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 && time_is_before_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 *data_dev;
235 struct block_device *md_dev;
236 struct dm_pool_metadata *pmd;
238 dm_block_t low_water_blocks;
239 uint32_t sectors_per_block;
240 int sectors_per_block_shift;
242 struct pool_features pf;
243 bool low_water_triggered:1; /* A dm event has been sent */
245 bool out_of_data_space:1;
247 struct dm_bio_prison *prison;
248 struct dm_kcopyd_client *copier;
250 struct work_struct worker;
251 struct workqueue_struct *wq;
252 struct throttle throttle;
253 struct delayed_work waker;
254 struct delayed_work no_space_timeout;
256 unsigned long last_commit_jiffies;
260 struct bio_list deferred_flush_bios;
261 struct bio_list deferred_flush_completions;
262 struct list_head prepared_mappings;
263 struct list_head prepared_discards;
264 struct list_head prepared_discards_pt2;
265 struct list_head active_thins;
267 struct dm_deferred_set *shared_read_ds;
268 struct dm_deferred_set *all_io_ds;
270 struct dm_thin_new_mapping *next_mapping;
272 process_bio_fn process_bio;
273 process_bio_fn process_discard;
275 process_cell_fn process_cell;
276 process_cell_fn process_discard_cell;
278 process_mapping_fn process_prepared_mapping;
279 process_mapping_fn process_prepared_discard;
280 process_mapping_fn process_prepared_discard_pt2;
282 struct dm_bio_prison_cell **cell_sort_array;
284 mempool_t mapping_pool;
287 static void metadata_operation_failed(struct pool *pool, const char *op, int r);
289 static enum pool_mode get_pool_mode(struct pool *pool)
291 return pool->pf.mode;
294 static void notify_of_pool_mode_change(struct pool *pool)
296 const char *descs[] = {
303 const char *extra_desc = NULL;
304 enum pool_mode mode = get_pool_mode(pool);
306 if (mode == PM_OUT_OF_DATA_SPACE) {
307 if (!pool->pf.error_if_no_space)
308 extra_desc = " (queue IO)";
310 extra_desc = " (error IO)";
313 dm_table_event(pool->ti->table);
314 DMINFO("%s: switching pool to %s%s mode",
315 dm_device_name(pool->pool_md),
316 descs[(int)mode], extra_desc ? : "");
320 * Target context for a pool.
323 struct dm_target *ti;
325 struct dm_dev *data_dev;
326 struct dm_dev *metadata_dev;
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, GFP_NOWAIT,
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;
613 bio_list_init(&bios);
615 spin_lock_irq(&tc->lock);
616 __merge_bio_list(&bios, master);
617 spin_unlock_irq(&tc->lock);
619 error_bio_list(&bios, error);
622 static void requeue_deferred_cells(struct thin_c *tc)
624 struct pool *pool = tc->pool;
625 struct list_head cells;
626 struct dm_bio_prison_cell *cell, *tmp;
628 INIT_LIST_HEAD(&cells);
630 spin_lock_irq(&tc->lock);
631 list_splice_init(&tc->deferred_cells, &cells);
632 spin_unlock_irq(&tc->lock);
634 list_for_each_entry_safe(cell, tmp, &cells, user_list)
635 cell_requeue(pool, cell);
638 static void requeue_io(struct thin_c *tc)
640 struct bio_list bios;
642 bio_list_init(&bios);
644 spin_lock_irq(&tc->lock);
645 __merge_bio_list(&bios, &tc->deferred_bio_list);
646 __merge_bio_list(&bios, &tc->retry_on_resume_list);
647 spin_unlock_irq(&tc->lock);
649 error_bio_list(&bios, BLK_STS_DM_REQUEUE);
650 requeue_deferred_cells(tc);
653 static void error_retry_list_with_code(struct pool *pool, blk_status_t error)
658 list_for_each_entry_rcu(tc, &pool->active_thins, list)
659 error_thin_bio_list(tc, &tc->retry_on_resume_list, error);
663 static void error_retry_list(struct pool *pool)
665 error_retry_list_with_code(pool, get_pool_io_error_code(pool));
669 * This section of code contains the logic for processing a thin device's IO.
670 * Much of the code depends on pool object resources (lists, workqueues, etc)
671 * but most is exclusively called from the thin target rather than the thin-pool
675 static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
677 struct pool *pool = tc->pool;
678 sector_t block_nr = bio->bi_iter.bi_sector;
680 if (block_size_is_power_of_two(pool))
681 block_nr >>= pool->sectors_per_block_shift;
683 (void) sector_div(block_nr, pool->sectors_per_block);
689 * Returns the _complete_ blocks that this bio covers.
691 static void get_bio_block_range(struct thin_c *tc, struct bio *bio,
692 dm_block_t *begin, dm_block_t *end)
694 struct pool *pool = tc->pool;
695 sector_t b = bio->bi_iter.bi_sector;
696 sector_t e = b + (bio->bi_iter.bi_size >> SECTOR_SHIFT);
698 b += pool->sectors_per_block - 1ull; /* so we round up */
700 if (block_size_is_power_of_two(pool)) {
701 b >>= pool->sectors_per_block_shift;
702 e >>= pool->sectors_per_block_shift;
704 (void) sector_div(b, pool->sectors_per_block);
705 (void) sector_div(e, pool->sectors_per_block);
709 /* Can happen if the bio is within a single block. */
716 static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
718 struct pool *pool = tc->pool;
719 sector_t bi_sector = bio->bi_iter.bi_sector;
721 bio_set_dev(bio, tc->pool_dev->bdev);
722 if (block_size_is_power_of_two(pool))
723 bio->bi_iter.bi_sector =
724 (block << pool->sectors_per_block_shift) |
725 (bi_sector & (pool->sectors_per_block - 1));
727 bio->bi_iter.bi_sector = (block * pool->sectors_per_block) +
728 sector_div(bi_sector, pool->sectors_per_block);
731 static void remap_to_origin(struct thin_c *tc, struct bio *bio)
733 bio_set_dev(bio, tc->origin_dev->bdev);
736 static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
738 return op_is_flush(bio->bi_opf) &&
739 dm_thin_changed_this_transaction(tc->td);
742 static void inc_all_io_entry(struct pool *pool, struct bio *bio)
744 struct dm_thin_endio_hook *h;
746 if (bio_op(bio) == REQ_OP_DISCARD)
749 h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
750 h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
753 static void issue(struct thin_c *tc, struct bio *bio)
755 struct pool *pool = tc->pool;
757 if (!bio_triggers_commit(tc, bio)) {
758 dm_submit_bio_remap(bio, NULL);
763 * Complete bio with an error if earlier I/O caused changes to
764 * the metadata that can't be committed e.g, due to I/O errors
765 * on the metadata device.
767 if (dm_thin_aborted_changes(tc->td)) {
773 * Batch together any bios that trigger commits and then issue a
774 * single commit for them in process_deferred_bios().
776 spin_lock_irq(&pool->lock);
777 bio_list_add(&pool->deferred_flush_bios, bio);
778 spin_unlock_irq(&pool->lock);
781 static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
783 remap_to_origin(tc, bio);
787 static void remap_and_issue(struct thin_c *tc, struct bio *bio,
790 remap(tc, bio, block);
794 /*----------------------------------------------------------------*/
797 * Bio endio functions.
799 struct dm_thin_new_mapping {
800 struct list_head list;
806 * Track quiescing, copying and zeroing preparation actions. When this
807 * counter hits zero the block is prepared and can be inserted into the
810 atomic_t prepare_actions;
814 dm_block_t virt_begin, virt_end;
815 dm_block_t data_block;
816 struct dm_bio_prison_cell *cell;
819 * If the bio covers the whole area of a block then we can avoid
820 * zeroing or copying. Instead this bio is hooked. The bio will
821 * still be in the cell, so care has to be taken to avoid issuing
825 bio_end_io_t *saved_bi_end_io;
828 static void __complete_mapping_preparation(struct dm_thin_new_mapping *m)
830 struct pool *pool = m->tc->pool;
832 if (atomic_dec_and_test(&m->prepare_actions)) {
833 list_add_tail(&m->list, &pool->prepared_mappings);
838 static void complete_mapping_preparation(struct dm_thin_new_mapping *m)
841 struct pool *pool = m->tc->pool;
843 spin_lock_irqsave(&pool->lock, flags);
844 __complete_mapping_preparation(m);
845 spin_unlock_irqrestore(&pool->lock, flags);
848 static void copy_complete(int read_err, unsigned long write_err, void *context)
850 struct dm_thin_new_mapping *m = context;
852 m->status = read_err || write_err ? BLK_STS_IOERR : 0;
853 complete_mapping_preparation(m);
856 static void overwrite_endio(struct bio *bio)
858 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
859 struct dm_thin_new_mapping *m = h->overwrite_mapping;
861 bio->bi_end_io = m->saved_bi_end_io;
863 m->status = bio->bi_status;
864 complete_mapping_preparation(m);
867 /*----------------------------------------------------------------*/
874 * Prepared mapping jobs.
878 * This sends the bios in the cell, except the original holder, back
879 * to the deferred_bios list.
881 static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
883 struct pool *pool = tc->pool;
887 spin_lock_irqsave(&tc->lock, flags);
888 cell_release_no_holder(pool, cell, &tc->deferred_bio_list);
889 has_work = !bio_list_empty(&tc->deferred_bio_list);
890 spin_unlock_irqrestore(&tc->lock, flags);
896 static void thin_defer_bio(struct thin_c *tc, struct bio *bio);
900 struct bio_list defer_bios;
901 struct bio_list issue_bios;
904 static void __inc_remap_and_issue_cell(void *context,
905 struct dm_bio_prison_cell *cell)
907 struct remap_info *info = context;
910 while ((bio = bio_list_pop(&cell->bios))) {
911 if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD)
912 bio_list_add(&info->defer_bios, bio);
914 inc_all_io_entry(info->tc->pool, bio);
917 * We can't issue the bios with the bio prison lock
918 * held, so we add them to a list to issue on
919 * return from this function.
921 bio_list_add(&info->issue_bios, bio);
926 static void inc_remap_and_issue_cell(struct thin_c *tc,
927 struct dm_bio_prison_cell *cell,
931 struct remap_info info;
934 bio_list_init(&info.defer_bios);
935 bio_list_init(&info.issue_bios);
938 * We have to be careful to inc any bios we're about to issue
939 * before the cell is released, and avoid a race with new bios
940 * being added to the cell.
942 cell_visit_release(tc->pool, __inc_remap_and_issue_cell,
945 while ((bio = bio_list_pop(&info.defer_bios)))
946 thin_defer_bio(tc, bio);
948 while ((bio = bio_list_pop(&info.issue_bios)))
949 remap_and_issue(info.tc, bio, block);
952 static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
954 cell_error(m->tc->pool, m->cell);
956 mempool_free(m, &m->tc->pool->mapping_pool);
959 static void complete_overwrite_bio(struct thin_c *tc, struct bio *bio)
961 struct pool *pool = tc->pool;
964 * If the bio has the REQ_FUA flag set we must commit the metadata
965 * before signaling its completion.
967 if (!bio_triggers_commit(tc, bio)) {
973 * Complete bio with an error if earlier I/O caused changes to the
974 * metadata that can't be committed, e.g, due to I/O errors on the
977 if (dm_thin_aborted_changes(tc->td)) {
983 * Batch together any bios that trigger commits and then issue a
984 * single commit for them in process_deferred_bios().
986 spin_lock_irq(&pool->lock);
987 bio_list_add(&pool->deferred_flush_completions, bio);
988 spin_unlock_irq(&pool->lock);
991 static void process_prepared_mapping(struct dm_thin_new_mapping *m)
993 struct thin_c *tc = m->tc;
994 struct pool *pool = tc->pool;
995 struct bio *bio = m->bio;
999 cell_error(pool, m->cell);
1004 * Commit the prepared block into the mapping btree.
1005 * Any I/O for this block arriving after this point will get
1006 * remapped to it directly.
1008 r = dm_thin_insert_block(tc->td, m->virt_begin, m->data_block);
1010 metadata_operation_failed(pool, "dm_thin_insert_block", r);
1011 cell_error(pool, m->cell);
1016 * Release any bios held while the block was being provisioned.
