projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
dm thin metadata: factor out __write_initial_superblock
[deliverable/linux.git] / drivers / md / dm-thin-metadata.c
CommitLineData
991d9fa0
JT		 1/*
2 * Copyright (C) 2011 Red Hat, Inc.
3 *
4 * This file is released under the GPL.
5 */
6
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"
12
13#include <linux/list.h>
14#include <linux/device-mapper.h>
15#include <linux/workqueue.h>
16
17/*--------------------------------------------------------------------------
18 * As far as the metadata goes, there is:
19 *
20 * - A superblock in block zero, taking up fewer than 512 bytes for
21 * atomic writes.
22 *
23 * - A space map managing the metadata blocks.
24 *
25 * - A space map managing the data blocks.
26 *
27 * - A btree mapping our internal thin dev ids onto struct disk_device_details.
28 *
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.
33 *
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.
41 *
42 * Space maps have 2 btrees:
43 *
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.
47 *
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:
50 *
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
55 *
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.
59 *
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.
65 *
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.
70 *
71 * All metadata io is in THIN_METADATA_BLOCK_SIZE sized/aligned chunks
72 * from the block manager.
73 *--------------------------------------------------------------------------*/
74
75#define DM_MSG_PREFIX "thin metadata"
76
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
82
8c971178
JT		 83/*
84 * 3 for btree insert +
85 * 2 for btree lookup used within space map
86 */
87#define THIN_MAX_CONCURRENT_LOCKS 5
88
991d9fa0
JT		 89/* This should be plenty */
90#define SPACE_MAP_ROOT_SIZE 128
91
92/*
93 * Little endian on-disk superblock and device details.
94 */
95struct thin_disk_superblock {
96 __le32 csum; /* Checksum of superblock except for this field. */
97 __le32 flags;
98 __le64 blocknr; /* This block number, dm_block_t. */
99
100 __u8 uuid[16];
101 __le64 magic;
102 __le32 version;
103 __le32 time;
104
105 __le64 trans_id;
106
107 /*
108 * Root held by userspace transactions.
109 */
110 __le64 held_root;
111
112 __u8 data_space_map_root[SPACE_MAP_ROOT_SIZE];
113 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
114
115 /*
116 * 2-level btree mapping (dev_id, (dev block, time)) -> data block
117 */
118 __le64 data_mapping_root;
119
120 /*
121 * Device detail root mapping dev_id -> device_details
122 */
123 __le64 device_details_root;
124
125 __le32 data_block_size; /* In 512-byte sectors. */
126
127 __le32 metadata_block_size; /* In 512-byte sectors. */
128 __le64 metadata_nr_blocks;
129
130 __le32 compat_flags;
131 __le32 compat_ro_flags;
132 __le32 incompat_flags;
133} __packed;
134
135struct disk_device_details {
136 __le64 mapped_blocks;
137 __le64 transaction_id; /* When created. */
138 __le32 creation_time;
139 __le32 snapshotted_time;
140} __packed;
141
142struct dm_pool_metadata {
143 struct hlist_node hash;
144
145 struct block_device *bdev;
146 struct dm_block_manager *bm;
147 struct dm_space_map *metadata_sm;
148 struct dm_space_map *data_sm;
149 struct dm_transaction_manager *tm;
150 struct dm_transaction_manager *nb_tm;
151
152 /*
153 * Two-level btree.
154 * First level holds thin_dev_t.
155 * Second level holds mappings.
156 */
157 struct dm_btree_info info;
158
159 /*
160 * Non-blocking version of the above.
161 */
162 struct dm_btree_info nb_info;
163
164 /*
165 * Just the top level for deleting whole devices.
166 */
167 struct dm_btree_info tl_info;
168
169 /*
170 * Just the bottom level for creating new devices.
171 */
172 struct dm_btree_info bl_info;
173
174 /*
175 * Describes the device details btree.
176 */
177 struct dm_btree_info details_info;
178
179 struct rw_semaphore root_lock;
180 uint32_t time;
991d9fa0
JT		 181 dm_block_t root;
182 dm_block_t details_root;
183 struct list_head thin_devices;
184 uint64_t trans_id;
185 unsigned long flags;
186 sector_t data_block_size;
187};
188
189struct dm_thin_device {
190 struct list_head list;
191 struct dm_pool_metadata *pmd;
192 dm_thin_id id;
193
194 int open_count;
195 int changed;
196 uint64_t mapped_blocks;
197 uint64_t transaction_id;
198 uint32_t creation_time;
199 uint32_t snapshotted_time;
200};
201
202/*----------------------------------------------------------------
203 * superblock validator
204 *--------------------------------------------------------------*/
205
206#define SUPERBLOCK_CSUM_XOR 160774
207
208static void sb_prepare_for_write(struct dm_block_validator *v,
209 struct dm_block *b,
210 size_t block_size)
211{
212 struct thin_disk_superblock *disk_super = dm_block_data(b);
213
214 disk_super->blocknr = cpu_to_le64(dm_block_location(b));
215 disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
216 block_size - sizeof(__le32),
217 SUPERBLOCK_CSUM_XOR));
218}
219
220static int sb_check(struct dm_block_validator *v,
221 struct dm_block *b,
222 size_t block_size)
223{
224 struct thin_disk_superblock *disk_super = dm_block_data(b);
225 __le32 csum_le;
226
227 if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
228 DMERR("sb_check failed: blocknr %llu: "
