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
;
230 sector_t origin_size
;
234 struct dm_thin_device
*td
;
237 struct bio_list deferred_bio_list
;
238 struct bio_list retry_on_resume_list
;
239 struct rb_root sort_bio_list
; /* sorted list of deferred bios */
242 * Ensures the thin is not destroyed until the worker has finished
243 * iterating the active_thins list.
246 struct completion can_destroy
;
249 /*----------------------------------------------------------------*/
252 * wake_worker() is used when new work is queued and when pool_resume is
253 * ready to continue deferred IO processing.
255 static void wake_worker(struct pool
*pool
)
257 queue_work(pool
->wq
, &pool
->worker
);
260 /*----------------------------------------------------------------*/
262 static int bio_detain(struct pool
*pool
, struct dm_cell_key
*key
, struct bio
*bio
,
263 struct dm_bio_prison_cell
**cell_result
)
266 struct dm_bio_prison_cell
*cell_prealloc
;
269 * Allocate a cell from the prison's mempool.
270 * This might block but it can't fail.
272 cell_prealloc
= dm_bio_prison_alloc_cell(pool
->prison
, GFP_NOIO
);
274 r
= dm_bio_detain(pool
->prison
, key
, bio
, cell_prealloc
, cell_result
);
277 * We reused an old cell; we can get rid of
280 dm_bio_prison_free_cell(pool
->prison
, cell_prealloc
);
285 static void cell_release(struct pool
*pool
,
286 struct dm_bio_prison_cell
*cell
,
287 struct bio_list
*bios
)
289 dm_cell_release(pool
->prison
, cell
, bios
);
290 dm_bio_prison_free_cell(pool
->prison
, cell
);
293 static void cell_release_no_holder(struct pool
*pool
,
294 struct dm_bio_prison_cell
*cell
,
295 struct bio_list
*bios
)
297 dm_cell_release_no_holder(pool
->prison
, cell
, bios
);
298 dm_bio_prison_free_cell(pool
->prison
, cell
);
301 static void cell_defer_no_holder_no_free(struct thin_c
*tc
,
302 struct dm_bio_prison_cell
*cell
)
304 struct pool
*pool
= tc
->pool
;
307 spin_lock_irqsave(&tc
->lock
, flags
);
308 dm_cell_release_no_holder(pool
->prison
, cell
, &tc
->deferred_bio_list
);
309 spin_unlock_irqrestore(&tc
->lock
, flags
);
314 static void cell_error_with_code(struct pool
*pool
,
315 struct dm_bio_prison_cell
*cell
, int error_code
)
317 dm_cell_error(pool
->prison
, cell
, error_code
);
318 dm_bio_prison_free_cell(pool
->prison
, cell
);
321 static void cell_error(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
323 cell_error_with_code(pool
, cell
, -EIO
);
326 /*----------------------------------------------------------------*/
329 * A global list of pools that uses a struct mapped_device as a key.
331 static struct dm_thin_pool_table
{
333 struct list_head pools
;
334 } dm_thin_pool_table
;
336 static void pool_table_init(void)
338 mutex_init(&dm_thin_pool_table
.mutex
);
339 INIT_LIST_HEAD(&dm_thin_pool_table
.pools
);
342 static void __pool_table_insert(struct pool
*pool
)
344 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
345 list_add(&pool
->list
, &dm_thin_pool_table
.pools
);
348 static void __pool_table_remove(struct pool
*pool
)
350 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
351 list_del(&pool
->list
);
354 static struct pool
*__pool_table_lookup(struct mapped_device
*md
)
356 struct pool
*pool
= NULL
, *tmp
;
358 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
360 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
361 if (tmp
->pool_md
== md
) {
370 static struct pool
*__pool_table_lookup_metadata_dev(struct block_device
*md_dev
)
372 struct pool
*pool
= NULL
, *tmp
;
374 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
376 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
377 if (tmp
->md_dev
== md_dev
) {
386 /*----------------------------------------------------------------*/
388 struct dm_thin_endio_hook
{
390 struct dm_deferred_entry
*shared_read_entry
;
391 struct dm_deferred_entry
*all_io_entry
;
392 struct dm_thin_new_mapping
*overwrite_mapping
;
393 struct rb_node rb_node
;
396 static void requeue_bio_list(struct thin_c
*tc
, struct bio_list
*master
)
399 struct bio_list bios
;
402 bio_list_init(&bios
);
404 spin_lock_irqsave(&tc
->lock
, flags
);
405 bio_list_merge(&bios
, master
);
406 bio_list_init(master
);
407 spin_unlock_irqrestore(&tc
->lock
, flags
);
409 while ((bio
= bio_list_pop(&bios
)))
410 bio_endio(bio
, DM_ENDIO_REQUEUE
);
413 static void requeue_io(struct thin_c
*tc
)
415 requeue_bio_list(tc
, &tc
->deferred_bio_list
);
416 requeue_bio_list(tc
, &tc
->retry_on_resume_list
);
419 static void error_thin_retry_list(struct thin_c
*tc
)
423 struct bio_list bios
;
425 bio_list_init(&bios
);
427 spin_lock_irqsave(&tc
->lock
, flags
);
428 bio_list_merge(&bios
, &tc
->retry_on_resume_list
);
429 bio_list_init(&tc
->retry_on_resume_list
);
430 spin_unlock_irqrestore(&tc
->lock
, flags
);
432 while ((bio
= bio_list_pop(&bios
)))
436 static void error_retry_list(struct pool
*pool
)
441 list_for_each_entry_rcu(tc
, &pool
->active_thins
, list
)
442 error_thin_retry_list(tc
);
447 * This section of code contains the logic for processing a thin device's IO.
448 * Much of the code depends on pool object resources (lists, workqueues, etc)
449 * but most is exclusively called from the thin target rather than the thin-pool
453 static bool block_size_is_power_of_two(struct pool
*pool
)
455 return pool
->sectors_per_block_shift
>= 0;
458 static dm_block_t
get_bio_block(struct thin_c
*tc
, struct bio
*bio
)
460 struct pool
*pool
= tc
->pool
;
461 sector_t block_nr
= bio
->bi_iter
.bi_sector
;
463 if (block_size_is_power_of_two(pool
))
464 block_nr
>>= pool
->sectors_per_block_shift
;
466 (void) sector_div(block_nr
, pool
->sectors_per_block
);
471 static void remap(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
)
473 struct pool
*pool
= tc
->pool
;
474 sector_t bi_sector
= bio
->bi_iter
.bi_sector
;
476 bio
->bi_bdev
= tc
->pool_dev
->bdev
;
477 if (block_size_is_power_of_two(pool
))
478 bio
->bi_iter
.bi_sector
=
479 (block
<< pool
->sectors_per_block_shift
) |
480 (bi_sector
& (pool
->sectors_per_block
- 1));
482 bio
->bi_iter
.bi_sector
= (block
* pool
->sectors_per_block
) +
483 sector_div(bi_sector
, pool
->sectors_per_block
);
486 static void remap_to_origin(struct thin_c
*tc
, struct bio
*bio
)
488 bio
->bi_bdev
= tc
->origin_dev
->bdev
;
491 static int bio_triggers_commit(struct thin_c
*tc
, struct bio
*bio
)
493 return (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) &&
494 dm_thin_changed_this_transaction(tc
->td
);
497 static void inc_all_io_entry(struct pool
*pool
, struct bio
*bio
)
499 struct dm_thin_endio_hook
*h
;
501 if (bio
->bi_rw
& REQ_DISCARD
)
504 h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
505 h
->all_io_entry
= dm_deferred_entry_inc(pool
->all_io_ds
);
508 static void issue(struct thin_c
*tc
, struct bio
*bio
)
510 struct pool
*pool
= tc
->pool
;
513 if (!bio_triggers_commit(tc
, bio
)) {
514 generic_make_request(bio
);
519 * Complete bio with an error if earlier I/O caused changes to
520 * the metadata that can't be committed e.g, due to I/O errors
521 * on the metadata device.
523 if (dm_thin_aborted_changes(tc
->td
)) {
529 * Batch together any bios that trigger commits and then issue a
530 * single commit for them in process_deferred_bios().
532 spin_lock_irqsave(&pool
->lock
, flags
);
533 bio_list_add(&pool
->deferred_flush_bios
, bio
);
534 spin_unlock_irqrestore(&pool
->lock
, flags
);
537 static void remap_to_origin_and_issue(struct thin_c
*tc
, struct bio
*bio
)
539 remap_to_origin(tc
, bio
);
543 static void remap_and_issue(struct thin_c
*tc
, struct bio
*bio
,
546 remap(tc
, bio
, block
);
550 /*----------------------------------------------------------------*/
553 * Bio endio functions.
555 struct dm_thin_new_mapping
{
556 struct list_head list
;
559 bool definitely_not_shared
:1;
562 * Track quiescing, copying and zeroing preparation actions. When this
563 * counter hits zero the block is prepared and can be inserted into the
566 atomic_t prepare_actions
;
570 dm_block_t virt_block
;
571 dm_block_t data_block
;
572 struct dm_bio_prison_cell
*cell
, *cell2
;
575 * If the bio covers the whole area of a block then we can avoid
576 * zeroing or copying. Instead this bio is hooked. The bio will
577 * still be in the cell, so care has to be taken to avoid issuing
581 bio_end_io_t
*saved_bi_end_io
;
584 static void __complete_mapping_preparation(struct dm_thin_new_mapping
*m
)
586 struct pool
*pool
= m
->tc
->pool
;
588 if (atomic_dec_and_test(&m
->prepare_actions
)) {
589 list_add_tail(&m
->list
, &pool
->prepared_mappings
);
594 static void complete_mapping_preparation(struct dm_thin_new_mapping
*m
)
597 struct pool
*pool
= m
->tc
->pool
;
599 spin_lock_irqsave(&pool
->lock
, flags
);
600 __complete_mapping_preparation(m
);
601 spin_unlock_irqrestore(&pool
->lock
, flags
);
604 static void copy_complete(int read_err
, unsigned long write_err
, void *context
)
606 struct dm_thin_new_mapping
*m
= context
;
608 m
->err
= read_err
|| write_err
? -EIO
: 0;
609 complete_mapping_preparation(m
);
612 static void overwrite_endio(struct bio
*bio
, int err
)
614 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
615 struct dm_thin_new_mapping
*m
= h
->overwrite_mapping
;
618 complete_mapping_preparation(m
);
621 /*----------------------------------------------------------------*/
628 * Prepared mapping jobs.
632 * This sends the bios in the cell back to the deferred_bios list.
634 static void cell_defer(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
636 struct pool
*pool
= tc
->pool
;
639 spin_lock_irqsave(&tc
->lock
, flags
);
640 cell_release(pool
, cell
, &tc
->deferred_bio_list
);
641 spin_unlock_irqrestore(&tc
->lock
, flags
);
647 * Same as cell_defer above, except it omits the original holder of the cell.
649 static void cell_defer_no_holder(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
651 struct pool
*pool
= tc
->pool
;
654 spin_lock_irqsave(&tc
->lock
, flags
);
655 cell_release_no_holder(pool
, cell
, &tc
->deferred_bio_list
);
656 spin_unlock_irqrestore(&tc
->lock
, flags
);
661 static void process_prepared_mapping_fail(struct dm_thin_new_mapping
*m
)
664 m
->bio
->bi_end_io
= m
->saved_bi_end_io
;
665 atomic_inc(&m
->bio
->bi_remaining
);
667 cell_error(m
->tc
->pool
, m
->cell
);
669 mempool_free(m
, m
->tc
->pool
->mapping_pool
);
672 static void process_prepared_mapping(struct dm_thin_new_mapping
*m
)
674 struct thin_c
*tc
= m
->tc
;
675 struct pool
*pool
= tc
->pool
;
681 bio
->bi_end_io
= m
->saved_bi_end_io
;
682 atomic_inc(&bio
->bi_remaining
);
686 cell_error(pool
, m
->cell
);
691 * Commit the prepared block into the mapping btree.
692 * Any I/O for this block arriving after this point will get
693 * remapped to it directly.
