2 * Copyright (C) 2011-2012 Red Hat UK.
4 * This file is released under the GPL.
7 #include "dm-thin-metadata.h"
8 #include "dm-bio-prison.h"
11 #include <linux/device-mapper.h>
12 #include <linux/dm-io.h>
13 #include <linux/dm-kcopyd.h>
14 #include <linux/list.h>
15 #include <linux/init.h>
16 #include <linux/module.h>
17 #include <linux/slab.h>
19 #define DM_MSG_PREFIX "thin"
24 #define ENDIO_HOOK_POOL_SIZE 1024
25 #define MAPPING_POOL_SIZE 1024
26 #define PRISON_CELLS 1024
27 #define COMMIT_PERIOD HZ
29 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle
,
30 "A percentage of time allocated for copy on write");
33 * The block size of the device holding pool data must be
34 * between 64KB and 1GB.
36 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
37 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
40 * Device id is restricted to 24 bits.
42 #define MAX_DEV_ID ((1 << 24) - 1)
45 * How do we handle breaking sharing of data blocks?
46 * =================================================
48 * We use a standard copy-on-write btree to store the mappings for the
49 * devices (note I'm talking about copy-on-write of the metadata here, not
50 * the data). When you take an internal snapshot you clone the root node
51 * of the origin btree. After this there is no concept of an origin or a
52 * snapshot. They are just two device trees that happen to point to the
55 * When we get a write in we decide if it's to a shared data block using
56 * some timestamp magic. If it is, we have to break sharing.
58 * Let's say we write to a shared block in what was the origin. The
61 * i) plug io further to this physical block. (see bio_prison code).
63 * ii) quiesce any read io to that shared data block. Obviously
64 * including all devices that share this block. (see dm_deferred_set code)
66 * iii) copy the data block to a newly allocate block. This step can be
67 * missed out if the io covers the block. (schedule_copy).
69 * iv) insert the new mapping into the origin's btree
70 * (process_prepared_mapping). This act of inserting breaks some
71 * sharing of btree nodes between the two devices. Breaking sharing only
72 * effects the btree of that specific device. Btrees for the other
73 * devices that share the block never change. The btree for the origin
74 * device as it was after the last commit is untouched, ie. we're using
75 * persistent data structures in the functional programming sense.
77 * v) unplug io to this physical block, including the io that triggered
78 * the breaking of sharing.
80 * Steps (ii) and (iii) occur in parallel.
82 * The metadata _doesn't_ need to be committed before the io continues. We
83 * get away with this because the io is always written to a _new_ block.
84 * If there's a crash, then:
86 * - The origin mapping will point to the old origin block (the shared
87 * one). This will contain the data as it was before the io that triggered
88 * the breaking of sharing came in.
90 * - The snap mapping still points to the old block. As it would after
93 * The downside of this scheme is the timestamp magic isn't perfect, and
94 * will continue to think that data block in the snapshot device is shared
95 * even after the write to the origin has broken sharing. I suspect data
96 * blocks will typically be shared by many different devices, so we're
97 * breaking sharing n + 1 times, rather than n, where n is the number of
98 * devices that reference this data block. At the moment I think the
99 * benefits far, far outweigh the disadvantages.
102 /*----------------------------------------------------------------*/
107 static void build_data_key(struct dm_thin_device
*td
,
108 dm_block_t b
, struct dm_cell_key
*key
)
111 key
->dev
= dm_thin_dev_id(td
);
115 static void build_virtual_key(struct dm_thin_device
*td
, dm_block_t b
,
116 struct dm_cell_key
*key
)
119 key
->dev
= dm_thin_dev_id(td
);
123 /*----------------------------------------------------------------*/
126 * A pool device ties together a metadata device and a data device. It
127 * also provides the interface for creating and destroying internal
130 struct dm_thin_new_mapping
;
133 * The pool runs in 3 modes. Ordered in degraded order for comparisons.
136 PM_WRITE
, /* metadata may be changed */
137 PM_READ_ONLY
, /* metadata may not be changed */
138 PM_FAIL
, /* all I/O fails */
141 struct pool_features
{
144 bool zero_new_blocks
:1;
145 bool discard_enabled
:1;
146 bool discard_passdown
:1;
150 typedef void (*process_bio_fn
)(struct thin_c
*tc
, struct bio
*bio
);
151 typedef void (*process_mapping_fn
)(struct dm_thin_new_mapping
*m
);
154 struct list_head list
;
155 struct dm_target
*ti
; /* Only set if a pool target is bound */
157 struct mapped_device
*pool_md
;
158 struct block_device
*md_dev
;
159 struct dm_pool_metadata
*pmd
;
161 dm_block_t low_water_blocks
;
162 uint32_t sectors_per_block
;
163 int sectors_per_block_shift
;
165 struct pool_features pf
;
166 unsigned low_water_triggered
:1; /* A dm event has been sent */
167 unsigned no_free_space
:1; /* A -ENOSPC warning has been issued */
169 struct dm_bio_prison
*prison
;
170 struct dm_kcopyd_client
*copier
;
172 struct workqueue_struct
*wq
;
173 struct work_struct worker
;
174 struct delayed_work waker
;
176 unsigned long last_commit_jiffies
;
180 struct bio_list deferred_bios
;
181 struct bio_list deferred_flush_bios
;
182 struct list_head prepared_mappings
;
183 struct list_head prepared_discards
;
185 struct bio_list retry_on_resume_list
;
187 struct dm_deferred_set
*shared_read_ds
;
188 struct dm_deferred_set
*all_io_ds
;
190 struct dm_thin_new_mapping
*next_mapping
;
191 mempool_t
*mapping_pool
;
193 process_bio_fn process_bio
;
194 process_bio_fn process_discard
;
196 process_mapping_fn process_prepared_mapping
;
197 process_mapping_fn process_prepared_discard
;
200 static enum pool_mode
get_pool_mode(struct pool
*pool
);
201 static void set_pool_mode(struct pool
*pool
, enum pool_mode mode
);
204 * Target context for a pool.
207 struct dm_target
*ti
;
209 struct dm_dev
*data_dev
;
210 struct dm_dev
*metadata_dev
;
211 struct dm_target_callbacks callbacks
;
213 dm_block_t low_water_blocks
;
214 struct pool_features requested_pf
; /* Features requested during table load */
215 struct pool_features adjusted_pf
; /* Features used after adjusting for constituent devices */
219 * Target context for a thin.
222 struct dm_dev
*pool_dev
;
223 struct dm_dev
*origin_dev
;
227 struct dm_thin_device
*td
;
230 /*----------------------------------------------------------------*/
233 * wake_worker() is used when new work is queued and when pool_resume is
234 * ready to continue deferred IO processing.
236 static void wake_worker(struct pool
*pool
)
238 queue_work(pool
->wq
, &pool
->worker
);
241 /*----------------------------------------------------------------*/
243 static int bio_detain(struct pool
*pool
, struct dm_cell_key
*key
, struct bio
*bio
,
244 struct dm_bio_prison_cell
**cell_result
)
247 struct dm_bio_prison_cell
*cell_prealloc
;
250 * Allocate a cell from the prison's mempool.
251 * This might block but it can't fail.
253 cell_prealloc
= dm_bio_prison_alloc_cell(pool
->prison
, GFP_NOIO
);
255 r
= dm_bio_detain(pool
->prison
, key
, bio
, cell_prealloc
, cell_result
);
258 * We reused an old cell; we can get rid of
261 dm_bio_prison_free_cell(pool
->prison
, cell_prealloc
);
266 static void cell_release(struct pool
*pool
,
267 struct dm_bio_prison_cell
*cell
,
268 struct bio_list
*bios
)
270 dm_cell_release(pool
->prison
, cell
, bios
);
271 dm_bio_prison_free_cell(pool
->prison
, cell
);
274 static void cell_release_no_holder(struct pool
*pool
,
275 struct dm_bio_prison_cell
*cell
,
276 struct bio_list
*bios
)
278 dm_cell_release_no_holder(pool
->prison
, cell
, bios
);
279 dm_bio_prison_free_cell(pool
->prison
, cell
);
282 static void cell_defer_no_holder_no_free(struct thin_c
*tc
,
283 struct dm_bio_prison_cell
*cell
)
285 struct pool
*pool
= tc
->pool
;
288 spin_lock_irqsave(&pool
->lock
, flags
);
289 dm_cell_release_no_holder(pool
->prison
, cell
, &pool
->deferred_bios
);
290 spin_unlock_irqrestore(&pool
->lock
, flags
);
295 static void cell_error(struct pool
*pool
,
296 struct dm_bio_prison_cell
*cell
)
298 dm_cell_error(pool
->prison
, cell
);
299 dm_bio_prison_free_cell(pool
->prison
, cell
);
302 /*----------------------------------------------------------------*/
305 * A global list of pools that uses a struct mapped_device as a key.
307 static struct dm_thin_pool_table
{
309 struct list_head pools
;
310 } dm_thin_pool_table
;
312 static void pool_table_init(void)
314 mutex_init(&dm_thin_pool_table
.mutex
);
315 INIT_LIST_HEAD(&dm_thin_pool_table
.pools
);
318 static void __pool_table_insert(struct pool
*pool
)
320 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
321 list_add(&pool
->list
, &dm_thin_pool_table
.pools
);
324 static void __pool_table_remove(struct pool
*pool
)
326 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
327 list_del(&pool
->list
);
330 static struct pool
*__pool_table_lookup(struct mapped_device
*md
)
332 struct pool
*pool
= NULL
, *tmp
;
334 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
336 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
337 if (tmp
->pool_md
== md
) {
346 static struct pool
*__pool_table_lookup_metadata_dev(struct block_device
*md_dev
)
348 struct pool
*pool
= NULL
, *tmp
;
350 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
352 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
353 if (tmp
->md_dev
== md_dev
) {
362 /*----------------------------------------------------------------*/
364 struct dm_thin_endio_hook
{
366 struct dm_deferred_entry
*shared_read_entry
;
367 struct dm_deferred_entry
*all_io_entry
;
368 struct dm_thin_new_mapping
*overwrite_mapping
;
371 static void __requeue_bio_list(struct thin_c
*tc
, struct bio_list
*master
)
374 struct bio_list bios
;
376 bio_list_init(&bios
);
377 bio_list_merge(&bios
, master
);
378 bio_list_init(master
);
380 while ((bio
= bio_list_pop(&bios
))) {
381 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
384 bio_endio(bio
, DM_ENDIO_REQUEUE
);
386 bio_list_add(master
, bio
);
390 static void requeue_io(struct thin_c
*tc
)
392 struct pool
*pool
= tc
->pool
;
395 spin_lock_irqsave(&pool
->lock
, flags
);
396 __requeue_bio_list(tc
, &pool
->deferred_bios
);
397 __requeue_bio_list(tc
, &pool
->retry_on_resume_list
);
398 spin_unlock_irqrestore(&pool
->lock
, flags
);
402 * This section of code contains the logic for processing a thin device's IO.
