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/rculist.h>
16 #include <linux/init.h>
17 #include <linux/module.h>
18 #include <linux/slab.h>
19 #include <linux/rbtree.h>
21 #define DM_MSG_PREFIX "thin"
26 #define ENDIO_HOOK_POOL_SIZE 1024
27 #define MAPPING_POOL_SIZE 1024
28 #define PRISON_CELLS 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 /*----------------------------------------------------------------*/
131 * A pool device ties together a metadata device and a data device. It
132 * also provides the interface for creating and destroying internal
135 struct dm_thin_new_mapping;
138 * The pool runs in 4 modes. Ordered in degraded order for comparisons.
141 PM_WRITE, /* metadata may be changed */
142 PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */
143 PM_READ_ONLY, /* metadata may not be changed */
144 PM_FAIL, /* all I/O fails */
147 struct pool_features {
150 bool zero_new_blocks:1;
151 bool discard_enabled:1;
152 bool discard_passdown:1;
153 bool error_if_no_space:1;
157 typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
158 typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);
161 struct list_head list;
162 struct dm_target *ti; /* Only set if a pool target is bound */
164 struct mapped_device *pool_md;
165 struct block_device *md_dev;
166 struct dm_pool_metadata *pmd;
168 dm_block_t low_water_blocks;
169 uint32_t sectors_per_block;
170 int sectors_per_block_shift;
172 struct pool_features pf;
173 bool low_water_triggered:1; /* A dm event has been sent */
175 struct dm_bio_prison *prison;
176 struct dm_kcopyd_client *copier;
178 struct workqueue_struct *wq;
179 struct work_struct worker;
180 struct delayed_work waker;
181 struct delayed_work no_space_timeout;
183 unsigned long last_commit_jiffies;
187 struct bio_list deferred_flush_bios;
188 struct list_head prepared_mappings;
189 struct list_head prepared_discards;
190 struct list_head active_thins;
192 struct dm_deferred_set *shared_read_ds;
193 struct dm_deferred_set *all_io_ds;
195 struct dm_thin_new_mapping *next_mapping;
196 mempool_t *mapping_pool;
198 process_bio_fn process_bio;
199 process_bio_fn process_discard;
201 process_mapping_fn process_prepared_mapping;
202 process_mapping_fn process_prepared_discard;
205 static enum pool_mode get_pool_mode(struct pool *pool);
206 static void metadata_operation_failed(struct pool *pool, const char *op, int r);
209 * Target context for a pool.
212 struct dm_target *ti;
214 struct dm_dev *data_dev;
215 struct dm_dev *metadata_dev;
216 struct dm_target_callbacks callbacks;
218 dm_block_t low_water_blocks;
219 struct pool_features requested_pf; /* Features requested during table load */
220 struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */
224 * Target context for a thin.
227 struct list_head list;
228 struct dm_dev *pool_dev;
229 struct dm_dev *origin_dev;
230 sector_t origin_size;
234 struct dm_thin_device *td;
237 struct bio_list deferred_bio_list;
238 struct bio_list retry_on_resume_list;
239 struct rb_root sort_bio_list; /* sorted list of deferred bios */
242 * Ensures the thin is not destroyed until the worker has finished
243 * iterating the active_thins list.
246 struct completion can_destroy;
249 /*----------------------------------------------------------------*/
252 * wake_worker() is used when new work is queued and when pool_resume is
253 * ready to continue deferred IO processing.
255 static void wake_worker(struct pool *pool)
257 queue_work(pool->wq, &pool->worker);
260 /*----------------------------------------------------------------*/
262 static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio,
263 struct dm_bio_prison_cell **cell_result)
266 struct dm_bio_prison_cell *cell_prealloc;
269 * Allocate a cell from the prison's mempool.
270 * This might block but it can't fail.
272 cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO);
274 r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result);
277 * We reused an old cell; we can get rid of
280 dm_bio_prison_free_cell(pool->prison, cell_prealloc);
285 static void cell_release(struct pool *pool,
286 struct dm_bio_prison_cell *cell,
287 struct bio_list *bios)
289 dm_cell_release(pool->prison, cell, bios);
290 dm_bio_prison_free_cell(pool->prison, cell);
293 static void cell_release_no_holder(struct pool *pool,
294 struct dm_bio_prison_cell *cell,
295 struct bio_list *bios)
297 dm_cell_release_no_holder(pool->prison, cell, bios);
298 dm_bio_prison_free_cell(pool->prison, cell);
301 static void cell_defer_no_holder_no_free(struct thin_c *tc,
302 struct dm_bio_prison_cell *cell)
304 struct pool *pool = tc->pool;
307 spin_lock_irqsave(&tc->lock, flags);
308 dm_cell_release_no_holder(pool->prison, cell, &tc->deferred_bio_list);
309 spin_unlock_irqrestore(&tc->lock, flags);
314 static void cell_error_with_code(struct pool *pool,
315 struct dm_bio_prison_cell *cell, int error_code)
317 dm_cell_error(pool->prison, cell, error_code);
318 dm_bio_prison_free_cell(pool->prison, cell);
321 static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell)
323 cell_error_with_code(pool, cell, -EIO);
326 /*----------------------------------------------------------------*/
329 * A global list of pools that uses a struct mapped_device as a key.
331 static struct dm_thin_pool_table {
333 struct list_head pools;
334 } dm_thin_pool_table;
336 static void pool_table_init(void)
338 mutex_init(&dm_thin_pool_table.mutex);
339 INIT_LIST_HEAD(&dm_thin_pool_table.pools);
342 static void __pool_table_insert(struct pool *pool)
344 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
345 list_add(&pool->list, &dm_thin_pool_table.pools);
348 static void __pool_table_remove(struct pool *pool)
350 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
351 list_del(&pool->list);
354 static struct pool *__pool_table_lookup(struct mapped_device *md)
356 struct pool *pool = NULL, *tmp;
358 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
360 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
361 if (tmp->pool_md == md) {
370 static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
372 struct pool *pool = NULL, *tmp;
374 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
376 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
377 if (tmp->md_dev == md_dev) {
386 /*----------------------------------------------------------------*/
388 struct dm_thin_endio_hook {
390 struct dm_deferred_entry *shared_read_entry;
391 struct dm_deferred_entry *all_io_entry;
392 struct dm_thin_new_mapping *overwrite_mapping;
393 struct rb_node rb_node;
396 static void requeue_bio_list(struct thin_c *tc, struct bio_list *master)
399 struct bio_list bios;
402 bio_list_init(&bios);
404 spin_lock_irqsave(&tc->lock, flags);
405 bio_list_merge(&bios, master);
406 bio_list_init(master);
407 spin_unlock_irqrestore(&tc->lock, flags);
409 while ((bio = bio_list_pop(&bios)))
410 bio_endio(bio, DM_ENDIO_REQUEUE);
413 static void requeue_io(struct thin_c *tc)
415 requeue_bio_list(tc, &tc->deferred_bio_list);
416 requeue_bio_list(tc, &tc->retry_on_resume_list);
419 static void error_thin_retry_list(struct thin_c *tc)
423 struct bio_list bios;
425 bio_list_init(&bios);
427 spin_lock_irqsave(&tc->lock, flags);
428 bio_list_merge(&bios, &tc->retry_on_resume_list);
429 bio_list_init(&tc->retry_on_resume_list);
430 spin_unlock_irqrestore(&tc->lock, flags);
432 while ((bio = bio_list_pop(&bios)))
436 static void error_retry_list(struct pool *pool)
441 list_for_each_entry_rcu(tc, &pool->active_thins, list)
442 error_thin_retry_list(tc);
447 * This section of code contains the logic for processing a thin device's IO.
448 * Much of the code depends on pool object resources (lists, workqueues, etc)
449 * but most is exclusively called from the thin target rather than the thin-pool
453 static bool block_size_is_power_of_two(struct pool *pool)
455 return pool->sectors_per_block_shift >= 0;
458 static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
460 struct pool *pool = tc->pool;
461 sector_t block_nr = bio->bi_iter.bi_sector;
463 if (block_size_is_power_of_two(pool))
464 block_nr >>= pool->sectors_per_block_shift;
466 (void) sector_div(block_nr, pool->sectors_per_block);
471 static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
473 struct pool *pool = tc->pool;
474 sector_t bi_sector = bio->bi_iter.bi_sector;
476 bio->bi_bdev = tc->pool_dev->bdev;
477 if (block_size_is_power_of_two(pool))
478 bio->bi_iter.bi_sector =
479 (block << pool->sectors_per_block_shift) |
480 (bi_sector & (pool->sectors_per_block - 1));
482 bio->bi_iter.bi_sector = (block * pool->sectors_per_block) +
483 sector_div(bi_sector, pool->sectors_per_block);
486 static void remap_to_origin(struct thin_c *tc, struct bio *bio)
488 bio->bi_bdev = tc->origin_dev->bdev;
491 static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
493 return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) &&
494 dm_thin_changed_this_transaction(tc->td);
497 static void inc_all_io_entry(struct pool *pool, struct bio *bio)
499 struct dm_thin_endio_hook *h;
501 if (bio->bi_rw & REQ_DISCARD)
504 h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
505 h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
508 static void issue(struct thin_c *tc, struct bio *bio)
510 struct pool *pool = tc->pool;
513 if (!bio_triggers_commit(tc, bio)) {
514 generic_make_request(bio);
519 * Complete bio with an error if earlier I/O caused changes to
520 * the metadata that can't be committed e.g, due to I/O errors
521 * on the metadata device.
523 if (dm_thin_aborted_changes(tc->td)) {
529 * Batch together any bios that trigger commits and then issue a
530 * single commit for them in process_deferred_bios().
532 spin_lock_irqsave(&pool->lock, flags);
533 bio_list_add(&pool->deferred_flush_bios, bio);
534 spin_unlock_irqrestore(&pool->lock, flags);
537 static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
539 remap_to_origin(tc, bio);
543 static void remap_and_issue(struct thin_c *tc, struct bio *bio,
546 remap(tc, bio, block);
550 /*----------------------------------------------------------------*/
553 * Bio endio functions.
