2 * Copyright (C) 2011-2012 Red Hat UK.
4 * This file is released under the GPL.
7 #include "dm-thin-metadata.h"
8 #include "dm-bio-prison-v1.h"
11 #include <linux/device-mapper.h>
12 #include <linux/dm-io.h>
13 #include <linux/dm-kcopyd.h>
14 #include <linux/jiffies.h>
15 #include <linux/log2.h>
16 #include <linux/list.h>
17 #include <linux/rculist.h>
18 #include <linux/init.h>
19 #include <linux/module.h>
20 #include <linux/slab.h>
21 #include <linux/vmalloc.h>
22 #include <linux/sort.h>
23 #include <linux/rbtree.h>
25 #define DM_MSG_PREFIX "thin"
30 #define ENDIO_HOOK_POOL_SIZE 1024
31 #define MAPPING_POOL_SIZE 1024
32 #define COMMIT_PERIOD HZ
33 #define NO_SPACE_TIMEOUT_SECS 60
35 static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS;
37 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
38 "A percentage of time allocated for copy on write");
41 * The block size of the device holding pool data must be
42 * between 64KB and 1GB.
44 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
45 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
48 * Device id is restricted to 24 bits.
50 #define MAX_DEV_ID ((1 << 24) - 1)
53 * How do we handle breaking sharing of data blocks?
54 * =================================================
56 * We use a standard copy-on-write btree to store the mappings for the
57 * devices (note I'm talking about copy-on-write of the metadata here, not
58 * the data). When you take an internal snapshot you clone the root node
59 * of the origin btree. After this there is no concept of an origin or a
60 * snapshot. They are just two device trees that happen to point to the
63 * When we get a write in we decide if it's to a shared data block using
64 * some timestamp magic. If it is, we have to break sharing.
66 * Let's say we write to a shared block in what was the origin. The
69 * i) plug io further to this physical block. (see bio_prison code).
71 * ii) quiesce any read io to that shared data block. Obviously
72 * including all devices that share this block. (see dm_deferred_set code)
74 * iii) copy the data block to a newly allocate block. This step can be
75 * missed out if the io covers the block. (schedule_copy).
77 * iv) insert the new mapping into the origin's btree
78 * (process_prepared_mapping). This act of inserting breaks some
79 * sharing of btree nodes between the two devices. Breaking sharing only
80 * effects the btree of that specific device. Btrees for the other
81 * devices that share the block never change. The btree for the origin
82 * device as it was after the last commit is untouched, ie. we're using
83 * persistent data structures in the functional programming sense.
85 * v) unplug io to this physical block, including the io that triggered
86 * the breaking of sharing.
88 * Steps (ii) and (iii) occur in parallel.
90 * The metadata _doesn't_ need to be committed before the io continues. We
91 * get away with this because the io is always written to a _new_ block.
92 * If there's a crash, then:
94 * - The origin mapping will point to the old origin block (the shared
95 * one). This will contain the data as it was before the io that triggered
96 * the breaking of sharing came in.
98 * - The snap mapping still points to the old block. As it would after
101 * The downside of this scheme is the timestamp magic isn't perfect, and
102 * will continue to think that data block in the snapshot device is shared
103 * even after the write to the origin has broken sharing. I suspect data
104 * blocks will typically be shared by many different devices, so we're
105 * breaking sharing n + 1 times, rather than n, where n is the number of
106 * devices that reference this data block. At the moment I think the
107 * benefits far, far outweigh the disadvantages.
110 /*----------------------------------------------------------------*/
120 static void build_key(struct dm_thin_device *td, enum lock_space ls,
121 dm_block_t b, dm_block_t e, struct dm_cell_key *key)
123 key->virtual = (ls == VIRTUAL);
124 key->dev = dm_thin_dev_id(td);
125 key->block_begin = b;
129 static void build_data_key(struct dm_thin_device *td, dm_block_t b,
130 struct dm_cell_key *key)
132 build_key(td, PHYSICAL, b, b + 1llu, key);
135 static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
136 struct dm_cell_key *key)
138 build_key(td, VIRTUAL, b, b + 1llu, key);
141 /*----------------------------------------------------------------*/
143 #define THROTTLE_THRESHOLD (1 * HZ)
146 struct rw_semaphore lock;
147 unsigned long threshold;
148 bool throttle_applied;
151 static void throttle_init(struct throttle *t)
153 init_rwsem(&t->lock);
154 t->throttle_applied = false;
157 static void throttle_work_start(struct throttle *t)
159 t->threshold = jiffies + THROTTLE_THRESHOLD;
162 static void throttle_work_update(struct throttle *t)
164 if (!t->throttle_applied && jiffies > t->threshold) {
165 down_write(&t->lock);
166 t->throttle_applied = true;
170 static void throttle_work_complete(struct throttle *t)
172 if (t->throttle_applied) {
173 t->throttle_applied = false;
178 static void throttle_lock(struct throttle *t)
183 static void throttle_unlock(struct throttle *t)
188 /*----------------------------------------------------------------*/
191 * A pool device ties together a metadata device and a data device. It
192 * also provides the interface for creating and destroying internal
195 struct dm_thin_new_mapping;
198 * The pool runs in various modes. Ordered in degraded order for comparisons.
201 PM_WRITE, /* metadata may be changed */
202 PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */
205 * Like READ_ONLY, except may switch back to WRITE on metadata resize. Reported as READ_ONLY.
207 PM_OUT_OF_METADATA_SPACE,
208 PM_READ_ONLY, /* metadata may not be changed */
210 PM_FAIL, /* all I/O fails */
213 struct pool_features {
216 bool zero_new_blocks:1;
217 bool discard_enabled:1;
218 bool discard_passdown:1;
219 bool error_if_no_space:1;
223 typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
224 typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell);
225 typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);
227 #define CELL_SORT_ARRAY_SIZE 8192
230 struct list_head list;
231 struct dm_target *ti; /* Only set if a pool target is bound */
233 struct mapped_device *pool_md;
234 struct block_device *md_dev;
235 struct dm_pool_metadata *pmd;
237 dm_block_t low_water_blocks;
238 uint32_t sectors_per_block;
239 int sectors_per_block_shift;
241 struct pool_features pf;
242 bool low_water_triggered:1; /* A dm event has been sent */
244 bool out_of_data_space:1;
246 struct dm_bio_prison *prison;
247 struct dm_kcopyd_client *copier;
249 struct work_struct worker;
250 struct workqueue_struct *wq;
251 struct throttle throttle;
252 struct delayed_work waker;
253 struct delayed_work no_space_timeout;
255 unsigned long last_commit_jiffies;
259 struct bio_list deferred_flush_bios;
260 struct bio_list deferred_flush_completions;
261 struct list_head prepared_mappings;
262 struct list_head prepared_discards;
263 struct list_head prepared_discards_pt2;
264 struct list_head active_thins;
266 struct dm_deferred_set *shared_read_ds;
267 struct dm_deferred_set *all_io_ds;
269 struct dm_thin_new_mapping *next_mapping;
271 process_bio_fn process_bio;
272 process_bio_fn process_discard;
274 process_cell_fn process_cell;
275 process_cell_fn process_discard_cell;
277 process_mapping_fn process_prepared_mapping;
278 process_mapping_fn process_prepared_discard;
279 process_mapping_fn process_prepared_discard_pt2;
281 struct dm_bio_prison_cell **cell_sort_array;
283 mempool_t mapping_pool;
286 static void metadata_operation_failed(struct pool *pool, const char *op, int r);
288 static enum pool_mode get_pool_mode(struct pool *pool)
290 return pool->pf.mode;
293 static void notify_of_pool_mode_change(struct pool *pool)
295 const char *descs[] = {
302 const char *extra_desc = NULL;
303 enum pool_mode mode = get_pool_mode(pool);
305 if (mode == PM_OUT_OF_DATA_SPACE) {
306 if (!pool->pf.error_if_no_space)
307 extra_desc = " (queue IO)";
309 extra_desc = " (error IO)";
312 dm_table_event(pool->ti->table);
313 DMINFO("%s: switching pool to %s%s mode",
314 dm_device_name(pool->pool_md),
315 descs[(int)mode], extra_desc ? : "");
319 * Target context for a pool.
322 struct dm_target *ti;
324 struct dm_dev *data_dev;
325 struct dm_dev *metadata_dev;
326 struct dm_target_callbacks callbacks;
328 dm_block_t low_water_blocks;
329 struct pool_features requested_pf; /* Features requested during table load */
330 struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */
334 * Target context for a thin.
337 struct list_head list;
338 struct dm_dev *pool_dev;
339 struct dm_dev *origin_dev;
340 sector_t origin_size;
344 struct dm_thin_device *td;
345 struct mapped_device *thin_md;
349 struct list_head deferred_cells;
350 struct bio_list deferred_bio_list;
351 struct bio_list retry_on_resume_list;
352 struct rb_root sort_bio_list; /* sorted list of deferred bios */
355 * Ensures the thin is not destroyed until the worker has finished
356 * iterating the active_thins list.
359 struct completion can_destroy;
362 /*----------------------------------------------------------------*/
364 static bool block_size_is_power_of_two(struct pool *pool)
366 return pool->sectors_per_block_shift >= 0;
369 static sector_t block_to_sectors(struct pool *pool, dm_block_t b)
371 return block_size_is_power_of_two(pool) ?
372 (b << pool->sectors_per_block_shift) :
373 (b * pool->sectors_per_block);
376 /*----------------------------------------------------------------*/
380 struct blk_plug plug;
381 struct bio *parent_bio;
385 static void begin_discard(struct discard_op *op, struct thin_c *tc, struct bio *parent)
390 blk_start_plug(&op->plug);
391 op->parent_bio = parent;
395 static int issue_discard(struct discard_op *op, dm_block_t data_b, dm_block_t data_e)
397 struct thin_c *tc = op->tc;
398 sector_t s = block_to_sectors(tc->pool, data_b);
399 sector_t len = block_to_sectors(tc->pool, data_e - data_b);
401 return __blkdev_issue_discard(tc->pool_dev->bdev, s, len,
402 GFP_NOWAIT, 0, &op->bio);
405 static void end_discard(struct discard_op *op, int r)
409 * Even if one of the calls to issue_discard failed, we
410 * need to wait for the chain to complete.
412 bio_chain(op->bio, op->parent_bio);
413 bio_set_op_attrs(op->bio, REQ_OP_DISCARD, 0);
417 blk_finish_plug(&op->plug);
420 * Even if r is set, there could be sub discards in flight that we
423 if (r && !op->parent_bio->bi_status)
424 op->parent_bio->bi_status = errno_to_blk_status(r);
425 bio_endio(op->parent_bio);
428 /*----------------------------------------------------------------*/
431 * wake_worker() is used when new work is queued and when pool_resume is
432 * ready to continue deferred IO processing.
434 static void wake_worker(struct pool *pool)
436 queue_work(pool->wq, &pool->worker);
439 /*----------------------------------------------------------------*/
441 static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio,
442 struct dm_bio_prison_cell **cell_result)
445 struct dm_bio_prison_cell *cell_prealloc;
448 * Allocate a cell from the prison's mempool.
449 * This might block but it can't fail.
451 cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO);
453 r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result);
456 * We reused an old cell; we can get rid of
459 dm_bio_prison_free_cell(pool->prison, cell_prealloc);
464 static void cell_release(struct pool *pool,
465 struct dm_bio_prison_cell *cell,
466 struct bio_list *bios)
468 dm_cell_release(pool->prison, cell, bios);
469 dm_bio_prison_free_cell(pool->prison, cell);
472 static void cell_visit_release(struct pool *pool,
473 void (*fn)(void *, struct dm_bio_prison_cell *),
475 struct dm_bio_prison_cell *cell)
477 dm_cell_visit_release(pool->prison, fn, context, cell);
478 dm_bio_prison_free_cell(pool->prison, cell);
481 static void cell_release_no_holder(struct pool *pool,
482 struct dm_bio_prison_cell *cell,
483 struct bio_list *bios)
485 dm_cell_release_no_holder(pool->prison, cell, bios);
486 dm_bio_prison_free_cell(pool->prison, cell);
489 static void cell_error_with_code(struct pool *pool,
490 struct dm_bio_prison_cell *cell, blk_status_t error_code)
492 dm_cell_error(pool->prison, cell, error_code);
493 dm_bio_prison_free_cell(pool->prison, cell);
496 static blk_status_t get_pool_io_error_code(struct pool *pool)
498 return pool->out_of_data_space ? BLK_STS_NOSPC : BLK_STS_IOERR;
501 static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell)
503 cell_error_with_code(pool, cell, get_pool_io_error_code(pool));
506 static void cell_success(struct pool *pool, struct dm_bio_prison_cell *cell)
508 cell_error_with_code(pool, cell, 0);
511 static void cell_requeue(struct pool *pool, struct dm_bio_prison_cell *cell)
513 cell_error_with_code(pool, cell, BLK_STS_DM_REQUEUE);
516 /*----------------------------------------------------------------*/
519 * A global list of pools that uses a struct mapped_device as a key.
