1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2011-2012 Red Hat UK.
5 * This file is released under the GPL.
8 #include "dm-thin-metadata.h"
9 #include "dm-bio-prison-v1.h"
12 #include <linux/device-mapper.h>
13 #include <linux/dm-io.h>
14 #include <linux/dm-kcopyd.h>
15 #include <linux/jiffies.h>
16 #include <linux/log2.h>
17 #include <linux/list.h>
18 #include <linux/rculist.h>
19 #include <linux/init.h>
20 #include <linux/module.h>
21 #include <linux/slab.h>
22 #include <linux/vmalloc.h>
23 #include <linux/sort.h>
24 #include <linux/rbtree.h>
26 #define DM_MSG_PREFIX "thin"
31 #define ENDIO_HOOK_POOL_SIZE 1024
32 #define MAPPING_POOL_SIZE 1024
33 #define COMMIT_PERIOD HZ
34 #define NO_SPACE_TIMEOUT_SECS 60
36 static unsigned int no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS;
38 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
39 "A percentage of time allocated for copy on write");
42 * The block size of the device holding pool data must be
43 * between 64KB and 1GB.
45 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
46 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
49 * Device id is restricted to 24 bits.
51 #define MAX_DEV_ID ((1 << 24) - 1)
54 * How do we handle breaking sharing of data blocks?
55 * =================================================
57 * We use a standard copy-on-write btree to store the mappings for the
58 * devices (note I'm talking about copy-on-write of the metadata here, not
59 * the data). When you take an internal snapshot you clone the root node
60 * of the origin btree. After this there is no concept of an origin or a
61 * snapshot. They are just two device trees that happen to point to the
64 * When we get a write in we decide if it's to a shared data block using
65 * some timestamp magic. If it is, we have to break sharing.
67 * Let's say we write to a shared block in what was the origin. The
70 * i) plug io further to this physical block. (see bio_prison code).
72 * ii) quiesce any read io to that shared data block. Obviously
73 * including all devices that share this block. (see dm_deferred_set code)
75 * iii) copy the data block to a newly allocate block. This step can be
76 * missed out if the io covers the block. (schedule_copy).
78 * iv) insert the new mapping into the origin's btree
79 * (process_prepared_mapping). This act of inserting breaks some
80 * sharing of btree nodes between the two devices. Breaking sharing only
81 * effects the btree of that specific device. Btrees for the other
82 * devices that share the block never change. The btree for the origin
83 * device as it was after the last commit is untouched, ie. we're using
84 * persistent data structures in the functional programming sense.
86 * v) unplug io to this physical block, including the io that triggered
87 * the breaking of sharing.
89 * Steps (ii) and (iii) occur in parallel.
91 * The metadata _doesn't_ need to be committed before the io continues. We
92 * get away with this because the io is always written to a _new_ block.
93 * If there's a crash, then:
95 * - The origin mapping will point to the old origin block (the shared
96 * one). This will contain the data as it was before the io that triggered
97 * the breaking of sharing came in.
99 * - The snap mapping still points to the old block. As it would after
102 * The downside of this scheme is the timestamp magic isn't perfect, and
103 * will continue to think that data block in the snapshot device is shared
104 * even after the write to the origin has broken sharing. I suspect data
105 * blocks will typically be shared by many different devices, so we're
106 * breaking sharing n + 1 times, rather than n, where n is the number of
107 * devices that reference this data block. At the moment I think the
108 * benefits far, far outweigh the disadvantages.
111 /*----------------------------------------------------------------*/
121 static void build_key(struct dm_thin_device *td, enum lock_space ls,
122 dm_block_t b, dm_block_t e, struct dm_cell_key *key)
124 key->virtual = (ls == VIRTUAL);
125 key->dev = dm_thin_dev_id(td);
126 key->block_begin = b;
130 static void build_data_key(struct dm_thin_device *td, dm_block_t b,
131 struct dm_cell_key *key)
133 build_key(td, PHYSICAL, b, b + 1llu, key);
136 static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
137 struct dm_cell_key *key)
139 build_key(td, VIRTUAL, b, b + 1llu, key);
142 /*----------------------------------------------------------------*/
144 #define THROTTLE_THRESHOLD (1 * HZ)
147 struct rw_semaphore lock;
148 unsigned long threshold;
149 bool throttle_applied;
152 static void throttle_init(struct throttle *t)
154 init_rwsem(&t->lock);
155 t->throttle_applied = false;
158 static void throttle_work_start(struct throttle *t)
160 t->threshold = jiffies + THROTTLE_THRESHOLD;
163 static void throttle_work_update(struct throttle *t)
165 if (!t->throttle_applied && time_is_before_jiffies(t->threshold)) {
166 down_write(&t->lock);
167 t->throttle_applied = true;
171 static void throttle_work_complete(struct throttle *t)
173 if (t->throttle_applied) {
174 t->throttle_applied = false;
179 static void throttle_lock(struct throttle *t)
184 static void throttle_unlock(struct throttle *t)
189 /*----------------------------------------------------------------*/
192 * A pool device ties together a metadata device and a data device. It
193 * also provides the interface for creating and destroying internal
196 struct dm_thin_new_mapping;
199 * The pool runs in various modes. Ordered in degraded order for comparisons.
202 PM_WRITE, /* metadata may be changed */
203 PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */
206 * Like READ_ONLY, except may switch back to WRITE on metadata resize. Reported as READ_ONLY.
208 PM_OUT_OF_METADATA_SPACE,
209 PM_READ_ONLY, /* metadata may not be changed */
211 PM_FAIL, /* all I/O fails */
214 struct pool_features {
217 bool zero_new_blocks:1;
218 bool discard_enabled:1;
219 bool discard_passdown:1;
220 bool error_if_no_space:1;
224 typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
225 typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell);
226 typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);
228 #define CELL_SORT_ARRAY_SIZE 8192
231 struct list_head list;
232 struct dm_target *ti; /* Only set if a pool target is bound */
234 struct mapped_device *pool_md;
235 struct block_device *data_dev;
236 struct block_device *md_dev;
237 struct dm_pool_metadata *pmd;
239 dm_block_t low_water_blocks;
240 uint32_t sectors_per_block;
241 int sectors_per_block_shift;
243 struct pool_features pf;
244 bool low_water_triggered:1; /* A dm event has been sent */
246 bool out_of_data_space:1;
248 struct dm_bio_prison *prison;
249 struct dm_kcopyd_client *copier;
251 struct work_struct worker;
252 struct workqueue_struct *wq;
253 struct throttle throttle;
254 struct delayed_work waker;
255 struct delayed_work no_space_timeout;
257 unsigned long last_commit_jiffies;
258 unsigned int ref_count;
261 struct bio_list deferred_flush_bios;
262 struct bio_list deferred_flush_completions;
263 struct list_head prepared_mappings;
264 struct list_head prepared_discards;
265 struct list_head prepared_discards_pt2;
266 struct list_head active_thins;
268 struct dm_deferred_set *shared_read_ds;
269 struct dm_deferred_set *all_io_ds;
271 struct dm_thin_new_mapping *next_mapping;
273 process_bio_fn process_bio;
274 process_bio_fn process_discard;
276 process_cell_fn process_cell;
277 process_cell_fn process_discard_cell;
279 process_mapping_fn process_prepared_mapping;
280 process_mapping_fn process_prepared_discard;
281 process_mapping_fn process_prepared_discard_pt2;
283 struct dm_bio_prison_cell **cell_sort_array;
285 mempool_t mapping_pool;
288 static void metadata_operation_failed(struct pool *pool, const char *op, int r);
290 static enum pool_mode get_pool_mode(struct pool *pool)
292 return pool->pf.mode;
295 static void notify_of_pool_mode_change(struct pool *pool)
297 static const char *descs[] = {
304 const char *extra_desc = NULL;
305 enum pool_mode mode = get_pool_mode(pool);
307 if (mode == PM_OUT_OF_DATA_SPACE) {
308 if (!pool->pf.error_if_no_space)
309 extra_desc = " (queue IO)";
311 extra_desc = " (error IO)";
314 dm_table_event(pool->ti->table);
315 DMINFO("%s: switching pool to %s%s mode",
316 dm_device_name(pool->pool_md),
317 descs[(int)mode], extra_desc ? : "");
321 * Target context for a pool.
324 struct dm_target *ti;
326 struct dm_dev *data_dev;
327 struct dm_dev *metadata_dev;
329 dm_block_t low_water_blocks;
330 struct pool_features requested_pf; /* Features requested during table load */
331 struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */
335 * Target context for a thin.
338 struct list_head list;
339 struct dm_dev *pool_dev;
340 struct dm_dev *origin_dev;
341 sector_t origin_size;
345 struct dm_thin_device *td;
346 struct mapped_device *thin_md;
350 struct list_head deferred_cells;
351 struct bio_list deferred_bio_list;
352 struct bio_list retry_on_resume_list;
353 struct rb_root sort_bio_list; /* sorted list of deferred bios */
356 * Ensures the thin is not destroyed until the worker has finished
357 * iterating the active_thins list.
360 struct completion can_destroy;
363 /*----------------------------------------------------------------*/
365 static bool block_size_is_power_of_two(struct pool *pool)
367 return pool->sectors_per_block_shift >= 0;
370 static sector_t block_to_sectors(struct pool *pool, dm_block_t b)
372 return block_size_is_power_of_two(pool) ?
373 (b << pool->sectors_per_block_shift) :
374 (b * pool->sectors_per_block);
377 /*----------------------------------------------------------------*/
381 struct blk_plug plug;
382 struct bio *parent_bio;
386 static void begin_discard(struct discard_op *op, struct thin_c *tc, struct bio *parent)
391 blk_start_plug(&op->plug);
392 op->parent_bio = parent;
396 static int issue_discard(struct discard_op *op, dm_block_t data_b, dm_block_t data_e)
398 struct thin_c *tc = op->tc;
399 sector_t s = block_to_sectors(tc->pool, data_b);
400 sector_t len = block_to_sectors(tc->pool, data_e - data_b);
402 return __blkdev_issue_discard(tc->pool_dev->bdev, s, len, GFP_NOWAIT,
406 static void end_discard(struct discard_op *op, int r)
410 * Even if one of the calls to issue_discard failed, we
411 * need to wait for the chain to complete.
413 bio_chain(op->bio, op->parent_bio);
414 op->bio->bi_opf = REQ_OP_DISCARD;
418 blk_finish_plug(&op->plug);
421 * Even if r is set, there could be sub discards in flight that we
424 if (r && !op->parent_bio->bi_status)
425 op->parent_bio->bi_status = errno_to_blk_status(r);
426 bio_endio(op->parent_bio);
429 /*----------------------------------------------------------------*/
432 * wake_worker() is used when new work is queued and when pool_resume is
433 * ready to continue deferred IO processing.
435 static void wake_worker(struct pool *pool)
437 queue_work(pool->wq, &pool->worker);
440 /*----------------------------------------------------------------*/
442 static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio,
443 struct dm_bio_prison_cell **cell_result)
446 struct dm_bio_prison_cell *cell_prealloc;
449 * Allocate a cell from the prison's mempool.
450 * This might block but it can't fail.
452 cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO);
454 r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result);
457 * We reused an old cell; we can get rid of
460 dm_bio_prison_free_cell(pool->prison, cell_prealloc);
465 static void cell_release(struct pool *pool,
466 struct dm_bio_prison_cell *cell,
467 struct bio_list *bios)
469 dm_cell_release(pool->prison, cell, bios);
470 dm_bio_prison_free_cell(pool->prison, cell);
473 static void cell_visit_release(struct pool *pool,
474 void (*fn)(void *, struct dm_bio_prison_cell *),
476 struct dm_bio_prison_cell *cell)
478 dm_cell_visit_release(pool->prison, fn, context, cell);
479 dm_bio_prison_free_cell(pool->prison, cell);
482 static void cell_release_no_holder(struct pool *pool,
483 struct dm_bio_prison_cell *cell,
484 struct bio_list *bios)
486 dm_cell_release_no_holder(pool->prison, cell, bios);
487 dm_bio_prison_free_cell(pool->prison, cell);
490 static void cell_error_with_code(struct pool *pool,
491 struct dm_bio_prison_cell *cell, blk_status_t error_code)
493 dm_cell_error(pool->prison, cell, error_code);
494 dm_bio_prison_free_cell(pool->prison, cell);
497 static blk_status_t get_pool_io_error_code(struct pool *pool)
499 return pool->out_of_data_space ? BLK_STS_NOSPC : BLK_STS_IOERR;
502 static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell)
504 cell_error_with_code(pool, cell, get_pool_io_error_code(pool));
507 static void cell_success(struct pool *pool, struct dm_bio_prison_cell *cell)
509 cell_error_with_code(pool, cell, 0);
512 static void cell_requeue(struct pool *pool, struct dm_bio_prison_cell *cell)
514 cell_error_with_code(pool, cell, BLK_STS_DM_REQUEUE);
517 /*----------------------------------------------------------------*/
520 * A global list of pools that uses a struct mapped_device as a key.
