projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
dm thin metadata: remove pointless label from __commit_transaction
[GitHub/mt8127/android_kernel_alcatel_ttab.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
83/* This should be plenty */
84#define SPACE_MAP_ROOT_SIZE 128
85
86/*
87 * Little endian on-disk superblock and device details.
88 */
89struct thin_disk_superblock {
90 __le32 csum; /* Checksum of superblock except for this field. */
91 __le32 flags;
92 __le64 blocknr; /* This block number, dm_block_t. */
93
94 __u8 uuid[16];
95 __le64 magic;
96 __le32 version;
97 __le32 time;
98
99 __le64 trans_id;
100
101 /*
102 * Root held by userspace transactions.
103 */
104 __le64 held_root;
105
106 __u8 data_space_map_root[SPACE_MAP_ROOT_SIZE];
107 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
108
109 /*
110 * 2-level btree mapping (dev_id, (dev block, time)) -> data block
111 */
112 __le64 data_mapping_root;
113
114 /*
115 * Device detail root mapping dev_id -> device_details
116 */
117 __le64 device_details_root;
118
119 __le32 data_block_size; /* In 512-byte sectors. */
120
121 __le32 metadata_block_size; /* In 512-byte sectors. */
122 __le64 metadata_nr_blocks;
123
124 __le32 compat_flags;
125 __le32 compat_ro_flags;
126 __le32 incompat_flags;
127} __packed;
128
129struct disk_device_details {
130 __le64 mapped_blocks;
131 __le64 transaction_id; /* When created. */
132 __le32 creation_time;
133 __le32 snapshotted_time;
134} __packed;
135
136struct dm_pool_metadata {
137 struct hlist_node hash;
138
139 struct block_device *bdev;
140 struct dm_block_manager *bm;
141 struct dm_space_map *metadata_sm;
142 struct dm_space_map *data_sm;
143 struct dm_transaction_manager *tm;
144 struct dm_transaction_manager *nb_tm;
145
146 /*
147 * Two-level btree.
148 * First level holds thin_dev_t.
149 * Second level holds mappings.
150 */
151 struct dm_btree_info info;
152
153 /*
154 * Non-blocking version of the above.
155 */
156 struct dm_btree_info nb_info;
157
158 /*
159 * Just the top level for deleting whole devices.
160 */
161 struct dm_btree_info tl_info;
162
163 /*
164 * Just the bottom level for creating new devices.
165 */
166 struct dm_btree_info bl_info;
167
168 /*
169 * Describes the device details btree.
170 */
171 struct dm_btree_info details_info;
172
173 struct rw_semaphore root_lock;
174 uint32_t time;
175 int need_commit;
176 dm_block_t root;
177 dm_block_t details_root;
178 struct list_head thin_devices;
179 uint64_t trans_id;
180 unsigned long flags;
181 sector_t data_block_size;
182};
183
184struct dm_thin_device {
185 struct list_head list;
186 struct dm_pool_metadata *pmd;
187 dm_thin_id id;
188
189 int open_count;
190 int changed;
191 uint64_t mapped_blocks;
192 uint64_t transaction_id;
193 uint32_t creation_time;
194 uint32_t snapshotted_time;
195};
196
197/*----------------------------------------------------------------
198 * superblock validator
199 *--------------------------------------------------------------*/
200
201#define SUPERBLOCK_CSUM_XOR 160774
202
203static void sb_prepare_for_write(struct dm_block_validator *v,
204 struct dm_block *b,
205 size_t block_size)
206{
207 struct thin_disk_superblock *disk_super = dm_block_data(b);
208
209 disk_super->blocknr = cpu_to_le64(dm_block_location(b));
210 disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
211 block_size - sizeof(__le32),
212 SUPERBLOCK_CSUM_XOR));
213}
214
215static int sb_check(struct dm_block_validator *v,
216 struct dm_block *b,
217 size_t block_size)
218{
219 struct thin_disk_superblock *disk_super = dm_block_data(b);
220 __le32 csum_le;
221
222 if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
223 DMERR("sb_check failed: blocknr %llu: "
224 "wanted %llu", le64_to_cpu(disk_super->blocknr),
225 (unsigned long long)dm_block_location(b));
226 return -ENOTBLK;
227 }
228
229 if (le64_to_cpu(disk_super->magic) != THIN_SUPERBLOCK_MAGIC) {
230 DMERR("sb_check failed: magic %llu: "
231 "wanted %llu", le64_to_cpu(disk_super->magic),
232 (unsigned long long)THIN_SUPERBLOCK_MAGIC);
233 return -EILSEQ;
234 }
235
236 csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
237 block_size - sizeof(__le32),
238 SUPERBLOCK_CSUM_XOR));
239 if (csum_le != disk_super->csum) {
240 DMERR("sb_check failed: csum %u: wanted %u",
241 le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
242 return -EILSEQ;
243 }
244
245 return 0;
246}
247
248static struct dm_block_validator sb_validator = {
249 .name = "superblock",
250 .prepare_for_write = sb_prepare_for_write,
251 .check = sb_check
252};
253
