2 * Copyright (C) 2011-2012 Red Hat UK.
4 * This file is released under the GPL.
7 #include "dm-thin-metadata.h"
8 #include "dm-bio-prison-v1.h"
11 #include <linux/device-mapper.h>
12 #include <linux/dm-io.h>
13 #include <linux/dm-kcopyd.h>
14 #include <linux/jiffies.h>
15 #include <linux/log2.h>
16 #include <linux/list.h>
17 #include <linux/rculist.h>
18 #include <linux/init.h>
19 #include <linux/module.h>
20 #include <linux/slab.h>
21 #include <linux/vmalloc.h>
22 #include <linux/sort.h>
23 #include <linux/rbtree.h>
25 #define DM_MSG_PREFIX "thin"
30 #define ENDIO_HOOK_POOL_SIZE 1024
31 #define MAPPING_POOL_SIZE 1024
32 #define COMMIT_PERIOD HZ
33 #define NO_SPACE_TIMEOUT_SECS 60
35 static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS;
37 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
38 "A percentage of time allocated for copy on write");
41 * The block size of the device holding pool data must be
42 * between 64KB and 1GB.
44 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
45 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
48 * Device id is restricted to 24 bits.
50 #define MAX_DEV_ID ((1 << 24) - 1)
53 * How do we handle breaking sharing of data blocks?
54 * =================================================
56 * We use a standard copy-on-write btree to store the mappings for the
57 * devices (note I'm talking about copy-on-write of the metadata here, not
58 * the data). When you take an internal snapshot you clone the root node
59 * of the origin btree. After this there is no concept of an origin or a
60 * snapshot. They are just two device trees that happen to point to the
63 * When we get a write in we decide if it's to a shared data block using
64 * some timestamp magic. If it is, we have to break sharing.
66 * Let's say we write to a shared block in what was the origin. The
69 * i) plug io further to this physical block. (see bio_prison code).
71 * ii) quiesce any read io to that shared data block. Obviously
72 * including all devices that share this block. (see dm_deferred_set code)
74 * iii) copy the data block to a newly allocate block. This step can be
75 * missed out if the io covers the block. (schedule_copy).
77 * iv) insert the new mapping into the origin's btree
78 * (process_prepared_mapping). This act of inserting breaks some
79 * sharing of btree nodes between the two devices. Breaking sharing only
80 * effects the btree of that specific device. Btrees for the other
81 * devices that share the block never change. The btree for the origin
82 * device as it was after the last commit is untouched, ie. we're using
83 * persistent data structures in the functional programming sense.
85 * v) unplug io to this physical block, including the io that triggered
86 * the breaking of sharing.
88 * Steps (ii) and (iii) occur in parallel.
90 * The metadata _doesn't_ need to be committed before the io continues. We
91 * get away with this because the io is always written to a _new_ block.
92 * If there's a crash, then:
94 * - The origin mapping will point to the old origin block (the shared
95 * one). This will contain the data as it was before the io that triggered
96 * the breaking of sharing came in.
98 * - The snap mapping still points to the old block. As it would after
101 * The downside of this scheme is the timestamp magic isn't perfect, and
102 * will continue to think that data block in the snapshot device is shared
103 * even after the write to the origin has broken sharing. I suspect data
104 * blocks will typically be shared by many different devices, so we're
105 * breaking sharing n + 1 times, rather than n, where n is the number of
106 * devices that reference this data block. At the moment I think the
107 * benefits far, far outweigh the disadvantages.
110 /*----------------------------------------------------------------*/
120 static void build_key(struct dm_thin_device *td, enum lock_space ls,
121 dm_block_t b, dm_block_t e, struct dm_cell_key *key)
123 key->virtual = (ls == VIRTUAL);
124 key->dev = dm_thin_dev_id(td);
125 key->block_begin = b;
129 static void build_data_key(struct dm_thin_device *td, dm_block_t b,
130 struct dm_cell_key *key)
132 build_key(td, PHYSICAL, b, b + 1llu, key);
135 static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
136 struct dm_cell_key *key)
138 build_key(td, VIRTUAL, b, b + 1llu, key);
141 /*----------------------------------------------------------------*/
143 #define THROTTLE_THRESHOLD (1 * HZ)
146 struct rw_semaphore lock;
147 unsigned long threshold;
148 bool throttle_applied;
151 static void throttle_init(struct throttle *t)
153 init_rwsem(&t->lock);
154 t->throttle_applied = false;
157 static void throttle_work_start(struct throttle *t)
159 t->threshold = jiffies + THROTTLE_THRESHOLD;
162 static void throttle_work_update(struct throttle *t)
164 if (!t->throttle_applied && jiffies > t->threshold) {
165 down_write(&t->lock);
166 t->throttle_applied = true;
170 static void throttle_work_complete(struct throttle *t)
172 if (t->throttle_applied) {
173 t->throttle_applied = false;
178 static void throttle_lock(struct throttle *t)
183 static void throttle_unlock(struct throttle *t)
188 /*----------------------------------------------------------------*/
191 * A pool device ties together a metadata device and a data device. It
192 * also provides the interface for creating and destroying internal
195 struct dm_thin_new_mapping;
198 * The pool runs in various modes. Ordered in degraded order for comparisons.
201 PM_WRITE, /* metadata may be changed */
202 PM_OUT_OF_DATA_SPACE, /* metadata may be changed, though data may not be allocated */
205 * Like READ_ONLY, except may switch back to WRITE on metadata resize. Reported as READ_ONLY.
207 PM_OUT_OF_METADATA_SPACE,
208 PM_READ_ONLY, /* metadata may not be changed */
210 PM_FAIL, /* all I/O fails */
213 struct pool_features {
216 bool zero_new_blocks:1;
217 bool discard_enabled:1;
218 bool discard_passdown:1;
219 bool error_if_no_space:1;
223 typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
224 typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell);
225 typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);
227 #define CELL_SORT_ARRAY_SIZE 8192
230 struct list_head list;
231 struct dm_target *ti; /* Only set if a pool target is bound */
233 struct mapped_device *pool_md;
234 struct block_device *data_dev;
235 struct block_device *md_dev;
236 struct dm_pool_metadata *pmd;
238 dm_block_t low_water_blocks;
239 uint32_t sectors_per_block;
240 int sectors_per_block_shift;
242 struct pool_features pf;
243 bool low_water_triggered:1; /* A dm event has been sent */
245 bool out_of_data_space:1;
247 struct dm_bio_prison *prison;
248 struct dm_kcopyd_client *copier;
250 struct work_struct worker;
251 struct workqueue_struct *wq;
252 struct throttle throttle;
253 struct delayed_work waker;
254 struct delayed_work no_space_timeout;
256 unsigned long last_commit_jiffies;
260 struct bio_list deferred_flush_bios;
261 struct bio_list deferred_flush_completions;
262 struct list_head prepared_mappings;
263 struct list_head prepared_discards;
264 struct list_head prepared_discards_pt2;
265 struct list_head active_thins;
267 struct dm_deferred_set *shared_read_ds;
268 struct dm_deferred_set *all_io_ds;
270 struct dm_thin_new_mapping *next_mapping;
272 process_bio_fn process_bio;
273 process_bio_fn process_discard;
275 process_cell_fn process_cell;
276 process_cell_fn process_discard_cell;
278 process_mapping_fn process_prepared_mapping;
279 process_mapping_fn process_prepared_discard;
280 process_mapping_fn process_prepared_discard_pt2;
282 struct dm_bio_prison_cell **cell_sort_array;
284 mempool_t mapping_pool;
286 struct bio flush_bio;
289 static void metadata_operation_failed(struct pool *pool, const char *op, int r);
291 static enum pool_mode get_pool_mode(struct pool *pool)
293 return pool->pf.mode;
296 static void notify_of_pool_mode_change(struct pool *pool)
298 const char *descs[] = {
305 const char *extra_desc = NULL;
306 enum pool_mode mode = get_pool_mode(pool);
308 if (mode == PM_OUT_OF_DATA_SPACE) {
309 if (!pool->pf.error_if_no_space)
310 extra_desc = " (queue IO)";
312 extra_desc = " (error IO)";
315 dm_table_event(pool->ti->table);
316 DMINFO("%s: switching pool to %s%s mode",
317 dm_device_name(pool->pool_md),
318 descs[(int)mode], extra_desc ? : "");
322 * Target context for a pool.
325 struct dm_target *ti;
327 struct dm_dev *data_dev;
328 struct dm_dev *metadata_dev;
330 dm_block_t low_water_blocks;
331 struct pool_features requested_pf; /* Features requested during table load */
332 struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */
336 * Target context for a thin.
339 struct list_head list;
340 struct dm_dev *pool_dev;
341 struct dm_dev *origin_dev;
342 sector_t origin_size;
346 struct dm_thin_device *td;
347 struct mapped_device *thin_md;
351 struct list_head deferred_cells;
352 struct bio_list deferred_bio_list;
353 struct bio_list retry_on_resume_list;
354 struct rb_root sort_bio_list; /* sorted list of deferred bios */
357 * Ensures the thin is not destroyed until the worker has finished
358 * iterating the active_thins list.
361 struct completion can_destroy;
364 /*----------------------------------------------------------------*/
366 static bool block_size_is_power_of_two(struct pool *pool)
368 return pool->sectors_per_block_shift >= 0;
371 static sector_t block_to_sectors(struct pool *pool, dm_block_t b)
373 return block_size_is_power_of_two(pool) ?
374 (b << pool->sectors_per_block_shift) :
375 (b * pool->sectors_per_block);
378 /*----------------------------------------------------------------*/
382 struct blk_plug plug;
383 struct bio *parent_bio;
387 static void begin_discard(struct discard_op *op, struct thin_c *tc, struct bio *parent)
392 blk_start_plug(&op->plug);
393 op->parent_bio = parent;
397 static int issue_discard(struct discard_op *op, dm_block_t data_b, dm_block_t data_e)
399 struct thin_c *tc = op->tc;
400 sector_t s = block_to_sectors(tc->pool, data_b);
401 sector_t len = block_to_sectors(tc->pool, data_e - data_b);
403 return __blkdev_issue_discard(tc->pool_dev->bdev, s, len,
404 GFP_NOWAIT, 0, &op->bio);
407 static void end_discard(struct discard_op *op, int r)
411 * Even if one of the calls to issue_discard failed, we
412 * need to wait for the chain to complete.
414 bio_chain(op->bio, op->parent_bio);
415 bio_set_op_attrs(op->bio, REQ_OP_DISCARD, 0);
419 blk_finish_plug(&op->plug);
422 * Even if r is set, there could be sub discards in flight that we
425 if (r && !op->parent_bio->bi_status)
426 op->parent_bio->bi_status = errno_to_blk_status(r);
427 bio_endio(op->parent_bio);
430 /*----------------------------------------------------------------*/
433 * wake_worker() is used when new work is queued and when pool_resume is
434 * ready to continue deferred IO processing.
436 static void wake_worker(struct pool *pool)
438 queue_work(pool->wq, &pool->worker);
441 /*----------------------------------------------------------------*/
443 static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio,
444 struct dm_bio_prison_cell **cell_result)
447 struct dm_bio_prison_cell *cell_prealloc;
450 * Allocate a cell from the prison's mempool.
451 * This might block but it can't fail.
453 cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO);
455 r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result);
458 * We reused an old cell; we can get rid of
461 dm_bio_prison_free_cell(pool->prison, cell_prealloc);
466 static void cell_release(struct pool *pool,
467 struct dm_bio_prison_cell *cell,
468 struct bio_list *bios)
470 dm_cell_release(pool->prison, cell, bios);
471 dm_bio_prison_free_cell(pool->prison, cell);
474 static void cell_visit_release(struct pool *pool,
475 void (*fn)(void *, struct dm_bio_prison_cell *),
477 struct dm_bio_prison_cell *cell)
479 dm_cell_visit_release(pool->prison, fn, context, cell);
480 dm_bio_prison_free_cell(pool->prison, cell);
483 static void cell_release_no_holder(struct pool *pool,
484 struct dm_bio_prison_cell *cell,
485 struct bio_list *bios)
487 dm_cell_release_no_holder(pool->prison, cell, bios);
488 dm_bio_prison_free_cell(pool->prison, cell);
491 static void cell_error_with_code(struct pool *pool,
492 struct dm_bio_prison_cell *cell, blk_status_t error_code)
494 dm_cell_error(pool->prison, cell, error_code);
495 dm_bio_prison_free_cell(pool->prison, cell);
498 static blk_status_t get_pool_io_error_code(struct pool *pool)
500 return pool->out_of_data_space ? BLK_STS_NOSPC : BLK_STS_IOERR;
503 static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell)
505 cell_error_with_code(pool, cell, get_pool_io_error_code(pool));
508 static void cell_success(struct pool *pool, struct dm_bio_prison_cell *cell)
510 cell_error_with_code(pool, cell, 0);
513 static void cell_requeue(struct pool *pool, struct dm_bio_prison_cell *cell)
515 cell_error_with_code(pool, cell, BLK_STS_DM_REQUEUE);
518 /*----------------------------------------------------------------*/
521 * A global list of pools that uses a struct mapped_device as a key.
