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
83 /* This should be plenty */
84 #define SPACE_MAP_ROOT_SIZE 128
87 * Little endian on-disk superblock and device details.
89 struct thin_disk_superblock {
90 __le32 csum; /* Checksum of superblock except for this field. */
92 __le64 blocknr; /* This block number, dm_block_t. */
102 * Root held by userspace transactions.
106 __u8 data_space_map_root[SPACE_MAP_ROOT_SIZE];
107 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
110 * 2-level btree mapping (dev_id, (dev block, time)) -> data block
112 __le64 data_mapping_root;
115 * Device detail root mapping dev_id -> device_details
117 __le64 device_details_root;
119 __le32 data_block_size; /* In 512-byte sectors. */
121 __le32 metadata_block_size; /* In 512-byte sectors. */
122 __le64 metadata_nr_blocks;
125 __le32 compat_ro_flags;
126 __le32 incompat_flags;
129 struct disk_device_details {
130 __le64 mapped_blocks;
131 __le64 transaction_id; /* When created. */
132 __le32 creation_time;
133 __le32 snapshotted_time;
136 struct dm_pool_metadata {
137 struct hlist_node hash;
139 struct block_device *bdev;
140 struct dm_block_manager *bm;
141 struct dm_space_map *metadata_sm;
142 struct dm_space_map *data_sm;
143 struct dm_transaction_manager *tm;
144 struct dm_transaction_manager *nb_tm;
148 * First level holds thin_dev_t.
149 * Second level holds mappings.
151 struct dm_btree_info info;
154 * Non-blocking version of the above.
156 struct dm_btree_info nb_info;
159 * Just the top level for deleting whole devices.
161 struct dm_btree_info tl_info;
164 * Just the bottom level for creating new devices.
166 struct dm_btree_info bl_info;
169 * Describes the device details btree.
171 struct dm_btree_info details_info;
173 struct rw_semaphore root_lock;
177 dm_block_t details_root;
178 struct list_head thin_devices;
181 sector_t data_block_size;
184 struct dm_thin_device {
185 struct list_head list;
186 struct dm_pool_metadata *pmd;
191 uint64_t mapped_blocks;
192 uint64_t transaction_id;
193 uint32_t creation_time;
194 uint32_t snapshotted_time;
197 /*----------------------------------------------------------------
198 * superblock validator
199 *--------------------------------------------------------------*/
201 #define SUPERBLOCK_CSUM_XOR 160774
203 static void sb_prepare_for_write(struct dm_block_validator *v,
207 struct thin_disk_superblock *disk_super = dm_block_data(b);
209 disk_super->blocknr = cpu_to_le64(dm_block_location(b));
210 disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
211 block_size - sizeof(__le32),
212 SUPERBLOCK_CSUM_XOR));
215 static int sb_check(struct dm_block_validator *v,
219 struct thin_disk_superblock *disk_super = dm_block_data(b);
222 if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
223 DMERR("sb_check failed: blocknr %llu: "
224 "wanted %llu", le64_to_cpu(disk_super->blocknr),
225 (unsigned long long)dm_block_location(b));
229 if (le64_to_cpu(disk_super->magic) != THIN_SUPERBLOCK_MAGIC) {
230 DMERR("sb_check failed: magic %llu: "
231 "wanted %llu", le64_to_cpu(disk_super->magic),
232 (unsigned long long)THIN_SUPERBLOCK_MAGIC);
236 csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
237 block_size - sizeof(__le32),
238 SUPERBLOCK_CSUM_XOR));
239 if (csum_le != disk_super->csum) {
240 DMERR("sb_check failed: csum %u: wanted %u",
241 le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
248 static struct dm_block_validator sb_validator = {
249 .name = "superblock",
250 .prepare_for_write = sb_prepare_for_write,
254 /*----------------------------------------------------------------
