xref: /linux/fs/btrfs/disk-io.c (revision d786bdf2a70545a868cd0b06b5603cd5a5fec011)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5 
6 #include <linux/fs.h>
7 #include <linux/blkdev.h>
8 #include <linux/writeback.h>
9 #include <linux/workqueue.h>
10 #include <linux/kthread.h>
11 #include <linux/slab.h>
12 #include <linux/migrate.h>
13 #include <linux/ratelimit.h>
14 #include <linux/uuid.h>
15 #include <linux/semaphore.h>
16 #include <linux/error-injection.h>
17 #include <linux/crc32c.h>
18 #include <linux/sched/mm.h>
19 #include <asm/unaligned.h>
20 #include <crypto/hash.h>
21 #include "ctree.h"
22 #include "disk-io.h"
23 #include "transaction.h"
24 #include "btrfs_inode.h"
25 #include "volumes.h"
26 #include "print-tree.h"
27 #include "locking.h"
28 #include "tree-log.h"
29 #include "free-space-cache.h"
30 #include "free-space-tree.h"
31 #include "check-integrity.h"
32 #include "rcu-string.h"
33 #include "dev-replace.h"
34 #include "raid56.h"
35 #include "sysfs.h"
36 #include "qgroup.h"
37 #include "compression.h"
38 #include "tree-checker.h"
39 #include "ref-verify.h"
40 #include "block-group.h"
41 #include "discard.h"
42 #include "space-info.h"
43 #include "zoned.h"
44 #include "subpage.h"
45 
46 #define BTRFS_SUPER_FLAG_SUPP	(BTRFS_HEADER_FLAG_WRITTEN |\
47 				 BTRFS_HEADER_FLAG_RELOC |\
48 				 BTRFS_SUPER_FLAG_ERROR |\
49 				 BTRFS_SUPER_FLAG_SEEDING |\
50 				 BTRFS_SUPER_FLAG_METADUMP |\
51 				 BTRFS_SUPER_FLAG_METADUMP_V2)
52 
53 static void end_workqueue_fn(struct btrfs_work *work);
54 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
55 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56 				      struct btrfs_fs_info *fs_info);
57 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
59 					struct extent_io_tree *dirty_pages,
60 					int mark);
61 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
62 				       struct extent_io_tree *pinned_extents);
63 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
64 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
65 
66 /*
67  * btrfs_end_io_wq structs are used to do processing in task context when an IO
68  * is complete.  This is used during reads to verify checksums, and it is used
69  * by writes to insert metadata for new file extents after IO is complete.
70  */
71 struct btrfs_end_io_wq {
72 	struct bio *bio;
73 	bio_end_io_t *end_io;
74 	void *private;
75 	struct btrfs_fs_info *info;
76 	blk_status_t status;
77 	enum btrfs_wq_endio_type metadata;
78 	struct btrfs_work work;
79 };
80 
81 static struct kmem_cache *btrfs_end_io_wq_cache;
82 
83 int __init btrfs_end_io_wq_init(void)
84 {
85 	btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
86 					sizeof(struct btrfs_end_io_wq),
87 					0,
88 					SLAB_MEM_SPREAD,
89 					NULL);
90 	if (!btrfs_end_io_wq_cache)
91 		return -ENOMEM;
92 	return 0;
93 }
94 
95 void __cold btrfs_end_io_wq_exit(void)
96 {
97 	kmem_cache_destroy(btrfs_end_io_wq_cache);
98 }
99 
100 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
101 {
102 	if (fs_info->csum_shash)
103 		crypto_free_shash(fs_info->csum_shash);
104 }
105 
106 /*
107  * async submit bios are used to offload expensive checksumming
108  * onto the worker threads.  They checksum file and metadata bios
109  * just before they are sent down the IO stack.
110  */
111 struct async_submit_bio {
112 	struct inode *inode;
113 	struct bio *bio;
114 	extent_submit_bio_start_t *submit_bio_start;
115 	int mirror_num;
116 
117 	/* Optional parameter for submit_bio_start used by direct io */
118 	u64 dio_file_offset;
119 	struct btrfs_work work;
120 	blk_status_t status;
121 };
122 
123 /*
124  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
125  * eb, the lockdep key is determined by the btrfs_root it belongs to and
126  * the level the eb occupies in the tree.
127  *
128  * Different roots are used for different purposes and may nest inside each
129  * other and they require separate keysets.  As lockdep keys should be
130  * static, assign keysets according to the purpose of the root as indicated
131  * by btrfs_root->root_key.objectid.  This ensures that all special purpose
132  * roots have separate keysets.
133  *
134  * Lock-nesting across peer nodes is always done with the immediate parent
135  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
136  * subclass to avoid triggering lockdep warning in such cases.
137  *
138  * The key is set by the readpage_end_io_hook after the buffer has passed
139  * csum validation but before the pages are unlocked.  It is also set by
140  * btrfs_init_new_buffer on freshly allocated blocks.
141  *
142  * We also add a check to make sure the highest level of the tree is the
143  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
144  * needs update as well.
145  */
146 #ifdef CONFIG_DEBUG_LOCK_ALLOC
147 # if BTRFS_MAX_LEVEL != 8
148 #  error
149 # endif
150 
151 #define DEFINE_LEVEL(stem, level)					\
152 	.names[level] = "btrfs-" stem "-0" #level,
153 
154 #define DEFINE_NAME(stem)						\
155 	DEFINE_LEVEL(stem, 0)						\
156 	DEFINE_LEVEL(stem, 1)						\
157 	DEFINE_LEVEL(stem, 2)						\
158 	DEFINE_LEVEL(stem, 3)						\
159 	DEFINE_LEVEL(stem, 4)						\
160 	DEFINE_LEVEL(stem, 5)						\
161 	DEFINE_LEVEL(stem, 6)						\
162 	DEFINE_LEVEL(stem, 7)
163 
164 static struct btrfs_lockdep_keyset {
165 	u64			id;		/* root objectid */
166 	/* Longest entry: btrfs-free-space-00 */
167 	char			names[BTRFS_MAX_LEVEL][20];
168 	struct lock_class_key	keys[BTRFS_MAX_LEVEL];
169 } btrfs_lockdep_keysets[] = {
170 	{ .id = BTRFS_ROOT_TREE_OBJECTID,	DEFINE_NAME("root")	},
171 	{ .id = BTRFS_EXTENT_TREE_OBJECTID,	DEFINE_NAME("extent")	},
172 	{ .id = BTRFS_CHUNK_TREE_OBJECTID,	DEFINE_NAME("chunk")	},
173 	{ .id = BTRFS_DEV_TREE_OBJECTID,	DEFINE_NAME("dev")	},
174 	{ .id = BTRFS_CSUM_TREE_OBJECTID,	DEFINE_NAME("csum")	},
175 	{ .id = BTRFS_QUOTA_TREE_OBJECTID,	DEFINE_NAME("quota")	},
176 	{ .id = BTRFS_TREE_LOG_OBJECTID,	DEFINE_NAME("log")	},
177 	{ .id = BTRFS_TREE_RELOC_OBJECTID,	DEFINE_NAME("treloc")	},
178 	{ .id = BTRFS_DATA_RELOC_TREE_OBJECTID,	DEFINE_NAME("dreloc")	},
179 	{ .id = BTRFS_UUID_TREE_OBJECTID,	DEFINE_NAME("uuid")	},
180 	{ .id = BTRFS_FREE_SPACE_TREE_OBJECTID,	DEFINE_NAME("free-space") },
181 	{ .id = 0,				DEFINE_NAME("tree")	},
182 };
183 
184 #undef DEFINE_LEVEL
185 #undef DEFINE_NAME
186 
187 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
188 				    int level)
189 {
190 	struct btrfs_lockdep_keyset *ks;
191 
192 	BUG_ON(level >= ARRAY_SIZE(ks->keys));
193 
194 	/* find the matching keyset, id 0 is the default entry */
195 	for (ks = btrfs_lockdep_keysets; ks->id; ks++)
196 		if (ks->id == objectid)
197 			break;
198 
199 	lockdep_set_class_and_name(&eb->lock,
200 				   &ks->keys[level], ks->names[level]);
201 }
202 
203 #endif
204 
205 /*
206  * Compute the csum of a btree block and store the result to provided buffer.
207  */
208 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
209 {
210 	struct btrfs_fs_info *fs_info = buf->fs_info;
211 	const int num_pages = num_extent_pages(buf);
212 	const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
213 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
214 	char *kaddr;
215 	int i;
216 
217 	shash->tfm = fs_info->csum_shash;
218 	crypto_shash_init(shash);
219 	kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
220 	crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
221 			    first_page_part - BTRFS_CSUM_SIZE);
222 
223 	for (i = 1; i < num_pages; i++) {
224 		kaddr = page_address(buf->pages[i]);
225 		crypto_shash_update(shash, kaddr, PAGE_SIZE);
226 	}
227 	memset(result, 0, BTRFS_CSUM_SIZE);
228 	crypto_shash_final(shash, result);
229 }
230 
231 /*
232  * we can't consider a given block up to date unless the transid of the
233  * block matches the transid in the parent node's pointer.  This is how we
234  * detect blocks that either didn't get written at all or got written
235  * in the wrong place.
236  */
237 static int verify_parent_transid(struct extent_io_tree *io_tree,
238 				 struct extent_buffer *eb, u64 parent_transid,
239 				 int atomic)
240 {
241 	struct extent_state *cached_state = NULL;
242 	int ret;
243 
244 	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
245 		return 0;
246 
247 	if (atomic)
248 		return -EAGAIN;
249 
250 	lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
251 			 &cached_state);
252 	if (extent_buffer_uptodate(eb) &&
253 	    btrfs_header_generation(eb) == parent_transid) {
254 		ret = 0;
255 		goto out;
256 	}
257 	btrfs_err_rl(eb->fs_info,
258 		"parent transid verify failed on %llu wanted %llu found %llu",
259 			eb->start,
260 			parent_transid, btrfs_header_generation(eb));
261 	ret = 1;
262 	clear_extent_buffer_uptodate(eb);
263 out:
264 	unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
265 			     &cached_state);
266 	return ret;
267 }
268 
269 static bool btrfs_supported_super_csum(u16 csum_type)
270 {
271 	switch (csum_type) {
272 	case BTRFS_CSUM_TYPE_CRC32:
273 	case BTRFS_CSUM_TYPE_XXHASH:
274 	case BTRFS_CSUM_TYPE_SHA256:
275 	case BTRFS_CSUM_TYPE_BLAKE2:
276 		return true;
277 	default:
278 		return false;
279 	}
280 }
281 
282 /*
283  * Return 0 if the superblock checksum type matches the checksum value of that
284  * algorithm. Pass the raw disk superblock data.
285  */
286 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
287 				  char *raw_disk_sb)
288 {
289 	struct btrfs_super_block *disk_sb =
290 		(struct btrfs_super_block *)raw_disk_sb;
291 	char result[BTRFS_CSUM_SIZE];
292 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
293 
294 	shash->tfm = fs_info->csum_shash;
295 
296 	/*
297 	 * The super_block structure does not span the whole
298 	 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
299 	 * filled with zeros and is included in the checksum.
300 	 */
301 	crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
302 			    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
303 
304 	if (memcmp(disk_sb->csum, result, fs_info->csum_size))
305 		return 1;
306 
307 	return 0;
308 }
309 
310 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
311 			   struct btrfs_key *first_key, u64 parent_transid)
312 {
313 	struct btrfs_fs_info *fs_info = eb->fs_info;
314 	int found_level;
315 	struct btrfs_key found_key;
316 	int ret;
317 
318 	found_level = btrfs_header_level(eb);
319 	if (found_level != level) {
320 		WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
321 		     KERN_ERR "BTRFS: tree level check failed\n");
322 		btrfs_err(fs_info,
323 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
324 			  eb->start, level, found_level);
325 		return -EIO;
326 	}
327 
328 	if (!first_key)
329 		return 0;
330 
331 	/*
332 	 * For live tree block (new tree blocks in current transaction),
333 	 * we need proper lock context to avoid race, which is impossible here.
334 	 * So we only checks tree blocks which is read from disk, whose
335 	 * generation <= fs_info->last_trans_committed.
336 	 */
337 	if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
338 		return 0;
339 
340 	/* We have @first_key, so this @eb must have at least one item */
341 	if (btrfs_header_nritems(eb) == 0) {
342 		btrfs_err(fs_info,
343 		"invalid tree nritems, bytenr=%llu nritems=0 expect >0",
344 			  eb->start);
345 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
346 		return -EUCLEAN;
347 	}
348 
349 	if (found_level)
350 		btrfs_node_key_to_cpu(eb, &found_key, 0);
351 	else
352 		btrfs_item_key_to_cpu(eb, &found_key, 0);
353 	ret = btrfs_comp_cpu_keys(first_key, &found_key);
354 
355 	if (ret) {
356 		WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
357 		     KERN_ERR "BTRFS: tree first key check failed\n");
358 		btrfs_err(fs_info,
359 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
360 			  eb->start, parent_transid, first_key->objectid,
361 			  first_key->type, first_key->offset,
362 			  found_key.objectid, found_key.type,
363 			  found_key.offset);
364 	}
365 	return ret;
366 }
367 
368 /*
369  * helper to read a given tree block, doing retries as required when
370  * the checksums don't match and we have alternate mirrors to try.
371  *
372  * @parent_transid:	expected transid, skip check if 0
373  * @level:		expected level, mandatory check
374  * @first_key:		expected key of first slot, skip check if NULL
375  */
376 int btrfs_read_extent_buffer(struct extent_buffer *eb,
377 			     u64 parent_transid, int level,
378 			     struct btrfs_key *first_key)
379 {
380 	struct btrfs_fs_info *fs_info = eb->fs_info;
381 	struct extent_io_tree *io_tree;
382 	int failed = 0;
383 	int ret;
384 	int num_copies = 0;
385 	int mirror_num = 0;
386 	int failed_mirror = 0;
387 
388 	io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
389 	while (1) {
390 		clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
391 		ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
392 		if (!ret) {
393 			if (verify_parent_transid(io_tree, eb,
394 						   parent_transid, 0))
395 				ret = -EIO;
396 			else if (btrfs_verify_level_key(eb, level,
397 						first_key, parent_transid))
398 				ret = -EUCLEAN;
399 			else
400 				break;
401 		}
402 
403 		num_copies = btrfs_num_copies(fs_info,
404 					      eb->start, eb->len);
405 		if (num_copies == 1)
406 			break;
407 
408 		if (!failed_mirror) {
409 			failed = 1;
410 			failed_mirror = eb->read_mirror;
411 		}
412 
413 		mirror_num++;
414 		if (mirror_num == failed_mirror)
415 			mirror_num++;
416 
417 		if (mirror_num > num_copies)
418 			break;
419 	}
420 
421 	if (failed && !ret && failed_mirror)
422 		btrfs_repair_eb_io_failure(eb, failed_mirror);
423 
424 	return ret;
425 }
426 
427 static int csum_one_extent_buffer(struct extent_buffer *eb)
428 {
429 	struct btrfs_fs_info *fs_info = eb->fs_info;
430 	u8 result[BTRFS_CSUM_SIZE];
431 	int ret;
432 
433 	ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
434 				    offsetof(struct btrfs_header, fsid),
435 				    BTRFS_FSID_SIZE) == 0);
436 	csum_tree_block(eb, result);
437 
438 	if (btrfs_header_level(eb))
439 		ret = btrfs_check_node(eb);
440 	else
441 		ret = btrfs_check_leaf_full(eb);
442 
443 	if (ret < 0)
444 		goto error;
445 
446 	/*
447 	 * Also check the generation, the eb reached here must be newer than
448 	 * last committed. Or something seriously wrong happened.
449 	 */
450 	if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) {
451 		ret = -EUCLEAN;
452 		btrfs_err(fs_info,
453 			"block=%llu bad generation, have %llu expect > %llu",
454 			  eb->start, btrfs_header_generation(eb),
455 			  fs_info->last_trans_committed);
456 		goto error;
457 	}
458 	write_extent_buffer(eb, result, 0, fs_info->csum_size);
459 
460 	return 0;
461 
462 error:
463 	btrfs_print_tree(eb, 0);
464 	btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
465 		  eb->start);
466 	WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
467 	return ret;
468 }
469 
470 /* Checksum all dirty extent buffers in one bio_vec */
471 static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
472 				      struct bio_vec *bvec)
473 {
474 	struct page *page = bvec->bv_page;
475 	u64 bvec_start = page_offset(page) + bvec->bv_offset;
476 	u64 cur;
477 	int ret = 0;
478 
479 	for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
480 	     cur += fs_info->nodesize) {
481 		struct extent_buffer *eb;
482 		bool uptodate;
483 
484 		eb = find_extent_buffer(fs_info, cur);
485 		uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
486 						       fs_info->nodesize);
487 
488 		/* A dirty eb shouldn't disappear from extent_buffers */
489 		if (WARN_ON(!eb))
490 			return -EUCLEAN;
491 
492 		if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
493 			free_extent_buffer(eb);
494 			return -EUCLEAN;
495 		}
496 		if (WARN_ON(!uptodate)) {
497 			free_extent_buffer(eb);
498 			return -EUCLEAN;
499 		}
500 
501 		ret = csum_one_extent_buffer(eb);
502 		free_extent_buffer(eb);
503 		if (ret < 0)
504 			return ret;
505 	}
506 	return ret;
507 }
508 
509 /*
510  * Checksum a dirty tree block before IO.  This has extra checks to make sure
511  * we only fill in the checksum field in the first page of a multi-page block.
512  * For subpage extent buffers we need bvec to also read the offset in the page.
513  */
514 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
515 {
516 	struct page *page = bvec->bv_page;
517 	u64 start = page_offset(page);
518 	u64 found_start;
519 	struct extent_buffer *eb;
520 
521 	if (fs_info->nodesize < PAGE_SIZE)
522 		return csum_dirty_subpage_buffers(fs_info, bvec);
523 
524 	eb = (struct extent_buffer *)page->private;
525 	if (page != eb->pages[0])
526 		return 0;
527 
528 	found_start = btrfs_header_bytenr(eb);
529 
530 	if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
531 		WARN_ON(found_start != 0);
532 		return 0;
533 	}
534 
535 	/*
536 	 * Please do not consolidate these warnings into a single if.
537 	 * It is useful to know what went wrong.
538 	 */
539 	if (WARN_ON(found_start != start))
540 		return -EUCLEAN;
541 	if (WARN_ON(!PageUptodate(page)))
542 		return -EUCLEAN;
543 
544 	return csum_one_extent_buffer(eb);
545 }
546 
547 static int check_tree_block_fsid(struct extent_buffer *eb)
548 {
549 	struct btrfs_fs_info *fs_info = eb->fs_info;
550 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
551 	u8 fsid[BTRFS_FSID_SIZE];
552 	u8 *metadata_uuid;
553 
554 	read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
555 			   BTRFS_FSID_SIZE);
556 	/*
557 	 * Checking the incompat flag is only valid for the current fs. For
558 	 * seed devices it's forbidden to have their uuid changed so reading
559 	 * ->fsid in this case is fine
560 	 */
561 	if (btrfs_fs_incompat(fs_info, METADATA_UUID))
562 		metadata_uuid = fs_devices->metadata_uuid;
563 	else
564 		metadata_uuid = fs_devices->fsid;
565 
566 	if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
567 		return 0;
568 
569 	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
570 		if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
571 			return 0;
572 
573 	return 1;
574 }
575 
576 /* Do basic extent buffer checks at read time */
577 static int validate_extent_buffer(struct extent_buffer *eb)
578 {
579 	struct btrfs_fs_info *fs_info = eb->fs_info;
580 	u64 found_start;
581 	const u32 csum_size = fs_info->csum_size;
582 	u8 found_level;
583 	u8 result[BTRFS_CSUM_SIZE];
584 	const u8 *header_csum;
585 	int ret = 0;
586 
587 	found_start = btrfs_header_bytenr(eb);
588 	if (found_start != eb->start) {
589 		btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
590 			     eb->start, found_start);
591 		ret = -EIO;
592 		goto out;
593 	}
594 	if (check_tree_block_fsid(eb)) {
595 		btrfs_err_rl(fs_info, "bad fsid on block %llu",
596 			     eb->start);
597 		ret = -EIO;
598 		goto out;
599 	}
600 	found_level = btrfs_header_level(eb);
601 	if (found_level >= BTRFS_MAX_LEVEL) {
602 		btrfs_err(fs_info, "bad tree block level %d on %llu",
603 			  (int)btrfs_header_level(eb), eb->start);
604 		ret = -EIO;
605 		goto out;
606 	}
607 
608 	csum_tree_block(eb, result);
609 	header_csum = page_address(eb->pages[0]) +
610 		get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
611 
612 	if (memcmp(result, header_csum, csum_size) != 0) {
613 		btrfs_warn_rl(fs_info,
614 	"checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
615 			      eb->start,
616 			      CSUM_FMT_VALUE(csum_size, header_csum),
617 			      CSUM_FMT_VALUE(csum_size, result),
618 			      btrfs_header_level(eb));
619 		ret = -EUCLEAN;
620 		goto out;
621 	}
622 
623 	/*
624 	 * If this is a leaf block and it is corrupt, set the corrupt bit so
625 	 * that we don't try and read the other copies of this block, just
626 	 * return -EIO.
627 	 */
628 	if (found_level == 0 && btrfs_check_leaf_full(eb)) {
629 		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
630 		ret = -EIO;
631 	}
632 
633 	if (found_level > 0 && btrfs_check_node(eb))
634 		ret = -EIO;
635 
636 	if (!ret)
637 		set_extent_buffer_uptodate(eb);
638 	else
639 		btrfs_err(fs_info,
640 			  "block=%llu read time tree block corruption detected",
641 			  eb->start);
642 out:
643 	return ret;
644 }
645 
646 static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
647 				   int mirror)
648 {
649 	struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
650 	struct extent_buffer *eb;
651 	bool reads_done;
652 	int ret = 0;
653 
654 	/*
655 	 * We don't allow bio merge for subpage metadata read, so we should
656 	 * only get one eb for each endio hook.
657 	 */
658 	ASSERT(end == start + fs_info->nodesize - 1);
659 	ASSERT(PagePrivate(page));
660 
661 	eb = find_extent_buffer(fs_info, start);
662 	/*
663 	 * When we are reading one tree block, eb must have been inserted into
664 	 * the radix tree. If not, something is wrong.
665 	 */
666 	ASSERT(eb);
667 
668 	reads_done = atomic_dec_and_test(&eb->io_pages);
669 	/* Subpage read must finish in page read */
670 	ASSERT(reads_done);
671 
672 	eb->read_mirror = mirror;
673 	if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
674 		ret = -EIO;
675 		goto err;
676 	}
677 	ret = validate_extent_buffer(eb);
678 	if (ret < 0)
679 		goto err;
680 
681 	set_extent_buffer_uptodate(eb);
682 
683 	free_extent_buffer(eb);
684 	return ret;
685 err:
686 	/*
687 	 * end_bio_extent_readpage decrements io_pages in case of error,
688 	 * make sure it has something to decrement.
