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/log2.h>
15 #include <linux/list.h>
16 #include <linux/rculist.h>
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
20 #include <linux/rbtree.h>
22 #define DM_MSG_PREFIX "thin"
27 #define ENDIO_HOOK_POOL_SIZE 1024
28 #define MAPPING_POOL_SIZE 1024
29 #define COMMIT_PERIOD HZ
30 #define NO_SPACE_TIMEOUT_SECS 60
32 static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS;
34 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
35 "A percentage of time allocated for copy on write");
38 * The block size of the device holding pool data must be
39 * between 64KB and 1GB.
41 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
42 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
45 * Device id is restricted to 24 bits.
47 #define MAX_DEV_ID ((1 << 24) - 1)
50 * How do we handle breaking sharing of data blocks?
51 * =================================================
53 * We use a standard copy-on-write btree to store the mappings for the
54 * devices (note I'm talking about copy-on-write of the metadata here, not
55 * the data). When you take an internal snapshot you clone the root node
56 * of the origin btree. After this there is no concept of an origin or a
57 * snapshot. They are just two device trees that happen to point to the
60 * When we get a write in we decide if it's to a shared data block using
61 * some timestamp magic. If it is, we have to break sharing.
63 * Let's say we write to a shared block in what was the origin. The
66 * i) plug io further to this physical block. (see bio_prison code).
68 * ii) quiesce any read io to that shared data block. Obviously
69 * including all devices that share this block. (see dm_deferred_set code)
71 * iii) copy the data block to a newly allocate block. This step can be
72 * missed out if the io covers the block. (schedule_copy).
74 * iv) insert the new mapping into the origin's btree
75 * (process_prepared_mapping). This act of inserting breaks some
76 * sharing of btree nodes between the two devices. Breaking sharing only
77 * effects the btree of that specific device. Btrees for the other
78 * devices that share the block never change. The btree for the origin
79 * device as it was after the last commit is untouched, ie. we're using
80 * persistent data structures in the functional programming sense.
82 * v) unplug io to this physical block, including the io that triggered
83 * the breaking of sharing.
85 * Steps (ii) and (iii) occur in parallel.
87 * The metadata _doesn't_ need to be committed before the io continues. We
88 * get away with this because the io is always written to a _new_ block.
89 * If there's a crash, then:
91 * - The origin mapping will point to the old origin block (the shared
92 * one). This will contain the data as it was before the io that triggered
93 * the breaking of sharing came in.
95 * - The snap mapping still points to the old block. As it would after
98 * The downside of this scheme is the timestamp magic isn't perfect, and
99 * will continue to think that data block in the snapshot device is shared
100 * even after the write to the origin has broken sharing. I suspect data
101 * blocks will typically be shared by many different devices, so we're
102 * breaking sharing n + 1 times, rather than n, where n is the number of
103 * devices that reference this data block. At the moment I think the
104 * benefits far, far outweigh the disadvantages.
107 /*----------------------------------------------------------------*/
112 static void build_data_key(struct dm_thin_device *td,
113 dm_block_t b, struct dm_cell_key *key)
116 key->dev = dm_thin_dev_id(td);
120 static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
121 struct dm_cell_key *key)
124 key->dev = dm_thin_dev_id(td);
128 /*----------------------------------------------------------------*/
130 #define THROTTLE_THRESHOLD (1 * HZ)
133 struct rw_semaphore lock;
134 unsigned long threshold;
135 bool throttle_applied;
138 static void throttle_init(struct throttle *t)
140 init_rwsem(&t->lock);
141 t->throttle_applied = false;
144 static void throttle_work_start(struct throttle *t)
146 t->threshold = jiffies + THROTTLE_THRESHOLD;
149 static void throttle_work_update(struct throttle *t)
151 if (!t->throttle_applied && jiffies > t->threshold) {
152 down_write(&t->lock);
153 t->throttle_applied = true;
157 static void throttle_work_complete(struct throttle *t)
159 if (t->throttle_applied) {
160 t->throttle_applied = false;
165 static void throttle_lock(struct throttle *t)
170 static void throttle_unlock(struct throttle *t)
175 /*----------------------------------------------------------------*/
178 * A pool device ties together a metadata device and a data device. It
179 * also provides the interface for creating and destroying internal
182 struct dm_thin_new_mapping;
185 * The pool runs in 4 modes. Ordered in degraded order for comparisons.
188 PM_WRITE, /* metadata may be changed */
189 PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */
190 PM_READ_ONLY, /* metadata may not be changed */
191 PM_FAIL, /* all I/O fails */
194 struct pool_features {
197 bool zero_new_blocks:1;
198 bool discard_enabled:1;
199 bool discard_passdown:1;
200 bool error_if_no_space:1;
204 typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
205 typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);
208 struct list_head list;
209 struct dm_target *ti; /* Only set if a pool target is bound */
211 struct mapped_device *pool_md;
212 struct block_device *md_dev;
213 struct dm_pool_metadata *pmd;
215 dm_block_t low_water_blocks;
216 uint32_t sectors_per_block;
217 int sectors_per_block_shift;
219 struct pool_features pf;
220 bool low_water_triggered:1; /* A dm event has been sent */
222 struct dm_bio_prison *prison;
223 struct dm_kcopyd_client *copier;
225 struct workqueue_struct *wq;
226 struct throttle throttle;
227 struct work_struct worker;
228 struct delayed_work waker;
229 struct delayed_work no_space_timeout;
231 unsigned long last_commit_jiffies;
235 struct bio_list deferred_flush_bios;
236 struct list_head prepared_mappings;
237 struct list_head prepared_discards;
238 struct list_head active_thins;
240 struct dm_deferred_set *shared_read_ds;
241 struct dm_deferred_set *all_io_ds;
243 struct dm_thin_new_mapping *next_mapping;
244 mempool_t *mapping_pool;
246 process_bio_fn process_bio;
247 process_bio_fn process_discard;
249 process_mapping_fn process_prepared_mapping;
250 process_mapping_fn process_prepared_discard;
253 static enum pool_mode get_pool_mode(struct pool *pool);
254 static void metadata_operation_failed(struct pool *pool, const char *op, int r);
257 * Target context for a pool.
260 struct dm_target *ti;
262 struct dm_dev *data_dev;
263 struct dm_dev *metadata_dev;
264 struct dm_target_callbacks callbacks;
266 dm_block_t low_water_blocks;
267 struct pool_features requested_pf; /* Features requested during table load */
268 struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */
272 * Target context for a thin.
275 struct list_head list;
276 struct dm_dev *pool_dev;
277 struct dm_dev *origin_dev;
278 sector_t origin_size;
282 struct dm_thin_device *td;
285 struct bio_list deferred_bio_list;
286 struct bio_list retry_on_resume_list;
287 struct rb_root sort_bio_list; /* sorted list of deferred bios */
290 * Ensures the thin is not destroyed until the worker has finished
291 * iterating the active_thins list.
294 struct completion can_destroy;
297 /*----------------------------------------------------------------*/
300 * wake_worker() is used when new work is queued and when pool_resume is
301 * ready to continue deferred IO processing.
303 static void wake_worker(struct pool *pool)
305 queue_work(pool->wq, &pool->worker);
308 /*----------------------------------------------------------------*/
310 static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio,
311 struct dm_bio_prison_cell **cell_result)
314 struct dm_bio_prison_cell *cell_prealloc;
317 * Allocate a cell from the prison's mempool.
318 * This might block but it can't fail.
320 cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO);
322 r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result);
325 * We reused an old cell; we can get rid of
328 dm_bio_prison_free_cell(pool->prison, cell_prealloc);
333 static void cell_release(struct pool *pool,
334 struct dm_bio_prison_cell *cell,
335 struct bio_list *bios)
337 dm_cell_release(pool->prison, cell, bios);
338 dm_bio_prison_free_cell(pool->prison, cell);
341 static void cell_release_no_holder(struct pool *pool,
342 struct dm_bio_prison_cell *cell,
343 struct bio_list *bios)
345 dm_cell_release_no_holder(pool->prison, cell, bios);
346 dm_bio_prison_free_cell(pool->prison, cell);
349 static void cell_defer_no_holder_no_free(struct thin_c *tc,
350 struct dm_bio_prison_cell *cell)
352 struct pool *pool = tc->pool;
355 spin_lock_irqsave(&tc->lock, flags);
356 dm_cell_release_no_holder(pool->prison, cell, &tc->deferred_bio_list);
357 spin_unlock_irqrestore(&tc->lock, flags);
362 static void cell_error_with_code(struct pool *pool,
363 struct dm_bio_prison_cell *cell, int error_code)
365 dm_cell_error(pool->prison, cell, error_code);
366 dm_bio_prison_free_cell(pool->prison, cell);
369 static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell)
371 cell_error_with_code(pool, cell, -EIO);
374 /*----------------------------------------------------------------*/
377 * A global list of pools that uses a struct mapped_device as a key.
379 static struct dm_thin_pool_table {
381 struct list_head pools;
382 } dm_thin_pool_table;
384 static void pool_table_init(void)
386 mutex_init(&dm_thin_pool_table.mutex);
387 INIT_LIST_HEAD(&dm_thin_pool_table.pools);
390 static void __pool_table_insert(struct pool *pool)
392 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
393 list_add(&pool->list, &dm_thin_pool_table.pools);
396 static void __pool_table_remove(struct pool *pool)
398 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
399 list_del(&pool->list);
402 static struct pool *__pool_table_lookup(struct mapped_device *md)
404 struct pool *pool = NULL, *tmp;
406 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
408 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
409 if (tmp->pool_md == md) {
418 static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
420 struct pool *pool = NULL, *tmp;
422 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
424 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
425 if (tmp->md_dev == md_dev) {
434 /*----------------------------------------------------------------*/
436 struct dm_thin_endio_hook {
438 struct dm_deferred_entry *shared_read_entry;
439 struct dm_deferred_entry *all_io_entry;
440 struct dm_thin_new_mapping *overwrite_mapping;
441 struct rb_node rb_node;
444 static void requeue_bio_list(struct thin_c *tc, struct bio_list *master)
447 struct bio_list bios;
450 bio_list_init(&bios);
452 spin_lock_irqsave(&tc->lock, flags);
453 bio_list_merge(&bios, master);
454 bio_list_init(master);
455 spin_unlock_irqrestore(&tc->lock, flags);
457 while ((bio = bio_list_pop(&bios)))
458 bio_endio(bio, DM_ENDIO_REQUEUE);
461 static void requeue_io(struct thin_c *tc)
463 requeue_bio_list(tc, &tc->deferred_bio_list);
464 requeue_bio_list(tc, &tc->retry_on_resume_list);
467 static void error_thin_retry_list(struct thin_c *tc)
471 struct bio_list bios;
473 bio_list_init(&bios);
475 spin_lock_irqsave(&tc->lock, flags);
476 bio_list_merge(&bios, &tc->retry_on_resume_list);
477 bio_list_init(&tc->retry_on_resume_list);
478 spin_unlock_irqrestore(&tc->lock, flags);
480 while ((bio = bio_list_pop(&bios)))
484 static void error_retry_list(struct pool *pool)
489 list_for_each_entry_rcu(tc, &pool->active_thins, list)
490 error_thin_retry_list(tc);
495 * This section of code contains the logic for processing a thin device's IO.
