2 * Copyright (C) 2011-2012 Red Hat UK.
4 * This file is released under the GPL.
7 #include "dm-thin-metadata.h"
8 #include "dm-bio-prison.h"
11 #include <linux/device-mapper.h>
12 #include <linux/dm-io.h>
13 #include <linux/dm-kcopyd.h>
14 #include <linux/list.h>
15 #include <linux/init.h>
16 #include <linux/module.h>
17 #include <linux/slab.h>
19 #define DM_MSG_PREFIX "thin"
24 #define ENDIO_HOOK_POOL_SIZE 1024
25 #define MAPPING_POOL_SIZE 1024
26 #define PRISON_CELLS 1024
27 #define COMMIT_PERIOD HZ
29 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle
,
30 "A percentage of time allocated for copy on write");
33 * The block size of the device holding pool data must be
34 * between 64KB and 1GB.
36 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
37 #define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
40 * Device id is restricted to 24 bits.
42 #define MAX_DEV_ID ((1 << 24) - 1)
45 * How do we handle breaking sharing of data blocks?
46 * =================================================
48 * We use a standard copy-on-write btree to store the mappings for the
49 * devices (note I'm talking about copy-on-write of the metadata here, not
50 * the data). When you take an internal snapshot you clone the root node
51 * of the origin btree. After this there is no concept of an origin or a
52 * snapshot. They are just two device trees that happen to point to the
55 * When we get a write in we decide if it's to a shared data block using
56 * some timestamp magic. If it is, we have to break sharing.
58 * Let's say we write to a shared block in what was the origin. The
61 * i) plug io further to this physical block. (see bio_prison code).
63 * ii) quiesce any read io to that shared data block. Obviously
64 * including all devices that share this block. (see dm_deferred_set code)
66 * iii) copy the data block to a newly allocate block. This step can be
67 * missed out if the io covers the block. (schedule_copy).
69 * iv) insert the new mapping into the origin's btree
70 * (process_prepared_mapping). This act of inserting breaks some
71 * sharing of btree nodes between the two devices. Breaking sharing only
72 * effects the btree of that specific device. Btrees for the other
73 * devices that share the block never change. The btree for the origin
74 * device as it was after the last commit is untouched, ie. we're using
75 * persistent data structures in the functional programming sense.
77 * v) unplug io to this physical block, including the io that triggered
78 * the breaking of sharing.
80 * Steps (ii) and (iii) occur in parallel.
82 * The metadata _doesn't_ need to be committed before the io continues. We
83 * get away with this because the io is always written to a _new_ block.
84 * If there's a crash, then:
86 * - The origin mapping will point to the old origin block (the shared
87 * one). This will contain the data as it was before the io that triggered
88 * the breaking of sharing came in.
90 * - The snap mapping still points to the old block. As it would after
93 * The downside of this scheme is the timestamp magic isn't perfect, and
94 * will continue to think that data block in the snapshot device is shared
95 * even after the write to the origin has broken sharing. I suspect data
96 * blocks will typically be shared by many different devices, so we're
97 * breaking sharing n + 1 times, rather than n, where n is the number of
98 * devices that reference this data block. At the moment I think the
99 * benefits far, far outweigh the disadvantages.
102 /*----------------------------------------------------------------*/
107 static void build_data_key(struct dm_thin_device
*td
,
108 dm_block_t b
, struct dm_cell_key
*key
)
111 key
->dev
= dm_thin_dev_id(td
);
115 static void build_virtual_key(struct dm_thin_device
*td
, dm_block_t b
,
116 struct dm_cell_key
*key
)
119 key
->dev
= dm_thin_dev_id(td
);
123 /*----------------------------------------------------------------*/
126 * A pool device ties together a metadata device and a data device. It
127 * also provides the interface for creating and destroying internal
130 struct dm_thin_new_mapping
;
133 * The pool runs in 3 modes. Ordered in degraded order for comparisons.
136 PM_WRITE
, /* metadata may be changed */
137 PM_READ_ONLY
, /* metadata may not be changed */
138 PM_FAIL
, /* all I/O fails */
141 struct pool_features
{
144 bool zero_new_blocks
:1;
145 bool discard_enabled
:1;
146 bool discard_passdown
:1;
147 bool error_if_no_space
:1;
151 typedef void (*process_bio_fn
)(struct thin_c
*tc
, struct bio
*bio
);
152 typedef void (*process_mapping_fn
)(struct dm_thin_new_mapping
*m
);
155 struct list_head list
;
156 struct dm_target
*ti
; /* Only set if a pool target is bound */
158 struct mapped_device
*pool_md
;
159 struct block_device
*md_dev
;
160 struct dm_pool_metadata
*pmd
;
162 dm_block_t low_water_blocks
;
163 uint32_t sectors_per_block
;
164 int sectors_per_block_shift
;
166 struct pool_features pf
;
167 bool low_water_triggered
:1; /* A dm event has been sent */
168 bool no_free_space
:1; /* bios will be requeued if set */
170 struct dm_bio_prison
*prison
;
171 struct dm_kcopyd_client
*copier
;
173 struct workqueue_struct
*wq
;
174 struct work_struct worker
;
175 struct delayed_work waker
;
177 unsigned long last_commit_jiffies
;
181 struct bio_list deferred_bios
;
182 struct bio_list deferred_flush_bios
;
183 struct list_head prepared_mappings
;
184 struct list_head prepared_discards
;
186 struct bio_list retry_on_resume_list
;
188 struct dm_deferred_set
*shared_read_ds
;
189 struct dm_deferred_set
*all_io_ds
;
191 struct dm_thin_new_mapping
*next_mapping
;
192 mempool_t
*mapping_pool
;
194 process_bio_fn process_bio
;
195 process_bio_fn process_discard
;
197 process_mapping_fn process_prepared_mapping
;
198 process_mapping_fn process_prepared_discard
;
201 static enum pool_mode
get_pool_mode(struct pool
*pool
);
202 static void out_of_data_space(struct pool
*pool
);
203 static void metadata_operation_failed(struct pool
*pool
, const char *op
, int r
);
206 * Target context for a pool.
209 struct dm_target
*ti
;
211 struct dm_dev
*data_dev
;
212 struct dm_dev
*metadata_dev
;
213 struct dm_target_callbacks callbacks
;
215 dm_block_t low_water_blocks
;
216 struct pool_features requested_pf
; /* Features requested during table load */
217 struct pool_features adjusted_pf
; /* Features used after adjusting for constituent devices */
221 * Target context for a thin.
224 struct dm_dev
*pool_dev
;
225 struct dm_dev
*origin_dev
;
229 struct dm_thin_device
*td
;
232 /*----------------------------------------------------------------*/
235 * wake_worker() is used when new work is queued and when pool_resume is
236 * ready to continue deferred IO processing.
238 static void wake_worker(struct pool
*pool
)
240 queue_work(pool
->wq
, &pool
->worker
);
243 /*----------------------------------------------------------------*/
245 static int bio_detain(struct pool
*pool
, struct dm_cell_key
*key
, struct bio
*bio
,
246 struct dm_bio_prison_cell
**cell_result
)
249 struct dm_bio_prison_cell
*cell_prealloc
;
252 * Allocate a cell from the prison's mempool.
253 * This might block but it can't fail.
255 cell_prealloc
= dm_bio_prison_alloc_cell(pool
->prison
, GFP_NOIO
);
257 r
= dm_bio_detain(pool
->prison
, key
, bio
, cell_prealloc
, cell_result
);
260 * We reused an old cell; we can get rid of
263 dm_bio_prison_free_cell(pool
->prison
, cell_prealloc
);
268 static void cell_release(struct pool
*pool
,
269 struct dm_bio_prison_cell
*cell
,
270 struct bio_list
*bios
)
272 dm_cell_release(pool
->prison
, cell
, bios
);
273 dm_bio_prison_free_cell(pool
->prison
, cell
);
276 static void cell_release_no_holder(struct pool
*pool
,
277 struct dm_bio_prison_cell
*cell
,
278 struct bio_list
*bios
)
280 dm_cell_release_no_holder(pool
->prison
, cell
, bios
);
281 dm_bio_prison_free_cell(pool
->prison
, cell
);
284 static void cell_defer_no_holder_no_free(struct thin_c
*tc
,
285 struct dm_bio_prison_cell
*cell
)
287 struct pool
*pool
= tc
->pool
;
290 spin_lock_irqsave(&pool
->lock
, flags
);
291 dm_cell_release_no_holder(pool
->prison
, cell
, &pool
->deferred_bios
);
292 spin_unlock_irqrestore(&pool
->lock
, flags
);
297 static void cell_error(struct pool
*pool
,
298 struct dm_bio_prison_cell
*cell
)
300 dm_cell_error(pool
->prison
, cell
);
301 dm_bio_prison_free_cell(pool
->prison
, cell
);
304 /*----------------------------------------------------------------*/
307 * A global list of pools that uses a struct mapped_device as a key.
