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/init.h>
16 #include <linux/module.h>
17 #include <linux/slab.h>
19 #define DM_MSG_PREFIX "thin"
24 #define ENDIO_HOOK_POOL_SIZE 1024
25 #define MAPPING_POOL_SIZE 1024
26 #define PRISON_CELLS 1024
27 #define COMMIT_PERIOD HZ
29 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle
,
30 "A percentage of time allocated for copy on write");
33 * The block size of the device holding pool data must be
34 * between 64KB and 1GB.
36 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
37 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
40 * Device id is restricted to 24 bits.
42 #define MAX_DEV_ID ((1 << 24) - 1)
45 * How do we handle breaking sharing of data blocks?
46 * =================================================
48 * We use a standard copy-on-write btree to store the mappings for the
49 * devices (note I'm talking about copy-on-write of the metadata here, not
50 * the data). When you take an internal snapshot you clone the root node
51 * of the origin btree. After this there is no concept of an origin or a
52 * snapshot. They are just two device trees that happen to point to the
55 * When we get a write in we decide if it's to a shared data block using
56 * some timestamp magic. If it is, we have to break sharing.
58 * Let's say we write to a shared block in what was the origin. The
61 * i) plug io further to this physical block. (see bio_prison code).
63 * ii) quiesce any read io to that shared data block. Obviously
64 * including all devices that share this block. (see dm_deferred_set code)
66 * iii) copy the data block to a newly allocate block. This step can be
67 * missed out if the io covers the block. (schedule_copy).
69 * iv) insert the new mapping into the origin's btree
70 * (process_prepared_mapping). This act of inserting breaks some
71 * sharing of btree nodes between the two devices. Breaking sharing only
72 * effects the btree of that specific device. Btrees for the other
73 * devices that share the block never change. The btree for the origin
74 * device as it was after the last commit is untouched, ie. we're using
75 * persistent data structures in the functional programming sense.
77 * v) unplug io to this physical block, including the io that triggered
78 * the breaking of sharing.
80 * Steps (ii) and (iii) occur in parallel.
82 * The metadata _doesn't_ need to be committed before the io continues. We
83 * get away with this because the io is always written to a _new_ block.
84 * If there's a crash, then:
86 * - The origin mapping will point to the old origin block (the shared
87 * one). This will contain the data as it was before the io that triggered
88 * the breaking of sharing came in.
90 * - The snap mapping still points to the old block. As it would after
93 * The downside of this scheme is the timestamp magic isn't perfect, and
94 * will continue to think that data block in the snapshot device is shared
95 * even after the write to the origin has broken sharing. I suspect data
96 * blocks will typically be shared by many different devices, so we're
97 * breaking sharing n + 1 times, rather than n, where n is the number of
98 * devices that reference this data block. At the moment I think the
99 * benefits far, far outweigh the disadvantages.
102 /*----------------------------------------------------------------*/
107 static void build_data_key(struct dm_thin_device
*td
,
108 dm_block_t b
, struct dm_cell_key
*key
)
111 key
->dev
= dm_thin_dev_id(td
);
115 static void build_virtual_key(struct dm_thin_device
*td
, dm_block_t b
,
116 struct dm_cell_key
*key
)
119 key
->dev
= dm_thin_dev_id(td
);
123 /*----------------------------------------------------------------*/
126 * A pool device ties together a metadata device and a data device. It
127 * also provides the interface for creating and destroying internal
130 struct dm_thin_new_mapping
;
133 * The pool runs in 3 modes. Ordered in degraded order for comparisons.
136 PM_WRITE
, /* metadata may be changed */
137 PM_READ_ONLY
, /* metadata may not be changed */
138 PM_FAIL
, /* all I/O fails */
141 struct pool_features
{
144 bool zero_new_blocks
:1;
145 bool discard_enabled
:1;
146 bool discard_passdown
:1;
150 typedef void (*process_bio_fn
)(struct thin_c
*tc
, struct bio
*bio
);
151 typedef void (*process_mapping_fn
)(struct dm_thin_new_mapping
*m
);
154 struct list_head list
;
155 struct dm_target
*ti
; /* Only set if a pool target is bound */
157 struct mapped_device
*pool_md
;
158 struct block_device
*md_dev
;
159 struct dm_pool_metadata
*pmd
;
161 dm_block_t low_water_blocks
;
162 uint32_t sectors_per_block
;
163 int sectors_per_block_shift
;
165 struct pool_features pf
;
166 unsigned low_water_triggered
:1; /* A dm event has been sent */
167 unsigned no_free_space
:1; /* A -ENOSPC warning has been issued */
169 struct dm_bio_prison
*prison
;
170 struct dm_kcopyd_client
*copier
;
172 struct workqueue_struct
*wq
;
173 struct work_struct worker
;
174 struct delayed_work waker
;
176 unsigned long last_commit_jiffies
;
180 struct bio_list deferred_bios
;
181 struct bio_list deferred_flush_bios
;
182 struct list_head prepared_mappings
;
183 struct list_head prepared_discards
;
185 struct bio_list retry_on_resume_list
;
187 struct dm_deferred_set
*shared_read_ds
;
188 struct dm_deferred_set
*all_io_ds
;
190 struct dm_thin_new_mapping
*next_mapping
;
191 mempool_t
*mapping_pool
;
193 process_bio_fn process_bio
;
194 process_bio_fn process_discard
;
196 process_mapping_fn process_prepared_mapping
;
197 process_mapping_fn process_prepared_discard
;
200 static enum pool_mode
get_pool_mode(struct pool
*pool
);
201 static void set_pool_mode(struct pool
*pool
, enum pool_mode mode
);
204 * Target context for a pool.
207 struct dm_target
*ti
;
209 struct dm_dev
*data_dev
;
210 struct dm_dev
*metadata_dev
;
211 struct dm_target_callbacks callbacks
;
213 dm_block_t low_water_blocks
;
214 struct pool_features requested_pf
; /* Features requested during table load */
215 struct pool_features adjusted_pf
; /* Features used after adjusting for constituent devices */
219 * Target context for a thin.
222 struct dm_dev
*pool_dev
;
223 struct dm_dev
*origin_dev
;
227 struct dm_thin_device
*td
;
230 /*----------------------------------------------------------------*/
233 * wake_worker() is used when new work is queued and when pool_resume is
234 * ready to continue deferred IO processing.
236 static void wake_worker(struct pool
*pool
)
238 queue_work(pool
->wq
, &pool
->worker
);
241 /*----------------------------------------------------------------*/
243 static int bio_detain(struct pool
*pool
, struct dm_cell_key
*key
, struct bio
*bio
,
244 struct dm_bio_prison_cell
**cell_result
)
247 struct dm_bio_prison_cell
*cell_prealloc
;
250 * Allocate a cell from the prison's mempool.
251 * This might block but it can't fail.
253 cell_prealloc
= dm_bio_prison_alloc_cell(pool
->prison
, GFP_NOIO
);
255 r
= dm_bio_detain(pool
->prison
, key
, bio
, cell_prealloc
, cell_result
);
258 * We reused an old cell; we can get rid of
261 dm_bio_prison_free_cell(pool
->prison
, cell_prealloc
);
266 static void cell_release(struct pool
*pool
,
267 struct dm_bio_prison_cell
*cell
,
268 struct bio_list
*bios
)
270 dm_cell_release(pool
->prison
, cell
, bios
);
271 dm_bio_prison_free_cell(pool
->prison
, cell
);
274 static void cell_release_no_holder(struct pool
*pool
,
275 struct dm_bio_prison_cell
*cell
,
276 struct bio_list
*bios
)
278 dm_cell_release_no_holder(pool
->prison
, cell
, bios
);
279 dm_bio_prison_free_cell(pool
->prison
, cell
);
282 static void cell_defer_no_holder_no_free(struct thin_c
*tc
,
283 struct dm_bio_prison_cell
*cell
)
285 struct pool
*pool
= tc
->pool
;
288 spin_lock_irqsave(&pool
->lock
, flags
);
289 dm_cell_release_no_holder(pool
->prison
, cell
, &pool
->deferred_bios
);
290 spin_unlock_irqrestore(&pool
->lock
, flags
);
295 static void cell_error(struct pool
*pool
,
296 struct dm_bio_prison_cell
*cell
)
298 dm_cell_error(pool
->prison
, cell
);
299 dm_bio_prison_free_cell(pool
->prison
, cell
);
302 /*----------------------------------------------------------------*/
305 * A global list of pools that uses a struct mapped_device as a key.
307 static struct dm_thin_pool_table
{
309 struct list_head pools
;
310 } dm_thin_pool_table
;
312 static void pool_table_init(void)
314 mutex_init(&dm_thin_pool_table
.mutex
);
315 INIT_LIST_HEAD(&dm_thin_pool_table
.pools
);
318 static void __pool_table_insert(struct pool
*pool
)
320 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
321 list_add(&pool
->list
, &dm_thin_pool_table
.pools
);
324 static void __pool_table_remove(struct pool
*pool
)
326 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
327 list_del(&pool
->list
);
330 static struct pool
*__pool_table_lookup(struct mapped_device
*md
)
332 struct pool
*pool
= NULL
, *tmp
;
334 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
336 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
337 if (tmp
->pool_md
== md
) {
346 static struct pool
*__pool_table_lookup_metadata_dev(struct block_device
*md_dev
)
348 struct pool
*pool
= NULL
, *tmp
;
350 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
352 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
353 if (tmp
->md_dev
== md_dev
) {
362 /*----------------------------------------------------------------*/
364 struct dm_thin_endio_hook
{
366 struct dm_deferred_entry
*shared_read_entry
;
367 struct dm_deferred_entry
*all_io_entry
;
368 struct dm_thin_new_mapping
*overwrite_mapping
;
371 static void __requeue_bio_list(struct thin_c
*tc
, struct bio_list
*master
)
374 struct bio_list bios
;
376 bio_list_init(&bios
);
377 bio_list_merge(&bios
, master
);
378 bio_list_init(master
);
380 while ((bio
= bio_list_pop(&bios
))) {
381 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
384 bio_endio(bio
, DM_ENDIO_REQUEUE
);
386 bio_list_add(master
, bio
);
390 static void requeue_io(struct thin_c
*tc
)
392 struct pool
*pool
= tc
->pool
;
395 spin_lock_irqsave(&pool
->lock
, flags
);
396 __requeue_bio_list(tc
, &pool
->deferred_bios
);
397 __requeue_bio_list(tc
, &pool
->retry_on_resume_list
);
398 spin_unlock_irqrestore(&pool
->lock
, flags
);
402 * This section of code contains the logic for processing a thin device's IO.