254/*----------------------------------------------------------------
255 * Methods for the btree value types
256 *--------------------------------------------------------------*/
257
258static uint64_t pack_block_time(dm_block_t b, uint32_t t)
259{
260 return (b << 24) | t;
261}
262
263static void unpack_block_time(uint64_t v, dm_block_t *b, uint32_t *t)
264{
265 *b = v >> 24;
266 *t = v & ((1 << 24) - 1);
267}
268
269static void data_block_inc(void *context, void *value_le)
270{
271 struct dm_space_map *sm = context;
272 __le64 v_le;
273 uint64_t b;
274 uint32_t t;
275
276 memcpy(&v_le, value_le, sizeof(v_le));
277 unpack_block_time(le64_to_cpu(v_le), &b, &t);
278 dm_sm_inc_block(sm, b);
279}
280
281static void data_block_dec(void *context, void *value_le)
282{
283 struct dm_space_map *sm = context;
284 __le64 v_le;
285 uint64_t b;
286 uint32_t t;
287
288 memcpy(&v_le, value_le, sizeof(v_le));
289 unpack_block_time(le64_to_cpu(v_le), &b, &t);
290 dm_sm_dec_block(sm, b);
291}
292
293static int data_block_equal(void *context, void *value1_le, void *value2_le)
294{
295 __le64 v1_le, v2_le;
296 uint64_t b1, b2;
297 uint32_t t;
298
299 memcpy(&v1_le, value1_le, sizeof(v1_le));
300 memcpy(&v2_le, value2_le, sizeof(v2_le));
301 unpack_block_time(le64_to_cpu(v1_le), &b1, &t);
302 unpack_block_time(le64_to_cpu(v2_le), &b2, &t);
303
304 return b1 == b2;
305}
306
307static void subtree_inc(void *context, void *value)
308{
309 struct dm_btree_info *info = context;
310 __le64 root_le;
311 uint64_t root;
312
313 memcpy(&root_le, value, sizeof(root_le));
314 root = le64_to_cpu(root_le);
315 dm_tm_inc(info->tm, root);
316}
317
318static void subtree_dec(void *context, void *value)
319{
320 struct dm_btree_info *info = context;
321 __le64 root_le;
322 uint64_t root;
323
324 memcpy(&root_le, value, sizeof(root_le));
325 root = le64_to_cpu(root_le);
326 if (dm_btree_del(info, root))
327 DMERR("btree delete failed\n");
328}
329
330static int subtree_equal(void *context, void *value1_le, void *value2_le)
331{
332 __le64 v1_le, v2_le;
333 memcpy(&v1_le, value1_le, sizeof(v1_le));
334 memcpy(&v2_le, value2_le, sizeof(v2_le));
335
336 return v1_le == v2_le;
337}
338
339/*----------------------------------------------------------------*/
340
341static int superblock_all_zeroes(struct dm_block_manager *bm, int *result)
342{
343 int r;
344 unsigned i;
345 struct dm_block *b;
346 __le64 *data_le, zero = cpu_to_le64(0);
347 unsigned block_size = dm_bm_block_size(bm) / sizeof(__le64);
348
349 /*
350 * We can't use a validator here - it may be all zeroes.
351 */
352 r = dm_bm_read_lock(bm, THIN_SUPERBLOCK_LOCATION, NULL, &b);
353 if (r)
354 return r;
355
356 data_le = dm_block_data(b);
357 *result = 1;
358 for (i = 0; i < block_size; i++) {
359 if (data_le[i] != zero) {
360 *result = 0;
361 break;
362 }
363 }
364
365 return dm_bm_unlock(b);
366}
367
368static int init_pmd(struct dm_pool_metadata *pmd,
369 struct dm_block_manager *bm,
370 dm_block_t nr_blocks, int create)
371{
372 int r;
373 struct dm_space_map *sm, *data_sm;
374 struct dm_transaction_manager *tm;
375 struct dm_block *sblock;
376
377 if (create) {
378 r = dm_tm_create_with_sm(bm, THIN_SUPERBLOCK_LOCATION,
379 &sb_validator, &tm, &sm, &sblock);
380 if (r < 0) {
381 DMERR("tm_create_with_sm failed");
382 return r;
383 }
384
385 data_sm = dm_sm_disk_create(tm, nr_blocks);
386 if (IS_ERR(data_sm)) {
387 DMERR("sm_disk_create failed");
4469a5f3 388 dm_tm_unlock(tm, sblock);
991d9fa0
JT		 389 r = PTR_ERR(data_sm);
390 goto bad;
391 }
392 } else {
393 struct thin_disk_superblock *disk_super = NULL;
394 size_t space_map_root_offset =
395 offsetof(struct thin_disk_superblock, metadata_space_map_root);
396
397 r = dm_tm_open_with_sm(bm, THIN_SUPERBLOCK_LOCATION,
398 &sb_validator, space_map_root_offset,
399 SPACE_MAP_ROOT_SIZE, &tm, &sm, &sblock);
400 if (r < 0) {
401 DMERR("tm_open_with_sm failed");
402 return r;
403 }
404
405 disk_super = dm_block_data(sblock);
406 data_sm = dm_sm_disk_open(tm, disk_super->data_space_map_root,
407 sizeof(disk_super->data_space_map_root));
408 if (IS_ERR(data_sm)) {
409 DMERR("sm_disk_open failed");
410 r = PTR_ERR(data_sm);
411 goto bad;
412 }
413 }
414
415
416 r = dm_tm_unlock(tm, sblock);
417 if (r < 0) {
418 DMERR("couldn't unlock superblock");
419 goto bad_data_sm;
420 }
421
422 pmd->bm = bm;
423 pmd->metadata_sm = sm;
424 pmd->data_sm = data_sm;
425 pmd->tm = tm;
426 pmd->nb_tm = dm_tm_create_non_blocking_clone(tm);
427 if (!pmd->nb_tm) {
428 DMERR("could not create clone tm");
429 r = -ENOMEM;
430 goto bad_data_sm;
