xref: /openbmc/linux/fs/btrfs/disk-io.c (revision 1fd02f66)
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/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/workqueue.h>
11 #include <linux/kthread.h>
12 #include <linux/slab.h>
13 #include <linux/migrate.h>
14 #include <linux/ratelimit.h>
15 #include <linux/uuid.h>
16 #include <linux/semaphore.h>
17 #include <linux/error-injection.h>
18 #include <linux/crc32c.h>
19 #include <linux/sched/mm.h>
20 #include <asm/unaligned.h>
21 #include <crypto/hash.h>
22 #include "ctree.h"
23 #include "disk-io.h"
24 #include "transaction.h"
25 #include "btrfs_inode.h"
26 #include "volumes.h"
27 #include "print-tree.h"
28 #include "locking.h"
29 #include "tree-log.h"
30 #include "free-space-cache.h"
31 #include "free-space-tree.h"
32 #include "check-integrity.h"
33 #include "rcu-string.h"
34 #include "dev-replace.h"
35 #include "raid56.h"
36 #include "sysfs.h"
37 #include "qgroup.h"
38 #include "compression.h"
39 #include "tree-checker.h"
40 #include "ref-verify.h"
41 #include "block-group.h"
42 #include "discard.h"
43 #include "space-info.h"
44 #include "zoned.h"
45 #include "subpage.h"
46 
47 #define BTRFS_SUPER_FLAG_SUPP	(BTRFS_HEADER_FLAG_WRITTEN |\
48 				 BTRFS_HEADER_FLAG_RELOC |\
49 				 BTRFS_SUPER_FLAG_ERROR |\
50 				 BTRFS_SUPER_FLAG_SEEDING |\
51 				 BTRFS_SUPER_FLAG_METADUMP |\
52 				 BTRFS_SUPER_FLAG_METADUMP_V2)
53 
54 static void end_workqueue_fn(struct btrfs_work *work);
55 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
56 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
57 				      struct btrfs_fs_info *fs_info);
58 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
59 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
60 					struct extent_io_tree *dirty_pages,
61 					int mark);
62 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
63 				       struct extent_io_tree *pinned_extents);
64 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
65 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
66 
67 /*
68  * btrfs_end_io_wq structs are used to do processing in task context when an IO
69  * is complete.  This is used during reads to verify checksums, and it is used
70  * by writes to insert metadata for new file extents after IO is complete.
71  */
72 struct btrfs_end_io_wq {
73 	struct bio *bio;
74 	bio_end_io_t *end_io;
75 	void *private;
76 	struct btrfs_fs_info *info;
77 	blk_status_t status;
78 	enum btrfs_wq_endio_type metadata;
79 	struct btrfs_work work;
80 };
81 
82 static struct kmem_cache *btrfs_end_io_wq_cache;
83 
84 int __init btrfs_end_io_wq_init(void)
85 {
86 	btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
87 					sizeof(struct btrfs_end_io_wq),
88 					0,
89 					SLAB_MEM_SPREAD,
90 					NULL);
91 	if (!btrfs_end_io_wq_cache)
92 		return -ENOMEM;
93 	return 0;
94 }
95 
96 void __cold btrfs_end_io_wq_exit(void)
97 {
98 	kmem_cache_destroy(btrfs_end_io_wq_cache);
99 }
100 
101 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
102 {
103 	if (fs_info->csum_shash)
104 		crypto_free_shash(fs_info->csum_shash);
105 }
106 
107 /*
108  * async submit bios are used to offload expensive checksumming
109  * onto the worker threads.  They checksum file and metadata bios
110  * just before they are sent down the IO stack.
111  */
112 struct async_submit_bio {
113 	struct inode *inode;
114 	struct bio *bio;
115 	extent_submit_bio_start_t *submit_bio_start;
116 	int mirror_num;
117 
118 	/* Optional parameter for submit_bio_start used by direct io */
119 	u64 dio_file_offset;
120 	struct btrfs_work work;
121 	blk_status_t status;
122 };
123 
124 /*
125  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
126  * eb, the lockdep key is determined by the btrfs_root it belongs to and
127  * the level the eb occupies in the tree.
128  *
129  * Different roots are used for different purposes and may nest inside each
130  * other and they require separate keysets.  As lockdep keys should be
131  * static, assign keysets according to the purpose of the root as indicated
132  * by btrfs_root->root_key.objectid.  This ensures that all special purpose
133  * roots have separate keysets.
134  *
135  * Lock-nesting across peer nodes is always done with the immediate parent
136  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
137  * subclass to avoid triggering lockdep warning in such cases.
138  *
139  * The key is set by the readpage_end_io_hook after the buffer has passed
140  * csum validation but before the pages are unlocked.  It is also set by
141  * btrfs_init_new_buffer on freshly allocated blocks.
142  *
143  * We also add a check to make sure the highest level of the tree is the
144  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
145  * needs update as well.
146  */
147 #ifdef CONFIG_DEBUG_LOCK_ALLOC
148 # if BTRFS_MAX_LEVEL != 8
149 #  error
150 # endif
151 
152 #define DEFINE_LEVEL(stem, level)					\
153 	.names[level] = "btrfs-" stem "-0" #level,
154 
155 #define DEFINE_NAME(stem)						\
156 	DEFINE_LEVEL(stem, 0)						\
157 	DEFINE_LEVEL(stem, 1)						\
158 	DEFINE_LEVEL(stem, 2)						\
159 	DEFINE_LEVEL(stem, 3)						\
160 	DEFINE_LEVEL(stem, 4)						\
161 	DEFINE_LEVEL(stem, 5)						\
162 	DEFINE_LEVEL(stem, 6)						\
163 	DEFINE_LEVEL(stem, 7)
164 
165 static struct btrfs_lockdep_keyset {
166 	u64			id;		/* root objectid */
167 	/* Longest entry: btrfs-free-space-00 */
168 	char			names[BTRFS_MAX_LEVEL][20];
169 	struct lock_class_key	keys[BTRFS_MAX_LEVEL];
170 } btrfs_lockdep_keysets[] = {
171 	{ .id = BTRFS_ROOT_TREE_OBJECTID,	DEFINE_NAME("root")	},
172 	{ .id = BTRFS_EXTENT_TREE_OBJECTID,	DEFINE_NAME("extent")	},
173 	{ .id = BTRFS_CHUNK_TREE_OBJECTID,	DEFINE_NAME("chunk")	},
174 	{ .id = BTRFS_DEV_TREE_OBJECTID,	DEFINE_NAME("dev")	},
175 	{ .id = BTRFS_CSUM_TREE_OBJECTID,	DEFINE_NAME("csum")	},
176 	{ .id = BTRFS_QUOTA_TREE_OBJECTID,	DEFINE_NAME("quota")	},
177 	{ .id = BTRFS_TREE_LOG_OBJECTID,	DEFINE_NAME("log")	},
178 	{ .id = BTRFS_TREE_RELOC_OBJECTID,	DEFINE_NAME("treloc")	},
179 	{ .id = BTRFS_DATA_RELOC_TREE_OBJECTID,	DEFINE_NAME("dreloc")	},
180 	{ .id = BTRFS_UUID_TREE_OBJECTID,	DEFINE_NAME("uuid")	},
181 	{ .id = BTRFS_FREE_SPACE_TREE_OBJECTID,	DEFINE_NAME("free-space") },
182 	{ .id = 0,				DEFINE_NAME("tree")	},
183 };
184 
185 #undef DEFINE_LEVEL
186 #undef DEFINE_NAME
187 
188 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
189 				    int level)
190 {
191 	struct btrfs_lockdep_keyset *ks;
192 
193 	BUG_ON(level >= ARRAY_SIZE(ks->keys));
194 
195 	/* find the matching keyset, id 0 is the default entry */
196 	for (ks = btrfs_lockdep_keysets; ks->id; ks++)
197 		if (ks->id == objectid)
198 			break;
199 
200 	lockdep_set_class_and_name(&eb->lock,
201 				   &ks->keys[level], ks->names[level]);
202 }
203 
204 #endif
205 
206 /*
207  * Compute the csum of a btree block and store the result to provided buffer.
208  */
209 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
210 {
211 	struct btrfs_fs_info *fs_info = buf->fs_info;
212 	const int num_pages = num_extent_pages(buf);
213 	const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
214 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
215 	char *kaddr;
216 	int i;
217 
218 	shash->tfm = fs_info->csum_shash;
219 	crypto_shash_init(shash);
220 	kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
221 	crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
222 			    first_page_part - BTRFS_CSUM_SIZE);
223 
224 	for (i = 1; i < num_pages; i++) {
225 		kaddr = page_address(buf->pages[i]);
226 		crypto_shash_update(shash, kaddr, PAGE_SIZE);
227 	}
228 	memset(result, 0, BTRFS_CSUM_SIZE);
229 	crypto_shash_final(shash, result);
230 }
231 
232 /*
233  * we can't consider a given block up to date unless the transid of the
234  * block matches the transid in the parent node's pointer.  This is how we
235  * detect blocks that either didn't get written at all or got written
236  * in the wrong place.
237  */
238 static int verify_parent_transid(struct extent_io_tree *io_tree,
239 				 struct extent_buffer *eb, u64 parent_transid,
240 				 int atomic)
241 {
242 	struct extent_state *cached_state = NULL;
243 	int ret;
244 
245 	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
246 		return 0;
247 
248 	if (atomic)
249 		return -EAGAIN;
250 
251 	lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
252 			 &cached_state);
253 	if (extent_buffer_uptodate(eb) &&
254 	    btrfs_header_generation(eb) == parent_transid) {
255 		ret = 0;
256 		goto out;
257 	}
258 	btrfs_err_rl(eb->fs_info,
259 		"parent transid verify failed on %llu wanted %llu found %llu",
260 			eb->start,
261 			parent_transid, btrfs_header_generation(eb));
262 	ret = 1;
263 	clear_extent_buffer_uptodate(eb);
264 out:
265 	unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
266 			     &cached_state);
267 	return ret;
268 }
269 
270 static bool btrfs_supported_super_csum(u16 csum_type)
271 {
272 	switch (csum_type) {
273 	case BTRFS_CSUM_TYPE_CRC32:
274 	case BTRFS_CSUM_TYPE_XXHASH:
275 	case BTRFS_CSUM_TYPE_SHA256:
276 	case BTRFS_CSUM_TYPE_BLAKE2:
277 		return true;
278 	default:
279 		return false;
280 	}
281 }
282 
283 /*
284  * Return 0 if the superblock checksum type matches the checksum value of that
285  * algorithm. Pass the raw disk superblock data.
286  */
287 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
288 				  char *raw_disk_sb)
289 {
290 	struct btrfs_super_block *disk_sb =
291 		(struct btrfs_super_block *)raw_disk_sb;
292 	char result[BTRFS_CSUM_SIZE];
293 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
294 
295 	shash->tfm = fs_info->csum_shash;
296 
297 	/*
298 	 * The super_block structure does not span the whole
299 	 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
300 	 * filled with zeros and is included in the checksum.
301 	 */
302 	crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
303 			    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
304 
305 	if (memcmp(disk_sb->csum, result, fs_info->csum_size))
306 		return 1;
307 
308 	return 0;
309 }
310 
311 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
312 			   struct btrfs_key *first_key, u64 parent_transid)
313 {
314 	struct btrfs_fs_info *fs_info = eb->fs_info;
315 	int found_level;
316 	struct btrfs_key found_key;
317 	int ret;
318 
319 	found_level = btrfs_header_level(eb);
320 	if (found_level != level) {
321 		WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
322 		     KERN_ERR "BTRFS: tree level check failed\n");
323 		btrfs_err(fs_info,
324 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
325 			  eb->start, level, found_level);
326 		return -EIO;
327 	}
328 
329 	if (!first_key)
330 		return 0;
331 
332 	/*
333 	 * For live tree block (new tree blocks in current transaction),
334 	 * we need proper lock context to avoid race, which is impossible here.
335 	 * So we only checks tree blocks which is read from disk, whose
336 	 * generation <= fs_info->last_trans_committed.
337 	 */
338 	if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
339 		return 0;
340 
341 	/* We have @first_key, so this @eb must have at least one item */
342 	if (btrfs_header_nritems(eb) == 0) {
343 		btrfs_err(fs_info,
344 		"invalid tree nritems, bytenr=%llu nritems=0 expect >0",
345 			  eb->start);
346 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
347 		return -EUCLEAN;
348 	}
349 
350 	if (found_level)
351 		btrfs_node_key_to_cpu(eb, &found_key, 0);
352 	else
353 		btrfs_item_key_to_cpu(eb, &found_key, 0);
354 	ret = btrfs_comp_cpu_keys(first_key, &found_key);
355 
356 	if (ret) {
357 		WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
358 		     KERN_ERR "BTRFS: tree first key check failed\n");
359 		btrfs_err(fs_info,
360 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
361 			  eb->start, parent_transid, first_key->objectid,
362 			  first_key->type, first_key->offset,
363 			  found_key.objectid, found_key.type,
364 			  found_key.offset);
365 	}
366 	return ret;
367 }
368 
369 /*
370  * helper to read a given tree block, doing retries as required when
371  * the checksums don't match and we have alternate mirrors to try.
372  *
373  * @parent_transid:	expected transid, skip check if 0
374  * @level:		expected level, mandatory check
375  * @first_key:		expected key of first slot, skip check if NULL
376  */
377 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
378 					  u64 parent_transid, int level,
379 					  struct btrfs_key *first_key)
380 {
381 	struct btrfs_fs_info *fs_info = eb->fs_info;
382 	struct extent_io_tree *io_tree;
383 	int failed = 0;
384 	int ret;
385 	int num_copies = 0;
386 	int mirror_num = 0;
387 	int failed_mirror = 0;
388 
389 	io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
390 	while (1) {
391 		clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
392 		ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
393 		if (!ret) {
394 			if (verify_parent_transid(io_tree, eb,
395 						   parent_transid, 0))
396 				ret = -EIO;
397 			else if (btrfs_verify_level_key(eb, level,
398 						first_key, parent_transid))
399 				ret = -EUCLEAN;
400 			else
401 				break;
402 		}
403 
404 		num_copies = btrfs_num_copies(fs_info,
405 					      eb->start, eb->len);
406 		if (num_copies == 1)
407 			break;
408 
409 		if (!failed_mirror) {
410 			failed = 1;
411 			failed_mirror = eb->read_mirror;
412 		}
413 
414 		mirror_num++;
415 		if (mirror_num == failed_mirror)
416 			mirror_num++;
417 
418 		if (mirror_num > num_copies)
419 			break;
420 	}
421 
422 	if (failed && !ret && failed_mirror)
423 		btrfs_repair_eb_io_failure(eb, failed_mirror);
424 
425 	return ret;
426 }
427 
428 static int csum_one_extent_buffer(struct extent_buffer *eb)
429 {
430 	struct btrfs_fs_info *fs_info = eb->fs_info;
431 	u8 result[BTRFS_CSUM_SIZE];
432 	int ret;
433 
434 	ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
435 				    offsetof(struct btrfs_header, fsid),
436 				    BTRFS_FSID_SIZE) == 0);
437 	csum_tree_block(eb, result);
438 
439 	if (btrfs_header_level(eb))
440 		ret = btrfs_check_node(eb);
441 	else
442 		ret = btrfs_check_leaf_full(eb);
443 
444 	if (ret < 0)
445 		goto error;
446 
447 	/*
448 	 * Also check the generation, the eb reached here must be newer than
449 	 * last committed. Or something seriously wrong happened.
