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 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;
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 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;
1042 cell_defer_no_holder(tc, m->cell);
1043 mempool_free(m, &tc->pool->mapping_pool);
1046 static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
1048 bio_io_error(m->bio);
1049 free_discard_mapping(m);
1052 static void process_prepared_discard_success(struct dm_thin_new_mapping *m)
1055 free_discard_mapping(m);
1058 static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m)
1061 struct thin_c *tc = m->tc;
1063 r = dm_thin_remove_range(tc->td, m->cell->key.block_begin, m->cell->key.block_end);
1065 metadata_operation_failed(tc->pool, "dm_thin_remove_range", r);
1066 bio_io_error(m->bio);
1070 cell_defer_no_holder(tc, m->cell);
1071 mempool_free(m, &tc->pool->mapping_pool);
1074 /*----------------------------------------------------------------*/
1076 static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m,
1077 struct bio *discard_parent)
1080 * We've already unmapped this range of blocks, but before we
1081 * passdown we have to check that these blocks are now unused.
1085 struct thin_c *tc = m->tc;
1086 struct pool *pool = tc->pool;
1087 dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin;
1088 struct discard_op op;
1090 begin_discard(&op, tc, discard_parent);
1092 /* find start of unmapped run */
1093 for (; b < end; b++) {
1094 r = dm_pool_block_is_shared(pool->pmd, b, &shared);
1105 /* find end of run */
1106 for (e = b + 1; e != end; e++) {
1107 r = dm_pool_block_is_shared(pool->pmd, e, &shared);
1115 r = issue_discard(&op, b, e);
1122 end_discard(&op, r);
1125 static void queue_passdown_pt2(struct dm_thin_new_mapping *m)
1127 unsigned long flags;
1128 struct pool *pool = m->tc->pool;
1130 spin_lock_irqsave(&pool->lock, flags);
1131 list_add_tail(&m->list, &pool->prepared_discards_pt2);
1132 spin_unlock_irqrestore(&pool->lock, flags);
1136 static void passdown_endio(struct bio *bio)
1139 * It doesn't matter if the passdown discard failed, we still want
1140 * to unmap (we ignore err).
1142 queue_passdown_pt2(bio->bi_private);
1146 static void process_prepared_discard_passdown_pt1(struct dm_thin_new_mapping *m)
1149 struct thin_c *tc = m->tc;
1150 struct pool *pool = tc->pool;
1151 struct bio *discard_parent;
1152 dm_block_t data_end = m->data_block + (m->virt_end - m->virt_begin);
1155 * Only this thread allocates blocks, so we can be sure that the
1156 * newly unmapped blocks will not be allocated before the end of
1159 r = dm_thin_remove_range(tc->td, m->virt_begin, m->virt_end);
1161 metadata_operation_failed(pool, "dm_thin_remove_range", r);
1162 bio_io_error(m->bio);
1163 cell_defer_no_holder(tc, m->cell);
1164 mempool_free(m, &pool->mapping_pool);
1169 * Increment the unmapped blocks. This prevents a race between the
1170 * passdown io and reallocation of freed blocks.
1172 r = dm_pool_inc_data_range(pool->pmd, m->data_block, data_end);
1174 metadata_operation_failed(pool, "dm_pool_inc_data_range", r);
1175 bio_io_error(m->bio);
1176 cell_defer_no_holder(tc, m->cell);
1177 mempool_free(m, &pool->mapping_pool);
1181 discard_parent = bio_alloc(NULL, 1, 0, GFP_NOIO);
1182 discard_parent->bi_end_io = passdown_endio;
1183 discard_parent->bi_private = m;
1184 if (m->maybe_shared)
1185 passdown_double_checking_shared_status(m, discard_parent);
1187 struct discard_op op;
1189 begin_discard(&op, tc, discard_parent);
1190 r = issue_discard(&op, m->data_block, data_end);
1191 end_discard(&op, r);
1195 static void process_prepared_discard_passdown_pt2(struct dm_thin_new_mapping *m)
1198 struct thin_c *tc = m->tc;
1199 struct pool *pool = tc->pool;
1202 * The passdown has completed, so now we can decrement all those
1205 r = dm_pool_dec_data_range(pool->pmd, m->data_block,
1206 m->data_block + (m->virt_end - m->virt_begin));
1208 metadata_operation_failed(pool, "dm_pool_dec_data_range", r);
1209 bio_io_error(m->bio);
1213 cell_defer_no_holder(tc, m->cell);
1214 mempool_free(m, &pool->mapping_pool);
1217 static void process_prepared(struct pool *pool, struct list_head *head,
1218 process_mapping_fn *fn)
1220 struct list_head maps;
1221 struct dm_thin_new_mapping *m, *tmp;
1223 INIT_LIST_HEAD(&maps);
1224 spin_lock_irq(&pool->lock);
1225 list_splice_init(head, &maps);
1226 spin_unlock_irq(&pool->lock);
1228 list_for_each_entry_safe(m, tmp, &maps, list)
1233 * Deferred bio jobs.
1235 static int io_overlaps_block(struct pool *pool, struct bio *bio)
1237 return bio->bi_iter.bi_size ==
1238 (pool->sectors_per_block << SECTOR_SHIFT);
1241 static int io_overwrites_block(struct pool *pool, struct bio *bio)
1243 return (bio_data_dir(bio) == WRITE) &&
1244 io_overlaps_block(pool, bio);
1247 static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
1250 *save = bio->bi_end_io;
1251 bio->bi_end_io = fn;
1254 static int ensure_next_mapping(struct pool *pool)
1256 if (pool->next_mapping)
1259 pool->next_mapping = mempool_alloc(&pool->mapping_pool, GFP_ATOMIC);
1261 return pool->next_mapping ? 0 : -ENOMEM;
1264 static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
1266 struct dm_thin_new_mapping *m = pool->next_mapping;
1268 BUG_ON(!pool->next_mapping);
1270 memset(m, 0, sizeof(struct dm_thin_new_mapping));
1271 INIT_LIST_HEAD(&m->list);
1274 pool->next_mapping = NULL;
1279 static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m,
1280 sector_t begin, sector_t end)
1282 struct dm_io_region to;
1284 to.bdev = tc->pool_dev->bdev;
1286 to.count = end - begin;
1288 dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m);
1291 static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio,
1292 dm_block_t data_begin,
1293 struct dm_thin_new_mapping *m)
1295 struct pool *pool = tc->pool;
1296 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1298 h->overwrite_mapping = m;
1300 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
1301 inc_all_io_entry(pool, bio);
1302 remap_and_issue(tc, bio, data_begin);
1306 * A partial copy also needs to zero the uncopied region.
1308 static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
1309 struct dm_dev *origin, dm_block_t data_origin,
1310 dm_block_t data_dest,
1311 struct dm_bio_prison_cell *cell, struct bio *bio,
1314 struct pool *pool = tc->pool;
1315 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1318 m->virt_begin = virt_block;
1319 m->virt_end = virt_block + 1u;
1320 m->data_block = data_dest;
1324 * quiesce action + copy action + an extra reference held for the
1325 * duration of this function (we may need to inc later for a
1328 atomic_set(&m->prepare_actions, 3);
1330 if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
1331 complete_mapping_preparation(m); /* already quiesced */
1334 * IO to pool_dev remaps to the pool target's data_dev.
1336 * If the whole block of data is being overwritten, we can issue the
1337 * bio immediately. Otherwise we use kcopyd to clone the data first.
1339 if (io_overwrites_block(pool, bio))
1340 remap_and_issue_overwrite(tc, bio, data_dest, m);
1342 struct dm_io_region from, to;
1344 from.bdev = origin->bdev;
1345 from.sector = data_origin * pool->sectors_per_block;
1348 to.bdev = tc->pool_dev->bdev;
1349 to.sector = data_dest * pool->sectors_per_block;
1352 dm_kcopyd_copy(pool->copier, &from, 1, &to,
1353 0, copy_complete, m);
1356 * Do we need to zero a tail region?
1358 if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) {
1359 atomic_inc(&m->prepare_actions);
1361 data_dest * pool->sectors_per_block + len,
1362 (data_dest + 1) * pool->sectors_per_block);
1366 complete_mapping_preparation(m); /* drop our ref */
1369 static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
1370 dm_block_t data_origin, dm_block_t data_dest,
1371 struct dm_bio_prison_cell *cell, struct bio *bio)
1373 schedule_copy(tc, virt_block, tc->pool_dev,
1374 data_origin, data_dest, cell, bio,
1375 tc->pool->sectors_per_block);
1378 static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
1379 dm_block_t data_block, struct dm_bio_prison_cell *cell,
1382 struct pool *pool = tc->pool;
1383 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1385 atomic_set(&m->prepare_actions, 1); /* no need to quiesce */
1387 m->virt_begin = virt_block;
1388 m->virt_end = virt_block + 1u;
1389 m->data_block = data_block;
1393 * If the whole block of data is being overwritten or we are not
1394 * zeroing pre-existing data, we can issue the bio immediately.
1395 * Otherwise we use kcopyd to zero the data first.
1397 if (pool->pf.zero_new_blocks) {
1398 if (io_overwrites_block(pool, bio))
1399 remap_and_issue_overwrite(tc, bio, data_block, m);
1401 ll_zero(tc, m, data_block * pool->sectors_per_block,
1402 (data_block + 1) * pool->sectors_per_block);
1404 process_prepared_mapping(m);
1407 static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
1408 dm_block_t data_dest,
1409 struct dm_bio_prison_cell *cell, struct bio *bio)
1411 struct pool *pool = tc->pool;
1412 sector_t virt_block_begin = virt_block * pool->sectors_per_block;
1413 sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block;
1415 if (virt_block_end <= tc->origin_size)
1416 schedule_copy(tc, virt_block, tc->origin_dev,
1417 virt_block, data_dest, cell, bio,
1418 pool->sectors_per_block);
1420 else if (virt_block_begin < tc->origin_size)
1421 schedule_copy(tc, virt_block, tc->origin_dev,
1422 virt_block, data_dest, cell, bio,
1423 tc->origin_size - virt_block_begin);
1426 schedule_zero(tc, virt_block, data_dest, cell, bio);
1429 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode);
1431 static void requeue_bios(struct pool *pool);
1433 static bool is_read_only_pool_mode(enum pool_mode mode)
1435 return (mode == PM_OUT_OF_METADATA_SPACE || mode == PM_READ_ONLY);
1438 static bool is_read_only(struct pool *pool)
1440 return is_read_only_pool_mode(get_pool_mode(pool));
1443 static void check_for_metadata_space(struct pool *pool)
1446 const char *ooms_reason = NULL;
1449 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free);
1451 ooms_reason = "Could not get free metadata blocks";
1453 ooms_reason = "No free metadata blocks";
1455 if (ooms_reason && !is_read_only(pool)) {
1456 DMERR("%s", ooms_reason);
1457 set_pool_mode(pool, PM_OUT_OF_METADATA_SPACE);
1461 static void check_for_data_space(struct pool *pool)
1466 if (get_pool_mode(pool) != PM_OUT_OF_DATA_SPACE)
1469 r = dm_pool_get_free_block_count(pool->pmd, &nr_free);
1474 set_pool_mode(pool, PM_WRITE);
1480 * A non-zero return indicates read_only or fail_io mode.
