blob: e89f8e7d8a33ad2b9ff29b0d1d2de17d5181d5aa [file] [log] [blame]
Joe Thornber991d9fa2011-10-31 20:21:18 +00001/*
2 * Copyright (C) 2011 Red Hat UK.
3 *
4 * This file is released under the GPL.
5 */
6
7#include "dm-thin-metadata.h"
8
9#include <linux/device-mapper.h>
10#include <linux/dm-io.h>
11#include <linux/dm-kcopyd.h>
12#include <linux/list.h>
13#include <linux/init.h>
14#include <linux/module.h>
15#include <linux/slab.h>
16
17#define DM_MSG_PREFIX "thin"
18
19/*
20 * Tunable constants
21 */
Alasdair G Kergon7768ed32012-07-27 15:07:57 +010022#define ENDIO_HOOK_POOL_SIZE 1024
Joe Thornber991d9fa2011-10-31 20:21:18 +000023#define DEFERRED_SET_SIZE 64
24#define MAPPING_POOL_SIZE 1024
25#define PRISON_CELLS 1024
Joe Thornber905e51b2012-03-28 18:41:27 +010026#define COMMIT_PERIOD HZ
Joe Thornber991d9fa2011-10-31 20:21:18 +000027
28/*
29 * The block size of the device holding pool data must be
30 * between 64KB and 1GB.
31 */
32#define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
33#define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
34
35/*
Joe Thornber991d9fa2011-10-31 20:21:18 +000036 * Device id is restricted to 24 bits.
37 */
38#define MAX_DEV_ID ((1 << 24) - 1)
39
40/*
41 * How do we handle breaking sharing of data blocks?
42 * =================================================
43 *
44 * We use a standard copy-on-write btree to store the mappings for the
45 * devices (note I'm talking about copy-on-write of the metadata here, not
46 * the data). When you take an internal snapshot you clone the root node
47 * of the origin btree. After this there is no concept of an origin or a
48 * snapshot. They are just two device trees that happen to point to the
49 * same data blocks.
50 *
51 * When we get a write in we decide if it's to a shared data block using
52 * some timestamp magic. If it is, we have to break sharing.
53 *
54 * Let's say we write to a shared block in what was the origin. The
55 * steps are:
56 *
57 * i) plug io further to this physical block. (see bio_prison code).
58 *
59 * ii) quiesce any read io to that shared data block. Obviously
60 * including all devices that share this block. (see deferred_set code)
61 *
62 * iii) copy the data block to a newly allocate block. This step can be
63 * missed out if the io covers the block. (schedule_copy).
64 *
65 * iv) insert the new mapping into the origin's btree
Joe Thornberfe878f32012-03-28 18:41:24 +010066 * (process_prepared_mapping). This act of inserting breaks some
Joe Thornber991d9fa2011-10-31 20:21:18 +000067 * sharing of btree nodes between the two devices. Breaking sharing only
68 * effects the btree of that specific device. Btrees for the other
69 * devices that share the block never change. The btree for the origin
70 * device as it was after the last commit is untouched, ie. we're using
71 * persistent data structures in the functional programming sense.
72 *
73 * v) unplug io to this physical block, including the io that triggered
74 * the breaking of sharing.
75 *
76 * Steps (ii) and (iii) occur in parallel.
77 *
78 * The metadata _doesn't_ need to be committed before the io continues. We
79 * get away with this because the io is always written to a _new_ block.
80 * If there's a crash, then:
81 *
82 * - The origin mapping will point to the old origin block (the shared
83 * one). This will contain the data as it was before the io that triggered
84 * the breaking of sharing came in.
85 *
86 * - The snap mapping still points to the old block. As it would after
87 * the commit.
88 *
89 * The downside of this scheme is the timestamp magic isn't perfect, and
90 * will continue to think that data block in the snapshot device is shared
91 * even after the write to the origin has broken sharing. I suspect data
92 * blocks will typically be shared by many different devices, so we're
93 * breaking sharing n + 1 times, rather than n, where n is the number of
94 * devices that reference this data block. At the moment I think the
95 * benefits far, far outweigh the disadvantages.
96 */
97
98/*----------------------------------------------------------------*/
99
100/*
101 * Sometimes we can't deal with a bio straight away. We put them in prison
102 * where they can't cause any mischief. Bios are put in a cell identified
103 * by a key, multiple bios can be in the same cell. When the cell is
104 * subsequently unlocked the bios become available.
105 */
106struct bio_prison;
107
108struct cell_key {
109 int virtual;
110 dm_thin_id dev;
111 dm_block_t block;
112};
113
Mike Snitzera24c2562012-06-03 00:30:00 +0100114struct dm_bio_prison_cell {
Joe Thornber991d9fa2011-10-31 20:21:18 +0000115 struct hlist_node list;
116 struct bio_prison *prison;
117 struct cell_key key;
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100118 struct bio *holder;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000119 struct bio_list bios;
120};
121
122struct bio_prison {
123 spinlock_t lock;
124 mempool_t *cell_pool;
125
126 unsigned nr_buckets;
127 unsigned hash_mask;
128 struct hlist_head *cells;
129};
130
131static uint32_t calc_nr_buckets(unsigned nr_cells)
132{
133 uint32_t n = 128;
134
135 nr_cells /= 4;
136 nr_cells = min(nr_cells, 8192u);
137
138 while (n < nr_cells)
139 n <<= 1;
140
141 return n;
142}
143
Mike Snitzera24c2562012-06-03 00:30:00 +0100144static struct kmem_cache *_cell_cache;
145
Joe Thornber991d9fa2011-10-31 20:21:18 +0000146/*
147 * @nr_cells should be the number of cells you want in use _concurrently_.
148 * Don't confuse it with the number of distinct keys.
149 */
150static struct bio_prison *prison_create(unsigned nr_cells)
151{
152 unsigned i;
153 uint32_t nr_buckets = calc_nr_buckets(nr_cells);
154 size_t len = sizeof(struct bio_prison) +
155 (sizeof(struct hlist_head) * nr_buckets);
156 struct bio_prison *prison = kmalloc(len, GFP_KERNEL);
157
158 if (!prison)
159 return NULL;
160
161 spin_lock_init(&prison->lock);
Mike Snitzera24c2562012-06-03 00:30:00 +0100162 prison->cell_pool = mempool_create_slab_pool(nr_cells, _cell_cache);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000163 if (!prison->cell_pool) {
164 kfree(prison);
165 return NULL;
166 }
167
168 prison->nr_buckets = nr_buckets;
169 prison->hash_mask = nr_buckets - 1;
170 prison->cells = (struct hlist_head *) (prison + 1);
171 for (i = 0; i < nr_buckets; i++)
172 INIT_HLIST_HEAD(prison->cells + i);
173
174 return prison;
175}
176
177static void prison_destroy(struct bio_prison *prison)
178{
179 mempool_destroy(prison->cell_pool);
180 kfree(prison);
181}
182
183static uint32_t hash_key(struct bio_prison *prison, struct cell_key *key)
184{
185 const unsigned long BIG_PRIME = 4294967291UL;
186 uint64_t hash = key->block * BIG_PRIME;
187
188 return (uint32_t) (hash & prison->hash_mask);
189}
190
191static int keys_equal(struct cell_key *lhs, struct cell_key *rhs)
192{
193 return (lhs->virtual == rhs->virtual) &&
194 (lhs->dev == rhs->dev) &&
195 (lhs->block == rhs->block);
196}
197
Mike Snitzera24c2562012-06-03 00:30:00 +0100198static struct dm_bio_prison_cell *__search_bucket(struct hlist_head *bucket,
199 struct cell_key *key)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000200{
Mike Snitzera24c2562012-06-03 00:30:00 +0100201 struct dm_bio_prison_cell *cell;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000202 struct hlist_node *tmp;
203
204 hlist_for_each_entry(cell, tmp, bucket, list)
205 if (keys_equal(&cell->key, key))
206 return cell;
207
208 return NULL;
209}
210
211/*
212 * This may block if a new cell needs allocating. You must ensure that
213 * cells will be unlocked even if the calling thread is blocked.
214 *
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100215 * Returns 1 if the cell was already held, 0 if @inmate is the new holder.
Joe Thornber991d9fa2011-10-31 20:21:18 +0000216 */
217static int bio_detain(struct bio_prison *prison, struct cell_key *key,
Mike Snitzera24c2562012-06-03 00:30:00 +0100218 struct bio *inmate, struct dm_bio_prison_cell **ref)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000219{
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100220 int r = 1;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000221 unsigned long flags;
222 uint32_t hash = hash_key(prison, key);
Mike Snitzera24c2562012-06-03 00:30:00 +0100223 struct dm_bio_prison_cell *cell, *cell2;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000224
225 BUG_ON(hash > prison->nr_buckets);
226
227 spin_lock_irqsave(&prison->lock, flags);
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100228
Joe Thornber991d9fa2011-10-31 20:21:18 +0000229 cell = __search_bucket(prison->cells + hash, key);
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100230 if (cell) {
231 bio_list_add(&cell->bios, inmate);
232 goto out;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000233 }
234
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100235 /*
236 * Allocate a new cell
237 */
Joe Thornber991d9fa2011-10-31 20:21:18 +0000238 spin_unlock_irqrestore(&prison->lock, flags);
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100239 cell2 = mempool_alloc(prison->cell_pool, GFP_NOIO);
240 spin_lock_irqsave(&prison->lock, flags);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000241
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100242 /*
243 * We've been unlocked, so we have to double check that
244 * nobody else has inserted this cell in the meantime.
245 */
246 cell = __search_bucket(prison->cells + hash, key);
247 if (cell) {
Joe Thornber991d9fa2011-10-31 20:21:18 +0000248 mempool_free(cell2, prison->cell_pool);
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100249 bio_list_add(&cell->bios, inmate);
250 goto out;
251 }
252
253 /*
254 * Use new cell.
255 */
256 cell = cell2;
257
258 cell->prison = prison;
259 memcpy(&cell->key, key, sizeof(cell->key));
260 cell->holder = inmate;
261 bio_list_init(&cell->bios);
262 hlist_add_head(&cell->list, prison->cells + hash);
263
264 r = 0;
265
266out:
267 spin_unlock_irqrestore(&prison->lock, flags);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000268
269 *ref = cell;
270
271 return r;
272}
273
274/*
275 * @inmates must have been initialised prior to this call
276 */
Mike Snitzera24c2562012-06-03 00:30:00 +0100277static void __cell_release(struct dm_bio_prison_cell *cell, struct bio_list *inmates)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000278{
279 struct bio_prison *prison = cell->prison;
280
281 hlist_del(&cell->list);
282
Mike Snitzer03aaae72012-05-12 01:43:12 +0100283 if (inmates) {
284 bio_list_add(inmates, cell->holder);
285 bio_list_merge(inmates, &cell->bios);
286 }
Joe Thornber991d9fa2011-10-31 20:21:18 +0000287
288 mempool_free(cell, prison->cell_pool);
289}
290
Mike Snitzera24c2562012-06-03 00:30:00 +0100291static void cell_release(struct dm_bio_prison_cell *cell, struct bio_list *bios)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000292{
293 unsigned long flags;
294 struct bio_prison *prison = cell->prison;
295
296 spin_lock_irqsave(&prison->lock, flags);
297 __cell_release(cell, bios);
298 spin_unlock_irqrestore(&prison->lock, flags);
299}
300
301/*
302 * There are a couple of places where we put a bio into a cell briefly
303 * before taking it out again. In these situations we know that no other
304 * bio may be in the cell. This function releases the cell, and also does
305 * a sanity check.
306 */
Mike Snitzera24c2562012-06-03 00:30:00 +0100307static void __cell_release_singleton(struct dm_bio_prison_cell *cell, struct bio *bio)
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100308{
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100309 BUG_ON(cell->holder != bio);
310 BUG_ON(!bio_list_empty(&cell->bios));
Mike Snitzer03aaae72012-05-12 01:43:12 +0100311
312 __cell_release(cell, NULL);
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100313}
314
Mike Snitzera24c2562012-06-03 00:30:00 +0100315static void cell_release_singleton(struct dm_bio_prison_cell *cell, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000316{
Joe Thornber991d9fa2011-10-31 20:21:18 +0000317 unsigned long flags;
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100318 struct bio_prison *prison = cell->prison;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000319
320 spin_lock_irqsave(&prison->lock, flags);
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100321 __cell_release_singleton(cell, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000322 spin_unlock_irqrestore(&prison->lock, flags);
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100323}
Joe Thornber991d9fa2011-10-31 20:21:18 +0000324
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100325/*
326 * Sometimes we don't want the holder, just the additional bios.
