2 * Copyright (C) 2011-2012 Red Hat UK.
4 * This file is released under the GPL.
7 #include "dm-thin-metadata.h"
8 #include "dm-bio-prison.h"
11 #include <linux/device-mapper.h>
12 #include <linux/dm-io.h>
13 #include <linux/dm-kcopyd.h>
14 #include <linux/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_error
)
387 op
->parent_bio
->bi_error
= 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
, int error_code
)
455 dm_cell_error(pool
->prison
, cell
, error_code
);
456 dm_bio_prison_free_cell(pool
->prison
, cell
);
459 static int get_pool_io_error_code(struct pool
*pool
)
461 return pool
->out_of_data_space
? -ENOSPC
: -EIO
;
464 static void cell_error(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
466 int error
= get_pool_io_error_code(pool
);
468 cell_error_with_code(pool
, cell
, error
);
471 static void cell_success(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
473 cell_error_with_code(pool
, cell
, 0);
476 static void cell_requeue(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
478 cell_error_with_code(pool
, cell
, DM_ENDIO_REQUEUE
);
481 /*----------------------------------------------------------------*/
484 * A global list of pools that uses a struct mapped_device as a key.
486 static struct dm_thin_pool_table
{
488 struct list_head pools
;
489 } dm_thin_pool_table
;
491 static void pool_table_init(void)
493 mutex_init(&dm_thin_pool_table
.mutex
);
494 INIT_LIST_HEAD(&dm_thin_pool_table
.pools
);
497 static void __pool_table_insert(struct pool
*pool
)
499 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
500 list_add(&pool
->list
, &dm_thin_pool_table
.pools
);
503 static void __pool_table_remove(struct pool
*pool
)
505 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
506 list_del(&pool
->list
);
509 static struct pool
*__pool_table_lookup(struct mapped_device
*md
)
511 struct pool
*pool
= NULL
, *tmp
;
513 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
515 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
516 if (tmp
->pool_md
== md
) {
525 static struct pool
*__pool_table_lookup_metadata_dev(struct block_device
*md_dev
)
527 struct pool
*pool
= NULL
, *tmp
;
529 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
531 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
532 if (tmp
->md_dev
== md_dev
) {
541 /*----------------------------------------------------------------*/
543 struct dm_thin_endio_hook
{
545 struct dm_deferred_entry
*shared_read_entry
;
546 struct dm_deferred_entry
*all_io_entry
;
547 struct dm_thin_new_mapping
*overwrite_mapping
;
548 struct rb_node rb_node
;
549 struct dm_bio_prison_cell
*cell
;
552 static void __merge_bio_list(struct bio_list
*bios
, struct bio_list
*master
)
554 bio_list_merge(bios
, master
);
555 bio_list_init(master
);
558 static void error_bio_list(struct bio_list
*bios
, int error
)
562 while ((bio
= bio_list_pop(bios
))) {
563 bio
->bi_error
= error
;
568 static void error_thin_bio_list(struct thin_c
*tc
, struct bio_list
*master
, int error
)
570 struct bio_list bios
;
573 bio_list_init(&bios
);
575 spin_lock_irqsave(&tc
->lock
, flags
);
576 __merge_bio_list(&bios
, master
);
577 spin_unlock_irqrestore(&tc
->lock
, flags
);
579 error_bio_list(&bios
, error
);
582 static void requeue_deferred_cells(struct thin_c
*tc
)
584 struct pool
*pool
= tc
->pool
;
586 struct list_head cells
;
587 struct dm_bio_prison_cell
*cell
, *tmp
;
589 INIT_LIST_HEAD(&cells
);
591 spin_lock_irqsave(&tc
->lock
, flags
);
592 list_splice_init(&tc
->deferred_cells
, &cells
);
593 spin_unlock_irqrestore(&tc
->lock
, flags
);
595 list_for_each_entry_safe(cell
, tmp
, &cells
, user_list
)
596 cell_requeue(pool
, cell
);
599 static void requeue_io(struct thin_c
*tc
)
601 struct bio_list bios
;
604 bio_list_init(&bios
);
606 spin_lock_irqsave(&tc
->lock
, flags
);
607 __merge_bio_list(&bios
, &tc
->deferred_bio_list
);
608 __merge_bio_list(&bios
, &tc
->retry_on_resume_list
);
609 spin_unlock_irqrestore(&tc
->lock
, flags
);
611 error_bio_list(&bios
, DM_ENDIO_REQUEUE
);
612 requeue_deferred_cells(tc
);
615 static void error_retry_list_with_code(struct pool
*pool
, int error
)
620 list_for_each_entry_rcu(tc
, &pool
->active_thins
, list
)
621 error_thin_bio_list(tc
, &tc
->retry_on_resume_list
, error
);
625 static void error_retry_list(struct pool
*pool
)
627 int error
= get_pool_io_error_code(pool
);
629 error_retry_list_with_code(pool
, error
);
633 * This section of code contains the logic for processing a thin device's IO.
634 * Much of the code depends on pool object resources (lists, workqueues, etc)
635 * but most is exclusively called from the thin target rather than the thin-pool
639 static dm_block_t
get_bio_block(struct thin_c
*tc
, struct bio
*bio
)
641 struct pool
*pool
= tc
->pool
;
642 sector_t block_nr
= bio
->bi_iter
.bi_sector
;
644 if (block_size_is_power_of_two(pool
))
645 block_nr
>>= pool
->sectors_per_block_shift
;
647 (void) sector_div(block_nr
, pool
->sectors_per_block
);
653 * Returns the _complete_ blocks that this bio covers.
655 static void get_bio_block_range(struct thin_c
*tc
, struct bio
*bio
,
656 dm_block_t
*begin
, dm_block_t
*end
)
658 struct pool
*pool
= tc
->pool
;
659 sector_t b
= bio
->bi_iter
.bi_sector
;
660 sector_t e
= b
+ (bio
->bi_iter
.bi_size
>> SECTOR_SHIFT
);
662 b
+= pool
->sectors_per_block
- 1ull; /* so we round up */
664 if (block_size_is_power_of_two(pool
)) {
665 b
>>= pool
->sectors_per_block_shift
;
666 e
>>= pool
->sectors_per_block_shift
;
668 (void) sector_div(b
, pool
->sectors_per_block
);
669 (void) sector_div(e
, pool
->sectors_per_block
);
673 /* Can happen if the bio is within a single block. */
680 static void remap(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
)
682 struct pool
*pool
= tc
->pool
;
683 sector_t bi_sector
= bio
->bi_iter
.bi_sector
;
685 bio
->bi_bdev
= tc
->pool_dev
->bdev
;
686 if (block_size_is_power_of_two(pool
))
687 bio
->bi_iter
.bi_sector
=
688 (block
<< pool
->sectors_per_block_shift
) |
689 (bi_sector
& (pool
->sectors_per_block
- 1));
691 bio
->bi_iter
.bi_sector
= (block
* pool
->sectors_per_block
) +
692 sector_div(bi_sector
, pool
->sectors_per_block
);
695 static void remap_to_origin(struct thin_c
*tc
, struct bio
*bio
)
697 bio
->bi_bdev
= tc
->origin_dev
->bdev
;
700 static int bio_triggers_commit(struct thin_c
*tc
, struct bio
*bio
)
702 return (bio
->bi_opf
& (REQ_PREFLUSH
| REQ_FUA
)) &&
703 dm_thin_changed_this_transaction(tc
->td
);
706 static void inc_all_io_entry(struct pool
*pool
, struct bio
*bio
)
708 struct dm_thin_endio_hook
*h
;
710 if (bio_op(bio
) == REQ_OP_DISCARD
)
713 h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
714 h
->all_io_entry
= dm_deferred_entry_inc(pool
->all_io_ds
);
717 static void issue(struct thin_c
*tc
, struct bio
*bio
)
719 struct pool
*pool
= tc
->pool
;
722 if (!bio_triggers_commit(tc
, bio
)) {
723 generic_make_request(bio
);
728 * Complete bio with an error if earlier I/O caused changes to
729 * the metadata that can't be committed e.g, due to I/O errors
730 * on the metadata device.
732 if (dm_thin_aborted_changes(tc
->td
)) {
738 * Batch together any bios that trigger commits and then issue a
739 * single commit for them in process_deferred_bios().
741 spin_lock_irqsave(&pool
->lock
, flags
);
742 bio_list_add(&pool
->deferred_flush_bios
, bio
);
743 spin_unlock_irqrestore(&pool
->lock
, flags
);
746 static void remap_to_origin_and_issue(struct thin_c
*tc
, struct bio
*bio
)
748 remap_to_origin(tc
, bio
);
752 static void remap_and_issue(struct thin_c
*tc
, struct bio
*bio
,
755 remap(tc
, bio
, block
);
759 /*----------------------------------------------------------------*/
762 * Bio endio functions.
764 struct dm_thin_new_mapping
{
765 struct list_head list
;
771 * Track quiescing, copying and zeroing preparation actions. When this
772 * counter hits zero the block is prepared and can be inserted into the
775 atomic_t prepare_actions
;
779 dm_block_t virt_begin
, virt_end
;
780 dm_block_t data_block
;
781 struct dm_bio_prison_cell
*cell
;
784 * If the bio covers the whole area of a block then we can avoid
785 * zeroing or copying. Instead this bio is hooked. The bio will
786 * still be in the cell, so care has to be taken to avoid issuing
790 bio_end_io_t
*saved_bi_end_io
;
793 static void __complete_mapping_preparation(struct dm_thin_new_mapping
*m
)
795 struct pool
*pool
= m
->tc
->pool
;
797 if (atomic_dec_and_test(&m
->prepare_actions
)) {
798 list_add_tail(&m
->list
, &pool
->prepared_mappings
);
803 static void complete_mapping_preparation(struct dm_thin_new_mapping
*m
)
806 struct pool
*pool
= m
->tc
->pool
;
808 spin_lock_irqsave(&pool
->lock
, flags
);
809 __complete_mapping_preparation(m
);
810 spin_unlock_irqrestore(&pool
->lock
, flags
);
813 static void copy_complete(int read_err
, unsigned long write_err
, void *context
)
815 struct dm_thin_new_mapping
*m
= context
;
817 m
->err
= read_err
|| write_err
? -EIO
: 0;
818 complete_mapping_preparation(m
);
821 static void overwrite_endio(struct bio
*bio
)
823 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
824 struct dm_thin_new_mapping
*m
= h
->overwrite_mapping
;
826 bio
->bi_end_io
= m
->saved_bi_end_io
;
828 m
->err
= bio
->bi_error
;
829 complete_mapping_preparation(m
);
832 /*----------------------------------------------------------------*/
839 * Prepared mapping jobs.
843 * This sends the bios in the cell, except the original holder, back
844 * to the deferred_bios list.
846 static void cell_defer_no_holder(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
848 struct pool
*pool
= tc
->pool
;
851 spin_lock_irqsave(&tc
->lock
, flags
);
852 cell_release_no_holder(pool
, cell
, &tc
->deferred_bio_list
);
853 spin_unlock_irqrestore(&tc
->lock
, flags
);
858 static void thin_defer_bio(struct thin_c
*tc
, struct bio
*bio
);
862 struct bio_list defer_bios
;
863 struct bio_list issue_bios
;
866 static void __inc_remap_and_issue_cell(void *context
,
867 struct dm_bio_prison_cell
*cell
)
869 struct remap_info
*info
= context
;
872 while ((bio
= bio_list_pop(&cell
->bios
))) {
873 if (bio
->bi_opf
& (REQ_PREFLUSH
| REQ_FUA
) ||
874 bio_op(bio
) == REQ_OP_DISCARD
)
875 bio_list_add(&info
->defer_bios
, bio
);
877 inc_all_io_entry(info
->tc
->pool
, bio
);
880 * We can't issue the bios with the bio prison lock
881 * held, so we add them to a list to issue on
882 * return from this function.
884 bio_list_add(&info
->issue_bios
, bio
);
889 static void inc_remap_and_issue_cell(struct thin_c
*tc
,
890 struct dm_bio_prison_cell
*cell
,
894 struct remap_info info
;
897 bio_list_init(&info
.defer_bios
);
898 bio_list_init(&info
.issue_bios
);
901 * We have to be careful to inc any bios we're about to issue
902 * before the cell is released, and avoid a race with new bios
903 * being added to the cell.
905 cell_visit_release(tc
->pool
, __inc_remap_and_issue_cell
,
908 while ((bio
= bio_list_pop(&info
.defer_bios
)))
909 thin_defer_bio(tc
, bio
);
911 while ((bio
= bio_list_pop(&info
.issue_bios
)))
912 remap_and_issue(info
.tc
, bio
, block
);
915 static void process_prepared_mapping_fail(struct dm_thin_new_mapping
*m
)
917 cell_error(m
->tc
->pool
, m
->cell
);
919 mempool_free(m
, m
->tc
->pool
->mapping_pool
);
922 static void process_prepared_mapping(struct dm_thin_new_mapping
*m
)
924 struct thin_c
*tc
= m
->tc
;
925 struct pool
*pool
= tc
->pool
;
926 struct bio
*bio
= m
->bio
;
930 cell_error(pool
, m
->cell
);
935 * Commit the prepared block into the mapping btree.
936 * Any I/O for this block arriving after this point will get
937 * remapped to it directly.
939 r
= dm_thin_insert_block(tc
->td
, m
->virt_begin
, m
->data_block
);
941 metadata_operation_failed(pool
, "dm_thin_insert_block", r
);
942 cell_error(pool
, m
->cell
);
947 * Release any bios held while the block was being provisioned.
948 * If we are processing a write bio that completely covers the block,
949 * we already processed it so can ignore it now when processing
950 * the bios in the cell.
