2 * Copyright (C) 2011-2012 Red Hat UK.
4 * This file is released under the GPL.
7 #include "dm-thin-metadata.h"
8 #include "dm-bio-prison.h"
11 #include <linux/device-mapper.h>
12 #include <linux/dm-io.h>
13 #include <linux/dm-kcopyd.h>
14 #include <linux/list.h>
15 #include <linux/rculist.h>
16 #include <linux/init.h>
17 #include <linux/module.h>
18 #include <linux/slab.h>
19 #include <linux/rbtree.h>
21 #define DM_MSG_PREFIX "thin"
26 #define ENDIO_HOOK_POOL_SIZE 1024
27 #define MAPPING_POOL_SIZE 1024
28 #define PRISON_CELLS 1024
29 #define COMMIT_PERIOD HZ
30 #define NO_SPACE_TIMEOUT_SECS 60
32 static unsigned no_space_timeout_secs
= NO_SPACE_TIMEOUT_SECS
;
34 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle
,
35 "A percentage of time allocated for copy on write");
38 * The block size of the device holding pool data must be
39 * between 64KB and 1GB.
41 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
42 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
45 * Device id is restricted to 24 bits.
47 #define MAX_DEV_ID ((1 << 24) - 1)
50 * How do we handle breaking sharing of data blocks?
51 * =================================================
53 * We use a standard copy-on-write btree to store the mappings for the
54 * devices (note I'm talking about copy-on-write of the metadata here, not
55 * the data). When you take an internal snapshot you clone the root node
56 * of the origin btree. After this there is no concept of an origin or a
57 * snapshot. They are just two device trees that happen to point to the
60 * When we get a write in we decide if it's to a shared data block using
61 * some timestamp magic. If it is, we have to break sharing.
63 * Let's say we write to a shared block in what was the origin. The
66 * i) plug io further to this physical block. (see bio_prison code).
68 * ii) quiesce any read io to that shared data block. Obviously
69 * including all devices that share this block. (see dm_deferred_set code)
71 * iii) copy the data block to a newly allocate block. This step can be
72 * missed out if the io covers the block. (schedule_copy).
74 * iv) insert the new mapping into the origin's btree
75 * (process_prepared_mapping). This act of inserting breaks some
76 * sharing of btree nodes between the two devices. Breaking sharing only
77 * effects the btree of that specific device. Btrees for the other
78 * devices that share the block never change. The btree for the origin
79 * device as it was after the last commit is untouched, ie. we're using
80 * persistent data structures in the functional programming sense.
82 * v) unplug io to this physical block, including the io that triggered
83 * the breaking of sharing.
85 * Steps (ii) and (iii) occur in parallel.
87 * The metadata _doesn't_ need to be committed before the io continues. We
88 * get away with this because the io is always written to a _new_ block.
89 * If there's a crash, then:
91 * - The origin mapping will point to the old origin block (the shared
92 * one). This will contain the data as it was before the io that triggered
93 * the breaking of sharing came in.
95 * - The snap mapping still points to the old block. As it would after
98 * The downside of this scheme is the timestamp magic isn't perfect, and
99 * will continue to think that data block in the snapshot device is shared
100 * even after the write to the origin has broken sharing. I suspect data
101 * blocks will typically be shared by many different devices, so we're
102 * breaking sharing n + 1 times, rather than n, where n is the number of
103 * devices that reference this data block. At the moment I think the
104 * benefits far, far outweigh the disadvantages.
107 /*----------------------------------------------------------------*/
112 static void build_data_key(struct dm_thin_device
*td
,
113 dm_block_t b
, struct dm_cell_key
*key
)
116 key
->dev
= dm_thin_dev_id(td
);
120 static void build_virtual_key(struct dm_thin_device
*td
, dm_block_t b
,
121 struct dm_cell_key
*key
)
124 key
->dev
= dm_thin_dev_id(td
);
128 /*----------------------------------------------------------------*/
131 * A pool device ties together a metadata device and a data device. It
132 * also provides the interface for creating and destroying internal
135 struct dm_thin_new_mapping
;
138 * The pool runs in 4 modes. Ordered in degraded order for comparisons.
141 PM_WRITE
, /* metadata may be changed */
142 PM_OUT_OF_DATA_SPACE
, /* metadata may be changed, though data may not be allocated */
143 PM_READ_ONLY
, /* metadata may not be changed */
144 PM_FAIL
, /* all I/O fails */
147 struct pool_features
{
150 bool zero_new_blocks
:1;
151 bool discard_enabled
:1;
152 bool discard_passdown
:1;
153 bool error_if_no_space
:1;
157 typedef void (*process_bio_fn
)(struct thin_c
*tc
, struct bio
*bio
);
158 typedef void (*process_mapping_fn
)(struct dm_thin_new_mapping
*m
);
161 struct list_head list
;
162 struct dm_target
*ti
; /* Only set if a pool target is bound */
164 struct mapped_device
*pool_md
;
165 struct block_device
*md_dev
;
166 struct dm_pool_metadata
*pmd
;
168 dm_block_t low_water_blocks
;
169 uint32_t sectors_per_block
;
170 int sectors_per_block_shift
;
172 struct pool_features pf
;
173 bool low_water_triggered
:1; /* A dm event has been sent */
175 struct dm_bio_prison
*prison
;
176 struct dm_kcopyd_client
*copier
;
178 struct workqueue_struct
*wq
;
179 struct work_struct worker
;
180 struct delayed_work waker
;
181 struct delayed_work no_space_timeout
;
183 unsigned long last_commit_jiffies
;
187 struct bio_list deferred_flush_bios
;
188 struct list_head prepared_mappings
;
189 struct list_head prepared_discards
;
190 struct list_head active_thins
;
192 struct dm_deferred_set
*shared_read_ds
;
193 struct dm_deferred_set
*all_io_ds
;
195 struct dm_thin_new_mapping
*next_mapping
;
196 mempool_t
*mapping_pool
;
198 process_bio_fn process_bio
;
199 process_bio_fn process_discard
;
201 process_mapping_fn process_prepared_mapping
;
202 process_mapping_fn process_prepared_discard
;
205 static enum pool_mode
get_pool_mode(struct pool
*pool
);
206 static void metadata_operation_failed(struct pool
*pool
, const char *op
, int r
);
209 * Target context for a pool.
212 struct dm_target
*ti
;
214 struct dm_dev
*data_dev
;
215 struct dm_dev
*metadata_dev
;
216 struct dm_target_callbacks callbacks
;
218 dm_block_t low_water_blocks
;
219 struct pool_features requested_pf
; /* Features requested during table load */
220 struct pool_features adjusted_pf
; /* Features used after adjusting for constituent devices */
224 * Target context for a thin.
227 struct list_head list
;
228 struct dm_dev
*pool_dev
;
229 struct dm_dev
*origin_dev
;
233 struct dm_thin_device
*td
;
236 struct bio_list deferred_bio_list
;
237 struct bio_list retry_on_resume_list
;
238 struct rb_root sort_bio_list
; /* sorted list of deferred bios */
241 * Ensures the thin is not destroyed until the worker has finished
242 * iterating the active_thins list.
245 struct completion can_destroy
;
248 /*----------------------------------------------------------------*/
251 * wake_worker() is used when new work is queued and when pool_resume is
252 * ready to continue deferred IO processing.
254 static void wake_worker(struct pool
*pool
)
256 queue_work(pool
->wq
, &pool
->worker
);
259 /*----------------------------------------------------------------*/
261 static int bio_detain(struct pool
*pool
, struct dm_cell_key
*key
, struct bio
*bio
,
262 struct dm_bio_prison_cell
**cell_result
)
265 struct dm_bio_prison_cell
*cell_prealloc
;
268 * Allocate a cell from the prison's mempool.
269 * This might block but it can't fail.
271 cell_prealloc
= dm_bio_prison_alloc_cell(pool
->prison
, GFP_NOIO
);
273 r
= dm_bio_detain(pool
->prison
, key
, bio
, cell_prealloc
, cell_result
);
276 * We reused an old cell; we can get rid of
279 dm_bio_prison_free_cell(pool
->prison
, cell_prealloc
);
284 static void cell_release(struct pool
*pool
,
285 struct dm_bio_prison_cell
*cell
,
286 struct bio_list
*bios
)
288 dm_cell_release(pool
->prison
, cell
, bios
);
289 dm_bio_prison_free_cell(pool
->prison
, cell
);
292 static void cell_release_no_holder(struct pool
*pool
,
293 struct dm_bio_prison_cell
*cell
,
294 struct bio_list
*bios
)
296 dm_cell_release_no_holder(pool
->prison
, cell
, bios
);
297 dm_bio_prison_free_cell(pool
->prison
, cell
);
300 static void cell_defer_no_holder_no_free(struct thin_c
*tc
,
301 struct dm_bio_prison_cell
*cell
)
303 struct pool
*pool
= tc
->pool
;
306 spin_lock_irqsave(&tc
->lock
, flags
);
307 dm_cell_release_no_holder(pool
->prison
, cell
, &tc
->deferred_bio_list
);
308 spin_unlock_irqrestore(&tc
->lock
, flags
);
313 static void cell_error(struct pool
*pool
,
314 struct dm_bio_prison_cell
*cell
)
316 dm_cell_error(pool
->prison
, cell
);
317 dm_bio_prison_free_cell(pool
->prison
, cell
);
320 /*----------------------------------------------------------------*/
323 * A global list of pools that uses a struct mapped_device as a key.
325 static struct dm_thin_pool_table
{
327 struct list_head pools
;
328 } dm_thin_pool_table
;
330 static void pool_table_init(void)
332 mutex_init(&dm_thin_pool_table
.mutex
);
333 INIT_LIST_HEAD(&dm_thin_pool_table
.pools
);
336 static void __pool_table_insert(struct pool
*pool
)
338 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
339 list_add(&pool
->list
, &dm_thin_pool_table
.pools
);
342 static void __pool_table_remove(struct pool
*pool
)
344 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
345 list_del(&pool
->list
);
348 static struct pool
*__pool_table_lookup(struct mapped_device
*md
)
350 struct pool
*pool
= NULL
, *tmp
;
352 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
354 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
355 if (tmp
->pool_md
== md
) {
364 static struct pool
*__pool_table_lookup_metadata_dev(struct block_device
*md_dev
)
366 struct pool
*pool
= NULL
, *tmp
;
368 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
370 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
371 if (tmp
->md_dev
== md_dev
) {
380 /*----------------------------------------------------------------*/
382 struct dm_thin_endio_hook
{
384 struct dm_deferred_entry
*shared_read_entry
;
385 struct dm_deferred_entry
*all_io_entry
;
386 struct dm_thin_new_mapping
*overwrite_mapping
;
387 struct rb_node rb_node
;
390 static void requeue_bio_list(struct thin_c
*tc
, struct bio_list
*master
)
393 struct bio_list bios
;
396 bio_list_init(&bios
);
398 spin_lock_irqsave(&tc
->lock
, flags
);
399 bio_list_merge(&bios
, master
);
400 bio_list_init(master
);
401 spin_unlock_irqrestore(&tc
->lock
, flags
);
403 while ((bio
= bio_list_pop(&bios
)))
404 bio_endio(bio
, DM_ENDIO_REQUEUE
);
407 static void requeue_io(struct thin_c
*tc
)
409 requeue_bio_list(tc
, &tc
->deferred_bio_list
);
410 requeue_bio_list(tc
, &tc
->retry_on_resume_list
);
413 static void error_thin_retry_list(struct thin_c
*tc
)
417 struct bio_list bios
;
419 bio_list_init(&bios
);
421 spin_lock_irqsave(&tc
->lock
, flags
);
422 bio_list_merge(&bios
, &tc
->retry_on_resume_list
);
423 bio_list_init(&tc
->retry_on_resume_list
);
424 spin_unlock_irqrestore(&tc
->lock
, flags
);
426 while ((bio
= bio_list_pop(&bios
)))
430 static void error_retry_list(struct pool
*pool
)
435 list_for_each_entry_rcu(tc
, &pool
->active_thins
, list
)
436 error_thin_retry_list(tc
);
441 * This section of code contains the logic for processing a thin device's IO.