327 */
Mike Snitzera24c2562012-06-03 00:30:00 +0100328static void __cell_release_no_holder(struct dm_bio_prison_cell *cell,
329 struct bio_list *inmates)
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100330{
331 struct bio_prison *prison = cell->prison;
332
333 hlist_del(&cell->list);
334 bio_list_merge(inmates, &cell->bios);
335
336 mempool_free(cell, prison->cell_pool);
337}
338
Mike Snitzera24c2562012-06-03 00:30:00 +0100339static void cell_release_no_holder(struct dm_bio_prison_cell *cell,
340 struct bio_list *inmates)
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100341{
342 unsigned long flags;
343 struct bio_prison *prison = cell->prison;
344
345 spin_lock_irqsave(&prison->lock, flags);
346 __cell_release_no_holder(cell, inmates);
347 spin_unlock_irqrestore(&prison->lock, flags);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000348}
349
Mike Snitzera24c2562012-06-03 00:30:00 +0100350static void cell_error(struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000351{
352 struct bio_prison *prison = cell->prison;
353 struct bio_list bios;
354 struct bio *bio;
355 unsigned long flags;
356
357 bio_list_init(&bios);
358
359 spin_lock_irqsave(&prison->lock, flags);
360 __cell_release(cell, &bios);
361 spin_unlock_irqrestore(&prison->lock, flags);
362
363 while ((bio = bio_list_pop(&bios)))
364 bio_io_error(bio);
365}
366
367/*----------------------------------------------------------------*/
368
369/*
370 * We use the deferred set to keep track of pending reads to shared blocks.
371 * We do this to ensure the new mapping caused by a write isn't performed
372 * until these prior reads have completed. Otherwise the insertion of the
373 * new mapping could free the old block that the read bios are mapped to.
374 */
375
376struct deferred_set;
377struct deferred_entry {
378 struct deferred_set *ds;
379 unsigned count;
380 struct list_head work_items;
381};
382
383struct deferred_set {
384 spinlock_t lock;
385 unsigned current_entry;
386 unsigned sweeper;
387 struct deferred_entry entries[DEFERRED_SET_SIZE];
388};
389
390static void ds_init(struct deferred_set *ds)
391{
392 int i;
393
394 spin_lock_init(&ds->lock);
395 ds->current_entry = 0;
396 ds->sweeper = 0;
397 for (i = 0; i < DEFERRED_SET_SIZE; i++) {
398 ds->entries[i].ds = ds;
399 ds->entries[i].count = 0;
400 INIT_LIST_HEAD(&ds->entries[i].work_items);
401 }
402}
403
404static struct deferred_entry *ds_inc(struct deferred_set *ds)
405{
406 unsigned long flags;
407 struct deferred_entry *entry;
408
409 spin_lock_irqsave(&ds->lock, flags);
410 entry = ds->entries + ds->current_entry;
411 entry->count++;
412 spin_unlock_irqrestore(&ds->lock, flags);
413
414 return entry;
415}
416
417static unsigned ds_next(unsigned index)
418{
419 return (index + 1) % DEFERRED_SET_SIZE;
420}
421
422static void __sweep(struct deferred_set *ds, struct list_head *head)
423{
424 while ((ds->sweeper != ds->current_entry) &&
425 !ds->entries[ds->sweeper].count) {
426 list_splice_init(&ds->entries[ds->sweeper].work_items, head);
427 ds->sweeper = ds_next(ds->sweeper);
428 }
429
430 if ((ds->sweeper == ds->current_entry) && !ds->entries[ds->sweeper].count)
431 list_splice_init(&ds->entries[ds->sweeper].work_items, head);
432}
433
434static void ds_dec(struct deferred_entry *entry, struct list_head *head)
435{
436 unsigned long flags;
437
438 spin_lock_irqsave(&entry->ds->lock, flags);
439 BUG_ON(!entry->count);
440 --entry->count;
441 __sweep(entry->ds, head);
442 spin_unlock_irqrestore(&entry->ds->lock, flags);
443}
444
445/*
446 * Returns 1 if deferred or 0 if no pending items to delay job.
447 */
448static int ds_add_work(struct deferred_set *ds, struct list_head *work)
449{
450 int r = 1;
451 unsigned long flags;
452 unsigned next_entry;
453
454 spin_lock_irqsave(&ds->lock, flags);
455 if ((ds->sweeper == ds->current_entry) &&
456 !ds->entries[ds->current_entry].count)
457 r = 0;
458 else {
459 list_add(work, &ds->entries[ds->current_entry].work_items);
460 next_entry = ds_next(ds->current_entry);
461 if (!ds->entries[next_entry].count)
462 ds->current_entry = next_entry;
463 }
464 spin_unlock_irqrestore(&ds->lock, flags);
465
466 return r;
467}
468
469/*----------------------------------------------------------------*/
470
471/*
472 * Key building.
473 */
474static void build_data_key(struct dm_thin_device *td,
475 dm_block_t b, struct cell_key *key)
476{
477 key->virtual = 0;
478 key->dev = dm_thin_dev_id(td);
479 key->block = b;
480}
481
482static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
483 struct cell_key *key)
484{
485 key->virtual = 1;
486 key->dev = dm_thin_dev_id(td);
487 key->block = b;
488}
489
490/*----------------------------------------------------------------*/
491
492/*
493 * A pool device ties together a metadata device and a data device. It
494 * also provides the interface for creating and destroying internal
495 * devices.
496 */
Mike Snitzera24c2562012-06-03 00:30:00 +0100497struct dm_thin_new_mapping;
Joe Thornber67e2e2b2012-03-28 18:41:29 +0100498
499struct pool_features {
500 unsigned zero_new_blocks:1;
501 unsigned discard_enabled:1;
502 unsigned discard_passdown:1;
503};
504
Joe Thornber991d9fa2011-10-31 20:21:18 +0000505struct pool {
506 struct list_head list;
507 struct dm_target *ti; /* Only set if a pool target is bound */
508
509 struct mapped_device *pool_md;
510 struct block_device *md_dev;
511 struct dm_pool_metadata *pmd;
512
513 uint32_t sectors_per_block;
514 unsigned block_shift;
515 dm_block_t offset_mask;
516 dm_block_t low_water_blocks;
517
Joe Thornber67e2e2b2012-03-28 18:41:29 +0100518 struct pool_features pf;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000519 unsigned low_water_triggered:1; /* A dm event has been sent */
520 unsigned no_free_space:1; /* A -ENOSPC warning has been issued */
521
522 struct bio_prison *prison;
523 struct dm_kcopyd_client *copier;
524
525 struct workqueue_struct *wq;
526 struct work_struct worker;
Joe Thornber905e51b2012-03-28 18:41:27 +0100527 struct delayed_work waker;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000528
529 unsigned ref_count;
Joe Thornber905e51b2012-03-28 18:41:27 +0100530 unsigned long last_commit_jiffies;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000531
532 spinlock_t lock;
533 struct bio_list deferred_bios;
534 struct bio_list deferred_flush_bios;
535 struct list_head prepared_mappings;
Joe Thornber104655f2012-03-28 18:41:28 +0100536 struct list_head prepared_discards;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000537
538 struct bio_list retry_on_resume_list;
539
Joe Thornbereb2aa482012-03-28 18:41:28 +0100540 struct deferred_set shared_read_ds;
Joe Thornber104655f2012-03-28 18:41:28 +0100541 struct deferred_set all_io_ds;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000542
Mike Snitzera24c2562012-06-03 00:30:00 +0100543 struct dm_thin_new_mapping *next_mapping;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000544 mempool_t *mapping_pool;
545 mempool_t *endio_hook_pool;
546};
547
548/*
549 * Target context for a pool.
550 */
551struct pool_c {
552 struct dm_target *ti;
553 struct pool *pool;
554 struct dm_dev *data_dev;
555 struct dm_dev *metadata_dev;
556 struct dm_target_callbacks callbacks;
557
558 dm_block_t low_water_blocks;
Joe Thornber67e2e2b2012-03-28 18:41:29 +0100559 struct pool_features pf;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000560};
561
562/*
563 * Target context for a thin.
564 */
565struct thin_c {
566 struct dm_dev *pool_dev;
Joe Thornber2dd9c252012-03-28 18:41:28 +0100567 struct dm_dev *origin_dev;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000568 dm_thin_id dev_id;
569
570 struct pool *pool;
571 struct dm_thin_device *td;
572};
573
574/*----------------------------------------------------------------*/
575
576/*
577 * A global list of pools that uses a struct mapped_device as a key.
578 */
579static struct dm_thin_pool_table {
580 struct mutex mutex;
581 struct list_head pools;
582} dm_thin_pool_table;
583
584static void pool_table_init(void)
585{
586 mutex_init(&dm_thin_pool_table.mutex);
587 INIT_LIST_HEAD(&dm_thin_pool_table.pools);
588}
589
590static void __pool_table_insert(struct pool *pool)
591{
592 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
593 list_add(&pool->list, &dm_thin_pool_table.pools);
594}
595
596static void __pool_table_remove(struct pool *pool)
597{
598 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
599 list_del(&pool->list);
600}
601
602static struct pool *__pool_table_lookup(struct mapped_device *md)
603{
604 struct pool *pool = NULL, *tmp;
605
606 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
607
608 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
609 if (tmp->pool_md == md) {
610 pool = tmp;
611 break;
612 }
613 }
614
615 return pool;
616}
617
618static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
619{
620 struct pool *pool = NULL, *tmp;
621
622 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
623
624 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
625 if (tmp->md_dev == md_dev) {
626 pool = tmp;
627 break;
628 }
629 }
630
631 return pool;
632}
633
634/*----------------------------------------------------------------*/
635
Mike Snitzera24c2562012-06-03 00:30:00 +0100636struct dm_thin_endio_hook {
Joe Thornbereb2aa482012-03-28 18:41:28 +0100637 struct thin_c *tc;
638 struct deferred_entry *shared_read_entry;
Joe Thornber104655f2012-03-28 18:41:28 +0100639 struct deferred_entry *all_io_entry;
Mike Snitzera24c2562012-06-03 00:30:00 +0100640 struct dm_thin_new_mapping *overwrite_mapping;
Joe Thornbereb2aa482012-03-28 18:41:28 +0100641};
642
Joe Thornber991d9fa2011-10-31 20:21:18 +0000643static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master)
644{
645 struct bio *bio;
646 struct bio_list bios;
647
648 bio_list_init(&bios);
649 bio_list_merge(&bios, master);
650 bio_list_init(master);
651
652 while ((bio = bio_list_pop(&bios))) {
Mike Snitzera24c2562012-06-03 00:30:00 +0100653 struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
654
Joe Thornbereb2aa482012-03-28 18:41:28 +0100655 if (h->tc == tc)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000656 bio_endio(bio, DM_ENDIO_REQUEUE);
657 else
658 bio_list_add(master, bio);
659 }
660}
661
662static void requeue_io(struct thin_c *tc)
663{
664 struct pool *pool = tc->pool;
665 unsigned long flags;
666
667 spin_lock_irqsave(&pool->lock, flags);
668 __requeue_bio_list(tc, &pool->deferred_bios);
669 __requeue_bio_list(tc, &pool->retry_on_resume_list);
670 spin_unlock_irqrestore(&pool->lock, flags);
671}
672
673/*
674 * This section of code contains the logic for processing a thin device's IO.
675 * Much of the code depends on pool object resources (lists, workqueues, etc)
676 * but most is exclusively called from the thin target rather than the thin-pool
677 * target.
678 */
679
680static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
681{
682 return bio->bi_sector >> tc->pool->block_shift;
683}
684
685static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
686{
687 struct pool *pool = tc->pool;
688
689 bio->bi_bdev = tc->pool_dev->bdev;
690 bio->bi_sector = (block << pool->block_shift) +
691 (bio->bi_sector & pool->offset_mask);
692}
693
Joe Thornber2dd9c252012-03-28 18:41:28 +0100694static void remap_to_origin(struct thin_c *tc, struct bio *bio)
695{
696 bio->bi_bdev = tc->origin_dev->bdev;
697}
698
699static void issue(struct thin_c *tc, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000700{
701 struct pool *pool = tc->pool;
702 unsigned long flags;
703
Joe Thornber991d9fa2011-10-31 20:21:18 +0000704 /*
705 * Batch together any FUA/FLUSH bios we find and then issue
706 * a single commit for them in process_deferred_bios().
707 */
708 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
709 spin_lock_irqsave(&pool->lock, flags);
710 bio_list_add(&pool->deferred_flush_bios, bio);
711 spin_unlock_irqrestore(&pool->lock, flags);
712 } else
713 generic_make_request(bio);
714}
715
Joe Thornber2dd9c252012-03-28 18:41:28 +0100716static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
717{
718 remap_to_origin(tc, bio);
719 issue(tc, bio);
720}
721
722static void remap_and_issue(struct thin_c *tc, struct bio *bio,
723 dm_block_t block)
724{
725 remap(tc, bio, block);
726 issue(tc, bio);
727}
728
Joe Thornber991d9fa2011-10-31 20:21:18 +0000729/*
730 * wake_worker() is used when new work is queued and when pool_resume is
731 * ready to continue deferred IO processing.