1481 * Many callers don't care about the return value.
1483 static int commit(struct pool *pool)
1487 if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE)
1490 r = dm_pool_commit_metadata(pool->pmd);
1492 metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
1494 check_for_metadata_space(pool);
1495 check_for_data_space(pool);
1501 static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks)
1503 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
1504 DMWARN("%s: reached low water mark for data device: sending event.",
1505 dm_device_name(pool->pool_md));
1506 spin_lock_irq(&pool->lock);
1507 pool->low_water_triggered = true;
1508 spin_unlock_irq(&pool->lock);
1509 dm_table_event(pool->ti->table);
1513 static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
1516 dm_block_t free_blocks;
1517 struct pool *pool = tc->pool;
1519 if (WARN_ON(get_pool_mode(pool) != PM_WRITE))
1522 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1524 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1528 check_low_water_mark(pool, free_blocks);
1532 * Try to commit to see if that will free up some
1539 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1541 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1546 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1551 r = dm_pool_alloc_data_block(pool->pmd, result);
1554 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1556 metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
1560 r = dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks);
1562 metadata_operation_failed(pool, "dm_pool_get_free_metadata_block_count", r);
1567 /* Let's commit before we use up the metadata reserve. */
1577 * If we have run out of space, queue bios until the device is
1578 * resumed, presumably after having been reloaded with more space.
1580 static void retry_on_resume(struct bio *bio)
1582 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1583 struct thin_c *tc = h->tc;
1585 spin_lock_irq(&tc->lock);
1586 bio_list_add(&tc->retry_on_resume_list, bio);
1587 spin_unlock_irq(&tc->lock);
1590 static blk_status_t should_error_unserviceable_bio(struct pool *pool)
1592 enum pool_mode m = get_pool_mode(pool);
1596 /* Shouldn't get here */
1597 DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
1598 return BLK_STS_IOERR;
1600 case PM_OUT_OF_DATA_SPACE:
1601 return pool->pf.error_if_no_space ? BLK_STS_NOSPC : 0;
1603 case PM_OUT_OF_METADATA_SPACE:
1606 return BLK_STS_IOERR;
1608 /* Shouldn't get here */
1609 DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1610 return BLK_STS_IOERR;
1614 static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
1616 blk_status_t error = should_error_unserviceable_bio(pool);
1619 bio->bi_status = error;
1622 retry_on_resume(bio);
1625 static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell)
1628 struct bio_list bios;
1631 error = should_error_unserviceable_bio(pool);
1633 cell_error_with_code(pool, cell, error);
1637 bio_list_init(&bios);
1638 cell_release(pool, cell, &bios);
1640 while ((bio = bio_list_pop(&bios)))
1641 retry_on_resume(bio);
1644 static void process_discard_cell_no_passdown(struct thin_c *tc,
1645 struct dm_bio_prison_cell *virt_cell)
1647 struct pool *pool = tc->pool;
1648 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1651 * We don't need to lock the data blocks, since there's no
1652 * passdown. We only lock data blocks for allocation and breaking sharing.
1655 m->virt_begin = virt_cell->key.block_begin;
1656 m->virt_end = virt_cell->key.block_end;
1657 m->cell = virt_cell;
1658 m->bio = virt_cell->holder;
1660 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1661 pool->process_prepared_discard(m);
1664 static void break_up_discard_bio(struct thin_c *tc, dm_block_t begin, dm_block_t end,
1667 struct pool *pool = tc->pool;
1671 struct dm_cell_key data_key;
1672 struct dm_bio_prison_cell *data_cell;
1673 struct dm_thin_new_mapping *m;
1674 dm_block_t virt_begin, virt_end, data_begin;
1676 while (begin != end) {
1677 r = ensure_next_mapping(pool);
1679 /* we did our best */
1682 r = dm_thin_find_mapped_range(tc->td, begin, end, &virt_begin, &virt_end,
1683 &data_begin, &maybe_shared);
1686 * Silently fail, letting any mappings we've
1691 build_key(tc->td, PHYSICAL, data_begin, data_begin + (virt_end - virt_begin), &data_key);
1692 if (bio_detain(tc->pool, &data_key, NULL, &data_cell)) {
1693 /* contention, we'll give up with this range */
1699 * IO may still be going to the destination block. We must
1700 * quiesce before we can do the removal.
1702 m = get_next_mapping(pool);
1704 m->maybe_shared = maybe_shared;
1705 m->virt_begin = virt_begin;
1706 m->virt_end = virt_end;
1707 m->data_block = data_begin;
1708 m->cell = data_cell;
1712 * The parent bio must not complete before sub discard bios are
1713 * chained to it (see end_discard's bio_chain)!
1715 * This per-mapping bi_remaining increment is paired with
1716 * the implicit decrement that occurs via bio_endio() in
1719 bio_inc_remaining(bio);
1720 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1721 pool->process_prepared_discard(m);
1727 static void process_discard_cell_passdown(struct thin_c *tc, struct dm_bio_prison_cell *virt_cell)
1729 struct bio *bio = virt_cell->holder;
1730 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1733 * The virt_cell will only get freed once the origin bio completes.
1734 * This means it will remain locked while all the individual
1735 * passdown bios are in flight.
1737 h->cell = virt_cell;
1738 break_up_discard_bio(tc, virt_cell->key.block_begin, virt_cell->key.block_end, bio);
1741 * We complete the bio now, knowing that the bi_remaining field
1742 * will prevent completion until the sub range discards have
1748 static void process_discard_bio(struct thin_c *tc, struct bio *bio)
1750 dm_block_t begin, end;
1751 struct dm_cell_key virt_key;
1752 struct dm_bio_prison_cell *virt_cell;
1754 get_bio_block_range(tc, bio, &begin, &end);
1757 * The discard covers less than a block.
1763 build_key(tc->td, VIRTUAL, begin, end, &virt_key);
1764 if (bio_detain(tc->pool, &virt_key, bio, &virt_cell))
1766 * Potential starvation issue: We're relying on the
1767 * fs/application being well behaved, and not trying to
1768 * send IO to a region at the same time as discarding it.
1769 * If they do this persistently then it's possible this
1770 * cell will never be granted.
1774 tc->pool->process_discard_cell(tc, virt_cell);
1777 static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
1778 struct dm_cell_key *key,
1779 struct dm_thin_lookup_result *lookup_result,
1780 struct dm_bio_prison_cell *cell)
1783 dm_block_t data_block;
1784 struct pool *pool = tc->pool;
1786 r = alloc_data_block(tc, &data_block);
1789 schedule_internal_copy(tc, block, lookup_result->block,
1790 data_block, cell, bio);
1794 retry_bios_on_resume(pool, cell);
1798 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1800 cell_error(pool, cell);
1805 static void __remap_and_issue_shared_cell(void *context,
1806 struct dm_bio_prison_cell *cell)
1808 struct remap_info *info = context;
1811 while ((bio = bio_list_pop(&cell->bios))) {
1812 if (bio_data_dir(bio) == WRITE || op_is_flush(bio->bi_opf) ||
1813 bio_op(bio) == REQ_OP_DISCARD)
1814 bio_list_add(&info->defer_bios, bio);
1816 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1818 h->shared_read_entry = dm_deferred_entry_inc(info->tc->pool->shared_read_ds);
1819 inc_all_io_entry(info->tc->pool, bio);
1820 bio_list_add(&info->issue_bios, bio);
1825 static void remap_and_issue_shared_cell(struct thin_c *tc,
1826 struct dm_bio_prison_cell *cell,
1830 struct remap_info info;
1833 bio_list_init(&info.defer_bios);
1834 bio_list_init(&info.issue_bios);
1836 cell_visit_release(tc->pool, __remap_and_issue_shared_cell,
1839 while ((bio = bio_list_pop(&info.defer_bios)))
1840 thin_defer_bio(tc, bio);
1842 while ((bio = bio_list_pop(&info.issue_bios)))
1843 remap_and_issue(tc, bio, block);
1846 static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1848 struct dm_thin_lookup_result *lookup_result,
1849 struct dm_bio_prison_cell *virt_cell)
1851 struct dm_bio_prison_cell *data_cell;
1852 struct pool *pool = tc->pool;
1853 struct dm_cell_key key;
1856 * If cell is already occupied, then sharing is already in the process
1857 * of being broken so we have nothing further to do here.
1859 build_data_key(tc->td, lookup_result->block, &key);
1860 if (bio_detain(pool, &key, bio, &data_cell)) {
1861 cell_defer_no_holder(tc, virt_cell);
1865 if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) {
1866 break_sharing(tc, bio, block, &key, lookup_result, data_cell);
1867 cell_defer_no_holder(tc, virt_cell);
1869 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1871 h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
1872 inc_all_io_entry(pool, bio);
1873 remap_and_issue(tc, bio, lookup_result->block);
1875 remap_and_issue_shared_cell(tc, data_cell, lookup_result->block);
1876 remap_and_issue_shared_cell(tc, virt_cell, lookup_result->block);
1880 static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
1881 struct dm_bio_prison_cell *cell)
1884 dm_block_t data_block;
1885 struct pool *pool = tc->pool;
1888 * Remap empty bios (flushes) immediately, without provisioning.
1890 if (!bio->bi_iter.bi_size) {
1891 inc_all_io_entry(pool, bio);
1892 cell_defer_no_holder(tc, cell);
1894 remap_and_issue(tc, bio, 0);
1899 * Fill read bios with zeroes and complete them immediately.
