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 2
80 #define SECTOR_TO_BLOCK_SHIFT 3
84 * 3 for btree insert +
85 * 2 for btree lookup used within space map
87 * 2 for shadow spine +
88 * 4 for rebalance 3 child node
90 #define THIN_MAX_CONCURRENT_LOCKS 6
92 /* This should be plenty */
93 #define SPACE_MAP_ROOT_SIZE 128
96 * Little endian on-disk superblock and device details.
98 struct thin_disk_superblock {
99 __le32 csum; /* Checksum of superblock except for this field. */
101 __le64 blocknr; /* This block number, dm_block_t. */
111 * Root held by userspace transactions.
115 __u8 data_space_map_root[SPACE_MAP_ROOT_SIZE];
116 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
119 * 2-level btree mapping (dev_id, (dev block, time)) -> data block
121 __le64 data_mapping_root;
124 * Device detail root mapping dev_id -> device_details
126 __le64 device_details_root;
128 __le32 data_block_size; /* In 512-byte sectors. */
130 __le32 metadata_block_size; /* In 512-byte sectors. */
131 __le64 metadata_nr_blocks;
134 __le32 compat_ro_flags;
135 __le32 incompat_flags;
138 struct disk_device_details {
139 __le64 mapped_blocks;
140 __le64 transaction_id; /* When created. */
141 __le32 creation_time;
142 __le32 snapshotted_time;
145 struct dm_pool_metadata {
146 struct hlist_node hash;
148 struct block_device *bdev;
149 struct dm_block_manager *bm;
150 struct dm_space_map *metadata_sm;
151 struct dm_space_map *data_sm;
152 struct dm_transaction_manager *tm;
153 struct dm_transaction_manager *nb_tm;
157 * First level holds thin_dev_t.
158 * Second level holds mappings.
160 struct dm_btree_info info;
163 * Non-blocking version of the above.
165 struct dm_btree_info nb_info;
168 * Just the top level for deleting whole devices.
170 struct dm_btree_info tl_info;
173 * Just the bottom level for creating new devices.
175 struct dm_btree_info bl_info;
178 * Describes the device details btree.
180 struct dm_btree_info details_info;
182 struct rw_semaphore root_lock;
185 dm_block_t details_root;
186 struct list_head thin_devices;
189 sector_t data_block_size;
192 * We reserve a section of the metadata for commit overhead.
193 * All reported space does *not* include this.
195 dm_block_t metadata_reserve;
198 * Set if a transaction has to be aborted but the attempt to roll back
199 * to the previous (good) transaction failed. The only pool metadata
200 * operation possible in this state is the closing of the device.
205 * Set once a thin-pool has been accessed through one of the interfaces
206 * that imply the pool is in-service (e.g. thin devices created/deleted,
207 * thin-pool message, metadata snapshots, etc).
212 * Reading the space map roots can fail, so we read it into these
213 * buffers before the superblock is locked and updated.
215 __u8 data_space_map_root[SPACE_MAP_ROOT_SIZE];
216 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
219 struct dm_thin_device {
220 struct list_head list;
221 struct dm_pool_metadata *pmd;
226 bool aborted_with_changes:1;
227 uint64_t mapped_blocks;
228 uint64_t transaction_id;
229 uint32_t creation_time;
230 uint32_t snapshotted_time;
233 /*----------------------------------------------------------------
234 * superblock validator
235 *--------------------------------------------------------------*/
237 #define SUPERBLOCK_CSUM_XOR 160774
239 static void sb_prepare_for_write(struct dm_block_validator *v,
243 struct thin_disk_superblock *disk_super = dm_block_data(b);
245 disk_super->blocknr = cpu_to_le64(dm_block_location(b));
246 disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
247 block_size - sizeof(__le32),
248 SUPERBLOCK_CSUM_XOR));
251 static int sb_check(struct dm_block_validator *v,
255 struct thin_disk_superblock *disk_super = dm_block_data(b);
258 if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
259 DMERR("sb_check failed: blocknr %llu: "
260 "wanted %llu", le64_to_cpu(disk_super->blocknr),
261 (unsigned long long)dm_block_location(b));
265 if (le64_to_cpu(disk_super->magic) != THIN_SUPERBLOCK_MAGIC) {
266 DMERR("sb_check failed: magic %llu: "
267 "wanted %llu", le64_to_cpu(disk_super->magic),
268 (unsigned long long)THIN_SUPERBLOCK_MAGIC);
272 csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
273 block_size - sizeof(__le32),
274 SUPERBLOCK_CSUM_XOR));
275 if (csum_le != disk_super->csum) {
276 DMERR("sb_check failed: csum %u: wanted %u",
277 le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
284 static struct dm_block_validator sb_validator = {
285 .name = "superblock",
286 .prepare_for_write = sb_prepare_for_write,
290 /*----------------------------------------------------------------
291 * Methods for the btree value types
292 *--------------------------------------------------------------*/
294 static uint64_t pack_block_time(dm_block_t b, uint32_t t)
296 return (b << 24) | t;
299 static void unpack_block_time(uint64_t v, dm_block_t *b, uint32_t *t)
302 *t = v & ((1 << 24) - 1);
305 static void data_block_inc(void *context, const void *value_le)
307 struct dm_space_map *sm = context;
312 memcpy(&v_le, value_le, sizeof(v_le));
313 unpack_block_time(le64_to_cpu(v_le), &b, &t);
314 dm_sm_inc_block(sm, b);
317 static void data_block_dec(void *context, const void *value_le)
319 struct dm_space_map *sm = context;
324 memcpy(&v_le, value_le, sizeof(v_le));
325 unpack_block_time(le64_to_cpu(v_le), &b, &t);
326 dm_sm_dec_block(sm, b);
329 static int data_block_equal(void *context, const void *value1_le, const void *value2_le)
335 memcpy(&v1_le, value1_le, sizeof(v1_le));
336 memcpy(&v2_le, value2_le, sizeof(v2_le));
337 unpack_block_time(le64_to_cpu(v1_le), &b1, &t);
338 unpack_block_time(le64_to_cpu(v2_le), &b2, &t);
343 static void subtree_inc(void *context, const void *value)
345 struct dm_btree_info *info = context;
349 memcpy(&root_le, value, sizeof(root_le));
350 root = le64_to_cpu(root_le);
351 dm_tm_inc(info->tm, root);
354 static void subtree_dec(void *context, const void *value)
356 struct dm_btree_info *info = context;
360 memcpy(&root_le, value, sizeof(root_le));
361 root = le64_to_cpu(root_le);
362 if (dm_btree_del(info, root))
363 DMERR("btree delete failed");
366 static int subtree_equal(void *context, const void *value1_le, const void *value2_le)
369 memcpy(&v1_le, value1_le, sizeof(v1_le));
370 memcpy(&v2_le, value2_le, sizeof(v2_le));
372 return v1_le == v2_le;
375 /*----------------------------------------------------------------*/
378 * Variant that is used for in-core only changes or code that
379 * shouldn't put the pool in service on its own (e.g. commit).
