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 * Reading the space map roots can fail, so we read it into these
206 * buffers before the superblock is locked and updated.
208 __u8 data_space_map_root[SPACE_MAP_ROOT_SIZE];
209 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
212 struct dm_thin_device {
213 struct list_head list;
214 struct dm_pool_metadata *pmd;
219 bool aborted_with_changes:1;
220 uint64_t mapped_blocks;
221 uint64_t transaction_id;
222 uint32_t creation_time;
223 uint32_t snapshotted_time;
226 /*----------------------------------------------------------------
227 * superblock validator
228 *--------------------------------------------------------------*/
230 #define SUPERBLOCK_CSUM_XOR 160774
232 static void sb_prepare_for_write(struct dm_block_validator *v,
236 struct thin_disk_superblock *disk_super = dm_block_data(b);
238 disk_super->blocknr = cpu_to_le64(dm_block_location(b));
239 disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
240 block_size - sizeof(__le32),
241 SUPERBLOCK_CSUM_XOR));
244 static int sb_check(struct dm_block_validator *v,
248 struct thin_disk_superblock *disk_super = dm_block_data(b);
251 if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
252 DMERR("sb_check failed: blocknr %llu: "
253 "wanted %llu", le64_to_cpu(disk_super->blocknr),
254 (unsigned long long)dm_block_location(b));
258 if (le64_to_cpu(disk_super->magic) != THIN_SUPERBLOCK_MAGIC) {
259 DMERR("sb_check failed: magic %llu: "
260 "wanted %llu", le64_to_cpu(disk_super->magic),
261 (unsigned long long)THIN_SUPERBLOCK_MAGIC);
265 csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
266 block_size - sizeof(__le32),
267 SUPERBLOCK_CSUM_XOR));
268 if (csum_le != disk_super->csum) {
269 DMERR("sb_check failed: csum %u: wanted %u",
270 le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
277 static struct dm_block_validator sb_validator = {
278 .name = "superblock",
279 .prepare_for_write = sb_prepare_for_write,
283 /*----------------------------------------------------------------
284 * Methods for the btree value types
285 *--------------------------------------------------------------*/
287 static uint64_t pack_block_time(dm_block_t b, uint32_t t)
289 return (b << 24) | t;
292 static void unpack_block_time(uint64_t v, dm_block_t *b, uint32_t *t)
295 *t = v & ((1 << 24) - 1);
298 static void data_block_inc(void *context, const void *value_le)
300 struct dm_space_map *sm = context;
305 memcpy(&v_le, value_le, sizeof(v_le));
306 unpack_block_time(le64_to_cpu(v_le), &b, &t);
307 dm_sm_inc_block(sm, b);
310 static void data_block_dec(void *context, const void *value_le)
312 struct dm_space_map *sm = context;
317 memcpy(&v_le, value_le, sizeof(v_le));
318 unpack_block_time(le64_to_cpu(v_le), &b, &t);
319 dm_sm_dec_block(sm, b);
322 static int data_block_equal(void *context, const void *value1_le, const void *value2_le)
328 memcpy(&v1_le, value1_le, sizeof(v1_le));
329 memcpy(&v2_le, value2_le, sizeof(v2_le));
330 unpack_block_time(le64_to_cpu(v1_le), &b1, &t);
331 unpack_block_time(le64_to_cpu(v2_le), &b2, &t);
336 static void subtree_inc(void *context, const void *value)
338 struct dm_btree_info *info = context;
342 memcpy(&root_le, value, sizeof(root_le));
343 root = le64_to_cpu(root_le);
344 dm_tm_inc(info->tm, root);
347 static void subtree_dec(void *context, const void *value)
349 struct dm_btree_info *info = context;
353 memcpy(&root_le, value, sizeof(root_le));
354 root = le64_to_cpu(root_le);
355 if (dm_btree_del(info, root))
356 DMERR("btree delete failed");
359 static int subtree_equal(void *context, const void *value1_le, const void *value2_le)
362 memcpy(&v1_le, value1_le, sizeof(v1_le));
363 memcpy(&v2_le, value2_le, sizeof(v2_le));
365 return v1_le == v2_le;
368 /*----------------------------------------------------------------*/
370 static int superblock_lock_zero(struct dm_pool_metadata *pmd,
371 struct dm_block **sblock)
373 return dm_bm_write_lock_zero(pmd->bm, THIN_SUPERBLOCK_LOCATION,
374 &sb_validator, sblock);
377 static int superblock_lock(struct dm_pool_metadata *pmd,
378 struct dm_block **sblock)
380 return dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
381 &sb_validator, sblock);
384 static int __superblock_all_zeroes(struct dm_block_manager *bm, int *result)
389 __le64 *data_le, zero = cpu_to_le64(0);
390 unsigned block_size = dm_bm_block_size(bm) / sizeof(__le64);
393 * We can't use a validator here - it may be all zeroes.
