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;
515 unsigned definitely_not_shared:1;
518 dm_block_t virt_block;
519 dm_block_t data_block;
520 struct dm_bio_prison_cell *cell, *cell2;
524 * If the bio covers the whole area of a block then we can avoid
525 * zeroing or copying. Instead this bio is hooked. The bio will
526 * still be in the cell, so care has to be taken to avoid issuing
530 bio_end_io_t *saved_bi_end_io;
533 static void __maybe_add_mapping(struct dm_thin_new_mapping *m)
535 struct pool *pool = m->tc->pool;
537 if (m->quiesced && m->prepared) {
538 list_add(&m->list, &pool->prepared_mappings);
543 static void copy_complete(int read_err, unsigned long write_err, void *context)
546 struct dm_thin_new_mapping *m = context;
547 struct pool *pool = m->tc->pool;
549 m->err = read_err || write_err ? -EIO : 0;
551 spin_lock_irqsave(&pool->lock, flags);
553 __maybe_add_mapping(m);
554 spin_unlock_irqrestore(&pool->lock, flags);
557 static void overwrite_endio(struct bio *bio, int err)
560 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
561 struct dm_thin_new_mapping *m = h->overwrite_mapping;
562 struct pool *pool = m->tc->pool;
566 spin_lock_irqsave(&pool->lock, flags);
568 __maybe_add_mapping(m);
569 spin_unlock_irqrestore(&pool->lock, flags);
572 /*----------------------------------------------------------------*/
579 * Prepared mapping jobs.
583 * This sends the bios in the cell back to the deferred_bios list.
585 static void cell_defer(struct thin_c *tc, struct dm_bio_prison_cell *cell)
587 struct pool *pool = tc->pool;
590 spin_lock_irqsave(&pool->lock, flags);
591 cell_release(pool, cell, &pool->deferred_bios);
592 spin_unlock_irqrestore(&tc->pool->lock, flags);
598 * Same as cell_defer above, except it omits the original holder of the cell.
600 static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
602 struct pool *pool = tc->pool;
605 spin_lock_irqsave(&pool->lock, flags);
606 cell_release_no_holder(pool, cell, &pool->deferred_bios);
607 spin_unlock_irqrestore(&pool->lock, flags);
612 static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
615 m->bio->bi_end_io = m->saved_bi_end_io;
616 cell_error(m->tc->pool, m->cell);
618 mempool_free(m, m->tc->pool->mapping_pool);
621 static void process_prepared_mapping(struct dm_thin_new_mapping *m)
623 struct thin_c *tc = m->tc;
624 struct pool *pool = tc->pool;
630 bio->bi_end_io = m->saved_bi_end_io;
633 cell_error(pool, m->cell);
638 * Commit the prepared block into the mapping btree.
639 * Any I/O for this block arriving after this point will get
640 * remapped to it directly.
642 r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block);
644 DMERR_LIMIT("%s: dm_thin_insert_block() failed: error = %d",
645 dm_device_name(pool->pool_md), r);
646 set_pool_mode(pool, PM_READ_ONLY);
647 cell_error(pool, m->cell);
652 * Release any bios held while the block was being provisioned.
653 * If we are processing a write bio that completely covers the block,
654 * we already processed it so can ignore it now when processing
655 * the bios in the cell.
658 cell_defer_no_holder(tc, m->cell);
661 cell_defer(tc, m->cell);
665 mempool_free(m, pool->mapping_pool);
668 static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
670 struct thin_c *tc = m->tc;
672 bio_io_error(m->bio);
673 cell_defer_no_holder(tc, m->cell);
674 cell_defer_no_holder(tc, m->cell2);
675 mempool_free(m, tc->pool->mapping_pool);
678 static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m)
680 struct thin_c *tc = m->tc;
682 inc_all_io_entry(tc->pool, m->bio);
683 cell_defer_no_holder(tc, m->cell);
684 cell_defer_no_holder(tc, m->cell2);
687 if (m->definitely_not_shared)
688 remap_and_issue(tc, m->bio, m->data_block);
691 if (dm_pool_block_is_used(tc->pool->pmd, m->data_block, &used) || used)
692 bio_endio(m->bio, 0);
694 remap_and_issue(tc, m->bio, m->data_block);
697 bio_endio(m->bio, 0);
699 mempool_free(m, tc->pool->mapping_pool);
702 static void process_prepared_discard(struct dm_thin_new_mapping *m)
705 struct thin_c *tc = m->tc;
707 r = dm_thin_remove_block(tc->td, m->virt_block);
709 DMERR_LIMIT("dm_thin_remove_block() failed");
711 process_prepared_discard_passdown(m);
714 static void process_prepared(struct pool *pool, struct list_head *head,
715 process_mapping_fn *fn)
718 struct list_head maps;
719 struct dm_thin_new_mapping *m, *tmp;
721 INIT_LIST_HEAD(&maps);
722 spin_lock_irqsave(&pool->lock, flags);
723 list_splice_init(head, &maps);
724 spin_unlock_irqrestore(&pool->lock, flags);
726 list_for_each_entry_safe(m, tmp, &maps, list)
733 static int io_overlaps_block(struct pool *pool, struct bio *bio)
735 return bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT);
738 static int io_overwrites_block(struct pool *pool, struct bio *bio)
740 return (bio_data_dir(bio) == WRITE) &&
741 io_overlaps_block(pool, bio);
744 static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
747 *save = bio->bi_end_io;
751 static int ensure_next_mapping(struct pool *pool)
753 if (pool->next_mapping)
756 pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);
758 return pool->next_mapping ? 0 : -ENOMEM;
761 static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
763 struct dm_thin_new_mapping *m = pool->next_mapping;
765 BUG_ON(!pool->next_mapping);
767 memset(m, 0, sizeof(struct dm_thin_new_mapping));
768 INIT_LIST_HEAD(&m->list);
771 pool->next_mapping = NULL;
776 static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
777 struct dm_dev *origin, dm_block_t data_origin,
778 dm_block_t data_dest,
779 struct dm_bio_prison_cell *cell, struct bio *bio)
782 struct pool *pool = tc->pool;
783 struct dm_thin_new_mapping *m = get_next_mapping(pool);
786 m->virt_block = virt_block;
787 m->data_block = data_dest;
790 if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
794 * IO to pool_dev remaps to the pool target's data_dev.
796 * If the whole block of data is being overwritten, we can issue the
797 * bio immediately. Otherwise we use kcopyd to clone the data first.
