2 * Copyright (C) 2011 Red Hat UK.
4 * This file is released under the GPL.
7 #include "dm-thin-metadata.h"
9 #include <linux/device-mapper.h>
10 #include <linux/dm-io.h>
11 #include <linux/dm-kcopyd.h>
12 #include <linux/list.h>
13 #include <linux/init.h>
14 #include <linux/module.h>
15 #include <linux/slab.h>
17 #define DM_MSG_PREFIX "thin"
22 #define ENDIO_HOOK_POOL_SIZE 10240
23 #define DEFERRED_SET_SIZE 64
24 #define MAPPING_POOL_SIZE 1024
25 #define PRISON_CELLS 1024
26 #define COMMIT_PERIOD HZ
29 * The block size of the device holding pool data must be
30 * between 64KB and 1GB.
32 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
33 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
36 * Device id is restricted to 24 bits.
38 #define MAX_DEV_ID ((1 << 24) - 1)
41 * How do we handle breaking sharing of data blocks?
42 * =================================================
44 * We use a standard copy-on-write btree to store the mappings for the
45 * devices (note I'm talking about copy-on-write of the metadata here, not
46 * the data). When you take an internal snapshot you clone the root node
47 * of the origin btree. After this there is no concept of an origin or a
48 * snapshot. They are just two device trees that happen to point to the
51 * When we get a write in we decide if it's to a shared data block using
52 * some timestamp magic. If it is, we have to break sharing.
54 * Let's say we write to a shared block in what was the origin. The
57 * i) plug io further to this physical block. (see bio_prison code).
59 * ii) quiesce any read io to that shared data block. Obviously
60 * including all devices that share this block. (see deferred_set code)
62 * iii) copy the data block to a newly allocate block. This step can be
63 * missed out if the io covers the block. (schedule_copy).
65 * iv) insert the new mapping into the origin's btree
66 * (process_prepared_mapping). This act of inserting breaks some
67 * sharing of btree nodes between the two devices. Breaking sharing only
68 * effects the btree of that specific device. Btrees for the other
69 * devices that share the block never change. The btree for the origin
70 * device as it was after the last commit is untouched, ie. we're using
71 * persistent data structures in the functional programming sense.
73 * v) unplug io to this physical block, including the io that triggered
74 * the breaking of sharing.
76 * Steps (ii) and (iii) occur in parallel.
78 * The metadata _doesn't_ need to be committed before the io continues. We
79 * get away with this because the io is always written to a _new_ block.
80 * If there's a crash, then:
82 * - The origin mapping will point to the old origin block (the shared
83 * one). This will contain the data as it was before the io that triggered
84 * the breaking of sharing came in.
86 * - The snap mapping still points to the old block. As it would after
89 * The downside of this scheme is the timestamp magic isn't perfect, and
90 * will continue to think that data block in the snapshot device is shared
91 * even after the write to the origin has broken sharing. I suspect data
92 * blocks will typically be shared by many different devices, so we're
93 * breaking sharing n + 1 times, rather than n, where n is the number of
94 * devices that reference this data block. At the moment I think the
95 * benefits far, far outweigh the disadvantages.
98 /*----------------------------------------------------------------*/
101 * Sometimes we can't deal with a bio straight away. We put them in prison
102 * where they can't cause any mischief. Bios are put in a cell identified
103 * by a key, multiple bios can be in the same cell. When the cell is
104 * subsequently unlocked the bios become available.
115 struct hlist_node list
;
116 struct bio_prison
*prison
;
119 struct bio_list bios
;
124 mempool_t
*cell_pool
;
128 struct hlist_head
*cells
;
131 static uint32_t calc_nr_buckets(unsigned nr_cells
)
136 nr_cells
= min(nr_cells
, 8192u);
145 * @nr_cells should be the number of cells you want in use _concurrently_.
146 * Don't confuse it with the number of distinct keys.
148 static struct bio_prison
*prison_create(unsigned nr_cells
)
151 uint32_t nr_buckets
= calc_nr_buckets(nr_cells
);
152 size_t len
= sizeof(struct bio_prison
) +
153 (sizeof(struct hlist_head
) * nr_buckets
);
154 struct bio_prison
*prison
= kmalloc(len
, GFP_KERNEL
);
159 spin_lock_init(&prison
->lock
);
160 prison
->cell_pool
= mempool_create_kmalloc_pool(nr_cells
,
161 sizeof(struct cell
));
162 if (!prison
->cell_pool
) {
167 prison
->nr_buckets
= nr_buckets
;
168 prison
->hash_mask
= nr_buckets
- 1;
169 prison
->cells
= (struct hlist_head
*) (prison
+ 1);
170 for (i
= 0; i
< nr_buckets
; i
++)
171 INIT_HLIST_HEAD(prison
->cells
+ i
);
176 static void prison_destroy(struct bio_prison
*prison
)
178 mempool_destroy(prison
->cell_pool
);
182 static uint32_t hash_key(struct bio_prison
*prison
, struct cell_key
*key
)
184 const unsigned long BIG_PRIME
= 4294967291UL;
185 uint64_t hash
= key
->block
* BIG_PRIME
;
187 return (uint32_t) (hash
& prison
->hash_mask
);
190 static int keys_equal(struct cell_key
*lhs
, struct cell_key
*rhs
)
192 return (lhs
->virtual == rhs
->virtual) &&
193 (lhs
->dev
== rhs
->dev
) &&
194 (lhs
->block
== rhs
->block
);
197 static struct cell
*__search_bucket(struct hlist_head
*bucket
,
198 struct cell_key
*key
)
201 struct hlist_node
*tmp
;
203 hlist_for_each_entry(cell
, tmp
, bucket
, list
)
204 if (keys_equal(&cell
->key
, key
))
211 * This may block if a new cell needs allocating. You must ensure that
212 * cells will be unlocked even if the calling thread is blocked.
214 * Returns 1 if the cell was already held, 0 if @inmate is the new holder.
216 static int bio_detain(struct bio_prison
*prison
, struct cell_key
*key
,
217 struct bio
*inmate
, struct cell
**ref
)
221 uint32_t hash
= hash_key(prison
, key
);
222 struct cell
*cell
, *cell2
;
224 BUG_ON(hash
> prison
->nr_buckets
);
226 spin_lock_irqsave(&prison
->lock
, flags
);
228 cell
= __search_bucket(prison
->cells
+ hash
, key
);
230 bio_list_add(&cell
->bios
, inmate
);
235 * Allocate a new cell
237 spin_unlock_irqrestore(&prison
->lock
, flags
);
238 cell2
= mempool_alloc(prison
->cell_pool
, GFP_NOIO
);
239 spin_lock_irqsave(&prison
->lock
, flags
);
242 * We've been unlocked, so we have to double check that
243 * nobody else has inserted this cell in the meantime.
245 cell
= __search_bucket(prison
->cells
+ hash
, key
);
247 mempool_free(cell2
, prison
->cell_pool
);
248 bio_list_add(&cell
->bios
, inmate
);
257 cell
->prison
= prison
;
258 memcpy(&cell
->key
, key
, sizeof(cell
->key
));
259 cell
->holder
= inmate
;
260 bio_list_init(&cell
->bios
);
261 hlist_add_head(&cell
->list
, prison
->cells
+ hash
);
266 spin_unlock_irqrestore(&prison
->lock
, flags
);
274 * @inmates must have been initialised prior to this call
276 static void __cell_release(struct cell
*cell
, struct bio_list
*inmates
)
278 struct bio_prison
*prison
= cell
->prison
;
280 hlist_del(&cell
->list
);
282 bio_list_add(inmates
, cell
->holder
);
283 bio_list_merge(inmates
, &cell
->bios
);
285 mempool_free(cell
, prison
->cell_pool
);
288 static void cell_release(struct cell
*cell
, struct bio_list
*bios
)
291 struct bio_prison
*prison
= cell
->prison
;
293 spin_lock_irqsave(&prison
->lock
, flags
);
294 __cell_release(cell
, bios
);
295 spin_unlock_irqrestore(&prison
->lock
, flags
);
299 * There are a couple of places where we put a bio into a cell briefly
300 * before taking it out again. In these situations we know that no other
301 * bio may be in the cell. This function releases the cell, and also does
304 static void __cell_release_singleton(struct cell
*cell
, struct bio
*bio
)
306 hlist_del(&cell
->list
);
307 BUG_ON(cell
->holder
!= bio
);
308 BUG_ON(!bio_list_empty(&cell
->bios
));
311 static void cell_release_singleton(struct cell
*cell
, struct bio
*bio
)
314 struct bio_prison
*prison
= cell
->prison
;
316 spin_lock_irqsave(&prison
->lock
, flags
);
317 __cell_release_singleton(cell
, bio
);
318 spin_unlock_irqrestore(&prison
->lock
, flags
);
322 * Sometimes we don't want the holder, just the additional bios.
