2 * Copyright (C) 2011-2012 Red Hat, Inc.
4 * This file is released under the GPL.
7 #include "dm-thin-metadata.h"
8 #include "persistent-data/dm-btree.h"
9 #include "persistent-data/dm-space-map.h"
10 #include "persistent-data/dm-space-map-disk.h"
11 #include "persistent-data/dm-transaction-manager.h"
13 #include <linux/list.h>
14 #include <linux/device-mapper.h>
15 #include <linux/workqueue.h>
17 /*--------------------------------------------------------------------------
18 * As far as the metadata goes, there is:
20 * - A superblock in block zero, taking up fewer than 512 bytes for
23 * - A space map managing the metadata blocks.
25 * - A space map managing the data blocks.
27 * - A btree mapping our internal thin dev ids onto struct disk_device_details.
29 * - A hierarchical btree, with 2 levels which effectively maps (thin
30 * dev id, virtual block) -> block_time. Block time is a 64-bit
31 * field holding the time in the low 24 bits, and block in the top 48
34 * BTrees consist solely of btree_nodes, that fill a block. Some are
35 * internal nodes, as such their values are a __le64 pointing to other
36 * nodes. Leaf nodes can store data of any reasonable size (ie. much
37 * smaller than the block size). The nodes consist of the header,
38 * followed by an array of keys, followed by an array of values. We have
39 * to binary search on the keys so they're all held together to help the
42 * Space maps have 2 btrees:
44 * - One maps a uint64_t onto a struct index_entry. Which points to a
45 * bitmap block, and has some details about how many free entries there
48 * - The bitmap blocks have a header (for the checksum). Then the rest
49 * of the block is pairs of bits. With the meaning being:
54 * 3 - ref count is higher than 2
56 * - If the count is higher than 2 then the ref count is entered in a
57 * second btree that directly maps the block_address to a uint32_t ref
60 * The space map metadata variant doesn't have a bitmaps btree. Instead
61 * it has one single blocks worth of index_entries. This avoids
62 * recursive issues with the bitmap btree needing to allocate space in
63 * order to insert. With a small data block size such as 64k the
64 * metadata support data devices that are hundreds of terrabytes.
66 * The space maps allocate space linearly from front to back. Space that
67 * is freed in a transaction is never recycled within that transaction.
68 * To try and avoid fragmenting _free_ space the allocator always goes
69 * back and fills in gaps.
71 * All metadata io is in THIN_METADATA_BLOCK_SIZE sized/aligned chunks
72 * from the block manager.
73 *--------------------------------------------------------------------------*/
75 #define DM_MSG_PREFIX "thin metadata"
77 #define THIN_SUPERBLOCK_MAGIC 27022010
78 #define THIN_SUPERBLOCK_LOCATION 0
79 #define THIN_VERSION 1
80 #define THIN_METADATA_CACHE_SIZE 64
81 #define SECTOR_TO_BLOCK_SHIFT 3
84 * 3 for btree insert +
85 * 2 for btree lookup used within space map
87 #define THIN_MAX_CONCURRENT_LOCKS 5
89 /* This should be plenty */
90 #define SPACE_MAP_ROOT_SIZE 128
93 * Little endian on-disk superblock and device details.
95 struct thin_disk_superblock
{
96 __le32 csum
; /* Checksum of superblock except for this field. */
98 __le64 blocknr
; /* This block number, dm_block_t. */
108 * Root held by userspace transactions.
112 __u8 data_space_map_root
[SPACE_MAP_ROOT_SIZE
];
113 __u8 metadata_space_map_root
[SPACE_MAP_ROOT_SIZE
];
116 * 2-level btree mapping (dev_id, (dev block, time)) -> data block
118 __le64 data_mapping_root
;
121 * Device detail root mapping dev_id -> device_details
123 __le64 device_details_root
;
125 __le32 data_block_size
; /* In 512-byte sectors. */
127 __le32 metadata_block_size
; /* In 512-byte sectors. */
128 __le64 metadata_nr_blocks
;
131 __le32 compat_ro_flags
;
132 __le32 incompat_flags
;
135 struct disk_device_details
{
136 __le64 mapped_blocks
;
137 __le64 transaction_id
; /* When created. */
138 __le32 creation_time
;
139 __le32 snapshotted_time
;
142 struct dm_pool_metadata
{
143 struct hlist_node hash
;
145 struct block_device
*bdev
;
146 struct dm_block_manager
*bm
;
147 struct dm_space_map
*metadata_sm
;
148 struct dm_space_map
*data_sm
;
149 struct dm_transaction_manager
*tm
;
150 struct dm_transaction_manager
*nb_tm
;
154 * First level holds thin_dev_t.
155 * Second level holds mappings.
157 struct dm_btree_info info
;
160 * Non-blocking version of the above.
162 struct dm_btree_info nb_info
;
165 * Just the top level for deleting whole devices.
167 struct dm_btree_info tl_info
;
170 * Just the bottom level for creating new devices.
172 struct dm_btree_info bl_info
;
175 * Describes the device details btree.
177 struct dm_btree_info details_info
;
179 struct rw_semaphore root_lock
;
182 dm_block_t details_root
;
183 struct list_head thin_devices
;
186 sector_t data_block_size
;
190 * Set if a transaction has to be aborted but the attempt to roll back
191 * to the previous (good) transaction failed. The only pool metadata
192 * operation possible in this state is the closing of the device.