1017 * If we are processing a write bio that completely covers the block,
1018 * we already processed it so can ignore it now when processing
1019 * the bios in the cell.
1022 inc_remap_and_issue_cell(tc, m->cell, m->data_block);
1023 complete_overwrite_bio(tc, bio);
1025 inc_all_io_entry(tc->pool, m->cell->holder);
1026 remap_and_issue(tc, m->cell->holder, m->data_block);
1027 inc_remap_and_issue_cell(tc, m->cell, m->data_block);
1032 mempool_free(m, &pool->mapping_pool);
1035 /*----------------------------------------------------------------*/
1037 static void free_discard_mapping(struct dm_thin_new_mapping *m)
1039 struct thin_c *tc = m->tc;
1041 cell_defer_no_holder(tc, m->cell);
1042 mempool_free(m, &tc->pool->mapping_pool);
1045 static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
1047 bio_io_error(m->bio);
1048 free_discard_mapping(m);
1051 static void process_prepared_discard_success(struct dm_thin_new_mapping *m)
1054 free_discard_mapping(m);
1057 static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m)
1060 struct thin_c *tc = m->tc;
1062 r = dm_thin_remove_range(tc->td, m->cell->key.block_begin, m->cell->key.block_end);
1064 metadata_operation_failed(tc->pool, "dm_thin_remove_range", r);
1065 bio_io_error(m->bio);
1069 cell_defer_no_holder(tc, m->cell);
1070 mempool_free(m, &tc->pool->mapping_pool);
1073 /*----------------------------------------------------------------*/
1075 static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m,
1076 struct bio *discard_parent)
1079 * We've already unmapped this range of blocks, but before we
1080 * passdown we have to check that these blocks are now unused.
1084 struct thin_c *tc = m->tc;
1085 struct pool *pool = tc->pool;
1086 dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin;
1087 struct discard_op op;
1089 begin_discard(&op, tc, discard_parent);
1091 /* find start of unmapped run */
1092 for (; b < end; b++) {
1093 r = dm_pool_block_is_shared(pool->pmd, b, &shared);
1104 /* find end of run */
1105 for (e = b + 1; e != end; e++) {
1106 r = dm_pool_block_is_shared(pool->pmd, e, &shared);
1114 r = issue_discard(&op, b, e);
1121 end_discard(&op, r);
1124 static void queue_passdown_pt2(struct dm_thin_new_mapping *m)
1126 unsigned long flags;
1127 struct pool *pool = m->tc->pool;
1129 spin_lock_irqsave(&pool->lock, flags);
1130 list_add_tail(&m->list, &pool->prepared_discards_pt2);
1131 spin_unlock_irqrestore(&pool->lock, flags);
1135 static void passdown_endio(struct bio *bio)
1138 * It doesn't matter if the passdown discard failed, we still want
1139 * to unmap (we ignore err).
1141 queue_passdown_pt2(bio->bi_private);
1145 static void process_prepared_discard_passdown_pt1(struct dm_thin_new_mapping *m)
1148 struct thin_c *tc = m->tc;
1149 struct pool *pool = tc->pool;
1150 struct bio *discard_parent;
1151 dm_block_t data_end = m->data_block + (m->virt_end - m->virt_begin);
1154 * Only this thread allocates blocks, so we can be sure that the
1155 * newly unmapped blocks will not be allocated before the end of
1158 r = dm_thin_remove_range(tc->td, m->virt_begin, m->virt_end);
1160 metadata_operation_failed(pool, "dm_thin_remove_range", r);
1161 bio_io_error(m->bio);
1162 cell_defer_no_holder(tc, m->cell);
1163 mempool_free(m, &pool->mapping_pool);
1168 * Increment the unmapped blocks. This prevents a race between the
1169 * passdown io and reallocation of freed blocks.
1171 r = dm_pool_inc_data_range(pool->pmd, m->data_block, data_end);
1173 metadata_operation_failed(pool, "dm_pool_inc_data_range", r);
1174 bio_io_error(m->bio);
1175 cell_defer_no_holder(tc, m->cell);
1176 mempool_free(m, &pool->mapping_pool);
1180 discard_parent = bio_alloc(NULL, 1, 0, GFP_NOIO);
1181 discard_parent->bi_end_io = passdown_endio;
1182 discard_parent->bi_private = m;
1183 if (m->maybe_shared)
1184 passdown_double_checking_shared_status(m, discard_parent);
1186 struct discard_op op;
1188 begin_discard(&op, tc, discard_parent);
1189 r = issue_discard(&op, m->data_block, data_end);
1190 end_discard(&op, r);
1194 static void process_prepared_discard_passdown_pt2(struct dm_thin_new_mapping *m)
1197 struct thin_c *tc = m->tc;
1198 struct pool *pool = tc->pool;
1201 * The passdown has completed, so now we can decrement all those
1204 r = dm_pool_dec_data_range(pool->pmd, m->data_block,
1205 m->data_block + (m->virt_end - m->virt_begin));
1207 metadata_operation_failed(pool, "dm_pool_dec_data_range", r);
1208 bio_io_error(m->bio);
1212 cell_defer_no_holder(tc, m->cell);
1213 mempool_free(m, &pool->mapping_pool);
1216 static void process_prepared(struct pool *pool, struct list_head *head,
1217 process_mapping_fn *fn)
1219 struct list_head maps;
1220 struct dm_thin_new_mapping *m, *tmp;
1222 INIT_LIST_HEAD(&maps);
1223 spin_lock_irq(&pool->lock);
1224 list_splice_init(head, &maps);
1225 spin_unlock_irq(&pool->lock);
1227 list_for_each_entry_safe(m, tmp, &maps, list)
1232 * Deferred bio jobs.
1234 static int io_overlaps_block(struct pool *pool, struct bio *bio)
1236 return bio->bi_iter.bi_size ==
1237 (pool->sectors_per_block << SECTOR_SHIFT);
1240 static int io_overwrites_block(struct pool *pool, struct bio *bio)
1242 return (bio_data_dir(bio) == WRITE) &&
1243 io_overlaps_block(pool, bio);
1246 static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
1249 *save = bio->bi_end_io;
1250 bio->bi_end_io = fn;
1253 static int ensure_next_mapping(struct pool *pool)
1255 if (pool->next_mapping)
1258 pool->next_mapping = mempool_alloc(&pool->mapping_pool, GFP_ATOMIC);
1260 return pool->next_mapping ? 0 : -ENOMEM;
1263 static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
1265 struct dm_thin_new_mapping *m = pool->next_mapping;
1267 BUG_ON(!pool->next_mapping);
1269 memset(m, 0, sizeof(struct dm_thin_new_mapping));
1270 INIT_LIST_HEAD(&m->list);
1273 pool->next_mapping = NULL;
1278 static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m,
1279 sector_t begin, sector_t end)
1281 struct dm_io_region to;
1283 to.bdev = tc->pool_dev->bdev;
1285 to.count = end - begin;
1287 dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m);
1290 static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio,
1291 dm_block_t data_begin,
1292 struct dm_thin_new_mapping *m)
1294 struct pool *pool = tc->pool;
1295 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1297 h->overwrite_mapping = m;
1299 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
1300 inc_all_io_entry(pool, bio);
1301 remap_and_issue(tc, bio, data_begin);
1305 * A partial copy also needs to zero the uncopied region.
1307 static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
1308 struct dm_dev *origin, dm_block_t data_origin,
1309 dm_block_t data_dest,
1310 struct dm_bio_prison_cell *cell, struct bio *bio,
1313 struct pool *pool = tc->pool;
1314 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1317 m->virt_begin = virt_block;
1318 m->virt_end = virt_block + 1u;
1319 m->data_block = data_dest;
1323 * quiesce action + copy action + an extra reference held for the
1324 * duration of this function (we may need to inc later for a
1327 atomic_set(&m->prepare_actions, 3);
1329 if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
1330 complete_mapping_preparation(m); /* already quiesced */
1333 * IO to pool_dev remaps to the pool target's data_dev.
1335 * If the whole block of data is being overwritten, we can issue the
1336 * bio immediately. Otherwise we use kcopyd to clone the data first.
1338 if (io_overwrites_block(pool, bio))
1339 remap_and_issue_overwrite(tc, bio, data_dest, m);
1341 struct dm_io_region from, to;
1343 from.bdev = origin->bdev;
1344 from.sector = data_origin * pool->sectors_per_block;
1347 to.bdev = tc->pool_dev->bdev;
1348 to.sector = data_dest * pool->sectors_per_block;
1351 dm_kcopyd_copy(pool->copier, &from, 1, &to,
1352 0, copy_complete, m);
1355 * Do we need to zero a tail region?
1357 if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) {
1358 atomic_inc(&m->prepare_actions);
1360 data_dest * pool->sectors_per_block + len,
1361 (data_dest + 1) * pool->sectors_per_block);
1365 complete_mapping_preparation(m); /* drop our ref */
1368 static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
1369 dm_block_t data_origin, dm_block_t data_dest,
1370 struct dm_bio_prison_cell *cell, struct bio *bio)
1372 schedule_copy(tc, virt_block, tc->pool_dev,
1373 data_origin, data_dest, cell, bio,
1374 tc->pool->sectors_per_block);
1377 static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
1378 dm_block_t data_block, struct dm_bio_prison_cell *cell,
1381 struct pool *pool = tc->pool;
1382 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1384 atomic_set(&m->prepare_actions, 1); /* no need to quiesce */
1386 m->virt_begin = virt_block;
1387 m->virt_end = virt_block + 1u;
1388 m->data_block = data_block;
1392 * If the whole block of data is being overwritten or we are not
1393 * zeroing pre-existing data, we can issue the bio immediately.
1394 * Otherwise we use kcopyd to zero the data first.
1396 if (pool->pf.zero_new_blocks) {
1397 if (io_overwrites_block(pool, bio))
1398 remap_and_issue_overwrite(tc, bio, data_block, m);
1400 ll_zero(tc, m, data_block * pool->sectors_per_block,
1401 (data_block + 1) * pool->sectors_per_block);
1403 process_prepared_mapping(m);
1406 static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
1407 dm_block_t data_dest,
1408 struct dm_bio_prison_cell *cell, struct bio *bio)
1410 struct pool *pool = tc->pool;
1411 sector_t virt_block_begin = virt_block * pool->sectors_per_block;
1412 sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block;
1414 if (virt_block_end <= tc->origin_size)
1415 schedule_copy(tc, virt_block, tc->origin_dev,
1416 virt_block, data_dest, cell, bio,
1417 pool->sectors_per_block);
1419 else if (virt_block_begin < tc->origin_size)
1420 schedule_copy(tc, virt_block, tc->origin_dev,
1421 virt_block, data_dest, cell, bio,
1422 tc->origin_size - virt_block_begin);
1425 schedule_zero(tc, virt_block, data_dest, cell, bio);
1428 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode);
1430 static void requeue_bios(struct pool *pool);
1432 static bool is_read_only_pool_mode(enum pool_mode mode)
1434 return (mode == PM_OUT_OF_METADATA_SPACE || mode == PM_READ_ONLY);
1437 static bool is_read_only(struct pool *pool)
1439 return is_read_only_pool_mode(get_pool_mode(pool));
1442 static void check_for_metadata_space(struct pool *pool)
1445 const char *ooms_reason = NULL;
1448 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free);
1450 ooms_reason = "Could not get free metadata blocks";
1452 ooms_reason = "No free metadata blocks";
1454 if (ooms_reason && !is_read_only(pool)) {
1455 DMERR("%s", ooms_reason);
1456 set_pool_mode(pool, PM_OUT_OF_METADATA_SPACE);
1460 static void check_for_data_space(struct pool *pool)
1465 if (get_pool_mode(pool) != PM_OUT_OF_DATA_SPACE)
1468 r = dm_pool_get_free_block_count(pool->pmd, &nr_free);
1473 set_pool_mode(pool, PM_WRITE);
1479 * A non-zero return indicates read_only or fail_io mode.
1480 * Many callers don't care about the return value.
