2 * Copyright (C) 2011 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
;
189 struct dm_thin_device
{
190 struct list_head list
;
191 struct dm_pool_metadata
*pmd
;
196 uint64_t mapped_blocks
;
197 uint64_t transaction_id
;
198 uint32_t creation_time
;
199 uint32_t snapshotted_time
;
202 /*----------------------------------------------------------------
203 * superblock validator
204 *--------------------------------------------------------------*/
206 #define SUPERBLOCK_CSUM_XOR 160774
208 static void sb_prepare_for_write(struct dm_block_validator
*v
,
212 struct thin_disk_superblock
*disk_super
= dm_block_data(b
);
214 disk_super
->blocknr
= cpu_to_le64(dm_block_location(b
));
215 disk_super
->csum
= cpu_to_le32(dm_bm_checksum(&disk_super
->flags
,
216 block_size
- sizeof(__le32
),
217 SUPERBLOCK_CSUM_XOR
));
220 static int sb_check(struct dm_block_validator
*v
,
224 struct thin_disk_superblock
*disk_super
= dm_block_data(b
);
227 if (dm_block_location(b
) != le64_to_cpu(disk_super
->blocknr
)) {
228 DMERR("sb_check failed: blocknr %llu: "
229 "wanted %llu", le64_to_cpu(disk_super
->blocknr
),
230 (unsigned long long)dm_block_location(b
));
234 if (le64_to_cpu(disk_super
->magic
) != THIN_SUPERBLOCK_MAGIC
) {
235 DMERR("sb_check failed: magic %llu: "
236 "wanted %llu", le64_to_cpu(disk_super
->magic
),
237 (unsigned long long)THIN_SUPERBLOCK_MAGIC
);
241 csum_le
= cpu_to_le32(dm_bm_checksum(&disk_super
->flags
,
242 block_size
- sizeof(__le32
),
243 SUPERBLOCK_CSUM_XOR
));
244 if (csum_le
!= disk_super
->csum
) {
245 DMERR("sb_check failed: csum %u: wanted %u",
246 le32_to_cpu(csum_le
), le32_to_cpu(disk_super
->csum
));
253 static struct dm_block_validator sb_validator
= {
254 .name
= "superblock",
255 .prepare_for_write
= sb_prepare_for_write
,
259 /*----------------------------------------------------------------
260 * Methods for the btree value types
261 *--------------------------------------------------------------*/
263 static uint64_t pack_block_time(dm_block_t b
, uint32_t t
)
265 return (b
<< 24) | t
;
268 static void unpack_block_time(uint64_t v
, dm_block_t
*b
, uint32_t *t
)
271 *t
= v
& ((1 << 24) - 1);
274 static void data_block_inc(void *context
, void *value_le
)
276 struct dm_space_map
*sm
= context
;
281 memcpy(&v_le
, value_le
, sizeof(v_le
));
282 unpack_block_time(le64_to_cpu(v_le
), &b
, &t
);
283 dm_sm_inc_block(sm
, b
);
286 static void data_block_dec(void *context
, void *value_le
)
288 struct dm_space_map
*sm
= context
;
293 memcpy(&v_le
, value_le
, sizeof(v_le
));
294 unpack_block_time(le64_to_cpu(v_le
), &b
, &t
);
295 dm_sm_dec_block(sm
, b
);
298 static int data_block_equal(void *context
, void *value1_le
, void *value2_le
)
304 memcpy(&v1_le
, value1_le
, sizeof(v1_le
));
305 memcpy(&v2_le
, value2_le
, sizeof(v2_le
));
306 unpack_block_time(le64_to_cpu(v1_le
), &b1
, &t
);
307 unpack_block_time(le64_to_cpu(v2_le
), &b2
, &t
);
312 static void subtree_inc(void *context
, void *value
)
314 struct dm_btree_info
*info
= context
;
318 memcpy(&root_le
, value
, sizeof(root_le
));
319 root
= le64_to_cpu(root_le
);
320 dm_tm_inc(info
->tm
, root
);
323 static void subtree_dec(void *context
, void *value
)
325 struct dm_btree_info
*info
= context
;
329 memcpy(&root_le
, value
, sizeof(root_le
));
330 root
= le64_to_cpu(root_le
);
331 if (dm_btree_del(info
, root
))
332 DMERR("btree delete failed\n");
335 static int subtree_equal(void *context
, void *value1_le
, void *value2_le
)
338 memcpy(&v1_le
, value1_le
, sizeof(v1_le
));
339 memcpy(&v2_le
, value2_le
, sizeof(v2_le
));
341 return v1_le
== v2_le
;
344 /*----------------------------------------------------------------*/
346 static int superblock_lock_zero(struct dm_pool_metadata
*pmd
,
347 struct dm_block
**sblock
)
349 return dm_bm_write_lock_zero(pmd
->bm
, THIN_SUPERBLOCK_LOCATION
,
350 &sb_validator
, sblock
);
353 static int superblock_lock(struct dm_pool_metadata
*pmd
,
354 struct dm_block
**sblock
)
356 return dm_bm_write_lock(pmd
->bm
, THIN_SUPERBLOCK_LOCATION
,
357 &sb_validator
, sblock
);
360 static int __superblock_all_zeroes(struct dm_block_manager
*bm
, int *result
)
365 __le64
*data_le
, zero
= cpu_to_le64(0);
366 unsigned block_size
= dm_bm_block_size(bm
) / sizeof(__le64
);
369 * We can't use a validator here - it may be all zeroes.
