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;
147 bool error_if_no_space
:1;
151 typedef void (*process_bio_fn
)(struct thin_c
*tc
, struct bio
*bio
);
152 typedef void (*process_mapping_fn
)(struct dm_thin_new_mapping
*m
);
155 struct list_head list
;
156 struct dm_target
*ti
; /* Only set if a pool target is bound */
158 struct mapped_device
*pool_md
;
159 struct block_device
*md_dev
;
160 struct dm_pool_metadata
*pmd
;
162 dm_block_t low_water_blocks
;
163 uint32_t sectors_per_block
;
164 int sectors_per_block_shift
;
166 struct pool_features pf
;
167 bool low_water_triggered
:1; /* A dm event has been sent */
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 out_of_data_space(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
;
231 /*----------------------------------------------------------------*/
234 * wake_worker() is used when new work is queued and when pool_resume is
235 * ready to continue deferred IO processing.
237 static void wake_worker(struct pool
*pool
)
239 queue_work(pool
->wq
, &pool
->worker
);
242 /*----------------------------------------------------------------*/
244 static int bio_detain(struct pool
*pool
, struct dm_cell_key
*key
, struct bio
*bio
,
245 struct dm_bio_prison_cell
**cell_result
)
248 struct dm_bio_prison_cell
*cell_prealloc
;
251 * Allocate a cell from the prison's mempool.
252 * This might block but it can't fail.
254 cell_prealloc
= dm_bio_prison_alloc_cell(pool
->prison
, GFP_NOIO
);
256 r
= dm_bio_detain(pool
->prison
, key
, bio
, cell_prealloc
, cell_result
);
259 * We reused an old cell; we can get rid of
262 dm_bio_prison_free_cell(pool
->prison
, cell_prealloc
);
267 static void cell_release(struct pool
*pool
,
268 struct dm_bio_prison_cell
*cell
,
269 struct bio_list
*bios
)
271 dm_cell_release(pool
->prison
, cell
, bios
);
272 dm_bio_prison_free_cell(pool
->prison
, cell
);
275 static void cell_release_no_holder(struct pool
*pool
,
276 struct dm_bio_prison_cell
*cell
,
277 struct bio_list
*bios
)
279 dm_cell_release_no_holder(pool
->prison
, cell
, bios
);
280 dm_bio_prison_free_cell(pool
->prison
, cell
);
283 static void cell_defer_no_holder_no_free(struct thin_c
*tc
,
284 struct dm_bio_prison_cell
*cell
)
286 struct pool
*pool
= tc
->pool
;
289 spin_lock_irqsave(&pool
->lock
, flags
);
290 dm_cell_release_no_holder(pool
->prison
, cell
, &pool
->deferred_bios
);
291 spin_unlock_irqrestore(&pool
->lock
, flags
);
296 static void cell_error(struct pool
*pool
,
297 struct dm_bio_prison_cell
*cell
)
299 dm_cell_error(pool
->prison
, cell
);
300 dm_bio_prison_free_cell(pool
->prison
, cell
);
303 /*----------------------------------------------------------------*/
306 * A global list of pools that uses a struct mapped_device as a key.
308 static struct dm_thin_pool_table
{
310 struct list_head pools
;
311 } dm_thin_pool_table
;
313 static void pool_table_init(void)
315 mutex_init(&dm_thin_pool_table
.mutex
);
316 INIT_LIST_HEAD(&dm_thin_pool_table
.pools
);
319 static void __pool_table_insert(struct pool
*pool
)
321 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
322 list_add(&pool
->list
, &dm_thin_pool_table
.pools
);
325 static void __pool_table_remove(struct pool
*pool
)
327 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
328 list_del(&pool
->list
);
331 static struct pool
*__pool_table_lookup(struct mapped_device
*md
)
333 struct pool
*pool
= NULL
, *tmp
;
335 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
337 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
338 if (tmp
->pool_md
== md
) {
347 static struct pool
*__pool_table_lookup_metadata_dev(struct block_device
*md_dev
)
349 struct pool
*pool
= NULL
, *tmp
;
351 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
353 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
354 if (tmp
->md_dev
== md_dev
) {
363 /*----------------------------------------------------------------*/
365 struct dm_thin_endio_hook
{
367 struct dm_deferred_entry
*shared_read_entry
;
368 struct dm_deferred_entry
*all_io_entry
;
369 struct dm_thin_new_mapping
*overwrite_mapping
;
372 static void __requeue_bio_list(struct thin_c
*tc
, struct bio_list
*master
)
375 struct bio_list bios
;
377 bio_list_init(&bios
);
378 bio_list_merge(&bios
, master
);
379 bio_list_init(master
);
381 while ((bio
= bio_list_pop(&bios
))) {
382 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
385 bio_endio(bio
, DM_ENDIO_REQUEUE
);
387 bio_list_add(master
, bio
);
391 static void requeue_io(struct thin_c
*tc
)
393 struct pool
*pool
= tc
->pool
;
396 spin_lock_irqsave(&pool
->lock
, flags
);
397 __requeue_bio_list(tc
, &pool
->deferred_bios
);
398 __requeue_bio_list(tc
, &pool
->retry_on_resume_list
);
399 spin_unlock_irqrestore(&pool
->lock
, flags
);
403 * This section of code contains the logic for processing a thin device's IO.
404 * Much of the code depends on pool object resources (lists, workqueues, etc)
405 * but most is exclusively called from the thin target rather than the thin-pool
409 static bool block_size_is_power_of_two(struct pool
*pool
)
411 return pool
->sectors_per_block_shift
>= 0;
414 static dm_block_t
get_bio_block(struct thin_c
*tc
, struct bio
*bio
)
416 struct pool
*pool
= tc
->pool
;
417 sector_t block_nr
= bio
->bi_iter
.bi_sector
;
419 if (block_size_is_power_of_two(pool
))
420 block_nr
>>= pool
->sectors_per_block_shift
;
422 (void) sector_div(block_nr
, pool
->sectors_per_block
);
427 static void remap(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
)
429 struct pool
*pool
= tc
->pool
;
430 sector_t bi_sector
= bio
->bi_iter
.bi_sector
;
432 bio
->bi_bdev
= tc
->pool_dev
->bdev
;
433 if (block_size_is_power_of_two(pool
))
434 bio
->bi_iter
.bi_sector
=
435 (block
<< pool
->sectors_per_block_shift
) |
436 (bi_sector
& (pool
->sectors_per_block
- 1));
438 bio
->bi_iter
.bi_sector
= (block
* pool
->sectors_per_block
) +
439 sector_div(bi_sector
, pool
->sectors_per_block
);
442 static void remap_to_origin(struct thin_c
*tc
, struct bio
*bio
)
444 bio
->bi_bdev
= tc
->origin_dev
->bdev
;
447 static int bio_triggers_commit(struct thin_c
*tc
, struct bio
*bio
)
449 return (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) &&
450 dm_thin_changed_this_transaction(tc
->td
);
453 static void inc_all_io_entry(struct pool
*pool
, struct bio
*bio
)
455 struct dm_thin_endio_hook
*h
;
457 if (bio
->bi_rw
& REQ_DISCARD
)
460 h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
461 h
->all_io_entry
= dm_deferred_entry_inc(pool
->all_io_ds
);
464 static void issue(struct thin_c
*tc
, struct bio
*bio
)
466 struct pool
*pool
= tc
->pool
;
469 if (!bio_triggers_commit(tc
, bio
)) {
470 generic_make_request(bio
);
475 * Complete bio with an error if earlier I/O caused changes to
476 * the metadata that can't be committed e.g, due to I/O errors
477 * on the metadata device.
479 if (dm_thin_aborted_changes(tc
->td
)) {
485 * Batch together any bios that trigger commits and then issue a
486 * single commit for them in process_deferred_bios().
488 spin_lock_irqsave(&pool
->lock
, flags
);
489 bio_list_add(&pool
->deferred_flush_bios
, bio
);
490 spin_unlock_irqrestore(&pool
->lock
, flags
);
493 static void remap_to_origin_and_issue(struct thin_c
*tc
, struct bio
*bio
)
495 remap_to_origin(tc
, bio
);
499 static void remap_and_issue(struct thin_c
*tc
, struct bio
*bio
,
502 remap(tc
, bio
, block
);
506 /*----------------------------------------------------------------*/
509 * Bio endio functions.
511 struct dm_thin_new_mapping
{
512 struct list_head list
;
517 bool definitely_not_shared
:1;
521 dm_block_t virt_block
;
522 dm_block_t data_block
;
523 struct dm_bio_prison_cell
*cell
, *cell2
;
526 * If the bio covers the whole area of a block then we can avoid
527 * zeroing or copying. Instead this bio is hooked. The bio will
528 * still be in the cell, so care has to be taken to avoid issuing
532 bio_end_io_t
*saved_bi_end_io
;
535 static void __maybe_add_mapping(struct dm_thin_new_mapping
*m
)
537 struct pool
*pool
= m
->tc
->pool
;
539 if (m
->quiesced
&& m
->prepared
) {
540 list_add_tail(&m
->list
, &pool
->prepared_mappings
);
545 static void copy_complete(int read_err
, unsigned long write_err
, void *context
)
548 struct dm_thin_new_mapping
*m
= context
;
549 struct pool
*pool
= m
->tc
->pool
;
551 m
->err
= read_err
|| write_err
? -EIO
: 0;
553 spin_lock_irqsave(&pool
->lock
, flags
);
555 __maybe_add_mapping(m
);
556 spin_unlock_irqrestore(&pool
->lock
, flags
);
559 static void overwrite_endio(struct bio
*bio
, int err
)
562 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
563 struct dm_thin_new_mapping
*m
= h
->overwrite_mapping
;
564 struct pool
*pool
= m
->tc
->pool
;
568 spin_lock_irqsave(&pool
->lock
, flags
);
570 __maybe_add_mapping(m
);
571 spin_unlock_irqrestore(&pool
->lock
, flags
);
574 /*----------------------------------------------------------------*/
581 * Prepared mapping jobs.
585 * This sends the bios in the cell back to the deferred_bios list.
587 static void cell_defer(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
589 struct pool
*pool
= tc
->pool
;
592 spin_lock_irqsave(&pool
->lock
, flags
);
593 cell_release(pool
, cell
, &pool
->deferred_bios
);
594 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
600 * Same as cell_defer above, except it omits the original holder of the cell.