555 struct dm_thin_new_mapping {
556 struct list_head list;
559 bool definitely_not_shared:1;
562 * Track quiescing, copying and zeroing preparation actions. When this
563 * counter hits zero the block is prepared and can be inserted into the
566 atomic_t prepare_actions;
570 dm_block_t virt_block;
571 dm_block_t data_block;
572 struct dm_bio_prison_cell *cell, *cell2;
575 * If the bio covers the whole area of a block then we can avoid
576 * zeroing or copying. Instead this bio is hooked. The bio will
577 * still be in the cell, so care has to be taken to avoid issuing
581 bio_end_io_t *saved_bi_end_io;
584 static void __complete_mapping_preparation(struct dm_thin_new_mapping *m)
586 struct pool *pool = m->tc->pool;
588 if (atomic_dec_and_test(&m->prepare_actions)) {
589 list_add_tail(&m->list, &pool->prepared_mappings);
594 static void complete_mapping_preparation(struct dm_thin_new_mapping *m)
597 struct pool *pool = m->tc->pool;
599 spin_lock_irqsave(&pool->lock, flags);
600 __complete_mapping_preparation(m);
601 spin_unlock_irqrestore(&pool->lock, flags);
604 static void copy_complete(int read_err, unsigned long write_err, void *context)
606 struct dm_thin_new_mapping *m = context;
608 m->err = read_err || write_err ? -EIO : 0;
609 complete_mapping_preparation(m);
612 static void overwrite_endio(struct bio *bio, int err)
614 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
615 struct dm_thin_new_mapping *m = h->overwrite_mapping;
618 complete_mapping_preparation(m);
621 /*----------------------------------------------------------------*/
628 * Prepared mapping jobs.
632 * This sends the bios in the cell back to the deferred_bios list.
634 static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell)
636 struct pool *pool = tc->pool;
639 spin_lock_irqsave(&tc->lock, flags);
640 cell_release(pool, cell, &tc->deferred_bio_list);
641 spin_unlock_irqrestore(&tc->lock, flags);
647 * Same as cell_defer above, except it omits the original holder of the cell.
649 static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
651 struct pool *pool = tc->pool;
654 spin_lock_irqsave(&tc->lock, flags);
655 cell_release_no_holder(pool, cell, &tc->deferred_bio_list);
656 spin_unlock_irqrestore(&tc->lock, flags);
661 static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
664 m->bio->bi_end_io = m->saved_bi_end_io;
665 atomic_inc(&m->bio->bi_remaining);
667 cell_error(m->tc->pool, m->cell);
669 mempool_free(m, m->tc->pool->mapping_pool);
672 static void process_prepared_mapping(struct dm_thin_new_mapping *m)
674 struct thin_c *tc = m->tc;
675 struct pool *pool = tc->pool;
681 bio->bi_end_io = m->saved_bi_end_io;
682 atomic_inc(&bio->bi_remaining);
686 cell_error(pool, m->cell);
691 * Commit the prepared block into the mapping btree.
692 * Any I/O for this block arriving after this point will get
693 * remapped to it directly.
695 r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block);
697 metadata_operation_failed(pool, "dm_thin_insert_block", r);
698 cell_error(pool, m->cell);
703 * Release any bios held while the block was being provisioned.
704 * If we are processing a write bio that completely covers the block,
705 * we already processed it so can ignore it now when processing
706 * the bios in the cell.
709 cell_defer_no_holder(tc, m->cell);
712 cell_defer(tc, m->cell);
716 mempool_free(m, pool->mapping_pool);
719 static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
721 struct thin_c *tc = m->tc;
723 bio_io_error(m->bio);
724 cell_defer_no_holder(tc, m->cell);
725 cell_defer_no_holder(tc, m->cell2);
726 mempool_free(m, tc->pool->mapping_pool);
729 static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m)
731 struct thin_c *tc = m->tc;
733 inc_all_io_entry(tc->pool, m->bio);
734 cell_defer_no_holder(tc, m->cell);
735 cell_defer_no_holder(tc, m->cell2);
738 if (m->definitely_not_shared)
739 remap_and_issue(tc, m->bio, m->data_block);
742 if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used)
743 bio_endio(m->bio, 0);
745 remap_and_issue(tc, m->bio, m->data_block);
748 bio_endio(m->bio, 0);
750 mempool_free(m, tc->pool->mapping_pool);
753 static void process_prepared_discard(struct dm_thin_new_mapping *m)
756 struct thin_c *tc = m->tc;
758 r = dm_thin_remove_block(tc->td, m->virt_block);
760 DMERR_LIMIT("dm_thin_remove_block() failed");
762 process_prepared_discard_passdown(m);
765 static void process_prepared(struct pool *pool, struct list_head *head,
766 process_mapping_fn *fn)
769 struct list_head maps;
770 struct dm_thin_new_mapping *m, *tmp;
772 INIT_LIST_HEAD(&maps);
773 spin_lock_irqsave(&pool->lock, flags);
774 list_splice_init(head, &maps);
775 spin_unlock_irqrestore(&pool->lock, flags);
777 list_for_each_entry_safe(m, tmp, &maps, list)
784 static int io_overlaps_block(struct pool *pool, struct bio *bio)
786 return bio->bi_iter.bi_size ==
787 (pool->sectors_per_block << SECTOR_SHIFT);
790 static int io_overwrites_block(struct pool *pool, struct bio *bio)
792 return (bio_data_dir(bio) == WRITE) &&
793 io_overlaps_block(pool, bio);
796 static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
799 *save = bio->bi_end_io;
803 static int ensure_next_mapping(struct pool *pool)
805 if (pool->next_mapping)
808 pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);
810 return pool->next_mapping ? 0 : -ENOMEM;
813 static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
815 struct dm_thin_new_mapping *m = pool->next_mapping;
817 BUG_ON(!pool->next_mapping);
819 memset(m, 0, sizeof(struct dm_thin_new_mapping));
820 INIT_LIST_HEAD(&m->list);
823 pool->next_mapping = NULL;
828 static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m,
829 sector_t begin, sector_t end)
832 struct dm_io_region to;
834 to.bdev = tc->pool_dev->bdev;
836 to.count = end - begin;
838 r = dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m);
840 DMERR_LIMIT("dm_kcopyd_zero() failed");
841 copy_complete(1, 1, m);
846 * A partial copy also needs to zero the uncopied region.
848 static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
849 struct dm_dev *origin, dm_block_t data_origin,
850 dm_block_t data_dest,
851 struct dm_bio_prison_cell *cell, struct bio *bio,
855 struct pool *pool = tc->pool;
856 struct dm_thin_new_mapping *m = get_next_mapping(pool);
859 m->virt_block = virt_block;
860 m->data_block = data_dest;
864 * quiesce action + copy action + an extra reference held for the
865 * duration of this function (we may need to inc later for a
868 atomic_set(&m->prepare_actions, 3);
870 if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
871 complete_mapping_preparation(m); /* already quiesced */
874 * IO to pool_dev remaps to the pool target's data_dev.
876 * If the whole block of data is being overwritten, we can issue the
877 * bio immediately. Otherwise we use kcopyd to clone the data first.
879 if (io_overwrites_block(pool, bio)) {
880 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
882 h->overwrite_mapping = m;
884 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
885 inc_all_io_entry(pool, bio);
886 remap_and_issue(tc, bio, data_dest);
888 struct dm_io_region from, to;
890 from.bdev = origin->bdev;
891 from.sector = data_origin * pool->sectors_per_block;
894 to.bdev = tc->pool_dev->bdev;
895 to.sector = data_dest * pool->sectors_per_block;
898 r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
899 0, copy_complete, m);
901 DMERR_LIMIT("dm_kcopyd_copy() failed");
902 copy_complete(1, 1, m);
905 * We allow the zero to be issued, to simplify the
906 * error path. Otherwise we'd need to start
907 * worrying about decrementing the prepare_actions
913 * Do we need to zero a tail region?
915 if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) {
916 atomic_inc(&m->prepare_actions);
918 data_dest * pool->sectors_per_block + len,
919 (data_dest + 1) * pool->sectors_per_block);
923 complete_mapping_preparation(m); /* drop our ref */
926 static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
927 dm_block_t data_origin, dm_block_t data_dest,
928 struct dm_bio_prison_cell *cell, struct bio *bio)
930 schedule_copy(tc, virt_block, tc->pool_dev,
931 data_origin, data_dest, cell, bio,
932 tc->pool->sectors_per_block);
935 static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
936 dm_block_t data_block, struct dm_bio_prison_cell *cell,
939 struct pool *pool = tc->pool;
940 struct dm_thin_new_mapping *m = get_next_mapping(pool);
942 atomic_set(&m->prepare_actions, 1); /* no need to quiesce */
944 m->virt_block = virt_block;
945 m->data_block = data_block;
949 * If the whole block of data is being overwritten or we are not
950 * zeroing pre-existing data, we can issue the bio immediately.
951 * Otherwise we use kcopyd to zero the data first.
953 if (!pool->pf.zero_new_blocks)
954 process_prepared_mapping(m);
956 else if (io_overwrites_block(pool, bio)) {
957 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
959 h->overwrite_mapping = m;
961 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
962 inc_all_io_entry(pool, bio);
963 remap_and_issue(tc, bio, data_block);
967 data_block * pool->sectors_per_block,
968 (data_block + 1) * pool->sectors_per_block);
971 static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
972 dm_block_t data_dest,
973 struct dm_bio_prison_cell *cell, struct bio *bio)
975 struct pool *pool = tc->pool;
976 sector_t virt_block_begin = virt_block * pool->sectors_per_block;
977 sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block;
979 if (virt_block_end <= tc->origin_size)
980 schedule_copy(tc, virt_block, tc->origin_dev,
981 virt_block, data_dest, cell, bio,
982 pool->sectors_per_block);
984 else if (virt_block_begin < tc->origin_size)
985 schedule_copy(tc, virt_block, tc->origin_dev,
986 virt_block, data_dest, cell, bio,
987 tc->origin_size - virt_block_begin);
990 schedule_zero(tc, virt_block, data_dest, cell, bio);
994 * A non-zero return indicates read_only or fail_io mode.