953 inc_remap_and_issue_cell(tc
, m
->cell
, m
->data_block
);
956 inc_all_io_entry(tc
->pool
, m
->cell
->holder
);
957 remap_and_issue(tc
, m
->cell
->holder
, m
->data_block
);
958 inc_remap_and_issue_cell(tc
, m
->cell
, m
->data_block
);
963 mempool_free(m
, pool
->mapping_pool
);
966 /*----------------------------------------------------------------*/
968 static void free_discard_mapping(struct dm_thin_new_mapping
*m
)
970 struct thin_c
*tc
= m
->tc
;
972 cell_defer_no_holder(tc
, m
->cell
);
973 mempool_free(m
, tc
->pool
->mapping_pool
);
976 static void process_prepared_discard_fail(struct dm_thin_new_mapping
*m
)
978 bio_io_error(m
->bio
);
979 free_discard_mapping(m
);
982 static void process_prepared_discard_success(struct dm_thin_new_mapping
*m
)
985 free_discard_mapping(m
);
988 static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping
*m
)
991 struct thin_c
*tc
= m
->tc
;
993 r
= dm_thin_remove_range(tc
->td
, m
->cell
->key
.block_begin
, m
->cell
->key
.block_end
);
995 metadata_operation_failed(tc
->pool
, "dm_thin_remove_range", r
);
996 bio_io_error(m
->bio
);
1000 cell_defer_no_holder(tc
, m
->cell
);
1001 mempool_free(m
, tc
->pool
->mapping_pool
);
1004 /*----------------------------------------------------------------*/
1006 static void passdown_double_checking_shared_status(struct dm_thin_new_mapping
*m
,
1007 struct bio
*discard_parent
)
1010 * We've already unmapped this range of blocks, but before we
1011 * passdown we have to check that these blocks are now unused.
1015 struct thin_c
*tc
= m
->tc
;
1016 struct pool
*pool
= tc
->pool
;
1017 dm_block_t b
= m
->data_block
, e
, end
= m
->data_block
+ m
->virt_end
- m
->virt_begin
;
1018 struct discard_op op
;
1020 begin_discard(&op
, tc
, discard_parent
);
1022 /* find start of unmapped run */
1023 for (; b
< end
; b
++) {
1024 r
= dm_pool_block_is_used(pool
->pmd
, b
, &used
);
1035 /* find end of run */
1036 for (e
= b
+ 1; e
!= end
; e
++) {
1037 r
= dm_pool_block_is_used(pool
->pmd
, e
, &used
);
1045 r
= issue_discard(&op
, b
, e
);
1052 end_discard(&op
, r
);
1055 static void queue_passdown_pt2(struct dm_thin_new_mapping
*m
)
1057 unsigned long flags
;
1058 struct pool
*pool
= m
->tc
->pool
;
1060 spin_lock_irqsave(&pool
->lock
, flags
);
1061 list_add_tail(&m
->list
, &pool
->prepared_discards_pt2
);
1062 spin_unlock_irqrestore(&pool
->lock
, flags
);
1066 static void passdown_endio(struct bio
*bio
)
1069 * It doesn't matter if the passdown discard failed, we still want
1070 * to unmap (we ignore err).
1072 queue_passdown_pt2(bio
->bi_private
);
1076 static void process_prepared_discard_passdown_pt1(struct dm_thin_new_mapping
*m
)
1079 struct thin_c
*tc
= m
->tc
;
1080 struct pool
*pool
= tc
->pool
;
1081 struct bio
*discard_parent
;
1082 dm_block_t data_end
= m
->data_block
+ (m
->virt_end
- m
->virt_begin
);
1085 * Only this thread allocates blocks, so we can be sure that the
1086 * newly unmapped blocks will not be allocated before the end of
1089 r
= dm_thin_remove_range(tc
->td
, m
->virt_begin
, m
->virt_end
);
1091 metadata_operation_failed(pool
, "dm_thin_remove_range", r
);
1092 bio_io_error(m
->bio
);
1093 cell_defer_no_holder(tc
, m
->cell
);
1094 mempool_free(m
, pool
->mapping_pool
);
1098 discard_parent
= bio_alloc(GFP_NOIO
, 1);
1099 if (!discard_parent
) {
1100 DMWARN("%s: unable to allocate top level discard bio for passdown. Skipping passdown.",
1101 dm_device_name(tc
->pool
->pool_md
));
1102 queue_passdown_pt2(m
);
1105 discard_parent
->bi_end_io
= passdown_endio
;
1106 discard_parent
->bi_private
= m
;
1108 if (m
->maybe_shared
)
1109 passdown_double_checking_shared_status(m
, discard_parent
);
1111 struct discard_op op
;
1113 begin_discard(&op
, tc
, discard_parent
);
1114 r
= issue_discard(&op
, m
->data_block
, data_end
);
1115 end_discard(&op
, r
);
1120 * Increment the unmapped blocks. This prevents a race between the
1121 * passdown io and reallocation of freed blocks.
1123 r
= dm_pool_inc_data_range(pool
->pmd
, m
->data_block
, data_end
);
1125 metadata_operation_failed(pool
, "dm_pool_inc_data_range", r
);
1126 bio_io_error(m
->bio
);
1127 cell_defer_no_holder(tc
, m
->cell
);
1128 mempool_free(m
, pool
->mapping_pool
);
1133 static void process_prepared_discard_passdown_pt2(struct dm_thin_new_mapping
*m
)
1136 struct thin_c
*tc
= m
->tc
;
1137 struct pool
*pool
= tc
->pool
;
1140 * The passdown has completed, so now we can decrement all those
1143 r
= dm_pool_dec_data_range(pool
->pmd
, m
->data_block
,
1144 m
->data_block
+ (m
->virt_end
- m
->virt_begin
));
1146 metadata_operation_failed(pool
, "dm_pool_dec_data_range", r
);
1147 bio_io_error(m
->bio
);
1151 cell_defer_no_holder(tc
, m
->cell
);
1152 mempool_free(m
, pool
->mapping_pool
);
1155 static void process_prepared(struct pool
*pool
, struct list_head
*head
,
1156 process_mapping_fn
*fn
)
1158 unsigned long flags
;
1159 struct list_head maps
;
1160 struct dm_thin_new_mapping
*m
, *tmp
;
1162 INIT_LIST_HEAD(&maps
);
1163 spin_lock_irqsave(&pool
->lock
, flags
);
1164 list_splice_init(head
, &maps
);
1165 spin_unlock_irqrestore(&pool
->lock
, flags
);
1167 list_for_each_entry_safe(m
, tmp
, &maps
, list
)
1172 * Deferred bio jobs.
1174 static int io_overlaps_block(struct pool
*pool
, struct bio
*bio
)
1176 return bio
->bi_iter
.bi_size
==
1177 (pool
->sectors_per_block
<< SECTOR_SHIFT
);
1180 static int io_overwrites_block(struct pool
*pool
, struct bio
*bio
)
1182 return (bio_data_dir(bio
) == WRITE
) &&
1183 io_overlaps_block(pool
, bio
);
1186 static void save_and_set_endio(struct bio
*bio
, bio_end_io_t
**save
,
1189 *save
= bio
->bi_end_io
;
1190 bio
->bi_end_io
= fn
;
1193 static int ensure_next_mapping(struct pool
*pool
)
1195 if (pool
->next_mapping
)
1198 pool
->next_mapping
= mempool_alloc(pool
->mapping_pool
, GFP_ATOMIC
);
1200 return pool
->next_mapping
? 0 : -ENOMEM
;
1203 static struct dm_thin_new_mapping
*get_next_mapping(struct pool
*pool
)
1205 struct dm_thin_new_mapping
*m
= pool
->next_mapping
;
1207 BUG_ON(!pool
->next_mapping
);
1209 memset(m
, 0, sizeof(struct dm_thin_new_mapping
));
1210 INIT_LIST_HEAD(&m
->list
);
1213 pool
->next_mapping
= NULL
;
1218 static void ll_zero(struct thin_c
*tc
, struct dm_thin_new_mapping
*m
,
1219 sector_t begin
, sector_t end
)
1222 struct dm_io_region to
;
1224 to
.bdev
= tc
->pool_dev
->bdev
;
1226 to
.count
= end
- begin
;
1228 r
= dm_kcopyd_zero(tc
->pool
->copier
, 1, &to
, 0, copy_complete
, m
);
1230 DMERR_LIMIT("dm_kcopyd_zero() failed");
1231 copy_complete(1, 1, m
);
1235 static void remap_and_issue_overwrite(struct thin_c
*tc
, struct bio
*bio
,
1236 dm_block_t data_begin
,
1237 struct dm_thin_new_mapping
*m
)
1239 struct pool
*pool
= tc
->pool
;
1240 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1242 h
->overwrite_mapping
= m
;
1244 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
1245 inc_all_io_entry(pool
, bio
);
1246 remap_and_issue(tc
, bio
, data_begin
);
1250 * A partial copy also needs to zero the uncopied region.
1252 static void schedule_copy(struct thin_c
*tc
, dm_block_t virt_block
,
1253 struct dm_dev
*origin
, dm_block_t data_origin
,
1254 dm_block_t data_dest
,
1255 struct dm_bio_prison_cell
*cell
, struct bio
*bio
,
1259 struct pool
*pool
= tc
->pool
;
1260 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
1263 m
->virt_begin
= virt_block
;
1264 m
->virt_end
= virt_block
+ 1u;
1265 m
->data_block
= data_dest
;
1269 * quiesce action + copy action + an extra reference held for the
1270 * duration of this function (we may need to inc later for a
1273 atomic_set(&m
->prepare_actions
, 3);
1275 if (!dm_deferred_set_add_work(pool
->shared_read_ds
, &m
->list
))
1276 complete_mapping_preparation(m
); /* already quiesced */
1279 * IO to pool_dev remaps to the pool target's data_dev.
1281 * If the whole block of data is being overwritten, we can issue the
1282 * bio immediately. Otherwise we use kcopyd to clone the data first.
1284 if (io_overwrites_block(pool
, bio
))
1285 remap_and_issue_overwrite(tc
, bio
, data_dest
, m
);
1287 struct dm_io_region from
, to
;
1289 from
.bdev
= origin
->bdev
;
1290 from
.sector
= data_origin
* pool
->sectors_per_block
;
1293 to
.bdev
= tc
->pool_dev
->bdev
;
1294 to
.sector
= data_dest
* pool
->sectors_per_block
;
1297 r
= dm_kcopyd_copy(pool
->copier
, &from
, 1, &to
,
1298 0, copy_complete
, m
);
1300 DMERR_LIMIT("dm_kcopyd_copy() failed");
1301 copy_complete(1, 1, m
);
1304 * We allow the zero to be issued, to simplify the
1305 * error path. Otherwise we'd need to start
1306 * worrying about decrementing the prepare_actions
1312 * Do we need to zero a tail region?
1314 if (len
< pool
->sectors_per_block
&& pool
->pf
.zero_new_blocks
) {
1315 atomic_inc(&m
->prepare_actions
);
1317 data_dest
* pool
->sectors_per_block
+ len
,
1318 (data_dest
+ 1) * pool
->sectors_per_block
);
1322 complete_mapping_preparation(m
); /* drop our ref */
1325 static void schedule_internal_copy(struct thin_c
*tc
, dm_block_t virt_block
,
1326 dm_block_t data_origin
, dm_block_t data_dest
,
1327 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
1329 schedule_copy(tc
, virt_block
, tc
->pool_dev
,
1330 data_origin
, data_dest
, cell
, bio
,
1331 tc
->pool
->sectors_per_block
);
1334 static void schedule_zero(struct thin_c
*tc
, dm_block_t virt_block
,
1335 dm_block_t data_block
, struct dm_bio_prison_cell
*cell
,
1338 struct pool
*pool
= tc
->pool
;
1339 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
1341 atomic_set(&m
->prepare_actions
, 1); /* no need to quiesce */
1343 m
->virt_begin
= virt_block
;
1344 m
->virt_end
= virt_block
+ 1u;
1345 m
->data_block
= data_block
;
1349 * If the whole block of data is being overwritten or we are not
1350 * zeroing pre-existing data, we can issue the bio immediately.
1351 * Otherwise we use kcopyd to zero the data first.
1353 if (pool
->pf
.zero_new_blocks
) {
1354 if (io_overwrites_block(pool
, bio
))
1355 remap_and_issue_overwrite(tc
, bio
, data_block
, m
);
1357 ll_zero(tc
, m
, data_block
* pool
->sectors_per_block
,
1358 (data_block
+ 1) * pool
->sectors_per_block
);
1360 process_prepared_mapping(m
);
1363 static void schedule_external_copy(struct thin_c
*tc
, dm_block_t virt_block
,
1364 dm_block_t data_dest
,
1365 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
1367 struct pool
*pool
= tc
->pool
;
1368 sector_t virt_block_begin
= virt_block
* pool
->sectors_per_block
;
1369 sector_t virt_block_end
= (virt_block
+ 1) * pool
->sectors_per_block
;
1371 if (virt_block_end
<= tc
->origin_size
)
1372 schedule_copy(tc
, virt_block
, tc
->origin_dev
,
1373 virt_block
, data_dest
, cell
, bio
,
1374 pool
->sectors_per_block
);
1376 else if (virt_block_begin
< tc
->origin_size
)
1377 schedule_copy(tc
, virt_block
, tc
->origin_dev
,
1378 virt_block
, data_dest
, cell
, bio
,
1379 tc
->origin_size
- virt_block_begin
);
1382 schedule_zero(tc
, virt_block
, data_dest
, cell
, bio
);
1385 static void set_pool_mode(struct pool
*pool
, enum pool_mode new_mode
);
1387 static void check_for_space(struct pool
*pool
)
1392 if (get_pool_mode(pool
) != PM_OUT_OF_DATA_SPACE
)
1395 r
= dm_pool_get_free_block_count(pool
->pmd
, &nr_free
);
1400 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
);
1477 metadata_operation_failed(pool
, "dm_pool_alloc_data_block", r
);
1485 * If we have run out of space, queue bios until the device is
1486 * resumed, presumably after having been reloaded with more space.