689 	 */
690 	atomic_inc(&eb->io_pages);
691 	clear_extent_buffer_uptodate(eb);
692 	free_extent_buffer(eb);
693 	return ret;
694 }
695 
696 int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio,
697 				   struct page *page, u64 start, u64 end,
698 				   int mirror)
699 {
700 	struct extent_buffer *eb;
701 	int ret = 0;
702 	int reads_done;
703 
704 	ASSERT(page->private);
705 
706 	if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
707 		return validate_subpage_buffer(page, start, end, mirror);
708 
709 	eb = (struct extent_buffer *)page->private;
710 
711 	/*
712 	 * The pending IO might have been the only thing that kept this buffer
713 	 * in memory.  Make sure we have a ref for all this other checks
714 	 */
715 	atomic_inc(&eb->refs);
716 
717 	reads_done = atomic_dec_and_test(&eb->io_pages);
718 	if (!reads_done)
719 		goto err;
720 
721 	eb->read_mirror = mirror;
722 	if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
723 		ret = -EIO;
724 		goto err;
725 	}
726 	ret = validate_extent_buffer(eb);
727 err:
728 	if (ret) {
729 		/*
730 		 * our io error hook is going to dec the io pages
731 		 * again, we have to make sure it has something
732 		 * to decrement
733 		 */
734 		atomic_inc(&eb->io_pages);
735 		clear_extent_buffer_uptodate(eb);
736 	}
737 	free_extent_buffer(eb);
738 
739 	return ret;
740 }
741 
742 static void end_workqueue_bio(struct bio *bio)
743 {
744 	struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
745 	struct btrfs_fs_info *fs_info;
746 	struct btrfs_workqueue *wq;
747 
748 	fs_info = end_io_wq->info;
749 	end_io_wq->status = bio->bi_status;
750 
751 	if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
752 		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
753 			wq = fs_info->endio_meta_write_workers;
754 		else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
755 			wq = fs_info->endio_freespace_worker;
756 		else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
757 			wq = fs_info->endio_raid56_workers;
758 		else
759 			wq = fs_info->endio_write_workers;
760 	} else {
761 		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
762 			wq = fs_info->endio_raid56_workers;
763 		else if (end_io_wq->metadata)
764 			wq = fs_info->endio_meta_workers;
765 		else
766 			wq = fs_info->endio_workers;
767 	}
768 
769 	btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
770 	btrfs_queue_work(wq, &end_io_wq->work);
771 }
772 
773 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
774 			enum btrfs_wq_endio_type metadata)
775 {
776 	struct btrfs_end_io_wq *end_io_wq;
777 
778 	end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
779 	if (!end_io_wq)
780 		return BLK_STS_RESOURCE;
781 
782 	end_io_wq->private = bio->bi_private;
783 	end_io_wq->end_io = bio->bi_end_io;
784 	end_io_wq->info = info;
785 	end_io_wq->status = 0;
786 	end_io_wq->bio = bio;
787 	end_io_wq->metadata = metadata;
788 
789 	bio->bi_private = end_io_wq;
790 	bio->bi_end_io = end_workqueue_bio;
791 	return 0;
792 }
793 
794 static void run_one_async_start(struct btrfs_work *work)
795 {
796 	struct async_submit_bio *async;
797 	blk_status_t ret;
798 
799 	async = container_of(work, struct  async_submit_bio, work);
800 	ret = async->submit_bio_start(async->inode, async->bio,
801 				      async->dio_file_offset);
802 	if (ret)
803 		async->status = ret;
804 }
805 
806 /*
807  * In order to insert checksums into the metadata in large chunks, we wait
808  * until bio submission time.   All the pages in the bio are checksummed and
809  * sums are attached onto the ordered extent record.
810  *
811  * At IO completion time the csums attached on the ordered extent record are
812  * inserted into the tree.
813  */
814 static void run_one_async_done(struct btrfs_work *work)
815 {
816 	struct async_submit_bio *async;
817 	struct inode *inode;
818 	blk_status_t ret;
819 
820 	async = container_of(work, struct  async_submit_bio, work);
821 	inode = async->inode;
822 
823 	/* If an error occurred we just want to clean up the bio and move on */
824 	if (async->status) {
825 		async->bio->bi_status = async->status;
826 		bio_endio(async->bio);
827 		return;
828 	}
829 
830 	/*
831 	 * All of the bios that pass through here are from async helpers.
832 	 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
833 	 * This changes nothing when cgroups aren't in use.
834 	 */
835 	async->bio->bi_opf |= REQ_CGROUP_PUNT;
836 	ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
837 	if (ret) {
838 		async->bio->bi_status = ret;
839 		bio_endio(async->bio);
840 	}
841 }
842 
843 static void run_one_async_free(struct btrfs_work *work)
844 {
845 	struct async_submit_bio *async;
846 
847 	async = container_of(work, struct  async_submit_bio, work);
848 	kfree(async);
849 }
850 
851 blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio,
852 				 int mirror_num, u64 dio_file_offset,
853 				 extent_submit_bio_start_t *submit_bio_start)
854 {
855 	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
856 	struct async_submit_bio *async;
857 
858 	async = kmalloc(sizeof(*async), GFP_NOFS);
859 	if (!async)
860 		return BLK_STS_RESOURCE;
861 
862 	async->inode = inode;
863 	async->bio = bio;
864 	async->mirror_num = mirror_num;
865 	async->submit_bio_start = submit_bio_start;
866 
867 	btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
868 			run_one_async_free);
869 
870 	async->dio_file_offset = dio_file_offset;
871 
872 	async->status = 0;
873 
874 	if (op_is_sync(bio->bi_opf))
875 		btrfs_queue_work(fs_info->hipri_workers, &async->work);
876 	else
877 		btrfs_queue_work(fs_info->workers, &async->work);
878 	return 0;
879 }
880 
881 static blk_status_t btree_csum_one_bio(struct bio *bio)
882 {
883 	struct bio_vec *bvec;
884 	struct btrfs_root *root;
885 	int ret = 0;
886 	struct bvec_iter_all iter_all;
887 
888 	ASSERT(!bio_flagged(bio, BIO_CLONED));
889 	bio_for_each_segment_all(bvec, bio, iter_all) {
890 		root = BTRFS_I(bvec->bv_page->mapping->host)->root;
891 		ret = csum_dirty_buffer(root->fs_info, bvec);
892 		if (ret)
893 			break;
894 	}
895 
896 	return errno_to_blk_status(ret);
897 }
898 
899 static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
900 					   u64 dio_file_offset)
901 {
902 	/*
903 	 * when we're called for a write, we're already in the async
904 	 * submission context.  Just jump into btrfs_map_bio
905 	 */
906 	return btree_csum_one_bio(bio);
907 }
908 
909 static bool should_async_write(struct btrfs_fs_info *fs_info,
910 			     struct btrfs_inode *bi)
911 {
912 	if (btrfs_is_zoned(fs_info))
913 		return false;
914 	if (atomic_read(&bi->sync_writers))
915 		return false;
916 	if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
917 		return false;
918 	return true;
919 }
920 
921 void btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, int mirror_num)
922 {
923 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
924 	blk_status_t ret;
925 
926 	if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
927 		/*
928 		 * called for a read, do the setup so that checksum validation
929 		 * can happen in the async kernel threads
930 		 */
931 		ret = btrfs_bio_wq_end_io(fs_info, bio,
932 					  BTRFS_WQ_ENDIO_METADATA);
933 		if (!ret)
934 			ret = btrfs_map_bio(fs_info, bio, mirror_num);
935 	} else if (!should_async_write(fs_info, BTRFS_I(inode))) {
936 		ret = btree_csum_one_bio(bio);
937 		if (!ret)
938 			ret = btrfs_map_bio(fs_info, bio, mirror_num);
939 	} else {
940 		/*
941 		 * kthread helpers are used to submit writes so that
942 		 * checksumming can happen in parallel across all CPUs
943 		 */
944 		ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
945 					  btree_submit_bio_start);
946 	}
947 
948 	if (ret) {
949 		bio->bi_status = ret;
950 		bio_endio(bio);
951 	}
952 }
953 
954 #ifdef CONFIG_MIGRATION
955 static int btree_migratepage(struct address_space *mapping,
956 			struct page *newpage, struct page *page,
957 			enum migrate_mode mode)
958 {
959 	/*
960 	 * we can't safely write a btree page from here,
961 	 * we haven't done the locking hook
962 	 */
963 	if (PageDirty(page))
964 		return -EAGAIN;
965 	/*
966 	 * Buffers may be managed in a filesystem specific way.
967 	 * We must have no buffers or drop them.
968 	 */
969 	if (page_has_private(page) &&
970 	    !try_to_release_page(page, GFP_KERNEL))
971 		return -EAGAIN;
972 	return migrate_page(mapping, newpage, page, mode);
973 }
974 #endif
975 
976 
977 static int btree_writepages(struct address_space *mapping,
978 			    struct writeback_control *wbc)
979 {
980 	struct btrfs_fs_info *fs_info;
981 	int ret;
982 
983 	if (wbc->sync_mode == WB_SYNC_NONE) {
984 
985 		if (wbc->for_kupdate)
986 			return 0;
987 
988 		fs_info = BTRFS_I(mapping->host)->root->fs_info;
989 		/* this is a bit racy, but that's ok */
990 		ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
991 					     BTRFS_DIRTY_METADATA_THRESH,
992 					     fs_info->dirty_metadata_batch);
993 		if (ret < 0)
994 			return 0;
995 	}
996 	return btree_write_cache_pages(mapping, wbc);
997 }
998 
999 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
1000 {
1001 	if (folio_test_writeback(folio) || folio_test_dirty(folio))
1002 		return false;
1003 
1004 	return try_release_extent_buffer(&folio->page);
1005 }
1006 
1007 static void btree_invalidate_folio(struct folio *folio, size_t offset,
1008 				 size_t length)
1009 {
1010 	struct extent_io_tree *tree;
1011 	tree = &BTRFS_I(folio->mapping->host)->io_tree;
1012 	extent_invalidate_folio(tree, folio, offset);
1013 	btree_release_folio(folio, GFP_NOFS);
1014 	if (folio_get_private(folio)) {
1015 		btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
1016 			   "folio private not zero on folio %llu",
1017 			   (unsigned long long)folio_pos(folio));
1018 		folio_detach_private(folio);
1019 	}
1020 }
1021 
1022 #ifdef DEBUG
1023 static bool btree_dirty_folio(struct address_space *mapping,
1024 		struct folio *folio)
1025 {
1026 	struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
1027 	struct btrfs_subpage *subpage;
1028 	struct extent_buffer *eb;
1029 	int cur_bit = 0;
1030 	u64 page_start = folio_pos(folio);
1031 
1032 	if (fs_info->sectorsize == PAGE_SIZE) {
1033 		eb = folio_get_private(folio);
1034 		BUG_ON(!eb);
1035 		BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1036 		BUG_ON(!atomic_read(&eb->refs));
1037 		btrfs_assert_tree_write_locked(eb);
1038 		return filemap_dirty_folio(mapping, folio);
1039 	}
1040 	subpage = folio_get_private(folio);
1041 
1042 	ASSERT(subpage->dirty_bitmap);
1043 	while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
1044 		unsigned long flags;
1045 		u64 cur;
1046 		u16 tmp = (1 << cur_bit);
1047 
1048 		spin_lock_irqsave(&subpage->lock, flags);
1049 		if (!(tmp & subpage->dirty_bitmap)) {
1050 			spin_unlock_irqrestore(&subpage->lock, flags);
1051 			cur_bit++;
1052 			continue;
1053 		}
1054 		spin_unlock_irqrestore(&subpage->lock, flags);
1055 		cur = page_start + cur_bit * fs_info->sectorsize;
1056 
1057 		eb = find_extent_buffer(fs_info, cur);
1058 		ASSERT(eb);
1059 		ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1060 		ASSERT(atomic_read(&eb->refs));
1061 		btrfs_assert_tree_write_locked(eb);
1062 		free_extent_buffer(eb);
1063 
1064 		cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
1065 	}
1066 	return filemap_dirty_folio(mapping, folio);
1067 }
1068 #else
1069 #define btree_dirty_folio filemap_dirty_folio
1070 #endif
1071 
1072 static const struct address_space_operations btree_aops = {
1073 	.writepages	= btree_writepages,
1074 	.release_folio	= btree_release_folio,
1075 	.invalidate_folio = btree_invalidate_folio,
1076 #ifdef CONFIG_MIGRATION
1077 	.migratepage	= btree_migratepage,
1078 #endif
1079 	.dirty_folio = btree_dirty_folio,
1080 };
1081 
1082 struct extent_buffer *btrfs_find_create_tree_block(
1083 						struct btrfs_fs_info *fs_info,
1084 						u64 bytenr, u64 owner_root,
1085 						int level)
1086 {
1087 	if (btrfs_is_testing(fs_info))
1088 		return alloc_test_extent_buffer(fs_info, bytenr);
1089 	return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
1090 }
1091 
1092 /*
1093  * Read tree block at logical address @bytenr and do variant basic but critical
1094  * verification.
1095  *
1096  * @owner_root:		the objectid of the root owner for this block.
1097  * @parent_transid:	expected transid of this tree block, skip check if 0
1098  * @level:		expected level, mandatory check
1099  * @first_key:		expected key in slot 0, skip check if NULL
1100  */
1101 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1102 				      u64 owner_root, u64 parent_transid,
1103 				      int level, struct btrfs_key *first_key)
1104 {
1105 	struct extent_buffer *buf = NULL;
1106 	int ret;
1107 
1108 	buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
1109 	if (IS_ERR(buf))
1110 		return buf;
1111 
1112 	ret = btrfs_read_extent_buffer(buf, parent_transid, level, first_key);
1113 	if (ret) {
1114 		free_extent_buffer_stale(buf);
1115 		return ERR_PTR(ret);
1116 	}
1117 	if (btrfs_check_eb_owner(buf, owner_root)) {
1118 		free_extent_buffer_stale(buf);
1119 		return ERR_PTR(-EUCLEAN);
1120 	}
1121 	return buf;
1122 
1123 }
1124 
1125 void btrfs_clean_tree_block(struct extent_buffer *buf)
1126 {
1127 	struct btrfs_fs_info *fs_info = buf->fs_info;
1128 	if (btrfs_header_generation(buf) ==
1129 	    fs_info->running_transaction->transid) {
1130 		btrfs_assert_tree_write_locked(buf);
1131 
1132 		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1133 			percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1134 						 -buf->len,
1135 						 fs_info->dirty_metadata_batch);
1136 			clear_extent_buffer_dirty(buf);
1137 		}
1138 	}
1139 }
1140 
1141 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1142 			 u64 objectid)
1143 {
1144 	bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1145 
1146 	memset(&root->root_key, 0, sizeof(root->root_key));
1147 	memset(&root->root_item, 0, sizeof(root->root_item));
1148 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1149 	root->fs_info = fs_info;
1150 	root->root_key.objectid = objectid;
1151 	root->node = NULL;
1152 	root->commit_root = NULL;
1153 	root->state = 0;
1154 	RB_CLEAR_NODE(&root->rb_node);
1155 
1156 	root->last_trans = 0;
1157 	root->free_objectid = 0;
1158 	root->nr_delalloc_inodes = 0;
1159 	root->nr_ordered_extents = 0;
1160 	root->inode_tree = RB_ROOT;
1161 	xa_init_flags(&root->delayed_nodes, GFP_ATOMIC);
1162 
1163 	btrfs_init_root_block_rsv(root);
1164 
1165 	INIT_LIST_HEAD(&root->dirty_list);
1166 	INIT_LIST_HEAD(&root->root_list);
1167 	INIT_LIST_HEAD(&root->delalloc_inodes);
1168 	INIT_LIST_HEAD(&root->delalloc_root);
1169 	INIT_LIST_HEAD(&root->ordered_extents);
1170 	INIT_LIST_HEAD(&root->ordered_root);
1171 	INIT_LIST_HEAD(&root->reloc_dirty_list);
1172 	INIT_LIST_HEAD(&root->logged_list[0]);
1173 	INIT_LIST_HEAD(&root->logged_list[1]);
1174 	spin_lock_init(&root->inode_lock);
1175 	spin_lock_init(&root->delalloc_lock);
1176 	spin_lock_init(&root->ordered_extent_lock);
1177 	spin_lock_init(&root->accounting_lock);
1178 	spin_lock_init(&root->log_extents_lock[0]);
1179 	spin_lock_init(&root->log_extents_lock[1]);
1180 	spin_lock_init(&root->qgroup_meta_rsv_lock);
1181 	mutex_init(&root->objectid_mutex);
1182 	mutex_init(&root->log_mutex);
1183 	mutex_init(&root->ordered_extent_mutex);
1184 	mutex_init(&root->delalloc_mutex);
1185 	init_waitqueue_head(&root->qgroup_flush_wait);
1186 	init_waitqueue_head(&root->log_writer_wait);
1187 	init_waitqueue_head(&root->log_commit_wait[0]);
1188 	init_waitqueue_head(&root->log_commit_wait[1]);
1189 	INIT_LIST_HEAD(&root->log_ctxs[0]);
1190 	INIT_LIST_HEAD(&root->log_ctxs[1]);
1191 	atomic_set(&root->log_commit[0], 0);
1192 	atomic_set(&root->log_commit[1], 0);
1193 	atomic_set(&root->log_writers, 0);
1194 	atomic_set(&root->log_batch, 0);
1195 	refcount_set(&root->refs, 1);
1196 	atomic_set(&root->snapshot_force_cow, 0);
1197 	atomic_set(&root->nr_swapfiles, 0);
1198 	root->log_transid = 0;
1199 	root->log_transid_committed = -1;
1200 	root->last_log_commit = 0;
1201 	root->anon_dev = 0;
1202 	if (!dummy) {
1203 		extent_io_tree_init(fs_info, &root->dirty_log_pages,
1204 				    IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1205 		extent_io_tree_init(fs_info, &root->log_csum_range,
1206 				    IO_TREE_LOG_CSUM_RANGE, NULL);
1207 	}
1208 
1209 	spin_lock_init(&root->root_item_lock);
1210 	btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1211 #ifdef CONFIG_BTRFS_DEBUG
1212 	INIT_LIST_HEAD(&root->leak_list);
1213 	spin_lock(&fs_info->fs_roots_lock);
1214 	list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1215 	spin_unlock(&fs_info->fs_roots_lock);
1216 #endif
1217 }
1218 
1219 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1220 					   u64 objectid, gfp_t flags)
1221 {
1222 	struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1223 	if (root)
1224 		__setup_root(root, fs_info, objectid);
1225 	return root;
1226 }
1227 
1228 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1229 /* Should only be used by the testing infrastructure */
1230 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1231 {
1232 	struct btrfs_root *root;
1233 
1234 	if (!fs_info)
1235 		return ERR_PTR(-EINVAL);
1236 
1237 	root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1238 	if (!root)
1239 		return ERR_PTR(-ENOMEM);
1240 
1241 	/* We don't use the stripesize in selftest, set it as sectorsize */
1242 	root->alloc_bytenr = 0;
1243 
1244 	return root;
1245 }
1246 #endif
1247 
1248 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
1249 {
1250 	const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
1251 	const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
1252 
1253 	return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
1254 }
1255 
1256 static int global_root_key_cmp(const void *k, const struct rb_node *node)
1257 {
1258 	const struct btrfs_key *key = k;
1259 	const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
1260 
1261 	return btrfs_comp_cpu_keys(key, &root->root_key);
1262 }
1263 
1264 int btrfs_global_root_insert(struct btrfs_root *root)
1265 {
1266 	struct btrfs_fs_info *fs_info = root->fs_info;
1267 	struct rb_node *tmp;
1268 
1269 	write_lock(&fs_info->global_root_lock);
1270 	tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
1271 	write_unlock(&fs_info->global_root_lock);
1272 	ASSERT(!tmp);
1273 
1274 	return tmp ? -EEXIST : 0;
1275 }
1276 
1277 void btrfs_global_root_delete(struct btrfs_root *root)
1278 {
1279 	struct btrfs_fs_info *fs_info = root->fs_info;
1280 
1281 	write_lock(&fs_info->global_root_lock);
1282 	rb_erase(&root->rb_node, &fs_info->global_root_tree);
1283 	write_unlock(&fs_info->global_root_lock);
1284 }
1285 
1286 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
1287 				     struct btrfs_key *key)
1288 {
1289 	struct rb_node *node;
1290 	struct btrfs_root *root = NULL;
1291 
1292 	read_lock(&fs_info->global_root_lock);
1293 	node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
1294 	if (node)
1295 		root = container_of(node, struct btrfs_root, rb_node);
1296 	read_unlock(&fs_info->global_root_lock);
1297 
1298 	return root;
1299 }
1300 
1301 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
1302 {
1303 	struct btrfs_block_group *block_group;
1304 	u64 ret;
1305 
1306 	if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
1307 		return 0;
1308 
1309 	if (bytenr)
1310 		block_group = btrfs_lookup_block_group(fs_info, bytenr);
1311 	else
1312 		block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
1313 	ASSERT(block_group);
1314 	if (!block_group)
1315 		return 0;
1316 	ret = block_group->global_root_id;
1317 	btrfs_put_block_group(block_group);
1318 
1319 	return ret;
1320 }
1321 
1322 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1323 {
1324 	struct btrfs_key key = {
1325 		.objectid = BTRFS_CSUM_TREE_OBJECTID,
1326 		.type = BTRFS_ROOT_ITEM_KEY,
1327 		.offset = btrfs_global_root_id(fs_info, bytenr),
1328 	};
1329 
1330 	return btrfs_global_root(fs_info, &key);
1331 }
1332 
1333 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1334 {
1335 	struct btrfs_key key = {
1336 		.objectid = BTRFS_EXTENT_TREE_OBJECTID,
1337 		.type = BTRFS_ROOT_ITEM_KEY,
1338 		.offset = btrfs_global_root_id(fs_info, bytenr),
1339 	};
1340 
1341 	return btrfs_global_root(fs_info, &key);
1342 }
1343 
1344 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1345 				     u64 objectid)
1346 {
1347 	struct btrfs_fs_info *fs_info = trans->fs_info;
1348 	struct extent_buffer *leaf;
1349 	struct btrfs_root *tree_root = fs_info->tree_root;
1350 	struct btrfs_root *root;
1351 	struct btrfs_key key;
1352 	unsigned int nofs_flag;
1353 	int ret = 0;
1354 
1355 	/*
1356 	 * We're holding a transaction handle, so use a NOFS memory allocation
1357 	 * context to avoid deadlock if reclaim happens.