229 "wanted %llu", le64_to_cpu(disk_super->blocknr),
230 (unsigned long long)dm_block_location(b));
231 return -ENOTBLK;
232 }
233
234 if (le64_to_cpu(disk_super->magic) != THIN_SUPERBLOCK_MAGIC) {
235 DMERR("sb_check failed: magic %llu: "
236 "wanted %llu", le64_to_cpu(disk_super->magic),
237 (unsigned long long)THIN_SUPERBLOCK_MAGIC);
238 return -EILSEQ;
239 }
240
241 csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
242 block_size - sizeof(__le32),
243 SUPERBLOCK_CSUM_XOR));
244 if (csum_le != disk_super->csum) {
245 DMERR("sb_check failed: csum %u: wanted %u",
246 le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
247 return -EILSEQ;
248 }
249
250 return 0;
251}
252
253static struct dm_block_validator sb_validator = {
254 .name = "superblock",
255 .prepare_for_write = sb_prepare_for_write,
256 .check = sb_check
257};
258
259/*----------------------------------------------------------------
260 * Methods for the btree value types
261 *--------------------------------------------------------------*/
262
263static uint64_t pack_block_time(dm_block_t b, uint32_t t)
264{
265 return (b << 24) | t;
266}
267
268static void unpack_block_time(uint64_t v, dm_block_t *b, uint32_t *t)
269{
270 *b = v >> 24;
271 *t = v & ((1 << 24) - 1);
272}
273
274static void data_block_inc(void *context, void *value_le)
275{
276 struct dm_space_map *sm = context;
277 __le64 v_le;
278 uint64_t b;
279 uint32_t t;
280
281 memcpy(&v_le, value_le, sizeof(v_le));
282 unpack_block_time(le64_to_cpu(v_le), &b, &t);
283 dm_sm_inc_block(sm, b);
284}
285
286static void data_block_dec(void *context, void *value_le)
287{
288 struct dm_space_map *sm = context;
289 __le64 v_le;
290 uint64_t b;
291 uint32_t t;
292
293 memcpy(&v_le, value_le, sizeof(v_le));
294 unpack_block_time(le64_to_cpu(v_le), &b, &t);
295 dm_sm_dec_block(sm, b);
296}
297
298static int data_block_equal(void *context, void *value1_le, void *value2_le)
299{
300 __le64 v1_le, v2_le;
301 uint64_t b1, b2;
302 uint32_t t;
303
304 memcpy(&v1_le, value1_le, sizeof(v1_le));
305 memcpy(&v2_le, value2_le, sizeof(v2_le));
306 unpack_block_time(le64_to_cpu(v1_le), &b1, &t);
307 unpack_block_time(le64_to_cpu(v2_le), &b2, &t);
308
309 return b1 == b2;
310}
311
312static void subtree_inc(void *context, void *value)
313{
314 struct dm_btree_info *info = context;
315 __le64 root_le;
316 uint64_t root;
317
318 memcpy(&root_le, value, sizeof(root_le));
319 root = le64_to_cpu(root_le);
320 dm_tm_inc(info->tm, root);
321}
322
323static void subtree_dec(void *context, void *value)
324{
325 struct dm_btree_info *info = context;
326 __le64 root_le;
327 uint64_t root;
328
329 memcpy(&root_le, value, sizeof(root_le));
330 root = le64_to_cpu(root_le);
331 if (dm_btree_del(info, root))
332 DMERR("btree delete failed\n");
333}
334
335static int subtree_equal(void *context, void *value1_le, void *value2_le)
336{
337 __le64 v1_le, v2_le;
338 memcpy(&v1_le, value1_le, sizeof(v1_le));
339 memcpy(&v2_le, value2_le, sizeof(v2_le));
340
341 return v1_le == v2_le;
342}
343
344/*----------------------------------------------------------------*/
345
25971192
JT		 346static int superblock_lock_zero(struct dm_pool_metadata *pmd,
347 struct dm_block **sblock)
348{
349 return dm_bm_write_lock_zero(pmd->bm, THIN_SUPERBLOCK_LOCATION,
350 &sb_validator, sblock);
351}
352
353static int superblock_lock(struct dm_pool_metadata *pmd,
354 struct dm_block **sblock)
355{
356 return dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
357 &sb_validator, sblock);
358}
359
332627db 360static int __superblock_all_zeroes(struct dm_block_manager *bm, int *result)
991d9fa0
JT		 361{
362 int r;
363 unsigned i;
364 struct dm_block *b;
365 __le64 *data_le, zero = cpu_to_le64(0);
366 unsigned block_size = dm_bm_block_size(bm) / sizeof(__le64);
367
368 /*
369 * We can't use a validator here - it may be all zeroes.
370 */
371 r = dm_bm_read_lock(bm, THIN_SUPERBLOCK_LOCATION, NULL, &b);
372 if (r)
373 return r;
374
375 data_le = dm_block_data(b);
376 *result = 1;
377 for (i = 0; i < block_size; i++) {
378 if (data_le[i] != zero) {
379 *result = 0;
380 break;
381 }
382 }
383
384 return dm_bm_unlock(b);
385}
386
41675aea
JT		 387static void __setup_btree_details(struct dm_pool_metadata *pmd)
388{
389 pmd->info.tm = pmd->tm;
390 pmd->info.levels = 2;
391 pmd->info.value_type.context = pmd->data_sm;
392 pmd->info.value_type.size = sizeof(__le64);
393 pmd->info.value_type.inc = data_block_inc;
394 pmd->info.value_type.dec = data_block_dec;
395 pmd->info.value_type.equal = data_block_equal;
396
397 memcpy(&pmd->nb_info, &pmd->info, sizeof(pmd->nb_info));
398 pmd->nb_info.tm = pmd->nb_tm;
399
400 pmd->tl_info.tm = pmd->tm;
401 pmd->tl_info.levels = 1;
402 pmd->tl_info.value_type.context = &pmd->info;
403 pmd->tl_info.value_type.size = sizeof(__le64);
404 pmd->tl_info.value_type.inc = subtree_inc;
405 pmd->tl_info.value_type.dec = subtree_dec;
406 pmd->tl_info.value_type.equal = subtree_equal;
407
408 pmd->bl_info.tm = pmd->tm;
409 pmd->bl_info.levels = 1;
410 pmd->bl_info.value_type.context = pmd->data_sm;
411 pmd->bl_info.value_type.size = sizeof(__le64);
412 pmd->bl_info.value_type.inc = data_block_inc;
413 pmd->bl_info.value_type.dec = data_block_dec;
414 pmd->bl_info.value_type.equal = data_block_equal;