522 static struct dm_thin_pool_table {
524 struct list_head pools;
525 } dm_thin_pool_table;
527 static void pool_table_init(void)
529 mutex_init(&dm_thin_pool_table.mutex);
530 INIT_LIST_HEAD(&dm_thin_pool_table.pools);
533 static void pool_table_exit(void)
535 mutex_destroy(&dm_thin_pool_table.mutex);
538 static void __pool_table_insert(struct pool *pool)
540 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
541 list_add(&pool->list, &dm_thin_pool_table.pools);
544 static void __pool_table_remove(struct pool *pool)
546 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
547 list_del(&pool->list);
550 static struct pool *__pool_table_lookup(struct mapped_device *md)
552 struct pool *pool = NULL, *tmp;
554 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
556 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
557 if (tmp->pool_md == md) {
566 static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
568 struct pool *pool = NULL, *tmp;
570 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
572 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
573 if (tmp->md_dev == md_dev) {
582 /*----------------------------------------------------------------*/
584 struct dm_thin_endio_hook {
586 struct dm_deferred_entry *shared_read_entry;
587 struct dm_deferred_entry *all_io_entry;
588 struct dm_thin_new_mapping *overwrite_mapping;
589 struct rb_node rb_node;
590 struct dm_bio_prison_cell *cell;
593 static void __merge_bio_list(struct bio_list *bios, struct bio_list *master)
595 bio_list_merge(bios, master);
596 bio_list_init(master);
599 static void error_bio_list(struct bio_list *bios, blk_status_t error)
603 while ((bio = bio_list_pop(bios))) {
604 bio->bi_status = error;
609 static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master,
612 struct bio_list bios;
614 bio_list_init(&bios);
616 spin_lock_irq(&tc->lock);
617 __merge_bio_list(&bios, master);
618 spin_unlock_irq(&tc->lock);
620 error_bio_list(&bios, error);
623 static void requeue_deferred_cells(struct thin_c *tc)
625 struct pool *pool = tc->pool;
626 struct list_head cells;
627 struct dm_bio_prison_cell *cell, *tmp;
629 INIT_LIST_HEAD(&cells);
631 spin_lock_irq(&tc->lock);
632 list_splice_init(&tc->deferred_cells, &cells);
633 spin_unlock_irq(&tc->lock);
635 list_for_each_entry_safe(cell, tmp, &cells, user_list)
636 cell_requeue(pool, cell);
639 static void requeue_io(struct thin_c *tc)
641 struct bio_list bios;
643 bio_list_init(&bios);
645 spin_lock_irq(&tc->lock);
646 __merge_bio_list(&bios, &tc->deferred_bio_list);
647 __merge_bio_list(&bios, &tc->retry_on_resume_list);
648 spin_unlock_irq(&tc->lock);
650 error_bio_list(&bios, BLK_STS_DM_REQUEUE);
651 requeue_deferred_cells(tc);
654 static void error_retry_list_with_code(struct pool *pool, blk_status_t error)
659 list_for_each_entry_rcu(tc, &pool->active_thins, list)
660 error_thin_bio_list(tc, &tc->retry_on_resume_list, error);
664 static void error_retry_list(struct pool *pool)
666 error_retry_list_with_code(pool, get_pool_io_error_code(pool));
670 * This section of code contains the logic for processing a thin device's IO.
671 * Much of the code depends on pool object resources (lists, workqueues, etc)
672 * but most is exclusively called from the thin target rather than the thin-pool
676 static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
678 struct pool *pool = tc->pool;
679 sector_t block_nr = bio->bi_iter.bi_sector;
681 if (block_size_is_power_of_two(pool))
682 block_nr >>= pool->sectors_per_block_shift;
684 (void) sector_div(block_nr, pool->sectors_per_block);
690 * Returns the _complete_ blocks that this bio covers.
692 static void get_bio_block_range(struct thin_c *tc, struct bio *bio,
693 dm_block_t *begin, dm_block_t *end)
695 struct pool *pool = tc->pool;
696 sector_t b = bio->bi_iter.bi_sector;
697 sector_t e = b + (bio->bi_iter.bi_size >> SECTOR_SHIFT);
699 b += pool->sectors_per_block - 1ull; /* so we round up */
701 if (block_size_is_power_of_two(pool)) {
702 b >>= pool->sectors_per_block_shift;
703 e >>= pool->sectors_per_block_shift;
705 (void) sector_div(b, pool->sectors_per_block);
706 (void) sector_div(e, pool->sectors_per_block);
710 /* Can happen if the bio is within a single block. */
717 static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
719 struct pool *pool = tc->pool;
720 sector_t bi_sector = bio->bi_iter.bi_sector;
722 bio_set_dev(bio, tc->pool_dev->bdev);
723 if (block_size_is_power_of_two(pool))
724 bio->bi_iter.bi_sector =
725 (block << pool->sectors_per_block_shift) |
726 (bi_sector & (pool->sectors_per_block - 1));
728 bio->bi_iter.bi_sector = (block * pool->sectors_per_block) +
729 sector_div(bi_sector, pool->sectors_per_block);
732 static void remap_to_origin(struct thin_c *tc, struct bio *bio)
734 bio_set_dev(bio, tc->origin_dev->bdev);
737 static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
739 return op_is_flush(bio->bi_opf) &&
740 dm_thin_changed_this_transaction(tc->td);
743 static void inc_all_io_entry(struct pool *pool, struct bio *bio)
745 struct dm_thin_endio_hook *h;
747 if (bio_op(bio) == REQ_OP_DISCARD)
750 h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
751 h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
754 static void issue(struct thin_c *tc, struct bio *bio)
756 struct pool *pool = tc->pool;
758 if (!bio_triggers_commit(tc, bio)) {
759 dm_submit_bio_remap(bio, NULL);
764 * Complete bio with an error if earlier I/O caused changes to
765 * the metadata that can't be committed e.g, due to I/O errors
766 * on the metadata device.
768 if (dm_thin_aborted_changes(tc->td)) {
774 * Batch together any bios that trigger commits and then issue a
775 * single commit for them in process_deferred_bios().
777 spin_lock_irq(&pool->lock);
778 bio_list_add(&pool->deferred_flush_bios, bio);
779 spin_unlock_irq(&pool->lock);
782 static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
784 remap_to_origin(tc, bio);
788 static void remap_and_issue(struct thin_c *tc, struct bio *bio,
791 remap(tc, bio, block);
795 /*----------------------------------------------------------------*/
798 * Bio endio functions.
800 struct dm_thin_new_mapping {
801 struct list_head list;
807 * Track quiescing, copying and zeroing preparation actions. When this
808 * counter hits zero the block is prepared and can be inserted into the
811 atomic_t prepare_actions;
815 dm_block_t virt_begin, virt_end;
816 dm_block_t data_block;
817 struct dm_bio_prison_cell *cell;
820 * If the bio covers the whole area of a block then we can avoid
821 * zeroing or copying. Instead this bio is hooked. The bio will
822 * still be in the cell, so care has to be taken to avoid issuing
826 bio_end_io_t *saved_bi_end_io;
829 static void __complete_mapping_preparation(struct dm_thin_new_mapping *m)
831 struct pool *pool = m->tc->pool;
833 if (atomic_dec_and_test(&m->prepare_actions)) {
834 list_add_tail(&m->list, &pool->prepared_mappings);
839 static void complete_mapping_preparation(struct dm_thin_new_mapping *m)
842 struct pool *pool = m->tc->pool;
844 spin_lock_irqsave(&pool->lock, flags);
845 __complete_mapping_preparation(m);
846 spin_unlock_irqrestore(&pool->lock, flags);
849 static void copy_complete(int read_err, unsigned long write_err, void *context)
851 struct dm_thin_new_mapping *m = context;
853 m->status = read_err || write_err ? BLK_STS_IOERR : 0;
854 complete_mapping_preparation(m);
857 static void overwrite_endio(struct bio *bio)
859 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
860 struct dm_thin_new_mapping *m = h->overwrite_mapping;
862 bio->bi_end_io = m->saved_bi_end_io;
864 m->status = bio->bi_status;
865 complete_mapping_preparation(m);
868 /*----------------------------------------------------------------*/
875 * Prepared mapping jobs.
879 * This sends the bios in the cell, except the original holder, back
880 * to the deferred_bios list.
882 static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
884 struct pool *pool = tc->pool;
888 spin_lock_irqsave(&tc->lock, flags);
889 cell_release_no_holder(pool, cell, &tc->deferred_bio_list);
890 has_work = !bio_list_empty(&tc->deferred_bio_list);
891 spin_unlock_irqrestore(&tc->lock, flags);
897 static void thin_defer_bio(struct thin_c *tc, struct bio *bio);
901 struct bio_list defer_bios;
902 struct bio_list issue_bios;
905 static void __inc_remap_and_issue_cell(void *context,
906 struct dm_bio_prison_cell *cell)
908 struct remap_info *info = context;
911 while ((bio = bio_list_pop(&cell->bios))) {
912 if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD)
913 bio_list_add(&info->defer_bios, bio);
915 inc_all_io_entry(info->tc->pool, bio);
918 * We can't issue the bios with the bio prison lock
919 * held, so we add them to a list to issue on
920 * return from this function.
922 bio_list_add(&info->issue_bios, bio);
927 static void inc_remap_and_issue_cell(struct thin_c *tc,
928 struct dm_bio_prison_cell *cell,
932 struct remap_info info;
935 bio_list_init(&info.defer_bios);
936 bio_list_init(&info.issue_bios);
939 * We have to be careful to inc any bios we're about to issue
940 * before the cell is released, and avoid a race with new bios
941 * being added to the cell.
943 cell_visit_release(tc->pool, __inc_remap_and_issue_cell,
946 while ((bio = bio_list_pop(&info.defer_bios)))
947 thin_defer_bio(tc, bio);
949 while ((bio = bio_list_pop(&info.issue_bios)))
950 remap_and_issue(info.tc, bio, block);
953 static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
955 cell_error(m->tc->pool, m->cell);
957 mempool_free(m, &m->tc->pool->mapping_pool);
960 static void complete_overwrite_bio(struct thin_c *tc, struct bio *bio)
962 struct pool *pool = tc->pool;
965 * If the bio has the REQ_FUA flag set we must commit the metadata
966 * before signaling its completion.
968 if (!bio_triggers_commit(tc, bio)) {
974 * Complete bio with an error if earlier I/O caused changes to the
975 * metadata that can't be committed, e.g, due to I/O errors on the
978 if (dm_thin_aborted_changes(tc->td)) {
984 * Batch together any bios that trigger commits and then issue a
985 * single commit for them in process_deferred_bios().
987 spin_lock_irq(&pool->lock);
988 bio_list_add(&pool->deferred_flush_completions, bio);
989 spin_unlock_irq(&pool->lock);
992 static void process_prepared_mapping(struct dm_thin_new_mapping *m)
994 struct thin_c *tc = m->tc;
995 struct pool *pool = tc->pool;
996 struct bio *bio = m->bio;
1000 cell_error(pool, m->cell);
1005 * Commit the prepared block into the mapping btree.
1006 * Any I/O for this block arriving after this point will get
1007 * remapped to it directly.
1009 r = dm_thin_insert_block(tc->td, m->virt_begin, m->data_block);
1011 metadata_operation_failed(pool, "dm_thin_insert_block", r);
1012 cell_error(pool, m->cell);
1017 * Release any bios held while the block was being provisioned.
1018 * If we are processing a write bio that completely covers the block,
1019 * we already processed it so can ignore it now when processing
1020 * the bios in the cell.
1023 inc_remap_and_issue_cell(tc, m->cell, m->data_block);
1024 complete_overwrite_bio(tc, bio);
1026 inc_all_io_entry(tc->pool, m->cell->holder);
1027 remap_and_issue(tc, m->cell->holder, m->data_block);
1028 inc_remap_and_issue_cell(tc, m->cell, m->data_block);
1033 mempool_free(m, &pool->mapping_pool);
1036 /*----------------------------------------------------------------*/
1038 static void free_discard_mapping(struct dm_thin_new_mapping *m)
1040 struct thin_c *tc = m->tc;
1043 cell_defer_no_holder(tc, m->cell);
1044 mempool_free(m, &tc->pool->mapping_pool);
1047 static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
1049 bio_io_error(m->bio);
1050 free_discard_mapping(m);
1053 static void process_prepared_discard_success(struct dm_thin_new_mapping *m)
1056 free_discard_mapping(m);
1059 static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m)
1062 struct thin_c *tc = m->tc;
1064 r = dm_thin_remove_range(tc->td, m->cell->key.block_begin, m->cell->key.block_end);
1066 metadata_operation_failed(tc->pool, "dm_thin_remove_range", r);
1067 bio_io_error(m->bio);
1071 cell_defer_no_holder(tc, m->cell);
1072 mempool_free(m, &tc->pool->mapping_pool);
1075 /*----------------------------------------------------------------*/
1077 static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m,
1078 struct bio *discard_parent)
1081 * We've already unmapped this range of blocks, but before we
1082 * passdown we have to check that these blocks are now unused.
1086 struct thin_c *tc = m->tc;
1087 struct pool *pool = tc->pool;
1088 dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin;
1089 struct discard_op op;
1091 begin_discard(&op, tc, discard_parent);
1093 /* find start of unmapped run */
1094 for (; b < end; b++) {
1095 r = dm_pool_block_is_shared(pool->pmd, b, &shared);
1106 /* find end of run */
1107 for (e = b + 1; e != end; e++) {
1108 r = dm_pool_block_is_shared(pool->pmd, e, &shared);
1116 r = issue_discard(&op, b, e);
1123 end_discard(&op, r);
1126 static void queue_passdown_pt2(struct dm_thin_new_mapping *m)
1128 unsigned long flags;
1129 struct pool *pool = m->tc->pool;
1131 spin_lock_irqsave(&pool->lock, flags);
1132 list_add_tail(&m->list, &pool->prepared_discards_pt2);
1133 spin_unlock_irqrestore(&pool->lock, flags);
1137 static void passdown_endio(struct bio *bio)
1140 * It doesn't matter if the passdown discard failed, we still want
1141 * to unmap (we ignore err).