799 if (io_overwrites_block(pool, bio)) {
800 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
802 h->overwrite_mapping = m;
804 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
805 inc_all_io_entry(pool, bio);
806 remap_and_issue(tc, bio, data_dest);
808 struct dm_io_region from, to;
810 from.bdev = origin->bdev;
811 from.sector = data_origin * pool->sectors_per_block;
812 from.count = pool->sectors_per_block;
814 to.bdev = tc->pool_dev->bdev;
815 to.sector = data_dest * pool->sectors_per_block;
816 to.count = pool->sectors_per_block;
818 r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
819 0, copy_complete, m);
821 mempool_free(m, pool->mapping_pool);
822 DMERR_LIMIT("dm_kcopyd_copy() failed");
823 cell_error(pool, cell);
828 static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
829 dm_block_t data_origin, dm_block_t data_dest,
830 struct dm_bio_prison_cell *cell, struct bio *bio)
832 schedule_copy(tc, virt_block, tc->pool_dev,
833 data_origin, data_dest, cell, bio);
836 static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
837 dm_block_t data_dest,
838 struct dm_bio_prison_cell *cell, struct bio *bio)
840 schedule_copy(tc, virt_block, tc->origin_dev,
841 virt_block, data_dest, cell, bio);
844 static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
845 dm_block_t data_block, struct dm_bio_prison_cell *cell,
848 struct pool *pool = tc->pool;
849 struct dm_thin_new_mapping *m = get_next_mapping(pool);
854 m->virt_block = virt_block;
855 m->data_block = data_block;
859 * If the whole block of data is being overwritten or we are not
860 * zeroing pre-existing data, we can issue the bio immediately.
861 * Otherwise we use kcopyd to zero the data first.
863 if (!pool->pf.zero_new_blocks)
864 process_prepared_mapping(m);
866 else if (io_overwrites_block(pool, bio)) {
867 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
869 h->overwrite_mapping = m;
871 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
872 inc_all_io_entry(pool, bio);
873 remap_and_issue(tc, bio, data_block);
876 struct dm_io_region to;
878 to.bdev = tc->pool_dev->bdev;
879 to.sector = data_block * pool->sectors_per_block;
880 to.count = pool->sectors_per_block;
882 r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m);
884 mempool_free(m, pool->mapping_pool);
885 DMERR_LIMIT("dm_kcopyd_zero() failed");
886 cell_error(pool, cell);
891 static int commit(struct pool *pool)
895 r = dm_pool_commit_metadata(pool->pmd);
897 DMERR_LIMIT("commit failed: error = %d", r);
903 * A non-zero return indicates read_only or fail_io mode.
904 * Many callers don't care about the return value.
906 static int commit_or_fallback(struct pool *pool)
910 if (get_pool_mode(pool) != PM_WRITE)
915 set_pool_mode(pool, PM_READ_ONLY);
920 static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
923 dm_block_t free_blocks;
925 struct pool *pool = tc->pool;
927 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
931 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
932 DMWARN("%s: reached low water mark for data device: sending event.",
933 dm_device_name(pool->pool_md));
934 spin_lock_irqsave(&pool->lock, flags);
935 pool->low_water_triggered = 1;
936 spin_unlock_irqrestore(&pool->lock, flags);
937 dm_table_event(pool->ti->table);
941 if (pool->no_free_space)
945 * Try to commit to see if that will free up some
948 (void) commit_or_fallback(pool);
950 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
955 * If we still have no space we set a flag to avoid
956 * doing all this checking and return -ENOSPC.
959 DMWARN("%s: no free space available.",
960 dm_device_name(pool->pool_md));
961 spin_lock_irqsave(&pool->lock, flags);
962 pool->no_free_space = 1;
963 spin_unlock_irqrestore(&pool->lock, flags);
969 r = dm_pool_alloc_data_block(pool->pmd, result);
977 * If we have run out of space, queue bios until the device is
978 * resumed, presumably after having been reloaded with more space.
980 static void retry_on_resume(struct bio *bio)
982 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
983 struct thin_c *tc = h->tc;
984 struct pool *pool = tc->pool;
987 spin_lock_irqsave(&pool->lock, flags);
988 bio_list_add(&pool->retry_on_resume_list, bio);
989 spin_unlock_irqrestore(&pool->lock, flags);
992 static void no_space(struct pool *pool, struct dm_bio_prison_cell *cell)
995 struct bio_list bios;
997 bio_list_init(&bios);
998 cell_release(pool, cell, &bios);
1000 while ((bio = bio_list_pop(&bios)))
1001 retry_on_resume(bio);
1004 static void process_discard(struct thin_c *tc, struct bio *bio)
1007 unsigned long flags;
1008 struct pool *pool = tc->pool;
1009 struct dm_bio_prison_cell *cell, *cell2;
1010 struct dm_cell_key key, key2;
1011 dm_block_t block = get_bio_block(tc, bio);
1012 struct dm_thin_lookup_result lookup_result;
1013 struct dm_thin_new_mapping *m;
1015 build_virtual_key(tc->td, block, &key);
1016 if (bio_detain(tc->pool, &key, bio, &cell))
1019 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1023 * Check nobody is fiddling with this pool block. This can
1024 * happen if someone's in the process of breaking sharing
1027 build_data_key(tc->td, lookup_result.block, &key2);
1028 if (bio_detain(tc->pool, &key2, bio, &cell2)) {
1029 cell_defer_no_holder(tc, cell);
1033 if (io_overlaps_block(pool, bio)) {
1035 * IO may still be going to the destination block. We must
1036 * quiesce before we can do the removal.
1038 m = get_next_mapping(pool);
1040 m->pass_discard = pool->pf.discard_passdown;
1041 m->definitely_not_shared = !lookup_result.shared;
1042 m->virt_block = block;
1043 m->data_block = lookup_result.block;
1048 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) {
1049 spin_lock_irqsave(&pool->lock, flags);
1050 list_add(&m->list, &pool->prepared_discards);
1051 spin_unlock_irqrestore(&pool->lock, flags);
1055 inc_all_io_entry(pool, bio);
1056 cell_defer_no_holder(tc, cell);
1057 cell_defer_no_holder(tc, cell2);
1060 * The DM core makes sure that the discard doesn't span
1061 * a block boundary. So we submit the discard of a
1062 * partial block appropriately.
1064 if ((!lookup_result.shared) && pool->pf.discard_passdown)
1065 remap_and_issue(tc, bio, lookup_result.block);
1073 * It isn't provisioned, just forget it.
1075 cell_defer_no_holder(tc, cell);
1080 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1082 cell_defer_no_holder(tc, cell);
1088 static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
1089 struct dm_cell_key *key,
1090 struct dm_thin_lookup_result *lookup_result,
1091 struct dm_bio_prison_cell *cell)
1094 dm_block_t data_block;
1096 r = alloc_data_block(tc, &data_block);
1099 schedule_internal_copy(tc, block, lookup_result->block,
1100 data_block, cell, bio);
1104 no_space(tc->pool, cell);
1108 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1110 cell_error(tc->pool, cell);
1115 static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1117 struct dm_thin_lookup_result *lookup_result)
1119 struct dm_bio_prison_cell *cell;
1120 struct pool *pool = tc->pool;
1121 struct dm_cell_key key;
1124 * If cell is already occupied, then sharing is already in the process
1125 * of being broken so we have nothing further to do here.
1127 build_data_key(tc->td, lookup_result->block, &key);
1128 if (bio_detain(pool, &key, bio, &cell))
1131 if (bio_data_dir(bio) == WRITE && bio->bi_size)
1132 break_sharing(tc, bio, block, &key, lookup_result, cell);
1134 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1136 h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
1137 inc_all_io_entry(pool, bio);
1138 cell_defer_no_holder(tc, cell);
1140 remap_and_issue(tc, bio, lookup_result->block);
1144 static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
1145 struct dm_bio_prison_cell *cell)
1148 dm_block_t data_block;
1149 struct pool *pool = tc->pool;
1152 * Remap empty bios (flushes) immediately, without provisioning.