415
416 pmd->details_info.tm = pmd->tm;
417 pmd->details_info.levels = 1;
418 pmd->details_info.value_type.context = NULL;
419 pmd->details_info.value_type.size = sizeof(struct disk_device_details);
420 pmd->details_info.value_type.inc = NULL;
421 pmd->details_info.value_type.dec = NULL;
422 pmd->details_info.value_type.equal = NULL;
423}
424
9cb6653f
JT		 425static int __write_initial_superblock(struct dm_pool_metadata *pmd)
426{
427 int r;
428 struct dm_block *sblock;
429 struct thin_disk_superblock *disk_super;
430 sector_t bdev_size = i_size_read(pmd->bdev->bd_inode) >> SECTOR_SHIFT;
431
432 if (bdev_size > THIN_METADATA_MAX_SECTORS)
433 bdev_size = THIN_METADATA_MAX_SECTORS;
434
435 r = superblock_lock_zero(pmd, &sblock);
436 if (r)
437 return r;
438
439 disk_super = dm_block_data(sblock);
440 disk_super->magic = cpu_to_le64(THIN_SUPERBLOCK_MAGIC);
441 disk_super->version = cpu_to_le32(THIN_VERSION);
442 disk_super->time = 0;
443 disk_super->metadata_block_size = cpu_to_le32(THIN_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
444 disk_super->metadata_nr_blocks = cpu_to_le64(bdev_size >> SECTOR_TO_BLOCK_SHIFT);
445 disk_super->data_block_size = cpu_to_le32(pmd->data_block_size);
446
447 r = dm_bm_unlock(sblock);
448 if (r)
449 return r;
450
451 pmd->flags = 0;
452 r = dm_pool_commit_metadata(pmd);
453 if (r < 0)
454 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
455 __func__, r);
456
457 return r;
458}
459
332627db
JT		 460static int __open_or_format_metadata(struct dm_pool_metadata *pmd,
461 struct dm_block_manager *bm,
462 dm_block_t nr_blocks, int create)
991d9fa0
JT		 463{
464 int r;
465 struct dm_space_map *sm, *data_sm;
466 struct dm_transaction_manager *tm;
467 struct dm_block *sblock;
468
469 if (create) {
384ef0e6 470 r = dm_tm_create_with_sm(bm, THIN_SUPERBLOCK_LOCATION, &tm, &sm);
991d9fa0
JT		 471 if (r < 0) {
472 DMERR("tm_create_with_sm failed");
473 return r;
474 }
475
476 data_sm = dm_sm_disk_create(tm, nr_blocks);
477 if (IS_ERR(data_sm)) {
478 DMERR("sm_disk_create failed");
479 r = PTR_ERR(data_sm);
480 goto bad;
481 }
482 } else {
384ef0e6 483 struct thin_disk_superblock *disk_super;
991d9fa0 484
25971192
JT		 485 r = dm_bm_read_lock(bm, THIN_SUPERBLOCK_LOCATION,
486 &sb_validator, &sblock);
384ef0e6
JT		 487 if (r < 0) {
488 DMERR("couldn't read superblock");
489 return r;
490 }
491
492 disk_super = dm_block_data(sblock);
991d9fa0 493 r = dm_tm_open_with_sm(bm, THIN_SUPERBLOCK_LOCATION,
384ef0e6
JT		 494 disk_super->metadata_space_map_root,
495 sizeof(disk_super->metadata_space_map_root),
496 &tm, &sm);
991d9fa0
JT		 497 if (r < 0) {
498 DMERR("tm_open_with_sm failed");
384ef0e6 499 dm_bm_unlock(sblock);
991d9fa0
JT		 500 return r;
501 }
502
991d9fa0
JT		 503 data_sm = dm_sm_disk_open(tm, disk_super->data_space_map_root,
504 sizeof(disk_super->data_space_map_root));
505 if (IS_ERR(data_sm)) {
506 DMERR("sm_disk_open failed");
384ef0e6 507 dm_bm_unlock(sblock);
991d9fa0
JT		 508 r = PTR_ERR(data_sm);
509 goto bad;
510 }
991d9fa0 511
384ef0e6 512 dm_bm_unlock(sblock);
991d9fa0
JT		 513 }
514
515 pmd->bm = bm;
516 pmd->metadata_sm = sm;
517 pmd->data_sm = data_sm;
518 pmd->tm = tm;
519 pmd->nb_tm = dm_tm_create_non_blocking_clone(tm);
520 if (!pmd->nb_tm) {
521 DMERR("could not create clone tm");
522 r = -ENOMEM;
523 goto bad_data_sm;
524 }
525
41675aea 526 __setup_btree_details(pmd);
991d9fa0 527
6a0ebd31 528 pmd->root = 0;
991d9fa0
JT		 529 pmd->details_root = 0;
530 pmd->trans_id = 0;
531 pmd->flags = 0;
991d9fa0 532
9cb6653f
JT		 533 if (!create)
534 return 0;
535
536 r = dm_btree_empty(&pmd->info, &pmd->root);
537 if (r < 0)
538 goto bad_data_sm;
539
540 r = dm_btree_empty(&pmd->details_info, &pmd->details_root);
541 if (r < 0) {
542 DMERR("couldn't create devices root");
543 goto bad_data_sm;
544 }
545
546 r = __write_initial_superblock(pmd);
547 if (r)
548 goto bad_data_sm;
549
991d9fa0
JT		 550 return 0;
551
552bad_data_sm:
553 dm_sm_destroy(data_sm);
554bad:
555 dm_tm_destroy(tm);
556 dm_sm_destroy(sm);
557
558 return r;
559}
560
332627db
JT		 561static int __create_persistent_data_objects(struct dm_pool_metadata *pmd,
562 dm_block_t nr_blocks, int *create)
563{
564 int r;
565
566 pmd->bm = dm_block_manager_create(pmd->bdev, THIN_METADATA_BLOCK_SIZE,
567 THIN_METADATA_CACHE_SIZE,
568 THIN_MAX_CONCURRENT_LOCKS);
569 if (IS_ERR(pmd->bm)) {
570 DMERR("could not create block manager");
571 return PTR_ERR(pmd->bm);
572 }
573
574 r = __superblock_all_zeroes(pmd->bm, create);
575 if (r) {
576 dm_block_manager_destroy(pmd->bm);
577 return r;
578 }
579
580 r = __open_or_format_metadata(pmd, pmd->bm, nr_blocks, *create);
581 if (r)
582 dm_block_manager_destroy(pmd->bm);
583
584 return r;
585}
586
f9dd9352
JT		 587static void __destroy_persistent_data_objects(struct dm_pool_metadata *pmd)
588{
589 dm_sm_destroy(pmd->data_sm);
590 dm_sm_destroy(pmd->metadata_sm);
591 dm_tm_destroy(pmd->nb_tm);
592 dm_tm_destroy(pmd->tm);
593 dm_block_manager_destroy(pmd->bm);
594}
595
991d9fa0
JT		 596static int __begin_transaction(struct dm_pool_metadata *pmd)
597{
598 int r;
599 u32 features;
600 struct thin_disk_superblock *disk_super;
601 struct dm_block *sblock;
602
991d9fa0
JT		 603 /*
604 * We re-read the superblock every time. Shouldn't need to do this
605 * really.