324 static void __cell_release_no_holder(struct cell
*cell
, struct bio_list
*inmates
)
326 struct bio_prison
*prison
= cell
->prison
;
328 hlist_del(&cell
->list
);
329 bio_list_merge(inmates
, &cell
->bios
);
331 mempool_free(cell
, prison
->cell_pool
);
334 static void cell_release_no_holder(struct cell
*cell
, struct bio_list
*inmates
)
337 struct bio_prison
*prison
= cell
->prison
;
339 spin_lock_irqsave(&prison
->lock
, flags
);
340 __cell_release_no_holder(cell
, inmates
);
341 spin_unlock_irqrestore(&prison
->lock
, flags
);
344 static void cell_error(struct cell
*cell
)
346 struct bio_prison
*prison
= cell
->prison
;
347 struct bio_list bios
;
351 bio_list_init(&bios
);
353 spin_lock_irqsave(&prison
->lock
, flags
);
354 __cell_release(cell
, &bios
);
355 spin_unlock_irqrestore(&prison
->lock
, flags
);
357 while ((bio
= bio_list_pop(&bios
)))
361 /*----------------------------------------------------------------*/
364 * We use the deferred set to keep track of pending reads to shared blocks.
365 * We do this to ensure the new mapping caused by a write isn't performed
366 * until these prior reads have completed. Otherwise the insertion of the
367 * new mapping could free the old block that the read bios are mapped to.
371 struct deferred_entry
{
372 struct deferred_set
*ds
;
374 struct list_head work_items
;
377 struct deferred_set
{
379 unsigned current_entry
;
381 struct deferred_entry entries
[DEFERRED_SET_SIZE
];
384 static void ds_init(struct deferred_set
*ds
)
388 spin_lock_init(&ds
->lock
);
389 ds
->current_entry
= 0;
391 for (i
= 0; i
< DEFERRED_SET_SIZE
; i
++) {
392 ds
->entries
[i
].ds
= ds
;
393 ds
->entries
[i
].count
= 0;
394 INIT_LIST_HEAD(&ds
->entries
[i
].work_items
);
398 static struct deferred_entry
*ds_inc(struct deferred_set
*ds
)
401 struct deferred_entry
*entry
;
403 spin_lock_irqsave(&ds
->lock
, flags
);
404 entry
= ds
->entries
+ ds
->current_entry
;
406 spin_unlock_irqrestore(&ds
->lock
, flags
);
411 static unsigned ds_next(unsigned index
)
413 return (index
+ 1) % DEFERRED_SET_SIZE
;
416 static void __sweep(struct deferred_set
*ds
, struct list_head
*head
)
418 while ((ds
->sweeper
!= ds
->current_entry
) &&
419 !ds
->entries
[ds
->sweeper
].count
) {
420 list_splice_init(&ds
->entries
[ds
->sweeper
].work_items
, head
);
421 ds
->sweeper
= ds_next(ds
->sweeper
);
424 if ((ds
->sweeper
== ds
->current_entry
) && !ds
->entries
[ds
->sweeper
].count
)
425 list_splice_init(&ds
->entries
[ds
->sweeper
].work_items
, head
);
428 static void ds_dec(struct deferred_entry
*entry
, struct list_head
*head
)
432 spin_lock_irqsave(&entry
->ds
->lock
, flags
);
433 BUG_ON(!entry
->count
);
435 __sweep(entry
->ds
, head
);
436 spin_unlock_irqrestore(&entry
->ds
->lock
, flags
);
440 * Returns 1 if deferred or 0 if no pending items to delay job.
442 static int ds_add_work(struct deferred_set
*ds
, struct list_head
*work
)
448 spin_lock_irqsave(&ds
->lock
, flags
);
449 if ((ds
->sweeper
== ds
->current_entry
) &&
450 !ds
->entries
[ds
->current_entry
].count
)
453 list_add(work
, &ds
->entries
[ds
->current_entry
].work_items
);
454 next_entry
= ds_next(ds
->current_entry
);
455 if (!ds
->entries
[next_entry
].count
)
456 ds
->current_entry
= next_entry
;
458 spin_unlock_irqrestore(&ds
->lock
, flags
);
463 /*----------------------------------------------------------------*/
468 static void build_data_key(struct dm_thin_device
*td
,
469 dm_block_t b
, struct cell_key
*key
)
472 key
->dev
= dm_thin_dev_id(td
);
476 static void build_virtual_key(struct dm_thin_device
*td
, dm_block_t b
,
477 struct cell_key
*key
)
480 key
->dev
= dm_thin_dev_id(td
);
484 /*----------------------------------------------------------------*/
487 * A pool device ties together a metadata device and a data device. It
488 * also provides the interface for creating and destroying internal
493 struct pool_features
{
494 unsigned zero_new_blocks
:1;
495 unsigned discard_enabled
:1;
496 unsigned discard_passdown
:1;
500 struct list_head list
;
501 struct dm_target
*ti
; /* Only set if a pool target is bound */
503 struct mapped_device
*pool_md
;
504 struct block_device
*md_dev
;
505 struct dm_pool_metadata
*pmd
;
507 uint32_t sectors_per_block
;
508 unsigned block_shift
;
509 dm_block_t offset_mask
;
510 dm_block_t low_water_blocks
;
512 struct pool_features pf
;
513 unsigned low_water_triggered
:1; /* A dm event has been sent */
514 unsigned no_free_space
:1; /* A -ENOSPC warning has been issued */
516 struct bio_prison
*prison
;
517 struct dm_kcopyd_client
*copier
;
519 struct workqueue_struct
*wq
;
520 struct work_struct worker
;
521 struct delayed_work waker
;
524 unsigned long last_commit_jiffies
;
527 struct bio_list deferred_bios
;
528 struct bio_list deferred_flush_bios
;
529 struct list_head prepared_mappings
;
530 struct list_head prepared_discards
;
532 struct bio_list retry_on_resume_list
;
534 struct deferred_set shared_read_ds
;
535 struct deferred_set all_io_ds
;
537 struct new_mapping
*next_mapping
;
538 mempool_t
*mapping_pool
;
539 mempool_t
*endio_hook_pool
;
543 * Target context for a pool.
546 struct dm_target
*ti
;
548 struct dm_dev
*data_dev
;
549 struct dm_dev
*metadata_dev
;
550 struct dm_target_callbacks callbacks
;
552 dm_block_t low_water_blocks
;
553 struct pool_features pf
;
557 * Target context for a thin.
560 struct dm_dev
*pool_dev
;
561 struct dm_dev
*origin_dev
;
565 struct dm_thin_device
*td
;
568 /*----------------------------------------------------------------*/
571 * A global list of pools that uses a struct mapped_device as a key.
573 static struct dm_thin_pool_table
{
575 struct list_head pools
;
576 } dm_thin_pool_table
;
578 static void pool_table_init(void)
580 mutex_init(&dm_thin_pool_table
.mutex
);
581 INIT_LIST_HEAD(&dm_thin_pool_table
.pools
);
584 static void __pool_table_insert(struct pool
*pool
)
586 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
587 list_add(&pool
->list
, &dm_thin_pool_table
.pools
);
590 static void __pool_table_remove(struct pool
*pool
)
592 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
593 list_del(&pool
->list
);
596 static struct pool
*__pool_table_lookup(struct mapped_device
*md
)
598 struct pool
*pool
= NULL
, *tmp
;
600 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
602 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
603 if (tmp
->pool_md
== md
) {
612 static struct pool
*__pool_table_lookup_metadata_dev(struct block_device
*md_dev
)
614 struct pool
*pool
= NULL
, *tmp
;
616 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
618 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
619 if (tmp
->md_dev
== md_dev
) {
628 /*----------------------------------------------------------------*/
632 struct deferred_entry
*shared_read_entry
;
633 struct deferred_entry
*all_io_entry
;
634 struct new_mapping
*overwrite_mapping
;
637 static void __requeue_bio_list(struct thin_c
*tc
, struct bio_list
*master
)
640 struct bio_list bios
;
642 bio_list_init(&bios
);
643 bio_list_merge(&bios
, master
);
644 bio_list_init(master
);
646 while ((bio
= bio_list_pop(&bios
))) {
647 struct endio_hook
*h
= dm_get_mapinfo(bio
)->ptr
;
649 bio_endio(bio
, DM_ENDIO_REQUEUE
);
651 bio_list_add(master
, bio
);
655 static void requeue_io(struct thin_c
*tc
)
657 struct pool
*pool
= tc
->pool
;
660 spin_lock_irqsave(&pool
->lock
, flags
);
661 __requeue_bio_list(tc
, &pool
->deferred_bios
);
662 __requeue_bio_list(tc
, &pool
->retry_on_resume_list
);
663 spin_unlock_irqrestore(&pool
->lock
, flags
);
667 * This section of code contains the logic for processing a thin device's IO.
668 * Much of the code depends on pool object resources (lists, workqueues, etc)
669 * but most is exclusively called from the thin target rather than the thin-pool
673 static dm_block_t
get_bio_block(struct thin_c
*tc
, struct bio
*bio
)
675 return bio
->bi_sector
>> tc
->pool
->block_shift
;
678 static void remap(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
)
680 struct pool
*pool
= tc
->pool
;
682 bio
->bi_bdev
= tc
->pool_dev
->bdev
;
683 bio
->bi_sector
= (block
<< pool
->block_shift
) +
684 (bio
->bi_sector
& pool
->offset_mask
);
687 static void remap_to_origin(struct thin_c
*tc
, struct bio
*bio
)
689 bio
->bi_bdev
= tc
->origin_dev
->bdev
;
692 static void issue(struct thin_c
*tc
, struct bio
*bio
)
694 struct pool
*pool
= tc
->pool
;
698 * Batch together any FUA/FLUSH bios we find and then issue
699 * a single commit for them in process_deferred_bios().