732 */
733static void wake_worker(struct pool *pool)
734{
735 queue_work(pool->wq, &pool->worker);
736}
737
738/*----------------------------------------------------------------*/
739
740/*
741 * Bio endio functions.
742 */
Mike Snitzera24c2562012-06-03 00:30:00 +0100743struct dm_thin_new_mapping {
Joe Thornber991d9fa2011-10-31 20:21:18 +0000744 struct list_head list;
745
Joe Thornbereb2aa482012-03-28 18:41:28 +0100746 unsigned quiesced:1;
747 unsigned prepared:1;
Joe Thornber104655f2012-03-28 18:41:28 +0100748 unsigned pass_discard:1;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000749
750 struct thin_c *tc;
751 dm_block_t virt_block;
752 dm_block_t data_block;
Mike Snitzera24c2562012-06-03 00:30:00 +0100753 struct dm_bio_prison_cell *cell, *cell2;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000754 int err;
755
756 /*
757 * If the bio covers the whole area of a block then we can avoid
758 * zeroing or copying. Instead this bio is hooked. The bio will
759 * still be in the cell, so care has to be taken to avoid issuing
760 * the bio twice.
761 */
762 struct bio *bio;
763 bio_end_io_t *saved_bi_end_io;
764};
765
Mike Snitzera24c2562012-06-03 00:30:00 +0100766static void __maybe_add_mapping(struct dm_thin_new_mapping *m)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000767{
768 struct pool *pool = m->tc->pool;
769
Joe Thornbereb2aa482012-03-28 18:41:28 +0100770 if (m->quiesced && m->prepared) {
Joe Thornber991d9fa2011-10-31 20:21:18 +0000771 list_add(&m->list, &pool->prepared_mappings);
772 wake_worker(pool);
773 }
774}
775
776static void copy_complete(int read_err, unsigned long write_err, void *context)
777{
778 unsigned long flags;
Mike Snitzera24c2562012-06-03 00:30:00 +0100779 struct dm_thin_new_mapping *m = context;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000780 struct pool *pool = m->tc->pool;
781
782 m->err = read_err || write_err ? -EIO : 0;
783
784 spin_lock_irqsave(&pool->lock, flags);
785 m->prepared = 1;
786 __maybe_add_mapping(m);
787 spin_unlock_irqrestore(&pool->lock, flags);
788}
789
790static void overwrite_endio(struct bio *bio, int err)
791{
792 unsigned long flags;
Mike Snitzera24c2562012-06-03 00:30:00 +0100793 struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
794 struct dm_thin_new_mapping *m = h->overwrite_mapping;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000795 struct pool *pool = m->tc->pool;
796
797 m->err = err;
798
799 spin_lock_irqsave(&pool->lock, flags);
800 m->prepared = 1;
801 __maybe_add_mapping(m);
802 spin_unlock_irqrestore(&pool->lock, flags);
803}
804
Joe Thornber991d9fa2011-10-31 20:21:18 +0000805/*----------------------------------------------------------------*/
806
807/*
808 * Workqueue.
809 */
810
811/*
812 * Prepared mapping jobs.
813 */
814
815/*
816 * This sends the bios in the cell back to the deferred_bios list.
817 */
Mike Snitzera24c2562012-06-03 00:30:00 +0100818static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell,
Joe Thornber991d9fa2011-10-31 20:21:18 +0000819 dm_block_t data_block)
820{
821 struct pool *pool = tc->pool;
822 unsigned long flags;
823
824 spin_lock_irqsave(&pool->lock, flags);
825 cell_release(cell, &pool->deferred_bios);
826 spin_unlock_irqrestore(&tc->pool->lock, flags);
827
828 wake_worker(pool);
829}
830
831/*
832 * Same as cell_defer above, except it omits one particular detainee,
833 * a write bio that covers the block and has already been processed.
834 */
Mike Snitzera24c2562012-06-03 00:30:00 +0100835static void cell_defer_except(struct thin_c *tc, struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000836{
837 struct bio_list bios;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000838 struct pool *pool = tc->pool;
839 unsigned long flags;
840
841 bio_list_init(&bios);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000842
843 spin_lock_irqsave(&pool->lock, flags);
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100844 cell_release_no_holder(cell, &pool->deferred_bios);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000845 spin_unlock_irqrestore(&pool->lock, flags);
846
847 wake_worker(pool);
848}
849
Mike Snitzera24c2562012-06-03 00:30:00 +0100850static void process_prepared_mapping(struct dm_thin_new_mapping *m)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000851{
852 struct thin_c *tc = m->tc;
853 struct bio *bio;
854 int r;
855
856 bio = m->bio;
857 if (bio)
858 bio->bi_end_io = m->saved_bi_end_io;
859
860 if (m->err) {
861 cell_error(m->cell);
862 return;
863 }
864
865 /*
866 * Commit the prepared block into the mapping btree.
867 * Any I/O for this block arriving after this point will get
868 * remapped to it directly.
869 */
870 r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block);
871 if (r) {
872 DMERR("dm_thin_insert_block() failed");
873 cell_error(m->cell);
874 return;
875 }
876
877 /*
878 * Release any bios held while the block was being provisioned.
879 * If we are processing a write bio that completely covers the block,
880 * we already processed it so can ignore it now when processing
881 * the bios in the cell.
882 */
883 if (bio) {
Joe Thornber6f94a4c2012-03-28 18:41:23 +0100884 cell_defer_except(tc, m->cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000885 bio_endio(bio, 0);
886 } else
887 cell_defer(tc, m->cell, m->data_block);
888
889 list_del(&m->list);
890 mempool_free(m, tc->pool->mapping_pool);
891}
892
Mike Snitzera24c2562012-06-03 00:30:00 +0100893static void process_prepared_discard(struct dm_thin_new_mapping *m)
Joe Thornber104655f2012-03-28 18:41:28 +0100894{
895 int r;
896 struct thin_c *tc = m->tc;
897
898 r = dm_thin_remove_block(tc->td, m->virt_block);
899 if (r)
900 DMERR("dm_thin_remove_block() failed");
901
902 /*
903 * Pass the discard down to the underlying device?
904 */
905 if (m->pass_discard)
906 remap_and_issue(tc, m->bio, m->data_block);
907 else
908 bio_endio(m->bio, 0);
909
910 cell_defer_except(tc, m->cell);
911 cell_defer_except(tc, m->cell2);
912 mempool_free(m, tc->pool->mapping_pool);
913}
914
915static void process_prepared(struct pool *pool, struct list_head *head,
Mike Snitzera24c2562012-06-03 00:30:00 +0100916 void (*fn)(struct dm_thin_new_mapping *))
Joe Thornber991d9fa2011-10-31 20:21:18 +0000917{
918 unsigned long flags;
919 struct list_head maps;
Mike Snitzera24c2562012-06-03 00:30:00 +0100920 struct dm_thin_new_mapping *m, *tmp;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000921
922 INIT_LIST_HEAD(&maps);
923 spin_lock_irqsave(&pool->lock, flags);
Joe Thornber104655f2012-03-28 18:41:28 +0100924 list_splice_init(head, &maps);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000925 spin_unlock_irqrestore(&pool->lock, flags);
926
927 list_for_each_entry_safe(m, tmp, &maps, list)
Joe Thornber104655f2012-03-28 18:41:28 +0100928 fn(m);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000929}
930
931/*
932 * Deferred bio jobs.
933 */
Joe Thornber104655f2012-03-28 18:41:28 +0100934static int io_overlaps_block(struct pool *pool, struct bio *bio)
935{
936 return !(bio->bi_sector & pool->offset_mask) &&
937 (bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT));
938
939}
940
Joe Thornber991d9fa2011-10-31 20:21:18 +0000941static int io_overwrites_block(struct pool *pool, struct bio *bio)
942{
Joe Thornber104655f2012-03-28 18:41:28 +0100943 return (bio_data_dir(bio) == WRITE) &&
944 io_overlaps_block(pool, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000945}
946
947static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
948 bio_end_io_t *fn)
949{
950 *save = bio->bi_end_io;
951 bio->bi_end_io = fn;
952}
953
954static int ensure_next_mapping(struct pool *pool)
955{
956 if (pool->next_mapping)
957 return 0;
958
959 pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);
960
961 return pool->next_mapping ? 0 : -ENOMEM;
962}
963
Mike Snitzera24c2562012-06-03 00:30:00 +0100964static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000965{
Mike Snitzera24c2562012-06-03 00:30:00 +0100966 struct dm_thin_new_mapping *r = pool->next_mapping;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000967
968 BUG_ON(!pool->next_mapping);
969
970 pool->next_mapping = NULL;
971
972 return r;
973}
974
975static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
Joe Thornber2dd9c252012-03-28 18:41:28 +0100976 struct dm_dev *origin, dm_block_t data_origin,
977 dm_block_t data_dest,
Mike Snitzera24c2562012-06-03 00:30:00 +0100978 struct dm_bio_prison_cell *cell, struct bio *bio)
Joe Thornber991d9fa2011-10-31 20:21:18 +0000979{
980 int r;
981 struct pool *pool = tc->pool;
Mike Snitzera24c2562012-06-03 00:30:00 +0100982 struct dm_thin_new_mapping *m = get_next_mapping(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +0000983
984 INIT_LIST_HEAD(&m->list);
Joe Thornbereb2aa482012-03-28 18:41:28 +0100985 m->quiesced = 0;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000986 m->prepared = 0;
987 m->tc = tc;
988 m->virt_block = virt_block;
989 m->data_block = data_dest;
990 m->cell = cell;
991 m->err = 0;
992 m->bio = NULL;
993
Joe Thornbereb2aa482012-03-28 18:41:28 +0100994 if (!ds_add_work(&pool->shared_read_ds, &m->list))
995 m->quiesced = 1;
Joe Thornber991d9fa2011-10-31 20:21:18 +0000996
997 /*
998 * IO to pool_dev remaps to the pool target's data_dev.
999 *
1000 * If the whole block of data is being overwritten, we can issue the
1001 * bio immediately. Otherwise we use kcopyd to clone the data first.
1002 */
1003 if (io_overwrites_block(pool, bio)) {
Mike Snitzera24c2562012-06-03 00:30:00 +01001004 struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
1005
Joe Thornbereb2aa482012-03-28 18:41:28 +01001006 h->overwrite_mapping = m;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001007 m->bio = bio;
1008 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001009 remap_and_issue(tc, bio, data_dest);
1010 } else {
1011 struct dm_io_region from, to;
1012
Joe Thornber2dd9c252012-03-28 18:41:28 +01001013 from.bdev = origin->bdev;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001014 from.sector = data_origin * pool->sectors_per_block;
1015 from.count = pool->sectors_per_block;
1016
1017 to.bdev = tc->pool_dev->bdev;
1018 to.sector = data_dest * pool->sectors_per_block;
1019 to.count = pool->sectors_per_block;
1020
1021 r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
1022 0, copy_complete, m);
1023 if (r < 0) {
1024 mempool_free(m, pool->mapping_pool);
1025 DMERR("dm_kcopyd_copy() failed");
1026 cell_error(cell);
1027 }
1028 }
1029}
1030
Joe Thornber2dd9c252012-03-28 18:41:28 +01001031static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
1032 dm_block_t data_origin, dm_block_t data_dest,
Mike Snitzera24c2562012-06-03 00:30:00 +01001033 struct dm_bio_prison_cell *cell, struct bio *bio)
Joe Thornber2dd9c252012-03-28 18:41:28 +01001034{
1035 schedule_copy(tc, virt_block, tc->pool_dev,
1036 data_origin, data_dest, cell, bio);
1037}
1038
1039static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
1040 dm_block_t data_dest,
Mike Snitzera24c2562012-06-03 00:30:00 +01001041 struct dm_bio_prison_cell *cell, struct bio *bio)
Joe Thornber2dd9c252012-03-28 18:41:28 +01001042{
1043 schedule_copy(tc, virt_block, tc->origin_dev,
1044 virt_block, data_dest, cell, bio);
1045}
1046
Joe Thornber991d9fa2011-10-31 20:21:18 +00001047static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
Mike Snitzera24c2562012-06-03 00:30:00 +01001048 dm_block_t data_block, struct dm_bio_prison_cell *cell,
Joe Thornber991d9fa2011-10-31 20:21:18 +00001049 struct bio *bio)
1050{
1051 struct pool *pool = tc->pool;
Mike Snitzera24c2562012-06-03 00:30:00 +01001052 struct dm_thin_new_mapping *m = get_next_mapping(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001053
1054 INIT_LIST_HEAD(&m->list);
Joe Thornbereb2aa482012-03-28 18:41:28 +01001055 m->quiesced = 1;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001056 m->prepared = 0;
1057 m->tc = tc;
1058 m->virt_block = virt_block;
1059 m->data_block = data_block;
1060 m->cell = cell;
1061 m->err = 0;
1062 m->bio = NULL;
1063
1064 /*
1065 * If the whole block of data is being overwritten or we are not
1066 * zeroing pre-existing data, we can issue the bio immediately.