521 static struct dm_thin_pool_table {
523 struct list_head pools;
524 } dm_thin_pool_table;
526 static void pool_table_init(void)
528 mutex_init(&dm_thin_pool_table.mutex);
529 INIT_LIST_HEAD(&dm_thin_pool_table.pools);
532 static void pool_table_exit(void)
534 mutex_destroy(&dm_thin_pool_table.mutex);
537 static void __pool_table_insert(struct pool *pool)
539 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
540 list_add(&pool->list, &dm_thin_pool_table.pools);
543 static void __pool_table_remove(struct pool *pool)
545 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
546 list_del(&pool->list);
549 static struct pool *__pool_table_lookup(struct mapped_device *md)
551 struct pool *pool = NULL, *tmp;
553 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
555 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
556 if (tmp->pool_md == md) {
565 static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
567 struct pool *pool = NULL, *tmp;
569 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
571 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
572 if (tmp->md_dev == md_dev) {
581 /*----------------------------------------------------------------*/
583 struct dm_thin_endio_hook {
585 struct dm_deferred_entry *shared_read_entry;
586 struct dm_deferred_entry *all_io_entry;
587 struct dm_thin_new_mapping *overwrite_mapping;
588 struct rb_node rb_node;
589 struct dm_bio_prison_cell *cell;
592 static void __merge_bio_list(struct bio_list *bios, struct bio_list *master)
594 bio_list_merge(bios, master);
595 bio_list_init(master);
598 static void error_bio_list(struct bio_list *bios, blk_status_t error)
602 while ((bio = bio_list_pop(bios))) {
603 bio->bi_status = error;
608 static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master,
611 struct bio_list bios;
614 bio_list_init(&bios);
616 spin_lock_irqsave(&tc->lock, flags);
617 __merge_bio_list(&bios, master);
618 spin_unlock_irqrestore(&tc->lock, flags);
620 error_bio_list(&bios, error);
623 static void requeue_deferred_cells(struct thin_c *tc)
625 struct pool *pool = tc->pool;
627 struct list_head cells;
628 struct dm_bio_prison_cell *cell, *tmp;
630 INIT_LIST_HEAD(&cells);
632 spin_lock_irqsave(&tc->lock, flags);
633 list_splice_init(&tc->deferred_cells, &cells);
634 spin_unlock_irqrestore(&tc->lock, flags);
636 list_for_each_entry_safe(cell, tmp, &cells, user_list)
637 cell_requeue(pool, cell);
640 static void requeue_io(struct thin_c *tc)
642 struct bio_list bios;
645 bio_list_init(&bios);
647 spin_lock_irqsave(&tc->lock, flags);
648 __merge_bio_list(&bios, &tc->deferred_bio_list);
649 __merge_bio_list(&bios, &tc->retry_on_resume_list);
650 spin_unlock_irqrestore(&tc->lock, flags);
652 error_bio_list(&bios, BLK_STS_DM_REQUEUE);
653 requeue_deferred_cells(tc);
656 static void error_retry_list_with_code(struct pool *pool, blk_status_t error)
661 list_for_each_entry_rcu(tc, &pool->active_thins, list)
662 error_thin_bio_list(tc, &tc->retry_on_resume_list, error);
666 static void error_retry_list(struct pool *pool)
668 error_retry_list_with_code(pool, get_pool_io_error_code(pool));
672 * This section of code contains the logic for processing a thin device's IO.
673 * Much of the code depends on pool object resources (lists, workqueues, etc)
674 * but most is exclusively called from the thin target rather than the thin-pool
678 static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
680 struct pool *pool = tc->pool;
681 sector_t block_nr = bio->bi_iter.bi_sector;
683 if (block_size_is_power_of_two(pool))
684 block_nr >>= pool->sectors_per_block_shift;
686 (void) sector_div(block_nr, pool->sectors_per_block);
692 * Returns the _complete_ blocks that this bio covers.
694 static void get_bio_block_range(struct thin_c *tc, struct bio *bio,
695 dm_block_t *begin, dm_block_t *end)
697 struct pool *pool = tc->pool;
698 sector_t b = bio->bi_iter.bi_sector;
699 sector_t e = b + (bio->bi_iter.bi_size >> SECTOR_SHIFT);
701 b += pool->sectors_per_block - 1ull; /* so we round up */
703 if (block_size_is_power_of_two(pool)) {
704 b >>= pool->sectors_per_block_shift;
705 e >>= pool->sectors_per_block_shift;
707 (void) sector_div(b, pool->sectors_per_block);
708 (void) sector_div(e, pool->sectors_per_block);
712 /* Can happen if the bio is within a single block. */
719 static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
721 struct pool *pool = tc->pool;
722 sector_t bi_sector = bio->bi_iter.bi_sector;
724 bio_set_dev(bio, tc->pool_dev->bdev);
725 if (block_size_is_power_of_two(pool))
726 bio->bi_iter.bi_sector =
727 (block << pool->sectors_per_block_shift) |
728 (bi_sector & (pool->sectors_per_block - 1));
730 bio->bi_iter.bi_sector = (block * pool->sectors_per_block) +
731 sector_div(bi_sector, pool->sectors_per_block);
734 static void remap_to_origin(struct thin_c *tc, struct bio *bio)
736 bio_set_dev(bio, tc->origin_dev->bdev);
739 static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
741 return op_is_flush(bio->bi_opf) &&
742 dm_thin_changed_this_transaction(tc->td);
745 static void inc_all_io_entry(struct pool *pool, struct bio *bio)
747 struct dm_thin_endio_hook *h;
749 if (bio_op(bio) == REQ_OP_DISCARD)
752 h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
753 h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
756 static void issue(struct thin_c *tc, struct bio *bio)
758 struct pool *pool = tc->pool;
761 if (!bio_triggers_commit(tc, bio)) {
762 generic_make_request(bio);
767 * Complete bio with an error if earlier I/O caused changes to
768 * the metadata that can't be committed e.g, due to I/O errors
769 * on the metadata device.
771 if (dm_thin_aborted_changes(tc->td)) {
777 * Batch together any bios that trigger commits and then issue a
778 * single commit for them in process_deferred_bios().
780 spin_lock_irqsave(&pool->lock, flags);
781 bio_list_add(&pool->deferred_flush_bios, bio);
782 spin_unlock_irqrestore(&pool->lock, flags);
785 static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
787 remap_to_origin(tc, bio);
791 static void remap_and_issue(struct thin_c *tc, struct bio *bio,
794 remap(tc, bio, block);
798 /*----------------------------------------------------------------*/
801 * Bio endio functions.
803 struct dm_thin_new_mapping {
804 struct list_head list;
810 * Track quiescing, copying and zeroing preparation actions. When this
811 * counter hits zero the block is prepared and can be inserted into the
814 atomic_t prepare_actions;
818 dm_block_t virt_begin, virt_end;
819 dm_block_t data_block;
820 struct dm_bio_prison_cell *cell;
823 * If the bio covers the whole area of a block then we can avoid
824 * zeroing or copying. Instead this bio is hooked. The bio will
825 * still be in the cell, so care has to be taken to avoid issuing
829 bio_end_io_t *saved_bi_end_io;
832 static void __complete_mapping_preparation(struct dm_thin_new_mapping *m)
834 struct pool *pool = m->tc->pool;
836 if (atomic_dec_and_test(&m->prepare_actions)) {
837 list_add_tail(&m->list, &pool->prepared_mappings);
842 static void complete_mapping_preparation(struct dm_thin_new_mapping *m)
845 struct pool *pool = m->tc->pool;
847 spin_lock_irqsave(&pool->lock, flags);
848 __complete_mapping_preparation(m);
849 spin_unlock_irqrestore(&pool->lock, flags);
852 static void copy_complete(int read_err, unsigned long write_err, void *context)
854 struct dm_thin_new_mapping *m = context;
856 m->status = read_err || write_err ? BLK_STS_IOERR : 0;
857 complete_mapping_preparation(m);
860 static void overwrite_endio(struct bio *bio)
862 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
863 struct dm_thin_new_mapping *m = h->overwrite_mapping;
865 bio->bi_end_io = m->saved_bi_end_io;
867 m->status = bio->bi_status;
868 complete_mapping_preparation(m);
871 /*----------------------------------------------------------------*/
878 * Prepared mapping jobs.
882 * This sends the bios in the cell, except the original holder, back
883 * to the deferred_bios list.
885 static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
887 struct pool *pool = tc->pool;
890 spin_lock_irqsave(&tc->lock, flags);
891 cell_release_no_holder(pool, cell, &tc->deferred_bio_list);
892 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;
966 * If the bio has the REQ_FUA flag set we must commit the metadata
967 * before signaling its completion.
969 if (!bio_triggers_commit(tc, bio)) {
975 * Complete bio with an error if earlier I/O caused changes to the
976 * metadata that can't be committed, e.g, due to I/O errors on the
979 if (dm_thin_aborted_changes(tc->td)) {
985 * Batch together any bios that trigger commits and then issue a
986 * single commit for them in process_deferred_bios().
988 spin_lock_irqsave(&pool->lock, flags);
989 bio_list_add(&pool->deferred_flush_completions, bio);
990 spin_unlock_irqrestore(&pool->lock, flags);
993 static void process_prepared_mapping(struct dm_thin_new_mapping *m)
995 struct thin_c *tc = m->tc;
996 struct pool *pool = tc->pool;
997 struct bio *bio = m->bio;
1001 cell_error(pool, m->cell);
1006 * Commit the prepared block into the mapping btree.
1007 * Any I/O for this block arriving after this point will get
1008 * remapped to it directly.
1010 r = dm_thin_insert_block(tc->td, m->virt_begin, m->data_block);
1012 metadata_operation_failed(pool, "dm_thin_insert_block", r);
1013 cell_error(pool, m->cell);
1018 * Release any bios held while the block was being provisioned.
1019 * If we are processing a write bio that completely covers the block,
1020 * we already processed it so can ignore it now when processing
1021 * the bios in the cell.
1024 inc_remap_and_issue_cell(tc, m->cell, m->data_block);
1025 complete_overwrite_bio(tc, bio);
1027 inc_all_io_entry(tc->pool, m->cell->holder);
1028 remap_and_issue(tc, m->cell->holder, m->data_block);
1029 inc_remap_and_issue_cell(tc, m->cell, m->data_block);
1034 mempool_free(m, &pool->mapping_pool);
1037 /*----------------------------------------------------------------*/
1039 static void free_discard_mapping(struct dm_thin_new_mapping *m)
1041 struct thin_c *tc = m->tc;
1043 cell_defer_no_holder(tc, m->cell);
1044 mempool_free(m, &tc->pool->mapping_pool);
1047 static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
1049 bio_io_error(m->bio);
1050 free_discard_mapping(m);
1053 static void process_prepared_discard_success(struct dm_thin_new_mapping *m)
1056 free_discard_mapping(m);
1059 static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m)
1062 struct thin_c *tc = m->tc;
1064 r = dm_thin_remove_range(tc->td, m->cell->key.block_begin, m->cell->key.block_end);
1066 metadata_operation_failed(tc->pool, "dm_thin_remove_range", r);
1067 bio_io_error(m->bio);
1071 cell_defer_no_holder(tc, m->cell);
1072 mempool_free(m, &tc->pool->mapping_pool);
1075 /*----------------------------------------------------------------*/
1077 static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m,
1078 struct bio *discard_parent)
1081 * We've already unmapped this range of blocks, but before we
1082 * passdown we have to check that these blocks are now unused.
1086 struct thin_c *tc = m->tc;
1087 struct pool *pool = tc->pool;
1088 dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin;
1089 struct discard_op op;
1091 begin_discard(&op, tc, discard_parent);
1093 /* find start of unmapped run */
1094 for (; b < end; b++) {
1095 r = dm_pool_block_is_shared(pool->pmd, b, &shared);
1106 /* find end of run */
1107 for (e = b + 1; e != end; e++) {
1108 r = dm_pool_block_is_shared(pool->pmd, e, &shared);
1116 r = issue_discard(&op, b, e);
1123 end_discard(&op, r);
1126 static void queue_passdown_pt2(struct dm_thin_new_mapping *m)
1128 unsigned long flags;
1129 struct pool *pool = m->tc->pool;
1131 spin_lock_irqsave(&pool->lock, flags);
1132 list_add_tail(&m->list, &pool->prepared_discards_pt2);
1133 spin_unlock_irqrestore(&pool->lock, flags);
1137 static void passdown_endio(struct bio *bio)
1140 * It doesn't matter if the passdown discard failed, we still want
1141 * to unmap (we ignore err).
1143 queue_passdown_pt2(bio->bi_private);
1147 static void process_prepared_discard_passdown_pt1(struct dm_thin_new_mapping *m)
1150 struct thin_c *tc = m->tc;
1151 struct pool *pool = tc->pool;
1152 struct bio *discard_parent;
1153 dm_block_t data_end = m->data_block + (m->virt_end - m->virt_begin);
1156 * Only this thread allocates blocks, so we can be sure that the
1157 * newly unmapped blocks will not be allocated before the end of
1160 r = dm_thin_remove_range(tc->td, m->virt_begin, m->virt_end);
1162 metadata_operation_failed(pool, "dm_thin_remove_range", r);
1163 bio_io_error(m->bio);
1164 cell_defer_no_holder(tc, m->cell);
1165 mempool_free(m, &pool->mapping_pool);
1170 * Increment the unmapped blocks. This prevents a race between the
1171 * passdown io and reallocation of freed blocks.