1143 queue_passdown_pt2(bio->bi_private);
1147 static void process_prepared_discard_passdown_pt1(struct dm_thin_new_mapping *m)
1150 struct thin_c *tc = m->tc;
1151 struct pool *pool = tc->pool;
1152 struct bio *discard_parent;
1153 dm_block_t data_end = m->data_block + (m->virt_end - m->virt_begin);
1156 * Only this thread allocates blocks, so we can be sure that the
1157 * newly unmapped blocks will not be allocated before the end of
1160 r = dm_thin_remove_range(tc->td, m->virt_begin, m->virt_end);
1162 metadata_operation_failed(pool, "dm_thin_remove_range", r);
1163 bio_io_error(m->bio);
1164 cell_defer_no_holder(tc, m->cell);
1165 mempool_free(m, &pool->mapping_pool);
1170 * Increment the unmapped blocks. This prevents a race between the
1171 * passdown io and reallocation of freed blocks.
1173 r = dm_pool_inc_data_range(pool->pmd, m->data_block, data_end);
1175 metadata_operation_failed(pool, "dm_pool_inc_data_range", r);
1176 bio_io_error(m->bio);
1177 cell_defer_no_holder(tc, m->cell);
1178 mempool_free(m, &pool->mapping_pool);
1182 discard_parent = bio_alloc(NULL, 1, 0, GFP_NOIO);
1183 discard_parent->bi_end_io = passdown_endio;
1184 discard_parent->bi_private = m;
1185 if (m->maybe_shared)
1186 passdown_double_checking_shared_status(m, discard_parent);
1188 struct discard_op op;
1190 begin_discard(&op, tc, discard_parent);
1191 r = issue_discard(&op, m->data_block, data_end);
1192 end_discard(&op, r);
1196 static void process_prepared_discard_passdown_pt2(struct dm_thin_new_mapping *m)
1199 struct thin_c *tc = m->tc;
1200 struct pool *pool = tc->pool;
1203 * The passdown has completed, so now we can decrement all those
1206 r = dm_pool_dec_data_range(pool->pmd, m->data_block,
1207 m->data_block + (m->virt_end - m->virt_begin));
1209 metadata_operation_failed(pool, "dm_pool_dec_data_range", r);
1210 bio_io_error(m->bio);
1214 cell_defer_no_holder(tc, m->cell);
1215 mempool_free(m, &pool->mapping_pool);
1218 static void process_prepared(struct pool *pool, struct list_head *head,
1219 process_mapping_fn *fn)
1221 struct list_head maps;
1222 struct dm_thin_new_mapping *m, *tmp;
1224 INIT_LIST_HEAD(&maps);
1225 spin_lock_irq(&pool->lock);
1226 list_splice_init(head, &maps);
1227 spin_unlock_irq(&pool->lock);
1229 list_for_each_entry_safe(m, tmp, &maps, list)
1234 * Deferred bio jobs.
1236 static int io_overlaps_block(struct pool *pool, struct bio *bio)
1238 return bio->bi_iter.bi_size ==
1239 (pool->sectors_per_block << SECTOR_SHIFT);
1242 static int io_overwrites_block(struct pool *pool, struct bio *bio)
1244 return (bio_data_dir(bio) == WRITE) &&
1245 io_overlaps_block(pool, bio);
1248 static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
1251 *save = bio->bi_end_io;
1252 bio->bi_end_io = fn;
1255 static int ensure_next_mapping(struct pool *pool)
1257 if (pool->next_mapping)
1260 pool->next_mapping = mempool_alloc(&pool->mapping_pool, GFP_ATOMIC);
1262 return pool->next_mapping ? 0 : -ENOMEM;
1265 static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
1267 struct dm_thin_new_mapping *m = pool->next_mapping;
1269 BUG_ON(!pool->next_mapping);
1271 memset(m, 0, sizeof(struct dm_thin_new_mapping));
1272 INIT_LIST_HEAD(&m->list);
1275 pool->next_mapping = NULL;
1280 static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m,
1281 sector_t begin, sector_t end)
1283 struct dm_io_region to;
1285 to.bdev = tc->pool_dev->bdev;
1287 to.count = end - begin;
1289 dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m);
1292 static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio,
1293 dm_block_t data_begin,
1294 struct dm_thin_new_mapping *m)
1296 struct pool *pool = tc->pool;
1297 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1299 h->overwrite_mapping = m;
1301 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
1302 inc_all_io_entry(pool, bio);
1303 remap_and_issue(tc, bio, data_begin);
1307 * A partial copy also needs to zero the uncopied region.
1309 static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
1310 struct dm_dev *origin, dm_block_t data_origin,
1311 dm_block_t data_dest,
1312 struct dm_bio_prison_cell *cell, struct bio *bio,
1315 struct pool *pool = tc->pool;
1316 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1319 m->virt_begin = virt_block;
1320 m->virt_end = virt_block + 1u;
1321 m->data_block = data_dest;
1325 * quiesce action + copy action + an extra reference held for the
1326 * duration of this function (we may need to inc later for a
1329 atomic_set(&m->prepare_actions, 3);
1331 if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
1332 complete_mapping_preparation(m); /* already quiesced */
1335 * IO to pool_dev remaps to the pool target's data_dev.
1337 * If the whole block of data is being overwritten, we can issue the
1338 * bio immediately. Otherwise we use kcopyd to clone the data first.
1340 if (io_overwrites_block(pool, bio))
1341 remap_and_issue_overwrite(tc, bio, data_dest, m);
1343 struct dm_io_region from, to;
1345 from.bdev = origin->bdev;
1346 from.sector = data_origin * pool->sectors_per_block;
1349 to.bdev = tc->pool_dev->bdev;
1350 to.sector = data_dest * pool->sectors_per_block;
1353 dm_kcopyd_copy(pool->copier, &from, 1, &to,
1354 0, copy_complete, m);
1357 * Do we need to zero a tail region?
1359 if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) {
1360 atomic_inc(&m->prepare_actions);
1362 data_dest * pool->sectors_per_block + len,
1363 (data_dest + 1) * pool->sectors_per_block);
1367 complete_mapping_preparation(m); /* drop our ref */
1370 static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
1371 dm_block_t data_origin, dm_block_t data_dest,
1372 struct dm_bio_prison_cell *cell, struct bio *bio)
1374 schedule_copy(tc, virt_block, tc->pool_dev,
1375 data_origin, data_dest, cell, bio,
1376 tc->pool->sectors_per_block);
1379 static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
1380 dm_block_t data_block, struct dm_bio_prison_cell *cell,
1383 struct pool *pool = tc->pool;
1384 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1386 atomic_set(&m->prepare_actions, 1); /* no need to quiesce */
1388 m->virt_begin = virt_block;
1389 m->virt_end = virt_block + 1u;
1390 m->data_block = data_block;
1394 * If the whole block of data is being overwritten or we are not
1395 * zeroing pre-existing data, we can issue the bio immediately.
1396 * Otherwise we use kcopyd to zero the data first.
1398 if (pool->pf.zero_new_blocks) {
1399 if (io_overwrites_block(pool, bio))
1400 remap_and_issue_overwrite(tc, bio, data_block, m);
1402 ll_zero(tc, m, data_block * pool->sectors_per_block,
1403 (data_block + 1) * pool->sectors_per_block);
1405 process_prepared_mapping(m);
1408 static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
1409 dm_block_t data_dest,
1410 struct dm_bio_prison_cell *cell, struct bio *bio)
1412 struct pool *pool = tc->pool;
1413 sector_t virt_block_begin = virt_block * pool->sectors_per_block;
1414 sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block;
1416 if (virt_block_end <= tc->origin_size)
1417 schedule_copy(tc, virt_block, tc->origin_dev,
1418 virt_block, data_dest, cell, bio,
1419 pool->sectors_per_block);
1421 else if (virt_block_begin < tc->origin_size)
1422 schedule_copy(tc, virt_block, tc->origin_dev,
1423 virt_block, data_dest, cell, bio,
1424 tc->origin_size - virt_block_begin);
1427 schedule_zero(tc, virt_block, data_dest, cell, bio);
1430 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode);
1432 static void requeue_bios(struct pool *pool);
1434 static bool is_read_only_pool_mode(enum pool_mode mode)
1436 return (mode == PM_OUT_OF_METADATA_SPACE || mode == PM_READ_ONLY);
1439 static bool is_read_only(struct pool *pool)
1441 return is_read_only_pool_mode(get_pool_mode(pool));
1444 static void check_for_metadata_space(struct pool *pool)
1447 const char *ooms_reason = NULL;
1450 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free);
1452 ooms_reason = "Could not get free metadata blocks";
1454 ooms_reason = "No free metadata blocks";
1456 if (ooms_reason && !is_read_only(pool)) {
1457 DMERR("%s", ooms_reason);
1458 set_pool_mode(pool, PM_OUT_OF_METADATA_SPACE);
1462 static void check_for_data_space(struct pool *pool)
1467 if (get_pool_mode(pool) != PM_OUT_OF_DATA_SPACE)
1470 r = dm_pool_get_free_block_count(pool->pmd, &nr_free);
1475 set_pool_mode(pool, PM_WRITE);
1481 * A non-zero return indicates read_only or fail_io mode.
1482 * Many callers don't care about the return value.
1484 static int commit(struct pool *pool)
1488 if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE)
1491 r = dm_pool_commit_metadata(pool->pmd);
1493 metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
1495 check_for_metadata_space(pool);
1496 check_for_data_space(pool);
1502 static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks)
1504 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
1505 DMWARN("%s: reached low water mark for data device: sending event.",
1506 dm_device_name(pool->pool_md));
1507 spin_lock_irq(&pool->lock);
1508 pool->low_water_triggered = true;
1509 spin_unlock_irq(&pool->lock);
1510 dm_table_event(pool->ti->table);
1514 static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
1517 dm_block_t free_blocks;
1518 struct pool *pool = tc->pool;
1520 if (WARN_ON(get_pool_mode(pool) != PM_WRITE))
1523 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1525 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1529 check_low_water_mark(pool, free_blocks);
1533 * Try to commit to see if that will free up some
1540 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1542 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1547 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1552 r = dm_pool_alloc_data_block(pool->pmd, result);
1555 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1557 metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
1561 r = dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks);
1563 metadata_operation_failed(pool, "dm_pool_get_free_metadata_block_count", r);
1568 /* Let's commit before we use up the metadata reserve. */
1578 * If we have run out of space, queue bios until the device is
1579 * resumed, presumably after having been reloaded with more space.
1581 static void retry_on_resume(struct bio *bio)
1583 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1584 struct thin_c *tc = h->tc;
1586 spin_lock_irq(&tc->lock);
1587 bio_list_add(&tc->retry_on_resume_list, bio);
1588 spin_unlock_irq(&tc->lock);
1591 static blk_status_t should_error_unserviceable_bio(struct pool *pool)
1593 enum pool_mode m = get_pool_mode(pool);
1597 /* Shouldn't get here */
1598 DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
1599 return BLK_STS_IOERR;
1601 case PM_OUT_OF_DATA_SPACE:
1602 return pool->pf.error_if_no_space ? BLK_STS_NOSPC : 0;
1604 case PM_OUT_OF_METADATA_SPACE:
1607 return BLK_STS_IOERR;
1609 /* Shouldn't get here */
1610 DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1611 return BLK_STS_IOERR;
1615 static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
1617 blk_status_t error = should_error_unserviceable_bio(pool);
1620 bio->bi_status = error;
1623 retry_on_resume(bio);
1626 static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell)
1629 struct bio_list bios;
1632 error = should_error_unserviceable_bio(pool);
1634 cell_error_with_code(pool, cell, error);
1638 bio_list_init(&bios);
1639 cell_release(pool, cell, &bios);
1641 while ((bio = bio_list_pop(&bios)))
1642 retry_on_resume(bio);
1645 static void process_discard_cell_no_passdown(struct thin_c *tc,
1646 struct dm_bio_prison_cell *virt_cell)
1648 struct pool *pool = tc->pool;
1649 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1652 * We don't need to lock the data blocks, since there's no
1653 * passdown. We only lock data blocks for allocation and breaking sharing.
1656 m->virt_begin = virt_cell->key.block_begin;
1657 m->virt_end = virt_cell->key.block_end;
1658 m->cell = virt_cell;
1659 m->bio = virt_cell->holder;
1661 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1662 pool->process_prepared_discard(m);
1665 static void break_up_discard_bio(struct thin_c *tc, dm_block_t begin, dm_block_t end,
1668 struct pool *pool = tc->pool;
1672 struct dm_cell_key data_key;
1673 struct dm_bio_prison_cell *data_cell;
1674 struct dm_thin_new_mapping *m;
1675 dm_block_t virt_begin, virt_end, data_begin;
1677 while (begin != end) {
1678 r = ensure_next_mapping(pool);
1680 /* we did our best */
1683 r = dm_thin_find_mapped_range(tc->td, begin, end, &virt_begin, &virt_end,
1684 &data_begin, &maybe_shared);
1687 * Silently fail, letting any mappings we've
1692 build_key(tc->td, PHYSICAL, data_begin, data_begin + (virt_end - virt_begin), &data_key);
1693 if (bio_detain(tc->pool, &data_key, NULL, &data_cell)) {
1694 /* contention, we'll give up with this range */
1700 * IO may still be going to the destination block. We must
1701 * quiesce before we can do the removal.
1703 m = get_next_mapping(pool);
1705 m->maybe_shared = maybe_shared;
1706 m->virt_begin = virt_begin;
1707 m->virt_end = virt_end;
1708 m->data_block = data_begin;
1709 m->cell = data_cell;
1713 * The parent bio must not complete before sub discard bios are
1714 * chained to it (see end_discard's bio_chain)!
1716 * This per-mapping bi_remaining increment is paired with
1717 * the implicit decrement that occurs via bio_endio() in
1720 bio_inc_remaining(bio);
1721 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1722 pool->process_prepared_discard(m);
1728 static void process_discard_cell_passdown(struct thin_c *tc, struct dm_bio_prison_cell *virt_cell)
1730 struct bio *bio = virt_cell->holder;
1731 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1734 * The virt_cell will only get freed once the origin bio completes.
