2 * Copyright (C) 2011-2012 Red Hat UK.
4 * This file is released under the GPL.
7 #include "dm-thin-metadata.h"
8 #include "dm-bio-prison.h"
11 #include <linux/device-mapper.h>
12 #include <linux/dm-io.h>
13 #include <linux/dm-kcopyd.h>
14 #include <linux/list.h>
15 #include <linux/init.h>
16 #include <linux/module.h>
17 #include <linux/slab.h>
19 #define DM_MSG_PREFIX "thin"
24 #define ENDIO_HOOK_POOL_SIZE 1024
25 #define MAPPING_POOL_SIZE 1024
26 #define PRISON_CELLS 1024
27 #define COMMIT_PERIOD HZ
29 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
30 "A percentage of time allocated for copy on write");
33 * The block size of the device holding pool data must be
34 * between 64KB and 1GB.
36 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
37 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
40 * Device id is restricted to 24 bits.
42 #define MAX_DEV_ID ((1 << 24) - 1)
45 * How do we handle breaking sharing of data blocks?
46 * =================================================
48 * We use a standard copy-on-write btree to store the mappings for the
49 * devices (note I'm talking about copy-on-write of the metadata here, not
50 * the data). When you take an internal snapshot you clone the root node
51 * of the origin btree. After this there is no concept of an origin or a
52 * snapshot. They are just two device trees that happen to point to the
55 * When we get a write in we decide if it's to a shared data block using
56 * some timestamp magic. If it is, we have to break sharing.
58 * Let's say we write to a shared block in what was the origin. The
61 * i) plug io further to this physical block. (see bio_prison code).
63 * ii) quiesce any read io to that shared data block. Obviously
64 * including all devices that share this block. (see dm_deferred_set code)
66 * iii) copy the data block to a newly allocate block. This step can be
67 * missed out if the io covers the block. (schedule_copy).
69 * iv) insert the new mapping into the origin's btree
70 * (process_prepared_mapping). This act of inserting breaks some
71 * sharing of btree nodes between the two devices. Breaking sharing only
72 * effects the btree of that specific device. Btrees for the other
73 * devices that share the block never change. The btree for the origin
74 * device as it was after the last commit is untouched, ie. we're using
75 * persistent data structures in the functional programming sense.
77 * v) unplug io to this physical block, including the io that triggered
78 * the breaking of sharing.
80 * Steps (ii) and (iii) occur in parallel.
82 * The metadata _doesn't_ need to be committed before the io continues. We
83 * get away with this because the io is always written to a _new_ block.
84 * If there's a crash, then:
86 * - The origin mapping will point to the old origin block (the shared
87 * one). This will contain the data as it was before the io that triggered
88 * the breaking of sharing came in.
90 * - The snap mapping still points to the old block. As it would after
93 * The downside of this scheme is the timestamp magic isn't perfect, and
94 * will continue to think that data block in the snapshot device is shared
95 * even after the write to the origin has broken sharing. I suspect data
96 * blocks will typically be shared by many different devices, so we're
97 * breaking sharing n + 1 times, rather than n, where n is the number of
98 * devices that reference this data block. At the moment I think the
99 * benefits far, far outweigh the disadvantages.
102 /*----------------------------------------------------------------*/
107 static void build_data_key(struct dm_thin_device *td,
108 dm_block_t b, struct dm_cell_key *key)
111 key->dev = dm_thin_dev_id(td);
115 static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
116 struct dm_cell_key *key)
119 key->dev = dm_thin_dev_id(td);
123 /*----------------------------------------------------------------*/
126 * A pool device ties together a metadata device and a data device. It
127 * also provides the interface for creating and destroying internal
130 struct dm_thin_new_mapping;
133 * The pool runs in 3 modes. Ordered in degraded order for comparisons.
136 PM_WRITE, /* metadata may be changed */
137 PM_READ_ONLY, /* metadata may not be changed */
138 PM_FAIL, /* all I/O fails */
141 struct pool_features {
144 bool zero_new_blocks:1;
145 bool discard_enabled:1;
146 bool discard_passdown:1;
150 typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
151 typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);
154 struct list_head list;
155 struct dm_target *ti; /* Only set if a pool target is bound */
157 struct mapped_device *pool_md;
158 struct block_device *md_dev;
159 struct dm_pool_metadata *pmd;
161 dm_block_t low_water_blocks;
162 uint32_t sectors_per_block;
163 int sectors_per_block_shift;
165 struct pool_features pf;
166 unsigned low_water_triggered:1; /* A dm event has been sent */
167 unsigned no_free_space:1; /* A -ENOSPC warning has been issued */
169 struct dm_bio_prison *prison;
170 struct dm_kcopyd_client *copier;
172 struct workqueue_struct *wq;
173 struct work_struct worker;
174 struct delayed_work waker;
176 unsigned long last_commit_jiffies;
180 struct bio_list deferred_bios;
181 struct bio_list deferred_flush_bios;
182 struct list_head prepared_mappings;
183 struct list_head prepared_discards;
185 struct bio_list retry_on_resume_list;
187 struct dm_deferred_set *shared_read_ds;
188 struct dm_deferred_set *all_io_ds;
190 struct dm_thin_new_mapping *next_mapping;
191 mempool_t *mapping_pool;
193 process_bio_fn process_bio;
194 process_bio_fn process_discard;
196 process_mapping_fn process_prepared_mapping;
197 process_mapping_fn process_prepared_discard;
200 static enum pool_mode get_pool_mode(struct pool *pool);
201 static void set_pool_mode(struct pool *pool, enum pool_mode mode);
204 * Target context for a pool.
207 struct dm_target *ti;
209 struct dm_dev *data_dev;
210 struct dm_dev *metadata_dev;
211 struct dm_target_callbacks callbacks;
213 dm_block_t low_water_blocks;
214 struct pool_features requested_pf; /* Features requested during table load */
215 struct pool_features adjusted_pf; /* Features used after adjusting for constituent devices */
219 * Target context for a thin.
222 struct dm_dev *pool_dev;
223 struct dm_dev *origin_dev;
227 struct dm_thin_device *td;
230 /*----------------------------------------------------------------*/
233 * wake_worker() is used when new work is queued and when pool_resume is
234 * ready to continue deferred IO processing.
236 static void wake_worker(struct pool *pool)
238 queue_work(pool->wq, &pool->worker);
241 /*----------------------------------------------------------------*/
243 static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio,
244 struct dm_bio_prison_cell **cell_result)
247 struct dm_bio_prison_cell *cell_prealloc;
250 * Allocate a cell from the prison's mempool.
251 * This might block but it can't fail.
253 cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO);
255 r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result);
258 * We reused an old cell; we can get rid of
261 dm_bio_prison_free_cell(pool->prison, cell_prealloc);
266 static void cell_release(struct pool *pool,
267 struct dm_bio_prison_cell *cell,
268 struct bio_list *bios)
270 dm_cell_release(pool->prison, cell, bios);
271 dm_bio_prison_free_cell(pool->prison, cell);
274 static void cell_release_no_holder(struct pool *pool,
275 struct dm_bio_prison_cell *cell,
276 struct bio_list *bios)
278 dm_cell_release_no_holder(pool->prison, cell, bios);
279 dm_bio_prison_free_cell(pool->prison, cell);
282 static void cell_defer_no_holder_no_free(struct thin_c *tc,
283 struct dm_bio_prison_cell *cell)
285 struct pool *pool = tc->pool;
288 spin_lock_irqsave(&pool->lock, flags);
289 dm_cell_release_no_holder(pool->prison, cell, &pool->deferred_bios);
290 spin_unlock_irqrestore(&pool->lock, flags);
295 static void cell_error(struct pool *pool,
296 struct dm_bio_prison_cell *cell)
298 dm_cell_error(pool->prison, cell);
299 dm_bio_prison_free_cell(pool->prison, cell);
302 /*----------------------------------------------------------------*/
305 * A global list of pools that uses a struct mapped_device as a key.