381 static inline void __pmd_write_lock(struct dm_pool_metadata *pmd)
382 __acquires(pmd->root_lock)
384 down_write(&pmd->root_lock);
386 #define pmd_write_lock_in_core(pmd) __pmd_write_lock((pmd))
388 static inline void pmd_write_lock(struct dm_pool_metadata *pmd)
390 __pmd_write_lock(pmd);
391 if (unlikely(!pmd->in_service))
392 pmd->in_service = true;
395 static inline void pmd_write_unlock(struct dm_pool_metadata *pmd)
396 __releases(pmd->root_lock)
398 up_write(&pmd->root_lock);
401 /*----------------------------------------------------------------*/
403 static int superblock_lock_zero(struct dm_pool_metadata *pmd,
404 struct dm_block **sblock)
406 return dm_bm_write_lock_zero(pmd->bm, THIN_SUPERBLOCK_LOCATION,
407 &sb_validator, sblock);
410 static int superblock_lock(struct dm_pool_metadata *pmd,
411 struct dm_block **sblock)
413 return dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
414 &sb_validator, sblock);
417 static int __superblock_all_zeroes(struct dm_block_manager *bm, int *result)
422 __le64 *data_le, zero = cpu_to_le64(0);
423 unsigned block_size = dm_bm_block_size(bm) / sizeof(__le64);
426 * We can't use a validator here - it may be all zeroes.
428 r = dm_bm_read_lock(bm, THIN_SUPERBLOCK_LOCATION, NULL, &b);
432 data_le = dm_block_data(b);
434 for (i = 0; i < block_size; i++) {
435 if (data_le[i] != zero) {
446 static void __setup_btree_details(struct dm_pool_metadata *pmd)
448 pmd->info.tm = pmd->tm;
449 pmd->info.levels = 2;
450 pmd->info.value_type.context = pmd->data_sm;
451 pmd->info.value_type.size = sizeof(__le64);
452 pmd->info.value_type.inc = data_block_inc;
453 pmd->info.value_type.dec = data_block_dec;
454 pmd->info.value_type.equal = data_block_equal;
456 memcpy(&pmd->nb_info, &pmd->info, sizeof(pmd->nb_info));
457 pmd->nb_info.tm = pmd->nb_tm;
459 pmd->tl_info.tm = pmd->tm;
460 pmd->tl_info.levels = 1;
461 pmd->tl_info.value_type.context = &pmd->bl_info;
462 pmd->tl_info.value_type.size = sizeof(__le64);
463 pmd->tl_info.value_type.inc = subtree_inc;
464 pmd->tl_info.value_type.dec = subtree_dec;
465 pmd->tl_info.value_type.equal = subtree_equal;
467 pmd->bl_info.tm = pmd->tm;
468 pmd->bl_info.levels = 1;
469 pmd->bl_info.value_type.context = pmd->data_sm;
470 pmd->bl_info.value_type.size = sizeof(__le64);
471 pmd->bl_info.value_type.inc = data_block_inc;
472 pmd->bl_info.value_type.dec = data_block_dec;
473 pmd->bl_info.value_type.equal = data_block_equal;
475 pmd->details_info.tm = pmd->tm;
476 pmd->details_info.levels = 1;
477 pmd->details_info.value_type.context = NULL;
478 pmd->details_info.value_type.size = sizeof(struct disk_device_details);
479 pmd->details_info.value_type.inc = NULL;
480 pmd->details_info.value_type.dec = NULL;
481 pmd->details_info.value_type.equal = NULL;
484 static int save_sm_roots(struct dm_pool_metadata *pmd)
489 r = dm_sm_root_size(pmd->metadata_sm, &len);
493 r = dm_sm_copy_root(pmd->metadata_sm, &pmd->metadata_space_map_root, len);
497 r = dm_sm_root_size(pmd->data_sm, &len);
501 return dm_sm_copy_root(pmd->data_sm, &pmd->data_space_map_root, len);
504 static void copy_sm_roots(struct dm_pool_metadata *pmd,
505 struct thin_disk_superblock *disk)
507 memcpy(&disk->metadata_space_map_root,
508 &pmd->metadata_space_map_root,
509 sizeof(pmd->metadata_space_map_root));
511 memcpy(&disk->data_space_map_root,
512 &pmd->data_space_map_root,
513 sizeof(pmd->data_space_map_root));
516 static int __write_initial_superblock(struct dm_pool_metadata *pmd)
519 struct dm_block *sblock;
520 struct thin_disk_superblock *disk_super;
521 sector_t bdev_size = i_size_read(pmd->bdev->bd_inode) >> SECTOR_SHIFT;
523 if (bdev_size > THIN_METADATA_MAX_SECTORS)
524 bdev_size = THIN_METADATA_MAX_SECTORS;
526 r = dm_sm_commit(pmd->data_sm);
530 r = dm_tm_pre_commit(pmd->tm);
534 r = save_sm_roots(pmd);
538 r = superblock_lock_zero(pmd, &sblock);
542 disk_super = dm_block_data(sblock);
543 disk_super->flags = 0;
544 memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
545 disk_super->magic = cpu_to_le64(THIN_SUPERBLOCK_MAGIC);
546 disk_super->version = cpu_to_le32(THIN_VERSION);
547 disk_super->time = 0;
548 disk_super->trans_id = 0;
549 disk_super->held_root = 0;
551 copy_sm_roots(pmd, disk_super);
553 disk_super->data_mapping_root = cpu_to_le64(pmd->root);
554 disk_super->device_details_root = cpu_to_le64(pmd->details_root);
555 disk_super->metadata_block_size = cpu_to_le32(THIN_METADATA_BLOCK_SIZE);
556 disk_super->metadata_nr_blocks = cpu_to_le64(bdev_size >> SECTOR_TO_BLOCK_SHIFT);
557 disk_super->data_block_size = cpu_to_le32(pmd->data_block_size);
559 return dm_tm_commit(pmd->tm, sblock);