701 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
702 spin_lock_irqsave(&pool
->lock
, flags
);
703 bio_list_add(&pool
->deferred_flush_bios
, bio
);
704 spin_unlock_irqrestore(&pool
->lock
, flags
);
706 generic_make_request(bio
);
709 static void remap_to_origin_and_issue(struct thin_c
*tc
, struct bio
*bio
)
711 remap_to_origin(tc
, bio
);
715 static void remap_and_issue(struct thin_c
*tc
, struct bio
*bio
,
718 remap(tc
, bio
, block
);
723 * wake_worker() is used when new work is queued and when pool_resume is
724 * ready to continue deferred IO processing.
726 static void wake_worker(struct pool
*pool
)
728 queue_work(pool
->wq
, &pool
->worker
);
731 /*----------------------------------------------------------------*/
734 * Bio endio functions.
737 struct list_head list
;
741 unsigned pass_discard
:1;
744 dm_block_t virt_block
;
745 dm_block_t data_block
;
746 struct cell
*cell
, *cell2
;
750 * If the bio covers the whole area of a block then we can avoid
751 * zeroing or copying. Instead this bio is hooked. The bio will
752 * still be in the cell, so care has to be taken to avoid issuing
756 bio_end_io_t
*saved_bi_end_io
;
759 static void __maybe_add_mapping(struct new_mapping
*m
)
761 struct pool
*pool
= m
->tc
->pool
;
763 if (m
->quiesced
&& m
->prepared
) {
764 list_add(&m
->list
, &pool
->prepared_mappings
);
769 static void copy_complete(int read_err
, unsigned long write_err
, void *context
)
772 struct new_mapping
*m
= context
;
773 struct pool
*pool
= m
->tc
->pool
;
775 m
->err
= read_err
|| write_err
? -EIO
: 0;
777 spin_lock_irqsave(&pool
->lock
, flags
);
779 __maybe_add_mapping(m
);
780 spin_unlock_irqrestore(&pool
->lock
, flags
);
783 static void overwrite_endio(struct bio
*bio
, int err
)
786 struct endio_hook
*h
= dm_get_mapinfo(bio
)->ptr
;
787 struct new_mapping
*m
= h
->overwrite_mapping
;
788 struct pool
*pool
= m
->tc
->pool
;
792 spin_lock_irqsave(&pool
->lock
, flags
);
794 __maybe_add_mapping(m
);
795 spin_unlock_irqrestore(&pool
->lock
, flags
);
798 /*----------------------------------------------------------------*/
805 * Prepared mapping jobs.
809 * This sends the bios in the cell back to the deferred_bios list.
811 static void cell_defer(struct thin_c
*tc
, struct cell
*cell
,
812 dm_block_t data_block
)
814 struct pool
*pool
= tc
->pool
;
817 spin_lock_irqsave(&pool
->lock
, flags
);
818 cell_release(cell
, &pool
->deferred_bios
);
819 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
825 * Same as cell_defer above, except it omits one particular detainee,
826 * a write bio that covers the block and has already been processed.
828 static void cell_defer_except(struct thin_c
*tc
, struct cell
*cell
)
830 struct bio_list bios
;
831 struct pool
*pool
= tc
->pool
;
834 bio_list_init(&bios
);
836 spin_lock_irqsave(&pool
->lock
, flags
);
837 cell_release_no_holder(cell
, &pool
->deferred_bios
);
838 spin_unlock_irqrestore(&pool
->lock
, flags
);
843 static void process_prepared_mapping(struct new_mapping
*m
)
845 struct thin_c
*tc
= m
->tc
;
851 bio
->bi_end_io
= m
->saved_bi_end_io
;
859 * Commit the prepared block into the mapping btree.
860 * Any I/O for this block arriving after this point will get
861 * remapped to it directly.
863 r
= dm_thin_insert_block(tc
->td
, m
->virt_block
, m
->data_block
);
865 DMERR("dm_thin_insert_block() failed");
871 * Release any bios held while the block was being provisioned.
872 * If we are processing a write bio that completely covers the block,
873 * we already processed it so can ignore it now when processing
874 * the bios in the cell.
877 cell_defer_except(tc
, m
->cell
);
880 cell_defer(tc
, m
->cell
, m
->data_block
);
883 mempool_free(m
, tc
->pool
->mapping_pool
);
886 static void process_prepared_discard(struct new_mapping
*m
)
889 struct thin_c
*tc
= m
->tc
;
891 r
= dm_thin_remove_block(tc
->td
, m
->virt_block
);
893 DMERR("dm_thin_remove_block() failed");
896 * Pass the discard down to the underlying device?
899 remap_and_issue(tc
, m
->bio
, m
->data_block
);
901 bio_endio(m
->bio
, 0);
903 cell_defer_except(tc
, m
->cell
);
904 cell_defer_except(tc
, m
->cell2
);
905 mempool_free(m
, tc
->pool
->mapping_pool
);
908 static void process_prepared(struct pool
*pool
, struct list_head
*head
,
909 void (*fn
)(struct new_mapping
*))
912 struct list_head maps
;
913 struct new_mapping
*m
, *tmp
;
915 INIT_LIST_HEAD(&maps
);
916 spin_lock_irqsave(&pool
->lock
, flags
);
917 list_splice_init(head
, &maps
);
918 spin_unlock_irqrestore(&pool
->lock
, flags
);
920 list_for_each_entry_safe(m
, tmp
, &maps
, list
)
927 static int io_overlaps_block(struct pool
*pool
, struct bio
*bio
)
929 return !(bio
->bi_sector
& pool
->offset_mask
) &&
930 (bio
->bi_size
== (pool
->sectors_per_block
<< SECTOR_SHIFT
));
934 static int io_overwrites_block(struct pool
*pool
, struct bio
*bio
)
936 return (bio_data_dir(bio
) == WRITE
) &&
937 io_overlaps_block(pool
, bio
);
940 static void save_and_set_endio(struct bio
*bio
, bio_end_io_t
**save
,
943 *save
= bio
->bi_end_io
;
947 static int ensure_next_mapping(struct pool
*pool
)
949 if (pool
->next_mapping
)
952 pool
->next_mapping
= mempool_alloc(pool
->mapping_pool
, GFP_ATOMIC
);
954 return pool
->next_mapping
? 0 : -ENOMEM
;
957 static struct new_mapping
*get_next_mapping(struct pool
*pool
)
959 struct new_mapping
*r
= pool
->next_mapping
;
961 BUG_ON(!pool
->next_mapping
);
963 pool
->next_mapping
= NULL
;
968 static void schedule_copy(struct thin_c
*tc
, dm_block_t virt_block
,
969 struct dm_dev
*origin
, dm_block_t data_origin
,
970 dm_block_t data_dest
,
971 struct cell
*cell
, struct bio
*bio
)
974 struct pool
*pool
= tc
->pool
;
975 struct new_mapping
*m
= get_next_mapping(pool
);
977 INIT_LIST_HEAD(&m
->list
);
981 m
->virt_block
= virt_block
;
982 m
->data_block
= data_dest
;
987 if (!ds_add_work(&pool
->shared_read_ds
, &m
->list
))
991 * IO to pool_dev remaps to the pool target's data_dev.
993 * If the whole block of data is being overwritten, we can issue the
994 * bio immediately. Otherwise we use kcopyd to clone the data first.
996 if (io_overwrites_block(pool
, bio
)) {
997 struct endio_hook
*h
= dm_get_mapinfo(bio
)->ptr
;
998 h
->overwrite_mapping
= m
;
1000 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
1001 remap_and_issue(tc
, bio
, data_dest
);
1003 struct dm_io_region from
, to
;
1005 from
.bdev
= origin
->bdev
;
1006 from
.sector
= data_origin
* pool
->sectors_per_block
;
1007 from
.count
= pool
->sectors_per_block
;
1009 to
.bdev
= tc
->pool_dev
->bdev
;
1010 to
.sector
= data_dest
* pool
->sectors_per_block
;
1011 to
.count
= pool
->sectors_per_block
;
1013 r
= dm_kcopyd_copy(pool
->copier
, &from
, 1, &to
,
1014 0, copy_complete
, m
);
1016 mempool_free(m
, pool
->mapping_pool
);
1017 DMERR("dm_kcopyd_copy() failed");
1023 static void schedule_internal_copy(struct thin_c
*tc
, dm_block_t virt_block
,
1024 dm_block_t data_origin
, dm_block_t data_dest
,
1025 struct cell
*cell
, struct bio
*bio
)
1027 schedule_copy(tc
, virt_block
, tc
->pool_dev
,
1028 data_origin
, data_dest
, cell
, bio
);
1031 static void schedule_external_copy(struct thin_c
*tc
, dm_block_t virt_block
,
1032 dm_block_t data_dest
,
1033 struct cell
*cell
, struct bio
*bio
)
1035 schedule_copy(tc
, virt_block
, tc
->origin_dev
,
1036 virt_block
, data_dest
, cell
, bio
);
1039 static void schedule_zero(struct thin_c
*tc
, dm_block_t virt_block
,
1040 dm_block_t data_block
, struct cell
*cell
,
1043 struct pool
*pool
= tc
->pool
;
1044 struct new_mapping
*m
= get_next_mapping(pool
);
1046 INIT_LIST_HEAD(&m
->list
);
1050 m
->virt_block
= virt_block
;
1051 m
->data_block
= data_block
;
1057 * If the whole block of data is being overwritten or we are not
1058 * zeroing pre-existing data, we can issue the bio immediately.
