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 active_thins
;
258 struct dm_deferred_set
*shared_read_ds
;
259 struct dm_deferred_set
*all_io_ds
;
261 struct dm_thin_new_mapping
*next_mapping
;
262 mempool_t
*mapping_pool
;
264 process_bio_fn process_bio
;
265 process_bio_fn process_discard
;
267 process_cell_fn process_cell
;
268 process_cell_fn process_discard_cell
;
270 process_mapping_fn process_prepared_mapping
;
271 process_mapping_fn process_prepared_discard
;
273 struct dm_bio_prison_cell
**cell_sort_array
;
276 static enum pool_mode
get_pool_mode(struct pool
*pool
);
277 static void metadata_operation_failed(struct pool
*pool
, const char *op
, int r
);
280 * Target context for a pool.
283 struct dm_target
*ti
;
285 struct dm_dev
*data_dev
;
286 struct dm_dev
*metadata_dev
;
287 struct dm_target_callbacks callbacks
;
289 dm_block_t low_water_blocks
;
290 struct pool_features requested_pf
; /* Features requested during table load */
291 struct pool_features adjusted_pf
; /* Features used after adjusting for constituent devices */
295 * Target context for a thin.
298 struct list_head list
;
299 struct dm_dev
*pool_dev
;
300 struct dm_dev
*origin_dev
;
301 sector_t origin_size
;
305 struct dm_thin_device
*td
;
306 struct mapped_device
*thin_md
;
310 struct list_head deferred_cells
;
311 struct bio_list deferred_bio_list
;
312 struct bio_list retry_on_resume_list
;
313 struct rb_root sort_bio_list
; /* sorted list of deferred bios */
316 * Ensures the thin is not destroyed until the worker has finished
317 * iterating the active_thins list.
320 struct completion can_destroy
;
323 /*----------------------------------------------------------------*/
325 static bool block_size_is_power_of_two(struct pool
*pool
)
327 return pool
->sectors_per_block_shift
>= 0;
330 static sector_t
block_to_sectors(struct pool
*pool
, dm_block_t b
)
332 return block_size_is_power_of_two(pool
) ?
333 (b
<< pool
->sectors_per_block_shift
) :
334 (b
* pool
->sectors_per_block
);
337 /*----------------------------------------------------------------*/
341 struct blk_plug plug
;
342 struct bio
*parent_bio
;
346 static void begin_discard(struct discard_op
*op
, struct thin_c
*tc
, struct bio
*parent
)
351 blk_start_plug(&op
->plug
);
352 op
->parent_bio
= parent
;
356 static int issue_discard(struct discard_op
*op
, dm_block_t data_b
, dm_block_t data_e
)
358 struct thin_c
*tc
= op
->tc
;
359 sector_t s
= block_to_sectors(tc
->pool
, data_b
);
360 sector_t len
= block_to_sectors(tc
->pool
, data_e
- data_b
);
362 return __blkdev_issue_discard(tc
->pool_dev
->bdev
, s
, len
,
363 GFP_NOWAIT
, 0, &op
->bio
);
366 static void end_discard(struct discard_op
*op
, int r
)
370 * Even if one of the calls to issue_discard failed, we
371 * need to wait for the chain to complete.
373 bio_chain(op
->bio
, op
->parent_bio
);
374 bio_set_op_attrs(op
->bio
, REQ_OP_DISCARD
, 0);
378 blk_finish_plug(&op
->plug
);
381 * Even if r is set, there could be sub discards in flight that we
384 if (r
&& !op
->parent_bio
->bi_error
)
385 op
->parent_bio
->bi_error
= r
;
386 bio_endio(op
->parent_bio
);
389 /*----------------------------------------------------------------*/
392 * wake_worker() is used when new work is queued and when pool_resume is
393 * ready to continue deferred IO processing.
395 static void wake_worker(struct pool
*pool
)
397 queue_work(pool
->wq
, &pool
->worker
);
400 /*----------------------------------------------------------------*/
402 static int bio_detain(struct pool
*pool
, struct dm_cell_key
*key
, struct bio
*bio
,
403 struct dm_bio_prison_cell
**cell_result
)
406 struct dm_bio_prison_cell
*cell_prealloc
;
409 * Allocate a cell from the prison's mempool.
410 * This might block but it can't fail.
412 cell_prealloc
= dm_bio_prison_alloc_cell(pool
->prison
, GFP_NOIO
);
414 r
= dm_bio_detain(pool
->prison
, key
, bio
, cell_prealloc
, cell_result
);
417 * We reused an old cell; we can get rid of
420 dm_bio_prison_free_cell(pool
->prison
, cell_prealloc
);
425 static void cell_release(struct pool
*pool
,
426 struct dm_bio_prison_cell
*cell
,
427 struct bio_list
*bios
)
429 dm_cell_release(pool
->prison
, cell
, bios
);
430 dm_bio_prison_free_cell(pool
->prison
, cell
);
433 static void cell_visit_release(struct pool
*pool
,
434 void (*fn
)(void *, struct dm_bio_prison_cell
*),
436 struct dm_bio_prison_cell
*cell
)
438 dm_cell_visit_release(pool
->prison
, fn
, context
, cell
);
439 dm_bio_prison_free_cell(pool
->prison
, cell
);
442 static void cell_release_no_holder(struct pool
*pool
,
443 struct dm_bio_prison_cell
*cell
,
444 struct bio_list
*bios
)
446 dm_cell_release_no_holder(pool
->prison
, cell
, bios
);
447 dm_bio_prison_free_cell(pool
->prison
, cell
);
450 static void cell_error_with_code(struct pool
*pool
,
451 struct dm_bio_prison_cell
*cell
, int error_code
)
453 dm_cell_error(pool
->prison
, cell
, error_code
);
454 dm_bio_prison_free_cell(pool
->prison
, cell
);
457 static int get_pool_io_error_code(struct pool
*pool
)
459 return pool
->out_of_data_space
? -ENOSPC
: -EIO
;
462 static void cell_error(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
464 int error
= get_pool_io_error_code(pool
);
466 cell_error_with_code(pool
, cell
, error
);
469 static void cell_success(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
471 cell_error_with_code(pool
, cell
, 0);
474 static void cell_requeue(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
476 cell_error_with_code(pool
, cell
, DM_ENDIO_REQUEUE
);
479 /*----------------------------------------------------------------*/
482 * A global list of pools that uses a struct mapped_device as a key.
484 static struct dm_thin_pool_table
{
486 struct list_head pools
;
487 } dm_thin_pool_table
;
489 static void pool_table_init(void)
491 mutex_init(&dm_thin_pool_table
.mutex
);
492 INIT_LIST_HEAD(&dm_thin_pool_table
.pools
);
495 static void __pool_table_insert(struct pool
*pool
)
497 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
498 list_add(&pool
->list
, &dm_thin_pool_table
.pools
);
501 static void __pool_table_remove(struct pool
*pool
)
503 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
504 list_del(&pool
->list
);
507 static struct pool
*__pool_table_lookup(struct mapped_device
*md
)
509 struct pool
*pool
= NULL
, *tmp
;
511 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
513 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
514 if (tmp
->pool_md
== md
) {
523 static struct pool
*__pool_table_lookup_metadata_dev(struct block_device
*md_dev
)
525 struct pool
*pool
= NULL
, *tmp
;
527 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
529 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
530 if (tmp
->md_dev
== md_dev
) {
539 /*----------------------------------------------------------------*/
541 struct dm_thin_endio_hook
{
543 struct dm_deferred_entry
*shared_read_entry
;
544 struct dm_deferred_entry
*all_io_entry
;
545 struct dm_thin_new_mapping
*overwrite_mapping
;
546 struct rb_node rb_node
;
547 struct dm_bio_prison_cell
*cell
;
550 static void __merge_bio_list(struct bio_list
*bios
, struct bio_list
*master
)
552 bio_list_merge(bios
, master
);
553 bio_list_init(master
);
556 static void error_bio_list(struct bio_list
*bios
, int error
)
560 while ((bio
= bio_list_pop(bios
))) {
561 bio
->bi_error
= error
;
566 static void error_thin_bio_list(struct thin_c
*tc
, struct bio_list
*master
, int error
)
568 struct bio_list bios
;
571 bio_list_init(&bios
);
573 spin_lock_irqsave(&tc
->lock
, flags
);
574 __merge_bio_list(&bios
, master
);
575 spin_unlock_irqrestore(&tc
->lock
, flags
);
577 error_bio_list(&bios
, error
);
580 static void requeue_deferred_cells(struct thin_c
*tc
)
582 struct pool
*pool
= tc
->pool
;
584 struct list_head cells
;
585 struct dm_bio_prison_cell
*cell
, *tmp
;
587 INIT_LIST_HEAD(&cells
);
589 spin_lock_irqsave(&tc
->lock
, flags
);
590 list_splice_init(&tc
->deferred_cells
, &cells
);
591 spin_unlock_irqrestore(&tc
->lock
, flags
);
593 list_for_each_entry_safe(cell
, tmp
, &cells
, user_list
)
594 cell_requeue(pool
, cell
);
597 static void requeue_io(struct thin_c
*tc
)
599 struct bio_list bios
;
602 bio_list_init(&bios
);
604 spin_lock_irqsave(&tc
->lock
, flags
);
605 __merge_bio_list(&bios
, &tc
->deferred_bio_list
);
606 __merge_bio_list(&bios
, &tc
->retry_on_resume_list
);
607 spin_unlock_irqrestore(&tc
->lock
, flags
);
609 error_bio_list(&bios
, DM_ENDIO_REQUEUE
);
610 requeue_deferred_cells(tc
);
613 static void error_retry_list_with_code(struct pool
*pool
, int error
)
618 list_for_each_entry_rcu(tc
, &pool
->active_thins
, list
)
619 error_thin_bio_list(tc
, &tc
->retry_on_resume_list
, error
);
623 static void error_retry_list(struct pool
*pool
)
625 int error
= get_pool_io_error_code(pool
);
627 error_retry_list_with_code(pool
, error
);
631 * This section of code contains the logic for processing a thin device's IO.
632 * Much of the code depends on pool object resources (lists, workqueues, etc)
633 * but most is exclusively called from the thin target rather than the thin-pool
637 static dm_block_t
get_bio_block(struct thin_c
*tc
, struct bio
*bio
)
639 struct pool
*pool
= tc
->pool
;
640 sector_t block_nr
= bio
->bi_iter
.bi_sector
;
642 if (block_size_is_power_of_two(pool
))
643 block_nr
>>= pool
->sectors_per_block_shift
;
645 (void) sector_div(block_nr
, pool
->sectors_per_block
);
651 * Returns the _complete_ blocks that this bio covers.
653 static void get_bio_block_range(struct thin_c
*tc
, struct bio
*bio
,
654 dm_block_t
*begin
, dm_block_t
*end
)
656 struct pool
*pool
= tc
->pool
;
657 sector_t b
= bio
->bi_iter
.bi_sector
;
658 sector_t e
= b
+ (bio
->bi_iter
.bi_size
>> SECTOR_SHIFT
);
660 b
+= pool
->sectors_per_block
- 1ull; /* so we round up */
662 if (block_size_is_power_of_two(pool
)) {
663 b
>>= pool
->sectors_per_block_shift
;
664 e
>>= pool
->sectors_per_block_shift
;
666 (void) sector_div(b
, pool
->sectors_per_block
);
667 (void) sector_div(e
, pool
->sectors_per_block
);
671 /* Can happen if the bio is within a single block. */
678 static void remap(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
)
680 struct pool
*pool
= tc
->pool
;
681 sector_t bi_sector
= bio
->bi_iter
.bi_sector
;
683 bio
->bi_bdev
= tc
->pool_dev
->bdev
;
684 if (block_size_is_power_of_two(pool
))
685 bio
->bi_iter
.bi_sector
=
686 (block
<< pool
->sectors_per_block_shift
) |
687 (bi_sector
& (pool
->sectors_per_block
- 1));
689 bio
->bi_iter
.bi_sector
= (block
* pool
->sectors_per_block
) +
690 sector_div(bi_sector
, pool
->sectors_per_block
);
693 static void remap_to_origin(struct thin_c
*tc
, struct bio
*bio
)
695 bio
->bi_bdev
= tc
->origin_dev
->bdev
;
698 static int bio_triggers_commit(struct thin_c
*tc
, struct bio
*bio
)
700 return (bio
->bi_rw
& (REQ_PREFLUSH
| REQ_FUA
)) &&
701 dm_thin_changed_this_transaction(tc
->td
);
704 static void inc_all_io_entry(struct pool
*pool
, struct bio
*bio
)
706 struct dm_thin_endio_hook
*h
;
708 if (bio_op(bio
) == REQ_OP_DISCARD
)
711 h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
712 h
->all_io_entry
= dm_deferred_entry_inc(pool
->all_io_ds
);
715 static void issue(struct thin_c
*tc
, struct bio
*bio
)
717 struct pool
*pool
= tc
->pool
;
720 if (!bio_triggers_commit(tc
, bio
)) {
721 generic_make_request(bio
);
726 * Complete bio with an error if earlier I/O caused changes to
727 * the metadata that can't be committed e.g, due to I/O errors
728 * on the metadata device.
730 if (dm_thin_aborted_changes(tc
->td
)) {
736 * Batch together any bios that trigger commits and then issue a
737 * single commit for them in process_deferred_bios().
739 spin_lock_irqsave(&pool
->lock
, flags
);
740 bio_list_add(&pool
->deferred_flush_bios
, bio
);
741 spin_unlock_irqrestore(&pool
->lock
, flags
);
744 static void remap_to_origin_and_issue(struct thin_c
*tc
, struct bio
*bio
)
746 remap_to_origin(tc
, bio
);
750 static void remap_and_issue(struct thin_c
*tc
, struct bio
*bio
,
753 remap(tc
, bio
, block
);
757 /*----------------------------------------------------------------*/
760 * Bio endio functions.
