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-v1.h"
11 #include <linux/device-mapper.h>
12 #include <linux/dm-io.h>
13 #include <linux/dm-kcopyd.h>
14 #include <linux/jiffies.h>
15 #include <linux/log2.h>
16 #include <linux/list.h>
17 #include <linux/rculist.h>
18 #include <linux/init.h>
19 #include <linux/module.h>
20 #include <linux/slab.h>
21 #include <linux/vmalloc.h>
22 #include <linux/sort.h>
23 #include <linux/rbtree.h>
25 #define DM_MSG_PREFIX "thin"
30 #define ENDIO_HOOK_POOL_SIZE 1024
31 #define MAPPING_POOL_SIZE 1024
32 #define COMMIT_PERIOD HZ
33 #define NO_SPACE_TIMEOUT_SECS 60
35 static unsigned no_space_timeout_secs
= NO_SPACE_TIMEOUT_SECS
;
37 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle
,
38 "A percentage of time allocated for copy on write");
41 * The block size of the device holding pool data must be
42 * between 64KB and 1GB.
44 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
45 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
48 * Device id is restricted to 24 bits.
50 #define MAX_DEV_ID ((1 << 24) - 1)
53 * How do we handle breaking sharing of data blocks?
54 * =================================================
56 * We use a standard copy-on-write btree to store the mappings for the
57 * devices (note I'm talking about copy-on-write of the metadata here, not
58 * the data). When you take an internal snapshot you clone the root node
59 * of the origin btree. After this there is no concept of an origin or a
60 * snapshot. They are just two device trees that happen to point to the
63 * When we get a write in we decide if it's to a shared data block using
64 * some timestamp magic. If it is, we have to break sharing.
66 * Let's say we write to a shared block in what was the origin. The
69 * i) plug io further to this physical block. (see bio_prison code).
71 * ii) quiesce any read io to that shared data block. Obviously
72 * including all devices that share this block. (see dm_deferred_set code)
74 * iii) copy the data block to a newly allocate block. This step can be
75 * missed out if the io covers the block. (schedule_copy).
77 * iv) insert the new mapping into the origin's btree
78 * (process_prepared_mapping). This act of inserting breaks some
79 * sharing of btree nodes between the two devices. Breaking sharing only
80 * effects the btree of that specific device. Btrees for the other
81 * devices that share the block never change. The btree for the origin
82 * device as it was after the last commit is untouched, ie. we're using
83 * persistent data structures in the functional programming sense.
85 * v) unplug io to this physical block, including the io that triggered
86 * the breaking of sharing.
88 * Steps (ii) and (iii) occur in parallel.
90 * The metadata _doesn't_ need to be committed before the io continues. We
91 * get away with this because the io is always written to a _new_ block.
92 * If there's a crash, then:
94 * - The origin mapping will point to the old origin block (the shared
95 * one). This will contain the data as it was before the io that triggered
96 * the breaking of sharing came in.
98 * - The snap mapping still points to the old block. As it would after
101 * The downside of this scheme is the timestamp magic isn't perfect, and
102 * will continue to think that data block in the snapshot device is shared
103 * even after the write to the origin has broken sharing. I suspect data
104 * blocks will typically be shared by many different devices, so we're
105 * breaking sharing n + 1 times, rather than n, where n is the number of
106 * devices that reference this data block. At the moment I think the
107 * benefits far, far outweigh the disadvantages.
110 /*----------------------------------------------------------------*/
120 static void build_key(struct dm_thin_device
*td
, enum lock_space ls
,
121 dm_block_t b
, dm_block_t e
, struct dm_cell_key
*key
)
123 key
->virtual = (ls
== VIRTUAL
);
124 key
->dev
= dm_thin_dev_id(td
);
125 key
->block_begin
= b
;
129 static void build_data_key(struct dm_thin_device
*td
, dm_block_t b
,
130 struct dm_cell_key
*key
)
132 build_key(td
, PHYSICAL
, b
, b
+ 1llu, key
);
135 static void build_virtual_key(struct dm_thin_device
*td
, dm_block_t b
,
136 struct dm_cell_key
*key
)
138 build_key(td
, VIRTUAL
, b
, b
+ 1llu, key
);
141 /*----------------------------------------------------------------*/
143 #define THROTTLE_THRESHOLD (1 * HZ)
146 struct rw_semaphore lock
;
147 unsigned long threshold
;
148 bool throttle_applied
;
151 static void throttle_init(struct throttle
*t
)
153 init_rwsem(&t
->lock
);
154 t
->throttle_applied
= false;
157 static void throttle_work_start(struct throttle
*t
)
159 t
->threshold
= jiffies
+ THROTTLE_THRESHOLD
;
162 static void throttle_work_update(struct throttle
*t
)
164 if (!t
->throttle_applied
&& jiffies
> t
->threshold
) {
165 down_write(&t
->lock
);
166 t
->throttle_applied
= true;
170 static void throttle_work_complete(struct throttle
*t
)
172 if (t
->throttle_applied
) {
173 t
->throttle_applied
= false;
178 static void throttle_lock(struct throttle
*t
)
183 static void throttle_unlock(struct throttle
*t
)
188 /*----------------------------------------------------------------*/
191 * A pool device ties together a metadata device and a data device. It
192 * also provides the interface for creating and destroying internal
195 struct dm_thin_new_mapping
;
198 * The pool runs in 4 modes. Ordered in degraded order for comparisons.
201 PM_WRITE
, /* metadata may be changed */
202 PM_OUT_OF_DATA_SPACE
, /* metadata may be changed, though data may not be allocated */
203 PM_READ_ONLY
, /* metadata may not be changed */
204 PM_FAIL
, /* all I/O fails */
207 struct pool_features
{
210 bool zero_new_blocks
:1;
211 bool discard_enabled
:1;
212 bool discard_passdown
:1;
213 bool error_if_no_space
:1;
217 typedef void (*process_bio_fn
)(struct thin_c
*tc
, struct bio
*bio
);
218 typedef void (*process_cell_fn
)(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
);
219 typedef void (*process_mapping_fn
)(struct dm_thin_new_mapping
*m
);
221 #define CELL_SORT_ARRAY_SIZE 8192
224 struct list_head list
;
225 struct dm_target
*ti
; /* Only set if a pool target is bound */
227 struct mapped_device
*pool_md
;
228 struct block_device
*md_dev
;
229 struct dm_pool_metadata
*pmd
;
231 dm_block_t low_water_blocks
;
232 uint32_t sectors_per_block
;
233 int sectors_per_block_shift
;
235 struct pool_features pf
;
236 bool low_water_triggered
:1; /* A dm event has been sent */
238 bool out_of_data_space
:1;
240 struct dm_bio_prison
*prison
;
241 struct dm_kcopyd_client
*copier
;
243 struct workqueue_struct
*wq
;
244 struct throttle throttle
;
245 struct work_struct worker
;
246 struct delayed_work waker
;
247 struct delayed_work no_space_timeout
;
249 unsigned long last_commit_jiffies
;
253 struct bio_list deferred_flush_bios
;
254 struct list_head prepared_mappings
;
255 struct list_head prepared_discards
;
256 struct list_head prepared_discards_pt2
;
257 struct list_head active_thins
;
259 struct dm_deferred_set
*shared_read_ds
;
260 struct dm_deferred_set
*all_io_ds
;
262 struct dm_thin_new_mapping
*next_mapping
;
263 mempool_t
*mapping_pool
;
265 process_bio_fn process_bio
;
266 process_bio_fn process_discard
;
268 process_cell_fn process_cell
;
269 process_cell_fn process_discard_cell
;
271 process_mapping_fn process_prepared_mapping
;
272 process_mapping_fn process_prepared_discard
;
273 process_mapping_fn process_prepared_discard_pt2
;
275 struct dm_bio_prison_cell
**cell_sort_array
;
278 static enum pool_mode
get_pool_mode(struct pool
*pool
);
279 static void metadata_operation_failed(struct pool
*pool
, const char *op
, int r
);
282 * Target context for a pool.
285 struct dm_target
*ti
;
287 struct dm_dev
*data_dev
;
288 struct dm_dev
*metadata_dev
;
289 struct dm_target_callbacks callbacks
;
291 dm_block_t low_water_blocks
;
292 struct pool_features requested_pf
; /* Features requested during table load */
293 struct pool_features adjusted_pf
; /* Features used after adjusting for constituent devices */
297 * Target context for a thin.
300 struct list_head list
;
301 struct dm_dev
*pool_dev
;
302 struct dm_dev
*origin_dev
;
303 sector_t origin_size
;
307 struct dm_thin_device
*td
;
308 struct mapped_device
*thin_md
;
312 struct list_head deferred_cells
;
313 struct bio_list deferred_bio_list
;
314 struct bio_list retry_on_resume_list
;
315 struct rb_root sort_bio_list
; /* sorted list of deferred bios */
318 * Ensures the thin is not destroyed until the worker has finished
319 * iterating the active_thins list.
322 struct completion can_destroy
;
325 /*----------------------------------------------------------------*/
327 static bool block_size_is_power_of_two(struct pool
*pool
)
329 return pool
->sectors_per_block_shift
>= 0;
332 static sector_t
block_to_sectors(struct pool
*pool
, dm_block_t b
)
334 return block_size_is_power_of_two(pool
) ?
335 (b
<< pool
->sectors_per_block_shift
) :
336 (b
* pool
->sectors_per_block
);
339 /*----------------------------------------------------------------*/
343 struct blk_plug plug
;
344 struct bio
*parent_bio
;
348 static void begin_discard(struct discard_op
*op
, struct thin_c
*tc
, struct bio
*parent
)
353 blk_start_plug(&op
->plug
);
354 op
->parent_bio
= parent
;
358 static int issue_discard(struct discard_op
*op
, dm_block_t data_b
, dm_block_t data_e
)
360 struct thin_c
*tc
= op
->tc
;
361 sector_t s
= block_to_sectors(tc
->pool
, data_b
);
362 sector_t len
= block_to_sectors(tc
->pool
, data_e
- data_b
);
364 return __blkdev_issue_discard(tc
->pool_dev
->bdev
, s
, len
,
365 GFP_NOWAIT
, 0, &op
->bio
);
368 static void end_discard(struct discard_op
*op
, int r
)
372 * Even if one of the calls to issue_discard failed, we
373 * need to wait for the chain to complete.
375 bio_chain(op
->bio
, op
->parent_bio
);
376 bio_set_op_attrs(op
->bio
, REQ_OP_DISCARD
, 0);
380 blk_finish_plug(&op
->plug
);
383 * Even if r is set, there could be sub discards in flight that we
386 if (r
&& !op
->parent_bio
->bi_status
)
387 op
->parent_bio
->bi_status
= errno_to_blk_status(r
);
388 bio_endio(op
->parent_bio
);
391 /*----------------------------------------------------------------*/
394 * wake_worker() is used when new work is queued and when pool_resume is
395 * ready to continue deferred IO processing.
397 static void wake_worker(struct pool
*pool
)
399 queue_work(pool
->wq
, &pool
->worker
);
402 /*----------------------------------------------------------------*/
404 static int bio_detain(struct pool
*pool
, struct dm_cell_key
*key
, struct bio
*bio
,
405 struct dm_bio_prison_cell
**cell_result
)
408 struct dm_bio_prison_cell
*cell_prealloc
;
411 * Allocate a cell from the prison's mempool.
412 * This might block but it can't fail.
414 cell_prealloc
= dm_bio_prison_alloc_cell(pool
->prison
, GFP_NOIO
);
416 r
= dm_bio_detain(pool
->prison
, key
, bio
, cell_prealloc
, cell_result
);
419 * We reused an old cell; we can get rid of
422 dm_bio_prison_free_cell(pool
->prison
, cell_prealloc
);
427 static void cell_release(struct pool
*pool
,
428 struct dm_bio_prison_cell
*cell
,
429 struct bio_list
*bios
)
431 dm_cell_release(pool
->prison
, cell
, bios
);
432 dm_bio_prison_free_cell(pool
->prison
, cell
);
435 static void cell_visit_release(struct pool
*pool
,
436 void (*fn
)(void *, struct dm_bio_prison_cell
*),
438 struct dm_bio_prison_cell
*cell
)
440 dm_cell_visit_release(pool
->prison
, fn
, context
, cell
);
441 dm_bio_prison_free_cell(pool
->prison
, cell
);
444 static void cell_release_no_holder(struct pool
*pool
,
445 struct dm_bio_prison_cell
*cell
,
446 struct bio_list
*bios
)
448 dm_cell_release_no_holder(pool
->prison
, cell
, bios
);
449 dm_bio_prison_free_cell(pool
->prison
, cell
);
452 static void cell_error_with_code(struct pool
*pool
,
453 struct dm_bio_prison_cell
*cell
, blk_status_t error_code
)
455 dm_cell_error(pool
->prison
, cell
, error_code
);
456 dm_bio_prison_free_cell(pool
->prison
, cell
);
459 static blk_status_t
get_pool_io_error_code(struct pool
*pool
)
461 return pool
->out_of_data_space
? BLK_STS_NOSPC
: BLK_STS_IOERR
;
464 static void cell_error(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
466 cell_error_with_code(pool
, cell
, get_pool_io_error_code(pool
));
469 static void cell_success(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
471 cell_error_with_code(pool
, cell
, 0);
474 static void cell_requeue(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
476 cell_error_with_code(pool
, cell
, BLK_STS_DM_REQUEUE
);
479 /*----------------------------------------------------------------*/
482 * A global list of pools that uses a struct mapped_device as a key.
484 static struct dm_thin_pool_table
{
486 struct list_head pools
;
487 } dm_thin_pool_table
;
489 static void pool_table_init(void)
491 mutex_init(&dm_thin_pool_table
.mutex
);
492 INIT_LIST_HEAD(&dm_thin_pool_table
.pools
);
495 static void __pool_table_insert(struct pool
*pool
)
497 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
498 list_add(&pool
->list
, &dm_thin_pool_table
.pools
);
501 static void __pool_table_remove(struct pool
*pool
)
503 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
504 list_del(&pool
->list
);
507 static struct pool
*__pool_table_lookup(struct mapped_device
*md
)
509 struct pool
*pool
= NULL
, *tmp
;
511 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
513 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
514 if (tmp
->pool_md
== md
) {
523 static struct pool
*__pool_table_lookup_metadata_dev(struct block_device
*md_dev
)
525 struct pool
*pool
= NULL
, *tmp
;
527 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
529 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
530 if (tmp
->md_dev
== md_dev
) {
539 /*----------------------------------------------------------------*/
541 struct dm_thin_endio_hook
{
543 struct dm_deferred_entry
*shared_read_entry
;
544 struct dm_deferred_entry
*all_io_entry
;
545 struct dm_thin_new_mapping
*overwrite_mapping
;
546 struct rb_node rb_node
;
547 struct dm_bio_prison_cell
*cell
;
550 static void __merge_bio_list(struct bio_list
*bios
, struct bio_list
*master
)
552 bio_list_merge(bios
, master
);
553 bio_list_init(master
);
556 static void error_bio_list(struct bio_list
*bios
, blk_status_t error
)
560 while ((bio
= bio_list_pop(bios
))) {
561 bio
->bi_status
= error
;
566 static void error_thin_bio_list(struct thin_c
*tc
, struct bio_list
*master
,
569 struct bio_list bios
;
572 bio_list_init(&bios
);
574 spin_lock_irqsave(&tc
->lock
, flags
);
575 __merge_bio_list(&bios
, master
);
576 spin_unlock_irqrestore(&tc
->lock
, flags
);
578 error_bio_list(&bios
, error
);
581 static void requeue_deferred_cells(struct thin_c
*tc
)
583 struct pool
*pool
= tc
->pool
;
585 struct list_head cells
;
586 struct dm_bio_prison_cell
*cell
, *tmp
;
588 INIT_LIST_HEAD(&cells
);
590 spin_lock_irqsave(&tc
->lock
, flags
);
591 list_splice_init(&tc
->deferred_cells
, &cells
);
592 spin_unlock_irqrestore(&tc
->lock
, flags
);
594 list_for_each_entry_safe(cell
, tmp
, &cells
, user_list
)
595 cell_requeue(pool
, cell
);
598 static void requeue_io(struct thin_c
*tc
)
600 struct bio_list bios
;
603 bio_list_init(&bios
);
605 spin_lock_irqsave(&tc
->lock
, flags
);
606 __merge_bio_list(&bios
, &tc
->deferred_bio_list
);
607 __merge_bio_list(&bios
, &tc
->retry_on_resume_list
);
608 spin_unlock_irqrestore(&tc
->lock
, flags
);
610 error_bio_list(&bios
, BLK_STS_DM_REQUEUE
);
611 requeue_deferred_cells(tc
);
614 static void error_retry_list_with_code(struct pool
*pool
, blk_status_t error
)
619 list_for_each_entry_rcu(tc
, &pool
->active_thins
, list
)
620 error_thin_bio_list(tc
, &tc
->retry_on_resume_list
, error
);
624 static void error_retry_list(struct pool
*pool
)
626 error_retry_list_with_code(pool
, get_pool_io_error_code(pool
));
630 * This section of code contains the logic for processing a thin device's IO.