1154 if (!bio->bi_size) {
1155 inc_all_io_entry(pool, bio);
1156 cell_defer_no_holder(tc, cell);
1158 remap_and_issue(tc, bio, 0);
1163 * Fill read bios with zeroes and complete them immediately.
1165 if (bio_data_dir(bio) == READ) {
1167 cell_defer_no_holder(tc, cell);
1172 r = alloc_data_block(tc, &data_block);
1176 schedule_external_copy(tc, block, data_block, cell, bio);
1178 schedule_zero(tc, block, data_block, cell, bio);
1182 no_space(pool, cell);
1186 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1188 set_pool_mode(pool, PM_READ_ONLY);
1189 cell_error(pool, cell);
1194 static void process_bio(struct thin_c *tc, struct bio *bio)
1197 struct pool *pool = tc->pool;
1198 dm_block_t block = get_bio_block(tc, bio);
1199 struct dm_bio_prison_cell *cell;
1200 struct dm_cell_key key;
1201 struct dm_thin_lookup_result lookup_result;
1204 * If cell is already occupied, then the block is already
1205 * being provisioned so we have nothing further to do here.
1207 build_virtual_key(tc->td, block, &key);
1208 if (bio_detain(pool, &key, bio, &cell))
1211 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1214 if (lookup_result.shared) {
1215 process_shared_bio(tc, bio, block, &lookup_result);
1216 cell_defer_no_holder(tc, cell); /* FIXME: pass this cell into process_shared? */
1218 inc_all_io_entry(pool, bio);
1219 cell_defer_no_holder(tc, cell);
1221 remap_and_issue(tc, bio, lookup_result.block);
1226 if (bio_data_dir(bio) == READ && tc->origin_dev) {
1227 inc_all_io_entry(pool, bio);
1228 cell_defer_no_holder(tc, cell);
1230 remap_to_origin_and_issue(tc, bio);
1232 provision_block(tc, bio, block, cell);
1236 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1238 cell_defer_no_holder(tc, cell);
1244 static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
1247 int rw = bio_data_dir(bio);
1248 dm_block_t block = get_bio_block(tc, bio);
1249 struct dm_thin_lookup_result lookup_result;
1251 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1254 if (lookup_result.shared && (rw == WRITE) && bio->bi_size)
1257 inc_all_io_entry(tc->pool, bio);
1258 remap_and_issue(tc, bio, lookup_result.block);
1268 if (tc->origin_dev) {
1269 inc_all_io_entry(tc->pool, bio);
1270 remap_to_origin_and_issue(tc, bio);
1279 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1286 static void process_bio_fail(struct thin_c *tc, struct bio *bio)
1292 * FIXME: should we also commit due to size of transaction, measured in
1295 static int need_commit_due_to_time(struct pool *pool)
1297 return jiffies < pool->last_commit_jiffies ||
1298 jiffies > pool->last_commit_jiffies + COMMIT_PERIOD;
1301 static void process_deferred_bios(struct pool *pool)
1303 unsigned long flags;
1305 struct bio_list bios;
1307 bio_list_init(&bios);
1309 spin_lock_irqsave(&pool->lock, flags);
1310 bio_list_merge(&bios, &pool->deferred_bios);
1311 bio_list_init(&pool->deferred_bios);
1312 spin_unlock_irqrestore(&pool->lock, flags);
1314 while ((bio = bio_list_pop(&bios))) {
1315 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1316 struct thin_c *tc = h->tc;
1319 * If we've got no free new_mapping structs, and processing
1320 * this bio might require one, we pause until there are some
1321 * prepared mappings to process.
1323 if (ensure_next_mapping(pool)) {
1324 spin_lock_irqsave(&pool->lock, flags);
1325 bio_list_merge(&pool->deferred_bios, &bios);
1326 spin_unlock_irqrestore(&pool->lock, flags);
1331 if (bio->bi_rw & REQ_DISCARD)
1332 pool->process_discard(tc, bio);
1334 pool->process_bio(tc, bio);
1338 * If there are any deferred flush bios, we must commit
1339 * the metadata before issuing them.
1341 bio_list_init(&bios);
1342 spin_lock_irqsave(&pool->lock, flags);
1343 bio_list_merge(&bios, &pool->deferred_flush_bios);
1344 bio_list_init(&pool->deferred_flush_bios);
1345 spin_unlock_irqrestore(&pool->lock, flags);
1347 if (bio_list_empty(&bios) &&
1348 !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool)))
1351 if (commit_or_fallback(pool)) {
1352 while ((bio = bio_list_pop(&bios)))
1356 pool->last_commit_jiffies = jiffies;
1358 while ((bio = bio_list_pop(&bios)))
1359 generic_make_request(bio);
1362 static void do_worker(struct work_struct *ws)
1364 struct pool *pool = container_of(ws, struct pool, worker);
1366 process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
1367 process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
1368 process_deferred_bios(pool);
1372 * We want to commit periodically so that not too much
1373 * unwritten data builds up.
1375 static void do_waker(struct work_struct *ws)
1377 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
1379 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
1382 /*----------------------------------------------------------------*/
1384 static enum pool_mode get_pool_mode(struct pool *pool)
1386 return pool->pf.mode;
1389 static void set_pool_mode(struct pool *pool, enum pool_mode mode)
1393 pool->pf.mode = mode;
1397 DMERR("switching pool to failure mode");
1398 pool->process_bio = process_bio_fail;
1399 pool->process_discard = process_bio_fail;
1400 pool->process_prepared_mapping = process_prepared_mapping_fail;
1401 pool->process_prepared_discard = process_prepared_discard_fail;
1405 DMERR("switching pool to read-only mode");
1406 r = dm_pool_abort_metadata(pool->pmd);
1408 DMERR("aborting transaction failed");
1409 set_pool_mode(pool, PM_FAIL);
1411 dm_pool_metadata_read_only(pool->pmd);
1412 pool->process_bio = process_bio_read_only;
1413 pool->process_discard = process_discard;
1414 pool->process_prepared_mapping = process_prepared_mapping_fail;
1415 pool->process_prepared_discard = process_prepared_discard_passdown;
1420 pool->process_bio = process_bio;
1421 pool->process_discard = process_discard;
1422 pool->process_prepared_mapping = process_prepared_mapping;
1423 pool->process_prepared_discard = process_prepared_discard;
1428 /*----------------------------------------------------------------*/
1431 * Mapping functions.
1435 * Called only while mapping a thin bio to hand it over to the workqueue.
1437 static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
1439 unsigned long flags;
1440 struct pool *pool = tc->pool;
1442 spin_lock_irqsave(&pool->lock, flags);
1443 bio_list_add(&pool->deferred_bios, bio);
1444 spin_unlock_irqrestore(&pool->lock, flags);
1449 static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
1451 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1454 h->shared_read_entry = NULL;
1455 h->all_io_entry = NULL;
1456 h->overwrite_mapping = NULL;
1460 * Non-blocking function called from the thin target's map function.
