2 * Copyright (C) 2011-2012 Red Hat UK.
4 * This file is released under the GPL.
7 #include "dm-thin-metadata.h"
8 #include "dm-bio-prison.h"
11 #include <linux/device-mapper.h>
12 #include <linux/dm-io.h>
13 #include <linux/dm-kcopyd.h>
14 #include <linux/list.h>
15 #include <linux/init.h>
16 #include <linux/module.h>
17 #include <linux/slab.h>
19 #define DM_MSG_PREFIX "thin"
24 #define ENDIO_HOOK_POOL_SIZE 1024
25 #define MAPPING_POOL_SIZE 1024
26 #define PRISON_CELLS 1024
27 #define COMMIT_PERIOD HZ
29 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
30 "A percentage of time allocated for copy on write");
33 * The block size of the device holding pool data must be
34 * between 64KB and 1GB.
36 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
37 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
40 * Device id is restricted to 24 bits.
42 #define MAX_DEV_ID ((1 << 24) - 1)
45 * How do we handle breaking sharing of data blocks?
46 * =================================================
48 * We use a standard copy-on-write btree to store the mappings for the
49 * devices (note I'm talking about copy-on-write of the metadata here, not
50 * the data). When you take an internal snapshot you clone the root node
51 * of the origin btree. After this there is no concept of an origin or a
52 * snapshot. They are just two device trees that happen to point to the
55 * When we get a write in we decide if it's to a shared data block using
56 * some timestamp magic. If it is, we have to break sharing.
58 * Let's say we write to a shared block in what was the origin. The
61 * i) plug io further to this physical block. (see bio_prison code).
63 * ii) quiesce any read io to that shared data block. Obviously
64 * including all devices that share this block. (see dm_deferred_set code)
66 * iii) copy the data block to a newly allocate block. This step can be
67 * missed out if the io covers the block. (schedule_copy).
69 * iv) insert the new mapping into the origin's btree
70 * (process_prepared_mapping). This act of inserting breaks some
71 * sharing of btree nodes between the two devices. Breaking sharing only
72 * effects the btree of that specific device. Btrees for the other
73 * devices that share the block never change. The btree for the origin
74 * device as it was after the last commit is untouched, ie. we're using
75 * persistent data structures in the functional programming sense.
77 * v) unplug io to this physical block, including the io that triggered
78 * the breaking of sharing.
80 * Steps (ii) and (iii) occur in parallel.
82 * The metadata _doesn't_ need to be committed before the io continues. We
83 * get away with this because the io is always written to a _new_ block.
84 * If there's a crash, then:
86 * - The origin mapping will point to the old origin block (the shared
87 * one). This will contain the data as it was before the io that triggered
88 * the breaking of sharing came in.
90 * - The snap mapping still points to the old block. As it would after
93 * The downside of this scheme is the timestamp magic isn't perfect, and
94 * will continue to think that data block in the snapshot device is shared
95 * even after the write to the origin has broken sharing. I suspect data
96 * blocks will typically be shared by many different devices, so we're
97 * breaking sharing n + 1 times, rather than n, where n is the number of
98 * devices that reference this data block. At the moment I think the
99 * benefits far, far outweigh the disadvantages.
102 /*----------------------------------------------------------------*/
107 static void build_data_key(struct dm_thin_device *td,
108 dm_block_t b, struct dm_cell_key *key)
111 key->dev = dm_thin_dev_id(td);
115 static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
116 struct dm_cell_key *key)
119 key->dev = dm_thin_dev_id(td);
123 /*----------------------------------------------------------------*/
126 * A pool device ties together a metadata device and a data device. It
127 * also provides the interface for creating and destroying internal
130 struct dm_thin_new_mapping;
133 * The pool runs in 3 modes. Ordered in degraded order for comparisons.
136 PM_WRITE, /* metadata may be changed */
137 PM_READ_ONLY, /* metadata may not be changed */
138 PM_FAIL, /* all I/O fails */
141 struct pool_features {
144 bool zero_new_blocks:1;
145 bool discard_enabled:1;
146 bool discard_passdown:1;
147 bool error_if_no_space:1;
151 typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
152 typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);
155 struct list_head list;
156 struct dm_target *ti; /* Only set if a pool target is bound */
158 struct mapped_device *pool_md;
159 struct block_device *md_dev;
160 struct dm_pool_metadata *pmd;
162 dm_block_t low_water_blocks;
163 uint32_t sectors_per_block;
164 int sectors_per_block_shift;
166 struct pool_features pf;
167 bool low_water_triggered:1; /* A dm event has been sent */
169 struct dm_bio_prison *prison;
170 struct dm_kcopyd_client *copier;
172 struct workqueue_struct *wq;
173 struct work_struct worker;
174 struct delayed_work waker;
176 unsigned long last_commit_jiffies;
180 struct bio_list deferred_bios;
181 struct bio_list deferred_flush_bios;
182 struct list_head prepared_mappings;
183 struct list_head prepared_discards;
185 struct bio_list retry_on_resume_list;
187 struct dm_deferred_set *shared_read_ds;
188 struct dm_deferred_set *all_io_ds;
190 struct dm_thin_new_mapping *next_mapping;
191 mempool_t *mapping_pool;
193 process_bio_fn process_bio;
194 process_bio_fn process_discard;
196 process_mapping_fn process_prepared_mapping;
197 process_mapping_fn process_prepared_discard;
200 static enum pool_mode get_pool_mode(struct pool *pool);
201 static void out_of_data_space(struct pool *pool);
202 static void metadata_operation_failed(struct pool *pool, const char *op, int r);
205 * Target context for a pool.
208 struct dm_target *ti;
210 struct dm_dev *data_dev;
211 struct dm_dev *metadata_dev;
212 struct dm_target_callbacks callbacks;
214 dm_block_t low_water_blocks;
215 struct pool_features requested_pf; /* Features requested during table load */
216 struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */
220 * Target context for a thin.
223 struct dm_dev *pool_dev;
224 struct dm_dev *origin_dev;
228 struct dm_thin_device *td;
231 /*----------------------------------------------------------------*/
234 * wake_worker() is used when new work is queued and when pool_resume is
235 * ready to continue deferred IO processing.
237 static void wake_worker(struct pool *pool)
239 queue_work(pool->wq, &pool->worker);
242 /*----------------------------------------------------------------*/
244 static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio,
245 struct dm_bio_prison_cell **cell_result)
248 struct dm_bio_prison_cell *cell_prealloc;
251 * Allocate a cell from the prison's mempool.
252 * This might block but it can't fail.
254 cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO);
256 r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result);
259 * We reused an old cell; we can get rid of
262 dm_bio_prison_free_cell(pool->prison, cell_prealloc);
267 static void cell_release(struct pool *pool,
268 struct dm_bio_prison_cell *cell,
269 struct bio_list *bios)
271 dm_cell_release(pool->prison, cell, bios);
272 dm_bio_prison_free_cell(pool->prison, cell);
275 static void cell_release_no_holder(struct pool *pool,
276 struct dm_bio_prison_cell *cell,
277 struct bio_list *bios)
279 dm_cell_release_no_holder(pool->prison, cell, bios);
280 dm_bio_prison_free_cell(pool->prison, cell);
283 static void cell_defer_no_holder_no_free(struct thin_c *tc,
284 struct dm_bio_prison_cell *cell)
286 struct pool *pool = tc->pool;
289 spin_lock_irqsave(&pool->lock, flags);
290 dm_cell_release_no_holder(pool->prison, cell, &pool->deferred_bios);
291 spin_unlock_irqrestore(&pool->lock, flags);
296 static void cell_error(struct pool *pool,
297 struct dm_bio_prison_cell *cell)
299 dm_cell_error(pool->prison, cell);
300 dm_bio_prison_free_cell(pool->prison, cell);
303 /*----------------------------------------------------------------*/
306 * A global list of pools that uses a struct mapped_device as a key.
308 static struct dm_thin_pool_table {
310 struct list_head pools;
311 } dm_thin_pool_table;
313 static void pool_table_init(void)
315 mutex_init(&dm_thin_pool_table.mutex);
316 INIT_LIST_HEAD(&dm_thin_pool_table.pools);
319 static void __pool_table_insert(struct pool *pool)
321 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
322 list_add(&pool->list, &dm_thin_pool_table.pools);
325 static void __pool_table_remove(struct pool *pool)
327 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
328 list_del(&pool->list);
331 static struct pool *__pool_table_lookup(struct mapped_device *md)
333 struct pool *pool = NULL, *tmp;
335 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
337 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
338 if (tmp->pool_md == md) {
347 static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
349 struct pool *pool = NULL, *tmp;
351 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
353 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
354 if (tmp->md_dev == md_dev) {
363 /*----------------------------------------------------------------*/
365 struct dm_thin_endio_hook {
367 struct dm_deferred_entry *shared_read_entry;
368 struct dm_deferred_entry *all_io_entry;
369 struct dm_thin_new_mapping *overwrite_mapping;
372 static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master)
375 struct bio_list bios;
377 bio_list_init(&bios);
378 bio_list_merge(&bios, master);
379 bio_list_init(master);
381 while ((bio = bio_list_pop(&bios))) {
382 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
385 bio_endio(bio, DM_ENDIO_REQUEUE);
387 bio_list_add(master, bio);
391 static void requeue_io(struct thin_c *tc)
393 struct pool *pool = tc->pool;
396 spin_lock_irqsave(&pool->lock, flags);
397 __requeue_bio_list(tc, &pool->deferred_bios);
398 __requeue_bio_list(tc, &pool->retry_on_resume_list);
399 spin_unlock_irqrestore(&pool->lock, flags);
403 * This section of code contains the logic for processing a thin device's IO.
