xref: /linux/drivers/md/dm-thin-metadata.c (revision 905e46acd3272d04566fec49afbd7ad9e2ed9ae3)
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 2
80 #define SECTOR_TO_BLOCK_SHIFT 3
81 
82 /*
83  *  3 for btree insert +
84  *  2 for btree lookup used within space map
85  */
86 #define THIN_MAX_CONCURRENT_LOCKS 5
87 
88 /* This should be plenty */
89 #define SPACE_MAP_ROOT_SIZE 128
90 
91 /*
92  * Little endian on-disk superblock and device details.
93  */
94 struct thin_disk_superblock {
95 	__le32 csum;	/* Checksum of superblock except for this field. */
96 	__le32 flags;
97 	__le64 blocknr;	/* This block number, dm_block_t. */
98 
99 	__u8 uuid[16];
100 	__le64 magic;
101 	__le32 version;
102 	__le32 time;
103 
104 	__le64 trans_id;
105 
106 	/*
107 	 * Root held by userspace transactions.
108 	 */
109 	__le64 held_root;
110 
111 	__u8 data_space_map_root[SPACE_MAP_ROOT_SIZE];
112 	__u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
113 
114 	/*
115 	 * 2-level btree mapping (dev_id, (dev block, time)) -> data block
116 	 */
117 	__le64 data_mapping_root;
118 
119 	/*
120 	 * Device detail root mapping dev_id -> device_details
121 	 */
122 	__le64 device_details_root;
123 
124 	__le32 data_block_size;		/* In 512-byte sectors. */
125 
126 	__le32 metadata_block_size;	/* In 512-byte sectors. */
127 	__le64 metadata_nr_blocks;
128 
129 	__le32 compat_flags;
130 	__le32 compat_ro_flags;
131 	__le32 incompat_flags;
132 } __packed;
133 
134 struct disk_device_details {
135 	__le64 mapped_blocks;
136 	__le64 transaction_id;		/* When created. */
137 	__le32 creation_time;
138 	__le32 snapshotted_time;
139 } __packed;
140 
141 struct dm_pool_metadata {
142 	struct hlist_node hash;
143 
144 	struct block_device *bdev;
145 	struct dm_block_manager *bm;
146 	struct dm_space_map *metadata_sm;
147 	struct dm_space_map *data_sm;
148 	struct dm_transaction_manager *tm;
149 	struct dm_transaction_manager *nb_tm;
150 
151 	/*
152 	 * Two-level btree.
153 	 * First level holds thin_dev_t.
154 	 * Second level holds mappings.
155 	 */
156 	struct dm_btree_info info;
157 
158 	/*
159 	 * Non-blocking version of the above.
160 	 */
161 	struct dm_btree_info nb_info;
162 
163 	/*
164 	 * Just the top level for deleting whole devices.
165 	 */
166 	struct dm_btree_info tl_info;
167 
168 	/*
169 	 * Just the bottom level for creating new devices.
170 	 */
171 	struct dm_btree_info bl_info;
172 
173 	/*
174 	 * Describes the device details btree.
175 	 */
176 	struct dm_btree_info details_info;
177 
178 	struct rw_semaphore root_lock;
179 	uint32_t time;
180 	dm_block_t root;
181 	dm_block_t details_root;
182 	struct list_head thin_devices;
183 	uint64_t trans_id;
184 	unsigned long flags;
185 	sector_t data_block_size;
186 
187 	/*
188 	 * Set if a transaction has to be aborted but the attempt to roll back
189 	 * to the previous (good) transaction failed.  The only pool metadata
190 	 * operation possible in this state is the closing of the device.
191 	 */
192 	bool fail_io:1;
193 
194 	/*
195 	 * Reading the space map roots can fail, so we read it into these
196 	 * buffers before the superblock is locked and updated.
197 	 */
198 	__u8 data_space_map_root[SPACE_MAP_ROOT_SIZE];
199 	__u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
200 };
201 
202 struct dm_thin_device {
203 	struct list_head list;
204 	struct dm_pool_metadata *pmd;
205 	dm_thin_id id;
206 
207 	int open_count;
208 	bool changed:1;
209 	bool aborted_with_changes:1;
210 	uint64_t mapped_blocks;
211 	uint64_t transaction_id;
212 	uint32_t creation_time;
213 	uint32_t snapshotted_time;
214 };
215 
216 /*----------------------------------------------------------------
217  * superblock validator
218  *--------------------------------------------------------------*/
219 
220 #define SUPERBLOCK_CSUM_XOR 160774
221 
222 static void sb_prepare_for_write(struct dm_block_validator *v,
223 				 struct dm_block *b,
224 				 size_t block_size)
225 {
226 	struct thin_disk_superblock *disk_super = dm_block_data(b);
227 
228 	disk_super->blocknr = cpu_to_le64(dm_block_location(b));
229 	disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
230 						      block_size - sizeof(__le32),
231 						      SUPERBLOCK_CSUM_XOR));
232 }
233 
234 static int sb_check(struct dm_block_validator *v,
235 		    struct dm_block *b,
236 		    size_t block_size)
237 {
238 	struct thin_disk_superblock *disk_super = dm_block_data(b);
239 	__le32 csum_le;
240 
241 	if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
242 		DMERR("sb_check failed: blocknr %llu: "
243 		      "wanted %llu", le64_to_cpu(disk_super->blocknr),
244 		      (unsigned long long)dm_block_location(b));
245 		return -ENOTBLK;
246 	}
247 
248 	if (le64_to_cpu(disk_super->magic) != THIN_SUPERBLOCK_MAGIC) {
249 		DMERR("sb_check failed: magic %llu: "
250 		      "wanted %llu", le64_to_cpu(disk_super->magic),
251 		      (unsigned long long)THIN_SUPERBLOCK_MAGIC);
252 		return -EILSEQ;
253 	}
254 
255 	csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
256 					     block_size - sizeof(__le32),
257 					     SUPERBLOCK_CSUM_XOR));
258 	if (csum_le != disk_super->csum) {
259 		DMERR("sb_check failed: csum %u: wanted %u",
260 		      le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
261 		return -EILSEQ;
262 	}
263 
264 	return 0;
265 }
266 
267 static struct dm_block_validator sb_validator = {
268 	.name = "superblock",
269 	.prepare_for_write = sb_prepare_for_write,
270 	.check = sb_check
271 };
272 
273 /*----------------------------------------------------------------
274  * Methods for the btree value types
275  *--------------------------------------------------------------*/
276 
277 static uint64_t pack_block_time(dm_block_t b, uint32_t t)
278 {
279 	return (b << 24) | t;
280 }
281 
282 static void unpack_block_time(uint64_t v, dm_block_t *b, uint32_t *t)
283 {
284 	*b = v >> 24;
285 	*t = v & ((1 << 24) - 1);
286 }
287 
288 static void data_block_inc(void *context, const void *value_le)
289 {
290 	struct dm_space_map *sm = context;
291 	__le64 v_le;
292 	uint64_t b;
293 	uint32_t t;
294 
295 	memcpy(&v_le, value_le, sizeof(v_le));
296 	unpack_block_time(le64_to_cpu(v_le), &b, &t);
297 	dm_sm_inc_block(sm, b);
298 }
299 
300 static void data_block_dec(void *context, const void *value_le)
301 {
302 	struct dm_space_map *sm = context;
303 	__le64 v_le;
304 	uint64_t b;
305 	uint32_t t;
306 
307 	memcpy(&v_le, value_le, sizeof(v_le));
308 	unpack_block_time(le64_to_cpu(v_le), &b, &t);
309 	dm_sm_dec_block(sm, b);
310 }
311 
312 static int data_block_equal(void *context, const void *value1_le, const void *value2_le)
313 {
314 	__le64 v1_le, v2_le;
315 	uint64_t b1, b2;
316 	uint32_t t;
317 
318 	memcpy(&v1_le, value1_le, sizeof(v1_le));
319 	memcpy(&v2_le, value2_le, sizeof(v2_le));
320 	unpack_block_time(le64_to_cpu(v1_le), &b1, &t);
321 	unpack_block_time(le64_to_cpu(v2_le), &b2, &t);
322 
323 	return b1 == b2;
324 }
325 
326 static void subtree_inc(void *context, const void *value)
327 {
328 	struct dm_btree_info *info = context;
329 	__le64 root_le;
330 	uint64_t root;
331 
332 	memcpy(&root_le, value, sizeof(root_le));
333 	root = le64_to_cpu(root_le);
334 	dm_tm_inc(info->tm, root);
335 }
336 
337 static void subtree_dec(void *context, const void *value)
338 {
339 	struct dm_btree_info *info = context;
340 	__le64 root_le;
341 	uint64_t root;
342 
343 	memcpy(&root_le, value, sizeof(root_le));
344 	root = le64_to_cpu(root_le);
345 	if (dm_btree_del(info, root))
346 		DMERR("btree delete failed");
347 }
348 
349 static int subtree_equal(void *context, const void *value1_le, const void *value2_le)
350 {
351 	__le64 v1_le, v2_le;
352 	memcpy(&v1_le, value1_le, sizeof(v1_le));
353 	memcpy(&v2_le, value2_le, sizeof(v2_le));
354 
355 	return v1_le == v2_le;
356 }
357 
358 /*----------------------------------------------------------------*/
359 
360 static int superblock_lock_zero(struct dm_pool_metadata *pmd,
361 				struct dm_block **sblock)
362 {
363 	return dm_bm_write_lock_zero(pmd->bm, THIN_SUPERBLOCK_LOCATION,
364 				     &sb_validator, sblock);
365 }
366 
367 static int superblock_lock(struct dm_pool_metadata *pmd,
368 			   struct dm_block **sblock)
369 {
370 	return dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
371 				&sb_validator, sblock);
372 }
373 
374 static int __superblock_all_zeroes(struct dm_block_manager *bm, int *result)
375 {
376 	int r;
377 	unsigned i;
378 	struct dm_block *b;
379 	__le64 *data_le, zero = cpu_to_le64(0);
380 	unsigned block_size = dm_bm_block_size(bm) / sizeof(__le64);
381 
382 	/*
383 	 * We can't use a validator here - it may be all zeroes.
