2 * Copyright (C) 2011-2012 Red Hat UK.
4 * This file is released under the GPL.
7 #include "dm-thin-metadata.h"
8 #include "dm-bio-prison.h"
11 #include <linux/device-mapper.h>
12 #include <linux/dm-io.h>
13 #include <linux/dm-kcopyd.h>
14 #include <linux/list.h>
15 #include <linux/init.h>
16 #include <linux/module.h>
17 #include <linux/slab.h>
19 #define DM_MSG_PREFIX "thin"
24 #define ENDIO_HOOK_POOL_SIZE 1024
25 #define MAPPING_POOL_SIZE 1024
26 #define PRISON_CELLS 1024
27 #define COMMIT_PERIOD HZ
29 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle
,
30 "A percentage of time allocated for copy on write");
33 * The block size of the device holding pool data must be
34 * between 64KB and 1GB.
36 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
37 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
40 * Device id is restricted to 24 bits.
42 #define MAX_DEV_ID ((1 << 24) - 1)
45 * How do we handle breaking sharing of data blocks?
46 * =================================================
48 * We use a standard copy-on-write btree to store the mappings for the
49 * devices (note I'm talking about copy-on-write of the metadata here, not
50 * the data). When you take an internal snapshot you clone the root node
51 * of the origin btree. After this there is no concept of an origin or a
52 * snapshot. They are just two device trees that happen to point to the
55 * When we get a write in we decide if it's to a shared data block using
56 * some timestamp magic. If it is, we have to break sharing.
58 * Let's say we write to a shared block in what was the origin. The
61 * i) plug io further to this physical block. (see bio_prison code).
63 * ii) quiesce any read io to that shared data block. Obviously
64 * including all devices that share this block. (see dm_deferred_set code)
66 * iii) copy the data block to a newly allocate block. This step can be
67 * missed out if the io covers the block. (schedule_copy).
69 * iv) insert the new mapping into the origin's btree
70 * (process_prepared_mapping). This act of inserting breaks some
71 * sharing of btree nodes between the two devices. Breaking sharing only
72 * effects the btree of that specific device. Btrees for the other
73 * devices that share the block never change. The btree for the origin
74 * device as it was after the last commit is untouched, ie. we're using
75 * persistent data structures in the functional programming sense.
77 * v) unplug io to this physical block, including the io that triggered
78 * the breaking of sharing.
80 * Steps (ii) and (iii) occur in parallel.
82 * The metadata _doesn't_ need to be committed before the io continues. We
83 * get away with this because the io is always written to a _new_ block.
84 * If there's a crash, then:
86 * - The origin mapping will point to the old origin block (the shared
87 * one). This will contain the data as it was before the io that triggered
88 * the breaking of sharing came in.
90 * - The snap mapping still points to the old block. As it would after
93 * The downside of this scheme is the timestamp magic isn't perfect, and
94 * will continue to think that data block in the snapshot device is shared
95 * even after the write to the origin has broken sharing. I suspect data
96 * blocks will typically be shared by many different devices, so we're
97 * breaking sharing n + 1 times, rather than n, where n is the number of
98 * devices that reference this data block. At the moment I think the
99 * benefits far, far outweigh the disadvantages.
102 /*----------------------------------------------------------------*/
107 static void build_data_key(struct dm_thin_device
*td
,
108 dm_block_t b
, struct dm_cell_key
*key
)
111 key
->dev
= dm_thin_dev_id(td
);
115 static void build_virtual_key(struct dm_thin_device
*td
, dm_block_t b
,
116 struct dm_cell_key
*key
)
119 key
->dev
= dm_thin_dev_id(td
);
123 /*----------------------------------------------------------------*/
126 * A pool device ties together a metadata device and a data device. It
127 * also provides the interface for creating and destroying internal
130 struct dm_thin_new_mapping
;
133 * The pool runs in 3 modes. Ordered in degraded order for comparisons.
136 PM_WRITE
, /* metadata may be changed */
137 PM_READ_ONLY
, /* metadata may not be changed */
138 PM_FAIL
, /* all I/O fails */
141 struct pool_features
{
144 bool zero_new_blocks
:1;
145 bool discard_enabled
:1;
146 bool discard_passdown
:1;
150 typedef void (*process_bio_fn
)(struct thin_c
*tc
, struct bio
*bio
);
151 typedef void (*process_mapping_fn
)(struct dm_thin_new_mapping
*m
);
154 struct list_head list
;
155 struct dm_target
*ti
; /* Only set if a pool target is bound */
157 struct mapped_device
*pool_md
;
158 struct block_device
*md_dev
;
159 struct dm_pool_metadata
*pmd
;
161 dm_block_t low_water_blocks
;
162 uint32_t sectors_per_block
;
163 int sectors_per_block_shift
;
165 struct pool_features pf
;
166 bool low_water_triggered
:1; /* A dm event has been sent */
167 bool no_free_space
:1; /* A -ENOSPC warning has been issued */
169 struct dm_bio_prison
*prison
;
170 struct dm_kcopyd_client
*copier
;
172 struct workqueue_struct
*wq
;
173 struct work_struct worker
;
174 struct delayed_work waker
;
176 unsigned long last_commit_jiffies
;
180 struct bio_list deferred_bios
;
181 struct bio_list deferred_flush_bios
;
182 struct list_head prepared_mappings
;
183 struct list_head prepared_discards
;
185 struct bio_list retry_on_resume_list
;
187 struct dm_deferred_set
*shared_read_ds
;
188 struct dm_deferred_set
*all_io_ds
;
190 struct dm_thin_new_mapping
*next_mapping
;
191 mempool_t
*mapping_pool
;
193 process_bio_fn process_bio
;
194 process_bio_fn process_discard
;
196 process_mapping_fn process_prepared_mapping
;
197 process_mapping_fn process_prepared_discard
;
200 static enum pool_mode
get_pool_mode(struct pool
*pool
);
201 static void metadata_operation_failed(struct pool
*pool
, const char *op
, int r
);
204 * Target context for a pool.
207 struct dm_target
*ti
;
209 struct dm_dev
*data_dev
;
210 struct dm_dev
*metadata_dev
;
211 struct dm_target_callbacks callbacks
;
213 dm_block_t low_water_blocks
;
214 struct pool_features requested_pf
; /* Features requested during table load */
215 struct pool_features adjusted_pf
; /* Features used after adjusting for constituent devices */
219 * Target context for a thin.
222 struct dm_dev
*pool_dev
;
223 struct dm_dev
*origin_dev
;
227 struct dm_thin_device
*td
;
230 /*----------------------------------------------------------------*/
233 * wake_worker() is used when new work is queued and when pool_resume is
234 * ready to continue deferred IO processing.
236 static void wake_worker(struct pool
*pool
)
238 queue_work(pool
->wq
, &pool
->worker
);
241 /*----------------------------------------------------------------*/
243 static int bio_detain(struct pool
*pool
, struct dm_cell_key
*key
, struct bio
*bio
,
244 struct dm_bio_prison_cell
**cell_result
)
247 struct dm_bio_prison_cell
*cell_prealloc
;
250 * Allocate a cell from the prison's mempool.
251 * This might block but it can't fail.
253 cell_prealloc
= dm_bio_prison_alloc_cell(pool
->prison
, GFP_NOIO
);
255 r
= dm_bio_detain(pool
->prison
, key
, bio
, cell_prealloc
, cell_result
);
258 * We reused an old cell; we can get rid of
261 dm_bio_prison_free_cell(pool
->prison
, cell_prealloc
);
266 static void cell_release(struct pool
*pool
,
267 struct dm_bio_prison_cell
*cell
,
268 struct bio_list
*bios
)
270 dm_cell_release(pool
->prison
, cell
, bios
);
271 dm_bio_prison_free_cell(pool
->prison
, cell
);
274 static void cell_release_no_holder(struct pool
*pool
,
275 struct dm_bio_prison_cell
*cell
,
276 struct bio_list
*bios
)
278 dm_cell_release_no_holder(pool
->prison
, cell
, bios
);
279 dm_bio_prison_free_cell(pool
->prison
, cell
);
282 static void cell_defer_no_holder_no_free(struct thin_c
*tc
,
283 struct dm_bio_prison_cell
*cell
)
285 struct pool
*pool
= tc
->pool
;
288 spin_lock_irqsave(&pool
->lock
, flags
);
289 dm_cell_release_no_holder(pool
->prison
, cell
, &pool
->deferred_bios
);
290 spin_unlock_irqrestore(&pool
->lock
, flags
);
295 static void cell_error(struct pool
*pool
,
296 struct dm_bio_prison_cell
*cell
)
298 dm_cell_error(pool
->prison
, cell
);
299 dm_bio_prison_free_cell(pool
->prison
, cell
);
302 /*----------------------------------------------------------------*/
305 * A global list of pools that uses a struct mapped_device as a key.
307 static struct dm_thin_pool_table
{
309 struct list_head pools
;
310 } dm_thin_pool_table
;
312 static void pool_table_init(void)
314 mutex_init(&dm_thin_pool_table
.mutex
);
315 INIT_LIST_HEAD(&dm_thin_pool_table
.pools
);
318 static void __pool_table_insert(struct pool
*pool
)
320 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
321 list_add(&pool
->list
, &dm_thin_pool_table
.pools
);
324 static void __pool_table_remove(struct pool
*pool
)
326 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
327 list_del(&pool
->list
);
330 static struct pool
*__pool_table_lookup(struct mapped_device
*md
)
332 struct pool
*pool
= NULL
, *tmp
;
334 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
336 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
337 if (tmp
->pool_md
== md
) {
346 static struct pool
*__pool_table_lookup_metadata_dev(struct block_device
*md_dev
)
348 struct pool
*pool
= NULL
, *tmp
;
350 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
352 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
353 if (tmp
->md_dev
== md_dev
) {
362 /*----------------------------------------------------------------*/
364 struct dm_thin_endio_hook
{
366 struct dm_deferred_entry
*shared_read_entry
;
367 struct dm_deferred_entry
*all_io_entry
;
368 struct dm_thin_new_mapping
*overwrite_mapping
;
371 static void __requeue_bio_list(struct thin_c
*tc
, struct bio_list
*master
)
374 struct bio_list bios
;
376 bio_list_init(&bios
);
377 bio_list_merge(&bios
, master
);
378 bio_list_init(master
);
380 while ((bio
= bio_list_pop(&bios
))) {
381 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
384 bio_endio(bio
, DM_ENDIO_REQUEUE
);
386 bio_list_add(master
, bio
);
390 static void requeue_io(struct thin_c
*tc
)
392 struct pool
*pool
= tc
->pool
;
395 spin_lock_irqsave(&pool
->lock
, flags
);
396 __requeue_bio_list(tc
, &pool
->deferred_bios
);
397 __requeue_bio_list(tc
, &pool
->retry_on_resume_list
);
398 spin_unlock_irqrestore(&pool
->lock
, flags
);
402 * This section of code contains the logic for processing a thin device's IO.