307 static struct dm_thin_pool_table {
309 struct list_head pools;
310 } dm_thin_pool_table;
312 static void pool_table_init(void)
314 mutex_init(&dm_thin_pool_table.mutex);
315 INIT_LIST_HEAD(&dm_thin_pool_table.pools);
318 static void __pool_table_insert(struct pool *pool)
320 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
321 list_add(&pool->list, &dm_thin_pool_table.pools);
324 static void __pool_table_remove(struct pool *pool)
326 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
327 list_del(&pool->list);
330 static struct pool *__pool_table_lookup(struct mapped_device *md)
332 struct pool *pool = NULL, *tmp;
334 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
336 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
337 if (tmp->pool_md == md) {
346 static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
348 struct pool *pool = NULL, *tmp;
350 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
352 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
353 if (tmp->md_dev == md_dev) {
362 /*----------------------------------------------------------------*/
364 struct dm_thin_endio_hook {
366 struct dm_deferred_entry *shared_read_entry;
367 struct dm_deferred_entry *all_io_entry;
368 struct dm_thin_new_mapping *overwrite_mapping;
371 static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master)
374 struct bio_list bios;
376 bio_list_init(&bios);
377 bio_list_merge(&bios, master);
378 bio_list_init(master);
380 while ((bio = bio_list_pop(&bios))) {
381 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
384 bio_endio(bio, DM_ENDIO_REQUEUE);
386 bio_list_add(master, bio);
390 static void requeue_io(struct thin_c *tc)
392 struct pool *pool = tc->pool;
395 spin_lock_irqsave(&pool->lock, flags);
396 __requeue_bio_list(tc, &pool->deferred_bios);
397 __requeue_bio_list(tc, &pool->retry_on_resume_list);
398 spin_unlock_irqrestore(&pool->lock, flags);
402 * This section of code contains the logic for processing a thin device's IO.
403 * Much of the code depends on pool object resources (lists, workqueues, etc)
404 * but most is exclusively called from the thin target rather than the thin-pool
408 static bool block_size_is_power_of_two(struct pool *pool)
410 return pool->sectors_per_block_shift >= 0;
413 static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
415 struct pool *pool = tc->pool;
416 sector_t block_nr = bio->bi_sector;
418 if (block_size_is_power_of_two(pool))
419 block_nr >>= pool->sectors_per_block_shift;
421 (void) sector_div(block_nr, pool->sectors_per_block);
426 static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
428 struct pool *pool = tc->pool;
429 sector_t bi_sector = bio->bi_sector;
431 bio->bi_bdev = tc->pool_dev->bdev;
432 if (block_size_is_power_of_two(pool))
433 bio->bi_sector = (block << pool->sectors_per_block_shift) |
434 (bi_sector & (pool->sectors_per_block - 1));
436 bio->bi_sector = (block * pool->sectors_per_block) +
437 sector_div(bi_sector, pool->sectors_per_block);
440 static void remap_to_origin(struct thin_c *tc, struct bio *bio)
442 bio->bi_bdev = tc->origin_dev->bdev;
445 static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
447 return (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) &&
448 dm_thin_changed_this_transaction(tc->td);
451 static void inc_all_io_entry(struct pool *pool, struct bio *bio)
453 struct dm_thin_endio_hook *h;
455 if (bio->bi_rw & REQ_DISCARD)
458 h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
459 h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
462 static void issue(struct thin_c *tc, struct bio *bio)
464 struct pool *pool = tc->pool;
467 if (!bio_triggers_commit(tc, bio)) {
468 generic_make_request(bio);
473 * Complete bio with an error if earlier I/O caused changes to
474 * the metadata that can't be committed e.g, due to I/O errors
475 * on the metadata device.
477 if (dm_thin_aborted_changes(tc->td)) {
483 * Batch together any bios that trigger commits and then issue a
484 * single commit for them in process_deferred_bios().
486 spin_lock_irqsave(&pool->lock, flags);
487 bio_list_add(&pool->deferred_flush_bios, bio);
488 spin_unlock_irqrestore(&pool->lock, flags);
491 static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
493 remap_to_origin(tc, bio);
497 static void remap_and_issue(struct thin_c *tc, struct bio *bio,
500 remap(tc, bio, block);
504 /*----------------------------------------------------------------*/
507 * Bio endio functions.
509 struct dm_thin_new_mapping {
510 struct list_head list;
514 unsigned pass_discard:1;
517 dm_block_t virt_block;
518 dm_block_t data_block;
519 struct dm_bio_prison_cell *cell, *cell2;
523 * If the bio covers the whole area of a block then we can avoid
524 * zeroing or copying. Instead this bio is hooked. The bio will
525 * still be in the cell, so care has to be taken to avoid issuing
529 bio_end_io_t *saved_bi_end_io;
532 static void __maybe_add_mapping(struct dm_thin_new_mapping *m)
534 struct pool *pool = m->tc->pool;
536 if (m->quiesced && m->prepared) {
537 list_add(&m->list, &pool->prepared_mappings);
542 static void copy_complete(int read_err, unsigned long write_err, void *context)
545 struct dm_thin_new_mapping *m = context;
546 struct pool *pool = m->tc->pool;
548 m->err = read_err || write_err ? -EIO : 0;
550 spin_lock_irqsave(&pool->lock, flags);
552 __maybe_add_mapping(m);
553 spin_unlock_irqrestore(&pool->lock, flags);
556 static void overwrite_endio(struct bio *bio, int err)
559 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
560 struct dm_thin_new_mapping *m = h->overwrite_mapping;
561 struct pool *pool = m->tc->pool;
565 spin_lock_irqsave(&pool->lock, flags);
567 __maybe_add_mapping(m);
568 spin_unlock_irqrestore(&pool->lock, flags);
571 /*----------------------------------------------------------------*/
578 * Prepared mapping jobs.
582 * This sends the bios in the cell back to the deferred_bios list.
584 static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell)
586 struct pool *pool = tc->pool;
589 spin_lock_irqsave(&pool->lock, flags);
590 cell_release(pool, cell, &pool->deferred_bios);
591 spin_unlock_irqrestore(&tc->pool->lock, flags);
597 * Same as cell_defer above, except it omits the original holder of the cell.
599 static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
601 struct pool *pool = tc->pool;
604 spin_lock_irqsave(&pool->lock, flags);
605 cell_release_no_holder(pool, cell, &pool->deferred_bios);
606 spin_unlock_irqrestore(&pool->lock, flags);
611 static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
614 m->bio->bi_end_io = m->saved_bi_end_io;
615 cell_error(m->tc->pool, m->cell);
617 mempool_free(m, m->tc->pool->mapping_pool);
620 static void process_prepared_mapping(struct dm_thin_new_mapping *m)
622 struct thin_c *tc = m->tc;
623 struct pool *pool = tc->pool;
629 bio->bi_end_io = m->saved_bi_end_io;
632 cell_error(pool, m->cell);
637 * Commit the prepared block into the mapping btree.
638 * Any I/O for this block arriving after this point will get
639 * remapped to it directly.
641 r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block);
643 DMERR_LIMIT("%s: dm_thin_insert_block() failed: error = %d",
644 dm_device_name(pool->pool_md), r);
645 set_pool_mode(pool, PM_READ_ONLY);
646 cell_error(pool, m->cell);
651 * Release any bios held while the block was being provisioned.
652 * If we are processing a write bio that completely covers the block,
653 * we already processed it so can ignore it now when processing
654 * the bios in the cell.
657 cell_defer_no_holder(tc, m->cell);
660 cell_defer(tc, m->cell);
664 mempool_free(m, pool->mapping_pool);
667 static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
669 struct thin_c *tc = m->tc;
671 bio_io_error(m->bio);
672 cell_defer_no_holder(tc, m->cell);
673 cell_defer_no_holder(tc, m->cell2);
674 mempool_free(m, tc->pool->mapping_pool);
677 static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m)
679 struct thin_c *tc = m->tc;
681 inc_all_io_entry(tc->pool, m->bio);
682 cell_defer_no_holder(tc, m->cell);
683 cell_defer_no_holder(tc, m->cell2);
686 remap_and_issue(tc, m->bio, m->data_block);
688 bio_endio(m->bio, 0);
690 mempool_free(m, tc->pool->mapping_pool);
693 static void process_prepared_discard(struct dm_thin_new_mapping *m)
696 struct thin_c *tc = m->tc;
698 r = dm_thin_remove_block(tc->td, m->virt_block);
700 DMERR_LIMIT("dm_thin_remove_block() failed");
702 process_prepared_discard_passdown(m);
705 static void process_prepared(struct pool *pool, struct list_head *head,
706 process_mapping_fn *fn)
709 struct list_head maps;
710 struct dm_thin_new_mapping *m, *tmp;
712 INIT_LIST_HEAD(&maps);
713 spin_lock_irqsave(&pool->lock, flags);
714 list_splice_init(head, &maps);
715 spin_unlock_irqrestore(&pool->lock, flags);
717 list_for_each_entry_safe(m, tmp, &maps, list)
724 static int io_overlaps_block(struct pool *pool, struct bio *bio)
726 return bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT);
729 static int io_overwrites_block(struct pool *pool, struct bio *bio)
731 return (bio_data_dir(bio) == WRITE) &&
732 io_overlaps_block(pool, bio);
735 static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
738 *save = bio->bi_end_io;
742 static int ensure_next_mapping(struct pool *pool)
744 if (pool->next_mapping)
747 pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);
749 return pool->next_mapping ? 0 : -ENOMEM;
752 static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
754 struct dm_thin_new_mapping *r = pool->next_mapping;
756 BUG_ON(!pool->next_mapping);
758 pool->next_mapping = NULL;
763 static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
764 struct dm_dev *origin, dm_block_t data_origin,
765 dm_block_t data_dest,
766 struct dm_bio_prison_cell *cell, struct bio *bio)
769 struct pool *pool = tc->pool;
770 struct dm_thin_new_mapping *m = get_next_mapping(pool);
772 INIT_LIST_HEAD(&m->list);
776 m->virt_block = virt_block;
777 m->data_block = data_dest;
782 if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
786 * IO to pool_dev remaps to the pool target's data_dev.