695 r
= dm_thin_insert_block(tc
->td
, m
->virt_block
, m
->data_block
);
697 metadata_operation_failed(pool
, "dm_thin_insert_block", r
);
698 cell_error(pool
, m
->cell
);
703 * Release any bios held while the block was being provisioned.
704 * If we are processing a write bio that completely covers the block,
705 * we already processed it so can ignore it now when processing
706 * the bios in the cell.
709 cell_defer_no_holder(tc
, m
->cell
);
712 cell_defer(tc
, m
->cell
);
716 mempool_free(m
, pool
->mapping_pool
);
719 static void process_prepared_discard_fail(struct dm_thin_new_mapping
*m
)
721 struct thin_c
*tc
= m
->tc
;
723 bio_io_error(m
->bio
);
724 cell_defer_no_holder(tc
, m
->cell
);
725 cell_defer_no_holder(tc
, m
->cell2
);
726 mempool_free(m
, tc
->pool
->mapping_pool
);
729 static void process_prepared_discard_passdown(struct dm_thin_new_mapping
*m
)
731 struct thin_c
*tc
= m
->tc
;
733 inc_all_io_entry(tc
->pool
, m
->bio
);
734 cell_defer_no_holder(tc
, m
->cell
);
735 cell_defer_no_holder(tc
, m
->cell2
);
738 if (m
->definitely_not_shared
)
739 remap_and_issue(tc
, m
->bio
, m
->data_block
);
742 if (dm_pool_block_is_used(tc
->pool
->pmd
, m
->data_block
, &used
) || used
)
743 bio_endio(m
->bio
, 0);
745 remap_and_issue(tc
, m
->bio
, m
->data_block
);
748 bio_endio(m
->bio
, 0);
750 mempool_free(m
, tc
->pool
->mapping_pool
);
753 static void process_prepared_discard(struct dm_thin_new_mapping
*m
)
756 struct thin_c
*tc
= m
->tc
;
758 r
= dm_thin_remove_block(tc
->td
, m
->virt_block
);
760 DMERR_LIMIT("dm_thin_remove_block() failed");
762 process_prepared_discard_passdown(m
);
765 static void process_prepared(struct pool
*pool
, struct list_head
*head
,
766 process_mapping_fn
*fn
)
769 struct list_head maps
;
770 struct dm_thin_new_mapping
*m
, *tmp
;
772 INIT_LIST_HEAD(&maps
);
773 spin_lock_irqsave(&pool
->lock
, flags
);
774 list_splice_init(head
, &maps
);
775 spin_unlock_irqrestore(&pool
->lock
, flags
);
777 list_for_each_entry_safe(m
, tmp
, &maps
, list
)
784 static int io_overlaps_block(struct pool
*pool
, struct bio
*bio
)
786 return bio
->bi_iter
.bi_size
==
787 (pool
->sectors_per_block
<< SECTOR_SHIFT
);
790 static int io_overwrites_block(struct pool
*pool
, struct bio
*bio
)
792 return (bio_data_dir(bio
) == WRITE
) &&
793 io_overlaps_block(pool
, bio
);
796 static void save_and_set_endio(struct bio
*bio
, bio_end_io_t
**save
,
799 *save
= bio
->bi_end_io
;
803 static int ensure_next_mapping(struct pool
*pool
)
805 if (pool
->next_mapping
)
808 pool
->next_mapping
= mempool_alloc(pool
->mapping_pool
, GFP_ATOMIC
);
810 return pool
->next_mapping
? 0 : -ENOMEM
;
813 static struct dm_thin_new_mapping
*get_next_mapping(struct pool
*pool
)
815 struct dm_thin_new_mapping
*m
= pool
->next_mapping
;
817 BUG_ON(!pool
->next_mapping
);
819 memset(m
, 0, sizeof(struct dm_thin_new_mapping
));
820 INIT_LIST_HEAD(&m
->list
);
823 pool
->next_mapping
= NULL
;
828 static void ll_zero(struct thin_c
*tc
, struct dm_thin_new_mapping
*m
,
829 sector_t begin
, sector_t end
)
832 struct dm_io_region to
;
834 to
.bdev
= tc
->pool_dev
->bdev
;
836 to
.count
= end
- begin
;
838 r
= dm_kcopyd_zero(tc
->pool
->copier
, 1, &to
, 0, copy_complete
, m
);
840 DMERR_LIMIT("dm_kcopyd_zero() failed");
841 copy_complete(1, 1, m
);
846 * A partial copy also needs to zero the uncopied region.
848 static void schedule_copy(struct thin_c
*tc
, dm_block_t virt_block
,
849 struct dm_dev
*origin
, dm_block_t data_origin
,
850 dm_block_t data_dest
,
851 struct dm_bio_prison_cell
*cell
, struct bio
*bio
,
855 struct pool
*pool
= tc
->pool
;
856 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
859 m
->virt_block
= virt_block
;
860 m
->data_block
= data_dest
;
864 * quiesce action + copy action + an extra reference held for the
865 * duration of this function (we may need to inc later for a
868 atomic_set(&m
->prepare_actions
, 3);
870 if (!dm_deferred_set_add_work(pool
->shared_read_ds
, &m
->list
))
871 complete_mapping_preparation(m
); /* already quiesced */
874 * IO to pool_dev remaps to the pool target's data_dev.
876 * If the whole block of data is being overwritten, we can issue the
877 * bio immediately. Otherwise we use kcopyd to clone the data first.
879 if (io_overwrites_block(pool
, bio
)) {
880 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
882 h
->overwrite_mapping
= m
;
884 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
885 inc_all_io_entry(pool
, bio
);
886 remap_and_issue(tc
, bio
, data_dest
);
888 struct dm_io_region from
, to
;
890 from
.bdev
= origin
->bdev
;
891 from
.sector
= data_origin
* pool
->sectors_per_block
;
894 to
.bdev
= tc
->pool_dev
->bdev
;
895 to
.sector
= data_dest
* pool
->sectors_per_block
;
898 r
= dm_kcopyd_copy(pool
->copier
, &from
, 1, &to
,
899 0, copy_complete
, m
);
901 DMERR_LIMIT("dm_kcopyd_copy() failed");
902 copy_complete(1, 1, m
);
905 * We allow the zero to be issued, to simplify the
906 * error path. Otherwise we'd need to start
907 * worrying about decrementing the prepare_actions
913 * Do we need to zero a tail region?
915 if (len
< pool
->sectors_per_block
&& pool
->pf
.zero_new_blocks
) {
916 atomic_inc(&m
->prepare_actions
);
918 data_dest
* pool
->sectors_per_block
+ len
,
919 (data_dest
+ 1) * pool
->sectors_per_block
);
923 complete_mapping_preparation(m
); /* drop our ref */
926 static void schedule_internal_copy(struct thin_c
*tc
, dm_block_t virt_block
,
927 dm_block_t data_origin
, dm_block_t data_dest
,
928 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
930 schedule_copy(tc
, virt_block
, tc
->pool_dev
,
931 data_origin
, data_dest
, cell
, bio
,
932 tc
->pool
->sectors_per_block
);
935 static void schedule_zero(struct thin_c
*tc
, dm_block_t virt_block
,
936 dm_block_t data_block
, struct dm_bio_prison_cell
*cell
,
939 struct pool
*pool
= tc
->pool
;
940 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
942 atomic_set(&m
->prepare_actions
, 1); /* no need to quiesce */
944 m
->virt_block
= virt_block
;
945 m
->data_block
= data_block
;
949 * If the whole block of data is being overwritten or we are not
950 * zeroing pre-existing data, we can issue the bio immediately.
951 * Otherwise we use kcopyd to zero the data first.
953 if (!pool
->pf
.zero_new_blocks
)
954 process_prepared_mapping(m
);
956 else if (io_overwrites_block(pool
, bio
)) {
957 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
959 h
->overwrite_mapping
= m
;
961 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
962 inc_all_io_entry(pool
, bio
);
963 remap_and_issue(tc
, bio
, data_block
);
967 data_block
* pool
->sectors_per_block
,
968 (data_block
+ 1) * pool
->sectors_per_block
);
971 static void schedule_external_copy(struct thin_c
*tc
, dm_block_t virt_block
,
972 dm_block_t data_dest
,
973 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
975 struct pool
*pool
= tc
->pool
;
976 sector_t virt_block_begin
= virt_block
* pool
->sectors_per_block
;
977 sector_t virt_block_end
= (virt_block
+ 1) * pool
->sectors_per_block
;
979 if (virt_block_end
<= tc
->origin_size
)
980 schedule_copy(tc
, virt_block
, tc
->origin_dev
,
981 virt_block
, data_dest
, cell
, bio
,
982 pool
->sectors_per_block
);
984 else if (virt_block_begin
< tc
->origin_size
)
985 schedule_copy(tc
, virt_block
, tc
->origin_dev
,
986 virt_block
, data_dest
, cell
, bio
,
987 tc
->origin_size
- virt_block_begin
);
990 schedule_zero(tc
, virt_block
, data_dest
, cell
, bio
);
994 * A non-zero return indicates read_only or fail_io mode.
995 * Many callers don't care about the return value.
997 static int commit(struct pool
*pool
)
1001 if (get_pool_mode(pool
) >= PM_READ_ONLY
)
1004 r
= dm_pool_commit_metadata(pool
->pmd
);
1006 metadata_operation_failed(pool
, "dm_pool_commit_metadata", r
);
1011 static void check_low_water_mark(struct pool
*pool
, dm_block_t free_blocks
)
1013 unsigned long flags
;
1015 if (free_blocks
<= pool
->low_water_blocks
&& !pool
->low_water_triggered
) {
1016 DMWARN("%s: reached low water mark for data device: sending event.",
1017 dm_device_name(pool
->pool_md
));
1018 spin_lock_irqsave(&pool
->lock
, flags
);
1019 pool
->low_water_triggered
= true;
1020 spin_unlock_irqrestore(&pool
->lock
, flags
);
1021 dm_table_event(pool
->ti
->table
);
1025 static void set_pool_mode(struct pool
*pool
, enum pool_mode new_mode
);
1027 static int alloc_data_block(struct thin_c
*tc
, dm_block_t
*result
)
1030 dm_block_t free_blocks
;
1031 struct pool
*pool
= tc
->pool
;
1033 if (WARN_ON(get_pool_mode(pool
) != PM_WRITE
))
1036 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
1038 metadata_operation_failed(pool
, "dm_pool_get_free_block_count", r
);
1042 check_low_water_mark(pool
, free_blocks
);
1046 * Try to commit to see if that will free up some
1053 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
1055 metadata_operation_failed(pool
, "dm_pool_get_free_block_count", r
);
1060 set_pool_mode(pool
, PM_OUT_OF_DATA_SPACE
);
1065 r
= dm_pool_alloc_data_block(pool
->pmd
, result
);
1067 metadata_operation_failed(pool
, "dm_pool_alloc_data_block", r
);
1075 * If we have run out of space, queue bios until the device is
1076 * resumed, presumably after having been reloaded with more space.
