2 * Copyright (C) 2011-2012 Red Hat UK.
4 * This file is released under the GPL.
7 #include "dm-thin-metadata.h"
8 #include "dm-bio-prison-v1.h"
11 #include <linux/device-mapper.h>
12 #include <linux/dm-io.h>
13 #include <linux/dm-kcopyd.h>
14 #include <linux/jiffies.h>
15 #include <linux/log2.h>
16 #include <linux/list.h>
17 #include <linux/rculist.h>
18 #include <linux/init.h>
19 #include <linux/module.h>
20 #include <linux/slab.h>
21 #include <linux/vmalloc.h>
22 #include <linux/sort.h>
23 #include <linux/rbtree.h>
25 #define DM_MSG_PREFIX "thin"
30 #define ENDIO_HOOK_POOL_SIZE 1024
31 #define MAPPING_POOL_SIZE 1024
32 #define COMMIT_PERIOD HZ
33 #define NO_SPACE_TIMEOUT_SECS 60
35 static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS;
37 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
38 "A percentage of time allocated for copy on write");
41 * The block size of the device holding pool data must be
42 * between 64KB and 1GB.
44 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
45 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
48 * Device id is restricted to 24 bits.
50 #define MAX_DEV_ID ((1 << 24) - 1)
53 * How do we handle breaking sharing of data blocks?
54 * =================================================
56 * We use a standard copy-on-write btree to store the mappings for the
57 * devices (note I'm talking about copy-on-write of the metadata here, not
58 * the data). When you take an internal snapshot you clone the root node
59 * of the origin btree. After this there is no concept of an origin or a
60 * snapshot. They are just two device trees that happen to point to the
63 * When we get a write in we decide if it's to a shared data block using
64 * some timestamp magic. If it is, we have to break sharing.
66 * Let's say we write to a shared block in what was the origin. The
69 * i) plug io further to this physical block. (see bio_prison code).
71 * ii) quiesce any read io to that shared data block. Obviously
72 * including all devices that share this block. (see dm_deferred_set code)
74 * iii) copy the data block to a newly allocate block. This step can be
75 * missed out if the io covers the block. (schedule_copy).
77 * iv) insert the new mapping into the origin's btree
78 * (process_prepared_mapping). This act of inserting breaks some
79 * sharing of btree nodes between the two devices. Breaking sharing only
80 * effects the btree of that specific device. Btrees for the other
81 * devices that share the block never change. The btree for the origin
82 * device as it was after the last commit is untouched, ie. we're using
83 * persistent data structures in the functional programming sense.
85 * v) unplug io to this physical block, including the io that triggered
86 * the breaking of sharing.
88 * Steps (ii) and (iii) occur in parallel.
90 * The metadata _doesn't_ need to be committed before the io continues. We
91 * get away with this because the io is always written to a _new_ block.
92 * If there's a crash, then:
94 * - The origin mapping will point to the old origin block (the shared
95 * one). This will contain the data as it was before the io that triggered
96 * the breaking of sharing came in.
98 * - The snap mapping still points to the old block. As it would after
101 * The downside of this scheme is the timestamp magic isn't perfect, and
102 * will continue to think that data block in the snapshot device is shared
103 * even after the write to the origin has broken sharing. I suspect data
104 * blocks will typically be shared by many different devices, so we're
105 * breaking sharing n + 1 times, rather than n, where n is the number of
106 * devices that reference this data block. At the moment I think the
107 * benefits far, far outweigh the disadvantages.
110 /*----------------------------------------------------------------*/
120 static void build_key(struct dm_thin_device *td, enum lock_space ls,
121 dm_block_t b, dm_block_t e, struct dm_cell_key *key)
123 key->virtual = (ls == VIRTUAL);
124 key->dev = dm_thin_dev_id(td);
125 key->block_begin = b;
129 static void build_data_key(struct dm_thin_device *td, dm_block_t b,
130 struct dm_cell_key *key)
132 build_key(td, PHYSICAL, b, b + 1llu, key);
135 static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
136 struct dm_cell_key *key)
138 build_key(td, VIRTUAL, b, b + 1llu, key);
141 /*----------------------------------------------------------------*/
143 #define THROTTLE_THRESHOLD (1 * HZ)
146 struct rw_semaphore lock;
147 unsigned long threshold;
148 bool throttle_applied;
151 static void throttle_init(struct throttle *t)
153 init_rwsem(&t->lock);
154 t->throttle_applied = false;
157 static void throttle_work_start(struct throttle *t)
159 t->threshold = jiffies + THROTTLE_THRESHOLD;
162 static void throttle_work_update(struct throttle *t)
164 if (!t->throttle_applied && jiffies > t->threshold) {
165 down_write(&t->lock);
166 t->throttle_applied = true;
170 static void throttle_work_complete(struct throttle *t)
172 if (t->throttle_applied) {
173 t->throttle_applied = false;
178 static void throttle_lock(struct throttle *t)
183 static void throttle_unlock(struct throttle *t)
188 /*----------------------------------------------------------------*/
191 * A pool device ties together a metadata device and a data device. It
192 * also provides the interface for creating and destroying internal
195 struct dm_thin_new_mapping;
198 * The pool runs in various modes. Ordered in degraded order for comparisons.
201 PM_WRITE, /* metadata may be changed */
202 PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */
205 * Like READ_ONLY, except may switch back to WRITE on metadata resize. Reported as READ_ONLY.
207 PM_OUT_OF_METADATA_SPACE,
208 PM_READ_ONLY, /* metadata may not be changed */
210 PM_FAIL, /* all I/O fails */
213 struct pool_features {
216 bool zero_new_blocks:1;
217 bool discard_enabled:1;
218 bool discard_passdown:1;
219 bool error_if_no_space:1;
223 typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
224 typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell);
225 typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);
227 #define CELL_SORT_ARRAY_SIZE 8192
230 struct list_head list;
231 struct dm_target *ti; /* Only set if a pool target is bound */
233 struct mapped_device *pool_md;
234 struct block_device *md_dev;
235 struct dm_pool_metadata *pmd;
237 dm_block_t low_water_blocks;
238 uint32_t sectors_per_block;
239 int sectors_per_block_shift;
241 struct pool_features pf;
242 bool low_water_triggered:1; /* A dm event has been sent */
244 bool out_of_data_space:1;
246 struct dm_bio_prison *prison;
247 struct dm_kcopyd_client *copier;
249 struct workqueue_struct *wq;
250 struct throttle throttle;
251 struct work_struct worker;
252 struct delayed_work waker;
253 struct delayed_work no_space_timeout;
255 unsigned long last_commit_jiffies;
259 struct bio_list deferred_flush_bios;
260 struct bio_list deferred_flush_completions;
261 struct list_head prepared_mappings;
262 struct list_head prepared_discards;
263 struct list_head prepared_discards_pt2;
264 struct list_head active_thins;
266 struct dm_deferred_set *shared_read_ds;
267 struct dm_deferred_set *all_io_ds;
269 struct dm_thin_new_mapping *next_mapping;
270 mempool_t *mapping_pool;
272 process_bio_fn process_bio;
273 process_bio_fn process_discard;
275 process_cell_fn process_cell;
276 process_cell_fn process_discard_cell;
278 process_mapping_fn process_prepared_mapping;
279 process_mapping_fn process_prepared_discard;
280 process_mapping_fn process_prepared_discard_pt2;
282 struct dm_bio_prison_cell **cell_sort_array;
285 static void metadata_operation_failed(struct pool *pool, const char *op, int r);
287 static enum pool_mode get_pool_mode(struct pool *pool)
289 return pool->pf.mode;
292 static void notify_of_pool_mode_change(struct pool *pool)
294 const char *descs[] = {
301 const char *extra_desc = NULL;
302 enum pool_mode mode = get_pool_mode(pool);
304 if (mode == PM_OUT_OF_DATA_SPACE) {
305 if (!pool->pf.error_if_no_space)
306 extra_desc = " (queue IO)";
308 extra_desc = " (error IO)";
311 dm_table_event(pool->ti->table);
312 DMINFO("%s: switching pool to %s%s mode",
313 dm_device_name(pool->pool_md),
314 descs[(int)mode], extra_desc ? : "");
318 * Target context for a pool.
321 struct dm_target *ti;
323 struct dm_dev *data_dev;
324 struct dm_dev *metadata_dev;
325 struct dm_target_callbacks callbacks;
327 dm_block_t low_water_blocks;
328 struct pool_features requested_pf; /* Features requested during table load */
329 struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */
333 * Target context for a thin.
336 struct list_head list;
337 struct dm_dev *pool_dev;
338 struct dm_dev *origin_dev;
339 sector_t origin_size;
343 struct dm_thin_device *td;
344 struct mapped_device *thin_md;
348 struct list_head deferred_cells;
349 struct bio_list deferred_bio_list;
350 struct bio_list retry_on_resume_list;
351 struct rb_root sort_bio_list; /* sorted list of deferred bios */
354 * Ensures the thin is not destroyed until the worker has finished
355 * iterating the active_thins list.
358 struct completion can_destroy;
361 /*----------------------------------------------------------------*/
363 static bool block_size_is_power_of_two(struct pool *pool)
365 return pool->sectors_per_block_shift >= 0;
368 static sector_t block_to_sectors(struct pool *pool, dm_block_t b)
370 return block_size_is_power_of_two(pool) ?
371 (b << pool->sectors_per_block_shift) :
372 (b * pool->sectors_per_block);
375 /*----------------------------------------------------------------*/
379 struct blk_plug plug;
380 struct bio *parent_bio;
384 static void begin_discard(struct discard_op *op, struct thin_c *tc, struct bio *parent)
389 blk_start_plug(&op->plug);
390 op->parent_bio = parent;
394 static int issue_discard(struct discard_op *op, dm_block_t data_b, dm_block_t data_e)
396 struct thin_c *tc = op->tc;
397 sector_t s = block_to_sectors(tc->pool, data_b);
398 sector_t len = block_to_sectors(tc->pool, data_e - data_b);
400 return __blkdev_issue_discard(tc->pool_dev->bdev, s, len,
401 GFP_NOWAIT, 0, &op->bio);
404 static void end_discard(struct discard_op *op, int r)
408 * Even if one of the calls to issue_discard failed, we
409 * need to wait for the chain to complete.
411 bio_chain(op->bio, op->parent_bio);
412 bio_set_op_attrs(op->bio, REQ_OP_DISCARD, 0);
416 blk_finish_plug(&op->plug);
419 * Even if r is set, there could be sub discards in flight that we
422 if (r && !op->parent_bio->bi_status)
423 op->parent_bio->bi_status = errno_to_blk_status(r);
424 bio_endio(op->parent_bio);
427 /*----------------------------------------------------------------*/
430 * wake_worker() is used when new work is queued and when pool_resume is
431 * ready to continue deferred IO processing.
433 static void wake_worker(struct pool *pool)
435 queue_work(pool->wq, &pool->worker);
438 /*----------------------------------------------------------------*/
440 static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio,
441 struct dm_bio_prison_cell **cell_result)
444 struct dm_bio_prison_cell *cell_prealloc;
447 * Allocate a cell from the prison's mempool.
448 * This might block but it can't fail.
450 cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO);
452 r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result);
455 * We reused an old cell; we can get rid of
458 dm_bio_prison_free_cell(pool->prison, cell_prealloc);
463 static void cell_release(struct pool *pool,
464 struct dm_bio_prison_cell *cell,
465 struct bio_list *bios)
467 dm_cell_release(pool->prison, cell, bios);
468 dm_bio_prison_free_cell(pool->prison, cell);
471 static void cell_visit_release(struct pool *pool,
472 void (*fn)(void *, struct dm_bio_prison_cell *),
474 struct dm_bio_prison_cell *cell)
476 dm_cell_visit_release(pool->prison, fn, context, cell);
477 dm_bio_prison_free_cell(pool->prison, cell);
480 static void cell_release_no_holder(struct pool *pool,
481 struct dm_bio_prison_cell *cell,
482 struct bio_list *bios)
484 dm_cell_release_no_holder(pool->prison, cell, bios);
485 dm_bio_prison_free_cell(pool->prison, cell);
488 static void cell_error_with_code(struct pool *pool,
489 struct dm_bio_prison_cell *cell, blk_status_t error_code)
491 dm_cell_error(pool->prison, cell, error_code);
492 dm_bio_prison_free_cell(pool->prison, cell);
495 static blk_status_t get_pool_io_error_code(struct pool *pool)
497 return pool->out_of_data_space ? BLK_STS_NOSPC : BLK_STS_IOERR;
500 static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell)
502 cell_error_with_code(pool, cell, get_pool_io_error_code(pool));
505 static void cell_success(struct pool *pool, struct dm_bio_prison_cell *cell)
507 cell_error_with_code(pool, cell, 0);
510 static void cell_requeue(struct pool *pool, struct dm_bio_prison_cell *cell)
512 cell_error_with_code(pool, cell, BLK_STS_DM_REQUEUE);
515 /*----------------------------------------------------------------*/
518 * A global list of pools that uses a struct mapped_device as a key.