602 static void cell_defer_no_holder(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
604 struct pool
*pool
= tc
->pool
;
607 spin_lock_irqsave(&pool
->lock
, flags
);
608 cell_release_no_holder(pool
, cell
, &pool
->deferred_bios
);
609 spin_unlock_irqrestore(&pool
->lock
, flags
);
614 static void process_prepared_mapping_fail(struct dm_thin_new_mapping
*m
)
617 m
->bio
->bi_end_io
= m
->saved_bi_end_io
;
618 atomic_inc(&m
->bio
->bi_remaining
);
620 cell_error(m
->tc
->pool
, m
->cell
);
622 mempool_free(m
, m
->tc
->pool
->mapping_pool
);
625 static void process_prepared_mapping(struct dm_thin_new_mapping
*m
)
627 struct thin_c
*tc
= m
->tc
;
628 struct pool
*pool
= tc
->pool
;
634 bio
->bi_end_io
= m
->saved_bi_end_io
;
635 atomic_inc(&bio
->bi_remaining
);
639 cell_error(pool
, m
->cell
);
644 * Commit the prepared block into the mapping btree.
645 * Any I/O for this block arriving after this point will get
646 * remapped to it directly.
648 r
= dm_thin_insert_block(tc
->td
, m
->virt_block
, m
->data_block
);
650 metadata_operation_failed(pool
, "dm_thin_insert_block", r
);
651 cell_error(pool
, m
->cell
);
656 * Release any bios held while the block was being provisioned.
657 * If we are processing a write bio that completely covers the block,
658 * we already processed it so can ignore it now when processing
659 * the bios in the cell.
662 cell_defer_no_holder(tc
, m
->cell
);
665 cell_defer(tc
, m
->cell
);
669 mempool_free(m
, pool
->mapping_pool
);
672 static void process_prepared_discard_fail(struct dm_thin_new_mapping
*m
)
674 struct thin_c
*tc
= m
->tc
;
676 bio_io_error(m
->bio
);
677 cell_defer_no_holder(tc
, m
->cell
);
678 cell_defer_no_holder(tc
, m
->cell2
);
679 mempool_free(m
, tc
->pool
->mapping_pool
);
682 static void process_prepared_discard_passdown(struct dm_thin_new_mapping
*m
)
684 struct thin_c
*tc
= m
->tc
;
686 inc_all_io_entry(tc
->pool
, m
->bio
);
687 cell_defer_no_holder(tc
, m
->cell
);
688 cell_defer_no_holder(tc
, m
->cell2
);
691 if (m
->definitely_not_shared
)
692 remap_and_issue(tc
, m
->bio
, m
->data_block
);
695 if (dm_pool_block_is_used(tc
->pool
->pmd
, m
->data_block
, &used
) || used
)
696 bio_endio(m
->bio
, 0);
698 remap_and_issue(tc
, m
->bio
, m
->data_block
);
701 bio_endio(m
->bio
, 0);
703 mempool_free(m
, tc
->pool
->mapping_pool
);
706 static void process_prepared_discard(struct dm_thin_new_mapping
*m
)
709 struct thin_c
*tc
= m
->tc
;
711 r
= dm_thin_remove_block(tc
->td
, m
->virt_block
);
713 DMERR_LIMIT("dm_thin_remove_block() failed");
715 process_prepared_discard_passdown(m
);
718 static void process_prepared(struct pool
*pool
, struct list_head
*head
,
719 process_mapping_fn
*fn
)
722 struct list_head maps
;
723 struct dm_thin_new_mapping
*m
, *tmp
;
725 INIT_LIST_HEAD(&maps
);
726 spin_lock_irqsave(&pool
->lock
, flags
);
727 list_splice_init(head
, &maps
);
728 spin_unlock_irqrestore(&pool
->lock
, flags
);
730 list_for_each_entry_safe(m
, tmp
, &maps
, list
)
737 static int io_overlaps_block(struct pool
*pool
, struct bio
*bio
)
739 return bio
->bi_iter
.bi_size
==
740 (pool
->sectors_per_block
<< SECTOR_SHIFT
);
743 static int io_overwrites_block(struct pool
*pool
, struct bio
*bio
)
745 return (bio_data_dir(bio
) == WRITE
) &&
746 io_overlaps_block(pool
, bio
);
749 static void save_and_set_endio(struct bio
*bio
, bio_end_io_t
**save
,
752 *save
= bio
->bi_end_io
;
756 static int ensure_next_mapping(struct pool
*pool
)
758 if (pool
->next_mapping
)
761 pool
->next_mapping
= mempool_alloc(pool
->mapping_pool
, GFP_ATOMIC
);
763 return pool
->next_mapping
? 0 : -ENOMEM
;
766 static struct dm_thin_new_mapping
*get_next_mapping(struct pool
*pool
)
768 struct dm_thin_new_mapping
*m
= pool
->next_mapping
;
770 BUG_ON(!pool
->next_mapping
);
772 memset(m
, 0, sizeof(struct dm_thin_new_mapping
));
773 INIT_LIST_HEAD(&m
->list
);
776 pool
->next_mapping
= NULL
;
781 static void schedule_copy(struct thin_c
*tc
, dm_block_t virt_block
,
782 struct dm_dev
*origin
, dm_block_t data_origin
,
783 dm_block_t data_dest
,
784 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
787 struct pool
*pool
= tc
->pool
;
788 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
791 m
->virt_block
= virt_block
;
792 m
->data_block
= data_dest
;
795 if (!dm_deferred_set_add_work(pool
->shared_read_ds
, &m
->list
))
799 * IO to pool_dev remaps to the pool target's data_dev.
801 * If the whole block of data is being overwritten, we can issue the
802 * bio immediately. Otherwise we use kcopyd to clone the data first.
804 if (io_overwrites_block(pool
, bio
)) {
805 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
807 h
->overwrite_mapping
= m
;
809 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
810 inc_all_io_entry(pool
, bio
);
811 remap_and_issue(tc
, bio
, data_dest
);
813 struct dm_io_region from
, to
;
815 from
.bdev
= origin
->bdev
;
816 from
.sector
= data_origin
* pool
->sectors_per_block
;
817 from
.count
= pool
->sectors_per_block
;
819 to
.bdev
= tc
->pool_dev
->bdev
;
820 to
.sector
= data_dest
* pool
->sectors_per_block
;
821 to
.count
= pool
->sectors_per_block
;
823 r
= dm_kcopyd_copy(pool
->copier
, &from
, 1, &to
,
824 0, copy_complete
, m
);
826 mempool_free(m
, pool
->mapping_pool
);
827 DMERR_LIMIT("dm_kcopyd_copy() failed");
828 cell_error(pool
, cell
);
833 static void schedule_internal_copy(struct thin_c
*tc
, dm_block_t virt_block
,
834 dm_block_t data_origin
, dm_block_t data_dest
,
835 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
837 schedule_copy(tc
, virt_block
, tc
->pool_dev
,
838 data_origin
, data_dest
, cell
, bio
);
841 static void schedule_external_copy(struct thin_c
*tc
, dm_block_t virt_block
,
842 dm_block_t data_dest
,
843 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
845 schedule_copy(tc
, virt_block
, tc
->origin_dev
,
846 virt_block
, data_dest
, cell
, bio
);
849 static void schedule_zero(struct thin_c
*tc
, dm_block_t virt_block
,
850 dm_block_t data_block
, struct dm_bio_prison_cell
*cell
,
853 struct pool
*pool
= tc
->pool
;
854 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
859 m
->virt_block
= virt_block
;
860 m
->data_block
= data_block
;
864 * If the whole block of data is being overwritten or we are not
865 * zeroing pre-existing data, we can issue the bio immediately.
866 * Otherwise we use kcopyd to zero the data first.
868 if (!pool
->pf
.zero_new_blocks
)
869 process_prepared_mapping(m
);
871 else if (io_overwrites_block(pool
, bio
)) {
872 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
874 h
->overwrite_mapping
= m
;
876 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
877 inc_all_io_entry(pool
, bio
);
878 remap_and_issue(tc
, bio
, data_block
);
881 struct dm_io_region to
;
883 to
.bdev
= tc
->pool_dev
->bdev
;
884 to
.sector
= data_block
* pool
->sectors_per_block
;
885 to
.count
= pool
->sectors_per_block
;
887 r
= dm_kcopyd_zero(pool
->copier
, 1, &to
, 0, copy_complete
, m
);
889 mempool_free(m
, pool
->mapping_pool
);
890 DMERR_LIMIT("dm_kcopyd_zero() failed");
891 cell_error(pool
, cell
);
897 * A non-zero return indicates read_only or fail_io mode.
898 * Many callers don't care about the return value.
900 static int commit(struct pool
*pool
)
904 if (get_pool_mode(pool
) != PM_WRITE
)
907 r
= dm_pool_commit_metadata(pool
->pmd
);
909 metadata_operation_failed(pool
, "dm_pool_commit_metadata", r
);
914 static void check_low_water_mark(struct pool
*pool
, dm_block_t free_blocks
)
918 if (free_blocks
<= pool
->low_water_blocks
&& !pool
->low_water_triggered
) {
919 DMWARN("%s: reached low water mark for data device: sending event.",
920 dm_device_name(pool
->pool_md
));
921 spin_lock_irqsave(&pool
->lock
, flags
);
922 pool
->low_water_triggered
= true;
923 spin_unlock_irqrestore(&pool
->lock
, flags
);
924 dm_table_event(pool
->ti
->table
);
928 static int alloc_data_block(struct thin_c
*tc
, dm_block_t
*result
)
931 dm_block_t free_blocks
;
932 struct pool
*pool
= tc
->pool
;
934 if (get_pool_mode(pool
) != PM_WRITE
)
937 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
939 metadata_operation_failed(pool
, "dm_pool_get_free_block_count", r
);
943 check_low_water_mark(pool
, free_blocks
);
947 * Try to commit to see if that will free up some
954 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
956 metadata_operation_failed(pool
, "dm_pool_get_free_block_count", r
);
961 out_of_data_space(pool
);
966 r
= dm_pool_alloc_data_block(pool
->pmd
, result
);
968 metadata_operation_failed(pool
, "dm_pool_alloc_data_block", r
);
976 * If we have run out of space, queue bios until the device is
977 * resumed, presumably after having been reloaded with more space.