995 * Many callers don't care about the return value.
997 static int commit(struct pool *pool)
1001 if (get_pool_mode(pool) >= PM_READ_ONLY)
1004 r = dm_pool_commit_metadata(pool->pmd);
1006 metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
1011 static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks)
1013 unsigned long flags;
1015 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
1016 DMWARN("%s: reached low water mark for data device: sending event.",
1017 dm_device_name(pool->pool_md));
1018 spin_lock_irqsave(&pool->lock, flags);
1019 pool->low_water_triggered = true;
1020 spin_unlock_irqrestore(&pool->lock, flags);
1021 dm_table_event(pool->ti->table);
1025 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode);
1027 static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
1030 dm_block_t free_blocks;
1031 struct pool *pool = tc->pool;
1033 if (WARN_ON(get_pool_mode(pool) != PM_WRITE))
1036 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1038 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1042 check_low_water_mark(pool, free_blocks);
1046 * Try to commit to see if that will free up some
1053 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1055 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1060 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1065 r = dm_pool_alloc_data_block(pool->pmd, result);
1067 metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
1075 * If we have run out of space, queue bios until the device is
1076 * resumed, presumably after having been reloaded with more space.
1078 static void retry_on_resume(struct bio *bio)
1080 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1081 struct thin_c *tc = h->tc;
1082 unsigned long flags;
1084 spin_lock_irqsave(&tc->lock, flags);
1085 bio_list_add(&tc->retry_on_resume_list, bio);
1086 spin_unlock_irqrestore(&tc->lock, flags);
1089 static int should_error_unserviceable_bio(struct pool *pool)
1091 enum pool_mode m = get_pool_mode(pool);
1095 /* Shouldn't get here */
1096 DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
1099 case PM_OUT_OF_DATA_SPACE:
1100 return pool->pf.error_if_no_space ? -ENOSPC : 0;
1106 /* Shouldn't get here */
1107 DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1112 static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
1114 int error = should_error_unserviceable_bio(pool);
1117 bio_endio(bio, error);
1119 retry_on_resume(bio);
1122 static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell)
1125 struct bio_list bios;
1128 error = should_error_unserviceable_bio(pool);
1130 cell_error_with_code(pool, cell, error);
1134 bio_list_init(&bios);
1135 cell_release(pool, cell, &bios);
1137 error = should_error_unserviceable_bio(pool);
1139 while ((bio = bio_list_pop(&bios)))
1140 bio_endio(bio, error);
1142 while ((bio = bio_list_pop(&bios)))
1143 retry_on_resume(bio);
1146 static void process_discard(struct thin_c *tc, struct bio *bio)
1149 unsigned long flags;
1150 struct pool *pool = tc->pool;
1151 struct dm_bio_prison_cell *cell, *cell2;
1152 struct dm_cell_key key, key2;
1153 dm_block_t block = get_bio_block(tc, bio);
1154 struct dm_thin_lookup_result lookup_result;
1155 struct dm_thin_new_mapping *m;
1157 build_virtual_key(tc->td, block, &key);
1158 if (bio_detain(tc->pool, &key, bio, &cell))
1161 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1165 * Check nobody is fiddling with this pool block. This can
1166 * happen if someone's in the process of breaking sharing
1169 build_data_key(tc->td, lookup_result.block, &key2);
1170 if (bio_detain(tc->pool, &key2, bio, &cell2)) {
1171 cell_defer_no_holder(tc, cell);
1175 if (io_overlaps_block(pool, bio)) {
1177 * IO may still be going to the destination block. We must
1178 * quiesce before we can do the removal.
1180 m = get_next_mapping(pool);
1182 m->pass_discard = pool->pf.discard_passdown;
1183 m->definitely_not_shared = !lookup_result.shared;
1184 m->virt_block = block;
1185 m->data_block = lookup_result.block;
1190 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) {
1191 spin_lock_irqsave(&pool->lock, flags);
1192 list_add_tail(&m->list, &pool->prepared_discards);
1193 spin_unlock_irqrestore(&pool->lock, flags);
1197 inc_all_io_entry(pool, bio);
1198 cell_defer_no_holder(tc, cell);
1199 cell_defer_no_holder(tc, cell2);
1202 * The DM core makes sure that the discard doesn't span
1203 * a block boundary. So we submit the discard of a
1204 * partial block appropriately.
1206 if ((!lookup_result.shared) && pool->pf.discard_passdown)
1207 remap_and_issue(tc, bio, lookup_result.block);
1215 * It isn't provisioned, just forget it.
1217 cell_defer_no_holder(tc, cell);
1222 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1224 cell_defer_no_holder(tc, cell);
1230 static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
1231 struct dm_cell_key *key,
1232 struct dm_thin_lookup_result *lookup_result,
1233 struct dm_bio_prison_cell *cell)
1236 dm_block_t data_block;
1237 struct pool *pool = tc->pool;
1239 r = alloc_data_block(tc, &data_block);
1242 schedule_internal_copy(tc, block, lookup_result->block,
1243 data_block, cell, bio);
1247 retry_bios_on_resume(pool, cell);
1251 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1253 cell_error(pool, cell);
1258 static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1260 struct dm_thin_lookup_result *lookup_result)
1262 struct dm_bio_prison_cell *cell;
1263 struct pool *pool = tc->pool;
1264 struct dm_cell_key key;
1267 * If cell is already occupied, then sharing is already in the process
1268 * of being broken so we have nothing further to do here.
1270 build_data_key(tc->td, lookup_result->block, &key);
1271 if (bio_detain(pool, &key, bio, &cell))
1274 if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size)
1275 break_sharing(tc, bio, block, &key, lookup_result, cell);
1277 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1279 h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
1280 inc_all_io_entry(pool, bio);
1281 cell_defer_no_holder(tc, cell);
1283 remap_and_issue(tc, bio, lookup_result->block);
1287 static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
1288 struct dm_bio_prison_cell *cell)
1291 dm_block_t data_block;
1292 struct pool *pool = tc->pool;
1295 * Remap empty bios (flushes) immediately, without provisioning.
1297 if (!bio->bi_iter.bi_size) {
1298 inc_all_io_entry(pool, bio);
1299 cell_defer_no_holder(tc, cell);
1301 remap_and_issue(tc, bio, 0);
1306 * Fill read bios with zeroes and complete them immediately.
1308 if (bio_data_dir(bio) == READ) {
1310 cell_defer_no_holder(tc, cell);
1315 r = alloc_data_block(tc, &data_block);
1319 schedule_external_copy(tc, block, data_block, cell, bio);
1321 schedule_zero(tc, block, data_block, cell, bio);
1325 retry_bios_on_resume(pool, cell);
1329 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1331 cell_error(pool, cell);
1336 static void process_bio(struct thin_c *tc, struct bio *bio)
1339 struct pool *pool = tc->pool;
1340 dm_block_t block = get_bio_block(tc, bio);
1341 struct dm_bio_prison_cell *cell;
1342 struct dm_cell_key key;
1343 struct dm_thin_lookup_result lookup_result;
1346 * If cell is already occupied, then the block is already
1347 * being provisioned so we have nothing further to do here.
1349 build_virtual_key(tc->td, block, &key);
1350 if (bio_detain(pool, &key, bio, &cell))
1353 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1356 if (lookup_result.shared) {
1357 process_shared_bio(tc, bio, block, &lookup_result);
1358 cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */
1360 inc_all_io_entry(pool, bio);
1361 cell_defer_no_holder(tc, cell);
1363 remap_and_issue(tc, bio, lookup_result.block);
1368 if (bio_data_dir(bio) == READ && tc->origin_dev) {
1369 inc_all_io_entry(pool, bio);
1370 cell_defer_no_holder(tc, cell);
1372 if (bio_end_sector(bio) <= tc->origin_size)
1373 remap_to_origin_and_issue(tc, bio);
1375 else if (bio->bi_iter.bi_sector < tc->origin_size) {
1377 bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT;
1378 remap_to_origin_and_issue(tc, bio);
1385 provision_block(tc, bio, block, cell);
1389 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1391 cell_defer_no_holder(tc, cell);
1397 static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
1400 int rw = bio_data_dir(bio);
1401 dm_block_t block = get_bio_block(tc, bio);
1402 struct dm_thin_lookup_result lookup_result;
1404 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1407 if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size)
1408 handle_unserviceable_bio(tc->pool, bio);
1410 inc_all_io_entry(tc->pool, bio);
1411 remap_and_issue(tc, bio, lookup_result.block);
1417 handle_unserviceable_bio(tc->pool, bio);
1421 if (tc->origin_dev) {
1422 inc_all_io_entry(tc->pool, bio);
1423 remap_to_origin_and_issue(tc, bio);
1432 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1439 static void process_bio_success(struct thin_c *tc, struct bio *bio)
1444 static void process_bio_fail(struct thin_c *tc, struct bio *bio)
1450 * FIXME: should we also commit due to size of transaction, measured in
1453 static int need_commit_due_to_time(struct pool *pool)
1455 return jiffies < pool->last_commit_jiffies ||
1456 jiffies > pool->last_commit_jiffies + COMMIT_PERIOD;
1459 #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node)
1460 #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook))
1462 static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio)
1464 struct rb_node **rbp, *parent;
1465 struct dm_thin_endio_hook *pbd;
1466 sector_t bi_sector = bio->bi_iter.bi_sector;
1468 rbp = &tc->sort_bio_list.rb_node;
1472 pbd = thin_pbd(parent);
1474 if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector)
1475 rbp = &(*rbp)->rb_left;
1477 rbp = &(*rbp)->rb_right;
1480 pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1481 rb_link_node(&pbd->rb_node, parent, rbp);
1482 rb_insert_color(&pbd->rb_node, &tc->sort_bio_list);
1485 static void __extract_sorted_bios(struct thin_c *tc)
1487 struct rb_node *node;
1488 struct dm_thin_endio_hook *pbd;
1491 for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) {
1492 pbd = thin_pbd(node);
1493 bio = thin_bio(pbd);
1495 bio_list_add(&tc->deferred_bio_list, bio);
1496 rb_erase(&pbd->rb_node, &tc->sort_bio_list);
1499 WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list));
1502 static void __sort_thin_deferred_bios(struct thin_c *tc)
1505 struct bio_list bios;
1507 bio_list_init(&bios);
1508 bio_list_merge(&bios, &tc->deferred_bio_list);
1509 bio_list_init(&tc->deferred_bio_list);
1511 /* Sort deferred_bio_list using rb-tree */
1512 while ((bio = bio_list_pop(&bios)))
1513 __thin_bio_rb_add(tc, bio);
1516 * Transfer the sorted bios in sort_bio_list back to
1517 * deferred_bio_list to allow lockless submission of
1520 __extract_sorted_bios(tc);
1523 static void process_thin_deferred_bios(struct thin_c *tc)
1525 struct pool *pool = tc->pool;
1526 unsigned long flags;
1528 struct bio_list bios;
1529 struct blk_plug plug;
1531 if (tc->requeue_mode) {
1532 requeue_bio_list(tc, &tc->deferred_bio_list);
1536 bio_list_init(&bios);
1538 spin_lock_irqsave(&tc->lock, flags);
1540 if (bio_list_empty(&tc->deferred_bio_list)) {
1541 spin_unlock_irqrestore(&tc->lock, flags);
1545 __sort_thin_deferred_bios(tc);
1547 bio_list_merge(&bios, &tc->deferred_bio_list);
1548 bio_list_init(&tc->deferred_bio_list);
1550 spin_unlock_irqrestore(&tc->lock, flags);
1552 blk_start_plug(&plug);
1553 while ((bio = bio_list_pop(&bios))) {
1555 * If we've got no free new_mapping structs, and processing
1556 * this bio might require one, we pause until there are some
1557 * prepared mappings to process.
