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;
183 dm_block_t details_root;
184 struct list_head thin_devices;
187 sector_t data_block_size;
190 struct dm_thin_device {
191 struct list_head list;
192 struct dm_pool_metadata *pmd;
197 uint64_t mapped_blocks;
198 uint64_t transaction_id;
199 uint32_t creation_time;
200 uint32_t snapshotted_time;
203 /*----------------------------------------------------------------
204 * superblock validator
205 *--------------------------------------------------------------*/
207 #define SUPERBLOCK_CSUM_XOR 160774
209 static void sb_prepare_for_write(struct dm_block_validator *v,
213 struct thin_disk_superblock *disk_super = dm_block_data(b);
215 disk_super->blocknr = cpu_to_le64(dm_block_location(b));
216 disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
217 block_size - sizeof(__le32),
218 SUPERBLOCK_CSUM_XOR));
221 static int sb_check(struct dm_block_validator *v,
225 struct thin_disk_superblock *disk_super = dm_block_data(b);
228 if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
229 DMERR("sb_check failed: blocknr %llu: "
230 "wanted %llu", le64_to_cpu(disk_super->blocknr),
231 (unsigned long long)dm_block_location(b));
235 if (le64_to_cpu(disk_super->magic) != THIN_SUPERBLOCK_MAGIC) {
236 DMERR("sb_check failed: magic %llu: "
237 "wanted %llu", le64_to_cpu(disk_super->magic),
238 (unsigned long long)THIN_SUPERBLOCK_MAGIC);
242 csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
243 block_size - sizeof(__le32),
244 SUPERBLOCK_CSUM_XOR));
245 if (csum_le != disk_super->csum) {
246 DMERR("sb_check failed: csum %u: wanted %u",
247 le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
254 static struct dm_block_validator sb_validator = {
255 .name = "superblock",
256 .prepare_for_write = sb_prepare_for_write,
260 /*----------------------------------------------------------------
261 * Methods for the btree value types
262 *--------------------------------------------------------------*/
264 static uint64_t pack_block_time(dm_block_t b, uint32_t t)
266 return (b << 24) | t;
269 static void unpack_block_time(uint64_t v, dm_block_t *b, uint32_t *t)
272 *t = v & ((1 << 24) - 1);
275 static void data_block_inc(void *context, void *value_le)
277 struct dm_space_map *sm = context;
282 memcpy(&v_le, value_le, sizeof(v_le));
283 unpack_block_time(le64_to_cpu(v_le), &b, &t);
284 dm_sm_inc_block(sm, b);
287 static void data_block_dec(void *context, void *value_le)
289 struct dm_space_map *sm = context;
294 memcpy(&v_le, value_le, sizeof(v_le));
295 unpack_block_time(le64_to_cpu(v_le), &b, &t);
296 dm_sm_dec_block(sm, b);
299 static int data_block_equal(void *context, void *value1_le, void *value2_le)
305 memcpy(&v1_le, value1_le, sizeof(v1_le));
306 memcpy(&v2_le, value2_le, sizeof(v2_le));
307 unpack_block_time(le64_to_cpu(v1_le), &b1, &t);
308 unpack_block_time(le64_to_cpu(v2_le), &b2, &t);
313 static void subtree_inc(void *context, void *value)
315 struct dm_btree_info *info = context;
319 memcpy(&root_le, value, sizeof(root_le));
320 root = le64_to_cpu(root_le);
321 dm_tm_inc(info->tm, root);
324 static void subtree_dec(void *context, void *value)
326 struct dm_btree_info *info = context;
330 memcpy(&root_le, value, sizeof(root_le));
331 root = le64_to_cpu(root_le);
332 if (dm_btree_del(info, root))
333 DMERR("btree delete failed\n");
336 static int subtree_equal(void *context, void *value1_le, void *value2_le)
339 memcpy(&v1_le, value1_le, sizeof(v1_le));
340 memcpy(&v2_le, value2_le, sizeof(v2_le));
342 return v1_le == v2_le;
345 /*----------------------------------------------------------------*/
347 static int superblock_all_zeroes(struct dm_block_manager *bm, int *result)
352 __le64 *data_le, zero = cpu_to_le64(0);
353 unsigned block_size = dm_bm_block_size(bm) / sizeof(__le64);
356 * We can't use a validator here - it may be all zeroes.