1358 	 */
1359 	nofs_flag = memalloc_nofs_save();
1360 	root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1361 	memalloc_nofs_restore(nofs_flag);
1362 	if (!root)
1363 		return ERR_PTR(-ENOMEM);
1364 
1365 	root->root_key.objectid = objectid;
1366 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1367 	root->root_key.offset = 0;
1368 
1369 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1370 				      BTRFS_NESTING_NORMAL);
1371 	if (IS_ERR(leaf)) {
1372 		ret = PTR_ERR(leaf);
1373 		leaf = NULL;
1374 		goto fail_unlock;
1375 	}
1376 
1377 	root->node = leaf;
1378 	btrfs_mark_buffer_dirty(leaf);
1379 
1380 	root->commit_root = btrfs_root_node(root);
1381 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1382 
1383 	btrfs_set_root_flags(&root->root_item, 0);
1384 	btrfs_set_root_limit(&root->root_item, 0);
1385 	btrfs_set_root_bytenr(&root->root_item, leaf->start);
1386 	btrfs_set_root_generation(&root->root_item, trans->transid);
1387 	btrfs_set_root_level(&root->root_item, 0);
1388 	btrfs_set_root_refs(&root->root_item, 1);
1389 	btrfs_set_root_used(&root->root_item, leaf->len);
1390 	btrfs_set_root_last_snapshot(&root->root_item, 0);
1391 	btrfs_set_root_dirid(&root->root_item, 0);
1392 	if (is_fstree(objectid))
1393 		generate_random_guid(root->root_item.uuid);
1394 	else
1395 		export_guid(root->root_item.uuid, &guid_null);
1396 	btrfs_set_root_drop_level(&root->root_item, 0);
1397 
1398 	btrfs_tree_unlock(leaf);
1399 
1400 	key.objectid = objectid;
1401 	key.type = BTRFS_ROOT_ITEM_KEY;
1402 	key.offset = 0;
1403 	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1404 	if (ret)
1405 		goto fail;
1406 
1407 	return root;
1408 
1409 fail_unlock:
1410 	if (leaf)
1411 		btrfs_tree_unlock(leaf);
1412 fail:
1413 	btrfs_put_root(root);
1414 
1415 	return ERR_PTR(ret);
1416 }
1417 
1418 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1419 					 struct btrfs_fs_info *fs_info)
1420 {
1421 	struct btrfs_root *root;
1422 
1423 	root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1424 	if (!root)
1425 		return ERR_PTR(-ENOMEM);
1426 
1427 	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1428 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1429 	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1430 
1431 	return root;
1432 }
1433 
1434 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1435 			      struct btrfs_root *root)
1436 {
1437 	struct extent_buffer *leaf;
1438 
1439 	/*
1440 	 * DON'T set SHAREABLE bit for log trees.
1441 	 *
1442 	 * Log trees are not exposed to user space thus can't be snapshotted,
1443 	 * and they go away before a real commit is actually done.
1444 	 *
1445 	 * They do store pointers to file data extents, and those reference
1446 	 * counts still get updated (along with back refs to the log tree).
1447 	 */
1448 
1449 	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1450 			NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1451 	if (IS_ERR(leaf))
1452 		return PTR_ERR(leaf);
1453 
1454 	root->node = leaf;
1455 
1456 	btrfs_mark_buffer_dirty(root->node);
1457 	btrfs_tree_unlock(root->node);
1458 
1459 	return 0;
1460 }
1461 
1462 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1463 			     struct btrfs_fs_info *fs_info)
1464 {
1465 	struct btrfs_root *log_root;
1466 
1467 	log_root = alloc_log_tree(trans, fs_info);
1468 	if (IS_ERR(log_root))
1469 		return PTR_ERR(log_root);
1470 
1471 	if (!btrfs_is_zoned(fs_info)) {
1472 		int ret = btrfs_alloc_log_tree_node(trans, log_root);
1473 
1474 		if (ret) {
1475 			btrfs_put_root(log_root);
1476 			return ret;
1477 		}
1478 	}
1479 
1480 	WARN_ON(fs_info->log_root_tree);
1481 	fs_info->log_root_tree = log_root;
1482 	return 0;
1483 }
1484 
1485 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1486 		       struct btrfs_root *root)
1487 {
1488 	struct btrfs_fs_info *fs_info = root->fs_info;
1489 	struct btrfs_root *log_root;
1490 	struct btrfs_inode_item *inode_item;
1491 	int ret;
1492 
1493 	log_root = alloc_log_tree(trans, fs_info);
1494 	if (IS_ERR(log_root))
1495 		return PTR_ERR(log_root);
1496 
1497 	ret = btrfs_alloc_log_tree_node(trans, log_root);
1498 	if (ret) {
1499 		btrfs_put_root(log_root);
1500 		return ret;
1501 	}
1502 
1503 	log_root->last_trans = trans->transid;
1504 	log_root->root_key.offset = root->root_key.objectid;
1505 
1506 	inode_item = &log_root->root_item.inode;
1507 	btrfs_set_stack_inode_generation(inode_item, 1);
1508 	btrfs_set_stack_inode_size(inode_item, 3);
1509 	btrfs_set_stack_inode_nlink(inode_item, 1);
1510 	btrfs_set_stack_inode_nbytes(inode_item,
1511 				     fs_info->nodesize);
1512 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1513 
1514 	btrfs_set_root_node(&log_root->root_item, log_root->node);
1515 
1516 	WARN_ON(root->log_root);
1517 	root->log_root = log_root;
1518 	root->log_transid = 0;
1519 	root->log_transid_committed = -1;
1520 	root->last_log_commit = 0;
1521 	return 0;
1522 }
1523 
1524 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1525 					      struct btrfs_path *path,
1526 					      struct btrfs_key *key)
1527 {
1528 	struct btrfs_root *root;
1529 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1530 	u64 generation;
1531 	int ret;
1532 	int level;
1533 
1534 	root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1535 	if (!root)
1536 		return ERR_PTR(-ENOMEM);
1537 
1538 	ret = btrfs_find_root(tree_root, key, path,
1539 			      &root->root_item, &root->root_key);
1540 	if (ret) {
1541 		if (ret > 0)
1542 			ret = -ENOENT;
1543 		goto fail;
1544 	}
1545 
1546 	generation = btrfs_root_generation(&root->root_item);
1547 	level = btrfs_root_level(&root->root_item);
1548 	root->node = read_tree_block(fs_info,
1549 				     btrfs_root_bytenr(&root->root_item),
1550 				     key->objectid, generation, level, NULL);
1551 	if (IS_ERR(root->node)) {
1552 		ret = PTR_ERR(root->node);
1553 		root->node = NULL;
1554 		goto fail;
1555 	}
1556 	if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1557 		ret = -EIO;
1558 		goto fail;
1559 	}
1560 
1561 	/*
1562 	 * For real fs, and not log/reloc trees, root owner must
1563 	 * match its root node owner
1564 	 */
1565 	if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1566 	    root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1567 	    root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1568 	    root->root_key.objectid != btrfs_header_owner(root->node)) {
1569 		btrfs_crit(fs_info,
1570 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1571 			   root->root_key.objectid, root->node->start,
1572 			   btrfs_header_owner(root->node),
1573 			   root->root_key.objectid);
1574 		ret = -EUCLEAN;
1575 		goto fail;
1576 	}
1577 	root->commit_root = btrfs_root_node(root);
1578 	return root;
1579 fail:
1580 	btrfs_put_root(root);
1581 	return ERR_PTR(ret);
1582 }
1583 
1584 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1585 					struct btrfs_key *key)
1586 {
1587 	struct btrfs_root *root;
1588 	struct btrfs_path *path;
1589 
1590 	path = btrfs_alloc_path();
1591 	if (!path)
1592 		return ERR_PTR(-ENOMEM);
1593 	root = read_tree_root_path(tree_root, path, key);
1594 	btrfs_free_path(path);
1595 
1596 	return root;
1597 }
1598 
1599 /*
1600  * Initialize subvolume root in-memory structure
1601  *
1602  * @anon_dev:	anonymous device to attach to the root, if zero, allocate new
1603  */
1604 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1605 {
1606 	int ret;
1607 	unsigned int nofs_flag;
1608 
1609 	/*
1610 	 * We might be called under a transaction (e.g. indirect backref
1611 	 * resolution) which could deadlock if it triggers memory reclaim
1612 	 */
1613 	nofs_flag = memalloc_nofs_save();
1614 	ret = btrfs_drew_lock_init(&root->snapshot_lock);
1615 	memalloc_nofs_restore(nofs_flag);
1616 	if (ret)
1617 		goto fail;
1618 
1619 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1620 	    !btrfs_is_data_reloc_root(root)) {
1621 		set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1622 		btrfs_check_and_init_root_item(&root->root_item);
1623 	}
1624 
1625 	/*
1626 	 * Don't assign anonymous block device to roots that are not exposed to
1627 	 * userspace, the id pool is limited to 1M
1628 	 */
1629 	if (is_fstree(root->root_key.objectid) &&
1630 	    btrfs_root_refs(&root->root_item) > 0) {
1631 		if (!anon_dev) {
1632 			ret = get_anon_bdev(&root->anon_dev);
1633 			if (ret)
1634 				goto fail;
1635 		} else {
1636 			root->anon_dev = anon_dev;
1637 		}
1638 	}
1639 
1640 	mutex_lock(&root->objectid_mutex);
1641 	ret = btrfs_init_root_free_objectid(root);
1642 	if (ret) {
1643 		mutex_unlock(&root->objectid_mutex);
1644 		goto fail;
1645 	}
1646 
1647 	ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1648 
1649 	mutex_unlock(&root->objectid_mutex);
1650 
1651 	return 0;
1652 fail:
1653 	/* The caller is responsible to call btrfs_free_fs_root */
1654 	return ret;
1655 }
1656 
1657 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1658 					       u64 root_id)
1659 {
1660 	struct btrfs_root *root;
1661 
1662 	spin_lock(&fs_info->fs_roots_lock);
1663 	root = xa_load(&fs_info->fs_roots, (unsigned long)root_id);
1664 	if (root)
1665 		root = btrfs_grab_root(root);
1666 	spin_unlock(&fs_info->fs_roots_lock);
1667 	return root;
1668 }
1669 
1670 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1671 						u64 objectid)
1672 {
1673 	struct btrfs_key key = {
1674 		.objectid = objectid,
1675 		.type = BTRFS_ROOT_ITEM_KEY,
1676 		.offset = 0,
1677 	};
1678 
1679 	if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1680 		return btrfs_grab_root(fs_info->tree_root);
1681 	if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1682 		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1683 	if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1684 		return btrfs_grab_root(fs_info->chunk_root);
1685 	if (objectid == BTRFS_DEV_TREE_OBJECTID)
1686 		return btrfs_grab_root(fs_info->dev_root);
1687 	if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1688 		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1689 	if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1690 		return btrfs_grab_root(fs_info->quota_root) ?
1691 			fs_info->quota_root : ERR_PTR(-ENOENT);
1692 	if (objectid == BTRFS_UUID_TREE_OBJECTID)
1693 		return btrfs_grab_root(fs_info->uuid_root) ?
1694 			fs_info->uuid_root : ERR_PTR(-ENOENT);
1695 	if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) {
1696 		struct btrfs_root *root = btrfs_global_root(fs_info, &key);
1697 
1698 		return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT);
1699 	}
1700 	return NULL;
1701 }
1702 
1703 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1704 			 struct btrfs_root *root)
1705 {
1706 	int ret;
1707 
1708 	spin_lock(&fs_info->fs_roots_lock);
1709 	ret = xa_insert(&fs_info->fs_roots, (unsigned long)root->root_key.objectid,
1710 			root, GFP_NOFS);
1711 	if (ret == 0) {
1712 		btrfs_grab_root(root);
1713 		set_bit(BTRFS_ROOT_REGISTERED, &root->state);
1714 	}
1715 	spin_unlock(&fs_info->fs_roots_lock);
1716 
1717 	return ret;
1718 }
1719 
1720 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1721 {
1722 #ifdef CONFIG_BTRFS_DEBUG
1723 	struct btrfs_root *root;
1724 
1725 	while (!list_empty(&fs_info->allocated_roots)) {
1726 		char buf[BTRFS_ROOT_NAME_BUF_LEN];
1727 
1728 		root = list_first_entry(&fs_info->allocated_roots,
1729 					struct btrfs_root, leak_list);
1730 		btrfs_err(fs_info, "leaked root %s refcount %d",
1731 			  btrfs_root_name(&root->root_key, buf),
1732 			  refcount_read(&root->refs));
1733 		while (refcount_read(&root->refs) > 1)
1734 			btrfs_put_root(root);
1735 		btrfs_put_root(root);
1736 	}
1737 #endif
1738 }
1739 
1740 static void free_global_roots(struct btrfs_fs_info *fs_info)
1741 {
1742 	struct btrfs_root *root;
1743 	struct rb_node *node;
1744 
1745 	while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1746 		root = rb_entry(node, struct btrfs_root, rb_node);
1747 		rb_erase(&root->rb_node, &fs_info->global_root_tree);
1748 		btrfs_put_root(root);
1749 	}
1750 }
1751 
1752 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1753 {
1754 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1755 	percpu_counter_destroy(&fs_info->delalloc_bytes);
1756 	percpu_counter_destroy(&fs_info->ordered_bytes);
1757 	percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1758 	btrfs_free_csum_hash(fs_info);
1759 	btrfs_free_stripe_hash_table(fs_info);
1760 	btrfs_free_ref_cache(fs_info);
1761 	kfree(fs_info->balance_ctl);
1762 	kfree(fs_info->delayed_root);
1763 	free_global_roots(fs_info);
1764 	btrfs_put_root(fs_info->tree_root);
1765 	btrfs_put_root(fs_info->chunk_root);
1766 	btrfs_put_root(fs_info->dev_root);
1767 	btrfs_put_root(fs_info->quota_root);
1768 	btrfs_put_root(fs_info->uuid_root);
1769 	btrfs_put_root(fs_info->fs_root);
1770 	btrfs_put_root(fs_info->data_reloc_root);
1771 	btrfs_put_root(fs_info->block_group_root);
1772 	btrfs_check_leaked_roots(fs_info);
1773 	btrfs_extent_buffer_leak_debug_check(fs_info);
1774 	kfree(fs_info->super_copy);
1775 	kfree(fs_info->super_for_commit);
1776 	kfree(fs_info->subpage_info);
1777 	kvfree(fs_info);
1778 }
1779 
1780 
1781 /*
1782  * Get an in-memory reference of a root structure.
1783  *
1784  * For essential trees like root/extent tree, we grab it from fs_info directly.
1785  * For subvolume trees, we check the cached filesystem roots first. If not
1786  * found, then read it from disk and add it to cached fs roots.
1787  *
1788  * Caller should release the root by calling btrfs_put_root() after the usage.
1789  *
1790  * NOTE: Reloc and log trees can't be read by this function as they share the
1791  *	 same root objectid.
1792  *
1793  * @objectid:	root id
1794  * @anon_dev:	preallocated anonymous block device number for new roots,
1795  * 		pass 0 for new allocation.
1796  * @check_ref:	whether to check root item references, If true, return -ENOENT
1797  *		for orphan roots
1798  */
1799 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1800 					     u64 objectid, dev_t anon_dev,
1801 					     bool check_ref)
1802 {
1803 	struct btrfs_root *root;
1804 	struct btrfs_path *path;
1805 	struct btrfs_key key;
1806 	int ret;
1807 
1808 	root = btrfs_get_global_root(fs_info, objectid);
1809 	if (root)
1810 		return root;
1811 again:
1812 	root = btrfs_lookup_fs_root(fs_info, objectid);
1813 	if (root) {
1814 		/* Shouldn't get preallocated anon_dev for cached roots */
1815 		ASSERT(!anon_dev);
1816 		if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1817 			btrfs_put_root(root);
1818 			return ERR_PTR(-ENOENT);
1819 		}
1820 		return root;
1821 	}
1822 
1823 	key.objectid = objectid;
1824 	key.type = BTRFS_ROOT_ITEM_KEY;
1825 	key.offset = (u64)-1;
1826 	root = btrfs_read_tree_root(fs_info->tree_root, &key);
1827 	if (IS_ERR(root))
1828 		return root;
1829 
1830 	if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1831 		ret = -ENOENT;
1832 		goto fail;
1833 	}
1834 
1835 	ret = btrfs_init_fs_root(root, anon_dev);
1836 	if (ret)
1837 		goto fail;
1838 
1839 	path = btrfs_alloc_path();
1840 	if (!path) {
1841 		ret = -ENOMEM;
1842 		goto fail;
1843 	}
1844 	key.objectid = BTRFS_ORPHAN_OBJECTID;
1845 	key.type = BTRFS_ORPHAN_ITEM_KEY;
1846 	key.offset = objectid;
1847 
1848 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1849 	btrfs_free_path(path);
1850 	if (ret < 0)
1851 		goto fail;
1852 	if (ret == 0)
1853 		set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1854 
1855 	ret = btrfs_insert_fs_root(fs_info, root);
1856 	if (ret) {
1857 		if (ret == -EEXIST) {
1858 			btrfs_put_root(root);
1859 			goto again;
1860 		}
1861 		goto fail;
1862 	}
1863 	return root;
1864 fail:
1865 	/*
1866 	 * If our caller provided us an anonymous device, then it's his
1867 	 * responsability to free it in case we fail. So we have to set our
1868 	 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1869 	 * and once again by our caller.
1870 	 */
1871 	if (anon_dev)
1872 		root->anon_dev = 0;
1873 	btrfs_put_root(root);
1874 	return ERR_PTR(ret);
1875 }
1876 
1877 /*
1878  * Get in-memory reference of a root structure
1879  *
1880  * @objectid:	tree objectid
1881  * @check_ref:	if set, verify that the tree exists and the item has at least
1882  *		one reference
1883  */
1884 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1885 				     u64 objectid, bool check_ref)
1886 {
1887 	return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1888 }
1889 
1890 /*
1891  * Get in-memory reference of a root structure, created as new, optionally pass
1892  * the anonymous block device id
1893  *
1894  * @objectid:	tree objectid
1895  * @anon_dev:	if zero, allocate a new anonymous block device or use the
1896  *		parameter value
1897  */
1898 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1899 					 u64 objectid, dev_t anon_dev)
1900 {
1901 	return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1902 }
1903 
1904 /*
1905  * btrfs_get_fs_root_commit_root - return a root for the given objectid
1906  * @fs_info:	the fs_info
1907  * @objectid:	the objectid we need to lookup
1908  *
1909  * This is exclusively used for backref walking, and exists specifically because
1910  * of how qgroups does lookups.  Qgroups will do a backref lookup at delayed ref
1911  * creation time, which means we may have to read the tree_root in order to look
1912  * up a fs root that is not in memory.  If the root is not in memory we will
1913  * read the tree root commit root and look up the fs root from there.  This is a
1914  * temporary root, it will not be inserted into the radix tree as it doesn't
1915  * have the most uptodate information, it'll simply be discarded once the
1916  * backref code is finished using the root.
1917  */
1918 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1919 						 struct btrfs_path *path,
1920 						 u64 objectid)
1921 {
1922 	struct btrfs_root *root;
1923 	struct btrfs_key key;
1924 
1925 	ASSERT(path->search_commit_root && path->skip_locking);
1926 
1927 	/*
1928 	 * This can return -ENOENT if we ask for a root that doesn't exist, but
1929 	 * since this is called via the backref walking code we won't be looking
1930 	 * up a root that doesn't exist, unless there's corruption.  So if root
1931 	 * != NULL just return it.
1932 	 */
1933 	root = btrfs_get_global_root(fs_info, objectid);
1934 	if (root)
1935 		return root;
1936 
1937 	root = btrfs_lookup_fs_root(fs_info, objectid);
1938 	if (root)
1939 		return root;
1940 
1941 	key.objectid = objectid;
1942 	key.type = BTRFS_ROOT_ITEM_KEY;
1943 	key.offset = (u64)-1;
1944 	root = read_tree_root_path(fs_info->tree_root, path, &key);
1945 	btrfs_release_path(path);
1946 
1947 	return root;
1948 }
1949 
1950 /*
1951  * called by the kthread helper functions to finally call the bio end_io
1952  * functions.  This is where read checksum verification actually happens
1953  */
1954 static void end_workqueue_fn(struct btrfs_work *work)
1955 {
1956 	struct bio *bio;
1957 	struct btrfs_end_io_wq *end_io_wq;
1958 
1959 	end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1960 	bio = end_io_wq->bio;
1961 
1962 	bio->bi_status = end_io_wq->status;
1963 	bio->bi_private = end_io_wq->private;
1964 	bio->bi_end_io = end_io_wq->end_io;
1965 	bio_endio(bio);
1966 	kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1967 }
1968 
1969 static int cleaner_kthread(void *arg)
1970 {
1971 	struct btrfs_fs_info *fs_info = arg;
1972 	int again;
1973 
1974 	while (1) {
1975 		again = 0;
1976 
1977 		set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1978 
1979 		/* Make the cleaner go to sleep early. */
1980 		if (btrfs_need_cleaner_sleep(fs_info))
1981 			goto sleep;
1982 
1983 		/*
1984 		 * Do not do anything if we might cause open_ctree() to block
1985 		 * before we have finished mounting the filesystem.
1986 		 */
1987 		if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1988 			goto sleep;
1989 
1990 		if (!mutex_trylock(&fs_info->cleaner_mutex))
1991 			goto sleep;
1992 
1993 		/*
1994 		 * Avoid the problem that we change the status of the fs
1995 		 * during the above check and trylock.
1996 		 */
1997 		if (btrfs_need_cleaner_sleep(fs_info)) {
1998 			mutex_unlock(&fs_info->cleaner_mutex);
1999 			goto sleep;
2000 		}
2001 
2002 		btrfs_run_delayed_iputs(fs_info);
2003 
2004 		again = btrfs_clean_one_deleted_snapshot(fs_info);
2005 		mutex_unlock(&fs_info->cleaner_mutex);
2006 
2007 		/*
2008 		 * The defragger has dealt with the R/O remount and umount,
2009 		 * needn't do anything special here.
2010 		 */
2011 		btrfs_run_defrag_inodes(fs_info);
2012 
2013 		/*
2014 		 * Acquires fs_info->reclaim_bgs_lock to avoid racing
2015 		 * with relocation (btrfs_relocate_chunk) and relocation
2016 		 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
2017 		 * after acquiring fs_info->reclaim_bgs_lock. So we
2018 		 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
2019 		 * unused block groups.
