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;
515 unsigned definitely_not_shared
:1;
518 dm_block_t virt_block
;
519 dm_block_t data_block
;
520 struct dm_bio_prison_cell
*cell
, *cell2
;
524 * If the bio covers the whole area of a block then we can avoid
525 * zeroing or copying. Instead this bio is hooked. The bio will
526 * still be in the cell, so care has to be taken to avoid issuing
530 bio_end_io_t
*saved_bi_end_io
;
533 static void __maybe_add_mapping(struct dm_thin_new_mapping
*m
)
535 struct pool
*pool
= m
->tc
->pool
;
537 if (m
->quiesced
&& m
->prepared
) {
538 list_add(&m
->list
, &pool
->prepared_mappings
);
543 static void copy_complete(int read_err
, unsigned long write_err
, void *context
)
546 struct dm_thin_new_mapping
*m
= context
;
547 struct pool
*pool
= m
->tc
->pool
;
549 m
->err
= read_err
|| write_err
? -EIO
: 0;
551 spin_lock_irqsave(&pool
->lock
, flags
);
553 __maybe_add_mapping(m
);
554 spin_unlock_irqrestore(&pool
->lock
, flags
);
557 static void overwrite_endio(struct bio
*bio
, int err
)
560 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
561 struct dm_thin_new_mapping
*m
= h
->overwrite_mapping
;
562 struct pool
*pool
= m
->tc
->pool
;
566 spin_lock_irqsave(&pool
->lock
, flags
);
568 __maybe_add_mapping(m
);
569 spin_unlock_irqrestore(&pool
->lock
, flags
);
572 /*----------------------------------------------------------------*/
579 * Prepared mapping jobs.
583 * This sends the bios in the cell back to the deferred_bios list.
585 static void cell_defer(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
587 struct pool
*pool
= tc
->pool
;
590 spin_lock_irqsave(&pool
->lock
, flags
);
591 cell_release(pool
, cell
, &pool
->deferred_bios
);
592 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
598 * Same as cell_defer above, except it omits the original holder of the cell.
600 static void cell_defer_no_holder(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
602 struct pool
*pool
= tc
->pool
;
605 spin_lock_irqsave(&pool
->lock
, flags
);
606 cell_release_no_holder(pool
, cell
, &pool
->deferred_bios
);
607 spin_unlock_irqrestore(&pool
->lock
, flags
);
612 static void process_prepared_mapping_fail(struct dm_thin_new_mapping
*m
)
615 m
->bio
->bi_end_io
= m
->saved_bi_end_io
;
616 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 DMERR_LIMIT("%s: dm_thin_insert_block() failed: error = %d",
645 dm_device_name(pool
->pool_md
), r
);
646 set_pool_mode(pool
, PM_READ_ONLY
);
647 cell_error(pool
, m
->cell
);
652 * Release any bios held while the block was being provisioned.
653 * If we are processing a write bio that completely covers the block,
654 * we already processed it so can ignore it now when processing
655 * the bios in the cell.
658 cell_defer_no_holder(tc
, m
->cell
);
661 cell_defer(tc
, m
->cell
);
665 mempool_free(m
, pool
->mapping_pool
);
668 static void process_prepared_discard_fail(struct dm_thin_new_mapping
*m
)
670 struct thin_c
*tc
= m
->tc
;
672 bio_io_error(m
->bio
);
673 cell_defer_no_holder(tc
, m
->cell
);
674 cell_defer_no_holder(tc
, m
->cell2
);
675 mempool_free(m
, tc
->pool
->mapping_pool
);
678 static void process_prepared_discard_passdown(struct dm_thin_new_mapping
*m
)
680 struct thin_c
*tc
= m
->tc
;
682 inc_all_io_entry(tc
->pool
, m
->bio
);
683 cell_defer_no_holder(tc
, m
->cell
);
684 cell_defer_no_holder(tc
, m
->cell2
);
687 if (m
->definitely_not_shared
)
688 remap_and_issue(tc
, m
->bio
, m
->data_block
);
691 if (dm_pool_block_is_used(tc
->pool
->pmd
, m
->data_block
, &used
) || used
)
692 bio_endio(m
->bio
, 0);
694 remap_and_issue(tc
, m
->bio
, m
->data_block
);
697 bio_endio(m
->bio
, 0);
699 mempool_free(m
, tc
->pool
->mapping_pool
);
702 static void process_prepared_discard(struct dm_thin_new_mapping
*m
)
705 struct thin_c
*tc
= m
->tc
;
707 r
= dm_thin_remove_block(tc
->td
, m
->virt_block
);
709 DMERR_LIMIT("dm_thin_remove_block() failed");
711 process_prepared_discard_passdown(m
);
714 static void process_prepared(struct pool
*pool
, struct list_head
*head
,
715 process_mapping_fn
*fn
)
718 struct list_head maps
;
719 struct dm_thin_new_mapping
*m
, *tmp
;
721 INIT_LIST_HEAD(&maps
);
722 spin_lock_irqsave(&pool
->lock
, flags
);
723 list_splice_init(head
, &maps
);
724 spin_unlock_irqrestore(&pool
->lock
, flags
);
726 list_for_each_entry_safe(m
, tmp
, &maps
, list
)
733 static int io_overlaps_block(struct pool
*pool
, struct bio
*bio
)
735 return bio
->bi_size
== (pool
->sectors_per_block
<< SECTOR_SHIFT
);
738 static int io_overwrites_block(struct pool
*pool
, struct bio
*bio
)
740 return (bio_data_dir(bio
) == WRITE
) &&
741 io_overlaps_block(pool
, bio
);
744 static void save_and_set_endio(struct bio
*bio
, bio_end_io_t
**save
,
747 *save
= bio
->bi_end_io
;
751 static int ensure_next_mapping(struct pool
*pool
)
753 if (pool
->next_mapping
)
756 pool
->next_mapping
= mempool_alloc(pool
->mapping_pool
, GFP_ATOMIC
);
758 return pool
->next_mapping
? 0 : -ENOMEM
;
761 static struct dm_thin_new_mapping
*get_next_mapping(struct pool
*pool
)
763 struct dm_thin_new_mapping
*m
= pool
->next_mapping
;
765 BUG_ON(!pool
->next_mapping
);
767 memset(m
, 0, sizeof(struct dm_thin_new_mapping
));
768 INIT_LIST_HEAD(&m
->list
);
771 pool
->next_mapping
= NULL
;
776 static void schedule_copy(struct thin_c
*tc
, dm_block_t virt_block
,
777 struct dm_dev
*origin
, dm_block_t data_origin
,
778 dm_block_t data_dest
,
779 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
782 struct pool
*pool
= tc
->pool
;
783 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
786 m
->virt_block
= virt_block
;
787 m
->data_block
= data_dest
;
790 if (!dm_deferred_set_add_work(pool
->shared_read_ds
, &m
->list
))
794 * IO to pool_dev remaps to the pool target's data_dev.
796 * If the whole block of data is being overwritten, we can issue the
797 * bio immediately. Otherwise we use kcopyd to clone the data first.
799 if (io_overwrites_block(pool
, bio
)) {
800 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
802 h
->overwrite_mapping
= m
;
804 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
805 inc_all_io_entry(pool
, bio
);
806 remap_and_issue(tc
, bio
, data_dest
);
808 struct dm_io_region from
, to
;
810 from
.bdev
= origin
->bdev
;
811 from
.sector
= data_origin
* pool
->sectors_per_block
;
812 from
.count
= pool
->sectors_per_block
;
814 to
.bdev
= tc
->pool_dev
->bdev
;
815 to
.sector
= data_dest
* pool
->sectors_per_block
;
816 to
.count
= pool
->sectors_per_block
;
818 r
= dm_kcopyd_copy(pool
->copier
, &from
, 1, &to
,
819 0, copy_complete
, m
);
821 mempool_free(m
, pool
->mapping_pool
);
822 DMERR_LIMIT("dm_kcopyd_copy() failed");
823 cell_error(pool
, cell
);
828 static void schedule_internal_copy(struct thin_c
*tc
, dm_block_t virt_block
,
829 dm_block_t data_origin
, dm_block_t data_dest
,
830 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
832 schedule_copy(tc
, virt_block
, tc
->pool_dev
,
833 data_origin
, data_dest
, cell
, bio
);
836 static void schedule_external_copy(struct thin_c
*tc
, dm_block_t virt_block
,
837 dm_block_t data_dest
,
838 struct dm_bio_prison_cell
*cell
, struct bio
*bio
)
840 schedule_copy(tc
, virt_block
, tc
->origin_dev
,
841 virt_block
, data_dest
, cell
, bio
);
844 static void schedule_zero(struct thin_c
*tc
, dm_block_t virt_block
,
845 dm_block_t data_block
, struct dm_bio_prison_cell
*cell
,
848 struct pool
*pool
= tc
->pool
;
849 struct dm_thin_new_mapping
*m
= get_next_mapping(pool
);
854 m
->virt_block
= virt_block
;
855 m
->data_block
= data_block
;
859 * If the whole block of data is being overwritten or we are not
860 * zeroing pre-existing data, we can issue the bio immediately.