1482 static int commit(struct pool *pool)
1486 if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE)
1489 r = dm_pool_commit_metadata(pool->pmd);
1491 metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
1493 check_for_metadata_space(pool);
1494 check_for_data_space(pool);
1500 static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks)
1502 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
1503 DMWARN("%s: reached low water mark for data device: sending event.",
1504 dm_device_name(pool->pool_md));
1505 spin_lock_irq(&pool->lock);
1506 pool->low_water_triggered = true;
1507 spin_unlock_irq(&pool->lock);
1508 dm_table_event(pool->ti->table);
1512 static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
1515 dm_block_t free_blocks;
1516 struct pool *pool = tc->pool;
1518 if (WARN_ON(get_pool_mode(pool) != PM_WRITE))
1521 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1523 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1527 check_low_water_mark(pool, free_blocks);
1531 * Try to commit to see if that will free up some
1538 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1540 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1545 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1550 r = dm_pool_alloc_data_block(pool->pmd, result);
1553 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1555 metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
1559 r = dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks);
1561 metadata_operation_failed(pool, "dm_pool_get_free_metadata_block_count", r);
1566 /* Let's commit before we use up the metadata reserve. */
1576 * If we have run out of space, queue bios until the device is
1577 * resumed, presumably after having been reloaded with more space.
1579 static void retry_on_resume(struct bio *bio)
1581 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1582 struct thin_c *tc = h->tc;
1584 spin_lock_irq(&tc->lock);
1585 bio_list_add(&tc->retry_on_resume_list, bio);
1586 spin_unlock_irq(&tc->lock);
1589 static blk_status_t should_error_unserviceable_bio(struct pool *pool)
1591 enum pool_mode m = get_pool_mode(pool);
1595 /* Shouldn't get here */
1596 DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
1597 return BLK_STS_IOERR;
1599 case PM_OUT_OF_DATA_SPACE:
1600 return pool->pf.error_if_no_space ? BLK_STS_NOSPC : 0;
1602 case PM_OUT_OF_METADATA_SPACE:
1605 return BLK_STS_IOERR;
1607 /* Shouldn't get here */
1608 DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1609 return BLK_STS_IOERR;
1613 static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
1615 blk_status_t error = should_error_unserviceable_bio(pool);
1618 bio->bi_status = error;
1621 retry_on_resume(bio);
1624 static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell)
1627 struct bio_list bios;
1630 error = should_error_unserviceable_bio(pool);
1632 cell_error_with_code(pool, cell, error);
1636 bio_list_init(&bios);
1637 cell_release(pool, cell, &bios);
1639 while ((bio = bio_list_pop(&bios)))
1640 retry_on_resume(bio);
1643 static void process_discard_cell_no_passdown(struct thin_c *tc,
1644 struct dm_bio_prison_cell *virt_cell)
1646 struct pool *pool = tc->pool;
1647 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1650 * We don't need to lock the data blocks, since there's no
1651 * passdown. We only lock data blocks for allocation and breaking sharing.
1654 m->virt_begin = virt_cell->key.block_begin;
1655 m->virt_end = virt_cell->key.block_end;
1656 m->cell = virt_cell;
1657 m->bio = virt_cell->holder;
1659 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1660 pool->process_prepared_discard(m);
1663 static void break_up_discard_bio(struct thin_c *tc, dm_block_t begin, dm_block_t end,
1666 struct pool *pool = tc->pool;
1670 struct dm_cell_key data_key;
1671 struct dm_bio_prison_cell *data_cell;
1672 struct dm_thin_new_mapping *m;
1673 dm_block_t virt_begin, virt_end, data_begin;
1675 while (begin != end) {
1676 r = ensure_next_mapping(pool);
1678 /* we did our best */
1681 r = dm_thin_find_mapped_range(tc->td, begin, end, &virt_begin, &virt_end,
1682 &data_begin, &maybe_shared);
1685 * Silently fail, letting any mappings we've
1690 build_key(tc->td, PHYSICAL, data_begin, data_begin + (virt_end - virt_begin), &data_key);
1691 if (bio_detain(tc->pool, &data_key, NULL, &data_cell)) {
1692 /* contention, we'll give up with this range */
1698 * IO may still be going to the destination block. We must
1699 * quiesce before we can do the removal.
1701 m = get_next_mapping(pool);
1703 m->maybe_shared = maybe_shared;
1704 m->virt_begin = virt_begin;
1705 m->virt_end = virt_end;
1706 m->data_block = data_begin;
1707 m->cell = data_cell;
1711 * The parent bio must not complete before sub discard bios are
1712 * chained to it (see end_discard's bio_chain)!
1714 * This per-mapping bi_remaining increment is paired with
1715 * the implicit decrement that occurs via bio_endio() in
1718 bio_inc_remaining(bio);
1719 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1720 pool->process_prepared_discard(m);
1726 static void process_discard_cell_passdown(struct thin_c *tc, struct dm_bio_prison_cell *virt_cell)
1728 struct bio *bio = virt_cell->holder;
1729 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1732 * The virt_cell will only get freed once the origin bio completes.
1733 * This means it will remain locked while all the individual
1734 * passdown bios are in flight.
1736 h->cell = virt_cell;
1737 break_up_discard_bio(tc, virt_cell->key.block_begin, virt_cell->key.block_end, bio);
1740 * We complete the bio now, knowing that the bi_remaining field
1741 * will prevent completion until the sub range discards have
1747 static void process_discard_bio(struct thin_c *tc, struct bio *bio)
1749 dm_block_t begin, end;
1750 struct dm_cell_key virt_key;
1751 struct dm_bio_prison_cell *virt_cell;
1753 get_bio_block_range(tc, bio, &begin, &end);
1756 * The discard covers less than a block.
1762 build_key(tc->td, VIRTUAL, begin, end, &virt_key);
1763 if (bio_detain(tc->pool, &virt_key, bio, &virt_cell))
1765 * Potential starvation issue: We're relying on the
1766 * fs/application being well behaved, and not trying to
1767 * send IO to a region at the same time as discarding it.
1768 * If they do this persistently then it's possible this
1769 * cell will never be granted.
1773 tc->pool->process_discard_cell(tc, virt_cell);
1776 static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
1777 struct dm_cell_key *key,
1778 struct dm_thin_lookup_result *lookup_result,
1779 struct dm_bio_prison_cell *cell)
1782 dm_block_t data_block;
1783 struct pool *pool = tc->pool;
1785 r = alloc_data_block(tc, &data_block);
1788 schedule_internal_copy(tc, block, lookup_result->block,
1789 data_block, cell, bio);
1793 retry_bios_on_resume(pool, cell);
1797 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1799 cell_error(pool, cell);
1804 static void __remap_and_issue_shared_cell(void *context,
1805 struct dm_bio_prison_cell *cell)
1807 struct remap_info *info = context;
1810 while ((bio = bio_list_pop(&cell->bios))) {
1811 if (bio_data_dir(bio) == WRITE || op_is_flush(bio->bi_opf) ||
1812 bio_op(bio) == REQ_OP_DISCARD)
1813 bio_list_add(&info->defer_bios, bio);
1815 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1817 h->shared_read_entry = dm_deferred_entry_inc(info->tc->pool->shared_read_ds);
1818 inc_all_io_entry(info->tc->pool, bio);
1819 bio_list_add(&info->issue_bios, bio);
1824 static void remap_and_issue_shared_cell(struct thin_c *tc,
1825 struct dm_bio_prison_cell *cell,
1829 struct remap_info info;
1832 bio_list_init(&info.defer_bios);
1833 bio_list_init(&info.issue_bios);
1835 cell_visit_release(tc->pool, __remap_and_issue_shared_cell,
1838 while ((bio = bio_list_pop(&info.defer_bios)))
1839 thin_defer_bio(tc, bio);
1841 while ((bio = bio_list_pop(&info.issue_bios)))
1842 remap_and_issue(tc, bio, block);
1845 static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1847 struct dm_thin_lookup_result *lookup_result,
1848 struct dm_bio_prison_cell *virt_cell)
1850 struct dm_bio_prison_cell *data_cell;
1851 struct pool *pool = tc->pool;
1852 struct dm_cell_key key;
1855 * If cell is already occupied, then sharing is already in the process
1856 * of being broken so we have nothing further to do here.
1858 build_data_key(tc->td, lookup_result->block, &key);
1859 if (bio_detain(pool, &key, bio, &data_cell)) {
1860 cell_defer_no_holder(tc, virt_cell);
1864 if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) {
1865 break_sharing(tc, bio, block, &key, lookup_result, data_cell);
1866 cell_defer_no_holder(tc, virt_cell);
1868 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1870 h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
1871 inc_all_io_entry(pool, bio);
1872 remap_and_issue(tc, bio, lookup_result->block);
1874 remap_and_issue_shared_cell(tc, data_cell, lookup_result->block);
1875 remap_and_issue_shared_cell(tc, virt_cell, lookup_result->block);
1879 static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
1880 struct dm_bio_prison_cell *cell)
1883 dm_block_t data_block;
1884 struct pool *pool = tc->pool;
1887 * Remap empty bios (flushes) immediately, without provisioning.
1889 if (!bio->bi_iter.bi_size) {
1890 inc_all_io_entry(pool, bio);
1891 cell_defer_no_holder(tc, cell);
1893 remap_and_issue(tc, bio, 0);
1898 * Fill read bios with zeroes and complete them immediately.
1900 if (bio_data_dir(bio) == READ) {
1902 cell_defer_no_holder(tc, cell);
1907 r = alloc_data_block(tc, &data_block);
1911 schedule_external_copy(tc, block, data_block, cell, bio);
1913 schedule_zero(tc, block, data_block, cell, bio);
1917 retry_bios_on_resume(pool, cell);
1921 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1923 cell_error(pool, cell);
1928 static void process_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
1931 struct pool *pool = tc->pool;
1932 struct bio *bio = cell->holder;
1933 dm_block_t block = get_bio_block(tc, bio);
1934 struct dm_thin_lookup_result lookup_result;
1936 if (tc->requeue_mode) {
1937 cell_requeue(pool, cell);
1941 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1944 if (lookup_result.shared)
1945 process_shared_bio(tc, bio, block, &lookup_result, cell);
1947 inc_all_io_entry(pool, bio);
1948 remap_and_issue(tc, bio, lookup_result.block);
1949 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
1954 if (bio_data_dir(bio) == READ && tc->origin_dev) {
1955 inc_all_io_entry(pool, bio);
1956 cell_defer_no_holder(tc, cell);
1958 if (bio_end_sector(bio) <= tc->origin_size)
1959 remap_to_origin_and_issue(tc, bio);
1961 else if (bio->bi_iter.bi_sector < tc->origin_size) {
1963 bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT;
1964 remap_to_origin_and_issue(tc, bio);
1971 provision_block(tc, bio, block, cell);
1975 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1977 cell_defer_no_holder(tc, cell);
1983 static void process_bio(struct thin_c *tc, struct bio *bio)
1985 struct pool *pool = tc->pool;
1986 dm_block_t block = get_bio_block(tc, bio);
1987 struct dm_bio_prison_cell *cell;
1988 struct dm_cell_key key;
1991 * If cell is already occupied, then the block is already
1992 * being provisioned so we have nothing further to do here.