1488 static void retry_on_resume(struct bio
*bio
)
1490 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1491 struct thin_c
*tc
= h
->tc
;
1492 unsigned long flags
;
1494 spin_lock_irqsave(&tc
->lock
, flags
);
1495 bio_list_add(&tc
->retry_on_resume_list
, bio
);
1496 spin_unlock_irqrestore(&tc
->lock
, flags
);
1499 static int should_error_unserviceable_bio(struct pool
*pool
)
1501 enum pool_mode m
= get_pool_mode(pool
);
1505 /* Shouldn't get here */
1506 DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
1509 case PM_OUT_OF_DATA_SPACE
:
1510 return pool
->pf
.error_if_no_space
? -ENOSPC
: 0;
1516 /* Shouldn't get here */
1517 DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1522 static void handle_unserviceable_bio(struct pool
*pool
, struct bio
*bio
)
1524 int error
= should_error_unserviceable_bio(pool
);
1527 bio
->bi_error
= error
;
1530 retry_on_resume(bio
);
1533 static void retry_bios_on_resume(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
1536 struct bio_list bios
;
1539 error
= should_error_unserviceable_bio(pool
);
1541 cell_error_with_code(pool
, cell
, error
);
1545 bio_list_init(&bios
);
1546 cell_release(pool
, cell
, &bios
);
1548 while ((bio
= bio_list_pop(&bios
)))
1549 retry_on_resume(bio
);
1552 static void process_discard_cell_no_passdown(struct thin_c
*tc
,
1553 struct dm_bio_prison_cell
*virt_cell
)
1555 struct pool
*pool
= tc
->pool
;
1556 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
1559 * We don't need to lock the data blocks, since there's no
1560 * passdown. We only lock data blocks for allocation and breaking sharing.
1563 m
->virt_begin
= virt_cell
->key
.block_begin
;
1564 m
->virt_end
= virt_cell
->key
.block_end
;
1565 m
->cell
= virt_cell
;
1566 m
->bio
= virt_cell
->holder
;
1568 if (!dm_deferred_set_add_work(pool
->all_io_ds
, &m
->list
))
1569 pool
->process_prepared_discard(m
);
1572 static void break_up_discard_bio(struct thin_c
*tc
, dm_block_t begin
, dm_block_t end
,
1575 struct pool
*pool
= tc
->pool
;
1579 struct dm_cell_key data_key
;
1580 struct dm_bio_prison_cell
*data_cell
;
1581 struct dm_thin_new_mapping
*m
;
1582 dm_block_t virt_begin
, virt_end
, data_begin
;
1584 while (begin
!= end
) {
1585 r
= ensure_next_mapping(pool
);
1587 /* we did our best */
1590 r
= dm_thin_find_mapped_range(tc
->td
, begin
, end
, &virt_begin
, &virt_end
,
1591 &data_begin
, &maybe_shared
);
1594 * Silently fail, letting any mappings we've
1599 build_key(tc
->td
, PHYSICAL
, data_begin
, data_begin
+ (virt_end
- virt_begin
), &data_key
);
1600 if (bio_detain(tc
->pool
, &data_key
, NULL
, &data_cell
)) {
1601 /* contention, we'll give up with this range */
1607 * IO may still be going to the destination block. We must
1608 * quiesce before we can do the removal.
1610 m
= get_next_mapping(pool
);
1612 m
->maybe_shared
= maybe_shared
;
1613 m
->virt_begin
= virt_begin
;
1614 m
->virt_end
= virt_end
;
1615 m
->data_block
= data_begin
;
1616 m
->cell
= data_cell
;
1620 * The parent bio must not complete before sub discard bios are
1621 * chained to it (see end_discard's bio_chain)!
1623 * This per-mapping bi_remaining increment is paired with
1624 * the implicit decrement that occurs via bio_endio() in
1627 bio_inc_remaining(bio
);
1628 if (!dm_deferred_set_add_work(pool
->all_io_ds
, &m
->list
))
1629 pool
->process_prepared_discard(m
);
1635 static void process_discard_cell_passdown(struct thin_c
*tc
, struct dm_bio_prison_cell
*virt_cell
)
1637 struct bio
*bio
= virt_cell
->holder
;
1638 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1641 * The virt_cell will only get freed once the origin bio completes.
1642 * This means it will remain locked while all the individual
1643 * passdown bios are in flight.
1645 h
->cell
= virt_cell
;
1646 break_up_discard_bio(tc
, virt_cell
->key
.block_begin
, virt_cell
->key
.block_end
, bio
);
1649 * We complete the bio now, knowing that the bi_remaining field
1650 * will prevent completion until the sub range discards have
1656 static void process_discard_bio(struct thin_c
*tc
, struct bio
*bio
)
1658 dm_block_t begin
, end
;
1659 struct dm_cell_key virt_key
;
1660 struct dm_bio_prison_cell
*virt_cell
;
1662 get_bio_block_range(tc
, bio
, &begin
, &end
);
1665 * The discard covers less than a block.
1671 build_key(tc
->td
, VIRTUAL
, begin
, end
, &virt_key
);
1672 if (bio_detain(tc
->pool
, &virt_key
, bio
, &virt_cell
))
1674 * Potential starvation issue: We're relying on the
1675 * fs/application being well behaved, and not trying to
1676 * send IO to a region at the same time as discarding it.
1677 * If they do this persistently then it's possible this
1678 * cell will never be granted.
1682 tc
->pool
->process_discard_cell(tc
, virt_cell
);
1685 static void break_sharing(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1686 struct dm_cell_key
*key
,
1687 struct dm_thin_lookup_result
*lookup_result
,
1688 struct dm_bio_prison_cell
*cell
)
1691 dm_block_t data_block
;
1692 struct pool
*pool
= tc
->pool
;
1694 r
= alloc_data_block(tc
, &data_block
);
1697 schedule_internal_copy(tc
, block
, lookup_result
->block
,
1698 data_block
, cell
, bio
);
1702 retry_bios_on_resume(pool
, cell
);
1706 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1708 cell_error(pool
, cell
);
1713 static void __remap_and_issue_shared_cell(void *context
,
1714 struct dm_bio_prison_cell
*cell
)
1716 struct remap_info
*info
= context
;
1719 while ((bio
= bio_list_pop(&cell
->bios
))) {
1720 if ((bio_data_dir(bio
) == WRITE
) ||
1721 (bio
->bi_opf
& (REQ_PREFLUSH
| REQ_FUA
) ||
1722 bio_op(bio
) == REQ_OP_DISCARD
))
1723 bio_list_add(&info
->defer_bios
, bio
);
1725 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));;
1727 h
->shared_read_entry
= dm_deferred_entry_inc(info
->tc
->pool
->shared_read_ds
);
1728 inc_all_io_entry(info
->tc
->pool
, bio
);
1729 bio_list_add(&info
->issue_bios
, bio
);
1734 static void remap_and_issue_shared_cell(struct thin_c
*tc
,
1735 struct dm_bio_prison_cell
*cell
,
1739 struct remap_info info
;
1742 bio_list_init(&info
.defer_bios
);
1743 bio_list_init(&info
.issue_bios
);
1745 cell_visit_release(tc
->pool
, __remap_and_issue_shared_cell
,
1748 while ((bio
= bio_list_pop(&info
.defer_bios
)))
1749 thin_defer_bio(tc
, bio
);
1751 while ((bio
= bio_list_pop(&info
.issue_bios
)))
1752 remap_and_issue(tc
, bio
, block
);
1755 static void process_shared_bio(struct thin_c
*tc
, struct bio
*bio
,
1757 struct dm_thin_lookup_result
*lookup_result
,
1758 struct dm_bio_prison_cell
*virt_cell
)
1760 struct dm_bio_prison_cell
*data_cell
;
1761 struct pool
*pool
= tc
->pool
;
1762 struct dm_cell_key key
;
1765 * If cell is already occupied, then sharing is already in the process
1766 * of being broken so we have nothing further to do here.
1768 build_data_key(tc
->td
, lookup_result
->block
, &key
);
1769 if (bio_detain(pool
, &key
, bio
, &data_cell
)) {
1770 cell_defer_no_holder(tc
, virt_cell
);
1774 if (bio_data_dir(bio
) == WRITE
&& bio
->bi_iter
.bi_size
) {
1775 break_sharing(tc
, bio
, block
, &key
, lookup_result
, data_cell
);
1776 cell_defer_no_holder(tc
, virt_cell
);
1778 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1780 h
->shared_read_entry
= dm_deferred_entry_inc(pool
->shared_read_ds
);
1781 inc_all_io_entry(pool
, bio
);
1782 remap_and_issue(tc
, bio
, lookup_result
->block
);
1784 remap_and_issue_shared_cell(tc
, data_cell
, lookup_result
->block
);
1785 remap_and_issue_shared_cell(tc
, virt_cell
, lookup_result
->block
);
1789 static void provision_block(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1790 struct dm_bio_prison_cell
*cell
)
1793 dm_block_t data_block
;
1794 struct pool
*pool
= tc
->pool
;
1797 * Remap empty bios (flushes) immediately, without provisioning.
1799 if (!bio
->bi_iter
.bi_size
) {
1800 inc_all_io_entry(pool
, bio
);
1801 cell_defer_no_holder(tc
, cell
);
1803 remap_and_issue(tc
, bio
, 0);
1808 * Fill read bios with zeroes and complete them immediately.
1810 if (bio_data_dir(bio
) == READ
) {
1812 cell_defer_no_holder(tc
, cell
);
1817 r
= alloc_data_block(tc
, &data_block
);
1821 schedule_external_copy(tc
, block
, data_block
, cell
, bio
);
1823 schedule_zero(tc
, block
, data_block
, cell
, bio
);
1827 retry_bios_on_resume(pool
, cell
);
1831 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1833 cell_error(pool
, cell
);
1838 static void process_cell(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
1841 struct pool
*pool
= tc
->pool
;
1842 struct bio
*bio
= cell
->holder
;
1843 dm_block_t block
= get_bio_block(tc
, bio
);
1844 struct dm_thin_lookup_result lookup_result
;
1846 if (tc
->requeue_mode
) {
1847 cell_requeue(pool
, cell
);
1851 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1854 if (lookup_result
.shared
)
1855 process_shared_bio(tc
, bio
, block
, &lookup_result
, cell
);
1857 inc_all_io_entry(pool
, bio
);
1858 remap_and_issue(tc
, bio
, lookup_result
.block
);
1859 inc_remap_and_issue_cell(tc
, cell
, lookup_result
.block
);
1864 if (bio_data_dir(bio
) == READ
&& tc
->origin_dev
) {
1865 inc_all_io_entry(pool
, bio
);
1866 cell_defer_no_holder(tc
, cell
);
1868 if (bio_end_sector(bio
) <= tc
->origin_size
)
1869 remap_to_origin_and_issue(tc
, bio
);
1871 else if (bio
->bi_iter
.bi_sector
< tc
->origin_size
) {
1873 bio
->bi_iter
.bi_size
= (tc
->origin_size
- bio
->bi_iter
.bi_sector
) << SECTOR_SHIFT
;
1874 remap_to_origin_and_issue(tc
, bio
);
1881 provision_block(tc
, bio
, block
, cell
);
1885 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1887 cell_defer_no_holder(tc
, cell
);
1893 static void process_bio(struct thin_c
*tc
, struct bio
*bio
)
1895 struct pool
*pool
= tc
->pool
;
1896 dm_block_t block
= get_bio_block(tc
, bio
);
1897 struct dm_bio_prison_cell
*cell
;
1898 struct dm_cell_key key
;
1901 * If cell is already occupied, then the block is already
1902 * being provisioned so we have nothing further to do here.