520 static struct dm_thin_pool_table {
522 struct list_head pools;
523 } dm_thin_pool_table;
525 static void pool_table_init(void)
527 mutex_init(&dm_thin_pool_table.mutex);
528 INIT_LIST_HEAD(&dm_thin_pool_table.pools);
531 static void __pool_table_insert(struct pool *pool)
533 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
534 list_add(&pool->list, &dm_thin_pool_table.pools);
537 static void __pool_table_remove(struct pool *pool)
539 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
540 list_del(&pool->list);
543 static struct pool *__pool_table_lookup(struct mapped_device *md)
545 struct pool *pool = NULL, *tmp;
547 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
549 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
550 if (tmp->pool_md == md) {
559 static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
561 struct pool *pool = NULL, *tmp;
563 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
565 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
566 if (tmp->md_dev == md_dev) {
575 /*----------------------------------------------------------------*/
577 struct dm_thin_endio_hook {
579 struct dm_deferred_entry *shared_read_entry;
580 struct dm_deferred_entry *all_io_entry;
581 struct dm_thin_new_mapping *overwrite_mapping;
582 struct rb_node rb_node;
583 struct dm_bio_prison_cell *cell;
586 static void __merge_bio_list(struct bio_list *bios, struct bio_list *master)
588 bio_list_merge(bios, master);
589 bio_list_init(master);
592 static void error_bio_list(struct bio_list *bios, blk_status_t error)
596 while ((bio = bio_list_pop(bios))) {
597 bio->bi_status = error;
602 static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master,
605 struct bio_list bios;
608 bio_list_init(&bios);
610 spin_lock_irqsave(&tc->lock, flags);
611 __merge_bio_list(&bios, master);
612 spin_unlock_irqrestore(&tc->lock, flags);
614 error_bio_list(&bios, error);
617 static void requeue_deferred_cells(struct thin_c *tc)
619 struct pool *pool = tc->pool;
621 struct list_head cells;
622 struct dm_bio_prison_cell *cell, *tmp;
624 INIT_LIST_HEAD(&cells);
626 spin_lock_irqsave(&tc->lock, flags);
627 list_splice_init(&tc->deferred_cells, &cells);
628 spin_unlock_irqrestore(&tc->lock, flags);
630 list_for_each_entry_safe(cell, tmp, &cells, user_list)
631 cell_requeue(pool, cell);
634 static void requeue_io(struct thin_c *tc)
636 struct bio_list bios;
639 bio_list_init(&bios);
641 spin_lock_irqsave(&tc->lock, flags);
642 __merge_bio_list(&bios, &tc->deferred_bio_list);
643 __merge_bio_list(&bios, &tc->retry_on_resume_list);
644 spin_unlock_irqrestore(&tc->lock, flags);
646 error_bio_list(&bios, BLK_STS_DM_REQUEUE);
647 requeue_deferred_cells(tc);
650 static void error_retry_list_with_code(struct pool *pool, blk_status_t error)
655 list_for_each_entry_rcu(tc, &pool->active_thins, list)
656 error_thin_bio_list(tc, &tc->retry_on_resume_list, error);
660 static void error_retry_list(struct pool *pool)
662 error_retry_list_with_code(pool, get_pool_io_error_code(pool));
666 * This section of code contains the logic for processing a thin device's IO.
667 * Much of the code depends on pool object resources (lists, workqueues, etc)
668 * but most is exclusively called from the thin target rather than the thin-pool
672 static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
674 struct pool *pool = tc->pool;
675 sector_t block_nr = bio->bi_iter.bi_sector;
677 if (block_size_is_power_of_two(pool))
678 block_nr >>= pool->sectors_per_block_shift;
680 (void) sector_div(block_nr, pool->sectors_per_block);
686 * Returns the _complete_ blocks that this bio covers.
688 static void get_bio_block_range(struct thin_c *tc, struct bio *bio,
689 dm_block_t *begin, dm_block_t *end)
691 struct pool *pool = tc->pool;
692 sector_t b = bio->bi_iter.bi_sector;
693 sector_t e = b + (bio->bi_iter.bi_size >> SECTOR_SHIFT);
695 b += pool->sectors_per_block - 1ull; /* so we round up */
697 if (block_size_is_power_of_two(pool)) {
698 b >>= pool->sectors_per_block_shift;
699 e >>= pool->sectors_per_block_shift;
701 (void) sector_div(b, pool->sectors_per_block);
702 (void) sector_div(e, pool->sectors_per_block);
706 /* Can happen if the bio is within a single block. */
713 static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
715 struct pool *pool = tc->pool;
716 sector_t bi_sector = bio->bi_iter.bi_sector;
718 bio_set_dev(bio, tc->pool_dev->bdev);
719 if (block_size_is_power_of_two(pool))
720 bio->bi_iter.bi_sector =
721 (block << pool->sectors_per_block_shift) |
722 (bi_sector & (pool->sectors_per_block - 1));
724 bio->bi_iter.bi_sector = (block * pool->sectors_per_block) +
725 sector_div(bi_sector, pool->sectors_per_block);
728 static void remap_to_origin(struct thin_c *tc, struct bio *bio)
730 bio_set_dev(bio, tc->origin_dev->bdev);
733 static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
735 return op_is_flush(bio->bi_opf) &&
736 dm_thin_changed_this_transaction(tc->td);
739 static void inc_all_io_entry(struct pool *pool, struct bio *bio)
741 struct dm_thin_endio_hook *h;
743 if (bio_op(bio) == REQ_OP_DISCARD)
746 h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
747 h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
750 static void issue(struct thin_c *tc, struct bio *bio)
752 struct pool *pool = tc->pool;
755 if (!bio_triggers_commit(tc, bio)) {
756 generic_make_request(bio);
761 * Complete bio with an error if earlier I/O caused changes to
762 * the metadata that can't be committed e.g, due to I/O errors
763 * on the metadata device.
765 if (dm_thin_aborted_changes(tc->td)) {
771 * Batch together any bios that trigger commits and then issue a
772 * single commit for them in process_deferred_bios().
774 spin_lock_irqsave(&pool->lock, flags);
775 bio_list_add(&pool->deferred_flush_bios, bio);
776 spin_unlock_irqrestore(&pool->lock, flags);
779 static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
781 remap_to_origin(tc, bio);
785 static void remap_and_issue(struct thin_c *tc, struct bio *bio,
788 remap(tc, bio, block);
792 /*----------------------------------------------------------------*/
795 * Bio endio functions.
797 struct dm_thin_new_mapping {
798 struct list_head list;
804 * Track quiescing, copying and zeroing preparation actions. When this
805 * counter hits zero the block is prepared and can be inserted into the
808 atomic_t prepare_actions;
812 dm_block_t virt_begin, virt_end;
813 dm_block_t data_block;
814 struct dm_bio_prison_cell *cell;
817 * If the bio covers the whole area of a block then we can avoid
818 * zeroing or copying. Instead this bio is hooked. The bio will
819 * still be in the cell, so care has to be taken to avoid issuing
823 bio_end_io_t *saved_bi_end_io;
826 static void __complete_mapping_preparation(struct dm_thin_new_mapping *m)
828 struct pool *pool = m->tc->pool;
830 if (atomic_dec_and_test(&m->prepare_actions)) {
831 list_add_tail(&m->list, &pool->prepared_mappings);
836 static void complete_mapping_preparation(struct dm_thin_new_mapping *m)
839 struct pool *pool = m->tc->pool;
841 spin_lock_irqsave(&pool->lock, flags);
842 __complete_mapping_preparation(m);
843 spin_unlock_irqrestore(&pool->lock, flags);
846 static void copy_complete(int read_err, unsigned long write_err, void *context)
848 struct dm_thin_new_mapping *m = context;
850 m->status = read_err || write_err ? BLK_STS_IOERR : 0;
851 complete_mapping_preparation(m);
854 static void overwrite_endio(struct bio *bio)
856 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
857 struct dm_thin_new_mapping *m = h->overwrite_mapping;
859 bio->bi_end_io = m->saved_bi_end_io;
861 m->status = bio->bi_status;
862 complete_mapping_preparation(m);
865 /*----------------------------------------------------------------*/
872 * Prepared mapping jobs.
876 * This sends the bios in the cell, except the original holder, back
877 * to the deferred_bios list.
879 static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
881 struct pool *pool = tc->pool;
884 spin_lock_irqsave(&tc->lock, flags);
885 cell_release_no_holder(pool, cell, &tc->deferred_bio_list);
886 spin_unlock_irqrestore(&tc->lock, flags);
891 static void thin_defer_bio(struct thin_c *tc, struct bio *bio);
895 struct bio_list defer_bios;
896 struct bio_list issue_bios;
899 static void __inc_remap_and_issue_cell(void *context,
900 struct dm_bio_prison_cell *cell)
902 struct remap_info *info = context;
905 while ((bio = bio_list_pop(&cell->bios))) {
906 if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD)
907 bio_list_add(&info->defer_bios, bio);
909 inc_all_io_entry(info->tc->pool, bio);
912 * We can't issue the bios with the bio prison lock
913 * held, so we add them to a list to issue on
914 * return from this function.
916 bio_list_add(&info->issue_bios, bio);
921 static void inc_remap_and_issue_cell(struct thin_c *tc,
922 struct dm_bio_prison_cell *cell,
926 struct remap_info info;
929 bio_list_init(&info.defer_bios);
930 bio_list_init(&info.issue_bios);
933 * We have to be careful to inc any bios we're about to issue
934 * before the cell is released, and avoid a race with new bios
935 * being added to the cell.
937 cell_visit_release(tc->pool, __inc_remap_and_issue_cell,
940 while ((bio = bio_list_pop(&info.defer_bios)))
941 thin_defer_bio(tc, bio);
943 while ((bio = bio_list_pop(&info.issue_bios)))
944 remap_and_issue(info.tc, bio, block);
947 static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
949 cell_error(m->tc->pool, m->cell);
951 mempool_free(m, m->tc->pool->mapping_pool);
954 static void complete_overwrite_bio(struct thin_c *tc, struct bio *bio)
956 struct pool *pool = tc->pool;
960 * If the bio has the REQ_FUA flag set we must commit the metadata
961 * before signaling its completion.
963 if (!bio_triggers_commit(tc, bio)) {
969 * Complete bio with an error if earlier I/O caused changes to the
970 * metadata that can't be committed, e.g, due to I/O errors on the
973 if (dm_thin_aborted_changes(tc->td)) {
979 * Batch together any bios that trigger commits and then issue a
980 * single commit for them in process_deferred_bios().
982 spin_lock_irqsave(&pool->lock, flags);
983 bio_list_add(&pool->deferred_flush_completions, bio);
984 spin_unlock_irqrestore(&pool->lock, flags);
987 static void process_prepared_mapping(struct dm_thin_new_mapping *m)
989 struct thin_c *tc = m->tc;
990 struct pool *pool = tc->pool;
991 struct bio *bio = m->bio;
995 cell_error(pool, m->cell);
1000 * Commit the prepared block into the mapping btree.
1001 * Any I/O for this block arriving after this point will get
1002 * remapped to it directly.
1004 r = dm_thin_insert_block(tc->td, m->virt_begin, m->data_block);
1006 metadata_operation_failed(pool, "dm_thin_insert_block", r);
1007 cell_error(pool, m->cell);
1012 * Release any bios held while the block was being provisioned.
1013 * If we are processing a write bio that completely covers the block,
1014 * we already processed it so can ignore it now when processing
1015 * the bios in the cell.
1018 inc_remap_and_issue_cell(tc, m->cell, m->data_block);
1019 complete_overwrite_bio(tc, bio);
1021 inc_all_io_entry(tc->pool, m->cell->holder);
1022 remap_and_issue(tc, m->cell->holder, m->data_block);
1023 inc_remap_and_issue_cell(tc, m->cell, m->data_block);
1028 mempool_free(m, pool->mapping_pool);
1031 /*----------------------------------------------------------------*/
1033 static void free_discard_mapping(struct dm_thin_new_mapping *m)
1035 struct thin_c *tc = m->tc;
1037 cell_defer_no_holder(tc, m->cell);
1038 mempool_free(m, tc->pool->mapping_pool);
1041 static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
1043 bio_io_error(m->bio);
1044 free_discard_mapping(m);
1047 static void process_prepared_discard_success(struct dm_thin_new_mapping *m)
1050 free_discard_mapping(m);
1053 static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m)
1056 struct thin_c *tc = m->tc;
1058 r = dm_thin_remove_range(tc->td, m->cell->key.block_begin, m->cell->key.block_end);
1060 metadata_operation_failed(tc->pool, "dm_thin_remove_range", r);
1061 bio_io_error(m->bio);
1065 cell_defer_no_holder(tc, m->cell);
1066 mempool_free(m, tc->pool->mapping_pool);
1069 /*----------------------------------------------------------------*/
1071 static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m,
1072 struct bio *discard_parent)
1075 * We've already unmapped this range of blocks, but before we
1076 * passdown we have to check that these blocks are now unused.
1080 struct thin_c *tc = m->tc;
1081 struct pool *pool = tc->pool;
1082 dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin;
1083 struct discard_op op;
1085 begin_discard(&op, tc, discard_parent);
1087 /* find start of unmapped run */
1088 for (; b < end; b++) {
1089 r = dm_pool_block_is_shared(pool->pmd, b, &shared);
1100 /* find end of run */
1101 for (e = b + 1; e != end; e++) {
1102 r = dm_pool_block_is_shared(pool->pmd, e, &shared);
1110 r = issue_discard(&op, b, e);
1117 end_discard(&op, r);
1120 static void queue_passdown_pt2(struct dm_thin_new_mapping *m)
1122 unsigned long flags;
1123 struct pool *pool = m->tc->pool;
1125 spin_lock_irqsave(&pool->lock, flags);
1126 list_add_tail(&m->list, &pool->prepared_discards_pt2);
1127 spin_unlock_irqrestore(&pool->lock, flags);
1131 static void passdown_endio(struct bio *bio)
1134 * It doesn't matter if the passdown discard failed, we still want
1135 * to unmap (we ignore err).
