1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2011-2012 Red Hat UK.
5 * This file is released under the GPL.
8 #include "dm-thin-metadata.h"
9 #include "dm-bio-prison-v1.h"
12 #include <linux/device-mapper.h>
13 #include <linux/dm-io.h>
14 #include <linux/dm-kcopyd.h>
15 #include <linux/jiffies.h>
16 #include <linux/log2.h>
17 #include <linux/list.h>
18 #include <linux/rculist.h>
19 #include <linux/init.h>
20 #include <linux/module.h>
21 #include <linux/slab.h>
22 #include <linux/vmalloc.h>
23 #include <linux/sort.h>
24 #include <linux/rbtree.h>
26 #define DM_MSG_PREFIX "thin"
31 #define ENDIO_HOOK_POOL_SIZE 1024
32 #define MAPPING_POOL_SIZE 1024
33 #define COMMIT_PERIOD HZ
34 #define NO_SPACE_TIMEOUT_SECS 60
36 static unsigned int no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS;
38 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
39 "A percentage of time allocated for copy on write");
42 * The block size of the device holding pool data must be
43 * between 64KB and 1GB.
45 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
46 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
49 * Device id is restricted to 24 bits.
51 #define MAX_DEV_ID ((1 << 24) - 1)
54 * How do we handle breaking sharing of data blocks?
55 * =================================================
57 * We use a standard copy-on-write btree to store the mappings for the
58 * devices (note I'm talking about copy-on-write of the metadata here, not
59 * the data). When you take an internal snapshot you clone the root node
60 * of the origin btree. After this there is no concept of an origin or a
61 * snapshot. They are just two device trees that happen to point to the
64 * When we get a write in we decide if it's to a shared data block using
65 * some timestamp magic. If it is, we have to break sharing.
67 * Let's say we write to a shared block in what was the origin. The
70 * i) plug io further to this physical block. (see bio_prison code).
72 * ii) quiesce any read io to that shared data block. Obviously
73 * including all devices that share this block. (see dm_deferred_set code)
75 * iii) copy the data block to a newly allocate block. This step can be
76 * missed out if the io covers the block. (schedule_copy).
78 * iv) insert the new mapping into the origin's btree
79 * (process_prepared_mapping). This act of inserting breaks some
80 * sharing of btree nodes between the two devices. Breaking sharing only
81 * effects the btree of that specific device. Btrees for the other
82 * devices that share the block never change. The btree for the origin
83 * device as it was after the last commit is untouched, ie. we're using
84 * persistent data structures in the functional programming sense.
86 * v) unplug io to this physical block, including the io that triggered
87 * the breaking of sharing.
89 * Steps (ii) and (iii) occur in parallel.
91 * The metadata _doesn't_ need to be committed before the io continues. We
92 * get away with this because the io is always written to a _new_ block.
93 * If there's a crash, then:
95 * - The origin mapping will point to the old origin block (the shared
96 * one). This will contain the data as it was before the io that triggered
97 * the breaking of sharing came in.
99 * - The snap mapping still points to the old block. As it would after
102 * The downside of this scheme is the timestamp magic isn't perfect, and
103 * will continue to think that data block in the snapshot device is shared
104 * even after the write to the origin has broken sharing. I suspect data
105 * blocks will typically be shared by many different devices, so we're
106 * breaking sharing n + 1 times, rather than n, where n is the number of
107 * devices that reference this data block. At the moment I think the
108 * benefits far, far outweigh the disadvantages.
111 /*----------------------------------------------------------------*/
121 static bool build_key(struct dm_thin_device *td, enum lock_space ls,
122 dm_block_t b, dm_block_t e, struct dm_cell_key *key)
124 key->virtual = (ls == VIRTUAL);
125 key->dev = dm_thin_dev_id(td);
126 key->block_begin = b;
129 return dm_cell_key_has_valid_range(key);
132 static void build_data_key(struct dm_thin_device *td, dm_block_t b,
133 struct dm_cell_key *key)
135 (void) build_key(td, PHYSICAL, b, b + 1llu, key);
138 static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
139 struct dm_cell_key *key)
141 (void) build_key(td, VIRTUAL, b, b + 1llu, key);
144 /*----------------------------------------------------------------*/
146 #define THROTTLE_THRESHOLD (1 * HZ)
149 struct rw_semaphore lock;
150 unsigned long threshold;
151 bool throttle_applied;
154 static void throttle_init(struct throttle *t)
156 init_rwsem(&t->lock);
157 t->throttle_applied = false;
160 static void throttle_work_start(struct throttle *t)
162 t->threshold = jiffies + THROTTLE_THRESHOLD;
165 static void throttle_work_update(struct throttle *t)
167 if (!t->throttle_applied && time_is_before_jiffies(t->threshold)) {
168 down_write(&t->lock);
169 t->throttle_applied = true;
173 static void throttle_work_complete(struct throttle *t)
175 if (t->throttle_applied) {
176 t->throttle_applied = false;
181 static void throttle_lock(struct throttle *t)
186 static void throttle_unlock(struct throttle *t)
191 /*----------------------------------------------------------------*/
194 * A pool device ties together a metadata device and a data device. It
195 * also provides the interface for creating and destroying internal
198 struct dm_thin_new_mapping;
201 * The pool runs in various modes. Ordered in degraded order for comparisons.
204 PM_WRITE, /* metadata may be changed */
205 PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */
208 * Like READ_ONLY, except may switch back to WRITE on metadata resize. Reported as READ_ONLY.
210 PM_OUT_OF_METADATA_SPACE,
211 PM_READ_ONLY, /* metadata may not be changed */
213 PM_FAIL, /* all I/O fails */
216 struct pool_features {
219 bool zero_new_blocks:1;
220 bool discard_enabled:1;
221 bool discard_passdown:1;
222 bool error_if_no_space:1;
226 typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
227 typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell);
228 typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);
230 #define CELL_SORT_ARRAY_SIZE 8192
233 struct list_head list;
234 struct dm_target *ti; /* Only set if a pool target is bound */
236 struct mapped_device *pool_md;
237 struct block_device *data_dev;
238 struct block_device *md_dev;
239 struct dm_pool_metadata *pmd;
241 dm_block_t low_water_blocks;
242 uint32_t sectors_per_block;
243 int sectors_per_block_shift;
245 struct pool_features pf;
246 bool low_water_triggered:1; /* A dm event has been sent */
248 bool out_of_data_space:1;
250 struct dm_bio_prison *prison;
251 struct dm_kcopyd_client *copier;
253 struct work_struct worker;
254 struct workqueue_struct *wq;
255 struct throttle throttle;
256 struct delayed_work waker;
257 struct delayed_work no_space_timeout;
259 unsigned long last_commit_jiffies;
260 unsigned int ref_count;
263 struct bio_list deferred_flush_bios;
264 struct bio_list deferred_flush_completions;
265 struct list_head prepared_mappings;
266 struct list_head prepared_discards;
267 struct list_head prepared_discards_pt2;
268 struct list_head active_thins;
270 struct dm_deferred_set *shared_read_ds;
271 struct dm_deferred_set *all_io_ds;
273 struct dm_thin_new_mapping *next_mapping;
275 process_bio_fn process_bio;
276 process_bio_fn process_discard;
278 process_cell_fn process_cell;
279 process_cell_fn process_discard_cell;
281 process_mapping_fn process_prepared_mapping;
282 process_mapping_fn process_prepared_discard;
283 process_mapping_fn process_prepared_discard_pt2;
285 struct dm_bio_prison_cell **cell_sort_array;
287 mempool_t mapping_pool;
290 static void metadata_operation_failed(struct pool *pool, const char *op, int r);
292 static enum pool_mode get_pool_mode(struct pool *pool)
294 return pool->pf.mode;
297 static void notify_of_pool_mode_change(struct pool *pool)
299 static const char *descs[] = {
306 const char *extra_desc = NULL;
307 enum pool_mode mode = get_pool_mode(pool);
309 if (mode == PM_OUT_OF_DATA_SPACE) {
310 if (!pool->pf.error_if_no_space)
311 extra_desc = " (queue IO)";
313 extra_desc = " (error IO)";
316 dm_table_event(pool->ti->table);
317 DMINFO("%s: switching pool to %s%s mode",
318 dm_device_name(pool->pool_md),
319 descs[(int)mode], extra_desc ? : "");
323 * Target context for a pool.
326 struct dm_target *ti;
328 struct dm_dev *data_dev;
329 struct dm_dev *metadata_dev;
331 dm_block_t low_water_blocks;
332 struct pool_features requested_pf; /* Features requested during table load */
333 struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */
337 * Target context for a thin.
340 struct list_head list;
341 struct dm_dev *pool_dev;
342 struct dm_dev *origin_dev;
343 sector_t origin_size;
347 struct dm_thin_device *td;
348 struct mapped_device *thin_md;
352 struct list_head deferred_cells;
353 struct bio_list deferred_bio_list;
354 struct bio_list retry_on_resume_list;
355 struct rb_root sort_bio_list; /* sorted list of deferred bios */
358 * Ensures the thin is not destroyed until the worker has finished
359 * iterating the active_thins list.
362 struct completion can_destroy;
365 /*----------------------------------------------------------------*/
367 static bool block_size_is_power_of_two(struct pool *pool)
369 return pool->sectors_per_block_shift >= 0;
372 static sector_t block_to_sectors(struct pool *pool, dm_block_t b)
374 return block_size_is_power_of_two(pool) ?
375 (b << pool->sectors_per_block_shift) :
376 (b * pool->sectors_per_block);
379 /*----------------------------------------------------------------*/
383 struct blk_plug plug;
384 struct bio *parent_bio;
388 static void begin_discard(struct discard_op *op, struct thin_c *tc, struct bio *parent)
393 blk_start_plug(&op->plug);
394 op->parent_bio = parent;
398 static int issue_discard(struct discard_op *op, dm_block_t data_b, dm_block_t data_e)
400 struct thin_c *tc = op->tc;
401 sector_t s = block_to_sectors(tc->pool, data_b);
402 sector_t len = block_to_sectors(tc->pool, data_e - data_b);
404 return __blkdev_issue_discard(tc->pool_dev->bdev, s, len, GFP_NOWAIT,
408 static void end_discard(struct discard_op *op, int r)
412 * Even if one of the calls to issue_discard failed, we
413 * need to wait for the chain to complete.
415 bio_chain(op->bio, op->parent_bio);
416 op->bio->bi_opf = REQ_OP_DISCARD;
420 blk_finish_plug(&op->plug);
423 * Even if r is set, there could be sub discards in flight that we
426 if (r && !op->parent_bio->bi_status)
427 op->parent_bio->bi_status = errno_to_blk_status(r);
428 bio_endio(op->parent_bio);
431 /*----------------------------------------------------------------*/
434 * wake_worker() is used when new work is queued and when pool_resume is
435 * ready to continue deferred IO processing.
437 static void wake_worker(struct pool *pool)
439 queue_work(pool->wq, &pool->worker);
442 /*----------------------------------------------------------------*/
444 static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio,
445 struct dm_bio_prison_cell **cell_result)
448 struct dm_bio_prison_cell *cell_prealloc;
451 * Allocate a cell from the prison's mempool.
452 * This might block but it can't fail.
454 cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO);
456 r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result);
459 * We reused an old cell; we can get rid of
462 dm_bio_prison_free_cell(pool->prison, cell_prealloc);
467 static void cell_release(struct pool *pool,
468 struct dm_bio_prison_cell *cell,
469 struct bio_list *bios)
471 dm_cell_release(pool->prison, cell, bios);
472 dm_bio_prison_free_cell(pool->prison, cell);
475 static void cell_visit_release(struct pool *pool,
476 void (*fn)(void *, struct dm_bio_prison_cell *),
478 struct dm_bio_prison_cell *cell)
480 dm_cell_visit_release(pool->prison, fn, context, cell);
481 dm_bio_prison_free_cell(pool->prison, cell);
484 static void cell_release_no_holder(struct pool *pool,
485 struct dm_bio_prison_cell *cell,
486 struct bio_list *bios)
488 dm_cell_release_no_holder(pool->prison, cell, bios);
489 dm_bio_prison_free_cell(pool->prison, cell);
492 static void cell_error_with_code(struct pool *pool,
493 struct dm_bio_prison_cell *cell, blk_status_t error_code)
495 dm_cell_error(pool->prison, cell, error_code);
496 dm_bio_prison_free_cell(pool->prison, cell);
499 static blk_status_t get_pool_io_error_code(struct pool *pool)
501 return pool->out_of_data_space ? BLK_STS_NOSPC : BLK_STS_IOERR;
504 static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell)
506 cell_error_with_code(pool, cell, get_pool_io_error_code(pool));
509 static void cell_success(struct pool *pool, struct dm_bio_prison_cell *cell)
511 cell_error_with_code(pool, cell, 0);
514 static void cell_requeue(struct pool *pool, struct dm_bio_prison_cell *cell)
516 cell_error_with_code(pool, cell, BLK_STS_DM_REQUEUE);
519 /*----------------------------------------------------------------*/
522 * A global list of pools that uses a struct mapped_device as a key.