1735 * This means it will remain locked while all the individual
1736 * passdown bios are in flight.
1738 h->cell = virt_cell;
1739 break_up_discard_bio(tc, virt_cell->key.block_begin, virt_cell->key.block_end, bio);
1742 * We complete the bio now, knowing that the bi_remaining field
1743 * will prevent completion until the sub range discards have
1749 static void process_discard_bio(struct thin_c *tc, struct bio *bio)
1751 dm_block_t begin, end;
1752 struct dm_cell_key virt_key;
1753 struct dm_bio_prison_cell *virt_cell;
1755 get_bio_block_range(tc, bio, &begin, &end);
1758 * The discard covers less than a block.
1764 build_key(tc->td, VIRTUAL, begin, end, &virt_key);
1765 if (bio_detain(tc->pool, &virt_key, bio, &virt_cell))
1767 * Potential starvation issue: We're relying on the
1768 * fs/application being well behaved, and not trying to
1769 * send IO to a region at the same time as discarding it.
1770 * If they do this persistently then it's possible this
1771 * cell will never be granted.
1775 tc->pool->process_discard_cell(tc, virt_cell);
1778 static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
1779 struct dm_cell_key *key,
1780 struct dm_thin_lookup_result *lookup_result,
1781 struct dm_bio_prison_cell *cell)
1784 dm_block_t data_block;
1785 struct pool *pool = tc->pool;
1787 r = alloc_data_block(tc, &data_block);
1790 schedule_internal_copy(tc, block, lookup_result->block,
1791 data_block, cell, bio);
1795 retry_bios_on_resume(pool, cell);
1799 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1801 cell_error(pool, cell);
1806 static void __remap_and_issue_shared_cell(void *context,
1807 struct dm_bio_prison_cell *cell)
1809 struct remap_info *info = context;
1812 while ((bio = bio_list_pop(&cell->bios))) {
1813 if (bio_data_dir(bio) == WRITE || op_is_flush(bio->bi_opf) ||
1814 bio_op(bio) == REQ_OP_DISCARD)
1815 bio_list_add(&info->defer_bios, bio);
1817 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1819 h->shared_read_entry = dm_deferred_entry_inc(info->tc->pool->shared_read_ds);
1820 inc_all_io_entry(info->tc->pool, bio);
1821 bio_list_add(&info->issue_bios, bio);
1826 static void remap_and_issue_shared_cell(struct thin_c *tc,
1827 struct dm_bio_prison_cell *cell,
1831 struct remap_info info;
1834 bio_list_init(&info.defer_bios);
1835 bio_list_init(&info.issue_bios);
1837 cell_visit_release(tc->pool, __remap_and_issue_shared_cell,
1840 while ((bio = bio_list_pop(&info.defer_bios)))
1841 thin_defer_bio(tc, bio);
1843 while ((bio = bio_list_pop(&info.issue_bios)))
1844 remap_and_issue(tc, bio, block);
1847 static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1849 struct dm_thin_lookup_result *lookup_result,
1850 struct dm_bio_prison_cell *virt_cell)
1852 struct dm_bio_prison_cell *data_cell;
1853 struct pool *pool = tc->pool;
1854 struct dm_cell_key key;
1857 * If cell is already occupied, then sharing is already in the process
1858 * of being broken so we have nothing further to do here.
1860 build_data_key(tc->td, lookup_result->block, &key);
1861 if (bio_detain(pool, &key, bio, &data_cell)) {
1862 cell_defer_no_holder(tc, virt_cell);
1866 if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) {
1867 break_sharing(tc, bio, block, &key, lookup_result, data_cell);
1868 cell_defer_no_holder(tc, virt_cell);
1870 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1872 h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
1873 inc_all_io_entry(pool, bio);
1874 remap_and_issue(tc, bio, lookup_result->block);
1876 remap_and_issue_shared_cell(tc, data_cell, lookup_result->block);
1877 remap_and_issue_shared_cell(tc, virt_cell, lookup_result->block);
1881 static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
1882 struct dm_bio_prison_cell *cell)
1885 dm_block_t data_block;
1886 struct pool *pool = tc->pool;
1889 * Remap empty bios (flushes) immediately, without provisioning.
1891 if (!bio->bi_iter.bi_size) {
1892 inc_all_io_entry(pool, bio);
1893 cell_defer_no_holder(tc, cell);
1895 remap_and_issue(tc, bio, 0);
1900 * Fill read bios with zeroes and complete them immediately.
1902 if (bio_data_dir(bio) == READ) {
1904 cell_defer_no_holder(tc, cell);
1909 r = alloc_data_block(tc, &data_block);
1913 schedule_external_copy(tc, block, data_block, cell, bio);
1915 schedule_zero(tc, block, data_block, cell, bio);
1919 retry_bios_on_resume(pool, cell);
1923 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1925 cell_error(pool, cell);
1930 static void process_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
1933 struct pool *pool = tc->pool;
1934 struct bio *bio = cell->holder;
1935 dm_block_t block = get_bio_block(tc, bio);
1936 struct dm_thin_lookup_result lookup_result;
1938 if (tc->requeue_mode) {
1939 cell_requeue(pool, cell);
1943 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1946 if (lookup_result.shared)
1947 process_shared_bio(tc, bio, block, &lookup_result, cell);
1949 inc_all_io_entry(pool, bio);
1950 remap_and_issue(tc, bio, lookup_result.block);
1951 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
1956 if (bio_data_dir(bio) == READ && tc->origin_dev) {
1957 inc_all_io_entry(pool, bio);
1958 cell_defer_no_holder(tc, cell);
1960 if (bio_end_sector(bio) <= tc->origin_size)
1961 remap_to_origin_and_issue(tc, bio);
1963 else if (bio->bi_iter.bi_sector < tc->origin_size) {
1965 bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT;
1966 remap_to_origin_and_issue(tc, bio);
1973 provision_block(tc, bio, block, cell);
1977 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1979 cell_defer_no_holder(tc, cell);
1985 static void process_bio(struct thin_c *tc, struct bio *bio)
1987 struct pool *pool = tc->pool;
1988 dm_block_t block = get_bio_block(tc, bio);
1989 struct dm_bio_prison_cell *cell;
1990 struct dm_cell_key key;
1993 * If cell is already occupied, then the block is already
1994 * being provisioned so we have nothing further to do here.
1996 build_virtual_key(tc->td, block, &key);
1997 if (bio_detain(pool, &key, bio, &cell))
2000 process_cell(tc, cell);
2003 static void __process_bio_read_only(struct thin_c *tc, struct bio *bio,
2004 struct dm_bio_prison_cell *cell)
2007 int rw = bio_data_dir(bio);
2008 dm_block_t block = get_bio_block(tc, bio);
2009 struct dm_thin_lookup_result lookup_result;
2011 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
2014 if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) {
2015 handle_unserviceable_bio(tc->pool, bio);
2017 cell_defer_no_holder(tc, cell);
2019 inc_all_io_entry(tc->pool, bio);
2020 remap_and_issue(tc, bio, lookup_result.block);
2022 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
2028 cell_defer_no_holder(tc, cell);
2030 handle_unserviceable_bio(tc->pool, bio);
2034 if (tc->origin_dev) {
2035 inc_all_io_entry(tc->pool, bio);
2036 remap_to_origin_and_issue(tc, bio);
2045 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
2048 cell_defer_no_holder(tc, cell);
2054 static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
2056 __process_bio_read_only(tc, bio, NULL);
2059 static void process_cell_read_only(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2061 __process_bio_read_only(tc, cell->holder, cell);
2064 static void process_bio_success(struct thin_c *tc, struct bio *bio)
2069 static void process_bio_fail(struct thin_c *tc, struct bio *bio)
2074 static void process_cell_success(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2076 cell_success(tc->pool, cell);
2079 static void process_cell_fail(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2081 cell_error(tc->pool, cell);
2085 * FIXME: should we also commit due to size of transaction, measured in
2088 static int need_commit_due_to_time(struct pool *pool)
2090 return !time_in_range(jiffies, pool->last_commit_jiffies,
2091 pool->last_commit_jiffies + COMMIT_PERIOD);
2094 #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node)
2095 #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook))
2097 static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio)
2099 struct rb_node **rbp, *parent;
2100 struct dm_thin_endio_hook *pbd;
2101 sector_t bi_sector = bio->bi_iter.bi_sector;
2103 rbp = &tc->sort_bio_list.rb_node;
2107 pbd = thin_pbd(parent);
2109 if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector)
2110 rbp = &(*rbp)->rb_left;
2112 rbp = &(*rbp)->rb_right;
2115 pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2116 rb_link_node(&pbd->rb_node, parent, rbp);
2117 rb_insert_color(&pbd->rb_node, &tc->sort_bio_list);
2120 static void __extract_sorted_bios(struct thin_c *tc)
2122 struct rb_node *node;
2123 struct dm_thin_endio_hook *pbd;
2126 for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) {
2127 pbd = thin_pbd(node);
2128 bio = thin_bio(pbd);
2130 bio_list_add(&tc->deferred_bio_list, bio);
2131 rb_erase(&pbd->rb_node, &tc->sort_bio_list);
2134 WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list));
2137 static void __sort_thin_deferred_bios(struct thin_c *tc)
2140 struct bio_list bios;
2142 bio_list_init(&bios);
2143 bio_list_merge(&bios, &tc->deferred_bio_list);
2144 bio_list_init(&tc->deferred_bio_list);
2146 /* Sort deferred_bio_list using rb-tree */
2147 while ((bio = bio_list_pop(&bios)))
2148 __thin_bio_rb_add(tc, bio);
2151 * Transfer the sorted bios in sort_bio_list back to
2152 * deferred_bio_list to allow lockless submission of
2155 __extract_sorted_bios(tc);
2158 static void process_thin_deferred_bios(struct thin_c *tc)
2160 struct pool *pool = tc->pool;
2162 struct bio_list bios;
2163 struct blk_plug plug;
2164 unsigned int count = 0;
2166 if (tc->requeue_mode) {
2167 error_thin_bio_list(tc, &tc->deferred_bio_list,
2168 BLK_STS_DM_REQUEUE);
2172 bio_list_init(&bios);
2174 spin_lock_irq(&tc->lock);
2176 if (bio_list_empty(&tc->deferred_bio_list)) {
2177 spin_unlock_irq(&tc->lock);
2181 __sort_thin_deferred_bios(tc);
2183 bio_list_merge(&bios, &tc->deferred_bio_list);
2184 bio_list_init(&tc->deferred_bio_list);
2186 spin_unlock_irq(&tc->lock);
2188 blk_start_plug(&plug);
2189 while ((bio = bio_list_pop(&bios))) {
2191 * If we've got no free new_mapping structs, and processing
2192 * this bio might require one, we pause until there are some
2193 * prepared mappings to process.
2195 if (ensure_next_mapping(pool)) {
2196 spin_lock_irq(&tc->lock);
2197 bio_list_add(&tc->deferred_bio_list, bio);
2198 bio_list_merge(&tc->deferred_bio_list, &bios);
2199 spin_unlock_irq(&tc->lock);
2203 if (bio_op(bio) == REQ_OP_DISCARD)
2204 pool->process_discard(tc, bio);
2206 pool->process_bio(tc, bio);
2208 if ((count++ & 127) == 0) {
2209 throttle_work_update(&pool->throttle);
2210 dm_pool_issue_prefetches(pool->pmd);
2214 blk_finish_plug(&plug);
2217 static int cmp_cells(const void *lhs, const void *rhs)
2219 struct dm_bio_prison_cell *lhs_cell = *((struct dm_bio_prison_cell **) lhs);
2220 struct dm_bio_prison_cell *rhs_cell = *((struct dm_bio_prison_cell **) rhs);
2222 BUG_ON(!lhs_cell->holder);
2223 BUG_ON(!rhs_cell->holder);
2225 if (lhs_cell->holder->bi_iter.bi_sector < rhs_cell->holder->bi_iter.bi_sector)
2228 if (lhs_cell->holder->bi_iter.bi_sector > rhs_cell->holder->bi_iter.bi_sector)
2234 static unsigned int sort_cells(struct pool *pool, struct list_head *cells)
2236 unsigned int count = 0;
2237 struct dm_bio_prison_cell *cell, *tmp;
2239 list_for_each_entry_safe(cell, tmp, cells, user_list) {
2240 if (count >= CELL_SORT_ARRAY_SIZE)
2243 pool->cell_sort_array[count++] = cell;
2244 list_del(&cell->user_list);
2247 sort(pool->cell_sort_array, count, sizeof(cell), cmp_cells, NULL);
2252 static void process_thin_deferred_cells(struct thin_c *tc)
2254 struct pool *pool = tc->pool;
2255 struct list_head cells;
2256 struct dm_bio_prison_cell *cell;
2257 unsigned int i, j, count;
2259 INIT_LIST_HEAD(&cells);
2261 spin_lock_irq(&tc->lock);
2262 list_splice_init(&tc->deferred_cells, &cells);
2263 spin_unlock_irq(&tc->lock);
2265 if (list_empty(&cells))
2269 count = sort_cells(tc->pool, &cells);
2271 for (i = 0; i < count; i++) {
2272 cell = pool->cell_sort_array[i];
2273 BUG_ON(!cell->holder);
2276 * If we've got no free new_mapping structs, and processing
2277 * this bio might require one, we pause until there are some
2278 * prepared mappings to process.