403 * Much of the code depends on pool object resources (lists, workqueues, etc)
404 * but most is exclusively called from the thin target rather than the thin-pool
408 static bool block_size_is_power_of_two(struct pool
*pool
)
410 return pool
->sectors_per_block_shift
>= 0;
413 static dm_block_t
get_bio_block(struct thin_c
*tc
, struct bio
*bio
)
415 struct pool
*pool
= tc
->pool
;
416 sector_t block_nr
= bio
->bi_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
;
514 unsigned pass_discard
:1;
517 dm_block_t virt_block
;
518 dm_block_t data_block
;
519 struct dm_bio_prison_cell
*cell
, *cell2
;
523 * If the bio covers the whole area of a block then we can avoid
524 * zeroing or copying. Instead this bio is hooked. The bio will
525 * still be in the cell, so care has to be taken to avoid issuing
529 bio_end_io_t
*saved_bi_end_io
;
532 static void __maybe_add_mapping(struct dm_thin_new_mapping
*m
)
534 struct pool
*pool
= m
->tc
->pool
;
536 if (m
->quiesced
&& m
->prepared
) {
537 list_add(&m
->list
, &pool
->prepared_mappings
);
542 static void copy_complete(int read_err
, unsigned long write_err
, void *context
)
545 struct dm_thin_new_mapping
*m
= context
;
546 struct pool
*pool
= m
->tc
->pool
;
548 m
->err
= read_err
|| write_err
? -EIO
: 0;
550 spin_lock_irqsave(&pool
->lock
, flags
);
552 __maybe_add_mapping(m
);
553 spin_unlock_irqrestore(&pool
->lock
, flags
);
556 static void overwrite_endio(struct bio
*bio
, int err
)
559 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
560 struct dm_thin_new_mapping
*m
= h
->overwrite_mapping
;
561 struct pool
*pool
= m
->tc
->pool
;
565 spin_lock_irqsave(&pool
->lock
, flags
);
567 __maybe_add_mapping(m
);
568 spin_unlock_irqrestore(&pool
->lock
, flags
);
571 /*----------------------------------------------------------------*/
578 * Prepared mapping jobs.
582 * This sends the bios in the cell back to the deferred_bios list.
584 static void cell_defer(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
586 struct pool
*pool
= tc
->pool
;
589 spin_lock_irqsave(&pool
->lock
, flags
);
590 cell_release(pool
, cell
, &pool
->deferred_bios
);
591 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
597 * Same as cell_defer above, except it omits the original holder of the cell.
599 static void cell_defer_no_holder(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
601 struct pool
*pool
= tc
->pool
;
604 spin_lock_irqsave(&pool
->lock
, flags
);
605 cell_release_no_holder(pool
, cell
, &pool
->deferred_bios
);
606 spin_unlock_irqrestore(&pool
->lock
, flags
);
611 static void process_prepared_mapping_fail(struct dm_thin_new_mapping
*m
)
614 m
->bio
->bi_end_io
= m
->saved_bi_end_io
;
615 cell_error(m
->tc
->pool
, m
->cell
);
617 mempool_free(m
, m
->tc
->pool
->mapping_pool
);
620 static void process_prepared_mapping(struct dm_thin_new_mapping
*m
)
622 struct thin_c
*tc
= m
->tc
;
623 struct pool
*pool
= tc
->pool
;
629 bio
->bi_end_io
= m
->saved_bi_end_io
;
632 cell_error(pool
, m
->cell
);
637 * Commit the prepared block into the mapping btree.
638 * Any I/O for this block arriving after this point will get
639 * remapped to it directly.
641 r
= dm_thin_insert_block(tc
->td
, m
->virt_block
, m
->data_block
);
643 DMERR_LIMIT("%s: dm_thin_insert_block() failed: error = %d",
644 dm_device_name(pool
->pool_md
), r
);
645 set_pool_mode(pool
, PM_READ_ONLY
);
646 cell_error(pool
, m
->cell
);
651 * Release any bios held while the block was being provisioned.
652 * If we are processing a write bio that completely covers the block,
653 * we already processed it so can ignore it now when processing
654 * the bios in the cell.
657 cell_defer_no_holder(tc
, m
->cell
);
660 cell_defer(tc
, m
->cell
);
664 mempool_free(m
, pool
->mapping_pool
);
667 static void process_prepared_discard_fail(struct dm_thin_new_mapping
*m
)
669 struct thin_c
*tc
= m
->tc
;
671 bio_io_error(m
->bio
);
672 cell_defer_no_holder(tc
, m
->cell
);
673 cell_defer_no_holder(tc
, m
->cell2
);
674 mempool_free(m
, tc
->pool
->mapping_pool
);
677 static void process_prepared_discard_passdown(struct dm_thin_new_mapping
*m
)
679 struct thin_c
*tc
= m
->tc
;
681 inc_all_io_entry(tc
->pool
, m
->bio
);
682 cell_defer_no_holder(tc
, m
->cell
);
683 cell_defer_no_holder(tc
, m
->cell2
);
686 remap_and_issue(tc
, m
->bio
, m
->data_block
);
688 bio_endio(m
->bio
, 0);
690 mempool_free(m
, tc
->pool
->mapping_pool
);
693 static void process_prepared_discard(struct dm_thin_new_mapping
*m
)
696 struct thin_c
*tc
= m
->tc
;
698 r
= dm_thin_remove_block(tc
->td
, m
->virt_block
);
700 DMERR_LIMIT("dm_thin_remove_block() failed");
702 process_prepared_discard_passdown(m
);
705 static void process_prepared(struct pool
*pool
, struct list_head
*head
,
706 process_mapping_fn
*fn
)
709 struct list_head maps
;
710 struct dm_thin_new_mapping
*m
, *tmp
;
712 INIT_LIST_HEAD(&maps
);
713 spin_lock_irqsave(&pool
->lock
, flags
);
714 list_splice_init(head
, &maps
);
715 spin_unlock_irqrestore(&pool
->lock
, flags
);
717 list_for_each_entry_safe(m
, tmp
, &maps
, list
)
724 static int io_overlaps_block(struct pool
*pool
, struct bio
*bio
)
726 return bio
->bi_size
== (pool
->sectors_per_block
<< SECTOR_SHIFT
);
729 static int io_overwrites_block(struct pool
*pool
, struct bio
*bio
)
731 return (bio_data_dir(bio
) == WRITE
) &&
732 io_overlaps_block(pool
, bio
);
735 static void save_and_set_endio(struct bio
*bio
, bio_end_io_t
**save
,
738 *save
= bio
->bi_end_io
;
742 static int ensure_next_mapping(struct pool
*pool
)
744 if (pool
->next_mapping
)
747 pool
->next_mapping
= mempool_alloc(pool
->mapping_pool
, GFP_ATOMIC
);
749 return pool
->next_mapping
? 0 : -ENOMEM
;
752 static struct dm_thin_new_mapping
*get_next_mapping(struct pool
*pool
)
754 struct dm_thin_new_mapping
*m
= pool
->next_mapping
;
756 BUG_ON(!pool
->next_mapping
);
758 memset(m
, 0, sizeof(struct dm_thin_new_mapping
));
759 INIT_LIST_HEAD(&m
->list
);
762 pool
->next_mapping
= NULL
;
767 static void schedule_copy(struct thin_c
*tc
, dm_block_t virt_block
,
768 struct dm_dev
*origin
, dm_block_t data_origin
,
769 dm_block_t data_dest
,
770 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
773 struct pool
*pool
= tc
->pool
;
774 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
777 m
->virt_block
= virt_block
;
778 m
->data_block
= data_dest
;
781 if (!dm_deferred_set_add_work(pool
->shared_read_ds
, &m
->list
))
785 * IO to pool_dev remaps to the pool target's data_dev.
787 * If the whole block of data is being overwritten, we can issue the
788 * bio immediately. Otherwise we use kcopyd to clone the data first.
790 if (io_overwrites_block(pool
, bio
)) {
791 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
793 h
->overwrite_mapping
= m
;
795 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
796 inc_all_io_entry(pool
, bio
);
797 remap_and_issue(tc
, bio
, data_dest
);
799 struct dm_io_region from
, to
;
801 from
.bdev
= origin
->bdev
;
802 from
.sector
= data_origin
* pool
->sectors_per_block
;
803 from
.count
= pool
->sectors_per_block
;
805 to
.bdev
= tc
->pool_dev
->bdev
;
806 to
.sector
= data_dest
* pool
->sectors_per_block
;
807 to
.count
= pool
->sectors_per_block
;
809 r
= dm_kcopyd_copy(pool
->copier
, &from
, 1, &to
,
810 0, copy_complete
, m
);
812 mempool_free(m
, pool
->mapping_pool
);
813 DMERR_LIMIT("dm_kcopyd_copy() failed");
814 cell_error(pool
, cell
);
819 static void schedule_internal_copy(struct thin_c
*tc
, dm_block_t virt_block
,
820 dm_block_t data_origin
, dm_block_t data_dest
,
821 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
823 schedule_copy(tc
, virt_block
, tc
->pool_dev
,
824 data_origin
, data_dest
, cell
, bio
);
827 static void schedule_external_copy(struct thin_c
*tc
, dm_block_t virt_block
,
828 dm_block_t data_dest
,
829 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
831 schedule_copy(tc
, virt_block
, tc
->origin_dev
,
832 virt_block
, data_dest
, cell
, bio
);
835 static void schedule_zero(struct thin_c
*tc
, dm_block_t virt_block
,
836 dm_block_t data_block
, struct dm_bio_prison_cell
*cell
,
839 struct pool
*pool
= tc
->pool
;
840 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
845 m
->virt_block
= virt_block
;
846 m
->data_block
= data_block
;
850 * If the whole block of data is being overwritten or we are not
851 * zeroing pre-existing data, we can issue the bio immediately.