631 * Much of the code depends on pool object resources (lists, workqueues, etc)
632 * but most is exclusively called from the thin target rather than the thin-pool
636 static dm_block_t
get_bio_block(struct thin_c
*tc
, struct bio
*bio
)
638 struct pool
*pool
= tc
->pool
;
639 sector_t block_nr
= bio
->bi_iter
.bi_sector
;
641 if (block_size_is_power_of_two(pool
))
642 block_nr
>>= pool
->sectors_per_block_shift
;
644 (void) sector_div(block_nr
, pool
->sectors_per_block
);
650 * Returns the _complete_ blocks that this bio covers.
652 static void get_bio_block_range(struct thin_c
*tc
, struct bio
*bio
,
653 dm_block_t
*begin
, dm_block_t
*end
)
655 struct pool
*pool
= tc
->pool
;
656 sector_t b
= bio
->bi_iter
.bi_sector
;
657 sector_t e
= b
+ (bio
->bi_iter
.bi_size
>> SECTOR_SHIFT
);
659 b
+= pool
->sectors_per_block
- 1ull; /* so we round up */
661 if (block_size_is_power_of_two(pool
)) {
662 b
>>= pool
->sectors_per_block_shift
;
663 e
>>= pool
->sectors_per_block_shift
;
665 (void) sector_div(b
, pool
->sectors_per_block
);
666 (void) sector_div(e
, pool
->sectors_per_block
);
670 /* Can happen if the bio is within a single block. */
677 static void remap(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
)
679 struct pool
*pool
= tc
->pool
;
680 sector_t bi_sector
= bio
->bi_iter
.bi_sector
;
682 bio_set_dev(bio
, tc
->pool_dev
->bdev
);
683 if (block_size_is_power_of_two(pool
))
684 bio
->bi_iter
.bi_sector
=
685 (block
<< pool
->sectors_per_block_shift
) |
686 (bi_sector
& (pool
->sectors_per_block
- 1));
688 bio
->bi_iter
.bi_sector
= (block
* pool
->sectors_per_block
) +
689 sector_div(bi_sector
, pool
->sectors_per_block
);
692 static void remap_to_origin(struct thin_c
*tc
, struct bio
*bio
)
694 bio_set_dev(bio
, tc
->origin_dev
->bdev
);
697 static int bio_triggers_commit(struct thin_c
*tc
, struct bio
*bio
)
699 return op_is_flush(bio
->bi_opf
) &&
700 dm_thin_changed_this_transaction(tc
->td
);
703 static void inc_all_io_entry(struct pool
*pool
, struct bio
*bio
)
705 struct dm_thin_endio_hook
*h
;
707 if (bio_op(bio
) == REQ_OP_DISCARD
)
710 h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
711 h
->all_io_entry
= dm_deferred_entry_inc(pool
->all_io_ds
);
714 static void issue(struct thin_c
*tc
, struct bio
*bio
)
716 struct pool
*pool
= tc
->pool
;
719 if (!bio_triggers_commit(tc
, bio
)) {
720 generic_make_request(bio
);
725 * Complete bio with an error if earlier I/O caused changes to
726 * the metadata that can't be committed e.g, due to I/O errors
727 * on the metadata device.
729 if (dm_thin_aborted_changes(tc
->td
)) {
735 * Batch together any bios that trigger commits and then issue a
736 * single commit for them in process_deferred_bios().
738 spin_lock_irqsave(&pool
->lock
, flags
);
739 bio_list_add(&pool
->deferred_flush_bios
, bio
);
740 spin_unlock_irqrestore(&pool
->lock
, flags
);
743 static void remap_to_origin_and_issue(struct thin_c
*tc
, struct bio
*bio
)
745 remap_to_origin(tc
, bio
);
749 static void remap_and_issue(struct thin_c
*tc
, struct bio
*bio
,
752 remap(tc
, bio
, block
);
756 /*----------------------------------------------------------------*/
759 * Bio endio functions.
761 struct dm_thin_new_mapping
{
762 struct list_head list
;
768 * Track quiescing, copying and zeroing preparation actions. When this
769 * counter hits zero the block is prepared and can be inserted into the
772 atomic_t prepare_actions
;
776 dm_block_t virt_begin
, virt_end
;
777 dm_block_t data_block
;
778 struct dm_bio_prison_cell
*cell
;
781 * If the bio covers the whole area of a block then we can avoid
782 * zeroing or copying. Instead this bio is hooked. The bio will
783 * still be in the cell, so care has to be taken to avoid issuing
787 bio_end_io_t
*saved_bi_end_io
;
790 static void __complete_mapping_preparation(struct dm_thin_new_mapping
*m
)
792 struct pool
*pool
= m
->tc
->pool
;
794 if (atomic_dec_and_test(&m
->prepare_actions
)) {
795 list_add_tail(&m
->list
, &pool
->prepared_mappings
);
800 static void complete_mapping_preparation(struct dm_thin_new_mapping
*m
)
803 struct pool
*pool
= m
->tc
->pool
;
805 spin_lock_irqsave(&pool
->lock
, flags
);
806 __complete_mapping_preparation(m
);
807 spin_unlock_irqrestore(&pool
->lock
, flags
);
810 static void copy_complete(int read_err
, unsigned long write_err
, void *context
)
812 struct dm_thin_new_mapping
*m
= context
;
814 m
->status
= read_err
|| write_err
? BLK_STS_IOERR
: 0;
815 complete_mapping_preparation(m
);
818 static void overwrite_endio(struct bio
*bio
)
820 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
821 struct dm_thin_new_mapping
*m
= h
->overwrite_mapping
;
823 bio
->bi_end_io
= m
->saved_bi_end_io
;
825 m
->status
= bio
->bi_status
;
826 complete_mapping_preparation(m
);
829 /*----------------------------------------------------------------*/
836 * Prepared mapping jobs.
840 * This sends the bios in the cell, except the original holder, back
841 * to the deferred_bios list.
843 static void cell_defer_no_holder(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
845 struct pool
*pool
= tc
->pool
;
848 spin_lock_irqsave(&tc
->lock
, flags
);
849 cell_release_no_holder(pool
, cell
, &tc
->deferred_bio_list
);
850 spin_unlock_irqrestore(&tc
->lock
, flags
);
855 static void thin_defer_bio(struct thin_c
*tc
, struct bio
*bio
);
859 struct bio_list defer_bios
;
860 struct bio_list issue_bios
;
863 static void __inc_remap_and_issue_cell(void *context
,
864 struct dm_bio_prison_cell
*cell
)
866 struct remap_info
*info
= context
;
869 while ((bio
= bio_list_pop(&cell
->bios
))) {
870 if (op_is_flush(bio
->bi_opf
) || bio_op(bio
) == REQ_OP_DISCARD
)
871 bio_list_add(&info
->defer_bios
, bio
);
873 inc_all_io_entry(info
->tc
->pool
, bio
);
876 * We can't issue the bios with the bio prison lock
877 * held, so we add them to a list to issue on
878 * return from this function.
880 bio_list_add(&info
->issue_bios
, bio
);
885 static void inc_remap_and_issue_cell(struct thin_c
*tc
,
886 struct dm_bio_prison_cell
*cell
,
890 struct remap_info info
;
893 bio_list_init(&info
.defer_bios
);
894 bio_list_init(&info
.issue_bios
);
897 * We have to be careful to inc any bios we're about to issue
898 * before the cell is released, and avoid a race with new bios
899 * being added to the cell.
901 cell_visit_release(tc
->pool
, __inc_remap_and_issue_cell
,
904 while ((bio
= bio_list_pop(&info
.defer_bios
)))
905 thin_defer_bio(tc
, bio
);
907 while ((bio
= bio_list_pop(&info
.issue_bios
)))
908 remap_and_issue(info
.tc
, bio
, block
);
911 static void process_prepared_mapping_fail(struct dm_thin_new_mapping
*m
)
913 cell_error(m
->tc
->pool
, m
->cell
);
915 mempool_free(m
, m
->tc
->pool
->mapping_pool
);
918 static void process_prepared_mapping(struct dm_thin_new_mapping
*m
)
920 struct thin_c
*tc
= m
->tc
;
921 struct pool
*pool
= tc
->pool
;
922 struct bio
*bio
= m
->bio
;
926 cell_error(pool
, m
->cell
);
931 * Commit the prepared block into the mapping btree.
932 * Any I/O for this block arriving after this point will get
933 * remapped to it directly.
935 r
= dm_thin_insert_block(tc
->td
, m
->virt_begin
, m
->data_block
);
937 metadata_operation_failed(pool
, "dm_thin_insert_block", r
);
938 cell_error(pool
, m
->cell
);
943 * Release any bios held while the block was being provisioned.
944 * If we are processing a write bio that completely covers the block,
945 * we already processed it so can ignore it now when processing
946 * the bios in the cell.
949 inc_remap_and_issue_cell(tc
, m
->cell
, m
->data_block
);
952 inc_all_io_entry(tc
->pool
, m
->cell
->holder
);
953 remap_and_issue(tc
, m
->cell
->holder
, m
->data_block
);
954 inc_remap_and_issue_cell(tc
, m
->cell
, m
->data_block
);
959 mempool_free(m
, pool
->mapping_pool
);
962 /*----------------------------------------------------------------*/
964 static void free_discard_mapping(struct dm_thin_new_mapping
*m
)
966 struct thin_c
*tc
= m
->tc
;
968 cell_defer_no_holder(tc
, m
->cell
);
969 mempool_free(m
, tc
->pool
->mapping_pool
);
972 static void process_prepared_discard_fail(struct dm_thin_new_mapping
*m
)
974 bio_io_error(m
->bio
);
975 free_discard_mapping(m
);
978 static void process_prepared_discard_success(struct dm_thin_new_mapping
*m
)
981 free_discard_mapping(m
);
984 static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping
*m
)
987 struct thin_c
*tc
= m
->tc
;
989 r
= dm_thin_remove_range(tc
->td
, m
->cell
->key
.block_begin
, m
->cell
->key
.block_end
);
991 metadata_operation_failed(tc
->pool
, "dm_thin_remove_range", r
);
992 bio_io_error(m
->bio
);
996 cell_defer_no_holder(tc
, m
->cell
);
997 mempool_free(m
, tc
->pool
->mapping_pool
);
1000 /*----------------------------------------------------------------*/
1002 static void passdown_double_checking_shared_status(struct dm_thin_new_mapping
*m
,
1003 struct bio
*discard_parent
)
1006 * We've already unmapped this range of blocks, but before we
1007 * passdown we have to check that these blocks are now unused.
1011 struct thin_c
*tc
= m
->tc
;
1012 struct pool
*pool
= tc
->pool
;
1013 dm_block_t b
= m
->data_block
, e
, end
= m
->data_block
+ m
->virt_end
- m
->virt_begin
;
1014 struct discard_op op
;
1016 begin_discard(&op
, tc
, discard_parent
);
1018 /* find start of unmapped run */
1019 for (; b
< end
; b
++) {
1020 r
= dm_pool_block_is_used(pool
->pmd
, b
, &used
);
1031 /* find end of run */
1032 for (e
= b
+ 1; e
!= end
; e
++) {
1033 r
= dm_pool_block_is_used(pool
->pmd
, e
, &used
);
1041 r
= issue_discard(&op
, b
, e
);
1048 end_discard(&op
, r
);
1051 static void queue_passdown_pt2(struct dm_thin_new_mapping
*m
)
1053 unsigned long flags
;
1054 struct pool
*pool
= m
->tc
->pool
;
1056 spin_lock_irqsave(&pool
->lock
, flags
);
1057 list_add_tail(&m
->list
, &pool
->prepared_discards_pt2
);
1058 spin_unlock_irqrestore(&pool
->lock
, flags
);
1062 static void passdown_endio(struct bio
*bio
)
1065 * It doesn't matter if the passdown discard failed, we still want
1066 * to unmap (we ignore err).
1068 queue_passdown_pt2(bio
->bi_private
);
1072 static void process_prepared_discard_passdown_pt1(struct dm_thin_new_mapping
*m
)
1075 struct thin_c
*tc
= m
->tc
;
1076 struct pool
*pool
= tc
->pool
;
1077 struct bio
*discard_parent
;
1078 dm_block_t data_end
= m
->data_block
+ (m
->virt_end
- m
->virt_begin
);
1081 * Only this thread allocates blocks, so we can be sure that the
1082 * newly unmapped blocks will not be allocated before the end of
1085 r
= dm_thin_remove_range(tc
->td
, m
->virt_begin
, m
->virt_end
);
1087 metadata_operation_failed(pool
, "dm_thin_remove_range", r
);
1088 bio_io_error(m
->bio
);
1089 cell_defer_no_holder(tc
, m
->cell
);
1090 mempool_free(m
, pool
->mapping_pool
);
1095 * Increment the unmapped blocks. This prevents a race between the
1096 * passdown io and reallocation of freed blocks.
1098 r
= dm_pool_inc_data_range(pool
->pmd
, m
->data_block
, data_end
);
1100 metadata_operation_failed(pool
, "dm_pool_inc_data_range", r
);
1101 bio_io_error(m
->bio
);
1102 cell_defer_no_holder(tc
, m
->cell
);
1103 mempool_free(m
, pool
->mapping_pool
);
1107 discard_parent
= bio_alloc(GFP_NOIO
, 1);
1108 if (!discard_parent
) {
1109 DMWARN("%s: unable to allocate top level discard bio for passdown. Skipping passdown.",
1110 dm_device_name(tc
->pool
->pool_md
));
1111 queue_passdown_pt2(m
);
1114 discard_parent
->bi_end_io
= passdown_endio
;
1115 discard_parent
->bi_private
= m
;
1117 if (m
->maybe_shared
)
1118 passdown_double_checking_shared_status(m
, discard_parent
);
1120 struct discard_op op
;
1122 begin_discard(&op
, tc
, discard_parent
);
1123 r
= issue_discard(&op
, m
->data_block
, data_end
);
1124 end_discard(&op
, r
);
1129 static void process_prepared_discard_passdown_pt2(struct dm_thin_new_mapping
*m
)
1132 struct thin_c
*tc
= m
->tc
;
1133 struct pool
*pool
= tc
->pool
;
1136 * The passdown has completed, so now we can decrement all those
1139 r
= dm_pool_dec_data_range(pool
->pmd
, m
->data_block
,
1140 m
->data_block
+ (m
->virt_end
- m
->virt_begin
));
1142 metadata_operation_failed(pool
, "dm_pool_dec_data_range", r
);
1143 bio_io_error(m
->bio
);
1147 cell_defer_no_holder(tc
, m
->cell
);
1148 mempool_free(m
, pool
->mapping_pool
);
1151 static void process_prepared(struct pool
*pool
, struct list_head
*head
,
1152 process_mapping_fn
*fn
)
1154 unsigned long flags
;
1155 struct list_head maps
;
1156 struct dm_thin_new_mapping
*m
, *tmp
;
1158 INIT_LIST_HEAD(&maps
);
1159 spin_lock_irqsave(&pool
->lock
, flags
);
1160 list_splice_init(head
, &maps
);
1161 spin_unlock_irqrestore(&pool
->lock
, flags
);
1163 list_for_each_entry_safe(m
, tmp
, &maps
, list
)
1168 * Deferred bio jobs.
