2 * Copyright (C) 2011-2012 Red Hat UK.
4 * This file is released under the GPL.
7 #include "dm-thin-metadata.h"
8 #include "dm-bio-prison.h"
11 #include <linux/device-mapper.h>
12 #include <linux/dm-io.h>
13 #include <linux/dm-kcopyd.h>
14 #include <linux/list.h>
15 #include <linux/init.h>
16 #include <linux/module.h>
17 #include <linux/slab.h>
19 #define DM_MSG_PREFIX "thin"
24 #define ENDIO_HOOK_POOL_SIZE 1024
25 #define MAPPING_POOL_SIZE 1024
26 #define PRISON_CELLS 1024
27 #define COMMIT_PERIOD HZ
29 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle
,
30 "A percentage of time allocated for copy on write");
33 * The block size of the device holding pool data must be
34 * between 64KB and 1GB.
36 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
37 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
40 * Device id is restricted to 24 bits.
42 #define MAX_DEV_ID ((1 << 24) - 1)
45 * How do we handle breaking sharing of data blocks?
46 * =================================================
48 * We use a standard copy-on-write btree to store the mappings for the
49 * devices (note I'm talking about copy-on-write of the metadata here, not
50 * the data). When you take an internal snapshot you clone the root node
51 * of the origin btree. After this there is no concept of an origin or a
52 * snapshot. They are just two device trees that happen to point to the
55 * When we get a write in we decide if it's to a shared data block using
56 * some timestamp magic. If it is, we have to break sharing.
58 * Let's say we write to a shared block in what was the origin. The
61 * i) plug io further to this physical block. (see bio_prison code).
63 * ii) quiesce any read io to that shared data block. Obviously
64 * including all devices that share this block. (see dm_deferred_set code)
66 * iii) copy the data block to a newly allocate block. This step can be
67 * missed out if the io covers the block. (schedule_copy).
69 * iv) insert the new mapping into the origin's btree
70 * (process_prepared_mapping). This act of inserting breaks some
71 * sharing of btree nodes between the two devices. Breaking sharing only
72 * effects the btree of that specific device. Btrees for the other
73 * devices that share the block never change. The btree for the origin
74 * device as it was after the last commit is untouched, ie. we're using
75 * persistent data structures in the functional programming sense.
77 * v) unplug io to this physical block, including the io that triggered
78 * the breaking of sharing.
80 * Steps (ii) and (iii) occur in parallel.
82 * The metadata _doesn't_ need to be committed before the io continues. We
83 * get away with this because the io is always written to a _new_ block.
84 * If there's a crash, then:
86 * - The origin mapping will point to the old origin block (the shared
87 * one). This will contain the data as it was before the io that triggered
88 * the breaking of sharing came in.
90 * - The snap mapping still points to the old block. As it would after
93 * The downside of this scheme is the timestamp magic isn't perfect, and
94 * will continue to think that data block in the snapshot device is shared
95 * even after the write to the origin has broken sharing. I suspect data
96 * blocks will typically be shared by many different devices, so we're
97 * breaking sharing n + 1 times, rather than n, where n is the number of
98 * devices that reference this data block. At the moment I think the
99 * benefits far, far outweigh the disadvantages.
102 /*----------------------------------------------------------------*/
107 static void build_data_key(struct dm_thin_device
*td
,
108 dm_block_t b
, struct dm_cell_key
*key
)
111 key
->dev
= dm_thin_dev_id(td
);
115 static void build_virtual_key(struct dm_thin_device
*td
, dm_block_t b
,
116 struct dm_cell_key
*key
)
119 key
->dev
= dm_thin_dev_id(td
);
123 /*----------------------------------------------------------------*/
126 * A pool device ties together a metadata device and a data device. It
127 * also provides the interface for creating and destroying internal
130 struct dm_thin_new_mapping
;
133 * The pool runs in 3 modes. Ordered in degraded order for comparisons.
136 PM_WRITE
, /* metadata may be changed */
137 PM_READ_ONLY
, /* metadata may not be changed */
138 PM_FAIL
, /* all I/O fails */
141 struct pool_features
{
144 bool zero_new_blocks
:1;
145 bool discard_enabled
:1;
146 bool discard_passdown
:1;
150 typedef void (*process_bio_fn
)(struct thin_c
*tc
, struct bio
*bio
);
151 typedef void (*process_mapping_fn
)(struct dm_thin_new_mapping
*m
);
154 struct list_head list
;
155 struct dm_target
*ti
; /* Only set if a pool target is bound */
157 struct mapped_device
*pool_md
;
158 struct block_device
*md_dev
;
159 struct dm_pool_metadata
*pmd
;
161 dm_block_t low_water_blocks
;
162 uint32_t sectors_per_block
;
163 int sectors_per_block_shift
;
165 struct pool_features pf
;
166 unsigned low_water_triggered
:1; /* A dm event has been sent */
167 unsigned no_free_space
:1; /* A -ENOSPC warning has been issued */
169 struct dm_bio_prison
*prison
;
170 struct dm_kcopyd_client
*copier
;
172 struct workqueue_struct
*wq
;
173 struct work_struct worker
;
174 struct delayed_work waker
;
176 unsigned long last_commit_jiffies
;
180 struct bio_list deferred_bios
;
181 struct bio_list deferred_flush_bios
;
182 struct list_head prepared_mappings
;
183 struct list_head prepared_discards
;
185 struct bio_list retry_on_resume_list
;
187 struct dm_deferred_set
*shared_read_ds
;
188 struct dm_deferred_set
*all_io_ds
;
190 struct dm_thin_new_mapping
*next_mapping
;
191 mempool_t
*mapping_pool
;
193 process_bio_fn process_bio
;
194 process_bio_fn process_discard
;
196 process_mapping_fn process_prepared_mapping
;
197 process_mapping_fn process_prepared_discard
;
200 static enum pool_mode
get_pool_mode(struct pool
*pool
);
201 static void set_pool_mode(struct pool
*pool
, enum pool_mode mode
);
204 * Target context for a pool.
207 struct dm_target
*ti
;
209 struct dm_dev
*data_dev
;
210 struct dm_dev
*metadata_dev
;
211 struct dm_target_callbacks callbacks
;
213 dm_block_t low_water_blocks
;
214 struct pool_features requested_pf
; /* Features requested during table load */
215 struct pool_features adjusted_pf
; /* Features used after adjusting for constituent devices */
219 * Target context for a thin.
222 struct dm_dev
*pool_dev
;
223 struct dm_dev
*origin_dev
;
227 struct dm_thin_device
*td
;
230 /*----------------------------------------------------------------*/
233 * wake_worker() is used when new work is queued and when pool_resume is
234 * ready to continue deferred IO processing.
236 static void wake_worker(struct pool
*pool
)
238 queue_work(pool
->wq
, &pool
->worker
);
241 /*----------------------------------------------------------------*/
243 static int bio_detain(struct pool
*pool
, struct dm_cell_key
*key
, struct bio
*bio
,
244 struct dm_bio_prison_cell
**cell_result
)
247 struct dm_bio_prison_cell
*cell_prealloc
;
250 * Allocate a cell from the prison's mempool.
251 * This might block but it can't fail.
253 cell_prealloc
= dm_bio_prison_alloc_cell(pool
->prison
, GFP_NOIO
);
255 r
= dm_bio_detain(pool
->prison
, key
, bio
, cell_prealloc
, cell_result
);
258 * We reused an old cell; we can get rid of
261 dm_bio_prison_free_cell(pool
->prison
, cell_prealloc
);
266 static void cell_release(struct pool
*pool
,
267 struct dm_bio_prison_cell
*cell
,
268 struct bio_list
*bios
)
270 dm_cell_release(pool
->prison
, cell
, bios
);
271 dm_bio_prison_free_cell(pool
->prison
, cell
);
274 static void cell_release_no_holder(struct pool
*pool
,
275 struct dm_bio_prison_cell
*cell
,
276 struct bio_list
*bios
)
278 dm_cell_release_no_holder(pool
->prison
, cell
, bios
);
279 dm_bio_prison_free_cell(pool
->prison
, cell
);
282 static void cell_defer_no_holder_no_free(struct thin_c
*tc
,
283 struct dm_bio_prison_cell
*cell
)
285 struct pool
*pool
= tc
->pool
;
288 spin_lock_irqsave(&pool
->lock
, flags
);
289 dm_cell_release_no_holder(pool
->prison
, cell
, &pool
->deferred_bios
);
290 spin_unlock_irqrestore(&pool
->lock
, flags
);
295 static void cell_error(struct pool
*pool
,
296 struct dm_bio_prison_cell
*cell
)
298 dm_cell_error(pool
->prison
, cell
);
299 dm_bio_prison_free_cell(pool
->prison
, cell
);
302 /*----------------------------------------------------------------*/
305 * A global list of pools that uses a struct mapped_device as a key.
307 static struct dm_thin_pool_table
{
309 struct list_head pools
;
310 } dm_thin_pool_table
;
312 static void pool_table_init(void)
314 mutex_init(&dm_thin_pool_table
.mutex
);
315 INIT_LIST_HEAD(&dm_thin_pool_table
.pools
);
318 static void __pool_table_insert(struct pool
*pool
)
320 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
321 list_add(&pool
->list
, &dm_thin_pool_table
.pools
);
324 static void __pool_table_remove(struct pool
*pool
)
326 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
327 list_del(&pool
->list
);
330 static struct pool
*__pool_table_lookup(struct mapped_device
*md
)
332 struct pool
*pool
= NULL
, *tmp
;
334 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
336 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
337 if (tmp
->pool_md
== md
) {
346 static struct pool
*__pool_table_lookup_metadata_dev(struct block_device
*md_dev
)
348 struct pool
*pool
= NULL
, *tmp
;
350 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
352 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
353 if (tmp
->md_dev
== md_dev
) {
362 /*----------------------------------------------------------------*/
364 struct dm_thin_endio_hook
{
366 struct dm_deferred_entry
*shared_read_entry
;
367 struct dm_deferred_entry
*all_io_entry
;
368 struct dm_thin_new_mapping
*overwrite_mapping
;
371 static void __requeue_bio_list(struct thin_c
*tc
, struct bio_list
*master
)
374 struct bio_list bios
;
376 bio_list_init(&bios
);
377 bio_list_merge(&bios
, master
);
378 bio_list_init(master
);
380 while ((bio
= bio_list_pop(&bios
))) {
381 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
384 bio_endio(bio
, DM_ENDIO_REQUEUE
);
386 bio_list_add(master
, bio
);
390 static void requeue_io(struct thin_c
*tc
)
392 struct pool
*pool
= tc
->pool
;
395 spin_lock_irqsave(&pool
->lock
, flags
);
396 __requeue_bio_list(tc
, &pool
->deferred_bios
);
397 __requeue_bio_list(tc
, &pool
->retry_on_resume_list
);
398 spin_unlock_irqrestore(&pool
->lock
, flags
);
402 * This section of code contains the logic for processing a thin device's IO.