403 * Much of the code depends on pool object resources (lists, workqueues, etc)
404 * but most is exclusively called from the thin target rather than the thin-pool
408 static bool block_size_is_power_of_two(struct pool
*pool
)
410 return pool
->sectors_per_block_shift
>= 0;
413 static dm_block_t
get_bio_block(struct thin_c
*tc
, struct bio
*bio
)
415 struct pool
*pool
= tc
->pool
;
416 sector_t block_nr
= bio
->bi_sector
;
418 if (block_size_is_power_of_two(pool
))
419 block_nr
>>= pool
->sectors_per_block_shift
;
421 (void) sector_div(block_nr
, pool
->sectors_per_block
);
426 static void remap(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
)
428 struct pool
*pool
= tc
->pool
;
429 sector_t bi_sector
= bio
->bi_sector
;
431 bio
->bi_bdev
= tc
->pool_dev
->bdev
;
432 if (block_size_is_power_of_two(pool
))
433 bio
->bi_sector
= (block
<< pool
->sectors_per_block_shift
) |
434 (bi_sector
& (pool
->sectors_per_block
- 1));
436 bio
->bi_sector
= (block
* pool
->sectors_per_block
) +
437 sector_div(bi_sector
, pool
->sectors_per_block
);
440 static void remap_to_origin(struct thin_c
*tc
, struct bio
*bio
)
442 bio
->bi_bdev
= tc
->origin_dev
->bdev
;
445 static int bio_triggers_commit(struct thin_c
*tc
, struct bio
*bio
)
447 return (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) &&
448 dm_thin_changed_this_transaction(tc
->td
);
451 static void inc_all_io_entry(struct pool
*pool
, struct bio
*bio
)
453 struct dm_thin_endio_hook
*h
;
455 if (bio
->bi_rw
& REQ_DISCARD
)
458 h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
459 h
->all_io_entry
= dm_deferred_entry_inc(pool
->all_io_ds
);
462 static void issue(struct thin_c
*tc
, struct bio
*bio
)
464 struct pool
*pool
= tc
->pool
;
467 if (!bio_triggers_commit(tc
, bio
)) {
468 generic_make_request(bio
);
473 * Complete bio with an error if earlier I/O caused changes to
474 * the metadata that can't be committed e.g, due to I/O errors
475 * on the metadata device.
477 if (dm_thin_aborted_changes(tc
->td
)) {
483 * Batch together any bios that trigger commits and then issue a
484 * single commit for them in process_deferred_bios().
486 spin_lock_irqsave(&pool
->lock
, flags
);
487 bio_list_add(&pool
->deferred_flush_bios
, bio
);
488 spin_unlock_irqrestore(&pool
->lock
, flags
);
491 static void remap_to_origin_and_issue(struct thin_c
*tc
, struct bio
*bio
)
493 remap_to_origin(tc
, bio
);
497 static void remap_and_issue(struct thin_c
*tc
, struct bio
*bio
,
500 remap(tc
, bio
, block
);
504 /*----------------------------------------------------------------*/
507 * Bio endio functions.
509 struct dm_thin_new_mapping
{
510 struct list_head list
;
515 bool definitely_not_shared
:1;
519 dm_block_t virt_block
;
520 dm_block_t data_block
;
521 struct dm_bio_prison_cell
*cell
, *cell2
;
524 * If the bio covers the whole area of a block then we can avoid
525 * zeroing or copying. Instead this bio is hooked. The bio will
526 * still be in the cell, so care has to be taken to avoid issuing
530 bio_end_io_t
*saved_bi_end_io
;
533 static void __maybe_add_mapping(struct dm_thin_new_mapping
*m
)
535 struct pool
*pool
= m
->tc
->pool
;
537 if (m
->quiesced
&& m
->prepared
) {
538 list_add_tail(&m
->list
, &pool
->prepared_mappings
);
543 static void copy_complete(int read_err
, unsigned long write_err
, void *context
)
546 struct dm_thin_new_mapping
*m
= context
;
547 struct pool
*pool
= m
->tc
->pool
;
549 m
->err
= read_err
|| write_err
? -EIO
: 0;
551 spin_lock_irqsave(&pool
->lock
, flags
);
553 __maybe_add_mapping(m
);
554 spin_unlock_irqrestore(&pool
->lock
, flags
);
557 static void overwrite_endio(struct bio
*bio
, int err
)
560 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
561 struct dm_thin_new_mapping
*m
= h
->overwrite_mapping
;
562 struct pool
*pool
= m
->tc
->pool
;
566 spin_lock_irqsave(&pool
->lock
, flags
);
568 __maybe_add_mapping(m
);
569 spin_unlock_irqrestore(&pool
->lock
, flags
);
572 /*----------------------------------------------------------------*/
579 * Prepared mapping jobs.
583 * This sends the bios in the cell back to the deferred_bios list.
585 static void cell_defer(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
587 struct pool
*pool
= tc
->pool
;
590 spin_lock_irqsave(&pool
->lock
, flags
);
591 cell_release(pool
, cell
, &pool
->deferred_bios
);
592 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
598 * Same as cell_defer above, except it omits the original holder of the cell.
600 static void cell_defer_no_holder(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
602 struct pool
*pool
= tc
->pool
;
605 spin_lock_irqsave(&pool
->lock
, flags
);
606 cell_release_no_holder(pool
, cell
, &pool
->deferred_bios
);
607 spin_unlock_irqrestore(&pool
->lock
, flags
);
612 static void process_prepared_mapping_fail(struct dm_thin_new_mapping
*m
)
615 m
->bio
->bi_end_io
= m
->saved_bi_end_io
;
616 cell_error(m
->tc
->pool
, m
->cell
);
618 mempool_free(m
, m
->tc
->pool
->mapping_pool
);
621 static void process_prepared_mapping(struct dm_thin_new_mapping
*m
)
623 struct thin_c
*tc
= m
->tc
;
624 struct pool
*pool
= tc
->pool
;
630 bio
->bi_end_io
= m
->saved_bi_end_io
;
633 cell_error(pool
, m
->cell
);
638 * Commit the prepared block into the mapping btree.
639 * Any I/O for this block arriving after this point will get
640 * remapped to it directly.
642 r
= dm_thin_insert_block(tc
->td
, m
->virt_block
, m
->data_block
);
644 metadata_operation_failed(pool
, "dm_thin_insert_block", r
);
645 cell_error(pool
, m
->cell
);
650 * Release any bios held while the block was being provisioned.
651 * If we are processing a write bio that completely covers the block,
652 * we already processed it so can ignore it now when processing
653 * the bios in the cell.
656 cell_defer_no_holder(tc
, m
->cell
);
659 cell_defer(tc
, m
->cell
);
663 mempool_free(m
, pool
->mapping_pool
);
666 static void process_prepared_discard_fail(struct dm_thin_new_mapping
*m
)
668 struct thin_c
*tc
= m
->tc
;
670 bio_io_error(m
->bio
);
671 cell_defer_no_holder(tc
, m
->cell
);
672 cell_defer_no_holder(tc
, m
->cell2
);
673 mempool_free(m
, tc
->pool
->mapping_pool
);
676 static void process_prepared_discard_passdown(struct dm_thin_new_mapping
*m
)
678 struct thin_c
*tc
= m
->tc
;
680 inc_all_io_entry(tc
->pool
, m
->bio
);
681 cell_defer_no_holder(tc
, m
->cell
);
682 cell_defer_no_holder(tc
, m
->cell2
);
685 if (m
->definitely_not_shared
)
686 remap_and_issue(tc
, m
->bio
, m
->data_block
);
689 if (dm_pool_block_is_used(tc
->pool
->pmd
, m
->data_block
, &used
) || used
)
690 bio_endio(m
->bio
, 0);
692 remap_and_issue(tc
, m
->bio
, m
->data_block
);
695 bio_endio(m
->bio
, 0);
697 mempool_free(m
, tc
->pool
->mapping_pool
);
700 static void process_prepared_discard(struct dm_thin_new_mapping
*m
)
703 struct thin_c
*tc
= m
->tc
;
705 r
= dm_thin_remove_block(tc
->td
, m
->virt_block
);
707 DMERR_LIMIT("dm_thin_remove_block() failed");
709 process_prepared_discard_passdown(m
);
712 static void process_prepared(struct pool
*pool
, struct list_head
*head
,
713 process_mapping_fn
*fn
)
716 struct list_head maps
;
717 struct dm_thin_new_mapping
*m
, *tmp
;
719 INIT_LIST_HEAD(&maps
);
720 spin_lock_irqsave(&pool
->lock
, flags
);
721 list_splice_init(head
, &maps
);
722 spin_unlock_irqrestore(&pool
->lock
, flags
);
724 list_for_each_entry_safe(m
, tmp
, &maps
, list
)
731 static int io_overlaps_block(struct pool
*pool
, struct bio
*bio
)
733 return bio
->bi_size
== (pool
->sectors_per_block
<< SECTOR_SHIFT
);
736 static int io_overwrites_block(struct pool
*pool
, struct bio
*bio
)
738 return (bio_data_dir(bio
) == WRITE
) &&
739 io_overlaps_block(pool
, bio
);
742 static void save_and_set_endio(struct bio
*bio
, bio_end_io_t
**save
,
745 *save
= bio
->bi_end_io
;
749 static int ensure_next_mapping(struct pool
*pool
)
751 if (pool
->next_mapping
)
754 pool
->next_mapping
= mempool_alloc(pool
->mapping_pool
, GFP_ATOMIC
);
756 return pool
->next_mapping
? 0 : -ENOMEM
;
759 static struct dm_thin_new_mapping
*get_next_mapping(struct pool
*pool
)
761 struct dm_thin_new_mapping
*m
= pool
->next_mapping
;
763 BUG_ON(!pool
->next_mapping
);
765 memset(m
, 0, sizeof(struct dm_thin_new_mapping
));
766 INIT_LIST_HEAD(&m
->list
);
769 pool
->next_mapping
= NULL
;
774 static void schedule_copy(struct thin_c
*tc
, dm_block_t virt_block
,
775 struct dm_dev
*origin
, dm_block_t data_origin
,
776 dm_block_t data_dest
,
777 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
780 struct pool
*pool
= tc
->pool
;
781 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
784 m
->virt_block
= virt_block
;
785 m
->data_block
= data_dest
;
788 if (!dm_deferred_set_add_work(pool
->shared_read_ds
, &m
->list
))
792 * IO to pool_dev remaps to the pool target's data_dev.
