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 list_head list
;
494 struct dm_target
*ti
; /* Only set if a pool target is bound */
496 struct mapped_device
*pool_md
;
497 struct block_device
*md_dev
;
498 struct dm_pool_metadata
*pmd
;
500 uint32_t sectors_per_block
;
501 unsigned block_shift
;
502 dm_block_t offset_mask
;
503 dm_block_t low_water_blocks
;
505 unsigned zero_new_blocks
:1;
506 unsigned low_water_triggered
:1; /* A dm event has been sent */
507 unsigned no_free_space
:1; /* A -ENOSPC warning has been issued */
509 struct bio_prison
*prison
;
510 struct dm_kcopyd_client
*copier
;
512 struct workqueue_struct
*wq
;
513 struct work_struct worker
;
514 struct delayed_work waker
;
517 unsigned long last_commit_jiffies
;
520 struct bio_list deferred_bios
;
521 struct bio_list deferred_flush_bios
;
522 struct list_head prepared_mappings
;
524 struct bio_list retry_on_resume_list
;
526 struct deferred_set ds
; /* FIXME: move to thin_c */
528 struct new_mapping
*next_mapping
;
529 mempool_t
*mapping_pool
;
530 mempool_t
*endio_hook_pool
;
534 * Target context for a pool.
537 struct dm_target
*ti
;
539 struct dm_dev
*data_dev
;
540 struct dm_dev
*metadata_dev
;
541 struct dm_target_callbacks callbacks
;
543 dm_block_t low_water_blocks
;
544 unsigned zero_new_blocks
:1;
548 * Target context for a thin.
551 struct dm_dev
*pool_dev
;
555 struct dm_thin_device
*td
;
558 /*----------------------------------------------------------------*/
561 * A global list of pools that uses a struct mapped_device as a key.
563 static struct dm_thin_pool_table
{
565 struct list_head pools
;
566 } dm_thin_pool_table
;
568 static void pool_table_init(void)
570 mutex_init(&dm_thin_pool_table
.mutex
);
571 INIT_LIST_HEAD(&dm_thin_pool_table
.pools
);
574 static void __pool_table_insert(struct pool
*pool
)
576 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
577 list_add(&pool
->list
, &dm_thin_pool_table
.pools
);
580 static void __pool_table_remove(struct pool
*pool
)
582 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
583 list_del(&pool
->list
);
586 static struct pool
*__pool_table_lookup(struct mapped_device
*md
)
588 struct pool
*pool
= NULL
, *tmp
;
590 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
592 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
593 if (tmp
->pool_md
== md
) {
602 static struct pool
*__pool_table_lookup_metadata_dev(struct block_device
*md_dev
)
604 struct pool
*pool
= NULL
, *tmp
;
606 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
608 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
609 if (tmp
->md_dev
== md_dev
) {
618 /*----------------------------------------------------------------*/
620 static void __requeue_bio_list(struct thin_c
*tc
, struct bio_list
*master
)
623 struct bio_list bios
;
625 bio_list_init(&bios
);
626 bio_list_merge(&bios
, master
);
627 bio_list_init(master
);
629 while ((bio
= bio_list_pop(&bios
))) {
630 if (dm_get_mapinfo(bio
)->ptr
== tc
)
631 bio_endio(bio
, DM_ENDIO_REQUEUE
);
633 bio_list_add(master
, bio
);
637 static void requeue_io(struct thin_c
*tc
)
639 struct pool
*pool
= tc
->pool
;
642 spin_lock_irqsave(&pool
->lock
, flags
);
643 __requeue_bio_list(tc
, &pool
->deferred_bios
);
644 __requeue_bio_list(tc
, &pool
->retry_on_resume_list
);
645 spin_unlock_irqrestore(&pool
->lock
, flags
);
649 * This section of code contains the logic for processing a thin device's IO.
650 * Much of the code depends on pool object resources (lists, workqueues, etc)
651 * but most is exclusively called from the thin target rather than the thin-pool
655 static dm_block_t
get_bio_block(struct thin_c
*tc
, struct bio
*bio
)
657 return bio
->bi_sector
>> tc
->pool
->block_shift
;
660 static void remap(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
)
662 struct pool
*pool
= tc
->pool
;
664 bio
->bi_bdev
= tc
->pool_dev
->bdev
;
665 bio
->bi_sector
= (block
<< pool
->block_shift
) +
666 (bio
->bi_sector
& pool
->offset_mask
);
669 static void remap_and_issue(struct thin_c
*tc
, struct bio
*bio
,
672 struct pool
*pool
= tc
->pool
;
675 remap(tc
, bio
, block
);
678 * Batch together any FUA/FLUSH bios we find and then issue
679 * a single commit for them in process_deferred_bios().
681 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
682 spin_lock_irqsave(&pool
->lock
, flags
);
683 bio_list_add(&pool
->deferred_flush_bios
, bio
);
684 spin_unlock_irqrestore(&pool
->lock
, flags
);
686 generic_make_request(bio
);
690 * wake_worker() is used when new work is queued and when pool_resume is
691 * ready to continue deferred IO processing.
693 static void wake_worker(struct pool
*pool
)
695 queue_work(pool
->wq
, &pool
->worker
);
698 /*----------------------------------------------------------------*/
701 * Bio endio functions.