762 struct dm_thin_new_mapping
{
763 struct list_head list
;
769 * Track quiescing, copying and zeroing preparation actions. When this
770 * counter hits zero the block is prepared and can be inserted into the
773 atomic_t prepare_actions
;
777 dm_block_t virt_begin
, virt_end
;
778 dm_block_t data_block
;
779 struct dm_bio_prison_cell
*cell
;
782 * If the bio covers the whole area of a block then we can avoid
783 * zeroing or copying. Instead this bio is hooked. The bio will
784 * still be in the cell, so care has to be taken to avoid issuing
788 bio_end_io_t
*saved_bi_end_io
;
791 static void __complete_mapping_preparation(struct dm_thin_new_mapping
*m
)
793 struct pool
*pool
= m
->tc
->pool
;
795 if (atomic_dec_and_test(&m
->prepare_actions
)) {
796 list_add_tail(&m
->list
, &pool
->prepared_mappings
);
801 static void complete_mapping_preparation(struct dm_thin_new_mapping
*m
)
804 struct pool
*pool
= m
->tc
->pool
;
806 spin_lock_irqsave(&pool
->lock
, flags
);
807 __complete_mapping_preparation(m
);
808 spin_unlock_irqrestore(&pool
->lock
, flags
);
811 static void copy_complete(int read_err
, unsigned long write_err
, void *context
)
813 struct dm_thin_new_mapping
*m
= context
;
815 m
->err
= read_err
|| write_err
? -EIO
: 0;
816 complete_mapping_preparation(m
);
819 static void overwrite_endio(struct bio
*bio
)
821 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
822 struct dm_thin_new_mapping
*m
= h
->overwrite_mapping
;
824 bio
->bi_end_io
= m
->saved_bi_end_io
;
826 m
->err
= bio
->bi_error
;
827 complete_mapping_preparation(m
);
830 /*----------------------------------------------------------------*/
837 * Prepared mapping jobs.
841 * This sends the bios in the cell, except the original holder, back
842 * to the deferred_bios list.
844 static void cell_defer_no_holder(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
846 struct pool
*pool
= tc
->pool
;
849 spin_lock_irqsave(&tc
->lock
, flags
);
850 cell_release_no_holder(pool
, cell
, &tc
->deferred_bio_list
);
851 spin_unlock_irqrestore(&tc
->lock
, flags
);
856 static void thin_defer_bio(struct thin_c
*tc
, struct bio
*bio
);
860 struct bio_list defer_bios
;
861 struct bio_list issue_bios
;
864 static void __inc_remap_and_issue_cell(void *context
,
865 struct dm_bio_prison_cell
*cell
)
867 struct remap_info
*info
= context
;
870 while ((bio
= bio_list_pop(&cell
->bios
))) {
871 if (bio
->bi_rw
& (REQ_PREFLUSH
| REQ_FUA
) ||
872 bio_op(bio
) == REQ_OP_DISCARD
)
873 bio_list_add(&info
->defer_bios
, bio
);
875 inc_all_io_entry(info
->tc
->pool
, bio
);
878 * We can't issue the bios with the bio prison lock
879 * held, so we add them to a list to issue on
880 * return from this function.
882 bio_list_add(&info
->issue_bios
, bio
);
887 static void inc_remap_and_issue_cell(struct thin_c
*tc
,
888 struct dm_bio_prison_cell
*cell
,
892 struct remap_info info
;
895 bio_list_init(&info
.defer_bios
);
896 bio_list_init(&info
.issue_bios
);
899 * We have to be careful to inc any bios we're about to issue
900 * before the cell is released, and avoid a race with new bios
901 * being added to the cell.
903 cell_visit_release(tc
->pool
, __inc_remap_and_issue_cell
,
906 while ((bio
= bio_list_pop(&info
.defer_bios
)))
907 thin_defer_bio(tc
, bio
);
909 while ((bio
= bio_list_pop(&info
.issue_bios
)))
910 remap_and_issue(info
.tc
, bio
, block
);
913 static void process_prepared_mapping_fail(struct dm_thin_new_mapping
*m
)
915 cell_error(m
->tc
->pool
, m
->cell
);
917 mempool_free(m
, m
->tc
->pool
->mapping_pool
);
920 static void process_prepared_mapping(struct dm_thin_new_mapping
*m
)
922 struct thin_c
*tc
= m
->tc
;
923 struct pool
*pool
= tc
->pool
;
924 struct bio
*bio
= m
->bio
;
928 cell_error(pool
, m
->cell
);
933 * Commit the prepared block into the mapping btree.
934 * Any I/O for this block arriving after this point will get
935 * remapped to it directly.
937 r
= dm_thin_insert_block(tc
->td
, m
->virt_begin
, m
->data_block
);
939 metadata_operation_failed(pool
, "dm_thin_insert_block", r
);
940 cell_error(pool
, m
->cell
);
945 * Release any bios held while the block was being provisioned.
946 * If we are processing a write bio that completely covers the block,
947 * we already processed it so can ignore it now when processing
948 * the bios in the cell.
951 inc_remap_and_issue_cell(tc
, m
->cell
, m
->data_block
);
954 inc_all_io_entry(tc
->pool
, m
->cell
->holder
);
955 remap_and_issue(tc
, m
->cell
->holder
, m
->data_block
);
956 inc_remap_and_issue_cell(tc
, m
->cell
, m
->data_block
);
961 mempool_free(m
, pool
->mapping_pool
);
964 /*----------------------------------------------------------------*/
966 static void free_discard_mapping(struct dm_thin_new_mapping
*m
)
968 struct thin_c
*tc
= m
->tc
;
970 cell_defer_no_holder(tc
, m
->cell
);
971 mempool_free(m
, tc
->pool
->mapping_pool
);
974 static void process_prepared_discard_fail(struct dm_thin_new_mapping
*m
)
976 bio_io_error(m
->bio
);
977 free_discard_mapping(m
);
980 static void process_prepared_discard_success(struct dm_thin_new_mapping
*m
)
983 free_discard_mapping(m
);
986 static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping
*m
)
989 struct thin_c
*tc
= m
->tc
;
991 r
= dm_thin_remove_range(tc
->td
, m
->cell
->key
.block_begin
, m
->cell
->key
.block_end
);
993 metadata_operation_failed(tc
->pool
, "dm_thin_remove_range", r
);
994 bio_io_error(m
->bio
);
998 cell_defer_no_holder(tc
, m
->cell
);
999 mempool_free(m
, tc
->pool
->mapping_pool
);
1002 /*----------------------------------------------------------------*/
1004 static void passdown_double_checking_shared_status(struct dm_thin_new_mapping
*m
)
1007 * We've already unmapped this range of blocks, but before we
1008 * passdown we have to check that these blocks are now unused.
1012 struct thin_c
*tc
= m
->tc
;
1013 struct pool
*pool
= tc
->pool
;
1014 dm_block_t b
= m
->data_block
, e
, end
= m
->data_block
+ m
->virt_end
- m
->virt_begin
;
1015 struct discard_op op
;
1017 begin_discard(&op
, tc
, m
->bio
);
1019 /* find start of unmapped run */
1020 for (; b
< end
; b
++) {
1021 r
= dm_pool_block_is_used(pool
->pmd
, b
, &used
);
1032 /* find end of run */
1033 for (e
= b
+ 1; e
!= end
; e
++) {
1034 r
= dm_pool_block_is_used(pool
->pmd
, e
, &used
);
1042 r
= issue_discard(&op
, b
, e
);
1049 end_discard(&op
, r
);
1052 static void process_prepared_discard_passdown(struct dm_thin_new_mapping
*m
)
1055 struct thin_c
*tc
= m
->tc
;
1056 struct pool
*pool
= tc
->pool
;
1058 r
= dm_thin_remove_range(tc
->td
, m
->virt_begin
, m
->virt_end
);
1060 metadata_operation_failed(pool
, "dm_thin_remove_range", r
);
1061 bio_io_error(m
->bio
);
1063 } else if (m
->maybe_shared
) {
1064 passdown_double_checking_shared_status(m
);
1067 struct discard_op op
;
1068 begin_discard(&op
, tc
, m
->bio
);
1069 r
= issue_discard(&op
, m
->data_block
,
1070 m
->data_block
+ (m
->virt_end
- m
->virt_begin
));
1071 end_discard(&op
, r
);
1074 cell_defer_no_holder(tc
, m
->cell
);
1075 mempool_free(m
, pool
->mapping_pool
);
1078 static void process_prepared(struct pool
*pool
, struct list_head
*head
,
1079 process_mapping_fn
*fn
)
1081 unsigned long flags
;
1082 struct list_head maps
;
1083 struct dm_thin_new_mapping
*m
, *tmp
;
1085 INIT_LIST_HEAD(&maps
);
1086 spin_lock_irqsave(&pool
->lock
, flags
);
1087 list_splice_init(head
, &maps
);
1088 spin_unlock_irqrestore(&pool
->lock
, flags
);
1090 list_for_each_entry_safe(m
, tmp
, &maps
, list
)
1095 * Deferred bio jobs.
1097 static int io_overlaps_block(struct pool
*pool
, struct bio
*bio
)
1099 return bio
->bi_iter
.bi_size
==
1100 (pool
->sectors_per_block
<< SECTOR_SHIFT
);
1103 static int io_overwrites_block(struct pool
*pool
, struct bio
*bio
)
1105 return (bio_data_dir(bio
) == WRITE
) &&
1106 io_overlaps_block(pool
, bio
);
1109 static void save_and_set_endio(struct bio
*bio
, bio_end_io_t
**save
,
1112 *save
= bio
->bi_end_io
;
1113 bio
->bi_end_io
= fn
;
1116 static int ensure_next_mapping(struct pool
*pool
)
1118 if (pool
->next_mapping
)
1121 pool
->next_mapping
= mempool_alloc(pool
->mapping_pool
, GFP_ATOMIC
);
1123 return pool
->next_mapping
? 0 : -ENOMEM
;
1126 static struct dm_thin_new_mapping
*get_next_mapping(struct pool
*pool
)
1128 struct dm_thin_new_mapping
*m
= pool
->next_mapping
;
1130 BUG_ON(!pool
->next_mapping
);
1132 memset(m
, 0, sizeof(struct dm_thin_new_mapping
));
1133 INIT_LIST_HEAD(&m
->list
);
1136 pool
->next_mapping
= NULL
;
1141 static void ll_zero(struct thin_c
*tc
, struct dm_thin_new_mapping
*m
,
1142 sector_t begin
, sector_t end
)
1145 struct dm_io_region to
;
1147 to
.bdev
= tc
->pool_dev
->bdev
;
1149 to
.count
= end
- begin
;
1151 r
= dm_kcopyd_zero(tc
->pool
->copier
, 1, &to
, 0, copy_complete
, m
);
1153 DMERR_LIMIT("dm_kcopyd_zero() failed");
1154 copy_complete(1, 1, m
);
1158 static void remap_and_issue_overwrite(struct thin_c
*tc
, struct bio
*bio
,
1159 dm_block_t data_begin
,
1160 struct dm_thin_new_mapping
*m
)
1162 struct pool
*pool
= tc
->pool
;
1163 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1165 h
->overwrite_mapping
= m
;
1167 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
1168 inc_all_io_entry(pool
, bio
);
1169 remap_and_issue(tc
, bio
, data_begin
);
1173 * A partial copy also needs to zero the uncopied region.
1175 static void schedule_copy(struct thin_c
*tc
, dm_block_t virt_block
,
1176 struct dm_dev
*origin
, dm_block_t data_origin
,
1177 dm_block_t data_dest
,
1178 struct dm_bio_prison_cell
*cell
, struct bio
*bio
,
1182 struct pool
*pool
= tc
->pool
;
1183 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
1186 m
->virt_begin
= virt_block
;
1187 m
->virt_end
= virt_block
+ 1u;
1188 m
->data_block
= data_dest
;
1192 * quiesce action + copy action + an extra reference held for the
1193 * duration of this function (we may need to inc later for a
1196 atomic_set(&m
->prepare_actions
, 3);
1198 if (!dm_deferred_set_add_work(pool
->shared_read_ds
, &m
->list
))
1199 complete_mapping_preparation(m
); /* already quiesced */
1202 * IO to pool_dev remaps to the pool target's data_dev.
1204 * If the whole block of data is being overwritten, we can issue the
1205 * bio immediately. Otherwise we use kcopyd to clone the data first.
1207 if (io_overwrites_block(pool
, bio
))
1208 remap_and_issue_overwrite(tc
, bio
, data_dest
, m
);
1210 struct dm_io_region from
, to
;
1212 from
.bdev
= origin
->bdev
;
1213 from
.sector
= data_origin
* pool
->sectors_per_block
;
1216 to
.bdev
= tc
->pool_dev
->bdev
;
1217 to
.sector
= data_dest
* pool
->sectors_per_block
;
1220 r
= dm_kcopyd_copy(pool
->copier
, &from
, 1, &to
,
1221 0, copy_complete
, m
);
1223 DMERR_LIMIT("dm_kcopyd_copy() failed");
1224 copy_complete(1, 1, m
);
1227 * We allow the zero to be issued, to simplify the
1228 * error path. Otherwise we'd need to start
1229 * worrying about decrementing the prepare_actions
1235 * Do we need to zero a tail region?
1237 if (len
< pool
->sectors_per_block
&& pool
->pf
.zero_new_blocks
) {
1238 atomic_inc(&m
->prepare_actions
);
1240 data_dest
* pool
->sectors_per_block
+ len
,
1241 (data_dest
+ 1) * pool
->sectors_per_block
);
1245 complete_mapping_preparation(m
); /* drop our ref */
1248 static void schedule_internal_copy(struct thin_c
*tc
, dm_block_t virt_block
,
1249 dm_block_t data_origin
, dm_block_t data_dest
,
1250 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
1252 schedule_copy(tc
, virt_block
, tc
->pool_dev
,
1253 data_origin
, data_dest
, cell
, bio
,
1254 tc
->pool
->sectors_per_block
);
1257 static void schedule_zero(struct thin_c
*tc
, dm_block_t virt_block
,
1258 dm_block_t data_block
, struct dm_bio_prison_cell
*cell
,
1261 struct pool
*pool
= tc
->pool
;
1262 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
1264 atomic_set(&m
->prepare_actions
, 1); /* no need to quiesce */
1266 m
->virt_begin
= virt_block
;
1267 m
->virt_end
= virt_block
+ 1u;
1268 m
->data_block
= data_block
;
1272 * If the whole block of data is being overwritten or we are not
1273 * zeroing pre-existing data, we can issue the bio immediately.
1274 * Otherwise we use kcopyd to zero the data first.
1276 if (pool
->pf
.zero_new_blocks
) {
1277 if (io_overwrites_block(pool
, bio
))
1278 remap_and_issue_overwrite(tc
, bio
, data_block
, m
);
1280 ll_zero(tc
, m
, data_block
* pool
->sectors_per_block
,
1281 (data_block
+ 1) * pool
->sectors_per_block
);
1283 process_prepared_mapping(m
);
1286 static void schedule_external_copy(struct thin_c
*tc
, dm_block_t virt_block
,
1287 dm_block_t data_dest
,
1288 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
1290 struct pool
*pool
= tc
->pool
;
1291 sector_t virt_block_begin
= virt_block
* pool
->sectors_per_block
;
1292 sector_t virt_block_end
= (virt_block
+ 1) * pool
->sectors_per_block
;
1294 if (virt_block_end
<= tc
->origin_size
)
1295 schedule_copy(tc
, virt_block
, tc
->origin_dev
,
1296 virt_block
, data_dest
, cell
, bio
,
1297 pool
->sectors_per_block
);
1299 else if (virt_block_begin
< tc
->origin_size
)
1300 schedule_copy(tc
, virt_block
, tc
->origin_dev
,
1301 virt_block
, data_dest
, cell
, bio
,
1302 tc
->origin_size
- virt_block_begin
);
1305 schedule_zero(tc
, virt_block
, data_dest
, cell
, bio
);
1308 static void set_pool_mode(struct pool
*pool
, enum pool_mode new_mode
);
1310 static void check_for_space(struct pool
*pool
)
1315 if (get_pool_mode(pool
) != PM_OUT_OF_DATA_SPACE
)
1318 r
= dm_pool_get_free_block_count(pool
->pmd
, &nr_free
);
1323 set_pool_mode(pool
, PM_WRITE
);
1327 * A non-zero return indicates read_only or fail_io mode.