794 * If the whole block of data is being overwritten, we can issue the
795 * bio immediately. Otherwise we use kcopyd to clone the data first.
797 if (io_overwrites_block(pool
, bio
)) {
798 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
800 h
->overwrite_mapping
= m
;
802 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
803 inc_all_io_entry(pool
, bio
);
804 remap_and_issue(tc
, bio
, data_dest
);
806 struct dm_io_region from
, to
;
808 from
.bdev
= origin
->bdev
;
809 from
.sector
= data_origin
* pool
->sectors_per_block
;
810 from
.count
= pool
->sectors_per_block
;
812 to
.bdev
= tc
->pool_dev
->bdev
;
813 to
.sector
= data_dest
* pool
->sectors_per_block
;
814 to
.count
= pool
->sectors_per_block
;
816 r
= dm_kcopyd_copy(pool
->copier
, &from
, 1, &to
,
817 0, copy_complete
, m
);
819 mempool_free(m
, pool
->mapping_pool
);
820 DMERR_LIMIT("dm_kcopyd_copy() failed");
821 cell_error(pool
, cell
);
826 static void schedule_internal_copy(struct thin_c
*tc
, dm_block_t virt_block
,
827 dm_block_t data_origin
, dm_block_t data_dest
,
828 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
830 schedule_copy(tc
, virt_block
, tc
->pool_dev
,
831 data_origin
, data_dest
, cell
, bio
);
834 static void schedule_external_copy(struct thin_c
*tc
, dm_block_t virt_block
,
835 dm_block_t data_dest
,
836 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
838 schedule_copy(tc
, virt_block
, tc
->origin_dev
,
839 virt_block
, data_dest
, cell
, bio
);
842 static void schedule_zero(struct thin_c
*tc
, dm_block_t virt_block
,
843 dm_block_t data_block
, struct dm_bio_prison_cell
*cell
,
846 struct pool
*pool
= tc
->pool
;
847 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
852 m
->virt_block
= virt_block
;
853 m
->data_block
= data_block
;
857 * If the whole block of data is being overwritten or we are not
858 * zeroing pre-existing data, we can issue the bio immediately.
859 * Otherwise we use kcopyd to zero the data first.
861 if (!pool
->pf
.zero_new_blocks
)
862 process_prepared_mapping(m
);
864 else if (io_overwrites_block(pool
, bio
)) {
865 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
867 h
->overwrite_mapping
= m
;
869 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
870 inc_all_io_entry(pool
, bio
);
871 remap_and_issue(tc
, bio
, data_block
);
874 struct dm_io_region to
;
876 to
.bdev
= tc
->pool_dev
->bdev
;
877 to
.sector
= data_block
* pool
->sectors_per_block
;
878 to
.count
= pool
->sectors_per_block
;
880 r
= dm_kcopyd_zero(pool
->copier
, 1, &to
, 0, copy_complete
, m
);
882 mempool_free(m
, pool
->mapping_pool
);
883 DMERR_LIMIT("dm_kcopyd_zero() failed");
884 cell_error(pool
, cell
);
890 * A non-zero return indicates read_only or fail_io mode.
891 * Many callers don't care about the return value.
893 static int commit(struct pool
*pool
)
897 if (get_pool_mode(pool
) != PM_WRITE
)
900 r
= dm_pool_commit_metadata(pool
->pmd
);
902 metadata_operation_failed(pool
, "dm_pool_commit_metadata", r
);
907 static void check_low_water_mark(struct pool
*pool
, dm_block_t free_blocks
)
911 if (free_blocks
<= pool
->low_water_blocks
&& !pool
->low_water_triggered
) {
912 DMWARN("%s: reached low water mark for data device: sending event.",
913 dm_device_name(pool
->pool_md
));
914 spin_lock_irqsave(&pool
->lock
, flags
);
915 pool
->low_water_triggered
= true;
916 spin_unlock_irqrestore(&pool
->lock
, flags
);
917 dm_table_event(pool
->ti
->table
);
921 static int alloc_data_block(struct thin_c
*tc
, dm_block_t
*result
)
924 dm_block_t free_blocks
;
926 struct pool
*pool
= tc
->pool
;
929 * Once no_free_space is set we must not allow allocation to succeed.
930 * Otherwise it is difficult to explain, debug, test and support.
932 if (pool
->no_free_space
)
935 if (get_pool_mode(pool
) != PM_WRITE
)
938 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
940 metadata_operation_failed(pool
, "dm_pool_get_free_block_count", r
);
944 check_low_water_mark(pool
, free_blocks
);
948 * Try to commit to see if that will free up some
955 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
957 metadata_operation_failed(pool
, "dm_pool_get_free_block_count", r
);
962 * If we still have no space we set a flag to avoid
963 * doing all this checking and return -ENOSPC. This
964 * flag serves as a latch that disallows allocations from
965 * this pool until the admin takes action (e.g. resize or
969 DMWARN("%s: no free data space available.",
970 dm_device_name(pool
->pool_md
));
971 spin_lock_irqsave(&pool
->lock
, flags
);
972 pool
->no_free_space
= true;
973 spin_unlock_irqrestore(&pool
->lock
, flags
);
978 r
= dm_pool_alloc_data_block(pool
->pmd
, result
);
981 !dm_pool_get_free_metadata_block_count(pool
->pmd
, &free_blocks
) &&
983 DMWARN("%s: no free metadata space available.",
984 dm_device_name(pool
->pool_md
));
986 metadata_operation_failed(pool
, "dm_pool_alloc_data_block", r
);
994 * If we have run out of space, queue bios until the device is
995 * resumed, presumably after having been reloaded with more space.
997 static void retry_on_resume(struct bio
*bio
)
999 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1000 struct thin_c
*tc
= h
->tc
;
1001 struct pool
*pool
= tc
->pool
;
1002 unsigned long flags
;
1004 spin_lock_irqsave(&pool
->lock
, flags
);
1005 bio_list_add(&pool
->retry_on_resume_list
, bio
);
1006 spin_unlock_irqrestore(&pool
->lock
, flags
);
1009 static void no_space(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
1012 struct bio_list bios
;
1014 bio_list_init(&bios
);
1015 cell_release(pool
, cell
, &bios
);
1017 while ((bio
= bio_list_pop(&bios
)))
1018 retry_on_resume(bio
);
1021 static void process_discard(struct thin_c
*tc
, struct bio
*bio
)
1024 unsigned long flags
;
1025 struct pool
*pool
= tc
->pool
;
1026 struct dm_bio_prison_cell
*cell
, *cell2
;
1027 struct dm_cell_key key
, key2
;
1028 dm_block_t block
= get_bio_block(tc
, bio
);
1029 struct dm_thin_lookup_result lookup_result
;
1030 struct dm_thin_new_mapping
*m
;
1032 build_virtual_key(tc
->td
, block
, &key
);
1033 if (bio_detain(tc
->pool
, &key
, bio
, &cell
))
1036 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1040 * Check nobody is fiddling with this pool block. This can
1041 * happen if someone's in the process of breaking sharing
1044 build_data_key(tc
->td
, lookup_result
.block
, &key2
);
1045 if (bio_detain(tc
->pool
, &key2
, bio
, &cell2
)) {
1046 cell_defer_no_holder(tc
, cell
);
1050 if (io_overlaps_block(pool
, bio
)) {
1052 * IO may still be going to the destination block. We must
1053 * quiesce before we can do the removal.
1055 m
= get_next_mapping(pool
);
1057 m
->pass_discard
= pool
->pf
.discard_passdown
;
1058 m
->definitely_not_shared
= !lookup_result
.shared
;
1059 m
->virt_block
= block
;
1060 m
->data_block
= lookup_result
.block
;
1065 if (!dm_deferred_set_add_work(pool
->all_io_ds
, &m
->list
)) {
1066 spin_lock_irqsave(&pool
->lock
, flags
);
1067 list_add_tail(&m
->list
, &pool
->prepared_discards
);
1068 spin_unlock_irqrestore(&pool
->lock
, flags
);
1072 inc_all_io_entry(pool
, bio
);
1073 cell_defer_no_holder(tc
, cell
);
1074 cell_defer_no_holder(tc
, cell2
);
1077 * The DM core makes sure that the discard doesn't span
1078 * a block boundary. So we submit the discard of a
1079 * partial block appropriately.
1081 if ((!lookup_result
.shared
) && pool
->pf
.discard_passdown
)
1082 remap_and_issue(tc
, bio
, lookup_result
.block
);
1090 * It isn't provisioned, just forget it.
1092 cell_defer_no_holder(tc
, cell
);
1097 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1099 cell_defer_no_holder(tc
, cell
);
1105 static void break_sharing(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1106 struct dm_cell_key
*key
,
1107 struct dm_thin_lookup_result
*lookup_result
,
1108 struct dm_bio_prison_cell
*cell
)
1111 dm_block_t data_block
;
1112 struct pool
*pool
= tc
->pool
;
1114 r
= alloc_data_block(tc
, &data_block
);
1117 schedule_internal_copy(tc
, block
, lookup_result
->block
,
1118 data_block
, cell
, bio
);
1122 no_space(pool
, cell
);
1126 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1128 cell_error(pool
, cell
);
1133 static void process_shared_bio(struct thin_c
*tc
, struct bio
*bio
,
1135 struct dm_thin_lookup_result
*lookup_result
)
1137 struct dm_bio_prison_cell
*cell
;
1138 struct pool
*pool
= tc
->pool
;
1139 struct dm_cell_key key
;
1142 * If cell is already occupied, then sharing is already in the process
1143 * of being broken so we have nothing further to do here.
1145 build_data_key(tc
->td
, lookup_result
->block
, &key
);
1146 if (bio_detain(pool
, &key
, bio
, &cell
))
1149 if (bio_data_dir(bio
) == WRITE
&& bio
->bi_size
)
1150 break_sharing(tc
, bio
, block
, &key
, lookup_result
, cell
);
1152 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1154 h
->shared_read_entry
= dm_deferred_entry_inc(pool
->shared_read_ds
);
1155 inc_all_io_entry(pool
, bio
);
1156 cell_defer_no_holder(tc
, cell
);
1158 remap_and_issue(tc
, bio
, lookup_result
->block
);
1162 static void provision_block(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1163 struct dm_bio_prison_cell
*cell
)
1166 dm_block_t data_block
;
1167 struct pool
*pool
= tc
->pool
;
1170 * Remap empty bios (flushes) immediately, without provisioning.