371 r
= dm_bm_read_lock(bm
, THIN_SUPERBLOCK_LOCATION
, NULL
, &b
);
375 data_le
= dm_block_data(b
);
377 for (i
= 0; i
< block_size
; i
++) {
378 if (data_le
[i
] != zero
) {
384 return dm_bm_unlock(b
);
387 static void __setup_btree_details(struct dm_pool_metadata
*pmd
)
389 pmd
->info
.tm
= pmd
->tm
;
390 pmd
->info
.levels
= 2;
391 pmd
->info
.value_type
.context
= pmd
->data_sm
;
392 pmd
->info
.value_type
.size
= sizeof(__le64
);
393 pmd
->info
.value_type
.inc
= data_block_inc
;
394 pmd
->info
.value_type
.dec
= data_block_dec
;
395 pmd
->info
.value_type
.equal
= data_block_equal
;
397 memcpy(&pmd
->nb_info
, &pmd
->info
, sizeof(pmd
->nb_info
));
398 pmd
->nb_info
.tm
= pmd
->nb_tm
;
400 pmd
->tl_info
.tm
= pmd
->tm
;
401 pmd
->tl_info
.levels
= 1;
402 pmd
->tl_info
.value_type
.context
= &pmd
->info
;
403 pmd
->tl_info
.value_type
.size
= sizeof(__le64
);
404 pmd
->tl_info
.value_type
.inc
= subtree_inc
;
405 pmd
->tl_info
.value_type
.dec
= subtree_dec
;
406 pmd
->tl_info
.value_type
.equal
= subtree_equal
;
408 pmd
->bl_info
.tm
= pmd
->tm
;
409 pmd
->bl_info
.levels
= 1;
410 pmd
->bl_info
.value_type
.context
= pmd
->data_sm
;
411 pmd
->bl_info
.value_type
.size
= sizeof(__le64
);
412 pmd
->bl_info
.value_type
.inc
= data_block_inc
;
413 pmd
->bl_info
.value_type
.dec
= data_block_dec
;
414 pmd
->bl_info
.value_type
.equal
= data_block_equal
;
416 pmd
->details_info
.tm
= pmd
->tm
;
417 pmd
->details_info
.levels
= 1;
418 pmd
->details_info
.value_type
.context
= NULL
;
419 pmd
->details_info
.value_type
.size
= sizeof(struct disk_device_details
);
420 pmd
->details_info
.value_type
.inc
= NULL
;
421 pmd
->details_info
.value_type
.dec
= NULL
;
422 pmd
->details_info
.value_type
.equal
= NULL
;
425 static int __write_initial_superblock(struct dm_pool_metadata
*pmd
)
428 struct dm_block
*sblock
;
429 size_t metadata_len
, data_len
;
430 struct thin_disk_superblock
*disk_super
;
431 sector_t bdev_size
= i_size_read(pmd
->bdev
->bd_inode
) >> SECTOR_SHIFT
;
433 if (bdev_size
> THIN_METADATA_MAX_SECTORS
)
434 bdev_size
= THIN_METADATA_MAX_SECTORS
;
436 r
= dm_sm_root_size(pmd
->metadata_sm
, &metadata_len
);
440 r
= dm_sm_root_size(pmd
->data_sm
, &data_len
);
444 r
= dm_sm_commit(pmd
->data_sm
);
448 r
= dm_tm_pre_commit(pmd
->tm
);
452 r
= superblock_lock_zero(pmd
, &sblock
);
456 disk_super
= dm_block_data(sblock
);
457 disk_super
->flags
= 0;
458 memset(disk_super
->uuid
, 0, sizeof(disk_super
->uuid
));
459 disk_super
->magic
= cpu_to_le64(THIN_SUPERBLOCK_MAGIC
);
460 disk_super
->version
= cpu_to_le32(THIN_VERSION
);
461 disk_super
->time
= 0;
462 disk_super
->trans_id
= 0;
463 disk_super
->held_root
= 0;
465 r
= dm_sm_copy_root(pmd
->metadata_sm
, &disk_super
->metadata_space_map_root
,
470 r
= dm_sm_copy_root(pmd
->data_sm
, &disk_super
->data_space_map_root
,
475 disk_super
->data_mapping_root
= cpu_to_le64(pmd
->root
);
476 disk_super
->device_details_root
= cpu_to_le64(pmd
->details_root
);
477 disk_super
->metadata_block_size
= cpu_to_le32(THIN_METADATA_BLOCK_SIZE
>> SECTOR_SHIFT
);
478 disk_super
->metadata_nr_blocks
= cpu_to_le64(bdev_size
>> SECTOR_TO_BLOCK_SHIFT
);
479 disk_super
->data_block_size
= cpu_to_le32(pmd
->data_block_size
);
481 return dm_tm_commit(pmd
->tm
, sblock
);
484 dm_bm_unlock(sblock
);
488 static int __format_metadata(struct dm_pool_metadata
*pmd
)
492 r
= dm_tm_create_with_sm(pmd
->bm
, THIN_SUPERBLOCK_LOCATION
,
493 &pmd
->tm
, &pmd
->metadata_sm
);
495 DMERR("tm_create_with_sm failed");
499 pmd
->data_sm
= dm_sm_disk_create(pmd
->tm
, 0);
500 if (IS_ERR(pmd
->data_sm
)) {
501 DMERR("sm_disk_create failed");