496 * Much of the code depends on pool object resources (lists, workqueues, etc)
497 * but most is exclusively called from the thin target rather than the thin-pool
501 static bool block_size_is_power_of_two(struct pool *pool)
503 return pool->sectors_per_block_shift >= 0;
506 static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
508 struct pool *pool = tc->pool;
509 sector_t block_nr = bio->bi_iter.bi_sector;
511 if (block_size_is_power_of_two(pool))
512 block_nr >>= pool->sectors_per_block_shift;
514 (void) sector_div(block_nr, pool->sectors_per_block);
519 static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
521 struct pool *pool = tc->pool;
522 sector_t bi_sector = bio->bi_iter.bi_sector;
524 bio->bi_bdev = tc->pool_dev->bdev;
525 if (block_size_is_power_of_two(pool))
526 bio->bi_iter.bi_sector =
527 (block << pool->sectors_per_block_shift) |
528 (bi_sector & (pool->sectors_per_block - 1));
530 bio->bi_iter.bi_sector = (block * pool->sectors_per_block) +
531 sector_div(bi_sector, pool->sectors_per_block);
534 static void remap_to_origin(struct thin_c *tc, struct bio *bio)
536 bio->bi_bdev = tc->origin_dev->bdev;
539 static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
541 return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) &&
542 dm_thin_changed_this_transaction(tc->td);
545 static void inc_all_io_entry(struct pool *pool, struct bio *bio)
547 struct dm_thin_endio_hook *h;
549 if (bio->bi_rw & REQ_DISCARD)
552 h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
553 h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
556 static void issue(struct thin_c *tc, struct bio *bio)
558 struct pool *pool = tc->pool;
561 if (!bio_triggers_commit(tc, bio)) {
562 generic_make_request(bio);
567 * Complete bio with an error if earlier I/O caused changes to
568 * the metadata that can't be committed e.g, due to I/O errors
569 * on the metadata device.
571 if (dm_thin_aborted_changes(tc->td)) {
577 * Batch together any bios that trigger commits and then issue a
578 * single commit for them in process_deferred_bios().
580 spin_lock_irqsave(&pool->lock, flags);
581 bio_list_add(&pool->deferred_flush_bios, bio);
582 spin_unlock_irqrestore(&pool->lock, flags);
585 static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
587 remap_to_origin(tc, bio);
591 static void remap_and_issue(struct thin_c *tc, struct bio *bio,
594 remap(tc, bio, block);
598 /*----------------------------------------------------------------*/
601 * Bio endio functions.
603 struct dm_thin_new_mapping {
604 struct list_head list;
607 bool definitely_not_shared:1;
610 * Track quiescing, copying and zeroing preparation actions. When this
611 * counter hits zero the block is prepared and can be inserted into the
614 atomic_t prepare_actions;
618 dm_block_t virt_block;
619 dm_block_t data_block;
620 struct dm_bio_prison_cell *cell, *cell2;
623 * If the bio covers the whole area of a block then we can avoid
624 * zeroing or copying. Instead this bio is hooked. The bio will
625 * still be in the cell, so care has to be taken to avoid issuing
629 bio_end_io_t *saved_bi_end_io;
632 static void __complete_mapping_preparation(struct dm_thin_new_mapping *m)
634 struct pool *pool = m->tc->pool;
636 if (atomic_dec_and_test(&m->prepare_actions)) {
637 list_add_tail(&m->list, &pool->prepared_mappings);
642 static void complete_mapping_preparation(struct dm_thin_new_mapping *m)
645 struct pool *pool = m->tc->pool;
647 spin_lock_irqsave(&pool->lock, flags);
648 __complete_mapping_preparation(m);
649 spin_unlock_irqrestore(&pool->lock, flags);
652 static void copy_complete(int read_err, unsigned long write_err, void *context)
654 struct dm_thin_new_mapping *m = context;
656 m->err = read_err || write_err ? -EIO : 0;
657 complete_mapping_preparation(m);
660 static void overwrite_endio(struct bio *bio, int err)
662 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
663 struct dm_thin_new_mapping *m = h->overwrite_mapping;
666 complete_mapping_preparation(m);
669 /*----------------------------------------------------------------*/
676 * Prepared mapping jobs.
680 * This sends the bios in the cell back to the deferred_bios list.
682 static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell)
684 struct pool *pool = tc->pool;
687 spin_lock_irqsave(&tc->lock, flags);
688 cell_release(pool, cell, &tc->deferred_bio_list);
689 spin_unlock_irqrestore(&tc->lock, flags);
695 * Same as cell_defer above, except it omits the original holder of the cell.
697 static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
699 struct pool *pool = tc->pool;
702 spin_lock_irqsave(&tc->lock, flags);
703 cell_release_no_holder(pool, cell, &tc->deferred_bio_list);
704 spin_unlock_irqrestore(&tc->lock, flags);
709 static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
712 m->bio->bi_end_io = m->saved_bi_end_io;
713 atomic_inc(&m->bio->bi_remaining);
715 cell_error(m->tc->pool, m->cell);
717 mempool_free(m, m->tc->pool->mapping_pool);
720 static void process_prepared_mapping(struct dm_thin_new_mapping *m)
722 struct thin_c *tc = m->tc;
723 struct pool *pool = tc->pool;
729 bio->bi_end_io = m->saved_bi_end_io;
730 atomic_inc(&bio->bi_remaining);
734 cell_error(pool, m->cell);
739 * Commit the prepared block into the mapping btree.
740 * Any I/O for this block arriving after this point will get
741 * remapped to it directly.
743 r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block);
745 metadata_operation_failed(pool, "dm_thin_insert_block", r);
746 cell_error(pool, m->cell);
751 * Release any bios held while the block was being provisioned.
752 * If we are processing a write bio that completely covers the block,
753 * we already processed it so can ignore it now when processing
754 * the bios in the cell.
757 cell_defer_no_holder(tc, m->cell);
760 cell_defer(tc, m->cell);
764 mempool_free(m, pool->mapping_pool);
767 static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
769 struct thin_c *tc = m->tc;
771 bio_io_error(m->bio);
772 cell_defer_no_holder(tc, m->cell);
773 cell_defer_no_holder(tc, m->cell2);
774 mempool_free(m, tc->pool->mapping_pool);
777 static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m)
779 struct thin_c *tc = m->tc;
781 inc_all_io_entry(tc->pool, m->bio);
782 cell_defer_no_holder(tc, m->cell);
783 cell_defer_no_holder(tc, m->cell2);
786 if (m->definitely_not_shared)
787 remap_and_issue(tc, m->bio, m->data_block);
790 if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used)
791 bio_endio(m->bio, 0);
793 remap_and_issue(tc, m->bio, m->data_block);
796 bio_endio(m->bio, 0);
798 mempool_free(m, tc->pool->mapping_pool);
801 static void process_prepared_discard(struct dm_thin_new_mapping *m)
804 struct thin_c *tc = m->tc;
806 r = dm_thin_remove_block(tc->td, m->virt_block);
808 DMERR_LIMIT("dm_thin_remove_block() failed");
810 process_prepared_discard_passdown(m);
813 static void process_prepared(struct pool *pool, struct list_head *head,
814 process_mapping_fn *fn)
817 struct list_head maps;
818 struct dm_thin_new_mapping *m, *tmp;
820 INIT_LIST_HEAD(&maps);
821 spin_lock_irqsave(&pool->lock, flags);
822 list_splice_init(head, &maps);
823 spin_unlock_irqrestore(&pool->lock, flags);
825 list_for_each_entry_safe(m, tmp, &maps, list)
832 static int io_overlaps_block(struct pool *pool, struct bio *bio)
834 return bio->bi_iter.bi_size ==
835 (pool->sectors_per_block << SECTOR_SHIFT);
838 static int io_overwrites_block(struct pool *pool, struct bio *bio)
840 return (bio_data_dir(bio) == WRITE) &&
841 io_overlaps_block(pool, bio);
844 static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
847 *save = bio->bi_end_io;
851 static int ensure_next_mapping(struct pool *pool)
853 if (pool->next_mapping)
856 pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);
858 return pool->next_mapping ? 0 : -ENOMEM;
861 static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
863 struct dm_thin_new_mapping *m = pool->next_mapping;
865 BUG_ON(!pool->next_mapping);
867 memset(m, 0, sizeof(struct dm_thin_new_mapping));
868 INIT_LIST_HEAD(&m->list);
871 pool->next_mapping = NULL;
876 static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m,
877 sector_t begin, sector_t end)
880 struct dm_io_region to;
882 to.bdev = tc->pool_dev->bdev;
884 to.count = end - begin;
886 r = dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m);
888 DMERR_LIMIT("dm_kcopyd_zero() failed");
889 copy_complete(1, 1, m);
894 * A partial copy also needs to zero the uncopied region.
896 static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
897 struct dm_dev *origin, dm_block_t data_origin,
898 dm_block_t data_dest,
899 struct dm_bio_prison_cell *cell, struct bio *bio,
903 struct pool *pool = tc->pool;
904 struct dm_thin_new_mapping *m = get_next_mapping(pool);
907 m->virt_block = virt_block;
908 m->data_block = data_dest;
912 * quiesce action + copy action + an extra reference held for the
913 * duration of this function (we may need to inc later for a
916 atomic_set(&m->prepare_actions, 3);
918 if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
919 complete_mapping_preparation(m); /* already quiesced */
922 * IO to pool_dev remaps to the pool target's data_dev.
924 * If the whole block of data is being overwritten, we can issue the
925 * bio immediately. Otherwise we use kcopyd to clone the data first.
927 if (io_overwrites_block(pool, bio)) {
928 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
930 h->overwrite_mapping = m;
932 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
933 inc_all_io_entry(pool, bio);
934 remap_and_issue(tc, bio, data_dest);
936 struct dm_io_region from, to;
938 from.bdev = origin->bdev;
939 from.sector = data_origin * pool->sectors_per_block;
942 to.bdev = tc->pool_dev->bdev;
943 to.sector = data_dest * pool->sectors_per_block;
946 r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
947 0, copy_complete, m);
949 DMERR_LIMIT("dm_kcopyd_copy() failed");
950 copy_complete(1, 1, m);
953 * We allow the zero to be issued, to simplify the
954 * error path. Otherwise we'd need to start
955 * worrying about decrementing the prepare_actions
961 * Do we need to zero a tail region?