1078 static void retry_on_resume(struct bio
*bio
)
1080 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1081 struct thin_c
*tc
= h
->tc
;
1082 unsigned long flags
;
1084 spin_lock_irqsave(&tc
->lock
, flags
);
1085 bio_list_add(&tc
->retry_on_resume_list
, bio
);
1086 spin_unlock_irqrestore(&tc
->lock
, flags
);
1089 static int should_error_unserviceable_bio(struct pool
*pool
)
1091 enum pool_mode m
= get_pool_mode(pool
);
1095 /* Shouldn't get here */
1096 DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
1099 case PM_OUT_OF_DATA_SPACE
:
1100 return pool
->pf
.error_if_no_space
? -ENOSPC
: 0;
1106 /* Shouldn't get here */
1107 DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1112 static void handle_unserviceable_bio(struct pool
*pool
, struct bio
*bio
)
1114 int error
= should_error_unserviceable_bio(pool
);
1117 bio_endio(bio
, error
);
1119 retry_on_resume(bio
);
1122 static void retry_bios_on_resume(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
1125 struct bio_list bios
;
1128 error
= should_error_unserviceable_bio(pool
);
1130 cell_error_with_code(pool
, cell
, error
);
1134 bio_list_init(&bios
);
1135 cell_release(pool
, cell
, &bios
);
1137 error
= should_error_unserviceable_bio(pool
);
1139 while ((bio
= bio_list_pop(&bios
)))
1140 bio_endio(bio
, error
);
1142 while ((bio
= bio_list_pop(&bios
)))
1143 retry_on_resume(bio
);
1146 static void process_discard(struct thin_c
*tc
, struct bio
*bio
)
1149 unsigned long flags
;
1150 struct pool
*pool
= tc
->pool
;
1151 struct dm_bio_prison_cell
*cell
, *cell2
;
1152 struct dm_cell_key key
, key2
;
1153 dm_block_t block
= get_bio_block(tc
, bio
);
1154 struct dm_thin_lookup_result lookup_result
;
1155 struct dm_thin_new_mapping
*m
;
1157 build_virtual_key(tc
->td
, block
, &key
);
1158 if (bio_detain(tc
->pool
, &key
, bio
, &cell
))
1161 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1165 * Check nobody is fiddling with this pool block. This can
1166 * happen if someone's in the process of breaking sharing
1169 build_data_key(tc
->td
, lookup_result
.block
, &key2
);
1170 if (bio_detain(tc
->pool
, &key2
, bio
, &cell2
)) {
1171 cell_defer_no_holder(tc
, cell
);
1175 if (io_overlaps_block(pool
, bio
)) {
1177 * IO may still be going to the destination block. We must
1178 * quiesce before we can do the removal.
1180 m
= get_next_mapping(pool
);
1182 m
->pass_discard
= pool
->pf
.discard_passdown
;
1183 m
->definitely_not_shared
= !lookup_result
.shared
;
1184 m
->virt_block
= block
;
1185 m
->data_block
= lookup_result
.block
;
1190 if (!dm_deferred_set_add_work(pool
->all_io_ds
, &m
->list
)) {
1191 spin_lock_irqsave(&pool
->lock
, flags
);
1192 list_add_tail(&m
->list
, &pool
->prepared_discards
);
1193 spin_unlock_irqrestore(&pool
->lock
, flags
);
1197 inc_all_io_entry(pool
, bio
);
1198 cell_defer_no_holder(tc
, cell
);
1199 cell_defer_no_holder(tc
, cell2
);
1202 * The DM core makes sure that the discard doesn't span
1203 * a block boundary. So we submit the discard of a
1204 * partial block appropriately.
1206 if ((!lookup_result
.shared
) && pool
->pf
.discard_passdown
)
1207 remap_and_issue(tc
, bio
, lookup_result
.block
);
1215 * It isn't provisioned, just forget it.
1217 cell_defer_no_holder(tc
, cell
);
1222 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1224 cell_defer_no_holder(tc
, cell
);
1230 static void break_sharing(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1231 struct dm_cell_key
*key
,
1232 struct dm_thin_lookup_result
*lookup_result
,
1233 struct dm_bio_prison_cell
*cell
)
1236 dm_block_t data_block
;
1237 struct pool
*pool
= tc
->pool
;
1239 r
= alloc_data_block(tc
, &data_block
);
1242 schedule_internal_copy(tc
, block
, lookup_result
->block
,
1243 data_block
, cell
, bio
);
1247 retry_bios_on_resume(pool
, cell
);
1251 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1253 cell_error(pool
, cell
);
1258 static void process_shared_bio(struct thin_c
*tc
, struct bio
*bio
,
1260 struct dm_thin_lookup_result
*lookup_result
)
1262 struct dm_bio_prison_cell
*cell
;
1263 struct pool
*pool
= tc
->pool
;
1264 struct dm_cell_key key
;
1267 * If cell is already occupied, then sharing is already in the process
1268 * of being broken so we have nothing further to do here.
1270 build_data_key(tc
->td
, lookup_result
->block
, &key
);
1271 if (bio_detain(pool
, &key
, bio
, &cell
))
1274 if (bio_data_dir(bio
) == WRITE
&& bio
->bi_iter
.bi_size
)
1275 break_sharing(tc
, bio
, block
, &key
, lookup_result
, cell
);
1277 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1279 h
->shared_read_entry
= dm_deferred_entry_inc(pool
->shared_read_ds
);
1280 inc_all_io_entry(pool
, bio
);
1281 cell_defer_no_holder(tc
, cell
);
1283 remap_and_issue(tc
, bio
, lookup_result
->block
);
1287 static void provision_block(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1288 struct dm_bio_prison_cell
*cell
)
1291 dm_block_t data_block
;
1292 struct pool
*pool
= tc
->pool
;
1295 * Remap empty bios (flushes) immediately, without provisioning.
1297 if (!bio
->bi_iter
.bi_size
) {
1298 inc_all_io_entry(pool
, bio
);
1299 cell_defer_no_holder(tc
, cell
);
1301 remap_and_issue(tc
, bio
, 0);
1306 * Fill read bios with zeroes and complete them immediately.
1308 if (bio_data_dir(bio
) == READ
) {
1310 cell_defer_no_holder(tc
, cell
);
1315 r
= alloc_data_block(tc
, &data_block
);
1319 schedule_external_copy(tc
, block
, data_block
, cell
, bio
);
1321 schedule_zero(tc
, block
, data_block
, cell
, bio
);
1325 retry_bios_on_resume(pool
, cell
);
1329 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1331 cell_error(pool
, cell
);
1336 static void process_bio(struct thin_c
*tc
, struct bio
*bio
)
1339 struct pool
*pool
= tc
->pool
;
1340 dm_block_t block
= get_bio_block(tc
, bio
);
1341 struct dm_bio_prison_cell
*cell
;
1342 struct dm_cell_key key
;
1343 struct dm_thin_lookup_result lookup_result
;
1346 * If cell is already occupied, then the block is already
1347 * being provisioned so we have nothing further to do here.
1349 build_virtual_key(tc
->td
, block
, &key
);
1350 if (bio_detain(pool
, &key
, bio
, &cell
))
1353 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1356 if (lookup_result
.shared
) {
1357 process_shared_bio(tc
, bio
, block
, &lookup_result
);
1358 cell_defer_no_holder(tc
, cell
); /* FIXME: pass this cell into process_shared? */
1360 inc_all_io_entry(pool
, bio
);
1361 cell_defer_no_holder(tc
, cell
);
1363 remap_and_issue(tc
, bio
, lookup_result
.block
);
1368 if (bio_data_dir(bio
) == READ
&& tc
->origin_dev
) {
1369 inc_all_io_entry(pool
, bio
);
1370 cell_defer_no_holder(tc
, cell
);
1372 if (bio_end_sector(bio
) <= tc
->origin_size
)
1373 remap_to_origin_and_issue(tc
, bio
);
1375 else if (bio
->bi_iter
.bi_sector
< tc
->origin_size
) {
1377 bio
->bi_iter
.bi_size
= (tc
->origin_size
- bio
->bi_iter
.bi_sector
) << SECTOR_SHIFT
;
1378 remap_to_origin_and_issue(tc
, bio
);
1385 provision_block(tc
, bio
, block
, cell
);
1389 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1391 cell_defer_no_holder(tc
, cell
);
1397 static void process_bio_read_only(struct thin_c
*tc
, struct bio
*bio
)
1400 int rw
= bio_data_dir(bio
);
1401 dm_block_t block
= get_bio_block(tc
, bio
);
1402 struct dm_thin_lookup_result lookup_result
;
1404 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1407 if (lookup_result
.shared
&& (rw
== WRITE
) && bio
->bi_iter
.bi_size
)
1408 handle_unserviceable_bio(tc
->pool
, bio
);
1410 inc_all_io_entry(tc
->pool
, bio
);
1411 remap_and_issue(tc
, bio
, lookup_result
.block
);
1417 handle_unserviceable_bio(tc
->pool
, bio
);
1421 if (tc
->origin_dev
) {
1422 inc_all_io_entry(tc
->pool
, bio
);
1423 remap_to_origin_and_issue(tc
, bio
);
1432 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1439 static void process_bio_success(struct thin_c
*tc
, struct bio
*bio
)
1444 static void process_bio_fail(struct thin_c
*tc
, struct bio
*bio
)
1450 * FIXME: should we also commit due to size of transaction, measured in
1453 static int need_commit_due_to_time(struct pool
*pool
)
1455 return jiffies
< pool
->last_commit_jiffies
||
1456 jiffies
> pool
->last_commit_jiffies
+ COMMIT_PERIOD
;
1459 #define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node)
1460 #define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook))
1462 static void __thin_bio_rb_add(struct thin_c
*tc
, struct bio
*bio
)
1464 struct rb_node
**rbp
, *parent
;
1465 struct dm_thin_endio_hook
*pbd
;
1466 sector_t bi_sector
= bio
->bi_iter
.bi_sector
;
1468 rbp
= &tc
->sort_bio_list
.rb_node
;
1472 pbd
= thin_pbd(parent
);
1474 if (bi_sector
< thin_bio(pbd
)->bi_iter
.bi_sector
)
1475 rbp
= &(*rbp
)->rb_left
;
1477 rbp
= &(*rbp
)->rb_right
;
1480 pbd
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1481 rb_link_node(&pbd
->rb_node
, parent
, rbp
);
1482 rb_insert_color(&pbd
->rb_node
, &tc
->sort_bio_list
);
1485 static void __extract_sorted_bios(struct thin_c
*tc
)
1487 struct rb_node
*node
;
1488 struct dm_thin_endio_hook
*pbd
;
1491 for (node
= rb_first(&tc
->sort_bio_list
); node
; node
= rb_next(node
)) {
1492 pbd
= thin_pbd(node
);
1493 bio
= thin_bio(pbd
);
1495 bio_list_add(&tc
->deferred_bio_list
, bio
);
1496 rb_erase(&pbd
->rb_node
, &tc
->sort_bio_list
);
1499 WARN_ON(!RB_EMPTY_ROOT(&tc
->sort_bio_list
));
1502 static void __sort_thin_deferred_bios(struct thin_c
*tc
)
1505 struct bio_list bios
;
1507 bio_list_init(&bios
);
1508 bio_list_merge(&bios
, &tc
->deferred_bio_list
);
1509 bio_list_init(&tc
->deferred_bio_list
);
1511 /* Sort deferred_bio_list using rb-tree */
1512 while ((bio
= bio_list_pop(&bios
)))
1513 __thin_bio_rb_add(tc
, bio
);
1516 * Transfer the sorted bios in sort_bio_list back to
1517 * deferred_bio_list to allow lockless submission of
1520 __extract_sorted_bios(tc
);
1523 static void process_thin_deferred_bios(struct thin_c
*tc
)
1525 struct pool
*pool
= tc
->pool
;
1526 unsigned long flags
;
1528 struct bio_list bios
;
1529 struct blk_plug plug
;
1531 if (tc
->requeue_mode
) {
1532 requeue_bio_list(tc
, &tc
->deferred_bio_list
);
1536 bio_list_init(&bios
);
1538 spin_lock_irqsave(&tc
->lock
, flags
);
1540 if (bio_list_empty(&tc
->deferred_bio_list
)) {
1541 spin_unlock_irqrestore(&tc
->lock
, flags
);
1545 __sort_thin_deferred_bios(tc
);
1547 bio_list_merge(&bios
, &tc
->deferred_bio_list
);
1548 bio_list_init(&tc
->deferred_bio_list
);
1550 spin_unlock_irqrestore(&tc
->lock
, flags
);
1552 blk_start_plug(&plug
);
1553 while ((bio
= bio_list_pop(&bios
))) {
1555 * If we've got no free new_mapping structs, and processing
1556 * this bio might require one, we pause until there are some
1557 * prepared mappings to process.
