2 * Copyright (C) 2011 Red Hat UK.
4 * This file is released under the GPL.
7 #include "dm-thin-metadata.h"
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>
17 #define DM_MSG_PREFIX "thin"
22 #define ENDIO_HOOK_POOL_SIZE 1024
23 #define DEFERRED_SET_SIZE 64
24 #define MAPPING_POOL_SIZE 1024
25 #define PRISON_CELLS 1024
26 #define COMMIT_PERIOD HZ
29 * The block size of the device holding pool data must be
30 * between 64KB and 1GB.
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)
36 * Device id is restricted to 24 bits.
38 #define MAX_DEV_ID ((1 << 24) - 1)
41 * How do we handle breaking sharing of data blocks?
42 * =================================================
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
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.
54 * Let's say we write to a shared block in what was the origin. The
57 * i) plug io further to this physical block. (see bio_prison code).
59 * ii) quiesce any read io to that shared data block. Obviously
60 * including all devices that share this block. (see deferred_set code)
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).
65 * iv) insert the new mapping into the origin's btree
66 * (process_prepared_mapping). This act of inserting breaks some
67 * 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.
73 * v) unplug io to this physical block, including the io that triggered
74 * the breaking of sharing.
76 * Steps (ii) and (iii) occur in parallel.
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:
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.
86 * - The snap mapping still points to the old block. As it would after
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.
98 /*----------------------------------------------------------------*/
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.
114 struct dm_bio_prison_cell {
115 struct hlist_node list;
116 struct bio_prison *prison;
119 struct bio_list bios;
124 mempool_t *cell_pool;
128 struct hlist_head *cells;
131 static uint32_t calc_nr_buckets(unsigned nr_cells)
136 nr_cells = min(nr_cells, 8192u);
144 static struct kmem_cache *_cell_cache;
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.
150 static struct bio_prison *prison_create(unsigned nr_cells)
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);
161 spin_lock_init(&prison->lock);
162 prison->cell_pool = mempool_create_slab_pool(nr_cells, _cell_cache);
163 if (!prison->cell_pool) {
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);
177 static void prison_destroy(struct bio_prison *prison)
179 mempool_destroy(prison->cell_pool);
183 static uint32_t hash_key(struct bio_prison *prison, struct cell_key *key)
185 const unsigned long BIG_PRIME = 4294967291UL;
186 uint64_t hash = key->block * BIG_PRIME;
188 return (uint32_t) (hash & prison->hash_mask);
191 static int keys_equal(struct cell_key *lhs, struct cell_key *rhs)
193 return (lhs->virtual == rhs->virtual) &&
194 (lhs->dev == rhs->dev) &&
195 (lhs->block == rhs->block);
198 static struct dm_bio_prison_cell *__search_bucket(struct hlist_head *bucket,
199 struct cell_key *key)
201 struct dm_bio_prison_cell *cell;
202 struct hlist_node *tmp;
204 hlist_for_each_entry(cell, tmp, bucket, list)
205 if (keys_equal(&cell->key, key))
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.
215 * Returns 1 if the cell was already held, 0 if @inmate is the new holder.
217 static int bio_detain(struct bio_prison *prison, struct cell_key *key,
218 struct bio *inmate, struct dm_bio_prison_cell **ref)
222 uint32_t hash = hash_key(prison, key);
223 struct dm_bio_prison_cell *cell, *cell2;
225 BUG_ON(hash > prison->nr_buckets);
227 spin_lock_irqsave(&prison->lock, flags);
229 cell = __search_bucket(prison->cells + hash, key);
231 bio_list_add(&cell->bios, inmate);
236 * Allocate a new cell
238 spin_unlock_irqrestore(&prison->lock, flags);
239 cell2 = mempool_alloc(prison->cell_pool, GFP_NOIO);
240 spin_lock_irqsave(&prison->lock, flags);
243 * We've been unlocked, so we have to double check that
244 * nobody else has inserted this cell in the meantime.
246 cell = __search_bucket(prison->cells + hash, key);
248 mempool_free(cell2, prison->cell_pool);
249 bio_list_add(&cell->bios, inmate);
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);
267 spin_unlock_irqrestore(&prison->lock, flags);
275 * @inmates must have been initialised prior to this call
277 static void __cell_release(struct dm_bio_prison_cell *cell, struct bio_list *inmates)
279 struct bio_prison *prison = cell->prison;
281 hlist_del(&cell->list);
284 bio_list_add(inmates, cell->holder);
285 bio_list_merge(inmates, &cell->bios);
288 mempool_free(cell, prison->cell_pool);
291 static void cell_release(struct dm_bio_prison_cell *cell, struct bio_list *bios)
294 struct bio_prison *prison = cell->prison;
296 spin_lock_irqsave(&prison->lock, flags);
297 __cell_release(cell, bios);
298 spin_unlock_irqrestore(&prison->lock, flags);
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
307 static void __cell_release_singleton(struct dm_bio_prison_cell *cell, struct bio *bio)
309 BUG_ON(cell->holder != bio);
310 BUG_ON(!bio_list_empty(&cell->bios));
312 __cell_release(cell, NULL);
315 static void cell_release_singleton(struct dm_bio_prison_cell *cell, struct bio *bio)
318 struct bio_prison *prison = cell->prison;
320 spin_lock_irqsave(&prison->lock, flags);
321 __cell_release_singleton(cell, bio);
322 spin_unlock_irqrestore(&prison->lock, flags);
326 * Sometimes we don't want the holder, just the additional bios.
328 static void __cell_release_no_holder(struct dm_bio_prison_cell *cell,
329 struct bio_list *inmates)
331 struct bio_prison *prison = cell->prison;
333 hlist_del(&cell->list);
334 bio_list_merge(inmates, &cell->bios);
336 mempool_free(cell, prison->cell_pool);
339 static void cell_release_no_holder(struct dm_bio_prison_cell *cell,
340 struct bio_list *inmates)
343 struct bio_prison *prison = cell->prison;
345 spin_lock_irqsave(&prison->lock, flags);
346 __cell_release_no_holder(cell, inmates);
347 spin_unlock_irqrestore(&prison->lock, flags);
350 static void cell_error(struct dm_bio_prison_cell *cell)
352 struct bio_prison *prison = cell->prison;
353 struct bio_list bios;
357 bio_list_init(&bios);
359 spin_lock_irqsave(&prison->lock, flags);
360 __cell_release(cell, &bios);
361 spin_unlock_irqrestore(&prison->lock, flags);
363 while ((bio = bio_list_pop(&bios)))
367 /*----------------------------------------------------------------*/
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.
377 struct deferred_entry {
378 struct deferred_set *ds;
380 struct list_head work_items;
383 struct deferred_set {
385 unsigned current_entry;
387 struct deferred_entry entries[DEFERRED_SET_SIZE];
390 static void ds_init(struct deferred_set *ds)
394 spin_lock_init(&ds->lock);
395 ds->current_entry = 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);
404 static struct deferred_entry *ds_inc(struct deferred_set *ds)
407 struct deferred_entry *entry;
409 spin_lock_irqsave(&ds->lock, flags);
410 entry = ds->entries + ds->current_entry;
412 spin_unlock_irqrestore(&ds->lock, flags);
417 static unsigned ds_next(unsigned index)
419 return (index + 1) % DEFERRED_SET_SIZE;
422 static void __sweep(struct deferred_set *ds, struct list_head *head)
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);
430 if ((ds->sweeper == ds->current_entry) && !ds->entries[ds->sweeper].count)
431 list_splice_init(&ds->entries[ds->sweeper].work_items, head);
434 static void ds_dec(struct deferred_entry *entry, struct list_head *head)
438 spin_lock_irqsave(&entry->ds->lock, flags);
439 BUG_ON(!entry->count);
441 __sweep(entry->ds, head);
442 spin_unlock_irqrestore(&entry->ds->lock, flags);
446 * Returns 1 if deferred or 0 if no pending items to delay job.
