dm thin metadata: fix __udivdi3 undefined on 32-bit
[linux/fpc-iii.git] / drivers / md / dm-thin-metadata.c
blob2711aa965445c073a91fbfcd2c67e4461d8f84a8
1 /*
2 * Copyright (C) 2011-2012 Red Hat, Inc.
4 * This file is released under the GPL.
5 */
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
21 * atomic writes.
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
32 * bits.
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
40 * cpu cache.
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
46 * are etc.
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:
51 * 0 - ref count is 0
52 * 1 - ref count is 1
53 * 2 - ref count is 2
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
58 * count.
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 THIN_METADATA_CACHE_SIZE 64
81 #define SECTOR_TO_BLOCK_SHIFT 3
84 * For btree insert:
85 * 3 for btree insert +
86 * 2 for btree lookup used within space map
87 * For btree remove:
88 * 2 for shadow spine +
89 * 4 for rebalance 3 child node
91 #define THIN_MAX_CONCURRENT_LOCKS 6
93 /* This should be plenty */
94 #define SPACE_MAP_ROOT_SIZE 128
97 * Little endian on-disk superblock and device details.
99 struct thin_disk_superblock {
100 __le32 csum; /* Checksum of superblock except for this field. */
101 __le32 flags;
102 __le64 blocknr; /* This block number, dm_block_t. */
104 __u8 uuid[16];
105 __le64 magic;
106 __le32 version;
107 __le32 time;
109 __le64 trans_id;
112 * Root held by userspace transactions.
114 __le64 held_root;
116 __u8 data_space_map_root[SPACE_MAP_ROOT_SIZE];
117 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
120 * 2-level btree mapping (dev_id, (dev block, time)) -> data block
122 __le64 data_mapping_root;
125 * Device detail root mapping dev_id -> device_details
127 __le64 device_details_root;
129 __le32 data_block_size; /* In 512-byte sectors. */
131 __le32 metadata_block_size; /* In 512-byte sectors. */
132 __le64 metadata_nr_blocks;
134 __le32 compat_flags;
135 __le32 compat_ro_flags;
136 __le32 incompat_flags;
137 } __packed;
139 struct disk_device_details {
140 __le64 mapped_blocks;
141 __le64 transaction_id; /* When created. */
142 __le32 creation_time;
143 __le32 snapshotted_time;
144 } __packed;
146 struct dm_pool_metadata {
147 struct hlist_node hash;
149 struct block_device *bdev;
150 struct dm_block_manager *bm;
151 struct dm_space_map *metadata_sm;
152 struct dm_space_map *data_sm;
153 struct dm_transaction_manager *tm;
154 struct dm_transaction_manager *nb_tm;
157 * Two-level btree.
158 * First level holds thin_dev_t.
159 * Second level holds mappings.
161 struct dm_btree_info info;
164 * Non-blocking version of the above.
166 struct dm_btree_info nb_info;
169 * Just the top level for deleting whole devices.
171 struct dm_btree_info tl_info;
174 * Just the bottom level for creating new devices.
176 struct dm_btree_info bl_info;
179 * Describes the device details btree.
181 struct dm_btree_info details_info;
183 struct rw_semaphore root_lock;
184 uint32_t time;
185 dm_block_t root;
186 dm_block_t details_root;
187 struct list_head thin_devices;
188 uint64_t trans_id;
189 unsigned long flags;
190 sector_t data_block_size;
193 * We reserve a section of the metadata for commit overhead.
194 * All reported space does *not* include this.
196 dm_block_t metadata_reserve;
199 * Set if a transaction has to be aborted but the attempt to roll back
200 * to the previous (good) transaction failed. The only pool metadata
201 * operation possible in this state is the closing of the device.
203 bool fail_io:1;
206 * Reading the space map roots can fail, so we read it into these
207 * buffers before the superblock is locked and updated.
209 __u8 data_space_map_root[SPACE_MAP_ROOT_SIZE];
210 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
213 struct dm_thin_device {
214 struct list_head list;
215 struct dm_pool_metadata *pmd;
216 dm_thin_id id;
218 int open_count;
219 bool changed:1;
220 bool aborted_with_changes:1;
221 uint64_t mapped_blocks;
222 uint64_t transaction_id;
223 uint32_t creation_time;
224 uint32_t snapshotted_time;
227 /*----------------------------------------------------------------
228 * superblock validator
229 *--------------------------------------------------------------*/
231 #define SUPERBLOCK_CSUM_XOR 160774
233 static void sb_prepare_for_write(struct dm_block_validator *v,
234 struct dm_block *b,
235 size_t block_size)
237 struct thin_disk_superblock *disk_super = dm_block_data(b);
239 disk_super->blocknr = cpu_to_le64(dm_block_location(b));
240 disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
241 block_size - sizeof(__le32),
242 SUPERBLOCK_CSUM_XOR));
245 static int sb_check(struct dm_block_validator *v,
246 struct dm_block *b,
247 size_t block_size)
249 struct thin_disk_superblock *disk_super = dm_block_data(b);
250 __le32 csum_le;
252 if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
253 DMERR("sb_check failed: blocknr %llu: "
254 "wanted %llu", le64_to_cpu(disk_super->blocknr),
255 (unsigned long long)dm_block_location(b));
256 return -ENOTBLK;
259 if (le64_to_cpu(disk_super->magic) != THIN_SUPERBLOCK_MAGIC) {
260 DMERR("sb_check failed: magic %llu: "
261 "wanted %llu", le64_to_cpu(disk_super->magic),
262 (unsigned long long)THIN_SUPERBLOCK_MAGIC);
263 return -EILSEQ;
266 csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
267 block_size - sizeof(__le32),
268 SUPERBLOCK_CSUM_XOR));
269 if (csum_le != disk_super->csum) {
270 DMERR("sb_check failed: csum %u: wanted %u",
271 le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
272 return -EILSEQ;
275 return 0;
278 static struct dm_block_validator sb_validator = {
279 .name = "superblock",
280 .prepare_for_write = sb_prepare_for_write,
281 .check = sb_check