1173 r = dm_pool_inc_data_range(pool->pmd, m->data_block, data_end);
1175 metadata_operation_failed(pool, "dm_pool_inc_data_range", r);
1176 bio_io_error(m->bio);
1177 cell_defer_no_holder(tc, m->cell);
1178 mempool_free(m, &pool->mapping_pool);
1182 discard_parent = bio_alloc(GFP_NOIO, 1);
1183 if (!discard_parent) {
1184 DMWARN("%s: unable to allocate top level discard bio for passdown. Skipping passdown.",
1185 dm_device_name(tc->pool->pool_md));
1186 queue_passdown_pt2(m);
1189 discard_parent->bi_end_io = passdown_endio;
1190 discard_parent->bi_private = m;
1192 if (m->maybe_shared)
1193 passdown_double_checking_shared_status(m, discard_parent);
1195 struct discard_op op;
1197 begin_discard(&op, tc, discard_parent);
1198 r = issue_discard(&op, m->data_block, data_end);
1199 end_discard(&op, r);
1204 static void process_prepared_discard_passdown_pt2(struct dm_thin_new_mapping *m)
1207 struct thin_c *tc = m->tc;
1208 struct pool *pool = tc->pool;
1211 * The passdown has completed, so now we can decrement all those
1214 r = dm_pool_dec_data_range(pool->pmd, m->data_block,
1215 m->data_block + (m->virt_end - m->virt_begin));
1217 metadata_operation_failed(pool, "dm_pool_dec_data_range", r);
1218 bio_io_error(m->bio);
1222 cell_defer_no_holder(tc, m->cell);
1223 mempool_free(m, &pool->mapping_pool);
1226 static void process_prepared(struct pool *pool, struct list_head *head,
1227 process_mapping_fn *fn)
1229 unsigned long flags;
1230 struct list_head maps;
1231 struct dm_thin_new_mapping *m, *tmp;
1233 INIT_LIST_HEAD(&maps);
1234 spin_lock_irqsave(&pool->lock, flags);
1235 list_splice_init(head, &maps);
1236 spin_unlock_irqrestore(&pool->lock, flags);
1238 list_for_each_entry_safe(m, tmp, &maps, list)
1243 * Deferred bio jobs.
1245 static int io_overlaps_block(struct pool *pool, struct bio *bio)
1247 return bio->bi_iter.bi_size ==
1248 (pool->sectors_per_block << SECTOR_SHIFT);
1251 static int io_overwrites_block(struct pool *pool, struct bio *bio)
1253 return (bio_data_dir(bio) == WRITE) &&
1254 io_overlaps_block(pool, bio);
1257 static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
1260 *save = bio->bi_end_io;
1261 bio->bi_end_io = fn;
1264 static int ensure_next_mapping(struct pool *pool)
1266 if (pool->next_mapping)
1269 pool->next_mapping = mempool_alloc(&pool->mapping_pool, GFP_ATOMIC);
1271 return pool->next_mapping ? 0 : -ENOMEM;
1274 static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
1276 struct dm_thin_new_mapping *m = pool->next_mapping;
1278 BUG_ON(!pool->next_mapping);
1280 memset(m, 0, sizeof(struct dm_thin_new_mapping));
1281 INIT_LIST_HEAD(&m->list);
1284 pool->next_mapping = NULL;
1289 static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m,
1290 sector_t begin, sector_t end)
1292 struct dm_io_region to;
1294 to.bdev = tc->pool_dev->bdev;
1296 to.count = end - begin;
1298 dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m);
1301 static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio,
1302 dm_block_t data_begin,
1303 struct dm_thin_new_mapping *m)
1305 struct pool *pool = tc->pool;
1306 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1308 h->overwrite_mapping = m;
1310 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
1311 inc_all_io_entry(pool, bio);
1312 remap_and_issue(tc, bio, data_begin);
1316 * A partial copy also needs to zero the uncopied region.
1318 static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
1319 struct dm_dev *origin, dm_block_t data_origin,
1320 dm_block_t data_dest,
1321 struct dm_bio_prison_cell *cell, struct bio *bio,
1324 struct pool *pool = tc->pool;
1325 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1328 m->virt_begin = virt_block;
1329 m->virt_end = virt_block + 1u;
1330 m->data_block = data_dest;
1334 * quiesce action + copy action + an extra reference held for the
1335 * duration of this function (we may need to inc later for a
1338 atomic_set(&m->prepare_actions, 3);
1340 if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
1341 complete_mapping_preparation(m); /* already quiesced */
1344 * IO to pool_dev remaps to the pool target's data_dev.
1346 * If the whole block of data is being overwritten, we can issue the
1347 * bio immediately. Otherwise we use kcopyd to clone the data first.
1349 if (io_overwrites_block(pool, bio))
1350 remap_and_issue_overwrite(tc, bio, data_dest, m);
1352 struct dm_io_region from, to;
1354 from.bdev = origin->bdev;
1355 from.sector = data_origin * pool->sectors_per_block;
1358 to.bdev = tc->pool_dev->bdev;
1359 to.sector = data_dest * pool->sectors_per_block;
1362 dm_kcopyd_copy(pool->copier, &from, 1, &to,
1363 0, copy_complete, m);
1366 * Do we need to zero a tail region?
1368 if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) {
1369 atomic_inc(&m->prepare_actions);
1371 data_dest * pool->sectors_per_block + len,
1372 (data_dest + 1) * pool->sectors_per_block);
1376 complete_mapping_preparation(m); /* drop our ref */
1379 static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
1380 dm_block_t data_origin, dm_block_t data_dest,
1381 struct dm_bio_prison_cell *cell, struct bio *bio)
1383 schedule_copy(tc, virt_block, tc->pool_dev,
1384 data_origin, data_dest, cell, bio,
1385 tc->pool->sectors_per_block);
1388 static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
1389 dm_block_t data_block, struct dm_bio_prison_cell *cell,
1392 struct pool *pool = tc->pool;
1393 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1395 atomic_set(&m->prepare_actions, 1); /* no need to quiesce */
1397 m->virt_begin = virt_block;
1398 m->virt_end = virt_block + 1u;
1399 m->data_block = data_block;
1403 * If the whole block of data is being overwritten or we are not
1404 * zeroing pre-existing data, we can issue the bio immediately.
1405 * Otherwise we use kcopyd to zero the data first.
1407 if (pool->pf.zero_new_blocks) {
1408 if (io_overwrites_block(pool, bio))
1409 remap_and_issue_overwrite(tc, bio, data_block, m);
1411 ll_zero(tc, m, data_block * pool->sectors_per_block,
1412 (data_block + 1) * pool->sectors_per_block);
1414 process_prepared_mapping(m);
1417 static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
1418 dm_block_t data_dest,
1419 struct dm_bio_prison_cell *cell, struct bio *bio)
1421 struct pool *pool = tc->pool;
1422 sector_t virt_block_begin = virt_block * pool->sectors_per_block;
1423 sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block;
1425 if (virt_block_end <= tc->origin_size)
1426 schedule_copy(tc, virt_block, tc->origin_dev,
1427 virt_block, data_dest, cell, bio,
1428 pool->sectors_per_block);
1430 else if (virt_block_begin < tc->origin_size)
1431 schedule_copy(tc, virt_block, tc->origin_dev,
1432 virt_block, data_dest, cell, bio,
1433 tc->origin_size - virt_block_begin);
1436 schedule_zero(tc, virt_block, data_dest, cell, bio);
1439 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode);
1441 static void requeue_bios(struct pool *pool);
1443 static bool is_read_only_pool_mode(enum pool_mode mode)
1445 return (mode == PM_OUT_OF_METADATA_SPACE || mode == PM_READ_ONLY);
1448 static bool is_read_only(struct pool *pool)
1450 return is_read_only_pool_mode(get_pool_mode(pool));
1453 static void check_for_metadata_space(struct pool *pool)
1456 const char *ooms_reason = NULL;
1459 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free);
1461 ooms_reason = "Could not get free metadata blocks";
1463 ooms_reason = "No free metadata blocks";
1465 if (ooms_reason && !is_read_only(pool)) {
1466 DMERR("%s", ooms_reason);
1467 set_pool_mode(pool, PM_OUT_OF_METADATA_SPACE);
1471 static void check_for_data_space(struct pool *pool)
1476 if (get_pool_mode(pool) != PM_OUT_OF_DATA_SPACE)
1479 r = dm_pool_get_free_block_count(pool->pmd, &nr_free);
1484 set_pool_mode(pool, PM_WRITE);
1490 * A non-zero return indicates read_only or fail_io mode.
1491 * Many callers don't care about the return value.
1493 static int commit(struct pool *pool)
1497 if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE)
1500 r = dm_pool_commit_metadata(pool->pmd);
1502 metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
1504 check_for_metadata_space(pool);
1505 check_for_data_space(pool);
1511 static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks)
1513 unsigned long flags;
1515 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
1516 DMWARN("%s: reached low water mark for data device: sending event.",
1517 dm_device_name(pool->pool_md));
1518 spin_lock_irqsave(&pool->lock, flags);
1519 pool->low_water_triggered = true;
1520 spin_unlock_irqrestore(&pool->lock, flags);
1521 dm_table_event(pool->ti->table);
1525 static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
1528 dm_block_t free_blocks;
1529 struct pool *pool = tc->pool;
1531 if (WARN_ON(get_pool_mode(pool) != PM_WRITE))
1534 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1536 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1540 check_low_water_mark(pool, free_blocks);
1544 * Try to commit to see if that will free up some
1551 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1553 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1558 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1563 r = dm_pool_alloc_data_block(pool->pmd, result);
1566 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1568 metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
1572 r = dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks);
1574 metadata_operation_failed(pool, "dm_pool_get_free_metadata_block_count", r);
1579 /* Let's commit before we use up the metadata reserve. */
1589 * If we have run out of space, queue bios until the device is
1590 * resumed, presumably after having been reloaded with more space.
1592 static void retry_on_resume(struct bio *bio)
1594 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1595 struct thin_c *tc = h->tc;
1596 unsigned long flags;
1598 spin_lock_irqsave(&tc->lock, flags);
1599 bio_list_add(&tc->retry_on_resume_list, bio);
1600 spin_unlock_irqrestore(&tc->lock, flags);
1603 static blk_status_t should_error_unserviceable_bio(struct pool *pool)
1605 enum pool_mode m = get_pool_mode(pool);
1609 /* Shouldn't get here */
1610 DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
1611 return BLK_STS_IOERR;
1613 case PM_OUT_OF_DATA_SPACE:
1614 return pool->pf.error_if_no_space ? BLK_STS_NOSPC : 0;
1616 case PM_OUT_OF_METADATA_SPACE:
1619 return BLK_STS_IOERR;
1621 /* Shouldn't get here */
1622 DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1623 return BLK_STS_IOERR;
1627 static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
1629 blk_status_t error = should_error_unserviceable_bio(pool);
1632 bio->bi_status = error;
1635 retry_on_resume(bio);
1638 static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell)
1641 struct bio_list bios;
1644 error = should_error_unserviceable_bio(pool);
1646 cell_error_with_code(pool, cell, error);
1650 bio_list_init(&bios);
1651 cell_release(pool, cell, &bios);
1653 while ((bio = bio_list_pop(&bios)))
1654 retry_on_resume(bio);
1657 static void process_discard_cell_no_passdown(struct thin_c *tc,
1658 struct dm_bio_prison_cell *virt_cell)
1660 struct pool *pool = tc->pool;
1661 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1664 * We don't need to lock the data blocks, since there's no
1665 * passdown. We only lock data blocks for allocation and breaking sharing.
1668 m->virt_begin = virt_cell->key.block_begin;
1669 m->virt_end = virt_cell->key.block_end;
1670 m->cell = virt_cell;
1671 m->bio = virt_cell->holder;
1673 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1674 pool->process_prepared_discard(m);
1677 static void break_up_discard_bio(struct thin_c *tc, dm_block_t begin, dm_block_t end,
1680 struct pool *pool = tc->pool;
1684 struct dm_cell_key data_key;
1685 struct dm_bio_prison_cell *data_cell;
1686 struct dm_thin_new_mapping *m;
1687 dm_block_t virt_begin, virt_end, data_begin;
1689 while (begin != end) {
1690 r = ensure_next_mapping(pool);
1692 /* we did our best */
1695 r = dm_thin_find_mapped_range(tc->td, begin, end, &virt_begin, &virt_end,
1696 &data_begin, &maybe_shared);
1699 * Silently fail, letting any mappings we've
1704 build_key(tc->td, PHYSICAL, data_begin, data_begin + (virt_end - virt_begin), &data_key);
1705 if (bio_detain(tc->pool, &data_key, NULL, &data_cell)) {
1706 /* contention, we'll give up with this range */
1712 * IO may still be going to the destination block. We must
1713 * quiesce before we can do the removal.
1715 m = get_next_mapping(pool);
1717 m->maybe_shared = maybe_shared;
1718 m->virt_begin = virt_begin;
1719 m->virt_end = virt_end;
1720 m->data_block = data_begin;
1721 m->cell = data_cell;
1725 * The parent bio must not complete before sub discard bios are
1726 * chained to it (see end_discard's bio_chain)!
1728 * This per-mapping bi_remaining increment is paired with
1729 * the implicit decrement that occurs via bio_endio() in
1732 bio_inc_remaining(bio);
1733 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1734 pool->process_prepared_discard(m);
1740 static void process_discard_cell_passdown(struct thin_c *tc, struct dm_bio_prison_cell *virt_cell)
1742 struct bio *bio = virt_cell->holder;
1743 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1746 * The virt_cell will only get freed once the origin bio completes.
1747 * This means it will remain locked while all the individual
1748 * passdown bios are in flight.
1750 h->cell = virt_cell;
1751 break_up_discard_bio(tc, virt_cell->key.block_begin, virt_cell->key.block_end, bio);
1754 * We complete the bio now, knowing that the bi_remaining field
1755 * will prevent completion until the sub range discards have
1761 static void process_discard_bio(struct thin_c *tc, struct bio *bio)
1763 dm_block_t begin, end;
1764 struct dm_cell_key virt_key;
1765 struct dm_bio_prison_cell *virt_cell;
1767 get_bio_block_range(tc, bio, &begin, &end);
1770 * The discard covers less than a block.