442 * Much of the code depends on pool object resources (lists, workqueues, etc)
443 * but most is exclusively called from the thin target rather than the thin-pool
447 static bool block_size_is_power_of_two(struct pool
*pool
)
449 return pool
->sectors_per_block_shift
>= 0;
452 static dm_block_t
get_bio_block(struct thin_c
*tc
, struct bio
*bio
)
454 struct pool
*pool
= tc
->pool
;
455 sector_t block_nr
= bio
->bi_iter
.bi_sector
;
457 if (block_size_is_power_of_two(pool
))
458 block_nr
>>= pool
->sectors_per_block_shift
;
460 (void) sector_div(block_nr
, pool
->sectors_per_block
);
465 static void remap(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
)
467 struct pool
*pool
= tc
->pool
;
468 sector_t bi_sector
= bio
->bi_iter
.bi_sector
;
470 bio
->bi_bdev
= tc
->pool_dev
->bdev
;
471 if (block_size_is_power_of_two(pool
))
472 bio
->bi_iter
.bi_sector
=
473 (block
<< pool
->sectors_per_block_shift
) |
474 (bi_sector
& (pool
->sectors_per_block
- 1));
476 bio
->bi_iter
.bi_sector
= (block
* pool
->sectors_per_block
) +
477 sector_div(bi_sector
, pool
->sectors_per_block
);
480 static void remap_to_origin(struct thin_c
*tc
, struct bio
*bio
)
482 bio
->bi_bdev
= tc
->origin_dev
->bdev
;
485 static int bio_triggers_commit(struct thin_c
*tc
, struct bio
*bio
)
487 return (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) &&
488 dm_thin_changed_this_transaction(tc
->td
);
491 static void inc_all_io_entry(struct pool
*pool
, struct bio
*bio
)
493 struct dm_thin_endio_hook
*h
;
495 if (bio
->bi_rw
& REQ_DISCARD
)
498 h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
499 h
->all_io_entry
= dm_deferred_entry_inc(pool
->all_io_ds
);
502 static void issue(struct thin_c
*tc
, struct bio
*bio
)
504 struct pool
*pool
= tc
->pool
;
507 if (!bio_triggers_commit(tc
, bio
)) {
508 generic_make_request(bio
);
513 * Complete bio with an error if earlier I/O caused changes to
514 * the metadata that can't be committed e.g, due to I/O errors
515 * on the metadata device.
517 if (dm_thin_aborted_changes(tc
->td
)) {
523 * Batch together any bios that trigger commits and then issue a
524 * single commit for them in process_deferred_bios().
526 spin_lock_irqsave(&pool
->lock
, flags
);
527 bio_list_add(&pool
->deferred_flush_bios
, bio
);
528 spin_unlock_irqrestore(&pool
->lock
, flags
);
531 static void remap_to_origin_and_issue(struct thin_c
*tc
, struct bio
*bio
)
533 remap_to_origin(tc
, bio
);
537 static void remap_and_issue(struct thin_c
*tc
, struct bio
*bio
,
540 remap(tc
, bio
, block
);
544 /*----------------------------------------------------------------*/
547 * Bio endio functions.
549 struct dm_thin_new_mapping
{
550 struct list_head list
;
555 bool definitely_not_shared
:1;
559 dm_block_t virt_block
;
560 dm_block_t data_block
;
561 struct dm_bio_prison_cell
*cell
, *cell2
;
564 * If the bio covers the whole area of a block then we can avoid
565 * zeroing or copying. Instead this bio is hooked. The bio will
566 * still be in the cell, so care has to be taken to avoid issuing
570 bio_end_io_t
*saved_bi_end_io
;
573 static void __maybe_add_mapping(struct dm_thin_new_mapping
*m
)
575 struct pool
*pool
= m
->tc
->pool
;
577 if (m
->quiesced
&& m
->prepared
) {
578 list_add_tail(&m
->list
, &pool
->prepared_mappings
);
583 static void copy_complete(int read_err
, unsigned long write_err
, void *context
)
586 struct dm_thin_new_mapping
*m
= context
;
587 struct pool
*pool
= m
->tc
->pool
;
589 m
->err
= read_err
|| write_err
? -EIO
: 0;
591 spin_lock_irqsave(&pool
->lock
, flags
);
593 __maybe_add_mapping(m
);
594 spin_unlock_irqrestore(&pool
->lock
, flags
);
597 static void overwrite_endio(struct bio
*bio
, int err
)
600 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
601 struct dm_thin_new_mapping
*m
= h
->overwrite_mapping
;
602 struct pool
*pool
= m
->tc
->pool
;
606 spin_lock_irqsave(&pool
->lock
, flags
);
608 __maybe_add_mapping(m
);
609 spin_unlock_irqrestore(&pool
->lock
, flags
);
612 /*----------------------------------------------------------------*/
619 * Prepared mapping jobs.
623 * This sends the bios in the cell back to the deferred_bios list.
625 static void cell_defer(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
627 struct pool
*pool
= tc
->pool
;
630 spin_lock_irqsave(&tc
->lock
, flags
);
631 cell_release(pool
, cell
, &tc
->deferred_bio_list
);
632 spin_unlock_irqrestore(&tc
->lock
, flags
);
638 * Same as cell_defer above, except it omits the original holder of the cell.
640 static void cell_defer_no_holder(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
642 struct pool
*pool
= tc
->pool
;
645 spin_lock_irqsave(&tc
->lock
, flags
);
646 cell_release_no_holder(pool
, cell
, &tc
->deferred_bio_list
);
647 spin_unlock_irqrestore(&tc
->lock
, flags
);
652 static void process_prepared_mapping_fail(struct dm_thin_new_mapping
*m
)
655 m
->bio
->bi_end_io
= m
->saved_bi_end_io
;
656 atomic_inc(&m
->bio
->bi_remaining
);
658 cell_error(m
->tc
->pool
, m
->cell
);
660 mempool_free(m
, m
->tc
->pool
->mapping_pool
);
663 static void process_prepared_mapping(struct dm_thin_new_mapping
*m
)
665 struct thin_c
*tc
= m
->tc
;
666 struct pool
*pool
= tc
->pool
;
672 bio
->bi_end_io
= m
->saved_bi_end_io
;
673 atomic_inc(&bio
->bi_remaining
);
677 cell_error(pool
, m
->cell
);
682 * Commit the prepared block into the mapping btree.
683 * Any I/O for this block arriving after this point will get
684 * remapped to it directly.
686 r
= dm_thin_insert_block(tc
->td
, m
->virt_block
, m
->data_block
);
688 metadata_operation_failed(pool
, "dm_thin_insert_block", r
);
689 cell_error(pool
, m
->cell
);
694 * Release any bios held while the block was being provisioned.
695 * If we are processing a write bio that completely covers the block,
696 * we already processed it so can ignore it now when processing
697 * the bios in the cell.
700 cell_defer_no_holder(tc
, m
->cell
);
703 cell_defer(tc
, m
->cell
);
707 mempool_free(m
, pool
->mapping_pool
);
710 static void process_prepared_discard_fail(struct dm_thin_new_mapping
*m
)
712 struct thin_c
*tc
= m
->tc
;
714 bio_io_error(m
->bio
);
715 cell_defer_no_holder(tc
, m
->cell
);
716 cell_defer_no_holder(tc
, m
->cell2
);
717 mempool_free(m
, tc
->pool
->mapping_pool
);
720 static void process_prepared_discard_passdown(struct dm_thin_new_mapping
*m
)
722 struct thin_c
*tc
= m
->tc
;
724 inc_all_io_entry(tc
->pool
, m
->bio
);
725 cell_defer_no_holder(tc
, m
->cell
);
726 cell_defer_no_holder(tc
, m
->cell2
);
729 if (m
->definitely_not_shared
)
730 remap_and_issue(tc
, m
->bio
, m
->data_block
);
733 if (dm_pool_block_is_used(tc
->pool
->pmd
, m
->data_block
, &used
) || used
)
734 bio_endio(m
->bio
, 0);
736 remap_and_issue(tc
, m
->bio
, m
->data_block
);
739 bio_endio(m
->bio
, 0);
741 mempool_free(m
, tc
->pool
->mapping_pool
);
744 static void process_prepared_discard(struct dm_thin_new_mapping
*m
)
747 struct thin_c
*tc
= m
->tc
;
749 r
= dm_thin_remove_block(tc
->td
, m
->virt_block
);
751 DMERR_LIMIT("dm_thin_remove_block() failed");
753 process_prepared_discard_passdown(m
);
756 static void process_prepared(struct pool
*pool
, struct list_head
*head
,
757 process_mapping_fn
*fn
)
760 struct list_head maps
;
761 struct dm_thin_new_mapping
*m
, *tmp
;
763 INIT_LIST_HEAD(&maps
);
764 spin_lock_irqsave(&pool
->lock
, flags
);
765 list_splice_init(head
, &maps
);
766 spin_unlock_irqrestore(&pool
->lock
, flags
);
768 list_for_each_entry_safe(m
, tmp
, &maps
, list
)
775 static int io_overlaps_block(struct pool
*pool
, struct bio
*bio
)
777 return bio
->bi_iter
.bi_size
==
778 (pool
->sectors_per_block
<< SECTOR_SHIFT
);
781 static int io_overwrites_block(struct pool
*pool
, struct bio
*bio
)
783 return (bio_data_dir(bio
) == WRITE
) &&
784 io_overlaps_block(pool
, bio
);
787 static void save_and_set_endio(struct bio
*bio
, bio_end_io_t
**save
,
790 *save
= bio
->bi_end_io
;
794 static int ensure_next_mapping(struct pool
*pool
)
796 if (pool
->next_mapping
)
799 pool
->next_mapping
= mempool_alloc(pool
->mapping_pool
, GFP_ATOMIC
);
801 return pool
->next_mapping
? 0 : -ENOMEM
;
804 static struct dm_thin_new_mapping
*get_next_mapping(struct pool
*pool
)
806 struct dm_thin_new_mapping
*m
= pool
->next_mapping
;
808 BUG_ON(!pool
->next_mapping
);
810 memset(m
, 0, sizeof(struct dm_thin_new_mapping
));
811 INIT_LIST_HEAD(&m
->list
);
814 pool
->next_mapping
= NULL
;
819 static void schedule_copy(struct thin_c
*tc
, dm_block_t virt_block
,
820 struct dm_dev
*origin
, dm_block_t data_origin
,
821 dm_block_t data_dest
,
822 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
825 struct pool
*pool
= tc
->pool
;
826 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
829 m
->virt_block
= virt_block
;
830 m
->data_block
= data_dest
;
833 if (!dm_deferred_set_add_work(pool
->shared_read_ds
, &m
->list
))
837 * IO to pool_dev remaps to the pool target's data_dev.
839 * If the whole block of data is being overwritten, we can issue the
840 * bio immediately. Otherwise we use kcopyd to clone the data first.
842 if (io_overwrites_block(pool
, bio
)) {
843 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
845 h
->overwrite_mapping
= m
;
847 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
848 inc_all_io_entry(pool
, bio
);
849 remap_and_issue(tc
, bio
, data_dest
);
851 struct dm_io_region from
, to
;
853 from
.bdev
= origin
->bdev
;
854 from
.sector
= data_origin
* pool
->sectors_per_block
;
855 from
.count
= pool
->sectors_per_block
;
857 to
.bdev
= tc
->pool_dev
->bdev
;
858 to
.sector
= data_dest
* pool
->sectors_per_block
;
859 to
.count
= pool
->sectors_per_block
;
861 r
= dm_kcopyd_copy(pool
->copier
, &from
, 1, &to
,
862 0, copy_complete
, m
);
864 mempool_free(m
, pool
->mapping_pool
);
865 DMERR_LIMIT("dm_kcopyd_copy() failed");
866 cell_error(pool
, cell
);
871 static void schedule_internal_copy(struct thin_c
*tc
, dm_block_t virt_block
,
872 dm_block_t data_origin
, dm_block_t data_dest
,
873 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
875 schedule_copy(tc
, virt_block
, tc
->pool_dev
,
876 data_origin
, data_dest
, cell
, bio
);
879 static void schedule_external_copy(struct thin_c
*tc
, dm_block_t virt_block
,
880 dm_block_t data_dest
,
881 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
883 schedule_copy(tc
, virt_block
, tc
->origin_dev
,
884 virt_block
, data_dest
, cell
, bio
);
887 static void schedule_zero(struct thin_c
*tc
, dm_block_t virt_block
,
888 dm_block_t data_block
, struct dm_bio_prison_cell
*cell
,
891 struct pool
*pool
= tc
->pool
;
892 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
897 m
->virt_block
= virt_block
;
898 m
->data_block
= data_block
;
902 * If the whole block of data is being overwritten or we are not
903 * zeroing pre-existing data, we can issue the bio immediately.
904 * Otherwise we use kcopyd to zero the data first.
