xref: /openbmc/linux/drivers/md/dm-thin-metadata.c (revision 2792d42f)
1 /*
2  * Copyright (C) 2011-2012 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 /*
84  *  3 for btree insert +
85  *  2 for btree lookup used within space map
86  */
87 #define THIN_MAX_CONCURRENT_LOCKS 5
88 
89 /* This should be plenty */
90 #define SPACE_MAP_ROOT_SIZE 128
91 
92 /*
93  * Little endian on-disk superblock and device details.
94  */
95 struct 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 
135 struct disk_device_details {
136 	__le64 mapped_blocks;
137 	__le64 transaction_id;		/* When created. */
138 	__le32 creation_time;
139 	__le32 snapshotted_time;
140 } __packed;
141 
142 struct 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;
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 	bool read_only:1;
188 
189 	/*
190 	 * Set if a transaction has to be aborted but the attempt to roll back
191 	 * to the previous (good) transaction failed.  The only pool metadata
192 	 * operation possible in this state is the closing of the device.
193 	 */
194 	bool fail_io:1;
195 };
196 
197 struct dm_thin_device {
198 	struct list_head list;
199 	struct dm_pool_metadata *pmd;
200 	dm_thin_id id;
201 
202 	int open_count;
203 	bool changed:1;
204 	bool aborted_with_changes:1;
205 	uint64_t mapped_blocks;
206 	uint64_t transaction_id;
207 	uint32_t creation_time;
208 	uint32_t snapshotted_time;
209 };
210 
211 /*----------------------------------------------------------------
212  * superblock validator
213  *--------------------------------------------------------------*/
214 
215 #define SUPERBLOCK_CSUM_XOR 160774
216 
217 static void sb_prepare_for_write(struct dm_block_validator *v,
218 				 struct dm_block *b,
219 				 size_t block_size)
220 {
221 	struct thin_disk_superblock *disk_super = dm_block_data(b);
222 
223 	disk_super->blocknr = cpu_to_le64(dm_block_location(b));
224 	disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
225 						      block_size - sizeof(__le32),
226 						      SUPERBLOCK_CSUM_XOR));
227 }
228 
229 static int sb_check(struct dm_block_validator *v,
230 		    struct dm_block *b,
231 		    size_t block_size)
232 {
233 	struct thin_disk_superblock *disk_super = dm_block_data(b);
234 	__le32 csum_le;
235 
236 	if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
237 		DMERR("sb_check failed: blocknr %llu: "
238 		      "wanted %llu", le64_to_cpu(disk_super->blocknr),
239 		      (unsigned long long)dm_block_location(b));
240 		return -ENOTBLK;
241 	}
242 
243 	if (le64_to_cpu(disk_super->magic) != THIN_SUPERBLOCK_MAGIC) {
244 		DMERR("sb_check failed: magic %llu: "
245 		      "wanted %llu", le64_to_cpu(disk_super->magic),
246 		      (unsigned long long)THIN_SUPERBLOCK_MAGIC);
247 		return -EILSEQ;
248 	}
249 
250 	csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
251 					     block_size - sizeof(__le32),
252 					     SUPERBLOCK_CSUM_XOR));
253 	if (csum_le != disk_super->csum) {
254 		DMERR("sb_check failed: csum %u: wanted %u",
255 		      le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
256 		return -EILSEQ;
257 	}
258 
259 	return 0;
260 }
261 
262 static struct dm_block_validator sb_validator = {
263 	.name = "superblock",
264 	.prepare_for_write = sb_prepare_for_write,
265 	.check = sb_check
266 };
267 
268 /*----------------------------------------------------------------
269  * Methods for the btree value types
270  *--------------------------------------------------------------*/
271 
272 static uint64_t pack_block_time(dm_block_t b, uint32_t t)
273 {
274 	return (b << 24) | t;
275 }
276 
277 static void unpack_block_time(uint64_t v, dm_block_t *b, uint32_t *t)
278 {
279 	*b = v >> 24;
280 	*t = v & ((1 << 24) - 1);
281 }
282 
283 static void data_block_inc(void *context, const void *value_le)
284 {
285 	struct dm_space_map *sm = context;
286 	__le64 v_le;
287 	uint64_t b;
288 	uint32_t t;
289 
290 	memcpy(&v_le, value_le, sizeof(v_le));
291 	unpack_block_time(le64_to_cpu(v_le), &b, &t);
292 	dm_sm_inc_block(sm, b);
293 }
294 
295 static void data_block_dec(void *context, const void *value_le)
296 {
297 	struct dm_space_map *sm = context;
298 	__le64 v_le;
299 	uint64_t b;
300 	uint32_t t;
301 
302 	memcpy(&v_le, value_le, sizeof(v_le));
303 	unpack_block_time(le64_to_cpu(v_le), &b, &t);
304 	dm_sm_dec_block(sm, b);
305 }
306 
307 static int data_block_equal(void *context, const void *value1_le, const void *value2_le)
308 {
309 	__le64 v1_le, v2_le;
310 	uint64_t b1, b2;
311 	uint32_t t;
312 
313 	memcpy(&v1_le, value1_le, sizeof(v1_le));
314 	memcpy(&v2_le, value2_le, sizeof(v2_le));
315 	unpack_block_time(le64_to_cpu(v1_le), &b1, &t);
316 	unpack_block_time(le64_to_cpu(v2_le), &b2, &t);
317 
318 	return b1 == b2;
319 }
320 
321 static void subtree_inc(void *context, const void *value)
322 {
323 	struct dm_btree_info *info = context;
324 	__le64 root_le;
325 	uint64_t root;
326 
327 	memcpy(&root_le, value, sizeof(root_le));
328 	root = le64_to_cpu(root_le);
329 	dm_tm_inc(info->tm, root);
330 }
331 
332 static void subtree_dec(void *context, const void *value)
333 {
334 	struct dm_btree_info *info = context;
335 	__le64 root_le;
336 	uint64_t root;
337 
338 	memcpy(&root_le, value, sizeof(root_le));
339 	root = le64_to_cpu(root_le);
340 	if (dm_btree_del(info, root))
341 		DMERR("btree delete failed\n");
342 }
343 
344 static int subtree_equal(void *context, const void *value1_le, const void *value2_le)
345 {
346 	__le64 v1_le, v2_le;
347 	memcpy(&v1_le, value1_le, sizeof(v1_le));
348 	memcpy(&v2_le, value2_le, sizeof(v2_le));
349 
350 	return v1_le == v2_le;
351 }
352 
353 /*----------------------------------------------------------------*/
354 
355 static int superblock_lock_zero(struct dm_pool_metadata *pmd,
356 				struct dm_block **sblock)
357 {
358 	return dm_bm_write_lock_zero(pmd->bm, THIN_SUPERBLOCK_LOCATION,
359 				     &sb_validator, sblock);
360 }
361 
362 static int superblock_lock(struct dm_pool_metadata *pmd,
363 			   struct dm_block **sblock)
364 {
365 	return dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
366 				&sb_validator, sblock);
367 }
368 
369 static int __superblock_all_zeroes(struct dm_block_manager *bm, int *result)
370 {
371 	int r;
372 	unsigned i;
373 	struct dm_block *b;
374 	__le64 *data_le, zero = cpu_to_le64(0);
375 	unsigned block_size = dm_bm_block_size(bm) / sizeof(__le64);
376 
377 	/*
378 	 * We can't use a validator here - it may be all zeroes.