524 static struct dm_thin_pool_table {
526 struct list_head pools;
527 } dm_thin_pool_table;
529 static void pool_table_init(void)
531 mutex_init(&dm_thin_pool_table.mutex);
532 INIT_LIST_HEAD(&dm_thin_pool_table.pools);
535 static void pool_table_exit(void)
537 mutex_destroy(&dm_thin_pool_table.mutex);
540 static void __pool_table_insert(struct pool *pool)
542 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
543 list_add(&pool->list, &dm_thin_pool_table.pools);
546 static void __pool_table_remove(struct pool *pool)
548 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
549 list_del(&pool->list);
552 static struct pool *__pool_table_lookup(struct mapped_device *md)
554 struct pool *pool = NULL, *tmp;
556 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
558 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
559 if (tmp->pool_md == md) {
568 static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
570 struct pool *pool = NULL, *tmp;
572 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
574 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
575 if (tmp->md_dev == md_dev) {
584 /*----------------------------------------------------------------*/
586 struct dm_thin_endio_hook {
588 struct dm_deferred_entry *shared_read_entry;
589 struct dm_deferred_entry *all_io_entry;
590 struct dm_thin_new_mapping *overwrite_mapping;
591 struct rb_node rb_node;
592 struct dm_bio_prison_cell *cell;
595 static void __merge_bio_list(struct bio_list *bios, struct bio_list *master)
597 bio_list_merge(bios, master);
598 bio_list_init(master);
601 static void error_bio_list(struct bio_list *bios, blk_status_t error)
605 while ((bio = bio_list_pop(bios))) {
606 bio->bi_status = error;
611 static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master,
614 struct bio_list bios;
616 bio_list_init(&bios);
618 spin_lock_irq(&tc->lock);
619 __merge_bio_list(&bios, master);
620 spin_unlock_irq(&tc->lock);
622 error_bio_list(&bios, error);
625 static void requeue_deferred_cells(struct thin_c *tc)
627 struct pool *pool = tc->pool;
628 struct list_head cells;
629 struct dm_bio_prison_cell *cell, *tmp;
631 INIT_LIST_HEAD(&cells);
633 spin_lock_irq(&tc->lock);
634 list_splice_init(&tc->deferred_cells, &cells);
635 spin_unlock_irq(&tc->lock);
637 list_for_each_entry_safe(cell, tmp, &cells, user_list)
638 cell_requeue(pool, cell);
641 static void requeue_io(struct thin_c *tc)
643 struct bio_list bios;
645 bio_list_init(&bios);
647 spin_lock_irq(&tc->lock);
648 __merge_bio_list(&bios, &tc->deferred_bio_list);
649 __merge_bio_list(&bios, &tc->retry_on_resume_list);
650 spin_unlock_irq(&tc->lock);
652 error_bio_list(&bios, BLK_STS_DM_REQUEUE);
653 requeue_deferred_cells(tc);
656 static void error_retry_list_with_code(struct pool *pool, blk_status_t error)
661 list_for_each_entry_rcu(tc, &pool->active_thins, list)
662 error_thin_bio_list(tc, &tc->retry_on_resume_list, error);
666 static void error_retry_list(struct pool *pool)
668 error_retry_list_with_code(pool, get_pool_io_error_code(pool));
672 * This section of code contains the logic for processing a thin device's IO.
673 * Much of the code depends on pool object resources (lists, workqueues, etc)
674 * but most is exclusively called from the thin target rather than the thin-pool
678 static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
680 struct pool *pool = tc->pool;
681 sector_t block_nr = bio->bi_iter.bi_sector;
683 if (block_size_is_power_of_two(pool))
684 block_nr >>= pool->sectors_per_block_shift;
686 (void) sector_div(block_nr, pool->sectors_per_block);
692 * Returns the _complete_ blocks that this bio covers.
694 static void get_bio_block_range(struct thin_c *tc, struct bio *bio,
695 dm_block_t *begin, dm_block_t *end)
697 struct pool *pool = tc->pool;
698 sector_t b = bio->bi_iter.bi_sector;
699 sector_t e = b + (bio->bi_iter.bi_size >> SECTOR_SHIFT);
701 b += pool->sectors_per_block - 1ull; /* so we round up */
703 if (block_size_is_power_of_two(pool)) {
704 b >>= pool->sectors_per_block_shift;
705 e >>= pool->sectors_per_block_shift;
707 (void) sector_div(b, pool->sectors_per_block);
708 (void) sector_div(e, pool->sectors_per_block);
712 /* Can happen if the bio is within a single block. */
719 static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
721 struct pool *pool = tc->pool;
722 sector_t bi_sector = bio->bi_iter.bi_sector;
724 bio_set_dev(bio, tc->pool_dev->bdev);
725 if (block_size_is_power_of_two(pool))
726 bio->bi_iter.bi_sector =
727 (block << pool->sectors_per_block_shift) |
728 (bi_sector & (pool->sectors_per_block - 1));
730 bio->bi_iter.bi_sector = (block * pool->sectors_per_block) +
731 sector_div(bi_sector, pool->sectors_per_block);
734 static void remap_to_origin(struct thin_c *tc, struct bio *bio)
736 bio_set_dev(bio, tc->origin_dev->bdev);
739 static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
741 return op_is_flush(bio->bi_opf) &&
742 dm_thin_changed_this_transaction(tc->td);
745 static void inc_all_io_entry(struct pool *pool, struct bio *bio)
747 struct dm_thin_endio_hook *h;
749 if (bio_op(bio) == REQ_OP_DISCARD)
752 h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
753 h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
756 static void issue(struct thin_c *tc, struct bio *bio)
758 struct pool *pool = tc->pool;
760 if (!bio_triggers_commit(tc, bio)) {
761 dm_submit_bio_remap(bio, NULL);
766 * Complete bio with an error if earlier I/O caused changes to
767 * the metadata that can't be committed e.g, due to I/O errors
768 * on the metadata device.
770 if (dm_thin_aborted_changes(tc->td)) {
776 * Batch together any bios that trigger commits and then issue a
777 * single commit for them in process_deferred_bios().
779 spin_lock_irq(&pool->lock);
780 bio_list_add(&pool->deferred_flush_bios, bio);
781 spin_unlock_irq(&pool->lock);
784 static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
786 remap_to_origin(tc, bio);
790 static void remap_and_issue(struct thin_c *tc, struct bio *bio,
793 remap(tc, bio, block);
797 /*----------------------------------------------------------------*/
800 * Bio endio functions.
802 struct dm_thin_new_mapping {
803 struct list_head list;
809 * Track quiescing, copying and zeroing preparation actions. When this
810 * counter hits zero the block is prepared and can be inserted into the
813 atomic_t prepare_actions;
817 dm_block_t virt_begin, virt_end;
818 dm_block_t data_block;
819 struct dm_bio_prison_cell *cell;
822 * If the bio covers the whole area of a block then we can avoid
823 * zeroing or copying. Instead this bio is hooked. The bio will
824 * still be in the cell, so care has to be taken to avoid issuing
828 bio_end_io_t *saved_bi_end_io;
831 static void __complete_mapping_preparation(struct dm_thin_new_mapping *m)
833 struct pool *pool = m->tc->pool;
835 if (atomic_dec_and_test(&m->prepare_actions)) {
836 list_add_tail(&m->list, &pool->prepared_mappings);
841 static void complete_mapping_preparation(struct dm_thin_new_mapping *m)
844 struct pool *pool = m->tc->pool;
846 spin_lock_irqsave(&pool->lock, flags);
847 __complete_mapping_preparation(m);
848 spin_unlock_irqrestore(&pool->lock, flags);
851 static void copy_complete(int read_err, unsigned long write_err, void *context)
853 struct dm_thin_new_mapping *m = context;
855 m->status = read_err || write_err ? BLK_STS_IOERR : 0;
856 complete_mapping_preparation(m);
859 static void overwrite_endio(struct bio *bio)
861 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
862 struct dm_thin_new_mapping *m = h->overwrite_mapping;
864 bio->bi_end_io = m->saved_bi_end_io;
866 m->status = bio->bi_status;
867 complete_mapping_preparation(m);
870 /*----------------------------------------------------------------*/
877 * Prepared mapping jobs.
881 * This sends the bios in the cell, except the original holder, back
882 * to the deferred_bios list.
884 static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
886 struct pool *pool = tc->pool;
888 struct bio_list bios;
890 bio_list_init(&bios);
891 cell_release_no_holder(pool, cell, &bios);
893 if (!bio_list_empty(&bios)) {
894 spin_lock_irqsave(&tc->lock, flags);
895 bio_list_merge(&tc->deferred_bio_list, &bios);
896 spin_unlock_irqrestore(&tc->lock, flags);
901 static void thin_defer_bio(struct thin_c *tc, struct bio *bio);
905 struct bio_list defer_bios;
906 struct bio_list issue_bios;
909 static void __inc_remap_and_issue_cell(void *context,
910 struct dm_bio_prison_cell *cell)
912 struct remap_info *info = context;
915 while ((bio = bio_list_pop(&cell->bios))) {
916 if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD)
917 bio_list_add(&info->defer_bios, bio);
919 inc_all_io_entry(info->tc->pool, bio);
922 * We can't issue the bios with the bio prison lock
923 * held, so we add them to a list to issue on
924 * return from this function.
926 bio_list_add(&info->issue_bios, bio);
931 static void inc_remap_and_issue_cell(struct thin_c *tc,
932 struct dm_bio_prison_cell *cell,
936 struct remap_info info;
939 bio_list_init(&info.defer_bios);
940 bio_list_init(&info.issue_bios);
943 * We have to be careful to inc any bios we're about to issue
944 * before the cell is released, and avoid a race with new bios
945 * being added to the cell.
947 cell_visit_release(tc->pool, __inc_remap_and_issue_cell,
950 while ((bio = bio_list_pop(&info.defer_bios)))
951 thin_defer_bio(tc, bio);
953 while ((bio = bio_list_pop(&info.issue_bios)))
954 remap_and_issue(info.tc, bio, block);
957 static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
959 cell_error(m->tc->pool, m->cell);
961 mempool_free(m, &m->tc->pool->mapping_pool);
964 static void complete_overwrite_bio(struct thin_c *tc, struct bio *bio)
966 struct pool *pool = tc->pool;
969 * If the bio has the REQ_FUA flag set we must commit the metadata
970 * before signaling its completion.
972 if (!bio_triggers_commit(tc, bio)) {
978 * Complete bio with an error if earlier I/O caused changes to the
979 * metadata that can't be committed, e.g, due to I/O errors on the
982 if (dm_thin_aborted_changes(tc->td)) {
988 * Batch together any bios that trigger commits and then issue a
989 * single commit for them in process_deferred_bios().
991 spin_lock_irq(&pool->lock);
992 bio_list_add(&pool->deferred_flush_completions, bio);
993 spin_unlock_irq(&pool->lock);
996 static void process_prepared_mapping(struct dm_thin_new_mapping *m)
998 struct thin_c *tc = m->tc;
999 struct pool *pool = tc->pool;
1000 struct bio *bio = m->bio;
1004 cell_error(pool, m->cell);
1009 * Commit the prepared block into the mapping btree.
1010 * Any I/O for this block arriving after this point will get
1011 * remapped to it directly.
1013 r = dm_thin_insert_block(tc->td, m->virt_begin, m->data_block);
1015 metadata_operation_failed(pool, "dm_thin_insert_block", r);
1016 cell_error(pool, m->cell);
1021 * Release any bios held while the block was being provisioned.
1022 * If we are processing a write bio that completely covers the block,
1023 * we already processed it so can ignore it now when processing
1024 * the bios in the cell.
1027 inc_remap_and_issue_cell(tc, m->cell, m->data_block);
1028 complete_overwrite_bio(tc, bio);
1030 inc_all_io_entry(tc->pool, m->cell->holder);
1031 remap_and_issue(tc, m->cell->holder, m->data_block);
1032 inc_remap_and_issue_cell(tc, m->cell, m->data_block);
1037 mempool_free(m, &pool->mapping_pool);
1040 /*----------------------------------------------------------------*/
1042 static void free_discard_mapping(struct dm_thin_new_mapping *m)
1044 struct thin_c *tc = m->tc;
1047 cell_defer_no_holder(tc, m->cell);
1048 mempool_free(m, &tc->pool->mapping_pool);
1051 static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
1053 bio_io_error(m->bio);
1054 free_discard_mapping(m);
1057 static void process_prepared_discard_success(struct dm_thin_new_mapping *m)
1060 free_discard_mapping(m);
1063 static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m)
1066 struct thin_c *tc = m->tc;
1068 r = dm_thin_remove_range(tc->td, m->cell->key.block_begin, m->cell->key.block_end);
1070 metadata_operation_failed(tc->pool, "dm_thin_remove_range", r);
1071 bio_io_error(m->bio);
1075 cell_defer_no_holder(tc, m->cell);
1076 mempool_free(m, &tc->pool->mapping_pool);
1079 /*----------------------------------------------------------------*/
1081 static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m,
1082 struct bio *discard_parent)
1085 * We've already unmapped this range of blocks, but before we
1086 * passdown we have to check that these blocks are now unused.
1090 struct thin_c *tc = m->tc;
1091 struct pool *pool = tc->pool;
1092 dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin;
1093 struct discard_op op;
1095 begin_discard(&op, tc, discard_parent);
1097 /* find start of unmapped run */
1098 for (; b < end; b++) {
1099 r = dm_pool_block_is_shared(pool->pmd, b, &shared);
1110 /* find end of run */
1111 for (e = b + 1; e != end; e++) {
1112 r = dm_pool_block_is_shared(pool->pmd, e, &shared);
1120 r = issue_discard(&op, b, e);
1127 end_discard(&op, r);
1130 static void queue_passdown_pt2(struct dm_thin_new_mapping *m)
1132 unsigned long flags;
1133 struct pool *pool = m->tc->pool;
1135 spin_lock_irqsave(&pool->lock, flags);
1136 list_add_tail(&m->list, &pool->prepared_discards_pt2);
1137 spin_unlock_irqrestore(&pool->lock, flags);
1141 static void passdown_endio(struct bio *bio)
1144 * It doesn't matter if the passdown discard failed, we still want
1145 * to unmap (we ignore err).