1559 if (ensure_next_mapping(pool)) {
1560 spin_lock_irqsave(&tc->lock, flags);
1561 bio_list_add(&tc->deferred_bio_list, bio);
1562 bio_list_merge(&tc->deferred_bio_list, &bios);
1563 spin_unlock_irqrestore(&tc->lock, flags);
1567 if (bio->bi_rw & REQ_DISCARD)
1568 pool->process_discard(tc, bio);
1570 pool->process_bio(tc, bio);
1572 blk_finish_plug(&plug);
1575 static void thin_get(struct thin_c *tc);
1576 static void thin_put(struct thin_c *tc);
1579 * We can't hold rcu_read_lock() around code that can block. So we
1580 * find a thin with the rcu lock held; bump a refcount; then drop
1583 static struct thin_c *get_first_thin(struct pool *pool)
1585 struct thin_c *tc = NULL;
1588 if (!list_empty(&pool->active_thins)) {
1589 tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list);
1597 static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc)
1599 struct thin_c *old_tc = tc;
1602 list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) {
1614 static void process_deferred_bios(struct pool *pool)
1616 unsigned long flags;
1618 struct bio_list bios;
1621 tc = get_first_thin(pool);
1623 process_thin_deferred_bios(tc);
1624 tc = get_next_thin(pool, tc);
1628 * If there are any deferred flush bios, we must commit
1629 * the metadata before issuing them.
1631 bio_list_init(&bios);
1632 spin_lock_irqsave(&pool->lock, flags);
1633 bio_list_merge(&bios, &pool->deferred_flush_bios);
1634 bio_list_init(&pool->deferred_flush_bios);
1635 spin_unlock_irqrestore(&pool->lock, flags);
1637 if (bio_list_empty(&bios) &&
1638 !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool)))
1642 while ((bio = bio_list_pop(&bios)))
1646 pool->last_commit_jiffies = jiffies;
1648 while ((bio = bio_list_pop(&bios)))
1649 generic_make_request(bio);
1652 static void do_worker(struct work_struct *ws)
1654 struct pool *pool = container_of(ws, struct pool, worker);
1656 process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
1657 process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
1658 process_deferred_bios(pool);
1662 * We want to commit periodically so that not too much
1663 * unwritten data builds up.
1665 static void do_waker(struct work_struct *ws)
1667 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
1669 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
1673 * We're holding onto IO to allow userland time to react. After the
1674 * timeout either the pool will have been resized (and thus back in
1675 * PM_WRITE mode), or we degrade to PM_READ_ONLY and start erroring IO.
1677 static void do_no_space_timeout(struct work_struct *ws)
1679 struct pool *pool = container_of(to_delayed_work(ws), struct pool,
1682 if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space)
1683 set_pool_mode(pool, PM_READ_ONLY);
1686 /*----------------------------------------------------------------*/
1689 struct work_struct worker;
1690 struct completion complete;
1693 static struct pool_work *to_pool_work(struct work_struct *ws)
1695 return container_of(ws, struct pool_work, worker);
1698 static void pool_work_complete(struct pool_work *pw)
1700 complete(&pw->complete);
1703 static void pool_work_wait(struct pool_work *pw, struct pool *pool,
1704 void (*fn)(struct work_struct *))
1706 INIT_WORK_ONSTACK(&pw->worker, fn);
1707 init_completion(&pw->complete);
1708 queue_work(pool->wq, &pw->worker);
1709 wait_for_completion(&pw->complete);
1712 /*----------------------------------------------------------------*/
1714 struct noflush_work {
1715 struct pool_work pw;
1719 static struct noflush_work *to_noflush(struct work_struct *ws)
1721 return container_of(to_pool_work(ws), struct noflush_work, pw);
1724 static void do_noflush_start(struct work_struct *ws)
1726 struct noflush_work *w = to_noflush(ws);
1727 w->tc->requeue_mode = true;
1729 pool_work_complete(&w->pw);
1732 static void do_noflush_stop(struct work_struct *ws)
1734 struct noflush_work *w = to_noflush(ws);
1735 w->tc->requeue_mode = false;
1736 pool_work_complete(&w->pw);
1739 static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *))
1741 struct noflush_work w;
1744 pool_work_wait(&w.pw, tc->pool, fn);
1747 /*----------------------------------------------------------------*/
1749 static enum pool_mode get_pool_mode(struct pool *pool)
1751 return pool->pf.mode;
1754 static void notify_of_pool_mode_change(struct pool *pool, const char *new_mode)
1756 dm_table_event(pool->ti->table);
1757 DMINFO("%s: switching pool to %s mode",
1758 dm_device_name(pool->pool_md), new_mode);
1761 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode)
1763 struct pool_c *pt = pool->ti->private;
1764 bool needs_check = dm_pool_metadata_needs_check(pool->pmd);
1765 enum pool_mode old_mode = get_pool_mode(pool);
1766 unsigned long no_space_timeout = ACCESS_ONCE(no_space_timeout_secs) * HZ;
1769 * Never allow the pool to transition to PM_WRITE mode if user
1770 * intervention is required to verify metadata and data consistency.
1772 if (new_mode == PM_WRITE && needs_check) {
1773 DMERR("%s: unable to switch pool to write mode until repaired.",
1774 dm_device_name(pool->pool_md));
1775 if (old_mode != new_mode)
1776 new_mode = old_mode;
1778 new_mode = PM_READ_ONLY;
1781 * If we were in PM_FAIL mode, rollback of metadata failed. We're
1782 * not going to recover without a thin_repair. So we never let the
1783 * pool move out of the old mode.
1785 if (old_mode == PM_FAIL)
1786 new_mode = old_mode;
1790 if (old_mode != new_mode)
1791 notify_of_pool_mode_change(pool, "failure");
1792 dm_pool_metadata_read_only(pool->pmd);
1793 pool->process_bio = process_bio_fail;
1794 pool->process_discard = process_bio_fail;
1795 pool->process_prepared_mapping = process_prepared_mapping_fail;
1796 pool->process_prepared_discard = process_prepared_discard_fail;
1798 error_retry_list(pool);
1802 if (old_mode != new_mode)
1803 notify_of_pool_mode_change(pool, "read-only");
1804 dm_pool_metadata_read_only(pool->pmd);
1805 pool->process_bio = process_bio_read_only;
1806 pool->process_discard = process_bio_success;
1807 pool->process_prepared_mapping = process_prepared_mapping_fail;
1808 pool->process_prepared_discard = process_prepared_discard_passdown;
1810 error_retry_list(pool);
1813 case PM_OUT_OF_DATA_SPACE:
1815 * Ideally we'd never hit this state; the low water mark
1816 * would trigger userland to extend the pool before we
1817 * completely run out of data space. However, many small
1818 * IOs to unprovisioned space can consume data space at an
1819 * alarming rate. Adjust your low water mark if you're
1820 * frequently seeing this mode.
1822 if (old_mode != new_mode)
1823 notify_of_pool_mode_change(pool, "out-of-data-space");
1824 pool->process_bio = process_bio_read_only;
1825 pool->process_discard = process_discard;
1826 pool->process_prepared_mapping = process_prepared_mapping;
1827 pool->process_prepared_discard = process_prepared_discard_passdown;
1829 if (!pool->pf.error_if_no_space && no_space_timeout)
1830 queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout);
1834 if (old_mode != new_mode)
1835 notify_of_pool_mode_change(pool, "write");
1836 dm_pool_metadata_read_write(pool->pmd);
1837 pool->process_bio = process_bio;
1838 pool->process_discard = process_discard;
1839 pool->process_prepared_mapping = process_prepared_mapping;
1840 pool->process_prepared_discard = process_prepared_discard;
1844 pool->pf.mode = new_mode;
1846 * The pool mode may have changed, sync it so bind_control_target()
1847 * doesn't cause an unexpected mode transition on resume.
1849 pt->adjusted_pf.mode = new_mode;
1852 static void abort_transaction(struct pool *pool)
1854 const char *dev_name = dm_device_name(pool->pool_md);
1856 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name);
1857 if (dm_pool_abort_metadata(pool->pmd)) {
1858 DMERR("%s: failed to abort metadata transaction", dev_name);
1859 set_pool_mode(pool, PM_FAIL);
1862 if (dm_pool_metadata_set_needs_check(pool->pmd)) {
1863 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name);
1864 set_pool_mode(pool, PM_FAIL);
1868 static void metadata_operation_failed(struct pool *pool, const char *op, int r)
1870 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
1871 dm_device_name(pool->pool_md), op, r);
1873 abort_transaction(pool);
1874 set_pool_mode(pool, PM_READ_ONLY);
1877 /*----------------------------------------------------------------*/
1880 * Mapping functions.