379 	 */
380 	r = dm_bm_read_lock(bm, THIN_SUPERBLOCK_LOCATION, NULL, &b);
381 	if (r)
382 		return r;
383 
384 	data_le = dm_block_data(b);
385 	*result = 1;
386 	for (i = 0; i < block_size; i++) {
387 		if (data_le[i] != zero) {
388 			*result = 0;
389 			break;
390 		}
391 	}
392 
393 	return dm_bm_unlock(b);
394 }
395 
396 static void __setup_btree_details(struct dm_pool_metadata *pmd)
397 {
398 	pmd->info.tm = pmd->tm;
399 	pmd->info.levels = 2;
400 	pmd->info.value_type.context = pmd->data_sm;
401 	pmd->info.value_type.size = sizeof(__le64);
402 	pmd->info.value_type.inc = data_block_inc;
403 	pmd->info.value_type.dec = data_block_dec;
404 	pmd->info.value_type.equal = data_block_equal;
405 
406 	memcpy(&pmd->nb_info, &pmd->info, sizeof(pmd->nb_info));
407 	pmd->nb_info.tm = pmd->nb_tm;
408 
409 	pmd->tl_info.tm = pmd->tm;
410 	pmd->tl_info.levels = 1;
411 	pmd->tl_info.value_type.context = &pmd->bl_info;
412 	pmd->tl_info.value_type.size = sizeof(__le64);
413 	pmd->tl_info.value_type.inc = subtree_inc;
414 	pmd->tl_info.value_type.dec = subtree_dec;
415 	pmd->tl_info.value_type.equal = subtree_equal;
416 
417 	pmd->bl_info.tm = pmd->tm;
418 	pmd->bl_info.levels = 1;
419 	pmd->bl_info.value_type.context = pmd->data_sm;
420 	pmd->bl_info.value_type.size = sizeof(__le64);
421 	pmd->bl_info.value_type.inc = data_block_inc;
422 	pmd->bl_info.value_type.dec = data_block_dec;
423 	pmd->bl_info.value_type.equal = data_block_equal;
424 
425 	pmd->details_info.tm = pmd->tm;
426 	pmd->details_info.levels = 1;
427 	pmd->details_info.value_type.context = NULL;
428 	pmd->details_info.value_type.size = sizeof(struct disk_device_details);
429 	pmd->details_info.value_type.inc = NULL;
430 	pmd->details_info.value_type.dec = NULL;
431 	pmd->details_info.value_type.equal = NULL;
432 }
433 
434 static int __write_initial_superblock(struct dm_pool_metadata *pmd)
435 {
436 	int r;
437 	struct dm_block *sblock;
438 	size_t metadata_len, data_len;
439 	struct thin_disk_superblock *disk_super;
440 	sector_t bdev_size = i_size_read(pmd->bdev->bd_inode) >> SECTOR_SHIFT;
441 
442 	if (bdev_size > THIN_METADATA_MAX_SECTORS)
443 		bdev_size = THIN_METADATA_MAX_SECTORS;
444 
445 	r = dm_sm_root_size(pmd->metadata_sm, &metadata_len);
446 	if (r < 0)
447 		return r;
448 
449 	r = dm_sm_root_size(pmd->data_sm, &data_len);
450 	if (r < 0)
451 		return r;
452 
453 	r = dm_sm_commit(pmd->data_sm);
454 	if (r < 0)
455 		return r;
456 
457 	r = dm_tm_pre_commit(pmd->tm);
458 	if (r < 0)
459 		return r;
460 
461 	r = superblock_lock_zero(pmd, &sblock);
462 	if (r)
463 		return r;
464 
465 	disk_super = dm_block_data(sblock);
466 	disk_super->flags = 0;
467 	memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
468 	disk_super->magic = cpu_to_le64(THIN_SUPERBLOCK_MAGIC);
469 	disk_super->version = cpu_to_le32(THIN_VERSION);
470 	disk_super->time = 0;
471 	disk_super->trans_id = 0;
472 	disk_super->held_root = 0;
473 
474 	r = dm_sm_copy_root(pmd->metadata_sm, &disk_super->metadata_space_map_root,
475 			    metadata_len);
476 	if (r < 0)
477 		goto bad_locked;
478 
479 	r = dm_sm_copy_root(pmd->data_sm, &disk_super->data_space_map_root,
480 			    data_len);
481 	if (r < 0)
482 		goto bad_locked;
483 
484 	disk_super->data_mapping_root = cpu_to_le64(pmd->root);
485 	disk_super->device_details_root = cpu_to_le64(pmd->details_root);
486 	disk_super->metadata_block_size = cpu_to_le32(THIN_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
487 	disk_super->metadata_nr_blocks = cpu_to_le64(bdev_size >> SECTOR_TO_BLOCK_SHIFT);
488 	disk_super->data_block_size = cpu_to_le32(pmd->data_block_size);
489 
490 	return dm_tm_commit(pmd->tm, sblock);
491 
492 bad_locked:
493 	dm_bm_unlock(sblock);
494 	return r;
495 }
496 
497 static int __format_metadata(struct dm_pool_metadata *pmd)
498 {
499 	int r;
500 
501 	r = dm_tm_create_with_sm(pmd->bm, THIN_SUPERBLOCK_LOCATION,
502 				 &pmd->tm, &pmd->metadata_sm);
503 	if (r < 0) {
504 		DMERR("tm_create_with_sm failed");
505 		return r;
506 	}
507 
508 	pmd->data_sm = dm_sm_disk_create(pmd->tm, 0);
509 	if (IS_ERR(pmd->data_sm)) {
510 		DMERR("sm_disk_create failed");
511 		r = PTR_ERR(pmd->data_sm);
512 		goto bad_cleanup_tm;
513 	}
514 
515 	pmd->nb_tm = dm_tm_create_non_blocking_clone(pmd->tm);
516 	if (!pmd->nb_tm) {
517 		DMERR("could not create non-blocking clone tm");
518 		r = -ENOMEM;
519 		goto bad_cleanup_data_sm;
520 	}
521 
522 	__setup_btree_details(pmd);
523 
524 	r = dm_btree_empty(&pmd->info, &pmd->root);
525 	if (r < 0)
526 		goto bad_cleanup_nb_tm;
527 
528 	r = dm_btree_empty(&pmd->details_info, &pmd->details_root);
529 	if (r < 0) {
530 		DMERR("couldn't create devices root");
531 		goto bad_cleanup_nb_tm;
532 	}
533 
534 	r = __write_initial_superblock(pmd);
535 	if (r)
536 		goto bad_cleanup_nb_tm;
537 
538 	return 0;
539 
540 bad_cleanup_nb_tm:
541 	dm_tm_destroy(pmd->nb_tm);
542 bad_cleanup_data_sm:
543 	dm_sm_destroy(pmd->data_sm);
544 bad_cleanup_tm:
545 	dm_tm_destroy(pmd->tm);
546 	dm_sm_destroy(pmd->metadata_sm);
547 
548 	return r;
549 }
550 
551 static int __check_incompat_features(struct thin_disk_superblock *disk_super,
552 				     struct dm_pool_metadata *pmd)
553 {
554 	uint32_t features;
555 
556 	features = le32_to_cpu(disk_super->incompat_flags) & ~THIN_FEATURE_INCOMPAT_SUPP;
557 	if (features) {
558 		DMERR("could not access metadata due to unsupported optional features (%lx).",
559 		      (unsigned long)features);
560 		return -EINVAL;
561 	}
562 
563 	/*
564 	 * Check for read-only metadata to skip the following RDWR checks.