403 * Much of the code depends on pool object resources (lists, workqueues, etc)
404 * but most is exclusively called from the thin target rather than the thin-pool
408 static bool block_size_is_power_of_two(struct pool
*pool
)
410 return pool
->sectors_per_block_shift
>= 0;
413 static dm_block_t
get_bio_block(struct thin_c
*tc
, struct bio
*bio
)
415 struct pool
*pool
= tc
->pool
;
416 sector_t block_nr
= bio
->bi_iter
.bi_sector
;
418 if (block_size_is_power_of_two(pool
))
419 block_nr
>>= pool
->sectors_per_block_shift
;
421 (void) sector_div(block_nr
, pool
->sectors_per_block
);
426 static void remap(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
)
428 struct pool
*pool
= tc
->pool
;
429 sector_t bi_sector
= bio
->bi_iter
.bi_sector
;
431 bio
->bi_bdev
= tc
->pool_dev
->bdev
;
432 if (block_size_is_power_of_two(pool
))
433 bio
->bi_iter
.bi_sector
=
434 (block
<< pool
->sectors_per_block_shift
) |
435 (bi_sector
& (pool
->sectors_per_block
- 1));
437 bio
->bi_iter
.bi_sector
= (block
* pool
->sectors_per_block
) +
438 sector_div(bi_sector
, pool
->sectors_per_block
);
441 static void remap_to_origin(struct thin_c
*tc
, struct bio
*bio
)
443 bio
->bi_bdev
= tc
->origin_dev
->bdev
;
446 static int bio_triggers_commit(struct thin_c
*tc
, struct bio
*bio
)
448 return (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) &&
449 dm_thin_changed_this_transaction(tc
->td
);
452 static void inc_all_io_entry(struct pool
*pool
, struct bio
*bio
)
454 struct dm_thin_endio_hook
*h
;
456 if (bio
->bi_rw
& REQ_DISCARD
)
459 h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
460 h
->all_io_entry
= dm_deferred_entry_inc(pool
->all_io_ds
);
463 static void issue(struct thin_c
*tc
, struct bio
*bio
)
465 struct pool
*pool
= tc
->pool
;
468 if (!bio_triggers_commit(tc
, bio
)) {
469 generic_make_request(bio
);
474 * Complete bio with an error if earlier I/O caused changes to
475 * the metadata that can't be committed e.g, due to I/O errors
476 * on the metadata device.
478 if (dm_thin_aborted_changes(tc
->td
)) {
484 * Batch together any bios that trigger commits and then issue a
485 * single commit for them in process_deferred_bios().
487 spin_lock_irqsave(&pool
->lock
, flags
);
488 bio_list_add(&pool
->deferred_flush_bios
, bio
);
489 spin_unlock_irqrestore(&pool
->lock
, flags
);
492 static void remap_to_origin_and_issue(struct thin_c
*tc
, struct bio
*bio
)
494 remap_to_origin(tc
, bio
);
498 static void remap_and_issue(struct thin_c
*tc
, struct bio
*bio
,
501 remap(tc
, bio
, block
);
505 /*----------------------------------------------------------------*/
508 * Bio endio functions.
510 struct dm_thin_new_mapping
{
511 struct list_head list
;
515 unsigned pass_discard
:1;
518 dm_block_t virt_block
;
519 dm_block_t data_block
;
520 struct dm_bio_prison_cell
*cell
, *cell2
;
524 * If the bio covers the whole area of a block then we can avoid
525 * zeroing or copying. Instead this bio is hooked. The bio will
526 * still be in the cell, so care has to be taken to avoid issuing
530 bio_end_io_t
*saved_bi_end_io
;
533 static void __maybe_add_mapping(struct dm_thin_new_mapping
*m
)
535 struct pool
*pool
= m
->tc
->pool
;
537 if (m
->quiesced
&& m
->prepared
) {
538 list_add(&m
->list
, &pool
->prepared_mappings
);
543 static void copy_complete(int read_err
, unsigned long write_err
, void *context
)
546 struct dm_thin_new_mapping
*m
= context
;
547 struct pool
*pool
= m
->tc
->pool
;
549 m
->err
= read_err
|| write_err
? -EIO
: 0;
551 spin_lock_irqsave(&pool
->lock
, flags
);
553 __maybe_add_mapping(m
);
554 spin_unlock_irqrestore(&pool
->lock
, flags
);
557 static void overwrite_endio(struct bio
*bio
, int err
)
560 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
561 struct dm_thin_new_mapping
*m
= h
->overwrite_mapping
;
562 struct pool
*pool
= m
->tc
->pool
;
566 spin_lock_irqsave(&pool
->lock
, flags
);
568 __maybe_add_mapping(m
);
569 spin_unlock_irqrestore(&pool
->lock
, flags
);
572 /*----------------------------------------------------------------*/
579 * Prepared mapping jobs.
583 * This sends the bios in the cell back to the deferred_bios list.
585 static void cell_defer(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
587 struct pool
*pool
= tc
->pool
;
590 spin_lock_irqsave(&pool
->lock
, flags
);
591 cell_release(pool
, cell
, &pool
->deferred_bios
);
592 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
598 * Same as cell_defer above, except it omits the original holder of the cell.
600 static void cell_defer_no_holder(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
602 struct pool
*pool
= tc
->pool
;
605 spin_lock_irqsave(&pool
->lock
, flags
);
606 cell_release_no_holder(pool
, cell
, &pool
->deferred_bios
);
607 spin_unlock_irqrestore(&pool
->lock
, flags
);
612 static void process_prepared_mapping_fail(struct dm_thin_new_mapping
*m
)
615 m
->bio
->bi_end_io
= m
->saved_bi_end_io
;
616 atomic_inc(&m
->bio
->bi_remaining
);
618 cell_error(m
->tc
->pool
, m
->cell
);
620 mempool_free(m
, m
->tc
->pool
->mapping_pool
);
623 static void process_prepared_mapping(struct dm_thin_new_mapping
*m
)
625 struct thin_c
*tc
= m
->tc
;
626 struct pool
*pool
= tc
->pool
;
632 bio
->bi_end_io
= m
->saved_bi_end_io
;
633 atomic_inc(&bio
->bi_remaining
);
637 cell_error(pool
, m
->cell
);
642 * Commit the prepared block into the mapping btree.
643 * Any I/O for this block arriving after this point will get
644 * remapped to it directly.
646 r
= dm_thin_insert_block(tc
->td
, m
->virt_block
, m
->data_block
);
648 DMERR_LIMIT("dm_thin_insert_block() failed");
649 cell_error(pool
, m
->cell
);
654 * Release any bios held while the block was being provisioned.
655 * If we are processing a write bio that completely covers the block,
656 * we already processed it so can ignore it now when processing
657 * the bios in the cell.
660 cell_defer_no_holder(tc
, m
->cell
);
663 cell_defer(tc
, m
->cell
);
667 mempool_free(m
, pool
->mapping_pool
);
670 static void process_prepared_discard_fail(struct dm_thin_new_mapping
*m
)
672 struct thin_c
*tc
= m
->tc
;
674 bio_io_error(m
->bio
);
675 cell_defer_no_holder(tc
, m
->cell
);
676 cell_defer_no_holder(tc
, m
->cell2
);
677 mempool_free(m
, tc
->pool
->mapping_pool
);
680 static void process_prepared_discard_passdown(struct dm_thin_new_mapping
*m
)
682 struct thin_c
*tc
= m
->tc
;
684 inc_all_io_entry(tc
->pool
, m
->bio
);
685 cell_defer_no_holder(tc
, m
->cell
);
686 cell_defer_no_holder(tc
, m
->cell2
);
689 remap_and_issue(tc
, m
->bio
, m
->data_block
);
691 bio_endio(m
->bio
, 0);
693 mempool_free(m
, tc
->pool
->mapping_pool
);
696 static void process_prepared_discard(struct dm_thin_new_mapping
*m
)
699 struct thin_c
*tc
= m
->tc
;
701 r
= dm_thin_remove_block(tc
->td
, m
->virt_block
);
703 DMERR_LIMIT("dm_thin_remove_block() failed");
705 process_prepared_discard_passdown(m
);
708 static void process_prepared(struct pool
*pool
, struct list_head
*head
,
709 process_mapping_fn
*fn
)
712 struct list_head maps
;
713 struct dm_thin_new_mapping
*m
, *tmp
;
715 INIT_LIST_HEAD(&maps
);
716 spin_lock_irqsave(&pool
->lock
, flags
);
717 list_splice_init(head
, &maps
);
718 spin_unlock_irqrestore(&pool
->lock
, flags
);
720 list_for_each_entry_safe(m
, tmp
, &maps
, list
)
727 static int io_overlaps_block(struct pool
*pool
, struct bio
*bio
)
729 return bio
->bi_iter
.bi_size
==
730 (pool
->sectors_per_block
<< SECTOR_SHIFT
);
733 static int io_overwrites_block(struct pool
*pool
, struct bio
*bio
)
735 return (bio_data_dir(bio
) == WRITE
) &&
736 io_overlaps_block(pool
, bio
);
739 static void save_and_set_endio(struct bio
*bio
, bio_end_io_t
**save
,
742 *save
= bio
->bi_end_io
;
746 static int ensure_next_mapping(struct pool
*pool
)
748 if (pool
->next_mapping
)
751 pool
->next_mapping
= mempool_alloc(pool
->mapping_pool
, GFP_ATOMIC
);
753 return pool
->next_mapping
? 0 : -ENOMEM
;
756 static struct dm_thin_new_mapping
*get_next_mapping(struct pool
*pool
)
758 struct dm_thin_new_mapping
*r
= pool
->next_mapping
;
760 BUG_ON(!pool
->next_mapping
);
762 pool
->next_mapping
= NULL
;
767 static void schedule_copy(struct thin_c
*tc
, dm_block_t virt_block
,
768 struct dm_dev
*origin
, dm_block_t data_origin
,
769 dm_block_t data_dest
,
770 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
773 struct pool
*pool
= tc
->pool
;
774 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
776 INIT_LIST_HEAD(&m
->list
);
780 m
->virt_block
= virt_block
;
781 m
->data_block
= data_dest
;
786 if (!dm_deferred_set_add_work(pool
->shared_read_ds
, &m
->list
))
790 * IO to pool_dev remaps to the pool target's data_dev.
792 * If the whole block of data is being overwritten, we can issue the
793 * bio immediately. Otherwise we use kcopyd to clone the data first.