395 r = dm_bm_read_lock(bm, THIN_SUPERBLOCK_LOCATION, NULL, &b);
399 data_le = dm_block_data(b);
401 for (i = 0; i < block_size; i++) {
402 if (data_le[i] != zero) {
413 static void __setup_btree_details(struct dm_pool_metadata *pmd)
415 pmd->info.tm = pmd->tm;
416 pmd->info.levels = 2;
417 pmd->info.value_type.context = pmd->data_sm;
418 pmd->info.value_type.size = sizeof(__le64);
419 pmd->info.value_type.inc = data_block_inc;
420 pmd->info.value_type.dec = data_block_dec;
421 pmd->info.value_type.equal = data_block_equal;
423 memcpy(&pmd->nb_info, &pmd->info, sizeof(pmd->nb_info));
424 pmd->nb_info.tm = pmd->nb_tm;
426 pmd->tl_info.tm = pmd->tm;
427 pmd->tl_info.levels = 1;
428 pmd->tl_info.value_type.context = &pmd->bl_info;
429 pmd->tl_info.value_type.size = sizeof(__le64);
430 pmd->tl_info.value_type.inc = subtree_inc;
431 pmd->tl_info.value_type.dec = subtree_dec;
432 pmd->tl_info.value_type.equal = subtree_equal;
434 pmd->bl_info.tm = pmd->tm;
435 pmd->bl_info.levels = 1;
436 pmd->bl_info.value_type.context = pmd->data_sm;
437 pmd->bl_info.value_type.size = sizeof(__le64);
438 pmd->bl_info.value_type.inc = data_block_inc;
439 pmd->bl_info.value_type.dec = data_block_dec;
440 pmd->bl_info.value_type.equal = data_block_equal;
442 pmd->details_info.tm = pmd->tm;
443 pmd->details_info.levels = 1;
444 pmd->details_info.value_type.context = NULL;
445 pmd->details_info.value_type.size = sizeof(struct disk_device_details);
446 pmd->details_info.value_type.inc = NULL;
447 pmd->details_info.value_type.dec = NULL;
448 pmd->details_info.value_type.equal = NULL;
451 static int save_sm_roots(struct dm_pool_metadata *pmd)
456 r = dm_sm_root_size(pmd->metadata_sm, &len);
460 r = dm_sm_copy_root(pmd->metadata_sm, &pmd->metadata_space_map_root, len);
464 r = dm_sm_root_size(pmd->data_sm, &len);
468 return dm_sm_copy_root(pmd->data_sm, &pmd->data_space_map_root, len);
471 static void copy_sm_roots(struct dm_pool_metadata *pmd,
472 struct thin_disk_superblock *disk)
474 memcpy(&disk->metadata_space_map_root,
475 &pmd->metadata_space_map_root,
476 sizeof(pmd->metadata_space_map_root));
478 memcpy(&disk->data_space_map_root,
479 &pmd->data_space_map_root,
480 sizeof(pmd->data_space_map_root));
483 static int __write_initial_superblock(struct dm_pool_metadata *pmd)
486 struct dm_block *sblock;
487 struct thin_disk_superblock *disk_super;
488 sector_t bdev_size = i_size_read(pmd->bdev->bd_inode) >> SECTOR_SHIFT;
490 if (bdev_size > THIN_METADATA_MAX_SECTORS)
491 bdev_size = THIN_METADATA_MAX_SECTORS;
493 r = dm_sm_commit(pmd->data_sm);
497 r = dm_tm_pre_commit(pmd->tm);
501 r = save_sm_roots(pmd);
505 r = superblock_lock_zero(pmd, &sblock);
509 disk_super = dm_block_data(sblock);
510 disk_super->flags = 0;
511 memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
512 disk_super->magic = cpu_to_le64(THIN_SUPERBLOCK_MAGIC);
513 disk_super->version = cpu_to_le32(THIN_VERSION);
514 disk_super->time = 0;
515 disk_super->trans_id = 0;
516 disk_super->held_root = 0;
518 copy_sm_roots(pmd, disk_super);
520 disk_super->data_mapping_root = cpu_to_le64(pmd->root);
521 disk_super->device_details_root = cpu_to_le64(pmd->details_root);
522 disk_super->metadata_block_size = cpu_to_le32(THIN_METADATA_BLOCK_SIZE);
523 disk_super->metadata_nr_blocks = cpu_to_le64(bdev_size >> SECTOR_TO_BLOCK_SHIFT);
524 disk_super->data_block_size = cpu_to_le32(pmd->data_block_size);
526 return dm_tm_commit(pmd->tm, sblock);
529 static int __format_metadata(struct dm_pool_metadata *pmd)
533 r = dm_tm_create_with_sm(pmd->bm, THIN_SUPERBLOCK_LOCATION,
534 &pmd->tm, &pmd->metadata_sm);
536 DMERR("tm_create_with_sm failed");
540 pmd->data_sm = dm_sm_disk_create(pmd->tm, 0);
541 if (IS_ERR(pmd->data_sm)) {
542 DMERR("sm_disk_create failed");
543 r = PTR_ERR(pmd->data_sm);
547 pmd->nb_tm = dm_tm_create_non_blocking_clone(pmd->tm);
549 DMERR("could not create non-blocking clone tm");
551 goto bad_cleanup_data_sm;
554 __setup_btree_details(pmd);
556 r = dm_btree_empty(&pmd->info, &pmd->root);
558 goto bad_cleanup_nb_tm;
560 r = dm_btree_empty(&pmd->details_info, &pmd->details_root);
562 DMERR("couldn't create devices root");
563 goto bad_cleanup_nb_tm;
566 r = __write_initial_superblock(pmd);
568 goto bad_cleanup_nb_tm;
573 dm_tm_destroy(pmd->nb_tm);
575 dm_sm_destroy(pmd->data_sm);
577 dm_tm_destroy(pmd->tm);
578 dm_sm_destroy(pmd->metadata_sm);
583 static int __check_incompat_features(struct thin_disk_superblock *disk_super,
584 struct dm_pool_metadata *pmd)
588 features = le32_to_cpu(disk_super->incompat_flags) & ~THIN_FEATURE_INCOMPAT_SUPP;
590 DMERR("could not access metadata due to unsupported optional features (%lx).",
591 (unsigned long)features);
596 * Check for read-only metadata to skip the following RDWR checks.