448 static int ds_add_work(struct deferred_set *ds, struct list_head *work)
454 spin_lock_irqsave(&ds->lock, flags);
455 if ((ds->sweeper == ds->current_entry) &&
456 !ds->entries[ds->current_entry].count)
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;
464 spin_unlock_irqrestore(&ds->lock, flags);
469 /*----------------------------------------------------------------*/
474 static void build_data_key(struct dm_thin_device *td,
475 dm_block_t b, struct cell_key *key)
478 key->dev = dm_thin_dev_id(td);
482 static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
483 struct cell_key *key)
486 key->dev = dm_thin_dev_id(td);
490 /*----------------------------------------------------------------*/
493 * A pool device ties together a metadata device and a data device. It
494 * also provides the interface for creating and destroying internal
497 struct dm_thin_new_mapping;
499 struct pool_features {
500 unsigned zero_new_blocks:1;
501 unsigned discard_enabled:1;
502 unsigned discard_passdown:1;
506 struct list_head list;
507 struct dm_target *ti; /* Only set if a pool target is bound */
509 struct mapped_device *pool_md;
510 struct block_device *md_dev;
511 struct dm_pool_metadata *pmd;
513 dm_block_t low_water_blocks;
514 uint32_t sectors_per_block;
515 int sectors_per_block_shift;
517 struct pool_features pf;
518 unsigned low_water_triggered:1; /* A dm event has been sent */
519 unsigned no_free_space:1; /* A -ENOSPC warning has been issued */
521 struct bio_prison *prison;
522 struct dm_kcopyd_client *copier;
524 struct workqueue_struct *wq;
525 struct work_struct worker;
526 struct delayed_work waker;
528 unsigned long last_commit_jiffies;
532 struct bio_list deferred_bios;
533 struct bio_list deferred_flush_bios;
534 struct list_head prepared_mappings;
535 struct list_head prepared_discards;
537 struct bio_list retry_on_resume_list;
539 struct deferred_set shared_read_ds;
540 struct deferred_set all_io_ds;
542 struct dm_thin_new_mapping *next_mapping;
543 mempool_t *mapping_pool;
544 mempool_t *endio_hook_pool;
548 * Target context for a pool.
551 struct dm_target *ti;
553 struct dm_dev *data_dev;
554 struct dm_dev *metadata_dev;
555 struct dm_target_callbacks callbacks;
557 dm_block_t low_water_blocks;
558 struct pool_features pf;
562 * Target context for a thin.
565 struct dm_dev *pool_dev;
566 struct dm_dev *origin_dev;
570 struct dm_thin_device *td;
573 /*----------------------------------------------------------------*/
576 * A global list of pools that uses a struct mapped_device as a key.
578 static struct dm_thin_pool_table {
580 struct list_head pools;
581 } dm_thin_pool_table;
583 static void pool_table_init(void)
585 mutex_init(&dm_thin_pool_table.mutex);
586 INIT_LIST_HEAD(&dm_thin_pool_table.pools);
589 static void __pool_table_insert(struct pool *pool)
591 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
592 list_add(&pool->list, &dm_thin_pool_table.pools);
595 static void __pool_table_remove(struct pool *pool)
597 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
598 list_del(&pool->list);
601 static struct pool *__pool_table_lookup(struct mapped_device *md)
603 struct pool *pool = NULL, *tmp;
605 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
607 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
608 if (tmp->pool_md == md) {
617 static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
619 struct pool *pool = NULL, *tmp;
621 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
623 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
624 if (tmp->md_dev == md_dev) {
633 /*----------------------------------------------------------------*/
635 struct dm_thin_endio_hook {
637 struct deferred_entry *shared_read_entry;
638 struct deferred_entry *all_io_entry;
639 struct dm_thin_new_mapping *overwrite_mapping;
642 static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master)
645 struct bio_list bios;
647 bio_list_init(&bios);
648 bio_list_merge(&bios, master);
649 bio_list_init(master);
651 while ((bio = bio_list_pop(&bios))) {
652 struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
655 bio_endio(bio, DM_ENDIO_REQUEUE);
657 bio_list_add(master, bio);
661 static void requeue_io(struct thin_c *tc)
663 struct pool *pool = tc->pool;
666 spin_lock_irqsave(&pool->lock, flags);
667 __requeue_bio_list(tc, &pool->deferred_bios);
668 __requeue_bio_list(tc, &pool->retry_on_resume_list);
669 spin_unlock_irqrestore(&pool->lock, flags);
673 * This section of code contains the logic for processing a thin device's IO.
674 * Much of the code depends on pool object resources (lists, workqueues, etc)
675 * but most is exclusively called from the thin target rather than the thin-pool
679 static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
681 sector_t block_nr = bio->bi_sector;
683 if (tc->pool->sectors_per_block_shift < 0)
684 (void) sector_div(block_nr, tc->pool->sectors_per_block);
686 block_nr >>= tc->pool->sectors_per_block_shift;
691 static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
693 struct pool *pool = tc->pool;
694 sector_t bi_sector = bio->bi_sector;
696 bio->bi_bdev = tc->pool_dev->bdev;
697 if (tc->pool->sectors_per_block_shift < 0)
698 bio->bi_sector = (block * pool->sectors_per_block) +
699 sector_div(bi_sector, pool->sectors_per_block);
701 bio->bi_sector = (block << pool->sectors_per_block_shift) |
702 (bi_sector & (pool->sectors_per_block - 1));
705 static void remap_to_origin(struct thin_c *tc, struct bio *bio)
707 bio->bi_bdev = tc->origin_dev->bdev;
710 static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
712 return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) &&
713 dm_thin_changed_this_transaction(tc->td);
716 static void issue(struct thin_c *tc, struct bio *bio)
718 struct pool *pool = tc->pool;
722 * Batch together any FUA/FLUSH bios we find and then issue
723 * a single commit for them in process_deferred_bios().
725 if (bio_triggers_commit(tc, bio)) {
726 spin_lock_irqsave(&pool->lock, flags);
727 bio_list_add(&pool->deferred_flush_bios, bio);
728 spin_unlock_irqrestore(&pool->lock, flags);
730 generic_make_request(bio);
733 static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
735 remap_to_origin(tc, bio);
739 static void remap_and_issue(struct thin_c *tc, struct bio *bio,
742 remap(tc, bio, block);
747 * wake_worker() is used when new work is queued and when pool_resume is
748 * ready to continue deferred IO processing.
750 static void wake_worker(struct pool *pool)
752 queue_work(pool->wq, &pool->worker);
755 /*----------------------------------------------------------------*/
758 * Bio endio functions.
760 struct dm_thin_new_mapping {
761 struct list_head list;
765 unsigned pass_discard:1;
768 dm_block_t virt_block;
769 dm_block_t data_block;
770 struct dm_bio_prison_cell *cell, *cell2;
774 * If the bio covers the whole area of a block then we can avoid
775 * zeroing or copying. Instead this bio is hooked. The bio will
776 * still be in the cell, so care has to be taken to avoid issuing
780 bio_end_io_t *saved_bi_end_io;
783 static void __maybe_add_mapping(struct dm_thin_new_mapping *m)
785 struct pool *pool = m->tc->pool;
787 if (m->quiesced && m->prepared) {
788 list_add(&m->list, &pool->prepared_mappings);
793 static void copy_complete(int read_err, unsigned long write_err, void *context)
796 struct dm_thin_new_mapping *m = context;
797 struct pool *pool = m->tc->pool;
799 m->err = read_err || write_err ? -EIO : 0;
801 spin_lock_irqsave(&pool->lock, flags);
803 __maybe_add_mapping(m);
804 spin_unlock_irqrestore(&pool->lock, flags);
807 static void overwrite_endio(struct bio *bio, int err)
810 struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
811 struct dm_thin_new_mapping *m = h->overwrite_mapping;
812 struct pool *pool = m->tc->pool;
816 spin_lock_irqsave(&pool->lock, flags);
818 __maybe_add_mapping(m);
819 spin_unlock_irqrestore(&pool->lock, flags);
822 /*----------------------------------------------------------------*/
829 * Prepared mapping jobs.