1170 static int io_overlaps_block(struct pool
*pool
, struct bio
*bio
)
1172 return bio
->bi_iter
.bi_size
==
1173 (pool
->sectors_per_block
<< SECTOR_SHIFT
);
1176 static int io_overwrites_block(struct pool
*pool
, struct bio
*bio
)
1178 return (bio_data_dir(bio
) == WRITE
) &&
1179 io_overlaps_block(pool
, bio
);
1182 static void save_and_set_endio(struct bio
*bio
, bio_end_io_t
**save
,
1185 *save
= bio
->bi_end_io
;
1186 bio
->bi_end_io
= fn
;
1189 static int ensure_next_mapping(struct pool
*pool
)
1191 if (pool
->next_mapping
)
1194 pool
->next_mapping
= mempool_alloc(pool
->mapping_pool
, GFP_ATOMIC
);
1196 return pool
->next_mapping
? 0 : -ENOMEM
;
1199 static struct dm_thin_new_mapping
*get_next_mapping(struct pool
*pool
)
1201 struct dm_thin_new_mapping
*m
= pool
->next_mapping
;
1203 BUG_ON(!pool
->next_mapping
);
1205 memset(m
, 0, sizeof(struct dm_thin_new_mapping
));
1206 INIT_LIST_HEAD(&m
->list
);
1209 pool
->next_mapping
= NULL
;
1214 static void ll_zero(struct thin_c
*tc
, struct dm_thin_new_mapping
*m
,
1215 sector_t begin
, sector_t end
)
1218 struct dm_io_region to
;
1220 to
.bdev
= tc
->pool_dev
->bdev
;
1222 to
.count
= end
- begin
;
1224 r
= dm_kcopyd_zero(tc
->pool
->copier
, 1, &to
, 0, copy_complete
, m
);
1226 DMERR_LIMIT("dm_kcopyd_zero() failed");
1227 copy_complete(1, 1, m
);
1231 static void remap_and_issue_overwrite(struct thin_c
*tc
, struct bio
*bio
,
1232 dm_block_t data_begin
,
1233 struct dm_thin_new_mapping
*m
)
1235 struct pool
*pool
= tc
->pool
;
1236 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1238 h
->overwrite_mapping
= m
;
1240 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
1241 inc_all_io_entry(pool
, bio
);
1242 remap_and_issue(tc
, bio
, data_begin
);
1246 * A partial copy also needs to zero the uncopied region.
1248 static void schedule_copy(struct thin_c
*tc
, dm_block_t virt_block
,
1249 struct dm_dev
*origin
, dm_block_t data_origin
,
1250 dm_block_t data_dest
,
1251 struct dm_bio_prison_cell
*cell
, struct bio
*bio
,
1255 struct pool
*pool
= tc
->pool
;
1256 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
1259 m
->virt_begin
= virt_block
;
1260 m
->virt_end
= virt_block
+ 1u;
1261 m
->data_block
= data_dest
;
1265 * quiesce action + copy action + an extra reference held for the
1266 * duration of this function (we may need to inc later for a
1269 atomic_set(&m
->prepare_actions
, 3);
1271 if (!dm_deferred_set_add_work(pool
->shared_read_ds
, &m
->list
))
1272 complete_mapping_preparation(m
); /* already quiesced */
1275 * IO to pool_dev remaps to the pool target's data_dev.
1277 * If the whole block of data is being overwritten, we can issue the
1278 * bio immediately. Otherwise we use kcopyd to clone the data first.
1280 if (io_overwrites_block(pool
, bio
))
1281 remap_and_issue_overwrite(tc
, bio
, data_dest
, m
);
1283 struct dm_io_region from
, to
;
1285 from
.bdev
= origin
->bdev
;
1286 from
.sector
= data_origin
* pool
->sectors_per_block
;
1289 to
.bdev
= tc
->pool_dev
->bdev
;
1290 to
.sector
= data_dest
* pool
->sectors_per_block
;
1293 r
= dm_kcopyd_copy(pool
->copier
, &from
, 1, &to
,
1294 0, copy_complete
, m
);
1296 DMERR_LIMIT("dm_kcopyd_copy() failed");
1297 copy_complete(1, 1, m
);
1300 * We allow the zero to be issued, to simplify the
1301 * error path. Otherwise we'd need to start
1302 * worrying about decrementing the prepare_actions
1308 * Do we need to zero a tail region?
1310 if (len
< pool
->sectors_per_block
&& pool
->pf
.zero_new_blocks
) {
1311 atomic_inc(&m
->prepare_actions
);
1313 data_dest
* pool
->sectors_per_block
+ len
,
1314 (data_dest
+ 1) * pool
->sectors_per_block
);
1318 complete_mapping_preparation(m
); /* drop our ref */
1321 static void schedule_internal_copy(struct thin_c
*tc
, dm_block_t virt_block
,
1322 dm_block_t data_origin
, dm_block_t data_dest
,
1323 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
1325 schedule_copy(tc
, virt_block
, tc
->pool_dev
,
1326 data_origin
, data_dest
, cell
, bio
,
1327 tc
->pool
->sectors_per_block
);
1330 static void schedule_zero(struct thin_c
*tc
, dm_block_t virt_block
,
1331 dm_block_t data_block
, struct dm_bio_prison_cell
*cell
,
1334 struct pool
*pool
= tc
->pool
;
1335 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
1337 atomic_set(&m
->prepare_actions
, 1); /* no need to quiesce */
1339 m
->virt_begin
= virt_block
;
1340 m
->virt_end
= virt_block
+ 1u;
1341 m
->data_block
= data_block
;
1345 * If the whole block of data is being overwritten or we are not
1346 * zeroing pre-existing data, we can issue the bio immediately.
1347 * Otherwise we use kcopyd to zero the data first.
1349 if (pool
->pf
.zero_new_blocks
) {
1350 if (io_overwrites_block(pool
, bio
))
1351 remap_and_issue_overwrite(tc
, bio
, data_block
, m
);
1353 ll_zero(tc
, m
, data_block
* pool
->sectors_per_block
,
1354 (data_block
+ 1) * pool
->sectors_per_block
);
1356 process_prepared_mapping(m
);
1359 static void schedule_external_copy(struct thin_c
*tc
, dm_block_t virt_block
,
1360 dm_block_t data_dest
,
1361 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
1363 struct pool
*pool
= tc
->pool
;
1364 sector_t virt_block_begin
= virt_block
* pool
->sectors_per_block
;
1365 sector_t virt_block_end
= (virt_block
+ 1) * pool
->sectors_per_block
;
1367 if (virt_block_end
<= tc
->origin_size
)
1368 schedule_copy(tc
, virt_block
, tc
->origin_dev
,
1369 virt_block
, data_dest
, cell
, bio
,
1370 pool
->sectors_per_block
);
1372 else if (virt_block_begin
< tc
->origin_size
)
1373 schedule_copy(tc
, virt_block
, tc
->origin_dev
,
1374 virt_block
, data_dest
, cell
, bio
,
1375 tc
->origin_size
- virt_block_begin
);
1378 schedule_zero(tc
, virt_block
, data_dest
, cell
, bio
);
1381 static void set_pool_mode(struct pool
*pool
, enum pool_mode new_mode
);
1383 static void requeue_bios(struct pool
*pool
);
1385 static void check_for_space(struct pool
*pool
)
1390 if (get_pool_mode(pool
) != PM_OUT_OF_DATA_SPACE
)
1393 r
= dm_pool_get_free_block_count(pool
->pmd
, &nr_free
);
1398 set_pool_mode(pool
, PM_WRITE
);
1404 * A non-zero return indicates read_only or fail_io mode.
1405 * Many callers don't care about the return value.
1407 static int commit(struct pool
*pool
)
1411 if (get_pool_mode(pool
) >= PM_READ_ONLY
)
1414 r
= dm_pool_commit_metadata(pool
->pmd
);
1416 metadata_operation_failed(pool
, "dm_pool_commit_metadata", r
);
1418 check_for_space(pool
);
1423 static void check_low_water_mark(struct pool
*pool
, dm_block_t free_blocks
)
1425 unsigned long flags
;
1427 if (free_blocks
<= pool
->low_water_blocks
&& !pool
->low_water_triggered
) {
1428 DMWARN("%s: reached low water mark for data device: sending event.",
1429 dm_device_name(pool
->pool_md
));
1430 spin_lock_irqsave(&pool
->lock
, flags
);
1431 pool
->low_water_triggered
= true;
1432 spin_unlock_irqrestore(&pool
->lock
, flags
);
1433 dm_table_event(pool
->ti
->table
);
1437 static int alloc_data_block(struct thin_c
*tc
, dm_block_t
*result
)
1440 dm_block_t free_blocks
;
1441 struct pool
*pool
= tc
->pool
;
1443 if (WARN_ON(get_pool_mode(pool
) != PM_WRITE
))
1446 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
1448 metadata_operation_failed(pool
, "dm_pool_get_free_block_count", r
);
1452 check_low_water_mark(pool
, free_blocks
);
1456 * Try to commit to see if that will free up some
1463 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
1465 metadata_operation_failed(pool
, "dm_pool_get_free_block_count", r
);
1470 set_pool_mode(pool
, PM_OUT_OF_DATA_SPACE
);
1475 r
= dm_pool_alloc_data_block(pool
->pmd
, result
);
1478 set_pool_mode(pool
, PM_OUT_OF_DATA_SPACE
);
1480 metadata_operation_failed(pool
, "dm_pool_alloc_data_block", r
);
1488 * If we have run out of space, queue bios until the device is
1489 * resumed, presumably after having been reloaded with more space.
1491 static void retry_on_resume(struct bio
*bio
)
1493 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1494 struct thin_c
*tc
= h
->tc
;
1495 unsigned long flags
;
1497 spin_lock_irqsave(&tc
->lock
, flags
);
1498 bio_list_add(&tc
->retry_on_resume_list
, bio
);
1499 spin_unlock_irqrestore(&tc
->lock
, flags
);
1502 static blk_status_t
should_error_unserviceable_bio(struct pool
*pool
)
1504 enum pool_mode m
= get_pool_mode(pool
);
1508 /* Shouldn't get here */
1509 DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
1510 return BLK_STS_IOERR
;
1512 case PM_OUT_OF_DATA_SPACE
:
1513 return pool
->pf
.error_if_no_space
? BLK_STS_NOSPC
: 0;
1517 return BLK_STS_IOERR
;
1519 /* Shouldn't get here */
1520 DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1521 return BLK_STS_IOERR
;
1525 static void handle_unserviceable_bio(struct pool
*pool
, struct bio
*bio
)
1527 blk_status_t error
= should_error_unserviceable_bio(pool
);
1530 bio
->bi_status
= error
;
1533 retry_on_resume(bio
);
1536 static void retry_bios_on_resume(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
1539 struct bio_list bios
;
1542 error
= should_error_unserviceable_bio(pool
);
1544 cell_error_with_code(pool
, cell
, error
);
1548 bio_list_init(&bios
);
1549 cell_release(pool
, cell
, &bios
);
1551 while ((bio
= bio_list_pop(&bios
)))
1552 retry_on_resume(bio
);
1555 static void process_discard_cell_no_passdown(struct thin_c
*tc
,
1556 struct dm_bio_prison_cell
*virt_cell
)
1558 struct pool
*pool
= tc
->pool
;
1559 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
1562 * We don't need to lock the data blocks, since there's no
1563 * passdown. We only lock data blocks for allocation and breaking sharing.
1566 m
->virt_begin
= virt_cell
->key
.block_begin
;
1567 m
->virt_end
= virt_cell
->key
.block_end
;
1568 m
->cell
= virt_cell
;
1569 m
->bio
= virt_cell
->holder
;
1571 if (!dm_deferred_set_add_work(pool
->all_io_ds
, &m
->list
))
1572 pool
->process_prepared_discard(m
);
1575 static void break_up_discard_bio(struct thin_c
*tc
, dm_block_t begin
, dm_block_t end
,
1578 struct pool
*pool
= tc
->pool
;
1582 struct dm_cell_key data_key
;
1583 struct dm_bio_prison_cell
*data_cell
;
1584 struct dm_thin_new_mapping
*m
;
1585 dm_block_t virt_begin
, virt_end
, data_begin
;
1587 while (begin
!= end
) {
1588 r
= ensure_next_mapping(pool
);
1590 /* we did our best */
1593 r
= dm_thin_find_mapped_range(tc
->td
, begin
, end
, &virt_begin
, &virt_end
,
1594 &data_begin
, &maybe_shared
);
1597 * Silently fail, letting any mappings we've
1602 build_key(tc
->td
, PHYSICAL
, data_begin
, data_begin
+ (virt_end
- virt_begin
), &data_key
);
1603 if (bio_detain(tc
->pool
, &data_key
, NULL
, &data_cell
)) {
1604 /* contention, we'll give up with this range */
1610 * IO may still be going to the destination block. We must
1611 * quiesce before we can do the removal.
1613 m
= get_next_mapping(pool
);
1615 m
->maybe_shared
= maybe_shared
;
1616 m
->virt_begin
= virt_begin
;
1617 m
->virt_end
= virt_end
;
1618 m
->data_block
= data_begin
;
1619 m
->cell
= data_cell
;
1623 * The parent bio must not complete before sub discard bios are
1624 * chained to it (see end_discard's bio_chain)!
1626 * This per-mapping bi_remaining increment is paired with
1627 * the implicit decrement that occurs via bio_endio() in
1630 bio_inc_remaining(bio
);
1631 if (!dm_deferred_set_add_work(pool
->all_io_ds
, &m
->list
))
1632 pool
->process_prepared_discard(m
);
1638 static void process_discard_cell_passdown(struct thin_c
*tc
, struct dm_bio_prison_cell
*virt_cell
)
1640 struct bio
*bio
= virt_cell
->holder
;
1641 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1644 * The virt_cell will only get freed once the origin bio completes.
1645 * This means it will remain locked while all the individual
1646 * passdown bios are in flight.
1648 h
->cell
= virt_cell
;
1649 break_up_discard_bio(tc
, virt_cell
->key
.block_begin
, virt_cell
->key
.block_end
, bio
);
1652 * We complete the bio now, knowing that the bi_remaining field
1653 * will prevent completion until the sub range discards have
1659 static void process_discard_bio(struct thin_c
*tc
, struct bio
*bio
)
1661 dm_block_t begin
, end
;
1662 struct dm_cell_key virt_key
;
1663 struct dm_bio_prison_cell
*virt_cell
;
1665 get_bio_block_range(tc
, bio
, &begin
, &end
);
1668 * The discard covers less than a block.
1674 build_key(tc
->td
, VIRTUAL
, begin
, end
, &virt_key
);
1675 if (bio_detain(tc
->pool
, &virt_key
, bio
, &virt_cell
))
1677 * Potential starvation issue: We're relying on the
1678 * fs/application being well behaved, and not trying to
1679 * send IO to a region at the same time as discarding it.
1680 * If they do this persistently then it's possible this
1681 * cell will never be granted.