1901 if (bio_data_dir(bio) == READ) {
1903 cell_defer_no_holder(tc, cell);
1908 r = alloc_data_block(tc, &data_block);
1912 schedule_external_copy(tc, block, data_block, cell, bio);
1914 schedule_zero(tc, block, data_block, cell, bio);
1918 retry_bios_on_resume(pool, cell);
1922 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1924 cell_error(pool, cell);
1929 static void process_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
1932 struct pool *pool = tc->pool;
1933 struct bio *bio = cell->holder;
1934 dm_block_t block = get_bio_block(tc, bio);
1935 struct dm_thin_lookup_result lookup_result;
1937 if (tc->requeue_mode) {
1938 cell_requeue(pool, cell);
1942 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1945 if (lookup_result.shared)
1946 process_shared_bio(tc, bio, block, &lookup_result, cell);
1948 inc_all_io_entry(pool, bio);
1949 remap_and_issue(tc, bio, lookup_result.block);
1950 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
1955 if (bio_data_dir(bio) == READ && tc->origin_dev) {
1956 inc_all_io_entry(pool, bio);
1957 cell_defer_no_holder(tc, cell);
1959 if (bio_end_sector(bio) <= tc->origin_size)
1960 remap_to_origin_and_issue(tc, bio);
1962 else if (bio->bi_iter.bi_sector < tc->origin_size) {
1964 bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT;
1965 remap_to_origin_and_issue(tc, bio);
1972 provision_block(tc, bio, block, cell);
1976 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1978 cell_defer_no_holder(tc, cell);
1984 static void process_bio(struct thin_c *tc, struct bio *bio)
1986 struct pool *pool = tc->pool;
1987 dm_block_t block = get_bio_block(tc, bio);
1988 struct dm_bio_prison_cell *cell;
1989 struct dm_cell_key key;
1992 * If cell is already occupied, then the block is already
1993 * being provisioned so we have nothing further to do here.
1995 build_virtual_key(tc->td, block, &key);
1996 if (bio_detain(pool, &key, bio, &cell))
1999 process_cell(tc, cell);
2002 static void __process_bio_read_only(struct thin_c *tc, struct bio *bio,
2003 struct dm_bio_prison_cell *cell)
2006 int rw = bio_data_dir(bio);
2007 dm_block_t block = get_bio_block(tc, bio);
2008 struct dm_thin_lookup_result lookup_result;
2010 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
2013 if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) {
2014 handle_unserviceable_bio(tc->pool, bio);
2016 cell_defer_no_holder(tc, cell);
2018 inc_all_io_entry(tc->pool, bio);
2019 remap_and_issue(tc, bio, lookup_result.block);
2021 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
2027 cell_defer_no_holder(tc, cell);
2029 handle_unserviceable_bio(tc->pool, bio);
2033 if (tc->origin_dev) {
2034 inc_all_io_entry(tc->pool, bio);
2035 remap_to_origin_and_issue(tc, bio);
2044 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
2047 cell_defer_no_holder(tc, cell);
2053 static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
2055 __process_bio_read_only(tc, bio, NULL);
2058 static void process_cell_read_only(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2060 __process_bio_read_only(tc, cell->holder, cell);
2063 static void process_bio_success(struct thin_c *tc, struct bio *bio)
2068 static void process_bio_fail(struct thin_c *tc, struct bio *bio)
2073 static void process_cell_success(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2075 cell_success(tc->pool, cell);
2078 static void process_cell_fail(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2080 cell_error(tc->pool, cell);
2084 * FIXME: should we also commit due to size of transaction, measured in
2087 static int need_commit_due_to_time(struct pool *pool)
2089 return !time_in_range(jiffies, pool->last_commit_jiffies,
2090 pool->last_commit_jiffies + COMMIT_PERIOD);
2093 #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node)
2094 #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook))
2096 static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio)
2098 struct rb_node **rbp, *parent;
2099 struct dm_thin_endio_hook *pbd;
2100 sector_t bi_sector = bio->bi_iter.bi_sector;
2102 rbp = &tc->sort_bio_list.rb_node;
2106 pbd = thin_pbd(parent);
2108 if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector)
2109 rbp = &(*rbp)->rb_left;
2111 rbp = &(*rbp)->rb_right;
2114 pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2115 rb_link_node(&pbd->rb_node, parent, rbp);
2116 rb_insert_color(&pbd->rb_node, &tc->sort_bio_list);
2119 static void __extract_sorted_bios(struct thin_c *tc)
2121 struct rb_node *node;
2122 struct dm_thin_endio_hook *pbd;
2125 for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) {
2126 pbd = thin_pbd(node);
2127 bio = thin_bio(pbd);
2129 bio_list_add(&tc->deferred_bio_list, bio);
2130 rb_erase(&pbd->rb_node, &tc->sort_bio_list);
2133 WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list));
2136 static void __sort_thin_deferred_bios(struct thin_c *tc)
2139 struct bio_list bios;
2141 bio_list_init(&bios);
2142 bio_list_merge(&bios, &tc->deferred_bio_list);
2143 bio_list_init(&tc->deferred_bio_list);
2145 /* Sort deferred_bio_list using rb-tree */
2146 while ((bio = bio_list_pop(&bios)))
2147 __thin_bio_rb_add(tc, bio);
2150 * Transfer the sorted bios in sort_bio_list back to
2151 * deferred_bio_list to allow lockless submission of
2154 __extract_sorted_bios(tc);
2157 static void process_thin_deferred_bios(struct thin_c *tc)
2159 struct pool *pool = tc->pool;
2161 struct bio_list bios;
2162 struct blk_plug plug;
2165 if (tc->requeue_mode) {
2166 error_thin_bio_list(tc, &tc->deferred_bio_list,
2167 BLK_STS_DM_REQUEUE);
2171 bio_list_init(&bios);
2173 spin_lock_irq(&tc->lock);
2175 if (bio_list_empty(&tc->deferred_bio_list)) {
2176 spin_unlock_irq(&tc->lock);
2180 __sort_thin_deferred_bios(tc);
2182 bio_list_merge(&bios, &tc->deferred_bio_list);
2183 bio_list_init(&tc->deferred_bio_list);
2185 spin_unlock_irq(&tc->lock);
2187 blk_start_plug(&plug);
2188 while ((bio = bio_list_pop(&bios))) {
2190 * If we've got no free new_mapping structs, and processing
2191 * this bio might require one, we pause until there are some
2192 * prepared mappings to process.
2194 if (ensure_next_mapping(pool)) {
2195 spin_lock_irq(&tc->lock);
2196 bio_list_add(&tc->deferred_bio_list, bio);
2197 bio_list_merge(&tc->deferred_bio_list, &bios);
2198 spin_unlock_irq(&tc->lock);
2202 if (bio_op(bio) == REQ_OP_DISCARD)
2203 pool->process_discard(tc, bio);
2205 pool->process_bio(tc, bio);
2207 if ((count++ & 127) == 0) {
2208 throttle_work_update(&pool->throttle);
2209 dm_pool_issue_prefetches(pool->pmd);
2212 blk_finish_plug(&plug);
2215 static int cmp_cells(const void *lhs, const void *rhs)
2217 struct dm_bio_prison_cell *lhs_cell = *((struct dm_bio_prison_cell **) lhs);
2218 struct dm_bio_prison_cell *rhs_cell = *((struct dm_bio_prison_cell **) rhs);
2220 BUG_ON(!lhs_cell->holder);
2221 BUG_ON(!rhs_cell->holder);
2223 if (lhs_cell->holder->bi_iter.bi_sector < rhs_cell->holder->bi_iter.bi_sector)
2226 if (lhs_cell->holder->bi_iter.bi_sector > rhs_cell->holder->bi_iter.bi_sector)
2232 static unsigned sort_cells(struct pool *pool, struct list_head *cells)
2235 struct dm_bio_prison_cell *cell, *tmp;
2237 list_for_each_entry_safe(cell, tmp, cells, user_list) {
2238 if (count >= CELL_SORT_ARRAY_SIZE)
2241 pool->cell_sort_array[count++] = cell;
2242 list_del(&cell->user_list);
2245 sort(pool->cell_sort_array, count, sizeof(cell), cmp_cells, NULL);
2250 static void process_thin_deferred_cells(struct thin_c *tc)
2252 struct pool *pool = tc->pool;
2253 struct list_head cells;
2254 struct dm_bio_prison_cell *cell;
2255 unsigned i, j, count;
2257 INIT_LIST_HEAD(&cells);
2259 spin_lock_irq(&tc->lock);
2260 list_splice_init(&tc->deferred_cells, &cells);
2261 spin_unlock_irq(&tc->lock);
2263 if (list_empty(&cells))
2267 count = sort_cells(tc->pool, &cells);
2269 for (i = 0; i < count; i++) {
2270 cell = pool->cell_sort_array[i];
2271 BUG_ON(!cell->holder);
2274 * If we've got no free new_mapping structs, and processing
2275 * this bio might require one, we pause until there are some
2276 * prepared mappings to process.
2278 if (ensure_next_mapping(pool)) {
2279 for (j = i; j < count; j++)
2280 list_add(&pool->cell_sort_array[j]->user_list, &cells);
2282 spin_lock_irq(&tc->lock);
2283 list_splice(&cells, &tc->deferred_cells);
2284 spin_unlock_irq(&tc->lock);
2288 if (bio_op(cell->holder) == REQ_OP_DISCARD)
2289 pool->process_discard_cell(tc, cell);
2291 pool->process_cell(tc, cell);
2293 } while (!list_empty(&cells));
2296 static void thin_get(struct thin_c *tc);
2297 static void thin_put(struct thin_c *tc);
2300 * We can't hold rcu_read_lock() around code that can block. So we
2301 * find a thin with the rcu lock held; bump a refcount; then drop
2304 static struct thin_c *get_first_thin(struct pool *pool)
2306 struct thin_c *tc = NULL;
2309 if (!list_empty(&pool->active_thins)) {
2310 tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list);
2318 static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc)
2320 struct thin_c *old_tc = tc;
2323 list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) {
2335 static void process_deferred_bios(struct pool *pool)
2338 struct bio_list bios, bio_completions;
2341 tc = get_first_thin(pool);
2343 process_thin_deferred_cells(tc);
2344 process_thin_deferred_bios(tc);
2345 tc = get_next_thin(pool, tc);
2349 * If there are any deferred flush bios, we must commit the metadata
2350 * before issuing them or signaling their completion.
2352 bio_list_init(&bios);
2353 bio_list_init(&bio_completions);
2355 spin_lock_irq(&pool->lock);
2356 bio_list_merge(&bios, &pool->deferred_flush_bios);
2357 bio_list_init(&pool->deferred_flush_bios);
2359 bio_list_merge(&bio_completions, &pool->deferred_flush_completions);
2360 bio_list_init(&pool->deferred_flush_completions);
2361 spin_unlock_irq(&pool->lock);
2363 if (bio_list_empty(&bios) && bio_list_empty(&bio_completions) &&
2364 !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool)))
2368 bio_list_merge(&bios, &bio_completions);
2370 while ((bio = bio_list_pop(&bios)))
2374 pool->last_commit_jiffies = jiffies;
2376 while ((bio = bio_list_pop(&bio_completions)))
2379 while ((bio = bio_list_pop(&bios))) {
2381 * The data device was flushed as part of metadata commit,
2382 * so complete redundant flushes immediately.