562 static int __format_metadata(struct dm_pool_metadata *pmd)
566 r = dm_tm_create_with_sm(pmd->bm, THIN_SUPERBLOCK_LOCATION,
567 &pmd->tm, &pmd->metadata_sm);
569 DMERR("tm_create_with_sm failed");
573 pmd->data_sm = dm_sm_disk_create(pmd->tm, 0);
574 if (IS_ERR(pmd->data_sm)) {
575 DMERR("sm_disk_create failed");
576 r = PTR_ERR(pmd->data_sm);
580 pmd->nb_tm = dm_tm_create_non_blocking_clone(pmd->tm);
582 DMERR("could not create non-blocking clone tm");
584 goto bad_cleanup_data_sm;
587 __setup_btree_details(pmd);
589 r = dm_btree_empty(&pmd->info, &pmd->root);
591 goto bad_cleanup_nb_tm;
593 r = dm_btree_empty(&pmd->details_info, &pmd->details_root);
595 DMERR("couldn't create devices root");
596 goto bad_cleanup_nb_tm;
599 r = __write_initial_superblock(pmd);
601 goto bad_cleanup_nb_tm;
606 dm_tm_destroy(pmd->nb_tm);
608 dm_sm_destroy(pmd->data_sm);
610 dm_tm_destroy(pmd->tm);
611 dm_sm_destroy(pmd->metadata_sm);
616 static int __check_incompat_features(struct thin_disk_superblock *disk_super,
617 struct dm_pool_metadata *pmd)
621 features = le32_to_cpu(disk_super->incompat_flags) & ~THIN_FEATURE_INCOMPAT_SUPP;
623 DMERR("could not access metadata due to unsupported optional features (%lx).",
624 (unsigned long)features);
629 * Check for read-only metadata to skip the following RDWR checks.
631 if (get_disk_ro(pmd->bdev->bd_disk))
634 features = le32_to_cpu(disk_super->compat_ro_flags) & ~THIN_FEATURE_COMPAT_RO_SUPP;
636 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
637 (unsigned long)features);
644 static int __open_metadata(struct dm_pool_metadata *pmd)
647 struct dm_block *sblock;
648 struct thin_disk_superblock *disk_super;
650 r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
651 &sb_validator, &sblock);
653 DMERR("couldn't read superblock");
657 disk_super = dm_block_data(sblock);
659 /* Verify the data block size hasn't changed */
660 if (le32_to_cpu(disk_super->data_block_size) != pmd->data_block_size) {
661 DMERR("changing the data block size (from %u to %llu) is not supported",
662 le32_to_cpu(disk_super->data_block_size),
663 (unsigned long long)pmd->data_block_size);
665 goto bad_unlock_sblock;
668 r = __check_incompat_features(disk_super, pmd);
670 goto bad_unlock_sblock;
672 r = dm_tm_open_with_sm(pmd->bm, THIN_SUPERBLOCK_LOCATION,
673 disk_super->metadata_space_map_root,
674 sizeof(disk_super->metadata_space_map_root),
675 &pmd->tm, &pmd->metadata_sm);
677 DMERR("tm_open_with_sm failed");
678 goto bad_unlock_sblock;
681 pmd->data_sm = dm_sm_disk_open(pmd->tm, disk_super->data_space_map_root,
682 sizeof(disk_super->data_space_map_root));
683 if (IS_ERR(pmd->data_sm)) {
684 DMERR("sm_disk_open failed");
685 r = PTR_ERR(pmd->data_sm);
689 pmd->nb_tm = dm_tm_create_non_blocking_clone(pmd->tm);
691 DMERR("could not create non-blocking clone tm");
693 goto bad_cleanup_data_sm;
696 __setup_btree_details(pmd);
697 dm_bm_unlock(sblock);
702 dm_sm_destroy(pmd->data_sm);
704 dm_tm_destroy(pmd->tm);
705 dm_sm_destroy(pmd->metadata_sm);
707 dm_bm_unlock(sblock);
712 static int __open_or_format_metadata(struct dm_pool_metadata *pmd, bool format_device)
716 r = __superblock_all_zeroes(pmd->bm, &unformatted);
721 return format_device ? __format_metadata(pmd) : -EPERM;
723 return __open_metadata(pmd);
726 static int __create_persistent_data_objects(struct dm_pool_metadata *pmd, bool format_device)
730 pmd->bm = dm_block_manager_create(pmd->bdev, THIN_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
731 THIN_MAX_CONCURRENT_LOCKS);
732 if (IS_ERR(pmd->bm)) {
733 DMERR("could not create block manager");
734 return PTR_ERR(pmd->bm);
737 r = __open_or_format_metadata(pmd, format_device);
739 dm_block_manager_destroy(pmd->bm);
744 static void __destroy_persistent_data_objects(struct dm_pool_metadata *pmd)
746 dm_sm_destroy(pmd->data_sm);
747 dm_sm_destroy(pmd->metadata_sm);
748 dm_tm_destroy(pmd->nb_tm);
749 dm_tm_destroy(pmd->tm);
750 dm_block_manager_destroy(pmd->bm);
753 static int __begin_transaction(struct dm_pool_metadata *pmd)
756 struct thin_disk_superblock *disk_super;
757 struct dm_block *sblock;
760 * We re-read the superblock every time. Shouldn't need to do this
763 r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
764 &sb_validator, &sblock);
768 disk_super = dm_block_data(sblock);
769 pmd->time = le32_to_cpu(disk_super->time);
770 pmd->root = le64_to_cpu(disk_super->data_mapping_root);
771 pmd->details_root = le64_to_cpu(disk_super->device_details_root);
772 pmd->trans_id = le64_to_cpu(disk_super->trans_id);
773 pmd->flags = le32_to_cpu(disk_super->flags);
774 pmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
776 dm_bm_unlock(sblock);
