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>
16 #include <linux/vmalloc.h>
18 #define DM_MSG_PREFIX "thin"
23 #define ENDIO_HOOK_POOL_SIZE 1024
24 #define DEFERRED_SET_SIZE 64
25 #define MAPPING_POOL_SIZE 1024
26 #define PRISON_CELLS 1024
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 * The metadata device is currently limited in size. The limitation is
37 * checked lower down in dm-space-map-metadata, but we also check it here
38 * so we can fail early.
40 * We have one block of index, which can hold 255 index entries. Each
41 * index entry contains allocation info about 16k metadata blocks.
43 #define METADATA_DEV_MAX_SECTORS (255 * (1 << 14) * (THIN_METADATA_BLOCK_SIZE / (1 << SECTOR_SHIFT)))
46 * Device id is restricted to 24 bits.
48 #define MAX_DEV_ID ((1 << 24) - 1)
51 * How do we handle breaking sharing of data blocks?
52 * =================================================
54 * We use a standard copy-on-write btree to store the mappings for the
55 * devices (note I'm talking about copy-on-write of the metadata here, not
56 * the data). When you take an internal snapshot you clone the root node
57 * of the origin btree. After this there is no concept of an origin or a
58 * snapshot. They are just two device trees that happen to point to the
61 * When we get a write in we decide if it's to a shared data block using
62 * some timestamp magic. If it is, we have to break sharing.
64 * Let's say we write to a shared block in what was the origin. The
67 * i) plug io further to this physical block. (see bio_prison code).
69 * ii) quiesce any read io to that shared data block. Obviously
70 * including all devices that share this block. (see deferred_set code)
72 * iii) copy the data block to a newly allocate block. This step can be
73 * missed out if the io covers the block. (schedule_copy).
75 * iv) insert the new mapping into the origin's btree
76 * (process_prepared_mappings). This act of inserting breaks some
77 * sharing of btree nodes between the two devices. Breaking sharing only
78 * effects the btree of that specific device. Btrees for the other
79 * devices that share the block never change. The btree for the origin
80 * device as it was after the last commit is untouched, ie. we're using
81 * persistent data structures in the functional programming sense.
83 * v) unplug io to this physical block, including the io that triggered
84 * the breaking of sharing.
86 * Steps (ii) and (iii) occur in parallel.
88 * The metadata _doesn't_ need to be committed before the io continues. We
89 * get away with this because the io is always written to a _new_ block.
90 * If there's a crash, then:
92 * - The origin mapping will point to the old origin block (the shared
93 * one). This will contain the data as it was before the io that triggered
94 * the breaking of sharing came in.
96 * - The snap mapping still points to the old block. As it would after
99 * The downside of this scheme is the timestamp magic isn't perfect, and
100 * will continue to think that data block in the snapshot device is shared
101 * even after the write to the origin has broken sharing. I suspect data
102 * blocks will typically be shared by many different devices, so we're
103 * breaking sharing n + 1 times, rather than n, where n is the number of
104 * devices that reference this data block. At the moment I think the
105 * benefits far, far outweigh the disadvantages.
108 /*----------------------------------------------------------------*/
111 * Sometimes we can't deal with a bio straight away. We put them in prison
112 * where they can't cause any mischief. Bios are put in a cell identified
113 * by a key, multiple bios can be in the same cell. When the cell is
114 * subsequently unlocked the bios become available.
125 struct hlist_node list;
126 struct bio_prison *prison;
129 struct bio_list bios;
134 mempool_t *cell_pool;
138 struct hlist_head *cells;
141 static uint32_t calc_nr_buckets(unsigned nr_cells)
146 nr_cells = min(nr_cells, 8192u);
155 * @nr_cells should be the number of cells you want in use _concurrently_.
156 * Don't confuse it with the number of distinct keys.
158 static struct bio_prison *prison_create(unsigned nr_cells)
161 uint32_t nr_buckets = calc_nr_buckets(nr_cells);
162 struct bio_prison *prison = kmalloc(sizeof(*prison), GFP_KERNEL);
167 spin_lock_init(&prison->lock);
168 prison->cell_pool = mempool_create_kmalloc_pool(nr_cells,
169 sizeof(struct cell));
170 if (!prison->cell_pool) {
175 prison->cells = vmalloc(sizeof(*prison->cells) * nr_buckets);
176 if (!prison->cells) {
177 mempool_destroy(prison->cell_pool);
182 prison->nr_buckets = nr_buckets;
183 prison->hash_mask = nr_buckets - 1;
184 for (i = 0; i < nr_buckets; i++)
185 INIT_HLIST_HEAD(prison->cells + i);
190 static void prison_destroy(struct bio_prison *prison)
192 vfree(prison->cells);
193 mempool_destroy(prison->cell_pool);
197 static uint32_t hash_key(struct bio_prison *prison, struct cell_key *key)
199 const unsigned long BIG_PRIME = 4294967291UL;
200 uint64_t hash = key->block * BIG_PRIME;
202 return (uint32_t) (hash & prison->hash_mask);
205 static int keys_equal(struct cell_key *lhs, struct cell_key *rhs)
207 return (lhs->virtual == rhs->virtual) &&
208 (lhs->dev == rhs->dev) &&
209 (lhs->block == rhs->block);
212 static struct cell *__search_bucket(struct hlist_head *bucket,
213 struct cell_key *key)
216 struct hlist_node *tmp;
218 hlist_for_each_entry(cell, tmp, bucket, list)
219 if (keys_equal(&cell->key, key))
226 * This may block if a new cell needs allocating. You must ensure that
227 * cells will be unlocked even if the calling thread is blocked.
229 * Returns 1 if the cell was already held, 0 if @inmate is the new holder.
231 static int bio_detain(struct bio_prison *prison, struct cell_key *key,
232 struct bio *inmate, struct cell **ref)
236 uint32_t hash = hash_key(prison, key);
237 struct cell *cell, *cell2;
239 BUG_ON(hash > prison->nr_buckets);
241 spin_lock_irqsave(&prison->lock, flags);
243 cell = __search_bucket(prison->cells + hash, key);
245 bio_list_add(&cell->bios, inmate);
250 * Allocate a new cell
252 spin_unlock_irqrestore(&prison->lock, flags);
253 cell2 = mempool_alloc(prison->cell_pool, GFP_NOIO);
254 spin_lock_irqsave(&prison->lock, flags);
257 * We've been unlocked, so we have to double check that
258 * nobody else has inserted this cell in the meantime.
