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 4 modes. Ordered in degraded order for comparisons.
136 PM_WRITE
, /* metadata may be changed */
137 PM_OUT_OF_DATA_SPACE
, /* metadata may be changed, though data may not be allocated */
138 PM_READ_ONLY
, /* metadata may not be changed */
139 PM_FAIL
, /* all I/O fails */
142 struct pool_features
{
145 bool zero_new_blocks
:1;
146 bool discard_enabled
:1;
147 bool discard_passdown
:1;
148 bool error_if_no_space
:1;
152 typedef void (*process_bio_fn
)(struct thin_c
*tc
, struct bio
*bio
);
153 typedef void (*process_mapping_fn
)(struct dm_thin_new_mapping
*m
);
156 struct list_head list
;
157 struct dm_target
*ti
; /* Only set if a pool target is bound */
159 struct mapped_device
*pool_md
;
160 struct block_device
*md_dev
;
161 struct dm_pool_metadata
*pmd
;
163 dm_block_t low_water_blocks
;
164 uint32_t sectors_per_block
;
165 int sectors_per_block_shift
;
167 struct pool_features pf
;
168 bool low_water_triggered
:1; /* A dm event has been sent */
170 struct dm_bio_prison
*prison
;
171 struct dm_kcopyd_client
*copier
;
173 struct workqueue_struct
*wq
;
174 struct work_struct worker
;
175 struct delayed_work waker
;
177 unsigned long last_commit_jiffies
;
181 struct bio_list deferred_bios
;
182 struct bio_list deferred_flush_bios
;
183 struct list_head prepared_mappings
;
184 struct list_head prepared_discards
;
186 struct bio_list retry_on_resume_list
;
188 struct dm_deferred_set
*shared_read_ds
;
189 struct dm_deferred_set
*all_io_ds
;
191 struct dm_thin_new_mapping
*next_mapping
;
192 mempool_t
*mapping_pool
;
194 process_bio_fn process_bio
;
195 process_bio_fn process_discard
;
197 process_mapping_fn process_prepared_mapping
;
198 process_mapping_fn process_prepared_discard
;
201 static enum pool_mode
get_pool_mode(struct pool
*pool
);
202 static void metadata_operation_failed(struct pool
*pool
, const char *op
, int r
);
205 * Target context for a pool.
208 struct dm_target
*ti
;
210 struct dm_dev
*data_dev
;
211 struct dm_dev
*metadata_dev
;
212 struct dm_target_callbacks callbacks
;
214 dm_block_t low_water_blocks
;
215 struct pool_features requested_pf
; /* Features requested during table load */
216 struct pool_features adjusted_pf
; /* Features used after adjusting for constituent devices */
220 * Target context for a thin.
223 struct dm_dev
*pool_dev
;
224 struct dm_dev
*origin_dev
;
228 struct dm_thin_device
*td
;
232 /*----------------------------------------------------------------*/
235 * wake_worker() is used when new work is queued and when pool_resume is
236 * ready to continue deferred IO processing.
238 static void wake_worker(struct pool
*pool
)
240 queue_work(pool
->wq
, &pool
->worker
);
243 /*----------------------------------------------------------------*/
245 static int bio_detain(struct pool
*pool
, struct dm_cell_key
*key
, struct bio
*bio
,
246 struct dm_bio_prison_cell
**cell_result
)
249 struct dm_bio_prison_cell
*cell_prealloc
;
252 * Allocate a cell from the prison's mempool.
253 * This might block but it can't fail.
255 cell_prealloc
= dm_bio_prison_alloc_cell(pool
->prison
, GFP_NOIO
);
257 r
= dm_bio_detain(pool
->prison
, key
, bio
, cell_prealloc
, cell_result
);
260 * We reused an old cell; we can get rid of
263 dm_bio_prison_free_cell(pool
->prison
, cell_prealloc
);
268 static void cell_release(struct pool
*pool
,
269 struct dm_bio_prison_cell
*cell
,
270 struct bio_list
*bios
)
272 dm_cell_release(pool
->prison
, cell
, bios
);
273 dm_bio_prison_free_cell(pool
->prison
, cell
);
276 static void cell_release_no_holder(struct pool
*pool
,
277 struct dm_bio_prison_cell
*cell
,
278 struct bio_list
*bios
)
280 dm_cell_release_no_holder(pool
->prison
, cell
, bios
);
281 dm_bio_prison_free_cell(pool
->prison
, cell
);
284 static void cell_defer_no_holder_no_free(struct thin_c
*tc
,
285 struct dm_bio_prison_cell
*cell
)
287 struct pool
*pool
= tc
->pool
;
290 spin_lock_irqsave(&pool
->lock
, flags
);
291 dm_cell_release_no_holder(pool
->prison
, cell
, &pool
->deferred_bios
);
292 spin_unlock_irqrestore(&pool
->lock
, flags
);
297 static void cell_error(struct pool
*pool
,
298 struct dm_bio_prison_cell
*cell
)
300 dm_cell_error(pool
->prison
, cell
);
301 dm_bio_prison_free_cell(pool
->prison
, cell
);
304 /*----------------------------------------------------------------*/
307 * A global list of pools that uses a struct mapped_device as a key.
309 static struct dm_thin_pool_table
{
311 struct list_head pools
;
312 } dm_thin_pool_table
;
314 static void pool_table_init(void)
316 mutex_init(&dm_thin_pool_table
.mutex
);
317 INIT_LIST_HEAD(&dm_thin_pool_table
.pools
);
320 static void __pool_table_insert(struct pool
*pool
)
322 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
323 list_add(&pool
->list
, &dm_thin_pool_table
.pools
);
326 static void __pool_table_remove(struct pool
*pool
)
328 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
329 list_del(&pool
->list
);
332 static struct pool
*__pool_table_lookup(struct mapped_device
*md
)
334 struct pool
*pool
= NULL
, *tmp
;
336 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
338 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
339 if (tmp
->pool_md
== md
) {
348 static struct pool
*__pool_table_lookup_metadata_dev(struct block_device
*md_dev
)
350 struct pool
*pool
= NULL
, *tmp
;
352 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
354 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
355 if (tmp
->md_dev
== md_dev
) {
364 /*----------------------------------------------------------------*/
366 struct dm_thin_endio_hook
{
368 struct dm_deferred_entry
*shared_read_entry
;
369 struct dm_deferred_entry
*all_io_entry
;
370 struct dm_thin_new_mapping
*overwrite_mapping
;
373 static void requeue_bio_list(struct thin_c
*tc
, struct bio_list
*master
)
376 struct bio_list bios
;
379 bio_list_init(&bios
);
381 spin_lock_irqsave(&tc
->pool
->lock
, flags
);
382 bio_list_merge(&bios
, master
);
383 bio_list_init(master
);
384 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
386 while ((bio
= bio_list_pop(&bios
))) {
387 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
390 bio_endio(bio
, DM_ENDIO_REQUEUE
);
392 bio_list_add(master
, bio
);
396 static void requeue_io(struct thin_c
*tc
)
398 struct pool
*pool
= tc
->pool
;
400 requeue_bio_list(tc
, &pool
->deferred_bios
);
401 requeue_bio_list(tc
, &pool
->retry_on_resume_list
);
404 static void error_retry_list(struct pool
*pool
)
408 struct bio_list bios
;
410 bio_list_init(&bios
);
412 spin_lock_irqsave(&pool
->lock
, flags
);
413 bio_list_merge(&bios
, &pool
->retry_on_resume_list
);
414 bio_list_init(&pool
->retry_on_resume_list
);
415 spin_unlock_irqrestore(&pool
->lock
, flags
);
417 while ((bio
= bio_list_pop(&bios
)))
422 * This section of code contains the logic for processing a thin device's IO.
423 * Much of the code depends on pool object resources (lists, workqueues, etc)
424 * but most is exclusively called from the thin target rather than the thin-pool
428 static bool block_size_is_power_of_two(struct pool
*pool
)
430 return pool
->sectors_per_block_shift
>= 0;
433 static dm_block_t
get_bio_block(struct thin_c
*tc
, struct bio
*bio
)
435 struct pool
*pool
= tc
->pool
;
436 sector_t block_nr
= bio
->bi_iter
.bi_sector
;
438 if (block_size_is_power_of_two(pool
))
439 block_nr
>>= pool
->sectors_per_block_shift
;
441 (void) sector_div(block_nr
, pool
->sectors_per_block
);
446 static void remap(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
)
448 struct pool
*pool
= tc
->pool
;
449 sector_t bi_sector
= bio
->bi_iter
.bi_sector
;
451 bio
->bi_bdev
= tc
->pool_dev
->bdev
;
452 if (block_size_is_power_of_two(pool
))
453 bio
->bi_iter
.bi_sector
=
454 (block
<< pool
->sectors_per_block_shift
) |
455 (bi_sector
& (pool
->sectors_per_block
- 1));
457 bio
->bi_iter
.bi_sector
= (block
* pool
->sectors_per_block
) +
458 sector_div(bi_sector
, pool
->sectors_per_block
);
461 static void remap_to_origin(struct thin_c
*tc
, struct bio
*bio
)
463 bio
->bi_bdev
= tc
->origin_dev
->bdev
;
466 static int bio_triggers_commit(struct thin_c
*tc
, struct bio
*bio
)
468 return (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) &&
469 dm_thin_changed_this_transaction(tc
->td
);
472 static void inc_all_io_entry(struct pool
*pool
, struct bio
*bio
)
474 struct dm_thin_endio_hook
*h
;
476 if (bio
->bi_rw
& REQ_DISCARD
)
479 h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
480 h
->all_io_entry
= dm_deferred_entry_inc(pool
->all_io_ds
);
483 static void issue(struct thin_c
*tc
, struct bio
*bio
)
485 struct pool
*pool
= tc
->pool
;
488 if (!bio_triggers_commit(tc
, bio
)) {
489 generic_make_request(bio
);
494 * Complete bio with an error if earlier I/O caused changes to
495 * the metadata that can't be committed e.g, due to I/O errors
496 * on the metadata device.
498 if (dm_thin_aborted_changes(tc
->td
)) {
504 * Batch together any bios that trigger commits and then issue a
505 * single commit for them in process_deferred_bios().
507 spin_lock_irqsave(&pool
->lock
, flags
);
508 bio_list_add(&pool
->deferred_flush_bios
, bio
);
509 spin_unlock_irqrestore(&pool
->lock
, flags
);
512 static void remap_to_origin_and_issue(struct thin_c
*tc
, struct bio
*bio
)
514 remap_to_origin(tc
, bio
);
518 static void remap_and_issue(struct thin_c
*tc
, struct bio
*bio
,
521 remap(tc
, bio
, block
);
525 /*----------------------------------------------------------------*/
528 * Bio endio functions.