431 }
432
433 pmd->info.tm = tm;
434 pmd->info.levels = 2;
435 pmd->info.value_type.context = pmd->data_sm;
436 pmd->info.value_type.size = sizeof(__le64);
437 pmd->info.value_type.inc = data_block_inc;
438 pmd->info.value_type.dec = data_block_dec;
439 pmd->info.value_type.equal = data_block_equal;
440
441 memcpy(&pmd->nb_info, &pmd->info, sizeof(pmd->nb_info));
442 pmd->nb_info.tm = pmd->nb_tm;
443
444 pmd->tl_info.tm = tm;
445 pmd->tl_info.levels = 1;
446 pmd->tl_info.value_type.context = &pmd->info;
447 pmd->tl_info.value_type.size = sizeof(__le64);
448 pmd->tl_info.value_type.inc = subtree_inc;
449 pmd->tl_info.value_type.dec = subtree_dec;
450 pmd->tl_info.value_type.equal = subtree_equal;
451
452 pmd->bl_info.tm = tm;
453 pmd->bl_info.levels = 1;
454 pmd->bl_info.value_type.context = pmd->data_sm;
455 pmd->bl_info.value_type.size = sizeof(__le64);
456 pmd->bl_info.value_type.inc = data_block_inc;
457 pmd->bl_info.value_type.dec = data_block_dec;
458 pmd->bl_info.value_type.equal = data_block_equal;
459
460 pmd->details_info.tm = tm;
461 pmd->details_info.levels = 1;
462 pmd->details_info.value_type.context = NULL;
463 pmd->details_info.value_type.size = sizeof(struct disk_device_details);
464 pmd->details_info.value_type.inc = NULL;
465 pmd->details_info.value_type.dec = NULL;
466 pmd->details_info.value_type.equal = NULL;
467
468 pmd->root = 0;
469
470 init_rwsem(&pmd->root_lock);
471 pmd->time = 0;
472 pmd->need_commit = 0;
473 pmd->details_root = 0;
474 pmd->trans_id = 0;
475 pmd->flags = 0;
476 INIT_LIST_HEAD(&pmd->thin_devices);
477
478 return 0;
479
480bad_data_sm:
481 dm_sm_destroy(data_sm);
482bad:
483 dm_tm_destroy(tm);
484 dm_sm_destroy(sm);
485
486 return r;
487}
488
489static int __begin_transaction(struct dm_pool_metadata *pmd)
490{
491 int r;
492 u32 features;
493 struct thin_disk_superblock *disk_super;
494 struct dm_block *sblock;
495
496 /*
497 * __maybe_commit_transaction() resets these
498 */
499 WARN_ON(pmd->need_commit);
500
501 /*
502 * We re-read the superblock every time. Shouldn't need to do this
503 * really.
504 */
505 r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
506 &sb_validator, &sblock);
507 if (r)
508 return r;
509
510 disk_super = dm_block_data(sblock);
511 pmd->time = le32_to_cpu(disk_super->time);
512 pmd->root = le64_to_cpu(disk_super->data_mapping_root);
513 pmd->details_root = le64_to_cpu(disk_super->device_details_root);
514 pmd->trans_id = le64_to_cpu(disk_super->trans_id);
515 pmd->flags = le32_to_cpu(disk_super->flags);
516 pmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
517
518 features = le32_to_cpu(disk_super->incompat_flags) & ~THIN_FEATURE_INCOMPAT_SUPP;
519 if (features) {
520 DMERR("could not access metadata due to "
521 "unsupported optional features (%lx).",
522 (unsigned long)features);
523 r = -EINVAL;
524 goto out;
525 }
526
527 /*
528 * Check for read-only metadata to skip the following RDWR checks.
529 */
530 if (get_disk_ro(pmd->bdev->bd_disk))
531 goto out;
532
533 features = le32_to_cpu(disk_super->compat_ro_flags) & ~THIN_FEATURE_COMPAT_RO_SUPP;
534 if (features) {
535 DMERR("could not access metadata RDWR due to "
536 "unsupported optional features (%lx).",
537 (unsigned long)features);
538 r = -EINVAL;
539 }
540
541out:
542 dm_bm_unlock(sblock);
543 return r;
544}
545
546static int __write_changed_details(struct dm_pool_metadata *pmd)
547{
548 int r;
549 struct dm_thin_device *td, *tmp;
550 struct disk_device_details details;
551 uint64_t key;
552
553 list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
554 if (!td->changed)
555 continue;
556
557 key = td->id;
558
559 details.mapped_blocks = cpu_to_le64(td->mapped_blocks);
560 details.transaction_id = cpu_to_le64(td->transaction_id);
561 details.creation_time = cpu_to_le32(td->creation_time);
562 details.snapshotted_time = cpu_to_le32(td->snapshotted_time);
563 __dm_bless_for_disk(&details);
564
565 r = dm_btree_insert(&pmd->details_info, pmd->details_root,
566 &key, &details, &pmd->details_root);
567 if (r)
568 return r;
569
570 if (td->open_count)
571 td->changed = 0;
572 else {
573 list_del(&td->list);
574 kfree(td);
575 }
576
577 pmd->need_commit = 1;
578 }
579
580 return 0;
581}
582
583static int __commit_transaction(struct dm_pool_metadata *pmd)
584{
585 /*
586 * FIXME: Associated pool should be made read-only on failure.
587 */
588 int r;
589 size_t metadata_len, data_len;
590 struct thin_disk_superblock *disk_super;
591 struct dm_block *sblock;
592
593 /*
594 * We need to know if the thin_disk_superblock exceeds a 512-byte sector.