606 */
607 r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
608 &sb_validator, &sblock);
609 if (r)
610 return r;
611
612 disk_super = dm_block_data(sblock);
613 pmd->time = le32_to_cpu(disk_super->time);
614 pmd->root = le64_to_cpu(disk_super->data_mapping_root);
615 pmd->details_root = le64_to_cpu(disk_super->device_details_root);
616 pmd->trans_id = le64_to_cpu(disk_super->trans_id);
617 pmd->flags = le32_to_cpu(disk_super->flags);
618 pmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
619
620 features = le32_to_cpu(disk_super->incompat_flags) & ~THIN_FEATURE_INCOMPAT_SUPP;
621 if (features) {
622 DMERR("could not access metadata due to "
623 "unsupported optional features (%lx).",
624 (unsigned long)features);
625 r = -EINVAL;
626 goto out;
627 }
628
629 /*
630 * Check for read-only metadata to skip the following RDWR checks.
631 */
632 if (get_disk_ro(pmd->bdev->bd_disk))
633 goto out;
634
635 features = le32_to_cpu(disk_super->compat_ro_flags) & ~THIN_FEATURE_COMPAT_RO_SUPP;
636 if (features) {
637 DMERR("could not access metadata RDWR due to "
638 "unsupported optional features (%lx).",
639 (unsigned long)features);
640 r = -EINVAL;
641 }
642
643out:
644 dm_bm_unlock(sblock);
645 return r;
646}
647
648static int __write_changed_details(struct dm_pool_metadata *pmd)
649{
650 int r;
651 struct dm_thin_device *td, *tmp;
652 struct disk_device_details details;
653 uint64_t key;
654
655 list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
656 if (!td->changed)
657 continue;
658
659 key = td->id;
660
661 details.mapped_blocks = cpu_to_le64(td->mapped_blocks);
662 details.transaction_id = cpu_to_le64(td->transaction_id);
663 details.creation_time = cpu_to_le32(td->creation_time);
664 details.snapshotted_time = cpu_to_le32(td->snapshotted_time);
665 __dm_bless_for_disk(&details);
666
667 r = dm_btree_insert(&pmd->details_info, pmd->details_root,
668 &key, &details, &pmd->details_root);
669 if (r)
670 return r;
671
672 if (td->open_count)
673 td->changed = 0;
674 else {
675 list_del(&td->list);
676 kfree(td);
677 }
991d9fa0
JT		 678 }
679
680 return 0;
681}
682
683static int __commit_transaction(struct dm_pool_metadata *pmd)
684{
685 /*
686 * FIXME: Associated pool should be made read-only on failure.
687 */
688 int r;
689 size_t metadata_len, data_len;
690 struct thin_disk_superblock *disk_super;
691 struct dm_block *sblock;
692
693 /*
694 * We need to know if the thin_disk_superblock exceeds a 512-byte sector.
695 */
696 BUILD_BUG_ON(sizeof(struct thin_disk_superblock) > 512);
697
698 r = __write_changed_details(pmd);
699 if (r < 0)
d973ac19 700 return r;
991d9fa0 701
991d9fa0
JT		 702 r = dm_sm_commit(pmd->data_sm);
703 if (r < 0)
d973ac19 704 return r;
991d9fa0
JT		 705
706 r = dm_tm_pre_commit(pmd->tm);
707 if (r < 0)
d973ac19 708 return r;
991d9fa0
JT		 709
710 r = dm_sm_root_size(pmd->metadata_sm, &metadata_len);
711 if (r < 0)
d973ac19 712 return r;
991d9fa0 713
fef838cc 714 r = dm_sm_root_size(pmd->data_sm, &data_len);
991d9fa0 715 if (r < 0)
d973ac19 716 return r;
991d9fa0 717
25971192 718 r = superblock_lock(pmd, &sblock);
991d9fa0 719 if (r)
d973ac19 720 return r;
991d9fa0
JT		 721
722 disk_super = dm_block_data(sblock);
723 disk_super->time = cpu_to_le32(pmd->time);
724 disk_super->data_mapping_root = cpu_to_le64(pmd->root);
725 disk_super->device_details_root = cpu_to_le64(pmd->details_root);
726 disk_super->trans_id = cpu_to_le64(pmd->trans_id);
727 disk_super->flags = cpu_to_le32(pmd->flags);
728
729 r = dm_sm_copy_root(pmd->metadata_sm, &disk_super->metadata_space_map_root,
730 metadata_len);
731 if (r < 0)
732 goto out_locked;
733
734 r = dm_sm_copy_root(pmd->data_sm, &disk_super->data_space_map_root,
735 data_len);
736 if (r < 0)
737 goto out_locked;
738
eb04cf63 739 return dm_tm_commit(pmd->tm, sblock);
991d9fa0
JT		 740
741out_locked:
742 dm_bm_unlock(sblock);
743 return r;
744}
745
746struct dm_pool_metadata *dm_pool_metadata_open(struct block_device *bdev,
747 sector_t data_block_size)
748{
749 int r;
991d9fa0 750 struct dm_pool_metadata *pmd;
991d9fa0 751 int create;