255 * Methods for the btree value types
256 *--------------------------------------------------------------*/
258 static uint64_t pack_block_time(dm_block_t b, uint32_t t)
260 return (b << 24) | t;
263 static void unpack_block_time(uint64_t v, dm_block_t *b, uint32_t *t)
266 *t = v & ((1 << 24) - 1);
269 static void data_block_inc(void *context, void *value_le)
271 struct dm_space_map *sm = context;
276 memcpy(&v_le, value_le, sizeof(v_le));
277 unpack_block_time(le64_to_cpu(v_le), &b, &t);
278 dm_sm_inc_block(sm, b);
281 static void data_block_dec(void *context, void *value_le)
283 struct dm_space_map *sm = context;
288 memcpy(&v_le, value_le, sizeof(v_le));
289 unpack_block_time(le64_to_cpu(v_le), &b, &t);
290 dm_sm_dec_block(sm, b);
293 static int data_block_equal(void *context, void *value1_le, void *value2_le)
299 memcpy(&v1_le, value1_le, sizeof(v1_le));
300 memcpy(&v2_le, value2_le, sizeof(v2_le));
301 unpack_block_time(le64_to_cpu(v1_le), &b1, &t);
302 unpack_block_time(le64_to_cpu(v2_le), &b2, &t);
307 static void subtree_inc(void *context, void *value)
309 struct dm_btree_info *info = context;
313 memcpy(&root_le, value, sizeof(root_le));
314 root = le64_to_cpu(root_le);
315 dm_tm_inc(info->tm, root);
318 static void subtree_dec(void *context, void *value)
320 struct dm_btree_info *info = context;
324 memcpy(&root_le, value, sizeof(root_le));
325 root = le64_to_cpu(root_le);
326 if (dm_btree_del(info, root))
327 DMERR("btree delete failed\n");
330 static int subtree_equal(void *context, void *value1_le, void *value2_le)
333 memcpy(&v1_le, value1_le, sizeof(v1_le));
334 memcpy(&v2_le, value2_le, sizeof(v2_le));
336 return v1_le == v2_le;
339 /*----------------------------------------------------------------*/
341 static int superblock_all_zeroes(struct dm_block_manager *bm, int *result)
346 __le64 *data_le, zero = cpu_to_le64(0);
347 unsigned block_size = dm_bm_block_size(bm) / sizeof(__le64);
350 * We can't use a validator here - it may be all zeroes.
352 r = dm_bm_read_lock(bm, THIN_SUPERBLOCK_LOCATION, NULL, &b);
356 data_le = dm_block_data(b);
358 for (i = 0; i < block_size; i++) {
359 if (data_le[i] != zero) {
365 return dm_bm_unlock(b);
368 static int init_pmd(struct dm_pool_metadata *pmd,
369 struct dm_block_manager *bm,
370 dm_block_t nr_blocks, int create)
373 struct dm_space_map *sm, *data_sm;
374 struct dm_transaction_manager *tm;
375 struct dm_block *sblock;
378 r = dm_tm_create_with_sm(bm, THIN_SUPERBLOCK_LOCATION,
379 &sb_validator, &tm, &sm, &sblock);
381 DMERR("tm_create_with_sm failed");
385 data_sm = dm_sm_disk_create(tm, nr_blocks);
386 if (IS_ERR(data_sm)) {
387 DMERR("sm_disk_create failed");
388 dm_tm_unlock(tm, sblock);
389 r = PTR_ERR(data_sm);
393 struct thin_disk_superblock *disk_super = NULL;
394 size_t space_map_root_offset =
395 offsetof(struct thin_disk_superblock, metadata_space_map_root);
397 r = dm_tm_open_with_sm(bm, THIN_SUPERBLOCK_LOCATION,
398 &sb_validator, space_map_root_offset,
399 SPACE_MAP_ROOT_SIZE, &tm, &sm, &sblock);
401 DMERR("tm_open_with_sm failed");
405 disk_super = dm_block_data(sblock);
406 data_sm = dm_sm_disk_open(tm, disk_super->data_space_map_root,
407 sizeof(disk_super->data_space_map_root));
408 if (IS_ERR(data_sm)) {
409 DMERR("sm_disk_open failed");
410 r = PTR_ERR(data_sm);
416 r = dm_tm_unlock(tm, sblock);
418 DMERR("couldn't unlock superblock");
423 pmd->metadata_sm = sm;
424 pmd->data_sm = data_sm;
426 pmd->nb_tm = dm_tm_create_non_blocking_clone(tm);
428 DMERR("could not create clone tm");
434 pmd->info.levels = 2;
435 pmd->info.value_type.context = pmd->data_sm;