906 if (!pool
->pf
.zero_new_blocks
)
907 process_prepared_mapping(m
);
909 else if (io_overwrites_block(pool
, bio
)) {
910 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
912 h
->overwrite_mapping
= m
;
914 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
915 inc_all_io_entry(pool
, bio
);
916 remap_and_issue(tc
, bio
, data_block
);
919 struct dm_io_region to
;
921 to
.bdev
= tc
->pool_dev
->bdev
;
922 to
.sector
= data_block
* pool
->sectors_per_block
;
923 to
.count
= pool
->sectors_per_block
;
925 r
= dm_kcopyd_zero(pool
->copier
, 1, &to
, 0, copy_complete
, m
);
927 mempool_free(m
, pool
->mapping_pool
);
928 DMERR_LIMIT("dm_kcopyd_zero() failed");
929 cell_error(pool
, cell
);
935 * A non-zero return indicates read_only or fail_io mode.
936 * Many callers don't care about the return value.
938 static int commit(struct pool
*pool
)
942 if (get_pool_mode(pool
) >= PM_READ_ONLY
)
945 r
= dm_pool_commit_metadata(pool
->pmd
);
947 metadata_operation_failed(pool
, "dm_pool_commit_metadata", r
);
952 static void check_low_water_mark(struct pool
*pool
, dm_block_t free_blocks
)
956 if (free_blocks
<= pool
->low_water_blocks
&& !pool
->low_water_triggered
) {
957 DMWARN("%s: reached low water mark for data device: sending event.",
958 dm_device_name(pool
->pool_md
));
959 spin_lock_irqsave(&pool
->lock
, flags
);
960 pool
->low_water_triggered
= true;
961 spin_unlock_irqrestore(&pool
->lock
, flags
);
962 dm_table_event(pool
->ti
->table
);
966 static void set_pool_mode(struct pool
*pool
, enum pool_mode new_mode
);
968 static int alloc_data_block(struct thin_c
*tc
, dm_block_t
*result
)
971 dm_block_t free_blocks
;
972 struct pool
*pool
= tc
->pool
;
974 if (WARN_ON(get_pool_mode(pool
) != PM_WRITE
))
977 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
979 metadata_operation_failed(pool
, "dm_pool_get_free_block_count", r
);
983 check_low_water_mark(pool
, free_blocks
);
987 * Try to commit to see if that will free up some
994 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
996 metadata_operation_failed(pool
, "dm_pool_get_free_block_count", r
);
1001 set_pool_mode(pool
, PM_OUT_OF_DATA_SPACE
);
1006 r
= dm_pool_alloc_data_block(pool
->pmd
, result
);
1008 metadata_operation_failed(pool
, "dm_pool_alloc_data_block", r
);
1016 * If we have run out of space, queue bios until the device is
1017 * resumed, presumably after having been reloaded with more space.
1019 static void retry_on_resume(struct bio
*bio
)
1021 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1022 struct thin_c
*tc
= h
->tc
;
1023 unsigned long flags
;
1025 spin_lock_irqsave(&tc
->lock
, flags
);
1026 bio_list_add(&tc
->retry_on_resume_list
, bio
);
1027 spin_unlock_irqrestore(&tc
->lock
, flags
);
1030 static bool should_error_unserviceable_bio(struct pool
*pool
)
1032 enum pool_mode m
= get_pool_mode(pool
);
1036 /* Shouldn't get here */
1037 DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
1040 case PM_OUT_OF_DATA_SPACE
:
1041 return pool
->pf
.error_if_no_space
;
1047 /* Shouldn't get here */
1048 DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1053 static void handle_unserviceable_bio(struct pool
*pool
, struct bio
*bio
)
1055 if (should_error_unserviceable_bio(pool
))
1058 retry_on_resume(bio
);
1061 static void retry_bios_on_resume(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
1064 struct bio_list bios
;
1066 if (should_error_unserviceable_bio(pool
)) {
1067 cell_error(pool
, cell
);
1071 bio_list_init(&bios
);
1072 cell_release(pool
, cell
, &bios
);
1074 if (should_error_unserviceable_bio(pool
))
1075 while ((bio
= bio_list_pop(&bios
)))
1078 while ((bio
= bio_list_pop(&bios
)))
1079 retry_on_resume(bio
);
1082 static void process_discard(struct thin_c
*tc
, struct bio
*bio
)
1085 unsigned long flags
;
1086 struct pool
*pool
= tc
->pool
;
1087 struct dm_bio_prison_cell
*cell
, *cell2
;
1088 struct dm_cell_key key
, key2
;
1089 dm_block_t block
= get_bio_block(tc
, bio
);
1090 struct dm_thin_lookup_result lookup_result
;
1091 struct dm_thin_new_mapping
*m
;
1093 build_virtual_key(tc
->td
, block
, &key
);
1094 if (bio_detain(tc
->pool
, &key
, bio
, &cell
))
1097 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1101 * Check nobody is fiddling with this pool block. This can
1102 * happen if someone's in the process of breaking sharing
1105 build_data_key(tc
->td
, lookup_result
.block
, &key2
);
1106 if (bio_detain(tc
->pool
, &key2
, bio
, &cell2
)) {
1107 cell_defer_no_holder(tc
, cell
);
1111 if (io_overlaps_block(pool
, bio
)) {
1113 * IO may still be going to the destination block. We must
1114 * quiesce before we can do the removal.
1116 m
= get_next_mapping(pool
);
1118 m
->pass_discard
= pool
->pf
.discard_passdown
;
1119 m
->definitely_not_shared
= !lookup_result
.shared
;
1120 m
->virt_block
= block
;
1121 m
->data_block
= lookup_result
.block
;
1126 if (!dm_deferred_set_add_work(pool
->all_io_ds
, &m
->list
)) {
1127 spin_lock_irqsave(&pool
->lock
, flags
);
1128 list_add_tail(&m
->list
, &pool
->prepared_discards
);
1129 spin_unlock_irqrestore(&pool
->lock
, flags
);
1133 inc_all_io_entry(pool
, bio
);
1134 cell_defer_no_holder(tc
, cell
);
1135 cell_defer_no_holder(tc
, cell2
);
1138 * The DM core makes sure that the discard doesn't span
1139 * a block boundary. So we submit the discard of a
1140 * partial block appropriately.
1142 if ((!lookup_result
.shared
) && pool
->pf
.discard_passdown
)
1143 remap_and_issue(tc
, bio
, lookup_result
.block
);
1151 * It isn't provisioned, just forget it.
1153 cell_defer_no_holder(tc
, cell
);
1158 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1160 cell_defer_no_holder(tc
, cell
);
1166 static void break_sharing(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1167 struct dm_cell_key
*key
,
1168 struct dm_thin_lookup_result
*lookup_result
,
1169 struct dm_bio_prison_cell
*cell
)
1172 dm_block_t data_block
;
1173 struct pool
*pool
= tc
->pool
;
1175 r
= alloc_data_block(tc
, &data_block
);
1178 schedule_internal_copy(tc
, block
, lookup_result
->block
,
1179 data_block
, cell
, bio
);
1183 retry_bios_on_resume(pool
, cell
);
1187 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1189 cell_error(pool
, cell
);
1194 static void process_shared_bio(struct thin_c
*tc
, struct bio
*bio
,
1196 struct dm_thin_lookup_result
*lookup_result
)
1198 struct dm_bio_prison_cell
*cell
;
1199 struct pool
*pool
= tc
->pool
;
1200 struct dm_cell_key key
;
1203 * If cell is already occupied, then sharing is already in the process
1204 * of being broken so we have nothing further to do here.
1206 build_data_key(tc
->td
, lookup_result
->block
, &key
);
1207 if (bio_detain(pool
, &key
, bio
, &cell
))
1210 if (bio_data_dir(bio
) == WRITE
&& bio
->bi_iter
.bi_size
)
1211 break_sharing(tc
, bio
, block
, &key
, lookup_result
, cell
);
1213 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1215 h
->shared_read_entry
= dm_deferred_entry_inc(pool
->shared_read_ds
);
1216 inc_all_io_entry(pool
, bio
);
1217 cell_defer_no_holder(tc
, cell
);
1219 remap_and_issue(tc
, bio
, lookup_result
->block
);
1223 static void provision_block(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1224 struct dm_bio_prison_cell
*cell
)
1227 dm_block_t data_block
;
1228 struct pool
*pool
= tc
->pool
;
1231 * Remap empty bios (flushes) immediately, without provisioning.
1233 if (!bio
->bi_iter
.bi_size
) {
1234 inc_all_io_entry(pool
, bio
);
1235 cell_defer_no_holder(tc
, cell
);
1237 remap_and_issue(tc
, bio
, 0);
1242 * Fill read bios with zeroes and complete them immediately.
1244 if (bio_data_dir(bio
) == READ
) {
1246 cell_defer_no_holder(tc
, cell
);
1251 r
= alloc_data_block(tc
, &data_block
);
1255 schedule_external_copy(tc
, block
, data_block
, cell
, bio
);
1257 schedule_zero(tc
, block
, data_block
, cell
, bio
);
1261 retry_bios_on_resume(pool
, cell
);
1265 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1267 cell_error(pool
, cell
);
1272 static void process_bio(struct thin_c
*tc
, struct bio
*bio
)
1275 struct pool
*pool
= tc
->pool
;
1276 dm_block_t block
= get_bio_block(tc
, bio
);
1277 struct dm_bio_prison_cell
*cell
;
1278 struct dm_cell_key key
;
1279 struct dm_thin_lookup_result lookup_result
;
1282 * If cell is already occupied, then the block is already
1283 * being provisioned so we have nothing further to do here.
1285 build_virtual_key(tc
->td
, block
, &key
);
1286 if (bio_detain(pool
, &key
, bio
, &cell
))
1289 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1292 if (lookup_result
.shared
) {
1293 process_shared_bio(tc
, bio
, block
, &lookup_result
);
1294 cell_defer_no_holder(tc
, cell
); /* FIXME: pass this cell into process_shared? */
1296 inc_all_io_entry(pool
, bio
);
1297 cell_defer_no_holder(tc
, cell
);
1299 remap_and_issue(tc
, bio
, lookup_result
.block
);
1304 if (bio_data_dir(bio
) == READ
&& tc
->origin_dev
) {
1305 inc_all_io_entry(pool
, bio
);
1306 cell_defer_no_holder(tc
, cell
);
1308 remap_to_origin_and_issue(tc
, bio
);
1310 provision_block(tc
, bio
, block
, cell
);
1314 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1316 cell_defer_no_holder(tc
, cell
);
1322 static void process_bio_read_only(struct thin_c
*tc
, struct bio
*bio
)
1325 int rw
= bio_data_dir(bio
);
1326 dm_block_t block
= get_bio_block(tc
, bio
);
1327 struct dm_thin_lookup_result lookup_result
;
1329 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1332 if (lookup_result
.shared
&& (rw
== WRITE
) && bio
->bi_iter
.bi_size
)
1333 handle_unserviceable_bio(tc
->pool
, bio
);
1335 inc_all_io_entry(tc
->pool
, bio
);
1336 remap_and_issue(tc
, bio
, lookup_result
.block
);
1342 handle_unserviceable_bio(tc
->pool
, bio
);
1346 if (tc
->origin_dev
) {
1347 inc_all_io_entry(tc
->pool
, bio
);
1348 remap_to_origin_and_issue(tc
, bio
);
1357 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1364 static void process_bio_success(struct thin_c
*tc
, struct bio
*bio
)
1369 static void process_bio_fail(struct thin_c
*tc
, struct bio
*bio
)
1375 * FIXME: should we also commit due to size of transaction, measured in
1378 static int need_commit_due_to_time(struct pool
*pool
)
1380 return jiffies
< pool
->last_commit_jiffies
||
1381 jiffies
> pool
->last_commit_jiffies
+ COMMIT_PERIOD
;
1384 #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node)
1385 #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook))
1387 static void __thin_bio_rb_add(struct thin_c
*tc
, struct bio
*bio
)
1389 struct rb_node
**rbp
, *parent
;
1390 struct dm_thin_endio_hook
*pbd
;
1391 sector_t bi_sector
= bio
->bi_iter
.bi_sector
;
1393 rbp
= &tc
->sort_bio_list
.rb_node
;
1397 pbd
= thin_pbd(parent
);
1399 if (bi_sector
< thin_bio(pbd
)->bi_iter
.bi_sector
)
1400 rbp
= &(*rbp
)->rb_left
;
1402 rbp
= &(*rbp
)->rb_right
;
1405 pbd
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1406 rb_link_node(&pbd
->rb_node
, parent
, rbp
);
1407 rb_insert_color(&pbd
->rb_node
, &tc
->sort_bio_list
);
1410 static void __extract_sorted_bios(struct thin_c
*tc
)
1412 struct rb_node
*node
;
1413 struct dm_thin_endio_hook
*pbd
;
1416 for (node
= rb_first(&tc
->sort_bio_list
); node
; node
= rb_next(node
)) {
1417 pbd
= thin_pbd(node
);
1418 bio
= thin_bio(pbd
);
1420 bio_list_add(&tc
->deferred_bio_list
, bio
);
1421 rb_erase(&pbd
->rb_node
, &tc
->sort_bio_list
);
1424 WARN_ON(!RB_EMPTY_ROOT(&tc
->sort_bio_list
));
1427 static void __sort_thin_deferred_bios(struct thin_c
*tc
)
1430 struct bio_list bios
;
1432 bio_list_init(&bios
);
1433 bio_list_merge(&bios
, &tc
->deferred_bio_list
);
1434 bio_list_init(&tc
->deferred_bio_list
);
1436 /* Sort deferred_bio_list using rb-tree */
1437 while ((bio
= bio_list_pop(&bios
)))
1438 __thin_bio_rb_add(tc
, bio
);
1441 * Transfer the sorted bios in sort_bio_list back to
1442 * deferred_bio_list to allow lockless submission of
1445 __extract_sorted_bios(tc
);
1448 static void process_thin_deferred_bios(struct thin_c
*tc
)
1450 struct pool
*pool
= tc
->pool
;
1451 unsigned long flags
;
1453 struct bio_list bios
;
1454 struct blk_plug plug
;
1456 if (tc
->requeue_mode
) {
1457 requeue_bio_list(tc
, &tc
->deferred_bio_list
);
1461 bio_list_init(&bios
);
1463 spin_lock_irqsave(&tc
->lock
, flags
);
1465 if (bio_list_empty(&tc
->deferred_bio_list
)) {
1466 spin_unlock_irqrestore(&tc
->lock
, flags
);
1470 __sort_thin_deferred_bios(tc
);
1472 bio_list_merge(&bios
, &tc
->deferred_bio_list
);
1473 bio_list_init(&tc
->deferred_bio_list
);
1475 spin_unlock_irqrestore(&tc
->lock
, flags
);
1477 blk_start_plug(&plug
);
1478 while ((bio
= bio_list_pop(&bios
))) {
1480 * If we've got no free new_mapping structs, and processing
1481 * this bio might require one, we pause until there are some
1482 * prepared mappings to process.