197 struct dm_thin_device
{
198 struct list_head list
;
199 struct dm_pool_metadata
*pmd
;
204 bool aborted_with_changes
:1;
205 uint64_t mapped_blocks
;
206 uint64_t transaction_id
;
207 uint32_t creation_time
;
208 uint32_t snapshotted_time
;
211 /*----------------------------------------------------------------
212 * superblock validator
213 *--------------------------------------------------------------*/
215 #define SUPERBLOCK_CSUM_XOR 160774
217 static void sb_prepare_for_write(struct dm_block_validator
*v
,
221 struct thin_disk_superblock
*disk_super
= dm_block_data(b
);
223 disk_super
->blocknr
= cpu_to_le64(dm_block_location(b
));
224 disk_super
->csum
= cpu_to_le32(dm_bm_checksum(&disk_super
->flags
,
225 block_size
- sizeof(__le32
),
226 SUPERBLOCK_CSUM_XOR
));
229 static int sb_check(struct dm_block_validator
*v
,
233 struct thin_disk_superblock
*disk_super
= dm_block_data(b
);
236 if (dm_block_location(b
) != le64_to_cpu(disk_super
->blocknr
)) {
237 DMERR("sb_check failed: blocknr %llu: "
238 "wanted %llu", le64_to_cpu(disk_super
->blocknr
),
239 (unsigned long long)dm_block_location(b
));
243 if (le64_to_cpu(disk_super
->magic
) != THIN_SUPERBLOCK_MAGIC
) {
244 DMERR("sb_check failed: magic %llu: "
245 "wanted %llu", le64_to_cpu(disk_super
->magic
),
246 (unsigned long long)THIN_SUPERBLOCK_MAGIC
);
250 csum_le
= cpu_to_le32(dm_bm_checksum(&disk_super
->flags
,
251 block_size
- sizeof(__le32
),
252 SUPERBLOCK_CSUM_XOR
));
253 if (csum_le
!= disk_super
->csum
) {
254 DMERR("sb_check failed: csum %u: wanted %u",
255 le32_to_cpu(csum_le
), le32_to_cpu(disk_super
->csum
));
262 static struct dm_block_validator sb_validator
= {
263 .name
= "superblock",
264 .prepare_for_write
= sb_prepare_for_write
,
268 /*----------------------------------------------------------------
269 * Methods for the btree value types
270 *--------------------------------------------------------------*/
272 static uint64_t pack_block_time(dm_block_t b
, uint32_t t
)
274 return (b
<< 24) | t
;
277 static void unpack_block_time(uint64_t v
, dm_block_t
*b
, uint32_t *t
)
280 *t
= v
& ((1 << 24) - 1);
283 static void data_block_inc(void *context
, const void *value_le
)
285 struct dm_space_map
*sm
= context
;
290 memcpy(&v_le
, value_le
, sizeof(v_le
));
291 unpack_block_time(le64_to_cpu(v_le
), &b
, &t
);
292 dm_sm_inc_block(sm
, b
);
295 static void data_block_dec(void *context
, const void *value_le
)
297 struct dm_space_map
*sm
= context
;
302 memcpy(&v_le
, value_le
, sizeof(v_le
));
303 unpack_block_time(le64_to_cpu(v_le
), &b
, &t
);
304 dm_sm_dec_block(sm
, b
);
307 static int data_block_equal(void *context
, const void *value1_le
, const void *value2_le
)
313 memcpy(&v1_le
, value1_le
, sizeof(v1_le
));
314 memcpy(&v2_le
, value2_le
, sizeof(v2_le
));
315 unpack_block_time(le64_to_cpu(v1_le
), &b1
, &t
);
316 unpack_block_time(le64_to_cpu(v2_le
), &b2
, &t
);
321 static void subtree_inc(void *context
, const void *value
)
323 struct dm_btree_info
*info
= context
;
327 memcpy(&root_le
, value
, sizeof(root_le
));
328 root
= le64_to_cpu(root_le
);
329 dm_tm_inc(info
->tm
, root
);
332 static void subtree_dec(void *context
, const void *value
)
334 struct dm_btree_info
*info
= context
;
338 memcpy(&root_le
, value
, sizeof(root_le
));
339 root
= le64_to_cpu(root_le
);
340 if (dm_btree_del(info
, root
))
341 DMERR("btree delete failed\n");
344 static int subtree_equal(void *context
, const void *value1_le
, const void *value2_le
)
347 memcpy(&v1_le
, value1_le
, sizeof(v1_le
));
348 memcpy(&v2_le
, value2_le
, sizeof(v2_le
));
350 return v1_le
== v2_le
;
353 /*----------------------------------------------------------------*/
355 static int superblock_lock_zero(struct dm_pool_metadata
*pmd
,
356 struct dm_block
**sblock
)
358 return dm_bm_write_lock_zero(pmd
->bm
, THIN_SUPERBLOCK_LOCATION
,
359 &sb_validator
, sblock
);
362 static int superblock_lock(struct dm_pool_metadata
*pmd
,
363 struct dm_block
**sblock
)
365 return dm_bm_write_lock(pmd
->bm
, THIN_SUPERBLOCK_LOCATION
,
366 &sb_validator
, sblock
);
369 static int __superblock_all_zeroes(struct dm_block_manager
*bm
, int *result
)
374 __le64
*data_le
, zero
= cpu_to_le64(0);
375 unsigned block_size
= dm_bm_block_size(bm
) / sizeof(__le64
);
378 * We can't use a validator here - it may be all zeroes.