436 pmd->info.value_type.size = sizeof(__le64);
437 pmd->info.value_type.inc = data_block_inc;
438 pmd->info.value_type.dec = data_block_dec;
439 pmd->info.value_type.equal = data_block_equal;
441 memcpy(&pmd->nb_info, &pmd->info, sizeof(pmd->nb_info));
442 pmd->nb_info.tm = pmd->nb_tm;
444 pmd->tl_info.tm = tm;
445 pmd->tl_info.levels = 1;
446 pmd->tl_info.value_type.context = &pmd->info;
447 pmd->tl_info.value_type.size = sizeof(__le64);
448 pmd->tl_info.value_type.inc = subtree_inc;
449 pmd->tl_info.value_type.dec = subtree_dec;
450 pmd->tl_info.value_type.equal = subtree_equal;
452 pmd->bl_info.tm = tm;
453 pmd->bl_info.levels = 1;
454 pmd->bl_info.value_type.context = pmd->data_sm;
455 pmd->bl_info.value_type.size = sizeof(__le64);
456 pmd->bl_info.value_type.inc = data_block_inc;
457 pmd->bl_info.value_type.dec = data_block_dec;
458 pmd->bl_info.value_type.equal = data_block_equal;
460 pmd->details_info.tm = tm;
461 pmd->details_info.levels = 1;
462 pmd->details_info.value_type.context = NULL;
463 pmd->details_info.value_type.size = sizeof(struct disk_device_details);
464 pmd->details_info.value_type.inc = NULL;
465 pmd->details_info.value_type.dec = NULL;
466 pmd->details_info.value_type.equal = NULL;
470 init_rwsem(&pmd->root_lock);
472 pmd->need_commit = 0;
473 pmd->details_root = 0;
476 INIT_LIST_HEAD(&pmd->thin_devices);
481 dm_sm_destroy(data_sm);
489 static int __begin_transaction(struct dm_pool_metadata *pmd)
493 struct thin_disk_superblock *disk_super;
494 struct dm_block *sblock;
497 * __maybe_commit_transaction() resets these
499 WARN_ON(pmd->need_commit);
502 * We re-read the superblock every time. Shouldn't need to do this
505 r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
506 &sb_validator, &sblock);
510 disk_super = dm_block_data(sblock);
511 pmd->time = le32_to_cpu(disk_super->time);
512 pmd->root = le64_to_cpu(disk_super->data_mapping_root);
513 pmd->details_root = le64_to_cpu(disk_super->device_details_root);
514 pmd->trans_id = le64_to_cpu(disk_super->trans_id);
515 pmd->flags = le32_to_cpu(disk_super->flags);
516 pmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
518 features = le32_to_cpu(disk_super->incompat_flags) & ~THIN_FEATURE_INCOMPAT_SUPP;
520 DMERR("could not access metadata due to "
521 "unsupported optional features (%lx).",
522 (unsigned long)features);
528 * Check for read-only metadata to skip the following RDWR checks.
530 if (get_disk_ro(pmd->bdev->bd_disk))
533 features = le32_to_cpu(disk_super->compat_ro_flags) & ~THIN_FEATURE_COMPAT_RO_SUPP;
535 DMERR("could not access metadata RDWR due to "
536 "unsupported optional features (%lx).",
537 (unsigned long)features);
542 dm_bm_unlock(sblock);
546 static int __write_changed_details(struct dm_pool_metadata *pmd)
549 struct dm_thin_device *td, *tmp;
550 struct disk_device_details details;
553 list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
559 details.mapped_blocks = cpu_to_le64(td->mapped_blocks);
560 details.transaction_id = cpu_to_le64(td->transaction_id);
561 details.creation_time = cpu_to_le32(td->creation_time);
562 details.snapshotted_time = cpu_to_le32(td->snapshotted_time);
563 __dm_bless_for_disk(&details);
565 r = dm_btree_insert(&pmd->details_info, pmd->details_root,
566 &key, &details, &pmd->details_root);
577 pmd->need_commit = 1;
583 static int __commit_transaction(struct dm_pool_metadata *pmd)
586 * FIXME: Associated pool should be made read-only on failure.
589 size_t metadata_len, data_len;
590 struct thin_disk_superblock *disk_super;
591 struct dm_block *sblock;
594 * We need to know if the thin_disk_superblock exceeds a 512-byte sector.