991d9fa0
JT		 752
753 pmd = kmalloc(sizeof(*pmd), GFP_KERNEL);
754 if (!pmd) {
755 DMERR("could not allocate metadata struct");
756 return ERR_PTR(-ENOMEM);
757 }
758
6a0ebd31
JT		 759 init_rwsem(&pmd->root_lock);
760 pmd->time = 0;
761 INIT_LIST_HEAD(&pmd->thin_devices);
332627db 762 pmd->bdev = bdev;
9cb6653f 763 pmd->data_block_size = data_block_size;
991d9fa0 764
332627db 765 r = __create_persistent_data_objects(pmd, 0, &create);
991d9fa0 766 if (r) {
991d9fa0
JT		 767 kfree(pmd);
768 return ERR_PTR(r);
769 }
991d9fa0
JT		 770
771 if (!create) {
772 r = __begin_transaction(pmd);
9cb6653f
JT		 773 if (r < 0) {
774 if (dm_pool_metadata_close(pmd) < 0)
775 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
776 return ERR_PTR(r);
777 }
991d9fa0
JT		 778 }
779
780 return pmd;
991d9fa0
JT		 781}
782
783int dm_pool_metadata_close(struct dm_pool_metadata *pmd)
784{
785 int r;
786 unsigned open_devices = 0;
787 struct dm_thin_device *td, *tmp;
788
789 down_read(&pmd->root_lock);
790 list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
791 if (td->open_count)
792 open_devices++;
793 else {
794 list_del(&td->list);
795 kfree(td);
796 }
797 }
798 up_read(&pmd->root_lock);
799
800 if (open_devices) {
801 DMERR("attempt to close pmd when %u device(s) are still open",
802 open_devices);
803 return -EBUSY;
804 }
805
806 r = __commit_transaction(pmd);
807 if (r < 0)
808 DMWARN("%s: __commit_transaction() failed, error = %d",
809 __func__, r);
810
f9dd9352 811 __destroy_persistent_data_objects(pmd);
991d9fa0
JT		 812 kfree(pmd);
813
814 return 0;
815}
816
1f3db25d
MS		 817/*
818 * __open_device: Returns @td corresponding to device with id @dev,
819 * creating it if @create is set and incrementing @td->open_count.
820 * On failure, @td is undefined.
821 */
991d9fa0
JT		 822static int __open_device(struct dm_pool_metadata *pmd,
823 dm_thin_id dev, int create,
824 struct dm_thin_device **td)
825{
826 int r, changed = 0;
827 struct dm_thin_device *td2;
828 uint64_t key = dev;
829 struct disk_device_details details_le;
830
831 /*
1f3db25d 832 * If the device is already open, return it.
991d9fa0
JT		 833 */
834 list_for_each_entry(td2, &pmd->thin_devices, list)
835 if (td2->id == dev) {
1f3db25d
MS		 836 /*
837 * May not create an already-open device.
838 */
839 if (create)
840 return -EEXIST;
841
991d9fa0
JT		 842 td2->open_count++;
843 *td = td2;
844 return 0;
845 }
846
847 /*
848 * Check the device exists.
849 */
850 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
851 &key, &details_le);
852 if (r) {
853 if (r != -ENODATA || !create)
854 return r;
855
1f3db25d
MS		 856 /*
857 * Create new device.
858 */
991d9fa0
JT		 859 changed = 1;
860 details_le.mapped_blocks = 0;
861 details_le.transaction_id = cpu_to_le64(pmd->trans_id);
862 details_le.creation_time = cpu_to_le32(pmd->time);
863 details_le.snapshotted_time = cpu_to_le32(pmd->time);
864 }
865
866 *td = kmalloc(sizeof(**td), GFP_NOIO);
867 if (!*td)
868 return -ENOMEM;
869
870 (*td)->pmd = pmd;
871 (*td)->id = dev;
872 (*td)->open_count = 1;
873 (*td)->changed = changed;
874 (*td)->mapped_blocks = le64_to_cpu(details_le.mapped_blocks);
875 (*td)->transaction_id = le64_to_cpu(details_le.transaction_id);
876 (*td)->creation_time = le32_to_cpu(details_le.creation_time);
877 (*td)->snapshotted_time = le32_to_cpu(details_le.snapshotted_time);
878
879 list_add(&(*td)->list, &pmd->thin_devices);
880
881 return 0;
882}
883
884static void __close_device(struct dm_thin_device *td)
885{
886 --td->open_count;
887}
888
889static int __create_thin(struct dm_pool_metadata *pmd,
890 dm_thin_id dev)
891{
892 int r;
893 dm_block_t dev_root;
894 uint64_t key = dev;
895 struct disk_device_details details_le;
896 struct dm_thin_device *td;
897 __le64 value;
898
899 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
900 &key, &details_le);
901 if (!r)
902 return -EEXIST;
903
904 /*
905 * Create an empty btree for the mappings.
906 */
907 r = dm_btree_empty(&pmd->bl_info, &dev_root);
908 if (r)
909 return r;
910
911 /*
912 * Insert it into the main mapping tree.