530 struct dm_thin_new_mapping
{
531 struct list_head list
;
536 bool definitely_not_shared
:1;
540 dm_block_t virt_block
;
541 dm_block_t data_block
;
542 struct dm_bio_prison_cell
*cell
, *cell2
;
545 * If the bio covers the whole area of a block then we can avoid
546 * zeroing or copying. Instead this bio is hooked. The bio will
547 * still be in the cell, so care has to be taken to avoid issuing
551 bio_end_io_t
*saved_bi_end_io
;
554 static void __maybe_add_mapping(struct dm_thin_new_mapping
*m
)
556 struct pool
*pool
= m
->tc
->pool
;
558 if (m
->quiesced
&& m
->prepared
) {
559 list_add_tail(&m
->list
, &pool
->prepared_mappings
);
564 static void copy_complete(int read_err
, unsigned long write_err
, void *context
)
567 struct dm_thin_new_mapping
*m
= context
;
568 struct pool
*pool
= m
->tc
->pool
;
570 m
->err
= read_err
|| write_err
? -EIO
: 0;
572 spin_lock_irqsave(&pool
->lock
, flags
);
574 __maybe_add_mapping(m
);
575 spin_unlock_irqrestore(&pool
->lock
, flags
);
578 static void overwrite_endio(struct bio
*bio
, int err
)
581 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
582 struct dm_thin_new_mapping
*m
= h
->overwrite_mapping
;
583 struct pool
*pool
= m
->tc
->pool
;
587 spin_lock_irqsave(&pool
->lock
, flags
);
589 __maybe_add_mapping(m
);
590 spin_unlock_irqrestore(&pool
->lock
, flags
);
593 /*----------------------------------------------------------------*/
600 * Prepared mapping jobs.
604 * This sends the bios in the cell back to the deferred_bios list.
606 static void cell_defer(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
608 struct pool
*pool
= tc
->pool
;
611 spin_lock_irqsave(&pool
->lock
, flags
);
612 cell_release(pool
, cell
, &pool
->deferred_bios
);
613 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
619 * Same as cell_defer above, except it omits the original holder of the cell.
621 static void cell_defer_no_holder(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
623 struct pool
*pool
= tc
->pool
;
626 spin_lock_irqsave(&pool
->lock
, flags
);
627 cell_release_no_holder(pool
, cell
, &pool
->deferred_bios
);
628 spin_unlock_irqrestore(&pool
->lock
, flags
);
633 static void process_prepared_mapping_fail(struct dm_thin_new_mapping
*m
)
636 m
->bio
->bi_end_io
= m
->saved_bi_end_io
;
637 atomic_inc(&m
->bio
->bi_remaining
);
639 cell_error(m
->tc
->pool
, m
->cell
);
641 mempool_free(m
, m
->tc
->pool
->mapping_pool
);
644 static void process_prepared_mapping(struct dm_thin_new_mapping
*m
)
646 struct thin_c
*tc
= m
->tc
;
647 struct pool
*pool
= tc
->pool
;
653 bio
->bi_end_io
= m
->saved_bi_end_io
;
654 atomic_inc(&bio
->bi_remaining
);
658 cell_error(pool
, m
->cell
);
663 * Commit the prepared block into the mapping btree.
664 * Any I/O for this block arriving after this point will get
665 * remapped to it directly.
667 r
= dm_thin_insert_block(tc
->td
, m
->virt_block
, m
->data_block
);
669 metadata_operation_failed(pool
, "dm_thin_insert_block", r
);
670 cell_error(pool
, m
->cell
);
675 * Release any bios held while the block was being provisioned.
676 * If we are processing a write bio that completely covers the block,
677 * we already processed it so can ignore it now when processing
678 * the bios in the cell.
681 cell_defer_no_holder(tc
, m
->cell
);
684 cell_defer(tc
, m
->cell
);
688 mempool_free(m
, pool
->mapping_pool
);
691 static void process_prepared_discard_fail(struct dm_thin_new_mapping
*m
)
693 struct thin_c
*tc
= m
->tc
;
695 bio_io_error(m
->bio
);
696 cell_defer_no_holder(tc
, m
->cell
);
697 cell_defer_no_holder(tc
, m
->cell2
);
698 mempool_free(m
, tc
->pool
->mapping_pool
);
701 static void process_prepared_discard_passdown(struct dm_thin_new_mapping
*m
)
703 struct thin_c
*tc
= m
->tc
;
705 inc_all_io_entry(tc
->pool
, m
->bio
);
706 cell_defer_no_holder(tc
, m
->cell
);
707 cell_defer_no_holder(tc
, m
->cell2
);
710 if (m
->definitely_not_shared
)
711 remap_and_issue(tc
, m
->bio
, m
->data_block
);
714 if (dm_pool_block_is_used(tc
->pool
->pmd
, m
->data_block
, &used
) || used
)
715 bio_endio(m
->bio
, 0);
717 remap_and_issue(tc
, m
->bio
, m
->data_block
);
720 bio_endio(m
->bio
, 0);
722 mempool_free(m
, tc
->pool
->mapping_pool
);
725 static void process_prepared_discard(struct dm_thin_new_mapping
*m
)
728 struct thin_c
*tc
= m
->tc
;
730 r
= dm_thin_remove_block(tc
->td
, m
->virt_block
);
732 DMERR_LIMIT("dm_thin_remove_block() failed");
734 process_prepared_discard_passdown(m
);
737 static void process_prepared(struct pool
*pool
, struct list_head
*head
,
738 process_mapping_fn
*fn
)
741 struct list_head maps
;
742 struct dm_thin_new_mapping
*m
, *tmp
;
744 INIT_LIST_HEAD(&maps
);
745 spin_lock_irqsave(&pool
->lock
, flags
);
746 list_splice_init(head
, &maps
);
747 spin_unlock_irqrestore(&pool
->lock
, flags
);
749 list_for_each_entry_safe(m
, tmp
, &maps
, list
)
756 static int io_overlaps_block(struct pool
*pool
, struct bio
*bio
)
758 return bio
->bi_iter
.bi_size
==
759 (pool
->sectors_per_block
<< SECTOR_SHIFT
);
762 static int io_overwrites_block(struct pool
*pool
, struct bio
*bio
)
764 return (bio_data_dir(bio
) == WRITE
) &&
765 io_overlaps_block(pool
, bio
);
768 static void save_and_set_endio(struct bio
*bio
, bio_end_io_t
**save
,
771 *save
= bio
->bi_end_io
;
775 static int ensure_next_mapping(struct pool
*pool
)
777 if (pool
->next_mapping
)
780 pool
->next_mapping
= mempool_alloc(pool
->mapping_pool
, GFP_ATOMIC
);
782 return pool
->next_mapping
? 0 : -ENOMEM
;
785 static struct dm_thin_new_mapping
*get_next_mapping(struct pool
*pool
)
787 struct dm_thin_new_mapping
*m
= pool
->next_mapping
;
789 BUG_ON(!pool
->next_mapping
);
791 memset(m
, 0, sizeof(struct dm_thin_new_mapping
));
792 INIT_LIST_HEAD(&m
->list
);
795 pool
->next_mapping
= NULL
;
800 static void schedule_copy(struct thin_c
*tc
, dm_block_t virt_block
,
801 struct dm_dev
*origin
, dm_block_t data_origin
,
802 dm_block_t data_dest
,
803 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
806 struct pool
*pool
= tc
->pool
;
807 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
810 m
->virt_block
= virt_block
;
811 m
->data_block
= data_dest
;
814 if (!dm_deferred_set_add_work(pool
->shared_read_ds
, &m
->list
))
818 * IO to pool_dev remaps to the pool target's data_dev.
820 * If the whole block of data is being overwritten, we can issue the
821 * bio immediately. Otherwise we use kcopyd to clone the data first.
823 if (io_overwrites_block(pool
, bio
)) {
824 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
826 h
->overwrite_mapping
= m
;
828 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
829 inc_all_io_entry(pool
, bio
);
830 remap_and_issue(tc
, bio
, data_dest
);
832 struct dm_io_region from
, to
;
834 from
.bdev
= origin
->bdev
;
835 from
.sector
= data_origin
* pool
->sectors_per_block
;
836 from
.count
= pool
->sectors_per_block
;
838 to
.bdev
= tc
->pool_dev
->bdev
;
839 to
.sector
= data_dest
* pool
->sectors_per_block
;
840 to
.count
= pool
->sectors_per_block
;
842 r
= dm_kcopyd_copy(pool
->copier
, &from
, 1, &to
,
843 0, copy_complete
, m
);
845 mempool_free(m
, pool
->mapping_pool
);
846 DMERR_LIMIT("dm_kcopyd_copy() failed");
847 cell_error(pool
, cell
);
852 static void schedule_internal_copy(struct thin_c
*tc
, dm_block_t virt_block
,
853 dm_block_t data_origin
, dm_block_t data_dest
,
854 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
856 schedule_copy(tc
, virt_block
, tc
->pool_dev
,
857 data_origin
, data_dest
, cell
, bio
);
860 static void schedule_external_copy(struct thin_c
*tc
, dm_block_t virt_block
,
861 dm_block_t data_dest
,
862 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
864 schedule_copy(tc
, virt_block
, tc
->origin_dev
,
865 virt_block
, data_dest
, cell
, bio
);
868 static void schedule_zero(struct thin_c
*tc
, dm_block_t virt_block
,
869 dm_block_t data_block
, struct dm_bio_prison_cell
*cell
,
872 struct pool
*pool
= tc
->pool
;
873 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
878 m
->virt_block
= virt_block
;
879 m
->data_block
= data_block
;
883 * If the whole block of data is being overwritten or we are not
884 * zeroing pre-existing data, we can issue the bio immediately.
885 * Otherwise we use kcopyd to zero the data first.
887 if (!pool
->pf
.zero_new_blocks
)
888 process_prepared_mapping(m
);
890 else if (io_overwrites_block(pool
, bio
)) {
891 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
893 h
->overwrite_mapping
= m
;
895 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
896 inc_all_io_entry(pool
, bio
);
897 remap_and_issue(tc
, bio
, data_block
);
900 struct dm_io_region to
;
902 to
.bdev
= tc
->pool_dev
->bdev
;
903 to
.sector
= data_block
* pool
->sectors_per_block
;
904 to
.count
= pool
->sectors_per_block
;
906 r
= dm_kcopyd_zero(pool
->copier
, 1, &to
, 0, copy_complete
, m
);
908 mempool_free(m
, pool
->mapping_pool
);
909 DMERR_LIMIT("dm_kcopyd_zero() failed");
910 cell_error(pool
, cell
);
916 * A non-zero return indicates read_only or fail_io mode.
917 * Many callers don't care about the return value.