1067 * Otherwise we use kcopyd to zero the data first.
1068 */
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001069 if (!pool->pf.zero_new_blocks)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001070 process_prepared_mapping(m);
1071
1072 else if (io_overwrites_block(pool, bio)) {
Mike Snitzera24c2562012-06-03 00:30:00 +01001073 struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
1074
Joe Thornbereb2aa482012-03-28 18:41:28 +01001075 h->overwrite_mapping = m;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001076 m->bio = bio;
1077 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001078 remap_and_issue(tc, bio, data_block);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001079 } else {
1080 int r;
1081 struct dm_io_region to;
1082
1083 to.bdev = tc->pool_dev->bdev;
1084 to.sector = data_block * pool->sectors_per_block;
1085 to.count = pool->sectors_per_block;
1086
1087 r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m);
1088 if (r < 0) {
1089 mempool_free(m, pool->mapping_pool);
1090 DMERR("dm_kcopyd_zero() failed");
1091 cell_error(cell);
1092 }
1093 }
1094}
1095
1096static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
1097{
1098 int r;
1099 dm_block_t free_blocks;
1100 unsigned long flags;
1101 struct pool *pool = tc->pool;
1102
1103 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1104 if (r)
1105 return r;
1106
1107 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
1108 DMWARN("%s: reached low water mark, sending event.",
1109 dm_device_name(pool->pool_md));
1110 spin_lock_irqsave(&pool->lock, flags);
1111 pool->low_water_triggered = 1;
1112 spin_unlock_irqrestore(&pool->lock, flags);
1113 dm_table_event(pool->ti->table);
1114 }
1115
1116 if (!free_blocks) {
1117 if (pool->no_free_space)
1118 return -ENOSPC;
1119 else {
1120 /*
1121 * Try to commit to see if that will free up some
1122 * more space.
1123 */
1124 r = dm_pool_commit_metadata(pool->pmd);
1125 if (r) {
1126 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
1127 __func__, r);
1128 return r;
1129 }
1130
1131 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1132 if (r)
1133 return r;
1134
1135 /*
1136 * If we still have no space we set a flag to avoid
1137 * doing all this checking and return -ENOSPC.
1138 */
1139 if (!free_blocks) {
1140 DMWARN("%s: no free space available.",
1141 dm_device_name(pool->pool_md));
1142 spin_lock_irqsave(&pool->lock, flags);
1143 pool->no_free_space = 1;
1144 spin_unlock_irqrestore(&pool->lock, flags);
1145 return -ENOSPC;
1146 }
1147 }
1148 }
1149
1150 r = dm_pool_alloc_data_block(pool->pmd, result);
1151 if (r)
1152 return r;
1153
1154 return 0;
1155}
1156
1157/*
1158 * If we have run out of space, queue bios until the device is
1159 * resumed, presumably after having been reloaded with more space.
1160 */
1161static void retry_on_resume(struct bio *bio)
1162{
Mike Snitzera24c2562012-06-03 00:30:00 +01001163 struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
Joe Thornbereb2aa482012-03-28 18:41:28 +01001164 struct thin_c *tc = h->tc;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001165 struct pool *pool = tc->pool;
1166 unsigned long flags;
1167
1168 spin_lock_irqsave(&pool->lock, flags);
1169 bio_list_add(&pool->retry_on_resume_list, bio);
1170 spin_unlock_irqrestore(&pool->lock, flags);
1171}
1172
Mike Snitzera24c2562012-06-03 00:30:00 +01001173static void no_space(struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001174{
1175 struct bio *bio;
1176 struct bio_list bios;
1177
1178 bio_list_init(&bios);
1179 cell_release(cell, &bios);
1180
1181 while ((bio = bio_list_pop(&bios)))
1182 retry_on_resume(bio);
1183}
1184
Joe Thornber104655f2012-03-28 18:41:28 +01001185static void process_discard(struct thin_c *tc, struct bio *bio)
1186{
1187 int r;
Mike Snitzerc3a0ce22012-05-12 01:43:16 +01001188 unsigned long flags;
Joe Thornber104655f2012-03-28 18:41:28 +01001189 struct pool *pool = tc->pool;
Mike Snitzera24c2562012-06-03 00:30:00 +01001190 struct dm_bio_prison_cell *cell, *cell2;
Joe Thornber104655f2012-03-28 18:41:28 +01001191 struct cell_key key, key2;
1192 dm_block_t block = get_bio_block(tc, bio);
1193 struct dm_thin_lookup_result lookup_result;
Mike Snitzera24c2562012-06-03 00:30:00 +01001194 struct dm_thin_new_mapping *m;
Joe Thornber104655f2012-03-28 18:41:28 +01001195
1196 build_virtual_key(tc->td, block, &key);
1197 if (bio_detain(tc->pool->prison, &key, bio, &cell))
1198 return;
1199
1200 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1201 switch (r) {
1202 case 0:
1203 /*
1204 * Check nobody is fiddling with this pool block. This can
1205 * happen if someone's in the process of breaking sharing
1206 * on this block.
1207 */
1208 build_data_key(tc->td, lookup_result.block, &key2);
1209 if (bio_detain(tc->pool->prison, &key2, bio, &cell2)) {
1210 cell_release_singleton(cell, bio);
1211 break;
1212 }
1213
1214 if (io_overlaps_block(pool, bio)) {
1215 /*
1216 * IO may still be going to the destination block. We must
1217 * quiesce before we can do the removal.
1218 */
1219 m = get_next_mapping(pool);
1220 m->tc = tc;
Mike Snitzer17b7d632012-07-27 15:07:57 +01001221 m->pass_discard = (!lookup_result.shared) && pool->pf.discard_passdown;
Joe Thornber104655f2012-03-28 18:41:28 +01001222 m->virt_block = block;
1223 m->data_block = lookup_result.block;
1224 m->cell = cell;
1225 m->cell2 = cell2;
1226 m->err = 0;
1227 m->bio = bio;
1228
1229 if (!ds_add_work(&pool->all_io_ds, &m->list)) {
Mike Snitzerc3a0ce22012-05-12 01:43:16 +01001230 spin_lock_irqsave(&pool->lock, flags);
Joe Thornber104655f2012-03-28 18:41:28 +01001231 list_add(&m->list, &pool->prepared_discards);
Mike Snitzerc3a0ce22012-05-12 01:43:16 +01001232 spin_unlock_irqrestore(&pool->lock, flags);
Joe Thornber104655f2012-03-28 18:41:28 +01001233 wake_worker(pool);
1234 }
1235 } else {
1236 /*
1237 * This path is hit if people are ignoring
1238 * limits->discard_granularity. It ignores any
1239 * part of the discard that is in a subsequent
1240 * block.
1241 */
1242 sector_t offset = bio->bi_sector - (block << pool->block_shift);
1243 unsigned remaining = (pool->sectors_per_block - offset) << 9;
1244 bio->bi_size = min(bio->bi_size, remaining);
1245
1246 cell_release_singleton(cell, bio);
1247 cell_release_singleton(cell2, bio);
Mikulas Patocka650d2a02012-07-20 14:25:05 +01001248 if ((!lookup_result.shared) && pool->pf.discard_passdown)
1249 remap_and_issue(tc, bio, lookup_result.block);
1250 else
1251 bio_endio(bio, 0);
Joe Thornber104655f2012-03-28 18:41:28 +01001252 }
1253 break;
1254
1255 case -ENODATA:
1256 /*
1257 * It isn't provisioned, just forget it.
1258 */
1259 cell_release_singleton(cell, bio);
1260 bio_endio(bio, 0);
1261 break;
1262
1263 default:
1264 DMERR("discard: find block unexpectedly returned %d", r);
1265 cell_release_singleton(cell, bio);
1266 bio_io_error(bio);
1267 break;
1268 }
1269}
1270
Joe Thornber991d9fa2011-10-31 20:21:18 +00001271static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
1272 struct cell_key *key,
1273 struct dm_thin_lookup_result *lookup_result,
Mike Snitzera24c2562012-06-03 00:30:00 +01001274 struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001275{
1276 int r;
1277 dm_block_t data_block;
1278
1279 r = alloc_data_block(tc, &data_block);
1280 switch (r) {
1281 case 0:
Joe Thornber2dd9c252012-03-28 18:41:28 +01001282 schedule_internal_copy(tc, block, lookup_result->block,
1283 data_block, cell, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001284 break;
1285
1286 case -ENOSPC:
1287 no_space(cell);
1288 break;
1289
1290 default:
1291 DMERR("%s: alloc_data_block() failed, error = %d", __func__, r);
1292 cell_error(cell);
1293 break;
1294 }
1295}
1296
1297static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1298 dm_block_t block,
1299 struct dm_thin_lookup_result *lookup_result)
1300{
Mike Snitzera24c2562012-06-03 00:30:00 +01001301 struct dm_bio_prison_cell *cell;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001302 struct pool *pool = tc->pool;
1303 struct cell_key key;
1304
1305 /*
1306 * If cell is already occupied, then sharing is already in the process
1307 * of being broken so we have nothing further to do here.
1308 */
1309 build_data_key(tc->td, lookup_result->block, &key);
1310 if (bio_detain(pool->prison, &key, bio, &cell))
1311 return;
1312
1313 if (bio_data_dir(bio) == WRITE)
1314 break_sharing(tc, bio, block, &key, lookup_result, cell);
1315 else {
Mike Snitzera24c2562012-06-03 00:30:00 +01001316 struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001317
Joe Thornbereb2aa482012-03-28 18:41:28 +01001318 h->shared_read_entry = ds_inc(&pool->shared_read_ds);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001319
1320 cell_release_singleton(cell, bio);
1321 remap_and_issue(tc, bio, lookup_result->block);
1322 }
1323}
1324
1325static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
Mike Snitzera24c2562012-06-03 00:30:00 +01001326 struct dm_bio_prison_cell *cell)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001327{
1328 int r;
1329 dm_block_t data_block;
1330
1331 /*
1332 * Remap empty bios (flushes) immediately, without provisioning.
1333 */
1334 if (!bio->bi_size) {
1335 cell_release_singleton(cell, bio);
1336 remap_and_issue(tc, bio, 0);
1337 return;
1338 }
1339
1340 /*
1341 * Fill read bios with zeroes and complete them immediately.
1342 */
1343 if (bio_data_dir(bio) == READ) {
1344 zero_fill_bio(bio);
1345 cell_release_singleton(cell, bio);
1346 bio_endio(bio, 0);
1347 return;
1348 }
1349
1350 r = alloc_data_block(tc, &data_block);
1351 switch (r) {
1352 case 0:
Joe Thornber2dd9c252012-03-28 18:41:28 +01001353 if (tc->origin_dev)
1354 schedule_external_copy(tc, block, data_block, cell, bio);
1355 else
1356 schedule_zero(tc, block, data_block, cell, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001357 break;
1358
1359 case -ENOSPC:
1360 no_space(cell);
1361 break;
1362
1363 default:
1364 DMERR("%s: alloc_data_block() failed, error = %d", __func__, r);
1365 cell_error(cell);
1366 break;
1367 }
1368}
1369
1370static void process_bio(struct thin_c *tc, struct bio *bio)
1371{
1372 int r;
1373 dm_block_t block = get_bio_block(tc, bio);
Mike Snitzera24c2562012-06-03 00:30:00 +01001374 struct dm_bio_prison_cell *cell;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001375 struct cell_key key;
1376 struct dm_thin_lookup_result lookup_result;
1377
1378 /*
1379 * If cell is already occupied, then the block is already
1380 * being provisioned so we have nothing further to do here.
1381 */
1382 build_virtual_key(tc->td, block, &key);
1383 if (bio_detain(tc->pool->prison, &key, bio, &cell))
1384 return;
1385
1386 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1387 switch (r) {
1388 case 0:
1389 /*
1390 * We can release this cell now. This thread is the only
1391 * one that puts bios into a cell, and we know there were
1392 * no preceding bios.
1393 */
1394 /*
1395 * TODO: this will probably have to change when discard goes
1396 * back in.