979 static void retry_on_resume(struct bio
*bio
)
981 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
982 struct thin_c
*tc
= h
->tc
;
983 struct pool
*pool
= tc
->pool
;
986 spin_lock_irqsave(&pool
->lock
, flags
);
987 bio_list_add(&pool
->retry_on_resume_list
, bio
);
988 spin_unlock_irqrestore(&pool
->lock
, flags
);
991 static void handle_unserviceable_bio(struct pool
*pool
, struct bio
*bio
)
994 * When pool is read-only, no cell locking is needed because
995 * nothing is changing.
997 WARN_ON_ONCE(get_pool_mode(pool
) != PM_READ_ONLY
);
999 if (pool
->pf
.error_if_no_space
)
1002 retry_on_resume(bio
);
1005 static void retry_bios_on_resume(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
1008 struct bio_list bios
;
1010 bio_list_init(&bios
);
1011 cell_release(pool
, cell
, &bios
);
1013 while ((bio
= bio_list_pop(&bios
)))
1014 handle_unserviceable_bio(pool
, bio
);
1017 static void process_discard(struct thin_c
*tc
, struct bio
*bio
)
1020 unsigned long flags
;
1021 struct pool
*pool
= tc
->pool
;
1022 struct dm_bio_prison_cell
*cell
, *cell2
;
1023 struct dm_cell_key key
, key2
;
1024 dm_block_t block
= get_bio_block(tc
, bio
);
1025 struct dm_thin_lookup_result lookup_result
;
1026 struct dm_thin_new_mapping
*m
;
1028 build_virtual_key(tc
->td
, block
, &key
);
1029 if (bio_detain(tc
->pool
, &key
, bio
, &cell
))
1032 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1036 * Check nobody is fiddling with this pool block. This can
1037 * happen if someone's in the process of breaking sharing
1040 build_data_key(tc
->td
, lookup_result
.block
, &key2
);
1041 if (bio_detain(tc
->pool
, &key2
, bio
, &cell2
)) {
1042 cell_defer_no_holder(tc
, cell
);
1046 if (io_overlaps_block(pool
, bio
)) {
1048 * IO may still be going to the destination block. We must
1049 * quiesce before we can do the removal.
1051 m
= get_next_mapping(pool
);
1053 m
->pass_discard
= pool
->pf
.discard_passdown
;
1054 m
->definitely_not_shared
= !lookup_result
.shared
;
1055 m
->virt_block
= block
;
1056 m
->data_block
= lookup_result
.block
;
1061 if (!dm_deferred_set_add_work(pool
->all_io_ds
, &m
->list
)) {
1062 spin_lock_irqsave(&pool
->lock
, flags
);
1063 list_add_tail(&m
->list
, &pool
->prepared_discards
);
1064 spin_unlock_irqrestore(&pool
->lock
, flags
);
1068 inc_all_io_entry(pool
, bio
);
1069 cell_defer_no_holder(tc
, cell
);
1070 cell_defer_no_holder(tc
, cell2
);
1073 * The DM core makes sure that the discard doesn't span
1074 * a block boundary. So we submit the discard of a
1075 * partial block appropriately.
1077 if ((!lookup_result
.shared
) && pool
->pf
.discard_passdown
)
1078 remap_and_issue(tc
, bio
, lookup_result
.block
);
1086 * It isn't provisioned, just forget it.
1088 cell_defer_no_holder(tc
, cell
);
1093 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1095 cell_defer_no_holder(tc
, cell
);
1101 static void break_sharing(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1102 struct dm_cell_key
*key
,
1103 struct dm_thin_lookup_result
*lookup_result
,
1104 struct dm_bio_prison_cell
*cell
)
1107 dm_block_t data_block
;
1108 struct pool
*pool
= tc
->pool
;
1110 r
= alloc_data_block(tc
, &data_block
);
1113 schedule_internal_copy(tc
, block
, lookup_result
->block
,
1114 data_block
, cell
, bio
);
1118 retry_bios_on_resume(pool
, cell
);
1122 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1124 cell_error(pool
, cell
);
1129 static void process_shared_bio(struct thin_c
*tc
, struct bio
*bio
,
1131 struct dm_thin_lookup_result
*lookup_result
)
1133 struct dm_bio_prison_cell
*cell
;
1134 struct pool
*pool
= tc
->pool
;
1135 struct dm_cell_key key
;
1138 * If cell is already occupied, then sharing is already in the process
1139 * of being broken so we have nothing further to do here.
1141 build_data_key(tc
->td
, lookup_result
->block
, &key
);
1142 if (bio_detain(pool
, &key
, bio
, &cell
))
1145 if (bio_data_dir(bio
) == WRITE
&& bio
->bi_iter
.bi_size
)
1146 break_sharing(tc
, bio
, block
, &key
, lookup_result
, cell
);
1148 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1150 h
->shared_read_entry
= dm_deferred_entry_inc(pool
->shared_read_ds
);
1151 inc_all_io_entry(pool
, bio
);
1152 cell_defer_no_holder(tc
, cell
);
1154 remap_and_issue(tc
, bio
, lookup_result
->block
);
1158 static void provision_block(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1159 struct dm_bio_prison_cell
*cell
)
1162 dm_block_t data_block
;
1163 struct pool
*pool
= tc
->pool
;
1166 * Remap empty bios (flushes) immediately, without provisioning.
1168 if (!bio
->bi_iter
.bi_size
) {
1169 inc_all_io_entry(pool
, bio
);
1170 cell_defer_no_holder(tc
, cell
);
1172 remap_and_issue(tc
, bio
, 0);
1177 * Fill read bios with zeroes and complete them immediately.
1179 if (bio_data_dir(bio
) == READ
) {
1181 cell_defer_no_holder(tc
, cell
);
1186 r
= alloc_data_block(tc
, &data_block
);
1190 schedule_external_copy(tc
, block
, data_block
, cell
, bio
);
1192 schedule_zero(tc
, block
, data_block
, cell
, bio
);
1196 retry_bios_on_resume(pool
, cell
);
1200 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1202 cell_error(pool
, cell
);
1207 static void process_bio(struct thin_c
*tc
, struct bio
*bio
)
1210 struct pool
*pool
= tc
->pool
;
1211 dm_block_t block
= get_bio_block(tc
, bio
);
1212 struct dm_bio_prison_cell
*cell
;
1213 struct dm_cell_key key
;
1214 struct dm_thin_lookup_result lookup_result
;
1217 * If cell is already occupied, then the block is already
1218 * being provisioned so we have nothing further to do here.
1220 build_virtual_key(tc
->td
, block
, &key
);
1221 if (bio_detain(pool
, &key
, bio
, &cell
))
1224 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1227 if (lookup_result
.shared
) {
1228 process_shared_bio(tc
, bio
, block
, &lookup_result
);
1229 cell_defer_no_holder(tc
, cell
); /* FIXME: pass this cell into process_shared? */
1231 inc_all_io_entry(pool
, bio
);
1232 cell_defer_no_holder(tc
, cell
);
1234 remap_and_issue(tc
, bio
, lookup_result
.block
);
1239 if (bio_data_dir(bio
) == READ
&& tc
->origin_dev
) {
1240 inc_all_io_entry(pool
, bio
);
1241 cell_defer_no_holder(tc
, cell
);
1243 remap_to_origin_and_issue(tc
, bio
);
1245 provision_block(tc
, bio
, block
, cell
);
1249 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1251 cell_defer_no_holder(tc
, cell
);
1257 static void process_bio_read_only(struct thin_c
*tc
, struct bio
*bio
)
1260 int rw
= bio_data_dir(bio
);
1261 dm_block_t block
= get_bio_block(tc
, bio
);
1262 struct dm_thin_lookup_result lookup_result
;
1264 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1267 if (lookup_result
.shared
&& (rw
== WRITE
) && bio
->bi_iter
.bi_size
)
1268 handle_unserviceable_bio(tc
->pool
, bio
);
1270 inc_all_io_entry(tc
->pool
, bio
);
1271 remap_and_issue(tc
, bio
, lookup_result
.block
);
1277 handle_unserviceable_bio(tc
->pool
, bio
);
1281 if (tc
->origin_dev
) {
1282 inc_all_io_entry(tc
->pool
, bio
);
1283 remap_to_origin_and_issue(tc
, bio
);
1292 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1299 static void process_bio_fail(struct thin_c
*tc
, struct bio
*bio
)
1305 * FIXME: should we also commit due to size of transaction, measured in
1308 static int need_commit_due_to_time(struct pool
*pool
)
1310 return jiffies
< pool
->last_commit_jiffies
||
1311 jiffies
> pool
->last_commit_jiffies
+ COMMIT_PERIOD
;
1314 static void process_deferred_bios(struct pool
*pool
)
1316 unsigned long flags
;
1318 struct bio_list bios
;
1320 bio_list_init(&bios
);
1322 spin_lock_irqsave(&pool
->lock
, flags
);
1323 bio_list_merge(&bios
, &pool
->deferred_bios
);
1324 bio_list_init(&pool
->deferred_bios
);
1325 spin_unlock_irqrestore(&pool
->lock
, flags
);
1327 while ((bio
= bio_list_pop(&bios
))) {
1328 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1329 struct thin_c
*tc
= h
->tc
;
1332 * If we've got no free new_mapping structs, and processing
1333 * this bio might require one, we pause until there are some
1334 * prepared mappings to process.
1336 if (ensure_next_mapping(pool
)) {
1337 spin_lock_irqsave(&pool
->lock
, flags
);
1338 bio_list_merge(&pool
->deferred_bios
, &bios
);
1339 spin_unlock_irqrestore(&pool
->lock
, flags
);
1344 if (bio
->bi_rw
& REQ_DISCARD
)
1345 pool
->process_discard(tc
, bio
);
1347 pool
->process_bio(tc
, bio
);
1351 * If there are any deferred flush bios, we must commit
1352 * the metadata before issuing them.