861 * Otherwise we use kcopyd to zero the data first.
863 if (!pool
->pf
.zero_new_blocks
)
864 process_prepared_mapping(m
);
866 else if (io_overwrites_block(pool
, bio
)) {
867 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
869 h
->overwrite_mapping
= m
;
871 save_and_set_endio(bio
, &m
->saved_bi_end_io
, overwrite_endio
);
872 inc_all_io_entry(pool
, bio
);
873 remap_and_issue(tc
, bio
, data_block
);
876 struct dm_io_region to
;
878 to
.bdev
= tc
->pool_dev
->bdev
;
879 to
.sector
= data_block
* pool
->sectors_per_block
;
880 to
.count
= pool
->sectors_per_block
;
882 r
= dm_kcopyd_zero(pool
->copier
, 1, &to
, 0, copy_complete
, m
);
884 mempool_free(m
, pool
->mapping_pool
);
885 DMERR_LIMIT("dm_kcopyd_zero() failed");
886 cell_error(pool
, cell
);
892 * A non-zero return indicates read_only or fail_io mode.
893 * Many callers don't care about the return value.
895 static int commit(struct pool
*pool
)
899 if (get_pool_mode(pool
) != PM_WRITE
)
902 r
= dm_pool_commit_metadata(pool
->pmd
);
904 DMERR_LIMIT("%s: dm_pool_commit_metadata failed: error = %d",
905 dm_device_name(pool
->pool_md
), r
);
906 set_pool_mode(pool
, PM_READ_ONLY
);
912 static int alloc_data_block(struct thin_c
*tc
, dm_block_t
*result
)
915 dm_block_t free_blocks
;
917 struct pool
*pool
= tc
->pool
;
920 * Once no_free_space is set we must not allow allocation to succeed.
921 * Otherwise it is difficult to explain, debug, test and support.
923 if (pool
->no_free_space
)
926 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
930 if (free_blocks
<= pool
->low_water_blocks
&& !pool
->low_water_triggered
) {
931 DMWARN("%s: reached low water mark for data device: sending event.",
932 dm_device_name(pool
->pool_md
));
933 spin_lock_irqsave(&pool
->lock
, flags
);
934 pool
->low_water_triggered
= 1;
935 spin_unlock_irqrestore(&pool
->lock
, flags
);
936 dm_table_event(pool
->ti
->table
);
941 * Try to commit to see if that will free up some
948 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
953 * If we still have no space we set a flag to avoid
954 * doing all this checking and return -ENOSPC. This
955 * flag serves as a latch that disallows allocations from
956 * this pool until the admin takes action (e.g. resize or
960 DMWARN("%s: no free data space available.",
961 dm_device_name(pool
->pool_md
));
962 spin_lock_irqsave(&pool
->lock
, flags
);
963 pool
->no_free_space
= 1;
964 spin_unlock_irqrestore(&pool
->lock
, flags
);
969 r
= dm_pool_alloc_data_block(pool
->pmd
, result
);
972 !dm_pool_get_free_metadata_block_count(pool
->pmd
, &free_blocks
) &&
974 DMWARN("%s: no free metadata space available.",
975 dm_device_name(pool
->pool_md
));
976 set_pool_mode(pool
, PM_READ_ONLY
);
985 * If we have run out of space, queue bios until the device is
986 * resumed, presumably after having been reloaded with more space.
988 static void retry_on_resume(struct bio
*bio
)
990 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
991 struct thin_c
*tc
= h
->tc
;
992 struct pool
*pool
= tc
->pool
;
995 spin_lock_irqsave(&pool
->lock
, flags
);
996 bio_list_add(&pool
->retry_on_resume_list
, bio
);
997 spin_unlock_irqrestore(&pool
->lock
, flags
);
1000 static void no_space(struct pool
*pool
, struct dm_bio_prison_cell
*cell
)
1003 struct bio_list bios
;
1005 bio_list_init(&bios
);
1006 cell_release(pool
, cell
, &bios
);
1008 while ((bio
= bio_list_pop(&bios
)))
1009 retry_on_resume(bio
);
1012 static void process_discard(struct thin_c
*tc
, struct bio
*bio
)
1015 unsigned long flags
;
1016 struct pool
*pool
= tc
->pool
;
1017 struct dm_bio_prison_cell
*cell
, *cell2
;
1018 struct dm_cell_key key
, key2
;
1019 dm_block_t block
= get_bio_block(tc
, bio
);
1020 struct dm_thin_lookup_result lookup_result
;
1021 struct dm_thin_new_mapping
*m
;
1023 build_virtual_key(tc
->td
, block
, &key
);
1024 if (bio_detain(tc
->pool
, &key
, bio
, &cell
))
1027 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1031 * Check nobody is fiddling with this pool block. This can
1032 * happen if someone's in the process of breaking sharing
1035 build_data_key(tc
->td
, lookup_result
.block
, &key2
);
1036 if (bio_detain(tc
->pool
, &key2
, bio
, &cell2
)) {
1037 cell_defer_no_holder(tc
, cell
);
1041 if (io_overlaps_block(pool
, bio
)) {
1043 * IO may still be going to the destination block. We must
1044 * quiesce before we can do the removal.
1046 m
= get_next_mapping(pool
);
1048 m
->pass_discard
= pool
->pf
.discard_passdown
;
1049 m
->definitely_not_shared
= !lookup_result
.shared
;
1050 m
->virt_block
= block
;
1051 m
->data_block
= lookup_result
.block
;
1056 if (!dm_deferred_set_add_work(pool
->all_io_ds
, &m
->list
)) {
1057 spin_lock_irqsave(&pool
->lock
, flags
);
1058 list_add(&m
->list
, &pool
->prepared_discards
);
1059 spin_unlock_irqrestore(&pool
->lock
, flags
);
1063 inc_all_io_entry(pool
, bio
);
1064 cell_defer_no_holder(tc
, cell
);
1065 cell_defer_no_holder(tc
, cell2
);
1068 * The DM core makes sure that the discard doesn't span
1069 * a block boundary. So we submit the discard of a
1070 * partial block appropriately.
1072 if ((!lookup_result
.shared
) && pool
->pf
.discard_passdown
)
1073 remap_and_issue(tc
, bio
, lookup_result
.block
);
1081 * It isn't provisioned, just forget it.
1083 cell_defer_no_holder(tc
, cell
);
1088 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1090 cell_defer_no_holder(tc
, cell
);
1096 static void break_sharing(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1097 struct dm_cell_key
*key
,
1098 struct dm_thin_lookup_result
*lookup_result
,
1099 struct dm_bio_prison_cell
*cell
)
1102 dm_block_t data_block
;
1103 struct pool
*pool
= tc
->pool
;
1105 r
= alloc_data_block(tc
, &data_block
);
1108 schedule_internal_copy(tc
, block
, lookup_result
->block
,
1109 data_block
, cell
, bio
);
1113 no_space(pool
, cell
);
1117 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1119 set_pool_mode(pool
, PM_READ_ONLY
);
1120 cell_error(pool
, cell
);
1125 static void process_shared_bio(struct thin_c
*tc
, struct bio
*bio
,
1127 struct dm_thin_lookup_result
*lookup_result
)
1129 struct dm_bio_prison_cell
*cell
;
1130 struct pool
*pool
= tc
->pool
;
1131 struct dm_cell_key key
;
1134 * If cell is already occupied, then sharing is already in the process
1135 * of being broken so we have nothing further to do here.
1137 build_data_key(tc
->td
, lookup_result
->block
, &key
);
1138 if (bio_detain(pool
, &key
, bio
, &cell
))
1141 if (bio_data_dir(bio
) == WRITE
&& bio
->bi_size
)
1142 break_sharing(tc
, bio
, block
, &key
, lookup_result
, cell
);
1144 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1146 h
->shared_read_entry
= dm_deferred_entry_inc(pool
->shared_read_ds
);
1147 inc_all_io_entry(pool
, bio
);
1148 cell_defer_no_holder(tc
, cell
);
1150 remap_and_issue(tc
, bio
, lookup_result
->block
);
1154 static void provision_block(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1155 struct dm_bio_prison_cell
*cell
)
1158 dm_block_t data_block
;
1159 struct pool
*pool
= tc
->pool
;
1162 * Remap empty bios (flushes) immediately, without provisioning.
1164 if (!bio
->bi_size
) {
1165 inc_all_io_entry(pool
, bio
);
1166 cell_defer_no_holder(tc
, cell
);
1168 remap_and_issue(tc
, bio
, 0);
1173 * Fill read bios with zeroes and complete them immediately.