1484 if (ensure_next_mapping(pool
)) {
1485 spin_lock_irqsave(&tc
->lock
, flags
);
1486 bio_list_add(&tc
->deferred_bio_list
, bio
);
1487 bio_list_merge(&tc
->deferred_bio_list
, &bios
);
1488 spin_unlock_irqrestore(&tc
->lock
, flags
);
1492 if (bio
->bi_rw
& REQ_DISCARD
)
1493 pool
->process_discard(tc
, bio
);
1495 pool
->process_bio(tc
, bio
);
1497 blk_finish_plug(&plug
);
1500 static void thin_get(struct thin_c
*tc
);
1501 static void thin_put(struct thin_c
*tc
);
1504 * We can't hold rcu_read_lock() around code that can block. So we
1505 * find a thin with the rcu lock held; bump a refcount; then drop
1508 static struct thin_c
*get_first_thin(struct pool
*pool
)
1510 struct thin_c
*tc
= NULL
;
1513 if (!list_empty(&pool
->active_thins
)) {
1514 tc
= list_entry_rcu(pool
->active_thins
.next
, struct thin_c
, list
);
1522 static struct thin_c
*get_next_thin(struct pool
*pool
, struct thin_c
*tc
)
1524 struct thin_c
*old_tc
= tc
;
1527 list_for_each_entry_continue_rcu(tc
, &pool
->active_thins
, list
) {
1539 static void process_deferred_bios(struct pool
*pool
)
1541 unsigned long flags
;
1543 struct bio_list bios
;
1546 tc
= get_first_thin(pool
);
1548 process_thin_deferred_bios(tc
);
1549 tc
= get_next_thin(pool
, tc
);
1553 * If there are any deferred flush bios, we must commit
1554 * the metadata before issuing them.
1556 bio_list_init(&bios
);
1557 spin_lock_irqsave(&pool
->lock
, flags
);
1558 bio_list_merge(&bios
, &pool
->deferred_flush_bios
);
1559 bio_list_init(&pool
->deferred_flush_bios
);
1560 spin_unlock_irqrestore(&pool
->lock
, flags
);
1562 if (bio_list_empty(&bios
) &&
1563 !(dm_pool_changed_this_transaction(pool
->pmd
) && need_commit_due_to_time(pool
)))
1567 while ((bio
= bio_list_pop(&bios
)))
1571 pool
->last_commit_jiffies
= jiffies
;
1573 while ((bio
= bio_list_pop(&bios
)))
1574 generic_make_request(bio
);
1577 static void do_worker(struct work_struct
*ws
)
1579 struct pool
*pool
= container_of(ws
, struct pool
, worker
);
1581 process_prepared(pool
, &pool
->prepared_mappings
, &pool
->process_prepared_mapping
);
1582 process_prepared(pool
, &pool
->prepared_discards
, &pool
->process_prepared_discard
);
1583 process_deferred_bios(pool
);
1587 * We want to commit periodically so that not too much
1588 * unwritten data builds up.
1590 static void do_waker(struct work_struct
*ws
)
1592 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
, waker
);
1594 queue_delayed_work(pool
->wq
, &pool
->waker
, COMMIT_PERIOD
);
1598 * We're holding onto IO to allow userland time to react. After the
1599 * timeout either the pool will have been resized (and thus back in
1600 * PM_WRITE mode), or we degrade to PM_READ_ONLY and start erroring IO.
1602 static void do_no_space_timeout(struct work_struct
*ws
)
1604 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
,
1607 if (get_pool_mode(pool
) == PM_OUT_OF_DATA_SPACE
&& !pool
->pf
.error_if_no_space
)
1608 set_pool_mode(pool
, PM_READ_ONLY
);
1611 /*----------------------------------------------------------------*/
1614 struct work_struct worker
;
1615 struct completion complete
;
1618 static struct pool_work
*to_pool_work(struct work_struct
*ws
)
1620 return container_of(ws
, struct pool_work
, worker
);
1623 static void pool_work_complete(struct pool_work
*pw
)
1625 complete(&pw
->complete
);
1628 static void pool_work_wait(struct pool_work
*pw
, struct pool
*pool
,
1629 void (*fn
)(struct work_struct
*))
1631 INIT_WORK_ONSTACK(&pw
->worker
, fn
);
1632 init_completion(&pw
->complete
);
1633 queue_work(pool
->wq
, &pw
->worker
);
1634 wait_for_completion(&pw
->complete
);
1637 /*----------------------------------------------------------------*/
1639 struct noflush_work
{
1640 struct pool_work pw
;
1644 static struct noflush_work
*to_noflush(struct work_struct
*ws
)
1646 return container_of(to_pool_work(ws
), struct noflush_work
, pw
);
1649 static void do_noflush_start(struct work_struct
*ws
)
1651 struct noflush_work
*w
= to_noflush(ws
);
1652 w
->tc
->requeue_mode
= true;
1654 pool_work_complete(&w
->pw
);
1657 static void do_noflush_stop(struct work_struct
*ws
)
1659 struct noflush_work
*w
= to_noflush(ws
);
1660 w
->tc
->requeue_mode
= false;
1661 pool_work_complete(&w
->pw
);
1664 static void noflush_work(struct thin_c
*tc
, void (*fn
)(struct work_struct
*))
1666 struct noflush_work w
;
1669 pool_work_wait(&w
.pw
, tc
->pool
, fn
);
1672 /*----------------------------------------------------------------*/
1674 static enum pool_mode
get_pool_mode(struct pool
*pool
)
1676 return pool
->pf
.mode
;
1679 static void notify_of_pool_mode_change(struct pool
*pool
, const char *new_mode
)
1681 dm_table_event(pool
->ti
->table
);
1682 DMINFO("%s: switching pool to %s mode",
1683 dm_device_name(pool
->pool_md
), new_mode
);
1686 static void set_pool_mode(struct pool
*pool
, enum pool_mode new_mode
)
1688 struct pool_c
*pt
= pool
->ti
->private;
1689 bool needs_check
= dm_pool_metadata_needs_check(pool
->pmd
);
1690 enum pool_mode old_mode
= get_pool_mode(pool
);
1691 unsigned long no_space_timeout
= ACCESS_ONCE(no_space_timeout_secs
) * HZ
;
1694 * Never allow the pool to transition to PM_WRITE mode if user
1695 * intervention is required to verify metadata and data consistency.
1697 if (new_mode
== PM_WRITE
&& needs_check
) {
1698 DMERR("%s: unable to switch pool to write mode until repaired.",
1699 dm_device_name(pool
->pool_md
));
1700 if (old_mode
!= new_mode
)
1701 new_mode
= old_mode
;
1703 new_mode
= PM_READ_ONLY
;
1706 * If we were in PM_FAIL mode, rollback of metadata failed. We're
1707 * not going to recover without a thin_repair. So we never let the
1708 * pool move out of the old mode.
1710 if (old_mode
== PM_FAIL
)
1711 new_mode
= old_mode
;
1715 if (old_mode
!= new_mode
)
1716 notify_of_pool_mode_change(pool
, "failure");
1717 dm_pool_metadata_read_only(pool
->pmd
);
1718 pool
->process_bio
= process_bio_fail
;
1719 pool
->process_discard
= process_bio_fail
;
1720 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1721 pool
->process_prepared_discard
= process_prepared_discard_fail
;
1723 error_retry_list(pool
);
1727 if (old_mode
!= new_mode
)
1728 notify_of_pool_mode_change(pool
, "read-only");
1729 dm_pool_metadata_read_only(pool
->pmd
);
1730 pool
->process_bio
= process_bio_read_only
;
1731 pool
->process_discard
= process_bio_success
;
1732 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1733 pool
->process_prepared_discard
= process_prepared_discard_passdown
;
1735 error_retry_list(pool
);
1738 case PM_OUT_OF_DATA_SPACE
:
1740 * Ideally we'd never hit this state; the low water mark
1741 * would trigger userland to extend the pool before we
1742 * completely run out of data space. However, many small
1743 * IOs to unprovisioned space can consume data space at an
1744 * alarming rate. Adjust your low water mark if you're
1745 * frequently seeing this mode.
1747 if (old_mode
!= new_mode
)
1748 notify_of_pool_mode_change(pool
, "out-of-data-space");
1749 pool
->process_bio
= process_bio_read_only
;
1750 pool
->process_discard
= process_discard
;
1751 pool
->process_prepared_mapping
= process_prepared_mapping
;
1752 pool
->process_prepared_discard
= process_prepared_discard_passdown
;
1754 if (!pool
->pf
.error_if_no_space
&& no_space_timeout
)
1755 queue_delayed_work(pool
->wq
, &pool
->no_space_timeout
, no_space_timeout
);
1759 if (old_mode
!= new_mode
)
1760 notify_of_pool_mode_change(pool
, "write");
1761 dm_pool_metadata_read_write(pool
->pmd
);
1762 pool
->process_bio
= process_bio
;
1763 pool
->process_discard
= process_discard
;
1764 pool
->process_prepared_mapping
= process_prepared_mapping
;
1765 pool
->process_prepared_discard
= process_prepared_discard
;
1769 pool
->pf
.mode
= new_mode
;
1771 * The pool mode may have changed, sync it so bind_control_target()
1772 * doesn't cause an unexpected mode transition on resume.
1774 pt
->adjusted_pf
.mode
= new_mode
;
1777 static void abort_transaction(struct pool
*pool
)
1779 const char *dev_name
= dm_device_name(pool
->pool_md
);
1781 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name
);
1782 if (dm_pool_abort_metadata(pool
->pmd
)) {
1783 DMERR("%s: failed to abort metadata transaction", dev_name
);
1784 set_pool_mode(pool
, PM_FAIL
);
1787 if (dm_pool_metadata_set_needs_check(pool
->pmd
)) {
1788 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name
);
1789 set_pool_mode(pool
, PM_FAIL
);
1793 static void metadata_operation_failed(struct pool
*pool
, const char *op
, int r
)
1795 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
1796 dm_device_name(pool
->pool_md
), op
, r
);
1798 abort_transaction(pool
);
1799 set_pool_mode(pool
, PM_READ_ONLY
);
1802 /*----------------------------------------------------------------*/
1805 * Mapping functions.