1884 * Called only while mapping a thin bio to hand it over to the workqueue.
1886 static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
1888 unsigned long flags;
1889 struct pool *pool = tc->pool;
1891 spin_lock_irqsave(&tc->lock, flags);
1892 bio_list_add(&tc->deferred_bio_list, bio);
1893 spin_unlock_irqrestore(&tc->lock, flags);
1898 static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
1900 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1903 h->shared_read_entry = NULL;
1904 h->all_io_entry = NULL;
1905 h->overwrite_mapping = NULL;
1909 * Non-blocking function called from the thin target's map function.
1911 static int thin_bio_map(struct dm_target *ti, struct bio *bio)
1914 struct thin_c *tc = ti->private;
1915 dm_block_t block = get_bio_block(tc, bio);
1916 struct dm_thin_device *td = tc->td;
1917 struct dm_thin_lookup_result result;
1918 struct dm_bio_prison_cell cell1, cell2;
1919 struct dm_bio_prison_cell *cell_result;
1920 struct dm_cell_key key;
1922 thin_hook_bio(tc, bio);
1924 if (tc->requeue_mode) {
1925 bio_endio(bio, DM_ENDIO_REQUEUE);
1926 return DM_MAPIO_SUBMITTED;
1929 if (get_pool_mode(tc->pool) == PM_FAIL) {
1931 return DM_MAPIO_SUBMITTED;
1934 if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) {
1935 thin_defer_bio(tc, bio);
1936 return DM_MAPIO_SUBMITTED;
1940 * We must hold the virtual cell before doing the lookup, otherwise
1941 * there's a race with discard.
1943 build_virtual_key(tc->td, block, &key);
1944 if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result))
1945 return DM_MAPIO_SUBMITTED;
1947 r = dm_thin_find_block(td, block, 0, &result);
1950 * Note that we defer readahead too.
1954 if (unlikely(result.shared)) {
1956 * We have a race condition here between the
1957 * result.shared value returned by the lookup and
1958 * snapshot creation, which may cause new
1961 * To avoid this always quiesce the origin before
1962 * taking the snap. You want to do this anyway to
1963 * ensure a consistent application view
1966 * More distant ancestors are irrelevant. The
1967 * shared flag will be set in their case.
1969 thin_defer_bio(tc, bio);
1970 cell_defer_no_holder_no_free(tc, &cell1);
1971 return DM_MAPIO_SUBMITTED;
1974 build_data_key(tc->td, result.block, &key);
1975 if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) {
1976 cell_defer_no_holder_no_free(tc, &cell1);
1977 return DM_MAPIO_SUBMITTED;
1980 inc_all_io_entry(tc->pool, bio);
1981 cell_defer_no_holder_no_free(tc, &cell2);
1982 cell_defer_no_holder_no_free(tc, &cell1);
1984 remap(tc, bio, result.block);
1985 return DM_MAPIO_REMAPPED;
1988 if (get_pool_mode(tc->pool) == PM_READ_ONLY) {
1990 * This block isn't provisioned, and we have no way
1993 handle_unserviceable_bio(tc->pool, bio);
1994 cell_defer_no_holder_no_free(tc, &cell1);
1995 return DM_MAPIO_SUBMITTED;
2001 * In future, the failed dm_thin_find_block above could
2002 * provide the hint to load the metadata into cache.
2004 thin_defer_bio(tc, bio);
2005 cell_defer_no_holder_no_free(tc, &cell1);
2006 return DM_MAPIO_SUBMITTED;
2010 * Must always call bio_io_error on failure.
2011 * dm_thin_find_block can fail with -EINVAL if the
2012 * pool is switched to fail-io mode.
2015 cell_defer_no_holder_no_free(tc, &cell1);
2016 return DM_MAPIO_SUBMITTED;
2020 static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
2022 struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
2023 struct request_queue *q;
2025 if (get_pool_mode(pt->pool) == PM_OUT_OF_DATA_SPACE)
2028 q = bdev_get_queue(pt->data_dev->bdev);
2029 return bdi_congested(&q->backing_dev_info, bdi_bits);
2032 static void requeue_bios(struct pool *pool)
2034 unsigned long flags;
2038 list_for_each_entry_rcu(tc, &pool->active_thins, list) {
2039 spin_lock_irqsave(&tc->lock, flags);
2040 bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list);
2041 bio_list_init(&tc->retry_on_resume_list);
2042 spin_unlock_irqrestore(&tc->lock, flags);
2047 /*----------------------------------------------------------------
2048 * Binding of control targets to a pool object
2049 *--------------------------------------------------------------*/
2050 static bool data_dev_supports_discard(struct pool_c *pt)
2052 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
2054 return q && blk_queue_discard(q);
2057 static bool is_factor(sector_t block_size, uint32_t n)
2059 return !sector_div(block_size, n);
2063 * If discard_passdown was enabled verify that the data device
2064 * supports discards. Disable discard_passdown if not.
2066 static void disable_passdown_if_not_supported(struct pool_c *pt)
2068 struct pool *pool = pt->pool;
2069 struct block_device *data_bdev = pt->data_dev->bdev;
2070 struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
2071 sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT;
2072 const char *reason = NULL;
2073 char buf[BDEVNAME_SIZE];
2075 if (!pt->adjusted_pf.discard_passdown)
2078 if (!data_dev_supports_discard(pt))
2079 reason = "discard unsupported";
2081 else if (data_limits->max_discard_sectors < pool->sectors_per_block)
2082 reason = "max discard sectors smaller than a block";
2084 else if (data_limits->discard_granularity > block_size)
2085 reason = "discard granularity larger than a block";
2087 else if (!is_factor(block_size, data_limits->discard_granularity))
2088 reason = "discard granularity not a factor of block size";
2091 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason);
2092 pt->adjusted_pf.discard_passdown = false;
2096 static int bind_control_target(struct pool *pool, struct dm_target *ti)
2098 struct pool_c *pt = ti->private;
2101 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
2103 enum pool_mode old_mode = get_pool_mode(pool);
2104 enum pool_mode new_mode = pt->adjusted_pf.mode;
2107 * Don't change the pool's mode until set_pool_mode() below.
2108 * Otherwise the pool's process_* function pointers may
2109 * not match the desired pool mode.
2111 pt->adjusted_pf.mode = old_mode;
2114 pool->pf = pt->adjusted_pf;
2115 pool->low_water_blocks = pt->low_water_blocks;
2117 set_pool_mode(pool, new_mode);
2122 static void unbind_control_target(struct pool *pool, struct dm_target *ti)
2128 /*----------------------------------------------------------------
2130 *--------------------------------------------------------------*/
2131 /* Initialize pool features. */
2132 static void pool_features_init(struct pool_features *pf)
2134 pf->mode = PM_WRITE;
2135 pf->zero_new_blocks = true;
2136 pf->discard_enabled = true;
2137 pf->discard_passdown = true;
2138 pf->error_if_no_space = false;
2141 static void __pool_destroy(struct pool *pool)
2143 __pool_table_remove(pool);
2145 if (dm_pool_metadata_close(pool->pmd) < 0)
2146 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
2148 dm_bio_prison_destroy(pool->prison);
2149 dm_kcopyd_client_destroy(pool->copier);
2152 destroy_workqueue(pool->wq);
2154 if (pool->next_mapping)
2155 mempool_free(pool->next_mapping, pool->mapping_pool);
2156 mempool_destroy(pool->mapping_pool);
2157 dm_deferred_set_destroy(pool->shared_read_ds);
2158 dm_deferred_set_destroy(pool->all_io_ds);
2162 static struct kmem_cache *_new_mapping_cache;
2164 static struct pool *pool_create(struct mapped_device *pool_md,
2165 struct block_device *metadata_dev,
2166 unsigned long block_size,
2167 int read_only, char **error)
2172 struct dm_pool_metadata *pmd;
2173 bool format_device = read_only ? false : true;
2175 pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
2177 *error = "Error creating metadata object";
2178 return (struct pool *)pmd;
2181 pool = kmalloc(sizeof(*pool), GFP_KERNEL);
2183 *error = "Error allocating memory for pool";
2184 err_p = ERR_PTR(-ENOMEM);
2189 pool->sectors_per_block = block_size;
2190 if (block_size & (block_size - 1))
2191 pool->sectors_per_block_shift = -1;
2193 pool->sectors_per_block_shift = __ffs(block_size);
2194 pool->low_water_blocks = 0;
2195 pool_features_init(&pool->pf);
2196 pool->prison = dm_bio_prison_create(PRISON_CELLS);
2197 if (!pool->prison) {
2198 *error = "Error creating pool's bio prison";
2199 err_p = ERR_PTR(-ENOMEM);
2203 pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2204 if (IS_ERR(pool->copier)) {
2205 r = PTR_ERR(pool->copier);
2206 *error = "Error creating pool's kcopyd client";
2208 goto bad_kcopyd_client;
2212 * Create singlethreaded workqueue that will service all devices
2213 * that use this metadata.