1462 static int thin_bio_map(struct dm_target *ti, struct bio *bio)
1465 struct thin_c *tc = ti->private;
1466 dm_block_t block = get_bio_block(tc, bio);
1467 struct dm_thin_device *td = tc->td;
1468 struct dm_thin_lookup_result result;
1469 struct dm_bio_prison_cell cell1, cell2;
1470 struct dm_bio_prison_cell *cell_result;
1471 struct dm_cell_key key;
1473 thin_hook_bio(tc, bio);
1475 if (get_pool_mode(tc->pool) == PM_FAIL) {
1477 return DM_MAPIO_SUBMITTED;
1480 if (bio->bi_rw & (REQ_DISCARD | REQ_FLUSH | REQ_FUA)) {
1481 thin_defer_bio(tc, bio);
1482 return DM_MAPIO_SUBMITTED;
1485 r = dm_thin_find_block(td, block, 0, &result);
1488 * Note that we defer readahead too.
1492 if (unlikely(result.shared)) {
1494 * We have a race condition here between the
1495 * result.shared value returned by the lookup and
1496 * snapshot creation, which may cause new
1499 * To avoid this always quiesce the origin before
1500 * taking the snap. You want to do this anyway to
1501 * ensure a consistent application view
1504 * More distant ancestors are irrelevant. The
1505 * shared flag will be set in their case.
1507 thin_defer_bio(tc, bio);
1508 return DM_MAPIO_SUBMITTED;
1511 build_virtual_key(tc->td, block, &key);
1512 if (dm_bio_detain(tc->pool->prison, &key, bio, &cell1, &cell_result))
1513 return DM_MAPIO_SUBMITTED;
1515 build_data_key(tc->td, result.block, &key);
1516 if (dm_bio_detain(tc->pool->prison, &key, bio, &cell2, &cell_result)) {
1517 cell_defer_no_holder_no_free(tc, &cell1);
1518 return DM_MAPIO_SUBMITTED;
1521 inc_all_io_entry(tc->pool, bio);
1522 cell_defer_no_holder_no_free(tc, &cell2);
1523 cell_defer_no_holder_no_free(tc, &cell1);
1525 remap(tc, bio, result.block);
1526 return DM_MAPIO_REMAPPED;
1529 if (get_pool_mode(tc->pool) == PM_READ_ONLY) {
1531 * This block isn't provisioned, and we have no way
1532 * of doing so. Just error it.
1535 return DM_MAPIO_SUBMITTED;
1541 * In future, the failed dm_thin_find_block above could
1542 * provide the hint to load the metadata into cache.
1544 thin_defer_bio(tc, bio);
1545 return DM_MAPIO_SUBMITTED;
1549 * Must always call bio_io_error on failure.
1550 * dm_thin_find_block can fail with -EINVAL if the
1551 * pool is switched to fail-io mode.
1554 return DM_MAPIO_SUBMITTED;
1558 static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
1561 unsigned long flags;
1562 struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
1564 spin_lock_irqsave(&pt->pool->lock, flags);
1565 r = !bio_list_empty(&pt->pool->retry_on_resume_list);
1566 spin_unlock_irqrestore(&pt->pool->lock, flags);
1569 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1570 r = bdi_congested(&q->backing_dev_info, bdi_bits);
1576 static void __requeue_bios(struct pool *pool)
1578 bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list);
1579 bio_list_init(&pool->retry_on_resume_list);
1582 /*----------------------------------------------------------------
1583 * Binding of control targets to a pool object
1584 *--------------------------------------------------------------*/
1585 static bool data_dev_supports_discard(struct pool_c *pt)
1587 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
1589 return q && blk_queue_discard(q);
1592 static bool is_factor(sector_t block_size, uint32_t n)
1594 return !sector_div(block_size, n);
1598 * If discard_passdown was enabled verify that the data device
1599 * supports discards. Disable discard_passdown if not.
1601 static void disable_passdown_if_not_supported(struct pool_c *pt)
1603 struct pool *pool = pt->pool;
1604 struct block_device *data_bdev = pt->data_dev->bdev;
1605 struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
1606 sector_t block_size = pool->sectors_per_block << SECTOR_SHIFT;
1607 const char *reason = NULL;
1608 char buf[BDEVNAME_SIZE];
1610 if (!pt->adjusted_pf.discard_passdown)
1613 if (!data_dev_supports_discard(pt))
1614 reason = "discard unsupported";
1616 else if (data_limits->max_discard_sectors < pool->sectors_per_block)
1617 reason = "max discard sectors smaller than a block";
1619 else if (data_limits->discard_granularity > block_size)
1620 reason = "discard granularity larger than a block";
1622 else if (!is_factor(block_size, data_limits->discard_granularity))
1623 reason = "discard granularity not a factor of block size";
1626 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason);
1627 pt->adjusted_pf.discard_passdown = false;
1631 static int bind_control_target(struct pool *pool, struct dm_target *ti)
1633 struct pool_c *pt = ti->private;
1636 * We want to make sure that degraded pools are never upgraded.
1638 enum pool_mode old_mode = pool->pf.mode;
1639 enum pool_mode new_mode = pt->adjusted_pf.mode;
1641 if (old_mode > new_mode)
1642 new_mode = old_mode;
1645 pool->low_water_blocks = pt->low_water_blocks;
1646 pool->pf = pt->adjusted_pf;
1648 set_pool_mode(pool, new_mode);
1653 static void unbind_control_target(struct pool *pool, struct dm_target *ti)
1659 /*----------------------------------------------------------------
1661 *--------------------------------------------------------------*/
1662 /* Initialize pool features. */
1663 static void pool_features_init(struct pool_features *pf)
1665 pf->mode = PM_WRITE;
1666 pf->zero_new_blocks = true;
1667 pf->discard_enabled = true;
1668 pf->discard_passdown = true;
1671 static void __pool_destroy(struct pool *pool)
1673 __pool_table_remove(pool);
1675 if (dm_pool_metadata_close(pool->pmd) < 0)
1676 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1678 dm_bio_prison_destroy(pool->prison);
1679 dm_kcopyd_client_destroy(pool->copier);
1682 destroy_workqueue(pool->wq);
1684 if (pool->next_mapping)
1685 mempool_free(pool->next_mapping, pool->mapping_pool);
1686 mempool_destroy(pool->mapping_pool);
1687 dm_deferred_set_destroy(pool->shared_read_ds);
1688 dm_deferred_set_destroy(pool->all_io_ds);
1692 static struct kmem_cache *_new_mapping_cache;
1694 static struct pool *pool_create(struct mapped_device *pool_md,
1695 struct block_device *metadata_dev,
1696 unsigned long block_size,
1697 int read_only, char **error)
1702 struct dm_pool_metadata *pmd;
1703 bool format_device = read_only ? false : true;
1705 pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
1707 *error = "Error creating metadata object";
1708 return (struct pool *)pmd;
1711 pool = kmalloc(sizeof(*pool), GFP_KERNEL);
1713 *error = "Error allocating memory for pool";
1714 err_p = ERR_PTR(-ENOMEM);
1719 pool->sectors_per_block = block_size;
1720 if (block_size & (block_size - 1))
1721 pool->sectors_per_block_shift = -1;
1723 pool->sectors_per_block_shift = __ffs(block_size);
1724 pool->low_water_blocks = 0;
1725 pool_features_init(&pool->pf);
1726 pool->prison = dm_bio_prison_create(PRISON_CELLS);
1727 if (!pool->prison) {
1728 *error = "Error creating pool's bio prison";
1729 err_p = ERR_PTR(-ENOMEM);
1733 pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
1734 if (IS_ERR(pool->copier)) {
1735 r = PTR_ERR(pool->copier);
1736 *error = "Error creating pool's kcopyd client";
1738 goto bad_kcopyd_client;
1742 * Create singlethreaded workqueue that will service all devices
1743 * that use this metadata.