596 BUILD_BUG_ON(sizeof(struct thin_disk_superblock) > 512);
598 r = __write_changed_details(pmd);
602 if (!pmd->need_commit)
605 r = dm_sm_commit(pmd->data_sm);
609 r = dm_tm_pre_commit(pmd->tm);
613 r = dm_sm_root_size(pmd->metadata_sm, &metadata_len);
617 r = dm_sm_root_size(pmd->metadata_sm, &data_len);
621 r = dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
622 &sb_validator, &sblock);
626 disk_super = dm_block_data(sblock);
627 disk_super->time = cpu_to_le32(pmd->time);
628 disk_super->data_mapping_root = cpu_to_le64(pmd->root);
629 disk_super->device_details_root = cpu_to_le64(pmd->details_root);
630 disk_super->trans_id = cpu_to_le64(pmd->trans_id);
631 disk_super->flags = cpu_to_le32(pmd->flags);
633 r = dm_sm_copy_root(pmd->metadata_sm, &disk_super->metadata_space_map_root,
638 r = dm_sm_copy_root(pmd->data_sm, &disk_super->data_space_map_root,
643 r = dm_tm_commit(pmd->tm, sblock);
645 pmd->need_commit = 0;
651 dm_bm_unlock(sblock);
655 struct dm_pool_metadata *dm_pool_metadata_open(struct block_device *bdev,
656 sector_t data_block_size)
659 struct thin_disk_superblock *disk_super;
660 struct dm_pool_metadata *pmd;
661 sector_t bdev_size = i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
662 struct dm_block_manager *bm;
664 struct dm_block *sblock;
666 pmd = kmalloc(sizeof(*pmd), GFP_KERNEL);
668 DMERR("could not allocate metadata struct");
669 return ERR_PTR(-ENOMEM);
674 * 3 for btree insert +
675 * 2 for btree lookup used within space map
677 bm = dm_block_manager_create(bdev, THIN_METADATA_BLOCK_SIZE,
678 THIN_METADATA_CACHE_SIZE, 5);
680 DMERR("could not create block manager");
682 return ERR_PTR(-ENOMEM);
685 r = superblock_all_zeroes(bm, &create);
687 dm_block_manager_destroy(bm);
693 r = init_pmd(pmd, bm, 0, create);
695 dm_block_manager_destroy(bm);
702 r = __begin_transaction(pmd);
711 r = dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
712 &sb_validator, &sblock);
716 disk_super = dm_block_data(sblock);
717 disk_super->magic = cpu_to_le64(THIN_SUPERBLOCK_MAGIC);
718 disk_super->version = cpu_to_le32(THIN_VERSION);
719 disk_super->time = 0;
720 disk_super->metadata_block_size = cpu_to_le32(THIN_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
721 disk_super->metadata_nr_blocks = cpu_to_le64(bdev_size >> SECTOR_TO_BLOCK_SHIFT);
722 disk_super->data_block_size = cpu_to_le32(data_block_size);
724 r = dm_bm_unlock(sblock);
728 r = dm_btree_empty(&pmd->info, &pmd->root);
732 r = dm_btree_empty(&pmd->details_info, &pmd->details_root);
734 DMERR("couldn't create devices root");
739 pmd->need_commit = 1;
740 r = dm_pool_commit_metadata(pmd);
742 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
750 if (dm_pool_metadata_close(pmd) < 0)
751 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
755 int dm_pool_metadata_close(struct dm_pool_metadata *pmd)
758 unsigned open_devices = 0;
759 struct dm_thin_device *td, *tmp;
761 down_read(&pmd->root_lock);
762 list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
770 up_read(&pmd->root_lock);
773 DMERR("attempt to close pmd when %u device(s) are still open",
778 r = __commit_transaction(pmd);
780 DMWARN("%s: __commit_transaction() failed, error = %d",
783 dm_tm_destroy(pmd->tm);
784 dm_tm_destroy(pmd->nb_tm);
785 dm_block_manager_destroy(pmd->bm);
786 dm_sm_destroy(pmd->metadata_sm);
787 dm_sm_destroy(pmd->data_sm);
794 * __open_device: Returns @td corresponding to device with id @dev,
795 * creating it if @create is set and incrementing @td->open_count.