403 * Much of the code depends on pool object resources (lists, workqueues, etc)
404 * but most is exclusively called from the thin target rather than the thin-pool
408 static bool block_size_is_power_of_two(struct pool
*pool
)
410 return pool
->sectors_per_block_shift
>= 0;
413 static dm_block_t
get_bio_block(struct thin_c
*tc
, struct bio
*bio
)
415 struct pool
*pool
= tc
->pool
;
416 sector_t block_nr
= bio
->bi_sector
;
418 if (block_size_is_power_of_two(pool
))
419 block_nr
>>= pool
->sectors_per_block_shift
;
421 (void) sector_div(block_nr
, pool
->sectors_per_block
);
426 static void remap(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
)
428 struct pool
*pool
= tc
->pool
;
429 sector_t bi_sector
= bio
->bi_sector
;
431 bio
->bi_bdev
= tc
->pool_dev
->bdev
;
432 if (block_size_is_power_of_two(pool
))
433 bio
->bi_sector
= (block
<< pool
->sectors_per_block_shift
) |
434 (bi_sector
& (pool
->sectors_per_block
- 1));
436 bio
->bi_sector
= (block
* pool
->sectors_per_block
) +
437 sector_div(bi_sector
, pool
->sectors_per_block
);
440 static void remap_to_origin(struct thin_c
*tc
, struct bio
*bio
)
442 bio
->bi_bdev
= tc
->origin_dev
->bdev
;
445 static int bio_triggers_commit(struct thin_c
*tc
, struct bio
*bio
)
447 return (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) &&
448 dm_thin_changed_this_transaction(tc
->td
);
451 static void inc_all_io_entry(struct pool
*pool
, struct bio
*bio
)
453 struct dm_thin_endio_hook
*h
;
455 if (bio
->bi_rw
& REQ_DISCARD
)
458 h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
459 h
->all_io_entry
= dm_deferred_entry_inc(pool
->all_io_ds
);
462 static void issue(struct thin_c
*tc
, struct bio
*bio
)
464 struct pool
*pool
= tc
->pool
;
467 if (!bio_triggers_commit(tc
, bio
)) {
468 generic_make_request(bio
);
473 * Complete bio with an error if earlier I/O caused changes to
474 * the metadata that can't be committed e.g, due to I/O errors
475 * on the metadata device.
477 if (dm_thin_aborted_changes(tc
->td
)) {
483 * Batch together any bios that trigger commits and then issue a
484 * single commit for them in process_deferred_bios().
486 spin_lock_irqsave(&pool
->lock
, flags
);
487 bio_list_add(&pool
->deferred_flush_bios
, bio
);
488 spin_unlock_irqrestore(&pool
->lock
, flags
);
491 static void remap_to_origin_and_issue(struct thin_c
*tc
, struct bio
*bio
)
493 remap_to_origin(tc
, bio
);
497 static void remap_and_issue(struct thin_c
*tc
, struct bio
*bio
,
500 remap(tc
, bio
, block
);
504 /*----------------------------------------------------------------*/
507 * Bio endio functions.
509 struct dm_thin_new_mapping
{
510 struct list_head list
;
514 unsigned pass_discard
:1;
517 dm_block_t virt_block
;
518 dm_block_t data_block
;
519 struct dm_bio_prison_cell
*cell
, *cell2
;
523 * If the bio covers the whole area of a block then we can avoid
524 * zeroing or copying. Instead this bio is hooked. The bio will
525 * still be in the cell, so care has to be taken to avoid issuing
529 bio_end_io_t
*saved_bi_end_io
;
532 static void __maybe_add_mapping(struct dm_thin_new_mapping
*m
)
534 struct pool
*pool
= m
->tc
->pool
;
536 if (m
->quiesced
&& m
->prepared
) {
537 list_add(&m
->list
, &pool
->prepared_mappings
);
542 static void copy_complete(int read_err
, unsigned long write_err
, void *context
)
545 struct dm_thin_new_mapping
*m
= context
;
546 struct pool
*pool
= m
->tc
->pool
;
548 m
->err
= read_err
|| write_err
? -EIO
: 0;
550 spin_lock_irqsave(&pool
->lock
, flags
);
552 __maybe_add_mapping(m
);
553 spin_unlock_irqrestore(&pool
->lock
, flags
);
556 static void overwrite_endio(struct bio
*bio
, int err
)
559 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
560 struct dm_thin_new_mapping
*m
= h
->overwrite_mapping
;
561 struct pool
*pool
= m
->tc
->pool
;
565 spin_lock_irqsave(&pool
->lock
, flags
);
567 __maybe_add_mapping(m
);
568 spin_unlock_irqrestore(&pool
->lock
, flags
);
571 /*----------------------------------------------------------------*/
578 * Prepared mapping jobs.
582 * This sends the bios in the cell back to the deferred_bios list.
584 static void cell_defer(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
586 struct pool
*pool
= tc
->pool
;
589 spin_lock_irqsave(&pool
->lock
, flags
);
590 cell_release(pool
, cell
, &pool
->deferred_bios
);
591 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
597 * Same as cell_defer above, except it omits the original holder of the cell.
599 static void cell_defer_no_holder(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
601 struct pool
*pool
= tc
->pool
;
604 spin_lock_irqsave(&pool
->lock
, flags
);
605 cell_release_no_holder(pool
, cell
, &pool
->deferred_bios
);
606 spin_unlock_irqrestore(&pool
->lock
, flags
);
611 static void process_prepared_mapping_fail(struct dm_thin_new_mapping
*m
)
614 m
->bio
->bi_end_io
= m
->saved_bi_end_io
;
615 cell_error(m
->tc
->pool
, m
->cell
);
617 mempool_free(m
, m
->tc
->pool
->mapping_pool
);
620 static void process_prepared_mapping(struct dm_thin_new_mapping
*m
)
622 struct thin_c
*tc
= m
->tc
;
623 struct pool
*pool
= tc
->pool
;
629 bio
->bi_end_io
= m
->saved_bi_end_io
;
632 cell_error(pool
, m
->cell
);
637 * Commit the prepared block into the mapping btree.
638 * Any I/O for this block arriving after this point will get
639 * remapped to it directly.
641 r
= dm_thin_insert_block(tc
->td
, m
->virt_block
, m
->data_block
);
643 DMERR_LIMIT("dm_thin_insert_block() failed");
644 cell_error(pool
, m
->cell
);
649 * Release any bios held while the block was being provisioned.
650 * If we are processing a write bio that completely covers the block,
651 * we already processed it so can ignore it now when processing
652 * the bios in the cell.
655 cell_defer_no_holder(tc
, m
->cell
);
658 cell_defer(tc
, m
->cell
);
662 mempool_free(m
, pool
->mapping_pool
);
665 static void process_prepared_discard_fail(struct dm_thin_new_mapping
*m
)
667 struct thin_c
*tc
= m
->tc
;
669 bio_io_error(m
->bio
);
670 cell_defer_no_holder(tc
, m
->cell
);
671 cell_defer_no_holder(tc
, m
->cell2
);
672 mempool_free(m
, tc
->pool
->mapping_pool
);
675 static void process_prepared_discard_passdown(struct dm_thin_new_mapping
*m
)
677 struct thin_c
*tc
= m
->tc
;
679 inc_all_io_entry(tc
->pool
, m
->bio
);
680 cell_defer_no_holder(tc
, m
->cell
);
681 cell_defer_no_holder(tc
, m
->cell2
);
684 remap_and_issue(tc
, m
->bio
, m
->data_block
);
686 bio_endio(m
->bio
, 0);
688 mempool_free(m
, tc
->pool
->mapping_pool
);
691 static void process_prepared_discard(struct dm_thin_new_mapping
*m
)
694 struct thin_c
*tc
= m
->tc
;
696 r
= dm_thin_remove_block(tc
->td
, m
->virt_block
);
698 DMERR_LIMIT("dm_thin_remove_block() failed");
700 process_prepared_discard_passdown(m
);
703 static void process_prepared(struct pool
*pool
, struct list_head
*head
,
704 process_mapping_fn
*fn
)
707 struct list_head maps
;
708 struct dm_thin_new_mapping
*m
, *tmp
;
710 INIT_LIST_HEAD(&maps
);
711 spin_lock_irqsave(&pool
->lock
, flags
);
712 list_splice_init(head
, &maps
);
713 spin_unlock_irqrestore(&pool
->lock
, flags
);
715 list_for_each_entry_safe(m
, tmp
, &maps
, list
)
722 static int io_overlaps_block(struct pool
*pool
, struct bio
*bio
)
724 return bio
->bi_size
== (pool
->sectors_per_block
<< SECTOR_SHIFT
);
727 static int io_overwrites_block(struct pool
*pool
, struct bio
*bio
)
729 return (bio_data_dir(bio
) == WRITE
) &&
730 io_overlaps_block(pool
, bio
);
733 static void save_and_set_endio(struct bio
*bio
, bio_end_io_t
**save
,
736 *save
= bio
->bi_end_io
;
740 static int ensure_next_mapping(struct pool
*pool
)
742 if (pool
->next_mapping
)
745 pool
->next_mapping
= mempool_alloc(pool
->mapping_pool
, GFP_ATOMIC
);
747 return pool
->next_mapping
? 0 : -ENOMEM
;
750 static struct dm_thin_new_mapping
*get_next_mapping(struct pool
*pool
)
752 struct dm_thin_new_mapping
*r
= pool
->next_mapping
;
754 BUG_ON(!pool
->next_mapping
);
756 pool
->next_mapping
= NULL
;
761 static void schedule_copy(struct thin_c
*tc
, dm_block_t virt_block
,
762 struct dm_dev
*origin
, dm_block_t data_origin
,
763 dm_block_t data_dest
,
764 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
767 struct pool
*pool
= tc
->pool
;
768 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
770 INIT_LIST_HEAD(&m
->list
);
774 m
->virt_block
= virt_block
;
775 m
->data_block
= data_dest
;
780 if (!dm_deferred_set_add_work(pool
->shared_read_ds
, &m
->list
))
784 * IO to pool_dev remaps to the pool target's data_dev.