788 * If the whole block of data is being overwritten, we can issue the
789 * bio immediately. Otherwise we use kcopyd to clone the data first.
791 if (io_overwrites_block(pool, bio)) {
792 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
794 h->overwrite_mapping = m;
796 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
797 inc_all_io_entry(pool, bio);
798 remap_and_issue(tc, bio, data_dest);
800 struct dm_io_region from, to;
802 from.bdev = origin->bdev;
803 from.sector = data_origin * pool->sectors_per_block;
804 from.count = pool->sectors_per_block;
806 to.bdev = tc->pool_dev->bdev;
807 to.sector = data_dest * pool->sectors_per_block;
808 to.count = pool->sectors_per_block;
810 r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
811 0, copy_complete, m);
813 mempool_free(m, pool->mapping_pool);
814 DMERR_LIMIT("dm_kcopyd_copy() failed");
815 cell_error(pool, cell);
820 static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
821 dm_block_t data_origin, dm_block_t data_dest,
822 struct dm_bio_prison_cell *cell, struct bio *bio)
824 schedule_copy(tc, virt_block, tc->pool_dev,
825 data_origin, data_dest, cell, bio);
828 static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
829 dm_block_t data_dest,
830 struct dm_bio_prison_cell *cell, struct bio *bio)
832 schedule_copy(tc, virt_block, tc->origin_dev,
833 virt_block, data_dest, cell, bio);
836 static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
837 dm_block_t data_block, struct dm_bio_prison_cell *cell,
840 struct pool *pool = tc->pool;
841 struct dm_thin_new_mapping *m = get_next_mapping(pool);
843 INIT_LIST_HEAD(&m->list);
847 m->virt_block = virt_block;
848 m->data_block = data_block;
854 * If the whole block of data is being overwritten or we are not
855 * zeroing pre-existing data, we can issue the bio immediately.
856 * Otherwise we use kcopyd to zero the data first.
858 if (!pool->pf.zero_new_blocks)
859 process_prepared_mapping(m);
861 else if (io_overwrites_block(pool, bio)) {
862 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
864 h->overwrite_mapping = m;
866 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
867 inc_all_io_entry(pool, bio);
868 remap_and_issue(tc, bio, data_block);
871 struct dm_io_region to;
873 to.bdev = tc->pool_dev->bdev;
874 to.sector = data_block * pool->sectors_per_block;
875 to.count = pool->sectors_per_block;
877 r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m);
879 mempool_free(m, pool->mapping_pool);
880 DMERR_LIMIT("dm_kcopyd_zero() failed");
881 cell_error(pool, cell);
887 * A non-zero return indicates read_only or fail_io mode.
888 * Many callers don't care about the return value.
890 static int commit(struct pool *pool)
894 if (get_pool_mode(pool) != PM_WRITE)
897 r = dm_pool_commit_metadata(pool->pmd);
899 DMERR_LIMIT("%s: dm_pool_commit_metadata failed: error = %d",
900 dm_device_name(pool->pool_md), r);
901 set_pool_mode(pool, PM_READ_ONLY);
907 static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
910 dm_block_t free_blocks;
912 struct pool *pool = tc->pool;
915 * Once no_free_space is set we must not allow allocation to succeed.
916 * Otherwise it is difficult to explain, debug, test and support.
918 if (pool->no_free_space)
921 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
925 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
926 DMWARN("%s: reached low water mark for data device: sending event.",
927 dm_device_name(pool->pool_md));
928 spin_lock_irqsave(&pool->lock, flags);
929 pool->low_water_triggered = 1;
930 spin_unlock_irqrestore(&pool->lock, flags);
931 dm_table_event(pool->ti->table);
936 * Try to commit to see if that will free up some
943 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
948 * If we still have no space we set a flag to avoid
949 * doing all this checking and return -ENOSPC. This
950 * flag serves as a latch that disallows allocations from
951 * this pool until the admin takes action (e.g. resize or
955 DMWARN("%s: no free data space available.",
956 dm_device_name(pool->pool_md));
957 spin_lock_irqsave(&pool->lock, flags);
958 pool->no_free_space = 1;
959 spin_unlock_irqrestore(&pool->lock, flags);
964 r = dm_pool_alloc_data_block(pool->pmd, result);
967 !dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks) &&
969 DMWARN("%s: no free metadata space available.",
970 dm_device_name(pool->pool_md));
971 set_pool_mode(pool, PM_READ_ONLY);
980 * If we have run out of space, queue bios until the device is
981 * resumed, presumably after having been reloaded with more space.
983 static void retry_on_resume(struct bio *bio)
985 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
986 struct thin_c *tc = h->tc;
987 struct pool *pool = tc->pool;
990 spin_lock_irqsave(&pool->lock, flags);
991 bio_list_add(&pool->retry_on_resume_list, bio);
992 spin_unlock_irqrestore(&pool->lock, flags);
995 static void no_space(struct pool *pool, struct dm_bio_prison_cell *cell)
998 struct bio_list bios;
1000 bio_list_init(&bios);
1001 cell_release(pool, cell, &bios);
1003 while ((bio = bio_list_pop(&bios)))
1004 retry_on_resume(bio);
1007 static void process_discard(struct thin_c *tc, struct bio *bio)
1010 unsigned long flags;
1011 struct pool *pool = tc->pool;
1012 struct dm_bio_prison_cell *cell, *cell2;
1013 struct dm_cell_key key, key2;
1014 dm_block_t block = get_bio_block(tc, bio);
1015 struct dm_thin_lookup_result lookup_result;
1016 struct dm_thin_new_mapping *m;
1018 build_virtual_key(tc->td, block, &key);
1019 if (bio_detain(tc->pool, &key, bio, &cell))
1022 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1026 * Check nobody is fiddling with this pool block. This can
1027 * happen if someone's in the process of breaking sharing
1030 build_data_key(tc->td, lookup_result.block, &key2);
1031 if (bio_detain(tc->pool, &key2, bio, &cell2)) {
1032 cell_defer_no_holder(tc, cell);
1036 if (io_overlaps_block(pool, bio)) {
1038 * IO may still be going to the destination block. We must
1039 * quiesce before we can do the removal.
1041 m = get_next_mapping(pool);
1043 m->pass_discard = (!lookup_result.shared) && pool->pf.discard_passdown;
1044 m->virt_block = block;
1045 m->data_block = lookup_result.block;
1051 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) {
1052 spin_lock_irqsave(&pool->lock, flags);
1053 list_add(&m->list, &pool->prepared_discards);
1054 spin_unlock_irqrestore(&pool->lock, flags);
1058 inc_all_io_entry(pool, bio);
1059 cell_defer_no_holder(tc, cell);
1060 cell_defer_no_holder(tc, cell2);
1063 * The DM core makes sure that the discard doesn't span
1064 * a block boundary. So we submit the discard of a
1065 * partial block appropriately.
1067 if ((!lookup_result.shared) && pool->pf.discard_passdown)
1068 remap_and_issue(tc, bio, lookup_result.block);
1076 * It isn't provisioned, just forget it.
1078 cell_defer_no_holder(tc, cell);
1083 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1085 cell_defer_no_holder(tc, cell);
1091 static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
1092 struct dm_cell_key *key,
1093 struct dm_thin_lookup_result *lookup_result,
1094 struct dm_bio_prison_cell *cell)
1097 dm_block_t data_block;
1098 struct pool *pool = tc->pool;
1100 r = alloc_data_block(tc, &data_block);
1103 schedule_internal_copy(tc, block, lookup_result->block,
1104 data_block, cell, bio);
1108 no_space(pool, cell);
1112 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1114 set_pool_mode(pool, PM_READ_ONLY);
1115 cell_error(pool, cell);
1120 static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1122 struct dm_thin_lookup_result *lookup_result)
1124 struct dm_bio_prison_cell *cell;
1125 struct pool *pool = tc->pool;
1126 struct dm_cell_key key;
1129 * If cell is already occupied, then sharing is already in the process
1130 * of being broken so we have nothing further to do here.