523 static struct dm_thin_pool_table {
525 struct list_head pools;
526 } dm_thin_pool_table;
528 static void pool_table_init(void)
530 mutex_init(&dm_thin_pool_table.mutex);
531 INIT_LIST_HEAD(&dm_thin_pool_table.pools);
534 static void pool_table_exit(void)
536 mutex_destroy(&dm_thin_pool_table.mutex);
539 static void __pool_table_insert(struct pool *pool)
541 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
542 list_add(&pool->list, &dm_thin_pool_table.pools);
545 static void __pool_table_remove(struct pool *pool)
547 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
548 list_del(&pool->list);
551 static struct pool *__pool_table_lookup(struct mapped_device *md)
553 struct pool *pool = NULL, *tmp;
555 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
557 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
558 if (tmp->pool_md == md) {
567 static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
569 struct pool *pool = NULL, *tmp;
571 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
573 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
574 if (tmp->md_dev == md_dev) {
583 /*----------------------------------------------------------------*/
585 struct dm_thin_endio_hook {
587 struct dm_deferred_entry *shared_read_entry;
588 struct dm_deferred_entry *all_io_entry;
589 struct dm_thin_new_mapping *overwrite_mapping;
590 struct rb_node rb_node;
591 struct dm_bio_prison_cell *cell;
594 static void __merge_bio_list(struct bio_list *bios, struct bio_list *master)
596 bio_list_merge(bios, master);
597 bio_list_init(master);
600 static void error_bio_list(struct bio_list *bios, blk_status_t error)
604 while ((bio = bio_list_pop(bios))) {
605 bio->bi_status = error;
610 static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master,
613 struct bio_list bios;
615 bio_list_init(&bios);
617 spin_lock_irq(&tc->lock);
618 __merge_bio_list(&bios, master);
619 spin_unlock_irq(&tc->lock);
621 error_bio_list(&bios, error);
624 static void requeue_deferred_cells(struct thin_c *tc)
626 struct pool *pool = tc->pool;
627 struct list_head cells;
628 struct dm_bio_prison_cell *cell, *tmp;
630 INIT_LIST_HEAD(&cells);
632 spin_lock_irq(&tc->lock);
633 list_splice_init(&tc->deferred_cells, &cells);
634 spin_unlock_irq(&tc->lock);
636 list_for_each_entry_safe(cell, tmp, &cells, user_list)
637 cell_requeue(pool, cell);
640 static void requeue_io(struct thin_c *tc)
642 struct bio_list bios;
644 bio_list_init(&bios);
646 spin_lock_irq(&tc->lock);
647 __merge_bio_list(&bios, &tc->deferred_bio_list);
648 __merge_bio_list(&bios, &tc->retry_on_resume_list);
649 spin_unlock_irq(&tc->lock);
651 error_bio_list(&bios, BLK_STS_DM_REQUEUE);
652 requeue_deferred_cells(tc);
655 static void error_retry_list_with_code(struct pool *pool, blk_status_t error)
660 list_for_each_entry_rcu(tc, &pool->active_thins, list)
661 error_thin_bio_list(tc, &tc->retry_on_resume_list, error);
665 static void error_retry_list(struct pool *pool)
667 error_retry_list_with_code(pool, get_pool_io_error_code(pool));
671 * This section of code contains the logic for processing a thin device's IO.
672 * Much of the code depends on pool object resources (lists, workqueues, etc)
673 * but most is exclusively called from the thin target rather than the thin-pool
677 static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
679 struct pool *pool = tc->pool;
680 sector_t block_nr = bio->bi_iter.bi_sector;
682 if (block_size_is_power_of_two(pool))
683 block_nr >>= pool->sectors_per_block_shift;
685 (void) sector_div(block_nr, pool->sectors_per_block);
691 * Returns the _complete_ blocks that this bio covers.
693 static void get_bio_block_range(struct thin_c *tc, struct bio *bio,
694 dm_block_t *begin, dm_block_t *end)
696 struct pool *pool = tc->pool;
697 sector_t b = bio->bi_iter.bi_sector;
698 sector_t e = b + (bio->bi_iter.bi_size >> SECTOR_SHIFT);
700 b += pool->sectors_per_block - 1ull; /* so we round up */
702 if (block_size_is_power_of_two(pool)) {
703 b >>= pool->sectors_per_block_shift;
704 e >>= pool->sectors_per_block_shift;
706 (void) sector_div(b, pool->sectors_per_block);
707 (void) sector_div(e, pool->sectors_per_block);
711 /* Can happen if the bio is within a single block. */
718 static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
720 struct pool *pool = tc->pool;
721 sector_t bi_sector = bio->bi_iter.bi_sector;
723 bio_set_dev(bio, tc->pool_dev->bdev);
724 if (block_size_is_power_of_two(pool))
725 bio->bi_iter.bi_sector =
726 (block << pool->sectors_per_block_shift) |
727 (bi_sector & (pool->sectors_per_block - 1));
729 bio->bi_iter.bi_sector = (block * pool->sectors_per_block) +
730 sector_div(bi_sector, pool->sectors_per_block);
733 static void remap_to_origin(struct thin_c *tc, struct bio *bio)
735 bio_set_dev(bio, tc->origin_dev->bdev);
738 static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
740 return op_is_flush(bio->bi_opf) &&
741 dm_thin_changed_this_transaction(tc->td);
744 static void inc_all_io_entry(struct pool *pool, struct bio *bio)
746 struct dm_thin_endio_hook *h;
748 if (bio_op(bio) == REQ_OP_DISCARD)
751 h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
752 h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
755 static void issue(struct thin_c *tc, struct bio *bio)
757 struct pool *pool = tc->pool;
759 if (!bio_triggers_commit(tc, bio)) {
760 submit_bio_noacct(bio);
765 * Complete bio with an error if earlier I/O caused changes to
766 * the metadata that can't be committed e.g, due to I/O errors
767 * on the metadata device.
769 if (dm_thin_aborted_changes(tc->td)) {
775 * Batch together any bios that trigger commits and then issue a
776 * single commit for them in process_deferred_bios().
778 spin_lock_irq(&pool->lock);
779 bio_list_add(&pool->deferred_flush_bios, bio);
780 spin_unlock_irq(&pool->lock);
783 static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
785 remap_to_origin(tc, bio);
789 static void remap_and_issue(struct thin_c *tc, struct bio *bio,
792 remap(tc, bio, block);
796 /*----------------------------------------------------------------*/
799 * Bio endio functions.
801 struct dm_thin_new_mapping {
802 struct list_head list;
808 * Track quiescing, copying and zeroing preparation actions. When this
809 * counter hits zero the block is prepared and can be inserted into the
812 atomic_t prepare_actions;
816 dm_block_t virt_begin, virt_end;
817 dm_block_t data_block;
818 struct dm_bio_prison_cell *cell;
821 * If the bio covers the whole area of a block then we can avoid
822 * zeroing or copying. Instead this bio is hooked. The bio will
823 * still be in the cell, so care has to be taken to avoid issuing
827 bio_end_io_t *saved_bi_end_io;
830 static void __complete_mapping_preparation(struct dm_thin_new_mapping *m)
832 struct pool *pool = m->tc->pool;
834 if (atomic_dec_and_test(&m->prepare_actions)) {
835 list_add_tail(&m->list, &pool->prepared_mappings);
840 static void complete_mapping_preparation(struct dm_thin_new_mapping *m)
843 struct pool *pool = m->tc->pool;
845 spin_lock_irqsave(&pool->lock, flags);
846 __complete_mapping_preparation(m);
847 spin_unlock_irqrestore(&pool->lock, flags);
850 static void copy_complete(int read_err, unsigned long write_err, void *context)
852 struct dm_thin_new_mapping *m = context;
854 m->status = read_err || write_err ? BLK_STS_IOERR : 0;
855 complete_mapping_preparation(m);
858 static void overwrite_endio(struct bio *bio)
860 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
861 struct dm_thin_new_mapping *m = h->overwrite_mapping;
863 bio->bi_end_io = m->saved_bi_end_io;
865 m->status = bio->bi_status;
866 complete_mapping_preparation(m);
869 /*----------------------------------------------------------------*/
876 * Prepared mapping jobs.
880 * This sends the bios in the cell, except the original holder, back
881 * to the deferred_bios list.
883 static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
885 struct pool *pool = tc->pool;
889 spin_lock_irqsave(&tc->lock, flags);
890 cell_release_no_holder(pool, cell, &tc->deferred_bio_list);
891 has_work = !bio_list_empty(&tc->deferred_bio_list);
892 spin_unlock_irqrestore(&tc->lock, flags);
898 static void thin_defer_bio(struct thin_c *tc, struct bio *bio);
902 struct bio_list defer_bios;
903 struct bio_list issue_bios;
906 static void __inc_remap_and_issue_cell(void *context,
907 struct dm_bio_prison_cell *cell)
909 struct remap_info *info = context;
912 while ((bio = bio_list_pop(&cell->bios))) {
913 if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD)
914 bio_list_add(&info->defer_bios, bio);
916 inc_all_io_entry(info->tc->pool, bio);
919 * We can't issue the bios with the bio prison lock
920 * held, so we add them to a list to issue on
921 * return from this function.
923 bio_list_add(&info->issue_bios, bio);
928 static void inc_remap_and_issue_cell(struct thin_c *tc,
929 struct dm_bio_prison_cell *cell,
933 struct remap_info info;
936 bio_list_init(&info.defer_bios);
937 bio_list_init(&info.issue_bios);
940 * We have to be careful to inc any bios we're about to issue
941 * before the cell is released, and avoid a race with new bios
942 * being added to the cell.
944 cell_visit_release(tc->pool, __inc_remap_and_issue_cell,
947 while ((bio = bio_list_pop(&info.defer_bios)))
948 thin_defer_bio(tc, bio);
950 while ((bio = bio_list_pop(&info.issue_bios)))
951 remap_and_issue(info.tc, bio, block);
954 static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
956 cell_error(m->tc->pool, m->cell);
958 mempool_free(m, &m->tc->pool->mapping_pool);
961 static void complete_overwrite_bio(struct thin_c *tc, struct bio *bio)
963 struct pool *pool = tc->pool;
966 * If the bio has the REQ_FUA flag set we must commit the metadata
967 * before signaling its completion.
969 if (!bio_triggers_commit(tc, bio)) {
975 * Complete bio with an error if earlier I/O caused changes to the
976 * metadata that can't be committed, e.g, due to I/O errors on the
979 if (dm_thin_aborted_changes(tc->td)) {
985 * Batch together any bios that trigger commits and then issue a
986 * single commit for them in process_deferred_bios().
988 spin_lock_irq(&pool->lock);
989 bio_list_add(&pool->deferred_flush_completions, bio);
990 spin_unlock_irq(&pool->lock);
993 static void process_prepared_mapping(struct dm_thin_new_mapping *m)
995 struct thin_c *tc = m->tc;
996 struct pool *pool = tc->pool;
997 struct bio *bio = m->bio;
1001 cell_error(pool, m->cell);
1006 * Commit the prepared block into the mapping btree.
1007 * Any I/O for this block arriving after this point will get
1008 * remapped to it directly.
1010 r = dm_thin_insert_block(tc->td, m->virt_begin, m->data_block);
1012 metadata_operation_failed(pool, "dm_thin_insert_block", r);
1013 cell_error(pool, m->cell);
1018 * Release any bios held while the block was being provisioned.
1019 * If we are processing a write bio that completely covers the block,
1020 * we already processed it so can ignore it now when processing
1021 * the bios in the cell.
1024 inc_remap_and_issue_cell(tc, m->cell, m->data_block);
1025 complete_overwrite_bio(tc, bio);
1027 inc_all_io_entry(tc->pool, m->cell->holder);
1028 remap_and_issue(tc, m->cell->holder, m->data_block);
1029 inc_remap_and_issue_cell(tc, m->cell, m->data_block);
1034 mempool_free(m, &pool->mapping_pool);
1037 /*----------------------------------------------------------------*/
1039 static void free_discard_mapping(struct dm_thin_new_mapping *m)
1041 struct thin_c *tc = m->tc;
1043 cell_defer_no_holder(tc, m->cell);
1044 mempool_free(m, &tc->pool->mapping_pool);
1047 static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
1049 bio_io_error(m->bio);
1050 free_discard_mapping(m);
1053 static void process_prepared_discard_success(struct dm_thin_new_mapping *m)
1056 free_discard_mapping(m);
1059 static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m)
1062 struct thin_c *tc = m->tc;
1064 r = dm_thin_remove_range(tc->td, m->cell->key.block_begin, m->cell->key.block_end);
1066 metadata_operation_failed(tc->pool, "dm_thin_remove_range", r);
1067 bio_io_error(m->bio);
1071 cell_defer_no_holder(tc, m->cell);
1072 mempool_free(m, &tc->pool->mapping_pool);
1075 /*----------------------------------------------------------------*/
1077 static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m,
1078 struct bio *discard_parent)
1081 * We've already unmapped this range of blocks, but before we
1082 * passdown we have to check that these blocks are now unused.
1086 struct thin_c *tc = m->tc;
1087 struct pool *pool = tc->pool;
1088 dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin;
1089 struct discard_op op;
1091 begin_discard(&op, tc, discard_parent);
1093 /* find start of unmapped run */
1094 for (; b < end; b++) {
1095 r = dm_pool_block_is_shared(pool->pmd, b, &shared);
1106 /* find end of run */
1107 for (e = b + 1; e != end; e++) {
1108 r = dm_pool_block_is_shared(pool->pmd, e, &shared);
1116 r = issue_discard(&op, b, e);
1123 end_discard(&op, r);
1126 static void queue_passdown_pt2(struct dm_thin_new_mapping *m)
1128 unsigned long flags;
1129 struct pool *pool = m->tc->pool;
1131 spin_lock_irqsave(&pool->lock, flags);
1132 list_add_tail(&m->list, &pool->prepared_discards_pt2);
1133 spin_unlock_irqrestore(&pool->lock, flags);
1137 static void passdown_endio(struct bio *bio)
1140 * It doesn't matter if the passdown discard failed, we still want
1141 * to unmap (we ignore err).