1776 build_key(tc->td, VIRTUAL, begin, end, &virt_key);
1777 if (bio_detain(tc->pool, &virt_key, bio, &virt_cell))
1779 * Potential starvation issue: We're relying on the
1780 * fs/application being well behaved, and not trying to
1781 * send IO to a region at the same time as discarding it.
1782 * If they do this persistently then it's possible this
1783 * cell will never be granted.
1787 tc->pool->process_discard_cell(tc, virt_cell);
1790 static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
1791 struct dm_cell_key *key,
1792 struct dm_thin_lookup_result *lookup_result,
1793 struct dm_bio_prison_cell *cell)
1796 dm_block_t data_block;
1797 struct pool *pool = tc->pool;
1799 r = alloc_data_block(tc, &data_block);
1802 schedule_internal_copy(tc, block, lookup_result->block,
1803 data_block, cell, bio);
1807 retry_bios_on_resume(pool, cell);
1811 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1813 cell_error(pool, cell);
1818 static void __remap_and_issue_shared_cell(void *context,
1819 struct dm_bio_prison_cell *cell)
1821 struct remap_info *info = context;
1824 while ((bio = bio_list_pop(&cell->bios))) {
1825 if (bio_data_dir(bio) == WRITE || op_is_flush(bio->bi_opf) ||
1826 bio_op(bio) == REQ_OP_DISCARD)
1827 bio_list_add(&info->defer_bios, bio);
1829 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1831 h->shared_read_entry = dm_deferred_entry_inc(info->tc->pool->shared_read_ds);
1832 inc_all_io_entry(info->tc->pool, bio);
1833 bio_list_add(&info->issue_bios, bio);
1838 static void remap_and_issue_shared_cell(struct thin_c *tc,
1839 struct dm_bio_prison_cell *cell,
1843 struct remap_info info;
1846 bio_list_init(&info.defer_bios);
1847 bio_list_init(&info.issue_bios);
1849 cell_visit_release(tc->pool, __remap_and_issue_shared_cell,
1852 while ((bio = bio_list_pop(&info.defer_bios)))
1853 thin_defer_bio(tc, bio);
1855 while ((bio = bio_list_pop(&info.issue_bios)))
1856 remap_and_issue(tc, bio, block);
1859 static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1861 struct dm_thin_lookup_result *lookup_result,
1862 struct dm_bio_prison_cell *virt_cell)
1864 struct dm_bio_prison_cell *data_cell;
1865 struct pool *pool = tc->pool;
1866 struct dm_cell_key key;
1869 * If cell is already occupied, then sharing is already in the process
1870 * of being broken so we have nothing further to do here.
1872 build_data_key(tc->td, lookup_result->block, &key);
1873 if (bio_detain(pool, &key, bio, &data_cell)) {
1874 cell_defer_no_holder(tc, virt_cell);
1878 if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) {
1879 break_sharing(tc, bio, block, &key, lookup_result, data_cell);
1880 cell_defer_no_holder(tc, virt_cell);
1882 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1884 h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
1885 inc_all_io_entry(pool, bio);
1886 remap_and_issue(tc, bio, lookup_result->block);
1888 remap_and_issue_shared_cell(tc, data_cell, lookup_result->block);
1889 remap_and_issue_shared_cell(tc, virt_cell, lookup_result->block);
1893 static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
1894 struct dm_bio_prison_cell *cell)
1897 dm_block_t data_block;
1898 struct pool *pool = tc->pool;
1901 * Remap empty bios (flushes) immediately, without provisioning.
1903 if (!bio->bi_iter.bi_size) {
1904 inc_all_io_entry(pool, bio);
1905 cell_defer_no_holder(tc, cell);
1907 remap_and_issue(tc, bio, 0);
1912 * Fill read bios with zeroes and complete them immediately.
1914 if (bio_data_dir(bio) == READ) {
1916 cell_defer_no_holder(tc, cell);
1921 r = alloc_data_block(tc, &data_block);
1925 schedule_external_copy(tc, block, data_block, cell, bio);
1927 schedule_zero(tc, block, data_block, cell, bio);
1931 retry_bios_on_resume(pool, cell);
1935 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1937 cell_error(pool, cell);
1942 static void process_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
1945 struct pool *pool = tc->pool;
1946 struct bio *bio = cell->holder;
1947 dm_block_t block = get_bio_block(tc, bio);
1948 struct dm_thin_lookup_result lookup_result;
1950 if (tc->requeue_mode) {
1951 cell_requeue(pool, cell);
1955 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1958 if (lookup_result.shared)
1959 process_shared_bio(tc, bio, block, &lookup_result, cell);
1961 inc_all_io_entry(pool, bio);
1962 remap_and_issue(tc, bio, lookup_result.block);
1963 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
1968 if (bio_data_dir(bio) == READ && tc->origin_dev) {
1969 inc_all_io_entry(pool, bio);
1970 cell_defer_no_holder(tc, cell);
1972 if (bio_end_sector(bio) <= tc->origin_size)
1973 remap_to_origin_and_issue(tc, bio);
1975 else if (bio->bi_iter.bi_sector < tc->origin_size) {
1977 bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT;
1978 remap_to_origin_and_issue(tc, bio);
1985 provision_block(tc, bio, block, cell);
1989 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1991 cell_defer_no_holder(tc, cell);
1997 static void process_bio(struct thin_c *tc, struct bio *bio)
1999 struct pool *pool = tc->pool;
2000 dm_block_t block = get_bio_block(tc, bio);
2001 struct dm_bio_prison_cell *cell;
2002 struct dm_cell_key key;
2005 * If cell is already occupied, then the block is already
2006 * being provisioned so we have nothing further to do here.
2008 build_virtual_key(tc->td, block, &key);
2009 if (bio_detain(pool, &key, bio, &cell))
2012 process_cell(tc, cell);
2015 static void __process_bio_read_only(struct thin_c *tc, struct bio *bio,
2016 struct dm_bio_prison_cell *cell)
2019 int rw = bio_data_dir(bio);
2020 dm_block_t block = get_bio_block(tc, bio);
2021 struct dm_thin_lookup_result lookup_result;
2023 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
2026 if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) {
2027 handle_unserviceable_bio(tc->pool, bio);
2029 cell_defer_no_holder(tc, cell);
2031 inc_all_io_entry(tc->pool, bio);
2032 remap_and_issue(tc, bio, lookup_result.block);
2034 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
2040 cell_defer_no_holder(tc, cell);
2042 handle_unserviceable_bio(tc->pool, bio);
2046 if (tc->origin_dev) {
2047 inc_all_io_entry(tc->pool, bio);
2048 remap_to_origin_and_issue(tc, bio);
2057 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
2060 cell_defer_no_holder(tc, cell);
2066 static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
2068 __process_bio_read_only(tc, bio, NULL);
2071 static void process_cell_read_only(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2073 __process_bio_read_only(tc, cell->holder, cell);
2076 static void process_bio_success(struct thin_c *tc, struct bio *bio)
2081 static void process_bio_fail(struct thin_c *tc, struct bio *bio)
2086 static void process_cell_success(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2088 cell_success(tc->pool, cell);
2091 static void process_cell_fail(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2093 cell_error(tc->pool, cell);
2097 * FIXME: should we also commit due to size of transaction, measured in
2100 static int need_commit_due_to_time(struct pool *pool)
2102 return !time_in_range(jiffies, pool->last_commit_jiffies,
2103 pool->last_commit_jiffies + COMMIT_PERIOD);
2106 #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node)
2107 #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook))
2109 static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio)
2111 struct rb_node **rbp, *parent;
2112 struct dm_thin_endio_hook *pbd;
2113 sector_t bi_sector = bio->bi_iter.bi_sector;
2115 rbp = &tc->sort_bio_list.rb_node;
2119 pbd = thin_pbd(parent);
2121 if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector)
2122 rbp = &(*rbp)->rb_left;
2124 rbp = &(*rbp)->rb_right;
2127 pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2128 rb_link_node(&pbd->rb_node, parent, rbp);
2129 rb_insert_color(&pbd->rb_node, &tc->sort_bio_list);
2132 static void __extract_sorted_bios(struct thin_c *tc)
2134 struct rb_node *node;
2135 struct dm_thin_endio_hook *pbd;
2138 for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) {
2139 pbd = thin_pbd(node);
2140 bio = thin_bio(pbd);
2142 bio_list_add(&tc->deferred_bio_list, bio);
2143 rb_erase(&pbd->rb_node, &tc->sort_bio_list);
2146 WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list));
2149 static void __sort_thin_deferred_bios(struct thin_c *tc)
2152 struct bio_list bios;
2154 bio_list_init(&bios);
2155 bio_list_merge(&bios, &tc->deferred_bio_list);
2156 bio_list_init(&tc->deferred_bio_list);
2158 /* Sort deferred_bio_list using rb-tree */
2159 while ((bio = bio_list_pop(&bios)))
2160 __thin_bio_rb_add(tc, bio);
2163 * Transfer the sorted bios in sort_bio_list back to
2164 * deferred_bio_list to allow lockless submission of
2167 __extract_sorted_bios(tc);
2170 static void process_thin_deferred_bios(struct thin_c *tc)
2172 struct pool *pool = tc->pool;
2173 unsigned long flags;
2175 struct bio_list bios;
2176 struct blk_plug plug;
2179 if (tc->requeue_mode) {
2180 error_thin_bio_list(tc, &tc->deferred_bio_list,
2181 BLK_STS_DM_REQUEUE);
2185 bio_list_init(&bios);
2187 spin_lock_irqsave(&tc->lock, flags);
2189 if (bio_list_empty(&tc->deferred_bio_list)) {
2190 spin_unlock_irqrestore(&tc->lock, flags);
2194 __sort_thin_deferred_bios(tc);
2196 bio_list_merge(&bios, &tc->deferred_bio_list);
2197 bio_list_init(&tc->deferred_bio_list);
2199 spin_unlock_irqrestore(&tc->lock, flags);
2201 blk_start_plug(&plug);
2202 while ((bio = bio_list_pop(&bios))) {
2204 * If we've got no free new_mapping structs, and processing
2205 * this bio might require one, we pause until there are some
2206 * prepared mappings to process.
2208 if (ensure_next_mapping(pool)) {
2209 spin_lock_irqsave(&tc->lock, flags);
2210 bio_list_add(&tc->deferred_bio_list, bio);
2211 bio_list_merge(&tc->deferred_bio_list, &bios);
2212 spin_unlock_irqrestore(&tc->lock, flags);
2216 if (bio_op(bio) == REQ_OP_DISCARD)
2217 pool->process_discard(tc, bio);
2219 pool->process_bio(tc, bio);
2221 if ((count++ & 127) == 0) {
2222 throttle_work_update(&pool->throttle);
2223 dm_pool_issue_prefetches(pool->pmd);
2227 blk_finish_plug(&plug);
2230 static int cmp_cells(const void *lhs, const void *rhs)
2232 struct dm_bio_prison_cell *lhs_cell = *((struct dm_bio_prison_cell **) lhs);
2233 struct dm_bio_prison_cell *rhs_cell = *((struct dm_bio_prison_cell **) rhs);
2235 BUG_ON(!lhs_cell->holder);
2236 BUG_ON(!rhs_cell->holder);
2238 if (lhs_cell->holder->bi_iter.bi_sector < rhs_cell->holder->bi_iter.bi_sector)
2241 if (lhs_cell->holder->bi_iter.bi_sector > rhs_cell->holder->bi_iter.bi_sector)
2247 static unsigned sort_cells(struct pool *pool, struct list_head *cells)
2250 struct dm_bio_prison_cell *cell, *tmp;
2252 list_for_each_entry_safe(cell, tmp, cells, user_list) {
2253 if (count >= CELL_SORT_ARRAY_SIZE)
2256 pool->cell_sort_array[count++] = cell;
2257 list_del(&cell->user_list);
2260 sort(pool->cell_sort_array, count, sizeof(cell), cmp_cells, NULL);
2265 static void process_thin_deferred_cells(struct thin_c *tc)
2267 struct pool *pool = tc->pool;
2268 unsigned long flags;
2269 struct list_head cells;
2270 struct dm_bio_prison_cell *cell;
2271 unsigned i, j, count;
2273 INIT_LIST_HEAD(&cells);
2275 spin_lock_irqsave(&tc->lock, flags);
2276 list_splice_init(&tc->deferred_cells, &cells);
2277 spin_unlock_irqrestore(&tc->lock, flags);
2279 if (list_empty(&cells))
2283 count = sort_cells(tc->pool, &cells);
2285 for (i = 0; i < count; i++) {
2286 cell = pool->cell_sort_array[i];
2287 BUG_ON(!cell->holder);
2290 * If we've got no free new_mapping structs, and processing
2291 * this bio might require one, we pause until there are some
2292 * prepared mappings to process.