1147 queue_passdown_pt2(bio->bi_private);
1151 static void process_prepared_discard_passdown_pt1(struct dm_thin_new_mapping *m)
1154 struct thin_c *tc = m->tc;
1155 struct pool *pool = tc->pool;
1156 struct bio *discard_parent;
1157 dm_block_t data_end = m->data_block + (m->virt_end - m->virt_begin);
1160 * Only this thread allocates blocks, so we can be sure that the
1161 * newly unmapped blocks will not be allocated before the end of
1164 r = dm_thin_remove_range(tc->td, m->virt_begin, m->virt_end);
1166 metadata_operation_failed(pool, "dm_thin_remove_range", r);
1167 bio_io_error(m->bio);
1168 cell_defer_no_holder(tc, m->cell);
1169 mempool_free(m, &pool->mapping_pool);
1174 * Increment the unmapped blocks. This prevents a race between the
1175 * passdown io and reallocation of freed blocks.
1177 r = dm_pool_inc_data_range(pool->pmd, m->data_block, data_end);
1179 metadata_operation_failed(pool, "dm_pool_inc_data_range", r);
1180 bio_io_error(m->bio);
1181 cell_defer_no_holder(tc, m->cell);
1182 mempool_free(m, &pool->mapping_pool);
1186 discard_parent = bio_alloc(NULL, 1, 0, GFP_NOIO);
1187 discard_parent->bi_end_io = passdown_endio;
1188 discard_parent->bi_private = m;
1189 if (m->maybe_shared)
1190 passdown_double_checking_shared_status(m, discard_parent);
1192 struct discard_op op;
1194 begin_discard(&op, tc, discard_parent);
1195 r = issue_discard(&op, m->data_block, data_end);
1196 end_discard(&op, r);
1200 static void process_prepared_discard_passdown_pt2(struct dm_thin_new_mapping *m)
1203 struct thin_c *tc = m->tc;
1204 struct pool *pool = tc->pool;
1207 * The passdown has completed, so now we can decrement all those
1210 r = dm_pool_dec_data_range(pool->pmd, m->data_block,
1211 m->data_block + (m->virt_end - m->virt_begin));
1213 metadata_operation_failed(pool, "dm_pool_dec_data_range", r);
1214 bio_io_error(m->bio);
1218 cell_defer_no_holder(tc, m->cell);
1219 mempool_free(m, &pool->mapping_pool);
1222 static void process_prepared(struct pool *pool, struct list_head *head,
1223 process_mapping_fn *fn)
1225 struct list_head maps;
1226 struct dm_thin_new_mapping *m, *tmp;
1228 INIT_LIST_HEAD(&maps);
1229 spin_lock_irq(&pool->lock);
1230 list_splice_init(head, &maps);
1231 spin_unlock_irq(&pool->lock);
1233 list_for_each_entry_safe(m, tmp, &maps, list)
1238 * Deferred bio jobs.
1240 static int io_overlaps_block(struct pool *pool, struct bio *bio)
1242 return bio->bi_iter.bi_size ==
1243 (pool->sectors_per_block << SECTOR_SHIFT);
1246 static int io_overwrites_block(struct pool *pool, struct bio *bio)
1248 return (bio_data_dir(bio) == WRITE) &&
1249 io_overlaps_block(pool, bio);
1252 static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
1255 *save = bio->bi_end_io;
1256 bio->bi_end_io = fn;
1259 static int ensure_next_mapping(struct pool *pool)
1261 if (pool->next_mapping)
1264 pool->next_mapping = mempool_alloc(&pool->mapping_pool, GFP_ATOMIC);
1266 return pool->next_mapping ? 0 : -ENOMEM;
1269 static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
1271 struct dm_thin_new_mapping *m = pool->next_mapping;
1273 BUG_ON(!pool->next_mapping);
1275 memset(m, 0, sizeof(struct dm_thin_new_mapping));
1276 INIT_LIST_HEAD(&m->list);
1279 pool->next_mapping = NULL;
1284 static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m,
1285 sector_t begin, sector_t end)
1287 struct dm_io_region to;
1289 to.bdev = tc->pool_dev->bdev;
1291 to.count = end - begin;
1293 dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m);
1296 static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio,
1297 dm_block_t data_begin,
1298 struct dm_thin_new_mapping *m)
1300 struct pool *pool = tc->pool;
1301 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1303 h->overwrite_mapping = m;
1305 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
1306 inc_all_io_entry(pool, bio);
1307 remap_and_issue(tc, bio, data_begin);
1311 * A partial copy also needs to zero the uncopied region.
1313 static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
1314 struct dm_dev *origin, dm_block_t data_origin,
1315 dm_block_t data_dest,
1316 struct dm_bio_prison_cell *cell, struct bio *bio,
1319 struct pool *pool = tc->pool;
1320 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1323 m->virt_begin = virt_block;
1324 m->virt_end = virt_block + 1u;
1325 m->data_block = data_dest;
1329 * quiesce action + copy action + an extra reference held for the
1330 * duration of this function (we may need to inc later for a
1333 atomic_set(&m->prepare_actions, 3);
1335 if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
1336 complete_mapping_preparation(m); /* already quiesced */
1339 * IO to pool_dev remaps to the pool target's data_dev.
1341 * If the whole block of data is being overwritten, we can issue the
1342 * bio immediately. Otherwise we use kcopyd to clone the data first.
1344 if (io_overwrites_block(pool, bio))
1345 remap_and_issue_overwrite(tc, bio, data_dest, m);
1347 struct dm_io_region from, to;
1349 from.bdev = origin->bdev;
1350 from.sector = data_origin * pool->sectors_per_block;
1353 to.bdev = tc->pool_dev->bdev;
1354 to.sector = data_dest * pool->sectors_per_block;
1357 dm_kcopyd_copy(pool->copier, &from, 1, &to,
1358 0, copy_complete, m);
1361 * Do we need to zero a tail region?
1363 if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) {
1364 atomic_inc(&m->prepare_actions);
1366 data_dest * pool->sectors_per_block + len,
1367 (data_dest + 1) * pool->sectors_per_block);
1371 complete_mapping_preparation(m); /* drop our ref */
1374 static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
1375 dm_block_t data_origin, dm_block_t data_dest,
1376 struct dm_bio_prison_cell *cell, struct bio *bio)
1378 schedule_copy(tc, virt_block, tc->pool_dev,
1379 data_origin, data_dest, cell, bio,
1380 tc->pool->sectors_per_block);
1383 static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
1384 dm_block_t data_block, struct dm_bio_prison_cell *cell,
1387 struct pool *pool = tc->pool;
1388 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1390 atomic_set(&m->prepare_actions, 1); /* no need to quiesce */
1392 m->virt_begin = virt_block;
1393 m->virt_end = virt_block + 1u;
1394 m->data_block = data_block;
1398 * If the whole block of data is being overwritten or we are not
1399 * zeroing pre-existing data, we can issue the bio immediately.
1400 * Otherwise we use kcopyd to zero the data first.
1402 if (pool->pf.zero_new_blocks) {
1403 if (io_overwrites_block(pool, bio))
1404 remap_and_issue_overwrite(tc, bio, data_block, m);
1406 ll_zero(tc, m, data_block * pool->sectors_per_block,
1407 (data_block + 1) * pool->sectors_per_block);
1409 process_prepared_mapping(m);
1412 static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
1413 dm_block_t data_dest,
1414 struct dm_bio_prison_cell *cell, struct bio *bio)
1416 struct pool *pool = tc->pool;
1417 sector_t virt_block_begin = virt_block * pool->sectors_per_block;
1418 sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block;
1420 if (virt_block_end <= tc->origin_size)
1421 schedule_copy(tc, virt_block, tc->origin_dev,
1422 virt_block, data_dest, cell, bio,
1423 pool->sectors_per_block);
1425 else if (virt_block_begin < tc->origin_size)
1426 schedule_copy(tc, virt_block, tc->origin_dev,
1427 virt_block, data_dest, cell, bio,
1428 tc->origin_size - virt_block_begin);
1431 schedule_zero(tc, virt_block, data_dest, cell, bio);
1434 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode);
1436 static void requeue_bios(struct pool *pool);
1438 static bool is_read_only_pool_mode(enum pool_mode mode)
1440 return (mode == PM_OUT_OF_METADATA_SPACE || mode == PM_READ_ONLY);
1443 static bool is_read_only(struct pool *pool)
1445 return is_read_only_pool_mode(get_pool_mode(pool));
1448 static void check_for_metadata_space(struct pool *pool)
1451 const char *ooms_reason = NULL;
1454 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free);
1456 ooms_reason = "Could not get free metadata blocks";
1458 ooms_reason = "No free metadata blocks";
1460 if (ooms_reason && !is_read_only(pool)) {
1461 DMERR("%s", ooms_reason);
1462 set_pool_mode(pool, PM_OUT_OF_METADATA_SPACE);
1466 static void check_for_data_space(struct pool *pool)
1471 if (get_pool_mode(pool) != PM_OUT_OF_DATA_SPACE)
1474 r = dm_pool_get_free_block_count(pool->pmd, &nr_free);
1479 set_pool_mode(pool, PM_WRITE);
1485 * A non-zero return indicates read_only or fail_io mode.
1486 * Many callers don't care about the return value.
1488 static int commit(struct pool *pool)
1492 if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE)
1495 r = dm_pool_commit_metadata(pool->pmd);
1497 metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
1499 check_for_metadata_space(pool);
1500 check_for_data_space(pool);
1506 static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks)
1508 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
1509 DMWARN("%s: reached low water mark for data device: sending event.",
1510 dm_device_name(pool->pool_md));
1511 spin_lock_irq(&pool->lock);
1512 pool->low_water_triggered = true;
1513 spin_unlock_irq(&pool->lock);
1514 dm_table_event(pool->ti->table);
1518 static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
1521 dm_block_t free_blocks;
1522 struct pool *pool = tc->pool;
1524 if (WARN_ON(get_pool_mode(pool) != PM_WRITE))
1527 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1529 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1533 check_low_water_mark(pool, free_blocks);
1537 * Try to commit to see if that will free up some
1544 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1546 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1551 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1556 r = dm_pool_alloc_data_block(pool->pmd, result);
1559 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1561 metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
1565 r = dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks);
1567 metadata_operation_failed(pool, "dm_pool_get_free_metadata_block_count", r);
1572 /* Let's commit before we use up the metadata reserve. */
1582 * If we have run out of space, queue bios until the device is
1583 * resumed, presumably after having been reloaded with more space.
1585 static void retry_on_resume(struct bio *bio)
1587 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1588 struct thin_c *tc = h->tc;
1590 spin_lock_irq(&tc->lock);
1591 bio_list_add(&tc->retry_on_resume_list, bio);
1592 spin_unlock_irq(&tc->lock);
1595 static blk_status_t should_error_unserviceable_bio(struct pool *pool)
1597 enum pool_mode m = get_pool_mode(pool);
1601 /* Shouldn't get here */
1602 DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
1603 return BLK_STS_IOERR;
1605 case PM_OUT_OF_DATA_SPACE:
1606 return pool->pf.error_if_no_space ? BLK_STS_NOSPC : 0;
1608 case PM_OUT_OF_METADATA_SPACE:
1611 return BLK_STS_IOERR;
1613 /* Shouldn't get here */
1614 DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1615 return BLK_STS_IOERR;
1619 static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
1621 blk_status_t error = should_error_unserviceable_bio(pool);
1624 bio->bi_status = error;
1627 retry_on_resume(bio);
1630 static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell)
1633 struct bio_list bios;
1636 error = should_error_unserviceable_bio(pool);
1638 cell_error_with_code(pool, cell, error);
1642 bio_list_init(&bios);
1643 cell_release(pool, cell, &bios);
1645 while ((bio = bio_list_pop(&bios)))
1646 retry_on_resume(bio);
1649 static void process_discard_cell_no_passdown(struct thin_c *tc,
1650 struct dm_bio_prison_cell *virt_cell)
1652 struct pool *pool = tc->pool;
1653 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1656 * We don't need to lock the data blocks, since there's no
1657 * passdown. We only lock data blocks for allocation and breaking sharing.
1660 m->virt_begin = virt_cell->key.block_begin;
1661 m->virt_end = virt_cell->key.block_end;
1662 m->cell = virt_cell;
1663 m->bio = virt_cell->holder;
1665 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1666 pool->process_prepared_discard(m);
1669 static void break_up_discard_bio(struct thin_c *tc, dm_block_t begin, dm_block_t end,
1672 struct pool *pool = tc->pool;
1676 struct dm_cell_key data_key;
1677 struct dm_bio_prison_cell *data_cell;
1678 struct dm_thin_new_mapping *m;
1679 dm_block_t virt_begin, virt_end, data_begin, data_end;
1680 dm_block_t len, next_boundary;
1682 while (begin != end) {
1683 r = dm_thin_find_mapped_range(tc->td, begin, end, &virt_begin, &virt_end,
1684 &data_begin, &maybe_shared);
1687 * Silently fail, letting any mappings we've
1693 data_end = data_begin + (virt_end - virt_begin);
1696 * Make sure the data region obeys the bio prison restrictions.
1698 while (data_begin < data_end) {
1699 r = ensure_next_mapping(pool);
1701 return; /* we did our best */
1703 next_boundary = ((data_begin >> BIO_PRISON_MAX_RANGE_SHIFT) + 1)
1704 << BIO_PRISON_MAX_RANGE_SHIFT;
1705 len = min_t(sector_t, data_end - data_begin, next_boundary - data_begin);
1707 /* This key is certainly within range given the above splitting */
1708 (void) build_key(tc->td, PHYSICAL, data_begin, data_begin + len, &data_key);
1709 if (bio_detain(tc->pool, &data_key, NULL, &data_cell)) {
1710 /* contention, we'll give up with this range */
1716 * IO may still be going to the destination block. We must
1717 * quiesce before we can do the removal.
1719 m = get_next_mapping(pool);
1721 m->maybe_shared = maybe_shared;
1722 m->virt_begin = virt_begin;
1723 m->virt_end = virt_begin + len;
1724 m->data_block = data_begin;
1725 m->cell = data_cell;
1729 * The parent bio must not complete before sub discard bios are
1730 * chained to it (see end_discard's bio_chain)!