595 */
596 BUILD_BUG_ON(sizeof(struct thin_disk_superblock) > 512);
597
598 r = __write_changed_details(pmd);
599 if (r < 0)
d973ac19 600 return r;
991d9fa0
JT		 601
602 if (!pmd->need_commit)
d973ac19 603 return r;
991d9fa0
JT		 604
605 r = dm_sm_commit(pmd->data_sm);
606 if (r < 0)
d973ac19 607 return r;
991d9fa0
JT		 608
609 r = dm_tm_pre_commit(pmd->tm);
610 if (r < 0)
d973ac19 611 return r;
991d9fa0
JT		 612
613 r = dm_sm_root_size(pmd->metadata_sm, &metadata_len);
614 if (r < 0)
d973ac19 615 return r;
991d9fa0 616
fef838cc 617 r = dm_sm_root_size(pmd->data_sm, &data_len);
991d9fa0 618 if (r < 0)
d973ac19 619 return r;
991d9fa0
JT		 620
621 r = dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
622 &sb_validator, &sblock);
623 if (r)
d973ac19 624 return r;
991d9fa0
JT		 625
626 disk_super = dm_block_data(sblock);
627 disk_super->time = cpu_to_le32(pmd->time);
628 disk_super->data_mapping_root = cpu_to_le64(pmd->root);
629 disk_super->device_details_root = cpu_to_le64(pmd->details_root);
630 disk_super->trans_id = cpu_to_le64(pmd->trans_id);
631 disk_super->flags = cpu_to_le32(pmd->flags);
632
633 r = dm_sm_copy_root(pmd->metadata_sm, &disk_super->metadata_space_map_root,
634 metadata_len);
635 if (r < 0)
636 goto out_locked;
637
638 r = dm_sm_copy_root(pmd->data_sm, &disk_super->data_space_map_root,
639 data_len);
640 if (r < 0)
641 goto out_locked;
642
643 r = dm_tm_commit(pmd->tm, sblock);
644 if (!r)
645 pmd->need_commit = 0;
646
991d9fa0
JT		 647 return r;
648
649out_locked:
650 dm_bm_unlock(sblock);
651 return r;
652}
653
654struct dm_pool_metadata *dm_pool_metadata_open(struct block_device *bdev,
655 sector_t data_block_size)
656{
657 int r;
658 struct thin_disk_superblock *disk_super;
659 struct dm_pool_metadata *pmd;
660 sector_t bdev_size = i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
661 struct dm_block_manager *bm;
662 int create;
663 struct dm_block *sblock;
664
665 pmd = kmalloc(sizeof(*pmd), GFP_KERNEL);
666 if (!pmd) {
667 DMERR("could not allocate metadata struct");
668 return ERR_PTR(-ENOMEM);
669 }
670
671 /*
672 * Max hex locks:
673 * 3 for btree insert +
674 * 2 for btree lookup used within space map
675 */
676 bm = dm_block_manager_create(bdev, THIN_METADATA_BLOCK_SIZE,
677 THIN_METADATA_CACHE_SIZE, 5);
678 if (!bm) {
679 DMERR("could not create block manager");
680 kfree(pmd);
681 return ERR_PTR(-ENOMEM);
682 }
683
684 r = superblock_all_zeroes(bm, &create);
685 if (r) {
686 dm_block_manager_destroy(bm);
687 kfree(pmd);
688 return ERR_PTR(r);
689 }
690
691
692 r = init_pmd(pmd, bm, 0, create);
693 if (r) {
694 dm_block_manager_destroy(bm);
695 kfree(pmd);
696 return ERR_PTR(r);
697 }
698 pmd->bdev = bdev;
699
700 if (!create) {
701 r = __begin_transaction(pmd);
702 if (r < 0)
703 goto bad;
704 return pmd;
705 }
706
707 /*
708 * Create.
709 */
710 r = dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
711 &sb_validator, &sblock);
712 if (r)
713 goto bad;
714
c4a69ecd
MS		 715 if (bdev_size > THIN_METADATA_MAX_SECTORS)
716 bdev_size = THIN_METADATA_MAX_SECTORS;
717
991d9fa0
JT		 718 disk_super = dm_block_data(sblock);
719 disk_super->magic = cpu_to_le64(THIN_SUPERBLOCK_MAGIC);
720 disk_super->version = cpu_to_le32(THIN_VERSION);
721 disk_super->time = 0;
722 disk_super->metadata_block_size = cpu_to_le32(THIN_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
723 disk_super->metadata_nr_blocks = cpu_to_le64(bdev_size >> SECTOR_TO_BLOCK_SHIFT);
724 disk_super->data_block_size = cpu_to_le32(data_block_size);
725
726 r = dm_bm_unlock(sblock);
727 if (r < 0)
728 goto bad;
729
730 r = dm_btree_empty(&pmd->info, &pmd->root);
731 if (r < 0)
732 goto bad;
733
734 r = dm_btree_empty(&pmd->details_info, &pmd->details_root);
735 if (r < 0) {
736 DMERR("couldn't create devices root");
737 goto bad;
738 }
739
740 pmd->flags = 0;
741 pmd->need_commit = 1;
742 r = dm_pool_commit_metadata(pmd);
743 if (r < 0) {
744 DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
745 __func__, r);
746 goto bad;
747 }
748
749 return pmd;
750
751bad:
752 if (dm_pool_metadata_close(pmd) < 0)
753 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
754 return ERR_PTR(r);
755}
756
757int dm_pool_metadata_close(struct dm_pool_metadata *pmd)
758{
759 int r;
760 unsigned open_devices = 0;
761 struct dm_thin_device *td, *tmp;
762
763 down_read(&pmd->root_lock);
764 list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
765 if (td->open_count)
766 open_devices++;
767 else {
768 list_del(&td->list);
769 kfree(td);
770 }
771 }
772 up_read(&pmd->root_lock);
773
774 if (open_devices) {
775 DMERR("attempt to close pmd when %u device(s) are still open",
776 open_devices);
777 return -EBUSY;
778 }
779
780 r = __commit_transaction(pmd);
781 if (r < 0)
782 DMWARN("%s: __commit_transaction() failed, error = %d",
783 __func__, r);
784
785 dm_tm_destroy(pmd->tm);
786 dm_tm_destroy(pmd->nb_tm);
787 dm_block_manager_destroy(pmd->bm);
788 dm_sm_destroy(pmd->metadata_sm);
789 dm_sm_destroy(pmd->data_sm);
790 kfree(pmd);
791
792 return 0;
793}
794
1f3db25d
MS		 795/*
796 * __open_device: Returns @td corresponding to device with id @dev,
797 * creating it if @create is set and incrementing @td->open_count.