260 cell = __search_bucket(prison->cells + hash, key);
262 mempool_free(cell2, prison->cell_pool);
263 bio_list_add(&cell->bios, inmate);
272 cell->prison = prison;
273 memcpy(&cell->key, key, sizeof(cell->key));
274 cell->holder = inmate;
275 bio_list_init(&cell->bios);
276 hlist_add_head(&cell->list, prison->cells + hash);
281 spin_unlock_irqrestore(&prison->lock, flags);
289 * @inmates must have been initialised prior to this call
291 static void __cell_release(struct cell *cell, struct bio_list *inmates)
293 struct bio_prison *prison = cell->prison;
295 hlist_del(&cell->list);
298 bio_list_add(inmates, cell->holder);
299 bio_list_merge(inmates, &cell->bios);
302 mempool_free(cell, prison->cell_pool);
305 static void cell_release(struct cell *cell, struct bio_list *bios)
308 struct bio_prison *prison = cell->prison;
310 spin_lock_irqsave(&prison->lock, flags);
311 __cell_release(cell, bios);
312 spin_unlock_irqrestore(&prison->lock, flags);
316 * There are a couple of places where we put a bio into a cell briefly
317 * before taking it out again. In these situations we know that no other
318 * bio may be in the cell. This function releases the cell, and also does
321 static void __cell_release_singleton(struct cell *cell, struct bio *bio)
323 BUG_ON(cell->holder != bio);
324 BUG_ON(!bio_list_empty(&cell->bios));
326 __cell_release(cell, NULL);
329 static void cell_release_singleton(struct cell *cell, struct bio *bio)
332 struct bio_prison *prison = cell->prison;
334 spin_lock_irqsave(&prison->lock, flags);
335 __cell_release_singleton(cell, bio);
336 spin_unlock_irqrestore(&prison->lock, flags);
340 * Sometimes we don't want the holder, just the additional bios.
342 static void __cell_release_no_holder(struct cell *cell, struct bio_list *inmates)
344 struct bio_prison *prison = cell->prison;
346 hlist_del(&cell->list);
347 bio_list_merge(inmates, &cell->bios);
349 mempool_free(cell, prison->cell_pool);
352 static void cell_release_no_holder(struct cell *cell, struct bio_list *inmates)
355 struct bio_prison *prison = cell->prison;
357 spin_lock_irqsave(&prison->lock, flags);
358 __cell_release_no_holder(cell, inmates);
359 spin_unlock_irqrestore(&prison->lock, flags);
362 static void cell_error(struct cell *cell)
364 struct bio_prison *prison = cell->prison;
365 struct bio_list bios;
369 bio_list_init(&bios);
371 spin_lock_irqsave(&prison->lock, flags);
372 __cell_release(cell, &bios);
373 spin_unlock_irqrestore(&prison->lock, flags);
375 while ((bio = bio_list_pop(&bios)))
379 /*----------------------------------------------------------------*/
382 * We use the deferred set to keep track of pending reads to shared blocks.
383 * We do this to ensure the new mapping caused by a write isn't performed
384 * until these prior reads have completed. Otherwise the insertion of the
385 * new mapping could free the old block that the read bios are mapped to.
389 struct deferred_entry {
390 struct deferred_set *ds;
392 struct list_head work_items;
395 struct deferred_set {
397 unsigned current_entry;
399 struct deferred_entry entries[DEFERRED_SET_SIZE];
402 static void ds_init(struct deferred_set *ds)
406 spin_lock_init(&ds->lock);
407 ds->current_entry = 0;
409 for (i = 0; i < DEFERRED_SET_SIZE; i++) {
410 ds->entries[i].ds = ds;
411 ds->entries[i].count = 0;
412 INIT_LIST_HEAD(&ds->entries[i].work_items);
416 static struct deferred_entry *ds_inc(struct deferred_set *ds)
419 struct deferred_entry *entry;
421 spin_lock_irqsave(&ds->lock, flags);
422 entry = ds->entries + ds->current_entry;
424 spin_unlock_irqrestore(&ds->lock, flags);
429 static unsigned ds_next(unsigned index)
431 return (index + 1) % DEFERRED_SET_SIZE;
434 static void __sweep(struct deferred_set *ds, struct list_head *head)
436 while ((ds->sweeper != ds->current_entry) &&
437 !ds->entries[ds->sweeper].count) {
438 list_splice_init(&ds->entries[ds->sweeper].work_items, head);
439 ds->sweeper = ds_next(ds->sweeper);
442 if ((ds->sweeper == ds->current_entry) && !ds->entries[ds->sweeper].count)
443 list_splice_init(&ds->entries[ds->sweeper].work_items, head);
446 static void ds_dec(struct deferred_entry *entry, struct list_head *head)
450 spin_lock_irqsave(&entry->ds->lock, flags);
451 BUG_ON(!entry->count);
453 __sweep(entry->ds, head);
454 spin_unlock_irqrestore(&entry->ds->lock, flags);
458 * Returns 1 if deferred or 0 if no pending items to delay job.
460 static int ds_add_work(struct deferred_set *ds, struct list_head *work)
466 spin_lock_irqsave(&ds->lock, flags);
467 if ((ds->sweeper == ds->current_entry) &&
468 !ds->entries[ds->current_entry].count)
471 list_add(work, &ds->entries[ds->current_entry].work_items);
472 next_entry = ds_next(ds->current_entry);
473 if (!ds->entries[next_entry].count)
474 ds->current_entry = next_entry;
476 spin_unlock_irqrestore(&ds->lock, flags);
481 /*----------------------------------------------------------------*/
486 static void build_data_key(struct dm_thin_device *td,
487 dm_block_t b, struct cell_key *key)
490 key->dev = dm_thin_dev_id(td);
494 static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
495 struct cell_key *key)
498 key->dev = dm_thin_dev_id(td);
502 /*----------------------------------------------------------------*/
505 * A pool device ties together a metadata device and a data device. It
506 * also provides the interface for creating and destroying internal
511 struct list_head list;
512 struct dm_target *ti; /* Only set if a pool target is bound */
514 struct mapped_device *pool_md;
515 struct block_device *md_dev;
516 struct dm_pool_metadata *pmd;
518 uint32_t sectors_per_block;
519 unsigned block_shift;
520 dm_block_t offset_mask;
521 dm_block_t low_water_blocks;
523 unsigned zero_new_blocks:1;
524 unsigned low_water_triggered:1; /* A dm event has been sent */
525 unsigned no_free_space:1; /* A -ENOSPC warning has been issued */
527 struct bio_prison *prison;
528 struct dm_kcopyd_client *copier;
530 struct workqueue_struct *wq;
531 struct work_struct worker;
536 struct bio_list deferred_bios;
537 struct bio_list deferred_flush_bios;
538 struct list_head prepared_mappings;
540 struct bio_list retry_on_resume_list;
542 struct deferred_set ds; /* FIXME: move to thin_c */
544 struct new_mapping *next_mapping;
545 mempool_t *mapping_pool;
546 mempool_t *endio_hook_pool;
550 * Target context for a pool.
553 struct dm_target *ti;
555 struct dm_dev *data_dev;
556 struct dm_dev *metadata_dev;
557 struct dm_target_callbacks callbacks;
559 dm_block_t low_water_blocks;
560 unsigned zero_new_blocks:1;
564 * Target context for a thin.