852 * Otherwise we use kcopyd to zero the data first.
854 if (!pool
->pf
.zero_new_blocks
)
855 process_prepared_mapping(m
);
857 else if (io_overwrites_block(pool
, bio
)) {
858 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
860 h
->overwrite_mapping
= m
;
862 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
863 inc_all_io_entry(pool
, bio
);
864 remap_and_issue(tc
, bio
, data_block
);
867 struct dm_io_region to
;
869 to
.bdev
= tc
->pool_dev
->bdev
;
870 to
.sector
= data_block
* pool
->sectors_per_block
;
871 to
.count
= pool
->sectors_per_block
;
873 r
= dm_kcopyd_zero(pool
->copier
, 1, &to
, 0, copy_complete
, m
);
875 mempool_free(m
, pool
->mapping_pool
);
876 DMERR_LIMIT("dm_kcopyd_zero() failed");
877 cell_error(pool
, cell
);
883 * A non-zero return indicates read_only or fail_io mode.
884 * Many callers don't care about the return value.
886 static int commit(struct pool
*pool
)
890 if (get_pool_mode(pool
) != PM_WRITE
)
893 r
= dm_pool_commit_metadata(pool
->pmd
);
895 DMERR_LIMIT("%s: dm_pool_commit_metadata failed: error = %d",
896 dm_device_name(pool
->pool_md
), r
);
897 set_pool_mode(pool
, PM_READ_ONLY
);
903 static int alloc_data_block(struct thin_c
*tc
, dm_block_t
*result
)
906 dm_block_t free_blocks
;
908 struct pool
*pool
= tc
->pool
;
911 * Once no_free_space is set we must not allow allocation to succeed.
912 * Otherwise it is difficult to explain, debug, test and support.
914 if (pool
->no_free_space
)
917 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
921 if (free_blocks
<= pool
->low_water_blocks
&& !pool
->low_water_triggered
) {
922 DMWARN("%s: reached low water mark for data device: sending event.",
923 dm_device_name(pool
->pool_md
));
924 spin_lock_irqsave(&pool
->lock
, flags
);
925 pool
->low_water_triggered
= 1;
926 spin_unlock_irqrestore(&pool
->lock
, flags
);
927 dm_table_event(pool
->ti
->table
);
932 * Try to commit to see if that will free up some
939 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
944 * If we still have no space we set a flag to avoid
945 * doing all this checking and return -ENOSPC. This
946 * flag serves as a latch that disallows allocations from
947 * this pool until the admin takes action (e.g. resize or
951 DMWARN("%s: no free data space available.",
952 dm_device_name(pool
->pool_md
));
953 spin_lock_irqsave(&pool
->lock
, flags
);
954 pool
->no_free_space
= 1;
955 spin_unlock_irqrestore(&pool
->lock
, flags
);
960 r
= dm_pool_alloc_data_block(pool
->pmd
, result
);
963 !dm_pool_get_free_metadata_block_count(pool
->pmd
, &free_blocks
) &&
965 DMWARN("%s: no free metadata space available.",
966 dm_device_name(pool
->pool_md
));
967 set_pool_mode(pool
, PM_READ_ONLY
);
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 no_space(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
994 struct bio_list bios
;
996 bio_list_init(&bios
);
997 cell_release(pool
, cell
, &bios
);
999 while ((bio
= bio_list_pop(&bios
)))
1000 retry_on_resume(bio
);
1003 static void process_discard(struct thin_c
*tc
, struct bio
*bio
)
1006 unsigned long flags
;
1007 struct pool
*pool
= tc
->pool
;
1008 struct dm_bio_prison_cell
*cell
, *cell2
;
1009 struct dm_cell_key key
, key2
;
1010 dm_block_t block
= get_bio_block(tc
, bio
);
1011 struct dm_thin_lookup_result lookup_result
;
1012 struct dm_thin_new_mapping
*m
;
1014 build_virtual_key(tc
->td
, block
, &key
);
1015 if (bio_detain(tc
->pool
, &key
, bio
, &cell
))
1018 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1022 * Check nobody is fiddling with this pool block. This can
1023 * happen if someone's in the process of breaking sharing
1026 build_data_key(tc
->td
, lookup_result
.block
, &key2
);
1027 if (bio_detain(tc
->pool
, &key2
, bio
, &cell2
)) {
1028 cell_defer_no_holder(tc
, cell
);
1032 if (io_overlaps_block(pool
, bio
)) {
1034 * IO may still be going to the destination block. We must
1035 * quiesce before we can do the removal.
1037 m
= get_next_mapping(pool
);
1039 m
->pass_discard
= (!lookup_result
.shared
) && pool
->pf
.discard_passdown
;
1040 m
->virt_block
= block
;
1041 m
->data_block
= lookup_result
.block
;
1046 if (!dm_deferred_set_add_work(pool
->all_io_ds
, &m
->list
)) {
1047 spin_lock_irqsave(&pool
->lock
, flags
);
1048 list_add(&m
->list
, &pool
->prepared_discards
);
1049 spin_unlock_irqrestore(&pool
->lock
, flags
);
1053 inc_all_io_entry(pool
, bio
);
1054 cell_defer_no_holder(tc
, cell
);
1055 cell_defer_no_holder(tc
, cell2
);
1058 * The DM core makes sure that the discard doesn't span
1059 * a block boundary. So we submit the discard of a
1060 * partial block appropriately.
1062 if ((!lookup_result
.shared
) && pool
->pf
.discard_passdown
)
1063 remap_and_issue(tc
, bio
, lookup_result
.block
);
1071 * It isn't provisioned, just forget it.
1073 cell_defer_no_holder(tc
, cell
);
1078 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1080 cell_defer_no_holder(tc
, cell
);
1086 static void break_sharing(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1087 struct dm_cell_key
*key
,
1088 struct dm_thin_lookup_result
*lookup_result
,
1089 struct dm_bio_prison_cell
*cell
)
1092 dm_block_t data_block
;
1093 struct pool
*pool
= tc
->pool
;
1095 r
= alloc_data_block(tc
, &data_block
);
1098 schedule_internal_copy(tc
, block
, lookup_result
->block
,
1099 data_block
, cell
, bio
);
1103 no_space(pool
, cell
);
1107 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1109 set_pool_mode(pool
, PM_READ_ONLY
);
1110 cell_error(pool
, cell
);
1115 static void process_shared_bio(struct thin_c
*tc
, struct bio
*bio
,
1117 struct dm_thin_lookup_result
*lookup_result
)
1119 struct dm_bio_prison_cell
*cell
;
1120 struct pool
*pool
= tc
->pool
;
1121 struct dm_cell_key key
;
1124 * If cell is already occupied, then sharing is already in the process
1125 * of being broken so we have nothing further to do here.
1127 build_data_key(tc
->td
, lookup_result
->block
, &key
);
1128 if (bio_detain(pool
, &key
, bio
, &cell
))
1131 if (bio_data_dir(bio
) == WRITE
&& bio
->bi_size
)
1132 break_sharing(tc
, bio
, block
, &key
, lookup_result
, cell
);
1134 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1136 h
->shared_read_entry
= dm_deferred_entry_inc(pool
->shared_read_ds
);
1137 inc_all_io_entry(pool
, bio
);
1138 cell_defer_no_holder(tc
, cell
);
1140 remap_and_issue(tc
, bio
, lookup_result
->block
);
1144 static void provision_block(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1145 struct dm_bio_prison_cell
*cell
)
1148 dm_block_t data_block
;
1149 struct pool
*pool
= tc
->pool
;
1152 * Remap empty bios (flushes) immediately, without provisioning.
1154 if (!bio
->bi_size
) {
1155 inc_all_io_entry(pool
, bio
);
1156 cell_defer_no_holder(tc
, cell
);
1158 remap_and_issue(tc
, bio
, 0);
1163 * Fill read bios with zeroes and complete them immediately.
1165 if (bio_data_dir(bio
) == READ
) {
1167 cell_defer_no_holder(tc
, cell
);
1172 r
= alloc_data_block(tc
, &data_block
);
1176 schedule_external_copy(tc
, block
, data_block
, cell
, bio
);
1178 schedule_zero(tc
, block
, data_block
, cell
, bio
);
1182 no_space(pool
, cell
);
1186 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1188 set_pool_mode(pool
, PM_READ_ONLY
);
1189 cell_error(pool
, cell
);
1194 static void process_bio(struct thin_c
*tc
, struct bio
*bio
)
1197 struct pool
*pool
= tc
->pool
;
1198 dm_block_t block
= get_bio_block(tc
, bio
);
1199 struct dm_bio_prison_cell
*cell
;
1200 struct dm_cell_key key
;
1201 struct dm_thin_lookup_result lookup_result
;
1204 * If cell is already occupied, then the block is already
1205 * being provisioned so we have nothing further to do here.