833 * This sends the bios in the cell back to the deferred_bios list.
835 static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell,
836 dm_block_t data_block)
838 struct pool *pool = tc->pool;
841 spin_lock_irqsave(&pool->lock, flags);
842 cell_release(cell, &pool->deferred_bios);
843 spin_unlock_irqrestore(&tc->pool->lock, flags);
849 * Same as cell_defer above, except it omits one particular detainee,
850 * a write bio that covers the block and has already been processed.
852 static void cell_defer_except(struct thin_c *tc, struct dm_bio_prison_cell *cell)
854 struct bio_list bios;
855 struct pool *pool = tc->pool;
858 bio_list_init(&bios);
860 spin_lock_irqsave(&pool->lock, flags);
861 cell_release_no_holder(cell, &pool->deferred_bios);
862 spin_unlock_irqrestore(&pool->lock, flags);
867 static void process_prepared_mapping(struct dm_thin_new_mapping *m)
869 struct thin_c *tc = m->tc;
875 bio->bi_end_io = m->saved_bi_end_io;
883 * Commit the prepared block into the mapping btree.
884 * Any I/O for this block arriving after this point will get
885 * remapped to it directly.
887 r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block);
889 DMERR("dm_thin_insert_block() failed");
895 * Release any bios held while the block was being provisioned.
896 * If we are processing a write bio that completely covers the block,
897 * we already processed it so can ignore it now when processing
898 * the bios in the cell.
901 cell_defer_except(tc, m->cell);
904 cell_defer(tc, m->cell, m->data_block);
908 mempool_free(m, tc->pool->mapping_pool);
911 static void process_prepared_discard(struct dm_thin_new_mapping *m)
914 struct thin_c *tc = m->tc;
916 r = dm_thin_remove_block(tc->td, m->virt_block);
918 DMERR("dm_thin_remove_block() failed");
921 * Pass the discard down to the underlying device?
924 remap_and_issue(tc, m->bio, m->data_block);
926 bio_endio(m->bio, 0);
928 cell_defer_except(tc, m->cell);
929 cell_defer_except(tc, m->cell2);
930 mempool_free(m, tc->pool->mapping_pool);
933 static void process_prepared(struct pool *pool, struct list_head *head,
934 void (*fn)(struct dm_thin_new_mapping *))
937 struct list_head maps;
938 struct dm_thin_new_mapping *m, *tmp;
940 INIT_LIST_HEAD(&maps);
941 spin_lock_irqsave(&pool->lock, flags);
942 list_splice_init(head, &maps);
943 spin_unlock_irqrestore(&pool->lock, flags);
945 list_for_each_entry_safe(m, tmp, &maps, list)
952 static int io_overlaps_block(struct pool *pool, struct bio *bio)
954 return bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT);
957 static int io_overwrites_block(struct pool *pool, struct bio *bio)
959 return (bio_data_dir(bio) == WRITE) &&
960 io_overlaps_block(pool, bio);
963 static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
966 *save = bio->bi_end_io;
970 static int ensure_next_mapping(struct pool *pool)
972 if (pool->next_mapping)
975 pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);
977 return pool->next_mapping ? 0 : -ENOMEM;
980 static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
982 struct dm_thin_new_mapping *r = pool->next_mapping;
984 BUG_ON(!pool->next_mapping);
986 pool->next_mapping = NULL;
991 static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
992 struct dm_dev *origin, dm_block_t data_origin,
993 dm_block_t data_dest,
994 struct dm_bio_prison_cell *cell, struct bio *bio)
997 struct pool *pool = tc->pool;
998 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1000 INIT_LIST_HEAD(&m->list);
1004 m->virt_block = virt_block;
1005 m->data_block = data_dest;
1010 if (!ds_add_work(&pool->shared_read_ds, &m->list))
1014 * IO to pool_dev remaps to the pool target's data_dev.
1016 * If the whole block of data is being overwritten, we can issue the
1017 * bio immediately. Otherwise we use kcopyd to clone the data first.
1019 if (io_overwrites_block(pool, bio)) {
1020 struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
1022 h->overwrite_mapping = m;
1024 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
1025 remap_and_issue(tc, bio, data_dest);
1027 struct dm_io_region from, to;
1029 from.bdev = origin->bdev;
1030 from.sector = data_origin * pool->sectors_per_block;
1031 from.count = pool->sectors_per_block;
1033 to.bdev = tc->pool_dev->bdev;
1034 to.sector = data_dest * pool->sectors_per_block;
1035 to.count = pool->sectors_per_block;
1037 r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
1038 0, copy_complete, m);
1040 mempool_free(m, pool->mapping_pool);
1041 DMERR("dm_kcopyd_copy() failed");
1047 static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
1048 dm_block_t data_origin, dm_block_t data_dest,
1049 struct dm_bio_prison_cell *cell, struct bio *bio)
1051 schedule_copy(tc, virt_block, tc->pool_dev,
1052 data_origin, data_dest, cell, bio);
1055 static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
1056 dm_block_t data_dest,
1057 struct dm_bio_prison_cell *cell, struct bio *bio)
1059 schedule_copy(tc, virt_block, tc->origin_dev,
1060 virt_block, data_dest, cell, bio);
1063 static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
1064 dm_block_t data_block, struct dm_bio_prison_cell *cell,
1067 struct pool *pool = tc->pool;
1068 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1070 INIT_LIST_HEAD(&m->list);
1074 m->virt_block = virt_block;
1075 m->data_block = data_block;
1081 * If the whole block of data is being overwritten or we are not
1082 * zeroing pre-existing data, we can issue the bio immediately.
1083 * Otherwise we use kcopyd to zero the data first.
1085 if (!pool->pf.zero_new_blocks)
1086 process_prepared_mapping(m);
1088 else if (io_overwrites_block(pool, bio)) {
1089 struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
1091 h->overwrite_mapping = m;
1093 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
1094 remap_and_issue(tc, bio, data_block);
1097 struct dm_io_region to;
1099 to.bdev = tc->pool_dev->bdev;
1100 to.sector = data_block * pool->sectors_per_block;
1101 to.count = pool->sectors_per_block;
1103 r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m);
1105 mempool_free(m, pool->mapping_pool);
1106 DMERR("dm_kcopyd_zero() failed");
1112 static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
1115 dm_block_t free_blocks;
1116 unsigned long flags;
1117 struct pool *pool = tc->pool;
1119 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1123 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
1124 DMWARN("%s: reached low water mark, sending event.",
1125 dm_device_name(pool->pool_md));
1126 spin_lock_irqsave(&pool->lock, flags);
1127 pool->low_water_triggered = 1;
1128 spin_unlock_irqrestore(&pool->lock, flags);
1129 dm_table_event(pool->ti->table);
1133 if (pool->no_free_space)
1137 * Try to commit to see if that will free up some
1140 r = dm_pool_commit_metadata(pool->pmd);
1142 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
1147 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1152 * If we still have no space we set a flag to avoid
1153 * doing all this checking and return -ENOSPC.
1156 DMWARN("%s: no free space available.",
1157 dm_device_name(pool->pool_md));
1158 spin_lock_irqsave(&pool->lock, flags);
1159 pool->no_free_space = 1;
1160 spin_unlock_irqrestore(&pool->lock, flags);
1166 r = dm_pool_alloc_data_block(pool->pmd, result);
1174 * If we have run out of space, queue bios until the device is
1175 * resumed, presumably after having been reloaded with more space.
