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.h"
11 #include <linux/device-mapper.h>
12 #include <linux/dm-io.h>
13 #include <linux/dm-kcopyd.h>
14 #include <linux/jiffies.h>
15 #include <linux/log2.h>
16 #include <linux/list.h>
17 #include <linux/rculist.h>
18 #include <linux/init.h>
19 #include <linux/module.h>
20 #include <linux/slab.h>
21 #include <linux/vmalloc.h>
22 #include <linux/sort.h>
23 #include <linux/rbtree.h>
25 #define DM_MSG_PREFIX "thin"
30 #define ENDIO_HOOK_POOL_SIZE 1024
31 #define MAPPING_POOL_SIZE 1024
32 #define COMMIT_PERIOD HZ
33 #define NO_SPACE_TIMEOUT_SECS 60
35 static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS;
37 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
38 "A percentage of time allocated for copy on write");
41 * The block size of the device holding pool data must be
42 * between 64KB and 1GB.
44 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
45 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
48 * Device id is restricted to 24 bits.
50 #define MAX_DEV_ID ((1 << 24) - 1)
53 * How do we handle breaking sharing of data blocks?
54 * =================================================
56 * We use a standard copy-on-write btree to store the mappings for the
57 * devices (note I'm talking about copy-on-write of the metadata here, not
58 * the data). When you take an internal snapshot you clone the root node
59 * of the origin btree. After this there is no concept of an origin or a
60 * snapshot. They are just two device trees that happen to point to the
63 * When we get a write in we decide if it's to a shared data block using
64 * some timestamp magic. If it is, we have to break sharing.
66 * Let's say we write to a shared block in what was the origin. The
69 * i) plug io further to this physical block. (see bio_prison code).
71 * ii) quiesce any read io to that shared data block. Obviously
72 * including all devices that share this block. (see dm_deferred_set code)
74 * iii) copy the data block to a newly allocate block. This step can be
75 * missed out if the io covers the block. (schedule_copy).
77 * iv) insert the new mapping into the origin's btree
78 * (process_prepared_mapping). This act of inserting breaks some
79 * sharing of btree nodes between the two devices. Breaking sharing only
80 * effects the btree of that specific device. Btrees for the other
81 * devices that share the block never change. The btree for the origin
82 * device as it was after the last commit is untouched, ie. we're using
83 * persistent data structures in the functional programming sense.
85 * v) unplug io to this physical block, including the io that triggered
86 * the breaking of sharing.
88 * Steps (ii) and (iii) occur in parallel.
90 * The metadata _doesn't_ need to be committed before the io continues. We
91 * get away with this because the io is always written to a _new_ block.
92 * If there's a crash, then:
94 * - The origin mapping will point to the old origin block (the shared
95 * one). This will contain the data as it was before the io that triggered
96 * the breaking of sharing came in.
98 * - The snap mapping still points to the old block. As it would after
101 * The downside of this scheme is the timestamp magic isn't perfect, and
102 * will continue to think that data block in the snapshot device is shared
103 * even after the write to the origin has broken sharing. I suspect data
104 * blocks will typically be shared by many different devices, so we're
105 * breaking sharing n + 1 times, rather than n, where n is the number of
106 * devices that reference this data block. At the moment I think the
107 * benefits far, far outweigh the disadvantages.
110 /*----------------------------------------------------------------*/
120 static void build_key(struct dm_thin_device *td, enum lock_space ls,
121 dm_block_t b, dm_block_t e, struct dm_cell_key *key)
123 key->virtual = (ls == VIRTUAL);
124 key->dev = dm_thin_dev_id(td);
125 key->block_begin = b;
129 static void build_data_key(struct dm_thin_device *td, dm_block_t b,
130 struct dm_cell_key *key)
132 build_key(td, PHYSICAL, b, b + 1llu, key);
135 static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
136 struct dm_cell_key *key)
138 build_key(td, VIRTUAL, b, b + 1llu, key);
141 /*----------------------------------------------------------------*/
143 #define THROTTLE_THRESHOLD (1 * HZ)
146 struct rw_semaphore lock;
147 unsigned long threshold;
148 bool throttle_applied;
151 static void throttle_init(struct throttle *t)
153 init_rwsem(&t->lock);
154 t->throttle_applied = false;
157 static void throttle_work_start(struct throttle *t)
159 t->threshold = jiffies + THROTTLE_THRESHOLD;
162 static void throttle_work_update(struct throttle *t)
164 if (!t->throttle_applied && jiffies > t->threshold) {
165 down_write(&t->lock);
166 t->throttle_applied = true;
170 static void throttle_work_complete(struct throttle *t)
172 if (t->throttle_applied) {
173 t->throttle_applied = false;
178 static void throttle_lock(struct throttle *t)
183 static void throttle_unlock(struct throttle *t)
188 /*----------------------------------------------------------------*/
191 * A pool device ties together a metadata device and a data device. It
192 * also provides the interface for creating and destroying internal
195 struct dm_thin_new_mapping;
198 * The pool runs in 4 modes. Ordered in degraded order for comparisons.
201 PM_WRITE, /* metadata may be changed */
202 PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */
205 * Like READ_ONLY, except may switch back to WRITE on metadata resize. Reported as READ_ONLY.
207 PM_OUT_OF_METADATA_SPACE,
208 PM_READ_ONLY, /* metadata may not be changed */
210 PM_FAIL, /* all I/O fails */
213 struct pool_features {
216 bool zero_new_blocks:1;
217 bool discard_enabled:1;
218 bool discard_passdown:1;
219 bool error_if_no_space:1;
223 typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
224 typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell);
225 typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);
227 #define CELL_SORT_ARRAY_SIZE 8192
230 struct list_head list;
231 struct dm_target *ti; /* Only set if a pool target is bound */
233 struct mapped_device *pool_md;
234 struct block_device *md_dev;
235 struct dm_pool_metadata *pmd;
237 dm_block_t low_water_blocks;
238 uint32_t sectors_per_block;
239 int sectors_per_block_shift;
241 struct pool_features pf;
242 bool low_water_triggered:1; /* A dm event has been sent */
244 bool out_of_data_space:1;
246 struct dm_bio_prison *prison;
247 struct dm_kcopyd_client *copier;
249 struct workqueue_struct *wq;
250 struct throttle throttle;
251 struct work_struct worker;
252 struct delayed_work waker;
253 struct delayed_work no_space_timeout;
255 unsigned long last_commit_jiffies;
259 struct bio_list deferred_flush_bios;
260 struct bio_list deferred_flush_completions;
261 struct list_head prepared_mappings;
262 struct list_head prepared_discards;
263 struct list_head prepared_discards_pt2;
264 struct list_head active_thins;
266 struct dm_deferred_set *shared_read_ds;
267 struct dm_deferred_set *all_io_ds;
269 struct dm_thin_new_mapping *next_mapping;
270 mempool_t *mapping_pool;
272 process_bio_fn process_bio;
273 process_bio_fn process_discard;
275 process_cell_fn process_cell;
276 process_cell_fn process_discard_cell;
278 process_mapping_fn process_prepared_mapping;
279 process_mapping_fn process_prepared_discard;
280 process_mapping_fn process_prepared_discard_pt2;
282 struct dm_bio_prison_cell **cell_sort_array;
285 static enum pool_mode get_pool_mode(struct pool *pool);
286 static void metadata_operation_failed(struct pool *pool, const char *op, int r);
289 * Target context for a pool.
292 struct dm_target *ti;
294 struct dm_dev *data_dev;
295 struct dm_dev *metadata_dev;
296 struct dm_target_callbacks callbacks;
298 dm_block_t low_water_blocks;
299 struct pool_features requested_pf; /* Features requested during table load */
300 struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */
304 * Target context for a thin.
307 struct list_head list;
308 struct dm_dev *pool_dev;
309 struct dm_dev *origin_dev;
310 sector_t origin_size;
314 struct dm_thin_device *td;
315 struct mapped_device *thin_md;
319 struct list_head deferred_cells;
320 struct bio_list deferred_bio_list;
321 struct bio_list retry_on_resume_list;
322 struct rb_root sort_bio_list; /* sorted list of deferred bios */
325 * Ensures the thin is not destroyed until the worker has finished
326 * iterating the active_thins list.
329 struct completion can_destroy;
332 /*----------------------------------------------------------------*/
334 static bool block_size_is_power_of_two(struct pool *pool)
336 return pool->sectors_per_block_shift >= 0;
339 static sector_t block_to_sectors(struct pool *pool, dm_block_t b)
341 return block_size_is_power_of_two(pool) ?
342 (b << pool->sectors_per_block_shift) :
343 (b * pool->sectors_per_block);
346 /*----------------------------------------------------------------*/
350 struct blk_plug plug;
351 struct bio *parent_bio;
355 static void begin_discard(struct discard_op *op, struct thin_c *tc, struct bio *parent)
360 blk_start_plug(&op->plug);
361 op->parent_bio = parent;
365 static int issue_discard(struct discard_op *op, dm_block_t data_b, dm_block_t data_e)
367 struct thin_c *tc = op->tc;
368 sector_t s = block_to_sectors(tc->pool, data_b);
369 sector_t len = block_to_sectors(tc->pool, data_e - data_b);
371 return __blkdev_issue_discard(tc->pool_dev->bdev, s, len,
372 GFP_NOWAIT, 0, &op->bio);
375 static void end_discard(struct discard_op *op, int r)
379 * Even if one of the calls to issue_discard failed, we
380 * need to wait for the chain to complete.
382 bio_chain(op->bio, op->parent_bio);
383 bio_set_op_attrs(op->bio, REQ_OP_DISCARD, 0);
387 blk_finish_plug(&op->plug);
390 * Even if r is set, there could be sub discards in flight that we
393 if (r && !op->parent_bio->bi_error)
394 op->parent_bio->bi_error = r;
395 bio_endio(op->parent_bio);
398 /*----------------------------------------------------------------*/
401 * wake_worker() is used when new work is queued and when pool_resume is
402 * ready to continue deferred IO processing.
404 static void wake_worker(struct pool *pool)
406 queue_work(pool->wq, &pool->worker);
409 /*----------------------------------------------------------------*/
411 static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio,
412 struct dm_bio_prison_cell **cell_result)
415 struct dm_bio_prison_cell *cell_prealloc;
418 * Allocate a cell from the prison's mempool.
419 * This might block but it can't fail.
421 cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO);
423 r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result);
426 * We reused an old cell; we can get rid of
429 dm_bio_prison_free_cell(pool->prison, cell_prealloc);
434 static void cell_release(struct pool *pool,
435 struct dm_bio_prison_cell *cell,
436 struct bio_list *bios)
438 dm_cell_release(pool->prison, cell, bios);
439 dm_bio_prison_free_cell(pool->prison, cell);
442 static void cell_visit_release(struct pool *pool,
443 void (*fn)(void *, struct dm_bio_prison_cell *),
445 struct dm_bio_prison_cell *cell)
447 dm_cell_visit_release(pool->prison, fn, context, cell);
448 dm_bio_prison_free_cell(pool->prison, cell);
451 static void cell_release_no_holder(struct pool *pool,
452 struct dm_bio_prison_cell *cell,
453 struct bio_list *bios)
455 dm_cell_release_no_holder(pool->prison, cell, bios);
456 dm_bio_prison_free_cell(pool->prison, cell);
459 static void cell_error_with_code(struct pool *pool,
460 struct dm_bio_prison_cell *cell, int error_code)
462 dm_cell_error(pool->prison, cell, error_code);
463 dm_bio_prison_free_cell(pool->prison, cell);
466 static int get_pool_io_error_code(struct pool *pool)
468 return pool->out_of_data_space ? -ENOSPC : -EIO;
471 static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell)
473 int error = get_pool_io_error_code(pool);
475 cell_error_with_code(pool, cell, error);
478 static void cell_success(struct pool *pool, struct dm_bio_prison_cell *cell)
480 cell_error_with_code(pool, cell, 0);
483 static void cell_requeue(struct pool *pool, struct dm_bio_prison_cell *cell)
485 cell_error_with_code(pool, cell, DM_ENDIO_REQUEUE);
488 /*----------------------------------------------------------------*/
491 * A global list of pools that uses a struct mapped_device as a key.