309 static struct dm_thin_pool_table
{
311 struct list_head pools
;
312 } dm_thin_pool_table
;
314 static void pool_table_init(void)
316 mutex_init(&dm_thin_pool_table
.mutex
);
317 INIT_LIST_HEAD(&dm_thin_pool_table
.pools
);
320 static void __pool_table_insert(struct pool
*pool
)
322 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
323 list_add(&pool
->list
, &dm_thin_pool_table
.pools
);
326 static void __pool_table_remove(struct pool
*pool
)
328 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
329 list_del(&pool
->list
);
332 static struct pool
*__pool_table_lookup(struct mapped_device
*md
)
334 struct pool
*pool
= NULL
, *tmp
;
336 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
338 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
339 if (tmp
->pool_md
== md
) {
348 static struct pool
*__pool_table_lookup_metadata_dev(struct block_device
*md_dev
)
350 struct pool
*pool
= NULL
, *tmp
;
352 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
354 list_for_each_entry(tmp
, &dm_thin_pool_table
.pools
, list
) {
355 if (tmp
->md_dev
== md_dev
) {
364 /*----------------------------------------------------------------*/
366 struct dm_thin_endio_hook
{
368 struct dm_deferred_entry
*shared_read_entry
;
369 struct dm_deferred_entry
*all_io_entry
;
370 struct dm_thin_new_mapping
*overwrite_mapping
;
373 static void __requeue_bio_list(struct thin_c
*tc
, struct bio_list
*master
)
376 struct bio_list bios
;
378 bio_list_init(&bios
);
379 bio_list_merge(&bios
, master
);
380 bio_list_init(master
);
382 while ((bio
= bio_list_pop(&bios
))) {
383 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
386 bio_endio(bio
, DM_ENDIO_REQUEUE
);
388 bio_list_add(master
, bio
);
392 static void requeue_io(struct thin_c
*tc
)
394 struct pool
*pool
= tc
->pool
;
397 spin_lock_irqsave(&pool
->lock
, flags
);
398 __requeue_bio_list(tc
, &pool
->deferred_bios
);
399 __requeue_bio_list(tc
, &pool
->retry_on_resume_list
);
400 spin_unlock_irqrestore(&pool
->lock
, flags
);
404 * This section of code contains the logic for processing a thin device's IO.
405 * Much of the code depends on pool object resources (lists, workqueues, etc)
406 * but most is exclusively called from the thin target rather than the thin-pool
410 static bool block_size_is_power_of_two(struct pool
*pool
)
412 return pool
->sectors_per_block_shift
>= 0;
415 static dm_block_t
get_bio_block(struct thin_c
*tc
, struct bio
*bio
)
417 struct pool
*pool
= tc
->pool
;
418 sector_t block_nr
= bio
->bi_sector
;
420 if (block_size_is_power_of_two(pool
))
421 block_nr
>>= pool
->sectors_per_block_shift
;
423 (void) sector_div(block_nr
, pool
->sectors_per_block
);
428 static void remap(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
)
430 struct pool
*pool
= tc
->pool
;
431 sector_t bi_sector
= bio
->bi_sector
;
433 bio
->bi_bdev
= tc
->pool_dev
->bdev
;
434 if (block_size_is_power_of_two(pool
))
435 bio
->bi_sector
= (block
<< pool
->sectors_per_block_shift
) |
436 (bi_sector
& (pool
->sectors_per_block
- 1));
438 bio
->bi_sector
= (block
* pool
->sectors_per_block
) +
439 sector_div(bi_sector
, pool
->sectors_per_block
);
442 static void remap_to_origin(struct thin_c
*tc
, struct bio
*bio
)
444 bio
->bi_bdev
= tc
->origin_dev
->bdev
;
447 static int bio_triggers_commit(struct thin_c
*tc
, struct bio
*bio
)
449 return (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) &&
450 dm_thin_changed_this_transaction(tc
->td
);
453 static void inc_all_io_entry(struct pool
*pool
, struct bio
*bio
)
455 struct dm_thin_endio_hook
*h
;
457 if (bio
->bi_rw
& REQ_DISCARD
)
460 h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
461 h
->all_io_entry
= dm_deferred_entry_inc(pool
->all_io_ds
);
464 static void issue(struct thin_c
*tc
, struct bio
*bio
)
466 struct pool
*pool
= tc
->pool
;
469 if (!bio_triggers_commit(tc
, bio
)) {
470 generic_make_request(bio
);
475 * Complete bio with an error if earlier I/O caused changes to
476 * the metadata that can't be committed e.g, due to I/O errors
477 * on the metadata device.
479 if (dm_thin_aborted_changes(tc
->td
)) {
485 * Batch together any bios that trigger commits and then issue a
486 * single commit for them in process_deferred_bios().
488 spin_lock_irqsave(&pool
->lock
, flags
);
489 bio_list_add(&pool
->deferred_flush_bios
, bio
);
490 spin_unlock_irqrestore(&pool
->lock
, flags
);
493 static void remap_to_origin_and_issue(struct thin_c
*tc
, struct bio
*bio
)
495 remap_to_origin(tc
, bio
);
499 static void remap_and_issue(struct thin_c
*tc
, struct bio
*bio
,
502 remap(tc
, bio
, block
);
506 /*----------------------------------------------------------------*/
509 * Bio endio functions.
511 struct dm_thin_new_mapping
{
512 struct list_head list
;
517 bool definitely_not_shared
:1;
521 dm_block_t virt_block
;
522 dm_block_t data_block
;
523 struct dm_bio_prison_cell
*cell
, *cell2
;
526 * If the bio covers the whole area of a block then we can avoid
527 * zeroing or copying. Instead this bio is hooked. The bio will
528 * still be in the cell, so care has to be taken to avoid issuing
532 bio_end_io_t
*saved_bi_end_io
;
535 static void __maybe_add_mapping(struct dm_thin_new_mapping
*m
)
537 struct pool
*pool
= m
->tc
->pool
;
539 if (m
->quiesced
&& m
->prepared
) {
540 list_add_tail(&m
->list
, &pool
->prepared_mappings
);
545 static void copy_complete(int read_err
, unsigned long write_err
, void *context
)
548 struct dm_thin_new_mapping
*m
= context
;
549 struct pool
*pool
= m
->tc
->pool
;
551 m
->err
= read_err
|| write_err
? -EIO
: 0;
553 spin_lock_irqsave(&pool
->lock
, flags
);
555 __maybe_add_mapping(m
);
556 spin_unlock_irqrestore(&pool
->lock
, flags
);
559 static void overwrite_endio(struct bio
*bio
, int err
)
562 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
563 struct dm_thin_new_mapping
*m
= h
->overwrite_mapping
;
564 struct pool
*pool
= m
->tc
->pool
;
568 spin_lock_irqsave(&pool
->lock
, flags
);
570 __maybe_add_mapping(m
);
571 spin_unlock_irqrestore(&pool
->lock
, flags
);
574 /*----------------------------------------------------------------*/
581 * Prepared mapping jobs.
585 * This sends the bios in the cell back to the deferred_bios list.
587 static void cell_defer(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
589 struct pool
*pool
= tc
->pool
;
592 spin_lock_irqsave(&pool
->lock
, flags
);
593 cell_release(pool
, cell
, &pool
->deferred_bios
);
594 spin_unlock_irqrestore(&tc
->pool
->lock
, flags
);
600 * Same as cell_defer above, except it omits the original holder of the cell.
602 static void cell_defer_no_holder(struct thin_c
*tc
, struct dm_bio_prison_cell
*cell
)
604 struct pool
*pool
= tc
->pool
;
607 spin_lock_irqsave(&pool
->lock
, flags
);
608 cell_release_no_holder(pool
, cell
, &pool
->deferred_bios
);
609 spin_unlock_irqrestore(&pool
->lock
, flags
);
614 static void process_prepared_mapping_fail(struct dm_thin_new_mapping
*m
)
617 m
->bio
->bi_end_io
= m
->saved_bi_end_io
;
618 cell_error(m
->tc
->pool
, m
->cell
);
620 mempool_free(m
, m
->tc
->pool
->mapping_pool
);
623 static void process_prepared_mapping(struct dm_thin_new_mapping
*m
)
625 struct thin_c
*tc
= m
->tc
;
626 struct pool
*pool
= tc
->pool
;
632 bio
->bi_end_io
= m
->saved_bi_end_io
;
635 cell_error(pool
, m
->cell
);
640 * Commit the prepared block into the mapping btree.
641 * Any I/O for this block arriving after this point will get
642 * remapped to it directly.
644 r
= dm_thin_insert_block(tc
->td
, m
->virt_block
, m
->data_block
);
646 metadata_operation_failed(pool
, "dm_thin_insert_block", r
);
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 metadata_operation_failed(pool
, "dm_pool_commit_metadata", r
);
909 static void check_low_water_mark(struct pool
*pool
, dm_block_t free_blocks
)
913 if (free_blocks
<= pool
->low_water_blocks
&& !pool
->low_water_triggered
) {
914 DMWARN("%s: reached low water mark for data device: sending event.",
915 dm_device_name(pool
->pool_md
));
916 spin_lock_irqsave(&pool
->lock
, flags
);
917 pool
->low_water_triggered
= true;
918 spin_unlock_irqrestore(&pool
->lock
, flags
);
919 dm_table_event(pool
->ti
->table
);
923 static int alloc_data_block(struct thin_c
*tc
, dm_block_t
*result
)
926 dm_block_t free_blocks
;
927 struct pool
*pool
= tc
->pool
;
929 if (get_pool_mode(pool
) != PM_WRITE
)
932 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
934 metadata_operation_failed(pool
, "dm_pool_get_free_block_count", r
);
938 check_low_water_mark(pool
, free_blocks
);
942 * Try to commit to see if that will free up some
949 r
= dm_pool_get_free_block_count(pool
->pmd
, &free_blocks
);
951 metadata_operation_failed(pool
, "dm_pool_get_free_block_count", r
);
956 out_of_data_space(pool
);
961 r
= dm_pool_alloc_data_block(pool
->pmd
, result
);
963 metadata_operation_failed(pool
, "dm_pool_alloc_data_block", r
);
971 * If we have run out of space, queue bios until the device is
972 * resumed, presumably after having been reloaded with more space.
974 static void retry_on_resume(struct bio
*bio
)
976 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
977 struct thin_c
*tc
= h
->tc
;
978 struct pool
*pool
= tc
->pool
;
981 spin_lock_irqsave(&pool
->lock
, flags
);
982 bio_list_add(&pool
->retry_on_resume_list
, bio
);
983 spin_unlock_irqrestore(&pool
->lock
, flags
);
986 static void handle_unserviceable_bio(struct pool
*pool
, struct bio
*bio
)
989 * When pool is read-only, no cell locking is needed because
990 * nothing is changing.