1745 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
1747 *error = "Error creating pool's workqueue";
1748 err_p = ERR_PTR(-ENOMEM);
1752 INIT_WORK(&pool->worker, do_worker);
1753 INIT_DELAYED_WORK(&pool->waker, do_waker);
1754 spin_lock_init(&pool->lock);
1755 bio_list_init(&pool->deferred_bios);
1756 bio_list_init(&pool->deferred_flush_bios);
1757 INIT_LIST_HEAD(&pool->prepared_mappings);
1758 INIT_LIST_HEAD(&pool->prepared_discards);
1759 pool->low_water_triggered = 0;
1760 pool->no_free_space = 0;
1761 bio_list_init(&pool->retry_on_resume_list);
1763 pool->shared_read_ds = dm_deferred_set_create();
1764 if (!pool->shared_read_ds) {
1765 *error = "Error creating pool's shared read deferred set";
1766 err_p = ERR_PTR(-ENOMEM);
1767 goto bad_shared_read_ds;
1770 pool->all_io_ds = dm_deferred_set_create();
1771 if (!pool->all_io_ds) {
1772 *error = "Error creating pool's all io deferred set";
1773 err_p = ERR_PTR(-ENOMEM);
1777 pool->next_mapping = NULL;
1778 pool->mapping_pool = mempool_create_slab_pool(MAPPING_POOL_SIZE,
1779 _new_mapping_cache);
1780 if (!pool->mapping_pool) {
1781 *error = "Error creating pool's mapping mempool";
1782 err_p = ERR_PTR(-ENOMEM);
1783 goto bad_mapping_pool;
1786 pool->ref_count = 1;
1787 pool->last_commit_jiffies = jiffies;
1788 pool->pool_md = pool_md;
1789 pool->md_dev = metadata_dev;
1790 __pool_table_insert(pool);
1795 dm_deferred_set_destroy(pool->all_io_ds);
1797 dm_deferred_set_destroy(pool->shared_read_ds);
1799 destroy_workqueue(pool->wq);
1801 dm_kcopyd_client_destroy(pool->copier);
1803 dm_bio_prison_destroy(pool->prison);
1807 if (dm_pool_metadata_close(pmd))
1808 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
1813 static void __pool_inc(struct pool *pool)
1815 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1819 static void __pool_dec(struct pool *pool)
1821 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
1822 BUG_ON(!pool->ref_count);
1823 if (!--pool->ref_count)
1824 __pool_destroy(pool);
1827 static struct pool *__pool_find(struct mapped_device *pool_md,
1828 struct block_device *metadata_dev,
1829 unsigned long block_size, int read_only,
1830 char **error, int *created)
1832 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
1835 if (pool->pool_md != pool_md) {
1836 *error = "metadata device already in use by a pool";
1837 return ERR_PTR(-EBUSY);
1842 pool = __pool_table_lookup(pool_md);
1844 if (pool->md_dev != metadata_dev) {
1845 *error = "different pool cannot replace a pool";
1846 return ERR_PTR(-EINVAL);
1851 pool = pool_create(pool_md, metadata_dev, block_size, read_only, error);
1859 /*----------------------------------------------------------------
1860 * Pool target methods
1861 *--------------------------------------------------------------*/
1862 static void pool_dtr(struct dm_target *ti)
1864 struct pool_c *pt = ti->private;
1866 mutex_lock(&dm_thin_pool_table.mutex);
1868 unbind_control_target(pt->pool, ti);
1869 __pool_dec(pt->pool);
1870 dm_put_device(ti, pt->metadata_dev);
1871 dm_put_device(ti, pt->data_dev);
1874 mutex_unlock(&dm_thin_pool_table.mutex);
1877 static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
1878 struct dm_target *ti)
1882 const char *arg_name;
1884 static struct dm_arg _args[] = {
1885 {0, 3, "Invalid number of pool feature arguments"},
1889 * No feature arguments supplied.
1894 r = dm_read_arg_group(_args, as, &argc, &ti->error);
1898 while (argc && !r) {
1899 arg_name = dm_shift_arg(as);
1902 if (!strcasecmp(arg_name, "skip_block_zeroing"))
1903 pf->zero_new_blocks = false;
1905 else if (!strcasecmp(arg_name, "ignore_discard"))
1906 pf->discard_enabled = false;
1908 else if (!strcasecmp(arg_name, "no_discard_passdown"))
1909 pf->discard_passdown = false;
1911 else if (!strcasecmp(arg_name, "read_only"))
1912 pf->mode = PM_READ_ONLY;
1915 ti->error = "Unrecognised pool feature requested";
1924 static void metadata_low_callback(void *context)
1926 struct pool *pool = context;
1928 DMWARN("%s: reached low water mark for metadata device: sending event.",
1929 dm_device_name(pool->pool_md));
1931 dm_table_event(pool->ti->table);
1934 static sector_t get_metadata_dev_size(struct block_device *bdev)
1936 sector_t metadata_dev_size = i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
1937 char buffer[BDEVNAME_SIZE];
1939 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING) {
1940 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
1941 bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS);
1942 metadata_dev_size = THIN_METADATA_MAX_SECTORS_WARNING;
1945 return metadata_dev_size;
1948 static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
1950 sector_t metadata_dev_size = get_metadata_dev_size(bdev);
1952 sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
1954 return metadata_dev_size;
1958 * When a metadata threshold is crossed a dm event is triggered, and
1959 * userland should respond by growing the metadata device. We could let
1960 * userland set the threshold, like we do with the data threshold, but I'm
1961 * not sure they know enough to do this well.
1963 static dm_block_t calc_metadata_threshold(struct pool_c *pt)
1966 * 4M is ample for all ops with the possible exception of thin
1967 * device deletion which is harmless if it fails (just retry the
1968 * delete after you've grown the device).
1970 dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
1971 return min((dm_block_t)1024ULL /* 4M */, quarter);
1975 * thin-pool <metadata dev> <data dev>
1976 * <data block size (sectors)>
1977 * <low water mark (blocks)>
1978 * [<#feature args> [<arg>]*]
1980 * Optional feature arguments are:
1981 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
1982 * ignore_discard: disable discard
1983 * no_discard_passdown: don't pass discards down to the data device
1985 static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
1987 int r, pool_created = 0;
1990 struct pool_features pf;
1991 struct dm_arg_set as;
1992 struct dm_dev *data_dev;
1993 unsigned long block_size;
1994 dm_block_t low_water_blocks;
1995 struct dm_dev *metadata_dev;
1996 fmode_t metadata_mode;
1999 * FIXME Remove validation from scope of lock.