502 r
= PTR_ERR(pmd
->data_sm
);
506 pmd
->nb_tm
= dm_tm_create_non_blocking_clone(pmd
->tm
);
508 DMERR("could not create non-blocking clone tm");
510 goto bad_cleanup_data_sm
;
513 __setup_btree_details(pmd
);
515 r
= dm_btree_empty(&pmd
->info
, &pmd
->root
);
517 goto bad_cleanup_nb_tm
;
519 r
= dm_btree_empty(&pmd
->details_info
, &pmd
->details_root
);
521 DMERR("couldn't create devices root");
522 goto bad_cleanup_nb_tm
;
525 r
= __write_initial_superblock(pmd
);
527 goto bad_cleanup_nb_tm
;
532 dm_tm_destroy(pmd
->nb_tm
);
534 dm_sm_destroy(pmd
->data_sm
);
536 dm_tm_destroy(pmd
->tm
);
537 dm_sm_destroy(pmd
->metadata_sm
);
542 static int __check_incompat_features(struct thin_disk_superblock
*disk_super
,
543 struct dm_pool_metadata
*pmd
)
547 features
= le32_to_cpu(disk_super
->incompat_flags
) & ~THIN_FEATURE_INCOMPAT_SUPP
;
549 DMERR("could not access metadata due to unsupported optional features (%lx).",
550 (unsigned long)features
);
555 * Check for read-only metadata to skip the following RDWR checks.
557 if (get_disk_ro(pmd
->bdev
->bd_disk
))
560 features
= le32_to_cpu(disk_super
->compat_ro_flags
) & ~THIN_FEATURE_COMPAT_RO_SUPP
;
562 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
563 (unsigned long)features
);
570 static int __open_metadata(struct dm_pool_metadata
*pmd
)
573 struct dm_block
*sblock
;
574 struct thin_disk_superblock
*disk_super
;
576 r
= dm_bm_read_lock(pmd
->bm
, THIN_SUPERBLOCK_LOCATION
,
577 &sb_validator
, &sblock
);
579 DMERR("couldn't read superblock");
583 disk_super
= dm_block_data(sblock
);
585 r
= __check_incompat_features(disk_super
, pmd
);
587 goto bad_unlock_sblock
;
589 r
= dm_tm_open_with_sm(pmd
->bm
, THIN_SUPERBLOCK_LOCATION
,
590 disk_super
->metadata_space_map_root
,
591 sizeof(disk_super
->metadata_space_map_root
),
592 &pmd
->tm
, &pmd
->metadata_sm
);
594 DMERR("tm_open_with_sm failed");
595 goto bad_unlock_sblock
;
598 pmd
->data_sm
= dm_sm_disk_open(pmd
->tm
, disk_super
->data_space_map_root
,
599 sizeof(disk_super
->data_space_map_root
));
600 if (IS_ERR(pmd
->data_sm
)) {
601 DMERR("sm_disk_open failed");
602 r
= PTR_ERR(pmd
->data_sm
);
606 pmd
->nb_tm
= dm_tm_create_non_blocking_clone(pmd
->tm
);
608 DMERR("could not create non-blocking clone tm");
610 goto bad_cleanup_data_sm
;
613 __setup_btree_details(pmd
);
614 return dm_bm_unlock(sblock
);
617 dm_sm_destroy(pmd
->data_sm
);
619 dm_tm_destroy(pmd
->tm
);
620 dm_sm_destroy(pmd
->metadata_sm
);
622 dm_bm_unlock(sblock
);
627 static int __open_or_format_metadata(struct dm_pool_metadata
*pmd
, bool format_device
)
631 r
= __superblock_all_zeroes(pmd
->bm
, &unformatted
);
636 return format_device
? __format_metadata(pmd
) : -EPERM
;
638 return __open_metadata(pmd
);
641 static int __create_persistent_data_objects(struct dm_pool_metadata
*pmd
, bool format_device
)
645 pmd
->bm
= dm_block_manager_create(pmd
->bdev
, THIN_METADATA_BLOCK_SIZE
,
646 THIN_METADATA_CACHE_SIZE
,
647 THIN_MAX_CONCURRENT_LOCKS
);
648 if (IS_ERR(pmd
->bm
)) {
649 DMERR("could not create block manager");
650 return PTR_ERR(pmd
->bm
);
653 r
= __open_or_format_metadata(pmd
, format_device
);
655 dm_block_manager_destroy(pmd
->bm
);
660 static void __destroy_persistent_data_objects(struct dm_pool_metadata
*pmd
)
662 dm_sm_destroy(pmd
->data_sm
);
663 dm_sm_destroy(pmd
->metadata_sm
);
664 dm_tm_destroy(pmd
->nb_tm
);
665 dm_tm_destroy(pmd
->tm
);
666 dm_block_manager_destroy(pmd
->bm
);
669 static int __begin_transaction(struct dm_pool_metadata
*pmd
)
672 struct thin_disk_superblock
*disk_super
;
673 struct dm_block
*sblock
;
676 * We re-read the superblock every time. Shouldn't need to do this