1172 if (!bio
->bi_size
) {
1173 inc_all_io_entry(pool
, bio
);
1174 cell_defer_no_holder(tc
, cell
);
1176 remap_and_issue(tc
, bio
, 0);
1181 * Fill read bios with zeroes and complete them immediately.
1183 if (bio_data_dir(bio
) == READ
) {
1185 cell_defer_no_holder(tc
, cell
);
1190 r
= alloc_data_block(tc
, &data_block
);
1194 schedule_external_copy(tc
, block
, data_block
, cell
, bio
);
1196 schedule_zero(tc
, block
, data_block
, cell
, bio
);
1200 no_space(pool
, cell
);
1204 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1206 cell_error(pool
, cell
);
1211 static void process_bio(struct thin_c
*tc
, struct bio
*bio
)
1214 struct pool
*pool
= tc
->pool
;
1215 dm_block_t block
= get_bio_block(tc
, bio
);
1216 struct dm_bio_prison_cell
*cell
;
1217 struct dm_cell_key key
;
1218 struct dm_thin_lookup_result lookup_result
;
1221 * If cell is already occupied, then the block is already
1222 * being provisioned so we have nothing further to do here.
1224 build_virtual_key(tc
->td
, block
, &key
);
1225 if (bio_detain(pool
, &key
, bio
, &cell
))
1228 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1231 if (lookup_result
.shared
) {
1232 process_shared_bio(tc
, bio
, block
, &lookup_result
);
1233 cell_defer_no_holder(tc
, cell
); /* FIXME: pass this cell into process_shared? */
1235 inc_all_io_entry(pool
, bio
);
1236 cell_defer_no_holder(tc
, cell
);
1238 remap_and_issue(tc
, bio
, lookup_result
.block
);
1243 if (bio_data_dir(bio
) == READ
&& tc
->origin_dev
) {
1244 inc_all_io_entry(pool
, bio
);
1245 cell_defer_no_holder(tc
, cell
);
1247 remap_to_origin_and_issue(tc
, bio
);
1249 provision_block(tc
, bio
, block
, cell
);
1253 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1255 cell_defer_no_holder(tc
, cell
);
1261 static void process_bio_read_only(struct thin_c
*tc
, struct bio
*bio
)
1264 int rw
= bio_data_dir(bio
);
1265 dm_block_t block
= get_bio_block(tc
, bio
);
1266 struct dm_thin_lookup_result lookup_result
;
1268 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1271 if (lookup_result
.shared
&& (rw
== WRITE
) && bio
->bi_size
)
1274 inc_all_io_entry(tc
->pool
, bio
);
1275 remap_and_issue(tc
, bio
, lookup_result
.block
);
1285 if (tc
->origin_dev
) {
1286 inc_all_io_entry(tc
->pool
, bio
);
1287 remap_to_origin_and_issue(tc
, bio
);
1296 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1303 static void process_bio_fail(struct thin_c
*tc
, struct bio
*bio
)
1309 * FIXME: should we also commit due to size of transaction, measured in
1312 static int need_commit_due_to_time(struct pool
*pool
)
1314 return jiffies
< pool
->last_commit_jiffies
||
1315 jiffies
> pool
->last_commit_jiffies
+ COMMIT_PERIOD
;
1318 static void process_deferred_bios(struct pool
*pool
)
1320 unsigned long flags
;
1322 struct bio_list bios
;
1324 bio_list_init(&bios
);
1326 spin_lock_irqsave(&pool
->lock
, flags
);
1327 bio_list_merge(&bios
, &pool
->deferred_bios
);
1328 bio_list_init(&pool
->deferred_bios
);
1329 spin_unlock_irqrestore(&pool
->lock
, flags
);
1331 while ((bio
= bio_list_pop(&bios
))) {
1332 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1333 struct thin_c
*tc
= h
->tc
;
1336 * If we've got no free new_mapping structs, and processing
1337 * this bio might require one, we pause until there are some
1338 * prepared mappings to process.
1340 if (ensure_next_mapping(pool
)) {
1341 spin_lock_irqsave(&pool
->lock
, flags
);
1342 bio_list_merge(&pool
->deferred_bios
, &bios
);
1343 spin_unlock_irqrestore(&pool
->lock
, flags
);
1348 if (bio
->bi_rw
& REQ_DISCARD
)
1349 pool
->process_discard(tc
, bio
);
1351 pool
->process_bio(tc
, bio
);
1355 * If there are any deferred flush bios, we must commit
1356 * the metadata before issuing them.
1358 bio_list_init(&bios
);
1359 spin_lock_irqsave(&pool
->lock
, flags
);
1360 bio_list_merge(&bios
, &pool
->deferred_flush_bios
);
1361 bio_list_init(&pool
->deferred_flush_bios
);
1362 spin_unlock_irqrestore(&pool
->lock
, flags
);
1364 if (bio_list_empty(&bios
) && !need_commit_due_to_time(pool
))
1368 while ((bio
= bio_list_pop(&bios
)))
1372 pool
->last_commit_jiffies
= jiffies
;
1374 while ((bio
= bio_list_pop(&bios
)))
1375 generic_make_request(bio
);
1378 static void do_worker(struct work_struct
*ws
)
1380 struct pool
*pool
= container_of(ws
, struct pool
, worker
);
1382 process_prepared(pool
, &pool
->prepared_mappings
, &pool
->process_prepared_mapping
);
1383 process_prepared(pool
, &pool
->prepared_discards
, &pool
->process_prepared_discard
);
1384 process_deferred_bios(pool
);
1388 * We want to commit periodically so that not too much
1389 * unwritten data builds up.
1391 static void do_waker(struct work_struct
*ws
)
1393 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
, waker
);
1395 queue_delayed_work(pool
->wq
, &pool
->waker
, COMMIT_PERIOD
);
1398 /*----------------------------------------------------------------*/
1400 static enum pool_mode
get_pool_mode(struct pool
*pool
)
1402 return pool
->pf
.mode
;
1405 static void set_pool_mode(struct pool
*pool
, enum pool_mode mode
)
1409 pool
->pf
.mode
= mode
;
1413 DMERR("%s: switching pool to failure mode",
1414 dm_device_name(pool
->pool_md
));
1415 dm_pool_metadata_read_only(pool
->pmd
);
1416 pool
->process_bio
= process_bio_fail
;
1417 pool
->process_discard
= process_bio_fail
;
1418 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1419 pool
->process_prepared_discard
= process_prepared_discard_fail
;
1423 DMERR("%s: switching pool to read-only mode",
1424 dm_device_name(pool
->pool_md
));
1425 r
= dm_pool_abort_metadata(pool
->pmd
);
1427 DMERR("%s: aborting transaction failed",
1428 dm_device_name(pool
->pool_md
));
1429 set_pool_mode(pool
, PM_FAIL
);
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 dm_pool_metadata_read_write(pool
->pmd
);
1441 pool
->process_bio
= process_bio
;
1442 pool
->process_discard
= process_discard
;
1443 pool
->process_prepared_mapping
= process_prepared_mapping
;
1444 pool
->process_prepared_discard
= process_prepared_discard
;
1450 * Rather than calling set_pool_mode directly, use these which describe the
1451 * reason for mode degradation.
1453 static void metadata_operation_failed(struct pool
*pool
, const char *op
, int r
)
1455 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
1456 dm_device_name(pool
->pool_md
), op
, r
);
1458 set_pool_mode(pool
, PM_READ_ONLY
);
1461 /*----------------------------------------------------------------*/
1464 * Mapping functions.
1468 * Called only while mapping a thin bio to hand it over to the workqueue.
1470 static void thin_defer_bio(struct thin_c
*tc
, struct bio
*bio
)
1472 unsigned long flags
;
1473 struct pool
*pool
= tc
->pool
;
1475 spin_lock_irqsave(&pool
->lock
, flags
);
1476 bio_list_add(&pool
->deferred_bios
, bio
);
1477 spin_unlock_irqrestore(&pool
->lock
, flags
);
1482 static void thin_hook_bio(struct thin_c
*tc
, struct bio
*bio
)
1484 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1487 h
->shared_read_entry
= NULL
;
1488 h
->all_io_entry
= NULL
;
1489 h
->overwrite_mapping
= NULL
;
1493 * Non-blocking function called from the thin target's map function.
1495 static int thin_bio_map(struct dm_target
*ti
, struct bio
*bio
)
1498 struct thin_c
*tc
= ti
->private;
1499 dm_block_t block
= get_bio_block(tc
, bio
);
1500 struct dm_thin_device
*td
= tc
->td
;
1501 struct dm_thin_lookup_result result
;
1502 struct dm_bio_prison_cell cell1
, cell2
;
1503 struct dm_bio_prison_cell
*cell_result
;
1504 struct dm_cell_key key
;
1506 thin_hook_bio(tc
, bio
);
1508 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
1510 return DM_MAPIO_SUBMITTED
;
1513 if (bio
->bi_rw
& (REQ_DISCARD
| REQ_FLUSH
| REQ_FUA
)) {
1514 thin_defer_bio(tc
, bio
);
1515 return DM_MAPIO_SUBMITTED
;
1518 r
= dm_thin_find_block(td
, block
, 0, &result
);
1521 * Note that we defer readahead too.
1525 if (unlikely(result
.shared
)) {
1527 * We have a race condition here between the
1528 * result.shared value returned by the lookup and
1529 * snapshot creation, which may cause new
1532 * To avoid this always quiesce the origin before
1533 * taking the snap. You want to do this anyway to
1534 * ensure a consistent application view
1537 * More distant ancestors are irrelevant. The
1538 * shared flag will be set in their case.
1540 thin_defer_bio(tc
, bio
);
1541 return DM_MAPIO_SUBMITTED
;
1544 build_virtual_key(tc
->td
, block
, &key
);
1545 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell1
, &cell_result
))
1546 return DM_MAPIO_SUBMITTED
;
1548 build_data_key(tc
->td
, result
.block
, &key
);
1549 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell2
, &cell_result
)) {
1550 cell_defer_no_holder_no_free(tc
, &cell1
);
1551 return DM_MAPIO_SUBMITTED
;
1554 inc_all_io_entry(tc
->pool
, bio
);
1555 cell_defer_no_holder_no_free(tc
, &cell2
);
1556 cell_defer_no_holder_no_free(tc
, &cell1
);
1558 remap(tc
, bio
, result
.block
);
1559 return DM_MAPIO_REMAPPED
;
1562 if (get_pool_mode(tc
->pool
) == PM_READ_ONLY
) {
1564 * This block isn't provisioned, and we have no way
1565 * of doing so. Just error it.