795 if (io_overwrites_block(pool
, bio
)) {
796 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
798 h
->overwrite_mapping
= m
;
800 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
801 inc_all_io_entry(pool
, bio
);
802 remap_and_issue(tc
, bio
, data_dest
);
804 struct dm_io_region from
, to
;
806 from
.bdev
= origin
->bdev
;
807 from
.sector
= data_origin
* pool
->sectors_per_block
;
808 from
.count
= pool
->sectors_per_block
;
810 to
.bdev
= tc
->pool_dev
->bdev
;
811 to
.sector
= data_dest
* pool
->sectors_per_block
;
812 to
.count
= pool
->sectors_per_block
;
814 r
= dm_kcopyd_copy(pool
->copier
, &from
, 1, &to
,
815 0, copy_complete
, m
);
817 mempool_free(m
, pool
->mapping_pool
);
818 DMERR_LIMIT("dm_kcopyd_copy() failed");
819 cell_error(pool
, cell
);
824 static void schedule_internal_copy(struct thin_c
*tc
, dm_block_t virt_block
,
825 dm_block_t data_origin
, dm_block_t data_dest
,
826 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
828 schedule_copy(tc
, virt_block
, tc
->pool_dev
,
829 data_origin
, data_dest
, cell
, bio
);
832 static void schedule_external_copy(struct thin_c
*tc
, dm_block_t virt_block
,
833 dm_block_t data_dest
,
834 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
836 schedule_copy(tc
, virt_block
, tc
->origin_dev
,
837 virt_block
, data_dest
, cell
, bio
);
840 static void schedule_zero(struct thin_c
*tc
, dm_block_t virt_block
,
841 dm_block_t data_block
, struct dm_bio_prison_cell
*cell
,
844 struct pool
*pool
= tc
->pool
;
845 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
847 INIT_LIST_HEAD(&m
->list
);
851 m
->virt_block
= virt_block
;
852 m
->data_block
= data_block
;
858 * If the whole block of data is being overwritten or we are not
859 * zeroing pre-existing data, we can issue the bio immediately.
860 * Otherwise we use kcopyd to zero the data first.
862 if (!pool
->pf
.zero_new_blocks
)
863 process_prepared_mapping(m
);
865 else if (io_overwrites_block(pool
, bio
)) {
866 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
868 h
->overwrite_mapping
= m
;
870 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
871 inc_all_io_entry(pool
, bio
);
872 remap_and_issue(tc
, bio
, data_block
);
875 struct dm_io_region to
;
877 to
.bdev
= tc
->pool_dev
->bdev
;
878 to
.sector
= data_block
* pool
->sectors_per_block
;
879 to
.count
= pool
->sectors_per_block
;
881 r
= dm_kcopyd_zero(pool
->copier
, 1, &to
, 0, copy_complete
, m
);
883 mempool_free(m
, pool
->mapping_pool
);
884 DMERR_LIMIT("dm_kcopyd_zero() failed");
885 cell_error(pool
, cell
);
890 static int commit(struct pool
*pool
)
894 r
= dm_pool_commit_metadata(pool
->pmd
);
896 DMERR_LIMIT("%s: commit failed: error = %d",
897 dm_device_name(pool
->pool_md
), r
);
903 * A non-zero return indicates read_only or fail_io mode.
904 * Many callers don't care about the return value.
906 static int commit_or_fallback(struct pool
*pool
)
910 if (get_pool_mode(pool
) != PM_WRITE
)
915 set_pool_mode(pool
, PM_READ_ONLY
);
920 static int alloc_data_block(struct thin_c
*tc
, dm_block_t
*result
)
923 dm_block_t free_blocks
;
925 struct pool
*pool
= tc
->pool
;
928 * Once no_free_space is set we must not allow allocation to succeed.
929 * Otherwise it is difficult to explain, debug, test and support.
931 if (pool
->no_free_space
)
934 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
938 if (free_blocks
<= pool
->low_water_blocks
&& !pool
->low_water_triggered
) {
939 DMWARN("%s: reached low water mark for data device: sending event.",
940 dm_device_name(pool
->pool_md
));
941 spin_lock_irqsave(&pool
->lock
, flags
);
942 pool
->low_water_triggered
= 1;
943 spin_unlock_irqrestore(&pool
->lock
, flags
);
944 dm_table_event(pool
->ti
->table
);
949 * Try to commit to see if that will free up some
952 (void) commit_or_fallback(pool
);
954 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
959 * If we still have no space we set a flag to avoid
960 * doing all this checking and return -ENOSPC. This
961 * flag serves as a latch that disallows allocations from
962 * this pool until the admin takes action (e.g. resize or
966 DMWARN("%s: no free space available.",
967 dm_device_name(pool
->pool_md
));
968 spin_lock_irqsave(&pool
->lock
, flags
);
969 pool
->no_free_space
= 1;
970 spin_unlock_irqrestore(&pool
->lock
, flags
);
975 r
= dm_pool_alloc_data_block(pool
->pmd
, result
);
983 * If we have run out of space, queue bios until the device is
984 * resumed, presumably after having been reloaded with more space.
986 static void retry_on_resume(struct bio
*bio
)
988 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
989 struct thin_c
*tc
= h
->tc
;
990 struct pool
*pool
= tc
->pool
;
993 spin_lock_irqsave(&pool
->lock
, flags
);
994 bio_list_add(&pool
->retry_on_resume_list
, bio
);
995 spin_unlock_irqrestore(&pool
->lock
, flags
);
998 static void no_space(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
1001 struct bio_list bios
;
1003 bio_list_init(&bios
);
1004 cell_release(pool
, cell
, &bios
);
1006 while ((bio
= bio_list_pop(&bios
)))
1007 retry_on_resume(bio
);
1010 static void process_discard(struct thin_c
*tc
, struct bio
*bio
)
1013 unsigned long flags
;
1014 struct pool
*pool
= tc
->pool
;
1015 struct dm_bio_prison_cell
*cell
, *cell2
;
1016 struct dm_cell_key key
, key2
;
1017 dm_block_t block
= get_bio_block(tc
, bio
);
1018 struct dm_thin_lookup_result lookup_result
;
1019 struct dm_thin_new_mapping
*m
;
1021 build_virtual_key(tc
->td
, block
, &key
);
1022 if (bio_detain(tc
->pool
, &key
, bio
, &cell
))
1025 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1029 * Check nobody is fiddling with this pool block. This can
1030 * happen if someone's in the process of breaking sharing
1033 build_data_key(tc
->td
, lookup_result
.block
, &key2
);
1034 if (bio_detain(tc
->pool
, &key2
, bio
, &cell2
)) {
1035 cell_defer_no_holder(tc
, cell
);
1039 if (io_overlaps_block(pool
, bio
)) {
1041 * IO may still be going to the destination block. We must
1042 * quiesce before we can do the removal.
1044 m
= get_next_mapping(pool
);
1046 m
->pass_discard
= (!lookup_result
.shared
) && pool
->pf
.discard_passdown
;
1047 m
->virt_block
= block
;
1048 m
->data_block
= lookup_result
.block
;
1054 if (!dm_deferred_set_add_work(pool
->all_io_ds
, &m
->list
)) {
1055 spin_lock_irqsave(&pool
->lock
, flags
);
1056 list_add(&m
->list
, &pool
->prepared_discards
);
1057 spin_unlock_irqrestore(&pool
->lock
, flags
);
1061 inc_all_io_entry(pool
, bio
);
1062 cell_defer_no_holder(tc
, cell
);
1063 cell_defer_no_holder(tc
, cell2
);
1066 * The DM core makes sure that the discard doesn't span
1067 * a block boundary. So we submit the discard of a
1068 * partial block appropriately.
1070 if ((!lookup_result
.shared
) && pool
->pf
.discard_passdown
)
1071 remap_and_issue(tc
, bio
, lookup_result
.block
);
1079 * It isn't provisioned, just forget it.
1081 cell_defer_no_holder(tc
, cell
);
1086 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1088 cell_defer_no_holder(tc
, cell
);
1094 static void break_sharing(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1095 struct dm_cell_key
*key
,
1096 struct dm_thin_lookup_result
*lookup_result
,
1097 struct dm_bio_prison_cell
*cell
)
1100 dm_block_t data_block
;
1101 struct pool
*pool
= tc
->pool
;
1103 r
= alloc_data_block(tc
, &data_block
);
1106 schedule_internal_copy(tc
, block
, lookup_result
->block
,
1107 data_block
, cell
, bio
);
1111 no_space(pool
, cell
);
1115 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1117 set_pool_mode(pool
, PM_READ_ONLY
);
1118 cell_error(pool
, cell
);
1123 static void process_shared_bio(struct thin_c
*tc
, struct bio
*bio
,
1125 struct dm_thin_lookup_result
*lookup_result
)
1127 struct dm_bio_prison_cell
*cell
;
1128 struct pool
*pool
= tc
->pool
;
1129 struct dm_cell_key key
;
1132 * If cell is already occupied, then sharing is already in the process
1133 * of being broken so we have nothing further to do here.
1135 build_data_key(tc
->td
, lookup_result
->block
, &key
);
1136 if (bio_detain(pool
, &key
, bio
, &cell
))
1139 if (bio_data_dir(bio
) == WRITE
&& bio
->bi_iter
.bi_size
)
1140 break_sharing(tc
, bio
, block
, &key
, lookup_result
, cell
);
1142 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1144 h
->shared_read_entry
= dm_deferred_entry_inc(pool
->shared_read_ds
);
1145 inc_all_io_entry(pool
, bio
);
1146 cell_defer_no_holder(tc
, cell
);
1148 remap_and_issue(tc
, bio
, lookup_result
->block
);
1152 static void provision_block(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1153 struct dm_bio_prison_cell
*cell
)
1156 dm_block_t data_block
;
1157 struct pool
*pool
= tc
->pool
;
1160 * Remap empty bios (flushes) immediately, without provisioning.
1162 if (!bio
->bi_iter
.bi_size
) {
1163 inc_all_io_entry(pool
, bio
);
1164 cell_defer_no_holder(tc
, cell
);
1166 remap_and_issue(tc
, bio
, 0);
1171 * Fill read bios with zeroes and complete them immediately.