358 r = dm_bm_read_lock(bm, THIN_SUPERBLOCK_LOCATION, NULL, &b);
362 data_le = dm_block_data(b);
364 for (i = 0; i < block_size; i++) {
365 if (data_le[i] != zero) {
371 return dm_bm_unlock(b);
374 static int init_pmd(struct dm_pool_metadata *pmd,
375 struct dm_block_manager *bm,
376 dm_block_t nr_blocks, int create)
379 struct dm_space_map *sm, *data_sm;
380 struct dm_transaction_manager *tm;
381 struct dm_block *sblock;
384 r = dm_tm_create_with_sm(bm, THIN_SUPERBLOCK_LOCATION,
385 &sb_validator, &tm, &sm, &sblock);
387 DMERR("tm_create_with_sm failed");
391 data_sm = dm_sm_disk_create(tm, nr_blocks);
392 if (IS_ERR(data_sm)) {
393 DMERR("sm_disk_create failed");
394 dm_tm_unlock(tm, sblock);
395 r = PTR_ERR(data_sm);
399 struct thin_disk_superblock *disk_super = NULL;
400 size_t space_map_root_offset =
401 offsetof(struct thin_disk_superblock, metadata_space_map_root);
403 r = dm_tm_open_with_sm(bm, THIN_SUPERBLOCK_LOCATION,
404 &sb_validator, space_map_root_offset,
405 SPACE_MAP_ROOT_SIZE, &tm, &sm, &sblock);
407 DMERR("tm_open_with_sm failed");
411 disk_super = dm_block_data(sblock);
412 data_sm = dm_sm_disk_open(tm, disk_super->data_space_map_root,
413 sizeof(disk_super->data_space_map_root));
414 if (IS_ERR(data_sm)) {
415 DMERR("sm_disk_open failed");
416 r = PTR_ERR(data_sm);
422 r = dm_tm_unlock(tm, sblock);
424 DMERR("couldn't unlock superblock");
429 pmd->metadata_sm = sm;
430 pmd->data_sm = data_sm;
432 pmd->nb_tm = dm_tm_create_non_blocking_clone(tm);
434 DMERR("could not create clone tm");
440 pmd->info.levels = 2;
441 pmd->info.value_type.context = pmd->data_sm;
442 pmd->info.value_type.size = sizeof(__le64);
443 pmd->info.value_type.inc = data_block_inc;
444 pmd->info.value_type.dec = data_block_dec;
445 pmd->info.value_type.equal = data_block_equal;
447 memcpy(&pmd->nb_info, &pmd->info, sizeof(pmd->nb_info));
448 pmd->nb_info.tm = pmd->nb_tm;
450 pmd->tl_info.tm = tm;
451 pmd->tl_info.levels = 1;
452 pmd->tl_info.value_type.context = &pmd->info;
453 pmd->tl_info.value_type.size = sizeof(__le64);
454 pmd->tl_info.value_type.inc = subtree_inc;
455 pmd->tl_info.value_type.dec = subtree_dec;
456 pmd->tl_info.value_type.equal = subtree_equal;
458 pmd->bl_info.tm = tm;
459 pmd->bl_info.levels = 1;
460 pmd->bl_info.value_type.context = pmd->data_sm;
461 pmd->bl_info.value_type.size = sizeof(__le64);
462 pmd->bl_info.value_type.inc = data_block_inc;
463 pmd->bl_info.value_type.dec = data_block_dec;
464 pmd->bl_info.value_type.equal = data_block_equal;
466 pmd->details_info.tm = tm;
467 pmd->details_info.levels = 1;
468 pmd->details_info.value_type.context = NULL;