2020 		 */
2021 		btrfs_delete_unused_bgs(fs_info);
2022 
2023 		/*
2024 		 * Reclaim block groups in the reclaim_bgs list after we deleted
2025 		 * all unused block_groups. This possibly gives us some more free
2026 		 * space.
2027 		 */
2028 		btrfs_reclaim_bgs(fs_info);
2029 sleep:
2030 		clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
2031 		if (kthread_should_park())
2032 			kthread_parkme();
2033 		if (kthread_should_stop())
2034 			return 0;
2035 		if (!again) {
2036 			set_current_state(TASK_INTERRUPTIBLE);
2037 			schedule();
2038 			__set_current_state(TASK_RUNNING);
2039 		}
2040 	}
2041 }
2042 
2043 static int transaction_kthread(void *arg)
2044 {
2045 	struct btrfs_root *root = arg;
2046 	struct btrfs_fs_info *fs_info = root->fs_info;
2047 	struct btrfs_trans_handle *trans;
2048 	struct btrfs_transaction *cur;
2049 	u64 transid;
2050 	time64_t delta;
2051 	unsigned long delay;
2052 	bool cannot_commit;
2053 
2054 	do {
2055 		cannot_commit = false;
2056 		delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
2057 		mutex_lock(&fs_info->transaction_kthread_mutex);
2058 
2059 		spin_lock(&fs_info->trans_lock);
2060 		cur = fs_info->running_transaction;
2061 		if (!cur) {
2062 			spin_unlock(&fs_info->trans_lock);
2063 			goto sleep;
2064 		}
2065 
2066 		delta = ktime_get_seconds() - cur->start_time;
2067 		if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
2068 		    cur->state < TRANS_STATE_COMMIT_START &&
2069 		    delta < fs_info->commit_interval) {
2070 			spin_unlock(&fs_info->trans_lock);
2071 			delay -= msecs_to_jiffies((delta - 1) * 1000);
2072 			delay = min(delay,
2073 				    msecs_to_jiffies(fs_info->commit_interval * 1000));
2074 			goto sleep;
2075 		}
2076 		transid = cur->transid;
2077 		spin_unlock(&fs_info->trans_lock);
2078 
2079 		/* If the file system is aborted, this will always fail. */
2080 		trans = btrfs_attach_transaction(root);
2081 		if (IS_ERR(trans)) {
2082 			if (PTR_ERR(trans) != -ENOENT)
2083 				cannot_commit = true;
2084 			goto sleep;
2085 		}
2086 		if (transid == trans->transid) {
2087 			btrfs_commit_transaction(trans);
2088 		} else {
2089 			btrfs_end_transaction(trans);
2090 		}
2091 sleep:
2092 		wake_up_process(fs_info->cleaner_kthread);
2093 		mutex_unlock(&fs_info->transaction_kthread_mutex);
2094 
2095 		if (BTRFS_FS_ERROR(fs_info))
2096 			btrfs_cleanup_transaction(fs_info);
2097 		if (!kthread_should_stop() &&
2098 				(!btrfs_transaction_blocked(fs_info) ||
2099 				 cannot_commit))
2100 			schedule_timeout_interruptible(delay);
2101 	} while (!kthread_should_stop());
2102 	return 0;
2103 }
2104 
2105 /*
2106  * This will find the highest generation in the array of root backups.  The
2107  * index of the highest array is returned, or -EINVAL if we can't find
2108  * anything.
2109  *
2110  * We check to make sure the array is valid by comparing the
2111  * generation of the latest  root in the array with the generation
2112  * in the super block.  If they don't match we pitch it.
2113  */
2114 static int find_newest_super_backup(struct btrfs_fs_info *info)
2115 {
2116 	const u64 newest_gen = btrfs_super_generation(info->super_copy);
2117 	u64 cur;
2118 	struct btrfs_root_backup *root_backup;
2119 	int i;
2120 
2121 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2122 		root_backup = info->super_copy->super_roots + i;
2123 		cur = btrfs_backup_tree_root_gen(root_backup);
2124 		if (cur == newest_gen)
2125 			return i;
2126 	}
2127 
2128 	return -EINVAL;
2129 }
2130 
2131 /*
2132  * copy all the root pointers into the super backup array.
2133  * this will bump the backup pointer by one when it is
2134  * done
2135  */
2136 static void backup_super_roots(struct btrfs_fs_info *info)
2137 {
2138 	const int next_backup = info->backup_root_index;
2139 	struct btrfs_root_backup *root_backup;
2140 
2141 	root_backup = info->super_for_commit->super_roots + next_backup;
2142 
2143 	/*
2144 	 * make sure all of our padding and empty slots get zero filled
2145 	 * regardless of which ones we use today
2146 	 */
2147 	memset(root_backup, 0, sizeof(*root_backup));
2148 
2149 	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2150 
2151 	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2152 	btrfs_set_backup_tree_root_gen(root_backup,
2153 			       btrfs_header_generation(info->tree_root->node));
2154 
2155 	btrfs_set_backup_tree_root_level(root_backup,
2156 			       btrfs_header_level(info->tree_root->node));
2157 
2158 	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2159 	btrfs_set_backup_chunk_root_gen(root_backup,
2160 			       btrfs_header_generation(info->chunk_root->node));
2161 	btrfs_set_backup_chunk_root_level(root_backup,
2162 			       btrfs_header_level(info->chunk_root->node));
2163 
2164 	if (btrfs_fs_incompat(info, EXTENT_TREE_V2)) {
2165 		btrfs_set_backup_block_group_root(root_backup,
2166 					info->block_group_root->node->start);
2167 		btrfs_set_backup_block_group_root_gen(root_backup,
2168 			btrfs_header_generation(info->block_group_root->node));
2169 		btrfs_set_backup_block_group_root_level(root_backup,
2170 			btrfs_header_level(info->block_group_root->node));
2171 	} else {
2172 		struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
2173 		struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
2174 
2175 		btrfs_set_backup_extent_root(root_backup,
2176 					     extent_root->node->start);
2177 		btrfs_set_backup_extent_root_gen(root_backup,
2178 				btrfs_header_generation(extent_root->node));
2179 		btrfs_set_backup_extent_root_level(root_backup,
2180 					btrfs_header_level(extent_root->node));
2181 
2182 		btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
2183 		btrfs_set_backup_csum_root_gen(root_backup,
2184 					       btrfs_header_generation(csum_root->node));
2185 		btrfs_set_backup_csum_root_level(root_backup,
2186 						 btrfs_header_level(csum_root->node));
2187 	}
2188 
2189 	/*
2190 	 * we might commit during log recovery, which happens before we set
2191 	 * the fs_root.  Make sure it is valid before we fill it in.
2192 	 */
2193 	if (info->fs_root && info->fs_root->node) {
2194 		btrfs_set_backup_fs_root(root_backup,
2195 					 info->fs_root->node->start);
2196 		btrfs_set_backup_fs_root_gen(root_backup,
2197 			       btrfs_header_generation(info->fs_root->node));
2198 		btrfs_set_backup_fs_root_level(root_backup,
2199 			       btrfs_header_level(info->fs_root->node));
2200 	}
2201 
2202 	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2203 	btrfs_set_backup_dev_root_gen(root_backup,
2204 			       btrfs_header_generation(info->dev_root->node));
2205 	btrfs_set_backup_dev_root_level(root_backup,
2206 				       btrfs_header_level(info->dev_root->node));
2207 
2208 	btrfs_set_backup_total_bytes(root_backup,
2209 			     btrfs_super_total_bytes(info->super_copy));
2210 	btrfs_set_backup_bytes_used(root_backup,
2211 			     btrfs_super_bytes_used(info->super_copy));
2212 	btrfs_set_backup_num_devices(root_backup,
2213 			     btrfs_super_num_devices(info->super_copy));
2214 
2215 	/*
2216 	 * if we don't copy this out to the super_copy, it won't get remembered
2217 	 * for the next commit
2218 	 */
2219 	memcpy(&info->super_copy->super_roots,
2220 	       &info->super_for_commit->super_roots,
2221 	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2222 }
2223 
2224 /*
2225  * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2226  * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2227  *
2228  * fs_info - filesystem whose backup roots need to be read
2229  * priority - priority of backup root required
2230  *
2231  * Returns backup root index on success and -EINVAL otherwise.
2232  */
2233 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2234 {
2235 	int backup_index = find_newest_super_backup(fs_info);
2236 	struct btrfs_super_block *super = fs_info->super_copy;
2237 	struct btrfs_root_backup *root_backup;
2238 
2239 	if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2240 		if (priority == 0)
2241 			return backup_index;
2242 
2243 		backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2244 		backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2245 	} else {
2246 		return -EINVAL;
2247 	}
2248 
2249 	root_backup = super->super_roots + backup_index;
2250 
2251 	btrfs_set_super_generation(super,
2252 				   btrfs_backup_tree_root_gen(root_backup));
2253 	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2254 	btrfs_set_super_root_level(super,
2255 				   btrfs_backup_tree_root_level(root_backup));
2256 	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2257 
2258 	/*
2259 	 * Fixme: the total bytes and num_devices need to match or we should
2260 	 * need a fsck
2261 	 */
2262 	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2263 	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2264 
2265 	return backup_index;
2266 }
2267 
2268 /* helper to cleanup workers */
2269 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2270 {
2271 	btrfs_destroy_workqueue(fs_info->fixup_workers);
2272 	btrfs_destroy_workqueue(fs_info->delalloc_workers);
2273 	btrfs_destroy_workqueue(fs_info->hipri_workers);
2274 	btrfs_destroy_workqueue(fs_info->workers);
2275 	btrfs_destroy_workqueue(fs_info->endio_workers);
2276 	btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2277 	if (fs_info->rmw_workers)
2278 		destroy_workqueue(fs_info->rmw_workers);
2279 	btrfs_destroy_workqueue(fs_info->endio_write_workers);
2280 	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2281 	btrfs_destroy_workqueue(fs_info->delayed_workers);
2282 	btrfs_destroy_workqueue(fs_info->caching_workers);
2283 	btrfs_destroy_workqueue(fs_info->flush_workers);
2284 	btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2285 	if (fs_info->discard_ctl.discard_workers)
2286 		destroy_workqueue(fs_info->discard_ctl.discard_workers);
2287 	/*
2288 	 * Now that all other work queues are destroyed, we can safely destroy
2289 	 * the queues used for metadata I/O, since tasks from those other work
2290 	 * queues can do metadata I/O operations.
2291 	 */
2292 	btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2293 	btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2294 }
2295 
2296 static void free_root_extent_buffers(struct btrfs_root *root)
2297 {
2298 	if (root) {
2299 		free_extent_buffer(root->node);
2300 		free_extent_buffer(root->commit_root);
2301 		root->node = NULL;
2302 		root->commit_root = NULL;
2303 	}
2304 }
2305 
2306 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
2307 {
2308 	struct btrfs_root *root, *tmp;
2309 
2310 	rbtree_postorder_for_each_entry_safe(root, tmp,
2311 					     &fs_info->global_root_tree,
2312 					     rb_node)
2313 		free_root_extent_buffers(root);
2314 }
2315 
2316 /* helper to cleanup tree roots */
2317 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2318 {
2319 	free_root_extent_buffers(info->tree_root);
2320 
2321 	free_global_root_pointers(info);
2322 	free_root_extent_buffers(info->dev_root);
2323 	free_root_extent_buffers(info->quota_root);
2324 	free_root_extent_buffers(info->uuid_root);
2325 	free_root_extent_buffers(info->fs_root);
2326 	free_root_extent_buffers(info->data_reloc_root);
2327 	free_root_extent_buffers(info->block_group_root);
2328 	if (free_chunk_root)
2329 		free_root_extent_buffers(info->chunk_root);
2330 }
2331 
2332 void btrfs_put_root(struct btrfs_root *root)
2333 {
2334 	if (!root)
2335 		return;
2336 
2337 	if (refcount_dec_and_test(&root->refs)) {
2338 		WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2339 		WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2340 		if (root->anon_dev)
2341 			free_anon_bdev(root->anon_dev);
2342 		btrfs_drew_lock_destroy(&root->snapshot_lock);
2343 		free_root_extent_buffers(root);
2344 #ifdef CONFIG_BTRFS_DEBUG
2345 		spin_lock(&root->fs_info->fs_roots_lock);
2346 		list_del_init(&root->leak_list);
2347 		spin_unlock(&root->fs_info->fs_roots_lock);
2348 #endif
2349 		kfree(root);
2350 	}
2351 }
2352 
2353 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2354 {
2355 	struct btrfs_root *root;
2356 	unsigned long index = 0;
2357 
2358 	while (!list_empty(&fs_info->dead_roots)) {
2359 		root = list_entry(fs_info->dead_roots.next,
2360 				  struct btrfs_root, root_list);
2361 		list_del(&root->root_list);
2362 
2363 		if (test_bit(BTRFS_ROOT_REGISTERED, &root->state))
2364 			btrfs_drop_and_free_fs_root(fs_info, root);
2365 		btrfs_put_root(root);
2366 	}
2367 
2368 	xa_for_each(&fs_info->fs_roots, index, root) {
2369 		btrfs_drop_and_free_fs_root(fs_info, root);
2370 	}
2371 }
2372 
2373 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2374 {
2375 	mutex_init(&fs_info->scrub_lock);
2376 	atomic_set(&fs_info->scrubs_running, 0);
2377 	atomic_set(&fs_info->scrub_pause_req, 0);
2378 	atomic_set(&fs_info->scrubs_paused, 0);
2379 	atomic_set(&fs_info->scrub_cancel_req, 0);
2380 	init_waitqueue_head(&fs_info->scrub_pause_wait);
2381 	refcount_set(&fs_info->scrub_workers_refcnt, 0);
2382 }
2383 
2384 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2385 {
2386 	spin_lock_init(&fs_info->balance_lock);
2387 	mutex_init(&fs_info->balance_mutex);
2388 	atomic_set(&fs_info->balance_pause_req, 0);
2389 	atomic_set(&fs_info->balance_cancel_req, 0);
2390 	fs_info->balance_ctl = NULL;
2391 	init_waitqueue_head(&fs_info->balance_wait_q);
2392 	atomic_set(&fs_info->reloc_cancel_req, 0);
2393 }
2394 
2395 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2396 {
2397 	struct inode *inode = fs_info->btree_inode;
2398 
2399 	inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2400 	set_nlink(inode, 1);
2401 	/*
2402 	 * we set the i_size on the btree inode to the max possible int.
2403 	 * the real end of the address space is determined by all of
2404 	 * the devices in the system
2405 	 */
2406 	inode->i_size = OFFSET_MAX;
2407 	inode->i_mapping->a_ops = &btree_aops;
2408 
2409 	RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2410 	extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2411 			    IO_TREE_BTREE_INODE_IO, inode);
2412 	BTRFS_I(inode)->io_tree.track_uptodate = false;
2413 	extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2414 
2415 	BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2416 	memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2417 	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2418 	btrfs_insert_inode_hash(inode);
2419 }
2420 
2421 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2422 {
2423 	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2424 	init_rwsem(&fs_info->dev_replace.rwsem);
2425 	init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2426 }
2427 
2428 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2429 {
2430 	spin_lock_init(&fs_info->qgroup_lock);
2431 	mutex_init(&fs_info->qgroup_ioctl_lock);
2432 	fs_info->qgroup_tree = RB_ROOT;
2433 	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2434 	fs_info->qgroup_seq = 1;
2435 	fs_info->qgroup_ulist = NULL;
2436 	fs_info->qgroup_rescan_running = false;
2437 	mutex_init(&fs_info->qgroup_rescan_lock);
2438 }
2439 
2440 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
2441 {
2442 	u32 max_active = fs_info->thread_pool_size;
2443 	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2444 
2445 	fs_info->workers =
2446 		btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
2447 	fs_info->hipri_workers =
2448 		btrfs_alloc_workqueue(fs_info, "worker-high",
2449 				      flags | WQ_HIGHPRI, max_active, 16);
2450 
2451 	fs_info->delalloc_workers =
2452 		btrfs_alloc_workqueue(fs_info, "delalloc",
2453 				      flags, max_active, 2);
2454 
2455 	fs_info->flush_workers =
2456 		btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2457 				      flags, max_active, 0);
2458 
2459 	fs_info->caching_workers =
2460 		btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2461 
2462 	fs_info->fixup_workers =
2463 		btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2464 
2465 	/*
2466 	 * endios are largely parallel and should have a very
2467 	 * low idle thresh
2468 	 */
2469 	fs_info->endio_workers =
2470 		btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2471 	fs_info->endio_meta_workers =
2472 		btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2473 				      max_active, 4);
2474 	fs_info->endio_meta_write_workers =
2475 		btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2476 				      max_active, 2);
2477 	fs_info->endio_raid56_workers =
2478 		btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2479 				      max_active, 4);
2480 	fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
2481 	fs_info->endio_write_workers =
2482 		btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2483 				      max_active, 2);
2484 	fs_info->endio_freespace_worker =
2485 		btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2486 				      max_active, 0);
2487 	fs_info->delayed_workers =
2488 		btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2489 				      max_active, 0);
2490 	fs_info->qgroup_rescan_workers =
2491 		btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2492 	fs_info->discard_ctl.discard_workers =
2493 		alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2494 
2495 	if (!(fs_info->workers && fs_info->hipri_workers &&
2496 	      fs_info->delalloc_workers && fs_info->flush_workers &&
2497 	      fs_info->endio_workers && fs_info->endio_meta_workers &&
2498 	      fs_info->endio_meta_write_workers &&
2499 	      fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2500 	      fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2501 	      fs_info->caching_workers && fs_info->fixup_workers &&
2502 	      fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2503 	      fs_info->discard_ctl.discard_workers)) {
2504 		return -ENOMEM;
2505 	}
2506 
2507 	return 0;
2508 }
2509 
2510 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2511 {
2512 	struct crypto_shash *csum_shash;
2513 	const char *csum_driver = btrfs_super_csum_driver(csum_type);
2514 
2515 	csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2516 
2517 	if (IS_ERR(csum_shash)) {
2518 		btrfs_err(fs_info, "error allocating %s hash for checksum",
2519 			  csum_driver);
2520 		return PTR_ERR(csum_shash);
2521 	}
2522 
2523 	fs_info->csum_shash = csum_shash;
2524 
2525 	return 0;
2526 }
2527 
2528 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2529 			    struct btrfs_fs_devices *fs_devices)
2530 {
2531 	int ret;
2532 	struct btrfs_root *log_tree_root;
2533 	struct btrfs_super_block *disk_super = fs_info->super_copy;
2534 	u64 bytenr = btrfs_super_log_root(disk_super);
2535 	int level = btrfs_super_log_root_level(disk_super);
2536 
2537 	if (fs_devices->rw_devices == 0) {
2538 		btrfs_warn(fs_info, "log replay required on RO media");
2539 		return -EIO;
2540 	}
2541 
2542 	log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2543 					 GFP_KERNEL);
2544 	if (!log_tree_root)
2545 		return -ENOMEM;
2546 
2547 	log_tree_root->node = read_tree_block(fs_info, bytenr,
2548 					      BTRFS_TREE_LOG_OBJECTID,
2549 					      fs_info->generation + 1, level,
2550 					      NULL);
2551 	if (IS_ERR(log_tree_root->node)) {
2552 		btrfs_warn(fs_info, "failed to read log tree");
2553 		ret = PTR_ERR(log_tree_root->node);
2554 		log_tree_root->node = NULL;
2555 		btrfs_put_root(log_tree_root);
2556 		return ret;
2557 	}
2558 	if (!extent_buffer_uptodate(log_tree_root->node)) {
2559 		btrfs_err(fs_info, "failed to read log tree");
2560 		btrfs_put_root(log_tree_root);
2561 		return -EIO;
2562 	}
2563 
2564 	/* returns with log_tree_root freed on success */
2565 	ret = btrfs_recover_log_trees(log_tree_root);
2566 	if (ret) {
2567 		btrfs_handle_fs_error(fs_info, ret,
2568 				      "Failed to recover log tree");
2569 		btrfs_put_root(log_tree_root);
2570 		return ret;
2571 	}
2572 
2573 	if (sb_rdonly(fs_info->sb)) {
2574 		ret = btrfs_commit_super(fs_info);
2575 		if (ret)
2576 			return ret;
2577 	}
2578 
2579 	return 0;
2580 }
2581 
2582 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2583 				      struct btrfs_path *path, u64 objectid,
2584 				      const char *name)
2585 {
2586 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
2587 	struct btrfs_root *root;
2588 	u64 max_global_id = 0;
2589 	int ret;
2590 	struct btrfs_key key = {
2591 		.objectid = objectid,
2592 		.type = BTRFS_ROOT_ITEM_KEY,
2593 		.offset = 0,
2594 	};
2595 	bool found = false;
2596 
2597 	/* If we have IGNOREDATACSUMS skip loading these roots. */
2598 	if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2599 	    btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2600 		set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2601 		return 0;
2602 	}
2603 
2604 	while (1) {
2605 		ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2606 		if (ret < 0)
2607 			break;
2608 
2609 		if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2610 			ret = btrfs_next_leaf(tree_root, path);
2611 			if (ret) {
2612 				if (ret > 0)
2613 					ret = 0;
2614 				break;
2615 			}
2616 		}
2617 		ret = 0;
2618 
2619 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2620 		if (key.objectid != objectid)
2621 			break;
2622 		btrfs_release_path(path);
2623 
2624 		/*
2625 		 * Just worry about this for extent tree, it'll be the same for
2626 		 * everybody.