1173 if (bio_data_dir(bio
) == READ
) {
1175 cell_defer_no_holder(tc
, cell
);
1180 r
= alloc_data_block(tc
, &data_block
);
1184 schedule_external_copy(tc
, block
, data_block
, cell
, bio
);
1186 schedule_zero(tc
, block
, data_block
, cell
, bio
);
1190 no_space(pool
, cell
);
1194 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1196 set_pool_mode(pool
, PM_READ_ONLY
);
1197 cell_error(pool
, cell
);
1202 static void process_bio(struct thin_c
*tc
, struct bio
*bio
)
1205 struct pool
*pool
= tc
->pool
;
1206 dm_block_t block
= get_bio_block(tc
, bio
);
1207 struct dm_bio_prison_cell
*cell
;
1208 struct dm_cell_key key
;
1209 struct dm_thin_lookup_result lookup_result
;
1212 * If cell is already occupied, then the block is already
1213 * being provisioned so we have nothing further to do here.
1215 build_virtual_key(tc
->td
, block
, &key
);
1216 if (bio_detain(pool
, &key
, bio
, &cell
))
1219 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1222 if (lookup_result
.shared
) {
1223 process_shared_bio(tc
, bio
, block
, &lookup_result
);
1224 cell_defer_no_holder(tc
, cell
); /* FIXME: pass this cell into process_shared? */
1226 inc_all_io_entry(pool
, bio
);
1227 cell_defer_no_holder(tc
, cell
);
1229 remap_and_issue(tc
, bio
, lookup_result
.block
);
1234 if (bio_data_dir(bio
) == READ
&& tc
->origin_dev
) {
1235 inc_all_io_entry(pool
, bio
);
1236 cell_defer_no_holder(tc
, cell
);
1238 remap_to_origin_and_issue(tc
, bio
);
1240 provision_block(tc
, bio
, block
, cell
);
1244 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1246 cell_defer_no_holder(tc
, cell
);
1252 static void process_bio_read_only(struct thin_c
*tc
, struct bio
*bio
)
1255 int rw
= bio_data_dir(bio
);
1256 dm_block_t block
= get_bio_block(tc
, bio
);
1257 struct dm_thin_lookup_result lookup_result
;
1259 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1262 if (lookup_result
.shared
&& (rw
== WRITE
) && bio
->bi_iter
.bi_size
)
1265 inc_all_io_entry(tc
->pool
, bio
);
1266 remap_and_issue(tc
, bio
, lookup_result
.block
);
1276 if (tc
->origin_dev
) {
1277 inc_all_io_entry(tc
->pool
, bio
);
1278 remap_to_origin_and_issue(tc
, bio
);
1287 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1294 static void process_bio_fail(struct thin_c
*tc
, struct bio
*bio
)
1300 * FIXME: should we also commit due to size of transaction, measured in
1303 static int need_commit_due_to_time(struct pool
*pool
)
1305 return jiffies
< pool
->last_commit_jiffies
||
1306 jiffies
> pool
->last_commit_jiffies
+ COMMIT_PERIOD
;
1309 static void process_deferred_bios(struct pool
*pool
)
1311 unsigned long flags
;
1313 struct bio_list bios
;
1315 bio_list_init(&bios
);
1317 spin_lock_irqsave(&pool
->lock
, flags
);
1318 bio_list_merge(&bios
, &pool
->deferred_bios
);
1319 bio_list_init(&pool
->deferred_bios
);
1320 spin_unlock_irqrestore(&pool
->lock
, flags
);
1322 while ((bio
= bio_list_pop(&bios
))) {
1323 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1324 struct thin_c
*tc
= h
->tc
;
1327 * If we've got no free new_mapping structs, and processing
1328 * this bio might require one, we pause until there are some
1329 * prepared mappings to process.
1331 if (ensure_next_mapping(pool
)) {
1332 spin_lock_irqsave(&pool
->lock
, flags
);
1333 bio_list_merge(&pool
->deferred_bios
, &bios
);
1334 spin_unlock_irqrestore(&pool
->lock
, flags
);
1339 if (bio
->bi_rw
& REQ_DISCARD
)
1340 pool
->process_discard(tc
, bio
);
1342 pool
->process_bio(tc
, bio
);
1346 * If there are any deferred flush bios, we must commit
1347 * the metadata before issuing them.
1349 bio_list_init(&bios
);
1350 spin_lock_irqsave(&pool
->lock
, flags
);
1351 bio_list_merge(&bios
, &pool
->deferred_flush_bios
);
1352 bio_list_init(&pool
->deferred_flush_bios
);
1353 spin_unlock_irqrestore(&pool
->lock
, flags
);
1355 if (bio_list_empty(&bios
) && !need_commit_due_to_time(pool
))
1358 if (commit_or_fallback(pool
)) {
1359 while ((bio
= bio_list_pop(&bios
)))
1363 pool
->last_commit_jiffies
= jiffies
;
1365 while ((bio
= bio_list_pop(&bios
)))
1366 generic_make_request(bio
);
1369 static void do_worker(struct work_struct
*ws
)
1371 struct pool
*pool
= container_of(ws
, struct pool
, worker
);
1373 process_prepared(pool
, &pool
->prepared_mappings
, &pool
->process_prepared_mapping
);
1374 process_prepared(pool
, &pool
->prepared_discards
, &pool
->process_prepared_discard
);
1375 process_deferred_bios(pool
);
1379 * We want to commit periodically so that not too much
1380 * unwritten data builds up.
1382 static void do_waker(struct work_struct
*ws
)
1384 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
, waker
);
1386 queue_delayed_work(pool
->wq
, &pool
->waker
, COMMIT_PERIOD
);
1389 /*----------------------------------------------------------------*/
1391 static enum pool_mode
get_pool_mode(struct pool
*pool
)
1393 return pool
->pf
.mode
;
1396 static void set_pool_mode(struct pool
*pool
, enum pool_mode mode
)
1400 pool
->pf
.mode
= mode
;
1404 DMERR("%s: switching pool to failure mode",
1405 dm_device_name(pool
->pool_md
));
1406 pool
->process_bio
= process_bio_fail
;
1407 pool
->process_discard
= process_bio_fail
;
1408 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1409 pool
->process_prepared_discard
= process_prepared_discard_fail
;
1413 DMERR("%s: switching pool to read-only mode",
1414 dm_device_name(pool
->pool_md
));
1415 r
= dm_pool_abort_metadata(pool
->pmd
);
1417 DMERR("%s: aborting transaction failed",
1418 dm_device_name(pool
->pool_md
));
1419 set_pool_mode(pool
, PM_FAIL
);
1421 dm_pool_metadata_read_only(pool
->pmd
);
1422 pool
->process_bio
= process_bio_read_only
;
1423 pool
->process_discard
= process_discard
;
1424 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1425 pool
->process_prepared_discard
= process_prepared_discard_passdown
;
1430 pool
->process_bio
= process_bio
;
1431 pool
->process_discard
= process_discard
;
1432 pool
->process_prepared_mapping
= process_prepared_mapping
;
1433 pool
->process_prepared_discard
= process_prepared_discard
;
1438 /*----------------------------------------------------------------*/
1441 * Mapping functions.
1445 * Called only while mapping a thin bio to hand it over to the workqueue.
1447 static void thin_defer_bio(struct thin_c
*tc
, struct bio
*bio
)
1449 unsigned long flags
;
1450 struct pool
*pool
= tc
->pool
;
1452 spin_lock_irqsave(&pool
->lock
, flags
);
1453 bio_list_add(&pool
->deferred_bios
, bio
);
1454 spin_unlock_irqrestore(&pool
->lock
, flags
);
1459 static void thin_hook_bio(struct thin_c
*tc
, struct bio
*bio
)
1461 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1464 h
->shared_read_entry
= NULL
;
1465 h
->all_io_entry
= NULL
;
1466 h
->overwrite_mapping
= NULL
;
1470 * Non-blocking function called from the thin target's map function.
1472 static int thin_bio_map(struct dm_target
*ti
, struct bio
*bio
)
1475 struct thin_c
*tc
= ti
->private;
1476 dm_block_t block
= get_bio_block(tc
, bio
);
1477 struct dm_thin_device
*td
= tc
->td
;
1478 struct dm_thin_lookup_result result
;
1479 struct dm_bio_prison_cell cell1
, cell2
;
1480 struct dm_bio_prison_cell
*cell_result
;
1481 struct dm_cell_key key
;
1483 thin_hook_bio(tc
, bio
);
1485 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
1487 return DM_MAPIO_SUBMITTED
;
1490 if (bio
->bi_rw
& (REQ_DISCARD
| REQ_FLUSH
| REQ_FUA
)) {
1491 thin_defer_bio(tc
, bio
);
1492 return DM_MAPIO_SUBMITTED
;
1495 r
= dm_thin_find_block(td
, block
, 0, &result
);
1498 * Note that we defer readahead too.
1502 if (unlikely(result
.shared
)) {
1504 * We have a race condition here between the
1505 * result.shared value returned by the lookup and
1506 * snapshot creation, which may cause new
1509 * To avoid this always quiesce the origin before
1510 * taking the snap. You want to do this anyway to
1511 * ensure a consistent application view
1514 * More distant ancestors are irrelevant. The
1515 * shared flag will be set in their case.
1517 thin_defer_bio(tc
, bio
);
1518 return DM_MAPIO_SUBMITTED
;
1521 build_virtual_key(tc
->td
, block
, &key
);
1522 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell1
, &cell_result
))
1523 return DM_MAPIO_SUBMITTED
;
1525 build_data_key(tc
->td
, result
.block
, &key
);
1526 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell2
, &cell_result
)) {
1527 cell_defer_no_holder_no_free(tc
, &cell1
);
1528 return DM_MAPIO_SUBMITTED
;
1531 inc_all_io_entry(tc
->pool
, bio
);
1532 cell_defer_no_holder_no_free(tc
, &cell2
);
1533 cell_defer_no_holder_no_free(tc
, &cell1
);
1535 remap(tc
, bio
, result
.block
);
1536 return DM_MAPIO_REMAPPED
;
1539 if (get_pool_mode(tc
->pool
) == PM_READ_ONLY
) {
1541 * This block isn't provisioned, and we have no way
1542 * of doing so. Just error it.
1545 return DM_MAPIO_SUBMITTED
;
1551 * In future, the failed dm_thin_find_block above could
1552 * provide the hint to load the metadata into cache.
1554 thin_defer_bio(tc
, bio
);
1555 return DM_MAPIO_SUBMITTED
;
1559 * Must always call bio_io_error on failure.
1560 * dm_thin_find_block can fail with -EINVAL if the
1561 * pool is switched to fail-io mode.