1994 build_virtual_key(tc->td, block, &key);
1995 if (bio_detain(pool, &key, bio, &cell))
1998 process_cell(tc, cell);
2001 static void __process_bio_read_only(struct thin_c *tc, struct bio *bio,
2002 struct dm_bio_prison_cell *cell)
2005 int rw = bio_data_dir(bio);
2006 dm_block_t block = get_bio_block(tc, bio);
2007 struct dm_thin_lookup_result lookup_result;
2009 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
2012 if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) {
2013 handle_unserviceable_bio(tc->pool, bio);
2015 cell_defer_no_holder(tc, cell);
2017 inc_all_io_entry(tc->pool, bio);
2018 remap_and_issue(tc, bio, lookup_result.block);
2020 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
2026 cell_defer_no_holder(tc, cell);
2028 handle_unserviceable_bio(tc->pool, bio);
2032 if (tc->origin_dev) {
2033 inc_all_io_entry(tc->pool, bio);
2034 remap_to_origin_and_issue(tc, bio);
2043 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
2046 cell_defer_no_holder(tc, cell);
2052 static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
2054 __process_bio_read_only(tc, bio, NULL);
2057 static void process_cell_read_only(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2059 __process_bio_read_only(tc, cell->holder, cell);
2062 static void process_bio_success(struct thin_c *tc, struct bio *bio)
2067 static void process_bio_fail(struct thin_c *tc, struct bio *bio)
2072 static void process_cell_success(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2074 cell_success(tc->pool, cell);
2077 static void process_cell_fail(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2079 cell_error(tc->pool, cell);
2083 * FIXME: should we also commit due to size of transaction, measured in
2086 static int need_commit_due_to_time(struct pool *pool)
2088 return !time_in_range(jiffies, pool->last_commit_jiffies,
2089 pool->last_commit_jiffies + COMMIT_PERIOD);
2092 #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node)
2093 #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook))
2095 static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio)
2097 struct rb_node **rbp, *parent;
2098 struct dm_thin_endio_hook *pbd;
2099 sector_t bi_sector = bio->bi_iter.bi_sector;
2101 rbp = &tc->sort_bio_list.rb_node;
2105 pbd = thin_pbd(parent);
2107 if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector)
2108 rbp = &(*rbp)->rb_left;
2110 rbp = &(*rbp)->rb_right;
2113 pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2114 rb_link_node(&pbd->rb_node, parent, rbp);
2115 rb_insert_color(&pbd->rb_node, &tc->sort_bio_list);
2118 static void __extract_sorted_bios(struct thin_c *tc)
2120 struct rb_node *node;
2121 struct dm_thin_endio_hook *pbd;
2124 for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) {
2125 pbd = thin_pbd(node);
2126 bio = thin_bio(pbd);
2128 bio_list_add(&tc->deferred_bio_list, bio);
2129 rb_erase(&pbd->rb_node, &tc->sort_bio_list);
2132 WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list));
2135 static void __sort_thin_deferred_bios(struct thin_c *tc)
2138 struct bio_list bios;
2140 bio_list_init(&bios);
2141 bio_list_merge(&bios, &tc->deferred_bio_list);
2142 bio_list_init(&tc->deferred_bio_list);
2144 /* Sort deferred_bio_list using rb-tree */
2145 while ((bio = bio_list_pop(&bios)))
2146 __thin_bio_rb_add(tc, bio);
2149 * Transfer the sorted bios in sort_bio_list back to
2150 * deferred_bio_list to allow lockless submission of
2153 __extract_sorted_bios(tc);
2156 static void process_thin_deferred_bios(struct thin_c *tc)
2158 struct pool *pool = tc->pool;
2160 struct bio_list bios;
2161 struct blk_plug plug;
2164 if (tc->requeue_mode) {
2165 error_thin_bio_list(tc, &tc->deferred_bio_list,
2166 BLK_STS_DM_REQUEUE);
2170 bio_list_init(&bios);
2172 spin_lock_irq(&tc->lock);
2174 if (bio_list_empty(&tc->deferred_bio_list)) {
2175 spin_unlock_irq(&tc->lock);
2179 __sort_thin_deferred_bios(tc);
2181 bio_list_merge(&bios, &tc->deferred_bio_list);
2182 bio_list_init(&tc->deferred_bio_list);
2184 spin_unlock_irq(&tc->lock);
2186 blk_start_plug(&plug);
2187 while ((bio = bio_list_pop(&bios))) {
2189 * If we've got no free new_mapping structs, and processing
2190 * this bio might require one, we pause until there are some
2191 * prepared mappings to process.
2193 if (ensure_next_mapping(pool)) {
2194 spin_lock_irq(&tc->lock);
2195 bio_list_add(&tc->deferred_bio_list, bio);
2196 bio_list_merge(&tc->deferred_bio_list, &bios);
2197 spin_unlock_irq(&tc->lock);
2201 if (bio_op(bio) == REQ_OP_DISCARD)
2202 pool->process_discard(tc, bio);
2204 pool->process_bio(tc, bio);
2206 if ((count++ & 127) == 0) {
2207 throttle_work_update(&pool->throttle);
2208 dm_pool_issue_prefetches(pool->pmd);
2211 blk_finish_plug(&plug);
2214 static int cmp_cells(const void *lhs, const void *rhs)
2216 struct dm_bio_prison_cell *lhs_cell = *((struct dm_bio_prison_cell **) lhs);
2217 struct dm_bio_prison_cell *rhs_cell = *((struct dm_bio_prison_cell **) rhs);
2219 BUG_ON(!lhs_cell->holder);
2220 BUG_ON(!rhs_cell->holder);
2222 if (lhs_cell->holder->bi_iter.bi_sector < rhs_cell->holder->bi_iter.bi_sector)
2225 if (lhs_cell->holder->bi_iter.bi_sector > rhs_cell->holder->bi_iter.bi_sector)
2231 static unsigned sort_cells(struct pool *pool, struct list_head *cells)
2234 struct dm_bio_prison_cell *cell, *tmp;
2236 list_for_each_entry_safe(cell, tmp, cells, user_list) {
2237 if (count >= CELL_SORT_ARRAY_SIZE)
2240 pool->cell_sort_array[count++] = cell;
2241 list_del(&cell->user_list);
2244 sort(pool->cell_sort_array, count, sizeof(cell), cmp_cells, NULL);
2249 static void process_thin_deferred_cells(struct thin_c *tc)
2251 struct pool *pool = tc->pool;
2252 struct list_head cells;
2253 struct dm_bio_prison_cell *cell;
2254 unsigned i, j, count;
2256 INIT_LIST_HEAD(&cells);
2258 spin_lock_irq(&tc->lock);
2259 list_splice_init(&tc->deferred_cells, &cells);
2260 spin_unlock_irq(&tc->lock);
2262 if (list_empty(&cells))
2266 count = sort_cells(tc->pool, &cells);
2268 for (i = 0; i < count; i++) {
2269 cell = pool->cell_sort_array[i];
2270 BUG_ON(!cell->holder);
2273 * If we've got no free new_mapping structs, and processing
2274 * this bio might require one, we pause until there are some
2275 * prepared mappings to process.
2277 if (ensure_next_mapping(pool)) {
2278 for (j = i; j < count; j++)
2279 list_add(&pool->cell_sort_array[j]->user_list, &cells);
2281 spin_lock_irq(&tc->lock);
2282 list_splice(&cells, &tc->deferred_cells);
2283 spin_unlock_irq(&tc->lock);
2287 if (bio_op(cell->holder) == REQ_OP_DISCARD)
2288 pool->process_discard_cell(tc, cell);
2290 pool->process_cell(tc, cell);
2292 } while (!list_empty(&cells));
2295 static void thin_get(struct thin_c *tc);
2296 static void thin_put(struct thin_c *tc);
2299 * We can't hold rcu_read_lock() around code that can block. So we
2300 * find a thin with the rcu lock held; bump a refcount; then drop
2303 static struct thin_c *get_first_thin(struct pool *pool)
2305 struct thin_c *tc = NULL;
2308 if (!list_empty(&pool->active_thins)) {
2309 tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list);
2317 static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc)
2319 struct thin_c *old_tc = tc;
2322 list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) {
2334 static void process_deferred_bios(struct pool *pool)
2337 struct bio_list bios, bio_completions;
2340 tc = get_first_thin(pool);
2342 process_thin_deferred_cells(tc);
2343 process_thin_deferred_bios(tc);
2344 tc = get_next_thin(pool, tc);
2348 * If there are any deferred flush bios, we must commit the metadata
2349 * before issuing them or signaling their completion.
2351 bio_list_init(&bios);
2352 bio_list_init(&bio_completions);
2354 spin_lock_irq(&pool->lock);
2355 bio_list_merge(&bios, &pool->deferred_flush_bios);
2356 bio_list_init(&pool->deferred_flush_bios);
2358 bio_list_merge(&bio_completions, &pool->deferred_flush_completions);
2359 bio_list_init(&pool->deferred_flush_completions);
2360 spin_unlock_irq(&pool->lock);
2362 if (bio_list_empty(&bios) && bio_list_empty(&bio_completions) &&
2363 !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool)))
2367 bio_list_merge(&bios, &bio_completions);
2369 while ((bio = bio_list_pop(&bios)))
2373 pool->last_commit_jiffies = jiffies;
2375 while ((bio = bio_list_pop(&bio_completions)))
2378 while ((bio = bio_list_pop(&bios))) {
2380 * The data device was flushed as part of metadata commit,
2381 * so complete redundant flushes immediately.
2383 if (bio->bi_opf & REQ_PREFLUSH)
2386 dm_submit_bio_remap(bio, NULL);
2390 static void do_worker(struct work_struct *ws)
2392 struct pool *pool = container_of(ws, struct pool, worker);
2394 throttle_work_start(&pool->throttle);
2395 dm_pool_issue_prefetches(pool->pmd);
2396 throttle_work_update(&pool->throttle);
2397 process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
2398 throttle_work_update(&pool->throttle);
2399 process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
2400 throttle_work_update(&pool->throttle);
2401 process_prepared(pool, &pool->prepared_discards_pt2, &pool->process_prepared_discard_pt2);
2402 throttle_work_update(&pool->throttle);
2403 process_deferred_bios(pool);
2404 throttle_work_complete(&pool->throttle);
2408 * We want to commit periodically so that not too much
2409 * unwritten data builds up.
2411 static void do_waker(struct work_struct *ws)
2413 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
2415 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
2419 * We're holding onto IO to allow userland time to react. After the
2420 * timeout either the pool will have been resized (and thus back in
2421 * PM_WRITE mode), or we degrade to PM_OUT_OF_DATA_SPACE w/ error_if_no_space.
2423 static void do_no_space_timeout(struct work_struct *ws)
2425 struct pool *pool = container_of(to_delayed_work(ws), struct pool,
2428 if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) {
2429 pool->pf.error_if_no_space = true;
2430 notify_of_pool_mode_change(pool);
2431 error_retry_list_with_code(pool, BLK_STS_NOSPC);
2435 /*----------------------------------------------------------------*/
2438 struct work_struct worker;
2439 struct completion complete;
2442 static struct pool_work *to_pool_work(struct work_struct *ws)
2444 return container_of(ws, struct pool_work, worker);
2447 static void pool_work_complete(struct pool_work *pw)
2449 complete(&pw->complete);
2452 static void pool_work_wait(struct pool_work *pw, struct pool *pool,
2453 void (*fn)(struct work_struct *))
2455 INIT_WORK_ONSTACK(&pw->worker, fn);
2456 init_completion(&pw->complete);
2457 queue_work(pool->wq, &pw->worker);
2458 wait_for_completion(&pw->complete);
2461 /*----------------------------------------------------------------*/
2463 struct noflush_work {
2464 struct pool_work pw;
2468 static struct noflush_work *to_noflush(struct work_struct *ws)
2470 return container_of(to_pool_work(ws), struct noflush_work, pw);
2473 static void do_noflush_start(struct work_struct *ws)
2475 struct noflush_work *w = to_noflush(ws);
2476 w->tc->requeue_mode = true;
2478 pool_work_complete(&w->pw);
2481 static void do_noflush_stop(struct work_struct *ws)
2483 struct noflush_work *w = to_noflush(ws);
2484 w->tc->requeue_mode = false;
2485 pool_work_complete(&w->pw);
2488 static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *))
2490 struct noflush_work w;
2493 pool_work_wait(&w.pw, tc->pool, fn);
2496 /*----------------------------------------------------------------*/
2498 static bool passdown_enabled(struct pool_c *pt)
2500 return pt->adjusted_pf.discard_passdown;
2503 static void set_discard_callbacks(struct pool *pool)
2505 struct pool_c *pt = pool->ti->private;
2507 if (passdown_enabled(pt)) {
2508 pool->process_discard_cell = process_discard_cell_passdown;
2509 pool->process_prepared_discard = process_prepared_discard_passdown_pt1;
2510 pool->process_prepared_discard_pt2 = process_prepared_discard_passdown_pt2;
2512 pool->process_discard_cell = process_discard_cell_no_passdown;
2513 pool->process_prepared_discard = process_prepared_discard_no_passdown;
2517 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode)
2519 struct pool_c *pt = pool->ti->private;
2520 bool needs_check = dm_pool_metadata_needs_check(pool->pmd);
2521 enum pool_mode old_mode = get_pool_mode(pool);
2522 unsigned long no_space_timeout = READ_ONCE(no_space_timeout_secs) * HZ;
2525 * Never allow the pool to transition to PM_WRITE mode if user
2526 * intervention is required to verify metadata and data consistency.