780 static int __write_changed_details(struct dm_pool_metadata *pmd)
783 struct dm_thin_device *td, *tmp;
784 struct disk_device_details details;
787 list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
793 details.mapped_blocks = cpu_to_le64(td->mapped_blocks);
794 details.transaction_id = cpu_to_le64(td->transaction_id);
795 details.creation_time = cpu_to_le32(td->creation_time);
796 details.snapshotted_time = cpu_to_le32(td->snapshotted_time);
797 __dm_bless_for_disk(&details);
799 r = dm_btree_insert(&pmd->details_info, pmd->details_root,
800 &key, &details, &pmd->details_root);
815 static int __commit_transaction(struct dm_pool_metadata *pmd)
818 struct thin_disk_superblock *disk_super;
819 struct dm_block *sblock;
822 * We need to know if the thin_disk_superblock exceeds a 512-byte sector.
824 BUILD_BUG_ON(sizeof(struct thin_disk_superblock) > 512);
826 if (unlikely(!pmd->in_service))
829 r = __write_changed_details(pmd);
833 r = dm_sm_commit(pmd->data_sm);
837 r = dm_tm_pre_commit(pmd->tm);
841 r = save_sm_roots(pmd);
845 r = superblock_lock(pmd, &sblock);
849 disk_super = dm_block_data(sblock);
850 disk_super->time = cpu_to_le32(pmd->time);
851 disk_super->data_mapping_root = cpu_to_le64(pmd->root);
852 disk_super->device_details_root = cpu_to_le64(pmd->details_root);
853 disk_super->trans_id = cpu_to_le64(pmd->trans_id);
854 disk_super->flags = cpu_to_le32(pmd->flags);
856 copy_sm_roots(pmd, disk_super);
858 return dm_tm_commit(pmd->tm, sblock);
861 static void __set_metadata_reserve(struct dm_pool_metadata *pmd)
865 dm_block_t max_blocks = 4096; /* 16M */
867 r = dm_sm_get_nr_blocks(pmd->metadata_sm, &total);
869 DMERR("could not get size of metadata device");
870 pmd->metadata_reserve = max_blocks;
872 pmd->metadata_reserve = min(max_blocks, div_u64(total, 10));
875 struct dm_pool_metadata *dm_pool_metadata_open(struct block_device *bdev,
876 sector_t data_block_size,
880 struct dm_pool_metadata *pmd;
882 pmd = kmalloc(sizeof(*pmd), GFP_KERNEL);
884 DMERR("could not allocate metadata struct");
885 return ERR_PTR(-ENOMEM);
888 init_rwsem(&pmd->root_lock);
890 INIT_LIST_HEAD(&pmd->thin_devices);
891 pmd->fail_io = false;
892 pmd->in_service = false;
894 pmd->data_block_size = data_block_size;
896 r = __create_persistent_data_objects(pmd, format_device);
902 r = __begin_transaction(pmd);
904 if (dm_pool_metadata_close(pmd) < 0)
905 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
909 __set_metadata_reserve(pmd);
914 int dm_pool_metadata_close(struct dm_pool_metadata *pmd)
917 unsigned open_devices = 0;
918 struct dm_thin_device *td, *tmp;
920 down_read(&pmd->root_lock);
921 list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
929 up_read(&pmd->root_lock);
932 DMERR("attempt to close pmd when %u device(s) are still open",
937 if (!dm_bm_is_read_only(pmd->bm) && !pmd->fail_io) {
938 r = __commit_transaction(pmd);
940 DMWARN("%s: __commit_transaction() failed, error = %d",
944 __destroy_persistent_data_objects(pmd);
951 * __open_device: Returns @td corresponding to device with id @dev,
952 * creating it if @create is set and incrementing @td->open_count.
953 * On failure, @td is undefined.
955 static int __open_device(struct dm_pool_metadata *pmd,
956 dm_thin_id dev, int create,
957 struct dm_thin_device **td)
960 struct dm_thin_device *td2;
962 struct disk_device_details details_le;
965 * If the device is already open, return it.
967 list_for_each_entry(td2, &pmd->thin_devices, list)
968 if (td2->id == dev) {
970 * May not create an already-open device.
981 * Check the device exists.
983 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
986 if (r != -ENODATA || !create)
993 details_le.mapped_blocks = 0;
994 details_le.transaction_id = cpu_to_le64(pmd->trans_id);
995 details_le.creation_time = cpu_to_le32(pmd->time);
996 details_le.snapshotted_time = cpu_to_le32(pmd->time);
999 *td = kmalloc(sizeof(**td), GFP_NOIO);
1005 (*td)->open_count = 1;
1006 (*td)->changed = changed;
1007 (*td)->aborted_with_changes = false;
1008 (*td)->mapped_blocks = le64_to_cpu(details_le.mapped_blocks);
1009 (*td)->transaction_id = le64_to_cpu(details_le.transaction_id);
1010 (*td)->creation_time = le32_to_cpu(details_le.creation_time);
1011 (*td)->snapshotted_time = le32_to_cpu(details_le.snapshotted_time);
1013 list_add(&(*td)->list, &pmd->thin_devices);
1018 static void __close_device(struct dm_thin_device *td)
1023 static int __create_thin(struct dm_pool_metadata *pmd,
1027 dm_block_t dev_root;
1029 struct disk_device_details details_le;
1030 struct dm_thin_device *td;
1033 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
1039 * Create an empty btree for the mappings.