679 r
= dm_bm_read_lock(pmd
->bm
, THIN_SUPERBLOCK_LOCATION
,
680 &sb_validator
, &sblock
);
684 disk_super
= dm_block_data(sblock
);
685 pmd
->time
= le32_to_cpu(disk_super
->time
);
686 pmd
->root
= le64_to_cpu(disk_super
->data_mapping_root
);
687 pmd
->details_root
= le64_to_cpu(disk_super
->device_details_root
);
688 pmd
->trans_id
= le64_to_cpu(disk_super
->trans_id
);
689 pmd
->flags
= le32_to_cpu(disk_super
->flags
);
690 pmd
->data_block_size
= le32_to_cpu(disk_super
->data_block_size
);
692 dm_bm_unlock(sblock
);
696 static int __write_changed_details(struct dm_pool_metadata
*pmd
)
699 struct dm_thin_device
*td
, *tmp
;
700 struct disk_device_details details
;
703 list_for_each_entry_safe(td
, tmp
, &pmd
->thin_devices
, list
) {
709 details
.mapped_blocks
= cpu_to_le64(td
->mapped_blocks
);
710 details
.transaction_id
= cpu_to_le64(td
->transaction_id
);
711 details
.creation_time
= cpu_to_le32(td
->creation_time
);
712 details
.snapshotted_time
= cpu_to_le32(td
->snapshotted_time
);
713 __dm_bless_for_disk(&details
);
715 r
= dm_btree_insert(&pmd
->details_info
, pmd
->details_root
,
716 &key
, &details
, &pmd
->details_root
);
731 static int __commit_transaction(struct dm_pool_metadata
*pmd
)
734 * FIXME: Associated pool should be made read-only on failure.
737 size_t metadata_len
, data_len
;
738 struct thin_disk_superblock
*disk_super
;
739 struct dm_block
*sblock
;
742 * We need to know if the thin_disk_superblock exceeds a 512-byte sector.
744 BUILD_BUG_ON(sizeof(struct thin_disk_superblock
) > 512);
746 r
= __write_changed_details(pmd
);
750 r
= dm_sm_commit(pmd
->data_sm
);
754 r
= dm_tm_pre_commit(pmd
->tm
);
758 r
= dm_sm_root_size(pmd
->metadata_sm
, &metadata_len
);
762 r
= dm_sm_root_size(pmd
->data_sm
, &data_len
);
766 r
= superblock_lock(pmd
, &sblock
);
770 disk_super
= dm_block_data(sblock
);
771 disk_super
->time
= cpu_to_le32(pmd
->time
);
772 disk_super
->data_mapping_root
= cpu_to_le64(pmd
->root
);
773 disk_super
->device_details_root
= cpu_to_le64(pmd
->details_root
);
774 disk_super
->trans_id
= cpu_to_le64(pmd
->trans_id
);
775 disk_super
->flags
= cpu_to_le32(pmd
->flags
);
777 r
= dm_sm_copy_root(pmd
->metadata_sm
, &disk_super
->metadata_space_map_root
,
782 r
= dm_sm_copy_root(pmd
->data_sm
, &disk_super
->data_space_map_root
,
787 return dm_tm_commit(pmd
->tm
, sblock
);
790 dm_bm_unlock(sblock
);
794 struct dm_pool_metadata
*dm_pool_metadata_open(struct block_device
*bdev
,
795 sector_t data_block_size
,
799 struct dm_pool_metadata
*pmd
;
801 pmd
= kmalloc(sizeof(*pmd
), GFP_KERNEL
);
803 DMERR("could not allocate metadata struct");
804 return ERR_PTR(-ENOMEM
);
807 init_rwsem(&pmd
->root_lock
);
809 INIT_LIST_HEAD(&pmd
->thin_devices
);
811 pmd
->data_block_size
= data_block_size
;
813 r
= __create_persistent_data_objects(pmd
, format_device
);
819 r
= __begin_transaction(pmd
);
821 if (dm_pool_metadata_close(pmd
) < 0)
822 DMWARN("%s: dm_pool_metadata_close() failed.", __func__
);
829 int dm_pool_metadata_close(struct dm_pool_metadata
*pmd
)
832 unsigned open_devices
= 0;
833 struct dm_thin_device
*td
, *tmp
;
835 down_read(&pmd
->root_lock
);
836 list_for_each_entry_safe(td
, tmp
, &pmd
->thin_devices
, list
) {
844 up_read(&pmd
->root_lock
);
847 DMERR("attempt to close pmd when %u device(s) are still open",
852 r
= __commit_transaction(pmd
);
854 DMWARN("%s: __commit_transaction() failed, error = %d",
857 __destroy_persistent_data_objects(pmd
);
864 * __open_device: Returns @td corresponding to device with id @dev,
865 * creating it if @create is set and incrementing @td->open_count.