565 	 */
566 	if (get_disk_ro(pmd->bdev->bd_disk))
567 		return 0;
568 
569 	features = le32_to_cpu(disk_super->compat_ro_flags) & ~THIN_FEATURE_COMPAT_RO_SUPP;
570 	if (features) {
571 		DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
572 		      (unsigned long)features);
573 		return -EINVAL;
574 	}
575 
576 	return 0;
577 }
578 
579 static int __open_metadata(struct dm_pool_metadata *pmd)
580 {
581 	int r;
582 	struct dm_block *sblock;
583 	struct thin_disk_superblock *disk_super;
584 
585 	r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
586 			    &sb_validator, &sblock);
587 	if (r < 0) {
588 		DMERR("couldn't read superblock");
589 		return r;
590 	}
591 
592 	disk_super = dm_block_data(sblock);
593 
594 	r = __check_incompat_features(disk_super, pmd);
595 	if (r < 0)
596 		goto bad_unlock_sblock;
597 
598 	r = dm_tm_open_with_sm(pmd->bm, THIN_SUPERBLOCK_LOCATION,
599 			       disk_super->metadata_space_map_root,
600 			       sizeof(disk_super->metadata_space_map_root),
601 			       &pmd->tm, &pmd->metadata_sm);
602 	if (r < 0) {
603 		DMERR("tm_open_with_sm failed");
604 		goto bad_unlock_sblock;
605 	}
606 
607 	pmd->data_sm = dm_sm_disk_open(pmd->tm, disk_super->data_space_map_root,
608 				       sizeof(disk_super->data_space_map_root));
609 	if (IS_ERR(pmd->data_sm)) {
610 		DMERR("sm_disk_open failed");
611 		r = PTR_ERR(pmd->data_sm);
612 		goto bad_cleanup_tm;
613 	}
614 
615 	pmd->nb_tm = dm_tm_create_non_blocking_clone(pmd->tm);
616 	if (!pmd->nb_tm) {
617 		DMERR("could not create non-blocking clone tm");
618 		r = -ENOMEM;
619 		goto bad_cleanup_data_sm;
620 	}
621 
622 	__setup_btree_details(pmd);
623 	return dm_bm_unlock(sblock);
624 
625 bad_cleanup_data_sm:
626 	dm_sm_destroy(pmd->data_sm);
627 bad_cleanup_tm:
628 	dm_tm_destroy(pmd->tm);
629 	dm_sm_destroy(pmd->metadata_sm);
630 bad_unlock_sblock:
631 	dm_bm_unlock(sblock);
632 
633 	return r;
634 }
635 
636 static int __open_or_format_metadata(struct dm_pool_metadata *pmd, bool format_device)
637 {
638 	int r, unformatted;
639 
640 	r = __superblock_all_zeroes(pmd->bm, &unformatted);
641 	if (r)
642 		return r;
643 
644 	if (unformatted)
645 		return format_device ? __format_metadata(pmd) : -EPERM;
646 
647 	return __open_metadata(pmd);
648 }
649 
650 static int __create_persistent_data_objects(struct dm_pool_metadata *pmd, bool format_device)
651 {
652 	int r;
653 
654 	pmd->bm = dm_block_manager_create(pmd->bdev, THIN_METADATA_BLOCK_SIZE,
655 					  THIN_METADATA_CACHE_SIZE,
656 					  THIN_MAX_CONCURRENT_LOCKS);
657 	if (IS_ERR(pmd->bm)) {
658 		DMERR("could not create block manager");
659 		return PTR_ERR(pmd->bm);
660 	}
661 
662 	r = __open_or_format_metadata(pmd, format_device);
663 	if (r)
664 		dm_block_manager_destroy(pmd->bm);
665 
666 	return r;
667 }
668 
669 static void __destroy_persistent_data_objects(struct dm_pool_metadata *pmd)
670 {
671 	dm_sm_destroy(pmd->data_sm);
672 	dm_sm_destroy(pmd->metadata_sm);
673 	dm_tm_destroy(pmd->nb_tm);
674 	dm_tm_destroy(pmd->tm);
675 	dm_block_manager_destroy(pmd->bm);
676 }
677 
678 static int __begin_transaction(struct dm_pool_metadata *pmd)
679 {
680 	int r;
681 	struct thin_disk_superblock *disk_super;
682 	struct dm_block *sblock;
683 
684 	/*
685 	 * We re-read the superblock every time.  Shouldn't need to do this
686 	 * really.