384 	 */
385 	r = dm_bm_read_lock(bm, THIN_SUPERBLOCK_LOCATION, NULL, &b);
386 	if (r)
387 		return r;
388 
389 	data_le = dm_block_data(b);
390 	*result = 1;
391 	for (i = 0; i < block_size; i++) {
392 		if (data_le[i] != zero) {
393 			*result = 0;
394 			break;
395 		}
396 	}
397 
398 	dm_bm_unlock(b);
399 
400 	return 0;
401 }
402 
403 static void __setup_btree_details(struct dm_pool_metadata *pmd)
404 {
405 	pmd->info.tm = pmd->tm;
406 	pmd->info.levels = 2;
407 	pmd->info.value_type.context = pmd->data_sm;
408 	pmd->info.value_type.size = sizeof(__le64);
409 	pmd->info.value_type.inc = data_block_inc;
410 	pmd->info.value_type.dec = data_block_dec;
411 	pmd->info.value_type.equal = data_block_equal;
412 
413 	memcpy(&pmd->nb_info, &pmd->info, sizeof(pmd->nb_info));
414 	pmd->nb_info.tm = pmd->nb_tm;
415 
416 	pmd->tl_info.tm = pmd->tm;
417 	pmd->tl_info.levels = 1;
418 	pmd->tl_info.value_type.context = &pmd->bl_info;
419 	pmd->tl_info.value_type.size = sizeof(__le64);
420 	pmd->tl_info.value_type.inc = subtree_inc;
421 	pmd->tl_info.value_type.dec = subtree_dec;
422 	pmd->tl_info.value_type.equal = subtree_equal;
423 
424 	pmd->bl_info.tm = pmd->tm;
425 	pmd->bl_info.levels = 1;
426 	pmd->bl_info.value_type.context = pmd->data_sm;
427 	pmd->bl_info.value_type.size = sizeof(__le64);
428 	pmd->bl_info.value_type.inc = data_block_inc;
429 	pmd->bl_info.value_type.dec = data_block_dec;
430 	pmd->bl_info.value_type.equal = data_block_equal;
431 
432 	pmd->details_info.tm = pmd->tm;
433 	pmd->details_info.levels = 1;
434 	pmd->details_info.value_type.context = NULL;
435 	pmd->details_info.value_type.size = sizeof(struct disk_device_details);
436 	pmd->details_info.value_type.inc = NULL;
437 	pmd->details_info.value_type.dec = NULL;
438 	pmd->details_info.value_type.equal = NULL;
439 }
440 
441 static int save_sm_roots(struct dm_pool_metadata *pmd)
442 {
443 	int r;
444 	size_t len;
445 
446 	r = dm_sm_root_size(pmd->metadata_sm, &len);
447 	if (r < 0)
448 		return r;
449 
450 	r = dm_sm_copy_root(pmd->metadata_sm, &pmd->metadata_space_map_root, len);
451 	if (r < 0)
452 		return r;
453 
454 	r = dm_sm_root_size(pmd->data_sm, &len);
455 	if (r < 0)
456 		return r;
457 
458 	return dm_sm_copy_root(pmd->data_sm, &pmd->data_space_map_root, len);
459 }
460 
461 static void copy_sm_roots(struct dm_pool_metadata *pmd,
462 			  struct thin_disk_superblock *disk)
463 {
464 	memcpy(&disk->metadata_space_map_root,
465 	       &pmd->metadata_space_map_root,
466 	       sizeof(pmd->metadata_space_map_root));
467 
468 	memcpy(&disk->data_space_map_root,
469 	       &pmd->data_space_map_root,
470 	       sizeof(pmd->data_space_map_root));
471 }
472 
473 static int __write_initial_superblock(struct dm_pool_metadata *pmd)
474 {
475 	int r;
476 	struct dm_block *sblock;
477 	struct thin_disk_superblock *disk_super;
478 	sector_t bdev_size = i_size_read(pmd->bdev->bd_inode) >> SECTOR_SHIFT;
479 
480 	if (bdev_size > THIN_METADATA_MAX_SECTORS)
481 		bdev_size = THIN_METADATA_MAX_SECTORS;
482 
483 	r = dm_sm_commit(pmd->data_sm);
484 	if (r < 0)
485 		return r;
486 
487 	r = save_sm_roots(pmd);
488 	if (r < 0)
489 		return r;
490 
491 	r = dm_tm_pre_commit(pmd->tm);
492 	if (r < 0)
493 		return r;
494 
495 	r = superblock_lock_zero(pmd, &sblock);
496 	if (r)
497 		return r;
498 
499 	disk_super = dm_block_data(sblock);
500 	disk_super->flags = 0;
501 	memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
502 	disk_super->magic = cpu_to_le64(THIN_SUPERBLOCK_MAGIC);
503 	disk_super->version = cpu_to_le32(THIN_VERSION);
504 	disk_super->time = 0;
505 	disk_super->trans_id = 0;
506 	disk_super->held_root = 0;
507 
508 	copy_sm_roots(pmd, disk_super);
509 
510 	disk_super->data_mapping_root = cpu_to_le64(pmd->root);
511 	disk_super->device_details_root = cpu_to_le64(pmd->details_root);
512 	disk_super->metadata_block_size = cpu_to_le32(THIN_METADATA_BLOCK_SIZE);
513 	disk_super->metadata_nr_blocks = cpu_to_le64(bdev_size >> SECTOR_TO_BLOCK_SHIFT);
514 	disk_super->data_block_size = cpu_to_le32(pmd->data_block_size);
515 
516 	return dm_tm_commit(pmd->tm, sblock);
517 }
518 
519 static int __format_metadata(struct dm_pool_metadata *pmd)
520 {
521 	int r;
522 
523 	r = dm_tm_create_with_sm(pmd->bm, THIN_SUPERBLOCK_LOCATION,
524 				 &pmd->tm, &pmd->metadata_sm);
525 	if (r < 0) {
526 		DMERR("tm_create_with_sm failed");
527 		return r;
528 	}
529 
530 	pmd->data_sm = dm_sm_disk_create(pmd->tm, 0);
531 	if (IS_ERR(pmd->data_sm)) {
532 		DMERR("sm_disk_create failed");
533 		r = PTR_ERR(pmd->data_sm);
534 		goto bad_cleanup_tm;
535 	}
536 
537 	pmd->nb_tm = dm_tm_create_non_blocking_clone(pmd->tm);
538 	if (!pmd->nb_tm) {
539 		DMERR("could not create non-blocking clone tm");
540 		r = -ENOMEM;
541 		goto bad_cleanup_data_sm;
542 	}
543 
544 	__setup_btree_details(pmd);
545 
546 	r = dm_btree_empty(&pmd->info, &pmd->root);
547 	if (r < 0)
548 		goto bad_cleanup_nb_tm;
549 
550 	r = dm_btree_empty(&pmd->details_info, &pmd->details_root);
551 	if (r < 0) {
552 		DMERR("couldn't create devices root");
553 		goto bad_cleanup_nb_tm;
554 	}
555 
556 	r = __write_initial_superblock(pmd);
557 	if (r)
558 		goto bad_cleanup_nb_tm;
559 
560 	return 0;
561 
562 bad_cleanup_nb_tm:
563 	dm_tm_destroy(pmd->nb_tm);
564 bad_cleanup_data_sm:
565 	dm_sm_destroy(pmd->data_sm);
566 bad_cleanup_tm:
567 	dm_tm_destroy(pmd->tm);
568 	dm_sm_destroy(pmd->metadata_sm);
569 
570 	return r;
571 }
572 
573 static int __check_incompat_features(struct thin_disk_superblock *disk_super,
574 				     struct dm_pool_metadata *pmd)
575 {
576 	uint32_t features;
577 
578 	features = le32_to_cpu(disk_super->incompat_flags) & ~THIN_FEATURE_INCOMPAT_SUPP;
579 	if (features) {
580 		DMERR("could not access metadata due to unsupported optional features (%lx).",
581 		      (unsigned long)features);
582 		return -EINVAL;
583 	}
584 
585 	/*
586 	 * Check for read-only metadata to skip the following RDWR checks.