2627 		 */
2628 		if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2629 			max_global_id = max(max_global_id, key.offset);
2630 
2631 		found = true;
2632 		root = read_tree_root_path(tree_root, path, &key);
2633 		if (IS_ERR(root)) {
2634 			if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2635 				ret = PTR_ERR(root);
2636 			break;
2637 		}
2638 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2639 		ret = btrfs_global_root_insert(root);
2640 		if (ret) {
2641 			btrfs_put_root(root);
2642 			break;
2643 		}
2644 		key.offset++;
2645 	}
2646 	btrfs_release_path(path);
2647 
2648 	if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2649 		fs_info->nr_global_roots = max_global_id + 1;
2650 
2651 	if (!found || ret) {
2652 		if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2653 			set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2654 
2655 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2656 			ret = ret ? ret : -ENOENT;
2657 		else
2658 			ret = 0;
2659 		btrfs_err(fs_info, "failed to load root %s", name);
2660 	}
2661 	return ret;
2662 }
2663 
2664 static int load_global_roots(struct btrfs_root *tree_root)
2665 {
2666 	struct btrfs_path *path;
2667 	int ret = 0;
2668 
2669 	path = btrfs_alloc_path();
2670 	if (!path)
2671 		return -ENOMEM;
2672 
2673 	ret = load_global_roots_objectid(tree_root, path,
2674 					 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2675 	if (ret)
2676 		goto out;
2677 	ret = load_global_roots_objectid(tree_root, path,
2678 					 BTRFS_CSUM_TREE_OBJECTID, "csum");
2679 	if (ret)
2680 		goto out;
2681 	if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2682 		goto out;
2683 	ret = load_global_roots_objectid(tree_root, path,
2684 					 BTRFS_FREE_SPACE_TREE_OBJECTID,
2685 					 "free space");
2686 out:
2687 	btrfs_free_path(path);
2688 	return ret;
2689 }
2690 
2691 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2692 {
2693 	struct btrfs_root *tree_root = fs_info->tree_root;
2694 	struct btrfs_root *root;
2695 	struct btrfs_key location;
2696 	int ret;
2697 
2698 	BUG_ON(!fs_info->tree_root);
2699 
2700 	ret = load_global_roots(tree_root);
2701 	if (ret)
2702 		return ret;
2703 
2704 	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2705 	location.type = BTRFS_ROOT_ITEM_KEY;
2706 	location.offset = 0;
2707 
2708 	root = btrfs_read_tree_root(tree_root, &location);
2709 	if (IS_ERR(root)) {
2710 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2711 			ret = PTR_ERR(root);
2712 			goto out;
2713 		}
2714 	} else {
2715 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2716 		fs_info->dev_root = root;
2717 	}
2718 	/* Initialize fs_info for all devices in any case */
2719 	btrfs_init_devices_late(fs_info);
2720 
2721 	/*
2722 	 * This tree can share blocks with some other fs tree during relocation
2723 	 * and we need a proper setup by btrfs_get_fs_root
2724 	 */
2725 	root = btrfs_get_fs_root(tree_root->fs_info,
2726 				 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2727 	if (IS_ERR(root)) {
2728 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2729 			ret = PTR_ERR(root);
2730 			goto out;
2731 		}
2732 	} else {
2733 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2734 		fs_info->data_reloc_root = root;
2735 	}
2736 
2737 	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2738 	root = btrfs_read_tree_root(tree_root, &location);
2739 	if (!IS_ERR(root)) {
2740 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2741 		set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2742 		fs_info->quota_root = root;
2743 	}
2744 
2745 	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2746 	root = btrfs_read_tree_root(tree_root, &location);
2747 	if (IS_ERR(root)) {
2748 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2749 			ret = PTR_ERR(root);
2750 			if (ret != -ENOENT)
2751 				goto out;
2752 		}
2753 	} else {
2754 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2755 		fs_info->uuid_root = root;
2756 	}
2757 
2758 	return 0;
2759 out:
2760 	btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2761 		   location.objectid, ret);
2762 	return ret;
2763 }
2764 
2765 /*
2766  * Real super block validation
2767  * NOTE: super csum type and incompat features will not be checked here.
2768  *
2769  * @sb:		super block to check
2770  * @mirror_num:	the super block number to check its bytenr:
2771  * 		0	the primary (1st) sb
2772  * 		1, 2	2nd and 3rd backup copy
2773  * 	       -1	skip bytenr check
2774  */
2775 static int validate_super(struct btrfs_fs_info *fs_info,
2776 			    struct btrfs_super_block *sb, int mirror_num)
2777 {
2778 	u64 nodesize = btrfs_super_nodesize(sb);
2779 	u64 sectorsize = btrfs_super_sectorsize(sb);
2780 	int ret = 0;
2781 
2782 	if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2783 		btrfs_err(fs_info, "no valid FS found");
2784 		ret = -EINVAL;
2785 	}
2786 	if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2787 		btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2788 				btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2789 		ret = -EINVAL;
2790 	}
2791 	if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2792 		btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2793 				btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2794 		ret = -EINVAL;
2795 	}
2796 	if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2797 		btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2798 				btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2799 		ret = -EINVAL;
2800 	}
2801 	if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2802 		btrfs_err(fs_info, "log_root level too big: %d >= %d",
2803 				btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2804 		ret = -EINVAL;
2805 	}
2806 
2807 	/*
2808 	 * Check sectorsize and nodesize first, other check will need it.
2809 	 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2810 	 */
2811 	if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2812 	    sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2813 		btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2814 		ret = -EINVAL;
2815 	}
2816 
2817 	/*
2818 	 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2819 	 *
2820 	 * We can support 16K sectorsize with 64K page size without problem,
2821 	 * but such sectorsize/pagesize combination doesn't make much sense.
2822 	 * 4K will be our future standard, PAGE_SIZE is supported from the very
2823 	 * beginning.
2824 	 */
2825 	if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2826 		btrfs_err(fs_info,
2827 			"sectorsize %llu not yet supported for page size %lu",
2828 			sectorsize, PAGE_SIZE);
2829 		ret = -EINVAL;
2830 	}
2831 
2832 	if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2833 	    nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2834 		btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2835 		ret = -EINVAL;
2836 	}
2837 	if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2838 		btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2839 			  le32_to_cpu(sb->__unused_leafsize), nodesize);
2840 		ret = -EINVAL;
2841 	}
2842 
2843 	/* Root alignment check */
2844 	if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2845 		btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2846 			   btrfs_super_root(sb));
2847 		ret = -EINVAL;
2848 	}
2849 	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2850 		btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2851 			   btrfs_super_chunk_root(sb));
2852 		ret = -EINVAL;
2853 	}
2854 	if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2855 		btrfs_warn(fs_info, "log_root block unaligned: %llu",
2856 			   btrfs_super_log_root(sb));
2857 		ret = -EINVAL;
2858 	}
2859 
2860 	if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2861 		   BTRFS_FSID_SIZE)) {
2862 		btrfs_err(fs_info,
2863 		"superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2864 			fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2865 		ret = -EINVAL;
2866 	}
2867 
2868 	if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2869 	    memcmp(fs_info->fs_devices->metadata_uuid,
2870 		   fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2871 		btrfs_err(fs_info,
2872 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2873 			fs_info->super_copy->metadata_uuid,
2874 			fs_info->fs_devices->metadata_uuid);
2875 		ret = -EINVAL;
2876 	}
2877 
2878 	if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2879 		   BTRFS_FSID_SIZE) != 0) {
2880 		btrfs_err(fs_info,
2881 			"dev_item UUID does not match metadata fsid: %pU != %pU",
2882 			fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2883 		ret = -EINVAL;
2884 	}
2885 
2886 	/*
2887 	 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2888 	 * done later
2889 	 */
2890 	if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2891 		btrfs_err(fs_info, "bytes_used is too small %llu",
2892 			  btrfs_super_bytes_used(sb));
2893 		ret = -EINVAL;
2894 	}
2895 	if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2896 		btrfs_err(fs_info, "invalid stripesize %u",
2897 			  btrfs_super_stripesize(sb));
2898 		ret = -EINVAL;
2899 	}
2900 	if (btrfs_super_num_devices(sb) > (1UL << 31))
2901 		btrfs_warn(fs_info, "suspicious number of devices: %llu",
2902 			   btrfs_super_num_devices(sb));
2903 	if (btrfs_super_num_devices(sb) == 0) {
2904 		btrfs_err(fs_info, "number of devices is 0");
2905 		ret = -EINVAL;
2906 	}
2907 
2908 	if (mirror_num >= 0 &&
2909 	    btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2910 		btrfs_err(fs_info, "super offset mismatch %llu != %u",
2911 			  btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2912 		ret = -EINVAL;
2913 	}
2914 
2915 	/*
2916 	 * Obvious sys_chunk_array corruptions, it must hold at least one key
2917 	 * and one chunk
2918 	 */
2919 	if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2920 		btrfs_err(fs_info, "system chunk array too big %u > %u",
2921 			  btrfs_super_sys_array_size(sb),
2922 			  BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2923 		ret = -EINVAL;
2924 	}
2925 	if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2926 			+ sizeof(struct btrfs_chunk)) {
2927 		btrfs_err(fs_info, "system chunk array too small %u < %zu",
2928 			  btrfs_super_sys_array_size(sb),
2929 			  sizeof(struct btrfs_disk_key)
2930 			  + sizeof(struct btrfs_chunk));
2931 		ret = -EINVAL;
2932 	}
2933 
2934 	/*
2935 	 * The generation is a global counter, we'll trust it more than the others
2936 	 * but it's still possible that it's the one that's wrong.
2937 	 */
2938 	if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2939 		btrfs_warn(fs_info,
2940 			"suspicious: generation < chunk_root_generation: %llu < %llu",
2941 			btrfs_super_generation(sb),
2942 			btrfs_super_chunk_root_generation(sb));
2943 	if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2944 	    && btrfs_super_cache_generation(sb) != (u64)-1)
2945 		btrfs_warn(fs_info,
2946 			"suspicious: generation < cache_generation: %llu < %llu",
2947 			btrfs_super_generation(sb),
2948 			btrfs_super_cache_generation(sb));
2949 
2950 	return ret;
2951 }
2952 
2953 /*
2954  * Validation of super block at mount time.
2955  * Some checks already done early at mount time, like csum type and incompat
2956  * flags will be skipped.
2957  */
2958 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2959 {
2960 	return validate_super(fs_info, fs_info->super_copy, 0);
2961 }
2962 
2963 /*
2964  * Validation of super block at write time.
2965  * Some checks like bytenr check will be skipped as their values will be
2966  * overwritten soon.
2967  * Extra checks like csum type and incompat flags will be done here.
2968  */
2969 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2970 				      struct btrfs_super_block *sb)
2971 {
2972 	int ret;
2973 
2974 	ret = validate_super(fs_info, sb, -1);
2975 	if (ret < 0)
2976 		goto out;
2977 	if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2978 		ret = -EUCLEAN;
2979 		btrfs_err(fs_info, "invalid csum type, has %u want %u",
2980 			  btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2981 		goto out;
2982 	}
2983 	if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2984 		ret = -EUCLEAN;
2985 		btrfs_err(fs_info,
2986 		"invalid incompat flags, has 0x%llx valid mask 0x%llx",
2987 			  btrfs_super_incompat_flags(sb),
2988 			  (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2989 		goto out;
2990 	}
2991 out:
2992 	if (ret < 0)
2993 		btrfs_err(fs_info,
2994 		"super block corruption detected before writing it to disk");
2995 	return ret;
2996 }
2997 
2998 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2999 {
3000 	int ret = 0;
3001 
3002 	root->node = read_tree_block(root->fs_info, bytenr,
3003 				     root->root_key.objectid, gen, level, NULL);
3004 	if (IS_ERR(root->node)) {
3005 		ret = PTR_ERR(root->node);
3006 		root->node = NULL;
3007 		return ret;
3008 	}
3009 	if (!extent_buffer_uptodate(root->node)) {
3010 		free_extent_buffer(root->node);
3011 		root->node = NULL;
3012 		return -EIO;
3013 	}
3014 
3015 	btrfs_set_root_node(&root->root_item, root->node);
3016 	root->commit_root = btrfs_root_node(root);
3017 	btrfs_set_root_refs(&root->root_item, 1);
3018 	return ret;
3019 }
3020 
3021 static int load_important_roots(struct btrfs_fs_info *fs_info)
3022 {
3023 	struct btrfs_super_block *sb = fs_info->super_copy;
3024 	u64 gen, bytenr;
3025 	int level, ret;
3026 
3027 	bytenr = btrfs_super_root(sb);
3028 	gen = btrfs_super_generation(sb);
3029 	level = btrfs_super_root_level(sb);
3030 	ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
3031 	if (ret) {
3032 		btrfs_warn(fs_info, "couldn't read tree root");
3033 		return ret;
3034 	}
3035 
3036 	if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
3037 		return 0;
3038 
3039 	bytenr = btrfs_super_block_group_root(sb);
3040 	gen = btrfs_super_block_group_root_generation(sb);
3041 	level = btrfs_super_block_group_root_level(sb);
3042 	ret = load_super_root(fs_info->block_group_root, bytenr, gen, level);
3043 	if (ret)
3044 		btrfs_warn(fs_info, "couldn't read block group root");
3045 	return ret;
3046 }
3047 
3048 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
3049 {
3050 	int backup_index = find_newest_super_backup(fs_info);
3051 	struct btrfs_super_block *sb = fs_info->super_copy;
3052 	struct btrfs_root *tree_root = fs_info->tree_root;
3053 	bool handle_error = false;
3054 	int ret = 0;
3055 	int i;
3056 
3057 	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3058 		struct btrfs_root *root;
3059 
3060 		root = btrfs_alloc_root(fs_info, BTRFS_BLOCK_GROUP_TREE_OBJECTID,
3061 					GFP_KERNEL);
3062 		if (!root)
3063 			return -ENOMEM;
3064 		fs_info->block_group_root = root;
3065 	}
3066 
3067 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
3068 		if (handle_error) {
3069 			if (!IS_ERR(tree_root->node))
3070 				free_extent_buffer(tree_root->node);
3071 			tree_root->node = NULL;
3072 
3073 			if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3074 				break;
3075 
3076 			free_root_pointers(fs_info, 0);
3077 
3078 			/*
3079 			 * Don't use the log in recovery mode, it won't be
3080 			 * valid
3081 			 */
3082 			btrfs_set_super_log_root(sb, 0);
3083 
3084 			/* We can't trust the free space cache either */
3085 			btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3086 
3087 			ret = read_backup_root(fs_info, i);
3088 			backup_index = ret;
3089 			if (ret < 0)
3090 				return ret;
3091 		}
3092 
3093 		ret = load_important_roots(fs_info);
3094 		if (ret) {
3095 			handle_error = true;
3096 			continue;
3097 		}
3098 
3099 		/*
3100 		 * No need to hold btrfs_root::objectid_mutex since the fs
3101 		 * hasn't been fully initialised and we are the only user
3102 		 */
3103 		ret = btrfs_init_root_free_objectid(tree_root);
3104 		if (ret < 0) {
3105 			handle_error = true;
3106 			continue;
3107 		}
3108 
3109 		ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
3110 
3111 		ret = btrfs_read_roots(fs_info);
3112 		if (ret < 0) {
3113 			handle_error = true;
3114 			continue;
3115 		}
3116 
3117 		/* All successful */
3118 		fs_info->generation = btrfs_header_generation(tree_root->node);
3119 		fs_info->last_trans_committed = fs_info->generation;
3120 		fs_info->last_reloc_trans = 0;
3121 
3122 		/* Always begin writing backup roots after the one being used */
3123 		if (backup_index < 0) {
3124 			fs_info->backup_root_index = 0;
3125 		} else {
3126 			fs_info->backup_root_index = backup_index + 1;
3127 			fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
3128 		}
3129 		break;
3130 	}
3131 
3132 	return ret;
3133 }
3134 
3135 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
3136 {
3137 	xa_init_flags(&fs_info->fs_roots, GFP_ATOMIC);
3138 	xa_init_flags(&fs_info->extent_buffers, GFP_ATOMIC);
3139 	INIT_LIST_HEAD(&fs_info->trans_list);
3140 	INIT_LIST_HEAD(&fs_info->dead_roots);
3141 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
3142 	INIT_LIST_HEAD(&fs_info->delalloc_roots);
3143 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
3144 	spin_lock_init(&fs_info->delalloc_root_lock);
3145 	spin_lock_init(&fs_info->trans_lock);
3146 	spin_lock_init(&fs_info->fs_roots_lock);
3147 	spin_lock_init(&fs_info->delayed_iput_lock);
3148 	spin_lock_init(&fs_info->defrag_inodes_lock);
3149 	spin_lock_init(&fs_info->super_lock);
3150 	spin_lock_init(&fs_info->buffer_lock);
3151 	spin_lock_init(&fs_info->unused_bgs_lock);
3152 	spin_lock_init(&fs_info->treelog_bg_lock);
3153 	spin_lock_init(&fs_info->zone_active_bgs_lock);
3154 	spin_lock_init(&fs_info->relocation_bg_lock);
3155 	rwlock_init(&fs_info->tree_mod_log_lock);
3156 	rwlock_init(&fs_info->global_root_lock);
3157 	mutex_init(&fs_info->unused_bg_unpin_mutex);
3158 	mutex_init(&fs_info->reclaim_bgs_lock);
3159 	mutex_init(&fs_info->reloc_mutex);
3160 	mutex_init(&fs_info->delalloc_root_mutex);
3161 	mutex_init(&fs_info->zoned_meta_io_lock);
3162 	mutex_init(&fs_info->zoned_data_reloc_io_lock);
3163 	seqlock_init(&fs_info->profiles_lock);
3164 
3165 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
3166 	INIT_LIST_HEAD(&fs_info->space_info);
3167 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
3168 	INIT_LIST_HEAD(&fs_info->unused_bgs);
3169 	INIT_LIST_HEAD(&fs_info->reclaim_bgs);
3170 	INIT_LIST_HEAD(&fs_info->zone_active_bgs);
3171 #ifdef CONFIG_BTRFS_DEBUG
3172 	INIT_LIST_HEAD(&fs_info->allocated_roots);
3173 	INIT_LIST_HEAD(&fs_info->allocated_ebs);
3174 	spin_lock_init(&fs_info->eb_leak_lock);
3175 #endif
3176 	extent_map_tree_init(&fs_info->mapping_tree);
3177 	btrfs_init_block_rsv(&fs_info->global_block_rsv,
3178 			     BTRFS_BLOCK_RSV_GLOBAL);
3179 	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
3180 	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
3181 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
3182 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
3183 			     BTRFS_BLOCK_RSV_DELOPS);
3184 	btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
3185 			     BTRFS_BLOCK_RSV_DELREFS);
3186 
3187 	atomic_set(&fs_info->async_delalloc_pages, 0);
3188 	atomic_set(&fs_info->defrag_running, 0);
3189 	atomic_set(&fs_info->nr_delayed_iputs, 0);
3190 	atomic64_set(&fs_info->tree_mod_seq, 0);
3191 	fs_info->global_root_tree = RB_ROOT;
3192 	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
3193 	fs_info->metadata_ratio = 0;
3194 	fs_info->defrag_inodes = RB_ROOT;
3195 	atomic64_set(&fs_info->free_chunk_space, 0);
3196 	fs_info->tree_mod_log = RB_ROOT;
3197 	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
3198 	fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
3199 	btrfs_init_ref_verify(fs_info);
3200 
3201 	fs_info->thread_pool_size = min_t(unsigned long,
3202 					  num_online_cpus() + 2, 8);
3203 
3204 	INIT_LIST_HEAD(&fs_info->ordered_roots);
3205 	spin_lock_init(&fs_info->ordered_root_lock);
3206 
3207 	btrfs_init_scrub(fs_info);
3208 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3209 	fs_info->check_integrity_print_mask = 0;
3210 #endif
3211 	btrfs_init_balance(fs_info);
3212 	btrfs_init_async_reclaim_work(fs_info);
3213 
3214 	rwlock_init(&fs_info->block_group_cache_lock);
3215 	fs_info->block_group_cache_tree = RB_ROOT_CACHED;
3216 
3217 	extent_io_tree_init(fs_info, &fs_info->excluded_extents,
3218 			    IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
3219 
3220 	mutex_init(&fs_info->ordered_operations_mutex);
3221 	mutex_init(&fs_info->tree_log_mutex);
3222 	mutex_init(&fs_info->chunk_mutex);
3223 	mutex_init(&fs_info->transaction_kthread_mutex);
3224 	mutex_init(&fs_info->cleaner_mutex);
3225 	mutex_init(&fs_info->ro_block_group_mutex);
3226 	init_rwsem(&fs_info->commit_root_sem);
3227 	init_rwsem(&fs_info->cleanup_work_sem);
3228 	init_rwsem(&fs_info->subvol_sem);
3229 	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
3230 
3231 	btrfs_init_dev_replace_locks(fs_info);
3232 	btrfs_init_qgroup(fs_info);
3233 	btrfs_discard_init(fs_info);
3234 
3235 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
3236 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
3237 
3238 	init_waitqueue_head(&fs_info->transaction_throttle);
3239 	init_waitqueue_head(&fs_info->transaction_wait);
3240 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
3241 	init_waitqueue_head(&fs_info->async_submit_wait);
3242 	init_waitqueue_head(&fs_info->delayed_iputs_wait);
3243 
3244 	/* Usable values until the real ones are cached from the superblock */
3245 	fs_info->nodesize = 4096;
3246 	fs_info->sectorsize = 4096;
3247 	fs_info->sectorsize_bits = ilog2(4096);
3248 	fs_info->stripesize = 4096;
3249 
3250 	spin_lock_init(&fs_info->swapfile_pins_lock);
3251 	fs_info->swapfile_pins = RB_ROOT;
3252 
3253 	fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
3254 	INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
3255 }
3256 
3257 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
3258 {
3259 	int ret;
3260 
3261 	fs_info->sb = sb;
3262 	sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
3263 	sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3264 
3265 	ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3266 	if (ret)
3267 		return ret;
3268 
3269 	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3270 	if (ret)
3271 		return ret;
3272 
3273 	fs_info->dirty_metadata_batch = PAGE_SIZE *
3274 					(1 + ilog2(nr_cpu_ids));
3275 
3276 	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3277 	if (ret)
3278 		return ret;
3279 
3280 	ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3281 			GFP_KERNEL);
3282 	if (ret)
3283 		return ret;
3284 
3285 	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3286 					GFP_KERNEL);
3287 	if (!fs_info->delayed_root)
3288 		return -ENOMEM;
3289 	btrfs_init_delayed_root(fs_info->delayed_root);
3290 
3291 	if (sb_rdonly(sb))
3292 		set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3293 
3294 	return btrfs_alloc_stripe_hash_table(fs_info);
3295 }
3296 
3297 static int btrfs_uuid_rescan_kthread(void *data)
3298 {
3299 	struct btrfs_fs_info *fs_info = data;
3300 	int ret;
3301 
3302 	/*
3303 	 * 1st step is to iterate through the existing UUID tree and
3304 	 * to delete all entries that contain outdated data.
3305 	 * 2nd step is to add all missing entries to the UUID tree.
3306 	 */
3307 	ret = btrfs_uuid_tree_iterate(fs_info);
3308 	if (ret < 0) {
3309 		if (ret != -EINTR)
3310 			btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3311 				   ret);
3312 		up(&fs_info->uuid_tree_rescan_sem);
3313 		return ret;
3314 	}
3315 	return btrfs_uuid_scan_kthread(data);
3316 }
3317 
3318 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3319 {
3320 	struct task_struct *task;
3321 
3322 	down(&fs_info->uuid_tree_rescan_sem);
3323 	task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3324 	if (IS_ERR(task)) {
3325 		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
3326 		btrfs_warn(fs_info, "failed to start uuid_rescan task");
3327 		up(&fs_info->uuid_tree_rescan_sem);
3328 		return PTR_ERR(task);
3329 	}
3330 
3331 	return 0;
3332 }
3333 
3334 /*
3335  * Some options only have meaning at mount time and shouldn't persist across
3336  * remounts, or be displayed. Clear these at the end of mount and remount
3337  * code paths.