1732 * This per-mapping bi_remaining increment is paired with
1733 * the implicit decrement that occurs via bio_endio() in
1736 bio_inc_remaining(bio);
1737 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1738 pool->process_prepared_discard(m);
1748 static void process_discard_cell_passdown(struct thin_c *tc, struct dm_bio_prison_cell *virt_cell)
1750 struct bio *bio = virt_cell->holder;
1751 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1754 * The virt_cell will only get freed once the origin bio completes.
1755 * This means it will remain locked while all the individual
1756 * passdown bios are in flight.
1758 h->cell = virt_cell;
1759 break_up_discard_bio(tc, virt_cell->key.block_begin, virt_cell->key.block_end, bio);
1762 * We complete the bio now, knowing that the bi_remaining field
1763 * will prevent completion until the sub range discards have
1769 static void process_discard_bio(struct thin_c *tc, struct bio *bio)
1771 dm_block_t begin, end;
1772 struct dm_cell_key virt_key;
1773 struct dm_bio_prison_cell *virt_cell;
1775 get_bio_block_range(tc, bio, &begin, &end);
1778 * The discard covers less than a block.
1784 if (unlikely(!build_key(tc->td, VIRTUAL, begin, end, &virt_key))) {
1785 DMERR_LIMIT("Discard doesn't respect bio prison limits");
1790 if (bio_detain(tc->pool, &virt_key, bio, &virt_cell)) {
1792 * Potential starvation issue: We're relying on the
1793 * fs/application being well behaved, and not trying to
1794 * send IO to a region at the same time as discarding it.
1795 * If they do this persistently then it's possible this
1796 * cell will never be granted.
1801 tc->pool->process_discard_cell(tc, virt_cell);
1804 static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
1805 struct dm_cell_key *key,
1806 struct dm_thin_lookup_result *lookup_result,
1807 struct dm_bio_prison_cell *cell)
1810 dm_block_t data_block;
1811 struct pool *pool = tc->pool;
1813 r = alloc_data_block(tc, &data_block);
1816 schedule_internal_copy(tc, block, lookup_result->block,
1817 data_block, cell, bio);
1821 retry_bios_on_resume(pool, cell);
1825 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1827 cell_error(pool, cell);
1832 static void __remap_and_issue_shared_cell(void *context,
1833 struct dm_bio_prison_cell *cell)
1835 struct remap_info *info = context;
1838 while ((bio = bio_list_pop(&cell->bios))) {
1839 if (bio_data_dir(bio) == WRITE || op_is_flush(bio->bi_opf) ||
1840 bio_op(bio) == REQ_OP_DISCARD)
1841 bio_list_add(&info->defer_bios, bio);
1843 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1845 h->shared_read_entry = dm_deferred_entry_inc(info->tc->pool->shared_read_ds);
1846 inc_all_io_entry(info->tc->pool, bio);
1847 bio_list_add(&info->issue_bios, bio);
1852 static void remap_and_issue_shared_cell(struct thin_c *tc,
1853 struct dm_bio_prison_cell *cell,
1857 struct remap_info info;
1860 bio_list_init(&info.defer_bios);
1861 bio_list_init(&info.issue_bios);
1863 cell_visit_release(tc->pool, __remap_and_issue_shared_cell,
1866 while ((bio = bio_list_pop(&info.defer_bios)))
1867 thin_defer_bio(tc, bio);
1869 while ((bio = bio_list_pop(&info.issue_bios)))
1870 remap_and_issue(tc, bio, block);
1873 static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1875 struct dm_thin_lookup_result *lookup_result,
1876 struct dm_bio_prison_cell *virt_cell)
1878 struct dm_bio_prison_cell *data_cell;
1879 struct pool *pool = tc->pool;
1880 struct dm_cell_key key;
1883 * If cell is already occupied, then sharing is already in the process
1884 * of being broken so we have nothing further to do here.
1886 build_data_key(tc->td, lookup_result->block, &key);
1887 if (bio_detain(pool, &key, bio, &data_cell)) {
1888 cell_defer_no_holder(tc, virt_cell);
1892 if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) {
1893 break_sharing(tc, bio, block, &key, lookup_result, data_cell);
1894 cell_defer_no_holder(tc, virt_cell);
1896 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1898 h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
1899 inc_all_io_entry(pool, bio);
1900 remap_and_issue(tc, bio, lookup_result->block);
1902 remap_and_issue_shared_cell(tc, data_cell, lookup_result->block);
1903 remap_and_issue_shared_cell(tc, virt_cell, lookup_result->block);
1907 static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
1908 struct dm_bio_prison_cell *cell)
1911 dm_block_t data_block;
1912 struct pool *pool = tc->pool;
1915 * Remap empty bios (flushes) immediately, without provisioning.
1917 if (!bio->bi_iter.bi_size) {
1918 inc_all_io_entry(pool, bio);
1919 cell_defer_no_holder(tc, cell);
1921 remap_and_issue(tc, bio, 0);
1926 * Fill read bios with zeroes and complete them immediately.
1928 if (bio_data_dir(bio) == READ) {
1930 cell_defer_no_holder(tc, cell);
1935 r = alloc_data_block(tc, &data_block);
1939 schedule_external_copy(tc, block, data_block, cell, bio);
1941 schedule_zero(tc, block, data_block, cell, bio);
1945 retry_bios_on_resume(pool, cell);
1949 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1951 cell_error(pool, cell);
1956 static void process_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
1959 struct pool *pool = tc->pool;
1960 struct bio *bio = cell->holder;
1961 dm_block_t block = get_bio_block(tc, bio);
1962 struct dm_thin_lookup_result lookup_result;
1964 if (tc->requeue_mode) {
1965 cell_requeue(pool, cell);
1969 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1972 if (lookup_result.shared)
1973 process_shared_bio(tc, bio, block, &lookup_result, cell);
1975 inc_all_io_entry(pool, bio);
1976 remap_and_issue(tc, bio, lookup_result.block);
1977 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
1982 if (bio_data_dir(bio) == READ && tc->origin_dev) {
1983 inc_all_io_entry(pool, bio);
1984 cell_defer_no_holder(tc, cell);
1986 if (bio_end_sector(bio) <= tc->origin_size)
1987 remap_to_origin_and_issue(tc, bio);
1989 else if (bio->bi_iter.bi_sector < tc->origin_size) {
1991 bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT;
1992 remap_to_origin_and_issue(tc, bio);
1999 provision_block(tc, bio, block, cell);
2003 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
2005 cell_defer_no_holder(tc, cell);
2011 static void process_bio(struct thin_c *tc, struct bio *bio)
2013 struct pool *pool = tc->pool;
2014 dm_block_t block = get_bio_block(tc, bio);
2015 struct dm_bio_prison_cell *cell;
2016 struct dm_cell_key key;
2019 * If cell is already occupied, then the block is already
2020 * being provisioned so we have nothing further to do here.
2022 build_virtual_key(tc->td, block, &key);
2023 if (bio_detain(pool, &key, bio, &cell))
2026 process_cell(tc, cell);
2029 static void __process_bio_read_only(struct thin_c *tc, struct bio *bio,
2030 struct dm_bio_prison_cell *cell)
2033 int rw = bio_data_dir(bio);
2034 dm_block_t block = get_bio_block(tc, bio);
2035 struct dm_thin_lookup_result lookup_result;
2037 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
2040 if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) {
2041 handle_unserviceable_bio(tc->pool, bio);
2043 cell_defer_no_holder(tc, cell);
2045 inc_all_io_entry(tc->pool, bio);
2046 remap_and_issue(tc, bio, lookup_result.block);
2048 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
2054 cell_defer_no_holder(tc, cell);
2056 handle_unserviceable_bio(tc->pool, bio);
2060 if (tc->origin_dev) {
2061 inc_all_io_entry(tc->pool, bio);
2062 remap_to_origin_and_issue(tc, bio);
2071 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
2074 cell_defer_no_holder(tc, cell);
2080 static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
2082 __process_bio_read_only(tc, bio, NULL);
2085 static void process_cell_read_only(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2087 __process_bio_read_only(tc, cell->holder, cell);
2090 static void process_bio_success(struct thin_c *tc, struct bio *bio)
2095 static void process_bio_fail(struct thin_c *tc, struct bio *bio)
2100 static void process_cell_success(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2102 cell_success(tc->pool, cell);
2105 static void process_cell_fail(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2107 cell_error(tc->pool, cell);
2111 * FIXME: should we also commit due to size of transaction, measured in
2114 static int need_commit_due_to_time(struct pool *pool)
2116 return !time_in_range(jiffies, pool->last_commit_jiffies,
2117 pool->last_commit_jiffies + COMMIT_PERIOD);
2120 #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node)
2121 #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook))
2123 static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio)
2125 struct rb_node **rbp, *parent;
2126 struct dm_thin_endio_hook *pbd;
2127 sector_t bi_sector = bio->bi_iter.bi_sector;
2129 rbp = &tc->sort_bio_list.rb_node;
2133 pbd = thin_pbd(parent);
2135 if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector)
2136 rbp = &(*rbp)->rb_left;
2138 rbp = &(*rbp)->rb_right;
2141 pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2142 rb_link_node(&pbd->rb_node, parent, rbp);
2143 rb_insert_color(&pbd->rb_node, &tc->sort_bio_list);
2146 static void __extract_sorted_bios(struct thin_c *tc)
2148 struct rb_node *node;
2149 struct dm_thin_endio_hook *pbd;
2152 for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) {
2153 pbd = thin_pbd(node);
2154 bio = thin_bio(pbd);
2156 bio_list_add(&tc->deferred_bio_list, bio);
2157 rb_erase(&pbd->rb_node, &tc->sort_bio_list);
2160 WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list));
2163 static void __sort_thin_deferred_bios(struct thin_c *tc)
2166 struct bio_list bios;
2168 bio_list_init(&bios);
2169 bio_list_merge(&bios, &tc->deferred_bio_list);
2170 bio_list_init(&tc->deferred_bio_list);
2172 /* Sort deferred_bio_list using rb-tree */
2173 while ((bio = bio_list_pop(&bios)))
2174 __thin_bio_rb_add(tc, bio);
2177 * Transfer the sorted bios in sort_bio_list back to
2178 * deferred_bio_list to allow lockless submission of
2181 __extract_sorted_bios(tc);
2184 static void process_thin_deferred_bios(struct thin_c *tc)
2186 struct pool *pool = tc->pool;
2188 struct bio_list bios;
2189 struct blk_plug plug;
2190 unsigned int count = 0;
2192 if (tc->requeue_mode) {
2193 error_thin_bio_list(tc, &tc->deferred_bio_list,
2194 BLK_STS_DM_REQUEUE);
2198 bio_list_init(&bios);
2200 spin_lock_irq(&tc->lock);
2202 if (bio_list_empty(&tc->deferred_bio_list)) {
2203 spin_unlock_irq(&tc->lock);
2207 __sort_thin_deferred_bios(tc);
2209 bio_list_merge(&bios, &tc->deferred_bio_list);
2210 bio_list_init(&tc->deferred_bio_list);
2212 spin_unlock_irq(&tc->lock);
2214 blk_start_plug(&plug);
2215 while ((bio = bio_list_pop(&bios))) {
2217 * If we've got no free new_mapping structs, and processing
2218 * this bio might require one, we pause until there are some
2219 * prepared mappings to process.
2221 if (ensure_next_mapping(pool)) {
2222 spin_lock_irq(&tc->lock);
2223 bio_list_add(&tc->deferred_bio_list, bio);
2224 bio_list_merge(&tc->deferred_bio_list, &bios);
2225 spin_unlock_irq(&tc->lock);
2229 if (bio_op(bio) == REQ_OP_DISCARD)
2230 pool->process_discard(tc, bio);
2232 pool->process_bio(tc, bio);
2234 if ((count++ & 127) == 0) {
2235 throttle_work_update(&pool->throttle);
2236 dm_pool_issue_prefetches(pool->pmd);
2240 blk_finish_plug(&plug);
2243 static int cmp_cells(const void *lhs, const void *rhs)
2245 struct dm_bio_prison_cell *lhs_cell = *((struct dm_bio_prison_cell **) lhs);
2246 struct dm_bio_prison_cell *rhs_cell = *((struct dm_bio_prison_cell **) rhs);
2248 BUG_ON(!lhs_cell->holder);
2249 BUG_ON(!rhs_cell->holder);
2251 if (lhs_cell->holder->bi_iter.bi_sector < rhs_cell->holder->bi_iter.bi_sector)
2254 if (lhs_cell->holder->bi_iter.bi_sector > rhs_cell->holder->bi_iter.bi_sector)
2260 static unsigned int sort_cells(struct pool *pool, struct list_head *cells)
2262 unsigned int count = 0;
2263 struct dm_bio_prison_cell *cell, *tmp;
2265 list_for_each_entry_safe(cell, tmp, cells, user_list) {
2266 if (count >= CELL_SORT_ARRAY_SIZE)
2269 pool->cell_sort_array[count++] = cell;
2270 list_del(&cell->user_list);
2273 sort(pool->cell_sort_array, count, sizeof(cell), cmp_cells, NULL);
2278 static void process_thin_deferred_cells(struct thin_c *tc)
2280 struct pool *pool = tc->pool;
2281 struct list_head cells;
2282 struct dm_bio_prison_cell *cell;
2283 unsigned int i, j, count;
2285 INIT_LIST_HEAD(&cells);
2287 spin_lock_irq(&tc->lock);
2288 list_splice_init(&tc->deferred_cells, &cells);
2289 spin_unlock_irq(&tc->lock);
2291 if (list_empty(&cells))
2295 count = sort_cells(tc->pool, &cells);
2297 for (i = 0; i < count; i++) {
2298 cell = pool->cell_sort_array[i];
2299 BUG_ON(!cell->holder);
2302 * If we've got no free new_mapping structs, and processing
2303 * this bio might require one, we pause until there are some
2304 * prepared mappings to process.