598 if (get_disk_ro(pmd->bdev->bd_disk))
601 features = le32_to_cpu(disk_super->compat_ro_flags) & ~THIN_FEATURE_COMPAT_RO_SUPP;
603 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
604 (unsigned long)features);
611 static int __open_metadata(struct dm_pool_metadata *pmd)
614 struct dm_block *sblock;
615 struct thin_disk_superblock *disk_super;
617 r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
618 &sb_validator, &sblock);
620 DMERR("couldn't read superblock");
624 disk_super = dm_block_data(sblock);
626 /* Verify the data block size hasn't changed */
627 if (le32_to_cpu(disk_super->data_block_size) != pmd->data_block_size) {
628 DMERR("changing the data block size (from %u to %llu) is not supported",
629 le32_to_cpu(disk_super->data_block_size),
630 (unsigned long long)pmd->data_block_size);
632 goto bad_unlock_sblock;
635 r = __check_incompat_features(disk_super, pmd);
637 goto bad_unlock_sblock;
639 r = dm_tm_open_with_sm(pmd->bm, THIN_SUPERBLOCK_LOCATION,
640 disk_super->metadata_space_map_root,
641 sizeof(disk_super->metadata_space_map_root),
642 &pmd->tm, &pmd->metadata_sm);
644 DMERR("tm_open_with_sm failed");
645 goto bad_unlock_sblock;
648 pmd->data_sm = dm_sm_disk_open(pmd->tm, disk_super->data_space_map_root,
649 sizeof(disk_super->data_space_map_root));
650 if (IS_ERR(pmd->data_sm)) {
651 DMERR("sm_disk_open failed");
652 r = PTR_ERR(pmd->data_sm);
656 pmd->nb_tm = dm_tm_create_non_blocking_clone(pmd->tm);
658 DMERR("could not create non-blocking clone tm");
660 goto bad_cleanup_data_sm;
663 __setup_btree_details(pmd);
664 dm_bm_unlock(sblock);
669 dm_sm_destroy(pmd->data_sm);
671 dm_tm_destroy(pmd->tm);
672 dm_sm_destroy(pmd->metadata_sm);
674 dm_bm_unlock(sblock);
679 static int __open_or_format_metadata(struct dm_pool_metadata *pmd, bool format_device)
683 r = __superblock_all_zeroes(pmd->bm, &unformatted);
688 return format_device ? __format_metadata(pmd) : -EPERM;
690 return __open_metadata(pmd);
693 static int __create_persistent_data_objects(struct dm_pool_metadata *pmd, bool format_device)
697 pmd->bm = dm_block_manager_create(pmd->bdev, THIN_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
698 THIN_MAX_CONCURRENT_LOCKS);
699 if (IS_ERR(pmd->bm)) {
700 DMERR("could not create block manager");
701 return PTR_ERR(pmd->bm);
704 r = __open_or_format_metadata(pmd, format_device);
706 dm_block_manager_destroy(pmd->bm);
711 static void __destroy_persistent_data_objects(struct dm_pool_metadata *pmd)
713 dm_sm_destroy(pmd->data_sm);
714 dm_sm_destroy(pmd->metadata_sm);
715 dm_tm_destroy(pmd->nb_tm);
716 dm_tm_destroy(pmd->tm);
717 dm_block_manager_destroy(pmd->bm);
720 static int __begin_transaction(struct dm_pool_metadata *pmd)
723 struct thin_disk_superblock *disk_super;
724 struct dm_block *sblock;
727 * We re-read the superblock every time. Shouldn't need to do this
730 r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
731 &sb_validator, &sblock);
735 disk_super = dm_block_data(sblock);
736 pmd->time = le32_to_cpu(disk_super->time);
737 pmd->root = le64_to_cpu(disk_super->data_mapping_root);
738 pmd->details_root = le64_to_cpu(disk_super->device_details_root);
739 pmd->trans_id = le64_to_cpu(disk_super->trans_id);
740 pmd->flags = le32_to_cpu(disk_super->flags);
741 pmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
743 dm_bm_unlock(sblock);
747 static int __write_changed_details(struct dm_pool_metadata *pmd)
750 struct dm_thin_device *td, *tmp;
751 struct disk_device_details details;
754 list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
760 details.mapped_blocks = cpu_to_le64(td->mapped_blocks);
761 details.transaction_id = cpu_to_le64(td->transaction_id);
762 details.creation_time = cpu_to_le32(td->creation_time);
763 details.snapshotted_time = cpu_to_le32(td->snapshotted_time);
764 __dm_bless_for_disk(&details);
766 r = dm_btree_insert(&pmd->details_info, pmd->details_root,
767 &key, &details, &pmd->details_root);
782 static int __commit_transaction(struct dm_pool_metadata *pmd)
785 struct thin_disk_superblock *disk_super;
786 struct dm_block *sblock;
789 * We need to know if the thin_disk_superblock exceeds a 512-byte sector.