2528 if (new_mode == PM_WRITE && needs_check) {
2529 DMERR("%s: unable to switch pool to write mode until repaired.",
2530 dm_device_name(pool->pool_md));
2531 if (old_mode != new_mode)
2532 new_mode = old_mode;
2534 new_mode = PM_READ_ONLY;
2537 * If we were in PM_FAIL mode, rollback of metadata failed. We're
2538 * not going to recover without a thin_repair. So we never let the
2539 * pool move out of the old mode.
2541 if (old_mode == PM_FAIL)
2542 new_mode = old_mode;
2546 dm_pool_metadata_read_only(pool->pmd);
2547 pool->process_bio = process_bio_fail;
2548 pool->process_discard = process_bio_fail;
2549 pool->process_cell = process_cell_fail;
2550 pool->process_discard_cell = process_cell_fail;
2551 pool->process_prepared_mapping = process_prepared_mapping_fail;
2552 pool->process_prepared_discard = process_prepared_discard_fail;
2554 error_retry_list(pool);
2557 case PM_OUT_OF_METADATA_SPACE:
2559 dm_pool_metadata_read_only(pool->pmd);
2560 pool->process_bio = process_bio_read_only;
2561 pool->process_discard = process_bio_success;
2562 pool->process_cell = process_cell_read_only;
2563 pool->process_discard_cell = process_cell_success;
2564 pool->process_prepared_mapping = process_prepared_mapping_fail;
2565 pool->process_prepared_discard = process_prepared_discard_success;
2567 error_retry_list(pool);
2570 case PM_OUT_OF_DATA_SPACE:
2572 * Ideally we'd never hit this state; the low water mark
2573 * would trigger userland to extend the pool before we
2574 * completely run out of data space. However, many small
2575 * IOs to unprovisioned space can consume data space at an
2576 * alarming rate. Adjust your low water mark if you're
2577 * frequently seeing this mode.
2579 pool->out_of_data_space = true;
2580 pool->process_bio = process_bio_read_only;
2581 pool->process_discard = process_discard_bio;
2582 pool->process_cell = process_cell_read_only;
2583 pool->process_prepared_mapping = process_prepared_mapping;
2584 set_discard_callbacks(pool);
2586 if (!pool->pf.error_if_no_space && no_space_timeout)
2587 queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout);
2591 if (old_mode == PM_OUT_OF_DATA_SPACE)
2592 cancel_delayed_work_sync(&pool->no_space_timeout);
2593 pool->out_of_data_space = false;
2594 pool->pf.error_if_no_space = pt->requested_pf.error_if_no_space;
2595 dm_pool_metadata_read_write(pool->pmd);
2596 pool->process_bio = process_bio;
2597 pool->process_discard = process_discard_bio;
2598 pool->process_cell = process_cell;
2599 pool->process_prepared_mapping = process_prepared_mapping;
2600 set_discard_callbacks(pool);
2604 pool->pf.mode = new_mode;
2606 * The pool mode may have changed, sync it so bind_control_target()
2607 * doesn't cause an unexpected mode transition on resume.
2609 pt->adjusted_pf.mode = new_mode;
2611 if (old_mode != new_mode)
2612 notify_of_pool_mode_change(pool);
2615 static void abort_transaction(struct pool *pool)
2617 const char *dev_name = dm_device_name(pool->pool_md);
2619 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name);
2620 if (dm_pool_abort_metadata(pool->pmd)) {
2621 DMERR("%s: failed to abort metadata transaction", dev_name);
2622 set_pool_mode(pool, PM_FAIL);
2625 if (dm_pool_metadata_set_needs_check(pool->pmd)) {
2626 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name);
2627 set_pool_mode(pool, PM_FAIL);
2631 static void metadata_operation_failed(struct pool *pool, const char *op, int r)
2633 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
2634 dm_device_name(pool->pool_md), op, r);
2636 abort_transaction(pool);
2637 set_pool_mode(pool, PM_READ_ONLY);
2640 /*----------------------------------------------------------------*/
2643 * Mapping functions.
2647 * Called only while mapping a thin bio to hand it over to the workqueue.
2649 static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
2651 struct pool *pool = tc->pool;
2653 spin_lock_irq(&tc->lock);
2654 bio_list_add(&tc->deferred_bio_list, bio);
2655 spin_unlock_irq(&tc->lock);
2660 static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio)
2662 struct pool *pool = tc->pool;
2664 throttle_lock(&pool->throttle);
2665 thin_defer_bio(tc, bio);
2666 throttle_unlock(&pool->throttle);
2669 static void thin_defer_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2671 struct pool *pool = tc->pool;
2673 throttle_lock(&pool->throttle);
2674 spin_lock_irq(&tc->lock);
2675 list_add_tail(&cell->user_list, &tc->deferred_cells);
2676 spin_unlock_irq(&tc->lock);
2677 throttle_unlock(&pool->throttle);
2682 static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
2684 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2687 h->shared_read_entry = NULL;
2688 h->all_io_entry = NULL;
2689 h->overwrite_mapping = NULL;
2694 * Non-blocking function called from the thin target's map function.
2696 static int thin_bio_map(struct dm_target *ti, struct bio *bio)
2699 struct thin_c *tc = ti->private;
2700 dm_block_t block = get_bio_block(tc, bio);
2701 struct dm_thin_device *td = tc->td;
2702 struct dm_thin_lookup_result result;
2703 struct dm_bio_prison_cell *virt_cell, *data_cell;
2704 struct dm_cell_key key;
2706 thin_hook_bio(tc, bio);
2708 if (tc->requeue_mode) {
2709 bio->bi_status = BLK_STS_DM_REQUEUE;
2711 return DM_MAPIO_SUBMITTED;
2714 if (get_pool_mode(tc->pool) == PM_FAIL) {
2716 return DM_MAPIO_SUBMITTED;
2719 if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD) {
2720 thin_defer_bio_with_throttle(tc, bio);
2721 return DM_MAPIO_SUBMITTED;
2725 * We must hold the virtual cell before doing the lookup, otherwise
2726 * there's a race with discard.
2728 build_virtual_key(tc->td, block, &key);
2729 if (bio_detain(tc->pool, &key, bio, &virt_cell))
2730 return DM_MAPIO_SUBMITTED;
2732 r = dm_thin_find_block(td, block, 0, &result);
2735 * Note that we defer readahead too.
2739 if (unlikely(result.shared)) {
2741 * We have a race condition here between the
2742 * result.shared value returned by the lookup and
2743 * snapshot creation, which may cause new
2746 * To avoid this always quiesce the origin before
2747 * taking the snap. You want to do this anyway to
2748 * ensure a consistent application view
2751 * More distant ancestors are irrelevant. The
2752 * shared flag will be set in their case.
2754 thin_defer_cell(tc, virt_cell);
2755 return DM_MAPIO_SUBMITTED;
2758 build_data_key(tc->td, result.block, &key);
2759 if (bio_detain(tc->pool, &key, bio, &data_cell)) {
2760 cell_defer_no_holder(tc, virt_cell);
2761 return DM_MAPIO_SUBMITTED;
2764 inc_all_io_entry(tc->pool, bio);
2765 cell_defer_no_holder(tc, data_cell);
2766 cell_defer_no_holder(tc, virt_cell);
2768 remap(tc, bio, result.block);
2769 return DM_MAPIO_REMAPPED;
2773 thin_defer_cell(tc, virt_cell);
2774 return DM_MAPIO_SUBMITTED;
2778 * Must always call bio_io_error on failure.
2779 * dm_thin_find_block can fail with -EINVAL if the
2780 * pool is switched to fail-io mode.
2783 cell_defer_no_holder(tc, virt_cell);
2784 return DM_MAPIO_SUBMITTED;
2788 static void requeue_bios(struct pool *pool)
2793 list_for_each_entry_rcu(tc, &pool->active_thins, list) {
2794 spin_lock_irq(&tc->lock);
2795 bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list);
2796 bio_list_init(&tc->retry_on_resume_list);
2797 spin_unlock_irq(&tc->lock);
2802 /*----------------------------------------------------------------
2803 * Binding of control targets to a pool object
2804 *--------------------------------------------------------------*/
2805 static bool is_factor(sector_t block_size, uint32_t n)
2807 return !sector_div(block_size, n);
2811 * If discard_passdown was enabled verify that the data device
2812 * supports discards. Disable discard_passdown if not.
2814 static void disable_passdown_if_not_supported(struct pool_c *pt)
2816 struct pool *pool = pt->pool;
2817 struct block_device *data_bdev = pt->data_dev->bdev;
2818 struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
2819 const char *reason = NULL;
2821 if (!pt->adjusted_pf.discard_passdown)
2824 if (!bdev_max_discard_sectors(pt->data_dev->bdev))
2825 reason = "discard unsupported";
2827 else if (data_limits->max_discard_sectors < pool->sectors_per_block)
2828 reason = "max discard sectors smaller than a block";
2831 DMWARN("Data device (%pg) %s: Disabling discard passdown.", data_bdev, reason);
2832 pt->adjusted_pf.discard_passdown = false;
2836 static int bind_control_target(struct pool *pool, struct dm_target *ti)
2838 struct pool_c *pt = ti->private;
2841 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
2843 enum pool_mode old_mode = get_pool_mode(pool);
2844 enum pool_mode new_mode = pt->adjusted_pf.mode;
2847 * Don't change the pool's mode until set_pool_mode() below.
2848 * Otherwise the pool's process_* function pointers may
2849 * not match the desired pool mode.
2851 pt->adjusted_pf.mode = old_mode;
2854 pool->pf = pt->adjusted_pf;
2855 pool->low_water_blocks = pt->low_water_blocks;
2857 set_pool_mode(pool, new_mode);
2862 static void unbind_control_target(struct pool *pool, struct dm_target *ti)
2868 /*----------------------------------------------------------------
2870 *--------------------------------------------------------------*/
2871 /* Initialize pool features. */
2872 static void pool_features_init(struct pool_features *pf)
2874 pf->mode = PM_WRITE;
2875 pf->zero_new_blocks = true;
2876 pf->discard_enabled = true;
2877 pf->discard_passdown = true;
2878 pf->error_if_no_space = false;
2881 static void __pool_destroy(struct pool *pool)
2883 __pool_table_remove(pool);
2885 vfree(pool->cell_sort_array);
2886 if (dm_pool_metadata_close(pool->pmd) < 0)
2887 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
2889 dm_bio_prison_destroy(pool->prison);
2890 dm_kcopyd_client_destroy(pool->copier);
2893 destroy_workqueue(pool->wq);
2895 if (pool->next_mapping)
2896 mempool_free(pool->next_mapping, &pool->mapping_pool);
2897 mempool_exit(&pool->mapping_pool);
2898 dm_deferred_set_destroy(pool->shared_read_ds);
2899 dm_deferred_set_destroy(pool->all_io_ds);
2903 static struct kmem_cache *_new_mapping_cache;
2905 static struct pool *pool_create(struct mapped_device *pool_md,
2906 struct block_device *metadata_dev,
2907 struct block_device *data_dev,
2908 unsigned long block_size,
2909 int read_only, char **error)
2914 struct dm_pool_metadata *pmd;
2915 bool format_device = read_only ? false : true;
2917 pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
2919 *error = "Error creating metadata object";
2920 return (struct pool *)pmd;
2923 pool = kzalloc(sizeof(*pool), GFP_KERNEL);
2925 *error = "Error allocating memory for pool";
2926 err_p = ERR_PTR(-ENOMEM);
2931 pool->sectors_per_block = block_size;
2932 if (block_size & (block_size - 1))
2933 pool->sectors_per_block_shift = -1;
2935 pool->sectors_per_block_shift = __ffs(block_size);
2936 pool->low_water_blocks = 0;
2937 pool_features_init(&pool->pf);
2938 pool->prison = dm_bio_prison_create();
2939 if (!pool->prison) {
2940 *error = "Error creating pool's bio prison";
2941 err_p = ERR_PTR(-ENOMEM);
2945 pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2946 if (IS_ERR(pool->copier)) {
2947 r = PTR_ERR(pool->copier);
2948 *error = "Error creating pool's kcopyd client";
2950 goto bad_kcopyd_client;
2954 * Create singlethreaded workqueue that will service all devices
2955 * that use this metadata.