1175 if (bio_data_dir(bio
) == READ
) {
1177 cell_defer_no_holder(tc
, cell
);
1182 r
= alloc_data_block(tc
, &data_block
);
1186 schedule_external_copy(tc
, block
, data_block
, cell
, bio
);
1188 schedule_zero(tc
, block
, data_block
, cell
, bio
);
1192 no_space(pool
, cell
);
1196 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1198 set_pool_mode(pool
, PM_READ_ONLY
);
1199 cell_error(pool
, cell
);
1204 static void process_bio(struct thin_c
*tc
, struct bio
*bio
)
1207 struct pool
*pool
= tc
->pool
;
1208 dm_block_t block
= get_bio_block(tc
, bio
);
1209 struct dm_bio_prison_cell
*cell
;
1210 struct dm_cell_key key
;
1211 struct dm_thin_lookup_result lookup_result
;
1214 * If cell is already occupied, then the block is already
1215 * being provisioned so we have nothing further to do here.
1217 build_virtual_key(tc
->td
, block
, &key
);
1218 if (bio_detain(pool
, &key
, bio
, &cell
))
1221 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1224 if (lookup_result
.shared
) {
1225 process_shared_bio(tc
, bio
, block
, &lookup_result
);
1226 cell_defer_no_holder(tc
, cell
); /* FIXME: pass this cell into process_shared? */
1228 inc_all_io_entry(pool
, bio
);
1229 cell_defer_no_holder(tc
, cell
);
1231 remap_and_issue(tc
, bio
, lookup_result
.block
);
1236 if (bio_data_dir(bio
) == READ
&& tc
->origin_dev
) {
1237 inc_all_io_entry(pool
, bio
);
1238 cell_defer_no_holder(tc
, cell
);
1240 remap_to_origin_and_issue(tc
, bio
);
1242 provision_block(tc
, bio
, block
, cell
);
1246 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1248 cell_defer_no_holder(tc
, cell
);
1254 static void process_bio_read_only(struct thin_c
*tc
, struct bio
*bio
)
1257 int rw
= bio_data_dir(bio
);
1258 dm_block_t block
= get_bio_block(tc
, bio
);
1259 struct dm_thin_lookup_result lookup_result
;
1261 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1264 if (lookup_result
.shared
&& (rw
== WRITE
) && bio
->bi_size
)
1267 inc_all_io_entry(tc
->pool
, bio
);
1268 remap_and_issue(tc
, bio
, lookup_result
.block
);
1278 if (tc
->origin_dev
) {
1279 inc_all_io_entry(tc
->pool
, bio
);
1280 remap_to_origin_and_issue(tc
, bio
);
1289 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1296 static void process_bio_fail(struct thin_c
*tc
, struct bio
*bio
)
1302 * FIXME: should we also commit due to size of transaction, measured in
1305 static int need_commit_due_to_time(struct pool
*pool
)
1307 return jiffies
< pool
->last_commit_jiffies
||
1308 jiffies
> pool
->last_commit_jiffies
+ COMMIT_PERIOD
;
1311 static void process_deferred_bios(struct pool
*pool
)
1313 unsigned long flags
;
1315 struct bio_list bios
;
1317 bio_list_init(&bios
);
1319 spin_lock_irqsave(&pool
->lock
, flags
);
1320 bio_list_merge(&bios
, &pool
->deferred_bios
);
1321 bio_list_init(&pool
->deferred_bios
);
1322 spin_unlock_irqrestore(&pool
->lock
, flags
);
1324 while ((bio
= bio_list_pop(&bios
))) {
1325 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1326 struct thin_c
*tc
= h
->tc
;
1329 * If we've got no free new_mapping structs, and processing
1330 * this bio might require one, we pause until there are some
1331 * prepared mappings to process.
1333 if (ensure_next_mapping(pool
)) {
1334 spin_lock_irqsave(&pool
->lock
, flags
);
1335 bio_list_merge(&pool
->deferred_bios
, &bios
);
1336 spin_unlock_irqrestore(&pool
->lock
, flags
);
1341 if (bio
->bi_rw
& REQ_DISCARD
)
1342 pool
->process_discard(tc
, bio
);
1344 pool
->process_bio(tc
, bio
);
1348 * If there are any deferred flush bios, we must commit
1349 * the metadata before issuing them.
1351 bio_list_init(&bios
);
1352 spin_lock_irqsave(&pool
->lock
, flags
);
1353 bio_list_merge(&bios
, &pool
->deferred_flush_bios
);
1354 bio_list_init(&pool
->deferred_flush_bios
);
1355 spin_unlock_irqrestore(&pool
->lock
, flags
);
1357 if (bio_list_empty(&bios
) && !need_commit_due_to_time(pool
))
1361 while ((bio
= bio_list_pop(&bios
)))
1365 pool
->last_commit_jiffies
= jiffies
;
1367 while ((bio
= bio_list_pop(&bios
)))
1368 generic_make_request(bio
);
1371 static void do_worker(struct work_struct
*ws
)
1373 struct pool
*pool
= container_of(ws
, struct pool
, worker
);
1375 process_prepared(pool
, &pool
->prepared_mappings
, &pool
->process_prepared_mapping
);
1376 process_prepared(pool
, &pool
->prepared_discards
, &pool
->process_prepared_discard
);
1377 process_deferred_bios(pool
);
1381 * We want to commit periodically so that not too much
1382 * unwritten data builds up.
1384 static void do_waker(struct work_struct
*ws
)
1386 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
, waker
);
1388 queue_delayed_work(pool
->wq
, &pool
->waker
, COMMIT_PERIOD
);
1391 /*----------------------------------------------------------------*/
1393 static enum pool_mode
get_pool_mode(struct pool
*pool
)
1395 return pool
->pf
.mode
;
1398 static void set_pool_mode(struct pool
*pool
, enum pool_mode mode
)
1402 pool
->pf
.mode
= mode
;
1406 DMERR("%s: switching pool to failure mode",
1407 dm_device_name(pool
->pool_md
));
1408 dm_pool_metadata_read_only(pool
->pmd
);
1409 pool
->process_bio
= process_bio_fail
;
1410 pool
->process_discard
= process_bio_fail
;
1411 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1412 pool
->process_prepared_discard
= process_prepared_discard_fail
;
1416 DMERR("%s: switching pool to read-only mode",
1417 dm_device_name(pool
->pool_md
));
1418 r
= dm_pool_abort_metadata(pool
->pmd
);
1420 DMERR("%s: aborting transaction failed",
1421 dm_device_name(pool
->pool_md
));
1422 set_pool_mode(pool
, PM_FAIL
);
1424 dm_pool_metadata_read_only(pool
->pmd
);
1425 pool
->process_bio
= process_bio_read_only
;
1426 pool
->process_discard
= process_discard
;
1427 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1428 pool
->process_prepared_discard
= process_prepared_discard_passdown
;
1433 dm_pool_metadata_read_write(pool
->pmd
);
1434 pool
->process_bio
= process_bio
;
1435 pool
->process_discard
= process_discard
;
1436 pool
->process_prepared_mapping
= process_prepared_mapping
;
1437 pool
->process_prepared_discard
= process_prepared_discard
;
1442 /*----------------------------------------------------------------*/
1445 * Mapping functions.
1449 * Called only while mapping a thin bio to hand it over to the workqueue.
1451 static void thin_defer_bio(struct thin_c
*tc
, struct bio
*bio
)
1453 unsigned long flags
;
1454 struct pool
*pool
= tc
->pool
;
1456 spin_lock_irqsave(&pool
->lock
, flags
);
1457 bio_list_add(&pool
->deferred_bios
, bio
);
1458 spin_unlock_irqrestore(&pool
->lock
, flags
);
1463 static void thin_hook_bio(struct thin_c
*tc
, struct bio
*bio
)
1465 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1468 h
->shared_read_entry
= NULL
;
1469 h
->all_io_entry
= NULL
;
1470 h
->overwrite_mapping
= NULL
;
1474 * Non-blocking function called from the thin target's map function.
1476 static int thin_bio_map(struct dm_target
*ti
, struct bio
*bio
)
1479 struct thin_c
*tc
= ti
->private;
1480 dm_block_t block
= get_bio_block(tc
, bio
);
1481 struct dm_thin_device
*td
= tc
->td
;
1482 struct dm_thin_lookup_result result
;
1483 struct dm_bio_prison_cell cell1
, cell2
;
1484 struct dm_bio_prison_cell
*cell_result
;
1485 struct dm_cell_key key
;
1487 thin_hook_bio(tc
, bio
);
1489 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
1491 return DM_MAPIO_SUBMITTED
;
1494 if (bio
->bi_rw
& (REQ_DISCARD
| REQ_FLUSH
| REQ_FUA
)) {
1495 thin_defer_bio(tc
, bio
);
1496 return DM_MAPIO_SUBMITTED
;
1499 r
= dm_thin_find_block(td
, block
, 0, &result
);
1502 * Note that we defer readahead too.