493 static struct dm_thin_pool_table {
495 struct list_head pools;
496 } dm_thin_pool_table;
498 static void pool_table_init(void)
500 mutex_init(&dm_thin_pool_table.mutex);
501 INIT_LIST_HEAD(&dm_thin_pool_table.pools);
504 static void __pool_table_insert(struct pool *pool)
506 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
507 list_add(&pool->list, &dm_thin_pool_table.pools);
510 static void __pool_table_remove(struct pool *pool)
512 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
513 list_del(&pool->list);
516 static struct pool *__pool_table_lookup(struct mapped_device *md)
518 struct pool *pool = NULL, *tmp;
520 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
522 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
523 if (tmp->pool_md == md) {
532 static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
534 struct pool *pool = NULL, *tmp;
536 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
538 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
539 if (tmp->md_dev == md_dev) {
548 /*----------------------------------------------------------------*/
550 struct dm_thin_endio_hook {
552 struct dm_deferred_entry *shared_read_entry;
553 struct dm_deferred_entry *all_io_entry;
554 struct dm_thin_new_mapping *overwrite_mapping;
555 struct rb_node rb_node;
556 struct dm_bio_prison_cell *cell;
559 static void __merge_bio_list(struct bio_list *bios, struct bio_list *master)
561 bio_list_merge(bios, master);
562 bio_list_init(master);
565 static void error_bio_list(struct bio_list *bios, int error)
569 while ((bio = bio_list_pop(bios))) {
570 bio->bi_error = error;
575 static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master, int error)
577 struct bio_list bios;
580 bio_list_init(&bios);
582 spin_lock_irqsave(&tc->lock, flags);
583 __merge_bio_list(&bios, master);
584 spin_unlock_irqrestore(&tc->lock, flags);
586 error_bio_list(&bios, error);
589 static void requeue_deferred_cells(struct thin_c *tc)
591 struct pool *pool = tc->pool;
593 struct list_head cells;
594 struct dm_bio_prison_cell *cell, *tmp;
596 INIT_LIST_HEAD(&cells);
598 spin_lock_irqsave(&tc->lock, flags);
599 list_splice_init(&tc->deferred_cells, &cells);
600 spin_unlock_irqrestore(&tc->lock, flags);
602 list_for_each_entry_safe(cell, tmp, &cells, user_list)
603 cell_requeue(pool, cell);
606 static void requeue_io(struct thin_c *tc)
608 struct bio_list bios;
611 bio_list_init(&bios);
613 spin_lock_irqsave(&tc->lock, flags);
614 __merge_bio_list(&bios, &tc->deferred_bio_list);
615 __merge_bio_list(&bios, &tc->retry_on_resume_list);
616 spin_unlock_irqrestore(&tc->lock, flags);
618 error_bio_list(&bios, DM_ENDIO_REQUEUE);
619 requeue_deferred_cells(tc);
622 static void error_retry_list_with_code(struct pool *pool, int error)
627 list_for_each_entry_rcu(tc, &pool->active_thins, list)
628 error_thin_bio_list(tc, &tc->retry_on_resume_list, error);
632 static void error_retry_list(struct pool *pool)
634 int error = get_pool_io_error_code(pool);
636 error_retry_list_with_code(pool, error);
640 * This section of code contains the logic for processing a thin device's IO.
641 * Much of the code depends on pool object resources (lists, workqueues, etc)
642 * but most is exclusively called from the thin target rather than the thin-pool
646 static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
648 struct pool *pool = tc->pool;
649 sector_t block_nr = bio->bi_iter.bi_sector;
651 if (block_size_is_power_of_two(pool))
652 block_nr >>= pool->sectors_per_block_shift;
654 (void) sector_div(block_nr, pool->sectors_per_block);
660 * Returns the _complete_ blocks that this bio covers.
662 static void get_bio_block_range(struct thin_c *tc, struct bio *bio,
663 dm_block_t *begin, dm_block_t *end)
665 struct pool *pool = tc->pool;
666 sector_t b = bio->bi_iter.bi_sector;
667 sector_t e = b + (bio->bi_iter.bi_size >> SECTOR_SHIFT);
669 b += pool->sectors_per_block - 1ull; /* so we round up */
671 if (block_size_is_power_of_two(pool)) {
672 b >>= pool->sectors_per_block_shift;
673 e >>= pool->sectors_per_block_shift;
675 (void) sector_div(b, pool->sectors_per_block);
676 (void) sector_div(e, pool->sectors_per_block);
680 /* Can happen if the bio is within a single block. */
687 static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
689 struct pool *pool = tc->pool;
690 sector_t bi_sector = bio->bi_iter.bi_sector;
692 bio->bi_bdev = tc->pool_dev->bdev;
693 if (block_size_is_power_of_two(pool))
694 bio->bi_iter.bi_sector =
695 (block << pool->sectors_per_block_shift) |
696 (bi_sector & (pool->sectors_per_block - 1));
698 bio->bi_iter.bi_sector = (block * pool->sectors_per_block) +
699 sector_div(bi_sector, pool->sectors_per_block);
702 static void remap_to_origin(struct thin_c *tc, struct bio *bio)
704 bio->bi_bdev = tc->origin_dev->bdev;
707 static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
709 return (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA)) &&
710 dm_thin_changed_this_transaction(tc->td);
713 static void inc_all_io_entry(struct pool *pool, struct bio *bio)
715 struct dm_thin_endio_hook *h;
717 if (bio_op(bio) == REQ_OP_DISCARD)
720 h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
721 h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
724 static void issue(struct thin_c *tc, struct bio *bio)
726 struct pool *pool = tc->pool;
729 if (!bio_triggers_commit(tc, bio)) {
730 generic_make_request(bio);
735 * Complete bio with an error if earlier I/O caused changes to
736 * the metadata that can't be committed e.g, due to I/O errors
737 * on the metadata device.
739 if (dm_thin_aborted_changes(tc->td)) {
745 * Batch together any bios that trigger commits and then issue a
746 * single commit for them in process_deferred_bios().
748 spin_lock_irqsave(&pool->lock, flags);
749 bio_list_add(&pool->deferred_flush_bios, bio);
750 spin_unlock_irqrestore(&pool->lock, flags);
753 static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
755 remap_to_origin(tc, bio);
759 static void remap_and_issue(struct thin_c *tc, struct bio *bio,
762 remap(tc, bio, block);
766 /*----------------------------------------------------------------*/
769 * Bio endio functions.
771 struct dm_thin_new_mapping {
772 struct list_head list;
778 * Track quiescing, copying and zeroing preparation actions. When this
779 * counter hits zero the block is prepared and can be inserted into the
782 atomic_t prepare_actions;
786 dm_block_t virt_begin, virt_end;
787 dm_block_t data_block;
788 struct dm_bio_prison_cell *cell;
791 * If the bio covers the whole area of a block then we can avoid
792 * zeroing or copying. Instead this bio is hooked. The bio will
793 * still be in the cell, so care has to be taken to avoid issuing
797 bio_end_io_t *saved_bi_end_io;
800 static void __complete_mapping_preparation(struct dm_thin_new_mapping *m)
802 struct pool *pool = m->tc->pool;
804 if (atomic_dec_and_test(&m->prepare_actions)) {
805 list_add_tail(&m->list, &pool->prepared_mappings);
810 static void complete_mapping_preparation(struct dm_thin_new_mapping *m)
813 struct pool *pool = m->tc->pool;
815 spin_lock_irqsave(&pool->lock, flags);
816 __complete_mapping_preparation(m);
817 spin_unlock_irqrestore(&pool->lock, flags);
820 static void copy_complete(int read_err, unsigned long write_err, void *context)
822 struct dm_thin_new_mapping *m = context;
824 m->err = read_err || write_err ? -EIO : 0;
825 complete_mapping_preparation(m);
828 static void overwrite_endio(struct bio *bio)
830 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
831 struct dm_thin_new_mapping *m = h->overwrite_mapping;
833 bio->bi_end_io = m->saved_bi_end_io;
835 m->err = bio->bi_error;
836 complete_mapping_preparation(m);
839 /*----------------------------------------------------------------*/
846 * Prepared mapping jobs.
850 * This sends the bios in the cell, except the original holder, back
851 * to the deferred_bios list.
853 static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
855 struct pool *pool = tc->pool;
858 spin_lock_irqsave(&tc->lock, flags);
859 cell_release_no_holder(pool, cell, &tc->deferred_bio_list);
860 spin_unlock_irqrestore(&tc->lock, flags);
865 static void thin_defer_bio(struct thin_c *tc, struct bio *bio);
869 struct bio_list defer_bios;
870 struct bio_list issue_bios;
873 static void __inc_remap_and_issue_cell(void *context,
874 struct dm_bio_prison_cell *cell)
876 struct remap_info *info = context;
879 while ((bio = bio_list_pop(&cell->bios))) {
880 if (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA) ||
881 bio_op(bio) == REQ_OP_DISCARD)
882 bio_list_add(&info->defer_bios, bio);
884 inc_all_io_entry(info->tc->pool, bio);
887 * We can't issue the bios with the bio prison lock
888 * held, so we add them to a list to issue on
889 * return from this function.
891 bio_list_add(&info->issue_bios, bio);
896 static void inc_remap_and_issue_cell(struct thin_c *tc,
897 struct dm_bio_prison_cell *cell,
901 struct remap_info info;
904 bio_list_init(&info.defer_bios);
905 bio_list_init(&info.issue_bios);
908 * We have to be careful to inc any bios we're about to issue
909 * before the cell is released, and avoid a race with new bios
910 * being added to the cell.
912 cell_visit_release(tc->pool, __inc_remap_and_issue_cell,
915 while ((bio = bio_list_pop(&info.defer_bios)))
916 thin_defer_bio(tc, bio);
918 while ((bio = bio_list_pop(&info.issue_bios)))
919 remap_and_issue(info.tc, bio, block);
922 static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
924 cell_error(m->tc->pool, m->cell);
926 mempool_free(m, m->tc->pool->mapping_pool);
929 static void complete_overwrite_bio(struct thin_c *tc, struct bio *bio)
931 struct pool *pool = tc->pool;
935 * If the bio has the REQ_FUA flag set we must commit the metadata
936 * before signaling its completion.
938 if (!bio_triggers_commit(tc, bio)) {
944 * Complete bio with an error if earlier I/O caused changes to the
945 * metadata that can't be committed, e.g, due to I/O errors on the
948 if (dm_thin_aborted_changes(tc->td)) {
954 * Batch together any bios that trigger commits and then issue a
955 * single commit for them in process_deferred_bios().
957 spin_lock_irqsave(&pool->lock, flags);
958 bio_list_add(&pool->deferred_flush_completions, bio);
959 spin_unlock_irqrestore(&pool->lock, flags);
962 static void process_prepared_mapping(struct dm_thin_new_mapping *m)
964 struct thin_c *tc = m->tc;
965 struct pool *pool = tc->pool;
966 struct bio *bio = m->bio;
970 cell_error(pool, m->cell);
975 * Commit the prepared block into the mapping btree.
976 * Any I/O for this block arriving after this point will get
977 * remapped to it directly.
979 r = dm_thin_insert_block(tc->td, m->virt_begin, m->data_block);
981 metadata_operation_failed(pool, "dm_thin_insert_block", r);
982 cell_error(pool, m->cell);
987 * Release any bios held while the block was being provisioned.
988 * If we are processing a write bio that completely covers the block,
989 * we already processed it so can ignore it now when processing
990 * the bios in the cell.
993 inc_remap_and_issue_cell(tc, m->cell, m->data_block);
994 complete_overwrite_bio(tc, bio);
996 inc_all_io_entry(tc->pool, m->cell->holder);
997 remap_and_issue(tc, m->cell->holder, m->data_block);
998 inc_remap_and_issue_cell(tc, m->cell, m->data_block);
1003 mempool_free(m, pool->mapping_pool);
1006 /*----------------------------------------------------------------*/
1008 static void free_discard_mapping(struct dm_thin_new_mapping *m)
1010 struct thin_c *tc = m->tc;
1012 cell_defer_no_holder(tc, m->cell);
1013 mempool_free(m, tc->pool->mapping_pool);
1016 static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
1018 bio_io_error(m->bio);
1019 free_discard_mapping(m);
1022 static void process_prepared_discard_success(struct dm_thin_new_mapping *m)
1025 free_discard_mapping(m);
1028 static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m)
1031 struct thin_c *tc = m->tc;
1033 r = dm_thin_remove_range(tc->td, m->cell->key.block_begin, m->cell->key.block_end);
1035 metadata_operation_failed(tc->pool, "dm_thin_remove_range", r);
1036 bio_io_error(m->bio);
1040 cell_defer_no_holder(tc, m->cell);
1041 mempool_free(m, tc->pool->mapping_pool);
1044 /*----------------------------------------------------------------*/
1046 static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m,
1047 struct bio *discard_parent)
1050 * We've already unmapped this range of blocks, but before we
1051 * passdown we have to check that these blocks are now unused.
1055 struct thin_c *tc = m->tc;
1056 struct pool *pool = tc->pool;
1057 dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin;
1058 struct discard_op op;
1060 begin_discard(&op, tc, discard_parent);
1062 /* find start of unmapped run */
1063 for (; b < end; b++) {
1064 r = dm_pool_block_is_shared(pool->pmd, b, &shared);
1075 /* find end of run */
1076 for (e = b + 1; e != end; e++) {
1077 r = dm_pool_block_is_shared(pool->pmd, e, &shared);
1085 r = issue_discard(&op, b, e);
1092 end_discard(&op, r);
1095 static void queue_passdown_pt2(struct dm_thin_new_mapping *m)
1097 unsigned long flags;
1098 struct pool *pool = m->tc->pool;
1100 spin_lock_irqsave(&pool->lock, flags);
1101 list_add_tail(&m->list, &pool->prepared_discards_pt2);
1102 spin_unlock_irqrestore(&pool->lock, flags);
1106 static void passdown_endio(struct bio *bio)
1109 * It doesn't matter if the passdown discard failed, we still want
1110 * to unmap (we ignore err).