567 struct dm_dev *pool_dev;
571 struct dm_thin_device *td;
574 /*----------------------------------------------------------------*/
577 * A global list of pools that uses a struct mapped_device as a key.
579 static struct dm_thin_pool_table {
581 struct list_head pools;
582 } dm_thin_pool_table;
584 static void pool_table_init(void)
586 mutex_init(&dm_thin_pool_table.mutex);
587 INIT_LIST_HEAD(&dm_thin_pool_table.pools);
590 static void __pool_table_insert(struct pool *pool)
592 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
593 list_add(&pool->list, &dm_thin_pool_table.pools);
596 static void __pool_table_remove(struct pool *pool)
598 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
599 list_del(&pool->list);
602 static struct pool *__pool_table_lookup(struct mapped_device *md)
604 struct pool *pool = NULL, *tmp;
606 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
608 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
609 if (tmp->pool_md == md) {
618 static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
620 struct pool *pool = NULL, *tmp;
622 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
624 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
625 if (tmp->md_dev == md_dev) {
634 /*----------------------------------------------------------------*/
636 static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master)
639 struct bio_list bios;
641 bio_list_init(&bios);
642 bio_list_merge(&bios, master);
643 bio_list_init(master);
645 while ((bio = bio_list_pop(&bios))) {
646 if (dm_get_mapinfo(bio)->ptr == tc)
647 bio_endio(bio, DM_ENDIO_REQUEUE);
649 bio_list_add(master, bio);
653 static void requeue_io(struct thin_c *tc)
655 struct pool *pool = tc->pool;
658 spin_lock_irqsave(&pool->lock, flags);
659 __requeue_bio_list(tc, &pool->deferred_bios);
660 __requeue_bio_list(tc, &pool->retry_on_resume_list);
661 spin_unlock_irqrestore(&pool->lock, flags);
665 * This section of code contains the logic for processing a thin device's IO.
666 * Much of the code depends on pool object resources (lists, workqueues, etc)
667 * but most is exclusively called from the thin target rather than the thin-pool
671 static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
673 return bio->bi_sector >> tc->pool->block_shift;
676 static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
678 struct pool *pool = tc->pool;
680 bio->bi_bdev = tc->pool_dev->bdev;
681 bio->bi_sector = (block << pool->block_shift) +
682 (bio->bi_sector & pool->offset_mask);
685 static void remap_and_issue(struct thin_c *tc, struct bio *bio,
688 struct pool *pool = tc->pool;
691 remap(tc, bio, block);
694 * Batch together any FUA/FLUSH bios we find and then issue
695 * a single commit for them in process_deferred_bios().
697 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
698 spin_lock_irqsave(&pool->lock, flags);
699 bio_list_add(&pool->deferred_flush_bios, bio);
700 spin_unlock_irqrestore(&pool->lock, flags);
702 generic_make_request(bio);
706 * wake_worker() is used when new work is queued and when pool_resume is
707 * ready to continue deferred IO processing.
709 static void wake_worker(struct pool *pool)
711 queue_work(pool->wq, &pool->worker);
714 /*----------------------------------------------------------------*/
717 * Bio endio functions.
721 bio_end_io_t *saved_bi_end_io;
722 struct deferred_entry *entry;
726 struct list_head list;
731 dm_block_t virt_block;
732 dm_block_t data_block;
737 * If the bio covers the whole area of a block then we can avoid
738 * zeroing or copying. Instead this bio is hooked. The bio will
739 * still be in the cell, so care has to be taken to avoid issuing
743 bio_end_io_t *saved_bi_end_io;
746 static void __maybe_add_mapping(struct new_mapping *m)
748 struct pool *pool = m->tc->pool;
750 if (list_empty(&m->list) && m->prepared) {
751 list_add(&m->list, &pool->prepared_mappings);
756 static void copy_complete(int read_err, unsigned long write_err, void *context)
759 struct new_mapping *m = context;
760 struct pool *pool = m->tc->pool;
762 m->err = read_err || write_err ? -EIO : 0;
764 spin_lock_irqsave(&pool->lock, flags);
766 __maybe_add_mapping(m);
767 spin_unlock_irqrestore(&pool->lock, flags);
770 static void overwrite_endio(struct bio *bio, int err)
773 struct new_mapping *m = dm_get_mapinfo(bio)->ptr;
774 struct pool *pool = m->tc->pool;
778 spin_lock_irqsave(&pool->lock, flags);
780 __maybe_add_mapping(m);
781 spin_unlock_irqrestore(&pool->lock, flags);
784 static void shared_read_endio(struct bio *bio, int err)
786 struct list_head mappings;
787 struct new_mapping *m, *tmp;
788 struct endio_hook *h = dm_get_mapinfo(bio)->ptr;
790 struct pool *pool = h->tc->pool;
792 bio->bi_end_io = h->saved_bi_end_io;
795 INIT_LIST_HEAD(&mappings);
796 ds_dec(h->entry, &mappings);
798 spin_lock_irqsave(&pool->lock, flags);
799 list_for_each_entry_safe(m, tmp, &mappings, list) {
801 INIT_LIST_HEAD(&m->list);
802 __maybe_add_mapping(m);
804 spin_unlock_irqrestore(&pool->lock, flags);
806 mempool_free(h, pool->endio_hook_pool);
809 /*----------------------------------------------------------------*/
816 * Prepared mapping jobs.
820 * This sends the bios in the cell back to the deferred_bios list.
822 static void cell_defer(struct thin_c *tc, struct cell *cell,
823 dm_block_t data_block)
825 struct pool *pool = tc->pool;
828 spin_lock_irqsave(&pool->lock, flags);
829 cell_release(cell, &pool->deferred_bios);
830 spin_unlock_irqrestore(&tc->pool->lock, flags);
836 * Same as cell_defer above, except it omits one particular detainee,
837 * a write bio that covers the block and has already been processed.
839 static void cell_defer_except(struct thin_c *tc, struct cell *cell)
841 struct bio_list bios;
842 struct pool *pool = tc->pool;
845 bio_list_init(&bios);
847 spin_lock_irqsave(&pool->lock, flags);
848 cell_release_no_holder(cell, &pool->deferred_bios);
849 spin_unlock_irqrestore(&pool->lock, flags);
854 static void process_prepared_mapping(struct new_mapping *m)
856 struct thin_c *tc = m->tc;
862 bio->bi_end_io = m->saved_bi_end_io;
870 * Commit the prepared block into the mapping btree.
871 * Any I/O for this block arriving after this point will get
872 * remapped to it directly.
874 r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block);
876 DMERR("dm_thin_insert_block() failed");
882 * Release any bios held while the block was being provisioned.
883 * If we are processing a write bio that completely covers the block,
884 * we already processed it so can ignore it now when processing
885 * the bios in the cell.