791 BUILD_BUG_ON(sizeof(struct thin_disk_superblock) > 512);
793 r = __write_changed_details(pmd);
797 r = dm_sm_commit(pmd->data_sm);
801 r = dm_tm_pre_commit(pmd->tm);
805 r = save_sm_roots(pmd);
809 r = superblock_lock(pmd, &sblock);
813 disk_super = dm_block_data(sblock);
814 disk_super->time = cpu_to_le32(pmd->time);
815 disk_super->data_mapping_root = cpu_to_le64(pmd->root);
816 disk_super->device_details_root = cpu_to_le64(pmd->details_root);
817 disk_super->trans_id = cpu_to_le64(pmd->trans_id);
818 disk_super->flags = cpu_to_le32(pmd->flags);
820 copy_sm_roots(pmd, disk_super);
822 return dm_tm_commit(pmd->tm, sblock);
825 static void __set_metadata_reserve(struct dm_pool_metadata *pmd)
829 dm_block_t max_blocks = 4096; /* 16M */
831 r = dm_sm_get_nr_blocks(pmd->metadata_sm, &total);
833 DMERR("could not get size of metadata device");
834 pmd->metadata_reserve = max_blocks;
836 pmd->metadata_reserve = min(max_blocks, div_u64(total, 10));
839 struct dm_pool_metadata *dm_pool_metadata_open(struct block_device *bdev,
840 sector_t data_block_size,
844 struct dm_pool_metadata *pmd;
846 pmd = kmalloc(sizeof(*pmd), GFP_KERNEL);
848 DMERR("could not allocate metadata struct");
849 return ERR_PTR(-ENOMEM);
852 init_rwsem(&pmd->root_lock);
854 INIT_LIST_HEAD(&pmd->thin_devices);
855 pmd->fail_io = false;
857 pmd->data_block_size = data_block_size;
859 r = __create_persistent_data_objects(pmd, format_device);
865 r = __begin_transaction(pmd);
867 if (dm_pool_metadata_close(pmd) < 0)
868 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
872 __set_metadata_reserve(pmd);
877 int dm_pool_metadata_close(struct dm_pool_metadata *pmd)
880 unsigned open_devices = 0;
881 struct dm_thin_device *td, *tmp;
883 down_read(&pmd->root_lock);
884 list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
892 up_read(&pmd->root_lock);
895 DMERR("attempt to close pmd when %u device(s) are still open",
900 if (!dm_bm_is_read_only(pmd->bm) && !pmd->fail_io) {
901 r = __commit_transaction(pmd);
903 DMWARN("%s: __commit_transaction() failed, error = %d",
908 __destroy_persistent_data_objects(pmd);
915 * __open_device: Returns @td corresponding to device with id @dev,
916 * creating it if @create is set and incrementing @td->open_count.
917 * On failure, @td is undefined.
919 static int __open_device(struct dm_pool_metadata *pmd,
920 dm_thin_id dev, int create,
921 struct dm_thin_device **td)
924 struct dm_thin_device *td2;
926 struct disk_device_details details_le;
929 * If the device is already open, return it.
931 list_for_each_entry(td2, &pmd->thin_devices, list)
932 if (td2->id == dev) {
934 * May not create an already-open device.
945 * Check the device exists.
947 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
950 if (r != -ENODATA || !create)
957 details_le.mapped_blocks = 0;
958 details_le.transaction_id = cpu_to_le64(pmd->trans_id);
959 details_le.creation_time = cpu_to_le32(pmd->time);
960 details_le.snapshotted_time = cpu_to_le32(pmd->time);
963 *td = kmalloc(sizeof(**td), GFP_NOIO);
969 (*td)->open_count = 1;
970 (*td)->changed = changed;
971 (*td)->aborted_with_changes = false;
972 (*td)->mapped_blocks = le64_to_cpu(details_le.mapped_blocks);
973 (*td)->transaction_id = le64_to_cpu(details_le.transaction_id);
974 (*td)->creation_time = le32_to_cpu(details_le.creation_time);
975 (*td)->snapshotted_time = le32_to_cpu(details_le.snapshotted_time);
977 list_add(&(*td)->list, &pmd->thin_devices);
982 static void __close_device(struct dm_thin_device *td)
987 static int __create_thin(struct dm_pool_metadata *pmd,
993 struct disk_device_details details_le;
994 struct dm_thin_device *td;
997 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
1003 * Create an empty btree for the mappings.
1005 r = dm_btree_empty(&pmd->bl_info, &dev_root);
1010 * Insert it into the main mapping tree.