2957 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
2959 *error = "Error creating pool's workqueue";
2960 err_p = ERR_PTR(-ENOMEM);
2964 throttle_init(&pool->throttle);
2965 INIT_WORK(&pool->worker, do_worker);
2966 INIT_DELAYED_WORK(&pool->waker, do_waker);
2967 INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout);
2968 spin_lock_init(&pool->lock);
2969 bio_list_init(&pool->deferred_flush_bios);
2970 bio_list_init(&pool->deferred_flush_completions);
2971 INIT_LIST_HEAD(&pool->prepared_mappings);
2972 INIT_LIST_HEAD(&pool->prepared_discards);
2973 INIT_LIST_HEAD(&pool->prepared_discards_pt2);
2974 INIT_LIST_HEAD(&pool->active_thins);
2975 pool->low_water_triggered = false;
2976 pool->suspended = true;
2977 pool->out_of_data_space = false;
2979 pool->shared_read_ds = dm_deferred_set_create();
2980 if (!pool->shared_read_ds) {
2981 *error = "Error creating pool's shared read deferred set";
2982 err_p = ERR_PTR(-ENOMEM);
2983 goto bad_shared_read_ds;
2986 pool->all_io_ds = dm_deferred_set_create();
2987 if (!pool->all_io_ds) {
2988 *error = "Error creating pool's all io deferred set";
2989 err_p = ERR_PTR(-ENOMEM);
2993 pool->next_mapping = NULL;
2994 r = mempool_init_slab_pool(&pool->mapping_pool, MAPPING_POOL_SIZE,
2995 _new_mapping_cache);
2997 *error = "Error creating pool's mapping mempool";
2999 goto bad_mapping_pool;
3002 pool->cell_sort_array =
3003 vmalloc(array_size(CELL_SORT_ARRAY_SIZE,
3004 sizeof(*pool->cell_sort_array)));
3005 if (!pool->cell_sort_array) {
3006 *error = "Error allocating cell sort array";
3007 err_p = ERR_PTR(-ENOMEM);
3008 goto bad_sort_array;
3011 pool->ref_count = 1;
3012 pool->last_commit_jiffies = jiffies;
3013 pool->pool_md = pool_md;
3014 pool->md_dev = metadata_dev;
3015 pool->data_dev = data_dev;
3016 __pool_table_insert(pool);
3021 mempool_exit(&pool->mapping_pool);
3023 dm_deferred_set_destroy(pool->all_io_ds);
3025 dm_deferred_set_destroy(pool->shared_read_ds);
3027 destroy_workqueue(pool->wq);
3029 dm_kcopyd_client_destroy(pool->copier);
3031 dm_bio_prison_destroy(pool->prison);
3035 if (dm_pool_metadata_close(pmd))
3036 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
3041 static void __pool_inc(struct pool *pool)
3043 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
3047 static void __pool_dec(struct pool *pool)
3049 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
3050 BUG_ON(!pool->ref_count);
3051 if (!--pool->ref_count)
3052 __pool_destroy(pool);
3055 static struct pool *__pool_find(struct mapped_device *pool_md,
3056 struct block_device *metadata_dev,
3057 struct block_device *data_dev,
3058 unsigned long block_size, int read_only,
3059 char **error, int *created)
3061 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
3064 if (pool->pool_md != pool_md) {
3065 *error = "metadata device already in use by a pool";
3066 return ERR_PTR(-EBUSY);
3068 if (pool->data_dev != data_dev) {
3069 *error = "data device already in use by a pool";
3070 return ERR_PTR(-EBUSY);
3075 pool = __pool_table_lookup(pool_md);
3077 if (pool->md_dev != metadata_dev || pool->data_dev != data_dev) {
3078 *error = "different pool cannot replace a pool";
3079 return ERR_PTR(-EINVAL);
3084 pool = pool_create(pool_md, metadata_dev, data_dev, block_size, read_only, error);
3092 /*----------------------------------------------------------------
3093 * Pool target methods
3094 *--------------------------------------------------------------*/
3095 static void pool_dtr(struct dm_target *ti)
3097 struct pool_c *pt = ti->private;
3099 mutex_lock(&dm_thin_pool_table.mutex);
3101 unbind_control_target(pt->pool, ti);
3102 __pool_dec(pt->pool);
3103 dm_put_device(ti, pt->metadata_dev);
3104 dm_put_device(ti, pt->data_dev);
3107 mutex_unlock(&dm_thin_pool_table.mutex);
3110 static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
3111 struct dm_target *ti)
3115 const char *arg_name;
3117 static const struct dm_arg _args[] = {
3118 {0, 4, "Invalid number of pool feature arguments"},
3122 * No feature arguments supplied.
3127 r = dm_read_arg_group(_args, as, &argc, &ti->error);
3131 while (argc && !r) {
3132 arg_name = dm_shift_arg(as);
3135 if (!strcasecmp(arg_name, "skip_block_zeroing"))
3136 pf->zero_new_blocks = false;
3138 else if (!strcasecmp(arg_name, "ignore_discard"))
3139 pf->discard_enabled = false;
3141 else if (!strcasecmp(arg_name, "no_discard_passdown"))
3142 pf->discard_passdown = false;
3144 else if (!strcasecmp(arg_name, "read_only"))
3145 pf->mode = PM_READ_ONLY;
3147 else if (!strcasecmp(arg_name, "error_if_no_space"))
3148 pf->error_if_no_space = true;
3151 ti->error = "Unrecognised pool feature requested";
3160 static void metadata_low_callback(void *context)
3162 struct pool *pool = context;
3164 DMWARN("%s: reached low water mark for metadata device: sending event.",
3165 dm_device_name(pool->pool_md));
3167 dm_table_event(pool->ti->table);
3171 * We need to flush the data device **before** committing the metadata.
3173 * This ensures that the data blocks of any newly inserted mappings are
3174 * properly written to non-volatile storage and won't be lost in case of a
3177 * Failure to do so can result in data corruption in the case of internal or
3178 * external snapshots and in the case of newly provisioned blocks, when block
3179 * zeroing is enabled.
3181 static int metadata_pre_commit_callback(void *context)
3183 struct pool *pool = context;
3185 return blkdev_issue_flush(pool->data_dev);
3188 static sector_t get_dev_size(struct block_device *bdev)
3190 return bdev_nr_sectors(bdev);
3193 static void warn_if_metadata_device_too_big(struct block_device *bdev)
3195 sector_t metadata_dev_size = get_dev_size(bdev);
3197 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING)
3198 DMWARN("Metadata device %pg is larger than %u sectors: excess space will not be used.",
3199 bdev, THIN_METADATA_MAX_SECTORS);
3202 static sector_t get_metadata_dev_size(struct block_device *bdev)
3204 sector_t metadata_dev_size = get_dev_size(bdev);
3206 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS)
3207 metadata_dev_size = THIN_METADATA_MAX_SECTORS;
3209 return metadata_dev_size;
3212 static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
3214 sector_t metadata_dev_size = get_metadata_dev_size(bdev);
3216 sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE);
3218 return metadata_dev_size;
3222 * When a metadata threshold is crossed a dm event is triggered, and
3223 * userland should respond by growing the metadata device. We could let
3224 * userland set the threshold, like we do with the data threshold, but I'm
3225 * not sure they know enough to do this well.
3227 static dm_block_t calc_metadata_threshold(struct pool_c *pt)
3230 * 4M is ample for all ops with the possible exception of thin
3231 * device deletion which is harmless if it fails (just retry the
3232 * delete after you've grown the device).
3234 dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
3235 return min((dm_block_t)1024ULL /* 4M */, quarter);
3239 * thin-pool <metadata dev> <data dev>
3240 * <data block size (sectors)>
3241 * <low water mark (blocks)>
3242 * [<#feature args> [<arg>]*]
3244 * Optional feature arguments are:
3245 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
3246 * ignore_discard: disable discard
3247 * no_discard_passdown: don't pass discards down to the data device
3248 * read_only: Don't allow any changes to be made to the pool metadata.
3249 * error_if_no_space: error IOs, instead of queueing, if no space.
3251 static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
3253 int r, pool_created = 0;
3256 struct pool_features pf;
3257 struct dm_arg_set as;
3258 struct dm_dev *data_dev;
3259 unsigned long block_size;
3260 dm_block_t low_water_blocks;
3261 struct dm_dev *metadata_dev;
3262 fmode_t metadata_mode;
3265 * FIXME Remove validation from scope of lock.
3267 mutex_lock(&dm_thin_pool_table.mutex);
3270 ti->error = "Invalid argument count";
3278 /* make sure metadata and data are different devices */
3279 if (!strcmp(argv[0], argv[1])) {
3280 ti->error = "Error setting metadata or data device";
3286 * Set default pool features.
3288 pool_features_init(&pf);
3290 dm_consume_args(&as, 4);
3291 r = parse_pool_features(&as, &pf, ti);
3295 metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
3296 r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
3298 ti->error = "Error opening metadata block device";
3301 warn_if_metadata_device_too_big(metadata_dev->bdev);
3303 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
3305 ti->error = "Error getting data device";
3309 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
3310 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
3311 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
3312 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
3313 ti->error = "Invalid block size";
3318 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
3319 ti->error = "Invalid low water mark";
3324 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
3330 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, data_dev->bdev,
3331 block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
3338 * 'pool_created' reflects whether this is the first table load.
3339 * Top level discard support is not allowed to be changed after
3340 * initial load. This would require a pool reload to trigger thin
3343 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
3344 ti->error = "Discard support cannot be disabled once enabled";
3346 goto out_flags_changed;
3351 pt->metadata_dev = metadata_dev;
3352 pt->data_dev = data_dev;
3353 pt->low_water_blocks = low_water_blocks;
3354 pt->adjusted_pf = pt->requested_pf = pf;
3355 ti->num_flush_bios = 1;
3358 * Only need to enable discards if the pool should pass
3359 * them down to the data device. The thin device's discard
3360 * processing will cause mappings to be removed from the btree.
3362 if (pf.discard_enabled && pf.discard_passdown) {
3363 ti->num_discard_bios = 1;
3366 * Setting 'discards_supported' circumvents the normal
3367 * stacking of discard limits (this keeps the pool and
3368 * thin devices' discard limits consistent).
3370 ti->discards_supported = true;
3374 r = dm_pool_register_metadata_threshold(pt->pool->pmd,
3375 calc_metadata_threshold(pt),
3376 metadata_low_callback,
3379 ti->error = "Error registering metadata threshold";
3380 goto out_flags_changed;
3383 dm_pool_register_pre_commit_callback(pool->pmd,
3384 metadata_pre_commit_callback, pool);
3386 mutex_unlock(&dm_thin_pool_table.mutex);
3395 dm_put_device(ti, data_dev);
3397 dm_put_device(ti, metadata_dev);
3399 mutex_unlock(&dm_thin_pool_table.mutex);
3404 static int pool_map(struct dm_target *ti, struct bio *bio)
3407 struct pool_c *pt = ti->private;
3408 struct pool *pool = pt->pool;
3411 * As this is a singleton target, ti->begin is always zero.