3338  */
3339 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3340 {
3341 	btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3342 	btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3343 }
3344 
3345 /*
3346  * Mounting logic specific to read-write file systems. Shared by open_ctree
3347  * and btrfs_remount when remounting from read-only to read-write.
3348  */
3349 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3350 {
3351 	int ret;
3352 	const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3353 	bool clear_free_space_tree = false;
3354 
3355 	if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3356 	    btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3357 		clear_free_space_tree = true;
3358 	} else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3359 		   !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3360 		btrfs_warn(fs_info, "free space tree is invalid");
3361 		clear_free_space_tree = true;
3362 	}
3363 
3364 	if (clear_free_space_tree) {
3365 		btrfs_info(fs_info, "clearing free space tree");
3366 		ret = btrfs_clear_free_space_tree(fs_info);
3367 		if (ret) {
3368 			btrfs_warn(fs_info,
3369 				   "failed to clear free space tree: %d", ret);
3370 			goto out;
3371 		}
3372 	}
3373 
3374 	/*
3375 	 * btrfs_find_orphan_roots() is responsible for finding all the dead
3376 	 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3377 	 * them into the fs_info->fs_roots. This must be done before
3378 	 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3379 	 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3380 	 * item before the root's tree is deleted - this means that if we unmount
3381 	 * or crash before the deletion completes, on the next mount we will not
3382 	 * delete what remains of the tree because the orphan item does not
3383 	 * exists anymore, which is what tells us we have a pending deletion.
3384 	 */
3385 	ret = btrfs_find_orphan_roots(fs_info);
3386 	if (ret)
3387 		goto out;
3388 
3389 	ret = btrfs_cleanup_fs_roots(fs_info);
3390 	if (ret)
3391 		goto out;
3392 
3393 	down_read(&fs_info->cleanup_work_sem);
3394 	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3395 	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3396 		up_read(&fs_info->cleanup_work_sem);
3397 		goto out;
3398 	}
3399 	up_read(&fs_info->cleanup_work_sem);
3400 
3401 	mutex_lock(&fs_info->cleaner_mutex);
3402 	ret = btrfs_recover_relocation(fs_info);
3403 	mutex_unlock(&fs_info->cleaner_mutex);
3404 	if (ret < 0) {
3405 		btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3406 		goto out;
3407 	}
3408 
3409 	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3410 	    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3411 		btrfs_info(fs_info, "creating free space tree");
3412 		ret = btrfs_create_free_space_tree(fs_info);
3413 		if (ret) {
3414 			btrfs_warn(fs_info,
3415 				"failed to create free space tree: %d", ret);
3416 			goto out;
3417 		}
3418 	}
3419 
3420 	if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3421 		ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3422 		if (ret)
3423 			goto out;
3424 	}
3425 
3426 	ret = btrfs_resume_balance_async(fs_info);
3427 	if (ret)
3428 		goto out;
3429 
3430 	ret = btrfs_resume_dev_replace_async(fs_info);
3431 	if (ret) {
3432 		btrfs_warn(fs_info, "failed to resume dev_replace");
3433 		goto out;
3434 	}
3435 
3436 	btrfs_qgroup_rescan_resume(fs_info);
3437 
3438 	if (!fs_info->uuid_root) {
3439 		btrfs_info(fs_info, "creating UUID tree");
3440 		ret = btrfs_create_uuid_tree(fs_info);
3441 		if (ret) {
3442 			btrfs_warn(fs_info,
3443 				   "failed to create the UUID tree %d", ret);
3444 			goto out;
3445 		}
3446 	}
3447 
3448 out:
3449 	return ret;
3450 }
3451 
3452 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3453 		      char *options)
3454 {
3455 	u32 sectorsize;
3456 	u32 nodesize;
3457 	u32 stripesize;
3458 	u64 generation;
3459 	u64 features;
3460 	u16 csum_type;
3461 	struct btrfs_super_block *disk_super;
3462 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3463 	struct btrfs_root *tree_root;
3464 	struct btrfs_root *chunk_root;
3465 	int ret;
3466 	int err = -EINVAL;
3467 	int level;
3468 
3469 	ret = init_mount_fs_info(fs_info, sb);
3470 	if (ret) {
3471 		err = ret;
3472 		goto fail;
3473 	}
3474 
3475 	/* These need to be init'ed before we start creating inodes and such. */
3476 	tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3477 				     GFP_KERNEL);
3478 	fs_info->tree_root = tree_root;
3479 	chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3480 				      GFP_KERNEL);
3481 	fs_info->chunk_root = chunk_root;
3482 	if (!tree_root || !chunk_root) {
3483 		err = -ENOMEM;
3484 		goto fail;
3485 	}
3486 
3487 	fs_info->btree_inode = new_inode(sb);
3488 	if (!fs_info->btree_inode) {
3489 		err = -ENOMEM;
3490 		goto fail;
3491 	}
3492 	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3493 	btrfs_init_btree_inode(fs_info);
3494 
3495 	invalidate_bdev(fs_devices->latest_dev->bdev);
3496 
3497 	/*
3498 	 * Read super block and check the signature bytes only
3499 	 */
3500 	disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3501 	if (IS_ERR(disk_super)) {
3502 		err = PTR_ERR(disk_super);
3503 		goto fail_alloc;
3504 	}
3505 
3506 	/*
3507 	 * Verify the type first, if that or the checksum value are
3508 	 * corrupted, we'll find out
3509 	 */
3510 	csum_type = btrfs_super_csum_type(disk_super);
3511 	if (!btrfs_supported_super_csum(csum_type)) {
3512 		btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3513 			  csum_type);
3514 		err = -EINVAL;
3515 		btrfs_release_disk_super(disk_super);
3516 		goto fail_alloc;
3517 	}
3518 
3519 	fs_info->csum_size = btrfs_super_csum_size(disk_super);
3520 
3521 	ret = btrfs_init_csum_hash(fs_info, csum_type);
3522 	if (ret) {
3523 		err = ret;
3524 		btrfs_release_disk_super(disk_super);
3525 		goto fail_alloc;
3526 	}
3527 
3528 	/*
3529 	 * We want to check superblock checksum, the type is stored inside.
3530 	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3531 	 */
3532 	if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
3533 		btrfs_err(fs_info, "superblock checksum mismatch");
3534 		err = -EINVAL;
3535 		btrfs_release_disk_super(disk_super);
3536 		goto fail_alloc;
3537 	}
3538 
3539 	/*
3540 	 * super_copy is zeroed at allocation time and we never touch the
3541 	 * following bytes up to INFO_SIZE, the checksum is calculated from
3542 	 * the whole block of INFO_SIZE
3543 	 */
3544 	memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3545 	btrfs_release_disk_super(disk_super);
3546 
3547 	disk_super = fs_info->super_copy;
3548 
3549 
3550 	features = btrfs_super_flags(disk_super);
3551 	if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3552 		features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3553 		btrfs_set_super_flags(disk_super, features);
3554 		btrfs_info(fs_info,
3555 			"found metadata UUID change in progress flag, clearing");
3556 	}
3557 
3558 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
3559 	       sizeof(*fs_info->super_for_commit));
3560 
3561 	ret = btrfs_validate_mount_super(fs_info);
3562 	if (ret) {
3563 		btrfs_err(fs_info, "superblock contains fatal errors");
3564 		err = -EINVAL;
3565 		goto fail_alloc;
3566 	}
3567 
3568 	if (!btrfs_super_root(disk_super))
3569 		goto fail_alloc;
3570 
3571 	/* check FS state, whether FS is broken. */
3572 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3573 		set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3574 
3575 	/*
3576 	 * In the long term, we'll store the compression type in the super
3577 	 * block, and it'll be used for per file compression control.
3578 	 */
3579 	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3580 
3581 	/*
3582 	 * Flag our filesystem as having big metadata blocks if they are bigger
3583 	 * than the page size.
3584 	 */
3585 	if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3586 		if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3587 			btrfs_info(fs_info,
3588 				"flagging fs with big metadata feature");
3589 		features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3590 	}
3591 
3592 	/* Set up fs_info before parsing mount options */
3593 	nodesize = btrfs_super_nodesize(disk_super);
3594 	sectorsize = btrfs_super_sectorsize(disk_super);
3595 	stripesize = sectorsize;
3596 	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3597 	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3598 
3599 	fs_info->nodesize = nodesize;
3600 	fs_info->sectorsize = sectorsize;
3601 	fs_info->sectorsize_bits = ilog2(sectorsize);
3602 	fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3603 	fs_info->stripesize = stripesize;
3604 
3605 	ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3606 	if (ret) {
3607 		err = ret;
3608 		goto fail_alloc;
3609 	}
3610 
3611 	features = btrfs_super_incompat_flags(disk_super) &
3612 		~BTRFS_FEATURE_INCOMPAT_SUPP;
3613 	if (features) {
3614 		btrfs_err(fs_info,
3615 		    "cannot mount because of unsupported optional features (0x%llx)",
3616 		    features);
3617 		err = -EINVAL;
3618 		goto fail_alloc;
3619 	}
3620 
3621 	features = btrfs_super_incompat_flags(disk_super);
3622 	features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3623 	if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3624 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3625 	else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3626 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3627 
3628 	if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3629 		btrfs_info(fs_info, "has skinny extents");
3630 
3631 	/*
3632 	 * mixed block groups end up with duplicate but slightly offset
3633 	 * extent buffers for the same range.  It leads to corruptions
3634 	 */
3635 	if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3636 	    (sectorsize != nodesize)) {
3637 		btrfs_err(fs_info,
3638 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3639 			nodesize, sectorsize);
3640 		goto fail_alloc;
3641 	}
3642 
3643 	/*
3644 	 * Needn't use the lock because there is no other task which will
3645 	 * update the flag.
3646 	 */
3647 	btrfs_set_super_incompat_flags(disk_super, features);
3648 
3649 	features = btrfs_super_compat_ro_flags(disk_super) &
3650 		~BTRFS_FEATURE_COMPAT_RO_SUPP;
3651 	if (!sb_rdonly(sb) && features) {
3652 		btrfs_err(fs_info,
3653 	"cannot mount read-write because of unsupported optional features (0x%llx)",
3654 		       features);
3655 		err = -EINVAL;
3656 		goto fail_alloc;
3657 	}
3658 
3659 	if (sectorsize < PAGE_SIZE) {
3660 		struct btrfs_subpage_info *subpage_info;
3661 
3662 		/*
3663 		 * V1 space cache has some hardcoded PAGE_SIZE usage, and is
3664 		 * going to be deprecated.
3665 		 *
3666 		 * Force to use v2 cache for subpage case.
3667 		 */
3668 		btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
3669 		btrfs_set_and_info(fs_info, FREE_SPACE_TREE,
3670 			"forcing free space tree for sector size %u with page size %lu",
3671 			sectorsize, PAGE_SIZE);
3672 
3673 		btrfs_warn(fs_info,
3674 		"read-write for sector size %u with page size %lu is experimental",
3675 			   sectorsize, PAGE_SIZE);
3676 		subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3677 		if (!subpage_info)
3678 			goto fail_alloc;
3679 		btrfs_init_subpage_info(subpage_info, sectorsize);
3680 		fs_info->subpage_info = subpage_info;
3681 	}
3682 
3683 	ret = btrfs_init_workqueues(fs_info);
3684 	if (ret) {
3685 		err = ret;
3686 		goto fail_sb_buffer;
3687 	}
3688 
3689 	sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3690 	sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3691 
3692 	sb->s_blocksize = sectorsize;
3693 	sb->s_blocksize_bits = blksize_bits(sectorsize);
3694 	memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3695 
3696 	mutex_lock(&fs_info->chunk_mutex);
3697 	ret = btrfs_read_sys_array(fs_info);
3698 	mutex_unlock(&fs_info->chunk_mutex);
3699 	if (ret) {
3700 		btrfs_err(fs_info, "failed to read the system array: %d", ret);
3701 		goto fail_sb_buffer;
3702 	}
3703 
3704 	generation = btrfs_super_chunk_root_generation(disk_super);
3705 	level = btrfs_super_chunk_root_level(disk_super);
3706 	ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3707 			      generation, level);
3708 	if (ret) {
3709 		btrfs_err(fs_info, "failed to read chunk root");
3710 		goto fail_tree_roots;
3711 	}
3712 
3713 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3714 			   offsetof(struct btrfs_header, chunk_tree_uuid),
3715 			   BTRFS_UUID_SIZE);
3716 
3717 	ret = btrfs_read_chunk_tree(fs_info);
3718 	if (ret) {
3719 		btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3720 		goto fail_tree_roots;
3721 	}
3722 
3723 	/*
3724 	 * At this point we know all the devices that make this filesystem,
3725 	 * including the seed devices but we don't know yet if the replace
3726 	 * target is required. So free devices that are not part of this
3727 	 * filesystem but skip the replace target device which is checked
3728 	 * below in btrfs_init_dev_replace().
3729 	 */
3730 	btrfs_free_extra_devids(fs_devices);
3731 	if (!fs_devices->latest_dev->bdev) {
3732 		btrfs_err(fs_info, "failed to read devices");
3733 		goto fail_tree_roots;
3734 	}
3735 
3736 	ret = init_tree_roots(fs_info);
3737 	if (ret)
3738 		goto fail_tree_roots;
3739 
3740 	/*
3741 	 * Get zone type information of zoned block devices. This will also
3742 	 * handle emulation of a zoned filesystem if a regular device has the
3743 	 * zoned incompat feature flag set.
3744 	 */
3745 	ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3746 	if (ret) {
3747 		btrfs_err(fs_info,
3748 			  "zoned: failed to read device zone info: %d",
3749 			  ret);
3750 		goto fail_block_groups;
3751 	}
3752 
3753 	/*
3754 	 * If we have a uuid root and we're not being told to rescan we need to
3755 	 * check the generation here so we can set the
3756 	 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3757 	 * transaction during a balance or the log replay without updating the
3758 	 * uuid generation, and then if we crash we would rescan the uuid tree,
3759 	 * even though it was perfectly fine.
3760 	 */
3761 	if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3762 	    fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3763 		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3764 
3765 	ret = btrfs_verify_dev_extents(fs_info);
3766 	if (ret) {
3767 		btrfs_err(fs_info,
3768 			  "failed to verify dev extents against chunks: %d",
3769 			  ret);
3770 		goto fail_block_groups;
3771 	}
3772 	ret = btrfs_recover_balance(fs_info);
3773 	if (ret) {
3774 		btrfs_err(fs_info, "failed to recover balance: %d", ret);
3775 		goto fail_block_groups;
3776 	}
3777 
3778 	ret = btrfs_init_dev_stats(fs_info);
3779 	if (ret) {
3780 		btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3781 		goto fail_block_groups;
3782 	}
3783 
3784 	ret = btrfs_init_dev_replace(fs_info);
3785 	if (ret) {
3786 		btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3787 		goto fail_block_groups;
3788 	}
3789 
3790 	ret = btrfs_check_zoned_mode(fs_info);
3791 	if (ret) {
3792 		btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3793 			  ret);
3794 		goto fail_block_groups;
3795 	}
3796 
3797 	ret = btrfs_sysfs_add_fsid(fs_devices);
3798 	if (ret) {
3799 		btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3800 				ret);
3801 		goto fail_block_groups;
3802 	}
3803 
3804 	ret = btrfs_sysfs_add_mounted(fs_info);
3805 	if (ret) {
3806 		btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3807 		goto fail_fsdev_sysfs;
3808 	}
3809 
3810 	ret = btrfs_init_space_info(fs_info);
3811 	if (ret) {
3812 		btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3813 		goto fail_sysfs;
3814 	}
3815 
3816 	ret = btrfs_read_block_groups(fs_info);
3817 	if (ret) {
3818 		btrfs_err(fs_info, "failed to read block groups: %d", ret);
3819 		goto fail_sysfs;
3820 	}
3821 
3822 	btrfs_free_zone_cache(fs_info);
3823 
3824 	if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3825 	    !btrfs_check_rw_degradable(fs_info, NULL)) {
3826 		btrfs_warn(fs_info,
3827 		"writable mount is not allowed due to too many missing devices");
3828 		goto fail_sysfs;
3829 	}
3830 
3831 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3832 					       "btrfs-cleaner");
3833 	if (IS_ERR(fs_info->cleaner_kthread))
3834 		goto fail_sysfs;
3835 
3836 	fs_info->transaction_kthread = kthread_run(transaction_kthread,
3837 						   tree_root,
3838 						   "btrfs-transaction");
3839 	if (IS_ERR(fs_info->transaction_kthread))
3840 		goto fail_cleaner;
3841 
3842 	if (!btrfs_test_opt(fs_info, NOSSD) &&
3843 	    !fs_info->fs_devices->rotating) {
3844 		btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3845 	}
3846 
3847 	/*
3848 	 * Mount does not set all options immediately, we can do it now and do
3849 	 * not have to wait for transaction commit
3850 	 */
3851 	btrfs_apply_pending_changes(fs_info);
3852 
3853 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3854 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3855 		ret = btrfsic_mount(fs_info, fs_devices,
3856 				    btrfs_test_opt(fs_info,
3857 					CHECK_INTEGRITY_DATA) ? 1 : 0,
3858 				    fs_info->check_integrity_print_mask);
3859 		if (ret)
3860 			btrfs_warn(fs_info,
3861 				"failed to initialize integrity check module: %d",
3862 				ret);
3863 	}
3864 #endif
3865 	ret = btrfs_read_qgroup_config(fs_info);
3866 	if (ret)
3867 		goto fail_trans_kthread;
3868 
3869 	if (btrfs_build_ref_tree(fs_info))
3870 		btrfs_err(fs_info, "couldn't build ref tree");
3871 
3872 	/* do not make disk changes in broken FS or nologreplay is given */
3873 	if (btrfs_super_log_root(disk_super) != 0 &&
3874 	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3875 		btrfs_info(fs_info, "start tree-log replay");
3876 		ret = btrfs_replay_log(fs_info, fs_devices);
3877 		if (ret) {
3878 			err = ret;
3879 			goto fail_qgroup;
3880 		}
3881 	}
3882 
3883 	fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3884 	if (IS_ERR(fs_info->fs_root)) {
3885 		err = PTR_ERR(fs_info->fs_root);
3886 		btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3887 		fs_info->fs_root = NULL;
3888 		goto fail_qgroup;
3889 	}
3890 
3891 	if (sb_rdonly(sb))
3892 		goto clear_oneshot;
3893 
3894 	ret = btrfs_start_pre_rw_mount(fs_info);
3895 	if (ret) {
3896 		close_ctree(fs_info);
3897 		return ret;
3898 	}
3899 	btrfs_discard_resume(fs_info);
3900 
3901 	if (fs_info->uuid_root &&
3902 	    (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3903 	     fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3904 		btrfs_info(fs_info, "checking UUID tree");
3905 		ret = btrfs_check_uuid_tree(fs_info);
3906 		if (ret) {
3907 			btrfs_warn(fs_info,
3908 				"failed to check the UUID tree: %d", ret);
3909 			close_ctree(fs_info);
3910 			return ret;
3911 		}
3912 	}
3913 
3914 	set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3915 
3916 	/* Kick the cleaner thread so it'll start deleting snapshots. */
3917 	if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3918 		wake_up_process(fs_info->cleaner_kthread);
3919 
3920 clear_oneshot:
3921 	btrfs_clear_oneshot_options(fs_info);
3922 	return 0;
3923 
3924 fail_qgroup:
3925 	btrfs_free_qgroup_config(fs_info);
3926 fail_trans_kthread:
3927 	kthread_stop(fs_info->transaction_kthread);
3928 	btrfs_cleanup_transaction(fs_info);
3929 	btrfs_free_fs_roots(fs_info);
3930 fail_cleaner:
3931 	kthread_stop(fs_info->cleaner_kthread);
3932 
3933 	/*
3934 	 * make sure we're done with the btree inode before we stop our
3935 	 * kthreads
3936 	 */
3937 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3938 
3939 fail_sysfs:
3940 	btrfs_sysfs_remove_mounted(fs_info);
3941 
3942 fail_fsdev_sysfs:
3943 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3944 
3945 fail_block_groups:
3946 	btrfs_put_block_group_cache(fs_info);
3947 
3948 fail_tree_roots:
3949 	if (fs_info->data_reloc_root)
3950 		btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3951 	free_root_pointers(fs_info, true);
3952 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3953 
3954 fail_sb_buffer:
3955 	btrfs_stop_all_workers(fs_info);
3956 	btrfs_free_block_groups(fs_info);
3957 fail_alloc:
3958 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3959 
3960 	iput(fs_info->btree_inode);
3961 fail:
3962 	btrfs_close_devices(fs_info->fs_devices);
3963 	return err;
3964 }
3965 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3966 
3967 static void btrfs_end_super_write(struct bio *bio)
3968 {
3969 	struct btrfs_device *device = bio->bi_private;
3970 	struct bio_vec *bvec;
3971 	struct bvec_iter_all iter_all;
3972 	struct page *page;
3973 
3974 	bio_for_each_segment_all(bvec, bio, iter_all) {
3975 		page = bvec->bv_page;
3976 
3977 		if (bio->bi_status) {
3978 			btrfs_warn_rl_in_rcu(device->fs_info,
3979 				"lost page write due to IO error on %s (%d)",
3980 				rcu_str_deref(device->name),
3981 				blk_status_to_errno(bio->bi_status));
3982 			ClearPageUptodate(page);
3983 			SetPageError(page);
3984 			btrfs_dev_stat_inc_and_print(device,
3985 						     BTRFS_DEV_STAT_WRITE_ERRS);
3986 		} else {
3987 			SetPageUptodate(page);
3988 		}
3989 
3990 		put_page(page);
3991 		unlock_page(page);
3992 	}
3993 
3994 	bio_put(bio);
3995 }
3996 
3997 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3998 						   int copy_num)
3999 {
4000 	struct btrfs_super_block *super;
4001 	struct page *page;
4002 	u64 bytenr, bytenr_orig;
4003 	struct address_space *mapping = bdev->bd_inode->i_mapping;
4004 	int ret;
4005 
4006 	bytenr_orig = btrfs_sb_offset(copy_num);
4007 	ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
4008 	if (ret == -ENOENT)
4009 		return ERR_PTR(-EINVAL);
4010 	else if (ret)
4011 		return ERR_PTR(ret);
4012 
4013 	if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
4014 		return ERR_PTR(-EINVAL);
4015 
4016 	page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
4017 	if (IS_ERR(page))
4018 		return ERR_CAST(page);
4019 
4020 	super = page_address(page);
4021 	if (btrfs_super_magic(super) != BTRFS_MAGIC) {
4022 		btrfs_release_disk_super(super);
4023 		return ERR_PTR(-ENODATA);
4024 	}
4025 
4026 	if (btrfs_super_bytenr(super) != bytenr_orig) {
4027 		btrfs_release_disk_super(super);
4028 		return ERR_PTR(-EINVAL);
4029 	}
4030 
4031 	return super;
4032 }
4033 
4034 
4035 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
4036 {
4037 	struct btrfs_super_block *super, *latest = NULL;
4038 	int i;
4039 	u64 transid = 0;
4040 
4041 	/* we would like to check all the supers, but that would make
4042 	 * a btrfs mount succeed after a mkfs from a different FS.