1059 * Otherwise we use kcopyd to zero the data first.
1061 if (!pool
->pf
.zero_new_blocks
)
1062 process_prepared_mapping(m
);
1064 else if (io_overwrites_block(pool
, bio
)) {
1065 struct endio_hook
*h
= dm_get_mapinfo(bio
)->ptr
;
1066 h
->overwrite_mapping
= m
;
1068 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
1069 remap_and_issue(tc
, bio
, data_block
);
1073 struct dm_io_region to
;
1075 to
.bdev
= tc
->pool_dev
->bdev
;
1076 to
.sector
= data_block
* pool
->sectors_per_block
;
1077 to
.count
= pool
->sectors_per_block
;
1079 r
= dm_kcopyd_zero(pool
->copier
, 1, &to
, 0, copy_complete
, m
);
1081 mempool_free(m
, pool
->mapping_pool
);
1082 DMERR("dm_kcopyd_zero() failed");
1088 static int alloc_data_block(struct thin_c
*tc
, dm_block_t
*result
)
1091 dm_block_t free_blocks
;
1092 unsigned long flags
;
1093 struct pool
*pool
= tc
->pool
;
1095 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
1099 if (free_blocks
<= pool
->low_water_blocks
&& !pool
->low_water_triggered
) {
1100 DMWARN("%s: reached low water mark, sending event.",
1101 dm_device_name(pool
->pool_md
));
1102 spin_lock_irqsave(&pool
->lock
, flags
);
1103 pool
->low_water_triggered
= 1;
1104 spin_unlock_irqrestore(&pool
->lock
, flags
);
1105 dm_table_event(pool
->ti
->table
);
1109 if (pool
->no_free_space
)
1113 * Try to commit to see if that will free up some
1116 r
= dm_pool_commit_metadata(pool
->pmd
);
1118 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
1123 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
1128 * If we still have no space we set a flag to avoid
1129 * doing all this checking and return -ENOSPC.
1132 DMWARN("%s: no free space available.",
1133 dm_device_name(pool
->pool_md
));
1134 spin_lock_irqsave(&pool
->lock
, flags
);
1135 pool
->no_free_space
= 1;
1136 spin_unlock_irqrestore(&pool
->lock
, flags
);
1142 r
= dm_pool_alloc_data_block(pool
->pmd
, result
);
1150 * If we have run out of space, queue bios until the device is
1151 * resumed, presumably after having been reloaded with more space.
1153 static void retry_on_resume(struct bio
*bio
)
1155 struct endio_hook
*h
= dm_get_mapinfo(bio
)->ptr
;
1156 struct thin_c
*tc
= h
->tc
;
1157 struct pool
*pool
= tc
->pool
;
1158 unsigned long flags
;
1160 spin_lock_irqsave(&pool
->lock
, flags
);
1161 bio_list_add(&pool
->retry_on_resume_list
, bio
);
1162 spin_unlock_irqrestore(&pool
->lock
, flags
);
1165 static void no_space(struct cell
*cell
)
1168 struct bio_list bios
;
1170 bio_list_init(&bios
);
1171 cell_release(cell
, &bios
);
1173 while ((bio
= bio_list_pop(&bios
)))
1174 retry_on_resume(bio
);
1177 static void process_discard(struct thin_c
*tc
, struct bio
*bio
)
1180 struct pool
*pool
= tc
->pool
;
1181 struct cell
*cell
, *cell2
;
1182 struct cell_key key
, key2
;
1183 dm_block_t block
= get_bio_block(tc
, bio
);
1184 struct dm_thin_lookup_result lookup_result
;
1185 struct new_mapping
*m
;
1187 build_virtual_key(tc
->td
, block
, &key
);
1188 if (bio_detain(tc
->pool
->prison
, &key
, bio
, &cell
))
1191 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1195 * Check nobody is fiddling with this pool block. This can
1196 * happen if someone's in the process of breaking sharing
1199 build_data_key(tc
->td
, lookup_result
.block
, &key2
);
1200 if (bio_detain(tc
->pool
->prison
, &key2
, bio
, &cell2
)) {
1201 cell_release_singleton(cell
, bio
);
1205 if (io_overlaps_block(pool
, bio
)) {
1207 * IO may still be going to the destination block. We must
1208 * quiesce before we can do the removal.
1210 m
= get_next_mapping(pool
);
1212 m
->pass_discard
= (!lookup_result
.shared
) & pool
->pf
.discard_passdown
;
1213 m
->virt_block
= block
;
1214 m
->data_block
= lookup_result
.block
;
1220 if (!ds_add_work(&pool
->all_io_ds
, &m
->list
)) {
1221 list_add(&m
->list
, &pool
->prepared_discards
);
1226 * This path is hit if people are ignoring
1227 * limits->discard_granularity. It ignores any
1228 * part of the discard that is in a subsequent
1231 sector_t offset
= bio
->bi_sector
- (block
<< pool
->block_shift
);
1232 unsigned remaining
= (pool
->sectors_per_block
- offset
) << 9;
1233 bio
->bi_size
= min(bio
->bi_size
, remaining
);
1235 cell_release_singleton(cell
, bio
);
1236 cell_release_singleton(cell2
, bio
);
1237 remap_and_issue(tc
, bio
, lookup_result
.block
);
1243 * It isn't provisioned, just forget it.
1245 cell_release_singleton(cell
, bio
);
1250 DMERR("discard: find block unexpectedly returned %d", r
);
1251 cell_release_singleton(cell
, bio
);
1257 static void break_sharing(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1258 struct cell_key
*key
,
1259 struct dm_thin_lookup_result
*lookup_result
,
1263 dm_block_t data_block
;
1265 r
= alloc_data_block(tc
, &data_block
);
1268 schedule_internal_copy(tc
, block
, lookup_result
->block
,
1269 data_block
, cell
, bio
);
1277 DMERR("%s: alloc_data_block() failed, error = %d", __func__
, r
);
1283 static void process_shared_bio(struct thin_c
*tc
, struct bio
*bio
,
1285 struct dm_thin_lookup_result
*lookup_result
)
1288 struct pool
*pool
= tc
->pool
;
1289 struct cell_key key
;
1292 * If cell is already occupied, then sharing is already in the process
1293 * of being broken so we have nothing further to do here.
1295 build_data_key(tc
->td
, lookup_result
->block
, &key
);
1296 if (bio_detain(pool
->prison
, &key
, bio
, &cell
))
1299 if (bio_data_dir(bio
) == WRITE
)
1300 break_sharing(tc
, bio
, block
, &key
, lookup_result
, cell
);
1302 struct endio_hook
*h
= dm_get_mapinfo(bio
)->ptr
;
1304 h
->shared_read_entry
= ds_inc(&pool
->shared_read_ds
);
1306 cell_release_singleton(cell
, bio
);
1307 remap_and_issue(tc
, bio
, lookup_result
->block
);
1311 static void provision_block(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1315 dm_block_t data_block
;
1318 * Remap empty bios (flushes) immediately, without provisioning.
1320 if (!bio
->bi_size
) {
1321 cell_release_singleton(cell
, bio
);
1322 remap_and_issue(tc
, bio
, 0);
1327 * Fill read bios with zeroes and complete them immediately.
1329 if (bio_data_dir(bio
) == READ
) {
1331 cell_release_singleton(cell
, bio
);
1336 r
= alloc_data_block(tc
, &data_block
);
1340 schedule_external_copy(tc
, block
, data_block
, cell
, bio
);
1342 schedule_zero(tc
, block
, data_block
, cell
, bio
);
1350 DMERR("%s: alloc_data_block() failed, error = %d", __func__
, r
);
1356 static void process_bio(struct thin_c
*tc
, struct bio
*bio
)
1359 dm_block_t block
= get_bio_block(tc
, bio
);
1361 struct cell_key key
;
1362 struct dm_thin_lookup_result lookup_result
;
1365 * If cell is already occupied, then the block is already
1366 * being provisioned so we have nothing further to do here.
1368 build_virtual_key(tc
->td
, block
, &key
);
1369 if (bio_detain(tc
->pool
->prison
, &key
, bio
, &cell
))
1372 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1376 * We can release this cell now. This thread is the only
1377 * one that puts bios into a cell, and we know there were
1378 * no preceding bios.
1381 * TODO: this will probably have to change when discard goes
1384 cell_release_singleton(cell
, bio
);
1386 if (lookup_result
.shared
)
1387 process_shared_bio(tc
, bio
, block
, &lookup_result
);
1389 remap_and_issue(tc
, bio
, lookup_result
.block
);
1393 if (bio_data_dir(bio
) == READ
&& tc
->origin_dev
) {
1394 cell_release_singleton(cell
, bio
);
1395 remap_to_origin_and_issue(tc
, bio
);
1397 provision_block(tc
, bio
, block
, cell
);
1401 DMERR("dm_thin_find_block() failed, error = %d", r
);
1402 cell_release_singleton(cell
, bio
);
1408 static int need_commit_due_to_time(struct pool
*pool
)
1410 return jiffies
< pool
->last_commit_jiffies
||
1411 jiffies
> pool
->last_commit_jiffies
+ COMMIT_PERIOD
;
1414 static void process_deferred_bios(struct pool
*pool
)
1416 unsigned long flags
;
1418 struct bio_list bios
;
1421 bio_list_init(&bios
);
1423 spin_lock_irqsave(&pool
->lock
, flags
);
1424 bio_list_merge(&bios
, &pool
->deferred_bios
);
1425 bio_list_init(&pool
->deferred_bios
);
1426 spin_unlock_irqrestore(&pool
->lock
, flags
);
1428 while ((bio
= bio_list_pop(&bios
))) {
1429 struct endio_hook
*h
= dm_get_mapinfo(bio
)->ptr
;
1430 struct thin_c
*tc
= h
->tc
;
1433 * If we've got no free new_mapping structs, and processing
1434 * this bio might require one, we pause until there are some
1435 * prepared mappings to process.