1685 tc
->pool
->process_discard_cell(tc
, virt_cell
);
1688 static void break_sharing(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1689 struct dm_cell_key
*key
,
1690 struct dm_thin_lookup_result
*lookup_result
,
1691 struct dm_bio_prison_cell
*cell
)
1694 dm_block_t data_block
;
1695 struct pool
*pool
= tc
->pool
;
1697 r
= alloc_data_block(tc
, &data_block
);
1700 schedule_internal_copy(tc
, block
, lookup_result
->block
,
1701 data_block
, cell
, bio
);
1705 retry_bios_on_resume(pool
, cell
);
1709 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1711 cell_error(pool
, cell
);
1716 static void __remap_and_issue_shared_cell(void *context
,
1717 struct dm_bio_prison_cell
*cell
)
1719 struct remap_info
*info
= context
;
1722 while ((bio
= bio_list_pop(&cell
->bios
))) {
1723 if (bio_data_dir(bio
) == WRITE
|| op_is_flush(bio
->bi_opf
) ||
1724 bio_op(bio
) == REQ_OP_DISCARD
)
1725 bio_list_add(&info
->defer_bios
, bio
);
1727 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));;
1729 h
->shared_read_entry
= dm_deferred_entry_inc(info
->tc
->pool
->shared_read_ds
);
1730 inc_all_io_entry(info
->tc
->pool
, bio
);
1731 bio_list_add(&info
->issue_bios
, bio
);
1736 static void remap_and_issue_shared_cell(struct thin_c
*tc
,
1737 struct dm_bio_prison_cell
*cell
,
1741 struct remap_info info
;
1744 bio_list_init(&info
.defer_bios
);
1745 bio_list_init(&info
.issue_bios
);
1747 cell_visit_release(tc
->pool
, __remap_and_issue_shared_cell
,
1750 while ((bio
= bio_list_pop(&info
.defer_bios
)))
1751 thin_defer_bio(tc
, bio
);
1753 while ((bio
= bio_list_pop(&info
.issue_bios
)))
1754 remap_and_issue(tc
, bio
, block
);
1757 static void process_shared_bio(struct thin_c
*tc
, struct bio
*bio
,
1759 struct dm_thin_lookup_result
*lookup_result
,
1760 struct dm_bio_prison_cell
*virt_cell
)
1762 struct dm_bio_prison_cell
*data_cell
;
1763 struct pool
*pool
= tc
->pool
;
1764 struct dm_cell_key key
;
1767 * If cell is already occupied, then sharing is already in the process
1768 * of being broken so we have nothing further to do here.
1770 build_data_key(tc
->td
, lookup_result
->block
, &key
);
1771 if (bio_detain(pool
, &key
, bio
, &data_cell
)) {
1772 cell_defer_no_holder(tc
, virt_cell
);
1776 if (bio_data_dir(bio
) == WRITE
&& bio
->bi_iter
.bi_size
) {
1777 break_sharing(tc
, bio
, block
, &key
, lookup_result
, data_cell
);
1778 cell_defer_no_holder(tc
, virt_cell
);
1780 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1782 h
->shared_read_entry
= dm_deferred_entry_inc(pool
->shared_read_ds
);
1783 inc_all_io_entry(pool
, bio
);
1784 remap_and_issue(tc
, bio
, lookup_result
->block
);
1786 remap_and_issue_shared_cell(tc
, data_cell
, lookup_result
->block
);
1787 remap_and_issue_shared_cell(tc
, virt_cell
, lookup_result
->block
);
1791 static void provision_block(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1792 struct dm_bio_prison_cell
*cell
)
1795 dm_block_t data_block
;
1796 struct pool
*pool
= tc
->pool
;
1799 * Remap empty bios (flushes) immediately, without provisioning.
1801 if (!bio
->bi_iter
.bi_size
) {
1802 inc_all_io_entry(pool
, bio
);
1803 cell_defer_no_holder(tc
, cell
);
1805 remap_and_issue(tc
, bio
, 0);
1810 * Fill read bios with zeroes and complete them immediately.
1812 if (bio_data_dir(bio
) == READ
) {
1814 cell_defer_no_holder(tc
, cell
);
1819 r
= alloc_data_block(tc
, &data_block
);
1823 schedule_external_copy(tc
, block
, data_block
, cell
, bio
);
1825 schedule_zero(tc
, block
, data_block
, cell
, bio
);
1829 retry_bios_on_resume(pool
, cell
);
1833 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1835 cell_error(pool
, cell
);
1840 static void process_cell(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
1843 struct pool
*pool
= tc
->pool
;
1844 struct bio
*bio
= cell
->holder
;
1845 dm_block_t block
= get_bio_block(tc
, bio
);
1846 struct dm_thin_lookup_result lookup_result
;
1848 if (tc
->requeue_mode
) {
1849 cell_requeue(pool
, cell
);
1853 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1856 if (lookup_result
.shared
)
1857 process_shared_bio(tc
, bio
, block
, &lookup_result
, cell
);
1859 inc_all_io_entry(pool
, bio
);
1860 remap_and_issue(tc
, bio
, lookup_result
.block
);
1861 inc_remap_and_issue_cell(tc
, cell
, lookup_result
.block
);
1866 if (bio_data_dir(bio
) == READ
&& tc
->origin_dev
) {
1867 inc_all_io_entry(pool
, bio
);
1868 cell_defer_no_holder(tc
, cell
);
1870 if (bio_end_sector(bio
) <= tc
->origin_size
)
1871 remap_to_origin_and_issue(tc
, bio
);
1873 else if (bio
->bi_iter
.bi_sector
< tc
->origin_size
) {
1875 bio
->bi_iter
.bi_size
= (tc
->origin_size
- bio
->bi_iter
.bi_sector
) << SECTOR_SHIFT
;
1876 remap_to_origin_and_issue(tc
, bio
);
1883 provision_block(tc
, bio
, block
, cell
);
1887 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1889 cell_defer_no_holder(tc
, cell
);
1895 static void process_bio(struct thin_c
*tc
, struct bio
*bio
)
1897 struct pool
*pool
= tc
->pool
;
1898 dm_block_t block
= get_bio_block(tc
, bio
);
1899 struct dm_bio_prison_cell
*cell
;
1900 struct dm_cell_key key
;
1903 * If cell is already occupied, then the block is already
1904 * being provisioned so we have nothing further to do here.
1906 build_virtual_key(tc
->td
, block
, &key
);
1907 if (bio_detain(pool
, &key
, bio
, &cell
))
1910 process_cell(tc
, cell
);
1913 static void __process_bio_read_only(struct thin_c
*tc
, struct bio
*bio
,
1914 struct dm_bio_prison_cell
*cell
)
1917 int rw
= bio_data_dir(bio
);
1918 dm_block_t block
= get_bio_block(tc
, bio
);
1919 struct dm_thin_lookup_result lookup_result
;
1921 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1924 if (lookup_result
.shared
&& (rw
== WRITE
) && bio
->bi_iter
.bi_size
) {
1925 handle_unserviceable_bio(tc
->pool
, bio
);
1927 cell_defer_no_holder(tc
, cell
);
1929 inc_all_io_entry(tc
->pool
, bio
);
1930 remap_and_issue(tc
, bio
, lookup_result
.block
);
1932 inc_remap_and_issue_cell(tc
, cell
, lookup_result
.block
);
1938 cell_defer_no_holder(tc
, cell
);
1940 handle_unserviceable_bio(tc
->pool
, bio
);
1944 if (tc
->origin_dev
) {
1945 inc_all_io_entry(tc
->pool
, bio
);
1946 remap_to_origin_and_issue(tc
, bio
);
1955 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1958 cell_defer_no_holder(tc
, cell
);
1964 static void process_bio_read_only(struct thin_c
*tc
, struct bio
*bio
)
1966 __process_bio_read_only(tc
, bio
, NULL
);
1969 static void process_cell_read_only(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
1971 __process_bio_read_only(tc
, cell
->holder
, cell
);
1974 static void process_bio_success(struct thin_c
*tc
, struct bio
*bio
)
1979 static void process_bio_fail(struct thin_c
*tc
, struct bio
*bio
)
1984 static void process_cell_success(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
1986 cell_success(tc
->pool
, cell
);
1989 static void process_cell_fail(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
1991 cell_error(tc
->pool
, cell
);
1995 * FIXME: should we also commit due to size of transaction, measured in
1998 static int need_commit_due_to_time(struct pool
*pool
)
2000 return !time_in_range(jiffies
, pool
->last_commit_jiffies
,
2001 pool
->last_commit_jiffies
+ COMMIT_PERIOD
);
2004 #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node)
2005 #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook))
2007 static void __thin_bio_rb_add(struct thin_c
*tc
, struct bio
*bio
)
2009 struct rb_node
**rbp
, *parent
;
2010 struct dm_thin_endio_hook
*pbd
;
2011 sector_t bi_sector
= bio
->bi_iter
.bi_sector
;
2013 rbp
= &tc
->sort_bio_list
.rb_node
;
2017 pbd
= thin_pbd(parent
);
2019 if (bi_sector
< thin_bio(pbd
)->bi_iter
.bi_sector
)
2020 rbp
= &(*rbp
)->rb_left
;
2022 rbp
= &(*rbp
)->rb_right
;
2025 pbd
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
2026 rb_link_node(&pbd
->rb_node
, parent
, rbp
);
2027 rb_insert_color(&pbd
->rb_node
, &tc
->sort_bio_list
);
2030 static void __extract_sorted_bios(struct thin_c
*tc
)
2032 struct rb_node
*node
;
2033 struct dm_thin_endio_hook
*pbd
;
2036 for (node
= rb_first(&tc
->sort_bio_list
); node
; node
= rb_next(node
)) {
2037 pbd
= thin_pbd(node
);
2038 bio
= thin_bio(pbd
);
2040 bio_list_add(&tc
->deferred_bio_list
, bio
);
2041 rb_erase(&pbd
->rb_node
, &tc
->sort_bio_list
);
2044 WARN_ON(!RB_EMPTY_ROOT(&tc
->sort_bio_list
));
2047 static void __sort_thin_deferred_bios(struct thin_c
*tc
)
2050 struct bio_list bios
;
2052 bio_list_init(&bios
);
2053 bio_list_merge(&bios
, &tc
->deferred_bio_list
);
2054 bio_list_init(&tc
->deferred_bio_list
);
2056 /* Sort deferred_bio_list using rb-tree */
2057 while ((bio
= bio_list_pop(&bios
)))
2058 __thin_bio_rb_add(tc
, bio
);
2061 * Transfer the sorted bios in sort_bio_list back to
2062 * deferred_bio_list to allow lockless submission of
2065 __extract_sorted_bios(tc
);
2068 static void process_thin_deferred_bios(struct thin_c
*tc
)
2070 struct pool
*pool
= tc
->pool
;
2071 unsigned long flags
;
2073 struct bio_list bios
;
2074 struct blk_plug plug
;
2077 if (tc
->requeue_mode
) {
2078 error_thin_bio_list(tc
, &tc
->deferred_bio_list
,
2079 BLK_STS_DM_REQUEUE
);
2083 bio_list_init(&bios
);
2085 spin_lock_irqsave(&tc
->lock
, flags
);
2087 if (bio_list_empty(&tc
->deferred_bio_list
)) {
2088 spin_unlock_irqrestore(&tc
->lock
, flags
);
2092 __sort_thin_deferred_bios(tc
);
2094 bio_list_merge(&bios
, &tc
->deferred_bio_list
);
2095 bio_list_init(&tc
->deferred_bio_list
);
2097 spin_unlock_irqrestore(&tc
->lock
, flags
);
2099 blk_start_plug(&plug
);
2100 while ((bio
= bio_list_pop(&bios
))) {
2102 * If we've got no free new_mapping structs, and processing
2103 * this bio might require one, we pause until there are some
2104 * prepared mappings to process.
2106 if (ensure_next_mapping(pool
)) {
2107 spin_lock_irqsave(&tc
->lock
, flags
);
2108 bio_list_add(&tc
->deferred_bio_list
, bio
);
2109 bio_list_merge(&tc
->deferred_bio_list
, &bios
);
2110 spin_unlock_irqrestore(&tc
->lock
, flags
);
2114 if (bio_op(bio
) == REQ_OP_DISCARD
)
2115 pool
->process_discard(tc
, bio
);
2117 pool
->process_bio(tc
, bio
);
2119 if ((count
++ & 127) == 0) {
2120 throttle_work_update(&pool
->throttle
);
2121 dm_pool_issue_prefetches(pool
->pmd
);
2124 blk_finish_plug(&plug
);
2127 static int cmp_cells(const void *lhs
, const void *rhs
)
2129 struct dm_bio_prison_cell
*lhs_cell
= *((struct dm_bio_prison_cell
**) lhs
);
2130 struct dm_bio_prison_cell
*rhs_cell
= *((struct dm_bio_prison_cell
**) rhs
);
2132 BUG_ON(!lhs_cell
->holder
);
2133 BUG_ON(!rhs_cell
->holder
);
2135 if (lhs_cell
->holder
->bi_iter
.bi_sector
< rhs_cell
->holder
->bi_iter
.bi_sector
)
2138 if (lhs_cell
->holder
->bi_iter
.bi_sector
> rhs_cell
->holder
->bi_iter
.bi_sector
)
2144 static unsigned sort_cells(struct pool
*pool
, struct list_head
*cells
)
2147 struct dm_bio_prison_cell
*cell
, *tmp
;
2149 list_for_each_entry_safe(cell
, tmp
, cells
, user_list
) {
2150 if (count
>= CELL_SORT_ARRAY_SIZE
)
2153 pool
->cell_sort_array
[count
++] = cell
;
2154 list_del(&cell
->user_list
);
2157 sort(pool
->cell_sort_array
, count
, sizeof(cell
), cmp_cells
, NULL
);
2162 static void process_thin_deferred_cells(struct thin_c
*tc
)
2164 struct pool
*pool
= tc
->pool
;
2165 unsigned long flags
;
2166 struct list_head cells
;
2167 struct dm_bio_prison_cell
*cell
;
2168 unsigned i
, j
, count
;
2170 INIT_LIST_HEAD(&cells
);
2172 spin_lock_irqsave(&tc
->lock
, flags
);
2173 list_splice_init(&tc
->deferred_cells
, &cells
);
2174 spin_unlock_irqrestore(&tc
->lock
, flags
);
2176 if (list_empty(&cells
))
2180 count
= sort_cells(tc
->pool
, &cells
);
2182 for (i
= 0; i
< count
; i
++) {
2183 cell
= pool
->cell_sort_array
[i
];
2184 BUG_ON(!cell
->holder
);
2187 * If we've got no free new_mapping structs, and processing
2188 * this bio might require one, we pause until there are some
2189 * prepared mappings to process.
2191 if (ensure_next_mapping(pool
)) {
2192 for (j
= i
; j
< count
; j
++)
2193 list_add(&pool
->cell_sort_array
[j
]->user_list
, &cells
);
2195 spin_lock_irqsave(&tc
->lock
, flags
);
2196 list_splice(&cells
, &tc
->deferred_cells
);
2197 spin_unlock_irqrestore(&tc
->lock
, flags
);
2201 if (bio_op(cell
->holder
) == REQ_OP_DISCARD
)
2202 pool
->process_discard_cell(tc
, cell
);
2204 pool
->process_cell(tc
, cell
);
2206 } while (!list_empty(&cells
));
2209 static void thin_get(struct thin_c
*tc
);
2210 static void thin_put(struct thin_c
*tc
);
2213 * We can't hold rcu_read_lock() around code that can block. So we
2214 * find a thin with the rcu lock held; bump a refcount; then drop
2217 static struct thin_c
*get_first_thin(struct pool
*pool
)
2219 struct thin_c
*tc
= NULL
;
2222 if (!list_empty(&pool
->active_thins
)) {
2223 tc
= list_entry_rcu(pool
->active_thins
.next
, struct thin_c
, list
);
2231 static struct thin_c
*get_next_thin(struct pool
*pool
, struct thin_c
*tc
)
2233 struct thin_c
*old_tc
= tc
;
2236 list_for_each_entry_continue_rcu(tc
, &pool
->active_thins
, list
) {
2248 static void process_deferred_bios(struct pool
*pool
)
2250 unsigned long flags
;
2252 struct bio_list bios
;
2255 tc
= get_first_thin(pool
);
2257 process_thin_deferred_cells(tc
);
2258 process_thin_deferred_bios(tc
);
2259 tc
= get_next_thin(pool
, tc
);
2263 * If there are any deferred flush bios, we must commit
2264 * the metadata before issuing them.