469 pmd->details_info.value_type.size = sizeof(struct disk_device_details);
470 pmd->details_info.value_type.inc = NULL;
471 pmd->details_info.value_type.dec = NULL;
472 pmd->details_info.value_type.equal = NULL;
476 init_rwsem(&pmd->root_lock);
478 pmd->need_commit = 0;
479 pmd->details_root = 0;
482 INIT_LIST_HEAD(&pmd->thin_devices);
487 dm_sm_destroy(data_sm);
495 static int __begin_transaction(struct dm_pool_metadata *pmd)
499 struct thin_disk_superblock *disk_super;
500 struct dm_block *sblock;
503 * __maybe_commit_transaction() resets these
505 WARN_ON(pmd->need_commit);
508 * We re-read the superblock every time. Shouldn't need to do this
511 r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
512 &sb_validator, &sblock);
516 disk_super = dm_block_data(sblock);
517 pmd->time = le32_to_cpu(disk_super->time);
518 pmd->root = le64_to_cpu(disk_super->data_mapping_root);
519 pmd->details_root = le64_to_cpu(disk_super->device_details_root);
520 pmd->trans_id = le64_to_cpu(disk_super->trans_id);
521 pmd->flags = le32_to_cpu(disk_super->flags);
522 pmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
524 features = le32_to_cpu(disk_super->incompat_flags) & ~THIN_FEATURE_INCOMPAT_SUPP;
526 DMERR("could not access metadata due to "
527 "unsupported optional features (%lx).",
528 (unsigned long)features);
534 * Check for read-only metadata to skip the following RDWR checks.
536 if (get_disk_ro(pmd->bdev->bd_disk))
539 features = le32_to_cpu(disk_super->compat_ro_flags) & ~THIN_FEATURE_COMPAT_RO_SUPP;
541 DMERR("could not access metadata RDWR due to "
542 "unsupported optional features (%lx).",
543 (unsigned long)features);
548 dm_bm_unlock(sblock);
552 static int __write_changed_details(struct dm_pool_metadata *pmd)
555 struct dm_thin_device *td, *tmp;
556 struct disk_device_details details;
559 list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
565 details.mapped_blocks = cpu_to_le64(td->mapped_blocks);
566 details.transaction_id = cpu_to_le64(td->transaction_id);
567 details.creation_time = cpu_to_le32(td->creation_time);
568 details.snapshotted_time = cpu_to_le32(td->snapshotted_time);
569 __dm_bless_for_disk(&details);
571 r = dm_btree_insert(&pmd->details_info, pmd->details_root,
572 &key, &details, &pmd->details_root);
583 pmd->need_commit = 1;
589 static int __commit_transaction(struct dm_pool_metadata *pmd)
592 * FIXME: Associated pool should be made read-only on failure.
595 size_t metadata_len, data_len;
596 struct thin_disk_superblock *disk_super;
597 struct dm_block *sblock;
600 * We need to know if the thin_disk_superblock exceeds a 512-byte sector.