888 cell_defer_except(tc, m->cell);
891 cell_defer(tc, m->cell, m->data_block);
895 mempool_free(m, tc->pool->mapping_pool);
898 static void process_prepared_mappings(struct pool *pool)
901 struct list_head maps;
902 struct new_mapping *m, *tmp;
904 INIT_LIST_HEAD(&maps);
905 spin_lock_irqsave(&pool->lock, flags);
906 list_splice_init(&pool->prepared_mappings, &maps);
907 spin_unlock_irqrestore(&pool->lock, flags);
909 list_for_each_entry_safe(m, tmp, &maps, list)
910 process_prepared_mapping(m);
916 static int io_overwrites_block(struct pool *pool, struct bio *bio)
918 return ((bio_data_dir(bio) == WRITE) &&
919 !(bio->bi_sector & pool->offset_mask)) &&
920 (bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT));
923 static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
926 *save = bio->bi_end_io;
930 static int ensure_next_mapping(struct pool *pool)
932 if (pool->next_mapping)
935 pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);
937 return pool->next_mapping ? 0 : -ENOMEM;
940 static struct new_mapping *get_next_mapping(struct pool *pool)
942 struct new_mapping *r = pool->next_mapping;
944 BUG_ON(!pool->next_mapping);
946 pool->next_mapping = NULL;
951 static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
952 dm_block_t data_origin, dm_block_t data_dest,
953 struct cell *cell, struct bio *bio)
956 struct pool *pool = tc->pool;
957 struct new_mapping *m = get_next_mapping(pool);
959 INIT_LIST_HEAD(&m->list);
962 m->virt_block = virt_block;
963 m->data_block = data_dest;
968 ds_add_work(&pool->ds, &m->list);
971 * IO to pool_dev remaps to the pool target's data_dev.
973 * If the whole block of data is being overwritten, we can issue the
974 * bio immediately. Otherwise we use kcopyd to clone the data first.
976 if (io_overwrites_block(pool, bio)) {
978 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
979 dm_get_mapinfo(bio)->ptr = m;
980 remap_and_issue(tc, bio, data_dest);
982 struct dm_io_region from, to;
984 from.bdev = tc->pool_dev->bdev;
985 from.sector = data_origin * pool->sectors_per_block;
986 from.count = pool->sectors_per_block;
988 to.bdev = tc->pool_dev->bdev;
989 to.sector = data_dest * pool->sectors_per_block;
990 to.count = pool->sectors_per_block;
992 r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
993 0, copy_complete, m);
995 mempool_free(m, pool->mapping_pool);
996 DMERR("dm_kcopyd_copy() failed");
1002 static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
1003 dm_block_t data_block, struct cell *cell,
1006 struct pool *pool = tc->pool;
1007 struct new_mapping *m = get_next_mapping(pool);
1009 INIT_LIST_HEAD(&m->list);
1012 m->virt_block = virt_block;
1013 m->data_block = data_block;
1019 * If the whole block of data is being overwritten or we are not
1020 * zeroing pre-existing data, we can issue the bio immediately.
1021 * Otherwise we use kcopyd to zero the data first.
1023 if (!pool->zero_new_blocks)
1024 process_prepared_mapping(m);
1026 else if (io_overwrites_block(pool, bio)) {
1028 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
1029 dm_get_mapinfo(bio)->ptr = m;
1030 remap_and_issue(tc, bio, data_block);
1034 struct dm_io_region to;
1036 to.bdev = tc->pool_dev->bdev;
1037 to.sector = data_block * pool->sectors_per_block;
1038 to.count = pool->sectors_per_block;
1040 r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m);
1042 mempool_free(m, pool->mapping_pool);
1043 DMERR("dm_kcopyd_zero() failed");
1049 static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
1052 dm_block_t free_blocks;
1053 unsigned long flags;
1054 struct pool *pool = tc->pool;
1056 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1060 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
1061 DMWARN("%s: reached low water mark, sending event.",
1062 dm_device_name(pool->pool_md));
1063 spin_lock_irqsave(&pool->lock, flags);
1064 pool->low_water_triggered = 1;
1065 spin_unlock_irqrestore(&pool->lock, flags);
1066 dm_table_event(pool->ti->table);
1070 if (pool->no_free_space)
1074 * Try to commit to see if that will free up some
1077 r = dm_pool_commit_metadata(pool->pmd);
1079 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
1084 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1089 * If we still have no space we set a flag to avoid
1090 * doing all this checking and return -ENOSPC.
1093 DMWARN("%s: no free space available.",
1094 dm_device_name(pool->pool_md));
1095 spin_lock_irqsave(&pool->lock, flags);
1096 pool->no_free_space = 1;
1097 spin_unlock_irqrestore(&pool->lock, flags);
1103 r = dm_pool_alloc_data_block(pool->pmd, result);
1111 * If we have run out of space, queue bios until the device is
1112 * resumed, presumably after having been reloaded with more space.
1114 static void retry_on_resume(struct bio *bio)
1116 struct thin_c *tc = dm_get_mapinfo(bio)->ptr;
1117 struct pool *pool = tc->pool;
1118 unsigned long flags;
1120 spin_lock_irqsave(&pool->lock, flags);
1121 bio_list_add(&pool->retry_on_resume_list, bio);
1122 spin_unlock_irqrestore(&pool->lock, flags);
1125 static void no_space(struct cell *cell)
1128 struct bio_list bios;
1130 bio_list_init(&bios);
1131 cell_release(cell, &bios);
1133 while ((bio = bio_list_pop(&bios)))
1134 retry_on_resume(bio);
1137 static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
1138 struct cell_key *key,
1139 struct dm_thin_lookup_result *lookup_result,
1143 dm_block_t data_block;
1145 r = alloc_data_block(tc, &data_block);
1148 schedule_copy(tc, block, lookup_result->block,
1149 data_block, cell, bio);
1157 DMERR("%s: alloc_data_block() failed, error = %d", __func__, r);
1163 static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1165 struct dm_thin_lookup_result *lookup_result)
1168 struct pool *pool = tc->pool;
1169 struct cell_key key;
1172 * If cell is already occupied, then sharing is already in the process
1173 * of being broken so we have nothing further to do here.
1175 build_data_key(tc->td, lookup_result->block, &key);
1176 if (bio_detain(pool->prison, &key, bio, &cell))
1179 if (bio_data_dir(bio) == WRITE)
1180 break_sharing(tc, bio, block, &key, lookup_result, cell);
1182 struct endio_hook *h;
1183 h = mempool_alloc(pool->endio_hook_pool, GFP_NOIO);
1186 h->entry = ds_inc(&pool->ds);
1187 save_and_set_endio(bio, &h->saved_bi_end_io, shared_read_endio);
1188 dm_get_mapinfo(bio)->ptr = h;
1190 cell_release_singleton(cell, bio);
1191 remap_and_issue(tc, bio, lookup_result->block);
1195 static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
1199 dm_block_t data_block;
1202 * Remap empty bios (flushes) immediately, without provisioning.