2001 mutex_lock(&dm_thin_pool_table.mutex);
2004 ti->error = "Invalid argument count";
2013 * Set default pool features.
2015 pool_features_init(&pf);
2017 dm_consume_args(&as, 4);
2018 r = parse_pool_features(&as, &pf, ti);
2022 metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
2023 r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
2025 ti->error = "Error opening metadata block device";
2030 * Run for the side-effect of possibly issuing a warning if the
2031 * device is too big.
2033 (void) get_metadata_dev_size(metadata_dev->bdev);
2035 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
2037 ti->error = "Error getting data device";
2041 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
2042 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
2043 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
2044 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
2045 ti->error = "Invalid block size";
2050 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
2051 ti->error = "Invalid low water mark";
2056 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
2062 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
2063 block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
2070 * 'pool_created' reflects whether this is the first table load.
2071 * Top level discard support is not allowed to be changed after
2072 * initial load. This would require a pool reload to trigger thin
2075 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
2076 ti->error = "Discard support cannot be disabled once enabled";
2078 goto out_flags_changed;
2083 pt->metadata_dev = metadata_dev;
2084 pt->data_dev = data_dev;
2085 pt->low_water_blocks = low_water_blocks;
2086 pt->adjusted_pf = pt->requested_pf = pf;
2087 ti->num_flush_bios = 1;
2090 * Only need to enable discards if the pool should pass
2091 * them down to the data device. The thin device's discard
2092 * processing will cause mappings to be removed from the btree.
2094 if (pf.discard_enabled && pf.discard_passdown) {
2095 ti->num_discard_bios = 1;
2098 * Setting 'discards_supported' circumvents the normal
2099 * stacking of discard limits (this keeps the pool and
2100 * thin devices' discard limits consistent).
2102 ti->discards_supported = true;
2103 ti->discard_zeroes_data_unsupported = true;
2107 r = dm_pool_register_metadata_threshold(pt->pool->pmd,
2108 calc_metadata_threshold(pt),
2109 metadata_low_callback,
2114 pt->callbacks.congested_fn = pool_is_congested;
2115 dm_table_add_target_callbacks(ti->table, &pt->callbacks);
2117 mutex_unlock(&dm_thin_pool_table.mutex);
2126 dm_put_device(ti, data_dev);
2128 dm_put_device(ti, metadata_dev);
2130 mutex_unlock(&dm_thin_pool_table.mutex);
2135 static int pool_map(struct dm_target *ti, struct bio *bio)
2138 struct pool_c *pt = ti->private;
2139 struct pool *pool = pt->pool;
2140 unsigned long flags;
2143 * As this is a singleton target, ti->begin is always zero.
2145 spin_lock_irqsave(&pool->lock, flags);
2146 bio->bi_bdev = pt->data_dev->bdev;
2147 r = DM_MAPIO_REMAPPED;
2148 spin_unlock_irqrestore(&pool->lock, flags);
2153 static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
2156 struct pool_c *pt = ti->private;
2157 struct pool *pool = pt->pool;
2158 sector_t data_size = ti->len;
2159 dm_block_t sb_data_size;
2161 *need_commit = false;
2163 (void) sector_div(data_size, pool->sectors_per_block);
2165 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
2167 DMERR("failed to retrieve data device size");
2171 if (data_size < sb_data_size) {
2172 DMERR("pool target (%llu blocks) too small: expected %llu",
2173 (unsigned long long)data_size, sb_data_size);
2176 } else if (data_size > sb_data_size) {
2177 r = dm_pool_resize_data_dev(pool->pmd, data_size);
2179 DMERR("failed to resize data device");
2180 set_pool_mode(pool, PM_READ_ONLY);
2184 *need_commit = true;
2190 static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
2193 struct pool_c *pt = ti->private;
2194 struct pool *pool = pt->pool;
2195 dm_block_t metadata_dev_size, sb_metadata_dev_size;
2197 *need_commit = false;
2199 metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
2201 r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
2203 DMERR("failed to retrieve data device size");
2207 if (metadata_dev_size < sb_metadata_dev_size) {
2208 DMERR("metadata device (%llu blocks) too small: expected %llu",
2209 metadata_dev_size, sb_metadata_dev_size);
2212 } else if (metadata_dev_size > sb_metadata_dev_size) {
2213 r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
2215 DMERR("failed to resize metadata device");
2219 *need_commit = true;
2226 * Retrieves the number of blocks of the data device from
2227 * the superblock and compares it to the actual device size,
2228 * thus resizing the data device in case it has grown.
2230 * This both copes with opening preallocated data devices in the ctr
2231 * being followed by a resume
2233 * calling the resume method individually after userspace has
2234 * grown the data device in reaction to a table event.
2236 static int pool_preresume(struct dm_target *ti)
2239 bool need_commit1, need_commit2;
2240 struct pool_c *pt = ti->private;
2241 struct pool *pool = pt->pool;
2244 * Take control of the pool object.