284 /*----------------------------------------------------------------
285 * Methods for the btree value types
286 *--------------------------------------------------------------*/
288 static uint64_t pack_block_time(dm_block_t b, uint32_t t)
290 return (b << 24) | t;
293 static void unpack_block_time(uint64_t v, dm_block_t *b, uint32_t *t)
295 *b = v >> 24;
296 *t = v & ((1 << 24) - 1);
299 static void data_block_inc(void *context, const void *value_le)
301 struct dm_space_map *sm = context;
302 __le64 v_le;
303 uint64_t b;
304 uint32_t t;
306 memcpy(&v_le, value_le, sizeof(v_le));
307 unpack_block_time(le64_to_cpu(v_le), &b, &t);
308 dm_sm_inc_block(sm, b);
311 static void data_block_dec(void *context, const void *value_le)
313 struct dm_space_map *sm = context;
314 __le64 v_le;
315 uint64_t b;
316 uint32_t t;
318 memcpy(&v_le, value_le, sizeof(v_le));
319 unpack_block_time(le64_to_cpu(v_le), &b, &t);
320 dm_sm_dec_block(sm, b);
323 static int data_block_equal(void *context, const void *value1_le, const void *value2_le)
325 __le64 v1_le, v2_le;
326 uint64_t b1, b2;
327 uint32_t t;
329 memcpy(&v1_le, value1_le, sizeof(v1_le));
330 memcpy(&v2_le, value2_le, sizeof(v2_le));
331 unpack_block_time(le64_to_cpu(v1_le), &b1, &t);
332 unpack_block_time(le64_to_cpu(v2_le), &b2, &t);
334 return b1 == b2;
337 static void subtree_inc(void *context, const void *value)
339 struct dm_btree_info *info = context;
340 __le64 root_le;
341 uint64_t root;
343 memcpy(&root_le, value, sizeof(root_le));
344 root = le64_to_cpu(root_le);
345 dm_tm_inc(info->tm, root);
348 static void subtree_dec(void *context, const void *value)
350 struct dm_btree_info *info = context;
351 __le64 root_le;
352 uint64_t root;
354 memcpy(&root_le, value, sizeof(root_le));
355 root = le64_to_cpu(root_le);
356 if (dm_btree_del(info, root))
357 DMERR("btree delete failed\n");
360 static int subtree_equal(void *context, const void *value1_le, const void *value2_le)
362 __le64 v1_le, v2_le;
363 memcpy(&v1_le, value1_le, sizeof(v1_le));
364 memcpy(&v2_le, value2_le, sizeof(v2_le));
366 return v1_le == v2_le;
369 /*----------------------------------------------------------------*/
371 static int superblock_lock_zero(struct dm_pool_metadata *pmd,
372 struct dm_block **sblock)
374 return dm_bm_write_lock_zero(pmd->bm, THIN_SUPERBLOCK_LOCATION,
375 &sb_validator, sblock);
378 static int superblock_lock(struct dm_pool_metadata *pmd,
379 struct dm_block **sblock)
381 return dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
382 &sb_validator, sblock);
385 static int __superblock_all_zeroes(struct dm_block_manager *bm, int *result)
387 int r;
388 unsigned i;
389 struct dm_block *b;
390 __le64 *data_le, zero = cpu_to_le64(0);
391 unsigned block_size = dm_bm_block_size(bm) / sizeof(__le64);
394 * We can't use a validator here - it may be all zeroes.
396 r = dm_bm_read_lock(bm, THIN_SUPERBLOCK_LOCATION, NULL, &b);
397 if (r)
398 return r;
400 data_le = dm_block_data(b);
401 *result = 1;
402 for (i = 0; i < block_size; i++) {
403 if (data_le[i] != zero) {
404 *result = 0;
405 break;
409 dm_bm_unlock(b);
411 return 0;
414 static void __setup_btree_details(struct dm_pool_metadata *pmd)
416 pmd->info.tm = pmd->tm;
417 pmd->info.levels = 2;
418 pmd->info.value_type.context = pmd->data_sm;
419 pmd->info.value_type.size = sizeof(__le64);
420 pmd->info.value_type.inc = data_block_inc;
421 pmd->info.value_type.dec = data_block_dec;
422 pmd->info.value_type.equal = data_block_equal;
424 memcpy(&pmd->nb_info, &pmd->info, sizeof(pmd->nb_info));
425 pmd->nb_info.tm = pmd->nb_tm;
427 pmd->tl_info.tm = pmd->tm;
428 pmd->tl_info.levels = 1;
429 pmd->tl_info.value_type.context = &pmd->bl_info;
430 pmd->tl_info.value_type.size = sizeof(__le64);
431 pmd->tl_info.value_type.inc = subtree_inc;
432 pmd->tl_info.value_type.dec = subtree_dec;
433 pmd->tl_info.value_type.equal = subtree_equal;
435 pmd->bl_info.tm = pmd->tm;
436 pmd->bl_info.levels = 1;
437 pmd->bl_info.value_type.context = pmd->data_sm;
438 pmd->bl_info.value_type.size = sizeof(__le64);
439 pmd->bl_info.value_type.inc = data_block_inc;
440 pmd->bl_info.value_type.dec = data_block_dec;
441 pmd->bl_info.value_type.equal = data_block_equal;
443 pmd->details_info.tm = pmd->tm;
444 pmd->details_info.levels = 1;
445 pmd->details_info.value_type.context = NULL;
446 pmd->details_info.value_type.size = sizeof(struct disk_device_details);
447 pmd->details_info.value_type.inc = NULL;
448 pmd->details_info.value_type.dec = NULL;
449 pmd->details_info.value_type.equal = NULL;
452 static int save_sm_roots(struct dm_pool_metadata *pmd)
454 int r;
455 size_t len;
457 r = dm_sm_root_size(pmd->metadata_sm, &len);
458 if (r < 0)
459 return r;
461 r = dm_sm_copy_root(pmd->metadata_sm, &pmd->metadata_space_map_root, len);
462 if (r < 0)
463 return r;
465 r = dm_sm_root_size(pmd->data_sm, &len);
466 if (r < 0)
467 return r;
469 return dm_sm_copy_root(pmd->data_sm, &pmd->data_space_map_root, len);
472 static void copy_sm_roots(struct dm_pool_metadata *pmd,
473 struct thin_disk_superblock *disk)
475 memcpy(&disk->metadata_space_map_root,
476 &pmd->metadata_space_map_root,
477 sizeof(pmd->metadata_space_map_root));
479 memcpy(&disk->data_space_map_root,
480 &pmd->data_space_map_root,
481 sizeof(pmd->data_space_map_root));
484 static int __write_initial_superblock(struct dm_pool_metadata *pmd)
486 int r;
487 struct dm_block *sblock;
488 struct thin_disk_superblock *disk_super;
489 sector_t bdev_size = i_size_read(pmd->bdev->bd_inode) >> SECTOR_SHIFT;
491 if (bdev_size > THIN_METADATA_MAX_SECTORS)
492 bdev_size = THIN_METADATA_MAX_SECTORS;
494 r = dm_sm_commit(pmd->data_sm);
495 if (r < 0)
496 return r;
498 r = dm_tm_pre_commit(pmd->tm);
499 if (r < 0)
500 return r;
502 r = save_sm_roots(pmd);
503 if (r < 0)
504 return r;