963 if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) {
964 atomic_inc(&m->prepare_actions);
966 data_dest * pool->sectors_per_block + len,
967 (data_dest + 1) * pool->sectors_per_block);
971 complete_mapping_preparation(m); /* drop our ref */
974 static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
975 dm_block_t data_origin, dm_block_t data_dest,
976 struct dm_bio_prison_cell *cell, struct bio *bio)
978 schedule_copy(tc, virt_block, tc->pool_dev,
979 data_origin, data_dest, cell, bio,
980 tc->pool->sectors_per_block);
983 static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
984 dm_block_t data_block, struct dm_bio_prison_cell *cell,
987 struct pool *pool = tc->pool;
988 struct dm_thin_new_mapping *m = get_next_mapping(pool);
990 atomic_set(&m->prepare_actions, 1); /* no need to quiesce */
992 m->virt_block = virt_block;
993 m->data_block = data_block;
997 * If the whole block of data is being overwritten or we are not
998 * zeroing pre-existing data, we can issue the bio immediately.
999 * Otherwise we use kcopyd to zero the data first.
1001 if (!pool->pf.zero_new_blocks)
1002 process_prepared_mapping(m);
1004 else if (io_overwrites_block(pool, bio)) {
1005 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1007 h->overwrite_mapping = m;
1009 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
1010 inc_all_io_entry(pool, bio);
1011 remap_and_issue(tc, bio, data_block);
1015 data_block * pool->sectors_per_block,
1016 (data_block + 1) * pool->sectors_per_block);
1019 static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
1020 dm_block_t data_dest,
1021 struct dm_bio_prison_cell *cell, struct bio *bio)
1023 struct pool *pool = tc->pool;
1024 sector_t virt_block_begin = virt_block * pool->sectors_per_block;
1025 sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block;
1027 if (virt_block_end <= tc->origin_size)
1028 schedule_copy(tc, virt_block, tc->origin_dev,
1029 virt_block, data_dest, cell, bio,
1030 pool->sectors_per_block);
1032 else if (virt_block_begin < tc->origin_size)
1033 schedule_copy(tc, virt_block, tc->origin_dev,
1034 virt_block, data_dest, cell, bio,
1035 tc->origin_size - virt_block_begin);
1038 schedule_zero(tc, virt_block, data_dest, cell, bio);
1042 * A non-zero return indicates read_only or fail_io mode.
1043 * Many callers don't care about the return value.
1045 static int commit(struct pool *pool)
1049 if (get_pool_mode(pool) >= PM_READ_ONLY)
1052 r = dm_pool_commit_metadata(pool->pmd);
1054 metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
1059 static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks)
1061 unsigned long flags;
1063 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
1064 DMWARN("%s: reached low water mark for data device: sending event.",
1065 dm_device_name(pool->pool_md));
1066 spin_lock_irqsave(&pool->lock, flags);
1067 pool->low_water_triggered = true;
1068 spin_unlock_irqrestore(&pool->lock, flags);
1069 dm_table_event(pool->ti->table);
1073 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode);
1075 static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
1078 dm_block_t free_blocks;
1079 struct pool *pool = tc->pool;
1081 if (WARN_ON(get_pool_mode(pool) != PM_WRITE))
1084 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1086 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1090 check_low_water_mark(pool, free_blocks);
1094 * Try to commit to see if that will free up some
1101 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1103 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1108 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1113 r = dm_pool_alloc_data_block(pool->pmd, result);
1115 metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
1123 * If we have run out of space, queue bios until the device is
1124 * resumed, presumably after having been reloaded with more space.
1126 static void retry_on_resume(struct bio *bio)
1128 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1129 struct thin_c *tc = h->tc;
1130 unsigned long flags;
1132 spin_lock_irqsave(&tc->lock, flags);
1133 bio_list_add(&tc->retry_on_resume_list, bio);
1134 spin_unlock_irqrestore(&tc->lock, flags);
1137 static int should_error_unserviceable_bio(struct pool *pool)
1139 enum pool_mode m = get_pool_mode(pool);
1143 /* Shouldn't get here */
1144 DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
1147 case PM_OUT_OF_DATA_SPACE:
1148 return pool->pf.error_if_no_space ? -ENOSPC : 0;
1154 /* Shouldn't get here */
1155 DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1160 static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
1162 int error = should_error_unserviceable_bio(pool);
1165 bio_endio(bio, error);
1167 retry_on_resume(bio);
1170 static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell)
1173 struct bio_list bios;
1176 error = should_error_unserviceable_bio(pool);
1178 cell_error_with_code(pool, cell, error);
1182 bio_list_init(&bios);
1183 cell_release(pool, cell, &bios);
1185 error = should_error_unserviceable_bio(pool);
1187 while ((bio = bio_list_pop(&bios)))
1188 bio_endio(bio, error);
1190 while ((bio = bio_list_pop(&bios)))
1191 retry_on_resume(bio);
1194 static void process_discard(struct thin_c *tc, struct bio *bio)
1197 struct pool *pool = tc->pool;
1198 struct dm_bio_prison_cell *cell, *cell2;
1199 struct dm_cell_key key, key2;
1200 dm_block_t block = get_bio_block(tc, bio);
1201 struct dm_thin_lookup_result lookup_result;
1202 struct dm_thin_new_mapping *m;
1204 build_virtual_key(tc->td, block, &key);
1205 if (bio_detain(tc->pool, &key, bio, &cell))
1208 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1212 * Check nobody is fiddling with this pool block. This can
1213 * happen if someone's in the process of breaking sharing
1216 build_data_key(tc->td, lookup_result.block, &key2);
1217 if (bio_detain(tc->pool, &key2, bio, &cell2)) {
1218 cell_defer_no_holder(tc, cell);
1222 if (io_overlaps_block(pool, bio)) {
1224 * IO may still be going to the destination block. We must
1225 * quiesce before we can do the removal.
1227 m = get_next_mapping(pool);
1229 m->pass_discard = pool->pf.discard_passdown;
1230 m->definitely_not_shared = !lookup_result.shared;
1231 m->virt_block = block;
1232 m->data_block = lookup_result.block;
1237 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1238 pool->process_prepared_discard(m);
1241 inc_all_io_entry(pool, bio);
1242 cell_defer_no_holder(tc, cell);
1243 cell_defer_no_holder(tc, cell2);
1246 * The DM core makes sure that the discard doesn't span
1247 * a block boundary. So we submit the discard of a
1248 * partial block appropriately.
1250 if ((!lookup_result.shared) && pool->pf.discard_passdown)
1251 remap_and_issue(tc, bio, lookup_result.block);
1259 * It isn't provisioned, just forget it.
1261 cell_defer_no_holder(tc, cell);
1266 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1268 cell_defer_no_holder(tc, cell);
1274 static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
1275 struct dm_cell_key *key,
1276 struct dm_thin_lookup_result *lookup_result,
1277 struct dm_bio_prison_cell *cell)
1280 dm_block_t data_block;
1281 struct pool *pool = tc->pool;
1283 r = alloc_data_block(tc, &data_block);
1286 schedule_internal_copy(tc, block, lookup_result->block,
1287 data_block, cell, bio);
1291 retry_bios_on_resume(pool, cell);
1295 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1297 cell_error(pool, cell);
1302 static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1304 struct dm_thin_lookup_result *lookup_result)
1306 struct dm_bio_prison_cell *cell;
1307 struct pool *pool = tc->pool;
1308 struct dm_cell_key key;
1311 * If cell is already occupied, then sharing is already in the process
1312 * of being broken so we have nothing further to do here.
1314 build_data_key(tc->td, lookup_result->block, &key);
1315 if (bio_detain(pool, &key, bio, &cell))
1318 if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size)
1319 break_sharing(tc, bio, block, &key, lookup_result, cell);
1321 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1323 h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
1324 inc_all_io_entry(pool, bio);
1325 cell_defer_no_holder(tc, cell);
1327 remap_and_issue(tc, bio, lookup_result->block);
1331 static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
1332 struct dm_bio_prison_cell *cell)
1335 dm_block_t data_block;
1336 struct pool *pool = tc->pool;
1339 * Remap empty bios (flushes) immediately, without provisioning.
1341 if (!bio->bi_iter.bi_size) {
1342 inc_all_io_entry(pool, bio);
1343 cell_defer_no_holder(tc, cell);
1345 remap_and_issue(tc, bio, 0);
1350 * Fill read bios with zeroes and complete them immediately.
1352 if (bio_data_dir(bio) == READ) {
1354 cell_defer_no_holder(tc, cell);
1359 r = alloc_data_block(tc, &data_block);
1363 schedule_external_copy(tc, block, data_block, cell, bio);
1365 schedule_zero(tc, block, data_block, cell, bio);
1369 retry_bios_on_resume(pool, cell);
1373 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1375 cell_error(pool, cell);
1380 static void process_bio(struct thin_c *tc, struct bio *bio)
1383 struct pool *pool = tc->pool;
1384 dm_block_t block = get_bio_block(tc, bio);
1385 struct dm_bio_prison_cell *cell;
1386 struct dm_cell_key key;
1387 struct dm_thin_lookup_result lookup_result;
1390 * If cell is already occupied, then the block is already
1391 * being provisioned so we have nothing further to do here.