2384 if (bio->bi_opf & REQ_PREFLUSH)
2387 dm_submit_bio_remap(bio, NULL);
2391 static void do_worker(struct work_struct *ws)
2393 struct pool *pool = container_of(ws, struct pool, worker);
2395 throttle_work_start(&pool->throttle);
2396 dm_pool_issue_prefetches(pool->pmd);
2397 throttle_work_update(&pool->throttle);
2398 process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
2399 throttle_work_update(&pool->throttle);
2400 process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
2401 throttle_work_update(&pool->throttle);
2402 process_prepared(pool, &pool->prepared_discards_pt2, &pool->process_prepared_discard_pt2);
2403 throttle_work_update(&pool->throttle);
2404 process_deferred_bios(pool);
2405 throttle_work_complete(&pool->throttle);
2409 * We want to commit periodically so that not too much
2410 * unwritten data builds up.
2412 static void do_waker(struct work_struct *ws)
2414 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
2416 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
2420 * We're holding onto IO to allow userland time to react. After the
2421 * timeout either the pool will have been resized (and thus back in
2422 * PM_WRITE mode), or we degrade to PM_OUT_OF_DATA_SPACE w/ error_if_no_space.
2424 static void do_no_space_timeout(struct work_struct *ws)
2426 struct pool *pool = container_of(to_delayed_work(ws), struct pool,
2429 if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) {
2430 pool->pf.error_if_no_space = true;
2431 notify_of_pool_mode_change(pool);
2432 error_retry_list_with_code(pool, BLK_STS_NOSPC);
2436 /*----------------------------------------------------------------*/
2439 struct work_struct worker;
2440 struct completion complete;
2443 static struct pool_work *to_pool_work(struct work_struct *ws)
2445 return container_of(ws, struct pool_work, worker);
2448 static void pool_work_complete(struct pool_work *pw)
2450 complete(&pw->complete);
2453 static void pool_work_wait(struct pool_work *pw, struct pool *pool,
2454 void (*fn)(struct work_struct *))
2456 INIT_WORK_ONSTACK(&pw->worker, fn);
2457 init_completion(&pw->complete);
2458 queue_work(pool->wq, &pw->worker);
2459 wait_for_completion(&pw->complete);
2462 /*----------------------------------------------------------------*/
2464 struct noflush_work {
2465 struct pool_work pw;
2469 static struct noflush_work *to_noflush(struct work_struct *ws)
2471 return container_of(to_pool_work(ws), struct noflush_work, pw);
2474 static void do_noflush_start(struct work_struct *ws)
2476 struct noflush_work *w = to_noflush(ws);
2477 w->tc->requeue_mode = true;
2479 pool_work_complete(&w->pw);
2482 static void do_noflush_stop(struct work_struct *ws)
2484 struct noflush_work *w = to_noflush(ws);
2485 w->tc->requeue_mode = false;
2486 pool_work_complete(&w->pw);
2489 static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *))
2491 struct noflush_work w;
2494 pool_work_wait(&w.pw, tc->pool, fn);
2497 /*----------------------------------------------------------------*/
2499 static bool passdown_enabled(struct pool_c *pt)
2501 return pt->adjusted_pf.discard_passdown;
2504 static void set_discard_callbacks(struct pool *pool)
2506 struct pool_c *pt = pool->ti->private;
2508 if (passdown_enabled(pt)) {
2509 pool->process_discard_cell = process_discard_cell_passdown;
2510 pool->process_prepared_discard = process_prepared_discard_passdown_pt1;
2511 pool->process_prepared_discard_pt2 = process_prepared_discard_passdown_pt2;
2513 pool->process_discard_cell = process_discard_cell_no_passdown;
2514 pool->process_prepared_discard = process_prepared_discard_no_passdown;
2518 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode)
2520 struct pool_c *pt = pool->ti->private;
2521 bool needs_check = dm_pool_metadata_needs_check(pool->pmd);
2522 enum pool_mode old_mode = get_pool_mode(pool);
2523 unsigned long no_space_timeout = READ_ONCE(no_space_timeout_secs) * HZ;
2526 * Never allow the pool to transition to PM_WRITE mode if user
2527 * intervention is required to verify metadata and data consistency.
2529 if (new_mode == PM_WRITE && needs_check) {
2530 DMERR("%s: unable to switch pool to write mode until repaired.",
2531 dm_device_name(pool->pool_md));
2532 if (old_mode != new_mode)
2533 new_mode = old_mode;
2535 new_mode = PM_READ_ONLY;
2538 * If we were in PM_FAIL mode, rollback of metadata failed. We're
2539 * not going to recover without a thin_repair. So we never let the
2540 * pool move out of the old mode.
2542 if (old_mode == PM_FAIL)
2543 new_mode = old_mode;
2547 dm_pool_metadata_read_only(pool->pmd);
2548 pool->process_bio = process_bio_fail;
2549 pool->process_discard = process_bio_fail;
2550 pool->process_cell = process_cell_fail;
2551 pool->process_discard_cell = process_cell_fail;
2552 pool->process_prepared_mapping = process_prepared_mapping_fail;
2553 pool->process_prepared_discard = process_prepared_discard_fail;
2555 error_retry_list(pool);
2558 case PM_OUT_OF_METADATA_SPACE:
2560 dm_pool_metadata_read_only(pool->pmd);
2561 pool->process_bio = process_bio_read_only;
2562 pool->process_discard = process_bio_success;
2563 pool->process_cell = process_cell_read_only;
2564 pool->process_discard_cell = process_cell_success;
2565 pool->process_prepared_mapping = process_prepared_mapping_fail;
2566 pool->process_prepared_discard = process_prepared_discard_success;
2568 error_retry_list(pool);
2571 case PM_OUT_OF_DATA_SPACE:
2573 * Ideally we'd never hit this state; the low water mark
2574 * would trigger userland to extend the pool before we
2575 * completely run out of data space. However, many small
2576 * IOs to unprovisioned space can consume data space at an
2577 * alarming rate. Adjust your low water mark if you're
2578 * frequently seeing this mode.
2580 pool->out_of_data_space = true;
2581 pool->process_bio = process_bio_read_only;
2582 pool->process_discard = process_discard_bio;
2583 pool->process_cell = process_cell_read_only;
2584 pool->process_prepared_mapping = process_prepared_mapping;
2585 set_discard_callbacks(pool);
2587 if (!pool->pf.error_if_no_space && no_space_timeout)
2588 queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout);
2592 if (old_mode == PM_OUT_OF_DATA_SPACE)
2593 cancel_delayed_work_sync(&pool->no_space_timeout);
2594 pool->out_of_data_space = false;
2595 pool->pf.error_if_no_space = pt->requested_pf.error_if_no_space;
2596 dm_pool_metadata_read_write(pool->pmd);
2597 pool->process_bio = process_bio;
2598 pool->process_discard = process_discard_bio;
2599 pool->process_cell = process_cell;
2600 pool->process_prepared_mapping = process_prepared_mapping;
2601 set_discard_callbacks(pool);
2605 pool->pf.mode = new_mode;
2607 * The pool mode may have changed, sync it so bind_control_target()
2608 * doesn't cause an unexpected mode transition on resume.
2610 pt->adjusted_pf.mode = new_mode;
2612 if (old_mode != new_mode)
2613 notify_of_pool_mode_change(pool);
2616 static void abort_transaction(struct pool *pool)
2618 const char *dev_name = dm_device_name(pool->pool_md);
2620 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name);
2621 if (dm_pool_abort_metadata(pool->pmd)) {
2622 DMERR("%s: failed to abort metadata transaction", dev_name);
2623 set_pool_mode(pool, PM_FAIL);
2626 if (dm_pool_metadata_set_needs_check(pool->pmd)) {
2627 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name);
2628 set_pool_mode(pool, PM_FAIL);
2632 static void metadata_operation_failed(struct pool *pool, const char *op, int r)
2634 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
2635 dm_device_name(pool->pool_md), op, r);
2637 abort_transaction(pool);
2638 set_pool_mode(pool, PM_READ_ONLY);
2641 /*----------------------------------------------------------------*/
2644 * Mapping functions.
2648 * Called only while mapping a thin bio to hand it over to the workqueue.
2650 static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
2652 struct pool *pool = tc->pool;
2654 spin_lock_irq(&tc->lock);
2655 bio_list_add(&tc->deferred_bio_list, bio);
2656 spin_unlock_irq(&tc->lock);
2661 static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio)
2663 struct pool *pool = tc->pool;
2665 throttle_lock(&pool->throttle);
2666 thin_defer_bio(tc, bio);
2667 throttle_unlock(&pool->throttle);
2670 static void thin_defer_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2672 struct pool *pool = tc->pool;
2674 throttle_lock(&pool->throttle);
2675 spin_lock_irq(&tc->lock);
2676 list_add_tail(&cell->user_list, &tc->deferred_cells);
2677 spin_unlock_irq(&tc->lock);
2678 throttle_unlock(&pool->throttle);
2683 static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
2685 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2688 h->shared_read_entry = NULL;
2689 h->all_io_entry = NULL;
2690 h->overwrite_mapping = NULL;
2695 * Non-blocking function called from the thin target's map function.
2697 static int thin_bio_map(struct dm_target *ti, struct bio *bio)
2700 struct thin_c *tc = ti->private;
2701 dm_block_t block = get_bio_block(tc, bio);
2702 struct dm_thin_device *td = tc->td;
2703 struct dm_thin_lookup_result result;
2704 struct dm_bio_prison_cell *virt_cell, *data_cell;
2705 struct dm_cell_key key;
2707 thin_hook_bio(tc, bio);
2709 if (tc->requeue_mode) {
2710 bio->bi_status = BLK_STS_DM_REQUEUE;
2712 return DM_MAPIO_SUBMITTED;
2715 if (get_pool_mode(tc->pool) == PM_FAIL) {
2717 return DM_MAPIO_SUBMITTED;
2720 if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD) {
2721 thin_defer_bio_with_throttle(tc, bio);
2722 return DM_MAPIO_SUBMITTED;
2726 * We must hold the virtual cell before doing the lookup, otherwise
2727 * there's a race with discard.
2729 build_virtual_key(tc->td, block, &key);
2730 if (bio_detain(tc->pool, &key, bio, &virt_cell))
2731 return DM_MAPIO_SUBMITTED;
2733 r = dm_thin_find_block(td, block, 0, &result);
2736 * Note that we defer readahead too.