705 bio_end_io_t
*saved_bi_end_io
;
706 struct deferred_entry
*entry
;
710 struct list_head list
;
715 dm_block_t virt_block
;
716 dm_block_t data_block
;
721 * If the bio covers the whole area of a block then we can avoid
722 * zeroing or copying. Instead this bio is hooked. The bio will
723 * still be in the cell, so care has to be taken to avoid issuing
727 bio_end_io_t
*saved_bi_end_io
;
730 static void __maybe_add_mapping(struct new_mapping
*m
)
732 struct pool
*pool
= m
->tc
->pool
;
734 if (list_empty(&m
->list
) && m
->prepared
) {
735 list_add(&m
->list
, &pool
->prepared_mappings
);
740 static void copy_complete(int read_err
, unsigned long write_err
, void *context
)
743 struct new_mapping
*m
= context
;
744 struct pool
*pool
= m
->tc
->pool
;
746 m
->err
= read_err
|| write_err
? -EIO
: 0;
748 spin_lock_irqsave(&pool
->lock
, flags
);
750 __maybe_add_mapping(m
);
751 spin_unlock_irqrestore(&pool
->lock
, flags
);
754 static void overwrite_endio(struct bio
*bio
, int err
)
757 struct new_mapping
*m
= dm_get_mapinfo(bio
)->ptr
;
758 struct pool
*pool
= m
->tc
->pool
;
762 spin_lock_irqsave(&pool
->lock
, flags
);
764 __maybe_add_mapping(m
);
765 spin_unlock_irqrestore(&pool
->lock
, flags
);
768 static void shared_read_endio(struct bio
*bio
, int err
)
770 struct list_head mappings
;
771 struct new_mapping
*m
, *tmp
;
772 struct endio_hook
*h
= dm_get_mapinfo(bio
)->ptr
;
774 struct pool
*pool
= h
->tc
->pool
;
776 bio
->bi_end_io
= h
->saved_bi_end_io
;
779 INIT_LIST_HEAD(&mappings
);
780 ds_dec(h
->entry
, &mappings
);
782 spin_lock_irqsave(&pool
->lock
, flags
);
783 list_for_each_entry_safe(m
, tmp
, &mappings
, list
) {
785 INIT_LIST_HEAD(&m
->list
);
786 __maybe_add_mapping(m
);
788 spin_unlock_irqrestore(&pool
->lock
, flags
);
790 mempool_free(h
, pool
->endio_hook_pool
);
793 /*----------------------------------------------------------------*/
800 * Prepared mapping jobs.
804 * This sends the bios in the cell back to the deferred_bios list.
806 static void cell_defer(struct thin_c
*tc
, struct cell
*cell
,
807 dm_block_t data_block
)
809 struct pool
*pool
= tc
->pool
;
812 spin_lock_irqsave(&pool
->lock
, flags
);
813 cell_release(cell
, &pool
->deferred_bios
);
814 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
820 * Same as cell_defer above, except it omits one particular detainee,
821 * a write bio that covers the block and has already been processed.
823 static void cell_defer_except(struct thin_c
*tc
, struct cell
*cell
)
825 struct bio_list bios
;
826 struct pool
*pool
= tc
->pool
;
829 bio_list_init(&bios
);
831 spin_lock_irqsave(&pool
->lock
, flags
);
832 cell_release_no_holder(cell
, &pool
->deferred_bios
);
833 spin_unlock_irqrestore(&pool
->lock
, flags
);
838 static void process_prepared_mapping(struct new_mapping
*m
)
840 struct thin_c
*tc
= m
->tc
;
846 bio
->bi_end_io
= m
->saved_bi_end_io
;
854 * Commit the prepared block into the mapping btree.
855 * Any I/O for this block arriving after this point will get
856 * remapped to it directly.
858 r
= dm_thin_insert_block(tc
->td
, m
->virt_block
, m
->data_block
);
860 DMERR("dm_thin_insert_block() failed");
866 * Release any bios held while the block was being provisioned.
867 * If we are processing a write bio that completely covers the block,
868 * we already processed it so can ignore it now when processing
869 * the bios in the cell.
872 cell_defer_except(tc
, m
->cell
);
875 cell_defer(tc
, m
->cell
, m
->data_block
);
878 mempool_free(m
, tc
->pool
->mapping_pool
);
881 static void process_prepared_mappings(struct pool
*pool
)
884 struct list_head maps
;
885 struct new_mapping
*m
, *tmp
;
887 INIT_LIST_HEAD(&maps
);
888 spin_lock_irqsave(&pool
->lock
, flags
);
889 list_splice_init(&pool
->prepared_mappings
, &maps
);
890 spin_unlock_irqrestore(&pool
->lock
, flags
);
892 list_for_each_entry_safe(m
, tmp
, &maps
, list
)
893 process_prepared_mapping(m
);
899 static int io_overwrites_block(struct pool
*pool
, struct bio
*bio
)
901 return ((bio_data_dir(bio
) == WRITE
) &&
902 !(bio
->bi_sector
& pool
->offset_mask
)) &&
903 (bio
->bi_size
== (pool
->sectors_per_block
<< SECTOR_SHIFT
));
906 static void save_and_set_endio(struct bio
*bio
, bio_end_io_t
**save
,
909 *save
= bio
->bi_end_io
;
913 static int ensure_next_mapping(struct pool
*pool
)
915 if (pool
->next_mapping
)
918 pool
->next_mapping
= mempool_alloc(pool
->mapping_pool
, GFP_ATOMIC
);
920 return pool
->next_mapping
? 0 : -ENOMEM
;
923 static struct new_mapping
*get_next_mapping(struct pool
*pool
)
925 struct new_mapping
*r
= pool
->next_mapping
;
927 BUG_ON(!pool
->next_mapping
);
929 pool
->next_mapping
= NULL
;
934 static void schedule_copy(struct thin_c
*tc
, dm_block_t virt_block
,
935 dm_block_t data_origin
, dm_block_t data_dest
,
936 struct cell
*cell
, struct bio
*bio
)
939 struct pool
*pool
= tc
->pool
;
940 struct new_mapping
*m
= get_next_mapping(pool
);
942 INIT_LIST_HEAD(&m
->list
);
945 m
->virt_block
= virt_block
;
946 m
->data_block
= data_dest
;
951 ds_add_work(&pool
->ds
, &m
->list
);
954 * IO to pool_dev remaps to the pool target's data_dev.
956 * If the whole block of data is being overwritten, we can issue the
957 * bio immediately. Otherwise we use kcopyd to clone the data first.
959 if (io_overwrites_block(pool
, bio
)) {
961 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
962 dm_get_mapinfo(bio
)->ptr
= m
;
963 remap_and_issue(tc
, bio
, data_dest
);
965 struct dm_io_region from
, to
;
967 from
.bdev
= tc
->pool_dev
->bdev
;
968 from
.sector
= data_origin
* pool
->sectors_per_block
;
969 from
.count
= pool
->sectors_per_block
;
971 to
.bdev
= tc
->pool_dev
->bdev
;
972 to
.sector
= data_dest
* pool
->sectors_per_block
;
973 to
.count
= pool
->sectors_per_block
;
975 r
= dm_kcopyd_copy(pool
->copier
, &from
, 1, &to
,
976 0, copy_complete
, m
);
978 mempool_free(m
, pool
->mapping_pool
);
979 DMERR("dm_kcopyd_copy() failed");
985 static void schedule_zero(struct thin_c
*tc
, dm_block_t virt_block
,
986 dm_block_t data_block
, struct cell
*cell
,
989 struct pool
*pool
= tc
->pool
;
990 struct new_mapping
*m
= get_next_mapping(pool
);
992 INIT_LIST_HEAD(&m
->list
);
995 m
->virt_block
= virt_block
;
996 m
->data_block
= data_block
;
1002 * If the whole block of data is being overwritten or we are not
1003 * zeroing pre-existing data, we can issue the bio immediately.