786 * If the whole block of data is being overwritten, we can issue the
787 * bio immediately. Otherwise we use kcopyd to clone the data first.
789 if (io_overwrites_block(pool
, bio
)) {
790 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
792 h
->overwrite_mapping
= m
;
794 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
795 inc_all_io_entry(pool
, bio
);
796 remap_and_issue(tc
, bio
, data_dest
);
798 struct dm_io_region from
, to
;
800 from
.bdev
= origin
->bdev
;
801 from
.sector
= data_origin
* pool
->sectors_per_block
;
802 from
.count
= pool
->sectors_per_block
;
804 to
.bdev
= tc
->pool_dev
->bdev
;
805 to
.sector
= data_dest
* pool
->sectors_per_block
;
806 to
.count
= pool
->sectors_per_block
;
808 r
= dm_kcopyd_copy(pool
->copier
, &from
, 1, &to
,
809 0, copy_complete
, m
);
811 mempool_free(m
, pool
->mapping_pool
);
812 DMERR_LIMIT("dm_kcopyd_copy() failed");
813 cell_error(pool
, cell
);
818 static void schedule_internal_copy(struct thin_c
*tc
, dm_block_t virt_block
,
819 dm_block_t data_origin
, dm_block_t data_dest
,
820 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
822 schedule_copy(tc
, virt_block
, tc
->pool_dev
,
823 data_origin
, data_dest
, cell
, bio
);
826 static void schedule_external_copy(struct thin_c
*tc
, dm_block_t virt_block
,
827 dm_block_t data_dest
,
828 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
830 schedule_copy(tc
, virt_block
, tc
->origin_dev
,
831 virt_block
, data_dest
, cell
, bio
);
834 static void schedule_zero(struct thin_c
*tc
, dm_block_t virt_block
,
835 dm_block_t data_block
, struct dm_bio_prison_cell
*cell
,
838 struct pool
*pool
= tc
->pool
;
839 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
841 INIT_LIST_HEAD(&m
->list
);
845 m
->virt_block
= virt_block
;
846 m
->data_block
= data_block
;
852 * If the whole block of data is being overwritten or we are not
853 * zeroing pre-existing data, we can issue the bio immediately.
854 * Otherwise we use kcopyd to zero the data first.
856 if (!pool
->pf
.zero_new_blocks
)
857 process_prepared_mapping(m
);
859 else if (io_overwrites_block(pool
, bio
)) {
860 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
862 h
->overwrite_mapping
= m
;
864 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
865 inc_all_io_entry(pool
, bio
);
866 remap_and_issue(tc
, bio
, data_block
);
869 struct dm_io_region to
;
871 to
.bdev
= tc
->pool_dev
->bdev
;
872 to
.sector
= data_block
* pool
->sectors_per_block
;
873 to
.count
= pool
->sectors_per_block
;
875 r
= dm_kcopyd_zero(pool
->copier
, 1, &to
, 0, copy_complete
, m
);
877 mempool_free(m
, pool
->mapping_pool
);
878 DMERR_LIMIT("dm_kcopyd_zero() failed");
879 cell_error(pool
, cell
);
884 static int commit(struct pool
*pool
)
888 r
= dm_pool_commit_metadata(pool
->pmd
);
890 DMERR_LIMIT("commit failed: error = %d", r
);
896 * A non-zero return indicates read_only or fail_io mode.
897 * Many callers don't care about the return value.
899 static int commit_or_fallback(struct pool
*pool
)
903 if (get_pool_mode(pool
) != PM_WRITE
)
908 set_pool_mode(pool
, PM_READ_ONLY
);
913 static int alloc_data_block(struct thin_c
*tc
, dm_block_t
*result
)
916 dm_block_t free_blocks
;
918 struct pool
*pool
= tc
->pool
;
920 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
924 if (free_blocks
<= pool
->low_water_blocks
&& !pool
->low_water_triggered
) {
925 DMWARN("%s: reached low water mark for data device: sending event.",
926 dm_device_name(pool
->pool_md
));
927 spin_lock_irqsave(&pool
->lock
, flags
);
928 pool
->low_water_triggered
= 1;
929 spin_unlock_irqrestore(&pool
->lock
, flags
);
930 dm_table_event(pool
->ti
->table
);
934 if (pool
->no_free_space
)
938 * Try to commit to see if that will free up some
941 (void) commit_or_fallback(pool
);
943 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
948 * If we still have no space we set a flag to avoid
949 * doing all this checking and return -ENOSPC.
952 DMWARN("%s: no free space available.",
953 dm_device_name(pool
->pool_md
));
954 spin_lock_irqsave(&pool
->lock
, flags
);
955 pool
->no_free_space
= 1;
956 spin_unlock_irqrestore(&pool
->lock
, flags
);
962 r
= dm_pool_alloc_data_block(pool
->pmd
, result
);
970 * If we have run out of space, queue bios until the device is
971 * resumed, presumably after having been reloaded with more space.
973 static void retry_on_resume(struct bio
*bio
)
975 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
976 struct thin_c
*tc
= h
->tc
;
977 struct pool
*pool
= tc
->pool
;
980 spin_lock_irqsave(&pool
->lock
, flags
);
981 bio_list_add(&pool
->retry_on_resume_list
, bio
);
982 spin_unlock_irqrestore(&pool
->lock
, flags
);
985 static void no_space(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
988 struct bio_list bios
;
990 bio_list_init(&bios
);
991 cell_release(pool
, cell
, &bios
);
993 while ((bio
= bio_list_pop(&bios
)))
994 retry_on_resume(bio
);
997 static void process_discard(struct thin_c
*tc
, struct bio
*bio
)
1000 unsigned long flags
;
1001 struct pool
*pool
= tc
->pool
;
1002 struct dm_bio_prison_cell
*cell
, *cell2
;
1003 struct dm_cell_key key
, key2
;
1004 dm_block_t block
= get_bio_block(tc
, bio
);
1005 struct dm_thin_lookup_result lookup_result
;
1006 struct dm_thin_new_mapping
*m
;
1008 build_virtual_key(tc
->td
, block
, &key
);
1009 if (bio_detain(tc
->pool
, &key
, bio
, &cell
))
1012 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1016 * Check nobody is fiddling with this pool block. This can
1017 * happen if someone's in the process of breaking sharing
1020 build_data_key(tc
->td
, lookup_result
.block
, &key2
);
1021 if (bio_detain(tc
->pool
, &key2
, bio
, &cell2
)) {
1022 cell_defer_no_holder(tc
, cell
);
1026 if (io_overlaps_block(pool
, bio
)) {
1028 * IO may still be going to the destination block. We must
1029 * quiesce before we can do the removal.
1031 m
= get_next_mapping(pool
);
1033 m
->pass_discard
= (!lookup_result
.shared
) && pool
->pf
.discard_passdown
;
1034 m
->virt_block
= block
;
1035 m
->data_block
= lookup_result
.block
;
1041 if (!dm_deferred_set_add_work(pool
->all_io_ds
, &m
->list
)) {
1042 spin_lock_irqsave(&pool
->lock
, flags
);
1043 list_add(&m
->list
, &pool
->prepared_discards
);
1044 spin_unlock_irqrestore(&pool
->lock
, flags
);
1048 inc_all_io_entry(pool
, bio
);
1049 cell_defer_no_holder(tc
, cell
);
1050 cell_defer_no_holder(tc
, cell2
);
1053 * The DM core makes sure that the discard doesn't span
1054 * a block boundary. So we submit the discard of a
1055 * partial block appropriately.
1057 if ((!lookup_result
.shared
) && pool
->pf
.discard_passdown
)
1058 remap_and_issue(tc
, bio
, lookup_result
.block
);
1066 * It isn't provisioned, just forget it.
1068 cell_defer_no_holder(tc
, cell
);
1073 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1075 cell_defer_no_holder(tc
, cell
);
1081 static void break_sharing(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1082 struct dm_cell_key
*key
,
1083 struct dm_thin_lookup_result
*lookup_result
,
1084 struct dm_bio_prison_cell
*cell
)
1087 dm_block_t data_block
;
1089 r
= alloc_data_block(tc
, &data_block
);
1092 schedule_internal_copy(tc
, block
, lookup_result
->block
,
1093 data_block
, cell
, bio
);
1097 no_space(tc
->pool
, cell
);
1101 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1103 cell_error(tc
->pool
, cell
);
1108 static void process_shared_bio(struct thin_c
*tc
, struct bio
*bio
,
1110 struct dm_thin_lookup_result
*lookup_result
)
1112 struct dm_bio_prison_cell
*cell
;
1113 struct pool
*pool
= tc
->pool
;
1114 struct dm_cell_key key
;
1117 * If cell is already occupied, then sharing is already in the process
1118 * of being broken so we have nothing further to do here.
1120 build_data_key(tc
->td
, lookup_result
->block
, &key
);
1121 if (bio_detain(pool
, &key
, bio
, &cell
))
1124 if (bio_data_dir(bio
) == WRITE
&& bio
->bi_size
)
1125 break_sharing(tc
, bio
, block
, &key
, lookup_result
, cell
);
1127 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1129 h
->shared_read_entry
= dm_deferred_entry_inc(pool
->shared_read_ds
);
1130 inc_all_io_entry(pool
, bio
);
1131 cell_defer_no_holder(tc
, cell
);
1133 remap_and_issue(tc
, bio
, lookup_result
->block
);
1137 static void provision_block(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1138 struct dm_bio_prison_cell
*cell
)
1141 dm_block_t data_block
;
1142 struct pool
*pool
= tc
->pool
;
1145 * Remap empty bios (flushes) immediately, without provisioning.
1147 if (!bio
->bi_size
) {
1148 inc_all_io_entry(pool
, bio
);
1149 cell_defer_no_holder(tc
, cell
);
1151 remap_and_issue(tc
, bio
, 0);
1156 * Fill read bios with zeroes and complete them immediately.