1397 */
1398 cell_release_singleton(cell, bio);
1399
1400 if (lookup_result.shared)
1401 process_shared_bio(tc, bio, block, &lookup_result);
1402 else
1403 remap_and_issue(tc, bio, lookup_result.block);
1404 break;
1405
1406 case -ENODATA:
Joe Thornber2dd9c252012-03-28 18:41:28 +01001407 if (bio_data_dir(bio) == READ && tc->origin_dev) {
1408 cell_release_singleton(cell, bio);
1409 remap_to_origin_and_issue(tc, bio);
1410 } else
1411 provision_block(tc, bio, block, cell);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001412 break;
1413
1414 default:
1415 DMERR("dm_thin_find_block() failed, error = %d", r);
Joe Thornber104655f2012-03-28 18:41:28 +01001416 cell_release_singleton(cell, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001417 bio_io_error(bio);
1418 break;
1419 }
1420}
1421
Joe Thornber905e51b2012-03-28 18:41:27 +01001422static int need_commit_due_to_time(struct pool *pool)
1423{
1424 return jiffies < pool->last_commit_jiffies ||
1425 jiffies > pool->last_commit_jiffies + COMMIT_PERIOD;
1426}
1427
Joe Thornber991d9fa2011-10-31 20:21:18 +00001428static void process_deferred_bios(struct pool *pool)
1429{
1430 unsigned long flags;
1431 struct bio *bio;
1432 struct bio_list bios;
1433 int r;
1434
1435 bio_list_init(&bios);
1436
1437 spin_lock_irqsave(&pool->lock, flags);
1438 bio_list_merge(&bios, &pool->deferred_bios);
1439 bio_list_init(&pool->deferred_bios);
1440 spin_unlock_irqrestore(&pool->lock, flags);
1441
1442 while ((bio = bio_list_pop(&bios))) {
Mike Snitzera24c2562012-06-03 00:30:00 +01001443 struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
Joe Thornbereb2aa482012-03-28 18:41:28 +01001444 struct thin_c *tc = h->tc;
1445
Joe Thornber991d9fa2011-10-31 20:21:18 +00001446 /*
1447 * If we've got no free new_mapping structs, and processing
1448 * this bio might require one, we pause until there are some
1449 * prepared mappings to process.
1450 */
1451 if (ensure_next_mapping(pool)) {
1452 spin_lock_irqsave(&pool->lock, flags);
1453 bio_list_merge(&pool->deferred_bios, &bios);
1454 spin_unlock_irqrestore(&pool->lock, flags);
1455
1456 break;
1457 }
Joe Thornber104655f2012-03-28 18:41:28 +01001458
1459 if (bio->bi_rw & REQ_DISCARD)
1460 process_discard(tc, bio);
1461 else
1462 process_bio(tc, bio);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001463 }
1464
1465 /*
1466 * If there are any deferred flush bios, we must commit
1467 * the metadata before issuing them.
1468 */
1469 bio_list_init(&bios);
1470 spin_lock_irqsave(&pool->lock, flags);
1471 bio_list_merge(&bios, &pool->deferred_flush_bios);
1472 bio_list_init(&pool->deferred_flush_bios);
1473 spin_unlock_irqrestore(&pool->lock, flags);
1474
Joe Thornber905e51b2012-03-28 18:41:27 +01001475 if (bio_list_empty(&bios) && !need_commit_due_to_time(pool))
Joe Thornber991d9fa2011-10-31 20:21:18 +00001476 return;
1477
1478 r = dm_pool_commit_metadata(pool->pmd);
1479 if (r) {
1480 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
1481 __func__, r);
1482 while ((bio = bio_list_pop(&bios)))
1483 bio_io_error(bio);
1484 return;
1485 }
Joe Thornber905e51b2012-03-28 18:41:27 +01001486 pool->last_commit_jiffies = jiffies;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001487
1488 while ((bio = bio_list_pop(&bios)))
1489 generic_make_request(bio);
1490}
1491
1492static void do_worker(struct work_struct *ws)
1493{
1494 struct pool *pool = container_of(ws, struct pool, worker);
1495
Joe Thornber104655f2012-03-28 18:41:28 +01001496 process_prepared(pool, &pool->prepared_mappings, process_prepared_mapping);
1497 process_prepared(pool, &pool->prepared_discards, process_prepared_discard);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001498 process_deferred_bios(pool);
1499}
1500
Joe Thornber905e51b2012-03-28 18:41:27 +01001501/*
1502 * We want to commit periodically so that not too much
1503 * unwritten data builds up.
1504 */
1505static void do_waker(struct work_struct *ws)
1506{
1507 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
1508 wake_worker(pool);
1509 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
1510}
1511
Joe Thornber991d9fa2011-10-31 20:21:18 +00001512/*----------------------------------------------------------------*/
1513
1514/*
1515 * Mapping functions.
1516 */
1517
1518/*
1519 * Called only while mapping a thin bio to hand it over to the workqueue.
1520 */
1521static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
1522{
1523 unsigned long flags;
1524 struct pool *pool = tc->pool;
1525
1526 spin_lock_irqsave(&pool->lock, flags);
1527 bio_list_add(&pool->deferred_bios, bio);
1528 spin_unlock_irqrestore(&pool->lock, flags);
1529
1530 wake_worker(pool);
1531}
1532
Mike Snitzera24c2562012-06-03 00:30:00 +01001533static struct dm_thin_endio_hook *thin_hook_bio(struct thin_c *tc, struct bio *bio)
Joe Thornbereb2aa482012-03-28 18:41:28 +01001534{
1535 struct pool *pool = tc->pool;
Mike Snitzera24c2562012-06-03 00:30:00 +01001536 struct dm_thin_endio_hook *h = mempool_alloc(pool->endio_hook_pool, GFP_NOIO);
Joe Thornbereb2aa482012-03-28 18:41:28 +01001537
1538 h->tc = tc;
1539 h->shared_read_entry = NULL;
Joe Thornber104655f2012-03-28 18:41:28 +01001540 h->all_io_entry = bio->bi_rw & REQ_DISCARD ? NULL : ds_inc(&pool->all_io_ds);
Joe Thornbereb2aa482012-03-28 18:41:28 +01001541 h->overwrite_mapping = NULL;
1542
1543 return h;
1544}
1545
Joe Thornber991d9fa2011-10-31 20:21:18 +00001546/*
1547 * Non-blocking function called from the thin target's map function.
1548 */
1549static int thin_bio_map(struct dm_target *ti, struct bio *bio,
1550 union map_info *map_context)
1551{
1552 int r;
1553 struct thin_c *tc = ti->private;
1554 dm_block_t block = get_bio_block(tc, bio);
1555 struct dm_thin_device *td = tc->td;
1556 struct dm_thin_lookup_result result;
1557
Joe Thornbereb2aa482012-03-28 18:41:28 +01001558 map_context->ptr = thin_hook_bio(tc, bio);
Joe Thornber104655f2012-03-28 18:41:28 +01001559 if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00001560 thin_defer_bio(tc, bio);
1561 return DM_MAPIO_SUBMITTED;
1562 }
1563
1564 r = dm_thin_find_block(td, block, 0, &result);
1565
1566 /*
1567 * Note that we defer readahead too.
1568 */
1569 switch (r) {
1570 case 0:
1571 if (unlikely(result.shared)) {
1572 /*
1573 * We have a race condition here between the
1574 * result.shared value returned by the lookup and
1575 * snapshot creation, which may cause new
1576 * sharing.
1577 *
1578 * To avoid this always quiesce the origin before
1579 * taking the snap. You want to do this anyway to
1580 * ensure a consistent application view
1581 * (i.e. lockfs).
1582 *
1583 * More distant ancestors are irrelevant. The
1584 * shared flag will be set in their case.
1585 */
1586 thin_defer_bio(tc, bio);
1587 r = DM_MAPIO_SUBMITTED;
1588 } else {
1589 remap(tc, bio, result.block);
1590 r = DM_MAPIO_REMAPPED;
1591 }
1592 break;
1593
1594 case -ENODATA:
1595 /*
1596 * In future, the failed dm_thin_find_block above could
1597 * provide the hint to load the metadata into cache.
1598 */
1599 case -EWOULDBLOCK:
1600 thin_defer_bio(tc, bio);
1601 r = DM_MAPIO_SUBMITTED;
1602 break;
1603 }
1604
1605 return r;
1606}
1607
1608static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
1609{
1610 int r;
1611 unsigned long flags;
1612 struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
1613
1614 spin_lock_irqsave(&pt->pool->lock, flags);
1615 r = !bio_list_empty(&pt->pool->retry_on_resume_list);
1616 spin_unlock_irqrestore(&pt->pool->lock, flags);
1617
1618 if (!r) {
1619 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1620 r = bdi_congested(&q->backing_dev_info, bdi_bits);
1621 }
1622
1623 return r;
1624}
1625
1626static void __requeue_bios(struct pool *pool)
1627{
1628 bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list);
1629 bio_list_init(&pool->retry_on_resume_list);
1630}
1631
1632/*----------------------------------------------------------------
1633 * Binding of control targets to a pool object
1634 *--------------------------------------------------------------*/
1635static int bind_control_target(struct pool *pool, struct dm_target *ti)
1636{
1637 struct pool_c *pt = ti->private;
1638
1639 pool->ti = ti;
1640 pool->low_water_blocks = pt->low_water_blocks;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001641 pool->pf = pt->pf;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001642
Mike Snitzerf4026932012-05-19 01:01:01 +01001643 /*
1644 * If discard_passdown was enabled verify that the data device
1645 * supports discards. Disable discard_passdown if not; otherwise
1646 * -EOPNOTSUPP will be returned.
1647 */
1648 if (pt->pf.discard_passdown) {
1649 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1650 if (!q || !blk_queue_discard(q)) {
1651 char buf[BDEVNAME_SIZE];
1652 DMWARN("Discard unsupported by data device (%s): Disabling discard passdown.",
1653 bdevname(pt->data_dev->bdev, buf));
1654 pool->pf.discard_passdown = 0;
1655 }
1656 }
1657
Joe Thornber991d9fa2011-10-31 20:21:18 +00001658 return 0;
1659}
1660
1661static void unbind_control_target(struct pool *pool, struct dm_target *ti)
1662{
1663 if (pool->ti == ti)
1664 pool->ti = NULL;
1665}
1666
1667/*----------------------------------------------------------------
1668 * Pool creation
1669 *--------------------------------------------------------------*/
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001670/* Initialize pool features. */
1671static void pool_features_init(struct pool_features *pf)
1672{
1673 pf->zero_new_blocks = 1;
1674 pf->discard_enabled = 1;
1675 pf->discard_passdown = 1;
1676}
1677
Joe Thornber991d9fa2011-10-31 20:21:18 +00001678static void __pool_destroy(struct pool *pool)
1679{
1680 __pool_table_remove(pool);
1681
1682 if (dm_pool_metadata_close(pool->pmd) < 0)
1683 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1684
1685 prison_destroy(pool->prison);
1686 dm_kcopyd_client_destroy(pool->copier);
1687
1688 if (pool->wq)
1689 destroy_workqueue(pool->wq);
1690
1691 if (pool->next_mapping)
1692 mempool_free(pool->next_mapping, pool->mapping_pool);
1693 mempool_destroy(pool->mapping_pool);
1694 mempool_destroy(pool->endio_hook_pool);
1695 kfree(pool);
1696}
1697
Mike Snitzera24c2562012-06-03 00:30:00 +01001698static struct kmem_cache *_new_mapping_cache;
1699static struct kmem_cache *_endio_hook_cache;
1700
Joe Thornber991d9fa2011-10-31 20:21:18 +00001701static struct pool *pool_create(struct mapped_device *pool_md,
1702 struct block_device *metadata_dev,
1703 unsigned long block_size, char **error)
1704{
1705 int r;
1706 void *err_p;
1707 struct pool *pool;
1708 struct dm_pool_metadata *pmd;
1709
1710 pmd = dm_pool_metadata_open(metadata_dev, block_size);
1711 if (IS_ERR(pmd)) {
1712 *error = "Error creating metadata object";
1713 return (struct pool *)pmd;
1714 }
1715
1716 pool = kmalloc(sizeof(*pool), GFP_KERNEL);
1717 if (!pool) {
1718 *error = "Error allocating memory for pool";
1719 err_p = ERR_PTR(-ENOMEM);
1720 goto bad_pool;
1721 }
1722
1723 pool->pmd = pmd;
1724 pool->sectors_per_block = block_size;
1725 pool->block_shift = ffs(block_size) - 1;
1726 pool->offset_mask = block_size - 1;
1727 pool->low_water_blocks = 0;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001728 pool_features_init(&pool->pf);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001729 pool->prison = prison_create(PRISON_CELLS);
1730 if (!pool->prison) {
1731 *error = "Error creating pool's bio prison";
1732 err_p = ERR_PTR(-ENOMEM);
1733 goto bad_prison;
1734 }
1735
1736 pool->copier = dm_kcopyd_client_create();
1737 if (IS_ERR(pool->copier)) {
1738 r = PTR_ERR(pool->copier);
1739 *error = "Error creating pool's kcopyd client";
1740 err_p = ERR_PTR(r);
1741 goto bad_kcopyd_client;
1742 }
1743
1744 /*
1745 * Create singlethreaded workqueue that will service all devices
1746 * that use this metadata.