1904 build_virtual_key(tc
->td
, block
, &key
);
1905 if (bio_detain(pool
, &key
, bio
, &cell
))
1908 process_cell(tc
, cell
);
1911 static void __process_bio_read_only(struct thin_c
*tc
, struct bio
*bio
,
1912 struct dm_bio_prison_cell
*cell
)
1915 int rw
= bio_data_dir(bio
);
1916 dm_block_t block
= get_bio_block(tc
, bio
);
1917 struct dm_thin_lookup_result lookup_result
;
1919 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1922 if (lookup_result
.shared
&& (rw
== WRITE
) && bio
->bi_iter
.bi_size
) {
1923 handle_unserviceable_bio(tc
->pool
, bio
);
1925 cell_defer_no_holder(tc
, cell
);
1927 inc_all_io_entry(tc
->pool
, bio
);
1928 remap_and_issue(tc
, bio
, lookup_result
.block
);
1930 inc_remap_and_issue_cell(tc
, cell
, lookup_result
.block
);
1936 cell_defer_no_holder(tc
, cell
);
1938 handle_unserviceable_bio(tc
->pool
, bio
);
1942 if (tc
->origin_dev
) {
1943 inc_all_io_entry(tc
->pool
, bio
);
1944 remap_to_origin_and_issue(tc
, bio
);
1953 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1956 cell_defer_no_holder(tc
, cell
);
1962 static void process_bio_read_only(struct thin_c
*tc
, struct bio
*bio
)
1964 __process_bio_read_only(tc
, bio
, NULL
);
1967 static void process_cell_read_only(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
1969 __process_bio_read_only(tc
, cell
->holder
, cell
);
1972 static void process_bio_success(struct thin_c
*tc
, struct bio
*bio
)
1977 static void process_bio_fail(struct thin_c
*tc
, struct bio
*bio
)
1982 static void process_cell_success(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
1984 cell_success(tc
->pool
, cell
);
1987 static void process_cell_fail(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
1989 cell_error(tc
->pool
, cell
);
1993 * FIXME: should we also commit due to size of transaction, measured in
1996 static int need_commit_due_to_time(struct pool
*pool
)
1998 return !time_in_range(jiffies
, pool
->last_commit_jiffies
,
1999 pool
->last_commit_jiffies
+ COMMIT_PERIOD
);
2002 #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node)
2003 #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook))
2005 static void __thin_bio_rb_add(struct thin_c
*tc
, struct bio
*bio
)
2007 struct rb_node
**rbp
, *parent
;
2008 struct dm_thin_endio_hook
*pbd
;
2009 sector_t bi_sector
= bio
->bi_iter
.bi_sector
;
2011 rbp
= &tc
->sort_bio_list
.rb_node
;
2015 pbd
= thin_pbd(parent
);
2017 if (bi_sector
< thin_bio(pbd
)->bi_iter
.bi_sector
)
2018 rbp
= &(*rbp
)->rb_left
;
2020 rbp
= &(*rbp
)->rb_right
;
2023 pbd
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
2024 rb_link_node(&pbd
->rb_node
, parent
, rbp
);
2025 rb_insert_color(&pbd
->rb_node
, &tc
->sort_bio_list
);
2028 static void __extract_sorted_bios(struct thin_c
*tc
)
2030 struct rb_node
*node
;
2031 struct dm_thin_endio_hook
*pbd
;
2034 for (node
= rb_first(&tc
->sort_bio_list
); node
; node
= rb_next(node
)) {
2035 pbd
= thin_pbd(node
);
2036 bio
= thin_bio(pbd
);
2038 bio_list_add(&tc
->deferred_bio_list
, bio
);
2039 rb_erase(&pbd
->rb_node
, &tc
->sort_bio_list
);
2042 WARN_ON(!RB_EMPTY_ROOT(&tc
->sort_bio_list
));
2045 static void __sort_thin_deferred_bios(struct thin_c
*tc
)
2048 struct bio_list bios
;
2050 bio_list_init(&bios
);
2051 bio_list_merge(&bios
, &tc
->deferred_bio_list
);
2052 bio_list_init(&tc
->deferred_bio_list
);
2054 /* Sort deferred_bio_list using rb-tree */
2055 while ((bio
= bio_list_pop(&bios
)))
2056 __thin_bio_rb_add(tc
, bio
);
2059 * Transfer the sorted bios in sort_bio_list back to
2060 * deferred_bio_list to allow lockless submission of
2063 __extract_sorted_bios(tc
);
2066 static void process_thin_deferred_bios(struct thin_c
*tc
)
2068 struct pool
*pool
= tc
->pool
;
2069 unsigned long flags
;
2071 struct bio_list bios
;
2072 struct blk_plug plug
;
2075 if (tc
->requeue_mode
) {
2076 error_thin_bio_list(tc
, &tc
->deferred_bio_list
, DM_ENDIO_REQUEUE
);
2080 bio_list_init(&bios
);
2082 spin_lock_irqsave(&tc
->lock
, flags
);
2084 if (bio_list_empty(&tc
->deferred_bio_list
)) {
2085 spin_unlock_irqrestore(&tc
->lock
, flags
);
2089 __sort_thin_deferred_bios(tc
);
2091 bio_list_merge(&bios
, &tc
->deferred_bio_list
);
2092 bio_list_init(&tc
->deferred_bio_list
);
2094 spin_unlock_irqrestore(&tc
->lock
, flags
);
2096 blk_start_plug(&plug
);
2097 while ((bio
= bio_list_pop(&bios
))) {
2099 * If we've got no free new_mapping structs, and processing
2100 * this bio might require one, we pause until there are some
2101 * prepared mappings to process.
2103 if (ensure_next_mapping(pool
)) {
2104 spin_lock_irqsave(&tc
->lock
, flags
);
2105 bio_list_add(&tc
->deferred_bio_list
, bio
);
2106 bio_list_merge(&tc
->deferred_bio_list
, &bios
);
2107 spin_unlock_irqrestore(&tc
->lock
, flags
);
2111 if (bio_op(bio
) == REQ_OP_DISCARD
)
2112 pool
->process_discard(tc
, bio
);
2114 pool
->process_bio(tc
, bio
);
2116 if ((count
++ & 127) == 0) {
2117 throttle_work_update(&pool
->throttle
);
2118 dm_pool_issue_prefetches(pool
->pmd
);
2121 blk_finish_plug(&plug
);
2124 static int cmp_cells(const void *lhs
, const void *rhs
)
2126 struct dm_bio_prison_cell
*lhs_cell
= *((struct dm_bio_prison_cell
**) lhs
);
2127 struct dm_bio_prison_cell
*rhs_cell
= *((struct dm_bio_prison_cell
**) rhs
);
2129 BUG_ON(!lhs_cell
->holder
);
2130 BUG_ON(!rhs_cell
->holder
);
2132 if (lhs_cell
->holder
->bi_iter
.bi_sector
< rhs_cell
->holder
->bi_iter
.bi_sector
)
2135 if (lhs_cell
->holder
->bi_iter
.bi_sector
> rhs_cell
->holder
->bi_iter
.bi_sector
)
2141 static unsigned sort_cells(struct pool
*pool
, struct list_head
*cells
)
2144 struct dm_bio_prison_cell
*cell
, *tmp
;
2146 list_for_each_entry_safe(cell
, tmp
, cells
, user_list
) {
2147 if (count
>= CELL_SORT_ARRAY_SIZE
)
2150 pool
->cell_sort_array
[count
++] = cell
;
2151 list_del(&cell
->user_list
);
2154 sort(pool
->cell_sort_array
, count
, sizeof(cell
), cmp_cells
, NULL
);
2159 static void process_thin_deferred_cells(struct thin_c
*tc
)
2161 struct pool
*pool
= tc
->pool
;
2162 unsigned long flags
;
2163 struct list_head cells
;
2164 struct dm_bio_prison_cell
*cell
;
2165 unsigned i
, j
, count
;
2167 INIT_LIST_HEAD(&cells
);
2169 spin_lock_irqsave(&tc
->lock
, flags
);
2170 list_splice_init(&tc
->deferred_cells
, &cells
);
2171 spin_unlock_irqrestore(&tc
->lock
, flags
);
2173 if (list_empty(&cells
))
2177 count
= sort_cells(tc
->pool
, &cells
);
2179 for (i
= 0; i
< count
; i
++) {
2180 cell
= pool
->cell_sort_array
[i
];
2181 BUG_ON(!cell
->holder
);
2184 * If we've got no free new_mapping structs, and processing
2185 * this bio might require one, we pause until there are some
2186 * prepared mappings to process.
2188 if (ensure_next_mapping(pool
)) {
2189 for (j
= i
; j
< count
; j
++)
2190 list_add(&pool
->cell_sort_array
[j
]->user_list
, &cells
);
2192 spin_lock_irqsave(&tc
->lock
, flags
);
2193 list_splice(&cells
, &tc
->deferred_cells
);
2194 spin_unlock_irqrestore(&tc
->lock
, flags
);
2198 if (bio_op(cell
->holder
) == REQ_OP_DISCARD
)
2199 pool
->process_discard_cell(tc
, cell
);
2201 pool
->process_cell(tc
, cell
);
2203 } while (!list_empty(&cells
));
2206 static void thin_get(struct thin_c
*tc
);
2207 static void thin_put(struct thin_c
*tc
);
2210 * We can't hold rcu_read_lock() around code that can block. So we
2211 * find a thin with the rcu lock held; bump a refcount; then drop
2214 static struct thin_c
*get_first_thin(struct pool
*pool
)
2216 struct thin_c
*tc
= NULL
;
2219 if (!list_empty(&pool
->active_thins
)) {
2220 tc
= list_entry_rcu(pool
->active_thins
.next
, struct thin_c
, list
);
2228 static struct thin_c
*get_next_thin(struct pool
*pool
, struct thin_c
*tc
)
2230 struct thin_c
*old_tc
= tc
;
2233 list_for_each_entry_continue_rcu(tc
, &pool
->active_thins
, list
) {
2245 static void process_deferred_bios(struct pool
*pool
)
2247 unsigned long flags
;
2249 struct bio_list bios
;
2252 tc
= get_first_thin(pool
);
2254 process_thin_deferred_cells(tc
);
2255 process_thin_deferred_bios(tc
);
2256 tc
= get_next_thin(pool
, tc
);
2260 * If there are any deferred flush bios, we must commit
2261 * the metadata before issuing them.
2263 bio_list_init(&bios
);
2264 spin_lock_irqsave(&pool
->lock
, flags
);
2265 bio_list_merge(&bios
, &pool
->deferred_flush_bios
);
2266 bio_list_init(&pool
->deferred_flush_bios
);
2267 spin_unlock_irqrestore(&pool
->lock
, flags
);
2269 if (bio_list_empty(&bios
) &&
2270 !(dm_pool_changed_this_transaction(pool
->pmd
) && need_commit_due_to_time(pool
)))
2274 while ((bio
= bio_list_pop(&bios
)))
2278 pool
->last_commit_jiffies
= jiffies
;
2280 while ((bio
= bio_list_pop(&bios
)))
2281 generic_make_request(bio
);
2284 static void do_worker(struct work_struct
*ws
)
2286 struct pool
*pool
= container_of(ws
, struct pool
, worker
);
2288 throttle_work_start(&pool
->throttle
);
2289 dm_pool_issue_prefetches(pool
->pmd
);
2290 throttle_work_update(&pool
->throttle
);
2291 process_prepared(pool
, &pool
->prepared_mappings
, &pool
->process_prepared_mapping
);
2292 throttle_work_update(&pool
->throttle
);
2293 process_prepared(pool
, &pool
->prepared_discards
, &pool
->process_prepared_discard
);
2294 throttle_work_update(&pool
->throttle
);
2295 process_prepared(pool
, &pool
->prepared_discards_pt2
, &pool
->process_prepared_discard_pt2
);
2296 throttle_work_update(&pool
->throttle
);
2297 process_deferred_bios(pool
);
2298 throttle_work_complete(&pool
->throttle
);
2302 * We want to commit periodically so that not too much
2303 * unwritten data builds up.
2305 static void do_waker(struct work_struct
*ws
)
2307 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
, waker
);
2309 queue_delayed_work(pool
->wq
, &pool
->waker
, COMMIT_PERIOD
);
2312 static void notify_of_pool_mode_change_to_oods(struct pool
*pool
);
2315 * We're holding onto IO to allow userland time to react. After the
2316 * timeout either the pool will have been resized (and thus back in
2317 * PM_WRITE mode), or we degrade to PM_OUT_OF_DATA_SPACE w/ error_if_no_space.
2319 static void do_no_space_timeout(struct work_struct
*ws
)
2321 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
,
2324 if (get_pool_mode(pool
) == PM_OUT_OF_DATA_SPACE
&& !pool
->pf
.error_if_no_space
) {
2325 pool
->pf
.error_if_no_space
= true;
2326 notify_of_pool_mode_change_to_oods(pool
);
2327 error_retry_list_with_code(pool
, -ENOSPC
);
2331 /*----------------------------------------------------------------*/
2334 struct work_struct worker
;
2335 struct completion complete
;
2338 static struct pool_work
*to_pool_work(struct work_struct
*ws
)
2340 return container_of(ws
, struct pool_work
, worker
);
2343 static void pool_work_complete(struct pool_work
*pw
)
2345 complete(&pw
->complete
);
2348 static void pool_work_wait(struct pool_work
*pw
, struct pool
*pool
,
2349 void (*fn
)(struct work_struct
*))
2351 INIT_WORK_ONSTACK(&pw
->worker
, fn
);
2352 init_completion(&pw
->complete
);
2353 queue_work(pool
->wq
, &pw
->worker
);
2354 wait_for_completion(&pw
->complete
);
2357 /*----------------------------------------------------------------*/
2359 struct noflush_work
{
2360 struct pool_work pw
;
2364 static struct noflush_work
*to_noflush(struct work_struct
*ws
)
2366 return container_of(to_pool_work(ws
), struct noflush_work
, pw
);
2369 static void do_noflush_start(struct work_struct
*ws
)
2371 struct noflush_work
*w
= to_noflush(ws
);
2372 w
->tc
->requeue_mode
= true;
2374 pool_work_complete(&w
->pw
);
2377 static void do_noflush_stop(struct work_struct
*ws
)
2379 struct noflush_work
*w
= to_noflush(ws
);
2380 w
->tc
->requeue_mode
= false;
2381 pool_work_complete(&w
->pw
);
2384 static void noflush_work(struct thin_c
*tc
, void (*fn
)(struct work_struct
*))
2386 struct noflush_work w
;
2389 pool_work_wait(&w
.pw
, tc
->pool
, fn
);
2392 /*----------------------------------------------------------------*/
2394 static enum pool_mode
get_pool_mode(struct pool
*pool
)
2396 return pool
->pf
.mode
;
2399 static void notify_of_pool_mode_change(struct pool
*pool
, const char *new_mode
)
2401 dm_table_event(pool
->ti
->table
);
2402 DMINFO("%s: switching pool to %s mode",
2403 dm_device_name(pool
->pool_md
), new_mode
);
2406 static void notify_of_pool_mode_change_to_oods(struct pool
*pool
)
2408 if (!pool
->pf
.error_if_no_space
)
2409 notify_of_pool_mode_change(pool
, "out-of-data-space (queue IO)");
2411 notify_of_pool_mode_change(pool
, "out-of-data-space (error IO)");
2414 static bool passdown_enabled(struct pool_c
*pt
)
2416 return pt
->adjusted_pf
.discard_passdown
;
2419 static void set_discard_callbacks(struct pool
*pool
)
2421 struct pool_c
*pt
= pool
->ti
->private;
2423 if (passdown_enabled(pt
)) {
2424 pool
->process_discard_cell
= process_discard_cell_passdown
;
2425 pool
->process_prepared_discard
= process_prepared_discard_passdown_pt1
;
2426 pool
->process_prepared_discard_pt2
= process_prepared_discard_passdown_pt2
;
2428 pool
->process_discard_cell
= process_discard_cell_no_passdown
;
2429 pool
->process_prepared_discard
= process_prepared_discard_no_passdown
;
2433 static void set_pool_mode(struct pool
*pool
, enum pool_mode new_mode
)
2435 struct pool_c
*pt
= pool
->ti
->private;
2436 bool needs_check
= dm_pool_metadata_needs_check(pool
->pmd
);
2437 enum pool_mode old_mode
= get_pool_mode(pool
);
2438 unsigned long no_space_timeout
= ACCESS_ONCE(no_space_timeout_secs
) * HZ
;
2441 * Never allow the pool to transition to PM_WRITE mode if user
2442 * intervention is required to verify metadata and data consistency.