919 static int commit(struct pool
*pool
)
923 if (get_pool_mode(pool
) != PM_WRITE
)
926 r
= dm_pool_commit_metadata(pool
->pmd
);
928 metadata_operation_failed(pool
, "dm_pool_commit_metadata", r
);
933 static void check_low_water_mark(struct pool
*pool
, dm_block_t free_blocks
)
937 if (free_blocks
<= pool
->low_water_blocks
&& !pool
->low_water_triggered
) {
938 DMWARN("%s: reached low water mark for data device: sending event.",
939 dm_device_name(pool
->pool_md
));
940 spin_lock_irqsave(&pool
->lock
, flags
);
941 pool
->low_water_triggered
= true;
942 spin_unlock_irqrestore(&pool
->lock
, flags
);
943 dm_table_event(pool
->ti
->table
);
947 static void set_pool_mode(struct pool
*pool
, enum pool_mode new_mode
);
949 static int alloc_data_block(struct thin_c
*tc
, dm_block_t
*result
)
952 dm_block_t free_blocks
;
953 struct pool
*pool
= tc
->pool
;
955 if (WARN_ON(get_pool_mode(pool
) != PM_WRITE
))
958 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
960 metadata_operation_failed(pool
, "dm_pool_get_free_block_count", r
);
964 check_low_water_mark(pool
, free_blocks
);
968 * Try to commit to see if that will free up some
975 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
977 metadata_operation_failed(pool
, "dm_pool_get_free_block_count", r
);
982 set_pool_mode(pool
, PM_OUT_OF_DATA_SPACE
);
987 r
= dm_pool_alloc_data_block(pool
->pmd
, result
);
989 metadata_operation_failed(pool
, "dm_pool_alloc_data_block", r
);
997 * If we have run out of space, queue bios until the device is
998 * resumed, presumably after having been reloaded with more space.
1000 static void retry_on_resume(struct bio
*bio
)
1002 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1003 struct thin_c
*tc
= h
->tc
;
1004 struct pool
*pool
= tc
->pool
;
1005 unsigned long flags
;
1007 spin_lock_irqsave(&pool
->lock
, flags
);
1008 bio_list_add(&pool
->retry_on_resume_list
, bio
);
1009 spin_unlock_irqrestore(&pool
->lock
, flags
);
1012 static bool should_error_unserviceable_bio(struct pool
*pool
)
1014 enum pool_mode m
= get_pool_mode(pool
);
1018 /* Shouldn't get here */
1019 DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
1022 case PM_OUT_OF_DATA_SPACE
:
1023 return pool
->pf
.error_if_no_space
;
1029 /* Shouldn't get here */
1030 DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1035 static void handle_unserviceable_bio(struct pool
*pool
, struct bio
*bio
)
1037 if (should_error_unserviceable_bio(pool
))
1040 retry_on_resume(bio
);
1043 static void retry_bios_on_resume(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
1046 struct bio_list bios
;
1048 if (should_error_unserviceable_bio(pool
)) {
1049 cell_error(pool
, cell
);
1053 bio_list_init(&bios
);
1054 cell_release(pool
, cell
, &bios
);
1056 if (should_error_unserviceable_bio(pool
))
1057 while ((bio
= bio_list_pop(&bios
)))
1060 while ((bio
= bio_list_pop(&bios
)))
1061 retry_on_resume(bio
);
1064 static void process_discard(struct thin_c
*tc
, struct bio
*bio
)
1067 unsigned long flags
;
1068 struct pool
*pool
= tc
->pool
;
1069 struct dm_bio_prison_cell
*cell
, *cell2
;
1070 struct dm_cell_key key
, key2
;
1071 dm_block_t block
= get_bio_block(tc
, bio
);
1072 struct dm_thin_lookup_result lookup_result
;
1073 struct dm_thin_new_mapping
*m
;
1075 build_virtual_key(tc
->td
, block
, &key
);
1076 if (bio_detain(tc
->pool
, &key
, bio
, &cell
))
1079 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1083 * Check nobody is fiddling with this pool block. This can
1084 * happen if someone's in the process of breaking sharing
1087 build_data_key(tc
->td
, lookup_result
.block
, &key2
);
1088 if (bio_detain(tc
->pool
, &key2
, bio
, &cell2
)) {
1089 cell_defer_no_holder(tc
, cell
);
1093 if (io_overlaps_block(pool
, bio
)) {
1095 * IO may still be going to the destination block. We must
1096 * quiesce before we can do the removal.
1098 m
= get_next_mapping(pool
);
1100 m
->pass_discard
= pool
->pf
.discard_passdown
;
1101 m
->definitely_not_shared
= !lookup_result
.shared
;
1102 m
->virt_block
= block
;
1103 m
->data_block
= lookup_result
.block
;
1108 if (!dm_deferred_set_add_work(pool
->all_io_ds
, &m
->list
)) {
1109 spin_lock_irqsave(&pool
->lock
, flags
);
1110 list_add_tail(&m
->list
, &pool
->prepared_discards
);
1111 spin_unlock_irqrestore(&pool
->lock
, flags
);
1115 inc_all_io_entry(pool
, bio
);
1116 cell_defer_no_holder(tc
, cell
);
1117 cell_defer_no_holder(tc
, cell2
);
1120 * The DM core makes sure that the discard doesn't span
1121 * a block boundary. So we submit the discard of a
1122 * partial block appropriately.
1124 if ((!lookup_result
.shared
) && pool
->pf
.discard_passdown
)
1125 remap_and_issue(tc
, bio
, lookup_result
.block
);
1133 * It isn't provisioned, just forget it.
1135 cell_defer_no_holder(tc
, cell
);
1140 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1142 cell_defer_no_holder(tc
, cell
);
1148 static void break_sharing(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1149 struct dm_cell_key
*key
,
1150 struct dm_thin_lookup_result
*lookup_result
,
1151 struct dm_bio_prison_cell
*cell
)
1154 dm_block_t data_block
;
1155 struct pool
*pool
= tc
->pool
;
1157 r
= alloc_data_block(tc
, &data_block
);
1160 schedule_internal_copy(tc
, block
, lookup_result
->block
,
1161 data_block
, cell
, bio
);
1165 retry_bios_on_resume(pool
, cell
);
1169 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1171 cell_error(pool
, cell
);
1176 static void process_shared_bio(struct thin_c
*tc
, struct bio
*bio
,
1178 struct dm_thin_lookup_result
*lookup_result
)
1180 struct dm_bio_prison_cell
*cell
;
1181 struct pool
*pool
= tc
->pool
;
1182 struct dm_cell_key key
;
1185 * If cell is already occupied, then sharing is already in the process
1186 * of being broken so we have nothing further to do here.
1188 build_data_key(tc
->td
, lookup_result
->block
, &key
);
1189 if (bio_detain(pool
, &key
, bio
, &cell
))
1192 if (bio_data_dir(bio
) == WRITE
&& bio
->bi_iter
.bi_size
)
1193 break_sharing(tc
, bio
, block
, &key
, lookup_result
, cell
);
1195 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1197 h
->shared_read_entry
= dm_deferred_entry_inc(pool
->shared_read_ds
);
1198 inc_all_io_entry(pool
, bio
);
1199 cell_defer_no_holder(tc
, cell
);
1201 remap_and_issue(tc
, bio
, lookup_result
->block
);
1205 static void provision_block(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1206 struct dm_bio_prison_cell
*cell
)
1209 dm_block_t data_block
;
1210 struct pool
*pool
= tc
->pool
;
1213 * Remap empty bios (flushes) immediately, without provisioning.
1215 if (!bio
->bi_iter
.bi_size
) {
1216 inc_all_io_entry(pool
, bio
);
1217 cell_defer_no_holder(tc
, cell
);
1219 remap_and_issue(tc
, bio
, 0);
1224 * Fill read bios with zeroes and complete them immediately.
1226 if (bio_data_dir(bio
) == READ
) {
1228 cell_defer_no_holder(tc
, cell
);
1233 r
= alloc_data_block(tc
, &data_block
);
1237 schedule_external_copy(tc
, block
, data_block
, cell
, bio
);
1239 schedule_zero(tc
, block
, data_block
, cell
, bio
);
1243 retry_bios_on_resume(pool
, cell
);
1247 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1249 cell_error(pool
, cell
);
1254 static void process_bio(struct thin_c
*tc
, struct bio
*bio
)
1257 struct pool
*pool
= tc
->pool
;
1258 dm_block_t block
= get_bio_block(tc
, bio
);
1259 struct dm_bio_prison_cell
*cell
;
1260 struct dm_cell_key key
;
1261 struct dm_thin_lookup_result lookup_result
;
1264 * If cell is already occupied, then the block is already
1265 * being provisioned so we have nothing further to do here.
1267 build_virtual_key(tc
->td
, block
, &key
);
1268 if (bio_detain(pool
, &key
, bio
, &cell
))
1271 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1274 if (lookup_result
.shared
) {
1275 process_shared_bio(tc
, bio
, block
, &lookup_result
);
1276 cell_defer_no_holder(tc
, cell
); /* FIXME: pass this cell into process_shared? */
1278 inc_all_io_entry(pool
, bio
);
1279 cell_defer_no_holder(tc
, cell
);
1281 remap_and_issue(tc
, bio
, lookup_result
.block
);
1286 if (bio_data_dir(bio
) == READ
&& tc
->origin_dev
) {
1287 inc_all_io_entry(pool
, bio
);
1288 cell_defer_no_holder(tc
, cell
);
1290 remap_to_origin_and_issue(tc
, bio
);
1292 provision_block(tc
, bio
, block
, cell
);
1296 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1298 cell_defer_no_holder(tc
, cell
);
1304 static void process_bio_read_only(struct thin_c
*tc
, struct bio
*bio
)
1307 int rw
= bio_data_dir(bio
);
1308 dm_block_t block
= get_bio_block(tc
, bio
);
1309 struct dm_thin_lookup_result lookup_result
;
1311 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1314 if (lookup_result
.shared
&& (rw
== WRITE
) && bio
->bi_iter
.bi_size
)
1315 handle_unserviceable_bio(tc
->pool
, bio
);
1317 inc_all_io_entry(tc
->pool
, bio
);
1318 remap_and_issue(tc
, bio
, lookup_result
.block
);
1324 handle_unserviceable_bio(tc
->pool
, bio
);
1328 if (tc
->origin_dev
) {
1329 inc_all_io_entry(tc
->pool
, bio
);
1330 remap_to_origin_and_issue(tc
, bio
);
1339 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1346 static void process_bio_success(struct thin_c
*tc
, struct bio
*bio
)
1351 static void process_bio_fail(struct thin_c
*tc
, struct bio
*bio
)
1357 * FIXME: should we also commit due to size of transaction, measured in
1360 static int need_commit_due_to_time(struct pool
*pool
)
1362 return jiffies
< pool
->last_commit_jiffies
||
1363 jiffies
> pool
->last_commit_jiffies
+ COMMIT_PERIOD
;
1366 static void process_deferred_bios(struct pool
*pool
)
1368 unsigned long flags
;
1370 struct bio_list bios
;
1372 bio_list_init(&bios
);
1374 spin_lock_irqsave(&pool
->lock
, flags
);
1375 bio_list_merge(&bios
, &pool
->deferred_bios
);
1376 bio_list_init(&pool
->deferred_bios
);
1377 spin_unlock_irqrestore(&pool
->lock
, flags
);
1379 while ((bio
= bio_list_pop(&bios
))) {
1380 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1381 struct thin_c
*tc
= h
->tc
;
1383 if (tc
->requeue_mode
) {
1384 bio_endio(bio
, DM_ENDIO_REQUEUE
);
1389 * If we've got no free new_mapping structs, and processing
1390 * this bio might require one, we pause until there are some
1391 * prepared mappings to process.