687 	 */
688 	r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
689 			    &sb_validator, &sblock);
690 	if (r)
691 		return r;
692 
693 	disk_super = dm_block_data(sblock);
694 	pmd->time = le32_to_cpu(disk_super->time);
695 	pmd->root = le64_to_cpu(disk_super->data_mapping_root);
696 	pmd->details_root = le64_to_cpu(disk_super->device_details_root);
697 	pmd->trans_id = le64_to_cpu(disk_super->trans_id);
698 	pmd->flags = le32_to_cpu(disk_super->flags);
699 	pmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
700 
701 	dm_bm_unlock(sblock);
702 	return 0;
703 }
704 
705 static int __write_changed_details(struct dm_pool_metadata *pmd)
706 {
707 	int r;
708 	struct dm_thin_device *td, *tmp;
709 	struct disk_device_details details;
710 	uint64_t key;
711 
712 	list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
713 		if (!td->changed)
714 			continue;
715 
716 		key = td->id;
717 
718 		details.mapped_blocks = cpu_to_le64(td->mapped_blocks);
719 		details.transaction_id = cpu_to_le64(td->transaction_id);
720 		details.creation_time = cpu_to_le32(td->creation_time);
721 		details.snapshotted_time = cpu_to_le32(td->snapshotted_time);
722 		__dm_bless_for_disk(&details);
723 
724 		r = dm_btree_insert(&pmd->details_info, pmd->details_root,
725 				    &key, &details, &pmd->details_root);
726 		if (r)
727 			return r;
728 
729 		if (td->open_count)
730 			td->changed = 0;
731 		else {
732 			list_del(&td->list);
733 			kfree(td);
734 		}
735 	}
736 
737 	return 0;
738 }
739 
740 static int __commit_transaction(struct dm_pool_metadata *pmd)
741 {
742 	int r;
743 	size_t metadata_len, data_len;
744 	struct thin_disk_superblock *disk_super;
745 	struct dm_block *sblock;
746 
747 	/*
748 	 * We need to know if the thin_disk_superblock exceeds a 512-byte sector.
749 	 */
750 	BUILD_BUG_ON(sizeof(struct thin_disk_superblock) > 512);
751 
752 	r = __write_changed_details(pmd);
753 	if (r < 0)
754 		return r;
755 
756 	r = dm_sm_commit(pmd->data_sm);
757 	if (r < 0)
758 		return r;
759 
760 	r = dm_tm_pre_commit(pmd->tm);
761 	if (r < 0)
762 		return r;
763 
764 	r = dm_sm_root_size(pmd->metadata_sm, &metadata_len);
765 	if (r < 0)
766 		return r;
767 
768 	r = dm_sm_root_size(pmd->data_sm, &data_len);
769 	if (r < 0)
770 		return r;
771 
772 	r = superblock_lock(pmd, &sblock);
773 	if (r)
774 		return r;
775 
776 	disk_super = dm_block_data(sblock);
777 	disk_super->time = cpu_to_le32(pmd->time);
778 	disk_super->data_mapping_root = cpu_to_le64(pmd->root);
779 	disk_super->device_details_root = cpu_to_le64(pmd->details_root);
780 	disk_super->trans_id = cpu_to_le64(pmd->trans_id);
781 	disk_super->flags = cpu_to_le32(pmd->flags);
782 
783 	r = dm_sm_copy_root(pmd->metadata_sm, &disk_super->metadata_space_map_root,
784 			    metadata_len);
785 	if (r < 0)
786 		goto out_locked;
787 
788 	r = dm_sm_copy_root(pmd->data_sm, &disk_super->data_space_map_root,
789 			    data_len);
790 	if (r < 0)
791 		goto out_locked;
792 
793 	return dm_tm_commit(pmd->tm, sblock);
794 
795 out_locked:
796 	dm_bm_unlock(sblock);
797 	return r;
798 }
799 
800 struct dm_pool_metadata *dm_pool_metadata_open(struct block_device *bdev,
801 					       sector_t data_block_size,
802 					       bool format_device)
803 {
804 	int r;
805 	struct dm_pool_metadata *pmd;
806 
807 	pmd = kmalloc(sizeof(*pmd), GFP_KERNEL);
808 	if (!pmd) {
809 		DMERR("could not allocate metadata struct");
810 		return ERR_PTR(-ENOMEM);
811 	}
812 
813 	init_rwsem(&pmd->root_lock);
814 	pmd->time = 0;
815 	INIT_LIST_HEAD(&pmd->thin_devices);
816 	pmd->read_only = false;
817 	pmd->fail_io = false;
818 	pmd->bdev = bdev;
819 	pmd->data_block_size = data_block_size;
820 
821 	r = __create_persistent_data_objects(pmd, format_device);
822 	if (r) {
823 		kfree(pmd);
824 		return ERR_PTR(r);
825 	}
826 
827 	r = __begin_transaction(pmd);
828 	if (r < 0) {
829 		if (dm_pool_metadata_close(pmd) < 0)
830 			DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
831 		return ERR_PTR(r);
832 	}
833 
834 	return pmd;
835 }
836 
837 int dm_pool_metadata_close(struct dm_pool_metadata *pmd)
838 {
839 	int r;
840 	unsigned open_devices = 0;
841 	struct dm_thin_device *td, *tmp;
842 
843 	down_read(&pmd->root_lock);
844 	list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
845 		if (td->open_count)
846 			open_devices++;
847 		else {
848 			list_del(&td->list);
849 			kfree(td);
850 		}
851 	}
852 	up_read(&pmd->root_lock);
853 
854 	if (open_devices) {
855 		DMERR("attempt to close pmd when %u device(s) are still open",
856 		       open_devices);
857 		return -EBUSY;
858 	}
859 
860 	if (!pmd->read_only && !pmd->fail_io) {
861 		r = __commit_transaction(pmd);
862 		if (r < 0)
863 			DMWARN("%s: __commit_transaction() failed, error = %d",
864 			       __func__, r);
865 	}
866 
867 	if (!pmd->fail_io)
868 		__destroy_persistent_data_objects(pmd);
869 
870 	kfree(pmd);
871 	return 0;
872 }
873 
874 /*
875  * __open_device: Returns @td corresponding to device with id @dev,
876  * creating it if @create is set and incrementing @td->open_count.
877  * On failure, @td is undefined.
878  */
879 static int __open_device(struct dm_pool_metadata *pmd,
880 			 dm_thin_id dev, int create,
881 			 struct dm_thin_device **td)
882 {
883 	int r, changed = 0;
884 	struct dm_thin_device *td2;
885 	uint64_t key = dev;
886 	struct disk_device_details details_le;
887 
888 	/*
889 	 * If the device is already open, return it.
890 	 */
891 	list_for_each_entry(td2, &pmd->thin_devices, list)
892 		if (td2->id == dev) {
893 			/*
894 			 * May not create an already-open device.
895 			 */
896 			if (create)
897 				return -EEXIST;
898 
899 			td2->open_count++;
900 			*td = td2;
901 			return 0;
902 		}
903 
904 	/*
905 	 * Check the device exists.
906 	 */
907 	r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
908 			    &key, &details_le);
909 	if (r) {
910 		if (r != -ENODATA || !create)
911 			return r;
912 
913 		/*
914 		 * Create new device.