450 	 */
451 	if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) {
452 		ret = -EUCLEAN;
453 		btrfs_err(fs_info,
454 			"block=%llu bad generation, have %llu expect > %llu",
455 			  eb->start, btrfs_header_generation(eb),
456 			  fs_info->last_trans_committed);
457 		goto error;
458 	}
459 	write_extent_buffer(eb, result, 0, fs_info->csum_size);
460 
461 	return 0;
462 
463 error:
464 	btrfs_print_tree(eb, 0);
465 	btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
466 		  eb->start);
467 	WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
468 	return ret;
469 }
470 
471 /* Checksum all dirty extent buffers in one bio_vec */
472 static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
473 				      struct bio_vec *bvec)
474 {
475 	struct page *page = bvec->bv_page;
476 	u64 bvec_start = page_offset(page) + bvec->bv_offset;
477 	u64 cur;
478 	int ret = 0;
479 
480 	for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
481 	     cur += fs_info->nodesize) {
482 		struct extent_buffer *eb;
483 		bool uptodate;
484 
485 		eb = find_extent_buffer(fs_info, cur);
486 		uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
487 						       fs_info->nodesize);
488 
489 		/* A dirty eb shouldn't disappear from buffer_radix */
490 		if (WARN_ON(!eb))
491 			return -EUCLEAN;
492 
493 		if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
494 			free_extent_buffer(eb);
495 			return -EUCLEAN;
496 		}
497 		if (WARN_ON(!uptodate)) {
498 			free_extent_buffer(eb);
499 			return -EUCLEAN;
500 		}
501 
502 		ret = csum_one_extent_buffer(eb);
503 		free_extent_buffer(eb);
504 		if (ret < 0)
505 			return ret;
506 	}
507 	return ret;
508 }
509 
510 /*
511  * Checksum a dirty tree block before IO.  This has extra checks to make sure
512  * we only fill in the checksum field in the first page of a multi-page block.
513  * For subpage extent buffers we need bvec to also read the offset in the page.
514  */
515 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
516 {
517 	struct page *page = bvec->bv_page;
518 	u64 start = page_offset(page);
519 	u64 found_start;
520 	struct extent_buffer *eb;
521 
522 	if (fs_info->sectorsize < PAGE_SIZE)
523 		return csum_dirty_subpage_buffers(fs_info, bvec);
524 
525 	eb = (struct extent_buffer *)page->private;
526 	if (page != eb->pages[0])
527 		return 0;
528 
529 	found_start = btrfs_header_bytenr(eb);
530 
531 	if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
532 		WARN_ON(found_start != 0);
533 		return 0;
534 	}
535 
536 	/*
537 	 * Please do not consolidate these warnings into a single if.
538 	 * It is useful to know what went wrong.
539 	 */
540 	if (WARN_ON(found_start != start))
541 		return -EUCLEAN;
542 	if (WARN_ON(!PageUptodate(page)))
543 		return -EUCLEAN;
544 
545 	return csum_one_extent_buffer(eb);
546 }
547 
548 static int check_tree_block_fsid(struct extent_buffer *eb)
549 {
550 	struct btrfs_fs_info *fs_info = eb->fs_info;
551 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
552 	u8 fsid[BTRFS_FSID_SIZE];
553 	u8 *metadata_uuid;
554 
555 	read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
556 			   BTRFS_FSID_SIZE);
557 	/*
558 	 * Checking the incompat flag is only valid for the current fs. For
559 	 * seed devices it's forbidden to have their uuid changed so reading
560 	 * ->fsid in this case is fine
561 	 */
562 	if (btrfs_fs_incompat(fs_info, METADATA_UUID))
563 		metadata_uuid = fs_devices->metadata_uuid;
564 	else
565 		metadata_uuid = fs_devices->fsid;
566 
567 	if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
568 		return 0;
569 
570 	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
571 		if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
572 			return 0;
573 
574 	return 1;
575 }
576 
577 /* Do basic extent buffer checks at read time */
578 static int validate_extent_buffer(struct extent_buffer *eb)
579 {
580 	struct btrfs_fs_info *fs_info = eb->fs_info;
581 	u64 found_start;
582 	const u32 csum_size = fs_info->csum_size;
583 	u8 found_level;
584 	u8 result[BTRFS_CSUM_SIZE];
585 	const u8 *header_csum;
586 	int ret = 0;
587 
588 	found_start = btrfs_header_bytenr(eb);
589 	if (found_start != eb->start) {
590 		btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
591 			     eb->start, found_start);
592 		ret = -EIO;
593 		goto out;
594 	}
595 	if (check_tree_block_fsid(eb)) {
596 		btrfs_err_rl(fs_info, "bad fsid on block %llu",
597 			     eb->start);
598 		ret = -EIO;
599 		goto out;
600 	}
601 	found_level = btrfs_header_level(eb);
602 	if (found_level >= BTRFS_MAX_LEVEL) {
603 		btrfs_err(fs_info, "bad tree block level %d on %llu",
604 			  (int)btrfs_header_level(eb), eb->start);
605 		ret = -EIO;
606 		goto out;
607 	}
608 
609 	csum_tree_block(eb, result);
610 	header_csum = page_address(eb->pages[0]) +
611 		get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
612 
613 	if (memcmp(result, header_csum, csum_size) != 0) {
614 		btrfs_warn_rl(fs_info,
615 	"checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
616 			      eb->start,
617 			      CSUM_FMT_VALUE(csum_size, header_csum),
618 			      CSUM_FMT_VALUE(csum_size, result),
619 			      btrfs_header_level(eb));
620 		ret = -EUCLEAN;
621 		goto out;
622 	}
623 
624 	/*
625 	 * If this is a leaf block and it is corrupt, set the corrupt bit so
626 	 * that we don't try and read the other copies of this block, just
627 	 * return -EIO.
628 	 */
629 	if (found_level == 0 && btrfs_check_leaf_full(eb)) {
630 		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
631 		ret = -EIO;
632 	}
633 
634 	if (found_level > 0 && btrfs_check_node(eb))
635 		ret = -EIO;
636 
637 	if (!ret)
638 		set_extent_buffer_uptodate(eb);
639 	else
640 		btrfs_err(fs_info,
641 			  "block=%llu read time tree block corruption detected",
642 			  eb->start);
643 out:
644 	return ret;
645 }
646 
647 static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
648 				   int mirror)
649 {
650 	struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
651 	struct extent_buffer *eb;
652 	bool reads_done;
653 	int ret = 0;
654 
655 	/*
656 	 * We don't allow bio merge for subpage metadata read, so we should
657 	 * only get one eb for each endio hook.
658 	 */
659 	ASSERT(end == start + fs_info->nodesize - 1);
660 	ASSERT(PagePrivate(page));
661 
662 	eb = find_extent_buffer(fs_info, start);
663 	/*
664 	 * When we are reading one tree block, eb must have been inserted into
665 	 * the radix tree. If not, something is wrong.
666 	 */
667 	ASSERT(eb);
668 
669 	reads_done = atomic_dec_and_test(&eb->io_pages);
670 	/* Subpage read must finish in page read */
671 	ASSERT(reads_done);
672 
673 	eb->read_mirror = mirror;
674 	if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
675 		ret = -EIO;
676 		goto err;
677 	}
678 	ret = validate_extent_buffer(eb);
679 	if (ret < 0)
680 		goto err;
681 
682 	set_extent_buffer_uptodate(eb);
683 
684 	free_extent_buffer(eb);
685 	return ret;
686 err:
687 	/*
688 	 * end_bio_extent_readpage decrements io_pages in case of error,
689 	 * make sure it has something to decrement.
690 	 */
691 	atomic_inc(&eb->io_pages);
692 	clear_extent_buffer_uptodate(eb);
693 	free_extent_buffer(eb);
694 	return ret;
695 }
696 
697 int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio,
698 				   struct page *page, u64 start, u64 end,
699 				   int mirror)
700 {
701 	struct extent_buffer *eb;
702 	int ret = 0;
703 	int reads_done;
704 
705 	ASSERT(page->private);
706 
707 	if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE)
708 		return validate_subpage_buffer(page, start, end, mirror);
709 
710 	eb = (struct extent_buffer *)page->private;
711 
712 	/*
713 	 * The pending IO might have been the only thing that kept this buffer
714 	 * in memory.  Make sure we have a ref for all this other checks
715 	 */
716 	atomic_inc(&eb->refs);
717 
718 	reads_done = atomic_dec_and_test(&eb->io_pages);
719 	if (!reads_done)
720 		goto err;
721 
722 	eb->read_mirror = mirror;
723 	if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
724 		ret = -EIO;
725 		goto err;
726 	}
727 	ret = validate_extent_buffer(eb);
728 err:
729 	if (ret) {
730 		/*
731 		 * our io error hook is going to dec the io pages
732 		 * again, we have to make sure it has something
733 		 * to decrement
734 		 */
735 		atomic_inc(&eb->io_pages);
736 		clear_extent_buffer_uptodate(eb);
737 	}
738 	free_extent_buffer(eb);
739 
740 	return ret;
741 }
742 
743 static void end_workqueue_bio(struct bio *bio)
744 {
745 	struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
746 	struct btrfs_fs_info *fs_info;
747 	struct btrfs_workqueue *wq;
748 
749 	fs_info = end_io_wq->info;
750 	end_io_wq->status = bio->bi_status;
751 
752 	if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
753 		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
754 			wq = fs_info->endio_meta_write_workers;
755 		else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
756 			wq = fs_info->endio_freespace_worker;
757 		else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
758 			wq = fs_info->endio_raid56_workers;
759 		else
760 			wq = fs_info->endio_write_workers;
761 	} else {
762 		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
763 			wq = fs_info->endio_raid56_workers;
764 		else if (end_io_wq->metadata)
765 			wq = fs_info->endio_meta_workers;
766 		else
767 			wq = fs_info->endio_workers;
768 	}
769 
770 	btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
771 	btrfs_queue_work(wq, &end_io_wq->work);
772 }
773 
774 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
775 			enum btrfs_wq_endio_type metadata)
776 {
777 	struct btrfs_end_io_wq *end_io_wq;
778 
779 	end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
780 	if (!end_io_wq)
781 		return BLK_STS_RESOURCE;
782 
783 	end_io_wq->private = bio->bi_private;
784 	end_io_wq->end_io = bio->bi_end_io;
785 	end_io_wq->info = info;
786 	end_io_wq->status = 0;
787 	end_io_wq->bio = bio;
788 	end_io_wq->metadata = metadata;
789 
790 	bio->bi_private = end_io_wq;
791 	bio->bi_end_io = end_workqueue_bio;
792 	return 0;
793 }
794 
795 static void run_one_async_start(struct btrfs_work *work)
796 {
797 	struct async_submit_bio *async;
798 	blk_status_t ret;
799 
800 	async = container_of(work, struct  async_submit_bio, work);
801 	ret = async->submit_bio_start(async->inode, async->bio,
802 				      async->dio_file_offset);
803 	if (ret)
804 		async->status = ret;
805 }
806 
807 /*
808  * In order to insert checksums into the metadata in large chunks, we wait
809  * until bio submission time.   All the pages in the bio are checksummed and
810  * sums are attached onto the ordered extent record.
811  *
812  * At IO completion time the csums attached on the ordered extent record are
813  * inserted into the tree.
814  */
815 static void run_one_async_done(struct btrfs_work *work)
816 {
817 	struct async_submit_bio *async;
818 	struct inode *inode;
819 	blk_status_t ret;
820 
821 	async = container_of(work, struct  async_submit_bio, work);
822 	inode = async->inode;
823 
824 	/* If an error occurred we just want to clean up the bio and move on */
825 	if (async->status) {
826 		async->bio->bi_status = async->status;
827 		bio_endio(async->bio);
828 		return;
829 	}
830 
831 	/*
832 	 * All of the bios that pass through here are from async helpers.
833 	 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
834 	 * This changes nothing when cgroups aren't in use.