1207 build_virtual_key(tc
->td
, block
, &key
);
1208 if (bio_detain(pool
, &key
, bio
, &cell
))
1211 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1214 if (lookup_result
.shared
) {
1215 process_shared_bio(tc
, bio
, block
, &lookup_result
);
1216 cell_defer_no_holder(tc
, cell
); /* FIXME: pass this cell into process_shared? */
1218 inc_all_io_entry(pool
, bio
);
1219 cell_defer_no_holder(tc
, cell
);
1221 remap_and_issue(tc
, bio
, lookup_result
.block
);
1226 if (bio_data_dir(bio
) == READ
&& tc
->origin_dev
) {
1227 inc_all_io_entry(pool
, bio
);
1228 cell_defer_no_holder(tc
, cell
);
1230 remap_to_origin_and_issue(tc
, bio
);
1232 provision_block(tc
, bio
, block
, cell
);
1236 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1238 cell_defer_no_holder(tc
, cell
);
1244 static void process_bio_read_only(struct thin_c
*tc
, struct bio
*bio
)
1247 int rw
= bio_data_dir(bio
);
1248 dm_block_t block
= get_bio_block(tc
, bio
);
1249 struct dm_thin_lookup_result lookup_result
;
1251 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1254 if (lookup_result
.shared
&& (rw
== WRITE
) && bio
->bi_size
)
1257 inc_all_io_entry(tc
->pool
, bio
);
1258 remap_and_issue(tc
, bio
, lookup_result
.block
);
1268 if (tc
->origin_dev
) {
1269 inc_all_io_entry(tc
->pool
, bio
);
1270 remap_to_origin_and_issue(tc
, bio
);
1279 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1286 static void process_bio_fail(struct thin_c
*tc
, struct bio
*bio
)
1292 * FIXME: should we also commit due to size of transaction, measured in
1295 static int need_commit_due_to_time(struct pool
*pool
)
1297 return jiffies
< pool
->last_commit_jiffies
||
1298 jiffies
> pool
->last_commit_jiffies
+ COMMIT_PERIOD
;
1301 static void process_deferred_bios(struct pool
*pool
)
1303 unsigned long flags
;
1305 struct bio_list bios
;
1307 bio_list_init(&bios
);
1309 spin_lock_irqsave(&pool
->lock
, flags
);
1310 bio_list_merge(&bios
, &pool
->deferred_bios
);
1311 bio_list_init(&pool
->deferred_bios
);
1312 spin_unlock_irqrestore(&pool
->lock
, flags
);
1314 while ((bio
= bio_list_pop(&bios
))) {
1315 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1316 struct thin_c
*tc
= h
->tc
;
1319 * If we've got no free new_mapping structs, and processing
1320 * this bio might require one, we pause until there are some
1321 * prepared mappings to process.
1323 if (ensure_next_mapping(pool
)) {
1324 spin_lock_irqsave(&pool
->lock
, flags
);
1325 bio_list_merge(&pool
->deferred_bios
, &bios
);
1326 spin_unlock_irqrestore(&pool
->lock
, flags
);
1331 if (bio
->bi_rw
& REQ_DISCARD
)
1332 pool
->process_discard(tc
, bio
);
1334 pool
->process_bio(tc
, bio
);
1338 * If there are any deferred flush bios, we must commit
1339 * the metadata before issuing them.
1341 bio_list_init(&bios
);
1342 spin_lock_irqsave(&pool
->lock
, flags
);
1343 bio_list_merge(&bios
, &pool
->deferred_flush_bios
);
1344 bio_list_init(&pool
->deferred_flush_bios
);
1345 spin_unlock_irqrestore(&pool
->lock
, flags
);
1347 if (bio_list_empty(&bios
) && !need_commit_due_to_time(pool
))
1351 while ((bio
= bio_list_pop(&bios
)))
1355 pool
->last_commit_jiffies
= jiffies
;
1357 while ((bio
= bio_list_pop(&bios
)))
1358 generic_make_request(bio
);
1361 static void do_worker(struct work_struct
*ws
)
1363 struct pool
*pool
= container_of(ws
, struct pool
, worker
);
1365 process_prepared(pool
, &pool
->prepared_mappings
, &pool
->process_prepared_mapping
);
1366 process_prepared(pool
, &pool
->prepared_discards
, &pool
->process_prepared_discard
);
1367 process_deferred_bios(pool
);
1371 * We want to commit periodically so that not too much
1372 * unwritten data builds up.
1374 static void do_waker(struct work_struct
*ws
)
1376 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
, waker
);
1378 queue_delayed_work(pool
->wq
, &pool
->waker
, COMMIT_PERIOD
);
1381 /*----------------------------------------------------------------*/
1383 static enum pool_mode
get_pool_mode(struct pool
*pool
)
1385 return pool
->pf
.mode
;
1388 static void set_pool_mode(struct pool
*pool
, enum pool_mode mode
)
1392 pool
->pf
.mode
= mode
;
1396 DMERR("%s: switching pool to failure mode",
1397 dm_device_name(pool
->pool_md
));
1398 dm_pool_metadata_read_only(pool
->pmd
);
1399 pool
->process_bio
= process_bio_fail
;
1400 pool
->process_discard
= process_bio_fail
;
1401 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1402 pool
->process_prepared_discard
= process_prepared_discard_fail
;
1406 DMERR("%s: switching pool to read-only mode",
1407 dm_device_name(pool
->pool_md
));
1408 r
= dm_pool_abort_metadata(pool
->pmd
);
1410 DMERR("%s: aborting transaction failed",
1411 dm_device_name(pool
->pool_md
));
1412 set_pool_mode(pool
, PM_FAIL
);
1414 dm_pool_metadata_read_only(pool
->pmd
);
1415 pool
->process_bio
= process_bio_read_only
;
1416 pool
->process_discard
= process_discard
;
1417 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1418 pool
->process_prepared_discard
= process_prepared_discard_passdown
;
1423 dm_pool_metadata_read_write(pool
->pmd
);
1424 pool
->process_bio
= process_bio
;
1425 pool
->process_discard
= process_discard
;
1426 pool
->process_prepared_mapping
= process_prepared_mapping
;
1427 pool
->process_prepared_discard
= process_prepared_discard
;
1432 /*----------------------------------------------------------------*/
1435 * Mapping functions.
1439 * Called only while mapping a thin bio to hand it over to the workqueue.
1441 static void thin_defer_bio(struct thin_c
*tc
, struct bio
*bio
)
1443 unsigned long flags
;
1444 struct pool
*pool
= tc
->pool
;
1446 spin_lock_irqsave(&pool
->lock
, flags
);
1447 bio_list_add(&pool
->deferred_bios
, bio
);
1448 spin_unlock_irqrestore(&pool
->lock
, flags
);
1453 static void thin_hook_bio(struct thin_c
*tc
, struct bio
*bio
)
1455 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1458 h
->shared_read_entry
= NULL
;
1459 h
->all_io_entry
= NULL
;
1460 h
->overwrite_mapping
= NULL
;
1464 * Non-blocking function called from the thin target's map function.
1466 static int thin_bio_map(struct dm_target
*ti
, struct bio
*bio
)
1469 struct thin_c
*tc
= ti
->private;
1470 dm_block_t block
= get_bio_block(tc
, bio
);
1471 struct dm_thin_device
*td
= tc
->td
;
1472 struct dm_thin_lookup_result result
;
1473 struct dm_bio_prison_cell cell1
, cell2
;
1474 struct dm_bio_prison_cell
*cell_result
;
1475 struct dm_cell_key key
;
1477 thin_hook_bio(tc
, bio
);
1479 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
1481 return DM_MAPIO_SUBMITTED
;
1484 if (bio
->bi_rw
& (REQ_DISCARD
| REQ_FLUSH
| REQ_FUA
)) {
1485 thin_defer_bio(tc
, bio
);
1486 return DM_MAPIO_SUBMITTED
;
1489 r
= dm_thin_find_block(td
, block
, 0, &result
);
1492 * Note that we defer readahead too.
1496 if (unlikely(result
.shared
)) {
1498 * We have a race condition here between the
1499 * result.shared value returned by the lookup and
1500 * snapshot creation, which may cause new
1503 * To avoid this always quiesce the origin before
1504 * taking the snap. You want to do this anyway to
1505 * ensure a consistent application view
1508 * More distant ancestors are irrelevant. The
1509 * shared flag will be set in their case.
1511 thin_defer_bio(tc
, bio
);
1512 return DM_MAPIO_SUBMITTED
;
1515 build_virtual_key(tc
->td
, block
, &key
);
1516 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell1
, &cell_result
))
1517 return DM_MAPIO_SUBMITTED
;
1519 build_data_key(tc
->td
, result
.block
, &key
);
1520 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell2
, &cell_result
)) {
1521 cell_defer_no_holder_no_free(tc
, &cell1
);
1522 return DM_MAPIO_SUBMITTED
;
1525 inc_all_io_entry(tc
->pool
, bio
);
1526 cell_defer_no_holder_no_free(tc
, &cell2
);
1527 cell_defer_no_holder_no_free(tc
, &cell1
);
1529 remap(tc
, bio
, result
.block
);
1530 return DM_MAPIO_REMAPPED
;
1533 if (get_pool_mode(tc
->pool
) == PM_READ_ONLY
) {
1535 * This block isn't provisioned, and we have no way
1536 * of doing so. Just error it.
1539 return DM_MAPIO_SUBMITTED
;
1545 * In future, the failed dm_thin_find_block above could
1546 * provide the hint to load the metadata into cache.
1548 thin_defer_bio(tc
, bio
);
1549 return DM_MAPIO_SUBMITTED
;
1553 * Must always call bio_io_error on failure.
1554 * dm_thin_find_block can fail with -EINVAL if the
1555 * pool is switched to fail-io mode.
1558 return DM_MAPIO_SUBMITTED
;
1562 static int pool_is_congested(struct dm_target_callbacks
*cb
, int bdi_bits
)
1565 unsigned long flags
;
1566 struct pool_c
*pt
= container_of(cb
, struct pool_c
, callbacks
);
1568 spin_lock_irqsave(&pt
->pool
->lock
, flags
);
1569 r
= !bio_list_empty(&pt
->pool
->retry_on_resume_list
);
1570 spin_unlock_irqrestore(&pt
->pool
->lock
, flags
);
1573 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1574 r
= bdi_congested(&q
->backing_dev_info
, bdi_bits
);
1580 static void __requeue_bios(struct pool
*pool
)
1582 bio_list_merge(&pool
->deferred_bios
, &pool
->retry_on_resume_list
);
1583 bio_list_init(&pool
->retry_on_resume_list
);
1586 /*----------------------------------------------------------------
1587 * Binding of control targets to a pool object
1588 *--------------------------------------------------------------*/
1589 static bool data_dev_supports_discard(struct pool_c
*pt
)
1591 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1593 return q
&& blk_queue_discard(q
);
1596 static bool is_factor(sector_t block_size
, uint32_t n
)
1598 return !sector_div(block_size
, n
);
1602 * If discard_passdown was enabled verify that the data device
1603 * supports discards. Disable discard_passdown if not.