1809 * Called only while mapping a thin bio to hand it over to the workqueue.
1811 static void thin_defer_bio(struct thin_c
*tc
, struct bio
*bio
)
1813 unsigned long flags
;
1814 struct pool
*pool
= tc
->pool
;
1816 spin_lock_irqsave(&tc
->lock
, flags
);
1817 bio_list_add(&tc
->deferred_bio_list
, bio
);
1818 spin_unlock_irqrestore(&tc
->lock
, flags
);
1823 static void thin_hook_bio(struct thin_c
*tc
, struct bio
*bio
)
1825 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1828 h
->shared_read_entry
= NULL
;
1829 h
->all_io_entry
= NULL
;
1830 h
->overwrite_mapping
= NULL
;
1834 * Non-blocking function called from the thin target's map function.
1836 static int thin_bio_map(struct dm_target
*ti
, struct bio
*bio
)
1839 struct thin_c
*tc
= ti
->private;
1840 dm_block_t block
= get_bio_block(tc
, bio
);
1841 struct dm_thin_device
*td
= tc
->td
;
1842 struct dm_thin_lookup_result result
;
1843 struct dm_bio_prison_cell cell1
, cell2
;
1844 struct dm_bio_prison_cell
*cell_result
;
1845 struct dm_cell_key key
;
1847 thin_hook_bio(tc
, bio
);
1849 if (tc
->requeue_mode
) {
1850 bio_endio(bio
, DM_ENDIO_REQUEUE
);
1851 return DM_MAPIO_SUBMITTED
;
1854 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
1856 return DM_MAPIO_SUBMITTED
;
1859 if (bio
->bi_rw
& (REQ_DISCARD
| REQ_FLUSH
| REQ_FUA
)) {
1860 thin_defer_bio(tc
, bio
);
1861 return DM_MAPIO_SUBMITTED
;
1864 r
= dm_thin_find_block(td
, block
, 0, &result
);
1867 * Note that we defer readahead too.
1871 if (unlikely(result
.shared
)) {
1873 * We have a race condition here between the
1874 * result.shared value returned by the lookup and
1875 * snapshot creation, which may cause new
1878 * To avoid this always quiesce the origin before
1879 * taking the snap. You want to do this anyway to
1880 * ensure a consistent application view
1883 * More distant ancestors are irrelevant. The
1884 * shared flag will be set in their case.
1886 thin_defer_bio(tc
, bio
);
1887 return DM_MAPIO_SUBMITTED
;
1890 build_virtual_key(tc
->td
, block
, &key
);
1891 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell1
, &cell_result
))
1892 return DM_MAPIO_SUBMITTED
;
1894 build_data_key(tc
->td
, result
.block
, &key
);
1895 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell2
, &cell_result
)) {
1896 cell_defer_no_holder_no_free(tc
, &cell1
);
1897 return DM_MAPIO_SUBMITTED
;
1900 inc_all_io_entry(tc
->pool
, bio
);
1901 cell_defer_no_holder_no_free(tc
, &cell2
);
1902 cell_defer_no_holder_no_free(tc
, &cell1
);
1904 remap(tc
, bio
, result
.block
);
1905 return DM_MAPIO_REMAPPED
;
1908 if (get_pool_mode(tc
->pool
) == PM_READ_ONLY
) {
1910 * This block isn't provisioned, and we have no way
1913 handle_unserviceable_bio(tc
->pool
, bio
);
1914 return DM_MAPIO_SUBMITTED
;
1920 * In future, the failed dm_thin_find_block above could
1921 * provide the hint to load the metadata into cache.
1923 thin_defer_bio(tc
, bio
);
1924 return DM_MAPIO_SUBMITTED
;
1928 * Must always call bio_io_error on failure.
1929 * dm_thin_find_block can fail with -EINVAL if the
1930 * pool is switched to fail-io mode.
1933 return DM_MAPIO_SUBMITTED
;
1937 static int pool_is_congested(struct dm_target_callbacks
*cb
, int bdi_bits
)
1939 struct pool_c
*pt
= container_of(cb
, struct pool_c
, callbacks
);
1940 struct request_queue
*q
;
1942 if (get_pool_mode(pt
->pool
) == PM_OUT_OF_DATA_SPACE
)
1945 q
= bdev_get_queue(pt
->data_dev
->bdev
);
1946 return bdi_congested(&q
->backing_dev_info
, bdi_bits
);
1949 static void requeue_bios(struct pool
*pool
)
1951 unsigned long flags
;
1955 list_for_each_entry_rcu(tc
, &pool
->active_thins
, list
) {
1956 spin_lock_irqsave(&tc
->lock
, flags
);
1957 bio_list_merge(&tc
->deferred_bio_list
, &tc
->retry_on_resume_list
);
1958 bio_list_init(&tc
->retry_on_resume_list
);
1959 spin_unlock_irqrestore(&tc
->lock
, flags
);
1964 /*----------------------------------------------------------------
1965 * Binding of control targets to a pool object
1966 *--------------------------------------------------------------*/
1967 static bool data_dev_supports_discard(struct pool_c
*pt
)
1969 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1971 return q
&& blk_queue_discard(q
);
1974 static bool is_factor(sector_t block_size
, uint32_t n
)
1976 return !sector_div(block_size
, n
);
1980 * If discard_passdown was enabled verify that the data device
1981 * supports discards. Disable discard_passdown if not.
1983 static void disable_passdown_if_not_supported(struct pool_c
*pt
)
1985 struct pool
*pool
= pt
->pool
;
1986 struct block_device
*data_bdev
= pt
->data_dev
->bdev
;
1987 struct queue_limits
*data_limits
= &bdev_get_queue(data_bdev
)->limits
;
1988 sector_t block_size
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
1989 const char *reason
= NULL
;
1990 char buf
[BDEVNAME_SIZE
];
1992 if (!pt
->adjusted_pf
.discard_passdown
)
1995 if (!data_dev_supports_discard(pt
))
1996 reason
= "discard unsupported";
1998 else if (data_limits
->max_discard_sectors
< pool
->sectors_per_block
)
1999 reason
= "max discard sectors smaller than a block";
2001 else if (data_limits
->discard_granularity
> block_size
)
2002 reason
= "discard granularity larger than a block";
2004 else if (!is_factor(block_size
, data_limits
->discard_granularity
))
2005 reason
= "discard granularity not a factor of block size";
2008 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev
, buf
), reason
);
2009 pt
->adjusted_pf
.discard_passdown
= false;
2013 static int bind_control_target(struct pool
*pool
, struct dm_target
*ti
)
2015 struct pool_c
*pt
= ti
->private;
2018 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
2020 enum pool_mode old_mode
= get_pool_mode(pool
);
2021 enum pool_mode new_mode
= pt
->adjusted_pf
.mode
;
2024 * Don't change the pool's mode until set_pool_mode() below.
2025 * Otherwise the pool's process_* function pointers may
2026 * not match the desired pool mode.
2028 pt
->adjusted_pf
.mode
= old_mode
;
2031 pool
->pf
= pt
->adjusted_pf
;
2032 pool
->low_water_blocks
= pt
->low_water_blocks
;
2034 set_pool_mode(pool
, new_mode
);
2039 static void unbind_control_target(struct pool
*pool
, struct dm_target
*ti
)
2045 /*----------------------------------------------------------------
2047 *--------------------------------------------------------------*/
2048 /* Initialize pool features. */
2049 static void pool_features_init(struct pool_features
*pf
)
2051 pf
->mode
= PM_WRITE
;
2052 pf
->zero_new_blocks
= true;
2053 pf
->discard_enabled
= true;
2054 pf
->discard_passdown
= true;
2055 pf
->error_if_no_space
= false;
2058 static void __pool_destroy(struct pool
*pool
)
2060 __pool_table_remove(pool
);
2062 if (dm_pool_metadata_close(pool
->pmd
) < 0)
2063 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
2065 dm_bio_prison_destroy(pool
->prison
);
2066 dm_kcopyd_client_destroy(pool
->copier
);
2069 destroy_workqueue(pool
->wq
);
2071 if (pool
->next_mapping
)
2072 mempool_free(pool
->next_mapping
, pool
->mapping_pool
);
2073 mempool_destroy(pool
->mapping_pool
);
2074 dm_deferred_set_destroy(pool
->shared_read_ds
);
2075 dm_deferred_set_destroy(pool
->all_io_ds
);
2079 static struct kmem_cache
*_new_mapping_cache
;
2081 static struct pool
*pool_create(struct mapped_device
*pool_md
,
2082 struct block_device
*metadata_dev
,
2083 unsigned long block_size
,
2084 int read_only
, char **error
)
2089 struct dm_pool_metadata
*pmd
;
2090 bool format_device
= read_only
? false : true;
2092 pmd
= dm_pool_metadata_open(metadata_dev
, block_size
, format_device
);
2094 *error
= "Error creating metadata object";
2095 return (struct pool
*)pmd
;
2098 pool
= kmalloc(sizeof(*pool
), GFP_KERNEL
);
2100 *error
= "Error allocating memory for pool";
2101 err_p
= ERR_PTR(-ENOMEM
);
2106 pool
->sectors_per_block
= block_size
;
2107 if (block_size
& (block_size
- 1))
2108 pool
->sectors_per_block_shift
= -1;
2110 pool
->sectors_per_block_shift
= __ffs(block_size
);
2111 pool
->low_water_blocks
= 0;
2112 pool_features_init(&pool
->pf
);
2113 pool
->prison
= dm_bio_prison_create(PRISON_CELLS
);
2114 if (!pool
->prison
) {
2115 *error
= "Error creating pool's bio prison";
2116 err_p
= ERR_PTR(-ENOMEM
);
2120 pool
->copier
= dm_kcopyd_client_create(&dm_kcopyd_throttle
);
2121 if (IS_ERR(pool
->copier
)) {
2122 r
= PTR_ERR(pool
->copier
);
2123 *error
= "Error creating pool's kcopyd client";
2125 goto bad_kcopyd_client
;
2129 * Create singlethreaded workqueue that will service all devices
2130 * that use this metadata.