1354 bio_list_init(&bios
);
1355 spin_lock_irqsave(&pool
->lock
, flags
);
1356 bio_list_merge(&bios
, &pool
->deferred_flush_bios
);
1357 bio_list_init(&pool
->deferred_flush_bios
);
1358 spin_unlock_irqrestore(&pool
->lock
, flags
);
1360 if (bio_list_empty(&bios
) &&
1361 !(dm_pool_changed_this_transaction(pool
->pmd
) && need_commit_due_to_time(pool
)))
1365 while ((bio
= bio_list_pop(&bios
)))
1369 pool
->last_commit_jiffies
= jiffies
;
1371 while ((bio
= bio_list_pop(&bios
)))
1372 generic_make_request(bio
);
1375 static void do_worker(struct work_struct
*ws
)
1377 struct pool
*pool
= container_of(ws
, struct pool
, worker
);
1379 process_prepared(pool
, &pool
->prepared_mappings
, &pool
->process_prepared_mapping
);
1380 process_prepared(pool
, &pool
->prepared_discards
, &pool
->process_prepared_discard
);
1381 process_deferred_bios(pool
);
1385 * We want to commit periodically so that not too much
1386 * unwritten data builds up.
1388 static void do_waker(struct work_struct
*ws
)
1390 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
, waker
);
1392 queue_delayed_work(pool
->wq
, &pool
->waker
, COMMIT_PERIOD
);
1395 /*----------------------------------------------------------------*/
1397 static enum pool_mode
get_pool_mode(struct pool
*pool
)
1399 return pool
->pf
.mode
;
1402 static void set_pool_mode(struct pool
*pool
, enum pool_mode new_mode
)
1405 struct pool_c
*pt
= pool
->ti
->private;
1406 enum pool_mode old_mode
= pool
->pf
.mode
;
1410 if (old_mode
!= new_mode
)
1411 DMERR("%s: switching pool to failure mode",
1412 dm_device_name(pool
->pool_md
));
1413 dm_pool_metadata_read_only(pool
->pmd
);
1414 pool
->process_bio
= process_bio_fail
;
1415 pool
->process_discard
= process_bio_fail
;
1416 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1417 pool
->process_prepared_discard
= process_prepared_discard_fail
;
1421 if (old_mode
!= new_mode
)
1422 DMERR("%s: switching pool to read-only mode",
1423 dm_device_name(pool
->pool_md
));
1424 r
= dm_pool_abort_metadata(pool
->pmd
);
1426 DMERR("%s: aborting transaction failed",
1427 dm_device_name(pool
->pool_md
));
1429 set_pool_mode(pool
, new_mode
);
1431 dm_pool_metadata_read_only(pool
->pmd
);
1432 pool
->process_bio
= process_bio_read_only
;
1433 pool
->process_discard
= process_discard
;
1434 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1435 pool
->process_prepared_discard
= process_prepared_discard_passdown
;
1440 if (old_mode
!= new_mode
)
1441 DMINFO("%s: switching pool to write mode",
1442 dm_device_name(pool
->pool_md
));
1443 dm_pool_metadata_read_write(pool
->pmd
);
1444 pool
->process_bio
= process_bio
;
1445 pool
->process_discard
= process_discard
;
1446 pool
->process_prepared_mapping
= process_prepared_mapping
;
1447 pool
->process_prepared_discard
= process_prepared_discard
;
1451 pool
->pf
.mode
= new_mode
;
1453 * The pool mode may have changed, sync it so bind_control_target()
1454 * doesn't cause an unexpected mode transition on resume.
1456 pt
->adjusted_pf
.mode
= new_mode
;
1460 * Rather than calling set_pool_mode directly, use these which describe the
1461 * reason for mode degradation.
1463 static void out_of_data_space(struct pool
*pool
)
1465 DMERR_LIMIT("%s: no free data space available.",
1466 dm_device_name(pool
->pool_md
));
1467 set_pool_mode(pool
, PM_READ_ONLY
);
1470 static void metadata_operation_failed(struct pool
*pool
, const char *op
, int r
)
1472 dm_block_t free_blocks
;
1474 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
1475 dm_device_name(pool
->pool_md
), op
, r
);
1478 !dm_pool_get_free_metadata_block_count(pool
->pmd
, &free_blocks
) &&
1480 DMERR_LIMIT("%s: no free metadata space available.",
1481 dm_device_name(pool
->pool_md
));
1483 set_pool_mode(pool
, PM_READ_ONLY
);
1486 /*----------------------------------------------------------------*/
1489 * Mapping functions.
1493 * Called only while mapping a thin bio to hand it over to the workqueue.
1495 static void thin_defer_bio(struct thin_c
*tc
, struct bio
*bio
)
1497 unsigned long flags
;
1498 struct pool
*pool
= tc
->pool
;
1500 spin_lock_irqsave(&pool
->lock
, flags
);
1501 bio_list_add(&pool
->deferred_bios
, bio
);
1502 spin_unlock_irqrestore(&pool
->lock
, flags
);
1507 static void thin_hook_bio(struct thin_c
*tc
, struct bio
*bio
)
1509 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1512 h
->shared_read_entry
= NULL
;
1513 h
->all_io_entry
= NULL
;
1514 h
->overwrite_mapping
= NULL
;
1518 * Non-blocking function called from the thin target's map function.
1520 static int thin_bio_map(struct dm_target
*ti
, struct bio
*bio
)
1523 struct thin_c
*tc
= ti
->private;
1524 dm_block_t block
= get_bio_block(tc
, bio
);
1525 struct dm_thin_device
*td
= tc
->td
;
1526 struct dm_thin_lookup_result result
;
1527 struct dm_bio_prison_cell cell1
, cell2
;
1528 struct dm_bio_prison_cell
*cell_result
;
1529 struct dm_cell_key key
;
1531 thin_hook_bio(tc
, bio
);
1533 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
1535 return DM_MAPIO_SUBMITTED
;
1538 if (bio
->bi_rw
& (REQ_DISCARD
| REQ_FLUSH
| REQ_FUA
)) {
1539 thin_defer_bio(tc
, bio
);
1540 return DM_MAPIO_SUBMITTED
;
1543 r
= dm_thin_find_block(td
, block
, 0, &result
);
1546 * Note that we defer readahead too.
1550 if (unlikely(result
.shared
)) {
1552 * We have a race condition here between the
1553 * result.shared value returned by the lookup and
1554 * snapshot creation, which may cause new
1557 * To avoid this always quiesce the origin before
1558 * taking the snap. You want to do this anyway to
1559 * ensure a consistent application view
1562 * More distant ancestors are irrelevant. The
1563 * shared flag will be set in their case.
1565 thin_defer_bio(tc
, bio
);
1566 return DM_MAPIO_SUBMITTED
;
1569 build_virtual_key(tc
->td
, block
, &key
);
1570 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell1
, &cell_result
))
1571 return DM_MAPIO_SUBMITTED
;
1573 build_data_key(tc
->td
, result
.block
, &key
);
1574 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell2
, &cell_result
)) {
1575 cell_defer_no_holder_no_free(tc
, &cell1
);
1576 return DM_MAPIO_SUBMITTED
;
1579 inc_all_io_entry(tc
->pool
, bio
);
1580 cell_defer_no_holder_no_free(tc
, &cell2
);
1581 cell_defer_no_holder_no_free(tc
, &cell1
);
1583 remap(tc
, bio
, result
.block
);
1584 return DM_MAPIO_REMAPPED
;
1587 if (get_pool_mode(tc
->pool
) == PM_READ_ONLY
) {
1589 * This block isn't provisioned, and we have no way
1592 handle_unserviceable_bio(tc
->pool
, bio
);
1593 return DM_MAPIO_SUBMITTED
;
1599 * In future, the failed dm_thin_find_block above could
1600 * provide the hint to load the metadata into cache.
1602 thin_defer_bio(tc
, bio
);
1603 return DM_MAPIO_SUBMITTED
;
1607 * Must always call bio_io_error on failure.
1608 * dm_thin_find_block can fail with -EINVAL if the
1609 * pool is switched to fail-io mode.
1612 return DM_MAPIO_SUBMITTED
;
1616 static int pool_is_congested(struct dm_target_callbacks
*cb
, int bdi_bits
)
1619 unsigned long flags
;
1620 struct pool_c
*pt
= container_of(cb
, struct pool_c
, callbacks
);
1622 spin_lock_irqsave(&pt
->pool
->lock
, flags
);
1623 r
= !bio_list_empty(&pt
->pool
->retry_on_resume_list
);
1624 spin_unlock_irqrestore(&pt
->pool
->lock
, flags
);
1627 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1628 r
= bdi_congested(&q
->backing_dev_info
, bdi_bits
);
1634 static void __requeue_bios(struct pool
*pool
)
1636 bio_list_merge(&pool
->deferred_bios
, &pool
->retry_on_resume_list
);
1637 bio_list_init(&pool
->retry_on_resume_list
);
1640 /*----------------------------------------------------------------
1641 * Binding of control targets to a pool object
1642 *--------------------------------------------------------------*/
1643 static bool data_dev_supports_discard(struct pool_c
*pt
)
1645 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1647 return q
&& blk_queue_discard(q
);
1650 static bool is_factor(sector_t block_size
, uint32_t n
)
1652 return !sector_div(block_size
, n
);
1656 * If discard_passdown was enabled verify that the data device
1657 * supports discards. Disable discard_passdown if not.
1659 static void disable_passdown_if_not_supported(struct pool_c
*pt
)
1661 struct pool
*pool
= pt
->pool
;
1662 struct block_device
*data_bdev
= pt
->data_dev
->bdev
;
1663 struct queue_limits
*data_limits
= &bdev_get_queue(data_bdev
)->limits
;
1664 sector_t block_size
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
1665 const char *reason
= NULL
;
1666 char buf
[BDEVNAME_SIZE
];
1668 if (!pt
->adjusted_pf
.discard_passdown
)
1671 if (!data_dev_supports_discard(pt
))
1672 reason
= "discard unsupported";
1674 else if (data_limits
->max_discard_sectors
< pool
->sectors_per_block
)
1675 reason
= "max discard sectors smaller than a block";
1677 else if (data_limits
->discard_granularity
> block_size
)
1678 reason
= "discard granularity larger than a block";
1680 else if (!is_factor(block_size
, data_limits
->discard_granularity
))
1681 reason
= "discard granularity not a factor of block size";
1684 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev
, buf
), reason
);
1685 pt
->adjusted_pf
.discard_passdown
= false;
1689 static int bind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1691 struct pool_c
*pt
= ti
->private;
1694 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
1696 enum pool_mode old_mode
= pool
->pf
.mode
;
1697 enum pool_mode new_mode
= pt
->adjusted_pf
.mode
;
1700 * Don't change the pool's mode until set_pool_mode() below.