1112 queue_passdown_pt2(bio->bi_private);
1116 static void process_prepared_discard_passdown_pt1(struct dm_thin_new_mapping *m)
1119 struct thin_c *tc = m->tc;
1120 struct pool *pool = tc->pool;
1121 struct bio *discard_parent;
1122 dm_block_t data_end = m->data_block + (m->virt_end - m->virt_begin);
1125 * Only this thread allocates blocks, so we can be sure that the
1126 * newly unmapped blocks will not be allocated before the end of
1129 r = dm_thin_remove_range(tc->td, m->virt_begin, m->virt_end);
1131 metadata_operation_failed(pool, "dm_thin_remove_range", r);
1132 bio_io_error(m->bio);
1133 cell_defer_no_holder(tc, m->cell);
1134 mempool_free(m, pool->mapping_pool);
1139 * Increment the unmapped blocks. This prevents a race between the
1140 * passdown io and reallocation of freed blocks.
1142 r = dm_pool_inc_data_range(pool->pmd, m->data_block, data_end);
1144 metadata_operation_failed(pool, "dm_pool_inc_data_range", r);
1145 bio_io_error(m->bio);
1146 cell_defer_no_holder(tc, m->cell);
1147 mempool_free(m, pool->mapping_pool);
1151 discard_parent = bio_alloc(GFP_NOIO, 1);
1152 if (!discard_parent) {
1153 DMWARN("%s: unable to allocate top level discard bio for passdown. Skipping passdown.",
1154 dm_device_name(tc->pool->pool_md));
1155 queue_passdown_pt2(m);
1158 discard_parent->bi_end_io = passdown_endio;
1159 discard_parent->bi_private = m;
1161 if (m->maybe_shared)
1162 passdown_double_checking_shared_status(m, discard_parent);
1164 struct discard_op op;
1166 begin_discard(&op, tc, discard_parent);
1167 r = issue_discard(&op, m->data_block, data_end);
1168 end_discard(&op, r);
1173 static void process_prepared_discard_passdown_pt2(struct dm_thin_new_mapping *m)
1176 struct thin_c *tc = m->tc;
1177 struct pool *pool = tc->pool;
1180 * The passdown has completed, so now we can decrement all those
1183 r = dm_pool_dec_data_range(pool->pmd, m->data_block,
1184 m->data_block + (m->virt_end - m->virt_begin));
1186 metadata_operation_failed(pool, "dm_pool_dec_data_range", r);
1187 bio_io_error(m->bio);
1191 cell_defer_no_holder(tc, m->cell);
1192 mempool_free(m, pool->mapping_pool);
1195 static void process_prepared(struct pool *pool, struct list_head *head,
1196 process_mapping_fn *fn)
1198 unsigned long flags;
1199 struct list_head maps;
1200 struct dm_thin_new_mapping *m, *tmp;
1202 INIT_LIST_HEAD(&maps);
1203 spin_lock_irqsave(&pool->lock, flags);
1204 list_splice_init(head, &maps);
1205 spin_unlock_irqrestore(&pool->lock, flags);
1207 list_for_each_entry_safe(m, tmp, &maps, list)
1212 * Deferred bio jobs.
1214 static int io_overlaps_block(struct pool *pool, struct bio *bio)
1216 return bio->bi_iter.bi_size ==
1217 (pool->sectors_per_block << SECTOR_SHIFT);
1220 static int io_overwrites_block(struct pool *pool, struct bio *bio)
1222 return (bio_data_dir(bio) == WRITE) &&
1223 io_overlaps_block(pool, bio);
1226 static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
1229 *save = bio->bi_end_io;
1230 bio->bi_end_io = fn;
1233 static int ensure_next_mapping(struct pool *pool)
1235 if (pool->next_mapping)
1238 pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);
1240 return pool->next_mapping ? 0 : -ENOMEM;
1243 static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
1245 struct dm_thin_new_mapping *m = pool->next_mapping;
1247 BUG_ON(!pool->next_mapping);
1249 memset(m, 0, sizeof(struct dm_thin_new_mapping));
1250 INIT_LIST_HEAD(&m->list);
1253 pool->next_mapping = NULL;
1258 static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m,
1259 sector_t begin, sector_t end)
1262 struct dm_io_region to;
1264 to.bdev = tc->pool_dev->bdev;
1266 to.count = end - begin;
1268 r = dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m);
1270 DMERR_LIMIT("dm_kcopyd_zero() failed");
1271 copy_complete(1, 1, m);
1275 static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio,
1276 dm_block_t data_begin,
1277 struct dm_thin_new_mapping *m)
1279 struct pool *pool = tc->pool;
1280 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1282 h->overwrite_mapping = m;
1284 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
1285 inc_all_io_entry(pool, bio);
1286 remap_and_issue(tc, bio, data_begin);
1290 * A partial copy also needs to zero the uncopied region.
1292 static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
1293 struct dm_dev *origin, dm_block_t data_origin,
1294 dm_block_t data_dest,
1295 struct dm_bio_prison_cell *cell, struct bio *bio,
1299 struct pool *pool = tc->pool;
1300 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1303 m->virt_begin = virt_block;
1304 m->virt_end = virt_block + 1u;
1305 m->data_block = data_dest;
1309 * quiesce action + copy action + an extra reference held for the
1310 * duration of this function (we may need to inc later for a
1313 atomic_set(&m->prepare_actions, 3);
1315 if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
1316 complete_mapping_preparation(m); /* already quiesced */
1319 * IO to pool_dev remaps to the pool target's data_dev.
1321 * If the whole block of data is being overwritten, we can issue the
1322 * bio immediately. Otherwise we use kcopyd to clone the data first.
1324 if (io_overwrites_block(pool, bio))
1325 remap_and_issue_overwrite(tc, bio, data_dest, m);
1327 struct dm_io_region from, to;
1329 from.bdev = origin->bdev;
1330 from.sector = data_origin * pool->sectors_per_block;
1333 to.bdev = tc->pool_dev->bdev;
1334 to.sector = data_dest * pool->sectors_per_block;
1337 r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
1338 0, copy_complete, m);
1340 DMERR_LIMIT("dm_kcopyd_copy() failed");
1341 copy_complete(1, 1, m);
1344 * We allow the zero to be issued, to simplify the
1345 * error path. Otherwise we'd need to start
1346 * worrying about decrementing the prepare_actions
1352 * Do we need to zero a tail region?
1354 if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) {
1355 atomic_inc(&m->prepare_actions);
1357 data_dest * pool->sectors_per_block + len,
1358 (data_dest + 1) * pool->sectors_per_block);
1362 complete_mapping_preparation(m); /* drop our ref */
1365 static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
1366 dm_block_t data_origin, dm_block_t data_dest,
1367 struct dm_bio_prison_cell *cell, struct bio *bio)
1369 schedule_copy(tc, virt_block, tc->pool_dev,
1370 data_origin, data_dest, cell, bio,
1371 tc->pool->sectors_per_block);
1374 static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
1375 dm_block_t data_block, struct dm_bio_prison_cell *cell,
1378 struct pool *pool = tc->pool;
1379 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1381 atomic_set(&m->prepare_actions, 1); /* no need to quiesce */
1383 m->virt_begin = virt_block;
1384 m->virt_end = virt_block + 1u;
1385 m->data_block = data_block;
1389 * If the whole block of data is being overwritten or we are not
1390 * zeroing pre-existing data, we can issue the bio immediately.
1391 * Otherwise we use kcopyd to zero the data first.
1393 if (pool->pf.zero_new_blocks) {
1394 if (io_overwrites_block(pool, bio))
1395 remap_and_issue_overwrite(tc, bio, data_block, m);
1397 ll_zero(tc, m, data_block * pool->sectors_per_block,
1398 (data_block + 1) * pool->sectors_per_block);
1400 process_prepared_mapping(m);
1403 static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
1404 dm_block_t data_dest,
1405 struct dm_bio_prison_cell *cell, struct bio *bio)
1407 struct pool *pool = tc->pool;
1408 sector_t virt_block_begin = virt_block * pool->sectors_per_block;
1409 sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block;
1411 if (virt_block_end <= tc->origin_size)
1412 schedule_copy(tc, virt_block, tc->origin_dev,
1413 virt_block, data_dest, cell, bio,
1414 pool->sectors_per_block);
1416 else if (virt_block_begin < tc->origin_size)
1417 schedule_copy(tc, virt_block, tc->origin_dev,
1418 virt_block, data_dest, cell, bio,
1419 tc->origin_size - virt_block_begin);
1422 schedule_zero(tc, virt_block, data_dest, cell, bio);
1425 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode);
1427 static void requeue_bios(struct pool *pool);
1429 static bool is_read_only_pool_mode(enum pool_mode mode)
1431 return (mode == PM_OUT_OF_METADATA_SPACE || mode == PM_READ_ONLY);
1434 static bool is_read_only(struct pool *pool)
1436 return is_read_only_pool_mode(get_pool_mode(pool));
1439 static void check_for_metadata_space(struct pool *pool)
1442 const char *ooms_reason = NULL;
1445 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free);
1447 ooms_reason = "Could not get free metadata blocks";
1449 ooms_reason = "No free metadata blocks";
1451 if (ooms_reason && !is_read_only(pool)) {
1452 DMERR("%s", ooms_reason);
1453 set_pool_mode(pool, PM_OUT_OF_METADATA_SPACE);
1457 static void check_for_data_space(struct pool *pool)
1462 if (get_pool_mode(pool) != PM_OUT_OF_DATA_SPACE)
1465 r = dm_pool_get_free_block_count(pool->pmd, &nr_free);
1470 set_pool_mode(pool, PM_WRITE);
1476 * A non-zero return indicates read_only or fail_io mode.
1477 * Many callers don't care about the return value.
1479 static int commit(struct pool *pool)
1483 if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE)
1486 r = dm_pool_commit_metadata(pool->pmd);
1488 metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
1490 check_for_metadata_space(pool);
1491 check_for_data_space(pool);
1497 static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks)
1499 unsigned long flags;
1501 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
1502 DMWARN("%s: reached low water mark for data device: sending event.",
1503 dm_device_name(pool->pool_md));
1504 spin_lock_irqsave(&pool->lock, flags);
1505 pool->low_water_triggered = true;
1506 spin_unlock_irqrestore(&pool->lock, flags);
1507 dm_table_event(pool->ti->table);
1511 static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
1514 dm_block_t free_blocks;
1515 struct pool *pool = tc->pool;
1517 if (WARN_ON(get_pool_mode(pool) != PM_WRITE))
1520 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1522 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1526 check_low_water_mark(pool, free_blocks);
1530 * Try to commit to see if that will free up some
1537 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1539 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1544 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1549 r = dm_pool_alloc_data_block(pool->pmd, result);
1552 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1554 metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
1558 r = dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks);
1560 metadata_operation_failed(pool, "dm_pool_get_free_metadata_block_count", r);
1565 /* Let's commit before we use up the metadata reserve. */
1575 * If we have run out of space, queue bios until the device is
1576 * resumed, presumably after having been reloaded with more space.
1578 static void retry_on_resume(struct bio *bio)
1580 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1581 struct thin_c *tc = h->tc;
1582 unsigned long flags;
1584 spin_lock_irqsave(&tc->lock, flags);
1585 bio_list_add(&tc->retry_on_resume_list, bio);
1586 spin_unlock_irqrestore(&tc->lock, flags);
1589 static int 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");
1599 case PM_OUT_OF_DATA_SPACE:
1600 return pool->pf.error_if_no_space ? -ENOSPC : 0;
1602 case PM_OUT_OF_METADATA_SPACE:
1607 /* Shouldn't get here */
1608 DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1613 static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
1615 int error = should_error_unserviceable_bio(pool);
1618 bio->bi_error = 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) ||
1812 (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA) ||
1813 bio_op(bio) == REQ_OP_DISCARD))
1814 bio_list_add(&info->defer_bios, bio);
1816 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));;
1818 h->shared_read_entry = dm_deferred_entry_inc(info->tc->pool->shared_read_ds);
1819 inc_all_io_entry(info->tc->pool, bio);
1820 bio_list_add(&info->issue_bios, bio);
1825 static void remap_and_issue_shared_cell(struct thin_c *tc,
1826 struct dm_bio_prison_cell *cell,
1830 struct remap_info info;
1833 bio_list_init(&info.defer_bios);
1834 bio_list_init(&info.issue_bios);
1836 cell_visit_release(tc->pool, __remap_and_issue_shared_cell,
1839 while ((bio = bio_list_pop(&info.defer_bios)))
1840 thin_defer_bio(tc, bio);
1842 while ((bio = bio_list_pop(&info.issue_bios)))
1843 remap_and_issue(tc, bio, block);
1846 static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1848 struct dm_thin_lookup_result *lookup_result,
1849 struct dm_bio_prison_cell *virt_cell)
1851 struct dm_bio_prison_cell *data_cell;
1852 struct pool *pool = tc->pool;
1853 struct dm_cell_key key;
1856 * If cell is already occupied, then sharing is already in the process
1857 * of being broken so we have nothing further to do here.