1328 * Many callers don't care about the return value.
1330 static int commit(struct pool
*pool
)
1334 if (get_pool_mode(pool
) >= PM_READ_ONLY
)
1337 r
= dm_pool_commit_metadata(pool
->pmd
);
1339 metadata_operation_failed(pool
, "dm_pool_commit_metadata", r
);
1341 check_for_space(pool
);
1346 static void check_low_water_mark(struct pool
*pool
, dm_block_t free_blocks
)
1348 unsigned long flags
;
1350 if (free_blocks
<= pool
->low_water_blocks
&& !pool
->low_water_triggered
) {
1351 DMWARN("%s: reached low water mark for data device: sending event.",
1352 dm_device_name(pool
->pool_md
));
1353 spin_lock_irqsave(&pool
->lock
, flags
);
1354 pool
->low_water_triggered
= true;
1355 spin_unlock_irqrestore(&pool
->lock
, flags
);
1356 dm_table_event(pool
->ti
->table
);
1360 static int alloc_data_block(struct thin_c
*tc
, dm_block_t
*result
)
1363 dm_block_t free_blocks
;
1364 struct pool
*pool
= tc
->pool
;
1366 if (WARN_ON(get_pool_mode(pool
) != PM_WRITE
))
1369 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
1371 metadata_operation_failed(pool
, "dm_pool_get_free_block_count", r
);
1375 check_low_water_mark(pool
, free_blocks
);
1379 * Try to commit to see if that will free up some
1386 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
1388 metadata_operation_failed(pool
, "dm_pool_get_free_block_count", r
);
1393 set_pool_mode(pool
, PM_OUT_OF_DATA_SPACE
);
1398 r
= dm_pool_alloc_data_block(pool
->pmd
, result
);
1400 metadata_operation_failed(pool
, "dm_pool_alloc_data_block", r
);
1408 * If we have run out of space, queue bios until the device is
1409 * resumed, presumably after having been reloaded with more space.
1411 static void retry_on_resume(struct bio
*bio
)
1413 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1414 struct thin_c
*tc
= h
->tc
;
1415 unsigned long flags
;
1417 spin_lock_irqsave(&tc
->lock
, flags
);
1418 bio_list_add(&tc
->retry_on_resume_list
, bio
);
1419 spin_unlock_irqrestore(&tc
->lock
, flags
);
1422 static int should_error_unserviceable_bio(struct pool
*pool
)
1424 enum pool_mode m
= get_pool_mode(pool
);
1428 /* Shouldn't get here */
1429 DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
1432 case PM_OUT_OF_DATA_SPACE
:
1433 return pool
->pf
.error_if_no_space
? -ENOSPC
: 0;
1439 /* Shouldn't get here */
1440 DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1445 static void handle_unserviceable_bio(struct pool
*pool
, struct bio
*bio
)
1447 int error
= should_error_unserviceable_bio(pool
);
1450 bio
->bi_error
= error
;
1453 retry_on_resume(bio
);
1456 static void retry_bios_on_resume(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
1459 struct bio_list bios
;
1462 error
= should_error_unserviceable_bio(pool
);
1464 cell_error_with_code(pool
, cell
, error
);
1468 bio_list_init(&bios
);
1469 cell_release(pool
, cell
, &bios
);
1471 while ((bio
= bio_list_pop(&bios
)))
1472 retry_on_resume(bio
);
1475 static void process_discard_cell_no_passdown(struct thin_c
*tc
,
1476 struct dm_bio_prison_cell
*virt_cell
)
1478 struct pool
*pool
= tc
->pool
;
1479 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
1482 * We don't need to lock the data blocks, since there's no
1483 * passdown. We only lock data blocks for allocation and breaking sharing.
1486 m
->virt_begin
= virt_cell
->key
.block_begin
;
1487 m
->virt_end
= virt_cell
->key
.block_end
;
1488 m
->cell
= virt_cell
;
1489 m
->bio
= virt_cell
->holder
;
1491 if (!dm_deferred_set_add_work(pool
->all_io_ds
, &m
->list
))
1492 pool
->process_prepared_discard(m
);
1495 static void break_up_discard_bio(struct thin_c
*tc
, dm_block_t begin
, dm_block_t end
,
1498 struct pool
*pool
= tc
->pool
;
1502 struct dm_cell_key data_key
;
1503 struct dm_bio_prison_cell
*data_cell
;
1504 struct dm_thin_new_mapping
*m
;
1505 dm_block_t virt_begin
, virt_end
, data_begin
;
1507 while (begin
!= end
) {
1508 r
= ensure_next_mapping(pool
);
1510 /* we did our best */
1513 r
= dm_thin_find_mapped_range(tc
->td
, begin
, end
, &virt_begin
, &virt_end
,
1514 &data_begin
, &maybe_shared
);
1517 * Silently fail, letting any mappings we've
1522 build_key(tc
->td
, PHYSICAL
, data_begin
, data_begin
+ (virt_end
- virt_begin
), &data_key
);
1523 if (bio_detain(tc
->pool
, &data_key
, NULL
, &data_cell
)) {
1524 /* contention, we'll give up with this range */
1530 * IO may still be going to the destination block. We must
1531 * quiesce before we can do the removal.
1533 m
= get_next_mapping(pool
);
1535 m
->maybe_shared
= maybe_shared
;
1536 m
->virt_begin
= virt_begin
;
1537 m
->virt_end
= virt_end
;
1538 m
->data_block
= data_begin
;
1539 m
->cell
= data_cell
;
1543 * The parent bio must not complete before sub discard bios are
1544 * chained to it (see end_discard's bio_chain)!
1546 * This per-mapping bi_remaining increment is paired with
1547 * the implicit decrement that occurs via bio_endio() in
1550 bio_inc_remaining(bio
);
1551 if (!dm_deferred_set_add_work(pool
->all_io_ds
, &m
->list
))
1552 pool
->process_prepared_discard(m
);
1558 static void process_discard_cell_passdown(struct thin_c
*tc
, struct dm_bio_prison_cell
*virt_cell
)
1560 struct bio
*bio
= virt_cell
->holder
;
1561 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1564 * The virt_cell will only get freed once the origin bio completes.
1565 * This means it will remain locked while all the individual
1566 * passdown bios are in flight.
1568 h
->cell
= virt_cell
;
1569 break_up_discard_bio(tc
, virt_cell
->key
.block_begin
, virt_cell
->key
.block_end
, bio
);
1572 * We complete the bio now, knowing that the bi_remaining field
1573 * will prevent completion until the sub range discards have
1579 static void process_discard_bio(struct thin_c
*tc
, struct bio
*bio
)
1581 dm_block_t begin
, end
;
1582 struct dm_cell_key virt_key
;
1583 struct dm_bio_prison_cell
*virt_cell
;
1585 get_bio_block_range(tc
, bio
, &begin
, &end
);
1588 * The discard covers less than a block.
1594 build_key(tc
->td
, VIRTUAL
, begin
, end
, &virt_key
);
1595 if (bio_detain(tc
->pool
, &virt_key
, bio
, &virt_cell
))
1597 * Potential starvation issue: We're relying on the
1598 * fs/application being well behaved, and not trying to
1599 * send IO to a region at the same time as discarding it.
1600 * If they do this persistently then it's possible this
1601 * cell will never be granted.
1605 tc
->pool
->process_discard_cell(tc
, virt_cell
);
1608 static void break_sharing(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1609 struct dm_cell_key
*key
,
1610 struct dm_thin_lookup_result
*lookup_result
,
1611 struct dm_bio_prison_cell
*cell
)
1614 dm_block_t data_block
;
1615 struct pool
*pool
= tc
->pool
;
1617 r
= alloc_data_block(tc
, &data_block
);
1620 schedule_internal_copy(tc
, block
, lookup_result
->block
,
1621 data_block
, cell
, bio
);
1625 retry_bios_on_resume(pool
, cell
);
1629 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1631 cell_error(pool
, cell
);
1636 static void __remap_and_issue_shared_cell(void *context
,
1637 struct dm_bio_prison_cell
*cell
)
1639 struct remap_info
*info
= context
;
1642 while ((bio
= bio_list_pop(&cell
->bios
))) {
1643 if ((bio_data_dir(bio
) == WRITE
) ||
1644 (bio
->bi_rw
& (REQ_PREFLUSH
| REQ_FUA
) ||
1645 bio_op(bio
) == REQ_OP_DISCARD
))
1646 bio_list_add(&info
->defer_bios
, bio
);
1648 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));;
1650 h
->shared_read_entry
= dm_deferred_entry_inc(info
->tc
->pool
->shared_read_ds
);
1651 inc_all_io_entry(info
->tc
->pool
, bio
);
1652 bio_list_add(&info
->issue_bios
, bio
);
1657 static void remap_and_issue_shared_cell(struct thin_c
*tc
,
1658 struct dm_bio_prison_cell
*cell
,
1662 struct remap_info info
;
1665 bio_list_init(&info
.defer_bios
);
1666 bio_list_init(&info
.issue_bios
);
1668 cell_visit_release(tc
->pool
, __remap_and_issue_shared_cell
,
1671 while ((bio
= bio_list_pop(&info
.defer_bios
)))
1672 thin_defer_bio(tc
, bio
);
1674 while ((bio
= bio_list_pop(&info
.issue_bios
)))
1675 remap_and_issue(tc
, bio
, block
);
1678 static void process_shared_bio(struct thin_c
*tc
, struct bio
*bio
,
1680 struct dm_thin_lookup_result
*lookup_result
,
1681 struct dm_bio_prison_cell
*virt_cell
)
1683 struct dm_bio_prison_cell
*data_cell
;
1684 struct pool
*pool
= tc
->pool
;
1685 struct dm_cell_key key
;
1688 * If cell is already occupied, then sharing is already in the process
1689 * of being broken so we have nothing further to do here.
1691 build_data_key(tc
->td
, lookup_result
->block
, &key
);
1692 if (bio_detain(pool
, &key
, bio
, &data_cell
)) {
1693 cell_defer_no_holder(tc
, virt_cell
);
1697 if (bio_data_dir(bio
) == WRITE
&& bio
->bi_iter
.bi_size
) {
1698 break_sharing(tc
, bio
, block
, &key
, lookup_result
, data_cell
);
1699 cell_defer_no_holder(tc
, virt_cell
);
1701 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1703 h
->shared_read_entry
= dm_deferred_entry_inc(pool
->shared_read_ds
);
1704 inc_all_io_entry(pool
, bio
);
1705 remap_and_issue(tc
, bio
, lookup_result
->block
);
1707 remap_and_issue_shared_cell(tc
, data_cell
, lookup_result
->block
);
1708 remap_and_issue_shared_cell(tc
, virt_cell
, lookup_result
->block
);
1712 static void provision_block(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1713 struct dm_bio_prison_cell
*cell
)
1716 dm_block_t data_block
;
1717 struct pool
*pool
= tc
->pool
;
1720 * Remap empty bios (flushes) immediately, without provisioning.
1722 if (!bio
->bi_iter
.bi_size
) {
1723 inc_all_io_entry(pool
, bio
);
1724 cell_defer_no_holder(tc
, cell
);
1726 remap_and_issue(tc
, bio
, 0);
1731 * Fill read bios with zeroes and complete them immediately.
1733 if (bio_data_dir(bio
) == READ
) {
1735 cell_defer_no_holder(tc
, cell
);
1740 r
= alloc_data_block(tc
, &data_block
);
1744 schedule_external_copy(tc
, block
, data_block
, cell
, bio
);
1746 schedule_zero(tc
, block
, data_block
, cell
, bio
);
1750 retry_bios_on_resume(pool
, cell
);
1754 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1756 cell_error(pool
, cell
);
1761 static void process_cell(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
1764 struct pool
*pool
= tc
->pool
;
1765 struct bio
*bio
= cell
->holder
;
1766 dm_block_t block
= get_bio_block(tc
, bio
);
1767 struct dm_thin_lookup_result lookup_result
;
1769 if (tc
->requeue_mode
) {
1770 cell_requeue(pool
, cell
);
1774 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1777 if (lookup_result
.shared
)
1778 process_shared_bio(tc
, bio
, block
, &lookup_result
, cell
);
1780 inc_all_io_entry(pool
, bio
);
1781 remap_and_issue(tc
, bio
, lookup_result
.block
);
1782 inc_remap_and_issue_cell(tc
, cell
, lookup_result
.block
);
1787 if (bio_data_dir(bio
) == READ
&& tc
->origin_dev
) {
1788 inc_all_io_entry(pool
, bio
);
1789 cell_defer_no_holder(tc
, cell
);
1791 if (bio_end_sector(bio
) <= tc
->origin_size
)
1792 remap_to_origin_and_issue(tc
, bio
);
1794 else if (bio
->bi_iter
.bi_sector
< tc
->origin_size
) {
1796 bio
->bi_iter
.bi_size
= (tc
->origin_size
- bio
->bi_iter
.bi_sector
) << SECTOR_SHIFT
;
1797 remap_to_origin_and_issue(tc
, bio
);
1804 provision_block(tc
, bio
, block
, cell
);
1808 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1810 cell_defer_no_holder(tc
, cell
);
1816 static void process_bio(struct thin_c
*tc
, struct bio
*bio
)
1818 struct pool
*pool
= tc
->pool
;
1819 dm_block_t block
= get_bio_block(tc
, bio
);
1820 struct dm_bio_prison_cell
*cell
;
1821 struct dm_cell_key key
;
1824 * If cell is already occupied, then the block is already
1825 * being provisioned so we have nothing further to do here.