2294 if (ensure_next_mapping(pool)) {
2295 for (j = i; j < count; j++)
2296 list_add(&pool->cell_sort_array[j]->user_list, &cells);
2298 spin_lock_irqsave(&tc->lock, flags);
2299 list_splice(&cells, &tc->deferred_cells);
2300 spin_unlock_irqrestore(&tc->lock, flags);
2304 if (bio_op(cell->holder) == REQ_OP_DISCARD)
2305 pool->process_discard_cell(tc, cell);
2307 pool->process_cell(tc, cell);
2310 } while (!list_empty(&cells));
2313 static void thin_get(struct thin_c *tc);
2314 static void thin_put(struct thin_c *tc);
2317 * We can't hold rcu_read_lock() around code that can block. So we
2318 * find a thin with the rcu lock held; bump a refcount; then drop
2321 static struct thin_c *get_first_thin(struct pool *pool)
2323 struct thin_c *tc = NULL;
2326 if (!list_empty(&pool->active_thins)) {
2327 tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list);
2335 static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc)
2337 struct thin_c *old_tc = tc;
2340 list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) {
2352 static void process_deferred_bios(struct pool *pool)
2354 unsigned long flags;
2356 struct bio_list bios, bio_completions;
2359 tc = get_first_thin(pool);
2361 process_thin_deferred_cells(tc);
2362 process_thin_deferred_bios(tc);
2363 tc = get_next_thin(pool, tc);
2367 * If there are any deferred flush bios, we must commit the metadata
2368 * before issuing them or signaling their completion.
2370 bio_list_init(&bios);
2371 bio_list_init(&bio_completions);
2373 spin_lock_irqsave(&pool->lock, flags);
2374 bio_list_merge(&bios, &pool->deferred_flush_bios);
2375 bio_list_init(&pool->deferred_flush_bios);
2377 bio_list_merge(&bio_completions, &pool->deferred_flush_completions);
2378 bio_list_init(&pool->deferred_flush_completions);
2379 spin_unlock_irqrestore(&pool->lock, flags);
2381 if (bio_list_empty(&bios) && bio_list_empty(&bio_completions) &&
2382 !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool)))
2386 bio_list_merge(&bios, &bio_completions);
2388 while ((bio = bio_list_pop(&bios)))
2392 pool->last_commit_jiffies = jiffies;
2394 while ((bio = bio_list_pop(&bio_completions)))
2397 while ((bio = bio_list_pop(&bios)))
2398 generic_make_request(bio);
2401 static void do_worker(struct work_struct *ws)
2403 struct pool *pool = container_of(ws, struct pool, worker);
2405 throttle_work_start(&pool->throttle);
2406 dm_pool_issue_prefetches(pool->pmd);
2407 throttle_work_update(&pool->throttle);
2408 process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
2409 throttle_work_update(&pool->throttle);
2410 process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
2411 throttle_work_update(&pool->throttle);
2412 process_prepared(pool, &pool->prepared_discards_pt2, &pool->process_prepared_discard_pt2);
2413 throttle_work_update(&pool->throttle);
2414 process_deferred_bios(pool);
2415 throttle_work_complete(&pool->throttle);
2419 * We want to commit periodically so that not too much
2420 * unwritten data builds up.
2422 static void do_waker(struct work_struct *ws)
2424 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
2426 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
2430 * We're holding onto IO to allow userland time to react. After the
2431 * timeout either the pool will have been resized (and thus back in
2432 * PM_WRITE mode), or we degrade to PM_OUT_OF_DATA_SPACE w/ error_if_no_space.
2434 static void do_no_space_timeout(struct work_struct *ws)
2436 struct pool *pool = container_of(to_delayed_work(ws), struct pool,
2439 if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) {
2440 pool->pf.error_if_no_space = true;
2441 notify_of_pool_mode_change(pool);
2442 error_retry_list_with_code(pool, BLK_STS_NOSPC);
2446 /*----------------------------------------------------------------*/
2449 struct work_struct worker;
2450 struct completion complete;
2453 static struct pool_work *to_pool_work(struct work_struct *ws)
2455 return container_of(ws, struct pool_work, worker);
2458 static void pool_work_complete(struct pool_work *pw)
2460 complete(&pw->complete);
2463 static void pool_work_wait(struct pool_work *pw, struct pool *pool,
2464 void (*fn)(struct work_struct *))
2466 INIT_WORK_ONSTACK(&pw->worker, fn);
2467 init_completion(&pw->complete);
2468 queue_work(pool->wq, &pw->worker);
2469 wait_for_completion(&pw->complete);
2472 /*----------------------------------------------------------------*/
2474 struct noflush_work {
2475 struct pool_work pw;
2479 static struct noflush_work *to_noflush(struct work_struct *ws)
2481 return container_of(to_pool_work(ws), struct noflush_work, pw);
2484 static void do_noflush_start(struct work_struct *ws)
2486 struct noflush_work *w = to_noflush(ws);
2487 w->tc->requeue_mode = true;
2489 pool_work_complete(&w->pw);
2492 static void do_noflush_stop(struct work_struct *ws)
2494 struct noflush_work *w = to_noflush(ws);
2495 w->tc->requeue_mode = false;
2496 pool_work_complete(&w->pw);
2499 static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *))
2501 struct noflush_work w;
2504 pool_work_wait(&w.pw, tc->pool, fn);
2507 /*----------------------------------------------------------------*/
2509 static bool passdown_enabled(struct pool_c *pt)
2511 return pt->adjusted_pf.discard_passdown;
2514 static void set_discard_callbacks(struct pool *pool)
2516 struct pool_c *pt = pool->ti->private;
2518 if (passdown_enabled(pt)) {
2519 pool->process_discard_cell = process_discard_cell_passdown;
2520 pool->process_prepared_discard = process_prepared_discard_passdown_pt1;
2521 pool->process_prepared_discard_pt2 = process_prepared_discard_passdown_pt2;
2523 pool->process_discard_cell = process_discard_cell_no_passdown;
2524 pool->process_prepared_discard = process_prepared_discard_no_passdown;
2528 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode)
2530 struct pool_c *pt = pool->ti->private;
2531 bool needs_check = dm_pool_metadata_needs_check(pool->pmd);
2532 enum pool_mode old_mode = get_pool_mode(pool);
2533 unsigned long no_space_timeout = READ_ONCE(no_space_timeout_secs) * HZ;
2536 * Never allow the pool to transition to PM_WRITE mode if user
2537 * intervention is required to verify metadata and data consistency.
2539 if (new_mode == PM_WRITE && needs_check) {
2540 DMERR("%s: unable to switch pool to write mode until repaired.",
2541 dm_device_name(pool->pool_md));
2542 if (old_mode != new_mode)
2543 new_mode = old_mode;
2545 new_mode = PM_READ_ONLY;
2548 * If we were in PM_FAIL mode, rollback of metadata failed. We're
2549 * not going to recover without a thin_repair. So we never let the
2550 * pool move out of the old mode.
2552 if (old_mode == PM_FAIL)
2553 new_mode = old_mode;
2557 dm_pool_metadata_read_only(pool->pmd);
2558 pool->process_bio = process_bio_fail;
2559 pool->process_discard = process_bio_fail;
2560 pool->process_cell = process_cell_fail;
2561 pool->process_discard_cell = process_cell_fail;
2562 pool->process_prepared_mapping = process_prepared_mapping_fail;
2563 pool->process_prepared_discard = process_prepared_discard_fail;
2565 error_retry_list(pool);
2568 case PM_OUT_OF_METADATA_SPACE:
2570 dm_pool_metadata_read_only(pool->pmd);
2571 pool->process_bio = process_bio_read_only;
2572 pool->process_discard = process_bio_success;
2573 pool->process_cell = process_cell_read_only;
2574 pool->process_discard_cell = process_cell_success;
2575 pool->process_prepared_mapping = process_prepared_mapping_fail;
2576 pool->process_prepared_discard = process_prepared_discard_success;
2578 error_retry_list(pool);
2581 case PM_OUT_OF_DATA_SPACE:
2583 * Ideally we'd never hit this state; the low water mark
2584 * would trigger userland to extend the pool before we
2585 * completely run out of data space. However, many small
2586 * IOs to unprovisioned space can consume data space at an
2587 * alarming rate. Adjust your low water mark if you're
2588 * frequently seeing this mode.
2590 pool->out_of_data_space = true;
2591 pool->process_bio = process_bio_read_only;
2592 pool->process_discard = process_discard_bio;
2593 pool->process_cell = process_cell_read_only;
2594 pool->process_prepared_mapping = process_prepared_mapping;
2595 set_discard_callbacks(pool);
2597 if (!pool->pf.error_if_no_space && no_space_timeout)
2598 queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout);
2602 if (old_mode == PM_OUT_OF_DATA_SPACE)
2603 cancel_delayed_work_sync(&pool->no_space_timeout);
2604 pool->out_of_data_space = false;
2605 pool->pf.error_if_no_space = pt->requested_pf.error_if_no_space;
2606 dm_pool_metadata_read_write(pool->pmd);
2607 pool->process_bio = process_bio;
2608 pool->process_discard = process_discard_bio;
2609 pool->process_cell = process_cell;
2610 pool->process_prepared_mapping = process_prepared_mapping;
2611 set_discard_callbacks(pool);
2615 pool->pf.mode = new_mode;
2617 * The pool mode may have changed, sync it so bind_control_target()
2618 * doesn't cause an unexpected mode transition on resume.
2620 pt->adjusted_pf.mode = new_mode;
2622 if (old_mode != new_mode)
2623 notify_of_pool_mode_change(pool);
2626 static void abort_transaction(struct pool *pool)
2628 const char *dev_name = dm_device_name(pool->pool_md);
2630 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name);
2631 if (dm_pool_abort_metadata(pool->pmd)) {
2632 DMERR("%s: failed to abort metadata transaction", dev_name);
2633 set_pool_mode(pool, PM_FAIL);
2636 if (dm_pool_metadata_set_needs_check(pool->pmd)) {
2637 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name);
2638 set_pool_mode(pool, PM_FAIL);
2642 static void metadata_operation_failed(struct pool *pool, const char *op, int r)
2644 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
2645 dm_device_name(pool->pool_md), op, r);
2647 abort_transaction(pool);
2648 set_pool_mode(pool, PM_READ_ONLY);
2651 /*----------------------------------------------------------------*/
2654 * Mapping functions.
2658 * Called only while mapping a thin bio to hand it over to the workqueue.
2660 static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
2662 unsigned long flags;
2663 struct pool *pool = tc->pool;
2665 spin_lock_irqsave(&tc->lock, flags);
2666 bio_list_add(&tc->deferred_bio_list, bio);
2667 spin_unlock_irqrestore(&tc->lock, flags);
2672 static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio)
2674 struct pool *pool = tc->pool;
2676 throttle_lock(&pool->throttle);
2677 thin_defer_bio(tc, bio);
2678 throttle_unlock(&pool->throttle);
2681 static void thin_defer_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2683 unsigned long flags;
2684 struct pool *pool = tc->pool;
2686 throttle_lock(&pool->throttle);
2687 spin_lock_irqsave(&tc->lock, flags);
2688 list_add_tail(&cell->user_list, &tc->deferred_cells);
2689 spin_unlock_irqrestore(&tc->lock, flags);
2690 throttle_unlock(&pool->throttle);
2695 static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
2697 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2700 h->shared_read_entry = NULL;
2701 h->all_io_entry = NULL;
2702 h->overwrite_mapping = NULL;
2707 * Non-blocking function called from the thin target's map function.
2709 static int thin_bio_map(struct dm_target *ti, struct bio *bio)
2712 struct thin_c *tc = ti->private;
2713 dm_block_t block = get_bio_block(tc, bio);
2714 struct dm_thin_device *td = tc->td;
2715 struct dm_thin_lookup_result result;
2716 struct dm_bio_prison_cell *virt_cell, *data_cell;
2717 struct dm_cell_key key;
2719 thin_hook_bio(tc, bio);
2721 if (tc->requeue_mode) {
2722 bio->bi_status = BLK_STS_DM_REQUEUE;
2724 return DM_MAPIO_SUBMITTED;
2727 if (get_pool_mode(tc->pool) == PM_FAIL) {
2729 return DM_MAPIO_SUBMITTED;
2732 if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD) {
2733 thin_defer_bio_with_throttle(tc, bio);
2734 return DM_MAPIO_SUBMITTED;
2738 * We must hold the virtual cell before doing the lookup, otherwise
2739 * there's a race with discard.
2741 build_virtual_key(tc->td, block, &key);
2742 if (bio_detain(tc->pool, &key, bio, &virt_cell))
2743 return DM_MAPIO_SUBMITTED;
2745 r = dm_thin_find_block(td, block, 0, &result);
2748 * Note that we defer readahead too.
2752 if (unlikely(result.shared)) {
2754 * We have a race condition here between the
2755 * result.shared value returned by the lookup and
2756 * snapshot creation, which may cause new
2759 * To avoid this always quiesce the origin before
2760 * taking the snap. You want to do this anyway to
2761 * ensure a consistent application view
2764 * More distant ancestors are irrelevant. The
2765 * shared flag will be set in their case.
2767 thin_defer_cell(tc, virt_cell);
2768 return DM_MAPIO_SUBMITTED;
2771 build_data_key(tc->td, result.block, &key);
2772 if (bio_detain(tc->pool, &key, bio, &data_cell)) {
2773 cell_defer_no_holder(tc, virt_cell);
2774 return DM_MAPIO_SUBMITTED;
2777 inc_all_io_entry(tc->pool, bio);
2778 cell_defer_no_holder(tc, data_cell);
2779 cell_defer_no_holder(tc, virt_cell);
2781 remap(tc, bio, result.block);
2782 return DM_MAPIO_REMAPPED;
2786 thin_defer_cell(tc, virt_cell);
2787 return DM_MAPIO_SUBMITTED;
2791 * Must always call bio_io_error on failure.
2792 * dm_thin_find_block can fail with -EINVAL if the
2793 * pool is switched to fail-io mode.