1158 if (bio_data_dir(bio
) == READ
) {
1160 cell_defer_no_holder(tc
, cell
);
1165 r
= alloc_data_block(tc
, &data_block
);
1169 schedule_external_copy(tc
, block
, data_block
, cell
, bio
);
1171 schedule_zero(tc
, block
, data_block
, cell
, bio
);
1175 no_space(pool
, cell
);
1179 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1181 set_pool_mode(pool
, PM_READ_ONLY
);
1182 cell_error(pool
, cell
);
1187 static void process_bio(struct thin_c
*tc
, struct bio
*bio
)
1190 struct pool
*pool
= tc
->pool
;
1191 dm_block_t block
= get_bio_block(tc
, bio
);
1192 struct dm_bio_prison_cell
*cell
;
1193 struct dm_cell_key key
;
1194 struct dm_thin_lookup_result lookup_result
;
1197 * If cell is already occupied, then the block is already
1198 * being provisioned so we have nothing further to do here.
1200 build_virtual_key(tc
->td
, block
, &key
);
1201 if (bio_detain(pool
, &key
, bio
, &cell
))
1204 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1207 if (lookup_result
.shared
) {
1208 process_shared_bio(tc
, bio
, block
, &lookup_result
);
1209 cell_defer_no_holder(tc
, cell
); /* FIXME: pass this cell into process_shared? */
1211 inc_all_io_entry(pool
, bio
);
1212 cell_defer_no_holder(tc
, cell
);
1214 remap_and_issue(tc
, bio
, lookup_result
.block
);
1219 if (bio_data_dir(bio
) == READ
&& tc
->origin_dev
) {
1220 inc_all_io_entry(pool
, bio
);
1221 cell_defer_no_holder(tc
, cell
);
1223 remap_to_origin_and_issue(tc
, bio
);
1225 provision_block(tc
, bio
, block
, cell
);
1229 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1231 cell_defer_no_holder(tc
, cell
);
1237 static void process_bio_read_only(struct thin_c
*tc
, struct bio
*bio
)
1240 int rw
= bio_data_dir(bio
);
1241 dm_block_t block
= get_bio_block(tc
, bio
);
1242 struct dm_thin_lookup_result lookup_result
;
1244 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1247 if (lookup_result
.shared
&& (rw
== WRITE
) && bio
->bi_size
)
1250 inc_all_io_entry(tc
->pool
, bio
);
1251 remap_and_issue(tc
, bio
, lookup_result
.block
);
1261 if (tc
->origin_dev
) {
1262 inc_all_io_entry(tc
->pool
, bio
);
1263 remap_to_origin_and_issue(tc
, bio
);
1272 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1279 static void process_bio_fail(struct thin_c
*tc
, struct bio
*bio
)
1285 * FIXME: should we also commit due to size of transaction, measured in
1288 static int need_commit_due_to_time(struct pool
*pool
)
1290 return jiffies
< pool
->last_commit_jiffies
||
1291 jiffies
> pool
->last_commit_jiffies
+ COMMIT_PERIOD
;
1294 static void process_deferred_bios(struct pool
*pool
)
1296 unsigned long flags
;
1298 struct bio_list bios
;
1300 bio_list_init(&bios
);
1302 spin_lock_irqsave(&pool
->lock
, flags
);
1303 bio_list_merge(&bios
, &pool
->deferred_bios
);
1304 bio_list_init(&pool
->deferred_bios
);
1305 spin_unlock_irqrestore(&pool
->lock
, flags
);
1307 while ((bio
= bio_list_pop(&bios
))) {
1308 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1309 struct thin_c
*tc
= h
->tc
;
1312 * If we've got no free new_mapping structs, and processing
1313 * this bio might require one, we pause until there are some
1314 * prepared mappings to process.
1316 if (ensure_next_mapping(pool
)) {
1317 spin_lock_irqsave(&pool
->lock
, flags
);
1318 bio_list_merge(&pool
->deferred_bios
, &bios
);
1319 spin_unlock_irqrestore(&pool
->lock
, flags
);
1324 if (bio
->bi_rw
& REQ_DISCARD
)
1325 pool
->process_discard(tc
, bio
);
1327 pool
->process_bio(tc
, bio
);
1331 * If there are any deferred flush bios, we must commit
1332 * the metadata before issuing them.
1334 bio_list_init(&bios
);
1335 spin_lock_irqsave(&pool
->lock
, flags
);
1336 bio_list_merge(&bios
, &pool
->deferred_flush_bios
);
1337 bio_list_init(&pool
->deferred_flush_bios
);
1338 spin_unlock_irqrestore(&pool
->lock
, flags
);
1340 if (bio_list_empty(&bios
) && !need_commit_due_to_time(pool
))
1343 if (commit_or_fallback(pool
)) {
1344 while ((bio
= bio_list_pop(&bios
)))
1348 pool
->last_commit_jiffies
= jiffies
;
1350 while ((bio
= bio_list_pop(&bios
)))
1351 generic_make_request(bio
);
1354 static void do_worker(struct work_struct
*ws
)
1356 struct pool
*pool
= container_of(ws
, struct pool
, worker
);
1358 process_prepared(pool
, &pool
->prepared_mappings
, &pool
->process_prepared_mapping
);
1359 process_prepared(pool
, &pool
->prepared_discards
, &pool
->process_prepared_discard
);
1360 process_deferred_bios(pool
);
1364 * We want to commit periodically so that not too much
1365 * unwritten data builds up.
1367 static void do_waker(struct work_struct
*ws
)
1369 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
, waker
);
1371 queue_delayed_work(pool
->wq
, &pool
->waker
, COMMIT_PERIOD
);
1374 /*----------------------------------------------------------------*/
1376 static enum pool_mode
get_pool_mode(struct pool
*pool
)
1378 return pool
->pf
.mode
;
1381 static void set_pool_mode(struct pool
*pool
, enum pool_mode mode
)
1385 pool
->pf
.mode
= mode
;
1389 DMERR("switching pool to failure mode");
1390 pool
->process_bio
= process_bio_fail
;
1391 pool
->process_discard
= process_bio_fail
;
1392 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1393 pool
->process_prepared_discard
= process_prepared_discard_fail
;
1397 DMERR("switching pool to read-only mode");
1398 r
= dm_pool_abort_metadata(pool
->pmd
);
1400 DMERR("aborting transaction failed");
1401 set_pool_mode(pool
, PM_FAIL
);
1403 dm_pool_metadata_read_only(pool
->pmd
);
1404 pool
->process_bio
= process_bio_read_only
;
1405 pool
->process_discard
= process_discard
;
1406 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1407 pool
->process_prepared_discard
= process_prepared_discard_passdown
;
1412 pool
->process_bio
= process_bio
;
1413 pool
->process_discard
= process_discard
;
1414 pool
->process_prepared_mapping
= process_prepared_mapping
;
1415 pool
->process_prepared_discard
= process_prepared_discard
;
1420 /*----------------------------------------------------------------*/
1423 * Mapping functions.
1427 * Called only while mapping a thin bio to hand it over to the workqueue.
1429 static void thin_defer_bio(struct thin_c
*tc
, struct bio
*bio
)
1431 unsigned long flags
;
1432 struct pool
*pool
= tc
->pool
;
1434 spin_lock_irqsave(&pool
->lock
, flags
);
1435 bio_list_add(&pool
->deferred_bios
, bio
);
1436 spin_unlock_irqrestore(&pool
->lock
, flags
);
1441 static void thin_hook_bio(struct thin_c
*tc
, struct bio
*bio
)
1443 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1446 h
->shared_read_entry
= NULL
;
1447 h
->all_io_entry
= NULL
;
1448 h
->overwrite_mapping
= NULL
;
1452 * Non-blocking function called from the thin target's map function.
1454 static int thin_bio_map(struct dm_target
*ti
, struct bio
*bio
)
1457 struct thin_c
*tc
= ti
->private;
1458 dm_block_t block
= get_bio_block(tc
, bio
);
1459 struct dm_thin_device
*td
= tc
->td
;
1460 struct dm_thin_lookup_result result
;
1461 struct dm_bio_prison_cell cell1
, cell2
;
1462 struct dm_bio_prison_cell
*cell_result
;
1463 struct dm_cell_key key
;
1465 thin_hook_bio(tc
, bio
);
1467 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
1469 return DM_MAPIO_SUBMITTED
;
1472 if (bio
->bi_rw
& (REQ_DISCARD
| REQ_FLUSH
| REQ_FUA
)) {
1473 thin_defer_bio(tc
, bio
);
1474 return DM_MAPIO_SUBMITTED
;
1477 r
= dm_thin_find_block(td
, block
, 0, &result
);
1480 * Note that we defer readahead too.
1484 if (unlikely(result
.shared
)) {
1486 * We have a race condition here between the
1487 * result.shared value returned by the lookup and
1488 * snapshot creation, which may cause new
1491 * To avoid this always quiesce the origin before
1492 * taking the snap. You want to do this anyway to
1493 * ensure a consistent application view
1496 * More distant ancestors are irrelevant. The
1497 * shared flag will be set in their case.
1499 thin_defer_bio(tc
, bio
);
1500 return DM_MAPIO_SUBMITTED
;
1503 build_virtual_key(tc
->td
, block
, &key
);
1504 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell1
, &cell_result
))
1505 return DM_MAPIO_SUBMITTED
;
1507 build_data_key(tc
->td
, result
.block
, &key
);
1508 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell2
, &cell_result
)) {
1509 cell_defer_no_holder_no_free(tc
, &cell1
);
1510 return DM_MAPIO_SUBMITTED
;
1513 inc_all_io_entry(tc
->pool
, bio
);
1514 cell_defer_no_holder_no_free(tc
, &cell2
);
1515 cell_defer_no_holder_no_free(tc
, &cell1
);
1517 remap(tc
, bio
, result
.block
);
1518 return DM_MAPIO_REMAPPED
;
1521 if (get_pool_mode(tc
->pool
) == PM_READ_ONLY
) {
1523 * This block isn't provisioned, and we have no way
1524 * of doing so. Just error it.
1527 return DM_MAPIO_SUBMITTED
;
1533 * In future, the failed dm_thin_find_block above could
1534 * provide the hint to load the metadata into cache.
1536 thin_defer_bio(tc
, bio
);
1537 return DM_MAPIO_SUBMITTED
;
1541 * Must always call bio_io_error on failure.