1041 r = dm_btree_empty(&pmd->bl_info, &dev_root);
1046 * Insert it into the main mapping tree.
1048 value = cpu_to_le64(dev_root);
1049 __dm_bless_for_disk(&value);
1050 r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
1052 dm_btree_del(&pmd->bl_info, dev_root);
1056 r = __open_device(pmd, dev, 1, &td);
1058 dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1059 dm_btree_del(&pmd->bl_info, dev_root);
1067 int dm_pool_create_thin(struct dm_pool_metadata *pmd, dm_thin_id dev)
1071 pmd_write_lock(pmd);
1073 r = __create_thin(pmd, dev);
1074 pmd_write_unlock(pmd);
1079 static int __set_snapshot_details(struct dm_pool_metadata *pmd,
1080 struct dm_thin_device *snap,
1081 dm_thin_id origin, uint32_t time)
1084 struct dm_thin_device *td;
1086 r = __open_device(pmd, origin, 0, &td);
1091 td->snapshotted_time = time;
1093 snap->mapped_blocks = td->mapped_blocks;
1094 snap->snapshotted_time = time;
1100 static int __create_snap(struct dm_pool_metadata *pmd,
1101 dm_thin_id dev, dm_thin_id origin)
1104 dm_block_t origin_root;
1105 uint64_t key = origin, dev_key = dev;
1106 struct dm_thin_device *td;
1107 struct disk_device_details details_le;
1110 /* check this device is unused */
1111 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
1112 &dev_key, &details_le);
1116 /* find the mapping tree for the origin */
1117 r = dm_btree_lookup(&pmd->tl_info, pmd->root, &key, &value);
1120 origin_root = le64_to_cpu(value);
1122 /* clone the origin, an inc will do */
1123 dm_tm_inc(pmd->tm, origin_root);
1125 /* insert into the main mapping tree */
1126 value = cpu_to_le64(origin_root);
1127 __dm_bless_for_disk(&value);
1129 r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
1131 dm_tm_dec(pmd->tm, origin_root);
1137 r = __open_device(pmd, dev, 1, &td);
1141 r = __set_snapshot_details(pmd, td, origin, pmd->time);
1150 dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1151 dm_btree_remove(&pmd->details_info, pmd->details_root,
1152 &key, &pmd->details_root);
1156 int dm_pool_create_snap(struct dm_pool_metadata *pmd,
1162 pmd_write_lock(pmd);
1164 r = __create_snap(pmd, dev, origin);
1165 pmd_write_unlock(pmd);
1170 static int __delete_device(struct dm_pool_metadata *pmd, dm_thin_id dev)
1174 struct dm_thin_device *td;
1176 /* TODO: failure should mark the transaction invalid */
1177 r = __open_device(pmd, dev, 0, &td);
1181 if (td->open_count > 1) {
1186 list_del(&td->list);
1188 r = dm_btree_remove(&pmd->details_info, pmd->details_root,
1189 &key, &pmd->details_root);
1193 r = dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1200 int dm_pool_delete_thin_device(struct dm_pool_metadata *pmd,
1205 pmd_write_lock(pmd);
1207 r = __delete_device(pmd, dev);
1208 pmd_write_unlock(pmd);
1213 int dm_pool_set_metadata_transaction_id(struct dm_pool_metadata *pmd,
1214 uint64_t current_id,
1219 pmd_write_lock(pmd);
1224 if (pmd->trans_id != current_id) {
1225 DMERR("mismatched transaction id");
1229 pmd->trans_id = new_id;
1233 pmd_write_unlock(pmd);
1238 int dm_pool_get_metadata_transaction_id(struct dm_pool_metadata *pmd,
1243 down_read(&pmd->root_lock);
1244 if (!pmd->fail_io) {
1245 *result = pmd->trans_id;
1248 up_read(&pmd->root_lock);
1253 static int __reserve_metadata_snap(struct dm_pool_metadata *pmd)
1256 struct thin_disk_superblock *disk_super;
1257 struct dm_block *copy, *sblock;
1258 dm_block_t held_root;
1261 * We commit to ensure the btree roots which we increment in a
1262 * moment are up to date.
1264 r = __commit_transaction(pmd);
1266 DMWARN("%s: __commit_transaction() failed, error = %d",
1272 * Copy the superblock.
1274 dm_sm_inc_block(pmd->metadata_sm, THIN_SUPERBLOCK_LOCATION);
1275 r = dm_tm_shadow_block(pmd->tm, THIN_SUPERBLOCK_LOCATION,
1276 &sb_validator, ©, &inc);
1282 held_root = dm_block_location(copy);
1283 disk_super = dm_block_data(copy);
1285 if (le64_to_cpu(disk_super->held_root)) {
1286 DMWARN("Pool metadata snapshot already exists: release this before taking another.");
1288 dm_tm_dec(pmd->tm, held_root);
1289 dm_tm_unlock(pmd->tm, copy);
1294 * Wipe the spacemap since we're not publishing this.
1296 memset(&disk_super->data_space_map_root, 0,
1297 sizeof(disk_super->data_space_map_root));
1298 memset(&disk_super->metadata_space_map_root, 0,
1299 sizeof(disk_super->metadata_space_map_root));
1302 * Increment the data structures that need to be preserved.
1304 dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->data_mapping_root));
1305 dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->device_details_root));
1306 dm_tm_unlock(pmd->tm, copy);
1309 * Write the held root into the superblock.