835 	 */
836 	async->bio->bi_opf |= REQ_CGROUP_PUNT;
837 	ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
838 	if (ret) {
839 		async->bio->bi_status = ret;
840 		bio_endio(async->bio);
841 	}
842 }
843 
844 static void run_one_async_free(struct btrfs_work *work)
845 {
846 	struct async_submit_bio *async;
847 
848 	async = container_of(work, struct  async_submit_bio, work);
849 	kfree(async);
850 }
851 
852 blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio,
853 				 int mirror_num, unsigned long bio_flags,
854 				 u64 dio_file_offset,
855 				 extent_submit_bio_start_t *submit_bio_start)
856 {
857 	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
858 	struct async_submit_bio *async;
859 
860 	async = kmalloc(sizeof(*async), GFP_NOFS);
861 	if (!async)
862 		return BLK_STS_RESOURCE;
863 
864 	async->inode = inode;
865 	async->bio = bio;
866 	async->mirror_num = mirror_num;
867 	async->submit_bio_start = submit_bio_start;
868 
869 	btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
870 			run_one_async_free);
871 
872 	async->dio_file_offset = dio_file_offset;
873 
874 	async->status = 0;
875 
876 	if (op_is_sync(bio->bi_opf))
877 		btrfs_set_work_high_priority(&async->work);
878 
879 	btrfs_queue_work(fs_info->workers, &async->work);
880 	return 0;
881 }
882 
883 static blk_status_t btree_csum_one_bio(struct bio *bio)
884 {
885 	struct bio_vec *bvec;
886 	struct btrfs_root *root;
887 	int ret = 0;
888 	struct bvec_iter_all iter_all;
889 
890 	ASSERT(!bio_flagged(bio, BIO_CLONED));
891 	bio_for_each_segment_all(bvec, bio, iter_all) {
892 		root = BTRFS_I(bvec->bv_page->mapping->host)->root;
893 		ret = csum_dirty_buffer(root->fs_info, bvec);
894 		if (ret)
895 			break;
896 	}
897 
898 	return errno_to_blk_status(ret);
899 }
900 
901 static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
902 					   u64 dio_file_offset)
903 {
904 	/*
905 	 * when we're called for a write, we're already in the async
906 	 * submission context.  Just jump into btrfs_map_bio
907 	 */
908 	return btree_csum_one_bio(bio);
909 }
910 
911 static bool should_async_write(struct btrfs_fs_info *fs_info,
912 			     struct btrfs_inode *bi)
913 {
914 	if (btrfs_is_zoned(fs_info))
915 		return false;
916 	if (atomic_read(&bi->sync_writers))
917 		return false;
918 	if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
919 		return false;
920 	return true;
921 }
922 
923 blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio,
924 				       int mirror_num, unsigned long bio_flags)
925 {
926 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
927 	blk_status_t ret;
928 
929 	if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
930 		/*
931 		 * called for a read, do the setup so that checksum validation
932 		 * can happen in the async kernel threads
933 		 */
934 		ret = btrfs_bio_wq_end_io(fs_info, bio,
935 					  BTRFS_WQ_ENDIO_METADATA);
936 		if (ret)
937 			goto out_w_error;
938 		ret = btrfs_map_bio(fs_info, bio, mirror_num);
939 	} else if (!should_async_write(fs_info, BTRFS_I(inode))) {
940 		ret = btree_csum_one_bio(bio);
941 		if (ret)
942 			goto out_w_error;
943 		ret = btrfs_map_bio(fs_info, bio, mirror_num);
944 	} else {
945 		/*
946 		 * kthread helpers are used to submit writes so that
947 		 * checksumming can happen in parallel across all CPUs
948 		 */
949 		ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
950 					  0, btree_submit_bio_start);
951 	}
952 
953 	if (ret)
954 		goto out_w_error;
955 	return 0;
956 
957 out_w_error:
958 	bio->bi_status = ret;
959 	bio_endio(bio);
960 	return ret;
961 }
962 
963 #ifdef CONFIG_MIGRATION
964 static int btree_migratepage(struct address_space *mapping,
965 			struct page *newpage, struct page *page,
966 			enum migrate_mode mode)
967 {
968 	/*
969 	 * we can't safely write a btree page from here,
970 	 * we haven't done the locking hook
971 	 */
972 	if (PageDirty(page))
973 		return -EAGAIN;
974 	/*
975 	 * Buffers may be managed in a filesystem specific way.
976 	 * We must have no buffers or drop them.
977 	 */
978 	if (page_has_private(page) &&
979 	    !try_to_release_page(page, GFP_KERNEL))
980 		return -EAGAIN;
981 	return migrate_page(mapping, newpage, page, mode);
982 }
983 #endif
984 
985 
986 static int btree_writepages(struct address_space *mapping,
987 			    struct writeback_control *wbc)
988 {
989 	struct btrfs_fs_info *fs_info;
990 	int ret;
991 
992 	if (wbc->sync_mode == WB_SYNC_NONE) {
993 
994 		if (wbc->for_kupdate)
995 			return 0;
996 
997 		fs_info = BTRFS_I(mapping->host)->root->fs_info;
998 		/* this is a bit racy, but that's ok */
999 		ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1000 					     BTRFS_DIRTY_METADATA_THRESH,
1001 					     fs_info->dirty_metadata_batch);
1002 		if (ret < 0)
1003 			return 0;
1004 	}
1005 	return btree_write_cache_pages(mapping, wbc);
1006 }
1007 
1008 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1009 {
1010 	if (PageWriteback(page) || PageDirty(page))
1011 		return 0;
1012 
1013 	return try_release_extent_buffer(page);
1014 }
1015 
1016 static void btree_invalidate_folio(struct folio *folio, size_t offset,
1017 				 size_t length)
1018 {
1019 	struct extent_io_tree *tree;
1020 	tree = &BTRFS_I(folio->mapping->host)->io_tree;
1021 	extent_invalidate_folio(tree, folio, offset);
1022 	btree_releasepage(&folio->page, GFP_NOFS);
1023 	if (folio_get_private(folio)) {
1024 		btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
1025 			   "folio private not zero on folio %llu",
1026 			   (unsigned long long)folio_pos(folio));
1027 		folio_detach_private(folio);
1028 	}
1029 }
1030 
1031 #ifdef DEBUG
1032 static bool btree_dirty_folio(struct address_space *mapping,
1033 		struct folio *folio)
1034 {
1035 	struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
1036 	struct btrfs_subpage *subpage;
1037 	struct extent_buffer *eb;
1038 	int cur_bit = 0;
1039 	u64 page_start = folio_pos(folio);
1040 
1041 	if (fs_info->sectorsize == PAGE_SIZE) {
1042 		eb = folio_get_private(folio);
1043 		BUG_ON(!eb);
1044 		BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1045 		BUG_ON(!atomic_read(&eb->refs));
1046 		btrfs_assert_tree_write_locked(eb);
1047 		return filemap_dirty_folio(mapping, folio);
1048 	}
1049 	subpage = folio_get_private(folio);
1050 
1051 	ASSERT(subpage->dirty_bitmap);
1052 	while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
1053 		unsigned long flags;
1054 		u64 cur;
1055 		u16 tmp = (1 << cur_bit);
1056 
1057 		spin_lock_irqsave(&subpage->lock, flags);
1058 		if (!(tmp & subpage->dirty_bitmap)) {
1059 			spin_unlock_irqrestore(&subpage->lock, flags);
1060 			cur_bit++;
1061 			continue;
1062 		}
1063 		spin_unlock_irqrestore(&subpage->lock, flags);
1064 		cur = page_start + cur_bit * fs_info->sectorsize;
1065 
1066 		eb = find_extent_buffer(fs_info, cur);
1067 		ASSERT(eb);
1068 		ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1069 		ASSERT(atomic_read(&eb->refs));
1070 		btrfs_assert_tree_write_locked(eb);
1071 		free_extent_buffer(eb);
1072 
1073 		cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
1074 	}
1075 	return filemap_dirty_folio(mapping, folio);
1076 }
1077 #else
1078 #define btree_dirty_folio filemap_dirty_folio
1079 #endif
1080 
1081 static const struct address_space_operations btree_aops = {
1082 	.writepages	= btree_writepages,
1083 	.releasepage	= btree_releasepage,
1084 	.invalidate_folio = btree_invalidate_folio,
1085 #ifdef CONFIG_MIGRATION
1086 	.migratepage	= btree_migratepage,
1087 #endif
1088 	.dirty_folio = btree_dirty_folio,
1089 };
1090 
1091 struct extent_buffer *btrfs_find_create_tree_block(
1092 						struct btrfs_fs_info *fs_info,
1093 						u64 bytenr, u64 owner_root,
1094 						int level)
1095 {
1096 	if (btrfs_is_testing(fs_info))
1097 		return alloc_test_extent_buffer(fs_info, bytenr);
1098 	return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
1099 }
1100 
1101 /*
1102  * Read tree block at logical address @bytenr and do variant basic but critical
1103  * verification.
1104  *
1105  * @owner_root:		the objectid of the root owner for this block.
1106  * @parent_transid:	expected transid of this tree block, skip check if 0
1107  * @level:		expected level, mandatory check
1108  * @first_key:		expected key in slot 0, skip check if NULL
1109  */
1110 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1111 				      u64 owner_root, u64 parent_transid,
1112 				      int level, struct btrfs_key *first_key)
1113 {
1114 	struct extent_buffer *buf = NULL;
1115 	int ret;
1116 
1117 	buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
1118 	if (IS_ERR(buf))
1119 		return buf;
1120 
1121 	ret = btree_read_extent_buffer_pages(buf, parent_transid,
1122 					     level, first_key);
1123 	if (ret) {
1124 		free_extent_buffer_stale(buf);
1125 		return ERR_PTR(ret);
1126 	}
1127 	return buf;
1128 
1129 }
1130 
1131 void btrfs_clean_tree_block(struct extent_buffer *buf)
1132 {
1133 	struct btrfs_fs_info *fs_info = buf->fs_info;
1134 	if (btrfs_header_generation(buf) ==
1135 	    fs_info->running_transaction->transid) {
1136 		btrfs_assert_tree_write_locked(buf);
1137 
1138 		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1139 			percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1140 						 -buf->len,
1141 						 fs_info->dirty_metadata_batch);
1142 			clear_extent_buffer_dirty(buf);
1143 		}
1144 	}
1145 }
1146 
1147 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1148 			 u64 objectid)
1149 {
1150 	bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1151 
1152 	memset(&root->root_key, 0, sizeof(root->root_key));
1153 	memset(&root->root_item, 0, sizeof(root->root_item));
1154 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1155 	root->fs_info = fs_info;
1156 	root->root_key.objectid = objectid;
1157 	root->node = NULL;
1158 	root->commit_root = NULL;
1159 	root->state = 0;
1160 	RB_CLEAR_NODE(&root->rb_node);
1161 
1162 	root->last_trans = 0;
1163 	root->free_objectid = 0;
1164 	root->nr_delalloc_inodes = 0;
1165 	root->nr_ordered_extents = 0;
1166 	root->inode_tree = RB_ROOT;
1167 	INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1168 
1169 	btrfs_init_root_block_rsv(root);
1170 
1171 	INIT_LIST_HEAD(&root->dirty_list);
1172 	INIT_LIST_HEAD(&root->root_list);
1173 	INIT_LIST_HEAD(&root->delalloc_inodes);
1174 	INIT_LIST_HEAD(&root->delalloc_root);
1175 	INIT_LIST_HEAD(&root->ordered_extents);
1176 	INIT_LIST_HEAD(&root->ordered_root);
1177 	INIT_LIST_HEAD(&root->reloc_dirty_list);
1178 	INIT_LIST_HEAD(&root->logged_list[0]);
1179 	INIT_LIST_HEAD(&root->logged_list[1]);
1180 	spin_lock_init(&root->inode_lock);
1181 	spin_lock_init(&root->delalloc_lock);
1182 	spin_lock_init(&root->ordered_extent_lock);
1183 	spin_lock_init(&root->accounting_lock);
1184 	spin_lock_init(&root->log_extents_lock[0]);
1185 	spin_lock_init(&root->log_extents_lock[1]);
1186 	spin_lock_init(&root->qgroup_meta_rsv_lock);
1187 	mutex_init(&root->objectid_mutex);
1188 	mutex_init(&root->log_mutex);
1189 	mutex_init(&root->ordered_extent_mutex);
1190 	mutex_init(&root->delalloc_mutex);
1191 	init_waitqueue_head(&root->qgroup_flush_wait);
1192 	init_waitqueue_head(&root->log_writer_wait);
1193 	init_waitqueue_head(&root->log_commit_wait[0]);
1194 	init_waitqueue_head(&root->log_commit_wait[1]);
1195 	INIT_LIST_HEAD(&root->log_ctxs[0]);
1196 	INIT_LIST_HEAD(&root->log_ctxs[1]);
1197 	atomic_set(&root->log_commit[0], 0);
1198 	atomic_set(&root->log_commit[1], 0);
1199 	atomic_set(&root->log_writers, 0);
1200 	atomic_set(&root->log_batch, 0);
1201 	refcount_set(&root->refs, 1);
1202 	atomic_set(&root->snapshot_force_cow, 0);
1203 	atomic_set(&root->nr_swapfiles, 0);
1204 	root->log_transid = 0;
1205 	root->log_transid_committed = -1;
1206 	root->last_log_commit = 0;
1207 	root->anon_dev = 0;
1208 	if (!dummy) {
1209 		extent_io_tree_init(fs_info, &root->dirty_log_pages,
1210 				    IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1211 		extent_io_tree_init(fs_info, &root->log_csum_range,
1212 				    IO_TREE_LOG_CSUM_RANGE, NULL);
1213 	}
1214 
1215 	spin_lock_init(&root->root_item_lock);
1216 	btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1217 #ifdef CONFIG_BTRFS_DEBUG
1218 	INIT_LIST_HEAD(&root->leak_list);
1219 	spin_lock(&fs_info->fs_roots_radix_lock);
1220 	list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1221 	spin_unlock(&fs_info->fs_roots_radix_lock);
1222 #endif
1223 }
1224 
1225 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1226 					   u64 objectid, gfp_t flags)
1227 {
1228 	struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1229 	if (root)
1230 		__setup_root(root, fs_info, objectid);
1231 	return root;
1232 }
1233 
1234 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1235 /* Should only be used by the testing infrastructure */
1236 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1237 {
1238 	struct btrfs_root *root;
1239 
1240 	if (!fs_info)
1241 		return ERR_PTR(-EINVAL);
1242 
1243 	root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1244 	if (!root)
1245 		return ERR_PTR(-ENOMEM);
1246 
1247 	/* We don't use the stripesize in selftest, set it as sectorsize */
1248 	root->alloc_bytenr = 0;
1249 
1250 	return root;
1251 }
1252 #endif
1253 
1254 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
1255 {
1256 	const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
1257 	const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
1258 
1259 	return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
1260 }
1261 
1262 static int global_root_key_cmp(const void *k, const struct rb_node *node)
1263 {
1264 	const struct btrfs_key *key = k;
1265 	const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
1266 
1267 	return btrfs_comp_cpu_keys(key, &root->root_key);
1268 }
1269 
1270 int btrfs_global_root_insert(struct btrfs_root *root)
1271 {
1272 	struct btrfs_fs_info *fs_info = root->fs_info;
1273 	struct rb_node *tmp;
1274 
1275 	write_lock(&fs_info->global_root_lock);
1276 	tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
1277 	write_unlock(&fs_info->global_root_lock);
1278 	ASSERT(!tmp);
1279 
1280 	return tmp ? -EEXIST : 0;
1281 }
1282 
1283 void btrfs_global_root_delete(struct btrfs_root *root)
1284 {
1285 	struct btrfs_fs_info *fs_info = root->fs_info;
1286 
1287 	write_lock(&fs_info->global_root_lock);
1288 	rb_erase(&root->rb_node, &fs_info->global_root_tree);
1289 	write_unlock(&fs_info->global_root_lock);
1290 }
1291 
1292 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
1293 				     struct btrfs_key *key)
1294 {
1295 	struct rb_node *node;
1296 	struct btrfs_root *root = NULL;
1297 
1298 	read_lock(&fs_info->global_root_lock);
1299 	node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
1300 	if (node)
1301 		root = container_of(node, struct btrfs_root, rb_node);
1302 	read_unlock(&fs_info->global_root_lock);
1303 
1304 	return root;
1305 }
1306 
1307 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
1308 {
1309 	struct btrfs_block_group *block_group;
1310 	u64 ret;
1311 
1312 	if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
1313 		return 0;
1314 
1315 	if (bytenr)
1316 		block_group = btrfs_lookup_block_group(fs_info, bytenr);
1317 	else
1318 		block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
1319 	ASSERT(block_group);
1320 	if (!block_group)
1321 		return 0;
1322 	ret = block_group->global_root_id;
1323 	btrfs_put_block_group(block_group);
1324 
1325 	return ret;
1326 }
1327 
1328 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1329 {
1330 	struct btrfs_key key = {
1331 		.objectid = BTRFS_CSUM_TREE_OBJECTID,
1332 		.type = BTRFS_ROOT_ITEM_KEY,
1333 		.offset = btrfs_global_root_id(fs_info, bytenr),
1334 	};
1335 
1336 	return btrfs_global_root(fs_info, &key);
1337 }
1338 
1339 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1340 {
1341 	struct btrfs_key key = {
1342 		.objectid = BTRFS_EXTENT_TREE_OBJECTID,
1343 		.type = BTRFS_ROOT_ITEM_KEY,
1344 		.offset = btrfs_global_root_id(fs_info, bytenr),
1345 	};
1346 
1347 	return btrfs_global_root(fs_info, &key);
1348 }
1349 
1350 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1351 				     u64 objectid)
1352 {
1353 	struct btrfs_fs_info *fs_info = trans->fs_info;
1354 	struct extent_buffer *leaf;
1355 	struct btrfs_root *tree_root = fs_info->tree_root;
1356 	struct btrfs_root *root;
1357 	struct btrfs_key key;
1358 	unsigned int nofs_flag;
1359 	int ret = 0;
1360 
1361 	/*
1362 	 * We're holding a transaction handle, so use a NOFS memory allocation
1363 	 * context to avoid deadlock if reclaim happens.