1542 * dm_thin_find_block can fail with -EINVAL if the
1543 * pool is switched to fail-io mode.
1546 return DM_MAPIO_SUBMITTED
;
1550 static int pool_is_congested(struct dm_target_callbacks
*cb
, int bdi_bits
)
1553 unsigned long flags
;
1554 struct pool_c
*pt
= container_of(cb
, struct pool_c
, callbacks
);
1556 spin_lock_irqsave(&pt
->pool
->lock
, flags
);
1557 r
= !bio_list_empty(&pt
->pool
->retry_on_resume_list
);
1558 spin_unlock_irqrestore(&pt
->pool
->lock
, flags
);
1561 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1562 r
= bdi_congested(&q
->backing_dev_info
, bdi_bits
);
1568 static void __requeue_bios(struct pool
*pool
)
1570 bio_list_merge(&pool
->deferred_bios
, &pool
->retry_on_resume_list
);
1571 bio_list_init(&pool
->retry_on_resume_list
);
1574 /*----------------------------------------------------------------
1575 * Binding of control targets to a pool object
1576 *--------------------------------------------------------------*/
1577 static bool data_dev_supports_discard(struct pool_c
*pt
)
1579 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1581 return q
&& blk_queue_discard(q
);
1584 static bool is_factor(sector_t block_size
, uint32_t n
)
1586 return !sector_div(block_size
, n
);
1590 * If discard_passdown was enabled verify that the data device
1591 * supports discards. Disable discard_passdown if not.
1593 static void disable_passdown_if_not_supported(struct pool_c
*pt
)
1595 struct pool
*pool
= pt
->pool
;
1596 struct block_device
*data_bdev
= pt
->data_dev
->bdev
;
1597 struct queue_limits
*data_limits
= &bdev_get_queue(data_bdev
)->limits
;
1598 sector_t block_size
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
1599 const char *reason
= NULL
;
1600 char buf
[BDEVNAME_SIZE
];
1602 if (!pt
->adjusted_pf
.discard_passdown
)
1605 if (!data_dev_supports_discard(pt
))
1606 reason
= "discard unsupported";
1608 else if (data_limits
->max_discard_sectors
< pool
->sectors_per_block
)
1609 reason
= "max discard sectors smaller than a block";
1611 else if (data_limits
->discard_granularity
> block_size
)
1612 reason
= "discard granularity larger than a block";
1614 else if (!is_factor(block_size
, data_limits
->discard_granularity
))
1615 reason
= "discard granularity not a factor of block size";
1618 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev
, buf
), reason
);
1619 pt
->adjusted_pf
.discard_passdown
= false;
1623 static int bind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1625 struct pool_c
*pt
= ti
->private;
1628 * We want to make sure that degraded pools are never upgraded.
1630 enum pool_mode old_mode
= pool
->pf
.mode
;
1631 enum pool_mode new_mode
= pt
->adjusted_pf
.mode
;
1633 if (old_mode
> new_mode
)
1634 new_mode
= old_mode
;
1637 pool
->low_water_blocks
= pt
->low_water_blocks
;
1638 pool
->pf
= pt
->adjusted_pf
;
1640 set_pool_mode(pool
, new_mode
);
1645 static void unbind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1651 /*----------------------------------------------------------------
1653 *--------------------------------------------------------------*/
1654 /* Initialize pool features. */
1655 static void pool_features_init(struct pool_features
*pf
)
1657 pf
->mode
= PM_WRITE
;
1658 pf
->zero_new_blocks
= true;
1659 pf
->discard_enabled
= true;
1660 pf
->discard_passdown
= true;
1663 static void __pool_destroy(struct pool
*pool
)
1665 __pool_table_remove(pool
);
1667 if (dm_pool_metadata_close(pool
->pmd
) < 0)
1668 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1670 dm_bio_prison_destroy(pool
->prison
);
1671 dm_kcopyd_client_destroy(pool
->copier
);
1674 destroy_workqueue(pool
->wq
);
1676 if (pool
->next_mapping
)
1677 mempool_free(pool
->next_mapping
, pool
->mapping_pool
);
1678 mempool_destroy(pool
->mapping_pool
);
1679 dm_deferred_set_destroy(pool
->shared_read_ds
);
1680 dm_deferred_set_destroy(pool
->all_io_ds
);
1684 static struct kmem_cache
*_new_mapping_cache
;
1686 static struct pool
*pool_create(struct mapped_device
*pool_md
,
1687 struct block_device
*metadata_dev
,
1688 unsigned long block_size
,
1689 int read_only
, char **error
)
1694 struct dm_pool_metadata
*pmd
;
1695 bool format_device
= read_only
? false : true;
1697 pmd
= dm_pool_metadata_open(metadata_dev
, block_size
, format_device
);
1699 *error
= "Error creating metadata object";
1700 return (struct pool
*)pmd
;
1703 pool
= kmalloc(sizeof(*pool
), GFP_KERNEL
);
1705 *error
= "Error allocating memory for pool";
1706 err_p
= ERR_PTR(-ENOMEM
);
1711 pool
->sectors_per_block
= block_size
;
1712 if (block_size
& (block_size
- 1))
1713 pool
->sectors_per_block_shift
= -1;
1715 pool
->sectors_per_block_shift
= __ffs(block_size
);
1716 pool
->low_water_blocks
= 0;
1717 pool_features_init(&pool
->pf
);
1718 pool
->prison
= dm_bio_prison_create(PRISON_CELLS
);
1719 if (!pool
->prison
) {
1720 *error
= "Error creating pool's bio prison";
1721 err_p
= ERR_PTR(-ENOMEM
);
1725 pool
->copier
= dm_kcopyd_client_create(&dm_kcopyd_throttle
);
1726 if (IS_ERR(pool
->copier
)) {
1727 r
= PTR_ERR(pool
->copier
);
1728 *error
= "Error creating pool's kcopyd client";
1730 goto bad_kcopyd_client
;
1734 * Create singlethreaded workqueue that will service all devices
1735 * that use this metadata.
1737 pool
->wq
= alloc_ordered_workqueue("dm-" DM_MSG_PREFIX
, WQ_MEM_RECLAIM
);
1739 *error
= "Error creating pool's workqueue";
1740 err_p
= ERR_PTR(-ENOMEM
);
1744 INIT_WORK(&pool
->worker
, do_worker
);
1745 INIT_DELAYED_WORK(&pool
->waker
, do_waker
);
1746 spin_lock_init(&pool
->lock
);
1747 bio_list_init(&pool
->deferred_bios
);
1748 bio_list_init(&pool
->deferred_flush_bios
);
1749 INIT_LIST_HEAD(&pool
->prepared_mappings
);
1750 INIT_LIST_HEAD(&pool
->prepared_discards
);
1751 pool
->low_water_triggered
= 0;
1752 pool
->no_free_space
= 0;
1753 bio_list_init(&pool
->retry_on_resume_list
);
1755 pool
->shared_read_ds
= dm_deferred_set_create();
1756 if (!pool
->shared_read_ds
) {
1757 *error
= "Error creating pool's shared read deferred set";
1758 err_p
= ERR_PTR(-ENOMEM
);
1759 goto bad_shared_read_ds
;
1762 pool
->all_io_ds
= dm_deferred_set_create();
1763 if (!pool
->all_io_ds
) {
1764 *error
= "Error creating pool's all io deferred set";
1765 err_p
= ERR_PTR(-ENOMEM
);
1769 pool
->next_mapping
= NULL
;
1770 pool
->mapping_pool
= mempool_create_slab_pool(MAPPING_POOL_SIZE
,
1771 _new_mapping_cache
);
1772 if (!pool
->mapping_pool
) {
1773 *error
= "Error creating pool's mapping mempool";
1774 err_p
= ERR_PTR(-ENOMEM
);
1775 goto bad_mapping_pool
;
1778 pool
->ref_count
= 1;
1779 pool
->last_commit_jiffies
= jiffies
;
1780 pool
->pool_md
= pool_md
;
1781 pool
->md_dev
= metadata_dev
;
1782 __pool_table_insert(pool
);
1787 dm_deferred_set_destroy(pool
->all_io_ds
);
1789 dm_deferred_set_destroy(pool
->shared_read_ds
);
1791 destroy_workqueue(pool
->wq
);
1793 dm_kcopyd_client_destroy(pool
->copier
);
1795 dm_bio_prison_destroy(pool
->prison
);
1799 if (dm_pool_metadata_close(pmd
))
1800 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1805 static void __pool_inc(struct pool
*pool
)
1807 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
1811 static void __pool_dec(struct pool
*pool
)
1813 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
1814 BUG_ON(!pool
->ref_count
);
1815 if (!--pool
->ref_count
)
1816 __pool_destroy(pool
);
1819 static struct pool
*__pool_find(struct mapped_device
*pool_md
,
1820 struct block_device
*metadata_dev
,
1821 unsigned long block_size
, int read_only
,
1822 char **error
, int *created
)
1824 struct pool
*pool
= __pool_table_lookup_metadata_dev(metadata_dev
);
1827 if (pool
->pool_md
!= pool_md
) {
1828 *error
= "metadata device already in use by a pool";
1829 return ERR_PTR(-EBUSY
);
1834 pool
= __pool_table_lookup(pool_md
);
1836 if (pool
->md_dev
!= metadata_dev
) {
1837 *error
= "different pool cannot replace a pool";
1838 return ERR_PTR(-EINVAL
);
1843 pool
= pool_create(pool_md
, metadata_dev
, block_size
, read_only
, error
);
1851 /*----------------------------------------------------------------
1852 * Pool target methods
1853 *--------------------------------------------------------------*/
1854 static void pool_dtr(struct dm_target
*ti
)
1856 struct pool_c
*pt
= ti
->private;
1858 mutex_lock(&dm_thin_pool_table
.mutex
);
1860 unbind_control_target(pt
->pool
, ti
);
1861 __pool_dec(pt
->pool
);
1862 dm_put_device(ti
, pt
->metadata_dev
);
1863 dm_put_device(ti
, pt
->data_dev
);
1866 mutex_unlock(&dm_thin_pool_table
.mutex
);
1869 static int parse_pool_features(struct dm_arg_set
*as
, struct pool_features
*pf
,
1870 struct dm_target
*ti
)
1874 const char *arg_name
;
1876 static struct dm_arg _args
[] = {
1877 {0, 3, "Invalid number of pool feature arguments"},
1881 * No feature arguments supplied.