1132 build_data_key(tc->td, lookup_result->block, &key);
1133 if (bio_detain(pool, &key, bio, &cell))
1136 if (bio_data_dir(bio) == WRITE && bio->bi_size)
1137 break_sharing(tc, bio, block, &key, lookup_result, cell);
1139 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1141 h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
1142 inc_all_io_entry(pool, bio);
1143 cell_defer_no_holder(tc, cell);
1145 remap_and_issue(tc, bio, lookup_result->block);
1149 static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
1150 struct dm_bio_prison_cell *cell)
1153 dm_block_t data_block;
1154 struct pool *pool = tc->pool;
1157 * Remap empty bios (flushes) immediately, without provisioning.
1159 if (!bio->bi_size) {
1160 inc_all_io_entry(pool, bio);
1161 cell_defer_no_holder(tc, cell);
1163 remap_and_issue(tc, bio, 0);
1168 * Fill read bios with zeroes and complete them immediately.
1170 if (bio_data_dir(bio) == READ) {
1172 cell_defer_no_holder(tc, cell);
1177 r = alloc_data_block(tc, &data_block);
1181 schedule_external_copy(tc, block, data_block, cell, bio);
1183 schedule_zero(tc, block, data_block, cell, bio);
1187 no_space(pool, cell);
1191 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1193 set_pool_mode(pool, PM_READ_ONLY);
1194 cell_error(pool, cell);
1199 static void process_bio(struct thin_c *tc, struct bio *bio)
1202 struct pool *pool = tc->pool;
1203 dm_block_t block = get_bio_block(tc, bio);
1204 struct dm_bio_prison_cell *cell;
1205 struct dm_cell_key key;
1206 struct dm_thin_lookup_result lookup_result;
1209 * If cell is already occupied, then the block is already
1210 * being provisioned so we have nothing further to do here.
1212 build_virtual_key(tc->td, block, &key);
1213 if (bio_detain(pool, &key, bio, &cell))
1216 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1219 if (lookup_result.shared) {
1220 process_shared_bio(tc, bio, block, &lookup_result);
1221 cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */
1223 inc_all_io_entry(pool, bio);
1224 cell_defer_no_holder(tc, cell);
1226 remap_and_issue(tc, bio, lookup_result.block);
1231 if (bio_data_dir(bio) == READ && tc->origin_dev) {
1232 inc_all_io_entry(pool, bio);
1233 cell_defer_no_holder(tc, cell);
1235 remap_to_origin_and_issue(tc, bio);
1237 provision_block(tc, bio, block, cell);
1241 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1243 cell_defer_no_holder(tc, cell);
1249 static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
1252 int rw = bio_data_dir(bio);
1253 dm_block_t block = get_bio_block(tc, bio);
1254 struct dm_thin_lookup_result lookup_result;
1256 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1259 if (lookup_result.shared && (rw == WRITE) && bio->bi_size)
1262 inc_all_io_entry(tc->pool, bio);
1263 remap_and_issue(tc, bio, lookup_result.block);
1273 if (tc->origin_dev) {
1274 inc_all_io_entry(tc->pool, bio);
1275 remap_to_origin_and_issue(tc, bio);
1284 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1291 static void process_bio_fail(struct thin_c *tc, struct bio *bio)
1297 * FIXME: should we also commit due to size of transaction, measured in
1300 static int need_commit_due_to_time(struct pool *pool)
1302 return jiffies < pool->last_commit_jiffies ||
1303 jiffies > pool->last_commit_jiffies + COMMIT_PERIOD;
1306 static void process_deferred_bios(struct pool *pool)
1308 unsigned long flags;
1310 struct bio_list bios;
1312 bio_list_init(&bios);
1314 spin_lock_irqsave(&pool->lock, flags);
1315 bio_list_merge(&bios, &pool->deferred_bios);
1316 bio_list_init(&pool->deferred_bios);
1317 spin_unlock_irqrestore(&pool->lock, flags);
1319 while ((bio = bio_list_pop(&bios))) {
1320 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1321 struct thin_c *tc = h->tc;
1324 * If we've got no free new_mapping structs, and processing
1325 * this bio might require one, we pause until there are some
1326 * prepared mappings to process.
1328 if (ensure_next_mapping(pool)) {
1329 spin_lock_irqsave(&pool->lock, flags);
1330 bio_list_merge(&pool->deferred_bios, &bios);
1331 spin_unlock_irqrestore(&pool->lock, flags);
1336 if (bio->bi_rw & REQ_DISCARD)
1337 pool->process_discard(tc, bio);
1339 pool->process_bio(tc, bio);
1343 * If there are any deferred flush bios, we must commit
1344 * the metadata before issuing them.
1346 bio_list_init(&bios);
1347 spin_lock_irqsave(&pool->lock, flags);
1348 bio_list_merge(&bios, &pool->deferred_flush_bios);
1349 bio_list_init(&pool->deferred_flush_bios);
1350 spin_unlock_irqrestore(&pool->lock, flags);
1352 if (bio_list_empty(&bios) && !need_commit_due_to_time(pool))
1356 while ((bio = bio_list_pop(&bios)))
1360 pool->last_commit_jiffies = jiffies;
1362 while ((bio = bio_list_pop(&bios)))
1363 generic_make_request(bio);
1366 static void do_worker(struct work_struct *ws)
1368 struct pool *pool = container_of(ws, struct pool, worker);
1370 process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
1371 process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
1372 process_deferred_bios(pool);
1376 * We want to commit periodically so that not too much
1377 * unwritten data builds up.
1379 static void do_waker(struct work_struct *ws)
1381 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
1383 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
1386 /*----------------------------------------------------------------*/
1388 static enum pool_mode get_pool_mode(struct pool *pool)
1390 return pool->pf.mode;
1393 static void set_pool_mode(struct pool *pool, enum pool_mode mode)
1397 pool->pf.mode = mode;
1401 DMERR("%s: switching pool to failure mode",
1402 dm_device_name(pool->pool_md));
1403 dm_pool_metadata_read_only(pool->pmd);
1404 pool->process_bio = process_bio_fail;
1405 pool->process_discard = process_bio_fail;
1406 pool->process_prepared_mapping = process_prepared_mapping_fail;
1407 pool->process_prepared_discard = process_prepared_discard_fail;
1411 DMERR("%s: switching pool to read-only mode",
1412 dm_device_name(pool->pool_md));
1413 r = dm_pool_abort_metadata(pool->pmd);
1415 DMERR("%s: aborting transaction failed",
1416 dm_device_name(pool->pool_md));
1417 set_pool_mode(pool, PM_FAIL);
1419 dm_pool_metadata_read_only(pool->pmd);
1420 pool->process_bio = process_bio_read_only;
1421 pool->process_discard = process_discard;
1422 pool->process_prepared_mapping = process_prepared_mapping_fail;
1423 pool->process_prepared_discard = process_prepared_discard_passdown;
1428 dm_pool_metadata_read_write(pool->pmd);
1429 pool->process_bio = process_bio;
1430 pool->process_discard = process_discard;
1431 pool->process_prepared_mapping = process_prepared_mapping;
1432 pool->process_prepared_discard = process_prepared_discard;
1437 /*----------------------------------------------------------------*/
1440 * Mapping functions.
1444 * Called only while mapping a thin bio to hand it over to the workqueue.
1446 static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
1448 unsigned long flags;
1449 struct pool *pool = tc->pool;
1451 spin_lock_irqsave(&pool->lock, flags);
1452 bio_list_add(&pool->deferred_bios, bio);
1453 spin_unlock_irqrestore(&pool->lock, flags);
1458 static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
1460 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1463 h->shared_read_entry = NULL;
1464 h->all_io_entry = NULL;
1465 h->overwrite_mapping = NULL;
1469 * Non-blocking function called from the thin target's map function.
1471 static int thin_bio_map(struct dm_target *ti, struct bio *bio)
1474 struct thin_c *tc = ti->private;
1475 dm_block_t block = get_bio_block(tc, bio);
1476 struct dm_thin_device *td = tc->td;
1477 struct dm_thin_lookup_result result;
1478 struct dm_bio_prison_cell cell1, cell2;
1479 struct dm_bio_prison_cell *cell_result;
1480 struct dm_cell_key key;
1482 thin_hook_bio(tc, bio);
1484 if (get_pool_mode(tc->pool) == PM_FAIL) {
1486 return DM_MAPIO_SUBMITTED;
1489 if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) {
1490 thin_defer_bio(tc, bio);
1491 return DM_MAPIO_SUBMITTED;
1494 r = dm_thin_find_block(td, block, 0, &result);
1497 * Note that we defer readahead too.