1559 if (ensure_next_mapping(pool
)) {
1560 spin_lock_irqsave(&tc
->lock
, flags
);
1561 bio_list_add(&tc
->deferred_bio_list
, bio
);
1562 bio_list_merge(&tc
->deferred_bio_list
, &bios
);
1563 spin_unlock_irqrestore(&tc
->lock
, flags
);
1567 if (bio
->bi_rw
& REQ_DISCARD
)
1568 pool
->process_discard(tc
, bio
);
1570 pool
->process_bio(tc
, bio
);
1572 blk_finish_plug(&plug
);
1575 static void thin_get(struct thin_c
*tc
);
1576 static void thin_put(struct thin_c
*tc
);
1579 * We can't hold rcu_read_lock() around code that can block. So we
1580 * find a thin with the rcu lock held; bump a refcount; then drop
1583 static struct thin_c
*get_first_thin(struct pool
*pool
)
1585 struct thin_c
*tc
= NULL
;
1588 if (!list_empty(&pool
->active_thins
)) {
1589 tc
= list_entry_rcu(pool
->active_thins
.next
, struct thin_c
, list
);
1597 static struct thin_c
*get_next_thin(struct pool
*pool
, struct thin_c
*tc
)
1599 struct thin_c
*old_tc
= tc
;
1602 list_for_each_entry_continue_rcu(tc
, &pool
->active_thins
, list
) {
1614 static void process_deferred_bios(struct pool
*pool
)
1616 unsigned long flags
;
1618 struct bio_list bios
;
1621 tc
= get_first_thin(pool
);
1623 process_thin_deferred_bios(tc
);
1624 tc
= get_next_thin(pool
, tc
);
1628 * If there are any deferred flush bios, we must commit
1629 * the metadata before issuing them.
1631 bio_list_init(&bios
);
1632 spin_lock_irqsave(&pool
->lock
, flags
);
1633 bio_list_merge(&bios
, &pool
->deferred_flush_bios
);
1634 bio_list_init(&pool
->deferred_flush_bios
);
1635 spin_unlock_irqrestore(&pool
->lock
, flags
);
1637 if (bio_list_empty(&bios
) &&
1638 !(dm_pool_changed_this_transaction(pool
->pmd
) && need_commit_due_to_time(pool
)))
1642 while ((bio
= bio_list_pop(&bios
)))
1646 pool
->last_commit_jiffies
= jiffies
;
1648 while ((bio
= bio_list_pop(&bios
)))
1649 generic_make_request(bio
);
1652 static void do_worker(struct work_struct
*ws
)
1654 struct pool
*pool
= container_of(ws
, struct pool
, worker
);
1656 process_prepared(pool
, &pool
->prepared_mappings
, &pool
->process_prepared_mapping
);
1657 process_prepared(pool
, &pool
->prepared_discards
, &pool
->process_prepared_discard
);
1658 process_deferred_bios(pool
);
1662 * We want to commit periodically so that not too much
1663 * unwritten data builds up.
1665 static void do_waker(struct work_struct
*ws
)
1667 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
, waker
);
1669 queue_delayed_work(pool
->wq
, &pool
->waker
, COMMIT_PERIOD
);
1673 * We're holding onto IO to allow userland time to react. After the
1674 * timeout either the pool will have been resized (and thus back in
1675 * PM_WRITE mode), or we degrade to PM_READ_ONLY and start erroring IO.
1677 static void do_no_space_timeout(struct work_struct
*ws
)
1679 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
,
1682 if (get_pool_mode(pool
) == PM_OUT_OF_DATA_SPACE
&& !pool
->pf
.error_if_no_space
)
1683 set_pool_mode(pool
, PM_READ_ONLY
);
1686 /*----------------------------------------------------------------*/
1689 struct work_struct worker
;
1690 struct completion complete
;
1693 static struct pool_work
*to_pool_work(struct work_struct
*ws
)
1695 return container_of(ws
, struct pool_work
, worker
);
1698 static void pool_work_complete(struct pool_work
*pw
)
1700 complete(&pw
->complete
);
1703 static void pool_work_wait(struct pool_work
*pw
, struct pool
*pool
,
1704 void (*fn
)(struct work_struct
*))
1706 INIT_WORK_ONSTACK(&pw
->worker
, fn
);
1707 init_completion(&pw
->complete
);
1708 queue_work(pool
->wq
, &pw
->worker
);
1709 wait_for_completion(&pw
->complete
);
1712 /*----------------------------------------------------------------*/
1714 struct noflush_work
{
1715 struct pool_work pw
;
1719 static struct noflush_work
*to_noflush(struct work_struct
*ws
)
1721 return container_of(to_pool_work(ws
), struct noflush_work
, pw
);
1724 static void do_noflush_start(struct work_struct
*ws
)
1726 struct noflush_work
*w
= to_noflush(ws
);
1727 w
->tc
->requeue_mode
= true;
1729 pool_work_complete(&w
->pw
);
1732 static void do_noflush_stop(struct work_struct
*ws
)
1734 struct noflush_work
*w
= to_noflush(ws
);
1735 w
->tc
->requeue_mode
= false;
1736 pool_work_complete(&w
->pw
);
1739 static void noflush_work(struct thin_c
*tc
, void (*fn
)(struct work_struct
*))
1741 struct noflush_work w
;
1744 pool_work_wait(&w
.pw
, tc
->pool
, fn
);
1747 /*----------------------------------------------------------------*/
1749 static enum pool_mode
get_pool_mode(struct pool
*pool
)
1751 return pool
->pf
.mode
;
1754 static void notify_of_pool_mode_change(struct pool
*pool
, const char *new_mode
)
1756 dm_table_event(pool
->ti
->table
);
1757 DMINFO("%s: switching pool to %s mode",
1758 dm_device_name(pool
->pool_md
), new_mode
);
1761 static void set_pool_mode(struct pool
*pool
, enum pool_mode new_mode
)
1763 struct pool_c
*pt
= pool
->ti
->private;
1764 bool needs_check
= dm_pool_metadata_needs_check(pool
->pmd
);
1765 enum pool_mode old_mode
= get_pool_mode(pool
);
1766 unsigned long no_space_timeout
= ACCESS_ONCE(no_space_timeout_secs
) * HZ
;
1769 * Never allow the pool to transition to PM_WRITE mode if user
1770 * intervention is required to verify metadata and data consistency.
1772 if (new_mode
== PM_WRITE
&& needs_check
) {
1773 DMERR("%s: unable to switch pool to write mode until repaired.",
1774 dm_device_name(pool
->pool_md
));
1775 if (old_mode
!= new_mode
)
1776 new_mode
= old_mode
;
1778 new_mode
= PM_READ_ONLY
;
1781 * If we were in PM_FAIL mode, rollback of metadata failed. We're
1782 * not going to recover without a thin_repair. So we never let the
1783 * pool move out of the old mode.
1785 if (old_mode
== PM_FAIL
)
1786 new_mode
= old_mode
;
1790 if (old_mode
!= new_mode
)
1791 notify_of_pool_mode_change(pool
, "failure");
1792 dm_pool_metadata_read_only(pool
->pmd
);
1793 pool
->process_bio
= process_bio_fail
;
1794 pool
->process_discard
= process_bio_fail
;
1795 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1796 pool
->process_prepared_discard
= process_prepared_discard_fail
;
1798 error_retry_list(pool
);
1802 if (old_mode
!= new_mode
)
1803 notify_of_pool_mode_change(pool
, "read-only");
1804 dm_pool_metadata_read_only(pool
->pmd
);
1805 pool
->process_bio
= process_bio_read_only
;
1806 pool
->process_discard
= process_bio_success
;
1807 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1808 pool
->process_prepared_discard
= process_prepared_discard_passdown
;
1810 error_retry_list(pool
);
1813 case PM_OUT_OF_DATA_SPACE
:
1815 * Ideally we'd never hit this state; the low water mark
1816 * would trigger userland to extend the pool before we
1817 * completely run out of data space. However, many small
1818 * IOs to unprovisioned space can consume data space at an
1819 * alarming rate. Adjust your low water mark if you're
1820 * frequently seeing this mode.
1822 if (old_mode
!= new_mode
)
1823 notify_of_pool_mode_change(pool
, "out-of-data-space");
1824 pool
->process_bio
= process_bio_read_only
;
1825 pool
->process_discard
= process_discard
;
1826 pool
->process_prepared_mapping
= process_prepared_mapping
;
1827 pool
->process_prepared_discard
= process_prepared_discard_passdown
;
1829 if (!pool
->pf
.error_if_no_space
&& no_space_timeout
)
1830 queue_delayed_work(pool
->wq
, &pool
->no_space_timeout
, no_space_timeout
);
1834 if (old_mode
!= new_mode
)
1835 notify_of_pool_mode_change(pool
, "write");
1836 dm_pool_metadata_read_write(pool
->pmd
);
1837 pool
->process_bio
= process_bio
;
1838 pool
->process_discard
= process_discard
;
1839 pool
->process_prepared_mapping
= process_prepared_mapping
;
1840 pool
->process_prepared_discard
= process_prepared_discard
;
1844 pool
->pf
.mode
= new_mode
;
1846 * The pool mode may have changed, sync it so bind_control_target()
1847 * doesn't cause an unexpected mode transition on resume.
1849 pt
->adjusted_pf
.mode
= new_mode
;
1852 static void abort_transaction(struct pool
*pool
)
1854 const char *dev_name
= dm_device_name(pool
->pool_md
);
1856 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name
);
1857 if (dm_pool_abort_metadata(pool
->pmd
)) {
1858 DMERR("%s: failed to abort metadata transaction", dev_name
);
1859 set_pool_mode(pool
, PM_FAIL
);
1862 if (dm_pool_metadata_set_needs_check(pool
->pmd
)) {
1863 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name
);
1864 set_pool_mode(pool
, PM_FAIL
);
1868 static void metadata_operation_failed(struct pool
*pool
, const char *op
, int r
)
1870 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
1871 dm_device_name(pool
->pool_md
), op
, r
);
1873 abort_transaction(pool
);
1874 set_pool_mode(pool
, PM_READ_ONLY
);
1877 /*----------------------------------------------------------------*/
1880 * Mapping functions.
1884 * Called only while mapping a thin bio to hand it over to the workqueue.
1886 static void thin_defer_bio(struct thin_c
*tc
, struct bio
*bio
)
1888 unsigned long flags
;
1889 struct pool
*pool
= tc
->pool
;
1891 spin_lock_irqsave(&tc
->lock
, flags
);
1892 bio_list_add(&tc
->deferred_bio_list
, bio
);
1893 spin_unlock_irqrestore(&tc
->lock
, flags
);
1898 static void thin_hook_bio(struct thin_c
*tc
, struct bio
*bio
)
1900 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1903 h
->shared_read_entry
= NULL
;
1904 h
->all_io_entry
= NULL
;
1905 h
->overwrite_mapping
= NULL
;
1909 * Non-blocking function called from the thin target's map function.
1911 static int thin_bio_map(struct dm_target
*ti
, struct bio
*bio
)
1914 struct thin_c
*tc
= ti
->private;
1915 dm_block_t block
= get_bio_block(tc
, bio
);
1916 struct dm_thin_device
*td
= tc
->td
;
1917 struct dm_thin_lookup_result result
;
1918 struct dm_bio_prison_cell cell1
, cell2
;
1919 struct dm_bio_prison_cell
*cell_result
;
1920 struct dm_cell_key key
;
1922 thin_hook_bio(tc
, bio
);
1924 if (tc
->requeue_mode
) {
1925 bio_endio(bio
, DM_ENDIO_REQUEUE
);
1926 return DM_MAPIO_SUBMITTED
;
1929 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
1931 return DM_MAPIO_SUBMITTED
;
1934 if (bio
->bi_rw
& (REQ_DISCARD
| REQ_FLUSH
| REQ_FUA
)) {
1935 thin_defer_bio(tc
, bio
);
1936 return DM_MAPIO_SUBMITTED
;
1939 r
= dm_thin_find_block(td
, block
, 0, &result
);
1942 * Note that we defer readahead too.