1177 static void retry_on_resume(struct bio *bio)
1179 struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
1180 struct thin_c *tc = h->tc;
1181 struct pool *pool = tc->pool;
1182 unsigned long flags;
1184 spin_lock_irqsave(&pool->lock, flags);
1185 bio_list_add(&pool->retry_on_resume_list, bio);
1186 spin_unlock_irqrestore(&pool->lock, flags);
1189 static void no_space(struct dm_bio_prison_cell *cell)
1192 struct bio_list bios;
1194 bio_list_init(&bios);
1195 cell_release(cell, &bios);
1197 while ((bio = bio_list_pop(&bios)))
1198 retry_on_resume(bio);
1201 static void process_discard(struct thin_c *tc, struct bio *bio)
1204 unsigned long flags;
1205 struct pool *pool = tc->pool;
1206 struct dm_bio_prison_cell *cell, *cell2;
1207 struct cell_key key, key2;
1208 dm_block_t block = get_bio_block(tc, bio);
1209 struct dm_thin_lookup_result lookup_result;
1210 struct dm_thin_new_mapping *m;
1212 build_virtual_key(tc->td, block, &key);
1213 if (bio_detain(tc->pool->prison, &key, bio, &cell))
1216 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1220 * Check nobody is fiddling with this pool block. This can
1221 * happen if someone's in the process of breaking sharing
1224 build_data_key(tc->td, lookup_result.block, &key2);
1225 if (bio_detain(tc->pool->prison, &key2, bio, &cell2)) {
1226 cell_release_singleton(cell, bio);
1230 if (io_overlaps_block(pool, bio)) {
1232 * IO may still be going to the destination block. We must
1233 * quiesce before we can do the removal.
1235 m = get_next_mapping(pool);
1237 m->pass_discard = (!lookup_result.shared) && pool->pf.discard_passdown;
1238 m->virt_block = block;
1239 m->data_block = lookup_result.block;
1245 if (!ds_add_work(&pool->all_io_ds, &m->list)) {
1246 spin_lock_irqsave(&pool->lock, flags);
1247 list_add(&m->list, &pool->prepared_discards);
1248 spin_unlock_irqrestore(&pool->lock, flags);
1253 * The DM core makes sure that the discard doesn't span
1254 * a block boundary. So we submit the discard of a
1255 * partial block appropriately.
1257 cell_release_singleton(cell, bio);
1258 cell_release_singleton(cell2, bio);
1259 if ((!lookup_result.shared) && pool->pf.discard_passdown)
1260 remap_and_issue(tc, bio, lookup_result.block);
1268 * It isn't provisioned, just forget it.
1270 cell_release_singleton(cell, bio);
1275 DMERR("discard: find block unexpectedly returned %d", r);
1276 cell_release_singleton(cell, bio);
1282 static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
1283 struct cell_key *key,
1284 struct dm_thin_lookup_result *lookup_result,
1285 struct dm_bio_prison_cell *cell)
1288 dm_block_t data_block;
1290 r = alloc_data_block(tc, &data_block);
1293 schedule_internal_copy(tc, block, lookup_result->block,
1294 data_block, cell, bio);
1302 DMERR("%s: alloc_data_block() failed, error = %d", __func__, r);
1308 static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1310 struct dm_thin_lookup_result *lookup_result)
1312 struct dm_bio_prison_cell *cell;
1313 struct pool *pool = tc->pool;
1314 struct cell_key key;
1317 * If cell is already occupied, then sharing is already in the process
1318 * of being broken so we have nothing further to do here.
1320 build_data_key(tc->td, lookup_result->block, &key);
1321 if (bio_detain(pool->prison, &key, bio, &cell))
1324 if (bio_data_dir(bio) == WRITE && bio->bi_size)
1325 break_sharing(tc, bio, block, &key, lookup_result, cell);
1327 struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
1329 h->shared_read_entry = ds_inc(&pool->shared_read_ds);
1331 cell_release_singleton(cell, bio);
1332 remap_and_issue(tc, bio, lookup_result->block);
1336 static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
1337 struct dm_bio_prison_cell *cell)
1340 dm_block_t data_block;
1343 * Remap empty bios (flushes) immediately, without provisioning.
1345 if (!bio->bi_size) {
1346 cell_release_singleton(cell, bio);
1347 remap_and_issue(tc, bio, 0);
1352 * Fill read bios with zeroes and complete them immediately.
1354 if (bio_data_dir(bio) == READ) {
1356 cell_release_singleton(cell, bio);
1361 r = alloc_data_block(tc, &data_block);
1365 schedule_external_copy(tc, block, data_block, cell, bio);
1367 schedule_zero(tc, block, data_block, cell, bio);
1375 DMERR("%s: alloc_data_block() failed, error = %d", __func__, r);
1381 static void process_bio(struct thin_c *tc, struct bio *bio)
1384 dm_block_t block = get_bio_block(tc, bio);
1385 struct dm_bio_prison_cell *cell;
1386 struct cell_key key;
1387 struct dm_thin_lookup_result lookup_result;
1390 * If cell is already occupied, then the block is already
1391 * being provisioned so we have nothing further to do here.
1393 build_virtual_key(tc->td, block, &key);
1394 if (bio_detain(tc->pool->prison, &key, bio, &cell))
1397 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1401 * We can release this cell now. This thread is the only
1402 * one that puts bios into a cell, and we know there were
1403 * no preceding bios.
1406 * TODO: this will probably have to change when discard goes
1409 cell_release_singleton(cell, bio);
1411 if (lookup_result.shared)
1412 process_shared_bio(tc, bio, block, &lookup_result);
1414 remap_and_issue(tc, bio, lookup_result.block);
1418 if (bio_data_dir(bio) == READ && tc->origin_dev) {
1419 cell_release_singleton(cell, bio);
1420 remap_to_origin_and_issue(tc, bio);
1422 provision_block(tc, bio, block, cell);
1426 DMERR("dm_thin_find_block() failed, error = %d", r);
1427 cell_release_singleton(cell, bio);
1433 static int need_commit_due_to_time(struct pool *pool)
1435 return jiffies < pool->last_commit_jiffies ||
1436 jiffies > pool->last_commit_jiffies + COMMIT_PERIOD;
1439 static void process_deferred_bios(struct pool *pool)
1441 unsigned long flags;
1443 struct bio_list bios;
1446 bio_list_init(&bios);
1448 spin_lock_irqsave(&pool->lock, flags);
1449 bio_list_merge(&bios, &pool->deferred_bios);
1450 bio_list_init(&pool->deferred_bios);
1451 spin_unlock_irqrestore(&pool->lock, flags);
1453 while ((bio = bio_list_pop(&bios))) {
1454 struct dm_thin_endio_hook *h = dm_get_mapinfo(bio)->ptr;
1455 struct thin_c *tc = h->tc;
1458 * If we've got no free new_mapping structs, and processing
1459 * this bio might require one, we pause until there are some
1460 * prepared mappings to process.
1462 if (ensure_next_mapping(pool)) {
1463 spin_lock_irqsave(&pool->lock, flags);
1464 bio_list_merge(&pool->deferred_bios, &bios);
1465 spin_unlock_irqrestore(&pool->lock, flags);
1470 if (bio->bi_rw & REQ_DISCARD)
1471 process_discard(tc, bio);
1473 process_bio(tc, bio);
1477 * If there are any deferred flush bios, we must commit
1478 * the metadata before issuing them.