2280 if (ensure_next_mapping(pool)) {
2281 for (j = i; j < count; j++)
2282 list_add(&pool->cell_sort_array[j]->user_list, &cells);
2284 spin_lock_irq(&tc->lock);
2285 list_splice(&cells, &tc->deferred_cells);
2286 spin_unlock_irq(&tc->lock);
2290 if (bio_op(cell->holder) == REQ_OP_DISCARD)
2291 pool->process_discard_cell(tc, cell);
2293 pool->process_cell(tc, cell);
2296 } while (!list_empty(&cells));
2299 static void thin_get(struct thin_c *tc);
2300 static void thin_put(struct thin_c *tc);
2303 * We can't hold rcu_read_lock() around code that can block. So we
2304 * find a thin with the rcu lock held; bump a refcount; then drop
2307 static struct thin_c *get_first_thin(struct pool *pool)
2309 struct thin_c *tc = NULL;
2312 if (!list_empty(&pool->active_thins)) {
2313 tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list);
2321 static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc)
2323 struct thin_c *old_tc = tc;
2326 list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) {
2338 static void process_deferred_bios(struct pool *pool)
2341 struct bio_list bios, bio_completions;
2344 tc = get_first_thin(pool);
2346 process_thin_deferred_cells(tc);
2347 process_thin_deferred_bios(tc);
2348 tc = get_next_thin(pool, tc);
2352 * If there are any deferred flush bios, we must commit the metadata
2353 * before issuing them or signaling their completion.
2355 bio_list_init(&bios);
2356 bio_list_init(&bio_completions);
2358 spin_lock_irq(&pool->lock);
2359 bio_list_merge(&bios, &pool->deferred_flush_bios);
2360 bio_list_init(&pool->deferred_flush_bios);
2362 bio_list_merge(&bio_completions, &pool->deferred_flush_completions);
2363 bio_list_init(&pool->deferred_flush_completions);
2364 spin_unlock_irq(&pool->lock);
2366 if (bio_list_empty(&bios) && bio_list_empty(&bio_completions) &&
2367 !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool)))
2371 bio_list_merge(&bios, &bio_completions);
2373 while ((bio = bio_list_pop(&bios)))
2377 pool->last_commit_jiffies = jiffies;
2379 while ((bio = bio_list_pop(&bio_completions)))
2382 while ((bio = bio_list_pop(&bios))) {
2384 * The data device was flushed as part of metadata commit,
2385 * so complete redundant flushes immediately.
2387 if (bio->bi_opf & REQ_PREFLUSH)
2390 dm_submit_bio_remap(bio, NULL);
2394 static void do_worker(struct work_struct *ws)
2396 struct pool *pool = container_of(ws, struct pool, worker);
2398 throttle_work_start(&pool->throttle);
2399 dm_pool_issue_prefetches(pool->pmd);
2400 throttle_work_update(&pool->throttle);
2401 process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
2402 throttle_work_update(&pool->throttle);
2403 process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
2404 throttle_work_update(&pool->throttle);
2405 process_prepared(pool, &pool->prepared_discards_pt2, &pool->process_prepared_discard_pt2);
2406 throttle_work_update(&pool->throttle);
2407 process_deferred_bios(pool);
2408 throttle_work_complete(&pool->throttle);
2412 * We want to commit periodically so that not too much
2413 * unwritten data builds up.
2415 static void do_waker(struct work_struct *ws)
2417 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
2420 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
2424 * We're holding onto IO to allow userland time to react. After the
2425 * timeout either the pool will have been resized (and thus back in
2426 * PM_WRITE mode), or we degrade to PM_OUT_OF_DATA_SPACE w/ error_if_no_space.
2428 static void do_no_space_timeout(struct work_struct *ws)
2430 struct pool *pool = container_of(to_delayed_work(ws), struct pool,
2433 if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) {
2434 pool->pf.error_if_no_space = true;
2435 notify_of_pool_mode_change(pool);
2436 error_retry_list_with_code(pool, BLK_STS_NOSPC);
2440 /*----------------------------------------------------------------*/
2443 struct work_struct worker;
2444 struct completion complete;
2447 static struct pool_work *to_pool_work(struct work_struct *ws)
2449 return container_of(ws, struct pool_work, worker);
2452 static void pool_work_complete(struct pool_work *pw)
2454 complete(&pw->complete);
2457 static void pool_work_wait(struct pool_work *pw, struct pool *pool,
2458 void (*fn)(struct work_struct *))
2460 INIT_WORK_ONSTACK(&pw->worker, fn);
2461 init_completion(&pw->complete);
2462 queue_work(pool->wq, &pw->worker);
2463 wait_for_completion(&pw->complete);
2466 /*----------------------------------------------------------------*/
2468 struct noflush_work {
2469 struct pool_work pw;
2473 static struct noflush_work *to_noflush(struct work_struct *ws)
2475 return container_of(to_pool_work(ws), struct noflush_work, pw);
2478 static void do_noflush_start(struct work_struct *ws)
2480 struct noflush_work *w = to_noflush(ws);
2482 w->tc->requeue_mode = true;
2484 pool_work_complete(&w->pw);
2487 static void do_noflush_stop(struct work_struct *ws)
2489 struct noflush_work *w = to_noflush(ws);
2491 w->tc->requeue_mode = false;
2492 pool_work_complete(&w->pw);
2495 static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *))
2497 struct noflush_work w;
2500 pool_work_wait(&w.pw, tc->pool, fn);
2503 /*----------------------------------------------------------------*/
2505 static bool passdown_enabled(struct pool_c *pt)
2507 return pt->adjusted_pf.discard_passdown;
2510 static void set_discard_callbacks(struct pool *pool)
2512 struct pool_c *pt = pool->ti->private;
2514 if (passdown_enabled(pt)) {
2515 pool->process_discard_cell = process_discard_cell_passdown;
2516 pool->process_prepared_discard = process_prepared_discard_passdown_pt1;
2517 pool->process_prepared_discard_pt2 = process_prepared_discard_passdown_pt2;
2519 pool->process_discard_cell = process_discard_cell_no_passdown;
2520 pool->process_prepared_discard = process_prepared_discard_no_passdown;
2524 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode)
2526 struct pool_c *pt = pool->ti->private;
2527 bool needs_check = dm_pool_metadata_needs_check(pool->pmd);
2528 enum pool_mode old_mode = get_pool_mode(pool);
2529 unsigned long no_space_timeout = READ_ONCE(no_space_timeout_secs) * HZ;
2532 * Never allow the pool to transition to PM_WRITE mode if user
2533 * intervention is required to verify metadata and data consistency.
2535 if (new_mode == PM_WRITE && needs_check) {
2536 DMERR("%s: unable to switch pool to write mode until repaired.",
2537 dm_device_name(pool->pool_md));
2538 if (old_mode != new_mode)
2539 new_mode = old_mode;
2541 new_mode = PM_READ_ONLY;
2544 * If we were in PM_FAIL mode, rollback of metadata failed. We're
2545 * not going to recover without a thin_repair. So we never let the
2546 * pool move out of the old mode.
2548 if (old_mode == PM_FAIL)
2549 new_mode = old_mode;
2553 dm_pool_metadata_read_only(pool->pmd);
2554 pool->process_bio = process_bio_fail;
2555 pool->process_discard = process_bio_fail;
2556 pool->process_cell = process_cell_fail;
2557 pool->process_discard_cell = process_cell_fail;
2558 pool->process_prepared_mapping = process_prepared_mapping_fail;
2559 pool->process_prepared_discard = process_prepared_discard_fail;
2561 error_retry_list(pool);
2564 case PM_OUT_OF_METADATA_SPACE:
2566 dm_pool_metadata_read_only(pool->pmd);
2567 pool->process_bio = process_bio_read_only;
2568 pool->process_discard = process_bio_success;
2569 pool->process_cell = process_cell_read_only;
2570 pool->process_discard_cell = process_cell_success;
2571 pool->process_prepared_mapping = process_prepared_mapping_fail;
2572 pool->process_prepared_discard = process_prepared_discard_success;
2574 error_retry_list(pool);
2577 case PM_OUT_OF_DATA_SPACE:
2579 * Ideally we'd never hit this state; the low water mark
2580 * would trigger userland to extend the pool before we
2581 * completely run out of data space. However, many small
2582 * IOs to unprovisioned space can consume data space at an
2583 * alarming rate. Adjust your low water mark if you're
2584 * frequently seeing this mode.
2586 pool->out_of_data_space = true;
2587 pool->process_bio = process_bio_read_only;
2588 pool->process_discard = process_discard_bio;
2589 pool->process_cell = process_cell_read_only;
2590 pool->process_prepared_mapping = process_prepared_mapping;
2591 set_discard_callbacks(pool);
2593 if (!pool->pf.error_if_no_space && no_space_timeout)
2594 queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout);
2598 if (old_mode == PM_OUT_OF_DATA_SPACE)
2599 cancel_delayed_work_sync(&pool->no_space_timeout);
2600 pool->out_of_data_space = false;
2601 pool->pf.error_if_no_space = pt->requested_pf.error_if_no_space;
2602 dm_pool_metadata_read_write(pool->pmd);
2603 pool->process_bio = process_bio;
2604 pool->process_discard = process_discard_bio;
2605 pool->process_cell = process_cell;
2606 pool->process_prepared_mapping = process_prepared_mapping;
2607 set_discard_callbacks(pool);
2611 pool->pf.mode = new_mode;
2613 * The pool mode may have changed, sync it so bind_control_target()
2614 * doesn't cause an unexpected mode transition on resume.
2616 pt->adjusted_pf.mode = new_mode;
2618 if (old_mode != new_mode)
2619 notify_of_pool_mode_change(pool);
2622 static void abort_transaction(struct pool *pool)
2624 const char *dev_name = dm_device_name(pool->pool_md);
2626 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name);
2627 if (dm_pool_abort_metadata(pool->pmd)) {
2628 DMERR("%s: failed to abort metadata transaction", dev_name);
2629 set_pool_mode(pool, PM_FAIL);
2632 if (dm_pool_metadata_set_needs_check(pool->pmd)) {
2633 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name);
2634 set_pool_mode(pool, PM_FAIL);
2638 static void metadata_operation_failed(struct pool *pool, const char *op, int r)
2640 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
2641 dm_device_name(pool->pool_md), op, r);
2643 abort_transaction(pool);
2644 set_pool_mode(pool, PM_READ_ONLY);
2647 /*----------------------------------------------------------------*/
2650 * Mapping functions.
2654 * Called only while mapping a thin bio to hand it over to the workqueue.
2656 static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
2658 struct pool *pool = tc->pool;
2660 spin_lock_irq(&tc->lock);
2661 bio_list_add(&tc->deferred_bio_list, bio);
2662 spin_unlock_irq(&tc->lock);
2667 static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio)
2669 struct pool *pool = tc->pool;
2671 throttle_lock(&pool->throttle);
2672 thin_defer_bio(tc, bio);
2673 throttle_unlock(&pool->throttle);
2676 static void thin_defer_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2678 struct pool *pool = tc->pool;
2680 throttle_lock(&pool->throttle);
2681 spin_lock_irq(&tc->lock);
2682 list_add_tail(&cell->user_list, &tc->deferred_cells);
2683 spin_unlock_irq(&tc->lock);
2684 throttle_unlock(&pool->throttle);
2689 static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
2691 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2694 h->shared_read_entry = NULL;
2695 h->all_io_entry = NULL;
2696 h->overwrite_mapping = NULL;
2701 * Non-blocking function called from the thin target's map function.
2703 static int thin_bio_map(struct dm_target *ti, struct bio *bio)
2706 struct thin_c *tc = ti->private;
2707 dm_block_t block = get_bio_block(tc, bio);
2708 struct dm_thin_device *td = tc->td;
2709 struct dm_thin_lookup_result result;
2710 struct dm_bio_prison_cell *virt_cell, *data_cell;
2711 struct dm_cell_key key;
2713 thin_hook_bio(tc, bio);
2715 if (tc->requeue_mode) {
2716 bio->bi_status = BLK_STS_DM_REQUEUE;
2718 return DM_MAPIO_SUBMITTED;
2721 if (get_pool_mode(tc->pool) == PM_FAIL) {
2723 return DM_MAPIO_SUBMITTED;
2726 if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD) {
2727 thin_defer_bio_with_throttle(tc, bio);
2728 return DM_MAPIO_SUBMITTED;
2732 * We must hold the virtual cell before doing the lookup, otherwise
2733 * there's a race with discard.
2735 build_virtual_key(tc->td, block, &key);
2736 if (bio_detain(tc->pool, &key, bio, &virt_cell))
2737 return DM_MAPIO_SUBMITTED;
2739 r = dm_thin_find_block(td, block, 0, &result);
2742 * Note that we defer readahead too.
2746 if (unlikely(result.shared)) {
2748 * We have a race condition here between the
2749 * result.shared value returned by the lookup and
2750 * snapshot creation, which may cause new
2753 * To avoid this always quiesce the origin before
2754 * taking the snap. You want to do this anyway to
2755 * ensure a consistent application view
2758 * More distant ancestors are irrelevant. The
2759 * shared flag will be set in their case.
2761 thin_defer_cell(tc, virt_cell);
2762 return DM_MAPIO_SUBMITTED;
2765 build_data_key(tc->td, result.block, &key);
2766 if (bio_detain(tc->pool, &key, bio, &data_cell)) {
2767 cell_defer_no_holder(tc, virt_cell);
2768 return DM_MAPIO_SUBMITTED;
2771 inc_all_io_entry(tc->pool, bio);
2772 cell_defer_no_holder(tc, data_cell);
2773 cell_defer_no_holder(tc, virt_cell);
2775 remap(tc, bio, result.block);
2776 return DM_MAPIO_REMAPPED;
2780 thin_defer_cell(tc, virt_cell);
2781 return DM_MAPIO_SUBMITTED;
2785 * Must always call bio_io_error on failure.
2786 * dm_thin_find_block can fail with -EINVAL if the
2787 * pool is switched to fail-io mode.