506 r = superblock_lock_zero(pmd, &sblock);
507 if (r)
508 return r;
510 disk_super = dm_block_data(sblock);
511 disk_super->flags = 0;
512 memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
513 disk_super->magic = cpu_to_le64(THIN_SUPERBLOCK_MAGIC);
514 disk_super->version = cpu_to_le32(THIN_VERSION);
515 disk_super->time = 0;
516 disk_super->trans_id = 0;
517 disk_super->held_root = 0;
519 copy_sm_roots(pmd, disk_super);
521 disk_super->data_mapping_root = cpu_to_le64(pmd->root);
522 disk_super->device_details_root = cpu_to_le64(pmd->details_root);
523 disk_super->metadata_block_size = cpu_to_le32(THIN_METADATA_BLOCK_SIZE);
524 disk_super->metadata_nr_blocks = cpu_to_le64(bdev_size >> SECTOR_TO_BLOCK_SHIFT);
525 disk_super->data_block_size = cpu_to_le32(pmd->data_block_size);
527 return dm_tm_commit(pmd->tm, sblock);
530 static int __format_metadata(struct dm_pool_metadata *pmd)
532 int r;
534 r = dm_tm_create_with_sm(pmd->bm, THIN_SUPERBLOCK_LOCATION,
535 &pmd->tm, &pmd->metadata_sm);
536 if (r < 0) {
537 DMERR("tm_create_with_sm failed");
538 return r;
541 pmd->data_sm = dm_sm_disk_create(pmd->tm, 0);
542 if (IS_ERR(pmd->data_sm)) {
543 DMERR("sm_disk_create failed");
544 r = PTR_ERR(pmd->data_sm);
545 goto bad_cleanup_tm;
548 pmd->nb_tm = dm_tm_create_non_blocking_clone(pmd->tm);
549 if (!pmd->nb_tm) {
550 DMERR("could not create non-blocking clone tm");
551 r = -ENOMEM;
552 goto bad_cleanup_data_sm;
555 __setup_btree_details(pmd);
557 r = dm_btree_empty(&pmd->info, &pmd->root);
558 if (r < 0)
559 goto bad_cleanup_nb_tm;
561 r = dm_btree_empty(&pmd->details_info, &pmd->details_root);
562 if (r < 0) {
563 DMERR("couldn't create devices root");
564 goto bad_cleanup_nb_tm;
567 r = __write_initial_superblock(pmd);
568 if (r)
569 goto bad_cleanup_nb_tm;
571 return 0;
573 bad_cleanup_nb_tm:
574 dm_tm_destroy(pmd->nb_tm);
575 bad_cleanup_data_sm:
576 dm_sm_destroy(pmd->data_sm);
577 bad_cleanup_tm:
578 dm_tm_destroy(pmd->tm);
579 dm_sm_destroy(pmd->metadata_sm);
581 return r;
584 static int __check_incompat_features(struct thin_disk_superblock *disk_super,
585 struct dm_pool_metadata *pmd)
587 uint32_t features;
589 features = le32_to_cpu(disk_super->incompat_flags) & ~THIN_FEATURE_INCOMPAT_SUPP;
590 if (features) {
591 DMERR("could not access metadata due to unsupported optional features (%lx).",
592 (unsigned long)features);
593 return -EINVAL;
597 * Check for read-only metadata to skip the following RDWR checks.
599 if (get_disk_ro(pmd->bdev->bd_disk))
600 return 0;
602 features = le32_to_cpu(disk_super->compat_ro_flags) & ~THIN_FEATURE_COMPAT_RO_SUPP;
603 if (features) {
604 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
605 (unsigned long)features);
606 return -EINVAL;
609 return 0;
612 static int __open_metadata(struct dm_pool_metadata *pmd)
614 int r;
615 struct dm_block *sblock;
616 struct thin_disk_superblock *disk_super;
618 r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
619 &sb_validator, &sblock);
620 if (r < 0) {
621 DMERR("couldn't read superblock");
622 return r;
625 disk_super = dm_block_data(sblock);
627 /* Verify the data block size hasn't changed */
628 if (le32_to_cpu(disk_super->data_block_size) != pmd->data_block_size) {
629 DMERR("changing the data block size (from %u to %llu) is not supported",
630 le32_to_cpu(disk_super->data_block_size),
631 (unsigned long long)pmd->data_block_size);
632 r = -EINVAL;
633 goto bad_unlock_sblock;
636 r = __check_incompat_features(disk_super, pmd);
637 if (r < 0)
638 goto bad_unlock_sblock;
640 r = dm_tm_open_with_sm(pmd->bm, THIN_SUPERBLOCK_LOCATION,
641 disk_super->metadata_space_map_root,
642 sizeof(disk_super->metadata_space_map_root),
643 &pmd->tm, &pmd->metadata_sm);
644 if (r < 0) {
645 DMERR("tm_open_with_sm failed");
646 goto bad_unlock_sblock;
649 pmd->data_sm = dm_sm_disk_open(pmd->tm, disk_super->data_space_map_root,
650 sizeof(disk_super->data_space_map_root));
651 if (IS_ERR(pmd->data_sm)) {
652 DMERR("sm_disk_open failed");
653 r = PTR_ERR(pmd->data_sm);
654 goto bad_cleanup_tm;
657 pmd->nb_tm = dm_tm_create_non_blocking_clone(pmd->tm);
658 if (!pmd->nb_tm) {
659 DMERR("could not create non-blocking clone tm");
660 r = -ENOMEM;
661 goto bad_cleanup_data_sm;
664 __setup_btree_details(pmd);
665 dm_bm_unlock(sblock);
667 return 0;
669 bad_cleanup_data_sm:
670 dm_sm_destroy(pmd->data_sm);
671 bad_cleanup_tm:
672 dm_tm_destroy(pmd->tm);
673 dm_sm_destroy(pmd->metadata_sm);
674 bad_unlock_sblock:
675 dm_bm_unlock(sblock);
677 return r;
680 static int __open_or_format_metadata(struct dm_pool_metadata *pmd, bool format_device)
682 int r, unformatted;
684 r = __superblock_all_zeroes(pmd->bm, &unformatted);
685 if (r)
686 return r;
688 if (unformatted)
689 return format_device ? __format_metadata(pmd) : -EPERM;
691 return __open_metadata(pmd);
694 static int __create_persistent_data_objects(struct dm_pool_metadata *pmd, bool format_device)
696 int r;
698 pmd->bm = dm_block_manager_create(pmd->bdev, THIN_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
699 THIN_METADATA_CACHE_SIZE,
700 THIN_MAX_CONCURRENT_LOCKS);
701 if (IS_ERR(pmd->bm)) {
702 DMERR("could not create block manager");
703 return PTR_ERR(pmd->bm);
706 r = __open_or_format_metadata(pmd, format_device);
707 if (r)
708 dm_block_manager_destroy(pmd->bm);
710 return r;
713 static void __destroy_persistent_data_objects(struct dm_pool_metadata *pmd)
715 dm_sm_destroy(pmd->data_sm);
716 dm_sm_destroy(pmd->metadata_sm);
717 dm_tm_destroy(pmd->nb_tm);
718 dm_tm_destroy(pmd->tm);
719 dm_block_manager_destroy(pmd->bm);
722 static int __begin_transaction(struct dm_pool_metadata *pmd)
724 int r;
725 struct thin_disk_superblock *disk_super;
726 struct dm_block *sblock;
729 * We re-read the superblock every time. Shouldn't need to do this
730 * really.