1946 if (unlikely(result
.shared
)) {
1948 * We have a race condition here between the
1949 * result.shared value returned by the lookup and
1950 * snapshot creation, which may cause new
1953 * To avoid this always quiesce the origin before
1954 * taking the snap. You want to do this anyway to
1955 * ensure a consistent application view
1958 * More distant ancestors are irrelevant. The
1959 * shared flag will be set in their case.
1961 thin_defer_bio(tc
, bio
);
1962 return DM_MAPIO_SUBMITTED
;
1965 build_virtual_key(tc
->td
, block
, &key
);
1966 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell1
, &cell_result
))
1967 return DM_MAPIO_SUBMITTED
;
1969 build_data_key(tc
->td
, result
.block
, &key
);
1970 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell2
, &cell_result
)) {
1971 cell_defer_no_holder_no_free(tc
, &cell1
);
1972 return DM_MAPIO_SUBMITTED
;
1975 inc_all_io_entry(tc
->pool
, bio
);
1976 cell_defer_no_holder_no_free(tc
, &cell2
);
1977 cell_defer_no_holder_no_free(tc
, &cell1
);
1979 remap(tc
, bio
, result
.block
);
1980 return DM_MAPIO_REMAPPED
;
1983 if (get_pool_mode(tc
->pool
) == PM_READ_ONLY
) {
1985 * This block isn't provisioned, and we have no way
1988 handle_unserviceable_bio(tc
->pool
, bio
);
1989 return DM_MAPIO_SUBMITTED
;
1995 * In future, the failed dm_thin_find_block above could
1996 * provide the hint to load the metadata into cache.
1998 thin_defer_bio(tc
, bio
);
1999 return DM_MAPIO_SUBMITTED
;
2003 * Must always call bio_io_error on failure.
2004 * dm_thin_find_block can fail with -EINVAL if the
2005 * pool is switched to fail-io mode.
2008 return DM_MAPIO_SUBMITTED
;
2012 static int pool_is_congested(struct dm_target_callbacks
*cb
, int bdi_bits
)
2014 struct pool_c
*pt
= container_of(cb
, struct pool_c
, callbacks
);
2015 struct request_queue
*q
;
2017 if (get_pool_mode(pt
->pool
) == PM_OUT_OF_DATA_SPACE
)
2020 q
= bdev_get_queue(pt
->data_dev
->bdev
);
2021 return bdi_congested(&q
->backing_dev_info
, bdi_bits
);
2024 static void requeue_bios(struct pool
*pool
)
2026 unsigned long flags
;
2030 list_for_each_entry_rcu(tc
, &pool
->active_thins
, list
) {
2031 spin_lock_irqsave(&tc
->lock
, flags
);
2032 bio_list_merge(&tc
->deferred_bio_list
, &tc
->retry_on_resume_list
);
2033 bio_list_init(&tc
->retry_on_resume_list
);
2034 spin_unlock_irqrestore(&tc
->lock
, flags
);
2039 /*----------------------------------------------------------------
2040 * Binding of control targets to a pool object
2041 *--------------------------------------------------------------*/
2042 static bool data_dev_supports_discard(struct pool_c
*pt
)
2044 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
2046 return q
&& blk_queue_discard(q
);
2049 static bool is_factor(sector_t block_size
, uint32_t n
)
2051 return !sector_div(block_size
, n
);
2055 * If discard_passdown was enabled verify that the data device
2056 * supports discards. Disable discard_passdown if not.
2058 static void disable_passdown_if_not_supported(struct pool_c
*pt
)
2060 struct pool
*pool
= pt
->pool
;
2061 struct block_device
*data_bdev
= pt
->data_dev
->bdev
;
2062 struct queue_limits
*data_limits
= &bdev_get_queue(data_bdev
)->limits
;
2063 sector_t block_size
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
2064 const char *reason
= NULL
;
2065 char buf
[BDEVNAME_SIZE
];
2067 if (!pt
->adjusted_pf
.discard_passdown
)
2070 if (!data_dev_supports_discard(pt
))
2071 reason
= "discard unsupported";
2073 else if (data_limits
->max_discard_sectors
< pool
->sectors_per_block
)
2074 reason
= "max discard sectors smaller than a block";
2076 else if (data_limits
->discard_granularity
> block_size
)
2077 reason
= "discard granularity larger than a block";
2079 else if (!is_factor(block_size
, data_limits
->discard_granularity
))
2080 reason
= "discard granularity not a factor of block size";
2083 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev
, buf
), reason
);
2084 pt
->adjusted_pf
.discard_passdown
= false;
2088 static int bind_control_target(struct pool
*pool
, struct dm_target
*ti
)
2090 struct pool_c
*pt
= ti
->private;
2093 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
2095 enum pool_mode old_mode
= get_pool_mode(pool
);
2096 enum pool_mode new_mode
= pt
->adjusted_pf
.mode
;
2099 * Don't change the pool's mode until set_pool_mode() below.
2100 * Otherwise the pool's process_* function pointers may
2101 * not match the desired pool mode.
2103 pt
->adjusted_pf
.mode
= old_mode
;
2106 pool
->pf
= pt
->adjusted_pf
;
2107 pool
->low_water_blocks
= pt
->low_water_blocks
;
2109 set_pool_mode(pool
, new_mode
);
2114 static void unbind_control_target(struct pool
*pool
, struct dm_target
*ti
)
2120 /*----------------------------------------------------------------
2122 *--------------------------------------------------------------*/
2123 /* Initialize pool features. */
2124 static void pool_features_init(struct pool_features
*pf
)
2126 pf
->mode
= PM_WRITE
;
2127 pf
->zero_new_blocks
= true;
2128 pf
->discard_enabled
= true;
2129 pf
->discard_passdown
= true;
2130 pf
->error_if_no_space
= false;
2133 static void __pool_destroy(struct pool
*pool
)
2135 __pool_table_remove(pool
);
2137 if (dm_pool_metadata_close(pool
->pmd
) < 0)
2138 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
2140 dm_bio_prison_destroy(pool
->prison
);
2141 dm_kcopyd_client_destroy(pool
->copier
);
2144 destroy_workqueue(pool
->wq
);
2146 if (pool
->next_mapping
)
2147 mempool_free(pool
->next_mapping
, pool
->mapping_pool
);
2148 mempool_destroy(pool
->mapping_pool
);
2149 dm_deferred_set_destroy(pool
->shared_read_ds
);
2150 dm_deferred_set_destroy(pool
->all_io_ds
);
2154 static struct kmem_cache
*_new_mapping_cache
;
2156 static struct pool
*pool_create(struct mapped_device
*pool_md
,
2157 struct block_device
*metadata_dev
,
2158 unsigned long block_size
,
2159 int read_only
, char **error
)
2164 struct dm_pool_metadata
*pmd
;
2165 bool format_device
= read_only
? false : true;
2167 pmd
= dm_pool_metadata_open(metadata_dev
, block_size
, format_device
);
2169 *error
= "Error creating metadata object";
2170 return (struct pool
*)pmd
;
2173 pool
= kmalloc(sizeof(*pool
), GFP_KERNEL
);
2175 *error
= "Error allocating memory for pool";
2176 err_p
= ERR_PTR(-ENOMEM
);
2181 pool
->sectors_per_block
= block_size
;
2182 if (block_size
& (block_size
- 1))
2183 pool
->sectors_per_block_shift
= -1;
2185 pool
->sectors_per_block_shift
= __ffs(block_size
);
2186 pool
->low_water_blocks
= 0;
2187 pool_features_init(&pool
->pf
);
2188 pool
->prison
= dm_bio_prison_create(PRISON_CELLS
);
2189 if (!pool
->prison
) {
2190 *error
= "Error creating pool's bio prison";
2191 err_p
= ERR_PTR(-ENOMEM
);
2195 pool
->copier
= dm_kcopyd_client_create(&dm_kcopyd_throttle
);
2196 if (IS_ERR(pool
->copier
)) {
2197 r
= PTR_ERR(pool
->copier
);
2198 *error
= "Error creating pool's kcopyd client";
2200 goto bad_kcopyd_client
;
2204 * Create singlethreaded workqueue that will service all devices
2205 * that use this metadata.
2207 pool
->wq
= alloc_ordered_workqueue("dm-" DM_MSG_PREFIX
, WQ_MEM_RECLAIM
);
2209 *error
= "Error creating pool's workqueue";
2210 err_p
= ERR_PTR(-ENOMEM
);
2214 INIT_WORK(&pool
->worker
, do_worker
);
2215 INIT_DELAYED_WORK(&pool
->waker
, do_waker
);
2216 INIT_DELAYED_WORK(&pool
->no_space_timeout
, do_no_space_timeout
);
2217 spin_lock_init(&pool
->lock
);
2218 bio_list_init(&pool
->deferred_flush_bios
);
2219 INIT_LIST_HEAD(&pool
->prepared_mappings
);
2220 INIT_LIST_HEAD(&pool
->prepared_discards
);
2221 INIT_LIST_HEAD(&pool
->active_thins
);
2222 pool
->low_water_triggered
= false;
2224 pool
->shared_read_ds
= dm_deferred_set_create();
2225 if (!pool
->shared_read_ds
) {
2226 *error
= "Error creating pool's shared read deferred set";
2227 err_p
= ERR_PTR(-ENOMEM
);
2228 goto bad_shared_read_ds
;
2231 pool
->all_io_ds
= dm_deferred_set_create();
2232 if (!pool
->all_io_ds
) {
2233 *error
= "Error creating pool's all io deferred set";
2234 err_p
= ERR_PTR(-ENOMEM
);
2238 pool
->next_mapping
= NULL
;
2239 pool
->mapping_pool
= mempool_create_slab_pool(MAPPING_POOL_SIZE
,
2240 _new_mapping_cache
);
2241 if (!pool
->mapping_pool
) {
2242 *error
= "Error creating pool's mapping mempool";
2243 err_p
= ERR_PTR(-ENOMEM
);
2244 goto bad_mapping_pool
;
2247 pool
->ref_count
= 1;
2248 pool
->last_commit_jiffies
= jiffies
;
2249 pool
->pool_md
= pool_md
;
2250 pool
->md_dev
= metadata_dev
;
2251 __pool_table_insert(pool
);
2256 dm_deferred_set_destroy(pool
->all_io_ds
);
2258 dm_deferred_set_destroy(pool
->shared_read_ds
);
2260 destroy_workqueue(pool
->wq
);
2262 dm_kcopyd_client_destroy(pool
->copier
);
2264 dm_bio_prison_destroy(pool
->prison
);
2268 if (dm_pool_metadata_close(pmd
))
2269 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
2274 static void __pool_inc(struct pool
*pool
)
2276 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
2280 static void __pool_dec(struct pool
*pool
)
2282 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
2283 BUG_ON(!pool
->ref_count
);
2284 if (!--pool
->ref_count
)
2285 __pool_destroy(pool
);
2288 static struct pool
*__pool_find(struct mapped_device
*pool_md
,
2289 struct block_device
*metadata_dev
,
2290 unsigned long block_size
, int read_only
,
2291 char **error
, int *created
)
2293 struct pool
*pool
= __pool_table_lookup_metadata_dev(metadata_dev
);
2296 if (pool
->pool_md
!= pool_md
) {
2297 *error
= "metadata device already in use by a pool";
2298 return ERR_PTR(-EBUSY
);
2303 pool
= __pool_table_lookup(pool_md
);
2305 if (pool
->md_dev
!= metadata_dev
) {
2306 *error
= "different pool cannot replace a pool";
2307 return ERR_PTR(-EINVAL
);
2312 pool
= pool_create(pool_md
, metadata_dev
, block_size
, read_only
, error
);
2320 /*----------------------------------------------------------------
2321 * Pool target methods
2322 *--------------------------------------------------------------*/
2323 static void pool_dtr(struct dm_target
*ti
)
2325 struct pool_c
*pt
= ti
->private;
2327 mutex_lock(&dm_thin_pool_table
.mutex
);
2329 unbind_control_target(pt
->pool
, ti
);
2330 __pool_dec(pt
->pool
);
2331 dm_put_device(ti
, pt
->metadata_dev
);
2332 dm_put_device(ti
, pt
->data_dev
);
2335 mutex_unlock(&dm_thin_pool_table
.mutex
);
2338 static int parse_pool_features(struct dm_arg_set
*as
, struct pool_features
*pf
,
2339 struct dm_target
*ti
)
2343 const char *arg_name
;
2345 static struct dm_arg _args
[] = {
2346 {0, 4, "Invalid number of pool feature arguments"},
2350 * No feature arguments supplied.