2796 cell_defer_no_holder(tc, virt_cell);
2797 return DM_MAPIO_SUBMITTED;
2801 static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
2803 struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
2804 struct request_queue *q;
2806 if (get_pool_mode(pt->pool) == PM_OUT_OF_DATA_SPACE)
2809 q = bdev_get_queue(pt->data_dev->bdev);
2810 return bdi_congested(q->backing_dev_info, bdi_bits);
2813 static void requeue_bios(struct pool *pool)
2815 unsigned long flags;
2819 list_for_each_entry_rcu(tc, &pool->active_thins, list) {
2820 spin_lock_irqsave(&tc->lock, flags);
2821 bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list);
2822 bio_list_init(&tc->retry_on_resume_list);
2823 spin_unlock_irqrestore(&tc->lock, flags);
2828 /*----------------------------------------------------------------
2829 * Binding of control targets to a pool object
2830 *--------------------------------------------------------------*/
2831 static bool data_dev_supports_discard(struct pool_c *pt)
2833 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
2835 return q && blk_queue_discard(q);
2838 static bool is_factor(sector_t block_size, uint32_t n)
2840 return !sector_div(block_size, n);
2844 * If discard_passdown was enabled verify that the data device
2845 * supports discards. Disable discard_passdown if not.
2847 static void disable_passdown_if_not_supported(struct pool_c *pt)
2849 struct pool *pool = pt->pool;
2850 struct block_device *data_bdev = pt->data_dev->bdev;
2851 struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
2852 const char *reason = NULL;
2853 char buf[BDEVNAME_SIZE];
2855 if (!pt->adjusted_pf.discard_passdown)
2858 if (!data_dev_supports_discard(pt))
2859 reason = "discard unsupported";
2861 else if (data_limits->max_discard_sectors < pool->sectors_per_block)
2862 reason = "max discard sectors smaller than a block";
2865 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason);
2866 pt->adjusted_pf.discard_passdown = false;
2870 static int bind_control_target(struct pool *pool, struct dm_target *ti)
2872 struct pool_c *pt = ti->private;
2875 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
2877 enum pool_mode old_mode = get_pool_mode(pool);
2878 enum pool_mode new_mode = pt->adjusted_pf.mode;
2881 * Don't change the pool's mode until set_pool_mode() below.
2882 * Otherwise the pool's process_* function pointers may
2883 * not match the desired pool mode.
2885 pt->adjusted_pf.mode = old_mode;
2888 pool->pf = pt->adjusted_pf;
2889 pool->low_water_blocks = pt->low_water_blocks;
2891 set_pool_mode(pool, new_mode);
2896 static void unbind_control_target(struct pool *pool, struct dm_target *ti)
2902 /*----------------------------------------------------------------
2904 *--------------------------------------------------------------*/
2905 /* Initialize pool features. */
2906 static void pool_features_init(struct pool_features *pf)
2908 pf->mode = PM_WRITE;
2909 pf->zero_new_blocks = true;
2910 pf->discard_enabled = true;
2911 pf->discard_passdown = true;
2912 pf->error_if_no_space = false;
2915 static void __pool_destroy(struct pool *pool)
2917 __pool_table_remove(pool);
2919 vfree(pool->cell_sort_array);
2920 if (dm_pool_metadata_close(pool->pmd) < 0)
2921 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
2923 dm_bio_prison_destroy(pool->prison);
2924 dm_kcopyd_client_destroy(pool->copier);
2926 cancel_delayed_work_sync(&pool->waker);
2927 cancel_delayed_work_sync(&pool->no_space_timeout);
2929 destroy_workqueue(pool->wq);
2931 if (pool->next_mapping)
2932 mempool_free(pool->next_mapping, &pool->mapping_pool);
2933 mempool_exit(&pool->mapping_pool);
2934 dm_deferred_set_destroy(pool->shared_read_ds);
2935 dm_deferred_set_destroy(pool->all_io_ds);
2939 static struct kmem_cache *_new_mapping_cache;
2941 static struct pool *pool_create(struct mapped_device *pool_md,
2942 struct block_device *metadata_dev,
2943 unsigned long block_size,
2944 int read_only, char **error)
2949 struct dm_pool_metadata *pmd;
2950 bool format_device = read_only ? false : true;
2952 pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
2954 *error = "Error creating metadata object";
2955 return (struct pool *)pmd;
2958 pool = kzalloc(sizeof(*pool), GFP_KERNEL);
2960 *error = "Error allocating memory for pool";
2961 err_p = ERR_PTR(-ENOMEM);
2966 pool->sectors_per_block = block_size;
2967 if (block_size & (block_size - 1))
2968 pool->sectors_per_block_shift = -1;
2970 pool->sectors_per_block_shift = __ffs(block_size);
2971 pool->low_water_blocks = 0;
2972 pool_features_init(&pool->pf);
2973 pool->prison = dm_bio_prison_create();
2974 if (!pool->prison) {
2975 *error = "Error creating pool's bio prison";
2976 err_p = ERR_PTR(-ENOMEM);
2980 pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2981 if (IS_ERR(pool->copier)) {
2982 r = PTR_ERR(pool->copier);
2983 *error = "Error creating pool's kcopyd client";
2985 goto bad_kcopyd_client;
2989 * Create singlethreaded workqueue that will service all devices
2990 * that use this metadata.
2992 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
2994 *error = "Error creating pool's workqueue";
2995 err_p = ERR_PTR(-ENOMEM);
2999 throttle_init(&pool->throttle);
3000 INIT_WORK(&pool->worker, do_worker);
3001 INIT_DELAYED_WORK(&pool->waker, do_waker);
3002 INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout);
3003 spin_lock_init(&pool->lock);
3004 bio_list_init(&pool->deferred_flush_bios);
3005 bio_list_init(&pool->deferred_flush_completions);
3006 INIT_LIST_HEAD(&pool->prepared_mappings);
3007 INIT_LIST_HEAD(&pool->prepared_discards);
3008 INIT_LIST_HEAD(&pool->prepared_discards_pt2);
3009 INIT_LIST_HEAD(&pool->active_thins);
3010 pool->low_water_triggered = false;
3011 pool->suspended = true;
3012 pool->out_of_data_space = false;
3014 pool->shared_read_ds = dm_deferred_set_create();
3015 if (!pool->shared_read_ds) {
3016 *error = "Error creating pool's shared read deferred set";
3017 err_p = ERR_PTR(-ENOMEM);
3018 goto bad_shared_read_ds;
3021 pool->all_io_ds = dm_deferred_set_create();
3022 if (!pool->all_io_ds) {
3023 *error = "Error creating pool's all io deferred set";
3024 err_p = ERR_PTR(-ENOMEM);
3028 pool->next_mapping = NULL;
3029 r = mempool_init_slab_pool(&pool->mapping_pool, MAPPING_POOL_SIZE,
3030 _new_mapping_cache);
3032 *error = "Error creating pool's mapping mempool";
3034 goto bad_mapping_pool;
3037 pool->cell_sort_array =
3038 vmalloc(array_size(CELL_SORT_ARRAY_SIZE,
3039 sizeof(*pool->cell_sort_array)));
3040 if (!pool->cell_sort_array) {
3041 *error = "Error allocating cell sort array";
3042 err_p = ERR_PTR(-ENOMEM);
3043 goto bad_sort_array;
3046 pool->ref_count = 1;
3047 pool->last_commit_jiffies = jiffies;
3048 pool->pool_md = pool_md;
3049 pool->md_dev = metadata_dev;
3050 __pool_table_insert(pool);
3055 mempool_exit(&pool->mapping_pool);
3057 dm_deferred_set_destroy(pool->all_io_ds);
3059 dm_deferred_set_destroy(pool->shared_read_ds);
3061 destroy_workqueue(pool->wq);
3063 dm_kcopyd_client_destroy(pool->copier);
3065 dm_bio_prison_destroy(pool->prison);
3069 if (dm_pool_metadata_close(pmd))
3070 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
3075 static void __pool_inc(struct pool *pool)
3077 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
3081 static void __pool_dec(struct pool *pool)
3083 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
3084 BUG_ON(!pool->ref_count);
3085 if (!--pool->ref_count)
3086 __pool_destroy(pool);
3089 static struct pool *__pool_find(struct mapped_device *pool_md,
3090 struct block_device *metadata_dev,
3091 unsigned long block_size, int read_only,
3092 char **error, int *created)
3094 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
3097 if (pool->pool_md != pool_md) {
3098 *error = "metadata device already in use by a pool";
3099 return ERR_PTR(-EBUSY);
3104 pool = __pool_table_lookup(pool_md);
3106 if (pool->md_dev != metadata_dev) {
3107 *error = "different pool cannot replace a pool";
3108 return ERR_PTR(-EINVAL);
3113 pool = pool_create(pool_md, metadata_dev, block_size, read_only, error);
3121 /*----------------------------------------------------------------
3122 * Pool target methods
3123 *--------------------------------------------------------------*/
3124 static void pool_dtr(struct dm_target *ti)
3126 struct pool_c *pt = ti->private;
3128 mutex_lock(&dm_thin_pool_table.mutex);
3130 unbind_control_target(pt->pool, ti);
3131 __pool_dec(pt->pool);
3132 dm_put_device(ti, pt->metadata_dev);
3133 dm_put_device(ti, pt->data_dev);
3136 mutex_unlock(&dm_thin_pool_table.mutex);
3139 static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
3140 struct dm_target *ti)
3144 const char *arg_name;
3146 static const struct dm_arg _args[] = {
3147 {0, 4, "Invalid number of pool feature arguments"},
3151 * No feature arguments supplied.
3156 r = dm_read_arg_group(_args, as, &argc, &ti->error);
3160 while (argc && !r) {
3161 arg_name = dm_shift_arg(as);
3164 if (!strcasecmp(arg_name, "skip_block_zeroing"))
3165 pf->zero_new_blocks = false;
3167 else if (!strcasecmp(arg_name, "ignore_discard"))
3168 pf->discard_enabled = false;
3170 else if (!strcasecmp(arg_name, "no_discard_passdown"))
3171 pf->discard_passdown = false;
3173 else if (!strcasecmp(arg_name, "read_only"))
3174 pf->mode = PM_READ_ONLY;
3176 else if (!strcasecmp(arg_name, "error_if_no_space"))
3177 pf->error_if_no_space = true;
3180 ti->error = "Unrecognised pool feature requested";
3189 static void metadata_low_callback(void *context)
3191 struct pool *pool = context;
3193 DMWARN("%s: reached low water mark for metadata device: sending event.",
3194 dm_device_name(pool->pool_md));
3196 dm_table_event(pool->ti->table);
3199 static sector_t get_dev_size(struct block_device *bdev)
3201 return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
3204 static void warn_if_metadata_device_too_big(struct block_device *bdev)
3206 sector_t metadata_dev_size = get_dev_size(bdev);
3207 char buffer[BDEVNAME_SIZE];
3209 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING)
3210 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
3211 bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS);
3214 static sector_t get_metadata_dev_size(struct block_device *bdev)
3216 sector_t metadata_dev_size = get_dev_size(bdev);
3218 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS)
3219 metadata_dev_size = THIN_METADATA_MAX_SECTORS;
3221 return metadata_dev_size;
3224 static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
3226 sector_t metadata_dev_size = get_metadata_dev_size(bdev);
3228 sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE);
3230 return metadata_dev_size;
3234 * When a metadata threshold is crossed a dm event is triggered, and
3235 * userland should respond by growing the metadata device. We could let
3236 * userland set the threshold, like we do with the data threshold, but I'm
3237 * not sure they know enough to do this well.
3239 static dm_block_t calc_metadata_threshold(struct pool_c *pt)
3242 * 4M is ample for all ops with the possible exception of thin
3243 * device deletion which is harmless if it fails (just retry the
3244 * delete after you've grown the device).
3246 dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
3247 return min((dm_block_t)1024ULL /* 4M */, quarter);
3251 * thin-pool <metadata dev> <data dev>
3252 * <data block size (sectors)>
3253 * <low water mark (blocks)>
3254 * [<#feature args> [<arg>]*]
3256 * Optional feature arguments are:
3257 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
3258 * ignore_discard: disable discard
3259 * no_discard_passdown: don't pass discards down to the data device
3260 * read_only: Don't allow any changes to be made to the pool metadata.
3261 * error_if_no_space: error IOs, instead of queueing, if no space.
3263 static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
3265 int r, pool_created = 0;
3268 struct pool_features pf;
3269 struct dm_arg_set as;
3270 struct dm_dev *data_dev;
3271 unsigned long block_size;
3272 dm_block_t low_water_blocks;
3273 struct dm_dev *metadata_dev;
3274 fmode_t metadata_mode;
3277 * FIXME Remove validation from scope of lock.
3279 mutex_lock(&dm_thin_pool_table.mutex);
3282 ti->error = "Invalid argument count";
3290 /* make sure metadata and data are different devices */
3291 if (!strcmp(argv[0], argv[1])) {
3292 ti->error = "Error setting metadata or data device";
3298 * Set default pool features.
3300 pool_features_init(&pf);
3302 dm_consume_args(&as, 4);
3303 r = parse_pool_features(&as, &pf, ti);
3307 metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
3308 r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
3310 ti->error = "Error opening metadata block device";
3313 warn_if_metadata_device_too_big(metadata_dev->bdev);
3315 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
3317 ti->error = "Error getting data device";
3321 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
3322 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
3323 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
3324 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
3325 ti->error = "Invalid block size";
3330 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
3331 ti->error = "Invalid low water mark";
3336 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
3342 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
3343 block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
3350 * 'pool_created' reflects whether this is the first table load.