1143 queue_passdown_pt2(bio->bi_private);
1147 static void process_prepared_discard_passdown_pt1(struct dm_thin_new_mapping *m)
1150 struct thin_c *tc = m->tc;
1151 struct pool *pool = tc->pool;
1152 struct bio *discard_parent;
1153 dm_block_t data_end = m->data_block + (m->virt_end - m->virt_begin);
1156 * Only this thread allocates blocks, so we can be sure that the
1157 * newly unmapped blocks will not be allocated before the end of
1160 r = dm_thin_remove_range(tc->td, m->virt_begin, m->virt_end);
1162 metadata_operation_failed(pool, "dm_thin_remove_range", r);
1163 bio_io_error(m->bio);
1164 cell_defer_no_holder(tc, m->cell);
1165 mempool_free(m, &pool->mapping_pool);
1170 * Increment the unmapped blocks. This prevents a race between the
1171 * passdown io and reallocation of freed blocks.
1173 r = dm_pool_inc_data_range(pool->pmd, m->data_block, data_end);
1175 metadata_operation_failed(pool, "dm_pool_inc_data_range", r);
1176 bio_io_error(m->bio);
1177 cell_defer_no_holder(tc, m->cell);
1178 mempool_free(m, &pool->mapping_pool);
1182 discard_parent = bio_alloc(GFP_NOIO, 1);
1183 if (!discard_parent) {
1184 DMWARN("%s: unable to allocate top level discard bio for passdown. Skipping passdown.",
1185 dm_device_name(tc->pool->pool_md));
1186 queue_passdown_pt2(m);
1189 discard_parent->bi_end_io = passdown_endio;
1190 discard_parent->bi_private = m;
1192 if (m->maybe_shared)
1193 passdown_double_checking_shared_status(m, discard_parent);
1195 struct discard_op op;
1197 begin_discard(&op, tc, discard_parent);
1198 r = issue_discard(&op, m->data_block, data_end);
1199 end_discard(&op, r);
1204 static void process_prepared_discard_passdown_pt2(struct dm_thin_new_mapping *m)
1207 struct thin_c *tc = m->tc;
1208 struct pool *pool = tc->pool;
1211 * The passdown has completed, so now we can decrement all those
1214 r = dm_pool_dec_data_range(pool->pmd, m->data_block,
1215 m->data_block + (m->virt_end - m->virt_begin));
1217 metadata_operation_failed(pool, "dm_pool_dec_data_range", r);
1218 bio_io_error(m->bio);
1222 cell_defer_no_holder(tc, m->cell);
1223 mempool_free(m, &pool->mapping_pool);
1226 static void process_prepared(struct pool *pool, struct list_head *head,
1227 process_mapping_fn *fn)
1229 struct list_head maps;
1230 struct dm_thin_new_mapping *m, *tmp;
1232 INIT_LIST_HEAD(&maps);
1233 spin_lock_irq(&pool->lock);
1234 list_splice_init(head, &maps);
1235 spin_unlock_irq(&pool->lock);
1237 list_for_each_entry_safe(m, tmp, &maps, list)
1242 * Deferred bio jobs.
1244 static int io_overlaps_block(struct pool *pool, struct bio *bio)
1246 return bio->bi_iter.bi_size ==
1247 (pool->sectors_per_block << SECTOR_SHIFT);
1250 static int io_overwrites_block(struct pool *pool, struct bio *bio)
1252 return (bio_data_dir(bio) == WRITE) &&
1253 io_overlaps_block(pool, bio);
1256 static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
1259 *save = bio->bi_end_io;
1260 bio->bi_end_io = fn;
1263 static int ensure_next_mapping(struct pool *pool)
1265 if (pool->next_mapping)
1268 pool->next_mapping = mempool_alloc(&pool->mapping_pool, GFP_ATOMIC);
1270 return pool->next_mapping ? 0 : -ENOMEM;
1273 static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
1275 struct dm_thin_new_mapping *m = pool->next_mapping;
1277 BUG_ON(!pool->next_mapping);
1279 memset(m, 0, sizeof(struct dm_thin_new_mapping));
1280 INIT_LIST_HEAD(&m->list);
1283 pool->next_mapping = NULL;
1288 static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m,
1289 sector_t begin, sector_t end)
1291 struct dm_io_region to;
1293 to.bdev = tc->pool_dev->bdev;
1295 to.count = end - begin;
1297 dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m);
1300 static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio,
1301 dm_block_t data_begin,
1302 struct dm_thin_new_mapping *m)
1304 struct pool *pool = tc->pool;
1305 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1307 h->overwrite_mapping = m;
1309 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
1310 inc_all_io_entry(pool, bio);
1311 remap_and_issue(tc, bio, data_begin);
1315 * A partial copy also needs to zero the uncopied region.
1317 static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
1318 struct dm_dev *origin, dm_block_t data_origin,
1319 dm_block_t data_dest,
1320 struct dm_bio_prison_cell *cell, struct bio *bio,
1323 struct pool *pool = tc->pool;
1324 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1327 m->virt_begin = virt_block;
1328 m->virt_end = virt_block + 1u;
1329 m->data_block = data_dest;
1333 * quiesce action + copy action + an extra reference held for the
1334 * duration of this function (we may need to inc later for a
1337 atomic_set(&m->prepare_actions, 3);
1339 if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
1340 complete_mapping_preparation(m); /* already quiesced */
1343 * IO to pool_dev remaps to the pool target's data_dev.
1345 * If the whole block of data is being overwritten, we can issue the
1346 * bio immediately. Otherwise we use kcopyd to clone the data first.
1348 if (io_overwrites_block(pool, bio))
1349 remap_and_issue_overwrite(tc, bio, data_dest, m);
1351 struct dm_io_region from, to;
1353 from.bdev = origin->bdev;
1354 from.sector = data_origin * pool->sectors_per_block;
1357 to.bdev = tc->pool_dev->bdev;
1358 to.sector = data_dest * pool->sectors_per_block;
1361 dm_kcopyd_copy(pool->copier, &from, 1, &to,
1362 0, copy_complete, m);
1365 * Do we need to zero a tail region?
1367 if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) {
1368 atomic_inc(&m->prepare_actions);
1370 data_dest * pool->sectors_per_block + len,
1371 (data_dest + 1) * pool->sectors_per_block);
1375 complete_mapping_preparation(m); /* drop our ref */
1378 static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
1379 dm_block_t data_origin, dm_block_t data_dest,
1380 struct dm_bio_prison_cell *cell, struct bio *bio)
1382 schedule_copy(tc, virt_block, tc->pool_dev,
1383 data_origin, data_dest, cell, bio,
1384 tc->pool->sectors_per_block);
1387 static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
1388 dm_block_t data_block, struct dm_bio_prison_cell *cell,
1391 struct pool *pool = tc->pool;
1392 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1394 atomic_set(&m->prepare_actions, 1); /* no need to quiesce */
1396 m->virt_begin = virt_block;
1397 m->virt_end = virt_block + 1u;
1398 m->data_block = data_block;
1402 * If the whole block of data is being overwritten or we are not
1403 * zeroing pre-existing data, we can issue the bio immediately.
1404 * Otherwise we use kcopyd to zero the data first.
1406 if (pool->pf.zero_new_blocks) {
1407 if (io_overwrites_block(pool, bio))
1408 remap_and_issue_overwrite(tc, bio, data_block, m);
1410 ll_zero(tc, m, data_block * pool->sectors_per_block,
1411 (data_block + 1) * pool->sectors_per_block);
1413 process_prepared_mapping(m);
1416 static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
1417 dm_block_t data_dest,
1418 struct dm_bio_prison_cell *cell, struct bio *bio)
1420 struct pool *pool = tc->pool;
1421 sector_t virt_block_begin = virt_block * pool->sectors_per_block;
1422 sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block;
1424 if (virt_block_end <= tc->origin_size)
1425 schedule_copy(tc, virt_block, tc->origin_dev,
1426 virt_block, data_dest, cell, bio,
1427 pool->sectors_per_block);
1429 else if (virt_block_begin < tc->origin_size)
1430 schedule_copy(tc, virt_block, tc->origin_dev,
1431 virt_block, data_dest, cell, bio,
1432 tc->origin_size - virt_block_begin);
1435 schedule_zero(tc, virt_block, data_dest, cell, bio);
1438 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode);
1440 static void requeue_bios(struct pool *pool);
1442 static bool is_read_only_pool_mode(enum pool_mode mode)
1444 return (mode == PM_OUT_OF_METADATA_SPACE || mode == PM_READ_ONLY);
1447 static bool is_read_only(struct pool *pool)
1449 return is_read_only_pool_mode(get_pool_mode(pool));
1452 static void check_for_metadata_space(struct pool *pool)
1455 const char *ooms_reason = NULL;
1458 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free);
1460 ooms_reason = "Could not get free metadata blocks";
1462 ooms_reason = "No free metadata blocks";
1464 if (ooms_reason && !is_read_only(pool)) {
1465 DMERR("%s", ooms_reason);
1466 set_pool_mode(pool, PM_OUT_OF_METADATA_SPACE);
1470 static void check_for_data_space(struct pool *pool)
1475 if (get_pool_mode(pool) != PM_OUT_OF_DATA_SPACE)
1478 r = dm_pool_get_free_block_count(pool->pmd, &nr_free);
1483 set_pool_mode(pool, PM_WRITE);
1489 * A non-zero return indicates read_only or fail_io mode.
1490 * Many callers don't care about the return value.
1492 static int commit(struct pool *pool)
1496 if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE)
1499 r = dm_pool_commit_metadata(pool->pmd);
1501 metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
1503 check_for_metadata_space(pool);
1504 check_for_data_space(pool);
1510 static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks)
1512 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
1513 DMWARN("%s: reached low water mark for data device: sending event.",
1514 dm_device_name(pool->pool_md));
1515 spin_lock_irq(&pool->lock);
1516 pool->low_water_triggered = true;
1517 spin_unlock_irq(&pool->lock);
1518 dm_table_event(pool->ti->table);
1522 static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
1525 dm_block_t free_blocks;
1526 struct pool *pool = tc->pool;
1528 if (WARN_ON(get_pool_mode(pool) != PM_WRITE))
1531 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1533 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1537 check_low_water_mark(pool, free_blocks);
1541 * Try to commit to see if that will free up some
1548 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1550 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1555 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1560 r = dm_pool_alloc_data_block(pool->pmd, result);
1563 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1565 metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
1569 r = dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks);
1571 metadata_operation_failed(pool, "dm_pool_get_free_metadata_block_count", r);
1576 /* Let's commit before we use up the metadata reserve. */
1586 * If we have run out of space, queue bios until the device is
1587 * resumed, presumably after having been reloaded with more space.
1589 static void retry_on_resume(struct bio *bio)
1591 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1592 struct thin_c *tc = h->tc;
1594 spin_lock_irq(&tc->lock);
1595 bio_list_add(&tc->retry_on_resume_list, bio);
1596 spin_unlock_irq(&tc->lock);
1599 static blk_status_t should_error_unserviceable_bio(struct pool *pool)
1601 enum pool_mode m = get_pool_mode(pool);
1605 /* Shouldn't get here */
1606 DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
1607 return BLK_STS_IOERR;
1609 case PM_OUT_OF_DATA_SPACE:
1610 return pool->pf.error_if_no_space ? BLK_STS_NOSPC : 0;
1612 case PM_OUT_OF_METADATA_SPACE:
1615 return BLK_STS_IOERR;
1617 /* Shouldn't get here */
1618 DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1619 return BLK_STS_IOERR;
1623 static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
1625 blk_status_t error = should_error_unserviceable_bio(pool);
1628 bio->bi_status = error;
1631 retry_on_resume(bio);
1634 static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell)
1637 struct bio_list bios;
1640 error = should_error_unserviceable_bio(pool);
1642 cell_error_with_code(pool, cell, error);
1646 bio_list_init(&bios);
1647 cell_release(pool, cell, &bios);
1649 while ((bio = bio_list_pop(&bios)))
1650 retry_on_resume(bio);
1653 static void process_discard_cell_no_passdown(struct thin_c *tc,
1654 struct dm_bio_prison_cell *virt_cell)
1656 struct pool *pool = tc->pool;
1657 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1660 * We don't need to lock the data blocks, since there's no
1661 * passdown. We only lock data blocks for allocation and breaking sharing.
1664 m->virt_begin = virt_cell->key.block_begin;
1665 m->virt_end = virt_cell->key.block_end;
1666 m->cell = virt_cell;
1667 m->bio = virt_cell->holder;
1669 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1670 pool->process_prepared_discard(m);
1673 static void break_up_discard_bio(struct thin_c *tc, dm_block_t begin, dm_block_t end,
1676 struct pool *pool = tc->pool;
1680 struct dm_cell_key data_key;
1681 struct dm_bio_prison_cell *data_cell;
1682 struct dm_thin_new_mapping *m;
1683 dm_block_t virt_begin, virt_end, data_begin;
1685 while (begin != end) {
1686 r = ensure_next_mapping(pool);
1688 /* we did our best */
1691 r = dm_thin_find_mapped_range(tc->td, begin, end, &virt_begin, &virt_end,
1692 &data_begin, &maybe_shared);
1695 * Silently fail, letting any mappings we've
1700 build_key(tc->td, PHYSICAL, data_begin, data_begin + (virt_end - virt_begin), &data_key);
1701 if (bio_detain(tc->pool, &data_key, NULL, &data_cell)) {
1702 /* contention, we'll give up with this range */
1708 * IO may still be going to the destination block. We must
1709 * quiesce before we can do the removal.
1711 m = get_next_mapping(pool);
1713 m->maybe_shared = maybe_shared;
1714 m->virt_begin = virt_begin;
1715 m->virt_end = virt_end;
1716 m->data_block = data_begin;
1717 m->cell = data_cell;
1721 * The parent bio must not complete before sub discard bios are
1722 * chained to it (see end_discard's bio_chain)!