2215 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
2217 *error = "Error creating pool's workqueue";
2218 err_p = ERR_PTR(-ENOMEM);
2222 INIT_WORK(&pool->worker, do_worker);
2223 INIT_DELAYED_WORK(&pool->waker, do_waker);
2224 INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout);
2225 spin_lock_init(&pool->lock);
2226 bio_list_init(&pool->deferred_flush_bios);
2227 INIT_LIST_HEAD(&pool->prepared_mappings);
2228 INIT_LIST_HEAD(&pool->prepared_discards);
2229 INIT_LIST_HEAD(&pool->active_thins);
2230 pool->low_water_triggered = false;
2232 pool->shared_read_ds = dm_deferred_set_create();
2233 if (!pool->shared_read_ds) {
2234 *error = "Error creating pool's shared read deferred set";
2235 err_p = ERR_PTR(-ENOMEM);
2236 goto bad_shared_read_ds;
2239 pool->all_io_ds = dm_deferred_set_create();
2240 if (!pool->all_io_ds) {
2241 *error = "Error creating pool's all io deferred set";
2242 err_p = ERR_PTR(-ENOMEM);
2246 pool->next_mapping = NULL;
2247 pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE,
2248 _new_mapping_cache);
2249 if (!pool->mapping_pool) {
2250 *error = "Error creating pool's mapping mempool";
2251 err_p = ERR_PTR(-ENOMEM);
2252 goto bad_mapping_pool;
2255 pool->ref_count = 1;
2256 pool->last_commit_jiffies = jiffies;
2257 pool->pool_md = pool_md;
2258 pool->md_dev = metadata_dev;
2259 __pool_table_insert(pool);
2264 dm_deferred_set_destroy(pool->all_io_ds);
2266 dm_deferred_set_destroy(pool->shared_read_ds);
2268 destroy_workqueue(pool->wq);
2270 dm_kcopyd_client_destroy(pool->copier);
2272 dm_bio_prison_destroy(pool->prison);
2276 if (dm_pool_metadata_close(pmd))
2277 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
2282 static void __pool_inc(struct pool *pool)
2284 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
2288 static void __pool_dec(struct pool *pool)
2290 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
2291 BUG_ON(!pool->ref_count);
2292 if (!--pool->ref_count)
2293 __pool_destroy(pool);
2296 static struct pool *__pool_find(struct mapped_device *pool_md,
2297 struct block_device *metadata_dev,
2298 unsigned long block_size, int read_only,
2299 char **error, int *created)
2301 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
2304 if (pool->pool_md != pool_md) {
2305 *error = "metadata device already in use by a pool";
2306 return ERR_PTR(-EBUSY);
2311 pool = __pool_table_lookup(pool_md);
2313 if (pool->md_dev != metadata_dev) {
2314 *error = "different pool cannot replace a pool";
2315 return ERR_PTR(-EINVAL);
2320 pool = pool_create(pool_md, metadata_dev, block_size, read_only, error);
2328 /*----------------------------------------------------------------
2329 * Pool target methods
2330 *--------------------------------------------------------------*/
2331 static void pool_dtr(struct dm_target *ti)
2333 struct pool_c *pt = ti->private;
2335 mutex_lock(&dm_thin_pool_table.mutex);
2337 unbind_control_target(pt->pool, ti);
2338 __pool_dec(pt->pool);
2339 dm_put_device(ti, pt->metadata_dev);
2340 dm_put_device(ti, pt->data_dev);
2343 mutex_unlock(&dm_thin_pool_table.mutex);
2346 static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
2347 struct dm_target *ti)
2351 const char *arg_name;
2353 static struct dm_arg _args[] = {
2354 {0, 4, "Invalid number of pool feature arguments"},
2358 * No feature arguments supplied.
2363 r = dm_read_arg_group(_args, as, &argc, &ti->error);
2367 while (argc && !r) {
2368 arg_name = dm_shift_arg(as);
2371 if (!strcasecmp(arg_name, "skip_block_zeroing"))
2372 pf->zero_new_blocks = false;
2374 else if (!strcasecmp(arg_name, "ignore_discard"))
2375 pf->discard_enabled = false;
2377 else if (!strcasecmp(arg_name, "no_discard_passdown"))
2378 pf->discard_passdown = false;
2380 else if (!strcasecmp(arg_name, "read_only"))
2381 pf->mode = PM_READ_ONLY;
2383 else if (!strcasecmp(arg_name, "error_if_no_space"))
2384 pf->error_if_no_space = true;
2387 ti->error = "Unrecognised pool feature requested";
2396 static void metadata_low_callback(void *context)
2398 struct pool *pool = context;
2400 DMWARN("%s: reached low water mark for metadata device: sending event.",
2401 dm_device_name(pool->pool_md));
2403 dm_table_event(pool->ti->table);
2406 static sector_t get_dev_size(struct block_device *bdev)
2408 return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
2411 static void warn_if_metadata_device_too_big(struct block_device *bdev)
2413 sector_t metadata_dev_size = get_dev_size(bdev);
2414 char buffer[BDEVNAME_SIZE];
2416 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING)
2417 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
2418 bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS);
2421 static sector_t get_metadata_dev_size(struct block_device *bdev)
2423 sector_t metadata_dev_size = get_dev_size(bdev);
2425 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS)
2426 metadata_dev_size = THIN_METADATA_MAX_SECTORS;
2428 return metadata_dev_size;
2431 static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
2433 sector_t metadata_dev_size = get_metadata_dev_size(bdev);
2435 sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE);
2437 return metadata_dev_size;
2441 * When a metadata threshold is crossed a dm event is triggered, and
2442 * userland should respond by growing the metadata device. We could let
2443 * userland set the threshold, like we do with the data threshold, but I'm
2444 * not sure they know enough to do this well.
2446 static dm_block_t calc_metadata_threshold(struct pool_c *pt)
2449 * 4M is ample for all ops with the possible exception of thin
2450 * device deletion which is harmless if it fails (just retry the
2451 * delete after you've grown the device).
2453 dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
2454 return min((dm_block_t)1024ULL /* 4M */, quarter);
2458 * thin-pool <metadata dev> <data dev>
2459 * <data block size (sectors)>
2460 * <low water mark (blocks)>
2461 * [<#feature args> [<arg>]*]
2463 * Optional feature arguments are:
2464 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
2465 * ignore_discard: disable discard
2466 * no_discard_passdown: don't pass discards down to the data device
2467 * read_only: Don't allow any changes to be made to the pool metadata.
2468 * error_if_no_space: error IOs, instead of queueing, if no space.
2470 static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
2472 int r, pool_created = 0;
2475 struct pool_features pf;
2476 struct dm_arg_set as;
2477 struct dm_dev *data_dev;
2478 unsigned long block_size;
2479 dm_block_t low_water_blocks;
2480 struct dm_dev *metadata_dev;
2481 fmode_t metadata_mode;
2484 * FIXME Remove validation from scope of lock.
2486 mutex_lock(&dm_thin_pool_table.mutex);
2489 ti->error = "Invalid argument count";
2498 * Set default pool features.
2500 pool_features_init(&pf);
2502 dm_consume_args(&as, 4);
2503 r = parse_pool_features(&as, &pf, ti);
2507 metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
2508 r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
2510 ti->error = "Error opening metadata block device";
2513 warn_if_metadata_device_too_big(metadata_dev->bdev);
2515 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
2517 ti->error = "Error getting data device";
2521 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
2522 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
2523 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
2524 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
2525 ti->error = "Invalid block size";
2530 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
2531 ti->error = "Invalid low water mark";
2536 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
2542 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
2543 block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
2550 * 'pool_created' reflects whether this is the first table load.
2551 * Top level discard support is not allowed to be changed after
2552 * initial load. This would require a pool reload to trigger thin
2555 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
2556 ti->error = "Discard support cannot be disabled once enabled";
2558 goto out_flags_changed;
2563 pt->metadata_dev = metadata_dev;
2564 pt->data_dev = data_dev;
2565 pt->low_water_blocks = low_water_blocks;
2566 pt->adjusted_pf = pt->requested_pf = pf;
2567 ti->num_flush_bios = 1;
2570 * Only need to enable discards if the pool should pass
2571 * them down to the data device. The thin device's discard
2572 * processing will cause mappings to be removed from the btree.
2574 ti->discard_zeroes_data_unsupported = true;
2575 if (pf.discard_enabled && pf.discard_passdown) {
2576 ti->num_discard_bios = 1;
2579 * Setting 'discards_supported' circumvents the normal
2580 * stacking of discard limits (this keeps the pool and
2581 * thin devices' discard limits consistent).
2583 ti->discards_supported = true;
2587 r = dm_pool_register_metadata_threshold(pt->pool->pmd,
2588 calc_metadata_threshold(pt),
2589 metadata_low_callback,
2594 pt->callbacks.congested_fn = pool_is_congested;
2595 dm_table_add_target_callbacks(ti->table, &pt->callbacks);
2597 mutex_unlock(&dm_thin_pool_table.mutex);
2606 dm_put_device(ti, data_dev);
2608 dm_put_device(ti, metadata_dev);
2610 mutex_unlock(&dm_thin_pool_table.mutex);
2615 static int pool_map(struct dm_target *ti, struct bio *bio)
2618 struct pool_c *pt = ti->private;
2619 struct pool *pool = pt->pool;
2620 unsigned long flags;
2623 * As this is a singleton target, ti->begin is always zero.
2625 spin_lock_irqsave(&pool->lock, flags);
2626 bio->bi_bdev = pt->data_dev->bdev;
2627 r = DM_MAPIO_REMAPPED;
2628 spin_unlock_irqrestore(&pool->lock, flags);
2633 static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
2636 struct pool_c *pt = ti->private;
2637 struct pool *pool = pt->pool;
2638 sector_t data_size = ti->len;
2639 dm_block_t sb_data_size;
2641 *need_commit = false;
2643 (void) sector_div(data_size, pool->sectors_per_block);
2645 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
2647 DMERR("%s: failed to retrieve data device size",
2648 dm_device_name(pool->pool_md));
2652 if (data_size < sb_data_size) {
2653 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
2654 dm_device_name(pool->pool_md),
2655 (unsigned long long)data_size, sb_data_size);
2658 } else if (data_size > sb_data_size) {
2659 if (dm_pool_metadata_needs_check(pool->pmd)) {
2660 DMERR("%s: unable to grow the data device until repaired.",
2661 dm_device_name(pool->pool_md));
2666 DMINFO("%s: growing the data device from %llu to %llu blocks",
2667 dm_device_name(pool->pool_md),
2668 sb_data_size, (unsigned long long)data_size);
2669 r = dm_pool_resize_data_dev(pool->pmd, data_size);
2671 metadata_operation_failed(pool, "dm_pool_resize_data_dev", r);
2675 *need_commit = true;
2681 static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
2684 struct pool_c *pt = ti->private;
2685 struct pool *pool = pt->pool;
2686 dm_block_t metadata_dev_size, sb_metadata_dev_size;
2688 *need_commit = false;
2690 metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
2692 r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
2694 DMERR("%s: failed to retrieve metadata device size",
2695 dm_device_name(pool->pool_md));
2699 if (metadata_dev_size < sb_metadata_dev_size) {
2700 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
2701 dm_device_name(pool->pool_md),
2702 metadata_dev_size, sb_metadata_dev_size);
2705 } else if (metadata_dev_size > sb_metadata_dev_size) {
2706 if (dm_pool_metadata_needs_check(pool->pmd)) {
2707 DMERR("%s: unable to grow the metadata device until repaired.",
2708 dm_device_name(pool->pool_md));
2712 warn_if_metadata_device_too_big(pool->md_dev);
2713 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
2714 dm_device_name(pool->pool_md),
2715 sb_metadata_dev_size, metadata_dev_size);
2716 r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
2718 metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r);
2722 *need_commit = true;
2729 * Retrieves the number of blocks of the data device from
2730 * the superblock and compares it to the actual device size,
2731 * thus resizing the data device in case it has grown.