1827 build_virtual_key(tc
->td
, block
, &key
);
1828 if (bio_detain(pool
, &key
, bio
, &cell
))
1831 process_cell(tc
, cell
);
1834 static void __process_bio_read_only(struct thin_c
*tc
, struct bio
*bio
,
1835 struct dm_bio_prison_cell
*cell
)
1838 int rw
= bio_data_dir(bio
);
1839 dm_block_t block
= get_bio_block(tc
, bio
);
1840 struct dm_thin_lookup_result lookup_result
;
1842 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1845 if (lookup_result
.shared
&& (rw
== WRITE
) && bio
->bi_iter
.bi_size
) {
1846 handle_unserviceable_bio(tc
->pool
, bio
);
1848 cell_defer_no_holder(tc
, cell
);
1850 inc_all_io_entry(tc
->pool
, bio
);
1851 remap_and_issue(tc
, bio
, lookup_result
.block
);
1853 inc_remap_and_issue_cell(tc
, cell
, lookup_result
.block
);
1859 cell_defer_no_holder(tc
, cell
);
1861 handle_unserviceable_bio(tc
->pool
, bio
);
1865 if (tc
->origin_dev
) {
1866 inc_all_io_entry(tc
->pool
, bio
);
1867 remap_to_origin_and_issue(tc
, bio
);
1876 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1879 cell_defer_no_holder(tc
, cell
);
1885 static void process_bio_read_only(struct thin_c
*tc
, struct bio
*bio
)
1887 __process_bio_read_only(tc
, bio
, NULL
);
1890 static void process_cell_read_only(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
1892 __process_bio_read_only(tc
, cell
->holder
, cell
);
1895 static void process_bio_success(struct thin_c
*tc
, struct bio
*bio
)
1900 static void process_bio_fail(struct thin_c
*tc
, struct bio
*bio
)
1905 static void process_cell_success(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
1907 cell_success(tc
->pool
, cell
);
1910 static void process_cell_fail(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
1912 cell_error(tc
->pool
, cell
);
1916 * FIXME: should we also commit due to size of transaction, measured in
1919 static int need_commit_due_to_time(struct pool
*pool
)
1921 return !time_in_range(jiffies
, pool
->last_commit_jiffies
,
1922 pool
->last_commit_jiffies
+ COMMIT_PERIOD
);
1925 #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node)
1926 #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook))
1928 static void __thin_bio_rb_add(struct thin_c
*tc
, struct bio
*bio
)
1930 struct rb_node
**rbp
, *parent
;
1931 struct dm_thin_endio_hook
*pbd
;
1932 sector_t bi_sector
= bio
->bi_iter
.bi_sector
;
1934 rbp
= &tc
->sort_bio_list
.rb_node
;
1938 pbd
= thin_pbd(parent
);
1940 if (bi_sector
< thin_bio(pbd
)->bi_iter
.bi_sector
)
1941 rbp
= &(*rbp
)->rb_left
;
1943 rbp
= &(*rbp
)->rb_right
;
1946 pbd
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1947 rb_link_node(&pbd
->rb_node
, parent
, rbp
);
1948 rb_insert_color(&pbd
->rb_node
, &tc
->sort_bio_list
);
1951 static void __extract_sorted_bios(struct thin_c
*tc
)
1953 struct rb_node
*node
;
1954 struct dm_thin_endio_hook
*pbd
;
1957 for (node
= rb_first(&tc
->sort_bio_list
); node
; node
= rb_next(node
)) {
1958 pbd
= thin_pbd(node
);
1959 bio
= thin_bio(pbd
);
1961 bio_list_add(&tc
->deferred_bio_list
, bio
);
1962 rb_erase(&pbd
->rb_node
, &tc
->sort_bio_list
);
1965 WARN_ON(!RB_EMPTY_ROOT(&tc
->sort_bio_list
));
1968 static void __sort_thin_deferred_bios(struct thin_c
*tc
)
1971 struct bio_list bios
;
1973 bio_list_init(&bios
);
1974 bio_list_merge(&bios
, &tc
->deferred_bio_list
);
1975 bio_list_init(&tc
->deferred_bio_list
);
1977 /* Sort deferred_bio_list using rb-tree */
1978 while ((bio
= bio_list_pop(&bios
)))
1979 __thin_bio_rb_add(tc
, bio
);
1982 * Transfer the sorted bios in sort_bio_list back to
1983 * deferred_bio_list to allow lockless submission of
1986 __extract_sorted_bios(tc
);
1989 static void process_thin_deferred_bios(struct thin_c
*tc
)
1991 struct pool
*pool
= tc
->pool
;
1992 unsigned long flags
;
1994 struct bio_list bios
;
1995 struct blk_plug plug
;
1998 if (tc
->requeue_mode
) {
1999 error_thin_bio_list(tc
, &tc
->deferred_bio_list
, DM_ENDIO_REQUEUE
);
2003 bio_list_init(&bios
);
2005 spin_lock_irqsave(&tc
->lock
, flags
);
2007 if (bio_list_empty(&tc
->deferred_bio_list
)) {
2008 spin_unlock_irqrestore(&tc
->lock
, flags
);
2012 __sort_thin_deferred_bios(tc
);
2014 bio_list_merge(&bios
, &tc
->deferred_bio_list
);
2015 bio_list_init(&tc
->deferred_bio_list
);
2017 spin_unlock_irqrestore(&tc
->lock
, flags
);
2019 blk_start_plug(&plug
);
2020 while ((bio
= bio_list_pop(&bios
))) {
2022 * If we've got no free new_mapping structs, and processing
2023 * this bio might require one, we pause until there are some
2024 * prepared mappings to process.
2026 if (ensure_next_mapping(pool
)) {
2027 spin_lock_irqsave(&tc
->lock
, flags
);
2028 bio_list_add(&tc
->deferred_bio_list
, bio
);
2029 bio_list_merge(&tc
->deferred_bio_list
, &bios
);
2030 spin_unlock_irqrestore(&tc
->lock
, flags
);
2034 if (bio_op(bio
) == REQ_OP_DISCARD
)
2035 pool
->process_discard(tc
, bio
);
2037 pool
->process_bio(tc
, bio
);
2039 if ((count
++ & 127) == 0) {
2040 throttle_work_update(&pool
->throttle
);
2041 dm_pool_issue_prefetches(pool
->pmd
);
2044 blk_finish_plug(&plug
);
2047 static int cmp_cells(const void *lhs
, const void *rhs
)
2049 struct dm_bio_prison_cell
*lhs_cell
= *((struct dm_bio_prison_cell
**) lhs
);
2050 struct dm_bio_prison_cell
*rhs_cell
= *((struct dm_bio_prison_cell
**) rhs
);
2052 BUG_ON(!lhs_cell
->holder
);
2053 BUG_ON(!rhs_cell
->holder
);
2055 if (lhs_cell
->holder
->bi_iter
.bi_sector
< rhs_cell
->holder
->bi_iter
.bi_sector
)
2058 if (lhs_cell
->holder
->bi_iter
.bi_sector
> rhs_cell
->holder
->bi_iter
.bi_sector
)
2064 static unsigned sort_cells(struct pool
*pool
, struct list_head
*cells
)
2067 struct dm_bio_prison_cell
*cell
, *tmp
;
2069 list_for_each_entry_safe(cell
, tmp
, cells
, user_list
) {
2070 if (count
>= CELL_SORT_ARRAY_SIZE
)
2073 pool
->cell_sort_array
[count
++] = cell
;
2074 list_del(&cell
->user_list
);
2077 sort(pool
->cell_sort_array
, count
, sizeof(cell
), cmp_cells
, NULL
);
2082 static void process_thin_deferred_cells(struct thin_c
*tc
)
2084 struct pool
*pool
= tc
->pool
;
2085 unsigned long flags
;
2086 struct list_head cells
;
2087 struct dm_bio_prison_cell
*cell
;
2088 unsigned i
, j
, count
;
2090 INIT_LIST_HEAD(&cells
);
2092 spin_lock_irqsave(&tc
->lock
, flags
);
2093 list_splice_init(&tc
->deferred_cells
, &cells
);
2094 spin_unlock_irqrestore(&tc
->lock
, flags
);
2096 if (list_empty(&cells
))
2100 count
= sort_cells(tc
->pool
, &cells
);
2102 for (i
= 0; i
< count
; i
++) {
2103 cell
= pool
->cell_sort_array
[i
];
2104 BUG_ON(!cell
->holder
);
2107 * If we've got no free new_mapping structs, and processing
2108 * this bio might require one, we pause until there are some
2109 * prepared mappings to process.
2111 if (ensure_next_mapping(pool
)) {
2112 for (j
= i
; j
< count
; j
++)
2113 list_add(&pool
->cell_sort_array
[j
]->user_list
, &cells
);
2115 spin_lock_irqsave(&tc
->lock
, flags
);
2116 list_splice(&cells
, &tc
->deferred_cells
);
2117 spin_unlock_irqrestore(&tc
->lock
, flags
);
2121 if (bio_op(cell
->holder
) == REQ_OP_DISCARD
)
2122 pool
->process_discard_cell(tc
, cell
);
2124 pool
->process_cell(tc
, cell
);
2126 } while (!list_empty(&cells
));
2129 static void thin_get(struct thin_c
*tc
);
2130 static void thin_put(struct thin_c
*tc
);
2133 * We can't hold rcu_read_lock() around code that can block. So we
2134 * find a thin with the rcu lock held; bump a refcount; then drop
2137 static struct thin_c
*get_first_thin(struct pool
*pool
)
2139 struct thin_c
*tc
= NULL
;
2142 if (!list_empty(&pool
->active_thins
)) {
2143 tc
= list_entry_rcu(pool
->active_thins
.next
, struct thin_c
, list
);
2151 static struct thin_c
*get_next_thin(struct pool
*pool
, struct thin_c
*tc
)
2153 struct thin_c
*old_tc
= tc
;
2156 list_for_each_entry_continue_rcu(tc
, &pool
->active_thins
, list
) {
2168 static void process_deferred_bios(struct pool
*pool
)
2170 unsigned long flags
;
2172 struct bio_list bios
;
2175 tc
= get_first_thin(pool
);
2177 process_thin_deferred_cells(tc
);
2178 process_thin_deferred_bios(tc
);
2179 tc
= get_next_thin(pool
, tc
);
2183 * If there are any deferred flush bios, we must commit
2184 * the metadata before issuing them.
2186 bio_list_init(&bios
);
2187 spin_lock_irqsave(&pool
->lock
, flags
);
2188 bio_list_merge(&bios
, &pool
->deferred_flush_bios
);
2189 bio_list_init(&pool
->deferred_flush_bios
);
2190 spin_unlock_irqrestore(&pool
->lock
, flags
);
2192 if (bio_list_empty(&bios
) &&
2193 !(dm_pool_changed_this_transaction(pool
->pmd
) && need_commit_due_to_time(pool
)))
2197 while ((bio
= bio_list_pop(&bios
)))
2201 pool
->last_commit_jiffies
= jiffies
;
2203 while ((bio
= bio_list_pop(&bios
)))
2204 generic_make_request(bio
);
2207 static void do_worker(struct work_struct
*ws
)
2209 struct pool
*pool
= container_of(ws
, struct pool
, worker
);
2211 throttle_work_start(&pool
->throttle
);
2212 dm_pool_issue_prefetches(pool
->pmd
);
2213 throttle_work_update(&pool
->throttle
);
2214 process_prepared(pool
, &pool
->prepared_mappings
, &pool
->process_prepared_mapping
);
2215 throttle_work_update(&pool
->throttle
);
2216 process_prepared(pool
, &pool
->prepared_discards
, &pool
->process_prepared_discard
);
2217 throttle_work_update(&pool
->throttle
);
2218 process_deferred_bios(pool
);
2219 throttle_work_complete(&pool
->throttle
);
2223 * We want to commit periodically so that not too much
2224 * unwritten data builds up.
2226 static void do_waker(struct work_struct
*ws
)
2228 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
, waker
);
2230 queue_delayed_work(pool
->wq
, &pool
->waker
, COMMIT_PERIOD
);
2233 static void notify_of_pool_mode_change_to_oods(struct pool
*pool
);
2236 * We're holding onto IO to allow userland time to react. After the
2237 * timeout either the pool will have been resized (and thus back in
2238 * PM_WRITE mode), or we degrade to PM_OUT_OF_DATA_SPACE w/ error_if_no_space.
2240 static void do_no_space_timeout(struct work_struct
*ws
)
2242 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
,
2245 if (get_pool_mode(pool
) == PM_OUT_OF_DATA_SPACE
&& !pool
->pf
.error_if_no_space
) {
2246 pool
->pf
.error_if_no_space
= true;
2247 notify_of_pool_mode_change_to_oods(pool
);
2248 error_retry_list_with_code(pool
, -ENOSPC
);
2252 /*----------------------------------------------------------------*/
2255 struct work_struct worker
;
2256 struct completion complete
;
2259 static struct pool_work
*to_pool_work(struct work_struct
*ws
)
2261 return container_of(ws
, struct pool_work
, worker
);
2264 static void pool_work_complete(struct pool_work
*pw
)
2266 complete(&pw
->complete
);
2269 static void pool_work_wait(struct pool_work
*pw
, struct pool
*pool
,
2270 void (*fn
)(struct work_struct
*))
2272 INIT_WORK_ONSTACK(&pw
->worker
, fn
);
2273 init_completion(&pw
->complete
);
2274 queue_work(pool
->wq
, &pw
->worker
);
2275 wait_for_completion(&pw
->complete
);
2278 /*----------------------------------------------------------------*/
2280 struct noflush_work
{
2281 struct pool_work pw
;
2285 static struct noflush_work
*to_noflush(struct work_struct
*ws
)
2287 return container_of(to_pool_work(ws
), struct noflush_work
, pw
);
2290 static void do_noflush_start(struct work_struct
*ws
)
2292 struct noflush_work
*w
= to_noflush(ws
);
2293 w
->tc
->requeue_mode
= true;
2295 pool_work_complete(&w
->pw
);
2298 static void do_noflush_stop(struct work_struct
*ws
)
2300 struct noflush_work
*w
= to_noflush(ws
);
2301 w
->tc
->requeue_mode
= false;
2302 pool_work_complete(&w
->pw
);
2305 static void noflush_work(struct thin_c
*tc
, void (*fn
)(struct work_struct
*))
2307 struct noflush_work w
;
2310 pool_work_wait(&w
.pw
, tc
->pool
, fn
);
2313 /*----------------------------------------------------------------*/
2315 static enum pool_mode
get_pool_mode(struct pool
*pool
)
2317 return pool
->pf
.mode
;
2320 static void notify_of_pool_mode_change(struct pool
*pool
, const char *new_mode
)
2322 dm_table_event(pool
->ti
->table
);
2323 DMINFO("%s: switching pool to %s mode",
2324 dm_device_name(pool
->pool_md
), new_mode
);
2327 static void notify_of_pool_mode_change_to_oods(struct pool
*pool
)
2329 if (!pool
->pf
.error_if_no_space
)
2330 notify_of_pool_mode_change(pool
, "out-of-data-space (queue IO)");
2332 notify_of_pool_mode_change(pool
, "out-of-data-space (error IO)");
2335 static bool passdown_enabled(struct pool_c
*pt
)
2337 return pt
->adjusted_pf
.discard_passdown
;
2340 static void set_discard_callbacks(struct pool
*pool
)
2342 struct pool_c
*pt
= pool
->ti
->private;
2344 if (passdown_enabled(pt
)) {
2345 pool
->process_discard_cell
= process_discard_cell_passdown
;
2346 pool
->process_prepared_discard
= process_prepared_discard_passdown
;
2348 pool
->process_discard_cell
= process_discard_cell_no_passdown
;
2349 pool
->process_prepared_discard
= process_prepared_discard_no_passdown
;
2353 static void set_pool_mode(struct pool
*pool
, enum pool_mode new_mode
)
2355 struct pool_c
*pt
= pool
->ti
->private;
2356 bool needs_check
= dm_pool_metadata_needs_check(pool
->pmd
);
2357 enum pool_mode old_mode
= get_pool_mode(pool
);
2358 unsigned long no_space_timeout
= ACCESS_ONCE(no_space_timeout_secs
) * HZ
;
2361 * Never allow the pool to transition to PM_WRITE mode if user
2362 * intervention is required to verify metadata and data consistency.