1724 * This per-mapping bi_remaining increment is paired with
1725 * the implicit decrement that occurs via bio_endio() in
1728 bio_inc_remaining(bio);
1729 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1730 pool->process_prepared_discard(m);
1736 static void process_discard_cell_passdown(struct thin_c *tc, struct dm_bio_prison_cell *virt_cell)
1738 struct bio *bio = virt_cell->holder;
1739 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1742 * The virt_cell will only get freed once the origin bio completes.
1743 * This means it will remain locked while all the individual
1744 * passdown bios are in flight.
1746 h->cell = virt_cell;
1747 break_up_discard_bio(tc, virt_cell->key.block_begin, virt_cell->key.block_end, bio);
1750 * We complete the bio now, knowing that the bi_remaining field
1751 * will prevent completion until the sub range discards have
1757 static void process_discard_bio(struct thin_c *tc, struct bio *bio)
1759 dm_block_t begin, end;
1760 struct dm_cell_key virt_key;
1761 struct dm_bio_prison_cell *virt_cell;
1763 get_bio_block_range(tc, bio, &begin, &end);
1766 * The discard covers less than a block.
1772 build_key(tc->td, VIRTUAL, begin, end, &virt_key);
1773 if (bio_detain(tc->pool, &virt_key, bio, &virt_cell))
1775 * Potential starvation issue: We're relying on the
1776 * fs/application being well behaved, and not trying to
1777 * send IO to a region at the same time as discarding it.
1778 * If they do this persistently then it's possible this
1779 * cell will never be granted.
1783 tc->pool->process_discard_cell(tc, virt_cell);
1786 static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
1787 struct dm_cell_key *key,
1788 struct dm_thin_lookup_result *lookup_result,
1789 struct dm_bio_prison_cell *cell)
1792 dm_block_t data_block;
1793 struct pool *pool = tc->pool;
1795 r = alloc_data_block(tc, &data_block);
1798 schedule_internal_copy(tc, block, lookup_result->block,
1799 data_block, cell, bio);
1803 retry_bios_on_resume(pool, cell);
1807 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1809 cell_error(pool, cell);
1814 static void __remap_and_issue_shared_cell(void *context,
1815 struct dm_bio_prison_cell *cell)
1817 struct remap_info *info = context;
1820 while ((bio = bio_list_pop(&cell->bios))) {
1821 if (bio_data_dir(bio) == WRITE || op_is_flush(bio->bi_opf) ||
1822 bio_op(bio) == REQ_OP_DISCARD)
1823 bio_list_add(&info->defer_bios, bio);
1825 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1827 h->shared_read_entry = dm_deferred_entry_inc(info->tc->pool->shared_read_ds);
1828 inc_all_io_entry(info->tc->pool, bio);
1829 bio_list_add(&info->issue_bios, bio);
1834 static void remap_and_issue_shared_cell(struct thin_c *tc,
1835 struct dm_bio_prison_cell *cell,
1839 struct remap_info info;
1842 bio_list_init(&info.defer_bios);
1843 bio_list_init(&info.issue_bios);
1845 cell_visit_release(tc->pool, __remap_and_issue_shared_cell,
1848 while ((bio = bio_list_pop(&info.defer_bios)))
1849 thin_defer_bio(tc, bio);
1851 while ((bio = bio_list_pop(&info.issue_bios)))
1852 remap_and_issue(tc, bio, block);
1855 static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1857 struct dm_thin_lookup_result *lookup_result,
1858 struct dm_bio_prison_cell *virt_cell)
1860 struct dm_bio_prison_cell *data_cell;
1861 struct pool *pool = tc->pool;
1862 struct dm_cell_key key;
1865 * If cell is already occupied, then sharing is already in the process
1866 * of being broken so we have nothing further to do here.
1868 build_data_key(tc->td, lookup_result->block, &key);
1869 if (bio_detain(pool, &key, bio, &data_cell)) {
1870 cell_defer_no_holder(tc, virt_cell);
1874 if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) {
1875 break_sharing(tc, bio, block, &key, lookup_result, data_cell);
1876 cell_defer_no_holder(tc, virt_cell);
1878 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1880 h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
1881 inc_all_io_entry(pool, bio);
1882 remap_and_issue(tc, bio, lookup_result->block);
1884 remap_and_issue_shared_cell(tc, data_cell, lookup_result->block);
1885 remap_and_issue_shared_cell(tc, virt_cell, lookup_result->block);
1889 static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
1890 struct dm_bio_prison_cell *cell)
1893 dm_block_t data_block;
1894 struct pool *pool = tc->pool;
1897 * Remap empty bios (flushes) immediately, without provisioning.
1899 if (!bio->bi_iter.bi_size) {
1900 inc_all_io_entry(pool, bio);
1901 cell_defer_no_holder(tc, cell);
1903 remap_and_issue(tc, bio, 0);
1908 * Fill read bios with zeroes and complete them immediately.
1910 if (bio_data_dir(bio) == READ) {
1912 cell_defer_no_holder(tc, cell);
1917 r = alloc_data_block(tc, &data_block);
1921 schedule_external_copy(tc, block, data_block, cell, bio);
1923 schedule_zero(tc, block, data_block, cell, bio);
1927 retry_bios_on_resume(pool, cell);
1931 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1933 cell_error(pool, cell);
1938 static void process_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
1941 struct pool *pool = tc->pool;
1942 struct bio *bio = cell->holder;
1943 dm_block_t block = get_bio_block(tc, bio);
1944 struct dm_thin_lookup_result lookup_result;
1946 if (tc->requeue_mode) {
1947 cell_requeue(pool, cell);
1951 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1954 if (lookup_result.shared)
1955 process_shared_bio(tc, bio, block, &lookup_result, cell);
1957 inc_all_io_entry(pool, bio);
1958 remap_and_issue(tc, bio, lookup_result.block);
1959 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
1964 if (bio_data_dir(bio) == READ && tc->origin_dev) {
1965 inc_all_io_entry(pool, bio);
1966 cell_defer_no_holder(tc, cell);
1968 if (bio_end_sector(bio) <= tc->origin_size)
1969 remap_to_origin_and_issue(tc, bio);
1971 else if (bio->bi_iter.bi_sector < tc->origin_size) {
1973 bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT;
1974 remap_to_origin_and_issue(tc, bio);
1981 provision_block(tc, bio, block, cell);
1985 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1987 cell_defer_no_holder(tc, cell);
1993 static void process_bio(struct thin_c *tc, struct bio *bio)
1995 struct pool *pool = tc->pool;
1996 dm_block_t block = get_bio_block(tc, bio);
1997 struct dm_bio_prison_cell *cell;
1998 struct dm_cell_key key;
2001 * If cell is already occupied, then the block is already
2002 * being provisioned so we have nothing further to do here.
2004 build_virtual_key(tc->td, block, &key);
2005 if (bio_detain(pool, &key, bio, &cell))
2008 process_cell(tc, cell);
2011 static void __process_bio_read_only(struct thin_c *tc, struct bio *bio,
2012 struct dm_bio_prison_cell *cell)
2015 int rw = bio_data_dir(bio);
2016 dm_block_t block = get_bio_block(tc, bio);
2017 struct dm_thin_lookup_result lookup_result;
2019 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
2022 if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) {
2023 handle_unserviceable_bio(tc->pool, bio);
2025 cell_defer_no_holder(tc, cell);
2027 inc_all_io_entry(tc->pool, bio);
2028 remap_and_issue(tc, bio, lookup_result.block);
2030 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
2036 cell_defer_no_holder(tc, cell);
2038 handle_unserviceable_bio(tc->pool, bio);
2042 if (tc->origin_dev) {
2043 inc_all_io_entry(tc->pool, bio);
2044 remap_to_origin_and_issue(tc, bio);
2053 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
2056 cell_defer_no_holder(tc, cell);
2062 static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
2064 __process_bio_read_only(tc, bio, NULL);
2067 static void process_cell_read_only(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2069 __process_bio_read_only(tc, cell->holder, cell);
2072 static void process_bio_success(struct thin_c *tc, struct bio *bio)
2077 static void process_bio_fail(struct thin_c *tc, struct bio *bio)
2082 static void process_cell_success(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2084 cell_success(tc->pool, cell);
2087 static void process_cell_fail(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2089 cell_error(tc->pool, cell);
2093 * FIXME: should we also commit due to size of transaction, measured in
2096 static int need_commit_due_to_time(struct pool *pool)
2098 return !time_in_range(jiffies, pool->last_commit_jiffies,
2099 pool->last_commit_jiffies + COMMIT_PERIOD);
2102 #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node)
2103 #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook))
2105 static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio)
2107 struct rb_node **rbp, *parent;
2108 struct dm_thin_endio_hook *pbd;
2109 sector_t bi_sector = bio->bi_iter.bi_sector;
2111 rbp = &tc->sort_bio_list.rb_node;
2115 pbd = thin_pbd(parent);
2117 if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector)
2118 rbp = &(*rbp)->rb_left;
2120 rbp = &(*rbp)->rb_right;
2123 pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2124 rb_link_node(&pbd->rb_node, parent, rbp);
2125 rb_insert_color(&pbd->rb_node, &tc->sort_bio_list);
2128 static void __extract_sorted_bios(struct thin_c *tc)
2130 struct rb_node *node;
2131 struct dm_thin_endio_hook *pbd;
2134 for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) {
2135 pbd = thin_pbd(node);
2136 bio = thin_bio(pbd);
2138 bio_list_add(&tc->deferred_bio_list, bio);
2139 rb_erase(&pbd->rb_node, &tc->sort_bio_list);
2142 WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list));
2145 static void __sort_thin_deferred_bios(struct thin_c *tc)
2148 struct bio_list bios;
2150 bio_list_init(&bios);
2151 bio_list_merge(&bios, &tc->deferred_bio_list);
2152 bio_list_init(&tc->deferred_bio_list);
2154 /* Sort deferred_bio_list using rb-tree */
2155 while ((bio = bio_list_pop(&bios)))
2156 __thin_bio_rb_add(tc, bio);
2159 * Transfer the sorted bios in sort_bio_list back to
2160 * deferred_bio_list to allow lockless submission of
2163 __extract_sorted_bios(tc);
2166 static void process_thin_deferred_bios(struct thin_c *tc)
2168 struct pool *pool = tc->pool;
2170 struct bio_list bios;
2171 struct blk_plug plug;
2174 if (tc->requeue_mode) {
2175 error_thin_bio_list(tc, &tc->deferred_bio_list,
2176 BLK_STS_DM_REQUEUE);
2180 bio_list_init(&bios);
2182 spin_lock_irq(&tc->lock);
2184 if (bio_list_empty(&tc->deferred_bio_list)) {
2185 spin_unlock_irq(&tc->lock);
2189 __sort_thin_deferred_bios(tc);
2191 bio_list_merge(&bios, &tc->deferred_bio_list);
2192 bio_list_init(&tc->deferred_bio_list);
2194 spin_unlock_irq(&tc->lock);
2196 blk_start_plug(&plug);
2197 while ((bio = bio_list_pop(&bios))) {
2199 * If we've got no free new_mapping structs, and processing
2200 * this bio might require one, we pause until there are some
2201 * prepared mappings to process.
2203 if (ensure_next_mapping(pool)) {
2204 spin_lock_irq(&tc->lock);
2205 bio_list_add(&tc->deferred_bio_list, bio);
2206 bio_list_merge(&tc->deferred_bio_list, &bios);
2207 spin_unlock_irq(&tc->lock);
2211 if (bio_op(bio) == REQ_OP_DISCARD)
2212 pool->process_discard(tc, bio);
2214 pool->process_bio(tc, bio);
2216 if ((count++ & 127) == 0) {
2217 throttle_work_update(&pool->throttle);
2218 dm_pool_issue_prefetches(pool->pmd);
2222 blk_finish_plug(&plug);
2225 static int cmp_cells(const void *lhs, const void *rhs)
2227 struct dm_bio_prison_cell *lhs_cell = *((struct dm_bio_prison_cell **) lhs);
2228 struct dm_bio_prison_cell *rhs_cell = *((struct dm_bio_prison_cell **) rhs);
2230 BUG_ON(!lhs_cell->holder);
2231 BUG_ON(!rhs_cell->holder);
2233 if (lhs_cell->holder->bi_iter.bi_sector < rhs_cell->holder->bi_iter.bi_sector)
2236 if (lhs_cell->holder->bi_iter.bi_sector > rhs_cell->holder->bi_iter.bi_sector)
2242 static unsigned sort_cells(struct pool *pool, struct list_head *cells)
2245 struct dm_bio_prison_cell *cell, *tmp;
2247 list_for_each_entry_safe(cell, tmp, cells, user_list) {
2248 if (count >= CELL_SORT_ARRAY_SIZE)
2251 pool->cell_sort_array[count++] = cell;
2252 list_del(&cell->user_list);
2255 sort(pool->cell_sort_array, count, sizeof(cell), cmp_cells, NULL);
2260 static void process_thin_deferred_cells(struct thin_c *tc)
2262 struct pool *pool = tc->pool;
2263 struct list_head cells;
2264 struct dm_bio_prison_cell *cell;
2265 unsigned i, j, count;
2267 INIT_LIST_HEAD(&cells);
2269 spin_lock_irq(&tc->lock);
2270 list_splice_init(&tc->deferred_cells, &cells);
2271 spin_unlock_irq(&tc->lock);
2273 if (list_empty(&cells))
2277 count = sort_cells(tc->pool, &cells);
2279 for (i = 0; i < count; i++) {
2280 cell = pool->cell_sort_array[i];
2281 BUG_ON(!cell->holder);
2284 * If we've got no free new_mapping structs, and processing
2285 * this bio might require one, we pause until there are some
2286 * prepared mappings to process.