2266 bio_list_init(&bios
);
2267 spin_lock_irqsave(&pool
->lock
, flags
);
2268 bio_list_merge(&bios
, &pool
->deferred_flush_bios
);
2269 bio_list_init(&pool
->deferred_flush_bios
);
2270 spin_unlock_irqrestore(&pool
->lock
, flags
);
2272 if (bio_list_empty(&bios
) &&
2273 !(dm_pool_changed_this_transaction(pool
->pmd
) && need_commit_due_to_time(pool
)))
2277 while ((bio
= bio_list_pop(&bios
)))
2281 pool
->last_commit_jiffies
= jiffies
;
2283 while ((bio
= bio_list_pop(&bios
)))
2284 generic_make_request(bio
);
2287 static void do_worker(struct work_struct
*ws
)
2289 struct pool
*pool
= container_of(ws
, struct pool
, worker
);
2291 throttle_work_start(&pool
->throttle
);
2292 dm_pool_issue_prefetches(pool
->pmd
);
2293 throttle_work_update(&pool
->throttle
);
2294 process_prepared(pool
, &pool
->prepared_mappings
, &pool
->process_prepared_mapping
);
2295 throttle_work_update(&pool
->throttle
);
2296 process_prepared(pool
, &pool
->prepared_discards
, &pool
->process_prepared_discard
);
2297 throttle_work_update(&pool
->throttle
);
2298 process_prepared(pool
, &pool
->prepared_discards_pt2
, &pool
->process_prepared_discard_pt2
);
2299 throttle_work_update(&pool
->throttle
);
2300 process_deferred_bios(pool
);
2301 throttle_work_complete(&pool
->throttle
);
2305 * We want to commit periodically so that not too much
2306 * unwritten data builds up.
2308 static void do_waker(struct work_struct
*ws
)
2310 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
, waker
);
2312 queue_delayed_work(pool
->wq
, &pool
->waker
, COMMIT_PERIOD
);
2315 static void notify_of_pool_mode_change_to_oods(struct pool
*pool
);
2318 * We're holding onto IO to allow userland time to react. After the
2319 * timeout either the pool will have been resized (and thus back in
2320 * PM_WRITE mode), or we degrade to PM_OUT_OF_DATA_SPACE w/ error_if_no_space.
2322 static void do_no_space_timeout(struct work_struct
*ws
)
2324 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
,
2327 if (get_pool_mode(pool
) == PM_OUT_OF_DATA_SPACE
&& !pool
->pf
.error_if_no_space
) {
2328 pool
->pf
.error_if_no_space
= true;
2329 notify_of_pool_mode_change_to_oods(pool
);
2330 error_retry_list_with_code(pool
, BLK_STS_NOSPC
);
2334 /*----------------------------------------------------------------*/
2337 struct work_struct worker
;
2338 struct completion complete
;
2341 static struct pool_work
*to_pool_work(struct work_struct
*ws
)
2343 return container_of(ws
, struct pool_work
, worker
);
2346 static void pool_work_complete(struct pool_work
*pw
)
2348 complete(&pw
->complete
);
2351 static void pool_work_wait(struct pool_work
*pw
, struct pool
*pool
,
2352 void (*fn
)(struct work_struct
*))
2354 INIT_WORK_ONSTACK(&pw
->worker
, fn
);
2355 init_completion(&pw
->complete
);
2356 queue_work(pool
->wq
, &pw
->worker
);
2357 wait_for_completion(&pw
->complete
);
2360 /*----------------------------------------------------------------*/
2362 struct noflush_work
{
2363 struct pool_work pw
;
2367 static struct noflush_work
*to_noflush(struct work_struct
*ws
)
2369 return container_of(to_pool_work(ws
), struct noflush_work
, pw
);
2372 static void do_noflush_start(struct work_struct
*ws
)
2374 struct noflush_work
*w
= to_noflush(ws
);
2375 w
->tc
->requeue_mode
= true;
2377 pool_work_complete(&w
->pw
);
2380 static void do_noflush_stop(struct work_struct
*ws
)
2382 struct noflush_work
*w
= to_noflush(ws
);
2383 w
->tc
->requeue_mode
= false;
2384 pool_work_complete(&w
->pw
);
2387 static void noflush_work(struct thin_c
*tc
, void (*fn
)(struct work_struct
*))
2389 struct noflush_work w
;
2392 pool_work_wait(&w
.pw
, tc
->pool
, fn
);
2395 /*----------------------------------------------------------------*/
2397 static enum pool_mode
get_pool_mode(struct pool
*pool
)
2399 return pool
->pf
.mode
;
2402 static void notify_of_pool_mode_change(struct pool
*pool
, const char *new_mode
)
2404 dm_table_event(pool
->ti
->table
);
2405 DMINFO("%s: switching pool to %s mode",
2406 dm_device_name(pool
->pool_md
), new_mode
);
2409 static void notify_of_pool_mode_change_to_oods(struct pool
*pool
)
2411 if (!pool
->pf
.error_if_no_space
)
2412 notify_of_pool_mode_change(pool
, "out-of-data-space (queue IO)");
2414 notify_of_pool_mode_change(pool
, "out-of-data-space (error IO)");
2417 static bool passdown_enabled(struct pool_c
*pt
)
2419 return pt
->adjusted_pf
.discard_passdown
;
2422 static void set_discard_callbacks(struct pool
*pool
)
2424 struct pool_c
*pt
= pool
->ti
->private;
2426 if (passdown_enabled(pt
)) {
2427 pool
->process_discard_cell
= process_discard_cell_passdown
;
2428 pool
->process_prepared_discard
= process_prepared_discard_passdown_pt1
;
2429 pool
->process_prepared_discard_pt2
= process_prepared_discard_passdown_pt2
;
2431 pool
->process_discard_cell
= process_discard_cell_no_passdown
;
2432 pool
->process_prepared_discard
= process_prepared_discard_no_passdown
;
2436 static void set_pool_mode(struct pool
*pool
, enum pool_mode new_mode
)
2438 struct pool_c
*pt
= pool
->ti
->private;
2439 bool needs_check
= dm_pool_metadata_needs_check(pool
->pmd
);
2440 enum pool_mode old_mode
= get_pool_mode(pool
);
2441 unsigned long no_space_timeout
= READ_ONCE(no_space_timeout_secs
) * HZ
;
2444 * Never allow the pool to transition to PM_WRITE mode if user
2445 * intervention is required to verify metadata and data consistency.
2447 if (new_mode
== PM_WRITE
&& needs_check
) {
2448 DMERR("%s: unable to switch pool to write mode until repaired.",
2449 dm_device_name(pool
->pool_md
));
2450 if (old_mode
!= new_mode
)
2451 new_mode
= old_mode
;
2453 new_mode
= PM_READ_ONLY
;
2456 * If we were in PM_FAIL mode, rollback of metadata failed. We're
2457 * not going to recover without a thin_repair. So we never let the
2458 * pool move out of the old mode.
2460 if (old_mode
== PM_FAIL
)
2461 new_mode
= old_mode
;
2465 if (old_mode
!= new_mode
)
2466 notify_of_pool_mode_change(pool
, "failure");
2467 dm_pool_metadata_read_only(pool
->pmd
);
2468 pool
->process_bio
= process_bio_fail
;
2469 pool
->process_discard
= process_bio_fail
;
2470 pool
->process_cell
= process_cell_fail
;
2471 pool
->process_discard_cell
= process_cell_fail
;
2472 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
2473 pool
->process_prepared_discard
= process_prepared_discard_fail
;
2475 error_retry_list(pool
);
2479 if (old_mode
!= new_mode
)
2480 notify_of_pool_mode_change(pool
, "read-only");
2481 dm_pool_metadata_read_only(pool
->pmd
);
2482 pool
->process_bio
= process_bio_read_only
;
2483 pool
->process_discard
= process_bio_success
;
2484 pool
->process_cell
= process_cell_read_only
;
2485 pool
->process_discard_cell
= process_cell_success
;
2486 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
2487 pool
->process_prepared_discard
= process_prepared_discard_success
;
2489 error_retry_list(pool
);
2492 case PM_OUT_OF_DATA_SPACE
:
2494 * Ideally we'd never hit this state; the low water mark
2495 * would trigger userland to extend the pool before we
2496 * completely run out of data space. However, many small
2497 * IOs to unprovisioned space can consume data space at an
2498 * alarming rate. Adjust your low water mark if you're
2499 * frequently seeing this mode.
2501 if (old_mode
!= new_mode
)
2502 notify_of_pool_mode_change_to_oods(pool
);
2503 pool
->out_of_data_space
= true;
2504 pool
->process_bio
= process_bio_read_only
;
2505 pool
->process_discard
= process_discard_bio
;
2506 pool
->process_cell
= process_cell_read_only
;
2507 pool
->process_prepared_mapping
= process_prepared_mapping
;
2508 set_discard_callbacks(pool
);
2510 if (!pool
->pf
.error_if_no_space
&& no_space_timeout
)
2511 queue_delayed_work(pool
->wq
, &pool
->no_space_timeout
, no_space_timeout
);
2515 if (old_mode
!= new_mode
)
2516 notify_of_pool_mode_change(pool
, "write");
2517 pool
->out_of_data_space
= false;
2518 pool
->pf
.error_if_no_space
= pt
->requested_pf
.error_if_no_space
;
2519 dm_pool_metadata_read_write(pool
->pmd
);
2520 pool
->process_bio
= process_bio
;
2521 pool
->process_discard
= process_discard_bio
;
2522 pool
->process_cell
= process_cell
;
2523 pool
->process_prepared_mapping
= process_prepared_mapping
;
2524 set_discard_callbacks(pool
);
2528 pool
->pf
.mode
= new_mode
;
2530 * The pool mode may have changed, sync it so bind_control_target()
2531 * doesn't cause an unexpected mode transition on resume.
2533 pt
->adjusted_pf
.mode
= new_mode
;
2536 static void abort_transaction(struct pool
*pool
)
2538 const char *dev_name
= dm_device_name(pool
->pool_md
);
2540 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name
);
2541 if (dm_pool_abort_metadata(pool
->pmd
)) {
2542 DMERR("%s: failed to abort metadata transaction", dev_name
);
2543 set_pool_mode(pool
, PM_FAIL
);
2546 if (dm_pool_metadata_set_needs_check(pool
->pmd
)) {
2547 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name
);
2548 set_pool_mode(pool
, PM_FAIL
);
2552 static void metadata_operation_failed(struct pool
*pool
, const char *op
, int r
)
2554 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
2555 dm_device_name(pool
->pool_md
), op
, r
);
2557 abort_transaction(pool
);
2558 set_pool_mode(pool
, PM_READ_ONLY
);
2561 /*----------------------------------------------------------------*/
2564 * Mapping functions.
2568 * Called only while mapping a thin bio to hand it over to the workqueue.
2570 static void thin_defer_bio(struct thin_c
*tc
, struct bio
*bio
)
2572 unsigned long flags
;
2573 struct pool
*pool
= tc
->pool
;
2575 spin_lock_irqsave(&tc
->lock
, flags
);
2576 bio_list_add(&tc
->deferred_bio_list
, bio
);
2577 spin_unlock_irqrestore(&tc
->lock
, flags
);
2582 static void thin_defer_bio_with_throttle(struct thin_c
*tc
, struct bio
*bio
)
2584 struct pool
*pool
= tc
->pool
;
2586 throttle_lock(&pool
->throttle
);
2587 thin_defer_bio(tc
, bio
);
2588 throttle_unlock(&pool
->throttle
);
2591 static void thin_defer_cell(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
2593 unsigned long flags
;
2594 struct pool
*pool
= tc
->pool
;
2596 throttle_lock(&pool
->throttle
);
2597 spin_lock_irqsave(&tc
->lock
, flags
);
2598 list_add_tail(&cell
->user_list
, &tc
->deferred_cells
);
2599 spin_unlock_irqrestore(&tc
->lock
, flags
);
2600 throttle_unlock(&pool
->throttle
);
2605 static void thin_hook_bio(struct thin_c
*tc
, struct bio
*bio
)
2607 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
2610 h
->shared_read_entry
= NULL
;
2611 h
->all_io_entry
= NULL
;
2612 h
->overwrite_mapping
= NULL
;
2617 * Non-blocking function called from the thin target's map function.
2619 static int thin_bio_map(struct dm_target
*ti
, struct bio
*bio
)
2622 struct thin_c
*tc
= ti
->private;
2623 dm_block_t block
= get_bio_block(tc
, bio
);
2624 struct dm_thin_device
*td
= tc
->td
;
2625 struct dm_thin_lookup_result result
;
2626 struct dm_bio_prison_cell
*virt_cell
, *data_cell
;
2627 struct dm_cell_key key
;
2629 thin_hook_bio(tc
, bio
);
2631 if (tc
->requeue_mode
) {
2632 bio
->bi_status
= BLK_STS_DM_REQUEUE
;
2634 return DM_MAPIO_SUBMITTED
;
2637 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
2639 return DM_MAPIO_SUBMITTED
;
2642 if (op_is_flush(bio
->bi_opf
) || bio_op(bio
) == REQ_OP_DISCARD
) {
2643 thin_defer_bio_with_throttle(tc
, bio
);
2644 return DM_MAPIO_SUBMITTED
;
2648 * We must hold the virtual cell before doing the lookup, otherwise
2649 * there's a race with discard.
2651 build_virtual_key(tc
->td
, block
, &key
);
2652 if (bio_detain(tc
->pool
, &key
, bio
, &virt_cell
))
2653 return DM_MAPIO_SUBMITTED
;
2655 r
= dm_thin_find_block(td
, block
, 0, &result
);
2658 * Note that we defer readahead too.
2662 if (unlikely(result
.shared
)) {
2664 * We have a race condition here between the
2665 * result.shared value returned by the lookup and
2666 * snapshot creation, which may cause new
2669 * To avoid this always quiesce the origin before
2670 * taking the snap. You want to do this anyway to
2671 * ensure a consistent application view
2674 * More distant ancestors are irrelevant. The
2675 * shared flag will be set in their case.
2677 thin_defer_cell(tc
, virt_cell
);
2678 return DM_MAPIO_SUBMITTED
;
2681 build_data_key(tc
->td
, result
.block
, &key
);
2682 if (bio_detain(tc
->pool
, &key
, bio
, &data_cell
)) {
2683 cell_defer_no_holder(tc
, virt_cell
);
2684 return DM_MAPIO_SUBMITTED
;
2687 inc_all_io_entry(tc
->pool
, bio
);
2688 cell_defer_no_holder(tc
, data_cell
);
2689 cell_defer_no_holder(tc
, virt_cell
);
2691 remap(tc
, bio
, result
.block
);
2692 return DM_MAPIO_REMAPPED
;
2696 thin_defer_cell(tc
, virt_cell
);
2697 return DM_MAPIO_SUBMITTED
;
2701 * Must always call bio_io_error on failure.
2702 * dm_thin_find_block can fail with -EINVAL if the
2703 * pool is switched to fail-io mode.
2706 cell_defer_no_holder(tc
, virt_cell
);
2707 return DM_MAPIO_SUBMITTED
;
2711 static int pool_is_congested(struct dm_target_callbacks
*cb
, int bdi_bits
)
2713 struct pool_c
*pt
= container_of(cb
, struct pool_c
, callbacks
);
2714 struct request_queue
*q
;
2716 if (get_pool_mode(pt
->pool
) == PM_OUT_OF_DATA_SPACE
)
2719 q
= bdev_get_queue(pt
->data_dev
->bdev
);
2720 return bdi_congested(q
->backing_dev_info
, bdi_bits
);
2723 static void requeue_bios(struct pool
*pool
)
2725 unsigned long flags
;
2729 list_for_each_entry_rcu(tc
, &pool
->active_thins
, list
) {
2730 spin_lock_irqsave(&tc
->lock
, flags
);
2731 bio_list_merge(&tc
->deferred_bio_list
, &tc
->retry_on_resume_list
);
2732 bio_list_init(&tc
->retry_on_resume_list
);
2733 spin_unlock_irqrestore(&tc
->lock
, flags
);
2738 /*----------------------------------------------------------------
2739 * Binding of control targets to a pool object
2740 *--------------------------------------------------------------*/
2741 static bool data_dev_supports_discard(struct pool_c
*pt
)
2743 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
2745 return q
&& blk_queue_discard(q
);
2748 static bool is_factor(sector_t block_size
, uint32_t n
)
2750 return !sector_div(block_size
, n
);
2754 * If discard_passdown was enabled verify that the data device
2755 * supports discards. Disable discard_passdown if not.