1859 build_data_key(tc->td, lookup_result->block, &key);
1860 if (bio_detain(pool, &key, bio, &data_cell)) {
1861 cell_defer_no_holder(tc, virt_cell);
1865 if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) {
1866 break_sharing(tc, bio, block, &key, lookup_result, data_cell);
1867 cell_defer_no_holder(tc, virt_cell);
1869 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1871 h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
1872 inc_all_io_entry(pool, bio);
1873 remap_and_issue(tc, bio, lookup_result->block);
1875 remap_and_issue_shared_cell(tc, data_cell, lookup_result->block);
1876 remap_and_issue_shared_cell(tc, virt_cell, lookup_result->block);
1880 static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
1881 struct dm_bio_prison_cell *cell)
1884 dm_block_t data_block;
1885 struct pool *pool = tc->pool;
1888 * Remap empty bios (flushes) immediately, without provisioning.
1890 if (!bio->bi_iter.bi_size) {
1891 inc_all_io_entry(pool, bio);
1892 cell_defer_no_holder(tc, cell);
1894 remap_and_issue(tc, bio, 0);
1899 * Fill read bios with zeroes and complete them immediately.
1901 if (bio_data_dir(bio) == READ) {
1903 cell_defer_no_holder(tc, cell);
1908 r = alloc_data_block(tc, &data_block);
1912 schedule_external_copy(tc, block, data_block, cell, bio);
1914 schedule_zero(tc, block, data_block, cell, bio);
1918 retry_bios_on_resume(pool, cell);
1922 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1924 cell_error(pool, cell);
1929 static void process_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
1932 struct pool *pool = tc->pool;
1933 struct bio *bio = cell->holder;
1934 dm_block_t block = get_bio_block(tc, bio);
1935 struct dm_thin_lookup_result lookup_result;
1937 if (tc->requeue_mode) {
1938 cell_requeue(pool, cell);
1942 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1945 if (lookup_result.shared)
1946 process_shared_bio(tc, bio, block, &lookup_result, cell);
1948 inc_all_io_entry(pool, bio);
1949 remap_and_issue(tc, bio, lookup_result.block);
1950 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
1955 if (bio_data_dir(bio) == READ && tc->origin_dev) {
1956 inc_all_io_entry(pool, bio);
1957 cell_defer_no_holder(tc, cell);
1959 if (bio_end_sector(bio) <= tc->origin_size)
1960 remap_to_origin_and_issue(tc, bio);
1962 else if (bio->bi_iter.bi_sector < tc->origin_size) {
1964 bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT;
1965 remap_to_origin_and_issue(tc, bio);
1972 provision_block(tc, bio, block, cell);
1976 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1978 cell_defer_no_holder(tc, cell);
1984 static void process_bio(struct thin_c *tc, struct bio *bio)
1986 struct pool *pool = tc->pool;
1987 dm_block_t block = get_bio_block(tc, bio);
1988 struct dm_bio_prison_cell *cell;
1989 struct dm_cell_key key;
1992 * If cell is already occupied, then the block is already
1993 * being provisioned so we have nothing further to do here.
1995 build_virtual_key(tc->td, block, &key);
1996 if (bio_detain(pool, &key, bio, &cell))
1999 process_cell(tc, cell);
2002 static void __process_bio_read_only(struct thin_c *tc, struct bio *bio,
2003 struct dm_bio_prison_cell *cell)
2006 int rw = bio_data_dir(bio);
2007 dm_block_t block = get_bio_block(tc, bio);
2008 struct dm_thin_lookup_result lookup_result;
2010 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
2013 if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) {
2014 handle_unserviceable_bio(tc->pool, bio);
2016 cell_defer_no_holder(tc, cell);
2018 inc_all_io_entry(tc->pool, bio);
2019 remap_and_issue(tc, bio, lookup_result.block);
2021 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
2027 cell_defer_no_holder(tc, cell);
2029 handle_unserviceable_bio(tc->pool, bio);
2033 if (tc->origin_dev) {
2034 inc_all_io_entry(tc->pool, bio);
2035 remap_to_origin_and_issue(tc, bio);
2044 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
2047 cell_defer_no_holder(tc, cell);
2053 static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
2055 __process_bio_read_only(tc, bio, NULL);
2058 static void process_cell_read_only(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2060 __process_bio_read_only(tc, cell->holder, cell);
2063 static void process_bio_success(struct thin_c *tc, struct bio *bio)
2068 static void process_bio_fail(struct thin_c *tc, struct bio *bio)
2073 static void process_cell_success(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2075 cell_success(tc->pool, cell);
2078 static void process_cell_fail(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2080 cell_error(tc->pool, cell);
2084 * FIXME: should we also commit due to size of transaction, measured in
2087 static int need_commit_due_to_time(struct pool *pool)
2089 return !time_in_range(jiffies, pool->last_commit_jiffies,
2090 pool->last_commit_jiffies + COMMIT_PERIOD);
2093 #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node)
2094 #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook))
2096 static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio)
2098 struct rb_node **rbp, *parent;
2099 struct dm_thin_endio_hook *pbd;
2100 sector_t bi_sector = bio->bi_iter.bi_sector;
2102 rbp = &tc->sort_bio_list.rb_node;
2106 pbd = thin_pbd(parent);
2108 if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector)
2109 rbp = &(*rbp)->rb_left;
2111 rbp = &(*rbp)->rb_right;
2114 pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2115 rb_link_node(&pbd->rb_node, parent, rbp);
2116 rb_insert_color(&pbd->rb_node, &tc->sort_bio_list);
2119 static void __extract_sorted_bios(struct thin_c *tc)
2121 struct rb_node *node;
2122 struct dm_thin_endio_hook *pbd;
2125 for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) {
2126 pbd = thin_pbd(node);
2127 bio = thin_bio(pbd);
2129 bio_list_add(&tc->deferred_bio_list, bio);
2130 rb_erase(&pbd->rb_node, &tc->sort_bio_list);
2133 WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list));
2136 static void __sort_thin_deferred_bios(struct thin_c *tc)
2139 struct bio_list bios;
2141 bio_list_init(&bios);
2142 bio_list_merge(&bios, &tc->deferred_bio_list);
2143 bio_list_init(&tc->deferred_bio_list);
2145 /* Sort deferred_bio_list using rb-tree */
2146 while ((bio = bio_list_pop(&bios)))
2147 __thin_bio_rb_add(tc, bio);
2150 * Transfer the sorted bios in sort_bio_list back to
2151 * deferred_bio_list to allow lockless submission of
2154 __extract_sorted_bios(tc);
2157 static void process_thin_deferred_bios(struct thin_c *tc)
2159 struct pool *pool = tc->pool;
2160 unsigned long flags;
2162 struct bio_list bios;
2163 struct blk_plug plug;
2166 if (tc->requeue_mode) {
2167 error_thin_bio_list(tc, &tc->deferred_bio_list, DM_ENDIO_REQUEUE);
2171 bio_list_init(&bios);
2173 spin_lock_irqsave(&tc->lock, flags);
2175 if (bio_list_empty(&tc->deferred_bio_list)) {
2176 spin_unlock_irqrestore(&tc->lock, flags);
2180 __sort_thin_deferred_bios(tc);
2182 bio_list_merge(&bios, &tc->deferred_bio_list);
2183 bio_list_init(&tc->deferred_bio_list);
2185 spin_unlock_irqrestore(&tc->lock, flags);
2187 blk_start_plug(&plug);
2188 while ((bio = bio_list_pop(&bios))) {
2190 * If we've got no free new_mapping structs, and processing
2191 * this bio might require one, we pause until there are some
2192 * prepared mappings to process.
2194 if (ensure_next_mapping(pool)) {
2195 spin_lock_irqsave(&tc->lock, flags);
2196 bio_list_add(&tc->deferred_bio_list, bio);
2197 bio_list_merge(&tc->deferred_bio_list, &bios);
2198 spin_unlock_irqrestore(&tc->lock, flags);
2202 if (bio_op(bio) == REQ_OP_DISCARD)
2203 pool->process_discard(tc, bio);
2205 pool->process_bio(tc, bio);
2207 if ((count++ & 127) == 0) {
2208 throttle_work_update(&pool->throttle);
2209 dm_pool_issue_prefetches(pool->pmd);
2212 blk_finish_plug(&plug);
2215 static int cmp_cells(const void *lhs, const void *rhs)
2217 struct dm_bio_prison_cell *lhs_cell = *((struct dm_bio_prison_cell **) lhs);
2218 struct dm_bio_prison_cell *rhs_cell = *((struct dm_bio_prison_cell **) rhs);
2220 BUG_ON(!lhs_cell->holder);
2221 BUG_ON(!rhs_cell->holder);
2223 if (lhs_cell->holder->bi_iter.bi_sector < rhs_cell->holder->bi_iter.bi_sector)
2226 if (lhs_cell->holder->bi_iter.bi_sector > rhs_cell->holder->bi_iter.bi_sector)
2232 static unsigned sort_cells(struct pool *pool, struct list_head *cells)
2235 struct dm_bio_prison_cell *cell, *tmp;
2237 list_for_each_entry_safe(cell, tmp, cells, user_list) {
2238 if (count >= CELL_SORT_ARRAY_SIZE)
2241 pool->cell_sort_array[count++] = cell;
2242 list_del(&cell->user_list);
2245 sort(pool->cell_sort_array, count, sizeof(cell), cmp_cells, NULL);
2250 static void process_thin_deferred_cells(struct thin_c *tc)
2252 struct pool *pool = tc->pool;
2253 unsigned long flags;
2254 struct list_head cells;
2255 struct dm_bio_prison_cell *cell;
2256 unsigned i, j, count;
2258 INIT_LIST_HEAD(&cells);
2260 spin_lock_irqsave(&tc->lock, flags);
2261 list_splice_init(&tc->deferred_cells, &cells);
2262 spin_unlock_irqrestore(&tc->lock, flags);
2264 if (list_empty(&cells))
2268 count = sort_cells(tc->pool, &cells);
2270 for (i = 0; i < count; i++) {
2271 cell = pool->cell_sort_array[i];
2272 BUG_ON(!cell->holder);
2275 * If we've got no free new_mapping structs, and processing
2276 * this bio might require one, we pause until there are some
2277 * prepared mappings to process.
2279 if (ensure_next_mapping(pool)) {
2280 for (j = i; j < count; j++)
2281 list_add(&pool->cell_sort_array[j]->user_list, &cells);
2283 spin_lock_irqsave(&tc->lock, flags);
2284 list_splice(&cells, &tc->deferred_cells);
2285 spin_unlock_irqrestore(&tc->lock, flags);
2289 if (bio_op(cell->holder) == REQ_OP_DISCARD)
2290 pool->process_discard_cell(tc, cell);
2292 pool->process_cell(tc, cell);
2294 } while (!list_empty(&cells));
2297 static void thin_get(struct thin_c *tc);
2298 static void thin_put(struct thin_c *tc);
2301 * We can't hold rcu_read_lock() around code that can block. So we
2302 * find a thin with the rcu lock held; bump a refcount; then drop
2305 static struct thin_c *get_first_thin(struct pool *pool)
2307 struct thin_c *tc = NULL;
2310 if (!list_empty(&pool->active_thins)) {
2311 tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list);
2319 static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc)
2321 struct thin_c *old_tc = tc;
2324 list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) {
2336 static void process_deferred_bios(struct pool *pool)
2338 unsigned long flags;
2340 struct bio_list bios, bio_completions;
2343 tc = get_first_thin(pool);
2345 process_thin_deferred_cells(tc);
2346 process_thin_deferred_bios(tc);
2347 tc = get_next_thin(pool, tc);
2351 * If there are any deferred flush bios, we must commit the metadata
2352 * before issuing them or signaling their completion.
2354 bio_list_init(&bios);
2355 bio_list_init(&bio_completions);
2357 spin_lock_irqsave(&pool->lock, flags);
2358 bio_list_merge(&bios, &pool->deferred_flush_bios);
2359 bio_list_init(&pool->deferred_flush_bios);
2361 bio_list_merge(&bio_completions, &pool->deferred_flush_completions);
2362 bio_list_init(&pool->deferred_flush_completions);
2363 spin_unlock_irqrestore(&pool->lock, flags);
2365 if (bio_list_empty(&bios) && bio_list_empty(&bio_completions) &&
2366 !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool)))
2370 bio_list_merge(&bios, &bio_completions);
2372 while ((bio = bio_list_pop(&bios)))
2376 pool->last_commit_jiffies = jiffies;
2378 while ((bio = bio_list_pop(&bio_completions)))
2381 while ((bio = bio_list_pop(&bios)))
2382 generic_make_request(bio);
2385 static void do_worker(struct work_struct *ws)
2387 struct pool *pool = container_of(ws, struct pool, worker);
2389 throttle_work_start(&pool->throttle);
2390 dm_pool_issue_prefetches(pool->pmd);
2391 throttle_work_update(&pool->throttle);
2392 process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
2393 throttle_work_update(&pool->throttle);
2394 process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
2395 throttle_work_update(&pool->throttle);
2396 process_prepared(pool, &pool->prepared_discards_pt2, &pool->process_prepared_discard_pt2);
2397 throttle_work_update(&pool->throttle);
2398 process_deferred_bios(pool);
2399 throttle_work_complete(&pool->throttle);
2403 * We want to commit periodically so that not too much
2404 * unwritten data builds up.