2740 if (unlikely(result.shared)) {
2742 * We have a race condition here between the
2743 * result.shared value returned by the lookup and
2744 * snapshot creation, which may cause new
2747 * To avoid this always quiesce the origin before
2748 * taking the snap. You want to do this anyway to
2749 * ensure a consistent application view
2752 * More distant ancestors are irrelevant. The
2753 * shared flag will be set in their case.
2755 thin_defer_cell(tc, virt_cell);
2756 return DM_MAPIO_SUBMITTED;
2759 build_data_key(tc->td, result.block, &key);
2760 if (bio_detain(tc->pool, &key, bio, &data_cell)) {
2761 cell_defer_no_holder(tc, virt_cell);
2762 return DM_MAPIO_SUBMITTED;
2765 inc_all_io_entry(tc->pool, bio);
2766 cell_defer_no_holder(tc, data_cell);
2767 cell_defer_no_holder(tc, virt_cell);
2769 remap(tc, bio, result.block);
2770 return DM_MAPIO_REMAPPED;
2774 thin_defer_cell(tc, virt_cell);
2775 return DM_MAPIO_SUBMITTED;
2779 * Must always call bio_io_error on failure.
2780 * dm_thin_find_block can fail with -EINVAL if the
2781 * pool is switched to fail-io mode.
2784 cell_defer_no_holder(tc, virt_cell);
2785 return DM_MAPIO_SUBMITTED;
2789 static void requeue_bios(struct pool *pool)
2794 list_for_each_entry_rcu(tc, &pool->active_thins, list) {
2795 spin_lock_irq(&tc->lock);
2796 bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list);
2797 bio_list_init(&tc->retry_on_resume_list);
2798 spin_unlock_irq(&tc->lock);
2803 /*----------------------------------------------------------------
2804 * Binding of control targets to a pool object
2805 *--------------------------------------------------------------*/
2806 static bool is_factor(sector_t block_size, uint32_t n)
2808 return !sector_div(block_size, n);
2812 * If discard_passdown was enabled verify that the data device
2813 * supports discards. Disable discard_passdown if not.
2815 static void disable_passdown_if_not_supported(struct pool_c *pt)
2817 struct pool *pool = pt->pool;
2818 struct block_device *data_bdev = pt->data_dev->bdev;
2819 struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
2820 const char *reason = NULL;
2822 if (!pt->adjusted_pf.discard_passdown)
2825 if (!bdev_max_discard_sectors(pt->data_dev->bdev))
2826 reason = "discard unsupported";
2828 else if (data_limits->max_discard_sectors < pool->sectors_per_block)
2829 reason = "max discard sectors smaller than a block";
2832 DMWARN("Data device (%pg) %s: Disabling discard passdown.", data_bdev, reason);
2833 pt->adjusted_pf.discard_passdown = false;
2837 static int bind_control_target(struct pool *pool, struct dm_target *ti)
2839 struct pool_c *pt = ti->private;
2842 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
2844 enum pool_mode old_mode = get_pool_mode(pool);
2845 enum pool_mode new_mode = pt->adjusted_pf.mode;
2848 * Don't change the pool's mode until set_pool_mode() below.
2849 * Otherwise the pool's process_* function pointers may
2850 * not match the desired pool mode.
2852 pt->adjusted_pf.mode = old_mode;
2855 pool->pf = pt->adjusted_pf;
2856 pool->low_water_blocks = pt->low_water_blocks;
2858 set_pool_mode(pool, new_mode);
2863 static void unbind_control_target(struct pool *pool, struct dm_target *ti)
2869 /*----------------------------------------------------------------
2871 *--------------------------------------------------------------*/
2872 /* Initialize pool features. */
2873 static void pool_features_init(struct pool_features *pf)
2875 pf->mode = PM_WRITE;
2876 pf->zero_new_blocks = true;
2877 pf->discard_enabled = true;
2878 pf->discard_passdown = true;
2879 pf->error_if_no_space = false;
2882 static void __pool_destroy(struct pool *pool)
2884 __pool_table_remove(pool);
2886 vfree(pool->cell_sort_array);
2887 if (dm_pool_metadata_close(pool->pmd) < 0)
2888 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
2890 dm_bio_prison_destroy(pool->prison);
2891 dm_kcopyd_client_destroy(pool->copier);
2893 cancel_delayed_work_sync(&pool->waker);
2894 cancel_delayed_work_sync(&pool->no_space_timeout);
2896 destroy_workqueue(pool->wq);
2898 if (pool->next_mapping)
2899 mempool_free(pool->next_mapping, &pool->mapping_pool);
2900 mempool_exit(&pool->mapping_pool);
2901 dm_deferred_set_destroy(pool->shared_read_ds);
2902 dm_deferred_set_destroy(pool->all_io_ds);
2906 static struct kmem_cache *_new_mapping_cache;
2908 static struct pool *pool_create(struct mapped_device *pool_md,
2909 struct block_device *metadata_dev,
2910 struct block_device *data_dev,
2911 unsigned long block_size,
2912 int read_only, char **error)
2917 struct dm_pool_metadata *pmd;
2918 bool format_device = read_only ? false : true;
2920 pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
2922 *error = "Error creating metadata object";
2923 return (struct pool *)pmd;
2926 pool = kzalloc(sizeof(*pool), GFP_KERNEL);
2928 *error = "Error allocating memory for pool";
2929 err_p = ERR_PTR(-ENOMEM);
2934 pool->sectors_per_block = block_size;
2935 if (block_size & (block_size - 1))
2936 pool->sectors_per_block_shift = -1;
2938 pool->sectors_per_block_shift = __ffs(block_size);
2939 pool->low_water_blocks = 0;
2940 pool_features_init(&pool->pf);
2941 pool->prison = dm_bio_prison_create();
2942 if (!pool->prison) {
2943 *error = "Error creating pool's bio prison";
2944 err_p = ERR_PTR(-ENOMEM);
2948 pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2949 if (IS_ERR(pool->copier)) {
2950 r = PTR_ERR(pool->copier);
2951 *error = "Error creating pool's kcopyd client";
2953 goto bad_kcopyd_client;
2957 * Create singlethreaded workqueue that will service all devices
2958 * that use this metadata.
2960 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
2962 *error = "Error creating pool's workqueue";
2963 err_p = ERR_PTR(-ENOMEM);
2967 throttle_init(&pool->throttle);
2968 INIT_WORK(&pool->worker, do_worker);
2969 INIT_DELAYED_WORK(&pool->waker, do_waker);
2970 INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout);
2971 spin_lock_init(&pool->lock);
2972 bio_list_init(&pool->deferred_flush_bios);
2973 bio_list_init(&pool->deferred_flush_completions);
2974 INIT_LIST_HEAD(&pool->prepared_mappings);
2975 INIT_LIST_HEAD(&pool->prepared_discards);
2976 INIT_LIST_HEAD(&pool->prepared_discards_pt2);
2977 INIT_LIST_HEAD(&pool->active_thins);
2978 pool->low_water_triggered = false;
2979 pool->suspended = true;
2980 pool->out_of_data_space = false;
2982 pool->shared_read_ds = dm_deferred_set_create();
2983 if (!pool->shared_read_ds) {
2984 *error = "Error creating pool's shared read deferred set";
2985 err_p = ERR_PTR(-ENOMEM);
2986 goto bad_shared_read_ds;
2989 pool->all_io_ds = dm_deferred_set_create();
2990 if (!pool->all_io_ds) {
2991 *error = "Error creating pool's all io deferred set";
2992 err_p = ERR_PTR(-ENOMEM);
2996 pool->next_mapping = NULL;
2997 r = mempool_init_slab_pool(&pool->mapping_pool, MAPPING_POOL_SIZE,
2998 _new_mapping_cache);
3000 *error = "Error creating pool's mapping mempool";
3002 goto bad_mapping_pool;
3005 pool->cell_sort_array =
3006 vmalloc(array_size(CELL_SORT_ARRAY_SIZE,
3007 sizeof(*pool->cell_sort_array)));
3008 if (!pool->cell_sort_array) {
3009 *error = "Error allocating cell sort array";
3010 err_p = ERR_PTR(-ENOMEM);
3011 goto bad_sort_array;
3014 pool->ref_count = 1;
3015 pool->last_commit_jiffies = jiffies;
3016 pool->pool_md = pool_md;
3017 pool->md_dev = metadata_dev;
3018 pool->data_dev = data_dev;
3019 __pool_table_insert(pool);
3024 mempool_exit(&pool->mapping_pool);
3026 dm_deferred_set_destroy(pool->all_io_ds);
3028 dm_deferred_set_destroy(pool->shared_read_ds);
3030 destroy_workqueue(pool->wq);
3032 dm_kcopyd_client_destroy(pool->copier);
3034 dm_bio_prison_destroy(pool->prison);
3038 if (dm_pool_metadata_close(pmd))
3039 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
3044 static void __pool_inc(struct pool *pool)
3046 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
3050 static void __pool_dec(struct pool *pool)
3052 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
3053 BUG_ON(!pool->ref_count);
3054 if (!--pool->ref_count)
3055 __pool_destroy(pool);
3058 static struct pool *__pool_find(struct mapped_device *pool_md,
3059 struct block_device *metadata_dev,
3060 struct block_device *data_dev,
3061 unsigned long block_size, int read_only,
3062 char **error, int *created)
3064 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
3067 if (pool->pool_md != pool_md) {
3068 *error = "metadata device already in use by a pool";
3069 return ERR_PTR(-EBUSY);
3071 if (pool->data_dev != data_dev) {
3072 *error = "data device already in use by a pool";
3073 return ERR_PTR(-EBUSY);
3078 pool = __pool_table_lookup(pool_md);
3080 if (pool->md_dev != metadata_dev || pool->data_dev != data_dev) {
3081 *error = "different pool cannot replace a pool";
3082 return ERR_PTR(-EINVAL);
3087 pool = pool_create(pool_md, metadata_dev, data_dev, block_size, read_only, error);
3095 /*----------------------------------------------------------------
3096 * Pool target methods
3097 *--------------------------------------------------------------*/
3098 static void pool_dtr(struct dm_target *ti)
3100 struct pool_c *pt = ti->private;
3102 mutex_lock(&dm_thin_pool_table.mutex);
3104 unbind_control_target(pt->pool, ti);
3105 __pool_dec(pt->pool);
3106 dm_put_device(ti, pt->metadata_dev);
3107 dm_put_device(ti, pt->data_dev);
3110 mutex_unlock(&dm_thin_pool_table.mutex);
3113 static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
3114 struct dm_target *ti)
3118 const char *arg_name;
3120 static const struct dm_arg _args[] = {
3121 {0, 4, "Invalid number of pool feature arguments"},
3125 * No feature arguments supplied.