2790 cell_defer_no_holder(tc, virt_cell);
2791 return DM_MAPIO_SUBMITTED;
2795 static void requeue_bios(struct pool *pool)
2800 list_for_each_entry_rcu(tc, &pool->active_thins, list) {
2801 spin_lock_irq(&tc->lock);
2802 bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list);
2803 bio_list_init(&tc->retry_on_resume_list);
2804 spin_unlock_irq(&tc->lock);
2810 *--------------------------------------------------------------
2811 * Binding of control targets to a pool object
2812 *--------------------------------------------------------------
2814 static bool is_factor(sector_t block_size, uint32_t n)
2816 return !sector_div(block_size, n);
2820 * If discard_passdown was enabled verify that the data device
2821 * supports discards. Disable discard_passdown if not.
2823 static void disable_passdown_if_not_supported(struct pool_c *pt)
2825 struct pool *pool = pt->pool;
2826 struct block_device *data_bdev = pt->data_dev->bdev;
2827 struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
2828 const char *reason = NULL;
2830 if (!pt->adjusted_pf.discard_passdown)
2833 if (!bdev_max_discard_sectors(pt->data_dev->bdev))
2834 reason = "discard unsupported";
2836 else if (data_limits->max_discard_sectors < pool->sectors_per_block)
2837 reason = "max discard sectors smaller than a block";
2840 DMWARN("Data device (%pg) %s: Disabling discard passdown.", data_bdev, reason);
2841 pt->adjusted_pf.discard_passdown = false;
2845 static int bind_control_target(struct pool *pool, struct dm_target *ti)
2847 struct pool_c *pt = ti->private;
2850 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
2852 enum pool_mode old_mode = get_pool_mode(pool);
2853 enum pool_mode new_mode = pt->adjusted_pf.mode;
2856 * Don't change the pool's mode until set_pool_mode() below.
2857 * Otherwise the pool's process_* function pointers may
2858 * not match the desired pool mode.
2860 pt->adjusted_pf.mode = old_mode;
2863 pool->pf = pt->adjusted_pf;
2864 pool->low_water_blocks = pt->low_water_blocks;
2866 set_pool_mode(pool, new_mode);
2871 static void unbind_control_target(struct pool *pool, struct dm_target *ti)
2878 *--------------------------------------------------------------
2880 *--------------------------------------------------------------
2882 /* Initialize pool features. */
2883 static void pool_features_init(struct pool_features *pf)
2885 pf->mode = PM_WRITE;
2886 pf->zero_new_blocks = true;
2887 pf->discard_enabled = true;
2888 pf->discard_passdown = true;
2889 pf->error_if_no_space = false;
2892 static void __pool_destroy(struct pool *pool)
2894 __pool_table_remove(pool);
2896 vfree(pool->cell_sort_array);
2897 if (dm_pool_metadata_close(pool->pmd) < 0)
2898 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
2900 dm_bio_prison_destroy(pool->prison);
2901 dm_kcopyd_client_destroy(pool->copier);
2903 cancel_delayed_work_sync(&pool->waker);
2904 cancel_delayed_work_sync(&pool->no_space_timeout);
2906 destroy_workqueue(pool->wq);
2908 if (pool->next_mapping)
2909 mempool_free(pool->next_mapping, &pool->mapping_pool);
2910 mempool_exit(&pool->mapping_pool);
2911 dm_deferred_set_destroy(pool->shared_read_ds);
2912 dm_deferred_set_destroy(pool->all_io_ds);
2916 static struct kmem_cache *_new_mapping_cache;
2918 static struct pool *pool_create(struct mapped_device *pool_md,
2919 struct block_device *metadata_dev,
2920 struct block_device *data_dev,
2921 unsigned long block_size,
2922 int read_only, char **error)
2927 struct dm_pool_metadata *pmd;
2928 bool format_device = read_only ? false : true;
2930 pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
2932 *error = "Error creating metadata object";
2933 return (struct pool *)pmd;
2936 pool = kzalloc(sizeof(*pool), GFP_KERNEL);
2938 *error = "Error allocating memory for pool";
2939 err_p = ERR_PTR(-ENOMEM);
2944 pool->sectors_per_block = block_size;
2945 if (block_size & (block_size - 1))
2946 pool->sectors_per_block_shift = -1;
2948 pool->sectors_per_block_shift = __ffs(block_size);
2949 pool->low_water_blocks = 0;
2950 pool_features_init(&pool->pf);
2951 pool->prison = dm_bio_prison_create();
2952 if (!pool->prison) {
2953 *error = "Error creating pool's bio prison";
2954 err_p = ERR_PTR(-ENOMEM);
2958 pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2959 if (IS_ERR(pool->copier)) {
2960 r = PTR_ERR(pool->copier);
2961 *error = "Error creating pool's kcopyd client";
2963 goto bad_kcopyd_client;
2967 * Create singlethreaded workqueue that will service all devices
2968 * that use this metadata.
2970 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
2972 *error = "Error creating pool's workqueue";
2973 err_p = ERR_PTR(-ENOMEM);
2977 throttle_init(&pool->throttle);
2978 INIT_WORK(&pool->worker, do_worker);
2979 INIT_DELAYED_WORK(&pool->waker, do_waker);
2980 INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout);
2981 spin_lock_init(&pool->lock);
2982 bio_list_init(&pool->deferred_flush_bios);
2983 bio_list_init(&pool->deferred_flush_completions);
2984 INIT_LIST_HEAD(&pool->prepared_mappings);
2985 INIT_LIST_HEAD(&pool->prepared_discards);
2986 INIT_LIST_HEAD(&pool->prepared_discards_pt2);
2987 INIT_LIST_HEAD(&pool->active_thins);
2988 pool->low_water_triggered = false;
2989 pool->suspended = true;
2990 pool->out_of_data_space = false;
2992 pool->shared_read_ds = dm_deferred_set_create();
2993 if (!pool->shared_read_ds) {
2994 *error = "Error creating pool's shared read deferred set";
2995 err_p = ERR_PTR(-ENOMEM);
2996 goto bad_shared_read_ds;
2999 pool->all_io_ds = dm_deferred_set_create();
3000 if (!pool->all_io_ds) {
3001 *error = "Error creating pool's all io deferred set";
3002 err_p = ERR_PTR(-ENOMEM);
3006 pool->next_mapping = NULL;
3007 r = mempool_init_slab_pool(&pool->mapping_pool, MAPPING_POOL_SIZE,
3008 _new_mapping_cache);
3010 *error = "Error creating pool's mapping mempool";
3012 goto bad_mapping_pool;
3015 pool->cell_sort_array =
3016 vmalloc(array_size(CELL_SORT_ARRAY_SIZE,
3017 sizeof(*pool->cell_sort_array)));
3018 if (!pool->cell_sort_array) {
3019 *error = "Error allocating cell sort array";
3020 err_p = ERR_PTR(-ENOMEM);
3021 goto bad_sort_array;
3024 pool->ref_count = 1;
3025 pool->last_commit_jiffies = jiffies;
3026 pool->pool_md = pool_md;
3027 pool->md_dev = metadata_dev;
3028 pool->data_dev = data_dev;
3029 __pool_table_insert(pool);
3034 mempool_exit(&pool->mapping_pool);
3036 dm_deferred_set_destroy(pool->all_io_ds);
3038 dm_deferred_set_destroy(pool->shared_read_ds);
3040 destroy_workqueue(pool->wq);
3042 dm_kcopyd_client_destroy(pool->copier);
3044 dm_bio_prison_destroy(pool->prison);
3048 if (dm_pool_metadata_close(pmd))
3049 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
3054 static void __pool_inc(struct pool *pool)
3056 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
3060 static void __pool_dec(struct pool *pool)
3062 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
3063 BUG_ON(!pool->ref_count);
3064 if (!--pool->ref_count)
3065 __pool_destroy(pool);
3068 static struct pool *__pool_find(struct mapped_device *pool_md,
3069 struct block_device *metadata_dev,
3070 struct block_device *data_dev,
3071 unsigned long block_size, int read_only,
3072 char **error, int *created)
3074 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
3077 if (pool->pool_md != pool_md) {
3078 *error = "metadata device already in use by a pool";
3079 return ERR_PTR(-EBUSY);
3081 if (pool->data_dev != data_dev) {
3082 *error = "data device already in use by a pool";
3083 return ERR_PTR(-EBUSY);
3088 pool = __pool_table_lookup(pool_md);
3090 if (pool->md_dev != metadata_dev || pool->data_dev != data_dev) {
3091 *error = "different pool cannot replace a pool";
3092 return ERR_PTR(-EINVAL);
3097 pool = pool_create(pool_md, metadata_dev, data_dev, block_size, read_only, error);
3106 *--------------------------------------------------------------
3107 * Pool target methods
3108 *--------------------------------------------------------------
3110 static void pool_dtr(struct dm_target *ti)
3112 struct pool_c *pt = ti->private;
3114 mutex_lock(&dm_thin_pool_table.mutex);
3116 unbind_control_target(pt->pool, ti);
3117 __pool_dec(pt->pool);
3118 dm_put_device(ti, pt->metadata_dev);
3119 dm_put_device(ti, pt->data_dev);
3122 mutex_unlock(&dm_thin_pool_table.mutex);
3125 static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
3126 struct dm_target *ti)
3130 const char *arg_name;
3132 static const struct dm_arg _args[] = {
3133 {0, 4, "Invalid number of pool feature arguments"},
3137 * No feature arguments supplied.
3142 r = dm_read_arg_group(_args, as, &argc, &ti->error);
3146 while (argc && !r) {
3147 arg_name = dm_shift_arg(as);
3150 if (!strcasecmp(arg_name, "skip_block_zeroing"))
3151 pf->zero_new_blocks = false;
3153 else if (!strcasecmp(arg_name, "ignore_discard"))
3154 pf->discard_enabled = false;
3156 else if (!strcasecmp(arg_name, "no_discard_passdown"))
3157 pf->discard_passdown = false;
3159 else if (!strcasecmp(arg_name, "read_only"))
3160 pf->mode = PM_READ_ONLY;
3162 else if (!strcasecmp(arg_name, "error_if_no_space"))
3163 pf->error_if_no_space = true;
3166 ti->error = "Unrecognised pool feature requested";
3175 static void metadata_low_callback(void *context)
3177 struct pool *pool = context;
3179 DMWARN("%s: reached low water mark for metadata device: sending event.",
3180 dm_device_name(pool->pool_md));
3182 dm_table_event(pool->ti->table);
3186 * We need to flush the data device **before** committing the metadata.
3188 * This ensures that the data blocks of any newly inserted mappings are
3189 * properly written to non-volatile storage and won't be lost in case of a
3192 * Failure to do so can result in data corruption in the case of internal or
3193 * external snapshots and in the case of newly provisioned blocks, when block
3194 * zeroing is enabled.
3196 static int metadata_pre_commit_callback(void *context)
3198 struct pool *pool = context;
3200 return blkdev_issue_flush(pool->data_dev);
3203 static sector_t get_dev_size(struct block_device *bdev)
3205 return bdev_nr_sectors(bdev);
3208 static void warn_if_metadata_device_too_big(struct block_device *bdev)
3210 sector_t metadata_dev_size = get_dev_size(bdev);
3212 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING)
3213 DMWARN("Metadata device %pg is larger than %u sectors: excess space will not be used.",
3214 bdev, THIN_METADATA_MAX_SECTORS);
3217 static sector_t get_metadata_dev_size(struct block_device *bdev)
3219 sector_t metadata_dev_size = get_dev_size(bdev);
3221 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS)
3222 metadata_dev_size = THIN_METADATA_MAX_SECTORS;
3224 return metadata_dev_size;
3227 static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
3229 sector_t metadata_dev_size = get_metadata_dev_size(bdev);
3231 sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE);
3233 return metadata_dev_size;
3237 * When a metadata threshold is crossed a dm event is triggered, and
3238 * userland should respond by growing the metadata device. We could let
3239 * userland set the threshold, like we do with the data threshold, but I'm
3240 * not sure they know enough to do this well.
3242 static dm_block_t calc_metadata_threshold(struct pool_c *pt)
3245 * 4M is ample for all ops with the possible exception of thin
3246 * device deletion which is harmless if it fails (just retry the
3247 * delete after you've grown the device).
3249 dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
3251 return min((dm_block_t)1024ULL /* 4M */, quarter);
3255 * thin-pool <metadata dev> <data dev>
3256 * <data block size (sectors)>
3257 * <low water mark (blocks)>
3258 * [<#feature args> [<arg>]*]
3260 * Optional feature arguments are:
3261 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
3262 * ignore_discard: disable discard
3263 * no_discard_passdown: don't pass discards down to the data device
3264 * read_only: Don't allow any changes to be made to the pool metadata.
3265 * error_if_no_space: error IOs, instead of queueing, if no space.
3267 static int pool_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3269 int r, pool_created = 0;
3272 struct pool_features pf;
3273 struct dm_arg_set as;
3274 struct dm_dev *data_dev;
3275 unsigned long block_size;
3276 dm_block_t low_water_blocks;
3277 struct dm_dev *metadata_dev;
3278 fmode_t metadata_mode;
3281 * FIXME Remove validation from scope of lock.
3283 mutex_lock(&dm_thin_pool_table.mutex);
3286 ti->error = "Invalid argument count";
3294 /* make sure metadata and data are different devices */
3295 if (!strcmp(argv[0], argv[1])) {
3296 ti->error = "Error setting metadata or data device";
3302 * Set default pool features.
3304 pool_features_init(&pf);
3306 dm_consume_args(&as, 4);
3307 r = parse_pool_features(&as, &pf, ti);
3311 metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
3312 r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
3314 ti->error = "Error opening metadata block device";
3317 warn_if_metadata_device_too_big(metadata_dev->bdev);
3319 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
3321 ti->error = "Error getting data device";
3325 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
3326 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
3327 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
3328 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
3329 ti->error = "Invalid block size";
3334 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
3335 ti->error = "Invalid low water mark";
3340 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
3346 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, data_dev->bdev,
3347 block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
3354 * 'pool_created' reflects whether this is the first table load.