1605 static void disable_passdown_if_not_supported(struct pool_c
*pt
)
1607 struct pool
*pool
= pt
->pool
;
1608 struct block_device
*data_bdev
= pt
->data_dev
->bdev
;
1609 struct queue_limits
*data_limits
= &bdev_get_queue(data_bdev
)->limits
;
1610 sector_t block_size
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
1611 const char *reason
= NULL
;
1612 char buf
[BDEVNAME_SIZE
];
1614 if (!pt
->adjusted_pf
.discard_passdown
)
1617 if (!data_dev_supports_discard(pt
))
1618 reason
= "discard unsupported";
1620 else if (data_limits
->max_discard_sectors
< pool
->sectors_per_block
)
1621 reason
= "max discard sectors smaller than a block";
1623 else if (data_limits
->discard_granularity
> block_size
)
1624 reason
= "discard granularity larger than a block";
1626 else if (!is_factor(block_size
, data_limits
->discard_granularity
))
1627 reason
= "discard granularity not a factor of block size";
1630 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev
, buf
), reason
);
1631 pt
->adjusted_pf
.discard_passdown
= false;
1635 static int bind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1637 struct pool_c
*pt
= ti
->private;
1640 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
1642 enum pool_mode old_mode
= pool
->pf
.mode
;
1643 enum pool_mode new_mode
= pt
->adjusted_pf
.mode
;
1646 * If we were in PM_FAIL mode, rollback of metadata failed. We're
1647 * not going to recover without a thin_repair. So we never let the
1648 * pool move out of the old mode. On the other hand a PM_READ_ONLY
1649 * may have been due to a lack of metadata or data space, and may
1650 * now work (ie. if the underlying devices have been resized).
1652 if (old_mode
== PM_FAIL
)
1653 new_mode
= old_mode
;
1656 pool
->low_water_blocks
= pt
->low_water_blocks
;
1657 pool
->pf
= pt
->adjusted_pf
;
1659 set_pool_mode(pool
, new_mode
);
1664 static void unbind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1670 /*----------------------------------------------------------------
1672 *--------------------------------------------------------------*/
1673 /* Initialize pool features. */
1674 static void pool_features_init(struct pool_features
*pf
)
1676 pf
->mode
= PM_WRITE
;
1677 pf
->zero_new_blocks
= true;
1678 pf
->discard_enabled
= true;
1679 pf
->discard_passdown
= true;
1682 static void __pool_destroy(struct pool
*pool
)
1684 __pool_table_remove(pool
);
1686 if (dm_pool_metadata_close(pool
->pmd
) < 0)
1687 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1689 dm_bio_prison_destroy(pool
->prison
);
1690 dm_kcopyd_client_destroy(pool
->copier
);
1693 destroy_workqueue(pool
->wq
);
1695 if (pool
->next_mapping
)
1696 mempool_free(pool
->next_mapping
, pool
->mapping_pool
);
1697 mempool_destroy(pool
->mapping_pool
);
1698 dm_deferred_set_destroy(pool
->shared_read_ds
);
1699 dm_deferred_set_destroy(pool
->all_io_ds
);
1703 static struct kmem_cache
*_new_mapping_cache
;
1705 static struct pool
*pool_create(struct mapped_device
*pool_md
,
1706 struct block_device
*metadata_dev
,
1707 unsigned long block_size
,
1708 int read_only
, char **error
)
1713 struct dm_pool_metadata
*pmd
;
1714 bool format_device
= read_only
? false : true;
1716 pmd
= dm_pool_metadata_open(metadata_dev
, block_size
, format_device
);
1718 *error
= "Error creating metadata object";
1719 return (struct pool
*)pmd
;
1722 pool
= kmalloc(sizeof(*pool
), GFP_KERNEL
);
1724 *error
= "Error allocating memory for pool";
1725 err_p
= ERR_PTR(-ENOMEM
);
1730 pool
->sectors_per_block
= block_size
;
1731 if (block_size
& (block_size
- 1))
1732 pool
->sectors_per_block_shift
= -1;
1734 pool
->sectors_per_block_shift
= __ffs(block_size
);
1735 pool
->low_water_blocks
= 0;
1736 pool_features_init(&pool
->pf
);
1737 pool
->prison
= dm_bio_prison_create(PRISON_CELLS
);
1738 if (!pool
->prison
) {
1739 *error
= "Error creating pool's bio prison";
1740 err_p
= ERR_PTR(-ENOMEM
);
1744 pool
->copier
= dm_kcopyd_client_create(&dm_kcopyd_throttle
);
1745 if (IS_ERR(pool
->copier
)) {
1746 r
= PTR_ERR(pool
->copier
);
1747 *error
= "Error creating pool's kcopyd client";
1749 goto bad_kcopyd_client
;
1753 * Create singlethreaded workqueue that will service all devices
1754 * that use this metadata.
1756 pool
->wq
= alloc_ordered_workqueue("dm-" DM_MSG_PREFIX
, WQ_MEM_RECLAIM
);
1758 *error
= "Error creating pool's workqueue";
1759 err_p
= ERR_PTR(-ENOMEM
);
1763 INIT_WORK(&pool
->worker
, do_worker
);
1764 INIT_DELAYED_WORK(&pool
->waker
, do_waker
);
1765 spin_lock_init(&pool
->lock
);
1766 bio_list_init(&pool
->deferred_bios
);
1767 bio_list_init(&pool
->deferred_flush_bios
);
1768 INIT_LIST_HEAD(&pool
->prepared_mappings
);
1769 INIT_LIST_HEAD(&pool
->prepared_discards
);
1770 pool
->low_water_triggered
= 0;
1771 pool
->no_free_space
= 0;
1772 bio_list_init(&pool
->retry_on_resume_list
);
1774 pool
->shared_read_ds
= dm_deferred_set_create();
1775 if (!pool
->shared_read_ds
) {
1776 *error
= "Error creating pool's shared read deferred set";
1777 err_p
= ERR_PTR(-ENOMEM
);
1778 goto bad_shared_read_ds
;
1781 pool
->all_io_ds
= dm_deferred_set_create();
1782 if (!pool
->all_io_ds
) {
1783 *error
= "Error creating pool's all io deferred set";
1784 err_p
= ERR_PTR(-ENOMEM
);
1788 pool
->next_mapping
= NULL
;
1789 pool
->mapping_pool
= mempool_create_slab_pool(MAPPING_POOL_SIZE
,
1790 _new_mapping_cache
);
1791 if (!pool
->mapping_pool
) {
1792 *error
= "Error creating pool's mapping mempool";
1793 err_p
= ERR_PTR(-ENOMEM
);
1794 goto bad_mapping_pool
;
1797 pool
->ref_count
= 1;
1798 pool
->last_commit_jiffies
= jiffies
;
1799 pool
->pool_md
= pool_md
;
1800 pool
->md_dev
= metadata_dev
;
1801 __pool_table_insert(pool
);
1806 dm_deferred_set_destroy(pool
->all_io_ds
);
1808 dm_deferred_set_destroy(pool
->shared_read_ds
);
1810 destroy_workqueue(pool
->wq
);
1812 dm_kcopyd_client_destroy(pool
->copier
);
1814 dm_bio_prison_destroy(pool
->prison
);
1818 if (dm_pool_metadata_close(pmd
))
1819 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1824 static void __pool_inc(struct pool
*pool
)
1826 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
1830 static void __pool_dec(struct pool
*pool
)
1832 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
1833 BUG_ON(!pool
->ref_count
);
1834 if (!--pool
->ref_count
)
1835 __pool_destroy(pool
);
1838 static struct pool
*__pool_find(struct mapped_device
*pool_md
,
1839 struct block_device
*metadata_dev
,
1840 unsigned long block_size
, int read_only
,
1841 char **error
, int *created
)
1843 struct pool
*pool
= __pool_table_lookup_metadata_dev(metadata_dev
);
1846 if (pool
->pool_md
!= pool_md
) {
1847 *error
= "metadata device already in use by a pool";
1848 return ERR_PTR(-EBUSY
);
1853 pool
= __pool_table_lookup(pool_md
);
1855 if (pool
->md_dev
!= metadata_dev
) {
1856 *error
= "different pool cannot replace a pool";
1857 return ERR_PTR(-EINVAL
);
1862 pool
= pool_create(pool_md
, metadata_dev
, block_size
, read_only
, error
);
1870 /*----------------------------------------------------------------
1871 * Pool target methods
1872 *--------------------------------------------------------------*/
1873 static void pool_dtr(struct dm_target
*ti
)
1875 struct pool_c
*pt
= ti
->private;
1877 mutex_lock(&dm_thin_pool_table
.mutex
);
1879 unbind_control_target(pt
->pool
, ti
);
1880 __pool_dec(pt
->pool
);
1881 dm_put_device(ti
, pt
->metadata_dev
);
1882 dm_put_device(ti
, pt
->data_dev
);
1885 mutex_unlock(&dm_thin_pool_table
.mutex
);
1888 static int parse_pool_features(struct dm_arg_set
*as
, struct pool_features
*pf
,
1889 struct dm_target
*ti
)
1893 const char *arg_name
;
1895 static struct dm_arg _args
[] = {
1896 {0, 3, "Invalid number of pool feature arguments"},
1900 * No feature arguments supplied.
1905 r
= dm_read_arg_group(_args
, as
, &argc
, &ti
->error
);
1909 while (argc
&& !r
) {
1910 arg_name
= dm_shift_arg(as
);
1913 if (!strcasecmp(arg_name
, "skip_block_zeroing"))
1914 pf
->zero_new_blocks
= false;
1916 else if (!strcasecmp(arg_name
, "ignore_discard"))
1917 pf
->discard_enabled
= false;
1919 else if (!strcasecmp(arg_name
, "no_discard_passdown"))
1920 pf
->discard_passdown
= false;
1922 else if (!strcasecmp(arg_name
, "read_only"))
1923 pf
->mode
= PM_READ_ONLY
;
1926 ti
->error
= "Unrecognised pool feature requested";
1935 static void metadata_low_callback(void *context
)
1937 struct pool
*pool
= context
;
1939 DMWARN("%s: reached low water mark for metadata device: sending event.",
1940 dm_device_name(pool
->pool_md
));
1942 dm_table_event(pool
->ti
->table
);
1945 static sector_t
get_metadata_dev_size(struct block_device
*bdev
)
1947 sector_t metadata_dev_size
= i_size_read(bdev
->bd_inode
) >> SECTOR_SHIFT
;
1948 char buffer
[BDEVNAME_SIZE
];
1950 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS_WARNING
) {
1951 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
1952 bdevname(bdev
, buffer
), THIN_METADATA_MAX_SECTORS
);
1953 metadata_dev_size
= THIN_METADATA_MAX_SECTORS_WARNING
;
1956 return metadata_dev_size
;
1959 static dm_block_t
get_metadata_dev_size_in_blocks(struct block_device
*bdev
)
1961 sector_t metadata_dev_size
= get_metadata_dev_size(bdev
);
1963 sector_div(metadata_dev_size
, THIN_METADATA_BLOCK_SIZE
>> SECTOR_SHIFT
);
1965 return metadata_dev_size
;
1969 * When a metadata threshold is crossed a dm event is triggered, and
1970 * userland should respond by growing the metadata device. We could let
1971 * userland set the threshold, like we do with the data threshold, but I'm
1972 * not sure they know enough to do this well.