3130 r = dm_read_arg_group(_args, as, &argc, &ti->error);
3134 while (argc && !r) {
3135 arg_name = dm_shift_arg(as);
3138 if (!strcasecmp(arg_name, "skip_block_zeroing"))
3139 pf->zero_new_blocks = false;
3141 else if (!strcasecmp(arg_name, "ignore_discard"))
3142 pf->discard_enabled = false;
3144 else if (!strcasecmp(arg_name, "no_discard_passdown"))
3145 pf->discard_passdown = false;
3147 else if (!strcasecmp(arg_name, "read_only"))
3148 pf->mode = PM_READ_ONLY;
3150 else if (!strcasecmp(arg_name, "error_if_no_space"))
3151 pf->error_if_no_space = true;
3154 ti->error = "Unrecognised pool feature requested";
3163 static void metadata_low_callback(void *context)
3165 struct pool *pool = context;
3167 DMWARN("%s: reached low water mark for metadata device: sending event.",
3168 dm_device_name(pool->pool_md));
3170 dm_table_event(pool->ti->table);
3174 * We need to flush the data device **before** committing the metadata.
3176 * This ensures that the data blocks of any newly inserted mappings are
3177 * properly written to non-volatile storage and won't be lost in case of a
3180 * Failure to do so can result in data corruption in the case of internal or
3181 * external snapshots and in the case of newly provisioned blocks, when block
3182 * zeroing is enabled.
3184 static int metadata_pre_commit_callback(void *context)
3186 struct pool *pool = context;
3188 return blkdev_issue_flush(pool->data_dev);
3191 static sector_t get_dev_size(struct block_device *bdev)
3193 return bdev_nr_sectors(bdev);
3196 static void warn_if_metadata_device_too_big(struct block_device *bdev)
3198 sector_t metadata_dev_size = get_dev_size(bdev);
3200 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING)
3201 DMWARN("Metadata device %pg is larger than %u sectors: excess space will not be used.",
3202 bdev, THIN_METADATA_MAX_SECTORS);
3205 static sector_t get_metadata_dev_size(struct block_device *bdev)
3207 sector_t metadata_dev_size = get_dev_size(bdev);
3209 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS)
3210 metadata_dev_size = THIN_METADATA_MAX_SECTORS;
3212 return metadata_dev_size;
3215 static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
3217 sector_t metadata_dev_size = get_metadata_dev_size(bdev);
3219 sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE);
3221 return metadata_dev_size;
3225 * When a metadata threshold is crossed a dm event is triggered, and
3226 * userland should respond by growing the metadata device. We could let
3227 * userland set the threshold, like we do with the data threshold, but I'm
3228 * not sure they know enough to do this well.
3230 static dm_block_t calc_metadata_threshold(struct pool_c *pt)
3233 * 4M is ample for all ops with the possible exception of thin
3234 * device deletion which is harmless if it fails (just retry the
3235 * delete after you've grown the device).
3237 dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
3238 return min((dm_block_t)1024ULL /* 4M */, quarter);
3242 * thin-pool <metadata dev> <data dev>
3243 * <data block size (sectors)>
3244 * <low water mark (blocks)>
3245 * [<#feature args> [<arg>]*]
3247 * Optional feature arguments are:
3248 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
3249 * ignore_discard: disable discard
3250 * no_discard_passdown: don't pass discards down to the data device
3251 * read_only: Don't allow any changes to be made to the pool metadata.
3252 * error_if_no_space: error IOs, instead of queueing, if no space.
3254 static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
3256 int r, pool_created = 0;
3259 struct pool_features pf;
3260 struct dm_arg_set as;
3261 struct dm_dev *data_dev;
3262 unsigned long block_size;
3263 dm_block_t low_water_blocks;
3264 struct dm_dev *metadata_dev;
3265 fmode_t metadata_mode;
3268 * FIXME Remove validation from scope of lock.
3270 mutex_lock(&dm_thin_pool_table.mutex);
3273 ti->error = "Invalid argument count";
3281 /* make sure metadata and data are different devices */
3282 if (!strcmp(argv[0], argv[1])) {
3283 ti->error = "Error setting metadata or data device";
3289 * Set default pool features.
3291 pool_features_init(&pf);
3293 dm_consume_args(&as, 4);
3294 r = parse_pool_features(&as, &pf, ti);
3298 metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
3299 r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
3301 ti->error = "Error opening metadata block device";
3304 warn_if_metadata_device_too_big(metadata_dev->bdev);
3306 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
3308 ti->error = "Error getting data device";
3312 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
3313 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
3314 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
3315 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
3316 ti->error = "Invalid block size";
3321 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
3322 ti->error = "Invalid low water mark";
3327 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
3333 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, data_dev->bdev,
3334 block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
3341 * 'pool_created' reflects whether this is the first table load.
3342 * Top level discard support is not allowed to be changed after
3343 * initial load. This would require a pool reload to trigger thin
3346 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
3347 ti->error = "Discard support cannot be disabled once enabled";
3349 goto out_flags_changed;
3354 pt->metadata_dev = metadata_dev;
3355 pt->data_dev = data_dev;
3356 pt->low_water_blocks = low_water_blocks;
3357 pt->adjusted_pf = pt->requested_pf = pf;
3358 ti->num_flush_bios = 1;
3361 * Only need to enable discards if the pool should pass
3362 * them down to the data device. The thin device's discard
3363 * processing will cause mappings to be removed from the btree.
3365 if (pf.discard_enabled && pf.discard_passdown) {
3366 ti->num_discard_bios = 1;
3369 * Setting 'discards_supported' circumvents the normal
3370 * stacking of discard limits (this keeps the pool and
3371 * thin devices' discard limits consistent).
3373 ti->discards_supported = true;
3377 r = dm_pool_register_metadata_threshold(pt->pool->pmd,
3378 calc_metadata_threshold(pt),
3379 metadata_low_callback,
3382 ti->error = "Error registering metadata threshold";
3383 goto out_flags_changed;
3386 dm_pool_register_pre_commit_callback(pool->pmd,
3387 metadata_pre_commit_callback, pool);
3389 mutex_unlock(&dm_thin_pool_table.mutex);
3398 dm_put_device(ti, data_dev);
3400 dm_put_device(ti, metadata_dev);
3402 mutex_unlock(&dm_thin_pool_table.mutex);
3407 static int pool_map(struct dm_target *ti, struct bio *bio)
3410 struct pool_c *pt = ti->private;
3411 struct pool *pool = pt->pool;
3414 * As this is a singleton target, ti->begin is always zero.
3416 spin_lock_irq(&pool->lock);
3417 bio_set_dev(bio, pt->data_dev->bdev);
3418 r = DM_MAPIO_REMAPPED;
3419 spin_unlock_irq(&pool->lock);
3424 static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
3427 struct pool_c *pt = ti->private;
3428 struct pool *pool = pt->pool;
3429 sector_t data_size = ti->len;
3430 dm_block_t sb_data_size;
3432 *need_commit = false;
3434 (void) sector_div(data_size, pool->sectors_per_block);
3436 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
3438 DMERR("%s: failed to retrieve data device size",
3439 dm_device_name(pool->pool_md));
3443 if (data_size < sb_data_size) {
3444 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
3445 dm_device_name(pool->pool_md),
3446 (unsigned long long)data_size, sb_data_size);
3449 } else if (data_size > sb_data_size) {
3450 if (dm_pool_metadata_needs_check(pool->pmd)) {
3451 DMERR("%s: unable to grow the data device until repaired.",
3452 dm_device_name(pool->pool_md));
3457 DMINFO("%s: growing the data device from %llu to %llu blocks",
3458 dm_device_name(pool->pool_md),
3459 sb_data_size, (unsigned long long)data_size);
3460 r = dm_pool_resize_data_dev(pool->pmd, data_size);
3462 metadata_operation_failed(pool, "dm_pool_resize_data_dev", r);
3466 *need_commit = true;
3472 static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
3475 struct pool_c *pt = ti->private;
3476 struct pool *pool = pt->pool;
3477 dm_block_t metadata_dev_size, sb_metadata_dev_size;
3479 *need_commit = false;
3481 metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
3483 r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
3485 DMERR("%s: failed to retrieve metadata device size",
3486 dm_device_name(pool->pool_md));
3490 if (metadata_dev_size < sb_metadata_dev_size) {
3491 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
3492 dm_device_name(pool->pool_md),
3493 metadata_dev_size, sb_metadata_dev_size);
3496 } else if (metadata_dev_size > sb_metadata_dev_size) {
3497 if (dm_pool_metadata_needs_check(pool->pmd)) {
3498 DMERR("%s: unable to grow the metadata device until repaired.",
3499 dm_device_name(pool->pool_md));
3503 warn_if_metadata_device_too_big(pool->md_dev);
3504 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
3505 dm_device_name(pool->pool_md),
3506 sb_metadata_dev_size, metadata_dev_size);
3508 if (get_pool_mode(pool) == PM_OUT_OF_METADATA_SPACE)
3509 set_pool_mode(pool, PM_WRITE);
3511 r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
3513 metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r);
3517 *need_commit = true;
3524 * Retrieves the number of blocks of the data device from
3525 * the superblock and compares it to the actual device size,
3526 * thus resizing the data device in case it has grown.
3528 * This both copes with opening preallocated data devices in the ctr
3529 * being followed by a resume
3531 * calling the resume method individually after userspace has
3532 * grown the data device in reaction to a table event.
3534 static int pool_preresume(struct dm_target *ti)
3537 bool need_commit1, need_commit2;
3538 struct pool_c *pt = ti->private;
3539 struct pool *pool = pt->pool;
3542 * Take control of the pool object.
3544 r = bind_control_target(pool, ti);
3548 r = maybe_resize_data_dev(ti, &need_commit1);
3552 r = maybe_resize_metadata_dev(ti, &need_commit2);
3556 if (need_commit1 || need_commit2)
3557 (void) commit(pool);
3560 * When a thin-pool is PM_FAIL, it cannot be rebuilt if
3561 * bio is in deferred list. Therefore need to return 0
3562 * to allow pool_resume() to flush IO.