1506 if (unlikely(result
.shared
)) {
1508 * We have a race condition here between the
1509 * result.shared value returned by the lookup and
1510 * snapshot creation, which may cause new
1513 * To avoid this always quiesce the origin before
1514 * taking the snap. You want to do this anyway to
1515 * ensure a consistent application view
1518 * More distant ancestors are irrelevant. The
1519 * shared flag will be set in their case.
1521 thin_defer_bio(tc
, bio
);
1522 return DM_MAPIO_SUBMITTED
;
1525 build_virtual_key(tc
->td
, block
, &key
);
1526 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell1
, &cell_result
))
1527 return DM_MAPIO_SUBMITTED
;
1529 build_data_key(tc
->td
, result
.block
, &key
);
1530 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell2
, &cell_result
)) {
1531 cell_defer_no_holder_no_free(tc
, &cell1
);
1532 return DM_MAPIO_SUBMITTED
;
1535 inc_all_io_entry(tc
->pool
, bio
);
1536 cell_defer_no_holder_no_free(tc
, &cell2
);
1537 cell_defer_no_holder_no_free(tc
, &cell1
);
1539 remap(tc
, bio
, result
.block
);
1540 return DM_MAPIO_REMAPPED
;
1543 if (get_pool_mode(tc
->pool
) == PM_READ_ONLY
) {
1545 * This block isn't provisioned, and we have no way
1546 * of doing so. Just error it.
1549 return DM_MAPIO_SUBMITTED
;
1555 * In future, the failed dm_thin_find_block above could
1556 * provide the hint to load the metadata into cache.
1558 thin_defer_bio(tc
, bio
);
1559 return DM_MAPIO_SUBMITTED
;
1563 * Must always call bio_io_error on failure.
1564 * dm_thin_find_block can fail with -EINVAL if the
1565 * pool is switched to fail-io mode.
1568 return DM_MAPIO_SUBMITTED
;
1572 static int pool_is_congested(struct dm_target_callbacks
*cb
, int bdi_bits
)
1575 unsigned long flags
;
1576 struct pool_c
*pt
= container_of(cb
, struct pool_c
, callbacks
);
1578 spin_lock_irqsave(&pt
->pool
->lock
, flags
);
1579 r
= !bio_list_empty(&pt
->pool
->retry_on_resume_list
);
1580 spin_unlock_irqrestore(&pt
->pool
->lock
, flags
);
1583 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1584 r
= bdi_congested(&q
->backing_dev_info
, bdi_bits
);
1590 static void __requeue_bios(struct pool
*pool
)
1592 bio_list_merge(&pool
->deferred_bios
, &pool
->retry_on_resume_list
);
1593 bio_list_init(&pool
->retry_on_resume_list
);
1596 /*----------------------------------------------------------------
1597 * Binding of control targets to a pool object
1598 *--------------------------------------------------------------*/
1599 static bool data_dev_supports_discard(struct pool_c
*pt
)
1601 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1603 return q
&& blk_queue_discard(q
);
1606 static bool is_factor(sector_t block_size
, uint32_t n
)
1608 return !sector_div(block_size
, n
);
1612 * If discard_passdown was enabled verify that the data device
1613 * supports discards. Disable discard_passdown if not.
1615 static void disable_passdown_if_not_supported(struct pool_c
*pt
)
1617 struct pool
*pool
= pt
->pool
;
1618 struct block_device
*data_bdev
= pt
->data_dev
->bdev
;
1619 struct queue_limits
*data_limits
= &bdev_get_queue(data_bdev
)->limits
;
1620 sector_t block_size
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
1621 const char *reason
= NULL
;
1622 char buf
[BDEVNAME_SIZE
];
1624 if (!pt
->adjusted_pf
.discard_passdown
)
1627 if (!data_dev_supports_discard(pt
))
1628 reason
= "discard unsupported";
1630 else if (data_limits
->max_discard_sectors
< pool
->sectors_per_block
)
1631 reason
= "max discard sectors smaller than a block";
1633 else if (data_limits
->discard_granularity
> block_size
)
1634 reason
= "discard granularity larger than a block";
1636 else if (!is_factor(block_size
, data_limits
->discard_granularity
))
1637 reason
= "discard granularity not a factor of block size";
1640 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev
, buf
), reason
);
1641 pt
->adjusted_pf
.discard_passdown
= false;
1645 static int bind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1647 struct pool_c
*pt
= ti
->private;
1650 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
1652 enum pool_mode old_mode
= pool
->pf
.mode
;
1653 enum pool_mode new_mode
= pt
->adjusted_pf
.mode
;
1656 * If we were in PM_FAIL mode, rollback of metadata failed. We're
1657 * not going to recover without a thin_repair. So we never let the
1658 * pool move out of the old mode. On the other hand a PM_READ_ONLY
1659 * may have been due to a lack of metadata or data space, and may
1660 * now work (ie. if the underlying devices have been resized).
1662 if (old_mode
== PM_FAIL
)
1663 new_mode
= old_mode
;
1666 pool
->low_water_blocks
= pt
->low_water_blocks
;
1667 pool
->pf
= pt
->adjusted_pf
;
1669 set_pool_mode(pool
, new_mode
);
1674 static void unbind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1680 /*----------------------------------------------------------------
1682 *--------------------------------------------------------------*/
1683 /* Initialize pool features. */
1684 static void pool_features_init(struct pool_features
*pf
)
1686 pf
->mode
= PM_WRITE
;
1687 pf
->zero_new_blocks
= true;
1688 pf
->discard_enabled
= true;
1689 pf
->discard_passdown
= true;
1692 static void __pool_destroy(struct pool
*pool
)
1694 __pool_table_remove(pool
);
1696 if (dm_pool_metadata_close(pool
->pmd
) < 0)
1697 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1699 dm_bio_prison_destroy(pool
->prison
);
1700 dm_kcopyd_client_destroy(pool
->copier
);
1703 destroy_workqueue(pool
->wq
);
1705 if (pool
->next_mapping
)
1706 mempool_free(pool
->next_mapping
, pool
->mapping_pool
);
1707 mempool_destroy(pool
->mapping_pool
);
1708 dm_deferred_set_destroy(pool
->shared_read_ds
);
1709 dm_deferred_set_destroy(pool
->all_io_ds
);
1713 static struct kmem_cache
*_new_mapping_cache
;
1715 static struct pool
*pool_create(struct mapped_device
*pool_md
,
1716 struct block_device
*metadata_dev
,
1717 unsigned long block_size
,
1718 int read_only
, char **error
)
1723 struct dm_pool_metadata
*pmd
;
1724 bool format_device
= read_only
? false : true;
1726 pmd
= dm_pool_metadata_open(metadata_dev
, block_size
, format_device
);
1728 *error
= "Error creating metadata object";
1729 return (struct pool
*)pmd
;
1732 pool
= kmalloc(sizeof(*pool
), GFP_KERNEL
);
1734 *error
= "Error allocating memory for pool";
1735 err_p
= ERR_PTR(-ENOMEM
);
1740 pool
->sectors_per_block
= block_size
;
1741 if (block_size
& (block_size
- 1))
1742 pool
->sectors_per_block_shift
= -1;
1744 pool
->sectors_per_block_shift
= __ffs(block_size
);
1745 pool
->low_water_blocks
= 0;
1746 pool_features_init(&pool
->pf
);
1747 pool
->prison
= dm_bio_prison_create(PRISON_CELLS
);
1748 if (!pool
->prison
) {
1749 *error
= "Error creating pool's bio prison";
1750 err_p
= ERR_PTR(-ENOMEM
);
1754 pool
->copier
= dm_kcopyd_client_create(&dm_kcopyd_throttle
);
1755 if (IS_ERR(pool
->copier
)) {
1756 r
= PTR_ERR(pool
->copier
);
1757 *error
= "Error creating pool's kcopyd client";
1759 goto bad_kcopyd_client
;
1763 * Create singlethreaded workqueue that will service all devices
1764 * that use this metadata.