1204 if (!bio->bi_size) {
1205 cell_release_singleton(cell, bio);
1206 remap_and_issue(tc, bio, 0);
1211 * Fill read bios with zeroes and complete them immediately.
1213 if (bio_data_dir(bio) == READ) {
1215 cell_release_singleton(cell, bio);
1220 r = alloc_data_block(tc, &data_block);
1223 schedule_zero(tc, block, data_block, cell, bio);
1231 DMERR("%s: alloc_data_block() failed, error = %d", __func__, r);
1237 static void process_bio(struct thin_c *tc, struct bio *bio)
1240 dm_block_t block = get_bio_block(tc, bio);
1242 struct cell_key key;
1243 struct dm_thin_lookup_result lookup_result;
1246 * If cell is already occupied, then the block is already
1247 * being provisioned so we have nothing further to do here.
1249 build_virtual_key(tc->td, block, &key);
1250 if (bio_detain(tc->pool->prison, &key, bio, &cell))
1253 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1257 * We can release this cell now. This thread is the only
1258 * one that puts bios into a cell, and we know there were
1259 * no preceding bios.
1262 * TODO: this will probably have to change when discard goes
1265 cell_release_singleton(cell, bio);
1267 if (lookup_result.shared)
1268 process_shared_bio(tc, bio, block, &lookup_result);
1270 remap_and_issue(tc, bio, lookup_result.block);
1274 provision_block(tc, bio, block, cell);
1278 DMERR("dm_thin_find_block() failed, error = %d", r);
1284 static void process_deferred_bios(struct pool *pool)
1286 unsigned long flags;
1288 struct bio_list bios;
1291 bio_list_init(&bios);
1293 spin_lock_irqsave(&pool->lock, flags);
1294 bio_list_merge(&bios, &pool->deferred_bios);
1295 bio_list_init(&pool->deferred_bios);
1296 spin_unlock_irqrestore(&pool->lock, flags);
1298 while ((bio = bio_list_pop(&bios))) {
1299 struct thin_c *tc = dm_get_mapinfo(bio)->ptr;
1301 * If we've got no free new_mapping structs, and processing
1302 * this bio might require one, we pause until there are some
1303 * prepared mappings to process.
1305 if (ensure_next_mapping(pool)) {
1306 spin_lock_irqsave(&pool->lock, flags);
1307 bio_list_add(&pool->deferred_bios, bio);
1308 bio_list_merge(&pool->deferred_bios, &bios);
1309 spin_unlock_irqrestore(&pool->lock, flags);
1312 process_bio(tc, bio);
1316 * If there are any deferred flush bios, we must commit
1317 * the metadata before issuing them.
1319 bio_list_init(&bios);
1320 spin_lock_irqsave(&pool->lock, flags);
1321 bio_list_merge(&bios, &pool->deferred_flush_bios);
1322 bio_list_init(&pool->deferred_flush_bios);
1323 spin_unlock_irqrestore(&pool->lock, flags);
1325 if (bio_list_empty(&bios))
1328 r = dm_pool_commit_metadata(pool->pmd);
1330 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
1332 while ((bio = bio_list_pop(&bios)))
1337 while ((bio = bio_list_pop(&bios)))
1338 generic_make_request(bio);
1341 static void do_worker(struct work_struct *ws)
1343 struct pool *pool = container_of(ws, struct pool, worker);
1345 process_prepared_mappings(pool);
1346 process_deferred_bios(pool);
1349 /*----------------------------------------------------------------*/
1352 * Mapping functions.
1356 * Called only while mapping a thin bio to hand it over to the workqueue.
1358 static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
1360 unsigned long flags;
1361 struct pool *pool = tc->pool;
1363 spin_lock_irqsave(&pool->lock, flags);
1364 bio_list_add(&pool->deferred_bios, bio);
1365 spin_unlock_irqrestore(&pool->lock, flags);
1371 * Non-blocking function called from the thin target's map function.
1373 static int thin_bio_map(struct dm_target *ti, struct bio *bio,
1374 union map_info *map_context)
1377 struct thin_c *tc = ti->private;
1378 dm_block_t block = get_bio_block(tc, bio);
1379 struct dm_thin_device *td = tc->td;
1380 struct dm_thin_lookup_result result;
1383 * Save the thin context for easy access from the deferred bio later.
1385 map_context->ptr = tc;
1387 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1388 thin_defer_bio(tc, bio);
1389 return DM_MAPIO_SUBMITTED;
1392 r = dm_thin_find_block(td, block, 0, &result);
1395 * Note that we defer readahead too.
1399 if (unlikely(result.shared)) {
1401 * We have a race condition here between the
1402 * result.shared value returned by the lookup and
1403 * snapshot creation, which may cause new
1406 * To avoid this always quiesce the origin before
1407 * taking the snap. You want to do this anyway to
1408 * ensure a consistent application view
1411 * More distant ancestors are irrelevant. The
1412 * shared flag will be set in their case.
1414 thin_defer_bio(tc, bio);
1415 r = DM_MAPIO_SUBMITTED;
1417 remap(tc, bio, result.block);
1418 r = DM_MAPIO_REMAPPED;
1424 * In future, the failed dm_thin_find_block above could
1425 * provide the hint to load the metadata into cache.