1004 * Otherwise we use kcopyd to zero the data first.
1006 if (!pool
->zero_new_blocks
)
1007 process_prepared_mapping(m
);
1009 else if (io_overwrites_block(pool
, bio
)) {
1011 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
1012 dm_get_mapinfo(bio
)->ptr
= m
;
1013 remap_and_issue(tc
, bio
, data_block
);
1017 struct dm_io_region to
;
1019 to
.bdev
= tc
->pool_dev
->bdev
;
1020 to
.sector
= data_block
* pool
->sectors_per_block
;
1021 to
.count
= pool
->sectors_per_block
;
1023 r
= dm_kcopyd_zero(pool
->copier
, 1, &to
, 0, copy_complete
, m
);
1025 mempool_free(m
, pool
->mapping_pool
);
1026 DMERR("dm_kcopyd_zero() failed");
1032 static int alloc_data_block(struct thin_c
*tc
, dm_block_t
*result
)
1035 dm_block_t free_blocks
;
1036 unsigned long flags
;
1037 struct pool
*pool
= tc
->pool
;
1039 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
1043 if (free_blocks
<= pool
->low_water_blocks
&& !pool
->low_water_triggered
) {
1044 DMWARN("%s: reached low water mark, sending event.",
1045 dm_device_name(pool
->pool_md
));
1046 spin_lock_irqsave(&pool
->lock
, flags
);
1047 pool
->low_water_triggered
= 1;
1048 spin_unlock_irqrestore(&pool
->lock
, flags
);
1049 dm_table_event(pool
->ti
->table
);
1053 if (pool
->no_free_space
)
1057 * Try to commit to see if that will free up some
1060 r
= dm_pool_commit_metadata(pool
->pmd
);
1062 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
1067 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
1072 * If we still have no space we set a flag to avoid
1073 * doing all this checking and return -ENOSPC.
1076 DMWARN("%s: no free space available.",
1077 dm_device_name(pool
->pool_md
));
1078 spin_lock_irqsave(&pool
->lock
, flags
);
1079 pool
->no_free_space
= 1;
1080 spin_unlock_irqrestore(&pool
->lock
, flags
);
1086 r
= dm_pool_alloc_data_block(pool
->pmd
, result
);
1094 * If we have run out of space, queue bios until the device is
1095 * resumed, presumably after having been reloaded with more space.
1097 static void retry_on_resume(struct bio
*bio
)
1099 struct thin_c
*tc
= dm_get_mapinfo(bio
)->ptr
;
1100 struct pool
*pool
= tc
->pool
;
1101 unsigned long flags
;
1103 spin_lock_irqsave(&pool
->lock
, flags
);
1104 bio_list_add(&pool
->retry_on_resume_list
, bio
);
1105 spin_unlock_irqrestore(&pool
->lock
, flags
);
1108 static void no_space(struct cell
*cell
)
1111 struct bio_list bios
;
1113 bio_list_init(&bios
);
1114 cell_release(cell
, &bios
);
1116 while ((bio
= bio_list_pop(&bios
)))
1117 retry_on_resume(bio
);
1120 static void break_sharing(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1121 struct cell_key
*key
,
1122 struct dm_thin_lookup_result
*lookup_result
,
1126 dm_block_t data_block
;
1128 r
= alloc_data_block(tc
, &data_block
);
1131 schedule_copy(tc
, block
, lookup_result
->block
,
1132 data_block
, cell
, bio
);
1140 DMERR("%s: alloc_data_block() failed, error = %d", __func__
, r
);
1146 static void process_shared_bio(struct thin_c
*tc
, struct bio
*bio
,
1148 struct dm_thin_lookup_result
*lookup_result
)
1151 struct pool
*pool
= tc
->pool
;
1152 struct cell_key key
;
1155 * If cell is already occupied, then sharing is already in the process
1156 * of being broken so we have nothing further to do here.
1158 build_data_key(tc
->td
, lookup_result
->block
, &key
);
1159 if (bio_detain(pool
->prison
, &key
, bio
, &cell
))
1162 if (bio_data_dir(bio
) == WRITE
)
1163 break_sharing(tc
, bio
, block
, &key
, lookup_result
, cell
);
1165 struct endio_hook
*h
;
1166 h
= mempool_alloc(pool
->endio_hook_pool
, GFP_NOIO
);
1169 h
->entry
= ds_inc(&pool
->ds
);
1170 save_and_set_endio(bio
, &h
->saved_bi_end_io
, shared_read_endio
);
1171 dm_get_mapinfo(bio
)->ptr
= h
;
1173 cell_release_singleton(cell
, bio
);
1174 remap_and_issue(tc
, bio
, lookup_result
->block
);
1178 static void provision_block(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1182 dm_block_t data_block
;
1185 * Remap empty bios (flushes) immediately, without provisioning.
1187 if (!bio
->bi_size
) {
1188 cell_release_singleton(cell
, bio
);
1189 remap_and_issue(tc
, bio
, 0);
1194 * Fill read bios with zeroes and complete them immediately.
1196 if (bio_data_dir(bio
) == READ
) {
1198 cell_release_singleton(cell
, bio
);
1203 r
= alloc_data_block(tc
, &data_block
);
1206 schedule_zero(tc
, block
, data_block
, cell
, bio
);
1214 DMERR("%s: alloc_data_block() failed, error = %d", __func__
, r
);
1220 static void process_bio(struct thin_c
*tc
, struct bio
*bio
)
1223 dm_block_t block
= get_bio_block(tc
, bio
);
1225 struct cell_key key
;
1226 struct dm_thin_lookup_result lookup_result
;
1229 * If cell is already occupied, then the block is already
1230 * being provisioned so we have nothing further to do here.
1232 build_virtual_key(tc
->td
, block
, &key
);
1233 if (bio_detain(tc
->pool
->prison
, &key
, bio
, &cell
))
1236 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1240 * We can release this cell now. This thread is the only
1241 * one that puts bios into a cell, and we know there were
1242 * no preceding bios.