992 WARN_ON_ONCE(get_pool_mode(pool
) != PM_READ_ONLY
);
994 if (pool
->no_free_space
)
995 retry_on_resume(bio
);
1000 static void retry_bios_on_resume(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 handle_unserviceable_bio(pool
, 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_tail(&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 retry_bios_on_resume(pool
, cell
);
1117 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1119 cell_error(pool
, cell
);
1124 static void process_shared_bio(struct thin_c
*tc
, struct bio
*bio
,
1126 struct dm_thin_lookup_result
*lookup_result
)
1128 struct dm_bio_prison_cell
*cell
;
1129 struct pool
*pool
= tc
->pool
;
1130 struct dm_cell_key key
;
1133 * If cell is already occupied, then sharing is already in the process
1134 * of being broken so we have nothing further to do here.
1136 build_data_key(tc
->td
, lookup_result
->block
, &key
);
1137 if (bio_detain(pool
, &key
, bio
, &cell
))
1140 if (bio_data_dir(bio
) == WRITE
&& bio
->bi_size
)
1141 break_sharing(tc
, bio
, block
, &key
, lookup_result
, cell
);
1143 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1145 h
->shared_read_entry
= dm_deferred_entry_inc(pool
->shared_read_ds
);
1146 inc_all_io_entry(pool
, bio
);
1147 cell_defer_no_holder(tc
, cell
);
1149 remap_and_issue(tc
, bio
, lookup_result
->block
);
1153 static void provision_block(struct thin_c
*tc
, struct bio
*bio
, dm_block_t block
,
1154 struct dm_bio_prison_cell
*cell
)
1157 dm_block_t data_block
;
1158 struct pool
*pool
= tc
->pool
;
1161 * Remap empty bios (flushes) immediately, without provisioning.
1163 if (!bio
->bi_size
) {
1164 inc_all_io_entry(pool
, bio
);
1165 cell_defer_no_holder(tc
, cell
);
1167 remap_and_issue(tc
, bio
, 0);
1172 * Fill read bios with zeroes and complete them immediately.
1174 if (bio_data_dir(bio
) == READ
) {
1176 cell_defer_no_holder(tc
, cell
);
1181 r
= alloc_data_block(tc
, &data_block
);
1185 schedule_external_copy(tc
, block
, data_block
, cell
, bio
);
1187 schedule_zero(tc
, block
, data_block
, cell
, bio
);
1191 retry_bios_on_resume(pool
, cell
);
1195 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1197 cell_error(pool
, cell
);
1202 static void process_bio(struct thin_c
*tc
, struct bio
*bio
)
1205 struct pool
*pool
= tc
->pool
;
1206 dm_block_t block
= get_bio_block(tc
, bio
);
1207 struct dm_bio_prison_cell
*cell
;
1208 struct dm_cell_key key
;
1209 struct dm_thin_lookup_result lookup_result
;
1212 * If cell is already occupied, then the block is already
1213 * being provisioned so we have nothing further to do here.
1215 build_virtual_key(tc
->td
, block
, &key
);
1216 if (bio_detain(pool
, &key
, bio
, &cell
))
1219 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1222 if (lookup_result
.shared
) {
1223 process_shared_bio(tc
, bio
, block
, &lookup_result
);
1224 cell_defer_no_holder(tc
, cell
); /* FIXME: pass this cell into process_shared? */
1226 inc_all_io_entry(pool
, bio
);
1227 cell_defer_no_holder(tc
, cell
);
1229 remap_and_issue(tc
, bio
, lookup_result
.block
);
1234 if (bio_data_dir(bio
) == READ
&& tc
->origin_dev
) {
1235 inc_all_io_entry(pool
, bio
);
1236 cell_defer_no_holder(tc
, cell
);
1238 remap_to_origin_and_issue(tc
, bio
);
1240 provision_block(tc
, bio
, block
, cell
);
1244 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1246 cell_defer_no_holder(tc
, cell
);
1252 static void process_bio_read_only(struct thin_c
*tc
, struct bio
*bio
)
1255 int rw
= bio_data_dir(bio
);
1256 dm_block_t block
= get_bio_block(tc
, bio
);
1257 struct dm_thin_lookup_result lookup_result
;
1259 r
= dm_thin_find_block(tc
->td
, block
, 1, &lookup_result
);
1262 if (lookup_result
.shared
&& (rw
== WRITE
) && bio
->bi_size
)
1263 handle_unserviceable_bio(tc
->pool
, bio
);
1265 inc_all_io_entry(tc
->pool
, bio
);
1266 remap_and_issue(tc
, bio
, lookup_result
.block
);
1272 handle_unserviceable_bio(tc
->pool
, bio
);
1276 if (tc
->origin_dev
) {
1277 inc_all_io_entry(tc
->pool
, bio
);
1278 remap_to_origin_and_issue(tc
, bio
);
1287 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1294 static void process_bio_fail(struct thin_c
*tc
, struct bio
*bio
)
1300 * FIXME: should we also commit due to size of transaction, measured in
1303 static int need_commit_due_to_time(struct pool
*pool
)
1305 return jiffies
< pool
->last_commit_jiffies
||
1306 jiffies
> pool
->last_commit_jiffies
+ COMMIT_PERIOD
;
1309 static void process_deferred_bios(struct pool
*pool
)
1311 unsigned long flags
;
1313 struct bio_list bios
;
1315 bio_list_init(&bios
);
1317 spin_lock_irqsave(&pool
->lock
, flags
);
1318 bio_list_merge(&bios
, &pool
->deferred_bios
);
1319 bio_list_init(&pool
->deferred_bios
);
1320 spin_unlock_irqrestore(&pool
->lock
, flags
);
1322 while ((bio
= bio_list_pop(&bios
))) {
1323 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1324 struct thin_c
*tc
= h
->tc
;
1327 * If we've got no free new_mapping structs, and processing
1328 * this bio might require one, we pause until there are some
1329 * prepared mappings to process.
1331 if (ensure_next_mapping(pool
)) {
1332 spin_lock_irqsave(&pool
->lock
, flags
);
1333 bio_list_merge(&pool
->deferred_bios
, &bios
);
1334 spin_unlock_irqrestore(&pool
->lock
, flags
);
1339 if (bio
->bi_rw
& REQ_DISCARD
)
1340 pool
->process_discard(tc
, bio
);
1342 pool
->process_bio(tc
, bio
);
1346 * If there are any deferred flush bios, we must commit
1347 * the metadata before issuing them.
1349 bio_list_init(&bios
);
1350 spin_lock_irqsave(&pool
->lock
, flags
);
1351 bio_list_merge(&bios
, &pool
->deferred_flush_bios
);
1352 bio_list_init(&pool
->deferred_flush_bios
);
1353 spin_unlock_irqrestore(&pool
->lock
, flags
);
1355 if (bio_list_empty(&bios
) && !need_commit_due_to_time(pool
))
1359 while ((bio
= bio_list_pop(&bios
)))
1363 pool
->last_commit_jiffies
= jiffies
;
1365 while ((bio
= bio_list_pop(&bios
)))
1366 generic_make_request(bio
);
1369 static void do_worker(struct work_struct
*ws
)
1371 struct pool
*pool
= container_of(ws
, struct pool
, worker
);
1373 process_prepared(pool
, &pool
->prepared_mappings
, &pool
->process_prepared_mapping
);
1374 process_prepared(pool
, &pool
->prepared_discards
, &pool
->process_prepared_discard
);
1375 process_deferred_bios(pool
);
1379 * We want to commit periodically so that not too much
1380 * unwritten data builds up.
1382 static void do_waker(struct work_struct
*ws
)
1384 struct pool
*pool
= container_of(to_delayed_work(ws
), struct pool
, waker
);
1386 queue_delayed_work(pool
->wq
, &pool
->waker
, COMMIT_PERIOD
);
1389 /*----------------------------------------------------------------*/
1391 static enum pool_mode
get_pool_mode(struct pool
*pool
)
1393 return pool
->pf
.mode
;
1396 static void set_pool_mode(struct pool
*pool
, enum pool_mode mode
)
1400 pool
->pf
.mode
= mode
;
1404 DMERR("%s: switching pool to failure mode",
1405 dm_device_name(pool
->pool_md
));
1406 dm_pool_metadata_read_only(pool
->pmd
);
1407 pool
->process_bio
= process_bio_fail
;
1408 pool
->process_discard
= process_bio_fail
;
1409 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1410 pool
->process_prepared_discard
= process_prepared_discard_fail
;
1414 DMERR("%s: switching pool to read-only mode",
1415 dm_device_name(pool
->pool_md
));
1416 r
= dm_pool_abort_metadata(pool
->pmd
);
1418 DMERR("%s: aborting transaction failed",
1419 dm_device_name(pool
->pool_md
));
1420 set_pool_mode(pool
, PM_FAIL
);
1422 dm_pool_metadata_read_only(pool
->pmd
);
1423 pool
->process_bio
= process_bio_read_only
;
1424 pool
->process_discard
= process_discard
;
1425 pool
->process_prepared_mapping
= process_prepared_mapping_fail
;
1426 pool
->process_prepared_discard
= process_prepared_discard_passdown
;
1431 dm_pool_metadata_read_write(pool
->pmd
);
1432 pool
->process_bio
= process_bio
;
1433 pool
->process_discard
= process_discard
;
1434 pool
->process_prepared_mapping
= process_prepared_mapping
;
1435 pool
->process_prepared_discard
= process_prepared_discard
;
1440 static void set_no_free_space(struct pool
*pool
)
1442 unsigned long flags
;
1444 if (pool
->pf
.error_if_no_space
)
1447 spin_lock_irqsave(&pool
->lock
, flags
);
1448 pool
->no_free_space
= true;
1449 spin_unlock_irqrestore(&pool
->lock
, flags
);
1453 * Rather than calling set_pool_mode directly, use these which describe the
1454 * reason for mode degradation.