796 * On failure, @td is undefined.
798 static int __open_device(struct dm_pool_metadata *pmd,
799 dm_thin_id dev, int create,
800 struct dm_thin_device **td)
803 struct dm_thin_device *td2;
805 struct disk_device_details details_le;
808 * If the device is already open, return it.
810 list_for_each_entry(td2, &pmd->thin_devices, list)
811 if (td2->id == dev) {
813 * May not create an already-open device.
824 * Check the device exists.
826 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
829 if (r != -ENODATA || !create)
836 details_le.mapped_blocks = 0;
837 details_le.transaction_id = cpu_to_le64(pmd->trans_id);
838 details_le.creation_time = cpu_to_le32(pmd->time);
839 details_le.snapshotted_time = cpu_to_le32(pmd->time);
842 *td = kmalloc(sizeof(**td), GFP_NOIO);
848 (*td)->open_count = 1;
849 (*td)->changed = changed;
850 (*td)->mapped_blocks = le64_to_cpu(details_le.mapped_blocks);
851 (*td)->transaction_id = le64_to_cpu(details_le.transaction_id);
852 (*td)->creation_time = le32_to_cpu(details_le.creation_time);
853 (*td)->snapshotted_time = le32_to_cpu(details_le.snapshotted_time);
855 list_add(&(*td)->list, &pmd->thin_devices);
860 static void __close_device(struct dm_thin_device *td)
865 static int __create_thin(struct dm_pool_metadata *pmd,
871 struct disk_device_details details_le;
872 struct dm_thin_device *td;
875 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
881 * Create an empty btree for the mappings.
883 r = dm_btree_empty(&pmd->bl_info, &dev_root);
888 * Insert it into the main mapping tree.
890 value = cpu_to_le64(dev_root);
891 __dm_bless_for_disk(&value);
892 r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
894 dm_btree_del(&pmd->bl_info, dev_root);
898 r = __open_device(pmd, dev, 1, &td);
900 dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
901 dm_btree_del(&pmd->bl_info, dev_root);
909 int dm_pool_create_thin(struct dm_pool_metadata *pmd, dm_thin_id dev)
913 down_write(&pmd->root_lock);
914 r = __create_thin(pmd, dev);
915 up_write(&pmd->root_lock);
920 static int __set_snapshot_details(struct dm_pool_metadata *pmd,
921 struct dm_thin_device *snap,
922 dm_thin_id origin, uint32_t time)
925 struct dm_thin_device *td;
927 r = __open_device(pmd, origin, 0, &td);
932 td->snapshotted_time = time;
934 snap->mapped_blocks = td->mapped_blocks;
935 snap->snapshotted_time = time;
941 static int __create_snap(struct dm_pool_metadata *pmd,
942 dm_thin_id dev, dm_thin_id origin)
945 dm_block_t origin_root;
946 uint64_t key = origin, dev_key = dev;
947 struct dm_thin_device *td;
948 struct disk_device_details details_le;
951 /* check this device is unused */
952 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
953 &dev_key, &details_le);
957 /* find the mapping tree for the origin */
958 r = dm_btree_lookup(&pmd->tl_info, pmd->root, &key, &value);
961 origin_root = le64_to_cpu(value);
963 /* clone the origin, an inc will do */
964 dm_tm_inc(pmd->tm, origin_root);
966 /* insert into the main mapping tree */
967 value = cpu_to_le64(origin_root);
968 __dm_bless_for_disk(&value);
970 r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
972 dm_tm_dec(pmd->tm, origin_root);
978 r = __open_device(pmd, dev, 1, &td);
982 r = __set_snapshot_details(pmd, td, origin, pmd->time);
991 dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
992 dm_btree_remove(&pmd->details_info, pmd->details_root,
993 &key, &pmd->details_root);
997 int dm_pool_create_snap(struct dm_pool_metadata *pmd,
1003 down_write(&pmd->root_lock);
1004 r = __create_snap(pmd, dev, origin);
1005 up_write(&pmd->root_lock);
1010 static int __delete_device(struct dm_pool_metadata *pmd, dm_thin_id dev)
1014 struct dm_thin_device *td;
1016 /* TODO: failure should mark the transaction invalid */
1017 r = __open_device(pmd, dev, 0, &td);
1021 if (td->open_count > 1) {
1026 list_del(&td->list);
1028 r = dm_btree_remove(&pmd->details_info, pmd->details_root,
1029 &key, &pmd->details_root);
1033 r = dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1037 pmd->need_commit = 1;
1042 int dm_pool_delete_thin_device(struct dm_pool_metadata *pmd,