2246 r = bind_control_target(pool, ti);
2250 r = maybe_resize_data_dev(ti, &need_commit1);
2254 r = maybe_resize_metadata_dev(ti, &need_commit2);
2258 if (need_commit1 || need_commit2)
2259 (void) commit_or_fallback(pool);
2264 static void pool_resume(struct dm_target *ti)
2266 struct pool_c *pt = ti->private;
2267 struct pool *pool = pt->pool;
2268 unsigned long flags;
2270 spin_lock_irqsave(&pool->lock, flags);
2271 pool->low_water_triggered = 0;
2272 pool->no_free_space = 0;
2273 __requeue_bios(pool);
2274 spin_unlock_irqrestore(&pool->lock, flags);
2276 do_waker(&pool->waker.work);
2279 static void pool_postsuspend(struct dm_target *ti)
2281 struct pool_c *pt = ti->private;
2282 struct pool *pool = pt->pool;
2284 cancel_delayed_work(&pool->waker);
2285 flush_workqueue(pool->wq);
2286 (void) commit_or_fallback(pool);
2289 static int check_arg_count(unsigned argc, unsigned args_required)
2291 if (argc != args_required) {
2292 DMWARN("Message received with %u arguments instead of %u.",
2293 argc, args_required);
2300 static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
2302 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
2303 *dev_id <= MAX_DEV_ID)
2307 DMWARN("Message received with invalid device id: %s", arg);
2312 static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
2317 r = check_arg_count(argc, 2);
2321 r = read_dev_id(argv[1], &dev_id, 1);
2325 r = dm_pool_create_thin(pool->pmd, dev_id);
2327 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2335 static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2338 dm_thin_id origin_dev_id;
2341 r = check_arg_count(argc, 3);
2345 r = read_dev_id(argv[1], &dev_id, 1);
2349 r = read_dev_id(argv[2], &origin_dev_id, 1);
2353 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
2355 DMWARN("Creation of new snapshot %s of device %s failed.",
2363 static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
2368 r = check_arg_count(argc, 2);
2372 r = read_dev_id(argv[1], &dev_id, 1);
2376 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
2378 DMWARN("Deletion of thin device %s failed.", argv[1]);
2383 static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
2385 dm_thin_id old_id, new_id;
2388 r = check_arg_count(argc, 3);
2392 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
2393 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
2397 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
2398 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
2402 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
2404 DMWARN("Failed to change transaction id from %s to %s.",
2412 static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2416 r = check_arg_count(argc, 1);
2420 (void) commit_or_fallback(pool);
2422 r = dm_pool_reserve_metadata_snap(pool->pmd);
2424 DMWARN("reserve_metadata_snap message failed.");
2429 static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
2433 r = check_arg_count(argc, 1);
2437 r = dm_pool_release_metadata_snap(pool->pmd);
2439 DMWARN("release_metadata_snap message failed.");
2445 * Messages supported:
2446 * create_thin <dev_id>
2447 * create_snap <dev_id> <origin_id>
2449 * trim <dev_id> <new_size_in_sectors>
2450 * set_transaction_id <current_trans_id> <new_trans_id>
2451 * reserve_metadata_snap
2452 * release_metadata_snap
2454 static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
2457 struct pool_c *pt = ti->private;
2458 struct pool *pool = pt->pool;
2460 if (!strcasecmp(argv[0], "create_thin"))
2461 r = process_create_thin_mesg(argc, argv, pool);
2463 else if (!strcasecmp(argv[0], "create_snap"))
2464 r = process_create_snap_mesg(argc, argv, pool);
2466 else if (!strcasecmp(argv[0], "delete"))
2467 r = process_delete_mesg(argc, argv, pool);
2469 else if (!strcasecmp(argv[0], "set_transaction_id"))
2470 r = process_set_transaction_id_mesg(argc, argv, pool);
2472 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
2473 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
2475 else if (!strcasecmp(argv[0], "release_metadata_snap"))
2476 r = process_release_metadata_snap_mesg(argc, argv, pool);
2479 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
2482 (void) commit_or_fallback(pool);
2487 static void emit_flags(struct pool_features *pf, char *result,
2488 unsigned sz, unsigned maxlen)
2490 unsigned count = !pf->zero_new_blocks + !pf->discard_enabled +
2491 !pf->discard_passdown + (pf->mode == PM_READ_ONLY);
2492 DMEMIT("%u ", count);
2494 if (!pf->zero_new_blocks)
2495 DMEMIT("skip_block_zeroing ");
2497 if (!pf->discard_enabled)
2498 DMEMIT("ignore_discard ");
2500 if (!pf->discard_passdown)
2501 DMEMIT("no_discard_passdown ");
2503 if (pf->mode == PM_READ_ONLY)
2504 DMEMIT("read_only ");
2509 * <transaction id> <used metadata sectors>/<total metadata sectors>
2510 * <used data sectors>/<total data sectors> <held metadata root>
2512 static void pool_status(struct dm_target *ti, status_type_t type,
2513 unsigned status_flags, char *result, unsigned maxlen)
2517 uint64_t transaction_id;
2518 dm_block_t nr_free_blocks_data;
2519 dm_block_t nr_free_blocks_metadata;
2520 dm_block_t nr_blocks_data;
2521 dm_block_t nr_blocks_metadata;
2522 dm_block_t held_root;
2523 char buf[BDEVNAME_SIZE];
2524 char buf2[BDEVNAME_SIZE];
2525 struct pool_c *pt = ti->private;
2526 struct pool *pool = pt->pool;
2529 case STATUSTYPE_INFO:
2530 if (get_pool_mode(pool) == PM_FAIL) {
2535 /* Commit to ensure statistics aren't out-of-date */
2536 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
2537 (void) commit_or_fallback(pool);
2539 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
2541 DMERR("dm_pool_get_metadata_transaction_id returned %d", r);
2545 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
2547 DMERR("dm_pool_get_free_metadata_block_count returned %d", r);
2551 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
2553 DMERR("dm_pool_get_metadata_dev_size returned %d", r);
2557 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
2559 DMERR("dm_pool_get_free_block_count returned %d", r);
2563 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
2565 DMERR("dm_pool_get_data_dev_size returned %d", r);
2569 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
2571 DMERR("dm_pool_get_metadata_snap returned %d", r);
2575 DMEMIT("%llu %llu/%llu %llu/%llu ",
2576 (unsigned long long)transaction_id,
2577 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
2578 (unsigned long long)nr_blocks_metadata,
2579 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
2580 (unsigned long long)nr_blocks_data);
2583 DMEMIT("%llu ", held_root);
2587 if (pool->pf.mode == PM_READ_ONLY)
2592 if (!pool->pf.discard_enabled)
2593 DMEMIT("ignore_discard");
2594 else if (pool->pf.discard_passdown)
2595 DMEMIT("discard_passdown");
2597 DMEMIT("no_discard_passdown");
2601 case STATUSTYPE_TABLE:
2602 DMEMIT("%s %s %lu %llu ",
2603 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
2604 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
2605 (unsigned long)pool->sectors_per_block,
2606 (unsigned long long)pt->low_water_blocks);
2607 emit_flags(&pt->requested_pf, result, sz, maxlen);
2616 static int pool_iterate_devices(struct dm_target *ti,
2617 iterate_devices_callout_fn fn, void *data)
2619 struct pool_c *pt = ti->private;
2621 return fn(ti, pt->data_dev, 0, ti->len, data);
2624 static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2625 struct bio_vec *biovec, int max_size)
2627 struct pool_c *pt = ti->private;
2628 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
2630 if (!q->merge_bvec_fn)
2633 bvm->bi_bdev = pt->data_dev->bdev;
2635 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2638 static void set_discard_limits(struct pool_c *pt, struct queue_limits *limits)
2640 struct pool *pool = pt->pool;
2641 struct queue_limits *data_limits;
2643 limits->max_discard_sectors = pool->sectors_per_block;
2646 * discard_granularity is just a hint, and not enforced.
2648 if (pt->adjusted_pf.discard_passdown) {
2649 data_limits = &bdev_get_queue(pt->data_dev->bdev)->limits;
2650 limits->discard_granularity = data_limits->discard_granularity;
2652 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
2655 static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
2657 struct pool_c *pt = ti->private;
2658 struct pool *pool = pt->pool;
2660 blk_limits_io_min(limits, 0);
2661 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
2664 * pt->adjusted_pf is a staging area for the actual features to use.
2665 * They get transferred to the live pool in bind_control_target()
2666 * called from pool_preresume().