1456 static void out_of_data_space(struct pool
*pool
)
1458 DMERR_LIMIT("%s: no free data space available.",
1459 dm_device_name(pool
->pool_md
));
1460 set_no_free_space(pool
);
1461 set_pool_mode(pool
, PM_READ_ONLY
);
1464 static void metadata_operation_failed(struct pool
*pool
, const char *op
, int r
)
1466 dm_block_t free_blocks
;
1468 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
1469 dm_device_name(pool
->pool_md
), op
, r
);
1472 !dm_pool_get_free_metadata_block_count(pool
->pmd
, &free_blocks
) &&
1474 DMERR_LIMIT("%s: no free metadata space available.",
1475 dm_device_name(pool
->pool_md
));
1476 set_no_free_space(pool
);
1479 set_pool_mode(pool
, PM_READ_ONLY
);
1482 /*----------------------------------------------------------------*/
1485 * Mapping functions.
1489 * Called only while mapping a thin bio to hand it over to the workqueue.
1491 static void thin_defer_bio(struct thin_c
*tc
, struct bio
*bio
)
1493 unsigned long flags
;
1494 struct pool
*pool
= tc
->pool
;
1496 spin_lock_irqsave(&pool
->lock
, flags
);
1497 bio_list_add(&pool
->deferred_bios
, bio
);
1498 spin_unlock_irqrestore(&pool
->lock
, flags
);
1503 static void thin_hook_bio(struct thin_c
*tc
, struct bio
*bio
)
1505 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
1508 h
->shared_read_entry
= NULL
;
1509 h
->all_io_entry
= NULL
;
1510 h
->overwrite_mapping
= NULL
;
1514 * Non-blocking function called from the thin target's map function.
1516 static int thin_bio_map(struct dm_target
*ti
, struct bio
*bio
)
1519 struct thin_c
*tc
= ti
->private;
1520 dm_block_t block
= get_bio_block(tc
, bio
);
1521 struct dm_thin_device
*td
= tc
->td
;
1522 struct dm_thin_lookup_result result
;
1523 struct dm_bio_prison_cell cell1
, cell2
;
1524 struct dm_bio_prison_cell
*cell_result
;
1525 struct dm_cell_key key
;
1527 thin_hook_bio(tc
, bio
);
1529 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
1531 return DM_MAPIO_SUBMITTED
;
1534 if (bio
->bi_rw
& (REQ_DISCARD
| REQ_FLUSH
| REQ_FUA
)) {
1535 thin_defer_bio(tc
, bio
);
1536 return DM_MAPIO_SUBMITTED
;
1539 r
= dm_thin_find_block(td
, block
, 0, &result
);
1542 * Note that we defer readahead too.
1546 if (unlikely(result
.shared
)) {
1548 * We have a race condition here between the
1549 * result.shared value returned by the lookup and
1550 * snapshot creation, which may cause new
1553 * To avoid this always quiesce the origin before
1554 * taking the snap. You want to do this anyway to
1555 * ensure a consistent application view
1558 * More distant ancestors are irrelevant. The
1559 * shared flag will be set in their case.
1561 thin_defer_bio(tc
, bio
);
1562 return DM_MAPIO_SUBMITTED
;
1565 build_virtual_key(tc
->td
, block
, &key
);
1566 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell1
, &cell_result
))
1567 return DM_MAPIO_SUBMITTED
;
1569 build_data_key(tc
->td
, result
.block
, &key
);
1570 if (dm_bio_detain(tc
->pool
->prison
, &key
, bio
, &cell2
, &cell_result
)) {
1571 cell_defer_no_holder_no_free(tc
, &cell1
);
1572 return DM_MAPIO_SUBMITTED
;
1575 inc_all_io_entry(tc
->pool
, bio
);
1576 cell_defer_no_holder_no_free(tc
, &cell2
);
1577 cell_defer_no_holder_no_free(tc
, &cell1
);
1579 remap(tc
, bio
, result
.block
);
1580 return DM_MAPIO_REMAPPED
;
1583 if (get_pool_mode(tc
->pool
) == PM_READ_ONLY
) {
1585 * This block isn't provisioned, and we have no way
1588 handle_unserviceable_bio(tc
->pool
, bio
);
1589 return DM_MAPIO_SUBMITTED
;
1595 * In future, the failed dm_thin_find_block above could
1596 * provide the hint to load the metadata into cache.
1598 thin_defer_bio(tc
, bio
);
1599 return DM_MAPIO_SUBMITTED
;
1603 * Must always call bio_io_error on failure.
1604 * dm_thin_find_block can fail with -EINVAL if the
1605 * pool is switched to fail-io mode.
1608 return DM_MAPIO_SUBMITTED
;
1612 static int pool_is_congested(struct dm_target_callbacks
*cb
, int bdi_bits
)
1615 unsigned long flags
;
1616 struct pool_c
*pt
= container_of(cb
, struct pool_c
, callbacks
);
1618 spin_lock_irqsave(&pt
->pool
->lock
, flags
);
1619 r
= !bio_list_empty(&pt
->pool
->retry_on_resume_list
);
1620 spin_unlock_irqrestore(&pt
->pool
->lock
, flags
);
1623 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1624 r
= bdi_congested(&q
->backing_dev_info
, bdi_bits
);
1630 static void __requeue_bios(struct pool
*pool
)
1632 bio_list_merge(&pool
->deferred_bios
, &pool
->retry_on_resume_list
);
1633 bio_list_init(&pool
->retry_on_resume_list
);
1636 /*----------------------------------------------------------------
1637 * Binding of control targets to a pool object
1638 *--------------------------------------------------------------*/
1639 static bool data_dev_supports_discard(struct pool_c
*pt
)
1641 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
1643 return q
&& blk_queue_discard(q
);
1646 static bool is_factor(sector_t block_size
, uint32_t n
)
1648 return !sector_div(block_size
, n
);
1652 * If discard_passdown was enabled verify that the data device
1653 * supports discards. Disable discard_passdown if not.
1655 static void disable_passdown_if_not_supported(struct pool_c
*pt
)
1657 struct pool
*pool
= pt
->pool
;
1658 struct block_device
*data_bdev
= pt
->data_dev
->bdev
;
1659 struct queue_limits
*data_limits
= &bdev_get_queue(data_bdev
)->limits
;
1660 sector_t block_size
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
1661 const char *reason
= NULL
;
1662 char buf
[BDEVNAME_SIZE
];
1664 if (!pt
->adjusted_pf
.discard_passdown
)
1667 if (!data_dev_supports_discard(pt
))
1668 reason
= "discard unsupported";
1670 else if (data_limits
->max_discard_sectors
< pool
->sectors_per_block
)
1671 reason
= "max discard sectors smaller than a block";
1673 else if (data_limits
->discard_granularity
> block_size
)
1674 reason
= "discard granularity larger than a block";
1676 else if (!is_factor(block_size
, data_limits
->discard_granularity
))
1677 reason
= "discard granularity not a factor of block size";
1680 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev
, buf
), reason
);
1681 pt
->adjusted_pf
.discard_passdown
= false;
1685 static int bind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1687 struct pool_c
*pt
= ti
->private;
1690 * We want to make sure that a pool in PM_FAIL mode is never upgraded.
1692 enum pool_mode old_mode
= pool
->pf
.mode
;
1693 enum pool_mode new_mode
= pt
->adjusted_pf
.mode
;
1696 * If we were in PM_FAIL mode, rollback of metadata failed. We're
1697 * not going to recover without a thin_repair. So we never let the
1698 * pool move out of the old mode. On the other hand a PM_READ_ONLY
1699 * may have been due to a lack of metadata or data space, and may
1700 * now work (ie. if the underlying devices have been resized).
1702 if (old_mode
== PM_FAIL
)
1703 new_mode
= old_mode
;
1706 pool
->low_water_blocks
= pt
->low_water_blocks
;
1707 pool
->pf
= pt
->adjusted_pf
;
1709 set_pool_mode(pool
, new_mode
);
1714 static void unbind_control_target(struct pool
*pool
, struct dm_target
*ti
)
1720 /*----------------------------------------------------------------
1722 *--------------------------------------------------------------*/
1723 /* Initialize pool features. */
1724 static void pool_features_init(struct pool_features
*pf
)
1726 pf
->mode
= PM_WRITE
;
1727 pf
->zero_new_blocks
= true;
1728 pf
->discard_enabled
= true;
1729 pf
->discard_passdown
= true;
1730 pf
->error_if_no_space
= false;
1733 static void __pool_destroy(struct pool
*pool
)
1735 __pool_table_remove(pool
);
1737 if (dm_pool_metadata_close(pool
->pmd
) < 0)
1738 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1740 dm_bio_prison_destroy(pool
->prison
);
1741 dm_kcopyd_client_destroy(pool
->copier
);
1744 destroy_workqueue(pool
->wq
);
1746 if (pool
->next_mapping
)
1747 mempool_free(pool
->next_mapping
, pool
->mapping_pool
);
1748 mempool_destroy(pool
->mapping_pool
);
1749 dm_deferred_set_destroy(pool
->shared_read_ds
);
1750 dm_deferred_set_destroy(pool
->all_io_ds
);
1754 static struct kmem_cache
*_new_mapping_cache
;
1756 static struct pool
*pool_create(struct mapped_device
*pool_md
,
1757 struct block_device
*metadata_dev
,
1758 unsigned long block_size
,
1759 int read_only
, char **error
)
1764 struct dm_pool_metadata
*pmd
;
1765 bool format_device
= read_only
? false : true;
1767 pmd
= dm_pool_metadata_open(metadata_dev
, block_size
, format_device
);
1769 *error
= "Error creating metadata object";
1770 return (struct pool
*)pmd
;
1773 pool
= kmalloc(sizeof(*pool
), GFP_KERNEL
);
1775 *error
= "Error allocating memory for pool";
1776 err_p
= ERR_PTR(-ENOMEM
);
1781 pool
->sectors_per_block
= block_size
;
1782 if (block_size
& (block_size
- 1))
1783 pool
->sectors_per_block_shift
= -1;
1785 pool
->sectors_per_block_shift
= __ffs(block_size
);
1786 pool
->low_water_blocks
= 0;
1787 pool_features_init(&pool
->pf
);
1788 pool
->prison
= dm_bio_prison_create(PRISON_CELLS
);
1789 if (!pool
->prison
) {
1790 *error
= "Error creating pool's bio prison";
1791 err_p
= ERR_PTR(-ENOMEM
);
1795 pool
->copier
= dm_kcopyd_client_create(&dm_kcopyd_throttle
);
1796 if (IS_ERR(pool
->copier
)) {
1797 r
= PTR_ERR(pool
->copier
);
1798 *error
= "Error creating pool's kcopyd client";
1800 goto bad_kcopyd_client
;
1804 * Create singlethreaded workqueue that will service all devices
1805 * that use this metadata.