1364 	 */
1365 	nofs_flag = memalloc_nofs_save();
1366 	root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1367 	memalloc_nofs_restore(nofs_flag);
1368 	if (!root)
1369 		return ERR_PTR(-ENOMEM);
1370 
1371 	root->root_key.objectid = objectid;
1372 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1373 	root->root_key.offset = 0;
1374 
1375 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1376 				      BTRFS_NESTING_NORMAL);
1377 	if (IS_ERR(leaf)) {
1378 		ret = PTR_ERR(leaf);
1379 		leaf = NULL;
1380 		goto fail_unlock;
1381 	}
1382 
1383 	root->node = leaf;
1384 	btrfs_mark_buffer_dirty(leaf);
1385 
1386 	root->commit_root = btrfs_root_node(root);
1387 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1388 
1389 	btrfs_set_root_flags(&root->root_item, 0);
1390 	btrfs_set_root_limit(&root->root_item, 0);
1391 	btrfs_set_root_bytenr(&root->root_item, leaf->start);
1392 	btrfs_set_root_generation(&root->root_item, trans->transid);
1393 	btrfs_set_root_level(&root->root_item, 0);
1394 	btrfs_set_root_refs(&root->root_item, 1);
1395 	btrfs_set_root_used(&root->root_item, leaf->len);
1396 	btrfs_set_root_last_snapshot(&root->root_item, 0);
1397 	btrfs_set_root_dirid(&root->root_item, 0);
1398 	if (is_fstree(objectid))
1399 		generate_random_guid(root->root_item.uuid);
1400 	else
1401 		export_guid(root->root_item.uuid, &guid_null);
1402 	btrfs_set_root_drop_level(&root->root_item, 0);
1403 
1404 	btrfs_tree_unlock(leaf);
1405 
1406 	key.objectid = objectid;
1407 	key.type = BTRFS_ROOT_ITEM_KEY;
1408 	key.offset = 0;
1409 	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1410 	if (ret)
1411 		goto fail;
1412 
1413 	return root;
1414 
1415 fail_unlock:
1416 	if (leaf)
1417 		btrfs_tree_unlock(leaf);
1418 fail:
1419 	btrfs_put_root(root);
1420 
1421 	return ERR_PTR(ret);
1422 }
1423 
1424 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1425 					 struct btrfs_fs_info *fs_info)
1426 {
1427 	struct btrfs_root *root;
1428 
1429 	root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1430 	if (!root)
1431 		return ERR_PTR(-ENOMEM);
1432 
1433 	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1434 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1435 	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1436 
1437 	return root;
1438 }
1439 
1440 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1441 			      struct btrfs_root *root)
1442 {
1443 	struct extent_buffer *leaf;
1444 
1445 	/*
1446 	 * DON'T set SHAREABLE bit for log trees.
1447 	 *
1448 	 * Log trees are not exposed to user space thus can't be snapshotted,
1449 	 * and they go away before a real commit is actually done.
1450 	 *
1451 	 * They do store pointers to file data extents, and those reference
1452 	 * counts still get updated (along with back refs to the log tree).
1453 	 */
1454 
1455 	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1456 			NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1457 	if (IS_ERR(leaf))
1458 		return PTR_ERR(leaf);
1459 
1460 	root->node = leaf;
1461 
1462 	btrfs_mark_buffer_dirty(root->node);
1463 	btrfs_tree_unlock(root->node);
1464 
1465 	return 0;
1466 }
1467 
1468 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1469 			     struct btrfs_fs_info *fs_info)
1470 {
1471 	struct btrfs_root *log_root;
1472 
1473 	log_root = alloc_log_tree(trans, fs_info);
1474 	if (IS_ERR(log_root))
1475 		return PTR_ERR(log_root);
1476 
1477 	if (!btrfs_is_zoned(fs_info)) {
1478 		int ret = btrfs_alloc_log_tree_node(trans, log_root);
1479 
1480 		if (ret) {
1481 			btrfs_put_root(log_root);
1482 			return ret;
1483 		}
1484 	}
1485 
1486 	WARN_ON(fs_info->log_root_tree);
1487 	fs_info->log_root_tree = log_root;
1488 	return 0;
1489 }
1490 
1491 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1492 		       struct btrfs_root *root)
1493 {
1494 	struct btrfs_fs_info *fs_info = root->fs_info;
1495 	struct btrfs_root *log_root;
1496 	struct btrfs_inode_item *inode_item;
1497 	int ret;
1498 
1499 	log_root = alloc_log_tree(trans, fs_info);
1500 	if (IS_ERR(log_root))
1501 		return PTR_ERR(log_root);
1502 
1503 	ret = btrfs_alloc_log_tree_node(trans, log_root);
1504 	if (ret) {
1505 		btrfs_put_root(log_root);
1506 		return ret;
1507 	}
1508 
1509 	log_root->last_trans = trans->transid;
1510 	log_root->root_key.offset = root->root_key.objectid;
1511 
1512 	inode_item = &log_root->root_item.inode;
1513 	btrfs_set_stack_inode_generation(inode_item, 1);
1514 	btrfs_set_stack_inode_size(inode_item, 3);
1515 	btrfs_set_stack_inode_nlink(inode_item, 1);
1516 	btrfs_set_stack_inode_nbytes(inode_item,
1517 				     fs_info->nodesize);
1518 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1519 
1520 	btrfs_set_root_node(&log_root->root_item, log_root->node);
1521 
1522 	WARN_ON(root->log_root);
1523 	root->log_root = log_root;
1524 	root->log_transid = 0;
1525 	root->log_transid_committed = -1;
1526 	root->last_log_commit = 0;
1527 	return 0;
1528 }
1529 
1530 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1531 					      struct btrfs_path *path,
1532 					      struct btrfs_key *key)
1533 {
1534 	struct btrfs_root *root;
1535 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1536 	u64 generation;
1537 	int ret;
1538 	int level;
1539 
1540 	root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1541 	if (!root)
1542 		return ERR_PTR(-ENOMEM);
1543 
1544 	ret = btrfs_find_root(tree_root, key, path,
1545 			      &root->root_item, &root->root_key);
1546 	if (ret) {
1547 		if (ret > 0)
1548 			ret = -ENOENT;
1549 		goto fail;
1550 	}
1551 
1552 	generation = btrfs_root_generation(&root->root_item);
1553 	level = btrfs_root_level(&root->root_item);
1554 	root->node = read_tree_block(fs_info,
1555 				     btrfs_root_bytenr(&root->root_item),
1556 				     key->objectid, generation, level, NULL);
1557 	if (IS_ERR(root->node)) {
1558 		ret = PTR_ERR(root->node);
1559 		root->node = NULL;
1560 		goto fail;
1561 	}
1562 	if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1563 		ret = -EIO;
1564 		goto fail;
1565 	}
1566 	root->commit_root = btrfs_root_node(root);
1567 	return root;
1568 fail:
1569 	btrfs_put_root(root);
1570 	return ERR_PTR(ret);
1571 }
1572 
1573 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1574 					struct btrfs_key *key)
1575 {
1576 	struct btrfs_root *root;
1577 	struct btrfs_path *path;
1578 
1579 	path = btrfs_alloc_path();
1580 	if (!path)
1581 		return ERR_PTR(-ENOMEM);
1582 	root = read_tree_root_path(tree_root, path, key);
1583 	btrfs_free_path(path);
1584 
1585 	return root;
1586 }
1587 
1588 /*
1589  * Initialize subvolume root in-memory structure
1590  *
1591  * @anon_dev:	anonymous device to attach to the root, if zero, allocate new
1592  */
1593 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1594 {
1595 	int ret;
1596 	unsigned int nofs_flag;
1597 
1598 	/*
1599 	 * We might be called under a transaction (e.g. indirect backref
1600 	 * resolution) which could deadlock if it triggers memory reclaim
1601 	 */
1602 	nofs_flag = memalloc_nofs_save();
1603 	ret = btrfs_drew_lock_init(&root->snapshot_lock);
1604 	memalloc_nofs_restore(nofs_flag);
1605 	if (ret)
1606 		goto fail;
1607 
1608 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1609 	    !btrfs_is_data_reloc_root(root)) {
1610 		set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1611 		btrfs_check_and_init_root_item(&root->root_item);
1612 	}
1613 
1614 	/*
1615 	 * Don't assign anonymous block device to roots that are not exposed to
1616 	 * userspace, the id pool is limited to 1M
1617 	 */
1618 	if (is_fstree(root->root_key.objectid) &&
1619 	    btrfs_root_refs(&root->root_item) > 0) {
1620 		if (!anon_dev) {
1621 			ret = get_anon_bdev(&root->anon_dev);
1622 			if (ret)
1623 				goto fail;
1624 		} else {
1625 			root->anon_dev = anon_dev;
1626 		}
1627 	}
1628 
1629 	mutex_lock(&root->objectid_mutex);
1630 	ret = btrfs_init_root_free_objectid(root);
1631 	if (ret) {
1632 		mutex_unlock(&root->objectid_mutex);
1633 		goto fail;
1634 	}
1635 
1636 	ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1637 
1638 	mutex_unlock(&root->objectid_mutex);
1639 
1640 	return 0;
1641 fail:
1642 	/* The caller is responsible to call btrfs_free_fs_root */
1643 	return ret;
1644 }
1645 
1646 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1647 					       u64 root_id)
1648 {
1649 	struct btrfs_root *root;
1650 
1651 	spin_lock(&fs_info->fs_roots_radix_lock);
1652 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1653 				 (unsigned long)root_id);
1654 	if (root)
1655 		root = btrfs_grab_root(root);
1656 	spin_unlock(&fs_info->fs_roots_radix_lock);
1657 	return root;
1658 }
1659 
1660 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1661 						u64 objectid)
1662 {
1663 	struct btrfs_key key = {
1664 		.objectid = objectid,
1665 		.type = BTRFS_ROOT_ITEM_KEY,
1666 		.offset = 0,
1667 	};
1668 
1669 	if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1670 		return btrfs_grab_root(fs_info->tree_root);
1671 	if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1672 		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1673 	if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1674 		return btrfs_grab_root(fs_info->chunk_root);
1675 	if (objectid == BTRFS_DEV_TREE_OBJECTID)
1676 		return btrfs_grab_root(fs_info->dev_root);
1677 	if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1678 		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1679 	if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1680 		return btrfs_grab_root(fs_info->quota_root) ?
1681 			fs_info->quota_root : ERR_PTR(-ENOENT);
1682 	if (objectid == BTRFS_UUID_TREE_OBJECTID)
1683 		return btrfs_grab_root(fs_info->uuid_root) ?
1684 			fs_info->uuid_root : ERR_PTR(-ENOENT);
1685 	if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) {
1686 		struct btrfs_root *root = btrfs_global_root(fs_info, &key);
1687 
1688 		return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT);
1689 	}
1690 	return NULL;
1691 }
1692 
1693 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1694 			 struct btrfs_root *root)
1695 {
1696 	int ret;
1697 
1698 	ret = radix_tree_preload(GFP_NOFS);
1699 	if (ret)
1700 		return ret;
1701 
1702 	spin_lock(&fs_info->fs_roots_radix_lock);
1703 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1704 				(unsigned long)root->root_key.objectid,
1705 				root);
1706 	if (ret == 0) {
1707 		btrfs_grab_root(root);
1708 		set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1709 	}
1710 	spin_unlock(&fs_info->fs_roots_radix_lock);
1711 	radix_tree_preload_end();
1712 
1713 	return ret;
1714 }
1715 
1716 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1717 {
1718 #ifdef CONFIG_BTRFS_DEBUG
1719 	struct btrfs_root *root;
1720 
1721 	while (!list_empty(&fs_info->allocated_roots)) {
1722 		char buf[BTRFS_ROOT_NAME_BUF_LEN];
1723 
1724 		root = list_first_entry(&fs_info->allocated_roots,
1725 					struct btrfs_root, leak_list);
1726 		btrfs_err(fs_info, "leaked root %s refcount %d",
1727 			  btrfs_root_name(&root->root_key, buf),
1728 			  refcount_read(&root->refs));
1729 		while (refcount_read(&root->refs) > 1)
1730 			btrfs_put_root(root);
1731 		btrfs_put_root(root);
1732 	}
1733 #endif
1734 }
1735 
1736 static void free_global_roots(struct btrfs_fs_info *fs_info)
1737 {
1738 	struct btrfs_root *root;
1739 	struct rb_node *node;
1740 
1741 	while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1742 		root = rb_entry(node, struct btrfs_root, rb_node);
1743 		rb_erase(&root->rb_node, &fs_info->global_root_tree);
1744 		btrfs_put_root(root);
1745 	}
1746 }
1747 
1748 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1749 {
1750 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1751 	percpu_counter_destroy(&fs_info->delalloc_bytes);
1752 	percpu_counter_destroy(&fs_info->ordered_bytes);
1753 	percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1754 	btrfs_free_csum_hash(fs_info);
1755 	btrfs_free_stripe_hash_table(fs_info);
1756 	btrfs_free_ref_cache(fs_info);
1757 	kfree(fs_info->balance_ctl);
1758 	kfree(fs_info->delayed_root);
1759 	free_global_roots(fs_info);
1760 	btrfs_put_root(fs_info->tree_root);
1761 	btrfs_put_root(fs_info->chunk_root);
1762 	btrfs_put_root(fs_info->dev_root);
1763 	btrfs_put_root(fs_info->quota_root);
1764 	btrfs_put_root(fs_info->uuid_root);
1765 	btrfs_put_root(fs_info->fs_root);
1766 	btrfs_put_root(fs_info->data_reloc_root);
1767 	btrfs_put_root(fs_info->block_group_root);
1768 	btrfs_check_leaked_roots(fs_info);
1769 	btrfs_extent_buffer_leak_debug_check(fs_info);
1770 	kfree(fs_info->super_copy);
1771 	kfree(fs_info->super_for_commit);
1772 	kfree(fs_info->subpage_info);
1773 	kvfree(fs_info);
1774 }
1775 
1776 
1777 /*
1778  * Get an in-memory reference of a root structure.