913 */
914 value = cpu_to_le64(dev_root);
915 __dm_bless_for_disk(&value);
916 r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
917 if (r) {
918 dm_btree_del(&pmd->bl_info, dev_root);
919 return r;
920 }
921
922 r = __open_device(pmd, dev, 1, &td);
923 if (r) {
991d9fa0
JT		 924 dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
925 dm_btree_del(&pmd->bl_info, dev_root);
926 return r;
927 }
991d9fa0
JT		 928 __close_device(td);
929
930 return r;
931}
932
933int dm_pool_create_thin(struct dm_pool_metadata *pmd, dm_thin_id dev)
934{
935 int r;
936
937 down_write(&pmd->root_lock);
938 r = __create_thin(pmd, dev);
939 up_write(&pmd->root_lock);
940
941 return r;
942}
943
944static int __set_snapshot_details(struct dm_pool_metadata *pmd,
945 struct dm_thin_device *snap,
946 dm_thin_id origin, uint32_t time)
947{
948 int r;
949 struct dm_thin_device *td;
950
951 r = __open_device(pmd, origin, 0, &td);
952 if (r)
953 return r;
954
955 td->changed = 1;
956 td->snapshotted_time = time;
957
958 snap->mapped_blocks = td->mapped_blocks;
959 snap->snapshotted_time = time;
960 __close_device(td);
961
962 return 0;
963}
964
965static int __create_snap(struct dm_pool_metadata *pmd,
966 dm_thin_id dev, dm_thin_id origin)
967{
968 int r;
969 dm_block_t origin_root;
970 uint64_t key = origin, dev_key = dev;
971 struct dm_thin_device *td;
972 struct disk_device_details details_le;
973 __le64 value;
974
975 /* check this device is unused */
976 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
977 &dev_key, &details_le);
978 if (!r)
979 return -EEXIST;
980
981 /* find the mapping tree for the origin */
982 r = dm_btree_lookup(&pmd->tl_info, pmd->root, &key, &value);
983 if (r)
984 return r;
985 origin_root = le64_to_cpu(value);
986
987 /* clone the origin, an inc will do */
988 dm_tm_inc(pmd->tm, origin_root);
989
990 /* insert into the main mapping tree */
991 value = cpu_to_le64(origin_root);
992 __dm_bless_for_disk(&value);
993 key = dev;
994 r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
995 if (r) {
996 dm_tm_dec(pmd->tm, origin_root);
997 return r;
998 }
999
1000 pmd->time++;
1001
1002 r = __open_device(pmd, dev, 1, &td);
1003 if (r)
1004 goto bad;
1005
1006 r = __set_snapshot_details(pmd, td, origin, pmd->time);
1f3db25d
MS		 1007 __close_device(td);
1008
991d9fa0
JT		 1009 if (r)
1010 goto bad;
1011
991d9fa0
JT		 1012 return 0;
1013
1014bad:
991d9fa0
JT		 1015 dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1016 dm_btree_remove(&pmd->details_info, pmd->details_root,
1017 &key, &pmd->details_root);
1018 return r;
1019}
1020
1021int dm_pool_create_snap(struct dm_pool_metadata *pmd,
1022 dm_thin_id dev,
1023 dm_thin_id origin)
1024{
1025 int r;
1026
1027 down_write(&pmd->root_lock);
1028 r = __create_snap(pmd, dev, origin);
1029 up_write(&pmd->root_lock);
1030
1031 return r;
1032}
1033
1034static int __delete_device(struct dm_pool_metadata *pmd, dm_thin_id dev)
1035{
1036 int r;
1037 uint64_t key = dev;
1038 struct dm_thin_device *td;
1039
1040 /* TODO: failure should mark the transaction invalid */
1041 r = __open_device(pmd, dev, 0, &td);
1042 if (r)
1043 return r;
1044
1045 if (td->open_count > 1) {
1046 __close_device(td);
1047 return -EBUSY;
1048 }
1049
1050 list_del(&td->list);
1051 kfree(td);
1052 r = dm_btree_remove(&pmd->details_info, pmd->details_root,
1053 &key, &pmd->details_root);
1054 if (r)
1055 return r;
1056
1057 r = dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1058 if (r)
1059 return r;
1060
991d9fa0
JT		 1061 return 0;
1062}
1063
1064int dm_pool_delete_thin_device(struct dm_pool_metadata *pmd,
1065 dm_thin_id dev)
1066{
1067 int r;
1068
1069 down_write(&pmd->root_lock);
1070 r = __delete_device(pmd, dev);
1071 up_write(&pmd->root_lock);
1072
1073 return r;
1074}
1075
1076int dm_pool_set_metadata_transaction_id(struct dm_pool_metadata *pmd,
1077 uint64_t current_id,
1078 uint64_t new_id)
1079{
1080 down_write(&pmd->root_lock);
1081 if (pmd->trans_id != current_id) {
1082 up_write(&pmd->root_lock);
1083 DMERR("mismatched transaction id");
1084 return -EINVAL;
1085 }
1086
1087 pmd->trans_id = new_id;
991d9fa0
JT		 1088 up_write(&pmd->root_lock);
1089
1090 return 0;
1091}
1092
1093int dm_pool_get_metadata_transaction_id(struct dm_pool_metadata *pmd,
1094 uint64_t *result)
1095{
1096 down_read(&pmd->root_lock);
1097 *result = pmd->trans_id;
1098 up_read(&pmd->root_lock);
1099
1100 return 0;
1101}
1102
cc8394d8
JT		 1103static int __reserve_metadata_snap(struct dm_pool_metadata *pmd)
1104{
1105 int r, inc;
1106 struct thin_disk_superblock *disk_super;
1107 struct dm_block *copy, *sblock;
1108 dm_block_t held_root;
1109
1110 /*
1111 * Copy the superblock.
1112 */
1113 dm_sm_inc_block(pmd->metadata_sm, THIN_SUPERBLOCK_LOCATION);
1114 r = dm_tm_shadow_block(pmd->tm, THIN_SUPERBLOCK_LOCATION,
1115 &sb_validator, &copy, &inc);
1116 if (r)
1117 return r;
1118
1119 BUG_ON(!inc);
1120
1121 held_root = dm_block_location(copy);
1122 disk_super = dm_block_data(copy);
1123
1124 if (le64_to_cpu(disk_super->held_root)) {
1125 DMWARN("Pool metadata snapshot already exists: release this before taking another.");
1126
1127 dm_tm_dec(pmd->tm, held_root);
1128 dm_tm_unlock(pmd->tm, copy);
cc8394d8
JT		 1129 return -EBUSY;
1130 }
1131
1132 /*
1133 * Wipe the spacemap since we're not publishing this.
1134 */
1135 memset(&disk_super->data_space_map_root, 0,
1136 sizeof(disk_super->data_space_map_root));
1137 memset(&disk_super->metadata_space_map_root, 0,
1138 sizeof(disk_super->metadata_space_map_root));
1139
1140 /*
1141 * Increment the data structures that need to be preserved.
1142 */
1143 dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->data_mapping_root));
1144 dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->device_details_root));
1145 dm_tm_unlock(pmd->tm, copy);
1146
1147 /*
1148 * Write the held root into the superblock.