380 r
= dm_bm_read_lock(bm
, THIN_SUPERBLOCK_LOCATION
, NULL
, &b
);
384 data_le
= dm_block_data(b
);
386 for (i
= 0; i
< block_size
; i
++) {
387 if (data_le
[i
] != zero
) {
393 return dm_bm_unlock(b
);
396 static void __setup_btree_details(struct dm_pool_metadata
*pmd
)
398 pmd
->info
.tm
= pmd
->tm
;
399 pmd
->info
.levels
= 2;
400 pmd
->info
.value_type
.context
= pmd
->data_sm
;
401 pmd
->info
.value_type
.size
= sizeof(__le64
);
402 pmd
->info
.value_type
.inc
= data_block_inc
;
403 pmd
->info
.value_type
.dec
= data_block_dec
;
404 pmd
->info
.value_type
.equal
= data_block_equal
;
406 memcpy(&pmd
->nb_info
, &pmd
->info
, sizeof(pmd
->nb_info
));
407 pmd
->nb_info
.tm
= pmd
->nb_tm
;
409 pmd
->tl_info
.tm
= pmd
->tm
;
410 pmd
->tl_info
.levels
= 1;
411 pmd
->tl_info
.value_type
.context
= &pmd
->bl_info
;
412 pmd
->tl_info
.value_type
.size
= sizeof(__le64
);
413 pmd
->tl_info
.value_type
.inc
= subtree_inc
;
414 pmd
->tl_info
.value_type
.dec
= subtree_dec
;
415 pmd
->tl_info
.value_type
.equal
= subtree_equal
;
417 pmd
->bl_info
.tm
= pmd
->tm
;
418 pmd
->bl_info
.levels
= 1;
419 pmd
->bl_info
.value_type
.context
= pmd
->data_sm
;
420 pmd
->bl_info
.value_type
.size
= sizeof(__le64
);
421 pmd
->bl_info
.value_type
.inc
= data_block_inc
;
422 pmd
->bl_info
.value_type
.dec
= data_block_dec
;
423 pmd
->bl_info
.value_type
.equal
= data_block_equal
;
425 pmd
->details_info
.tm
= pmd
->tm
;
426 pmd
->details_info
.levels
= 1;
427 pmd
->details_info
.value_type
.context
= NULL
;
428 pmd
->details_info
.value_type
.size
= sizeof(struct disk_device_details
);
429 pmd
->details_info
.value_type
.inc
= NULL
;
430 pmd
->details_info
.value_type
.dec
= NULL
;
431 pmd
->details_info
.value_type
.equal
= NULL
;
434 static int __write_initial_superblock(struct dm_pool_metadata
*pmd
)
437 struct dm_block
*sblock
;
438 size_t metadata_len
, data_len
;
439 struct thin_disk_superblock
*disk_super
;
440 sector_t bdev_size
= i_size_read(pmd
->bdev
->bd_inode
) >> SECTOR_SHIFT
;
442 if (bdev_size
> THIN_METADATA_MAX_SECTORS
)
443 bdev_size
= THIN_METADATA_MAX_SECTORS
;
445 r
= dm_sm_root_size(pmd
->metadata_sm
, &metadata_len
);
449 r
= dm_sm_root_size(pmd
->data_sm
, &data_len
);
453 r
= dm_sm_commit(pmd
->data_sm
);
457 r
= dm_tm_pre_commit(pmd
->tm
);
461 r
= superblock_lock_zero(pmd
, &sblock
);
465 disk_super
= dm_block_data(sblock
);
466 disk_super
->flags
= 0;
467 memset(disk_super
->uuid
, 0, sizeof(disk_super
->uuid
));
468 disk_super
->magic
= cpu_to_le64(THIN_SUPERBLOCK_MAGIC
);
469 disk_super
->version
= cpu_to_le32(THIN_VERSION
);
470 disk_super
->time
= 0;
471 disk_super
->trans_id
= 0;
472 disk_super
->held_root
= 0;
474 r
= dm_sm_copy_root(pmd
->metadata_sm
, &disk_super
->metadata_space_map_root
,
479 r
= dm_sm_copy_root(pmd
->data_sm
, &disk_super
->data_space_map_root
,
484 disk_super
->data_mapping_root
= cpu_to_le64(pmd
->root
);
485 disk_super
->device_details_root
= cpu_to_le64(pmd
->details_root
);
486 disk_super
->metadata_block_size
= cpu_to_le32(THIN_METADATA_BLOCK_SIZE
>> SECTOR_SHIFT
);
487 disk_super
->metadata_nr_blocks
= cpu_to_le64(bdev_size
>> SECTOR_TO_BLOCK_SHIFT
);
488 disk_super
->data_block_size
= cpu_to_le32(pmd
->data_block_size
);
490 return dm_tm_commit(pmd
->tm
, sblock
);
493 dm_bm_unlock(sblock
);
497 static int __format_metadata(struct dm_pool_metadata
*pmd
)
501 r
= dm_tm_create_with_sm(pmd
->bm
, THIN_SUPERBLOCK_LOCATION
,
502 &pmd
->tm
, &pmd
->metadata_sm
);
504 DMERR("tm_create_with_sm failed");
508 pmd
->data_sm
= dm_sm_disk_create(pmd
->tm
, 0);
509 if (IS_ERR(pmd
->data_sm
)) {
510 DMERR("sm_disk_create failed");
511 r
= PTR_ERR(pmd
->data_sm
);
515 pmd
->nb_tm
= dm_tm_create_non_blocking_clone(pmd
->tm
);
517 DMERR("could not create non-blocking clone tm");
519 goto bad_cleanup_data_sm
;
522 __setup_btree_details(pmd
);
524 r
= dm_btree_empty(&pmd
->info
, &pmd
->root
);
526 goto bad_cleanup_nb_tm
;
528 r
= dm_btree_empty(&pmd
->details_info
, &pmd
->details_root
);
530 DMERR("couldn't create devices root");
531 goto bad_cleanup_nb_tm
;
534 r
= __write_initial_superblock(pmd
);
536 goto bad_cleanup_nb_tm
;
541 dm_tm_destroy(pmd
->nb_tm
);
543 dm_sm_destroy(pmd
->data_sm
);
545 dm_tm_destroy(pmd
->tm
);
546 dm_sm_destroy(pmd
->metadata_sm
);
551 static int __check_incompat_features(struct thin_disk_superblock
*disk_super
,
552 struct dm_pool_metadata
*pmd
)
556 features
= le32_to_cpu(disk_super
->incompat_flags
) & ~THIN_FEATURE_INCOMPAT_SUPP
;
558 DMERR("could not access metadata due to unsupported optional features (%lx).",
559 (unsigned long)features
);
564 * Check for read-only metadata to skip the following RDWR checks.