1012 value = cpu_to_le64(dev_root);
1013 __dm_bless_for_disk(&value);
1014 r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
1016 dm_btree_del(&pmd->bl_info, dev_root);
1020 r = __open_device(pmd, dev, 1, &td);
1022 dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1023 dm_btree_del(&pmd->bl_info, dev_root);
1031 int dm_pool_create_thin(struct dm_pool_metadata *pmd, dm_thin_id dev)
1035 down_write(&pmd->root_lock);
1037 r = __create_thin(pmd, dev);
1038 up_write(&pmd->root_lock);
1043 static int __set_snapshot_details(struct dm_pool_metadata *pmd,
1044 struct dm_thin_device *snap,
1045 dm_thin_id origin, uint32_t time)
1048 struct dm_thin_device *td;
1050 r = __open_device(pmd, origin, 0, &td);
1055 td->snapshotted_time = time;
1057 snap->mapped_blocks = td->mapped_blocks;
1058 snap->snapshotted_time = time;
1064 static int __create_snap(struct dm_pool_metadata *pmd,
1065 dm_thin_id dev, dm_thin_id origin)
1068 dm_block_t origin_root;
1069 uint64_t key = origin, dev_key = dev;
1070 struct dm_thin_device *td;
1071 struct disk_device_details details_le;
1074 /* check this device is unused */
1075 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
1076 &dev_key, &details_le);
1080 /* find the mapping tree for the origin */
1081 r = dm_btree_lookup(&pmd->tl_info, pmd->root, &key, &value);
1084 origin_root = le64_to_cpu(value);
1086 /* clone the origin, an inc will do */
1087 dm_tm_inc(pmd->tm, origin_root);
1089 /* insert into the main mapping tree */
1090 value = cpu_to_le64(origin_root);
1091 __dm_bless_for_disk(&value);
1093 r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
1095 dm_tm_dec(pmd->tm, origin_root);
1101 r = __open_device(pmd, dev, 1, &td);
1105 r = __set_snapshot_details(pmd, td, origin, pmd->time);
1114 dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1115 dm_btree_remove(&pmd->details_info, pmd->details_root,
1116 &key, &pmd->details_root);
1120 int dm_pool_create_snap(struct dm_pool_metadata *pmd,
1126 down_write(&pmd->root_lock);
1128 r = __create_snap(pmd, dev, origin);
1129 up_write(&pmd->root_lock);
1134 static int __delete_device(struct dm_pool_metadata *pmd, dm_thin_id dev)
1138 struct dm_thin_device *td;
1140 /* TODO: failure should mark the transaction invalid */
1141 r = __open_device(pmd, dev, 0, &td);
1145 if (td->open_count > 1) {
1150 list_del(&td->list);
1152 r = dm_btree_remove(&pmd->details_info, pmd->details_root,
1153 &key, &pmd->details_root);
1157 r = dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1164 int dm_pool_delete_thin_device(struct dm_pool_metadata *pmd,
1169 down_write(&pmd->root_lock);
1171 r = __delete_device(pmd, dev);
1172 up_write(&pmd->root_lock);
1177 int dm_pool_set_metadata_transaction_id(struct dm_pool_metadata *pmd,
1178 uint64_t current_id,
1183 down_write(&pmd->root_lock);
1188 if (pmd->trans_id != current_id) {
1189 DMERR("mismatched transaction id");
1193 pmd->trans_id = new_id;
1197 up_write(&pmd->root_lock);
1202 int dm_pool_get_metadata_transaction_id(struct dm_pool_metadata *pmd,
1207 down_read(&pmd->root_lock);
1208 if (!pmd->fail_io) {
1209 *result = pmd->trans_id;
1212 up_read(&pmd->root_lock);
1217 static int __reserve_metadata_snap(struct dm_pool_metadata *pmd)
1220 struct thin_disk_superblock *disk_super;
1221 struct dm_block *copy, *sblock;
1222 dm_block_t held_root;
1225 * We commit to ensure the btree roots which we increment in a
1226 * moment are up to date.
1228 __commit_transaction(pmd);
1231 * Copy the superblock.
1233 dm_sm_inc_block(pmd->metadata_sm, THIN_SUPERBLOCK_LOCATION);
1234 r = dm_tm_shadow_block(pmd->tm, THIN_SUPERBLOCK_LOCATION,
1235 &sb_validator, ©, &inc);
1241 held_root = dm_block_location(copy);
1242 disk_super = dm_block_data(copy);
1244 if (le64_to_cpu(disk_super->held_root)) {
1245 DMWARN("Pool metadata snapshot already exists: release this before taking another.");
1247 dm_tm_dec(pmd->tm, held_root);
1248 dm_tm_unlock(pmd->tm, copy);
1253 * Wipe the spacemap since we're not publishing this.
1255 memset(&disk_super->data_space_map_root, 0,
1256 sizeof(disk_super->data_space_map_root));
1257 memset(&disk_super->metadata_space_map_root, 0,
1258 sizeof(disk_super->metadata_space_map_root));
1261 * Increment the data structures that need to be preserved.
1263 dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->data_mapping_root));
1264 dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->device_details_root));
1265 dm_tm_unlock(pmd->tm, copy);
1268 * Write the held root into the superblock.