1766 pool
->wq
= alloc_ordered_workqueue("dm-" DM_MSG_PREFIX
, WQ_MEM_RECLAIM
);
1768 *error
= "Error creating pool's workqueue";
1769 err_p
= ERR_PTR(-ENOMEM
);
1773 INIT_WORK(&pool
->worker
, do_worker
);
1774 INIT_DELAYED_WORK(&pool
->waker
, do_waker
);
1775 spin_lock_init(&pool
->lock
);
1776 bio_list_init(&pool
->deferred_bios
);
1777 bio_list_init(&pool
->deferred_flush_bios
);
1778 INIT_LIST_HEAD(&pool
->prepared_mappings
);
1779 INIT_LIST_HEAD(&pool
->prepared_discards
);
1780 pool
->low_water_triggered
= 0;
1781 pool
->no_free_space
= 0;
1782 bio_list_init(&pool
->retry_on_resume_list
);
1784 pool
->shared_read_ds
= dm_deferred_set_create();
1785 if (!pool
->shared_read_ds
) {
1786 *error
= "Error creating pool's shared read deferred set";
1787 err_p
= ERR_PTR(-ENOMEM
);
1788 goto bad_shared_read_ds
;
1791 pool
->all_io_ds
= dm_deferred_set_create();
1792 if (!pool
->all_io_ds
) {
1793 *error
= "Error creating pool's all io deferred set";
1794 err_p
= ERR_PTR(-ENOMEM
);
1798 pool
->next_mapping
= NULL
;
1799 pool
->mapping_pool
= mempool_create_slab_pool(MAPPING_POOL_SIZE
,
1800 _new_mapping_cache
);
1801 if (!pool
->mapping_pool
) {
1802 *error
= "Error creating pool's mapping mempool";
1803 err_p
= ERR_PTR(-ENOMEM
);
1804 goto bad_mapping_pool
;
1807 pool
->ref_count
= 1;
1808 pool
->last_commit_jiffies
= jiffies
;
1809 pool
->pool_md
= pool_md
;
1810 pool
->md_dev
= metadata_dev
;
1811 __pool_table_insert(pool
);
1816 dm_deferred_set_destroy(pool
->all_io_ds
);
1818 dm_deferred_set_destroy(pool
->shared_read_ds
);
1820 destroy_workqueue(pool
->wq
);
1822 dm_kcopyd_client_destroy(pool
->copier
);
1824 dm_bio_prison_destroy(pool
->prison
);
1828 if (dm_pool_metadata_close(pmd
))
1829 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1834 static void __pool_inc(struct pool
*pool
)
1836 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
1840 static void __pool_dec(struct pool
*pool
)
1842 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
1843 BUG_ON(!pool
->ref_count
);
1844 if (!--pool
->ref_count
)
1845 __pool_destroy(pool
);
1848 static struct pool
*__pool_find(struct mapped_device
*pool_md
,
1849 struct block_device
*metadata_dev
,
1850 unsigned long block_size
, int read_only
,
1851 char **error
, int *created
)
1853 struct pool
*pool
= __pool_table_lookup_metadata_dev(metadata_dev
);
1856 if (pool
->pool_md
!= pool_md
) {
1857 *error
= "metadata device already in use by a pool";
1858 return ERR_PTR(-EBUSY
);
1863 pool
= __pool_table_lookup(pool_md
);
1865 if (pool
->md_dev
!= metadata_dev
) {
1866 *error
= "different pool cannot replace a pool";
1867 return ERR_PTR(-EINVAL
);
1872 pool
= pool_create(pool_md
, metadata_dev
, block_size
, read_only
, error
);
1880 /*----------------------------------------------------------------
1881 * Pool target methods
1882 *--------------------------------------------------------------*/
1883 static void pool_dtr(struct dm_target
*ti
)
1885 struct pool_c
*pt
= ti
->private;
1887 mutex_lock(&dm_thin_pool_table
.mutex
);
1889 unbind_control_target(pt
->pool
, ti
);
1890 __pool_dec(pt
->pool
);
1891 dm_put_device(ti
, pt
->metadata_dev
);
1892 dm_put_device(ti
, pt
->data_dev
);
1895 mutex_unlock(&dm_thin_pool_table
.mutex
);
1898 static int parse_pool_features(struct dm_arg_set
*as
, struct pool_features
*pf
,
1899 struct dm_target
*ti
)
1903 const char *arg_name
;
1905 static struct dm_arg _args
[] = {
1906 {0, 3, "Invalid number of pool feature arguments"},
1910 * No feature arguments supplied.
1915 r
= dm_read_arg_group(_args
, as
, &argc
, &ti
->error
);
1919 while (argc
&& !r
) {
1920 arg_name
= dm_shift_arg(as
);
1923 if (!strcasecmp(arg_name
, "skip_block_zeroing"))
1924 pf
->zero_new_blocks
= false;
1926 else if (!strcasecmp(arg_name
, "ignore_discard"))
1927 pf
->discard_enabled
= false;
1929 else if (!strcasecmp(arg_name
, "no_discard_passdown"))
1930 pf
->discard_passdown
= false;
1932 else if (!strcasecmp(arg_name
, "read_only"))
1933 pf
->mode
= PM_READ_ONLY
;
1936 ti
->error
= "Unrecognised pool feature requested";
1945 static void metadata_low_callback(void *context
)
1947 struct pool
*pool
= context
;
1949 DMWARN("%s: reached low water mark for metadata device: sending event.",
1950 dm_device_name(pool
->pool_md
));
1952 dm_table_event(pool
->ti
->table
);
1955 static sector_t
get_metadata_dev_size(struct block_device
*bdev
)
1957 sector_t metadata_dev_size
= i_size_read(bdev
->bd_inode
) >> SECTOR_SHIFT
;
1958 char buffer
[BDEVNAME_SIZE
];
1960 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS_WARNING
) {
1961 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
1962 bdevname(bdev
, buffer
), THIN_METADATA_MAX_SECTORS
);
1963 metadata_dev_size
= THIN_METADATA_MAX_SECTORS_WARNING
;
1966 return metadata_dev_size
;
1969 static dm_block_t
get_metadata_dev_size_in_blocks(struct block_device
*bdev
)
1971 sector_t metadata_dev_size
= get_metadata_dev_size(bdev
);
1973 sector_div(metadata_dev_size
, THIN_METADATA_BLOCK_SIZE
>> SECTOR_SHIFT
);
1975 return metadata_dev_size
;
1979 * When a metadata threshold is crossed a dm event is triggered, and
1980 * userland should respond by growing the metadata device. We could let
1981 * userland set the threshold, like we do with the data threshold, but I'm
1982 * not sure they know enough to do this well.
1984 static dm_block_t
calc_metadata_threshold(struct pool_c
*pt
)
1987 * 4M is ample for all ops with the possible exception of thin
1988 * device deletion which is harmless if it fails (just retry the
1989 * delete after you've grown the device).
1991 dm_block_t quarter
= get_metadata_dev_size_in_blocks(pt
->metadata_dev
->bdev
) / 4;
1992 return min((dm_block_t
)1024ULL /* 4M */, quarter
);
1996 * thin-pool <metadata dev> <data dev>
1997 * <data block size (sectors)>
1998 * <low water mark (blocks)>
1999 * [<#feature args> [<arg>]*]
2001 * Optional feature arguments are:
2002 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
2003 * ignore_discard: disable discard
2004 * no_discard_passdown: don't pass discards down to the data device
2006 static int pool_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2008 int r
, pool_created
= 0;
2011 struct pool_features pf
;
2012 struct dm_arg_set as
;
2013 struct dm_dev
*data_dev
;
2014 unsigned long block_size
;
2015 dm_block_t low_water_blocks
;
2016 struct dm_dev
*metadata_dev
;
2017 fmode_t metadata_mode
;
2020 * FIXME Remove validation from scope of lock.
2022 mutex_lock(&dm_thin_pool_table
.mutex
);
2025 ti
->error
= "Invalid argument count";
2034 * Set default pool features.
2036 pool_features_init(&pf
);
2038 dm_consume_args(&as
, 4);
2039 r
= parse_pool_features(&as
, &pf
, ti
);
2043 metadata_mode
= FMODE_READ
| ((pf
.mode
== PM_READ_ONLY
) ? 0 : FMODE_WRITE
);
2044 r
= dm_get_device(ti
, argv
[0], metadata_mode
, &metadata_dev
);
2046 ti
->error
= "Error opening metadata block device";
2051 * Run for the side-effect of possibly issuing a warning if the
2052 * device is too big.
2054 (void) get_metadata_dev_size(metadata_dev
->bdev
);
2056 r
= dm_get_device(ti
, argv
[1], FMODE_READ
| FMODE_WRITE
, &data_dev
);
2058 ti
->error
= "Error getting data device";
2062 if (kstrtoul(argv
[2], 10, &block_size
) || !block_size
||
2063 block_size
< DATA_DEV_BLOCK_SIZE_MIN_SECTORS
||
2064 block_size
> DATA_DEV_BLOCK_SIZE_MAX_SECTORS
||
2065 block_size
& (DATA_DEV_BLOCK_SIZE_MIN_SECTORS
- 1)) {
2066 ti
->error
= "Invalid block size";
2071 if (kstrtoull(argv
[3], 10, (unsigned long long *)&low_water_blocks
)) {
2072 ti
->error
= "Invalid low water mark";
2077 pt
= kzalloc(sizeof(*pt
), GFP_KERNEL
);
2083 pool
= __pool_find(dm_table_get_md(ti
->table
), metadata_dev
->bdev
,
2084 block_size
, pf
.mode
== PM_READ_ONLY
, &ti
->error
, &pool_created
);
2091 * 'pool_created' reflects whether this is the first table load.