1568 return DM_MAPIO_SUBMITTED
;
1574 * In future, the failed dm_thin_find_block above could
1575 * provide the hint to load the metadata into cache.
1577 thin_defer_bio(tc
, bio
);
1578 return DM_MAPIO_SUBMITTED
;
1582 * Must always call bio_io_error on failure.
1583 * dm_thin_find_block can fail with -EINVAL if the
1584 * pool is switched to fail-io mode.
1587 return DM_MAPIO_SUBMITTED
;
1591 static int pool_is_congested(struct dm_target_callbacks
*cb
, int bdi_bits
)
1594 unsigned long flags
;
1595 struct pool_c
*pt
= container_of(cb
, struct pool_c
, callbacks
);
1597 spin_lock_irqsave(&pt
->pool
->lock
, flags
);
1598 r
= !bio_list_empty(&pt
->pool
->retry_on_resume_list
);
1599 spin_unlock_irqrestore(&pt
->pool
->lock
, flags
);
1602 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1603 r
= bdi_congested(&q
->backing_dev_info
, bdi_bits
);
1609 static void __requeue_bios(struct pool
*pool
)
1611 bio_list_merge(&pool
->deferred_bios
, &pool
->retry_on_resume_list
);
1612 bio_list_init(&pool
->retry_on_resume_list
);
1615 /*----------------------------------------------------------------
1616 * Binding of control targets to a pool object
1617 *--------------------------------------------------------------*/
1618 static bool data_dev_supports_discard(struct pool_c
*pt
)
1620 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1622 return q
&& blk_queue_discard(q
);
1625 static bool is_factor(sector_t block_size
, uint32_t n
)
1627 return !sector_div(block_size
, n
);
1631 * If discard_passdown was enabled verify that the data device
1632 * supports discards. Disable discard_passdown if not.
1634 static void disable_passdown_if_not_supported(struct pool_c
*pt
)
1636 struct pool
*pool
= pt
->pool
;
1637 struct block_device
*data_bdev
= pt
->data_dev
->bdev
;
1638 struct queue_limits
*data_limits
= &bdev_get_queue(data_bdev
)->limits
;
1639 sector_t block_size
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
1640 const char *reason
= NULL
;
1641 char buf
[BDEVNAME_SIZE
];
1643 if (!pt
->adjusted_pf
.discard_passdown
)
1646 if (!data_dev_supports_discard(pt
))
1647 reason
= "discard unsupported";
1649 else if (data_limits
->max_discard_sectors
< pool
->sectors_per_block
)
1650 reason
= "max discard sectors smaller than a block";
1652 else if (data_limits
->discard_granularity
> block_size
)
1653 reason
= "discard granularity larger than a block";
1655 else if (!is_factor(block_size
, data_limits
->discard_granularity
))
1656 reason
= "discard granularity not a factor of block size";
1659 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev
, buf
), reason
);
1660 pt
->adjusted_pf
.discard_passdown
= false;
1664 static int bind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1666 struct pool_c
*pt
= ti
->private;
1669 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
1671 enum pool_mode old_mode
= pool
->pf
.mode
;
1672 enum pool_mode new_mode
= pt
->adjusted_pf
.mode
;
1675 * If we were in PM_FAIL mode, rollback of metadata failed. We're
1676 * not going to recover without a thin_repair. So we never let the
1677 * pool move out of the old mode. On the other hand a PM_READ_ONLY
1678 * may have been due to a lack of metadata or data space, and may
1679 * now work (ie. if the underlying devices have been resized).
1681 if (old_mode
== PM_FAIL
)
1682 new_mode
= old_mode
;
1685 pool
->low_water_blocks
= pt
->low_water_blocks
;
1686 pool
->pf
= pt
->adjusted_pf
;
1688 set_pool_mode(pool
, new_mode
);
1693 static void unbind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1699 /*----------------------------------------------------------------
1701 *--------------------------------------------------------------*/
1702 /* Initialize pool features. */
1703 static void pool_features_init(struct pool_features
*pf
)
1705 pf
->mode
= PM_WRITE
;
1706 pf
->zero_new_blocks
= true;
1707 pf
->discard_enabled
= true;
1708 pf
->discard_passdown
= true;
1711 static void __pool_destroy(struct pool
*pool
)
1713 __pool_table_remove(pool
);
1715 if (dm_pool_metadata_close(pool
->pmd
) < 0)
1716 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1718 dm_bio_prison_destroy(pool
->prison
);
1719 dm_kcopyd_client_destroy(pool
->copier
);
1722 destroy_workqueue(pool
->wq
);
1724 if (pool
->next_mapping
)
1725 mempool_free(pool
->next_mapping
, pool
->mapping_pool
);
1726 mempool_destroy(pool
->mapping_pool
);
1727 dm_deferred_set_destroy(pool
->shared_read_ds
);
1728 dm_deferred_set_destroy(pool
->all_io_ds
);
1732 static struct kmem_cache
*_new_mapping_cache
;
1734 static struct pool
*pool_create(struct mapped_device
*pool_md
,
1735 struct block_device
*metadata_dev
,
1736 unsigned long block_size
,
1737 int read_only
, char **error
)
1742 struct dm_pool_metadata
*pmd
;
1743 bool format_device
= read_only
? false : true;
1745 pmd
= dm_pool_metadata_open(metadata_dev
, block_size
, format_device
);
1747 *error
= "Error creating metadata object";
1748 return (struct pool
*)pmd
;
1751 pool
= kmalloc(sizeof(*pool
), GFP_KERNEL
);
1753 *error
= "Error allocating memory for pool";
1754 err_p
= ERR_PTR(-ENOMEM
);
1759 pool
->sectors_per_block
= block_size
;
1760 if (block_size
& (block_size
- 1))
1761 pool
->sectors_per_block_shift
= -1;
1763 pool
->sectors_per_block_shift
= __ffs(block_size
);
1764 pool
->low_water_blocks
= 0;
1765 pool_features_init(&pool
->pf
);
1766 pool
->prison
= dm_bio_prison_create(PRISON_CELLS
);
1767 if (!pool
->prison
) {
1768 *error
= "Error creating pool's bio prison";
1769 err_p
= ERR_PTR(-ENOMEM
);
1773 pool
->copier
= dm_kcopyd_client_create(&dm_kcopyd_throttle
);
1774 if (IS_ERR(pool
->copier
)) {
1775 r
= PTR_ERR(pool
->copier
);
1776 *error
= "Error creating pool's kcopyd client";
1778 goto bad_kcopyd_client
;
1782 * Create singlethreaded workqueue that will service all devices
1783 * that use this metadata.
1785 pool
->wq
= alloc_ordered_workqueue("dm-" DM_MSG_PREFIX
, WQ_MEM_RECLAIM
);
1787 *error
= "Error creating pool's workqueue";
1788 err_p
= ERR_PTR(-ENOMEM
);
1792 INIT_WORK(&pool
->worker
, do_worker
);
1793 INIT_DELAYED_WORK(&pool
->waker
, do_waker
);
1794 spin_lock_init(&pool
->lock
);
1795 bio_list_init(&pool
->deferred_bios
);
1796 bio_list_init(&pool
->deferred_flush_bios
);
1797 INIT_LIST_HEAD(&pool
->prepared_mappings
);
1798 INIT_LIST_HEAD(&pool
->prepared_discards
);
1799 pool
->low_water_triggered
= false;
1800 pool
->no_free_space
= false;
1801 bio_list_init(&pool
->retry_on_resume_list
);
1803 pool
->shared_read_ds
= dm_deferred_set_create();
1804 if (!pool
->shared_read_ds
) {
1805 *error
= "Error creating pool's shared read deferred set";
1806 err_p
= ERR_PTR(-ENOMEM
);
1807 goto bad_shared_read_ds
;
1810 pool
->all_io_ds
= dm_deferred_set_create();
1811 if (!pool
->all_io_ds
) {
1812 *error
= "Error creating pool's all io deferred set";
1813 err_p
= ERR_PTR(-ENOMEM
);
1817 pool
->next_mapping
= NULL
;
1818 pool
->mapping_pool
= mempool_create_slab_pool(MAPPING_POOL_SIZE
,
1819 _new_mapping_cache
);
1820 if (!pool
->mapping_pool
) {
1821 *error
= "Error creating pool's mapping mempool";
1822 err_p
= ERR_PTR(-ENOMEM
);
1823 goto bad_mapping_pool
;
1826 pool
->ref_count
= 1;
1827 pool
->last_commit_jiffies
= jiffies
;
1828 pool
->pool_md
= pool_md
;
1829 pool
->md_dev
= metadata_dev
;
1830 __pool_table_insert(pool
);
1835 dm_deferred_set_destroy(pool
->all_io_ds
);
1837 dm_deferred_set_destroy(pool
->shared_read_ds
);
1839 destroy_workqueue(pool
->wq
);
1841 dm_kcopyd_client_destroy(pool
->copier
);
1843 dm_bio_prison_destroy(pool
->prison
);
1847 if (dm_pool_metadata_close(pmd
))
1848 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1853 static void __pool_inc(struct pool
*pool
)
1855 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
1859 static void __pool_dec(struct pool
*pool
)
1861 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
1862 BUG_ON(!pool
->ref_count
);
1863 if (!--pool
->ref_count
)
1864 __pool_destroy(pool
);
1867 static struct pool
*__pool_find(struct mapped_device
*pool_md
,
1868 struct block_device
*metadata_dev
,
1869 unsigned long block_size
, int read_only
,
1870 char **error
, int *created
)
1872 struct pool
*pool
= __pool_table_lookup_metadata_dev(metadata_dev
);
1875 if (pool
->pool_md
!= pool_md
) {
1876 *error
= "metadata device already in use by a pool";
1877 return ERR_PTR(-EBUSY
);
1882 pool
= __pool_table_lookup(pool_md
);
1884 if (pool
->md_dev
!= metadata_dev
) {
1885 *error
= "different pool cannot replace a pool";
1886 return ERR_PTR(-EINVAL
);
1891 pool
= pool_create(pool_md
, metadata_dev
, block_size
, read_only
, error
);
1899 /*----------------------------------------------------------------
1900 * Pool target methods
1901 *--------------------------------------------------------------*/
1902 static void pool_dtr(struct dm_target
*ti
)
1904 struct pool_c
*pt
= ti
->private;
1906 mutex_lock(&dm_thin_pool_table
.mutex
);
1908 unbind_control_target(pt
->pool
, ti
);
1909 __pool_dec(pt
->pool
);
1910 dm_put_device(ti
, pt
->metadata_dev
);
1911 dm_put_device(ti
, pt
->data_dev
);
1914 mutex_unlock(&dm_thin_pool_table
.mutex
);
1917 static int parse_pool_features(struct dm_arg_set
*as
, struct pool_features
*pf
,
1918 struct dm_target
*ti
)
1922 const char *arg_name
;
1924 static struct dm_arg _args
[] = {
1925 {0, 3, "Invalid number of pool feature arguments"},
1929 * No feature arguments supplied.