602 BUILD_BUG_ON(sizeof(struct thin_disk_superblock) > 512);
604 r = __write_changed_details(pmd);
608 if (!pmd->need_commit)
611 r = dm_sm_commit(pmd->data_sm);
615 r = dm_tm_pre_commit(pmd->tm);
619 r = dm_sm_root_size(pmd->metadata_sm, &metadata_len);
623 r = dm_sm_root_size(pmd->data_sm, &data_len);
627 r = dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
628 &sb_validator, &sblock);
632 disk_super = dm_block_data(sblock);
633 disk_super->time = cpu_to_le32(pmd->time);
634 disk_super->data_mapping_root = cpu_to_le64(pmd->root);
635 disk_super->device_details_root = cpu_to_le64(pmd->details_root);
636 disk_super->trans_id = cpu_to_le64(pmd->trans_id);
637 disk_super->flags = cpu_to_le32(pmd->flags);
639 r = dm_sm_copy_root(pmd->metadata_sm, &disk_super->metadata_space_map_root,
644 r = dm_sm_copy_root(pmd->data_sm, &disk_super->data_space_map_root,
649 r = dm_tm_commit(pmd->tm, sblock);
651 pmd->need_commit = 0;
656 dm_bm_unlock(sblock);
660 struct dm_pool_metadata *dm_pool_metadata_open(struct block_device *bdev,
661 sector_t data_block_size)
664 struct thin_disk_superblock *disk_super;
665 struct dm_pool_metadata *pmd;
666 sector_t bdev_size = i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
667 struct dm_block_manager *bm;
669 struct dm_block *sblock;
671 pmd = kmalloc(sizeof(*pmd), GFP_KERNEL);
673 DMERR("could not allocate metadata struct");
674 return ERR_PTR(-ENOMEM);
677 bm = dm_block_manager_create(bdev, THIN_METADATA_BLOCK_SIZE,
678 THIN_METADATA_CACHE_SIZE,
679 THIN_MAX_CONCURRENT_LOCKS);
681 DMERR("could not create block manager");
683 return ERR_PTR(-ENOMEM);
686 r = superblock_all_zeroes(bm, &create);
688 dm_block_manager_destroy(bm);
694 r = init_pmd(pmd, bm, 0, create);
696 dm_block_manager_destroy(bm);
703 r = __begin_transaction(pmd);
712 r = dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
713 &sb_validator, &sblock);
717 if (bdev_size > THIN_METADATA_MAX_SECTORS)
718 bdev_size = THIN_METADATA_MAX_SECTORS;
720 disk_super = dm_block_data(sblock);
721 disk_super->magic = cpu_to_le64(THIN_SUPERBLOCK_MAGIC);
722 disk_super->version = cpu_to_le32(THIN_VERSION);
723 disk_super->time = 0;
724 disk_super->metadata_block_size = cpu_to_le32(THIN_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
725 disk_super->metadata_nr_blocks = cpu_to_le64(bdev_size >> SECTOR_TO_BLOCK_SHIFT);
726 disk_super->data_block_size = cpu_to_le32(data_block_size);
728 r = dm_bm_unlock(sblock);
732 r = dm_btree_empty(&pmd->info, &pmd->root);
736 r = dm_btree_empty(&pmd->details_info, &pmd->details_root);
738 DMERR("couldn't create devices root");
743 pmd->need_commit = 1;
744 r = dm_pool_commit_metadata(pmd);
746 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
754 if (dm_pool_metadata_close(pmd) < 0)
755 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
759 int dm_pool_metadata_close(struct dm_pool_metadata *pmd)
762 unsigned open_devices = 0;
763 struct dm_thin_device *td, *tmp;
765 down_read(&pmd->root_lock);
766 list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
774 up_read(&pmd->root_lock);
777 DMERR("attempt to close pmd when %u device(s) are still open",
782 r = __commit_transaction(pmd);
784 DMWARN("%s: __commit_transaction() failed, error = %d",
787 dm_tm_destroy(pmd->tm);
788 dm_tm_destroy(pmd->nb_tm);
789 dm_block_manager_destroy(pmd->bm);
790 dm_sm_destroy(pmd->metadata_sm);
791 dm_sm_destroy(pmd->data_sm);
798 * __open_device: Returns @td corresponding to device with id @dev,
799 * creating it if @create is set and incrementing @td->open_count.
800 * On failure, @td is undefined.
802 static int __open_device(struct dm_pool_metadata *pmd,
803 dm_thin_id dev, int create,
804 struct dm_thin_device **td)
807 struct dm_thin_device *td2;
809 struct disk_device_details details_le;
812 * If the device is already open, return it.
814 list_for_each_entry(td2, &pmd->thin_devices, list)
815 if (td2->id == dev) {
817 * May not create an already-open device.
828 * Check the device exists.