732 r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
733 &sb_validator, &sblock);
734 if (r)
735 return r;
737 disk_super = dm_block_data(sblock);
738 pmd->time = le32_to_cpu(disk_super->time);
739 pmd->root = le64_to_cpu(disk_super->data_mapping_root);
740 pmd->details_root = le64_to_cpu(disk_super->device_details_root);
741 pmd->trans_id = le64_to_cpu(disk_super->trans_id);
742 pmd->flags = le32_to_cpu(disk_super->flags);
743 pmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
745 dm_bm_unlock(sblock);
746 return 0;
749 static int __write_changed_details(struct dm_pool_metadata *pmd)
751 int r;
752 struct dm_thin_device *td, *tmp;
753 struct disk_device_details details;
754 uint64_t key;
756 list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
757 if (!td->changed)
758 continue;
760 key = td->id;
762 details.mapped_blocks = cpu_to_le64(td->mapped_blocks);
763 details.transaction_id = cpu_to_le64(td->transaction_id);
764 details.creation_time = cpu_to_le32(td->creation_time);
765 details.snapshotted_time = cpu_to_le32(td->snapshotted_time);
766 __dm_bless_for_disk(&details);
768 r = dm_btree_insert(&pmd->details_info, pmd->details_root,
769 &key, &details, &pmd->details_root);
770 if (r)
771 return r;
773 if (td->open_count)
774 td->changed = 0;
775 else {
776 list_del(&td->list);
777 kfree(td);
781 return 0;
784 static int __commit_transaction(struct dm_pool_metadata *pmd)
786 int r;
787 size_t metadata_len, data_len;
788 struct thin_disk_superblock *disk_super;
789 struct dm_block *sblock;
792 * We need to know if the thin_disk_superblock exceeds a 512-byte sector.
794 BUILD_BUG_ON(sizeof(struct thin_disk_superblock) > 512);
796 r = __write_changed_details(pmd);
797 if (r < 0)
798 return r;
800 r = dm_sm_commit(pmd->data_sm);
801 if (r < 0)
802 return r;
804 r = dm_tm_pre_commit(pmd->tm);
805 if (r < 0)
806 return r;
808 r = dm_sm_root_size(pmd->metadata_sm, &metadata_len);
809 if (r < 0)
810 return r;
812 r = dm_sm_root_size(pmd->data_sm, &data_len);
813 if (r < 0)
814 return r;
816 r = save_sm_roots(pmd);
817 if (r < 0)
818 return r;
820 r = superblock_lock(pmd, &sblock);
821 if (r)
822 return r;
824 disk_super = dm_block_data(sblock);
825 disk_super->time = cpu_to_le32(pmd->time);
826 disk_super->data_mapping_root = cpu_to_le64(pmd->root);
827 disk_super->device_details_root = cpu_to_le64(pmd->details_root);
828 disk_super->trans_id = cpu_to_le64(pmd->trans_id);
829 disk_super->flags = cpu_to_le32(pmd->flags);
831 copy_sm_roots(pmd, disk_super);
833 return dm_tm_commit(pmd->tm, sblock);
836 static void __set_metadata_reserve(struct dm_pool_metadata *pmd)
838 int r;
839 dm_block_t total;
840 dm_block_t max_blocks = 4096; /* 16M */
842 r = dm_sm_get_nr_blocks(pmd->metadata_sm, &total);
843 if (r) {
844 DMERR("could not get size of metadata device");
845 pmd->metadata_reserve = max_blocks;
846 } else
847 pmd->metadata_reserve = min(max_blocks, div_u64(total, 10));
850 struct dm_pool_metadata *dm_pool_metadata_open(struct block_device *bdev,
851 sector_t data_block_size,
852 bool format_device)
854 int r;
855 struct dm_pool_metadata *pmd;
857 pmd = kmalloc(sizeof(*pmd), GFP_KERNEL);
858 if (!pmd) {
859 DMERR("could not allocate metadata struct");
860 return ERR_PTR(-ENOMEM);
863 init_rwsem(&pmd->root_lock);
864 pmd->time = 0;
865 INIT_LIST_HEAD(&pmd->thin_devices);
866 pmd->fail_io = false;
867 pmd->bdev = bdev;
868 pmd->data_block_size = data_block_size;
870 r = __create_persistent_data_objects(pmd, format_device);
871 if (r) {
872 kfree(pmd);
873 return ERR_PTR(r);
876 r = __begin_transaction(pmd);
877 if (r < 0) {
878 if (dm_pool_metadata_close(pmd) < 0)
879 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
880 return ERR_PTR(r);
883 __set_metadata_reserve(pmd);
885 return pmd;
888 int dm_pool_metadata_close(struct dm_pool_metadata *pmd)
890 int r;
891 unsigned open_devices = 0;
892 struct dm_thin_device *td, *tmp;
894 down_read(&pmd->root_lock);
895 list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
896 if (td->open_count)
897 open_devices++;
898 else {
899 list_del(&td->list);
900 kfree(td);
903 up_read(&pmd->root_lock);
905 if (open_devices) {
906 DMERR("attempt to close pmd when %u device(s) are still open",
907 open_devices);
908 return -EBUSY;
911 if (!dm_bm_is_read_only(pmd->bm) && !pmd->fail_io) {
912 r = __commit_transaction(pmd);
913 if (r < 0)
914 DMWARN("%s: __commit_transaction() failed, error = %d",
915 __func__, r);
918 if (!pmd->fail_io)
919 __destroy_persistent_data_objects(pmd);
921 kfree(pmd);
922 return 0;
926 * __open_device: Returns @td corresponding to device with id @dev,
927 * creating it if @create is set and incrementing @td->open_count.
928 * On failure, @td is undefined.
930 static int __open_device(struct dm_pool_metadata *pmd,
931 dm_thin_id dev, int create,
932 struct dm_thin_device **td)
934 int r, changed = 0;
935 struct dm_thin_device *td2;
936 uint64_t key = dev;
937 struct disk_device_details details_le;
940 * If the device is already open, return it.
942 list_for_each_entry(td2, &pmd->thin_devices, list)
943 if (td2->id == dev) {
945 * May not create an already-open device.
947 if (create)
948 return -EEXIST;
950 td2->open_count++;
951 *td = td2;
952 return 0;
956 * Check the device exists.
958 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
959 &key, &details_le);
960 if (r) {
961 if (r != -ENODATA || !create)
962 return r;
965 * Create new device.