1393 build_virtual_key(tc->td, block, &key);
1394 if (bio_detain(pool, &key, bio, &cell))
1397 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1400 if (lookup_result.shared) {
1401 process_shared_bio(tc, bio, block, &lookup_result);
1402 cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */
1404 inc_all_io_entry(pool, bio);
1405 cell_defer_no_holder(tc, cell);
1407 remap_and_issue(tc, bio, lookup_result.block);
1412 if (bio_data_dir(bio) == READ && tc->origin_dev) {
1413 inc_all_io_entry(pool, bio);
1414 cell_defer_no_holder(tc, cell);
1416 if (bio_end_sector(bio) <= tc->origin_size)
1417 remap_to_origin_and_issue(tc, bio);
1419 else if (bio->bi_iter.bi_sector < tc->origin_size) {
1421 bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT;
1422 remap_to_origin_and_issue(tc, bio);
1429 provision_block(tc, bio, block, cell);
1433 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1435 cell_defer_no_holder(tc, cell);
1441 static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
1444 int rw = bio_data_dir(bio);
1445 dm_block_t block = get_bio_block(tc, bio);
1446 struct dm_thin_lookup_result lookup_result;
1448 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1451 if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size)
1452 handle_unserviceable_bio(tc->pool, bio);
1454 inc_all_io_entry(tc->pool, bio);
1455 remap_and_issue(tc, bio, lookup_result.block);
1461 handle_unserviceable_bio(tc->pool, bio);
1465 if (tc->origin_dev) {
1466 inc_all_io_entry(tc->pool, bio);
1467 remap_to_origin_and_issue(tc, bio);
1476 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1483 static void process_bio_success(struct thin_c *tc, struct bio *bio)
1488 static void process_bio_fail(struct thin_c *tc, struct bio *bio)
1494 * FIXME: should we also commit due to size of transaction, measured in
1497 static int need_commit_due_to_time(struct pool *pool)
1499 return jiffies < pool->last_commit_jiffies ||
1500 jiffies > pool->last_commit_jiffies + COMMIT_PERIOD;
1503 #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node)
1504 #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook))
1506 static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio)
1508 struct rb_node **rbp, *parent;
1509 struct dm_thin_endio_hook *pbd;
1510 sector_t bi_sector = bio->bi_iter.bi_sector;
1512 rbp = &tc->sort_bio_list.rb_node;
1516 pbd = thin_pbd(parent);
1518 if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector)
1519 rbp = &(*rbp)->rb_left;
1521 rbp = &(*rbp)->rb_right;
1524 pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1525 rb_link_node(&pbd->rb_node, parent, rbp);
1526 rb_insert_color(&pbd->rb_node, &tc->sort_bio_list);
1529 static void __extract_sorted_bios(struct thin_c *tc)
1531 struct rb_node *node;
1532 struct dm_thin_endio_hook *pbd;
1535 for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) {
1536 pbd = thin_pbd(node);
1537 bio = thin_bio(pbd);
1539 bio_list_add(&tc->deferred_bio_list, bio);
1540 rb_erase(&pbd->rb_node, &tc->sort_bio_list);
1543 WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list));
1546 static void __sort_thin_deferred_bios(struct thin_c *tc)
1549 struct bio_list bios;
1551 bio_list_init(&bios);
1552 bio_list_merge(&bios, &tc->deferred_bio_list);
1553 bio_list_init(&tc->deferred_bio_list);
1555 /* Sort deferred_bio_list using rb-tree */
1556 while ((bio = bio_list_pop(&bios)))
1557 __thin_bio_rb_add(tc, bio);
1560 * Transfer the sorted bios in sort_bio_list back to
1561 * deferred_bio_list to allow lockless submission of
1564 __extract_sorted_bios(tc);
1567 static void process_thin_deferred_bios(struct thin_c *tc)
1569 struct pool *pool = tc->pool;
1570 unsigned long flags;
1572 struct bio_list bios;
1573 struct blk_plug plug;
1576 if (tc->requeue_mode) {
1577 requeue_bio_list(tc, &tc->deferred_bio_list);
1581 bio_list_init(&bios);
1583 spin_lock_irqsave(&tc->lock, flags);
1585 if (bio_list_empty(&tc->deferred_bio_list)) {
1586 spin_unlock_irqrestore(&tc->lock, flags);
1590 __sort_thin_deferred_bios(tc);
1592 bio_list_merge(&bios, &tc->deferred_bio_list);
1593 bio_list_init(&tc->deferred_bio_list);
1595 spin_unlock_irqrestore(&tc->lock, flags);
1597 blk_start_plug(&plug);
1598 while ((bio = bio_list_pop(&bios))) {
1600 * If we've got no free new_mapping structs, and processing
1601 * this bio might require one, we pause until there are some
1602 * prepared mappings to process.
1604 if (ensure_next_mapping(pool)) {
1605 spin_lock_irqsave(&tc->lock, flags);
1606 bio_list_add(&tc->deferred_bio_list, bio);
1607 bio_list_merge(&tc->deferred_bio_list, &bios);
1608 spin_unlock_irqrestore(&tc->lock, flags);
1612 if (bio->bi_rw & REQ_DISCARD)
1613 pool->process_discard(tc, bio);
1615 pool->process_bio(tc, bio);
1617 if ((count++ & 127) == 0) {
1618 throttle_work_update(&pool->throttle);
1619 dm_pool_issue_prefetches(pool->pmd);
1622 blk_finish_plug(&plug);
1625 static void thin_get(struct thin_c *tc);
1626 static void thin_put(struct thin_c *tc);
1629 * We can't hold rcu_read_lock() around code that can block. So we
1630 * find a thin with the rcu lock held; bump a refcount; then drop
1633 static struct thin_c *get_first_thin(struct pool *pool)
1635 struct thin_c *tc = NULL;
1638 if (!list_empty(&pool->active_thins)) {
1639 tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list);
1647 static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc)
1649 struct thin_c *old_tc = tc;
1652 list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) {
1664 static void process_deferred_bios(struct pool *pool)
1666 unsigned long flags;
1668 struct bio_list bios;
1671 tc = get_first_thin(pool);
1673 process_thin_deferred_bios(tc);
1674 tc = get_next_thin(pool, tc);
1678 * If there are any deferred flush bios, we must commit
1679 * the metadata before issuing them.
1681 bio_list_init(&bios);
1682 spin_lock_irqsave(&pool->lock, flags);
1683 bio_list_merge(&bios, &pool->deferred_flush_bios);
1684 bio_list_init(&pool->deferred_flush_bios);
1685 spin_unlock_irqrestore(&pool->lock, flags);
1687 if (bio_list_empty(&bios) &&
1688 !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool)))
1692 while ((bio = bio_list_pop(&bios)))
1696 pool->last_commit_jiffies = jiffies;
1698 while ((bio = bio_list_pop(&bios)))
1699 generic_make_request(bio);
1702 static void do_worker(struct work_struct *ws)
1704 struct pool *pool = container_of(ws, struct pool, worker);
1706 throttle_work_start(&pool->throttle);
1707 dm_pool_issue_prefetches(pool->pmd);
1708 throttle_work_update(&pool->throttle);
1709 process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
1710 throttle_work_update(&pool->throttle);
1711 process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
1712 throttle_work_update(&pool->throttle);
1713 process_deferred_bios(pool);
1714 throttle_work_complete(&pool->throttle);
1718 * We want to commit periodically so that not too much
1719 * unwritten data builds up.
1721 static void do_waker(struct work_struct *ws)
1723 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
1725 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
1729 * We're holding onto IO to allow userland time to react. After the
1730 * timeout either the pool will have been resized (and thus back in
1731 * PM_WRITE mode), or we degrade to PM_READ_ONLY and start erroring IO.
1733 static void do_no_space_timeout(struct work_struct *ws)
1735 struct pool *pool = container_of(to_delayed_work(ws), struct pool,
1738 if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space)
1739 set_pool_mode(pool, PM_READ_ONLY);
1742 /*----------------------------------------------------------------*/
1745 struct work_struct worker;
1746 struct completion complete;
1749 static struct pool_work *to_pool_work(struct work_struct *ws)
1751 return container_of(ws, struct pool_work, worker);
1754 static void pool_work_complete(struct pool_work *pw)
1756 complete(&pw->complete);
1759 static void pool_work_wait(struct pool_work *pw, struct pool *pool,
1760 void (*fn)(struct work_struct *))
1762 INIT_WORK_ONSTACK(&pw->worker, fn);
1763 init_completion(&pw->complete);
1764 queue_work(pool->wq, &pw->worker);
1765 wait_for_completion(&pw->complete);
1768 /*----------------------------------------------------------------*/
1770 struct noflush_work {
1771 struct pool_work pw;
1775 static struct noflush_work *to_noflush(struct work_struct *ws)
1777 return container_of(to_pool_work(ws), struct noflush_work, pw);
1780 static void do_noflush_start(struct work_struct *ws)
1782 struct noflush_work *w = to_noflush(ws);
1783 w->tc->requeue_mode = true;
1785 pool_work_complete(&w->pw);
1788 static void do_noflush_stop(struct work_struct *ws)
1790 struct noflush_work *w = to_noflush(ws);
1791 w->tc->requeue_mode = false;
1792 pool_work_complete(&w->pw);
1795 static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *))
1797 struct noflush_work w;
1800 pool_work_wait(&w.pw, tc->pool, fn);
1803 /*----------------------------------------------------------------*/
1805 static enum pool_mode get_pool_mode(struct pool *pool)
1807 return pool->pf.mode;
1810 static void notify_of_pool_mode_change(struct pool *pool, const char *new_mode)
1812 dm_table_event(pool->ti->table);
1813 DMINFO("%s: switching pool to %s mode",
1814 dm_device_name(pool->pool_md), new_mode);
1817 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode)
1819 struct pool_c *pt = pool->ti->private;
1820 bool needs_check = dm_pool_metadata_needs_check(pool->pmd);
1821 enum pool_mode old_mode = get_pool_mode(pool);
1822 unsigned long no_space_timeout = ACCESS_ONCE(no_space_timeout_secs) * HZ;
1825 * Never allow the pool to transition to PM_WRITE mode if user
1826 * intervention is required to verify metadata and data consistency.
1828 if (new_mode == PM_WRITE && needs_check) {
1829 DMERR("%s: unable to switch pool to write mode until repaired.",
1830 dm_device_name(pool->pool_md));
1831 if (old_mode != new_mode)
1832 new_mode = old_mode;
1834 new_mode = PM_READ_ONLY;
1837 * If we were in PM_FAIL mode, rollback of metadata failed. We're
1838 * not going to recover without a thin_repair. So we never let the
1839 * pool move out of the old mode.
1841 if (old_mode == PM_FAIL)
1842 new_mode = old_mode;
1846 if (old_mode != new_mode)
1847 notify_of_pool_mode_change(pool, "failure");
1848 dm_pool_metadata_read_only(pool->pmd);
1849 pool->process_bio = process_bio_fail;
1850 pool->process_discard = process_bio_fail;
1851 pool->process_prepared_mapping = process_prepared_mapping_fail;
1852 pool->process_prepared_discard = process_prepared_discard_fail;
1854 error_retry_list(pool);
1858 if (old_mode != new_mode)
1859 notify_of_pool_mode_change(pool, "read-only");
1860 dm_pool_metadata_read_only(pool->pmd);
1861 pool->process_bio = process_bio_read_only;
1862 pool->process_discard = process_bio_success;
1863 pool->process_prepared_mapping = process_prepared_mapping_fail;
1864 pool->process_prepared_discard = process_prepared_discard_passdown;
1866 error_retry_list(pool);
1869 case PM_OUT_OF_DATA_SPACE:
1871 * Ideally we'd never hit this state; the low water mark
1872 * would trigger userland to extend the pool before we
1873 * completely run out of data space. However, many small
1874 * IOs to unprovisioned space can consume data space at an
1875 * alarming rate. Adjust your low water mark if you're
1876 * frequently seeing this mode.