2444 if (new_mode
== PM_WRITE
&& needs_check
) {
2445 DMERR("%s: unable to switch pool to write mode until repaired.",
2446 dm_device_name(pool
->pool_md
));
2447 if (old_mode
!= new_mode
)
2448 new_mode
= old_mode
;
2450 new_mode
= PM_READ_ONLY
;
2453 * If we were in PM_FAIL mode, rollback of metadata failed. We're
2454 * not going to recover without a thin_repair. So we never let the
2455 * pool move out of the old mode.
2457 if (old_mode
== PM_FAIL
)
2458 new_mode
= old_mode
;
2462 if (old_mode
!= new_mode
)
2463 notify_of_pool_mode_change(pool
, "failure");
2464 dm_pool_metadata_read_only(pool
->pmd
);
2465 pool
->process_bio
= process_bio_fail
;
2466 pool
->process_discard
= process_bio_fail
;
2467 pool
->process_cell
= process_cell_fail
;
2468 pool
->process_discard_cell
= process_cell_fail
;
2469 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
2470 pool
->process_prepared_discard
= process_prepared_discard_fail
;
2472 error_retry_list(pool
);
2476 if (old_mode
!= new_mode
)
2477 notify_of_pool_mode_change(pool
, "read-only");
2478 dm_pool_metadata_read_only(pool
->pmd
);
2479 pool
->process_bio
= process_bio_read_only
;
2480 pool
->process_discard
= process_bio_success
;
2481 pool
->process_cell
= process_cell_read_only
;
2482 pool
->process_discard_cell
= process_cell_success
;
2483 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
2484 pool
->process_prepared_discard
= process_prepared_discard_success
;
2486 error_retry_list(pool
);
2489 case PM_OUT_OF_DATA_SPACE
:
2491 * Ideally we'd never hit this state; the low water mark
2492 * would trigger userland to extend the pool before we
2493 * completely run out of data space. However, many small
2494 * IOs to unprovisioned space can consume data space at an
2495 * alarming rate. Adjust your low water mark if you're
2496 * frequently seeing this mode.
2498 if (old_mode
!= new_mode
)
2499 notify_of_pool_mode_change_to_oods(pool
);
2500 pool
->out_of_data_space
= true;
2501 pool
->process_bio
= process_bio_read_only
;
2502 pool
->process_discard
= process_discard_bio
;
2503 pool
->process_cell
= process_cell_read_only
;
2504 pool
->process_prepared_mapping
= process_prepared_mapping
;
2505 set_discard_callbacks(pool
);
2507 if (!pool
->pf
.error_if_no_space
&& no_space_timeout
)
2508 queue_delayed_work(pool
->wq
, &pool
->no_space_timeout
, no_space_timeout
);
2512 if (old_mode
!= new_mode
)
2513 notify_of_pool_mode_change(pool
, "write");
2514 pool
->out_of_data_space
= false;
2515 pool
->pf
.error_if_no_space
= pt
->requested_pf
.error_if_no_space
;
2516 dm_pool_metadata_read_write(pool
->pmd
);
2517 pool
->process_bio
= process_bio
;
2518 pool
->process_discard
= process_discard_bio
;
2519 pool
->process_cell
= process_cell
;
2520 pool
->process_prepared_mapping
= process_prepared_mapping
;
2521 set_discard_callbacks(pool
);
2525 pool
->pf
.mode
= new_mode
;
2527 * The pool mode may have changed, sync it so bind_control_target()
2528 * doesn't cause an unexpected mode transition on resume.
2530 pt
->adjusted_pf
.mode
= new_mode
;
2533 static void abort_transaction(struct pool
*pool
)
2535 const char *dev_name
= dm_device_name(pool
->pool_md
);
2537 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name
);
2538 if (dm_pool_abort_metadata(pool
->pmd
)) {
2539 DMERR("%s: failed to abort metadata transaction", dev_name
);
2540 set_pool_mode(pool
, PM_FAIL
);
2543 if (dm_pool_metadata_set_needs_check(pool
->pmd
)) {
2544 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name
);
2545 set_pool_mode(pool
, PM_FAIL
);
2549 static void metadata_operation_failed(struct pool
*pool
, const char *op
, int r
)
2551 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
2552 dm_device_name(pool
->pool_md
), op
, r
);
2554 abort_transaction(pool
);
2555 set_pool_mode(pool
, PM_READ_ONLY
);
2558 /*----------------------------------------------------------------*/
2561 * Mapping functions.
2565 * Called only while mapping a thin bio to hand it over to the workqueue.
2567 static void thin_defer_bio(struct thin_c
*tc
, struct bio
*bio
)
2569 unsigned long flags
;
2570 struct pool
*pool
= tc
->pool
;
2572 spin_lock_irqsave(&tc
->lock
, flags
);
2573 bio_list_add(&tc
->deferred_bio_list
, bio
);
2574 spin_unlock_irqrestore(&tc
->lock
, flags
);
2579 static void thin_defer_bio_with_throttle(struct thin_c
*tc
, struct bio
*bio
)
2581 struct pool
*pool
= tc
->pool
;
2583 throttle_lock(&pool
->throttle
);
2584 thin_defer_bio(tc
, bio
);
2585 throttle_unlock(&pool
->throttle
);
2588 static void thin_defer_cell(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
2590 unsigned long flags
;
2591 struct pool
*pool
= tc
->pool
;
2593 throttle_lock(&pool
->throttle
);
2594 spin_lock_irqsave(&tc
->lock
, flags
);
2595 list_add_tail(&cell
->user_list
, &tc
->deferred_cells
);
2596 spin_unlock_irqrestore(&tc
->lock
, flags
);
2597 throttle_unlock(&pool
->throttle
);
2602 static void thin_hook_bio(struct thin_c
*tc
, struct bio
*bio
)
2604 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
2607 h
->shared_read_entry
= NULL
;
2608 h
->all_io_entry
= NULL
;
2609 h
->overwrite_mapping
= NULL
;
2614 * Non-blocking function called from the thin target's map function.
2616 static int thin_bio_map(struct dm_target
*ti
, struct bio
*bio
)
2619 struct thin_c
*tc
= ti
->private;
2620 dm_block_t block
= get_bio_block(tc
, bio
);
2621 struct dm_thin_device
*td
= tc
->td
;
2622 struct dm_thin_lookup_result result
;
2623 struct dm_bio_prison_cell
*virt_cell
, *data_cell
;
2624 struct dm_cell_key key
;
2626 thin_hook_bio(tc
, bio
);
2628 if (tc
->requeue_mode
) {
2629 bio
->bi_error
= DM_ENDIO_REQUEUE
;
2631 return DM_MAPIO_SUBMITTED
;
2634 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
2636 return DM_MAPIO_SUBMITTED
;
2639 if (bio
->bi_opf
& (REQ_PREFLUSH
| REQ_FUA
) ||
2640 bio_op(bio
) == REQ_OP_DISCARD
) {
2641 thin_defer_bio_with_throttle(tc
, bio
);
2642 return DM_MAPIO_SUBMITTED
;
2646 * We must hold the virtual cell before doing the lookup, otherwise
2647 * there's a race with discard.
2649 build_virtual_key(tc
->td
, block
, &key
);
2650 if (bio_detain(tc
->pool
, &key
, bio
, &virt_cell
))
2651 return DM_MAPIO_SUBMITTED
;
2653 r
= dm_thin_find_block(td
, block
, 0, &result
);
2656 * Note that we defer readahead too.
2660 if (unlikely(result
.shared
)) {
2662 * We have a race condition here between the
2663 * result.shared value returned by the lookup and
2664 * snapshot creation, which may cause new
2667 * To avoid this always quiesce the origin before
2668 * taking the snap. You want to do this anyway to
2669 * ensure a consistent application view
2672 * More distant ancestors are irrelevant. The
2673 * shared flag will be set in their case.
2675 thin_defer_cell(tc
, virt_cell
);
2676 return DM_MAPIO_SUBMITTED
;
2679 build_data_key(tc
->td
, result
.block
, &key
);
2680 if (bio_detain(tc
->pool
, &key
, bio
, &data_cell
)) {
2681 cell_defer_no_holder(tc
, virt_cell
);
2682 return DM_MAPIO_SUBMITTED
;
2685 inc_all_io_entry(tc
->pool
, bio
);
2686 cell_defer_no_holder(tc
, data_cell
);
2687 cell_defer_no_holder(tc
, virt_cell
);
2689 remap(tc
, bio
, result
.block
);
2690 return DM_MAPIO_REMAPPED
;
2694 thin_defer_cell(tc
, virt_cell
);
2695 return DM_MAPIO_SUBMITTED
;
2699 * Must always call bio_io_error on failure.
2700 * dm_thin_find_block can fail with -EINVAL if the
2701 * pool is switched to fail-io mode.
2704 cell_defer_no_holder(tc
, virt_cell
);
2705 return DM_MAPIO_SUBMITTED
;
2709 static int pool_is_congested(struct dm_target_callbacks
*cb
, int bdi_bits
)
2711 struct pool_c
*pt
= container_of(cb
, struct pool_c
, callbacks
);
2712 struct request_queue
*q
;
2714 if (get_pool_mode(pt
->pool
) == PM_OUT_OF_DATA_SPACE
)
2717 q
= bdev_get_queue(pt
->data_dev
->bdev
);
2718 return bdi_congested(&q
->backing_dev_info
, bdi_bits
);
2721 static void requeue_bios(struct pool
*pool
)
2723 unsigned long flags
;
2727 list_for_each_entry_rcu(tc
, &pool
->active_thins
, list
) {
2728 spin_lock_irqsave(&tc
->lock
, flags
);
2729 bio_list_merge(&tc
->deferred_bio_list
, &tc
->retry_on_resume_list
);
2730 bio_list_init(&tc
->retry_on_resume_list
);
2731 spin_unlock_irqrestore(&tc
->lock
, flags
);
2736 /*----------------------------------------------------------------
2737 * Binding of control targets to a pool object
2738 *--------------------------------------------------------------*/
2739 static bool data_dev_supports_discard(struct pool_c
*pt
)
2741 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
2743 return q
&& blk_queue_discard(q
);
2746 static bool is_factor(sector_t block_size
, uint32_t n
)
2748 return !sector_div(block_size
, n
);
2752 * If discard_passdown was enabled verify that the data device
2753 * supports discards. Disable discard_passdown if not.
2755 static void disable_passdown_if_not_supported(struct pool_c
*pt
)
2757 struct pool
*pool
= pt
->pool
;
2758 struct block_device
*data_bdev
= pt
->data_dev
->bdev
;
2759 struct queue_limits
*data_limits
= &bdev_get_queue(data_bdev
)->limits
;
2760 const char *reason
= NULL
;
2761 char buf
[BDEVNAME_SIZE
];
2763 if (!pt
->adjusted_pf
.discard_passdown
)
2766 if (!data_dev_supports_discard(pt
))
2767 reason
= "discard unsupported";
2769 else if (data_limits
->max_discard_sectors
< pool
->sectors_per_block
)
2770 reason
= "max discard sectors smaller than a block";
2773 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev
, buf
), reason
);
2774 pt
->adjusted_pf
.discard_passdown
= false;
2778 static int bind_control_target(struct pool
*pool
, struct dm_target
*ti
)
2780 struct pool_c
*pt
= ti
->private;
2783 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
2785 enum pool_mode old_mode
= get_pool_mode(pool
);
2786 enum pool_mode new_mode
= pt
->adjusted_pf
.mode
;
2789 * Don't change the pool's mode until set_pool_mode() below.
2790 * Otherwise the pool's process_* function pointers may
2791 * not match the desired pool mode.
2793 pt
->adjusted_pf
.mode
= old_mode
;
2796 pool
->pf
= pt
->adjusted_pf
;
2797 pool
->low_water_blocks
= pt
->low_water_blocks
;
2799 set_pool_mode(pool
, new_mode
);
2804 static void unbind_control_target(struct pool
*pool
, struct dm_target
*ti
)
2810 /*----------------------------------------------------------------
2812 *--------------------------------------------------------------*/
2813 /* Initialize pool features. */
2814 static void pool_features_init(struct pool_features
*pf
)
2816 pf
->mode
= PM_WRITE
;
2817 pf
->zero_new_blocks
= true;
2818 pf
->discard_enabled
= true;
2819 pf
->discard_passdown
= true;
2820 pf
->error_if_no_space
= false;
2823 static void __pool_destroy(struct pool
*pool
)
2825 __pool_table_remove(pool
);
2827 vfree(pool
->cell_sort_array
);
2828 if (dm_pool_metadata_close(pool
->pmd
) < 0)
2829 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
2831 dm_bio_prison_destroy(pool
->prison
);
2832 dm_kcopyd_client_destroy(pool
->copier
);
2835 destroy_workqueue(pool
->wq
);
2837 if (pool
->next_mapping
)
2838 mempool_free(pool
->next_mapping
, pool
->mapping_pool
);
2839 mempool_destroy(pool
->mapping_pool
);
2840 dm_deferred_set_destroy(pool
->shared_read_ds
);
2841 dm_deferred_set_destroy(pool
->all_io_ds
);
2845 static struct kmem_cache
*_new_mapping_cache
;
2847 static struct pool
*pool_create(struct mapped_device
*pool_md
,
2848 struct block_device
*metadata_dev
,
2849 unsigned long block_size
,
2850 int read_only
, char **error
)
2855 struct dm_pool_metadata
*pmd
;
2856 bool format_device
= read_only
? false : true;
2858 pmd
= dm_pool_metadata_open(metadata_dev
, block_size
, format_device
);
2860 *error
= "Error creating metadata object";
2861 return (struct pool
*)pmd
;
2864 pool
= kmalloc(sizeof(*pool
), GFP_KERNEL
);
2866 *error
= "Error allocating memory for pool";
2867 err_p
= ERR_PTR(-ENOMEM
);
2872 pool
->sectors_per_block
= block_size
;
2873 if (block_size
& (block_size
- 1))
2874 pool
->sectors_per_block_shift
= -1;
2876 pool
->sectors_per_block_shift
= __ffs(block_size
);
2877 pool
->low_water_blocks
= 0;
2878 pool_features_init(&pool
->pf
);
2879 pool
->prison
= dm_bio_prison_create();
2880 if (!pool
->prison
) {
2881 *error
= "Error creating pool's bio prison";
2882 err_p
= ERR_PTR(-ENOMEM
);
2886 pool
->copier
= dm_kcopyd_client_create(&dm_kcopyd_throttle
);
2887 if (IS_ERR(pool
->copier
)) {
2888 r
= PTR_ERR(pool
->copier
);
2889 *error
= "Error creating pool's kcopyd client";
2891 goto bad_kcopyd_client
;
2895 * Create singlethreaded workqueue that will service all devices
2896 * that use this metadata.