1501 if (unlikely(result.shared)) {
1503 * We have a race condition here between the
1504 * result.shared value returned by the lookup and
1505 * snapshot creation, which may cause new
1508 * To avoid this always quiesce the origin before
1509 * taking the snap. You want to do this anyway to
1510 * ensure a consistent application view
1513 * More distant ancestors are irrelevant. The
1514 * shared flag will be set in their case.
1516 thin_defer_bio(tc, bio);
1517 return DM_MAPIO_SUBMITTED;
1520 build_virtual_key(tc->td, block, &key);
1521 if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result))
1522 return DM_MAPIO_SUBMITTED;
1524 build_data_key(tc->td, result.block, &key);
1525 if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) {
1526 cell_defer_no_holder_no_free(tc, &cell1);
1527 return DM_MAPIO_SUBMITTED;
1530 inc_all_io_entry(tc->pool, bio);
1531 cell_defer_no_holder_no_free(tc, &cell2);
1532 cell_defer_no_holder_no_free(tc, &cell1);
1534 remap(tc, bio, result.block);
1535 return DM_MAPIO_REMAPPED;
1538 if (get_pool_mode(tc->pool) == PM_READ_ONLY) {
1540 * This block isn't provisioned, and we have no way
1541 * of doing so. Just error it.
1544 return DM_MAPIO_SUBMITTED;
1550 * In future, the failed dm_thin_find_block above could
1551 * provide the hint to load the metadata into cache.
1553 thin_defer_bio(tc, bio);
1554 return DM_MAPIO_SUBMITTED;
1558 * Must always call bio_io_error on failure.
1559 * dm_thin_find_block can fail with -EINVAL if the
1560 * pool is switched to fail-io mode.
1563 return DM_MAPIO_SUBMITTED;
1567 static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
1570 unsigned long flags;
1571 struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
1573 spin_lock_irqsave(&pt->pool->lock, flags);
1574 r = !bio_list_empty(&pt->pool->retry_on_resume_list);
1575 spin_unlock_irqrestore(&pt->pool->lock, flags);
1578 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1579 r = bdi_congested(&q->backing_dev_info, bdi_bits);
1585 static void __requeue_bios(struct pool *pool)
1587 bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list);
1588 bio_list_init(&pool->retry_on_resume_list);
1591 /*----------------------------------------------------------------
1592 * Binding of control targets to a pool object
1593 *--------------------------------------------------------------*/
1594 static bool data_dev_supports_discard(struct pool_c *pt)
1596 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1598 return q && blk_queue_discard(q);
1601 static bool is_factor(sector_t block_size, uint32_t n)
1603 return !sector_div(block_size, n);
1607 * If discard_passdown was enabled verify that the data device
1608 * supports discards. Disable discard_passdown if not.
1610 static void disable_passdown_if_not_supported(struct pool_c *pt)
1612 struct pool *pool = pt->pool;
1613 struct block_device *data_bdev = pt->data_dev->bdev;
1614 struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
1615 sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT;
1616 const char *reason = NULL;
1617 char buf[BDEVNAME_SIZE];
1619 if (!pt->adjusted_pf.discard_passdown)
1622 if (!data_dev_supports_discard(pt))
1623 reason = "discard unsupported";
1625 else if (data_limits->max_discard_sectors < pool->sectors_per_block)
1626 reason = "max discard sectors smaller than a block";
1628 else if (data_limits->discard_granularity > block_size)
1629 reason = "discard granularity larger than a block";
1631 else if (!is_factor(block_size, data_limits->discard_granularity))
1632 reason = "discard granularity not a factor of block size";
1635 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason);
1636 pt->adjusted_pf.discard_passdown = false;
1640 static int bind_control_target(struct pool *pool, struct dm_target *ti)
1642 struct pool_c *pt = ti->private;
1645 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
1647 enum pool_mode old_mode = pool->pf.mode;
1648 enum pool_mode new_mode = pt->adjusted_pf.mode;
1651 * If we were in PM_FAIL mode, rollback of metadata failed. We're
1652 * not going to recover without a thin_repair. So we never let the
1653 * pool move out of the old mode. On the other hand a PM_READ_ONLY
1654 * may have been due to a lack of metadata or data space, and may
1655 * now work (ie. if the underlying devices have been resized).
1657 if (old_mode == PM_FAIL)
1658 new_mode = old_mode;
1661 pool->low_water_blocks = pt->low_water_blocks;
1662 pool->pf = pt->adjusted_pf;
1664 set_pool_mode(pool, new_mode);
1669 static void unbind_control_target(struct pool *pool, struct dm_target *ti)
1675 /*----------------------------------------------------------------
1677 *--------------------------------------------------------------*/
1678 /* Initialize pool features. */
1679 static void pool_features_init(struct pool_features *pf)
1681 pf->mode = PM_WRITE;
1682 pf->zero_new_blocks = true;
1683 pf->discard_enabled = true;
1684 pf->discard_passdown = true;
1687 static void __pool_destroy(struct pool *pool)
1689 __pool_table_remove(pool);
1691 if (dm_pool_metadata_close(pool->pmd) < 0)
1692 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1694 dm_bio_prison_destroy(pool->prison);
1695 dm_kcopyd_client_destroy(pool->copier);
1698 destroy_workqueue(pool->wq);
1700 if (pool->next_mapping)
1701 mempool_free(pool->next_mapping, pool->mapping_pool);
1702 mempool_destroy(pool->mapping_pool);
1703 dm_deferred_set_destroy(pool->shared_read_ds);
1704 dm_deferred_set_destroy(pool->all_io_ds);
1708 static struct kmem_cache *_new_mapping_cache;
1710 static struct pool *pool_create(struct mapped_device *pool_md,
1711 struct block_device *metadata_dev,
1712 unsigned long block_size,
1713 int read_only, char **error)
1718 struct dm_pool_metadata *pmd;
1719 bool format_device = read_only ? false : true;
1721 pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
1723 *error = "Error creating metadata object";
1724 return (struct pool *)pmd;
1727 pool = kmalloc(sizeof(*pool), GFP_KERNEL);
1729 *error = "Error allocating memory for pool";
1730 err_p = ERR_PTR(-ENOMEM);
1735 pool->sectors_per_block = block_size;
1736 if (block_size & (block_size - 1))
1737 pool->sectors_per_block_shift = -1;
1739 pool->sectors_per_block_shift = __ffs(block_size);
1740 pool->low_water_blocks = 0;
1741 pool_features_init(&pool->pf);
1742 pool->prison = dm_bio_prison_create(PRISON_CELLS);
1743 if (!pool->prison) {
1744 *error = "Error creating pool's bio prison";
1745 err_p = ERR_PTR(-ENOMEM);
1749 pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
1750 if (IS_ERR(pool->copier)) {
1751 r = PTR_ERR(pool->copier);
1752 *error = "Error creating pool's kcopyd client";
1754 goto bad_kcopyd_client;
1758 * Create singlethreaded workqueue that will service all devices
1759 * that use this metadata.