1974 static dm_block_t
calc_metadata_threshold(struct pool_c
*pt
)
1977 * 4M is ample for all ops with the possible exception of thin
1978 * device deletion which is harmless if it fails (just retry the
1979 * delete after you've grown the device).
1981 dm_block_t quarter
= get_metadata_dev_size_in_blocks(pt
->metadata_dev
->bdev
) / 4;
1982 return min((dm_block_t
)1024ULL /* 4M */, quarter
);
1986 * thin-pool <metadata dev> <data dev>
1987 * <data block size (sectors)>
1988 * <low water mark (blocks)>
1989 * [<#feature args> [<arg>]*]
1991 * Optional feature arguments are:
1992 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
1993 * ignore_discard: disable discard
1994 * no_discard_passdown: don't pass discards down to the data device
1996 static int pool_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
1998 int r
, pool_created
= 0;
2001 struct pool_features pf
;
2002 struct dm_arg_set as
;
2003 struct dm_dev
*data_dev
;
2004 unsigned long block_size
;
2005 dm_block_t low_water_blocks
;
2006 struct dm_dev
*metadata_dev
;
2007 fmode_t metadata_mode
;
2010 * FIXME Remove validation from scope of lock.
2012 mutex_lock(&dm_thin_pool_table
.mutex
);
2015 ti
->error
= "Invalid argument count";
2024 * Set default pool features.
2026 pool_features_init(&pf
);
2028 dm_consume_args(&as
, 4);
2029 r
= parse_pool_features(&as
, &pf
, ti
);
2033 metadata_mode
= FMODE_READ
| ((pf
.mode
== PM_READ_ONLY
) ? 0 : FMODE_WRITE
);
2034 r
= dm_get_device(ti
, argv
[0], metadata_mode
, &metadata_dev
);
2036 ti
->error
= "Error opening metadata block device";
2041 * Run for the side-effect of possibly issuing a warning if the
2042 * device is too big.
2044 (void) get_metadata_dev_size(metadata_dev
->bdev
);
2046 r
= dm_get_device(ti
, argv
[1], FMODE_READ
| FMODE_WRITE
, &data_dev
);
2048 ti
->error
= "Error getting data device";
2052 if (kstrtoul(argv
[2], 10, &block_size
) || !block_size
||
2053 block_size
< DATA_DEV_BLOCK_SIZE_MIN_SECTORS
||
2054 block_size
> DATA_DEV_BLOCK_SIZE_MAX_SECTORS
||
2055 block_size
& (DATA_DEV_BLOCK_SIZE_MIN_SECTORS
- 1)) {
2056 ti
->error
= "Invalid block size";
2061 if (kstrtoull(argv
[3], 10, (unsigned long long *)&low_water_blocks
)) {
2062 ti
->error
= "Invalid low water mark";
2067 pt
= kzalloc(sizeof(*pt
), GFP_KERNEL
);
2073 pool
= __pool_find(dm_table_get_md(ti
->table
), metadata_dev
->bdev
,
2074 block_size
, pf
.mode
== PM_READ_ONLY
, &ti
->error
, &pool_created
);
2081 * 'pool_created' reflects whether this is the first table load.
2082 * Top level discard support is not allowed to be changed after
2083 * initial load. This would require a pool reload to trigger thin
2086 if (!pool_created
&& pf
.discard_enabled
!= pool
->pf
.discard_enabled
) {
2087 ti
->error
= "Discard support cannot be disabled once enabled";
2089 goto out_flags_changed
;
2094 pt
->metadata_dev
= metadata_dev
;
2095 pt
->data_dev
= data_dev
;
2096 pt
->low_water_blocks
= low_water_blocks
;
2097 pt
->adjusted_pf
= pt
->requested_pf
= pf
;
2098 ti
->num_flush_bios
= 1;
2101 * Only need to enable discards if the pool should pass
2102 * them down to the data device. The thin device's discard
2103 * processing will cause mappings to be removed from the btree.
2105 ti
->discard_zeroes_data_unsupported
= true;
2106 if (pf
.discard_enabled
&& pf
.discard_passdown
) {
2107 ti
->num_discard_bios
= 1;
2110 * Setting 'discards_supported' circumvents the normal
2111 * stacking of discard limits (this keeps the pool and
2112 * thin devices' discard limits consistent).
2114 ti
->discards_supported
= true;
2118 r
= dm_pool_register_metadata_threshold(pt
->pool
->pmd
,
2119 calc_metadata_threshold(pt
),
2120 metadata_low_callback
,
2125 pt
->callbacks
.congested_fn
= pool_is_congested
;
2126 dm_table_add_target_callbacks(ti
->table
, &pt
->callbacks
);
2128 mutex_unlock(&dm_thin_pool_table
.mutex
);
2137 dm_put_device(ti
, data_dev
);
2139 dm_put_device(ti
, metadata_dev
);
2141 mutex_unlock(&dm_thin_pool_table
.mutex
);
2146 static int pool_map(struct dm_target
*ti
, struct bio
*bio
)
2149 struct pool_c
*pt
= ti
->private;
2150 struct pool
*pool
= pt
->pool
;
2151 unsigned long flags
;
2154 * As this is a singleton target, ti->begin is always zero.
2156 spin_lock_irqsave(&pool
->lock
, flags
);
2157 bio
->bi_bdev
= pt
->data_dev
->bdev
;
2158 r
= DM_MAPIO_REMAPPED
;
2159 spin_unlock_irqrestore(&pool
->lock
, flags
);
2164 static int maybe_resize_data_dev(struct dm_target
*ti
, bool *need_commit
)
2167 struct pool_c
*pt
= ti
->private;
2168 struct pool
*pool
= pt
->pool
;
2169 sector_t data_size
= ti
->len
;
2170 dm_block_t sb_data_size
;
2172 *need_commit
= false;
2174 (void) sector_div(data_size
, pool
->sectors_per_block
);
2176 r
= dm_pool_get_data_dev_size(pool
->pmd
, &sb_data_size
);
2178 DMERR("%s: failed to retrieve data device size",
2179 dm_device_name(pool
->pool_md
));
2183 if (data_size
< sb_data_size
) {
2184 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
2185 dm_device_name(pool
->pool_md
),
2186 (unsigned long long)data_size
, sb_data_size
);
2189 } else if (data_size
> sb_data_size
) {
2190 r
= dm_pool_resize_data_dev(pool
->pmd
, data_size
);
2192 DMERR("%s: failed to resize data device",
2193 dm_device_name(pool
->pool_md
));
2194 set_pool_mode(pool
, PM_READ_ONLY
);
2198 *need_commit
= true;
2204 static int maybe_resize_metadata_dev(struct dm_target
*ti
, bool *need_commit
)
2207 struct pool_c
*pt
= ti
->private;
2208 struct pool
*pool
= pt
->pool
;
2209 dm_block_t metadata_dev_size
, sb_metadata_dev_size
;
2211 *need_commit
= false;
2213 metadata_dev_size
= get_metadata_dev_size_in_blocks(pool
->md_dev
);
2215 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &sb_metadata_dev_size
);
2217 DMERR("%s: failed to retrieve metadata device size",
2218 dm_device_name(pool
->pool_md
));
2222 if (metadata_dev_size
< sb_metadata_dev_size
) {
2223 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
2224 dm_device_name(pool
->pool_md
),
2225 metadata_dev_size
, sb_metadata_dev_size
);
2228 } else if (metadata_dev_size
> sb_metadata_dev_size
) {
2229 r
= dm_pool_resize_metadata_dev(pool
->pmd
, metadata_dev_size
);
2231 DMERR("%s: failed to resize metadata device",
2232 dm_device_name(pool
->pool_md
));
2236 *need_commit
= true;
2243 * Retrieves the number of blocks of the data device from
2244 * the superblock and compares it to the actual device size,
2245 * thus resizing the data device in case it has grown.
2247 * This both copes with opening preallocated data devices in the ctr
2248 * being followed by a resume
2250 * calling the resume method individually after userspace has
2251 * grown the data device in reaction to a table event.
2253 static int pool_preresume(struct dm_target
*ti
)
2256 bool need_commit1
, need_commit2
;
2257 struct pool_c
*pt
= ti
->private;
2258 struct pool
*pool
= pt
->pool
;
2261 * Take control of the pool object.