1934 r
= dm_read_arg_group(_args
, as
, &argc
, &ti
->error
);
1938 while (argc
&& !r
) {
1939 arg_name
= dm_shift_arg(as
);
1942 if (!strcasecmp(arg_name
, "skip_block_zeroing"))
1943 pf
->zero_new_blocks
= false;
1945 else if (!strcasecmp(arg_name
, "ignore_discard"))
1946 pf
->discard_enabled
= false;
1948 else if (!strcasecmp(arg_name
, "no_discard_passdown"))
1949 pf
->discard_passdown
= false;
1951 else if (!strcasecmp(arg_name
, "read_only"))
1952 pf
->mode
= PM_READ_ONLY
;
1955 ti
->error
= "Unrecognised pool feature requested";
1964 static void metadata_low_callback(void *context
)
1966 struct pool
*pool
= context
;
1968 DMWARN("%s: reached low water mark for metadata device: sending event.",
1969 dm_device_name(pool
->pool_md
));
1971 dm_table_event(pool
->ti
->table
);
1974 static sector_t
get_metadata_dev_size(struct block_device
*bdev
)
1976 sector_t metadata_dev_size
= i_size_read(bdev
->bd_inode
) >> SECTOR_SHIFT
;
1977 char buffer
[BDEVNAME_SIZE
];
1979 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS_WARNING
) {
1980 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
1981 bdevname(bdev
, buffer
), THIN_METADATA_MAX_SECTORS
);
1982 metadata_dev_size
= THIN_METADATA_MAX_SECTORS_WARNING
;
1985 return metadata_dev_size
;
1988 static dm_block_t
get_metadata_dev_size_in_blocks(struct block_device
*bdev
)
1990 sector_t metadata_dev_size
= get_metadata_dev_size(bdev
);
1992 sector_div(metadata_dev_size
, THIN_METADATA_BLOCK_SIZE
>> SECTOR_SHIFT
);
1994 return metadata_dev_size
;
1998 * When a metadata threshold is crossed a dm event is triggered, and
1999 * userland should respond by growing the metadata device. We could let
2000 * userland set the threshold, like we do with the data threshold, but I'm
2001 * not sure they know enough to do this well.
2003 static dm_block_t
calc_metadata_threshold(struct pool_c
*pt
)
2006 * 4M is ample for all ops with the possible exception of thin
2007 * device deletion which is harmless if it fails (just retry the
2008 * delete after you've grown the device).
2010 dm_block_t quarter
= get_metadata_dev_size_in_blocks(pt
->metadata_dev
->bdev
) / 4;
2011 return min((dm_block_t
)1024ULL /* 4M */, quarter
);
2015 * thin-pool <metadata dev> <data dev>
2016 * <data block size (sectors)>
2017 * <low water mark (blocks)>
2018 * [<#feature args> [<arg>]*]
2020 * Optional feature arguments are:
2021 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
2022 * ignore_discard: disable discard
2023 * no_discard_passdown: don't pass discards down to the data device
2025 static int pool_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2027 int r
, pool_created
= 0;
2030 struct pool_features pf
;
2031 struct dm_arg_set as
;
2032 struct dm_dev
*data_dev
;
2033 unsigned long block_size
;
2034 dm_block_t low_water_blocks
;
2035 struct dm_dev
*metadata_dev
;
2036 fmode_t metadata_mode
;
2039 * FIXME Remove validation from scope of lock.
2041 mutex_lock(&dm_thin_pool_table
.mutex
);
2044 ti
->error
= "Invalid argument count";
2053 * Set default pool features.
2055 pool_features_init(&pf
);
2057 dm_consume_args(&as
, 4);
2058 r
= parse_pool_features(&as
, &pf
, ti
);
2062 metadata_mode
= FMODE_READ
| ((pf
.mode
== PM_READ_ONLY
) ? 0 : FMODE_WRITE
);
2063 r
= dm_get_device(ti
, argv
[0], metadata_mode
, &metadata_dev
);
2065 ti
->error
= "Error opening metadata block device";
2070 * Run for the side-effect of possibly issuing a warning if the
2071 * device is too big.
2073 (void) get_metadata_dev_size(metadata_dev
->bdev
);
2075 r
= dm_get_device(ti
, argv
[1], FMODE_READ
| FMODE_WRITE
, &data_dev
);
2077 ti
->error
= "Error getting data device";
2081 if (kstrtoul(argv
[2], 10, &block_size
) || !block_size
||
2082 block_size
< DATA_DEV_BLOCK_SIZE_MIN_SECTORS
||
2083 block_size
> DATA_DEV_BLOCK_SIZE_MAX_SECTORS
||
2084 block_size
& (DATA_DEV_BLOCK_SIZE_MIN_SECTORS
- 1)) {
2085 ti
->error
= "Invalid block size";
2090 if (kstrtoull(argv
[3], 10, (unsigned long long *)&low_water_blocks
)) {
2091 ti
->error
= "Invalid low water mark";
2096 pt
= kzalloc(sizeof(*pt
), GFP_KERNEL
);
2102 pool
= __pool_find(dm_table_get_md(ti
->table
), metadata_dev
->bdev
,
2103 block_size
, pf
.mode
== PM_READ_ONLY
, &ti
->error
, &pool_created
);
2110 * 'pool_created' reflects whether this is the first table load.
2111 * Top level discard support is not allowed to be changed after
2112 * initial load. This would require a pool reload to trigger thin
2115 if (!pool_created
&& pf
.discard_enabled
!= pool
->pf
.discard_enabled
) {
2116 ti
->error
= "Discard support cannot be disabled once enabled";
2118 goto out_flags_changed
;
2123 pt
->metadata_dev
= metadata_dev
;
2124 pt
->data_dev
= data_dev
;
2125 pt
->low_water_blocks
= low_water_blocks
;
2126 pt
->adjusted_pf
= pt
->requested_pf
= pf
;
2127 ti
->num_flush_bios
= 1;
2130 * Only need to enable discards if the pool should pass
2131 * them down to the data device. The thin device's discard
2132 * processing will cause mappings to be removed from the btree.
2134 ti
->discard_zeroes_data_unsupported
= true;
2135 if (pf
.discard_enabled
&& pf
.discard_passdown
) {
2136 ti
->num_discard_bios
= 1;
2139 * Setting 'discards_supported' circumvents the normal
2140 * stacking of discard limits (this keeps the pool and
2141 * thin devices' discard limits consistent).
2143 ti
->discards_supported
= true;
2147 r
= dm_pool_register_metadata_threshold(pt
->pool
->pmd
,
2148 calc_metadata_threshold(pt
),
2149 metadata_low_callback
,
2154 pt
->callbacks
.congested_fn
= pool_is_congested
;
2155 dm_table_add_target_callbacks(ti
->table
, &pt
->callbacks
);
2157 mutex_unlock(&dm_thin_pool_table
.mutex
);
2166 dm_put_device(ti
, data_dev
);
2168 dm_put_device(ti
, metadata_dev
);
2170 mutex_unlock(&dm_thin_pool_table
.mutex
);
2175 static int pool_map(struct dm_target
*ti
, struct bio
*bio
)
2178 struct pool_c
*pt
= ti
->private;
2179 struct pool
*pool
= pt
->pool
;
2180 unsigned long flags
;
2183 * As this is a singleton target, ti->begin is always zero.
2185 spin_lock_irqsave(&pool
->lock
, flags
);
2186 bio
->bi_bdev
= pt
->data_dev
->bdev
;
2187 r
= DM_MAPIO_REMAPPED
;
2188 spin_unlock_irqrestore(&pool
->lock
, flags
);
2193 static int maybe_resize_data_dev(struct dm_target
*ti
, bool *need_commit
)
2196 struct pool_c
*pt
= ti
->private;
2197 struct pool
*pool
= pt
->pool
;
2198 sector_t data_size
= ti
->len
;
2199 dm_block_t sb_data_size
;
2201 *need_commit
= false;
2203 (void) sector_div(data_size
, pool
->sectors_per_block
);
2205 r
= dm_pool_get_data_dev_size(pool
->pmd
, &sb_data_size
);
2207 DMERR("%s: failed to retrieve data device size",
2208 dm_device_name(pool
->pool_md
));
2212 if (data_size
< sb_data_size
) {
2213 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
2214 dm_device_name(pool
->pool_md
),
2215 (unsigned long long)data_size
, sb_data_size
);
2218 } else if (data_size
> sb_data_size
) {
2219 r
= dm_pool_resize_data_dev(pool
->pmd
, data_size
);
2221 metadata_operation_failed(pool
, "dm_pool_resize_data_dev", r
);
2225 *need_commit
= true;
2231 static int maybe_resize_metadata_dev(struct dm_target
*ti
, bool *need_commit
)
2234 struct pool_c
*pt
= ti
->private;
2235 struct pool
*pool
= pt
->pool
;
2236 dm_block_t metadata_dev_size
, sb_metadata_dev_size
;
2238 *need_commit
= false;
2240 metadata_dev_size
= get_metadata_dev_size_in_blocks(pool
->md_dev
);
2242 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &sb_metadata_dev_size
);
2244 DMERR("%s: failed to retrieve metadata device size",
2245 dm_device_name(pool
->pool_md
));
2249 if (metadata_dev_size
< sb_metadata_dev_size
) {
2250 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
2251 dm_device_name(pool
->pool_md
),
2252 metadata_dev_size
, sb_metadata_dev_size
);
2255 } else if (metadata_dev_size
> sb_metadata_dev_size
) {
2256 r
= dm_pool_resize_metadata_dev(pool
->pmd
, metadata_dev_size
);
2258 metadata_operation_failed(pool
, "dm_pool_resize_metadata_dev", r
);
2262 *need_commit
= true;
2269 * Retrieves the number of blocks of the data device from
2270 * the superblock and compares it to the actual device size,
2271 * thus resizing the data device in case it has grown.
2273 * This both copes with opening preallocated data devices in the ctr
2274 * being followed by a resume
2276 * calling the resume method individually after userspace has
2277 * grown the data device in reaction to a table event.
2279 static int pool_preresume(struct dm_target
*ti
)
2282 bool need_commit1
, need_commit2
;
2283 struct pool_c
*pt
= ti
->private;
2284 struct pool
*pool
= pt
->pool
;
2287 * Take control of the pool object.