1245 * TODO: this will probably have to change when discard goes
1248 cell_release_singleton(cell
, bio
);
1250 if (lookup_result
.shared
)
1251 process_shared_bio(tc
, bio
, block
, &lookup_result
);
1253 remap_and_issue(tc
, bio
, lookup_result
.block
);
1257 provision_block(tc
, bio
, block
, cell
);
1261 DMERR("dm_thin_find_block() failed, error = %d", r
);
1267 static int need_commit_due_to_time(struct pool
*pool
)
1269 return jiffies
< pool
->last_commit_jiffies
||
1270 jiffies
> pool
->last_commit_jiffies
+ COMMIT_PERIOD
;
1273 static void process_deferred_bios(struct pool
*pool
)
1275 unsigned long flags
;
1277 struct bio_list bios
;
1280 bio_list_init(&bios
);
1282 spin_lock_irqsave(&pool
->lock
, flags
);
1283 bio_list_merge(&bios
, &pool
->deferred_bios
);
1284 bio_list_init(&pool
->deferred_bios
);
1285 spin_unlock_irqrestore(&pool
->lock
, flags
);
1287 while ((bio
= bio_list_pop(&bios
))) {
1288 struct thin_c
*tc
= dm_get_mapinfo(bio
)->ptr
;
1290 * If we've got no free new_mapping structs, and processing
1291 * this bio might require one, we pause until there are some
1292 * prepared mappings to process.
1294 if (ensure_next_mapping(pool
)) {
1295 spin_lock_irqsave(&pool
->lock
, flags
);
1296 bio_list_merge(&pool
->deferred_bios
, &bios
);
1297 spin_unlock_irqrestore(&pool
->lock
, flags
);
1301 process_bio(tc
, bio
);
1305 * If there are any deferred flush bios, we must commit
1306 * the metadata before issuing them.
1308 bio_list_init(&bios
);
1309 spin_lock_irqsave(&pool
->lock
, flags
);
1310 bio_list_merge(&bios
, &pool
->deferred_flush_bios
);
1311 bio_list_init(&pool
->deferred_flush_bios
);
1312 spin_unlock_irqrestore(&pool
->lock
, flags
);
1314 if (bio_list_empty(&bios
) && !need_commit_due_to_time(pool
))
1317 r
= dm_pool_commit_metadata(pool
->pmd
);
1319 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
1321 while ((bio
= bio_list_pop(&bios
)))
1325 pool
->last_commit_jiffies
= jiffies
;
1327 while ((bio
= bio_list_pop(&bios
)))
1328 generic_make_request(bio
);
1331 static void do_worker(struct work_struct
*ws
)
1333 struct pool
*pool
= container_of(ws
, struct pool
, worker
);
1335 process_prepared_mappings(pool
);
1336 process_deferred_bios(pool
);
1340 * We want to commit periodically so that not too much
1341 * unwritten data builds up.
1343 static void do_waker(struct work_struct
*ws
)
1345 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
, waker
);
1347 queue_delayed_work(pool
->wq
, &pool
->waker
, COMMIT_PERIOD
);
1350 /*----------------------------------------------------------------*/
1353 * Mapping functions.
1357 * Called only while mapping a thin bio to hand it over to the workqueue.
1359 static void thin_defer_bio(struct thin_c
*tc
, struct bio
*bio
)
1361 unsigned long flags
;
1362 struct pool
*pool
= tc
->pool
;
1364 spin_lock_irqsave(&pool
->lock
, flags
);
1365 bio_list_add(&pool
->deferred_bios
, bio
);
1366 spin_unlock_irqrestore(&pool
->lock
, flags
);
1372 * Non-blocking function called from the thin target's map function.
1374 static int thin_bio_map(struct dm_target
*ti
, struct bio
*bio
,
1375 union map_info
*map_context
)
1378 struct thin_c
*tc
= ti
->private;
1379 dm_block_t block
= get_bio_block(tc
, bio
);
1380 struct dm_thin_device
*td
= tc
->td
;
1381 struct dm_thin_lookup_result result
;
1384 * Save the thin context for easy access from the deferred bio later.
1386 map_context
->ptr
= tc
;
1388 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1389 thin_defer_bio(tc
, bio
);
1390 return DM_MAPIO_SUBMITTED
;
1393 r
= dm_thin_find_block(td
, block
, 0, &result
);
1396 * Note that we defer readahead too.
1400 if (unlikely(result
.shared
)) {
1402 * We have a race condition here between the
1403 * result.shared value returned by the lookup and
1404 * snapshot creation, which may cause new
1407 * To avoid this always quiesce the origin before
1408 * taking the snap. You want to do this anyway to
1409 * ensure a consistent application view
1412 * More distant ancestors are irrelevant. The
1413 * shared flag will be set in their case.
1415 thin_defer_bio(tc
, bio
);
1416 r
= DM_MAPIO_SUBMITTED
;
1418 remap(tc
, bio
, result
.block
);
1419 r
= DM_MAPIO_REMAPPED
;
1425 * In future, the failed dm_thin_find_block above could
1426 * provide the hint to load the metadata into cache.