2132 pool
->wq
= alloc_ordered_workqueue("dm-" DM_MSG_PREFIX
, WQ_MEM_RECLAIM
);
2134 *error
= "Error creating pool's workqueue";
2135 err_p
= ERR_PTR(-ENOMEM
);
2139 INIT_WORK(&pool
->worker
, do_worker
);
2140 INIT_DELAYED_WORK(&pool
->waker
, do_waker
);
2141 INIT_DELAYED_WORK(&pool
->no_space_timeout
, do_no_space_timeout
);
2142 spin_lock_init(&pool
->lock
);
2143 bio_list_init(&pool
->deferred_flush_bios
);
2144 INIT_LIST_HEAD(&pool
->prepared_mappings
);
2145 INIT_LIST_HEAD(&pool
->prepared_discards
);
2146 INIT_LIST_HEAD(&pool
->active_thins
);
2147 pool
->low_water_triggered
= false;
2149 pool
->shared_read_ds
= dm_deferred_set_create();
2150 if (!pool
->shared_read_ds
) {
2151 *error
= "Error creating pool's shared read deferred set";
2152 err_p
= ERR_PTR(-ENOMEM
);
2153 goto bad_shared_read_ds
;
2156 pool
->all_io_ds
= dm_deferred_set_create();
2157 if (!pool
->all_io_ds
) {
2158 *error
= "Error creating pool's all io deferred set";
2159 err_p
= ERR_PTR(-ENOMEM
);
2163 pool
->next_mapping
= NULL
;
2164 pool
->mapping_pool
= mempool_create_slab_pool(MAPPING_POOL_SIZE
,
2165 _new_mapping_cache
);
2166 if (!pool
->mapping_pool
) {
2167 *error
= "Error creating pool's mapping mempool";
2168 err_p
= ERR_PTR(-ENOMEM
);
2169 goto bad_mapping_pool
;
2172 pool
->ref_count
= 1;
2173 pool
->last_commit_jiffies
= jiffies
;
2174 pool
->pool_md
= pool_md
;
2175 pool
->md_dev
= metadata_dev
;
2176 __pool_table_insert(pool
);
2181 dm_deferred_set_destroy(pool
->all_io_ds
);
2183 dm_deferred_set_destroy(pool
->shared_read_ds
);
2185 destroy_workqueue(pool
->wq
);
2187 dm_kcopyd_client_destroy(pool
->copier
);
2189 dm_bio_prison_destroy(pool
->prison
);
2193 if (dm_pool_metadata_close(pmd
))
2194 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
2199 static void __pool_inc(struct pool
*pool
)
2201 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
2205 static void __pool_dec(struct pool
*pool
)
2207 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
2208 BUG_ON(!pool
->ref_count
);
2209 if (!--pool
->ref_count
)
2210 __pool_destroy(pool
);
2213 static struct pool
*__pool_find(struct mapped_device
*pool_md
,
2214 struct block_device
*metadata_dev
,
2215 unsigned long block_size
, int read_only
,
2216 char **error
, int *created
)
2218 struct pool
*pool
= __pool_table_lookup_metadata_dev(metadata_dev
);
2221 if (pool
->pool_md
!= pool_md
) {
2222 *error
= "metadata device already in use by a pool";
2223 return ERR_PTR(-EBUSY
);
2228 pool
= __pool_table_lookup(pool_md
);
2230 if (pool
->md_dev
!= metadata_dev
) {
2231 *error
= "different pool cannot replace a pool";
2232 return ERR_PTR(-EINVAL
);
2237 pool
= pool_create(pool_md
, metadata_dev
, block_size
, read_only
, error
);
2245 /*----------------------------------------------------------------
2246 * Pool target methods
2247 *--------------------------------------------------------------*/
2248 static void pool_dtr(struct dm_target
*ti
)
2250 struct pool_c
*pt
= ti
->private;
2252 mutex_lock(&dm_thin_pool_table
.mutex
);
2254 unbind_control_target(pt
->pool
, ti
);
2255 __pool_dec(pt
->pool
);
2256 dm_put_device(ti
, pt
->metadata_dev
);
2257 dm_put_device(ti
, pt
->data_dev
);
2260 mutex_unlock(&dm_thin_pool_table
.mutex
);
2263 static int parse_pool_features(struct dm_arg_set
*as
, struct pool_features
*pf
,
2264 struct dm_target
*ti
)
2268 const char *arg_name
;
2270 static struct dm_arg _args
[] = {
2271 {0, 4, "Invalid number of pool feature arguments"},
2275 * No feature arguments supplied.
2280 r
= dm_read_arg_group(_args
, as
, &argc
, &ti
->error
);
2284 while (argc
&& !r
) {
2285 arg_name
= dm_shift_arg(as
);
2288 if (!strcasecmp(arg_name
, "skip_block_zeroing"))
2289 pf
->zero_new_blocks
= false;
2291 else if (!strcasecmp(arg_name
, "ignore_discard"))
2292 pf
->discard_enabled
= false;
2294 else if (!strcasecmp(arg_name
, "no_discard_passdown"))
2295 pf
->discard_passdown
= false;
2297 else if (!strcasecmp(arg_name
, "read_only"))
2298 pf
->mode
= PM_READ_ONLY
;
2300 else if (!strcasecmp(arg_name
, "error_if_no_space"))
2301 pf
->error_if_no_space
= true;
2304 ti
->error
= "Unrecognised pool feature requested";
2313 static void metadata_low_callback(void *context
)
2315 struct pool
*pool
= context
;
2317 DMWARN("%s: reached low water mark for metadata device: sending event.",
2318 dm_device_name(pool
->pool_md
));
2320 dm_table_event(pool
->ti
->table
);
2323 static sector_t
get_dev_size(struct block_device
*bdev
)
2325 return i_size_read(bdev
->bd_inode
) >> SECTOR_SHIFT
;
2328 static void warn_if_metadata_device_too_big(struct block_device
*bdev
)
2330 sector_t metadata_dev_size
= get_dev_size(bdev
);
2331 char buffer
[BDEVNAME_SIZE
];
2333 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS_WARNING
)
2334 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
2335 bdevname(bdev
, buffer
), THIN_METADATA_MAX_SECTORS
);
2338 static sector_t
get_metadata_dev_size(struct block_device
*bdev
)
2340 sector_t metadata_dev_size
= get_dev_size(bdev
);
2342 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS
)
2343 metadata_dev_size
= THIN_METADATA_MAX_SECTORS
;
2345 return metadata_dev_size
;
2348 static dm_block_t
get_metadata_dev_size_in_blocks(struct block_device
*bdev
)
2350 sector_t metadata_dev_size
= get_metadata_dev_size(bdev
);
2352 sector_div(metadata_dev_size
, THIN_METADATA_BLOCK_SIZE
);
2354 return metadata_dev_size
;
2358 * When a metadata threshold is crossed a dm event is triggered, and
2359 * userland should respond by growing the metadata device. We could let
2360 * userland set the threshold, like we do with the data threshold, but I'm
2361 * not sure they know enough to do this well.
2363 static dm_block_t
calc_metadata_threshold(struct pool_c
*pt
)
2366 * 4M is ample for all ops with the possible exception of thin
2367 * device deletion which is harmless if it fails (just retry the
2368 * delete after you've grown the device).
2370 dm_block_t quarter
= get_metadata_dev_size_in_blocks(pt
->metadata_dev
->bdev
) / 4;
2371 return min((dm_block_t
)1024ULL /* 4M */, quarter
);
2375 * thin-pool <metadata dev> <data dev>
2376 * <data block size (sectors)>
2377 * <low water mark (blocks)>
2378 * [<#feature args> [<arg>]*]
2380 * Optional feature arguments are:
2381 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
2382 * ignore_discard: disable discard
2383 * no_discard_passdown: don't pass discards down to the data device
2384 * read_only: Don't allow any changes to be made to the pool metadata.
2385 * error_if_no_space: error IOs, instead of queueing, if no space.
2387 static int pool_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2389 int r
, pool_created
= 0;
2392 struct pool_features pf
;
2393 struct dm_arg_set as
;
2394 struct dm_dev
*data_dev
;
2395 unsigned long block_size
;
2396 dm_block_t low_water_blocks
;
2397 struct dm_dev
*metadata_dev
;
2398 fmode_t metadata_mode
;
2401 * FIXME Remove validation from scope of lock.
2403 mutex_lock(&dm_thin_pool_table
.mutex
);
2406 ti
->error
= "Invalid argument count";
2415 * Set default pool features.
2417 pool_features_init(&pf
);
2419 dm_consume_args(&as
, 4);
2420 r
= parse_pool_features(&as
, &pf
, ti
);
2424 metadata_mode
= FMODE_READ
| ((pf
.mode
== PM_READ_ONLY
) ? 0 : FMODE_WRITE
);
2425 r
= dm_get_device(ti
, argv
[0], metadata_mode
, &metadata_dev
);
2427 ti
->error
= "Error opening metadata block device";
2430 warn_if_metadata_device_too_big(metadata_dev
->bdev
);
2432 r
= dm_get_device(ti
, argv
[1], FMODE_READ
| FMODE_WRITE
, &data_dev
);
2434 ti
->error
= "Error getting data device";
2438 if (kstrtoul(argv
[2], 10, &block_size
) || !block_size
||
2439 block_size
< DATA_DEV_BLOCK_SIZE_MIN_SECTORS
||
2440 block_size
> DATA_DEV_BLOCK_SIZE_MAX_SECTORS
||
2441 block_size
& (DATA_DEV_BLOCK_SIZE_MIN_SECTORS
- 1)) {
2442 ti
->error
= "Invalid block size";
2447 if (kstrtoull(argv
[3], 10, (unsigned long long *)&low_water_blocks
)) {
2448 ti
->error
= "Invalid low water mark";
2453 pt
= kzalloc(sizeof(*pt
), GFP_KERNEL
);
2459 pool
= __pool_find(dm_table_get_md(ti
->table
), metadata_dev
->bdev
,
2460 block_size
, pf
.mode
== PM_READ_ONLY
, &ti
->error
, &pool_created
);
2467 * 'pool_created' reflects whether this is the first table load.
2468 * Top level discard support is not allowed to be changed after
2469 * initial load. This would require a pool reload to trigger thin
2472 if (!pool_created
&& pf
.discard_enabled
!= pool
->pf
.discard_enabled
) {
2473 ti
->error
= "Discard support cannot be disabled once enabled";
2475 goto out_flags_changed
;
2480 pt
->metadata_dev
= metadata_dev
;
2481 pt
->data_dev
= data_dev
;
2482 pt
->low_water_blocks
= low_water_blocks
;
2483 pt
->adjusted_pf
= pt
->requested_pf
= pf
;
2484 ti
->num_flush_bios
= 1;
2487 * Only need to enable discards if the pool should pass
2488 * them down to the data device. The thin device's discard
2489 * processing will cause mappings to be removed from the btree.
2491 ti
->discard_zeroes_data_unsupported
= true;
2492 if (pf
.discard_enabled
&& pf
.discard_passdown
) {
2493 ti
->num_discard_bios
= 1;
2496 * Setting 'discards_supported' circumvents the normal
2497 * stacking of discard limits (this keeps the pool and
2498 * thin devices' discard limits consistent).
2500 ti
->discards_supported
= true;
2504 r
= dm_pool_register_metadata_threshold(pt
->pool
->pmd
,
2505 calc_metadata_threshold(pt
),
2506 metadata_low_callback
,
2511 pt
->callbacks
.congested_fn
= pool_is_congested
;
2512 dm_table_add_target_callbacks(ti
->table
, &pt
->callbacks
);
2514 mutex_unlock(&dm_thin_pool_table
.mutex
);
2523 dm_put_device(ti
, data_dev
);
2525 dm_put_device(ti
, metadata_dev
);
2527 mutex_unlock(&dm_thin_pool_table
.mutex
);
2532 static int pool_map(struct dm_target
*ti
, struct bio
*bio
)
2535 struct pool_c
*pt
= ti
->private;
2536 struct pool
*pool
= pt
->pool
;
2537 unsigned long flags
;
2540 * As this is a singleton target, ti->begin is always zero.
2542 spin_lock_irqsave(&pool
->lock
, flags
);
2543 bio
->bi_bdev
= pt
->data_dev
->bdev
;
2544 r
= DM_MAPIO_REMAPPED
;
2545 spin_unlock_irqrestore(&pool
->lock
, flags
);
2550 static int maybe_resize_data_dev(struct dm_target
*ti
, bool *need_commit
)
2553 struct pool_c
*pt
= ti
->private;
2554 struct pool
*pool
= pt
->pool
;
2555 sector_t data_size
= ti
->len
;
2556 dm_block_t sb_data_size
;
2558 *need_commit
= false;
2560 (void) sector_div(data_size
, pool
->sectors_per_block
);
2562 r
= dm_pool_get_data_dev_size(pool
->pmd
, &sb_data_size
);
2564 DMERR("%s: failed to retrieve data device size",
2565 dm_device_name(pool
->pool_md
));
2569 if (data_size
< sb_data_size
) {
2570 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
2571 dm_device_name(pool
->pool_md
),
2572 (unsigned long long)data_size
, sb_data_size
);
2575 } else if (data_size
> sb_data_size
) {
2576 if (dm_pool_metadata_needs_check(pool
->pmd
)) {
2577 DMERR("%s: unable to grow the data device until repaired.",
2578 dm_device_name(pool
->pool_md
));
2583 DMINFO("%s: growing the data device from %llu to %llu blocks",
2584 dm_device_name(pool
->pool_md
),
2585 sb_data_size
, (unsigned long long)data_size
);
2586 r
= dm_pool_resize_data_dev(pool
->pmd
, data_size
);
2588 metadata_operation_failed(pool
, "dm_pool_resize_data_dev", r
);
2592 *need_commit
= true;
2598 static int maybe_resize_metadata_dev(struct dm_target
*ti
, bool *need_commit
)
2601 struct pool_c
*pt
= ti
->private;
2602 struct pool
*pool
= pt
->pool
;
2603 dm_block_t metadata_dev_size
, sb_metadata_dev_size
;
2605 *need_commit
= false;
2607 metadata_dev_size
= get_metadata_dev_size_in_blocks(pool
->md_dev
);
2609 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &sb_metadata_dev_size
);
2611 DMERR("%s: failed to retrieve metadata device size",
2612 dm_device_name(pool
->pool_md
));
2616 if (metadata_dev_size
< sb_metadata_dev_size
) {
2617 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
2618 dm_device_name(pool
->pool_md
),
2619 metadata_dev_size
, sb_metadata_dev_size
);
2622 } else if (metadata_dev_size
> sb_metadata_dev_size
) {
2623 if (dm_pool_metadata_needs_check(pool
->pmd
)) {
2624 DMERR("%s: unable to grow the metadata device until repaired.",
2625 dm_device_name(pool
->pool_md
));
2629 warn_if_metadata_device_too_big(pool
->md_dev
);
2630 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
2631 dm_device_name(pool
->pool_md
),
2632 sb_metadata_dev_size
, metadata_dev_size
);
2633 r
= dm_pool_resize_metadata_dev(pool
->pmd
, metadata_dev_size
);
2635 metadata_operation_failed(pool
, "dm_pool_resize_metadata_dev", r
);
2639 *need_commit
= true;
2646 * Retrieves the number of blocks of the data device from
2647 * the superblock and compares it to the actual device size,
2648 * thus resizing the data device in case it has grown.