1564 return DM_MAPIO_SUBMITTED
;
1568 static int pool_is_congested(struct dm_target_callbacks
*cb
, int bdi_bits
)
1571 unsigned long flags
;
1572 struct pool_c
*pt
= container_of(cb
, struct pool_c
, callbacks
);
1574 spin_lock_irqsave(&pt
->pool
->lock
, flags
);
1575 r
= !bio_list_empty(&pt
->pool
->retry_on_resume_list
);
1576 spin_unlock_irqrestore(&pt
->pool
->lock
, flags
);
1579 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1580 r
= bdi_congested(&q
->backing_dev_info
, bdi_bits
);
1586 static void __requeue_bios(struct pool
*pool
)
1588 bio_list_merge(&pool
->deferred_bios
, &pool
->retry_on_resume_list
);
1589 bio_list_init(&pool
->retry_on_resume_list
);
1592 /*----------------------------------------------------------------
1593 * Binding of control targets to a pool object
1594 *--------------------------------------------------------------*/
1595 static bool data_dev_supports_discard(struct pool_c
*pt
)
1597 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1599 return q
&& blk_queue_discard(q
);
1602 static bool is_factor(sector_t block_size
, uint32_t n
)
1604 return !sector_div(block_size
, n
);
1608 * If discard_passdown was enabled verify that the data device
1609 * supports discards. Disable discard_passdown if not.
1611 static void disable_passdown_if_not_supported(struct pool_c
*pt
)
1613 struct pool
*pool
= pt
->pool
;
1614 struct block_device
*data_bdev
= pt
->data_dev
->bdev
;
1615 struct queue_limits
*data_limits
= &bdev_get_queue(data_bdev
)->limits
;
1616 sector_t block_size
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
1617 const char *reason
= NULL
;
1618 char buf
[BDEVNAME_SIZE
];
1620 if (!pt
->adjusted_pf
.discard_passdown
)
1623 if (!data_dev_supports_discard(pt
))
1624 reason
= "discard unsupported";
1626 else if (data_limits
->max_discard_sectors
< pool
->sectors_per_block
)
1627 reason
= "max discard sectors smaller than a block";
1629 else if (data_limits
->discard_granularity
> block_size
)
1630 reason
= "discard granularity larger than a block";
1632 else if (!is_factor(block_size
, data_limits
->discard_granularity
))
1633 reason
= "discard granularity not a factor of block size";
1636 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev
, buf
), reason
);
1637 pt
->adjusted_pf
.discard_passdown
= false;
1641 static int bind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1643 struct pool_c
*pt
= ti
->private;
1646 * We want to make sure that degraded pools are never upgraded.
1648 enum pool_mode old_mode
= pool
->pf
.mode
;
1649 enum pool_mode new_mode
= pt
->adjusted_pf
.mode
;
1651 if (old_mode
> new_mode
)
1652 new_mode
= old_mode
;
1655 pool
->low_water_blocks
= pt
->low_water_blocks
;
1656 pool
->pf
= pt
->adjusted_pf
;
1658 set_pool_mode(pool
, new_mode
);
1663 static void unbind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1669 /*----------------------------------------------------------------
1671 *--------------------------------------------------------------*/
1672 /* Initialize pool features. */
1673 static void pool_features_init(struct pool_features
*pf
)
1675 pf
->mode
= PM_WRITE
;
1676 pf
->zero_new_blocks
= true;
1677 pf
->discard_enabled
= true;
1678 pf
->discard_passdown
= true;
1681 static void __pool_destroy(struct pool
*pool
)
1683 __pool_table_remove(pool
);
1685 if (dm_pool_metadata_close(pool
->pmd
) < 0)
1686 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1688 dm_bio_prison_destroy(pool
->prison
);
1689 dm_kcopyd_client_destroy(pool
->copier
);
1692 destroy_workqueue(pool
->wq
);
1694 if (pool
->next_mapping
)
1695 mempool_free(pool
->next_mapping
, pool
->mapping_pool
);
1696 mempool_destroy(pool
->mapping_pool
);
1697 dm_deferred_set_destroy(pool
->shared_read_ds
);
1698 dm_deferred_set_destroy(pool
->all_io_ds
);
1702 static struct kmem_cache
*_new_mapping_cache
;
1704 static struct pool
*pool_create(struct mapped_device
*pool_md
,
1705 struct block_device
*metadata_dev
,
1706 unsigned long block_size
,
1707 int read_only
, char **error
)
1712 struct dm_pool_metadata
*pmd
;
1713 bool format_device
= read_only
? false : true;
1715 pmd
= dm_pool_metadata_open(metadata_dev
, block_size
, format_device
);
1717 *error
= "Error creating metadata object";
1718 return (struct pool
*)pmd
;
1721 pool
= kmalloc(sizeof(*pool
), GFP_KERNEL
);
1723 *error
= "Error allocating memory for pool";
1724 err_p
= ERR_PTR(-ENOMEM
);
1729 pool
->sectors_per_block
= block_size
;
1730 if (block_size
& (block_size
- 1))
1731 pool
->sectors_per_block_shift
= -1;
1733 pool
->sectors_per_block_shift
= __ffs(block_size
);
1734 pool
->low_water_blocks
= 0;
1735 pool_features_init(&pool
->pf
);
1736 pool
->prison
= dm_bio_prison_create(PRISON_CELLS
);
1737 if (!pool
->prison
) {
1738 *error
= "Error creating pool's bio prison";
1739 err_p
= ERR_PTR(-ENOMEM
);
1743 pool
->copier
= dm_kcopyd_client_create(&dm_kcopyd_throttle
);
1744 if (IS_ERR(pool
->copier
)) {
1745 r
= PTR_ERR(pool
->copier
);
1746 *error
= "Error creating pool's kcopyd client";
1748 goto bad_kcopyd_client
;
1752 * Create singlethreaded workqueue that will service all devices
1753 * that use this metadata.
1755 pool
->wq
= alloc_ordered_workqueue("dm-" DM_MSG_PREFIX
, WQ_MEM_RECLAIM
);
1757 *error
= "Error creating pool's workqueue";
1758 err_p
= ERR_PTR(-ENOMEM
);
1762 INIT_WORK(&pool
->worker
, do_worker
);
1763 INIT_DELAYED_WORK(&pool
->waker
, do_waker
);
1764 spin_lock_init(&pool
->lock
);
1765 bio_list_init(&pool
->deferred_bios
);
1766 bio_list_init(&pool
->deferred_flush_bios
);
1767 INIT_LIST_HEAD(&pool
->prepared_mappings
);
1768 INIT_LIST_HEAD(&pool
->prepared_discards
);
1769 pool
->low_water_triggered
= 0;
1770 pool
->no_free_space
= 0;
1771 bio_list_init(&pool
->retry_on_resume_list
);
1773 pool
->shared_read_ds
= dm_deferred_set_create();
1774 if (!pool
->shared_read_ds
) {
1775 *error
= "Error creating pool's shared read deferred set";
1776 err_p
= ERR_PTR(-ENOMEM
);
1777 goto bad_shared_read_ds
;
1780 pool
->all_io_ds
= dm_deferred_set_create();
1781 if (!pool
->all_io_ds
) {
1782 *error
= "Error creating pool's all io deferred set";
1783 err_p
= ERR_PTR(-ENOMEM
);
1787 pool
->next_mapping
= NULL
;
1788 pool
->mapping_pool
= mempool_create_slab_pool(MAPPING_POOL_SIZE
,
1789 _new_mapping_cache
);
1790 if (!pool
->mapping_pool
) {
1791 *error
= "Error creating pool's mapping mempool";
1792 err_p
= ERR_PTR(-ENOMEM
);
1793 goto bad_mapping_pool
;
1796 pool
->ref_count
= 1;
1797 pool
->last_commit_jiffies
= jiffies
;
1798 pool
->pool_md
= pool_md
;
1799 pool
->md_dev
= metadata_dev
;
1800 __pool_table_insert(pool
);
1805 dm_deferred_set_destroy(pool
->all_io_ds
);
1807 dm_deferred_set_destroy(pool
->shared_read_ds
);
1809 destroy_workqueue(pool
->wq
);
1811 dm_kcopyd_client_destroy(pool
->copier
);
1813 dm_bio_prison_destroy(pool
->prison
);
1817 if (dm_pool_metadata_close(pmd
))
1818 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1823 static void __pool_inc(struct pool
*pool
)
1825 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
1829 static void __pool_dec(struct pool
*pool
)
1831 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
1832 BUG_ON(!pool
->ref_count
);
1833 if (!--pool
->ref_count
)
1834 __pool_destroy(pool
);
1837 static struct pool
*__pool_find(struct mapped_device
*pool_md
,
1838 struct block_device
*metadata_dev
,
1839 unsigned long block_size
, int read_only
,
1840 char **error
, int *created
)
1842 struct pool
*pool
= __pool_table_lookup_metadata_dev(metadata_dev
);
1845 if (pool
->pool_md
!= pool_md
) {
1846 *error
= "metadata device already in use by a pool";
1847 return ERR_PTR(-EBUSY
);
1852 pool
= __pool_table_lookup(pool_md
);
1854 if (pool
->md_dev
!= metadata_dev
) {
1855 *error
= "different pool cannot replace a pool";
1856 return ERR_PTR(-EINVAL
);
1861 pool
= pool_create(pool_md
, metadata_dev
, block_size
, read_only
, error
);
1869 /*----------------------------------------------------------------
1870 * Pool target methods
1871 *--------------------------------------------------------------*/
1872 static void pool_dtr(struct dm_target
*ti
)
1874 struct pool_c
*pt
= ti
->private;
1876 mutex_lock(&dm_thin_pool_table
.mutex
);
1878 unbind_control_target(pt
->pool
, ti
);
1879 __pool_dec(pt
->pool
);
1880 dm_put_device(ti
, pt
->metadata_dev
);
1881 dm_put_device(ti
, pt
->data_dev
);
1884 mutex_unlock(&dm_thin_pool_table
.mutex
);
1887 static int parse_pool_features(struct dm_arg_set
*as
, struct pool_features
*pf
,
1888 struct dm_target
*ti
)
1892 const char *arg_name
;
1894 static struct dm_arg _args
[] = {
1895 {0, 3, "Invalid number of pool feature arguments"},
1899 * No feature arguments supplied.