587 	 */
588 	if (get_disk_ro(pmd->bdev->bd_disk))
589 		return 0;
590 
591 	features = le32_to_cpu(disk_super->compat_ro_flags) & ~THIN_FEATURE_COMPAT_RO_SUPP;
592 	if (features) {
593 		DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
594 		      (unsigned long)features);
595 		return -EINVAL;
596 	}
597 
598 	return 0;
599 }
600 
601 static int __open_metadata(struct dm_pool_metadata *pmd)
602 {
603 	int r;
604 	struct dm_block *sblock;
605 	struct thin_disk_superblock *disk_super;
606 
607 	r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
608 			    &sb_validator, &sblock);
609 	if (r < 0) {
610 		DMERR("couldn't read superblock");
611 		return r;
612 	}
613 
614 	disk_super = dm_block_data(sblock);
615 
616 	/* Verify the data block size hasn't changed */
617 	if (le32_to_cpu(disk_super->data_block_size) != pmd->data_block_size) {
618 		DMERR("changing the data block size (from %u to %llu) is not supported",
619 		      le32_to_cpu(disk_super->data_block_size),
620 		      (unsigned long long)pmd->data_block_size);
621 		r = -EINVAL;
622 		goto bad_unlock_sblock;
623 	}
624 
625 	r = __check_incompat_features(disk_super, pmd);
626 	if (r < 0)
627 		goto bad_unlock_sblock;
628 
629 	r = dm_tm_open_with_sm(pmd->bm, THIN_SUPERBLOCK_LOCATION,
630 			       disk_super->metadata_space_map_root,
631 			       sizeof(disk_super->metadata_space_map_root),
632 			       &pmd->tm, &pmd->metadata_sm);
633 	if (r < 0) {
634 		DMERR("tm_open_with_sm failed");
635 		goto bad_unlock_sblock;
636 	}
637 
638 	pmd->data_sm = dm_sm_disk_open(pmd->tm, disk_super->data_space_map_root,
639 				       sizeof(disk_super->data_space_map_root));
640 	if (IS_ERR(pmd->data_sm)) {
641 		DMERR("sm_disk_open failed");
642 		r = PTR_ERR(pmd->data_sm);
643 		goto bad_cleanup_tm;
644 	}
645 
646 	pmd->nb_tm = dm_tm_create_non_blocking_clone(pmd->tm);
647 	if (!pmd->nb_tm) {
648 		DMERR("could not create non-blocking clone tm");
649 		r = -ENOMEM;
650 		goto bad_cleanup_data_sm;
651 	}
652 
653 	__setup_btree_details(pmd);
654 	dm_bm_unlock(sblock);
655 
656 	return 0;
657 
658 bad_cleanup_data_sm:
659 	dm_sm_destroy(pmd->data_sm);
660 bad_cleanup_tm:
661 	dm_tm_destroy(pmd->tm);
662 	dm_sm_destroy(pmd->metadata_sm);
663 bad_unlock_sblock:
664 	dm_bm_unlock(sblock);
665 
666 	return r;
667 }
668 
669 static int __open_or_format_metadata(struct dm_pool_metadata *pmd, bool format_device)
670 {
671 	int r, unformatted;
672 
673 	r = __superblock_all_zeroes(pmd->bm, &unformatted);
674 	if (r)
675 		return r;
676 
677 	if (unformatted)
678 		return format_device ? __format_metadata(pmd) : -EPERM;
679 
680 	return __open_metadata(pmd);
681 }
682 
683 static int __create_persistent_data_objects(struct dm_pool_metadata *pmd, bool format_device)
684 {
685 	int r;
686 
687 	pmd->bm = dm_block_manager_create(pmd->bdev, THIN_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
688 					  THIN_MAX_CONCURRENT_LOCKS);
689 	if (IS_ERR(pmd->bm)) {
690 		DMERR("could not create block manager");
691 		return PTR_ERR(pmd->bm);
692 	}
693 
694 	r = __open_or_format_metadata(pmd, format_device);
695 	if (r)
696 		dm_block_manager_destroy(pmd->bm);
697 
698 	return r;
699 }
700 
701 static void __destroy_persistent_data_objects(struct dm_pool_metadata *pmd)
702 {
703 	dm_sm_destroy(pmd->data_sm);
704 	dm_sm_destroy(pmd->metadata_sm);
705 	dm_tm_destroy(pmd->nb_tm);
706 	dm_tm_destroy(pmd->tm);
707 	dm_block_manager_destroy(pmd->bm);
708 }
709 
710 static int __begin_transaction(struct dm_pool_metadata *pmd)
711 {
712 	int r;
713 	struct thin_disk_superblock *disk_super;
714 	struct dm_block *sblock;
715 
716 	/*
717 	 * We re-read the superblock every time.  Shouldn't need to do this
718 	 * really.
719 	 */
720 	r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
721 			    &sb_validator, &sblock);
722 	if (r)
723 		return r;
724 
725 	disk_super = dm_block_data(sblock);
726 	pmd->time = le32_to_cpu(disk_super->time);
727 	pmd->root = le64_to_cpu(disk_super->data_mapping_root);
728 	pmd->details_root = le64_to_cpu(disk_super->device_details_root);
729 	pmd->trans_id = le64_to_cpu(disk_super->trans_id);
730 	pmd->flags = le32_to_cpu(disk_super->flags);
731 	pmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
732 
733 	dm_bm_unlock(sblock);
734 	return 0;
735 }
736 
737 static int __write_changed_details(struct dm_pool_metadata *pmd)
738 {
739 	int r;
740 	struct dm_thin_device *td, *tmp;
741 	struct disk_device_details details;
742 	uint64_t key;
743 
744 	list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
745 		if (!td->changed)
746 			continue;
747 
748 		key = td->id;
749 
750 		details.mapped_blocks = cpu_to_le64(td->mapped_blocks);
751 		details.transaction_id = cpu_to_le64(td->transaction_id);
752 		details.creation_time = cpu_to_le32(td->creation_time);
753 		details.snapshotted_time = cpu_to_le32(td->snapshotted_time);
754 		__dm_bless_for_disk(&details);
755 
756 		r = dm_btree_insert(&pmd->details_info, pmd->details_root,
757 				    &key, &details, &pmd->details_root);
758 		if (r)
759 			return r;
760 
761 		if (td->open_count)
762 			td->changed = 0;
763 		else {
764 			list_del(&td->list);
765 			kfree(td);
766 		}
767 	}
768 
769 	return 0;
770 }
771 
772 static int __commit_transaction(struct dm_pool_metadata *pmd)
773 {
774 	int r;
775 	size_t metadata_len, data_len;
776 	struct thin_disk_superblock *disk_super;
777 	struct dm_block *sblock;
778 
779 	/*
780 	 * We need to know if the thin_disk_superblock exceeds a 512-byte sector.