1429 thin_defer_bio(tc
, bio
);
1430 r
= DM_MAPIO_SUBMITTED
;
1437 static int pool_is_congested(struct dm_target_callbacks
*cb
, int bdi_bits
)
1440 unsigned long flags
;
1441 struct pool_c
*pt
= container_of(cb
, struct pool_c
, callbacks
);
1443 spin_lock_irqsave(&pt
->pool
->lock
, flags
);
1444 r
= !bio_list_empty(&pt
->pool
->retry_on_resume_list
);
1445 spin_unlock_irqrestore(&pt
->pool
->lock
, flags
);
1448 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1449 r
= bdi_congested(&q
->backing_dev_info
, bdi_bits
);
1455 static void __requeue_bios(struct pool
*pool
)
1457 bio_list_merge(&pool
->deferred_bios
, &pool
->retry_on_resume_list
);
1458 bio_list_init(&pool
->retry_on_resume_list
);
1461 /*----------------------------------------------------------------
1462 * Binding of control targets to a pool object
1463 *--------------------------------------------------------------*/
1464 static int bind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1466 struct pool_c
*pt
= ti
->private;
1469 pool
->low_water_blocks
= pt
->low_water_blocks
;
1470 pool
->zero_new_blocks
= pt
->zero_new_blocks
;
1475 static void unbind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1481 /*----------------------------------------------------------------
1483 *--------------------------------------------------------------*/
1484 static void __pool_destroy(struct pool
*pool
)
1486 __pool_table_remove(pool
);
1488 if (dm_pool_metadata_close(pool
->pmd
) < 0)
1489 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1491 prison_destroy(pool
->prison
);
1492 dm_kcopyd_client_destroy(pool
->copier
);
1495 destroy_workqueue(pool
->wq
);
1497 if (pool
->next_mapping
)
1498 mempool_free(pool
->next_mapping
, pool
->mapping_pool
);
1499 mempool_destroy(pool
->mapping_pool
);
1500 mempool_destroy(pool
->endio_hook_pool
);
1504 static struct pool
*pool_create(struct mapped_device
*pool_md
,
1505 struct block_device
*metadata_dev
,
1506 unsigned long block_size
, char **error
)
1511 struct dm_pool_metadata
*pmd
;
1513 pmd
= dm_pool_metadata_open(metadata_dev
, block_size
);
1515 *error
= "Error creating metadata object";
1516 return (struct pool
*)pmd
;
1519 pool
= kmalloc(sizeof(*pool
), GFP_KERNEL
);
1521 *error
= "Error allocating memory for pool";
1522 err_p
= ERR_PTR(-ENOMEM
);
1527 pool
->sectors_per_block
= block_size
;
1528 pool
->block_shift
= ffs(block_size
) - 1;
1529 pool
->offset_mask
= block_size
- 1;
1530 pool
->low_water_blocks
= 0;
1531 pool
->zero_new_blocks
= 1;
1532 pool
->prison
= prison_create(PRISON_CELLS
);
1533 if (!pool
->prison
) {
1534 *error
= "Error creating pool's bio prison";
1535 err_p
= ERR_PTR(-ENOMEM
);
1539 pool
->copier
= dm_kcopyd_client_create();
1540 if (IS_ERR(pool
->copier
)) {
1541 r
= PTR_ERR(pool
->copier
);
1542 *error
= "Error creating pool's kcopyd client";
1544 goto bad_kcopyd_client
;
1548 * Create singlethreaded workqueue that will service all devices
1549 * that use this metadata.
1551 pool
->wq
= alloc_ordered_workqueue("dm-" DM_MSG_PREFIX
, WQ_MEM_RECLAIM
);
1553 *error
= "Error creating pool's workqueue";
1554 err_p
= ERR_PTR(-ENOMEM
);
1558 INIT_WORK(&pool
->worker
, do_worker
);
1559 INIT_DELAYED_WORK(&pool
->waker
, do_waker
);
1560 spin_lock_init(&pool
->lock
);
1561 bio_list_init(&pool
->deferred_bios
);
1562 bio_list_init(&pool
->deferred_flush_bios
);
1563 INIT_LIST_HEAD(&pool
->prepared_mappings
);
1564 pool
->low_water_triggered
= 0;
1565 pool
->no_free_space
= 0;
1566 bio_list_init(&pool
->retry_on_resume_list
);
1569 pool
->next_mapping
= NULL
;
1570 pool
->mapping_pool
=
1571 mempool_create_kmalloc_pool(MAPPING_POOL_SIZE
, sizeof(struct new_mapping
));
1572 if (!pool
->mapping_pool
) {
1573 *error
= "Error creating pool's mapping mempool";
1574 err_p
= ERR_PTR(-ENOMEM
);
1575 goto bad_mapping_pool
;
1578 pool
->endio_hook_pool
=
1579 mempool_create_kmalloc_pool(ENDIO_HOOK_POOL_SIZE
, sizeof(struct endio_hook
));
1580 if (!pool
->endio_hook_pool
) {
1581 *error
= "Error creating pool's endio_hook mempool";
1582 err_p
= ERR_PTR(-ENOMEM
);
1583 goto bad_endio_hook_pool
;
1585 pool
->ref_count
= 1;
1586 pool
->last_commit_jiffies
= jiffies
;
1587 pool
->pool_md
= pool_md
;
1588 pool
->md_dev
= metadata_dev
;
1589 __pool_table_insert(pool
);
1593 bad_endio_hook_pool
:
1594 mempool_destroy(pool
->mapping_pool
);
1596 destroy_workqueue(pool
->wq
);
1598 dm_kcopyd_client_destroy(pool
->copier
);
1600 prison_destroy(pool
->prison
);
1604 if (dm_pool_metadata_close(pmd
))
1605 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1610 static void __pool_inc(struct pool
*pool
)
1612 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
1616 static void __pool_dec(struct pool
*pool
)
1618 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
1619 BUG_ON(!pool
->ref_count
);
1620 if (!--pool
->ref_count
)
1621 __pool_destroy(pool
);
1624 static struct pool
*__pool_find(struct mapped_device
*pool_md
,
1625 struct block_device
*metadata_dev
,
1626 unsigned long block_size
, char **error
)
1628 struct pool
*pool
= __pool_table_lookup_metadata_dev(metadata_dev
);
1631 if (pool
->pool_md
!= pool_md
)
1632 return ERR_PTR(-EBUSY
);
1636 pool
= __pool_table_lookup(pool_md
);
1638 if (pool
->md_dev
!= metadata_dev
)
1639 return ERR_PTR(-EINVAL
);
1643 pool
= pool_create(pool_md
, metadata_dev
, block_size
, error
);
1649 /*----------------------------------------------------------------
1650 * Pool target methods
1651 *--------------------------------------------------------------*/
1652 static void pool_dtr(struct dm_target
*ti
)
1654 struct pool_c
*pt
= ti
->private;
1656 mutex_lock(&dm_thin_pool_table
.mutex
);
1658 unbind_control_target(pt
->pool
, ti
);
1659 __pool_dec(pt
->pool
);
1660 dm_put_device(ti
, pt
->metadata_dev
);
1661 dm_put_device(ti
, pt
->data_dev
);
1664 mutex_unlock(&dm_thin_pool_table
.mutex
);
1667 struct pool_features
{
1668 unsigned zero_new_blocks
:1;
1671 static int parse_pool_features(struct dm_arg_set
*as
, struct pool_features
*pf
,
1672 struct dm_target
*ti
)
1676 const char *arg_name
;
1678 static struct dm_arg _args
[] = {
1679 {0, 1, "Invalid number of pool feature arguments"},
1683 * No feature arguments supplied.