1878 if (old_mode != new_mode)
1879 notify_of_pool_mode_change(pool, "out-of-data-space");
1880 pool->process_bio = process_bio_read_only;
1881 pool->process_discard = process_discard;
1882 pool->process_prepared_mapping = process_prepared_mapping;
1883 pool->process_prepared_discard = process_prepared_discard_passdown;
1885 if (!pool->pf.error_if_no_space && no_space_timeout)
1886 queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout);
1890 if (old_mode != new_mode)
1891 notify_of_pool_mode_change(pool, "write");
1892 dm_pool_metadata_read_write(pool->pmd);
1893 pool->process_bio = process_bio;
1894 pool->process_discard = process_discard;
1895 pool->process_prepared_mapping = process_prepared_mapping;
1896 pool->process_prepared_discard = process_prepared_discard;
1900 pool->pf.mode = new_mode;
1902 * The pool mode may have changed, sync it so bind_control_target()
1903 * doesn't cause an unexpected mode transition on resume.
1905 pt->adjusted_pf.mode = new_mode;
1908 static void abort_transaction(struct pool *pool)
1910 const char *dev_name = dm_device_name(pool->pool_md);
1912 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name);
1913 if (dm_pool_abort_metadata(pool->pmd)) {
1914 DMERR("%s: failed to abort metadata transaction", dev_name);
1915 set_pool_mode(pool, PM_FAIL);
1918 if (dm_pool_metadata_set_needs_check(pool->pmd)) {
1919 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name);
1920 set_pool_mode(pool, PM_FAIL);
1924 static void metadata_operation_failed(struct pool *pool, const char *op, int r)
1926 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
1927 dm_device_name(pool->pool_md), op, r);
1929 abort_transaction(pool);
1930 set_pool_mode(pool, PM_READ_ONLY);
1933 /*----------------------------------------------------------------*/
1936 * Mapping functions.
1940 * Called only while mapping a thin bio to hand it over to the workqueue.
1942 static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
1944 unsigned long flags;
1945 struct pool *pool = tc->pool;
1947 spin_lock_irqsave(&tc->lock, flags);
1948 bio_list_add(&tc->deferred_bio_list, bio);
1949 spin_unlock_irqrestore(&tc->lock, flags);
1954 static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio)
1956 struct pool *pool = tc->pool;
1958 throttle_lock(&pool->throttle);
1959 thin_defer_bio(tc, bio);
1960 throttle_unlock(&pool->throttle);
1963 static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
1965 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1968 h->shared_read_entry = NULL;
1969 h->all_io_entry = NULL;
1970 h->overwrite_mapping = NULL;
1974 * Non-blocking function called from the thin target's map function.
1976 static int thin_bio_map(struct dm_target *ti, struct bio *bio)
1979 struct thin_c *tc = ti->private;
1980 dm_block_t block = get_bio_block(tc, bio);
1981 struct dm_thin_device *td = tc->td;
1982 struct dm_thin_lookup_result result;
1983 struct dm_bio_prison_cell cell1, cell2;
1984 struct dm_bio_prison_cell *cell_result;
1985 struct dm_cell_key key;
1987 thin_hook_bio(tc, bio);
1989 if (tc->requeue_mode) {
1990 bio_endio(bio, DM_ENDIO_REQUEUE);
1991 return DM_MAPIO_SUBMITTED;
1994 if (get_pool_mode(tc->pool) == PM_FAIL) {
1996 return DM_MAPIO_SUBMITTED;
1999 if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) {
2000 thin_defer_bio_with_throttle(tc, bio);
2001 return DM_MAPIO_SUBMITTED;
2005 * We must hold the virtual cell before doing the lookup, otherwise
2006 * there's a race with discard.
2008 build_virtual_key(tc->td, block, &key);
2009 if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result))
2010 return DM_MAPIO_SUBMITTED;
2012 r = dm_thin_find_block(td, block, 0, &result);
2015 * Note that we defer readahead too.
2019 if (unlikely(result.shared)) {
2021 * We have a race condition here between the
2022 * result.shared value returned by the lookup and
2023 * snapshot creation, which may cause new
2026 * To avoid this always quiesce the origin before
2027 * taking the snap. You want to do this anyway to
2028 * ensure a consistent application view
2031 * More distant ancestors are irrelevant. The
2032 * shared flag will be set in their case.
2034 thin_defer_bio(tc, bio);
2035 cell_defer_no_holder_no_free(tc, &cell1);
2036 return DM_MAPIO_SUBMITTED;
2039 build_data_key(tc->td, result.block, &key);
2040 if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) {
2041 cell_defer_no_holder_no_free(tc, &cell1);
2042 return DM_MAPIO_SUBMITTED;
2045 inc_all_io_entry(tc->pool, bio);
2046 cell_defer_no_holder_no_free(tc, &cell2);
2047 cell_defer_no_holder_no_free(tc, &cell1);
2049 remap(tc, bio, result.block);
2050 return DM_MAPIO_REMAPPED;
2053 if (get_pool_mode(tc->pool) == PM_READ_ONLY) {
2055 * This block isn't provisioned, and we have no way
2058 handle_unserviceable_bio(tc->pool, bio);
2059 cell_defer_no_holder_no_free(tc, &cell1);
2060 return DM_MAPIO_SUBMITTED;
2065 thin_defer_bio(tc, bio);
2066 cell_defer_no_holder_no_free(tc, &cell1);
2067 return DM_MAPIO_SUBMITTED;
2071 * Must always call bio_io_error on failure.
2072 * dm_thin_find_block can fail with -EINVAL if the
2073 * pool is switched to fail-io mode.
2076 cell_defer_no_holder_no_free(tc, &cell1);
2077 return DM_MAPIO_SUBMITTED;
2081 static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
2083 struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
2084 struct request_queue *q;
2086 if (get_pool_mode(pt->pool) == PM_OUT_OF_DATA_SPACE)
2089 q = bdev_get_queue(pt->data_dev->bdev);
2090 return bdi_congested(&q->backing_dev_info, bdi_bits);
2093 static void requeue_bios(struct pool *pool)
2095 unsigned long flags;
2099 list_for_each_entry_rcu(tc, &pool->active_thins, list) {
2100 spin_lock_irqsave(&tc->lock, flags);
2101 bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list);
2102 bio_list_init(&tc->retry_on_resume_list);
2103 spin_unlock_irqrestore(&tc->lock, flags);
2108 /*----------------------------------------------------------------
2109 * Binding of control targets to a pool object
2110 *--------------------------------------------------------------*/
2111 static bool data_dev_supports_discard(struct pool_c *pt)
2113 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
2115 return q && blk_queue_discard(q);
2118 static bool is_factor(sector_t block_size, uint32_t n)
2120 return !sector_div(block_size, n);
2124 * If discard_passdown was enabled verify that the data device
2125 * supports discards. Disable discard_passdown if not.
2127 static void disable_passdown_if_not_supported(struct pool_c *pt)
2129 struct pool *pool = pt->pool;
2130 struct block_device *data_bdev = pt->data_dev->bdev;
2131 struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
2132 sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT;
2133 const char *reason = NULL;
2134 char buf[BDEVNAME_SIZE];
2136 if (!pt->adjusted_pf.discard_passdown)
2139 if (!data_dev_supports_discard(pt))
2140 reason = "discard unsupported";
2142 else if (data_limits->max_discard_sectors < pool->sectors_per_block)
2143 reason = "max discard sectors smaller than a block";
2145 else if (data_limits->discard_granularity > block_size)
2146 reason = "discard granularity larger than a block";
2148 else if (!is_factor(block_size, data_limits->discard_granularity))
2149 reason = "discard granularity not a factor of block size";
2152 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason);
2153 pt->adjusted_pf.discard_passdown = false;
2157 static int bind_control_target(struct pool *pool, struct dm_target *ti)
2159 struct pool_c *pt = ti->private;
2162 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
2164 enum pool_mode old_mode = get_pool_mode(pool);
2165 enum pool_mode new_mode = pt->adjusted_pf.mode;
2168 * Don't change the pool's mode until set_pool_mode() below.
2169 * Otherwise the pool's process_* function pointers may
2170 * not match the desired pool mode.
2172 pt->adjusted_pf.mode = old_mode;
2175 pool->pf = pt->adjusted_pf;
2176 pool->low_water_blocks = pt->low_water_blocks;
2178 set_pool_mode(pool, new_mode);
2183 static void unbind_control_target(struct pool *pool, struct dm_target *ti)
2189 /*----------------------------------------------------------------
2191 *--------------------------------------------------------------*/
2192 /* Initialize pool features. */
2193 static void pool_features_init(struct pool_features *pf)
2195 pf->mode = PM_WRITE;
2196 pf->zero_new_blocks = true;
2197 pf->discard_enabled = true;
2198 pf->discard_passdown = true;
2199 pf->error_if_no_space = false;
2202 static void __pool_destroy(struct pool *pool)
2204 __pool_table_remove(pool);
2206 if (dm_pool_metadata_close(pool->pmd) < 0)
2207 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
2209 dm_bio_prison_destroy(pool->prison);
2210 dm_kcopyd_client_destroy(pool->copier);
2213 destroy_workqueue(pool->wq);
2215 if (pool->next_mapping)
2216 mempool_free(pool->next_mapping, pool->mapping_pool);
2217 mempool_destroy(pool->mapping_pool);
2218 dm_deferred_set_destroy(pool->shared_read_ds);
2219 dm_deferred_set_destroy(pool->all_io_ds);
2223 static struct kmem_cache *_new_mapping_cache;
2225 static struct pool *pool_create(struct mapped_device *pool_md,
2226 struct block_device *metadata_dev,
2227 unsigned long block_size,
2228 int read_only, char **error)
2233 struct dm_pool_metadata *pmd;
2234 bool format_device = read_only ? false : true;
2236 pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
2238 *error = "Error creating metadata object";
2239 return (struct pool *)pmd;
2242 pool = kmalloc(sizeof(*pool), GFP_KERNEL);
2244 *error = "Error allocating memory for pool";
2245 err_p = ERR_PTR(-ENOMEM);
2250 pool->sectors_per_block = block_size;
2251 if (block_size & (block_size - 1))
2252 pool->sectors_per_block_shift = -1;
2254 pool->sectors_per_block_shift = __ffs(block_size);
2255 pool->low_water_blocks = 0;
2256 pool_features_init(&pool->pf);
2257 pool->prison = dm_bio_prison_create();
2258 if (!pool->prison) {
2259 *error = "Error creating pool's bio prison";
2260 err_p = ERR_PTR(-ENOMEM);
2264 pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2265 if (IS_ERR(pool->copier)) {
2266 r = PTR_ERR(pool->copier);
2267 *error = "Error creating pool's kcopyd client";
2269 goto bad_kcopyd_client;
2273 * Create singlethreaded workqueue that will service all devices
2274 * that use this metadata.