1393 if (ensure_next_mapping(pool
)) {
1394 spin_lock_irqsave(&pool
->lock
, flags
);
1395 bio_list_merge(&pool
->deferred_bios
, &bios
);
1396 spin_unlock_irqrestore(&pool
->lock
, flags
);
1401 if (bio
->bi_rw
& REQ_DISCARD
)
1402 pool
->process_discard(tc
, bio
);
1404 pool
->process_bio(tc
, bio
);
1408 * If there are any deferred flush bios, we must commit
1409 * the metadata before issuing them.
1411 bio_list_init(&bios
);
1412 spin_lock_irqsave(&pool
->lock
, flags
);
1413 bio_list_merge(&bios
, &pool
->deferred_flush_bios
);
1414 bio_list_init(&pool
->deferred_flush_bios
);
1415 spin_unlock_irqrestore(&pool
->lock
, flags
);
1417 if (bio_list_empty(&bios
) &&
1418 !(dm_pool_changed_this_transaction(pool
->pmd
) && need_commit_due_to_time(pool
)))
1422 while ((bio
= bio_list_pop(&bios
)))
1426 pool
->last_commit_jiffies
= jiffies
;
1428 while ((bio
= bio_list_pop(&bios
)))
1429 generic_make_request(bio
);
1432 static void do_worker(struct work_struct
*ws
)
1434 struct pool
*pool
= container_of(ws
, struct pool
, worker
);
1436 process_prepared(pool
, &pool
->prepared_mappings
, &pool
->process_prepared_mapping
);
1437 process_prepared(pool
, &pool
->prepared_discards
, &pool
->process_prepared_discard
);
1438 process_deferred_bios(pool
);
1442 * We want to commit periodically so that not too much
1443 * unwritten data builds up.
1445 static void do_waker(struct work_struct
*ws
)
1447 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
, waker
);
1449 queue_delayed_work(pool
->wq
, &pool
->waker
, COMMIT_PERIOD
);
1452 /*----------------------------------------------------------------*/
1454 struct noflush_work
{
1455 struct work_struct worker
;
1459 wait_queue_head_t wait
;
1462 static void complete_noflush_work(struct noflush_work
*w
)
1464 atomic_set(&w
->complete
, 1);
1468 static void do_noflush_start(struct work_struct
*ws
)
1470 struct noflush_work
*w
= container_of(ws
, struct noflush_work
, worker
);
1471 w
->tc
->requeue_mode
= true;
1473 complete_noflush_work(w
);
1476 static void do_noflush_stop(struct work_struct
*ws
)
1478 struct noflush_work
*w
= container_of(ws
, struct noflush_work
, worker
);
1479 w
->tc
->requeue_mode
= false;
1480 complete_noflush_work(w
);
1483 static void noflush_work(struct thin_c
*tc
, void (*fn
)(struct work_struct
*))
1485 struct noflush_work w
;
1487 INIT_WORK(&w
.worker
, fn
);
1489 atomic_set(&w
.complete
, 0);
1490 init_waitqueue_head(&w
.wait
);
1492 queue_work(tc
->pool
->wq
, &w
.worker
);
1494 wait_event(w
.wait
, atomic_read(&w
.complete
));
1497 /*----------------------------------------------------------------*/
1499 static enum pool_mode
get_pool_mode(struct pool
*pool
)
1501 return pool
->pf
.mode
;
1504 static void notify_of_pool_mode_change(struct pool
*pool
, const char *new_mode
)
1506 dm_table_event(pool
->ti
->table
);
1507 DMINFO("%s: switching pool to %s mode",
1508 dm_device_name(pool
->pool_md
), new_mode
);
1511 static void set_pool_mode(struct pool
*pool
, enum pool_mode new_mode
)
1513 struct pool_c
*pt
= pool
->ti
->private;
1514 bool needs_check
= dm_pool_metadata_needs_check(pool
->pmd
);
1515 enum pool_mode old_mode
= get_pool_mode(pool
);
1518 * Never allow the pool to transition to PM_WRITE mode if user
1519 * intervention is required to verify metadata and data consistency.
1521 if (new_mode
== PM_WRITE
&& needs_check
) {
1522 DMERR("%s: unable to switch pool to write mode until repaired.",
1523 dm_device_name(pool
->pool_md
));
1524 if (old_mode
!= new_mode
)
1525 new_mode
= old_mode
;
1527 new_mode
= PM_READ_ONLY
;
1530 * If we were in PM_FAIL mode, rollback of metadata failed. We're
1531 * not going to recover without a thin_repair. So we never let the
1532 * pool move out of the old mode.
1534 if (old_mode
== PM_FAIL
)
1535 new_mode
= old_mode
;
1539 if (old_mode
!= new_mode
)
1540 notify_of_pool_mode_change(pool
, "failure");
1541 dm_pool_metadata_read_only(pool
->pmd
);
1542 pool
->process_bio
= process_bio_fail
;
1543 pool
->process_discard
= process_bio_fail
;
1544 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1545 pool
->process_prepared_discard
= process_prepared_discard_fail
;
1547 error_retry_list(pool
);
1551 if (old_mode
!= new_mode
)
1552 notify_of_pool_mode_change(pool
, "read-only");
1553 dm_pool_metadata_read_only(pool
->pmd
);
1554 pool
->process_bio
= process_bio_read_only
;
1555 pool
->process_discard
= process_bio_success
;
1556 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1557 pool
->process_prepared_discard
= process_prepared_discard_passdown
;
1559 error_retry_list(pool
);
1562 case PM_OUT_OF_DATA_SPACE
:
1564 * Ideally we'd never hit this state; the low water mark
1565 * would trigger userland to extend the pool before we
1566 * completely run out of data space. However, many small
1567 * IOs to unprovisioned space can consume data space at an
1568 * alarming rate. Adjust your low water mark if you're
1569 * frequently seeing this mode.
1571 if (old_mode
!= new_mode
)
1572 notify_of_pool_mode_change(pool
, "out-of-data-space");
1573 pool
->process_bio
= process_bio_read_only
;
1574 pool
->process_discard
= process_discard
;
1575 pool
->process_prepared_mapping
= process_prepared_mapping
;
1576 pool
->process_prepared_discard
= process_prepared_discard_passdown
;
1580 if (old_mode
!= new_mode
)
1581 notify_of_pool_mode_change(pool
, "write");
1582 dm_pool_metadata_read_write(pool
->pmd
);
1583 pool
->process_bio
= process_bio
;
1584 pool
->process_discard
= process_discard
;
1585 pool
->process_prepared_mapping
= process_prepared_mapping
;
1586 pool
->process_prepared_discard
= process_prepared_discard
;
1590 pool
->pf
.mode
= new_mode
;
1592 * The pool mode may have changed, sync it so bind_control_target()
1593 * doesn't cause an unexpected mode transition on resume.
1595 pt
->adjusted_pf
.mode
= new_mode
;
1598 static void abort_transaction(struct pool
*pool
)
1600 const char *dev_name
= dm_device_name(pool
->pool_md
);
1602 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name
);
1603 if (dm_pool_abort_metadata(pool
->pmd
)) {
1604 DMERR("%s: failed to abort metadata transaction", dev_name
);
1605 set_pool_mode(pool
, PM_FAIL
);
1608 if (dm_pool_metadata_set_needs_check(pool
->pmd
)) {
1609 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name
);
1610 set_pool_mode(pool
, PM_FAIL
);
1614 static void metadata_operation_failed(struct pool
*pool
, const char *op
, int r
)
1616 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
1617 dm_device_name(pool
->pool_md
), op
, r
);
1619 abort_transaction(pool
);
1620 set_pool_mode(pool
, PM_READ_ONLY
);
1623 /*----------------------------------------------------------------*/
1626 * Mapping functions.
1630 * Called only while mapping a thin bio to hand it over to the workqueue.
1632 static void thin_defer_bio(struct thin_c
*tc
, struct bio
*bio
)
1634 unsigned long flags
;
1635 struct pool
*pool
= tc
->pool
;
1637 spin_lock_irqsave(&pool
->lock
, flags
);
1638 bio_list_add(&pool
->deferred_bios
, bio
);
1639 spin_unlock_irqrestore(&pool
->lock
, flags
);
1644 static void thin_hook_bio(struct thin_c
*tc
, struct bio
*bio
)
1646 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1649 h
->shared_read_entry
= NULL
;
1650 h
->all_io_entry
= NULL
;
1651 h
->overwrite_mapping
= NULL
;
1655 * Non-blocking function called from the thin target's map function.
1657 static int thin_bio_map(struct dm_target
*ti
, struct bio
*bio
)
1660 struct thin_c
*tc
= ti
->private;
1661 dm_block_t block
= get_bio_block(tc
, bio
);
1662 struct dm_thin_device
*td
= tc
->td
;
1663 struct dm_thin_lookup_result result
;
1664 struct dm_bio_prison_cell cell1
, cell2
;
1665 struct dm_bio_prison_cell
*cell_result
;
1666 struct dm_cell_key key
;
1668 thin_hook_bio(tc
, bio
);
1670 if (tc
->requeue_mode
) {
1671 bio_endio(bio
, DM_ENDIO_REQUEUE
);
1672 return DM_MAPIO_SUBMITTED
;
1675 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
1677 return DM_MAPIO_SUBMITTED
;
1680 if (bio
->bi_rw
& (REQ_DISCARD
| REQ_FLUSH
| REQ_FUA
)) {
1681 thin_defer_bio(tc
, bio
);
1682 return DM_MAPIO_SUBMITTED
;
1685 r
= dm_thin_find_block(td
, block
, 0, &result
);
1688 * Note that we defer readahead too.
1692 if (unlikely(result
.shared
)) {
1694 * We have a race condition here between the
1695 * result.shared value returned by the lookup and
1696 * snapshot creation, which may cause new
1699 * To avoid this always quiesce the origin before
1700 * taking the snap. You want to do this anyway to
1701 * ensure a consistent application view
1704 * More distant ancestors are irrelevant. The
1705 * shared flag will be set in their case.
1707 thin_defer_bio(tc
, bio
);
1708 return DM_MAPIO_SUBMITTED
;
1711 build_virtual_key(tc
->td
, block
, &key
);
1712 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell1
, &cell_result
))
1713 return DM_MAPIO_SUBMITTED
;
1715 build_data_key(tc
->td
, result
.block
, &key
);
1716 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell2
, &cell_result
)) {
1717 cell_defer_no_holder_no_free(tc
, &cell1
);
1718 return DM_MAPIO_SUBMITTED
;
1721 inc_all_io_entry(tc
->pool
, bio
);
1722 cell_defer_no_holder_no_free(tc
, &cell2
);
1723 cell_defer_no_holder_no_free(tc
, &cell1
);
1725 remap(tc
, bio
, result
.block
);
1726 return DM_MAPIO_REMAPPED
;
1729 if (get_pool_mode(tc
->pool
) == PM_READ_ONLY
) {
1731 * This block isn't provisioned, and we have no way
1734 handle_unserviceable_bio(tc
->pool
, bio
);
1735 return DM_MAPIO_SUBMITTED
;
1741 * In future, the failed dm_thin_find_block above could
1742 * provide the hint to load the metadata into cache.