566 if (get_disk_ro(pmd
->bdev
->bd_disk
))
569 features
= le32_to_cpu(disk_super
->compat_ro_flags
) & ~THIN_FEATURE_COMPAT_RO_SUPP
;
571 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
572 (unsigned long)features
);
579 static int __open_metadata(struct dm_pool_metadata
*pmd
)
582 struct dm_block
*sblock
;
583 struct thin_disk_superblock
*disk_super
;
585 r
= dm_bm_read_lock(pmd
->bm
, THIN_SUPERBLOCK_LOCATION
,
586 &sb_validator
, &sblock
);
588 DMERR("couldn't read superblock");
592 disk_super
= dm_block_data(sblock
);
594 /* Verify the data block size hasn't changed */
595 if (le32_to_cpu(disk_super
->data_block_size
) != pmd
->data_block_size
) {
596 DMERR("changing the data block size (from %u to %llu) is not supported",
597 le32_to_cpu(disk_super
->data_block_size
),
598 (unsigned long long)pmd
->data_block_size
);
600 goto bad_unlock_sblock
;
603 r
= __check_incompat_features(disk_super
, pmd
);
605 goto bad_unlock_sblock
;
607 r
= dm_tm_open_with_sm(pmd
->bm
, THIN_SUPERBLOCK_LOCATION
,
608 disk_super
->metadata_space_map_root
,
609 sizeof(disk_super
->metadata_space_map_root
),
610 &pmd
->tm
, &pmd
->metadata_sm
);
612 DMERR("tm_open_with_sm failed");
613 goto bad_unlock_sblock
;
616 pmd
->data_sm
= dm_sm_disk_open(pmd
->tm
, disk_super
->data_space_map_root
,
617 sizeof(disk_super
->data_space_map_root
));
618 if (IS_ERR(pmd
->data_sm
)) {
619 DMERR("sm_disk_open failed");
620 r
= PTR_ERR(pmd
->data_sm
);
624 pmd
->nb_tm
= dm_tm_create_non_blocking_clone(pmd
->tm
);
626 DMERR("could not create non-blocking clone tm");
628 goto bad_cleanup_data_sm
;
631 __setup_btree_details(pmd
);
632 return dm_bm_unlock(sblock
);
635 dm_sm_destroy(pmd
->data_sm
);
637 dm_tm_destroy(pmd
->tm
);
638 dm_sm_destroy(pmd
->metadata_sm
);
640 dm_bm_unlock(sblock
);
645 static int __open_or_format_metadata(struct dm_pool_metadata
*pmd
, bool format_device
)
649 r
= __superblock_all_zeroes(pmd
->bm
, &unformatted
);
654 return format_device
? __format_metadata(pmd
) : -EPERM
;
656 return __open_metadata(pmd
);
659 static int __create_persistent_data_objects(struct dm_pool_metadata
*pmd
, bool format_device
)
663 pmd
->bm
= dm_block_manager_create(pmd
->bdev
, THIN_METADATA_BLOCK_SIZE
,
664 THIN_METADATA_CACHE_SIZE
,
665 THIN_MAX_CONCURRENT_LOCKS
);
666 if (IS_ERR(pmd
->bm
)) {
667 DMERR("could not create block manager");
668 return PTR_ERR(pmd
->bm
);
671 r
= __open_or_format_metadata(pmd
, format_device
);
673 dm_block_manager_destroy(pmd
->bm
);
678 static void __destroy_persistent_data_objects(struct dm_pool_metadata
*pmd
)
680 dm_sm_destroy(pmd
->data_sm
);
681 dm_sm_destroy(pmd
->metadata_sm
);
682 dm_tm_destroy(pmd
->nb_tm
);
683 dm_tm_destroy(pmd
->tm
);
684 dm_block_manager_destroy(pmd
->bm
);
687 static int __begin_transaction(struct dm_pool_metadata
*pmd
)
690 struct thin_disk_superblock
*disk_super
;
691 struct dm_block
*sblock
;
694 * We re-read the superblock every time. Shouldn't need to do this
697 r
= dm_bm_read_lock(pmd
->bm
, THIN_SUPERBLOCK_LOCATION
,
698 &sb_validator
, &sblock
);
702 disk_super
= dm_block_data(sblock
);
703 pmd
->time
= le32_to_cpu(disk_super
->time
);
704 pmd
->root
= le64_to_cpu(disk_super
->data_mapping_root
);
705 pmd
->details_root
= le64_to_cpu(disk_super
->device_details_root
);
706 pmd
->trans_id
= le64_to_cpu(disk_super
->trans_id
);
707 pmd
->flags
= le32_to_cpu(disk_super
->flags
);
708 pmd
->data_block_size
= le32_to_cpu(disk_super
->data_block_size
);
710 dm_bm_unlock(sblock
);
714 static int __write_changed_details(struct dm_pool_metadata
*pmd
)
717 struct dm_thin_device
*td
, *tmp
;
718 struct disk_device_details details
;
721 list_for_each_entry_safe(td
, tmp
, &pmd
->thin_devices
, list
) {
727 details
.mapped_blocks
= cpu_to_le64(td
->mapped_blocks
);
728 details
.transaction_id
= cpu_to_le64(td
->transaction_id
);
729 details
.creation_time
= cpu_to_le32(td
->creation_time
);
730 details
.snapshotted_time
= cpu_to_le32(td
->snapshotted_time
);
731 __dm_bless_for_disk(&details
);
733 r
= dm_btree_insert(&pmd
->details_info
, pmd
->details_root
,
734 &key
, &details
, &pmd
->details_root
);
749 static int __commit_transaction(struct dm_pool_metadata
*pmd
)
752 size_t metadata_len
, data_len
;
753 struct thin_disk_superblock
*disk_super
;
754 struct dm_block
*sblock
;
757 * We need to know if the thin_disk_superblock exceeds a 512-byte sector.