830 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
833 if (r != -ENODATA || !create)
840 details_le.mapped_blocks = 0;
841 details_le.transaction_id = cpu_to_le64(pmd->trans_id);
842 details_le.creation_time = cpu_to_le32(pmd->time);
843 details_le.snapshotted_time = cpu_to_le32(pmd->time);
846 *td = kmalloc(sizeof(**td), GFP_NOIO);
852 (*td)->open_count = 1;
853 (*td)->changed = changed;
854 (*td)->mapped_blocks = le64_to_cpu(details_le.mapped_blocks);
855 (*td)->transaction_id = le64_to_cpu(details_le.transaction_id);
856 (*td)->creation_time = le32_to_cpu(details_le.creation_time);
857 (*td)->snapshotted_time = le32_to_cpu(details_le.snapshotted_time);
859 list_add(&(*td)->list, &pmd->thin_devices);
864 static void __close_device(struct dm_thin_device *td)
869 static int __create_thin(struct dm_pool_metadata *pmd,
875 struct disk_device_details details_le;
876 struct dm_thin_device *td;
879 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
885 * Create an empty btree for the mappings.
887 r = dm_btree_empty(&pmd->bl_info, &dev_root);
892 * Insert it into the main mapping tree.
894 value = cpu_to_le64(dev_root);
895 __dm_bless_for_disk(&value);
896 r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
898 dm_btree_del(&pmd->bl_info, dev_root);
902 r = __open_device(pmd, dev, 1, &td);
904 dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
905 dm_btree_del(&pmd->bl_info, dev_root);
913 int dm_pool_create_thin(struct dm_pool_metadata *pmd, dm_thin_id dev)
917 down_write(&pmd->root_lock);
918 r = __create_thin(pmd, dev);
919 up_write(&pmd->root_lock);
924 static int __set_snapshot_details(struct dm_pool_metadata *pmd,
925 struct dm_thin_device *snap,
926 dm_thin_id origin, uint32_t time)
929 struct dm_thin_device *td;
931 r = __open_device(pmd, origin, 0, &td);
936 td->snapshotted_time = time;
938 snap->mapped_blocks = td->mapped_blocks;
939 snap->snapshotted_time = time;
945 static int __create_snap(struct dm_pool_metadata *pmd,
946 dm_thin_id dev, dm_thin_id origin)
949 dm_block_t origin_root;
950 uint64_t key = origin, dev_key = dev;
951 struct dm_thin_device *td;
952 struct disk_device_details details_le;
955 /* check this device is unused */
956 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
957 &dev_key, &details_le);
961 /* find the mapping tree for the origin */
962 r = dm_btree_lookup(&pmd->tl_info, pmd->root, &key, &value);
965 origin_root = le64_to_cpu(value);
967 /* clone the origin, an inc will do */
968 dm_tm_inc(pmd->tm, origin_root);
970 /* insert into the main mapping tree */
971 value = cpu_to_le64(origin_root);
972 __dm_bless_for_disk(&value);
974 r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
976 dm_tm_dec(pmd->tm, origin_root);
982 r = __open_device(pmd, dev, 1, &td);
986 r = __set_snapshot_details(pmd, td, origin, pmd->time);
995 dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
996 dm_btree_remove(&pmd->details_info, pmd->details_root,
997 &key, &pmd->details_root);
1001 int dm_pool_create_snap(struct dm_pool_metadata *pmd,
1007 down_write(&pmd->root_lock);
1008 r = __create_snap(pmd, dev, origin);
1009 up_write(&pmd->root_lock);
1014 static int __delete_device(struct dm_pool_metadata *pmd, dm_thin_id dev)
1018 struct dm_thin_device *td;
1020 /* TODO: failure should mark the transaction invalid */
1021 r = __open_device(pmd, dev, 0, &td);
1025 if (td->open_count > 1) {
1030 list_del(&td->list);
1032 r = dm_btree_remove(&pmd->details_info, pmd->details_root,
1033 &key, &pmd->details_root);