1311 r = superblock_lock(pmd, &sblock);
1313 dm_tm_dec(pmd->tm, held_root);
1317 disk_super = dm_block_data(sblock);
1318 disk_super->held_root = cpu_to_le64(held_root);
1319 dm_bm_unlock(sblock);
1323 int dm_pool_reserve_metadata_snap(struct dm_pool_metadata *pmd)
1327 pmd_write_lock(pmd);
1329 r = __reserve_metadata_snap(pmd);
1330 pmd_write_unlock(pmd);
1335 static int __release_metadata_snap(struct dm_pool_metadata *pmd)
1338 struct thin_disk_superblock *disk_super;
1339 struct dm_block *sblock, *copy;
1340 dm_block_t held_root;
1342 r = superblock_lock(pmd, &sblock);
1346 disk_super = dm_block_data(sblock);
1347 held_root = le64_to_cpu(disk_super->held_root);
1348 disk_super->held_root = cpu_to_le64(0);
1350 dm_bm_unlock(sblock);
1353 DMWARN("No pool metadata snapshot found: nothing to release.");
1357 r = dm_tm_read_lock(pmd->tm, held_root, &sb_validator, ©);
1361 disk_super = dm_block_data(copy);
1362 dm_btree_del(&pmd->info, le64_to_cpu(disk_super->data_mapping_root));
1363 dm_btree_del(&pmd->details_info, le64_to_cpu(disk_super->device_details_root));
1364 dm_sm_dec_block(pmd->metadata_sm, held_root);
1366 dm_tm_unlock(pmd->tm, copy);
1371 int dm_pool_release_metadata_snap(struct dm_pool_metadata *pmd)
1375 pmd_write_lock(pmd);
1377 r = __release_metadata_snap(pmd);
1378 pmd_write_unlock(pmd);
1383 static int __get_metadata_snap(struct dm_pool_metadata *pmd,
1387 struct thin_disk_superblock *disk_super;
1388 struct dm_block *sblock;
1390 r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
1391 &sb_validator, &sblock);
1395 disk_super = dm_block_data(sblock);
1396 *result = le64_to_cpu(disk_super->held_root);
1398 dm_bm_unlock(sblock);
1403 int dm_pool_get_metadata_snap(struct dm_pool_metadata *pmd,
1408 down_read(&pmd->root_lock);
1410 r = __get_metadata_snap(pmd, result);
1411 up_read(&pmd->root_lock);
1416 int dm_pool_open_thin_device(struct dm_pool_metadata *pmd, dm_thin_id dev,
1417 struct dm_thin_device **td)
1421 pmd_write_lock_in_core(pmd);
1423 r = __open_device(pmd, dev, 0, td);
1424 pmd_write_unlock(pmd);
1429 int dm_pool_close_thin_device(struct dm_thin_device *td)
1431 pmd_write_lock_in_core(td->pmd);
1433 pmd_write_unlock(td->pmd);
1438 dm_thin_id dm_thin_dev_id(struct dm_thin_device *td)
1444 * Check whether @time (of block creation) is older than @td's last snapshot.
1445 * If so then the associated block is shared with the last snapshot device.
1446 * Any block on a device created *after* the device last got snapshotted is
1447 * necessarily not shared.
1449 static bool __snapshotted_since(struct dm_thin_device *td, uint32_t time)
1451 return td->snapshotted_time > time;
1454 static void unpack_lookup_result(struct dm_thin_device *td, __le64 value,
1455 struct dm_thin_lookup_result *result)
1457 uint64_t block_time = 0;
1458 dm_block_t exception_block;
1459 uint32_t exception_time;
1461 block_time = le64_to_cpu(value);
1462 unpack_block_time(block_time, &exception_block, &exception_time);
1463 result->block = exception_block;
1464 result->shared = __snapshotted_since(td, exception_time);
1467 static int __find_block(struct dm_thin_device *td, dm_block_t block,
1468 int can_issue_io, struct dm_thin_lookup_result *result)
1472 struct dm_pool_metadata *pmd = td->pmd;
1473 dm_block_t keys[2] = { td->id, block };
1474 struct dm_btree_info *info;
1479 info = &pmd->nb_info;
1481 r = dm_btree_lookup(info, pmd->root, keys, &value);
1483 unpack_lookup_result(td, value, result);
1488 int dm_thin_find_block(struct dm_thin_device *td, dm_block_t block,
1489 int can_issue_io, struct dm_thin_lookup_result *result)
1492 struct dm_pool_metadata *pmd = td->pmd;
1494 down_read(&pmd->root_lock);
1496 up_read(&pmd->root_lock);
1500 r = __find_block(td, block, can_issue_io, result);
1502 up_read(&pmd->root_lock);
1506 static int __find_next_mapped_block(struct dm_thin_device *td, dm_block_t block,
1508 struct dm_thin_lookup_result *result)
1512 struct dm_pool_metadata *pmd = td->pmd;
1513 dm_block_t keys[2] = { td->id, block };
1515 r = dm_btree_lookup_next(&pmd->info, pmd->root, keys, vblock, &value);
1517 unpack_lookup_result(td, value, result);
1522 static int __find_mapped_range(struct dm_thin_device *td,
1523 dm_block_t begin, dm_block_t end,
1524 dm_block_t *thin_begin, dm_block_t *thin_end,
1525 dm_block_t *pool_begin, bool *maybe_shared)
1528 dm_block_t pool_end;
1529 struct dm_thin_lookup_result lookup;
1534 r = __find_next_mapped_block(td, begin, &begin, &lookup);
1541 *thin_begin = begin;
1542 *pool_begin = lookup.block;
1543 *maybe_shared = lookup.shared;
1546 pool_end = *pool_begin + 1;
1547 while (begin != end) {