1688 r
= dm_read_arg_group(_args
, as
, &argc
, &ti
->error
);
1692 while (argc
&& !r
) {
1693 arg_name
= dm_shift_arg(as
);
1696 if (!strcasecmp(arg_name
, "skip_block_zeroing")) {
1697 pf
->zero_new_blocks
= 0;
1701 ti
->error
= "Unrecognised pool feature requested";
1709 * thin-pool <metadata dev> <data dev>
1710 * <data block size (sectors)>
1711 * <low water mark (blocks)>
1712 * [<#feature args> [<arg>]*]
1714 * Optional feature arguments are:
1715 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
1717 static int pool_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
1722 struct pool_features pf
;
1723 struct dm_arg_set as
;
1724 struct dm_dev
*data_dev
;
1725 unsigned long block_size
;
1726 dm_block_t low_water_blocks
;
1727 struct dm_dev
*metadata_dev
;
1728 sector_t metadata_dev_size
;
1729 char b
[BDEVNAME_SIZE
];
1732 * FIXME Remove validation from scope of lock.
1734 mutex_lock(&dm_thin_pool_table
.mutex
);
1737 ti
->error
= "Invalid argument count";
1744 r
= dm_get_device(ti
, argv
[0], FMODE_READ
| FMODE_WRITE
, &metadata_dev
);
1746 ti
->error
= "Error opening metadata block device";
1750 metadata_dev_size
= i_size_read(metadata_dev
->bdev
->bd_inode
) >> SECTOR_SHIFT
;
1751 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS_WARNING
)
1752 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
1753 bdevname(metadata_dev
->bdev
, b
), THIN_METADATA_MAX_SECTORS
);
1755 r
= dm_get_device(ti
, argv
[1], FMODE_READ
| FMODE_WRITE
, &data_dev
);
1757 ti
->error
= "Error getting data device";
1761 if (kstrtoul(argv
[2], 10, &block_size
) || !block_size
||
1762 block_size
< DATA_DEV_BLOCK_SIZE_MIN_SECTORS
||
1763 block_size
> DATA_DEV_BLOCK_SIZE_MAX_SECTORS
||
1764 !is_power_of_2(block_size
)) {
1765 ti
->error
= "Invalid block size";
1770 if (kstrtoull(argv
[3], 10, (unsigned long long *)&low_water_blocks
)) {
1771 ti
->error
= "Invalid low water mark";
1777 * Set default pool features.
1779 memset(&pf
, 0, sizeof(pf
));
1780 pf
.zero_new_blocks
= 1;
1782 dm_consume_args(&as
, 4);
1783 r
= parse_pool_features(&as
, &pf
, ti
);
1787 pt
= kzalloc(sizeof(*pt
), GFP_KERNEL
);
1793 pool
= __pool_find(dm_table_get_md(ti
->table
), metadata_dev
->bdev
,
1794 block_size
, &ti
->error
);
1802 pt
->metadata_dev
= metadata_dev
;
1803 pt
->data_dev
= data_dev
;
1804 pt
->low_water_blocks
= low_water_blocks
;
1805 pt
->zero_new_blocks
= pf
.zero_new_blocks
;
1806 ti
->num_flush_requests
= 1;
1807 ti
->num_discard_requests
= 0;
1810 pt
->callbacks
.congested_fn
= pool_is_congested
;
1811 dm_table_add_target_callbacks(ti
->table
, &pt
->callbacks
);
1813 mutex_unlock(&dm_thin_pool_table
.mutex
);
1820 dm_put_device(ti
, data_dev
);
1822 dm_put_device(ti
, metadata_dev
);
1824 mutex_unlock(&dm_thin_pool_table
.mutex
);
1829 static int pool_map(struct dm_target
*ti
, struct bio
*bio
,
1830 union map_info
*map_context
)
1833 struct pool_c
*pt
= ti
->private;
1834 struct pool
*pool
= pt
->pool
;
1835 unsigned long flags
;
1838 * As this is a singleton target, ti->begin is always zero.
1840 spin_lock_irqsave(&pool
->lock
, flags
);
1841 bio
->bi_bdev
= pt
->data_dev
->bdev
;
1842 r
= DM_MAPIO_REMAPPED
;
1843 spin_unlock_irqrestore(&pool
->lock
, flags
);
1849 * Retrieves the number of blocks of the data device from
1850 * the superblock and compares it to the actual device size,
1851 * thus resizing the data device in case it has grown.
1853 * This both copes with opening preallocated data devices in the ctr
1854 * being followed by a resume
1856 * calling the resume method individually after userspace has
1857 * grown the data device in reaction to a table event.
1859 static int pool_preresume(struct dm_target
*ti
)
1862 struct pool_c
*pt
= ti
->private;
1863 struct pool
*pool
= pt
->pool
;
1864 dm_block_t data_size
, sb_data_size
;
1867 * Take control of the pool object.
1869 r
= bind_control_target(pool
, ti
);
1873 data_size
= ti
->len
>> pool
->block_shift
;
1874 r
= dm_pool_get_data_dev_size(pool
->pmd
, &sb_data_size
);
1876 DMERR("failed to retrieve data device size");
1880 if (data_size
< sb_data_size
) {
1881 DMERR("pool target too small, is %llu blocks (expected %llu)",
1882 data_size
, sb_data_size
);
1885 } else if (data_size
> sb_data_size
) {
1886 r
= dm_pool_resize_data_dev(pool
->pmd
, data_size
);
1888 DMERR("failed to resize data device");
1892 r
= dm_pool_commit_metadata(pool
->pmd
);
1894 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
1903 static void pool_resume(struct dm_target
*ti
)
1905 struct pool_c
*pt
= ti
->private;
1906 struct pool
*pool
= pt
->pool
;
1907 unsigned long flags
;
1909 spin_lock_irqsave(&pool
->lock
, flags
);
1910 pool
->low_water_triggered
= 0;
1911 pool
->no_free_space
= 0;
1912 __requeue_bios(pool
);
1913 spin_unlock_irqrestore(&pool
->lock
, flags
);
1915 do_waker(&pool
->waker
.work
);
1918 static void pool_postsuspend(struct dm_target
*ti
)
1921 struct pool_c
*pt
= ti
->private;
1922 struct pool
*pool
= pt
->pool
;
1924 cancel_delayed_work(&pool
->waker
);
1925 flush_workqueue(pool
->wq
);
1927 r
= dm_pool_commit_metadata(pool
->pmd
);
1929 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
1931 /* FIXME: invalidate device? error the next FUA or FLUSH bio ?*/
1935 static int check_arg_count(unsigned argc