2263 r
= bind_control_target(pool
, ti
);
2267 r
= maybe_resize_data_dev(ti
, &need_commit1
);
2271 r
= maybe_resize_metadata_dev(ti
, &need_commit2
);
2275 if (need_commit1
|| need_commit2
)
2276 (void) commit(pool
);
2281 static void pool_resume(struct dm_target
*ti
)
2283 struct pool_c
*pt
= ti
->private;
2284 struct pool
*pool
= pt
->pool
;
2285 unsigned long flags
;
2287 spin_lock_irqsave(&pool
->lock
, flags
);
2288 pool
->low_water_triggered
= 0;
2289 pool
->no_free_space
= 0;
2290 __requeue_bios(pool
);
2291 spin_unlock_irqrestore(&pool
->lock
, flags
);
2293 do_waker(&pool
->waker
.work
);
2296 static void pool_postsuspend(struct dm_target
*ti
)
2298 struct pool_c
*pt
= ti
->private;
2299 struct pool
*pool
= pt
->pool
;
2301 cancel_delayed_work(&pool
->waker
);
2302 flush_workqueue(pool
->wq
);
2303 (void) commit(pool
);
2306 static int check_arg_count(unsigned argc
, unsigned args_required
)
2308 if (argc
!= args_required
) {
2309 DMWARN("Message received with %u arguments instead of %u.",
2310 argc
, args_required
);
2317 static int read_dev_id(char *arg
, dm_thin_id
*dev_id
, int warning
)
2319 if (!kstrtoull(arg
, 10, (unsigned long long *)dev_id
) &&
2320 *dev_id
<= MAX_DEV_ID
)
2324 DMWARN("Message received with invalid device id: %s", arg
);
2329 static int process_create_thin_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2334 r
= check_arg_count(argc
, 2);
2338 r
= read_dev_id(argv
[1], &dev_id
, 1);
2342 r
= dm_pool_create_thin(pool
->pmd
, dev_id
);
2344 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2352 static int process_create_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2355 dm_thin_id origin_dev_id
;
2358 r
= check_arg_count(argc
, 3);
2362 r
= read_dev_id(argv
[1], &dev_id
, 1);
2366 r
= read_dev_id(argv
[2], &origin_dev_id
, 1);
2370 r
= dm_pool_create_snap(pool
->pmd
, dev_id
, origin_dev_id
);
2372 DMWARN("Creation of new snapshot %s of device %s failed.",
2380 static int process_delete_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2385 r
= check_arg_count(argc
, 2);
2389 r
= read_dev_id(argv
[1], &dev_id
, 1);
2393 r
= dm_pool_delete_thin_device(pool
->pmd
, dev_id
);
2395 DMWARN("Deletion of thin device %s failed.", argv
[1]);
2400 static int process_set_transaction_id_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2402 dm_thin_id old_id
, new_id
;
2405 r
= check_arg_count(argc
, 3);
2409 if (kstrtoull(argv
[1], 10, (unsigned long long *)&old_id
)) {
2410 DMWARN("set_transaction_id message: Unrecognised id %s.", argv
[1]);
2414 if (kstrtoull(argv
[2], 10, (unsigned long long *)&new_id
)) {
2415 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv
[2]);
2419 r
= dm_pool_set_metadata_transaction_id(pool
->pmd
, old_id
, new_id
);
2421 DMWARN("Failed to change transaction id from %s to %s.",
2429 static int process_reserve_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2433 r
= check_arg_count(argc
, 1);
2437 (void) commit(pool
);
2439 r
= dm_pool_reserve_metadata_snap(pool
->pmd
);
2441 DMWARN("reserve_metadata_snap message failed.");
2446 static int process_release_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2450 r
= check_arg_count(argc
, 1);
2454 r
= dm_pool_release_metadata_snap(pool
->pmd
);
2456 DMWARN("release_metadata_snap message failed.");
2462 * Messages supported:
2463 * create_thin <dev_id>
2464 * create_snap <dev_id> <origin_id>
2466 * trim <dev_id> <new_size_in_sectors>
2467 * set_transaction_id <current_trans_id> <new_trans_id>
2468 * reserve_metadata_snap
2469 * release_metadata_snap
2471 static int pool_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
2474 struct pool_c
*pt
= ti
->private;
2475 struct pool
*pool
= pt
->pool
;
2477 if (!strcasecmp(argv
[0], "create_thin"))
2478 r
= process_create_thin_mesg(argc
, argv
, pool
);
2480 else if (!strcasecmp(argv
[0], "create_snap"))
2481 r
= process_create_snap_mesg(argc
, argv
, pool
);
2483 else if (!strcasecmp(argv
[0], "delete"))
2484 r
= process_delete_mesg(argc
, argv
, pool
);
2486 else if (!strcasecmp(argv
[0], "set_transaction_id"))
2487 r
= process_set_transaction_id_mesg(argc
, argv
, pool
);
2489 else if (!strcasecmp(argv
[0], "reserve_metadata_snap"))
2490 r
= process_reserve_metadata_snap_mesg(argc
, argv
, pool
);
2492 else if (!strcasecmp(argv
[0], "release_metadata_snap"))
2493 r
= process_release_metadata_snap_mesg(argc
, argv
, pool
);
2496 DMWARN("Unrecognised thin pool target message received: %s", argv
[0]);
2499 (void) commit(pool
);
2504 static void emit_flags(struct pool_features
*pf
, char *result
,
2505 unsigned sz
, unsigned maxlen
)
2507 unsigned count
= !pf
->zero_new_blocks
+ !pf
->discard_enabled
+
2508 !pf
->discard_passdown
+ (pf
->mode
== PM_READ_ONLY
);
2509 DMEMIT("%u ", count
);
2511 if (!pf
->zero_new_blocks
)
2512 DMEMIT("skip_block_zeroing ");
2514 if (!pf
->discard_enabled
)
2515 DMEMIT("ignore_discard ");
2517 if (!pf
->discard_passdown
)
2518 DMEMIT("no_discard_passdown ");
2520 if (pf
->mode
== PM_READ_ONLY
)
2521 DMEMIT("read_only ");
2526 * <transaction id> <used metadata sectors>/<total metadata sectors>
2527 * <used data sectors>/<total data sectors> <held metadata root>
2529 static void pool_status(struct dm_target
*ti
, status_type_t type
,
2530 unsigned status_flags
, char *result
, unsigned maxlen
)
2534 uint64_t transaction_id
;
2535 dm_block_t nr_free_blocks_data
;
2536 dm_block_t nr_free_blocks_metadata
;
2537 dm_block_t nr_blocks_data
;
2538 dm_block_t nr_blocks_metadata
;
2539 dm_block_t held_root
;
2540 char buf
[BDEVNAME_SIZE
];
2541 char buf2
[BDEVNAME_SIZE
];
2542 struct pool_c
*pt
= ti
->private;
2543 struct pool
*pool
= pt
->pool
;
2546 case STATUSTYPE_INFO
:
2547 if (get_pool_mode(pool
) == PM_FAIL
) {
2552 /* Commit to ensure statistics aren't out-of-date */
2553 if (!(status_flags
& DM_STATUS_NOFLUSH_FLAG
) && !dm_suspended(ti
))
2554 (void) commit(pool
);
2556 r
= dm_pool_get_metadata_transaction_id(pool
->pmd
, &transaction_id
);
2558 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
2559 dm_device_name(pool
->pool_md
), r
);
2563 r
= dm_pool_get_free_metadata_block_count(pool
->pmd
, &nr_free_blocks_metadata
);
2565 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
2566 dm_device_name(pool
->pool_md
), r
);
2570 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &nr_blocks_metadata
);
2572 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
2573 dm_device_name(pool
->pool_md
), r
);
2577 r
= dm_pool_get_free_block_count(pool
->pmd
, &nr_free_blocks_data
);
2579 DMERR("%s: dm_pool_get_free_block_count returned %d",
2580 dm_device_name(pool
->pool_md
), r
);
2584 r
= dm_pool_get_data_dev_size(pool
->pmd
, &nr_blocks_data
);
2586 DMERR("%s: dm_pool_get_data_dev_size returned %d",
2587 dm_device_name(pool
->pool_md
), r
);
2591 r
= dm_pool_get_metadata_snap(pool
->pmd
, &held_root
);
2593 DMERR("%s: dm_pool_get_metadata_snap returned %d",
2594 dm_device_name(pool
->pool_md
), r
);
2598 DMEMIT("%llu %llu/%llu %llu/%llu ",
2599 (unsigned long long)transaction_id
,
2600 (unsigned long long)(nr_blocks_metadata
- nr_free_blocks_metadata
),
2601 (unsigned long long)nr_blocks_metadata
,
2602 (unsigned long long)(nr_blocks_data
- nr_free_blocks_data
),
2603 (unsigned long long)nr_blocks_data
);
2606 DMEMIT("%llu ", held_root
);
2610 if (pool
->pf
.mode
== PM_READ_ONLY
)
2615 if (!pool
->pf
.discard_enabled
)
2616 DMEMIT("ignore_discard");
2617 else if (pool
->pf
.discard_passdown
)
2618 DMEMIT("discard_passdown");
2620 DMEMIT("no_discard_passdown");
2624 case STATUSTYPE_TABLE
:
2625 DMEMIT("%s %s %lu %llu ",
2626 format_dev_t(buf
, pt
->metadata_dev
->bdev
->bd_dev
),
2627 format_dev_t(buf2
, pt
->data_dev
->bdev
->bd_dev
),
2628 (unsigned long)pool
->sectors_per_block
,
2629 (unsigned long long)pt
->low_water_blocks
);
2630 emit_flags(&pt
->requested_pf
, result
, sz
, maxlen
);
2639 static int pool_iterate_devices(struct dm_target
*ti
,
2640 iterate_devices_callout_fn fn
, void *data
)
2642 struct pool_c
*pt
= ti
->private;
2644 return fn(ti
, pt
->data_dev
, 0, ti
->len
, data
);
2647 static int pool_merge(struct dm_target
*ti
, struct bvec_merge_data
*bvm
,
2648 struct bio_vec
*biovec
, int max_size
)
2650 struct pool_c
*pt
= ti
->private;
2651 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
2653 if (!q
->merge_bvec_fn
)
2656 bvm
->bi_bdev
= pt
->data_dev
->bdev
;
2658 return min(max_size
, q
->merge_bvec_fn(q
, bvm
, biovec
));
2661 static void set_discard_limits(struct pool_c
*pt
, struct queue_limits
*limits
)
2663 struct pool
*pool
= pt
->pool
;
2664 struct queue_limits
*data_limits
;
2666 limits
->max_discard_sectors
= pool
->sectors_per_block
;
2669 * discard_granularity is just a hint, and not enforced.
2671 if (pt
->adjusted_pf
.discard_passdown
) {
2672 data_limits
= &bdev_get_queue(pt
->data_dev
->bdev
)->limits
;
2673 limits
->discard_granularity
= data_limits
->discard_granularity
;
2675 limits
->discard_granularity
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
2678 static void pool_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
2680 struct pool_c
*pt
= ti
->private;
2681 struct pool
*pool
= pt
->pool
;
2682 uint64_t io_opt_sectors
= limits
->io_opt
>> SECTOR_SHIFT
;
2685 * If the system-determined stacked limits are compatible with the
2686 * pool's blocksize (io_opt is a factor) do not override them.