1270 r = superblock_lock(pmd, &sblock);
1272 dm_tm_dec(pmd->tm, held_root);
1276 disk_super = dm_block_data(sblock);
1277 disk_super->held_root = cpu_to_le64(held_root);
1278 dm_bm_unlock(sblock);
1282 int dm_pool_reserve_metadata_snap(struct dm_pool_metadata *pmd)
1286 down_write(&pmd->root_lock);
1288 r = __reserve_metadata_snap(pmd);
1289 up_write(&pmd->root_lock);
1294 static int __release_metadata_snap(struct dm_pool_metadata *pmd)
1297 struct thin_disk_superblock *disk_super;
1298 struct dm_block *sblock, *copy;
1299 dm_block_t held_root;
1301 r = superblock_lock(pmd, &sblock);
1305 disk_super = dm_block_data(sblock);
1306 held_root = le64_to_cpu(disk_super->held_root);
1307 disk_super->held_root = cpu_to_le64(0);
1309 dm_bm_unlock(sblock);
1312 DMWARN("No pool metadata snapshot found: nothing to release.");
1316 r = dm_tm_read_lock(pmd->tm, held_root, &sb_validator, ©);
1320 disk_super = dm_block_data(copy);
1321 dm_btree_del(&pmd->info, le64_to_cpu(disk_super->data_mapping_root));
1322 dm_btree_del(&pmd->details_info, le64_to_cpu(disk_super->device_details_root));
1323 dm_sm_dec_block(pmd->metadata_sm, held_root);
1325 dm_tm_unlock(pmd->tm, copy);
1330 int dm_pool_release_metadata_snap(struct dm_pool_metadata *pmd)
1334 down_write(&pmd->root_lock);
1336 r = __release_metadata_snap(pmd);
1337 up_write(&pmd->root_lock);
1342 static int __get_metadata_snap(struct dm_pool_metadata *pmd,
1346 struct thin_disk_superblock *disk_super;
1347 struct dm_block *sblock;
1349 r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
1350 &sb_validator, &sblock);
1354 disk_super = dm_block_data(sblock);
1355 *result = le64_to_cpu(disk_super->held_root);
1357 dm_bm_unlock(sblock);
1362 int dm_pool_get_metadata_snap(struct dm_pool_metadata *pmd,
1367 down_read(&pmd->root_lock);
1369 r = __get_metadata_snap(pmd, result);
1370 up_read(&pmd->root_lock);
1375 int dm_pool_open_thin_device(struct dm_pool_metadata *pmd, dm_thin_id dev,
1376 struct dm_thin_device **td)
1380 down_write(&pmd->root_lock);
1382 r = __open_device(pmd, dev, 0, td);
1383 up_write(&pmd->root_lock);
1388 int dm_pool_close_thin_device(struct dm_thin_device *td)
1390 down_write(&td->pmd->root_lock);
1392 up_write(&td->pmd->root_lock);
1397 dm_thin_id dm_thin_dev_id(struct dm_thin_device *td)
1403 * Check whether @time (of block creation) is older than @td's last snapshot.
1404 * If so then the associated block is shared with the last snapshot device.
1405 * Any block on a device created *after* the device last got snapshotted is
1406 * necessarily not shared.
1408 static bool __snapshotted_since(struct dm_thin_device *td, uint32_t time)
1410 return td->snapshotted_time > time;
1413 static void unpack_lookup_result(struct dm_thin_device *td, __le64 value,
1414 struct dm_thin_lookup_result *result)
1416 uint64_t block_time = 0;
1417 dm_block_t exception_block;
1418 uint32_t exception_time;
1420 block_time = le64_to_cpu(value);
1421 unpack_block_time(block_time, &exception_block, &exception_time);
1422 result->block = exception_block;
1423 result->shared = __snapshotted_since(td, exception_time);
1426 static int __find_block(struct dm_thin_device *td, dm_block_t block,
1427 int can_issue_io, struct dm_thin_lookup_result *result)
1431 struct dm_pool_metadata *pmd = td->pmd;
1432 dm_block_t keys[2] = { td->id, block };
1433 struct dm_btree_info *info;
1438 info = &pmd->nb_info;
1440 r = dm_btree_lookup(info, pmd->root, keys, &value);
1442 unpack_lookup_result(td, value, result);
1447 int dm_thin_find_block(struct dm_thin_device *td, dm_block_t block,
1448 int can_issue_io, struct dm_thin_lookup_result *result)
1451 struct dm_pool_metadata *pmd = td->pmd;
1453 down_read(&pmd->root_lock);
1455 up_read(&pmd->root_lock);
1459 r = __find_block(td, block, can_issue_io, result);
1461 up_read(&pmd->root_lock);
1465 static int __find_next_mapped_block(struct dm_thin_device *td, dm_block_t block,
1467 struct dm_thin_lookup_result *result)
1471 struct dm_pool_metadata *pmd = td->pmd;
1472 dm_block_t keys[2] = { td->id, block };
1474 r = dm_btree_lookup_next(&pmd->info, pmd->root, keys, vblock, &value);
1476 unpack_lookup_result(td, value, result);
1481 static int __find_mapped_range(struct dm_thin_device *td,
1482 dm_block_t begin, dm_block_t end,
1483 dm_block_t *thin_begin, dm_block_t *thin_end,
1484 dm_block_t *pool_begin, bool *maybe_shared)
1487 dm_block_t pool_end;
1488 struct dm_thin_lookup_result lookup;
1493 r = __find_next_mapped_block(td, begin, &begin, &lookup);
1500 *thin_begin = begin;
1501 *pool_begin = lookup.block;
1502 *maybe_shared = lookup.shared;
1505 pool_end = *pool_begin + 1;
1506 while (begin != end) {
1507 r = __find_block(td, begin, true, &lookup);
1515 if ((lookup.block != pool_end) ||
1516 (lookup.shared != *maybe_shared))
1527 int dm_thin_find_mapped_range(struct dm_thin_device *td,
1528 dm_block_t begin, dm_block_t end,
1529 dm_block_t *thin_begin, dm_block_t *thin_end,
1530 dm_block_t *pool_begin, bool *maybe_shared)
1533 struct dm_pool_metadata *pmd = td->pmd;
1535 down_read(&pmd->root_lock);
1536 if (!pmd->fail_io) {
1537 r = __find_mapped_range(td, begin, end, thin_begin, thin_end,
1538 pool_begin, maybe_shared);
1540 up_read(&pmd->root_lock);
1545 static int __insert(struct dm_thin_device *td, dm_block_t block,
1546 dm_block_t data_block)
1550 struct dm_pool_metadata *pmd = td->pmd;
1551 dm_block_t keys[2] = { td->id, block };
1553 value = cpu_to_le64(pack_block_time(data_block, pmd->time));
1554 __dm_bless_for_disk(&value);
1556 r = dm_btree_insert_notify(&pmd->info, pmd->root, keys, &value,
1557 &pmd->root, &inserted);
1563 td->mapped_blocks++;
1568 int dm_thin_insert_block(struct dm_thin_device *td, dm_block_t block,
1569 dm_block_t data_block)
1573 down_write(&td->pmd->root_lock);
1574 if (!td->pmd->fail_io)
1575 r = __insert(td, block, data_block);
1576 up_write(&td->pmd->root_lock);
1581 static int __remove(struct dm_thin_device *td, dm_block_t block)
1584 struct dm_pool_metadata *pmd = td->pmd;
1585 dm_block_t keys[2] = { td->id, block };
1587 r = dm_btree_remove(&pmd->info, pmd->root, keys, &pmd->root);
1591 td->mapped_blocks--;
1597 static int __remove_range(struct dm_thin_device *td, dm_block_t begin, dm_block_t end)
1600 unsigned count, total_count = 0;
1601 struct dm_pool_metadata *pmd = td->pmd;
1602 dm_block_t keys[1] = { td->id };
1604 dm_block_t mapping_root;
1607 * Find the mapping tree
1609 r = dm_btree_lookup(&pmd->tl_info, pmd->root, keys, &value);
1614 * Remove from the mapping tree, taking care to inc the
1615 * ref count so it doesn't get deleted.