967 changed = 1;
968 details_le.mapped_blocks = 0;
969 details_le.transaction_id = cpu_to_le64(pmd->trans_id);
970 details_le.creation_time = cpu_to_le32(pmd->time);
971 details_le.snapshotted_time = cpu_to_le32(pmd->time);
974 *td = kmalloc(sizeof(**td), GFP_NOIO);
975 if (!*td)
976 return -ENOMEM;
978 (*td)->pmd = pmd;
979 (*td)->id = dev;
980 (*td)->open_count = 1;
981 (*td)->changed = changed;
982 (*td)->aborted_with_changes = false;
983 (*td)->mapped_blocks = le64_to_cpu(details_le.mapped_blocks);
984 (*td)->transaction_id = le64_to_cpu(details_le.transaction_id);
985 (*td)->creation_time = le32_to_cpu(details_le.creation_time);
986 (*td)->snapshotted_time = le32_to_cpu(details_le.snapshotted_time);
988 list_add(&(*td)->list, &pmd->thin_devices);
990 return 0;
993 static void __close_device(struct dm_thin_device *td)
995 --td->open_count;
998 static int __create_thin(struct dm_pool_metadata *pmd,
999 dm_thin_id dev)
1001 int r;
1002 dm_block_t dev_root;
1003 uint64_t key = dev;
1004 struct disk_device_details details_le;
1005 struct dm_thin_device *td;
1006 __le64 value;
1008 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
1009 &key, &details_le);
1010 if (!r)
1011 return -EEXIST;
1014 * Create an empty btree for the mappings.
1016 r = dm_btree_empty(&pmd->bl_info, &dev_root);
1017 if (r)
1018 return r;
1021 * Insert it into the main mapping tree.
1023 value = cpu_to_le64(dev_root);
1024 __dm_bless_for_disk(&value);
1025 r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
1026 if (r) {
1027 dm_btree_del(&pmd->bl_info, dev_root);
1028 return r;
1031 r = __open_device(pmd, dev, 1, &td);
1032 if (r) {
1033 dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1034 dm_btree_del(&pmd->bl_info, dev_root);
1035 return r;
1037 __close_device(td);
1039 return r;
1042 int dm_pool_create_thin(struct dm_pool_metadata *pmd, dm_thin_id dev)
1044 int r = -EINVAL;
1046 down_write(&pmd->root_lock);
1047 if (!pmd->fail_io)
1048 r = __create_thin(pmd, dev);
1049 up_write(&pmd->root_lock);
1051 return r;
1054 static int __set_snapshot_details(struct dm_pool_metadata *pmd,
1055 struct dm_thin_device *snap,
1056 dm_thin_id origin, uint32_t time)
1058 int r;
1059 struct dm_thin_device *td;
1061 r = __open_device(pmd, origin, 0, &td);
1062 if (r)
1063 return r;
1065 td->changed = 1;
1066 td->snapshotted_time = time;
1068 snap->mapped_blocks = td->mapped_blocks;
1069 snap->snapshotted_time = time;
1070 __close_device(td);
1072 return 0;
1075 static int __create_snap(struct dm_pool_metadata *pmd,
1076 dm_thin_id dev, dm_thin_id origin)
1078 int r;
1079 dm_block_t origin_root;
1080 uint64_t key = origin, dev_key = dev;
1081 struct dm_thin_device *td;
1082 struct disk_device_details details_le;
1083 __le64 value;
1085 /* check this device is unused */
1086 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
1087 &dev_key, &details_le);
1088 if (!r)
1089 return -EEXIST;
1091 /* find the mapping tree for the origin */
1092 r = dm_btree_lookup(&pmd->tl_info, pmd->root, &key, &value);
1093 if (r)
1094 return r;
1095 origin_root = le64_to_cpu(value);
1097 /* clone the origin, an inc will do */
1098 dm_tm_inc(pmd->tm, origin_root);
1100 /* insert into the main mapping tree */
1101 value = cpu_to_le64(origin_root);
1102 __dm_bless_for_disk(&value);
1103 key = dev;
1104 r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
1105 if (r) {
1106 dm_tm_dec(pmd->tm, origin_root);
1107 return r;
1110 pmd->time++;
1112 r = __open_device(pmd, dev, 1, &td);
1113 if (r)
1114 goto bad;
1116 r = __set_snapshot_details(pmd, td, origin, pmd->time);
1117 __close_device(td);
1119 if (r)
1120 goto bad;
1122 return 0;
1124 bad:
1125 dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1126 dm_btree_remove(&pmd->details_info, pmd->details_root,
1127 &key, &pmd->details_root);
1128 return r;
1131 int dm_pool_create_snap(struct dm_pool_metadata *pmd,
1132 dm_thin_id dev,
1133 dm_thin_id origin)
1135 int r = -EINVAL;
1137 down_write(&pmd->root_lock);
1138 if (!pmd->fail_io)
1139 r = __create_snap(pmd, dev, origin);
1140 up_write(&pmd->root_lock);
1142 return r;
1145 static int __delete_device(struct dm_pool_metadata *pmd, dm_thin_id dev)
1147 int r;
1148 uint64_t key = dev;
1149 struct dm_thin_device *td;
1151 /* TODO: failure should mark the transaction invalid */
1152 r = __open_device(pmd, dev, 0, &td);
1153 if (r)
1154 return r;
1156 if (td->open_count > 1) {
1157 __close_device(td);
1158 return -EBUSY;
1161 list_del(&td->list);
1162 kfree(td);
1163 r = dm_btree_remove(&pmd->details_info, pmd->details_root,
1164 &key, &pmd->details_root);
1165 if (r)
1166 return r;
1168 r = dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1169 if (r)
1170 return r;
1172 return 0;
1175 int dm_pool_delete_thin_device(struct dm_pool_metadata *pmd,
1176 dm_thin_id dev)
1178 int r = -EINVAL;
1180 down_write(&pmd->root_lock);
1181 if (!pmd->fail_io)
1182 r = __delete_device(pmd, dev);
1183 up_write(&pmd->root_lock);
1185 return r;
1188 int dm_pool_set_metadata_transaction_id(struct dm_pool_metadata *pmd,
1189 uint64_t current_id,
1190 uint64_t new_id)
1192 int r = -EINVAL;
1194 down_write(&pmd->root_lock);
1196 if (pmd->fail_io)
1197 goto out;
1199 if (pmd->trans_id != current_id) {
1200 DMERR("mismatched transaction id");
1201 goto out;
1204 pmd->trans_id = new_id;
1205 r = 0;
1207 out:
1208 up_write(&pmd->root_lock);
1210 return r;
1213 int dm_pool_get_metadata_transaction_id(struct dm_pool_metadata *pmd,
1214 uint64_t *result)
1216 int r = -EINVAL;
1218 down_read(&pmd->root_lock);
1219 if (!pmd->fail_io) {
1220 *result = pmd->trans_id;
1221 r = 0;
1223 up_read(&pmd->root_lock);
1225 return r;
1228 static int __reserve_metadata_snap(struct dm_pool_metadata *pmd)
1230 int r, inc;
1231 struct thin_disk_superblock *disk_super;
1232 struct dm_block *copy, *sblock;
1233 dm_block_t held_root;
1236 * We commit to ensure the btree roots which we increment in a
1237 * moment are up to date.
1239 __commit_transaction(pmd);
1242 * Copy the superblock.