1149 */
25971192 1150 r = superblock_lock(pmd, &sblock);
cc8394d8
JT		 1151 if (r) {
1152 dm_tm_dec(pmd->tm, held_root);
cc8394d8
JT		 1153 return r;
1154 }
1155
1156 disk_super = dm_block_data(sblock);
1157 disk_super->held_root = cpu_to_le64(held_root);
1158 dm_bm_unlock(sblock);
cc8394d8
JT		 1159 return 0;
1160}
1161
1162int dm_pool_reserve_metadata_snap(struct dm_pool_metadata *pmd)
1163{
1164 int r;
1165
1166 down_write(&pmd->root_lock);
1167 r = __reserve_metadata_snap(pmd);
1168 up_write(&pmd->root_lock);
1169
1170 return r;
1171}
1172
1173static int __release_metadata_snap(struct dm_pool_metadata *pmd)
991d9fa0
JT		 1174{
1175 int r;
1176 struct thin_disk_superblock *disk_super;
cc8394d8
JT		 1177 struct dm_block *sblock, *copy;
1178 dm_block_t held_root;
991d9fa0 1179
25971192 1180 r = superblock_lock(pmd, &sblock);
991d9fa0
JT		 1181 if (r)
1182 return r;
1183
cc8394d8
JT		 1184 disk_super = dm_block_data(sblock);
1185 held_root = le64_to_cpu(disk_super->held_root);
1186 disk_super->held_root = cpu_to_le64(0);
cc8394d8
JT		 1187
1188 dm_bm_unlock(sblock);
1189
1190 if (!held_root) {
1191 DMWARN("No pool metadata snapshot found: nothing to release.");
1192 return -EINVAL;
1193 }
1194
1195 r = dm_tm_read_lock(pmd->tm, held_root, &sb_validator, &copy);
1196 if (r)
1197 return r;
1198
1199 disk_super = dm_block_data(copy);
1200 dm_sm_dec_block(pmd->metadata_sm, le64_to_cpu(disk_super->data_mapping_root));
1201 dm_sm_dec_block(pmd->metadata_sm, le64_to_cpu(disk_super->device_details_root));
1202 dm_sm_dec_block(pmd->metadata_sm, held_root);
1203
1204 return dm_tm_unlock(pmd->tm, copy);
1205}
1206
1207int dm_pool_release_metadata_snap(struct dm_pool_metadata *pmd)
1208{
1209 int r;
1210
1211 down_write(&pmd->root_lock);
1212 r = __release_metadata_snap(pmd);
1213 up_write(&pmd->root_lock);
1214
1215 return r;
1216}
1217
1218static int __get_metadata_snap(struct dm_pool_metadata *pmd,
1219 dm_block_t *result)
1220{
1221 int r;
1222 struct thin_disk_superblock *disk_super;
1223 struct dm_block *sblock;
1224
1225 r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
1226 &sb_validator, &sblock);
1227 if (r)
1228 return r;
1229
991d9fa0
JT		 1230 disk_super = dm_block_data(sblock);
1231 *result = le64_to_cpu(disk_super->held_root);
1232
1233 return dm_bm_unlock(sblock);
1234}
1235
cc8394d8
JT		 1236int dm_pool_get_metadata_snap(struct dm_pool_metadata *pmd,
1237 dm_block_t *result)
991d9fa0
JT		 1238{
1239 int r;
1240
1241 down_read(&pmd->root_lock);
cc8394d8 1242 r = __get_metadata_snap(pmd, result);
991d9fa0
JT		 1243 up_read(&pmd->root_lock);
1244
1245 return r;
1246}
1247
1248int dm_pool_open_thin_device(struct dm_pool_metadata *pmd, dm_thin_id dev,
1249 struct dm_thin_device **td)
1250{
1251 int r;
1252
1253 down_write(&pmd->root_lock);
1254 r = __open_device(pmd, dev, 0, td);
1255 up_write(&pmd->root_lock);
1256
1257 return r;
1258}
1259
1260int dm_pool_close_thin_device(struct dm_thin_device *td)
1261{
1262 down_write(&td->pmd->root_lock);
1263 __close_device(td);
1264 up_write(&td->pmd->root_lock);
1265
1266 return 0;
1267}
1268
1269dm_thin_id dm_thin_dev_id(struct dm_thin_device *td)
1270{
1271 return td->id;
1272}
1273
17b7d63f 1274static bool __snapshotted_since(struct dm_thin_device *td, uint32_t time)
991d9fa0
JT		 1275{
1276 return td->snapshotted_time > time;
1277}
1278
1279int dm_thin_find_block(struct dm_thin_device *td, dm_block_t block,
1280 int can_block, struct dm_thin_lookup_result *result)
1281{
1282 int r;
1283 uint64_t block_time = 0;
1284 __le64 value;
1285 struct dm_pool_metadata *pmd = td->pmd;
1286 dm_block_t keys[2] = { td->id, block };
1287
1288 if (can_block) {
1289 down_read(&pmd->root_lock);
1290 r = dm_btree_lookup(&pmd->info, pmd->root, keys, &value);
1291 if (!r)
1292 block_time = le64_to_cpu(value);
1293 up_read(&pmd->root_lock);
1294
1295 } else if (down_read_trylock(&pmd->root_lock)) {
1296 r = dm_btree_lookup(&pmd->nb_info, pmd->root, keys, &value);
1297 if (!r)
1298 block_time = le64_to_cpu(value);
1299 up_read(&pmd->root_lock);
1300
1301 } else
1302 return -EWOULDBLOCK;
1303
1304 if (!r) {
1305 dm_block_t exception_block;
1306 uint32_t exception_time;
1307 unpack_block_time(block_time, &exception_block,
1308 &exception_time);
1309 result->block = exception_block;
1310 result->shared = __snapshotted_since(td, exception_time);
1311 }
1312
1313 return r;
1314}
1315
1316static int __insert(struct dm_thin_device *td, dm_block_t block,
1317 dm_block_t data_block)
1318{
1319 int r, inserted;
1320 __le64 value;
1321 struct dm_pool_metadata *pmd = td->pmd;
1322 dm_block_t keys[2] = { td->id, block };
1323
991d9fa0
JT		 1324 value = cpu_to_le64(pack_block_time(data_block, pmd->time));
1325 __dm_bless_for_disk(&value);
1326