866 * On failure, @td is undefined.
868 static int __open_device(struct dm_pool_metadata
*pmd
,
869 dm_thin_id dev
, int create
,
870 struct dm_thin_device
**td
)
873 struct dm_thin_device
*td2
;
875 struct disk_device_details details_le
;
878 * If the device is already open, return it.
880 list_for_each_entry(td2
, &pmd
->thin_devices
, list
)
881 if (td2
->id
== dev
) {
883 * May not create an already-open device.
894 * Check the device exists.
896 r
= dm_btree_lookup(&pmd
->details_info
, pmd
->details_root
,
899 if (r
!= -ENODATA
|| !create
)
906 details_le
.mapped_blocks
= 0;
907 details_le
.transaction_id
= cpu_to_le64(pmd
->trans_id
);
908 details_le
.creation_time
= cpu_to_le32(pmd
->time
);
909 details_le
.snapshotted_time
= cpu_to_le32(pmd
->time
);
912 *td
= kmalloc(sizeof(**td
), GFP_NOIO
);
918 (*td
)->open_count
= 1;
919 (*td
)->changed
= changed
;
920 (*td
)->mapped_blocks
= le64_to_cpu(details_le
.mapped_blocks
);
921 (*td
)->transaction_id
= le64_to_cpu(details_le
.transaction_id
);
922 (*td
)->creation_time
= le32_to_cpu(details_le
.creation_time
);
923 (*td
)->snapshotted_time
= le32_to_cpu(details_le
.snapshotted_time
);
925 list_add(&(*td
)->list
, &pmd
->thin_devices
);
930 static void __close_device(struct dm_thin_device
*td
)
935 static int __create_thin(struct dm_pool_metadata
*pmd
,
941 struct disk_device_details details_le
;
942 struct dm_thin_device
*td
;
945 r
= dm_btree_lookup(&pmd
->details_info
, pmd
->details_root
,
951 * Create an empty btree for the mappings.
953 r
= dm_btree_empty(&pmd
->bl_info
, &dev_root
);
958 * Insert it into the main mapping tree.
960 value
= cpu_to_le64(dev_root
);
961 __dm_bless_for_disk(&value
);
962 r
= dm_btree_insert(&pmd
->tl_info
, pmd
->root
, &key
, &value
, &pmd
->root
);
964 dm_btree_del(&pmd
->bl_info
, dev_root
);
968 r
= __open_device(pmd
, dev
, 1, &td
);
970 dm_btree_remove(&pmd
->tl_info
, pmd
->root
, &key
, &pmd
->root
);
971 dm_btree_del(&pmd
->bl_info
, dev_root
);
979 int dm_pool_create_thin(struct dm_pool_metadata
*pmd
, dm_thin_id dev
)
983 down_write(&pmd
->root_lock
);
984 r
= __create_thin(pmd
, dev
);
985 up_write(&pmd
->root_lock
);
990 static int __set_snapshot_details(struct dm_pool_metadata
*pmd
,
991 struct dm_thin_device
*snap
,
992 dm_thin_id origin
, uint32_t time
)
995 struct dm_thin_device
*td
;
997 r
= __open_device(pmd
, origin
, 0, &td
);
1002 td
->snapshotted_time
= time
;
1004 snap
->mapped_blocks
= td
->mapped_blocks
;
1005 snap
->snapshotted_time
= time
;
1011 static int __create_snap(struct dm_pool_metadata
*pmd
,
1012 dm_thin_id dev
, dm_thin_id origin
)
1015 dm_block_t origin_root
;
1016 uint64_t key
= origin
, dev_key
= dev
;
1017 struct dm_thin_device
*td
;
1018 struct disk_device_details details_le
;
1021 /* check this device is unused */
1022 r
= dm_btree_lookup(&pmd
->details_info
, pmd
->details_root
,
1023 &dev_key
, &details_le
);
1027 /* find the mapping tree for the origin */
1028 r
= dm_btree_lookup(&pmd
->tl_info
, pmd
->root
, &key
, &value
);
1031 origin_root
= le64_to_cpu(value
);
1033 /* clone the origin, an inc will do */