4043 	 * So, we need to add a special mount option to scan for
4044 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
4045 	 */
4046 	for (i = 0; i < 1; i++) {
4047 		super = btrfs_read_dev_one_super(bdev, i);
4048 		if (IS_ERR(super))
4049 			continue;
4050 
4051 		if (!latest || btrfs_super_generation(super) > transid) {
4052 			if (latest)
4053 				btrfs_release_disk_super(super);
4054 
4055 			latest = super;
4056 			transid = btrfs_super_generation(super);
4057 		}
4058 	}
4059 
4060 	return super;
4061 }
4062 
4063 /*
4064  * Write superblock @sb to the @device. Do not wait for completion, all the
4065  * pages we use for writing are locked.
4066  *
4067  * Write @max_mirrors copies of the superblock, where 0 means default that fit
4068  * the expected device size at commit time. Note that max_mirrors must be
4069  * same for write and wait phases.
4070  *
4071  * Return number of errors when page is not found or submission fails.
4072  */
4073 static int write_dev_supers(struct btrfs_device *device,
4074 			    struct btrfs_super_block *sb, int max_mirrors)
4075 {
4076 	struct btrfs_fs_info *fs_info = device->fs_info;
4077 	struct address_space *mapping = device->bdev->bd_inode->i_mapping;
4078 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
4079 	int i;
4080 	int errors = 0;
4081 	int ret;
4082 	u64 bytenr, bytenr_orig;
4083 
4084 	if (max_mirrors == 0)
4085 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4086 
4087 	shash->tfm = fs_info->csum_shash;
4088 
4089 	for (i = 0; i < max_mirrors; i++) {
4090 		struct page *page;
4091 		struct bio *bio;
4092 		struct btrfs_super_block *disk_super;
4093 
4094 		bytenr_orig = btrfs_sb_offset(i);
4095 		ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
4096 		if (ret == -ENOENT) {
4097 			continue;
4098 		} else if (ret < 0) {
4099 			btrfs_err(device->fs_info,
4100 				"couldn't get super block location for mirror %d",
4101 				i);
4102 			errors++;
4103 			continue;
4104 		}
4105 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4106 		    device->commit_total_bytes)
4107 			break;
4108 
4109 		btrfs_set_super_bytenr(sb, bytenr_orig);
4110 
4111 		crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
4112 				    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
4113 				    sb->csum);
4114 
4115 		page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
4116 					   GFP_NOFS);
4117 		if (!page) {
4118 			btrfs_err(device->fs_info,
4119 			    "couldn't get super block page for bytenr %llu",
4120 			    bytenr);
4121 			errors++;
4122 			continue;
4123 		}
4124 
4125 		/* Bump the refcount for wait_dev_supers() */
4126 		get_page(page);
4127 
4128 		disk_super = page_address(page);
4129 		memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
4130 
4131 		/*
4132 		 * Directly use bios here instead of relying on the page cache
4133 		 * to do I/O, so we don't lose the ability to do integrity
4134 		 * checking.
4135 		 */
4136 		bio = bio_alloc(device->bdev, 1,
4137 				REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
4138 				GFP_NOFS);
4139 		bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
4140 		bio->bi_private = device;
4141 		bio->bi_end_io = btrfs_end_super_write;
4142 		__bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
4143 			       offset_in_page(bytenr));
4144 
4145 		/*
4146 		 * We FUA only the first super block.  The others we allow to
4147 		 * go down lazy and there's a short window where the on-disk
4148 		 * copies might still contain the older version.
4149 		 */
4150 		if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
4151 			bio->bi_opf |= REQ_FUA;
4152 
4153 		btrfsic_check_bio(bio);
4154 		submit_bio(bio);
4155 
4156 		if (btrfs_advance_sb_log(device, i))
4157 			errors++;
4158 	}
4159 	return errors < i ? 0 : -1;
4160 }
4161 
4162 /*
4163  * Wait for write completion of superblocks done by write_dev_supers,
4164  * @max_mirrors same for write and wait phases.
4165  *
4166  * Return number of errors when page is not found or not marked up to
4167  * date.
4168  */
4169 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
4170 {
4171 	int i;
4172 	int errors = 0;
4173 	bool primary_failed = false;
4174 	int ret;
4175 	u64 bytenr;
4176 
4177 	if (max_mirrors == 0)
4178 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4179 
4180 	for (i = 0; i < max_mirrors; i++) {
4181 		struct page *page;
4182 
4183 		ret = btrfs_sb_log_location(device, i, READ, &bytenr);
4184 		if (ret == -ENOENT) {
4185 			break;
4186 		} else if (ret < 0) {
4187 			errors++;
4188 			if (i == 0)
4189 				primary_failed = true;
4190 			continue;
4191 		}
4192 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4193 		    device->commit_total_bytes)
4194 			break;
4195 
4196 		page = find_get_page(device->bdev->bd_inode->i_mapping,
4197 				     bytenr >> PAGE_SHIFT);
4198 		if (!page) {
4199 			errors++;
4200 			if (i == 0)
4201 				primary_failed = true;
4202 			continue;
4203 		}
4204 		/* Page is submitted locked and unlocked once the IO completes */
4205 		wait_on_page_locked(page);
4206 		if (PageError(page)) {
4207 			errors++;
4208 			if (i == 0)
4209 				primary_failed = true;
4210 		}
4211 
4212 		/* Drop our reference */
4213 		put_page(page);
4214 
4215 		/* Drop the reference from the writing run */
4216 		put_page(page);
4217 	}
4218 
4219 	/* log error, force error return */
4220 	if (primary_failed) {
4221 		btrfs_err(device->fs_info, "error writing primary super block to device %llu",
4222 			  device->devid);
4223 		return -1;
4224 	}
4225 
4226 	return errors < i ? 0 : -1;
4227 }
4228 
4229 /*
4230  * endio for the write_dev_flush, this will wake anyone waiting
4231  * for the barrier when it is done
4232  */
4233 static void btrfs_end_empty_barrier(struct bio *bio)
4234 {
4235 	bio_uninit(bio);
4236 	complete(bio->bi_private);
4237 }
4238 
4239 /*
4240  * Submit a flush request to the device if it supports it. Error handling is
4241  * done in the waiting counterpart.
4242  */
4243 static void write_dev_flush(struct btrfs_device *device)
4244 {
4245 	struct bio *bio = &device->flush_bio;
4246 
4247 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4248 	/*
4249 	 * When a disk has write caching disabled, we skip submission of a bio
4250 	 * with flush and sync requests before writing the superblock, since
4251 	 * it's not needed. However when the integrity checker is enabled, this
4252 	 * results in reports that there are metadata blocks referred by a
4253 	 * superblock that were not properly flushed. So don't skip the bio
4254 	 * submission only when the integrity checker is enabled for the sake
4255 	 * of simplicity, since this is a debug tool and not meant for use in
4256 	 * non-debug builds.
4257 	 */
4258 	if (!bdev_write_cache(device->bdev))
4259 		return;
4260 #endif
4261 
4262 	bio_init(bio, device->bdev, NULL, 0,
4263 		 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
4264 	bio->bi_end_io = btrfs_end_empty_barrier;
4265 	init_completion(&device->flush_wait);
4266 	bio->bi_private = &device->flush_wait;
4267 
4268 	btrfsic_check_bio(bio);
4269 	submit_bio(bio);
4270 	set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4271 }
4272 
4273 /*
4274  * If the flush bio has been submitted by write_dev_flush, wait for it.
4275  */
4276 static blk_status_t wait_dev_flush(struct btrfs_device *device)
4277 {
4278 	struct bio *bio = &device->flush_bio;
4279 
4280 	if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
4281 		return BLK_STS_OK;
4282 
4283 	clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4284 	wait_for_completion_io(&device->flush_wait);
4285 
4286 	return bio->bi_status;
4287 }
4288 
4289 static int check_barrier_error(struct btrfs_fs_info *fs_info)
4290 {
4291 	if (!btrfs_check_rw_degradable(fs_info, NULL))
4292 		return -EIO;
4293 	return 0;
4294 }
4295 
4296 /*
4297  * send an empty flush down to each device in parallel,
4298  * then wait for them
4299  */
4300 static int barrier_all_devices(struct btrfs_fs_info *info)
4301 {
4302 	struct list_head *head;
4303 	struct btrfs_device *dev;
4304 	int errors_wait = 0;
4305 	blk_status_t ret;
4306 
4307 	lockdep_assert_held(&info->fs_devices->device_list_mutex);
4308 	/* send down all the barriers */
4309 	head = &info->fs_devices->devices;
4310 	list_for_each_entry(dev, head, dev_list) {
4311 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4312 			continue;
4313 		if (!dev->bdev)
4314 			continue;
4315 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4316 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4317 			continue;
4318 
4319 		write_dev_flush(dev);
4320 		dev->last_flush_error = BLK_STS_OK;
4321 	}
4322 
4323 	/* wait for all the barriers */
4324 	list_for_each_entry(dev, head, dev_list) {
4325 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4326 			continue;
4327 		if (!dev->bdev) {
4328 			errors_wait++;
4329 			continue;
4330 		}
4331 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4332 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4333 			continue;
4334 
4335 		ret = wait_dev_flush(dev);
4336 		if (ret) {
4337 			dev->last_flush_error = ret;
4338 			btrfs_dev_stat_inc_and_print(dev,
4339 					BTRFS_DEV_STAT_FLUSH_ERRS);
4340 			errors_wait++;
4341 		}
4342 	}
4343 
4344 	if (errors_wait) {
4345 		/*
4346 		 * At some point we need the status of all disks
4347 		 * to arrive at the volume status. So error checking
4348 		 * is being pushed to a separate loop.
4349 		 */
4350 		return check_barrier_error(info);
4351 	}
4352 	return 0;
4353 }
4354 
4355 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4356 {
4357 	int raid_type;
4358 	int min_tolerated = INT_MAX;
4359 
4360 	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4361 	    (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4362 		min_tolerated = min_t(int, min_tolerated,
4363 				    btrfs_raid_array[BTRFS_RAID_SINGLE].
4364 				    tolerated_failures);
4365 
4366 	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4367 		if (raid_type == BTRFS_RAID_SINGLE)
4368 			continue;
4369 		if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4370 			continue;
4371 		min_tolerated = min_t(int, min_tolerated,
4372 				    btrfs_raid_array[raid_type].
4373 				    tolerated_failures);
4374 	}
4375 
4376 	if (min_tolerated == INT_MAX) {
4377 		pr_warn("BTRFS: unknown raid flag: %llu", flags);
4378 		min_tolerated = 0;
4379 	}
4380 
4381 	return min_tolerated;
4382 }
4383 
4384 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4385 {
4386 	struct list_head *head;
4387 	struct btrfs_device *dev;
4388 	struct btrfs_super_block *sb;
4389 	struct btrfs_dev_item *dev_item;
4390 	int ret;
4391 	int do_barriers;
4392 	int max_errors;
4393 	int total_errors = 0;
4394 	u64 flags;
4395 
4396 	do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4397 
4398 	/*
4399 	 * max_mirrors == 0 indicates we're from commit_transaction,
4400 	 * not from fsync where the tree roots in fs_info have not
4401 	 * been consistent on disk.
4402 	 */
4403 	if (max_mirrors == 0)
4404 		backup_super_roots(fs_info);
4405 
4406 	sb = fs_info->super_for_commit;
4407 	dev_item = &sb->dev_item;
4408 
4409 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
4410 	head = &fs_info->fs_devices->devices;
4411 	max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4412 
4413 	if (do_barriers) {
4414 		ret = barrier_all_devices(fs_info);
4415 		if (ret) {
4416 			mutex_unlock(
4417 				&fs_info->fs_devices->device_list_mutex);
4418 			btrfs_handle_fs_error(fs_info, ret,
4419 					      "errors while submitting device barriers.");
4420 			return ret;
4421 		}
4422 	}
4423 
4424 	list_for_each_entry(dev, head, dev_list) {
4425 		if (!dev->bdev) {
4426 			total_errors++;
4427 			continue;
4428 		}
4429 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4430 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4431 			continue;
4432 
4433 		btrfs_set_stack_device_generation(dev_item, 0);
4434 		btrfs_set_stack_device_type(dev_item, dev->type);
4435 		btrfs_set_stack_device_id(dev_item, dev->devid);
4436 		btrfs_set_stack_device_total_bytes(dev_item,
4437 						   dev->commit_total_bytes);
4438 		btrfs_set_stack_device_bytes_used(dev_item,
4439 						  dev->commit_bytes_used);
4440 		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4441 		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4442 		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4443 		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4444 		memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4445 		       BTRFS_FSID_SIZE);
4446 
4447 		flags = btrfs_super_flags(sb);
4448 		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4449 
4450 		ret = btrfs_validate_write_super(fs_info, sb);
4451 		if (ret < 0) {
4452 			mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4453 			btrfs_handle_fs_error(fs_info, -EUCLEAN,
4454 				"unexpected superblock corruption detected");
4455 			return -EUCLEAN;
4456 		}
4457 
4458 		ret = write_dev_supers(dev, sb, max_mirrors);
4459 		if (ret)
4460 			total_errors++;
4461 	}
4462 	if (total_errors > max_errors) {
4463 		btrfs_err(fs_info, "%d errors while writing supers",
4464 			  total_errors);
4465 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4466 
4467 		/* FUA is masked off if unsupported and can't be the reason */
4468 		btrfs_handle_fs_error(fs_info, -EIO,
4469 				      "%d errors while writing supers",
4470 				      total_errors);
4471 		return -EIO;
4472 	}
4473 
4474 	total_errors = 0;
4475 	list_for_each_entry(dev, head, dev_list) {
4476 		if (!dev->bdev)
4477 			continue;
4478 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4479 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4480 			continue;
4481 
4482 		ret = wait_dev_supers(dev, max_mirrors);
4483 		if (ret)
4484 			total_errors++;
4485 	}
4486 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4487 	if (total_errors > max_errors) {
4488 		btrfs_handle_fs_error(fs_info, -EIO,
4489 				      "%d errors while writing supers",
4490 				      total_errors);
4491 		return -EIO;
4492 	}
4493 	return 0;
4494 }
4495 
4496 /* Drop a fs root from the radix tree and free it. */
4497 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4498 				  struct btrfs_root *root)
4499 {
4500 	bool drop_ref = false;
4501 
4502 	spin_lock(&fs_info->fs_roots_lock);
4503 	xa_erase(&fs_info->fs_roots, (unsigned long)root->root_key.objectid);
4504 	if (test_and_clear_bit(BTRFS_ROOT_REGISTERED, &root->state))
4505 		drop_ref = true;
4506 	spin_unlock(&fs_info->fs_roots_lock);
4507 
4508 	if (BTRFS_FS_ERROR(fs_info)) {
4509 		ASSERT(root->log_root == NULL);
4510 		if (root->reloc_root) {
4511 			btrfs_put_root(root->reloc_root);
4512 			root->reloc_root = NULL;
4513 		}
4514 	}
4515 
4516 	if (drop_ref)
4517 		btrfs_put_root(root);
4518 }
4519 
4520 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4521 {
4522 	struct btrfs_root *roots[8];
4523 	unsigned long index = 0;
4524 	int i;
4525 	int err = 0;
4526 	int grabbed;
4527 
4528 	while (1) {
4529 		struct btrfs_root *root;
4530 
4531 		spin_lock(&fs_info->fs_roots_lock);
4532 		if (!xa_find(&fs_info->fs_roots, &index, ULONG_MAX, XA_PRESENT)) {
4533 			spin_unlock(&fs_info->fs_roots_lock);
4534 			return err;
4535 		}
4536 
4537 		grabbed = 0;
4538 		xa_for_each_start(&fs_info->fs_roots, index, root, index) {
4539 			/* Avoid grabbing roots in dead_roots */
4540 			if (btrfs_root_refs(&root->root_item) > 0)
4541 				roots[grabbed++] = btrfs_grab_root(root);
4542 			if (grabbed >= ARRAY_SIZE(roots))
4543 				break;
4544 		}
4545 		spin_unlock(&fs_info->fs_roots_lock);
4546 
4547 		for (i = 0; i < grabbed; i++) {
4548 			if (!roots[i])
4549 				continue;
4550 			index = roots[i]->root_key.objectid;
4551 			err = btrfs_orphan_cleanup(roots[i]);
4552 			if (err)
4553 				goto out;
4554 			btrfs_put_root(roots[i]);
4555 		}
4556 		index++;
4557 	}
4558 
4559 out:
4560 	/* Release the roots that remain uncleaned due to error */
4561 	for (; i < grabbed; i++) {
4562 		if (roots[i])
4563 			btrfs_put_root(roots[i]);
4564 	}
4565 	return err;
4566 }
4567 
4568 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4569 {
4570 	struct btrfs_root *root = fs_info->tree_root;
4571 	struct btrfs_trans_handle *trans;
4572 
4573 	mutex_lock(&fs_info->cleaner_mutex);
4574 	btrfs_run_delayed_iputs(fs_info);
4575 	mutex_unlock(&fs_info->cleaner_mutex);
4576 	wake_up_process(fs_info->cleaner_kthread);
4577 
4578 	/* wait until ongoing cleanup work done */
4579 	down_write(&fs_info->cleanup_work_sem);
4580 	up_write(&fs_info->cleanup_work_sem);
4581 
4582 	trans = btrfs_join_transaction(root);
4583 	if (IS_ERR(trans))
4584 		return PTR_ERR(trans);
4585 	return btrfs_commit_transaction(trans);
4586 }
4587 
4588 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4589 {
4590 	struct btrfs_transaction *trans;
4591 	struct btrfs_transaction *tmp;
4592 	bool found = false;
4593 
4594 	if (list_empty(&fs_info->trans_list))
4595 		return;
4596 
4597 	/*
4598 	 * This function is only called at the very end of close_ctree(),
4599 	 * thus no other running transaction, no need to take trans_lock.
4600 	 */
4601 	ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4602 	list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4603 		struct extent_state *cached = NULL;
4604 		u64 dirty_bytes = 0;
4605 		u64 cur = 0;
4606 		u64 found_start;
4607 		u64 found_end;
4608 
4609 		found = true;
4610 		while (!find_first_extent_bit(&trans->dirty_pages, cur,
4611 			&found_start, &found_end, EXTENT_DIRTY, &cached)) {
4612 			dirty_bytes += found_end + 1 - found_start;
4613 			cur = found_end + 1;
4614 		}
4615 		btrfs_warn(fs_info,
4616 	"transaction %llu (with %llu dirty metadata bytes) is not committed",
4617 			   trans->transid, dirty_bytes);
4618 		btrfs_cleanup_one_transaction(trans, fs_info);
4619 
4620 		if (trans == fs_info->running_transaction)
4621 			fs_info->running_transaction = NULL;
4622 		list_del_init(&trans->list);
4623 
4624 		btrfs_put_transaction(trans);
4625 		trace_btrfs_transaction_commit(fs_info);
4626 	}
4627 	ASSERT(!found);
4628 }
4629 
4630 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4631 {
4632 	int ret;
4633 
4634 	set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4635 	/*
4636 	 * We don't want the cleaner to start new transactions, add more delayed
4637 	 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4638 	 * because that frees the task_struct, and the transaction kthread might
4639 	 * still try to wake up the cleaner.
4640 	 */
4641 	kthread_park(fs_info->cleaner_kthread);
4642 
4643 	/*
4644 	 * If we had UNFINISHED_DROPS we could still be processing them, so
4645 	 * clear that bit and wake up relocation so it can stop.
4646 	 */
4647 	btrfs_wake_unfinished_drop(fs_info);
4648 
4649 	/* wait for the qgroup rescan worker to stop */
4650 	btrfs_qgroup_wait_for_completion(fs_info, false);
4651 
4652 	/* wait for the uuid_scan task to finish */
4653 	down(&fs_info->uuid_tree_rescan_sem);
4654 	/* avoid complains from lockdep et al., set sem back to initial state */
4655 	up(&fs_info->uuid_tree_rescan_sem);
4656 
4657 	/* pause restriper - we want to resume on mount */
4658 	btrfs_pause_balance(fs_info);
4659 
4660 	btrfs_dev_replace_suspend_for_unmount(fs_info);
4661 
4662 	btrfs_scrub_cancel(fs_info);
4663 
4664 	/* wait for any defraggers to finish */
4665 	wait_event(fs_info->transaction_wait,
4666 		   (atomic_read(&fs_info->defrag_running) == 0));
4667 
4668 	/* clear out the rbtree of defraggable inodes */
4669 	btrfs_cleanup_defrag_inodes(fs_info);
4670 
4671 	cancel_work_sync(&fs_info->async_reclaim_work);
4672 	cancel_work_sync(&fs_info->async_data_reclaim_work);
4673 	cancel_work_sync(&fs_info->preempt_reclaim_work);
4674 
4675 	cancel_work_sync(&fs_info->reclaim_bgs_work);
4676 
4677 	/* Cancel or finish ongoing discard work */
4678 	btrfs_discard_cleanup(fs_info);
4679 
4680 	if (!sb_rdonly(fs_info->sb)) {
4681 		/*
4682 		 * The cleaner kthread is stopped, so do one final pass over
4683 		 * unused block groups.
4684 		 */
4685 		btrfs_delete_unused_bgs(fs_info);
4686 
4687 		/*
4688 		 * There might be existing delayed inode workers still running
4689 		 * and holding an empty delayed inode item. We must wait for
4690 		 * them to complete first because they can create a transaction.
4691 		 * This happens when someone calls btrfs_balance_delayed_items()
4692 		 * and then a transaction commit runs the same delayed nodes
4693 		 * before any delayed worker has done something with the nodes.
4694 		 * We must wait for any worker here and not at transaction
4695 		 * commit time since that could cause a deadlock.
4696 		 * This is a very rare case.