1037 r = dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1041 pmd->need_commit = 1;
1046 int dm_pool_delete_thin_device(struct dm_pool_metadata *pmd,
1051 down_write(&pmd->root_lock);
1052 r = __delete_device(pmd, dev);
1053 up_write(&pmd->root_lock);
1058 int dm_pool_set_metadata_transaction_id(struct dm_pool_metadata *pmd,
1059 uint64_t current_id,
1062 down_write(&pmd->root_lock);
1063 if (pmd->trans_id != current_id) {
1064 up_write(&pmd->root_lock);
1065 DMERR("mismatched transaction id");
1069 pmd->trans_id = new_id;
1070 pmd->need_commit = 1;
1071 up_write(&pmd->root_lock);
1076 int dm_pool_get_metadata_transaction_id(struct dm_pool_metadata *pmd,
1079 down_read(&pmd->root_lock);
1080 *result = pmd->trans_id;
1081 up_read(&pmd->root_lock);
1086 static int __reserve_metadata_snap(struct dm_pool_metadata *pmd)
1089 struct thin_disk_superblock *disk_super;
1090 struct dm_block *copy, *sblock;
1091 dm_block_t held_root;
1094 * Copy the superblock.
1096 dm_sm_inc_block(pmd->metadata_sm, THIN_SUPERBLOCK_LOCATION);
1097 r = dm_tm_shadow_block(pmd->tm, THIN_SUPERBLOCK_LOCATION,
1098 &sb_validator, ©, &inc);
1104 held_root = dm_block_location(copy);
1105 disk_super = dm_block_data(copy);
1107 if (le64_to_cpu(disk_super->held_root)) {
1108 DMWARN("Pool metadata snapshot already exists: release this before taking another.");
1110 dm_tm_dec(pmd->tm, held_root);
1111 dm_tm_unlock(pmd->tm, copy);
1112 pmd->need_commit = 1;
1118 * Wipe the spacemap since we're not publishing this.
1120 memset(&disk_super->data_space_map_root, 0,
1121 sizeof(disk_super->data_space_map_root));
1122 memset(&disk_super->metadata_space_map_root, 0,
1123 sizeof(disk_super->metadata_space_map_root));
1126 * Increment the data structures that need to be preserved.
1128 dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->data_mapping_root));
1129 dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->device_details_root));
1130 dm_tm_unlock(pmd->tm, copy);
1133 * Write the held root into the superblock.
1135 r = dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
1136 &sb_validator, &sblock);
1138 dm_tm_dec(pmd->tm, held_root);
1139 pmd->need_commit = 1;
1143 disk_super = dm_block_data(sblock);
1144 disk_super->held_root = cpu_to_le64(held_root);
1145 dm_bm_unlock(sblock);
1147 pmd->need_commit = 1;
1152 int dm_pool_reserve_metadata_snap(struct dm_pool_metadata *pmd)
1156 down_write(&pmd->root_lock);
1157 r = __reserve_metadata_snap(pmd);
1158 up_write(&pmd->root_lock);
1163 static int __release_metadata_snap(struct dm_pool_metadata *pmd)
1166 struct thin_disk_superblock *disk_super;
1167 struct dm_block *sblock, *copy;
1168 dm_block_t held_root;
1170 r = dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
1171 &sb_validator, &sblock);
1175 disk_super = dm_block_data(sblock);
1176 held_root = le64_to_cpu(disk_super->held_root);
1177 disk_super->held_root = cpu_to_le64(0);
1178 pmd->need_commit = 1;
1180 dm_bm_unlock(sblock);
1183 DMWARN("No pool metadata snapshot found: nothing to release.");
1187 r = dm_tm_read_lock(pmd->tm, held_root, &sb_validator, ©);
1191 disk_super = dm_block_data(copy);
1192 dm_sm_dec_block(pmd->metadata_sm, le64_to_cpu(disk_super->data_mapping_root));
1193 dm_sm_dec_block(pmd->metadata_sm, le64_to_cpu(disk_super->device_details_root));
1194 dm_sm_dec_block(pmd->metadata_sm, held_root);
1196 return dm_tm_unlock(pmd->tm, copy);
1199 int dm_pool_release_metadata_snap(struct dm_pool_metadata *pmd)
1203 down_write(&pmd->root_lock);