1437 if (ensure_next_mapping(pool
)) {
1438 spin_lock_irqsave(&pool
->lock
, flags
);
1439 bio_list_merge(&pool
->deferred_bios
, &bios
);
1440 spin_unlock_irqrestore(&pool
->lock
, flags
);
1445 if (bio
->bi_rw
& REQ_DISCARD
)
1446 process_discard(tc
, bio
);
1448 process_bio(tc
, bio
);
1452 * If there are any deferred flush bios, we must commit
1453 * the metadata before issuing them.
1455 bio_list_init(&bios
);
1456 spin_lock_irqsave(&pool
->lock
, flags
);
1457 bio_list_merge(&bios
, &pool
->deferred_flush_bios
);
1458 bio_list_init(&pool
->deferred_flush_bios
);
1459 spin_unlock_irqrestore(&pool
->lock
, flags
);
1461 if (bio_list_empty(&bios
) && !need_commit_due_to_time(pool
))
1464 r
= dm_pool_commit_metadata(pool
->pmd
);
1466 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
1468 while ((bio
= bio_list_pop(&bios
)))
1472 pool
->last_commit_jiffies
= jiffies
;
1474 while ((bio
= bio_list_pop(&bios
)))
1475 generic_make_request(bio
);
1478 static void do_worker(struct work_struct
*ws
)
1480 struct pool
*pool
= container_of(ws
, struct pool
, worker
);
1482 process_prepared(pool
, &pool
->prepared_mappings
, process_prepared_mapping
);
1483 process_prepared(pool
, &pool
->prepared_discards
, process_prepared_discard
);
1484 process_deferred_bios(pool
);
1488 * We want to commit periodically so that not too much
1489 * unwritten data builds up.
1491 static void do_waker(struct work_struct
*ws
)
1493 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
, waker
);
1495 queue_delayed_work(pool
->wq
, &pool
->waker
, COMMIT_PERIOD
);
1498 /*----------------------------------------------------------------*/
1501 * Mapping functions.
1505 * Called only while mapping a thin bio to hand it over to the workqueue.
1507 static void thin_defer_bio(struct thin_c
*tc
, struct bio
*bio
)
1509 unsigned long flags
;
1510 struct pool
*pool
= tc
->pool
;
1512 spin_lock_irqsave(&pool
->lock
, flags
);
1513 bio_list_add(&pool
->deferred_bios
, bio
);
1514 spin_unlock_irqrestore(&pool
->lock
, flags
);
1519 static struct endio_hook
*thin_hook_bio(struct thin_c
*tc
, struct bio
*bio
)
1521 struct pool
*pool
= tc
->pool
;
1522 struct endio_hook
*h
= mempool_alloc(pool
->endio_hook_pool
, GFP_NOIO
);
1525 h
->shared_read_entry
= NULL
;
1526 h
->all_io_entry
= bio
->bi_rw
& REQ_DISCARD
? NULL
: ds_inc(&pool
->all_io_ds
);
1527 h
->overwrite_mapping
= NULL
;
1533 * Non-blocking function called from the thin target's map function.
1535 static int thin_bio_map(struct dm_target
*ti
, struct bio
*bio
,
1536 union map_info
*map_context
)
1539 struct thin_c
*tc
= ti
->private;
1540 dm_block_t block
= get_bio_block(tc
, bio
);
1541 struct dm_thin_device
*td
= tc
->td
;
1542 struct dm_thin_lookup_result result
;
1544 map_context
->ptr
= thin_hook_bio(tc
, bio
);
1545 if (bio
->bi_rw
& (REQ_DISCARD
| REQ_FLUSH
| REQ_FUA
)) {
1546 thin_defer_bio(tc
, bio
);
1547 return DM_MAPIO_SUBMITTED
;
1550 r
= dm_thin_find_block(td
, block
, 0, &result
);
1553 * Note that we defer readahead too.
1557 if (unlikely(result
.shared
)) {
1559 * We have a race condition here between the
1560 * result.shared value returned by the lookup and
1561 * snapshot creation, which may cause new
1564 * To avoid this always quiesce the origin before
1565 * taking the snap. You want to do this anyway to
1566 * ensure a consistent application view
1569 * More distant ancestors are irrelevant. The
1570 * shared flag will be set in their case.
1572 thin_defer_bio(tc
, bio
);
1573 r
= DM_MAPIO_SUBMITTED
;
1575 remap(tc
, bio
, result
.block
);
1576 r
= DM_MAPIO_REMAPPED
;
1582 * In future, the failed dm_thin_find_block above could
1583 * provide the hint to load the metadata into cache.
1586 thin_defer_bio(tc
, bio
);
1587 r
= DM_MAPIO_SUBMITTED
;
1594 static int pool_is_congested(struct dm_target_callbacks
*cb
, int bdi_bits
)
1597 unsigned long flags
;
1598 struct pool_c
*pt
= container_of(cb
, struct pool_c
, callbacks
);
1600 spin_lock_irqsave(&pt
->pool
->lock
, flags
);
1601 r
= !bio_list_empty(&pt
->pool
->retry_on_resume_list
);
1602 spin_unlock_irqrestore(&pt
->pool
->lock
, flags
);
1605 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1606 r
= bdi_congested(&q
->backing_dev_info
, bdi_bits
);
1612 static void __requeue_bios(struct pool
*pool
)
1614 bio_list_merge(&pool
->deferred_bios
, &pool
->retry_on_resume_list
);
1615 bio_list_init(&pool
->retry_on_resume_list
);
1618 /*----------------------------------------------------------------
1619 * Binding of control targets to a pool object
1620 *--------------------------------------------------------------*/
1621 static int bind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1623 struct pool_c
*pt
= ti
->private;
1626 pool
->low_water_blocks
= pt
->low_water_blocks
;
1632 static void unbind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1638 /*----------------------------------------------------------------
1640 *--------------------------------------------------------------*/
1641 /* Initialize pool features. */
1642 static void pool_features_init(struct pool_features
*pf
)
1644 pf
->zero_new_blocks
= 1;
1645 pf
->discard_enabled
= 1;
1646 pf
->discard_passdown
= 1;
1649 static void __pool_destroy(struct pool
*pool
)
1651 __pool_table_remove(pool
);
1653 if (dm_pool_metadata_close(pool
->pmd
) < 0)
1654 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1656 prison_destroy(pool
->prison
);
1657 dm_kcopyd_client_destroy(pool
->copier
);
1660 destroy_workqueue(pool
->wq
);
1662 if (pool
->next_mapping
)
1663 mempool_free(pool
->next_mapping
, pool
->mapping_pool
);
1664 mempool_destroy(pool
->mapping_pool
);
1665 mempool_destroy(pool
->endio_hook_pool
);
1669 static struct pool
*pool_create(struct mapped_device
*pool_md
,
1670 struct block_device
*metadata_dev
,
1671 unsigned long block_size
, char **error
)
1676 struct dm_pool_metadata
*pmd
;
1678 pmd
= dm_pool_metadata_open(metadata_dev
, block_size
);
1680 *error
= "Error creating metadata object";
1681 return (struct pool
*)pmd
;
1684 pool
= kmalloc(sizeof(*pool
), GFP_KERNEL
);
1686 *error
= "Error allocating memory for pool";
1687 err_p
= ERR_PTR(-ENOMEM
);
1692 pool
->sectors_per_block
= block_size
;
1693 pool
->block_shift
= ffs(block_size
) - 1;
1694 pool
->offset_mask
= block_size
- 1;
1695 pool
->low_water_blocks
= 0;
1696 pool_features_init(&pool
->pf
);
1697 pool
->prison
= prison_create(PRISON_CELLS
);
1698 if (!pool
->prison
) {
1699 *error
= "Error creating pool's bio prison";
1700 err_p
= ERR_PTR(-ENOMEM
);
1704 pool
->copier
= dm_kcopyd_client_create();
1705 if (IS_ERR(pool
->copier
)) {
1706 r
= PTR_ERR(pool
->copier
);
1707 *error
= "Error creating pool's kcopyd client";
1709 goto bad_kcopyd_client
;
1713 * Create singlethreaded workqueue that will service all devices
1714 * that use this metadata.