1744 thin_defer_bio(tc
, bio
);
1745 return DM_MAPIO_SUBMITTED
;
1749 * Must always call bio_io_error on failure.
1750 * dm_thin_find_block can fail with -EINVAL if the
1751 * pool is switched to fail-io mode.
1754 return DM_MAPIO_SUBMITTED
;
1758 static int pool_is_congested(struct dm_target_callbacks
*cb
, int bdi_bits
)
1761 unsigned long flags
;
1762 struct pool_c
*pt
= container_of(cb
, struct pool_c
, callbacks
);
1764 spin_lock_irqsave(&pt
->pool
->lock
, flags
);
1765 r
= !bio_list_empty(&pt
->pool
->retry_on_resume_list
);
1766 spin_unlock_irqrestore(&pt
->pool
->lock
, flags
);
1769 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1770 r
= bdi_congested(&q
->backing_dev_info
, bdi_bits
);
1776 static void __requeue_bios(struct pool
*pool
)
1778 bio_list_merge(&pool
->deferred_bios
, &pool
->retry_on_resume_list
);
1779 bio_list_init(&pool
->retry_on_resume_list
);
1782 /*----------------------------------------------------------------
1783 * Binding of control targets to a pool object
1784 *--------------------------------------------------------------*/
1785 static bool data_dev_supports_discard(struct pool_c
*pt
)
1787 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1789 return q
&& blk_queue_discard(q
);
1792 static bool is_factor(sector_t block_size
, uint32_t n
)
1794 return !sector_div(block_size
, n
);
1798 * If discard_passdown was enabled verify that the data device
1799 * supports discards. Disable discard_passdown if not.
1801 static void disable_passdown_if_not_supported(struct pool_c
*pt
)
1803 struct pool
*pool
= pt
->pool
;
1804 struct block_device
*data_bdev
= pt
->data_dev
->bdev
;
1805 struct queue_limits
*data_limits
= &bdev_get_queue(data_bdev
)->limits
;
1806 sector_t block_size
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
1807 const char *reason
= NULL
;
1808 char buf
[BDEVNAME_SIZE
];
1810 if (!pt
->adjusted_pf
.discard_passdown
)
1813 if (!data_dev_supports_discard(pt
))
1814 reason
= "discard unsupported";
1816 else if (data_limits
->max_discard_sectors
< pool
->sectors_per_block
)
1817 reason
= "max discard sectors smaller than a block";
1819 else if (data_limits
->discard_granularity
> block_size
)
1820 reason
= "discard granularity larger than a block";
1822 else if (!is_factor(block_size
, data_limits
->discard_granularity
))
1823 reason
= "discard granularity not a factor of block size";
1826 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev
, buf
), reason
);
1827 pt
->adjusted_pf
.discard_passdown
= false;
1831 static int bind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1833 struct pool_c
*pt
= ti
->private;
1836 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
1838 enum pool_mode old_mode
= get_pool_mode(pool
);
1839 enum pool_mode new_mode
= pt
->adjusted_pf
.mode
;
1842 * Don't change the pool's mode until set_pool_mode() below.
1843 * Otherwise the pool's process_* function pointers may
1844 * not match the desired pool mode.
1846 pt
->adjusted_pf
.mode
= old_mode
;
1849 pool
->pf
= pt
->adjusted_pf
;
1850 pool
->low_water_blocks
= pt
->low_water_blocks
;
1852 set_pool_mode(pool
, new_mode
);
1857 static void unbind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1863 /*----------------------------------------------------------------
1865 *--------------------------------------------------------------*/
1866 /* Initialize pool features. */
1867 static void pool_features_init(struct pool_features
*pf
)
1869 pf
->mode
= PM_WRITE
;
1870 pf
->zero_new_blocks
= true;
1871 pf
->discard_enabled
= true;
1872 pf
->discard_passdown
= true;
1873 pf
->error_if_no_space
= false;
1876 static void __pool_destroy(struct pool
*pool
)
1878 __pool_table_remove(pool
);
1880 if (dm_pool_metadata_close(pool
->pmd
) < 0)
1881 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1883 dm_bio_prison_destroy(pool
->prison
);
1884 dm_kcopyd_client_destroy(pool
->copier
);
1887 destroy_workqueue(pool
->wq
);
1889 if (pool
->next_mapping
)
1890 mempool_free(pool
->next_mapping
, pool
->mapping_pool
);
1891 mempool_destroy(pool
->mapping_pool
);
1892 dm_deferred_set_destroy(pool
->shared_read_ds
);
1893 dm_deferred_set_destroy(pool
->all_io_ds
);
1897 static struct kmem_cache
*_new_mapping_cache
;
1899 static struct pool
*pool_create(struct mapped_device
*pool_md
,
1900 struct block_device
*metadata_dev
,
1901 unsigned long block_size
,
1902 int read_only
, char **error
)
1907 struct dm_pool_metadata
*pmd
;
1908 bool format_device
= read_only
? false : true;
1910 pmd
= dm_pool_metadata_open(metadata_dev
, block_size
, format_device
);
1912 *error
= "Error creating metadata object";
1913 return (struct pool
*)pmd
;
1916 pool
= kmalloc(sizeof(*pool
), GFP_KERNEL
);
1918 *error
= "Error allocating memory for pool";
1919 err_p
= ERR_PTR(-ENOMEM
);
1924 pool
->sectors_per_block
= block_size
;
1925 if (block_size
& (block_size
- 1))
1926 pool
->sectors_per_block_shift
= -1;
1928 pool
->sectors_per_block_shift
= __ffs(block_size
);
1929 pool
->low_water_blocks
= 0;
1930 pool_features_init(&pool
->pf
);
1931 pool
->prison
= dm_bio_prison_create(PRISON_CELLS
);
1932 if (!pool
->prison
) {
1933 *error
= "Error creating pool's bio prison";
1934 err_p
= ERR_PTR(-ENOMEM
);
1938 pool
->copier
= dm_kcopyd_client_create(&dm_kcopyd_throttle
);
1939 if (IS_ERR(pool
->copier
)) {
1940 r
= PTR_ERR(pool
->copier
);
1941 *error
= "Error creating pool's kcopyd client";
1943 goto bad_kcopyd_client
;
1947 * Create singlethreaded workqueue that will service all devices
1948 * that use this metadata.
1950 pool
->wq
= alloc_ordered_workqueue("dm-" DM_MSG_PREFIX
, WQ_MEM_RECLAIM
);
1952 *error
= "Error creating pool's workqueue";
1953 err_p
= ERR_PTR(-ENOMEM
);
1957 INIT_WORK(&pool
->worker
, do_worker
);
1958 INIT_DELAYED_WORK(&pool
->waker
, do_waker
);
1959 spin_lock_init(&pool
->lock
);
1960 bio_list_init(&pool
->deferred_bios
);
1961 bio_list_init(&pool
->deferred_flush_bios
);
1962 INIT_LIST_HEAD(&pool
->prepared_mappings
);
1963 INIT_LIST_HEAD(&pool
->prepared_discards
);
1964 pool
->low_water_triggered
= false;
1965 bio_list_init(&pool
->retry_on_resume_list
);
1967 pool
->shared_read_ds
= dm_deferred_set_create();
1968 if (!pool
->shared_read_ds
) {
1969 *error
= "Error creating pool's shared read deferred set";
1970 err_p
= ERR_PTR(-ENOMEM
);
1971 goto bad_shared_read_ds
;
1974 pool
->all_io_ds
= dm_deferred_set_create();
1975 if (!pool
->all_io_ds
) {
1976 *error
= "Error creating pool's all io deferred set";
1977 err_p
= ERR_PTR(-ENOMEM
);
1981 pool
->next_mapping
= NULL
;
1982 pool
->mapping_pool
= mempool_create_slab_pool(MAPPING_POOL_SIZE
,
1983 _new_mapping_cache
);
1984 if (!pool
->mapping_pool
) {
1985 *error
= "Error creating pool's mapping mempool";
1986 err_p
= ERR_PTR(-ENOMEM
);
1987 goto bad_mapping_pool
;
1990 pool
->ref_count
= 1;
1991 pool
->last_commit_jiffies
= jiffies
;
1992 pool
->pool_md
= pool_md
;
1993 pool
->md_dev
= metadata_dev
;
1994 __pool_table_insert(pool
);
1999 dm_deferred_set_destroy(pool
->all_io_ds
);
2001 dm_deferred_set_destroy(pool
->shared_read_ds
);
2003 destroy_workqueue(pool
->wq
);
2005 dm_kcopyd_client_destroy(pool
->copier
);
2007 dm_bio_prison_destroy(pool
->prison
);
2011 if (dm_pool_metadata_close(pmd
))
2012 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
2017 static void __pool_inc(struct pool
*pool
)
2019 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
2023 static void __pool_dec(struct pool
*pool
)
2025 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
2026 BUG_ON(!pool
->ref_count
);
2027 if (!--pool
->ref_count
)
2028 __pool_destroy(pool
);
2031 static struct pool
*__pool_find(struct mapped_device
*pool_md
,
2032 struct block_device
*metadata_dev
,
2033 unsigned long block_size
, int read_only
,
2034 char **error
, int *created
)
2036 struct pool
*pool
= __pool_table_lookup_metadata_dev(metadata_dev
);
2039 if (pool
->pool_md
!= pool_md
) {
2040 *error
= "metadata device already in use by a pool";
2041 return ERR_PTR(-EBUSY
);
2046 pool
= __pool_table_lookup(pool_md
);
2048 if (pool
->md_dev
!= metadata_dev
) {
2049 *error
= "different pool cannot replace a pool";
2050 return ERR_PTR(-EINVAL
);
2055 pool
= pool_create(pool_md
, metadata_dev
, block_size
, read_only
, error
);
2063 /*----------------------------------------------------------------
2064 * Pool target methods
2065 *--------------------------------------------------------------*/
2066 static void pool_dtr(struct dm_target
*ti
)
2068 struct pool_c
*pt
= ti
->private;
2070 mutex_lock(&dm_thin_pool_table
.mutex
);
2072 unbind_control_target(pt
->pool
, ti
);
2073 __pool_dec(pt
->pool
);
2074 dm_put_device(ti
, pt
->metadata_dev
);
2075 dm_put_device(ti
, pt
->data_dev
);
2078 mutex_unlock(&dm_thin_pool_table
.mutex
);
2081 static int parse_pool_features(struct dm_arg_set
*as
, struct pool_features
*pf
,
2082 struct dm_target
*ti
)
2086 const char *arg_name
;
2088 static struct dm_arg _args
[] = {
2089 {0, 4, "Invalid number of pool feature arguments"},
2093 * No feature arguments supplied.