2276 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
2278 *error = "Error creating pool's workqueue";
2279 err_p = ERR_PTR(-ENOMEM);
2283 throttle_init(&pool->throttle);
2284 INIT_WORK(&pool->worker, do_worker);
2285 INIT_DELAYED_WORK(&pool->waker, do_waker);
2286 INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout);
2287 spin_lock_init(&pool->lock);
2288 bio_list_init(&pool->deferred_flush_bios);
2289 INIT_LIST_HEAD(&pool->prepared_mappings);
2290 INIT_LIST_HEAD(&pool->prepared_discards);
2291 INIT_LIST_HEAD(&pool->active_thins);
2292 pool->low_water_triggered = false;
2294 pool->shared_read_ds = dm_deferred_set_create();
2295 if (!pool->shared_read_ds) {
2296 *error = "Error creating pool's shared read deferred set";
2297 err_p = ERR_PTR(-ENOMEM);
2298 goto bad_shared_read_ds;
2301 pool->all_io_ds = dm_deferred_set_create();
2302 if (!pool->all_io_ds) {
2303 *error = "Error creating pool's all io deferred set";
2304 err_p = ERR_PTR(-ENOMEM);
2308 pool->next_mapping = NULL;
2309 pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE,
2310 _new_mapping_cache);
2311 if (!pool->mapping_pool) {
2312 *error = "Error creating pool's mapping mempool";
2313 err_p = ERR_PTR(-ENOMEM);
2314 goto bad_mapping_pool;
2317 pool->ref_count = 1;
2318 pool->last_commit_jiffies = jiffies;
2319 pool->pool_md = pool_md;
2320 pool->md_dev = metadata_dev;
2321 __pool_table_insert(pool);
2326 dm_deferred_set_destroy(pool->all_io_ds);
2328 dm_deferred_set_destroy(pool->shared_read_ds);
2330 destroy_workqueue(pool->wq);
2332 dm_kcopyd_client_destroy(pool->copier);
2334 dm_bio_prison_destroy(pool->prison);
2338 if (dm_pool_metadata_close(pmd))
2339 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
2344 static void __pool_inc(struct pool *pool)
2346 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
2350 static void __pool_dec(struct pool *pool)
2352 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
2353 BUG_ON(!pool->ref_count);
2354 if (!--pool->ref_count)
2355 __pool_destroy(pool);
2358 static struct pool *__pool_find(struct mapped_device *pool_md,
2359 struct block_device *metadata_dev,
2360 unsigned long block_size, int read_only,
2361 char **error, int *created)
2363 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
2366 if (pool->pool_md != pool_md) {
2367 *error = "metadata device already in use by a pool";
2368 return ERR_PTR(-EBUSY);
2373 pool = __pool_table_lookup(pool_md);
2375 if (pool->md_dev != metadata_dev) {
2376 *error = "different pool cannot replace a pool";
2377 return ERR_PTR(-EINVAL);
2382 pool = pool_create(pool_md, metadata_dev, block_size, read_only, error);
2390 /*----------------------------------------------------------------
2391 * Pool target methods
2392 *--------------------------------------------------------------*/
2393 static void pool_dtr(struct dm_target *ti)
2395 struct pool_c *pt = ti->private;
2397 mutex_lock(&dm_thin_pool_table.mutex);
2399 unbind_control_target(pt->pool, ti);
2400 __pool_dec(pt->pool);
2401 dm_put_device(ti, pt->metadata_dev);
2402 dm_put_device(ti, pt->data_dev);
2405 mutex_unlock(&dm_thin_pool_table.mutex);
2408 static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
2409 struct dm_target *ti)
2413 const char *arg_name;
2415 static struct dm_arg _args[] = {
2416 {0, 4, "Invalid number of pool feature arguments"},
2420 * No feature arguments supplied.
2425 r = dm_read_arg_group(_args, as, &argc, &ti->error);
2429 while (argc && !r) {
2430 arg_name = dm_shift_arg(as);
2433 if (!strcasecmp(arg_name, "skip_block_zeroing"))
2434 pf->zero_new_blocks = false;
2436 else if (!strcasecmp(arg_name, "ignore_discard"))
2437 pf->discard_enabled = false;
2439 else if (!strcasecmp(arg_name, "no_discard_passdown"))
2440 pf->discard_passdown = false;
2442 else if (!strcasecmp(arg_name, "read_only"))
2443 pf->mode = PM_READ_ONLY;
2445 else if (!strcasecmp(arg_name, "error_if_no_space"))
2446 pf->error_if_no_space = true;
2449 ti->error = "Unrecognised pool feature requested";
2458 static void metadata_low_callback(void *context)
2460 struct pool *pool = context;
2462 DMWARN("%s: reached low water mark for metadata device: sending event.",
2463 dm_device_name(pool->pool_md));
2465 dm_table_event(pool->ti->table);
2468 static sector_t get_dev_size(struct block_device *bdev)
2470 return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
2473 static void warn_if_metadata_device_too_big(struct block_device *bdev)
2475 sector_t metadata_dev_size = get_dev_size(bdev);
2476 char buffer[BDEVNAME_SIZE];
2478 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING)
2479 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
2480 bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS);
2483 static sector_t get_metadata_dev_size(struct block_device *bdev)
2485 sector_t metadata_dev_size = get_dev_size(bdev);
2487 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS)
2488 metadata_dev_size = THIN_METADATA_MAX_SECTORS;
2490 return metadata_dev_size;
2493 static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
2495 sector_t metadata_dev_size = get_metadata_dev_size(bdev);
2497 sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE);
2499 return metadata_dev_size;
2503 * When a metadata threshold is crossed a dm event is triggered, and
2504 * userland should respond by growing the metadata device. We could let
2505 * userland set the threshold, like we do with the data threshold, but I'm
2506 * not sure they know enough to do this well.
2508 static dm_block_t calc_metadata_threshold(struct pool_c *pt)
2511 * 4M is ample for all ops with the possible exception of thin
2512 * device deletion which is harmless if it fails (just retry the
2513 * delete after you've grown the device).
2515 dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
2516 return min((dm_block_t)1024ULL /* 4M */, quarter);
2520 * thin-pool <metadata dev> <data dev>
2521 * <data block size (sectors)>
2522 * <low water mark (blocks)>
2523 * [<#feature args> [<arg>]*]
2525 * Optional feature arguments are:
2526 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
2527 * ignore_discard: disable discard
2528 * no_discard_passdown: don't pass discards down to the data device
2529 * read_only: Don't allow any changes to be made to the pool metadata.
2530 * error_if_no_space: error IOs, instead of queueing, if no space.
2532 static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
2534 int r, pool_created = 0;
2537 struct pool_features pf;
2538 struct dm_arg_set as;
2539 struct dm_dev *data_dev;
2540 unsigned long block_size;
2541 dm_block_t low_water_blocks;
2542 struct dm_dev *metadata_dev;
2543 fmode_t metadata_mode;
2546 * FIXME Remove validation from scope of lock.
2548 mutex_lock(&dm_thin_pool_table.mutex);
2551 ti->error = "Invalid argument count";
2560 * Set default pool features.
2562 pool_features_init(&pf);
2564 dm_consume_args(&as, 4);
2565 r = parse_pool_features(&as, &pf, ti);
2569 metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
2570 r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
2572 ti->error = "Error opening metadata block device";
2575 warn_if_metadata_device_too_big(metadata_dev->bdev);
2577 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
2579 ti->error = "Error getting data device";
2583 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
2584 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
2585 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
2586 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
2587 ti->error = "Invalid block size";
2592 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
2593 ti->error = "Invalid low water mark";
2598 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
2604 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
2605 block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
2612 * 'pool_created' reflects whether this is the first table load.
2613 * Top level discard support is not allowed to be changed after
2614 * initial load. This would require a pool reload to trigger thin
2617 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
2618 ti->error = "Discard support cannot be disabled once enabled";
2620 goto out_flags_changed;
2625 pt->metadata_dev = metadata_dev;
2626 pt->data_dev = data_dev;
2627 pt->low_water_blocks = low_water_blocks;
2628 pt->adjusted_pf = pt->requested_pf = pf;
2629 ti->num_flush_bios = 1;
2632 * Only need to enable discards if the pool should pass
2633 * them down to the data device. The thin device's discard
2634 * processing will cause mappings to be removed from the btree.
2636 ti->discard_zeroes_data_unsupported = true;
2637 if (pf.discard_enabled && pf.discard_passdown) {
2638 ti->num_discard_bios = 1;
2641 * Setting 'discards_supported' circumvents the normal
2642 * stacking of discard limits (this keeps the pool and
2643 * thin devices' discard limits consistent).
2645 ti->discards_supported = true;
2649 r = dm_pool_register_metadata_threshold(pt->pool->pmd,
2650 calc_metadata_threshold(pt),
2651 metadata_low_callback,
2656 pt->callbacks.congested_fn = pool_is_congested;
2657 dm_table_add_target_callbacks(ti->table, &pt->callbacks);
2659 mutex_unlock(&dm_thin_pool_table.mutex);
2668 dm_put_device(ti, data_dev);
2670 dm_put_device(ti, metadata_dev);
2672 mutex_unlock(&dm_thin_pool_table.mutex);
2677 static int pool_map(struct dm_target *ti, struct bio *bio)
2680 struct pool_c *pt = ti->private;
2681 struct pool *pool = pt->pool;
2682 unsigned long flags;
2685 * As this is a singleton target, ti->begin is always zero.
2687 spin_lock_irqsave(&pool->lock, flags);
2688 bio->bi_bdev = pt->data_dev->bdev;
2689 r = DM_MAPIO_REMAPPED;
2690 spin_unlock_irqrestore(&pool->lock, flags);
2695 static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
2698 struct pool_c *pt = ti->private;
2699 struct pool *pool = pt->pool;
2700 sector_t data_size = ti->len;
2701 dm_block_t sb_data_size;
2703 *need_commit = false;
2705 (void) sector_div(data_size, pool->sectors_per_block);
2707 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
2709 DMERR("%s: failed to retrieve data device size",
2710 dm_device_name(pool->pool_md));
2714 if (data_size < sb_data_size) {
2715 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
2716 dm_device_name(pool->pool_md),
2717 (unsigned long long)data_size, sb_data_size);
2720 } else if (data_size > sb_data_size) {
2721 if (dm_pool_metadata_needs_check(pool->pmd)) {
2722 DMERR("%s: unable to grow the data device until repaired.",
2723 dm_device_name(pool->pool_md));
2728 DMINFO("%s: growing the data device from %llu to %llu blocks",
2729 dm_device_name(pool->pool_md),
2730 sb_data_size, (unsigned long long)data_size);
2731 r = dm_pool_resize_data_dev(pool->pmd, data_size);
2733 metadata_operation_failed(pool, "dm_pool_resize_data_dev", r);
2737 *need_commit = true;
2743 static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
2746 struct pool_c *pt = ti->private;
2747 struct pool *pool = pt->pool;
2748 dm_block_t metadata_dev_size, sb_metadata_dev_size;
2750 *need_commit = false;
2752 metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
2754 r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
2756 DMERR("%s: failed to retrieve metadata device size",
2757 dm_device_name(pool->pool_md));
2761 if (metadata_dev_size < sb_metadata_dev_size) {
2762 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
2763 dm_device_name(pool->pool_md),
2764 metadata_dev_size, sb_metadata_dev_size);
2767 } else if (metadata_dev_size > sb_metadata_dev_size) {
2768 if (dm_pool_metadata_needs_check(pool->pmd)) {
2769 DMERR("%s: unable to grow the metadata device until repaired.",
2770 dm_device_name(pool->pool_md));
2774 warn_if_metadata_device_too_big(pool->md_dev);
2775 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
2776 dm_device_name(pool->pool_md),
2777 sb_metadata_dev_size, metadata_dev_size);
2778 r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
2780 metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r);
2784 *need_commit = true;
2791 * Retrieves the number of blocks of the data device from
2792 * the superblock and compares it to the actual device size,
2793 * thus resizing the data device in case it has grown.