2092 * Top level discard support is not allowed to be changed after
2093 * initial load. This would require a pool reload to trigger thin
2096 if (!pool_created
&& pf
.discard_enabled
!= pool
->pf
.discard_enabled
) {
2097 ti
->error
= "Discard support cannot be disabled once enabled";
2099 goto out_flags_changed
;
2104 pt
->metadata_dev
= metadata_dev
;
2105 pt
->data_dev
= data_dev
;
2106 pt
->low_water_blocks
= low_water_blocks
;
2107 pt
->adjusted_pf
= pt
->requested_pf
= pf
;
2108 ti
->num_flush_bios
= 1;
2111 * Only need to enable discards if the pool should pass
2112 * them down to the data device. The thin device's discard
2113 * processing will cause mappings to be removed from the btree.
2115 ti
->discard_zeroes_data_unsupported
= true;
2116 if (pf
.discard_enabled
&& pf
.discard_passdown
) {
2117 ti
->num_discard_bios
= 1;
2120 * Setting 'discards_supported' circumvents the normal
2121 * stacking of discard limits (this keeps the pool and
2122 * thin devices' discard limits consistent).
2124 ti
->discards_supported
= true;
2128 r
= dm_pool_register_metadata_threshold(pt
->pool
->pmd
,
2129 calc_metadata_threshold(pt
),
2130 metadata_low_callback
,
2135 pt
->callbacks
.congested_fn
= pool_is_congested
;
2136 dm_table_add_target_callbacks(ti
->table
, &pt
->callbacks
);
2138 mutex_unlock(&dm_thin_pool_table
.mutex
);
2147 dm_put_device(ti
, data_dev
);
2149 dm_put_device(ti
, metadata_dev
);
2151 mutex_unlock(&dm_thin_pool_table
.mutex
);
2156 static int pool_map(struct dm_target
*ti
, struct bio
*bio
)
2159 struct pool_c
*pt
= ti
->private;
2160 struct pool
*pool
= pt
->pool
;
2161 unsigned long flags
;
2164 * As this is a singleton target, ti->begin is always zero.
2166 spin_lock_irqsave(&pool
->lock
, flags
);
2167 bio
->bi_bdev
= pt
->data_dev
->bdev
;
2168 r
= DM_MAPIO_REMAPPED
;
2169 spin_unlock_irqrestore(&pool
->lock
, flags
);
2174 static int maybe_resize_data_dev(struct dm_target
*ti
, bool *need_commit
)
2177 struct pool_c
*pt
= ti
->private;
2178 struct pool
*pool
= pt
->pool
;
2179 sector_t data_size
= ti
->len
;
2180 dm_block_t sb_data_size
;
2182 *need_commit
= false;
2184 (void) sector_div(data_size
, pool
->sectors_per_block
);
2186 r
= dm_pool_get_data_dev_size(pool
->pmd
, &sb_data_size
);
2188 DMERR("%s: failed to retrieve data device size",
2189 dm_device_name(pool
->pool_md
));
2193 if (data_size
< sb_data_size
) {
2194 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
2195 dm_device_name(pool
->pool_md
),
2196 (unsigned long long)data_size
, sb_data_size
);
2199 } else if (data_size
> sb_data_size
) {
2200 r
= dm_pool_resize_data_dev(pool
->pmd
, data_size
);
2202 DMERR("%s: failed to resize data device",
2203 dm_device_name(pool
->pool_md
));
2204 set_pool_mode(pool
, PM_READ_ONLY
);
2208 *need_commit
= true;
2214 static int maybe_resize_metadata_dev(struct dm_target
*ti
, bool *need_commit
)
2217 struct pool_c
*pt
= ti
->private;
2218 struct pool
*pool
= pt
->pool
;
2219 dm_block_t metadata_dev_size
, sb_metadata_dev_size
;
2221 *need_commit
= false;
2223 metadata_dev_size
= get_metadata_dev_size_in_blocks(pool
->md_dev
);
2225 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &sb_metadata_dev_size
);
2227 DMERR("%s: failed to retrieve metadata device size",
2228 dm_device_name(pool
->pool_md
));
2232 if (metadata_dev_size
< sb_metadata_dev_size
) {
2233 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
2234 dm_device_name(pool
->pool_md
),
2235 metadata_dev_size
, sb_metadata_dev_size
);
2238 } else if (metadata_dev_size
> sb_metadata_dev_size
) {
2239 r
= dm_pool_resize_metadata_dev(pool
->pmd
, metadata_dev_size
);
2241 DMERR("%s: failed to resize metadata device",
2242 dm_device_name(pool
->pool_md
));
2246 *need_commit
= true;
2253 * Retrieves the number of blocks of the data device from
2254 * the superblock and compares it to the actual device size,
2255 * thus resizing the data device in case it has grown.
2257 * This both copes with opening preallocated data devices in the ctr
2258 * being followed by a resume
2260 * calling the resume method individually after userspace has
2261 * grown the data device in reaction to a table event.
2263 static int pool_preresume(struct dm_target
*ti
)
2266 bool need_commit1
, need_commit2
;
2267 struct pool_c
*pt
= ti
->private;
2268 struct pool
*pool
= pt
->pool
;
2271 * Take control of the pool object.
2273 r
= bind_control_target(pool
, ti
);
2277 r
= maybe_resize_data_dev(ti
, &need_commit1
);
2281 r
= maybe_resize_metadata_dev(ti
, &need_commit2
);
2285 if (need_commit1
|| need_commit2
)
2286 (void) commit(pool
);
2291 static void pool_resume(struct dm_target
*ti
)
2293 struct pool_c
*pt
= ti
->private;
2294 struct pool
*pool
= pt
->pool
;
2295 unsigned long flags
;
2297 spin_lock_irqsave(&pool
->lock
, flags
);
2298 pool
->low_water_triggered
= 0;
2299 pool
->no_free_space
= 0;
2300 __requeue_bios(pool
);
2301 spin_unlock_irqrestore(&pool
->lock
, flags
);
2303 do_waker(&pool
->waker
.work
);
2306 static void pool_postsuspend(struct dm_target
*ti
)
2308 struct pool_c
*pt
= ti
->private;
2309 struct pool
*pool
= pt
->pool
;
2311 cancel_delayed_work(&pool
->waker
);
2312 flush_workqueue(pool
->wq
);
2313 (void) commit(pool
);
2316 static int check_arg_count(unsigned argc
, unsigned args_required
)
2318 if (argc
!= args_required
) {
2319 DMWARN("Message received with %u arguments instead of %u.",
2320 argc
, args_required
);
2327 static int read_dev_id(char *arg
, dm_thin_id
*dev_id
, int warning
)
2329 if (!kstrtoull(arg
, 10, (unsigned long long *)dev_id
) &&
2330 *dev_id
<= MAX_DEV_ID
)
2334 DMWARN("Message received with invalid device id: %s", arg
);
2339 static int process_create_thin_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2344 r
= check_arg_count(argc
, 2);
2348 r
= read_dev_id(argv
[1], &dev_id
, 1);
2352 r
= dm_pool_create_thin(pool
->pmd
, dev_id
);
2354 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2362 static int process_create_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2365 dm_thin_id origin_dev_id
;
2368 r
= check_arg_count(argc
, 3);
2372 r
= read_dev_id(argv
[1], &dev_id
, 1);
2376 r
= read_dev_id(argv
[2], &origin_dev_id
, 1);
2380 r
= dm_pool_create_snap(pool
->pmd
, dev_id
, origin_dev_id
);
2382 DMWARN("Creation of new snapshot %s of device %s failed.",
2390 static int process_delete_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2395 r
= check_arg_count(argc
, 2);
2399 r
= read_dev_id(argv
[1], &dev_id
, 1);
2403 r
= dm_pool_delete_thin_device(pool
->pmd
, dev_id
);
2405 DMWARN("Deletion of thin device %s failed.", argv
[1]);
2410 static int process_set_transaction_id_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2412 dm_thin_id old_id
, new_id
;
2415 r
= check_arg_count(argc
, 3);
2419 if (kstrtoull(argv
[1], 10, (unsigned long long *)&old_id
)) {
2420 DMWARN("set_transaction_id message: Unrecognised id %s.", argv
[1]);
2424 if (kstrtoull(argv
[2], 10, (unsigned long long *)&new_id
)) {
2425 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv
[2]);
2429 r
= dm_pool_set_metadata_transaction_id(pool
->pmd
, old_id
, new_id
);
2431 DMWARN("Failed to change transaction id from %s to %s.",