1548 r = __find_block(td, begin, true, &lookup);
1556 if ((lookup.block != pool_end) ||
1557 (lookup.shared != *maybe_shared))
1568 int dm_thin_find_mapped_range(struct dm_thin_device *td,
1569 dm_block_t begin, dm_block_t end,
1570 dm_block_t *thin_begin, dm_block_t *thin_end,
1571 dm_block_t *pool_begin, bool *maybe_shared)
1574 struct dm_pool_metadata *pmd = td->pmd;
1576 down_read(&pmd->root_lock);
1577 if (!pmd->fail_io) {
1578 r = __find_mapped_range(td, begin, end, thin_begin, thin_end,
1579 pool_begin, maybe_shared);
1581 up_read(&pmd->root_lock);
1586 static int __insert(struct dm_thin_device *td, dm_block_t block,
1587 dm_block_t data_block)
1591 struct dm_pool_metadata *pmd = td->pmd;
1592 dm_block_t keys[2] = { td->id, block };
1594 value = cpu_to_le64(pack_block_time(data_block, pmd->time));
1595 __dm_bless_for_disk(&value);
1597 r = dm_btree_insert_notify(&pmd->info, pmd->root, keys, &value,
1598 &pmd->root, &inserted);
1604 td->mapped_blocks++;
1609 int dm_thin_insert_block(struct dm_thin_device *td, dm_block_t block,
1610 dm_block_t data_block)
1614 pmd_write_lock(td->pmd);
1615 if (!td->pmd->fail_io)
1616 r = __insert(td, block, data_block);
1617 pmd_write_unlock(td->pmd);
1622 static int __remove(struct dm_thin_device *td, dm_block_t block)
1625 struct dm_pool_metadata *pmd = td->pmd;
1626 dm_block_t keys[2] = { td->id, block };
1628 r = dm_btree_remove(&pmd->info, pmd->root, keys, &pmd->root);
1632 td->mapped_blocks--;
1638 static int __remove_range(struct dm_thin_device *td, dm_block_t begin, dm_block_t end)
1641 unsigned count, total_count = 0;
1642 struct dm_pool_metadata *pmd = td->pmd;
1643 dm_block_t keys[1] = { td->id };
1645 dm_block_t mapping_root;
1648 * Find the mapping tree
1650 r = dm_btree_lookup(&pmd->tl_info, pmd->root, keys, &value);
1655 * Remove from the mapping tree, taking care to inc the
1656 * ref count so it doesn't get deleted.
1658 mapping_root = le64_to_cpu(value);
1659 dm_tm_inc(pmd->tm, mapping_root);
1660 r = dm_btree_remove(&pmd->tl_info, pmd->root, keys, &pmd->root);
1665 * Remove leaves stops at the first unmapped entry, so we have to
1666 * loop round finding mapped ranges.
1668 while (begin < end) {
1669 r = dm_btree_lookup_next(&pmd->bl_info, mapping_root, &begin, &begin, &value);
1679 r = dm_btree_remove_leaves(&pmd->bl_info, mapping_root, &begin, end, &mapping_root, &count);
1683 total_count += count;
1686 td->mapped_blocks -= total_count;
1690 * Reinsert the mapping tree.
1692 value = cpu_to_le64(mapping_root);
1693 __dm_bless_for_disk(&value);
1694 return dm_btree_insert(&pmd->tl_info, pmd->root, keys, &value, &pmd->root);
1697 int dm_thin_remove_block(struct dm_thin_device *td, dm_block_t block)
1701 pmd_write_lock(td->pmd);
1702 if (!td->pmd->fail_io)
1703 r = __remove(td, block);
1704 pmd_write_unlock(td->pmd);
1709 int dm_thin_remove_range(struct dm_thin_device *td,
1710 dm_block_t begin, dm_block_t end)
1714 pmd_write_lock(td->pmd);
1715 if (!td->pmd->fail_io)
1716 r = __remove_range(td, begin, end);
1717 pmd_write_unlock(td->pmd);
1722 int dm_pool_block_is_shared(struct dm_pool_metadata *pmd, dm_block_t b, bool *result)
1727 down_read(&pmd->root_lock);
1728 r = dm_sm_get_count(pmd->data_sm, b, &ref_count);
1730 *result = (ref_count > 1);
1731 up_read(&pmd->root_lock);
1736 int dm_pool_inc_data_range(struct dm_pool_metadata *pmd, dm_block_t b, dm_block_t e)
1740 pmd_write_lock(pmd);
1741 for (; b != e; b++) {
1742 r = dm_sm_inc_block(pmd->data_sm, b);
1746 pmd_write_unlock(pmd);
1751 int dm_pool_dec_data_range(struct dm_pool_metadata *pmd, dm_block_t b, dm_block_t e)
1755 pmd_write_lock(pmd);
1756 for (; b != e; b++) {
1757 r = dm_sm_dec_block(pmd->data_sm, b);
1761 pmd_write_unlock(pmd);
1766 bool dm_thin_changed_this_transaction(struct dm_thin_device *td)
1770 down_read(&td->pmd->root_lock);
1772 up_read(&td->pmd->root_lock);
1777 bool dm_pool_changed_this_transaction(struct dm_pool_metadata *pmd)
1780 struct dm_thin_device *td, *tmp;
1782 down_read(&pmd->root_lock);
1783 list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
1789 up_read(&pmd->root_lock);
1794 bool dm_thin_aborted_changes(struct dm_thin_device *td)
1798 down_read(&td->pmd->root_lock);
1799 r = td->aborted_with_changes;
1800 up_read(&td->pmd->root_lock);
1805 int dm_pool_alloc_data_block(struct dm_pool_metadata *pmd, dm_block_t *result)
1809 pmd_write_lock(pmd);
1811 r = dm_sm_new_block(pmd->data_sm, result);
1812 pmd_write_unlock(pmd);
1817 int dm_pool_commit_metadata(struct dm_pool_metadata *pmd)
1822 * Care is taken to not have commit be what
1823 * triggers putting the thin-pool in-service.