1701 * Otherwise the pool's process_* function pointers may
1702 * not match the desired pool mode.
1704 pt
->adjusted_pf
.mode
= old_mode
;
1707 pool
->pf
= pt
->adjusted_pf
;
1708 pool
->low_water_blocks
= pt
->low_water_blocks
;
1711 * If we were in PM_FAIL mode, rollback of metadata failed. We're
1712 * not going to recover without a thin_repair. So we never let the
1713 * pool move out of the old mode. On the other hand a PM_READ_ONLY
1714 * may have been due to a lack of metadata or data space, and may
1715 * now work (ie. if the underlying devices have been resized).
1717 if (old_mode
== PM_FAIL
)
1718 new_mode
= old_mode
;
1720 set_pool_mode(pool
, new_mode
);
1725 static void unbind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1731 /*----------------------------------------------------------------
1733 *--------------------------------------------------------------*/
1734 /* Initialize pool features. */
1735 static void pool_features_init(struct pool_features
*pf
)
1737 pf
->mode
= PM_WRITE
;
1738 pf
->zero_new_blocks
= true;
1739 pf
->discard_enabled
= true;
1740 pf
->discard_passdown
= true;
1741 pf
->error_if_no_space
= false;
1744 static void __pool_destroy(struct pool
*pool
)
1746 __pool_table_remove(pool
);
1748 if (dm_pool_metadata_close(pool
->pmd
) < 0)
1749 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1751 dm_bio_prison_destroy(pool
->prison
);
1752 dm_kcopyd_client_destroy(pool
->copier
);
1755 destroy_workqueue(pool
->wq
);
1757 if (pool
->next_mapping
)
1758 mempool_free(pool
->next_mapping
, pool
->mapping_pool
);
1759 mempool_destroy(pool
->mapping_pool
);
1760 dm_deferred_set_destroy(pool
->shared_read_ds
);
1761 dm_deferred_set_destroy(pool
->all_io_ds
);
1765 static struct kmem_cache
*_new_mapping_cache
;
1767 static struct pool
*pool_create(struct mapped_device
*pool_md
,
1768 struct block_device
*metadata_dev
,
1769 unsigned long block_size
,
1770 int read_only
, char **error
)
1775 struct dm_pool_metadata
*pmd
;
1776 bool format_device
= read_only
? false : true;
1778 pmd
= dm_pool_metadata_open(metadata_dev
, block_size
, format_device
);
1780 *error
= "Error creating metadata object";
1781 return (struct pool
*)pmd
;
1784 pool
= kmalloc(sizeof(*pool
), GFP_KERNEL
);
1786 *error
= "Error allocating memory for pool";
1787 err_p
= ERR_PTR(-ENOMEM
);
1792 pool
->sectors_per_block
= block_size
;
1793 if (block_size
& (block_size
- 1))
1794 pool
->sectors_per_block_shift
= -1;
1796 pool
->sectors_per_block_shift
= __ffs(block_size
);
1797 pool
->low_water_blocks
= 0;
1798 pool_features_init(&pool
->pf
);
1799 pool
->prison
= dm_bio_prison_create(PRISON_CELLS
);
1800 if (!pool
->prison
) {
1801 *error
= "Error creating pool's bio prison";
1802 err_p
= ERR_PTR(-ENOMEM
);
1806 pool
->copier
= dm_kcopyd_client_create(&dm_kcopyd_throttle
);
1807 if (IS_ERR(pool
->copier
)) {
1808 r
= PTR_ERR(pool
->copier
);
1809 *error
= "Error creating pool's kcopyd client";
1811 goto bad_kcopyd_client
;
1815 * Create singlethreaded workqueue that will service all devices
1816 * that use this metadata.
1818 pool
->wq
= alloc_ordered_workqueue("dm-" DM_MSG_PREFIX
, WQ_MEM_RECLAIM
);
1820 *error
= "Error creating pool's workqueue";
1821 err_p
= ERR_PTR(-ENOMEM
);
1825 INIT_WORK(&pool
->worker
, do_worker
);
1826 INIT_DELAYED_WORK(&pool
->waker
, do_waker
);
1827 spin_lock_init(&pool
->lock
);
1828 bio_list_init(&pool
->deferred_bios
);
1829 bio_list_init(&pool
->deferred_flush_bios
);
1830 INIT_LIST_HEAD(&pool
->prepared_mappings
);
1831 INIT_LIST_HEAD(&pool
->prepared_discards
);
1832 pool
->low_water_triggered
= false;
1833 bio_list_init(&pool
->retry_on_resume_list
);
1835 pool
->shared_read_ds
= dm_deferred_set_create();
1836 if (!pool
->shared_read_ds
) {
1837 *error
= "Error creating pool's shared read deferred set";
1838 err_p
= ERR_PTR(-ENOMEM
);
1839 goto bad_shared_read_ds
;
1842 pool
->all_io_ds
= dm_deferred_set_create();
1843 if (!pool
->all_io_ds
) {
1844 *error
= "Error creating pool's all io deferred set";
1845 err_p
= ERR_PTR(-ENOMEM
);
1849 pool
->next_mapping
= NULL
;
1850 pool
->mapping_pool
= mempool_create_slab_pool(MAPPING_POOL_SIZE
,
1851 _new_mapping_cache
);
1852 if (!pool
->mapping_pool
) {
1853 *error
= "Error creating pool's mapping mempool";
1854 err_p
= ERR_PTR(-ENOMEM
);
1855 goto bad_mapping_pool
;
1858 pool
->ref_count
= 1;
1859 pool
->last_commit_jiffies
= jiffies
;
1860 pool
->pool_md
= pool_md
;
1861 pool
->md_dev
= metadata_dev
;
1862 __pool_table_insert(pool
);
1867 dm_deferred_set_destroy(pool
->all_io_ds
);
1869 dm_deferred_set_destroy(pool
->shared_read_ds
);
1871 destroy_workqueue(pool
->wq
);
1873 dm_kcopyd_client_destroy(pool
->copier
);
1875 dm_bio_prison_destroy(pool
->prison
);
1879 if (dm_pool_metadata_close(pmd
))
1880 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1885 static void __pool_inc(struct pool
*pool
)
1887 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
1891 static void __pool_dec(struct pool
*pool
)
1893 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
1894 BUG_ON(!pool
->ref_count
);
1895 if (!--pool
->ref_count
)
1896 __pool_destroy(pool
);
1899 static struct pool
*__pool_find(struct mapped_device
*pool_md
,
1900 struct block_device
*metadata_dev
,
1901 unsigned long block_size
, int read_only
,
1902 char **error
, int *created
)
1904 struct pool
*pool
= __pool_table_lookup_metadata_dev(metadata_dev
);
1907 if (pool
->pool_md
!= pool_md
) {
1908 *error
= "metadata device already in use by a pool";
1909 return ERR_PTR(-EBUSY
);
1914 pool
= __pool_table_lookup(pool_md
);
1916 if (pool
->md_dev
!= metadata_dev
) {
1917 *error
= "different pool cannot replace a pool";
1918 return ERR_PTR(-EINVAL
);
1923 pool
= pool_create(pool_md
, metadata_dev
, block_size
, read_only
, error
);
1931 /*----------------------------------------------------------------
1932 * Pool target methods
1933 *--------------------------------------------------------------*/
1934 static void pool_dtr(struct dm_target
*ti
)
1936 struct pool_c
*pt
= ti
->private;
1938 mutex_lock(&dm_thin_pool_table
.mutex
);
1940 unbind_control_target(pt
->pool
, ti
);
1941 __pool_dec(pt
->pool
);
1942 dm_put_device(ti
, pt
->metadata_dev
);
1943 dm_put_device(ti
, pt
->data_dev
);
1946 mutex_unlock(&dm_thin_pool_table
.mutex
);
1949 static int parse_pool_features(struct dm_arg_set
*as
, struct pool_features
*pf
,
1950 struct dm_target
*ti
)
1954 const char *arg_name
;
1956 static struct dm_arg _args
[] = {
1957 {0, 4, "Invalid number of pool feature arguments"},
1961 * No feature arguments supplied.
1966 r
= dm_read_arg_group(_args
, as
, &argc
, &ti
->error
);
1970 while (argc
&& !r
) {
1971 arg_name
= dm_shift_arg(as
);
1974 if (!strcasecmp(arg_name
, "skip_block_zeroing"))
1975 pf
->zero_new_blocks
= false;
1977 else if (!strcasecmp(arg_name
, "ignore_discard"))
1978 pf
->discard_enabled
= false;
1980 else if (!strcasecmp(arg_name
, "no_discard_passdown"))
1981 pf
->discard_passdown
= false;
1983 else if (!strcasecmp(arg_name
, "read_only"))
1984 pf
->mode
= PM_READ_ONLY
;
1986 else if (!strcasecmp(arg_name
, "error_if_no_space"))
1987 pf
->error_if_no_space
= true;
1990 ti
->error
= "Unrecognised pool feature requested";
1999 static void metadata_low_callback(void *context
)
2001 struct pool
*pool
= context
;
2003 DMWARN("%s: reached low water mark for metadata device: sending event.",
2004 dm_device_name(pool
->pool_md
));
2006 dm_table_event(pool
->ti
->table
);
2009 static sector_t
get_dev_size(struct block_device
*bdev
)
2011 return i_size_read(bdev
->bd_inode
) >> SECTOR_SHIFT
;
2014 static void warn_if_metadata_device_too_big(struct block_device
*bdev
)
2016 sector_t metadata_dev_size
= get_dev_size(bdev
);
2017 char buffer
[BDEVNAME_SIZE
];
2019 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS_WARNING
)
2020 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
2021 bdevname(bdev
, buffer
), THIN_METADATA_MAX_SECTORS
);
2024 static sector_t
get_metadata_dev_size(struct block_device
*bdev
)
2026 sector_t metadata_dev_size
= get_dev_size(bdev
);
2028 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS
)
2029 metadata_dev_size
= THIN_METADATA_MAX_SECTORS
;
2031 return metadata_dev_size
;
2034 static dm_block_t
get_metadata_dev_size_in_blocks(struct block_device
*bdev
)
2036 sector_t metadata_dev_size
= get_metadata_dev_size(bdev
);
2038 sector_div(metadata_dev_size
, THIN_METADATA_BLOCK_SIZE
);
2040 return metadata_dev_size
;
2044 * When a metadata threshold is crossed a dm event is triggered, and
2045 * userland should respond by growing the metadata device. We could let
2046 * userland set the threshold, like we do with the data threshold, but I'm
2047 * not sure they know enough to do this well.