1807 pool
->wq
= alloc_ordered_workqueue("dm-" DM_MSG_PREFIX
, WQ_MEM_RECLAIM
);
1809 *error
= "Error creating pool's workqueue";
1810 err_p
= ERR_PTR(-ENOMEM
);
1814 INIT_WORK(&pool
->worker
, do_worker
);
1815 INIT_DELAYED_WORK(&pool
->waker
, do_waker
);
1816 spin_lock_init(&pool
->lock
);
1817 bio_list_init(&pool
->deferred_bios
);
1818 bio_list_init(&pool
->deferred_flush_bios
);
1819 INIT_LIST_HEAD(&pool
->prepared_mappings
);
1820 INIT_LIST_HEAD(&pool
->prepared_discards
);
1821 pool
->low_water_triggered
= false;
1822 pool
->no_free_space
= false;
1823 bio_list_init(&pool
->retry_on_resume_list
);
1825 pool
->shared_read_ds
= dm_deferred_set_create();
1826 if (!pool
->shared_read_ds
) {
1827 *error
= "Error creating pool's shared read deferred set";
1828 err_p
= ERR_PTR(-ENOMEM
);
1829 goto bad_shared_read_ds
;
1832 pool
->all_io_ds
= dm_deferred_set_create();
1833 if (!pool
->all_io_ds
) {
1834 *error
= "Error creating pool's all io deferred set";
1835 err_p
= ERR_PTR(-ENOMEM
);
1839 pool
->next_mapping
= NULL
;
1840 pool
->mapping_pool
= mempool_create_slab_pool(MAPPING_POOL_SIZE
,
1841 _new_mapping_cache
);
1842 if (!pool
->mapping_pool
) {
1843 *error
= "Error creating pool's mapping mempool";
1844 err_p
= ERR_PTR(-ENOMEM
);
1845 goto bad_mapping_pool
;
1848 pool
->ref_count
= 1;
1849 pool
->last_commit_jiffies
= jiffies
;
1850 pool
->pool_md
= pool_md
;
1851 pool
->md_dev
= metadata_dev
;
1852 __pool_table_insert(pool
);
1857 dm_deferred_set_destroy(pool
->all_io_ds
);
1859 dm_deferred_set_destroy(pool
->shared_read_ds
);
1861 destroy_workqueue(pool
->wq
);
1863 dm_kcopyd_client_destroy(pool
->copier
);
1865 dm_bio_prison_destroy(pool
->prison
);
1869 if (dm_pool_metadata_close(pmd
))
1870 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
1875 static void __pool_inc(struct pool
*pool
)
1877 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
1881 static void __pool_dec(struct pool
*pool
)
1883 BUG_ON(!mutex_is_locked(&dm_thin_pool_table
.mutex
));
1884 BUG_ON(!pool
->ref_count
);
1885 if (!--pool
->ref_count
)
1886 __pool_destroy(pool
);
1889 static struct pool
*__pool_find(struct mapped_device
*pool_md
,
1890 struct block_device
*metadata_dev
,
1891 unsigned long block_size
, int read_only
,
1892 char **error
, int *created
)
1894 struct pool
*pool
= __pool_table_lookup_metadata_dev(metadata_dev
);
1897 if (pool
->pool_md
!= pool_md
) {
1898 *error
= "metadata device already in use by a pool";
1899 return ERR_PTR(-EBUSY
);
1904 pool
= __pool_table_lookup(pool_md
);
1906 if (pool
->md_dev
!= metadata_dev
) {
1907 *error
= "different pool cannot replace a pool";
1908 return ERR_PTR(-EINVAL
);
1913 pool
= pool_create(pool_md
, metadata_dev
, block_size
, read_only
, error
);
1921 /*----------------------------------------------------------------
1922 * Pool target methods
1923 *--------------------------------------------------------------*/
1924 static void pool_dtr(struct dm_target
*ti
)
1926 struct pool_c
*pt
= ti
->private;
1928 mutex_lock(&dm_thin_pool_table
.mutex
);
1930 unbind_control_target(pt
->pool
, ti
);
1931 __pool_dec(pt
->pool
);
1932 dm_put_device(ti
, pt
->metadata_dev
);
1933 dm_put_device(ti
, pt
->data_dev
);
1936 mutex_unlock(&dm_thin_pool_table
.mutex
);
1939 static int parse_pool_features(struct dm_arg_set
*as
, struct pool_features
*pf
,
1940 struct dm_target
*ti
)
1944 const char *arg_name
;
1946 static struct dm_arg _args
[] = {
1947 {0, 3, "Invalid number of pool feature arguments"},
1951 * No feature arguments supplied.
1956 r
= dm_read_arg_group(_args
, as
, &argc
, &ti
->error
);
1960 while (argc
&& !r
) {
1961 arg_name
= dm_shift_arg(as
);
1964 if (!strcasecmp(arg_name
, "skip_block_zeroing"))
1965 pf
->zero_new_blocks
= false;
1967 else if (!strcasecmp(arg_name
, "ignore_discard"))
1968 pf
->discard_enabled
= false;
1970 else if (!strcasecmp(arg_name
, "no_discard_passdown"))
1971 pf
->discard_passdown
= false;
1973 else if (!strcasecmp(arg_name
, "read_only"))
1974 pf
->mode
= PM_READ_ONLY
;
1976 else if (!strcasecmp(arg_name
, "error_if_no_space"))
1977 pf
->error_if_no_space
= true;
1980 ti
->error
= "Unrecognised pool feature requested";
1989 static void metadata_low_callback(void *context
)
1991 struct pool
*pool
= context
;
1993 DMWARN("%s: reached low water mark for metadata device: sending event.",
1994 dm_device_name(pool
->pool_md
));
1996 dm_table_event(pool
->ti
->table
);
1999 static sector_t
get_metadata_dev_size(struct block_device
*bdev
)
2001 sector_t metadata_dev_size
= i_size_read(bdev
->bd_inode
) >> SECTOR_SHIFT
;
2002 char buffer
[BDEVNAME_SIZE
];
2004 if (metadata_dev_size
> THIN_METADATA_MAX_SECTORS_WARNING
) {
2005 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
2006 bdevname(bdev
, buffer
), THIN_METADATA_MAX_SECTORS
);
2007 metadata_dev_size
= THIN_METADATA_MAX_SECTORS_WARNING
;
2010 return metadata_dev_size
;
2013 static dm_block_t
get_metadata_dev_size_in_blocks(struct block_device
*bdev
)
2015 sector_t metadata_dev_size
= get_metadata_dev_size(bdev
);
2017 sector_div(metadata_dev_size
, THIN_METADATA_BLOCK_SIZE
>> SECTOR_SHIFT
);
2019 return metadata_dev_size
;
2023 * When a metadata threshold is crossed a dm event is triggered, and
2024 * userland should respond by growing the metadata device. We could let
2025 * userland set the threshold, like we do with the data threshold, but I'm
2026 * not sure they know enough to do this well.
2028 static dm_block_t
calc_metadata_threshold(struct pool_c
*pt
)
2031 * 4M is ample for all ops with the possible exception of thin
2032 * device deletion which is harmless if it fails (just retry the
2033 * delete after you've grown the device).
2035 dm_block_t quarter
= get_metadata_dev_size_in_blocks(pt
->metadata_dev
->bdev
) / 4;
2036 return min((dm_block_t
)1024ULL /* 4M */, quarter
);
2040 * thin-pool <metadata dev> <data dev>
2041 * <data block size (sectors)>
2042 * <low water mark (blocks)>
2043 * [<#feature args> [<arg>]*]
2045 * Optional feature arguments are:
2046 * skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
2047 * ignore_discard: disable discard
2048 * no_discard_passdown: don't pass discards down to the data device
2049 * read_only: Don't allow any changes to be made to the pool metadata.
2050 * error_if_no_space: error IOs, instead of queueing, if no space.
2052 static int pool_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2054 int r
, pool_created
= 0;
2057 struct pool_features pf
;
2058 struct dm_arg_set as
;
2059 struct dm_dev
*data_dev
;
2060 unsigned long block_size
;
2061 dm_block_t low_water_blocks
;
2062 struct dm_dev
*metadata_dev
;
2063 fmode_t metadata_mode
;
2066 * FIXME Remove validation from scope of lock.
2068 mutex_lock(&dm_thin_pool_table
.mutex
);
2071 ti
->error
= "Invalid argument count";
2080 * Set default pool features.