3413 spin_lock_irq(&pool->lock);
3414 bio_set_dev(bio, pt->data_dev->bdev);
3415 r = DM_MAPIO_REMAPPED;
3416 spin_unlock_irq(&pool->lock);
3421 static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
3424 struct pool_c *pt = ti->private;
3425 struct pool *pool = pt->pool;
3426 sector_t data_size = ti->len;
3427 dm_block_t sb_data_size;
3429 *need_commit = false;
3431 (void) sector_div(data_size, pool->sectors_per_block);
3433 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
3435 DMERR("%s: failed to retrieve data device size",
3436 dm_device_name(pool->pool_md));
3440 if (data_size < sb_data_size) {
3441 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
3442 dm_device_name(pool->pool_md),
3443 (unsigned long long)data_size, sb_data_size);
3446 } else if (data_size > sb_data_size) {
3447 if (dm_pool_metadata_needs_check(pool->pmd)) {
3448 DMERR("%s: unable to grow the data device until repaired.",
3449 dm_device_name(pool->pool_md));
3454 DMINFO("%s: growing the data device from %llu to %llu blocks",
3455 dm_device_name(pool->pool_md),
3456 sb_data_size, (unsigned long long)data_size);
3457 r = dm_pool_resize_data_dev(pool->pmd, data_size);
3459 metadata_operation_failed(pool, "dm_pool_resize_data_dev", r);
3463 *need_commit = true;
3469 static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
3472 struct pool_c *pt = ti->private;
3473 struct pool *pool = pt->pool;
3474 dm_block_t metadata_dev_size, sb_metadata_dev_size;
3476 *need_commit = false;
3478 metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
3480 r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
3482 DMERR("%s: failed to retrieve metadata device size",
3483 dm_device_name(pool->pool_md));
3487 if (metadata_dev_size < sb_metadata_dev_size) {
3488 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
3489 dm_device_name(pool->pool_md),
3490 metadata_dev_size, sb_metadata_dev_size);
3493 } else if (metadata_dev_size > sb_metadata_dev_size) {
3494 if (dm_pool_metadata_needs_check(pool->pmd)) {
3495 DMERR("%s: unable to grow the metadata device until repaired.",
3496 dm_device_name(pool->pool_md));
3500 warn_if_metadata_device_too_big(pool->md_dev);
3501 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
3502 dm_device_name(pool->pool_md),
3503 sb_metadata_dev_size, metadata_dev_size);
3505 if (get_pool_mode(pool) == PM_OUT_OF_METADATA_SPACE)
3506 set_pool_mode(pool, PM_WRITE);
3508 r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
3510 metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r);
3514 *need_commit = true;
3521 * Retrieves the number of blocks of the data device from
3522 * the superblock and compares it to the actual device size,
3523 * thus resizing the data device in case it has grown.
3525 * This both copes with opening preallocated data devices in the ctr
3526 * being followed by a resume
3528 * calling the resume method individually after userspace has
3529 * grown the data device in reaction to a table event.
3531 static int pool_preresume(struct dm_target *ti)
3534 bool need_commit1, need_commit2;
3535 struct pool_c *pt = ti->private;
3536 struct pool *pool = pt->pool;
3539 * Take control of the pool object.
3541 r = bind_control_target(pool, ti);
3545 r = maybe_resize_data_dev(ti, &need_commit1);
3549 r = maybe_resize_metadata_dev(ti, &need_commit2);
3553 if (need_commit1 || need_commit2)
3554 (void) commit(pool);
3557 * When a thin-pool is PM_FAIL, it cannot be rebuilt if
3558 * bio is in deferred list. Therefore need to return 0
3559 * to allow pool_resume() to flush IO.
3561 if (r && get_pool_mode(pool) == PM_FAIL)
3567 static void pool_suspend_active_thins(struct pool *pool)
3571 /* Suspend all active thin devices */
3572 tc = get_first_thin(pool);
3574 dm_internal_suspend_noflush(tc->thin_md);
3575 tc = get_next_thin(pool, tc);
3579 static void pool_resume_active_thins(struct pool *pool)
3583 /* Resume all active thin devices */
3584 tc = get_first_thin(pool);
3586 dm_internal_resume(tc->thin_md);
3587 tc = get_next_thin(pool, tc);
3591 static void pool_resume(struct dm_target *ti)
3593 struct pool_c *pt = ti->private;
3594 struct pool *pool = pt->pool;
3597 * Must requeue active_thins' bios and then resume
3598 * active_thins _before_ clearing 'suspend' flag.
3601 pool_resume_active_thins(pool);
3603 spin_lock_irq(&pool->lock);
3604 pool->low_water_triggered = false;
3605 pool->suspended = false;
3606 spin_unlock_irq(&pool->lock);
3608 do_waker(&pool->waker.work);
3611 static void pool_presuspend(struct dm_target *ti)
3613 struct pool_c *pt = ti->private;
3614 struct pool *pool = pt->pool;
3616 spin_lock_irq(&pool->lock);
3617 pool->suspended = true;
3618 spin_unlock_irq(&pool->lock);
3620 pool_suspend_active_thins(pool);
3623 static void pool_presuspend_undo(struct dm_target *ti)
3625 struct pool_c *pt = ti->private;
3626 struct pool *pool = pt->pool;
3628 pool_resume_active_thins(pool);
3630 spin_lock_irq(&pool->lock);
3631 pool->suspended = false;
3632 spin_unlock_irq(&pool->lock);
3635 static void pool_postsuspend(struct dm_target *ti)
3637 struct pool_c *pt = ti->private;
3638 struct pool *pool = pt->pool;
3640 cancel_delayed_work_sync(&pool->waker);
3641 cancel_delayed_work_sync(&pool->no_space_timeout);
3642 flush_workqueue(pool->wq);
3643 (void) commit(pool);
3646 static int check_arg_count(unsigned argc, unsigned args_required)
3648 if (argc != args_required) {
3649 DMWARN("Message received with %u arguments instead of %u.",
3650 argc, args_required);
3657 static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
3659 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
3660 *dev_id <= MAX_DEV_ID)
3664 DMWARN("Message received with invalid device id: %s", arg);
3669 static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
3674 r = check_arg_count(argc, 2);
3678 r = read_dev_id(argv[1], &dev_id, 1);
3682 r = dm_pool_create_thin(pool->pmd, dev_id);
3684 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
3692 static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3695 dm_thin_id origin_dev_id;
3698 r = check_arg_count(argc, 3);
3702 r = read_dev_id(argv[1], &dev_id, 1);
3706 r = read_dev_id(argv[2], &origin_dev_id, 1);
3710 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
3712 DMWARN("Creation of new snapshot %s of device %s failed.",
3720 static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
3725 r = check_arg_count(argc, 2);
3729 r = read_dev_id(argv[1], &dev_id, 1);
3733 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
3735 DMWARN("Deletion of thin device %s failed.", argv[1]);
3740 static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
3742 dm_thin_id old_id, new_id;
3745 r = check_arg_count(argc, 3);
3749 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
3750 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
3754 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
3755 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
3759 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
3761 DMWARN("Failed to change transaction id from %s to %s.",
3769 static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3773 r = check_arg_count(argc, 1);
3777 (void) commit(pool);
3779 r = dm_pool_reserve_metadata_snap(pool->pmd);
3781 DMWARN("reserve_metadata_snap message failed.");
3786 static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3790 r = check_arg_count(argc, 1);
3794 r = dm_pool_release_metadata_snap(pool->pmd);
3796 DMWARN("release_metadata_snap message failed.");
3802 * Messages supported:
3803 * create_thin <dev_id>
3804 * create_snap <dev_id> <origin_id>
3806 * set_transaction_id <current_trans_id> <new_trans_id>
3807 * reserve_metadata_snap
3808 * release_metadata_snap
3810 static int pool_message(struct dm_target *ti, unsigned argc, char **argv,
3811 char *result, unsigned maxlen)
3814 struct pool_c *pt = ti->private;
3815 struct pool *pool = pt->pool;
3817 if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE) {
3818 DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode",
3819 dm_device_name(pool->pool_md));
3823 if (!strcasecmp(argv[0], "create_thin"))
3824 r = process_create_thin_mesg(argc, argv, pool);
3826 else if (!strcasecmp(argv[0], "create_snap"))
3827 r = process_create_snap_mesg(argc, argv, pool);
3829 else if (!strcasecmp(argv[0], "delete"))
3830 r = process_delete_mesg(argc, argv, pool);
3832 else if (!strcasecmp(argv[0], "set_transaction_id"))
3833 r = process_set_transaction_id_mesg(argc, argv, pool);
3835 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
3836 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
3838 else if (!strcasecmp(argv[0], "release_metadata_snap"))
3839 r = process_release_metadata_snap_mesg(argc, argv, pool);
3842 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
3845 (void) commit(pool);
3850 static void emit_flags(struct pool_features *pf, char *result,
3851 unsigned sz, unsigned maxlen)
3853 unsigned count = !pf->zero_new_blocks + !pf->discard_enabled +
3854 !pf->discard_passdown + (pf->mode == PM_READ_ONLY) +
3855 pf->error_if_no_space;
3856 DMEMIT("%u ", count);
3858 if (!pf->zero_new_blocks)
3859 DMEMIT("skip_block_zeroing ");
3861 if (!pf->discard_enabled)
3862 DMEMIT("ignore_discard ");
3864 if (!pf->discard_passdown)
3865 DMEMIT("no_discard_passdown ");
3867 if (pf->mode == PM_READ_ONLY)
3868 DMEMIT("read_only ");
3870 if (pf->error_if_no_space)
3871 DMEMIT("error_if_no_space ");
3876 * <transaction id> <used metadata sectors>/<total metadata sectors>
3877 * <used data sectors>/<total data sectors> <held metadata root>
3878 * <pool mode> <discard config> <no space config> <needs_check>
3880 static void pool_status(struct dm_target *ti, status_type_t type,
3881 unsigned status_flags, char *result, unsigned maxlen)
3885 uint64_t transaction_id;
3886 dm_block_t nr_free_blocks_data;
3887 dm_block_t nr_free_blocks_metadata;
3888 dm_block_t nr_blocks_data;
3889 dm_block_t nr_blocks_metadata;
3890 dm_block_t held_root;
3891 enum pool_mode mode;
3892 char buf[BDEVNAME_SIZE];
3893 char buf2[BDEVNAME_SIZE];
3894 struct pool_c *pt = ti->private;
3895 struct pool *pool = pt->pool;
3898 case STATUSTYPE_INFO:
3899 if (get_pool_mode(pool) == PM_FAIL) {
3904 /* Commit to ensure statistics aren't out-of-date */
3905 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
3906 (void) commit(pool);
3908 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
3910 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
3911 dm_device_name(pool->pool_md), r);
3915 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
3917 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
3918 dm_device_name(pool->pool_md), r);
3922 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
3924 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
3925 dm_device_name(pool->pool_md), r);
3929 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
3931 DMERR("%s: dm_pool_get_free_block_count returned %d",
3932 dm_device_name(pool->pool_md), r);
3936 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
3938 DMERR("%s: dm_pool_get_data_dev_size returned %d",
3939 dm_device_name(pool->pool_md), r);
3943 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
3945 DMERR("%s: dm_pool_get_metadata_snap returned %d",
3946 dm_device_name(pool->pool_md), r);
3950 DMEMIT("%llu %llu/%llu %llu/%llu ",
3951 (unsigned long long)transaction_id,
3952 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
3953 (unsigned long long)nr_blocks_metadata,
3954 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
3955 (unsigned long long)nr_blocks_data);
3958 DMEMIT("%llu ", held_root);
3962 mode = get_pool_mode(pool);
3963 if (mode == PM_OUT_OF_DATA_SPACE)
3964 DMEMIT("out_of_data_space ");
3965 else if (is_read_only_pool_mode(mode))
3970 if (!pool->pf.discard_enabled)
3971 DMEMIT("ignore_discard ");
3972 else if (pool->pf.discard_passdown)
3973 DMEMIT("discard_passdown ");
3975 DMEMIT("no_discard_passdown ");
3977 if (pool->pf.error_if_no_space)
3978 DMEMIT("error_if_no_space ");
3980 DMEMIT("queue_if_no_space ");
3982 if (dm_pool_metadata_needs_check(pool->pmd))
3983 DMEMIT("needs_check ");
3987 DMEMIT("%llu ", (unsigned long long)calc_metadata_threshold(pt));
3991 case STATUSTYPE_TABLE:
3992 DMEMIT("%s %s %lu %llu ",
3993 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
3994 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
3995 (unsigned long)pool->sectors_per_block,
3996 (unsigned long long)pt->low_water_blocks);
3997 emit_flags(&pt->requested_pf, result, sz, maxlen);
4000 case STATUSTYPE_IMA:
4010 static int pool_iterate_devices(struct dm_target *ti,
4011 iterate_devices_callout_fn fn, void *data)
4013 struct pool_c *pt = ti->private;
4015 return fn(ti, pt->data_dev, 0, ti->len, data);
4018 static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
4020 struct pool_c *pt = ti->private;
4021 struct pool *pool = pt->pool;
4022 sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
4025 * If max_sectors is smaller than pool->sectors_per_block adjust it
4026 * to the highest possible power-of-2 factor of pool->sectors_per_block.