2364 if (new_mode
== PM_WRITE
&& needs_check
) {
2365 DMERR("%s: unable to switch pool to write mode until repaired.",
2366 dm_device_name(pool
->pool_md
));
2367 if (old_mode
!= new_mode
)
2368 new_mode
= old_mode
;
2370 new_mode
= PM_READ_ONLY
;
2373 * If we were in PM_FAIL mode, rollback of metadata failed. We're
2374 * not going to recover without a thin_repair. So we never let the
2375 * pool move out of the old mode.
2377 if (old_mode
== PM_FAIL
)
2378 new_mode
= old_mode
;
2382 if (old_mode
!= new_mode
)
2383 notify_of_pool_mode_change(pool
, "failure");
2384 dm_pool_metadata_read_only(pool
->pmd
);
2385 pool
->process_bio
= process_bio_fail
;
2386 pool
->process_discard
= process_bio_fail
;
2387 pool
->process_cell
= process_cell_fail
;
2388 pool
->process_discard_cell
= process_cell_fail
;
2389 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
2390 pool
->process_prepared_discard
= process_prepared_discard_fail
;
2392 error_retry_list(pool
);
2396 if (old_mode
!= new_mode
)
2397 notify_of_pool_mode_change(pool
, "read-only");
2398 dm_pool_metadata_read_only(pool
->pmd
);
2399 pool
->process_bio
= process_bio_read_only
;
2400 pool
->process_discard
= process_bio_success
;
2401 pool
->process_cell
= process_cell_read_only
;
2402 pool
->process_discard_cell
= process_cell_success
;
2403 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
2404 pool
->process_prepared_discard
= process_prepared_discard_success
;
2406 error_retry_list(pool
);
2409 case PM_OUT_OF_DATA_SPACE
:
2411 * Ideally we'd never hit this state; the low water mark
2412 * would trigger userland to extend the pool before we
2413 * completely run out of data space. However, many small
2414 * IOs to unprovisioned space can consume data space at an
2415 * alarming rate. Adjust your low water mark if you're
2416 * frequently seeing this mode.
2418 if (old_mode
!= new_mode
)
2419 notify_of_pool_mode_change_to_oods(pool
);
2420 pool
->out_of_data_space
= true;
2421 pool
->process_bio
= process_bio_read_only
;
2422 pool
->process_discard
= process_discard_bio
;
2423 pool
->process_cell
= process_cell_read_only
;
2424 pool
->process_prepared_mapping
= process_prepared_mapping
;
2425 set_discard_callbacks(pool
);
2427 if (!pool
->pf
.error_if_no_space
&& no_space_timeout
)
2428 queue_delayed_work(pool
->wq
, &pool
->no_space_timeout
, no_space_timeout
);
2432 if (old_mode
!= new_mode
)
2433 notify_of_pool_mode_change(pool
, "write");
2434 pool
->out_of_data_space
= false;
2435 pool
->pf
.error_if_no_space
= pt
->requested_pf
.error_if_no_space
;
2436 dm_pool_metadata_read_write(pool
->pmd
);
2437 pool
->process_bio
= process_bio
;
2438 pool
->process_discard
= process_discard_bio
;
2439 pool
->process_cell
= process_cell
;
2440 pool
->process_prepared_mapping
= process_prepared_mapping
;
2441 set_discard_callbacks(pool
);
2445 pool
->pf
.mode
= new_mode
;
2447 * The pool mode may have changed, sync it so bind_control_target()
2448 * doesn't cause an unexpected mode transition on resume.
2450 pt
->adjusted_pf
.mode
= new_mode
;
2453 static void abort_transaction(struct pool
*pool
)
2455 const char *dev_name
= dm_device_name(pool
->pool_md
);
2457 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name
);
2458 if (dm_pool_abort_metadata(pool
->pmd
)) {
2459 DMERR("%s: failed to abort metadata transaction", dev_name
);
2460 set_pool_mode(pool
, PM_FAIL
);
2463 if (dm_pool_metadata_set_needs_check(pool
->pmd
)) {
2464 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name
);
2465 set_pool_mode(pool
, PM_FAIL
);
2469 static void metadata_operation_failed(struct pool
*pool
, const char *op
, int r
)
2471 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
2472 dm_device_name(pool
->pool_md
), op
, r
);
2474 abort_transaction(pool
);
2475 set_pool_mode(pool
, PM_READ_ONLY
);
2478 /*----------------------------------------------------------------*/
2481 * Mapping functions.
2485 * Called only while mapping a thin bio to hand it over to the workqueue.
2487 static void thin_defer_bio(struct thin_c
*tc
, struct bio
*bio
)
2489 unsigned long flags
;
2490 struct pool
*pool
= tc
->pool
;
2492 spin_lock_irqsave(&tc
->lock
, flags
);
2493 bio_list_add(&tc
->deferred_bio_list
, bio
);
2494 spin_unlock_irqrestore(&tc
->lock
, flags
);
2499 static void thin_defer_bio_with_throttle(struct thin_c
*tc
, struct bio
*bio
)
2501 struct pool
*pool
= tc
->pool
;
2503 throttle_lock(&pool
->throttle
);
2504 thin_defer_bio(tc
, bio
);
2505 throttle_unlock(&pool
->throttle
);
2508 static void thin_defer_cell(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
2510 unsigned long flags
;
2511 struct pool
*pool
= tc
->pool
;
2513 throttle_lock(&pool
->throttle
);
2514 spin_lock_irqsave(&tc
->lock
, flags
);
2515 list_add_tail(&cell
->user_list
, &tc
->deferred_cells
);
2516 spin_unlock_irqrestore(&tc
->lock
, flags
);
2517 throttle_unlock(&pool
->throttle
);
2522 static void thin_hook_bio(struct thin_c
*tc
, struct bio
*bio
)
2524 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
2527 h
->shared_read_entry
= NULL
;
2528 h
->all_io_entry
= NULL
;
2529 h
->overwrite_mapping
= NULL
;
2534 * Non-blocking function called from the thin target's map function.
2536 static int thin_bio_map(struct dm_target
*ti
, struct bio
*bio
)
2539 struct thin_c
*tc
= ti
->private;
2540 dm_block_t block
= get_bio_block(tc
, bio
);
2541 struct dm_thin_device
*td
= tc
->td
;
2542 struct dm_thin_lookup_result result
;
2543 struct dm_bio_prison_cell
*virt_cell
, *data_cell
;
2544 struct dm_cell_key key
;
2546 thin_hook_bio(tc
, bio
);
2548 if (tc
->requeue_mode
) {
2549 bio
->bi_error
= DM_ENDIO_REQUEUE
;
2551 return DM_MAPIO_SUBMITTED
;
2554 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
2556 return DM_MAPIO_SUBMITTED
;
2559 if (bio
->bi_rw
& (REQ_PREFLUSH
| REQ_FUA
) ||
2560 bio_op(bio
) == REQ_OP_DISCARD
) {
2561 thin_defer_bio_with_throttle(tc
, bio
);
2562 return DM_MAPIO_SUBMITTED
;
2566 * We must hold the virtual cell before doing the lookup, otherwise
2567 * there's a race with discard.
2569 build_virtual_key(tc
->td
, block
, &key
);
2570 if (bio_detain(tc
->pool
, &key
, bio
, &virt_cell
))
2571 return DM_MAPIO_SUBMITTED
;
2573 r
= dm_thin_find_block(td
, block
, 0, &result
);
2576 * Note that we defer readahead too.
2580 if (unlikely(result
.shared
)) {
2582 * We have a race condition here between the
2583 * result.shared value returned by the lookup and
2584 * snapshot creation, which may cause new
2587 * To avoid this always quiesce the origin before
2588 * taking the snap. You want to do this anyway to
2589 * ensure a consistent application view
2592 * More distant ancestors are irrelevant. The
2593 * shared flag will be set in their case.
2595 thin_defer_cell(tc
, virt_cell
);
2596 return DM_MAPIO_SUBMITTED
;
2599 build_data_key(tc
->td
, result
.block
, &key
);
2600 if (bio_detain(tc
->pool
, &key
, bio
, &data_cell
)) {
2601 cell_defer_no_holder(tc
, virt_cell
);
2602 return DM_MAPIO_SUBMITTED
;
2605 inc_all_io_entry(tc
->pool
, bio
);
2606 cell_defer_no_holder(tc
, data_cell
);
2607 cell_defer_no_holder(tc
, virt_cell
);
2609 remap(tc
, bio
, result
.block
);
2610 return DM_MAPIO_REMAPPED
;
2614 thin_defer_cell(tc
, virt_cell
);
2615 return DM_MAPIO_SUBMITTED
;
2619 * Must always call bio_io_error on failure.
2620 * dm_thin_find_block can fail with -EINVAL if the
2621 * pool is switched to fail-io mode.
2624 cell_defer_no_holder(tc
, virt_cell
);
2625 return DM_MAPIO_SUBMITTED
;
2629 static int pool_is_congested(struct dm_target_callbacks
*cb
, int bdi_bits
)
2631 struct pool_c
*pt
= container_of(cb
, struct pool_c
, callbacks
);
2632 struct request_queue
*q
;
2634 if (get_pool_mode(pt
->pool
) == PM_OUT_OF_DATA_SPACE
)
2637 q
= bdev_get_queue(pt
->data_dev
->bdev
);
2638 return bdi_congested(&q
->backing_dev_info
, bdi_bits
);
2641 static void requeue_bios(struct pool
*pool
)
2643 unsigned long flags
;
2647 list_for_each_entry_rcu(tc
, &pool
->active_thins
, list
) {
2648 spin_lock_irqsave(&tc
->lock
, flags
);
2649 bio_list_merge(&tc
->deferred_bio_list
, &tc
->retry_on_resume_list
);
2650 bio_list_init(&tc
->retry_on_resume_list
);
2651 spin_unlock_irqrestore(&tc
->lock
, flags
);
2656 /*----------------------------------------------------------------
2657 * Binding of control targets to a pool object
2658 *--------------------------------------------------------------*/
2659 static bool data_dev_supports_discard(struct pool_c
*pt
)
2661 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
2663 return q
&& blk_queue_discard(q
);
2666 static bool is_factor(sector_t block_size
, uint32_t n
)
2668 return !sector_div(block_size
, n
);
2672 * If discard_passdown was enabled verify that the data device
2673 * supports discards. Disable discard_passdown if not.
2675 static void disable_passdown_if_not_supported(struct pool_c
*pt
)
2677 struct pool
*pool
= pt
->pool
;
2678 struct block_device
*data_bdev
= pt
->data_dev
->bdev
;
2679 struct queue_limits
*data_limits
= &bdev_get_queue(data_bdev
)->limits
;
2680 const char *reason
= NULL
;
2681 char buf
[BDEVNAME_SIZE
];
2683 if (!pt
->adjusted_pf
.discard_passdown
)
2686 if (!data_dev_supports_discard(pt
))
2687 reason
= "discard unsupported";
2689 else if (data_limits
->max_discard_sectors
< pool
->sectors_per_block
)
2690 reason
= "max discard sectors smaller than a block";
2693 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev
, buf
), reason
);
2694 pt
->adjusted_pf
.discard_passdown
= false;
2698 static int bind_control_target(struct pool
*pool
, struct dm_target
*ti
)
2700 struct pool_c
*pt
= ti
->private;
2703 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
2705 enum pool_mode old_mode
= get_pool_mode(pool
);
2706 enum pool_mode new_mode
= pt
->adjusted_pf
.mode
;
2709 * Don't change the pool's mode until set_pool_mode() below.
2710 * Otherwise the pool's process_* function pointers may
2711 * not match the desired pool mode.
2713 pt
->adjusted_pf
.mode
= old_mode
;
2716 pool
->pf
= pt
->adjusted_pf
;
2717 pool
->low_water_blocks
= pt
->low_water_blocks
;
2719 set_pool_mode(pool
, new_mode
);
2724 static void unbind_control_target(struct pool
*pool
, struct dm_target
*ti
)
2730 /*----------------------------------------------------------------
2732 *--------------------------------------------------------------*/
2733 /* Initialize pool features. */
2734 static void pool_features_init(struct pool_features
*pf
)
2736 pf
->mode
= PM_WRITE
;
2737 pf
->zero_new_blocks
= true;
2738 pf
->discard_enabled
= true;
2739 pf
->discard_passdown
= true;
2740 pf
->error_if_no_space
= false;
2743 static void __pool_destroy(struct pool
*pool
)
2745 __pool_table_remove(pool
);
2747 vfree(pool
->cell_sort_array
);
2748 if (dm_pool_metadata_close(pool
->pmd
) < 0)
2749 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
2751 dm_bio_prison_destroy(pool
->prison
);
2752 dm_kcopyd_client_destroy(pool
->copier
);
2755 destroy_workqueue(pool
->wq
);
2757 if (pool
->next_mapping
)
2758 mempool_free(pool
->next_mapping
, pool
->mapping_pool
);
2759 mempool_destroy(pool
->mapping_pool
);
2760 dm_deferred_set_destroy(pool
->shared_read_ds
);
2761 dm_deferred_set_destroy(pool
->all_io_ds
);
2765 static struct kmem_cache
*_new_mapping_cache
;
2767 static struct pool
*pool_create(struct mapped_device
*pool_md
,
2768 struct block_device
*metadata_dev
,
2769 unsigned long block_size
,
2770 int read_only
, char **error
)
2775 struct dm_pool_metadata
*pmd
;
2776 bool format_device
= read_only
? false : true;
2778 pmd
= dm_pool_metadata_open(metadata_dev
, block_size
, format_device
);
2780 *error
= "Error creating metadata object";
2781 return (struct pool
*)pmd
;
2784 pool
= kmalloc(sizeof(*pool
), GFP_KERNEL
);
2786 *error
= "Error allocating memory for pool";
2787 err_p
= ERR_PTR(-ENOMEM
);
2792 pool
->sectors_per_block
= block_size
;
2793 if (block_size
& (block_size
- 1))
2794 pool
->sectors_per_block_shift
= -1;
2796 pool
->sectors_per_block_shift
= __ffs(block_size
);
2797 pool
->low_water_blocks
= 0;
2798 pool_features_init(&pool
->pf
);
2799 pool
->prison
= dm_bio_prison_create();
2800 if (!pool
->prison
) {
2801 *error
= "Error creating pool's bio prison";
2802 err_p
= ERR_PTR(-ENOMEM
);
2806 pool
->copier
= dm_kcopyd_client_create(&dm_kcopyd_throttle
);
2807 if (IS_ERR(pool
->copier
)) {
2808 r
= PTR_ERR(pool
->copier
);
2809 *error
= "Error creating pool's kcopyd client";
2811 goto bad_kcopyd_client
;
2815 * Create singlethreaded workqueue that will service all devices
2816 * that use this metadata.