1480 bio_list_init(&bios);
1481 spin_lock_irqsave(&pool->lock, flags);
1482 bio_list_merge(&bios, &pool->deferred_flush_bios);
1483 bio_list_init(&pool->deferred_flush_bios);
1484 spin_unlock_irqrestore(&pool->lock, flags);
1486 if (bio_list_empty(&bios) && !need_commit_due_to_time(pool))
1489 r = dm_pool_commit_metadata(pool->pmd);
1491 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
1493 while ((bio = bio_list_pop(&bios)))
1497 pool->last_commit_jiffies = jiffies;
1499 while ((bio = bio_list_pop(&bios)))
1500 generic_make_request(bio);
1503 static void do_worker(struct work_struct *ws)
1505 struct pool *pool = container_of(ws, struct pool, worker);
1507 process_prepared(pool, &pool->prepared_mappings, process_prepared_mapping);
1508 process_prepared(pool, &pool->prepared_discards, process_prepared_discard);
1509 process_deferred_bios(pool);
1513 * We want to commit periodically so that not too much
1514 * unwritten data builds up.
1516 static void do_waker(struct work_struct *ws)
1518 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
1520 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
1523 /*----------------------------------------------------------------*/
1526 * Mapping functions.
1530 * Called only while mapping a thin bio to hand it over to the workqueue.
1532 static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
1534 unsigned long flags;
1535 struct pool *pool = tc->pool;
1537 spin_lock_irqsave(&pool->lock, flags);
1538 bio_list_add(&pool->deferred_bios, bio);
1539 spin_unlock_irqrestore(&pool->lock, flags);
1544 static struct dm_thin_endio_hook *thin_hook_bio(struct thin_c *tc, struct bio *bio)
1546 struct pool *pool = tc->pool;
1547 struct dm_thin_endio_hook *h = mempool_alloc(pool->endio_hook_pool, GFP_NOIO);
1550 h->shared_read_entry = NULL;
1551 h->all_io_entry = bio->bi_rw & REQ_DISCARD ? NULL : ds_inc(&pool->all_io_ds);
1552 h->overwrite_mapping = NULL;
1558 * Non-blocking function called from the thin target's map function.
1560 static int thin_bio_map(struct dm_target *ti, struct bio *bio,
1561 union map_info *map_context)
1564 struct thin_c *tc = ti->private;
1565 dm_block_t block = get_bio_block(tc, bio);
1566 struct dm_thin_device *td = tc->td;
1567 struct dm_thin_lookup_result result;
1569 map_context->ptr = thin_hook_bio(tc, bio);
1570 if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) {
1571 thin_defer_bio(tc, bio);
1572 return DM_MAPIO_SUBMITTED;
1575 r = dm_thin_find_block(td, block, 0, &result);
1578 * Note that we defer readahead too.
1582 if (unlikely(result.shared)) {
1584 * We have a race condition here between the
1585 * result.shared value returned by the lookup and
1586 * snapshot creation, which may cause new
1589 * To avoid this always quiesce the origin before
1590 * taking the snap. You want to do this anyway to
1591 * ensure a consistent application view
1594 * More distant ancestors are irrelevant. The
1595 * shared flag will be set in their case.
1597 thin_defer_bio(tc, bio);
1598 r = DM_MAPIO_SUBMITTED;
1600 remap(tc, bio, result.block);
1601 r = DM_MAPIO_REMAPPED;
1607 * In future, the failed dm_thin_find_block above could
1608 * provide the hint to load the metadata into cache.
1611 thin_defer_bio(tc, bio);
1612 r = DM_MAPIO_SUBMITTED;
1619 static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
1622 unsigned long flags;
1623 struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
1625 spin_lock_irqsave(&pt->pool->lock, flags);
1626 r = !bio_list_empty(&pt->pool->retry_on_resume_list);
1627 spin_unlock_irqrestore(&pt->pool->lock, flags);
1630 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1631 r = bdi_congested(&q->backing_dev_info, bdi_bits);
1637 static void __requeue_bios(struct pool *pool)
1639 bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list);
1640 bio_list_init(&pool->retry_on_resume_list);
1643 /*----------------------------------------------------------------
1644 * Binding of control targets to a pool object
1645 *--------------------------------------------------------------*/
1646 static int bind_control_target(struct pool *pool, struct dm_target *ti)
1648 struct pool_c *pt = ti->private;
1651 pool->low_water_blocks = pt->low_water_blocks;
1655 * If discard_passdown was enabled verify that the data device
1656 * supports discards. Disable discard_passdown if not; otherwise
1657 * -EOPNOTSUPP will be returned.
1659 if (pt->pf.discard_passdown) {
1660 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1661 if (!q || !blk_queue_discard(q)) {
1662 char buf[BDEVNAME_SIZE];
1663 DMWARN("Discard unsupported by data device (%s): Disabling discard passdown.",
1664 bdevname(pt->data_dev->bdev, buf));
1665 pool->pf.discard_passdown = 0;
1672 static void unbind_control_target(struct pool *pool, struct dm_target *ti)
1678 /*----------------------------------------------------------------
1680 *--------------------------------------------------------------*/
1681 /* Initialize pool features. */
1682 static void pool_features_init(struct pool_features *pf)
1684 pf->zero_new_blocks = 1;
1685 pf->discard_enabled = 1;
1686 pf->discard_passdown = 1;
1689 static void __pool_destroy(struct pool *pool)
1691 __pool_table_remove(pool);
1693 if (dm_pool_metadata_close(pool->pmd) < 0)
1694 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1696 prison_destroy(pool->prison);
1697 dm_kcopyd_client_destroy(pool->copier);
1700 destroy_workqueue(pool->wq);
1702 if (pool->next_mapping)
1703 mempool_free(pool->next_mapping, pool->mapping_pool);
1704 mempool_destroy(pool->mapping_pool);
1705 mempool_destroy(pool->endio_hook_pool);
1709 static struct kmem_cache *_new_mapping_cache;
1710 static struct kmem_cache *_endio_hook_cache;
1712 static struct pool *pool_create(struct mapped_device *pool_md,
1713 struct block_device *metadata_dev,
1714 unsigned long block_size, char **error)
1719 struct dm_pool_metadata *pmd;
1721 pmd = dm_pool_metadata_open(metadata_dev, block_size, true);
1723 *error = "Error creating metadata object";
1724 return (struct pool *)pmd;
1727 pool = kmalloc(sizeof(*pool), GFP_KERNEL);
1729 *error = "Error allocating memory for pool";
1730 err_p = ERR_PTR(-ENOMEM);
1735 pool->sectors_per_block = block_size;
1736 if (block_size & (block_size - 1))
1737 pool->sectors_per_block_shift = -1;
1739 pool->sectors_per_block_shift = __ffs(block_size);
1740 pool->low_water_blocks = 0;
1741 pool_features_init(&pool->pf);
1742 pool->prison = prison_create(PRISON_CELLS);
1743 if (!pool->prison) {
1744 *error = "Error creating pool's bio prison";
1745 err_p = ERR_PTR(-ENOMEM);
1749 pool->copier = dm_kcopyd_client_create();
1750 if (IS_ERR(pool->copier)) {
1751 r = PTR_ERR(pool->copier);
1752 *error = "Error creating pool's kcopyd client";
1754 goto bad_kcopyd_client;
1758 * Create singlethreaded workqueue that will service all devices
1759 * that use this metadata.