1428 thin_defer_bio(tc, bio);
1429 r = DM_MAPIO_SUBMITTED;
1436 static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
1439 unsigned long flags;
1440 struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
1442 spin_lock_irqsave(&pt->pool->lock, flags);
1443 r = !bio_list_empty(&pt->pool->retry_on_resume_list);
1444 spin_unlock_irqrestore(&pt->pool->lock, flags);
1447 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1448 r = bdi_congested(&q->backing_dev_info, bdi_bits);
1454 static void __requeue_bios(struct pool *pool)
1456 bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list);
1457 bio_list_init(&pool->retry_on_resume_list);
1460 /*----------------------------------------------------------------
1461 * Binding of control targets to a pool object
1462 *--------------------------------------------------------------*/
1463 static int bind_control_target(struct pool *pool, struct dm_target *ti)
1465 struct pool_c *pt = ti->private;
1468 pool->low_water_blocks = pt->low_water_blocks;
1469 pool->zero_new_blocks = pt->zero_new_blocks;
1474 static void unbind_control_target(struct pool *pool, struct dm_target *ti)
1480 /*----------------------------------------------------------------
1482 *--------------------------------------------------------------*/
1483 static void __pool_destroy(struct pool *pool)
1485 __pool_table_remove(pool);
1487 if (dm_pool_metadata_close(pool->pmd) < 0)
1488 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1490 prison_destroy(pool->prison);
1491 dm_kcopyd_client_destroy(pool->copier);
1494 destroy_workqueue(pool->wq);
1496 if (pool->next_mapping)
1497 mempool_free(pool->next_mapping, pool->mapping_pool);
1498 mempool_destroy(pool->mapping_pool);
1499 mempool_destroy(pool->endio_hook_pool);
1503 static struct pool *pool_create(struct mapped_device *pool_md,
1504 struct block_device *metadata_dev,
1505 unsigned long block_size, char **error)
1510 struct dm_pool_metadata *pmd;
1512 pmd = dm_pool_metadata_open(metadata_dev, block_size);
1514 *error = "Error creating metadata object";
1515 return (struct pool *)pmd;
1518 pool = kmalloc(sizeof(*pool), GFP_KERNEL);
1520 *error = "Error allocating memory for pool";
1521 err_p = ERR_PTR(-ENOMEM);
1526 pool->sectors_per_block = block_size;
1527 pool->block_shift = ffs(block_size) - 1;
1528 pool->offset_mask = block_size - 1;
1529 pool->low_water_blocks = 0;
1530 pool->zero_new_blocks = 1;
1531 pool->prison = prison_create(PRISON_CELLS);
1532 if (!pool->prison) {
1533 *error = "Error creating pool's bio prison";
1534 err_p = ERR_PTR(-ENOMEM);
1538 pool->copier = dm_kcopyd_client_create();
1539 if (IS_ERR(pool->copier)) {
1540 r = PTR_ERR(pool->copier);
1541 *error = "Error creating pool's kcopyd client";
1543 goto bad_kcopyd_client;
1547 * Create singlethreaded workqueue that will service all devices
1548 * that use this metadata.
1550 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
1552 *error = "Error creating pool's workqueue";
1553 err_p = ERR_PTR(-ENOMEM);
1557 INIT_WORK(&pool->worker, do_worker);
1558 spin_lock_init(&pool->lock);
1559 bio_list_init(&pool->deferred_bios);
1560 bio_list_init(&pool->deferred_flush_bios);
1561 INIT_LIST_HEAD(&pool->prepared_mappings);
1562 pool->low_water_triggered = 0;
1563 pool->no_free_space = 0;
1564 bio_list_init(&pool->retry_on_resume_list);
1567 pool->next_mapping = NULL;
1568 pool->mapping_pool =
1569 mempool_create_kmalloc_pool(MAPPING_POOL_SIZE, sizeof(struct new_mapping));
1570 if (!pool->mapping_pool) {
1571 *error = "Error creating pool's mapping mempool";
1572 err_p = ERR_PTR(-ENOMEM);
1573 goto bad_mapping_pool;
1576 pool->endio_hook_pool =
1577 mempool_create_kmalloc_pool(ENDIO_HOOK_POOL_SIZE, sizeof(struct endio_hook));
1578 if (!pool->endio_hook_pool) {
1579 *error = "Error creating pool's endio_hook mempool";
1580 err_p = ERR_PTR(-ENOMEM);
1581 goto bad_endio_hook_pool;
1583 pool->ref_count = 1;
1584 pool->pool_md = pool_md;
1585 pool->md_dev = metadata_dev;
1586 __pool_table_insert(pool);
1590 bad_endio_hook_pool:
1591 mempool_destroy(pool->mapping_pool);
1593 destroy_workqueue(pool->wq);
1595 dm_kcopyd_client_destroy(pool->copier);
1597 prison_destroy(pool->prison);
1601 if (dm_pool_metadata_close(pmd))
1602 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1607 static void __pool_inc(struct pool *pool)
1609 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1613 static void __pool_dec(struct pool *pool)
1615 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1616 BUG_ON(!pool->ref_count);
1617 if (!--pool->ref_count)
1618 __pool_destroy(pool);
1621 static struct pool *__pool_find(struct mapped_device *pool_md,
1622 struct block_device *metadata_dev,
1623 unsigned long block_size, char **error)
1625 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
1628 if (pool->pool_md != pool_md)
1629 return ERR_PTR(-EBUSY);
1633 pool = __pool_table_lookup(pool_md);
1635 if (pool->md_dev != metadata_dev)
1636 return ERR_PTR(-EINVAL);
1640 pool = pool_create(pool_md, metadata_dev, block_size, error);
1646 /*----------------------------------------------------------------
1647 * Pool target methods
1648 *--------------------------------------------------------------*/
1649 static void pool_dtr(struct dm_target *ti)
1651 struct pool_c *pt = ti->private;
1653 mutex_lock(&dm_thin_pool_table.mutex);
1655 unbind_control_target(pt->pool, ti);
1656 __pool_dec(pt->pool);
1657 dm_put_device(ti, pt->metadata_dev);
1658 dm_put_device(ti, pt->data_dev);
1661 mutex_unlock(&dm_thin_pool_table.mutex);
1664 struct pool_features {
1665 unsigned zero_new_blocks:1;
1668 static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
1669 struct dm_target *ti)
1673 const char *arg_name;
1675 static struct dm_arg _args[] = {
1676 {0, 1, "Invalid number of pool feature arguments"},
1680 * No feature arguments supplied.
1685 r = dm_read_arg_group(_args, as, &argc, &ti->error);
1689 while (argc && !r) {
1690 arg_name = dm_shift_arg(as);
1693 if (!strcasecmp(arg_name, "skip_block_zeroing")) {
1694 pf->zero_new_blocks = 0;
1698 ti->error = "Unrecognised pool feature requested";
1706 * thin-pool <metadata dev> <data dev>
1707 * <data block size (sectors)>
1708 * <low water mark (blocks)>
1709 * [<#feature args> [<arg>]*]
1711 * Optional feature arguments are:
1712 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
1714 static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
1719 struct pool_features pf;
1720 struct dm_arg_set as;
1721 struct dm_dev *data_dev;
1722 unsigned long block_size;
1723 dm_block_t low_water_blocks;
1724 struct dm_dev *metadata_dev;
1725 sector_t metadata_dev_size;
1728 * FIXME Remove validation from scope of lock.
1730 mutex_lock(&dm_thin_pool_table.mutex);
1733 ti->error = "Invalid argument count";
1740 r = dm_get_device(ti, argv[0], FMODE_READ | FMODE_WRITE, &metadata_dev);
1742 ti->error = "Error opening metadata block device";
1746 metadata_dev_size = i_size_read(metadata_dev->bdev->bd_inode) >> SECTOR_SHIFT;
1747 if (metadata_dev_size > METADATA_DEV_MAX_SECTORS) {
1748 ti->error = "Metadata device is too large";
1753 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
1755 ti->error = "Error getting data device";
1759 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
1760 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
1761 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
1762 !is_power_of_2(block_size)) {
1763 ti->error = "Invalid block size";
1768 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
1769 ti->error = "Invalid low water mark";
1775 * Set default pool features.