915 		 */
916 		changed = 1;
917 		details_le.mapped_blocks = 0;
918 		details_le.transaction_id = cpu_to_le64(pmd->trans_id);
919 		details_le.creation_time = cpu_to_le32(pmd->time);
920 		details_le.snapshotted_time = cpu_to_le32(pmd->time);
921 	}
922 
923 	*td = kmalloc(sizeof(**td), GFP_NOIO);
924 	if (!*td)
925 		return -ENOMEM;
926 
927 	(*td)->pmd = pmd;
928 	(*td)->id = dev;
929 	(*td)->open_count = 1;
930 	(*td)->changed = changed;
931 	(*td)->aborted_with_changes = false;
932 	(*td)->mapped_blocks = le64_to_cpu(details_le.mapped_blocks);
933 	(*td)->transaction_id = le64_to_cpu(details_le.transaction_id);
934 	(*td)->creation_time = le32_to_cpu(details_le.creation_time);
935 	(*td)->snapshotted_time = le32_to_cpu(details_le.snapshotted_time);
936 
937 	list_add(&(*td)->list, &pmd->thin_devices);
938 
939 	return 0;
940 }
941 
942 static void __close_device(struct dm_thin_device *td)
943 {
944 	--td->open_count;
945 }
946 
947 static int __create_thin(struct dm_pool_metadata *pmd,
948 			 dm_thin_id dev)
949 {
950 	int r;
951 	dm_block_t dev_root;
952 	uint64_t key = dev;
953 	struct disk_device_details details_le;
954 	struct dm_thin_device *td;
955 	__le64 value;
956 
957 	r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
958 			    &key, &details_le);
959 	if (!r)
960 		return -EEXIST;
961 
962 	/*
963 	 * Create an empty btree for the mappings.
964 	 */
965 	r = dm_btree_empty(&pmd->bl_info, &dev_root);
966 	if (r)
967 		return r;
968 
969 	/*
970 	 * Insert it into the main mapping tree.
971 	 */
972 	value = cpu_to_le64(dev_root);
973 	__dm_bless_for_disk(&value);
974 	r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
975 	if (r) {
976 		dm_btree_del(&pmd->bl_info, dev_root);
977 		return r;
978 	}
979 
980 	r = __open_device(pmd, dev, 1, &td);
981 	if (r) {
982 		dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
983 		dm_btree_del(&pmd->bl_info, dev_root);
984 		return r;
985 	}
986 	__close_device(td);
987 
988 	return r;
989 }
990 
991 int dm_pool_create_thin(struct dm_pool_metadata *pmd, dm_thin_id dev)
992 {
993 	int r = -EINVAL;
994 
995 	down_write(&pmd->root_lock);
996 	if (!pmd->fail_io)
997 		r = __create_thin(pmd, dev);
998 	up_write(&pmd->root_lock);
999 
1000 	return r;
1001 }
1002 
1003 static int __set_snapshot_details(struct dm_pool_metadata *pmd,
1004 				  struct dm_thin_device *snap,
1005 				  dm_thin_id origin, uint32_t time)
1006 {
1007 	int r;
1008 	struct dm_thin_device *td;
1009 
1010 	r = __open_device(pmd, origin, 0, &td);
1011 	if (r)
1012 		return r;
1013 
1014 	td->changed = 1;
1015 	td->snapshotted_time = time;
1016 
1017 	snap->mapped_blocks = td->mapped_blocks;
1018 	snap->snapshotted_time = time;
1019 	__close_device(td);
1020 
1021 	return 0;
1022 }
1023 
1024 static int __create_snap(struct dm_pool_metadata *pmd,
1025 			 dm_thin_id dev, dm_thin_id origin)
1026 {
1027 	int r;
1028 	dm_block_t origin_root;
1029 	uint64_t key = origin, dev_key = dev;
1030 	struct dm_thin_device *td;
1031 	struct disk_device_details details_le;
1032 	__le64 value;
1033 
1034 	/* check this device is unused */
1035 	r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
1036 			    &dev_key, &details_le);
1037 	if (!r)
1038 		return -EEXIST;
1039 
1040 	/* find the mapping tree for the origin */
1041 	r = dm_btree_lookup(&pmd->tl_info, pmd->root, &key, &value);
1042 	if (r)
1043 		return r;
1044 	origin_root = le64_to_cpu(value);
1045 
1046 	/* clone the origin, an inc will do */
1047 	dm_tm_inc(pmd->tm, origin_root);
1048 
1049 	/* insert into the main mapping tree */
1050 	value = cpu_to_le64(origin_root);
1051 	__dm_bless_for_disk(&value);
1052 	key = dev;
1053 	r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
1054 	if (r) {
1055 		dm_tm_dec(pmd->tm, origin_root);
1056 		return r;
1057 	}
1058 
1059 	pmd->time++;
1060 
1061 	r = __open_device(pmd, dev, 1, &td);
1062 	if (r)
1063 		goto bad;
1064 
1065 	r = __set_snapshot_details(pmd, td, origin, pmd->time);
1066 	__close_device(td);
1067 
1068 	if (r)
1069 		goto bad;
1070 
1071 	return 0;
1072 
1073 bad:
1074 	dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1075 	dm_btree_remove(&pmd->details_info, pmd->details_root,
1076 			&key, &pmd->details_root);
1077 	return r;
1078 }
1079 
1080 int dm_pool_create_snap(struct dm_pool_metadata *pmd,
1081 				 dm_thin_id dev,
1082 				 dm_thin_id origin)
1083 {
1084 	int r = -EINVAL;
1085 
1086 	down_write(&pmd->root_lock);
1087 	if (!pmd->fail_io)
1088 		r = __create_snap(pmd, dev, origin);
1089 	up_write(&pmd->root_lock);
1090 
1091 	return r;
1092 }
1093 
1094 static int __delete_device(struct dm_pool_metadata *pmd, dm_thin_id dev)
1095 {
1096 	int r;
1097 	uint64_t key = dev;
1098 	struct dm_thin_device *td;
1099 
1100 	/* TODO: failure should mark the transaction invalid */
1101 	r = __open_device(pmd, dev, 0, &td);
1102 	if (r)
1103 		return r;
1104 
1105 	if (td->open_count > 1) {
1106 		__close_device(td);
1107 		return -EBUSY;
1108 	}
1109 
1110 	list_del(&td->list);
1111 	kfree(td);
1112 	r = dm_btree_remove(&pmd->details_info, pmd->details_root,
1113 			    &key, &pmd->details_root);
1114 	if (r)
1115 		return r;
1116 
1117 	r = dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1118 	if (r)
1119 		return r;
1120 
1121 	return 0;
1122 }
1123 
1124 int dm_pool_delete_thin_device(struct dm_pool_metadata *pmd,
1125 			       dm_thin_id dev)
1126 {
1127 	int r = -EINVAL;
1128 
1129 	down_write(&pmd->root_lock);
1130 	if (!pmd->fail_io)
1131 		r = __delete_device(pmd, dev);
1132 	up_write(&pmd->root_lock);
1133 
1134 	return r;
1135 }
1136 
1137 int dm_pool_set_metadata_transaction_id(struct dm_pool_metadata *pmd,
1138 					uint64_t current_id,
1139 					uint64_t new_id)
1140 {
1141 	int r = -EINVAL;
1142 
1143 	down_write(&pmd->root_lock);
1144 
1145 	if (pmd->fail_io)
1146 		goto out;
1147 
1148 	if (pmd->trans_id != current_id) {
1149 		DMERR("mismatched transaction id");
1150 		goto out;
1151 	}
1152 
1153 	pmd->trans_id = new_id;
1154 	r = 0;
1155 
1156 out:
1157 	up_write(&pmd->root_lock);
1158 
1159 	return r;
1160 }
1161 
1162 int dm_pool_get_metadata_transaction_id(struct dm_pool_metadata *pmd,
1163 					uint64_t *result)
1164 {
1165 	int r = -EINVAL;
1166 
1167 	down_read(&pmd->root_lock);
1168 	if (!pmd->fail_io) {
1169 		*result = pmd->trans_id;
1170 		r = 0;
1171 	}
1172 	up_read(&pmd->root_lock);
1173 
1174 	return r;
1175 }
1176 
1177 static int __reserve_metadata_snap(struct dm_pool_metadata *pmd)
1178 {
1179 	int r, inc;
1180 	struct thin_disk_superblock *disk_super;
1181 	struct dm_block *copy, *sblock;
1182 	dm_block_t held_root;
1183 
1184 	/*
1185 	 * Copy the superblock.