1204 r = __release_metadata_snap(pmd);
1205 up_write(&pmd->root_lock);
1210 static int __get_metadata_snap(struct dm_pool_metadata *pmd,
1214 struct thin_disk_superblock *disk_super;
1215 struct dm_block *sblock;
1217 r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
1218 &sb_validator, &sblock);
1222 disk_super = dm_block_data(sblock);
1223 *result = le64_to_cpu(disk_super->held_root);
1225 return dm_bm_unlock(sblock);
1228 int dm_pool_get_metadata_snap(struct dm_pool_metadata *pmd,
1233 down_read(&pmd->root_lock);
1234 r = __get_metadata_snap(pmd, result);
1235 up_read(&pmd->root_lock);
1240 int dm_pool_open_thin_device(struct dm_pool_metadata *pmd, dm_thin_id dev,
1241 struct dm_thin_device **td)
1245 down_write(&pmd->root_lock);
1246 r = __open_device(pmd, dev, 0, td);
1247 up_write(&pmd->root_lock);
1252 int dm_pool_close_thin_device(struct dm_thin_device *td)
1254 down_write(&td->pmd->root_lock);
1256 up_write(&td->pmd->root_lock);
1261 dm_thin_id dm_thin_dev_id(struct dm_thin_device *td)
1266 static bool __snapshotted_since(struct dm_thin_device *td, uint32_t time)
1268 return td->snapshotted_time > time;
1271 int dm_thin_find_block(struct dm_thin_device *td, dm_block_t block,
1272 int can_block, struct dm_thin_lookup_result *result)
1275 uint64_t block_time = 0;
1277 struct dm_pool_metadata *pmd = td->pmd;
1278 dm_block_t keys[2] = { td->id, block };
1281 down_read(&pmd->root_lock);
1282 r = dm_btree_lookup(&pmd->info, pmd->root, keys, &value);
1284 block_time = le64_to_cpu(value);
1285 up_read(&pmd->root_lock);
1287 } else if (down_read_trylock(&pmd->root_lock)) {
1288 r = dm_btree_lookup(&pmd->nb_info, pmd->root, keys, &value);
1290 block_time = le64_to_cpu(value);
1291 up_read(&pmd->root_lock);
1294 return -EWOULDBLOCK;
1297 dm_block_t exception_block;
1298 uint32_t exception_time;
1299 unpack_block_time(block_time, &exception_block,
1301 result->block = exception_block;
1302 result->shared = __snapshotted_since(td, exception_time);
1308 static int __insert(struct dm_thin_device *td, dm_block_t block,
1309 dm_block_t data_block)
1313 struct dm_pool_metadata *pmd = td->pmd;
1314 dm_block_t keys[2] = { td->id, block };
1316 pmd->need_commit = 1;
1317 value = cpu_to_le64(pack_block_time(data_block, pmd->time));
1318 __dm_bless_for_disk(&value);
1320 r = dm_btree_insert_notify(&pmd->info, pmd->root, keys, &value,
1321 &pmd->root, &inserted);
1326 td->mapped_blocks++;
1333 int dm_thin_insert_block(struct dm_thin_device *td, dm_block_t block,
1334 dm_block_t data_block)
1338 down_write(&td->pmd->root_lock);
1339 r = __insert(td, block, data_block);
1340 up_write(&td->pmd->root_lock);
1345 static int __remove(struct dm_thin_device *td, dm_block_t block)
1348 struct dm_pool_metadata *pmd = td->pmd;
1349 dm_block_t keys[2] = { td->id, block };
1351 r = dm_btree_remove(&pmd->info, pmd->root, keys, &pmd->root);
1355 td->mapped_blocks--;
1357 pmd->need_commit = 1;
1362 int dm_thin_remove_block(struct dm_thin_device *td, dm_block_t block)
1366 down_write(&td->pmd->root_lock);
1367 r = __remove(td, block);
1368 up_write(&td->pmd->root_lock);
1373 int dm_pool_alloc_data_block(struct dm_pool_metadata *pmd, dm_block_t *result)
1377 down_write(&pmd->root_lock);
1379 r = dm_sm_new_block(pmd->data_sm, result);
1380 pmd->need_commit = 1;
1382 up_write(&pmd->root_lock);
1387 int dm_pool_commit_metadata(struct dm_pool_metadata *pmd)
1391 down_write(&pmd->root_lock);
1393 r = __commit_transaction(pmd);
1398 * Open the next transaction.