1747 */
1748 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
1749 if (!pool->wq) {
1750 *error = "Error creating pool's workqueue";
1751 err_p = ERR_PTR(-ENOMEM);
1752 goto bad_wq;
1753 }
1754
1755 INIT_WORK(&pool->worker, do_worker);
Joe Thornber905e51b2012-03-28 18:41:27 +01001756 INIT_DELAYED_WORK(&pool->waker, do_waker);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001757 spin_lock_init(&pool->lock);
1758 bio_list_init(&pool->deferred_bios);
1759 bio_list_init(&pool->deferred_flush_bios);
1760 INIT_LIST_HEAD(&pool->prepared_mappings);
Joe Thornber104655f2012-03-28 18:41:28 +01001761 INIT_LIST_HEAD(&pool->prepared_discards);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001762 pool->low_water_triggered = 0;
1763 pool->no_free_space = 0;
1764 bio_list_init(&pool->retry_on_resume_list);
Joe Thornbereb2aa482012-03-28 18:41:28 +01001765 ds_init(&pool->shared_read_ds);
Joe Thornber104655f2012-03-28 18:41:28 +01001766 ds_init(&pool->all_io_ds);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001767
1768 pool->next_mapping = NULL;
Mike Snitzera24c2562012-06-03 00:30:00 +01001769 pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE,
1770 _new_mapping_cache);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001771 if (!pool->mapping_pool) {
1772 *error = "Error creating pool's mapping mempool";
1773 err_p = ERR_PTR(-ENOMEM);
1774 goto bad_mapping_pool;
1775 }
1776
Mike Snitzera24c2562012-06-03 00:30:00 +01001777 pool->endio_hook_pool = mempool_create_slab_pool(ENDIO_HOOK_POOL_SIZE,
1778 _endio_hook_cache);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001779 if (!pool->endio_hook_pool) {
1780 *error = "Error creating pool's endio_hook mempool";
1781 err_p = ERR_PTR(-ENOMEM);
1782 goto bad_endio_hook_pool;
1783 }
1784 pool->ref_count = 1;
Joe Thornber905e51b2012-03-28 18:41:27 +01001785 pool->last_commit_jiffies = jiffies;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001786 pool->pool_md = pool_md;
1787 pool->md_dev = metadata_dev;
1788 __pool_table_insert(pool);
1789
1790 return pool;
1791
1792bad_endio_hook_pool:
1793 mempool_destroy(pool->mapping_pool);
1794bad_mapping_pool:
1795 destroy_workqueue(pool->wq);
1796bad_wq:
1797 dm_kcopyd_client_destroy(pool->copier);
1798bad_kcopyd_client:
1799 prison_destroy(pool->prison);
1800bad_prison:
1801 kfree(pool);
1802bad_pool:
1803 if (dm_pool_metadata_close(pmd))
1804 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1805
1806 return err_p;
1807}
1808
1809static void __pool_inc(struct pool *pool)
1810{
1811 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1812 pool->ref_count++;
1813}
1814
1815static void __pool_dec(struct pool *pool)
1816{
1817 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1818 BUG_ON(!pool->ref_count);
1819 if (!--pool->ref_count)
1820 __pool_destroy(pool);
1821}
1822
1823static struct pool *__pool_find(struct mapped_device *pool_md,
1824 struct block_device *metadata_dev,
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001825 unsigned long block_size, char **error,
1826 int *created)
Joe Thornber991d9fa2011-10-31 20:21:18 +00001827{
1828 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
1829
1830 if (pool) {
Mike Snitzerf09996c2012-07-27 15:07:59 +01001831 if (pool->pool_md != pool_md) {
1832 *error = "metadata device already in use by a pool";
Joe Thornber991d9fa2011-10-31 20:21:18 +00001833 return ERR_PTR(-EBUSY);
Mike Snitzerf09996c2012-07-27 15:07:59 +01001834 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001835 __pool_inc(pool);
1836
1837 } else {
1838 pool = __pool_table_lookup(pool_md);
1839 if (pool) {
Mike Snitzerf09996c2012-07-27 15:07:59 +01001840 if (pool->md_dev != metadata_dev) {
1841 *error = "different pool cannot replace a pool";
Joe Thornber991d9fa2011-10-31 20:21:18 +00001842 return ERR_PTR(-EINVAL);
Mike Snitzerf09996c2012-07-27 15:07:59 +01001843 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001844 __pool_inc(pool);
1845
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001846 } else {
Joe Thornber991d9fa2011-10-31 20:21:18 +00001847 pool = pool_create(pool_md, metadata_dev, block_size, error);
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001848 *created = 1;
1849 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00001850 }
1851
1852 return pool;
1853}
1854
1855/*----------------------------------------------------------------
1856 * Pool target methods
1857 *--------------------------------------------------------------*/
1858static void pool_dtr(struct dm_target *ti)
1859{
1860 struct pool_c *pt = ti->private;
1861
1862 mutex_lock(&dm_thin_pool_table.mutex);
1863
1864 unbind_control_target(pt->pool, ti);
1865 __pool_dec(pt->pool);
1866 dm_put_device(ti, pt->metadata_dev);
1867 dm_put_device(ti, pt->data_dev);
1868 kfree(pt);
1869
1870 mutex_unlock(&dm_thin_pool_table.mutex);
1871}
1872
Joe Thornber991d9fa2011-10-31 20:21:18 +00001873static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
1874 struct dm_target *ti)
1875{
1876 int r;
1877 unsigned argc;
1878 const char *arg_name;
1879
1880 static struct dm_arg _args[] = {
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001881 {0, 3, "Invalid number of pool feature arguments"},
Joe Thornber991d9fa2011-10-31 20:21:18 +00001882 };
1883
1884 /*
1885 * No feature arguments supplied.
1886 */
1887 if (!as->argc)
1888 return 0;
1889
1890 r = dm_read_arg_group(_args, as, &argc, &ti->error);
1891 if (r)
1892 return -EINVAL;
1893
1894 while (argc && !r) {
1895 arg_name = dm_shift_arg(as);
1896 argc--;
1897
1898 if (!strcasecmp(arg_name, "skip_block_zeroing")) {
1899 pf->zero_new_blocks = 0;
1900 continue;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001901 } else if (!strcasecmp(arg_name, "ignore_discard")) {
1902 pf->discard_enabled = 0;
1903 continue;
1904 } else if (!strcasecmp(arg_name, "no_discard_passdown")) {
1905 pf->discard_passdown = 0;
1906 continue;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001907 }
1908
1909 ti->error = "Unrecognised pool feature requested";
1910 r = -EINVAL;
1911 }
1912
1913 return r;
1914}
1915
1916/*
1917 * thin-pool <metadata dev> <data dev>
1918 * <data block size (sectors)>
1919 * <low water mark (blocks)>
1920 * [<#feature args> [<arg>]*]
1921 *
1922 * Optional feature arguments are:
1923 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001924 * ignore_discard: disable discard
1925 * no_discard_passdown: don't pass discards down to the data device
Joe Thornber991d9fa2011-10-31 20:21:18 +00001926 */
1927static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
1928{
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001929 int r, pool_created = 0;
Joe Thornber991d9fa2011-10-31 20:21:18 +00001930 struct pool_c *pt;
1931 struct pool *pool;
1932 struct pool_features pf;
1933 struct dm_arg_set as;
1934 struct dm_dev *data_dev;
1935 unsigned long block_size;
1936 dm_block_t low_water_blocks;
1937 struct dm_dev *metadata_dev;
1938 sector_t metadata_dev_size;
Mike Snitzerc4a69ec2012-03-28 18:41:28 +01001939 char b[BDEVNAME_SIZE];
Joe Thornber991d9fa2011-10-31 20:21:18 +00001940
1941 /*
1942 * FIXME Remove validation from scope of lock.
1943 */
1944 mutex_lock(&dm_thin_pool_table.mutex);
1945
1946 if (argc < 4) {
1947 ti->error = "Invalid argument count";
1948 r = -EINVAL;
1949 goto out_unlock;
1950 }
1951 as.argc = argc;
1952 as.argv = argv;
1953
1954 r = dm_get_device(ti, argv[0], FMODE_READ | FMODE_WRITE, &metadata_dev);
1955 if (r) {
1956 ti->error = "Error opening metadata block device";
1957 goto out_unlock;
1958 }
1959
1960 metadata_dev_size = i_size_read(metadata_dev->bdev->bd_inode) >> SECTOR_SHIFT;
Mike Snitzerc4a69ec2012-03-28 18:41:28 +01001961 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING)
1962 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
1963 bdevname(metadata_dev->bdev, b), THIN_METADATA_MAX_SECTORS);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001964
1965 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
1966 if (r) {
1967 ti->error = "Error getting data device";
1968 goto out_metadata;
1969 }
1970
1971 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
1972 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
1973 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
1974 !is_power_of_2(block_size)) {
1975 ti->error = "Invalid block size";
1976 r = -EINVAL;
1977 goto out;
1978 }
1979
1980 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
1981 ti->error = "Invalid low water mark";
1982 r = -EINVAL;
1983 goto out;
1984 }
1985
1986 /*
1987 * Set default pool features.
1988 */
Joe Thornber67e2e2b2012-03-28 18:41:29 +01001989 pool_features_init(&pf);
Joe Thornber991d9fa2011-10-31 20:21:18 +00001990
1991 dm_consume_args(&as, 4);
1992 r = parse_pool_features(&as, &pf, ti);
1993 if (r)
1994 goto out;
1995
1996 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
1997 if (!pt) {
1998 r = -ENOMEM;
1999 goto out;
2000 }
2001
2002 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002003 block_size, &ti->error, &pool_created);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002004 if (IS_ERR(pool)) {
2005 r = PTR_ERR(pool);
2006 goto out_free_pt;
2007 }
2008
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002009 /*
2010 * 'pool_created' reflects whether this is the first table load.
2011 * Top level discard support is not allowed to be changed after
2012 * initial load. This would require a pool reload to trigger thin
2013 * device changes.
2014 */
2015 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
2016 ti->error = "Discard support cannot be disabled once enabled";
2017 r = -EINVAL;
2018 goto out_flags_changed;
2019 }
2020
Joe Thornber991d9fa2011-10-31 20:21:18 +00002021 pt->pool = pool;
2022 pt->ti = ti;
2023 pt->metadata_dev = metadata_dev;
2024 pt->data_dev = data_dev;
2025 pt->low_water_blocks = low_water_blocks;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002026 pt->pf = pf;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002027 ti->num_flush_requests = 1;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002028 /*
2029 * Only need to enable discards if the pool should pass
2030 * them down to the data device. The thin device's discard
2031 * processing will cause mappings to be removed from the btree.
2032 */
2033 if (pf.discard_enabled && pf.discard_passdown) {
2034 ti->num_discard_requests = 1;
2035 /*
2036 * Setting 'discards_supported' circumvents the normal
2037 * stacking of discard limits (this keeps the pool and
2038 * thin devices' discard limits consistent).
2039 */
2040 ti->discards_supported = 1;
2041 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002042 ti->private = pt;
2043
2044 pt->callbacks.congested_fn = pool_is_congested;
2045 dm_table_add_target_callbacks(ti->table, &pt->callbacks);
2046
2047 mutex_unlock(&dm_thin_pool_table.mutex);
2048
2049 return 0;
2050
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002051out_flags_changed:
2052 __pool_dec(pool);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002053out_free_pt:
2054 kfree(pt);
2055out:
2056 dm_put_device(ti, data_dev);
2057out_metadata:
2058 dm_put_device(ti, metadata_dev);
2059out_unlock:
2060 mutex_unlock(&dm_thin_pool_table.mutex);
2061
2062 return r;
2063}
2064
2065static int pool_map(struct dm_target *ti, struct bio *bio,
2066 union map_info *map_context)
2067{
2068 int r;
2069 struct pool_c *pt = ti->private;
2070 struct pool *pool = pt->pool;
2071 unsigned long flags;
2072
2073 /*
2074 * As this is a singleton target, ti->begin is always zero.
2075 */
2076 spin_lock_irqsave(&pool->lock, flags);
2077 bio->bi_bdev = pt->data_dev->bdev;
2078 r = DM_MAPIO_REMAPPED;
2079 spin_unlock_irqrestore(&pool->lock, flags);
2080
2081 return r;
2082}
2083
2084/*
2085 * Retrieves the number of blocks of the data device from
2086 * the superblock and compares it to the actual device size,
2087 * thus resizing the data device in case it has grown.
2088 *
2089 * This both copes with opening preallocated data devices in the ctr
2090 * being followed by a resume
2091 * -and-
2092 * calling the resume method individually after userspace has
2093 * grown the data device in reaction to a table event.
2094 */
2095static int pool_preresume(struct dm_target *ti)
2096{
2097 int r;
2098 struct pool_c *pt = ti->private;
2099 struct pool *pool = pt->pool;
2100 dm_block_t data_size, sb_data_size;
2101
2102 /*
2103 * Take control of the pool object.