1137 queue_passdown_pt2(bio->bi_private);
1141 static void process_prepared_discard_passdown_pt1(struct dm_thin_new_mapping *m)
1144 struct thin_c *tc = m->tc;
1145 struct pool *pool = tc->pool;
1146 struct bio *discard_parent;
1147 dm_block_t data_end = m->data_block + (m->virt_end - m->virt_begin);
1150 * Only this thread allocates blocks, so we can be sure that the
1151 * newly unmapped blocks will not be allocated before the end of
1154 r = dm_thin_remove_range(tc->td, m->virt_begin, m->virt_end);
1156 metadata_operation_failed(pool, "dm_thin_remove_range", r);
1157 bio_io_error(m->bio);
1158 cell_defer_no_holder(tc, m->cell);
1159 mempool_free(m, pool->mapping_pool);
1164 * Increment the unmapped blocks. This prevents a race between the
1165 * passdown io and reallocation of freed blocks.
1167 r = dm_pool_inc_data_range(pool->pmd, m->data_block, data_end);
1169 metadata_operation_failed(pool, "dm_pool_inc_data_range", r);
1170 bio_io_error(m->bio);
1171 cell_defer_no_holder(tc, m->cell);
1172 mempool_free(m, pool->mapping_pool);
1176 discard_parent = bio_alloc(GFP_NOIO, 1);
1177 if (!discard_parent) {
1178 DMWARN("%s: unable to allocate top level discard bio for passdown. Skipping passdown.",
1179 dm_device_name(tc->pool->pool_md));
1180 queue_passdown_pt2(m);
1183 discard_parent->bi_end_io = passdown_endio;
1184 discard_parent->bi_private = m;
1186 if (m->maybe_shared)
1187 passdown_double_checking_shared_status(m, discard_parent);
1189 struct discard_op op;
1191 begin_discard(&op, tc, discard_parent);
1192 r = issue_discard(&op, m->data_block, data_end);
1193 end_discard(&op, r);
1198 static void process_prepared_discard_passdown_pt2(struct dm_thin_new_mapping *m)
1201 struct thin_c *tc = m->tc;
1202 struct pool *pool = tc->pool;
1205 * The passdown has completed, so now we can decrement all those
1208 r = dm_pool_dec_data_range(pool->pmd, m->data_block,
1209 m->data_block + (m->virt_end - m->virt_begin));
1211 metadata_operation_failed(pool, "dm_pool_dec_data_range", r);
1212 bio_io_error(m->bio);
1216 cell_defer_no_holder(tc, m->cell);
1217 mempool_free(m, pool->mapping_pool);
1220 static void process_prepared(struct pool *pool, struct list_head *head,
1221 process_mapping_fn *fn)
1223 unsigned long flags;
1224 struct list_head maps;
1225 struct dm_thin_new_mapping *m, *tmp;
1227 INIT_LIST_HEAD(&maps);
1228 spin_lock_irqsave(&pool->lock, flags);
1229 list_splice_init(head, &maps);
1230 spin_unlock_irqrestore(&pool->lock, flags);
1232 list_for_each_entry_safe(m, tmp, &maps, list)
1237 * Deferred bio jobs.
1239 static int io_overlaps_block(struct pool *pool, struct bio *bio)
1241 return bio->bi_iter.bi_size ==
1242 (pool->sectors_per_block << SECTOR_SHIFT);
1245 static int io_overwrites_block(struct pool *pool, struct bio *bio)
1247 return (bio_data_dir(bio) == WRITE) &&
1248 io_overlaps_block(pool, bio);
1251 static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
1254 *save = bio->bi_end_io;
1255 bio->bi_end_io = fn;
1258 static int ensure_next_mapping(struct pool *pool)
1260 if (pool->next_mapping)
1263 pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);
1265 return pool->next_mapping ? 0 : -ENOMEM;
1268 static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
1270 struct dm_thin_new_mapping *m = pool->next_mapping;
1272 BUG_ON(!pool->next_mapping);
1274 memset(m, 0, sizeof(struct dm_thin_new_mapping));
1275 INIT_LIST_HEAD(&m->list);
1278 pool->next_mapping = NULL;
1283 static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m,
1284 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 r = dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m);
1295 DMERR_LIMIT("dm_kcopyd_zero() failed");
1296 copy_complete(1, 1, m);
1300 static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio,
1301 dm_block_t data_begin,
1302 struct dm_thin_new_mapping *m)
1304 struct pool *pool = tc->pool;
1305 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1307 h->overwrite_mapping = m;
1309 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
1310 inc_all_io_entry(pool, bio);
1311 remap_and_issue(tc, bio, data_begin);
1315 * A partial copy also needs to zero the uncopied region.
1317 static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
1318 struct dm_dev *origin, dm_block_t data_origin,
1319 dm_block_t data_dest,
1320 struct dm_bio_prison_cell *cell, struct bio *bio,
1324 struct pool *pool = tc->pool;
1325 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1328 m->virt_begin = virt_block;
1329 m->virt_end = virt_block + 1u;
1330 m->data_block = data_dest;
1334 * quiesce action + copy action + an extra reference held for the
1335 * duration of this function (we may need to inc later for a
1338 atomic_set(&m->prepare_actions, 3);
1340 if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
1341 complete_mapping_preparation(m); /* already quiesced */
1344 * IO to pool_dev remaps to the pool target's data_dev.
1346 * If the whole block of data is being overwritten, we can issue the
1347 * bio immediately. Otherwise we use kcopyd to clone the data first.
1349 if (io_overwrites_block(pool, bio))
1350 remap_and_issue_overwrite(tc, bio, data_dest, m);
1352 struct dm_io_region from, to;
1354 from.bdev = origin->bdev;
1355 from.sector = data_origin * pool->sectors_per_block;
1358 to.bdev = tc->pool_dev->bdev;
1359 to.sector = data_dest * pool->sectors_per_block;
1362 r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
1363 0, copy_complete, m);
1365 DMERR_LIMIT("dm_kcopyd_copy() failed");
1366 copy_complete(1, 1, m);
1369 * We allow the zero to be issued, to simplify the
1370 * error path. Otherwise we'd need to start
1371 * worrying about decrementing the prepare_actions
1377 * Do we need to zero a tail region?
1379 if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) {
1380 atomic_inc(&m->prepare_actions);
1382 data_dest * pool->sectors_per_block + len,
1383 (data_dest + 1) * pool->sectors_per_block);
1387 complete_mapping_preparation(m); /* drop our ref */
1390 static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
1391 dm_block_t data_origin, dm_block_t data_dest,
1392 struct dm_bio_prison_cell *cell, struct bio *bio)
1394 schedule_copy(tc, virt_block, tc->pool_dev,
1395 data_origin, data_dest, cell, bio,
1396 tc->pool->sectors_per_block);
1399 static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
1400 dm_block_t data_block, struct dm_bio_prison_cell *cell,
1403 struct pool *pool = tc->pool;
1404 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1406 atomic_set(&m->prepare_actions, 1); /* no need to quiesce */
1408 m->virt_begin = virt_block;
1409 m->virt_end = virt_block + 1u;
1410 m->data_block = data_block;
1414 * If the whole block of data is being overwritten or we are not
1415 * zeroing pre-existing data, we can issue the bio immediately.
1416 * Otherwise we use kcopyd to zero the data first.
1418 if (pool->pf.zero_new_blocks) {
1419 if (io_overwrites_block(pool, bio))
1420 remap_and_issue_overwrite(tc, bio, data_block, m);
1422 ll_zero(tc, m, data_block * pool->sectors_per_block,
1423 (data_block + 1) * pool->sectors_per_block);
1425 process_prepared_mapping(m);
1428 static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
1429 dm_block_t data_dest,
1430 struct dm_bio_prison_cell *cell, struct bio *bio)
1432 struct pool *pool = tc->pool;
1433 sector_t virt_block_begin = virt_block * pool->sectors_per_block;
1434 sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block;
1436 if (virt_block_end <= tc->origin_size)
1437 schedule_copy(tc, virt_block, tc->origin_dev,
1438 virt_block, data_dest, cell, bio,
1439 pool->sectors_per_block);
1441 else if (virt_block_begin < tc->origin_size)
1442 schedule_copy(tc, virt_block, tc->origin_dev,
1443 virt_block, data_dest, cell, bio,
1444 tc->origin_size - virt_block_begin);
1447 schedule_zero(tc, virt_block, data_dest, cell, bio);
1450 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode);
1452 static void requeue_bios(struct pool *pool);
1454 static bool is_read_only_pool_mode(enum pool_mode mode)
1456 return (mode == PM_OUT_OF_METADATA_SPACE || mode == PM_READ_ONLY);
1459 static bool is_read_only(struct pool *pool)
1461 return is_read_only_pool_mode(get_pool_mode(pool));
1464 static void check_for_metadata_space(struct pool *pool)
1467 const char *ooms_reason = NULL;
1470 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free);
1472 ooms_reason = "Could not get free metadata blocks";
1474 ooms_reason = "No free metadata blocks";
1476 if (ooms_reason && !is_read_only(pool)) {
1477 DMERR("%s", ooms_reason);
1478 set_pool_mode(pool, PM_OUT_OF_METADATA_SPACE);
1482 static void check_for_data_space(struct pool *pool)
1487 if (get_pool_mode(pool) != PM_OUT_OF_DATA_SPACE)
1490 r = dm_pool_get_free_block_count(pool->pmd, &nr_free);
1495 set_pool_mode(pool, PM_WRITE);
1501 * A non-zero return indicates read_only or fail_io mode.
1502 * Many callers don't care about the return value.
1504 static int commit(struct pool *pool)
1508 if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE)
1511 r = dm_pool_commit_metadata(pool->pmd);
1513 metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
1515 check_for_metadata_space(pool);
1516 check_for_data_space(pool);
1522 static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks)
1524 unsigned long flags;
1526 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
1527 DMWARN("%s: reached low water mark for data device: sending event.",
1528 dm_device_name(pool->pool_md));
1529 spin_lock_irqsave(&pool->lock, flags);
1530 pool->low_water_triggered = true;
1531 spin_unlock_irqrestore(&pool->lock, flags);
1532 dm_table_event(pool->ti->table);
1536 static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
1539 dm_block_t free_blocks;
1540 struct pool *pool = tc->pool;
1542 if (WARN_ON(get_pool_mode(pool) != PM_WRITE))
1545 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1547 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1551 check_low_water_mark(pool, free_blocks);
1555 * Try to commit to see if that will free up some
1562 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1564 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1569 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1574 r = dm_pool_alloc_data_block(pool->pmd, result);
1577 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1579 metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
1583 r = dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks);
1585 metadata_operation_failed(pool, "dm_pool_get_free_metadata_block_count", r);
1590 /* Let's commit before we use up the metadata reserve. */
1600 * If we have run out of space, queue bios until the device is
1601 * resumed, presumably after having been reloaded with more space.
1603 static void retry_on_resume(struct bio *bio)
1605 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1606 struct thin_c *tc = h->tc;
1607 unsigned long flags;
1609 spin_lock_irqsave(&tc->lock, flags);
1610 bio_list_add(&tc->retry_on_resume_list, bio);
1611 spin_unlock_irqrestore(&tc->lock, flags);
1614 static blk_status_t should_error_unserviceable_bio(struct pool *pool)
1616 enum pool_mode m = get_pool_mode(pool);
1620 /* Shouldn't get here */
1621 DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
1622 return BLK_STS_IOERR;
1624 case PM_OUT_OF_DATA_SPACE:
1625 return pool->pf.error_if_no_space ? BLK_STS_NOSPC : 0;
1627 case PM_OUT_OF_METADATA_SPACE:
1630 return BLK_STS_IOERR;
1632 /* Shouldn't get here */
1633 DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1634 return BLK_STS_IOERR;
1638 static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
1640 blk_status_t error = should_error_unserviceable_bio(pool);
1643 bio->bi_status = error;
1646 retry_on_resume(bio);
1649 static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell)
1652 struct bio_list bios;
1655 error = should_error_unserviceable_bio(pool);
1657 cell_error_with_code(pool, cell, error);
1661 bio_list_init(&bios);
1662 cell_release(pool, cell, &bios);
1664 while ((bio = bio_list_pop(&bios)))
1665 retry_on_resume(bio);
1668 static void process_discard_cell_no_passdown(struct thin_c *tc,
1669 struct dm_bio_prison_cell *virt_cell)
1671 struct pool *pool = tc->pool;
1672 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1675 * We don't need to lock the data blocks, since there's no
1676 * passdown. We only lock data blocks for allocation and breaking sharing.
1679 m->virt_begin = virt_cell->key.block_begin;
1680 m->virt_end = virt_cell->key.block_end;
1681 m->cell = virt_cell;
1682 m->bio = virt_cell->holder;
1684 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1685 pool->process_prepared_discard(m);
1688 static void break_up_discard_bio(struct thin_c *tc, dm_block_t begin, dm_block_t end,
1691 struct pool *pool = tc->pool;
1695 struct dm_cell_key data_key;
1696 struct dm_bio_prison_cell *data_cell;
1697 struct dm_thin_new_mapping *m;
1698 dm_block_t virt_begin, virt_end, data_begin;
1700 while (begin != end) {
1701 r = ensure_next_mapping(pool);
1703 /* we did our best */
1706 r = dm_thin_find_mapped_range(tc->td, begin, end, &virt_begin, &virt_end,
1707 &data_begin, &maybe_shared);
1710 * Silently fail, letting any mappings we've
1715 build_key(tc->td, PHYSICAL, data_begin, data_begin + (virt_end - virt_begin), &data_key);
1716 if (bio_detain(tc->pool, &data_key, NULL, &data_cell)) {
1717 /* contention, we'll give up with this range */
1723 * IO may still be going to the destination block. We must
1724 * quiesce before we can do the removal.