404 * Much of the code depends on pool object resources (lists, workqueues, etc)
405 * but most is exclusively called from the thin target rather than the thin-pool
409 static bool block_size_is_power_of_two(struct pool *pool)
411 return pool->sectors_per_block_shift >= 0;
414 static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
416 struct pool *pool = tc->pool;
417 sector_t block_nr = bio->bi_sector;
419 if (block_size_is_power_of_two(pool))
420 block_nr >>= pool->sectors_per_block_shift;
422 (void) sector_div(block_nr, pool->sectors_per_block);
427 static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
429 struct pool *pool = tc->pool;
430 sector_t bi_sector = bio->bi_sector;
432 bio->bi_bdev = tc->pool_dev->bdev;
433 if (block_size_is_power_of_two(pool))
434 bio->bi_sector = (block << pool->sectors_per_block_shift) |
435 (bi_sector & (pool->sectors_per_block - 1));
437 bio->bi_sector = (block * pool->sectors_per_block) +
438 sector_div(bi_sector, pool->sectors_per_block);
441 static void remap_to_origin(struct thin_c *tc, struct bio *bio)
443 bio->bi_bdev = tc->origin_dev->bdev;
446 static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
448 return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) &&
449 dm_thin_changed_this_transaction(tc->td);
452 static void inc_all_io_entry(struct pool *pool, struct bio *bio)
454 struct dm_thin_endio_hook *h;
456 if (bio->bi_rw & REQ_DISCARD)
459 h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
460 h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
463 static void issue(struct thin_c *tc, struct bio *bio)
465 struct pool *pool = tc->pool;
468 if (!bio_triggers_commit(tc, bio)) {
469 generic_make_request(bio);
474 * Complete bio with an error if earlier I/O caused changes to
475 * the metadata that can't be committed e.g, due to I/O errors
476 * on the metadata device.
478 if (dm_thin_aborted_changes(tc->td)) {
484 * Batch together any bios that trigger commits and then issue a
485 * single commit for them in process_deferred_bios().
487 spin_lock_irqsave(&pool->lock, flags);
488 bio_list_add(&pool->deferred_flush_bios, bio);
489 spin_unlock_irqrestore(&pool->lock, flags);
492 static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
494 remap_to_origin(tc, bio);
498 static void remap_and_issue(struct thin_c *tc, struct bio *bio,
501 remap(tc, bio, block);
505 /*----------------------------------------------------------------*/
508 * Bio endio functions.
510 struct dm_thin_new_mapping {
511 struct list_head list;
516 bool definitely_not_shared:1;
520 dm_block_t virt_block;
521 dm_block_t data_block;
522 struct dm_bio_prison_cell *cell, *cell2;
525 * If the bio covers the whole area of a block then we can avoid
526 * zeroing or copying. Instead this bio is hooked. The bio will
527 * still be in the cell, so care has to be taken to avoid issuing
531 bio_end_io_t *saved_bi_end_io;
534 static void __maybe_add_mapping(struct dm_thin_new_mapping *m)
536 struct pool *pool = m->tc->pool;
538 if (m->quiesced && m->prepared) {
539 list_add_tail(&m->list, &pool->prepared_mappings);
544 static void copy_complete(int read_err, unsigned long write_err, void *context)
547 struct dm_thin_new_mapping *m = context;
548 struct pool *pool = m->tc->pool;
550 m->err = read_err || write_err ? -EIO : 0;
552 spin_lock_irqsave(&pool->lock, flags);
554 __maybe_add_mapping(m);
555 spin_unlock_irqrestore(&pool->lock, flags);
558 static void overwrite_endio(struct bio *bio, int err)
561 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
562 struct dm_thin_new_mapping *m = h->overwrite_mapping;
563 struct pool *pool = m->tc->pool;
567 spin_lock_irqsave(&pool->lock, flags);
569 __maybe_add_mapping(m);
570 spin_unlock_irqrestore(&pool->lock, flags);
573 /*----------------------------------------------------------------*/
580 * Prepared mapping jobs.
584 * This sends the bios in the cell back to the deferred_bios list.
586 static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell)
588 struct pool *pool = tc->pool;
591 spin_lock_irqsave(&pool->lock, flags);
592 cell_release(pool, cell, &pool->deferred_bios);
593 spin_unlock_irqrestore(&tc->pool->lock, flags);
599 * Same as cell_defer above, except it omits the original holder of the cell.
601 static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
603 struct pool *pool = tc->pool;
606 spin_lock_irqsave(&pool->lock, flags);
607 cell_release_no_holder(pool, cell, &pool->deferred_bios);
608 spin_unlock_irqrestore(&pool->lock, flags);
613 static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
616 m->bio->bi_end_io = m->saved_bi_end_io;
617 cell_error(m->tc->pool, m->cell);
619 mempool_free(m, m->tc->pool->mapping_pool);
622 static void process_prepared_mapping(struct dm_thin_new_mapping *m)
624 struct thin_c *tc = m->tc;
625 struct pool *pool = tc->pool;
631 bio->bi_end_io = m->saved_bi_end_io;
634 cell_error(pool, m->cell);
639 * Commit the prepared block into the mapping btree.
640 * Any I/O for this block arriving after this point will get
641 * remapped to it directly.
643 r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block);
645 metadata_operation_failed(pool, "dm_thin_insert_block", r);
646 cell_error(pool, m->cell);
651 * Release any bios held while the block was being provisioned.
652 * If we are processing a write bio that completely covers the block,
653 * we already processed it so can ignore it now when processing
654 * the bios in the cell.
657 cell_defer_no_holder(tc, m->cell);
660 cell_defer(tc, m->cell);
664 mempool_free(m, pool->mapping_pool);
667 static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
669 struct thin_c *tc = m->tc;
671 bio_io_error(m->bio);
672 cell_defer_no_holder(tc, m->cell);
673 cell_defer_no_holder(tc, m->cell2);
674 mempool_free(m, tc->pool->mapping_pool);
677 static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m)
679 struct thin_c *tc = m->tc;
681 inc_all_io_entry(tc->pool, m->bio);
682 cell_defer_no_holder(tc, m->cell);
683 cell_defer_no_holder(tc, m->cell2);
686 if (m->definitely_not_shared)
687 remap_and_issue(tc, m->bio, m->data_block);
690 if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used)
691 bio_endio(m->bio, 0);
693 remap_and_issue(tc, m->bio, m->data_block);
696 bio_endio(m->bio, 0);
698 mempool_free(m, tc->pool->mapping_pool);
701 static void process_prepared_discard(struct dm_thin_new_mapping *m)
704 struct thin_c *tc = m->tc;
706 r = dm_thin_remove_block(tc->td, m->virt_block);
708 DMERR_LIMIT("dm_thin_remove_block() failed");
710 process_prepared_discard_passdown(m);
713 static void process_prepared(struct pool *pool, struct list_head *head,
714 process_mapping_fn *fn)
717 struct list_head maps;
718 struct dm_thin_new_mapping *m, *tmp;
720 INIT_LIST_HEAD(&maps);
721 spin_lock_irqsave(&pool->lock, flags);
722 list_splice_init(head, &maps);
723 spin_unlock_irqrestore(&pool->lock, flags);
725 list_for_each_entry_safe(m, tmp, &maps, list)
732 static int io_overlaps_block(struct pool *pool, struct bio *bio)
734 return bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT);
737 static int io_overwrites_block(struct pool *pool, struct bio *bio)
739 return (bio_data_dir(bio) == WRITE) &&
740 io_overlaps_block(pool, bio);
743 static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
746 *save = bio->bi_end_io;
750 static int ensure_next_mapping(struct pool *pool)
752 if (pool->next_mapping)
755 pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);
757 return pool->next_mapping ? 0 : -ENOMEM;
760 static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
762 struct dm_thin_new_mapping *m = pool->next_mapping;
764 BUG_ON(!pool->next_mapping);
766 memset(m, 0, sizeof(struct dm_thin_new_mapping));
767 INIT_LIST_HEAD(&m->list);
770 pool->next_mapping = NULL;
775 static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
776 struct dm_dev *origin, dm_block_t data_origin,
777 dm_block_t data_dest,
778 struct dm_bio_prison_cell *cell, struct bio *bio)
781 struct pool *pool = tc->pool;
782 struct dm_thin_new_mapping *m = get_next_mapping(pool);
785 m->virt_block = virt_block;
786 m->data_block = data_dest;
789 if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
793 * IO to pool_dev remaps to the pool target's data_dev.