1617 mapping_root = le64_to_cpu(value);
1618 dm_tm_inc(pmd->tm, mapping_root);
1619 r = dm_btree_remove(&pmd->tl_info, pmd->root, keys, &pmd->root);
1624 * Remove leaves stops at the first unmapped entry, so we have to
1625 * loop round finding mapped ranges.
1627 while (begin < end) {
1628 r = dm_btree_lookup_next(&pmd->bl_info, mapping_root, &begin, &begin, &value);
1638 r = dm_btree_remove_leaves(&pmd->bl_info, mapping_root, &begin, end, &mapping_root, &count);
1642 total_count += count;
1645 td->mapped_blocks -= total_count;
1649 * Reinsert the mapping tree.
1651 value = cpu_to_le64(mapping_root);
1652 __dm_bless_for_disk(&value);
1653 return dm_btree_insert(&pmd->tl_info, pmd->root, keys, &value, &pmd->root);
1656 int dm_thin_remove_block(struct dm_thin_device *td, dm_block_t block)
1660 down_write(&td->pmd->root_lock);
1661 if (!td->pmd->fail_io)
1662 r = __remove(td, block);
1663 up_write(&td->pmd->root_lock);
1668 int dm_thin_remove_range(struct dm_thin_device *td,
1669 dm_block_t begin, dm_block_t end)
1673 down_write(&td->pmd->root_lock);
1674 if (!td->pmd->fail_io)
1675 r = __remove_range(td, begin, end);
1676 up_write(&td->pmd->root_lock);
1681 int dm_pool_block_is_used(struct dm_pool_metadata *pmd, dm_block_t b, bool *result)
1686 down_read(&pmd->root_lock);
1687 r = dm_sm_get_count(pmd->data_sm, b, &ref_count);
1689 *result = (ref_count != 0);
1690 up_read(&pmd->root_lock);
1695 int dm_pool_inc_data_range(struct dm_pool_metadata *pmd, dm_block_t b, dm_block_t e)
1699 down_write(&pmd->root_lock);
1700 for (; b != e; b++) {
1701 r = dm_sm_inc_block(pmd->data_sm, b);
1705 up_write(&pmd->root_lock);
1710 int dm_pool_dec_data_range(struct dm_pool_metadata *pmd, dm_block_t b, dm_block_t e)
1714 down_write(&pmd->root_lock);
1715 for (; b != e; b++) {
1716 r = dm_sm_dec_block(pmd->data_sm, b);
1720 up_write(&pmd->root_lock);
1725 bool dm_thin_changed_this_transaction(struct dm_thin_device *td)
1729 down_read(&td->pmd->root_lock);
1731 up_read(&td->pmd->root_lock);
1736 bool dm_pool_changed_this_transaction(struct dm_pool_metadata *pmd)
1739 struct dm_thin_device *td, *tmp;
1741 down_read(&pmd->root_lock);
1742 list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
1748 up_read(&pmd->root_lock);
1753 bool dm_thin_aborted_changes(struct dm_thin_device *td)
1757 down_read(&td->pmd->root_lock);
1758 r = td->aborted_with_changes;
1759 up_read(&td->pmd->root_lock);
1764 int dm_pool_alloc_data_block(struct dm_pool_metadata *pmd, dm_block_t *result)
1768 down_write(&pmd->root_lock);
1770 r = dm_sm_new_block(pmd->data_sm, result);
1771 up_write(&pmd->root_lock);
1776 int dm_pool_commit_metadata(struct dm_pool_metadata *pmd)
1780 down_write(&pmd->root_lock);
1784 r = __commit_transaction(pmd);
1789 * Open the next transaction.