2818 pool
->wq
= alloc_ordered_workqueue("dm-" DM_MSG_PREFIX
, WQ_MEM_RECLAIM
);
2820 *error
= "Error creating pool's workqueue";
2821 err_p
= ERR_PTR(-ENOMEM
);
2825 throttle_init(&pool
->throttle
);
2826 INIT_WORK(&pool
->worker
, do_worker
);
2827 INIT_DELAYED_WORK(&pool
->waker
, do_waker
);
2828 INIT_DELAYED_WORK(&pool
->no_space_timeout
, do_no_space_timeout
);
2829 spin_lock_init(&pool
->lock
);
2830 bio_list_init(&pool
->deferred_flush_bios
);
2831 INIT_LIST_HEAD(&pool
->prepared_mappings
);
2832 INIT_LIST_HEAD(&pool
->prepared_discards
);
2833 INIT_LIST_HEAD(&pool
->active_thins
);
2834 pool
->low_water_triggered
= false;
2835 pool
->suspended
= true;
2836 pool
->out_of_data_space
= false;
2838 pool
->shared_read_ds
= dm_deferred_set_create();
2839 if (!pool
->shared_read_ds
) {
2840 *error
= "Error creating pool's shared read deferred set";
2841 err_p
= ERR_PTR(-ENOMEM
);
2842 goto bad_shared_read_ds
;
2845 pool
->all_io_ds
= dm_deferred_set_create();
2846 if (!pool
->all_io_ds
) {
2847 *error
= "Error creating pool's all io deferred set";
2848 err_p
= ERR_PTR(-ENOMEM
);
2852 pool
->next_mapping
= NULL
;
2853 pool
->mapping_pool
= mempool_create_slab_pool(MAPPING_POOL_SIZE
,
2854 _new_mapping_cache
);
2855 if (!pool
->mapping_pool
) {
2856 *error
= "Error creating pool's mapping mempool";
2857 err_p
= ERR_PTR(-ENOMEM
);
2858 goto bad_mapping_pool
;
2861 pool
->cell_sort_array
= vmalloc(sizeof(*pool
->cell_sort_array
) * CELL_SORT_ARRAY_SIZE
);
2862 if (!pool
->cell_sort_array
) {
2863 *error
= "Error allocating cell sort array";
2864 err_p
= ERR_PTR(-ENOMEM
);
2865 goto bad_sort_array
;
2868 pool
->ref_count
= 1;
2869 pool
->last_commit_jiffies
= jiffies
;
2870 pool
->pool_md
= pool_md
;
2871 pool
->md_dev
= metadata_dev
;
2872 __pool_table_insert(pool
);
2877 mempool_destroy(pool
->mapping_pool
);
2879 dm_deferred_set_destroy(pool
->all_io_ds
);
2881 dm_deferred_set_destroy(pool
->shared_read_ds
);
2883 destroy_workqueue(pool
->wq
);
2885 dm_kcopyd_client_destroy(pool
->copier
);
2887 dm_bio_prison_destroy(pool
->prison
);
2891 if (dm_pool_metadata_close(pmd
))
2892 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
2897 static void __pool_inc(struct pool
*pool
)
2899 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
2903 static void __pool_dec(struct pool
*pool
)
2905 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
2906 BUG_ON(!pool
->ref_count
);
2907 if (!--pool
->ref_count
)
2908 __pool_destroy(pool
);
2911 static struct pool
*__pool_find(struct mapped_device
*pool_md
,
2912 struct block_device
*metadata_dev
,
2913 unsigned long block_size
, int read_only
,
2914 char **error
, int *created
)
2916 struct pool
*pool
= __pool_table_lookup_metadata_dev(metadata_dev
);
2919 if (pool
->pool_md
!= pool_md
) {
2920 *error
= "metadata device already in use by a pool";
2921 return ERR_PTR(-EBUSY
);
2926 pool
= __pool_table_lookup(pool_md
);
2928 if (pool
->md_dev
!= metadata_dev
) {
2929 *error
= "different pool cannot replace a pool";
2930 return ERR_PTR(-EINVAL
);
2935 pool
= pool_create(pool_md
, metadata_dev
, block_size
, read_only
, error
);
2943 /*----------------------------------------------------------------
2944 * Pool target methods
2945 *--------------------------------------------------------------*/
2946 static void pool_dtr(struct dm_target
*ti
)
2948 struct pool_c
*pt
= ti
->private;
2950 mutex_lock(&dm_thin_pool_table
.mutex
);
2952 unbind_control_target(pt
->pool
, ti
);
2953 __pool_dec(pt
->pool
);
2954 dm_put_device(ti
, pt
->metadata_dev
);
2955 dm_put_device(ti
, pt
->data_dev
);
2958 mutex_unlock(&dm_thin_pool_table
.mutex
);
2961 static int parse_pool_features(struct dm_arg_set
*as
, struct pool_features
*pf
,
2962 struct dm_target
*ti
)
2966 const char *arg_name
;
2968 static struct dm_arg _args
[] = {
2969 {0, 4, "Invalid number of pool feature arguments"},
2973 * No feature arguments supplied.
2978 r
= dm_read_arg_group(_args
, as
, &argc
, &ti
->error
);
2982 while (argc
&& !r
) {
2983 arg_name
= dm_shift_arg(as
);
2986 if (!strcasecmp(arg_name
, "skip_block_zeroing"))
2987 pf
->zero_new_blocks
= false;
2989 else if (!strcasecmp(arg_name
, "ignore_discard"))
2990 pf
->discard_enabled
= false;
2992 else if (!strcasecmp(arg_name
, "no_discard_passdown"))
2993 pf
->discard_passdown
= false;
2995 else if (!strcasecmp(arg_name
, "read_only"))
2996 pf
->mode
= PM_READ_ONLY
;
2998 else if (!strcasecmp(arg_name
, "error_if_no_space"))
2999 pf
->error_if_no_space
= true;
3002 ti
->error
= "Unrecognised pool feature requested";
3011 static void metadata_low_callback(void *context
)
3013 struct pool
*pool
= context
;
3015 DMWARN("%s: reached low water mark for metadata device: sending event.",
3016 dm_device_name(pool
->pool_md
));
3018 dm_table_event(pool
->ti
->table
);
3021 static sector_t
get_dev_size(struct block_device
*bdev
)
3023 return i_size_read(bdev
->bd_inode
) >> SECTOR_SHIFT
;
3026 static void warn_if_metadata_device_too_big(struct block_device
*bdev
)
3028 sector_t metadata_dev_size
= get_dev_size(bdev
);
3029 char buffer
[BDEVNAME_SIZE
];
3031 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS_WARNING
)
3032 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
3033 bdevname(bdev
, buffer
), THIN_METADATA_MAX_SECTORS
);
3036 static sector_t
get_metadata_dev_size(struct block_device
*bdev
)
3038 sector_t metadata_dev_size
= get_dev_size(bdev
);
3040 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS
)
3041 metadata_dev_size
= THIN_METADATA_MAX_SECTORS
;
3043 return metadata_dev_size
;
3046 static dm_block_t
get_metadata_dev_size_in_blocks(struct block_device
*bdev
)
3048 sector_t metadata_dev_size
= get_metadata_dev_size(bdev
);
3050 sector_div(metadata_dev_size
, THIN_METADATA_BLOCK_SIZE
);
3052 return metadata_dev_size
;
3056 * When a metadata threshold is crossed a dm event is triggered, and
3057 * userland should respond by growing the metadata device. We could let
3058 * userland set the threshold, like we do with the data threshold, but I'm
3059 * not sure they know enough to do this well.
3061 static dm_block_t
calc_metadata_threshold(struct pool_c
*pt
)
3064 * 4M is ample for all ops with the possible exception of thin
3065 * device deletion which is harmless if it fails (just retry the
3066 * delete after you've grown the device).
3068 dm_block_t quarter
= get_metadata_dev_size_in_blocks(pt
->metadata_dev
->bdev
) / 4;
3069 return min((dm_block_t
)1024ULL /* 4M */, quarter
);
3073 * thin-pool <metadata dev> <data dev>
3074 * <data block size (sectors)>
3075 * <low water mark (blocks)>
3076 * [<#feature args> [<arg>]*]
3078 * Optional feature arguments are:
3079 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
3080 * ignore_discard: disable discard
3081 * no_discard_passdown: don't pass discards down to the data device
3082 * read_only: Don't allow any changes to be made to the pool metadata.
3083 * error_if_no_space: error IOs, instead of queueing, if no space.
3085 static int pool_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
3087 int r
, pool_created
= 0;
3090 struct pool_features pf
;
3091 struct dm_arg_set as
;
3092 struct dm_dev
*data_dev
;
3093 unsigned long block_size
;
3094 dm_block_t low_water_blocks
;
3095 struct dm_dev
*metadata_dev
;
3096 fmode_t metadata_mode
;
3099 * FIXME Remove validation from scope of lock.
3101 mutex_lock(&dm_thin_pool_table
.mutex
);
3104 ti
->error
= "Invalid argument count";
3113 * Set default pool features.
3115 pool_features_init(&pf
);
3117 dm_consume_args(&as
, 4);
3118 r
= parse_pool_features(&as
, &pf
, ti
);
3122 metadata_mode
= FMODE_READ
| ((pf
.mode
== PM_READ_ONLY
) ? 0 : FMODE_WRITE
);
3123 r
= dm_get_device(ti
, argv
[0], metadata_mode
, &metadata_dev
);
3125 ti
->error
= "Error opening metadata block device";
3128 warn_if_metadata_device_too_big(metadata_dev
->bdev
);
3130 r
= dm_get_device(ti
, argv
[1], FMODE_READ
| FMODE_WRITE
, &data_dev
);
3132 ti
->error
= "Error getting data device";
3136 if (kstrtoul(argv
[2], 10, &block_size
) || !block_size
||
3137 block_size
< DATA_DEV_BLOCK_SIZE_MIN_SECTORS
||
3138 block_size
> DATA_DEV_BLOCK_SIZE_MAX_SECTORS
||
3139 block_size
& (DATA_DEV_BLOCK_SIZE_MIN_SECTORS
- 1)) {
3140 ti
->error
= "Invalid block size";
3145 if (kstrtoull(argv
[3], 10, (unsigned long long *)&low_water_blocks
)) {
3146 ti
->error
= "Invalid low water mark";
3151 pt
= kzalloc(sizeof(*pt
), GFP_KERNEL
);
3157 pool
= __pool_find(dm_table_get_md(ti
->table
), metadata_dev
->bdev
,
3158 block_size
, pf
.mode
== PM_READ_ONLY
, &ti
->error
, &pool_created
);
3165 * 'pool_created' reflects whether this is the first table load.
3166 * Top level discard support is not allowed to be changed after
3167 * initial load. This would require a pool reload to trigger thin
3170 if (!pool_created
&& pf
.discard_enabled
!= pool
->pf
.discard_enabled
) {
3171 ti
->error
= "Discard support cannot be disabled once enabled";
3173 goto out_flags_changed
;
3178 pt
->metadata_dev
= metadata_dev
;
3179 pt
->data_dev
= data_dev
;
3180 pt
->low_water_blocks
= low_water_blocks
;
3181 pt
->adjusted_pf
= pt
->requested_pf
= pf
;
3182 ti
->num_flush_bios
= 1;
3185 * Only need to enable discards if the pool should pass
3186 * them down to the data device. The thin device's discard
3187 * processing will cause mappings to be removed from the btree.
3189 ti
->discard_zeroes_data_unsupported
= true;
3190 if (pf
.discard_enabled
&& pf
.discard_passdown
) {
3191 ti
->num_discard_bios
= 1;
3194 * Setting 'discards_supported' circumvents the normal
3195 * stacking of discard limits (this keeps the pool and
3196 * thin devices' discard limits consistent).
3198 ti
->discards_supported
= true;
3202 r
= dm_pool_register_metadata_threshold(pt
->pool
->pmd
,
3203 calc_metadata_threshold(pt
),
3204 metadata_low_callback
,
3207 goto out_flags_changed
;
3209 pt
->callbacks
.congested_fn
= pool_is_congested
;
3210 dm_table_add_target_callbacks(ti
->table
, &pt
->callbacks
);
3212 mutex_unlock(&dm_thin_pool_table
.mutex
);
3221 dm_put_device(ti
, data_dev
);
3223 dm_put_device(ti
, metadata_dev
);
3225 mutex_unlock(&dm_thin_pool_table
.mutex
);
3230 static int pool_map(struct dm_target
*ti
, struct bio
*bio
)
3233 struct pool_c
*pt
= ti
->private;
3234 struct pool
*pool
= pt
->pool
;
3235 unsigned long flags
;
3238 * As this is a singleton target, ti->begin is always zero.