2104 */
2105 r = bind_control_target(pool, ti);
2106 if (r)
2107 return r;
2108
2109 data_size = ti->len >> pool->block_shift;
2110 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
2111 if (r) {
2112 DMERR("failed to retrieve data device size");
2113 return r;
2114 }
2115
2116 if (data_size < sb_data_size) {
2117 DMERR("pool target too small, is %llu blocks (expected %llu)",
2118 data_size, sb_data_size);
2119 return -EINVAL;
2120
2121 } else if (data_size > sb_data_size) {
2122 r = dm_pool_resize_data_dev(pool->pmd, data_size);
2123 if (r) {
2124 DMERR("failed to resize data device");
2125 return r;
2126 }
2127
2128 r = dm_pool_commit_metadata(pool->pmd);
2129 if (r) {
2130 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
2131 __func__, r);
2132 return r;
2133 }
2134 }
2135
2136 return 0;
2137}
2138
2139static void pool_resume(struct dm_target *ti)
2140{
2141 struct pool_c *pt = ti->private;
2142 struct pool *pool = pt->pool;
2143 unsigned long flags;
2144
2145 spin_lock_irqsave(&pool->lock, flags);
2146 pool->low_water_triggered = 0;
2147 pool->no_free_space = 0;
2148 __requeue_bios(pool);
2149 spin_unlock_irqrestore(&pool->lock, flags);
2150
Joe Thornber905e51b2012-03-28 18:41:27 +01002151 do_waker(&pool->waker.work);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002152}
2153
2154static void pool_postsuspend(struct dm_target *ti)
2155{
2156 int r;
2157 struct pool_c *pt = ti->private;
2158 struct pool *pool = pt->pool;
2159
Joe Thornber905e51b2012-03-28 18:41:27 +01002160 cancel_delayed_work(&pool->waker);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002161 flush_workqueue(pool->wq);
2162
2163 r = dm_pool_commit_metadata(pool->pmd);
2164 if (r < 0) {
2165 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
2166 __func__, r);
2167 /* FIXME: invalidate device? error the next FUA or FLUSH bio ?*/
2168 }
2169}
2170
2171static int check_arg_count(unsigned argc, unsigned args_required)
2172{
2173 if (argc != args_required) {
2174 DMWARN("Message received with %u arguments instead of %u.",
2175 argc, args_required);
2176 return -EINVAL;
2177 }
2178
2179 return 0;
2180}
2181
2182static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
2183{
2184 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
2185 *dev_id <= MAX_DEV_ID)
2186 return 0;
2187
2188 if (warning)
2189 DMWARN("Message received with invalid device id: %s", arg);
2190
2191 return -EINVAL;
2192}
2193
2194static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
2195{
2196 dm_thin_id dev_id;
2197 int r;
2198
2199 r = check_arg_count(argc, 2);
2200 if (r)
2201 return r;
2202
2203 r = read_dev_id(argv[1], &dev_id, 1);
2204 if (r)
2205 return r;
2206
2207 r = dm_pool_create_thin(pool->pmd, dev_id);
2208 if (r) {
2209 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2210 argv[1]);
2211 return r;
2212 }
2213
2214 return 0;
2215}
2216
2217static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2218{
2219 dm_thin_id dev_id;
2220 dm_thin_id origin_dev_id;
2221 int r;
2222
2223 r = check_arg_count(argc, 3);
2224 if (r)
2225 return r;
2226
2227 r = read_dev_id(argv[1], &dev_id, 1);
2228 if (r)
2229 return r;
2230
2231 r = read_dev_id(argv[2], &origin_dev_id, 1);
2232 if (r)
2233 return r;
2234
2235 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
2236 if (r) {
2237 DMWARN("Creation of new snapshot %s of device %s failed.",
2238 argv[1], argv[2]);
2239 return r;
2240 }
2241
2242 return 0;
2243}
2244
2245static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
2246{
2247 dm_thin_id dev_id;
2248 int r;
2249
2250 r = check_arg_count(argc, 2);
2251 if (r)
2252 return r;
2253
2254 r = read_dev_id(argv[1], &dev_id, 1);
2255 if (r)
2256 return r;
2257
2258 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
2259 if (r)
2260 DMWARN("Deletion of thin device %s failed.", argv[1]);
2261
2262 return r;
2263}
2264
2265static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
2266{
2267 dm_thin_id old_id, new_id;
2268 int r;
2269
2270 r = check_arg_count(argc, 3);
2271 if (r)
2272 return r;
2273
2274 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
2275 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
2276 return -EINVAL;
2277 }
2278
2279 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
2280 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
2281 return -EINVAL;
2282 }
2283
2284 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
2285 if (r) {
2286 DMWARN("Failed to change transaction id from %s to %s.",
2287 argv[1], argv[2]);
2288 return r;
2289 }
2290
2291 return 0;
2292}
2293
Joe Thornbercc8394d2012-06-03 00:30:01 +01002294static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2295{
2296 int r;
2297
2298 r = check_arg_count(argc, 1);
2299 if (r)
2300 return r;
2301
Joe Thornber0d200ae2012-07-03 12:55:31 +01002302 r = dm_pool_commit_metadata(pool->pmd);
2303 if (r) {
2304 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
2305 __func__, r);
2306 return r;
2307 }
2308
Joe Thornbercc8394d2012-06-03 00:30:01 +01002309 r = dm_pool_reserve_metadata_snap(pool->pmd);
2310 if (r)
2311 DMWARN("reserve_metadata_snap message failed.");
2312
2313 return r;
2314}
2315
2316static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2317{
2318 int r;
2319
2320 r = check_arg_count(argc, 1);
2321 if (r)
2322 return r;
2323
2324 r = dm_pool_release_metadata_snap(pool->pmd);
2325 if (r)
2326 DMWARN("release_metadata_snap message failed.");
2327
2328 return r;
2329}
2330
Joe Thornber991d9fa2011-10-31 20:21:18 +00002331/*
2332 * Messages supported:
2333 * create_thin <dev_id>
2334 * create_snap <dev_id> <origin_id>
2335 * delete <dev_id>
2336 * trim <dev_id> <new_size_in_sectors>
2337 * set_transaction_id <current_trans_id> <new_trans_id>
Joe Thornbercc8394d2012-06-03 00:30:01 +01002338 * reserve_metadata_snap
2339 * release_metadata_snap
Joe Thornber991d9fa2011-10-31 20:21:18 +00002340 */
2341static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
2342{
2343 int r = -EINVAL;
2344 struct pool_c *pt = ti->private;
2345 struct pool *pool = pt->pool;
2346
2347 if (!strcasecmp(argv[0], "create_thin"))
2348 r = process_create_thin_mesg(argc, argv, pool);
2349
2350 else if (!strcasecmp(argv[0], "create_snap"))
2351 r = process_create_snap_mesg(argc, argv, pool);
2352
2353 else if (!strcasecmp(argv[0], "delete"))
2354 r = process_delete_mesg(argc, argv, pool);
2355
2356 else if (!strcasecmp(argv[0], "set_transaction_id"))
2357 r = process_set_transaction_id_mesg(argc, argv, pool);
2358
Joe Thornbercc8394d2012-06-03 00:30:01 +01002359 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
2360 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
2361
2362 else if (!strcasecmp(argv[0], "release_metadata_snap"))
2363 r = process_release_metadata_snap_mesg(argc, argv, pool);
2364
Joe Thornber991d9fa2011-10-31 20:21:18 +00002365 else
2366 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
2367
2368 if (!r) {
2369 r = dm_pool_commit_metadata(pool->pmd);
2370 if (r)
2371 DMERR("%s message: dm_pool_commit_metadata() failed, error = %d",
2372 argv[0], r);
2373 }
2374
2375 return r;
2376}
2377
2378/*
2379 * Status line is:
2380 * <transaction id> <used metadata sectors>/<total metadata sectors>
2381 * <used data sectors>/<total data sectors> <held metadata root>
2382 */
2383static int pool_status(struct dm_target *ti, status_type_t type,
2384 char *result, unsigned maxlen)
2385{
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002386 int r, count;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002387 unsigned sz = 0;
2388 uint64_t transaction_id;
2389 dm_block_t nr_free_blocks_data;
2390 dm_block_t nr_free_blocks_metadata;
2391 dm_block_t nr_blocks_data;
2392 dm_block_t nr_blocks_metadata;
2393 dm_block_t held_root;
2394 char buf[BDEVNAME_SIZE];
2395 char buf2[BDEVNAME_SIZE];
2396 struct pool_c *pt = ti->private;
2397 struct pool *pool = pt->pool;
2398
2399 switch (type) {
2400 case STATUSTYPE_INFO:
2401 r = dm_pool_get_metadata_transaction_id(pool->pmd,
2402 &transaction_id);
2403 if (r)
2404 return r;
2405
2406 r = dm_pool_get_free_metadata_block_count(pool->pmd,
2407 &nr_free_blocks_metadata);
2408 if (r)
2409 return r;
2410
2411 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
2412 if (r)
2413 return r;
2414
2415 r = dm_pool_get_free_block_count(pool->pmd,
2416 &nr_free_blocks_data);
2417 if (r)
2418 return r;
2419
2420 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
2421 if (r)
2422 return r;
2423
Joe Thornbercc8394d2012-06-03 00:30:01 +01002424 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002425 if (r)
2426 return r;
2427
2428 DMEMIT("%llu %llu/%llu %llu/%llu ",
2429 (unsigned long long)transaction_id,
2430 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
2431 (unsigned long long)nr_blocks_metadata,
2432 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
2433 (unsigned long long)nr_blocks_data);
2434
2435 if (held_root)
2436 DMEMIT("%llu", held_root);
2437 else
2438 DMEMIT("-");
2439
2440 break;
2441
2442 case STATUSTYPE_TABLE:
2443 DMEMIT("%s %s %lu %llu ",
2444 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
2445 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
2446 (unsigned long)pool->sectors_per_block,
2447 (unsigned long long)pt->low_water_blocks);
2448
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002449 count = !pool->pf.zero_new_blocks + !pool->pf.discard_enabled +
Mike Snitzerf4026932012-05-19 01:01:01 +01002450 !pt->pf.discard_passdown;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002451 DMEMIT("%u ", count);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002452
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002453 if (!pool->pf.zero_new_blocks)
Joe Thornber991d9fa2011-10-31 20:21:18 +00002454 DMEMIT("skip_block_zeroing ");
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002455
2456 if (!pool->pf.discard_enabled)
2457 DMEMIT("ignore_discard ");
2458
Mike Snitzerf4026932012-05-19 01:01:01 +01002459 if (!pt->pf.discard_passdown)
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002460 DMEMIT("no_discard_passdown ");
2461
Joe Thornber991d9fa2011-10-31 20:21:18 +00002462 break;
2463 }
2464
2465 return 0;
2466}
2467
2468static int pool_iterate_devices(struct dm_target *ti,
2469 iterate_devices_callout_fn fn, void *data)
2470{
2471 struct pool_c *pt = ti->private;
2472
2473 return fn(ti, pt->data_dev, 0, ti->len, data);
2474}
2475
2476static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2477 struct bio_vec *biovec, int max_size)
2478{
2479 struct pool_c *pt = ti->private;
2480 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
2481
2482 if (!q->merge_bvec_fn)
2483 return max_size;
2484
2485 bvm->bi_bdev = pt->data_dev->bdev;
2486
2487 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2488}
2489
Joe Thornber104655f2012-03-28 18:41:28 +01002490static void set_discard_limits(struct pool *pool, struct queue_limits *limits)
2491{
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002492 /*
2493 * FIXME: these limits may be incompatible with the pool's data device
2494 */
Joe Thornber104655f2012-03-28 18:41:28 +01002495 limits->max_discard_sectors = pool->sectors_per_block;
2496
2497 /*
2498 * This is just a hint, and not enforced. We have to cope with
2499 * bios that overlap 2 blocks.
2500 */
2501 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002502 limits->discard_zeroes_data = pool->pf.zero_new_blocks;
Joe Thornber104655f2012-03-28 18:41:28 +01002503}
2504
Joe Thornber991d9fa2011-10-31 20:21:18 +00002505static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
2506{
2507 struct pool_c *pt = ti->private;
2508 struct pool *pool = pt->pool;
2509
2510 blk_limits_io_min(limits, 0);
2511 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002512 if (pool->pf.discard_enabled)
2513 set_discard_limits(pool, limits);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002514}
2515
2516static struct target_type pool_target = {
2517 .name = "thin-pool",
2518 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
2519 DM_TARGET_IMMUTABLE,
Joe Thornbercc8394d2012-06-03 00:30:01 +01002520 .version = {1, 2, 0},
Joe Thornber991d9fa2011-10-31 20:21:18 +00002521 .module = THIS_MODULE,
2522 .ctr = pool_ctr,
2523 .dtr = pool_dtr,
2524 .map = pool_map,
2525 .postsuspend = pool_postsuspend,
2526 .preresume = pool_preresume,
2527 .resume = pool_resume,
2528 .message = pool_message,
2529 .status = pool_status,
2530 .merge = pool_merge,
2531 .iterate_devices = pool_iterate_devices,
2532 .io_hints = pool_io_hints,
2533};
2534
2535/*----------------------------------------------------------------
2536 * Thin target methods
2537 *--------------------------------------------------------------*/
2538static void thin_dtr(struct dm_target *ti)
2539{
2540 struct thin_c *tc = ti->private;
2541
2542 mutex_lock(&dm_thin_pool_table.mutex);
2543
2544 __pool_dec(tc->pool);
2545 dm_pool_close_thin_device(tc->td);
2546 dm_put_device(ti, tc->pool_dev);
Joe Thornber2dd9c252012-03-28 18:41:28 +01002547 if (tc->origin_dev)
2548 dm_put_device(ti, tc->origin_dev);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002549 kfree(tc);
2550
2551 mutex_unlock(&dm_thin_pool_table.mutex);
2552}
2553
2554/*
2555 * Thin target parameters:
2556 *
Joe Thornber2dd9c252012-03-28 18:41:28 +01002557 * <pool_dev> <dev_id> [origin_dev]
Joe Thornber991d9fa2011-10-31 20:21:18 +00002558 *
2559 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2560 * dev_id: the internal device identifier
Joe Thornber2dd9c252012-03-28 18:41:28 +01002561 * origin_dev: a device external to the pool that should act as the origin
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002562 *
2563 * If the pool device has discards disabled, they get disabled for the thin
2564 * device as well.