1034 dm_tm_inc(pmd
->tm
, origin_root
);
1036 /* insert into the main mapping tree */
1037 value
= cpu_to_le64(origin_root
);
1038 __dm_bless_for_disk(&value
);
1040 r
= dm_btree_insert(&pmd
->tl_info
, pmd
->root
, &key
, &value
, &pmd
->root
);
1042 dm_tm_dec(pmd
->tm
, origin_root
);
1048 r
= __open_device(pmd
, dev
, 1, &td
);
1052 r
= __set_snapshot_details(pmd
, td
, origin
, pmd
->time
);
1061 dm_btree_remove(&pmd
->tl_info
, pmd
->root
, &key
, &pmd
->root
);
1062 dm_btree_remove(&pmd
->details_info
, pmd
->details_root
,
1063 &key
, &pmd
->details_root
);
1067 int dm_pool_create_snap(struct dm_pool_metadata
*pmd
,
1073 down_write(&pmd
->root_lock
);
1074 r
= __create_snap(pmd
, dev
, origin
);
1075 up_write(&pmd
->root_lock
);
1080 static int __delete_device(struct dm_pool_metadata
*pmd
, dm_thin_id dev
)
1084 struct dm_thin_device
*td
;
1086 /* TODO: failure should mark the transaction invalid */
1087 r
= __open_device(pmd
, dev
, 0, &td
);
1091 if (td
->open_count
> 1) {
1096 list_del(&td
->list
);
1098 r
= dm_btree_remove(&pmd
->details_info
, pmd
->details_root
,
1099 &key
, &pmd
->details_root
);
1103 r
= dm_btree_remove(&pmd
->tl_info
, pmd
->root
, &key
, &pmd
->root
);
1110 int dm_pool_delete_thin_device(struct dm_pool_metadata
*pmd
,
1115 down_write(&pmd
->root_lock
);
1116 r
= __delete_device(pmd
, dev
);
1117 up_write(&pmd
->root_lock
);
1122 int dm_pool_set_metadata_transaction_id(struct dm_pool_metadata
*pmd
,
1123 uint64_t current_id
,
1126 down_write(&pmd
->root_lock
);
1127 if (pmd
->trans_id
!= current_id
) {
1128 up_write(&pmd
->root_lock
);
1129 DMERR("mismatched transaction id");
1133 pmd
->trans_id
= new_id
;
1134 up_write(&pmd
->root_lock
);
1139 int dm_pool_get_metadata_transaction_id(struct dm_pool_metadata
*pmd
,
1142 down_read(&pmd
->root_lock
);
1143 *result
= pmd
->trans_id
;
1144 up_read(&pmd
->root_lock
);
1149 static int __reserve_metadata_snap(struct dm_pool_metadata
*pmd
)
1152 struct thin_disk_superblock
*disk_super
;
1153 struct dm_block
*copy
, *sblock
;
1154 dm_block_t held_root
;
1157 * Copy the superblock.
1159 dm_sm_inc_block(pmd
->metadata_sm
, THIN_SUPERBLOCK_LOCATION
);
1160 r
= dm_tm_shadow_block(pmd
->tm
, THIN_SUPERBLOCK_LOCATION
,
1161 &sb_validator
, ©
, &inc
);
1167 held_root
= dm_block_location(copy
);
1168 disk_super
= dm_block_data(copy
);
1170 if (le64_to_cpu(disk_super
->held_root
)) {
1171 DMWARN("Pool metadata snapshot already exists: release this before taking another.");
1173 dm_tm_dec(pmd
->tm
, held_root
);
1174 dm_tm_unlock(pmd
->tm
, copy
);
1179 * Wipe the spacemap since we're not publishing this.
1181 memset(&disk_super
->data_space_map_root
, 0,
1182 sizeof(disk_super
->data_space_map_root
));
1183 memset(&disk_super
->metadata_space_map_root
, 0,
1184 sizeof(disk_super
->metadata_space_map_root
));
1187 * Increment the data structures that need to be preserved.
1189 dm_tm_inc(pmd
->tm
, le64_to_cpu(disk_super
->data_mapping_root
));
1190 dm_tm_inc(pmd
->tm
, le64_to_cpu(disk_super
->device_details_root
));
1191 dm_tm_unlock(pmd
->tm
, copy
);
1194 * Write the held root into the superblock.