3564 if (r && get_pool_mode(pool) == PM_FAIL)
3570 static void pool_suspend_active_thins(struct pool *pool)
3574 /* Suspend all active thin devices */
3575 tc = get_first_thin(pool);
3577 dm_internal_suspend_noflush(tc->thin_md);
3578 tc = get_next_thin(pool, tc);
3582 static void pool_resume_active_thins(struct pool *pool)
3586 /* Resume all active thin devices */
3587 tc = get_first_thin(pool);
3589 dm_internal_resume(tc->thin_md);
3590 tc = get_next_thin(pool, tc);
3594 static void pool_resume(struct dm_target *ti)
3596 struct pool_c *pt = ti->private;
3597 struct pool *pool = pt->pool;
3600 * Must requeue active_thins' bios and then resume
3601 * active_thins _before_ clearing 'suspend' flag.
3604 pool_resume_active_thins(pool);
3606 spin_lock_irq(&pool->lock);
3607 pool->low_water_triggered = false;
3608 pool->suspended = false;
3609 spin_unlock_irq(&pool->lock);
3611 do_waker(&pool->waker.work);
3614 static void pool_presuspend(struct dm_target *ti)
3616 struct pool_c *pt = ti->private;
3617 struct pool *pool = pt->pool;
3619 spin_lock_irq(&pool->lock);
3620 pool->suspended = true;
3621 spin_unlock_irq(&pool->lock);
3623 pool_suspend_active_thins(pool);
3626 static void pool_presuspend_undo(struct dm_target *ti)
3628 struct pool_c *pt = ti->private;
3629 struct pool *pool = pt->pool;
3631 pool_resume_active_thins(pool);
3633 spin_lock_irq(&pool->lock);
3634 pool->suspended = false;
3635 spin_unlock_irq(&pool->lock);
3638 static void pool_postsuspend(struct dm_target *ti)
3640 struct pool_c *pt = ti->private;
3641 struct pool *pool = pt->pool;
3643 cancel_delayed_work_sync(&pool->waker);
3644 cancel_delayed_work_sync(&pool->no_space_timeout);
3645 flush_workqueue(pool->wq);
3646 (void) commit(pool);
3649 static int check_arg_count(unsigned argc, unsigned args_required)
3651 if (argc != args_required) {
3652 DMWARN("Message received with %u arguments instead of %u.",
3653 argc, args_required);
3660 static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
3662 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
3663 *dev_id <= MAX_DEV_ID)
3667 DMWARN("Message received with invalid device id: %s", arg);
3672 static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
3677 r = check_arg_count(argc, 2);
3681 r = read_dev_id(argv[1], &dev_id, 1);
3685 r = dm_pool_create_thin(pool->pmd, dev_id);
3687 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
3695 static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3698 dm_thin_id origin_dev_id;
3701 r = check_arg_count(argc, 3);
3705 r = read_dev_id(argv[1], &dev_id, 1);
3709 r = read_dev_id(argv[2], &origin_dev_id, 1);
3713 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
3715 DMWARN("Creation of new snapshot %s of device %s failed.",
3723 static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
3728 r = check_arg_count(argc, 2);
3732 r = read_dev_id(argv[1], &dev_id, 1);
3736 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
3738 DMWARN("Deletion of thin device %s failed.", argv[1]);
3743 static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
3745 dm_thin_id old_id, new_id;
3748 r = check_arg_count(argc, 3);
3752 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
3753 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
3757 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
3758 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
3762 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
3764 DMWARN("Failed to change transaction id from %s to %s.",
3772 static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3776 r = check_arg_count(argc, 1);
3780 (void) commit(pool);
3782 r = dm_pool_reserve_metadata_snap(pool->pmd);
3784 DMWARN("reserve_metadata_snap message failed.");
3789 static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3793 r = check_arg_count(argc, 1);
3797 r = dm_pool_release_metadata_snap(pool->pmd);
3799 DMWARN("release_metadata_snap message failed.");
3805 * Messages supported:
3806 * create_thin <dev_id>
3807 * create_snap <dev_id> <origin_id>
3809 * set_transaction_id <current_trans_id> <new_trans_id>
3810 * reserve_metadata_snap
3811 * release_metadata_snap
3813 static int pool_message(struct dm_target *ti, unsigned argc, char **argv,
3814 char *result, unsigned maxlen)
3817 struct pool_c *pt = ti->private;
3818 struct pool *pool = pt->pool;
3820 if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE) {
3821 DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode",
3822 dm_device_name(pool->pool_md));
3826 if (!strcasecmp(argv[0], "create_thin"))
3827 r = process_create_thin_mesg(argc, argv, pool);
3829 else if (!strcasecmp(argv[0], "create_snap"))
3830 r = process_create_snap_mesg(argc, argv, pool);
3832 else if (!strcasecmp(argv[0], "delete"))
3833 r = process_delete_mesg(argc, argv, pool);
3835 else if (!strcasecmp(argv[0], "set_transaction_id"))
3836 r = process_set_transaction_id_mesg(argc, argv, pool);
3838 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
3839 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
3841 else if (!strcasecmp(argv[0], "release_metadata_snap"))
3842 r = process_release_metadata_snap_mesg(argc, argv, pool);
3845 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
3848 (void) commit(pool);
3853 static void emit_flags(struct pool_features *pf, char *result,
3854 unsigned sz, unsigned maxlen)
3856 unsigned count = !pf->zero_new_blocks + !pf->discard_enabled +
3857 !pf->discard_passdown + (pf->mode == PM_READ_ONLY) +
3858 pf->error_if_no_space;
3859 DMEMIT("%u ", count);
3861 if (!pf->zero_new_blocks)
3862 DMEMIT("skip_block_zeroing ");
3864 if (!pf->discard_enabled)
3865 DMEMIT("ignore_discard ");
3867 if (!pf->discard_passdown)
3868 DMEMIT("no_discard_passdown ");
3870 if (pf->mode == PM_READ_ONLY)
3871 DMEMIT("read_only ");
3873 if (pf->error_if_no_space)
3874 DMEMIT("error_if_no_space ");
3879 * <transaction id> <used metadata sectors>/<total metadata sectors>
3880 * <used data sectors>/<total data sectors> <held metadata root>
3881 * <pool mode> <discard config> <no space config> <needs_check>
3883 static void pool_status(struct dm_target *ti, status_type_t type,
3884 unsigned status_flags, char *result, unsigned maxlen)
3888 uint64_t transaction_id;
3889 dm_block_t nr_free_blocks_data;
3890 dm_block_t nr_free_blocks_metadata;
3891 dm_block_t nr_blocks_data;
3892 dm_block_t nr_blocks_metadata;
3893 dm_block_t held_root;
3894 enum pool_mode mode;
3895 char buf[BDEVNAME_SIZE];
3896 char buf2[BDEVNAME_SIZE];
3897 struct pool_c *pt = ti->private;
3898 struct pool *pool = pt->pool;
3901 case STATUSTYPE_INFO:
3902 if (get_pool_mode(pool) == PM_FAIL) {
3907 /* Commit to ensure statistics aren't out-of-date */
3908 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
3909 (void) commit(pool);
3911 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
3913 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
3914 dm_device_name(pool->pool_md), r);
3918 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
3920 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
3921 dm_device_name(pool->pool_md), r);
3925 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
3927 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
3928 dm_device_name(pool->pool_md), r);
3932 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
3934 DMERR("%s: dm_pool_get_free_block_count returned %d",
3935 dm_device_name(pool->pool_md), r);
3939 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
3941 DMERR("%s: dm_pool_get_data_dev_size returned %d",
3942 dm_device_name(pool->pool_md), r);
3946 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
3948 DMERR("%s: dm_pool_get_metadata_snap returned %d",
3949 dm_device_name(pool->pool_md), r);
3953 DMEMIT("%llu %llu/%llu %llu/%llu ",
3954 (unsigned long long)transaction_id,
3955 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
3956 (unsigned long long)nr_blocks_metadata,
3957 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
3958 (unsigned long long)nr_blocks_data);
3961 DMEMIT("%llu ", held_root);
3965 mode = get_pool_mode(pool);
3966 if (mode == PM_OUT_OF_DATA_SPACE)
3967 DMEMIT("out_of_data_space ");
3968 else if (is_read_only_pool_mode(mode))
3973 if (!pool->pf.discard_enabled)
3974 DMEMIT("ignore_discard ");
3975 else if (pool->pf.discard_passdown)
3976 DMEMIT("discard_passdown ");
3978 DMEMIT("no_discard_passdown ");
3980 if (pool->pf.error_if_no_space)
3981 DMEMIT("error_if_no_space ");
3983 DMEMIT("queue_if_no_space ");
3985 if (dm_pool_metadata_needs_check(pool->pmd))
3986 DMEMIT("needs_check ");
3990 DMEMIT("%llu ", (unsigned long long)calc_metadata_threshold(pt));
3994 case STATUSTYPE_TABLE:
3995 DMEMIT("%s %s %lu %llu ",
3996 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
3997 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
3998 (unsigned long)pool->sectors_per_block,
3999 (unsigned long long)pt->low_water_blocks);
4000 emit_flags(&pt->requested_pf, result, sz, maxlen);
4003 case STATUSTYPE_IMA:
4013 static int pool_iterate_devices(struct dm_target *ti,
4014 iterate_devices_callout_fn fn, void *data)
4016 struct pool_c *pt = ti->private;
4018 return fn(ti, pt->data_dev, 0, ti->len, data);
4021 static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
4023 struct pool_c *pt = ti->private;
4024 struct pool *pool = pt->pool;
4025 sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
4028 * If max_sectors is smaller than pool->sectors_per_block adjust it
4029 * to the highest possible power-of-2 factor of pool->sectors_per_block.
4030 * This is especially beneficial when the pool's data device is a RAID
4031 * device that has a full stripe width that matches pool->sectors_per_block
4032 * -- because even though partial RAID stripe-sized IOs will be issued to a
4033 * single RAID stripe; when aggregated they will end on a full RAID stripe
4034 * boundary.. which avoids additional partial RAID stripe writes cascading
4036 if (limits->max_sectors < pool->sectors_per_block) {
4037 while (!is_factor(pool->sectors_per_block, limits->max_sectors)) {
4038 if ((limits->max_sectors & (limits->max_sectors - 1)) == 0)
4039 limits->max_sectors--;
4040 limits->max_sectors = rounddown_pow_of_two(limits->max_sectors);
4045 * If the system-determined stacked limits are compatible with the
4046 * pool's blocksize (io_opt is a factor) do not override them.
4048 if (io_opt_sectors < pool->sectors_per_block ||
4049 !is_factor(io_opt_sectors, pool->sectors_per_block)) {
4050 if (is_factor(pool->sectors_per_block, limits->max_sectors))
4051 blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT);
4053 blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT);
4054 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
4058 * pt->adjusted_pf is a staging area for the actual features to use.