2406 static void do_waker(struct work_struct *ws)
2408 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
2410 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
2413 static void notify_of_pool_mode_change_to_oods(struct pool *pool);
2416 * We're holding onto IO to allow userland time to react. After the
2417 * timeout either the pool will have been resized (and thus back in
2418 * PM_WRITE mode), or we degrade to PM_OUT_OF_DATA_SPACE w/ error_if_no_space.
2420 static void do_no_space_timeout(struct work_struct *ws)
2422 struct pool *pool = container_of(to_delayed_work(ws), struct pool,
2425 if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) {
2426 pool->pf.error_if_no_space = true;
2427 notify_of_pool_mode_change_to_oods(pool);
2428 error_retry_list_with_code(pool, -ENOSPC);
2432 /*----------------------------------------------------------------*/
2435 struct work_struct worker;
2436 struct completion complete;
2439 static struct pool_work *to_pool_work(struct work_struct *ws)
2441 return container_of(ws, struct pool_work, worker);
2444 static void pool_work_complete(struct pool_work *pw)
2446 complete(&pw->complete);
2449 static void pool_work_wait(struct pool_work *pw, struct pool *pool,
2450 void (*fn)(struct work_struct *))
2452 INIT_WORK_ONSTACK(&pw->worker, fn);
2453 init_completion(&pw->complete);
2454 queue_work(pool->wq, &pw->worker);
2455 wait_for_completion(&pw->complete);
2458 /*----------------------------------------------------------------*/
2460 struct noflush_work {
2461 struct pool_work pw;
2465 static struct noflush_work *to_noflush(struct work_struct *ws)
2467 return container_of(to_pool_work(ws), struct noflush_work, pw);
2470 static void do_noflush_start(struct work_struct *ws)
2472 struct noflush_work *w = to_noflush(ws);
2473 w->tc->requeue_mode = true;
2475 pool_work_complete(&w->pw);
2478 static void do_noflush_stop(struct work_struct *ws)
2480 struct noflush_work *w = to_noflush(ws);
2481 w->tc->requeue_mode = false;
2482 pool_work_complete(&w->pw);
2485 static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *))
2487 struct noflush_work w;
2490 pool_work_wait(&w.pw, tc->pool, fn);
2493 /*----------------------------------------------------------------*/
2495 static enum pool_mode get_pool_mode(struct pool *pool)
2497 return pool->pf.mode;
2500 static void notify_of_pool_mode_change(struct pool *pool, const char *new_mode)
2502 dm_table_event(pool->ti->table);
2503 DMINFO("%s: switching pool to %s mode",
2504 dm_device_name(pool->pool_md), new_mode);
2507 static void notify_of_pool_mode_change_to_oods(struct pool *pool)
2509 if (!pool->pf.error_if_no_space)
2510 notify_of_pool_mode_change(pool, "out-of-data-space (queue IO)");
2512 notify_of_pool_mode_change(pool, "out-of-data-space (error IO)");
2515 static bool passdown_enabled(struct pool_c *pt)
2517 return pt->adjusted_pf.discard_passdown;
2520 static void set_discard_callbacks(struct pool *pool)
2522 struct pool_c *pt = pool->ti->private;
2524 if (passdown_enabled(pt)) {
2525 pool->process_discard_cell = process_discard_cell_passdown;
2526 pool->process_prepared_discard = process_prepared_discard_passdown_pt1;
2527 pool->process_prepared_discard_pt2 = process_prepared_discard_passdown_pt2;
2529 pool->process_discard_cell = process_discard_cell_no_passdown;
2530 pool->process_prepared_discard = process_prepared_discard_no_passdown;
2534 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode)
2536 struct pool_c *pt = pool->ti->private;
2537 bool needs_check = dm_pool_metadata_needs_check(pool->pmd);
2538 enum pool_mode old_mode = get_pool_mode(pool);
2539 unsigned long no_space_timeout = ACCESS_ONCE(no_space_timeout_secs) * HZ;
2542 * Never allow the pool to transition to PM_WRITE mode if user
2543 * intervention is required to verify metadata and data consistency.
2545 if (new_mode == PM_WRITE && needs_check) {
2546 DMERR("%s: unable to switch pool to write mode until repaired.",
2547 dm_device_name(pool->pool_md));
2548 if (old_mode != new_mode)
2549 new_mode = old_mode;
2551 new_mode = PM_READ_ONLY;
2554 * If we were in PM_FAIL mode, rollback of metadata failed. We're
2555 * not going to recover without a thin_repair. So we never let the
2556 * pool move out of the old mode.
2558 if (old_mode == PM_FAIL)
2559 new_mode = old_mode;
2563 if (old_mode != new_mode)
2564 notify_of_pool_mode_change(pool, "failure");
2565 dm_pool_metadata_read_only(pool->pmd);
2566 pool->process_bio = process_bio_fail;
2567 pool->process_discard = process_bio_fail;
2568 pool->process_cell = process_cell_fail;
2569 pool->process_discard_cell = process_cell_fail;
2570 pool->process_prepared_mapping = process_prepared_mapping_fail;
2571 pool->process_prepared_discard = process_prepared_discard_fail;
2573 error_retry_list(pool);
2576 case PM_OUT_OF_METADATA_SPACE:
2578 if (!is_read_only_pool_mode(old_mode))
2579 notify_of_pool_mode_change(pool, "read-only");
2580 dm_pool_metadata_read_only(pool->pmd);
2581 pool->process_bio = process_bio_read_only;
2582 pool->process_discard = process_bio_success;
2583 pool->process_cell = process_cell_read_only;
2584 pool->process_discard_cell = process_cell_success;
2585 pool->process_prepared_mapping = process_prepared_mapping_fail;
2586 pool->process_prepared_discard = process_prepared_discard_success;
2588 error_retry_list(pool);
2591 case PM_OUT_OF_DATA_SPACE:
2593 * Ideally we'd never hit this state; the low water mark
2594 * would trigger userland to extend the pool before we
2595 * completely run out of data space. However, many small
2596 * IOs to unprovisioned space can consume data space at an
2597 * alarming rate. Adjust your low water mark if you're
2598 * frequently seeing this mode.
2600 if (old_mode != new_mode)
2601 notify_of_pool_mode_change_to_oods(pool);
2602 pool->out_of_data_space = true;
2603 pool->process_bio = process_bio_read_only;
2604 pool->process_discard = process_discard_bio;
2605 pool->process_cell = process_cell_read_only;
2606 pool->process_prepared_mapping = process_prepared_mapping;
2607 set_discard_callbacks(pool);
2609 if (!pool->pf.error_if_no_space && no_space_timeout)
2610 queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout);
2614 if (old_mode != new_mode)
2615 notify_of_pool_mode_change(pool, "write");
2616 if (old_mode == PM_OUT_OF_DATA_SPACE)
2617 cancel_delayed_work_sync(&pool->no_space_timeout);
2618 pool->out_of_data_space = false;
2619 pool->pf.error_if_no_space = pt->requested_pf.error_if_no_space;
2620 dm_pool_metadata_read_write(pool->pmd);
2621 pool->process_bio = process_bio;
2622 pool->process_discard = process_discard_bio;
2623 pool->process_cell = process_cell;
2624 pool->process_prepared_mapping = process_prepared_mapping;
2625 set_discard_callbacks(pool);
2629 pool->pf.mode = new_mode;
2631 * The pool mode may have changed, sync it so bind_control_target()
2632 * doesn't cause an unexpected mode transition on resume.
2634 pt->adjusted_pf.mode = new_mode;
2637 static void abort_transaction(struct pool *pool)
2639 const char *dev_name = dm_device_name(pool->pool_md);
2641 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name);
2642 if (dm_pool_abort_metadata(pool->pmd)) {
2643 DMERR("%s: failed to abort metadata transaction", dev_name);
2644 set_pool_mode(pool, PM_FAIL);
2647 if (dm_pool_metadata_set_needs_check(pool->pmd)) {
2648 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name);
2649 set_pool_mode(pool, PM_FAIL);
2653 static void metadata_operation_failed(struct pool *pool, const char *op, int r)
2655 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
2656 dm_device_name(pool->pool_md), op, r);
2658 abort_transaction(pool);
2659 set_pool_mode(pool, PM_READ_ONLY);
2662 /*----------------------------------------------------------------*/
2665 * Mapping functions.
2669 * Called only while mapping a thin bio to hand it over to the workqueue.
2671 static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
2673 unsigned long flags;
2674 struct pool *pool = tc->pool;
2676 spin_lock_irqsave(&tc->lock, flags);
2677 bio_list_add(&tc->deferred_bio_list, bio);
2678 spin_unlock_irqrestore(&tc->lock, flags);
2683 static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio)
2685 struct pool *pool = tc->pool;
2687 throttle_lock(&pool->throttle);
2688 thin_defer_bio(tc, bio);
2689 throttle_unlock(&pool->throttle);
2692 static void thin_defer_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2694 unsigned long flags;
2695 struct pool *pool = tc->pool;
2697 throttle_lock(&pool->throttle);
2698 spin_lock_irqsave(&tc->lock, flags);
2699 list_add_tail(&cell->user_list, &tc->deferred_cells);
2700 spin_unlock_irqrestore(&tc->lock, flags);
2701 throttle_unlock(&pool->throttle);
2706 static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
2708 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2711 h->shared_read_entry = NULL;
2712 h->all_io_entry = NULL;
2713 h->overwrite_mapping = NULL;
2718 * Non-blocking function called from the thin target's map function.
2720 static int thin_bio_map(struct dm_target *ti, struct bio *bio)
2723 struct thin_c *tc = ti->private;
2724 dm_block_t block = get_bio_block(tc, bio);
2725 struct dm_thin_device *td = tc->td;
2726 struct dm_thin_lookup_result result;
2727 struct dm_bio_prison_cell *virt_cell, *data_cell;
2728 struct dm_cell_key key;
2730 thin_hook_bio(tc, bio);
2732 if (tc->requeue_mode) {
2733 bio->bi_error = DM_ENDIO_REQUEUE;
2735 return DM_MAPIO_SUBMITTED;
2738 if (get_pool_mode(tc->pool) == PM_FAIL) {
2740 return DM_MAPIO_SUBMITTED;
2743 if (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA) ||
2744 bio_op(bio) == REQ_OP_DISCARD) {
2745 thin_defer_bio_with_throttle(tc, bio);
2746 return DM_MAPIO_SUBMITTED;
2750 * We must hold the virtual cell before doing the lookup, otherwise
2751 * there's a race with discard.
2753 build_virtual_key(tc->td, block, &key);
2754 if (bio_detain(tc->pool, &key, bio, &virt_cell))
2755 return DM_MAPIO_SUBMITTED;
2757 r = dm_thin_find_block(td, block, 0, &result);
2760 * Note that we defer readahead too.
2764 if (unlikely(result.shared)) {
2766 * We have a race condition here between the
2767 * result.shared value returned by the lookup and
2768 * snapshot creation, which may cause new
2771 * To avoid this always quiesce the origin before
2772 * taking the snap. You want to do this anyway to
2773 * ensure a consistent application view
2776 * More distant ancestors are irrelevant. The
2777 * shared flag will be set in their case.
2779 thin_defer_cell(tc, virt_cell);
2780 return DM_MAPIO_SUBMITTED;
2783 build_data_key(tc->td, result.block, &key);
2784 if (bio_detain(tc->pool, &key, bio, &data_cell)) {
2785 cell_defer_no_holder(tc, virt_cell);
2786 return DM_MAPIO_SUBMITTED;
2789 inc_all_io_entry(tc->pool, bio);
2790 cell_defer_no_holder(tc, data_cell);
2791 cell_defer_no_holder(tc, virt_cell);
2793 remap(tc, bio, result.block);
2794 return DM_MAPIO_REMAPPED;
2798 thin_defer_cell(tc, virt_cell);
2799 return DM_MAPIO_SUBMITTED;
2803 * Must always call bio_io_error on failure.
2804 * dm_thin_find_block can fail with -EINVAL if the
2805 * pool is switched to fail-io mode.
2808 cell_defer_no_holder(tc, virt_cell);
2809 return DM_MAPIO_SUBMITTED;
2813 static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
2815 struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
2816 struct request_queue *q;
2818 if (get_pool_mode(pt->pool) == PM_OUT_OF_DATA_SPACE)
2821 q = bdev_get_queue(pt->data_dev->bdev);
2822 return bdi_congested(&q->backing_dev_info, bdi_bits);
2825 static void requeue_bios(struct pool *pool)
2827 unsigned long flags;
2831 list_for_each_entry_rcu(tc, &pool->active_thins, list) {
2832 spin_lock_irqsave(&tc->lock, flags);
2833 bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list);
2834 bio_list_init(&tc->retry_on_resume_list);
2835 spin_unlock_irqrestore(&tc->lock, flags);
2840 /*----------------------------------------------------------------
2841 * Binding of control targets to a pool object
2842 *--------------------------------------------------------------*/
2843 static bool data_dev_supports_discard(struct pool_c *pt)
2845 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
2847 return q && blk_queue_discard(q);
2850 static bool is_factor(sector_t block_size, uint32_t n)
2852 return !sector_div(block_size, n);
2856 * If discard_passdown was enabled verify that the data device
2857 * supports discards. Disable discard_passdown if not.