2082 pool_features_init(&pf
);
2084 dm_consume_args(&as
, 4);
2085 r
= parse_pool_features(&as
, &pf
, ti
);
2089 metadata_mode
= FMODE_READ
| ((pf
.mode
== PM_READ_ONLY
) ? 0 : FMODE_WRITE
);
2090 r
= dm_get_device(ti
, argv
[0], metadata_mode
, &metadata_dev
);
2092 ti
->error
= "Error opening metadata block device";
2097 * Run for the side-effect of possibly issuing a warning if the
2098 * device is too big.
2100 (void) get_metadata_dev_size(metadata_dev
->bdev
);
2102 r
= dm_get_device(ti
, argv
[1], FMODE_READ
| FMODE_WRITE
, &data_dev
);
2104 ti
->error
= "Error getting data device";
2108 if (kstrtoul(argv
[2], 10, &block_size
) || !block_size
||
2109 block_size
< DATA_DEV_BLOCK_SIZE_MIN_SECTORS
||
2110 block_size
> DATA_DEV_BLOCK_SIZE_MAX_SECTORS
||
2111 block_size
& (DATA_DEV_BLOCK_SIZE_MIN_SECTORS
- 1)) {
2112 ti
->error
= "Invalid block size";
2117 if (kstrtoull(argv
[3], 10, (unsigned long long *)&low_water_blocks
)) {
2118 ti
->error
= "Invalid low water mark";
2123 pt
= kzalloc(sizeof(*pt
), GFP_KERNEL
);
2129 pool
= __pool_find(dm_table_get_md(ti
->table
), metadata_dev
->bdev
,
2130 block_size
, pf
.mode
== PM_READ_ONLY
, &ti
->error
, &pool_created
);
2137 * 'pool_created' reflects whether this is the first table load.
2138 * Top level discard support is not allowed to be changed after
2139 * initial load. This would require a pool reload to trigger thin
2142 if (!pool_created
&& pf
.discard_enabled
!= pool
->pf
.discard_enabled
) {
2143 ti
->error
= "Discard support cannot be disabled once enabled";
2145 goto out_flags_changed
;
2150 pt
->metadata_dev
= metadata_dev
;
2151 pt
->data_dev
= data_dev
;
2152 pt
->low_water_blocks
= low_water_blocks
;
2153 pt
->adjusted_pf
= pt
->requested_pf
= pf
;
2154 ti
->num_flush_bios
= 1;
2157 * Only need to enable discards if the pool should pass
2158 * them down to the data device. The thin device's discard
2159 * processing will cause mappings to be removed from the btree.
2161 ti
->discard_zeroes_data_unsupported
= true;
2162 if (pf
.discard_enabled
&& pf
.discard_passdown
) {
2163 ti
->num_discard_bios
= 1;
2166 * Setting 'discards_supported' circumvents the normal
2167 * stacking of discard limits (this keeps the pool and
2168 * thin devices' discard limits consistent).
2170 ti
->discards_supported
= true;
2174 r
= dm_pool_register_metadata_threshold(pt
->pool
->pmd
,
2175 calc_metadata_threshold(pt
),
2176 metadata_low_callback
,
2181 pt
->callbacks
.congested_fn
= pool_is_congested
;
2182 dm_table_add_target_callbacks(ti
->table
, &pt
->callbacks
);
2184 mutex_unlock(&dm_thin_pool_table
.mutex
);
2193 dm_put_device(ti
, data_dev
);
2195 dm_put_device(ti
, metadata_dev
);
2197 mutex_unlock(&dm_thin_pool_table
.mutex
);
2202 static int pool_map(struct dm_target
*ti
, struct bio
*bio
)
2205 struct pool_c
*pt
= ti
->private;
2206 struct pool
*pool
= pt
->pool
;
2207 unsigned long flags
;
2210 * As this is a singleton target, ti->begin is always zero.
2212 spin_lock_irqsave(&pool
->lock
, flags
);
2213 bio
->bi_bdev
= pt
->data_dev
->bdev
;
2214 r
= DM_MAPIO_REMAPPED
;
2215 spin_unlock_irqrestore(&pool
->lock
, flags
);
2220 static int maybe_resize_data_dev(struct dm_target
*ti
, bool *need_commit
)
2223 struct pool_c
*pt
= ti
->private;
2224 struct pool
*pool
= pt
->pool
;
2225 sector_t data_size
= ti
->len
;
2226 dm_block_t sb_data_size
;
2228 *need_commit
= false;
2230 (void) sector_div(data_size
, pool
->sectors_per_block
);
2232 r
= dm_pool_get_data_dev_size(pool
->pmd
, &sb_data_size
);
2234 DMERR("%s: failed to retrieve data device size",
2235 dm_device_name(pool
->pool_md
));
2239 if (data_size
< sb_data_size
) {
2240 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
2241 dm_device_name(pool
->pool_md
),
2242 (unsigned long long)data_size
, sb_data_size
);
2245 } else if (data_size
> sb_data_size
) {
2247 DMINFO("%s: growing the data device from %llu to %llu blocks",
2248 dm_device_name(pool
->pool_md
),
2249 sb_data_size
, (unsigned long long)data_size
);
2250 r
= dm_pool_resize_data_dev(pool
->pmd
, data_size
);
2252 metadata_operation_failed(pool
, "dm_pool_resize_data_dev", r
);
2256 *need_commit
= true;
2262 static int maybe_resize_metadata_dev(struct dm_target
*ti
, bool *need_commit
)
2265 struct pool_c
*pt
= ti
->private;
2266 struct pool
*pool
= pt
->pool
;
2267 dm_block_t metadata_dev_size
, sb_metadata_dev_size
;
2269 *need_commit
= false;
2271 metadata_dev_size
= get_metadata_dev_size_in_blocks(pool
->md_dev
);
2273 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &sb_metadata_dev_size
);
2275 DMERR("%s: failed to retrieve metadata device size",
2276 dm_device_name(pool
->pool_md
));
2280 if (metadata_dev_size
< sb_metadata_dev_size
) {
2281 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
2282 dm_device_name(pool
->pool_md
),
2283 metadata_dev_size
, sb_metadata_dev_size
);
2286 } else if (metadata_dev_size
> sb_metadata_dev_size
) {
2287 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
2288 dm_device_name(pool
->pool_md
),
2289 sb_metadata_dev_size
, metadata_dev_size
);
2290 r
= dm_pool_resize_metadata_dev(pool
->pmd
, metadata_dev_size
);
2292 metadata_operation_failed(pool
, "dm_pool_resize_metadata_dev", r
);
2296 *need_commit
= true;
2303 * Retrieves the number of blocks of the data device from
2304 * the superblock and compares it to the actual device size,
2305 * thus resizing the data device in case it has grown.
2307 * This both copes with opening preallocated data devices in the ctr
2308 * being followed by a resume
2310 * calling the resume method individually after userspace has
2311 * grown the data device in reaction to a table event.
2313 static int pool_preresume(struct dm_target
*ti
)
2316 bool need_commit1
, need_commit2
;
2317 struct pool_c
*pt
= ti
->private;
2318 struct pool
*pool
= pt
->pool
;
2321 * Take control of the pool object.