1244 dm_sm_inc_block(pmd->metadata_sm, THIN_SUPERBLOCK_LOCATION);
1245 r = dm_tm_shadow_block(pmd->tm, THIN_SUPERBLOCK_LOCATION,
1246 &sb_validator, &copy, &inc);
1247 if (r)
1248 return r;
1250 BUG_ON(!inc);
1252 held_root = dm_block_location(copy);
1253 disk_super = dm_block_data(copy);
1255 if (le64_to_cpu(disk_super->held_root)) {
1256 DMWARN("Pool metadata snapshot already exists: release this before taking another.");
1258 dm_tm_dec(pmd->tm, held_root);
1259 dm_tm_unlock(pmd->tm, copy);
1260 return -EBUSY;
1264 * Wipe the spacemap since we're not publishing this.
1266 memset(&disk_super->data_space_map_root, 0,
1267 sizeof(disk_super->data_space_map_root));
1268 memset(&disk_super->metadata_space_map_root, 0,
1269 sizeof(disk_super->metadata_space_map_root));
1272 * Increment the data structures that need to be preserved.
1274 dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->data_mapping_root));
1275 dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->device_details_root));
1276 dm_tm_unlock(pmd->tm, copy);
1279 * Write the held root into the superblock.
1281 r = superblock_lock(pmd, &sblock);
1282 if (r) {
1283 dm_tm_dec(pmd->tm, held_root);
1284 return r;
1287 disk_super = dm_block_data(sblock);
1288 disk_super->held_root = cpu_to_le64(held_root);
1289 dm_bm_unlock(sblock);
1290 return 0;
1293 int dm_pool_reserve_metadata_snap(struct dm_pool_metadata *pmd)
1295 int r = -EINVAL;
1297 down_write(&pmd->root_lock);
1298 if (!pmd->fail_io)
1299 r = __reserve_metadata_snap(pmd);
1300 up_write(&pmd->root_lock);
1302 return r;
1305 static int __release_metadata_snap(struct dm_pool_metadata *pmd)
1307 int r;
1308 struct thin_disk_superblock *disk_super;
1309 struct dm_block *sblock, *copy;
1310 dm_block_t held_root;
1312 r = superblock_lock(pmd, &sblock);
1313 if (r)
1314 return r;
1316 disk_super = dm_block_data(sblock);
1317 held_root = le64_to_cpu(disk_super->held_root);
1318 disk_super->held_root = cpu_to_le64(0);
1320 dm_bm_unlock(sblock);
1322 if (!held_root) {
1323 DMWARN("No pool metadata snapshot found: nothing to release.");
1324 return -EINVAL;
1327 r = dm_tm_read_lock(pmd->tm, held_root, &sb_validator, &copy);
1328 if (r)
1329 return r;
1331 disk_super = dm_block_data(copy);
1332 dm_btree_del(&pmd->info, le64_to_cpu(disk_super->data_mapping_root));
1333 dm_btree_del(&pmd->details_info, le64_to_cpu(disk_super->device_details_root));
1334 dm_sm_dec_block(pmd->metadata_sm, held_root);
1336 dm_tm_unlock(pmd->tm, copy);
1338 return 0;
1341 int dm_pool_release_metadata_snap(struct dm_pool_metadata *pmd)
1343 int r = -EINVAL;
1345 down_write(&pmd->root_lock);
1346 if (!pmd->fail_io)
1347 r = __release_metadata_snap(pmd);
1348 up_write(&pmd->root_lock);
1350 return r;
1353 static int __get_metadata_snap(struct dm_pool_metadata *pmd,
1354 dm_block_t *result)
1356 int r;
1357 struct thin_disk_superblock *disk_super;
1358 struct dm_block *sblock;
1360 r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
1361 &sb_validator, &sblock);
1362 if (r)
1363 return r;
1365 disk_super = dm_block_data(sblock);
1366 *result = le64_to_cpu(disk_super->held_root);
1368 dm_bm_unlock(sblock);
1370 return 0;
1373 int dm_pool_get_metadata_snap(struct dm_pool_metadata *pmd,
1374 dm_block_t *result)
1376 int r = -EINVAL;
1378 down_read(&pmd->root_lock);
1379 if (!pmd->fail_io)
1380 r = __get_metadata_snap(pmd, result);
1381 up_read(&pmd->root_lock);
1383 return r;
1386 int dm_pool_open_thin_device(struct dm_pool_metadata *pmd, dm_thin_id dev,
1387 struct dm_thin_device **td)
1389 int r = -EINVAL;
1391 down_write(&pmd->root_lock);
1392 if (!pmd->fail_io)
1393 r = __open_device(pmd, dev, 0, td);
1394 up_write(&pmd->root_lock);
1396 return r;
1399 int dm_pool_close_thin_device(struct dm_thin_device *td)
1401 down_write(&td->pmd->root_lock);
1402 __close_device(td);
1403 up_write(&td->pmd->root_lock);
1405 return 0;
1408 dm_thin_id dm_thin_dev_id(struct dm_thin_device *td)
1410 return td->id;
1414 * Check whether @time (of block creation) is older than @td's last snapshot.
1415 * If so then the associated block is shared with the last snapshot device.
1416 * Any block on a device created *after* the device last got snapshotted is
1417 * necessarily not shared.
1419 static bool __snapshotted_since(struct dm_thin_device *td, uint32_t time)
1421 return td->snapshotted_time > time;
1424 int dm_thin_find_block(struct dm_thin_device *td, dm_block_t block,
1425 int can_issue_io, struct dm_thin_lookup_result *result)
1427 int r;
1428 __le64 value;
1429 struct dm_pool_metadata *pmd = td->pmd;
1430 dm_block_t keys[2] = { td->id, block };
1431 struct dm_btree_info *info;
1433 down_read(&pmd->root_lock);
1434 if (pmd->fail_io) {
1435 up_read(&pmd->root_lock);
1436 return -EINVAL;
1439 if (can_issue_io) {
1440 info = &pmd->info;
1441 } else
1442 info = &pmd->nb_info;
1444 r = dm_btree_lookup(info, pmd->root, keys, &value);
1445 if (!r) {
1446 uint64_t block_time = 0;
1447 dm_block_t exception_block;
1448 uint32_t exception_time;
1450 block_time = le64_to_cpu(value);
1451 unpack_block_time(block_time, &exception_block,
1452 &exception_time);
1453 result->block = exception_block;
1454 result->shared = __snapshotted_since(td, exception_time);
1457 up_read(&pmd->root_lock);
1458 return r;
1461 /* FIXME: write a more efficient one in btree */
1462 int dm_thin_find_mapped_range(struct dm_thin_device *td,
1463 dm_block_t begin, dm_block_t end,
1464 dm_block_t *thin_begin, dm_block_t *thin_end,
1465 dm_block_t *pool_begin, bool *maybe_shared)
1467 int r;
1468 dm_block_t pool_end;
1469 struct dm_thin_lookup_result lookup;
1471 if (end < begin)
1472 return -ENODATA;
1475 * Find first mapped block.