1726 m = get_next_mapping(pool);
1728 m->maybe_shared = maybe_shared;
1729 m->virt_begin = virt_begin;
1730 m->virt_end = virt_end;
1731 m->data_block = data_begin;
1732 m->cell = data_cell;
1736 * The parent bio must not complete before sub discard bios are
1737 * chained to it (see end_discard's bio_chain)!
1739 * This per-mapping bi_remaining increment is paired with
1740 * the implicit decrement that occurs via bio_endio() in
1743 bio_inc_remaining(bio);
1744 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1745 pool->process_prepared_discard(m);
1751 static void process_discard_cell_passdown(struct thin_c *tc, struct dm_bio_prison_cell *virt_cell)
1753 struct bio *bio = virt_cell->holder;
1754 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1757 * The virt_cell will only get freed once the origin bio completes.
1758 * This means it will remain locked while all the individual
1759 * passdown bios are in flight.
1761 h->cell = virt_cell;
1762 break_up_discard_bio(tc, virt_cell->key.block_begin, virt_cell->key.block_end, bio);
1765 * We complete the bio now, knowing that the bi_remaining field
1766 * will prevent completion until the sub range discards have
1772 static void process_discard_bio(struct thin_c *tc, struct bio *bio)
1774 dm_block_t begin, end;
1775 struct dm_cell_key virt_key;
1776 struct dm_bio_prison_cell *virt_cell;
1778 get_bio_block_range(tc, bio, &begin, &end);
1781 * The discard covers less than a block.
1787 build_key(tc->td, VIRTUAL, begin, end, &virt_key);
1788 if (bio_detain(tc->pool, &virt_key, bio, &virt_cell))
1790 * Potential starvation issue: We're relying on the
1791 * fs/application being well behaved, and not trying to
1792 * send IO to a region at the same time as discarding it.
1793 * If they do this persistently then it's possible this
1794 * cell will never be granted.
1798 tc->pool->process_discard_cell(tc, virt_cell);
1801 static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
1802 struct dm_cell_key *key,
1803 struct dm_thin_lookup_result *lookup_result,
1804 struct dm_bio_prison_cell *cell)
1807 dm_block_t data_block;
1808 struct pool *pool = tc->pool;
1810 r = alloc_data_block(tc, &data_block);
1813 schedule_internal_copy(tc, block, lookup_result->block,
1814 data_block, cell, bio);
1818 retry_bios_on_resume(pool, cell);
1822 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1824 cell_error(pool, cell);
1829 static void __remap_and_issue_shared_cell(void *context,
1830 struct dm_bio_prison_cell *cell)
1832 struct remap_info *info = context;
1835 while ((bio = bio_list_pop(&cell->bios))) {
1836 if (bio_data_dir(bio) == WRITE || op_is_flush(bio->bi_opf) ||
1837 bio_op(bio) == REQ_OP_DISCARD)
1838 bio_list_add(&info->defer_bios, bio);
1840 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));;
1842 h->shared_read_entry = dm_deferred_entry_inc(info->tc->pool->shared_read_ds);
1843 inc_all_io_entry(info->tc->pool, bio);
1844 bio_list_add(&info->issue_bios, bio);
1849 static void remap_and_issue_shared_cell(struct thin_c *tc,
1850 struct dm_bio_prison_cell *cell,
1854 struct remap_info info;
1857 bio_list_init(&info.defer_bios);
1858 bio_list_init(&info.issue_bios);
1860 cell_visit_release(tc->pool, __remap_and_issue_shared_cell,
1863 while ((bio = bio_list_pop(&info.defer_bios)))
1864 thin_defer_bio(tc, bio);
1866 while ((bio = bio_list_pop(&info.issue_bios)))
1867 remap_and_issue(tc, bio, block);
1870 static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1872 struct dm_thin_lookup_result *lookup_result,
1873 struct dm_bio_prison_cell *virt_cell)
1875 struct dm_bio_prison_cell *data_cell;
1876 struct pool *pool = tc->pool;
1877 struct dm_cell_key key;
1880 * If cell is already occupied, then sharing is already in the process
1881 * of being broken so we have nothing further to do here.
1883 build_data_key(tc->td, lookup_result->block, &key);
1884 if (bio_detain(pool, &key, bio, &data_cell)) {
1885 cell_defer_no_holder(tc, virt_cell);
1889 if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) {
1890 break_sharing(tc, bio, block, &key, lookup_result, data_cell);
1891 cell_defer_no_holder(tc, virt_cell);
1893 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1895 h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
1896 inc_all_io_entry(pool, bio);
1897 remap_and_issue(tc, bio, lookup_result->block);
1899 remap_and_issue_shared_cell(tc, data_cell, lookup_result->block);
1900 remap_and_issue_shared_cell(tc, virt_cell, lookup_result->block);
1904 static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
1905 struct dm_bio_prison_cell *cell)
1908 dm_block_t data_block;
1909 struct pool *pool = tc->pool;
1912 * Remap empty bios (flushes) immediately, without provisioning.
1914 if (!bio->bi_iter.bi_size) {
1915 inc_all_io_entry(pool, bio);
1916 cell_defer_no_holder(tc, cell);
1918 remap_and_issue(tc, bio, 0);
1923 * Fill read bios with zeroes and complete them immediately.
1925 if (bio_data_dir(bio) == READ) {
1927 cell_defer_no_holder(tc, cell);
1932 r = alloc_data_block(tc, &data_block);
1936 schedule_external_copy(tc, block, data_block, cell, bio);
1938 schedule_zero(tc, block, data_block, cell, bio);
1942 retry_bios_on_resume(pool, cell);
1946 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1948 cell_error(pool, cell);
1953 static void process_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
1956 struct pool *pool = tc->pool;
1957 struct bio *bio = cell->holder;
1958 dm_block_t block = get_bio_block(tc, bio);
1959 struct dm_thin_lookup_result lookup_result;
1961 if (tc->requeue_mode) {
1962 cell_requeue(pool, cell);
1966 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1969 if (lookup_result.shared)
1970 process_shared_bio(tc, bio, block, &lookup_result, cell);
1972 inc_all_io_entry(pool, bio);
1973 remap_and_issue(tc, bio, lookup_result.block);
1974 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
1979 if (bio_data_dir(bio) == READ && tc->origin_dev) {
1980 inc_all_io_entry(pool, bio);
1981 cell_defer_no_holder(tc, cell);
1983 if (bio_end_sector(bio) <= tc->origin_size)
1984 remap_to_origin_and_issue(tc, bio);
1986 else if (bio->bi_iter.bi_sector < tc->origin_size) {
1988 bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT;
1989 remap_to_origin_and_issue(tc, bio);
1996 provision_block(tc, bio, block, cell);
2000 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
2002 cell_defer_no_holder(tc, cell);
2008 static void process_bio(struct thin_c *tc, struct bio *bio)
2010 struct pool *pool = tc->pool;
2011 dm_block_t block = get_bio_block(tc, bio);
2012 struct dm_bio_prison_cell *cell;
2013 struct dm_cell_key key;
2016 * If cell is already occupied, then the block is already
2017 * being provisioned so we have nothing further to do here.
2019 build_virtual_key(tc->td, block, &key);
2020 if (bio_detain(pool, &key, bio, &cell))
2023 process_cell(tc, cell);
2026 static void __process_bio_read_only(struct thin_c *tc, struct bio *bio,
2027 struct dm_bio_prison_cell *cell)
2030 int rw = bio_data_dir(bio);
2031 dm_block_t block = get_bio_block(tc, bio);
2032 struct dm_thin_lookup_result lookup_result;
2034 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
2037 if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) {
2038 handle_unserviceable_bio(tc->pool, bio);
2040 cell_defer_no_holder(tc, cell);
2042 inc_all_io_entry(tc->pool, bio);
2043 remap_and_issue(tc, bio, lookup_result.block);
2045 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
2051 cell_defer_no_holder(tc, cell);
2053 handle_unserviceable_bio(tc->pool, bio);
2057 if (tc->origin_dev) {
2058 inc_all_io_entry(tc->pool, bio);
2059 remap_to_origin_and_issue(tc, bio);
2068 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
2071 cell_defer_no_holder(tc, cell);
2077 static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
2079 __process_bio_read_only(tc, bio, NULL);
2082 static void process_cell_read_only(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2084 __process_bio_read_only(tc, cell->holder, cell);
2087 static void process_bio_success(struct thin_c *tc, struct bio *bio)
2092 static void process_bio_fail(struct thin_c *tc, struct bio *bio)
2097 static void process_cell_success(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2099 cell_success(tc->pool, cell);
2102 static void process_cell_fail(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2104 cell_error(tc->pool, cell);
2108 * FIXME: should we also commit due to size of transaction, measured in
2111 static int need_commit_due_to_time(struct pool *pool)
2113 return !time_in_range(jiffies, pool->last_commit_jiffies,
2114 pool->last_commit_jiffies + COMMIT_PERIOD);
2117 #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node)
2118 #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook))
2120 static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio)
2122 struct rb_node **rbp, *parent;
2123 struct dm_thin_endio_hook *pbd;
2124 sector_t bi_sector = bio->bi_iter.bi_sector;
2126 rbp = &tc->sort_bio_list.rb_node;
2130 pbd = thin_pbd(parent);
2132 if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector)
2133 rbp = &(*rbp)->rb_left;
2135 rbp = &(*rbp)->rb_right;
2138 pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2139 rb_link_node(&pbd->rb_node, parent, rbp);
2140 rb_insert_color(&pbd->rb_node, &tc->sort_bio_list);
2143 static void __extract_sorted_bios(struct thin_c *tc)
2145 struct rb_node *node;
2146 struct dm_thin_endio_hook *pbd;
2149 for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) {
2150 pbd = thin_pbd(node);
2151 bio = thin_bio(pbd);
2153 bio_list_add(&tc->deferred_bio_list, bio);
2154 rb_erase(&pbd->rb_node, &tc->sort_bio_list);
2157 WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list));
2160 static void __sort_thin_deferred_bios(struct thin_c *tc)
2163 struct bio_list bios;
2165 bio_list_init(&bios);
2166 bio_list_merge(&bios, &tc->deferred_bio_list);
2167 bio_list_init(&tc->deferred_bio_list);
2169 /* Sort deferred_bio_list using rb-tree */
2170 while ((bio = bio_list_pop(&bios)))
2171 __thin_bio_rb_add(tc, bio);
2174 * Transfer the sorted bios in sort_bio_list back to
2175 * deferred_bio_list to allow lockless submission of
2178 __extract_sorted_bios(tc);
2181 static void process_thin_deferred_bios(struct thin_c *tc)
2183 struct pool *pool = tc->pool;
2184 unsigned long flags;
2186 struct bio_list bios;
2187 struct blk_plug plug;
2190 if (tc->requeue_mode) {
2191 error_thin_bio_list(tc, &tc->deferred_bio_list,
2192 BLK_STS_DM_REQUEUE);
2196 bio_list_init(&bios);
2198 spin_lock_irqsave(&tc->lock, flags);
2200 if (bio_list_empty(&tc->deferred_bio_list)) {
2201 spin_unlock_irqrestore(&tc->lock, flags);
2205 __sort_thin_deferred_bios(tc);
2207 bio_list_merge(&bios, &tc->deferred_bio_list);
2208 bio_list_init(&tc->deferred_bio_list);
2210 spin_unlock_irqrestore(&tc->lock, flags);
2212 blk_start_plug(&plug);
2213 while ((bio = bio_list_pop(&bios))) {
2215 * If we've got no free new_mapping structs, and processing
2216 * this bio might require one, we pause until there are some
2217 * prepared mappings to process.
2219 if (ensure_next_mapping(pool)) {
2220 spin_lock_irqsave(&tc->lock, flags);
2221 bio_list_add(&tc->deferred_bio_list, bio);
2222 bio_list_merge(&tc->deferred_bio_list, &bios);
2223 spin_unlock_irqrestore(&tc->lock, flags);
2227 if (bio_op(bio) == REQ_OP_DISCARD)
2228 pool->process_discard(tc, bio);
2230 pool->process_bio(tc, bio);
2232 if ((count++ & 127) == 0) {
2233 throttle_work_update(&pool->throttle);
2234 dm_pool_issue_prefetches(pool->pmd);
2238 blk_finish_plug(&plug);
2241 static int cmp_cells(const void *lhs, const void *rhs)
2243 struct dm_bio_prison_cell *lhs_cell = *((struct dm_bio_prison_cell **) lhs);
2244 struct dm_bio_prison_cell *rhs_cell = *((struct dm_bio_prison_cell **) rhs);
2246 BUG_ON(!lhs_cell->holder);
2247 BUG_ON(!rhs_cell->holder);
2249 if (lhs_cell->holder->bi_iter.bi_sector < rhs_cell->holder->bi_iter.bi_sector)
2252 if (lhs_cell->holder->bi_iter.bi_sector > rhs_cell->holder->bi_iter.bi_sector)
2258 static unsigned sort_cells(struct pool *pool, struct list_head *cells)
2261 struct dm_bio_prison_cell *cell, *tmp;
2263 list_for_each_entry_safe(cell, tmp, cells, user_list) {
2264 if (count >= CELL_SORT_ARRAY_SIZE)
2267 pool->cell_sort_array[count++] = cell;
2268 list_del(&cell->user_list);
2271 sort(pool->cell_sort_array, count, sizeof(cell), cmp_cells, NULL);
2276 static void process_thin_deferred_cells(struct thin_c *tc)
2278 struct pool *pool = tc->pool;
2279 unsigned long flags;
2280 struct list_head cells;
2281 struct dm_bio_prison_cell *cell;
2282 unsigned i, j, count;
2284 INIT_LIST_HEAD(&cells);
2286 spin_lock_irqsave(&tc->lock, flags);
2287 list_splice_init(&tc->deferred_cells, &cells);
2288 spin_unlock_irqrestore(&tc->lock, flags);
2290 if (list_empty(&cells))
2294 count = sort_cells(tc->pool, &cells);
2296 for (i = 0; i < count; i++) {
2297 cell = pool->cell_sort_array[i];
2298 BUG_ON(!cell->holder);
2301 * If we've got no free new_mapping structs, and processing
2302 * this bio might require one, we pause until there are some
2303 * prepared mappings to process.