2355 r
= dm_read_arg_group(_args
, as
, &argc
, &ti
->error
);
2359 while (argc
&& !r
) {
2360 arg_name
= dm_shift_arg(as
);
2363 if (!strcasecmp(arg_name
, "skip_block_zeroing"))
2364 pf
->zero_new_blocks
= false;
2366 else if (!strcasecmp(arg_name
, "ignore_discard"))
2367 pf
->discard_enabled
= false;
2369 else if (!strcasecmp(arg_name
, "no_discard_passdown"))
2370 pf
->discard_passdown
= false;
2372 else if (!strcasecmp(arg_name
, "read_only"))
2373 pf
->mode
= PM_READ_ONLY
;
2375 else if (!strcasecmp(arg_name
, "error_if_no_space"))
2376 pf
->error_if_no_space
= true;
2379 ti
->error
= "Unrecognised pool feature requested";
2388 static void metadata_low_callback(void *context
)
2390 struct pool
*pool
= context
;
2392 DMWARN("%s: reached low water mark for metadata device: sending event.",
2393 dm_device_name(pool
->pool_md
));
2395 dm_table_event(pool
->ti
->table
);
2398 static sector_t
get_dev_size(struct block_device
*bdev
)
2400 return i_size_read(bdev
->bd_inode
) >> SECTOR_SHIFT
;
2403 static void warn_if_metadata_device_too_big(struct block_device
*bdev
)
2405 sector_t metadata_dev_size
= get_dev_size(bdev
);
2406 char buffer
[BDEVNAME_SIZE
];
2408 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS_WARNING
)
2409 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
2410 bdevname(bdev
, buffer
), THIN_METADATA_MAX_SECTORS
);
2413 static sector_t
get_metadata_dev_size(struct block_device
*bdev
)
2415 sector_t metadata_dev_size
= get_dev_size(bdev
);
2417 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS
)
2418 metadata_dev_size
= THIN_METADATA_MAX_SECTORS
;
2420 return metadata_dev_size
;
2423 static dm_block_t
get_metadata_dev_size_in_blocks(struct block_device
*bdev
)
2425 sector_t metadata_dev_size
= get_metadata_dev_size(bdev
);
2427 sector_div(metadata_dev_size
, THIN_METADATA_BLOCK_SIZE
);
2429 return metadata_dev_size
;
2433 * When a metadata threshold is crossed a dm event is triggered, and
2434 * userland should respond by growing the metadata device. We could let
2435 * userland set the threshold, like we do with the data threshold, but I'm
2436 * not sure they know enough to do this well.
2438 static dm_block_t
calc_metadata_threshold(struct pool_c
*pt
)
2441 * 4M is ample for all ops with the possible exception of thin
2442 * device deletion which is harmless if it fails (just retry the
2443 * delete after you've grown the device).
2445 dm_block_t quarter
= get_metadata_dev_size_in_blocks(pt
->metadata_dev
->bdev
) / 4;
2446 return min((dm_block_t
)1024ULL /* 4M */, quarter
);
2450 * thin-pool <metadata dev> <data dev>
2451 * <data block size (sectors)>
2452 * <low water mark (blocks)>
2453 * [<#feature args> [<arg>]*]
2455 * Optional feature arguments are:
2456 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
2457 * ignore_discard: disable discard
2458 * no_discard_passdown: don't pass discards down to the data device
2459 * read_only: Don't allow any changes to be made to the pool metadata.
2460 * error_if_no_space: error IOs, instead of queueing, if no space.
2462 static int pool_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2464 int r
, pool_created
= 0;
2467 struct pool_features pf
;
2468 struct dm_arg_set as
;
2469 struct dm_dev
*data_dev
;
2470 unsigned long block_size
;
2471 dm_block_t low_water_blocks
;
2472 struct dm_dev
*metadata_dev
;
2473 fmode_t metadata_mode
;
2476 * FIXME Remove validation from scope of lock.
2478 mutex_lock(&dm_thin_pool_table
.mutex
);
2481 ti
->error
= "Invalid argument count";
2490 * Set default pool features.
2492 pool_features_init(&pf
);
2494 dm_consume_args(&as
, 4);
2495 r
= parse_pool_features(&as
, &pf
, ti
);
2499 metadata_mode
= FMODE_READ
| ((pf
.mode
== PM_READ_ONLY
) ? 0 : FMODE_WRITE
);
2500 r
= dm_get_device(ti
, argv
[0], metadata_mode
, &metadata_dev
);
2502 ti
->error
= "Error opening metadata block device";
2505 warn_if_metadata_device_too_big(metadata_dev
->bdev
);
2507 r
= dm_get_device(ti
, argv
[1], FMODE_READ
| FMODE_WRITE
, &data_dev
);
2509 ti
->error
= "Error getting data device";
2513 if (kstrtoul(argv
[2], 10, &block_size
) || !block_size
||
2514 block_size
< DATA_DEV_BLOCK_SIZE_MIN_SECTORS
||
2515 block_size
> DATA_DEV_BLOCK_SIZE_MAX_SECTORS
||
2516 block_size
& (DATA_DEV_BLOCK_SIZE_MIN_SECTORS
- 1)) {
2517 ti
->error
= "Invalid block size";
2522 if (kstrtoull(argv
[3], 10, (unsigned long long *)&low_water_blocks
)) {
2523 ti
->error
= "Invalid low water mark";
2528 pt
= kzalloc(sizeof(*pt
), GFP_KERNEL
);
2534 pool
= __pool_find(dm_table_get_md(ti
->table
), metadata_dev
->bdev
,
2535 block_size
, pf
.mode
== PM_READ_ONLY
, &ti
->error
, &pool_created
);
2542 * 'pool_created' reflects whether this is the first table load.
2543 * Top level discard support is not allowed to be changed after
2544 * initial load. This would require a pool reload to trigger thin
2547 if (!pool_created
&& pf
.discard_enabled
!= pool
->pf
.discard_enabled
) {
2548 ti
->error
= "Discard support cannot be disabled once enabled";
2550 goto out_flags_changed
;
2555 pt
->metadata_dev
= metadata_dev
;
2556 pt
->data_dev
= data_dev
;
2557 pt
->low_water_blocks
= low_water_blocks
;
2558 pt
->adjusted_pf
= pt
->requested_pf
= pf
;
2559 ti
->num_flush_bios
= 1;
2562 * Only need to enable discards if the pool should pass
2563 * them down to the data device. The thin device's discard
2564 * processing will cause mappings to be removed from the btree.
2566 ti
->discard_zeroes_data_unsupported
= true;
2567 if (pf
.discard_enabled
&& pf
.discard_passdown
) {
2568 ti
->num_discard_bios
= 1;
2571 * Setting 'discards_supported' circumvents the normal
2572 * stacking of discard limits (this keeps the pool and
2573 * thin devices' discard limits consistent).
2575 ti
->discards_supported
= true;
2579 r
= dm_pool_register_metadata_threshold(pt
->pool
->pmd
,
2580 calc_metadata_threshold(pt
),
2581 metadata_low_callback
,
2586 pt
->callbacks
.congested_fn
= pool_is_congested
;
2587 dm_table_add_target_callbacks(ti
->table
, &pt
->callbacks
);
2589 mutex_unlock(&dm_thin_pool_table
.mutex
);
2598 dm_put_device(ti
, data_dev
);
2600 dm_put_device(ti
, metadata_dev
);
2602 mutex_unlock(&dm_thin_pool_table
.mutex
);
2607 static int pool_map(struct dm_target
*ti
, struct bio
*bio
)
2610 struct pool_c
*pt
= ti
->private;
2611 struct pool
*pool
= pt
->pool
;
2612 unsigned long flags
;
2615 * As this is a singleton target, ti->begin is always zero.
2617 spin_lock_irqsave(&pool
->lock
, flags
);
2618 bio
->bi_bdev
= pt
->data_dev
->bdev
;
2619 r
= DM_MAPIO_REMAPPED
;
2620 spin_unlock_irqrestore(&pool
->lock
, flags
);
2625 static int maybe_resize_data_dev(struct dm_target
*ti
, bool *need_commit
)
2628 struct pool_c
*pt
= ti
->private;
2629 struct pool
*pool
= pt
->pool
;
2630 sector_t data_size
= ti
->len
;
2631 dm_block_t sb_data_size
;
2633 *need_commit
= false;
2635 (void) sector_div(data_size
, pool
->sectors_per_block
);
2637 r
= dm_pool_get_data_dev_size(pool
->pmd
, &sb_data_size
);
2639 DMERR("%s: failed to retrieve data device size",
2640 dm_device_name(pool
->pool_md
));
2644 if (data_size
< sb_data_size
) {
2645 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
2646 dm_device_name(pool
->pool_md
),
2647 (unsigned long long)data_size
, sb_data_size
);
2650 } else if (data_size
> sb_data_size
) {
2651 if (dm_pool_metadata_needs_check(pool
->pmd
)) {
2652 DMERR("%s: unable to grow the data device until repaired.",
2653 dm_device_name(pool
->pool_md
));
2658 DMINFO("%s: growing the data device from %llu to %llu blocks",
2659 dm_device_name(pool
->pool_md
),
2660 sb_data_size
, (unsigned long long)data_size
);
2661 r
= dm_pool_resize_data_dev(pool
->pmd
, data_size
);
2663 metadata_operation_failed(pool
, "dm_pool_resize_data_dev", r
);
2667 *need_commit
= true;
2673 static int maybe_resize_metadata_dev(struct dm_target
*ti
, bool *need_commit
)
2676 struct pool_c
*pt
= ti
->private;
2677 struct pool
*pool
= pt
->pool
;
2678 dm_block_t metadata_dev_size
, sb_metadata_dev_size
;
2680 *need_commit
= false;
2682 metadata_dev_size
= get_metadata_dev_size_in_blocks(pool
->md_dev
);
2684 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &sb_metadata_dev_size
);
2686 DMERR("%s: failed to retrieve metadata device size",
2687 dm_device_name(pool
->pool_md
));
2691 if (metadata_dev_size
< sb_metadata_dev_size
) {
2692 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
2693 dm_device_name(pool
->pool_md
),
2694 metadata_dev_size
, sb_metadata_dev_size
);
2697 } else if (metadata_dev_size
> sb_metadata_dev_size
) {
2698 if (dm_pool_metadata_needs_check(pool
->pmd
)) {
2699 DMERR("%s: unable to grow the metadata device until repaired.",
2700 dm_device_name(pool
->pool_md
));
2704 warn_if_metadata_device_too_big(pool
->md_dev
);
2705 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
2706 dm_device_name(pool
->pool_md
),
2707 sb_metadata_dev_size
, metadata_dev_size
);
2708 r
= dm_pool_resize_metadata_dev(pool
->pmd
, metadata_dev_size
);
2710 metadata_operation_failed(pool
, "dm_pool_resize_metadata_dev", r
);
2714 *need_commit
= true;
2721 * Retrieves the number of blocks of the data device from
2722 * the superblock and compares it to the actual device size,
2723 * thus resizing the data device in case it has grown.
2725 * This both copes with opening preallocated data devices in the ctr
2726 * being followed by a resume
2728 * calling the resume method individually after userspace has
2729 * grown the data device in reaction to a table event.
2731 static int pool_preresume(struct dm_target
*ti
)
2734 bool need_commit1
, need_commit2
;
2735 struct pool_c
*pt
= ti
->private;
2736 struct pool
*pool
= pt
->pool
;
2739 * Take control of the pool object.