2898 pool
->wq
= alloc_ordered_workqueue("dm-" DM_MSG_PREFIX
, WQ_MEM_RECLAIM
);
2900 *error
= "Error creating pool's workqueue";
2901 err_p
= ERR_PTR(-ENOMEM
);
2905 throttle_init(&pool
->throttle
);
2906 INIT_WORK(&pool
->worker
, do_worker
);
2907 INIT_DELAYED_WORK(&pool
->waker
, do_waker
);
2908 INIT_DELAYED_WORK(&pool
->no_space_timeout
, do_no_space_timeout
);
2909 spin_lock_init(&pool
->lock
);
2910 bio_list_init(&pool
->deferred_flush_bios
);
2911 INIT_LIST_HEAD(&pool
->prepared_mappings
);
2912 INIT_LIST_HEAD(&pool
->prepared_discards
);
2913 INIT_LIST_HEAD(&pool
->prepared_discards_pt2
);
2914 INIT_LIST_HEAD(&pool
->active_thins
);
2915 pool
->low_water_triggered
= false;
2916 pool
->suspended
= true;
2917 pool
->out_of_data_space
= false;
2919 pool
->shared_read_ds
= dm_deferred_set_create();
2920 if (!pool
->shared_read_ds
) {
2921 *error
= "Error creating pool's shared read deferred set";
2922 err_p
= ERR_PTR(-ENOMEM
);
2923 goto bad_shared_read_ds
;
2926 pool
->all_io_ds
= dm_deferred_set_create();
2927 if (!pool
->all_io_ds
) {
2928 *error
= "Error creating pool's all io deferred set";
2929 err_p
= ERR_PTR(-ENOMEM
);
2933 pool
->next_mapping
= NULL
;
2934 pool
->mapping_pool
= mempool_create_slab_pool(MAPPING_POOL_SIZE
,
2935 _new_mapping_cache
);
2936 if (!pool
->mapping_pool
) {
2937 *error
= "Error creating pool's mapping mempool";
2938 err_p
= ERR_PTR(-ENOMEM
);
2939 goto bad_mapping_pool
;
2942 pool
->cell_sort_array
= vmalloc(sizeof(*pool
->cell_sort_array
) * CELL_SORT_ARRAY_SIZE
);
2943 if (!pool
->cell_sort_array
) {
2944 *error
= "Error allocating cell sort array";
2945 err_p
= ERR_PTR(-ENOMEM
);
2946 goto bad_sort_array
;
2949 pool
->ref_count
= 1;
2950 pool
->last_commit_jiffies
= jiffies
;
2951 pool
->pool_md
= pool_md
;
2952 pool
->md_dev
= metadata_dev
;
2953 __pool_table_insert(pool
);
2958 mempool_destroy(pool
->mapping_pool
);
2960 dm_deferred_set_destroy(pool
->all_io_ds
);
2962 dm_deferred_set_destroy(pool
->shared_read_ds
);
2964 destroy_workqueue(pool
->wq
);
2966 dm_kcopyd_client_destroy(pool
->copier
);
2968 dm_bio_prison_destroy(pool
->prison
);
2972 if (dm_pool_metadata_close(pmd
))
2973 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
2978 static void __pool_inc(struct pool
*pool
)
2980 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
2984 static void __pool_dec(struct pool
*pool
)
2986 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
2987 BUG_ON(!pool
->ref_count
);
2988 if (!--pool
->ref_count
)
2989 __pool_destroy(pool
);
2992 static struct pool
*__pool_find(struct mapped_device
*pool_md
,
2993 struct block_device
*metadata_dev
,
2994 unsigned long block_size
, int read_only
,
2995 char **error
, int *created
)
2997 struct pool
*pool
= __pool_table_lookup_metadata_dev(metadata_dev
);
3000 if (pool
->pool_md
!= pool_md
) {
3001 *error
= "metadata device already in use by a pool";
3002 return ERR_PTR(-EBUSY
);
3007 pool
= __pool_table_lookup(pool_md
);
3009 if (pool
->md_dev
!= metadata_dev
) {
3010 *error
= "different pool cannot replace a pool";
3011 return ERR_PTR(-EINVAL
);
3016 pool
= pool_create(pool_md
, metadata_dev
, block_size
, read_only
, error
);
3024 /*----------------------------------------------------------------
3025 * Pool target methods
3026 *--------------------------------------------------------------*/
3027 static void pool_dtr(struct dm_target
*ti
)
3029 struct pool_c
*pt
= ti
->private;
3031 mutex_lock(&dm_thin_pool_table
.mutex
);
3033 unbind_control_target(pt
->pool
, ti
);
3034 __pool_dec(pt
->pool
);
3035 dm_put_device(ti
, pt
->metadata_dev
);
3036 dm_put_device(ti
, pt
->data_dev
);
3039 mutex_unlock(&dm_thin_pool_table
.mutex
);
3042 static int parse_pool_features(struct dm_arg_set
*as
, struct pool_features
*pf
,
3043 struct dm_target
*ti
)
3047 const char *arg_name
;
3049 static struct dm_arg _args
[] = {
3050 {0, 4, "Invalid number of pool feature arguments"},
3054 * No feature arguments supplied.
3059 r
= dm_read_arg_group(_args
, as
, &argc
, &ti
->error
);
3063 while (argc
&& !r
) {
3064 arg_name
= dm_shift_arg(as
);
3067 if (!strcasecmp(arg_name
, "skip_block_zeroing"))
3068 pf
->zero_new_blocks
= false;
3070 else if (!strcasecmp(arg_name
, "ignore_discard"))
3071 pf
->discard_enabled
= false;
3073 else if (!strcasecmp(arg_name
, "no_discard_passdown"))
3074 pf
->discard_passdown
= false;
3076 else if (!strcasecmp(arg_name
, "read_only"))
3077 pf
->mode
= PM_READ_ONLY
;
3079 else if (!strcasecmp(arg_name
, "error_if_no_space"))
3080 pf
->error_if_no_space
= true;
3083 ti
->error
= "Unrecognised pool feature requested";
3092 static void metadata_low_callback(void *context
)
3094 struct pool
*pool
= context
;
3096 DMWARN("%s: reached low water mark for metadata device: sending event.",
3097 dm_device_name(pool
->pool_md
));
3099 dm_table_event(pool
->ti
->table
);
3102 static sector_t
get_dev_size(struct block_device
*bdev
)
3104 return i_size_read(bdev
->bd_inode
) >> SECTOR_SHIFT
;
3107 static void warn_if_metadata_device_too_big(struct block_device
*bdev
)
3109 sector_t metadata_dev_size
= get_dev_size(bdev
);
3110 char buffer
[BDEVNAME_SIZE
];
3112 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS_WARNING
)
3113 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
3114 bdevname(bdev
, buffer
), THIN_METADATA_MAX_SECTORS
);
3117 static sector_t
get_metadata_dev_size(struct block_device
*bdev
)
3119 sector_t metadata_dev_size
= get_dev_size(bdev
);
3121 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS
)
3122 metadata_dev_size
= THIN_METADATA_MAX_SECTORS
;
3124 return metadata_dev_size
;
3127 static dm_block_t
get_metadata_dev_size_in_blocks(struct block_device
*bdev
)
3129 sector_t metadata_dev_size
= get_metadata_dev_size(bdev
);
3131 sector_div(metadata_dev_size
, THIN_METADATA_BLOCK_SIZE
);
3133 return metadata_dev_size
;
3137 * When a metadata threshold is crossed a dm event is triggered, and
3138 * userland should respond by growing the metadata device. We could let
3139 * userland set the threshold, like we do with the data threshold, but I'm
3140 * not sure they know enough to do this well.
3142 static dm_block_t
calc_metadata_threshold(struct pool_c
*pt
)
3145 * 4M is ample for all ops with the possible exception of thin
3146 * device deletion which is harmless if it fails (just retry the
3147 * delete after you've grown the device).
3149 dm_block_t quarter
= get_metadata_dev_size_in_blocks(pt
->metadata_dev
->bdev
) / 4;
3150 return min((dm_block_t
)1024ULL /* 4M */, quarter
);
3154 * thin-pool <metadata dev> <data dev>
3155 * <data block size (sectors)>
3156 * <low water mark (blocks)>
3157 * [<#feature args> [<arg>]*]
3159 * Optional feature arguments are:
3160 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
3161 * ignore_discard: disable discard
3162 * no_discard_passdown: don't pass discards down to the data device
3163 * read_only: Don't allow any changes to be made to the pool metadata.
3164 * error_if_no_space: error IOs, instead of queueing, if no space.
3166 static int pool_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
3168 int r
, pool_created
= 0;
3171 struct pool_features pf
;
3172 struct dm_arg_set as
;
3173 struct dm_dev
*data_dev
;
3174 unsigned long block_size
;
3175 dm_block_t low_water_blocks
;
3176 struct dm_dev
*metadata_dev
;
3177 fmode_t metadata_mode
;
3180 * FIXME Remove validation from scope of lock.
3182 mutex_lock(&dm_thin_pool_table
.mutex
);
3185 ti
->error
= "Invalid argument count";
3194 * Set default pool features.
3196 pool_features_init(&pf
);
3198 dm_consume_args(&as
, 4);
3199 r
= parse_pool_features(&as
, &pf
, ti
);
3203 metadata_mode
= FMODE_READ
| ((pf
.mode
== PM_READ_ONLY
) ? 0 : FMODE_WRITE
);
3204 r
= dm_get_device(ti
, argv
[0], metadata_mode
, &metadata_dev
);
3206 ti
->error
= "Error opening metadata block device";
3209 warn_if_metadata_device_too_big(metadata_dev
->bdev
);
3211 r
= dm_get_device(ti
, argv
[1], FMODE_READ
| FMODE_WRITE
, &data_dev
);
3213 ti
->error
= "Error getting data device";
3217 if (kstrtoul(argv
[2], 10, &block_size
) || !block_size
||
3218 block_size
< DATA_DEV_BLOCK_SIZE_MIN_SECTORS
||
3219 block_size
> DATA_DEV_BLOCK_SIZE_MAX_SECTORS
||
3220 block_size
& (DATA_DEV_BLOCK_SIZE_MIN_SECTORS
- 1)) {
3221 ti
->error
= "Invalid block size";
3226 if (kstrtoull(argv
[3], 10, (unsigned long long *)&low_water_blocks
)) {
3227 ti
->error
= "Invalid low water mark";
3232 pt
= kzalloc(sizeof(*pt
), GFP_KERNEL
);
3238 pool
= __pool_find(dm_table_get_md(ti
->table
), metadata_dev
->bdev
,
3239 block_size
, pf
.mode
== PM_READ_ONLY
, &ti
->error
, &pool_created
);
3246 * 'pool_created' reflects whether this is the first table load.
3247 * Top level discard support is not allowed to be changed after
3248 * initial load. This would require a pool reload to trigger thin
3251 if (!pool_created
&& pf
.discard_enabled
!= pool
->pf
.discard_enabled
) {
3252 ti
->error
= "Discard support cannot be disabled once enabled";
3254 goto out_flags_changed
;
3259 pt
->metadata_dev
= metadata_dev
;
3260 pt
->data_dev
= data_dev
;
3261 pt
->low_water_blocks
= low_water_blocks
;
3262 pt
->adjusted_pf
= pt
->requested_pf
= pf
;
3263 ti
->num_flush_bios
= 1;
3266 * Only need to enable discards if the pool should pass
3267 * them down to the data device. The thin device's discard
3268 * processing will cause mappings to be removed from the btree.
3270 ti
->discard_zeroes_data_unsupported
= true;
3271 if (pf
.discard_enabled
&& pf
.discard_passdown
) {
3272 ti
->num_discard_bios
= 1;
3275 * Setting 'discards_supported' circumvents the normal
3276 * stacking of discard limits (this keeps the pool and
3277 * thin devices' discard limits consistent).
3279 ti
->discards_supported
= true;
3283 r
= dm_pool_register_metadata_threshold(pt
->pool
->pmd
,
3284 calc_metadata_threshold(pt
),
3285 metadata_low_callback
,
3288 goto out_flags_changed
;
3290 pt
->callbacks
.congested_fn
= pool_is_congested
;
3291 dm_table_add_target_callbacks(ti
->table
, &pt
->callbacks
);
3293 mutex_unlock(&dm_thin_pool_table
.mutex
);
3302 dm_put_device(ti
, data_dev
);
3304 dm_put_device(ti
, metadata_dev
);
3306 mutex_unlock(&dm_thin_pool_table
.mutex
);
3311 static int pool_map(struct dm_target
*ti
, struct bio
*bio
)
3314 struct pool_c
*pt
= ti
->private;
3315 struct pool
*pool
= pt
->pool
;
3316 unsigned long flags
;
3319 * As this is a singleton target, ti->begin is always zero.