1777 memset(&pf, 0, sizeof(pf));
1778 pf.zero_new_blocks = 1;
1780 dm_consume_args(&as, 4);
1781 r = parse_pool_features(&as, &pf, ti);
1785 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
1791 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
1792 block_size, &ti->error);
1800 pt->metadata_dev = metadata_dev;
1801 pt->data_dev = data_dev;
1802 pt->low_water_blocks = low_water_blocks;
1803 pt->zero_new_blocks = pf.zero_new_blocks;
1804 ti->num_flush_requests = 1;
1805 ti->num_discard_requests = 0;
1808 pt->callbacks.congested_fn = pool_is_congested;
1809 dm_table_add_target_callbacks(ti->table, &pt->callbacks);
1811 mutex_unlock(&dm_thin_pool_table.mutex);
1818 dm_put_device(ti, data_dev);
1820 dm_put_device(ti, metadata_dev);
1822 mutex_unlock(&dm_thin_pool_table.mutex);
1827 static int pool_map(struct dm_target *ti, struct bio *bio,
1828 union map_info *map_context)
1831 struct pool_c *pt = ti->private;
1832 struct pool *pool = pt->pool;
1833 unsigned long flags;
1836 * As this is a singleton target, ti->begin is always zero.
1838 spin_lock_irqsave(&pool->lock, flags);
1839 bio->bi_bdev = pt->data_dev->bdev;
1840 r = DM_MAPIO_REMAPPED;
1841 spin_unlock_irqrestore(&pool->lock, flags);
1847 * Retrieves the number of blocks of the data device from
1848 * the superblock and compares it to the actual device size,
1849 * thus resizing the data device in case it has grown.
1851 * This both copes with opening preallocated data devices in the ctr
1852 * being followed by a resume
1854 * calling the resume method individually after userspace has
1855 * grown the data device in reaction to a table event.
1857 static int pool_preresume(struct dm_target *ti)
1860 struct pool_c *pt = ti->private;
1861 struct pool *pool = pt->pool;
1862 dm_block_t data_size, sb_data_size;
1865 * Take control of the pool object.
1867 r = bind_control_target(pool, ti);
1871 data_size = ti->len >> pool->block_shift;
1872 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
1874 DMERR("failed to retrieve data device size");
1878 if (data_size < sb_data_size) {
1879 DMERR("pool target too small, is %llu blocks (expected %llu)",
1880 data_size, sb_data_size);
1883 } else if (data_size > sb_data_size) {
1884 r = dm_pool_resize_data_dev(pool->pmd, data_size);
1886 DMERR("failed to resize data device");
1890 r = dm_pool_commit_metadata(pool->pmd);
1892 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
1901 static void pool_resume(struct dm_target *ti)
1903 struct pool_c *pt = ti->private;
1904 struct pool *pool = pt->pool;
1905 unsigned long flags;
1907 spin_lock_irqsave(&pool->lock, flags);
1908 pool->low_water_triggered = 0;
1909 pool->no_free_space = 0;
1910 __requeue_bios(pool);
1911 spin_unlock_irqrestore(&pool->lock, flags);
1916 static void pool_postsuspend(struct dm_target *ti)
1919 struct pool_c *pt = ti->private;
1920 struct pool *pool = pt->pool;
1922 flush_workqueue(pool->wq);
1924 r = dm_pool_commit_metadata(pool->pmd);
1926 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
1928 /* FIXME: invalidate device? error the next FUA or FLUSH bio ?*/
1932 static int check_arg_count(unsigned argc, unsigned args_required)
1934 if (argc != args_required) {
1935 DMWARN("Message received with %u arguments instead of %u.",
1936 argc, args_required);
1943 static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
1945 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
1946 *dev_id <= MAX_DEV_ID)
1950 DMWARN("Message received with invalid device id: %s", arg);
1955 static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
1960 r = check_arg_count(argc, 2);
1964 r = read_dev_id(argv[1], &dev_id, 1);
1968 r = dm_pool_create_thin(pool->pmd, dev_id);
1970 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
1978 static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
1981 dm_thin_id origin_dev_id;
1984 r = check_arg_count(argc, 3);
1988 r = read_dev_id(argv[1], &dev_id, 1);
1992 r = read_dev_id(argv[2], &origin_dev_id, 1);
1996 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
1998 DMWARN("Creation of new snapshot %s of device %s failed.",
2006 static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
2011 r = check_arg_count(argc, 2);
2015 r = read_dev_id(argv[1], &dev_id, 1);
2019 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
2021 DMWARN("Deletion of thin device %s failed.", argv[1]);
2026 static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
2028 dm_thin_id old_id, new_id;
2031 r = check_arg_count(argc, 3);
2035 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
2036 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
2040 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
2041 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
2045 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
2047 DMWARN("Failed to change transaction id from %s to %s.",
2056 * Messages supported:
2057 * create_thin <dev_id>
2058 * create_snap <dev_id> <origin_id>
2060 * trim <dev_id> <new_size_in_sectors>
2061 * set_transaction_id <current_trans_id> <new_trans_id>
2063 static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
2066 struct pool_c *pt = ti->private;
2067 struct pool *pool = pt->pool;
2069 if (!strcasecmp(argv[0], "create_thin"))
2070 r = process_create_thin_mesg(argc, argv, pool);
2072 else if (!strcasecmp(argv[0], "create_snap"))
2073 r = process_create_snap_mesg(argc, argv, pool);
2075 else if (!strcasecmp(argv[0], "delete"))
2076 r = process_delete_mesg(argc, argv, pool);
2078 else if (!strcasecmp(argv[0], "set_transaction_id"))
2079 r = process_set_transaction_id_mesg(argc, argv, pool);
2082 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
2085 r = dm_pool_commit_metadata(pool->pmd);
2087 DMERR("%s message: dm_pool_commit_metadata() failed, error = %d",
2096 * <transaction id> <used metadata sectors>/<total metadata sectors>
2097 * <used data sectors>/<total data sectors> <held metadata root>
2099 static void pool_status(struct dm_target *ti, status_type_t type,
2100 char *result, unsigned maxlen)
2104 uint64_t transaction_id;
2105 dm_block_t nr_free_blocks_data;
2106 dm_block_t nr_free_blocks_metadata;
2107 dm_block_t nr_blocks_data;
2108 dm_block_t nr_blocks_metadata;
2109 dm_block_t held_root;
2110 char buf[BDEVNAME_SIZE];
2111 char buf2[BDEVNAME_SIZE];
2112 struct pool_c *pt = ti->private;
2113 struct pool *pool = pt->pool;
2116 case STATUSTYPE_INFO:
2117 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
2119 DMERR("dm_pool_get_metadata_transaction_id returned %d", r);
2123 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
2125 DMERR("dm_pool_get_free_metadata_block_count returned %d", r);
2129 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
2131 DMERR("dm_pool_get_metadata_dev_size returned %d", r);