2741 r
= bind_control_target(pool
, ti
);
2745 r
= maybe_resize_data_dev(ti
, &need_commit1
);
2749 r
= maybe_resize_metadata_dev(ti
, &need_commit2
);
2753 if (need_commit1
|| need_commit2
)
2754 (void) commit(pool
);
2759 static void pool_resume(struct dm_target
*ti
)
2761 struct pool_c
*pt
= ti
->private;
2762 struct pool
*pool
= pt
->pool
;
2763 unsigned long flags
;
2765 spin_lock_irqsave(&pool
->lock
, flags
);
2766 pool
->low_water_triggered
= false;
2767 spin_unlock_irqrestore(&pool
->lock
, flags
);
2770 do_waker(&pool
->waker
.work
);
2773 static void pool_postsuspend(struct dm_target
*ti
)
2775 struct pool_c
*pt
= ti
->private;
2776 struct pool
*pool
= pt
->pool
;
2778 cancel_delayed_work(&pool
->waker
);
2779 cancel_delayed_work(&pool
->no_space_timeout
);
2780 flush_workqueue(pool
->wq
);
2781 (void) commit(pool
);
2784 static int check_arg_count(unsigned argc
, unsigned args_required
)
2786 if (argc
!= args_required
) {
2787 DMWARN("Message received with %u arguments instead of %u.",
2788 argc
, args_required
);
2795 static int read_dev_id(char *arg
, dm_thin_id
*dev_id
, int warning
)
2797 if (!kstrtoull(arg
, 10, (unsigned long long *)dev_id
) &&
2798 *dev_id
<= MAX_DEV_ID
)
2802 DMWARN("Message received with invalid device id: %s", arg
);
2807 static int process_create_thin_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2812 r
= check_arg_count(argc
, 2);
2816 r
= read_dev_id(argv
[1], &dev_id
, 1);
2820 r
= dm_pool_create_thin(pool
->pmd
, dev_id
);
2822 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2830 static int process_create_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2833 dm_thin_id origin_dev_id
;
2836 r
= check_arg_count(argc
, 3);
2840 r
= read_dev_id(argv
[1], &dev_id
, 1);
2844 r
= read_dev_id(argv
[2], &origin_dev_id
, 1);
2848 r
= dm_pool_create_snap(pool
->pmd
, dev_id
, origin_dev_id
);
2850 DMWARN("Creation of new snapshot %s of device %s failed.",
2858 static int process_delete_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2863 r
= check_arg_count(argc
, 2);
2867 r
= read_dev_id(argv
[1], &dev_id
, 1);
2871 r
= dm_pool_delete_thin_device(pool
->pmd
, dev_id
);
2873 DMWARN("Deletion of thin device %s failed.", argv
[1]);
2878 static int process_set_transaction_id_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2880 dm_thin_id old_id
, new_id
;
2883 r
= check_arg_count(argc
, 3);
2887 if (kstrtoull(argv
[1], 10, (unsigned long long *)&old_id
)) {
2888 DMWARN("set_transaction_id message: Unrecognised id %s.", argv
[1]);
2892 if (kstrtoull(argv
[2], 10, (unsigned long long *)&new_id
)) {
2893 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv
[2]);
2897 r
= dm_pool_set_metadata_transaction_id(pool
->pmd
, old_id
, new_id
);
2899 DMWARN("Failed to change transaction id from %s to %s.",
2907 static int process_reserve_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2911 r
= check_arg_count(argc
, 1);
2915 (void) commit(pool
);
2917 r
= dm_pool_reserve_metadata_snap(pool
->pmd
);
2919 DMWARN("reserve_metadata_snap message failed.");
2924 static int process_release_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2928 r
= check_arg_count(argc
, 1);
2932 r
= dm_pool_release_metadata_snap(pool
->pmd
);
2934 DMWARN("release_metadata_snap message failed.");
2940 * Messages supported:
2941 * create_thin <dev_id>
2942 * create_snap <dev_id> <origin_id>
2944 * trim <dev_id> <new_size_in_sectors>
2945 * set_transaction_id <current_trans_id> <new_trans_id>
2946 * reserve_metadata_snap
2947 * release_metadata_snap
2949 static int pool_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
2952 struct pool_c
*pt
= ti
->private;
2953 struct pool
*pool
= pt
->pool
;
2955 if (!strcasecmp(argv
[0], "create_thin"))
2956 r
= process_create_thin_mesg(argc
, argv
, pool
);
2958 else if (!strcasecmp(argv
[0], "create_snap"))
2959 r
= process_create_snap_mesg(argc
, argv
, pool
);
2961 else if (!strcasecmp(argv
[0], "delete"))
2962 r
= process_delete_mesg(argc
, argv
, pool
);
2964 else if (!strcasecmp(argv
[0], "set_transaction_id"))
2965 r
= process_set_transaction_id_mesg(argc
, argv
, pool
);
2967 else if (!strcasecmp(argv
[0], "reserve_metadata_snap"))
2968 r
= process_reserve_metadata_snap_mesg(argc
, argv
, pool
);
2970 else if (!strcasecmp(argv
[0], "release_metadata_snap"))
2971 r
= process_release_metadata_snap_mesg(argc
, argv
, pool
);
2974 DMWARN("Unrecognised thin pool target message received: %s", argv
[0]);
2977 (void) commit(pool
);
2982 static void emit_flags(struct pool_features
*pf
, char *result
,
2983 unsigned sz
, unsigned maxlen
)
2985 unsigned count
= !pf
->zero_new_blocks
+ !pf
->discard_enabled
+
2986 !pf
->discard_passdown
+ (pf
->mode
== PM_READ_ONLY
) +
2987 pf
->error_if_no_space
;
2988 DMEMIT("%u ", count
);
2990 if (!pf
->zero_new_blocks
)
2991 DMEMIT("skip_block_zeroing ");
2993 if (!pf
->discard_enabled
)
2994 DMEMIT("ignore_discard ");
2996 if (!pf
->discard_passdown
)
2997 DMEMIT("no_discard_passdown ");
2999 if (pf
->mode
== PM_READ_ONLY
)
3000 DMEMIT("read_only ");
3002 if (pf
->error_if_no_space
)
3003 DMEMIT("error_if_no_space ");
3008 * <transaction id> <used metadata sectors>/<total metadata sectors>
3009 * <used data sectors>/<total data sectors> <held metadata root>
3011 static void pool_status(struct dm_target
*ti
, status_type_t type
,
3012 unsigned status_flags
, char *result
, unsigned maxlen
)
3016 uint64_t transaction_id
;
3017 dm_block_t nr_free_blocks_data
;
3018 dm_block_t nr_free_blocks_metadata
;
3019 dm_block_t nr_blocks_data
;
3020 dm_block_t nr_blocks_metadata
;
3021 dm_block_t held_root
;
3022 char buf
[BDEVNAME_SIZE
];
3023 char buf2
[BDEVNAME_SIZE
];
3024 struct pool_c
*pt
= ti
->private;
3025 struct pool
*pool
= pt
->pool
;
3028 case STATUSTYPE_INFO
:
3029 if (get_pool_mode(pool
) == PM_FAIL
) {
3034 /* Commit to ensure statistics aren't out-of-date */
3035 if (!(status_flags
& DM_STATUS_NOFLUSH_FLAG
) && !dm_suspended(ti
))
3036 (void) commit(pool
);
3038 r
= dm_pool_get_metadata_transaction_id(pool
->pmd
, &transaction_id
);
3040 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
3041 dm_device_name(pool
->pool_md
), r
);
3045 r
= dm_pool_get_free_metadata_block_count(pool
->pmd
, &nr_free_blocks_metadata
);
3047 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
3048 dm_device_name(pool
->pool_md
), r
);
3052 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &nr_blocks_metadata
);
3054 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
3055 dm_device_name(pool
->pool_md
), r
);
3059 r
= dm_pool_get_free_block_count(pool
->pmd
, &nr_free_blocks_data
);
3061 DMERR("%s: dm_pool_get_free_block_count returned %d",
3062 dm_device_name(pool
->pool_md
), r
);
3066 r
= dm_pool_get_data_dev_size(pool
->pmd
, &nr_blocks_data
);
3068 DMERR("%s: dm_pool_get_data_dev_size returned %d",
3069 dm_device_name(pool
->pool_md
), r
);
3073 r
= dm_pool_get_metadata_snap(pool
->pmd
, &held_root
);
3075 DMERR("%s: dm_pool_get_metadata_snap returned %d",
3076 dm_device_name(pool
->pool_md
), r
);
3080 DMEMIT("%llu %llu/%llu %llu/%llu ",
3081 (unsigned long long)transaction_id
,
3082 (unsigned long long)(nr_blocks_metadata
- nr_free_blocks_metadata
),
3083 (unsigned long long)nr_blocks_metadata
,
3084 (unsigned long long)(nr_blocks_data
- nr_free_blocks_data
),
3085 (unsigned long long)nr_blocks_data
);
3088 DMEMIT("%llu ", held_root
);
3092 if (pool
->pf
.mode
== PM_OUT_OF_DATA_SPACE
)
3093 DMEMIT("out_of_data_space ");
3094 else if (pool
->pf
.mode
== PM_READ_ONLY
)
3099 if (!pool
->pf
.discard_enabled
)
3100 DMEMIT("ignore_discard ");
3101 else if (pool
->pf
.discard_passdown
)
3102 DMEMIT("discard_passdown ");
3104 DMEMIT("no_discard_passdown ");
3106 if (pool
->pf
.error_if_no_space
)
3107 DMEMIT("error_if_no_space ");
3109 DMEMIT("queue_if_no_space ");
3113 case STATUSTYPE_TABLE
:
3114 DMEMIT("%s %s %lu %llu ",
3115 format_dev_t(buf
, pt
->metadata_dev
->bdev
->bd_dev
),
3116 format_dev_t(buf2
, pt
->data_dev
->bdev
->bd_dev
),
3117 (unsigned long)pool
->sectors_per_block
,
3118 (unsigned long long)pt
->low_water_blocks
);
3119 emit_flags(&pt
->requested_pf
, result
, sz
, maxlen
);
3128 static int pool_iterate_devices(struct dm_target
*ti
,
3129 iterate_devices_callout_fn fn
, void *data
)
3131 struct pool_c
*pt
= ti
->private;
3133 return fn(ti
, pt
->data_dev
, 0, ti
->len
, data
);
3136 static int pool_merge(struct dm_target
*ti
, struct bvec_merge_data
*bvm
,
3137 struct bio_vec
*biovec
, int max_size
)
3139 struct pool_c
*pt
= ti
->private;
3140 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
3142 if (!q
->merge_bvec_fn
)
3145 bvm
->bi_bdev
= pt
->data_dev
->bdev
;
3147 return min(max_size
, q
->merge_bvec_fn(q
, bvm
, biovec
));
3150 static void set_discard_limits(struct pool_c
*pt
, struct queue_limits
*limits
)
3152 struct pool
*pool
= pt
->pool
;
3153 struct queue_limits
*data_limits
;
3155 limits
->max_discard_sectors
= pool
->sectors_per_block
;
3158 * discard_granularity is just a hint, and not enforced.
3160 if (pt
->adjusted_pf
.discard_passdown
) {
3161 data_limits
= &bdev_get_queue(pt
->data_dev
->bdev
)->limits
;
3162 limits
->discard_granularity
= max(data_limits
->discard_granularity
,
3163 pool
->sectors_per_block
<< SECTOR_SHIFT
);
3165 limits
->discard_granularity
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
3168 static void pool_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
3170 struct pool_c
*pt
= ti
->private;
3171 struct pool
*pool
= pt
->pool
;
3172 uint64_t io_opt_sectors
= limits
->io_opt
>> SECTOR_SHIFT
;
3175 * If the system-determined stacked limits are compatible with the
3176 * pool's blocksize (io_opt is a factor) do not override them.
3178 if (io_opt_sectors
< pool
->sectors_per_block
||
3179 do_div(io_opt_sectors
, pool
->sectors_per_block
)) {
3180 blk_limits_io_min(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
3181 blk_limits_io_opt(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
3185 * pt->adjusted_pf is a staging area for the actual features to use.