3321 spin_lock_irqsave(&pool
->lock
, flags
);
3322 bio
->bi_bdev
= pt
->data_dev
->bdev
;
3323 r
= DM_MAPIO_REMAPPED
;
3324 spin_unlock_irqrestore(&pool
->lock
, flags
);
3329 static int maybe_resize_data_dev(struct dm_target
*ti
, bool *need_commit
)
3332 struct pool_c
*pt
= ti
->private;
3333 struct pool
*pool
= pt
->pool
;
3334 sector_t data_size
= ti
->len
;
3335 dm_block_t sb_data_size
;
3337 *need_commit
= false;
3339 (void) sector_div(data_size
, pool
->sectors_per_block
);
3341 r
= dm_pool_get_data_dev_size(pool
->pmd
, &sb_data_size
);
3343 DMERR("%s: failed to retrieve data device size",
3344 dm_device_name(pool
->pool_md
));
3348 if (data_size
< sb_data_size
) {
3349 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
3350 dm_device_name(pool
->pool_md
),
3351 (unsigned long long)data_size
, sb_data_size
);
3354 } else if (data_size
> sb_data_size
) {
3355 if (dm_pool_metadata_needs_check(pool
->pmd
)) {
3356 DMERR("%s: unable to grow the data device until repaired.",
3357 dm_device_name(pool
->pool_md
));
3362 DMINFO("%s: growing the data device from %llu to %llu blocks",
3363 dm_device_name(pool
->pool_md
),
3364 sb_data_size
, (unsigned long long)data_size
);
3365 r
= dm_pool_resize_data_dev(pool
->pmd
, data_size
);
3367 metadata_operation_failed(pool
, "dm_pool_resize_data_dev", r
);
3371 *need_commit
= true;
3377 static int maybe_resize_metadata_dev(struct dm_target
*ti
, bool *need_commit
)
3380 struct pool_c
*pt
= ti
->private;
3381 struct pool
*pool
= pt
->pool
;
3382 dm_block_t metadata_dev_size
, sb_metadata_dev_size
;
3384 *need_commit
= false;
3386 metadata_dev_size
= get_metadata_dev_size_in_blocks(pool
->md_dev
);
3388 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &sb_metadata_dev_size
);
3390 DMERR("%s: failed to retrieve metadata device size",
3391 dm_device_name(pool
->pool_md
));
3395 if (metadata_dev_size
< sb_metadata_dev_size
) {
3396 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
3397 dm_device_name(pool
->pool_md
),
3398 metadata_dev_size
, sb_metadata_dev_size
);
3401 } else if (metadata_dev_size
> sb_metadata_dev_size
) {
3402 if (dm_pool_metadata_needs_check(pool
->pmd
)) {
3403 DMERR("%s: unable to grow the metadata device until repaired.",
3404 dm_device_name(pool
->pool_md
));
3408 warn_if_metadata_device_too_big(pool
->md_dev
);
3409 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
3410 dm_device_name(pool
->pool_md
),
3411 sb_metadata_dev_size
, metadata_dev_size
);
3412 r
= dm_pool_resize_metadata_dev(pool
->pmd
, metadata_dev_size
);
3414 metadata_operation_failed(pool
, "dm_pool_resize_metadata_dev", r
);
3418 *need_commit
= true;
3425 * Retrieves the number of blocks of the data device from
3426 * the superblock and compares it to the actual device size,
3427 * thus resizing the data device in case it has grown.
3429 * This both copes with opening preallocated data devices in the ctr
3430 * being followed by a resume
3432 * calling the resume method individually after userspace has
3433 * grown the data device in reaction to a table event.
3435 static int pool_preresume(struct dm_target
*ti
)
3438 bool need_commit1
, need_commit2
;
3439 struct pool_c
*pt
= ti
->private;
3440 struct pool
*pool
= pt
->pool
;
3443 * Take control of the pool object.
3445 r
= bind_control_target(pool
, ti
);
3449 r
= maybe_resize_data_dev(ti
, &need_commit1
);
3453 r
= maybe_resize_metadata_dev(ti
, &need_commit2
);
3457 if (need_commit1
|| need_commit2
)
3458 (void) commit(pool
);
3463 static void pool_suspend_active_thins(struct pool
*pool
)
3467 /* Suspend all active thin devices */
3468 tc
= get_first_thin(pool
);
3470 dm_internal_suspend_noflush(tc
->thin_md
);
3471 tc
= get_next_thin(pool
, tc
);
3475 static void pool_resume_active_thins(struct pool
*pool
)
3479 /* Resume all active thin devices */
3480 tc
= get_first_thin(pool
);
3482 dm_internal_resume(tc
->thin_md
);
3483 tc
= get_next_thin(pool
, tc
);
3487 static void pool_resume(struct dm_target
*ti
)
3489 struct pool_c
*pt
= ti
->private;
3490 struct pool
*pool
= pt
->pool
;
3491 unsigned long flags
;
3494 * Must requeue active_thins' bios and then resume
3495 * active_thins _before_ clearing 'suspend' flag.
3498 pool_resume_active_thins(pool
);
3500 spin_lock_irqsave(&pool
->lock
, flags
);
3501 pool
->low_water_triggered
= false;
3502 pool
->suspended
= false;
3503 spin_unlock_irqrestore(&pool
->lock
, flags
);
3505 do_waker(&pool
->waker
.work
);
3508 static void pool_presuspend(struct dm_target
*ti
)
3510 struct pool_c
*pt
= ti
->private;
3511 struct pool
*pool
= pt
->pool
;
3512 unsigned long flags
;
3514 spin_lock_irqsave(&pool
->lock
, flags
);
3515 pool
->suspended
= true;
3516 spin_unlock_irqrestore(&pool
->lock
, flags
);
3518 pool_suspend_active_thins(pool
);
3521 static void pool_presuspend_undo(struct dm_target
*ti
)
3523 struct pool_c
*pt
= ti
->private;
3524 struct pool
*pool
= pt
->pool
;
3525 unsigned long flags
;
3527 pool_resume_active_thins(pool
);
3529 spin_lock_irqsave(&pool
->lock
, flags
);
3530 pool
->suspended
= false;
3531 spin_unlock_irqrestore(&pool
->lock
, flags
);
3534 static void pool_postsuspend(struct dm_target
*ti
)
3536 struct pool_c
*pt
= ti
->private;
3537 struct pool
*pool
= pt
->pool
;
3539 cancel_delayed_work_sync(&pool
->waker
);
3540 cancel_delayed_work_sync(&pool
->no_space_timeout
);
3541 flush_workqueue(pool
->wq
);
3542 (void) commit(pool
);
3545 static int check_arg_count(unsigned argc
, unsigned args_required
)
3547 if (argc
!= args_required
) {
3548 DMWARN("Message received with %u arguments instead of %u.",
3549 argc
, args_required
);
3556 static int read_dev_id(char *arg
, dm_thin_id
*dev_id
, int warning
)
3558 if (!kstrtoull(arg
, 10, (unsigned long long *)dev_id
) &&
3559 *dev_id
<= MAX_DEV_ID
)
3563 DMWARN("Message received with invalid device id: %s", arg
);
3568 static int process_create_thin_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
3573 r
= check_arg_count(argc
, 2);
3577 r
= read_dev_id(argv
[1], &dev_id
, 1);
3581 r
= dm_pool_create_thin(pool
->pmd
, dev_id
);
3583 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
3591 static int process_create_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
3594 dm_thin_id origin_dev_id
;
3597 r
= check_arg_count(argc
, 3);
3601 r
= read_dev_id(argv
[1], &dev_id
, 1);
3605 r
= read_dev_id(argv
[2], &origin_dev_id
, 1);
3609 r
= dm_pool_create_snap(pool
->pmd
, dev_id
, origin_dev_id
);
3611 DMWARN("Creation of new snapshot %s of device %s failed.",
3619 static int process_delete_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
3624 r
= check_arg_count(argc
, 2);
3628 r
= read_dev_id(argv
[1], &dev_id
, 1);
3632 r
= dm_pool_delete_thin_device(pool
->pmd
, dev_id
);
3634 DMWARN("Deletion of thin device %s failed.", argv
[1]);
3639 static int process_set_transaction_id_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
3641 dm_thin_id old_id
, new_id
;
3644 r
= check_arg_count(argc
, 3);
3648 if (kstrtoull(argv
[1], 10, (unsigned long long *)&old_id
)) {
3649 DMWARN("set_transaction_id message: Unrecognised id %s.", argv
[1]);
3653 if (kstrtoull(argv
[2], 10, (unsigned long long *)&new_id
)) {
3654 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv
[2]);
3658 r
= dm_pool_set_metadata_transaction_id(pool
->pmd
, old_id
, new_id
);
3660 DMWARN("Failed to change transaction id from %s to %s.",
3668 static int process_reserve_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
3672 r
= check_arg_count(argc
, 1);
3676 (void) commit(pool
);
3678 r
= dm_pool_reserve_metadata_snap(pool
->pmd
);
3680 DMWARN("reserve_metadata_snap message failed.");
3685 static int process_release_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
3689 r
= check_arg_count(argc
, 1);
3693 r
= dm_pool_release_metadata_snap(pool
->pmd
);
3695 DMWARN("release_metadata_snap message failed.");
3701 * Messages supported:
3702 * create_thin <dev_id>
3703 * create_snap <dev_id> <origin_id>
3705 * set_transaction_id <current_trans_id> <new_trans_id>
3706 * reserve_metadata_snap
3707 * release_metadata_snap
3709 static int pool_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
3712 struct pool_c
*pt
= ti
->private;
3713 struct pool
*pool
= pt
->pool
;
3715 if (get_pool_mode(pool
) >= PM_READ_ONLY
) {
3716 DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode",
3717 dm_device_name(pool
->pool_md
));
3721 if (!strcasecmp(argv
[0], "create_thin"))
3722 r
= process_create_thin_mesg(argc
, argv
, pool
);
3724 else if (!strcasecmp(argv
[0], "create_snap"))
3725 r
= process_create_snap_mesg(argc
, argv
, pool
);
3727 else if (!strcasecmp(argv
[0], "delete"))
3728 r
= process_delete_mesg(argc
, argv
, pool
);
3730 else if (!strcasecmp(argv
[0], "set_transaction_id"))
3731 r
= process_set_transaction_id_mesg(argc
, argv
, pool
);
3733 else if (!strcasecmp(argv
[0], "reserve_metadata_snap"))
3734 r
= process_reserve_metadata_snap_mesg(argc
, argv
, pool
);
3736 else if (!strcasecmp(argv
[0], "release_metadata_snap"))
3737 r
= process_release_metadata_snap_mesg(argc
, argv
, pool
);
3740 DMWARN("Unrecognised thin pool target message received: %s", argv
[0]);
3743 (void) commit(pool
);
3748 static void emit_flags(struct pool_features
*pf
, char *result
,
3749 unsigned sz
, unsigned maxlen
)
3751 unsigned count
= !pf
->zero_new_blocks
+ !pf
->discard_enabled
+
3752 !pf
->discard_passdown
+ (pf
->mode
== PM_READ_ONLY
) +
3753 pf
->error_if_no_space
;
3754 DMEMIT("%u ", count
);
3756 if (!pf
->zero_new_blocks
)
3757 DMEMIT("skip_block_zeroing ");
3759 if (!pf
->discard_enabled
)
3760 DMEMIT("ignore_discard ");
3762 if (!pf
->discard_passdown
)
3763 DMEMIT("no_discard_passdown ");
3765 if (pf
->mode
== PM_READ_ONLY
)
3766 DMEMIT("read_only ");
3768 if (pf
->error_if_no_space
)
3769 DMEMIT("error_if_no_space ");
3774 * <transaction id> <used metadata sectors>/<total metadata sectors>
3775 * <used data sectors>/<total data sectors> <held metadata root>
3776 * <pool mode> <discard config> <no space config> <needs_check>
3778 static void pool_status(struct dm_target
*ti
, status_type_t type
,
3779 unsigned status_flags
, char *result
, unsigned maxlen
)
3783 uint64_t transaction_id
;
3784 dm_block_t nr_free_blocks_data
;
3785 dm_block_t nr_free_blocks_metadata
;
3786 dm_block_t nr_blocks_data
;
3787 dm_block_t nr_blocks_metadata
;
3788 dm_block_t held_root
;
3789 char buf
[BDEVNAME_SIZE
];
3790 char buf2
[BDEVNAME_SIZE
];
3791 struct pool_c
*pt
= ti
->private;
3792 struct pool
*pool
= pt
->pool
;
3795 case STATUSTYPE_INFO
:
3796 if (get_pool_mode(pool
) == PM_FAIL
) {
3801 /* Commit to ensure statistics aren't out-of-date */
3802 if (!(status_flags
& DM_STATUS_NOFLUSH_FLAG
) && !dm_suspended(ti
))
3803 (void) commit(pool
);
3805 r
= dm_pool_get_metadata_transaction_id(pool
->pmd
, &transaction_id
);
3807 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
3808 dm_device_name(pool
->pool_md
), r
);
3812 r
= dm_pool_get_free_metadata_block_count(pool
->pmd
, &nr_free_blocks_metadata
);
3814 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
3815 dm_device_name(pool
->pool_md
), r
);
3819 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &nr_blocks_metadata
);
3821 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
3822 dm_device_name(pool
->pool_md
), r
);
3826 r
= dm_pool_get_free_block_count(pool
->pmd
, &nr_free_blocks_data
);
3828 DMERR("%s: dm_pool_get_free_block_count returned %d",
3829 dm_device_name(pool
->pool_md
), r
);
3833 r
= dm_pool_get_data_dev_size(pool
->pmd
, &nr_blocks_data
);
3835 DMERR("%s: dm_pool_get_data_dev_size returned %d",
3836 dm_device_name(pool
->pool_md
), r
);
3840 r
= dm_pool_get_metadata_snap(pool
->pmd
, &held_root
);
3842 DMERR("%s: dm_pool_get_metadata_snap returned %d",
3843 dm_device_name(pool
->pool_md
), r
);
3847 DMEMIT("%llu %llu/%llu %llu/%llu ",
3848 (unsigned long long)transaction_id
,
3849 (unsigned long long)(nr_blocks_metadata
- nr_free_blocks_metadata
),
3850 (unsigned long long)nr_blocks_metadata
,
3851 (unsigned long long)(nr_blocks_data
- nr_free_blocks_data
),
3852 (unsigned long long)nr_blocks_data
);
3855 DMEMIT("%llu ", held_root
);
3859 if (pool
->pf
.mode
== PM_OUT_OF_DATA_SPACE
)
3860 DMEMIT("out_of_data_space ");
3861 else if (pool
->pf
.mode
== PM_READ_ONLY
)
3866 if (!pool
->pf
.discard_enabled
)
3867 DMEMIT("ignore_discard ");
3868 else if (pool
->pf
.discard_passdown
)
3869 DMEMIT("discard_passdown ");
3871 DMEMIT("no_discard_passdown ");
3873 if (pool
->pf
.error_if_no_space
)
3874 DMEMIT("error_if_no_space ");
3876 DMEMIT("queue_if_no_space ");
3878 if (dm_pool_metadata_needs_check(pool
->pmd
))
3879 DMEMIT("needs_check ");
3885 case STATUSTYPE_TABLE
:
3886 DMEMIT("%s %s %lu %llu ",
3887 format_dev_t(buf
, pt
->metadata_dev
->bdev
->bd_dev
),
3888 format_dev_t(buf2
, pt
->data_dev
->bdev
->bd_dev
),
3889 (unsigned long)pool
->sectors_per_block
,
3890 (unsigned long long)pt
->low_water_blocks
);
3891 emit_flags(&pt
->requested_pf
, result
, sz
, maxlen
);
3900 static int pool_iterate_devices(struct dm_target
*ti
,
3901 iterate_devices_callout_fn fn
, void *data
)
3903 struct pool_c
*pt
= ti
->private;
3905 return fn(ti
, pt
->data_dev
, 0, ti
->len
, data
);
3908 static void pool_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
3910 struct pool_c
*pt
= ti
->private;
3911 struct pool
*pool
= pt
->pool
;
3912 sector_t io_opt_sectors
= limits
->io_opt
>> SECTOR_SHIFT
;
3915 * If max_sectors is smaller than pool->sectors_per_block adjust it
3916 * to the highest possible power-of-2 factor of pool->sectors_per_block.