2305 if (ensure_next_mapping(pool)) {
2306 for (j = i; j < count; j++)
2307 list_add(&pool->cell_sort_array[j]->user_list, &cells);
2309 spin_lock_irqsave(&tc->lock, flags);
2310 list_splice(&cells, &tc->deferred_cells);
2311 spin_unlock_irqrestore(&tc->lock, flags);
2315 if (bio_op(cell->holder) == REQ_OP_DISCARD)
2316 pool->process_discard_cell(tc, cell);
2318 pool->process_cell(tc, cell);
2321 } while (!list_empty(&cells));
2324 static void thin_get(struct thin_c *tc);
2325 static void thin_put(struct thin_c *tc);
2328 * We can't hold rcu_read_lock() around code that can block. So we
2329 * find a thin with the rcu lock held; bump a refcount; then drop
2332 static struct thin_c *get_first_thin(struct pool *pool)
2334 struct thin_c *tc = NULL;
2337 if (!list_empty(&pool->active_thins)) {
2338 tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list);
2346 static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc)
2348 struct thin_c *old_tc = tc;
2351 list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) {
2363 static void process_deferred_bios(struct pool *pool)
2365 unsigned long flags;
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_irqsave(&pool->lock, flags);
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_irqrestore(&pool->lock, flags);
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)))
2409 generic_make_request(bio);
2412 static void do_worker(struct work_struct *ws)
2414 struct pool *pool = container_of(ws, struct pool, worker);
2416 throttle_work_start(&pool->throttle);
2417 dm_pool_issue_prefetches(pool->pmd);
2418 throttle_work_update(&pool->throttle);
2419 process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
2420 throttle_work_update(&pool->throttle);
2421 process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
2422 throttle_work_update(&pool->throttle);
2423 process_prepared(pool, &pool->prepared_discards_pt2, &pool->process_prepared_discard_pt2);
2424 throttle_work_update(&pool->throttle);
2425 process_deferred_bios(pool);
2426 throttle_work_complete(&pool->throttle);
2430 * We want to commit periodically so that not too much
2431 * unwritten data builds up.
2433 static void do_waker(struct work_struct *ws)
2435 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
2437 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
2441 * We're holding onto IO to allow userland time to react. After the
2442 * timeout either the pool will have been resized (and thus back in
2443 * PM_WRITE mode), or we degrade to PM_OUT_OF_DATA_SPACE w/ error_if_no_space.
2445 static void do_no_space_timeout(struct work_struct *ws)
2447 struct pool *pool = container_of(to_delayed_work(ws), struct pool,
2450 if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) {
2451 pool->pf.error_if_no_space = true;
2452 notify_of_pool_mode_change(pool);
2453 error_retry_list_with_code(pool, BLK_STS_NOSPC);
2457 /*----------------------------------------------------------------*/
2460 struct work_struct worker;
2461 struct completion complete;
2464 static struct pool_work *to_pool_work(struct work_struct *ws)
2466 return container_of(ws, struct pool_work, worker);
2469 static void pool_work_complete(struct pool_work *pw)
2471 complete(&pw->complete);
2474 static void pool_work_wait(struct pool_work *pw, struct pool *pool,
2475 void (*fn)(struct work_struct *))
2477 INIT_WORK_ONSTACK(&pw->worker, fn);
2478 init_completion(&pw->complete);
2479 queue_work(pool->wq, &pw->worker);
2480 wait_for_completion(&pw->complete);
2483 /*----------------------------------------------------------------*/
2485 struct noflush_work {
2486 struct pool_work pw;
2490 static struct noflush_work *to_noflush(struct work_struct *ws)
2492 return container_of(to_pool_work(ws), struct noflush_work, pw);
2495 static void do_noflush_start(struct work_struct *ws)
2497 struct noflush_work *w = to_noflush(ws);
2498 w->tc->requeue_mode = true;
2500 pool_work_complete(&w->pw);
2503 static void do_noflush_stop(struct work_struct *ws)
2505 struct noflush_work *w = to_noflush(ws);
2506 w->tc->requeue_mode = false;
2507 pool_work_complete(&w->pw);
2510 static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *))
2512 struct noflush_work w;
2515 pool_work_wait(&w.pw, tc->pool, fn);
2518 /*----------------------------------------------------------------*/
2520 static bool passdown_enabled(struct pool_c *pt)
2522 return pt->adjusted_pf.discard_passdown;
2525 static void set_discard_callbacks(struct pool *pool)
2527 struct pool_c *pt = pool->ti->private;
2529 if (passdown_enabled(pt)) {
2530 pool->process_discard_cell = process_discard_cell_passdown;
2531 pool->process_prepared_discard = process_prepared_discard_passdown_pt1;
2532 pool->process_prepared_discard_pt2 = process_prepared_discard_passdown_pt2;
2534 pool->process_discard_cell = process_discard_cell_no_passdown;
2535 pool->process_prepared_discard = process_prepared_discard_no_passdown;
2539 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode)
2541 struct pool_c *pt = pool->ti->private;
2542 bool needs_check = dm_pool_metadata_needs_check(pool->pmd);
2543 enum pool_mode old_mode = get_pool_mode(pool);
2544 unsigned long no_space_timeout = ACCESS_ONCE(no_space_timeout_secs) * HZ;
2547 * Never allow the pool to transition to PM_WRITE mode if user
2548 * intervention is required to verify metadata and data consistency.
2550 if (new_mode == PM_WRITE && needs_check) {
2551 DMERR("%s: unable to switch pool to write mode until repaired.",
2552 dm_device_name(pool->pool_md));
2553 if (old_mode != new_mode)
2554 new_mode = old_mode;
2556 new_mode = PM_READ_ONLY;
2559 * If we were in PM_FAIL mode, rollback of metadata failed. We're
2560 * not going to recover without a thin_repair. So we never let the
2561 * pool move out of the old mode.
2563 if (old_mode == PM_FAIL)
2564 new_mode = old_mode;
2568 dm_pool_metadata_read_only(pool->pmd);
2569 pool->process_bio = process_bio_fail;
2570 pool->process_discard = process_bio_fail;
2571 pool->process_cell = process_cell_fail;
2572 pool->process_discard_cell = process_cell_fail;
2573 pool->process_prepared_mapping = process_prepared_mapping_fail;
2574 pool->process_prepared_discard = process_prepared_discard_fail;
2576 error_retry_list(pool);
2579 case PM_OUT_OF_METADATA_SPACE:
2581 dm_pool_metadata_read_only(pool->pmd);
2582 pool->process_bio = process_bio_read_only;
2583 pool->process_discard = process_bio_success;
2584 pool->process_cell = process_cell_read_only;
2585 pool->process_discard_cell = process_cell_success;
2586 pool->process_prepared_mapping = process_prepared_mapping_fail;
2587 pool->process_prepared_discard = process_prepared_discard_success;
2589 error_retry_list(pool);
2592 case PM_OUT_OF_DATA_SPACE:
2594 * Ideally we'd never hit this state; the low water mark
2595 * would trigger userland to extend the pool before we
2596 * completely run out of data space. However, many small
2597 * IOs to unprovisioned space can consume data space at an
2598 * alarming rate. Adjust your low water mark if you're
2599 * frequently seeing this mode.
2601 pool->out_of_data_space = true;
2602 pool->process_bio = process_bio_read_only;
2603 pool->process_discard = process_discard_bio;
2604 pool->process_cell = process_cell_read_only;
2605 pool->process_prepared_mapping = process_prepared_mapping;
2606 set_discard_callbacks(pool);
2608 if (!pool->pf.error_if_no_space && no_space_timeout)
2609 queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout);
2613 if (old_mode == PM_OUT_OF_DATA_SPACE)
2614 cancel_delayed_work_sync(&pool->no_space_timeout);
2615 pool->out_of_data_space = false;
2616 pool->pf.error_if_no_space = pt->requested_pf.error_if_no_space;
2617 dm_pool_metadata_read_write(pool->pmd);
2618 pool->process_bio = process_bio;
2619 pool->process_discard = process_discard_bio;
2620 pool->process_cell = process_cell;
2621 pool->process_prepared_mapping = process_prepared_mapping;
2622 set_discard_callbacks(pool);
2626 pool->pf.mode = new_mode;
2628 * The pool mode may have changed, sync it so bind_control_target()
2629 * doesn't cause an unexpected mode transition on resume.
2631 pt->adjusted_pf.mode = new_mode;
2633 if (old_mode != new_mode)
2634 notify_of_pool_mode_change(pool);
2637 static void abort_transaction(struct pool *pool)
2639 const char *dev_name = dm_device_name(pool->pool_md);
2641 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name);
2642 if (dm_pool_abort_metadata(pool->pmd)) {
2643 DMERR("%s: failed to abort metadata transaction", dev_name);
2644 set_pool_mode(pool, PM_FAIL);
2647 if (dm_pool_metadata_set_needs_check(pool->pmd)) {
2648 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name);
2649 set_pool_mode(pool, PM_FAIL);
2653 static void metadata_operation_failed(struct pool *pool, const char *op, int r)
2655 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
2656 dm_device_name(pool->pool_md), op, r);
2658 abort_transaction(pool);
2659 set_pool_mode(pool, PM_READ_ONLY);
2662 /*----------------------------------------------------------------*/
2665 * Mapping functions.
2669 * Called only while mapping a thin bio to hand it over to the workqueue.
2671 static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
2673 unsigned long flags;
2674 struct pool *pool = tc->pool;
2676 spin_lock_irqsave(&tc->lock, flags);
2677 bio_list_add(&tc->deferred_bio_list, bio);
2678 spin_unlock_irqrestore(&tc->lock, flags);
2683 static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio)
2685 struct pool *pool = tc->pool;
2687 throttle_lock(&pool->throttle);
2688 thin_defer_bio(tc, bio);
2689 throttle_unlock(&pool->throttle);
2692 static void thin_defer_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2694 unsigned long flags;
2695 struct pool *pool = tc->pool;
2697 throttle_lock(&pool->throttle);
2698 spin_lock_irqsave(&tc->lock, flags);
2699 list_add_tail(&cell->user_list, &tc->deferred_cells);
2700 spin_unlock_irqrestore(&tc->lock, flags);
2701 throttle_unlock(&pool->throttle);
2706 static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
2708 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2711 h->shared_read_entry = NULL;
2712 h->all_io_entry = NULL;
2713 h->overwrite_mapping = NULL;
2718 * Non-blocking function called from the thin target's map function.
2720 static int thin_bio_map(struct dm_target *ti, struct bio *bio)
2723 struct thin_c *tc = ti->private;
2724 dm_block_t block = get_bio_block(tc, bio);
2725 struct dm_thin_device *td = tc->td;
2726 struct dm_thin_lookup_result result;
2727 struct dm_bio_prison_cell *virt_cell, *data_cell;
2728 struct dm_cell_key key;
2730 thin_hook_bio(tc, bio);
2732 if (tc->requeue_mode) {
2733 bio->bi_status = BLK_STS_DM_REQUEUE;
2735 return DM_MAPIO_SUBMITTED;
2738 if (get_pool_mode(tc->pool) == PM_FAIL) {
2740 return DM_MAPIO_SUBMITTED;
2743 if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD) {
2744 thin_defer_bio_with_throttle(tc, bio);
2745 return DM_MAPIO_SUBMITTED;
2749 * We must hold the virtual cell before doing the lookup, otherwise
2750 * there's a race with discard.
2752 build_virtual_key(tc->td, block, &key);
2753 if (bio_detain(tc->pool, &key, bio, &virt_cell))
2754 return DM_MAPIO_SUBMITTED;
2756 r = dm_thin_find_block(td, block, 0, &result);
2759 * Note that we defer readahead too.
2763 if (unlikely(result.shared)) {
2765 * We have a race condition here between the
2766 * result.shared value returned by the lookup and
2767 * snapshot creation, which may cause new
2770 * To avoid this always quiesce the origin before
2771 * taking the snap. You want to do this anyway to
2772 * ensure a consistent application view
2775 * More distant ancestors are irrelevant. The
2776 * shared flag will be set in their case.
2778 thin_defer_cell(tc, virt_cell);
2779 return DM_MAPIO_SUBMITTED;
2782 build_data_key(tc->td, result.block, &key);
2783 if (bio_detain(tc->pool, &key, bio, &data_cell)) {
2784 cell_defer_no_holder(tc, virt_cell);
2785 return DM_MAPIO_SUBMITTED;
2788 inc_all_io_entry(tc->pool, bio);
2789 cell_defer_no_holder(tc, data_cell);
2790 cell_defer_no_holder(tc, virt_cell);
2792 remap(tc, bio, result.block);
2793 return DM_MAPIO_REMAPPED;
2797 thin_defer_cell(tc, virt_cell);
2798 return DM_MAPIO_SUBMITTED;
2802 * Must always call bio_io_error on failure.