1886 r
= dm_read_arg_group(_args
, as
, &argc
, &ti
->error
);
1890 while (argc
&& !r
) {
1891 arg_name
= dm_shift_arg(as
);
1894 if (!strcasecmp(arg_name
, "skip_block_zeroing"))
1895 pf
->zero_new_blocks
= false;
1897 else if (!strcasecmp(arg_name
, "ignore_discard"))
1898 pf
->discard_enabled
= false;
1900 else if (!strcasecmp(arg_name
, "no_discard_passdown"))
1901 pf
->discard_passdown
= false;
1903 else if (!strcasecmp(arg_name
, "read_only"))
1904 pf
->mode
= PM_READ_ONLY
;
1907 ti
->error
= "Unrecognised pool feature requested";
1916 static void metadata_low_callback(void *context
)
1918 struct pool
*pool
= context
;
1920 DMWARN("%s: reached low water mark for metadata device: sending event.",
1921 dm_device_name(pool
->pool_md
));
1923 dm_table_event(pool
->ti
->table
);
1926 static sector_t
get_metadata_dev_size(struct block_device
*bdev
)
1928 sector_t metadata_dev_size
= i_size_read(bdev
->bd_inode
) >> SECTOR_SHIFT
;
1929 char buffer
[BDEVNAME_SIZE
];
1931 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS_WARNING
) {
1932 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
1933 bdevname(bdev
, buffer
), THIN_METADATA_MAX_SECTORS
);
1934 metadata_dev_size
= THIN_METADATA_MAX_SECTORS_WARNING
;
1937 return metadata_dev_size
;
1940 static dm_block_t
get_metadata_dev_size_in_blocks(struct block_device
*bdev
)
1942 sector_t metadata_dev_size
= get_metadata_dev_size(bdev
);
1944 sector_div(metadata_dev_size
, THIN_METADATA_BLOCK_SIZE
>> SECTOR_SHIFT
);
1946 return metadata_dev_size
;
1950 * When a metadata threshold is crossed a dm event is triggered, and
1951 * userland should respond by growing the metadata device. We could let
1952 * userland set the threshold, like we do with the data threshold, but I'm
1953 * not sure they know enough to do this well.
1955 static dm_block_t
calc_metadata_threshold(struct pool_c
*pt
)
1958 * 4M is ample for all ops with the possible exception of thin
1959 * device deletion which is harmless if it fails (just retry the
1960 * delete after you've grown the device).
1962 dm_block_t quarter
= get_metadata_dev_size_in_blocks(pt
->metadata_dev
->bdev
) / 4;
1963 return min((dm_block_t
)1024ULL /* 4M */, quarter
);
1967 * thin-pool <metadata dev> <data dev>
1968 * <data block size (sectors)>
1969 * <low water mark (blocks)>
1970 * [<#feature args> [<arg>]*]
1972 * Optional feature arguments are:
1973 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
1974 * ignore_discard: disable discard
1975 * no_discard_passdown: don't pass discards down to the data device
1977 static int pool_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
1979 int r
, pool_created
= 0;
1982 struct pool_features pf
;
1983 struct dm_arg_set as
;
1984 struct dm_dev
*data_dev
;
1985 unsigned long block_size
;
1986 dm_block_t low_water_blocks
;
1987 struct dm_dev
*metadata_dev
;
1988 fmode_t metadata_mode
;
1991 * FIXME Remove validation from scope of lock.
1993 mutex_lock(&dm_thin_pool_table
.mutex
);
1996 ti
->error
= "Invalid argument count";
2005 * Set default pool features.
2007 pool_features_init(&pf
);
2009 dm_consume_args(&as
, 4);
2010 r
= parse_pool_features(&as
, &pf
, ti
);
2014 metadata_mode
= FMODE_READ
| ((pf
.mode
== PM_READ_ONLY
) ? 0 : FMODE_WRITE
);
2015 r
= dm_get_device(ti
, argv
[0], metadata_mode
, &metadata_dev
);
2017 ti
->error
= "Error opening metadata block device";
2022 * Run for the side-effect of possibly issuing a warning if the
2023 * device is too big.
2025 (void) get_metadata_dev_size(metadata_dev
->bdev
);
2027 r
= dm_get_device(ti
, argv
[1], FMODE_READ
| FMODE_WRITE
, &data_dev
);
2029 ti
->error
= "Error getting data device";
2033 if (kstrtoul(argv
[2], 10, &block_size
) || !block_size
||
2034 block_size
< DATA_DEV_BLOCK_SIZE_MIN_SECTORS
||
2035 block_size
> DATA_DEV_BLOCK_SIZE_MAX_SECTORS
||
2036 block_size
& (DATA_DEV_BLOCK_SIZE_MIN_SECTORS
- 1)) {
2037 ti
->error
= "Invalid block size";
2042 if (kstrtoull(argv
[3], 10, (unsigned long long *)&low_water_blocks
)) {
2043 ti
->error
= "Invalid low water mark";
2048 pt
= kzalloc(sizeof(*pt
), GFP_KERNEL
);
2054 pool
= __pool_find(dm_table_get_md(ti
->table
), metadata_dev
->bdev
,
2055 block_size
, pf
.mode
== PM_READ_ONLY
, &ti
->error
, &pool_created
);
2062 * 'pool_created' reflects whether this is the first table load.
2063 * Top level discard support is not allowed to be changed after
2064 * initial load. This would require a pool reload to trigger thin
2067 if (!pool_created
&& pf
.discard_enabled
!= pool
->pf
.discard_enabled
) {
2068 ti
->error
= "Discard support cannot be disabled once enabled";
2070 goto out_flags_changed
;
2075 pt
->metadata_dev
= metadata_dev
;
2076 pt
->data_dev
= data_dev
;
2077 pt
->low_water_blocks
= low_water_blocks
;
2078 pt
->adjusted_pf
= pt
->requested_pf
= pf
;
2079 ti
->num_flush_bios
= 1;
2082 * Only need to enable discards if the pool should pass
2083 * them down to the data device. The thin device's discard
2084 * processing will cause mappings to be removed from the btree.
2086 if (pf
.discard_enabled
&& pf
.discard_passdown
) {
2087 ti
->num_discard_bios
= 1;
2090 * Setting 'discards_supported' circumvents the normal
2091 * stacking of discard limits (this keeps the pool and
2092 * thin devices' discard limits consistent).
2094 ti
->discards_supported
= true;
2095 ti
->discard_zeroes_data_unsupported
= true;
2099 r
= dm_pool_register_metadata_threshold(pt
->pool
->pmd
,
2100 calc_metadata_threshold(pt
),
2101 metadata_low_callback
,
2106 pt
->callbacks
.congested_fn
= pool_is_congested
;
2107 dm_table_add_target_callbacks(ti
->table
, &pt
->callbacks
);
2109 mutex_unlock(&dm_thin_pool_table
.mutex
);
2118 dm_put_device(ti
, data_dev
);
2120 dm_put_device(ti
, metadata_dev
);
2122 mutex_unlock(&dm_thin_pool_table
.mutex
);
2127 static int pool_map(struct dm_target
*ti
, struct bio
*bio
)
2130 struct pool_c
*pt
= ti
->private;
2131 struct pool
*pool
= pt
->pool
;
2132 unsigned long flags
;
2135 * As this is a singleton target, ti->begin is always zero.
2137 spin_lock_irqsave(&pool
->lock
, flags
);
2138 bio
->bi_bdev
= pt
->data_dev
->bdev
;
2139 r
= DM_MAPIO_REMAPPED
;
2140 spin_unlock_irqrestore(&pool
->lock
, flags
);
2145 static int maybe_resize_data_dev(struct dm_target
*ti
, bool *need_commit
)
2148 struct pool_c
*pt
= ti
->private;
2149 struct pool
*pool
= pt
->pool
;
2150 sector_t data_size
= ti
->len
;
2151 dm_block_t sb_data_size
;
2153 *need_commit
= false;
2155 (void) sector_div(data_size
, pool
->sectors_per_block
);
2157 r
= dm_pool_get_data_dev_size(pool
->pmd
, &sb_data_size
);
2159 DMERR("failed to retrieve data device size");
2163 if (data_size
< sb_data_size
) {
2164 DMERR("pool target (%llu blocks) too small: expected %llu",
2165 (unsigned long long)data_size
, sb_data_size
);
2168 } else if (data_size
> sb_data_size
) {
2169 r
= dm_pool_resize_data_dev(pool
->pmd
, data_size
);
2171 DMERR("failed to resize data device");
2172 set_pool_mode(pool
, PM_READ_ONLY
);
2176 *need_commit
= true;
2182 static int maybe_resize_metadata_dev(struct dm_target
*ti
, bool *need_commit
)
2185 struct pool_c
*pt
= ti
->private;
2186 struct pool
*pool
= pt
->pool
;
2187 dm_block_t metadata_dev_size
, sb_metadata_dev_size
;
2189 *need_commit
= false;
2191 metadata_dev_size
= get_metadata_dev_size(pool
->md_dev
);
2193 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &sb_metadata_dev_size
);
2195 DMERR("failed to retrieve data device size");
2199 if (metadata_dev_size
< sb_metadata_dev_size
) {
2200 DMERR("metadata device (%llu sectors) too small: expected %llu",
2201 metadata_dev_size
, sb_metadata_dev_size
);
2204 } else if (metadata_dev_size
> sb_metadata_dev_size
) {
2205 r
= dm_pool_resize_metadata_dev(pool
->pmd
, metadata_dev_size
);
2207 DMERR("failed to resize metadata device");
2211 *need_commit
= true;
2218 * Retrieves the number of blocks of the data device from
2219 * the superblock and compares it to the actual device size,
2220 * thus resizing the data device in case it has grown.
2222 * This both copes with opening preallocated data devices in the ctr
2223 * being followed by a resume
2225 * calling the resume method individually after userspace has
2226 * grown the data device in reaction to a table event.
2228 static int pool_preresume(struct dm_target
*ti
)
2231 bool need_commit1
, need_commit2
;
2232 struct pool_c
*pt
= ti
->private;
2233 struct pool
*pool
= pt
->pool
;
2236 * Take control of the pool object.