2650 * This both copes with opening preallocated data devices in the ctr
2651 * being followed by a resume
2653 * calling the resume method individually after userspace has
2654 * grown the data device in reaction to a table event.
2656 static int pool_preresume(struct dm_target
*ti
)
2659 bool need_commit1
, need_commit2
;
2660 struct pool_c
*pt
= ti
->private;
2661 struct pool
*pool
= pt
->pool
;
2664 * Take control of the pool object.
2666 r
= bind_control_target(pool
, ti
);
2670 r
= maybe_resize_data_dev(ti
, &need_commit1
);
2674 r
= maybe_resize_metadata_dev(ti
, &need_commit2
);
2678 if (need_commit1
|| need_commit2
)
2679 (void) commit(pool
);
2684 static void pool_resume(struct dm_target
*ti
)
2686 struct pool_c
*pt
= ti
->private;
2687 struct pool
*pool
= pt
->pool
;
2688 unsigned long flags
;
2690 spin_lock_irqsave(&pool
->lock
, flags
);
2691 pool
->low_water_triggered
= false;
2692 spin_unlock_irqrestore(&pool
->lock
, flags
);
2695 do_waker(&pool
->waker
.work
);
2698 static void pool_postsuspend(struct dm_target
*ti
)
2700 struct pool_c
*pt
= ti
->private;
2701 struct pool
*pool
= pt
->pool
;
2703 cancel_delayed_work(&pool
->waker
);
2704 cancel_delayed_work(&pool
->no_space_timeout
);
2705 flush_workqueue(pool
->wq
);
2706 (void) commit(pool
);
2709 static int check_arg_count(unsigned argc
, unsigned args_required
)
2711 if (argc
!= args_required
) {
2712 DMWARN("Message received with %u arguments instead of %u.",
2713 argc
, args_required
);
2720 static int read_dev_id(char *arg
, dm_thin_id
*dev_id
, int warning
)
2722 if (!kstrtoull(arg
, 10, (unsigned long long *)dev_id
) &&
2723 *dev_id
<= MAX_DEV_ID
)
2727 DMWARN("Message received with invalid device id: %s", arg
);
2732 static int process_create_thin_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2737 r
= check_arg_count(argc
, 2);
2741 r
= read_dev_id(argv
[1], &dev_id
, 1);
2745 r
= dm_pool_create_thin(pool
->pmd
, dev_id
);
2747 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2755 static int process_create_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2758 dm_thin_id origin_dev_id
;
2761 r
= check_arg_count(argc
, 3);
2765 r
= read_dev_id(argv
[1], &dev_id
, 1);
2769 r
= read_dev_id(argv
[2], &origin_dev_id
, 1);
2773 r
= dm_pool_create_snap(pool
->pmd
, dev_id
, origin_dev_id
);
2775 DMWARN("Creation of new snapshot %s of device %s failed.",
2783 static int process_delete_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2788 r
= check_arg_count(argc
, 2);
2792 r
= read_dev_id(argv
[1], &dev_id
, 1);
2796 r
= dm_pool_delete_thin_device(pool
->pmd
, dev_id
);
2798 DMWARN("Deletion of thin device %s failed.", argv
[1]);
2803 static int process_set_transaction_id_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2805 dm_thin_id old_id
, new_id
;
2808 r
= check_arg_count(argc
, 3);
2812 if (kstrtoull(argv
[1], 10, (unsigned long long *)&old_id
)) {
2813 DMWARN("set_transaction_id message: Unrecognised id %s.", argv
[1]);
2817 if (kstrtoull(argv
[2], 10, (unsigned long long *)&new_id
)) {
2818 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv
[2]);
2822 r
= dm_pool_set_metadata_transaction_id(pool
->pmd
, old_id
, new_id
);
2824 DMWARN("Failed to change transaction id from %s to %s.",
2832 static int process_reserve_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2836 r
= check_arg_count(argc
, 1);
2840 (void) commit(pool
);
2842 r
= dm_pool_reserve_metadata_snap(pool
->pmd
);
2844 DMWARN("reserve_metadata_snap message failed.");
2849 static int process_release_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2853 r
= check_arg_count(argc
, 1);
2857 r
= dm_pool_release_metadata_snap(pool
->pmd
);
2859 DMWARN("release_metadata_snap message failed.");
2865 * Messages supported:
2866 * create_thin <dev_id>
2867 * create_snap <dev_id> <origin_id>
2869 * trim <dev_id> <new_size_in_sectors>
2870 * set_transaction_id <current_trans_id> <new_trans_id>
2871 * reserve_metadata_snap
2872 * release_metadata_snap
2874 static int pool_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
2877 struct pool_c
*pt
= ti
->private;
2878 struct pool
*pool
= pt
->pool
;
2880 if (!strcasecmp(argv
[0], "create_thin"))
2881 r
= process_create_thin_mesg(argc
, argv
, pool
);
2883 else if (!strcasecmp(argv
[0], "create_snap"))
2884 r
= process_create_snap_mesg(argc
, argv
, pool
);
2886 else if (!strcasecmp(argv
[0], "delete"))
2887 r
= process_delete_mesg(argc
, argv
, pool
);
2889 else if (!strcasecmp(argv
[0], "set_transaction_id"))
2890 r
= process_set_transaction_id_mesg(argc
, argv
, pool
);
2892 else if (!strcasecmp(argv
[0], "reserve_metadata_snap"))
2893 r
= process_reserve_metadata_snap_mesg(argc
, argv
, pool
);
2895 else if (!strcasecmp(argv
[0], "release_metadata_snap"))
2896 r
= process_release_metadata_snap_mesg(argc
, argv
, pool
);
2899 DMWARN("Unrecognised thin pool target message received: %s", argv
[0]);
2902 (void) commit(pool
);
2907 static void emit_flags(struct pool_features
*pf
, char *result
,
2908 unsigned sz
, unsigned maxlen
)
2910 unsigned count
= !pf
->zero_new_blocks
+ !pf
->discard_enabled
+
2911 !pf
->discard_passdown
+ (pf
->mode
== PM_READ_ONLY
) +
2912 pf
->error_if_no_space
;
2913 DMEMIT("%u ", count
);
2915 if (!pf
->zero_new_blocks
)
2916 DMEMIT("skip_block_zeroing ");
2918 if (!pf
->discard_enabled
)
2919 DMEMIT("ignore_discard ");
2921 if (!pf
->discard_passdown
)
2922 DMEMIT("no_discard_passdown ");
2924 if (pf
->mode
== PM_READ_ONLY
)
2925 DMEMIT("read_only ");
2927 if (pf
->error_if_no_space
)
2928 DMEMIT("error_if_no_space ");
2933 * <transaction id> <used metadata sectors>/<total metadata sectors>
2934 * <used data sectors>/<total data sectors> <held metadata root>
2936 static void pool_status(struct dm_target
*ti
, status_type_t type
,
2937 unsigned status_flags
, char *result
, unsigned maxlen
)
2941 uint64_t transaction_id
;
2942 dm_block_t nr_free_blocks_data
;
2943 dm_block_t nr_free_blocks_metadata
;
2944 dm_block_t nr_blocks_data
;
2945 dm_block_t nr_blocks_metadata
;
2946 dm_block_t held_root
;
2947 char buf
[BDEVNAME_SIZE
];
2948 char buf2
[BDEVNAME_SIZE
];
2949 struct pool_c
*pt
= ti
->private;
2950 struct pool
*pool
= pt
->pool
;
2953 case STATUSTYPE_INFO
:
2954 if (get_pool_mode(pool
) == PM_FAIL
) {
2959 /* Commit to ensure statistics aren't out-of-date */
2960 if (!(status_flags
& DM_STATUS_NOFLUSH_FLAG
) && !dm_suspended(ti
))
2961 (void) commit(pool
);
2963 r
= dm_pool_get_metadata_transaction_id(pool
->pmd
, &transaction_id
);
2965 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
2966 dm_device_name(pool
->pool_md
), r
);
2970 r
= dm_pool_get_free_metadata_block_count(pool
->pmd
, &nr_free_blocks_metadata
);
2972 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
2973 dm_device_name(pool
->pool_md
), r
);
2977 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &nr_blocks_metadata
);
2979 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
2980 dm_device_name(pool
->pool_md
), r
);
2984 r
= dm_pool_get_free_block_count(pool
->pmd
, &nr_free_blocks_data
);
2986 DMERR("%s: dm_pool_get_free_block_count returned %d",
2987 dm_device_name(pool
->pool_md
), r
);
2991 r
= dm_pool_get_data_dev_size(pool
->pmd
, &nr_blocks_data
);
2993 DMERR("%s: dm_pool_get_data_dev_size returned %d",
2994 dm_device_name(pool
->pool_md
), r
);
2998 r
= dm_pool_get_metadata_snap(pool
->pmd
, &held_root
);
3000 DMERR("%s: dm_pool_get_metadata_snap returned %d",
3001 dm_device_name(pool
->pool_md
), r
);
3005 DMEMIT("%llu %llu/%llu %llu/%llu ",
3006 (unsigned long long)transaction_id
,
3007 (unsigned long long)(nr_blocks_metadata
- nr_free_blocks_metadata
),
3008 (unsigned long long)nr_blocks_metadata
,
3009 (unsigned long long)(nr_blocks_data
- nr_free_blocks_data
),
3010 (unsigned long long)nr_blocks_data
);
3013 DMEMIT("%llu ", held_root
);
3017 if (pool
->pf
.mode
== PM_OUT_OF_DATA_SPACE
)
3018 DMEMIT("out_of_data_space ");
3019 else if (pool
->pf
.mode
== PM_READ_ONLY
)
3024 if (!pool
->pf
.discard_enabled
)
3025 DMEMIT("ignore_discard ");
3026 else if (pool
->pf
.discard_passdown
)
3027 DMEMIT("discard_passdown ");
3029 DMEMIT("no_discard_passdown ");
3031 if (pool
->pf
.error_if_no_space
)
3032 DMEMIT("error_if_no_space ");
3034 DMEMIT("queue_if_no_space ");
3038 case STATUSTYPE_TABLE
:
3039 DMEMIT("%s %s %lu %llu ",
3040 format_dev_t(buf
, pt
->metadata_dev
->bdev
->bd_dev
),
3041 format_dev_t(buf2
, pt
->data_dev
->bdev
->bd_dev
),
3042 (unsigned long)pool
->sectors_per_block
,
3043 (unsigned long long)pt
->low_water_blocks
);
3044 emit_flags(&pt
->requested_pf
, result
, sz
, maxlen
);
3053 static int pool_iterate_devices(struct dm_target
*ti
,
3054 iterate_devices_callout_fn fn
, void *data
)
3056 struct pool_c
*pt
= ti
->private;
3058 return fn(ti
, pt
->data_dev
, 0, ti
->len
, data
);
3061 static int pool_merge(struct dm_target
*ti
, struct bvec_merge_data
*bvm
,
3062 struct bio_vec
*biovec
, int max_size
)
3064 struct pool_c
*pt
= ti
->private;
3065 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
3067 if (!q
->merge_bvec_fn
)
3070 bvm
->bi_bdev
= pt
->data_dev
->bdev
;
3072 return min(max_size
, q
->merge_bvec_fn(q
, bvm
, biovec
));
3075 static void set_discard_limits(struct pool_c
*pt
, struct queue_limits
*limits
)
3077 struct pool
*pool
= pt
->pool
;
3078 struct queue_limits
*data_limits
;
3080 limits
->max_discard_sectors
= pool
->sectors_per_block
;
3083 * discard_granularity is just a hint, and not enforced.