1779  *
1780  * For essential trees like root/extent tree, we grab it from fs_info directly.
1781  * For subvolume trees, we check the cached filesystem roots first. If not
1782  * found, then read it from disk and add it to cached fs roots.
1783  *
1784  * Caller should release the root by calling btrfs_put_root() after the usage.
1785  *
1786  * NOTE: Reloc and log trees can't be read by this function as they share the
1787  *	 same root objectid.
1788  *
1789  * @objectid:	root id
1790  * @anon_dev:	preallocated anonymous block device number for new roots,
1791  * 		pass 0 for new allocation.
1792  * @check_ref:	whether to check root item references, If true, return -ENOENT
1793  *		for orphan roots
1794  */
1795 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1796 					     u64 objectid, dev_t anon_dev,
1797 					     bool check_ref)
1798 {
1799 	struct btrfs_root *root;
1800 	struct btrfs_path *path;
1801 	struct btrfs_key key;
1802 	int ret;
1803 
1804 	root = btrfs_get_global_root(fs_info, objectid);
1805 	if (root)
1806 		return root;
1807 again:
1808 	root = btrfs_lookup_fs_root(fs_info, objectid);
1809 	if (root) {
1810 		/* Shouldn't get preallocated anon_dev for cached roots */
1811 		ASSERT(!anon_dev);
1812 		if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1813 			btrfs_put_root(root);
1814 			return ERR_PTR(-ENOENT);
1815 		}
1816 		return root;
1817 	}
1818 
1819 	key.objectid = objectid;
1820 	key.type = BTRFS_ROOT_ITEM_KEY;
1821 	key.offset = (u64)-1;
1822 	root = btrfs_read_tree_root(fs_info->tree_root, &key);
1823 	if (IS_ERR(root))
1824 		return root;
1825 
1826 	if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1827 		ret = -ENOENT;
1828 		goto fail;
1829 	}
1830 
1831 	ret = btrfs_init_fs_root(root, anon_dev);
1832 	if (ret)
1833 		goto fail;
1834 
1835 	path = btrfs_alloc_path();
1836 	if (!path) {
1837 		ret = -ENOMEM;
1838 		goto fail;
1839 	}
1840 	key.objectid = BTRFS_ORPHAN_OBJECTID;
1841 	key.type = BTRFS_ORPHAN_ITEM_KEY;
1842 	key.offset = objectid;
1843 
1844 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1845 	btrfs_free_path(path);
1846 	if (ret < 0)
1847 		goto fail;
1848 	if (ret == 0)
1849 		set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1850 
1851 	ret = btrfs_insert_fs_root(fs_info, root);
1852 	if (ret) {
1853 		if (ret == -EEXIST) {
1854 			btrfs_put_root(root);
1855 			goto again;
1856 		}
1857 		goto fail;
1858 	}
1859 	return root;
1860 fail:
1861 	/*
1862 	 * If our caller provided us an anonymous device, then it's his
1863 	 * responsability to free it in case we fail. So we have to set our
1864 	 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1865 	 * and once again by our caller.
1866 	 */
1867 	if (anon_dev)
1868 		root->anon_dev = 0;
1869 	btrfs_put_root(root);
1870 	return ERR_PTR(ret);
1871 }
1872 
1873 /*
1874  * Get in-memory reference of a root structure
1875  *
1876  * @objectid:	tree objectid
1877  * @check_ref:	if set, verify that the tree exists and the item has at least
1878  *		one reference
1879  */
1880 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1881 				     u64 objectid, bool check_ref)
1882 {
1883 	return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1884 }
1885 
1886 /*
1887  * Get in-memory reference of a root structure, created as new, optionally pass
1888  * the anonymous block device id
1889  *
1890  * @objectid:	tree objectid
1891  * @anon_dev:	if zero, allocate a new anonymous block device or use the
1892  *		parameter value
1893  */
1894 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1895 					 u64 objectid, dev_t anon_dev)
1896 {
1897 	return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1898 }
1899 
1900 /*
1901  * btrfs_get_fs_root_commit_root - return a root for the given objectid
1902  * @fs_info:	the fs_info
1903  * @objectid:	the objectid we need to lookup
1904  *
1905  * This is exclusively used for backref walking, and exists specifically because
1906  * of how qgroups does lookups.  Qgroups will do a backref lookup at delayed ref
1907  * creation time, which means we may have to read the tree_root in order to look
1908  * up a fs root that is not in memory.  If the root is not in memory we will
1909  * read the tree root commit root and look up the fs root from there.  This is a
1910  * temporary root, it will not be inserted into the radix tree as it doesn't
1911  * have the most uptodate information, it'll simply be discarded once the
1912  * backref code is finished using the root.
1913  */
1914 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1915 						 struct btrfs_path *path,
1916 						 u64 objectid)
1917 {
1918 	struct btrfs_root *root;
1919 	struct btrfs_key key;
1920 
1921 	ASSERT(path->search_commit_root && path->skip_locking);
1922 
1923 	/*
1924 	 * This can return -ENOENT if we ask for a root that doesn't exist, but
1925 	 * since this is called via the backref walking code we won't be looking
1926 	 * up a root that doesn't exist, unless there's corruption.  So if root
1927 	 * != NULL just return it.
1928 	 */
1929 	root = btrfs_get_global_root(fs_info, objectid);
1930 	if (root)
1931 		return root;
1932 
1933 	root = btrfs_lookup_fs_root(fs_info, objectid);
1934 	if (root)
1935 		return root;
1936 
1937 	key.objectid = objectid;
1938 	key.type = BTRFS_ROOT_ITEM_KEY;
1939 	key.offset = (u64)-1;
1940 	root = read_tree_root_path(fs_info->tree_root, path, &key);
1941 	btrfs_release_path(path);
1942 
1943 	return root;
1944 }
1945 
1946 /*
1947  * called by the kthread helper functions to finally call the bio end_io
1948  * functions.  This is where read checksum verification actually happens
1949  */
1950 static void end_workqueue_fn(struct btrfs_work *work)
1951 {
1952 	struct bio *bio;
1953 	struct btrfs_end_io_wq *end_io_wq;
1954 
1955 	end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1956 	bio = end_io_wq->bio;
1957 
1958 	bio->bi_status = end_io_wq->status;
1959 	bio->bi_private = end_io_wq->private;
1960 	bio->bi_end_io = end_io_wq->end_io;
1961 	bio_endio(bio);
1962 	kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1963 }
1964 
1965 static int cleaner_kthread(void *arg)
1966 {
1967 	struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)arg;
1968 	int again;
1969 
1970 	while (1) {
1971 		again = 0;
1972 
1973 		set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1974 
1975 		/* Make the cleaner go to sleep early. */
1976 		if (btrfs_need_cleaner_sleep(fs_info))
1977 			goto sleep;
1978 
1979 		/*
1980 		 * Do not do anything if we might cause open_ctree() to block
1981 		 * before we have finished mounting the filesystem.
1982 		 */
1983 		if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1984 			goto sleep;
1985 
1986 		if (!mutex_trylock(&fs_info->cleaner_mutex))
1987 			goto sleep;
1988 
1989 		/*
1990 		 * Avoid the problem that we change the status of the fs
1991 		 * during the above check and trylock.
1992 		 */
1993 		if (btrfs_need_cleaner_sleep(fs_info)) {
1994 			mutex_unlock(&fs_info->cleaner_mutex);
1995 			goto sleep;
1996 		}
1997 
1998 		btrfs_run_delayed_iputs(fs_info);
1999 
2000 		again = btrfs_clean_one_deleted_snapshot(fs_info);
2001 		mutex_unlock(&fs_info->cleaner_mutex);
2002 
2003 		/*
2004 		 * The defragger has dealt with the R/O remount and umount,
2005 		 * needn't do anything special here.
2006 		 */
2007 		btrfs_run_defrag_inodes(fs_info);
2008 
2009 		/*
2010 		 * Acquires fs_info->reclaim_bgs_lock to avoid racing
2011 		 * with relocation (btrfs_relocate_chunk) and relocation
2012 		 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
2013 		 * after acquiring fs_info->reclaim_bgs_lock. So we
2014 		 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
2015 		 * unused block groups.
2016 		 */
2017 		btrfs_delete_unused_bgs(fs_info);
2018 
2019 		/*
2020 		 * Reclaim block groups in the reclaim_bgs list after we deleted
2021 		 * all unused block_groups. This possibly gives us some more free
2022 		 * space.
2023 		 */
2024 		btrfs_reclaim_bgs(fs_info);
2025 sleep:
2026 		clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
2027 		if (kthread_should_park())
2028 			kthread_parkme();
2029 		if (kthread_should_stop())
2030 			return 0;
2031 		if (!again) {
2032 			set_current_state(TASK_INTERRUPTIBLE);
2033 			schedule();
2034 			__set_current_state(TASK_RUNNING);
2035 		}
2036 	}
2037 }
2038 
2039 static int transaction_kthread(void *arg)
2040 {
2041 	struct btrfs_root *root = arg;
2042 	struct btrfs_fs_info *fs_info = root->fs_info;
2043 	struct btrfs_trans_handle *trans;
2044 	struct btrfs_transaction *cur;
2045 	u64 transid;
2046 	time64_t delta;
2047 	unsigned long delay;
2048 	bool cannot_commit;
2049 
2050 	do {
2051 		cannot_commit = false;
2052 		delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
2053 		mutex_lock(&fs_info->transaction_kthread_mutex);
2054 
2055 		spin_lock(&fs_info->trans_lock);
2056 		cur = fs_info->running_transaction;
2057 		if (!cur) {
2058 			spin_unlock(&fs_info->trans_lock);
2059 			goto sleep;
2060 		}
2061 
2062 		delta = ktime_get_seconds() - cur->start_time;
2063 		if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
2064 		    cur->state < TRANS_STATE_COMMIT_START &&
2065 		    delta < fs_info->commit_interval) {
2066 			spin_unlock(&fs_info->trans_lock);
2067 			delay -= msecs_to_jiffies((delta - 1) * 1000);
2068 			delay = min(delay,
2069 				    msecs_to_jiffies(fs_info->commit_interval * 1000));
2070 			goto sleep;
2071 		}
2072 		transid = cur->transid;
2073 		spin_unlock(&fs_info->trans_lock);
2074 
2075 		/* If the file system is aborted, this will always fail. */
2076 		trans = btrfs_attach_transaction(root);
2077 		if (IS_ERR(trans)) {
2078 			if (PTR_ERR(trans) != -ENOENT)
2079 				cannot_commit = true;
2080 			goto sleep;
2081 		}
2082 		if (transid == trans->transid) {
2083 			btrfs_commit_transaction(trans);
2084 		} else {
2085 			btrfs_end_transaction(trans);
2086 		}
2087 sleep:
2088 		wake_up_process(fs_info->cleaner_kthread);
2089 		mutex_unlock(&fs_info->transaction_kthread_mutex);
2090 
2091 		if (BTRFS_FS_ERROR(fs_info))
2092 			btrfs_cleanup_transaction(fs_info);
2093 		if (!kthread_should_stop() &&
2094 				(!btrfs_transaction_blocked(fs_info) ||
2095 				 cannot_commit))
2096 			schedule_timeout_interruptible(delay);
2097 	} while (!kthread_should_stop());
2098 	return 0;
2099 }
2100 
2101 /*
2102  * This will find the highest generation in the array of root backups.  The
2103  * index of the highest array is returned, or -EINVAL if we can't find
2104  * anything.
2105  *
2106  * We check to make sure the array is valid by comparing the
2107  * generation of the latest  root in the array with the generation
2108  * in the super block.  If they don't match we pitch it.
2109  */
2110 static int find_newest_super_backup(struct btrfs_fs_info *info)
2111 {
2112 	const u64 newest_gen = btrfs_super_generation(info->super_copy);
2113 	u64 cur;
2114 	struct btrfs_root_backup *root_backup;
2115 	int i;
2116 
2117 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2118 		root_backup = info->super_copy->super_roots + i;
2119 		cur = btrfs_backup_tree_root_gen(root_backup);
2120 		if (cur == newest_gen)
2121 			return i;
2122 	}
2123 
2124 	return -EINVAL;
2125 }
2126 
2127 /*
2128  * copy all the root pointers into the super backup array.