795 * If the whole block of data is being overwritten, we can issue the
796 * bio immediately. Otherwise we use kcopyd to clone the data first.
798 if (io_overwrites_block(pool, bio)) {
799 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
801 h->overwrite_mapping = m;
803 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
804 inc_all_io_entry(pool, bio);
805 remap_and_issue(tc, bio, data_dest);
807 struct dm_io_region from, to;
809 from.bdev = origin->bdev;
810 from.sector = data_origin * pool->sectors_per_block;
811 from.count = pool->sectors_per_block;
813 to.bdev = tc->pool_dev->bdev;
814 to.sector = data_dest * pool->sectors_per_block;
815 to.count = pool->sectors_per_block;
817 r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
818 0, copy_complete, m);
820 mempool_free(m, pool->mapping_pool);
821 DMERR_LIMIT("dm_kcopyd_copy() failed");
822 cell_error(pool, cell);
827 static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
828 dm_block_t data_origin, dm_block_t data_dest,
829 struct dm_bio_prison_cell *cell, struct bio *bio)
831 schedule_copy(tc, virt_block, tc->pool_dev,
832 data_origin, data_dest, cell, bio);
835 static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
836 dm_block_t data_dest,
837 struct dm_bio_prison_cell *cell, struct bio *bio)
839 schedule_copy(tc, virt_block, tc->origin_dev,
840 virt_block, data_dest, cell, bio);
843 static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
844 dm_block_t data_block, struct dm_bio_prison_cell *cell,
847 struct pool *pool = tc->pool;
848 struct dm_thin_new_mapping *m = get_next_mapping(pool);
853 m->virt_block = virt_block;
854 m->data_block = data_block;
858 * If the whole block of data is being overwritten or we are not
859 * zeroing pre-existing data, we can issue the bio immediately.
860 * Otherwise we use kcopyd to zero the data first.
862 if (!pool->pf.zero_new_blocks)
863 process_prepared_mapping(m);
865 else if (io_overwrites_block(pool, bio)) {
866 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
868 h->overwrite_mapping = m;
870 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
871 inc_all_io_entry(pool, bio);
872 remap_and_issue(tc, bio, data_block);
875 struct dm_io_region to;
877 to.bdev = tc->pool_dev->bdev;
878 to.sector = data_block * pool->sectors_per_block;
879 to.count = pool->sectors_per_block;
881 r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m);
883 mempool_free(m, pool->mapping_pool);
884 DMERR_LIMIT("dm_kcopyd_zero() failed");
885 cell_error(pool, cell);
891 * A non-zero return indicates read_only or fail_io mode.
892 * Many callers don't care about the return value.
894 static int commit(struct pool *pool)
898 if (get_pool_mode(pool) != PM_WRITE)
901 r = dm_pool_commit_metadata(pool->pmd);
903 metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
908 static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks)
912 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
913 DMWARN("%s: reached low water mark for data device: sending event.",
914 dm_device_name(pool->pool_md));
915 spin_lock_irqsave(&pool->lock, flags);
916 pool->low_water_triggered = true;
917 spin_unlock_irqrestore(&pool->lock, flags);
918 dm_table_event(pool->ti->table);
922 static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
925 dm_block_t free_blocks;
926 struct pool *pool = tc->pool;
928 if (get_pool_mode(pool) != PM_WRITE)
931 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
933 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
937 check_low_water_mark(pool, free_blocks);
941 * Try to commit to see if that will free up some
948 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
950 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
955 out_of_data_space(pool);
960 r = dm_pool_alloc_data_block(pool->pmd, result);
962 metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
970 * If we have run out of space, queue bios until the device is
971 * resumed, presumably after having been reloaded with more space.
973 static void retry_on_resume(struct bio *bio)
975 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
976 struct thin_c *tc = h->tc;
977 struct pool *pool = tc->pool;
980 spin_lock_irqsave(&pool->lock, flags);
981 bio_list_add(&pool->retry_on_resume_list, bio);
982 spin_unlock_irqrestore(&pool->lock, flags);
985 static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
988 * When pool is read-only, no cell locking is needed because
989 * nothing is changing.
991 WARN_ON_ONCE(get_pool_mode(pool) != PM_READ_ONLY);
993 if (pool->pf.error_if_no_space)
996 retry_on_resume(bio);
999 static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell)
1002 struct bio_list bios;
1004 bio_list_init(&bios);
1005 cell_release(pool, cell, &bios);
1007 while ((bio = bio_list_pop(&bios)))
1008 handle_unserviceable_bio(pool, bio);
1011 static void process_discard(struct thin_c *tc, struct bio *bio)
1014 unsigned long flags;
1015 struct pool *pool = tc->pool;
1016 struct dm_bio_prison_cell *cell, *cell2;
1017 struct dm_cell_key key, key2;
1018 dm_block_t block = get_bio_block(tc, bio);
1019 struct dm_thin_lookup_result lookup_result;
1020 struct dm_thin_new_mapping *m;
1022 build_virtual_key(tc->td, block, &key);
1023 if (bio_detain(tc->pool, &key, bio, &cell))
1026 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1030 * Check nobody is fiddling with this pool block. This can
1031 * happen if someone's in the process of breaking sharing
1034 build_data_key(tc->td, lookup_result.block, &key2);
1035 if (bio_detain(tc->pool, &key2, bio, &cell2)) {
1036 cell_defer_no_holder(tc, cell);
1040 if (io_overlaps_block(pool, bio)) {
1042 * IO may still be going to the destination block. We must
1043 * quiesce before we can do the removal.
1045 m = get_next_mapping(pool);
1047 m->pass_discard = pool->pf.discard_passdown;
1048 m->definitely_not_shared = !lookup_result.shared;
1049 m->virt_block = block;
1050 m->data_block = lookup_result.block;
1055 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) {
1056 spin_lock_irqsave(&pool->lock, flags);
1057 list_add_tail(&m->list, &pool->prepared_discards);
1058 spin_unlock_irqrestore(&pool->lock, flags);
1062 inc_all_io_entry(pool, bio);
1063 cell_defer_no_holder(tc, cell);
1064 cell_defer_no_holder(tc, cell2);
1067 * The DM core makes sure that the discard doesn't span
1068 * a block boundary. So we submit the discard of a
1069 * partial block appropriately.
1071 if ((!lookup_result.shared) && pool->pf.discard_passdown)
1072 remap_and_issue(tc, bio, lookup_result.block);
1080 * It isn't provisioned, just forget it.
1082 cell_defer_no_holder(tc, cell);
1087 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1089 cell_defer_no_holder(tc, cell);
1095 static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
1096 struct dm_cell_key *key,
1097 struct dm_thin_lookup_result *lookup_result,
1098 struct dm_bio_prison_cell *cell)
1101 dm_block_t data_block;
1102 struct pool *pool = tc->pool;
1104 r = alloc_data_block(tc, &data_block);
1107 schedule_internal_copy(tc, block, lookup_result->block,
1108 data_block, cell, bio);
1112 retry_bios_on_resume(pool, cell);
1116 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1118 cell_error(pool, cell);
1123 static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1125 struct dm_thin_lookup_result *lookup_result)
1127 struct dm_bio_prison_cell *cell;
1128 struct pool *pool = tc->pool;
1129 struct dm_cell_key key;
1132 * If cell is already occupied, then sharing is already in the process
1133 * of being broken so we have nothing further to do here.
1135 build_data_key(tc->td, lookup_result->block, &key);
1136 if (bio_detain(pool, &key, bio, &cell))
1139 if (bio_data_dir(bio) == WRITE && bio->bi_size)
1140 break_sharing(tc, bio, block, &key, lookup_result, cell);
1142 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1144 h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
1145 inc_all_io_entry(pool, bio);
1146 cell_defer_no_holder(tc, cell);
1148 remap_and_issue(tc, bio, lookup_result->block);
1152 static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
1153 struct dm_bio_prison_cell *cell)
1156 dm_block_t data_block;
1157 struct pool *pool = tc->pool;
1160 * Remap empty bios (flushes) immediately, without provisioning.
1162 if (!bio->bi_size) {
1163 inc_all_io_entry(pool, bio);
1164 cell_defer_no_holder(tc, cell);
1166 remap_and_issue(tc, bio, 0);
1171 * Fill read bios with zeroes and complete them immediately.