2306 if (ensure_next_mapping(pool)) {
2307 for (j = i; j < count; j++)
2308 list_add(&pool->cell_sort_array[j]->user_list, &cells);
2310 spin_lock_irq(&tc->lock);
2311 list_splice(&cells, &tc->deferred_cells);
2312 spin_unlock_irq(&tc->lock);
2316 if (bio_op(cell->holder) == REQ_OP_DISCARD)
2317 pool->process_discard_cell(tc, cell);
2319 pool->process_cell(tc, cell);
2322 } while (!list_empty(&cells));
2325 static void thin_get(struct thin_c *tc);
2326 static void thin_put(struct thin_c *tc);
2329 * We can't hold rcu_read_lock() around code that can block. So we
2330 * find a thin with the rcu lock held; bump a refcount; then drop
2333 static struct thin_c *get_first_thin(struct pool *pool)
2335 struct thin_c *tc = NULL;
2338 if (!list_empty(&pool->active_thins)) {
2339 tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list);
2347 static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc)
2349 struct thin_c *old_tc = tc;
2352 list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) {
2364 static void process_deferred_bios(struct pool *pool)
2367 struct bio_list bios, bio_completions;
2370 tc = get_first_thin(pool);
2372 process_thin_deferred_cells(tc);
2373 process_thin_deferred_bios(tc);
2374 tc = get_next_thin(pool, tc);
2378 * If there are any deferred flush bios, we must commit the metadata
2379 * before issuing them or signaling their completion.
2381 bio_list_init(&bios);
2382 bio_list_init(&bio_completions);
2384 spin_lock_irq(&pool->lock);
2385 bio_list_merge(&bios, &pool->deferred_flush_bios);
2386 bio_list_init(&pool->deferred_flush_bios);
2388 bio_list_merge(&bio_completions, &pool->deferred_flush_completions);
2389 bio_list_init(&pool->deferred_flush_completions);
2390 spin_unlock_irq(&pool->lock);
2392 if (bio_list_empty(&bios) && bio_list_empty(&bio_completions) &&
2393 !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool)))
2397 bio_list_merge(&bios, &bio_completions);
2399 while ((bio = bio_list_pop(&bios)))
2403 pool->last_commit_jiffies = jiffies;
2405 while ((bio = bio_list_pop(&bio_completions)))
2408 while ((bio = bio_list_pop(&bios))) {
2410 * The data device was flushed as part of metadata commit,
2411 * so complete redundant flushes immediately.
2413 if (bio->bi_opf & REQ_PREFLUSH)
2416 dm_submit_bio_remap(bio, NULL);
2420 static void do_worker(struct work_struct *ws)
2422 struct pool *pool = container_of(ws, struct pool, worker);
2424 throttle_work_start(&pool->throttle);
2425 dm_pool_issue_prefetches(pool->pmd);
2426 throttle_work_update(&pool->throttle);
2427 process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
2428 throttle_work_update(&pool->throttle);
2429 process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
2430 throttle_work_update(&pool->throttle);
2431 process_prepared(pool, &pool->prepared_discards_pt2, &pool->process_prepared_discard_pt2);
2432 throttle_work_update(&pool->throttle);
2433 process_deferred_bios(pool);
2434 throttle_work_complete(&pool->throttle);
2438 * We want to commit periodically so that not too much
2439 * unwritten data builds up.
2441 static void do_waker(struct work_struct *ws)
2443 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
2446 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
2450 * We're holding onto IO to allow userland time to react. After the
2451 * timeout either the pool will have been resized (and thus back in
2452 * PM_WRITE mode), or we degrade to PM_OUT_OF_DATA_SPACE w/ error_if_no_space.
2454 static void do_no_space_timeout(struct work_struct *ws)
2456 struct pool *pool = container_of(to_delayed_work(ws), struct pool,
2459 if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) {
2460 pool->pf.error_if_no_space = true;
2461 notify_of_pool_mode_change(pool);
2462 error_retry_list_with_code(pool, BLK_STS_NOSPC);
2466 /*----------------------------------------------------------------*/
2469 struct work_struct worker;
2470 struct completion complete;
2473 static struct pool_work *to_pool_work(struct work_struct *ws)
2475 return container_of(ws, struct pool_work, worker);
2478 static void pool_work_complete(struct pool_work *pw)
2480 complete(&pw->complete);
2483 static void pool_work_wait(struct pool_work *pw, struct pool *pool,
2484 void (*fn)(struct work_struct *))
2486 INIT_WORK_ONSTACK(&pw->worker, fn);
2487 init_completion(&pw->complete);
2488 queue_work(pool->wq, &pw->worker);
2489 wait_for_completion(&pw->complete);
2492 /*----------------------------------------------------------------*/
2494 struct noflush_work {
2495 struct pool_work pw;
2499 static struct noflush_work *to_noflush(struct work_struct *ws)
2501 return container_of(to_pool_work(ws), struct noflush_work, pw);
2504 static void do_noflush_start(struct work_struct *ws)
2506 struct noflush_work *w = to_noflush(ws);
2508 w->tc->requeue_mode = true;
2510 pool_work_complete(&w->pw);
2513 static void do_noflush_stop(struct work_struct *ws)
2515 struct noflush_work *w = to_noflush(ws);
2517 w->tc->requeue_mode = false;
2518 pool_work_complete(&w->pw);
2521 static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *))
2523 struct noflush_work w;
2526 pool_work_wait(&w.pw, tc->pool, fn);
2529 /*----------------------------------------------------------------*/
2531 static bool passdown_enabled(struct pool_c *pt)
2533 return pt->adjusted_pf.discard_passdown;
2536 static void set_discard_callbacks(struct pool *pool)
2538 struct pool_c *pt = pool->ti->private;
2540 if (passdown_enabled(pt)) {
2541 pool->process_discard_cell = process_discard_cell_passdown;
2542 pool->process_prepared_discard = process_prepared_discard_passdown_pt1;
2543 pool->process_prepared_discard_pt2 = process_prepared_discard_passdown_pt2;
2545 pool->process_discard_cell = process_discard_cell_no_passdown;
2546 pool->process_prepared_discard = process_prepared_discard_no_passdown;
2550 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode)
2552 struct pool_c *pt = pool->ti->private;
2553 bool needs_check = dm_pool_metadata_needs_check(pool->pmd);
2554 enum pool_mode old_mode = get_pool_mode(pool);
2555 unsigned long no_space_timeout = READ_ONCE(no_space_timeout_secs) * HZ;
2558 * Never allow the pool to transition to PM_WRITE mode if user
2559 * intervention is required to verify metadata and data consistency.
2561 if (new_mode == PM_WRITE && needs_check) {
2562 DMERR("%s: unable to switch pool to write mode until repaired.",
2563 dm_device_name(pool->pool_md));
2564 if (old_mode != new_mode)
2565 new_mode = old_mode;
2567 new_mode = PM_READ_ONLY;
2570 * If we were in PM_FAIL mode, rollback of metadata failed. We're
2571 * not going to recover without a thin_repair. So we never let the
2572 * pool move out of the old mode.
2574 if (old_mode == PM_FAIL)
2575 new_mode = old_mode;
2579 dm_pool_metadata_read_only(pool->pmd);
2580 pool->process_bio = process_bio_fail;
2581 pool->process_discard = process_bio_fail;
2582 pool->process_cell = process_cell_fail;
2583 pool->process_discard_cell = process_cell_fail;
2584 pool->process_prepared_mapping = process_prepared_mapping_fail;
2585 pool->process_prepared_discard = process_prepared_discard_fail;
2587 error_retry_list(pool);
2590 case PM_OUT_OF_METADATA_SPACE:
2592 dm_pool_metadata_read_only(pool->pmd);
2593 pool->process_bio = process_bio_read_only;
2594 pool->process_discard = process_bio_success;
2595 pool->process_cell = process_cell_read_only;
2596 pool->process_discard_cell = process_cell_success;
2597 pool->process_prepared_mapping = process_prepared_mapping_fail;
2598 pool->process_prepared_discard = process_prepared_discard_success;
2600 error_retry_list(pool);
2603 case PM_OUT_OF_DATA_SPACE:
2605 * Ideally we'd never hit this state; the low water mark
2606 * would trigger userland to extend the pool before we
2607 * completely run out of data space. However, many small
2608 * IOs to unprovisioned space can consume data space at an
2609 * alarming rate. Adjust your low water mark if you're
2610 * frequently seeing this mode.
2612 pool->out_of_data_space = true;
2613 pool->process_bio = process_bio_read_only;
2614 pool->process_discard = process_discard_bio;
2615 pool->process_cell = process_cell_read_only;
2616 pool->process_prepared_mapping = process_prepared_mapping;
2617 set_discard_callbacks(pool);
2619 if (!pool->pf.error_if_no_space && no_space_timeout)
2620 queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout);
2624 if (old_mode == PM_OUT_OF_DATA_SPACE)
2625 cancel_delayed_work_sync(&pool->no_space_timeout);
2626 pool->out_of_data_space = false;
2627 pool->pf.error_if_no_space = pt->requested_pf.error_if_no_space;
2628 dm_pool_metadata_read_write(pool->pmd);
2629 pool->process_bio = process_bio;
2630 pool->process_discard = process_discard_bio;
2631 pool->process_cell = process_cell;
2632 pool->process_prepared_mapping = process_prepared_mapping;
2633 set_discard_callbacks(pool);
2637 pool->pf.mode = new_mode;
2639 * The pool mode may have changed, sync it so bind_control_target()
2640 * doesn't cause an unexpected mode transition on resume.
2642 pt->adjusted_pf.mode = new_mode;
2644 if (old_mode != new_mode)
2645 notify_of_pool_mode_change(pool);
2648 static void abort_transaction(struct pool *pool)
2650 const char *dev_name = dm_device_name(pool->pool_md);
2652 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name);
2653 if (dm_pool_abort_metadata(pool->pmd)) {
2654 DMERR("%s: failed to abort metadata transaction", dev_name);
2655 set_pool_mode(pool, PM_FAIL);
2658 if (dm_pool_metadata_set_needs_check(pool->pmd)) {
2659 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name);
2660 set_pool_mode(pool, PM_FAIL);
2664 static void metadata_operation_failed(struct pool *pool, const char *op, int r)
2666 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
2667 dm_device_name(pool->pool_md), op, r);
2669 abort_transaction(pool);
2670 set_pool_mode(pool, PM_READ_ONLY);
2673 /*----------------------------------------------------------------*/
2676 * Mapping functions.
2680 * Called only while mapping a thin bio to hand it over to the workqueue.
2682 static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
2684 struct pool *pool = tc->pool;
2686 spin_lock_irq(&tc->lock);
2687 bio_list_add(&tc->deferred_bio_list, bio);
2688 spin_unlock_irq(&tc->lock);
2693 static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio)
2695 struct pool *pool = tc->pool;
2697 throttle_lock(&pool->throttle);
2698 thin_defer_bio(tc, bio);
2699 throttle_unlock(&pool->throttle);
2702 static void thin_defer_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2704 struct pool *pool = tc->pool;
2706 throttle_lock(&pool->throttle);
2707 spin_lock_irq(&tc->lock);
2708 list_add_tail(&cell->user_list, &tc->deferred_cells);
2709 spin_unlock_irq(&tc->lock);
2710 throttle_unlock(&pool->throttle);
2715 static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
2717 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2720 h->shared_read_entry = NULL;
2721 h->all_io_entry = NULL;
2722 h->overwrite_mapping = NULL;
2727 * Non-blocking function called from the thin target's map function.
2729 static int thin_bio_map(struct dm_target *ti, struct bio *bio)
2732 struct thin_c *tc = ti->private;
2733 dm_block_t block = get_bio_block(tc, bio);
2734 struct dm_thin_device *td = tc->td;
2735 struct dm_thin_lookup_result result;
2736 struct dm_bio_prison_cell *virt_cell, *data_cell;
2737 struct dm_cell_key key;
2739 thin_hook_bio(tc, bio);
2741 if (tc->requeue_mode) {
2742 bio->bi_status = BLK_STS_DM_REQUEUE;
2744 return DM_MAPIO_SUBMITTED;
2747 if (get_pool_mode(tc->pool) == PM_FAIL) {
2749 return DM_MAPIO_SUBMITTED;
2752 if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD) {
2753 thin_defer_bio_with_throttle(tc, bio);
2754 return DM_MAPIO_SUBMITTED;
2758 * We must hold the virtual cell before doing the lookup, otherwise
2759 * there's a race with discard.
2761 build_virtual_key(tc->td, block, &key);
2762 if (bio_detain(tc->pool, &key, bio, &virt_cell))
2763 return DM_MAPIO_SUBMITTED;
2765 r = dm_thin_find_block(td, block, 0, &result);
2768 * Note that we defer readahead too.
2772 if (unlikely(result.shared)) {
2774 * We have a race condition here between the
2775 * result.shared value returned by the lookup and
2776 * snapshot creation, which may cause new
2779 * To avoid this always quiesce the origin before
2780 * taking the snap. You want to do this anyway to
2781 * ensure a consistent application view
2784 * More distant ancestors are irrelevant. The
2785 * shared flag will be set in their case.
2787 thin_defer_cell(tc, virt_cell);
2788 return DM_MAPIO_SUBMITTED;
2791 build_data_key(tc->td, result.block, &key);
2792 if (bio_detain(tc->pool, &key, bio, &data_cell)) {
2793 cell_defer_no_holder(tc, virt_cell);
2794 return DM_MAPIO_SUBMITTED;
2797 inc_all_io_entry(tc->pool, bio);
2798 cell_defer_no_holder(tc, data_cell);
2799 cell_defer_no_holder(tc, virt_cell);
2801 remap(tc, bio, result.block);
2802 return DM_MAPIO_REMAPPED;
2806 thin_defer_cell(tc, virt_cell);
2807 return DM_MAPIO_SUBMITTED;
2811 * Must always call bio_io_error on failure.