1716 pool
->wq
= alloc_ordered_workqueue("dm-" DM_MSG_PREFIX
, WQ_MEM_RECLAIM
);
1718 *error
= "Error creating pool's workqueue";
1719 err_p
= ERR_PTR(-ENOMEM
);
1723 INIT_WORK(&pool
->worker
, do_worker
);
1724 INIT_DELAYED_WORK(&pool
->waker
, do_waker
);
1725 spin_lock_init(&pool
->lock
);
1726 bio_list_init(&pool
->deferred_bios
);
1727 bio_list_init(&pool
->deferred_flush_bios
);
1728 INIT_LIST_HEAD(&pool
->prepared_mappings
);
1729 INIT_LIST_HEAD(&pool
->prepared_discards
);
1730 pool
->low_water_triggered
= 0;
1731 pool
->no_free_space
= 0;
1732 bio_list_init(&pool
->retry_on_resume_list
);
1733 ds_init(&pool
->shared_read_ds
);
1734 ds_init(&pool
->all_io_ds
);
1736 pool
->next_mapping
= NULL
;
1737 pool
->mapping_pool
=
1738 mempool_create_kmalloc_pool(MAPPING_POOL_SIZE
, sizeof(struct new_mapping
));
1739 if (!pool
->mapping_pool
) {
1740 *error
= "Error creating pool's mapping mempool";
1741 err_p
= ERR_PTR(-ENOMEM
);
1742 goto bad_mapping_pool
;
1745 pool
->endio_hook_pool
=
1746 mempool_create_kmalloc_pool(ENDIO_HOOK_POOL_SIZE
, sizeof(struct endio_hook
));
1747 if (!pool
->endio_hook_pool
) {
1748 *error
= "Error creating pool's endio_hook mempool";
1749 err_p
= ERR_PTR(-ENOMEM
);
1750 goto bad_endio_hook_pool
;
1752 pool
->ref_count
= 1;
1753 pool
->last_commit_jiffies
= jiffies
;
1754 pool
->pool_md
= pool_md
;
1755 pool
->md_dev
= metadata_dev
;
1756 __pool_table_insert(pool
);
1760 bad_endio_hook_pool
:
1761 mempool_destroy(pool
->mapping_pool
);
1763 destroy_workqueue(pool
->wq
);
1765 dm_kcopyd_client_destroy(pool
->copier
);
1767 prison_destroy(pool
->prison
);
1771 if (dm_pool_metadata_close(pmd
))
1772 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1777 static void __pool_inc(struct pool
*pool
)
1779 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
1783 static void __pool_dec(struct pool
*pool
)
1785 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
1786 BUG_ON(!pool
->ref_count
);
1787 if (!--pool
->ref_count
)
1788 __pool_destroy(pool
);
1791 static struct pool
*__pool_find(struct mapped_device
*pool_md
,
1792 struct block_device
*metadata_dev
,
1793 unsigned long block_size
, char **error
,
1796 struct pool
*pool
= __pool_table_lookup_metadata_dev(metadata_dev
);
1799 if (pool
->pool_md
!= pool_md
)
1800 return ERR_PTR(-EBUSY
);
1804 pool
= __pool_table_lookup(pool_md
);
1806 if (pool
->md_dev
!= metadata_dev
)
1807 return ERR_PTR(-EINVAL
);
1811 pool
= pool_create(pool_md
, metadata_dev
, block_size
, error
);
1819 /*----------------------------------------------------------------
1820 * Pool target methods
1821 *--------------------------------------------------------------*/
1822 static void pool_dtr(struct dm_target
*ti
)
1824 struct pool_c
*pt
= ti
->private;
1826 mutex_lock(&dm_thin_pool_table
.mutex
);
1828 unbind_control_target(pt
->pool
, ti
);
1829 __pool_dec(pt
->pool
);
1830 dm_put_device(ti
, pt
->metadata_dev
);
1831 dm_put_device(ti
, pt
->data_dev
);
1834 mutex_unlock(&dm_thin_pool_table
.mutex
);
1837 static int parse_pool_features(struct dm_arg_set
*as
, struct pool_features
*pf
,
1838 struct dm_target
*ti
)
1842 const char *arg_name
;
1844 static struct dm_arg _args
[] = {
1845 {0, 3, "Invalid number of pool feature arguments"},
1849 * No feature arguments supplied.
1854 r
= dm_read_arg_group(_args
, as
, &argc
, &ti
->error
);
1858 while (argc
&& !r
) {
1859 arg_name
= dm_shift_arg(as
);
1862 if (!strcasecmp(arg_name
, "skip_block_zeroing")) {
1863 pf
->zero_new_blocks
= 0;
1865 } else if (!strcasecmp(arg_name
, "ignore_discard")) {
1866 pf
->discard_enabled
= 0;
1868 } else if (!strcasecmp(arg_name
, "no_discard_passdown")) {
1869 pf
->discard_passdown
= 0;
1873 ti
->error
= "Unrecognised pool feature requested";
1881 * thin-pool <metadata dev> <data dev>
1882 * <data block size (sectors)>
1883 * <low water mark (blocks)>
1884 * [<#feature args> [<arg>]*]
1886 * Optional feature arguments are:
1887 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
1888 * ignore_discard: disable discard
1889 * no_discard_passdown: don't pass discards down to the data device
1891 static int pool_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
1893 int r
, pool_created
= 0;
1896 struct pool_features pf
;
1897 struct dm_arg_set as
;
1898 struct dm_dev
*data_dev
;
1899 unsigned long block_size
;
1900 dm_block_t low_water_blocks
;
1901 struct dm_dev
*metadata_dev
;
1902 sector_t metadata_dev_size
;
1903 char b
[BDEVNAME_SIZE
];
1906 * FIXME Remove validation from scope of lock.
1908 mutex_lock(&dm_thin_pool_table
.mutex
);
1911 ti
->error
= "Invalid argument count";
1918 r
= dm_get_device(ti
, argv
[0], FMODE_READ
| FMODE_WRITE
, &metadata_dev
);
1920 ti
->error
= "Error opening metadata block device";
1924 metadata_dev_size
= i_size_read(metadata_dev
->bdev
->bd_inode
) >> SECTOR_SHIFT
;
1925 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS_WARNING
)
1926 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
1927 bdevname(metadata_dev
->bdev
, b
), THIN_METADATA_MAX_SECTORS
);
1929 r
= dm_get_device(ti
, argv
[1], FMODE_READ
| FMODE_WRITE
, &data_dev
);
1931 ti
->error
= "Error getting data device";
1935 if (kstrtoul(argv
[2], 10, &block_size
) || !block_size
||
1936 block_size
< DATA_DEV_BLOCK_SIZE_MIN_SECTORS
||
1937 block_size
> DATA_DEV_BLOCK_SIZE_MAX_SECTORS
||
1938 !is_power_of_2(block_size
)) {
1939 ti
->error
= "Invalid block size";
1944 if (kstrtoull(argv
[3], 10, (unsigned long long *)&low_water_blocks
)) {
1945 ti
->error
= "Invalid low water mark";
1951 * Set default pool features.
1953 pool_features_init(&pf
);
1955 dm_consume_args(&as
, 4);
1956 r
= parse_pool_features(&as
, &pf
, ti
);
1960 pt
= kzalloc(sizeof(*pt
), GFP_KERNEL
);
1966 pool
= __pool_find(dm_table_get_md(ti
->table
), metadata_dev
->bdev
,
1967 block_size
, &ti
->error
, &pool_created
);
1974 * 'pool_created' reflects whether this is the first table load.
1975 * Top level discard support is not allowed to be changed after
1976 * initial load. This would require a pool reload to trigger thin
1979 if (!pool_created
&& pf
.discard_enabled
!= pool
->pf
.discard_enabled
) {
1980 ti
->error
= "Discard support cannot be disabled once enabled";
1982 goto out_flags_changed
;
1986 * If discard_passdown was enabled verify that the data device
1987 * supports discards. Disable discard_passdown if not; otherwise
1988 * -EOPNOTSUPP will be returned.
1990 if (pf
.discard_passdown
) {
1991 struct request_queue
*q
= bdev_get_queue(data_dev
->bdev
);
1992 if (!q
|| !blk_queue_discard(q
)) {
1993 DMWARN("Discard unsupported by data device: Disabling discard passdown.");
1994 pf
.discard_passdown
= 0;
2000 pt
->metadata_dev
= metadata_dev
;
2001 pt
->data_dev
= data_dev
;
2002 pt
->low_water_blocks
= low_water_blocks
;
2004 ti
->num_flush_requests
= 1;
2006 * Only need to enable discards if the pool should pass
2007 * them down to the data device. The thin device's discard
2008 * processing will cause mappings to be removed from the btree.
2010 if (pf
.discard_enabled
&& pf
.discard_passdown
) {
2011 ti
->num_discard_requests
= 1;
2013 * Setting 'discards_supported' circumvents the normal
2014 * stacking of discard limits (this keeps the pool and
2015 * thin devices' discard limits consistent).
2017 ti
->discards_supported
= 1;
2021 pt
->callbacks
.congested_fn
= pool_is_congested
;
2022 dm_table_add_target_callbacks(ti
->table
, &pt
->callbacks
);
2024 mutex_unlock(&dm_thin_pool_table
.mutex
);
2033 dm_put_device(ti
, data_dev
);
2035 dm_put_device(ti
, metadata_dev
);
2037 mutex_unlock(&dm_thin_pool_table
.mutex
);
2042 static int pool_map(struct dm_target
*ti
, struct bio
*bio
,
2043 union map_info
*map_context
)
2046 struct pool_c
*pt
= ti
->private;
2047 struct pool
*pool
= pt
->pool
;
2048 unsigned long flags
;
2051 * As this is a singleton target, ti->begin is always zero.