2733 * This both copes with opening preallocated data devices in the ctr
2734 * being followed by a resume
2736 * calling the resume method individually after userspace has
2737 * grown the data device in reaction to a table event.
2739 static int pool_preresume(struct dm_target *ti)
2742 bool need_commit1, need_commit2;
2743 struct pool_c *pt = ti->private;
2744 struct pool *pool = pt->pool;
2747 * Take control of the pool object.
2749 r = bind_control_target(pool, ti);
2753 r = maybe_resize_data_dev(ti, &need_commit1);
2757 r = maybe_resize_metadata_dev(ti, &need_commit2);
2761 if (need_commit1 || need_commit2)
2762 (void) commit(pool);
2767 static void pool_resume(struct dm_target *ti)
2769 struct pool_c *pt = ti->private;
2770 struct pool *pool = pt->pool;
2771 unsigned long flags;
2773 spin_lock_irqsave(&pool->lock, flags);
2774 pool->low_water_triggered = false;
2775 spin_unlock_irqrestore(&pool->lock, flags);
2778 do_waker(&pool->waker.work);
2781 static void pool_postsuspend(struct dm_target *ti)
2783 struct pool_c *pt = ti->private;
2784 struct pool *pool = pt->pool;
2786 cancel_delayed_work(&pool->waker);
2787 cancel_delayed_work(&pool->no_space_timeout);
2788 flush_workqueue(pool->wq);
2789 (void) commit(pool);
2792 static int check_arg_count(unsigned argc, unsigned args_required)
2794 if (argc != args_required) {
2795 DMWARN("Message received with %u arguments instead of %u.",
2796 argc, args_required);
2803 static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
2805 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
2806 *dev_id <= MAX_DEV_ID)
2810 DMWARN("Message received with invalid device id: %s", arg);
2815 static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
2820 r = check_arg_count(argc, 2);
2824 r = read_dev_id(argv[1], &dev_id, 1);
2828 r = dm_pool_create_thin(pool->pmd, dev_id);
2830 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2838 static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2841 dm_thin_id origin_dev_id;
2844 r = check_arg_count(argc, 3);
2848 r = read_dev_id(argv[1], &dev_id, 1);
2852 r = read_dev_id(argv[2], &origin_dev_id, 1);
2856 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
2858 DMWARN("Creation of new snapshot %s of device %s failed.",
2866 static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
2871 r = check_arg_count(argc, 2);
2875 r = read_dev_id(argv[1], &dev_id, 1);
2879 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
2881 DMWARN("Deletion of thin device %s failed.", argv[1]);
2886 static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
2888 dm_thin_id old_id, new_id;
2891 r = check_arg_count(argc, 3);
2895 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
2896 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
2900 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
2901 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
2905 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
2907 DMWARN("Failed to change transaction id from %s to %s.",
2915 static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2919 r = check_arg_count(argc, 1);
2923 (void) commit(pool);
2925 r = dm_pool_reserve_metadata_snap(pool->pmd);
2927 DMWARN("reserve_metadata_snap message failed.");
2932 static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2936 r = check_arg_count(argc, 1);
2940 r = dm_pool_release_metadata_snap(pool->pmd);
2942 DMWARN("release_metadata_snap message failed.");
2948 * Messages supported:
2949 * create_thin <dev_id>
2950 * create_snap <dev_id> <origin_id>
2952 * trim <dev_id> <new_size_in_sectors>
2953 * set_transaction_id <current_trans_id> <new_trans_id>
2954 * reserve_metadata_snap
2955 * release_metadata_snap
2957 static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
2960 struct pool_c *pt = ti->private;
2961 struct pool *pool = pt->pool;
2963 if (!strcasecmp(argv[0], "create_thin"))
2964 r = process_create_thin_mesg(argc, argv, pool);
2966 else if (!strcasecmp(argv[0], "create_snap"))
2967 r = process_create_snap_mesg(argc, argv, pool);
2969 else if (!strcasecmp(argv[0], "delete"))
2970 r = process_delete_mesg(argc, argv, pool);
2972 else if (!strcasecmp(argv[0], "set_transaction_id"))
2973 r = process_set_transaction_id_mesg(argc, argv, pool);
2975 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
2976 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
2978 else if (!strcasecmp(argv[0], "release_metadata_snap"))
2979 r = process_release_metadata_snap_mesg(argc, argv, pool);
2982 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
2985 (void) commit(pool);
2990 static void emit_flags(struct pool_features *pf, char *result,
2991 unsigned sz, unsigned maxlen)
2993 unsigned count = !pf->zero_new_blocks + !pf->discard_enabled +
2994 !pf->discard_passdown + (pf->mode == PM_READ_ONLY) +
2995 pf->error_if_no_space;
2996 DMEMIT("%u ", count);
2998 if (!pf->zero_new_blocks)
2999 DMEMIT("skip_block_zeroing ");
3001 if (!pf->discard_enabled)
3002 DMEMIT("ignore_discard ");
3004 if (!pf->discard_passdown)
3005 DMEMIT("no_discard_passdown ");
3007 if (pf->mode == PM_READ_ONLY)
3008 DMEMIT("read_only ");
3010 if (pf->error_if_no_space)
3011 DMEMIT("error_if_no_space ");
3016 * <transaction id> <used metadata sectors>/<total metadata sectors>
3017 * <used data sectors>/<total data sectors> <held metadata root>
3019 static void pool_status(struct dm_target *ti, status_type_t type,
3020 unsigned status_flags, char *result, unsigned maxlen)
3024 uint64_t transaction_id;
3025 dm_block_t nr_free_blocks_data;
3026 dm_block_t nr_free_blocks_metadata;
3027 dm_block_t nr_blocks_data;
3028 dm_block_t nr_blocks_metadata;
3029 dm_block_t held_root;
3030 char buf[BDEVNAME_SIZE];
3031 char buf2[BDEVNAME_SIZE];
3032 struct pool_c *pt = ti->private;
3033 struct pool *pool = pt->pool;
3036 case STATUSTYPE_INFO:
3037 if (get_pool_mode(pool) == PM_FAIL) {
3042 /* Commit to ensure statistics aren't out-of-date */
3043 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
3044 (void) commit(pool);
3046 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
3048 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
3049 dm_device_name(pool->pool_md), r);
3053 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
3055 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
3056 dm_device_name(pool->pool_md), r);
3060 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
3062 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
3063 dm_device_name(pool->pool_md), r);
3067 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
3069 DMERR("%s: dm_pool_get_free_block_count returned %d",
3070 dm_device_name(pool->pool_md), r);
3074 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
3076 DMERR("%s: dm_pool_get_data_dev_size returned %d",
3077 dm_device_name(pool->pool_md), r);
3081 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
3083 DMERR("%s: dm_pool_get_metadata_snap returned %d",
3084 dm_device_name(pool->pool_md), r);
3088 DMEMIT("%llu %llu/%llu %llu/%llu ",
3089 (unsigned long long)transaction_id,
3090 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
3091 (unsigned long long)nr_blocks_metadata,
3092 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
3093 (unsigned long long)nr_blocks_data);
3096 DMEMIT("%llu ", held_root);
3100 if (pool->pf.mode == PM_OUT_OF_DATA_SPACE)
3101 DMEMIT("out_of_data_space ");
3102 else if (pool->pf.mode == PM_READ_ONLY)
3107 if (!pool->pf.discard_enabled)
3108 DMEMIT("ignore_discard ");
3109 else if (pool->pf.discard_passdown)
3110 DMEMIT("discard_passdown ");
3112 DMEMIT("no_discard_passdown ");
3114 if (pool->pf.error_if_no_space)
3115 DMEMIT("error_if_no_space ");
3117 DMEMIT("queue_if_no_space ");
3121 case STATUSTYPE_TABLE:
3122 DMEMIT("%s %s %lu %llu ",
3123 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
3124 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
3125 (unsigned long)pool->sectors_per_block,
3126 (unsigned long long)pt->low_water_blocks);
3127 emit_flags(&pt->requested_pf, result, sz, maxlen);
3136 static int pool_iterate_devices(struct dm_target *ti,
3137 iterate_devices_callout_fn fn, void *data)
3139 struct pool_c *pt = ti->private;
3141 return fn(ti, pt->data_dev, 0, ti->len, data);
3144 static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
3145 struct bio_vec *biovec, int max_size)
3147 struct pool_c *pt = ti->private;
3148 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
3150 if (!q->merge_bvec_fn)
3153 bvm->bi_bdev = pt->data_dev->bdev;
3155 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
3158 static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits)
3160 struct pool *pool = pt->pool;
3161 struct queue_limits *data_limits;
3163 limits->max_discard_sectors = pool->sectors_per_block;
3166 * discard_granularity is just a hint, and not enforced.
3168 if (pt->adjusted_pf.discard_passdown) {
3169 data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits;
3170 limits->discard_granularity = max(data_limits->discard_granularity,
3171 pool->sectors_per_block << SECTOR_SHIFT);
3173 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
3176 static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
3178 struct pool_c *pt = ti->private;
3179 struct pool *pool = pt->pool;
3180 uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
3183 * If the system-determined stacked limits are compatible with the
3184 * pool's blocksize (io_opt is a factor) do not override them.