1904 r
= dm_read_arg_group(_args
, as
, &argc
, &ti
->error
);
1908 while (argc
&& !r
) {
1909 arg_name
= dm_shift_arg(as
);
1912 if (!strcasecmp(arg_name
, "skip_block_zeroing"))
1913 pf
->zero_new_blocks
= false;
1915 else if (!strcasecmp(arg_name
, "ignore_discard"))
1916 pf
->discard_enabled
= false;
1918 else if (!strcasecmp(arg_name
, "no_discard_passdown"))
1919 pf
->discard_passdown
= false;
1921 else if (!strcasecmp(arg_name
, "read_only"))
1922 pf
->mode
= PM_READ_ONLY
;
1925 ti
->error
= "Unrecognised pool feature requested";
1934 static void metadata_low_callback(void *context
)
1936 struct pool
*pool
= context
;
1938 DMWARN("%s: reached low water mark for metadata device: sending event.",
1939 dm_device_name(pool
->pool_md
));
1941 dm_table_event(pool
->ti
->table
);
1944 static sector_t
get_metadata_dev_size(struct block_device
*bdev
)
1946 sector_t metadata_dev_size
= i_size_read(bdev
->bd_inode
) >> SECTOR_SHIFT
;
1947 char buffer
[BDEVNAME_SIZE
];
1949 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS_WARNING
) {
1950 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
1951 bdevname(bdev
, buffer
), THIN_METADATA_MAX_SECTORS
);
1952 metadata_dev_size
= THIN_METADATA_MAX_SECTORS_WARNING
;
1955 return metadata_dev_size
;
1958 static dm_block_t
get_metadata_dev_size_in_blocks(struct block_device
*bdev
)
1960 sector_t metadata_dev_size
= get_metadata_dev_size(bdev
);
1962 sector_div(metadata_dev_size
, THIN_METADATA_BLOCK_SIZE
>> SECTOR_SHIFT
);
1964 return metadata_dev_size
;
1968 * When a metadata threshold is crossed a dm event is triggered, and
1969 * userland should respond by growing the metadata device. We could let
1970 * userland set the threshold, like we do with the data threshold, but I'm
1971 * not sure they know enough to do this well.
1973 static dm_block_t
calc_metadata_threshold(struct pool_c
*pt
)
1976 * 4M is ample for all ops with the possible exception of thin
1977 * device deletion which is harmless if it fails (just retry the
1978 * delete after you've grown the device).
1980 dm_block_t quarter
= get_metadata_dev_size_in_blocks(pt
->metadata_dev
->bdev
) / 4;
1981 return min((dm_block_t
)1024ULL /* 4M */, quarter
);
1985 * thin-pool <metadata dev> <data dev>
1986 * <data block size (sectors)>
1987 * <low water mark (blocks)>
1988 * [<#feature args> [<arg>]*]
1990 * Optional feature arguments are:
1991 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
1992 * ignore_discard: disable discard
1993 * no_discard_passdown: don't pass discards down to the data device
1995 static int pool_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
1997 int r
, pool_created
= 0;
2000 struct pool_features pf
;
2001 struct dm_arg_set as
;
2002 struct dm_dev
*data_dev
;
2003 unsigned long block_size
;
2004 dm_block_t low_water_blocks
;
2005 struct dm_dev
*metadata_dev
;
2006 fmode_t metadata_mode
;
2009 * FIXME Remove validation from scope of lock.
2011 mutex_lock(&dm_thin_pool_table
.mutex
);
2014 ti
->error
= "Invalid argument count";
2023 * Set default pool features.
2025 pool_features_init(&pf
);
2027 dm_consume_args(&as
, 4);
2028 r
= parse_pool_features(&as
, &pf
, ti
);
2032 metadata_mode
= FMODE_READ
| ((pf
.mode
== PM_READ_ONLY
) ? 0 : FMODE_WRITE
);
2033 r
= dm_get_device(ti
, argv
[0], metadata_mode
, &metadata_dev
);
2035 ti
->error
= "Error opening metadata block device";
2040 * Run for the side-effect of possibly issuing a warning if the
2041 * device is too big.
2043 (void) get_metadata_dev_size(metadata_dev
->bdev
);
2045 r
= dm_get_device(ti
, argv
[1], FMODE_READ
| FMODE_WRITE
, &data_dev
);
2047 ti
->error
= "Error getting data device";
2051 if (kstrtoul(argv
[2], 10, &block_size
) || !block_size
||
2052 block_size
< DATA_DEV_BLOCK_SIZE_MIN_SECTORS
||
2053 block_size
> DATA_DEV_BLOCK_SIZE_MAX_SECTORS
||
2054 block_size
& (DATA_DEV_BLOCK_SIZE_MIN_SECTORS
- 1)) {
2055 ti
->error
= "Invalid block size";
2060 if (kstrtoull(argv
[3], 10, (unsigned long long *)&low_water_blocks
)) {
2061 ti
->error
= "Invalid low water mark";
2066 pt
= kzalloc(sizeof(*pt
), GFP_KERNEL
);
2072 pool
= __pool_find(dm_table_get_md(ti
->table
), metadata_dev
->bdev
,
2073 block_size
, pf
.mode
== PM_READ_ONLY
, &ti
->error
, &pool_created
);
2080 * 'pool_created' reflects whether this is the first table load.
2081 * Top level discard support is not allowed to be changed after
2082 * initial load. This would require a pool reload to trigger thin
2085 if (!pool_created
&& pf
.discard_enabled
!= pool
->pf
.discard_enabled
) {
2086 ti
->error
= "Discard support cannot be disabled once enabled";
2088 goto out_flags_changed
;
2093 pt
->metadata_dev
= metadata_dev
;
2094 pt
->data_dev
= data_dev
;
2095 pt
->low_water_blocks
= low_water_blocks
;
2096 pt
->adjusted_pf
= pt
->requested_pf
= pf
;
2097 ti
->num_flush_bios
= 1;
2100 * Only need to enable discards if the pool should pass
2101 * them down to the data device. The thin device's discard
2102 * processing will cause mappings to be removed from the btree.
2104 ti
->discard_zeroes_data_unsupported
= true;
2105 if (pf
.discard_enabled
&& pf
.discard_passdown
) {
2106 ti
->num_discard_bios
= 1;
2109 * Setting 'discards_supported' circumvents the normal
2110 * stacking of discard limits (this keeps the pool and
2111 * thin devices' discard limits consistent).
2113 ti
->discards_supported
= true;
2117 r
= dm_pool_register_metadata_threshold(pt
->pool
->pmd
,
2118 calc_metadata_threshold(pt
),
2119 metadata_low_callback
,
2124 pt
->callbacks
.congested_fn
= pool_is_congested
;
2125 dm_table_add_target_callbacks(ti
->table
, &pt
->callbacks
);
2127 mutex_unlock(&dm_thin_pool_table
.mutex
);
2136 dm_put_device(ti
, data_dev
);
2138 dm_put_device(ti
, metadata_dev
);
2140 mutex_unlock(&dm_thin_pool_table
.mutex
);
2145 static int pool_map(struct dm_target
*ti
, struct bio
*bio
)
2148 struct pool_c
*pt
= ti
->private;
2149 struct pool
*pool
= pt
->pool
;
2150 unsigned long flags
;
2153 * As this is a singleton target, ti->begin is always zero.
2155 spin_lock_irqsave(&pool
->lock
, flags
);
2156 bio
->bi_bdev
= pt
->data_dev
->bdev
;
2157 r
= DM_MAPIO_REMAPPED
;
2158 spin_unlock_irqrestore(&pool
->lock
, flags
);
2163 static int maybe_resize_data_dev(struct dm_target
*ti
, bool *need_commit
)
2166 struct pool_c
*pt
= ti
->private;
2167 struct pool
*pool
= pt
->pool
;
2168 sector_t data_size
= ti
->len
;
2169 dm_block_t sb_data_size
;
2171 *need_commit
= false;
2173 (void) sector_div(data_size
, pool
->sectors_per_block
);
2175 r
= dm_pool_get_data_dev_size(pool
->pmd
, &sb_data_size
);
2177 DMERR("%s: failed to retrieve data device size",
2178 dm_device_name(pool
->pool_md
));
2182 if (data_size
< sb_data_size
) {
2183 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
2184 dm_device_name(pool
->pool_md
),
2185 (unsigned long long)data_size
, sb_data_size
);
2188 } else if (data_size
> sb_data_size
) {
2189 r
= dm_pool_resize_data_dev(pool
->pmd
, data_size
);
2191 DMERR("%s: failed to resize data device",
2192 dm_device_name(pool
->pool_md
));
2193 set_pool_mode(pool
, PM_READ_ONLY
);
2197 *need_commit
= true;
2203 static int maybe_resize_metadata_dev(struct dm_target
*ti
, bool *need_commit
)
2206 struct pool_c
*pt
= ti
->private;
2207 struct pool
*pool
= pt
->pool
;
2208 dm_block_t metadata_dev_size
, sb_metadata_dev_size
;
2210 *need_commit
= false;
2212 metadata_dev_size
= get_metadata_dev_size_in_blocks(pool
->md_dev
);
2214 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &sb_metadata_dev_size
);
2216 DMERR("%s: failed to retrieve metadata device size",
2217 dm_device_name(pool
->pool_md
));
2221 if (metadata_dev_size
< sb_metadata_dev_size
) {
2222 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
2223 dm_device_name(pool
->pool_md
),
2224 metadata_dev_size
, sb_metadata_dev_size
);
2227 } else if (metadata_dev_size
> sb_metadata_dev_size
) {
2228 r
= dm_pool_resize_metadata_dev(pool
->pmd
, metadata_dev_size
);
2230 DMERR("%s: failed to resize metadata device",
2231 dm_device_name(pool
->pool_md
));
2235 *need_commit
= true;
2242 * Retrieves the number of blocks of the data device from
2243 * the superblock and compares it to the actual device size,
2244 * thus resizing the data device in case it has grown.
2246 * This both copes with opening preallocated data devices in the ctr
2247 * being followed by a resume
2249 * calling the resume method individually after userspace has
2250 * grown the data device in reaction to a table event.
2252 static int pool_preresume(struct dm_target
*ti
)
2255 bool need_commit1
, need_commit2
;
2256 struct pool_c
*pt
= ti
->private;
2257 struct pool
*pool
= pt
->pool
;
2260 * Take control of the pool object.