2812 * dm_thin_find_block can fail with -EINVAL if the
2813 * pool is switched to fail-io mode.
2816 cell_defer_no_holder(tc, virt_cell);
2817 return DM_MAPIO_SUBMITTED;
2821 static void requeue_bios(struct pool *pool)
2826 list_for_each_entry_rcu(tc, &pool->active_thins, list) {
2827 spin_lock_irq(&tc->lock);
2828 bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list);
2829 bio_list_init(&tc->retry_on_resume_list);
2830 spin_unlock_irq(&tc->lock);
2836 *--------------------------------------------------------------
2837 * Binding of control targets to a pool object
2838 *--------------------------------------------------------------
2840 static bool is_factor(sector_t block_size, uint32_t n)
2842 return !sector_div(block_size, n);
2846 * If discard_passdown was enabled verify that the data device
2847 * supports discards. Disable discard_passdown if not.
2849 static void disable_passdown_if_not_supported(struct pool_c *pt)
2851 struct pool *pool = pt->pool;
2852 struct block_device *data_bdev = pt->data_dev->bdev;
2853 struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
2854 const char *reason = NULL;
2856 if (!pt->adjusted_pf.discard_passdown)
2859 if (!bdev_max_discard_sectors(pt->data_dev->bdev))
2860 reason = "discard unsupported";
2862 else if (data_limits->max_discard_sectors < pool->sectors_per_block)
2863 reason = "max discard sectors smaller than a block";
2866 DMWARN("Data device (%pg) %s: Disabling discard passdown.", data_bdev, reason);
2867 pt->adjusted_pf.discard_passdown = false;
2871 static int bind_control_target(struct pool *pool, struct dm_target *ti)
2873 struct pool_c *pt = ti->private;
2876 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
2878 enum pool_mode old_mode = get_pool_mode(pool);
2879 enum pool_mode new_mode = pt->adjusted_pf.mode;
2882 * Don't change the pool's mode until set_pool_mode() below.
2883 * Otherwise the pool's process_* function pointers may
2884 * not match the desired pool mode.
2886 pt->adjusted_pf.mode = old_mode;
2889 pool->pf = pt->adjusted_pf;
2890 pool->low_water_blocks = pt->low_water_blocks;
2892 set_pool_mode(pool, new_mode);
2897 static void unbind_control_target(struct pool *pool, struct dm_target *ti)
2904 *--------------------------------------------------------------
2906 *--------------------------------------------------------------
2908 /* Initialize pool features. */
2909 static void pool_features_init(struct pool_features *pf)
2911 pf->mode = PM_WRITE;
2912 pf->zero_new_blocks = true;
2913 pf->discard_enabled = true;
2914 pf->discard_passdown = true;
2915 pf->error_if_no_space = false;
2918 static void __pool_destroy(struct pool *pool)
2920 __pool_table_remove(pool);
2922 vfree(pool->cell_sort_array);
2923 if (dm_pool_metadata_close(pool->pmd) < 0)
2924 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
2926 dm_bio_prison_destroy(pool->prison);
2927 dm_kcopyd_client_destroy(pool->copier);
2929 cancel_delayed_work_sync(&pool->waker);
2930 cancel_delayed_work_sync(&pool->no_space_timeout);
2932 destroy_workqueue(pool->wq);
2934 if (pool->next_mapping)
2935 mempool_free(pool->next_mapping, &pool->mapping_pool);
2936 mempool_exit(&pool->mapping_pool);
2937 dm_deferred_set_destroy(pool->shared_read_ds);
2938 dm_deferred_set_destroy(pool->all_io_ds);
2942 static struct kmem_cache *_new_mapping_cache;
2944 static struct pool *pool_create(struct mapped_device *pool_md,
2945 struct block_device *metadata_dev,
2946 struct block_device *data_dev,
2947 unsigned long block_size,
2948 int read_only, char **error)
2953 struct dm_pool_metadata *pmd;
2954 bool format_device = read_only ? false : true;
2956 pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
2958 *error = "Error creating metadata object";
2959 return (struct pool *)pmd;
2962 pool = kzalloc(sizeof(*pool), GFP_KERNEL);
2964 *error = "Error allocating memory for pool";
2965 err_p = ERR_PTR(-ENOMEM);
2970 pool->sectors_per_block = block_size;
2971 if (block_size & (block_size - 1))
2972 pool->sectors_per_block_shift = -1;
2974 pool->sectors_per_block_shift = __ffs(block_size);
2975 pool->low_water_blocks = 0;
2976 pool_features_init(&pool->pf);
2977 pool->prison = dm_bio_prison_create();
2978 if (!pool->prison) {
2979 *error = "Error creating pool's bio prison";
2980 err_p = ERR_PTR(-ENOMEM);
2984 pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2985 if (IS_ERR(pool->copier)) {
2986 r = PTR_ERR(pool->copier);
2987 *error = "Error creating pool's kcopyd client";
2989 goto bad_kcopyd_client;
2993 * Create singlethreaded workqueue that will service all devices
2994 * that use this metadata.
2996 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
2998 *error = "Error creating pool's workqueue";
2999 err_p = ERR_PTR(-ENOMEM);
3003 throttle_init(&pool->throttle);
3004 INIT_WORK(&pool->worker, do_worker);
3005 INIT_DELAYED_WORK(&pool->waker, do_waker);
3006 INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout);
3007 spin_lock_init(&pool->lock);
3008 bio_list_init(&pool->deferred_flush_bios);
3009 bio_list_init(&pool->deferred_flush_completions);
3010 INIT_LIST_HEAD(&pool->prepared_mappings);
3011 INIT_LIST_HEAD(&pool->prepared_discards);
3012 INIT_LIST_HEAD(&pool->prepared_discards_pt2);
3013 INIT_LIST_HEAD(&pool->active_thins);
3014 pool->low_water_triggered = false;
3015 pool->suspended = true;
3016 pool->out_of_data_space = false;
3018 pool->shared_read_ds = dm_deferred_set_create();
3019 if (!pool->shared_read_ds) {
3020 *error = "Error creating pool's shared read deferred set";
3021 err_p = ERR_PTR(-ENOMEM);
3022 goto bad_shared_read_ds;
3025 pool->all_io_ds = dm_deferred_set_create();
3026 if (!pool->all_io_ds) {
3027 *error = "Error creating pool's all io deferred set";
3028 err_p = ERR_PTR(-ENOMEM);
3032 pool->next_mapping = NULL;
3033 r = mempool_init_slab_pool(&pool->mapping_pool, MAPPING_POOL_SIZE,
3034 _new_mapping_cache);
3036 *error = "Error creating pool's mapping mempool";
3038 goto bad_mapping_pool;
3041 pool->cell_sort_array =
3042 vmalloc(array_size(CELL_SORT_ARRAY_SIZE,
3043 sizeof(*pool->cell_sort_array)));
3044 if (!pool->cell_sort_array) {
3045 *error = "Error allocating cell sort array";
3046 err_p = ERR_PTR(-ENOMEM);
3047 goto bad_sort_array;
3050 pool->ref_count = 1;
3051 pool->last_commit_jiffies = jiffies;
3052 pool->pool_md = pool_md;
3053 pool->md_dev = metadata_dev;
3054 pool->data_dev = data_dev;
3055 __pool_table_insert(pool);
3060 mempool_exit(&pool->mapping_pool);
3062 dm_deferred_set_destroy(pool->all_io_ds);
3064 dm_deferred_set_destroy(pool->shared_read_ds);
3066 destroy_workqueue(pool->wq);
3068 dm_kcopyd_client_destroy(pool->copier);
3070 dm_bio_prison_destroy(pool->prison);
3074 if (dm_pool_metadata_close(pmd))
3075 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
3080 static void __pool_inc(struct pool *pool)
3082 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
3086 static void __pool_dec(struct pool *pool)
3088 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
3089 BUG_ON(!pool->ref_count);
3090 if (!--pool->ref_count)
3091 __pool_destroy(pool);
3094 static struct pool *__pool_find(struct mapped_device *pool_md,
3095 struct block_device *metadata_dev,
3096 struct block_device *data_dev,
3097 unsigned long block_size, int read_only,
3098 char **error, int *created)
3100 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
3103 if (pool->pool_md != pool_md) {
3104 *error = "metadata device already in use by a pool";
3105 return ERR_PTR(-EBUSY);
3107 if (pool->data_dev != data_dev) {
3108 *error = "data device already in use by a pool";
3109 return ERR_PTR(-EBUSY);
3114 pool = __pool_table_lookup(pool_md);
3116 if (pool->md_dev != metadata_dev || pool->data_dev != data_dev) {
3117 *error = "different pool cannot replace a pool";
3118 return ERR_PTR(-EINVAL);
3123 pool = pool_create(pool_md, metadata_dev, data_dev, block_size, read_only, error);
3132 *--------------------------------------------------------------
3133 * Pool target methods
3134 *--------------------------------------------------------------
3136 static void pool_dtr(struct dm_target *ti)
3138 struct pool_c *pt = ti->private;
3140 mutex_lock(&dm_thin_pool_table.mutex);
3142 unbind_control_target(pt->pool, ti);
3143 __pool_dec(pt->pool);
3144 dm_put_device(ti, pt->metadata_dev);
3145 dm_put_device(ti, pt->data_dev);
3148 mutex_unlock(&dm_thin_pool_table.mutex);
3151 static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
3152 struct dm_target *ti)
3156 const char *arg_name;
3158 static const struct dm_arg _args[] = {
3159 {0, 4, "Invalid number of pool feature arguments"},
3163 * No feature arguments supplied.
3168 r = dm_read_arg_group(_args, as, &argc, &ti->error);
3172 while (argc && !r) {
3173 arg_name = dm_shift_arg(as);
3176 if (!strcasecmp(arg_name, "skip_block_zeroing"))
3177 pf->zero_new_blocks = false;
3179 else if (!strcasecmp(arg_name, "ignore_discard"))
3180 pf->discard_enabled = false;
3182 else if (!strcasecmp(arg_name, "no_discard_passdown"))
3183 pf->discard_passdown = false;
3185 else if (!strcasecmp(arg_name, "read_only"))
3186 pf->mode = PM_READ_ONLY;
3188 else if (!strcasecmp(arg_name, "error_if_no_space"))
3189 pf->error_if_no_space = true;
3192 ti->error = "Unrecognised pool feature requested";
3201 static void metadata_low_callback(void *context)
3203 struct pool *pool = context;
3205 DMWARN("%s: reached low water mark for metadata device: sending event.",
3206 dm_device_name(pool->pool_md));
3208 dm_table_event(pool->ti->table);
3212 * We need to flush the data device **before** committing the metadata.
3214 * This ensures that the data blocks of any newly inserted mappings are
3215 * properly written to non-volatile storage and won't be lost in case of a
3218 * Failure to do so can result in data corruption in the case of internal or
3219 * external snapshots and in the case of newly provisioned blocks, when block
3220 * zeroing is enabled.
3222 static int metadata_pre_commit_callback(void *context)
3224 struct pool *pool = context;
3226 return blkdev_issue_flush(pool->data_dev);
3229 static sector_t get_dev_size(struct block_device *bdev)
3231 return bdev_nr_sectors(bdev);
3234 static void warn_if_metadata_device_too_big(struct block_device *bdev)
3236 sector_t metadata_dev_size = get_dev_size(bdev);
3238 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING)
3239 DMWARN("Metadata device %pg is larger than %u sectors: excess space will not be used.",
3240 bdev, THIN_METADATA_MAX_SECTORS);
3243 static sector_t get_metadata_dev_size(struct block_device *bdev)
3245 sector_t metadata_dev_size = get_dev_size(bdev);
3247 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS)
3248 metadata_dev_size = THIN_METADATA_MAX_SECTORS;
3250 return metadata_dev_size;
3253 static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
3255 sector_t metadata_dev_size = get_metadata_dev_size(bdev);
3257 sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE);
3259 return metadata_dev_size;
3263 * When a metadata threshold is crossed a dm event is triggered, and
3264 * userland should respond by growing the metadata device. We could let
3265 * userland set the threshold, like we do with the data threshold, but I'm
3266 * not sure they know enough to do this well.
3268 static dm_block_t calc_metadata_threshold(struct pool_c *pt)
3271 * 4M is ample for all ops with the possible exception of thin
3272 * device deletion which is harmless if it fails (just retry the
3273 * delete after you've grown the device).
3275 dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
3277 return min((dm_block_t)1024ULL /* 4M */, quarter);
3281 * thin-pool <metadata dev> <data dev>
3282 * <data block size (sectors)>
3283 * <low water mark (blocks)>
3284 * [<#feature args> [<arg>]*]
3286 * Optional feature arguments are:
3287 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
3288 * ignore_discard: disable discard
3289 * no_discard_passdown: don't pass discards down to the data device
3290 * read_only: Don't allow any changes to be made to the pool metadata.
3291 * error_if_no_space: error IOs, instead of queueing, if no space.
3293 static int pool_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3295 int r, pool_created = 0;
3298 struct pool_features pf;
3299 struct dm_arg_set as;
3300 struct dm_dev *data_dev;
3301 unsigned long block_size;
3302 dm_block_t low_water_blocks;
3303 struct dm_dev *metadata_dev;
3304 fmode_t metadata_mode;
3307 * FIXME Remove validation from scope of lock.
3309 mutex_lock(&dm_thin_pool_table.mutex);
3312 ti->error = "Invalid argument count";
3320 /* make sure metadata and data are different devices */
3321 if (!strcmp(argv[0], argv[1])) {
3322 ti->error = "Error setting metadata or data device";
3328 * Set default pool features.
3330 pool_features_init(&pf);
3332 dm_consume_args(&as, 4);
3333 r = parse_pool_features(&as, &pf, ti);
3337 metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
3338 r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
3340 ti->error = "Error opening metadata block device";
3343 warn_if_metadata_device_too_big(metadata_dev->bdev);
3345 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
3347 ti->error = "Error getting data device";
3351 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
3352 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
3353 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
3354 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
3355 ti->error = "Invalid block size";
3360 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
3361 ti->error = "Invalid low water mark";
3366 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
3372 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, data_dev->bdev,
3373 block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
3380 * 'pool_created' reflects whether this is the first table load.