781 	 */
782 	BUILD_BUG_ON(sizeof(struct thin_disk_superblock) > 512);
783 
784 	r = __write_changed_details(pmd);
785 	if (r < 0)
786 		return r;
787 
788 	r = dm_sm_commit(pmd->data_sm);
789 	if (r < 0)
790 		return r;
791 
792 	r = dm_tm_pre_commit(pmd->tm);
793 	if (r < 0)
794 		return r;
795 
796 	r = dm_sm_root_size(pmd->metadata_sm, &metadata_len);
797 	if (r < 0)
798 		return r;
799 
800 	r = dm_sm_root_size(pmd->data_sm, &data_len);
801 	if (r < 0)
802 		return r;
803 
804 	r = save_sm_roots(pmd);
805 	if (r < 0)
806 		return r;
807 
808 	r = superblock_lock(pmd, &sblock);
809 	if (r)
810 		return r;
811 
812 	disk_super = dm_block_data(sblock);
813 	disk_super->time = cpu_to_le32(pmd->time);
814 	disk_super->data_mapping_root = cpu_to_le64(pmd->root);
815 	disk_super->device_details_root = cpu_to_le64(pmd->details_root);
816 	disk_super->trans_id = cpu_to_le64(pmd->trans_id);
817 	disk_super->flags = cpu_to_le32(pmd->flags);
818 
819 	copy_sm_roots(pmd, disk_super);
820 
821 	return dm_tm_commit(pmd->tm, sblock);
822 }
823 
824 struct dm_pool_metadata *dm_pool_metadata_open(struct block_device *bdev,
825 					       sector_t data_block_size,
826 					       bool format_device)
827 {
828 	int r;
829 	struct dm_pool_metadata *pmd;
830 
831 	pmd = kmalloc(sizeof(*pmd), GFP_KERNEL);
832 	if (!pmd) {
833 		DMERR("could not allocate metadata struct");
834 		return ERR_PTR(-ENOMEM);
835 	}
836 
837 	init_rwsem(&pmd->root_lock);
838 	pmd->time = 0;
839 	INIT_LIST_HEAD(&pmd->thin_devices);
840 	pmd->fail_io = false;
841 	pmd->bdev = bdev;
842 	pmd->data_block_size = data_block_size;
843 
844 	r = __create_persistent_data_objects(pmd, format_device);
845 	if (r) {
846 		kfree(pmd);
847 		return ERR_PTR(r);
848 	}
849 
850 	r = __begin_transaction(pmd);
851 	if (r < 0) {
852 		if (dm_pool_metadata_close(pmd) < 0)
853 			DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
854 		return ERR_PTR(r);
855 	}
856 
857 	return pmd;
858 }
859 
860 int dm_pool_metadata_close(struct dm_pool_metadata *pmd)
861 {
862 	int r;
863 	unsigned open_devices = 0;
864 	struct dm_thin_device *td, *tmp;
865 
866 	down_read(&pmd->root_lock);
867 	list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
868 		if (td->open_count)
869 			open_devices++;
870 		else {
871 			list_del(&td->list);
872 			kfree(td);
873 		}
874 	}
875 	up_read(&pmd->root_lock);
876 
877 	if (open_devices) {
878 		DMERR("attempt to close pmd when %u device(s) are still open",
879 		       open_devices);
880 		return -EBUSY;
881 	}
882 
883 	if (!dm_bm_is_read_only(pmd->bm) && !pmd->fail_io) {
884 		r = __commit_transaction(pmd);
885 		if (r < 0)
886 			DMWARN("%s: __commit_transaction() failed, error = %d",
887 			       __func__, r);
888 	}
889 
890 	if (!pmd->fail_io)
891 		__destroy_persistent_data_objects(pmd);
892 
893 	kfree(pmd);
894 	return 0;
895 }
896 
897 /*
898  * __open_device: Returns @td corresponding to device with id @dev,
899  * creating it if @create is set and incrementing @td->open_count.
900  * On failure, @td is undefined.
901  */
902 static int __open_device(struct dm_pool_metadata *pmd,
903 			 dm_thin_id dev, int create,
904 			 struct dm_thin_device **td)
905 {
906 	int r, changed = 0;
907 	struct dm_thin_device *td2;
908 	uint64_t key = dev;
909 	struct disk_device_details details_le;
910 
911 	/*
912 	 * If the device is already open, return it.
913 	 */
914 	list_for_each_entry(td2, &pmd->thin_devices, list)
915 		if (td2->id == dev) {
916 			/*
917 			 * May not create an already-open device.
918 			 */
919 			if (create)
920 				return -EEXIST;
921 
922 			td2->open_count++;
923 			*td = td2;
924 			return 0;
925 		}
926 
927 	/*
928 	 * Check the device exists.
929 	 */
930 	r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
931 			    &key, &details_le);
932 	if (r) {
933 		if (r != -ENODATA || !create)
934 			return r;
935 
936 		/*
937 		 * Create new device.
938 		 */
939 		changed = 1;
940 		details_le.mapped_blocks = 0;
941 		details_le.transaction_id = cpu_to_le64(pmd->trans_id);
942 		details_le.creation_time = cpu_to_le32(pmd->time);
943 		details_le.snapshotted_time = cpu_to_le32(pmd->time);
944 	}
945 
946 	*td = kmalloc(sizeof(**td), GFP_NOIO);
947 	if (!*td)
948 		return -ENOMEM;
949 
950 	(*td)->pmd = pmd;
951 	(*td)->id = dev;
952 	(*td)->open_count = 1;
953 	(*td)->changed = changed;
954 	(*td)->aborted_with_changes = false;
955 	(*td)->mapped_blocks = le64_to_cpu(details_le.mapped_blocks);
956 	(*td)->transaction_id = le64_to_cpu(details_le.transaction_id);
957 	(*td)->creation_time = le32_to_cpu(details_le.creation_time);
958 	(*td)->snapshotted_time = le32_to_cpu(details_le.snapshotted_time);
959 
960 	list_add(&(*td)->list, &pmd->thin_devices);
961 
962 	return 0;
963 }
964 
965 static void __close_device(struct dm_thin_device *td)
966 {
967 	--td->open_count;
968 }
969 
970 static int __create_thin(struct dm_pool_metadata *pmd,
971 			 dm_thin_id dev)
972 {
973 	int r;
974 	dm_block_t dev_root;
975 	uint64_t key = dev;
976 	struct disk_device_details details_le;
977 	struct dm_thin_device *td;
978 	__le64 value;
979 
980 	r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
981 			    &key, &details_le);
982 	if (!r)
983 		return -EEXIST;
984 
985 	/*
986 	 * Create an empty btree for the mappings.
987 	 */
988 	r = dm_btree_empty(&pmd->bl_info, &dev_root);
989 	if (r)
990 		return r;
991 
992 	/*
993 	 * Insert it into the main mapping tree.