2803 * dm_thin_find_block can fail with -EINVAL if the
2804 * pool is switched to fail-io mode.
2807 cell_defer_no_holder(tc, virt_cell);
2808 return DM_MAPIO_SUBMITTED;
2812 static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
2814 struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
2815 struct request_queue *q;
2817 if (get_pool_mode(pt->pool) == PM_OUT_OF_DATA_SPACE)
2820 q = bdev_get_queue(pt->data_dev->bdev);
2821 return bdi_congested(q->backing_dev_info, bdi_bits);
2824 static void requeue_bios(struct pool *pool)
2826 unsigned long flags;
2830 list_for_each_entry_rcu(tc, &pool->active_thins, list) {
2831 spin_lock_irqsave(&tc->lock, flags);
2832 bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list);
2833 bio_list_init(&tc->retry_on_resume_list);
2834 spin_unlock_irqrestore(&tc->lock, flags);
2839 /*----------------------------------------------------------------
2840 * Binding of control targets to a pool object
2841 *--------------------------------------------------------------*/
2842 static bool data_dev_supports_discard(struct pool_c *pt)
2844 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
2846 return q && blk_queue_discard(q);
2849 static bool is_factor(sector_t block_size, uint32_t n)
2851 return !sector_div(block_size, n);
2855 * If discard_passdown was enabled verify that the data device
2856 * supports discards. Disable discard_passdown if not.
2858 static void disable_passdown_if_not_supported(struct pool_c *pt)
2860 struct pool *pool = pt->pool;
2861 struct block_device *data_bdev = pt->data_dev->bdev;
2862 struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
2863 const char *reason = NULL;
2864 char buf[BDEVNAME_SIZE];
2866 if (!pt->adjusted_pf.discard_passdown)
2869 if (!data_dev_supports_discard(pt))
2870 reason = "discard unsupported";
2872 else if (data_limits->max_discard_sectors < pool->sectors_per_block)
2873 reason = "max discard sectors smaller than a block";
2876 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason);
2877 pt->adjusted_pf.discard_passdown = false;
2881 static int bind_control_target(struct pool *pool, struct dm_target *ti)
2883 struct pool_c *pt = ti->private;
2886 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
2888 enum pool_mode old_mode = get_pool_mode(pool);
2889 enum pool_mode new_mode = pt->adjusted_pf.mode;
2892 * Don't change the pool's mode until set_pool_mode() below.
2893 * Otherwise the pool's process_* function pointers may
2894 * not match the desired pool mode.
2896 pt->adjusted_pf.mode = old_mode;
2899 pool->pf = pt->adjusted_pf;
2900 pool->low_water_blocks = pt->low_water_blocks;
2902 set_pool_mode(pool, new_mode);
2907 static void unbind_control_target(struct pool *pool, struct dm_target *ti)
2913 /*----------------------------------------------------------------
2915 *--------------------------------------------------------------*/
2916 /* Initialize pool features. */
2917 static void pool_features_init(struct pool_features *pf)
2919 pf->mode = PM_WRITE;
2920 pf->zero_new_blocks = true;
2921 pf->discard_enabled = true;
2922 pf->discard_passdown = true;
2923 pf->error_if_no_space = false;
2926 static void __pool_destroy(struct pool *pool)
2928 __pool_table_remove(pool);
2930 vfree(pool->cell_sort_array);
2931 if (dm_pool_metadata_close(pool->pmd) < 0)
2932 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
2934 dm_bio_prison_destroy(pool->prison);
2935 dm_kcopyd_client_destroy(pool->copier);
2937 cancel_delayed_work_sync(&pool->waker);
2938 cancel_delayed_work_sync(&pool->no_space_timeout);
2940 destroy_workqueue(pool->wq);
2942 if (pool->next_mapping)
2943 mempool_free(pool->next_mapping, pool->mapping_pool);
2944 mempool_destroy(pool->mapping_pool);
2945 dm_deferred_set_destroy(pool->shared_read_ds);
2946 dm_deferred_set_destroy(pool->all_io_ds);
2950 static struct kmem_cache *_new_mapping_cache;
2952 static struct pool *pool_create(struct mapped_device *pool_md,
2953 struct block_device *metadata_dev,
2954 unsigned long block_size,
2955 int read_only, char **error)
2960 struct dm_pool_metadata *pmd;
2961 bool format_device = read_only ? false : true;
2963 pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
2965 *error = "Error creating metadata object";
2966 return (struct pool *)pmd;
2969 pool = kzalloc(sizeof(*pool), GFP_KERNEL);
2971 *error = "Error allocating memory for pool";
2972 err_p = ERR_PTR(-ENOMEM);
2977 pool->sectors_per_block = block_size;
2978 if (block_size & (block_size - 1))
2979 pool->sectors_per_block_shift = -1;
2981 pool->sectors_per_block_shift = __ffs(block_size);
2982 pool->low_water_blocks = 0;
2983 pool_features_init(&pool->pf);
2984 pool->prison = dm_bio_prison_create();
2985 if (!pool->prison) {
2986 *error = "Error creating pool's bio prison";
2987 err_p = ERR_PTR(-ENOMEM);
2991 pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2992 if (IS_ERR(pool->copier)) {
2993 r = PTR_ERR(pool->copier);
2994 *error = "Error creating pool's kcopyd client";
2996 goto bad_kcopyd_client;
3000 * Create singlethreaded workqueue that will service all devices
3001 * that use this metadata.
3003 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
3005 *error = "Error creating pool's workqueue";
3006 err_p = ERR_PTR(-ENOMEM);
3010 throttle_init(&pool->throttle);
3011 INIT_WORK(&pool->worker, do_worker);
3012 INIT_DELAYED_WORK(&pool->waker, do_waker);
3013 INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout);
3014 spin_lock_init(&pool->lock);
3015 bio_list_init(&pool->deferred_flush_bios);
3016 bio_list_init(&pool->deferred_flush_completions);
3017 INIT_LIST_HEAD(&pool->prepared_mappings);
3018 INIT_LIST_HEAD(&pool->prepared_discards);
3019 INIT_LIST_HEAD(&pool->prepared_discards_pt2);
3020 INIT_LIST_HEAD(&pool->active_thins);
3021 pool->low_water_triggered = false;
3022 pool->suspended = true;
3023 pool->out_of_data_space = false;
3025 pool->shared_read_ds = dm_deferred_set_create();
3026 if (!pool->shared_read_ds) {
3027 *error = "Error creating pool's shared read deferred set";
3028 err_p = ERR_PTR(-ENOMEM);
3029 goto bad_shared_read_ds;
3032 pool->all_io_ds = dm_deferred_set_create();
3033 if (!pool->all_io_ds) {
3034 *error = "Error creating pool's all io deferred set";
3035 err_p = ERR_PTR(-ENOMEM);
3039 pool->next_mapping = NULL;
3040 pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE,
3041 _new_mapping_cache);
3042 if (!pool->mapping_pool) {
3043 *error = "Error creating pool's mapping mempool";
3044 err_p = ERR_PTR(-ENOMEM);
3045 goto bad_mapping_pool;
3048 pool->cell_sort_array = vmalloc(sizeof(*pool->cell_sort_array) * CELL_SORT_ARRAY_SIZE);
3049 if (!pool->cell_sort_array) {
3050 *error = "Error allocating cell sort array";
3051 err_p = ERR_PTR(-ENOMEM);
3052 goto bad_sort_array;
3055 pool->ref_count = 1;
3056 pool->last_commit_jiffies = jiffies;
3057 pool->pool_md = pool_md;
3058 pool->md_dev = metadata_dev;
3059 __pool_table_insert(pool);
3064 mempool_destroy(pool->mapping_pool);
3066 dm_deferred_set_destroy(pool->all_io_ds);
3068 dm_deferred_set_destroy(pool->shared_read_ds);
3070 destroy_workqueue(pool->wq);
3072 dm_kcopyd_client_destroy(pool->copier);
3074 dm_bio_prison_destroy(pool->prison);
3078 if (dm_pool_metadata_close(pmd))
3079 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
3084 static void __pool_inc(struct pool *pool)
3086 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
3090 static void __pool_dec(struct pool *pool)
3092 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
3093 BUG_ON(!pool->ref_count);
3094 if (!--pool->ref_count)
3095 __pool_destroy(pool);
3098 static struct pool *__pool_find(struct mapped_device *pool_md,
3099 struct block_device *metadata_dev,
3100 unsigned long block_size, int read_only,
3101 char **error, int *created)
3103 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
3106 if (pool->pool_md != pool_md) {
3107 *error = "metadata device already in use by a pool";
3108 return ERR_PTR(-EBUSY);
3113 pool = __pool_table_lookup(pool_md);
3115 if (pool->md_dev != metadata_dev) {
3116 *error = "different pool cannot replace a pool";
3117 return ERR_PTR(-EINVAL);
3122 pool = pool_create(pool_md, metadata_dev, block_size, read_only, error);
3130 /*----------------------------------------------------------------
3131 * Pool target methods
3132 *--------------------------------------------------------------*/
3133 static void pool_dtr(struct dm_target *ti)
3135 struct pool_c *pt = ti->private;
3137 mutex_lock(&dm_thin_pool_table.mutex);
3139 unbind_control_target(pt->pool, ti);
3140 __pool_dec(pt->pool);
3141 dm_put_device(ti, pt->metadata_dev);
3142 dm_put_device(ti, pt->data_dev);
3145 mutex_unlock(&dm_thin_pool_table.mutex);
3148 static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
3149 struct dm_target *ti)
3153 const char *arg_name;
3155 static const struct dm_arg _args[] = {
3156 {0, 4, "Invalid number of pool feature arguments"},
3160 * No feature arguments supplied.
3165 r = dm_read_arg_group(_args, as, &argc, &ti->error);
3169 while (argc && !r) {
3170 arg_name = dm_shift_arg(as);
3173 if (!strcasecmp(arg_name, "skip_block_zeroing"))
3174 pf->zero_new_blocks = false;
3176 else if (!strcasecmp(arg_name, "ignore_discard"))
3177 pf->discard_enabled = false;
3179 else if (!strcasecmp(arg_name, "no_discard_passdown"))
3180 pf->discard_passdown = false;
3182 else if (!strcasecmp(arg_name, "read_only"))
3183 pf->mode = PM_READ_ONLY;
3185 else if (!strcasecmp(arg_name, "error_if_no_space"))
3186 pf->error_if_no_space = true;
3189 ti->error = "Unrecognised pool feature requested";
3198 static void metadata_low_callback(void *context)
3200 struct pool *pool = context;
3202 DMWARN("%s: reached low water mark for metadata device: sending event.",
3203 dm_device_name(pool->pool_md));
3205 dm_table_event(pool->ti->table);
3208 static sector_t get_dev_size(struct block_device *bdev)
3210 return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
3213 static void warn_if_metadata_device_too_big(struct block_device *bdev)
3215 sector_t metadata_dev_size = get_dev_size(bdev);
3216 char buffer[BDEVNAME_SIZE];
3218 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING)
3219 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
3220 bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS);
3223 static sector_t get_metadata_dev_size(struct block_device *bdev)
3225 sector_t metadata_dev_size = get_dev_size(bdev);
3227 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS)
3228 metadata_dev_size = THIN_METADATA_MAX_SECTORS;
3230 return metadata_dev_size;
3233 static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
3235 sector_t metadata_dev_size = get_metadata_dev_size(bdev);
3237 sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE);
3239 return metadata_dev_size;
3243 * When a metadata threshold is crossed a dm event is triggered, and
3244 * userland should respond by growing the metadata device. We could let
3245 * userland set the threshold, like we do with the data threshold, but I'm
3246 * not sure they know enough to do this well.
3248 static dm_block_t calc_metadata_threshold(struct pool_c *pt)
3251 * 4M is ample for all ops with the possible exception of thin
3252 * device deletion which is harmless if it fails (just retry the
3253 * delete after you've grown the device).
3255 dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
3256 return min((dm_block_t)1024ULL /* 4M */, quarter);
3260 * thin-pool <metadata dev> <data dev>
3261 * <data block size (sectors)>
3262 * <low water mark (blocks)>
3263 * [<#feature args> [<arg>]*]
3265 * Optional feature arguments are:
3266 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
3267 * ignore_discard: disable discard
3268 * no_discard_passdown: don't pass discards down to the data device
3269 * read_only: Don't allow any changes to be made to the pool metadata.
3270 * error_if_no_space: error IOs, instead of queueing, if no space.
3272 static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
3274 int r, pool_created = 0;
3277 struct pool_features pf;
3278 struct dm_arg_set as;
3279 struct dm_dev *data_dev;
3280 unsigned long block_size;
3281 dm_block_t low_water_blocks;
3282 struct dm_dev *metadata_dev;
3283 fmode_t metadata_mode;
3286 * FIXME Remove validation from scope of lock.
3288 mutex_lock(&dm_thin_pool_table.mutex);
3291 ti->error = "Invalid argument count";
3299 /* make sure metadata and data are different devices */
3300 if (!strcmp(argv[0], argv[1])) {
3301 ti->error = "Error setting metadata or data device";
3307 * Set default pool features.