1196 r
= superblock_lock(pmd
, &sblock
);
1198 dm_tm_dec(pmd
->tm
, held_root
);
1202 disk_super
= dm_block_data(sblock
);
1203 disk_super
->held_root
= cpu_to_le64(held_root
);
1204 dm_bm_unlock(sblock
);
1208 int dm_pool_reserve_metadata_snap(struct dm_pool_metadata
*pmd
)
1212 down_write(&pmd
->root_lock
);
1213 r
= __reserve_metadata_snap(pmd
);
1214 up_write(&pmd
->root_lock
);
1219 static int __release_metadata_snap(struct dm_pool_metadata
*pmd
)
1222 struct thin_disk_superblock
*disk_super
;
1223 struct dm_block
*sblock
, *copy
;
1224 dm_block_t held_root
;
1226 r
= superblock_lock(pmd
, &sblock
);
1230 disk_super
= dm_block_data(sblock
);
1231 held_root
= le64_to_cpu(disk_super
->held_root
);
1232 disk_super
->held_root
= cpu_to_le64(0);
1234 dm_bm_unlock(sblock
);
1237 DMWARN("No pool metadata snapshot found: nothing to release.");
1241 r
= dm_tm_read_lock(pmd
->tm
, held_root
, &sb_validator
, ©
);
1245 disk_super
= dm_block_data(copy
);
1246 dm_sm_dec_block(pmd
->metadata_sm
, le64_to_cpu(disk_super
->data_mapping_root
));
1247 dm_sm_dec_block(pmd
->metadata_sm
, le64_to_cpu(disk_super
->device_details_root
));
1248 dm_sm_dec_block(pmd
->metadata_sm
, held_root
);
1250 return dm_tm_unlock(pmd
->tm
, copy
);
1253 int dm_pool_release_metadata_snap(struct dm_pool_metadata
*pmd
)
1257 down_write(&pmd
->root_lock
);
1258 r
= __release_metadata_snap(pmd
);
1259 up_write(&pmd
->root_lock
);
1264 static int __get_metadata_snap(struct dm_pool_metadata
*pmd
,
1268 struct thin_disk_superblock
*disk_super
;
1269 struct dm_block
*sblock
;
1271 r
= dm_bm_read_lock(pmd
->bm
, THIN_SUPERBLOCK_LOCATION
,
1272 &sb_validator
, &sblock
);
1276 disk_super
= dm_block_data(sblock
);
1277 *result
= le64_to_cpu(disk_super
->held_root
);
1279 return dm_bm_unlock(sblock
);
1282 int dm_pool_get_metadata_snap(struct dm_pool_metadata
*pmd
,
1287 down_read(&pmd
->root_lock
);
1288 r
= __get_metadata_snap(pmd
, result
);
1289 up_read(&pmd
->root_lock
);
1294 int dm_pool_open_thin_device(struct dm_pool_metadata
*pmd
, dm_thin_id dev
,
1295 struct dm_thin_device
**td
)
1299 down_write(&pmd
->root_lock
);
1300 r
= __open_device(pmd
, dev
, 0, td
);
1301 up_write(&pmd
->root_lock
);
1306 int dm_pool_close_thin_device(struct dm_thin_device
*td
)
1308 down_write(&td
->pmd
->root_lock
);
1310 up_write(&td
->pmd
->root_lock
);
1315 dm_thin_id
dm_thin_dev_id(struct dm_thin_device
*td
)
1320 static bool __snapshotted_since(struct dm_thin_device
*td
, uint32_t time
)
1322 return td
->snapshotted_time
> time
;
1325 int dm_thin_find_block(struct dm_thin_device
*td
, dm_block_t block
,
1326 int can_block
, struct dm_thin_lookup_result
*result
)
1329 uint64_t block_time
= 0;
1331 struct dm_pool_metadata
*pmd
= td
->pmd
;
1332 dm_block_t keys
[2] = { td
->id
, block
};
1335 down_read(&pmd
->root_lock
);
1336 r
= dm_btree_lookup(&pmd
->info
, pmd
->root
, keys
, &value
);
1338 block_time
= le64_to_cpu(value
);
1339 up_read(&pmd
->root_lock
);
1341 } else if (down_read_trylock(&pmd
->root_lock
)) {
1342 r
= dm_btree_lookup(&pmd
->nb_info
, pmd
->root
, keys
, &value
);
1344 block_time
= le64_to_cpu(value
);
1345 up_read(&pmd
->root_lock
);
1348 return -EWOULDBLOCK
;
1351 dm_block_t exception_block
;
1352 uint32_t exception_time
;
1353 unpack_block_time(block_time
, &exception_block
,
1355 result
->block
= exception_block
;
1356 result
->shared
= __snapshotted_since(td
, exception_time
);
1362 static int __insert(struct dm_thin_device
*td
, dm_block_t block
,
1363 dm_block_t data_block
)
1367 struct dm_pool_metadata
*pmd
= td
->pmd
;
1368 dm_block_t keys
[2] = { td
->id
, block
};
1370 value
= cpu_to_le64(pack_block_time(data_block
, pmd
->time
));
1371 __dm_bless_for_disk(&value
);
1373 r
= dm_btree_insert_notify(&pmd
->info
, pmd
->root
, keys
, &value
,
1374 &pmd
->root
, &inserted
);
1379 td
->mapped_blocks
++;
1386 int dm_thin_insert_block(struct dm_thin_device
*td
, dm_block_t block
,
1387 dm_block_t data_block
)
1391 down_write(&td
->pmd
->root_lock
);
1392 r
= __insert(td
, block
, data_block
);
1393 up_write(&td
->pmd
->root_lock
);
1398 static int __remove(struct dm_thin_device
*td
, dm_block_t block
)
1401 struct dm_pool_metadata
*pmd
= td
->pmd
;
1402 dm_block_t keys
[2] = { td
->id
, block
};
1404 r
= dm_btree_remove(&pmd
->info
, pmd
->root
, keys
, &pmd
->root
);
1408 td
->mapped_blocks
--;
1414 int dm_thin_remove_block(struct dm_thin_device
*td
, dm_block_t block
)
1418 down_write(&td
->pmd
->root_lock
);
1419 r
= __remove(td
, block
);
1420 up_write(&td
->pmd
->root_lock
);
1425 int dm_pool_alloc_data_block(struct dm_pool_metadata
*pmd
, dm_block_t
*result
)
1429 down_write(&pmd
->root_lock
);
1430 r
= dm_sm_new_block(pmd
->data_sm
, result
);
1431 up_write(&pmd
->root_lock
);
1436 int dm_pool_commit_metadata(struct dm_pool_metadata
*pmd
)
1440 down_write(&pmd
->root_lock
);
1442 r
= __commit_transaction(pmd
);
1447 * Open the next transaction.