2795 * This both copes with opening preallocated data devices in the ctr
2796 * being followed by a resume
2798 * calling the resume method individually after userspace has
2799 * grown the data device in reaction to a table event.
2801 static int pool_preresume(struct dm_target *ti)
2804 bool need_commit1, need_commit2;
2805 struct pool_c *pt = ti->private;
2806 struct pool *pool = pt->pool;
2809 * Take control of the pool object.
2811 r = bind_control_target(pool, ti);
2815 r = maybe_resize_data_dev(ti, &need_commit1);
2819 r = maybe_resize_metadata_dev(ti, &need_commit2);
2823 if (need_commit1 || need_commit2)
2824 (void) commit(pool);
2829 static void pool_resume(struct dm_target *ti)
2831 struct pool_c *pt = ti->private;
2832 struct pool *pool = pt->pool;
2833 unsigned long flags;
2835 spin_lock_irqsave(&pool->lock, flags);
2836 pool->low_water_triggered = false;
2837 spin_unlock_irqrestore(&pool->lock, flags);
2840 do_waker(&pool->waker.work);
2843 static void pool_postsuspend(struct dm_target *ti)
2845 struct pool_c *pt = ti->private;
2846 struct pool *pool = pt->pool;
2848 cancel_delayed_work(&pool->waker);
2849 cancel_delayed_work(&pool->no_space_timeout);
2850 flush_workqueue(pool->wq);
2851 (void) commit(pool);
2854 static int check_arg_count(unsigned argc, unsigned args_required)
2856 if (argc != args_required) {
2857 DMWARN("Message received with %u arguments instead of %u.",
2858 argc, args_required);
2865 static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
2867 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
2868 *dev_id <= MAX_DEV_ID)
2872 DMWARN("Message received with invalid device id: %s", arg);
2877 static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
2882 r = check_arg_count(argc, 2);
2886 r = read_dev_id(argv[1], &dev_id, 1);
2890 r = dm_pool_create_thin(pool->pmd, dev_id);
2892 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2900 static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2903 dm_thin_id origin_dev_id;
2906 r = check_arg_count(argc, 3);
2910 r = read_dev_id(argv[1], &dev_id, 1);
2914 r = read_dev_id(argv[2], &origin_dev_id, 1);
2918 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
2920 DMWARN("Creation of new snapshot %s of device %s failed.",
2928 static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
2933 r = check_arg_count(argc, 2);
2937 r = read_dev_id(argv[1], &dev_id, 1);
2941 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
2943 DMWARN("Deletion of thin device %s failed.", argv[1]);
2948 static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
2950 dm_thin_id old_id, new_id;
2953 r = check_arg_count(argc, 3);
2957 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
2958 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
2962 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
2963 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
2967 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
2969 DMWARN("Failed to change transaction id from %s to %s.",
2977 static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2981 r = check_arg_count(argc, 1);
2985 (void) commit(pool);
2987 r = dm_pool_reserve_metadata_snap(pool->pmd);
2989 DMWARN("reserve_metadata_snap message failed.");
2994 static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2998 r = check_arg_count(argc, 1);
3002 r = dm_pool_release_metadata_snap(pool->pmd);
3004 DMWARN("release_metadata_snap message failed.");
3010 * Messages supported:
3011 * create_thin <dev_id>
3012 * create_snap <dev_id> <origin_id>
3014 * trim <dev_id> <new_size_in_sectors>
3015 * set_transaction_id <current_trans_id> <new_trans_id>
3016 * reserve_metadata_snap
3017 * release_metadata_snap
3019 static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
3022 struct pool_c *pt = ti->private;
3023 struct pool *pool = pt->pool;
3025 if (!strcasecmp(argv[0], "create_thin"))
3026 r = process_create_thin_mesg(argc, argv, pool);
3028 else if (!strcasecmp(argv[0], "create_snap"))
3029 r = process_create_snap_mesg(argc, argv, pool);
3031 else if (!strcasecmp(argv[0], "delete"))
3032 r = process_delete_mesg(argc, argv, pool);
3034 else if (!strcasecmp(argv[0], "set_transaction_id"))
3035 r = process_set_transaction_id_mesg(argc, argv, pool);
3037 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
3038 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
3040 else if (!strcasecmp(argv[0], "release_metadata_snap"))
3041 r = process_release_metadata_snap_mesg(argc, argv, pool);
3044 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
3047 (void) commit(pool);
3052 static void emit_flags(struct pool_features *pf, char *result,
3053 unsigned sz, unsigned maxlen)
3055 unsigned count = !pf->zero_new_blocks + !pf->discard_enabled +
3056 !pf->discard_passdown + (pf->mode == PM_READ_ONLY) +
3057 pf->error_if_no_space;
3058 DMEMIT("%u ", count);
3060 if (!pf->zero_new_blocks)
3061 DMEMIT("skip_block_zeroing ");
3063 if (!pf->discard_enabled)
3064 DMEMIT("ignore_discard ");
3066 if (!pf->discard_passdown)
3067 DMEMIT("no_discard_passdown ");
3069 if (pf->mode == PM_READ_ONLY)
3070 DMEMIT("read_only ");
3072 if (pf->error_if_no_space)
3073 DMEMIT("error_if_no_space ");
3078 * <transaction id> <used metadata sectors>/<total metadata sectors>
3079 * <used data sectors>/<total data sectors> <held metadata root>
3081 static void pool_status(struct dm_target *ti, status_type_t type,
3082 unsigned status_flags, char *result, unsigned maxlen)
3086 uint64_t transaction_id;
3087 dm_block_t nr_free_blocks_data;
3088 dm_block_t nr_free_blocks_metadata;
3089 dm_block_t nr_blocks_data;
3090 dm_block_t nr_blocks_metadata;
3091 dm_block_t held_root;
3092 char buf[BDEVNAME_SIZE];
3093 char buf2[BDEVNAME_SIZE];
3094 struct pool_c *pt = ti->private;
3095 struct pool *pool = pt->pool;
3098 case STATUSTYPE_INFO:
3099 if (get_pool_mode(pool) == PM_FAIL) {
3104 /* Commit to ensure statistics aren't out-of-date */
3105 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
3106 (void) commit(pool);
3108 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
3110 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
3111 dm_device_name(pool->pool_md), r);
3115 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
3117 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
3118 dm_device_name(pool->pool_md), r);
3122 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
3124 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
3125 dm_device_name(pool->pool_md), r);
3129 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
3131 DMERR("%s: dm_pool_get_free_block_count returned %d",
3132 dm_device_name(pool->pool_md), r);
3136 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
3138 DMERR("%s: dm_pool_get_data_dev_size returned %d",
3139 dm_device_name(pool->pool_md), r);
3143 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
3145 DMERR("%s: dm_pool_get_metadata_snap returned %d",
3146 dm_device_name(pool->pool_md), r);
3150 DMEMIT("%llu %llu/%llu %llu/%llu ",
3151 (unsigned long long)transaction_id,
3152 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
3153 (unsigned long long)nr_blocks_metadata,
3154 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
3155 (unsigned long long)nr_blocks_data);
3158 DMEMIT("%llu ", held_root);
3162 if (pool->pf.mode == PM_OUT_OF_DATA_SPACE)
3163 DMEMIT("out_of_data_space ");
3164 else if (pool->pf.mode == PM_READ_ONLY)
3169 if (!pool->pf.discard_enabled)
3170 DMEMIT("ignore_discard ");
3171 else if (pool->pf.discard_passdown)
3172 DMEMIT("discard_passdown ");
3174 DMEMIT("no_discard_passdown ");
3176 if (pool->pf.error_if_no_space)
3177 DMEMIT("error_if_no_space ");
3179 DMEMIT("queue_if_no_space ");
3183 case STATUSTYPE_TABLE:
3184 DMEMIT("%s %s %lu %llu ",
3185 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
3186 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
3187 (unsigned long)pool->sectors_per_block,
3188 (unsigned long long)pt->low_water_blocks);
3189 emit_flags(&pt->requested_pf, result, sz, maxlen);
3198 static int pool_iterate_devices(struct dm_target *ti,
3199 iterate_devices_callout_fn fn, void *data)
3201 struct pool_c *pt = ti->private;
3203 return fn(ti, pt->data_dev, 0, ti->len, data);
3206 static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
3207 struct bio_vec *biovec, int max_size)
3209 struct pool_c *pt = ti->private;
3210 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
3212 if (!q->merge_bvec_fn)
3215 bvm->bi_bdev = pt->data_dev->bdev;
3217 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
3220 static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits)
3222 struct pool *pool = pt->pool;
3223 struct queue_limits *data_limits;
3225 limits->max_discard_sectors = pool->sectors_per_block;
3228 * discard_granularity is just a hint, and not enforced.
3230 if (pt->adjusted_pf.discard_passdown) {
3231 data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits;
3232 limits->discard_granularity = max(data_limits->discard_granularity,
3233 pool->sectors_per_block << SECTOR_SHIFT);
3235 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
3238 static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
3240 struct pool_c *pt = ti->private;
3241 struct pool *pool = pt->pool;
3242 sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
3245 * Adjust max_sectors_kb to highest possible power-of-2
3246 * factor of pool->sectors_per_block.
3248 if (limits->max_hw_sectors & (limits->max_hw_sectors - 1))
3249 limits->max_sectors = rounddown_pow_of_two(limits->max_hw_sectors);
3251 limits->max_sectors = limits->max_hw_sectors;
3253 if (limits->max_sectors < pool->sectors_per_block) {
3254 while (!is_factor(pool->sectors_per_block, limits->max_sectors)) {
3255 if ((limits->max_sectors & (limits->max_sectors - 1)) == 0)
3256 limits->max_sectors--;
3257 limits->max_sectors = rounddown_pow_of_two(limits->max_sectors);
3259 } else if (block_size_is_power_of_two(pool)) {
3260 /* max_sectors_kb is >= power-of-2 thinp blocksize */
3261 while (!is_factor(limits->max_sectors, pool->sectors_per_block)) {
3262 if ((limits->max_sectors & (limits->max_sectors - 1)) == 0)
3263 limits->max_sectors--;
3264 limits->max_sectors = rounddown_pow_of_two(limits->max_sectors);
3269 * If the system-determined stacked limits are compatible with the
3270 * pool's blocksize (io_opt is a factor) do not override them.