2288 if (ensure_next_mapping(pool)) {
2289 for (j = i; j < count; j++)
2290 list_add(&pool->cell_sort_array[j]->user_list, &cells);
2292 spin_lock_irq(&tc->lock);
2293 list_splice(&cells, &tc->deferred_cells);
2294 spin_unlock_irq(&tc->lock);
2298 if (bio_op(cell->holder) == REQ_OP_DISCARD)
2299 pool->process_discard_cell(tc, cell);
2301 pool->process_cell(tc, cell);
2304 } while (!list_empty(&cells));
2307 static void thin_get(struct thin_c *tc);
2308 static void thin_put(struct thin_c *tc);
2311 * We can't hold rcu_read_lock() around code that can block. So we
2312 * find a thin with the rcu lock held; bump a refcount; then drop
2315 static struct thin_c *get_first_thin(struct pool *pool)
2317 struct thin_c *tc = NULL;
2320 if (!list_empty(&pool->active_thins)) {
2321 tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list);
2329 static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc)
2331 struct thin_c *old_tc = tc;
2334 list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) {
2346 static void process_deferred_bios(struct pool *pool)
2349 struct bio_list bios, bio_completions;
2352 tc = get_first_thin(pool);
2354 process_thin_deferred_cells(tc);
2355 process_thin_deferred_bios(tc);
2356 tc = get_next_thin(pool, tc);
2360 * If there are any deferred flush bios, we must commit the metadata
2361 * before issuing them or signaling their completion.
2363 bio_list_init(&bios);
2364 bio_list_init(&bio_completions);
2366 spin_lock_irq(&pool->lock);
2367 bio_list_merge(&bios, &pool->deferred_flush_bios);
2368 bio_list_init(&pool->deferred_flush_bios);
2370 bio_list_merge(&bio_completions, &pool->deferred_flush_completions);
2371 bio_list_init(&pool->deferred_flush_completions);
2372 spin_unlock_irq(&pool->lock);
2374 if (bio_list_empty(&bios) && bio_list_empty(&bio_completions) &&
2375 !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool)))
2379 bio_list_merge(&bios, &bio_completions);
2381 while ((bio = bio_list_pop(&bios)))
2385 pool->last_commit_jiffies = jiffies;
2387 while ((bio = bio_list_pop(&bio_completions)))
2390 while ((bio = bio_list_pop(&bios))) {
2392 * The data device was flushed as part of metadata commit,
2393 * so complete redundant flushes immediately.
2395 if (bio->bi_opf & REQ_PREFLUSH)
2398 submit_bio_noacct(bio);
2402 static void do_worker(struct work_struct *ws)
2404 struct pool *pool = container_of(ws, struct pool, worker);
2406 throttle_work_start(&pool->throttle);
2407 dm_pool_issue_prefetches(pool->pmd);
2408 throttle_work_update(&pool->throttle);
2409 process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
2410 throttle_work_update(&pool->throttle);
2411 process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
2412 throttle_work_update(&pool->throttle);
2413 process_prepared(pool, &pool->prepared_discards_pt2, &pool->process_prepared_discard_pt2);
2414 throttle_work_update(&pool->throttle);
2415 process_deferred_bios(pool);
2416 throttle_work_complete(&pool->throttle);
2420 * We want to commit periodically so that not too much
2421 * unwritten data builds up.
2423 static void do_waker(struct work_struct *ws)
2425 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
2427 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
2431 * We're holding onto IO to allow userland time to react. After the
2432 * timeout either the pool will have been resized (and thus back in
2433 * PM_WRITE mode), or we degrade to PM_OUT_OF_DATA_SPACE w/ error_if_no_space.
2435 static void do_no_space_timeout(struct work_struct *ws)
2437 struct pool *pool = container_of(to_delayed_work(ws), struct pool,
2440 if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) {
2441 pool->pf.error_if_no_space = true;
2442 notify_of_pool_mode_change(pool);
2443 error_retry_list_with_code(pool, BLK_STS_NOSPC);
2447 /*----------------------------------------------------------------*/
2450 struct work_struct worker;
2451 struct completion complete;
2454 static struct pool_work *to_pool_work(struct work_struct *ws)
2456 return container_of(ws, struct pool_work, worker);
2459 static void pool_work_complete(struct pool_work *pw)
2461 complete(&pw->complete);
2464 static void pool_work_wait(struct pool_work *pw, struct pool *pool,
2465 void (*fn)(struct work_struct *))
2467 INIT_WORK_ONSTACK(&pw->worker, fn);
2468 init_completion(&pw->complete);
2469 queue_work(pool->wq, &pw->worker);
2470 wait_for_completion(&pw->complete);
2473 /*----------------------------------------------------------------*/
2475 struct noflush_work {
2476 struct pool_work pw;
2480 static struct noflush_work *to_noflush(struct work_struct *ws)
2482 return container_of(to_pool_work(ws), struct noflush_work, pw);
2485 static void do_noflush_start(struct work_struct *ws)
2487 struct noflush_work *w = to_noflush(ws);
2488 w->tc->requeue_mode = true;
2490 pool_work_complete(&w->pw);
2493 static void do_noflush_stop(struct work_struct *ws)
2495 struct noflush_work *w = to_noflush(ws);
2496 w->tc->requeue_mode = false;
2497 pool_work_complete(&w->pw);
2500 static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *))
2502 struct noflush_work w;
2505 pool_work_wait(&w.pw, tc->pool, fn);
2508 /*----------------------------------------------------------------*/
2510 static bool passdown_enabled(struct pool_c *pt)
2512 return pt->adjusted_pf.discard_passdown;
2515 static void set_discard_callbacks(struct pool *pool)
2517 struct pool_c *pt = pool->ti->private;
2519 if (passdown_enabled(pt)) {
2520 pool->process_discard_cell = process_discard_cell_passdown;
2521 pool->process_prepared_discard = process_prepared_discard_passdown_pt1;
2522 pool->process_prepared_discard_pt2 = process_prepared_discard_passdown_pt2;
2524 pool->process_discard_cell = process_discard_cell_no_passdown;
2525 pool->process_prepared_discard = process_prepared_discard_no_passdown;
2529 static void set_pool_mode(struct pool *pool, enum pool_mode new_mode)
2531 struct pool_c *pt = pool->ti->private;
2532 bool needs_check = dm_pool_metadata_needs_check(pool->pmd);
2533 enum pool_mode old_mode = get_pool_mode(pool);
2534 unsigned long no_space_timeout = READ_ONCE(no_space_timeout_secs) * HZ;
2537 * Never allow the pool to transition to PM_WRITE mode if user
2538 * intervention is required to verify metadata and data consistency.
2540 if (new_mode == PM_WRITE && needs_check) {
2541 DMERR("%s: unable to switch pool to write mode until repaired.",
2542 dm_device_name(pool->pool_md));
2543 if (old_mode != new_mode)
2544 new_mode = old_mode;
2546 new_mode = PM_READ_ONLY;
2549 * If we were in PM_FAIL mode, rollback of metadata failed. We're
2550 * not going to recover without a thin_repair. So we never let the
2551 * pool move out of the old mode.
2553 if (old_mode == PM_FAIL)
2554 new_mode = old_mode;
2558 dm_pool_metadata_read_only(pool->pmd);
2559 pool->process_bio = process_bio_fail;
2560 pool->process_discard = process_bio_fail;
2561 pool->process_cell = process_cell_fail;
2562 pool->process_discard_cell = process_cell_fail;
2563 pool->process_prepared_mapping = process_prepared_mapping_fail;
2564 pool->process_prepared_discard = process_prepared_discard_fail;
2566 error_retry_list(pool);
2569 case PM_OUT_OF_METADATA_SPACE:
2571 dm_pool_metadata_read_only(pool->pmd);
2572 pool->process_bio = process_bio_read_only;
2573 pool->process_discard = process_bio_success;
2574 pool->process_cell = process_cell_read_only;
2575 pool->process_discard_cell = process_cell_success;
2576 pool->process_prepared_mapping = process_prepared_mapping_fail;
2577 pool->process_prepared_discard = process_prepared_discard_success;
2579 error_retry_list(pool);
2582 case PM_OUT_OF_DATA_SPACE:
2584 * Ideally we'd never hit this state; the low water mark
2585 * would trigger userland to extend the pool before we
2586 * completely run out of data space. However, many small
2587 * IOs to unprovisioned space can consume data space at an
2588 * alarming rate. Adjust your low water mark if you're
2589 * frequently seeing this mode.
2591 pool->out_of_data_space = true;
2592 pool->process_bio = process_bio_read_only;
2593 pool->process_discard = process_discard_bio;
2594 pool->process_cell = process_cell_read_only;
2595 pool->process_prepared_mapping = process_prepared_mapping;
2596 set_discard_callbacks(pool);
2598 if (!pool->pf.error_if_no_space && no_space_timeout)
2599 queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout);
2603 if (old_mode == PM_OUT_OF_DATA_SPACE)
2604 cancel_delayed_work_sync(&pool->no_space_timeout);
2605 pool->out_of_data_space = false;
2606 pool->pf.error_if_no_space = pt->requested_pf.error_if_no_space;
2607 dm_pool_metadata_read_write(pool->pmd);
2608 pool->process_bio = process_bio;
2609 pool->process_discard = process_discard_bio;
2610 pool->process_cell = process_cell;
2611 pool->process_prepared_mapping = process_prepared_mapping;
2612 set_discard_callbacks(pool);
2616 pool->pf.mode = new_mode;
2618 * The pool mode may have changed, sync it so bind_control_target()
2619 * doesn't cause an unexpected mode transition on resume.
2621 pt->adjusted_pf.mode = new_mode;
2623 if (old_mode != new_mode)
2624 notify_of_pool_mode_change(pool);
2627 static void abort_transaction(struct pool *pool)
2629 const char *dev_name = dm_device_name(pool->pool_md);
2631 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name);
2632 if (dm_pool_abort_metadata(pool->pmd)) {
2633 DMERR("%s: failed to abort metadata transaction", dev_name);
2634 set_pool_mode(pool, PM_FAIL);
2637 if (dm_pool_metadata_set_needs_check(pool->pmd)) {
2638 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name);
2639 set_pool_mode(pool, PM_FAIL);
2643 static void metadata_operation_failed(struct pool *pool, const char *op, int r)
2645 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
2646 dm_device_name(pool->pool_md), op, r);
2648 abort_transaction(pool);
2649 set_pool_mode(pool, PM_READ_ONLY);
2652 /*----------------------------------------------------------------*/
2655 * Mapping functions.
2659 * Called only while mapping a thin bio to hand it over to the workqueue.
2661 static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
2663 struct pool *pool = tc->pool;
2665 spin_lock_irq(&tc->lock);
2666 bio_list_add(&tc->deferred_bio_list, bio);
2667 spin_unlock_irq(&tc->lock);
2672 static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio)
2674 struct pool *pool = tc->pool;
2676 throttle_lock(&pool->throttle);
2677 thin_defer_bio(tc, bio);
2678 throttle_unlock(&pool->throttle);
2681 static void thin_defer_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2683 struct pool *pool = tc->pool;
2685 throttle_lock(&pool->throttle);
2686 spin_lock_irq(&tc->lock);
2687 list_add_tail(&cell->user_list, &tc->deferred_cells);
2688 spin_unlock_irq(&tc->lock);
2689 throttle_unlock(&pool->throttle);
2694 static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
2696 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2699 h->shared_read_entry = NULL;
2700 h->all_io_entry = NULL;
2701 h->overwrite_mapping = NULL;
2706 * Non-blocking function called from the thin target's map function.
2708 static int thin_bio_map(struct dm_target *ti, struct bio *bio)
2711 struct thin_c *tc = ti->private;
2712 dm_block_t block = get_bio_block(tc, bio);
2713 struct dm_thin_device *td = tc->td;
2714 struct dm_thin_lookup_result result;
2715 struct dm_bio_prison_cell *virt_cell, *data_cell;
2716 struct dm_cell_key key;
2718 thin_hook_bio(tc, bio);
2720 if (tc->requeue_mode) {
2721 bio->bi_status = BLK_STS_DM_REQUEUE;
2723 return DM_MAPIO_SUBMITTED;
2726 if (get_pool_mode(tc->pool) == PM_FAIL) {
2728 return DM_MAPIO_SUBMITTED;
2731 if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD) {
2732 thin_defer_bio_with_throttle(tc, bio);
2733 return DM_MAPIO_SUBMITTED;
2737 * We must hold the virtual cell before doing the lookup, otherwise
2738 * there's a race with discard.
2740 build_virtual_key(tc->td, block, &key);
2741 if (bio_detain(tc->pool, &key, bio, &virt_cell))
2742 return DM_MAPIO_SUBMITTED;
2744 r = dm_thin_find_block(td, block, 0, &result);
2747 * Note that we defer readahead too.
2751 if (unlikely(result.shared)) {
2753 * We have a race condition here between the
2754 * result.shared value returned by the lookup and
2755 * snapshot creation, which may cause new
2758 * To avoid this always quiesce the origin before
2759 * taking the snap. You want to do this anyway to
2760 * ensure a consistent application view
2763 * More distant ancestors are irrelevant. The
2764 * shared flag will be set in their case.
2766 thin_defer_cell(tc, virt_cell);
2767 return DM_MAPIO_SUBMITTED;
2770 build_data_key(tc->td, result.block, &key);
2771 if (bio_detain(tc->pool, &key, bio, &data_cell)) {
2772 cell_defer_no_holder(tc, virt_cell);
2773 return DM_MAPIO_SUBMITTED;
2776 inc_all_io_entry(tc->pool, bio);
2777 cell_defer_no_holder(tc, data_cell);
2778 cell_defer_no_holder(tc, virt_cell);
2780 remap(tc, bio, result.block);
2781 return DM_MAPIO_REMAPPED;
2785 thin_defer_cell(tc, virt_cell);
2786 return DM_MAPIO_SUBMITTED;
2790 * Must always call bio_io_error on failure.