1761 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
1763 *error = "Error creating pool's workqueue";
1764 err_p = ERR_PTR(-ENOMEM);
1768 INIT_WORK(&pool->worker, do_worker);
1769 INIT_DELAYED_WORK(&pool->waker, do_waker);
1770 spin_lock_init(&pool->lock);
1771 bio_list_init(&pool->deferred_bios);
1772 bio_list_init(&pool->deferred_flush_bios);
1773 INIT_LIST_HEAD(&pool->prepared_mappings);
1774 INIT_LIST_HEAD(&pool->prepared_discards);
1775 pool->low_water_triggered = 0;
1776 pool->no_free_space = 0;
1777 bio_list_init(&pool->retry_on_resume_list);
1779 pool->shared_read_ds = dm_deferred_set_create();
1780 if (!pool->shared_read_ds) {
1781 *error = "Error creating pool's shared read deferred set";
1782 err_p = ERR_PTR(-ENOMEM);
1783 goto bad_shared_read_ds;
1786 pool->all_io_ds = dm_deferred_set_create();
1787 if (!pool->all_io_ds) {
1788 *error = "Error creating pool's all io deferred set";
1789 err_p = ERR_PTR(-ENOMEM);
1793 pool->next_mapping = NULL;
1794 pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE,
1795 _new_mapping_cache);
1796 if (!pool->mapping_pool) {
1797 *error = "Error creating pool's mapping mempool";
1798 err_p = ERR_PTR(-ENOMEM);
1799 goto bad_mapping_pool;
1802 pool->ref_count = 1;
1803 pool->last_commit_jiffies = jiffies;
1804 pool->pool_md = pool_md;
1805 pool->md_dev = metadata_dev;
1806 __pool_table_insert(pool);
1811 dm_deferred_set_destroy(pool->all_io_ds);
1813 dm_deferred_set_destroy(pool->shared_read_ds);
1815 destroy_workqueue(pool->wq);
1817 dm_kcopyd_client_destroy(pool->copier);
1819 dm_bio_prison_destroy(pool->prison);
1823 if (dm_pool_metadata_close(pmd))
1824 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1829 static void __pool_inc(struct pool *pool)
1831 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1835 static void __pool_dec(struct pool *pool)
1837 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1838 BUG_ON(!pool->ref_count);
1839 if (!--pool->ref_count)
1840 __pool_destroy(pool);
1843 static struct pool *__pool_find(struct mapped_device *pool_md,
1844 struct block_device *metadata_dev,
1845 unsigned long block_size, int read_only,
1846 char **error, int *created)
1848 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
1851 if (pool->pool_md != pool_md) {
1852 *error = "metadata device already in use by a pool";
1853 return ERR_PTR(-EBUSY);
1858 pool = __pool_table_lookup(pool_md);
1860 if (pool->md_dev != metadata_dev) {
1861 *error = "different pool cannot replace a pool";
1862 return ERR_PTR(-EINVAL);
1867 pool = pool_create(pool_md, metadata_dev, block_size, read_only, error);
1875 /*----------------------------------------------------------------
1876 * Pool target methods
1877 *--------------------------------------------------------------*/
1878 static void pool_dtr(struct dm_target *ti)
1880 struct pool_c *pt = ti->private;
1882 mutex_lock(&dm_thin_pool_table.mutex);
1884 unbind_control_target(pt->pool, ti);
1885 __pool_dec(pt->pool);
1886 dm_put_device(ti, pt->metadata_dev);
1887 dm_put_device(ti, pt->data_dev);
1890 mutex_unlock(&dm_thin_pool_table.mutex);
1893 static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
1894 struct dm_target *ti)
1898 const char *arg_name;
1900 static struct dm_arg _args[] = {
1901 {0, 3, "Invalid number of pool feature arguments"},
1905 * No feature arguments supplied.
1910 r = dm_read_arg_group(_args, as, &argc, &ti->error);
1914 while (argc && !r) {
1915 arg_name = dm_shift_arg(as);
1918 if (!strcasecmp(arg_name, "skip_block_zeroing"))
1919 pf->zero_new_blocks = false;
1921 else if (!strcasecmp(arg_name, "ignore_discard"))
1922 pf->discard_enabled = false;
1924 else if (!strcasecmp(arg_name, "no_discard_passdown"))
1925 pf->discard_passdown = false;
1927 else if (!strcasecmp(arg_name, "read_only"))
1928 pf->mode = PM_READ_ONLY;
1931 ti->error = "Unrecognised pool feature requested";
1940 static void metadata_low_callback(void *context)
1942 struct pool *pool = context;
1944 DMWARN("%s: reached low water mark for metadata device: sending event.",
1945 dm_device_name(pool->pool_md));
1947 dm_table_event(pool->ti->table);
1950 static sector_t get_metadata_dev_size(struct block_device *bdev)
1952 sector_t metadata_dev_size = i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
1953 char buffer[BDEVNAME_SIZE];
1955 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) {
1956 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
1957 bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS);
1958 metadata_dev_size = THIN_METADATA_MAX_SECTORS_WARNING;
1961 return metadata_dev_size;
1964 static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
1966 sector_t metadata_dev_size = get_metadata_dev_size(bdev);
1968 sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
1970 return metadata_dev_size;
1974 * When a metadata threshold is crossed a dm event is triggered, and
1975 * userland should respond by growing the metadata device. We could let
1976 * userland set the threshold, like we do with the data threshold, but I'm
1977 * not sure they know enough to do this well.
1979 static dm_block_t calc_metadata_threshold(struct pool_c *pt)
1982 * 4M is ample for all ops with the possible exception of thin
1983 * device deletion which is harmless if it fails (just retry the
1984 * delete after you've grown the device).
1986 dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
1987 return min((dm_block_t)1024ULL /* 4M */, quarter);
1991 * thin-pool <metadata dev> <data dev>
1992 * <data block size (sectors)>
1993 * <low water mark (blocks)>
1994 * [<#feature args> [<arg>]*]
1996 * Optional feature arguments are:
1997 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
1998 * ignore_discard: disable discard
1999 * no_discard_passdown: don't pass discards down to the data device
2001 static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
2003 int r, pool_created = 0;
2006 struct pool_features pf;
2007 struct dm_arg_set as;
2008 struct dm_dev *data_dev;
2009 unsigned long block_size;
2010 dm_block_t low_water_blocks;
2011 struct dm_dev *metadata_dev;
2012 fmode_t metadata_mode;
2015 * FIXME Remove validation from scope of lock.
2017 mutex_lock(&dm_thin_pool_table.mutex);
2020 ti->error = "Invalid argument count";
2029 * Set default pool features.
2031 pool_features_init(&pf);
2033 dm_consume_args(&as, 4);
2034 r = parse_pool_features(&as, &pf, ti);
2038 metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
2039 r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
2041 ti->error = "Error opening metadata block device";
2046 * Run for the side-effect of possibly issuing a warning if the
2047 * device is too big.
2049 (void) get_metadata_dev_size(metadata_dev->bdev);
2051 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
2053 ti->error = "Error getting data device";
2057 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
2058 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
2059 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
2060 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
2061 ti->error = "Invalid block size";
2066 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
2067 ti->error = "Invalid low water mark";
2072 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
2078 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
2079 block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
2086 * 'pool_created' reflects whether this is the first table load.
2087 * Top level discard support is not allowed to be changed after
2088 * initial load. This would require a pool reload to trigger thin
2091 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
2092 ti->error = "Discard support cannot be disabled once enabled";
2094 goto out_flags_changed;
2099 pt->metadata_dev = metadata_dev;
2100 pt->data_dev = data_dev;
2101 pt->low_water_blocks = low_water_blocks;
2102 pt->adjusted_pf = pt->requested_pf = pf;
2103 ti->num_flush_bios = 1;
2106 * Only need to enable discards if the pool should pass
2107 * them down to the data device. The thin device's discard
2108 * processing will cause mappings to be removed from the btree.
2110 ti->discard_zeroes_data_unsupported = true;
2111 if (pf.discard_enabled && pf.discard_passdown) {
2112 ti->num_discard_bios = 1;
2115 * Setting 'discards_supported' circumvents the normal
2116 * stacking of discard limits (this keeps the pool and
2117 * thin devices' discard limits consistent).
2119 ti->discards_supported = true;
2123 r = dm_pool_register_metadata_threshold(pt->pool->pmd,
2124 calc_metadata_threshold(pt),
2125 metadata_low_callback,
2130 pt->callbacks.congested_fn = pool_is_congested;
2131 dm_table_add_target_callbacks(ti->table, &pt->callbacks);
2133 mutex_unlock(&dm_thin_pool_table.mutex);
2142 dm_put_device(ti, data_dev);
2144 dm_put_device(ti, metadata_dev);
2146 mutex_unlock(&dm_thin_pool_table.mutex);
2151 static int pool_map(struct dm_target *ti, struct bio *bio)
2154 struct pool_c *pt = ti->private;
2155 struct pool *pool = pt->pool;
2156 unsigned long flags;
2159 * As this is a singleton target, ti->begin is always zero.
2161 spin_lock_irqsave(&pool->lock, flags);
2162 bio->bi_bdev = pt->data_dev->bdev;
2163 r = DM_MAPIO_REMAPPED;
2164 spin_unlock_irqrestore(&pool->lock, flags);
2169 static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
2172 struct pool_c *pt = ti->private;
2173 struct pool *pool = pt->pool;
2174 sector_t data_size = ti->len;
2175 dm_block_t sb_data_size;
2177 *need_commit = false;
2179 (void) sector_div(data_size, pool->sectors_per_block);
2181 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
2183 DMERR("%s: failed to retrieve data device size",
2184 dm_device_name(pool->pool_md));
2188 if (data_size < sb_data_size) {
2189 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
2190 dm_device_name(pool->pool_md),
2191 (unsigned long long)data_size, sb_data_size);
2194 } else if (data_size > sb_data_size) {
2195 r = dm_pool_resize_data_dev(pool->pmd, data_size);
2197 DMERR("%s: failed to resize data device",
2198 dm_device_name(pool->pool_md));
2199 set_pool_mode(pool, PM_READ_ONLY);
2203 *need_commit = true;
2209 static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
2212 struct pool_c *pt = ti->private;
2213 struct pool *pool = pt->pool;
2214 dm_block_t metadata_dev_size, sb_metadata_dev_size;
2216 *need_commit = false;
2218 metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
2220 r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
2222 DMERR("%s: failed to retrieve metadata device size",
2223 dm_device_name(pool->pool_md));
2227 if (metadata_dev_size < sb_metadata_dev_size) {
2228 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
2229 dm_device_name(pool->pool_md),
2230 metadata_dev_size, sb_metadata_dev_size);
2233 } else if (metadata_dev_size > sb_metadata_dev_size) {
2234 r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
2236 DMERR("%s: failed to resize metadata device",
2237 dm_device_name(pool->pool_md));
2241 *need_commit = true;
2248 * Retrieves the number of blocks of the data device from
2249 * the superblock and compares it to the actual device size,
2250 * thus resizing the data device in case it has grown.