759 BUILD_BUG_ON(sizeof(struct thin_disk_superblock
) > 512);
761 r
= __write_changed_details(pmd
);
765 r
= dm_sm_commit(pmd
->data_sm
);
769 r
= dm_tm_pre_commit(pmd
->tm
);
773 r
= dm_sm_root_size(pmd
->metadata_sm
, &metadata_len
);
777 r
= dm_sm_root_size(pmd
->data_sm
, &data_len
);
781 r
= superblock_lock(pmd
, &sblock
);
785 disk_super
= dm_block_data(sblock
);
786 disk_super
->time
= cpu_to_le32(pmd
->time
);
787 disk_super
->data_mapping_root
= cpu_to_le64(pmd
->root
);
788 disk_super
->device_details_root
= cpu_to_le64(pmd
->details_root
);
789 disk_super
->trans_id
= cpu_to_le64(pmd
->trans_id
);
790 disk_super
->flags
= cpu_to_le32(pmd
->flags
);
792 r
= dm_sm_copy_root(pmd
->metadata_sm
, &disk_super
->metadata_space_map_root
,
797 r
= dm_sm_copy_root(pmd
->data_sm
, &disk_super
->data_space_map_root
,
802 return dm_tm_commit(pmd
->tm
, sblock
);
805 dm_bm_unlock(sblock
);
809 struct dm_pool_metadata
*dm_pool_metadata_open(struct block_device
*bdev
,
810 sector_t data_block_size
,
814 struct dm_pool_metadata
*pmd
;
816 pmd
= kmalloc(sizeof(*pmd
), GFP_KERNEL
);
818 DMERR("could not allocate metadata struct");
819 return ERR_PTR(-ENOMEM
);
822 init_rwsem(&pmd
->root_lock
);
824 INIT_LIST_HEAD(&pmd
->thin_devices
);
825 pmd
->read_only
= false;
826 pmd
->fail_io
= false;
828 pmd
->data_block_size
= data_block_size
;
830 r
= __create_persistent_data_objects(pmd
, format_device
);
836 r
= __begin_transaction(pmd
);
838 if (dm_pool_metadata_close(pmd
) < 0)
839 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
846 int dm_pool_metadata_close(struct dm_pool_metadata
*pmd
)
849 unsigned open_devices
= 0;
850 struct dm_thin_device
*td
, *tmp
;
852 down_read(&pmd
->root_lock
);
853 list_for_each_entry_safe(td
, tmp
, &pmd
->thin_devices
, list
) {
861 up_read(&pmd
->root_lock
);
864 DMERR("attempt to close pmd when %u device(s) are still open",
869 if (!pmd
->read_only
&& !pmd
->fail_io
) {
870 r
= __commit_transaction(pmd
);
872 DMWARN("%s: __commit_transaction() failed, error = %d",
877 __destroy_persistent_data_objects(pmd
);
884 * __open_device: Returns @td corresponding to device with id @dev,
885 * creating it if @create is set and incrementing @td->open_count.
886 * On failure, @td is undefined.
888 static int __open_device(struct dm_pool_metadata
*pmd
,
889 dm_thin_id dev
, int create
,
890 struct dm_thin_device
**td
)
893 struct dm_thin_device
*td2
;
895 struct disk_device_details details_le
;
898 * If the device is already open, return it.
900 list_for_each_entry(td2
, &pmd
->thin_devices
, list
)
901 if (td2
->id
== dev
) {
903 * May not create an already-open device.
914 * Check the device exists.
916 r
= dm_btree_lookup(&pmd
->details_info
, pmd
->details_root
,
919 if (r
!= -ENODATA
|| !create
)
926 details_le
.mapped_blocks
= 0;
927 details_le
.transaction_id
= cpu_to_le64(pmd
->trans_id
);
928 details_le
.creation_time
= cpu_to_le32(pmd
->time
);
929 details_le
.snapshotted_time
= cpu_to_le32(pmd
->time
);
932 *td
= kmalloc(sizeof(**td
), GFP_NOIO
);
938 (*td
)->open_count
= 1;
939 (*td
)->changed
= changed
;
940 (*td
)->aborted_with_changes
= false;
941 (*td
)->mapped_blocks
= le64_to_cpu(details_le
.mapped_blocks
);
942 (*td
)->transaction_id
= le64_to_cpu(details_le
.transaction_id
);
943 (*td
)->creation_time
= le32_to_cpu(details_le
.creation_time
);
944 (*td
)->snapshotted_time
= le32_to_cpu(details_le
.snapshotted_time
);
946 list_add(&(*td
)->list
, &pmd
->thin_devices
);
951 static void __close_device(struct dm_thin_device
*td
)
956 static int __create_thin(struct dm_pool_metadata
*pmd
,
962 struct disk_device_details details_le
;
963 struct dm_thin_device
*td
;
966 r
= dm_btree_lookup(&pmd
->details_info
, pmd
->details_root
,
972 * Create an empty btree for the mappings.
974 r
= dm_btree_empty(&pmd
->bl_info
, &dev_root
);
979 * Insert it into the main mapping tree.