2859 static void disable_passdown_if_not_supported(struct pool_c *pt)
2861 struct pool *pool = pt->pool;
2862 struct block_device *data_bdev = pt->data_dev->bdev;
2863 struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
2864 const char *reason = NULL;
2865 char buf[BDEVNAME_SIZE];
2867 if (!pt->adjusted_pf.discard_passdown)
2870 if (!data_dev_supports_discard(pt))
2871 reason = "discard unsupported";
2873 else if (data_limits->max_discard_sectors < pool->sectors_per_block)
2874 reason = "max discard sectors smaller than a block";
2877 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason);
2878 pt->adjusted_pf.discard_passdown = false;
2882 static int bind_control_target(struct pool *pool, struct dm_target *ti)
2884 struct pool_c *pt = ti->private;
2887 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
2889 enum pool_mode old_mode = get_pool_mode(pool);
2890 enum pool_mode new_mode = pt->adjusted_pf.mode;
2893 * Don't change the pool's mode until set_pool_mode() below.
2894 * Otherwise the pool's process_* function pointers may
2895 * not match the desired pool mode.
2897 pt->adjusted_pf.mode = old_mode;
2900 pool->pf = pt->adjusted_pf;
2901 pool->low_water_blocks = pt->low_water_blocks;
2903 set_pool_mode(pool, new_mode);
2908 static void unbind_control_target(struct pool *pool, struct dm_target *ti)
2914 /*----------------------------------------------------------------
2916 *--------------------------------------------------------------*/
2917 /* Initialize pool features. */
2918 static void pool_features_init(struct pool_features *pf)
2920 pf->mode = PM_WRITE;
2921 pf->zero_new_blocks = true;
2922 pf->discard_enabled = true;
2923 pf->discard_passdown = true;
2924 pf->error_if_no_space = false;
2927 static void __pool_destroy(struct pool *pool)
2929 __pool_table_remove(pool);
2931 vfree(pool->cell_sort_array);
2932 if (dm_pool_metadata_close(pool->pmd) < 0)
2933 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
2935 dm_bio_prison_destroy(pool->prison);
2936 dm_kcopyd_client_destroy(pool->copier);
2939 destroy_workqueue(pool->wq);
2941 if (pool->next_mapping)
2942 mempool_free(pool->next_mapping, pool->mapping_pool);
2943 mempool_destroy(pool->mapping_pool);
2944 dm_deferred_set_destroy(pool->shared_read_ds);
2945 dm_deferred_set_destroy(pool->all_io_ds);
2949 static struct kmem_cache *_new_mapping_cache;
2951 static struct pool *pool_create(struct mapped_device *pool_md,
2952 struct block_device *metadata_dev,
2953 unsigned long block_size,
2954 int read_only, char **error)
2959 struct dm_pool_metadata *pmd;
2960 bool format_device = read_only ? false : true;
2962 pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
2964 *error = "Error creating metadata object";
2965 return (struct pool *)pmd;
2968 pool = kzalloc(sizeof(*pool), GFP_KERNEL);
2970 *error = "Error allocating memory for pool";
2971 err_p = ERR_PTR(-ENOMEM);
2976 pool->sectors_per_block = block_size;
2977 if (block_size & (block_size - 1))
2978 pool->sectors_per_block_shift = -1;
2980 pool->sectors_per_block_shift = __ffs(block_size);
2981 pool->low_water_blocks = 0;
2982 pool_features_init(&pool->pf);
2983 pool->prison = dm_bio_prison_create();
2984 if (!pool->prison) {
2985 *error = "Error creating pool's bio prison";
2986 err_p = ERR_PTR(-ENOMEM);
2990 pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2991 if (IS_ERR(pool->copier)) {
2992 r = PTR_ERR(pool->copier);
2993 *error = "Error creating pool's kcopyd client";
2995 goto bad_kcopyd_client;
2999 * Create singlethreaded workqueue that will service all devices
3000 * that use this metadata.
3002 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
3004 *error = "Error creating pool's workqueue";
3005 err_p = ERR_PTR(-ENOMEM);
3009 throttle_init(&pool->throttle);
3010 INIT_WORK(&pool->worker, do_worker);
3011 INIT_DELAYED_WORK(&pool->waker, do_waker);
3012 INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout);
3013 spin_lock_init(&pool->lock);
3014 bio_list_init(&pool->deferred_flush_bios);
3015 bio_list_init(&pool->deferred_flush_completions);
3016 INIT_LIST_HEAD(&pool->prepared_mappings);
3017 INIT_LIST_HEAD(&pool->prepared_discards);
3018 INIT_LIST_HEAD(&pool->prepared_discards_pt2);
3019 INIT_LIST_HEAD(&pool->active_thins);
3020 pool->low_water_triggered = false;
3021 pool->suspended = true;
3022 pool->out_of_data_space = false;
3024 pool->shared_read_ds = dm_deferred_set_create();
3025 if (!pool->shared_read_ds) {
3026 *error = "Error creating pool's shared read deferred set";
3027 err_p = ERR_PTR(-ENOMEM);
3028 goto bad_shared_read_ds;
3031 pool->all_io_ds = dm_deferred_set_create();
3032 if (!pool->all_io_ds) {
3033 *error = "Error creating pool's all io deferred set";
3034 err_p = ERR_PTR(-ENOMEM);
3038 pool->next_mapping = NULL;
3039 pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE,
3040 _new_mapping_cache);
3041 if (!pool->mapping_pool) {
3042 *error = "Error creating pool's mapping mempool";
3043 err_p = ERR_PTR(-ENOMEM);
3044 goto bad_mapping_pool;
3047 pool->cell_sort_array = vmalloc(sizeof(*pool->cell_sort_array) * CELL_SORT_ARRAY_SIZE);
3048 if (!pool->cell_sort_array) {
3049 *error = "Error allocating cell sort array";
3050 err_p = ERR_PTR(-ENOMEM);
3051 goto bad_sort_array;
3054 pool->ref_count = 1;
3055 pool->last_commit_jiffies = jiffies;
3056 pool->pool_md = pool_md;
3057 pool->md_dev = metadata_dev;
3058 __pool_table_insert(pool);
3063 mempool_destroy(pool->mapping_pool);
3065 dm_deferred_set_destroy(pool->all_io_ds);
3067 dm_deferred_set_destroy(pool->shared_read_ds);
3069 destroy_workqueue(pool->wq);
3071 dm_kcopyd_client_destroy(pool->copier);
3073 dm_bio_prison_destroy(pool->prison);
3077 if (dm_pool_metadata_close(pmd))
3078 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
3083 static void __pool_inc(struct pool *pool)
3085 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
3089 static void __pool_dec(struct pool *pool)
3091 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
3092 BUG_ON(!pool->ref_count);
3093 if (!--pool->ref_count)
3094 __pool_destroy(pool);
3097 static struct pool *__pool_find(struct mapped_device *pool_md,
3098 struct block_device *metadata_dev,
3099 unsigned long block_size, int read_only,
3100 char **error, int *created)
3102 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
3105 if (pool->pool_md != pool_md) {
3106 *error = "metadata device already in use by a pool";
3107 return ERR_PTR(-EBUSY);
3112 pool = __pool_table_lookup(pool_md);
3114 if (pool->md_dev != metadata_dev) {
3115 *error = "different pool cannot replace a pool";
3116 return ERR_PTR(-EINVAL);
3121 pool = pool_create(pool_md, metadata_dev, block_size, read_only, error);
3129 /*----------------------------------------------------------------
3130 * Pool target methods
3131 *--------------------------------------------------------------*/
3132 static void pool_dtr(struct dm_target *ti)
3134 struct pool_c *pt = ti->private;
3136 mutex_lock(&dm_thin_pool_table.mutex);
3138 unbind_control_target(pt->pool, ti);
3139 __pool_dec(pt->pool);
3140 dm_put_device(ti, pt->metadata_dev);
3141 dm_put_device(ti, pt->data_dev);
3144 mutex_unlock(&dm_thin_pool_table.mutex);
3147 static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
3148 struct dm_target *ti)
3152 const char *arg_name;
3154 static struct dm_arg _args[] = {
3155 {0, 4, "Invalid number of pool feature arguments"},
3159 * No feature arguments supplied.
3164 r = dm_read_arg_group(_args, as, &argc, &ti->error);
3168 while (argc && !r) {
3169 arg_name = dm_shift_arg(as);
3172 if (!strcasecmp(arg_name, "skip_block_zeroing"))
3173 pf->zero_new_blocks = false;
3175 else if (!strcasecmp(arg_name, "ignore_discard"))
3176 pf->discard_enabled = false;
3178 else if (!strcasecmp(arg_name, "no_discard_passdown"))
3179 pf->discard_passdown = false;
3181 else if (!strcasecmp(arg_name, "read_only"))
3182 pf->mode = PM_READ_ONLY;
3184 else if (!strcasecmp(arg_name, "error_if_no_space"))
3185 pf->error_if_no_space = true;
3188 ti->error = "Unrecognised pool feature requested";
3197 static void metadata_low_callback(void *context)
3199 struct pool *pool = context;
3201 DMWARN("%s: reached low water mark for metadata device: sending event.",
3202 dm_device_name(pool->pool_md));
3204 dm_table_event(pool->ti->table);
3207 static sector_t get_dev_size(struct block_device *bdev)
3209 return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
3212 static void warn_if_metadata_device_too_big(struct block_device *bdev)
3214 sector_t metadata_dev_size = get_dev_size(bdev);
3215 char buffer[BDEVNAME_SIZE];
3217 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING)
3218 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
3219 bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS);
3222 static sector_t get_metadata_dev_size(struct block_device *bdev)
3224 sector_t metadata_dev_size = get_dev_size(bdev);
3226 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS)
3227 metadata_dev_size = THIN_METADATA_MAX_SECTORS;
3229 return metadata_dev_size;
3232 static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
3234 sector_t metadata_dev_size = get_metadata_dev_size(bdev);
3236 sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE);
3238 return metadata_dev_size;
3242 * When a metadata threshold is crossed a dm event is triggered, and
3243 * userland should respond by growing the metadata device. We could let
3244 * userland set the threshold, like we do with the data threshold, but I'm
3245 * not sure they know enough to do this well.
3247 static dm_block_t calc_metadata_threshold(struct pool_c *pt)
3250 * 4M is ample for all ops with the possible exception of thin
3251 * device deletion which is harmless if it fails (just retry the
3252 * delete after you've grown the device).
3254 dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
3255 return min((dm_block_t)1024ULL /* 4M */, quarter);
3259 * thin-pool <metadata dev> <data dev>
3260 * <data block size (sectors)>
3261 * <low water mark (blocks)>
3262 * [<#feature args> [<arg>]*]
3264 * Optional feature arguments are:
3265 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
3266 * ignore_discard: disable discard
3267 * no_discard_passdown: don't pass discards down to the data device
3268 * read_only: Don't allow any changes to be made to the pool metadata.
3269 * error_if_no_space: error IOs, instead of queueing, if no space.
3271 static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
3273 int r, pool_created = 0;
3276 struct pool_features pf;
3277 struct dm_arg_set as;
3278 struct dm_dev *data_dev;
3279 unsigned long block_size;
3280 dm_block_t low_water_blocks;
3281 struct dm_dev *metadata_dev;
3282 fmode_t metadata_mode;
3285 * FIXME Remove validation from scope of lock.
3287 mutex_lock(&dm_thin_pool_table.mutex);
3290 ti->error = "Invalid argument count";
3298 /* make sure metadata and data are different devices */
3299 if (!strcmp(argv[0], argv[1])) {
3300 ti->error = "Error setting metadata or data device";
3306 * Set default pool features.
3308 pool_features_init(&pf);
3310 dm_consume_args(&as, 4);
3311 r = parse_pool_features(&as, &pf, ti);
3315 metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
3316 r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
3318 ti->error = "Error opening metadata block device";
3321 warn_if_metadata_device_too_big(metadata_dev->bdev);
3323 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
3325 ti->error = "Error getting data device";
3329 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
3330 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
3331 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
3332 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
3333 ti->error = "Invalid block size";
3338 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
3339 ti->error = "Invalid low water mark";
3344 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
3350 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
3351 block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
3358 * 'pool_created' reflects whether this is the first table load.
3359 * Top level discard support is not allowed to be changed after
3360 * initial load. This would require a pool reload to trigger thin
3363 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
3364 ti->error = "Discard support cannot be disabled once enabled";
3366 goto out_flags_changed;
3371 pt->metadata_dev = metadata_dev;
3372 pt->data_dev = data_dev;
3373 pt->low_water_blocks = low_water_blocks;
3374 pt->adjusted_pf = pt->requested_pf = pf;
3375 ti->num_flush_bios = 1;
3378 * Only need to enable discards if the pool should pass
3379 * them down to the data device. The thin device's discard
3380 * processing will cause mappings to be removed from the btree.