2252 * This both copes with opening preallocated data devices in the ctr
2253 * being followed by a resume
2255 * calling the resume method individually after userspace has
2256 * grown the data device in reaction to a table event.
2258 static int pool_preresume(struct dm_target *ti)
2261 bool need_commit1, need_commit2;
2262 struct pool_c *pt = ti->private;
2263 struct pool *pool = pt->pool;
2266 * Take control of the pool object.
2268 r = bind_control_target(pool, ti);
2272 r = maybe_resize_data_dev(ti, &need_commit1);
2276 r = maybe_resize_metadata_dev(ti, &need_commit2);
2280 if (need_commit1 || need_commit2)
2281 (void) commit(pool);
2286 static void pool_resume(struct dm_target *ti)
2288 struct pool_c *pt = ti->private;
2289 struct pool *pool = pt->pool;
2290 unsigned long flags;
2292 spin_lock_irqsave(&pool->lock, flags);
2293 pool->low_water_triggered = 0;
2294 pool->no_free_space = 0;
2295 __requeue_bios(pool);
2296 spin_unlock_irqrestore(&pool->lock, flags);
2298 do_waker(&pool->waker.work);
2301 static void pool_postsuspend(struct dm_target *ti)
2303 struct pool_c *pt = ti->private;
2304 struct pool *pool = pt->pool;
2306 cancel_delayed_work(&pool->waker);
2307 flush_workqueue(pool->wq);
2308 (void) commit(pool);
2311 static int check_arg_count(unsigned argc, unsigned args_required)
2313 if (argc != args_required) {
2314 DMWARN("Message received with %u arguments instead of %u.",
2315 argc, args_required);
2322 static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
2324 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
2325 *dev_id <= MAX_DEV_ID)
2329 DMWARN("Message received with invalid device id: %s", arg);
2334 static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
2339 r = check_arg_count(argc, 2);
2343 r = read_dev_id(argv[1], &dev_id, 1);
2347 r = dm_pool_create_thin(pool->pmd, dev_id);
2349 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2357 static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2360 dm_thin_id origin_dev_id;
2363 r = check_arg_count(argc, 3);
2367 r = read_dev_id(argv[1], &dev_id, 1);
2371 r = read_dev_id(argv[2], &origin_dev_id, 1);
2375 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
2377 DMWARN("Creation of new snapshot %s of device %s failed.",
2385 static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
2390 r = check_arg_count(argc, 2);
2394 r = read_dev_id(argv[1], &dev_id, 1);
2398 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
2400 DMWARN("Deletion of thin device %s failed.", argv[1]);
2405 static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
2407 dm_thin_id old_id, new_id;
2410 r = check_arg_count(argc, 3);
2414 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
2415 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
2419 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
2420 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
2424 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
2426 DMWARN("Failed to change transaction id from %s to %s.",
2434 static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2438 r = check_arg_count(argc, 1);
2442 (void) commit(pool);
2444 r = dm_pool_reserve_metadata_snap(pool->pmd);
2446 DMWARN("reserve_metadata_snap message failed.");
2451 static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2455 r = check_arg_count(argc, 1);
2459 r = dm_pool_release_metadata_snap(pool->pmd);
2461 DMWARN("release_metadata_snap message failed.");
2467 * Messages supported:
2468 * create_thin <dev_id>
2469 * create_snap <dev_id> <origin_id>
2471 * trim <dev_id> <new_size_in_sectors>
2472 * set_transaction_id <current_trans_id> <new_trans_id>
2473 * reserve_metadata_snap
2474 * release_metadata_snap
2476 static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
2479 struct pool_c *pt = ti->private;
2480 struct pool *pool = pt->pool;
2482 if (!strcasecmp(argv[0], "create_thin"))
2483 r = process_create_thin_mesg(argc, argv, pool);
2485 else if (!strcasecmp(argv[0], "create_snap"))
2486 r = process_create_snap_mesg(argc, argv, pool);
2488 else if (!strcasecmp(argv[0], "delete"))
2489 r = process_delete_mesg(argc, argv, pool);
2491 else if (!strcasecmp(argv[0], "set_transaction_id"))
2492 r = process_set_transaction_id_mesg(argc, argv, pool);
2494 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
2495 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
2497 else if (!strcasecmp(argv[0], "release_metadata_snap"))
2498 r = process_release_metadata_snap_mesg(argc, argv, pool);
2501 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
2504 (void) commit(pool);
2509 static void emit_flags(struct pool_features *pf, char *result,
2510 unsigned sz, unsigned maxlen)
2512 unsigned count = !pf->zero_new_blocks + !pf->discard_enabled +
2513 !pf->discard_passdown + (pf->mode == PM_READ_ONLY);
2514 DMEMIT("%u ", count);
2516 if (!pf->zero_new_blocks)
2517 DMEMIT("skip_block_zeroing ");
2519 if (!pf->discard_enabled)
2520 DMEMIT("ignore_discard ");
2522 if (!pf->discard_passdown)
2523 DMEMIT("no_discard_passdown ");
2525 if (pf->mode == PM_READ_ONLY)
2526 DMEMIT("read_only ");
2531 * <transaction id> <used metadata sectors>/<total metadata sectors>
2532 * <used data sectors>/<total data sectors> <held metadata root>
2534 static void pool_status(struct dm_target *ti, status_type_t type,
2535 unsigned status_flags, char *result, unsigned maxlen)
2539 uint64_t transaction_id;
2540 dm_block_t nr_free_blocks_data;
2541 dm_block_t nr_free_blocks_metadata;
2542 dm_block_t nr_blocks_data;
2543 dm_block_t nr_blocks_metadata;
2544 dm_block_t held_root;
2545 char buf[BDEVNAME_SIZE];
2546 char buf2[BDEVNAME_SIZE];
2547 struct pool_c *pt = ti->private;
2548 struct pool *pool = pt->pool;
2551 case STATUSTYPE_INFO:
2552 if (get_pool_mode(pool) == PM_FAIL) {
2557 /* Commit to ensure statistics aren't out-of-date */
2558 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
2559 (void) commit(pool);
2561 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
2563 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
2564 dm_device_name(pool->pool_md), r);
2568 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
2570 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
2571 dm_device_name(pool->pool_md), r);
2575 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
2577 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
2578 dm_device_name(pool->pool_md), r);
2582 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
2584 DMERR("%s: dm_pool_get_free_block_count returned %d",
2585 dm_device_name(pool->pool_md), r);
2589 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
2591 DMERR("%s: dm_pool_get_data_dev_size returned %d",
2592 dm_device_name(pool->pool_md), r);
2596 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
2598 DMERR("%s: dm_pool_get_metadata_snap returned %d",
2599 dm_device_name(pool->pool_md), r);
2603 DMEMIT("%llu %llu/%llu %llu/%llu ",
2604 (unsigned long long)transaction_id,
2605 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
2606 (unsigned long long)nr_blocks_metadata,
2607 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
2608 (unsigned long long)nr_blocks_data);
2611 DMEMIT("%llu ", held_root);
2615 if (pool->pf.mode == PM_READ_ONLY)
2620 if (!pool->pf.discard_enabled)
2621 DMEMIT("ignore_discard");
2622 else if (pool->pf.discard_passdown)
2623 DMEMIT("discard_passdown");
2625 DMEMIT("no_discard_passdown");
2629 case STATUSTYPE_TABLE:
2630 DMEMIT("%s %s %lu %llu ",
2631 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
2632 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
2633 (unsigned long)pool->sectors_per_block,
2634 (unsigned long long)pt->low_water_blocks);
2635 emit_flags(&pt->requested_pf, result, sz, maxlen);
2644 static int pool_iterate_devices(struct dm_target *ti,
2645 iterate_devices_callout_fn fn, void *data)
2647 struct pool_c *pt = ti->private;
2649 return fn(ti, pt->data_dev, 0, ti->len, data);
2652 static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2653 struct bio_vec *biovec, int max_size)
2655 struct pool_c *pt = ti->private;
2656 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
2658 if (!q->merge_bvec_fn)
2661 bvm->bi_bdev = pt->data_dev->bdev;
2663 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2666 static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits)
2668 struct pool *pool = pt->pool;
2669 struct queue_limits *data_limits;
2671 limits->max_discard_sectors = pool->sectors_per_block;
2674 * discard_granularity is just a hint, and not enforced.