2049 static dm_block_t
calc_metadata_threshold(struct pool_c
*pt
)
2052 * 4M is ample for all ops with the possible exception of thin
2053 * device deletion which is harmless if it fails (just retry the
2054 * delete after you've grown the device).
2056 dm_block_t quarter
= get_metadata_dev_size_in_blocks(pt
->metadata_dev
->bdev
) / 4;
2057 return min((dm_block_t
)1024ULL /* 4M */, quarter
);
2061 * thin-pool <metadata dev> <data dev>
2062 * <data block size (sectors)>
2063 * <low water mark (blocks)>
2064 * [<#feature args> [<arg>]*]
2066 * Optional feature arguments are:
2067 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
2068 * ignore_discard: disable discard
2069 * no_discard_passdown: don't pass discards down to the data device
2070 * read_only: Don't allow any changes to be made to the pool metadata.
2071 * error_if_no_space: error IOs, instead of queueing, if no space.
2073 static int pool_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2075 int r
, pool_created
= 0;
2078 struct pool_features pf
;
2079 struct dm_arg_set as
;
2080 struct dm_dev
*data_dev
;
2081 unsigned long block_size
;
2082 dm_block_t low_water_blocks
;
2083 struct dm_dev
*metadata_dev
;
2084 fmode_t metadata_mode
;
2087 * FIXME Remove validation from scope of lock.
2089 mutex_lock(&dm_thin_pool_table
.mutex
);
2092 ti
->error
= "Invalid argument count";
2101 * Set default pool features.
2103 pool_features_init(&pf
);
2105 dm_consume_args(&as
, 4);
2106 r
= parse_pool_features(&as
, &pf
, ti
);
2110 metadata_mode
= FMODE_READ
| ((pf
.mode
== PM_READ_ONLY
) ? 0 : FMODE_WRITE
);
2111 r
= dm_get_device(ti
, argv
[0], metadata_mode
, &metadata_dev
);
2113 ti
->error
= "Error opening metadata block device";
2116 warn_if_metadata_device_too_big(metadata_dev
->bdev
);
2118 r
= dm_get_device(ti
, argv
[1], FMODE_READ
| FMODE_WRITE
, &data_dev
);
2120 ti
->error
= "Error getting data device";
2124 if (kstrtoul(argv
[2], 10, &block_size
) || !block_size
||
2125 block_size
< DATA_DEV_BLOCK_SIZE_MIN_SECTORS
||
2126 block_size
> DATA_DEV_BLOCK_SIZE_MAX_SECTORS
||
2127 block_size
& (DATA_DEV_BLOCK_SIZE_MIN_SECTORS
- 1)) {
2128 ti
->error
= "Invalid block size";
2133 if (kstrtoull(argv
[3], 10, (unsigned long long *)&low_water_blocks
)) {
2134 ti
->error
= "Invalid low water mark";
2139 pt
= kzalloc(sizeof(*pt
), GFP_KERNEL
);
2145 pool
= __pool_find(dm_table_get_md(ti
->table
), metadata_dev
->bdev
,
2146 block_size
, pf
.mode
== PM_READ_ONLY
, &ti
->error
, &pool_created
);
2153 * 'pool_created' reflects whether this is the first table load.
2154 * Top level discard support is not allowed to be changed after
2155 * initial load. This would require a pool reload to trigger thin
2158 if (!pool_created
&& pf
.discard_enabled
!= pool
->pf
.discard_enabled
) {
2159 ti
->error
= "Discard support cannot be disabled once enabled";
2161 goto out_flags_changed
;
2166 pt
->metadata_dev
= metadata_dev
;
2167 pt
->data_dev
= data_dev
;
2168 pt
->low_water_blocks
= low_water_blocks
;
2169 pt
->adjusted_pf
= pt
->requested_pf
= pf
;
2170 ti
->num_flush_bios
= 1;
2173 * Only need to enable discards if the pool should pass
2174 * them down to the data device. The thin device's discard
2175 * processing will cause mappings to be removed from the btree.
2177 ti
->discard_zeroes_data_unsupported
= true;
2178 if (pf
.discard_enabled
&& pf
.discard_passdown
) {
2179 ti
->num_discard_bios
= 1;
2182 * Setting 'discards_supported' circumvents the normal
2183 * stacking of discard limits (this keeps the pool and
2184 * thin devices' discard limits consistent).
2186 ti
->discards_supported
= true;
2190 r
= dm_pool_register_metadata_threshold(pt
->pool
->pmd
,
2191 calc_metadata_threshold(pt
),
2192 metadata_low_callback
,
2197 pt
->callbacks
.congested_fn
= pool_is_congested
;
2198 dm_table_add_target_callbacks(ti
->table
, &pt
->callbacks
);
2200 mutex_unlock(&dm_thin_pool_table
.mutex
);
2209 dm_put_device(ti
, data_dev
);
2211 dm_put_device(ti
, metadata_dev
);
2213 mutex_unlock(&dm_thin_pool_table
.mutex
);
2218 static int pool_map(struct dm_target
*ti
, struct bio
*bio
)
2221 struct pool_c
*pt
= ti
->private;
2222 struct pool
*pool
= pt
->pool
;
2223 unsigned long flags
;
2226 * As this is a singleton target, ti->begin is always zero.
2228 spin_lock_irqsave(&pool
->lock
, flags
);
2229 bio
->bi_bdev
= pt
->data_dev
->bdev
;
2230 r
= DM_MAPIO_REMAPPED
;
2231 spin_unlock_irqrestore(&pool
->lock
, flags
);
2236 static int maybe_resize_data_dev(struct dm_target
*ti
, bool *need_commit
)
2239 struct pool_c
*pt
= ti
->private;
2240 struct pool
*pool
= pt
->pool
;
2241 sector_t data_size
= ti
->len
;
2242 dm_block_t sb_data_size
;
2244 *need_commit
= false;
2246 (void) sector_div(data_size
, pool
->sectors_per_block
);
2248 r
= dm_pool_get_data_dev_size(pool
->pmd
, &sb_data_size
);
2250 DMERR("%s: failed to retrieve data device size",
2251 dm_device_name(pool
->pool_md
));
2255 if (data_size
< sb_data_size
) {
2256 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
2257 dm_device_name(pool
->pool_md
),
2258 (unsigned long long)data_size
, sb_data_size
);
2261 } else if (data_size
> sb_data_size
) {
2263 DMINFO("%s: growing the data device from %llu to %llu blocks",
2264 dm_device_name(pool
->pool_md
),
2265 sb_data_size
, (unsigned long long)data_size
);
2266 r
= dm_pool_resize_data_dev(pool
->pmd
, data_size
);
2268 metadata_operation_failed(pool
, "dm_pool_resize_data_dev", r
);
2272 *need_commit
= true;
2278 static int maybe_resize_metadata_dev(struct dm_target
*ti
, bool *need_commit
)
2281 struct pool_c
*pt
= ti
->private;
2282 struct pool
*pool
= pt
->pool
;
2283 dm_block_t metadata_dev_size
, sb_metadata_dev_size
;
2285 *need_commit
= false;
2287 metadata_dev_size
= get_metadata_dev_size_in_blocks(pool
->md_dev
);
2289 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &sb_metadata_dev_size
);
2291 DMERR("%s: failed to retrieve metadata device size",
2292 dm_device_name(pool
->pool_md
));
2296 if (metadata_dev_size
< sb_metadata_dev_size
) {
2297 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
2298 dm_device_name(pool
->pool_md
),
2299 metadata_dev_size
, sb_metadata_dev_size
);
2302 } else if (metadata_dev_size
> sb_metadata_dev_size
) {
2303 warn_if_metadata_device_too_big(pool
->md_dev
);
2304 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
2305 dm_device_name(pool
->pool_md
),
2306 sb_metadata_dev_size
, metadata_dev_size
);
2307 r
= dm_pool_resize_metadata_dev(pool
->pmd
, metadata_dev_size
);
2309 metadata_operation_failed(pool
, "dm_pool_resize_metadata_dev", r
);
2313 *need_commit
= true;
2320 * Retrieves the number of blocks of the data device from
2321 * the superblock and compares it to the actual device size,
2322 * thus resizing the data device in case it has grown.
2324 * This both copes with opening preallocated data devices in the ctr
2325 * being followed by a resume
2327 * calling the resume method individually after userspace has
2328 * grown the data device in reaction to a table event.
2330 static int pool_preresume(struct dm_target
*ti
)
2333 bool need_commit1
, need_commit2
;
2334 struct pool_c
*pt
= ti
->private;
2335 struct pool
*pool
= pt
->pool
;
2338 * Take control of the pool object.