2129  * this will bump the backup pointer by one when it is
2130  * done
2131  */
2132 static void backup_super_roots(struct btrfs_fs_info *info)
2133 {
2134 	const int next_backup = info->backup_root_index;
2135 	struct btrfs_root_backup *root_backup;
2136 
2137 	root_backup = info->super_for_commit->super_roots + next_backup;
2138 
2139 	/*
2140 	 * make sure all of our padding and empty slots get zero filled
2141 	 * regardless of which ones we use today
2142 	 */
2143 	memset(root_backup, 0, sizeof(*root_backup));
2144 
2145 	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2146 
2147 	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2148 	btrfs_set_backup_tree_root_gen(root_backup,
2149 			       btrfs_header_generation(info->tree_root->node));
2150 
2151 	btrfs_set_backup_tree_root_level(root_backup,
2152 			       btrfs_header_level(info->tree_root->node));
2153 
2154 	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2155 	btrfs_set_backup_chunk_root_gen(root_backup,
2156 			       btrfs_header_generation(info->chunk_root->node));
2157 	btrfs_set_backup_chunk_root_level(root_backup,
2158 			       btrfs_header_level(info->chunk_root->node));
2159 
2160 	if (btrfs_fs_incompat(info, EXTENT_TREE_V2)) {
2161 		btrfs_set_backup_block_group_root(root_backup,
2162 					info->block_group_root->node->start);
2163 		btrfs_set_backup_block_group_root_gen(root_backup,
2164 			btrfs_header_generation(info->block_group_root->node));
2165 		btrfs_set_backup_block_group_root_level(root_backup,
2166 			btrfs_header_level(info->block_group_root->node));
2167 	} else {
2168 		struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
2169 		struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
2170 
2171 		btrfs_set_backup_extent_root(root_backup,
2172 					     extent_root->node->start);
2173 		btrfs_set_backup_extent_root_gen(root_backup,
2174 				btrfs_header_generation(extent_root->node));
2175 		btrfs_set_backup_extent_root_level(root_backup,
2176 					btrfs_header_level(extent_root->node));
2177 
2178 		btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
2179 		btrfs_set_backup_csum_root_gen(root_backup,
2180 					       btrfs_header_generation(csum_root->node));
2181 		btrfs_set_backup_csum_root_level(root_backup,
2182 						 btrfs_header_level(csum_root->node));
2183 	}
2184 
2185 	/*
2186 	 * we might commit during log recovery, which happens before we set
2187 	 * the fs_root.  Make sure it is valid before we fill it in.
2188 	 */
2189 	if (info->fs_root && info->fs_root->node) {
2190 		btrfs_set_backup_fs_root(root_backup,
2191 					 info->fs_root->node->start);
2192 		btrfs_set_backup_fs_root_gen(root_backup,
2193 			       btrfs_header_generation(info->fs_root->node));
2194 		btrfs_set_backup_fs_root_level(root_backup,
2195 			       btrfs_header_level(info->fs_root->node));
2196 	}
2197 
2198 	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2199 	btrfs_set_backup_dev_root_gen(root_backup,
2200 			       btrfs_header_generation(info->dev_root->node));
2201 	btrfs_set_backup_dev_root_level(root_backup,
2202 				       btrfs_header_level(info->dev_root->node));
2203 
2204 	btrfs_set_backup_total_bytes(root_backup,
2205 			     btrfs_super_total_bytes(info->super_copy));
2206 	btrfs_set_backup_bytes_used(root_backup,
2207 			     btrfs_super_bytes_used(info->super_copy));
2208 	btrfs_set_backup_num_devices(root_backup,
2209 			     btrfs_super_num_devices(info->super_copy));
2210 
2211 	/*
2212 	 * if we don't copy this out to the super_copy, it won't get remembered
2213 	 * for the next commit
2214 	 */
2215 	memcpy(&info->super_copy->super_roots,
2216 	       &info->super_for_commit->super_roots,
2217 	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2218 }
2219 
2220 /*
2221  * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2222  * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2223  *
2224  * fs_info - filesystem whose backup roots need to be read
2225  * priority - priority of backup root required
2226  *
2227  * Returns backup root index on success and -EINVAL otherwise.
2228  */
2229 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2230 {
2231 	int backup_index = find_newest_super_backup(fs_info);
2232 	struct btrfs_super_block *super = fs_info->super_copy;
2233 	struct btrfs_root_backup *root_backup;
2234 
2235 	if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2236 		if (priority == 0)
2237 			return backup_index;
2238 
2239 		backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2240 		backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2241 	} else {
2242 		return -EINVAL;
2243 	}
2244 
2245 	root_backup = super->super_roots + backup_index;
2246 
2247 	btrfs_set_super_generation(super,
2248 				   btrfs_backup_tree_root_gen(root_backup));
2249 	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2250 	btrfs_set_super_root_level(super,
2251 				   btrfs_backup_tree_root_level(root_backup));
2252 	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2253 
2254 	/*
2255 	 * Fixme: the total bytes and num_devices need to match or we should
2256 	 * need a fsck
2257 	 */
2258 	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2259 	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2260 
2261 	return backup_index;
2262 }
2263 
2264 /* helper to cleanup workers */
2265 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2266 {
2267 	btrfs_destroy_workqueue(fs_info->fixup_workers);
2268 	btrfs_destroy_workqueue(fs_info->delalloc_workers);
2269 	btrfs_destroy_workqueue(fs_info->workers);
2270 	btrfs_destroy_workqueue(fs_info->endio_workers);
2271 	btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2272 	btrfs_destroy_workqueue(fs_info->rmw_workers);
2273 	btrfs_destroy_workqueue(fs_info->endio_write_workers);
2274 	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2275 	btrfs_destroy_workqueue(fs_info->delayed_workers);
2276 	btrfs_destroy_workqueue(fs_info->caching_workers);
2277 	btrfs_destroy_workqueue(fs_info->flush_workers);
2278 	btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2279 	if (fs_info->discard_ctl.discard_workers)
2280 		destroy_workqueue(fs_info->discard_ctl.discard_workers);
2281 	/*
2282 	 * Now that all other work queues are destroyed, we can safely destroy
2283 	 * the queues used for metadata I/O, since tasks from those other work
2284 	 * queues can do metadata I/O operations.
2285 	 */
2286 	btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2287 	btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2288 }
2289 
2290 static void free_root_extent_buffers(struct btrfs_root *root)
2291 {
2292 	if (root) {
2293 		free_extent_buffer(root->node);
2294 		free_extent_buffer(root->commit_root);
2295 		root->node = NULL;
2296 		root->commit_root = NULL;
2297 	}
2298 }
2299 
2300 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
2301 {
2302 	struct btrfs_root *root, *tmp;
2303 
2304 	rbtree_postorder_for_each_entry_safe(root, tmp,
2305 					     &fs_info->global_root_tree,
2306 					     rb_node)
2307 		free_root_extent_buffers(root);
2308 }
2309 
2310 /* helper to cleanup tree roots */
2311 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2312 {
2313 	free_root_extent_buffers(info->tree_root);
2314 
2315 	free_global_root_pointers(info);
2316 	free_root_extent_buffers(info->dev_root);
2317 	free_root_extent_buffers(info->quota_root);
2318 	free_root_extent_buffers(info->uuid_root);
2319 	free_root_extent_buffers(info->fs_root);
2320 	free_root_extent_buffers(info->data_reloc_root);
2321 	free_root_extent_buffers(info->block_group_root);
2322 	if (free_chunk_root)
2323 		free_root_extent_buffers(info->chunk_root);
2324 }
2325 
2326 void btrfs_put_root(struct btrfs_root *root)
2327 {
2328 	if (!root)
2329 		return;
2330 
2331 	if (refcount_dec_and_test(&root->refs)) {
2332 		WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2333 		WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2334 		if (root->anon_dev)
2335 			free_anon_bdev(root->anon_dev);
2336 		btrfs_drew_lock_destroy(&root->snapshot_lock);
2337 		free_root_extent_buffers(root);
2338 #ifdef CONFIG_BTRFS_DEBUG
2339 		spin_lock(&root->fs_info->fs_roots_radix_lock);
2340 		list_del_init(&root->leak_list);
2341 		spin_unlock(&root->fs_info->fs_roots_radix_lock);
2342 #endif
2343 		kfree(root);
2344 	}
2345 }
2346 
2347 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2348 {
2349 	int ret;
2350 	struct btrfs_root *gang[8];
2351 	int i;
2352 
2353 	while (!list_empty(&fs_info->dead_roots)) {
2354 		gang[0] = list_entry(fs_info->dead_roots.next,
2355 				     struct btrfs_root, root_list);
2356 		list_del(&gang[0]->root_list);
2357 
2358 		if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2359 			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2360 		btrfs_put_root(gang[0]);
2361 	}
2362 
2363 	while (1) {
2364 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2365 					     (void **)gang, 0,
2366 					     ARRAY_SIZE(gang));
2367 		if (!ret)
2368 			break;
2369 		for (i = 0; i < ret; i++)
2370 			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2371 	}
2372 }
2373 
2374 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2375 {
2376 	mutex_init(&fs_info->scrub_lock);
2377 	atomic_set(&fs_info->scrubs_running, 0);
2378 	atomic_set(&fs_info->scrub_pause_req, 0);
2379 	atomic_set(&fs_info->scrubs_paused, 0);
2380 	atomic_set(&fs_info->scrub_cancel_req, 0);
2381 	init_waitqueue_head(&fs_info->scrub_pause_wait);
2382 	refcount_set(&fs_info->scrub_workers_refcnt, 0);
2383 }
2384 
2385 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2386 {
2387 	spin_lock_init(&fs_info->balance_lock);
2388 	mutex_init(&fs_info->balance_mutex);
2389 	atomic_set(&fs_info->balance_pause_req, 0);
2390 	atomic_set(&fs_info->balance_cancel_req, 0);
2391 	fs_info->balance_ctl = NULL;
2392 	init_waitqueue_head(&fs_info->balance_wait_q);
2393 	atomic_set(&fs_info->reloc_cancel_req, 0);
2394 }
2395 
2396 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2397 {
2398 	struct inode *inode = fs_info->btree_inode;
2399 
2400 	inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2401 	set_nlink(inode, 1);
2402 	/*
2403 	 * we set the i_size on the btree inode to the max possible int.
2404 	 * the real end of the address space is determined by all of
2405 	 * the devices in the system
2406 	 */
2407 	inode->i_size = OFFSET_MAX;
2408 	inode->i_mapping->a_ops = &btree_aops;
2409 
2410 	RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2411 	extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2412 			    IO_TREE_BTREE_INODE_IO, inode);
2413 	BTRFS_I(inode)->io_tree.track_uptodate = false;
2414 	extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2415 
2416 	BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2417 	memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2418 	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2419 	btrfs_insert_inode_hash(inode);
2420 }
2421 
2422 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2423 {
2424 	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2425 	init_rwsem(&fs_info->dev_replace.rwsem);
2426 	init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2427 }
2428 
2429 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2430 {
2431 	spin_lock_init(&fs_info->qgroup_lock);
2432 	mutex_init(&fs_info->qgroup_ioctl_lock);
2433 	fs_info->qgroup_tree = RB_ROOT;
2434 	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2435 	fs_info->qgroup_seq = 1;
2436 	fs_info->qgroup_ulist = NULL;
2437 	fs_info->qgroup_rescan_running = false;
2438 	mutex_init(&fs_info->qgroup_rescan_lock);
2439 }
2440 
2441 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
2442 {
2443 	u32 max_active = fs_info->thread_pool_size;
2444 	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2445 
2446 	fs_info->workers =
2447 		btrfs_alloc_workqueue(fs_info, "worker",
2448 				      flags | WQ_HIGHPRI, max_active, 16);
2449 
2450 	fs_info->delalloc_workers =
2451 		btrfs_alloc_workqueue(fs_info, "delalloc",
2452 				      flags, max_active, 2);
2453 
2454 	fs_info->flush_workers =
2455 		btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2456 				      flags, max_active, 0);
2457 
2458 	fs_info->caching_workers =
2459 		btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2460 
2461 	fs_info->fixup_workers =
2462 		btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2463 
2464 	/*
2465 	 * endios are largely parallel and should have a very
2466 	 * low idle thresh
2467 	 */
2468 	fs_info->endio_workers =
2469 		btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2470 	fs_info->endio_meta_workers =
2471 		btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2472 				      max_active, 4);
2473 	fs_info->endio_meta_write_workers =
2474 		btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2475 				      max_active, 2);
2476 	fs_info->endio_raid56_workers =
2477 		btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2478 				      max_active, 4);
2479 	fs_info->rmw_workers =
2480 		btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
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->delalloc_workers &&
2496 	      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 	 * For 4K page size, we only support 4K sector size.
2819 	 * For 64K page size, we support 64K and 4K sector sizes.
2820 	 */
2821 	if ((PAGE_SIZE == SZ_4K && sectorsize != PAGE_SIZE) ||
2822 	    (PAGE_SIZE == SZ_64K && (sectorsize != SZ_4K &&
2823 				     sectorsize != SZ_64K))) {
2824 		btrfs_err(fs_info,
2825 			"sectorsize %llu not yet supported for page size %lu",
2826 			sectorsize, PAGE_SIZE);
2827 		ret = -EINVAL;
2828 	}
2829 
2830 	if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2831 	    nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2832 		btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2833 		ret = -EINVAL;
2834 	}
2835 	if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2836 		btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2837 			  le32_to_cpu(sb->__unused_leafsize), nodesize);
2838 		ret = -EINVAL;
2839 	}
2840 
2841 	/* Root alignment check */
2842 	if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2843 		btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2844 			   btrfs_super_root(sb));
2845 		ret = -EINVAL;
2846 	}
2847 	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2848 		btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2849 			   btrfs_super_chunk_root(sb));
2850 		ret = -EINVAL;
2851 	}
2852 	if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2853 		btrfs_warn(fs_info, "log_root block unaligned: %llu",
2854 			   btrfs_super_log_root(sb));
2855 		ret = -EINVAL;
2856 	}
2857 
2858 	if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2859 		   BTRFS_FSID_SIZE)) {
2860 		btrfs_err(fs_info,
2861 		"superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2862 			fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2863 		ret = -EINVAL;
2864 	}
2865 
2866 	if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2867 	    memcmp(fs_info->fs_devices->metadata_uuid,
2868 		   fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2869 		btrfs_err(fs_info,
2870 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2871 			fs_info->super_copy->metadata_uuid,
2872 			fs_info->fs_devices->metadata_uuid);
2873 		ret = -EINVAL;
2874 	}
2875 
2876 	if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2877 		   BTRFS_FSID_SIZE) != 0) {
2878 		btrfs_err(fs_info,
2879 			"dev_item UUID does not match metadata fsid: %pU != %pU",
2880 			fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2881 		ret = -EINVAL;
2882 	}
2883 
2884 	/*
2885 	 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2886 	 * done later
2887 	 */
2888 	if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2889 		btrfs_err(fs_info, "bytes_used is too small %llu",
2890 			  btrfs_super_bytes_used(sb));
2891 		ret = -EINVAL;
2892 	}
2893 	if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2894 		btrfs_err(fs_info, "invalid stripesize %u",
2895 			  btrfs_super_stripesize(sb));
2896 		ret = -EINVAL;
2897 	}
2898 	if (btrfs_super_num_devices(sb) > (1UL << 31))
2899 		btrfs_warn(fs_info, "suspicious number of devices: %llu",
2900 			   btrfs_super_num_devices(sb));
2901 	if (btrfs_super_num_devices(sb) == 0) {
2902 		btrfs_err(fs_info, "number of devices is 0");
2903 		ret = -EINVAL;
2904 	}
2905 
2906 	if (mirror_num >= 0 &&
2907 	    btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2908 		btrfs_err(fs_info, "super offset mismatch %llu != %u",
2909 			  btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2910 		ret = -EINVAL;
2911 	}
2912 
2913 	/*
2914 	 * Obvious sys_chunk_array corruptions, it must hold at least one key
2915 	 * and one chunk
2916 	 */
2917 	if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2918 		btrfs_err(fs_info, "system chunk array too big %u > %u",
2919 			  btrfs_super_sys_array_size(sb),
2920 			  BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2921 		ret = -EINVAL;
2922 	}
2923 	if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2924 			+ sizeof(struct btrfs_chunk)) {
2925 		btrfs_err(fs_info, "system chunk array too small %u < %zu",
2926 			  btrfs_super_sys_array_size(sb),
2927 			  sizeof(struct btrfs_disk_key)
2928 			  + sizeof(struct btrfs_chunk));
2929 		ret = -EINVAL;
2930 	}
2931 
2932 	/*
2933 	 * The generation is a global counter, we'll trust it more than the others
2934 	 * but it's still possible that it's the one that's wrong.