2676 if (pt->adjusted_pf.discard_passdown) {
2677 data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits;
2678 limits->discard_granularity = data_limits->discard_granularity;
2680 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
2683 static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
2685 struct pool_c *pt = ti->private;
2686 struct pool *pool = pt->pool;
2687 uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
2690 * If the system-determined stacked limits are compatible with the
2691 * pool's blocksize (io_opt is a factor) do not override them.
2693 if (io_opt_sectors < pool->sectors_per_block ||
2694 do_div(io_opt_sectors, pool->sectors_per_block)) {
2695 blk_limits_io_min(limits, 0);
2696 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
2700 * pt->adjusted_pf is a staging area for the actual features to use.
2701 * They get transferred to the live pool in bind_control_target()
2702 * called from pool_preresume().
2704 if (!pt->adjusted_pf.discard_enabled) {
2706 * Must explicitly disallow stacking discard limits otherwise the
2707 * block layer will stack them if pool's data device has support.
2708 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
2709 * user to see that, so make sure to set all discard limits to 0.
2711 limits->discard_granularity = 0;
2715 disable_passdown_if_not_supported(pt);
2717 set_discard_limits(pt, limits);
2720 static struct target_type pool_target = {
2721 .name = "thin-pool",
2722 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
2723 DM_TARGET_IMMUTABLE,
2724 .version = {1, 9, 0},
2725 .module = THIS_MODULE,
2729 .postsuspend = pool_postsuspend,
2730 .preresume = pool_preresume,
2731 .resume = pool_resume,
2732 .message = pool_message,
2733 .status = pool_status,
2734 .merge = pool_merge,
2735 .iterate_devices = pool_iterate_devices,
2736 .io_hints = pool_io_hints,
2739 /*----------------------------------------------------------------
2740 * Thin target methods
2741 *--------------------------------------------------------------*/
2742 static void thin_dtr(struct dm_target *ti)
2744 struct thin_c *tc = ti->private;
2746 mutex_lock(&dm_thin_pool_table.mutex);
2748 __pool_dec(tc->pool);
2749 dm_pool_close_thin_device(tc->td);
2750 dm_put_device(ti, tc->pool_dev);
2752 dm_put_device(ti, tc->origin_dev);
2755 mutex_unlock(&dm_thin_pool_table.mutex);
2759 * Thin target parameters:
2761 * <pool_dev> <dev_id> [origin_dev]
2763 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2764 * dev_id: the internal device identifier
2765 * origin_dev: a device external to the pool that should act as the origin
2767 * If the pool device has discards disabled, they get disabled for the thin
2770 static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
2774 struct dm_dev *pool_dev, *origin_dev;
2775 struct mapped_device *pool_md;
2777 mutex_lock(&dm_thin_pool_table.mutex);
2779 if (argc != 2 && argc != 3) {
2780 ti->error = "Invalid argument count";
2785 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
2787 ti->error = "Out of memory";
2793 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
2795 ti->error = "Error opening origin device";
2796 goto bad_origin_dev;
2798 tc->origin_dev = origin_dev;
2801 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
2803 ti->error = "Error opening pool device";
2806 tc->pool_dev = pool_dev;
2808 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
2809 ti->error = "Invalid device id";
2814 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
2816 ti->error = "Couldn't get pool mapped device";
2821 tc->pool = __pool_table_lookup(pool_md);
2823 ti->error = "Couldn't find pool object";
2825 goto bad_pool_lookup;
2827 __pool_inc(tc->pool);
2829 if (get_pool_mode(tc->pool) == PM_FAIL) {
2830 ti->error = "Couldn't open thin device, Pool is in fail mode";
2834 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
2836 ti->error = "Couldn't open thin internal device";
2840 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
2844 ti->num_flush_bios = 1;
2845 ti->flush_supported = true;
2846 ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook);
2848 /* In case the pool supports discards, pass them on. */
2849 ti->discard_zeroes_data_unsupported = true;
2850 if (tc->pool->pf.discard_enabled) {
2851 ti->discards_supported = true;
2852 ti->num_discard_bios = 1;
2853 /* Discard bios must be split on a block boundary */
2854 ti->split_discard_bios = true;
2859 mutex_unlock(&dm_thin_pool_table.mutex);
2864 __pool_dec(tc->pool);
2868 dm_put_device(ti, tc->pool_dev);
2871 dm_put_device(ti, tc->origin_dev);
2875 mutex_unlock(&dm_thin_pool_table.mutex);
2880 static int thin_map(struct dm_target *ti, struct bio *bio)
2882 bio->bi_sector = dm_target_offset(ti, bio->bi_sector);
2884 return thin_bio_map(ti, bio);
2887 static int thin_endio(struct dm_target *ti, struct bio *bio, int err)
2889 unsigned long flags;
2890 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2891 struct list_head work;
2892 struct dm_thin_new_mapping *m, *tmp;
2893 struct pool *pool = h->tc->pool;
2895 if (h->shared_read_entry) {
2896 INIT_LIST_HEAD(&work);
2897 dm_deferred_entry_dec(h->shared_read_entry, &work);
2899 spin_lock_irqsave(&pool->lock, flags);
2900 list_for_each_entry_safe(m, tmp, &work, list) {
2903 __maybe_add_mapping(m);
2905 spin_unlock_irqrestore(&pool->lock, flags);
2908 if (h->all_io_entry) {
2909 INIT_LIST_HEAD(&work);
2910 dm_deferred_entry_dec(h->all_io_entry, &work);
2911 if (!list_empty(&work)) {
2912 spin_lock_irqsave(&pool->lock, flags);
2913 list_for_each_entry_safe(m, tmp, &work, list)
2914 list_add(&m->list, &pool->prepared_discards);
2915 spin_unlock_irqrestore(&pool->lock, flags);
2923 static void thin_postsuspend(struct dm_target *ti)
2925 if (dm_noflush_suspending(ti))
2926 requeue_io((struct thin_c *)ti->private);
2930 * <nr mapped sectors> <highest mapped sector>
2932 static void thin_status(struct dm_target *ti, status_type_t type,
2933 unsigned status_flags, char *result, unsigned maxlen)
2937 dm_block_t mapped, highest;
2938 char buf[BDEVNAME_SIZE];
2939 struct thin_c *tc = ti->private;
2941 if (get_pool_mode(tc->pool) == PM_FAIL) {
2950 case STATUSTYPE_INFO:
2951 r = dm_thin_get_mapped_count(tc->td, &mapped);
2953 DMERR("dm_thin_get_mapped_count returned %d", r);
2957 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
2959 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
2963 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
2965 DMEMIT("%llu", ((highest + 1) *
2966 tc->pool->sectors_per_block) - 1);
2971 case STATUSTYPE_TABLE:
2973 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
2974 (unsigned long) tc->dev_id);
2976 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
2987 static int thin_iterate_devices(struct dm_target *ti,
2988 iterate_devices_callout_fn fn, void *data)
2991 struct thin_c *tc = ti->private;
2992 struct pool *pool = tc->pool;
2995 * We can't call dm_pool_get_data_dev_size() since that blocks. So
2996 * we follow a more convoluted path through to the pool's target.
2999 return 0; /* nothing is bound */
3001 blocks = pool->ti->len;
3002 (void) sector_div(blocks, pool->sectors_per_block);
3004 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
3009 static struct target_type thin_target = {
3011 .version = {1, 9, 0},
3012 .module = THIS_MODULE,
3016 .end_io = thin_endio,
3017 .postsuspend = thin_postsuspend,
3018 .status = thin_status,
3019 .iterate_devices = thin_iterate_devices,
3022 /*----------------------------------------------------------------*/
3024 static int __init dm_thin_init(void)
3030 r = dm_register_target(&thin_target);
3034 r = dm_register_target(&pool_target);
3036 goto bad_pool_target;
3040 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
3041 if (!_new_mapping_cache)
3042 goto bad_new_mapping_cache;
3046 bad_new_mapping_cache:
3047 dm_unregister_target(&pool_target);
3049 dm_unregister_target(&thin_target);
3054 static void dm_thin_exit(void)
3056 dm_unregister_target(&thin_target);
3057 dm_unregister_target(&pool_target);
3059 kmem_cache_destroy(_new_mapping_cache);
3062 module_init(dm_thin_init);
3063 module_exit(dm_thin_exit);
3065 MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
3066 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3067 MODULE_LICENSE("GPL");