2340 r
= bind_control_target(pool
, ti
);
2344 r
= maybe_resize_data_dev(ti
, &need_commit1
);
2348 r
= maybe_resize_metadata_dev(ti
, &need_commit2
);
2352 if (need_commit1
|| need_commit2
)
2353 (void) commit(pool
);
2358 static void pool_resume(struct dm_target
*ti
)
2360 struct pool_c
*pt
= ti
->private;
2361 struct pool
*pool
= pt
->pool
;
2362 unsigned long flags
;
2364 spin_lock_irqsave(&pool
->lock
, flags
);
2365 pool
->low_water_triggered
= false;
2366 __requeue_bios(pool
);
2367 spin_unlock_irqrestore(&pool
->lock
, flags
);
2369 do_waker(&pool
->waker
.work
);
2372 static void pool_postsuspend(struct dm_target
*ti
)
2374 struct pool_c
*pt
= ti
->private;
2375 struct pool
*pool
= pt
->pool
;
2377 cancel_delayed_work(&pool
->waker
);
2378 flush_workqueue(pool
->wq
);
2379 (void) commit(pool
);
2382 static int check_arg_count(unsigned argc
, unsigned args_required
)
2384 if (argc
!= args_required
) {
2385 DMWARN("Message received with %u arguments instead of %u.",
2386 argc
, args_required
);
2393 static int read_dev_id(char *arg
, dm_thin_id
*dev_id
, int warning
)
2395 if (!kstrtoull(arg
, 10, (unsigned long long *)dev_id
) &&
2396 *dev_id
<= MAX_DEV_ID
)
2400 DMWARN("Message received with invalid device id: %s", arg
);
2405 static int process_create_thin_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2410 r
= check_arg_count(argc
, 2);
2414 r
= read_dev_id(argv
[1], &dev_id
, 1);
2418 r
= dm_pool_create_thin(pool
->pmd
, dev_id
);
2420 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2428 static int process_create_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2431 dm_thin_id origin_dev_id
;
2434 r
= check_arg_count(argc
, 3);
2438 r
= read_dev_id(argv
[1], &dev_id
, 1);
2442 r
= read_dev_id(argv
[2], &origin_dev_id
, 1);
2446 r
= dm_pool_create_snap(pool
->pmd
, dev_id
, origin_dev_id
);
2448 DMWARN("Creation of new snapshot %s of device %s failed.",
2456 static int process_delete_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2461 r
= check_arg_count(argc
, 2);
2465 r
= read_dev_id(argv
[1], &dev_id
, 1);
2469 r
= dm_pool_delete_thin_device(pool
->pmd
, dev_id
);
2471 DMWARN("Deletion of thin device %s failed.", argv
[1]);
2476 static int process_set_transaction_id_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2478 dm_thin_id old_id
, new_id
;
2481 r
= check_arg_count(argc
, 3);
2485 if (kstrtoull(argv
[1], 10, (unsigned long long *)&old_id
)) {
2486 DMWARN("set_transaction_id message: Unrecognised id %s.", argv
[1]);
2490 if (kstrtoull(argv
[2], 10, (unsigned long long *)&new_id
)) {
2491 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv
[2]);
2495 r
= dm_pool_set_metadata_transaction_id(pool
->pmd
, old_id
, new_id
);
2497 DMWARN("Failed to change transaction id from %s to %s.",
2505 static int process_reserve_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2509 r
= check_arg_count(argc
, 1);
2513 (void) commit(pool
);
2515 r
= dm_pool_reserve_metadata_snap(pool
->pmd
);
2517 DMWARN("reserve_metadata_snap message failed.");
2522 static int process_release_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2526 r
= check_arg_count(argc
, 1);
2530 r
= dm_pool_release_metadata_snap(pool
->pmd
);
2532 DMWARN("release_metadata_snap message failed.");
2538 * Messages supported:
2539 * create_thin <dev_id>
2540 * create_snap <dev_id> <origin_id>
2542 * trim <dev_id> <new_size_in_sectors>
2543 * set_transaction_id <current_trans_id> <new_trans_id>
2544 * reserve_metadata_snap
2545 * release_metadata_snap
2547 static int pool_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
2550 struct pool_c
*pt
= ti
->private;
2551 struct pool
*pool
= pt
->pool
;
2553 if (!strcasecmp(argv
[0], "create_thin"))
2554 r
= process_create_thin_mesg(argc
, argv
, pool
);
2556 else if (!strcasecmp(argv
[0], "create_snap"))
2557 r
= process_create_snap_mesg(argc
, argv
, pool
);
2559 else if (!strcasecmp(argv
[0], "delete"))
2560 r
= process_delete_mesg(argc
, argv
, pool
);
2562 else if (!strcasecmp(argv
[0], "set_transaction_id"))
2563 r
= process_set_transaction_id_mesg(argc
, argv
, pool
);
2565 else if (!strcasecmp(argv
[0], "reserve_metadata_snap"))
2566 r
= process_reserve_metadata_snap_mesg(argc
, argv
, pool
);
2568 else if (!strcasecmp(argv
[0], "release_metadata_snap"))
2569 r
= process_release_metadata_snap_mesg(argc
, argv
, pool
);
2572 DMWARN("Unrecognised thin pool target message received: %s", argv
[0]);
2575 (void) commit(pool
);
2580 static void emit_flags(struct pool_features
*pf
, char *result
,
2581 unsigned sz
, unsigned maxlen
)
2583 unsigned count
= !pf
->zero_new_blocks
+ !pf
->discard_enabled
+
2584 !pf
->discard_passdown
+ (pf
->mode
== PM_READ_ONLY
) +
2585 pf
->error_if_no_space
;
2586 DMEMIT("%u ", count
);
2588 if (!pf
->zero_new_blocks
)
2589 DMEMIT("skip_block_zeroing ");
2591 if (!pf
->discard_enabled
)
2592 DMEMIT("ignore_discard ");
2594 if (!pf
->discard_passdown
)
2595 DMEMIT("no_discard_passdown ");
2597 if (pf
->mode
== PM_READ_ONLY
)
2598 DMEMIT("read_only ");
2600 if (pf
->error_if_no_space
)
2601 DMEMIT("error_if_no_space ");
2606 * <transaction id> <used metadata sectors>/<total metadata sectors>
2607 * <used data sectors>/<total data sectors> <held metadata root>
2609 static void pool_status(struct dm_target
*ti
, status_type_t type
,
2610 unsigned status_flags
, char *result
, unsigned maxlen
)
2614 uint64_t transaction_id
;
2615 dm_block_t nr_free_blocks_data
;
2616 dm_block_t nr_free_blocks_metadata
;
2617 dm_block_t nr_blocks_data
;
2618 dm_block_t nr_blocks_metadata
;
2619 dm_block_t held_root
;
2620 char buf
[BDEVNAME_SIZE
];
2621 char buf2
[BDEVNAME_SIZE
];
2622 struct pool_c
*pt
= ti
->private;
2623 struct pool
*pool
= pt
->pool
;
2626 case STATUSTYPE_INFO
:
2627 if (get_pool_mode(pool
) == PM_FAIL
) {
2632 /* Commit to ensure statistics aren't out-of-date */
2633 if (!(status_flags
& DM_STATUS_NOFLUSH_FLAG
) && !dm_suspended(ti
))
2634 (void) commit(pool
);
2636 r
= dm_pool_get_metadata_transaction_id(pool
->pmd
, &transaction_id
);
2638 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
2639 dm_device_name(pool
->pool_md
), r
);
2643 r
= dm_pool_get_free_metadata_block_count(pool
->pmd
, &nr_free_blocks_metadata
);
2645 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
2646 dm_device_name(pool
->pool_md
), r
);
2650 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &nr_blocks_metadata
);
2652 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
2653 dm_device_name(pool
->pool_md
), r
);
2657 r
= dm_pool_get_free_block_count(pool
->pmd
, &nr_free_blocks_data
);
2659 DMERR("%s: dm_pool_get_free_block_count returned %d",
2660 dm_device_name(pool
->pool_md
), r
);
2664 r
= dm_pool_get_data_dev_size(pool
->pmd
, &nr_blocks_data
);
2666 DMERR("%s: dm_pool_get_data_dev_size returned %d",
2667 dm_device_name(pool
->pool_md
), r
);
2671 r
= dm_pool_get_metadata_snap(pool
->pmd
, &held_root
);
2673 DMERR("%s: dm_pool_get_metadata_snap returned %d",
2674 dm_device_name(pool
->pool_md
), r
);
2678 DMEMIT("%llu %llu/%llu %llu/%llu ",
2679 (unsigned long long)transaction_id
,
2680 (unsigned long long)(nr_blocks_metadata
- nr_free_blocks_metadata
),
2681 (unsigned long long)nr_blocks_metadata
,
2682 (unsigned long long)(nr_blocks_data
- nr_free_blocks_data
),
2683 (unsigned long long)nr_blocks_data
);
2686 DMEMIT("%llu ", held_root
);
2690 if (pool
->pf
.mode
== PM_READ_ONLY
)
2695 if (!pool
->pf
.discard_enabled
)
2696 DMEMIT("ignore_discard ");
2697 else if (pool
->pf
.discard_passdown
)
2698 DMEMIT("discard_passdown ");
2700 DMEMIT("no_discard_passdown ");
2702 if (pool
->pf
.error_if_no_space
)
2703 DMEMIT("error_if_no_space ");
2705 DMEMIT("queue_if_no_space ");
2709 case STATUSTYPE_TABLE
:
2710 DMEMIT("%s %s %lu %llu ",
2711 format_dev_t(buf
, pt
->metadata_dev
->bdev
->bd_dev
),
2712 format_dev_t(buf2
, pt
->data_dev
->bdev
->bd_dev
),
2713 (unsigned long)pool
->sectors_per_block
,
2714 (unsigned long long)pt
->low_water_blocks
);
2715 emit_flags(&pt
->requested_pf
, result
, sz
, maxlen
);
2724 static int pool_iterate_devices(struct dm_target
*ti
,
2725 iterate_devices_callout_fn fn
, void *data
)
2727 struct pool_c
*pt
= ti
->private;
2729 return fn(ti
, pt
->data_dev
, 0, ti
->len
, data
);
2732 static int pool_merge(struct dm_target
*ti
, struct bvec_merge_data
*bvm
,
2733 struct bio_vec
*biovec
, int max_size
)
2735 struct pool_c
*pt
= ti
->private;
2736 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
2738 if (!q
->merge_bvec_fn
)
2741 bvm
->bi_bdev
= pt
->data_dev
->bdev
;
2743 return min(max_size
, q
->merge_bvec_fn(q
, bvm
, biovec
));
2746 static void set_discard_limits(struct pool_c
*pt
, struct queue_limits
*limits
)
2748 struct pool
*pool
= pt
->pool
;
2749 struct queue_limits
*data_limits
;
2751 limits
->max_discard_sectors
= pool
->sectors_per_block
;
2754 * discard_granularity is just a hint, and not enforced.
2756 if (pt
->adjusted_pf
.discard_passdown
) {
2757 data_limits
= &bdev_get_queue(pt
->data_dev
->bdev
)->limits
;
2758 limits
->discard_granularity
= data_limits
->discard_granularity
;
2760 limits
->discard_granularity
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
2763 static void pool_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
2765 struct pool_c
*pt
= ti
->private;
2766 struct pool
*pool
= pt
->pool
;
2767 uint64_t io_opt_sectors
= limits
->io_opt
>> SECTOR_SHIFT
;
2770 * If the system-determined stacked limits are compatible with the
2771 * pool's blocksize (io_opt is a factor) do not override them.