3382 ti->discard_zeroes_data_unsupported = true;
3383 if (pf.discard_enabled && pf.discard_passdown) {
3384 ti->num_discard_bios = 1;
3387 * Setting 'discards_supported' circumvents the normal
3388 * stacking of discard limits (this keeps the pool and
3389 * thin devices' discard limits consistent).
3391 ti->discards_supported = true;
3395 r = dm_pool_register_metadata_threshold(pt->pool->pmd,
3396 calc_metadata_threshold(pt),
3397 metadata_low_callback,
3400 goto out_flags_changed;
3402 pt->callbacks.congested_fn = pool_is_congested;
3403 dm_table_add_target_callbacks(ti->table, &pt->callbacks);
3405 mutex_unlock(&dm_thin_pool_table.mutex);
3414 dm_put_device(ti, data_dev);
3416 dm_put_device(ti, metadata_dev);
3418 mutex_unlock(&dm_thin_pool_table.mutex);
3423 static int pool_map(struct dm_target *ti, struct bio *bio)
3426 struct pool_c *pt = ti->private;
3427 struct pool *pool = pt->pool;
3428 unsigned long flags;
3431 * As this is a singleton target, ti->begin is always zero.
3433 spin_lock_irqsave(&pool->lock, flags);
3434 bio->bi_bdev = pt->data_dev->bdev;
3435 r = DM_MAPIO_REMAPPED;
3436 spin_unlock_irqrestore(&pool->lock, flags);
3441 static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
3444 struct pool_c *pt = ti->private;
3445 struct pool *pool = pt->pool;
3446 sector_t data_size = ti->len;
3447 dm_block_t sb_data_size;
3449 *need_commit = false;
3451 (void) sector_div(data_size, pool->sectors_per_block);
3453 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
3455 DMERR("%s: failed to retrieve data device size",
3456 dm_device_name(pool->pool_md));
3460 if (data_size < sb_data_size) {
3461 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
3462 dm_device_name(pool->pool_md),
3463 (unsigned long long)data_size, sb_data_size);
3466 } else if (data_size > sb_data_size) {
3467 if (dm_pool_metadata_needs_check(pool->pmd)) {
3468 DMERR("%s: unable to grow the data device until repaired.",
3469 dm_device_name(pool->pool_md));
3474 DMINFO("%s: growing the data device from %llu to %llu blocks",
3475 dm_device_name(pool->pool_md),
3476 sb_data_size, (unsigned long long)data_size);
3477 r = dm_pool_resize_data_dev(pool->pmd, data_size);
3479 metadata_operation_failed(pool, "dm_pool_resize_data_dev", r);
3483 *need_commit = true;
3489 static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
3492 struct pool_c *pt = ti->private;
3493 struct pool *pool = pt->pool;
3494 dm_block_t metadata_dev_size, sb_metadata_dev_size;
3496 *need_commit = false;
3498 metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
3500 r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
3502 DMERR("%s: failed to retrieve metadata device size",
3503 dm_device_name(pool->pool_md));
3507 if (metadata_dev_size < sb_metadata_dev_size) {
3508 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
3509 dm_device_name(pool->pool_md),
3510 metadata_dev_size, sb_metadata_dev_size);
3513 } else if (metadata_dev_size > sb_metadata_dev_size) {
3514 if (dm_pool_metadata_needs_check(pool->pmd)) {
3515 DMERR("%s: unable to grow the metadata device until repaired.",
3516 dm_device_name(pool->pool_md));
3520 warn_if_metadata_device_too_big(pool->md_dev);
3521 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
3522 dm_device_name(pool->pool_md),
3523 sb_metadata_dev_size, metadata_dev_size);
3525 if (get_pool_mode(pool) == PM_OUT_OF_METADATA_SPACE)
3526 set_pool_mode(pool, PM_WRITE);
3528 r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
3530 metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r);
3534 *need_commit = true;
3541 * Retrieves the number of blocks of the data device from
3542 * the superblock and compares it to the actual device size,
3543 * thus resizing the data device in case it has grown.
3545 * This both copes with opening preallocated data devices in the ctr
3546 * being followed by a resume
3548 * calling the resume method individually after userspace has
3549 * grown the data device in reaction to a table event.
3551 static int pool_preresume(struct dm_target *ti)
3554 bool need_commit1, need_commit2;
3555 struct pool_c *pt = ti->private;
3556 struct pool *pool = pt->pool;
3559 * Take control of the pool object.
3561 r = bind_control_target(pool, ti);
3565 r = maybe_resize_data_dev(ti, &need_commit1);
3569 r = maybe_resize_metadata_dev(ti, &need_commit2);
3573 if (need_commit1 || need_commit2)
3574 (void) commit(pool);
3579 static void pool_suspend_active_thins(struct pool *pool)
3583 /* Suspend all active thin devices */
3584 tc = get_first_thin(pool);
3586 dm_internal_suspend_noflush(tc->thin_md);
3587 tc = get_next_thin(pool, tc);
3591 static void pool_resume_active_thins(struct pool *pool)
3595 /* Resume all active thin devices */
3596 tc = get_first_thin(pool);
3598 dm_internal_resume(tc->thin_md);
3599 tc = get_next_thin(pool, tc);
3603 static void pool_resume(struct dm_target *ti)
3605 struct pool_c *pt = ti->private;
3606 struct pool *pool = pt->pool;
3607 unsigned long flags;
3610 * Must requeue active_thins' bios and then resume
3611 * active_thins _before_ clearing 'suspend' flag.
3614 pool_resume_active_thins(pool);
3616 spin_lock_irqsave(&pool->lock, flags);
3617 pool->low_water_triggered = false;
3618 pool->suspended = false;
3619 spin_unlock_irqrestore(&pool->lock, flags);
3621 do_waker(&pool->waker.work);
3624 static void pool_presuspend(struct dm_target *ti)
3626 struct pool_c *pt = ti->private;
3627 struct pool *pool = pt->pool;
3628 unsigned long flags;
3630 spin_lock_irqsave(&pool->lock, flags);
3631 pool->suspended = true;
3632 spin_unlock_irqrestore(&pool->lock, flags);
3634 pool_suspend_active_thins(pool);
3637 static void pool_presuspend_undo(struct dm_target *ti)
3639 struct pool_c *pt = ti->private;
3640 struct pool *pool = pt->pool;
3641 unsigned long flags;
3643 pool_resume_active_thins(pool);
3645 spin_lock_irqsave(&pool->lock, flags);
3646 pool->suspended = false;
3647 spin_unlock_irqrestore(&pool->lock, flags);
3650 static void pool_postsuspend(struct dm_target *ti)
3652 struct pool_c *pt = ti->private;
3653 struct pool *pool = pt->pool;
3655 cancel_delayed_work_sync(&pool->waker);
3656 cancel_delayed_work_sync(&pool->no_space_timeout);
3657 flush_workqueue(pool->wq);
3658 (void) commit(pool);
3661 static int check_arg_count(unsigned argc, unsigned args_required)
3663 if (argc != args_required) {
3664 DMWARN("Message received with %u arguments instead of %u.",
3665 argc, args_required);
3672 static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
3674 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
3675 *dev_id <= MAX_DEV_ID)
3679 DMWARN("Message received with invalid device id: %s", arg);
3684 static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
3689 r = check_arg_count(argc, 2);
3693 r = read_dev_id(argv[1], &dev_id, 1);
3697 r = dm_pool_create_thin(pool->pmd, dev_id);
3699 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
3707 static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3710 dm_thin_id origin_dev_id;
3713 r = check_arg_count(argc, 3);
3717 r = read_dev_id(argv[1], &dev_id, 1);
3721 r = read_dev_id(argv[2], &origin_dev_id, 1);
3725 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
3727 DMWARN("Creation of new snapshot %s of device %s failed.",
3735 static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
3740 r = check_arg_count(argc, 2);
3744 r = read_dev_id(argv[1], &dev_id, 1);
3748 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
3750 DMWARN("Deletion of thin device %s failed.", argv[1]);
3755 static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
3757 dm_thin_id old_id, new_id;
3760 r = check_arg_count(argc, 3);
3764 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
3765 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
3769 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
3770 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
3774 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
3776 DMWARN("Failed to change transaction id from %s to %s.",
3784 static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3788 r = check_arg_count(argc, 1);
3792 (void) commit(pool);
3794 r = dm_pool_reserve_metadata_snap(pool->pmd);
3796 DMWARN("reserve_metadata_snap message failed.");
3801 static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3805 r = check_arg_count(argc, 1);
3809 r = dm_pool_release_metadata_snap(pool->pmd);
3811 DMWARN("release_metadata_snap message failed.");
3817 * Messages supported:
3818 * create_thin <dev_id>
3819 * create_snap <dev_id> <origin_id>
3821 * set_transaction_id <current_trans_id> <new_trans_id>
3822 * reserve_metadata_snap
3823 * release_metadata_snap
3825 static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
3828 struct pool_c *pt = ti->private;
3829 struct pool *pool = pt->pool;
3831 if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE) {
3832 DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode",
3833 dm_device_name(pool->pool_md));
3837 if (!strcasecmp(argv[0], "create_thin"))
3838 r = process_create_thin_mesg(argc, argv, pool);
3840 else if (!strcasecmp(argv[0], "create_snap"))
3841 r = process_create_snap_mesg(argc, argv, pool);
3843 else if (!strcasecmp(argv[0], "delete"))
3844 r = process_delete_mesg(argc, argv, pool);
3846 else if (!strcasecmp(argv[0], "set_transaction_id"))
3847 r = process_set_transaction_id_mesg(argc, argv, pool);
3849 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
3850 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
3852 else if (!strcasecmp(argv[0], "release_metadata_snap"))
3853 r = process_release_metadata_snap_mesg(argc, argv, pool);
3856 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
3859 (void) commit(pool);
3864 static void emit_flags(struct pool_features *pf, char *result,
3865 unsigned sz, unsigned maxlen)
3867 unsigned count = !pf->zero_new_blocks + !pf->discard_enabled +
3868 !pf->discard_passdown + (pf->mode == PM_READ_ONLY) +
3869 pf->error_if_no_space;
3870 DMEMIT("%u ", count);
3872 if (!pf->zero_new_blocks)
3873 DMEMIT("skip_block_zeroing ");
3875 if (!pf->discard_enabled)
3876 DMEMIT("ignore_discard ");
3878 if (!pf->discard_passdown)
3879 DMEMIT("no_discard_passdown ");
3881 if (pf->mode == PM_READ_ONLY)
3882 DMEMIT("read_only ");
3884 if (pf->error_if_no_space)
3885 DMEMIT("error_if_no_space ");
3890 * <transaction id> <used metadata sectors>/<total metadata sectors>
3891 * <used data sectors>/<total data sectors> <held metadata root>
3892 * <pool mode> <discard config> <no space config> <needs_check>
3894 static void pool_status(struct dm_target *ti, status_type_t type,
3895 unsigned status_flags, char *result, unsigned maxlen)
3899 uint64_t transaction_id;
3900 dm_block_t nr_free_blocks_data;
3901 dm_block_t nr_free_blocks_metadata;
3902 dm_block_t nr_blocks_data;
3903 dm_block_t nr_blocks_metadata;
3904 dm_block_t held_root;
3905 enum pool_mode mode;
3906 char buf[BDEVNAME_SIZE];
3907 char buf2[BDEVNAME_SIZE];
3908 struct pool_c *pt = ti->private;
3909 struct pool *pool = pt->pool;
3912 case STATUSTYPE_INFO:
3913 if (get_pool_mode(pool) == PM_FAIL) {
3918 /* Commit to ensure statistics aren't out-of-date */
3919 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
3920 (void) commit(pool);
3922 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
3924 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
3925 dm_device_name(pool->pool_md), r);
3929 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
3931 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
3932 dm_device_name(pool->pool_md), r);
3936 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
3938 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
3939 dm_device_name(pool->pool_md), r);
3943 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
3945 DMERR("%s: dm_pool_get_free_block_count returned %d",
3946 dm_device_name(pool->pool_md), r);
3950 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
3952 DMERR("%s: dm_pool_get_data_dev_size returned %d",
3953 dm_device_name(pool->pool_md), r);
3957 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
3959 DMERR("%s: dm_pool_get_metadata_snap returned %d",
3960 dm_device_name(pool->pool_md), r);
3964 DMEMIT("%llu %llu/%llu %llu/%llu ",
3965 (unsigned long long)transaction_id,
3966 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
3967 (unsigned long long)nr_blocks_metadata,
3968 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
3969 (unsigned long long)nr_blocks_data);
3972 DMEMIT("%llu ", held_root);
3976 mode = get_pool_mode(pool);
3977 if (mode == PM_OUT_OF_DATA_SPACE)
3978 DMEMIT("out_of_data_space ");
3979 else if (is_read_only_pool_mode(mode))
3984 if (!pool->pf.discard_enabled)
3985 DMEMIT("ignore_discard ");
3986 else if (pool->pf.discard_passdown)
3987 DMEMIT("discard_passdown ");
3989 DMEMIT("no_discard_passdown ");
3991 if (pool->pf.error_if_no_space)
3992 DMEMIT("error_if_no_space ");
3994 DMEMIT("queue_if_no_space ");
3996 if (dm_pool_metadata_needs_check(pool->pmd))
3997 DMEMIT("needs_check ");
4003 case STATUSTYPE_TABLE:
4004 DMEMIT("%s %s %lu %llu ",
4005 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
4006 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
4007 (unsigned long)pool->sectors_per_block,
4008 (unsigned long long)pt->low_water_blocks);
4009 emit_flags(&pt->requested_pf, result, sz, maxlen);
4018 static int pool_iterate_devices(struct dm_target *ti,
4019 iterate_devices_callout_fn fn, void *data)
4021 struct pool_c *pt = ti->private;
4023 return fn(ti, pt->data_dev, 0, ti->len, data);
4026 static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
4028 struct pool_c *pt = ti->private;
4029 struct pool *pool = pt->pool;
4030 sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
4033 * If max_sectors is smaller than pool->sectors_per_block adjust it
4034 * to the highest possible power-of-2 factor of pool->sectors_per_block.