1825 __pmd_write_lock(pmd);
1829 r = __commit_transaction(pmd);
1834 * Open the next transaction.
1836 r = __begin_transaction(pmd);
1838 pmd_write_unlock(pmd);
1842 static void __set_abort_with_changes_flags(struct dm_pool_metadata *pmd)
1844 struct dm_thin_device *td;
1846 list_for_each_entry(td, &pmd->thin_devices, list)
1847 td->aborted_with_changes = td->changed;
1850 int dm_pool_abort_metadata(struct dm_pool_metadata *pmd)
1854 pmd_write_lock(pmd);
1858 __set_abort_with_changes_flags(pmd);
1859 __destroy_persistent_data_objects(pmd);
1860 r = __create_persistent_data_objects(pmd, false);
1862 pmd->fail_io = true;
1865 pmd_write_unlock(pmd);
1870 int dm_pool_get_free_block_count(struct dm_pool_metadata *pmd, dm_block_t *result)
1874 down_read(&pmd->root_lock);
1876 r = dm_sm_get_nr_free(pmd->data_sm, result);
1877 up_read(&pmd->root_lock);
1882 int dm_pool_get_free_metadata_block_count(struct dm_pool_metadata *pmd,
1887 down_read(&pmd->root_lock);
1889 r = dm_sm_get_nr_free(pmd->metadata_sm, result);
1892 if (*result < pmd->metadata_reserve)
1895 *result -= pmd->metadata_reserve;
1897 up_read(&pmd->root_lock);
1902 int dm_pool_get_metadata_dev_size(struct dm_pool_metadata *pmd,
1907 down_read(&pmd->root_lock);
1909 r = dm_sm_get_nr_blocks(pmd->metadata_sm, result);
1910 up_read(&pmd->root_lock);
1915 int dm_pool_get_data_dev_size(struct dm_pool_metadata *pmd, dm_block_t *result)
1919 down_read(&pmd->root_lock);
1921 r = dm_sm_get_nr_blocks(pmd->data_sm, result);
1922 up_read(&pmd->root_lock);
1927 int dm_thin_get_mapped_count(struct dm_thin_device *td, dm_block_t *result)
1930 struct dm_pool_metadata *pmd = td->pmd;
1932 down_read(&pmd->root_lock);
1933 if (!pmd->fail_io) {
1934 *result = td->mapped_blocks;
1937 up_read(&pmd->root_lock);
1942 static int __highest_block(struct dm_thin_device *td, dm_block_t *result)
1946 dm_block_t thin_root;
1947 struct dm_pool_metadata *pmd = td->pmd;
1949 r = dm_btree_lookup(&pmd->tl_info, pmd->root, &td->id, &value_le);
1953 thin_root = le64_to_cpu(value_le);
1955 return dm_btree_find_highest_key(&pmd->bl_info, thin_root, result);
1958 int dm_thin_get_highest_mapped_block(struct dm_thin_device *td,
1962 struct dm_pool_metadata *pmd = td->pmd;
1964 down_read(&pmd->root_lock);
1966 r = __highest_block(td, result);
1967 up_read(&pmd->root_lock);
1972 static int __resize_space_map(struct dm_space_map *sm, dm_block_t new_count)
1975 dm_block_t old_count;
1977 r = dm_sm_get_nr_blocks(sm, &old_count);
1981 if (new_count == old_count)
1984 if (new_count < old_count) {
1985 DMERR("cannot reduce size of space map");
1989 return dm_sm_extend(sm, new_count - old_count);
1992 int dm_pool_resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1996 pmd_write_lock(pmd);
1998 r = __resize_space_map(pmd->data_sm, new_count);
1999 pmd_write_unlock(pmd);
2004 int dm_pool_resize_metadata_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
2008 pmd_write_lock(pmd);
2009 if (!pmd->fail_io) {
2010 r = __resize_space_map(pmd->metadata_sm, new_count);
2012 __set_metadata_reserve(pmd);
2014 pmd_write_unlock(pmd);
2019 void dm_pool_metadata_read_only(struct dm_pool_metadata *pmd)
2021 pmd_write_lock_in_core(pmd);
2022 dm_bm_set_read_only(pmd->bm);
2023 pmd_write_unlock(pmd);
2026 void dm_pool_metadata_read_write(struct dm_pool_metadata *pmd)
2028 pmd_write_lock_in_core(pmd);
2029 dm_bm_set_read_write(pmd->bm);
2030 pmd_write_unlock(pmd);
2033 int dm_pool_register_metadata_threshold(struct dm_pool_metadata *pmd,
2034 dm_block_t threshold,
2035 dm_sm_threshold_fn fn,
2040 pmd_write_lock_in_core(pmd);
2041 r = dm_sm_register_threshold_callback(pmd->metadata_sm, threshold, fn, context);
2042 pmd_write_unlock(pmd);
2047 int dm_pool_metadata_set_needs_check(struct dm_pool_metadata *pmd)
2050 struct dm_block *sblock;
2051 struct thin_disk_superblock *disk_super;
2053 pmd_write_lock(pmd);
2057 pmd->flags |= THIN_METADATA_NEEDS_CHECK_FLAG;
2059 r = superblock_lock(pmd, &sblock);
2061 DMERR("couldn't lock superblock");
2065 disk_super = dm_block_data(sblock);
2066 disk_super->flags = cpu_to_le32(pmd->flags);
2068 dm_bm_unlock(sblock);
2070 pmd_write_unlock(pmd);
2074 bool dm_pool_metadata_needs_check(struct dm_pool_metadata *pmd)
2078 down_read(&pmd->root_lock);
2079 needs_check = pmd->flags & THIN_METADATA_NEEDS_CHECK_FLAG;
2080 up_read(&pmd->root_lock);
2085 void dm_pool_issue_prefetches(struct dm_pool_metadata *pmd)
2087 down_read(&pmd->root_lock);
2089 dm_tm_issue_prefetches(pmd->tm);
2090 up_read(&pmd->root_lock);