798 * On failure, @td is undefined.
799 */
991d9fa0
JT		 800static int __open_device(struct dm_pool_metadata *pmd,
801 dm_thin_id dev, int create,
802 struct dm_thin_device **td)
803{
804 int r, changed = 0;
805 struct dm_thin_device *td2;
806 uint64_t key = dev;
807 struct disk_device_details details_le;
808
809 /*
1f3db25d 810 * If the device is already open, return it.
991d9fa0
JT		 811 */
812 list_for_each_entry(td2, &pmd->thin_devices, list)
813 if (td2->id == dev) {
1f3db25d
MS		 814 /*
815 * May not create an already-open device.
816 */
817 if (create)
818 return -EEXIST;
819
991d9fa0
JT		 820 td2->open_count++;
821 *td = td2;
822 return 0;
823 }
824
825 /*
826 * Check the device exists.
827 */
828 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
829 &key, &details_le);
830 if (r) {
831 if (r != -ENODATA || !create)
832 return r;
833
1f3db25d
MS		 834 /*
835 * Create new device.
836 */
991d9fa0
JT		 837 changed = 1;
838 details_le.mapped_blocks = 0;
839 details_le.transaction_id = cpu_to_le64(pmd->trans_id);
840 details_le.creation_time = cpu_to_le32(pmd->time);
841 details_le.snapshotted_time = cpu_to_le32(pmd->time);
842 }
843
844 *td = kmalloc(sizeof(**td), GFP_NOIO);
845 if (!*td)
846 return -ENOMEM;
847
848 (*td)->pmd = pmd;
849 (*td)->id = dev;
850 (*td)->open_count = 1;
851 (*td)->changed = changed;
852 (*td)->mapped_blocks = le64_to_cpu(details_le.mapped_blocks);
853 (*td)->transaction_id = le64_to_cpu(details_le.transaction_id);
854 (*td)->creation_time = le32_to_cpu(details_le.creation_time);
855 (*td)->snapshotted_time = le32_to_cpu(details_le.snapshotted_time);
856
857 list_add(&(*td)->list, &pmd->thin_devices);
858
859 return 0;
860}
861
862static void __close_device(struct dm_thin_device *td)
863{
864 --td->open_count;
865}
866
867static int __create_thin(struct dm_pool_metadata *pmd,
868 dm_thin_id dev)
869{
870 int r;
871 dm_block_t dev_root;
872 uint64_t key = dev;
873 struct disk_device_details details_le;
874 struct dm_thin_device *td;
875 __le64 value;
876
877 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
878 &key, &details_le);
879 if (!r)
880 return -EEXIST;
881
882 /*
883 * Create an empty btree for the mappings.
884 */
885 r = dm_btree_empty(&pmd->bl_info, &dev_root);
886 if (r)
887 return r;
888
889 /*
890 * Insert it into the main mapping tree.
891 */
892 value = cpu_to_le64(dev_root);
893 __dm_bless_for_disk(&value);
894 r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
895 if (r) {
896 dm_btree_del(&pmd->bl_info, dev_root);
897 return r;
898 }
899
900 r = __open_device(pmd, dev, 1, &td);
901 if (r) {
991d9fa0
JT		 902 dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
903 dm_btree_del(&pmd->bl_info, dev_root);
904 return r;
905 }
991d9fa0
JT		 906 __close_device(td);
907
908 return r;
909}
910
911int dm_pool_create_thin(struct dm_pool_metadata *pmd, dm_thin_id dev)
912{
913 int r;
914
915 down_write(&pmd->root_lock);
916 r = __create_thin(pmd, dev);
917 up_write(&pmd->root_lock);
918
919 return r;
920}
921
922static int __set_snapshot_details(struct dm_pool_metadata *pmd,
923 struct dm_thin_device *snap,
924 dm_thin_id origin, uint32_t time)
925{
926 int r;
927 struct dm_thin_device *td;
928
929 r = __open_device(pmd, origin, 0, &td);
930 if (r)
931 return r;
932
933 td->changed = 1;
934 td->snapshotted_time = time;
935
936 snap->mapped_blocks = td->mapped_blocks;
937 snap->snapshotted_time = time;
938 __close_device(td);
939
940 return 0;
941}
942
943static int __create_snap(struct dm_pool_metadata *pmd,
944 dm_thin_id dev, dm_thin_id origin)
945{
946 int r;
947 dm_block_t origin_root;
948 uint64_t key = origin, dev_key = dev;
949 struct dm_thin_device *td;
950 struct disk_device_details details_le;
951 __le64 value;
952
953 /* check this device is unused */
954 r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
955 &dev_key, &details_le);
956 if (!r)
957 return -EEXIST;
958
959 /* find the mapping tree for the origin */
960 r = dm_btree_lookup(&pmd->tl_info, pmd->root, &key, &value);
961 if (r)
962 return r;
963 origin_root = le64_to_cpu(value);
964
965 /* clone the origin, an inc will do */