2053 spin_lock_irqsave(&pool
->lock
, flags
);
2054 bio
->bi_bdev
= pt
->data_dev
->bdev
;
2055 r
= DM_MAPIO_REMAPPED
;
2056 spin_unlock_irqrestore(&pool
->lock
, flags
);
2062 * Retrieves the number of blocks of the data device from
2063 * the superblock and compares it to the actual device size,
2064 * thus resizing the data device in case it has grown.
2066 * This both copes with opening preallocated data devices in the ctr
2067 * being followed by a resume
2069 * calling the resume method individually after userspace has
2070 * grown the data device in reaction to a table event.
2072 static int pool_preresume(struct dm_target
*ti
)
2075 struct pool_c
*pt
= ti
->private;
2076 struct pool
*pool
= pt
->pool
;
2077 dm_block_t data_size
, sb_data_size
;
2080 * Take control of the pool object.
2082 r
= bind_control_target(pool
, ti
);
2086 data_size
= ti
->len
>> pool
->block_shift
;
2087 r
= dm_pool_get_data_dev_size(pool
->pmd
, &sb_data_size
);
2089 DMERR("failed to retrieve data device size");
2093 if (data_size
< sb_data_size
) {
2094 DMERR("pool target too small, is %llu blocks (expected %llu)",
2095 data_size
, sb_data_size
);
2098 } else if (data_size
> sb_data_size
) {
2099 r
= dm_pool_resize_data_dev(pool
->pmd
, data_size
);
2101 DMERR("failed to resize data device");
2105 r
= dm_pool_commit_metadata(pool
->pmd
);
2107 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
2116 static void pool_resume(struct dm_target
*ti
)
2118 struct pool_c
*pt
= ti
->private;
2119 struct pool
*pool
= pt
->pool
;
2120 unsigned long flags
;
2122 spin_lock_irqsave(&pool
->lock
, flags
);
2123 pool
->low_water_triggered
= 0;
2124 pool
->no_free_space
= 0;
2125 __requeue_bios(pool
);
2126 spin_unlock_irqrestore(&pool
->lock
, flags
);
2128 do_waker(&pool
->waker
.work
);
2131 static void pool_postsuspend(struct dm_target
*ti
)
2134 struct pool_c
*pt
= ti
->private;
2135 struct pool
*pool
= pt
->pool
;
2137 cancel_delayed_work(&pool
->waker
);
2138 flush_workqueue(pool
->wq
);
2140 r
= dm_pool_commit_metadata(pool
->pmd
);
2142 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
2144 /* FIXME: invalidate device? error the next FUA or FLUSH bio ?*/
2148 static int check_arg_count(unsigned argc
, unsigned args_required
)
2150 if (argc
!= args_required
) {
2151 DMWARN("Message received with %u arguments instead of %u.",
2152 argc
, args_required
);
2159 static int read_dev_id(char *arg
, dm_thin_id
*dev_id
, int warning
)
2161 if (!kstrtoull(arg
, 10, (unsigned long long *)dev_id
) &&
2162 *dev_id
<= MAX_DEV_ID
)
2166 DMWARN("Message received with invalid device id: %s", arg
);
2171 static int process_create_thin_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2176 r
= check_arg_count(argc
, 2);
2180 r
= read_dev_id(argv
[1], &dev_id
, 1);
2184 r
= dm_pool_create_thin(pool
->pmd
, dev_id
);
2186 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2194 static int process_create_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2197 dm_thin_id origin_dev_id
;
2200 r
= check_arg_count(argc
, 3);
2204 r
= read_dev_id(argv
[1], &dev_id
, 1);
2208 r
= read_dev_id(argv
[2], &origin_dev_id
, 1);
2212 r
= dm_pool_create_snap(pool
->pmd
, dev_id
, origin_dev_id
);
2214 DMWARN("Creation of new snapshot %s of device %s failed.",
2222 static int process_delete_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2227 r
= check_arg_count(argc
, 2);
2231 r
= read_dev_id(argv
[1], &dev_id
, 1);
2235 r
= dm_pool_delete_thin_device(pool
->pmd
, dev_id
);
2237 DMWARN("Deletion of thin device %s failed.", argv
[1]);
2242 static int process_set_transaction_id_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2244 dm_thin_id old_id
, new_id
;
2247 r
= check_arg_count(argc
, 3);
2251 if (kstrtoull(argv
[1], 10, (unsigned long long *)&old_id
)) {
2252 DMWARN("set_transaction_id message: Unrecognised id %s.", argv
[1]);
2256 if (kstrtoull(argv
[2], 10, (unsigned long long *)&new_id
)) {
2257 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv
[2]);
2261 r
= dm_pool_set_metadata_transaction_id(pool
->pmd
, old_id
, new_id
);
2263 DMWARN("Failed to change transaction id from %s to %s.",
2272 * Messages supported:
2273 * create_thin <dev_id>
2274 * create_snap <dev_id> <origin_id>
2276 * trim <dev_id> <new_size_in_sectors>
2277 * set_transaction_id <current_trans_id> <new_trans_id>
2279 static int pool_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
2282 struct pool_c
*pt
= ti
->private;
2283 struct pool
*pool
= pt
->pool
;
2285 if (!strcasecmp(argv
[0], "create_thin"))
2286 r
= process_create_thin_mesg(argc
, argv
, pool
);
2288 else if (!strcasecmp(argv
[0], "create_snap"))
2289 r
= process_create_snap_mesg(argc
, argv
, pool
);
2291 else if (!strcasecmp(argv
[0], "delete"))
2292 r
= process_delete_mesg(argc
, argv
, pool
);
2294 else if (!strcasecmp(argv
[0], "set_transaction_id"))
2295 r
= process_set_transaction_id_mesg(argc
, argv
, pool
);
2298 DMWARN("Unrecognised thin pool target message received: %s", argv
[0]);
2301 r
= dm_pool_commit_metadata(pool
->pmd
);
2303 DMERR("%s message: dm_pool_commit_metadata() failed, error = %d",
2312 * <transaction id> <used metadata sectors>/<total metadata sectors>
2313 * <used data sectors>/<total data sectors> <held metadata root>
2315 static int pool_status(struct dm_target
*ti
, status_type_t type
,
2316 char *result
, unsigned maxlen
)
2320 uint64_t transaction_id
;
2321 dm_block_t nr_free_blocks_data
;
2322 dm_block_t nr_free_blocks_metadata
;
2323 dm_block_t nr_blocks_data
;
2324 dm_block_t nr_blocks_metadata
;
2325 dm_block_t held_root
;
2326 char buf
[BDEVNAME_SIZE
];
2327 char buf2
[BDEVNAME_SIZE
];
2328 struct pool_c
*pt
= ti
->private;
2329 struct pool
*pool
= pt
->pool
;
2332 case STATUSTYPE_INFO
:
2333 r
= dm_pool_get_metadata_transaction_id(pool
->pmd
,
2338 r
= dm_pool_get_free_metadata_block_count(pool
->pmd
,
2339 &nr_free_blocks_metadata
);
2343 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &nr_blocks_metadata
);
2347 r
= dm_pool_get_free_block_count(pool
->pmd
,
2348 &nr_free_blocks_data
);
2352 r
= dm_pool_get_data_dev_size(pool
->pmd
, &nr_blocks_data
);
2356 r
= dm_pool_get_held_metadata_root(pool
->pmd
, &held_root
);
2360 DMEMIT("%llu %llu/%llu %llu/%llu ",
2361 (unsigned long long)transaction_id
,
2362 (unsigned long long)(nr_blocks_metadata
- nr_free_blocks_metadata
),
2363 (unsigned long long)nr_blocks_metadata
,
2364 (unsigned long long)(nr_blocks_data
- nr_free_blocks_data
),
2365 (unsigned long long)nr_blocks_data
);
2368 DMEMIT("%llu", held_root
);
2374 case STATUSTYPE_TABLE
:
2375 DMEMIT("%s %s %lu %llu ",
2376 format_dev_t(buf
, pt
->metadata_dev
->bdev
->bd_dev
),
2377 format_dev_t(buf2
, pt
->data_dev
->bdev
->bd_dev
),
2378 (unsigned long)pool
->sectors_per_block
,
2379 (unsigned long long)pt
->low_water_blocks
);
2381 count
= !pool
->pf
.zero_new_blocks
+ !pool
->pf
.discard_enabled
+
2382 !pool
->pf
.discard_passdown
;
2383 DMEMIT("%u ", count
);
2385 if (!pool
->pf
.zero_new_blocks
)
2386 DMEMIT("skip_block_zeroing ");
2388 if (!pool
->pf
.discard_enabled
)
2389 DMEMIT("ignore_discard ");
2391 if (!pool
->pf
.discard_passdown
)
2392 DMEMIT("no_discard_passdown ");
2400 static int pool_iterate_devices(struct dm_target
*ti
,
2401 iterate_devices_callout_fn fn
, void *data
)
2403 struct pool_c
*pt
= ti
->private;
2405 return fn(ti
, pt
->data_dev
, 0, ti
->len
, data
);
2408 static int pool_merge(struct dm_target
*ti
, struct bvec_merge_data
*bvm
,
2409 struct bio_vec
*biovec
, int max_size
)
2411 struct pool_c
*pt
= ti
->private;
2412 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
2414 if (!q
->merge_bvec_fn
)
2417 bvm
->bi_bdev
= pt
->data_dev
->bdev
;
2419 return min(max_size
, q
->merge_bvec_fn(q
, bvm
, biovec
));
2422 static void set_discard_limits(struct pool
*pool
, struct queue_limits
*limits
)
2425 * FIXME: these limits may be incompatible with the pool's data device
2427 limits
->max_discard_sectors
= pool
->sectors_per_block
;
2430 * This is just a hint, and not enforced. We have to cope with
2431 * bios that overlap 2 blocks.