2238 r
= bind_control_target(pool
, ti
);
2242 r
= maybe_resize_data_dev(ti
, &need_commit1
);
2246 r
= maybe_resize_metadata_dev(ti
, &need_commit2
);
2250 if (need_commit1
|| need_commit2
)
2251 (void) commit_or_fallback(pool
);
2256 static void pool_resume(struct dm_target
*ti
)
2258 struct pool_c
*pt
= ti
->private;
2259 struct pool
*pool
= pt
->pool
;
2260 unsigned long flags
;
2262 spin_lock_irqsave(&pool
->lock
, flags
);
2263 pool
->low_water_triggered
= 0;
2264 pool
->no_free_space
= 0;
2265 __requeue_bios(pool
);
2266 spin_unlock_irqrestore(&pool
->lock
, flags
);
2268 do_waker(&pool
->waker
.work
);
2271 static void pool_postsuspend(struct dm_target
*ti
)
2273 struct pool_c
*pt
= ti
->private;
2274 struct pool
*pool
= pt
->pool
;
2276 cancel_delayed_work(&pool
->waker
);
2277 flush_workqueue(pool
->wq
);
2278 (void) commit_or_fallback(pool
);
2281 static int check_arg_count(unsigned argc
, unsigned args_required
)
2283 if (argc
!= args_required
) {
2284 DMWARN("Message received with %u arguments instead of %u.",
2285 argc
, args_required
);
2292 static int read_dev_id(char *arg
, dm_thin_id
*dev_id
, int warning
)
2294 if (!kstrtoull(arg
, 10, (unsigned long long *)dev_id
) &&
2295 *dev_id
<= MAX_DEV_ID
)
2299 DMWARN("Message received with invalid device id: %s", arg
);
2304 static int process_create_thin_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2309 r
= check_arg_count(argc
, 2);
2313 r
= read_dev_id(argv
[1], &dev_id
, 1);
2317 r
= dm_pool_create_thin(pool
->pmd
, dev_id
);
2319 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2327 static int process_create_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2330 dm_thin_id origin_dev_id
;
2333 r
= check_arg_count(argc
, 3);
2337 r
= read_dev_id(argv
[1], &dev_id
, 1);
2341 r
= read_dev_id(argv
[2], &origin_dev_id
, 1);
2345 r
= dm_pool_create_snap(pool
->pmd
, dev_id
, origin_dev_id
);
2347 DMWARN("Creation of new snapshot %s of device %s failed.",
2355 static int process_delete_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2360 r
= check_arg_count(argc
, 2);
2364 r
= read_dev_id(argv
[1], &dev_id
, 1);
2368 r
= dm_pool_delete_thin_device(pool
->pmd
, dev_id
);
2370 DMWARN("Deletion of thin device %s failed.", argv
[1]);
2375 static int process_set_transaction_id_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2377 dm_thin_id old_id
, new_id
;
2380 r
= check_arg_count(argc
, 3);
2384 if (kstrtoull(argv
[1], 10, (unsigned long long *)&old_id
)) {
2385 DMWARN("set_transaction_id message: Unrecognised id %s.", argv
[1]);
2389 if (kstrtoull(argv
[2], 10, (unsigned long long *)&new_id
)) {
2390 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv
[2]);
2394 r
= dm_pool_set_metadata_transaction_id(pool
->pmd
, old_id
, new_id
);
2396 DMWARN("Failed to change transaction id from %s to %s.",
2404 static int process_reserve_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2408 r
= check_arg_count(argc
, 1);
2412 (void) commit_or_fallback(pool
);
2414 r
= dm_pool_reserve_metadata_snap(pool
->pmd
);
2416 DMWARN("reserve_metadata_snap message failed.");
2421 static int process_release_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2425 r
= check_arg_count(argc
, 1);
2429 r
= dm_pool_release_metadata_snap(pool
->pmd
);
2431 DMWARN("release_metadata_snap message failed.");
2437 * Messages supported:
2438 * create_thin <dev_id>
2439 * create_snap <dev_id> <origin_id>
2441 * trim <dev_id> <new_size_in_sectors>
2442 * set_transaction_id <current_trans_id> <new_trans_id>
2443 * reserve_metadata_snap
2444 * release_metadata_snap
2446 static int pool_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
2449 struct pool_c
*pt
= ti
->private;
2450 struct pool
*pool
= pt
->pool
;
2452 if (!strcasecmp(argv
[0], "create_thin"))
2453 r
= process_create_thin_mesg(argc
, argv
, pool
);
2455 else if (!strcasecmp(argv
[0], "create_snap"))
2456 r
= process_create_snap_mesg(argc
, argv
, pool
);
2458 else if (!strcasecmp(argv
[0], "delete"))
2459 r
= process_delete_mesg(argc
, argv
, pool
);
2461 else if (!strcasecmp(argv
[0], "set_transaction_id"))
2462 r
= process_set_transaction_id_mesg(argc
, argv
, pool
);
2464 else if (!strcasecmp(argv
[0], "reserve_metadata_snap"))
2465 r
= process_reserve_metadata_snap_mesg(argc
, argv
, pool
);
2467 else if (!strcasecmp(argv
[0], "release_metadata_snap"))
2468 r
= process_release_metadata_snap_mesg(argc
, argv
, pool
);
2471 DMWARN("Unrecognised thin pool target message received: %s", argv
[0]);
2474 (void) commit_or_fallback(pool
);
2479 static void emit_flags(struct pool_features
*pf
, char *result
,
2480 unsigned sz
, unsigned maxlen
)
2482 unsigned count
= !pf
->zero_new_blocks
+ !pf
->discard_enabled
+
2483 !pf
->discard_passdown
+ (pf
->mode
== PM_READ_ONLY
);
2484 DMEMIT("%u ", count
);
2486 if (!pf
->zero_new_blocks
)
2487 DMEMIT("skip_block_zeroing ");
2489 if (!pf
->discard_enabled
)
2490 DMEMIT("ignore_discard ");
2492 if (!pf
->discard_passdown
)
2493 DMEMIT("no_discard_passdown ");
2495 if (pf
->mode
== PM_READ_ONLY
)
2496 DMEMIT("read_only ");
2501 * <transaction id> <used metadata sectors>/<total metadata sectors>
2502 * <used data sectors>/<total data sectors> <held metadata root>
2504 static void pool_status(struct dm_target
*ti
, status_type_t type
,
2505 unsigned status_flags
, char *result
, unsigned maxlen
)
2509 uint64_t transaction_id
;
2510 dm_block_t nr_free_blocks_data
;
2511 dm_block_t nr_free_blocks_metadata
;
2512 dm_block_t nr_blocks_data
;
2513 dm_block_t nr_blocks_metadata
;
2514 dm_block_t held_root
;
2515 char buf
[BDEVNAME_SIZE
];
2516 char buf2
[BDEVNAME_SIZE
];
2517 struct pool_c
*pt
= ti
->private;
2518 struct pool
*pool
= pt
->pool
;
2521 case STATUSTYPE_INFO
:
2522 if (get_pool_mode(pool
) == PM_FAIL
) {
2527 /* Commit to ensure statistics aren't out-of-date */
2528 if (!(status_flags
& DM_STATUS_NOFLUSH_FLAG
) && !dm_suspended(ti
))
2529 (void) commit_or_fallback(pool
);
2531 r
= dm_pool_get_metadata_transaction_id(pool
->pmd
, &transaction_id
);
2533 DMERR("dm_pool_get_metadata_transaction_id returned %d", r
);
2537 r
= dm_pool_get_free_metadata_block_count(pool
->pmd
, &nr_free_blocks_metadata
);
2539 DMERR("dm_pool_get_free_metadata_block_count returned %d", r
);
2543 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &nr_blocks_metadata
);
2545 DMERR("dm_pool_get_metadata_dev_size returned %d", r
);
2549 r
= dm_pool_get_free_block_count(pool
->pmd
, &nr_free_blocks_data
);
2551 DMERR("dm_pool_get_free_block_count returned %d", r
);
2555 r
= dm_pool_get_data_dev_size(pool
->pmd
, &nr_blocks_data
);
2557 DMERR("dm_pool_get_data_dev_size returned %d", r
);
2561 r
= dm_pool_get_metadata_snap(pool
->pmd
, &held_root
);
2563 DMERR("dm_pool_get_metadata_snap returned %d", r
);
2567 DMEMIT("%llu %llu/%llu %llu/%llu ",
2568 (unsigned long long)transaction_id
,
2569 (unsigned long long)(nr_blocks_metadata
- nr_free_blocks_metadata
),
2570 (unsigned long long)nr_blocks_metadata
,
2571 (unsigned long long)(nr_blocks_data
- nr_free_blocks_data
),
2572 (unsigned long long)nr_blocks_data
);
2575 DMEMIT("%llu ", held_root
);
2579 if (pool
->pf
.mode
== PM_READ_ONLY
)
2584 if (!pool
->pf
.discard_enabled
)
2585 DMEMIT("ignore_discard");
2586 else if (pool
->pf
.discard_passdown
)
2587 DMEMIT("discard_passdown");
2589 DMEMIT("no_discard_passdown");
2593 case STATUSTYPE_TABLE
:
2594 DMEMIT("%s %s %lu %llu ",
2595 format_dev_t(buf
, pt
->metadata_dev
->bdev
->bd_dev
),
2596 format_dev_t(buf2
, pt
->data_dev
->bdev
->bd_dev
),
2597 (unsigned long)pool
->sectors_per_block
,
2598 (unsigned long long)pt
->low_water_blocks
);
2599 emit_flags(&pt
->requested_pf
, result
, sz
, maxlen
);
2608 static int pool_iterate_devices(struct dm_target
*ti
,
2609 iterate_devices_callout_fn fn
, void *data
)
2611 struct pool_c
*pt
= ti
->private;
2613 return fn(ti
, pt
->data_dev
, 0, ti
->len
, data
);
2616 static int pool_merge(struct dm_target
*ti
, struct bvec_merge_data
*bvm
,
2617 struct bio_vec
*biovec
, int max_size
)
2619 struct pool_c
*pt
= ti
->private;
2620 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
2622 if (!q
->merge_bvec_fn
)
2625 bvm
->bi_bdev
= pt
->data_dev
->bdev
;
2627 return min(max_size
, q
->merge_bvec_fn(q
, bvm
, biovec
));
2630 static void set_discard_limits(struct pool_c
*pt
, struct queue_limits
*limits
)
2632 struct pool
*pool
= pt
->pool
;
2633 struct queue_limits
*data_limits
;
2635 limits
->max_discard_sectors
= pool
->sectors_per_block
;
2638 * discard_granularity is just a hint, and not enforced.