2323 r
= bind_control_target(pool
, ti
);
2327 r
= maybe_resize_data_dev(ti
, &need_commit1
);
2331 r
= maybe_resize_metadata_dev(ti
, &need_commit2
);
2335 if (need_commit1
|| need_commit2
)
2336 (void) commit(pool
);
2341 static void pool_resume(struct dm_target
*ti
)
2343 struct pool_c
*pt
= ti
->private;
2344 struct pool
*pool
= pt
->pool
;
2345 unsigned long flags
;
2347 spin_lock_irqsave(&pool
->lock
, flags
);
2348 pool
->low_water_triggered
= false;
2349 pool
->no_free_space
= false;
2350 __requeue_bios(pool
);
2351 spin_unlock_irqrestore(&pool
->lock
, flags
);
2353 do_waker(&pool
->waker
.work
);
2356 static void pool_postsuspend(struct dm_target
*ti
)
2358 struct pool_c
*pt
= ti
->private;
2359 struct pool
*pool
= pt
->pool
;
2361 cancel_delayed_work(&pool
->waker
);
2362 flush_workqueue(pool
->wq
);
2363 (void) commit(pool
);
2366 static int check_arg_count(unsigned argc
, unsigned args_required
)
2368 if (argc
!= args_required
) {
2369 DMWARN("Message received with %u arguments instead of %u.",
2370 argc
, args_required
);
2377 static int read_dev_id(char *arg
, dm_thin_id
*dev_id
, int warning
)
2379 if (!kstrtoull(arg
, 10, (unsigned long long *)dev_id
) &&
2380 *dev_id
<= MAX_DEV_ID
)
2384 DMWARN("Message received with invalid device id: %s", arg
);
2389 static int process_create_thin_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2394 r
= check_arg_count(argc
, 2);
2398 r
= read_dev_id(argv
[1], &dev_id
, 1);
2402 r
= dm_pool_create_thin(pool
->pmd
, dev_id
);
2404 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
2412 static int process_create_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2415 dm_thin_id origin_dev_id
;
2418 r
= check_arg_count(argc
, 3);
2422 r
= read_dev_id(argv
[1], &dev_id
, 1);
2426 r
= read_dev_id(argv
[2], &origin_dev_id
, 1);
2430 r
= dm_pool_create_snap(pool
->pmd
, dev_id
, origin_dev_id
);
2432 DMWARN("Creation of new snapshot %s of device %s failed.",
2440 static int process_delete_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2445 r
= check_arg_count(argc
, 2);
2449 r
= read_dev_id(argv
[1], &dev_id
, 1);
2453 r
= dm_pool_delete_thin_device(pool
->pmd
, dev_id
);
2455 DMWARN("Deletion of thin device %s failed.", argv
[1]);
2460 static int process_set_transaction_id_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2462 dm_thin_id old_id
, new_id
;
2465 r
= check_arg_count(argc
, 3);
2469 if (kstrtoull(argv
[1], 10, (unsigned long long *)&old_id
)) {
2470 DMWARN("set_transaction_id message: Unrecognised id %s.", argv
[1]);
2474 if (kstrtoull(argv
[2], 10, (unsigned long long *)&new_id
)) {
2475 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv
[2]);
2479 r
= dm_pool_set_metadata_transaction_id(pool
->pmd
, old_id
, new_id
);
2481 DMWARN("Failed to change transaction id from %s to %s.",
2489 static int process_reserve_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2493 r
= check_arg_count(argc
, 1);
2497 (void) commit(pool
);
2499 r
= dm_pool_reserve_metadata_snap(pool
->pmd
);
2501 DMWARN("reserve_metadata_snap message failed.");
2506 static int process_release_metadata_snap_mesg(unsigned argc
, char **argv
, struct pool
*pool
)
2510 r
= check_arg_count(argc
, 1);
2514 r
= dm_pool_release_metadata_snap(pool
->pmd
);
2516 DMWARN("release_metadata_snap message failed.");
2522 * Messages supported:
2523 * create_thin <dev_id>
2524 * create_snap <dev_id> <origin_id>
2526 * trim <dev_id> <new_size_in_sectors>
2527 * set_transaction_id <current_trans_id> <new_trans_id>
2528 * reserve_metadata_snap
2529 * release_metadata_snap
2531 static int pool_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
2534 struct pool_c
*pt
= ti
->private;
2535 struct pool
*pool
= pt
->pool
;
2537 if (!strcasecmp(argv
[0], "create_thin"))
2538 r
= process_create_thin_mesg(argc
, argv
, pool
);
2540 else if (!strcasecmp(argv
[0], "create_snap"))
2541 r
= process_create_snap_mesg(argc
, argv
, pool
);
2543 else if (!strcasecmp(argv
[0], "delete"))
2544 r
= process_delete_mesg(argc
, argv
, pool
);
2546 else if (!strcasecmp(argv
[0], "set_transaction_id"))
2547 r
= process_set_transaction_id_mesg(argc
, argv
, pool
);
2549 else if (!strcasecmp(argv
[0], "reserve_metadata_snap"))
2550 r
= process_reserve_metadata_snap_mesg(argc
, argv
, pool
);
2552 else if (!strcasecmp(argv
[0], "release_metadata_snap"))
2553 r
= process_release_metadata_snap_mesg(argc
, argv
, pool
);
2556 DMWARN("Unrecognised thin pool target message received: %s", argv
[0]);
2559 (void) commit(pool
);
2564 static void emit_flags(struct pool_features
*pf
, char *result
,
2565 unsigned sz
, unsigned maxlen
)
2567 unsigned count
= !pf
->zero_new_blocks
+ !pf
->discard_enabled
+
2568 !pf
->discard_passdown
+ (pf
->mode
== PM_READ_ONLY
) +
2569 pf
->error_if_no_space
;
2570 DMEMIT("%u ", count
);
2572 if (!pf
->zero_new_blocks
)
2573 DMEMIT("skip_block_zeroing ");
2575 if (!pf
->discard_enabled
)
2576 DMEMIT("ignore_discard ");
2578 if (!pf
->discard_passdown
)
2579 DMEMIT("no_discard_passdown ");
2581 if (pf
->mode
== PM_READ_ONLY
)
2582 DMEMIT("read_only ");
2584 if (pf
->error_if_no_space
)
2585 DMEMIT("error_if_no_space ");
2590 * <transaction id> <used metadata sectors>/<total metadata sectors>
2591 * <used data sectors>/<total data sectors> <held metadata root>
2593 static void pool_status(struct dm_target
*ti
, status_type_t type
,
2594 unsigned status_flags
, char *result
, unsigned maxlen
)
2598 uint64_t transaction_id
;
2599 dm_block_t nr_free_blocks_data
;
2600 dm_block_t nr_free_blocks_metadata
;
2601 dm_block_t nr_blocks_data
;
2602 dm_block_t nr_blocks_metadata
;
2603 dm_block_t held_root
;
2604 char buf
[BDEVNAME_SIZE
];
2605 char buf2
[BDEVNAME_SIZE
];
2606 struct pool_c
*pt
= ti
->private;
2607 struct pool
*pool
= pt
->pool
;
2610 case STATUSTYPE_INFO
:
2611 if (get_pool_mode(pool
) == PM_FAIL
) {
2616 /* Commit to ensure statistics aren't out-of-date */
2617 if (!(status_flags
& DM_STATUS_NOFLUSH_FLAG
) && !dm_suspended(ti
))
2618 (void) commit(pool
);
2620 r
= dm_pool_get_metadata_transaction_id(pool
->pmd
, &transaction_id
);
2622 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
2623 dm_device_name(pool
->pool_md
), r
);
2627 r
= dm_pool_get_free_metadata_block_count(pool
->pmd
, &nr_free_blocks_metadata
);
2629 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
2630 dm_device_name(pool
->pool_md
), r
);
2634 r
= dm_pool_get_metadata_dev_size(pool
->pmd
, &nr_blocks_metadata
);
2636 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
2637 dm_device_name(pool
->pool_md
), r
);
2641 r
= dm_pool_get_free_block_count(pool
->pmd
, &nr_free_blocks_data
);
2643 DMERR("%s: dm_pool_get_free_block_count returned %d",
2644 dm_device_name(pool
->pool_md
), r
);
2648 r
= dm_pool_get_data_dev_size(pool
->pmd
, &nr_blocks_data
);
2650 DMERR("%s: dm_pool_get_data_dev_size returned %d",
2651 dm_device_name(pool
->pool_md
), r
);
2655 r
= dm_pool_get_metadata_snap(pool
->pmd
, &held_root
);
2657 DMERR("%s: dm_pool_get_metadata_snap returned %d",
2658 dm_device_name(pool
->pool_md
), r
);
2662 DMEMIT("%llu %llu/%llu %llu/%llu ",
2663 (unsigned long long)transaction_id
,
2664 (unsigned long long)(nr_blocks_metadata
- nr_free_blocks_metadata
),
2665 (unsigned long long)nr_blocks_metadata
,
2666 (unsigned long long)(nr_blocks_data
- nr_free_blocks_data
),
2667 (unsigned long long)nr_blocks_data
);
2670 DMEMIT("%llu ", held_root
);
2674 if (pool
->pf
.mode
== PM_READ_ONLY
)
2679 if (!pool
->pf
.discard_enabled
)
2680 DMEMIT("ignore_discard ");
2681 else if (pool
->pf
.discard_passdown
)
2682 DMEMIT("discard_passdown ");
2684 DMEMIT("no_discard_passdown ");
2686 if (pool
->pf
.error_if_no_space
)
2687 DMEMIT("error_if_no_space ");
2689 DMEMIT("queue_if_no_space ");
2693 case STATUSTYPE_TABLE
:
2694 DMEMIT("%s %s %lu %llu ",
2695 format_dev_t(buf
, pt
->metadata_dev
->bdev
->bd_dev
),
2696 format_dev_t(buf2
, pt
->data_dev
->bdev
->bd_dev
),
2697 (unsigned long)pool
->sectors_per_block
,
2698 (unsigned long long)pt
->low_water_blocks
);
2699 emit_flags(&pt
->requested_pf
, result
, sz
, maxlen
);
2708 static int pool_iterate_devices(struct dm_target
*ti
,
2709 iterate_devices_callout_fn fn
, void *data
)
2711 struct pool_c
*pt
= ti
->private;
2713 return fn(ti
, pt
->data_dev
, 0, ti
->len
, data
);
2716 static int pool_merge(struct dm_target
*ti
, struct bvec_merge_data
*bvm
,
2717 struct bio_vec
*biovec
, int max_size
)
2719 struct pool_c
*pt
= ti
->private;
2720 struct request_queue
*q
= bdev_get_queue(pt
->data_dev
->bdev
);
2722 if (!q
->merge_bvec_fn
)
2725 bvm
->bi_bdev
= pt
->data_dev
->bdev
;
2727 return min(max_size
, q
->merge_bvec_fn(q
, bvm
, biovec
));
2730 static void set_discard_limits(struct pool_c
*pt
, struct queue_limits
*limits
)
2732 struct pool
*pool
= pt
->pool
;
2733 struct queue_limits
*data_limits
;
2735 limits
->max_discard_sectors
= pool
->sectors_per_block
;
2738 * discard_granularity is just a hint, and not enforced.
2740 if (pt
->adjusted_pf
.discard_passdown
) {
2741 data_limits
= &bdev_get_queue(pt
->data_dev
->bdev
)->limits
;
2742 limits
->discard_granularity
= data_limits
->discard_granularity
;
2744 limits
->discard_granularity
= pool
->sectors_per_block
<< SECTOR_SHIFT
;
2747 static void pool_io_hints(struct dm_target
*ti
, struct queue_limits
*limits
)
2749 struct pool_c
*pt
= ti
->private;
2750 struct pool
*pool
= pt
->pool
;
2751 uint64_t io_opt_sectors
= limits
->io_opt
>> SECTOR_SHIFT
;
2754 * If the system-determined stacked limits are compatible with the
2755 * pool's blocksize (io_opt is a factor) do not override them.
2757 if (io_opt_sectors
< pool
->sectors_per_block
||
2758 do_div(io_opt_sectors
, pool
->sectors_per_block
)) {
2759 blk_limits_io_min(limits
, 0);
2760 blk_limits_io_opt(limits
, pool
->sectors_per_block
<< SECTOR_SHIFT
);
2764 * pt->adjusted_pf is a staging area for the actual features to use.
2765 * They get transferred to the live pool in bind_control_target()
2766 * called from pool_preresume().
2768 if (!pt
->adjusted_pf
.discard_enabled
) {
2770 * Must explicitly disallow stacking discard limits otherwise the
2771 * block layer will stack them if pool's data device has support.