2098 r
= dm_read_arg_group(_args
, as
, &argc
, &ti
->error
);
2102 while (argc
&& !r
) {
2103 arg_name
= dm_shift_arg(as
);
2106 if (!strcasecmp(arg_name
, "skip_block_zeroing"))
2107 pf
->zero_new_blocks
= false;
2109 else if (!strcasecmp(arg_name
, "ignore_discard"))
2110 pf
->discard_enabled
= false;
2112 else if (!strcasecmp(arg_name
, "no_discard_passdown"))
2113 pf
->discard_passdown
= false;
2115 else if (!strcasecmp(arg_name
, "read_only"))
2116 pf
->mode
= PM_READ_ONLY
;
2118 else if (!strcasecmp(arg_name
, "error_if_no_space"))
2119 pf
->error_if_no_space
= true;
2122 ti
->error
= "Unrecognised pool feature requested";
2131 static void metadata_low_callback(void *context
)
2133 struct pool
*pool
= context
;
2135 DMWARN("%s: reached low water mark for metadata device: sending event.",
2136 dm_device_name(pool
->pool_md
));
2138 dm_table_event(pool
->ti
->table
);
2141 static sector_t
get_dev_size(struct block_device
*bdev
)
2143 return i_size_read(bdev
->bd_inode
) >> SECTOR_SHIFT
;
2146 static void warn_if_metadata_device_too_big(struct block_device
*bdev
)
2148 sector_t metadata_dev_size
= get_dev_size(bdev
);
2149 char buffer
[BDEVNAME_SIZE
];
2151 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS_WARNING
)
2152 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
2153 bdevname(bdev
, buffer
), THIN_METADATA_MAX_SECTORS
);
2156 static sector_t
get_metadata_dev_size(struct block_device
*bdev
)
2158 sector_t metadata_dev_size
= get_dev_size(bdev
);
2160 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS
)
2161 metadata_dev_size
= THIN_METADATA_MAX_SECTORS
;
2163 return metadata_dev_size
;
2166 static dm_block_t
get_metadata_dev_size_in_blocks(struct block_device
*bdev
)
2168 sector_t metadata_dev_size
= get_metadata_dev_size(bdev
);
2170 sector_div(metadata_dev_size
, THIN_METADATA_BLOCK_SIZE
);
2172 return metadata_dev_size
;
2176 * When a metadata threshold is crossed a dm event is triggered, and
2177 * userland should respond by growing the metadata device. We could let
2178 * userland set the threshold, like we do with the data threshold, but I'm
2179 * not sure they know enough to do this well.
2181 static dm_block_t
calc_metadata_threshold(struct pool_c
*pt
)
2184 * 4M is ample for all ops with the possible exception of thin
2185 * device deletion which is harmless if it fails (just retry the
2186 * delete after you've grown the device).
2188 dm_block_t quarter
= get_metadata_dev_size_in_blocks(pt
->metadata_dev
->bdev
) / 4;
2189 return min((dm_block_t
)1024ULL /* 4M */, quarter
);
2193 * thin-pool <metadata dev> <data dev>
2194 * <data block size (sectors)>
2195 * <low water mark (blocks)>
2196 * [<#feature args> [<arg>]*]
2198 * Optional feature arguments are:
2199 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
2200 * ignore_discard: disable discard
2201 * no_discard_passdown: don't pass discards down to the data device
2202 * read_only: Don't allow any changes to be made to the pool metadata.
2203 * error_if_no_space: error IOs, instead of queueing, if no space.
2205 static int pool_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2207 int r
, pool_created
= 0;
2210 struct pool_features pf
;
2211 struct dm_arg_set as
;
2212 struct dm_dev
*data_dev
;
2213 unsigned long block_size
;
2214 dm_block_t low_water_blocks
;
2215 struct dm_dev
*metadata_dev
;
2216 fmode_t metadata_mode
;
2219 * FIXME Remove validation from scope of lock.
2221 mutex_lock(&dm_thin_pool_table
.mutex
);
2224 ti
->error
= "Invalid argument count";
2233 * Set default pool features.
2235 pool_features_init(&pf
);
2237 dm_consume_args(&as
, 4);
2238 r
= parse_pool_features(&as
, &pf
, ti
);
2242 metadata_mode
= FMODE_READ
| ((pf
.mode
== PM_READ_ONLY
) ? 0 : FMODE_WRITE
);
2243 r
= dm_get_device(ti
, argv
[0], metadata_mode
, &metadata_dev
);
2245 ti
->error
= "Error opening metadata block device";
2248 warn_if_metadata_device_too_big(metadata_dev
->bdev
);
2250 r
= dm_get_device(ti
, argv
[1], FMODE_READ
| FMODE_WRITE
, &data_dev
);
2252 ti
->error
= "Error getting data device";
2256 if (kstrtoul(argv
[2], 10, &block_size
) || !block_size
||
2257 block_size
< DATA_DEV_BLOCK_SIZE_MIN_SECTORS
||
2258 block_size
> DATA_DEV_BLOCK_SIZE_MAX_SECTORS
||
2259 block_size
& (DATA_DEV_BLOCK_SIZE_MIN_SECTORS
- 1)) {
2260 ti
->error
= "Invalid block size";
2265 if (kstrtoull(argv
[3], 10, (unsigned long long *)&low_water_blocks
)) {
2266 ti
->error
= "Invalid low water mark";
2271 pt
= kzalloc(sizeof(*pt
), GFP_KERNEL
);
2277 pool
= __pool_find(dm_table_get_md(ti
->table
), metadata_dev
->bdev
,
2278 block_size
, pf
.mode
== PM_READ_ONLY
, &ti
->error
, &pool_created
);
2285 * 'pool_created' reflects whether this is the first table load.
2286 * Top level discard support is not allowed to be changed after
2287 * initial load. This would require a pool reload to trigger thin
2290 if (!pool_created
&& pf
.discard_enabled
!= pool
->pf
.discard_enabled
) {
2291 ti
->error
= "Discard support cannot be disabled once enabled";
2293 goto out_flags_changed
;
2298 pt
->metadata_dev
= metadata_dev
;
2299 pt
->data_dev
= data_dev
;
2300 pt
->low_water_blocks
= low_water_blocks
;
2301 pt
->adjusted_pf
= pt
->requested_pf
= pf
;
2302 ti
->num_flush_bios
= 1;
2305 * Only need to enable discards if the pool should pass
2306 * them down to the data device. The thin device's discard
2307 * processing will cause mappings to be removed from the btree.
2309 ti
->discard_zeroes_data_unsupported
= true;
2310 if (pf
.discard_enabled
&& pf
.discard_passdown
) {
2311 ti
->num_discard_bios
= 1;
2314 * Setting 'discards_supported' circumvents the normal
2315 * stacking of discard limits (this keeps the pool and
2316 * thin devices' discard limits consistent).
2318 ti
->discards_supported
= true;
2322 r
= dm_pool_register_metadata_threshold(pt
->pool
->pmd
,
2323 calc_metadata_threshold(pt
),
2324 metadata_low_callback
,
2329 pt
->callbacks
.congested_fn
= pool_is_congested
;
2330 dm_table_add_target_callbacks(ti
->table
, &pt
->callbacks
);
2332 mutex_unlock(&dm_thin_pool_table
.mutex
);
2341 dm_put_device(ti
, data_dev
);
2343 dm_put_device(ti
, metadata_dev
);
2345 mutex_unlock(&dm_thin_pool_table
.mutex
);
2350 static int pool_map(struct dm_target
*ti
, struct bio
*bio
)
2353 struct pool_c
*pt
= ti
->private;
2354 struct pool
*pool
= pt
->pool
;
2355 unsigned long flags
;
2358 * As this is a singleton target, ti->begin is always zero.
2360 spin_lock_irqsave(&pool
->lock
, flags
);
2361 bio
->bi_bdev
= pt
->data_dev
->bdev
;
2362 r
= DM_MAPIO_REMAPPED
;
2363 spin_unlock_irqrestore(&pool
->lock
, flags
);
2368 static int maybe_resize_data_dev(struct dm_target
*ti
, bool *need_commit
)
2371 struct pool_c
*pt
= ti
->private;
2372 struct pool
*pool
= pt
->pool
;
2373 sector_t data_size
= ti
->len
;
2374 dm_block_t sb_data_size
;
2376 *need_commit
= false;
2378 (void) sector_div(data_size
, pool
->sectors_per_block
);
2380 r
= dm_pool_get_data_dev_size(pool
->pmd
, &sb_data_size
);
2382 DMERR("%s: failed to retrieve data device size",
2383 dm_device_name(pool
->pool_md
));
2387 if (data_size
< sb_data_size
) {
2388 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
2389 dm_device_name(pool
->pool_md
),
2390 (unsigned long long)data_size
, sb_data_size
);
2393 } else if (data_size
> sb_data_size
) {
2394 if (dm_pool_metadata_needs_check(pool
->pmd
)) {
2395 DMERR("%s: unable to grow the data device until repaired.",
2396 dm_device_name(pool
->pool_md
));
2401 DMINFO("%s: growing the data device from %llu to %llu blocks",
2402 dm_device_name(pool
->pool_md
),
2403 sb_data_size
, (unsigned long long)data_size
);
2404 r
= dm_pool_resize_data_dev(pool
->pmd
, data_size
);
2406 metadata_operation_failed(pool
, "dm_pool_resize_data_dev", r
);
2410 *need_commit
= true;
2416 static int maybe_resize_metadata_dev(struct dm_target
*ti
, bool *need_commit
)
2419 struct pool_c
*pt
= ti
->private;
2420 struct pool
*pool
= pt
->pool
;
2421 dm_block_t metadata_dev_size
, sb_metadata_dev_size
;
2423 *need_commit
= false;
2425 metadata_dev_size
= get_metadata_dev_size_in_blocks(pool
->md_dev
);
2427 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &sb_metadata_dev_size
);
2429 DMERR("%s: failed to retrieve metadata device size",
2430 dm_device_name(pool
->pool_md
));
2434 if (metadata_dev_size
< sb_metadata_dev_size
) {
2435 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
2436 dm_device_name(pool
->pool_md
),
2437 metadata_dev_size
, sb_metadata_dev_size
);
2440 } else if (metadata_dev_size
> sb_metadata_dev_size
) {
2441 if (dm_pool_metadata_needs_check(pool
->pmd
)) {
2442 DMERR("%s: unable to grow the metadata device until repaired.",
2443 dm_device_name(pool
->pool_md
));
2447 warn_if_metadata_device_too_big(pool
->md_dev
);
2448 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
2449 dm_device_name(pool
->pool_md
),
2450 sb_metadata_dev_size
, metadata_dev_size
);
2451 r
= dm_pool_resize_metadata_dev(pool
->pmd
, metadata_dev_size
);
2453 metadata_operation_failed(pool
, "dm_pool_resize_metadata_dev", r
);
2457 *need_commit
= true;
2464 * Retrieves the number of blocks of the data device from
2465 * the superblock and compares it to the actual device size,
2466 * thus resizing the data device in case it has grown.
2468 * This both copes with opening preallocated data devices in the ctr
2469 * being followed by a resume
2471 * calling the resume method individually after userspace has
2472 * grown the data device in reaction to a table event.
2474 static int pool_preresume(struct dm_target
*ti
)
2477 bool need_commit1
, need_commit2
;
2478 struct pool_c
*pt
= ti
->private;
2479 struct pool
*pool
= pt
->pool
;
2482 * Take control of the pool object.