1173 if (bio_data_dir(bio) == READ) {
1175 cell_defer_no_holder(tc, cell);
1180 r = alloc_data_block(tc, &data_block);
1184 schedule_external_copy(tc, block, data_block, cell, bio);
1186 schedule_zero(tc, block, data_block, cell, bio);
1190 retry_bios_on_resume(pool, cell);
1194 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1196 cell_error(pool, cell);
1201 static void process_bio(struct thin_c *tc, struct bio *bio)
1204 struct pool *pool = tc->pool;
1205 dm_block_t block = get_bio_block(tc, bio);
1206 struct dm_bio_prison_cell *cell;
1207 struct dm_cell_key key;
1208 struct dm_thin_lookup_result lookup_result;
1211 * If cell is already occupied, then the block is already
1212 * being provisioned so we have nothing further to do here.
1214 build_virtual_key(tc->td, block, &key);
1215 if (bio_detain(pool, &key, bio, &cell))
1218 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1221 if (lookup_result.shared) {
1222 process_shared_bio(tc, bio, block, &lookup_result);
1223 cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */
1225 inc_all_io_entry(pool, bio);
1226 cell_defer_no_holder(tc, cell);
1228 remap_and_issue(tc, bio, lookup_result.block);
1233 if (bio_data_dir(bio) == READ && tc->origin_dev) {
1234 inc_all_io_entry(pool, bio);
1235 cell_defer_no_holder(tc, cell);
1237 remap_to_origin_and_issue(tc, bio);
1239 provision_block(tc, bio, block, cell);
1243 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1245 cell_defer_no_holder(tc, cell);
1251 static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
1254 int rw = bio_data_dir(bio);
1255 dm_block_t block = get_bio_block(tc, bio);
1256 struct dm_thin_lookup_result lookup_result;
1258 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1261 if (lookup_result.shared && (rw == WRITE) && bio->bi_size)
1262 handle_unserviceable_bio(tc->pool, bio);
1264 inc_all_io_entry(tc->pool, bio);
1265 remap_and_issue(tc, bio, lookup_result.block);
1271 handle_unserviceable_bio(tc->pool, bio);
1275 if (tc->origin_dev) {
1276 inc_all_io_entry(tc->pool, bio);
1277 remap_to_origin_and_issue(tc, bio);
1286 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1293 static void process_bio_fail(struct thin_c *tc, struct bio *bio)
1299 * FIXME: should we also commit due to size of transaction, measured in
1302 static int need_commit_due_to_time(struct pool *pool)
1304 return jiffies < pool->last_commit_jiffies ||
1305 jiffies > pool->last_commit_jiffies + COMMIT_PERIOD;
1308 static void process_deferred_bios(struct pool *pool)
1310 unsigned long flags;
1312 struct bio_list bios;
1314 bio_list_init(&bios);
1316 spin_lock_irqsave(&pool->lock, flags);
1317 bio_list_merge(&bios, &pool->deferred_bios);
1318 bio_list_init(&pool->deferred_bios);
1319 spin_unlock_irqrestore(&pool->lock, flags);
1321 while ((bio = bio_list_pop(&bios))) {
1322 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1323 struct thin_c *tc = h->tc;
1326 * If we've got no free new_mapping structs, and processing
1327 * this bio might require one, we pause until there are some
1328 * prepared mappings to process.
1330 if (ensure_next_mapping(pool)) {
1331 spin_lock_irqsave(&pool->lock, flags);
1332 bio_list_merge(&pool->deferred_bios, &bios);
1333 spin_unlock_irqrestore(&pool->lock, flags);
1338 if (bio->bi_rw & REQ_DISCARD)
1339 pool->process_discard(tc, bio);
1341 pool->process_bio(tc, bio);
1345 * If there are any deferred flush bios, we must commit
1346 * the metadata before issuing them.
1348 bio_list_init(&bios);
1349 spin_lock_irqsave(&pool->lock, flags);
1350 bio_list_merge(&bios, &pool->deferred_flush_bios);
1351 bio_list_init(&pool->deferred_flush_bios);
1352 spin_unlock_irqrestore(&pool->lock, flags);
1354 if (bio_list_empty(&bios) && !need_commit_due_to_time(pool))
1358 while ((bio = bio_list_pop(&bios)))
1362 pool->last_commit_jiffies = jiffies;
1364 while ((bio = bio_list_pop(&bios)))
1365 generic_make_request(bio);
1368 static void do_worker(struct work_struct *ws)
1370 struct pool *pool = container_of(ws, struct pool, worker);
1372 process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
1373 process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
1374 process_deferred_bios(pool);
1378 * We want to commit periodically so that not too much
1379 * unwritten data builds up.
1381 static void do_waker(struct work_struct *ws)
1383 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
1385 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
1388 /*----------------------------------------------------------------*/
1390 static enum pool_mode get_pool_mode(struct pool *pool)
1392 return pool->pf.mode;
1395 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode)
1398 enum pool_mode old_mode = pool->pf.mode;
1402 if (old_mode != new_mode)
1403 DMERR("%s: switching pool to failure mode",
1404 dm_device_name(pool->pool_md));
1405 dm_pool_metadata_read_only(pool->pmd);
1406 pool->process_bio = process_bio_fail;
1407 pool->process_discard = process_bio_fail;
1408 pool->process_prepared_mapping = process_prepared_mapping_fail;
1409 pool->process_prepared_discard = process_prepared_discard_fail;
1413 if (old_mode != new_mode)
1414 DMERR("%s: switching pool to read-only mode",
1415 dm_device_name(pool->pool_md));
1416 r = dm_pool_abort_metadata(pool->pmd);
1418 DMERR("%s: aborting transaction failed",
1419 dm_device_name(pool->pool_md));
1421 set_pool_mode(pool, new_mode);
1423 dm_pool_metadata_read_only(pool->pmd);
1424 pool->process_bio = process_bio_read_only;
1425 pool->process_discard = process_discard;
1426 pool->process_prepared_mapping = process_prepared_mapping_fail;
1427 pool->process_prepared_discard = process_prepared_discard_passdown;
1432 if (old_mode != new_mode)
1433 DMINFO("%s: switching pool to write mode",
1434 dm_device_name(pool->pool_md));
1435 dm_pool_metadata_read_write(pool->pmd);
1436 pool->process_bio = process_bio;
1437 pool->process_discard = process_discard;
1438 pool->process_prepared_mapping = process_prepared_mapping;
1439 pool->process_prepared_discard = process_prepared_discard;
1443 pool->pf.mode = new_mode;
1447 * Rather than calling set_pool_mode directly, use these which describe the
1448 * reason for mode degradation.
1450 static void out_of_data_space(struct pool *pool)
1452 DMERR_LIMIT("%s: no free data space available.",
1453 dm_device_name(pool->pool_md));
1454 set_pool_mode(pool, PM_READ_ONLY);
1457 static void metadata_operation_failed(struct pool *pool, const char *op, int r)
1459 dm_block_t free_blocks;
1461 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
1462 dm_device_name(pool->pool_md), op, r);
1465 !dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks) &&
1467 DMERR_LIMIT("%s: no free metadata space available.",
1468 dm_device_name(pool->pool_md));
1470 set_pool_mode(pool, PM_READ_ONLY);
1473 /*----------------------------------------------------------------*/
1476 * Mapping functions.
1480 * Called only while mapping a thin bio to hand it over to the workqueue.
1482 static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
1484 unsigned long flags;
1485 struct pool *pool = tc->pool;
1487 spin_lock_irqsave(&pool->lock, flags);
1488 bio_list_add(&pool->deferred_bios, bio);
1489 spin_unlock_irqrestore(&pool->lock, flags);
1494 static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
1496 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1499 h->shared_read_entry = NULL;
1500 h->all_io_entry = NULL;
1501 h->overwrite_mapping = NULL;
1505 * Non-blocking function called from the thin target's map function.
1507 static int thin_bio_map(struct dm_target *ti, struct bio *bio)
1510 struct thin_c *tc = ti->private;
1511 dm_block_t block = get_bio_block(tc, bio);
1512 struct dm_thin_device *td = tc->td;
1513 struct dm_thin_lookup_result result;
1514 struct dm_bio_prison_cell cell1, cell2;
1515 struct dm_bio_prison_cell *cell_result;
1516 struct dm_cell_key key;
1518 thin_hook_bio(tc, bio);
1520 if (get_pool_mode(tc->pool) == PM_FAIL) {
1522 return DM_MAPIO_SUBMITTED;
1525 if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) {
1526 thin_defer_bio(tc, bio);
1527 return DM_MAPIO_SUBMITTED;
1530 r = dm_thin_find_block(td, block, 0, &result);
1533 * Note that we defer readahead too.
1537 if (unlikely(result.shared)) {
1539 * We have a race condition here between the
1540 * result.shared value returned by the lookup and
1541 * snapshot creation, which may cause new
1544 * To avoid this always quiesce the origin before
1545 * taking the snap. You want to do this anyway to
1546 * ensure a consistent application view
1549 * More distant ancestors are irrelevant. The
1550 * shared flag will be set in their case.