994 	 */
995 	value = cpu_to_le64(dev_root);
996 	__dm_bless_for_disk(&value);
997 	r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
998 	if (r) {
999 		dm_btree_del(&pmd->bl_info, dev_root);
1000 		return r;
1001 	}
1002 
1003 	r = __open_device(pmd, dev, 1, &td);
1004 	if (r) {
1005 		dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1006 		dm_btree_del(&pmd->bl_info, dev_root);
1007 		return r;
1008 	}
1009 	__close_device(td);
1010 
1011 	return r;
1012 }
1013 
1014 int dm_pool_create_thin(struct dm_pool_metadata *pmd, dm_thin_id dev)
1015 {
1016 	int r = -EINVAL;
1017 
1018 	down_write(&pmd->root_lock);
1019 	if (!pmd->fail_io)
1020 		r = __create_thin(pmd, dev);
1021 	up_write(&pmd->root_lock);
1022 
1023 	return r;
1024 }
1025 
1026 static int __set_snapshot_details(struct dm_pool_metadata *pmd,
1027 				  struct dm_thin_device *snap,
1028 				  dm_thin_id origin, uint32_t time)
1029 {
1030 	int r;
1031 	struct dm_thin_device *td;
1032 
1033 	r = __open_device(pmd, origin, 0, &td);
1034 	if (r)
1035 		return r;
1036 
1037 	td->changed = 1;
1038 	td->snapshotted_time = time;
1039 
1040 	snap->mapped_blocks = td->mapped_blocks;
1041 	snap->snapshotted_time = time;
1042 	__close_device(td);
1043 
1044 	return 0;
1045 }
1046 
1047 static int __create_snap(struct dm_pool_metadata *pmd,
1048 			 dm_thin_id dev, dm_thin_id origin)
1049 {
1050 	int r;
1051 	dm_block_t origin_root;
1052 	uint64_t key = origin, dev_key = dev;
1053 	struct dm_thin_device *td;
1054 	struct disk_device_details details_le;
1055 	__le64 value;
1056 
1057 	/* check this device is unused */
1058 	r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
1059 			    &dev_key, &details_le);
1060 	if (!r)
1061 		return -EEXIST;
1062 
1063 	/* find the mapping tree for the origin */
1064 	r = dm_btree_lookup(&pmd->tl_info, pmd->root, &key, &value);
1065 	if (r)
1066 		return r;
1067 	origin_root = le64_to_cpu(value);
1068 
1069 	/* clone the origin, an inc will do */
1070 	dm_tm_inc(pmd->tm, origin_root);
1071 
1072 	/* insert into the main mapping tree */
1073 	value = cpu_to_le64(origin_root);
1074 	__dm_bless_for_disk(&value);
1075 	key = dev;
1076 	r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
1077 	if (r) {
1078 		dm_tm_dec(pmd->tm, origin_root);
1079 		return r;
1080 	}
1081 
1082 	pmd->time++;
1083 
1084 	r = __open_device(pmd, dev, 1, &td);
1085 	if (r)
1086 		goto bad;
1087 
1088 	r = __set_snapshot_details(pmd, td, origin, pmd->time);
1089 	__close_device(td);
1090 
1091 	if (r)
1092 		goto bad;
1093 
1094 	return 0;
1095 
1096 bad:
1097 	dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1098 	dm_btree_remove(&pmd->details_info, pmd->details_root,
1099 			&key, &pmd->details_root);
1100 	return r;
1101 }
1102 
1103 int dm_pool_create_snap(struct dm_pool_metadata *pmd,
1104 				 dm_thin_id dev,
1105 				 dm_thin_id origin)
1106 {
1107 	int r = -EINVAL;
1108 
1109 	down_write(&pmd->root_lock);
1110 	if (!pmd->fail_io)
1111 		r = __create_snap(pmd, dev, origin);
1112 	up_write(&pmd->root_lock);
1113 
1114 	return r;
1115 }
1116 
1117 static int __delete_device(struct dm_pool_metadata *pmd, dm_thin_id dev)
1118 {
1119 	int r;
1120 	uint64_t key = dev;
1121 	struct dm_thin_device *td;
1122 
1123 	/* TODO: failure should mark the transaction invalid */
1124 	r = __open_device(pmd, dev, 0, &td);
1125 	if (r)
1126 		return r;
1127 
1128 	if (td->open_count > 1) {
1129 		__close_device(td);
1130 		return -EBUSY;
1131 	}
1132 
1133 	list_del(&td->list);
1134 	kfree(td);
1135 	r = dm_btree_remove(&pmd->details_info, pmd->details_root,
1136 			    &key, &pmd->details_root);
1137 	if (r)
1138 		return r;
1139 
1140 	r = dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1141 	if (r)
1142 		return r;
1143 
1144 	return 0;
1145 }
1146 
1147 int dm_pool_delete_thin_device(struct dm_pool_metadata *pmd,
1148 			       dm_thin_id dev)
1149 {
1150 	int r = -EINVAL;
1151 
1152 	down_write(&pmd->root_lock);
1153 	if (!pmd->fail_io)
1154 		r = __delete_device(pmd, dev);
1155 	up_write(&pmd->root_lock);
1156 
1157 	return r;
1158 }
1159 
1160 int dm_pool_set_metadata_transaction_id(struct dm_pool_metadata *pmd,
1161 					uint64_t current_id,
1162 					uint64_t new_id)
1163 {
1164 	int r = -EINVAL;
1165 
1166 	down_write(&pmd->root_lock);
1167 
1168 	if (pmd->fail_io)
1169 		goto out;
1170 
1171 	if (pmd->trans_id != current_id) {
1172 		DMERR("mismatched transaction id");
1173 		goto out;
1174 	}
1175 
1176 	pmd->trans_id = new_id;
1177 	r = 0;
1178 
1179 out:
1180 	up_write(&pmd->root_lock);
1181 
1182 	return r;
1183 }
1184 
1185 int dm_pool_get_metadata_transaction_id(struct dm_pool_metadata *pmd,
1186 					uint64_t *result)
1187 {
1188 	int r = -EINVAL;
1189 
1190 	down_read(&pmd->root_lock);
1191 	if (!pmd->fail_io) {
1192 		*result = pmd->trans_id;
1193 		r = 0;
1194 	}
1195 	up_read(&pmd->root_lock);
1196 
1197 	return r;
1198 }
1199 
1200 static int __reserve_metadata_snap(struct dm_pool_metadata *pmd)
1201 {
1202 	int r, inc;
1203 	struct thin_disk_superblock *disk_super;
1204 	struct dm_block *copy, *sblock;
1205 	dm_block_t held_root;
1206 
1207 	/*
1208 	 * We commit to ensure the btree roots which we increment in a
1209 	 * moment are up to date.
1210 	 */
1211 	__commit_transaction(pmd);
1212 
1213 	/*
1214 	 * Copy the superblock.
1215 	 */
1216 	dm_sm_inc_block(pmd->metadata_sm, THIN_SUPERBLOCK_LOCATION);
1217 	r = dm_tm_shadow_block(pmd->tm, THIN_SUPERBLOCK_LOCATION,
1218 			       &sb_validator, &copy, &inc);
1219 	if (r)
1220 		return r;
1221 
1222 	BUG_ON(!inc);
1223 
1224 	held_root = dm_block_location(copy);
1225 	disk_super = dm_block_data(copy);
1226 
1227 	if (le64_to_cpu(disk_super->held_root)) {
1228 		DMWARN("Pool metadata snapshot already exists: release this before taking another.");
1229 
1230 		dm_tm_dec(pmd->tm, held_root);
1231 		dm_tm_unlock(pmd->tm, copy);
1232 		return -EBUSY;
1233 	}
1234 
1235 	/*
1236 	 * Wipe the spacemap since we're not publishing this.
1237 	 */
1238 	memset(&disk_super->data_space_map_root, 0,
1239 	       sizeof(disk_super->data_space_map_root));
1240 	memset(&disk_super->metadata_space_map_root, 0,
1241 	       sizeof(disk_super->metadata_space_map_root));
1242 
1243 	/*
1244 	 * Increment the data structures that need to be preserved.
1245 	 */
1246 	dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->data_mapping_root));
1247 	dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->device_details_root));
1248 	dm_tm_unlock(pmd->tm, copy);
1249 
1250 	/*
1251 	 * Write the held root into the superblock.