3186 if (io_opt_sectors < pool->sectors_per_block ||
3187 do_div(io_opt_sectors, pool->sectors_per_block)) {
3188 blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT);
3189 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
3193 * pt->adjusted_pf is a staging area for the actual features to use.
3194 * They get transferred to the live pool in bind_control_target()
3195 * called from pool_preresume().
3197 if (!pt->adjusted_pf.discard_enabled) {
3199 * Must explicitly disallow stacking discard limits otherwise the
3200 * block layer will stack them if pool's data device has support.
3201 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
3202 * user to see that, so make sure to set all discard limits to 0.
3204 limits->discard_granularity = 0;
3208 disable_passdown_if_not_supported(pt);
3210 set_discard_limits(pt, limits);
3213 static struct target_type pool_target = {
3214 .name = "thin-pool",
3215 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
3216 DM_TARGET_IMMUTABLE,
3217 .version = {1, 13, 0},
3218 .module = THIS_MODULE,
3222 .postsuspend = pool_postsuspend,
3223 .preresume = pool_preresume,
3224 .resume = pool_resume,
3225 .message = pool_message,
3226 .status = pool_status,
3227 .merge = pool_merge,
3228 .iterate_devices = pool_iterate_devices,
3229 .io_hints = pool_io_hints,
3232 /*----------------------------------------------------------------
3233 * Thin target methods
3234 *--------------------------------------------------------------*/
3235 static void thin_get(struct thin_c *tc)
3237 atomic_inc(&tc->refcount);
3240 static void thin_put(struct thin_c *tc)
3242 if (atomic_dec_and_test(&tc->refcount))
3243 complete(&tc->can_destroy);
3246 static void thin_dtr(struct dm_target *ti)
3248 struct thin_c *tc = ti->private;
3249 unsigned long flags;
3252 wait_for_completion(&tc->can_destroy);
3254 spin_lock_irqsave(&tc->pool->lock, flags);
3255 list_del_rcu(&tc->list);
3256 spin_unlock_irqrestore(&tc->pool->lock, flags);
3259 mutex_lock(&dm_thin_pool_table.mutex);
3261 __pool_dec(tc->pool);
3262 dm_pool_close_thin_device(tc->td);
3263 dm_put_device(ti, tc->pool_dev);
3265 dm_put_device(ti, tc->origin_dev);
3268 mutex_unlock(&dm_thin_pool_table.mutex);
3272 * Thin target parameters:
3274 * <pool_dev> <dev_id> [origin_dev]
3276 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
3277 * dev_id: the internal device identifier
3278 * origin_dev: a device external to the pool that should act as the origin
3280 * If the pool device has discards disabled, they get disabled for the thin
3283 static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
3287 struct dm_dev *pool_dev, *origin_dev;
3288 struct mapped_device *pool_md;
3289 unsigned long flags;
3291 mutex_lock(&dm_thin_pool_table.mutex);
3293 if (argc != 2 && argc != 3) {
3294 ti->error = "Invalid argument count";
3299 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
3301 ti->error = "Out of memory";
3305 spin_lock_init(&tc->lock);
3306 bio_list_init(&tc->deferred_bio_list);
3307 bio_list_init(&tc->retry_on_resume_list);
3308 tc->sort_bio_list = RB_ROOT;
3311 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
3313 ti->error = "Error opening origin device";
3314 goto bad_origin_dev;
3316 tc->origin_dev = origin_dev;
3319 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
3321 ti->error = "Error opening pool device";
3324 tc->pool_dev = pool_dev;
3326 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
3327 ti->error = "Invalid device id";
3332 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
3334 ti->error = "Couldn't get pool mapped device";
3339 tc->pool = __pool_table_lookup(pool_md);
3341 ti->error = "Couldn't find pool object";
3343 goto bad_pool_lookup;
3345 __pool_inc(tc->pool);
3347 if (get_pool_mode(tc->pool) == PM_FAIL) {
3348 ti->error = "Couldn't open thin device, Pool is in fail mode";
3353 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
3355 ti->error = "Couldn't open thin internal device";
3359 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
3361 goto bad_target_max_io_len;
3363 ti->num_flush_bios = 1;
3364 ti->flush_supported = true;
3365 ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook);
3367 /* In case the pool supports discards, pass them on. */
3368 ti->discard_zeroes_data_unsupported = true;
3369 if (tc->pool->pf.discard_enabled) {
3370 ti->discards_supported = true;
3371 ti->num_discard_bios = 1;
3372 /* Discard bios must be split on a block boundary */
3373 ti->split_discard_bios = true;
3378 mutex_unlock(&dm_thin_pool_table.mutex);
3380 atomic_set(&tc->refcount, 1);
3381 init_completion(&tc->can_destroy);
3383 spin_lock_irqsave(&tc->pool->lock, flags);
3384 list_add_tail_rcu(&tc->list, &tc->pool->active_thins);
3385 spin_unlock_irqrestore(&tc->pool->lock, flags);
3387 * This synchronize_rcu() call is needed here otherwise we risk a
3388 * wake_worker() call finding no bios to process (because the newly
3389 * added tc isn't yet visible). So this reduces latency since we
3390 * aren't then dependent on the periodic commit to wake_worker().
3396 bad_target_max_io_len:
3397 dm_pool_close_thin_device(tc->td);
3399 __pool_dec(tc->pool);
3403 dm_put_device(ti, tc->pool_dev);
3406 dm_put_device(ti, tc->origin_dev);
3410 mutex_unlock(&dm_thin_pool_table.mutex);
3415 static int thin_map(struct dm_target *ti, struct bio *bio)
3417 bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
3419 return thin_bio_map(ti, bio);
3422 static int thin_endio(struct dm_target *ti, struct bio *bio, int err)
3424 unsigned long flags;
3425 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
3426 struct list_head work;
3427 struct dm_thin_new_mapping *m, *tmp;
3428 struct pool *pool = h->tc->pool;
3430 if (h->shared_read_entry) {
3431 INIT_LIST_HEAD(&work);
3432 dm_deferred_entry_dec(h->shared_read_entry, &work);
3434 spin_lock_irqsave(&pool->lock, flags);
3435 list_for_each_entry_safe(m, tmp, &work, list) {
3437 __complete_mapping_preparation(m);
3439 spin_unlock_irqrestore(&pool->lock, flags);
3442 if (h->all_io_entry) {
3443 INIT_LIST_HEAD(&work);
3444 dm_deferred_entry_dec(h->all_io_entry, &work);
3445 if (!list_empty(&work)) {
3446 spin_lock_irqsave(&pool->lock, flags);
3447 list_for_each_entry_safe(m, tmp, &work, list)
3448 list_add_tail(&m->list, &pool->prepared_discards);
3449 spin_unlock_irqrestore(&pool->lock, flags);
3457 static void thin_presuspend(struct dm_target *ti)
3459 struct thin_c *tc = ti->private;
3461 if (dm_noflush_suspending(ti))
3462 noflush_work(tc, do_noflush_start);
3465 static void thin_postsuspend(struct dm_target *ti)
3467 struct thin_c *tc = ti->private;
3470 * The dm_noflush_suspending flag has been cleared by now, so
3471 * unfortunately we must always run this.
3473 noflush_work(tc, do_noflush_stop);
3476 static int thin_preresume(struct dm_target *ti)
3478 struct thin_c *tc = ti->private;
3481 tc->origin_size = get_dev_size(tc->origin_dev->bdev);
3487 * <nr mapped sectors> <highest mapped sector>
3489 static void thin_status(struct dm_target *ti, status_type_t type,
3490 unsigned status_flags, char *result, unsigned maxlen)
3494 dm_block_t mapped, highest;
3495 char buf[BDEVNAME_SIZE];
3496 struct thin_c *tc = ti->private;
3498 if (get_pool_mode(tc->pool) == PM_FAIL) {
3507 case STATUSTYPE_INFO:
3508 r = dm_thin_get_mapped_count(tc->td, &mapped);
3510 DMERR("dm_thin_get_mapped_count returned %d", r);
3514 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
3516 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
3520 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
3522 DMEMIT("%llu", ((highest + 1) *
3523 tc->pool->sectors_per_block) - 1);
3528 case STATUSTYPE_TABLE:
3530 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
3531 (unsigned long) tc->dev_id);
3533 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
3544 static int thin_iterate_devices(struct dm_target *ti,
3545 iterate_devices_callout_fn fn, void *data)
3548 struct thin_c *tc = ti->private;
3549 struct pool *pool = tc->pool;
3552 * We can't call dm_pool_get_data_dev_size() since that blocks. So
3553 * we follow a more convoluted path through to the pool's target.
3556 return 0; /* nothing is bound */
3558 blocks = pool->ti->len;
3559 (void) sector_div(blocks, pool->sectors_per_block);
3561 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
3566 static struct target_type thin_target = {
3568 .version = {1, 13, 0},
3569 .module = THIS_MODULE,
3573 .end_io = thin_endio,
3574 .preresume = thin_preresume,
3575 .presuspend = thin_presuspend,
3576 .postsuspend = thin_postsuspend,
3577 .status = thin_status,
3578 .iterate_devices = thin_iterate_devices,
3581 /*----------------------------------------------------------------*/
3583 static int __init dm_thin_init(void)
3589 r = dm_register_target(&thin_target);
3593 r = dm_register_target(&pool_target);
3595 goto bad_pool_target;
3599 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
3600 if (!_new_mapping_cache)
3601 goto bad_new_mapping_cache;
3605 bad_new_mapping_cache:
3606 dm_unregister_target(&pool_target);
3608 dm_unregister_target(&thin_target);
3613 static void dm_thin_exit(void)
3615 dm_unregister_target(&thin_target);
3616 dm_unregister_target(&pool_target);
3618 kmem_cache_destroy(_new_mapping_cache);
3621 module_init(dm_thin_init);
3622 module_exit(dm_thin_exit);
3624 module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR);
3625 MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds");
3627 MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
3628 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3629 MODULE_LICENSE("GPL");