2935 	 */
2936 	if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2937 		btrfs_warn(fs_info,
2938 			"suspicious: generation < chunk_root_generation: %llu < %llu",
2939 			btrfs_super_generation(sb),
2940 			btrfs_super_chunk_root_generation(sb));
2941 	if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2942 	    && btrfs_super_cache_generation(sb) != (u64)-1)
2943 		btrfs_warn(fs_info,
2944 			"suspicious: generation < cache_generation: %llu < %llu",
2945 			btrfs_super_generation(sb),
2946 			btrfs_super_cache_generation(sb));
2947 
2948 	return ret;
2949 }
2950 
2951 /*
2952  * Validation of super block at mount time.
2953  * Some checks already done early at mount time, like csum type and incompat
2954  * flags will be skipped.
2955  */
2956 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2957 {
2958 	return validate_super(fs_info, fs_info->super_copy, 0);
2959 }
2960 
2961 /*
2962  * Validation of super block at write time.
2963  * Some checks like bytenr check will be skipped as their values will be
2964  * overwritten soon.
2965  * Extra checks like csum type and incompat flags will be done here.
2966  */
2967 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2968 				      struct btrfs_super_block *sb)
2969 {
2970 	int ret;
2971 
2972 	ret = validate_super(fs_info, sb, -1);
2973 	if (ret < 0)
2974 		goto out;
2975 	if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2976 		ret = -EUCLEAN;
2977 		btrfs_err(fs_info, "invalid csum type, has %u want %u",
2978 			  btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2979 		goto out;
2980 	}
2981 	if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2982 		ret = -EUCLEAN;
2983 		btrfs_err(fs_info,
2984 		"invalid incompat flags, has 0x%llx valid mask 0x%llx",
2985 			  btrfs_super_incompat_flags(sb),
2986 			  (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2987 		goto out;
2988 	}
2989 out:
2990 	if (ret < 0)
2991 		btrfs_err(fs_info,
2992 		"super block corruption detected before writing it to disk");
2993 	return ret;
2994 }
2995 
2996 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2997 {
2998 	int ret = 0;
2999 
3000 	root->node = read_tree_block(root->fs_info, bytenr,
3001 				     root->root_key.objectid, gen, level, NULL);
3002 	if (IS_ERR(root->node)) {
3003 		ret = PTR_ERR(root->node);
3004 		root->node = NULL;
3005 		return ret;
3006 	}
3007 	if (!extent_buffer_uptodate(root->node)) {
3008 		free_extent_buffer(root->node);
3009 		root->node = NULL;
3010 		return -EIO;
3011 	}
3012 
3013 	btrfs_set_root_node(&root->root_item, root->node);
3014 	root->commit_root = btrfs_root_node(root);
3015 	btrfs_set_root_refs(&root->root_item, 1);
3016 	return ret;
3017 }
3018 
3019 static int load_important_roots(struct btrfs_fs_info *fs_info)
3020 {
3021 	struct btrfs_super_block *sb = fs_info->super_copy;
3022 	u64 gen, bytenr;
3023 	int level, ret;
3024 
3025 	bytenr = btrfs_super_root(sb);
3026 	gen = btrfs_super_generation(sb);
3027 	level = btrfs_super_root_level(sb);
3028 	ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
3029 	if (ret) {
3030 		btrfs_warn(fs_info, "couldn't read tree root");
3031 		return ret;
3032 	}
3033 
3034 	if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
3035 		return 0;
3036 
3037 	bytenr = btrfs_super_block_group_root(sb);
3038 	gen = btrfs_super_block_group_root_generation(sb);
3039 	level = btrfs_super_block_group_root_level(sb);
3040 	ret = load_super_root(fs_info->block_group_root, bytenr, gen, level);
3041 	if (ret)
3042 		btrfs_warn(fs_info, "couldn't read block group root");
3043 	return ret;
3044 }
3045 
3046 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
3047 {
3048 	int backup_index = find_newest_super_backup(fs_info);
3049 	struct btrfs_super_block *sb = fs_info->super_copy;
3050 	struct btrfs_root *tree_root = fs_info->tree_root;
3051 	bool handle_error = false;
3052 	int ret = 0;
3053 	int i;
3054 
3055 	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3056 		struct btrfs_root *root;
3057 
3058 		root = btrfs_alloc_root(fs_info, BTRFS_BLOCK_GROUP_TREE_OBJECTID,
3059 					GFP_KERNEL);
3060 		if (!root)
3061 			return -ENOMEM;
3062 		fs_info->block_group_root = root;
3063 	}
3064 
3065 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
3066 		if (handle_error) {
3067 			if (!IS_ERR(tree_root->node))
3068 				free_extent_buffer(tree_root->node);
3069 			tree_root->node = NULL;
3070 
3071 			if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3072 				break;
3073 
3074 			free_root_pointers(fs_info, 0);
3075 
3076 			/*
3077 			 * Don't use the log in recovery mode, it won't be
3078 			 * valid
3079 			 */
3080 			btrfs_set_super_log_root(sb, 0);
3081 
3082 			/* We can't trust the free space cache either */
3083 			btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3084 
3085 			ret = read_backup_root(fs_info, i);
3086 			backup_index = ret;
3087 			if (ret < 0)
3088 				return ret;
3089 		}
3090 
3091 		ret = load_important_roots(fs_info);
3092 		if (ret) {
3093 			handle_error = true;
3094 			continue;
3095 		}
3096 
3097 		/*
3098 		 * No need to hold btrfs_root::objectid_mutex since the fs
3099 		 * hasn't been fully initialised and we are the only user
3100 		 */
3101 		ret = btrfs_init_root_free_objectid(tree_root);
3102 		if (ret < 0) {
3103 			handle_error = true;
3104 			continue;
3105 		}
3106 
3107 		ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
3108 
3109 		ret = btrfs_read_roots(fs_info);
3110 		if (ret < 0) {
3111 			handle_error = true;
3112 			continue;
3113 		}
3114 
3115 		/* All successful */
3116 		fs_info->generation = btrfs_header_generation(tree_root->node);
3117 		fs_info->last_trans_committed = fs_info->generation;
3118 		fs_info->last_reloc_trans = 0;
3119 
3120 		/* Always begin writing backup roots after the one being used */
3121 		if (backup_index < 0) {
3122 			fs_info->backup_root_index = 0;
3123 		} else {
3124 			fs_info->backup_root_index = backup_index + 1;
3125 			fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
3126 		}
3127 		break;
3128 	}
3129 
3130 	return ret;
3131 }
3132 
3133 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
3134 {
3135 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
3136 	INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
3137 	INIT_LIST_HEAD(&fs_info->trans_list);
3138 	INIT_LIST_HEAD(&fs_info->dead_roots);
3139 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
3140 	INIT_LIST_HEAD(&fs_info->delalloc_roots);
3141 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
3142 	spin_lock_init(&fs_info->delalloc_root_lock);
3143 	spin_lock_init(&fs_info->trans_lock);
3144 	spin_lock_init(&fs_info->fs_roots_radix_lock);
3145 	spin_lock_init(&fs_info->delayed_iput_lock);
3146 	spin_lock_init(&fs_info->defrag_inodes_lock);
3147 	spin_lock_init(&fs_info->super_lock);
3148 	spin_lock_init(&fs_info->buffer_lock);
3149 	spin_lock_init(&fs_info->unused_bgs_lock);
3150 	spin_lock_init(&fs_info->treelog_bg_lock);
3151 	spin_lock_init(&fs_info->zone_active_bgs_lock);
3152 	spin_lock_init(&fs_info->relocation_bg_lock);
3153 	rwlock_init(&fs_info->tree_mod_log_lock);
3154 	rwlock_init(&fs_info->global_root_lock);
3155 	mutex_init(&fs_info->unused_bg_unpin_mutex);
3156 	mutex_init(&fs_info->reclaim_bgs_lock);
3157 	mutex_init(&fs_info->reloc_mutex);
3158 	mutex_init(&fs_info->delalloc_root_mutex);
3159 	mutex_init(&fs_info->zoned_meta_io_lock);
3160 	seqlock_init(&fs_info->profiles_lock);
3161 
3162 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
3163 	INIT_LIST_HEAD(&fs_info->space_info);
3164 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
3165 	INIT_LIST_HEAD(&fs_info->unused_bgs);
3166 	INIT_LIST_HEAD(&fs_info->reclaim_bgs);
3167 	INIT_LIST_HEAD(&fs_info->zone_active_bgs);
3168 #ifdef CONFIG_BTRFS_DEBUG
3169 	INIT_LIST_HEAD(&fs_info->allocated_roots);
3170 	INIT_LIST_HEAD(&fs_info->allocated_ebs);
3171 	spin_lock_init(&fs_info->eb_leak_lock);
3172 #endif
3173 	extent_map_tree_init(&fs_info->mapping_tree);
3174 	btrfs_init_block_rsv(&fs_info->global_block_rsv,
3175 			     BTRFS_BLOCK_RSV_GLOBAL);
3176 	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
3177 	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
3178 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
3179 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
3180 			     BTRFS_BLOCK_RSV_DELOPS);
3181 	btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
3182 			     BTRFS_BLOCK_RSV_DELREFS);
3183 
3184 	atomic_set(&fs_info->async_delalloc_pages, 0);
3185 	atomic_set(&fs_info->defrag_running, 0);
3186 	atomic_set(&fs_info->nr_delayed_iputs, 0);
3187 	atomic64_set(&fs_info->tree_mod_seq, 0);
3188 	fs_info->global_root_tree = RB_ROOT;
3189 	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
3190 	fs_info->metadata_ratio = 0;
3191 	fs_info->defrag_inodes = RB_ROOT;
3192 	atomic64_set(&fs_info->free_chunk_space, 0);
3193 	fs_info->tree_mod_log = RB_ROOT;
3194 	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
3195 	fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
3196 	btrfs_init_ref_verify(fs_info);
3197 
3198 	fs_info->thread_pool_size = min_t(unsigned long,
3199 					  num_online_cpus() + 2, 8);
3200 
3201 	INIT_LIST_HEAD(&fs_info->ordered_roots);
3202 	spin_lock_init(&fs_info->ordered_root_lock);
3203 
3204 	btrfs_init_scrub(fs_info);
3205 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3206 	fs_info->check_integrity_print_mask = 0;
3207 #endif
3208 	btrfs_init_balance(fs_info);
3209 	btrfs_init_async_reclaim_work(fs_info);
3210 
3211 	spin_lock_init(&fs_info->block_group_cache_lock);
3212 	fs_info->block_group_cache_tree = RB_ROOT;
3213 	fs_info->first_logical_byte = (u64)-1;
3214 
3215 	extent_io_tree_init(fs_info, &fs_info->excluded_extents,
3216 			    IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
3217 
3218 	mutex_init(&fs_info->ordered_operations_mutex);
3219 	mutex_init(&fs_info->tree_log_mutex);
3220 	mutex_init(&fs_info->chunk_mutex);
3221 	mutex_init(&fs_info->transaction_kthread_mutex);
3222 	mutex_init(&fs_info->cleaner_mutex);
3223 	mutex_init(&fs_info->ro_block_group_mutex);
3224 	init_rwsem(&fs_info->commit_root_sem);
3225 	init_rwsem(&fs_info->cleanup_work_sem);
3226 	init_rwsem(&fs_info->subvol_sem);
3227 	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
3228 
3229 	btrfs_init_dev_replace_locks(fs_info);
3230 	btrfs_init_qgroup(fs_info);
3231 	btrfs_discard_init(fs_info);
3232 
3233 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
3234 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
3235 
3236 	init_waitqueue_head(&fs_info->transaction_throttle);
3237 	init_waitqueue_head(&fs_info->transaction_wait);
3238 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
3239 	init_waitqueue_head(&fs_info->async_submit_wait);
3240 	init_waitqueue_head(&fs_info->delayed_iputs_wait);
3241 
3242 	/* Usable values until the real ones are cached from the superblock */
3243 	fs_info->nodesize = 4096;
3244 	fs_info->sectorsize = 4096;
3245 	fs_info->sectorsize_bits = ilog2(4096);
3246 	fs_info->stripesize = 4096;
3247 
3248 	spin_lock_init(&fs_info->swapfile_pins_lock);
3249 	fs_info->swapfile_pins = RB_ROOT;
3250 
3251 	fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
3252 	INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
3253 }
3254 
3255 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
3256 {
3257 	int ret;
3258 
3259 	fs_info->sb = sb;
3260 	sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
3261 	sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3262 
3263 	ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3264 	if (ret)
3265 		return ret;
3266 
3267 	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3268 	if (ret)
3269 		return ret;
3270 
3271 	fs_info->dirty_metadata_batch = PAGE_SIZE *
3272 					(1 + ilog2(nr_cpu_ids));
3273 
3274 	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3275 	if (ret)
3276 		return ret;
3277 
3278 	ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3279 			GFP_KERNEL);
3280 	if (ret)
3281 		return ret;
3282 
3283 	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3284 					GFP_KERNEL);
3285 	if (!fs_info->delayed_root)
3286 		return -ENOMEM;
3287 	btrfs_init_delayed_root(fs_info->delayed_root);
3288 
3289 	if (sb_rdonly(sb))
3290 		set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3291 
3292 	return btrfs_alloc_stripe_hash_table(fs_info);
3293 }
3294 
3295 static int btrfs_uuid_rescan_kthread(void *data)
3296 {
3297 	struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3298 	int ret;
3299 
3300 	/*
3301 	 * 1st step is to iterate through the existing UUID tree and
3302 	 * to delete all entries that contain outdated data.
3303 	 * 2nd step is to add all missing entries to the UUID tree.