4059 * They get transferred to the live pool in bind_control_target()
4060 * called from pool_preresume().
4062 if (!pt->adjusted_pf.discard_enabled) {
4064 * Must explicitly disallow stacking discard limits otherwise the
4065 * block layer will stack them if pool's data device has support.
4067 limits->discard_granularity = 0;
4071 disable_passdown_if_not_supported(pt);
4074 * The pool uses the same discard limits as the underlying data
4075 * device. DM core has already set this up.
4079 static struct target_type pool_target = {
4080 .name = "thin-pool",
4081 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
4082 DM_TARGET_IMMUTABLE,
4083 .version = {1, 22, 0},
4084 .module = THIS_MODULE,
4088 .presuspend = pool_presuspend,
4089 .presuspend_undo = pool_presuspend_undo,
4090 .postsuspend = pool_postsuspend,
4091 .preresume = pool_preresume,
4092 .resume = pool_resume,
4093 .message = pool_message,
4094 .status = pool_status,
4095 .iterate_devices = pool_iterate_devices,
4096 .io_hints = pool_io_hints,
4099 /*----------------------------------------------------------------
4100 * Thin target methods
4101 *--------------------------------------------------------------*/
4102 static void thin_get(struct thin_c *tc)
4104 refcount_inc(&tc->refcount);
4107 static void thin_put(struct thin_c *tc)
4109 if (refcount_dec_and_test(&tc->refcount))
4110 complete(&tc->can_destroy);
4113 static void thin_dtr(struct dm_target *ti)
4115 struct thin_c *tc = ti->private;
4117 spin_lock_irq(&tc->pool->lock);
4118 list_del_rcu(&tc->list);
4119 spin_unlock_irq(&tc->pool->lock);
4123 wait_for_completion(&tc->can_destroy);
4125 mutex_lock(&dm_thin_pool_table.mutex);
4127 __pool_dec(tc->pool);
4128 dm_pool_close_thin_device(tc->td);
4129 dm_put_device(ti, tc->pool_dev);
4131 dm_put_device(ti, tc->origin_dev);
4134 mutex_unlock(&dm_thin_pool_table.mutex);
4138 * Thin target parameters:
4140 * <pool_dev> <dev_id> [origin_dev]
4142 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
4143 * dev_id: the internal device identifier
4144 * origin_dev: a device external to the pool that should act as the origin
4146 * If the pool device has discards disabled, they get disabled for the thin
4149 static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
4153 struct dm_dev *pool_dev, *origin_dev;
4154 struct mapped_device *pool_md;
4156 mutex_lock(&dm_thin_pool_table.mutex);
4158 if (argc != 2 && argc != 3) {
4159 ti->error = "Invalid argument count";
4164 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
4166 ti->error = "Out of memory";
4170 tc->thin_md = dm_table_get_md(ti->table);
4171 spin_lock_init(&tc->lock);
4172 INIT_LIST_HEAD(&tc->deferred_cells);
4173 bio_list_init(&tc->deferred_bio_list);
4174 bio_list_init(&tc->retry_on_resume_list);
4175 tc->sort_bio_list = RB_ROOT;
4178 if (!strcmp(argv[0], argv[2])) {
4179 ti->error = "Error setting origin device";
4181 goto bad_origin_dev;
4184 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
4186 ti->error = "Error opening origin device";
4187 goto bad_origin_dev;
4189 tc->origin_dev = origin_dev;
4192 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
4194 ti->error = "Error opening pool device";
4197 tc->pool_dev = pool_dev;
4199 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
4200 ti->error = "Invalid device id";
4205 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
4207 ti->error = "Couldn't get pool mapped device";
4212 tc->pool = __pool_table_lookup(pool_md);
4214 ti->error = "Couldn't find pool object";
4216 goto bad_pool_lookup;
4218 __pool_inc(tc->pool);
4220 if (get_pool_mode(tc->pool) == PM_FAIL) {
4221 ti->error = "Couldn't open thin device, Pool is in fail mode";
4226 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
4228 ti->error = "Couldn't open thin internal device";
4232 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
4236 ti->num_flush_bios = 1;
4237 ti->flush_supported = true;
4238 ti->accounts_remapped_io = true;
4239 ti->per_io_data_size = sizeof(struct dm_thin_endio_hook);
4241 /* In case the pool supports discards, pass them on. */
4242 if (tc->pool->pf.discard_enabled) {
4243 ti->discards_supported = true;
4244 ti->num_discard_bios = 1;
4247 mutex_unlock(&dm_thin_pool_table.mutex);
4249 spin_lock_irq(&tc->pool->lock);
4250 if (tc->pool->suspended) {
4251 spin_unlock_irq(&tc->pool->lock);
4252 mutex_lock(&dm_thin_pool_table.mutex); /* reacquire for __pool_dec */
4253 ti->error = "Unable to activate thin device while pool is suspended";
4257 refcount_set(&tc->refcount, 1);
4258 init_completion(&tc->can_destroy);
4259 list_add_tail_rcu(&tc->list, &tc->pool->active_thins);
4260 spin_unlock_irq(&tc->pool->lock);
4262 * This synchronize_rcu() call is needed here otherwise we risk a
4263 * wake_worker() call finding no bios to process (because the newly
4264 * added tc isn't yet visible). So this reduces latency since we
4265 * aren't then dependent on the periodic commit to wake_worker().
4274 dm_pool_close_thin_device(tc->td);
4276 __pool_dec(tc->pool);
4280 dm_put_device(ti, tc->pool_dev);
4283 dm_put_device(ti, tc->origin_dev);
4287 mutex_unlock(&dm_thin_pool_table.mutex);
4292 static int thin_map(struct dm_target *ti, struct bio *bio)
4294 bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
4296 return thin_bio_map(ti, bio);
4299 static int thin_endio(struct dm_target *ti, struct bio *bio,
4302 unsigned long flags;
4303 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
4304 struct list_head work;
4305 struct dm_thin_new_mapping *m, *tmp;
4306 struct pool *pool = h->tc->pool;
4308 if (h->shared_read_entry) {
4309 INIT_LIST_HEAD(&work);
4310 dm_deferred_entry_dec(h->shared_read_entry, &work);
4312 spin_lock_irqsave(&pool->lock, flags);
4313 list_for_each_entry_safe(m, tmp, &work, list) {
4315 __complete_mapping_preparation(m);
4317 spin_unlock_irqrestore(&pool->lock, flags);
4320 if (h->all_io_entry) {
4321 INIT_LIST_HEAD(&work);
4322 dm_deferred_entry_dec(h->all_io_entry, &work);
4323 if (!list_empty(&work)) {
4324 spin_lock_irqsave(&pool->lock, flags);
4325 list_for_each_entry_safe(m, tmp, &work, list)
4326 list_add_tail(&m->list, &pool->prepared_discards);
4327 spin_unlock_irqrestore(&pool->lock, flags);
4333 cell_defer_no_holder(h->tc, h->cell);
4335 return DM_ENDIO_DONE;
4338 static void thin_presuspend(struct dm_target *ti)
4340 struct thin_c *tc = ti->private;
4342 if (dm_noflush_suspending(ti))
4343 noflush_work(tc, do_noflush_start);
4346 static void thin_postsuspend(struct dm_target *ti)
4348 struct thin_c *tc = ti->private;
4351 * The dm_noflush_suspending flag has been cleared by now, so
4352 * unfortunately we must always run this.
4354 noflush_work(tc, do_noflush_stop);
4357 static int thin_preresume(struct dm_target *ti)
4359 struct thin_c *tc = ti->private;
4362 tc->origin_size = get_dev_size(tc->origin_dev->bdev);
4368 * <nr mapped sectors> <highest mapped sector>
4370 static void thin_status(struct dm_target *ti, status_type_t type,
4371 unsigned status_flags, char *result, unsigned maxlen)
4375 dm_block_t mapped, highest;
4376 char buf[BDEVNAME_SIZE];
4377 struct thin_c *tc = ti->private;
4379 if (get_pool_mode(tc->pool) == PM_FAIL) {
4388 case STATUSTYPE_INFO:
4389 r = dm_thin_get_mapped_count(tc->td, &mapped);
4391 DMERR("dm_thin_get_mapped_count returned %d", r);
4395 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
4397 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
4401 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
4403 DMEMIT("%llu", ((highest + 1) *
4404 tc->pool->sectors_per_block) - 1);
4409 case STATUSTYPE_TABLE:
4411 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
4412 (unsigned long) tc->dev_id);
4414 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
4417 case STATUSTYPE_IMA:
4429 static int thin_iterate_devices(struct dm_target *ti,
4430 iterate_devices_callout_fn fn, void *data)
4433 struct thin_c *tc = ti->private;
4434 struct pool *pool = tc->pool;
4437 * We can't call dm_pool_get_data_dev_size() since that blocks. So
4438 * we follow a more convoluted path through to the pool's target.
4441 return 0; /* nothing is bound */
4443 blocks = pool->ti->len;
4444 (void) sector_div(blocks, pool->sectors_per_block);
4446 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
4451 static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits)
4453 struct thin_c *tc = ti->private;
4454 struct pool *pool = tc->pool;
4456 if (!pool->pf.discard_enabled)
4459 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
4460 limits->max_discard_sectors = 2048 * 1024 * 16; /* 16G */
4463 static struct target_type thin_target = {
4465 .version = {1, 22, 0},
4466 .module = THIS_MODULE,
4470 .end_io = thin_endio,
4471 .preresume = thin_preresume,
4472 .presuspend = thin_presuspend,
4473 .postsuspend = thin_postsuspend,
4474 .status = thin_status,
4475 .iterate_devices = thin_iterate_devices,
4476 .io_hints = thin_io_hints,
4479 /*----------------------------------------------------------------*/
4481 static int __init dm_thin_init(void)
4487 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
4488 if (!_new_mapping_cache)
4491 r = dm_register_target(&thin_target);
4493 goto bad_new_mapping_cache;
4495 r = dm_register_target(&pool_target);
4497 goto bad_thin_target;
4502 dm_unregister_target(&thin_target);
4503 bad_new_mapping_cache:
4504 kmem_cache_destroy(_new_mapping_cache);
4509 static void dm_thin_exit(void)
4511 dm_unregister_target(&thin_target);
4512 dm_unregister_target(&pool_target);
4514 kmem_cache_destroy(_new_mapping_cache);
4519 module_init(dm_thin_init);
4520 module_exit(dm_thin_exit);
4522 module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR);
4523 MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds");
4525 MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
4526 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
4527 MODULE_LICENSE("GPL");