2439 static int process_reserve_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2443 r
= check_arg_count(argc
, 1);
2447 (void) commit(pool
);
2449 r
= dm_pool_reserve_metadata_snap(pool
->pmd
);
2451 DMWARN("reserve_metadata_snap message failed.");
2456 static int process_release_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2460 r
= check_arg_count(argc
, 1);
2464 r
= dm_pool_release_metadata_snap(pool
->pmd
);
2466 DMWARN("release_metadata_snap message failed.");
2472 * Messages supported:
2473 * create_thin <dev_id>
2474 * create_snap <dev_id> <origin_id>
2476 * trim <dev_id> <new_size_in_sectors>
2477 * set_transaction_id <current_trans_id> <new_trans_id>
2478 * reserve_metadata_snap
2479 * release_metadata_snap
2481 static int pool_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
2484 struct pool_c
*pt
= ti
->private;
2485 struct pool
*pool
= pt
->pool
;
2487 if (!strcasecmp(argv
[0], "create_thin"))
2488 r
= process_create_thin_mesg(argc
, argv
, pool
);
2490 else if (!strcasecmp(argv
[0], "create_snap"))
2491 r
= process_create_snap_mesg(argc
, argv
, pool
);
2493 else if (!strcasecmp(argv
[0], "delete"))
2494 r
= process_delete_mesg(argc
, argv
, pool
);
2496 else if (!strcasecmp(argv
[0], "set_transaction_id"))
2497 r
= process_set_transaction_id_mesg(argc
, argv
, pool
);
2499 else if (!strcasecmp(argv
[0], "reserve_metadata_snap"))
2500 r
= process_reserve_metadata_snap_mesg(argc
, argv
, pool
);
2502 else if (!strcasecmp(argv
[0], "release_metadata_snap"))
2503 r
= process_release_metadata_snap_mesg(argc
, argv
, pool
);
2506 DMWARN("Unrecognised thin pool target message received: %s", argv
[0]);
2509 (void) commit(pool
);
2514 static void emit_flags(struct pool_features
*pf
, char *result
,
2515 unsigned sz
, unsigned maxlen
)
2517 unsigned count
= !pf
->zero_new_blocks
+ !pf
->discard_enabled
+
2518 !pf
->discard_passdown
+ (pf
->mode
== PM_READ_ONLY
);
2519 DMEMIT("%u ", count
);
2521 if (!pf
->zero_new_blocks
)
2522 DMEMIT("skip_block_zeroing ");
2524 if (!pf
->discard_enabled
)
2525 DMEMIT("ignore_discard ");
2527 if (!pf
->discard_passdown
)
2528 DMEMIT("no_discard_passdown ");
2530 if (pf
->mode
== PM_READ_ONLY
)
2531 DMEMIT("read_only ");
2536 * <transaction id> <used metadata sectors>/<total metadata sectors>
2537 * <used data sectors>/<total data sectors> <held metadata root>
2539 static void pool_status(struct dm_target
*ti
, status_type_t type
,
2540 unsigned status_flags
, char *result
, unsigned maxlen
)
2544 uint64_t transaction_id
;
2545 dm_block_t nr_free_blocks_data
;
2546 dm_block_t nr_free_blocks_metadata
;
2547 dm_block_t nr_blocks_data
;
2548 dm_block_t nr_blocks_metadata
;
2549 dm_block_t held_root
;
2550 char buf
[BDEVNAME_SIZE
];
2551 char buf2
[BDEVNAME_SIZE
];
2552 struct pool_c
*pt
= ti
->private;
2553 struct pool
*pool
= pt
->pool
;
2556 case STATUSTYPE_INFO
:
2557 if (get_pool_mode(pool
) == PM_FAIL
) {
2562 /* Commit to ensure statistics aren't out-of-date */
2563 if (!(status_flags
& DM_STATUS_NOFLUSH_FLAG
) && !dm_suspended(ti
))
2564 (void) commit(pool
);
2566 r
= dm_pool_get_metadata_transaction_id(pool
->pmd
, &transaction_id
);
2568 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
2569 dm_device_name(pool
->pool_md
), r
);
2573 r
= dm_pool_get_free_metadata_block_count(pool
->pmd
, &nr_free_blocks_metadata
);
2575 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
2576 dm_device_name(pool
->pool_md
), r
);
2580 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &nr_blocks_metadata
);
2582 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
2583 dm_device_name(pool
->pool_md
), r
);
2587 r
= dm_pool_get_free_block_count(pool
->pmd
, &nr_free_blocks_data
);
2589 DMERR("%s: dm_pool_get_free_block_count returned %d",
2590 dm_device_name(pool
->pool_md
), r
);
2594 r
= dm_pool_get_data_dev_size(pool
->pmd
, &nr_blocks_data
);
2596 DMERR("%s: dm_pool_get_data_dev_size returned %d",
2597 dm_device_name(pool
->pool_md
), r
);
2601 r
= dm_pool_get_metadata_snap(pool
->pmd
, &held_root
);
2603 DMERR("%s: dm_pool_get_metadata_snap returned %d",
2604 dm_device_name(pool
->pool_md
), r
);
2608 DMEMIT("%llu %llu/%llu %llu/%llu ",
2609 (unsigned long long)transaction_id
,
2610 (unsigned long long)(nr_blocks_metadata
- nr_free_blocks_metadata
),
2611 (unsigned long long)nr_blocks_metadata
,
2612 (unsigned long long)(nr_blocks_data
- nr_free_blocks_data
),
2613 (unsigned long long)nr_blocks_data
);
2616 DMEMIT("%llu ", held_root
);
2620 if (pool
->pf
.mode
== PM_READ_ONLY
)
2625 if (!pool
->pf
.discard_enabled
)
2626 DMEMIT("ignore_discard");
2627 else if (pool
->pf
.discard_passdown
)
2628 DMEMIT("discard_passdown");
2630 DMEMIT("no_discard_passdown");
2634 case STATUSTYPE_TABLE
:
2635 DMEMIT("%s %s %lu %llu ",
2636 format_dev_t(buf
, pt
->metadata_dev
->bdev
->bd_dev
),
2637 format_dev_t(buf2
, pt
->data_dev
->bdev
->bd_dev
),
2638 (unsigned long)pool
->sectors_per_block
,
2639 (unsigned long long)pt
->low_water_blocks
);
2640 emit_flags(&pt
->requested_pf
, result
, sz
, maxlen
);
2649 static int pool_iterate_devices(struct dm_target
*ti
,
2650 iterate_devices_callout_fn fn
, void *data
)
2652 struct pool_c
*pt
= ti
->private;
2654 return fn(ti
, pt
->data_dev
, 0, ti
->len
, data
);
2657 static int pool_merge(struct dm_target
*ti
, struct bvec_merge_data
*bvm
,
2658 struct bio_vec
*biovec
, int max_size
)
2660 struct pool_c
*pt
= ti
->private;
2661 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
2663 if (!q
->merge_bvec_fn
)
2666 bvm
->bi_bdev
= pt
->data_dev
->bdev
;
2668 return min(max_size
, q
->merge_bvec_fn(q
, bvm
, biovec
));
2671 static void set_discard_limits(struct pool_c
*pt
, struct queue_limits
*limits
)
2673 struct pool
*pool
= pt
->pool
;
2674 struct queue_limits
*data_limits
;
2676 limits
->max_discard_sectors
= pool
->sectors_per_block
;
2679 * discard_granularity is just a hint, and not enforced.
2681 if (pt
->adjusted_pf
.discard_passdown
) {
2682 data_limits
= &bdev_get_queue(pt
->data_dev
->bdev
)->limits
;
2683 limits
->discard_granularity
= data_limits
->discard_granularity
;
2685 limits
->discard_granularity
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
2688 static void pool_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
2690 struct pool_c
*pt
= ti
->private;
2691 struct pool
*pool
= pt
->pool
;
2692 uint64_t io_opt_sectors
= limits
->io_opt
>> SECTOR_SHIFT
;
2695 * If the system-determined stacked limits are compatible with the
2696 * pool's blocksize (io_opt is a factor) do not override them.
2698 if (io_opt_sectors
< pool
->sectors_per_block
||
2699 do_div(io_opt_sectors
, pool
->sectors_per_block
)) {
2700 blk_limits_io_min(limits
, 0);
2701 blk_limits_io_opt(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
2705 * pt->adjusted_pf is a staging area for the actual features to use.
2706 * They get transferred to the live pool in bind_control_target()
2707 * called from pool_preresume().
2709 if (!pt
->adjusted_pf
.discard_enabled
) {
2711 * Must explicitly disallow stacking discard limits otherwise the
2712 * block layer will stack them if pool's data device has support.
2713 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
2714 * user to see that, so make sure to set all discard limits to 0.