981 value
= cpu_to_le64(dev_root
);
982 __dm_bless_for_disk(&value
);
983 r
= dm_btree_insert(&pmd
->tl_info
, pmd
->root
, &key
, &value
, &pmd
->root
);
985 dm_btree_del(&pmd
->bl_info
, dev_root
);
989 r
= __open_device(pmd
, dev
, 1, &td
);
991 dm_btree_remove(&pmd
->tl_info
, pmd
->root
, &key
, &pmd
->root
);
992 dm_btree_del(&pmd
->bl_info
, dev_root
);
1000 int dm_pool_create_thin(struct dm_pool_metadata
*pmd
, dm_thin_id dev
)
1004 down_write(&pmd
->root_lock
);
1006 r
= __create_thin(pmd
, dev
);
1007 up_write(&pmd
->root_lock
);
1012 static int __set_snapshot_details(struct dm_pool_metadata
*pmd
,
1013 struct dm_thin_device
*snap
,
1014 dm_thin_id origin
, uint32_t time
)
1017 struct dm_thin_device
*td
;
1019 r
= __open_device(pmd
, origin
, 0, &td
);
1024 td
->snapshotted_time
= time
;
1026 snap
->mapped_blocks
= td
->mapped_blocks
;
1027 snap
->snapshotted_time
= time
;
1033 static int __create_snap(struct dm_pool_metadata
*pmd
,
1034 dm_thin_id dev
, dm_thin_id origin
)
1037 dm_block_t origin_root
;
1038 uint64_t key
= origin
, dev_key
= dev
;
1039 struct dm_thin_device
*td
;
1040 struct disk_device_details details_le
;
1043 /* check this device is unused */
1044 r
= dm_btree_lookup(&pmd
->details_info
, pmd
->details_root
,
1045 &dev_key
, &details_le
);
1049 /* find the mapping tree for the origin */
1050 r
= dm_btree_lookup(&pmd
->tl_info
, pmd
->root
, &key
, &value
);
1053 origin_root
= le64_to_cpu(value
);
1055 /* clone the origin, an inc will do */
1056 dm_tm_inc(pmd
->tm
, origin_root
);
1058 /* insert into the main mapping tree */
1059 value
= cpu_to_le64(origin_root
);
1060 __dm_bless_for_disk(&value
);
1062 r
= dm_btree_insert(&pmd
->tl_info
, pmd
->root
, &key
, &value
, &pmd
->root
);
1064 dm_tm_dec(pmd
->tm
, origin_root
);
1070 r
= __open_device(pmd
, dev
, 1, &td
);
1074 r
= __set_snapshot_details(pmd
, td
, origin
, pmd
->time
);
1083 dm_btree_remove(&pmd
->tl_info
, pmd
->root
, &key
, &pmd
->root
);
1084 dm_btree_remove(&pmd
->details_info
, pmd
->details_root
,
1085 &key
, &pmd
->details_root
);
1089 int dm_pool_create_snap(struct dm_pool_metadata
*pmd
,
1095 down_write(&pmd
->root_lock
);
1097 r
= __create_snap(pmd
, dev
, origin
);
1098 up_write(&pmd
->root_lock
);
1103 static int __delete_device(struct dm_pool_metadata
*pmd
, dm_thin_id dev
)
1107 struct dm_thin_device
*td
;
1109 /* TODO: failure should mark the transaction invalid */
1110 r
= __open_device(pmd
, dev
, 0, &td
);
1114 if (td
->open_count
> 1) {
1119 list_del(&td
->list
);
1121 r
= dm_btree_remove(&pmd
->details_info
, pmd
->details_root
,
1122 &key
, &pmd
->details_root
);
1126 r
= dm_btree_remove(&pmd
->tl_info
, pmd
->root
, &key
, &pmd
->root
);
1133 int dm_pool_delete_thin_device(struct dm_pool_metadata
*pmd
,
1138 down_write(&pmd
->root_lock
);
1140 r
= __delete_device(pmd
, dev
);
1141 up_write(&pmd
->root_lock
);
1146 int dm_pool_set_metadata_transaction_id(struct dm_pool_metadata
*pmd
,
1147 uint64_t current_id
,
1152 down_write(&pmd
->root_lock
);
1157 if (pmd
->trans_id
!= current_id
) {
1158 DMERR("mismatched transaction id");
1162 pmd
->trans_id
= new_id
;
1166 up_write(&pmd
->root_lock
);
1171 int dm_pool_get_metadata_transaction_id(struct dm_pool_metadata
*pmd
,
1176 down_read(&pmd
->root_lock
);
1177 if (!pmd
->fail_io
) {
1178 *result
= pmd
->trans_id
;
1181 up_read(&pmd
->root_lock
);
1186 static int __reserve_metadata_snap(struct dm_pool_metadata
*pmd
)
1189 struct thin_disk_superblock
*disk_super
;
1190 struct dm_block
*copy
, *sblock
;
1191 dm_block_t held_root
;
1194 * Copy the superblock.
1196 dm_sm_inc_block(pmd
->metadata_sm
, THIN_SUPERBLOCK_LOCATION
);
1197 r
= dm_tm_shadow_block(pmd
->tm
, THIN_SUPERBLOCK_LOCATION
,
1198 &sb_validator
, ©
, &inc
);
1204 held_root
= dm_block_location(copy
);
1205 disk_super
= dm_block_data(copy
);
1207 if (le64_to_cpu(disk_super
->held_root
)) {
1208 DMWARN("Pool metadata snapshot already exists: release this before taking another.");
1210 dm_tm_dec(pmd
->tm
, held_root
);
1211 dm_tm_unlock(pmd
->tm
, copy
);
1216 * Wipe the spacemap since we're not publishing this.
1218 memset(&disk_super
->data_space_map_root
, 0,
1219 sizeof(disk_super
->data_space_map_root
));
1220 memset(&disk_super
->metadata_space_map_root
, 0,
1221 sizeof(disk_super
->metadata_space_map_root
));
1224 * Increment the data structures that need to be preserved.
1226 dm_tm_inc(pmd
->tm
, le64_to_cpu(disk_super
->data_mapping_root
));
1227 dm_tm_inc(pmd
->tm
, le64_to_cpu(disk_super
->device_details_root
));
1228 dm_tm_unlock(pmd
->tm
, copy
);
1231 * Write the held root into the superblock.