1761 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
1763 *error = "Error creating pool's workqueue";
1764 err_p = ERR_PTR(-ENOMEM);
1768 INIT_WORK(&pool->worker, do_worker);
1769 INIT_DELAYED_WORK(&pool->waker, do_waker);
1770 spin_lock_init(&pool->lock);
1771 bio_list_init(&pool->deferred_bios);
1772 bio_list_init(&pool->deferred_flush_bios);
1773 INIT_LIST_HEAD(&pool->prepared_mappings);
1774 INIT_LIST_HEAD(&pool->prepared_discards);
1775 pool->low_water_triggered = 0;
1776 pool->no_free_space = 0;
1777 bio_list_init(&pool->retry_on_resume_list);
1778 ds_init(&pool->shared_read_ds);
1779 ds_init(&pool->all_io_ds);
1781 pool->next_mapping = NULL;
1782 pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE,
1783 _new_mapping_cache);
1784 if (!pool->mapping_pool) {
1785 *error = "Error creating pool's mapping mempool";
1786 err_p = ERR_PTR(-ENOMEM);
1787 goto bad_mapping_pool;
1790 pool->endio_hook_pool = mempool_create_slab_pool(ENDIO_HOOK_POOL_SIZE,
1792 if (!pool->endio_hook_pool) {
1793 *error = "Error creating pool's endio_hook mempool";
1794 err_p = ERR_PTR(-ENOMEM);
1795 goto bad_endio_hook_pool;
1797 pool->ref_count = 1;
1798 pool->last_commit_jiffies = jiffies;
1799 pool->pool_md = pool_md;
1800 pool->md_dev = metadata_dev;
1801 __pool_table_insert(pool);
1805 bad_endio_hook_pool:
1806 mempool_destroy(pool->mapping_pool);
1808 destroy_workqueue(pool->wq);
1810 dm_kcopyd_client_destroy(pool->copier);
1812 prison_destroy(pool->prison);
1816 if (dm_pool_metadata_close(pmd))
1817 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1822 static void __pool_inc(struct pool *pool)
1824 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1828 static void __pool_dec(struct pool *pool)
1830 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1831 BUG_ON(!pool->ref_count);
1832 if (!--pool->ref_count)
1833 __pool_destroy(pool);
1836 static struct pool *__pool_find(struct mapped_device *pool_md,
1837 struct block_device *metadata_dev,
1838 unsigned long block_size, char **error,
1841 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
1844 if (pool->pool_md != pool_md) {
1845 *error = "metadata device already in use by a pool";
1846 return ERR_PTR(-EBUSY);
1851 pool = __pool_table_lookup(pool_md);
1853 if (pool->md_dev != metadata_dev) {
1854 *error = "different pool cannot replace a pool";
1855 return ERR_PTR(-EINVAL);
1860 pool = pool_create(pool_md, metadata_dev, block_size, error);
1868 /*----------------------------------------------------------------
1869 * Pool target methods
1870 *--------------------------------------------------------------*/
1871 static void pool_dtr(struct dm_target *ti)
1873 struct pool_c *pt = ti->private;
1875 mutex_lock(&dm_thin_pool_table.mutex);
1877 unbind_control_target(pt->pool, ti);
1878 __pool_dec(pt->pool);
1879 dm_put_device(ti, pt->metadata_dev);
1880 dm_put_device(ti, pt->data_dev);
1883 mutex_unlock(&dm_thin_pool_table.mutex);
1886 static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
1887 struct dm_target *ti)
1891 const char *arg_name;
1893 static struct dm_arg _args[] = {
1894 {0, 3, "Invalid number of pool feature arguments"},
1898 * No feature arguments supplied.
1903 r = dm_read_arg_group(_args, as, &argc, &ti->error);
1907 while (argc && !r) {
1908 arg_name = dm_shift_arg(as);
1911 if (!strcasecmp(arg_name, "skip_block_zeroing")) {
1912 pf->zero_new_blocks = 0;
1914 } else if (!strcasecmp(arg_name, "ignore_discard")) {
1915 pf->discard_enabled = 0;
1917 } else if (!strcasecmp(arg_name, "no_discard_passdown")) {
1918 pf->discard_passdown = 0;
1922 ti->error = "Unrecognised pool feature requested";
1930 * thin-pool <metadata dev> <data dev>
1931 * <data block size (sectors)>
1932 * <low water mark (blocks)>
1933 * [<#feature args> [<arg>]*]
1935 * Optional feature arguments are:
1936 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
1937 * ignore_discard: disable discard
1938 * no_discard_passdown: don't pass discards down to the data device
1940 static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
1942 int r, pool_created = 0;
1945 struct pool_features pf;
1946 struct dm_arg_set as;
1947 struct dm_dev *data_dev;
1948 unsigned long block_size;
1949 dm_block_t low_water_blocks;
1950 struct dm_dev *metadata_dev;
1951 sector_t metadata_dev_size;
1952 char b[BDEVNAME_SIZE];
1955 * FIXME Remove validation from scope of lock.
1957 mutex_lock(&dm_thin_pool_table.mutex);
1960 ti->error = "Invalid argument count";
1967 r = dm_get_device(ti, argv[0], FMODE_READ | FMODE_WRITE, &metadata_dev);
1969 ti->error = "Error opening metadata block device";
1973 metadata_dev_size = i_size_read(metadata_dev->bdev->bd_inode) >> SECTOR_SHIFT;
1974 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING)
1975 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
1976 bdevname(metadata_dev->bdev, b), THIN_METADATA_MAX_SECTORS);
1978 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
1980 ti->error = "Error getting data device";
1984 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
1985 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
1986 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
1987 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
1988 ti->error = "Invalid block size";
1993 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
1994 ti->error = "Invalid low water mark";
2000 * Set default pool features.
2002 pool_features_init(&pf);
2004 dm_consume_args(&as, 4);
2005 r = parse_pool_features(&as, &pf, ti);
2009 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
2015 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
2016 block_size, &ti->error, &pool_created);
2023 * 'pool_created' reflects whether this is the first table load.
2024 * Top level discard support is not allowed to be changed after
2025 * initial load. This would require a pool reload to trigger thin
2028 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
2029 ti->error = "Discard support cannot be disabled once enabled";
2031 goto out_flags_changed;
2035 * The block layer requires discard_granularity to be a power of 2.
2037 if (pf.discard_enabled && !is_power_of_2(block_size)) {
2038 ti->error = "Discard support must be disabled when the block size is not a power of 2";
2040 goto out_flags_changed;
2045 pt->metadata_dev = metadata_dev;
2046 pt->data_dev = data_dev;
2047 pt->low_water_blocks = low_water_blocks;
2049 ti->num_flush_requests = 1;
2051 * Only need to enable discards if the pool should pass
2052 * them down to the data device. The thin device's discard
2053 * processing will cause mappings to be removed from the btree.
2055 if (pf.discard_enabled && pf.discard_passdown) {
2056 ti->num_discard_requests = 1;
2058 * Setting 'discards_supported' circumvents the normal
2059 * stacking of discard limits (this keeps the pool and
2060 * thin devices' discard limits consistent).
2062 ti->discards_supported = true;
2066 pt->callbacks.congested_fn = pool_is_congested;
2067 dm_table_add_target_callbacks(ti->table, &pt->callbacks);
2069 mutex_unlock(&dm_thin_pool_table.mutex);
2078 dm_put_device(ti, data_dev);
2080 dm_put_device(ti, metadata_dev);
2082 mutex_unlock(&dm_thin_pool_table.mutex);
2087 static int pool_map(struct dm_target *ti, struct bio *bio,
2088 union map_info *map_context)
2091 struct pool_c *pt = ti->private;
2092 struct pool *pool = pt->pool;
2093 unsigned long flags;
2096 * As this is a singleton target, ti->begin is always zero.
2098 spin_lock_irqsave(&pool->lock, flags);
2099 bio->bi_bdev = pt->data_dev->bdev;
2100 r = DM_MAPIO_REMAPPED;
2101 spin_unlock_irqrestore(&pool->lock, flags);
2107 * Retrieves the number of blocks of the data device from
2108 * the superblock and compares it to the actual device size,
2109 * thus resizing the data device in case it has grown.
2111 * This both copes with opening preallocated data devices in the ctr
2112 * being followed by a resume
2114 * calling the resume method individually after userspace has
2115 * grown the data device in reaction to a table event.
2117 static int pool_preresume(struct dm_target *ti)
2120 struct pool_c *pt = ti->private;
2121 struct pool *pool = pt->pool;
2122 sector_t data_size = ti->len;
2123 dm_block_t sb_data_size;
2126 * Take control of the pool object.