1477 while (begin < end) {
1478 r = dm_thin_find_block(td, begin, true, &lookup);
1479 if (r) {
1480 if (r != -ENODATA)
1481 return r;
1482 } else
1483 break;
1485 begin++;
1488 if (begin == end)
1489 return -ENODATA;
1491 *thin_begin = begin;
1492 *pool_begin = lookup.block;
1493 *maybe_shared = lookup.shared;
1495 begin++;
1496 pool_end = *pool_begin + 1;
1497 while (begin != end) {
1498 r = dm_thin_find_block(td, begin, true, &lookup);
1499 if (r) {
1500 if (r == -ENODATA)
1501 break;
1502 else
1503 return r;
1506 if ((lookup.block != pool_end) ||
1507 (lookup.shared != *maybe_shared))
1508 break;
1510 pool_end++;
1511 begin++;
1514 *thin_end = begin;
1515 return 0;
1518 static int __insert(struct dm_thin_device *td, dm_block_t block,
1519 dm_block_t data_block)
1521 int r, inserted;
1522 __le64 value;
1523 struct dm_pool_metadata *pmd = td->pmd;
1524 dm_block_t keys[2] = { td->id, block };
1526 value = cpu_to_le64(pack_block_time(data_block, pmd->time));
1527 __dm_bless_for_disk(&value);
1529 r = dm_btree_insert_notify(&pmd->info, pmd->root, keys, &value,
1530 &pmd->root, &inserted);
1531 if (r)
1532 return r;
1534 td->changed = 1;
1535 if (inserted)
1536 td->mapped_blocks++;
1538 return 0;
1541 int dm_thin_insert_block(struct dm_thin_device *td, dm_block_t block,
1542 dm_block_t data_block)
1544 int r = -EINVAL;
1546 down_write(&td->pmd->root_lock);
1547 if (!td->pmd->fail_io)
1548 r = __insert(td, block, data_block);
1549 up_write(&td->pmd->root_lock);
1551 return r;
1554 static int __remove(struct dm_thin_device *td, dm_block_t block)
1556 int r;
1557 struct dm_pool_metadata *pmd = td->pmd;
1558 dm_block_t keys[2] = { td->id, block };
1560 r = dm_btree_remove(&pmd->info, pmd->root, keys, &pmd->root);
1561 if (r)
1562 return r;
1564 td->mapped_blocks--;
1565 td->changed = 1;
1567 return 0;
1570 static int __remove_range(struct dm_thin_device *td, dm_block_t begin, dm_block_t end)
1572 int r;
1573 unsigned count, total_count = 0;
1574 struct dm_pool_metadata *pmd = td->pmd;
1575 dm_block_t keys[1] = { td->id };
1576 __le64 value;
1577 dm_block_t mapping_root;
1580 * Find the mapping tree
1582 r = dm_btree_lookup(&pmd->tl_info, pmd->root, keys, &value);
1583 if (r)
1584 return r;
1587 * Remove from the mapping tree, taking care to inc the
1588 * ref count so it doesn't get deleted.
1590 mapping_root = le64_to_cpu(value);
1591 dm_tm_inc(pmd->tm, mapping_root);
1592 r = dm_btree_remove(&pmd->tl_info, pmd->root, keys, &pmd->root);
1593 if (r)
1594 return r;
1597 * Remove leaves stops at the first unmapped entry, so we have to
1598 * loop round finding mapped ranges.
1600 while (begin < end) {
1601 r = dm_btree_lookup_next(&pmd->bl_info, mapping_root, &begin, &begin, &value);
1602 if (r == -ENODATA)
1603 break;
1605 if (r)
1606 return r;
1608 if (begin >= end)
1609 break;
1611 r = dm_btree_remove_leaves(&pmd->bl_info, mapping_root, &begin, end, &mapping_root, &count);
1612 if (r)
1613 return r;
1615 total_count += count;
1618 td->mapped_blocks -= total_count;
1619 td->changed = 1;
1622 * Reinsert the mapping tree.
1624 value = cpu_to_le64(mapping_root);
1625 __dm_bless_for_disk(&value);
1626 return dm_btree_insert(&pmd->tl_info, pmd->root, keys, &value, &pmd->root);
1629 int dm_thin_remove_block(struct dm_thin_device *td, dm_block_t block)
1631 int r = -EINVAL;
1633 down_write(&td->pmd->root_lock);
1634 if (!td->pmd->fail_io)
1635 r = __remove(td, block);
1636 up_write(&td->pmd->root_lock);
1638 return r;
1641 int dm_thin_remove_range(struct dm_thin_device *td,
1642 dm_block_t begin, dm_block_t end)
1644 int r = -EINVAL;
1646 down_write(&td->pmd->root_lock);
1647 if (!td->pmd->fail_io)
1648 r = __remove_range(td, begin, end);
1649 up_write(&td->pmd->root_lock);
1651 return r;
1654 int dm_pool_block_is_used(struct dm_pool_metadata *pmd, dm_block_t b, bool *result)
1656 int r;
1657 uint32_t ref_count;
1659 down_read(&pmd->root_lock);
1660 r = dm_sm_get_count(pmd->data_sm, b, &ref_count);
1661 if (!r)
1662 *result = (ref_count != 0);
1663 up_read(&pmd->root_lock);
1665 return r;
1668 bool dm_thin_changed_this_transaction(struct dm_thin_device *td)
1670 int r;
1672 down_read(&td->pmd->root_lock);
1673 r = td->changed;
1674 up_read(&td->pmd->root_lock);
1676 return r;
1679 bool dm_pool_changed_this_transaction(struct dm_pool_metadata *pmd)
1681 bool r = false;
1682 struct dm_thin_device *td, *tmp;
1684 down_read(&pmd->root_lock);
1685 list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
1686 if (td->changed) {
1687 r = td->changed;
1688 break;
1691 up_read(&pmd->root_lock);
1693 return r;
1696 bool dm_thin_aborted_changes(struct dm_thin_device *td)
1698 bool r;
1700 down_read(&td->pmd->root_lock);
1701 r = td->aborted_with_changes;
1702 up_read(&td->pmd->root_lock);
1704 return r;
1707 int dm_pool_alloc_data_block(struct dm_pool_metadata *pmd, dm_block_t *result)
1709 int r = -EINVAL;
1711 down_write(&pmd->root_lock);
1712 if (!pmd->fail_io)
1713 r = dm_sm_new_block(pmd->data_sm, result);
1714 up_write(&pmd->root_lock);
1716 return r;
1719 int dm_pool_commit_metadata(struct dm_pool_metadata *pmd)
1721 int r = -EINVAL;
1723 down_write(&pmd->root_lock);
1724 if (pmd->fail_io)
1725 goto out;
1727 r = __commit_transaction(pmd);
1728 if (r <= 0)
1729 goto out;
1732 * Open the next transaction.