4697 		 */
4698 		btrfs_flush_workqueue(fs_info->delayed_workers);
4699 
4700 		ret = btrfs_commit_super(fs_info);
4701 		if (ret)
4702 			btrfs_err(fs_info, "commit super ret %d", ret);
4703 	}
4704 
4705 	if (BTRFS_FS_ERROR(fs_info))
4706 		btrfs_error_commit_super(fs_info);
4707 
4708 	kthread_stop(fs_info->transaction_kthread);
4709 	kthread_stop(fs_info->cleaner_kthread);
4710 
4711 	ASSERT(list_empty(&fs_info->delayed_iputs));
4712 	set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4713 
4714 	if (btrfs_check_quota_leak(fs_info)) {
4715 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4716 		btrfs_err(fs_info, "qgroup reserved space leaked");
4717 	}
4718 
4719 	btrfs_free_qgroup_config(fs_info);
4720 	ASSERT(list_empty(&fs_info->delalloc_roots));
4721 
4722 	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4723 		btrfs_info(fs_info, "at unmount delalloc count %lld",
4724 		       percpu_counter_sum(&fs_info->delalloc_bytes));
4725 	}
4726 
4727 	if (percpu_counter_sum(&fs_info->ordered_bytes))
4728 		btrfs_info(fs_info, "at unmount dio bytes count %lld",
4729 			   percpu_counter_sum(&fs_info->ordered_bytes));
4730 
4731 	btrfs_sysfs_remove_mounted(fs_info);
4732 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4733 
4734 	btrfs_put_block_group_cache(fs_info);
4735 
4736 	/*
4737 	 * we must make sure there is not any read request to
4738 	 * submit after we stopping all workers.
4739 	 */
4740 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4741 	btrfs_stop_all_workers(fs_info);
4742 
4743 	/* We shouldn't have any transaction open at this point */
4744 	warn_about_uncommitted_trans(fs_info);
4745 
4746 	clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4747 	free_root_pointers(fs_info, true);
4748 	btrfs_free_fs_roots(fs_info);
4749 
4750 	/*
4751 	 * We must free the block groups after dropping the fs_roots as we could
4752 	 * have had an IO error and have left over tree log blocks that aren't
4753 	 * cleaned up until the fs roots are freed.  This makes the block group
4754 	 * accounting appear to be wrong because there's pending reserved bytes,
4755 	 * so make sure we do the block group cleanup afterwards.
4756 	 */
4757 	btrfs_free_block_groups(fs_info);
4758 
4759 	iput(fs_info->btree_inode);
4760 
4761 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4762 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4763 		btrfsic_unmount(fs_info->fs_devices);
4764 #endif
4765 
4766 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
4767 	btrfs_close_devices(fs_info->fs_devices);
4768 }
4769 
4770 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4771 			  int atomic)
4772 {
4773 	int ret;
4774 	struct inode *btree_inode = buf->pages[0]->mapping->host;
4775 
4776 	ret = extent_buffer_uptodate(buf);
4777 	if (!ret)
4778 		return ret;
4779 
4780 	ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4781 				    parent_transid, atomic);
4782 	if (ret == -EAGAIN)
4783 		return ret;
4784 	return !ret;
4785 }
4786 
4787 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4788 {
4789 	struct btrfs_fs_info *fs_info = buf->fs_info;
4790 	u64 transid = btrfs_header_generation(buf);
4791 	int was_dirty;
4792 
4793 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4794 	/*
4795 	 * This is a fast path so only do this check if we have sanity tests
4796 	 * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4797 	 * outside of the sanity tests.
4798 	 */
4799 	if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4800 		return;
4801 #endif
4802 	btrfs_assert_tree_write_locked(buf);
4803 	if (transid != fs_info->generation)
4804 		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4805 			buf->start, transid, fs_info->generation);
4806 	was_dirty = set_extent_buffer_dirty(buf);
4807 	if (!was_dirty)
4808 		percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4809 					 buf->len,
4810 					 fs_info->dirty_metadata_batch);
4811 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4812 	/*
4813 	 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4814 	 * but item data not updated.
4815 	 * So here we should only check item pointers, not item data.
4816 	 */
4817 	if (btrfs_header_level(buf) == 0 &&
4818 	    btrfs_check_leaf_relaxed(buf)) {
4819 		btrfs_print_leaf(buf);
4820 		ASSERT(0);
4821 	}
4822 #endif
4823 }
4824 
4825 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4826 					int flush_delayed)
4827 {
4828 	/*
4829 	 * looks as though older kernels can get into trouble with
4830 	 * this code, they end up stuck in balance_dirty_pages forever
4831 	 */
4832 	int ret;
4833 
4834 	if (current->flags & PF_MEMALLOC)
4835 		return;
4836 
4837 	if (flush_delayed)
4838 		btrfs_balance_delayed_items(fs_info);
4839 
4840 	ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4841 				     BTRFS_DIRTY_METADATA_THRESH,
4842 				     fs_info->dirty_metadata_batch);
4843 	if (ret > 0) {
4844 		balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4845 	}
4846 }
4847 
4848 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4849 {
4850 	__btrfs_btree_balance_dirty(fs_info, 1);
4851 }
4852 
4853 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4854 {
4855 	__btrfs_btree_balance_dirty(fs_info, 0);
4856 }
4857 
4858 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4859 {
4860 	/* cleanup FS via transaction */
4861 	btrfs_cleanup_transaction(fs_info);
4862 
4863 	mutex_lock(&fs_info->cleaner_mutex);
4864 	btrfs_run_delayed_iputs(fs_info);
4865 	mutex_unlock(&fs_info->cleaner_mutex);
4866 
4867 	down_write(&fs_info->cleanup_work_sem);
4868 	up_write(&fs_info->cleanup_work_sem);
4869 }
4870 
4871 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4872 {
4873 	unsigned long index = 0;
4874 	int grabbed = 0;
4875 	struct btrfs_root *roots[8];
4876 
4877 	spin_lock(&fs_info->fs_roots_lock);
4878 	while ((grabbed = xa_extract(&fs_info->fs_roots, (void **)roots, index,
4879 				     ULONG_MAX, 8, XA_PRESENT))) {
4880 		for (int i = 0; i < grabbed; i++)
4881 			roots[i] = btrfs_grab_root(roots[i]);
4882 		spin_unlock(&fs_info->fs_roots_lock);
4883 
4884 		for (int i = 0; i < grabbed; i++) {
4885 			if (!roots[i])
4886 				continue;
4887 			index = roots[i]->root_key.objectid;
4888 			btrfs_free_log(NULL, roots[i]);
4889 			btrfs_put_root(roots[i]);
4890 		}
4891 		index++;
4892 		spin_lock(&fs_info->fs_roots_lock);
4893 	}
4894 	spin_unlock(&fs_info->fs_roots_lock);
4895 	btrfs_free_log_root_tree(NULL, fs_info);
4896 }
4897 
4898 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4899 {
4900 	struct btrfs_ordered_extent *ordered;
4901 
4902 	spin_lock(&root->ordered_extent_lock);
4903 	/*
4904 	 * This will just short circuit the ordered completion stuff which will
4905 	 * make sure the ordered extent gets properly cleaned up.
4906 	 */
4907 	list_for_each_entry(ordered, &root->ordered_extents,
4908 			    root_extent_list)
4909 		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4910 	spin_unlock(&root->ordered_extent_lock);
4911 }
4912 
4913 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4914 {
4915 	struct btrfs_root *root;
4916 	struct list_head splice;
4917 
4918 	INIT_LIST_HEAD(&splice);
4919 
4920 	spin_lock(&fs_info->ordered_root_lock);
4921 	list_splice_init(&fs_info->ordered_roots, &splice);
4922 	while (!list_empty(&splice)) {
4923 		root = list_first_entry(&splice, struct btrfs_root,
4924 					ordered_root);
4925 		list_move_tail(&root->ordered_root,
4926 			       &fs_info->ordered_roots);
4927 
4928 		spin_unlock(&fs_info->ordered_root_lock);
4929 		btrfs_destroy_ordered_extents(root);
4930 
4931 		cond_resched();
4932 		spin_lock(&fs_info->ordered_root_lock);
4933 	}
4934 	spin_unlock(&fs_info->ordered_root_lock);
4935 
4936 	/*
4937 	 * We need this here because if we've been flipped read-only we won't
4938 	 * get sync() from the umount, so we need to make sure any ordered
4939 	 * extents that haven't had their dirty pages IO start writeout yet
4940 	 * actually get run and error out properly.
4941 	 */
4942 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4943 }
4944 
4945 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4946 				      struct btrfs_fs_info *fs_info)
4947 {
4948 	struct rb_node *node;
4949 	struct btrfs_delayed_ref_root *delayed_refs;
4950 	struct btrfs_delayed_ref_node *ref;
4951 	int ret = 0;
4952 
4953 	delayed_refs = &trans->delayed_refs;
4954 
4955 	spin_lock(&delayed_refs->lock);
4956 	if (atomic_read(&delayed_refs->num_entries) == 0) {
4957 		spin_unlock(&delayed_refs->lock);
4958 		btrfs_debug(fs_info, "delayed_refs has NO entry");
4959 		return ret;
4960 	}
4961 
4962 	while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4963 		struct btrfs_delayed_ref_head *head;
4964 		struct rb_node *n;
4965 		bool pin_bytes = false;
4966 
4967 		head = rb_entry(node, struct btrfs_delayed_ref_head,
4968 				href_node);
4969 		if (btrfs_delayed_ref_lock(delayed_refs, head))
4970 			continue;
4971 
4972 		spin_lock(&head->lock);
4973 		while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4974 			ref = rb_entry(n, struct btrfs_delayed_ref_node,
4975 				       ref_node);
4976 			ref->in_tree = 0;
4977 			rb_erase_cached(&ref->ref_node, &head->ref_tree);
4978 			RB_CLEAR_NODE(&ref->ref_node);
4979 			if (!list_empty(&ref->add_list))
4980 				list_del(&ref->add_list);
4981 			atomic_dec(&delayed_refs->num_entries);
4982 			btrfs_put_delayed_ref(ref);
4983 		}
4984 		if (head->must_insert_reserved)
4985 			pin_bytes = true;
4986 		btrfs_free_delayed_extent_op(head->extent_op);
4987 		btrfs_delete_ref_head(delayed_refs, head);
4988 		spin_unlock(&head->lock);
4989 		spin_unlock(&delayed_refs->lock);
4990 		mutex_unlock(&head->mutex);
4991 
4992 		if (pin_bytes) {
4993 			struct btrfs_block_group *cache;
4994 
4995 			cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4996 			BUG_ON(!cache);
4997 
4998 			spin_lock(&cache->space_info->lock);
4999 			spin_lock(&cache->lock);
5000 			cache->pinned += head->num_bytes;
5001 			btrfs_space_info_update_bytes_pinned(fs_info,
5002 				cache->space_info, head->num_bytes);
5003 			cache->reserved -= head->num_bytes;
5004 			cache->space_info->bytes_reserved -= head->num_bytes;
5005 			spin_unlock(&cache->lock);
5006 			spin_unlock(&cache->space_info->lock);
5007 
5008 			btrfs_put_block_group(cache);
5009 
5010 			btrfs_error_unpin_extent_range(fs_info, head->bytenr,
5011 				head->bytenr + head->num_bytes - 1);
5012 		}
5013 		btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
5014 		btrfs_put_delayed_ref_head(head);
5015 		cond_resched();
5016 		spin_lock(&delayed_refs->lock);
5017 	}
5018 	btrfs_qgroup_destroy_extent_records(trans);
5019 
5020 	spin_unlock(&delayed_refs->lock);
5021 
5022 	return ret;
5023 }
5024 
5025 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
5026 {
5027 	struct btrfs_inode *btrfs_inode;
5028 	struct list_head splice;
5029 
5030 	INIT_LIST_HEAD(&splice);
5031 
5032 	spin_lock(&root->delalloc_lock);
5033 	list_splice_init(&root->delalloc_inodes, &splice);
5034 
5035 	while (!list_empty(&splice)) {
5036 		struct inode *inode = NULL;
5037 		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
5038 					       delalloc_inodes);
5039 		__btrfs_del_delalloc_inode(root, btrfs_inode);
5040 		spin_unlock(&root->delalloc_lock);
5041 
5042 		/*
5043 		 * Make sure we get a live inode and that it'll not disappear
5044 		 * meanwhile.
5045 		 */
5046 		inode = igrab(&btrfs_inode->vfs_inode);
5047 		if (inode) {
5048 			invalidate_inode_pages2(inode->i_mapping);
5049 			iput(inode);
5050 		}
5051 		spin_lock(&root->delalloc_lock);
5052 	}
5053 	spin_unlock(&root->delalloc_lock);
5054 }
5055 
5056 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
5057 {
5058 	struct btrfs_root *root;
5059 	struct list_head splice;
5060 
5061 	INIT_LIST_HEAD(&splice);
5062 
5063 	spin_lock(&fs_info->delalloc_root_lock);
5064 	list_splice_init(&fs_info->delalloc_roots, &splice);
5065 	while (!list_empty(&splice)) {
5066 		root = list_first_entry(&splice, struct btrfs_root,
5067 					 delalloc_root);
5068 		root = btrfs_grab_root(root);
5069 		BUG_ON(!root);
5070 		spin_unlock(&fs_info->delalloc_root_lock);
5071 
5072 		btrfs_destroy_delalloc_inodes(root);
5073 		btrfs_put_root(root);
5074 
5075 		spin_lock(&fs_info->delalloc_root_lock);
5076 	}
5077 	spin_unlock(&fs_info->delalloc_root_lock);
5078 }
5079 
5080 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
5081 					struct extent_io_tree *dirty_pages,
5082 					int mark)
5083 {
5084 	int ret;
5085 	struct extent_buffer *eb;
5086 	u64 start = 0;
5087 	u64 end;
5088 
5089 	while (1) {
5090 		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
5091 					    mark, NULL);
5092 		if (ret)
5093 			break;
5094 
5095 		clear_extent_bits(dirty_pages, start, end, mark);
5096 		while (start <= end) {
5097 			eb = find_extent_buffer(fs_info, start);
5098 			start += fs_info->nodesize;
5099 			if (!eb)
5100 				continue;
5101 			wait_on_extent_buffer_writeback(eb);
5102 
5103 			if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
5104 					       &eb->bflags))
5105 				clear_extent_buffer_dirty(eb);
5106 			free_extent_buffer_stale(eb);
5107 		}
5108 	}
5109 
5110 	return ret;
5111 }
5112 
5113 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
5114 				       struct extent_io_tree *unpin)
5115 {
5116 	u64 start;
5117 	u64 end;
5118 	int ret;
5119 
5120 	while (1) {
5121 		struct extent_state *cached_state = NULL;
5122 
5123 		/*
5124 		 * The btrfs_finish_extent_commit() may get the same range as
5125 		 * ours between find_first_extent_bit and clear_extent_dirty.
5126 		 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
5127 		 * the same extent range.
5128 		 */
5129 		mutex_lock(&fs_info->unused_bg_unpin_mutex);
5130 		ret = find_first_extent_bit(unpin, 0, &start, &end,
5131 					    EXTENT_DIRTY, &cached_state);
5132 		if (ret) {
5133 			mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5134 			break;
5135 		}
5136 
5137 		clear_extent_dirty(unpin, start, end, &cached_state);
5138 		free_extent_state(cached_state);
5139 		btrfs_error_unpin_extent_range(fs_info, start, end);
5140 		mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5141 		cond_resched();
5142 	}
5143 
5144 	return 0;
5145 }
5146 
5147 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
5148 {
5149 	struct inode *inode;
5150 
5151 	inode = cache->io_ctl.inode;
5152 	if (inode) {
5153 		invalidate_inode_pages2(inode->i_mapping);
5154 		BTRFS_I(inode)->generation = 0;
5155 		cache->io_ctl.inode = NULL;
5156 		iput(inode);
5157 	}
5158 	ASSERT(cache->io_ctl.pages == NULL);
5159 	btrfs_put_block_group(cache);
5160 }
5161 
5162 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
5163 			     struct btrfs_fs_info *fs_info)
5164 {
5165 	struct btrfs_block_group *cache;
5166 
5167 	spin_lock(&cur_trans->dirty_bgs_lock);
5168 	while (!list_empty(&cur_trans->dirty_bgs)) {
5169 		cache = list_first_entry(&cur_trans->dirty_bgs,
5170 					 struct btrfs_block_group,
5171 					 dirty_list);
5172 
5173 		if (!list_empty(&cache->io_list)) {
5174 			spin_unlock(&cur_trans->dirty_bgs_lock);
5175 			list_del_init(&cache->io_list);
5176 			btrfs_cleanup_bg_io(cache);
5177 			spin_lock(&cur_trans->dirty_bgs_lock);
5178 		}
5179 
5180 		list_del_init(&cache->dirty_list);
5181 		spin_lock(&cache->lock);
5182 		cache->disk_cache_state = BTRFS_DC_ERROR;
5183 		spin_unlock(&cache->lock);
5184 
5185 		spin_unlock(&cur_trans->dirty_bgs_lock);
5186 		btrfs_put_block_group(cache);
5187 		btrfs_delayed_refs_rsv_release(fs_info, 1);
5188 		spin_lock(&cur_trans->dirty_bgs_lock);
5189 	}
5190 	spin_unlock(&cur_trans->dirty_bgs_lock);
5191 
5192 	/*
5193 	 * Refer to the definition of io_bgs member for details why it's safe
5194 	 * to use it without any locking
5195 	 */
5196 	while (!list_empty(&cur_trans->io_bgs)) {
5197 		cache = list_first_entry(&cur_trans->io_bgs,
5198 					 struct btrfs_block_group,
5199 					 io_list);
5200 
5201 		list_del_init(&cache->io_list);
5202 		spin_lock(&cache->lock);
5203 		cache->disk_cache_state = BTRFS_DC_ERROR;
5204 		spin_unlock(&cache->lock);
5205 		btrfs_cleanup_bg_io(cache);
5206 	}
5207 }
5208 
5209 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
5210 				   struct btrfs_fs_info *fs_info)
5211 {
5212 	struct btrfs_device *dev, *tmp;
5213 
5214 	btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
5215 	ASSERT(list_empty(&cur_trans->dirty_bgs));
5216 	ASSERT(list_empty(&cur_trans->io_bgs));
5217 
5218 	list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
5219 				 post_commit_list) {
5220 		list_del_init(&dev->post_commit_list);
5221 	}
5222 
5223 	btrfs_destroy_delayed_refs(cur_trans, fs_info);
5224 
5225 	cur_trans->state = TRANS_STATE_COMMIT_START;
5226 	wake_up(&fs_info->transaction_blocked_wait);
5227 
5228 	cur_trans->state = TRANS_STATE_UNBLOCKED;
5229 	wake_up(&fs_info->transaction_wait);
5230 
5231 	btrfs_destroy_delayed_inodes(fs_info);
5232 
5233 	btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
5234 				     EXTENT_DIRTY);
5235 	btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
5236 
5237 	btrfs_free_redirty_list(cur_trans);
5238 
5239 	cur_trans->state =TRANS_STATE_COMPLETED;
5240 	wake_up(&cur_trans->commit_wait);
5241 }
5242 
5243 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
5244 {
5245 	struct btrfs_transaction *t;
5246 
5247 	mutex_lock(&fs_info->transaction_kthread_mutex);
5248 
5249 	spin_lock(&fs_info->trans_lock);
5250 	while (!list_empty(&fs_info->trans_list)) {
5251 		t = list_first_entry(&fs_info->trans_list,
5252 				     struct btrfs_transaction, list);
5253 		if (t->state >= TRANS_STATE_COMMIT_START) {
5254 			refcount_inc(&t->use_count);
5255 			spin_unlock(&fs_info->trans_lock);
5256 			btrfs_wait_for_commit(fs_info, t->transid);
5257 			btrfs_put_transaction(t);
5258 			spin_lock(&fs_info->trans_lock);
5259 			continue;
5260 		}
5261 		if (t == fs_info->running_transaction) {
5262 			t->state = TRANS_STATE_COMMIT_DOING;
5263 			spin_unlock(&fs_info->trans_lock);
5264 			/*
5265 			 * We wait for 0 num_writers since we don't hold a trans
5266 			 * handle open currently for this transaction.
5267 			 */
5268 			wait_event(t->writer_wait,
5269 				   atomic_read(&t->num_writers) == 0);
5270 		} else {
5271 			spin_unlock(&fs_info->trans_lock);
5272 		}
5273 		btrfs_cleanup_one_transaction(t, fs_info);
5274 
5275 		spin_lock(&fs_info->trans_lock);
5276 		if (t == fs_info->running_transaction)
5277 			fs_info->running_transaction = NULL;
5278 		list_del_init(&t->list);
5279 		spin_unlock(&fs_info->trans_lock);
5280 
5281 		btrfs_put_transaction(t);
5282 		trace_btrfs_transaction_commit(fs_info);
5283 		spin_lock(&fs_info->trans_lock);
5284 	}
5285 	spin_unlock(&fs_info->trans_lock);
5286 	btrfs_destroy_all_ordered_extents(fs_info);
5287 	btrfs_destroy_delayed_inodes(fs_info);
5288 	btrfs_assert_delayed_root_empty(fs_info);
5289 	btrfs_destroy_all_delalloc_inodes(fs_info);
5290 	btrfs_drop_all_logs(fs_info);
5291 	mutex_unlock(&fs_info->transaction_kthread_mutex);
5292 
5293 	return 0;
5294 }
5295 
5296 int btrfs_init_root_free_objectid(struct btrfs_root *root)
5297 {
5298 	struct btrfs_path *path;
5299 	int ret;
5300 	struct extent_buffer *l;
5301 	struct btrfs_key search_key;
5302 	struct btrfs_key found_key;
5303 	int slot;
5304 
5305 	path = btrfs_alloc_path();
5306 	if (!path)
5307 		return -ENOMEM;
5308 
5309 	search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
5310 	search_key.type = -1;
5311 	search_key.offset = (u64)-1;
5312 	ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5313 	if (ret < 0)
5314 		goto error;
5315 	BUG_ON(ret == 0); /* Corruption */
5316 	if (path->slots[0] > 0) {
5317 		slot = path->slots[0] - 1;
5318 		l = path->nodes[0];
5319 		btrfs_item_key_to_cpu(l, &found_key, slot);
5320 		root->free_objectid = max_t(u64, found_key.objectid + 1,
5321 					    BTRFS_FIRST_FREE_OBJECTID);
5322 	} else {
5323 		root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5324 	}
5325 	ret = 0;
5326 error:
5327 	btrfs_free_path(path);
5328 	return ret;
5329 }
5330 
5331 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5332 {
5333 	int ret;
5334 	mutex_lock(&root->objectid_mutex);
5335 
5336 	if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5337 		btrfs_warn(root->fs_info,
5338 			   "the objectid of root %llu reaches its highest value",
5339 			   root->root_key.objectid);
5340 		ret = -ENOSPC;
5341 		goto out;
5342 	}
5343 
5344 	*objectid = root->free_objectid++;
5345 	ret = 0;
5346 out:
5347 	mutex_unlock(&root->objectid_mutex);
5348 	return ret;
5349 }
5350