966 dm_tm_inc(pmd->tm, origin_root);
967
968 /* insert into the main mapping tree */
969 value = cpu_to_le64(origin_root);
970 __dm_bless_for_disk(&value);
971 key = dev;
972 r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
973 if (r) {
974 dm_tm_dec(pmd->tm, origin_root);
975 return r;
976 }
977
978 pmd->time++;
979
980 r = __open_device(pmd, dev, 1, &td);
981 if (r)
982 goto bad;
983
984 r = __set_snapshot_details(pmd, td, origin, pmd->time);
1f3db25d
MS		 985 __close_device(td);
986
991d9fa0
JT		 987 if (r)
988 goto bad;
989
991d9fa0
JT		 990 return 0;
991
992bad:
991d9fa0
JT		 993 dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
994 dm_btree_remove(&pmd->details_info, pmd->details_root,
995 &key, &pmd->details_root);
996 return r;
997}
998
999int dm_pool_create_snap(struct dm_pool_metadata *pmd,
1000 dm_thin_id dev,
1001 dm_thin_id origin)
1002{
1003 int r;
1004
1005 down_write(&pmd->root_lock);
1006 r = __create_snap(pmd, dev, origin);
1007 up_write(&pmd->root_lock);
1008
1009 return r;
1010}
1011
1012static int __delete_device(struct dm_pool_metadata *pmd, dm_thin_id dev)
1013{
1014 int r;
1015 uint64_t key = dev;
1016 struct dm_thin_device *td;
1017
1018 /* TODO: failure should mark the transaction invalid */
1019 r = __open_device(pmd, dev, 0, &td);
1020 if (r)
1021 return r;
1022
1023 if (td->open_count > 1) {
1024 __close_device(td);
1025 return -EBUSY;
1026 }
1027
1028 list_del(&td->list);
1029 kfree(td);
1030 r = dm_btree_remove(&pmd->details_info, pmd->details_root,
1031 &key, &pmd->details_root);
1032 if (r)
1033 return r;
1034
1035 r = dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1036 if (r)
1037 return r;
1038
1039 pmd->need_commit = 1;
1040
1041 return 0;
1042}
1043
1044int dm_pool_delete_thin_device(struct dm_pool_metadata *pmd,
1045 dm_thin_id dev)
1046{
1047 int r;
1048
1049 down_write(&pmd->root_lock);
1050 r = __delete_device(pmd, dev);
1051 up_write(&pmd->root_lock);
1052
1053 return r;
1054}
1055
1056int dm_pool_set_metadata_transaction_id(struct dm_pool_metadata *pmd,
1057 uint64_t current_id,
1058 uint64_t new_id)
1059{
1060 down_write(&pmd->root_lock);
1061 if (pmd->trans_id != current_id) {
1062 up_write(&pmd->root_lock);
1063 DMERR("mismatched transaction id");
1064 return -EINVAL;
1065 }
1066
1067 pmd->trans_id = new_id;
1068 pmd->need_commit = 1;
1069 up_write(&pmd->root_lock);
1070
1071 return 0;
1072}
1073
1074int dm_pool_get_metadata_transaction_id(struct dm_pool_metadata *pmd,
1075 uint64_t *result)
1076{
1077 down_read(&pmd->root_lock);
1078 *result = pmd->trans_id;
1079 up_read(&pmd->root_lock);
1080
1081 return 0;
1082}
1083
cc8394d8
JT		 1084static int __reserve_metadata_snap(struct dm_pool_metadata *pmd)
1085{
1086 int r, inc;
1087 struct thin_disk_superblock *disk_super;
1088 struct dm_block *copy, *sblock;
1089 dm_block_t held_root;
1090
1091 /*
1092 * Copy the superblock.
1093 */
1094 dm_sm_inc_block(pmd->metadata_sm, THIN_SUPERBLOCK_LOCATION);
1095 r = dm_tm_shadow_block(pmd->tm, THIN_SUPERBLOCK_LOCATION,
1096 &sb_validator, &copy, &inc);
1097 if (r)
1098 return r;
1099
1100 BUG_ON(!inc);
1101
1102 held_root = dm_block_location(copy);
1103 disk_super = dm_block_data(copy);
1104
1105 if (le64_to_cpu(disk_super->held_root)) {
1106 DMWARN("Pool metadata snapshot already exists: release this before taking another.");
1107
1108 dm_tm_dec(pmd->tm, held_root);
1109 dm_tm_unlock(pmd->tm, copy);
1110 pmd->need_commit = 1;
1111
1112 return -EBUSY;
1113 }
1114
1115 /*
1116 * Wipe the spacemap since we're not publishing this.
1117 */
1118 memset(&disk_super->data_space_map_root, 0,
1119 sizeof(disk_super->data_space_map_root));
1120 memset(&disk_super->metadata_space_map_root, 0,
1121 sizeof(disk_super->metadata_space_map_root));
1122
1123 /*
1124 * Increment the data structures that need to be preserved.
1125 */
1126 dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->data_mapping_root));
1127 dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->device_details_root));
1128 dm_tm_unlock(pmd->tm, copy);
1129
1130 /*
1131 * Write the held root into the superblock.