3381 * Top level discard support is not allowed to be changed after
3382 * initial load. This would require a pool reload to trigger thin
3385 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
3386 ti->error = "Discard support cannot be disabled once enabled";
3388 goto out_flags_changed;
3393 pt->metadata_dev = metadata_dev;
3394 pt->data_dev = data_dev;
3395 pt->low_water_blocks = low_water_blocks;
3396 pt->adjusted_pf = pt->requested_pf = pf;
3397 ti->num_flush_bios = 1;
3398 ti->limit_swap_bios = true;
3401 * Only need to enable discards if the pool should pass
3402 * them down to the data device. The thin device's discard
3403 * processing will cause mappings to be removed from the btree.
3405 if (pf.discard_enabled && pf.discard_passdown) {
3406 ti->num_discard_bios = 1;
3408 * Setting 'discards_supported' circumvents the normal
3409 * stacking of discard limits (this keeps the pool and
3410 * thin devices' discard limits consistent).
3412 ti->discards_supported = true;
3413 ti->max_discard_granularity = true;
3417 r = dm_pool_register_metadata_threshold(pt->pool->pmd,
3418 calc_metadata_threshold(pt),
3419 metadata_low_callback,
3422 ti->error = "Error registering metadata threshold";
3423 goto out_flags_changed;
3426 dm_pool_register_pre_commit_callback(pool->pmd,
3427 metadata_pre_commit_callback, pool);
3429 mutex_unlock(&dm_thin_pool_table.mutex);
3438 dm_put_device(ti, data_dev);
3440 dm_put_device(ti, metadata_dev);
3442 mutex_unlock(&dm_thin_pool_table.mutex);
3447 static int pool_map(struct dm_target *ti, struct bio *bio)
3450 struct pool_c *pt = ti->private;
3451 struct pool *pool = pt->pool;
3454 * As this is a singleton target, ti->begin is always zero.
3456 spin_lock_irq(&pool->lock);
3457 bio_set_dev(bio, pt->data_dev->bdev);
3458 r = DM_MAPIO_REMAPPED;
3459 spin_unlock_irq(&pool->lock);
3464 static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
3467 struct pool_c *pt = ti->private;
3468 struct pool *pool = pt->pool;
3469 sector_t data_size = ti->len;
3470 dm_block_t sb_data_size;
3472 *need_commit = false;
3474 (void) sector_div(data_size, pool->sectors_per_block);
3476 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
3478 DMERR("%s: failed to retrieve data device size",
3479 dm_device_name(pool->pool_md));
3483 if (data_size < sb_data_size) {
3484 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
3485 dm_device_name(pool->pool_md),
3486 (unsigned long long)data_size, sb_data_size);
3489 } else if (data_size > sb_data_size) {
3490 if (dm_pool_metadata_needs_check(pool->pmd)) {
3491 DMERR("%s: unable to grow the data device until repaired.",
3492 dm_device_name(pool->pool_md));
3497 DMINFO("%s: growing the data device from %llu to %llu blocks",
3498 dm_device_name(pool->pool_md),
3499 sb_data_size, (unsigned long long)data_size);
3500 r = dm_pool_resize_data_dev(pool->pmd, data_size);
3502 metadata_operation_failed(pool, "dm_pool_resize_data_dev", r);
3506 *need_commit = true;
3512 static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
3515 struct pool_c *pt = ti->private;
3516 struct pool *pool = pt->pool;
3517 dm_block_t metadata_dev_size, sb_metadata_dev_size;
3519 *need_commit = false;
3521 metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
3523 r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
3525 DMERR("%s: failed to retrieve metadata device size",
3526 dm_device_name(pool->pool_md));
3530 if (metadata_dev_size < sb_metadata_dev_size) {
3531 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
3532 dm_device_name(pool->pool_md),
3533 metadata_dev_size, sb_metadata_dev_size);
3536 } else if (metadata_dev_size > sb_metadata_dev_size) {
3537 if (dm_pool_metadata_needs_check(pool->pmd)) {
3538 DMERR("%s: unable to grow the metadata device until repaired.",
3539 dm_device_name(pool->pool_md));
3543 warn_if_metadata_device_too_big(pool->md_dev);
3544 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
3545 dm_device_name(pool->pool_md),
3546 sb_metadata_dev_size, metadata_dev_size);
3548 if (get_pool_mode(pool) == PM_OUT_OF_METADATA_SPACE)
3549 set_pool_mode(pool, PM_WRITE);
3551 r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
3553 metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r);
3557 *need_commit = true;
3564 * Retrieves the number of blocks of the data device from
3565 * the superblock and compares it to the actual device size,
3566 * thus resizing the data device in case it has grown.
3568 * This both copes with opening preallocated data devices in the ctr
3569 * being followed by a resume
3571 * calling the resume method individually after userspace has
3572 * grown the data device in reaction to a table event.
3574 static int pool_preresume(struct dm_target *ti)
3577 bool need_commit1, need_commit2;
3578 struct pool_c *pt = ti->private;
3579 struct pool *pool = pt->pool;
3582 * Take control of the pool object.
3584 r = bind_control_target(pool, ti);
3588 r = maybe_resize_data_dev(ti, &need_commit1);
3592 r = maybe_resize_metadata_dev(ti, &need_commit2);
3596 if (need_commit1 || need_commit2)
3597 (void) commit(pool);
3600 * When a thin-pool is PM_FAIL, it cannot be rebuilt if
3601 * bio is in deferred list. Therefore need to return 0
3602 * to allow pool_resume() to flush IO.
3604 if (r && get_pool_mode(pool) == PM_FAIL)
3610 static void pool_suspend_active_thins(struct pool *pool)
3614 /* Suspend all active thin devices */
3615 tc = get_first_thin(pool);
3617 dm_internal_suspend_noflush(tc->thin_md);
3618 tc = get_next_thin(pool, tc);
3622 static void pool_resume_active_thins(struct pool *pool)
3626 /* Resume all active thin devices */
3627 tc = get_first_thin(pool);
3629 dm_internal_resume(tc->thin_md);
3630 tc = get_next_thin(pool, tc);
3634 static void pool_resume(struct dm_target *ti)
3636 struct pool_c *pt = ti->private;
3637 struct pool *pool = pt->pool;
3640 * Must requeue active_thins' bios and then resume
3641 * active_thins _before_ clearing 'suspend' flag.
3644 pool_resume_active_thins(pool);
3646 spin_lock_irq(&pool->lock);
3647 pool->low_water_triggered = false;
3648 pool->suspended = false;
3649 spin_unlock_irq(&pool->lock);
3651 do_waker(&pool->waker.work);
3654 static void pool_presuspend(struct dm_target *ti)
3656 struct pool_c *pt = ti->private;
3657 struct pool *pool = pt->pool;
3659 spin_lock_irq(&pool->lock);
3660 pool->suspended = true;
3661 spin_unlock_irq(&pool->lock);
3663 pool_suspend_active_thins(pool);
3666 static void pool_presuspend_undo(struct dm_target *ti)
3668 struct pool_c *pt = ti->private;
3669 struct pool *pool = pt->pool;
3671 pool_resume_active_thins(pool);
3673 spin_lock_irq(&pool->lock);
3674 pool->suspended = false;
3675 spin_unlock_irq(&pool->lock);
3678 static void pool_postsuspend(struct dm_target *ti)
3680 struct pool_c *pt = ti->private;
3681 struct pool *pool = pt->pool;
3683 cancel_delayed_work_sync(&pool->waker);
3684 cancel_delayed_work_sync(&pool->no_space_timeout);
3685 flush_workqueue(pool->wq);
3686 (void) commit(pool);
3689 static int check_arg_count(unsigned int argc, unsigned int args_required)
3691 if (argc != args_required) {
3692 DMWARN("Message received with %u arguments instead of %u.",
3693 argc, args_required);
3700 static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
3702 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
3703 *dev_id <= MAX_DEV_ID)
3707 DMWARN("Message received with invalid device id: %s", arg);
3712 static int process_create_thin_mesg(unsigned int argc, char **argv, struct pool *pool)
3717 r = check_arg_count(argc, 2);
3721 r = read_dev_id(argv[1], &dev_id, 1);
3725 r = dm_pool_create_thin(pool->pmd, dev_id);
3727 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
3735 static int process_create_snap_mesg(unsigned int argc, char **argv, struct pool *pool)
3738 dm_thin_id origin_dev_id;
3741 r = check_arg_count(argc, 3);
3745 r = read_dev_id(argv[1], &dev_id, 1);
3749 r = read_dev_id(argv[2], &origin_dev_id, 1);
3753 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
3755 DMWARN("Creation of new snapshot %s of device %s failed.",
3763 static int process_delete_mesg(unsigned int argc, char **argv, struct pool *pool)
3768 r = check_arg_count(argc, 2);
3772 r = read_dev_id(argv[1], &dev_id, 1);
3776 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
3778 DMWARN("Deletion of thin device %s failed.", argv[1]);
3783 static int process_set_transaction_id_mesg(unsigned int argc, char **argv, struct pool *pool)
3785 dm_thin_id old_id, new_id;
3788 r = check_arg_count(argc, 3);
3792 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
3793 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
3797 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
3798 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
3802 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
3804 DMWARN("Failed to change transaction id from %s to %s.",
3812 static int process_reserve_metadata_snap_mesg(unsigned int argc, char **argv, struct pool *pool)
3816 r = check_arg_count(argc, 1);
3820 (void) commit(pool);
3822 r = dm_pool_reserve_metadata_snap(pool->pmd);
3824 DMWARN("reserve_metadata_snap message failed.");
3829 static int process_release_metadata_snap_mesg(unsigned int argc, char **argv, struct pool *pool)
3833 r = check_arg_count(argc, 1);
3837 r = dm_pool_release_metadata_snap(pool->pmd);
3839 DMWARN("release_metadata_snap message failed.");
3845 * Messages supported:
3846 * create_thin <dev_id>
3847 * create_snap <dev_id> <origin_id>
3849 * set_transaction_id <current_trans_id> <new_trans_id>
3850 * reserve_metadata_snap
3851 * release_metadata_snap
3853 static int pool_message(struct dm_target *ti, unsigned int argc, char **argv,
3854 char *result, unsigned int maxlen)
3857 struct pool_c *pt = ti->private;
3858 struct pool *pool = pt->pool;
3860 if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE) {
3861 DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode",
3862 dm_device_name(pool->pool_md));
3866 if (!strcasecmp(argv[0], "create_thin"))
3867 r = process_create_thin_mesg(argc, argv, pool);
3869 else if (!strcasecmp(argv[0], "create_snap"))
3870 r = process_create_snap_mesg(argc, argv, pool);
3872 else if (!strcasecmp(argv[0], "delete"))
3873 r = process_delete_mesg(argc, argv, pool);
3875 else if (!strcasecmp(argv[0], "set_transaction_id"))
3876 r = process_set_transaction_id_mesg(argc, argv, pool);
3878 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
3879 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
3881 else if (!strcasecmp(argv[0], "release_metadata_snap"))
3882 r = process_release_metadata_snap_mesg(argc, argv, pool);
3885 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
3888 (void) commit(pool);
3893 static void emit_flags(struct pool_features *pf, char *result,
3894 unsigned int sz, unsigned int maxlen)
3896 unsigned int count = !pf->zero_new_blocks + !pf->discard_enabled +
3897 !pf->discard_passdown + (pf->mode == PM_READ_ONLY) +
3898 pf->error_if_no_space;
3899 DMEMIT("%u ", count);
3901 if (!pf->zero_new_blocks)
3902 DMEMIT("skip_block_zeroing ");
3904 if (!pf->discard_enabled)
3905 DMEMIT("ignore_discard ");
3907 if (!pf->discard_passdown)
3908 DMEMIT("no_discard_passdown ");
3910 if (pf->mode == PM_READ_ONLY)
3911 DMEMIT("read_only ");
3913 if (pf->error_if_no_space)
3914 DMEMIT("error_if_no_space ");
3919 * <transaction id> <used metadata sectors>/<total metadata sectors>
3920 * <used data sectors>/<total data sectors> <held metadata root>
3921 * <pool mode> <discard config> <no space config> <needs_check>
3923 static void pool_status(struct dm_target *ti, status_type_t type,
3924 unsigned int status_flags, char *result, unsigned int maxlen)
3927 unsigned int sz = 0;
3928 uint64_t transaction_id;
3929 dm_block_t nr_free_blocks_data;
3930 dm_block_t nr_free_blocks_metadata;
3931 dm_block_t nr_blocks_data;
3932 dm_block_t nr_blocks_metadata;
3933 dm_block_t held_root;
3934 enum pool_mode mode;
3935 char buf[BDEVNAME_SIZE];
3936 char buf2[BDEVNAME_SIZE];
3937 struct pool_c *pt = ti->private;
3938 struct pool *pool = pt->pool;
3941 case STATUSTYPE_INFO:
3942 if (get_pool_mode(pool) == PM_FAIL) {
3947 /* Commit to ensure statistics aren't out-of-date */
3948 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
3949 (void) commit(pool);
3951 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
3953 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
3954 dm_device_name(pool->pool_md), r);
3958 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
3960 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
3961 dm_device_name(pool->pool_md), r);
3965 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
3967 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
3968 dm_device_name(pool->pool_md), r);
3972 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
3974 DMERR("%s: dm_pool_get_free_block_count returned %d",
3975 dm_device_name(pool->pool_md), r);
3979 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
3981 DMERR("%s: dm_pool_get_data_dev_size returned %d",
3982 dm_device_name(pool->pool_md), r);
3986 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
3988 DMERR("%s: dm_pool_get_metadata_snap returned %d",
3989 dm_device_name(pool->pool_md), r);
3993 DMEMIT("%llu %llu/%llu %llu/%llu ",
3994 (unsigned long long)transaction_id,
3995 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
3996 (unsigned long long)nr_blocks_metadata,
3997 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
3998 (unsigned long long)nr_blocks_data);
4001 DMEMIT("%llu ", held_root);
4005 mode = get_pool_mode(pool);
4006 if (mode == PM_OUT_OF_DATA_SPACE)
4007 DMEMIT("out_of_data_space ");
4008 else if (is_read_only_pool_mode(mode))
4013 if (!pool->pf.discard_enabled)
4014 DMEMIT("ignore_discard ");
4015 else if (pool->pf.discard_passdown)
4016 DMEMIT("discard_passdown ");
4018 DMEMIT("no_discard_passdown ");
4020 if (pool->pf.error_if_no_space)
4021 DMEMIT("error_if_no_space ");
4023 DMEMIT("queue_if_no_space ");
4025 if (dm_pool_metadata_needs_check(pool->pmd))
4026 DMEMIT("needs_check ");
4030 DMEMIT("%llu ", (unsigned long long)calc_metadata_threshold(pt));
4034 case STATUSTYPE_TABLE:
4035 DMEMIT("%s %s %lu %llu ",
4036 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
4037 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
4038 (unsigned long)pool->sectors_per_block,
4039 (unsigned long long)pt->low_water_blocks);
4040 emit_flags(&pt->requested_pf, result, sz, maxlen);
4043 case STATUSTYPE_IMA:
4053 static int pool_iterate_devices(struct dm_target *ti,
4054 iterate_devices_callout_fn fn, void *data)
4056 struct pool_c *pt = ti->private;
4058 return fn(ti, pt->data_dev, 0, ti->len, data);
4061 static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
4063 struct pool_c *pt = ti->private;
4064 struct pool *pool = pt->pool;
4065 sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
4068 * If max_sectors is smaller than pool->sectors_per_block adjust it
4069 * to the highest possible power-of-2 factor of pool->sectors_per_block.