4027 * This is especially beneficial when the pool's data device is a RAID
4028 * device that has a full stripe width that matches pool->sectors_per_block
4029 * -- because even though partial RAID stripe-sized IOs will be issued to a
4030 * single RAID stripe; when aggregated they will end on a full RAID stripe
4031 * boundary.. which avoids additional partial RAID stripe writes cascading
4033 if (limits->max_sectors < pool->sectors_per_block) {
4034 while (!is_factor(pool->sectors_per_block, limits->max_sectors)) {
4035 if ((limits->max_sectors & (limits->max_sectors - 1)) == 0)
4036 limits->max_sectors--;
4037 limits->max_sectors = rounddown_pow_of_two(limits->max_sectors);
4042 * If the system-determined stacked limits are compatible with the
4043 * pool's blocksize (io_opt is a factor) do not override them.
4045 if (io_opt_sectors < pool->sectors_per_block ||
4046 !is_factor(io_opt_sectors, pool->sectors_per_block)) {
4047 if (is_factor(pool->sectors_per_block, limits->max_sectors))
4048 blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT);
4050 blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT);
4051 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
4055 * pt->adjusted_pf is a staging area for the actual features to use.
4056 * They get transferred to the live pool in bind_control_target()
4057 * called from pool_preresume().
4059 if (!pt->adjusted_pf.discard_enabled) {
4061 * Must explicitly disallow stacking discard limits otherwise the
4062 * block layer will stack them if pool's data device has support.
4064 limits->discard_granularity = 0;
4068 disable_passdown_if_not_supported(pt);
4071 * The pool uses the same discard limits as the underlying data
4072 * device. DM core has already set this up.
4076 static struct target_type pool_target = {
4077 .name = "thin-pool",
4078 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
4079 DM_TARGET_IMMUTABLE,
4080 .version = {1, 22, 0},
4081 .module = THIS_MODULE,
4085 .presuspend = pool_presuspend,
4086 .presuspend_undo = pool_presuspend_undo,
4087 .postsuspend = pool_postsuspend,
4088 .preresume = pool_preresume,
4089 .resume = pool_resume,
4090 .message = pool_message,
4091 .status = pool_status,
4092 .iterate_devices = pool_iterate_devices,
4093 .io_hints = pool_io_hints,
4096 /*----------------------------------------------------------------
4097 * Thin target methods
4098 *--------------------------------------------------------------*/
4099 static void thin_get(struct thin_c *tc)
4101 refcount_inc(&tc->refcount);
4104 static void thin_put(struct thin_c *tc)
4106 if (refcount_dec_and_test(&tc->refcount))
4107 complete(&tc->can_destroy);
4110 static void thin_dtr(struct dm_target *ti)
4112 struct thin_c *tc = ti->private;
4114 spin_lock_irq(&tc->pool->lock);
4115 list_del_rcu(&tc->list);
4116 spin_unlock_irq(&tc->pool->lock);
4120 wait_for_completion(&tc->can_destroy);
4122 mutex_lock(&dm_thin_pool_table.mutex);
4124 __pool_dec(tc->pool);
4125 dm_pool_close_thin_device(tc->td);
4126 dm_put_device(ti, tc->pool_dev);
4128 dm_put_device(ti, tc->origin_dev);
4131 mutex_unlock(&dm_thin_pool_table.mutex);
4135 * Thin target parameters:
4137 * <pool_dev> <dev_id> [origin_dev]
4139 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
4140 * dev_id: the internal device identifier
4141 * origin_dev: a device external to the pool that should act as the origin
4143 * If the pool device has discards disabled, they get disabled for the thin
4146 static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
4150 struct dm_dev *pool_dev, *origin_dev;
4151 struct mapped_device *pool_md;
4153 mutex_lock(&dm_thin_pool_table.mutex);
4155 if (argc != 2 && argc != 3) {
4156 ti->error = "Invalid argument count";
4161 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
4163 ti->error = "Out of memory";
4167 tc->thin_md = dm_table_get_md(ti->table);
4168 spin_lock_init(&tc->lock);
4169 INIT_LIST_HEAD(&tc->deferred_cells);
4170 bio_list_init(&tc->deferred_bio_list);
4171 bio_list_init(&tc->retry_on_resume_list);
4172 tc->sort_bio_list = RB_ROOT;
4175 if (!strcmp(argv[0], argv[2])) {
4176 ti->error = "Error setting origin device";
4178 goto bad_origin_dev;
4181 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
4183 ti->error = "Error opening origin device";
4184 goto bad_origin_dev;
4186 tc->origin_dev = origin_dev;
4189 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
4191 ti->error = "Error opening pool device";
4194 tc->pool_dev = pool_dev;
4196 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
4197 ti->error = "Invalid device id";
4202 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
4204 ti->error = "Couldn't get pool mapped device";
4209 tc->pool = __pool_table_lookup(pool_md);
4211 ti->error = "Couldn't find pool object";
4213 goto bad_pool_lookup;
4215 __pool_inc(tc->pool);
4217 if (get_pool_mode(tc->pool) == PM_FAIL) {
4218 ti->error = "Couldn't open thin device, Pool is in fail mode";
4223 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
4225 ti->error = "Couldn't open thin internal device";
4229 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
4233 ti->num_flush_bios = 1;
4234 ti->flush_supported = true;
4235 ti->accounts_remapped_io = true;
4236 ti->per_io_data_size = sizeof(struct dm_thin_endio_hook);
4238 /* In case the pool supports discards, pass them on. */
4239 if (tc->pool->pf.discard_enabled) {
4240 ti->discards_supported = true;
4241 ti->num_discard_bios = 1;
4244 mutex_unlock(&dm_thin_pool_table.mutex);
4246 spin_lock_irq(&tc->pool->lock);
4247 if (tc->pool->suspended) {
4248 spin_unlock_irq(&tc->pool->lock);
4249 mutex_lock(&dm_thin_pool_table.mutex); /* reacquire for __pool_dec */
4250 ti->error = "Unable to activate thin device while pool is suspended";
4254 refcount_set(&tc->refcount, 1);
4255 init_completion(&tc->can_destroy);
4256 list_add_tail_rcu(&tc->list, &tc->pool->active_thins);
4257 spin_unlock_irq(&tc->pool->lock);
4259 * This synchronize_rcu() call is needed here otherwise we risk a
4260 * wake_worker() call finding no bios to process (because the newly
4261 * added tc isn't yet visible). So this reduces latency since we
4262 * aren't then dependent on the periodic commit to wake_worker().
4271 dm_pool_close_thin_device(tc->td);
4273 __pool_dec(tc->pool);
4277 dm_put_device(ti, tc->pool_dev);
4280 dm_put_device(ti, tc->origin_dev);
4284 mutex_unlock(&dm_thin_pool_table.mutex);
4289 static int thin_map(struct dm_target *ti, struct bio *bio)
4291 bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
4293 return thin_bio_map(ti, bio);
4296 static int thin_endio(struct dm_target *ti, struct bio *bio,
4299 unsigned long flags;
4300 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
4301 struct list_head work;
4302 struct dm_thin_new_mapping *m, *tmp;
4303 struct pool *pool = h->tc->pool;
4305 if (h->shared_read_entry) {
4306 INIT_LIST_HEAD(&work);
4307 dm_deferred_entry_dec(h->shared_read_entry, &work);
4309 spin_lock_irqsave(&pool->lock, flags);
4310 list_for_each_entry_safe(m, tmp, &work, list) {
4312 __complete_mapping_preparation(m);
4314 spin_unlock_irqrestore(&pool->lock, flags);
4317 if (h->all_io_entry) {
4318 INIT_LIST_HEAD(&work);
4319 dm_deferred_entry_dec(h->all_io_entry, &work);
4320 if (!list_empty(&work)) {
4321 spin_lock_irqsave(&pool->lock, flags);
4322 list_for_each_entry_safe(m, tmp, &work, list)
4323 list_add_tail(&m->list, &pool->prepared_discards);
4324 spin_unlock_irqrestore(&pool->lock, flags);
4330 cell_defer_no_holder(h->tc, h->cell);
4332 return DM_ENDIO_DONE;
4335 static void thin_presuspend(struct dm_target *ti)
4337 struct thin_c *tc = ti->private;
4339 if (dm_noflush_suspending(ti))
4340 noflush_work(tc, do_noflush_start);
4343 static void thin_postsuspend(struct dm_target *ti)
4345 struct thin_c *tc = ti->private;
4348 * The dm_noflush_suspending flag has been cleared by now, so
4349 * unfortunately we must always run this.
4351 noflush_work(tc, do_noflush_stop);
4354 static int thin_preresume(struct dm_target *ti)
4356 struct thin_c *tc = ti->private;
4359 tc->origin_size = get_dev_size(tc->origin_dev->bdev);
4365 * <nr mapped sectors> <highest mapped sector>
4367 static void thin_status(struct dm_target *ti, status_type_t type,
4368 unsigned status_flags, char *result, unsigned maxlen)
4372 dm_block_t mapped, highest;
4373 char buf[BDEVNAME_SIZE];
4374 struct thin_c *tc = ti->private;
4376 if (get_pool_mode(tc->pool) == PM_FAIL) {
4385 case STATUSTYPE_INFO:
4386 r = dm_thin_get_mapped_count(tc->td, &mapped);
4388 DMERR("dm_thin_get_mapped_count returned %d", r);
4392 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
4394 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
4398 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
4400 DMEMIT("%llu", ((highest + 1) *
4401 tc->pool->sectors_per_block) - 1);
4406 case STATUSTYPE_TABLE:
4408 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
4409 (unsigned long) tc->dev_id);
4411 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
4414 case STATUSTYPE_IMA:
4426 static int thin_iterate_devices(struct dm_target *ti,
4427 iterate_devices_callout_fn fn, void *data)
4430 struct thin_c *tc = ti->private;
4431 struct pool *pool = tc->pool;
4434 * We can't call dm_pool_get_data_dev_size() since that blocks. So
4435 * we follow a more convoluted path through to the pool's target.
4438 return 0; /* nothing is bound */
4440 blocks = pool->ti->len;
4441 (void) sector_div(blocks, pool->sectors_per_block);
4443 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
4448 static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits)
4450 struct thin_c *tc = ti->private;
4451 struct pool *pool = tc->pool;
4453 if (!pool->pf.discard_enabled)
4456 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
4457 limits->max_discard_sectors = 2048 * 1024 * 16; /* 16G */
4460 static struct target_type thin_target = {
4462 .version = {1, 22, 0},
4463 .module = THIS_MODULE,
4467 .end_io = thin_endio,
4468 .preresume = thin_preresume,
4469 .presuspend = thin_presuspend,
4470 .postsuspend = thin_postsuspend,
4471 .status = thin_status,
4472 .iterate_devices = thin_iterate_devices,
4473 .io_hints = thin_io_hints,
4476 /*----------------------------------------------------------------*/
4478 static int __init dm_thin_init(void)
4484 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
4485 if (!_new_mapping_cache)
4488 r = dm_register_target(&thin_target);
4490 goto bad_new_mapping_cache;
4492 r = dm_register_target(&pool_target);
4494 goto bad_thin_target;
4499 dm_unregister_target(&thin_target);
4500 bad_new_mapping_cache:
4501 kmem_cache_destroy(_new_mapping_cache);
4506 static void dm_thin_exit(void)
4508 dm_unregister_target(&thin_target);
4509 dm_unregister_target(&pool_target);
4511 kmem_cache_destroy(_new_mapping_cache);
4516 module_init(dm_thin_init);
4517 module_exit(dm_thin_exit);
4519 module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR);
4520 MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds");
4522 MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
4523 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
4524 MODULE_LICENSE("GPL");