2484 r
= bind_control_target(pool
, ti
);
2488 r
= maybe_resize_data_dev(ti
, &need_commit1
);
2492 r
= maybe_resize_metadata_dev(ti
, &need_commit2
);
2496 if (need_commit1
|| need_commit2
)
2497 (void) commit(pool
);
2502 static void pool_resume(struct dm_target
*ti
)
2504 struct pool_c
*pt
= ti
->private;
2505 struct pool
*pool
= pt
->pool
;
2506 unsigned long flags
;
2508 spin_lock_irqsave(&pool
->lock
, flags
);
2509 pool
->low_water_triggered
= false;
2510 __requeue_bios(pool
);
2511 spin_unlock_irqrestore(&pool
->lock
, flags
);
2513 do_waker(&pool
->waker
.work
);
2516 static void pool_postsuspend(struct dm_target
*ti
)
2518 struct pool_c
*pt
= ti
->private;
2519 struct pool
*pool
= pt
->pool
;
2521 cancel_delayed_work(&pool
->waker
);
2522 flush_workqueue(pool
->wq
);
2523 (void) commit(pool
);
2526 static int check_arg_count(unsigned argc
, unsigned args_required
)
2528 if (argc
!= args_required
) {
2529 DMWARN("Message received with %u arguments instead of %u.",
2530 argc
, args_required
);
2537 static int read_dev_id(char *arg
, dm_thin_id
*dev_id
, int warning
)
2539 if (!kstrtoull(arg
, 10, (unsigned long long *)dev_id
) &&
2540 *dev_id
<= MAX_DEV_ID
)
2544 DMWARN("Message received with invalid device id: %s", arg
);
2549 static int process_create_thin_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2554 r
= check_arg_count(argc
, 2);
2558 r
= read_dev_id(argv
[1], &dev_id
, 1);
2562 r
= dm_pool_create_thin(pool
->pmd
, dev_id
);
2564 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2572 static int process_create_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2575 dm_thin_id origin_dev_id
;
2578 r
= check_arg_count(argc
, 3);
2582 r
= read_dev_id(argv
[1], &dev_id
, 1);
2586 r
= read_dev_id(argv
[2], &origin_dev_id
, 1);
2590 r
= dm_pool_create_snap(pool
->pmd
, dev_id
, origin_dev_id
);
2592 DMWARN("Creation of new snapshot %s of device %s failed.",
2600 static int process_delete_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2605 r
= check_arg_count(argc
, 2);
2609 r
= read_dev_id(argv
[1], &dev_id
, 1);
2613 r
= dm_pool_delete_thin_device(pool
->pmd
, dev_id
);
2615 DMWARN("Deletion of thin device %s failed.", argv
[1]);
2620 static int process_set_transaction_id_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2622 dm_thin_id old_id
, new_id
;
2625 r
= check_arg_count(argc
, 3);
2629 if (kstrtoull(argv
[1], 10, (unsigned long long *)&old_id
)) {
2630 DMWARN("set_transaction_id message: Unrecognised id %s.", argv
[1]);
2634 if (kstrtoull(argv
[2], 10, (unsigned long long *)&new_id
)) {
2635 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv
[2]);
2639 r
= dm_pool_set_metadata_transaction_id(pool
->pmd
, old_id
, new_id
);
2641 DMWARN("Failed to change transaction id from %s to %s.",
2649 static int process_reserve_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2653 r
= check_arg_count(argc
, 1);
2657 (void) commit(pool
);
2659 r
= dm_pool_reserve_metadata_snap(pool
->pmd
);
2661 DMWARN("reserve_metadata_snap message failed.");
2666 static int process_release_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2670 r
= check_arg_count(argc
, 1);
2674 r
= dm_pool_release_metadata_snap(pool
->pmd
);
2676 DMWARN("release_metadata_snap message failed.");
2682 * Messages supported:
2683 * create_thin <dev_id>
2684 * create_snap <dev_id> <origin_id>
2686 * trim <dev_id> <new_size_in_sectors>
2687 * set_transaction_id <current_trans_id> <new_trans_id>
2688 * reserve_metadata_snap
2689 * release_metadata_snap
2691 static int pool_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
2694 struct pool_c
*pt
= ti
->private;
2695 struct pool
*pool
= pt
->pool
;
2697 if (!strcasecmp(argv
[0], "create_thin"))
2698 r
= process_create_thin_mesg(argc
, argv
, pool
);
2700 else if (!strcasecmp(argv
[0], "create_snap"))
2701 r
= process_create_snap_mesg(argc
, argv
, pool
);
2703 else if (!strcasecmp(argv
[0], "delete"))
2704 r
= process_delete_mesg(argc
, argv
, pool
);
2706 else if (!strcasecmp(argv
[0], "set_transaction_id"))
2707 r
= process_set_transaction_id_mesg(argc
, argv
, pool
);
2709 else if (!strcasecmp(argv
[0], "reserve_metadata_snap"))
2710 r
= process_reserve_metadata_snap_mesg(argc
, argv
, pool
);
2712 else if (!strcasecmp(argv
[0], "release_metadata_snap"))
2713 r
= process_release_metadata_snap_mesg(argc
, argv
, pool
);
2716 DMWARN("Unrecognised thin pool target message received: %s", argv
[0]);
2719 (void) commit(pool
);
2724 static void emit_flags(struct pool_features
*pf
, char *result
,
2725 unsigned sz
, unsigned maxlen
)
2727 unsigned count
= !pf
->zero_new_blocks
+ !pf
->discard_enabled
+
2728 !pf
->discard_passdown
+ (pf
->mode
== PM_READ_ONLY
) +
2729 pf
->error_if_no_space
;
2730 DMEMIT("%u ", count
);
2732 if (!pf
->zero_new_blocks
)
2733 DMEMIT("skip_block_zeroing ");
2735 if (!pf
->discard_enabled
)
2736 DMEMIT("ignore_discard ");
2738 if (!pf
->discard_passdown
)
2739 DMEMIT("no_discard_passdown ");
2741 if (pf
->mode
== PM_READ_ONLY
)
2742 DMEMIT("read_only ");
2744 if (pf
->error_if_no_space
)
2745 DMEMIT("error_if_no_space ");
2750 * <transaction id> <used metadata sectors>/<total metadata sectors>
2751 * <used data sectors>/<total data sectors> <held metadata root>
2753 static void pool_status(struct dm_target
*ti
, status_type_t type
,
2754 unsigned status_flags
, char *result
, unsigned maxlen
)
2758 uint64_t transaction_id
;
2759 dm_block_t nr_free_blocks_data
;
2760 dm_block_t nr_free_blocks_metadata
;
2761 dm_block_t nr_blocks_data
;
2762 dm_block_t nr_blocks_metadata
;
2763 dm_block_t held_root
;
2764 char buf
[BDEVNAME_SIZE
];
2765 char buf2
[BDEVNAME_SIZE
];
2766 struct pool_c
*pt
= ti
->private;
2767 struct pool
*pool
= pt
->pool
;
2770 case STATUSTYPE_INFO
:
2771 if (get_pool_mode(pool
) == PM_FAIL
) {
2776 /* Commit to ensure statistics aren't out-of-date */
2777 if (!(status_flags
& DM_STATUS_NOFLUSH_FLAG
) && !dm_suspended(ti
))
2778 (void) commit(pool
);
2780 r
= dm_pool_get_metadata_transaction_id(pool
->pmd
, &transaction_id
);
2782 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
2783 dm_device_name(pool
->pool_md
), r
);
2787 r
= dm_pool_get_free_metadata_block_count(pool
->pmd
, &nr_free_blocks_metadata
);
2789 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
2790 dm_device_name(pool
->pool_md
), r
);
2794 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &nr_blocks_metadata
);
2796 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
2797 dm_device_name(pool
->pool_md
), r
);
2801 r
= dm_pool_get_free_block_count(pool
->pmd
, &nr_free_blocks_data
);
2803 DMERR("%s: dm_pool_get_free_block_count returned %d",
2804 dm_device_name(pool
->pool_md
), r
);
2808 r
= dm_pool_get_data_dev_size(pool
->pmd
, &nr_blocks_data
);
2810 DMERR("%s: dm_pool_get_data_dev_size returned %d",
2811 dm_device_name(pool
->pool_md
), r
);
2815 r
= dm_pool_get_metadata_snap(pool
->pmd
, &held_root
);
2817 DMERR("%s: dm_pool_get_metadata_snap returned %d",
2818 dm_device_name(pool
->pool_md
), r
);
2822 DMEMIT("%llu %llu/%llu %llu/%llu ",
2823 (unsigned long long)transaction_id
,
2824 (unsigned long long)(nr_blocks_metadata
- nr_free_blocks_metadata
),
2825 (unsigned long long)nr_blocks_metadata
,
2826 (unsigned long long)(nr_blocks_data
- nr_free_blocks_data
),
2827 (unsigned long long)nr_blocks_data
);
2830 DMEMIT("%llu ", held_root
);
2834 if (pool
->pf
.mode
== PM_OUT_OF_DATA_SPACE
)
2835 DMEMIT("out_of_data_space ");
2836 else if (pool
->pf
.mode
== PM_READ_ONLY
)
2841 if (!pool
->pf
.discard_enabled
)
2842 DMEMIT("ignore_discard ");
2843 else if (pool
->pf
.discard_passdown
)
2844 DMEMIT("discard_passdown ");
2846 DMEMIT("no_discard_passdown ");
2848 if (pool
->pf
.error_if_no_space
)
2849 DMEMIT("error_if_no_space ");
2851 DMEMIT("queue_if_no_space ");
2855 case STATUSTYPE_TABLE
:
2856 DMEMIT("%s %s %lu %llu ",
2857 format_dev_t(buf
, pt
->metadata_dev
->bdev
->bd_dev
),
2858 format_dev_t(buf2
, pt
->data_dev
->bdev
->bd_dev
),
2859 (unsigned long)pool
->sectors_per_block
,
2860 (unsigned long long)pt
->low_water_blocks
);
2861 emit_flags(&pt
->requested_pf
, result
, sz
, maxlen
);
2870 static int pool_iterate_devices(struct dm_target
*ti
,
2871 iterate_devices_callout_fn fn
, void *data
)
2873 struct pool_c
*pt
= ti
->private;
2875 return fn(ti
, pt
->data_dev
, 0, ti
->len
, data
);
2878 static int pool_merge(struct dm_target
*ti
, struct bvec_merge_data
*bvm
,
2879 struct bio_vec
*biovec
, int max_size
)
2881 struct pool_c
*pt
= ti
->private;
2882 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
2884 if (!q
->merge_bvec_fn
)
2887 bvm
->bi_bdev
= pt
->data_dev
->bdev
;
2889 return min(max_size
, q
->merge_bvec_fn(q
, bvm
, biovec
));
2892 static void set_discard_limits(struct pool_c
*pt
, struct queue_limits
*limits
)
2894 struct pool
*pool
= pt
->pool
;
2895 struct queue_limits
*data_limits
;
2897 limits
->max_discard_sectors
= pool
->sectors_per_block
;
2900 * discard_granularity is just a hint, and not enforced.