1132 */
1133 r = dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
1134 &sb_validator, &sblock);
1135 if (r) {
1136 dm_tm_dec(pmd->tm, held_root);
1137 pmd->need_commit = 1;
1138 return r;
1139 }
1140
1141 disk_super = dm_block_data(sblock);
1142 disk_super->held_root = cpu_to_le64(held_root);
1143 dm_bm_unlock(sblock);
1144
1145 pmd->need_commit = 1;
1146
1147 return 0;
1148}
1149
1150int dm_pool_reserve_metadata_snap(struct dm_pool_metadata *pmd)
1151{
1152 int r;
1153
1154 down_write(&pmd->root_lock);
1155 r = __reserve_metadata_snap(pmd);
1156 up_write(&pmd->root_lock);
1157
1158 return r;
1159}
1160
1161static int __release_metadata_snap(struct dm_pool_metadata *pmd)
991d9fa0
JT		 1162{
1163 int r;
1164 struct thin_disk_superblock *disk_super;
cc8394d8
JT		 1165 struct dm_block *sblock, *copy;
1166 dm_block_t held_root;
991d9fa0
JT		 1167
1168 r = dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
1169 &sb_validator, &sblock);
1170 if (r)
1171 return r;
1172
cc8394d8
JT		 1173 disk_super = dm_block_data(sblock);
1174 held_root = le64_to_cpu(disk_super->held_root);
1175 disk_super->held_root = cpu_to_le64(0);
1176 pmd->need_commit = 1;
1177
1178 dm_bm_unlock(sblock);
1179
1180 if (!held_root) {
1181 DMWARN("No pool metadata snapshot found: nothing to release.");
1182 return -EINVAL;
1183 }
1184
1185 r = dm_tm_read_lock(pmd->tm, held_root, &sb_validator, &copy);
1186 if (r)
1187 return r;
1188
1189 disk_super = dm_block_data(copy);
1190 dm_sm_dec_block(pmd->metadata_sm, le64_to_cpu(disk_super->data_mapping_root));
1191 dm_sm_dec_block(pmd->metadata_sm, le64_to_cpu(disk_super->device_details_root));
1192 dm_sm_dec_block(pmd->metadata_sm, held_root);
1193
1194 return dm_tm_unlock(pmd->tm, copy);
1195}
1196
1197int dm_pool_release_metadata_snap(struct dm_pool_metadata *pmd)
1198{
1199 int r;
1200
1201 down_write(&pmd->root_lock);
1202 r = __release_metadata_snap(pmd);
1203 up_write(&pmd->root_lock);
1204
1205 return r;
1206}
1207
1208static int __get_metadata_snap(struct dm_pool_metadata *pmd,
1209 dm_block_t *result)
1210{
1211 int r;
1212 struct thin_disk_superblock *disk_super;
1213 struct dm_block *sblock;
1214
1215 r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
1216 &sb_validator, &sblock);
1217 if (r)
1218 return r;
1219
991d9fa0
JT		 1220 disk_super = dm_block_data(sblock);
1221 *result = le64_to_cpu(disk_super->held_root);
1222
1223 return dm_bm_unlock(sblock);
1224}
1225
cc8394d8
JT		 1226int dm_pool_get_metadata_snap(struct dm_pool_metadata *pmd,
1227 dm_block_t *result)
991d9fa0
JT		 1228{
1229 int r;
1230
1231 down_read(&pmd->root_lock);
cc8394d8 1232 r = __get_metadata_snap(pmd, result);
991d9fa0
JT		 1233 up_read(&pmd->root_lock);
1234
1235 return r;
1236}
1237
1238int dm_pool_open_thin_device(struct dm_pool_metadata *pmd, dm_thin_id dev,
1239 struct dm_thin_device **td)
1240{
1241 int r;
1242
1243 down_write(&pmd->root_lock);
1244 r = __open_device(pmd, dev, 0, td);
1245 up_write(&pmd->root_lock);
1246
1247 return r;
1248}
1249
1250int dm_pool_close_thin_device(struct dm_thin_device *td)
1251{
1252 down_write(&td->pmd->root_lock);
1253 __close_device(td);
1254 up_write(&td->pmd->root_lock);
1255
1256 return 0;
1257}
1258
1259dm_thin_id dm_thin_dev_id(struct dm_thin_device *td)
1260{
1261 return td->id;
1262}
1263
17b7d63f 1264static bool __snapshotted_since(struct dm_thin_device *td, uint32_t time)
991d9fa0
JT		 1265{
1266 return td->snapshotted_time > time;
1267}
1268
1269int dm_thin_find_block(struct dm_thin_device *td, dm_block_t block,
1270 int can_block, struct dm_thin_lookup_result *result)
1271{
1272 int r;
1273 uint64_t block_time = 0;
1274 __le64 value;
1275 struct dm_pool_metadata *pmd = td->pmd;
1276 dm_block_t keys[2] = { td->id, block };
1277
1278 if (can_block) {
1279 down_read(&pmd->root_lock);
1280 r = dm_btree_lookup(&pmd->info, pmd->root, keys, &value);
1281 if (!r)
1282 block_time = le64_to_cpu(value);
1283 up_read(&pmd->root_lock);
1284
1285 } else if (down_read_trylock(&pmd->root_lock)) {
1286 r = dm_btree_lookup(&pmd->nb_info, pmd->root, keys, &value);
1287 if (!r)
1288 block_time = le64_to_cpu(value);
1289 up_read(&pmd->root_lock);
1290
1291 } else
1292 return -EWOULDBLOCK;
1293
1294 if (!r) {
1295 dm_block_t exception_block;
1296 uint32_t exception_time;
1297 unpack_block_time(block_time, &exception_block,
1298 &exception_time);
1299 result->block = exception_block;
1300 result->shared = __snapshotted_since(td, exception_time);
1301 }
1302
1303 return r;
1304}
1305
1306static int __insert(struct dm_thin_device *td, dm_block_t block,
1307 dm_block_t data_block)
1308{
1309 int r, inserted;
1310 __le64 value;
1311 struct dm_pool_metadata *pmd = td->pmd;
1312 dm_block_t keys[2] = { td->id, block };
1313
1314 pmd->need_commit = 1;
1315 value = cpu_to_le64(pack_block_time(data_block, pmd->time));
1316 __dm_bless_for_disk(&value);
1317
1318 r = dm_btree_insert_notify(&pmd->info, pmd->root, keys, &value,
1319 &pmd->root, &inserted);
1320 if (r)
1321 return r;
1322
1323 if (inserted) {
1324 td->mapped_blocks++;
1325 td->changed = 1;
1326 }
1327
1328 return 0;
1329}
1330
1331int dm_thin_insert_block(struct dm_thin_device *td, dm_block_t block,
1332 dm_block_t data_block)
1333{
1334 int r;
1335
1336 down_write(&td->pmd->root_lock);
1337 r = __insert(td, block, data_block);
1338 up_write(&td->pmd->root_lock);
1339
1340 return r;
1341}
1342
1343static int __remove(struct dm_thin_device *td, dm_block_t block)
1344{
1345 int r;
1346 struct dm_pool_metadata *pmd = td->pmd;
1347 dm_block_t keys[2] = { td->id, block };
1348
1349 r = dm_btree_remove(&pmd->info, pmd->root, keys, &pmd->root);
1350 if (r)
1351 return r;
1352
af63bcb8
JT		 1353 td->mapped_blocks--;
1354 td->changed = 1;
991d9fa0
JT		 1355 pmd->need_commit = 1;
1356
1357 return 0;
1358}
1359
1360int dm_thin_remove_block(struct dm_thin_device *td, dm_block_t block)
1361{
1362 int r;
1363
1364 down_write(&td->pmd->root_lock);
1365 r = __remove(td, block);
1366 up_write(&td->pmd->root_lock);
1367
1368 return r;
1369}
1370
1371int dm_pool_alloc_data_block(struct dm_pool_metadata *pmd, dm_block_t *result)
1372{
1373 int r;
1374
1375 down_write(&pmd->root_lock);
1376
1377 r = dm_sm_new_block(pmd->data_sm, result);
1378 pmd->need_commit = 1;
1379
1380 up_write(&pmd->root_lock);
1381
1382 return r;
1383}
1384
1385int dm_pool_commit_metadata(struct dm_pool_metadata *pmd)
1386{
1387 int r;
1388
1389 down_write(&pmd->root_lock);
1390
1391 r = __commit_transaction(pmd);
1392 if (r <= 0)
1393 goto out;
1394
1395 /*
1396 * Open the next transaction.