1047 down_write(&pmd->root_lock);
1048 r = __delete_device(pmd, dev);
1049 up_write(&pmd->root_lock);
1054 int dm_pool_set_metadata_transaction_id(struct dm_pool_metadata *pmd,
1055 uint64_t current_id,
1058 down_write(&pmd->root_lock);
1059 if (pmd->trans_id != current_id) {
1060 up_write(&pmd->root_lock);
1061 DMERR("mismatched transaction id");
1065 pmd->trans_id = new_id;
1066 pmd->need_commit = 1;
1067 up_write(&pmd->root_lock);
1072 int dm_pool_get_metadata_transaction_id(struct dm_pool_metadata *pmd,
1075 down_read(&pmd->root_lock);
1076 *result = pmd->trans_id;
1077 up_read(&pmd->root_lock);
1082 static int __get_held_metadata_root(struct dm_pool_metadata *pmd,
1086 struct thin_disk_superblock *disk_super;
1087 struct dm_block *sblock;
1089 r = dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
1090 &sb_validator, &sblock);
1094 disk_super = dm_block_data(sblock);
1095 *result = le64_to_cpu(disk_super->held_root);
1097 return dm_bm_unlock(sblock);
1100 int dm_pool_get_held_metadata_root(struct dm_pool_metadata *pmd,
1105 down_read(&pmd->root_lock);
1106 r = __get_held_metadata_root(pmd, result);
1107 up_read(&pmd->root_lock);
1112 int dm_pool_open_thin_device(struct dm_pool_metadata *pmd, dm_thin_id dev,
1113 struct dm_thin_device **td)
1117 down_write(&pmd->root_lock);
1118 r = __open_device(pmd, dev, 0, td);
1119 up_write(&pmd->root_lock);
1124 int dm_pool_close_thin_device(struct dm_thin_device *td)
1126 down_write(&td->pmd->root_lock);
1128 up_write(&td->pmd->root_lock);
1133 dm_thin_id dm_thin_dev_id(struct dm_thin_device *td)
1138 static int __snapshotted_since(struct dm_thin_device *td, uint32_t time)
1140 return td->snapshotted_time > time;
1143 int dm_thin_find_block(struct dm_thin_device *td, dm_block_t block,
1144 int can_block, struct dm_thin_lookup_result *result)
1147 uint64_t block_time = 0;
1149 struct dm_pool_metadata *pmd = td->pmd;
1150 dm_block_t keys[2] = { td->id, block };
1153 down_read(&pmd->root_lock);
1154 r = dm_btree_lookup(&pmd->info, pmd->root, keys, &value);
1156 block_time = le64_to_cpu(value);
1157 up_read(&pmd->root_lock);
1159 } else if (down_read_trylock(&pmd->root_lock)) {
1160 r = dm_btree_lookup(&pmd->nb_info, pmd->root, keys, &value);
1162 block_time = le64_to_cpu(value);
1163 up_read(&pmd->root_lock);
1166 return -EWOULDBLOCK;
1169 dm_block_t exception_block;
1170 uint32_t exception_time;
1171 unpack_block_time(block_time, &exception_block,
1173 result->block = exception_block;
1174 result->shared = __snapshotted_since(td, exception_time);
1180 static int __insert(struct dm_thin_device *td, dm_block_t block,
1181 dm_block_t data_block)
1185 struct dm_pool_metadata *pmd = td->pmd;
1186 dm_block_t keys[2] = { td->id, block };
1188 pmd->need_commit = 1;
1189 value = cpu_to_le64(pack_block_time(data_block, pmd->time));
1190 __dm_bless_for_disk(&value);
1192 r = dm_btree_insert_notify(&pmd->info, pmd->root, keys, &value,
1193 &pmd->root, &inserted);
1198 td->mapped_blocks++;
1205 int dm_thin_insert_block(struct dm_thin_device *td, dm_block_t block,
1206 dm_block_t data_block)
1210 down_write(&td->pmd->root_lock);
1211 r = __insert(td, block, data_block);
1212 up_write(&td->pmd->root_lock);
1217 static int __remove(struct dm_thin_device *td, dm_block_t block)
1220 struct dm_pool_metadata *pmd = td->pmd;
1221 dm_block_t keys[2] = { td->id, block };
1223 r = dm_btree_remove(&pmd->info, pmd->root, keys, &pmd->root);
1227 td->mapped_blocks--;
1229 pmd->need_commit = 1;
1234 int dm_thin_remove_block(struct dm_thin_device *td, dm_block_t block)
1238 down_write(&td->pmd->root_lock);
1239 r = __remove(td, block);
1240 up_write(&td->pmd->root_lock);
1245 int dm_pool_alloc_data_block(struct dm_pool_metadata *pmd, dm_block_t *result)
1249 down_write(&pmd->root_lock);
1251 r = dm_sm_new_block(pmd->data_sm, result);
1252 pmd->need_commit = 1;
1254 up_write(&pmd->root_lock);
1259 int dm_pool_commit_metadata(struct dm_pool_metadata *pmd)
1263 down_write(&pmd->root_lock);
1265 r = __commit_transaction(pmd);
1270 * Open the next transaction.