2433 limits
->discard_granularity
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
2434 limits
->discard_zeroes_data
= pool
->pf
.zero_new_blocks
;
2437 static void pool_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
2439 struct pool_c
*pt
= ti
->private;
2440 struct pool
*pool
= pt
->pool
;
2442 blk_limits_io_min(limits
, 0);
2443 blk_limits_io_opt(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
2444 if (pool
->pf
.discard_enabled
)
2445 set_discard_limits(pool
, limits
);
2448 static struct target_type pool_target
= {
2449 .name
= "thin-pool",
2450 .features
= DM_TARGET_SINGLETON
| DM_TARGET_ALWAYS_WRITEABLE
|
2451 DM_TARGET_IMMUTABLE
,
2452 .version
= {1, 1, 0},
2453 .module
= THIS_MODULE
,
2457 .postsuspend
= pool_postsuspend
,
2458 .preresume
= pool_preresume
,
2459 .resume
= pool_resume
,
2460 .message
= pool_message
,
2461 .status
= pool_status
,
2462 .merge
= pool_merge
,
2463 .iterate_devices
= pool_iterate_devices
,
2464 .io_hints
= pool_io_hints
,
2467 /*----------------------------------------------------------------
2468 * Thin target methods
2469 *--------------------------------------------------------------*/
2470 static void thin_dtr(struct dm_target
*ti
)
2472 struct thin_c
*tc
= ti
->private;
2474 mutex_lock(&dm_thin_pool_table
.mutex
);
2476 __pool_dec(tc
->pool
);
2477 dm_pool_close_thin_device(tc
->td
);
2478 dm_put_device(ti
, tc
->pool_dev
);
2480 dm_put_device(ti
, tc
->origin_dev
);
2483 mutex_unlock(&dm_thin_pool_table
.mutex
);
2487 * Thin target parameters:
2489 * <pool_dev> <dev_id> [origin_dev]
2491 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2492 * dev_id: the internal device identifier
2493 * origin_dev: a device external to the pool that should act as the origin
2495 * If the pool device has discards disabled, they get disabled for the thin
2498 static int thin_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2502 struct dm_dev
*pool_dev
, *origin_dev
;
2503 struct mapped_device
*pool_md
;
2505 mutex_lock(&dm_thin_pool_table
.mutex
);
2507 if (argc
!= 2 && argc
!= 3) {
2508 ti
->error
= "Invalid argument count";
2513 tc
= ti
->private = kzalloc(sizeof(*tc
), GFP_KERNEL
);
2515 ti
->error
= "Out of memory";
2521 r
= dm_get_device(ti
, argv
[2], FMODE_READ
, &origin_dev
);
2523 ti
->error
= "Error opening origin device";
2524 goto bad_origin_dev
;
2526 tc
->origin_dev
= origin_dev
;
2529 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &pool_dev
);
2531 ti
->error
= "Error opening pool device";
2534 tc
->pool_dev
= pool_dev
;
2536 if (read_dev_id(argv
[1], (unsigned long long *)&tc
->dev_id
, 0)) {
2537 ti
->error
= "Invalid device id";
2542 pool_md
= dm_get_md(tc
->pool_dev
->bdev
->bd_dev
);
2544 ti
->error
= "Couldn't get pool mapped device";
2549 tc
->pool
= __pool_table_lookup(pool_md
);
2551 ti
->error
= "Couldn't find pool object";
2553 goto bad_pool_lookup
;
2555 __pool_inc(tc
->pool
);
2557 r
= dm_pool_open_thin_device(tc
->pool
->pmd
, tc
->dev_id
, &tc
->td
);
2559 ti
->error
= "Couldn't open thin internal device";
2563 ti
->split_io
= tc
->pool
->sectors_per_block
;
2564 ti
->num_flush_requests
= 1;
2566 /* In case the pool supports discards, pass them on. */
2567 if (tc
->pool
->pf
.discard_enabled
) {
2568 ti
->discards_supported
= 1;
2569 ti
->num_discard_requests
= 1;
2574 mutex_unlock(&dm_thin_pool_table
.mutex
);
2579 __pool_dec(tc
->pool
);
2583 dm_put_device(ti
, tc
->pool_dev
);
2586 dm_put_device(ti
, tc
->origin_dev
);
2590 mutex_unlock(&dm_thin_pool_table
.mutex
);
2595 static int thin_map(struct dm_target
*ti
, struct bio
*bio
,
2596 union map_info
*map_context
)
2598 bio
->bi_sector
= dm_target_offset(ti
, bio
->bi_sector
);
2600 return thin_bio_map(ti
, bio
, map_context
);
2603 static int thin_endio(struct dm_target
*ti
,
2604 struct bio
*bio
, int err
,
2605 union map_info
*map_context
)
2607 unsigned long flags
;
2608 struct endio_hook
*h
= map_context
->ptr
;
2609 struct list_head work
;
2610 struct new_mapping
*m
, *tmp
;
2611 struct pool
*pool
= h
->tc
->pool
;
2613 if (h
->shared_read_entry
) {
2614 INIT_LIST_HEAD(&work
);
2615 ds_dec(h
->shared_read_entry
, &work
);
2617 spin_lock_irqsave(&pool
->lock
, flags
);
2618 list_for_each_entry_safe(m
, tmp
, &work
, list
) {
2621 __maybe_add_mapping(m
);
2623 spin_unlock_irqrestore(&pool
->lock
, flags
);
2626 if (h
->all_io_entry
) {
2627 INIT_LIST_HEAD(&work
);
2628 ds_dec(h
->all_io_entry
, &work
);
2629 list_for_each_entry_safe(m
, tmp
, &work
, list
)
2630 list_add(&m
->list
, &pool
->prepared_discards
);
2633 mempool_free(h
, pool
->endio_hook_pool
);
2638 static void thin_postsuspend(struct dm_target
*ti
)
2640 if (dm_noflush_suspending(ti
))
2641 requeue_io((struct thin_c
*)ti
->private);
2645 * <nr mapped sectors> <highest mapped sector>
2647 static int thin_status(struct dm_target
*ti
, status_type_t type
,
2648 char *result
, unsigned maxlen
)
2652 dm_block_t mapped
, highest
;
2653 char buf
[BDEVNAME_SIZE
];
2654 struct thin_c
*tc
= ti
->private;
2660 case STATUSTYPE_INFO
:
2661 r
= dm_thin_get_mapped_count(tc
->td
, &mapped
);
2665 r
= dm_thin_get_highest_mapped_block(tc
->td
, &highest
);
2669 DMEMIT("%llu ", mapped
* tc
->pool
->sectors_per_block
);
2671 DMEMIT("%llu", ((highest
+ 1) *
2672 tc
->pool
->sectors_per_block
) - 1);
2677 case STATUSTYPE_TABLE
:
2679 format_dev_t(buf
, tc
->pool_dev
->bdev
->bd_dev
),
2680 (unsigned long) tc
->dev_id
);
2682 DMEMIT(" %s", format_dev_t(buf
, tc
->origin_dev
->bdev
->bd_dev
));
2690 static int thin_iterate_devices(struct dm_target
*ti
,
2691 iterate_devices_callout_fn fn
, void *data
)
2694 struct thin_c
*tc
= ti
->private;
2697 * We can't call dm_pool_get_data_dev_size() since that blocks. So
2698 * we follow a more convoluted path through to the pool's target.
2701 return 0; /* nothing is bound */
2703 blocks
= tc
->pool
->ti
->len
>> tc
->pool
->block_shift
;
2705 return fn(ti
, tc
->pool_dev
, 0, tc
->pool
->sectors_per_block
* blocks
, data
);
2710 static void thin_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
2712 struct thin_c
*tc
= ti
->private;
2713 struct pool
*pool
= tc
->pool
;
2715 blk_limits_io_min(limits
, 0);
2716 blk_limits_io_opt(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
2717 set_discard_limits(pool
, limits
);
2720 static struct target_type thin_target
= {
2722 .version
= {1, 1, 0},
2723 .module
= THIS_MODULE
,
2727 .end_io
= thin_endio
,
2728 .postsuspend
= thin_postsuspend
,
2729 .status
= thin_status
,
2730 .iterate_devices
= thin_iterate_devices
,
2731 .io_hints
= thin_io_hints
,
2734 /*----------------------------------------------------------------*/
2736 static int __init
dm_thin_init(void)
2742 r
= dm_register_target(&thin_target
);
2746 r
= dm_register_target(&pool_target
);
2748 dm_unregister_target(&thin_target
);
2753 static void dm_thin_exit(void)
2755 dm_unregister_target(&thin_target
);
2756 dm_unregister_target(&pool_target
);
2759 module_init(dm_thin_init
);
2760 module_exit(dm_thin_exit
);
2762 MODULE_DESCRIPTION(DM_NAME
"device-mapper thin provisioning target");
2763 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2764 MODULE_LICENSE("GPL");