3272 if (io_opt_sectors < pool->sectors_per_block ||
3273 !is_factor(io_opt_sectors, pool->sectors_per_block)) {
3274 if (is_factor(pool->sectors_per_block, limits->max_sectors))
3275 blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT);
3277 blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT);
3278 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
3282 * pt->adjusted_pf is a staging area for the actual features to use.
3283 * They get transferred to the live pool in bind_control_target()
3284 * called from pool_preresume().
3286 if (!pt->adjusted_pf.discard_enabled) {
3288 * Must explicitly disallow stacking discard limits otherwise the
3289 * block layer will stack them if pool's data device has support.
3290 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
3291 * user to see that, so make sure to set all discard limits to 0.
3293 limits->discard_granularity = 0;
3297 disable_passdown_if_not_supported(pt);
3299 set_discard_limits(pt, limits);
3302 static struct target_type pool_target = {
3303 .name = "thin-pool",
3304 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
3305 DM_TARGET_IMMUTABLE,
3306 .version = {1, 14, 0},
3307 .module = THIS_MODULE,
3311 .postsuspend = pool_postsuspend,
3312 .preresume = pool_preresume,
3313 .resume = pool_resume,
3314 .message = pool_message,
3315 .status = pool_status,
3316 .merge = pool_merge,
3317 .iterate_devices = pool_iterate_devices,
3318 .io_hints = pool_io_hints,
3321 /*----------------------------------------------------------------
3322 * Thin target methods
3323 *--------------------------------------------------------------*/
3324 static void thin_get(struct thin_c *tc)
3326 atomic_inc(&tc->refcount);
3329 static void thin_put(struct thin_c *tc)
3331 if (atomic_dec_and_test(&tc->refcount))
3332 complete(&tc->can_destroy);
3335 static void thin_dtr(struct dm_target *ti)
3337 struct thin_c *tc = ti->private;
3338 unsigned long flags;
3341 wait_for_completion(&tc->can_destroy);
3343 spin_lock_irqsave(&tc->pool->lock, flags);
3344 list_del_rcu(&tc->list);
3345 spin_unlock_irqrestore(&tc->pool->lock, flags);
3348 mutex_lock(&dm_thin_pool_table.mutex);
3350 __pool_dec(tc->pool);
3351 dm_pool_close_thin_device(tc->td);
3352 dm_put_device(ti, tc->pool_dev);
3354 dm_put_device(ti, tc->origin_dev);
3357 mutex_unlock(&dm_thin_pool_table.mutex);
3361 * Thin target parameters:
3363 * <pool_dev> <dev_id> [origin_dev]
3365 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
3366 * dev_id: the internal device identifier
3367 * origin_dev: a device external to the pool that should act as the origin
3369 * If the pool device has discards disabled, they get disabled for the thin
3372 static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
3376 struct dm_dev *pool_dev, *origin_dev;
3377 struct mapped_device *pool_md;
3378 unsigned long flags;
3380 mutex_lock(&dm_thin_pool_table.mutex);
3382 if (argc != 2 && argc != 3) {
3383 ti->error = "Invalid argument count";
3388 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
3390 ti->error = "Out of memory";
3394 spin_lock_init(&tc->lock);
3395 bio_list_init(&tc->deferred_bio_list);
3396 bio_list_init(&tc->retry_on_resume_list);
3397 tc->sort_bio_list = RB_ROOT;
3400 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
3402 ti->error = "Error opening origin device";
3403 goto bad_origin_dev;
3405 tc->origin_dev = origin_dev;
3408 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
3410 ti->error = "Error opening pool device";
3413 tc->pool_dev = pool_dev;
3415 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
3416 ti->error = "Invalid device id";
3421 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
3423 ti->error = "Couldn't get pool mapped device";
3428 tc->pool = __pool_table_lookup(pool_md);
3430 ti->error = "Couldn't find pool object";
3432 goto bad_pool_lookup;
3434 __pool_inc(tc->pool);
3436 if (get_pool_mode(tc->pool) == PM_FAIL) {
3437 ti->error = "Couldn't open thin device, Pool is in fail mode";
3442 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
3444 ti->error = "Couldn't open thin internal device";
3448 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
3450 goto bad_target_max_io_len;
3452 ti->num_flush_bios = 1;
3453 ti->flush_supported = true;
3454 ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook);
3456 /* In case the pool supports discards, pass them on. */
3457 ti->discard_zeroes_data_unsupported = true;
3458 if (tc->pool->pf.discard_enabled) {
3459 ti->discards_supported = true;
3460 ti->num_discard_bios = 1;
3461 /* Discard bios must be split on a block boundary */
3462 ti->split_discard_bios = true;
3467 mutex_unlock(&dm_thin_pool_table.mutex);
3469 atomic_set(&tc->refcount, 1);
3470 init_completion(&tc->can_destroy);
3472 spin_lock_irqsave(&tc->pool->lock, flags);
3473 list_add_tail_rcu(&tc->list, &tc->pool->active_thins);
3474 spin_unlock_irqrestore(&tc->pool->lock, flags);
3476 * This synchronize_rcu() call is needed here otherwise we risk a
3477 * wake_worker() call finding no bios to process (because the newly
3478 * added tc isn't yet visible). So this reduces latency since we
3479 * aren't then dependent on the periodic commit to wake_worker().
3485 bad_target_max_io_len:
3486 dm_pool_close_thin_device(tc->td);
3488 __pool_dec(tc->pool);
3492 dm_put_device(ti, tc->pool_dev);
3495 dm_put_device(ti, tc->origin_dev);
3499 mutex_unlock(&dm_thin_pool_table.mutex);
3504 static int thin_map(struct dm_target *ti, struct bio *bio)
3506 bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
3508 return thin_bio_map(ti, bio);
3511 static int thin_endio(struct dm_target *ti, struct bio *bio, int err)
3513 unsigned long flags;
3514 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
3515 struct list_head work;
3516 struct dm_thin_new_mapping *m, *tmp;
3517 struct pool *pool = h->tc->pool;
3519 if (h->shared_read_entry) {
3520 INIT_LIST_HEAD(&work);
3521 dm_deferred_entry_dec(h->shared_read_entry, &work);
3523 spin_lock_irqsave(&pool->lock, flags);
3524 list_for_each_entry_safe(m, tmp, &work, list) {
3526 __complete_mapping_preparation(m);
3528 spin_unlock_irqrestore(&pool->lock, flags);
3531 if (h->all_io_entry) {
3532 INIT_LIST_HEAD(&work);
3533 dm_deferred_entry_dec(h->all_io_entry, &work);
3534 if (!list_empty(&work)) {
3535 spin_lock_irqsave(&pool->lock, flags);
3536 list_for_each_entry_safe(m, tmp, &work, list)
3537 list_add_tail(&m->list, &pool->prepared_discards);
3538 spin_unlock_irqrestore(&pool->lock, flags);
3546 static void thin_presuspend(struct dm_target *ti)
3548 struct thin_c *tc = ti->private;
3550 if (dm_noflush_suspending(ti))
3551 noflush_work(tc, do_noflush_start);
3554 static void thin_postsuspend(struct dm_target *ti)
3556 struct thin_c *tc = ti->private;
3559 * The dm_noflush_suspending flag has been cleared by now, so
3560 * unfortunately we must always run this.
3562 noflush_work(tc, do_noflush_stop);
3565 static int thin_preresume(struct dm_target *ti)
3567 struct thin_c *tc = ti->private;
3570 tc->origin_size = get_dev_size(tc->origin_dev->bdev);
3576 * <nr mapped sectors> <highest mapped sector>
3578 static void thin_status(struct dm_target *ti, status_type_t type,
3579 unsigned status_flags, char *result, unsigned maxlen)
3583 dm_block_t mapped, highest;
3584 char buf[BDEVNAME_SIZE];
3585 struct thin_c *tc = ti->private;
3587 if (get_pool_mode(tc->pool) == PM_FAIL) {
3596 case STATUSTYPE_INFO:
3597 r = dm_thin_get_mapped_count(tc->td, &mapped);
3599 DMERR("dm_thin_get_mapped_count returned %d", r);
3603 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
3605 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
3609 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
3611 DMEMIT("%llu", ((highest + 1) *
3612 tc->pool->sectors_per_block) - 1);
3617 case STATUSTYPE_TABLE:
3619 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
3620 (unsigned long) tc->dev_id);
3622 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
3633 static int thin_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
3634 struct bio_vec *biovec, int max_size)
3636 struct thin_c *tc = ti->private;
3637 struct request_queue *q = bdev_get_queue(tc->pool_dev->bdev);
3639 if (!q->merge_bvec_fn)
3642 bvm->bi_bdev = tc->pool_dev->bdev;
3643 bvm->bi_sector = dm_target_offset(ti, bvm->bi_sector);
3645 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
3648 static int thin_iterate_devices(struct dm_target *ti,
3649 iterate_devices_callout_fn fn, void *data)
3652 struct thin_c *tc = ti->private;
3653 struct pool *pool = tc->pool;
3656 * We can't call dm_pool_get_data_dev_size() since that blocks. So
3657 * we follow a more convoluted path through to the pool's target.
3660 return 0; /* nothing is bound */
3662 blocks = pool->ti->len;
3663 (void) sector_div(blocks, pool->sectors_per_block);
3665 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
3670 static struct target_type thin_target = {
3672 .version = {1, 14, 0},
3673 .module = THIS_MODULE,
3677 .end_io = thin_endio,
3678 .preresume = thin_preresume,
3679 .presuspend = thin_presuspend,
3680 .postsuspend = thin_postsuspend,
3681 .status = thin_status,
3682 .merge = thin_merge,
3683 .iterate_devices = thin_iterate_devices,
3686 /*----------------------------------------------------------------*/
3688 static int __init dm_thin_init(void)
3694 r = dm_register_target(&thin_target);
3698 r = dm_register_target(&pool_target);
3700 goto bad_pool_target;
3704 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
3705 if (!_new_mapping_cache)
3706 goto bad_new_mapping_cache;
3710 bad_new_mapping_cache:
3711 dm_unregister_target(&pool_target);
3713 dm_unregister_target(&thin_target);
3718 static void dm_thin_exit(void)
3720 dm_unregister_target(&thin_target);
3721 dm_unregister_target(&pool_target);
3723 kmem_cache_destroy(_new_mapping_cache);
3726 module_init(dm_thin_init);
3727 module_exit(dm_thin_exit);
3729 module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR);
3730 MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds");
3732 MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
3733 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3734 MODULE_LICENSE("GPL");