1449 r
= __begin_transaction(pmd
);
1451 up_write(&pmd
->root_lock
);
1455 int dm_pool_get_free_block_count(struct dm_pool_metadata
*pmd
, dm_block_t
*result
)
1459 down_read(&pmd
->root_lock
);
1460 r
= dm_sm_get_nr_free(pmd
->data_sm
, result
);
1461 up_read(&pmd
->root_lock
);
1466 int dm_pool_get_free_metadata_block_count(struct dm_pool_metadata
*pmd
,
1471 down_read(&pmd
->root_lock
);
1472 r
= dm_sm_get_nr_free(pmd
->metadata_sm
, result
);
1473 up_read(&pmd
->root_lock
);
1478 int dm_pool_get_metadata_dev_size(struct dm_pool_metadata
*pmd
,
1483 down_read(&pmd
->root_lock
);
1484 r
= dm_sm_get_nr_blocks(pmd
->metadata_sm
, result
);
1485 up_read(&pmd
->root_lock
);
1490 int dm_pool_get_data_block_size(struct dm_pool_metadata
*pmd
, sector_t
*result
)
1492 down_read(&pmd
->root_lock
);
1493 *result
= pmd
->data_block_size
;
1494 up_read(&pmd
->root_lock
);
1499 int dm_pool_get_data_dev_size(struct dm_pool_metadata
*pmd
, dm_block_t
*result
)
1503 down_read(&pmd
->root_lock
);
1504 r
= dm_sm_get_nr_blocks(pmd
->data_sm
, result
);
1505 up_read(&pmd
->root_lock
);
1510 int dm_thin_get_mapped_count(struct dm_thin_device
*td
, dm_block_t
*result
)
1512 struct dm_pool_metadata
*pmd
= td
->pmd
;
1514 down_read(&pmd
->root_lock
);
1515 *result
= td
->mapped_blocks
;
1516 up_read(&pmd
->root_lock
);
1521 static int __highest_block(struct dm_thin_device
*td
, dm_block_t
*result
)
1525 dm_block_t thin_root
;
1526 struct dm_pool_metadata
*pmd
= td
->pmd
;
1528 r
= dm_btree_lookup(&pmd
->tl_info
, pmd
->root
, &td
->id
, &value_le
);
1532 thin_root
= le64_to_cpu(value_le
);
1534 return dm_btree_find_highest_key(&pmd
->bl_info
, thin_root
, result
);
1537 int dm_thin_get_highest_mapped_block(struct dm_thin_device
*td
,
1541 struct dm_pool_metadata
*pmd
= td
->pmd
;
1543 down_read(&pmd
->root_lock
);
1544 r
= __highest_block(td
, result
);
1545 up_read(&pmd
->root_lock
);
1550 static int __resize_data_dev(struct dm_pool_metadata
*pmd
, dm_block_t new_count
)
1553 dm_block_t old_count
;
1555 r
= dm_sm_get_nr_blocks(pmd
->data_sm
, &old_count
);
1559 if (new_count
== old_count
)
1562 if (new_count
< old_count
) {
1563 DMERR("cannot reduce size of data device");
1567 return dm_sm_extend(pmd
->data_sm
, new_count
- old_count
);
1570 int dm_pool_resize_data_dev(struct dm_pool_metadata
*pmd
, dm_block_t new_count
)
1574 down_write(&pmd
->root_lock
);
1575 r
= __resize_data_dev(pmd
, new_count
);
1576 up_write(&pmd
->root_lock
);