3085 if (pt
->adjusted_pf
.discard_passdown
) {
3086 data_limits
= &bdev_get_queue(pt
->data_dev
->bdev
)->limits
;
3087 limits
->discard_granularity
= data_limits
->discard_granularity
;
3089 limits
->discard_granularity
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
3092 static void pool_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
3094 struct pool_c
*pt
= ti
->private;
3095 struct pool
*pool
= pt
->pool
;
3096 uint64_t io_opt_sectors
= limits
->io_opt
>> SECTOR_SHIFT
;
3099 * If the system-determined stacked limits are compatible with the
3100 * pool's blocksize (io_opt is a factor) do not override them.
3102 if (io_opt_sectors
< pool
->sectors_per_block
||
3103 do_div(io_opt_sectors
, pool
->sectors_per_block
)) {
3104 blk_limits_io_min(limits
, 0);
3105 blk_limits_io_opt(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
3109 * pt->adjusted_pf is a staging area for the actual features to use.
3110 * They get transferred to the live pool in bind_control_target()
3111 * called from pool_preresume().
3113 if (!pt
->adjusted_pf
.discard_enabled
) {
3115 * Must explicitly disallow stacking discard limits otherwise the
3116 * block layer will stack them if pool's data device has support.
3117 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
3118 * user to see that, so make sure to set all discard limits to 0.
3120 limits
->discard_granularity
= 0;
3124 disable_passdown_if_not_supported(pt
);
3126 set_discard_limits(pt
, limits
);
3129 static struct target_type pool_target
= {
3130 .name
= "thin-pool",
3131 .features
= DM_TARGET_SINGLETON
| DM_TARGET_ALWAYS_WRITEABLE
|
3132 DM_TARGET_IMMUTABLE
,
3133 .version
= {1, 12, 0},
3134 .module
= THIS_MODULE
,
3138 .postsuspend
= pool_postsuspend
,
3139 .preresume
= pool_preresume
,
3140 .resume
= pool_resume
,
3141 .message
= pool_message
,
3142 .status
= pool_status
,
3143 .merge
= pool_merge
,
3144 .iterate_devices
= pool_iterate_devices
,
3145 .io_hints
= pool_io_hints
,
3148 /*----------------------------------------------------------------
3149 * Thin target methods
3150 *--------------------------------------------------------------*/
3151 static void thin_get(struct thin_c
*tc
)
3153 atomic_inc(&tc
->refcount
);
3156 static void thin_put(struct thin_c
*tc
)
3158 if (atomic_dec_and_test(&tc
->refcount
))
3159 complete(&tc
->can_destroy
);
3162 static void thin_dtr(struct dm_target
*ti
)
3164 struct thin_c
*tc
= ti
->private;
3165 unsigned long flags
;
3168 wait_for_completion(&tc
->can_destroy
);
3170 spin_lock_irqsave(&tc
->pool
->lock
, flags
);
3171 list_del_rcu(&tc
->list
);
3172 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
3175 mutex_lock(&dm_thin_pool_table
.mutex
);
3177 __pool_dec(tc
->pool
);
3178 dm_pool_close_thin_device(tc
->td
);
3179 dm_put_device(ti
, tc
->pool_dev
);
3181 dm_put_device(ti
, tc
->origin_dev
);
3184 mutex_unlock(&dm_thin_pool_table
.mutex
);
3188 * Thin target parameters:
3190 * <pool_dev> <dev_id> [origin_dev]
3192 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
3193 * dev_id: the internal device identifier
3194 * origin_dev: a device external to the pool that should act as the origin
3196 * If the pool device has discards disabled, they get disabled for the thin
3199 static int thin_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
3203 struct dm_dev
*pool_dev
, *origin_dev
;
3204 struct mapped_device
*pool_md
;
3205 unsigned long flags
;
3207 mutex_lock(&dm_thin_pool_table
.mutex
);
3209 if (argc
!= 2 && argc
!= 3) {
3210 ti
->error
= "Invalid argument count";
3215 tc
= ti
->private = kzalloc(sizeof(*tc
), GFP_KERNEL
);
3217 ti
->error
= "Out of memory";
3221 spin_lock_init(&tc
->lock
);
3222 bio_list_init(&tc
->deferred_bio_list
);
3223 bio_list_init(&tc
->retry_on_resume_list
);
3224 tc
->sort_bio_list
= RB_ROOT
;
3227 r
= dm_get_device(ti
, argv
[2], FMODE_READ
, &origin_dev
);
3229 ti
->error
= "Error opening origin device";
3230 goto bad_origin_dev
;
3232 tc
->origin_dev
= origin_dev
;
3235 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &pool_dev
);
3237 ti
->error
= "Error opening pool device";
3240 tc
->pool_dev
= pool_dev
;
3242 if (read_dev_id(argv
[1], (unsigned long long *)&tc
->dev_id
, 0)) {
3243 ti
->error
= "Invalid device id";
3248 pool_md
= dm_get_md(tc
->pool_dev
->bdev
->bd_dev
);
3250 ti
->error
= "Couldn't get pool mapped device";
3255 tc
->pool
= __pool_table_lookup(pool_md
);
3257 ti
->error
= "Couldn't find pool object";
3259 goto bad_pool_lookup
;
3261 __pool_inc(tc
->pool
);
3263 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
3264 ti
->error
= "Couldn't open thin device, Pool is in fail mode";
3269 r
= dm_pool_open_thin_device(tc
->pool
->pmd
, tc
->dev_id
, &tc
->td
);
3271 ti
->error
= "Couldn't open thin internal device";
3275 r
= dm_set_target_max_io_len(ti
, tc
->pool
->sectors_per_block
);
3277 goto bad_target_max_io_len
;
3279 ti
->num_flush_bios
= 1;
3280 ti
->flush_supported
= true;
3281 ti
->per_bio_data_size
= sizeof(struct dm_thin_endio_hook
);
3283 /* In case the pool supports discards, pass them on. */
3284 ti
->discard_zeroes_data_unsupported
= true;
3285 if (tc
->pool
->pf
.discard_enabled
) {
3286 ti
->discards_supported
= true;
3287 ti
->num_discard_bios
= 1;
3288 /* Discard bios must be split on a block boundary */
3289 ti
->split_discard_bios
= true;
3294 mutex_unlock(&dm_thin_pool_table
.mutex
);
3296 atomic_set(&tc
->refcount
, 1);
3297 init_completion(&tc
->can_destroy
);
3299 spin_lock_irqsave(&tc
->pool
->lock
, flags
);
3300 list_add_tail_rcu(&tc
->list
, &tc
->pool
->active_thins
);
3301 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
3303 * This synchronize_rcu() call is needed here otherwise we risk a
3304 * wake_worker() call finding no bios to process (because the newly
3305 * added tc isn't yet visible). So this reduces latency since we
3306 * aren't then dependent on the periodic commit to wake_worker().
3312 bad_target_max_io_len
:
3313 dm_pool_close_thin_device(tc
->td
);
3315 __pool_dec(tc
->pool
);
3319 dm_put_device(ti
, tc
->pool_dev
);
3322 dm_put_device(ti
, tc
->origin_dev
);
3326 mutex_unlock(&dm_thin_pool_table
.mutex
);
3331 static int thin_map(struct dm_target
*ti
, struct bio
*bio
)
3333 bio
->bi_iter
.bi_sector
= dm_target_offset(ti
, bio
->bi_iter
.bi_sector
);
3335 return thin_bio_map(ti
, bio
);
3338 static int thin_endio(struct dm_target
*ti
, struct bio
*bio
, int err
)
3340 unsigned long flags
;
3341 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
3342 struct list_head work
;
3343 struct dm_thin_new_mapping
*m
, *tmp
;
3344 struct pool
*pool
= h
->tc
->pool
;
3346 if (h
->shared_read_entry
) {
3347 INIT_LIST_HEAD(&work
);
3348 dm_deferred_entry_dec(h
->shared_read_entry
, &work
);
3350 spin_lock_irqsave(&pool
->lock
, flags
);
3351 list_for_each_entry_safe(m
, tmp
, &work
, list
) {
3354 __maybe_add_mapping(m
);
3356 spin_unlock_irqrestore(&pool
->lock
, flags
);
3359 if (h
->all_io_entry
) {
3360 INIT_LIST_HEAD(&work
);
3361 dm_deferred_entry_dec(h
->all_io_entry
, &work
);
3362 if (!list_empty(&work
)) {
3363 spin_lock_irqsave(&pool
->lock
, flags
);
3364 list_for_each_entry_safe(m
, tmp
, &work
, list
)
3365 list_add_tail(&m
->list
, &pool
->prepared_discards
);
3366 spin_unlock_irqrestore(&pool
->lock
, flags
);
3374 static void thin_presuspend(struct dm_target
*ti
)
3376 struct thin_c
*tc
= ti
->private;
3378 if (dm_noflush_suspending(ti
))
3379 noflush_work(tc
, do_noflush_start
);
3382 static void thin_postsuspend(struct dm_target
*ti
)
3384 struct thin_c
*tc
= ti
->private;
3387 * The dm_noflush_suspending flag has been cleared by now, so
3388 * unfortunately we must always run this.
3390 noflush_work(tc
, do_noflush_stop
);
3394 * <nr mapped sectors> <highest mapped sector>
3396 static void thin_status(struct dm_target
*ti
, status_type_t type
,
3397 unsigned status_flags
, char *result
, unsigned maxlen
)
3401 dm_block_t mapped
, highest
;
3402 char buf
[BDEVNAME_SIZE
];
3403 struct thin_c
*tc
= ti
->private;
3405 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
3414 case STATUSTYPE_INFO
:
3415 r
= dm_thin_get_mapped_count(tc
->td
, &mapped
);
3417 DMERR("dm_thin_get_mapped_count returned %d", r
);
3421 r
= dm_thin_get_highest_mapped_block(tc
->td
, &highest
);
3423 DMERR("dm_thin_get_highest_mapped_block returned %d", r
);
3427 DMEMIT("%llu ", mapped
* tc
->pool
->sectors_per_block
);
3429 DMEMIT("%llu", ((highest
+ 1) *
3430 tc
->pool
->sectors_per_block
) - 1);
3435 case STATUSTYPE_TABLE
:
3437 format_dev_t(buf
, tc
->pool_dev
->bdev
->bd_dev
),
3438 (unsigned long) tc
->dev_id
);
3440 DMEMIT(" %s", format_dev_t(buf
, tc
->origin_dev
->bdev
->bd_dev
));
3451 static int thin_iterate_devices(struct dm_target
*ti
,
3452 iterate_devices_callout_fn fn
, void *data
)
3455 struct thin_c
*tc
= ti
->private;
3456 struct pool
*pool
= tc
->pool
;
3459 * We can't call dm_pool_get_data_dev_size() since that blocks. So
3460 * we follow a more convoluted path through to the pool's target.
3463 return 0; /* nothing is bound */
3465 blocks
= pool
->ti
->len
;
3466 (void) sector_div(blocks
, pool
->sectors_per_block
);
3468 return fn(ti
, tc
->pool_dev
, 0, pool
->sectors_per_block
* blocks
, data
);
3473 static struct target_type thin_target
= {
3475 .version
= {1, 12, 0},
3476 .module
= THIS_MODULE
,
3480 .end_io
= thin_endio
,
3481 .presuspend
= thin_presuspend
,
3482 .postsuspend
= thin_postsuspend
,
3483 .status
= thin_status
,
3484 .iterate_devices
= thin_iterate_devices
,
3487 /*----------------------------------------------------------------*/
3489 static int __init
dm_thin_init(void)
3495 r
= dm_register_target(&thin_target
);
3499 r
= dm_register_target(&pool_target
);
3501 goto bad_pool_target
;
3505 _new_mapping_cache
= KMEM_CACHE(dm_thin_new_mapping
, 0);
3506 if (!_new_mapping_cache
)
3507 goto bad_new_mapping_cache
;
3511 bad_new_mapping_cache
:
3512 dm_unregister_target(&pool_target
);
3514 dm_unregister_target(&thin_target
);
3519 static void dm_thin_exit(void)
3521 dm_unregister_target(&thin_target
);
3522 dm_unregister_target(&pool_target
);
3524 kmem_cache_destroy(_new_mapping_cache
);
3527 module_init(dm_thin_init
);
3528 module_exit(dm_thin_exit
);
3530 module_param_named(no_space_timeout
, no_space_timeout_secs
, uint
, S_IRUGO
| S_IWUSR
);
3531 MODULE_PARM_DESC(no_space_timeout
, "Out of data space queue IO timeout in seconds");
3533 MODULE_DESCRIPTION(DM_NAME
" thin provisioning target");
3534 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3535 MODULE_LICENSE("GPL");