3351 * Top level discard support is not allowed to be changed after
3352 * initial load. This would require a pool reload to trigger thin
3355 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
3356 ti->error = "Discard support cannot be disabled once enabled";
3358 goto out_flags_changed;
3363 pt->metadata_dev = metadata_dev;
3364 pt->data_dev = data_dev;
3365 pt->low_water_blocks = low_water_blocks;
3366 pt->adjusted_pf = pt->requested_pf = pf;
3367 ti->num_flush_bios = 1;
3368 ti->limit_swap_bios = true;
3371 * Only need to enable discards if the pool should pass
3372 * them down to the data device. The thin device's discard
3373 * processing will cause mappings to be removed from the btree.
3375 if (pf.discard_enabled && pf.discard_passdown) {
3376 ti->num_discard_bios = 1;
3379 * Setting 'discards_supported' circumvents the normal
3380 * stacking of discard limits (this keeps the pool and
3381 * thin devices' discard limits consistent).
3383 ti->discards_supported = true;
3387 r = dm_pool_register_metadata_threshold(pt->pool->pmd,
3388 calc_metadata_threshold(pt),
3389 metadata_low_callback,
3392 goto out_flags_changed;
3394 pt->callbacks.congested_fn = pool_is_congested;
3395 dm_table_add_target_callbacks(ti->table, &pt->callbacks);
3397 mutex_unlock(&dm_thin_pool_table.mutex);
3406 dm_put_device(ti, data_dev);
3408 dm_put_device(ti, metadata_dev);
3410 mutex_unlock(&dm_thin_pool_table.mutex);
3415 static int pool_map(struct dm_target *ti, struct bio *bio)
3418 struct pool_c *pt = ti->private;
3419 struct pool *pool = pt->pool;
3420 unsigned long flags;
3423 * As this is a singleton target, ti->begin is always zero.
3425 spin_lock_irqsave(&pool->lock, flags);
3426 bio_set_dev(bio, pt->data_dev->bdev);
3427 r = DM_MAPIO_REMAPPED;
3428 spin_unlock_irqrestore(&pool->lock, flags);
3433 static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
3436 struct pool_c *pt = ti->private;
3437 struct pool *pool = pt->pool;
3438 sector_t data_size = ti->len;
3439 dm_block_t sb_data_size;
3441 *need_commit = false;
3443 (void) sector_div(data_size, pool->sectors_per_block);
3445 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
3447 DMERR("%s: failed to retrieve data device size",
3448 dm_device_name(pool->pool_md));
3452 if (data_size < sb_data_size) {
3453 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
3454 dm_device_name(pool->pool_md),
3455 (unsigned long long)data_size, sb_data_size);
3458 } else if (data_size > sb_data_size) {
3459 if (dm_pool_metadata_needs_check(pool->pmd)) {
3460 DMERR("%s: unable to grow the data device until repaired.",
3461 dm_device_name(pool->pool_md));
3466 DMINFO("%s: growing the data device from %llu to %llu blocks",
3467 dm_device_name(pool->pool_md),
3468 sb_data_size, (unsigned long long)data_size);
3469 r = dm_pool_resize_data_dev(pool->pmd, data_size);
3471 metadata_operation_failed(pool, "dm_pool_resize_data_dev", r);
3475 *need_commit = true;
3481 static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
3484 struct pool_c *pt = ti->private;
3485 struct pool *pool = pt->pool;
3486 dm_block_t metadata_dev_size, sb_metadata_dev_size;
3488 *need_commit = false;
3490 metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
3492 r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
3494 DMERR("%s: failed to retrieve metadata device size",
3495 dm_device_name(pool->pool_md));
3499 if (metadata_dev_size < sb_metadata_dev_size) {
3500 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
3501 dm_device_name(pool->pool_md),
3502 metadata_dev_size, sb_metadata_dev_size);
3505 } else if (metadata_dev_size > sb_metadata_dev_size) {
3506 if (dm_pool_metadata_needs_check(pool->pmd)) {
3507 DMERR("%s: unable to grow the metadata device until repaired.",
3508 dm_device_name(pool->pool_md));
3512 warn_if_metadata_device_too_big(pool->md_dev);
3513 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
3514 dm_device_name(pool->pool_md),
3515 sb_metadata_dev_size, metadata_dev_size);
3517 if (get_pool_mode(pool) == PM_OUT_OF_METADATA_SPACE)
3518 set_pool_mode(pool, PM_WRITE);
3520 r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
3522 metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r);
3526 *need_commit = true;
3533 * Retrieves the number of blocks of the data device from
3534 * the superblock and compares it to the actual device size,
3535 * thus resizing the data device in case it has grown.
3537 * This both copes with opening preallocated data devices in the ctr
3538 * being followed by a resume
3540 * calling the resume method individually after userspace has
3541 * grown the data device in reaction to a table event.
3543 static int pool_preresume(struct dm_target *ti)
3546 bool need_commit1, need_commit2;
3547 struct pool_c *pt = ti->private;
3548 struct pool *pool = pt->pool;
3551 * Take control of the pool object.
3553 r = bind_control_target(pool, ti);
3557 r = maybe_resize_data_dev(ti, &need_commit1);
3561 r = maybe_resize_metadata_dev(ti, &need_commit2);
3565 if (need_commit1 || need_commit2)
3566 (void) commit(pool);
3569 * When a thin-pool is PM_FAIL, it cannot be rebuilt if
3570 * bio is in deferred list. Therefore need to return 0
3571 * to allow pool_resume() to flush IO.
3573 if (r && get_pool_mode(pool) == PM_FAIL)
3579 static void pool_suspend_active_thins(struct pool *pool)
3583 /* Suspend all active thin devices */
3584 tc = get_first_thin(pool);
3586 dm_internal_suspend_noflush(tc->thin_md);
3587 tc = get_next_thin(pool, tc);
3591 static void pool_resume_active_thins(struct pool *pool)
3595 /* Resume all active thin devices */
3596 tc = get_first_thin(pool);
3598 dm_internal_resume(tc->thin_md);
3599 tc = get_next_thin(pool, tc);
3603 static void pool_resume(struct dm_target *ti)
3605 struct pool_c *pt = ti->private;
3606 struct pool *pool = pt->pool;
3607 unsigned long flags;
3610 * Must requeue active_thins' bios and then resume
3611 * active_thins _before_ clearing 'suspend' flag.
3614 pool_resume_active_thins(pool);
3616 spin_lock_irqsave(&pool->lock, flags);
3617 pool->low_water_triggered = false;
3618 pool->suspended = false;
3619 spin_unlock_irqrestore(&pool->lock, flags);
3621 do_waker(&pool->waker.work);
3624 static void pool_presuspend(struct dm_target *ti)
3626 struct pool_c *pt = ti->private;
3627 struct pool *pool = pt->pool;
3628 unsigned long flags;
3630 spin_lock_irqsave(&pool->lock, flags);
3631 pool->suspended = true;
3632 spin_unlock_irqrestore(&pool->lock, flags);
3634 pool_suspend_active_thins(pool);
3637 static void pool_presuspend_undo(struct dm_target *ti)
3639 struct pool_c *pt = ti->private;
3640 struct pool *pool = pt->pool;
3641 unsigned long flags;
3643 pool_resume_active_thins(pool);
3645 spin_lock_irqsave(&pool->lock, flags);
3646 pool->suspended = false;
3647 spin_unlock_irqrestore(&pool->lock, flags);
3650 static void pool_postsuspend(struct dm_target *ti)
3652 struct pool_c *pt = ti->private;
3653 struct pool *pool = pt->pool;
3655 cancel_delayed_work_sync(&pool->waker);
3656 cancel_delayed_work_sync(&pool->no_space_timeout);
3657 flush_workqueue(pool->wq);
3658 (void) commit(pool);
3661 static int check_arg_count(unsigned argc, unsigned args_required)
3663 if (argc != args_required) {
3664 DMWARN("Message received with %u arguments instead of %u.",
3665 argc, args_required);
3672 static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
3674 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
3675 *dev_id <= MAX_DEV_ID)
3679 DMWARN("Message received with invalid device id: %s", arg);
3684 static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
3689 r = check_arg_count(argc, 2);
3693 r = read_dev_id(argv[1], &dev_id, 1);
3697 r = dm_pool_create_thin(pool->pmd, dev_id);
3699 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
3707 static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3710 dm_thin_id origin_dev_id;
3713 r = check_arg_count(argc, 3);
3717 r = read_dev_id(argv[1], &dev_id, 1);
3721 r = read_dev_id(argv[2], &origin_dev_id, 1);
3725 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
3727 DMWARN("Creation of new snapshot %s of device %s failed.",
3735 static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
3740 r = check_arg_count(argc, 2);
3744 r = read_dev_id(argv[1], &dev_id, 1);
3748 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
3750 DMWARN("Deletion of thin device %s failed.", argv[1]);
3755 static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
3757 dm_thin_id old_id, new_id;
3760 r = check_arg_count(argc, 3);
3764 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
3765 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
3769 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
3770 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
3774 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
3776 DMWARN("Failed to change transaction id from %s to %s.",
3784 static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3788 r = check_arg_count(argc, 1);
3792 (void) commit(pool);
3794 r = dm_pool_reserve_metadata_snap(pool->pmd);
3796 DMWARN("reserve_metadata_snap message failed.");
3801 static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3805 r = check_arg_count(argc, 1);
3809 r = dm_pool_release_metadata_snap(pool->pmd);
3811 DMWARN("release_metadata_snap message failed.");
3817 * Messages supported:
3818 * create_thin <dev_id>
3819 * create_snap <dev_id> <origin_id>
3821 * set_transaction_id <current_trans_id> <new_trans_id>
3822 * reserve_metadata_snap
3823 * release_metadata_snap
3825 static int pool_message(struct dm_target *ti, unsigned argc, char **argv,
3826 char *result, unsigned maxlen)
3829 struct pool_c *pt = ti->private;
3830 struct pool *pool = pt->pool;
3832 if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE) {
3833 DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode",
3834 dm_device_name(pool->pool_md));
3838 if (!strcasecmp(argv[0], "create_thin"))
3839 r = process_create_thin_mesg(argc, argv, pool);
3841 else if (!strcasecmp(argv[0], "create_snap"))
3842 r = process_create_snap_mesg(argc, argv, pool);
3844 else if (!strcasecmp(argv[0], "delete"))
3845 r = process_delete_mesg(argc, argv, pool);
3847 else if (!strcasecmp(argv[0], "set_transaction_id"))
3848 r = process_set_transaction_id_mesg(argc, argv, pool);
3850 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
3851 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
3853 else if (!strcasecmp(argv[0], "release_metadata_snap"))
3854 r = process_release_metadata_snap_mesg(argc, argv, pool);
3857 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
3860 (void) commit(pool);
3865 static void emit_flags(struct pool_features *pf, char *result,
3866 unsigned sz, unsigned maxlen)
3868 unsigned count = !pf->zero_new_blocks + !pf->discard_enabled +
3869 !pf->discard_passdown + (pf->mode == PM_READ_ONLY) +
3870 pf->error_if_no_space;
3871 DMEMIT("%u ", count);
3873 if (!pf->zero_new_blocks)
3874 DMEMIT("skip_block_zeroing ");
3876 if (!pf->discard_enabled)
3877 DMEMIT("ignore_discard ");
3879 if (!pf->discard_passdown)
3880 DMEMIT("no_discard_passdown ");
3882 if (pf->mode == PM_READ_ONLY)
3883 DMEMIT("read_only ");
3885 if (pf->error_if_no_space)
3886 DMEMIT("error_if_no_space ");
3891 * <transaction id> <used metadata sectors>/<total metadata sectors>
3892 * <used data sectors>/<total data sectors> <held metadata root>
3893 * <pool mode> <discard config> <no space config> <needs_check>
3895 static void pool_status(struct dm_target *ti, status_type_t type,
3896 unsigned status_flags, char *result, unsigned maxlen)
3900 uint64_t transaction_id;
3901 dm_block_t nr_free_blocks_data;
3902 dm_block_t nr_free_blocks_metadata;
3903 dm_block_t nr_blocks_data;
3904 dm_block_t nr_blocks_metadata;
3905 dm_block_t held_root;
3906 enum pool_mode mode;
3907 char buf[BDEVNAME_SIZE];
3908 char buf2[BDEVNAME_SIZE];
3909 struct pool_c *pt = ti->private;
3910 struct pool *pool = pt->pool;
3913 case STATUSTYPE_INFO:
3914 if (get_pool_mode(pool) == PM_FAIL) {
3919 /* Commit to ensure statistics aren't out-of-date */
3920 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
3921 (void) commit(pool);
3923 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
3925 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
3926 dm_device_name(pool->pool_md), r);
3930 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
3932 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
3933 dm_device_name(pool->pool_md), r);
3937 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
3939 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
3940 dm_device_name(pool->pool_md), r);
3944 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
3946 DMERR("%s: dm_pool_get_free_block_count returned %d",
3947 dm_device_name(pool->pool_md), r);
3951 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
3953 DMERR("%s: dm_pool_get_data_dev_size returned %d",
3954 dm_device_name(pool->pool_md), r);
3958 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
3960 DMERR("%s: dm_pool_get_metadata_snap returned %d",
3961 dm_device_name(pool->pool_md), r);
3965 DMEMIT("%llu %llu/%llu %llu/%llu ",
3966 (unsigned long long)transaction_id,
3967 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
3968 (unsigned long long)nr_blocks_metadata,
3969 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
3970 (unsigned long long)nr_blocks_data);
3973 DMEMIT("%llu ", held_root);
3977 mode = get_pool_mode(pool);
3978 if (mode == PM_OUT_OF_DATA_SPACE)
3979 DMEMIT("out_of_data_space ");
3980 else if (is_read_only_pool_mode(mode))
3985 if (!pool->pf.discard_enabled)
3986 DMEMIT("ignore_discard ");
3987 else if (pool->pf.discard_passdown)
3988 DMEMIT("discard_passdown ");
3990 DMEMIT("no_discard_passdown ");
3992 if (pool->pf.error_if_no_space)
3993 DMEMIT("error_if_no_space ");
3995 DMEMIT("queue_if_no_space ");
3997 if (dm_pool_metadata_needs_check(pool->pmd))
3998 DMEMIT("needs_check ");
4002 DMEMIT("%llu ", (unsigned long long)calc_metadata_threshold(pt));
4006 case STATUSTYPE_TABLE:
4007 DMEMIT("%s %s %lu %llu ",
4008 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
4009 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
4010 (unsigned long)pool->sectors_per_block,
4011 (unsigned long long)pt->low_water_blocks);
4012 emit_flags(&pt->requested_pf, result, sz, maxlen);
4021 static int pool_iterate_devices(struct dm_target *ti,
4022 iterate_devices_callout_fn fn, void *data)
4024 struct pool_c *pt = ti->private;
4026 return fn(ti, pt->data_dev, 0, ti->len, data);
4029 static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
4031 struct pool_c *pt = ti->private;
4032 struct pool *pool = pt->pool;
4033 sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
4036 * If max_sectors is smaller than pool->sectors_per_block adjust it
4037 * to the highest possible power-of-2 factor of pool->sectors_per_block.