2773 if (io_opt_sectors
< pool
->sectors_per_block
||
2774 do_div(io_opt_sectors
, pool
->sectors_per_block
)) {
2775 blk_limits_io_min(limits
, 0);
2776 blk_limits_io_opt(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
2780 * pt->adjusted_pf is a staging area for the actual features to use.
2781 * They get transferred to the live pool in bind_control_target()
2782 * called from pool_preresume().
2784 if (!pt
->adjusted_pf
.discard_enabled
) {
2786 * Must explicitly disallow stacking discard limits otherwise the
2787 * block layer will stack them if pool's data device has support.
2788 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
2789 * user to see that, so make sure to set all discard limits to 0.
2791 limits
->discard_granularity
= 0;
2795 disable_passdown_if_not_supported(pt
);
2797 set_discard_limits(pt
, limits
);
2800 static struct target_type pool_target
= {
2801 .name
= "thin-pool",
2802 .features
= DM_TARGET_SINGLETON
| DM_TARGET_ALWAYS_WRITEABLE
|
2803 DM_TARGET_IMMUTABLE
,
2804 .version
= {1, 10, 0},
2805 .module
= THIS_MODULE
,
2809 .postsuspend
= pool_postsuspend
,
2810 .preresume
= pool_preresume
,
2811 .resume
= pool_resume
,
2812 .message
= pool_message
,
2813 .status
= pool_status
,
2814 .merge
= pool_merge
,
2815 .iterate_devices
= pool_iterate_devices
,
2816 .io_hints
= pool_io_hints
,
2819 /*----------------------------------------------------------------
2820 * Thin target methods
2821 *--------------------------------------------------------------*/
2822 static void thin_dtr(struct dm_target
*ti
)
2824 struct thin_c
*tc
= ti
->private;
2826 mutex_lock(&dm_thin_pool_table
.mutex
);
2828 __pool_dec(tc
->pool
);
2829 dm_pool_close_thin_device(tc
->td
);
2830 dm_put_device(ti
, tc
->pool_dev
);
2832 dm_put_device(ti
, tc
->origin_dev
);
2835 mutex_unlock(&dm_thin_pool_table
.mutex
);
2839 * Thin target parameters:
2841 * <pool_dev> <dev_id> [origin_dev]
2843 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2844 * dev_id: the internal device identifier
2845 * origin_dev: a device external to the pool that should act as the origin
2847 * If the pool device has discards disabled, they get disabled for the thin
2850 static int thin_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2854 struct dm_dev
*pool_dev
, *origin_dev
;
2855 struct mapped_device
*pool_md
;
2857 mutex_lock(&dm_thin_pool_table
.mutex
);
2859 if (argc
!= 2 && argc
!= 3) {
2860 ti
->error
= "Invalid argument count";
2865 tc
= ti
->private = kzalloc(sizeof(*tc
), GFP_KERNEL
);
2867 ti
->error
= "Out of memory";
2873 r
= dm_get_device(ti
, argv
[2], FMODE_READ
, &origin_dev
);
2875 ti
->error
= "Error opening origin device";
2876 goto bad_origin_dev
;
2878 tc
->origin_dev
= origin_dev
;
2881 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &pool_dev
);
2883 ti
->error
= "Error opening pool device";
2886 tc
->pool_dev
= pool_dev
;
2888 if (read_dev_id(argv
[1], (unsigned long long *)&tc
->dev_id
, 0)) {
2889 ti
->error
= "Invalid device id";
2894 pool_md
= dm_get_md(tc
->pool_dev
->bdev
->bd_dev
);
2896 ti
->error
= "Couldn't get pool mapped device";
2901 tc
->pool
= __pool_table_lookup(pool_md
);
2903 ti
->error
= "Couldn't find pool object";
2905 goto bad_pool_lookup
;
2907 __pool_inc(tc
->pool
);
2909 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
2910 ti
->error
= "Couldn't open thin device, Pool is in fail mode";
2915 r
= dm_pool_open_thin_device(tc
->pool
->pmd
, tc
->dev_id
, &tc
->td
);
2917 ti
->error
= "Couldn't open thin internal device";
2921 r
= dm_set_target_max_io_len(ti
, tc
->pool
->sectors_per_block
);
2923 goto bad_target_max_io_len
;
2925 ti
->num_flush_bios
= 1;
2926 ti
->flush_supported
= true;
2927 ti
->per_bio_data_size
= sizeof(struct dm_thin_endio_hook
);
2929 /* In case the pool supports discards, pass them on. */
2930 ti
->discard_zeroes_data_unsupported
= true;
2931 if (tc
->pool
->pf
.discard_enabled
) {
2932 ti
->discards_supported
= true;
2933 ti
->num_discard_bios
= 1;
2934 /* Discard bios must be split on a block boundary */
2935 ti
->split_discard_bios
= true;
2940 mutex_unlock(&dm_thin_pool_table
.mutex
);
2944 bad_target_max_io_len
:
2945 dm_pool_close_thin_device(tc
->td
);
2947 __pool_dec(tc
->pool
);
2951 dm_put_device(ti
, tc
->pool_dev
);
2954 dm_put_device(ti
, tc
->origin_dev
);
2958 mutex_unlock(&dm_thin_pool_table
.mutex
);
2963 static int thin_map(struct dm_target
*ti
, struct bio
*bio
)
2965 bio
->bi_iter
.bi_sector
= dm_target_offset(ti
, bio
->bi_iter
.bi_sector
);
2967 return thin_bio_map(ti
, bio
);
2970 static int thin_endio(struct dm_target
*ti
, struct bio
*bio
, int err
)
2972 unsigned long flags
;
2973 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
2974 struct list_head work
;
2975 struct dm_thin_new_mapping
*m
, *tmp
;
2976 struct pool
*pool
= h
->tc
->pool
;
2978 if (h
->shared_read_entry
) {
2979 INIT_LIST_HEAD(&work
);
2980 dm_deferred_entry_dec(h
->shared_read_entry
, &work
);
2982 spin_lock_irqsave(&pool
->lock
, flags
);
2983 list_for_each_entry_safe(m
, tmp
, &work
, list
) {
2986 __maybe_add_mapping(m
);
2988 spin_unlock_irqrestore(&pool
->lock
, flags
);
2991 if (h
->all_io_entry
) {
2992 INIT_LIST_HEAD(&work
);
2993 dm_deferred_entry_dec(h
->all_io_entry
, &work
);
2994 if (!list_empty(&work
)) {
2995 spin_lock_irqsave(&pool
->lock
, flags
);
2996 list_for_each_entry_safe(m
, tmp
, &work
, list
)
2997 list_add_tail(&m
->list
, &pool
->prepared_discards
);
2998 spin_unlock_irqrestore(&pool
->lock
, flags
);
3006 static void thin_postsuspend(struct dm_target
*ti
)
3008 if (dm_noflush_suspending(ti
))
3009 requeue_io((struct thin_c
*)ti
->private);
3013 * <nr mapped sectors> <highest mapped sector>
3015 static void thin_status(struct dm_target
*ti
, status_type_t type
,
3016 unsigned status_flags
, char *result
, unsigned maxlen
)
3020 dm_block_t mapped
, highest
;
3021 char buf
[BDEVNAME_SIZE
];
3022 struct thin_c
*tc
= ti
->private;
3024 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
3033 case STATUSTYPE_INFO
:
3034 r
= dm_thin_get_mapped_count(tc
->td
, &mapped
);
3036 DMERR("dm_thin_get_mapped_count returned %d", r
);
3040 r
= dm_thin_get_highest_mapped_block(tc
->td
, &highest
);
3042 DMERR("dm_thin_get_highest_mapped_block returned %d", r
);
3046 DMEMIT("%llu ", mapped
* tc
->pool
->sectors_per_block
);
3048 DMEMIT("%llu", ((highest
+ 1) *
3049 tc
->pool
->sectors_per_block
) - 1);
3054 case STATUSTYPE_TABLE
:
3056 format_dev_t(buf
, tc
->pool_dev
->bdev
->bd_dev
),
3057 (unsigned long) tc
->dev_id
);
3059 DMEMIT(" %s", format_dev_t(buf
, tc
->origin_dev
->bdev
->bd_dev
));
3070 static int thin_iterate_devices(struct dm_target
*ti
,
3071 iterate_devices_callout_fn fn
, void *data
)
3074 struct thin_c
*tc
= ti
->private;
3075 struct pool
*pool
= tc
->pool
;
3078 * We can't call dm_pool_get_data_dev_size() since that blocks. So
3079 * we follow a more convoluted path through to the pool's target.
3082 return 0; /* nothing is bound */
3084 blocks
= pool
->ti
->len
;
3085 (void) sector_div(blocks
, pool
->sectors_per_block
);
3087 return fn(ti
, tc
->pool_dev
, 0, pool
->sectors_per_block
* blocks
, data
);
3092 static struct target_type thin_target
= {
3094 .version
= {1, 10, 0},
3095 .module
= THIS_MODULE
,
3099 .end_io
= thin_endio
,
3100 .postsuspend
= thin_postsuspend
,
3101 .status
= thin_status
,
3102 .iterate_devices
= thin_iterate_devices
,
3105 /*----------------------------------------------------------------*/
3107 static int __init
dm_thin_init(void)
3113 r
= dm_register_target(&thin_target
);
3117 r
= dm_register_target(&pool_target
);
3119 goto bad_pool_target
;
3123 _new_mapping_cache
= KMEM_CACHE(dm_thin_new_mapping
, 0);
3124 if (!_new_mapping_cache
)
3125 goto bad_new_mapping_cache
;
3129 bad_new_mapping_cache
:
3130 dm_unregister_target(&pool_target
);
3132 dm_unregister_target(&thin_target
);
3137 static void dm_thin_exit(void)
3139 dm_unregister_target(&thin_target
);
3140 dm_unregister_target(&pool_target
);
3142 kmem_cache_destroy(_new_mapping_cache
);
3145 module_init(dm_thin_init
);
3146 module_exit(dm_thin_exit
);
3148 MODULE_DESCRIPTION(DM_NAME
" thin provisioning target");
3149 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3150 MODULE_LICENSE("GPL");