2262 r
= bind_control_target(pool
, ti
);
2266 r
= maybe_resize_data_dev(ti
, &need_commit1
);
2270 r
= maybe_resize_metadata_dev(ti
, &need_commit2
);
2274 if (need_commit1
|| need_commit2
)
2275 (void) commit_or_fallback(pool
);
2280 static void pool_resume(struct dm_target
*ti
)
2282 struct pool_c
*pt
= ti
->private;
2283 struct pool
*pool
= pt
->pool
;
2284 unsigned long flags
;
2286 spin_lock_irqsave(&pool
->lock
, flags
);
2287 pool
->low_water_triggered
= 0;
2288 pool
->no_free_space
= 0;
2289 __requeue_bios(pool
);
2290 spin_unlock_irqrestore(&pool
->lock
, flags
);
2292 do_waker(&pool
->waker
.work
);
2295 static void pool_postsuspend(struct dm_target
*ti
)
2297 struct pool_c
*pt
= ti
->private;
2298 struct pool
*pool
= pt
->pool
;
2300 cancel_delayed_work(&pool
->waker
);
2301 flush_workqueue(pool
->wq
);
2302 (void) commit_or_fallback(pool
);
2305 static int check_arg_count(unsigned argc
, unsigned args_required
)
2307 if (argc
!= args_required
) {
2308 DMWARN("Message received with %u arguments instead of %u.",
2309 argc
, args_required
);
2316 static int read_dev_id(char *arg
, dm_thin_id
*dev_id
, int warning
)
2318 if (!kstrtoull(arg
, 10, (unsigned long long *)dev_id
) &&
2319 *dev_id
<= MAX_DEV_ID
)
2323 DMWARN("Message received with invalid device id: %s", arg
);
2328 static int process_create_thin_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2333 r
= check_arg_count(argc
, 2);
2337 r
= read_dev_id(argv
[1], &dev_id
, 1);
2341 r
= dm_pool_create_thin(pool
->pmd
, dev_id
);
2343 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2351 static int process_create_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2354 dm_thin_id origin_dev_id
;
2357 r
= check_arg_count(argc
, 3);
2361 r
= read_dev_id(argv
[1], &dev_id
, 1);
2365 r
= read_dev_id(argv
[2], &origin_dev_id
, 1);
2369 r
= dm_pool_create_snap(pool
->pmd
, dev_id
, origin_dev_id
);
2371 DMWARN("Creation of new snapshot %s of device %s failed.",
2379 static int process_delete_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2384 r
= check_arg_count(argc
, 2);
2388 r
= read_dev_id(argv
[1], &dev_id
, 1);
2392 r
= dm_pool_delete_thin_device(pool
->pmd
, dev_id
);
2394 DMWARN("Deletion of thin device %s failed.", argv
[1]);
2399 static int process_set_transaction_id_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2401 dm_thin_id old_id
, new_id
;
2404 r
= check_arg_count(argc
, 3);
2408 if (kstrtoull(argv
[1], 10, (unsigned long long *)&old_id
)) {
2409 DMWARN("set_transaction_id message: Unrecognised id %s.", argv
[1]);
2413 if (kstrtoull(argv
[2], 10, (unsigned long long *)&new_id
)) {
2414 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv
[2]);
2418 r
= dm_pool_set_metadata_transaction_id(pool
->pmd
, old_id
, new_id
);
2420 DMWARN("Failed to change transaction id from %s to %s.",
2428 static int process_reserve_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2432 r
= check_arg_count(argc
, 1);
2436 (void) commit_or_fallback(pool
);
2438 r
= dm_pool_reserve_metadata_snap(pool
->pmd
);
2440 DMWARN("reserve_metadata_snap message failed.");
2445 static int process_release_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2449 r
= check_arg_count(argc
, 1);
2453 r
= dm_pool_release_metadata_snap(pool
->pmd
);
2455 DMWARN("release_metadata_snap message failed.");
2461 * Messages supported:
2462 * create_thin <dev_id>
2463 * create_snap <dev_id> <origin_id>
2465 * trim <dev_id> <new_size_in_sectors>
2466 * set_transaction_id <current_trans_id> <new_trans_id>
2467 * reserve_metadata_snap
2468 * release_metadata_snap
2470 static int pool_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
2473 struct pool_c
*pt
= ti
->private;
2474 struct pool
*pool
= pt
->pool
;
2476 if (!strcasecmp(argv
[0], "create_thin"))
2477 r
= process_create_thin_mesg(argc
, argv
, pool
);
2479 else if (!strcasecmp(argv
[0], "create_snap"))
2480 r
= process_create_snap_mesg(argc
, argv
, pool
);
2482 else if (!strcasecmp(argv
[0], "delete"))
2483 r
= process_delete_mesg(argc
, argv
, pool
);
2485 else if (!strcasecmp(argv
[0], "set_transaction_id"))
2486 r
= process_set_transaction_id_mesg(argc
, argv
, pool
);
2488 else if (!strcasecmp(argv
[0], "reserve_metadata_snap"))
2489 r
= process_reserve_metadata_snap_mesg(argc
, argv
, pool
);
2491 else if (!strcasecmp(argv
[0], "release_metadata_snap"))
2492 r
= process_release_metadata_snap_mesg(argc
, argv
, pool
);
2495 DMWARN("Unrecognised thin pool target message received: %s", argv
[0]);
2498 (void) commit_or_fallback(pool
);
2503 static void emit_flags(struct pool_features
*pf
, char *result
,
2504 unsigned sz
, unsigned maxlen
)
2506 unsigned count
= !pf
->zero_new_blocks
+ !pf
->discard_enabled
+
2507 !pf
->discard_passdown
+ (pf
->mode
== PM_READ_ONLY
);
2508 DMEMIT("%u ", count
);
2510 if (!pf
->zero_new_blocks
)
2511 DMEMIT("skip_block_zeroing ");
2513 if (!pf
->discard_enabled
)
2514 DMEMIT("ignore_discard ");
2516 if (!pf
->discard_passdown
)
2517 DMEMIT("no_discard_passdown ");
2519 if (pf
->mode
== PM_READ_ONLY
)
2520 DMEMIT("read_only ");
2525 * <transaction id> <used metadata sectors>/<total metadata sectors>
2526 * <used data sectors>/<total data sectors> <held metadata root>
2528 static void pool_status(struct dm_target
*ti
, status_type_t type
,
2529 unsigned status_flags
, char *result
, unsigned maxlen
)
2533 uint64_t transaction_id
;
2534 dm_block_t nr_free_blocks_data
;
2535 dm_block_t nr_free_blocks_metadata
;
2536 dm_block_t nr_blocks_data
;
2537 dm_block_t nr_blocks_metadata
;
2538 dm_block_t held_root
;
2539 char buf
[BDEVNAME_SIZE
];
2540 char buf2
[BDEVNAME_SIZE
];
2541 struct pool_c
*pt
= ti
->private;
2542 struct pool
*pool
= pt
->pool
;
2545 case STATUSTYPE_INFO
:
2546 if (get_pool_mode(pool
) == PM_FAIL
) {
2551 /* Commit to ensure statistics aren't out-of-date */
2552 if (!(status_flags
& DM_STATUS_NOFLUSH_FLAG
) && !dm_suspended(ti
))
2553 (void) commit_or_fallback(pool
);
2555 r
= dm_pool_get_metadata_transaction_id(pool
->pmd
, &transaction_id
);
2557 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
2558 dm_device_name(pool
->pool_md
), r
);
2562 r
= dm_pool_get_free_metadata_block_count(pool
->pmd
, &nr_free_blocks_metadata
);
2564 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
2565 dm_device_name(pool
->pool_md
), r
);
2569 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &nr_blocks_metadata
);
2571 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
2572 dm_device_name(pool
->pool_md
), r
);
2576 r
= dm_pool_get_free_block_count(pool
->pmd
, &nr_free_blocks_data
);
2578 DMERR("%s: dm_pool_get_free_block_count returned %d",
2579 dm_device_name(pool
->pool_md
), r
);
2583 r
= dm_pool_get_data_dev_size(pool
->pmd
, &nr_blocks_data
);
2585 DMERR("%s: dm_pool_get_data_dev_size returned %d",
2586 dm_device_name(pool
->pool_md
), r
);
2590 r
= dm_pool_get_metadata_snap(pool
->pmd
, &held_root
);
2592 DMERR("%s: dm_pool_get_metadata_snap returned %d",
2593 dm_device_name(pool
->pool_md
), r
);
2597 DMEMIT("%llu %llu/%llu %llu/%llu ",
2598 (unsigned long long)transaction_id
,
2599 (unsigned long long)(nr_blocks_metadata
- nr_free_blocks_metadata
),
2600 (unsigned long long)nr_blocks_metadata
,
2601 (unsigned long long)(nr_blocks_data
- nr_free_blocks_data
),
2602 (unsigned long long)nr_blocks_data
);
2605 DMEMIT("%llu ", held_root
);
2609 if (pool
->pf
.mode
== PM_READ_ONLY
)
2614 if (!pool
->pf
.discard_enabled
)
2615 DMEMIT("ignore_discard");
2616 else if (pool
->pf
.discard_passdown
)
2617 DMEMIT("discard_passdown");
2619 DMEMIT("no_discard_passdown");
2623 case STATUSTYPE_TABLE
:
2624 DMEMIT("%s %s %lu %llu ",
2625 format_dev_t(buf
, pt
->metadata_dev
->bdev
->bd_dev
),
2626 format_dev_t(buf2
, pt
->data_dev
->bdev
->bd_dev
),
2627 (unsigned long)pool
->sectors_per_block
,
2628 (unsigned long long)pt
->low_water_blocks
);
2629 emit_flags(&pt
->requested_pf
, result
, sz
, maxlen
);
2638 static int pool_iterate_devices(struct dm_target
*ti
,
2639 iterate_devices_callout_fn fn
, void *data
)
2641 struct pool_c
*pt
= ti
->private;
2643 return fn(ti
, pt
->data_dev
, 0, ti
->len
, data
);
2646 static int pool_merge(struct dm_target
*ti
, struct bvec_merge_data
*bvm
,
2647 struct bio_vec
*biovec
, int max_size
)
2649 struct pool_c
*pt
= ti
->private;
2650 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
2652 if (!q
->merge_bvec_fn
)
2655 bvm
->bi_bdev
= pt
->data_dev
->bdev
;
2657 return min(max_size
, q
->merge_bvec_fn(q
, bvm
, biovec
));
2660 static void set_discard_limits(struct pool_c
*pt
, struct queue_limits
*limits
)
2662 struct pool
*pool
= pt
->pool
;
2663 struct queue_limits
*data_limits
;
2665 limits
->max_discard_sectors
= pool
->sectors_per_block
;
2668 * discard_granularity is just a hint, and not enforced.