4038 * This is especially beneficial when the pool's data device is a RAID
4039 * device that has a full stripe width that matches pool->sectors_per_block
4040 * -- because even though partial RAID stripe-sized IOs will be issued to a
4041 * single RAID stripe; when aggregated they will end on a full RAID stripe
4042 * boundary.. which avoids additional partial RAID stripe writes cascading
4044 if (limits->max_sectors < pool->sectors_per_block) {
4045 while (!is_factor(pool->sectors_per_block, limits->max_sectors)) {
4046 if ((limits->max_sectors & (limits->max_sectors - 1)) == 0)
4047 limits->max_sectors--;
4048 limits->max_sectors = rounddown_pow_of_two(limits->max_sectors);
4053 * If the system-determined stacked limits are compatible with the
4054 * pool's blocksize (io_opt is a factor) do not override them.
4056 if (io_opt_sectors < pool->sectors_per_block ||
4057 !is_factor(io_opt_sectors, pool->sectors_per_block)) {
4058 if (is_factor(pool->sectors_per_block, limits->max_sectors))
4059 blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT);
4061 blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT);
4062 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
4066 * pt->adjusted_pf is a staging area for the actual features to use.
4067 * They get transferred to the live pool in bind_control_target()
4068 * called from pool_preresume().
4070 if (!pt->adjusted_pf.discard_enabled) {
4072 * Must explicitly disallow stacking discard limits otherwise the
4073 * block layer will stack them if pool's data device has support.
4074 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
4075 * user to see that, so make sure to set all discard limits to 0.
4077 limits->discard_granularity = 0;
4081 disable_passdown_if_not_supported(pt);
4084 * The pool uses the same discard limits as the underlying data
4085 * device. DM core has already set this up.
4089 static struct target_type pool_target = {
4090 .name = "thin-pool",
4091 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
4092 DM_TARGET_IMMUTABLE,
4093 .version = {1, 20, 0},
4094 .module = THIS_MODULE,
4098 .presuspend = pool_presuspend,
4099 .presuspend_undo = pool_presuspend_undo,
4100 .postsuspend = pool_postsuspend,
4101 .preresume = pool_preresume,
4102 .resume = pool_resume,
4103 .message = pool_message,
4104 .status = pool_status,
4105 .iterate_devices = pool_iterate_devices,
4106 .io_hints = pool_io_hints,
4109 /*----------------------------------------------------------------
4110 * Thin target methods
4111 *--------------------------------------------------------------*/
4112 static void thin_get(struct thin_c *tc)
4114 atomic_inc(&tc->refcount);
4117 static void thin_put(struct thin_c *tc)
4119 if (atomic_dec_and_test(&tc->refcount))
4120 complete(&tc->can_destroy);
4123 static void thin_dtr(struct dm_target *ti)
4125 struct thin_c *tc = ti->private;
4126 unsigned long flags;
4128 spin_lock_irqsave(&tc->pool->lock, flags);
4129 list_del_rcu(&tc->list);
4130 spin_unlock_irqrestore(&tc->pool->lock, flags);
4134 wait_for_completion(&tc->can_destroy);
4136 mutex_lock(&dm_thin_pool_table.mutex);
4138 __pool_dec(tc->pool);
4139 dm_pool_close_thin_device(tc->td);
4140 dm_put_device(ti, tc->pool_dev);
4142 dm_put_device(ti, tc->origin_dev);
4145 mutex_unlock(&dm_thin_pool_table.mutex);
4149 * Thin target parameters:
4151 * <pool_dev> <dev_id> [origin_dev]
4153 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
4154 * dev_id: the internal device identifier
4155 * origin_dev: a device external to the pool that should act as the origin
4157 * If the pool device has discards disabled, they get disabled for the thin
4160 static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
4164 struct dm_dev *pool_dev, *origin_dev;
4165 struct mapped_device *pool_md;
4166 unsigned long flags;
4168 mutex_lock(&dm_thin_pool_table.mutex);
4170 if (argc != 2 && argc != 3) {
4171 ti->error = "Invalid argument count";
4176 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
4178 ti->error = "Out of memory";
4182 tc->thin_md = dm_table_get_md(ti->table);
4183 spin_lock_init(&tc->lock);
4184 INIT_LIST_HEAD(&tc->deferred_cells);
4185 bio_list_init(&tc->deferred_bio_list);
4186 bio_list_init(&tc->retry_on_resume_list);
4187 tc->sort_bio_list = RB_ROOT;
4190 if (!strcmp(argv[0], argv[2])) {
4191 ti->error = "Error setting origin device";
4193 goto bad_origin_dev;
4196 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
4198 ti->error = "Error opening origin device";
4199 goto bad_origin_dev;
4201 tc->origin_dev = origin_dev;
4204 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
4206 ti->error = "Error opening pool device";
4209 tc->pool_dev = pool_dev;
4211 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
4212 ti->error = "Invalid device id";
4217 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
4219 ti->error = "Couldn't get pool mapped device";
4224 tc->pool = __pool_table_lookup(pool_md);
4226 ti->error = "Couldn't find pool object";
4228 goto bad_pool_lookup;
4230 __pool_inc(tc->pool);
4232 if (get_pool_mode(tc->pool) == PM_FAIL) {
4233 ti->error = "Couldn't open thin device, Pool is in fail mode";
4238 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
4240 ti->error = "Couldn't open thin internal device";
4244 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
4248 ti->num_flush_bios = 1;
4249 ti->limit_swap_bios = true;
4250 ti->flush_supported = true;
4251 ti->per_io_data_size = sizeof(struct dm_thin_endio_hook);
4253 /* In case the pool supports discards, pass them on. */
4254 if (tc->pool->pf.discard_enabled) {
4255 ti->discards_supported = true;
4256 ti->num_discard_bios = 1;
4257 ti->split_discard_bios = false;
4260 mutex_unlock(&dm_thin_pool_table.mutex);
4262 spin_lock_irqsave(&tc->pool->lock, flags);
4263 if (tc->pool->suspended) {
4264 spin_unlock_irqrestore(&tc->pool->lock, flags);
4265 mutex_lock(&dm_thin_pool_table.mutex); /* reacquire for __pool_dec */
4266 ti->error = "Unable to activate thin device while pool is suspended";
4270 atomic_set(&tc->refcount, 1);
4271 init_completion(&tc->can_destroy);
4272 list_add_tail_rcu(&tc->list, &tc->pool->active_thins);
4273 spin_unlock_irqrestore(&tc->pool->lock, flags);
4275 * This synchronize_rcu() call is needed here otherwise we risk a
4276 * wake_worker() call finding no bios to process (because the newly
4277 * added tc isn't yet visible). So this reduces latency since we
4278 * aren't then dependent on the periodic commit to wake_worker().
4287 dm_pool_close_thin_device(tc->td);
4289 __pool_dec(tc->pool);
4293 dm_put_device(ti, tc->pool_dev);
4296 dm_put_device(ti, tc->origin_dev);
4300 mutex_unlock(&dm_thin_pool_table.mutex);
4305 static int thin_map(struct dm_target *ti, struct bio *bio)
4307 bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
4309 return thin_bio_map(ti, bio);
4312 static int thin_endio(struct dm_target *ti, struct bio *bio,
4315 unsigned long flags;
4316 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
4317 struct list_head work;
4318 struct dm_thin_new_mapping *m, *tmp;
4319 struct pool *pool = h->tc->pool;
4321 if (h->shared_read_entry) {
4322 INIT_LIST_HEAD(&work);
4323 dm_deferred_entry_dec(h->shared_read_entry, &work);
4325 spin_lock_irqsave(&pool->lock, flags);
4326 list_for_each_entry_safe(m, tmp, &work, list) {
4328 __complete_mapping_preparation(m);
4330 spin_unlock_irqrestore(&pool->lock, flags);
4333 if (h->all_io_entry) {
4334 INIT_LIST_HEAD(&work);
4335 dm_deferred_entry_dec(h->all_io_entry, &work);
4336 if (!list_empty(&work)) {
4337 spin_lock_irqsave(&pool->lock, flags);
4338 list_for_each_entry_safe(m, tmp, &work, list)
4339 list_add_tail(&m->list, &pool->prepared_discards);
4340 spin_unlock_irqrestore(&pool->lock, flags);
4346 cell_defer_no_holder(h->tc, h->cell);
4348 return DM_ENDIO_DONE;
4351 static void thin_presuspend(struct dm_target *ti)
4353 struct thin_c *tc = ti->private;
4355 if (dm_noflush_suspending(ti))
4356 noflush_work(tc, do_noflush_start);
4359 static void thin_postsuspend(struct dm_target *ti)
4361 struct thin_c *tc = ti->private;
4364 * The dm_noflush_suspending flag has been cleared by now, so
4365 * unfortunately we must always run this.
4367 noflush_work(tc, do_noflush_stop);
4370 static int thin_preresume(struct dm_target *ti)
4372 struct thin_c *tc = ti->private;
4375 tc->origin_size = get_dev_size(tc->origin_dev->bdev);
4381 * <nr mapped sectors> <highest mapped sector>
4383 static void thin_status(struct dm_target *ti, status_type_t type,
4384 unsigned status_flags, char *result, unsigned maxlen)
4388 dm_block_t mapped, highest;
4389 char buf[BDEVNAME_SIZE];
4390 struct thin_c *tc = ti->private;
4392 if (get_pool_mode(tc->pool) == PM_FAIL) {
4401 case STATUSTYPE_INFO:
4402 r = dm_thin_get_mapped_count(tc->td, &mapped);
4404 DMERR("dm_thin_get_mapped_count returned %d", r);
4408 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
4410 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
4414 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
4416 DMEMIT("%llu", ((highest + 1) *
4417 tc->pool->sectors_per_block) - 1);
4422 case STATUSTYPE_TABLE:
4424 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
4425 (unsigned long) tc->dev_id);
4427 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
4438 static int thin_iterate_devices(struct dm_target *ti,
4439 iterate_devices_callout_fn fn, void *data)
4442 struct thin_c *tc = ti->private;
4443 struct pool *pool = tc->pool;
4446 * We can't call dm_pool_get_data_dev_size() since that blocks. So
4447 * we follow a more convoluted path through to the pool's target.
4450 return 0; /* nothing is bound */
4452 blocks = pool->ti->len;
4453 (void) sector_div(blocks, pool->sectors_per_block);
4455 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
4460 static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits)
4462 struct thin_c *tc = ti->private;
4463 struct pool *pool = tc->pool;
4465 if (!pool->pf.discard_enabled)
4468 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
4469 limits->max_discard_sectors = 2048 * 1024 * 16; /* 16G */
4472 static struct target_type thin_target = {
4474 .version = {1, 20, 0},
4475 .module = THIS_MODULE,
4479 .end_io = thin_endio,
4480 .preresume = thin_preresume,
4481 .presuspend = thin_presuspend,
4482 .postsuspend = thin_postsuspend,
4483 .status = thin_status,
4484 .iterate_devices = thin_iterate_devices,
4485 .io_hints = thin_io_hints,
4488 /*----------------------------------------------------------------*/
4490 static int __init dm_thin_init(void)
4496 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
4497 if (!_new_mapping_cache)
4500 r = dm_register_target(&thin_target);
4502 goto bad_new_mapping_cache;
4504 r = dm_register_target(&pool_target);
4506 goto bad_thin_target;
4511 dm_unregister_target(&thin_target);
4512 bad_new_mapping_cache:
4513 kmem_cache_destroy(_new_mapping_cache);
4518 static void dm_thin_exit(void)
4520 dm_unregister_target(&thin_target);
4521 dm_unregister_target(&pool_target);
4523 kmem_cache_destroy(_new_mapping_cache);
4528 module_init(dm_thin_init);
4529 module_exit(dm_thin_exit);
4531 module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR);
4532 MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds");
4534 MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
4535 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
4536 MODULE_LICENSE("GPL");