2128 r = bind_control_target(pool, ti);
2132 (void) sector_div(data_size, pool->sectors_per_block);
2134 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
2136 DMERR("failed to retrieve data device size");
2140 if (data_size < sb_data_size) {
2141 DMERR("pool target too small, is %llu blocks (expected %llu)",
2142 (unsigned long long)data_size, sb_data_size);
2145 } else if (data_size > sb_data_size) {
2146 r = dm_pool_resize_data_dev(pool->pmd, data_size);
2148 DMERR("failed to resize data device");
2152 r = dm_pool_commit_metadata(pool->pmd);
2154 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
2163 static void pool_resume(struct dm_target *ti)
2165 struct pool_c *pt = ti->private;
2166 struct pool *pool = pt->pool;
2167 unsigned long flags;
2169 spin_lock_irqsave(&pool->lock, flags);
2170 pool->low_water_triggered = 0;
2171 pool->no_free_space = 0;
2172 __requeue_bios(pool);
2173 spin_unlock_irqrestore(&pool->lock, flags);
2175 do_waker(&pool->waker.work);
2178 static void pool_postsuspend(struct dm_target *ti)
2181 struct pool_c *pt = ti->private;
2182 struct pool *pool = pt->pool;
2184 cancel_delayed_work(&pool->waker);
2185 flush_workqueue(pool->wq);
2187 r = dm_pool_commit_metadata(pool->pmd);
2189 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
2191 /* FIXME: invalidate device? error the next FUA or FLUSH bio ?*/
2195 static int check_arg_count(unsigned argc, unsigned args_required)
2197 if (argc != args_required) {
2198 DMWARN("Message received with %u arguments instead of %u.",
2199 argc, args_required);
2206 static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
2208 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
2209 *dev_id <= MAX_DEV_ID)
2213 DMWARN("Message received with invalid device id: %s", arg);
2218 static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
2223 r = check_arg_count(argc, 2);
2227 r = read_dev_id(argv[1], &dev_id, 1);
2231 r = dm_pool_create_thin(pool->pmd, dev_id);
2233 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2241 static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2244 dm_thin_id origin_dev_id;
2247 r = check_arg_count(argc, 3);
2251 r = read_dev_id(argv[1], &dev_id, 1);
2255 r = read_dev_id(argv[2], &origin_dev_id, 1);
2259 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
2261 DMWARN("Creation of new snapshot %s of device %s failed.",
2269 static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
2274 r = check_arg_count(argc, 2);
2278 r = read_dev_id(argv[1], &dev_id, 1);
2282 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
2284 DMWARN("Deletion of thin device %s failed.", argv[1]);
2289 static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
2291 dm_thin_id old_id, new_id;
2294 r = check_arg_count(argc, 3);
2298 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
2299 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
2303 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
2304 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
2308 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
2310 DMWARN("Failed to change transaction id from %s to %s.",
2318 static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2322 r = check_arg_count(argc, 1);
2326 r = dm_pool_commit_metadata(pool->pmd);
2328 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
2333 r = dm_pool_reserve_metadata_snap(pool->pmd);
2335 DMWARN("reserve_metadata_snap message failed.");
2340 static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2344 r = check_arg_count(argc, 1);
2348 r = dm_pool_release_metadata_snap(pool->pmd);
2350 DMWARN("release_metadata_snap message failed.");
2356 * Messages supported:
2357 * create_thin <dev_id>
2358 * create_snap <dev_id> <origin_id>
2360 * trim <dev_id> <new_size_in_sectors>
2361 * set_transaction_id <current_trans_id> <new_trans_id>
2362 * reserve_metadata_snap
2363 * release_metadata_snap
2365 static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
2368 struct pool_c *pt = ti->private;
2369 struct pool *pool = pt->pool;
2371 if (!strcasecmp(argv[0], "create_thin"))
2372 r = process_create_thin_mesg(argc, argv, pool);
2374 else if (!strcasecmp(argv[0], "create_snap"))
2375 r = process_create_snap_mesg(argc, argv, pool);
2377 else if (!strcasecmp(argv[0], "delete"))
2378 r = process_delete_mesg(argc, argv, pool);
2380 else if (!strcasecmp(argv[0], "set_transaction_id"))
2381 r = process_set_transaction_id_mesg(argc, argv, pool);
2383 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
2384 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
2386 else if (!strcasecmp(argv[0], "release_metadata_snap"))
2387 r = process_release_metadata_snap_mesg(argc, argv, pool);
2390 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
2393 r = dm_pool_commit_metadata(pool->pmd);
2395 DMERR("%s message: dm_pool_commit_metadata() failed, error = %d",
2404 * <transaction id> <used metadata sectors>/<total metadata sectors>
2405 * <used data sectors>/<total data sectors> <held metadata root>
2407 static int pool_status(struct dm_target *ti, status_type_t type,
2408 char *result, unsigned maxlen)
2412 uint64_t transaction_id;
2413 dm_block_t nr_free_blocks_data;
2414 dm_block_t nr_free_blocks_metadata;
2415 dm_block_t nr_blocks_data;
2416 dm_block_t nr_blocks_metadata;
2417 dm_block_t held_root;
2418 char buf[BDEVNAME_SIZE];
2419 char buf2[BDEVNAME_SIZE];
2420 struct pool_c *pt = ti->private;
2421 struct pool *pool = pt->pool;
2424 case STATUSTYPE_INFO:
2425 r = dm_pool_get_metadata_transaction_id(pool->pmd,
2430 r = dm_pool_get_free_metadata_block_count(pool->pmd,
2431 &nr_free_blocks_metadata);
2435 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
2439 r = dm_pool_get_free_block_count(pool->pmd,
2440 &nr_free_blocks_data);
2444 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
2448 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
2452 DMEMIT("%llu %llu/%llu %llu/%llu ",
2453 (unsigned long long)transaction_id,
2454 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
2455 (unsigned long long)nr_blocks_metadata,
2456 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
2457 (unsigned long long)nr_blocks_data);
2460 DMEMIT("%llu", held_root);
2466 case STATUSTYPE_TABLE:
2467 DMEMIT("%s %s %lu %llu ",
2468 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
2469 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
2470 (unsigned long)pool->sectors_per_block,
2471 (unsigned long long)pt->low_water_blocks);
2473 count = !pool->pf.zero_new_blocks + !pool->pf.discard_enabled +
2474 !pt->pf.discard_passdown;
2475 DMEMIT("%u ", count);
2477 if (!pool->pf.zero_new_blocks)
2478 DMEMIT("skip_block_zeroing ");
2480 if (!pool->pf.discard_enabled)
2481 DMEMIT("ignore_discard ");
2483 if (!pt->pf.discard_passdown)
2484 DMEMIT("no_discard_passdown ");
2492 static int pool_iterate_devices(struct dm_target *ti,
2493 iterate_devices_callout_fn fn, void *data)
2495 struct pool_c *pt = ti->private;
2497 return fn(ti, pt->data_dev, 0, ti->len, data);
2500 static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2501 struct bio_vec *biovec, int max_size)
2503 struct pool_c *pt = ti->private;
2504 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
2506 if (!q->merge_bvec_fn)
2509 bvm->bi_bdev = pt->data_dev->bdev;
2511 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2514 static void set_discard_limits(struct pool *pool, struct queue_limits *limits)
2517 * FIXME: these limits may be incompatible with the pool's data device
2519 limits->max_discard_sectors = pool->sectors_per_block;
2522 * This is just a hint, and not enforced. We have to cope with
2523 * bios that cover a block partially. A discard that spans a block
2524 * boundary is not sent to this target.