1252 	 */
1253 	r = superblock_lock(pmd, &sblock);
1254 	if (r) {
1255 		dm_tm_dec(pmd->tm, held_root);
1256 		return r;
1257 	}
1258 
1259 	disk_super = dm_block_data(sblock);
1260 	disk_super->held_root = cpu_to_le64(held_root);
1261 	dm_bm_unlock(sblock);
1262 	return 0;
1263 }
1264 
1265 int dm_pool_reserve_metadata_snap(struct dm_pool_metadata *pmd)
1266 {
1267 	int r = -EINVAL;
1268 
1269 	down_write(&pmd->root_lock);
1270 	if (!pmd->fail_io)
1271 		r = __reserve_metadata_snap(pmd);
1272 	up_write(&pmd->root_lock);
1273 
1274 	return r;
1275 }
1276 
1277 static int __release_metadata_snap(struct dm_pool_metadata *pmd)
1278 {
1279 	int r;
1280 	struct thin_disk_superblock *disk_super;
1281 	struct dm_block *sblock, *copy;
1282 	dm_block_t held_root;
1283 
1284 	r = superblock_lock(pmd, &sblock);
1285 	if (r)
1286 		return r;
1287 
1288 	disk_super = dm_block_data(sblock);
1289 	held_root = le64_to_cpu(disk_super->held_root);
1290 	disk_super->held_root = cpu_to_le64(0);
1291 
1292 	dm_bm_unlock(sblock);
1293 
1294 	if (!held_root) {
1295 		DMWARN("No pool metadata snapshot found: nothing to release.");
1296 		return -EINVAL;
1297 	}
1298 
1299 	r = dm_tm_read_lock(pmd->tm, held_root, &sb_validator, &copy);
1300 	if (r)
1301 		return r;
1302 
1303 	disk_super = dm_block_data(copy);
1304 	dm_btree_del(&pmd->info, le64_to_cpu(disk_super->data_mapping_root));
1305 	dm_btree_del(&pmd->details_info, le64_to_cpu(disk_super->device_details_root));
1306 	dm_sm_dec_block(pmd->metadata_sm, held_root);
1307 
1308 	dm_tm_unlock(pmd->tm, copy);
1309 
1310 	return 0;
1311 }
1312 
1313 int dm_pool_release_metadata_snap(struct dm_pool_metadata *pmd)
1314 {
1315 	int r = -EINVAL;
1316 
1317 	down_write(&pmd->root_lock);
1318 	if (!pmd->fail_io)
1319 		r = __release_metadata_snap(pmd);
1320 	up_write(&pmd->root_lock);
1321 
1322 	return r;
1323 }
1324 
1325 static int __get_metadata_snap(struct dm_pool_metadata *pmd,
1326 			       dm_block_t *result)
1327 {
1328 	int r;
1329 	struct thin_disk_superblock *disk_super;
1330 	struct dm_block *sblock;
1331 
1332 	r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
1333 			    &sb_validator, &sblock);
1334 	if (r)
1335 		return r;
1336 
1337 	disk_super = dm_block_data(sblock);
1338 	*result = le64_to_cpu(disk_super->held_root);
1339 
1340 	dm_bm_unlock(sblock);
1341 
1342 	return 0;
1343 }
1344 
1345 int dm_pool_get_metadata_snap(struct dm_pool_metadata *pmd,
1346 			      dm_block_t *result)
1347 {
1348 	int r = -EINVAL;
1349 
1350 	down_read(&pmd->root_lock);
1351 	if (!pmd->fail_io)
1352 		r = __get_metadata_snap(pmd, result);
1353 	up_read(&pmd->root_lock);
1354 
1355 	return r;
1356 }
1357 
1358 int dm_pool_open_thin_device(struct dm_pool_metadata *pmd, dm_thin_id dev,
1359 			     struct dm_thin_device **td)
1360 {
1361 	int r = -EINVAL;
1362 
1363 	down_write(&pmd->root_lock);
1364 	if (!pmd->fail_io)
1365 		r = __open_device(pmd, dev, 0, td);
1366 	up_write(&pmd->root_lock);
1367 
1368 	return r;
1369 }
1370 
1371 int dm_pool_close_thin_device(struct dm_thin_device *td)
1372 {
1373 	down_write(&td->pmd->root_lock);
1374 	__close_device(td);
1375 	up_write(&td->pmd->root_lock);
1376 
1377 	return 0;
1378 }
1379 
1380 dm_thin_id dm_thin_dev_id(struct dm_thin_device *td)
1381 {
1382 	return td->id;
1383 }
1384 
1385 /*
1386  * Check whether @time (of block creation) is older than @td's last snapshot.
1387  * If so then the associated block is shared with the last snapshot device.
1388  * Any block on a device created *after* the device last got snapshotted is
1389  * necessarily not shared.
1390  */
1391 static bool __snapshotted_since(struct dm_thin_device *td, uint32_t time)
1392 {
1393 	return td->snapshotted_time > time;
1394 }
1395 
1396 static void unpack_lookup_result(struct dm_thin_device *td, __le64 value,
1397 				 struct dm_thin_lookup_result *result)
1398 {
1399 	uint64_t block_time = 0;
1400 	dm_block_t exception_block;
1401 	uint32_t exception_time;
1402 
1403 	block_time = le64_to_cpu(value);
1404 	unpack_block_time(block_time, &exception_block, &exception_time);
1405 	result->block = exception_block;
1406 	result->shared = __snapshotted_since(td, exception_time);
1407 }
1408 
1409 static int __find_block(struct dm_thin_device *td, dm_block_t block,
1410 			int can_issue_io, struct dm_thin_lookup_result *result)
1411 {
1412 	int r;
1413 	__le64 value;
1414 	struct dm_pool_metadata *pmd = td->pmd;
1415 	dm_block_t keys[2] = { td->id, block };
1416 	struct dm_btree_info *info;
1417 
1418 	if (can_issue_io) {
1419 		info = &pmd->info;
1420 	} else
1421 		info = &pmd->nb_info;
1422 
1423 	r = dm_btree_lookup(info, pmd->root, keys, &value);
1424 	if (!r)
1425 		unpack_lookup_result(td, value, result);
1426 
1427 	return r;
1428 }
1429 
1430 int dm_thin_find_block(struct dm_thin_device *td, dm_block_t block,
1431 		       int can_issue_io, struct dm_thin_lookup_result *result)
1432 {
1433 	int r;
1434 	struct dm_pool_metadata *pmd = td->pmd;
1435 
1436 	down_read(&pmd->root_lock);
1437 	if (pmd->fail_io) {
1438 		up_read(&pmd->root_lock);
1439 		return -EINVAL;
1440 	}
1441 
1442 	r = __find_block(td, block, can_issue_io, result);
1443 
1444 	up_read(&pmd->root_lock);
1445 	return r;
1446 }
1447 
1448 static int __find_next_mapped_block(struct dm_thin_device *td, dm_block_t block,
1449 					  dm_block_t *vblock,
1450 					  struct dm_thin_lookup_result *result)
1451 {
1452 	int r;
1453 	__le64 value;
1454 	struct dm_pool_metadata *pmd = td->pmd;
1455 	dm_block_t keys[2] = { td->id, block };
1456 
1457 	r = dm_btree_lookup_next(&pmd->info, pmd->root, keys, vblock, &value);
1458 	if (!r)
1459 		unpack_lookup_result(td, value, result);
1460 
1461 	return r;
1462 }
1463 
1464 static int __find_mapped_range(struct dm_thin_device *td,
1465 			       dm_block_t begin, dm_block_t end,
1466 			       dm_block_t *thin_begin, dm_block_t *thin_end,
1467 			       dm_block_t *pool_begin, bool *maybe_shared)
1468 {
1469 	int r;
1470 	dm_block_t pool_end;
1471 	struct dm_thin_lookup_result lookup;
1472 
1473 	if (end < begin)
1474 		return -ENODATA;
1475 
1476 	r = __find_next_mapped_block(td, begin, &begin, &lookup);
1477 	if (r)
1478 		return r;
1479 
1480 	if (begin >= end)
1481 		return -ENODATA;
1482 
1483 	*thin_begin = begin;
1484 	*pool_begin = lookup.block;
1485 	*maybe_shared = lookup.shared;
1486 
1487 	begin++;
1488 	pool_end = *pool_begin + 1;
1489 	while (begin != end) {
1490 		r = __find_block(td, begin, true, &lookup);
1491 		if (r) {
1492 			if (r == -ENODATA)
1493 				break;
1494 			else
1495 				return r;
1496 		}
1497 
1498 		if ((lookup.block != pool_end) ||
1499 		    (lookup.shared != *maybe_shared))