Joe Thornber991d9fa2011-10-31 20:21:18 +00002565 */
2566static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
2567{
2568 int r;
2569 struct thin_c *tc;
Joe Thornber2dd9c252012-03-28 18:41:28 +01002570 struct dm_dev *pool_dev, *origin_dev;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002571 struct mapped_device *pool_md;
2572
2573 mutex_lock(&dm_thin_pool_table.mutex);
2574
Joe Thornber2dd9c252012-03-28 18:41:28 +01002575 if (argc != 2 && argc != 3) {
Joe Thornber991d9fa2011-10-31 20:21:18 +00002576 ti->error = "Invalid argument count";
2577 r = -EINVAL;
2578 goto out_unlock;
2579 }
2580
2581 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
2582 if (!tc) {
2583 ti->error = "Out of memory";
2584 r = -ENOMEM;
2585 goto out_unlock;
2586 }
2587
Joe Thornber2dd9c252012-03-28 18:41:28 +01002588 if (argc == 3) {
2589 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
2590 if (r) {
2591 ti->error = "Error opening origin device";
2592 goto bad_origin_dev;
2593 }
2594 tc->origin_dev = origin_dev;
2595 }
2596
Joe Thornber991d9fa2011-10-31 20:21:18 +00002597 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
2598 if (r) {
2599 ti->error = "Error opening pool device";
2600 goto bad_pool_dev;
2601 }
2602 tc->pool_dev = pool_dev;
2603
2604 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
2605 ti->error = "Invalid device id";
2606 r = -EINVAL;
2607 goto bad_common;
2608 }
2609
2610 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
2611 if (!pool_md) {
2612 ti->error = "Couldn't get pool mapped device";
2613 r = -EINVAL;
2614 goto bad_common;
2615 }
2616
2617 tc->pool = __pool_table_lookup(pool_md);
2618 if (!tc->pool) {
2619 ti->error = "Couldn't find pool object";
2620 r = -EINVAL;
2621 goto bad_pool_lookup;
2622 }
2623 __pool_inc(tc->pool);
2624
2625 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
2626 if (r) {
2627 ti->error = "Couldn't open thin internal device";
2628 goto bad_thin_open;
2629 }
2630
2631 ti->split_io = tc->pool->sectors_per_block;
2632 ti->num_flush_requests = 1;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002633
2634 /* In case the pool supports discards, pass them on. */
2635 if (tc->pool->pf.discard_enabled) {
2636 ti->discards_supported = 1;
2637 ti->num_discard_requests = 1;
Mikulas Patocka650d2a02012-07-20 14:25:05 +01002638 ti->discard_zeroes_data_unsupported = 1;
Joe Thornber67e2e2b2012-03-28 18:41:29 +01002639 }
Joe Thornber991d9fa2011-10-31 20:21:18 +00002640
2641 dm_put(pool_md);
2642
2643 mutex_unlock(&dm_thin_pool_table.mutex);
2644
2645 return 0;
2646
2647bad_thin_open:
2648 __pool_dec(tc->pool);
2649bad_pool_lookup:
2650 dm_put(pool_md);
2651bad_common:
2652 dm_put_device(ti, tc->pool_dev);
2653bad_pool_dev:
Joe Thornber2dd9c252012-03-28 18:41:28 +01002654 if (tc->origin_dev)
2655 dm_put_device(ti, tc->origin_dev);
2656bad_origin_dev:
Joe Thornber991d9fa2011-10-31 20:21:18 +00002657 kfree(tc);
2658out_unlock:
2659 mutex_unlock(&dm_thin_pool_table.mutex);
2660
2661 return r;
2662}
2663
2664static int thin_map(struct dm_target *ti, struct bio *bio,
2665 union map_info *map_context)
2666{
Alasdair G Kergon6efd6e82012-03-28 18:41:28 +01002667 bio->bi_sector = dm_target_offset(ti, bio->bi_sector);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002668
2669 return thin_bio_map(ti, bio, map_context);
2670}
2671
Joe Thornbereb2aa482012-03-28 18:41:28 +01002672static int thin_endio(struct dm_target *ti,
2673 struct bio *bio, int err,
2674 union map_info *map_context)
2675{
2676 unsigned long flags;
Mike Snitzera24c2562012-06-03 00:30:00 +01002677 struct dm_thin_endio_hook *h = map_context->ptr;
Joe Thornbereb2aa482012-03-28 18:41:28 +01002678 struct list_head work;
Mike Snitzera24c2562012-06-03 00:30:00 +01002679 struct dm_thin_new_mapping *m, *tmp;
Joe Thornbereb2aa482012-03-28 18:41:28 +01002680 struct pool *pool = h->tc->pool;
2681
2682 if (h->shared_read_entry) {
2683 INIT_LIST_HEAD(&work);
2684 ds_dec(h->shared_read_entry, &work);
2685
2686 spin_lock_irqsave(&pool->lock, flags);
2687 list_for_each_entry_safe(m, tmp, &work, list) {
2688 list_del(&m->list);
2689 m->quiesced = 1;
2690 __maybe_add_mapping(m);
2691 }
2692 spin_unlock_irqrestore(&pool->lock, flags);
2693 }
2694
Joe Thornber104655f2012-03-28 18:41:28 +01002695 if (h->all_io_entry) {
2696 INIT_LIST_HEAD(&work);
2697 ds_dec(h->all_io_entry, &work);
Mike Snitzerc3a0ce22012-05-12 01:43:16 +01002698 spin_lock_irqsave(&pool->lock, flags);
Joe Thornber104655f2012-03-28 18:41:28 +01002699 list_for_each_entry_safe(m, tmp, &work, list)
2700 list_add(&m->list, &pool->prepared_discards);
Mike Snitzerc3a0ce22012-05-12 01:43:16 +01002701 spin_unlock_irqrestore(&pool->lock, flags);
Joe Thornber104655f2012-03-28 18:41:28 +01002702 }
2703
Joe Thornbereb2aa482012-03-28 18:41:28 +01002704 mempool_free(h, pool->endio_hook_pool);
2705
2706 return 0;
2707}
2708
Joe Thornber991d9fa2011-10-31 20:21:18 +00002709static void thin_postsuspend(struct dm_target *ti)
2710{
2711 if (dm_noflush_suspending(ti))
2712 requeue_io((struct thin_c *)ti->private);
2713}
2714
2715/*
2716 * <nr mapped sectors> <highest mapped sector>
2717 */
2718static int thin_status(struct dm_target *ti, status_type_t type,
2719 char *result, unsigned maxlen)
2720{
2721 int r;
2722 ssize_t sz = 0;
2723 dm_block_t mapped, highest;
2724 char buf[BDEVNAME_SIZE];
2725 struct thin_c *tc = ti->private;
2726
2727 if (!tc->td)
2728 DMEMIT("-");
2729 else {
2730 switch (type) {
2731 case STATUSTYPE_INFO:
2732 r = dm_thin_get_mapped_count(tc->td, &mapped);
2733 if (r)
2734 return r;
2735
2736 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
2737 if (r < 0)
2738 return r;
2739
2740 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
2741 if (r)
2742 DMEMIT("%llu", ((highest + 1) *
2743 tc->pool->sectors_per_block) - 1);
2744 else
2745 DMEMIT("-");
2746 break;
2747
2748 case STATUSTYPE_TABLE:
2749 DMEMIT("%s %lu",
2750 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
2751 (unsigned long) tc->dev_id);
Joe Thornber2dd9c252012-03-28 18:41:28 +01002752 if (tc->origin_dev)
2753 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
Joe Thornber991d9fa2011-10-31 20:21:18 +00002754 break;
2755 }
2756 }
2757
2758 return 0;
2759}
2760
2761static int thin_iterate_devices(struct dm_target *ti,
2762 iterate_devices_callout_fn fn, void *data)
2763{
2764 dm_block_t blocks;
2765 struct thin_c *tc = ti->private;
2766
2767 /*
2768 * We can't call dm_pool_get_data_dev_size() since that blocks. So
2769 * we follow a more convoluted path through to the pool's target.
2770 */
2771 if (!tc->pool->ti)
2772 return 0; /* nothing is bound */
2773
2774 blocks = tc->pool->ti->len >> tc->pool->block_shift;
2775 if (blocks)
2776 return fn(ti, tc->pool_dev, 0, tc->pool->sectors_per_block * blocks, data);
2777
2778 return 0;
2779}
2780
2781static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits)
2782{
2783 struct thin_c *tc = ti->private;
Joe Thornber104655f2012-03-28 18:41:28 +01002784 struct pool *pool = tc->pool;
Joe Thornber991d9fa2011-10-31 20:21:18 +00002785
2786 blk_limits_io_min(limits, 0);
Joe Thornber104655f2012-03-28 18:41:28 +01002787 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
2788 set_discard_limits(pool, limits);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002789}
2790
2791static struct target_type thin_target = {
2792 .name = "thin",
Joe Thornber2dd9c252012-03-28 18:41:28 +01002793 .version = {1, 1, 0},
Joe Thornber991d9fa2011-10-31 20:21:18 +00002794 .module = THIS_MODULE,
2795 .ctr = thin_ctr,
2796 .dtr = thin_dtr,
2797 .map = thin_map,
Joe Thornbereb2aa482012-03-28 18:41:28 +01002798 .end_io = thin_endio,
Joe Thornber991d9fa2011-10-31 20:21:18 +00002799 .postsuspend = thin_postsuspend,
2800 .status = thin_status,
2801 .iterate_devices = thin_iterate_devices,
2802 .io_hints = thin_io_hints,
2803};
2804
2805/*----------------------------------------------------------------*/
2806
2807static int __init dm_thin_init(void)
2808{
2809 int r;
2810
2811 pool_table_init();
2812
2813 r = dm_register_target(&thin_target);
2814 if (r)
2815 return r;
2816
2817 r = dm_register_target(&pool_target);
2818 if (r)
Mike Snitzera24c2562012-06-03 00:30:00 +01002819 goto bad_pool_target;
2820
2821 r = -ENOMEM;
2822
2823 _cell_cache = KMEM_CACHE(dm_bio_prison_cell, 0);
2824 if (!_cell_cache)
2825 goto bad_cell_cache;
2826
2827 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
2828 if (!_new_mapping_cache)
2829 goto bad_new_mapping_cache;
2830
2831 _endio_hook_cache = KMEM_CACHE(dm_thin_endio_hook, 0);
2832 if (!_endio_hook_cache)
2833 goto bad_endio_hook_cache;
2834
2835 return 0;
2836
2837bad_endio_hook_cache:
2838 kmem_cache_destroy(_new_mapping_cache);
2839bad_new_mapping_cache:
2840 kmem_cache_destroy(_cell_cache);
2841bad_cell_cache:
2842 dm_unregister_target(&pool_target);
2843bad_pool_target:
2844 dm_unregister_target(&thin_target);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002845
2846 return r;
2847}
2848
2849static void dm_thin_exit(void)
2850{
2851 dm_unregister_target(&thin_target);
2852 dm_unregister_target(&pool_target);
Mike Snitzera24c2562012-06-03 00:30:00 +01002853
2854 kmem_cache_destroy(_cell_cache);
2855 kmem_cache_destroy(_new_mapping_cache);
2856 kmem_cache_destroy(_endio_hook_cache);
Joe Thornber991d9fa2011-10-31 20:21:18 +00002857}
2858
2859module_init(dm_thin_init);
2860module_exit(dm_thin_exit);
2861
Alasdair G Kergon7cab8bf2012-05-12 01:43:19 +01002862MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
Joe Thornber991d9fa2011-10-31 20:21:18 +00002863MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2864MODULE_LICENSE("GPL");