2757 static void disable_passdown_if_not_supported(struct pool_c
*pt
)
2759 struct pool
*pool
= pt
->pool
;
2760 struct block_device
*data_bdev
= pt
->data_dev
->bdev
;
2761 struct queue_limits
*data_limits
= &bdev_get_queue(data_bdev
)->limits
;
2762 const char *reason
= NULL
;
2763 char buf
[BDEVNAME_SIZE
];
2765 if (!pt
->adjusted_pf
.discard_passdown
)
2768 if (!data_dev_supports_discard(pt
))
2769 reason
= "discard unsupported";
2771 else if (data_limits
->max_discard_sectors
< pool
->sectors_per_block
)
2772 reason
= "max discard sectors smaller than a block";
2775 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev
, buf
), reason
);
2776 pt
->adjusted_pf
.discard_passdown
= false;
2780 static int bind_control_target(struct pool
*pool
, struct dm_target
*ti
)
2782 struct pool_c
*pt
= ti
->private;
2785 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
2787 enum pool_mode old_mode
= get_pool_mode(pool
);
2788 enum pool_mode new_mode
= pt
->adjusted_pf
.mode
;
2791 * Don't change the pool's mode until set_pool_mode() below.
2792 * Otherwise the pool's process_* function pointers may
2793 * not match the desired pool mode.
2795 pt
->adjusted_pf
.mode
= old_mode
;
2798 pool
->pf
= pt
->adjusted_pf
;
2799 pool
->low_water_blocks
= pt
->low_water_blocks
;
2801 set_pool_mode(pool
, new_mode
);
2806 static void unbind_control_target(struct pool
*pool
, struct dm_target
*ti
)
2812 /*----------------------------------------------------------------
2814 *--------------------------------------------------------------*/
2815 /* Initialize pool features. */
2816 static void pool_features_init(struct pool_features
*pf
)
2818 pf
->mode
= PM_WRITE
;
2819 pf
->zero_new_blocks
= true;
2820 pf
->discard_enabled
= true;
2821 pf
->discard_passdown
= true;
2822 pf
->error_if_no_space
= false;
2825 static void __pool_destroy(struct pool
*pool
)
2827 __pool_table_remove(pool
);
2829 vfree(pool
->cell_sort_array
);
2830 if (dm_pool_metadata_close(pool
->pmd
) < 0)
2831 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
2833 dm_bio_prison_destroy(pool
->prison
);
2834 dm_kcopyd_client_destroy(pool
->copier
);
2837 destroy_workqueue(pool
->wq
);
2839 if (pool
->next_mapping
)
2840 mempool_free(pool
->next_mapping
, pool
->mapping_pool
);
2841 mempool_destroy(pool
->mapping_pool
);
2842 dm_deferred_set_destroy(pool
->shared_read_ds
);
2843 dm_deferred_set_destroy(pool
->all_io_ds
);
2847 static struct kmem_cache
*_new_mapping_cache
;
2849 static struct pool
*pool_create(struct mapped_device
*pool_md
,
2850 struct block_device
*metadata_dev
,
2851 unsigned long block_size
,
2852 int read_only
, char **error
)
2857 struct dm_pool_metadata
*pmd
;
2858 bool format_device
= read_only
? false : true;
2860 pmd
= dm_pool_metadata_open(metadata_dev
, block_size
, format_device
);
2862 *error
= "Error creating metadata object";
2863 return (struct pool
*)pmd
;
2866 pool
= kmalloc(sizeof(*pool
), GFP_KERNEL
);
2868 *error
= "Error allocating memory for pool";
2869 err_p
= ERR_PTR(-ENOMEM
);
2874 pool
->sectors_per_block
= block_size
;
2875 if (block_size
& (block_size
- 1))
2876 pool
->sectors_per_block_shift
= -1;
2878 pool
->sectors_per_block_shift
= __ffs(block_size
);
2879 pool
->low_water_blocks
= 0;
2880 pool_features_init(&pool
->pf
);
2881 pool
->prison
= dm_bio_prison_create();
2882 if (!pool
->prison
) {
2883 *error
= "Error creating pool's bio prison";
2884 err_p
= ERR_PTR(-ENOMEM
);
2888 pool
->copier
= dm_kcopyd_client_create(&dm_kcopyd_throttle
);
2889 if (IS_ERR(pool
->copier
)) {
2890 r
= PTR_ERR(pool
->copier
);
2891 *error
= "Error creating pool's kcopyd client";
2893 goto bad_kcopyd_client
;
2897 * Create singlethreaded workqueue that will service all devices
2898 * that use this metadata.
2900 pool
->wq
= alloc_ordered_workqueue("dm-" DM_MSG_PREFIX
, WQ_MEM_RECLAIM
);
2902 *error
= "Error creating pool's workqueue";
2903 err_p
= ERR_PTR(-ENOMEM
);
2907 throttle_init(&pool
->throttle
);
2908 INIT_WORK(&pool
->worker
, do_worker
);
2909 INIT_DELAYED_WORK(&pool
->waker
, do_waker
);
2910 INIT_DELAYED_WORK(&pool
->no_space_timeout
, do_no_space_timeout
);
2911 spin_lock_init(&pool
->lock
);
2912 bio_list_init(&pool
->deferred_flush_bios
);
2913 INIT_LIST_HEAD(&pool
->prepared_mappings
);
2914 INIT_LIST_HEAD(&pool
->prepared_discards
);
2915 INIT_LIST_HEAD(&pool
->prepared_discards_pt2
);
2916 INIT_LIST_HEAD(&pool
->active_thins
);
2917 pool
->low_water_triggered
= false;
2918 pool
->suspended
= true;
2919 pool
->out_of_data_space
= false;
2921 pool
->shared_read_ds
= dm_deferred_set_create();
2922 if (!pool
->shared_read_ds
) {
2923 *error
= "Error creating pool's shared read deferred set";
2924 err_p
= ERR_PTR(-ENOMEM
);
2925 goto bad_shared_read_ds
;
2928 pool
->all_io_ds
= dm_deferred_set_create();
2929 if (!pool
->all_io_ds
) {
2930 *error
= "Error creating pool's all io deferred set";
2931 err_p
= ERR_PTR(-ENOMEM
);
2935 pool
->next_mapping
= NULL
;
2936 pool
->mapping_pool
= mempool_create_slab_pool(MAPPING_POOL_SIZE
,
2937 _new_mapping_cache
);
2938 if (!pool
->mapping_pool
) {
2939 *error
= "Error creating pool's mapping mempool";
2940 err_p
= ERR_PTR(-ENOMEM
);
2941 goto bad_mapping_pool
;
2944 pool
->cell_sort_array
= vmalloc(sizeof(*pool
->cell_sort_array
) * CELL_SORT_ARRAY_SIZE
);
2945 if (!pool
->cell_sort_array
) {
2946 *error
= "Error allocating cell sort array";
2947 err_p
= ERR_PTR(-ENOMEM
);
2948 goto bad_sort_array
;
2951 pool
->ref_count
= 1;
2952 pool
->last_commit_jiffies
= jiffies
;
2953 pool
->pool_md
= pool_md
;
2954 pool
->md_dev
= metadata_dev
;
2955 __pool_table_insert(pool
);
2960 mempool_destroy(pool
->mapping_pool
);
2962 dm_deferred_set_destroy(pool
->all_io_ds
);
2964 dm_deferred_set_destroy(pool
->shared_read_ds
);
2966 destroy_workqueue(pool
->wq
);
2968 dm_kcopyd_client_destroy(pool
->copier
);
2970 dm_bio_prison_destroy(pool
->prison
);
2974 if (dm_pool_metadata_close(pmd
))
2975 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
2980 static void __pool_inc(struct pool
*pool
)
2982 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
2986 static void __pool_dec(struct pool
*pool
)
2988 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
2989 BUG_ON(!pool
->ref_count
);
2990 if (!--pool
->ref_count
)
2991 __pool_destroy(pool
);
2994 static struct pool
*__pool_find(struct mapped_device
*pool_md
,
2995 struct block_device
*metadata_dev
,
2996 unsigned long block_size
, int read_only
,
2997 char **error
, int *created
)
2999 struct pool
*pool
= __pool_table_lookup_metadata_dev(metadata_dev
);
3002 if (pool
->pool_md
!= pool_md
) {
3003 *error
= "metadata device already in use by a pool";
3004 return ERR_PTR(-EBUSY
);
3009 pool
= __pool_table_lookup(pool_md
);
3011 if (pool
->md_dev
!= metadata_dev
) {
3012 *error
= "different pool cannot replace a pool";
3013 return ERR_PTR(-EINVAL
);
3018 pool
= pool_create(pool_md
, metadata_dev
, block_size
, read_only
, error
);
3026 /*----------------------------------------------------------------
3027 * Pool target methods
3028 *--------------------------------------------------------------*/
3029 static void pool_dtr(struct dm_target
*ti
)
3031 struct pool_c
*pt
= ti
->private;
3033 mutex_lock(&dm_thin_pool_table
.mutex
);
3035 unbind_control_target(pt
->pool
, ti
);
3036 __pool_dec(pt
->pool
);
3037 dm_put_device(ti
, pt
->metadata_dev
);
3038 dm_put_device(ti
, pt
->data_dev
);
3041 mutex_unlock(&dm_thin_pool_table
.mutex
);
3044 static int parse_pool_features(struct dm_arg_set
*as
, struct pool_features
*pf
,
3045 struct dm_target
*ti
)
3049 const char *arg_name
;
3051 static const struct dm_arg _args
[] = {
3052 {0, 4, "Invalid number of pool feature arguments"},
3056 * No feature arguments supplied.
3061 r
= dm_read_arg_group(_args
, as
, &argc
, &ti
->error
);
3065 while (argc
&& !r
) {
3066 arg_name
= dm_shift_arg(as
);
3069 if (!strcasecmp(arg_name
, "skip_block_zeroing"))
3070 pf
->zero_new_blocks
= false;
3072 else if (!strcasecmp(arg_name
, "ignore_discard"))
3073 pf
->discard_enabled
= false;
3075 else if (!strcasecmp(arg_name
, "no_discard_passdown"))
3076 pf
->discard_passdown
= false;
3078 else if (!strcasecmp(arg_name
, "read_only"))
3079 pf
->mode
= PM_READ_ONLY
;
3081 else if (!strcasecmp(arg_name
, "error_if_no_space"))
3082 pf
->error_if_no_space
= true;
3085 ti
->error
= "Unrecognised pool feature requested";
3094 static void metadata_low_callback(void *context
)
3096 struct pool
*pool
= context
;
3098 DMWARN("%s: reached low water mark for metadata device: sending event.",
3099 dm_device_name(pool
->pool_md
));
3101 dm_table_event(pool
->ti
->table
);
3104 static sector_t
get_dev_size(struct block_device
*bdev
)
3106 return i_size_read(bdev
->bd_inode
) >> SECTOR_SHIFT
;
3109 static void warn_if_metadata_device_too_big(struct block_device
*bdev
)
3111 sector_t metadata_dev_size
= get_dev_size(bdev
);
3112 char buffer
[BDEVNAME_SIZE
];
3114 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS_WARNING
)
3115 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
3116 bdevname(bdev
, buffer
), THIN_METADATA_MAX_SECTORS
);
3119 static sector_t
get_metadata_dev_size(struct block_device
*bdev
)
3121 sector_t metadata_dev_size
= get_dev_size(bdev
);
3123 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS
)
3124 metadata_dev_size
= THIN_METADATA_MAX_SECTORS
;
3126 return metadata_dev_size
;
3129 static dm_block_t
get_metadata_dev_size_in_blocks(struct block_device
*bdev
)
3131 sector_t metadata_dev_size
= get_metadata_dev_size(bdev
);
3133 sector_div(metadata_dev_size
, THIN_METADATA_BLOCK_SIZE
);
3135 return metadata_dev_size
;
3139 * When a metadata threshold is crossed a dm event is triggered, and
3140 * userland should respond by growing the metadata device. We could let
3141 * userland set the threshold, like we do with the data threshold, but I'm
3142 * not sure they know enough to do this well.
3144 static dm_block_t
calc_metadata_threshold(struct pool_c
*pt
)
3147 * 4M is ample for all ops with the possible exception of thin
3148 * device deletion which is harmless if it fails (just retry the
3149 * delete after you've grown the device).
3151 dm_block_t quarter
= get_metadata_dev_size_in_blocks(pt
->metadata_dev
->bdev
) / 4;
3152 return min((dm_block_t
)1024ULL /* 4M */, quarter
);
3156 * thin-pool <metadata dev> <data dev>
3157 * <data block size (sectors)>
3158 * <low water mark (blocks)>
3159 * [<#feature args> [<arg>]*]
3161 * Optional feature arguments are:
3162 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
3163 * ignore_discard: disable discard
3164 * no_discard_passdown: don't pass discards down to the data device
3165 * read_only: Don't allow any changes to be made to the pool metadata.
3166 * error_if_no_space: error IOs, instead of queueing, if no space.
3168 static int pool_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
3170 int r
, pool_created
= 0;
3173 struct pool_features pf
;
3174 struct dm_arg_set as
;
3175 struct dm_dev
*data_dev
;
3176 unsigned long block_size
;
3177 dm_block_t low_water_blocks
;
3178 struct dm_dev
*metadata_dev
;
3179 fmode_t metadata_mode
;
3182 * FIXME Remove validation from scope of lock.
3184 mutex_lock(&dm_thin_pool_table
.mutex
);
3187 ti
->error
= "Invalid argument count";
3196 * Set default pool features.
3198 pool_features_init(&pf
);
3200 dm_consume_args(&as
, 4);
3201 r
= parse_pool_features(&as
, &pf
, ti
);
3205 metadata_mode
= FMODE_READ
| ((pf
.mode
== PM_READ_ONLY
) ? 0 : FMODE_WRITE
);
3206 r
= dm_get_device(ti
, argv
[0], metadata_mode
, &metadata_dev
);
3208 ti
->error
= "Error opening metadata block device";
3211 warn_if_metadata_device_too_big(metadata_dev
->bdev
);
3213 r
= dm_get_device(ti
, argv
[1], FMODE_READ
| FMODE_WRITE
, &data_dev
);
3215 ti
->error
= "Error getting data device";
3219 if (kstrtoul(argv
[2], 10, &block_size
) || !block_size
||
3220 block_size
< DATA_DEV_BLOCK_SIZE_MIN_SECTORS
||
3221 block_size
> DATA_DEV_BLOCK_SIZE_MAX_SECTORS
||
3222 block_size
& (DATA_DEV_BLOCK_SIZE_MIN_SECTORS
- 1)) {
3223 ti
->error
= "Invalid block size";
3228 if (kstrtoull(argv
[3], 10, (unsigned long long *)&low_water_blocks
)) {
3229 ti
->error
= "Invalid low water mark";
3234 pt
= kzalloc(sizeof(*pt
), GFP_KERNEL
);
3240 pool
= __pool_find(dm_table_get_md(ti
->table
), metadata_dev
->bdev
,
3241 block_size
, pf
.mode
== PM_READ_ONLY
, &ti
->error
, &pool_created
);
3248 * 'pool_created' reflects whether this is the first table load.
3249 * Top level discard support is not allowed to be changed after
3250 * initial load. This would require a pool reload to trigger thin
3253 if (!pool_created
&& pf
.discard_enabled
!= pool
->pf
.discard_enabled
) {
3254 ti
->error
= "Discard support cannot be disabled once enabled";
3256 goto out_flags_changed
;
3261 pt
->metadata_dev
= metadata_dev
;
3262 pt
->data_dev
= data_dev
;
3263 pt
->low_water_blocks
= low_water_blocks
;
3264 pt
->adjusted_pf
= pt
->requested_pf
= pf
;
3265 ti
->num_flush_bios
= 1;
3268 * Only need to enable discards if the pool should pass
3269 * them down to the data device. The thin device's discard
3270 * processing will cause mappings to be removed from the btree.