2668 if (!pt->adjusted_pf.discard_enabled)
2671 disable_passdown_if_not_supported(pt);
2673 set_discard_limits(pt, limits);
2676 static struct target_type pool_target = {
2677 .name = "thin-pool",
2678 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
2679 DM_TARGET_IMMUTABLE,
2680 .version = {1, 8, 0},
2681 .module = THIS_MODULE,
2685 .postsuspend = pool_postsuspend,
2686 .preresume = pool_preresume,
2687 .resume = pool_resume,
2688 .message = pool_message,
2689 .status = pool_status,
2690 .merge = pool_merge,
2691 .iterate_devices = pool_iterate_devices,
2692 .io_hints = pool_io_hints,
2695 /*----------------------------------------------------------------
2696 * Thin target methods
2697 *--------------------------------------------------------------*/
2698 static void thin_dtr(struct dm_target *ti)
2700 struct thin_c *tc = ti->private;
2702 mutex_lock(&dm_thin_pool_table.mutex);
2704 __pool_dec(tc->pool);
2705 dm_pool_close_thin_device(tc->td);
2706 dm_put_device(ti, tc->pool_dev);
2708 dm_put_device(ti, tc->origin_dev);
2711 mutex_unlock(&dm_thin_pool_table.mutex);
2715 * Thin target parameters:
2717 * <pool_dev> <dev_id> [origin_dev]
2719 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2720 * dev_id: the internal device identifier
2721 * origin_dev: a device external to the pool that should act as the origin
2723 * If the pool device has discards disabled, they get disabled for the thin
2726 static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
2730 struct dm_dev *pool_dev, *origin_dev;
2731 struct mapped_device *pool_md;
2733 mutex_lock(&dm_thin_pool_table.mutex);
2735 if (argc != 2 && argc != 3) {
2736 ti->error = "Invalid argument count";
2741 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
2743 ti->error = "Out of memory";
2749 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
2751 ti->error = "Error opening origin device";
2752 goto bad_origin_dev;
2754 tc->origin_dev = origin_dev;
2757 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
2759 ti->error = "Error opening pool device";
2762 tc->pool_dev = pool_dev;
2764 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
2765 ti->error = "Invalid device id";
2770 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
2772 ti->error = "Couldn't get pool mapped device";
2777 tc->pool = __pool_table_lookup(pool_md);
2779 ti->error = "Couldn't find pool object";
2781 goto bad_pool_lookup;
2783 __pool_inc(tc->pool);
2785 if (get_pool_mode(tc->pool) == PM_FAIL) {
2786 ti->error = "Couldn't open thin device, Pool is in fail mode";
2791 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
2793 ti->error = "Couldn't open thin internal device";
2797 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
2799 goto bad_target_max_io_len;
2801 ti->num_flush_bios = 1;
2802 ti->flush_supported = true;
2803 ti->per_bio_data_size = sizeof(struct dm_thin_endio_hook);
2805 /* In case the pool supports discards, pass them on. */
2806 if (tc->pool->pf.discard_enabled) {
2807 ti->discards_supported = true;
2808 ti->num_discard_bios = 1;
2809 ti->discard_zeroes_data_unsupported = true;
2810 /* Discard bios must be split on a block boundary */
2811 ti->split_discard_bios = true;
2816 mutex_unlock(&dm_thin_pool_table.mutex);
2820 bad_target_max_io_len:
2821 dm_pool_close_thin_device(tc->td);
2823 __pool_dec(tc->pool);
2827 dm_put_device(ti, tc->pool_dev);
2830 dm_put_device(ti, tc->origin_dev);
2834 mutex_unlock(&dm_thin_pool_table.mutex);
2839 static int thin_map(struct dm_target *ti, struct bio *bio)
2841 bio->bi_sector = dm_target_offset(ti, bio->bi_sector);
2843 return thin_bio_map(ti, bio);
2846 static int thin_endio(struct dm_target *ti, struct bio *bio, int err)
2848 unsigned long flags;
2849 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2850 struct list_head work;
2851 struct dm_thin_new_mapping *m, *tmp;
2852 struct pool *pool = h->tc->pool;
2854 if (h->shared_read_entry) {
2855 INIT_LIST_HEAD(&work);
2856 dm_deferred_entry_dec(h->shared_read_entry, &work);
2858 spin_lock_irqsave(&pool->lock, flags);
2859 list_for_each_entry_safe(m, tmp, &work, list) {
2862 __maybe_add_mapping(m);
2864 spin_unlock_irqrestore(&pool->lock, flags);
2867 if (h->all_io_entry) {
2868 INIT_LIST_HEAD(&work);
2869 dm_deferred_entry_dec(h->all_io_entry, &work);
2870 if (!list_empty(&work)) {
2871 spin_lock_irqsave(&pool->lock, flags);
2872 list_for_each_entry_safe(m, tmp, &work, list)
2873 list_add(&m->list, &pool->prepared_discards);
2874 spin_unlock_irqrestore(&pool->lock, flags);
2882 static void thin_postsuspend(struct dm_target *ti)
2884 if (dm_noflush_suspending(ti))
2885 requeue_io((struct thin_c *)ti->private);
2889 * <nr mapped sectors> <highest mapped sector>
2891 static void thin_status(struct dm_target *ti, status_type_t type,
2892 unsigned status_flags, char *result, unsigned maxlen)
2896 dm_block_t mapped, highest;
2897 char buf[BDEVNAME_SIZE];
2898 struct thin_c *tc = ti->private;
2900 if (get_pool_mode(tc->pool) == PM_FAIL) {
2909 case STATUSTYPE_INFO:
2910 r = dm_thin_get_mapped_count(tc->td, &mapped);
2912 DMERR("dm_thin_get_mapped_count returned %d", r);
2916 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
2918 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
2922 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
2924 DMEMIT("%llu", ((highest + 1) *
2925 tc->pool->sectors_per_block) - 1);
2930 case STATUSTYPE_TABLE:
2932 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
2933 (unsigned long) tc->dev_id);
2935 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
2946 static int thin_iterate_devices(struct dm_target *ti,
2947 iterate_devices_callout_fn fn, void *data)
2950 struct thin_c *tc = ti->private;
2951 struct pool *pool = tc->pool;
2954 * We can't call dm_pool_get_data_dev_size() since that blocks. So
2955 * we follow a more convoluted path through to the pool's target.
2958 return 0; /* nothing is bound */
2960 blocks = pool->ti->len;
2961 (void) sector_div(blocks, pool->sectors_per_block);
2963 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
2968 static struct target_type thin_target = {
2970 .version = {1, 8, 0},
2971 .module = THIS_MODULE,
2975 .end_io = thin_endio,
2976 .postsuspend = thin_postsuspend,
2977 .status = thin_status,
2978 .iterate_devices = thin_iterate_devices,
2981 /*----------------------------------------------------------------*/
2983 static int __init dm_thin_init(void)
2989 r = dm_register_target(&thin_target);
2993 r = dm_register_target(&pool_target);
2995 goto bad_pool_target;
2999 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
3000 if (!_new_mapping_cache)
3001 goto bad_new_mapping_cache;
3005 bad_new_mapping_cache:
3006 dm_unregister_target(&pool_target);
3008 dm_unregister_target(&thin_target);
3013 static void dm_thin_exit(void)
3015 dm_unregister_target(&thin_target);
3016 dm_unregister_target(&pool_target);
3018 kmem_cache_destroy(_new_mapping_cache);
3021 module_init(dm_thin_init);
3022 module_exit(dm_thin_exit);
3024 MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
3025 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3026 MODULE_LICENSE("GPL");