, unsigned args_required
)
1937 if (argc
!= args_required
) {
1938 DMWARN("Message received with %u arguments instead of %u.",
1939 argc
, args_required
);
1946 static int read_dev_id(char *arg
, dm_thin_id
*dev_id
, int warning
)
1948 if (!kstrtoull(arg
, 10, (unsigned long long *)dev_id
) &&
1949 *dev_id
<= MAX_DEV_ID
)
1953 DMWARN("Message received with invalid device id: %s", arg
);
1958 static int process_create_thin_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
1963 r
= check_arg_count(argc
, 2);
1967 r
= read_dev_id(argv
[1], &dev_id
, 1);
1971 r
= dm_pool_create_thin(pool
->pmd
, dev_id
);
1973 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
1981 static int process_create_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
1984 dm_thin_id origin_dev_id
;
1987 r
= check_arg_count(argc
, 3);
1991 r
= read_dev_id(argv
[1], &dev_id
, 1);
1995 r
= read_dev_id(argv
[2], &origin_dev_id
, 1);
1999 r
= dm_pool_create_snap(pool
->pmd
, dev_id
, origin_dev_id
);
2001 DMWARN("Creation of new snapshot %s of device %s failed.",
2009 static int process_delete_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2014 r
= check_arg_count(argc
, 2);
2018 r
= read_dev_id(argv
[1], &dev_id
, 1);
2022 r
= dm_pool_delete_thin_device(pool
->pmd
, dev_id
);
2024 DMWARN("Deletion of thin device %s failed.", argv
[1]);
2029 static int process_set_transaction_id_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2031 dm_thin_id old_id
, new_id
;
2034 r
= check_arg_count(argc
, 3);
2038 if (kstrtoull(argv
[1], 10, (unsigned long long *)&old_id
)) {
2039 DMWARN("set_transaction_id message: Unrecognised id %s.", argv
[1]);
2043 if (kstrtoull(argv
[2], 10, (unsigned long long *)&new_id
)) {
2044 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv
[2]);
2048 r
= dm_pool_set_metadata_transaction_id(pool
->pmd
, old_id
, new_id
);
2050 DMWARN("Failed to change transaction id from %s to %s.",
2059 * Messages supported:
2060 * create_thin <dev_id>
2061 * create_snap <dev_id> <origin_id>
2063 * trim <dev_id> <new_size_in_sectors>
2064 * set_transaction_id <current_trans_id> <new_trans_id>
2066 static int pool_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
2069 struct pool_c
*pt
= ti
->private;
2070 struct pool
*pool
= pt
->pool
;
2072 if (!strcasecmp(argv
[0], "create_thin"))
2073 r
= process_create_thin_mesg(argc
, argv
, pool
);
2075 else if (!strcasecmp(argv
[0], "create_snap"))
2076 r
= process_create_snap_mesg(argc
, argv
, pool
);
2078 else if (!strcasecmp(argv
[0], "delete"))
2079 r
= process_delete_mesg(argc
, argv
, pool
);
2081 else if (!strcasecmp(argv
[0], "set_transaction_id"))
2082 r
= process_set_transaction_id_mesg(argc
, argv
, pool
);
2085 DMWARN("Unrecognised thin pool target message received: %s", argv
[0]);
2088 r
= dm_pool_commit_metadata(pool
->pmd
);
2090 DMERR("%s message: dm_pool_commit_metadata() failed, error = %d",
2099 * <transaction id> <used metadata sectors>/<total metadata sectors>
2100 * <used data sectors>/<total data sectors> <held metadata root>
2102 static int pool_status(struct dm_target
*ti
, status_type_t type
,
2103 char *result
, unsigned maxlen
)
2107 uint64_t transaction_id
;
2108 dm_block_t nr_free_blocks_data
;
2109 dm_block_t nr_free_blocks_metadata
;
2110 dm_block_t nr_blocks_data
;
2111 dm_block_t nr_blocks_metadata
;
2112 dm_block_t held_root
;
2113 char buf
[BDEVNAME_SIZE
];
2114 char buf2
[BDEVNAME_SIZE
];
2115 struct pool_c
*pt
= ti
->private;
2116 struct pool
*pool
= pt
->pool
;
2119 case STATUSTYPE_INFO
:
2120 r
= dm_pool_get_metadata_transaction_id(pool
->pmd
,
2125 r
= dm_pool_get_free_metadata_block_count(pool
->pmd
,
2126 &nr_free_blocks_metadata
);
2130 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &nr_blocks_metadata
);
2134 r
= dm_pool_get_free_block_count(pool
->pmd
,
2135 &nr_free_blocks_data
);
2139 r
= dm_pool_get_data_dev_size(pool
->pmd
, &nr_blocks_data
);
2143 r
= dm_pool_get_held_metadata_root(pool
->pmd
, &held_root
);
2147 DMEMIT("%llu %llu/%llu %llu/%llu ",
2148 (unsigned long long)transaction_id
,
2149 (unsigned long long)(nr_blocks_metadata
- nr_free_blocks_metadata
),
2150 (unsigned long long)nr_blocks_metadata
,
2151 (unsigned long long)(nr_blocks_data
- nr_free_blocks_data
),
2152 (unsigned long long)nr_blocks_data
);
2155 DMEMIT("%llu", held_root
);
2161 case STATUSTYPE_TABLE
:
2162 DMEMIT("%s %s %lu %llu ",
2163 format_dev_t(buf
, pt
->metadata_dev
->bdev
->bd_dev
),
2164 format_dev_t(buf2
, pt
->data_dev
->bdev
->bd_dev
),
2165 (unsigned long)pool
->sectors_per_block
,
2166 (unsigned long long)pt
->low_water_blocks
);
2168 DMEMIT("%u ", !pool
->zero_new_blocks
);
2170 if (!pool
->zero_new_blocks
)
2171 DMEMIT("skip_block_zeroing ");
2178 static int pool_iterate_devices(struct dm_target
*ti
,
2179 iterate_devices_callout_fn fn
, void *data
)
2181 struct pool_c
*pt
= ti
->private;
2183 return fn(ti
, pt
->data_dev
, 0, ti
->len
, data
);
2186 static int pool_merge(struct dm_target
*ti
, struct bvec_merge_data
*bvm
,
2187 struct bio_vec
*biovec
, int max_size
)
2189 struct pool_c
*pt
= ti
->private;
2190 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
2192 if (!q
->merge_bvec_fn
)
2195 bvm
->bi_bdev
= pt
->data_dev
->bdev
;
2197 return min(max_size
, q
->merge_bvec_fn(q
, bvm
, biovec