2135 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
2137 DMERR("dm_pool_get_free_block_count returned %d", r);
2141 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
2143 DMERR("dm_pool_get_data_dev_size returned %d", r);
2147 r = dm_pool_get_held_metadata_root(pool->pmd, &held_root);
2149 DMERR("dm_pool_get_held_metadata_root returned %d", r);
2153 DMEMIT("%llu %llu/%llu %llu/%llu ",
2154 (unsigned long long)transaction_id,
2155 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
2156 (unsigned long long)nr_blocks_metadata,
2157 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
2158 (unsigned long long)nr_blocks_data);
2161 DMEMIT("%llu", held_root);
2167 case STATUSTYPE_TABLE:
2168 DMEMIT("%s %s %lu %llu ",
2169 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
2170 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
2171 (unsigned long)pool->sectors_per_block,
2172 (unsigned long long)pt->low_water_blocks);
2174 DMEMIT("%u ", !pool->zero_new_blocks);
2176 if (!pool->zero_new_blocks)
2177 DMEMIT("skip_block_zeroing ");
2186 static int pool_iterate_devices(struct dm_target *ti,
2187 iterate_devices_callout_fn fn, void *data)
2189 struct pool_c *pt = ti->private;
2191 return fn(ti, pt->data_dev, 0, ti->len, data);
2194 static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2195 struct bio_vec *biovec, int max_size)
2197 struct pool_c *pt = ti->private;
2198 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
2200 if (!q->merge_bvec_fn)
2203 bvm->bi_bdev = pt->data_dev->bdev;
2205 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2208 static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
2210 struct pool_c *pt = ti->private;
2211 struct pool *pool = pt->pool;
2213 blk_limits_io_min(limits, 0);
2214 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
2217 static struct target_type pool_target = {
2218 .name = "thin-pool",
2219 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
2220 DM_TARGET_IMMUTABLE,
2221 .version = {1, 0, 2},
2222 .module = THIS_MODULE,
2226 .postsuspend = pool_postsuspend,
2227 .preresume = pool_preresume,
2228 .resume = pool_resume,
2229 .message = pool_message,
2230 .status = pool_status,
2231 .merge = pool_merge,
2232 .iterate_devices = pool_iterate_devices,
2233 .io_hints = pool_io_hints,
2236 /*----------------------------------------------------------------
2237 * Thin target methods
2238 *--------------------------------------------------------------*/
2239 static void thin_dtr(struct dm_target *ti)
2241 struct thin_c *tc = ti->private;
2243 mutex_lock(&dm_thin_pool_table.mutex);
2245 __pool_dec(tc->pool);
2246 dm_pool_close_thin_device(tc->td);
2247 dm_put_device(ti, tc->pool_dev);
2250 mutex_unlock(&dm_thin_pool_table.mutex);
2254 * Thin target parameters:
2256 * <pool_dev> <dev_id>
2258 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2259 * dev_id: the internal device identifier
2261 static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
2265 struct dm_dev *pool_dev;
2266 struct mapped_device *pool_md;
2268 mutex_lock(&dm_thin_pool_table.mutex);
2271 ti->error = "Invalid argument count";
2276 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
2278 ti->error = "Out of memory";
2283 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
2285 ti->error = "Error opening pool device";
2288 tc->pool_dev = pool_dev;
2290 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
2291 ti->error = "Invalid device id";
2296 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
2298 ti->error = "Couldn't get pool mapped device";
2303 tc->pool = __pool_table_lookup(pool_md);
2305 ti->error = "Couldn't find pool object";
2307 goto bad_pool_lookup;
2309 __pool_inc(tc->pool);
2311 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
2313 ti->error = "Couldn't open thin internal device";
2317 ti->split_io = tc->pool->sectors_per_block;
2318 ti->num_flush_requests = 1;
2319 ti->num_discard_requests = 0;
2320 ti->discards_supported = 0;
2324 mutex_unlock(&dm_thin_pool_table.mutex);
2329 __pool_dec(tc->pool);
2333 dm_put_device(ti, tc->pool_dev);
2337 mutex_unlock(&dm_thin_pool_table.mutex);
2342 static int thin_map(struct dm_target *ti, struct bio *bio,
2343 union map_info *map_context)
2345 bio->bi_sector -= ti->begin;
2347 return thin_bio_map(ti, bio, map_context);
2350 static void thin_postsuspend(struct dm_target *ti)
2352 if (dm_noflush_suspending(ti))
2353 requeue_io((struct thin_c *)ti->private);
2357 * <nr mapped sectors> <highest mapped sector>
2359 static void thin_status(struct dm_target *ti, status_type_t type,
2360 char *result, unsigned maxlen)
2364 dm_block_t mapped, highest;
2365 char buf[BDEVNAME_SIZE];
2366 struct thin_c *tc = ti->private;
2372 case STATUSTYPE_INFO:
2373 r = dm_thin_get_mapped_count(tc->td, &mapped);
2375 DMERR("dm_thin_get_mapped_count returned %d", r);
2379 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
2381 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
2385 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
2387 DMEMIT("%llu", ((highest + 1) *
2388 tc->pool->sectors_per_block) - 1);
2393 case STATUSTYPE_TABLE:
2395 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
2396 (unsigned long) tc->dev_id);
2407 static int thin_iterate_devices(struct dm_target *ti,
2408 iterate_devices_callout_fn fn, void *data)
2411 struct thin_c *tc = ti->private;
2414 * We can't call dm_pool_get_data_dev_size() since that blocks. So
2415 * we follow a more convoluted path through to the pool's target.
2418 return 0; /* nothing is bound */
2420 blocks = tc->pool->ti->len >> tc->pool->block_shift;
2422 return fn(ti, tc->pool_dev, 0, tc->pool->sectors_per_block * blocks, data);
2427 static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits)
2429 struct thin_c *tc = ti->private;
2431 blk_limits_io_min(limits, 0);
2432 blk_limits_io_opt(limits, tc->pool->sectors_per_block << SECTOR_SHIFT);
2435 static struct target_type thin_target = {
2437 .version = {1, 0, 2},
2438 .module = THIS_MODULE,
2442 .postsuspend = thin_postsuspend,
2443 .status = thin_status,
2444 .iterate_devices = thin_iterate_devices,
2445 .io_hints = thin_io_hints,
2448 /*----------------------------------------------------------------*/
2450 static int __init dm_thin_init(void)
2456 r = dm_register_target(&thin_target);
2460 r = dm_register_target(&pool_target);
2462 dm_unregister_target(&thin_target);
2467 static void dm_thin_exit(void)
2469 dm_unregister_target(&thin_target);
2470 dm_unregister_target(&pool_target);
2473 module_init(dm_thin_init);
2474 module_exit(dm_thin_exit);
2476 MODULE_DESCRIPTION(DM_NAME "device-mapper thin provisioning target");
2477 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2478 MODULE_LICENSE("GPL");