3309 pool_features_init(&pf);
3311 dm_consume_args(&as, 4);
3312 r = parse_pool_features(&as, &pf, ti);
3316 metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
3317 r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
3319 ti->error = "Error opening metadata block device";
3322 warn_if_metadata_device_too_big(metadata_dev->bdev);
3324 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
3326 ti->error = "Error getting data device";
3330 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
3331 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
3332 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
3333 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
3334 ti->error = "Invalid block size";
3339 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
3340 ti->error = "Invalid low water mark";
3345 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
3351 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
3352 block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
3359 * 'pool_created' reflects whether this is the first table load.
3360 * Top level discard support is not allowed to be changed after
3361 * initial load. This would require a pool reload to trigger thin
3364 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
3365 ti->error = "Discard support cannot be disabled once enabled";
3367 goto out_flags_changed;
3372 pt->metadata_dev = metadata_dev;
3373 pt->data_dev = data_dev;
3374 pt->low_water_blocks = low_water_blocks;
3375 pt->adjusted_pf = pt->requested_pf = pf;
3376 ti->num_flush_bios = 1;
3379 * Only need to enable discards if the pool should pass
3380 * them down to the data device. The thin device's discard
3381 * processing will cause mappings to be removed from the btree.
3383 if (pf.discard_enabled && pf.discard_passdown) {
3384 ti->num_discard_bios = 1;
3387 * Setting 'discards_supported' circumvents the normal
3388 * stacking of discard limits (this keeps the pool and
3389 * thin devices' discard limits consistent).
3391 ti->discards_supported = true;
3395 r = dm_pool_register_metadata_threshold(pt->pool->pmd,
3396 calc_metadata_threshold(pt),
3397 metadata_low_callback,
3400 goto out_flags_changed;
3402 pt->callbacks.congested_fn = pool_is_congested;
3403 dm_table_add_target_callbacks(ti->table, &pt->callbacks);
3405 mutex_unlock(&dm_thin_pool_table.mutex);
3414 dm_put_device(ti, data_dev);
3416 dm_put_device(ti, metadata_dev);
3418 mutex_unlock(&dm_thin_pool_table.mutex);
3423 static int pool_map(struct dm_target *ti, struct bio *bio)
3426 struct pool_c *pt = ti->private;
3427 struct pool *pool = pt->pool;
3428 unsigned long flags;
3431 * As this is a singleton target, ti->begin is always zero.
3433 spin_lock_irqsave(&pool->lock, flags);
3434 bio_set_dev(bio, pt->data_dev->bdev);
3435 r = DM_MAPIO_REMAPPED;
3436 spin_unlock_irqrestore(&pool->lock, flags);
3441 static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
3444 struct pool_c *pt = ti->private;
3445 struct pool *pool = pt->pool;
3446 sector_t data_size = ti->len;
3447 dm_block_t sb_data_size;
3449 *need_commit = false;
3451 (void) sector_div(data_size, pool->sectors_per_block);
3453 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
3455 DMERR("%s: failed to retrieve data device size",
3456 dm_device_name(pool->pool_md));
3460 if (data_size < sb_data_size) {
3461 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
3462 dm_device_name(pool->pool_md),
3463 (unsigned long long)data_size, sb_data_size);
3466 } else if (data_size > sb_data_size) {
3467 if (dm_pool_metadata_needs_check(pool->pmd)) {
3468 DMERR("%s: unable to grow the data device until repaired.",
3469 dm_device_name(pool->pool_md));
3474 DMINFO("%s: growing the data device from %llu to %llu blocks",
3475 dm_device_name(pool->pool_md),
3476 sb_data_size, (unsigned long long)data_size);
3477 r = dm_pool_resize_data_dev(pool->pmd, data_size);
3479 metadata_operation_failed(pool, "dm_pool_resize_data_dev", r);
3483 *need_commit = true;
3489 static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
3492 struct pool_c *pt = ti->private;
3493 struct pool *pool = pt->pool;
3494 dm_block_t metadata_dev_size, sb_metadata_dev_size;
3496 *need_commit = false;
3498 metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
3500 r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
3502 DMERR("%s: failed to retrieve metadata device size",
3503 dm_device_name(pool->pool_md));
3507 if (metadata_dev_size < sb_metadata_dev_size) {
3508 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
3509 dm_device_name(pool->pool_md),
3510 metadata_dev_size, sb_metadata_dev_size);
3513 } else if (metadata_dev_size > sb_metadata_dev_size) {
3514 if (dm_pool_metadata_needs_check(pool->pmd)) {
3515 DMERR("%s: unable to grow the metadata device until repaired.",
3516 dm_device_name(pool->pool_md));
3520 warn_if_metadata_device_too_big(pool->md_dev);
3521 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
3522 dm_device_name(pool->pool_md),
3523 sb_metadata_dev_size, metadata_dev_size);
3525 if (get_pool_mode(pool) == PM_OUT_OF_METADATA_SPACE)
3526 set_pool_mode(pool, PM_WRITE);
3528 r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
3530 metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r);
3534 *need_commit = true;
3541 * Retrieves the number of blocks of the data device from
3542 * the superblock and compares it to the actual device size,
3543 * thus resizing the data device in case it has grown.
3545 * This both copes with opening preallocated data devices in the ctr
3546 * being followed by a resume
3548 * calling the resume method individually after userspace has
3549 * grown the data device in reaction to a table event.
3551 static int pool_preresume(struct dm_target *ti)
3554 bool need_commit1, need_commit2;
3555 struct pool_c *pt = ti->private;
3556 struct pool *pool = pt->pool;
3559 * Take control of the pool object.
3561 r = bind_control_target(pool, ti);
3565 r = maybe_resize_data_dev(ti, &need_commit1);
3569 r = maybe_resize_metadata_dev(ti, &need_commit2);
3573 if (need_commit1 || need_commit2)
3574 (void) commit(pool);
3579 static void pool_suspend_active_thins(struct pool *pool)
3583 /* Suspend all active thin devices */
3584 tc = get_first_thin(pool);
3586 dm_internal_suspend_noflush(tc->thin_md);
3587 tc = get_next_thin(pool, tc);
3591 static void pool_resume_active_thins(struct pool *pool)
3595 /* Resume all active thin devices */
3596 tc = get_first_thin(pool);
3598 dm_internal_resume(tc->thin_md);
3599 tc = get_next_thin(pool, tc);
3603 static void pool_resume(struct dm_target *ti)
3605 struct pool_c *pt = ti->private;
3606 struct pool *pool = pt->pool;
3607 unsigned long flags;
3610 * Must requeue active_thins' bios and then resume
3611 * active_thins _before_ clearing 'suspend' flag.
3614 pool_resume_active_thins(pool);
3616 spin_lock_irqsave(&pool->lock, flags);
3617 pool->low_water_triggered = false;
3618 pool->suspended = false;
3619 spin_unlock_irqrestore(&pool->lock, flags);
3621 do_waker(&pool->waker.work);
3624 static void pool_presuspend(struct dm_target *ti)
3626 struct pool_c *pt = ti->private;
3627 struct pool *pool = pt->pool;
3628 unsigned long flags;
3630 spin_lock_irqsave(&pool->lock, flags);
3631 pool->suspended = true;
3632 spin_unlock_irqrestore(&pool->lock, flags);
3634 pool_suspend_active_thins(pool);
3637 static void pool_presuspend_undo(struct dm_target *ti)
3639 struct pool_c *pt = ti->private;
3640 struct pool *pool = pt->pool;
3641 unsigned long flags;
3643 pool_resume_active_thins(pool);
3645 spin_lock_irqsave(&pool->lock, flags);
3646 pool->suspended = false;
3647 spin_unlock_irqrestore(&pool->lock, flags);
3650 static void pool_postsuspend(struct dm_target *ti)
3652 struct pool_c *pt = ti->private;
3653 struct pool *pool = pt->pool;
3655 cancel_delayed_work_sync(&pool->waker);
3656 cancel_delayed_work_sync(&pool->no_space_timeout);
3657 flush_workqueue(pool->wq);
3658 (void) commit(pool);
3661 static int check_arg_count(unsigned argc, unsigned args_required)
3663 if (argc != args_required) {
3664 DMWARN("Message received with %u arguments instead of %u.",
3665 argc, args_required);
3672 static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
3674 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
3675 *dev_id <= MAX_DEV_ID)
3679 DMWARN("Message received with invalid device id: %s", arg);
3684 static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
3689 r = check_arg_count(argc, 2);
3693 r = read_dev_id(argv[1], &dev_id, 1);
3697 r = dm_pool_create_thin(pool->pmd, dev_id);
3699 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
3707 static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3710 dm_thin_id origin_dev_id;
3713 r = check_arg_count(argc, 3);
3717 r = read_dev_id(argv[1], &dev_id, 1);
3721 r = read_dev_id(argv[2], &origin_dev_id, 1);
3725 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
3727 DMWARN("Creation of new snapshot %s of device %s failed.",
3735 static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
3740 r = check_arg_count(argc, 2);
3744 r = read_dev_id(argv[1], &dev_id, 1);
3748 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
3750 DMWARN("Deletion of thin device %s failed.", argv[1]);
3755 static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
3757 dm_thin_id old_id, new_id;
3760 r = check_arg_count(argc, 3);
3764 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
3765 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
3769 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
3770 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
3774 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
3776 DMWARN("Failed to change transaction id from %s to %s.",
3784 static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3788 r = check_arg_count(argc, 1);
3792 (void) commit(pool);
3794 r = dm_pool_reserve_metadata_snap(pool->pmd);
3796 DMWARN("reserve_metadata_snap message failed.");
3801 static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3805 r = check_arg_count(argc, 1);
3809 r = dm_pool_release_metadata_snap(pool->pmd);
3811 DMWARN("release_metadata_snap message failed.");
3817 * Messages supported:
3818 * create_thin <dev_id>
3819 * create_snap <dev_id> <origin_id>
3821 * set_transaction_id <current_trans_id> <new_trans_id>
3822 * reserve_metadata_snap
3823 * release_metadata_snap
3825 static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
3828 struct pool_c *pt = ti->private;
3829 struct pool *pool = pt->pool;
3831 if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE) {
3832 DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode",
3833 dm_device_name(pool->pool_md));
3837 if (!strcasecmp(argv[0], "create_thin"))
3838 r = process_create_thin_mesg(argc, argv, pool);
3840 else if (!strcasecmp(argv[0], "create_snap"))
3841 r = process_create_snap_mesg(argc, argv, pool);
3843 else if (!strcasecmp(argv[0], "delete"))
3844 r = process_delete_mesg(argc, argv, pool);
3846 else if (!strcasecmp(argv[0], "set_transaction_id"))
3847 r = process_set_transaction_id_mesg(argc, argv, pool);
3849 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
3850 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
3852 else if (!strcasecmp(argv[0], "release_metadata_snap"))
3853 r = process_release_metadata_snap_mesg(argc, argv, pool);
3856 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
3859 (void) commit(pool);
3864 static void emit_flags(struct pool_features *pf, char *result,
3865 unsigned sz, unsigned maxlen)
3867 unsigned count = !pf->zero_new_blocks + !pf->discard_enabled +
3868 !pf->discard_passdown + (pf->mode == PM_READ_ONLY) +
3869 pf->error_if_no_space;
3870 DMEMIT("%u ", count);
3872 if (!pf->zero_new_blocks)
3873 DMEMIT("skip_block_zeroing ");
3875 if (!pf->discard_enabled)
3876 DMEMIT("ignore_discard ");
3878 if (!pf->discard_passdown)
3879 DMEMIT("no_discard_passdown ");
3881 if (pf->mode == PM_READ_ONLY)
3882 DMEMIT("read_only ");
3884 if (pf->error_if_no_space)
3885 DMEMIT("error_if_no_space ");
3890 * <transaction id> <used metadata sectors>/<total metadata sectors>
3891 * <used data sectors>/<total data sectors> <held metadata root>
3892 * <pool mode> <discard config> <no space config> <needs_check>
3894 static void pool_status(struct dm_target *ti, status_type_t type,
3895 unsigned status_flags, char *result, unsigned maxlen)
3899 uint64_t transaction_id;
3900 dm_block_t nr_free_blocks_data;
3901 dm_block_t nr_free_blocks_metadata;
3902 dm_block_t nr_blocks_data;
3903 dm_block_t nr_blocks_metadata;
3904 dm_block_t held_root;
3905 enum pool_mode mode;
3906 char buf[BDEVNAME_SIZE];
3907 char buf2[BDEVNAME_SIZE];
3908 struct pool_c *pt = ti->private;
3909 struct pool *pool = pt->pool;
3912 case STATUSTYPE_INFO:
3913 if (get_pool_mode(pool) == PM_FAIL) {
3918 /* Commit to ensure statistics aren't out-of-date */
3919 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
3920 (void) commit(pool);
3922 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
3924 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
3925 dm_device_name(pool->pool_md), r);
3929 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
3931 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
3932 dm_device_name(pool->pool_md), r);
3936 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
3938 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
3939 dm_device_name(pool->pool_md), r);
3943 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
3945 DMERR("%s: dm_pool_get_free_block_count returned %d",
3946 dm_device_name(pool->pool_md), r);
3950 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
3952 DMERR("%s: dm_pool_get_data_dev_size returned %d",
3953 dm_device_name(pool->pool_md), r);
3957 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
3959 DMERR("%s: dm_pool_get_metadata_snap returned %d",
3960 dm_device_name(pool->pool_md), r);
3964 DMEMIT("%llu %llu/%llu %llu/%llu ",
3965 (unsigned long long)transaction_id,
3966 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
3967 (unsigned long long)nr_blocks_metadata,
3968 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
3969 (unsigned long long)nr_blocks_data);
3972 DMEMIT("%llu ", held_root);
3976 mode = get_pool_mode(pool);
3977 if (mode == PM_OUT_OF_DATA_SPACE)
3978 DMEMIT("out_of_data_space ");
3979 else if (is_read_only_pool_mode(mode))
3984 if (!pool->pf.discard_enabled)
3985 DMEMIT("ignore_discard ");
3986 else if (pool->pf.discard_passdown)
3987 DMEMIT("discard_passdown ");
3989 DMEMIT("no_discard_passdown ");
3991 if (pool->pf.error_if_no_space)
3992 DMEMIT("error_if_no_space ");
3994 DMEMIT("queue_if_no_space ");
3996 if (dm_pool_metadata_needs_check(pool->pmd))
3997 DMEMIT("needs_check ");
4003 case STATUSTYPE_TABLE:
4004 DMEMIT("%s %s %lu %llu ",
4005 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
4006 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
4007 (unsigned long)pool->sectors_per_block,
4008 (unsigned long long)pt->low_water_blocks);
4009 emit_flags(&pt->requested_pf, result, sz, maxlen);
4018 static int pool_iterate_devices(struct dm_target *ti,
4019 iterate_devices_callout_fn fn, void *data)
4021 struct pool_c *pt = ti->private;
4023 return fn(ti, pt->data_dev, 0, ti->len, data);
4026 static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
4028 struct pool_c *pt = ti->private;
4029 struct pool *pool = pt->pool;
4030 sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
4033 * If max_sectors is smaller than pool->sectors_per_block adjust it
4034 * to the highest possible power-of-2 factor of pool->sectors_per_block.