3240 spin_lock_irqsave(&pool
->lock
, flags
);
3241 bio
->bi_bdev
= pt
->data_dev
->bdev
;
3242 r
= DM_MAPIO_REMAPPED
;
3243 spin_unlock_irqrestore(&pool
->lock
, flags
);
3248 static int maybe_resize_data_dev(struct dm_target
*ti
, bool *need_commit
)
3251 struct pool_c
*pt
= ti
->private;
3252 struct pool
*pool
= pt
->pool
;
3253 sector_t data_size
= ti
->len
;
3254 dm_block_t sb_data_size
;
3256 *need_commit
= false;
3258 (void) sector_div(data_size
, pool
->sectors_per_block
);
3260 r
= dm_pool_get_data_dev_size(pool
->pmd
, &sb_data_size
);
3262 DMERR("%s: failed to retrieve data device size",
3263 dm_device_name(pool
->pool_md
));
3267 if (data_size
< sb_data_size
) {
3268 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
3269 dm_device_name(pool
->pool_md
),
3270 (unsigned long long)data_size
, sb_data_size
);
3273 } else if (data_size
> sb_data_size
) {
3274 if (dm_pool_metadata_needs_check(pool
->pmd
)) {
3275 DMERR("%s: unable to grow the data device until repaired.",
3276 dm_device_name(pool
->pool_md
));
3281 DMINFO("%s: growing the data device from %llu to %llu blocks",
3282 dm_device_name(pool
->pool_md
),
3283 sb_data_size
, (unsigned long long)data_size
);
3284 r
= dm_pool_resize_data_dev(pool
->pmd
, data_size
);
3286 metadata_operation_failed(pool
, "dm_pool_resize_data_dev", r
);
3290 *need_commit
= true;
3296 static int maybe_resize_metadata_dev(struct dm_target
*ti
, bool *need_commit
)
3299 struct pool_c
*pt
= ti
->private;
3300 struct pool
*pool
= pt
->pool
;
3301 dm_block_t metadata_dev_size
, sb_metadata_dev_size
;
3303 *need_commit
= false;
3305 metadata_dev_size
= get_metadata_dev_size_in_blocks(pool
->md_dev
);
3307 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &sb_metadata_dev_size
);
3309 DMERR("%s: failed to retrieve metadata device size",
3310 dm_device_name(pool
->pool_md
));
3314 if (metadata_dev_size
< sb_metadata_dev_size
) {
3315 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
3316 dm_device_name(pool
->pool_md
),
3317 metadata_dev_size
, sb_metadata_dev_size
);
3320 } else if (metadata_dev_size
> sb_metadata_dev_size
) {
3321 if (dm_pool_metadata_needs_check(pool
->pmd
)) {
3322 DMERR("%s: unable to grow the metadata device until repaired.",
3323 dm_device_name(pool
->pool_md
));
3327 warn_if_metadata_device_too_big(pool
->md_dev
);
3328 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
3329 dm_device_name(pool
->pool_md
),
3330 sb_metadata_dev_size
, metadata_dev_size
);
3331 r
= dm_pool_resize_metadata_dev(pool
->pmd
, metadata_dev_size
);
3333 metadata_operation_failed(pool
, "dm_pool_resize_metadata_dev", r
);
3337 *need_commit
= true;
3344 * Retrieves the number of blocks of the data device from
3345 * the superblock and compares it to the actual device size,
3346 * thus resizing the data device in case it has grown.
3348 * This both copes with opening preallocated data devices in the ctr
3349 * being followed by a resume
3351 * calling the resume method individually after userspace has
3352 * grown the data device in reaction to a table event.
3354 static int pool_preresume(struct dm_target
*ti
)
3357 bool need_commit1
, need_commit2
;
3358 struct pool_c
*pt
= ti
->private;
3359 struct pool
*pool
= pt
->pool
;
3362 * Take control of the pool object.
3364 r
= bind_control_target(pool
, ti
);
3368 r
= maybe_resize_data_dev(ti
, &need_commit1
);
3372 r
= maybe_resize_metadata_dev(ti
, &need_commit2
);
3376 if (need_commit1
|| need_commit2
)
3377 (void) commit(pool
);
3382 static void pool_suspend_active_thins(struct pool
*pool
)
3386 /* Suspend all active thin devices */
3387 tc
= get_first_thin(pool
);
3389 dm_internal_suspend_noflush(tc
->thin_md
);
3390 tc
= get_next_thin(pool
, tc
);
3394 static void pool_resume_active_thins(struct pool
*pool
)
3398 /* Resume all active thin devices */
3399 tc
= get_first_thin(pool
);
3401 dm_internal_resume(tc
->thin_md
);
3402 tc
= get_next_thin(pool
, tc
);
3406 static void pool_resume(struct dm_target
*ti
)
3408 struct pool_c
*pt
= ti
->private;
3409 struct pool
*pool
= pt
->pool
;
3410 unsigned long flags
;
3413 * Must requeue active_thins' bios and then resume
3414 * active_thins _before_ clearing 'suspend' flag.
3417 pool_resume_active_thins(pool
);
3419 spin_lock_irqsave(&pool
->lock
, flags
);
3420 pool
->low_water_triggered
= false;
3421 pool
->suspended
= false;
3422 spin_unlock_irqrestore(&pool
->lock
, flags
);
3424 do_waker(&pool
->waker
.work
);
3427 static void pool_presuspend(struct dm_target
*ti
)
3429 struct pool_c
*pt
= ti
->private;
3430 struct pool
*pool
= pt
->pool
;
3431 unsigned long flags
;
3433 spin_lock_irqsave(&pool
->lock
, flags
);
3434 pool
->suspended
= true;
3435 spin_unlock_irqrestore(&pool
->lock
, flags
);
3437 pool_suspend_active_thins(pool
);
3440 static void pool_presuspend_undo(struct dm_target
*ti
)
3442 struct pool_c
*pt
= ti
->private;
3443 struct pool
*pool
= pt
->pool
;
3444 unsigned long flags
;
3446 pool_resume_active_thins(pool
);
3448 spin_lock_irqsave(&pool
->lock
, flags
);
3449 pool
->suspended
= false;
3450 spin_unlock_irqrestore(&pool
->lock
, flags
);
3453 static void pool_postsuspend(struct dm_target
*ti
)
3455 struct pool_c
*pt
= ti
->private;
3456 struct pool
*pool
= pt
->pool
;
3458 cancel_delayed_work_sync(&pool
->waker
);
3459 cancel_delayed_work_sync(&pool
->no_space_timeout
);
3460 flush_workqueue(pool
->wq
);
3461 (void) commit(pool
);
3464 static int check_arg_count(unsigned argc
, unsigned args_required
)
3466 if (argc
!= args_required
) {
3467 DMWARN("Message received with %u arguments instead of %u.",
3468 argc
, args_required
);
3475 static int read_dev_id(char *arg
, dm_thin_id
*dev_id
, int warning
)
3477 if (!kstrtoull(arg
, 10, (unsigned long long *)dev_id
) &&
3478 *dev_id
<= MAX_DEV_ID
)
3482 DMWARN("Message received with invalid device id: %s", arg
);
3487 static int process_create_thin_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
3492 r
= check_arg_count(argc
, 2);
3496 r
= read_dev_id(argv
[1], &dev_id
, 1);
3500 r
= dm_pool_create_thin(pool
->pmd
, dev_id
);
3502 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
3510 static int process_create_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
3513 dm_thin_id origin_dev_id
;
3516 r
= check_arg_count(argc
, 3);
3520 r
= read_dev_id(argv
[1], &dev_id
, 1);
3524 r
= read_dev_id(argv
[2], &origin_dev_id
, 1);
3528 r
= dm_pool_create_snap(pool
->pmd
, dev_id
, origin_dev_id
);
3530 DMWARN("Creation of new snapshot %s of device %s failed.",
3538 static int process_delete_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
3543 r
= check_arg_count(argc
, 2);
3547 r
= read_dev_id(argv
[1], &dev_id
, 1);
3551 r
= dm_pool_delete_thin_device(pool
->pmd
, dev_id
);
3553 DMWARN("Deletion of thin device %s failed.", argv
[1]);
3558 static int process_set_transaction_id_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
3560 dm_thin_id old_id
, new_id
;
3563 r
= check_arg_count(argc
, 3);
3567 if (kstrtoull(argv
[1], 10, (unsigned long long *)&old_id
)) {
3568 DMWARN("set_transaction_id message: Unrecognised id %s.", argv
[1]);
3572 if (kstrtoull(argv
[2], 10, (unsigned long long *)&new_id
)) {
3573 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv
[2]);
3577 r
= dm_pool_set_metadata_transaction_id(pool
->pmd
, old_id
, new_id
);
3579 DMWARN("Failed to change transaction id from %s to %s.",
3587 static int process_reserve_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
3591 r
= check_arg_count(argc
, 1);
3595 (void) commit(pool
);
3597 r
= dm_pool_reserve_metadata_snap(pool
->pmd
);
3599 DMWARN("reserve_metadata_snap message failed.");
3604 static int process_release_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
3608 r
= check_arg_count(argc
, 1);
3612 r
= dm_pool_release_metadata_snap(pool
->pmd
);
3614 DMWARN("release_metadata_snap message failed.");
3620 * Messages supported:
3621 * create_thin <dev_id>
3622 * create_snap <dev_id> <origin_id>
3624 * set_transaction_id <current_trans_id> <new_trans_id>
3625 * reserve_metadata_snap
3626 * release_metadata_snap
3628 static int pool_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
3631 struct pool_c
*pt
= ti
->private;
3632 struct pool
*pool
= pt
->pool
;
3634 if (get_pool_mode(pool
) >= PM_READ_ONLY
) {
3635 DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode",
3636 dm_device_name(pool
->pool_md
));
3640 if (!strcasecmp(argv
[0], "create_thin"))
3641 r
= process_create_thin_mesg(argc
, argv
, pool
);
3643 else if (!strcasecmp(argv
[0], "create_snap"))
3644 r
= process_create_snap_mesg(argc
, argv
, pool
);
3646 else if (!strcasecmp(argv
[0], "delete"))
3647 r
= process_delete_mesg(argc
, argv
, pool
);
3649 else if (!strcasecmp(argv
[0], "set_transaction_id"))
3650 r
= process_set_transaction_id_mesg(argc
, argv
, pool
);
3652 else if (!strcasecmp(argv
[0], "reserve_metadata_snap"))
3653 r
= process_reserve_metadata_snap_mesg(argc
, argv
, pool
);
3655 else if (!strcasecmp(argv
[0], "release_metadata_snap"))
3656 r
= process_release_metadata_snap_mesg(argc
, argv
, pool
);
3659 DMWARN("Unrecognised thin pool target message received: %s", argv
[0]);
3662 (void) commit(pool
);
3667 static void emit_flags(struct pool_features
*pf
, char *result
,
3668 unsigned sz
, unsigned maxlen
)
3670 unsigned count
= !pf
->zero_new_blocks
+ !pf
->discard_enabled
+
3671 !pf
->discard_passdown
+ (pf
->mode
== PM_READ_ONLY
) +
3672 pf
->error_if_no_space
;
3673 DMEMIT("%u ", count
);
3675 if (!pf
->zero_new_blocks
)
3676 DMEMIT("skip_block_zeroing ");
3678 if (!pf
->discard_enabled
)
3679 DMEMIT("ignore_discard ");
3681 if (!pf
->discard_passdown
)
3682 DMEMIT("no_discard_passdown ");
3684 if (pf
->mode
== PM_READ_ONLY
)
3685 DMEMIT("read_only ");
3687 if (pf
->error_if_no_space
)
3688 DMEMIT("error_if_no_space ");
3693 * <transaction id> <used metadata sectors>/<total metadata sectors>
3694 * <used data sectors>/<total data sectors> <held metadata root>
3695 * <pool mode> <discard config> <no space config> <needs_check>
3697 static void pool_status(struct dm_target
*ti
, status_type_t type
,
3698 unsigned status_flags
, char *result
, unsigned maxlen
)
3702 uint64_t transaction_id
;
3703 dm_block_t nr_free_blocks_data
;
3704 dm_block_t nr_free_blocks_metadata
;
3705 dm_block_t nr_blocks_data
;
3706 dm_block_t nr_blocks_metadata
;
3707 dm_block_t held_root
;
3708 char buf
[BDEVNAME_SIZE
];
3709 char buf2
[BDEVNAME_SIZE
];
3710 struct pool_c
*pt
= ti
->private;
3711 struct pool
*pool
= pt
->pool
;
3714 case STATUSTYPE_INFO
:
3715 if (get_pool_mode(pool
) == PM_FAIL
) {
3720 /* Commit to ensure statistics aren't out-of-date */
3721 if (!(status_flags
& DM_STATUS_NOFLUSH_FLAG
) && !dm_suspended(ti
))
3722 (void) commit(pool
);
3724 r
= dm_pool_get_metadata_transaction_id(pool
->pmd
, &transaction_id
);
3726 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
3727 dm_device_name(pool
->pool_md
), r
);
3731 r
= dm_pool_get_free_metadata_block_count(pool
->pmd
, &nr_free_blocks_metadata
);
3733 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
3734 dm_device_name(pool
->pool_md
), r
);
3738 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &nr_blocks_metadata
);
3740 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
3741 dm_device_name(pool
->pool_md
), r
);
3745 r
= dm_pool_get_free_block_count(pool
->pmd
, &nr_free_blocks_data
);
3747 DMERR("%s: dm_pool_get_free_block_count returned %d",
3748 dm_device_name(pool
->pool_md
), r
);
3752 r
= dm_pool_get_data_dev_size(pool
->pmd
, &nr_blocks_data
);
3754 DMERR("%s: dm_pool_get_data_dev_size returned %d",
3755 dm_device_name(pool
->pool_md
), r
);
3759 r
= dm_pool_get_metadata_snap(pool
->pmd
, &held_root
);
3761 DMERR("%s: dm_pool_get_metadata_snap returned %d",
3762 dm_device_name(pool
->pool_md
), r
);
3766 DMEMIT("%llu %llu/%llu %llu/%llu ",
3767 (unsigned long long)transaction_id
,
3768 (unsigned long long)(nr_blocks_metadata
- nr_free_blocks_metadata
),
3769 (unsigned long long)nr_blocks_metadata
,
3770 (unsigned long long)(nr_blocks_data
- nr_free_blocks_data
),
3771 (unsigned long long)nr_blocks_data
);
3774 DMEMIT("%llu ", held_root
);
3778 if (pool
->pf
.mode
== PM_OUT_OF_DATA_SPACE
)
3779 DMEMIT("out_of_data_space ");
3780 else if (pool
->pf
.mode
== PM_READ_ONLY
)
3785 if (!pool
->pf
.discard_enabled
)
3786 DMEMIT("ignore_discard ");
3787 else if (pool
->pf
.discard_passdown
)
3788 DMEMIT("discard_passdown ");
3790 DMEMIT("no_discard_passdown ");
3792 if (pool
->pf
.error_if_no_space
)
3793 DMEMIT("error_if_no_space ");
3795 DMEMIT("queue_if_no_space ");
3797 if (dm_pool_metadata_needs_check(pool
->pmd
))
3798 DMEMIT("needs_check ");
3804 case STATUSTYPE_TABLE
:
3805 DMEMIT("%s %s %lu %llu ",
3806 format_dev_t(buf
, pt
->metadata_dev
->bdev
->bd_dev
),
3807 format_dev_t(buf2
, pt
->data_dev
->bdev
->bd_dev
),
3808 (unsigned long)pool
->sectors_per_block
,
3809 (unsigned long long)pt
->low_water_blocks
);
3810 emit_flags(&pt
->requested_pf
, result
, sz
, maxlen
);
3819 static int pool_iterate_devices(struct dm_target
*ti
,
3820 iterate_devices_callout_fn fn
, void *data
)
3822 struct pool_c
*pt
= ti
->private;
3824 return fn(ti
, pt
->data_dev
, 0, ti
->len
, data
);
3827 static void pool_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
3829 struct pool_c
*pt
= ti
->private;
3830 struct pool
*pool
= pt
->pool
;
3831 sector_t io_opt_sectors
= limits
->io_opt
>> SECTOR_SHIFT
;
3834 * If max_sectors is smaller than pool->sectors_per_block adjust it
3835 * to the highest possible power-of-2 factor of pool->sectors_per_block.