1233 r
= superblock_lock(pmd
, &sblock
);
1235 dm_tm_dec(pmd
->tm
, held_root
);
1239 disk_super
= dm_block_data(sblock
);
1240 disk_super
->held_root
= cpu_to_le64(held_root
);
1241 dm_bm_unlock(sblock
);
1245 int dm_pool_reserve_metadata_snap(struct dm_pool_metadata
*pmd
)
1249 down_write(&pmd
->root_lock
);
1251 r
= __reserve_metadata_snap(pmd
);
1252 up_write(&pmd
->root_lock
);
1257 static int __release_metadata_snap(struct dm_pool_metadata
*pmd
)
1260 struct thin_disk_superblock
*disk_super
;
1261 struct dm_block
*sblock
, *copy
;
1262 dm_block_t held_root
;
1264 r
= superblock_lock(pmd
, &sblock
);
1268 disk_super
= dm_block_data(sblock
);
1269 held_root
= le64_to_cpu(disk_super
->held_root
);
1270 disk_super
->held_root
= cpu_to_le64(0);
1272 dm_bm_unlock(sblock
);
1275 DMWARN("No pool metadata snapshot found: nothing to release.");
1279 r
= dm_tm_read_lock(pmd
->tm
, held_root
, &sb_validator
, ©
);
1283 disk_super
= dm_block_data(copy
);
1284 dm_sm_dec_block(pmd
->metadata_sm
, le64_to_cpu(disk_super
->data_mapping_root
));
1285 dm_sm_dec_block(pmd
->metadata_sm
, le64_to_cpu(disk_super
->device_details_root
));
1286 dm_sm_dec_block(pmd
->metadata_sm
, held_root
);
1288 return dm_tm_unlock(pmd
->tm
, copy
);
1291 int dm_pool_release_metadata_snap(struct dm_pool_metadata
*pmd
)
1295 down_write(&pmd
->root_lock
);
1297 r
= __release_metadata_snap(pmd
);
1298 up_write(&pmd
->root_lock
);
1303 static int __get_metadata_snap(struct dm_pool_metadata
*pmd
,
1307 struct thin_disk_superblock
*disk_super
;
1308 struct dm_block
*sblock
;
1310 r
= dm_bm_read_lock(pmd
->bm
, THIN_SUPERBLOCK_LOCATION
,
1311 &sb_validator
, &sblock
);
1315 disk_super
= dm_block_data(sblock
);
1316 *result
= le64_to_cpu(disk_super
->held_root
);
1318 return dm_bm_unlock(sblock
);
1321 int dm_pool_get_metadata_snap(struct dm_pool_metadata
*pmd
,
1326 down_read(&pmd
->root_lock
);
1328 r
= __get_metadata_snap(pmd
, result
);
1329 up_read(&pmd
->root_lock
);
1334 int dm_pool_open_thin_device(struct dm_pool_metadata
*pmd
, dm_thin_id dev
,
1335 struct dm_thin_device
**td
)
1339 down_write(&pmd
->root_lock
);
1341 r
= __open_device(pmd
, dev
, 0, td
);
1342 up_write(&pmd
->root_lock
);
1347 int dm_pool_close_thin_device(struct dm_thin_device
*td
)
1349 down_write(&td
->pmd
->root_lock
);
1351 up_write(&td
->pmd
->root_lock
);
1356 dm_thin_id
dm_thin_dev_id(struct dm_thin_device
*td
)
1362 * Check whether @time (of block creation) is older than @td's last snapshot.
1363 * If so then the associated block is shared with the last snapshot device.
1364 * Any block on a device created *after* the device last got snapshotted is
1365 * necessarily not shared.
1367 static bool __snapshotted_since(struct dm_thin_device
*td
, uint32_t time
)
1369 return td
->snapshotted_time
> time
;
1372 int dm_thin_find_block(struct dm_thin_device
*td
, dm_block_t block
,
1373 int can_block
, struct dm_thin_lookup_result
*result
)
1376 uint64_t block_time
= 0;
1378 struct dm_pool_metadata
*pmd
= td
->pmd
;
1379 dm_block_t keys
[2] = { td
->id
, block
};
1380 struct dm_btree_info
*info
;
1383 down_read(&pmd
->root_lock
);
1385 } else if (down_read_trylock(&pmd
->root_lock
))
1386 info
= &pmd
->nb_info
;
1388 return -EWOULDBLOCK
;
1393 r
= dm_btree_lookup(info
, pmd
->root
, keys
, &value
);
1395 block_time
= le64_to_cpu(value
);
1398 up_read(&pmd
->root_lock
);
1401 dm_block_t exception_block
;
1402 uint32_t exception_time
;
1403 unpack_block_time(block_time
, &exception_block
,
1405 result
->block
= exception_block
;
1406 result
->shared
= __snapshotted_since(td
, exception_time
);
1412 static int __insert(struct dm_thin_device
*td
, dm_block_t block
,
1413 dm_block_t data_block
)
1417 struct dm_pool_metadata
*pmd
= td
->pmd
;
1418 dm_block_t keys
[2] = { td
->id
, block
};
1420 value
= cpu_to_le64(pack_block_time(data_block
, pmd
->time
));
1421 __dm_bless_for_disk(&value
);
1423 r
= dm_btree_insert_notify(&pmd
->info
, pmd
->root
, keys
, &value
,
1424 &pmd
->root
, &inserted
);
1430 td
->mapped_blocks
++;
1435 int dm_thin_insert_block(struct dm_thin_device
*td
, dm_block_t block
,
1436 dm_block_t data_block
)
1440 down_write(&td
->pmd
->root_lock
);
1441 if (!td
->pmd
->fail_io
)
1442 r
= __insert(td
, block
, data_block
);
1443 up_write(&td
->pmd
->root_lock
);
1448 static int __remove(struct dm_thin_device
*td
, dm_block_t block
)
1451 struct dm_pool_metadata
*pmd
= td
->pmd
;
1452 dm_block_t keys
[2] = { td
->id
, block
};
1454 r
= dm_btree_remove(&pmd
->info
, pmd
->root
, keys
, &pmd
->root
);
1458 td
->mapped_blocks
--;
1464 int dm_thin_remove_block(struct dm_thin_device
*td
, dm_block_t block
)
1468 down_write(&td
->pmd
->root_lock
);
1469 if (!td
->pmd
->fail_io
)
1470 r
= __remove(td
, block
);
1471 up_write(&td
->pmd
->root_lock
);
1476 int dm_pool_block_is_used(struct dm_pool_metadata
*pmd
, dm_block_t b
, bool *result
)
1481 down_read(&pmd
->root_lock
);
1482 r
= dm_sm_get_count(pmd
->data_sm
, b
, &ref_count
);
1484 *result
= (ref_count
!= 0);
1485 up_read(&pmd
->root_lock
);
1490 bool dm_thin_changed_this_transaction(struct dm_thin_device
*td
)
1494 down_read(&td
->pmd
->root_lock
);
1496 up_read(&td
->pmd
->root_lock
);
1501 bool dm_pool_changed_this_transaction(struct dm_pool_metadata
*pmd
)
1504 struct dm_thin_device
*td
, *tmp
;
1506 down_read(&pmd
->root_lock
);
1507 list_for_each_entry_safe(td
, tmp
, &pmd
->thin_devices
, list
) {
1513 up_read(&pmd
->root_lock
);
1518 bool dm_thin_aborted_changes(struct dm_thin_device
*td
)
1522 down_read(&td
->pmd
->root_lock
);
1523 r
= td
->aborted_with_changes
;
1524 up_read(&td
->pmd
->root_lock
);
1529 int dm_pool_alloc_data_block(struct dm_pool_metadata
*pmd
, dm_block_t
*result
)
1533 down_write(&pmd
->root_lock
);
1535 r
= dm_sm_new_block(pmd
->data_sm
, result
);
1536 up_write(&pmd
->root_lock
);
1541 int dm_pool_commit_metadata(struct dm_pool_metadata
*pmd
)
1545 down_write(&pmd
->root_lock
);
1549 r
= __commit_transaction(pmd
);
1554 * Open the next transaction.