1186 	 */
1187 	dm_sm_inc_block(pmd->metadata_sm, THIN_SUPERBLOCK_LOCATION);
1188 	r = dm_tm_shadow_block(pmd->tm, THIN_SUPERBLOCK_LOCATION,
1189 			       &sb_validator, &copy, &inc);
1190 	if (r)
1191 		return r;
1192 
1193 	BUG_ON(!inc);
1194 
1195 	held_root = dm_block_location(copy);
1196 	disk_super = dm_block_data(copy);
1197 
1198 	if (le64_to_cpu(disk_super->held_root)) {
1199 		DMWARN("Pool metadata snapshot already exists: release this before taking another.");
1200 
1201 		dm_tm_dec(pmd->tm, held_root);
1202 		dm_tm_unlock(pmd->tm, copy);
1203 		return -EBUSY;
1204 	}
1205 
1206 	/*
1207 	 * Wipe the spacemap since we're not publishing this.
1208 	 */
1209 	memset(&disk_super->data_space_map_root, 0,
1210 	       sizeof(disk_super->data_space_map_root));
1211 	memset(&disk_super->metadata_space_map_root, 0,
1212 	       sizeof(disk_super->metadata_space_map_root));
1213 
1214 	/*
1215 	 * Increment the data structures that need to be preserved.
1216 	 */
1217 	dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->data_mapping_root));
1218 	dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->device_details_root));
1219 	dm_tm_unlock(pmd->tm, copy);
1220 
1221 	/*
1222 	 * Write the held root into the superblock.
1223 	 */
1224 	r = superblock_lock(pmd, &sblock);
1225 	if (r) {
1226 		dm_tm_dec(pmd->tm, held_root);
1227 		return r;
1228 	}
1229 
1230 	disk_super = dm_block_data(sblock);
1231 	disk_super->held_root = cpu_to_le64(held_root);
1232 	dm_bm_unlock(sblock);
1233 	return 0;
1234 }
1235 
1236 int dm_pool_reserve_metadata_snap(struct dm_pool_metadata *pmd)
1237 {
1238 	int r = -EINVAL;
1239 
1240 	down_write(&pmd->root_lock);
1241 	if (!pmd->fail_io)
1242 		r = __reserve_metadata_snap(pmd);
1243 	up_write(&pmd->root_lock);
1244 
1245 	return r;
1246 }
1247 
1248 static int __release_metadata_snap(struct dm_pool_metadata *pmd)
1249 {
1250 	int r;
1251 	struct thin_disk_superblock *disk_super;
1252 	struct dm_block *sblock, *copy;
1253 	dm_block_t held_root;
1254 
1255 	r = superblock_lock(pmd, &sblock);
1256 	if (r)
1257 		return r;
1258 
1259 	disk_super = dm_block_data(sblock);
1260 	held_root = le64_to_cpu(disk_super->held_root);
1261 	disk_super->held_root = cpu_to_le64(0);
1262 
1263 	dm_bm_unlock(sblock);
1264 
1265 	if (!held_root) {
1266 		DMWARN("No pool metadata snapshot found: nothing to release.");
1267 		return -EINVAL;
1268 	}
1269 
1270 	r = dm_tm_read_lock(pmd->tm, held_root, &sb_validator, &copy);
1271 	if (r)
1272 		return r;
1273 
1274 	disk_super = dm_block_data(copy);
1275 	dm_sm_dec_block(pmd->metadata_sm, le64_to_cpu(disk_super->data_mapping_root));
1276 	dm_sm_dec_block(pmd->metadata_sm, le64_to_cpu(disk_super->device_details_root));
1277 	dm_sm_dec_block(pmd->metadata_sm, held_root);
1278 
1279 	return dm_tm_unlock(pmd->tm, copy);
1280 }
1281 
1282 int dm_pool_release_metadata_snap(struct dm_pool_metadata *pmd)
1283 {
1284 	int r = -EINVAL;
1285 
1286 	down_write(&pmd->root_lock);
1287 	if (!pmd->fail_io)
1288 		r = __release_metadata_snap(pmd);
1289 	up_write(&pmd->root_lock);
1290 
1291 	return r;
1292 }
1293 
1294 static int __get_metadata_snap(struct dm_pool_metadata *pmd,
1295 			       dm_block_t *result)
1296 {
1297 	int r;
1298 	struct thin_disk_superblock *disk_super;
1299 	struct dm_block *sblock;
1300 
1301 	r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
1302 			    &sb_validator, &sblock);
1303 	if (r)
1304 		return r;
1305 
1306 	disk_super = dm_block_data(sblock);
1307 	*result = le64_to_cpu(disk_super->held_root);
1308 
1309 	return dm_bm_unlock(sblock);
1310 }
1311 
1312 int dm_pool_get_metadata_snap(struct dm_pool_metadata *pmd,
1313 			      dm_block_t *result)
1314 {
1315 	int r = -EINVAL;
1316 
1317 	down_read(&pmd->root_lock);
1318 	if (!pmd->fail_io)
1319 		r = __get_metadata_snap(pmd, result);
1320 	up_read(&pmd->root_lock);
1321 
1322 	return r;
1323 }
1324 
1325 int dm_pool_open_thin_device(struct dm_pool_metadata *pmd, dm_thin_id dev,
1326 			     struct dm_thin_device **td)
1327 {
1328 	int r = -EINVAL;
1329 
1330 	down_write(&pmd->root_lock);
1331 	if (!pmd->fail_io)
1332 		r = __open_device(pmd, dev, 0, td);
1333 	up_write(&pmd->root_lock);
1334 
1335 	return r;