2716 limits
->discard_granularity
= 0;
2720 disable_passdown_if_not_supported(pt
);
2722 set_discard_limits(pt
, limits
);
2725 static struct target_type pool_target
= {
2726 .name
= "thin-pool",
2727 .features
= DM_TARGET_SINGLETON
| DM_TARGET_ALWAYS_WRITEABLE
|
2728 DM_TARGET_IMMUTABLE
,
2729 .version
= {1, 9, 0},
2730 .module
= THIS_MODULE
,
2734 .postsuspend
= pool_postsuspend
,
2735 .preresume
= pool_preresume
,
2736 .resume
= pool_resume
,
2737 .message
= pool_message
,
2738 .status
= pool_status
,
2739 .merge
= pool_merge
,
2740 .iterate_devices
= pool_iterate_devices
,
2741 .io_hints
= pool_io_hints
,
2744 /*----------------------------------------------------------------
2745 * Thin target methods
2746 *--------------------------------------------------------------*/
2747 static void thin_dtr(struct dm_target
*ti
)
2749 struct thin_c
*tc
= ti
->private;
2751 mutex_lock(&dm_thin_pool_table
.mutex
);
2753 __pool_dec(tc
->pool
);
2754 dm_pool_close_thin_device(tc
->td
);
2755 dm_put_device(ti
, tc
->pool_dev
);
2757 dm_put_device(ti
, tc
->origin_dev
);
2760 mutex_unlock(&dm_thin_pool_table
.mutex
);
2764 * Thin target parameters:
2766 * <pool_dev> <dev_id> [origin_dev]
2768 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2769 * dev_id: the internal device identifier
2770 * origin_dev: a device external to the pool that should act as the origin
2772 * If the pool device has discards disabled, they get disabled for the thin
2775 static int thin_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2779 struct dm_dev
*pool_dev
, *origin_dev
;
2780 struct mapped_device
*pool_md
;
2782 mutex_lock(&dm_thin_pool_table
.mutex
);
2784 if (argc
!= 2 && argc
!= 3) {
2785 ti
->error
= "Invalid argument count";
2790 tc
= ti
->private = kzalloc(sizeof(*tc
), GFP_KERNEL
);
2792 ti
->error
= "Out of memory";
2798 r
= dm_get_device(ti
, argv
[2], FMODE_READ
, &origin_dev
);
2800 ti
->error
= "Error opening origin device";
2801 goto bad_origin_dev
;
2803 tc
->origin_dev
= origin_dev
;
2806 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &pool_dev
);
2808 ti
->error
= "Error opening pool device";
2811 tc
->pool_dev
= pool_dev
;
2813 if (read_dev_id(argv
[1], (unsigned long long *)&tc
->dev_id
, 0)) {
2814 ti
->error
= "Invalid device id";
2819 pool_md
= dm_get_md(tc
->pool_dev
->bdev
->bd_dev
);
2821 ti
->error
= "Couldn't get pool mapped device";
2826 tc
->pool
= __pool_table_lookup(pool_md
);
2828 ti
->error
= "Couldn't find pool object";
2830 goto bad_pool_lookup
;
2832 __pool_inc(tc
->pool
);
2834 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
2835 ti
->error
= "Couldn't open thin device, Pool is in fail mode";
2839 r
= dm_pool_open_thin_device(tc
->pool
->pmd
, tc
->dev_id
, &tc
->td
);
2841 ti
->error
= "Couldn't open thin internal device";
2845 r
= dm_set_target_max_io_len(ti
, tc
->pool
->sectors_per_block
);
2849 ti
->num_flush_bios
= 1;
2850 ti
->flush_supported
= true;
2851 ti
->per_bio_data_size
= sizeof(struct dm_thin_endio_hook
);
2853 /* In case the pool supports discards, pass them on. */
2854 ti
->discard_zeroes_data_unsupported
= true;
2855 if (tc
->pool
->pf
.discard_enabled
) {
2856 ti
->discards_supported
= true;
2857 ti
->num_discard_bios
= 1;
2858 /* Discard bios must be split on a block boundary */
2859 ti
->split_discard_bios
= true;
2864 mutex_unlock(&dm_thin_pool_table
.mutex
);
2869 __pool_dec(tc
->pool
);
2873 dm_put_device(ti
, tc
->pool_dev
);
2876 dm_put_device(ti
, tc
->origin_dev
);
2880 mutex_unlock(&dm_thin_pool_table
.mutex
);
2885 static int thin_map(struct dm_target
*ti
, struct bio
*bio
)
2887 bio
->bi_sector
= dm_target_offset(ti
, bio
->bi_sector
);
2889 return thin_bio_map(ti
, bio
);
2892 static int thin_endio(struct dm_target
*ti
, struct bio
*bio
, int err
)
2894 unsigned long flags
;
2895 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
2896 struct list_head work
;
2897 struct dm_thin_new_mapping
*m
, *tmp
;
2898 struct pool
*pool
= h
->tc
->pool
;
2900 if (h
->shared_read_entry
) {
2901 INIT_LIST_HEAD(&work
);
2902 dm_deferred_entry_dec(h
->shared_read_entry
, &work
);
2904 spin_lock_irqsave(&pool
->lock
, flags
);
2905 list_for_each_entry_safe(m
, tmp
, &work
, list
) {
2908 __maybe_add_mapping(m
);
2910 spin_unlock_irqrestore(&pool
->lock
, flags
);
2913 if (h
->all_io_entry
) {
2914 INIT_LIST_HEAD(&work
);
2915 dm_deferred_entry_dec(h
->all_io_entry
, &work
);
2916 if (!list_empty(&work
)) {
2917 spin_lock_irqsave(&pool
->lock
, flags
);
2918 list_for_each_entry_safe(m
, tmp
, &work
, list
)
2919 list_add(&m
->list
, &pool
->prepared_discards
);
2920 spin_unlock_irqrestore(&pool
->lock
, flags
);
2928 static void thin_postsuspend(struct dm_target
*ti
)
2930 if (dm_noflush_suspending(ti
))
2931 requeue_io((struct thin_c
*)ti
->private);
2935 * <nr mapped sectors> <highest mapped sector>
2937 static void thin_status(struct dm_target
*ti
, status_type_t type
,
2938 unsigned status_flags
, char *result
, unsigned maxlen
)
2942 dm_block_t mapped
, highest
;
2943 char buf
[BDEVNAME_SIZE
];
2944 struct thin_c
*tc
= ti
->private;
2946 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
2955 case STATUSTYPE_INFO
:
2956 r
= dm_thin_get_mapped_count(tc
->td
, &mapped
);
2958 DMERR("dm_thin_get_mapped_count returned %d", r
);
2962 r
= dm_thin_get_highest_mapped_block(tc
->td
, &highest
);
2964 DMERR("dm_thin_get_highest_mapped_block returned %d", r
);
2968 DMEMIT("%llu ", mapped
* tc
->pool
->sectors_per_block
);
2970 DMEMIT("%llu", ((highest
+ 1) *
2971 tc
->pool
->sectors_per_block
) - 1);
2976 case STATUSTYPE_TABLE
:
2978 format_dev_t(buf
, tc
->pool_dev
->bdev
->bd_dev
),
2979 (unsigned long) tc
->dev_id
);
2981 DMEMIT(" %s", format_dev_t(buf
, tc
->origin_dev
->bdev
->bd_dev
));
2992 static int thin_iterate_devices(struct dm_target
*ti
,
2993 iterate_devices_callout_fn fn
, void *data
)
2996 struct thin_c
*tc
= ti
->private;
2997 struct pool
*pool
= tc
->pool
;
3000 * We can't call dm_pool_get_data_dev_size() since that blocks. So
3001 * we follow a more convoluted path through to the pool's target.
3004 return 0; /* nothing is bound */
3006 blocks
= pool
->ti
->len
;
3007 (void) sector_div(blocks
, pool
->sectors_per_block
);
3009 return fn(ti
, tc
->pool_dev
, 0, pool
->sectors_per_block
* blocks
, data
);
3014 static struct target_type thin_target
= {
3016 .version
= {1, 9, 0},
3017 .module
= THIS_MODULE
,
3021 .end_io
= thin_endio
,
3022 .postsuspend
= thin_postsuspend
,
3023 .status
= thin_status
,
3024 .iterate_devices
= thin_iterate_devices
,
3027 /*----------------------------------------------------------------*/
3029 static int __init
dm_thin_init(void)
3035 r
= dm_register_target(&thin_target
);
3039 r
= dm_register_target(&pool_target
);
3041 goto bad_pool_target
;
3045 _new_mapping_cache
= KMEM_CACHE(dm_thin_new_mapping
, 0);
3046 if (!_new_mapping_cache
)
3047 goto bad_new_mapping_cache
;
3051 bad_new_mapping_cache
:
3052 dm_unregister_target(&pool_target
);
3054 dm_unregister_target(&thin_target
);
3059 static void dm_thin_exit(void)
3061 dm_unregister_target(&thin_target
);
3062 dm_unregister_target(&pool_target
);
3064 kmem_cache_destroy(_new_mapping_cache
);
3067 module_init(dm_thin_init
);
3068 module_exit(dm_thin_exit
);
3070 MODULE_DESCRIPTION(DM_NAME
" thin provisioning target");
3071 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3072 MODULE_LICENSE("GPL");