1556 r
= __begin_transaction(pmd
);
1558 up_write(&pmd
->root_lock
);
1562 static void __set_abort_with_changes_flags(struct dm_pool_metadata
*pmd
)
1564 struct dm_thin_device
*td
;
1566 list_for_each_entry(td
, &pmd
->thin_devices
, list
)
1567 td
->aborted_with_changes
= td
->changed
;
1570 int dm_pool_abort_metadata(struct dm_pool_metadata
*pmd
)
1574 down_write(&pmd
->root_lock
);
1578 __set_abort_with_changes_flags(pmd
);
1579 __destroy_persistent_data_objects(pmd
);
1580 r
= __create_persistent_data_objects(pmd
, false);
1582 pmd
->fail_io
= true;
1585 up_write(&pmd
->root_lock
);
1590 int dm_pool_get_free_block_count(struct dm_pool_metadata
*pmd
, dm_block_t
*result
)
1594 down_read(&pmd
->root_lock
);
1596 r
= dm_sm_get_nr_free(pmd
->data_sm
, result
);
1597 up_read(&pmd
->root_lock
);
1602 int dm_pool_get_free_metadata_block_count(struct dm_pool_metadata
*pmd
,
1607 down_read(&pmd
->root_lock
);
1609 r
= dm_sm_get_nr_free(pmd
->metadata_sm
, result
);
1610 up_read(&pmd
->root_lock
);
1615 int dm_pool_get_metadata_dev_size(struct dm_pool_metadata
*pmd
,
1620 down_read(&pmd
->root_lock
);
1622 r
= dm_sm_get_nr_blocks(pmd
->metadata_sm
, result
);
1623 up_read(&pmd
->root_lock
);
1628 int dm_pool_get_data_block_size(struct dm_pool_metadata
*pmd
, sector_t
*result
)
1630 down_read(&pmd
->root_lock
);
1631 *result
= pmd
->data_block_size
;
1632 up_read(&pmd
->root_lock
);
1637 int dm_pool_get_data_dev_size(struct dm_pool_metadata
*pmd
, dm_block_t
*result
)
1641 down_read(&pmd
->root_lock
);
1643 r
= dm_sm_get_nr_blocks(pmd
->data_sm
, result
);
1644 up_read(&pmd
->root_lock
);
1649 int dm_thin_get_mapped_count(struct dm_thin_device
*td
, dm_block_t
*result
)
1652 struct dm_pool_metadata
*pmd
= td
->pmd
;
1654 down_read(&pmd
->root_lock
);
1655 if (!pmd
->fail_io
) {
1656 *result
= td
->mapped_blocks
;
1659 up_read(&pmd
->root_lock
);
1664 static int __highest_block(struct dm_thin_device
*td
, dm_block_t
*result
)
1668 dm_block_t thin_root
;
1669 struct dm_pool_metadata
*pmd
= td
->pmd
;
1671 r
= dm_btree_lookup(&pmd
->tl_info
, pmd
->root
, &td
->id
, &value_le
);
1675 thin_root
= le64_to_cpu(value_le
);
1677 return dm_btree_find_highest_key(&pmd
->bl_info
, thin_root
, result
);
1680 int dm_thin_get_highest_mapped_block(struct dm_thin_device
*td
,
1684 struct dm_pool_metadata
*pmd
= td
->pmd
;
1686 down_read(&pmd
->root_lock
);
1688 r
= __highest_block(td
, result
);
1689 up_read(&pmd
->root_lock
);
1694 static int __resize_space_map(struct dm_space_map
*sm
, dm_block_t new_count
)
1697 dm_block_t old_count
;
1699 r
= dm_sm_get_nr_blocks(sm
, &old_count
);
1703 if (new_count
== old_count
)
1706 if (new_count
< old_count
) {
1707 DMERR("cannot reduce size of space map");
1711 return dm_sm_extend(sm
, new_count
- old_count
);
1714 int dm_pool_resize_data_dev(struct dm_pool_metadata
*pmd
, dm_block_t new_count
)
1718 down_write(&pmd
->root_lock
);
1720 r
= __resize_space_map(pmd
->data_sm
, new_count
);
1721 up_write(&pmd
->root_lock
);
1726 int dm_pool_resize_metadata_dev(struct dm_pool_metadata
*pmd
, dm_block_t new_count
)
1730 down_write(&pmd
->root_lock
);
1732 r
= __resize_space_map(pmd
->metadata_sm
, new_count
);
1733 up_write(&pmd
->root_lock
);
1738 void dm_pool_metadata_read_only(struct dm_pool_metadata
*pmd
)
1740 down_write(&pmd
->root_lock
);
1741 pmd
->read_only
= true;
1742 dm_bm_set_read_only(pmd
->bm
);
1743 up_write(&pmd
->root_lock
);
1746 int dm_pool_register_metadata_threshold(struct dm_pool_metadata
*pmd
,
1747 dm_block_t threshold
,
1748 dm_sm_threshold_fn fn
,
1753 down_write(&pmd
->root_lock
);
1754 r
= dm_sm_register_threshold_callback(pmd
->metadata_sm
, threshold
, fn
, context
);
1755 up_write(&pmd
->root_lock
);