2791 * dm_thin_find_block can fail with -EINVAL if the
2792 * pool is switched to fail-io mode.
2795 cell_defer_no_holder(tc, virt_cell);
2796 return DM_MAPIO_SUBMITTED;
2800 static void requeue_bios(struct pool *pool)
2805 list_for_each_entry_rcu(tc, &pool->active_thins, list) {
2806 spin_lock_irq(&tc->lock);
2807 bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list);
2808 bio_list_init(&tc->retry_on_resume_list);
2809 spin_unlock_irq(&tc->lock);
2814 /*----------------------------------------------------------------
2815 * Binding of control targets to a pool object
2816 *--------------------------------------------------------------*/
2817 static bool data_dev_supports_discard(struct pool_c *pt)
2819 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
2821 return q && blk_queue_discard(q);
2824 static bool is_factor(sector_t block_size, uint32_t n)
2826 return !sector_div(block_size, n);
2830 * If discard_passdown was enabled verify that the data device
2831 * supports discards. Disable discard_passdown if not.
2833 static void disable_passdown_if_not_supported(struct pool_c *pt)
2835 struct pool *pool = pt->pool;
2836 struct block_device *data_bdev = pt->data_dev->bdev;
2837 struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
2838 const char *reason = NULL;
2839 char buf[BDEVNAME_SIZE];
2841 if (!pt->adjusted_pf.discard_passdown)
2844 if (!data_dev_supports_discard(pt))
2845 reason = "discard unsupported";
2847 else if (data_limits->max_discard_sectors < pool->sectors_per_block)
2848 reason = "max discard sectors smaller than a block";
2851 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason);
2852 pt->adjusted_pf.discard_passdown = false;
2856 static int bind_control_target(struct pool *pool, struct dm_target *ti)
2858 struct pool_c *pt = ti->private;
2861 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
2863 enum pool_mode old_mode = get_pool_mode(pool);
2864 enum pool_mode new_mode = pt->adjusted_pf.mode;
2867 * Don't change the pool's mode until set_pool_mode() below.
2868 * Otherwise the pool's process_* function pointers may
2869 * not match the desired pool mode.
2871 pt->adjusted_pf.mode = old_mode;
2874 pool->pf = pt->adjusted_pf;
2875 pool->low_water_blocks = pt->low_water_blocks;
2877 set_pool_mode(pool, new_mode);
2882 static void unbind_control_target(struct pool *pool, struct dm_target *ti)
2888 /*----------------------------------------------------------------
2890 *--------------------------------------------------------------*/
2891 /* Initialize pool features. */
2892 static void pool_features_init(struct pool_features *pf)
2894 pf->mode = PM_WRITE;
2895 pf->zero_new_blocks = true;
2896 pf->discard_enabled = true;
2897 pf->discard_passdown = true;
2898 pf->error_if_no_space = false;
2901 static void __pool_destroy(struct pool *pool)
2903 __pool_table_remove(pool);
2905 vfree(pool->cell_sort_array);
2906 if (dm_pool_metadata_close(pool->pmd) < 0)
2907 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
2909 dm_bio_prison_destroy(pool->prison);
2910 dm_kcopyd_client_destroy(pool->copier);
2912 cancel_delayed_work_sync(&pool->waker);
2913 cancel_delayed_work_sync(&pool->no_space_timeout);
2915 destroy_workqueue(pool->wq);
2917 if (pool->next_mapping)
2918 mempool_free(pool->next_mapping, &pool->mapping_pool);
2919 mempool_exit(&pool->mapping_pool);
2920 bio_uninit(&pool->flush_bio);
2921 dm_deferred_set_destroy(pool->shared_read_ds);
2922 dm_deferred_set_destroy(pool->all_io_ds);
2926 static struct kmem_cache *_new_mapping_cache;
2928 static struct pool *pool_create(struct mapped_device *pool_md,
2929 struct block_device *metadata_dev,
2930 struct block_device *data_dev,
2931 unsigned long block_size,
2932 int read_only, char **error)
2937 struct dm_pool_metadata *pmd;
2938 bool format_device = read_only ? false : true;
2940 pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
2942 *error = "Error creating metadata object";
2943 return (struct pool *)pmd;
2946 pool = kzalloc(sizeof(*pool), GFP_KERNEL);
2948 *error = "Error allocating memory for pool";
2949 err_p = ERR_PTR(-ENOMEM);
2954 pool->sectors_per_block = block_size;
2955 if (block_size & (block_size - 1))
2956 pool->sectors_per_block_shift = -1;
2958 pool->sectors_per_block_shift = __ffs(block_size);
2959 pool->low_water_blocks = 0;
2960 pool_features_init(&pool->pf);
2961 pool->prison = dm_bio_prison_create();
2962 if (!pool->prison) {
2963 *error = "Error creating pool's bio prison";
2964 err_p = ERR_PTR(-ENOMEM);
2968 pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2969 if (IS_ERR(pool->copier)) {
2970 r = PTR_ERR(pool->copier);
2971 *error = "Error creating pool's kcopyd client";
2973 goto bad_kcopyd_client;
2977 * Create singlethreaded workqueue that will service all devices
2978 * that use this metadata.
2980 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
2982 *error = "Error creating pool's workqueue";
2983 err_p = ERR_PTR(-ENOMEM);
2987 throttle_init(&pool->throttle);
2988 INIT_WORK(&pool->worker, do_worker);
2989 INIT_DELAYED_WORK(&pool->waker, do_waker);
2990 INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout);
2991 spin_lock_init(&pool->lock);
2992 bio_list_init(&pool->deferred_flush_bios);
2993 bio_list_init(&pool->deferred_flush_completions);
2994 INIT_LIST_HEAD(&pool->prepared_mappings);
2995 INIT_LIST_HEAD(&pool->prepared_discards);
2996 INIT_LIST_HEAD(&pool->prepared_discards_pt2);
2997 INIT_LIST_HEAD(&pool->active_thins);
2998 pool->low_water_triggered = false;
2999 pool->suspended = true;
3000 pool->out_of_data_space = false;
3001 bio_init(&pool->flush_bio, NULL, 0);
3003 pool->shared_read_ds = dm_deferred_set_create();
3004 if (!pool->shared_read_ds) {
3005 *error = "Error creating pool's shared read deferred set";
3006 err_p = ERR_PTR(-ENOMEM);
3007 goto bad_shared_read_ds;
3010 pool->all_io_ds = dm_deferred_set_create();
3011 if (!pool->all_io_ds) {
3012 *error = "Error creating pool's all io deferred set";
3013 err_p = ERR_PTR(-ENOMEM);
3017 pool->next_mapping = NULL;
3018 r = mempool_init_slab_pool(&pool->mapping_pool, MAPPING_POOL_SIZE,
3019 _new_mapping_cache);
3021 *error = "Error creating pool's mapping mempool";
3023 goto bad_mapping_pool;
3026 pool->cell_sort_array =
3027 vmalloc(array_size(CELL_SORT_ARRAY_SIZE,
3028 sizeof(*pool->cell_sort_array)));
3029 if (!pool->cell_sort_array) {
3030 *error = "Error allocating cell sort array";
3031 err_p = ERR_PTR(-ENOMEM);
3032 goto bad_sort_array;
3035 pool->ref_count = 1;
3036 pool->last_commit_jiffies = jiffies;
3037 pool->pool_md = pool_md;
3038 pool->md_dev = metadata_dev;
3039 pool->data_dev = data_dev;
3040 __pool_table_insert(pool);
3045 mempool_exit(&pool->mapping_pool);
3047 dm_deferred_set_destroy(pool->all_io_ds);
3049 dm_deferred_set_destroy(pool->shared_read_ds);
3051 destroy_workqueue(pool->wq);
3053 dm_kcopyd_client_destroy(pool->copier);
3055 dm_bio_prison_destroy(pool->prison);
3059 if (dm_pool_metadata_close(pmd))
3060 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
3065 static void __pool_inc(struct pool *pool)
3067 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
3071 static void __pool_dec(struct pool *pool)
3073 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
3074 BUG_ON(!pool->ref_count);
3075 if (!--pool->ref_count)
3076 __pool_destroy(pool);
3079 static struct pool *__pool_find(struct mapped_device *pool_md,
3080 struct block_device *metadata_dev,
3081 struct block_device *data_dev,
3082 unsigned long block_size, int read_only,
3083 char **error, int *created)
3085 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
3088 if (pool->pool_md != pool_md) {
3089 *error = "metadata device already in use by a pool";
3090 return ERR_PTR(-EBUSY);
3092 if (pool->data_dev != data_dev) {
3093 *error = "data device already in use by a pool";
3094 return ERR_PTR(-EBUSY);
3099 pool = __pool_table_lookup(pool_md);
3101 if (pool->md_dev != metadata_dev || pool->data_dev != data_dev) {
3102 *error = "different pool cannot replace a pool";
3103 return ERR_PTR(-EINVAL);
3108 pool = pool_create(pool_md, metadata_dev, data_dev, block_size, read_only, error);
3116 /*----------------------------------------------------------------
3117 * Pool target methods
3118 *--------------------------------------------------------------*/
3119 static void pool_dtr(struct dm_target *ti)
3121 struct pool_c *pt = ti->private;
3123 mutex_lock(&dm_thin_pool_table.mutex);
3125 unbind_control_target(pt->pool, ti);
3126 __pool_dec(pt->pool);
3127 dm_put_device(ti, pt->metadata_dev);
3128 dm_put_device(ti, pt->data_dev);
3131 mutex_unlock(&dm_thin_pool_table.mutex);
3134 static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
3135 struct dm_target *ti)
3139 const char *arg_name;
3141 static const struct dm_arg _args[] = {
3142 {0, 4, "Invalid number of pool feature arguments"},
3146 * No feature arguments supplied.
3151 r = dm_read_arg_group(_args, as, &argc, &ti->error);
3155 while (argc && !r) {
3156 arg_name = dm_shift_arg(as);
3159 if (!strcasecmp(arg_name, "skip_block_zeroing"))
3160 pf->zero_new_blocks = false;
3162 else if (!strcasecmp(arg_name, "ignore_discard"))
3163 pf->discard_enabled = false;
3165 else if (!strcasecmp(arg_name, "no_discard_passdown"))
3166 pf->discard_passdown = false;
3168 else if (!strcasecmp(arg_name, "read_only"))
3169 pf->mode = PM_READ_ONLY;
3171 else if (!strcasecmp(arg_name, "error_if_no_space"))
3172 pf->error_if_no_space = true;
3175 ti->error = "Unrecognised pool feature requested";
3184 static void metadata_low_callback(void *context)
3186 struct pool *pool = context;
3188 DMWARN("%s: reached low water mark for metadata device: sending event.",
3189 dm_device_name(pool->pool_md));
3191 dm_table_event(pool->ti->table);
3195 * We need to flush the data device **before** committing the metadata.
3197 * This ensures that the data blocks of any newly inserted mappings are
3198 * properly written to non-volatile storage and won't be lost in case of a
3201 * Failure to do so can result in data corruption in the case of internal or
3202 * external snapshots and in the case of newly provisioned blocks, when block
3203 * zeroing is enabled.
3205 static int metadata_pre_commit_callback(void *context)
3207 struct pool *pool = context;
3208 struct bio *flush_bio = &pool->flush_bio;
3210 bio_reset(flush_bio);
3211 bio_set_dev(flush_bio, pool->data_dev);
3212 flush_bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
3214 return submit_bio_wait(flush_bio);
3217 static sector_t get_dev_size(struct block_device *bdev)
3219 return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
3222 static void warn_if_metadata_device_too_big(struct block_device *bdev)
3224 sector_t metadata_dev_size = get_dev_size(bdev);
3225 char buffer[BDEVNAME_SIZE];
3227 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING)
3228 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
3229 bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS);
3232 static sector_t get_metadata_dev_size(struct block_device *bdev)
3234 sector_t metadata_dev_size = get_dev_size(bdev);
3236 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS)
3237 metadata_dev_size = THIN_METADATA_MAX_SECTORS;
3239 return metadata_dev_size;
3242 static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
3244 sector_t metadata_dev_size = get_metadata_dev_size(bdev);
3246 sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE);
3248 return metadata_dev_size;
3252 * When a metadata threshold is crossed a dm event is triggered, and
3253 * userland should respond by growing the metadata device. We could let
3254 * userland set the threshold, like we do with the data threshold, but I'm
3255 * not sure they know enough to do this well.
3257 static dm_block_t calc_metadata_threshold(struct pool_c *pt)
3260 * 4M is ample for all ops with the possible exception of thin
3261 * device deletion which is harmless if it fails (just retry the
3262 * delete after you've grown the device).
3264 dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
3265 return min((dm_block_t)1024ULL /* 4M */, quarter);
3269 * thin-pool <metadata dev> <data dev>
3270 * <data block size (sectors)>
3271 * <low water mark (blocks)>
3272 * [<#feature args> [<arg>]*]
3274 * Optional feature arguments are:
3275 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
3276 * ignore_discard: disable discard
3277 * no_discard_passdown: don't pass discards down to the data device
3278 * read_only: Don't allow any changes to be made to the pool metadata.
3279 * error_if_no_space: error IOs, instead of queueing, if no space.
3281 static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
3283 int r, pool_created = 0;
3286 struct pool_features pf;
3287 struct dm_arg_set as;
3288 struct dm_dev *data_dev;
3289 unsigned long block_size;
3290 dm_block_t low_water_blocks;
3291 struct dm_dev *metadata_dev;
3292 fmode_t metadata_mode;
3295 * FIXME Remove validation from scope of lock.