2670 if (pt
->adjusted_pf
.discard_passdown
) {
2671 data_limits
= &bdev_get_queue(pt
->data_dev
->bdev
)->limits
;
2672 limits
->discard_granularity
= data_limits
->discard_granularity
;
2674 limits
->discard_granularity
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
2677 static void pool_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
2679 struct pool_c
*pt
= ti
->private;
2680 struct pool
*pool
= pt
->pool
;
2681 uint64_t io_opt_sectors
= limits
->io_opt
>> SECTOR_SHIFT
;
2684 * If the system-determined stacked limits are compatible with the
2685 * pool's blocksize (io_opt is a factor) do not override them.
2687 if (io_opt_sectors
< pool
->sectors_per_block
||
2688 do_div(io_opt_sectors
, pool
->sectors_per_block
)) {
2689 blk_limits_io_min(limits
, 0);
2690 blk_limits_io_opt(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
2694 * pt->adjusted_pf is a staging area for the actual features to use.
2695 * They get transferred to the live pool in bind_control_target()
2696 * called from pool_preresume().
2698 if (!pt
->adjusted_pf
.discard_enabled
) {
2700 * Must explicitly disallow stacking discard limits otherwise the
2701 * block layer will stack them if pool's data device has support.
2702 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
2703 * user to see that, so make sure to set all discard limits to 0.
2705 limits
->discard_granularity
= 0;
2709 disable_passdown_if_not_supported(pt
);
2711 set_discard_limits(pt
, limits
);
2714 static struct target_type pool_target
= {
2715 .name
= "thin-pool",
2716 .features
= DM_TARGET_SINGLETON
| DM_TARGET_ALWAYS_WRITEABLE
|
2717 DM_TARGET_IMMUTABLE
,
2718 .version
= {1, 9, 0},
2719 .module
= THIS_MODULE
,
2723 .postsuspend
= pool_postsuspend
,
2724 .preresume
= pool_preresume
,
2725 .resume
= pool_resume
,
2726 .message
= pool_message
,
2727 .status
= pool_status
,
2728 .merge
= pool_merge
,
2729 .iterate_devices
= pool_iterate_devices
,
2730 .io_hints
= pool_io_hints
,
2733 /*----------------------------------------------------------------
2734 * Thin target methods
2735 *--------------------------------------------------------------*/
2736 static void thin_dtr(struct dm_target
*ti
)
2738 struct thin_c
*tc
= ti
->private;
2740 mutex_lock(&dm_thin_pool_table
.mutex
);
2742 __pool_dec(tc
->pool
);
2743 dm_pool_close_thin_device(tc
->td
);
2744 dm_put_device(ti
, tc
->pool_dev
);
2746 dm_put_device(ti
, tc
->origin_dev
);
2749 mutex_unlock(&dm_thin_pool_table
.mutex
);
2753 * Thin target parameters:
2755 * <pool_dev> <dev_id> [origin_dev]
2757 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2758 * dev_id: the internal device identifier
2759 * origin_dev: a device external to the pool that should act as the origin
2761 * If the pool device has discards disabled, they get disabled for the thin
2764 static int thin_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2768 struct dm_dev
*pool_dev
, *origin_dev
;
2769 struct mapped_device
*pool_md
;
2771 mutex_lock(&dm_thin_pool_table
.mutex
);
2773 if (argc
!= 2 && argc
!= 3) {
2774 ti
->error
= "Invalid argument count";
2779 tc
= ti
->private = kzalloc(sizeof(*tc
), GFP_KERNEL
);
2781 ti
->error
= "Out of memory";
2787 r
= dm_get_device(ti
, argv
[2], FMODE_READ
, &origin_dev
);
2789 ti
->error
= "Error opening origin device";
2790 goto bad_origin_dev
;
2792 tc
->origin_dev
= origin_dev
;
2795 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &pool_dev
);
2797 ti
->error
= "Error opening pool device";
2800 tc
->pool_dev
= pool_dev
;
2802 if (read_dev_id(argv
[1], (unsigned long long *)&tc
->dev_id
, 0)) {
2803 ti
->error
= "Invalid device id";
2808 pool_md
= dm_get_md(tc
->pool_dev
->bdev
->bd_dev
);
2810 ti
->error
= "Couldn't get pool mapped device";
2815 tc
->pool
= __pool_table_lookup(pool_md
);
2817 ti
->error
= "Couldn't find pool object";
2819 goto bad_pool_lookup
;
2821 __pool_inc(tc
->pool
);
2823 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
2824 ti
->error
= "Couldn't open thin device, Pool is in fail mode";
2828 r
= dm_pool_open_thin_device(tc
->pool
->pmd
, tc
->dev_id
, &tc
->td
);
2830 ti
->error
= "Couldn't open thin internal device";
2834 r
= dm_set_target_max_io_len(ti
, tc
->pool
->sectors_per_block
);
2838 ti
->num_flush_bios
= 1;
2839 ti
->flush_supported
= true;
2840 ti
->per_bio_data_size
= sizeof(struct dm_thin_endio_hook
);
2842 /* In case the pool supports discards, pass them on. */
2843 ti
->discard_zeroes_data_unsupported
= true;
2844 if (tc
->pool
->pf
.discard_enabled
) {
2845 ti
->discards_supported
= true;
2846 ti
->num_discard_bios
= 1;
2847 /* Discard bios must be split on a block boundary */
2848 ti
->split_discard_bios
= true;
2853 mutex_unlock(&dm_thin_pool_table
.mutex
);
2858 __pool_dec(tc
->pool
);
2862 dm_put_device(ti
, tc
->pool_dev
);
2865 dm_put_device(ti
, tc
->origin_dev
);
2869 mutex_unlock(&dm_thin_pool_table
.mutex
);
2874 static int thin_map(struct dm_target
*ti
, struct bio
*bio
)
2876 bio
->bi_iter
.bi_sector
= dm_target_offset(ti
, bio
->bi_iter
.bi_sector
);
2878 return thin_bio_map(ti
, bio
);
2881 static int thin_endio(struct dm_target
*ti
, struct bio
*bio
, int err
)
2883 unsigned long flags
;
2884 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
2885 struct list_head work
;
2886 struct dm_thin_new_mapping
*m
, *tmp
;
2887 struct pool
*pool
= h
->tc
->pool
;
2889 if (h
->shared_read_entry
) {
2890 INIT_LIST_HEAD(&work
);
2891 dm_deferred_entry_dec(h
->shared_read_entry
, &work
);
2893 spin_lock_irqsave(&pool
->lock
, flags
);
2894 list_for_each_entry_safe(m
, tmp
, &work
, list
) {
2897 __maybe_add_mapping(m
);
2899 spin_unlock_irqrestore(&pool
->lock
, flags
);
2902 if (h
->all_io_entry
) {
2903 INIT_LIST_HEAD(&work
);
2904 dm_deferred_entry_dec(h
->all_io_entry
, &work
);
2905 if (!list_empty(&work
)) {
2906 spin_lock_irqsave(&pool
->lock
, flags
);
2907 list_for_each_entry_safe(m
, tmp
, &work
, list
)
2908 list_add(&m
->list
, &pool
->prepared_discards
);
2909 spin_unlock_irqrestore(&pool
->lock
, flags
);
2917 static void thin_postsuspend(struct dm_target
*ti
)
2919 if (dm_noflush_suspending(ti
))
2920 requeue_io((struct thin_c
*)ti
->private);
2924 * <nr mapped sectors> <highest mapped sector>
2926 static void thin_status(struct dm_target
*ti
, status_type_t type
,
2927 unsigned status_flags
, char *result
, unsigned maxlen
)
2931 dm_block_t mapped
, highest
;
2932 char buf
[BDEVNAME_SIZE
];
2933 struct thin_c
*tc
= ti
->private;
2935 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
2944 case STATUSTYPE_INFO
:
2945 r
= dm_thin_get_mapped_count(tc
->td
, &mapped
);
2947 DMERR("dm_thin_get_mapped_count returned %d", r
);
2951 r
= dm_thin_get_highest_mapped_block(tc
->td
, &highest
);
2953 DMERR("dm_thin_get_highest_mapped_block returned %d", r
);
2957 DMEMIT("%llu ", mapped
* tc
->pool
->sectors_per_block
);
2959 DMEMIT("%llu", ((highest
+ 1) *
2960 tc
->pool
->sectors_per_block
) - 1);
2965 case STATUSTYPE_TABLE
:
2967 format_dev_t(buf
, tc
->pool_dev
->bdev
->bd_dev
),
2968 (unsigned long) tc
->dev_id
);
2970 DMEMIT(" %s", format_dev_t(buf
, tc
->origin_dev
->bdev
->bd_dev
));
2981 static int thin_iterate_devices(struct dm_target
*ti
,
2982 iterate_devices_callout_fn fn
, void *data
)
2985 struct thin_c
*tc
= ti
->private;
2986 struct pool
*pool
= tc
->pool
;
2989 * We can't call dm_pool_get_data_dev_size() since that blocks. So
2990 * we follow a more convoluted path through to the pool's target.
2993 return 0; /* nothing is bound */
2995 blocks
= pool
->ti
->len
;
2996 (void) sector_div(blocks
, pool
->sectors_per_block
);
2998 return fn(ti
, tc
->pool_dev
, 0, pool
->sectors_per_block
* blocks
, data
);
3003 static struct target_type thin_target
= {
3005 .version
= {1, 9, 0},
3006 .module
= THIS_MODULE
,
3010 .end_io
= thin_endio
,
3011 .postsuspend
= thin_postsuspend
,
3012 .status
= thin_status
,
3013 .iterate_devices
= thin_iterate_devices
,
3016 /*----------------------------------------------------------------*/
3018 static int __init
dm_thin_init(void)
3024 r
= dm_register_target(&thin_target
);
3028 r
= dm_register_target(&pool_target
);
3030 goto bad_pool_target
;
3034 _new_mapping_cache
= KMEM_CACHE(dm_thin_new_mapping
, 0);
3035 if (!_new_mapping_cache
)
3036 goto bad_new_mapping_cache
;
3040 bad_new_mapping_cache
:
3041 dm_unregister_target(&pool_target
);
3043 dm_unregister_target(&thin_target
);
3048 static void dm_thin_exit(void)
3050 dm_unregister_target(&thin_target
);
3051 dm_unregister_target(&pool_target
);
3053 kmem_cache_destroy(_new_mapping_cache
);
3056 module_init(dm_thin_init
);
3057 module_exit(dm_thin_exit
);
3059 MODULE_DESCRIPTION(DM_NAME
" thin provisioning target");
3060 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3061 MODULE_LICENSE("GPL");