1400 r = __begin_transaction(pmd);
1402 up_write(&pmd->root_lock);
1406 int dm_pool_get_free_block_count(struct dm_pool_metadata *pmd, dm_block_t *result)
1410 down_read(&pmd->root_lock);
1411 r = dm_sm_get_nr_free(pmd->data_sm, result);
1412 up_read(&pmd->root_lock);
1417 int dm_pool_get_free_metadata_block_count(struct dm_pool_metadata *pmd,
1422 down_read(&pmd->root_lock);
1423 r = dm_sm_get_nr_free(pmd->metadata_sm, result);
1424 up_read(&pmd->root_lock);
1429 int dm_pool_get_metadata_dev_size(struct dm_pool_metadata *pmd,
1434 down_read(&pmd->root_lock);
1435 r = dm_sm_get_nr_blocks(pmd->metadata_sm, result);
1436 up_read(&pmd->root_lock);
1441 int dm_pool_get_data_block_size(struct dm_pool_metadata *pmd, sector_t *result)
1443 down_read(&pmd->root_lock);
1444 *result = pmd->data_block_size;
1445 up_read(&pmd->root_lock);
1450 int dm_pool_get_data_dev_size(struct dm_pool_metadata *pmd, dm_block_t *result)
1454 down_read(&pmd->root_lock);
1455 r = dm_sm_get_nr_blocks(pmd->data_sm, result);
1456 up_read(&pmd->root_lock);
1461 int dm_thin_get_mapped_count(struct dm_thin_device *td, dm_block_t *result)
1463 struct dm_pool_metadata *pmd = td->pmd;
1465 down_read(&pmd->root_lock);
1466 *result = td->mapped_blocks;
1467 up_read(&pmd->root_lock);
1472 static int __highest_block(struct dm_thin_device *td, dm_block_t *result)
1476 dm_block_t thin_root;
1477 struct dm_pool_metadata *pmd = td->pmd;
1479 r = dm_btree_lookup(&pmd->tl_info, pmd->root, &td->id, &value_le);
1483 thin_root = le64_to_cpu(value_le);
1485 return dm_btree_find_highest_key(&pmd->bl_info, thin_root, result);
1488 int dm_thin_get_highest_mapped_block(struct dm_thin_device *td,
1492 struct dm_pool_metadata *pmd = td->pmd;
1494 down_read(&pmd->root_lock);
1495 r = __highest_block(td, result);
1496 up_read(&pmd->root_lock);
1501 static int __resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1504 dm_block_t old_count;
1506 r = dm_sm_get_nr_blocks(pmd->data_sm, &old_count);
1510 if (new_count == old_count)
1513 if (new_count < old_count) {
1514 DMERR("cannot reduce size of data device");
1518 r = dm_sm_extend(pmd->data_sm, new_count - old_count);
1520 pmd->need_commit = 1;
1525 int dm_pool_resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1529 down_write(&pmd->root_lock);
1530 r = __resize_data_dev(pmd, new_count);
1531 up_write(&pmd->root_lock);