_io_error(bio);
		break;
	}
}

1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
{
	int r;
	int rw = bio_data_dir(bio);
	dm_block_t block = get_bio_block(tc, bio);
	struct dm_thin_lookup_result lookup_result;

	r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
	switch (r) {
	case 0:
		if (lookup_result.shared && (rw == WRITE) && bio->bi_size)
1262
			handle_unserviceable_bio(tc->pool, bio);
1263
1264
		else {
			inc_all_io_entry(tc->pool, bio);
1265
			remap_and_issue(tc, bio, lookup_result.block);
1266
		}
1267
1268
1269
1270
		break;

	case -ENODATA:
		if (rw != READ) {
1271
			handle_unserviceable_bio(tc->pool, bio);
1272
1273
1274
1275
			break;
		}

		if (tc->origin_dev) {
1276
			inc_all_io_entry(tc->pool, bio);
1277
1278
1279
1280
1281
1282
1283
1284
1285
			remap_to_origin_and_issue(tc, bio);
			break;
		}

		zero_fill_bio(bio);
		bio_endio(bio, 0);
		break;

	default:
1286
1287
		DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
			    __func__, r);
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
		bio_io_error(bio);
		break;
	}
}

static void process_bio_fail(struct thin_c *tc, struct bio *bio)
{
	bio_io_error(bio);
}

1298
1299
1300
1301
/*
 * FIXME: should we also commit due to size of transaction, measured in
 * metadata blocks?
 */