2526 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
2527 limits->discard_zeroes_data = pool->pf.zero_new_blocks;
2530 static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
2532 struct pool_c *pt = ti->private;
2533 struct pool *pool = pt->pool;
2535 blk_limits_io_min(limits, 0);
2536 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
2537 if (pool->pf.discard_enabled)
2538 set_discard_limits(pool, limits);
2541 static struct target_type pool_target = {
2542 .name = "thin-pool",
2543 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
2544 DM_TARGET_IMMUTABLE,
2545 .version = {1, 2, 0},
2546 .module = THIS_MODULE,
2550 .postsuspend = pool_postsuspend,
2551 .preresume = pool_preresume,
2552 .resume = pool_resume,
2553 .message = pool_message,
2554 .status = pool_status,
2555 .merge = pool_merge,
2556 .iterate_devices = pool_iterate_devices,
2557 .io_hints = pool_io_hints,
2560 /*----------------------------------------------------------------
2561 * Thin target methods
2562 *--------------------------------------------------------------*/
2563 static void thin_dtr(struct dm_target *ti)
2565 struct thin_c *tc = ti->private;
2567 mutex_lock(&dm_thin_pool_table.mutex);
2569 __pool_dec(tc->pool);
2570 dm_pool_close_thin_device(tc->td);
2571 dm_put_device(ti, tc->pool_dev);
2573 dm_put_device(ti, tc->origin_dev);
2576 mutex_unlock(&dm_thin_pool_table.mutex);
2580 * Thin target parameters:
2582 * <pool_dev> <dev_id> [origin_dev]
2584 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2585 * dev_id: the internal device identifier
2586 * origin_dev: a device external to the pool that should act as the origin
2588 * If the pool device has discards disabled, they get disabled for the thin
2591 static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
2595 struct dm_dev *pool_dev, *origin_dev;
2596 struct mapped_device *pool_md;
2598 mutex_lock(&dm_thin_pool_table.mutex);
2600 if (argc != 2 && argc != 3) {
2601 ti->error = "Invalid argument count";
2606 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
2608 ti->error = "Out of memory";
2614 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
2616 ti->error = "Error opening origin device";
2617 goto bad_origin_dev;
2619 tc->origin_dev = origin_dev;
2622 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
2624 ti->error = "Error opening pool device";
2627 tc->pool_dev = pool_dev;
2629 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
2630 ti->error = "Invalid device id";
2635 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
2637 ti->error = "Couldn't get pool mapped device";
2642 tc->pool = __pool_table_lookup(pool_md);
2644 ti->error = "Couldn't find pool object";
2646 goto bad_pool_lookup;
2648 __pool_inc(tc->pool);
2650 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
2652 ti->error = "Couldn't open thin internal device";
2656 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
2660 ti->num_flush_requests = 1;
2661 ti->flush_supported = true;
2663 /* In case the pool supports discards, pass them on. */
2664 if (tc->pool->pf.discard_enabled) {
2665 ti->discards_supported = true;
2666 ti->num_discard_requests = 1;
2667 ti->discard_zeroes_data_unsupported = true;
2668 /* Discard requests must be split on a block boundary */
2669 ti->split_discard_requests = true;
2674 mutex_unlock(&dm_thin_pool_table.mutex);
2679 __pool_dec(tc->pool);
2683 dm_put_device(ti, tc->pool_dev);
2686 dm_put_device(ti, tc->origin_dev);
2690 mutex_unlock(&dm_thin_pool_table.mutex);
2695 static int thin_map(struct dm_target *ti, struct bio *bio,
2696 union map_info *map_context)
2698 bio->bi_sector = dm_target_offset(ti, bio->bi_sector);
2700 return thin_bio_map(ti, bio, map_context);
2703 static int thin_endio(struct dm_target *ti,
2704 struct bio *bio, int err,
2705 union map_info *map_context)
2707 unsigned long flags;
2708 struct dm_thin_endio_hook *h = map_context->ptr;
2709 struct list_head work;
2710 struct dm_thin_new_mapping *m, *tmp;
2711 struct pool *pool = h->tc->pool;
2713 if (h->shared_read_entry) {
2714 INIT_LIST_HEAD(&work);
2715 ds_dec(h->shared_read_entry, &work);
2717 spin_lock_irqsave(&pool->lock, flags);
2718 list_for_each_entry_safe(m, tmp, &work, list) {
2721 __maybe_add_mapping(m);
2723 spin_unlock_irqrestore(&pool->lock, flags);
2726 if (h->all_io_entry) {
2727 INIT_LIST_HEAD(&work);
2728 ds_dec(h->all_io_entry, &work);
2729 spin_lock_irqsave(&pool->lock, flags);
2730 list_for_each_entry_safe(m, tmp, &work, list)
2731 list_add(&m->list, &pool->prepared_discards);
2732 spin_unlock_irqrestore(&pool->lock, flags);
2735 mempool_free(h, pool->endio_hook_pool);
2740 static void thin_postsuspend(struct dm_target *ti)
2742 if (dm_noflush_suspending(ti))
2743 requeue_io((struct thin_c *)ti->private);
2747 * <nr mapped sectors> <highest mapped sector>
2749 static int thin_status(struct dm_target *ti, status_type_t type,
2750 char *result, unsigned maxlen)
2754 dm_block_t mapped, highest;
2755 char buf[BDEVNAME_SIZE];
2756 struct thin_c *tc = ti->private;
2762 case STATUSTYPE_INFO:
2763 r = dm_thin_get_mapped_count(tc->td, &mapped);
2767 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
2771 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
2773 DMEMIT("%llu", ((highest + 1) *
2774 tc->pool->sectors_per_block) - 1);
2779 case STATUSTYPE_TABLE:
2781 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
2782 (unsigned long) tc->dev_id);
2784 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
2792 static int thin_iterate_devices(struct dm_target *ti,
2793 iterate_devices_callout_fn fn, void *data)
2796 struct thin_c *tc = ti->private;
2797 struct pool *pool = tc->pool;
2800 * We can't call dm_pool_get_data_dev_size() since that blocks. So
2801 * we follow a more convoluted path through to the pool's target.
2804 return 0; /* nothing is bound */
2806 blocks = pool->ti->len;
2807 (void) sector_div(blocks, pool->sectors_per_block);
2809 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
2814 static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits)
2816 struct thin_c *tc = ti->private;
2817 struct pool *pool = tc->pool;
2819 blk_limits_io_min(limits, 0);
2820 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
2821 set_discard_limits(pool, limits);
2824 static struct target_type thin_target = {
2826 .version = {1, 2, 0},
2827 .module = THIS_MODULE,
2831 .end_io = thin_endio,
2832 .postsuspend = thin_postsuspend,
2833 .status = thin_status,
2834 .iterate_devices = thin_iterate_devices,
2835 .io_hints = thin_io_hints,
2838 /*----------------------------------------------------------------*/
2840 static int __init dm_thin_init(void)
2846 r = dm_register_target(&thin_target);
2850 r = dm_register_target(&pool_target);
2852 goto bad_pool_target;
2856 _cell_cache = KMEM_CACHE(dm_bio_prison_cell, 0);
2858 goto bad_cell_cache;
2860 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
2861 if (!_new_mapping_cache)
2862 goto bad_new_mapping_cache;
2864 _endio_hook_cache = KMEM_CACHE(dm_thin_endio_hook, 0);
2865 if (!_endio_hook_cache)
2866 goto bad_endio_hook_cache;
2870 bad_endio_hook_cache:
2871 kmem_cache_destroy(_new_mapping_cache);
2872 bad_new_mapping_cache:
2873 kmem_cache_destroy(_cell_cache);
2875 dm_unregister_target(&pool_target);
2877 dm_unregister_target(&thin_target);
2882 static void dm_thin_exit(void)
2884 dm_unregister_target(&thin_target);
2885 dm_unregister_target(&pool_target);
2887 kmem_cache_destroy(_cell_cache);
2888 kmem_cache_destroy(_new_mapping_cache);
2889 kmem_cache_destroy(_endio_hook_cache);
2892 module_init(dm_thin_init);
2893 module_exit(dm_thin_exit);
2895 MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
2896 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2897 MODULE_LICENSE("GPL");