3272 if (pf
.discard_enabled
&& pf
.discard_passdown
) {
3273 ti
->num_discard_bios
= 1;
3276 * Setting 'discards_supported' circumvents the normal
3277 * stacking of discard limits (this keeps the pool and
3278 * thin devices' discard limits consistent).
3280 ti
->discards_supported
= true;
3284 r
= dm_pool_register_metadata_threshold(pt
->pool
->pmd
,
3285 calc_metadata_threshold(pt
),
3286 metadata_low_callback
,
3289 goto out_flags_changed
;
3291 pt
->callbacks
.congested_fn
= pool_is_congested
;
3292 dm_table_add_target_callbacks(ti
->table
, &pt
->callbacks
);
3294 mutex_unlock(&dm_thin_pool_table
.mutex
);
3303 dm_put_device(ti
, data_dev
);
3305 dm_put_device(ti
, metadata_dev
);
3307 mutex_unlock(&dm_thin_pool_table
.mutex
);
3312 static int pool_map(struct dm_target
*ti
, struct bio
*bio
)
3315 struct pool_c
*pt
= ti
->private;
3316 struct pool
*pool
= pt
->pool
;
3317 unsigned long flags
;
3320 * As this is a singleton target, ti->begin is always zero.
3322 spin_lock_irqsave(&pool
->lock
, flags
);
3323 bio_set_dev(bio
, pt
->data_dev
->bdev
);
3324 r
= DM_MAPIO_REMAPPED
;
3325 spin_unlock_irqrestore(&pool
->lock
, flags
);
3330 static int maybe_resize_data_dev(struct dm_target
*ti
, bool *need_commit
)
3333 struct pool_c
*pt
= ti
->private;
3334 struct pool
*pool
= pt
->pool
;
3335 sector_t data_size
= ti
->len
;
3336 dm_block_t sb_data_size
;
3338 *need_commit
= false;
3340 (void) sector_div(data_size
, pool
->sectors_per_block
);
3342 r
= dm_pool_get_data_dev_size(pool
->pmd
, &sb_data_size
);
3344 DMERR("%s: failed to retrieve data device size",
3345 dm_device_name(pool
->pool_md
));
3349 if (data_size
< sb_data_size
) {
3350 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
3351 dm_device_name(pool
->pool_md
),
3352 (unsigned long long)data_size
, sb_data_size
);
3355 } else if (data_size
> sb_data_size
) {
3356 if (dm_pool_metadata_needs_check(pool
->pmd
)) {
3357 DMERR("%s: unable to grow the data device until repaired.",
3358 dm_device_name(pool
->pool_md
));
3363 DMINFO("%s: growing the data device from %llu to %llu blocks",
3364 dm_device_name(pool
->pool_md
),
3365 sb_data_size
, (unsigned long long)data_size
);
3366 r
= dm_pool_resize_data_dev(pool
->pmd
, data_size
);
3368 metadata_operation_failed(pool
, "dm_pool_resize_data_dev", r
);
3372 *need_commit
= true;
3378 static int maybe_resize_metadata_dev(struct dm_target
*ti
, bool *need_commit
)
3381 struct pool_c
*pt
= ti
->private;
3382 struct pool
*pool
= pt
->pool
;
3383 dm_block_t metadata_dev_size
, sb_metadata_dev_size
;
3385 *need_commit
= false;
3387 metadata_dev_size
= get_metadata_dev_size_in_blocks(pool
->md_dev
);
3389 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &sb_metadata_dev_size
);
3391 DMERR("%s: failed to retrieve metadata device size",
3392 dm_device_name(pool
->pool_md
));
3396 if (metadata_dev_size
< sb_metadata_dev_size
) {
3397 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
3398 dm_device_name(pool
->pool_md
),
3399 metadata_dev_size
, sb_metadata_dev_size
);
3402 } else if (metadata_dev_size
> sb_metadata_dev_size
) {
3403 if (dm_pool_metadata_needs_check(pool
->pmd
)) {
3404 DMERR("%s: unable to grow the metadata device until repaired.",
3405 dm_device_name(pool
->pool_md
));
3409 warn_if_metadata_device_too_big(pool
->md_dev
);
3410 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
3411 dm_device_name(pool
->pool_md
),
3412 sb_metadata_dev_size
, metadata_dev_size
);
3413 r
= dm_pool_resize_metadata_dev(pool
->pmd
, metadata_dev_size
);
3415 metadata_operation_failed(pool
, "dm_pool_resize_metadata_dev", r
);
3419 *need_commit
= true;
3426 * Retrieves the number of blocks of the data device from
3427 * the superblock and compares it to the actual device size,
3428 * thus resizing the data device in case it has grown.
3430 * This both copes with opening preallocated data devices in the ctr
3431 * being followed by a resume
3433 * calling the resume method individually after userspace has
3434 * grown the data device in reaction to a table event.
3436 static int pool_preresume(struct dm_target
*ti
)
3439 bool need_commit1
, need_commit2
;
3440 struct pool_c
*pt
= ti
->private;
3441 struct pool
*pool
= pt
->pool
;
3444 * Take control of the pool object.
3446 r
= bind_control_target(pool
, ti
);
3450 r
= maybe_resize_data_dev(ti
, &need_commit1
);
3454 r
= maybe_resize_metadata_dev(ti
, &need_commit2
);
3458 if (need_commit1
|| need_commit2
)
3459 (void) commit(pool
);
3464 static void pool_suspend_active_thins(struct pool
*pool
)
3468 /* Suspend all active thin devices */
3469 tc
= get_first_thin(pool
);
3471 dm_internal_suspend_noflush(tc
->thin_md
);
3472 tc
= get_next_thin(pool
, tc
);
3476 static void pool_resume_active_thins(struct pool
*pool
)
3480 /* Resume all active thin devices */
3481 tc
= get_first_thin(pool
);
3483 dm_internal_resume(tc
->thin_md
);
3484 tc
= get_next_thin(pool
, tc
);
3488 static void pool_resume(struct dm_target
*ti
)
3490 struct pool_c
*pt
= ti
->private;
3491 struct pool
*pool
= pt
->pool
;
3492 unsigned long flags
;
3495 * Must requeue active_thins' bios and then resume
3496 * active_thins _before_ clearing 'suspend' flag.
3499 pool_resume_active_thins(pool
);
3501 spin_lock_irqsave(&pool
->lock
, flags
);
3502 pool
->low_water_triggered
= false;
3503 pool
->suspended
= false;
3504 spin_unlock_irqrestore(&pool
->lock
, flags
);
3506 do_waker(&pool
->waker
.work
);
3509 static void pool_presuspend(struct dm_target
*ti
)
3511 struct pool_c
*pt
= ti
->private;
3512 struct pool
*pool
= pt
->pool
;
3513 unsigned long flags
;
3515 spin_lock_irqsave(&pool
->lock
, flags
);
3516 pool
->suspended
= true;
3517 spin_unlock_irqrestore(&pool
->lock
, flags
);
3519 pool_suspend_active_thins(pool
);
3522 static void pool_presuspend_undo(struct dm_target
*ti
)
3524 struct pool_c
*pt
= ti
->private;
3525 struct pool
*pool
= pt
->pool
;
3526 unsigned long flags
;
3528 pool_resume_active_thins(pool
);
3530 spin_lock_irqsave(&pool
->lock
, flags
);
3531 pool
->suspended
= false;
3532 spin_unlock_irqrestore(&pool
->lock
, flags
);
3535 static void pool_postsuspend(struct dm_target
*ti
)
3537 struct pool_c
*pt
= ti
->private;
3538 struct pool
*pool
= pt
->pool
;
3540 cancel_delayed_work_sync(&pool
->waker
);
3541 cancel_delayed_work_sync(&pool
->no_space_timeout
);
3542 flush_workqueue(pool
->wq
);
3543 (void) commit(pool
);
3546 static int check_arg_count(unsigned argc
, unsigned args_required
)
3548 if (argc
!= args_required
) {
3549 DMWARN("Message received with %u arguments instead of %u.",
3550 argc
, args_required
);
3557 static int read_dev_id(char *arg
, dm_thin_id
*dev_id
, int warning
)
3559 if (!kstrtoull(arg
, 10, (unsigned long long *)dev_id
) &&
3560 *dev_id
<= MAX_DEV_ID
)
3564 DMWARN("Message received with invalid device id: %s", arg
);
3569 static int process_create_thin_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
3574 r
= check_arg_count(argc
, 2);
3578 r
= read_dev_id(argv
[1], &dev_id
, 1);
3582 r
= dm_pool_create_thin(pool
->pmd
, dev_id
);
3584 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
3592 static int process_create_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
3595 dm_thin_id origin_dev_id
;
3598 r
= check_arg_count(argc
, 3);
3602 r
= read_dev_id(argv
[1], &dev_id
, 1);
3606 r
= read_dev_id(argv
[2], &origin_dev_id
, 1);
3610 r
= dm_pool_create_snap(pool
->pmd
, dev_id
, origin_dev_id
);
3612 DMWARN("Creation of new snapshot %s of device %s failed.",
3620 static int process_delete_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
3625 r
= check_arg_count(argc
, 2);
3629 r
= read_dev_id(argv
[1], &dev_id
, 1);
3633 r
= dm_pool_delete_thin_device(pool
->pmd
, dev_id
);
3635 DMWARN("Deletion of thin device %s failed.", argv
[1]);
3640 static int process_set_transaction_id_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
3642 dm_thin_id old_id
, new_id
;
3645 r
= check_arg_count(argc
, 3);
3649 if (kstrtoull(argv
[1], 10, (unsigned long long *)&old_id
)) {
3650 DMWARN("set_transaction_id message: Unrecognised id %s.", argv
[1]);
3654 if (kstrtoull(argv
[2], 10, (unsigned long long *)&new_id
)) {
3655 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv
[2]);
3659 r
= dm_pool_set_metadata_transaction_id(pool
->pmd
, old_id
, new_id
);
3661 DMWARN("Failed to change transaction id from %s to %s.",
3669 static int process_reserve_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
3673 r
= check_arg_count(argc
, 1);
3677 (void) commit(pool
);
3679 r
= dm_pool_reserve_metadata_snap(pool
->pmd
);
3681 DMWARN("reserve_metadata_snap message failed.");
3686 static int process_release_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
3690 r
= check_arg_count(argc
, 1);
3694 r
= dm_pool_release_metadata_snap(pool
->pmd
);
3696 DMWARN("release_metadata_snap message failed.");
3702 * Messages supported:
3703 * create_thin <dev_id>
3704 * create_snap <dev_id> <origin_id>
3706 * set_transaction_id <current_trans_id> <new_trans_id>
3707 * reserve_metadata_snap
3708 * release_metadata_snap
3710 static int pool_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
3713 struct pool_c
*pt
= ti
->private;
3714 struct pool
*pool
= pt
->pool
;
3716 if (get_pool_mode(pool
) >= PM_READ_ONLY
) {
3717 DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode",
3718 dm_device_name(pool
->pool_md
));
3722 if (!strcasecmp(argv
[0], "create_thin"))
3723 r
= process_create_thin_mesg(argc
, argv
, pool
);
3725 else if (!strcasecmp(argv
[0], "create_snap"))
3726 r
= process_create_snap_mesg(argc
, argv
, pool
);
3728 else if (!strcasecmp(argv
[0], "delete"))
3729 r
= process_delete_mesg(argc
, argv
, pool
);
3731 else if (!strcasecmp(argv
[0], "set_transaction_id"))
3732 r
= process_set_transaction_id_mesg(argc
, argv
, pool
);
3734 else if (!strcasecmp(argv
[0], "reserve_metadata_snap"))
3735 r
= process_reserve_metadata_snap_mesg(argc
, argv
, pool
);
3737 else if (!strcasecmp(argv
[0], "release_metadata_snap"))
3738 r
= process_release_metadata_snap_mesg(argc
, argv
, pool
);
3741 DMWARN("Unrecognised thin pool target message received: %s", argv
[0]);
3744 (void) commit(pool
);
3749 static void emit_flags(struct pool_features
*pf
, char *result
,
3750 unsigned sz
, unsigned maxlen
)
3752 unsigned count
= !pf
->zero_new_blocks
+ !pf
->discard_enabled
+
3753 !pf
->discard_passdown
+ (pf
->mode
== PM_READ_ONLY
) +
3754 pf
->error_if_no_space
;
3755 DMEMIT("%u ", count
);
3757 if (!pf
->zero_new_blocks
)
3758 DMEMIT("skip_block_zeroing ");
3760 if (!pf
->discard_enabled
)
3761 DMEMIT("ignore_discard ");
3763 if (!pf
->discard_passdown
)
3764 DMEMIT("no_discard_passdown ");
3766 if (pf
->mode
== PM_READ_ONLY
)
3767 DMEMIT("read_only ");
3769 if (pf
->error_if_no_space
)
3770 DMEMIT("error_if_no_space ");
3775 * <transaction id> <used metadata sectors>/<total metadata sectors>
3776 * <used data sectors>/<total data sectors> <held metadata root>
3777 * <pool mode> <discard config> <no space config> <needs_check>
3779 static void pool_status(struct dm_target
*ti
, status_type_t type
,
3780 unsigned status_flags
, char *result
, unsigned maxlen
)
3784 uint64_t transaction_id
;
3785 dm_block_t nr_free_blocks_data
;
3786 dm_block_t nr_free_blocks_metadata
;
3787 dm_block_t nr_blocks_data
;
3788 dm_block_t nr_blocks_metadata
;
3789 dm_block_t held_root
;
3790 char buf
[BDEVNAME_SIZE
];
3791 char buf2
[BDEVNAME_SIZE
];
3792 struct pool_c
*pt
= ti
->private;
3793 struct pool
*pool
= pt
->pool
;
3796 case STATUSTYPE_INFO
:
3797 if (get_pool_mode(pool
) == PM_FAIL
) {
3802 /* Commit to ensure statistics aren't out-of-date */
3803 if (!(status_flags
& DM_STATUS_NOFLUSH_FLAG
) && !dm_suspended(ti
))
3804 (void) commit(pool
);
3806 r
= dm_pool_get_metadata_transaction_id(pool
->pmd
, &transaction_id
);
3808 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
3809 dm_device_name(pool
->pool_md
), r
);
3813 r
= dm_pool_get_free_metadata_block_count(pool
->pmd
, &nr_free_blocks_metadata
);
3815 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
3816 dm_device_name(pool
->pool_md
), r
);
3820 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &nr_blocks_metadata
);
3822 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
3823 dm_device_name(pool
->pool_md
), r
);
3827 r
= dm_pool_get_free_block_count(pool
->pmd
, &nr_free_blocks_data
);
3829 DMERR("%s: dm_pool_get_free_block_count returned %d",
3830 dm_device_name(pool
->pool_md
), r
);
3834 r
= dm_pool_get_data_dev_size(pool
->pmd
, &nr_blocks_data
);
3836 DMERR("%s: dm_pool_get_data_dev_size returned %d",
3837 dm_device_name(pool
->pool_md
), r
);
3841 r
= dm_pool_get_metadata_snap(pool
->pmd
, &held_root
);
3843 DMERR("%s: dm_pool_get_metadata_snap returned %d",
3844 dm_device_name(pool
->pool_md
), r
);
3848 DMEMIT("%llu %llu/%llu %llu/%llu ",
3849 (unsigned long long)transaction_id
,
3850 (unsigned long long)(nr_blocks_metadata
- nr_free_blocks_metadata
),
3851 (unsigned long long)nr_blocks_metadata
,
3852 (unsigned long long)(nr_blocks_data
- nr_free_blocks_data
),
3853 (unsigned long long)nr_blocks_data
);
3856 DMEMIT("%llu ", held_root
);
3860 if (pool
->pf
.mode
== PM_OUT_OF_DATA_SPACE
)
3861 DMEMIT("out_of_data_space ");
3862 else if (pool
->pf
.mode
== PM_READ_ONLY
)
3867 if (!pool
->pf
.discard_enabled
)
3868 DMEMIT("ignore_discard ");
3869 else if (pool
->pf
.discard_passdown
)
3870 DMEMIT("discard_passdown ");
3872 DMEMIT("no_discard_passdown ");
3874 if (pool
->pf
.error_if_no_space
)
3875 DMEMIT("error_if_no_space ");
3877 DMEMIT("queue_if_no_space ");
3879 if (dm_pool_metadata_needs_check(pool
->pmd
))
3880 DMEMIT("needs_check ");
3886 case STATUSTYPE_TABLE
:
3887 DMEMIT("%s %s %lu %llu ",
3888 format_dev_t(buf
, pt
->metadata_dev
->bdev
->bd_dev
),
3889 format_dev_t(buf2
, pt
->data_dev
->bdev
->bd_dev
),
3890 (unsigned long)pool
->sectors_per_block
,
3891 (unsigned long long)pt
->low_water_blocks
);
3892 emit_flags(&pt
->requested_pf
, result
, sz
, maxlen
);
3901 static int pool_iterate_devices(struct dm_target
*ti
,
3902 iterate_devices_callout_fn fn
, void *data
)
3904 struct pool_c
*pt
= ti
->private;
3906 return fn(ti
, pt
->data_dev
, 0, ti
->len
, data
);
3909 static void pool_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
3911 struct pool_c
*pt
= ti
->private;
3912 struct pool
*pool
= pt
->pool
;
3913 sector_t io_opt_sectors
= limits
->io_opt
>> SECTOR_SHIFT
;
3916 * If max_sectors is smaller than pool->sectors_per_block adjust it
3917 * to the highest possible power-of-2 factor of pool->sectors_per_block.