2772 * QUEUE_FLAG_DISCARD wouldn't be set but there is no way for the
2773 * user to see that, so make sure to set all discard limits to 0.
2775 limits
->discard_granularity
= 0;
2779 disable_passdown_if_not_supported(pt
);
2781 set_discard_limits(pt
, limits
);
2784 static struct target_type pool_target
= {
2785 .name
= "thin-pool",
2786 .features
= DM_TARGET_SINGLETON
| DM_TARGET_ALWAYS_WRITEABLE
|
2787 DM_TARGET_IMMUTABLE
,
2788 .version
= {1, 10, 0},
2789 .module
= THIS_MODULE
,
2793 .postsuspend
= pool_postsuspend
,
2794 .preresume
= pool_preresume
,
2795 .resume
= pool_resume
,
2796 .message
= pool_message
,
2797 .status
= pool_status
,
2798 .merge
= pool_merge
,
2799 .iterate_devices
= pool_iterate_devices
,
2800 .io_hints
= pool_io_hints
,
2803 /*----------------------------------------------------------------
2804 * Thin target methods
2805 *--------------------------------------------------------------*/
2806 static void thin_dtr(struct dm_target
*ti
)
2808 struct thin_c
*tc
= ti
->private;
2810 mutex_lock(&dm_thin_pool_table
.mutex
);
2812 __pool_dec(tc
->pool
);
2813 dm_pool_close_thin_device(tc
->td
);
2814 dm_put_device(ti
, tc
->pool_dev
);
2816 dm_put_device(ti
, tc
->origin_dev
);
2819 mutex_unlock(&dm_thin_pool_table
.mutex
);
2823 * Thin target parameters:
2825 * <pool_dev> <dev_id> [origin_dev]
2827 * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
2828 * dev_id: the internal device identifier
2829 * origin_dev: a device external to the pool that should act as the origin
2831 * If the pool device has discards disabled, they get disabled for the thin
2834 static int thin_ctr(struct dm_target
*ti
, unsigned argc
, char **argv
)
2838 struct dm_dev
*pool_dev
, *origin_dev
;
2839 struct mapped_device
*pool_md
;
2841 mutex_lock(&dm_thin_pool_table
.mutex
);
2843 if (argc
!= 2 && argc
!= 3) {
2844 ti
->error
= "Invalid argument count";
2849 tc
= ti
->private = kzalloc(sizeof(*tc
), GFP_KERNEL
);
2851 ti
->error
= "Out of memory";
2857 r
= dm_get_device(ti
, argv
[2], FMODE_READ
, &origin_dev
);
2859 ti
->error
= "Error opening origin device";
2860 goto bad_origin_dev
;
2862 tc
->origin_dev
= origin_dev
;
2865 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &pool_dev
);
2867 ti
->error
= "Error opening pool device";
2870 tc
->pool_dev
= pool_dev
;
2872 if (read_dev_id(argv
[1], (unsigned long long *)&tc
->dev_id
, 0)) {
2873 ti
->error
= "Invalid device id";
2878 pool_md
= dm_get_md(tc
->pool_dev
->bdev
->bd_dev
);
2880 ti
->error
= "Couldn't get pool mapped device";
2885 tc
->pool
= __pool_table_lookup(pool_md
);
2887 ti
->error
= "Couldn't find pool object";
2889 goto bad_pool_lookup
;
2891 __pool_inc(tc
->pool
);
2893 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
2894 ti
->error
= "Couldn't open thin device, Pool is in fail mode";
2898 r
= dm_pool_open_thin_device(tc
->pool
->pmd
, tc
->dev_id
, &tc
->td
);
2900 ti
->error
= "Couldn't open thin internal device";
2904 r
= dm_set_target_max_io_len(ti
, tc
->pool
->sectors_per_block
);
2908 ti
->num_flush_bios
= 1;
2909 ti
->flush_supported
= true;
2910 ti
->per_bio_data_size
= sizeof(struct dm_thin_endio_hook
);
2912 /* In case the pool supports discards, pass them on. */
2913 ti
->discard_zeroes_data_unsupported
= true;
2914 if (tc
->pool
->pf
.discard_enabled
) {
2915 ti
->discards_supported
= true;
2916 ti
->num_discard_bios
= 1;
2917 /* Discard bios must be split on a block boundary */
2918 ti
->split_discard_bios
= true;
2923 mutex_unlock(&dm_thin_pool_table
.mutex
);
2928 __pool_dec(tc
->pool
);
2932 dm_put_device(ti
, tc
->pool_dev
);
2935 dm_put_device(ti
, tc
->origin_dev
);
2939 mutex_unlock(&dm_thin_pool_table
.mutex
);
2944 static int thin_map(struct dm_target
*ti
, struct bio
*bio
)
2946 bio
->bi_sector
= dm_target_offset(ti
, bio
->bi_sector
);
2948 return thin_bio_map(ti
, bio
);
2951 static int thin_endio(struct dm_target
*ti
, struct bio
*bio
, int err
)
2953 unsigned long flags
;
2954 struct dm_thin_endio_hook
*h
= dm_per_bio_data(bio
, sizeof(struct dm_thin_endio_hook
));
2955 struct list_head work
;
2956 struct dm_thin_new_mapping
*m
, *tmp
;
2957 struct pool
*pool
= h
->tc
->pool
;
2959 if (h
->shared_read_entry
) {
2960 INIT_LIST_HEAD(&work
);
2961 dm_deferred_entry_dec(h
->shared_read_entry
, &work
);
2963 spin_lock_irqsave(&pool
->lock
, flags
);
2964 list_for_each_entry_safe(m
, tmp
, &work
, list
) {
2967 __maybe_add_mapping(m
);
2969 spin_unlock_irqrestore(&pool
->lock
, flags
);
2972 if (h
->all_io_entry
) {
2973 INIT_LIST_HEAD(&work
);
2974 dm_deferred_entry_dec(h
->all_io_entry
, &work
);
2975 if (!list_empty(&work
)) {
2976 spin_lock_irqsave(&pool
->lock
, flags
);
2977 list_for_each_entry_safe(m
, tmp
, &work
, list
)
2978 list_add_tail(&m
->list
, &pool
->prepared_discards
);
2979 spin_unlock_irqrestore(&pool
->lock
, flags
);
2987 static void thin_postsuspend(struct dm_target
*ti
)
2989 if (dm_noflush_suspending(ti
))
2990 requeue_io((struct thin_c
*)ti
->private);
2994 * <nr mapped sectors> <highest mapped sector>
2996 static void thin_status(struct dm_target
*ti
, status_type_t type
,
2997 unsigned status_flags
, char *result
, unsigned maxlen
)
3001 dm_block_t mapped
, highest
;
3002 char buf
[BDEVNAME_SIZE
];
3003 struct thin_c
*tc
= ti
->private;
3005 if (get_pool_mode(tc
->pool
) == PM_FAIL
) {
3014 case STATUSTYPE_INFO
:
3015 r
= dm_thin_get_mapped_count(tc
->td
, &mapped
);
3017 DMERR("dm_thin_get_mapped_count returned %d", r
);
3021 r
= dm_thin_get_highest_mapped_block(tc
->td
, &highest
);
3023 DMERR("dm_thin_get_highest_mapped_block returned %d", r
);
3027 DMEMIT("%llu ", mapped
* tc
->pool
->sectors_per_block
);
3029 DMEMIT("%llu", ((highest
+ 1) *
3030 tc
->pool
->sectors_per_block
) - 1);
3035 case STATUSTYPE_TABLE
:
3037 format_dev_t(buf
, tc
->pool_dev
->bdev
->bd_dev
),
3038 (unsigned long) tc
->dev_id
);
3040 DMEMIT(" %s", format_dev_t(buf
, tc
->origin_dev
->bdev
->bd_dev
));
3051 static int thin_iterate_devices(struct dm_target
*ti
,
3052 iterate_devices_callout_fn fn
, void *data
)
3055 struct thin_c
*tc
= ti
->private;
3056 struct pool
*pool
= tc
->pool
;
3059 * We can't call dm_pool_get_data_dev_size() since that blocks. So
3060 * we follow a more convoluted path through to the pool's target.
3063 return 0; /* nothing is bound */
3065 blocks
= pool
->ti
->len
;
3066 (void) sector_div(blocks
, pool
->sectors_per_block
);
3068 return fn(ti
, tc
->pool_dev
, 0, pool
->sectors_per_block
* blocks
, data
);
3073 static struct target_type thin_target
= {
3075 .version
= {1, 10, 0},
3076 .module
= THIS_MODULE
,
3080 .end_io
= thin_endio
,
3081 .postsuspend
= thin_postsuspend
,
3082 .status
= thin_status
,
3083 .iterate_devices
= thin_iterate_devices
,
3086 /*----------------------------------------------------------------*/
3088 static int __init
dm_thin_init(void)
3094 r
= dm_register_target(&thin_target
);
3098 r
= dm_register_target(&pool_target
);
3100 goto bad_pool_target
;
3104 _new_mapping_cache
= KMEM_CACHE(dm_thin_new_mapping
, 0);
3105 if (!_new_mapping_cache
)
3106 goto bad_new_mapping_cache
;
3110 bad_new_mapping_cache
:
3111 dm_unregister_target(&pool_target
);
3113 dm_unregister_target(&thin_target
);
3118 static void dm_thin_exit(void)
3120 dm_unregister_target(&thin_target
);
3121 dm_unregister_target(&pool_target
);
3123 kmem_cache_destroy(_new_mapping_cache
);
3126 module_init(dm_thin_init
);
3127 module_exit(dm_thin_exit
);
3129 MODULE_DESCRIPTION(DM_NAME
" thin provisioning target");
3130 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3131 MODULE_LICENSE("GPL");