1552 thin_defer_bio(tc, bio);
1553 return DM_MAPIO_SUBMITTED;
1556 build_virtual_key(tc->td, block, &key);
1557 if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result))
1558 return DM_MAPIO_SUBMITTED;
1560 build_data_key(tc->td, result.block, &key);
1561 if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) {
1562 cell_defer_no_holder_no_free(tc, &cell1);
1563 return DM_MAPIO_SUBMITTED;
1566 inc_all_io_entry(tc->pool, bio);
1567 cell_defer_no_holder_no_free(tc, &cell2);
1568 cell_defer_no_holder_no_free(tc, &cell1);
1570 remap(tc, bio, result.block);
1571 return DM_MAPIO_REMAPPED;
1574 if (get_pool_mode(tc->pool) == PM_READ_ONLY) {
1576 * This block isn't provisioned, and we have no way
1579 handle_unserviceable_bio(tc->pool, bio);
1580 return DM_MAPIO_SUBMITTED;
1586 * In future, the failed dm_thin_find_block above could
1587 * provide the hint to load the metadata into cache.
1589 thin_defer_bio(tc, bio);
1590 return DM_MAPIO_SUBMITTED;
1594 * Must always call bio_io_error on failure.
1595 * dm_thin_find_block can fail with -EINVAL if the
1596 * pool is switched to fail-io mode.
1599 return DM_MAPIO_SUBMITTED;
1603 static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
1606 unsigned long flags;
1607 struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
1609 spin_lock_irqsave(&pt->pool->lock, flags);
1610 r = !bio_list_empty(&pt->pool->retry_on_resume_list);
1611 spin_unlock_irqrestore(&pt->pool->lock, flags);
1614 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1615 r = bdi_congested(&q->backing_dev_info, bdi_bits);
1621 static void __requeue_bios(struct pool *pool)
1623 bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list);
1624 bio_list_init(&pool->retry_on_resume_list);
1627 /*----------------------------------------------------------------
1628 * Binding of control targets to a pool object
1629 *--------------------------------------------------------------*/
1630 static bool data_dev_supports_discard(struct pool_c *pt)
1632 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1634 return q && blk_queue_discard(q);
1637 static bool is_factor(sector_t block_size, uint32_t n)
1639 return !sector_div(block_size, n);
1643 * If discard_passdown was enabled verify that the data device
1644 * supports discards. Disable discard_passdown if not.
1646 static void disable_passdown_if_not_supported(struct pool_c *pt)
1648 struct pool *pool = pt->pool;
1649 struct block_device *data_bdev = pt->data_dev->bdev;
1650 struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
1651 sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT;
1652 const char *reason = NULL;
1653 char buf[BDEVNAME_SIZE];
1655 if (!pt->adjusted_pf.discard_passdown)
1658 if (!data_dev_supports_discard(pt))
1659 reason = "discard unsupported";
1661 else if (data_limits->max_discard_sectors < pool->sectors_per_block)
1662 reason = "max discard sectors smaller than a block";
1664 else if (data_limits->discard_granularity > block_size)
1665 reason = "discard granularity larger than a block";
1667 else if (!is_factor(block_size, data_limits->discard_granularity))
1668 reason = "discard granularity not a factor of block size";
1671 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason);
1672 pt->adjusted_pf.discard_passdown = false;
1676 static int bind_control_target(struct pool *pool, struct dm_target *ti)
1678 struct pool_c *pt = ti->private;
1681 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
1683 enum pool_mode old_mode = pool->pf.mode;
1684 enum pool_mode new_mode = pt->adjusted_pf.mode;
1687 * Don't change the pool's mode until set_pool_mode() below.
1688 * Otherwise the pool's process_* function pointers may
1689 * not match the desired pool mode.
1691 pt->adjusted_pf.mode = old_mode;
1694 pool->pf = pt->adjusted_pf;
1695 pool->low_water_blocks = pt->low_water_blocks;
1698 * If we were in PM_FAIL mode, rollback of metadata failed. We're
1699 * not going to recover without a thin_repair. So we never let the
1700 * pool move out of the old mode. On the other hand a PM_READ_ONLY
1701 * may have been due to a lack of metadata or data space, and may
1702 * now work (ie. if the underlying devices have been resized).
1704 if (old_mode == PM_FAIL)
1705 new_mode = old_mode;
1707 set_pool_mode(pool, new_mode);
1712 static void unbind_control_target(struct pool *pool, struct dm_target *ti)
1718 /*----------------------------------------------------------------
1720 *--------------------------------------------------------------*/
1721 /* Initialize pool features. */
1722 static void pool_features_init(struct pool_features *pf)
1724 pf->mode = PM_WRITE;
1725 pf->zero_new_blocks = true;
1726 pf->discard_enabled = true;
1727 pf->discard_passdown = true;
1728 pf->error_if_no_space = false;
1731 static void __pool_destroy(struct pool *pool)
1733 __pool_table_remove(pool);
1735 if (dm_pool_metadata_close(pool->pmd) < 0)
1736 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1738 dm_bio_prison_destroy(pool->prison);
1739 dm_kcopyd_client_destroy(pool->copier);
1742 destroy_workqueue(pool->wq);
1744 if (pool->next_mapping)
1745 mempool_free(pool->next_mapping, pool->mapping_pool);
1746 mempool_destroy(pool->mapping_pool);
1747 dm_deferred_set_destroy(pool->shared_read_ds);
1748 dm_deferred_set_destroy(pool->all_io_ds);
1752 static struct kmem_cache *_new_mapping_cache;
1754 static struct pool *pool_create(struct mapped_device *pool_md,
1755 struct block_device *metadata_dev,
1756 unsigned long block_size,
1757 int read_only, char **error)
1762 struct dm_pool_metadata *pmd;
1763 bool format_device = read_only ? false : true;
1765 pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
1767 *error = "Error creating metadata object";
1768 return (struct pool *)pmd;
1771 pool = kmalloc(sizeof(*pool), GFP_KERNEL);
1773 *error = "Error allocating memory for pool";
1774 err_p = ERR_PTR(-ENOMEM);
1779 pool->sectors_per_block = block_size;
1780 if (block_size & (block_size - 1))
1781 pool->sectors_per_block_shift = -1;
1783 pool->sectors_per_block_shift = __ffs(block_size);
1784 pool->low_water_blocks = 0;
1785 pool_features_init(&pool->pf);
1786 pool->prison = dm_bio_prison_create(PRISON_CELLS);
1787 if (!pool->prison) {
1788 *error = "Error creating pool's bio prison";
1789 err_p = ERR_PTR(-ENOMEM);
1793 pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
1794 if (IS_ERR(pool->copier)) {
1795 r = PTR_ERR(pool->copier);
1796 *error = "Error creating pool's kcopyd client";
1798 goto bad_kcopyd_client;
1802 * Create singlethreaded workqueue that will service all devices
1803 * that use this metadata.