1500 			break;
1501 
1502 		pool_end++;
1503 		begin++;
1504 	}
1505 
1506 	*thin_end = begin;
1507 	return 0;
1508 }
1509 
1510 int dm_thin_find_mapped_range(struct dm_thin_device *td,
1511 			      dm_block_t begin, dm_block_t end,
1512 			      dm_block_t *thin_begin, dm_block_t *thin_end,
1513 			      dm_block_t *pool_begin, bool *maybe_shared)
1514 {
1515 	int r = -EINVAL;
1516 	struct dm_pool_metadata *pmd = td->pmd;
1517 
1518 	down_read(&pmd->root_lock);
1519 	if (!pmd->fail_io) {
1520 		r = __find_mapped_range(td, begin, end, thin_begin, thin_end,
1521 					pool_begin, maybe_shared);
1522 	}
1523 	up_read(&pmd->root_lock);
1524 
1525 	return r;
1526 }
1527 
1528 static int __insert(struct dm_thin_device *td, dm_block_t block,
1529 		    dm_block_t data_block)
1530 {
1531 	int r, inserted;
1532 	__le64 value;
1533 	struct dm_pool_metadata *pmd = td->pmd;
1534 	dm_block_t keys[2] = { td->id, block };
1535 
1536 	value = cpu_to_le64(pack_block_time(data_block, pmd->time));
1537 	__dm_bless_for_disk(&value);
1538 
1539 	r = dm_btree_insert_notify(&pmd->info, pmd->root, keys, &value,
1540 				   &pmd->root, &inserted);
1541 	if (r)
1542 		return r;
1543 
1544 	td->changed = 1;
1545 	if (inserted)
1546 		td->mapped_blocks++;
1547 
1548 	return 0;
1549 }
1550 
1551 int dm_thin_insert_block(struct dm_thin_device *td, dm_block_t block,
1552 			 dm_block_t data_block)
1553 {
1554 	int r = -EINVAL;
1555 
1556 	down_write(&td->pmd->root_lock);
1557 	if (!td->pmd->fail_io)
1558 		r = __insert(td, block, data_block);
1559 	up_write(&td->pmd->root_lock);
1560 
1561 	return r;
1562 }
1563 
1564 static int __remove(struct dm_thin_device *td, dm_block_t block)
1565 {
1566 	int r;
1567 	struct dm_pool_metadata *pmd = td->pmd;
1568 	dm_block_t keys[2] = { td->id, block };
1569 
1570 	r = dm_btree_remove(&pmd->info, pmd->root, keys, &pmd->root);
1571 	if (r)
1572 		return r;
1573 
1574 	td->mapped_blocks--;
1575 	td->changed = 1;
1576 
1577 	return 0;
1578 }
1579 
1580 static int __remove_range(struct dm_thin_device *td, dm_block_t begin, dm_block_t end)
1581 {
1582 	int r;
1583 	unsigned count, total_count = 0;
1584 	struct dm_pool_metadata *pmd = td->pmd;
1585 	dm_block_t keys[1] = { td->id };
1586 	__le64 value;
1587 	dm_block_t mapping_root;
1588 
1589 	/*
1590 	 * Find the mapping tree
1591 	 */
1592 	r = dm_btree_lookup(&pmd->tl_info, pmd->root, keys, &value);
1593 	if (r)
1594 		return r;
1595 
1596 	/*
1597 	 * Remove from the mapping tree, taking care to inc the
1598 	 * ref count so it doesn't get deleted.
1599 	 */
1600 	mapping_root = le64_to_cpu(value);
1601 	dm_tm_inc(pmd->tm, mapping_root);
1602 	r = dm_btree_remove(&pmd->tl_info, pmd->root, keys, &pmd->root);
1603 	if (r)
1604 		return r;
1605 
1606 	/*
1607 	 * Remove leaves stops at the first unmapped entry, so we have to
1608 	 * loop round finding mapped ranges.
1609 	 */
1610 	while (begin < end) {
1611 		r = dm_btree_lookup_next(&pmd->bl_info, mapping_root, &begin, &begin, &value);
1612 		if (r == -ENODATA)
1613 			break;
1614 
1615 		if (r)
1616 			return r;
1617 
1618 		if (begin >= end)
1619 			break;
1620 
1621 		r = dm_btree_remove_leaves(&pmd->bl_info, mapping_root, &begin, end, &mapping_root, &count);
1622 		if (r)
1623 			return r;
1624 
1625 		total_count += count;
1626 	}
1627 
1628 	td->mapped_blocks -= total_count;
1629 	td->changed = 1;
1630 
1631 	/*
1632 	 * Reinsert the mapping tree.
1633 	 */
1634 	value = cpu_to_le64(mapping_root);
1635 	__dm_bless_for_disk(&value);
1636 	return dm_btree_insert(&pmd->tl_info, pmd->root, keys, &value, &pmd->root);
1637 }
1638 
1639 int dm_thin_remove_block(struct dm_thin_device *td, dm_block_t block)
1640 {
1641 	int r = -EINVAL;
1642 
1643 	down_write(&td->pmd->root_lock);
1644 	if (!td->pmd->fail_io)
1645 		r = __remove(td, block);
1646 	up_write(&td->pmd->root_lock);
1647 
1648 	return r;
1649 }
1650 
1651 int dm_thin_remove_range(struct dm_thin_device *td,
1652 			 dm_block_t begin, dm_block_t end)
1653 {
1654 	int r = -EINVAL;
1655 
1656 	down_write(&td->pmd->root_lock);
1657 	if (!td->pmd->fail_io)
1658 		r = __remove_range(td, begin, end);
1659 	up_write(&td->pmd->root_lock);
1660 
1661 	return r;
1662 }
1663 
1664 int dm_pool_block_is_used(struct dm_pool_metadata *pmd, dm_block_t b, bool *result)
1665 {
1666 	int r;
1667 	uint32_t ref_count;
1668 
1669 	down_read(&pmd->root_lock);
1670 	r = dm_sm_get_count(pmd->data_sm, b, &ref_count);
1671 	if (!r)
1672 		*result = (ref_count != 0);
1673 	up_read(&pmd->root_lock);
1674 
1675 	return r;
1676 }
1677 
1678 int dm_pool_inc_data_range(struct dm_pool_metadata *pmd, dm_block_t b, dm_block_t e)
1679 {
1680 	int r = 0;
1681 
1682 	down_write(&pmd->root_lock);
1683 	for (; b != e; b++) {
1684 		r = dm_sm_inc_block(pmd->data_sm, b);
1685 		if (r)
1686 			break;
1687 	}
1688 	up_write(&pmd->root_lock);
1689 
1690 	return r;
1691 }
1692 
1693 int dm_pool_dec_data_range(struct dm_pool_metadata *pmd, dm_block_t b, dm_block_t e)
1694 {
1695 	int r = 0;
1696 
1697 	down_write(&pmd->root_lock);
1698 	for (; b != e; b++) {
1699 		r = dm_sm_dec_block(pmd->data_sm, b);
1700 		if (r)
1701 			break;
1702 	}
1703 	up_write(&pmd->root_lock);
1704 
1705 	return r;
1706 }
1707 
1708 bool dm_thin_changed_this_transaction(struct dm_thin_device *td)
1709 {
1710 	int r;
1711 
1712 	down_read(&td->pmd->root_lock);
1713 	r = td->changed;
1714 	up_read(&td->pmd->root_lock);
1715 
1716 	return r;
1717 }
1718 
1719 bool dm_pool_changed_this_transaction(struct dm_pool_metadata *pmd)
1720 {
1721 	bool r = false;
1722 	struct dm_thin_device *td, *tmp;
1723 
1724 	down_read(&pmd->root_lock);
1725 	list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
1726 		if (td->changed) {
1727 			r = td->changed;
1728 			break;
1729 		}
1730 	}
1731 	up_read(&pmd->root_lock);
1732 
1733 	return r;
1734 }
1735 
1736 bool dm_thin_aborted_changes(struct dm_thin_device *td)
1737 {
1738 	bool r;
1739 
1740 	down_read(&td->pmd->root_lock);
1741 	r = td->aborted_with_changes;
1742 	up_read(&td->pmd->root_lock);
1743 
1744 	return r;
1745 }
1746 
1747 int dm_pool_alloc_data_block(struct dm_pool_metadata *pmd, dm_block_t *result)
1748 {
1749 	int r = -EINVAL;
1750 
1751 	down_write(&pmd->root_lock);
1752 	if (!pmd->fail_io)
1753 		r = dm_sm_new_block(pmd->data_sm, result);
1754 	up_write(&pmd->root_lock);
1755 
1756 	return r;
1757 }
1758 
1759 int dm_pool_commit_metadata(struct dm_pool_metadata *pmd)
1760 {
1761 	int r = -EINVAL;
1762 
1763 	down_write(&pmd->root_lock);
1764 	if (pmd->fail_io)
1765 		goto out;
1766 
1767 	r = __commit_transaction(pmd);
1768 	if (r <= 0)
1769 		goto out;
1770 
1771 	/*
1772 	 * Open the next transaction.