3297 mutex_lock(&dm_thin_pool_table.mutex);
3300 ti->error = "Invalid argument count";
3308 /* make sure metadata and data are different devices */
3309 if (!strcmp(argv[0], argv[1])) {
3310 ti->error = "Error setting metadata or data device";
3316 * Set default pool features.
3318 pool_features_init(&pf);
3320 dm_consume_args(&as, 4);
3321 r = parse_pool_features(&as, &pf, ti);
3325 metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
3326 r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
3328 ti->error = "Error opening metadata block device";
3331 warn_if_metadata_device_too_big(metadata_dev->bdev);
3333 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
3335 ti->error = "Error getting data device";
3339 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
3340 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
3341 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
3342 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
3343 ti->error = "Invalid block size";
3348 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
3349 ti->error = "Invalid low water mark";
3354 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
3360 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev, data_dev->bdev,
3361 block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
3368 * 'pool_created' reflects whether this is the first table load.
3369 * Top level discard support is not allowed to be changed after
3370 * initial load. This would require a pool reload to trigger thin
3373 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
3374 ti->error = "Discard support cannot be disabled once enabled";
3376 goto out_flags_changed;
3381 pt->metadata_dev = metadata_dev;
3382 pt->data_dev = data_dev;
3383 pt->low_water_blocks = low_water_blocks;
3384 pt->adjusted_pf = pt->requested_pf = pf;
3385 ti->num_flush_bios = 1;
3386 ti->limit_swap_bios = true;
3389 * Only need to enable discards if the pool should pass
3390 * them down to the data device. The thin device's discard
3391 * processing will cause mappings to be removed from the btree.
3393 if (pf.discard_enabled && pf.discard_passdown) {
3394 ti->num_discard_bios = 1;
3397 * Setting 'discards_supported' circumvents the normal
3398 * stacking of discard limits (this keeps the pool and
3399 * thin devices' discard limits consistent).
3401 ti->discards_supported = true;
3405 r = dm_pool_register_metadata_threshold(pt->pool->pmd,
3406 calc_metadata_threshold(pt),
3407 metadata_low_callback,
3410 ti->error = "Error registering metadata threshold";
3411 goto out_flags_changed;
3414 dm_pool_register_pre_commit_callback(pool->pmd,
3415 metadata_pre_commit_callback, pool);
3417 mutex_unlock(&dm_thin_pool_table.mutex);
3426 dm_put_device(ti, data_dev);
3428 dm_put_device(ti, metadata_dev);
3430 mutex_unlock(&dm_thin_pool_table.mutex);
3435 static int pool_map(struct dm_target *ti, struct bio *bio)
3438 struct pool_c *pt = ti->private;
3439 struct pool *pool = pt->pool;
3442 * As this is a singleton target, ti->begin is always zero.
3444 spin_lock_irq(&pool->lock);
3445 bio_set_dev(bio, pt->data_dev->bdev);
3446 r = DM_MAPIO_REMAPPED;
3447 spin_unlock_irq(&pool->lock);
3452 static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
3455 struct pool_c *pt = ti->private;
3456 struct pool *pool = pt->pool;
3457 sector_t data_size = ti->len;
3458 dm_block_t sb_data_size;
3460 *need_commit = false;
3462 (void) sector_div(data_size, pool->sectors_per_block);
3464 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
3466 DMERR("%s: failed to retrieve data device size",
3467 dm_device_name(pool->pool_md));
3471 if (data_size < sb_data_size) {
3472 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
3473 dm_device_name(pool->pool_md),
3474 (unsigned long long)data_size, sb_data_size);
3477 } else if (data_size > sb_data_size) {
3478 if (dm_pool_metadata_needs_check(pool->pmd)) {
3479 DMERR("%s: unable to grow the data device until repaired.",
3480 dm_device_name(pool->pool_md));
3485 DMINFO("%s: growing the data device from %llu to %llu blocks",
3486 dm_device_name(pool->pool_md),
3487 sb_data_size, (unsigned long long)data_size);
3488 r = dm_pool_resize_data_dev(pool->pmd, data_size);
3490 metadata_operation_failed(pool, "dm_pool_resize_data_dev", r);
3494 *need_commit = true;
3500 static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
3503 struct pool_c *pt = ti->private;
3504 struct pool *pool = pt->pool;
3505 dm_block_t metadata_dev_size, sb_metadata_dev_size;
3507 *need_commit = false;
3509 metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
3511 r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
3513 DMERR("%s: failed to retrieve metadata device size",
3514 dm_device_name(pool->pool_md));
3518 if (metadata_dev_size < sb_metadata_dev_size) {
3519 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
3520 dm_device_name(pool->pool_md),
3521 metadata_dev_size, sb_metadata_dev_size);
3524 } else if (metadata_dev_size > sb_metadata_dev_size) {
3525 if (dm_pool_metadata_needs_check(pool->pmd)) {
3526 DMERR("%s: unable to grow the metadata device until repaired.",
3527 dm_device_name(pool->pool_md));
3531 warn_if_metadata_device_too_big(pool->md_dev);
3532 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
3533 dm_device_name(pool->pool_md),
3534 sb_metadata_dev_size, metadata_dev_size);
3536 if (get_pool_mode(pool) == PM_OUT_OF_METADATA_SPACE)
3537 set_pool_mode(pool, PM_WRITE);
3539 r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
3541 metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r);
3545 *need_commit = true;
3552 * Retrieves the number of blocks of the data device from
3553 * the superblock and compares it to the actual device size,
3554 * thus resizing the data device in case it has grown.
3556 * This both copes with opening preallocated data devices in the ctr
3557 * being followed by a resume
3559 * calling the resume method individually after userspace has
3560 * grown the data device in reaction to a table event.
3562 static int pool_preresume(struct dm_target *ti)
3565 bool need_commit1, need_commit2;
3566 struct pool_c *pt = ti->private;
3567 struct pool *pool = pt->pool;
3570 * Take control of the pool object.
3572 r = bind_control_target(pool, ti);
3576 r = maybe_resize_data_dev(ti, &need_commit1);
3580 r = maybe_resize_metadata_dev(ti, &need_commit2);
3584 if (need_commit1 || need_commit2)
3585 (void) commit(pool);
3588 * When a thin-pool is PM_FAIL, it cannot be rebuilt if
3589 * bio is in deferred list. Therefore need to return 0
3590 * to allow pool_resume() to flush IO.
3592 if (r && get_pool_mode(pool) == PM_FAIL)
3598 static void pool_suspend_active_thins(struct pool *pool)
3602 /* Suspend all active thin devices */
3603 tc = get_first_thin(pool);
3605 dm_internal_suspend_noflush(tc->thin_md);
3606 tc = get_next_thin(pool, tc);
3610 static void pool_resume_active_thins(struct pool *pool)
3614 /* Resume all active thin devices */
3615 tc = get_first_thin(pool);
3617 dm_internal_resume(tc->thin_md);
3618 tc = get_next_thin(pool, tc);
3622 static void pool_resume(struct dm_target *ti)
3624 struct pool_c *pt = ti->private;
3625 struct pool *pool = pt->pool;
3628 * Must requeue active_thins' bios and then resume
3629 * active_thins _before_ clearing 'suspend' flag.
3632 pool_resume_active_thins(pool);
3634 spin_lock_irq(&pool->lock);
3635 pool->low_water_triggered = false;
3636 pool->suspended = false;
3637 spin_unlock_irq(&pool->lock);
3639 do_waker(&pool->waker.work);
3642 static void pool_presuspend(struct dm_target *ti)
3644 struct pool_c *pt = ti->private;
3645 struct pool *pool = pt->pool;
3647 spin_lock_irq(&pool->lock);
3648 pool->suspended = true;
3649 spin_unlock_irq(&pool->lock);
3651 pool_suspend_active_thins(pool);
3654 static void pool_presuspend_undo(struct dm_target *ti)
3656 struct pool_c *pt = ti->private;
3657 struct pool *pool = pt->pool;
3659 pool_resume_active_thins(pool);
3661 spin_lock_irq(&pool->lock);
3662 pool->suspended = false;
3663 spin_unlock_irq(&pool->lock);
3666 static void pool_postsuspend(struct dm_target *ti)
3668 struct pool_c *pt = ti->private;
3669 struct pool *pool = pt->pool;
3671 cancel_delayed_work_sync(&pool->waker);
3672 cancel_delayed_work_sync(&pool->no_space_timeout);
3673 flush_workqueue(pool->wq);
3674 (void) commit(pool);
3677 static int check_arg_count(unsigned argc, unsigned args_required)
3679 if (argc != args_required) {
3680 DMWARN("Message received with %u arguments instead of %u.",
3681 argc, args_required);
3688 static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
3690 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
3691 *dev_id <= MAX_DEV_ID)
3695 DMWARN("Message received with invalid device id: %s", arg);
3700 static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
3705 r = check_arg_count(argc, 2);
3709 r = read_dev_id(argv[1], &dev_id, 1);
3713 r = dm_pool_create_thin(pool->pmd, dev_id);
3715 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
3723 static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3726 dm_thin_id origin_dev_id;
3729 r = check_arg_count(argc, 3);
3733 r = read_dev_id(argv[1], &dev_id, 1);
3737 r = read_dev_id(argv[2], &origin_dev_id, 1);
3741 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
3743 DMWARN("Creation of new snapshot %s of device %s failed.",
3751 static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
3756 r = check_arg_count(argc, 2);
3760 r = read_dev_id(argv[1], &dev_id, 1);
3764 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
3766 DMWARN("Deletion of thin device %s failed.", argv[1]);
3771 static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
3773 dm_thin_id old_id, new_id;
3776 r = check_arg_count(argc, 3);
3780 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
3781 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
3785 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
3786 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
3790 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
3792 DMWARN("Failed to change transaction id from %s to %s.",
3800 static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3804 r = check_arg_count(argc, 1);
3808 (void) commit(pool);
3810 r = dm_pool_reserve_metadata_snap(pool->pmd);
3812 DMWARN("reserve_metadata_snap message failed.");
3817 static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3821 r = check_arg_count(argc, 1);
3825 r = dm_pool_release_metadata_snap(pool->pmd);
3827 DMWARN("release_metadata_snap message failed.");
3833 * Messages supported:
3834 * create_thin <dev_id>
3835 * create_snap <dev_id> <origin_id>
3837 * set_transaction_id <current_trans_id> <new_trans_id>
3838 * reserve_metadata_snap
3839 * release_metadata_snap
3841 static int pool_message(struct dm_target *ti, unsigned argc, char **argv,
3842 char *result, unsigned maxlen)
3845 struct pool_c *pt = ti->private;
3846 struct pool *pool = pt->pool;
3848 if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE) {
3849 DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode",
3850 dm_device_name(pool->pool_md));
3854 if (!strcasecmp(argv[0], "create_thin"))
3855 r = process_create_thin_mesg(argc, argv, pool);
3857 else if (!strcasecmp(argv[0], "create_snap"))
3858 r = process_create_snap_mesg(argc, argv, pool);
3860 else if (!strcasecmp(argv[0], "delete"))
3861 r = process_delete_mesg(argc, argv, pool);
3863 else if (!strcasecmp(argv[0], "set_transaction_id"))
3864 r = process_set_transaction_id_mesg(argc, argv, pool);
3866 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
3867 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
3869 else if (!strcasecmp(argv[0], "release_metadata_snap"))
3870 r = process_release_metadata_snap_mesg(argc, argv, pool);
3873 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
3876 (void) commit(pool);
3881 static void emit_flags(struct pool_features *pf, char *result,
3882 unsigned sz, unsigned maxlen)
3884 unsigned count = !pf->zero_new_blocks + !pf->discard_enabled +
3885 !pf->discard_passdown + (pf->mode == PM_READ_ONLY) +
3886 pf->error_if_no_space;
3887 DMEMIT("%u ", count);
3889 if (!pf->zero_new_blocks)
3890 DMEMIT("skip_block_zeroing ");
3892 if (!pf->discard_enabled)
3893 DMEMIT("ignore_discard ");
3895 if (!pf->discard_passdown)
3896 DMEMIT("no_discard_passdown ");
3898 if (pf->mode == PM_READ_ONLY)
3899 DMEMIT("read_only ");
3901 if (pf->error_if_no_space)
3902 DMEMIT("error_if_no_space ");
3907 * <transaction id> <used metadata sectors>/<total metadata sectors>
3908 * <used data sectors>/<total data sectors> <held metadata root>
3909 * <pool mode> <discard config> <no space config> <needs_check>
3911 static void pool_status(struct dm_target *ti, status_type_t type,
3912 unsigned status_flags, char *result, unsigned maxlen)
3916 uint64_t transaction_id;
3917 dm_block_t nr_free_blocks_data;
3918 dm_block_t nr_free_blocks_metadata;
3919 dm_block_t nr_blocks_data;
3920 dm_block_t nr_blocks_metadata;
3921 dm_block_t held_root;
3922 enum pool_mode mode;
3923 char buf[BDEVNAME_SIZE];
3924 char buf2[BDEVNAME_SIZE];
3925 struct pool_c *pt = ti->private;
3926 struct pool *pool = pt->pool;
3929 case STATUSTYPE_INFO:
3930 if (get_pool_mode(pool) == PM_FAIL) {
3935 /* Commit to ensure statistics aren't out-of-date */
3936 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
3937 (void) commit(pool);
3939 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
3941 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
3942 dm_device_name(pool->pool_md), r);
3946 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
3948 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
3949 dm_device_name(pool->pool_md), r);
3953 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
3955 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
3956 dm_device_name(pool->pool_md), r);
3960 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
3962 DMERR("%s: dm_pool_get_free_block_count returned %d",
3963 dm_device_name(pool->pool_md), r);
3967 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
3969 DMERR("%s: dm_pool_get_data_dev_size returned %d",
3970 dm_device_name(pool->pool_md), r);
3974 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
3976 DMERR("%s: dm_pool_get_metadata_snap returned %d",
3977 dm_device_name(pool->pool_md), r);
3981 DMEMIT("%llu %llu/%llu %llu/%llu ",
3982 (unsigned long long)transaction_id,
3983 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
3984 (unsigned long long)nr_blocks_metadata,
3985 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
3986 (unsigned long long)nr_blocks_data);
3989 DMEMIT("%llu ", held_root);
3993 mode = get_pool_mode(pool);
3994 if (mode == PM_OUT_OF_DATA_SPACE)
3995 DMEMIT("out_of_data_space ");
3996 else if (is_read_only_pool_mode(mode))
4001 if (!pool->pf.discard_enabled)
4002 DMEMIT("ignore_discard ");
4003 else if (pool->pf.discard_passdown)
4004 DMEMIT("discard_passdown ");
4006 DMEMIT("no_discard_passdown ");
4008 if (pool->pf.error_if_no_space)
4009 DMEMIT("error_if_no_space ");
4011 DMEMIT("queue_if_no_space ");
4013 if (dm_pool_metadata_needs_check(pool->pmd))
4014 DMEMIT("needs_check ");
4018 DMEMIT("%llu ", (unsigned long long)calc_metadata_threshold(pt));
4022 case STATUSTYPE_TABLE:
4023 DMEMIT("%s %s %lu %llu ",
4024 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
4025 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
4026 (unsigned long)pool->sectors_per_block,
4027 (unsigned long long)pt->low_water_blocks);
4028 emit_flags(&pt->requested_pf, result, sz, maxlen);
4037 static int pool_iterate_devices(struct dm_target *ti,
4038 iterate_devices_callout_fn fn, void *data)
4040 struct pool_c *pt = ti->private;
4042 return fn(ti, pt->data_dev, 0, ti->len, data);
4045 static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
4047 struct pool_c *pt = ti->private;
4048 struct pool *pool = pt->pool;
4049 sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
4052 * If max_sectors is smaller than pool->sectors_per_block adjust it
4053 * to the highest possible power-of-2 factor of pool->sectors_per_block.