3836 * This is especially beneficial when the pool's data device is a RAID
3837 * device that has a full stripe width that matches pool->sectors_per_block
3838 * -- because even though partial RAID stripe-sized IOs will be issued to a
3839 * single RAID stripe; when aggregated they will end on a full RAID stripe
3840 * boundary.. which avoids additional partial RAID stripe writes cascading
3842 if (limits
->max_sectors
< pool
->sectors_per_block
) {
3843 while (!is_factor(pool
->sectors_per_block
, limits
->max_sectors
)) {
3844 if ((limits
->max_sectors
& (limits
->max_sectors
- 1)) == 0)
3845 limits
->max_sectors
--;
3846 limits
->max_sectors
= rounddown_pow_of_two(limits
->max_sectors
);
3851 * If the system-determined stacked limits are compatible with the
3852 * pool's blocksize (io_opt is a factor) do not override them.
3854 if (io_opt_sectors
< pool
->sectors_per_block
||
3855 !is_factor(io_opt_sectors
, pool
->sectors_per_block
)) {
3856 if (is_factor(pool
->sectors_per_block
, limits
->max_sectors
))
3857 blk_limits_io_min(limits
, limits
->max_sectors
<< SECTOR_SHIFT
);
3859 blk_limits_io_min(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
3860 blk_limits_io_opt(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
3864 * pt->adjusted_pf is a staging area for the actual features to use.
3865 * They get transferred to the live pool in bind_control_target()
3866 * called from pool_preresume().
3868 if (!pt
->adjusted_pf
.discard_enabled
) {
3870 * Must explicitly disallow stacking discard limits otherwise the
3871 * block layer will stack them if pool's data device has support.
3872 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
3873 * user to see that, so make sure to set all discard limits to 0.
3875 limits
->discard_granularity
= 0;
3879 disable_passdown_if_not_supported(pt
);
3882 * The pool uses the same discard limits as the underlying data
3883 * device. DM core has already set this up.
3887 static struct target_type pool_target
= {
3888 .name
= "thin-pool",
3889 .features
= DM_TARGET_SINGLETON
| DM_TARGET_ALWAYS_WRITEABLE
|
3890 DM_TARGET_IMMUTABLE
,
3891 .version
= {1, 19, 0},
3892 .module
= THIS_MODULE
,
3896 .presuspend
= pool_presuspend
,
3897 .presuspend_undo
= pool_presuspend_undo
,
3898 .postsuspend
= pool_postsuspend
,
3899 .preresume
= pool_preresume
,
3900 .resume
= pool_resume
,
3901 .message
= pool_message
,
3902 .status
= pool_status
,
3903 .iterate_devices
= pool_iterate_devices
,
3904 .io_hints
= pool_io_hints
,
3907 /*----------------------------------------------------------------
3908 * Thin target methods
3909 *--------------------------------------------------------------*/
3910 static void thin_get(struct thin_c
*tc
)
3912 atomic_inc(&tc
->refcount
);
3915 static void thin_put(struct thin_c
*tc
)
3917 if (atomic_dec_and_test(&tc
->refcount
))
3918 complete(&tc
->can_destroy
);
3921 static void thin_dtr(struct dm_target
*ti
)
3923 struct thin_c
*tc
= ti
->private;
3924 unsigned long flags
;
3926 spin_lock_irqsave(&tc
->pool
->lock
, flags
);
3927 list_del_rcu(&tc
->list
);
3928 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
3932 wait_for_completion(&tc
->can_destroy
);
3934 mutex_lock(&dm_thin_pool_table
.mutex
);
3936 __pool_dec(tc
->pool
);
3937 dm_pool_close_thin_device(tc
->td
);
3938 dm_put_device(ti
, tc
->pool_dev
);
3940 dm_put_device(ti
, tc
->origin_dev
);
3943 mutex_unlock(&dm_thin_pool_table
.mutex
);
3947 * Thin target parameters:
3949 * <pool_dev> <dev_id> [origin_dev]
3951 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
3952 * dev_id: the internal device identifier
3953 * origin_dev: a device external to the pool that should act as the origin
3955 * If the pool device has discards disabled, they get disabled for the thin
3958 static int thin_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
3962 struct dm_dev
*pool_dev
, *origin_dev
;
3963 struct mapped_device
*pool_md
;
3964 unsigned long flags
;
3966 mutex_lock(&dm_thin_pool_table
.mutex
);
3968 if (argc
!= 2 && argc
!= 3) {
3969 ti
->error
= "Invalid argument count";
3974 tc
= ti
->private = kzalloc(sizeof(*tc
), GFP_KERNEL
);
3976 ti
->error
= "Out of memory";
3980 tc
->thin_md
= dm_table_get_md(ti
->table
);
3981 spin_lock_init(&tc
->lock
);
3982 INIT_LIST_HEAD(&tc
->deferred_cells
);
3983 bio_list_init(&tc
->deferred_bio_list
);
3984 bio_list_init(&tc
->retry_on_resume_list
);
3985 tc
->sort_bio_list
= RB_ROOT
;
3988 r
= dm_get_device(ti
, argv
[2], FMODE_READ
, &origin_dev
);
3990 ti
->error
= "Error opening origin device";
3991 goto bad_origin_dev
;
3993 tc
->origin_dev
= origin_dev
;
3996 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &pool_dev
);
3998 ti
->error
= "Error opening pool device";
4001 tc
->pool_dev
= pool_dev
;
4003 if (read_dev_id(argv
[1], (unsigned long long *)&tc
->dev_id
, 0)) {
4004 ti
->error
= "Invalid device id";
4009 pool_md
= dm_get_md(tc
->pool_dev
->bdev
->bd_dev
);
4011 ti
->error
= "Couldn't get pool mapped device";
4016 tc
->pool
= __pool_table_lookup(pool_md
);
4018 ti
->error
= "Couldn't find pool object";
4020 goto bad_pool_lookup
;
4022 __pool_inc(tc
->pool
);
4024 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
4025 ti
->error
= "Couldn't open thin device, Pool is in fail mode";
4030 r
= dm_pool_open_thin_device(tc
->pool
->pmd
, tc
->dev_id
, &tc
->td
);
4032 ti
->error
= "Couldn't open thin internal device";
4036 r
= dm_set_target_max_io_len(ti
, tc
->pool
->sectors_per_block
);
4040 ti
->num_flush_bios
= 1;
4041 ti
->flush_supported
= true;
4042 ti
->per_io_data_size
= sizeof(struct dm_thin_endio_hook
);
4044 /* In case the pool supports discards, pass them on. */
4045 ti
->discard_zeroes_data_unsupported
= true;
4046 if (tc
->pool
->pf
.discard_enabled
) {
4047 ti
->discards_supported
= true;
4048 ti
->num_discard_bios
= 1;
4049 ti
->split_discard_bios
= false;
4052 mutex_unlock(&dm_thin_pool_table
.mutex
);
4054 spin_lock_irqsave(&tc
->pool
->lock
, flags
);
4055 if (tc
->pool
->suspended
) {
4056 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
4057 mutex_lock(&dm_thin_pool_table
.mutex
); /* reacquire for __pool_dec */
4058 ti
->error
= "Unable to activate thin device while pool is suspended";
4062 atomic_set(&tc
->refcount
, 1);
4063 init_completion(&tc
->can_destroy
);
4064 list_add_tail_rcu(&tc
->list
, &tc
->pool
->active_thins
);
4065 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
4067 * This synchronize_rcu() call is needed here otherwise we risk a
4068 * wake_worker() call finding no bios to process (because the newly
4069 * added tc isn't yet visible). So this reduces latency since we
4070 * aren't then dependent on the periodic commit to wake_worker().
4079 dm_pool_close_thin_device(tc
->td
);
4081 __pool_dec(tc
->pool
);
4085 dm_put_device(ti
, tc
->pool_dev
);
4088 dm_put_device(ti
, tc
->origin_dev
);
4092 mutex_unlock(&dm_thin_pool_table
.mutex
);
4097 static int thin_map(struct dm_target
*ti
, struct bio
*bio
)
4099 bio
->bi_iter
.bi_sector
= dm_target_offset(ti
, bio
->bi_iter
.bi_sector
);
4101 return thin_bio_map(ti
, bio
);
4104 static int thin_endio(struct dm_target
*ti
, struct bio
*bio
, int err
)
4106 unsigned long flags
;
4107 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
4108 struct list_head work
;
4109 struct dm_thin_new_mapping
*m
, *tmp
;
4110 struct pool
*pool
= h
->tc
->pool
;
4112 if (h
->shared_read_entry
) {
4113 INIT_LIST_HEAD(&work
);
4114 dm_deferred_entry_dec(h
->shared_read_entry
, &work
);
4116 spin_lock_irqsave(&pool
->lock
, flags
);
4117 list_for_each_entry_safe(m
, tmp
, &work
, list
) {
4119 __complete_mapping_preparation(m
);
4121 spin_unlock_irqrestore(&pool
->lock
, flags
);
4124 if (h
->all_io_entry
) {
4125 INIT_LIST_HEAD(&work
);
4126 dm_deferred_entry_dec(h
->all_io_entry
, &work
);
4127 if (!list_empty(&work
)) {
4128 spin_lock_irqsave(&pool
->lock
, flags
);
4129 list_for_each_entry_safe(m
, tmp
, &work
, list
)
4130 list_add_tail(&m
->list
, &pool
->prepared_discards
);
4131 spin_unlock_irqrestore(&pool
->lock
, flags
);
4137 cell_defer_no_holder(h
->tc
, h
->cell
);
4142 static void thin_presuspend(struct dm_target
*ti
)
4144 struct thin_c
*tc
= ti
->private;
4146 if (dm_noflush_suspending(ti
))
4147 noflush_work(tc
, do_noflush_start
);
4150 static void thin_postsuspend(struct dm_target
*ti
)
4152 struct thin_c
*tc
= ti
->private;
4155 * The dm_noflush_suspending flag has been cleared by now, so
4156 * unfortunately we must always run this.
4158 noflush_work(tc
, do_noflush_stop
);
4161 static int thin_preresume(struct dm_target
*ti
)
4163 struct thin_c
*tc
= ti
->private;
4166 tc
->origin_size
= get_dev_size(tc
->origin_dev
->bdev
);
4172 * <nr mapped sectors> <highest mapped sector>
4174 static void thin_status(struct dm_target
*ti
, status_type_t type
,
4175 unsigned status_flags
, char *result
, unsigned maxlen
)
4179 dm_block_t mapped
, highest
;
4180 char buf
[BDEVNAME_SIZE
];
4181 struct thin_c
*tc
= ti
->private;
4183 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
4192 case STATUSTYPE_INFO
:
4193 r
= dm_thin_get_mapped_count(tc
->td
, &mapped
);
4195 DMERR("dm_thin_get_mapped_count returned %d", r
);
4199 r
= dm_thin_get_highest_mapped_block(tc
->td
, &highest
);
4201 DMERR("dm_thin_get_highest_mapped_block returned %d", r
);
4205 DMEMIT("%llu ", mapped
* tc
->pool
->sectors_per_block
);
4207 DMEMIT("%llu", ((highest
+ 1) *
4208 tc
->pool
->sectors_per_block
) - 1);
4213 case STATUSTYPE_TABLE
:
4215 format_dev_t(buf
, tc
->pool_dev
->bdev
->bd_dev
),
4216 (unsigned long) tc
->dev_id
);
4218 DMEMIT(" %s", format_dev_t(buf
, tc
->origin_dev
->bdev
->bd_dev
));
4229 static int thin_iterate_devices(struct dm_target
*ti
,
4230 iterate_devices_callout_fn fn
, void *data
)
4233 struct thin_c
*tc
= ti
->private;
4234 struct pool
*pool
= tc
->pool
;
4237 * We can't call dm_pool_get_data_dev_size() since that blocks. So
4238 * we follow a more convoluted path through to the pool's target.
4241 return 0; /* nothing is bound */
4243 blocks
= pool
->ti
->len
;
4244 (void) sector_div(blocks
, pool
->sectors_per_block
);
4246 return fn(ti
, tc
->pool_dev
, 0, pool
->sectors_per_block
* blocks
, data
);
4251 static void thin_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
4253 struct thin_c
*tc
= ti
->private;
4254 struct pool
*pool
= tc
->pool
;
4256 if (!pool
->pf
.discard_enabled
)
4259 limits
->discard_granularity
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
4260 limits
->max_discard_sectors
= 2048 * 1024 * 16; /* 16G */
4263 static struct target_type thin_target
= {
4265 .version
= {1, 19, 0},
4266 .module
= THIS_MODULE
,
4270 .end_io
= thin_endio
,
4271 .preresume
= thin_preresume
,
4272 .presuspend
= thin_presuspend
,
4273 .postsuspend
= thin_postsuspend
,
4274 .status
= thin_status
,
4275 .iterate_devices
= thin_iterate_devices
,
4276 .io_hints
= thin_io_hints
,
4279 /*----------------------------------------------------------------*/
4281 static int __init
dm_thin_init(void)
4287 r
= dm_register_target(&thin_target
);
4291 r
= dm_register_target(&pool_target
);
4293 goto bad_pool_target
;
4297 _new_mapping_cache
= KMEM_CACHE(dm_thin_new_mapping
, 0);
4298 if (!_new_mapping_cache
)
4299 goto bad_new_mapping_cache
;
4303 bad_new_mapping_cache
:
4304 dm_unregister_target(&pool_target
);
4306 dm_unregister_target(&thin_target
);
4311 static void dm_thin_exit(void)
4313 dm_unregister_target(&thin_target
);
4314 dm_unregister_target(&pool_target
);
4316 kmem_cache_destroy(_new_mapping_cache
);
4319 module_init(dm_thin_init
);
4320 module_exit(dm_thin_exit
);
4322 module_param_named(no_space_timeout
, no_space_timeout_secs
, uint
, S_IRUGO
| S_IWUSR
);
4323 MODULE_PARM_DESC(no_space_timeout
, "Out of data space queue IO timeout in seconds");
4325 MODULE_DESCRIPTION(DM_NAME
" thin provisioning target");
4326 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
4327 MODULE_LICENSE("GPL");