3186 * They get transferred to the live pool in bind_control_target()
3187 * called from pool_preresume().
3189 if (!pt
->adjusted_pf
.discard_enabled
) {
3191 * Must explicitly disallow stacking discard limits otherwise the
3192 * block layer will stack them if pool's data device has support.
3193 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
3194 * user to see that, so make sure to set all discard limits to 0.
3196 limits
->discard_granularity
= 0;
3200 disable_passdown_if_not_supported(pt
);
3202 set_discard_limits(pt
, limits
);
3205 static struct target_type pool_target
= {
3206 .name
= "thin-pool",
3207 .features
= DM_TARGET_SINGLETON
| DM_TARGET_ALWAYS_WRITEABLE
|
3208 DM_TARGET_IMMUTABLE
,
3209 .version
= {1, 13, 0},
3210 .module
= THIS_MODULE
,
3214 .postsuspend
= pool_postsuspend
,
3215 .preresume
= pool_preresume
,
3216 .resume
= pool_resume
,
3217 .message
= pool_message
,
3218 .status
= pool_status
,
3219 .merge
= pool_merge
,
3220 .iterate_devices
= pool_iterate_devices
,
3221 .io_hints
= pool_io_hints
,
3224 /*----------------------------------------------------------------
3225 * Thin target methods
3226 *--------------------------------------------------------------*/
3227 static void thin_get(struct thin_c
*tc
)
3229 atomic_inc(&tc
->refcount
);
3232 static void thin_put(struct thin_c
*tc
)
3234 if (atomic_dec_and_test(&tc
->refcount
))
3235 complete(&tc
->can_destroy
);
3238 static void thin_dtr(struct dm_target
*ti
)
3240 struct thin_c
*tc
= ti
->private;
3241 unsigned long flags
;
3244 wait_for_completion(&tc
->can_destroy
);
3246 spin_lock_irqsave(&tc
->pool
->lock
, flags
);
3247 list_del_rcu(&tc
->list
);
3248 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
3251 mutex_lock(&dm_thin_pool_table
.mutex
);
3253 __pool_dec(tc
->pool
);
3254 dm_pool_close_thin_device(tc
->td
);
3255 dm_put_device(ti
, tc
->pool_dev
);
3257 dm_put_device(ti
, tc
->origin_dev
);
3260 mutex_unlock(&dm_thin_pool_table
.mutex
);
3264 * Thin target parameters:
3266 * <pool_dev> <dev_id> [origin_dev]
3268 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
3269 * dev_id: the internal device identifier
3270 * origin_dev: a device external to the pool that should act as the origin
3272 * If the pool device has discards disabled, they get disabled for the thin
3275 static int thin_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
3279 struct dm_dev
*pool_dev
, *origin_dev
;
3280 struct mapped_device
*pool_md
;
3281 unsigned long flags
;
3283 mutex_lock(&dm_thin_pool_table
.mutex
);
3285 if (argc
!= 2 && argc
!= 3) {
3286 ti
->error
= "Invalid argument count";
3291 tc
= ti
->private = kzalloc(sizeof(*tc
), GFP_KERNEL
);
3293 ti
->error
= "Out of memory";
3297 spin_lock_init(&tc
->lock
);
3298 bio_list_init(&tc
->deferred_bio_list
);
3299 bio_list_init(&tc
->retry_on_resume_list
);
3300 tc
->sort_bio_list
= RB_ROOT
;
3303 r
= dm_get_device(ti
, argv
[2], FMODE_READ
, &origin_dev
);
3305 ti
->error
= "Error opening origin device";
3306 goto bad_origin_dev
;
3308 tc
->origin_dev
= origin_dev
;
3311 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &pool_dev
);
3313 ti
->error
= "Error opening pool device";
3316 tc
->pool_dev
= pool_dev
;
3318 if (read_dev_id(argv
[1], (unsigned long long *)&tc
->dev_id
, 0)) {
3319 ti
->error
= "Invalid device id";
3324 pool_md
= dm_get_md(tc
->pool_dev
->bdev
->bd_dev
);
3326 ti
->error
= "Couldn't get pool mapped device";
3331 tc
->pool
= __pool_table_lookup(pool_md
);
3333 ti
->error
= "Couldn't find pool object";
3335 goto bad_pool_lookup
;
3337 __pool_inc(tc
->pool
);
3339 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
3340 ti
->error
= "Couldn't open thin device, Pool is in fail mode";
3345 r
= dm_pool_open_thin_device(tc
->pool
->pmd
, tc
->dev_id
, &tc
->td
);
3347 ti
->error
= "Couldn't open thin internal device";
3351 r
= dm_set_target_max_io_len(ti
, tc
->pool
->sectors_per_block
);
3353 goto bad_target_max_io_len
;
3355 ti
->num_flush_bios
= 1;
3356 ti
->flush_supported
= true;
3357 ti
->per_bio_data_size
= sizeof(struct dm_thin_endio_hook
);
3359 /* In case the pool supports discards, pass them on. */
3360 ti
->discard_zeroes_data_unsupported
= true;
3361 if (tc
->pool
->pf
.discard_enabled
) {
3362 ti
->discards_supported
= true;
3363 ti
->num_discard_bios
= 1;
3364 /* Discard bios must be split on a block boundary */
3365 ti
->split_discard_bios
= true;
3370 mutex_unlock(&dm_thin_pool_table
.mutex
);
3372 atomic_set(&tc
->refcount
, 1);
3373 init_completion(&tc
->can_destroy
);
3375 spin_lock_irqsave(&tc
->pool
->lock
, flags
);
3376 list_add_tail_rcu(&tc
->list
, &tc
->pool
->active_thins
);
3377 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
3379 * This synchronize_rcu() call is needed here otherwise we risk a
3380 * wake_worker() call finding no bios to process (because the newly
3381 * added tc isn't yet visible). So this reduces latency since we
3382 * aren't then dependent on the periodic commit to wake_worker().
3388 bad_target_max_io_len
:
3389 dm_pool_close_thin_device(tc
->td
);
3391 __pool_dec(tc
->pool
);
3395 dm_put_device(ti
, tc
->pool_dev
);
3398 dm_put_device(ti
, tc
->origin_dev
);
3402 mutex_unlock(&dm_thin_pool_table
.mutex
);
3407 static int thin_map(struct dm_target
*ti
, struct bio
*bio
)
3409 bio
->bi_iter
.bi_sector
= dm_target_offset(ti
, bio
->bi_iter
.bi_sector
);
3411 return thin_bio_map(ti
, bio
);
3414 static int thin_endio(struct dm_target
*ti
, struct bio
*bio
, int err
)
3416 unsigned long flags
;
3417 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
3418 struct list_head work
;
3419 struct dm_thin_new_mapping
*m
, *tmp
;
3420 struct pool
*pool
= h
->tc
->pool
;
3422 if (h
->shared_read_entry
) {
3423 INIT_LIST_HEAD(&work
);
3424 dm_deferred_entry_dec(h
->shared_read_entry
, &work
);
3426 spin_lock_irqsave(&pool
->lock
, flags
);
3427 list_for_each_entry_safe(m
, tmp
, &work
, list
) {
3429 __complete_mapping_preparation(m
);
3431 spin_unlock_irqrestore(&pool
->lock
, flags
);
3434 if (h
->all_io_entry
) {
3435 INIT_LIST_HEAD(&work
);
3436 dm_deferred_entry_dec(h
->all_io_entry
, &work
);
3437 if (!list_empty(&work
)) {
3438 spin_lock_irqsave(&pool
->lock
, flags
);
3439 list_for_each_entry_safe(m
, tmp
, &work
, list
)
3440 list_add_tail(&m
->list
, &pool
->prepared_discards
);
3441 spin_unlock_irqrestore(&pool
->lock
, flags
);
3449 static void thin_presuspend(struct dm_target
*ti
)
3451 struct thin_c
*tc
= ti
->private;
3453 if (dm_noflush_suspending(ti
))
3454 noflush_work(tc
, do_noflush_start
);
3457 static void thin_postsuspend(struct dm_target
*ti
)
3459 struct thin_c
*tc
= ti
->private;
3462 * The dm_noflush_suspending flag has been cleared by now, so
3463 * unfortunately we must always run this.
3465 noflush_work(tc
, do_noflush_stop
);
3468 static int thin_preresume(struct dm_target
*ti
)
3470 struct thin_c
*tc
= ti
->private;
3473 tc
->origin_size
= get_dev_size(tc
->origin_dev
->bdev
);
3479 * <nr mapped sectors> <highest mapped sector>
3481 static void thin_status(struct dm_target
*ti
, status_type_t type
,
3482 unsigned status_flags
, char *result
, unsigned maxlen
)
3486 dm_block_t mapped
, highest
;
3487 char buf
[BDEVNAME_SIZE
];
3488 struct thin_c
*tc
= ti
->private;
3490 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
3499 case STATUSTYPE_INFO
:
3500 r
= dm_thin_get_mapped_count(tc
->td
, &mapped
);
3502 DMERR("dm_thin_get_mapped_count returned %d", r
);
3506 r
= dm_thin_get_highest_mapped_block(tc
->td
, &highest
);
3508 DMERR("dm_thin_get_highest_mapped_block returned %d", r
);
3512 DMEMIT("%llu ", mapped
* tc
->pool
->sectors_per_block
);
3514 DMEMIT("%llu", ((highest
+ 1) *
3515 tc
->pool
->sectors_per_block
) - 1);
3520 case STATUSTYPE_TABLE
:
3522 format_dev_t(buf
, tc
->pool_dev
->bdev
->bd_dev
),
3523 (unsigned long) tc
->dev_id
);
3525 DMEMIT(" %s", format_dev_t(buf
, tc
->origin_dev
->bdev
->bd_dev
));
3536 static int thin_iterate_devices(struct dm_target
*ti
,
3537 iterate_devices_callout_fn fn
, void *data
)
3540 struct thin_c
*tc
= ti
->private;
3541 struct pool
*pool
= tc
->pool
;
3544 * We can't call dm_pool_get_data_dev_size() since that blocks. So
3545 * we follow a more convoluted path through to the pool's target.
3548 return 0; /* nothing is bound */
3550 blocks
= pool
->ti
->len
;
3551 (void) sector_div(blocks
, pool
->sectors_per_block
);
3553 return fn(ti
, tc
->pool_dev
, 0, pool
->sectors_per_block
* blocks
, data
);
3558 static struct target_type thin_target
= {
3560 .version
= {1, 13, 0},
3561 .module
= THIS_MODULE
,
3565 .end_io
= thin_endio
,
3566 .preresume
= thin_preresume
,
3567 .presuspend
= thin_presuspend
,
3568 .postsuspend
= thin_postsuspend
,
3569 .status
= thin_status
,
3570 .iterate_devices
= thin_iterate_devices
,
3573 /*----------------------------------------------------------------*/
3575 static int __init
dm_thin_init(void)
3581 r
= dm_register_target(&thin_target
);
3585 r
= dm_register_target(&pool_target
);
3587 goto bad_pool_target
;
3591 _new_mapping_cache
= KMEM_CACHE(dm_thin_new_mapping
, 0);
3592 if (!_new_mapping_cache
)
3593 goto bad_new_mapping_cache
;
3597 bad_new_mapping_cache
:
3598 dm_unregister_target(&pool_target
);
3600 dm_unregister_target(&thin_target
);
3605 static void dm_thin_exit(void)
3607 dm_unregister_target(&thin_target
);
3608 dm_unregister_target(&pool_target
);
3610 kmem_cache_destroy(_new_mapping_cache
);
3613 module_init(dm_thin_init
);
3614 module_exit(dm_thin_exit
);
3616 module_param_named(no_space_timeout
, no_space_timeout_secs
, uint
, S_IRUGO
| S_IWUSR
);
3617 MODULE_PARM_DESC(no_space_timeout
, "Out of data space queue IO timeout in seconds");
3619 MODULE_DESCRIPTION(DM_NAME
" thin provisioning target");
3620 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3621 MODULE_LICENSE("GPL");