1805 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
1807 *error = "Error creating pool's workqueue";
1808 err_p = ERR_PTR(-ENOMEM);
1812 INIT_WORK(&pool->worker, do_worker);
1813 INIT_DELAYED_WORK(&pool->waker, do_waker);
1814 spin_lock_init(&pool->lock);
1815 bio_list_init(&pool->deferred_bios);
1816 bio_list_init(&pool->deferred_flush_bios);
1817 INIT_LIST_HEAD(&pool->prepared_mappings);
1818 INIT_LIST_HEAD(&pool->prepared_discards);
1819 pool->low_water_triggered = false;
1820 bio_list_init(&pool->retry_on_resume_list);
1822 pool->shared_read_ds = dm_deferred_set_create();
1823 if (!pool->shared_read_ds) {
1824 *error = "Error creating pool's shared read deferred set";
1825 err_p = ERR_PTR(-ENOMEM);
1826 goto bad_shared_read_ds;
1829 pool->all_io_ds = dm_deferred_set_create();
1830 if (!pool->all_io_ds) {
1831 *error = "Error creating pool's all io deferred set";
1832 err_p = ERR_PTR(-ENOMEM);
1836 pool->next_mapping = NULL;
1837 pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE,
1838 _new_mapping_cache);
1839 if (!pool->mapping_pool) {
1840 *error = "Error creating pool's mapping mempool";
1841 err_p = ERR_PTR(-ENOMEM);
1842 goto bad_mapping_pool;
1845 pool->ref_count = 1;
1846 pool->last_commit_jiffies = jiffies;
1847 pool->pool_md = pool_md;
1848 pool->md_dev = metadata_dev;
1849 __pool_table_insert(pool);
1854 dm_deferred_set_destroy(pool->all_io_ds);
1856 dm_deferred_set_destroy(pool->shared_read_ds);
1858 destroy_workqueue(pool->wq);
1860 dm_kcopyd_client_destroy(pool->copier);
1862 dm_bio_prison_destroy(pool->prison);
1866 if (dm_pool_metadata_close(pmd))
1867 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1872 static void __pool_inc(struct pool *pool)
1874 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1878 static void __pool_dec(struct pool *pool)
1880 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1881 BUG_ON(!pool->ref_count);
1882 if (!--pool->ref_count)
1883 __pool_destroy(pool);
1886 static struct pool *__pool_find(struct mapped_device *pool_md,
1887 struct block_device *metadata_dev,
1888 unsigned long block_size, int read_only,
1889 char **error, int *created)
1891 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
1894 if (pool->pool_md != pool_md) {
1895 *error = "metadata device already in use by a pool";
1896 return ERR_PTR(-EBUSY);
1901 pool = __pool_table_lookup(pool_md);
1903 if (pool->md_dev != metadata_dev) {
1904 *error = "different pool cannot replace a pool";
1905 return ERR_PTR(-EINVAL);
1910 pool = pool_create(pool_md, metadata_dev, block_size, read_only, error);
1918 /*----------------------------------------------------------------
1919 * Pool target methods
1920 *--------------------------------------------------------------*/
1921 static void pool_dtr(struct dm_target *ti)
1923 struct pool_c *pt = ti->private;
1925 mutex_lock(&dm_thin_pool_table.mutex);
1927 unbind_control_target(pt->pool, ti);
1928 __pool_dec(pt->pool);
1929 dm_put_device(ti, pt->metadata_dev);
1930 dm_put_device(ti, pt->data_dev);
1933 mutex_unlock(&dm_thin_pool_table.mutex);
1936 static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
1937 struct dm_target *ti)
1941 const char *arg_name;
1943 static struct dm_arg _args[] = {
1944 {0, 4, "Invalid number of pool feature arguments"},
1948 * No feature arguments supplied.
1953 r = dm_read_arg_group(_args, as, &argc, &ti->error);
1957 while (argc && !r) {
1958 arg_name = dm_shift_arg(as);
1961 if (!strcasecmp(arg_name, "skip_block_zeroing"))
1962 pf->zero_new_blocks = false;
1964 else if (!strcasecmp(arg_name, "ignore_discard"))
1965 pf->discard_enabled = false;
1967 else if (!strcasecmp(arg_name, "no_discard_passdown"))
1968 pf->discard_passdown = false;
1970 else if (!strcasecmp(arg_name, "read_only"))
1971 pf->mode = PM_READ_ONLY;
1973 else if (!strcasecmp(arg_name, "error_if_no_space"))
1974 pf->error_if_no_space = true;
1977 ti->error = "Unrecognised pool feature requested";
1986 static void metadata_low_callback(void *context)
1988 struct pool *pool = context;
1990 DMWARN("%s: reached low water mark for metadata device: sending event.",
1991 dm_device_name(pool->pool_md));
1993 dm_table_event(pool->ti->table);
1996 static sector_t get_metadata_dev_size(struct block_device *bdev)
1998 sector_t metadata_dev_size = i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
1999 char buffer[BDEVNAME_SIZE];
2001 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) {
2002 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
2003 bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS);
2004 metadata_dev_size = THIN_METADATA_MAX_SECTORS_WARNING;
2007 return metadata_dev_size;
2010 static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
2012 sector_t metadata_dev_size = get_metadata_dev_size(bdev);
2014 sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
2016 return metadata_dev_size;
2020 * When a metadata threshold is crossed a dm event is triggered, and
2021 * userland should respond by growing the metadata device. We could let
2022 * userland set the threshold, like we do with the data threshold, but I'm
2023 * not sure they know enough to do this well.
2025 static dm_block_t calc_metadata_threshold(struct pool_c *pt)
2028 * 4M is ample for all ops with the possible exception of thin
2029 * device deletion which is harmless if it fails (just retry the
2030 * delete after you've grown the device).
2032 dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
2033 return min((dm_block_t)1024ULL /* 4M */, quarter);
2037 * thin-pool <metadata dev> <data dev>
2038 * <data block size (sectors)>
2039 * <low water mark (blocks)>
2040 * [<#feature args> [<arg>]*]
2042 * Optional feature arguments are:
2043 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
2044 * ignore_discard: disable discard
2045 * no_discard_passdown: don't pass discards down to the data device
2046 * read_only: Don't allow any changes to be made to the pool metadata.
2047 * error_if_no_space: error IOs, instead of queueing, if no space.
2049 static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
2051 int r, pool_created = 0;
2054 struct pool_features pf;
2055 struct dm_arg_set as;
2056 struct dm_dev *data_dev;
2057 unsigned long block_size;
2058 dm_block_t low_water_blocks;
2059 struct dm_dev *metadata_dev;
2060 fmode_t metadata_mode;
2063 * FIXME Remove validation from scope of lock.
2065 mutex_lock(&dm_thin_pool_table.mutex);
2068 ti->error = "Invalid argument count";
2077 * Set default pool features.
2079 pool_features_init(&pf);
2081 dm_consume_args(&as, 4);
2082 r = parse_pool_features(&as, &pf, ti);
2086 metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
2087 r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
2089 ti->error = "Error opening metadata block device";
2094 * Run for the side-effect of possibly issuing a warning if the
2095 * device is too big.
2097 (void) get_metadata_dev_size(metadata_dev->bdev);
2099 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
2101 ti->error = "Error getting data device";
2105 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
2106 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
2107 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
2108 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
2109 ti->error = "Invalid block size";
2114 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
2115 ti->error = "Invalid low water mark";
2120 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
2126 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
2127 block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
2134 * 'pool_created' reflects whether this is the first table load.
2135 * Top level discard support is not allowed to be changed after
2136 * initial load. This would require a pool reload to trigger thin
2139 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
2140 ti->error = "Discard support cannot be disabled once enabled";
2142 goto out_flags_changed;
2147 pt->metadata_dev = metadata_dev;
2148 pt->data_dev = data_dev;
2149 pt->low_water_blocks = low_water_blocks;
2150 pt->adjusted_pf = pt->requested_pf = pf;
2151 ti->num_flush_bios = 1;
2154 * Only need to enable discards if the pool should pass
2155 * them down to the data device. The thin device's discard
2156 * processing will cause mappings to be removed from the btree.
2158 ti->discard_zeroes_data_unsupported = true;
2159 if (pf.discard_enabled && pf.discard_passdown) {
2160 ti->num_discard_bios = 1;
2163 * Setting 'discards_supported' circumvents the normal
2164 * stacking of discard limits (this keeps the pool and
2165 * thin devices' discard limits consistent).
2167 ti->discards_supported = true;
2171 r = dm_pool_register_metadata_threshold(pt->pool->pmd,
2172 calc_metadata_threshold(pt),
2173 metadata_low_callback,
2178 pt->callbacks.congested_fn = pool_is_congested;
2179 dm_table_add_target_callbacks(ti->table, &pt->callbacks);
2181 mutex_unlock(&dm_thin_pool_table.mutex);
2190 dm_put_device(ti, data_dev);
2192 dm_put_device(ti, metadata_dev);
2194 mutex_unlock(&dm_thin_pool_table.mutex);
2199 static int pool_map(struct dm_target *ti, struct bio *bio)
2202 struct pool_c *pt = ti->private;
2203 struct pool *pool = pt->pool;
2204 unsigned long flags;
2207 * As this is a singleton target, ti->begin is always zero.
2209 spin_lock_irqsave(&pool->lock, flags);
2210 bio->bi_bdev = pt->data_dev->bdev;
2211 r = DM_MAPIO_REMAPPED;
2212 spin_unlock_irqrestore(&pool->lock, flags);
2217 static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
2220 struct pool_c *pt = ti->private;
2221 struct pool *pool = pt->pool;
2222 sector_t data_size = ti->len;
2223 dm_block_t sb_data_size;
2225 *need_commit = false;
2227 (void) sector_div(data_size, pool->sectors_per_block);
2229 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
2231 DMERR("%s: failed to retrieve data device size",
2232 dm_device_name(pool->pool_md));
2236 if (data_size < sb_data_size) {
2237 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
2238 dm_device_name(pool->pool_md),
2239 (unsigned long long)data_size, sb_data_size);
2242 } else if (data_size > sb_data_size) {
2244 DMINFO("%s: growing the data device from %llu to %llu blocks",
2245 dm_device_name(pool->pool_md),
2246 sb_data_size, (unsigned long long)data_size);
2247 r = dm_pool_resize_data_dev(pool->pmd, data_size);
2249 metadata_operation_failed(pool, "dm_pool_resize_data_dev", r);
2253 *need_commit = true;
2259 static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
2262 struct pool_c *pt = ti->private;
2263 struct pool *pool = pt->pool;
2264 dm_block_t metadata_dev_size, sb_metadata_dev_size;
2266 *need_commit = false;
2268 metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
2270 r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
2272 DMERR("%s: failed to retrieve metadata device size",
2273 dm_device_name(pool->pool_md));
2277 if (metadata_dev_size < sb_metadata_dev_size) {
2278 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
2279 dm_device_name(pool->pool_md),
2280 metadata_dev_size, sb_metadata_dev_size);
2283 } else if (metadata_dev_size > sb_metadata_dev_size) {
2284 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
2285 dm_device_name(pool->pool_md),
2286 sb_metadata_dev_size, metadata_dev_size);
2287 r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
2289 metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r);
2293 *need_commit = true;
2300 * Retrieves the number of blocks of the data device from
2301 * the superblock and compares it to the actual device size,
2302 * thus resizing the data device in case it has grown.