3355 * Top level discard support is not allowed to be changed after
3356 * initial load. This would require a pool reload to trigger thin
3359 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
3360 ti->error = "Discard support cannot be disabled once enabled";
3362 goto out_flags_changed;
3367 pt->metadata_dev = metadata_dev;
3368 pt->data_dev = data_dev;
3369 pt->low_water_blocks = low_water_blocks;
3370 pt->adjusted_pf = pt->requested_pf = pf;
3371 ti->num_flush_bios = 1;
3372 ti->limit_swap_bios = true;
3375 * Only need to enable discards if the pool should pass
3376 * them down to the data device. The thin device's discard
3377 * processing will cause mappings to be removed from the btree.
3379 if (pf.discard_enabled && pf.discard_passdown) {
3380 ti->num_discard_bios = 1;
3383 * Setting 'discards_supported' circumvents the normal
3384 * stacking of discard limits (this keeps the pool and
3385 * thin devices' discard limits consistent).
3387 ti->discards_supported = true;
3391 r = dm_pool_register_metadata_threshold(pt->pool->pmd,
3392 calc_metadata_threshold(pt),
3393 metadata_low_callback,
3396 ti->error = "Error registering metadata threshold";
3397 goto out_flags_changed;
3400 dm_pool_register_pre_commit_callback(pool->pmd,
3401 metadata_pre_commit_callback, pool);
3403 mutex_unlock(&dm_thin_pool_table.mutex);
3412 dm_put_device(ti, data_dev);
3414 dm_put_device(ti, metadata_dev);
3416 mutex_unlock(&dm_thin_pool_table.mutex);
3421 static int pool_map(struct dm_target *ti, struct bio *bio)
3424 struct pool_c *pt = ti->private;
3425 struct pool *pool = pt->pool;
3428 * As this is a singleton target, ti->begin is always zero.
3430 spin_lock_irq(&pool->lock);
3431 bio_set_dev(bio, pt->data_dev->bdev);
3432 r = DM_MAPIO_REMAPPED;
3433 spin_unlock_irq(&pool->lock);
3438 static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
3441 struct pool_c *pt = ti->private;
3442 struct pool *pool = pt->pool;
3443 sector_t data_size = ti->len;
3444 dm_block_t sb_data_size;
3446 *need_commit = false;
3448 (void) sector_div(data_size, pool->sectors_per_block);
3450 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
3452 DMERR("%s: failed to retrieve data device size",
3453 dm_device_name(pool->pool_md));
3457 if (data_size < sb_data_size) {
3458 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
3459 dm_device_name(pool->pool_md),
3460 (unsigned long long)data_size, sb_data_size);
3463 } else if (data_size > sb_data_size) {
3464 if (dm_pool_metadata_needs_check(pool->pmd)) {
3465 DMERR("%s: unable to grow the data device until repaired.",
3466 dm_device_name(pool->pool_md));
3471 DMINFO("%s: growing the data device from %llu to %llu blocks",
3472 dm_device_name(pool->pool_md),
3473 sb_data_size, (unsigned long long)data_size);
3474 r = dm_pool_resize_data_dev(pool->pmd, data_size);
3476 metadata_operation_failed(pool, "dm_pool_resize_data_dev", r);
3480 *need_commit = true;
3486 static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
3489 struct pool_c *pt = ti->private;
3490 struct pool *pool = pt->pool;
3491 dm_block_t metadata_dev_size, sb_metadata_dev_size;
3493 *need_commit = false;
3495 metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
3497 r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
3499 DMERR("%s: failed to retrieve metadata device size",
3500 dm_device_name(pool->pool_md));
3504 if (metadata_dev_size < sb_metadata_dev_size) {
3505 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
3506 dm_device_name(pool->pool_md),
3507 metadata_dev_size, sb_metadata_dev_size);
3510 } else if (metadata_dev_size > sb_metadata_dev_size) {
3511 if (dm_pool_metadata_needs_check(pool->pmd)) {
3512 DMERR("%s: unable to grow the metadata device until repaired.",
3513 dm_device_name(pool->pool_md));
3517 warn_if_metadata_device_too_big(pool->md_dev);
3518 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
3519 dm_device_name(pool->pool_md),
3520 sb_metadata_dev_size, metadata_dev_size);
3522 if (get_pool_mode(pool) == PM_OUT_OF_METADATA_SPACE)
3523 set_pool_mode(pool, PM_WRITE);
3525 r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
3527 metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r);
3531 *need_commit = true;
3538 * Retrieves the number of blocks of the data device from
3539 * the superblock and compares it to the actual device size,
3540 * thus resizing the data device in case it has grown.
3542 * This both copes with opening preallocated data devices in the ctr
3543 * being followed by a resume
3545 * calling the resume method individually after userspace has
3546 * grown the data device in reaction to a table event.
3548 static int pool_preresume(struct dm_target *ti)
3551 bool need_commit1, need_commit2;
3552 struct pool_c *pt = ti->private;
3553 struct pool *pool = pt->pool;
3556 * Take control of the pool object.
3558 r = bind_control_target(pool, ti);
3562 r = maybe_resize_data_dev(ti, &need_commit1);
3566 r = maybe_resize_metadata_dev(ti, &need_commit2);
3570 if (need_commit1 || need_commit2)
3571 (void) commit(pool);
3574 * When a thin-pool is PM_FAIL, it cannot be rebuilt if
3575 * bio is in deferred list. Therefore need to return 0
3576 * to allow pool_resume() to flush IO.
3578 if (r && get_pool_mode(pool) == PM_FAIL)
3584 static void pool_suspend_active_thins(struct pool *pool)
3588 /* Suspend all active thin devices */
3589 tc = get_first_thin(pool);
3591 dm_internal_suspend_noflush(tc->thin_md);
3592 tc = get_next_thin(pool, tc);
3596 static void pool_resume_active_thins(struct pool *pool)
3600 /* Resume all active thin devices */
3601 tc = get_first_thin(pool);
3603 dm_internal_resume(tc->thin_md);
3604 tc = get_next_thin(pool, tc);
3608 static void pool_resume(struct dm_target *ti)
3610 struct pool_c *pt = ti->private;
3611 struct pool *pool = pt->pool;
3614 * Must requeue active_thins' bios and then resume
3615 * active_thins _before_ clearing 'suspend' flag.
3618 pool_resume_active_thins(pool);
3620 spin_lock_irq(&pool->lock);
3621 pool->low_water_triggered = false;
3622 pool->suspended = false;
3623 spin_unlock_irq(&pool->lock);
3625 do_waker(&pool->waker.work);
3628 static void pool_presuspend(struct dm_target *ti)
3630 struct pool_c *pt = ti->private;
3631 struct pool *pool = pt->pool;
3633 spin_lock_irq(&pool->lock);
3634 pool->suspended = true;
3635 spin_unlock_irq(&pool->lock);
3637 pool_suspend_active_thins(pool);
3640 static void pool_presuspend_undo(struct dm_target *ti)
3642 struct pool_c *pt = ti->private;
3643 struct pool *pool = pt->pool;
3645 pool_resume_active_thins(pool);
3647 spin_lock_irq(&pool->lock);
3648 pool->suspended = false;
3649 spin_unlock_irq(&pool->lock);
3652 static void pool_postsuspend(struct dm_target *ti)
3654 struct pool_c *pt = ti->private;
3655 struct pool *pool = pt->pool;
3657 cancel_delayed_work_sync(&pool->waker);
3658 cancel_delayed_work_sync(&pool->no_space_timeout);
3659 flush_workqueue(pool->wq);
3660 (void) commit(pool);
3663 static int check_arg_count(unsigned int argc, unsigned int args_required)
3665 if (argc != args_required) {
3666 DMWARN("Message received with %u arguments instead of %u.",
3667 argc, args_required);
3674 static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
3676 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
3677 *dev_id <= MAX_DEV_ID)
3681 DMWARN("Message received with invalid device id: %s", arg);
3686 static int process_create_thin_mesg(unsigned int argc, char **argv, struct pool *pool)
3691 r = check_arg_count(argc, 2);
3695 r = read_dev_id(argv[1], &dev_id, 1);
3699 r = dm_pool_create_thin(pool->pmd, dev_id);
3701 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
3709 static int process_create_snap_mesg(unsigned int argc, char **argv, struct pool *pool)
3712 dm_thin_id origin_dev_id;
3715 r = check_arg_count(argc, 3);
3719 r = read_dev_id(argv[1], &dev_id, 1);
3723 r = read_dev_id(argv[2], &origin_dev_id, 1);
3727 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
3729 DMWARN("Creation of new snapshot %s of device %s failed.",
3737 static int process_delete_mesg(unsigned int argc, char **argv, struct pool *pool)
3742 r = check_arg_count(argc, 2);
3746 r = read_dev_id(argv[1], &dev_id, 1);
3750 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
3752 DMWARN("Deletion of thin device %s failed.", argv[1]);
3757 static int process_set_transaction_id_mesg(unsigned int argc, char **argv, struct pool *pool)
3759 dm_thin_id old_id, new_id;
3762 r = check_arg_count(argc, 3);
3766 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
3767 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
3771 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
3772 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
3776 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
3778 DMWARN("Failed to change transaction id from %s to %s.",
3786 static int process_reserve_metadata_snap_mesg(unsigned int argc, char **argv, struct pool *pool)
3790 r = check_arg_count(argc, 1);
3794 (void) commit(pool);
3796 r = dm_pool_reserve_metadata_snap(pool->pmd);
3798 DMWARN("reserve_metadata_snap message failed.");
3803 static int process_release_metadata_snap_mesg(unsigned int argc, char **argv, struct pool *pool)
3807 r = check_arg_count(argc, 1);
3811 r = dm_pool_release_metadata_snap(pool->pmd);
3813 DMWARN("release_metadata_snap message failed.");
3819 * Messages supported:
3820 * create_thin <dev_id>
3821 * create_snap <dev_id> <origin_id>
3823 * set_transaction_id <current_trans_id> <new_trans_id>
3824 * reserve_metadata_snap
3825 * release_metadata_snap
3827 static int pool_message(struct dm_target *ti, unsigned int argc, char **argv,
3828 char *result, unsigned int maxlen)
3831 struct pool_c *pt = ti->private;
3832 struct pool *pool = pt->pool;
3834 if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE) {
3835 DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode",
3836 dm_device_name(pool->pool_md));
3840 if (!strcasecmp(argv[0], "create_thin"))
3841 r = process_create_thin_mesg(argc, argv, pool);
3843 else if (!strcasecmp(argv[0], "create_snap"))
3844 r = process_create_snap_mesg(argc, argv, pool);
3846 else if (!strcasecmp(argv[0], "delete"))
3847 r = process_delete_mesg(argc, argv, pool);
3849 else if (!strcasecmp(argv[0], "set_transaction_id"))
3850 r = process_set_transaction_id_mesg(argc, argv, pool);
3852 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
3853 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
3855 else if (!strcasecmp(argv[0], "release_metadata_snap"))
3856 r = process_release_metadata_snap_mesg(argc, argv, pool);
3859 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
3862 (void) commit(pool);
3867 static void emit_flags(struct pool_features *pf, char *result,
3868 unsigned int sz, unsigned int maxlen)
3870 unsigned int count = !pf->zero_new_blocks + !pf->discard_enabled +
3871 !pf->discard_passdown + (pf->mode == PM_READ_ONLY) +
3872 pf->error_if_no_space;
3873 DMEMIT("%u ", count);
3875 if (!pf->zero_new_blocks)
3876 DMEMIT("skip_block_zeroing ");
3878 if (!pf->discard_enabled)
3879 DMEMIT("ignore_discard ");
3881 if (!pf->discard_passdown)
3882 DMEMIT("no_discard_passdown ");
3884 if (pf->mode == PM_READ_ONLY)
3885 DMEMIT("read_only ");
3887 if (pf->error_if_no_space)
3888 DMEMIT("error_if_no_space ");
3893 * <transaction id> <used metadata sectors>/<total metadata sectors>
3894 * <used data sectors>/<total data sectors> <held metadata root>
3895 * <pool mode> <discard config> <no space config> <needs_check>
3897 static void pool_status(struct dm_target *ti, status_type_t type,
3898 unsigned int status_flags, char *result, unsigned int maxlen)
3901 unsigned int sz = 0;
3902 uint64_t transaction_id;
3903 dm_block_t nr_free_blocks_data;
3904 dm_block_t nr_free_blocks_metadata;
3905 dm_block_t nr_blocks_data;
3906 dm_block_t nr_blocks_metadata;
3907 dm_block_t held_root;
3908 enum pool_mode mode;
3909 char buf[BDEVNAME_SIZE];
3910 char buf2[BDEVNAME_SIZE];
3911 struct pool_c *pt = ti->private;
3912 struct pool *pool = pt->pool;
3915 case STATUSTYPE_INFO:
3916 if (get_pool_mode(pool) == PM_FAIL) {
3921 /* Commit to ensure statistics aren't out-of-date */
3922 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
3923 (void) commit(pool);
3925 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
3927 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
3928 dm_device_name(pool->pool_md), r);
3932 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
3934 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
3935 dm_device_name(pool->pool_md), r);
3939 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
3941 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
3942 dm_device_name(pool->pool_md), r);
3946 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
3948 DMERR("%s: dm_pool_get_free_block_count returned %d",
3949 dm_device_name(pool->pool_md), r);
3953 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
3955 DMERR("%s: dm_pool_get_data_dev_size returned %d",
3956 dm_device_name(pool->pool_md), r);
3960 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
3962 DMERR("%s: dm_pool_get_metadata_snap returned %d",
3963 dm_device_name(pool->pool_md), r);
3967 DMEMIT("%llu %llu/%llu %llu/%llu ",
3968 (unsigned long long)transaction_id,
3969 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
3970 (unsigned long long)nr_blocks_metadata,
3971 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
3972 (unsigned long long)nr_blocks_data);
3975 DMEMIT("%llu ", held_root);
3979 mode = get_pool_mode(pool);
3980 if (mode == PM_OUT_OF_DATA_SPACE)
3981 DMEMIT("out_of_data_space ");
3982 else if (is_read_only_pool_mode(mode))
3987 if (!pool->pf.discard_enabled)
3988 DMEMIT("ignore_discard ");
3989 else if (pool->pf.discard_passdown)
3990 DMEMIT("discard_passdown ");
3992 DMEMIT("no_discard_passdown ");
3994 if (pool->pf.error_if_no_space)
3995 DMEMIT("error_if_no_space ");
3997 DMEMIT("queue_if_no_space ");
3999 if (dm_pool_metadata_needs_check(pool->pmd))
4000 DMEMIT("needs_check ");
4004 DMEMIT("%llu ", (unsigned long long)calc_metadata_threshold(pt));
4008 case STATUSTYPE_TABLE:
4009 DMEMIT("%s %s %lu %llu ",
4010 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
4011 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
4012 (unsigned long)pool->sectors_per_block,
4013 (unsigned long long)pt->low_water_blocks);
4014 emit_flags(&pt->requested_pf, result, sz, maxlen);
4017 case STATUSTYPE_IMA:
4027 static int pool_iterate_devices(struct dm_target *ti,
4028 iterate_devices_callout_fn fn, void *data)
4030 struct pool_c *pt = ti->private;
4032 return fn(ti, pt->data_dev, 0, ti->len, data);
4035 static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
4037 struct pool_c *pt = ti->private;
4038 struct pool *pool = pt->pool;
4039 sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
4042 * If max_sectors is smaller than pool->sectors_per_block adjust it
4043 * to the highest possible power-of-2 factor of pool->sectors_per_block.