2688 if (io_opt_sectors
< pool
->sectors_per_block
||
2689 do_div(io_opt_sectors
, pool
->sectors_per_block
)) {
2690 blk_limits_io_min(limits
, 0);
2691 blk_limits_io_opt(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
2695 * pt->adjusted_pf is a staging area for the actual features to use.
2696 * They get transferred to the live pool in bind_control_target()
2697 * called from pool_preresume().
2699 if (!pt
->adjusted_pf
.discard_enabled
) {
2701 * Must explicitly disallow stacking discard limits otherwise the
2702 * block layer will stack them if pool's data device has support.
2703 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
2704 * user to see that, so make sure to set all discard limits to 0.
2706 limits
->discard_granularity
= 0;
2710 disable_passdown_if_not_supported(pt
);
2712 set_discard_limits(pt
, limits
);
2715 static struct target_type pool_target
= {
2716 .name
= "thin-pool",
2717 .features
= DM_TARGET_SINGLETON
| DM_TARGET_ALWAYS_WRITEABLE
|
2718 DM_TARGET_IMMUTABLE
,
2719 .version
= {1, 9, 0},
2720 .module
= THIS_MODULE
,
2724 .postsuspend
= pool_postsuspend
,
2725 .preresume
= pool_preresume
,
2726 .resume
= pool_resume
,
2727 .message
= pool_message
,
2728 .status
= pool_status
,
2729 .merge
= pool_merge
,
2730 .iterate_devices
= pool_iterate_devices
,
2731 .io_hints
= pool_io_hints
,
2734 /*----------------------------------------------------------------
2735 * Thin target methods
2736 *--------------------------------------------------------------*/
2737 static void thin_dtr(struct dm_target
*ti
)
2739 struct thin_c
*tc
= ti
->private;
2741 mutex_lock(&dm_thin_pool_table
.mutex
);
2743 __pool_dec(tc
->pool
);
2744 dm_pool_close_thin_device(tc
->td
);
2745 dm_put_device(ti
, tc
->pool_dev
);
2747 dm_put_device(ti
, tc
->origin_dev
);
2750 mutex_unlock(&dm_thin_pool_table
.mutex
);
2754 * Thin target parameters:
2756 * <pool_dev> <dev_id> [origin_dev]
2758 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2759 * dev_id: the internal device identifier
2760 * origin_dev: a device external to the pool that should act as the origin
2762 * If the pool device has discards disabled, they get disabled for the thin
2765 static int thin_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2769 struct dm_dev
*pool_dev
, *origin_dev
;
2770 struct mapped_device
*pool_md
;
2772 mutex_lock(&dm_thin_pool_table
.mutex
);
2774 if (argc
!= 2 && argc
!= 3) {
2775 ti
->error
= "Invalid argument count";
2780 tc
= ti
->private = kzalloc(sizeof(*tc
), GFP_KERNEL
);
2782 ti
->error
= "Out of memory";
2788 r
= dm_get_device(ti
, argv
[2], FMODE_READ
, &origin_dev
);
2790 ti
->error
= "Error opening origin device";
2791 goto bad_origin_dev
;
2793 tc
->origin_dev
= origin_dev
;
2796 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &pool_dev
);
2798 ti
->error
= "Error opening pool device";
2801 tc
->pool_dev
= pool_dev
;
2803 if (read_dev_id(argv
[1], (unsigned long long *)&tc
->dev_id
, 0)) {
2804 ti
->error
= "Invalid device id";
2809 pool_md
= dm_get_md(tc
->pool_dev
->bdev
->bd_dev
);
2811 ti
->error
= "Couldn't get pool mapped device";
2816 tc
->pool
= __pool_table_lookup(pool_md
);
2818 ti
->error
= "Couldn't find pool object";
2820 goto bad_pool_lookup
;
2822 __pool_inc(tc
->pool
);
2824 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
2825 ti
->error
= "Couldn't open thin device, Pool is in fail mode";
2829 r
= dm_pool_open_thin_device(tc
->pool
->pmd
, tc
->dev_id
, &tc
->td
);
2831 ti
->error
= "Couldn't open thin internal device";
2835 r
= dm_set_target_max_io_len(ti
, tc
->pool
->sectors_per_block
);
2839 ti
->num_flush_bios
= 1;
2840 ti
->flush_supported
= true;
2841 ti
->per_bio_data_size
= sizeof(struct dm_thin_endio_hook
);
2843 /* In case the pool supports discards, pass them on. */
2844 ti
->discard_zeroes_data_unsupported
= true;
2845 if (tc
->pool
->pf
.discard_enabled
) {
2846 ti
->discards_supported
= true;
2847 ti
->num_discard_bios
= 1;
2848 /* Discard bios must be split on a block boundary */
2849 ti
->split_discard_bios
= true;
2854 mutex_unlock(&dm_thin_pool_table
.mutex
);
2859 __pool_dec(tc
->pool
);
2863 dm_put_device(ti
, tc
->pool_dev
);
2866 dm_put_device(ti
, tc
->origin_dev
);
2870 mutex_unlock(&dm_thin_pool_table
.mutex
);
2875 static int thin_map(struct dm_target
*ti
, struct bio
*bio
)
2877 bio
->bi_sector
= dm_target_offset(ti
, bio
->bi_sector
);
2879 return thin_bio_map(ti
, bio
);
2882 static int thin_endio(struct dm_target
*ti
, struct bio
*bio
, int err
)
2884 unsigned long flags
;
2885 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
2886 struct list_head work
;
2887 struct dm_thin_new_mapping
*m
, *tmp
;
2888 struct pool
*pool
= h
->tc
->pool
;
2890 if (h
->shared_read_entry
) {
2891 INIT_LIST_HEAD(&work
);
2892 dm_deferred_entry_dec(h
->shared_read_entry
, &work
);
2894 spin_lock_irqsave(&pool
->lock
, flags
);
2895 list_for_each_entry_safe(m
, tmp
, &work
, list
) {
2898 __maybe_add_mapping(m
);
2900 spin_unlock_irqrestore(&pool
->lock
, flags
);
2903 if (h
->all_io_entry
) {
2904 INIT_LIST_HEAD(&work
);
2905 dm_deferred_entry_dec(h
->all_io_entry
, &work
);
2906 if (!list_empty(&work
)) {
2907 spin_lock_irqsave(&pool
->lock
, flags
);
2908 list_for_each_entry_safe(m
, tmp
, &work
, list
)
2909 list_add(&m
->list
, &pool
->prepared_discards
);
2910 spin_unlock_irqrestore(&pool
->lock
, flags
);
2918 static void thin_postsuspend(struct dm_target
*ti
)
2920 if (dm_noflush_suspending(ti
))
2921 requeue_io((struct thin_c
*)ti
->private);
2925 * <nr mapped sectors> <highest mapped sector>
2927 static void thin_status(struct dm_target
*ti
, status_type_t type
,
2928 unsigned status_flags
, char *result
, unsigned maxlen
)
2932 dm_block_t mapped
, highest
;
2933 char buf
[BDEVNAME_SIZE
];
2934 struct thin_c
*tc
= ti
->private;
2936 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
2945 case STATUSTYPE_INFO
:
2946 r
= dm_thin_get_mapped_count(tc
->td
, &mapped
);
2948 DMERR("dm_thin_get_mapped_count returned %d", r
);
2952 r
= dm_thin_get_highest_mapped_block(tc
->td
, &highest
);
2954 DMERR("dm_thin_get_highest_mapped_block returned %d", r
);
2958 DMEMIT("%llu ", mapped
* tc
->pool
->sectors_per_block
);
2960 DMEMIT("%llu", ((highest
+ 1) *
2961 tc
->pool
->sectors_per_block
) - 1);
2966 case STATUSTYPE_TABLE
:
2968 format_dev_t(buf
, tc
->pool_dev
->bdev
->bd_dev
),
2969 (unsigned long) tc
->dev_id
);
2971 DMEMIT(" %s", format_dev_t(buf
, tc
->origin_dev
->bdev
->bd_dev
));
2982 static int thin_iterate_devices(struct dm_target
*ti
,
2983 iterate_devices_callout_fn fn
, void *data
)
2986 struct thin_c
*tc
= ti
->private;
2987 struct pool
*pool
= tc
->pool
;
2990 * We can't call dm_pool_get_data_dev_size() since that blocks. So
2991 * we follow a more convoluted path through to the pool's target.
2994 return 0; /* nothing is bound */
2996 blocks
= pool
->ti
->len
;
2997 (void) sector_div(blocks
, pool
->sectors_per_block
);
2999 return fn(ti
, tc
->pool_dev
, 0, pool
->sectors_per_block
* blocks
, data
);
3004 static struct target_type thin_target
= {
3006 .version
= {1, 9, 0},
3007 .module
= THIS_MODULE
,
3011 .end_io
= thin_endio
,
3012 .postsuspend
= thin_postsuspend
,
3013 .status
= thin_status
,
3014 .iterate_devices
= thin_iterate_devices
,
3017 /*----------------------------------------------------------------*/
3019 static int __init
dm_thin_init(void)
3025 r
= dm_register_target(&thin_target
);
3029 r
= dm_register_target(&pool_target
);
3031 goto bad_pool_target
;
3035 _new_mapping_cache
= KMEM_CACHE(dm_thin_new_mapping
, 0);
3036 if (!_new_mapping_cache
)
3037 goto bad_new_mapping_cache
;
3041 bad_new_mapping_cache
:
3042 dm_unregister_target(&pool_target
);
3044 dm_unregister_target(&thin_target
);
3049 static void dm_thin_exit(void)
3051 dm_unregister_target(&thin_target
);
3052 dm_unregister_target(&pool_target
);
3054 kmem_cache_destroy(_new_mapping_cache
);
3057 module_init(dm_thin_init
);
3058 module_exit(dm_thin_exit
);
3060 MODULE_DESCRIPTION(DM_NAME
" thin provisioning target");
3061 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3062 MODULE_LICENSE("GPL");