2902 if (pt
->adjusted_pf
.discard_passdown
) {
2903 data_limits
= &bdev_get_queue(pt
->data_dev
->bdev
)->limits
;
2904 limits
->discard_granularity
= data_limits
->discard_granularity
;
2906 limits
->discard_granularity
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
2909 static void pool_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
2911 struct pool_c
*pt
= ti
->private;
2912 struct pool
*pool
= pt
->pool
;
2913 uint64_t io_opt_sectors
= limits
->io_opt
>> SECTOR_SHIFT
;
2916 * If the system-determined stacked limits are compatible with the
2917 * pool's blocksize (io_opt is a factor) do not override them.
2919 if (io_opt_sectors
< pool
->sectors_per_block
||
2920 do_div(io_opt_sectors
, pool
->sectors_per_block
)) {
2921 blk_limits_io_min(limits
, 0);
2922 blk_limits_io_opt(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
2926 * pt->adjusted_pf is a staging area for the actual features to use.
2927 * They get transferred to the live pool in bind_control_target()
2928 * called from pool_preresume().
2930 if (!pt
->adjusted_pf
.discard_enabled
) {
2932 * Must explicitly disallow stacking discard limits otherwise the
2933 * block layer will stack them if pool's data device has support.
2934 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
2935 * user to see that, so make sure to set all discard limits to 0.
2937 limits
->discard_granularity
= 0;
2941 disable_passdown_if_not_supported(pt
);
2943 set_discard_limits(pt
, limits
);
2946 static struct target_type pool_target
= {
2947 .name
= "thin-pool",
2948 .features
= DM_TARGET_SINGLETON
| DM_TARGET_ALWAYS_WRITEABLE
|
2949 DM_TARGET_IMMUTABLE
,
2950 .version
= {1, 11, 0},
2951 .module
= THIS_MODULE
,
2955 .postsuspend
= pool_postsuspend
,
2956 .preresume
= pool_preresume
,
2957 .resume
= pool_resume
,
2958 .message
= pool_message
,
2959 .status
= pool_status
,
2960 .merge
= pool_merge
,
2961 .iterate_devices
= pool_iterate_devices
,
2962 .io_hints
= pool_io_hints
,
2965 /*----------------------------------------------------------------
2966 * Thin target methods
2967 *--------------------------------------------------------------*/
2968 static void thin_dtr(struct dm_target
*ti
)
2970 struct thin_c
*tc
= ti
->private;
2972 mutex_lock(&dm_thin_pool_table
.mutex
);
2974 __pool_dec(tc
->pool
);
2975 dm_pool_close_thin_device(tc
->td
);
2976 dm_put_device(ti
, tc
->pool_dev
);
2978 dm_put_device(ti
, tc
->origin_dev
);
2981 mutex_unlock(&dm_thin_pool_table
.mutex
);
2985 * Thin target parameters:
2987 * <pool_dev> <dev_id> [origin_dev]
2989 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2990 * dev_id: the internal device identifier
2991 * origin_dev: a device external to the pool that should act as the origin
2993 * If the pool device has discards disabled, they get disabled for the thin
2996 static int thin_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
3000 struct dm_dev
*pool_dev
, *origin_dev
;
3001 struct mapped_device
*pool_md
;
3003 mutex_lock(&dm_thin_pool_table
.mutex
);
3005 if (argc
!= 2 && argc
!= 3) {
3006 ti
->error
= "Invalid argument count";
3011 tc
= ti
->private = kzalloc(sizeof(*tc
), GFP_KERNEL
);
3013 ti
->error
= "Out of memory";
3019 r
= dm_get_device(ti
, argv
[2], FMODE_READ
, &origin_dev
);
3021 ti
->error
= "Error opening origin device";
3022 goto bad_origin_dev
;
3024 tc
->origin_dev
= origin_dev
;
3027 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &pool_dev
);
3029 ti
->error
= "Error opening pool device";
3032 tc
->pool_dev
= pool_dev
;
3034 if (read_dev_id(argv
[1], (unsigned long long *)&tc
->dev_id
, 0)) {
3035 ti
->error
= "Invalid device id";
3040 pool_md
= dm_get_md(tc
->pool_dev
->bdev
->bd_dev
);
3042 ti
->error
= "Couldn't get pool mapped device";
3047 tc
->pool
= __pool_table_lookup(pool_md
);
3049 ti
->error
= "Couldn't find pool object";
3051 goto bad_pool_lookup
;
3053 __pool_inc(tc
->pool
);
3055 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
3056 ti
->error
= "Couldn't open thin device, Pool is in fail mode";
3061 r
= dm_pool_open_thin_device(tc
->pool
->pmd
, tc
->dev_id
, &tc
->td
);
3063 ti
->error
= "Couldn't open thin internal device";
3067 r
= dm_set_target_max_io_len(ti
, tc
->pool
->sectors_per_block
);
3069 goto bad_target_max_io_len
;
3071 ti
->num_flush_bios
= 1;
3072 ti
->flush_supported
= true;
3073 ti
->per_bio_data_size
= sizeof(struct dm_thin_endio_hook
);
3075 /* In case the pool supports discards, pass them on. */
3076 ti
->discard_zeroes_data_unsupported
= true;
3077 if (tc
->pool
->pf
.discard_enabled
) {
3078 ti
->discards_supported
= true;
3079 ti
->num_discard_bios
= 1;
3080 /* Discard bios must be split on a block boundary */
3081 ti
->split_discard_bios
= true;
3086 mutex_unlock(&dm_thin_pool_table
.mutex
);
3090 bad_target_max_io_len
:
3091 dm_pool_close_thin_device(tc
->td
);
3093 __pool_dec(tc
->pool
);
3097 dm_put_device(ti
, tc
->pool_dev
);
3100 dm_put_device(ti
, tc
->origin_dev
);
3104 mutex_unlock(&dm_thin_pool_table
.mutex
);
3109 static int thin_map(struct dm_target
*ti
, struct bio
*bio
)
3111 bio
->bi_iter
.bi_sector
= dm_target_offset(ti
, bio
->bi_iter
.bi_sector
);
3113 return thin_bio_map(ti
, bio
);
3116 static int thin_endio(struct dm_target
*ti
, struct bio
*bio
, int err
)
3118 unsigned long flags
;
3119 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
3120 struct list_head work
;
3121 struct dm_thin_new_mapping
*m
, *tmp
;
3122 struct pool
*pool
= h
->tc
->pool
;
3124 if (h
->shared_read_entry
) {
3125 INIT_LIST_HEAD(&work
);
3126 dm_deferred_entry_dec(h
->shared_read_entry
, &work
);
3128 spin_lock_irqsave(&pool
->lock
, flags
);
3129 list_for_each_entry_safe(m
, tmp
, &work
, list
) {
3132 __maybe_add_mapping(m
);
3134 spin_unlock_irqrestore(&pool
->lock
, flags
);
3137 if (h
->all_io_entry
) {
3138 INIT_LIST_HEAD(&work
);
3139 dm_deferred_entry_dec(h
->all_io_entry
, &work
);
3140 if (!list_empty(&work
)) {
3141 spin_lock_irqsave(&pool
->lock
, flags
);
3142 list_for_each_entry_safe(m
, tmp
, &work
, list
)
3143 list_add_tail(&m
->list
, &pool
->prepared_discards
);
3144 spin_unlock_irqrestore(&pool
->lock
, flags
);
3152 static void thin_presuspend(struct dm_target
*ti
)
3154 struct thin_c
*tc
= ti
->private;
3156 if (dm_noflush_suspending(ti
))
3157 noflush_work(tc
, do_noflush_start
);
3160 static void thin_postsuspend(struct dm_target
*ti
)
3162 struct thin_c
*tc
= ti
->private;
3165 * The dm_noflush_suspending flag has been cleared by now, so
3166 * unfortunately we must always run this.
3168 noflush_work(tc
, do_noflush_stop
);
3172 * <nr mapped sectors> <highest mapped sector>
3174 static void thin_status(struct dm_target
*ti
, status_type_t type
,
3175 unsigned status_flags
, char *result
, unsigned maxlen
)
3179 dm_block_t mapped
, highest
;
3180 char buf
[BDEVNAME_SIZE
];
3181 struct thin_c
*tc
= ti
->private;
3183 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
3192 case STATUSTYPE_INFO
:
3193 r
= dm_thin_get_mapped_count(tc
->td
, &mapped
);
3195 DMERR("dm_thin_get_mapped_count returned %d", r
);
3199 r
= dm_thin_get_highest_mapped_block(tc
->td
, &highest
);
3201 DMERR("dm_thin_get_highest_mapped_block returned %d", r
);
3205 DMEMIT("%llu ", mapped
* tc
->pool
->sectors_per_block
);
3207 DMEMIT("%llu", ((highest
+ 1) *
3208 tc
->pool
->sectors_per_block
) - 1);
3213 case STATUSTYPE_TABLE
:
3215 format_dev_t(buf
, tc
->pool_dev
->bdev
->bd_dev
),
3216 (unsigned long) tc
->dev_id
);
3218 DMEMIT(" %s", format_dev_t(buf
, tc
->origin_dev
->bdev
->bd_dev
));
3229 static int thin_iterate_devices(struct dm_target
*ti
,
3230 iterate_devices_callout_fn fn
, void *data
)
3233 struct thin_c
*tc
= ti
->private;
3234 struct pool
*pool
= tc
->pool
;
3237 * We can't call dm_pool_get_data_dev_size() since that blocks. So
3238 * we follow a more convoluted path through to the pool's target.
3241 return 0; /* nothing is bound */
3243 blocks
= pool
->ti
->len
;
3244 (void) sector_div(blocks
, pool
->sectors_per_block
);
3246 return fn(ti
, tc
->pool_dev
, 0, pool
->sectors_per_block
* blocks
, data
);
3251 static struct target_type thin_target
= {
3253 .version
= {1, 11, 0},
3254 .module
= THIS_MODULE
,
3258 .end_io
= thin_endio
,
3259 .presuspend
= thin_presuspend
,
3260 .postsuspend
= thin_postsuspend
,
3261 .status
= thin_status
,
3262 .iterate_devices
= thin_iterate_devices
,
3265 /*----------------------------------------------------------------*/
3267 static int __init
dm_thin_init(void)
3273 r
= dm_register_target(&thin_target
);
3277 r
= dm_register_target(&pool_target
);
3279 goto bad_pool_target
;
3283 _new_mapping_cache
= KMEM_CACHE(dm_thin_new_mapping
, 0);
3284 if (!_new_mapping_cache
)
3285 goto bad_new_mapping_cache
;
3289 bad_new_mapping_cache
:
3290 dm_unregister_target(&pool_target
);
3292 dm_unregister_target(&thin_target
);
3297 static void dm_thin_exit(void)
3299 dm_unregister_target(&thin_target
);
3300 dm_unregister_target(&pool_target
);
3302 kmem_cache_destroy(_new_mapping_cache
);
3305 module_init(dm_thin_init
);
3306 module_exit(dm_thin_exit
);
3308 MODULE_DESCRIPTION(DM_NAME
" thin provisioning target");
3309 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3310 MODULE_LICENSE("GPL");