1272 r = __begin_transaction(pmd);
1274 up_write(&pmd->root_lock);
1278 int dm_pool_get_free_block_count(struct dm_pool_metadata *pmd, dm_block_t *result)
1282 down_read(&pmd->root_lock);
1283 r = dm_sm_get_nr_free(pmd->data_sm, result);
1284 up_read(&pmd->root_lock);
1289 int dm_pool_get_free_metadata_block_count(struct dm_pool_metadata *pmd,
1294 down_read(&pmd->root_lock);
1295 r = dm_sm_get_nr_free(pmd->metadata_sm, result);
1296 up_read(&pmd->root_lock);
1301 int dm_pool_get_metadata_dev_size(struct dm_pool_metadata *pmd,
1306 down_read(&pmd->root_lock);
1307 r = dm_sm_get_nr_blocks(pmd->metadata_sm, result);
1308 up_read(&pmd->root_lock);
1313 int dm_pool_get_data_block_size(struct dm_pool_metadata *pmd, sector_t *result)
1315 down_read(&pmd->root_lock);
1316 *result = pmd->data_block_size;
1317 up_read(&pmd->root_lock);
1322 int dm_pool_get_data_dev_size(struct dm_pool_metadata *pmd, dm_block_t *result)
1326 down_read(&pmd->root_lock);
1327 r = dm_sm_get_nr_blocks(pmd->data_sm, result);
1328 up_read(&pmd->root_lock);
1333 int dm_thin_get_mapped_count(struct dm_thin_device *td, dm_block_t *result)
1335 struct dm_pool_metadata *pmd = td->pmd;
1337 down_read(&pmd->root_lock);
1338 *result = td->mapped_blocks;
1339 up_read(&pmd->root_lock);
1344 static int __highest_block(struct dm_thin_device *td, dm_block_t *result)
1348 dm_block_t thin_root;
1349 struct dm_pool_metadata *pmd = td->pmd;
1351 r = dm_btree_lookup(&pmd->tl_info, pmd->root, &td->id, &value_le);
1355 thin_root = le64_to_cpu(value_le);
1357 return dm_btree_find_highest_key(&pmd->bl_info, thin_root, result);
1360 int dm_thin_get_highest_mapped_block(struct dm_thin_device *td,
1364 struct dm_pool_metadata *pmd = td->pmd;
1366 down_read(&pmd->root_lock);
1367 r = __highest_block(td, result);
1368 up_read(&pmd->root_lock);
1373 static int __resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1376 dm_block_t old_count;
1378 r = dm_sm_get_nr_blocks(pmd->data_sm, &old_count);
1382 if (new_count == old_count)
1385 if (new_count < old_count) {
1386 DMERR("cannot reduce size of data device");
1390 r = dm_sm_extend(pmd->data_sm, new_count - old_count);
1392 pmd->need_commit = 1;
1397 int dm_pool_resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1401 down_write(&pmd->root_lock);
1402 r = __resize_data_dev(pmd, new_count);
1403 up_write(&pmd->root_lock);