1336 }
1337 
1338 int dm_pool_close_thin_device(struct dm_thin_device *td)
1339 {
1340 	down_write(&td->pmd->root_lock);
1341 	__close_device(td);
1342 	up_write(&td->pmd->root_lock);
1343 
1344 	return 0;
1345 }
1346 
1347 dm_thin_id dm_thin_dev_id(struct dm_thin_device *td)
1348 {
1349 	return td->id;
1350 }
1351 
1352 static bool __snapshotted_since(struct dm_thin_device *td, uint32_t time)
1353 {
1354 	return td->snapshotted_time > time;
1355 }
1356 
1357 int dm_thin_find_block(struct dm_thin_device *td, dm_block_t block,
1358 		       int can_block, struct dm_thin_lookup_result *result)
1359 {
1360 	int r = -EINVAL;
1361 	uint64_t block_time = 0;
1362 	__le64 value;
1363 	struct dm_pool_metadata *pmd = td->pmd;
1364 	dm_block_t keys[2] = { td->id, block };
1365 	struct dm_btree_info *info;
1366 
1367 	if (can_block) {
1368 		down_read(&pmd->root_lock);
1369 		info = &pmd->info;
1370 	} else if (down_read_trylock(&pmd->root_lock))
1371 		info = &pmd->nb_info;
1372 	else
1373 		return -EWOULDBLOCK;
1374 
1375 	if (pmd->fail_io)
1376 		goto out;
1377 
1378 	r = dm_btree_lookup(info, pmd->root, keys, &value);
1379 	if (!r)
1380 		block_time = le64_to_cpu(value);
1381 
1382 out:
1383 	up_read(&pmd->root_lock);
1384 
1385 	if (!r) {
1386 		dm_block_t exception_block;
1387 		uint32_t exception_time;
1388 		unpack_block_time(block_time, &exception_block,
1389 				  &exception_time);
1390 		result->block = exception_block;
1391 		result->shared = __snapshotted_since(td, exception_time);
1392 	}
1393 
1394 	return r;
1395 }
1396 
1397 static int __insert(struct dm_thin_device *td, dm_block_t block,
1398 		    dm_block_t data_block)
1399 {
1400 	int r, inserted;
1401 	__le64 value;
1402 	struct dm_pool_metadata *pmd = td->pmd;
1403 	dm_block_t keys[2] = { td->id, block };
1404 
1405 	value = cpu_to_le64(pack_block_time(data_block, pmd->time));
1406 	__dm_bless_for_disk(&value);
1407 
1408 	r = dm_btree_insert_notify(&pmd->info, pmd->root, keys, &value,
1409 				   &pmd->root, &inserted);
1410 	if (r)
1411 		return r;
1412 
1413 	td->changed = 1;
1414 	if (inserted)
1415 		td->mapped_blocks++;
1416 
1417 	return 0;
1418 }
1419 
1420 int dm_thin_insert_block(struct dm_thin_device *td, dm_block_t block,
1421 			 dm_block_t data_block)
1422 {
1423 	int r = -EINVAL;
1424 
1425 	down_write(&td->pmd->root_lock);
1426 	if (!td->pmd->fail_io)
1427 		r = __insert(td, block, data_block);
1428 	up_write(&td->pmd->root_lock);
1429 
1430 	return r;
1431 }
1432 
1433 static int __remove(struct dm_thin_device *td, dm_block_t block)
1434 {
1435 	int r;
1436 	struct dm_pool_metadata *pmd = td->pmd;
1437 	dm_block_t keys[2] = { td->id, block };
1438 
1439 	r = dm_btree_remove(&pmd->info, pmd->root, keys, &pmd->root);
1440 	if (r)
1441 		return r;
1442 
1443 	td->mapped_blocks--;
1444 	td->changed = 1;
1445 
1446 	return 0;
1447 }
1448 
1449 int dm_thin_remove_block(struct dm_thin_device *td, dm_block_t block)
1450 {
1451 	int r = -EINVAL;
1452 
1453 	down_write(&td->pmd->root_lock);
1454 	if (!td->pmd->fail_io)
1455 		r = __remove(td, block);
1456 	up_write(&td->pmd->root_lock);
1457 
1458 	return r;
1459 }
1460 
1461 bool dm_thin_changed_this_transaction(struct dm_thin_device *td)
1462 {
1463 	int r;
1464 
1465 	down_read(&td->pmd->root_lock);
1466 	r = td->changed;
1467 	up_read(&td->pmd->root_lock);
1468 
1469 	return r;
1470 }
1471 
1472 bool dm_thin_aborted_changes(struct dm_thin_device *td)
1473 {
1474 	bool r;
1475 
1476 	down_read(&td->pmd->root_lock);
1477 	r = td->aborted_with_changes;
1478 	up_read(&td->pmd->root_lock);
1479 
1480 	return r;
1481 }
1482 
1483 int dm_pool_alloc_data_block(struct dm_pool_metadata *pmd, dm_block_t *result)
1484 {
1485 	int r = -EINVAL;
1486 
1487 	down_write(&pmd->root_lock);
1488 	if (!pmd->fail_io)
1489 		r = dm_sm_new_block(pmd->data_sm, result);
1490 	up_write(&pmd->root_lock);
1491 
1492 	return r;
1493 }
1494 
1495 int dm_pool_commit_metadata(struct dm_pool_metadata *pmd)
1496 {
1497 	int r = -EINVAL;
1498 
1499 	down_write(&pmd->root_lock);
1500 	if (pmd->fail_io)
1501 		goto out;
1502 
1503 	r = __commit_transaction(pmd);
1504 	if (r <= 0)
1505 		goto out;
1506 
1507 	/*
1508 	 * Open the next transaction.