1397 */
1398 r = __begin_transaction(pmd);
1399out:
1400 up_write(&pmd->root_lock);
1401 return r;
1402}
1403
1404int dm_pool_get_free_block_count(struct dm_pool_metadata *pmd, dm_block_t *result)
1405{
1406 int r;
1407
1408 down_read(&pmd->root_lock);
1409 r = dm_sm_get_nr_free(pmd->data_sm, result);
1410 up_read(&pmd->root_lock);
1411
1412 return r;
1413}
1414
1415int dm_pool_get_free_metadata_block_count(struct dm_pool_metadata *pmd,
1416 dm_block_t *result)
1417{
1418 int r;
1419
1420 down_read(&pmd->root_lock);
1421 r = dm_sm_get_nr_free(pmd->metadata_sm, result);
1422 up_read(&pmd->root_lock);
1423
1424 return r;
1425}
1426
1427int dm_pool_get_metadata_dev_size(struct dm_pool_metadata *pmd,
1428 dm_block_t *result)
1429{
1430 int r;
1431
1432 down_read(&pmd->root_lock);
1433 r = dm_sm_get_nr_blocks(pmd->metadata_sm, result);
1434 up_read(&pmd->root_lock);
1435
1436 return r;
1437}
1438
1439int dm_pool_get_data_block_size(struct dm_pool_metadata *pmd, sector_t *result)
1440{
1441 down_read(&pmd->root_lock);
1442 *result = pmd->data_block_size;
1443 up_read(&pmd->root_lock);
1444
1445 return 0;
1446}
1447
1448int dm_pool_get_data_dev_size(struct dm_pool_metadata *pmd, dm_block_t *result)
1449{
1450 int r;
1451
1452 down_read(&pmd->root_lock);
1453 r = dm_sm_get_nr_blocks(pmd->data_sm, result);
1454 up_read(&pmd->root_lock);
1455
1456 return r;
1457}
1458
1459int dm_thin_get_mapped_count(struct dm_thin_device *td, dm_block_t *result)
1460{
1461 struct dm_pool_metadata *pmd = td->pmd;
1462
1463 down_read(&pmd->root_lock);
1464 *result = td->mapped_blocks;
1465 up_read(&pmd->root_lock);
1466
1467 return 0;
1468}
1469
1470static int __highest_block(struct dm_thin_device *td, dm_block_t *result)
1471{
1472 int r;
1473 __le64 value_le;
1474 dm_block_t thin_root;
1475 struct dm_pool_metadata *pmd = td->pmd;
1476
1477 r = dm_btree_lookup(&pmd->tl_info, pmd->root, &td->id, &value_le);
1478 if (r)
1479 return r;
1480
1481 thin_root = le64_to_cpu(value_le);
1482
1483 return dm_btree_find_highest_key(&pmd->bl_info, thin_root, result);
1484}
1485
1486int dm_thin_get_highest_mapped_block(struct dm_thin_device *td,
1487 dm_block_t *result)
1488{
1489 int r;
1490 struct dm_pool_metadata *pmd = td->pmd;
1491
1492 down_read(&pmd->root_lock);
1493 r = __highest_block(td, result);
1494 up_read(&pmd->root_lock);
1495
1496 return r;
1497}
1498
1499static int __resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1500{
1501 int r;
1502 dm_block_t old_count;
1503
1504 r = dm_sm_get_nr_blocks(pmd->data_sm, &old_count);
1505 if (r)
1506 return r;
1507
1508 if (new_count == old_count)
1509 return 0;
1510
1511 if (new_count < old_count) {
1512 DMERR("cannot reduce size of data device");
1513 return -EINVAL;
1514 }
1515
1516 r = dm_sm_extend(pmd->data_sm, new_count - old_count);
1517 if (!r)
1518 pmd->need_commit = 1;
1519
1520 return r;
1521}
1522
1523int dm_pool_resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1524{
1525 int r;
1526
1527 down_write(&pmd->root_lock);
1528 r = __resize_data_dev(pmd, new_count);
1529 up_write(&pmd->root_lock);
1530
1531 return r;
1532}
kernel source for telekom puls (alcatel/mediatek)