2289 r
= bind_control_target(pool
, ti
);
2293 r
= maybe_resize_data_dev(ti
, &need_commit1
);
2297 r
= maybe_resize_metadata_dev(ti
, &need_commit2
);
2301 if (need_commit1
|| need_commit2
)
2302 (void) commit(pool
);
2307 static void pool_resume(struct dm_target
*ti
)
2309 struct pool_c
*pt
= ti
->private;
2310 struct pool
*pool
= pt
->pool
;
2311 unsigned long flags
;
2313 spin_lock_irqsave(&pool
->lock
, flags
);
2314 pool
->low_water_triggered
= false;
2315 pool
->no_free_space
= false;
2316 __requeue_bios(pool
);
2317 spin_unlock_irqrestore(&pool
->lock
, flags
);
2319 do_waker(&pool
->waker
.work
);
2322 static void pool_postsuspend(struct dm_target
*ti
)
2324 struct pool_c
*pt
= ti
->private;
2325 struct pool
*pool
= pt
->pool
;
2327 cancel_delayed_work(&pool
->waker
);
2328 flush_workqueue(pool
->wq
);
2329 (void) commit(pool
);
2332 static int check_arg_count(unsigned argc
, unsigned args_required
)
2334 if (argc
!= args_required
) {
2335 DMWARN("Message received with %u arguments instead of %u.",
2336 argc
, args_required
);
2343 static int read_dev_id(char *arg
, dm_thin_id
*dev_id
, int warning
)
2345 if (!kstrtoull(arg
, 10, (unsigned long long *)dev_id
) &&
2346 *dev_id
<= MAX_DEV_ID
)
2350 DMWARN("Message received with invalid device id: %s", arg
);
2355 static int process_create_thin_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2360 r
= check_arg_count(argc
, 2);
2364 r
= read_dev_id(argv
[1], &dev_id
, 1);
2368 r
= dm_pool_create_thin(pool
->pmd
, dev_id
);
2370 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2378 static int process_create_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2381 dm_thin_id origin_dev_id
;
2384 r
= check_arg_count(argc
, 3);
2388 r
= read_dev_id(argv
[1], &dev_id
, 1);
2392 r
= read_dev_id(argv
[2], &origin_dev_id
, 1);
2396 r
= dm_pool_create_snap(pool
->pmd
, dev_id
, origin_dev_id
);
2398 DMWARN("Creation of new snapshot %s of device %s failed.",
2406 static int process_delete_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2411 r
= check_arg_count(argc
, 2);
2415 r
= read_dev_id(argv
[1], &dev_id
, 1);
2419 r
= dm_pool_delete_thin_device(pool
->pmd
, dev_id
);
2421 DMWARN("Deletion of thin device %s failed.", argv
[1]);
2426 static int process_set_transaction_id_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2428 dm_thin_id old_id
, new_id
;
2431 r
= check_arg_count(argc
, 3);
2435 if (kstrtoull(argv
[1], 10, (unsigned long long *)&old_id
)) {
2436 DMWARN("set_transaction_id message: Unrecognised id %s.", argv
[1]);
2440 if (kstrtoull(argv
[2], 10, (unsigned long long *)&new_id
)) {
2441 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv
[2]);
2445 r
= dm_pool_set_metadata_transaction_id(pool
->pmd
, old_id
, new_id
);
2447 DMWARN("Failed to change transaction id from %s to %s.",
2455 static int process_reserve_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2459 r
= check_arg_count(argc
, 1);
2463 (void) commit(pool
);
2465 r
= dm_pool_reserve_metadata_snap(pool
->pmd
);
2467 DMWARN("reserve_metadata_snap message failed.");
2472 static int process_release_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2476 r
= check_arg_count(argc
, 1);
2480 r
= dm_pool_release_metadata_snap(pool
->pmd
);
2482 DMWARN("release_metadata_snap message failed.");
2488 * Messages supported:
2489 * create_thin <dev_id>
2490 * create_snap <dev_id> <origin_id>
2492 * trim <dev_id> <new_size_in_sectors>
2493 * set_transaction_id <current_trans_id> <new_trans_id>
2494 * reserve_metadata_snap
2495 * release_metadata_snap
2497 static int pool_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
2500 struct pool_c
*pt
= ti
->private;
2501 struct pool
*pool
= pt
->pool
;
2503 if (!strcasecmp(argv
[0], "create_thin"))
2504 r
= process_create_thin_mesg(argc
, argv
, pool
);
2506 else if (!strcasecmp(argv
[0], "create_snap"))
2507 r
= process_create_snap_mesg(argc
, argv
, pool
);
2509 else if (!strcasecmp(argv
[0], "delete"))
2510 r
= process_delete_mesg(argc
, argv
, pool
);
2512 else if (!strcasecmp(argv
[0], "set_transaction_id"))
2513 r
= process_set_transaction_id_mesg(argc
, argv
, pool
);
2515 else if (!strcasecmp(argv
[0], "reserve_metadata_snap"))
2516 r
= process_reserve_metadata_snap_mesg(argc
, argv
, pool
);
2518 else if (!strcasecmp(argv
[0], "release_metadata_snap"))
2519 r
= process_release_metadata_snap_mesg(argc
, argv
, pool
);
2522 DMWARN("Unrecognised thin pool target message received: %s", argv
[0]);
2525 (void) commit(pool
);
2530 static void emit_flags(struct pool_features
*pf
, char *result
,
2531 unsigned sz
, unsigned maxlen
)
2533 unsigned count
= !pf
->zero_new_blocks
+ !pf
->discard_enabled
+
2534 !pf
->discard_passdown
+ (pf
->mode
== PM_READ_ONLY
);
2535 DMEMIT("%u ", count
);
2537 if (!pf
->zero_new_blocks
)
2538 DMEMIT("skip_block_zeroing ");
2540 if (!pf
->discard_enabled
)
2541 DMEMIT("ignore_discard ");
2543 if (!pf
->discard_passdown
)
2544 DMEMIT("no_discard_passdown ");
2546 if (pf
->mode
== PM_READ_ONLY
)
2547 DMEMIT("read_only ");
2552 * <transaction id> <used metadata sectors>/<total metadata sectors>
2553 * <used data sectors>/<total data sectors> <held metadata root>
2555 static void pool_status(struct dm_target
*ti
, status_type_t type
,
2556 unsigned status_flags
, char *result
, unsigned maxlen
)
2560 uint64_t transaction_id
;
2561 dm_block_t nr_free_blocks_data
;
2562 dm_block_t nr_free_blocks_metadata
;
2563 dm_block_t nr_blocks_data
;
2564 dm_block_t nr_blocks_metadata
;
2565 dm_block_t held_root
;
2566 char buf
[BDEVNAME_SIZE
];
2567 char buf2
[BDEVNAME_SIZE
];
2568 struct pool_c
*pt
= ti
->private;
2569 struct pool
*pool
= pt
->pool
;
2572 case STATUSTYPE_INFO
:
2573 if (get_pool_mode(pool
) == PM_FAIL
) {
2578 /* Commit to ensure statistics aren't out-of-date */
2579 if (!(status_flags
& DM_STATUS_NOFLUSH_FLAG
) && !dm_suspended(ti
))
2580 (void) commit(pool
);
2582 r
= dm_pool_get_metadata_transaction_id(pool
->pmd
, &transaction_id
);
2584 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
2585 dm_device_name(pool
->pool_md
), r
);
2589 r
= dm_pool_get_free_metadata_block_count(pool
->pmd
, &nr_free_blocks_metadata
);
2591 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
2592 dm_device_name(pool
->pool_md
), r
);
2596 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &nr_blocks_metadata
);
2598 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
2599 dm_device_name(pool
->pool_md
), r
);
2603 r
= dm_pool_get_free_block_count(pool
->pmd
, &nr_free_blocks_data
);
2605 DMERR("%s: dm_pool_get_free_block_count returned %d",
2606 dm_device_name(pool
->pool_md
), r
);
2610 r
= dm_pool_get_data_dev_size(pool
->pmd
, &nr_blocks_data
);
2612 DMERR("%s: dm_pool_get_data_dev_size returned %d",
2613 dm_device_name(pool
->pool_md
), r
);
2617 r
= dm_pool_get_metadata_snap(pool
->pmd
, &held_root
);
2619 DMERR("%s: dm_pool_get_metadata_snap returned %d",
2620 dm_device_name(pool
->pool_md
), r
);
2624 DMEMIT("%llu %llu/%llu %llu/%llu ",
2625 (unsigned long long)transaction_id
,
2626 (unsigned long long)(nr_blocks_metadata
- nr_free_blocks_metadata
),
2627 (unsigned long long)nr_blocks_metadata
,
2628 (unsigned long long)(nr_blocks_data
- nr_free_blocks_data
),
2629 (unsigned long long)nr_blocks_data
);
2632 DMEMIT("%llu ", held_root
);
2636 if (pool
->pf
.mode
== PM_READ_ONLY
)
2641 if (!pool
->pf
.discard_enabled
)
2642 DMEMIT("ignore_discard");
2643 else if (pool
->pf
.discard_passdown
)
2644 DMEMIT("discard_passdown");
2646 DMEMIT("no_discard_passdown");
2650 case STATUSTYPE_TABLE
:
2651 DMEMIT("%s %s %lu %llu ",
2652 format_dev_t(buf
, pt
->metadata_dev
->bdev
->bd_dev
),
2653 format_dev_t(buf2
, pt
->data_dev
->bdev
->bd_dev
),
2654 (unsigned long)pool
->sectors_per_block
,
2655 (unsigned long long)pt
->low_water_blocks
);
2656 emit_flags(&pt
->requested_pf
, result
, sz
, maxlen
);
2665 static int pool_iterate_devices(struct dm_target
*ti
,
2666 iterate_devices_callout_fn fn
, void *data
)
2668 struct pool_c
*pt
= ti
->private;
2670 return fn(ti
, pt
->data_dev
, 0, ti
->len
, data
);
2673 static int pool_merge(struct dm_target
*ti
, struct bvec_merge_data
*bvm
,
2674 struct bio_vec
*biovec
, int max_size
)
2676 struct pool_c
*pt
= ti
->private;
2677 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
2679 if (!q
->merge_bvec_fn
)
2682 bvm
->bi_bdev
= pt
->data_dev
->bdev
;
2684 return min(max_size
, q
->merge_bvec_fn(q
, bvm
, biovec
));
2687 static void set_discard_limits(struct pool_c
*pt
, struct queue_limits
*limits
)
2689 struct pool
*pool
= pt
->pool
;
2690 struct queue_limits
*data_limits
;
2692 limits
->max_discard_sectors
= pool
->sectors_per_block
;
2695 * discard_granularity is just a hint, and not enforced.