1734 r = __begin_transaction(pmd);
1735 out:
1736 up_write(&pmd->root_lock);
1737 return r;
1740 static void __set_abort_with_changes_flags(struct dm_pool_metadata *pmd)
1742 struct dm_thin_device *td;
1744 list_for_each_entry(td, &pmd->thin_devices, list)
1745 td->aborted_with_changes = td->changed;
1748 int dm_pool_abort_metadata(struct dm_pool_metadata *pmd)
1750 int r = -EINVAL;
1752 down_write(&pmd->root_lock);
1753 if (pmd->fail_io)
1754 goto out;
1756 __set_abort_with_changes_flags(pmd);
1757 __destroy_persistent_data_objects(pmd);
1758 r = __create_persistent_data_objects(pmd, false);
1759 if (r)
1760 pmd->fail_io = true;
1762 out:
1763 up_write(&pmd->root_lock);
1765 return r;
1768 int dm_pool_get_free_block_count(struct dm_pool_metadata *pmd, dm_block_t *result)
1770 int r = -EINVAL;
1772 down_read(&pmd->root_lock);
1773 if (!pmd->fail_io)
1774 r = dm_sm_get_nr_free(pmd->data_sm, result);
1775 up_read(&pmd->root_lock);
1777 return r;
1780 int dm_pool_get_free_metadata_block_count(struct dm_pool_metadata *pmd,
1781 dm_block_t *result)
1783 int r = -EINVAL;
1785 down_read(&pmd->root_lock);
1786 if (!pmd->fail_io)
1787 r = dm_sm_get_nr_free(pmd->metadata_sm, result);
1789 if (!r) {
1790 if (*result < pmd->metadata_reserve)
1791 *result = 0;
1792 else
1793 *result -= pmd->metadata_reserve;
1795 up_read(&pmd->root_lock);
1797 return r;
1800 int dm_pool_get_metadata_dev_size(struct dm_pool_metadata *pmd,
1801 dm_block_t *result)
1803 int r = -EINVAL;
1805 down_read(&pmd->root_lock);
1806 if (!pmd->fail_io)
1807 r = dm_sm_get_nr_blocks(pmd->metadata_sm, result);
1808 up_read(&pmd->root_lock);
1810 return r;
1813 int dm_pool_get_data_dev_size(struct dm_pool_metadata *pmd, dm_block_t *result)
1815 int r = -EINVAL;
1817 down_read(&pmd->root_lock);
1818 if (!pmd->fail_io)
1819 r = dm_sm_get_nr_blocks(pmd->data_sm, result);
1820 up_read(&pmd->root_lock);
1822 return r;
1825 int dm_thin_get_mapped_count(struct dm_thin_device *td, dm_block_t *result)
1827 int r = -EINVAL;
1828 struct dm_pool_metadata *pmd = td->pmd;
1830 down_read(&pmd->root_lock);
1831 if (!pmd->fail_io) {
1832 *result = td->mapped_blocks;
1833 r = 0;
1835 up_read(&pmd->root_lock);
1837 return r;
1840 static int __highest_block(struct dm_thin_device *td, dm_block_t *result)
1842 int r;
1843 __le64 value_le;
1844 dm_block_t thin_root;
1845 struct dm_pool_metadata *pmd = td->pmd;
1847 r = dm_btree_lookup(&pmd->tl_info, pmd->root, &td->id, &value_le);
1848 if (r)
1849 return r;
1851 thin_root = le64_to_cpu(value_le);
1853 return dm_btree_find_highest_key(&pmd->bl_info, thin_root, result);
1856 int dm_thin_get_highest_mapped_block(struct dm_thin_device *td,
1857 dm_block_t *result)
1859 int r = -EINVAL;
1860 struct dm_pool_metadata *pmd = td->pmd;
1862 down_read(&pmd->root_lock);
1863 if (!pmd->fail_io)
1864 r = __highest_block(td, result);
1865 up_read(&pmd->root_lock);
1867 return r;
1870 static int __resize_space_map(struct dm_space_map *sm, dm_block_t new_count)
1872 int r;
1873 dm_block_t old_count;
1875 r = dm_sm_get_nr_blocks(sm, &old_count);
1876 if (r)
1877 return r;
1879 if (new_count == old_count)
1880 return 0;
1882 if (new_count < old_count) {
1883 DMERR("cannot reduce size of space map");
1884 return -EINVAL;
1887 return dm_sm_extend(sm, new_count - old_count);
1890 int dm_pool_resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1892 int r = -EINVAL;
1894 down_write(&pmd->root_lock);
1895 if (!pmd->fail_io)
1896 r = __resize_space_map(pmd->data_sm, new_count);
1897 up_write(&pmd->root_lock);
1899 return r;
1902 int dm_pool_resize_metadata_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1904 int r = -EINVAL;
1906 down_write(&pmd->root_lock);
1907 if (!pmd->fail_io) {
1908 r = __resize_space_map(pmd->metadata_sm, new_count);
1909 if (!r)
1910 __set_metadata_reserve(pmd);
1912 up_write(&pmd->root_lock);
1914 return r;
1917 void dm_pool_metadata_read_only(struct dm_pool_metadata *pmd)
1919 down_write(&pmd->root_lock);
1920 dm_bm_set_read_only(pmd->bm);
1921 up_write(&pmd->root_lock);
1924 void dm_pool_metadata_read_write(struct dm_pool_metadata *pmd)
1926 down_write(&pmd->root_lock);
1927 dm_bm_set_read_write(pmd->bm);
1928 up_write(&pmd->root_lock);
1931 int dm_pool_register_metadata_threshold(struct dm_pool_metadata *pmd,
1932 dm_block_t threshold,
1933 dm_sm_threshold_fn fn,
1934 void *context)
1936 int r;
1938 down_write(&pmd->root_lock);
1939 r = dm_sm_register_threshold_callback(pmd->metadata_sm, threshold, fn, context);
1940 up_write(&pmd->root_lock);
1942 return r;
1945 int dm_pool_metadata_set_needs_check(struct dm_pool_metadata *pmd)
1947 int r;
1948 struct dm_block *sblock;
1949 struct thin_disk_superblock *disk_super;
1951 down_write(&pmd->root_lock);
1952 pmd->flags |= THIN_METADATA_NEEDS_CHECK_FLAG;
1954 r = superblock_lock(pmd, &sblock);
1955 if (r) {
1956 DMERR("couldn't read superblock");
1957 goto out;
1960 disk_super = dm_block_data(sblock);
1961 disk_super->flags = cpu_to_le32(pmd->flags);
1963 dm_bm_unlock(sblock);
1964 out:
1965 up_write(&pmd->root_lock);
1966 return r;
1969 bool dm_pool_metadata_needs_check(struct dm_pool_metadata *pmd)
1971 bool needs_check;
1973 down_read(&pmd->root_lock);
1974 needs_check = pmd->flags & THIN_METADATA_NEEDS_CHECK_FLAG;
1975 up_read(&pmd->root_lock);
1977 return needs_check;
1980 void dm_pool_issue_prefetches(struct dm_pool_metadata *pmd)
1982 down_read(&pmd->root_lock);
1983 if (!pmd->fail_io)
1984 dm_tm_issue_prefetches(pmd->tm);
1985 up_read(&pmd->root_lock);