1509 	 */
1510 	r = __begin_transaction(pmd);
1511 out:
1512 	up_write(&pmd->root_lock);
1513 	return r;
1514 }
1515 
1516 static void __set_abort_with_changes_flags(struct dm_pool_metadata *pmd)
1517 {
1518 	struct dm_thin_device *td;
1519 
1520 	list_for_each_entry(td, &pmd->thin_devices, list)
1521 		td->aborted_with_changes = td->changed;
1522 }
1523 
1524 int dm_pool_abort_metadata(struct dm_pool_metadata *pmd)
1525 {
1526 	int r = -EINVAL;
1527 
1528 	down_write(&pmd->root_lock);
1529 	if (pmd->fail_io)
1530 		goto out;
1531 
1532 	__set_abort_with_changes_flags(pmd);
1533 	__destroy_persistent_data_objects(pmd);
1534 	r = __create_persistent_data_objects(pmd, false);
1535 	if (r)
1536 		pmd->fail_io = true;
1537 
1538 out:
1539 	up_write(&pmd->root_lock);
1540 
1541 	return r;
1542 }
1543 
1544 int dm_pool_get_free_block_count(struct dm_pool_metadata *pmd, dm_block_t *result)
1545 {
1546 	int r = -EINVAL;
1547 
1548 	down_read(&pmd->root_lock);
1549 	if (!pmd->fail_io)
1550 		r = dm_sm_get_nr_free(pmd->data_sm, result);
1551 	up_read(&pmd->root_lock);
1552 
1553 	return r;
1554 }
1555 
1556 int dm_pool_get_free_metadata_block_count(struct dm_pool_metadata *pmd,
1557 					  dm_block_t *result)
1558 {
1559 	int r = -EINVAL;
1560 
1561 	down_read(&pmd->root_lock);
1562 	if (!pmd->fail_io)
1563 		r = dm_sm_get_nr_free(pmd->metadata_sm, result);
1564 	up_read(&pmd->root_lock);
1565 
1566 	return r;
1567 }
1568 
1569 int dm_pool_get_metadata_dev_size(struct dm_pool_metadata *pmd,
1570 				  dm_block_t *result)
1571 {
1572 	int r = -EINVAL;
1573 
1574 	down_read(&pmd->root_lock);
1575 	if (!pmd->fail_io)
1576 		r = dm_sm_get_nr_blocks(pmd->metadata_sm, result);
1577 	up_read(&pmd->root_lock);
1578 
1579 	return r;
1580 }
1581 
1582 int dm_pool_get_data_block_size(struct dm_pool_metadata *pmd, sector_t *result)
1583 {
1584 	down_read(&pmd->root_lock);
1585 	*result = pmd->data_block_size;
1586 	up_read(&pmd->root_lock);
1587 
1588 	return 0;
1589 }
1590 
1591 int dm_pool_get_data_dev_size(struct dm_pool_metadata *pmd, dm_block_t *result)
1592 {
1593 	int r = -EINVAL;
1594 
1595 	down_read(&pmd->root_lock);
1596 	if (!pmd->fail_io)
1597 		r = dm_sm_get_nr_blocks(pmd->data_sm, result);
1598 	up_read(&pmd->root_lock);
1599 
1600 	return r;
1601 }
1602 
1603 int dm_thin_get_mapped_count(struct dm_thin_device *td, dm_block_t *result)
1604 {
1605 	int r = -EINVAL;
1606 	struct dm_pool_metadata *pmd = td->pmd;
1607 
1608 	down_read(&pmd->root_lock);
1609 	if (!pmd->fail_io) {
1610 		*result = td->mapped_blocks;
1611 		r = 0;
1612 	}
1613 	up_read(&pmd->root_lock);
1614 
1615 	return r;
1616 }
1617 
1618 static int __highest_block(struct dm_thin_device *td, dm_block_t *result)
1619 {
1620 	int r;
1621 	__le64 value_le;
1622 	dm_block_t thin_root;
1623 	struct dm_pool_metadata *pmd = td->pmd;
1624 
1625 	r = dm_btree_lookup(&pmd->tl_info, pmd->root, &td->id, &value_le);
1626 	if (r)
1627 		return r;
1628 
1629 	thin_root = le64_to_cpu(value_le);
1630 
1631 	return dm_btree_find_highest_key(&pmd->bl_info, thin_root, result);
1632 }
1633 
1634 int dm_thin_get_highest_mapped_block(struct dm_thin_device *td,
1635 				     dm_block_t *result)
1636 {
1637 	int r = -EINVAL;
1638 	struct dm_pool_metadata *pmd = td->pmd;
1639 
1640 	down_read(&pmd->root_lock);
1641 	if (!pmd->fail_io)
1642 		r = __highest_block(td, result);
1643 	up_read(&pmd->root_lock);
1644 
1645 	return r;
1646 }
1647 
1648 static int __resize_space_map(struct dm_space_map *sm, dm_block_t new_count)
1649 {
1650 	int r;
1651 	dm_block_t old_count;
1652 
1653 	r = dm_sm_get_nr_blocks(sm, &old_count);
1654 	if (r)
1655 		return r;
1656 
1657 	if (new_count == old_count)
1658 		return 0;
1659 
1660 	if (new_count < old_count) {
1661 		DMERR("cannot reduce size of space map");
1662 		return -EINVAL;
1663 	}
1664 
1665 	return dm_sm_extend(sm, new_count - old_count);
1666 }
1667 
1668 int dm_pool_resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1669 {
1670 	int r = -EINVAL;
1671 
1672 	down_write(&pmd->root_lock);
1673 	if (!pmd->fail_io)
1674 		r = __resize_space_map(pmd->data_sm, new_count);
1675 	up_write(&pmd->root_lock);
1676 
1677 	return r;
1678 }
1679 
1680 int dm_pool_resize_metadata_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1681 {
1682 	int r = -EINVAL;
1683 
1684 	down_write(&pmd->root_lock);
1685 	if (!pmd->fail_io)
1686 		r = __resize_space_map(pmd->metadata_sm, new_count);
1687 	up_write(&pmd->root_lock);
1688 
1689 	return r;
1690 }
1691 
1692 void dm_pool_metadata_read_only(struct dm_pool_metadata *pmd)
1693 {
1694 	down_write(&pmd->root_lock);
1695 	pmd->read_only = true;
1696 	dm_bm_set_read_only(pmd->bm);
1697 	up_write(&pmd->root_lock);
1698 }
1699 
1700 int dm_pool_register_metadata_threshold(struct dm_pool_metadata *pmd,
1701 					dm_block_t threshold,
1702 					dm_sm_threshold_fn fn,
1703 					void *context)
1704 {
1705 	int r;
1706 
1707 	down_write(&pmd->root_lock);
1708 	r = dm_sm_register_threshold_callback(pmd->metadata_sm, threshold, fn, context);
1709 	up_write(&pmd->root_lock);
1710 
1711 	return r;
1712 }
1713