4035 * This is especially beneficial when the pool's data device is a RAID
4036 * device that has a full stripe width that matches pool->sectors_per_block
4037 * -- because even though partial RAID stripe-sized IOs will be issued to a
4038 * single RAID stripe; when aggregated they will end on a full RAID stripe
4039 * boundary.. which avoids additional partial RAID stripe writes cascading
4041 if (limits->max_sectors < pool->sectors_per_block) {
4042 while (!is_factor(pool->sectors_per_block, limits->max_sectors)) {
4043 if ((limits->max_sectors & (limits->max_sectors - 1)) == 0)
4044 limits->max_sectors--;
4045 limits->max_sectors = rounddown_pow_of_two(limits->max_sectors);
4050 * If the system-determined stacked limits are compatible with the
4051 * pool's blocksize (io_opt is a factor) do not override them.
4053 if (io_opt_sectors < pool->sectors_per_block ||
4054 !is_factor(io_opt_sectors, pool->sectors_per_block)) {
4055 if (is_factor(pool->sectors_per_block, limits->max_sectors))
4056 blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT);
4058 blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT);
4059 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
4063 * pt->adjusted_pf is a staging area for the actual features to use.
4064 * They get transferred to the live pool in bind_control_target()
4065 * called from pool_preresume().
4067 if (!pt->adjusted_pf.discard_enabled) {
4069 * Must explicitly disallow stacking discard limits otherwise the
4070 * block layer will stack them if pool's data device has support.
4071 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
4072 * user to see that, so make sure to set all discard limits to 0.
4074 limits->discard_granularity = 0;
4078 disable_passdown_if_not_supported(pt);
4081 * The pool uses the same discard limits as the underlying data
4082 * device. DM core has already set this up.
4086 static struct target_type pool_target = {
4087 .name = "thin-pool",
4088 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
4089 DM_TARGET_IMMUTABLE,
4090 .version = {1, 19, 0},
4091 .module = THIS_MODULE,
4095 .presuspend = pool_presuspend,
4096 .presuspend_undo = pool_presuspend_undo,
4097 .postsuspend = pool_postsuspend,
4098 .preresume = pool_preresume,
4099 .resume = pool_resume,
4100 .message = pool_message,
4101 .status = pool_status,
4102 .iterate_devices = pool_iterate_devices,
4103 .io_hints = pool_io_hints,
4106 /*----------------------------------------------------------------
4107 * Thin target methods
4108 *--------------------------------------------------------------*/
4109 static void thin_get(struct thin_c *tc)
4111 atomic_inc(&tc->refcount);
4114 static void thin_put(struct thin_c *tc)
4116 if (atomic_dec_and_test(&tc->refcount))
4117 complete(&tc->can_destroy);
4120 static void thin_dtr(struct dm_target *ti)
4122 struct thin_c *tc = ti->private;
4123 unsigned long flags;
4125 spin_lock_irqsave(&tc->pool->lock, flags);
4126 list_del_rcu(&tc->list);
4127 spin_unlock_irqrestore(&tc->pool->lock, flags);
4131 wait_for_completion(&tc->can_destroy);
4133 mutex_lock(&dm_thin_pool_table.mutex);
4135 __pool_dec(tc->pool);
4136 dm_pool_close_thin_device(tc->td);
4137 dm_put_device(ti, tc->pool_dev);
4139 dm_put_device(ti, tc->origin_dev);
4142 mutex_unlock(&dm_thin_pool_table.mutex);
4146 * Thin target parameters:
4148 * <pool_dev> <dev_id> [origin_dev]
4150 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
4151 * dev_id: the internal device identifier
4152 * origin_dev: a device external to the pool that should act as the origin
4154 * If the pool device has discards disabled, they get disabled for the thin
4157 static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
4161 struct dm_dev *pool_dev, *origin_dev;
4162 struct mapped_device *pool_md;
4163 unsigned long flags;
4165 mutex_lock(&dm_thin_pool_table.mutex);
4167 if (argc != 2 && argc != 3) {
4168 ti->error = "Invalid argument count";
4173 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
4175 ti->error = "Out of memory";
4179 tc->thin_md = dm_table_get_md(ti->table);
4180 spin_lock_init(&tc->lock);
4181 INIT_LIST_HEAD(&tc->deferred_cells);
4182 bio_list_init(&tc->deferred_bio_list);
4183 bio_list_init(&tc->retry_on_resume_list);
4184 tc->sort_bio_list = RB_ROOT;
4187 if (!strcmp(argv[0], argv[2])) {
4188 ti->error = "Error setting origin device";
4190 goto bad_origin_dev;
4193 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
4195 ti->error = "Error opening origin device";
4196 goto bad_origin_dev;
4198 tc->origin_dev = origin_dev;
4201 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
4203 ti->error = "Error opening pool device";
4206 tc->pool_dev = pool_dev;
4208 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
4209 ti->error = "Invalid device id";
4214 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
4216 ti->error = "Couldn't get pool mapped device";
4221 tc->pool = __pool_table_lookup(pool_md);
4223 ti->error = "Couldn't find pool object";
4225 goto bad_pool_lookup;
4227 __pool_inc(tc->pool);
4229 if (get_pool_mode(tc->pool) == PM_FAIL) {
4230 ti->error = "Couldn't open thin device, Pool is in fail mode";
4235 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
4237 ti->error = "Couldn't open thin internal device";
4241 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
4245 ti->num_flush_bios = 1;
4246 ti->flush_supported = true;
4247 ti->per_io_data_size = sizeof(struct dm_thin_endio_hook);
4249 /* In case the pool supports discards, pass them on. */
4250 if (tc->pool->pf.discard_enabled) {
4251 ti->discards_supported = true;
4252 ti->num_discard_bios = 1;
4253 ti->split_discard_bios = false;
4256 mutex_unlock(&dm_thin_pool_table.mutex);
4258 spin_lock_irqsave(&tc->pool->lock, flags);
4259 if (tc->pool->suspended) {
4260 spin_unlock_irqrestore(&tc->pool->lock, flags);
4261 mutex_lock(&dm_thin_pool_table.mutex); /* reacquire for __pool_dec */
4262 ti->error = "Unable to activate thin device while pool is suspended";
4266 atomic_set(&tc->refcount, 1);
4267 init_completion(&tc->can_destroy);
4268 list_add_tail_rcu(&tc->list, &tc->pool->active_thins);
4269 spin_unlock_irqrestore(&tc->pool->lock, flags);
4271 * This synchronize_rcu() call is needed here otherwise we risk a
4272 * wake_worker() call finding no bios to process (because the newly
4273 * added tc isn't yet visible). So this reduces latency since we
4274 * aren't then dependent on the periodic commit to wake_worker().
4283 dm_pool_close_thin_device(tc->td);
4285 __pool_dec(tc->pool);
4289 dm_put_device(ti, tc->pool_dev);
4292 dm_put_device(ti, tc->origin_dev);
4296 mutex_unlock(&dm_thin_pool_table.mutex);
4301 static int thin_map(struct dm_target *ti, struct bio *bio)
4303 bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
4305 return thin_bio_map(ti, bio);
4308 static int thin_endio(struct dm_target *ti, struct bio *bio,
4311 unsigned long flags;
4312 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
4313 struct list_head work;
4314 struct dm_thin_new_mapping *m, *tmp;
4315 struct pool *pool = h->tc->pool;
4317 if (h->shared_read_entry) {
4318 INIT_LIST_HEAD(&work);
4319 dm_deferred_entry_dec(h->shared_read_entry, &work);
4321 spin_lock_irqsave(&pool->lock, flags);
4322 list_for_each_entry_safe(m, tmp, &work, list) {
4324 __complete_mapping_preparation(m);
4326 spin_unlock_irqrestore(&pool->lock, flags);
4329 if (h->all_io_entry) {
4330 INIT_LIST_HEAD(&work);
4331 dm_deferred_entry_dec(h->all_io_entry, &work);
4332 if (!list_empty(&work)) {
4333 spin_lock_irqsave(&pool->lock, flags);
4334 list_for_each_entry_safe(m, tmp, &work, list)
4335 list_add_tail(&m->list, &pool->prepared_discards);
4336 spin_unlock_irqrestore(&pool->lock, flags);
4342 cell_defer_no_holder(h->tc, h->cell);
4344 return DM_ENDIO_DONE;
4347 static void thin_presuspend(struct dm_target *ti)
4349 struct thin_c *tc = ti->private;
4351 if (dm_noflush_suspending(ti))
4352 noflush_work(tc, do_noflush_start);
4355 static void thin_postsuspend(struct dm_target *ti)
4357 struct thin_c *tc = ti->private;
4360 * The dm_noflush_suspending flag has been cleared by now, so
4361 * unfortunately we must always run this.
4363 noflush_work(tc, do_noflush_stop);
4366 static int thin_preresume(struct dm_target *ti)
4368 struct thin_c *tc = ti->private;
4371 tc->origin_size = get_dev_size(tc->origin_dev->bdev);
4377 * <nr mapped sectors> <highest mapped sector>
4379 static void thin_status(struct dm_target *ti, status_type_t type,
4380 unsigned status_flags, char *result, unsigned maxlen)
4384 dm_block_t mapped, highest;
4385 char buf[BDEVNAME_SIZE];
4386 struct thin_c *tc = ti->private;
4388 if (get_pool_mode(tc->pool) == PM_FAIL) {
4397 case STATUSTYPE_INFO:
4398 r = dm_thin_get_mapped_count(tc->td, &mapped);
4400 DMERR("dm_thin_get_mapped_count returned %d", r);
4404 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
4406 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
4410 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
4412 DMEMIT("%llu", ((highest + 1) *
4413 tc->pool->sectors_per_block) - 1);
4418 case STATUSTYPE_TABLE:
4420 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
4421 (unsigned long) tc->dev_id);
4423 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
4434 static int thin_iterate_devices(struct dm_target *ti,
4435 iterate_devices_callout_fn fn, void *data)
4438 struct thin_c *tc = ti->private;
4439 struct pool *pool = tc->pool;
4442 * We can't call dm_pool_get_data_dev_size() since that blocks. So
4443 * we follow a more convoluted path through to the pool's target.
4446 return 0; /* nothing is bound */
4448 blocks = pool->ti->len;
4449 (void) sector_div(blocks, pool->sectors_per_block);
4451 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
4456 static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits)
4458 struct thin_c *tc = ti->private;
4459 struct pool *pool = tc->pool;
4461 if (!pool->pf.discard_enabled)
4464 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
4465 limits->max_discard_sectors = 2048 * 1024 * 16; /* 16G */
4468 static struct target_type thin_target = {
4470 .version = {1, 19, 0},
4471 .module = THIS_MODULE,
4475 .end_io = thin_endio,
4476 .preresume = thin_preresume,
4477 .presuspend = thin_presuspend,
4478 .postsuspend = thin_postsuspend,
4479 .status = thin_status,
4480 .iterate_devices = thin_iterate_devices,
4481 .io_hints = thin_io_hints,
4484 /*----------------------------------------------------------------*/
4486 static int __init dm_thin_init(void)
4492 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
4493 if (!_new_mapping_cache)
4496 r = dm_register_target(&thin_target);
4498 goto bad_new_mapping_cache;
4500 r = dm_register_target(&pool_target);
4502 goto bad_thin_target;
4507 dm_unregister_target(&thin_target);
4508 bad_new_mapping_cache:
4509 kmem_cache_destroy(_new_mapping_cache);
4514 static void dm_thin_exit(void)
4516 dm_unregister_target(&thin_target);
4517 dm_unregister_target(&pool_target);
4519 kmem_cache_destroy(_new_mapping_cache);
4522 module_init(dm_thin_init);
4523 module_exit(dm_thin_exit);
4525 module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR);
4526 MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds");
4528 MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
4529 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
4530 MODULE_LICENSE("GPL");