1327 r = dm_btree_insert_notify(&pmd->info, pmd->root, keys, &value,
1328 &pmd->root, &inserted);
1329 if (r)
1330 return r;
1331
1332 if (inserted) {
1333 td->mapped_blocks++;
1334 td->changed = 1;
1335 }
1336
1337 return 0;
1338}
1339
1340int dm_thin_insert_block(struct dm_thin_device *td, dm_block_t block,
1341 dm_block_t data_block)
1342{
1343 int r;
1344
1345 down_write(&td->pmd->root_lock);
1346 r = __insert(td, block, data_block);
1347 up_write(&td->pmd->root_lock);
1348
1349 return r;
1350}
1351
1352static int __remove(struct dm_thin_device *td, dm_block_t block)
1353{
1354 int r;
1355 struct dm_pool_metadata *pmd = td->pmd;
1356 dm_block_t keys[2] = { td->id, block };
1357
1358 r = dm_btree_remove(&pmd->info, pmd->root, keys, &pmd->root);
1359 if (r)
1360 return r;
1361
af63bcb8
JT		 1362 td->mapped_blocks--;
1363 td->changed = 1;
991d9fa0
JT		 1364
1365 return 0;
1366}
1367
1368int dm_thin_remove_block(struct dm_thin_device *td, dm_block_t block)
1369{
1370 int r;
1371
1372 down_write(&td->pmd->root_lock);
1373 r = __remove(td, block);
1374 up_write(&td->pmd->root_lock);
1375
1376 return r;
1377}
1378
1379int dm_pool_alloc_data_block(struct dm_pool_metadata *pmd, dm_block_t *result)
1380{
1381 int r;
1382
1383 down_write(&pmd->root_lock);
991d9fa0 1384 r = dm_sm_new_block(pmd->data_sm, result);
991d9fa0
JT		 1385 up_write(&pmd->root_lock);
1386
1387 return r;
1388}
1389
1390int dm_pool_commit_metadata(struct dm_pool_metadata *pmd)
1391{
1392 int r;
1393
1394 down_write(&pmd->root_lock);
1395
1396 r = __commit_transaction(pmd);
1397 if (r <= 0)
1398 goto out;
1399
1400 /*
1401 * Open the next transaction.
1402 */
1403 r = __begin_transaction(pmd);
1404out:
1405 up_write(&pmd->root_lock);
1406 return r;
1407}
1408
1409int dm_pool_get_free_block_count(struct dm_pool_metadata *pmd, dm_block_t *result)
1410{
1411 int r;
1412
1413 down_read(&pmd->root_lock);
1414 r = dm_sm_get_nr_free(pmd->data_sm, result);
1415 up_read(&pmd->root_lock);
1416
1417 return r;
1418}
1419
1420int dm_pool_get_free_metadata_block_count(struct dm_pool_metadata *pmd,
1421 dm_block_t *result)
1422{
1423 int r;
1424
1425 down_read(&pmd->root_lock);
1426 r = dm_sm_get_nr_free(pmd->metadata_sm, result);
1427 up_read(&pmd->root_lock);
1428
1429 return r;
1430}
1431
1432int dm_pool_get_metadata_dev_size(struct dm_pool_metadata *pmd,
1433 dm_block_t *result)
1434{
1435 int r;
1436
1437 down_read(&pmd->root_lock);
1438 r = dm_sm_get_nr_blocks(pmd->metadata_sm, result);
1439 up_read(&pmd->root_lock);
1440
1441 return r;
1442}
1443
1444int dm_pool_get_data_block_size(struct dm_pool_metadata *pmd, sector_t *result)
1445{
1446 down_read(&pmd->root_lock);
1447 *result = pmd->data_block_size;
1448 up_read(&pmd->root_lock);
1449
1450 return 0;
1451}
1452
1453int dm_pool_get_data_dev_size(struct dm_pool_metadata *pmd, dm_block_t *result)
1454{
1455 int r;
1456
1457 down_read(&pmd->root_lock);
1458 r = dm_sm_get_nr_blocks(pmd->data_sm, result);
1459 up_read(&pmd->root_lock);
1460
1461 return r;
1462}
1463
1464int dm_thin_get_mapped_count(struct dm_thin_device *td, dm_block_t *result)
1465{
1466 struct dm_pool_metadata *pmd = td->pmd;
1467
1468 down_read(&pmd->root_lock);
1469 *result = td->mapped_blocks;
1470 up_read(&pmd->root_lock);
1471
1472 return 0;
1473}
1474
1475static int __highest_block(struct dm_thin_device *td, dm_block_t *result)
1476{
1477 int r;
1478 __le64 value_le;
1479 dm_block_t thin_root;
1480 struct dm_pool_metadata *pmd = td->pmd;
1481
1482 r = dm_btree_lookup(&pmd->tl_info, pmd->root, &td->id, &value_le);
1483 if (r)
1484 return r;
1485
1486 thin_root = le64_to_cpu(value_le);
1487
1488 return dm_btree_find_highest_key(&pmd->bl_info, thin_root, result);
1489}
1490
1491int dm_thin_get_highest_mapped_block(struct dm_thin_device *td,
1492 dm_block_t *result)
1493{
1494 int r;
1495 struct dm_pool_metadata *pmd = td->pmd;
1496
1497 down_read(&pmd->root_lock);
1498 r = __highest_block(td, result);
1499 up_read(&pmd->root_lock);
1500
1501 return r;
1502}
1503
1504static int __resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1505{
1506 int r;
1507 dm_block_t old_count;
1508
1509 r = dm_sm_get_nr_blocks(pmd->data_sm, &old_count);
1510 if (r)
1511 return r;
1512
1513 if (new_count == old_count)
1514 return 0;
1515
1516 if (new_count < old_count) {
1517 DMERR("cannot reduce size of data device");
1518 return -EINVAL;
1519 }
1520
eb04cf63 1521 return dm_sm_extend(pmd->data_sm, new_count - old_count);
991d9fa0
JT		 1522}
1523
1524int dm_pool_resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1525{
1526 int r;
1527
1528 down_write(&pmd->root_lock);
1529 r = __resize_data_dev(pmd, new_count);
1530 up_write(&pmd->root_lock);
1531
1532 return r;
1533}
Linux kernel - Deliverables
This page took 0.134076 seconds and 5 git commands to generate.