2697 if (pt
->adjusted_pf
.discard_passdown
) {
2698 data_limits
= &bdev_get_queue(pt
->data_dev
->bdev
)->limits
;
2699 limits
->discard_granularity
= data_limits
->discard_granularity
;
2701 limits
->discard_granularity
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
2704 static void pool_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
2706 struct pool_c
*pt
= ti
->private;
2707 struct pool
*pool
= pt
->pool
;
2708 uint64_t io_opt_sectors
= limits
->io_opt
>> SECTOR_SHIFT
;
2711 * If the system-determined stacked limits are compatible with the
2712 * pool's blocksize (io_opt is a factor) do not override them.
2714 if (io_opt_sectors
< pool
->sectors_per_block
||
2715 do_div(io_opt_sectors
, pool
->sectors_per_block
)) {
2716 blk_limits_io_min(limits
, 0);
2717 blk_limits_io_opt(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
2721 * pt->adjusted_pf is a staging area for the actual features to use.
2722 * They get transferred to the live pool in bind_control_target()
2723 * called from pool_preresume().
2725 if (!pt
->adjusted_pf
.discard_enabled
) {
2727 * Must explicitly disallow stacking discard limits otherwise the
2728 * block layer will stack them if pool's data device has support.
2729 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
2730 * user to see that, so make sure to set all discard limits to 0.
2732 limits
->discard_granularity
= 0;
2736 disable_passdown_if_not_supported(pt
);
2738 set_discard_limits(pt
, limits
);
2741 static struct target_type pool_target
= {
2742 .name
= "thin-pool",
2743 .features
= DM_TARGET_SINGLETON
| DM_TARGET_ALWAYS_WRITEABLE
|
2744 DM_TARGET_IMMUTABLE
,
2745 .version
= {1, 9, 0},
2746 .module
= THIS_MODULE
,
2750 .postsuspend
= pool_postsuspend
,
2751 .preresume
= pool_preresume
,
2752 .resume
= pool_resume
,
2753 .message
= pool_message
,
2754 .status
= pool_status
,
2755 .merge
= pool_merge
,
2756 .iterate_devices
= pool_iterate_devices
,
2757 .io_hints
= pool_io_hints
,
2760 /*----------------------------------------------------------------
2761 * Thin target methods
2762 *--------------------------------------------------------------*/
2763 static void thin_dtr(struct dm_target
*ti
)
2765 struct thin_c
*tc
= ti
->private;
2767 mutex_lock(&dm_thin_pool_table
.mutex
);
2769 __pool_dec(tc
->pool
);
2770 dm_pool_close_thin_device(tc
->td
);
2771 dm_put_device(ti
, tc
->pool_dev
);
2773 dm_put_device(ti
, tc
->origin_dev
);
2776 mutex_unlock(&dm_thin_pool_table
.mutex
);
2780 * Thin target parameters:
2782 * <pool_dev> <dev_id> [origin_dev]
2784 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2785 * dev_id: the internal device identifier
2786 * origin_dev: a device external to the pool that should act as the origin
2788 * If the pool device has discards disabled, they get disabled for the thin
2791 static int thin_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2795 struct dm_dev
*pool_dev
, *origin_dev
;
2796 struct mapped_device
*pool_md
;
2798 mutex_lock(&dm_thin_pool_table
.mutex
);
2800 if (argc
!= 2 && argc
!= 3) {
2801 ti
->error
= "Invalid argument count";
2806 tc
= ti
->private = kzalloc(sizeof(*tc
), GFP_KERNEL
);
2808 ti
->error
= "Out of memory";
2814 r
= dm_get_device(ti
, argv
[2], FMODE_READ
, &origin_dev
);
2816 ti
->error
= "Error opening origin device";
2817 goto bad_origin_dev
;
2819 tc
->origin_dev
= origin_dev
;
2822 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &pool_dev
);
2824 ti
->error
= "Error opening pool device";
2827 tc
->pool_dev
= pool_dev
;
2829 if (read_dev_id(argv
[1], (unsigned long long *)&tc
->dev_id
, 0)) {
2830 ti
->error
= "Invalid device id";
2835 pool_md
= dm_get_md(tc
->pool_dev
->bdev
->bd_dev
);
2837 ti
->error
= "Couldn't get pool mapped device";
2842 tc
->pool
= __pool_table_lookup(pool_md
);
2844 ti
->error
= "Couldn't find pool object";
2846 goto bad_pool_lookup
;
2848 __pool_inc(tc
->pool
);
2850 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
2851 ti
->error
= "Couldn't open thin device, Pool is in fail mode";
2855 r
= dm_pool_open_thin_device(tc
->pool
->pmd
, tc
->dev_id
, &tc
->td
);
2857 ti
->error
= "Couldn't open thin internal device";
2861 r
= dm_set_target_max_io_len(ti
, tc
->pool
->sectors_per_block
);
2865 ti
->num_flush_bios
= 1;
2866 ti
->flush_supported
= true;
2867 ti
->per_bio_data_size
= sizeof(struct dm_thin_endio_hook
);
2869 /* In case the pool supports discards, pass them on. */
2870 ti
->discard_zeroes_data_unsupported
= true;
2871 if (tc
->pool
->pf
.discard_enabled
) {
2872 ti
->discards_supported
= true;
2873 ti
->num_discard_bios
= 1;
2874 /* Discard bios must be split on a block boundary */
2875 ti
->split_discard_bios
= true;
2880 mutex_unlock(&dm_thin_pool_table
.mutex
);
2885 __pool_dec(tc
->pool
);
2889 dm_put_device(ti
, tc
->pool_dev
);
2892 dm_put_device(ti
, tc
->origin_dev
);
2896 mutex_unlock(&dm_thin_pool_table
.mutex
);
2901 static int thin_map(struct dm_target
*ti
, struct bio
*bio
)
2903 bio
->bi_sector
= dm_target_offset(ti
, bio
->bi_sector
);
2905 return thin_bio_map(ti
, bio
);
2908 static int thin_endio(struct dm_target
*ti
, struct bio
*bio
, int err
)
2910 unsigned long flags
;
2911 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
2912 struct list_head work
;
2913 struct dm_thin_new_mapping
*m
, *tmp
;
2914 struct pool
*pool
= h
->tc
->pool
;
2916 if (h
->shared_read_entry
) {
2917 INIT_LIST_HEAD(&work
);
2918 dm_deferred_entry_dec(h
->shared_read_entry
, &work
);
2920 spin_lock_irqsave(&pool
->lock
, flags
);
2921 list_for_each_entry_safe(m
, tmp
, &work
, list
) {
2924 __maybe_add_mapping(m
);
2926 spin_unlock_irqrestore(&pool
->lock
, flags
);
2929 if (h
->all_io_entry
) {
2930 INIT_LIST_HEAD(&work
);
2931 dm_deferred_entry_dec(h
->all_io_entry
, &work
);
2932 if (!list_empty(&work
)) {
2933 spin_lock_irqsave(&pool
->lock
, flags
);
2934 list_for_each_entry_safe(m
, tmp
, &work
, list
)
2935 list_add_tail(&m
->list
, &pool
->prepared_discards
);
2936 spin_unlock_irqrestore(&pool
->lock
, flags
);
2944 static void thin_postsuspend(struct dm_target
*ti
)
2946 if (dm_noflush_suspending(ti
))
2947 requeue_io((struct thin_c
*)ti
->private);
2951 * <nr mapped sectors> <highest mapped sector>
2953 static void thin_status(struct dm_target
*ti
, status_type_t type
,
2954 unsigned status_flags
, char *result
, unsigned maxlen
)
2958 dm_block_t mapped
, highest
;
2959 char buf
[BDEVNAME_SIZE
];
2960 struct thin_c
*tc
= ti
->private;
2962 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
2971 case STATUSTYPE_INFO
:
2972 r
= dm_thin_get_mapped_count(tc
->td
, &mapped
);
2974 DMERR("dm_thin_get_mapped_count returned %d", r
);
2978 r
= dm_thin_get_highest_mapped_block(tc
->td
, &highest
);
2980 DMERR("dm_thin_get_highest_mapped_block returned %d", r
);
2984 DMEMIT("%llu ", mapped
* tc
->pool
->sectors_per_block
);
2986 DMEMIT("%llu", ((highest
+ 1) *
2987 tc
->pool
->sectors_per_block
) - 1);
2992 case STATUSTYPE_TABLE
:
2994 format_dev_t(buf
, tc
->pool_dev
->bdev
->bd_dev
),
2995 (unsigned long) tc
->dev_id
);
2997 DMEMIT(" %s", format_dev_t(buf
, tc
->origin_dev
->bdev
->bd_dev
));
3008 static int thin_iterate_devices(struct dm_target
*ti
,
3009 iterate_devices_callout_fn fn
, void *data
)
3012 struct thin_c
*tc
= ti
->private;
3013 struct pool
*pool
= tc
->pool
;
3016 * We can't call dm_pool_get_data_dev_size() since that blocks. So
3017 * we follow a more convoluted path through to the pool's target.
3020 return 0; /* nothing is bound */
3022 blocks
= pool
->ti
->len
;
3023 (void) sector_div(blocks
, pool
->sectors_per_block
);
3025 return fn(ti
, tc
->pool_dev
, 0, pool
->sectors_per_block
* blocks
, data
);
3030 static struct target_type thin_target
= {
3032 .version
= {1, 9, 0},
3033 .module
= THIS_MODULE
,
3037 .end_io
= thin_endio
,
3038 .postsuspend
= thin_postsuspend
,
3039 .status
= thin_status
,
3040 .iterate_devices
= thin_iterate_devices
,
3043 /*----------------------------------------------------------------*/
3045 static int __init
dm_thin_init(void)
3051 r
= dm_register_target(&thin_target
);
3055 r
= dm_register_target(&pool_target
);
3057 goto bad_pool_target
;
3061 _new_mapping_cache
= KMEM_CACHE(dm_thin_new_mapping
, 0);
3062 if (!_new_mapping_cache
)
3063 goto bad_new_mapping_cache
;
3067 bad_new_mapping_cache
:
3068 dm_unregister_target(&pool_target
);
3070 dm_unregister_target(&thin_target
);
3075 static void dm_thin_exit(void)
3077 dm_unregister_target(&thin_target
);
3078 dm_unregister_target(&pool_target
);
3080 kmem_cache_destroy(_new_mapping_cache
);
3083 module_init(dm_thin_init
);
3084 module_exit(dm_thin_exit
);
3086 MODULE_DESCRIPTION(DM_NAME
" thin provisioning target");
3087 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3088 MODULE_LICENSE("GPL");