3918 * This is especially beneficial when the pool's data device is a RAID
3919 * device that has a full stripe width that matches pool->sectors_per_block
3920 * -- because even though partial RAID stripe-sized IOs will be issued to a
3921 * single RAID stripe; when aggregated they will end on a full RAID stripe
3922 * boundary.. which avoids additional partial RAID stripe writes cascading
3924 if (limits
->max_sectors
< pool
->sectors_per_block
) {
3925 while (!is_factor(pool
->sectors_per_block
, limits
->max_sectors
)) {
3926 if ((limits
->max_sectors
& (limits
->max_sectors
- 1)) == 0)
3927 limits
->max_sectors
--;
3928 limits
->max_sectors
= rounddown_pow_of_two(limits
->max_sectors
);
3933 * If the system-determined stacked limits are compatible with the
3934 * pool's blocksize (io_opt is a factor) do not override them.
3936 if (io_opt_sectors
< pool
->sectors_per_block
||
3937 !is_factor(io_opt_sectors
, pool
->sectors_per_block
)) {
3938 if (is_factor(pool
->sectors_per_block
, limits
->max_sectors
))
3939 blk_limits_io_min(limits
, limits
->max_sectors
<< SECTOR_SHIFT
);
3941 blk_limits_io_min(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
3942 blk_limits_io_opt(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
3946 * pt->adjusted_pf is a staging area for the actual features to use.
3947 * They get transferred to the live pool in bind_control_target()
3948 * called from pool_preresume().
3950 if (!pt
->adjusted_pf
.discard_enabled
) {
3952 * Must explicitly disallow stacking discard limits otherwise the
3953 * block layer will stack them if pool's data device has support.
3954 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
3955 * user to see that, so make sure to set all discard limits to 0.
3957 limits
->discard_granularity
= 0;
3961 disable_passdown_if_not_supported(pt
);
3964 * The pool uses the same discard limits as the underlying data
3965 * device. DM core has already set this up.
3969 static struct target_type pool_target
= {
3970 .name
= "thin-pool",
3971 .features
= DM_TARGET_SINGLETON
| DM_TARGET_ALWAYS_WRITEABLE
|
3972 DM_TARGET_IMMUTABLE
,
3973 .version
= {1, 19, 0},
3974 .module
= THIS_MODULE
,
3978 .presuspend
= pool_presuspend
,
3979 .presuspend_undo
= pool_presuspend_undo
,
3980 .postsuspend
= pool_postsuspend
,
3981 .preresume
= pool_preresume
,
3982 .resume
= pool_resume
,
3983 .message
= pool_message
,
3984 .status
= pool_status
,
3985 .iterate_devices
= pool_iterate_devices
,
3986 .io_hints
= pool_io_hints
,
3989 /*----------------------------------------------------------------
3990 * Thin target methods
3991 *--------------------------------------------------------------*/
3992 static void thin_get(struct thin_c
*tc
)
3994 atomic_inc(&tc
->refcount
);
3997 static void thin_put(struct thin_c
*tc
)
3999 if (atomic_dec_and_test(&tc
->refcount
))
4000 complete(&tc
->can_destroy
);
4003 static void thin_dtr(struct dm_target
*ti
)
4005 struct thin_c
*tc
= ti
->private;
4006 unsigned long flags
;
4008 spin_lock_irqsave(&tc
->pool
->lock
, flags
);
4009 list_del_rcu(&tc
->list
);
4010 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
4014 wait_for_completion(&tc
->can_destroy
);
4016 mutex_lock(&dm_thin_pool_table
.mutex
);
4018 __pool_dec(tc
->pool
);
4019 dm_pool_close_thin_device(tc
->td
);
4020 dm_put_device(ti
, tc
->pool_dev
);
4022 dm_put_device(ti
, tc
->origin_dev
);
4025 mutex_unlock(&dm_thin_pool_table
.mutex
);
4029 * Thin target parameters:
4031 * <pool_dev> <dev_id> [origin_dev]
4033 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
4034 * dev_id: the internal device identifier
4035 * origin_dev: a device external to the pool that should act as the origin
4037 * If the pool device has discards disabled, they get disabled for the thin
4040 static int thin_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
4044 struct dm_dev
*pool_dev
, *origin_dev
;
4045 struct mapped_device
*pool_md
;
4046 unsigned long flags
;
4048 mutex_lock(&dm_thin_pool_table
.mutex
);
4050 if (argc
!= 2 && argc
!= 3) {
4051 ti
->error
= "Invalid argument count";
4056 tc
= ti
->private = kzalloc(sizeof(*tc
), GFP_KERNEL
);
4058 ti
->error
= "Out of memory";
4062 tc
->thin_md
= dm_table_get_md(ti
->table
);
4063 spin_lock_init(&tc
->lock
);
4064 INIT_LIST_HEAD(&tc
->deferred_cells
);
4065 bio_list_init(&tc
->deferred_bio_list
);
4066 bio_list_init(&tc
->retry_on_resume_list
);
4067 tc
->sort_bio_list
= RB_ROOT
;
4070 r
= dm_get_device(ti
, argv
[2], FMODE_READ
, &origin_dev
);
4072 ti
->error
= "Error opening origin device";
4073 goto bad_origin_dev
;
4075 tc
->origin_dev
= origin_dev
;
4078 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &pool_dev
);
4080 ti
->error
= "Error opening pool device";
4083 tc
->pool_dev
= pool_dev
;
4085 if (read_dev_id(argv
[1], (unsigned long long *)&tc
->dev_id
, 0)) {
4086 ti
->error
= "Invalid device id";
4091 pool_md
= dm_get_md(tc
->pool_dev
->bdev
->bd_dev
);
4093 ti
->error
= "Couldn't get pool mapped device";
4098 tc
->pool
= __pool_table_lookup(pool_md
);
4100 ti
->error
= "Couldn't find pool object";
4102 goto bad_pool_lookup
;
4104 __pool_inc(tc
->pool
);
4106 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
4107 ti
->error
= "Couldn't open thin device, Pool is in fail mode";
4112 r
= dm_pool_open_thin_device(tc
->pool
->pmd
, tc
->dev_id
, &tc
->td
);
4114 ti
->error
= "Couldn't open thin internal device";
4118 r
= dm_set_target_max_io_len(ti
, tc
->pool
->sectors_per_block
);
4122 ti
->num_flush_bios
= 1;
4123 ti
->flush_supported
= true;
4124 ti
->per_io_data_size
= sizeof(struct dm_thin_endio_hook
);
4126 /* In case the pool supports discards, pass them on. */
4127 if (tc
->pool
->pf
.discard_enabled
) {
4128 ti
->discards_supported
= true;
4129 ti
->num_discard_bios
= 1;
4130 ti
->split_discard_bios
= false;
4133 mutex_unlock(&dm_thin_pool_table
.mutex
);
4135 spin_lock_irqsave(&tc
->pool
->lock
, flags
);
4136 if (tc
->pool
->suspended
) {
4137 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
4138 mutex_lock(&dm_thin_pool_table
.mutex
); /* reacquire for __pool_dec */
4139 ti
->error
= "Unable to activate thin device while pool is suspended";
4143 atomic_set(&tc
->refcount
, 1);
4144 init_completion(&tc
->can_destroy
);
4145 list_add_tail_rcu(&tc
->list
, &tc
->pool
->active_thins
);
4146 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
4148 * This synchronize_rcu() call is needed here otherwise we risk a
4149 * wake_worker() call finding no bios to process (because the newly
4150 * added tc isn't yet visible). So this reduces latency since we
4151 * aren't then dependent on the periodic commit to wake_worker().
4160 dm_pool_close_thin_device(tc
->td
);
4162 __pool_dec(tc
->pool
);
4166 dm_put_device(ti
, tc
->pool_dev
);
4169 dm_put_device(ti
, tc
->origin_dev
);
4173 mutex_unlock(&dm_thin_pool_table
.mutex
);
4178 static int thin_map(struct dm_target
*ti
, struct bio
*bio
)
4180 bio
->bi_iter
.bi_sector
= dm_target_offset(ti
, bio
->bi_iter
.bi_sector
);
4182 return thin_bio_map(ti
, bio
);
4185 static int thin_endio(struct dm_target
*ti
, struct bio
*bio
,
4188 unsigned long flags
;
4189 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
4190 struct list_head work
;
4191 struct dm_thin_new_mapping
*m
, *tmp
;
4192 struct pool
*pool
= h
->tc
->pool
;
4194 if (h
->shared_read_entry
) {
4195 INIT_LIST_HEAD(&work
);
4196 dm_deferred_entry_dec(h
->shared_read_entry
, &work
);
4198 spin_lock_irqsave(&pool
->lock
, flags
);
4199 list_for_each_entry_safe(m
, tmp
, &work
, list
) {
4201 __complete_mapping_preparation(m
);
4203 spin_unlock_irqrestore(&pool
->lock
, flags
);
4206 if (h
->all_io_entry
) {
4207 INIT_LIST_HEAD(&work
);
4208 dm_deferred_entry_dec(h
->all_io_entry
, &work
);
4209 if (!list_empty(&work
)) {
4210 spin_lock_irqsave(&pool
->lock
, flags
);
4211 list_for_each_entry_safe(m
, tmp
, &work
, list
)
4212 list_add_tail(&m
->list
, &pool
->prepared_discards
);
4213 spin_unlock_irqrestore(&pool
->lock
, flags
);
4219 cell_defer_no_holder(h
->tc
, h
->cell
);
4221 return DM_ENDIO_DONE
;
4224 static void thin_presuspend(struct dm_target
*ti
)
4226 struct thin_c
*tc
= ti
->private;
4228 if (dm_noflush_suspending(ti
))
4229 noflush_work(tc
, do_noflush_start
);
4232 static void thin_postsuspend(struct dm_target
*ti
)
4234 struct thin_c
*tc
= ti
->private;
4237 * The dm_noflush_suspending flag has been cleared by now, so
4238 * unfortunately we must always run this.
4240 noflush_work(tc
, do_noflush_stop
);
4243 static int thin_preresume(struct dm_target
*ti
)
4245 struct thin_c
*tc
= ti
->private;
4248 tc
->origin_size
= get_dev_size(tc
->origin_dev
->bdev
);
4254 * <nr mapped sectors> <highest mapped sector>
4256 static void thin_status(struct dm_target
*ti
, status_type_t type
,
4257 unsigned status_flags
, char *result
, unsigned maxlen
)
4261 dm_block_t mapped
, highest
;
4262 char buf
[BDEVNAME_SIZE
];
4263 struct thin_c
*tc
= ti
->private;
4265 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
4274 case STATUSTYPE_INFO
:
4275 r
= dm_thin_get_mapped_count(tc
->td
, &mapped
);
4277 DMERR("dm_thin_get_mapped_count returned %d", r
);
4281 r
= dm_thin_get_highest_mapped_block(tc
->td
, &highest
);
4283 DMERR("dm_thin_get_highest_mapped_block returned %d", r
);
4287 DMEMIT("%llu ", mapped
* tc
->pool
->sectors_per_block
);
4289 DMEMIT("%llu", ((highest
+ 1) *
4290 tc
->pool
->sectors_per_block
) - 1);
4295 case STATUSTYPE_TABLE
:
4297 format_dev_t(buf
, tc
->pool_dev
->bdev
->bd_dev
),
4298 (unsigned long) tc
->dev_id
);
4300 DMEMIT(" %s", format_dev_t(buf
, tc
->origin_dev
->bdev
->bd_dev
));
4311 static int thin_iterate_devices(struct dm_target
*ti
,
4312 iterate_devices_callout_fn fn
, void *data
)
4315 struct thin_c
*tc
= ti
->private;
4316 struct pool
*pool
= tc
->pool
;
4319 * We can't call dm_pool_get_data_dev_size() since that blocks. So
4320 * we follow a more convoluted path through to the pool's target.
4323 return 0; /* nothing is bound */
4325 blocks
= pool
->ti
->len
;
4326 (void) sector_div(blocks
, pool
->sectors_per_block
);
4328 return fn(ti
, tc
->pool_dev
, 0, pool
->sectors_per_block
* blocks
, data
);
4333 static void thin_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
4335 struct thin_c
*tc
= ti
->private;
4336 struct pool
*pool
= tc
->pool
;
4338 if (!pool
->pf
.discard_enabled
)
4341 limits
->discard_granularity
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
4342 limits
->max_discard_sectors
= 2048 * 1024 * 16; /* 16G */
4345 static struct target_type thin_target
= {
4347 .version
= {1, 19, 0},
4348 .module
= THIS_MODULE
,
4352 .end_io
= thin_endio
,
4353 .preresume
= thin_preresume
,
4354 .presuspend
= thin_presuspend
,
4355 .postsuspend
= thin_postsuspend
,
4356 .status
= thin_status
,
4357 .iterate_devices
= thin_iterate_devices
,
4358 .io_hints
= thin_io_hints
,
4361 /*----------------------------------------------------------------*/
4363 static int __init
dm_thin_init(void)
4369 _new_mapping_cache
= KMEM_CACHE(dm_thin_new_mapping
, 0);
4370 if (!_new_mapping_cache
)
4373 r
= dm_register_target(&thin_target
);
4375 goto bad_new_mapping_cache
;
4377 r
= dm_register_target(&pool_target
);
4379 goto bad_thin_target
;
4384 dm_unregister_target(&thin_target
);
4385 bad_new_mapping_cache
:
4386 kmem_cache_destroy(_new_mapping_cache
);
4391 static void dm_thin_exit(void)
4393 dm_unregister_target(&thin_target
);
4394 dm_unregister_target(&pool_target
);
4396 kmem_cache_destroy(_new_mapping_cache
);
4399 module_init(dm_thin_init
);
4400 module_exit(dm_thin_exit
);
4402 module_param_named(no_space_timeout
, no_space_timeout_secs
, uint
, S_IRUGO
| S_IWUSR
);
4403 MODULE_PARM_DESC(no_space_timeout
, "Out of data space queue IO timeout in seconds");
4405 MODULE_DESCRIPTION(DM_NAME
" thin provisioning target");
4406 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
4407 MODULE_LICENSE("GPL");