2304 * This both copes with opening preallocated data devices in the ctr
2305 * being followed by a resume
2307 * calling the resume method individually after userspace has
2308 * grown the data device in reaction to a table event.
2310 static int pool_preresume(struct dm_target *ti)
2313 bool need_commit1, need_commit2;
2314 struct pool_c *pt = ti->private;
2315 struct pool *pool = pt->pool;
2318 * Take control of the pool object.
2320 r = bind_control_target(pool, ti);
2324 r = maybe_resize_data_dev(ti, &need_commit1);
2328 r = maybe_resize_metadata_dev(ti, &need_commit2);
2332 if (need_commit1 || need_commit2)
2333 (void) commit(pool);
2338 static void pool_resume(struct dm_target *ti)
2340 struct pool_c *pt = ti->private;
2341 struct pool *pool = pt->pool;
2342 unsigned long flags;
2344 spin_lock_irqsave(&pool->lock, flags);
2345 pool->low_water_triggered = false;
2346 __requeue_bios(pool);
2347 spin_unlock_irqrestore(&pool->lock, flags);
2349 do_waker(&pool->waker.work);
2352 static void pool_postsuspend(struct dm_target *ti)
2354 struct pool_c *pt = ti->private;
2355 struct pool *pool = pt->pool;
2357 cancel_delayed_work(&pool->waker);
2358 flush_workqueue(pool->wq);
2359 (void) commit(pool);
2362 static int check_arg_count(unsigned argc, unsigned args_required)
2364 if (argc != args_required) {
2365 DMWARN("Message received with %u arguments instead of %u.",
2366 argc, args_required);
2373 static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
2375 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
2376 *dev_id <= MAX_DEV_ID)
2380 DMWARN("Message received with invalid device id: %s", arg);
2385 static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
2390 r = check_arg_count(argc, 2);
2394 r = read_dev_id(argv[1], &dev_id, 1);
2398 r = dm_pool_create_thin(pool->pmd, dev_id);
2400 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2408 static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2411 dm_thin_id origin_dev_id;
2414 r = check_arg_count(argc, 3);
2418 r = read_dev_id(argv[1], &dev_id, 1);
2422 r = read_dev_id(argv[2], &origin_dev_id, 1);
2426 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
2428 DMWARN("Creation of new snapshot %s of device %s failed.",
2436 static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
2441 r = check_arg_count(argc, 2);
2445 r = read_dev_id(argv[1], &dev_id, 1);
2449 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
2451 DMWARN("Deletion of thin device %s failed.", argv[1]);
2456 static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
2458 dm_thin_id old_id, new_id;
2461 r = check_arg_count(argc, 3);
2465 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
2466 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
2470 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
2471 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
2475 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
2477 DMWARN("Failed to change transaction id from %s to %s.",
2485 static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2489 r = check_arg_count(argc, 1);
2493 (void) commit(pool);
2495 r = dm_pool_reserve_metadata_snap(pool->pmd);
2497 DMWARN("reserve_metadata_snap message failed.");
2502 static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2506 r = check_arg_count(argc, 1);
2510 r = dm_pool_release_metadata_snap(pool->pmd);
2512 DMWARN("release_metadata_snap message failed.");
2518 * Messages supported:
2519 * create_thin <dev_id>
2520 * create_snap <dev_id> <origin_id>
2522 * trim <dev_id> <new_size_in_sectors>
2523 * set_transaction_id <current_trans_id> <new_trans_id>
2524 * reserve_metadata_snap
2525 * release_metadata_snap
2527 static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
2530 struct pool_c *pt = ti->private;
2531 struct pool *pool = pt->pool;
2533 if (!strcasecmp(argv[0], "create_thin"))
2534 r = process_create_thin_mesg(argc, argv, pool);
2536 else if (!strcasecmp(argv[0], "create_snap"))
2537 r = process_create_snap_mesg(argc, argv, pool);
2539 else if (!strcasecmp(argv[0], "delete"))
2540 r = process_delete_mesg(argc, argv, pool);
2542 else if (!strcasecmp(argv[0], "set_transaction_id"))
2543 r = process_set_transaction_id_mesg(argc, argv, pool);
2545 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
2546 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
2548 else if (!strcasecmp(argv[0], "release_metadata_snap"))
2549 r = process_release_metadata_snap_mesg(argc, argv, pool);
2552 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
2555 (void) commit(pool);
2560 static void emit_flags(struct pool_features *pf, char *result,
2561 unsigned sz, unsigned maxlen)
2563 unsigned count = !pf->zero_new_blocks + !pf->discard_enabled +
2564 !pf->discard_passdown + (pf->mode == PM_READ_ONLY) +
2565 pf->error_if_no_space;
2566 DMEMIT("%u ", count);
2568 if (!pf->zero_new_blocks)
2569 DMEMIT("skip_block_zeroing ");
2571 if (!pf->discard_enabled)
2572 DMEMIT("ignore_discard ");
2574 if (!pf->discard_passdown)
2575 DMEMIT("no_discard_passdown ");
2577 if (pf->mode == PM_READ_ONLY)
2578 DMEMIT("read_only ");
2580 if (pf->error_if_no_space)
2581 DMEMIT("error_if_no_space ");
2586 * <transaction id> <used metadata sectors>/<total metadata sectors>
2587 * <used data sectors>/<total data sectors> <held metadata root>
2589 static void pool_status(struct dm_target *ti, status_type_t type,
2590 unsigned status_flags, char *result, unsigned maxlen)
2594 uint64_t transaction_id;
2595 dm_block_t nr_free_blocks_data;
2596 dm_block_t nr_free_blocks_metadata;
2597 dm_block_t nr_blocks_data;
2598 dm_block_t nr_blocks_metadata;
2599 dm_block_t held_root;
2600 char buf[BDEVNAME_SIZE];
2601 char buf2[BDEVNAME_SIZE];
2602 struct pool_c *pt = ti->private;
2603 struct pool *pool = pt->pool;
2606 case STATUSTYPE_INFO:
2607 if (get_pool_mode(pool) == PM_FAIL) {
2612 /* Commit to ensure statistics aren't out-of-date */
2613 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
2614 (void) commit(pool);
2616 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
2618 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
2619 dm_device_name(pool->pool_md), r);
2623 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
2625 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
2626 dm_device_name(pool->pool_md), r);
2630 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
2632 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
2633 dm_device_name(pool->pool_md), r);
2637 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
2639 DMERR("%s: dm_pool_get_free_block_count returned %d",
2640 dm_device_name(pool->pool_md), r);
2644 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
2646 DMERR("%s: dm_pool_get_data_dev_size returned %d",
2647 dm_device_name(pool->pool_md), r);
2651 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
2653 DMERR("%s: dm_pool_get_metadata_snap returned %d",
2654 dm_device_name(pool->pool_md), r);
2658 DMEMIT("%llu %llu/%llu %llu/%llu ",
2659 (unsigned long long)transaction_id,
2660 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
2661 (unsigned long long)nr_blocks_metadata,
2662 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
2663 (unsigned long long)nr_blocks_data);
2666 DMEMIT("%llu ", held_root);
2670 if (pool->pf.mode == PM_READ_ONLY)
2675 if (!pool->pf.discard_enabled)
2676 DMEMIT("ignore_discard ");
2677 else if (pool->pf.discard_passdown)
2678 DMEMIT("discard_passdown ");
2680 DMEMIT("no_discard_passdown ");
2682 if (pool->pf.error_if_no_space)
2683 DMEMIT("error_if_no_space ");
2685 DMEMIT("queue_if_no_space ");
2689 case STATUSTYPE_TABLE:
2690 DMEMIT("%s %s %lu %llu ",
2691 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
2692 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
2693 (unsigned long)pool->sectors_per_block,
2694 (unsigned long long)pt->low_water_blocks);
2695 emit_flags(&pt->requested_pf, result, sz, maxlen);
2704 static int pool_iterate_devices(struct dm_target *ti,
2705 iterate_devices_callout_fn fn, void *data)
2707 struct pool_c *pt = ti->private;
2709 return fn(ti, pt->data_dev, 0, ti->len, data);
2712 static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2713 struct bio_vec *biovec, int max_size)
2715 struct pool_c *pt = ti->private;
2716 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
2718 if (!q->merge_bvec_fn)
2721 bvm->bi_bdev = pt->data_dev->bdev;
2723 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2726 static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits)
2728 struct pool *pool = pt->pool;
2729 struct queue_limits *data_limits;
2731 limits->max_discard_sectors = pool->sectors_per_block;
2734 * discard_granularity is just a hint, and not enforced.