2640 if (pt
->adjusted_pf
.discard_passdown
) {
2641 data_limits
= &bdev_get_queue(pt
->data_dev
->bdev
)->limits
;
2642 limits
->discard_granularity
= data_limits
->discard_granularity
;
2644 limits
->discard_granularity
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
2647 static void pool_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
2649 struct pool_c
*pt
= ti
->private;
2650 struct pool
*pool
= pt
->pool
;
2652 blk_limits_io_min(limits
, 0);
2653 blk_limits_io_opt(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
2656 * pt->adjusted_pf is a staging area for the actual features to use.
2657 * They get transferred to the live pool in bind_control_target()
2658 * called from pool_preresume().
2660 if (!pt
->adjusted_pf
.discard_enabled
)
2663 disable_passdown_if_not_supported(pt
);
2665 set_discard_limits(pt
, limits
);
2668 static struct target_type pool_target
= {
2669 .name
= "thin-pool",
2670 .features
= DM_TARGET_SINGLETON
| DM_TARGET_ALWAYS_WRITEABLE
|
2671 DM_TARGET_IMMUTABLE
,
2672 .version
= {1, 8, 0},
2673 .module
= THIS_MODULE
,
2677 .postsuspend
= pool_postsuspend
,
2678 .preresume
= pool_preresume
,
2679 .resume
= pool_resume
,
2680 .message
= pool_message
,
2681 .status
= pool_status
,
2682 .merge
= pool_merge
,
2683 .iterate_devices
= pool_iterate_devices
,
2684 .io_hints
= pool_io_hints
,
2687 /*----------------------------------------------------------------
2688 * Thin target methods
2689 *--------------------------------------------------------------*/
2690 static void thin_dtr(struct dm_target
*ti
)
2692 struct thin_c
*tc
= ti
->private;
2694 mutex_lock(&dm_thin_pool_table
.mutex
);
2696 __pool_dec(tc
->pool
);
2697 dm_pool_close_thin_device(tc
->td
);
2698 dm_put_device(ti
, tc
->pool_dev
);
2700 dm_put_device(ti
, tc
->origin_dev
);
2703 mutex_unlock(&dm_thin_pool_table
.mutex
);
2707 * Thin target parameters:
2709 * <pool_dev> <dev_id> [origin_dev]
2711 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2712 * dev_id: the internal device identifier
2713 * origin_dev: a device external to the pool that should act as the origin
2715 * If the pool device has discards disabled, they get disabled for the thin
2718 static int thin_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2722 struct dm_dev
*pool_dev
, *origin_dev
;
2723 struct mapped_device
*pool_md
;
2725 mutex_lock(&dm_thin_pool_table
.mutex
);
2727 if (argc
!= 2 && argc
!= 3) {
2728 ti
->error
= "Invalid argument count";
2733 tc
= ti
->private = kzalloc(sizeof(*tc
), GFP_KERNEL
);
2735 ti
->error
= "Out of memory";
2741 r
= dm_get_device(ti
, argv
[2], FMODE_READ
, &origin_dev
);
2743 ti
->error
= "Error opening origin device";
2744 goto bad_origin_dev
;
2746 tc
->origin_dev
= origin_dev
;
2749 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &pool_dev
);
2751 ti
->error
= "Error opening pool device";
2754 tc
->pool_dev
= pool_dev
;
2756 if (read_dev_id(argv
[1], (unsigned long long *)&tc
->dev_id
, 0)) {
2757 ti
->error
= "Invalid device id";
2762 pool_md
= dm_get_md(tc
->pool_dev
->bdev
->bd_dev
);
2764 ti
->error
= "Couldn't get pool mapped device";
2769 tc
->pool
= __pool_table_lookup(pool_md
);
2771 ti
->error
= "Couldn't find pool object";
2773 goto bad_pool_lookup
;
2775 __pool_inc(tc
->pool
);
2777 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
2778 ti
->error
= "Couldn't open thin device, Pool is in fail mode";
2782 r
= dm_pool_open_thin_device(tc
->pool
->pmd
, tc
->dev_id
, &tc
->td
);
2784 ti
->error
= "Couldn't open thin internal device";
2788 r
= dm_set_target_max_io_len(ti
, tc
->pool
->sectors_per_block
);
2792 ti
->num_flush_bios
= 1;
2793 ti
->flush_supported
= true;
2794 ti
->per_bio_data_size
= sizeof(struct dm_thin_endio_hook
);
2796 /* In case the pool supports discards, pass them on. */
2797 if (tc
->pool
->pf
.discard_enabled
) {
2798 ti
->discards_supported
= true;
2799 ti
->num_discard_bios
= 1;
2800 ti
->discard_zeroes_data_unsupported
= true;
2801 /* Discard bios must be split on a block boundary */
2802 ti
->split_discard_bios
= true;
2807 mutex_unlock(&dm_thin_pool_table
.mutex
);
2812 __pool_dec(tc
->pool
);
2816 dm_put_device(ti
, tc
->pool_dev
);
2819 dm_put_device(ti
, tc
->origin_dev
);
2823 mutex_unlock(&dm_thin_pool_table
.mutex
);
2828 static int thin_map(struct dm_target
*ti
, struct bio
*bio
)
2830 bio
->bi_sector
= dm_target_offset(ti
, bio
->bi_sector
);
2832 return thin_bio_map(ti
, bio
);
2835 static int thin_endio(struct dm_target
*ti
, struct bio
*bio
, int err
)
2837 unsigned long flags
;
2838 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
2839 struct list_head work
;
2840 struct dm_thin_new_mapping
*m
, *tmp
;
2841 struct pool
*pool
= h
->tc
->pool
;
2843 if (h
->shared_read_entry
) {
2844 INIT_LIST_HEAD(&work
);
2845 dm_deferred_entry_dec(h
->shared_read_entry
, &work
);
2847 spin_lock_irqsave(&pool
->lock
, flags
);
2848 list_for_each_entry_safe(m
, tmp
, &work
, list
) {
2851 __maybe_add_mapping(m
);
2853 spin_unlock_irqrestore(&pool
->lock
, flags
);
2856 if (h
->all_io_entry
) {
2857 INIT_LIST_HEAD(&work
);
2858 dm_deferred_entry_dec(h
->all_io_entry
, &work
);
2859 if (!list_empty(&work
)) {
2860 spin_lock_irqsave(&pool
->lock
, flags
);
2861 list_for_each_entry_safe(m
, tmp
, &work
, list
)
2862 list_add(&m
->list
, &pool
->prepared_discards
);
2863 spin_unlock_irqrestore(&pool
->lock
, flags
);
2871 static void thin_postsuspend(struct dm_target
*ti
)
2873 if (dm_noflush_suspending(ti
))
2874 requeue_io((struct thin_c
*)ti
->private);
2878 * <nr mapped sectors> <highest mapped sector>
2880 static void thin_status(struct dm_target
*ti
, status_type_t type
,
2881 unsigned status_flags
, char *result
, unsigned maxlen
)
2885 dm_block_t mapped
, highest
;
2886 char buf
[BDEVNAME_SIZE
];
2887 struct thin_c
*tc
= ti
->private;
2889 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
2898 case STATUSTYPE_INFO
:
2899 r
= dm_thin_get_mapped_count(tc
->td
, &mapped
);
2901 DMERR("dm_thin_get_mapped_count returned %d", r
);
2905 r
= dm_thin_get_highest_mapped_block(tc
->td
, &highest
);
2907 DMERR("dm_thin_get_highest_mapped_block returned %d", r
);
2911 DMEMIT("%llu ", mapped
* tc
->pool
->sectors_per_block
);
2913 DMEMIT("%llu", ((highest
+ 1) *
2914 tc
->pool
->sectors_per_block
) - 1);
2919 case STATUSTYPE_TABLE
:
2921 format_dev_t(buf
, tc
->pool_dev
->bdev
->bd_dev
),
2922 (unsigned long) tc
->dev_id
);
2924 DMEMIT(" %s", format_dev_t(buf
, tc
->origin_dev
->bdev
->bd_dev
));
2935 static int thin_iterate_devices(struct dm_target
*ti
,
2936 iterate_devices_callout_fn fn
, void *data
)
2939 struct thin_c
*tc
= ti
->private;
2940 struct pool
*pool
= tc
->pool
;
2943 * We can't call dm_pool_get_data_dev_size() since that blocks. So
2944 * we follow a more convoluted path through to the pool's target.
2947 return 0; /* nothing is bound */
2949 blocks
= pool
->ti
->len
;
2950 (void) sector_div(blocks
, pool
->sectors_per_block
);
2952 return fn(ti
, tc
->pool_dev
, 0, pool
->sectors_per_block
* blocks
, data
);
2957 static struct target_type thin_target
= {
2959 .version
= {1, 8, 0},
2960 .module
= THIS_MODULE
,
2964 .end_io
= thin_endio
,
2965 .postsuspend
= thin_postsuspend
,
2966 .status
= thin_status
,
2967 .iterate_devices
= thin_iterate_devices
,
2970 /*----------------------------------------------------------------*/
2972 static int __init
dm_thin_init(void)
2978 r
= dm_register_target(&thin_target
);
2982 r
= dm_register_target(&pool_target
);
2984 goto bad_pool_target
;
2988 _new_mapping_cache
= KMEM_CACHE(dm_thin_new_mapping
, 0);
2989 if (!_new_mapping_cache
)
2990 goto bad_new_mapping_cache
;
2994 bad_new_mapping_cache
:
2995 dm_unregister_target(&pool_target
);
2997 dm_unregister_target(&thin_target
);
3002 static void dm_thin_exit(void)
3004 dm_unregister_target(&thin_target
);
3005 dm_unregister_target(&pool_target
);
3007 kmem_cache_destroy(_new_mapping_cache
);
3010 module_init(dm_thin_init
);
3011 module_exit(dm_thin_exit
);
3013 MODULE_DESCRIPTION(DM_NAME
" thin provisioning target");
3014 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3015 MODULE_LICENSE("GPL");