1773 	 */
1774 	r = __begin_transaction(pmd);
1775 out:
1776 	up_write(&pmd->root_lock);
1777 	return r;
1778 }
1779 
1780 static void __set_abort_with_changes_flags(struct dm_pool_metadata *pmd)
1781 {
1782 	struct dm_thin_device *td;
1783 
1784 	list_for_each_entry(td, &pmd->thin_devices, list)
1785 		td->aborted_with_changes = td->changed;
1786 }
1787 
1788 int dm_pool_abort_metadata(struct dm_pool_metadata *pmd)
1789 {
1790 	int r = -EINVAL;
1791 
1792 	down_write(&pmd->root_lock);
1793 	if (pmd->fail_io)
1794 		goto out;
1795 
1796 	__set_abort_with_changes_flags(pmd);
1797 	__destroy_persistent_data_objects(pmd);
1798 	r = __create_persistent_data_objects(pmd, false);
1799 	if (r)
1800 		pmd->fail_io = true;
1801 
1802 out:
1803 	up_write(&pmd->root_lock);
1804 
1805 	return r;
1806 }
1807 
1808 int dm_pool_get_free_block_count(struct dm_pool_metadata *pmd, dm_block_t *result)
1809 {
1810 	int r = -EINVAL;
1811 
1812 	down_read(&pmd->root_lock);
1813 	if (!pmd->fail_io)
1814 		r = dm_sm_get_nr_free(pmd->data_sm, result);
1815 	up_read(&pmd->root_lock);
1816 
1817 	return r;
1818 }
1819 
1820 int dm_pool_get_free_metadata_block_count(struct dm_pool_metadata *pmd,
1821 					  dm_block_t *result)
1822 {
1823 	int r = -EINVAL;
1824 
1825 	down_read(&pmd->root_lock);
1826 	if (!pmd->fail_io)
1827 		r = dm_sm_get_nr_free(pmd->metadata_sm, result);
1828 	up_read(&pmd->root_lock);
1829 
1830 	return r;
1831 }
1832 
1833 int dm_pool_get_metadata_dev_size(struct dm_pool_metadata *pmd,
1834 				  dm_block_t *result)
1835 {
1836 	int r = -EINVAL;
1837 
1838 	down_read(&pmd->root_lock);
1839 	if (!pmd->fail_io)
1840 		r = dm_sm_get_nr_blocks(pmd->metadata_sm, result);
1841 	up_read(&pmd->root_lock);
1842 
1843 	return r;
1844 }
1845 
1846 int dm_pool_get_data_dev_size(struct dm_pool_metadata *pmd, dm_block_t *result)
1847 {
1848 	int r = -EINVAL;
1849 
1850 	down_read(&pmd->root_lock);
1851 	if (!pmd->fail_io)
1852 		r = dm_sm_get_nr_blocks(pmd->data_sm, result);
1853 	up_read(&pmd->root_lock);
1854 
1855 	return r;
1856 }
1857 
1858 int dm_thin_get_mapped_count(struct dm_thin_device *td, dm_block_t *result)
1859 {
1860 	int r = -EINVAL;
1861 	struct dm_pool_metadata *pmd = td->pmd;
1862 
1863 	down_read(&pmd->root_lock);
1864 	if (!pmd->fail_io) {
1865 		*result = td->mapped_blocks;
1866 		r = 0;
1867 	}
1868 	up_read(&pmd->root_lock);
1869 
1870 	return r;
1871 }
1872 
1873 static int __highest_block(struct dm_thin_device *td, dm_block_t *result)
1874 {
1875 	int r;
1876 	__le64 value_le;
1877 	dm_block_t thin_root;
1878 	struct dm_pool_metadata *pmd = td->pmd;
1879 
1880 	r = dm_btree_lookup(&pmd->tl_info, pmd->root, &td->id, &value_le);
1881 	if (r)
1882 		return r;
1883 
1884 	thin_root = le64_to_cpu(value_le);
1885 
1886 	return dm_btree_find_highest_key(&pmd->bl_info, thin_root, result);
1887 }
1888 
1889 int dm_thin_get_highest_mapped_block(struct dm_thin_device *td,
1890 				     dm_block_t *result)
1891 {
1892 	int r = -EINVAL;
1893 	struct dm_pool_metadata *pmd = td->pmd;
1894 
1895 	down_read(&pmd->root_lock);
1896 	if (!pmd->fail_io)
1897 		r = __highest_block(td, result);
1898 	up_read(&pmd->root_lock);
1899 
1900 	return r;
1901 }
1902 
1903 static int __resize_space_map(struct dm_space_map *sm, dm_block_t new_count)
1904 {
1905 	int r;
1906 	dm_block_t old_count;
1907 
1908 	r = dm_sm_get_nr_blocks(sm, &old_count);
1909 	if (r)
1910 		return r;
1911 
1912 	if (new_count == old_count)
1913 		return 0;
1914 
1915 	if (new_count < old_count) {
1916 		DMERR("cannot reduce size of space map");
1917 		return -EINVAL;
1918 	}
1919 
1920 	return dm_sm_extend(sm, new_count - old_count);
1921 }
1922 
1923 int dm_pool_resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1924 {
1925 	int r = -EINVAL;
1926 
1927 	down_write(&pmd->root_lock);
1928 	if (!pmd->fail_io)
1929 		r = __resize_space_map(pmd->data_sm, new_count);
1930 	up_write(&pmd->root_lock);
1931 
1932 	return r;
1933 }
1934 
1935 int dm_pool_resize_metadata_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1936 {
1937 	int r = -EINVAL;
1938 
1939 	down_write(&pmd->root_lock);
1940 	if (!pmd->fail_io)
1941 		r = __resize_space_map(pmd->metadata_sm, new_count);
1942 	up_write(&pmd->root_lock);
1943 
1944 	return r;
1945 }
1946 
1947 void dm_pool_metadata_read_only(struct dm_pool_metadata *pmd)
1948 {
1949 	down_write(&pmd->root_lock);
1950 	dm_bm_set_read_only(pmd->bm);
1951 	up_write(&pmd->root_lock);
1952 }
1953 
1954 void dm_pool_metadata_read_write(struct dm_pool_metadata *pmd)
1955 {
1956 	down_write(&pmd->root_lock);
1957 	dm_bm_set_read_write(pmd->bm);
1958 	up_write(&pmd->root_lock);
1959 }
1960 
1961 int dm_pool_register_metadata_threshold(struct dm_pool_metadata *pmd,
1962 					dm_block_t threshold,
1963 					dm_sm_threshold_fn fn,
1964 					void *context)
1965 {
1966 	int r;
1967 
1968 	down_write(&pmd->root_lock);
1969 	r = dm_sm_register_threshold_callback(pmd->metadata_sm, threshold, fn, context);
1970 	up_write(&pmd->root_lock);
1971 
1972 	return r;
1973 }
1974 
1975 int dm_pool_metadata_set_needs_check(struct dm_pool_metadata *pmd)
1976 {
1977 	int r;
1978 	struct dm_block *sblock;
1979 	struct thin_disk_superblock *disk_super;
1980 
1981 	down_write(&pmd->root_lock);
1982 	pmd->flags |= THIN_METADATA_NEEDS_CHECK_FLAG;
1983 
1984 	r = superblock_lock(pmd, &sblock);
1985 	if (r) {
1986 		DMERR("couldn't read superblock");
1987 		goto out;
1988 	}
1989 
1990 	disk_super = dm_block_data(sblock);
1991 	disk_super->flags = cpu_to_le32(pmd->flags);
1992 
1993 	dm_bm_unlock(sblock);
1994 out:
1995 	up_write(&pmd->root_lock);
1996 	return r;
1997 }
1998 
1999 bool dm_pool_metadata_needs_check(struct dm_pool_metadata *pmd)
2000 {
2001 	bool needs_check;
2002 
2003 	down_read(&pmd->root_lock);
2004 	needs_check = pmd->flags & THIN_METADATA_NEEDS_CHECK_FLAG;
2005 	up_read(&pmd->root_lock);
2006 
2007 	return needs_check;
2008 }
2009 
2010 void dm_pool_issue_prefetches(struct dm_pool_metadata *pmd)
2011 {
2012 	down_read(&pmd->root_lock);
2013 	if (!pmd->fail_io)
2014 		dm_tm_issue_prefetches(pmd->tm);
2015 	up_read(&pmd->root_lock);
2016 }
2017