2736 if (pt->adjusted_pf.discard_passdown) {
2737 data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits;
2738 limits->discard_granularity = data_limits->discard_granularity;
2740 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
2743 static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
2745 struct pool_c *pt = ti->private;
2746 struct pool *pool = pt->pool;
2747 uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
2750 * If the system-determined stacked limits are compatible with the
2751 * pool's blocksize (io_opt is a factor) do not override them.
2753 if (io_opt_sectors < pool->sectors_per_block ||
2754 do_div(io_opt_sectors, pool->sectors_per_block)) {
2755 blk_limits_io_min(limits, 0);
2756 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
2760 * pt->adjusted_pf is a staging area for the actual features to use.
2761 * They get transferred to the live pool in bind_control_target()
2762 * called from pool_preresume().
2764 if (!pt->adjusted_pf.discard_enabled) {
2766 * Must explicitly disallow stacking discard limits otherwise the
2767 * block layer will stack them if pool's data device has support.
2768 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
2769 * user to see that, so make sure to set all discard limits to 0.
2771 limits->discard_granularity = 0;
2775 disable_passdown_if_not_supported(pt);
2777 set_discard_limits(pt, limits);
2780 static struct target_type pool_target = {
2781 .name = "thin-pool",
2782 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
2783 DM_TARGET_IMMUTABLE,
2784 .version = {1, 10, 0},
2785 .module = THIS_MODULE,
2789 .postsuspend = pool_postsuspend,
2790 .preresume = pool_preresume,
2791 .resume = pool_resume,
2792 .message = pool_message,
2793 .status = pool_status,
2794 .merge = pool_merge,
2795 .iterate_devices = pool_iterate_devices,
2796 .io_hints = pool_io_hints,
2799 /*----------------------------------------------------------------
2800 * Thin target methods
2801 *--------------------------------------------------------------*/
2802 static void thin_dtr(struct dm_target *ti)
2804 struct thin_c *tc = ti->private;
2806 mutex_lock(&dm_thin_pool_table.mutex);
2808 __pool_dec(tc->pool);
2809 dm_pool_close_thin_device(tc->td);
2810 dm_put_device(ti, tc->pool_dev);
2812 dm_put_device(ti, tc->origin_dev);
2815 mutex_unlock(&dm_thin_pool_table.mutex);
2819 * Thin target parameters:
2821 * <pool_dev> <dev_id> [origin_dev]
2823 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2824 * dev_id: the internal device identifier
2825 * origin_dev: a device external to the pool that should act as the origin
2827 * If the pool device has discards disabled, they get disabled for the thin
2830 static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
2834 struct dm_dev *pool_dev, *origin_dev;
2835 struct mapped_device *pool_md;
2837 mutex_lock(&dm_thin_pool_table.mutex);
2839 if (argc != 2 && argc != 3) {
2840 ti->error = "Invalid argument count";
2845 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
2847 ti->error = "Out of memory";
2853 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
2855 ti->error = "Error opening origin device";
2856 goto bad_origin_dev;
2858 tc->origin_dev = origin_dev;
2861 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
2863 ti->error = "Error opening pool device";
2866 tc->pool_dev = pool_dev;
2868 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
2869 ti->error = "Invalid device id";
2874 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
2876 ti->error = "Couldn't get pool mapped device";
2881 tc->pool = __pool_table_lookup(pool_md);
2883 ti->error = "Couldn't find pool object";
2885 goto bad_pool_lookup;
2887 __pool_inc(tc->pool);
2889 if (get_pool_mode(tc->pool) == PM_FAIL) {
2890 ti->error = "Couldn't open thin device, Pool is in fail mode";
2894 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
2896 ti->error = "Couldn't open thin internal device";
2900 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
2904 ti->num_flush_bios = 1;
2905 ti->flush_supported = true;
2906 ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook);
2908 /* In case the pool supports discards, pass them on. */
2909 ti->discard_zeroes_data_unsupported = true;
2910 if (tc->pool->pf.discard_enabled) {
2911 ti->discards_supported = true;
2912 ti->num_discard_bios = 1;
2913 /* Discard bios must be split on a block boundary */
2914 ti->split_discard_bios = true;
2919 mutex_unlock(&dm_thin_pool_table.mutex);
2924 __pool_dec(tc->pool);
2928 dm_put_device(ti, tc->pool_dev);
2931 dm_put_device(ti, tc->origin_dev);
2935 mutex_unlock(&dm_thin_pool_table.mutex);
2940 static int thin_map(struct dm_target *ti, struct bio *bio)
2942 bio->bi_sector = dm_target_offset(ti, bio->bi_sector);
2944 return thin_bio_map(ti, bio);
2947 static int thin_endio(struct dm_target *ti, struct bio *bio, int err)
2949 unsigned long flags;
2950 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2951 struct list_head work;
2952 struct dm_thin_new_mapping *m, *tmp;
2953 struct pool *pool = h->tc->pool;
2955 if (h->shared_read_entry) {
2956 INIT_LIST_HEAD(&work);
2957 dm_deferred_entry_dec(h->shared_read_entry, &work);
2959 spin_lock_irqsave(&pool->lock, flags);
2960 list_for_each_entry_safe(m, tmp, &work, list) {
2963 __maybe_add_mapping(m);
2965 spin_unlock_irqrestore(&pool->lock, flags);
2968 if (h->all_io_entry) {
2969 INIT_LIST_HEAD(&work);
2970 dm_deferred_entry_dec(h->all_io_entry, &work);
2971 if (!list_empty(&work)) {
2972 spin_lock_irqsave(&pool->lock, flags);
2973 list_for_each_entry_safe(m, tmp, &work, list)
2974 list_add_tail(&m->list, &pool->prepared_discards);
2975 spin_unlock_irqrestore(&pool->lock, flags);
2983 static void thin_postsuspend(struct dm_target *ti)
2985 if (dm_noflush_suspending(ti))
2986 requeue_io((struct thin_c *)ti->private);
2990 * <nr mapped sectors> <highest mapped sector>
2992 static void thin_status(struct dm_target *ti, status_type_t type,
2993 unsigned status_flags, char *result, unsigned maxlen)
2997 dm_block_t mapped, highest;
2998 char buf[BDEVNAME_SIZE];
2999 struct thin_c *tc = ti->private;
3001 if (get_pool_mode(tc->pool) == PM_FAIL) {
3010 case STATUSTYPE_INFO:
3011 r = dm_thin_get_mapped_count(tc->td, &mapped);
3013 DMERR("dm_thin_get_mapped_count returned %d", r);
3017 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
3019 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
3023 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
3025 DMEMIT("%llu", ((highest + 1) *
3026 tc->pool->sectors_per_block) - 1);
3031 case STATUSTYPE_TABLE:
3033 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
3034 (unsigned long) tc->dev_id);
3036 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
3047 static int thin_iterate_devices(struct dm_target *ti,
3048 iterate_devices_callout_fn fn, void *data)
3051 struct thin_c *tc = ti->private;
3052 struct pool *pool = tc->pool;
3055 * We can't call dm_pool_get_data_dev_size() since that blocks. So
3056 * we follow a more convoluted path through to the pool's target.
3059 return 0; /* nothing is bound */
3061 blocks = pool->ti->len;
3062 (void) sector_div(blocks, pool->sectors_per_block);
3064 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
3069 static struct target_type thin_target = {
3071 .version = {1, 10, 0},
3072 .module = THIS_MODULE,
3076 .end_io = thin_endio,
3077 .postsuspend = thin_postsuspend,
3078 .status = thin_status,
3079 .iterate_devices = thin_iterate_devices,
3082 /*----------------------------------------------------------------*/
3084 static int __init dm_thin_init(void)
3090 r = dm_register_target(&thin_target);
3094 r = dm_register_target(&pool_target);
3096 goto bad_pool_target;
3100 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
3101 if (!_new_mapping_cache)
3102 goto bad_new_mapping_cache;
3106 bad_new_mapping_cache:
3107 dm_unregister_target(&pool_target);
3109 dm_unregister_target(&thin_target);
3114 static void dm_thin_exit(void)
3116 dm_unregister_target(&thin_target);
3117 dm_unregister_target(&pool_target);
3119 kmem_cache_destroy(_new_mapping_cache);
3122 module_init(dm_thin_init);
3123 module_exit(dm_thin_exit);
3125 MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
3126 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3127 MODULE_LICENSE("GPL");