));
2200 static void pool_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
2202 struct pool_c
*pt
= ti
->private;
2203 struct pool
*pool
= pt
->pool
;
2205 blk_limits_io_min(limits
, 0);
2206 blk_limits_io_opt(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
2209 static struct target_type pool_target
= {
2210 .name
= "thin-pool",
2211 .features
= DM_TARGET_SINGLETON
| DM_TARGET_ALWAYS_WRITEABLE
|
2212 DM_TARGET_IMMUTABLE
,
2213 .version
= {1, 0, 0},
2214 .module
= THIS_MODULE
,
2218 .postsuspend
= pool_postsuspend
,
2219 .preresume
= pool_preresume
,
2220 .resume
= pool_resume
,
2221 .message
= pool_message
,
2222 .status
= pool_status
,
2223 .merge
= pool_merge
,
2224 .iterate_devices
= pool_iterate_devices
,
2225 .io_hints
= pool_io_hints
,
2228 /*----------------------------------------------------------------
2229 * Thin target methods
2230 *--------------------------------------------------------------*/
2231 static void thin_dtr(struct dm_target
*ti
)
2233 struct thin_c
*tc
= ti
->private;
2235 mutex_lock(&dm_thin_pool_table
.mutex
);
2237 __pool_dec(tc
->pool
);
2238 dm_pool_close_thin_device(tc
->td
);
2239 dm_put_device(ti
, tc
->pool_dev
);
2242 mutex_unlock(&dm_thin_pool_table
.mutex
);
2246 * Thin target parameters:
2248 * <pool_dev> <dev_id>
2250 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2251 * dev_id: the internal device identifier
2253 static int thin_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2257 struct dm_dev
*pool_dev
;
2258 struct mapped_device
*pool_md
;
2260 mutex_lock(&dm_thin_pool_table
.mutex
);
2263 ti
->error
= "Invalid argument count";
2268 tc
= ti
->private = kzalloc(sizeof(*tc
), GFP_KERNEL
);
2270 ti
->error
= "Out of memory";
2275 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &pool_dev
);
2277 ti
->error
= "Error opening pool device";
2280 tc
->pool_dev
= pool_dev
;
2282 if (read_dev_id(argv
[1], (unsigned long long *)&tc
->dev_id
, 0)) {
2283 ti
->error
= "Invalid device id";
2288 pool_md
= dm_get_md(tc
->pool_dev
->bdev
->bd_dev
);
2290 ti
->error
= "Couldn't get pool mapped device";
2295 tc
->pool
= __pool_table_lookup(pool_md
);
2297 ti
->error
= "Couldn't find pool object";
2299 goto bad_pool_lookup
;
2301 __pool_inc(tc
->pool
);
2303 r
= dm_pool_open_thin_device(tc
->pool
->pmd
, tc
->dev_id
, &tc
->td
);
2305 ti
->error
= "Couldn't open thin internal device";
2309 ti
->split_io
= tc
->pool
->sectors_per_block
;
2310 ti
->num_flush_requests
= 1;
2311 ti
->num_discard_requests
= 0;
2312 ti
->discards_supported
= 0;
2316 mutex_unlock(&dm_thin_pool_table
.mutex
);
2321 __pool_dec(tc
->pool
);
2325 dm_put_device(ti
, tc
->pool_dev
);
2329 mutex_unlock(&dm_thin_pool_table
.mutex
);
2334 static int thin_map(struct dm_target
*ti
, struct bio
*bio
,
2335 union map_info
*map_context
)
2337 bio
->bi_sector
-= ti
->begin
;
2339 return thin_bio_map(ti
, bio
, map_context
);
2342 static void thin_postsuspend(struct dm_target
*ti
)
2344 if (dm_noflush_suspending(ti
))
2345 requeue_io((struct thin_c
*)ti
->private);
2349 * <nr mapped sectors> <highest mapped sector>
2351 static int thin_status(struct dm_target
*ti
, status_type_t type
,
2352 char *result
, unsigned maxlen
)
2356 dm_block_t mapped
, highest
;
2357 char buf
[BDEVNAME_SIZE
];
2358 struct thin_c
*tc
= ti
->private;
2364 case STATUSTYPE_INFO
:
2365 r
= dm_thin_get_mapped_count(tc
->td
, &mapped
);
2369 r
= dm_thin_get_highest_mapped_block(tc
->td
, &highest
);
2373 DMEMIT("%llu ", mapped
* tc
->pool
->sectors_per_block
);
2375 DMEMIT("%llu", ((highest
+ 1) *
2376 tc
->pool
->sectors_per_block
) - 1);
2381 case STATUSTYPE_TABLE
:
2383 format_dev_t(buf
, tc
->pool_dev
->bdev
->bd_dev
),
2384 (unsigned long) tc
->dev_id
);
2392 static int thin_iterate_devices(struct dm_target
*ti
,
2393 iterate_devices_callout_fn fn
, void *data
)
2396 struct thin_c
*tc
= ti
->private;
2399 * We can't call dm_pool_get_data_dev_size() since that blocks. So
2400 * we follow a more convoluted path through to the pool's target.
2403 return 0; /* nothing is bound */
2405 blocks
= tc
->pool
->ti
->len
>> tc
->pool
->block_shift
;
2407 return fn(ti
, tc
->pool_dev
, 0, tc
->pool
->sectors_per_block
* blocks
, data
);
2412 static void thin_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
2414 struct thin_c
*tc
= ti
->private;
2416 blk_limits_io_min(limits
, 0);
2417 blk_limits_io_opt(limits
, tc
->pool
->sectors_per_block
<< SECTOR_SHIFT
);
2420 static struct target_type thin_target
= {
2422 .version
= {1, 0, 0},
2423 .module
= THIS_MODULE
,
2427 .postsuspend
= thin_postsuspend
,
2428 .status
= thin_status
,
2429 .iterate_devices
= thin_iterate_devices
,
2430 .io_hints
= thin_io_hints
,
2433 /*----------------------------------------------------------------*/
2435 static int __init
dm_thin_init(void)
2441 r
= dm_register_target(&thin_target
);
2445 r
= dm_register_target(&pool_target
);
2447 dm_unregister_target(&thin_target
);
2452 static void dm_thin_exit(void)
2454 dm_unregister_target(&thin_target
);
2455 dm_unregister_target(&pool_target
);
2458 module_init(dm_thin_init
);
2459 module_exit(dm_thin_exit
);
2461 MODULE_DESCRIPTION(DM_NAME
"device-mapper thin provisioning target");
2462 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2463 MODULE_LICENSE("GPL");