1791 r = __begin_transaction(pmd);
1793 up_write(&pmd->root_lock);
1797 static void __set_abort_with_changes_flags(struct dm_pool_metadata *pmd)
1799 struct dm_thin_device *td;
1801 list_for_each_entry(td, &pmd->thin_devices, list)
1802 td->aborted_with_changes = td->changed;
1805 int dm_pool_abort_metadata(struct dm_pool_metadata *pmd)
1809 down_write(&pmd->root_lock);
1813 __set_abort_with_changes_flags(pmd);
1814 __destroy_persistent_data_objects(pmd);
1815 r = __create_persistent_data_objects(pmd, false);
1817 pmd->fail_io = true;
1820 up_write(&pmd->root_lock);
1825 int dm_pool_get_free_block_count(struct dm_pool_metadata *pmd, dm_block_t *result)
1829 down_read(&pmd->root_lock);
1831 r = dm_sm_get_nr_free(pmd->data_sm, result);
1832 up_read(&pmd->root_lock);
1837 int dm_pool_get_free_metadata_block_count(struct dm_pool_metadata *pmd,
1842 down_read(&pmd->root_lock);
1844 r = dm_sm_get_nr_free(pmd->metadata_sm, result);
1847 if (*result < pmd->metadata_reserve)
1850 *result -= pmd->metadata_reserve;
1852 up_read(&pmd->root_lock);
1857 int dm_pool_get_metadata_dev_size(struct dm_pool_metadata *pmd,
1862 down_read(&pmd->root_lock);
1864 r = dm_sm_get_nr_blocks(pmd->metadata_sm, result);
1865 up_read(&pmd->root_lock);
1870 int dm_pool_get_data_dev_size(struct dm_pool_metadata *pmd, dm_block_t *result)
1874 down_read(&pmd->root_lock);
1876 r = dm_sm_get_nr_blocks(pmd->data_sm, result);
1877 up_read(&pmd->root_lock);
1882 int dm_thin_get_mapped_count(struct dm_thin_device *td, dm_block_t *result)
1885 struct dm_pool_metadata *pmd = td->pmd;
1887 down_read(&pmd->root_lock);
1888 if (!pmd->fail_io) {
1889 *result = td->mapped_blocks;
1892 up_read(&pmd->root_lock);
1897 static int __highest_block(struct dm_thin_device *td, dm_block_t *result)
1901 dm_block_t thin_root;
1902 struct dm_pool_metadata *pmd = td->pmd;
1904 r = dm_btree_lookup(&pmd->tl_info, pmd->root, &td->id, &value_le);
1908 thin_root = le64_to_cpu(value_le);
1910 return dm_btree_find_highest_key(&pmd->bl_info, thin_root, result);
1913 int dm_thin_get_highest_mapped_block(struct dm_thin_device *td,
1917 struct dm_pool_metadata *pmd = td->pmd;
1919 down_read(&pmd->root_lock);
1921 r = __highest_block(td, result);
1922 up_read(&pmd->root_lock);
1927 static int __resize_space_map(struct dm_space_map *sm, dm_block_t new_count)
1930 dm_block_t old_count;
1932 r = dm_sm_get_nr_blocks(sm, &old_count);
1936 if (new_count == old_count)
1939 if (new_count < old_count) {
1940 DMERR("cannot reduce size of space map");
1944 return dm_sm_extend(sm, new_count - old_count);
1947 int dm_pool_resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1951 down_write(&pmd->root_lock);
1953 r = __resize_space_map(pmd->data_sm, new_count);
1954 up_write(&pmd->root_lock);
1959 int dm_pool_resize_metadata_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1963 down_write(&pmd->root_lock);
1964 if (!pmd->fail_io) {
1965 r = __resize_space_map(pmd->metadata_sm, new_count);
1967 __set_metadata_reserve(pmd);
1969 up_write(&pmd->root_lock);
1974 void dm_pool_metadata_read_only(struct dm_pool_metadata *pmd)
1976 down_write(&pmd->root_lock);
1977 dm_bm_set_read_only(pmd->bm);
1978 up_write(&pmd->root_lock);
1981 void dm_pool_metadata_read_write(struct dm_pool_metadata *pmd)
1983 down_write(&pmd->root_lock);
1984 dm_bm_set_read_write(pmd->bm);
1985 up_write(&pmd->root_lock);
1988 int dm_pool_register_metadata_threshold(struct dm_pool_metadata *pmd,
1989 dm_block_t threshold,
1990 dm_sm_threshold_fn fn,
1995 down_write(&pmd->root_lock);
1996 r = dm_sm_register_threshold_callback(pmd->metadata_sm, threshold, fn, context);
1997 up_write(&pmd->root_lock);
2002 int dm_pool_metadata_set_needs_check(struct dm_pool_metadata *pmd)
2005 struct dm_block *sblock;
2006 struct thin_disk_superblock *disk_super;
2008 down_write(&pmd->root_lock);
2009 pmd->flags |= THIN_METADATA_NEEDS_CHECK_FLAG;
2011 r = superblock_lock(pmd, &sblock);
2013 DMERR("couldn't read superblock");
2017 disk_super = dm_block_data(sblock);
2018 disk_super->flags = cpu_to_le32(pmd->flags);
2020 dm_bm_unlock(sblock);
2022 up_write(&pmd->root_lock);
2026 bool dm_pool_metadata_needs_check(struct dm_pool_metadata *pmd)
2030 down_read(&pmd->root_lock);
2031 needs_check = pmd->flags & THIN_METADATA_NEEDS_CHECK_FLAG;
2032 up_read(&pmd->root_lock);
2037 void dm_pool_issue_prefetches(struct dm_pool_metadata *pmd)
2039 down_read(&pmd->root_lock);
2041 dm_tm_issue_prefetches(pmd->tm);
2042 up_read(&pmd->root_lock);