4044 * This is especially beneficial when the pool's data device is a RAID
4045 * device that has a full stripe width that matches pool->sectors_per_block
4046 * -- because even though partial RAID stripe-sized IOs will be issued to a
4047 * single RAID stripe; when aggregated they will end on a full RAID stripe
4048 * boundary.. which avoids additional partial RAID stripe writes cascading
4050 if (limits->max_sectors < pool->sectors_per_block) {
4051 while (!is_factor(pool->sectors_per_block, limits->max_sectors)) {
4052 if ((limits->max_sectors & (limits->max_sectors - 1)) == 0)
4053 limits->max_sectors--;
4054 limits->max_sectors = rounddown_pow_of_two(limits->max_sectors);
4059 * If the system-determined stacked limits are compatible with the
4060 * pool's blocksize (io_opt is a factor) do not override them.
4062 if (io_opt_sectors < pool->sectors_per_block ||
4063 !is_factor(io_opt_sectors, pool->sectors_per_block)) {
4064 if (is_factor(pool->sectors_per_block, limits->max_sectors))
4065 blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT);
4067 blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT);
4068 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
4072 * pt->adjusted_pf is a staging area for the actual features to use.
4073 * They get transferred to the live pool in bind_control_target()
4074 * called from pool_preresume().
4076 if (!pt->adjusted_pf.discard_enabled) {
4078 * Must explicitly disallow stacking discard limits otherwise the
4079 * block layer will stack them if pool's data device has support.
4081 limits->discard_granularity = 0;
4085 disable_passdown_if_not_supported(pt);
4088 * The pool uses the same discard limits as the underlying data
4089 * device. DM core has already set this up.
4093 static struct target_type pool_target = {
4094 .name = "thin-pool",
4095 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
4096 DM_TARGET_IMMUTABLE,
4097 .version = {1, 22, 0},
4098 .module = THIS_MODULE,
4102 .presuspend = pool_presuspend,
4103 .presuspend_undo = pool_presuspend_undo,
4104 .postsuspend = pool_postsuspend,
4105 .preresume = pool_preresume,
4106 .resume = pool_resume,
4107 .message = pool_message,
4108 .status = pool_status,
4109 .iterate_devices = pool_iterate_devices,
4110 .io_hints = pool_io_hints,
4114 *--------------------------------------------------------------
4115 * Thin target methods
4116 *--------------------------------------------------------------
4118 static void thin_get(struct thin_c *tc)
4120 refcount_inc(&tc->refcount);
4123 static void thin_put(struct thin_c *tc)
4125 if (refcount_dec_and_test(&tc->refcount))
4126 complete(&tc->can_destroy);
4129 static void thin_dtr(struct dm_target *ti)
4131 struct thin_c *tc = ti->private;
4133 spin_lock_irq(&tc->pool->lock);
4134 list_del_rcu(&tc->list);
4135 spin_unlock_irq(&tc->pool->lock);
4139 wait_for_completion(&tc->can_destroy);
4141 mutex_lock(&dm_thin_pool_table.mutex);
4143 __pool_dec(tc->pool);
4144 dm_pool_close_thin_device(tc->td);
4145 dm_put_device(ti, tc->pool_dev);
4147 dm_put_device(ti, tc->origin_dev);
4150 mutex_unlock(&dm_thin_pool_table.mutex);
4154 * Thin target parameters:
4156 * <pool_dev> <dev_id> [origin_dev]
4158 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
4159 * dev_id: the internal device identifier
4160 * origin_dev: a device external to the pool that should act as the origin
4162 * If the pool device has discards disabled, they get disabled for the thin
4165 static int thin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
4169 struct dm_dev *pool_dev, *origin_dev;
4170 struct mapped_device *pool_md;
4172 mutex_lock(&dm_thin_pool_table.mutex);
4174 if (argc != 2 && argc != 3) {
4175 ti->error = "Invalid argument count";
4180 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
4182 ti->error = "Out of memory";
4186 tc->thin_md = dm_table_get_md(ti->table);
4187 spin_lock_init(&tc->lock);
4188 INIT_LIST_HEAD(&tc->deferred_cells);
4189 bio_list_init(&tc->deferred_bio_list);
4190 bio_list_init(&tc->retry_on_resume_list);
4191 tc->sort_bio_list = RB_ROOT;
4194 if (!strcmp(argv[0], argv[2])) {
4195 ti->error = "Error setting origin device";
4197 goto bad_origin_dev;
4200 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
4202 ti->error = "Error opening origin device";
4203 goto bad_origin_dev;
4205 tc->origin_dev = origin_dev;
4208 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
4210 ti->error = "Error opening pool device";
4213 tc->pool_dev = pool_dev;
4215 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
4216 ti->error = "Invalid device id";
4221 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
4223 ti->error = "Couldn't get pool mapped device";
4228 tc->pool = __pool_table_lookup(pool_md);
4230 ti->error = "Couldn't find pool object";
4232 goto bad_pool_lookup;
4234 __pool_inc(tc->pool);
4236 if (get_pool_mode(tc->pool) == PM_FAIL) {
4237 ti->error = "Couldn't open thin device, Pool is in fail mode";
4242 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
4244 ti->error = "Couldn't open thin internal device";
4248 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
4252 ti->num_flush_bios = 1;
4253 ti->limit_swap_bios = true;
4254 ti->flush_supported = true;
4255 ti->accounts_remapped_io = true;
4256 ti->per_io_data_size = sizeof(struct dm_thin_endio_hook);
4258 /* In case the pool supports discards, pass them on. */
4259 if (tc->pool->pf.discard_enabled) {
4260 ti->discards_supported = true;
4261 ti->num_discard_bios = 1;
4264 mutex_unlock(&dm_thin_pool_table.mutex);
4266 spin_lock_irq(&tc->pool->lock);
4267 if (tc->pool->suspended) {
4268 spin_unlock_irq(&tc->pool->lock);
4269 mutex_lock(&dm_thin_pool_table.mutex); /* reacquire for __pool_dec */
4270 ti->error = "Unable to activate thin device while pool is suspended";
4274 refcount_set(&tc->refcount, 1);
4275 init_completion(&tc->can_destroy);
4276 list_add_tail_rcu(&tc->list, &tc->pool->active_thins);
4277 spin_unlock_irq(&tc->pool->lock);
4279 * This synchronize_rcu() call is needed here otherwise we risk a
4280 * wake_worker() call finding no bios to process (because the newly
4281 * added tc isn't yet visible). So this reduces latency since we
4282 * aren't then dependent on the periodic commit to wake_worker().
4291 dm_pool_close_thin_device(tc->td);
4293 __pool_dec(tc->pool);
4297 dm_put_device(ti, tc->pool_dev);
4300 dm_put_device(ti, tc->origin_dev);
4304 mutex_unlock(&dm_thin_pool_table.mutex);
4309 static int thin_map(struct dm_target *ti, struct bio *bio)
4311 bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
4313 return thin_bio_map(ti, bio);
4316 static int thin_endio(struct dm_target *ti, struct bio *bio,
4319 unsigned long flags;
4320 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
4321 struct list_head work;
4322 struct dm_thin_new_mapping *m, *tmp;
4323 struct pool *pool = h->tc->pool;
4325 if (h->shared_read_entry) {
4326 INIT_LIST_HEAD(&work);
4327 dm_deferred_entry_dec(h->shared_read_entry, &work);
4329 spin_lock_irqsave(&pool->lock, flags);
4330 list_for_each_entry_safe(m, tmp, &work, list) {
4332 __complete_mapping_preparation(m);
4334 spin_unlock_irqrestore(&pool->lock, flags);
4337 if (h->all_io_entry) {
4338 INIT_LIST_HEAD(&work);
4339 dm_deferred_entry_dec(h->all_io_entry, &work);
4340 if (!list_empty(&work)) {
4341 spin_lock_irqsave(&pool->lock, flags);
4342 list_for_each_entry_safe(m, tmp, &work, list)
4343 list_add_tail(&m->list, &pool->prepared_discards);
4344 spin_unlock_irqrestore(&pool->lock, flags);
4350 cell_defer_no_holder(h->tc, h->cell);
4352 return DM_ENDIO_DONE;
4355 static void thin_presuspend(struct dm_target *ti)
4357 struct thin_c *tc = ti->private;
4359 if (dm_noflush_suspending(ti))
4360 noflush_work(tc, do_noflush_start);
4363 static void thin_postsuspend(struct dm_target *ti)
4365 struct thin_c *tc = ti->private;
4368 * The dm_noflush_suspending flag has been cleared by now, so
4369 * unfortunately we must always run this.
4371 noflush_work(tc, do_noflush_stop);
4374 static int thin_preresume(struct dm_target *ti)
4376 struct thin_c *tc = ti->private;
4379 tc->origin_size = get_dev_size(tc->origin_dev->bdev);
4385 * <nr mapped sectors> <highest mapped sector>
4387 static void thin_status(struct dm_target *ti, status_type_t type,
4388 unsigned int status_flags, char *result, unsigned int maxlen)
4392 dm_block_t mapped, highest;
4393 char buf[BDEVNAME_SIZE];
4394 struct thin_c *tc = ti->private;
4396 if (get_pool_mode(tc->pool) == PM_FAIL) {
4405 case STATUSTYPE_INFO:
4406 r = dm_thin_get_mapped_count(tc->td, &mapped);
4408 DMERR("dm_thin_get_mapped_count returned %d", r);
4412 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
4414 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
4418 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
4420 DMEMIT("%llu", ((highest + 1) *
4421 tc->pool->sectors_per_block) - 1);
4426 case STATUSTYPE_TABLE:
4428 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
4429 (unsigned long) tc->dev_id);
4431 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
4434 case STATUSTYPE_IMA:
4446 static int thin_iterate_devices(struct dm_target *ti,
4447 iterate_devices_callout_fn fn, void *data)
4450 struct thin_c *tc = ti->private;
4451 struct pool *pool = tc->pool;
4454 * We can't call dm_pool_get_data_dev_size() since that blocks. So
4455 * we follow a more convoluted path through to the pool's target.
4458 return 0; /* nothing is bound */
4460 blocks = pool->ti->len;
4461 (void) sector_div(blocks, pool->sectors_per_block);
4463 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
4468 static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits)
4470 struct thin_c *tc = ti->private;
4471 struct pool *pool = tc->pool;
4473 if (!pool->pf.discard_enabled)
4476 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
4477 limits->max_discard_sectors = 2048 * 1024 * 16; /* 16G */
4480 static struct target_type thin_target = {
4482 .version = {1, 22, 0},
4483 .module = THIS_MODULE,
4487 .end_io = thin_endio,
4488 .preresume = thin_preresume,
4489 .presuspend = thin_presuspend,
4490 .postsuspend = thin_postsuspend,
4491 .status = thin_status,
4492 .iterate_devices = thin_iterate_devices,
4493 .io_hints = thin_io_hints,
4496 /*----------------------------------------------------------------*/
4498 static int __init dm_thin_init(void)
4504 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
4505 if (!_new_mapping_cache)
4508 r = dm_register_target(&thin_target);
4510 goto bad_new_mapping_cache;
4512 r = dm_register_target(&pool_target);
4514 goto bad_thin_target;
4519 dm_unregister_target(&thin_target);
4520 bad_new_mapping_cache:
4521 kmem_cache_destroy(_new_mapping_cache);
4526 static void dm_thin_exit(void)
4528 dm_unregister_target(&thin_target);
4529 dm_unregister_target(&pool_target);
4531 kmem_cache_destroy(_new_mapping_cache);
4536 module_init(dm_thin_init);
4537 module_exit(dm_thin_exit);
4539 module_param_named(no_space_timeout, no_space_timeout_secs, uint, 0644);
4540 MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds");
4542 MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
4543 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
4544 MODULE_LICENSE("GPL");