4070 * This is especially beneficial when the pool's data device is a RAID
4071 * device that has a full stripe width that matches pool->sectors_per_block
4072 * -- because even though partial RAID stripe-sized IOs will be issued to a
4073 * single RAID stripe; when aggregated they will end on a full RAID stripe
4074 * boundary.. which avoids additional partial RAID stripe writes cascading
4076 if (limits->max_sectors < pool->sectors_per_block) {
4077 while (!is_factor(pool->sectors_per_block, limits->max_sectors)) {
4078 if ((limits->max_sectors & (limits->max_sectors - 1)) == 0)
4079 limits->max_sectors--;
4080 limits->max_sectors = rounddown_pow_of_two(limits->max_sectors);
4085 * If the system-determined stacked limits are compatible with the
4086 * pool's blocksize (io_opt is a factor) do not override them.
4088 if (io_opt_sectors < pool->sectors_per_block ||
4089 !is_factor(io_opt_sectors, pool->sectors_per_block)) {
4090 if (is_factor(pool->sectors_per_block, limits->max_sectors))
4091 blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT);
4093 blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT);
4094 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
4098 * pt->adjusted_pf is a staging area for the actual features to use.
4099 * They get transferred to the live pool in bind_control_target()
4100 * called from pool_preresume().
4102 if (!pt->adjusted_pf.discard_enabled) {
4104 * Must explicitly disallow stacking discard limits otherwise the
4105 * block layer will stack them if pool's data device has support.
4107 limits->discard_granularity = 0;
4111 disable_passdown_if_not_supported(pt);
4114 * The pool uses the same discard limits as the underlying data
4115 * device. DM core has already set this up.
4119 static struct target_type pool_target = {
4120 .name = "thin-pool",
4121 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
4122 DM_TARGET_IMMUTABLE,
4123 .version = {1, 23, 0},
4124 .module = THIS_MODULE,
4128 .presuspend = pool_presuspend,
4129 .presuspend_undo = pool_presuspend_undo,
4130 .postsuspend = pool_postsuspend,
4131 .preresume = pool_preresume,
4132 .resume = pool_resume,
4133 .message = pool_message,
4134 .status = pool_status,
4135 .iterate_devices = pool_iterate_devices,
4136 .io_hints = pool_io_hints,
4140 *--------------------------------------------------------------
4141 * Thin target methods
4142 *--------------------------------------------------------------
4144 static void thin_get(struct thin_c *tc)
4146 refcount_inc(&tc->refcount);
4149 static void thin_put(struct thin_c *tc)
4151 if (refcount_dec_and_test(&tc->refcount))
4152 complete(&tc->can_destroy);
4155 static void thin_dtr(struct dm_target *ti)
4157 struct thin_c *tc = ti->private;
4159 spin_lock_irq(&tc->pool->lock);
4160 list_del_rcu(&tc->list);
4161 spin_unlock_irq(&tc->pool->lock);
4165 wait_for_completion(&tc->can_destroy);
4167 mutex_lock(&dm_thin_pool_table.mutex);
4169 __pool_dec(tc->pool);
4170 dm_pool_close_thin_device(tc->td);
4171 dm_put_device(ti, tc->pool_dev);
4173 dm_put_device(ti, tc->origin_dev);
4176 mutex_unlock(&dm_thin_pool_table.mutex);
4180 * Thin target parameters:
4182 * <pool_dev> <dev_id> [origin_dev]
4184 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
4185 * dev_id: the internal device identifier
4186 * origin_dev: a device external to the pool that should act as the origin
4188 * If the pool device has discards disabled, they get disabled for the thin
4191 static int thin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
4195 struct dm_dev *pool_dev, *origin_dev;
4196 struct mapped_device *pool_md;
4198 mutex_lock(&dm_thin_pool_table.mutex);
4200 if (argc != 2 && argc != 3) {
4201 ti->error = "Invalid argument count";
4206 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
4208 ti->error = "Out of memory";
4212 tc->thin_md = dm_table_get_md(ti->table);
4213 spin_lock_init(&tc->lock);
4214 INIT_LIST_HEAD(&tc->deferred_cells);
4215 bio_list_init(&tc->deferred_bio_list);
4216 bio_list_init(&tc->retry_on_resume_list);
4217 tc->sort_bio_list = RB_ROOT;
4220 if (!strcmp(argv[0], argv[2])) {
4221 ti->error = "Error setting origin device";
4223 goto bad_origin_dev;
4226 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
4228 ti->error = "Error opening origin device";
4229 goto bad_origin_dev;
4231 tc->origin_dev = origin_dev;
4234 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
4236 ti->error = "Error opening pool device";
4239 tc->pool_dev = pool_dev;
4241 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
4242 ti->error = "Invalid device id";
4247 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
4249 ti->error = "Couldn't get pool mapped device";
4254 tc->pool = __pool_table_lookup(pool_md);
4256 ti->error = "Couldn't find pool object";
4258 goto bad_pool_lookup;
4260 __pool_inc(tc->pool);
4262 if (get_pool_mode(tc->pool) == PM_FAIL) {
4263 ti->error = "Couldn't open thin device, Pool is in fail mode";
4268 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
4270 ti->error = "Couldn't open thin internal device";
4274 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
4278 ti->num_flush_bios = 1;
4279 ti->limit_swap_bios = true;
4280 ti->flush_supported = true;
4281 ti->accounts_remapped_io = true;
4282 ti->per_io_data_size = sizeof(struct dm_thin_endio_hook);
4284 /* In case the pool supports discards, pass them on. */
4285 if (tc->pool->pf.discard_enabled) {
4286 ti->discards_supported = true;
4287 ti->num_discard_bios = 1;
4288 ti->max_discard_granularity = true;
4291 mutex_unlock(&dm_thin_pool_table.mutex);
4293 spin_lock_irq(&tc->pool->lock);
4294 if (tc->pool->suspended) {
4295 spin_unlock_irq(&tc->pool->lock);
4296 mutex_lock(&dm_thin_pool_table.mutex); /* reacquire for __pool_dec */
4297 ti->error = "Unable to activate thin device while pool is suspended";
4301 refcount_set(&tc->refcount, 1);
4302 init_completion(&tc->can_destroy);
4303 list_add_tail_rcu(&tc->list, &tc->pool->active_thins);
4304 spin_unlock_irq(&tc->pool->lock);
4306 * This synchronize_rcu() call is needed here otherwise we risk a
4307 * wake_worker() call finding no bios to process (because the newly
4308 * added tc isn't yet visible). So this reduces latency since we
4309 * aren't then dependent on the periodic commit to wake_worker().
4318 dm_pool_close_thin_device(tc->td);
4320 __pool_dec(tc->pool);
4324 dm_put_device(ti, tc->pool_dev);
4327 dm_put_device(ti, tc->origin_dev);
4331 mutex_unlock(&dm_thin_pool_table.mutex);
4336 static int thin_map(struct dm_target *ti, struct bio *bio)
4338 bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
4340 return thin_bio_map(ti, bio);
4343 static int thin_endio(struct dm_target *ti, struct bio *bio,
4346 unsigned long flags;
4347 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
4348 struct list_head work;
4349 struct dm_thin_new_mapping *m, *tmp;
4350 struct pool *pool = h->tc->pool;
4352 if (h->shared_read_entry) {
4353 INIT_LIST_HEAD(&work);
4354 dm_deferred_entry_dec(h->shared_read_entry, &work);
4356 spin_lock_irqsave(&pool->lock, flags);
4357 list_for_each_entry_safe(m, tmp, &work, list) {
4359 __complete_mapping_preparation(m);
4361 spin_unlock_irqrestore(&pool->lock, flags);
4364 if (h->all_io_entry) {
4365 INIT_LIST_HEAD(&work);
4366 dm_deferred_entry_dec(h->all_io_entry, &work);
4367 if (!list_empty(&work)) {
4368 spin_lock_irqsave(&pool->lock, flags);
4369 list_for_each_entry_safe(m, tmp, &work, list)
4370 list_add_tail(&m->list, &pool->prepared_discards);
4371 spin_unlock_irqrestore(&pool->lock, flags);
4377 cell_defer_no_holder(h->tc, h->cell);
4379 return DM_ENDIO_DONE;
4382 static void thin_presuspend(struct dm_target *ti)
4384 struct thin_c *tc = ti->private;
4386 if (dm_noflush_suspending(ti))
4387 noflush_work(tc, do_noflush_start);
4390 static void thin_postsuspend(struct dm_target *ti)
4392 struct thin_c *tc = ti->private;
4395 * The dm_noflush_suspending flag has been cleared by now, so
4396 * unfortunately we must always run this.
4398 noflush_work(tc, do_noflush_stop);
4401 static int thin_preresume(struct dm_target *ti)
4403 struct thin_c *tc = ti->private;
4406 tc->origin_size = get_dev_size(tc->origin_dev->bdev);
4412 * <nr mapped sectors> <highest mapped sector>
4414 static void thin_status(struct dm_target *ti, status_type_t type,
4415 unsigned int status_flags, char *result, unsigned int maxlen)
4419 dm_block_t mapped, highest;
4420 char buf[BDEVNAME_SIZE];
4421 struct thin_c *tc = ti->private;
4423 if (get_pool_mode(tc->pool) == PM_FAIL) {
4432 case STATUSTYPE_INFO:
4433 r = dm_thin_get_mapped_count(tc->td, &mapped);
4435 DMERR("dm_thin_get_mapped_count returned %d", r);
4439 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
4441 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
4445 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
4447 DMEMIT("%llu", ((highest + 1) *
4448 tc->pool->sectors_per_block) - 1);
4453 case STATUSTYPE_TABLE:
4455 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
4456 (unsigned long) tc->dev_id);
4458 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
4461 case STATUSTYPE_IMA:
4473 static int thin_iterate_devices(struct dm_target *ti,
4474 iterate_devices_callout_fn fn, void *data)
4477 struct thin_c *tc = ti->private;
4478 struct pool *pool = tc->pool;
4481 * We can't call dm_pool_get_data_dev_size() since that blocks. So
4482 * we follow a more convoluted path through to the pool's target.
4485 return 0; /* nothing is bound */
4487 blocks = pool->ti->len;
4488 (void) sector_div(blocks, pool->sectors_per_block);
4490 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
4495 static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits)
4497 struct thin_c *tc = ti->private;
4498 struct pool *pool = tc->pool;
4500 if (!pool->pf.discard_enabled)
4503 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
4504 limits->max_discard_sectors = pool->sectors_per_block * BIO_PRISON_MAX_RANGE;
4507 static struct target_type thin_target = {
4509 .version = {1, 23, 0},
4510 .module = THIS_MODULE,
4514 .end_io = thin_endio,
4515 .preresume = thin_preresume,
4516 .presuspend = thin_presuspend,
4517 .postsuspend = thin_postsuspend,
4518 .status = thin_status,
4519 .iterate_devices = thin_iterate_devices,
4520 .io_hints = thin_io_hints,
4523 /*----------------------------------------------------------------*/
4525 static int __init dm_thin_init(void)
4531 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
4532 if (!_new_mapping_cache)
4535 r = dm_register_target(&thin_target);
4537 goto bad_new_mapping_cache;
4539 r = dm_register_target(&pool_target);
4541 goto bad_thin_target;
4546 dm_unregister_target(&thin_target);
4547 bad_new_mapping_cache:
4548 kmem_cache_destroy(_new_mapping_cache);
4553 static void dm_thin_exit(void)
4555 dm_unregister_target(&thin_target);
4556 dm_unregister_target(&pool_target);
4558 kmem_cache_destroy(_new_mapping_cache);
4563 module_init(dm_thin_init);
4564 module_exit(dm_thin_exit);
4566 module_param_named(no_space_timeout, no_space_timeout_secs, uint, 0644);
4567 MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds");
4569 MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
4570 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
4571 MODULE_LICENSE("GPL");