4035 * This is especially beneficial when the pool's data device is a RAID
4036 * device that has a full stripe width that matches pool->sectors_per_block
4037 * -- because even though partial RAID stripe-sized IOs will be issued to a
4038 * single RAID stripe; when aggregated they will end on a full RAID stripe
4039 * boundary.. which avoids additional partial RAID stripe writes cascading
4041 if (limits->max_sectors < pool->sectors_per_block) {
4042 while (!is_factor(pool->sectors_per_block, limits->max_sectors)) {
4043 if ((limits->max_sectors & (limits->max_sectors - 1)) == 0)
4044 limits->max_sectors--;
4045 limits->max_sectors = rounddown_pow_of_two(limits->max_sectors);
4050 * If the system-determined stacked limits are compatible with the
4051 * pool's blocksize (io_opt is a factor) do not override them.
4053 if (io_opt_sectors < pool->sectors_per_block ||
4054 !is_factor(io_opt_sectors, pool->sectors_per_block)) {
4055 if (is_factor(pool->sectors_per_block, limits->max_sectors))
4056 blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT);
4058 blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT);
4059 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
4063 * pt->adjusted_pf is a staging area for the actual features to use.
4064 * They get transferred to the live pool in bind_control_target()
4065 * called from pool_preresume().
4067 if (!pt->adjusted_pf.discard_enabled) {
4069 * Must explicitly disallow stacking discard limits otherwise the
4070 * block layer will stack them if pool's data device has support.
4071 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
4072 * user to see that, so make sure to set all discard limits to 0.
4074 limits->discard_granularity = 0;
4078 disable_passdown_if_not_supported(pt);
4081 * The pool uses the same discard limits as the underlying data
4082 * device. DM core has already set this up.
4086 static struct target_type pool_target = {
4087 .name = "thin-pool",
4088 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
4089 DM_TARGET_IMMUTABLE,
4090 .version = {1, 19, 0},
4091 .module = THIS_MODULE,
4095 .presuspend = pool_presuspend,
4096 .presuspend_undo = pool_presuspend_undo,
4097 .postsuspend = pool_postsuspend,
4098 .preresume = pool_preresume,
4099 .resume = pool_resume,
4100 .message = pool_message,
4101 .status = pool_status,
4102 .iterate_devices = pool_iterate_devices,
4103 .io_hints = pool_io_hints,
4106 /*----------------------------------------------------------------
4107 * Thin target methods
4108 *--------------------------------------------------------------*/
4109 static void thin_get(struct thin_c *tc)
4111 atomic_inc(&tc->refcount);
4114 static void thin_put(struct thin_c *tc)
4116 if (atomic_dec_and_test(&tc->refcount))
4117 complete(&tc->can_destroy);
4120 static void thin_dtr(struct dm_target *ti)
4122 struct thin_c *tc = ti->private;
4123 unsigned long flags;
4125 spin_lock_irqsave(&tc->pool->lock, flags);
4126 list_del_rcu(&tc->list);
4127 spin_unlock_irqrestore(&tc->pool->lock, flags);
4131 wait_for_completion(&tc->can_destroy);
4133 mutex_lock(&dm_thin_pool_table.mutex);
4135 __pool_dec(tc->pool);
4136 dm_pool_close_thin_device(tc->td);
4137 dm_put_device(ti, tc->pool_dev);
4139 dm_put_device(ti, tc->origin_dev);
4142 mutex_unlock(&dm_thin_pool_table.mutex);
4146 * Thin target parameters:
4148 * <pool_dev> <dev_id> [origin_dev]
4150 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
4151 * dev_id: the internal device identifier
4152 * origin_dev: a device external to the pool that should act as the origin
4154 * If the pool device has discards disabled, they get disabled for the thin
4157 static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
4161 struct dm_dev *pool_dev, *origin_dev;
4162 struct mapped_device *pool_md;
4163 unsigned long flags;
4165 mutex_lock(&dm_thin_pool_table.mutex);
4167 if (argc != 2 && argc != 3) {
4168 ti->error = "Invalid argument count";
4173 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
4175 ti->error = "Out of memory";
4179 tc->thin_md = dm_table_get_md(ti->table);
4180 spin_lock_init(&tc->lock);
4181 INIT_LIST_HEAD(&tc->deferred_cells);
4182 bio_list_init(&tc->deferred_bio_list);
4183 bio_list_init(&tc->retry_on_resume_list);
4184 tc->sort_bio_list = RB_ROOT;
4187 if (!strcmp(argv[0], argv[2])) {
4188 ti->error = "Error setting origin device";
4190 goto bad_origin_dev;
4193 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
4195 ti->error = "Error opening origin device";
4196 goto bad_origin_dev;
4198 tc->origin_dev = origin_dev;
4201 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
4203 ti->error = "Error opening pool device";
4206 tc->pool_dev = pool_dev;
4208 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
4209 ti->error = "Invalid device id";
4214 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
4216 ti->error = "Couldn't get pool mapped device";
4221 tc->pool = __pool_table_lookup(pool_md);
4223 ti->error = "Couldn't find pool object";
4225 goto bad_pool_lookup;
4227 __pool_inc(tc->pool);
4229 if (get_pool_mode(tc->pool) == PM_FAIL) {
4230 ti->error = "Couldn't open thin device, Pool is in fail mode";
4235 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
4237 ti->error = "Couldn't open thin internal device";
4241 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
4245 ti->num_flush_bios = 1;
4246 ti->flush_supported = true;
4247 ti->per_io_data_size = sizeof(struct dm_thin_endio_hook);
4249 /* In case the pool supports discards, pass them on. */
4250 ti->discard_zeroes_data_unsupported = true;
4251 if (tc->pool->pf.discard_enabled) {
4252 ti->discards_supported = true;
4253 ti->num_discard_bios = 1;
4254 ti->split_discard_bios = false;
4257 mutex_unlock(&dm_thin_pool_table.mutex);
4259 spin_lock_irqsave(&tc->pool->lock, flags);
4260 if (tc->pool->suspended) {
4261 spin_unlock_irqrestore(&tc->pool->lock, flags);
4262 mutex_lock(&dm_thin_pool_table.mutex); /* reacquire for __pool_dec */
4263 ti->error = "Unable to activate thin device while pool is suspended";
4267 atomic_set(&tc->refcount, 1);
4268 init_completion(&tc->can_destroy);
4269 list_add_tail_rcu(&tc->list, &tc->pool->active_thins);
4270 spin_unlock_irqrestore(&tc->pool->lock, flags);
4272 * This synchronize_rcu() call is needed here otherwise we risk a
4273 * wake_worker() call finding no bios to process (because the newly
4274 * added tc isn't yet visible). So this reduces latency since we
4275 * aren't then dependent on the periodic commit to wake_worker().
4284 dm_pool_close_thin_device(tc->td);
4286 __pool_dec(tc->pool);
4290 dm_put_device(ti, tc->pool_dev);
4293 dm_put_device(ti, tc->origin_dev);
4297 mutex_unlock(&dm_thin_pool_table.mutex);
4302 static int thin_map(struct dm_target *ti, struct bio *bio)
4304 bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
4306 return thin_bio_map(ti, bio);
4309 static int thin_endio(struct dm_target *ti, struct bio *bio, int err)
4311 unsigned long flags;
4312 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
4313 struct list_head work;
4314 struct dm_thin_new_mapping *m, *tmp;
4315 struct pool *pool = h->tc->pool;
4317 if (h->shared_read_entry) {
4318 INIT_LIST_HEAD(&work);
4319 dm_deferred_entry_dec(h->shared_read_entry, &work);
4321 spin_lock_irqsave(&pool->lock, flags);
4322 list_for_each_entry_safe(m, tmp, &work, list) {
4324 __complete_mapping_preparation(m);
4326 spin_unlock_irqrestore(&pool->lock, flags);
4329 if (h->all_io_entry) {
4330 INIT_LIST_HEAD(&work);
4331 dm_deferred_entry_dec(h->all_io_entry, &work);
4332 if (!list_empty(&work)) {
4333 spin_lock_irqsave(&pool->lock, flags);
4334 list_for_each_entry_safe(m, tmp, &work, list)
4335 list_add_tail(&m->list, &pool->prepared_discards);
4336 spin_unlock_irqrestore(&pool->lock, flags);
4342 cell_defer_no_holder(h->tc, h->cell);
4347 static void thin_presuspend(struct dm_target *ti)
4349 struct thin_c *tc = ti->private;
4351 if (dm_noflush_suspending(ti))
4352 noflush_work(tc, do_noflush_start);
4355 static void thin_postsuspend(struct dm_target *ti)
4357 struct thin_c *tc = ti->private;
4360 * The dm_noflush_suspending flag has been cleared by now, so
4361 * unfortunately we must always run this.
4363 noflush_work(tc, do_noflush_stop);
4366 static int thin_preresume(struct dm_target *ti)
4368 struct thin_c *tc = ti->private;
4371 tc->origin_size = get_dev_size(tc->origin_dev->bdev);
4377 * <nr mapped sectors> <highest mapped sector>
4379 static void thin_status(struct dm_target *ti, status_type_t type,
4380 unsigned status_flags, char *result, unsigned maxlen)
4384 dm_block_t mapped, highest;
4385 char buf[BDEVNAME_SIZE];
4386 struct thin_c *tc = ti->private;
4388 if (get_pool_mode(tc->pool) == PM_FAIL) {
4397 case STATUSTYPE_INFO:
4398 r = dm_thin_get_mapped_count(tc->td, &mapped);
4400 DMERR("dm_thin_get_mapped_count returned %d", r);
4404 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
4406 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
4410 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
4412 DMEMIT("%llu", ((highest + 1) *
4413 tc->pool->sectors_per_block) - 1);
4418 case STATUSTYPE_TABLE:
4420 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
4421 (unsigned long) tc->dev_id);
4423 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
4434 static int thin_iterate_devices(struct dm_target *ti,
4435 iterate_devices_callout_fn fn, void *data)
4438 struct thin_c *tc = ti->private;
4439 struct pool *pool = tc->pool;
4442 * We can't call dm_pool_get_data_dev_size() since that blocks. So
4443 * we follow a more convoluted path through to the pool's target.
4446 return 0; /* nothing is bound */
4448 blocks = pool->ti->len;
4449 (void) sector_div(blocks, pool->sectors_per_block);
4451 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
4456 static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits)
4458 struct thin_c *tc = ti->private;
4459 struct pool *pool = tc->pool;
4461 if (!pool->pf.discard_enabled)
4464 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
4465 limits->max_discard_sectors = 2048 * 1024 * 16; /* 16G */
4468 static struct target_type thin_target = {
4470 .version = {1, 19, 0},
4471 .module = THIS_MODULE,
4475 .end_io = thin_endio,
4476 .preresume = thin_preresume,
4477 .presuspend = thin_presuspend,
4478 .postsuspend = thin_postsuspend,
4479 .status = thin_status,
4480 .iterate_devices = thin_iterate_devices,
4481 .io_hints = thin_io_hints,
4484 /*----------------------------------------------------------------*/
4486 static int __init dm_thin_init(void)
4492 r = dm_register_target(&thin_target);
4496 r = dm_register_target(&pool_target);
4498 goto bad_pool_target;
4502 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
4503 if (!_new_mapping_cache)
4504 goto bad_new_mapping_cache;
4508 bad_new_mapping_cache:
4509 dm_unregister_target(&pool_target);
4511 dm_unregister_target(&thin_target);
4516 static void dm_thin_exit(void)
4518 dm_unregister_target(&thin_target);
4519 dm_unregister_target(&pool_target);
4521 kmem_cache_destroy(_new_mapping_cache);
4524 module_init(dm_thin_init);
4525 module_exit(dm_thin_exit);
4527 module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR);
4528 MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds");
4530 MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
4531 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
4532 MODULE_LICENSE("GPL");