4054 * This is especially beneficial when the pool's data device is a RAID
4055 * device that has a full stripe width that matches pool->sectors_per_block
4056 * -- because even though partial RAID stripe-sized IOs will be issued to a
4057 * single RAID stripe; when aggregated they will end on a full RAID stripe
4058 * boundary.. which avoids additional partial RAID stripe writes cascading
4060 if (limits->max_sectors < pool->sectors_per_block) {
4061 while (!is_factor(pool->sectors_per_block, limits->max_sectors)) {
4062 if ((limits->max_sectors & (limits->max_sectors - 1)) == 0)
4063 limits->max_sectors--;
4064 limits->max_sectors = rounddown_pow_of_two(limits->max_sectors);
4069 * If the system-determined stacked limits are compatible with the
4070 * pool's blocksize (io_opt is a factor) do not override them.
4072 if (io_opt_sectors < pool->sectors_per_block ||
4073 !is_factor(io_opt_sectors, pool->sectors_per_block)) {
4074 if (is_factor(pool->sectors_per_block, limits->max_sectors))
4075 blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT);
4077 blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT);
4078 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
4082 * pt->adjusted_pf is a staging area for the actual features to use.
4083 * They get transferred to the live pool in bind_control_target()
4084 * called from pool_preresume().
4086 if (!pt->adjusted_pf.discard_enabled) {
4088 * Must explicitly disallow stacking discard limits otherwise the
4089 * block layer will stack them if pool's data device has support.
4090 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
4091 * user to see that, so make sure to set all discard limits to 0.
4093 limits->discard_granularity = 0;
4097 disable_passdown_if_not_supported(pt);
4100 * The pool uses the same discard limits as the underlying data
4101 * device. DM core has already set this up.
4105 static struct target_type pool_target = {
4106 .name = "thin-pool",
4107 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
4108 DM_TARGET_IMMUTABLE,
4109 .version = {1, 22, 0},
4110 .module = THIS_MODULE,
4114 .presuspend = pool_presuspend,
4115 .presuspend_undo = pool_presuspend_undo,
4116 .postsuspend = pool_postsuspend,
4117 .preresume = pool_preresume,
4118 .resume = pool_resume,
4119 .message = pool_message,
4120 .status = pool_status,
4121 .iterate_devices = pool_iterate_devices,
4122 .io_hints = pool_io_hints,
4125 /*----------------------------------------------------------------
4126 * Thin target methods
4127 *--------------------------------------------------------------*/
4128 static void thin_get(struct thin_c *tc)
4130 refcount_inc(&tc->refcount);
4133 static void thin_put(struct thin_c *tc)
4135 if (refcount_dec_and_test(&tc->refcount))
4136 complete(&tc->can_destroy);
4139 static void thin_dtr(struct dm_target *ti)
4141 struct thin_c *tc = ti->private;
4143 spin_lock_irq(&tc->pool->lock);
4144 list_del_rcu(&tc->list);
4145 spin_unlock_irq(&tc->pool->lock);
4149 wait_for_completion(&tc->can_destroy);
4151 mutex_lock(&dm_thin_pool_table.mutex);
4153 __pool_dec(tc->pool);
4154 dm_pool_close_thin_device(tc->td);
4155 dm_put_device(ti, tc->pool_dev);
4157 dm_put_device(ti, tc->origin_dev);
4160 mutex_unlock(&dm_thin_pool_table.mutex);
4164 * Thin target parameters:
4166 * <pool_dev> <dev_id> [origin_dev]
4168 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
4169 * dev_id: the internal device identifier
4170 * origin_dev: a device external to the pool that should act as the origin
4172 * If the pool device has discards disabled, they get disabled for the thin
4175 static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
4179 struct dm_dev *pool_dev, *origin_dev;
4180 struct mapped_device *pool_md;
4182 mutex_lock(&dm_thin_pool_table.mutex);
4184 if (argc != 2 && argc != 3) {
4185 ti->error = "Invalid argument count";
4190 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
4192 ti->error = "Out of memory";
4196 tc->thin_md = dm_table_get_md(ti->table);
4197 spin_lock_init(&tc->lock);
4198 INIT_LIST_HEAD(&tc->deferred_cells);
4199 bio_list_init(&tc->deferred_bio_list);
4200 bio_list_init(&tc->retry_on_resume_list);
4201 tc->sort_bio_list = RB_ROOT;
4204 if (!strcmp(argv[0], argv[2])) {
4205 ti->error = "Error setting origin device";
4207 goto bad_origin_dev;
4210 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
4212 ti->error = "Error opening origin device";
4213 goto bad_origin_dev;
4215 tc->origin_dev = origin_dev;
4218 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
4220 ti->error = "Error opening pool device";
4223 tc->pool_dev = pool_dev;
4225 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
4226 ti->error = "Invalid device id";
4231 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
4233 ti->error = "Couldn't get pool mapped device";
4238 tc->pool = __pool_table_lookup(pool_md);
4240 ti->error = "Couldn't find pool object";
4242 goto bad_pool_lookup;
4244 __pool_inc(tc->pool);
4246 if (get_pool_mode(tc->pool) == PM_FAIL) {
4247 ti->error = "Couldn't open thin device, Pool is in fail mode";
4252 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
4254 ti->error = "Couldn't open thin internal device";
4258 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
4262 ti->num_flush_bios = 1;
4263 ti->limit_swap_bios = true;
4264 ti->flush_supported = true;
4265 ti->per_io_data_size = sizeof(struct dm_thin_endio_hook);
4267 /* In case the pool supports discards, pass them on. */
4268 if (tc->pool->pf.discard_enabled) {
4269 ti->discards_supported = true;
4270 ti->num_discard_bios = 1;
4273 mutex_unlock(&dm_thin_pool_table.mutex);
4275 spin_lock_irq(&tc->pool->lock);
4276 if (tc->pool->suspended) {
4277 spin_unlock_irq(&tc->pool->lock);
4278 mutex_lock(&dm_thin_pool_table.mutex); /* reacquire for __pool_dec */
4279 ti->error = "Unable to activate thin device while pool is suspended";
4283 refcount_set(&tc->refcount, 1);
4284 init_completion(&tc->can_destroy);
4285 list_add_tail_rcu(&tc->list, &tc->pool->active_thins);
4286 spin_unlock_irq(&tc->pool->lock);
4288 * This synchronize_rcu() call is needed here otherwise we risk a
4289 * wake_worker() call finding no bios to process (because the newly
4290 * added tc isn't yet visible). So this reduces latency since we
4291 * aren't then dependent on the periodic commit to wake_worker().
4300 dm_pool_close_thin_device(tc->td);
4302 __pool_dec(tc->pool);
4306 dm_put_device(ti, tc->pool_dev);
4309 dm_put_device(ti, tc->origin_dev);
4313 mutex_unlock(&dm_thin_pool_table.mutex);
4318 static int thin_map(struct dm_target *ti, struct bio *bio)
4320 bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
4322 return thin_bio_map(ti, bio);
4325 static int thin_endio(struct dm_target *ti, struct bio *bio,
4328 unsigned long flags;
4329 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
4330 struct list_head work;
4331 struct dm_thin_new_mapping *m, *tmp;
4332 struct pool *pool = h->tc->pool;
4334 if (h->shared_read_entry) {
4335 INIT_LIST_HEAD(&work);
4336 dm_deferred_entry_dec(h->shared_read_entry, &work);
4338 spin_lock_irqsave(&pool->lock, flags);
4339 list_for_each_entry_safe(m, tmp, &work, list) {
4341 __complete_mapping_preparation(m);
4343 spin_unlock_irqrestore(&pool->lock, flags);
4346 if (h->all_io_entry) {
4347 INIT_LIST_HEAD(&work);
4348 dm_deferred_entry_dec(h->all_io_entry, &work);
4349 if (!list_empty(&work)) {
4350 spin_lock_irqsave(&pool->lock, flags);
4351 list_for_each_entry_safe(m, tmp, &work, list)
4352 list_add_tail(&m->list, &pool->prepared_discards);
4353 spin_unlock_irqrestore(&pool->lock, flags);
4359 cell_defer_no_holder(h->tc, h->cell);
4361 return DM_ENDIO_DONE;
4364 static void thin_presuspend(struct dm_target *ti)
4366 struct thin_c *tc = ti->private;
4368 if (dm_noflush_suspending(ti))
4369 noflush_work(tc, do_noflush_start);
4372 static void thin_postsuspend(struct dm_target *ti)
4374 struct thin_c *tc = ti->private;
4377 * The dm_noflush_suspending flag has been cleared by now, so
4378 * unfortunately we must always run this.
4380 noflush_work(tc, do_noflush_stop);
4383 static int thin_preresume(struct dm_target *ti)
4385 struct thin_c *tc = ti->private;
4388 tc->origin_size = get_dev_size(tc->origin_dev->bdev);
4394 * <nr mapped sectors> <highest mapped sector>
4396 static void thin_status(struct dm_target *ti, status_type_t type,
4397 unsigned status_flags, char *result, unsigned maxlen)
4401 dm_block_t mapped, highest;
4402 char buf[BDEVNAME_SIZE];
4403 struct thin_c *tc = ti->private;
4405 if (get_pool_mode(tc->pool) == PM_FAIL) {
4414 case STATUSTYPE_INFO:
4415 r = dm_thin_get_mapped_count(tc->td, &mapped);
4417 DMERR("dm_thin_get_mapped_count returned %d", r);
4421 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
4423 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
4427 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
4429 DMEMIT("%llu", ((highest + 1) *
4430 tc->pool->sectors_per_block) - 1);
4435 case STATUSTYPE_TABLE:
4437 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
4438 (unsigned long) tc->dev_id);
4440 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
4451 static int thin_iterate_devices(struct dm_target *ti,
4452 iterate_devices_callout_fn fn, void *data)
4455 struct thin_c *tc = ti->private;
4456 struct pool *pool = tc->pool;
4459 * We can't call dm_pool_get_data_dev_size() since that blocks. So
4460 * we follow a more convoluted path through to the pool's target.
4463 return 0; /* nothing is bound */
4465 blocks = pool->ti->len;
4466 (void) sector_div(blocks, pool->sectors_per_block);
4468 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
4473 static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits)
4475 struct thin_c *tc = ti->private;
4476 struct pool *pool = tc->pool;
4478 if (!pool->pf.discard_enabled)
4481 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
4482 limits->max_discard_sectors = 2048 * 1024 * 16; /* 16G */
4485 static struct target_type thin_target = {
4487 .version = {1, 22, 0},
4488 .module = THIS_MODULE,
4492 .end_io = thin_endio,
4493 .preresume = thin_preresume,
4494 .presuspend = thin_presuspend,
4495 .postsuspend = thin_postsuspend,
4496 .status = thin_status,
4497 .iterate_devices = thin_iterate_devices,
4498 .io_hints = thin_io_hints,
4501 /*----------------------------------------------------------------*/
4503 static int __init dm_thin_init(void)
4509 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
4510 if (!_new_mapping_cache)
4513 r = dm_register_target(&thin_target);
4515 goto bad_new_mapping_cache;
4517 r = dm_register_target(&pool_target);
4519 goto bad_thin_target;
4524 dm_unregister_target(&thin_target);
4525 bad_new_mapping_cache:
4526 kmem_cache_destroy(_new_mapping_cache);
4531 static void dm_thin_exit(void)
4533 dm_unregister_target(&thin_target);
4534 dm_unregister_target(&pool_target);
4536 kmem_cache_destroy(_new_mapping_cache);
4541 module_init(dm_thin_init);
4542 module_exit(dm_thin_exit);
4544 module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR);
4545 MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds");
4547 MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
4548 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
4549 MODULE_LICENSE("GPL");