3917 * This is especially beneficial when the pool's data device is a RAID
3918 * device that has a full stripe width that matches pool->sectors_per_block
3919 * -- because even though partial RAID stripe-sized IOs will be issued to a
3920 * single RAID stripe; when aggregated they will end on a full RAID stripe
3921 * boundary.. which avoids additional partial RAID stripe writes cascading
3923 if (limits
->max_sectors
< pool
->sectors_per_block
) {
3924 while (!is_factor(pool
->sectors_per_block
, limits
->max_sectors
)) {
3925 if ((limits
->max_sectors
& (limits
->max_sectors
- 1)) == 0)
3926 limits
->max_sectors
--;
3927 limits
->max_sectors
= rounddown_pow_of_two(limits
->max_sectors
);
3932 * If the system-determined stacked limits are compatible with the
3933 * pool's blocksize (io_opt is a factor) do not override them.
3935 if (io_opt_sectors
< pool
->sectors_per_block
||
3936 !is_factor(io_opt_sectors
, pool
->sectors_per_block
)) {
3937 if (is_factor(pool
->sectors_per_block
, limits
->max_sectors
))
3938 blk_limits_io_min(limits
, limits
->max_sectors
<< SECTOR_SHIFT
);
3940 blk_limits_io_min(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
3941 blk_limits_io_opt(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
3945 * pt->adjusted_pf is a staging area for the actual features to use.
3946 * They get transferred to the live pool in bind_control_target()
3947 * called from pool_preresume().
3949 if (!pt
->adjusted_pf
.discard_enabled
) {
3951 * Must explicitly disallow stacking discard limits otherwise the
3952 * block layer will stack them if pool's data device has support.
3953 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
3954 * user to see that, so make sure to set all discard limits to 0.
3956 limits
->discard_granularity
= 0;
3960 disable_passdown_if_not_supported(pt
);
3963 * The pool uses the same discard limits as the underlying data
3964 * device. DM core has already set this up.
3968 static struct target_type pool_target
= {
3969 .name
= "thin-pool",
3970 .features
= DM_TARGET_SINGLETON
| DM_TARGET_ALWAYS_WRITEABLE
|
3971 DM_TARGET_IMMUTABLE
,
3972 .version
= {1, 19, 0},
3973 .module
= THIS_MODULE
,
3977 .presuspend
= pool_presuspend
,
3978 .presuspend_undo
= pool_presuspend_undo
,
3979 .postsuspend
= pool_postsuspend
,
3980 .preresume
= pool_preresume
,
3981 .resume
= pool_resume
,
3982 .message
= pool_message
,
3983 .status
= pool_status
,
3984 .iterate_devices
= pool_iterate_devices
,
3985 .io_hints
= pool_io_hints
,
3988 /*----------------------------------------------------------------
3989 * Thin target methods
3990 *--------------------------------------------------------------*/
3991 static void thin_get(struct thin_c
*tc
)
3993 atomic_inc(&tc
->refcount
);
3996 static void thin_put(struct thin_c
*tc
)
3998 if (atomic_dec_and_test(&tc
->refcount
))
3999 complete(&tc
->can_destroy
);
4002 static void thin_dtr(struct dm_target
*ti
)
4004 struct thin_c
*tc
= ti
->private;
4005 unsigned long flags
;
4007 spin_lock_irqsave(&tc
->pool
->lock
, flags
);
4008 list_del_rcu(&tc
->list
);
4009 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
4013 wait_for_completion(&tc
->can_destroy
);
4015 mutex_lock(&dm_thin_pool_table
.mutex
);
4017 __pool_dec(tc
->pool
);
4018 dm_pool_close_thin_device(tc
->td
);
4019 dm_put_device(ti
, tc
->pool_dev
);
4021 dm_put_device(ti
, tc
->origin_dev
);
4024 mutex_unlock(&dm_thin_pool_table
.mutex
);
4028 * Thin target parameters:
4030 * <pool_dev> <dev_id> [origin_dev]
4032 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
4033 * dev_id: the internal device identifier
4034 * origin_dev: a device external to the pool that should act as the origin
4036 * If the pool device has discards disabled, they get disabled for the thin
4039 static int thin_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
4043 struct dm_dev
*pool_dev
, *origin_dev
;
4044 struct mapped_device
*pool_md
;
4045 unsigned long flags
;
4047 mutex_lock(&dm_thin_pool_table
.mutex
);
4049 if (argc
!= 2 && argc
!= 3) {
4050 ti
->error
= "Invalid argument count";
4055 tc
= ti
->private = kzalloc(sizeof(*tc
), GFP_KERNEL
);
4057 ti
->error
= "Out of memory";
4061 tc
->thin_md
= dm_table_get_md(ti
->table
);
4062 spin_lock_init(&tc
->lock
);
4063 INIT_LIST_HEAD(&tc
->deferred_cells
);
4064 bio_list_init(&tc
->deferred_bio_list
);
4065 bio_list_init(&tc
->retry_on_resume_list
);
4066 tc
->sort_bio_list
= RB_ROOT
;
4069 r
= dm_get_device(ti
, argv
[2], FMODE_READ
, &origin_dev
);
4071 ti
->error
= "Error opening origin device";
4072 goto bad_origin_dev
;
4074 tc
->origin_dev
= origin_dev
;
4077 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &pool_dev
);
4079 ti
->error
= "Error opening pool device";
4082 tc
->pool_dev
= pool_dev
;
4084 if (read_dev_id(argv
[1], (unsigned long long *)&tc
->dev_id
, 0)) {
4085 ti
->error
= "Invalid device id";
4090 pool_md
= dm_get_md(tc
->pool_dev
->bdev
->bd_dev
);
4092 ti
->error
= "Couldn't get pool mapped device";
4097 tc
->pool
= __pool_table_lookup(pool_md
);
4099 ti
->error
= "Couldn't find pool object";
4101 goto bad_pool_lookup
;
4103 __pool_inc(tc
->pool
);
4105 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
4106 ti
->error
= "Couldn't open thin device, Pool is in fail mode";
4111 r
= dm_pool_open_thin_device(tc
->pool
->pmd
, tc
->dev_id
, &tc
->td
);
4113 ti
->error
= "Couldn't open thin internal device";
4117 r
= dm_set_target_max_io_len(ti
, tc
->pool
->sectors_per_block
);
4121 ti
->num_flush_bios
= 1;
4122 ti
->flush_supported
= true;
4123 ti
->per_io_data_size
= sizeof(struct dm_thin_endio_hook
);
4125 /* In case the pool supports discards, pass them on. */
4126 ti
->discard_zeroes_data_unsupported
= true;
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
, int err
)
4187 unsigned long flags
;
4188 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
4189 struct list_head work
;
4190 struct dm_thin_new_mapping
*m
, *tmp
;
4191 struct pool
*pool
= h
->tc
->pool
;
4193 if (h
->shared_read_entry
) {
4194 INIT_LIST_HEAD(&work
);
4195 dm_deferred_entry_dec(h
->shared_read_entry
, &work
);
4197 spin_lock_irqsave(&pool
->lock
, flags
);
4198 list_for_each_entry_safe(m
, tmp
, &work
, list
) {
4200 __complete_mapping_preparation(m
);
4202 spin_unlock_irqrestore(&pool
->lock
, flags
);
4205 if (h
->all_io_entry
) {
4206 INIT_LIST_HEAD(&work
);
4207 dm_deferred_entry_dec(h
->all_io_entry
, &work
);
4208 if (!list_empty(&work
)) {
4209 spin_lock_irqsave(&pool
->lock
, flags
);
4210 list_for_each_entry_safe(m
, tmp
, &work
, list
)
4211 list_add_tail(&m
->list
, &pool
->prepared_discards
);
4212 spin_unlock_irqrestore(&pool
->lock
, flags
);
4218 cell_defer_no_holder(h
->tc
, h
->cell
);
4223 static void thin_presuspend(struct dm_target
*ti
)
4225 struct thin_c
*tc
= ti
->private;
4227 if (dm_noflush_suspending(ti
))
4228 noflush_work(tc
, do_noflush_start
);
4231 static void thin_postsuspend(struct dm_target
*ti
)
4233 struct thin_c
*tc
= ti
->private;
4236 * The dm_noflush_suspending flag has been cleared by now, so
4237 * unfortunately we must always run this.
4239 noflush_work(tc
, do_noflush_stop
);
4242 static int thin_preresume(struct dm_target
*ti
)
4244 struct thin_c
*tc
= ti
->private;
4247 tc
->origin_size
= get_dev_size(tc
->origin_dev
->bdev
);
4253 * <nr mapped sectors> <highest mapped sector>
4255 static void thin_status(struct dm_target
*ti
, status_type_t type
,
4256 unsigned status_flags
, char *result
, unsigned maxlen
)
4260 dm_block_t mapped
, highest
;
4261 char buf
[BDEVNAME_SIZE
];
4262 struct thin_c
*tc
= ti
->private;
4264 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
4273 case STATUSTYPE_INFO
:
4274 r
= dm_thin_get_mapped_count(tc
->td
, &mapped
);
4276 DMERR("dm_thin_get_mapped_count returned %d", r
);
4280 r
= dm_thin_get_highest_mapped_block(tc
->td
, &highest
);
4282 DMERR("dm_thin_get_highest_mapped_block returned %d", r
);
4286 DMEMIT("%llu ", mapped
* tc
->pool
->sectors_per_block
);
4288 DMEMIT("%llu", ((highest
+ 1) *
4289 tc
->pool
->sectors_per_block
) - 1);
4294 case STATUSTYPE_TABLE
:
4296 format_dev_t(buf
, tc
->pool_dev
->bdev
->bd_dev
),
4297 (unsigned long) tc
->dev_id
);
4299 DMEMIT(" %s", format_dev_t(buf
, tc
->origin_dev
->bdev
->bd_dev
));
4310 static int thin_iterate_devices(struct dm_target
*ti
,
4311 iterate_devices_callout_fn fn
, void *data
)
4314 struct thin_c
*tc
= ti
->private;
4315 struct pool
*pool
= tc
->pool
;
4318 * We can't call dm_pool_get_data_dev_size() since that blocks. So
4319 * we follow a more convoluted path through to the pool's target.
4322 return 0; /* nothing is bound */
4324 blocks
= pool
->ti
->len
;
4325 (void) sector_div(blocks
, pool
->sectors_per_block
);
4327 return fn(ti
, tc
->pool_dev
, 0, pool
->sectors_per_block
* blocks
, data
);
4332 static void thin_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
4334 struct thin_c
*tc
= ti
->private;
4335 struct pool
*pool
= tc
->pool
;
4337 if (!pool
->pf
.discard_enabled
)
4340 limits
->discard_granularity
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
4341 limits
->max_discard_sectors
= 2048 * 1024 * 16; /* 16G */
4344 static struct target_type thin_target
= {
4346 .version
= {1, 19, 0},
4347 .module
= THIS_MODULE
,
4351 .end_io
= thin_endio
,
4352 .preresume
= thin_preresume
,
4353 .presuspend
= thin_presuspend
,
4354 .postsuspend
= thin_postsuspend
,
4355 .status
= thin_status
,
4356 .iterate_devices
= thin_iterate_devices
,
4357 .io_hints
= thin_io_hints
,
4360 /*----------------------------------------------------------------*/
4362 static int __init
dm_thin_init(void)
4368 r
= dm_register_target(&thin_target
);
4372 r
= dm_register_target(&pool_target
);
4374 goto bad_pool_target
;
4378 _new_mapping_cache
= KMEM_CACHE(dm_thin_new_mapping
, 0);
4379 if (!_new_mapping_cache
)
4380 goto bad_new_mapping_cache
;
4384 bad_new_mapping_cache
:
4385 dm_unregister_target(&pool_target
);
4387 dm_unregister_target(&thin_target
);
4392 static void dm_thin_exit(void)
4394 dm_unregister_target(&thin_target
);
4395 dm_unregister_target(&pool_target
);
4397 kmem_cache_destroy(_new_mapping_cache
);
4400 module_init(dm_thin_init
);
4401 module_exit(dm_thin_exit
);
4403 module_param_named(no_space_timeout
, no_space_timeout_secs
, uint
, S_IRUGO
| S_IWUSR
);
4404 MODULE_PARM_DESC(no_space_timeout
, "Out of data space queue IO timeout in seconds");
4406 MODULE_DESCRIPTION(DM_NAME
" thin provisioning target");
4407 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
4408 MODULE_LICENSE("GPL");