3304 	 */
3305 	ret = btrfs_uuid_tree_iterate(fs_info);
3306 	if (ret < 0) {
3307 		if (ret != -EINTR)
3308 			btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3309 				   ret);
3310 		up(&fs_info->uuid_tree_rescan_sem);
3311 		return ret;
3312 	}
3313 	return btrfs_uuid_scan_kthread(data);
3314 }
3315 
3316 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3317 {
3318 	struct task_struct *task;
3319 
3320 	down(&fs_info->uuid_tree_rescan_sem);
3321 	task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3322 	if (IS_ERR(task)) {
3323 		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
3324 		btrfs_warn(fs_info, "failed to start uuid_rescan task");
3325 		up(&fs_info->uuid_tree_rescan_sem);
3326 		return PTR_ERR(task);
3327 	}
3328 
3329 	return 0;
3330 }
3331 
3332 /*
3333  * Some options only have meaning at mount time and shouldn't persist across
3334  * remounts, or be displayed. Clear these at the end of mount and remount
3335  * code paths.
3336  */
3337 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3338 {
3339 	btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3340 	btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3341 }
3342 
3343 /*
3344  * Mounting logic specific to read-write file systems. Shared by open_ctree
3345  * and btrfs_remount when remounting from read-only to read-write.
3346  */
3347 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3348 {
3349 	int ret;
3350 	const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3351 	bool clear_free_space_tree = false;
3352 
3353 	if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3354 	    btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3355 		clear_free_space_tree = true;
3356 	} else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3357 		   !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3358 		btrfs_warn(fs_info, "free space tree is invalid");
3359 		clear_free_space_tree = true;
3360 	}
3361 
3362 	if (clear_free_space_tree) {
3363 		btrfs_info(fs_info, "clearing free space tree");
3364 		ret = btrfs_clear_free_space_tree(fs_info);
3365 		if (ret) {
3366 			btrfs_warn(fs_info,
3367 				   "failed to clear free space tree: %d", ret);
3368 			goto out;
3369 		}
3370 	}
3371 
3372 	/*
3373 	 * btrfs_find_orphan_roots() is responsible for finding all the dead
3374 	 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3375 	 * them into the fs_info->fs_roots_radix tree. This must be done before
3376 	 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3377 	 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3378 	 * item before the root's tree is deleted - this means that if we unmount
3379 	 * or crash before the deletion completes, on the next mount we will not
3380 	 * delete what remains of the tree because the orphan item does not
3381 	 * exists anymore, which is what tells us we have a pending deletion.
3382 	 */
3383 	ret = btrfs_find_orphan_roots(fs_info);
3384 	if (ret)
3385 		goto out;
3386 
3387 	ret = btrfs_cleanup_fs_roots(fs_info);
3388 	if (ret)
3389 		goto out;
3390 
3391 	down_read(&fs_info->cleanup_work_sem);
3392 	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3393 	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3394 		up_read(&fs_info->cleanup_work_sem);
3395 		goto out;
3396 	}
3397 	up_read(&fs_info->cleanup_work_sem);
3398 
3399 	mutex_lock(&fs_info->cleaner_mutex);
3400 	ret = btrfs_recover_relocation(fs_info);
3401 	mutex_unlock(&fs_info->cleaner_mutex);
3402 	if (ret < 0) {
3403 		btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3404 		goto out;
3405 	}
3406 
3407 	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3408 	    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3409 		btrfs_info(fs_info, "creating free space tree");
3410 		ret = btrfs_create_free_space_tree(fs_info);
3411 		if (ret) {
3412 			btrfs_warn(fs_info,
3413 				"failed to create free space tree: %d", ret);
3414 			goto out;
3415 		}
3416 	}
3417 
3418 	if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3419 		ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3420 		if (ret)
3421 			goto out;
3422 	}
3423 
3424 	ret = btrfs_resume_balance_async(fs_info);
3425 	if (ret)
3426 		goto out;
3427 
3428 	ret = btrfs_resume_dev_replace_async(fs_info);
3429 	if (ret) {
3430 		btrfs_warn(fs_info, "failed to resume dev_replace");
3431 		goto out;
3432 	}
3433 
3434 	btrfs_qgroup_rescan_resume(fs_info);
3435 
3436 	if (!fs_info->uuid_root) {
3437 		btrfs_info(fs_info, "creating UUID tree");
3438 		ret = btrfs_create_uuid_tree(fs_info);
3439 		if (ret) {
3440 			btrfs_warn(fs_info,
3441 				   "failed to create the UUID tree %d", ret);
3442 			goto out;
3443 		}
3444 	}
3445 
3446 out:
3447 	return ret;
3448 }
3449 
3450 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3451 		      char *options)
3452 {
3453 	u32 sectorsize;
3454 	u32 nodesize;
3455 	u32 stripesize;
3456 	u64 generation;
3457 	u64 features;
3458 	u16 csum_type;
3459 	struct btrfs_super_block *disk_super;
3460 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3461 	struct btrfs_root *tree_root;
3462 	struct btrfs_root *chunk_root;
3463 	int ret;
3464 	int err = -EINVAL;
3465 	int level;
3466 
3467 	ret = init_mount_fs_info(fs_info, sb);
3468 	if (ret) {
3469 		err = ret;
3470 		goto fail;
3471 	}
3472 
3473 	/* These need to be init'ed before we start creating inodes and such. */
3474 	tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3475 				     GFP_KERNEL);
3476 	fs_info->tree_root = tree_root;
3477 	chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3478 				      GFP_KERNEL);
3479 	fs_info->chunk_root = chunk_root;
3480 	if (!tree_root || !chunk_root) {
3481 		err = -ENOMEM;
3482 		goto fail;
3483 	}
3484 
3485 	fs_info->btree_inode = new_inode(sb);
3486 	if (!fs_info->btree_inode) {
3487 		err = -ENOMEM;
3488 		goto fail;
3489 	}
3490 	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3491 	btrfs_init_btree_inode(fs_info);
3492 
3493 	invalidate_bdev(fs_devices->latest_dev->bdev);
3494 
3495 	/*
3496 	 * Read super block and check the signature bytes only
3497 	 */
3498 	disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3499 	if (IS_ERR(disk_super)) {
3500 		err = PTR_ERR(disk_super);
3501 		goto fail_alloc;
3502 	}
3503 
3504 	/*
3505 	 * Verify the type first, if that or the checksum value are
3506 	 * corrupted, we'll find out
3507 	 */
3508 	csum_type = btrfs_super_csum_type(disk_super);
3509 	if (!btrfs_supported_super_csum(csum_type)) {
3510 		btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3511 			  csum_type);
3512 		err = -EINVAL;
3513 		btrfs_release_disk_super(disk_super);
3514 		goto fail_alloc;
3515 	}
3516 
3517 	fs_info->csum_size = btrfs_super_csum_size(disk_super);
3518 
3519 	ret = btrfs_init_csum_hash(fs_info, csum_type);
3520 	if (ret) {
3521 		err = ret;
3522 		btrfs_release_disk_super(disk_super);
3523 		goto fail_alloc;
3524 	}
3525 
3526 	/*
3527 	 * We want to check superblock checksum, the type is stored inside.
3528 	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3529 	 */
3530 	if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
3531 		btrfs_err(fs_info, "superblock checksum mismatch");
3532 		err = -EINVAL;
3533 		btrfs_release_disk_super(disk_super);
3534 		goto fail_alloc;
3535 	}
3536 
3537 	/*
3538 	 * super_copy is zeroed at allocation time and we never touch the
3539 	 * following bytes up to INFO_SIZE, the checksum is calculated from
3540 	 * the whole block of INFO_SIZE
3541 	 */
3542 	memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3543 	btrfs_release_disk_super(disk_super);
3544 
3545 	disk_super = fs_info->super_copy;
3546 
3547 
3548 	features = btrfs_super_flags(disk_super);
3549 	if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3550 		features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3551 		btrfs_set_super_flags(disk_super, features);
3552 		btrfs_info(fs_info,
3553 			"found metadata UUID change in progress flag, clearing");
3554 	}
3555 
3556 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
3557 	       sizeof(*fs_info->super_for_commit));
3558 
3559 	ret = btrfs_validate_mount_super(fs_info);
3560 	if (ret) {
3561 		btrfs_err(fs_info, "superblock contains fatal errors");
3562 		err = -EINVAL;
3563 		goto fail_alloc;
3564 	}
3565 
3566 	if (!btrfs_super_root(disk_super))
3567 		goto fail_alloc;
3568 
3569 	/* check FS state, whether FS is broken. */
3570 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3571 		set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3572 
3573 	/*
3574 	 * In the long term, we'll store the compression type in the super
3575 	 * block, and it'll be used for per file compression control.
3576 	 */
3577 	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3578 
3579 	/*
3580 	 * Flag our filesystem as having big metadata blocks if they are bigger
3581 	 * than the page size.
3582 	 */
3583 	if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3584 		if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3585 			btrfs_info(fs_info,
3586 				"flagging fs with big metadata feature");
3587 		features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3588 	}
3589 
3590 	/* Set up fs_info before parsing mount options */
3591 	nodesize = btrfs_super_nodesize(disk_super);
3592 	sectorsize = btrfs_super_sectorsize(disk_super);
3593 	stripesize = sectorsize;
3594 	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3595 	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3596 
3597 	fs_info->nodesize = nodesize;
3598 	fs_info->sectorsize = sectorsize;
3599 	fs_info->sectorsize_bits = ilog2(sectorsize);
3600 	fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3601 	fs_info->stripesize = stripesize;
3602 
3603 	ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3604 	if (ret) {
3605 		err = ret;
3606 		goto fail_alloc;
3607 	}
3608 
3609 	features = btrfs_super_incompat_flags(disk_super) &
3610 		~BTRFS_FEATURE_INCOMPAT_SUPP;
3611 	if (features) {
3612 		btrfs_err(fs_info,
3613 		    "cannot mount because of unsupported optional features (%llx)",
3614 		    features);
3615 		err = -EINVAL;
3616 		goto fail_alloc;
3617 	}
3618 
3619 	features = btrfs_super_incompat_flags(disk_super);
3620 	features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3621 	if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3622 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3623 	else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3624 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3625 
3626 	if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3627 		btrfs_info(fs_info, "has skinny extents");
3628 
3629 	/*
3630 	 * mixed block groups end up with duplicate but slightly offset
3631 	 * extent buffers for the same range.  It leads to corruptions
3632 	 */
3633 	if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3634 	    (sectorsize != nodesize)) {
3635 		btrfs_err(fs_info,
3636 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3637 			nodesize, sectorsize);
3638 		goto fail_alloc;
3639 	}
3640 
3641 	/*
3642 	 * Needn't use the lock because there is no other task which will
3643 	 * update the flag.
3644 	 */
3645 	btrfs_set_super_incompat_flags(disk_super, features);
3646 
3647 	features = btrfs_super_compat_ro_flags(disk_super) &
3648 		~BTRFS_FEATURE_COMPAT_RO_SUPP;
3649 	if (!sb_rdonly(sb) && features) {
3650 		btrfs_err(fs_info,
3651 	"cannot mount read-write because of unsupported optional features (%llx)",
3652 		       features);
3653 		err = -EINVAL;
3654 		goto fail_alloc;
3655 	}
3656 
3657 	if (sectorsize < PAGE_SIZE) {
3658 		struct btrfs_subpage_info *subpage_info;
3659 
3660 		btrfs_warn(fs_info,
3661 		"read-write for sector size %u with page size %lu is experimental",
3662 			   sectorsize, PAGE_SIZE);
3663 		if (btrfs_super_incompat_flags(fs_info->super_copy) &
3664 			BTRFS_FEATURE_INCOMPAT_RAID56) {
3665 			btrfs_err(fs_info,
3666 		"RAID56 is not yet supported for sector size %u with page size %lu",
3667 				sectorsize, PAGE_SIZE);
3668 			err = -EINVAL;
3669 			goto fail_alloc;
3670 		}
3671 		subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3672 		if (!subpage_info)
3673 			goto fail_alloc;
3674 		btrfs_init_subpage_info(subpage_info, sectorsize);
3675 		fs_info->subpage_info = subpage_info;
3676 	}
3677 
3678 	ret = btrfs_init_workqueues(fs_info);
3679 	if (ret) {
3680 		err = ret;
3681 		goto fail_sb_buffer;
3682 	}
3683 
3684 	sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3685 	sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3686 
3687 	sb->s_blocksize = sectorsize;
3688 	sb->s_blocksize_bits = blksize_bits(sectorsize);
3689 	memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3690 
3691 	mutex_lock(&fs_info->chunk_mutex);
3692 	ret = btrfs_read_sys_array(fs_info);
3693 	mutex_unlock(&fs_info->chunk_mutex);
3694 	if (ret) {
3695 		btrfs_err(fs_info, "failed to read the system array: %d", ret);
3696 		goto fail_sb_buffer;
3697 	}
3698 
3699 	generation = btrfs_super_chunk_root_generation(disk_super);
3700 	level = btrfs_super_chunk_root_level(disk_super);
3701 	ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3702 			      generation, level);
3703 	if (ret) {
3704 		btrfs_err(fs_info, "failed to read chunk root");
3705 		goto fail_tree_roots;
3706 	}
3707 
3708 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3709 			   offsetof(struct btrfs_header, chunk_tree_uuid),
3710 			   BTRFS_UUID_SIZE);
3711 
3712 	ret = btrfs_read_chunk_tree(fs_info);
3713 	if (ret) {
3714 		btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3715 		goto fail_tree_roots;
3716 	}
3717 
3718 	/*
3719 	 * At this point we know all the devices that make this filesystem,
3720 	 * including the seed devices but we don't know yet if the replace
3721 	 * target is required. So free devices that are not part of this
3722 	 * filesystem but skip the replace target device which is checked
3723 	 * below in btrfs_init_dev_replace().
3724 	 */
3725 	btrfs_free_extra_devids(fs_devices);
3726 	if (!fs_devices->latest_dev->bdev) {
3727 		btrfs_err(fs_info, "failed to read devices");
3728 		goto fail_tree_roots;
3729 	}
3730 
3731 	ret = init_tree_roots(fs_info);
3732 	if (ret)
3733 		goto fail_tree_roots;
3734 
3735 	/*
3736 	 * Get zone type information of zoned block devices. This will also
3737 	 * handle emulation of a zoned filesystem if a regular device has the
3738 	 * zoned incompat feature flag set.
3739 	 */
3740 	ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3741 	if (ret) {
3742 		btrfs_err(fs_info,
3743 			  "zoned: failed to read device zone info: %d",
3744 			  ret);
3745 		goto fail_block_groups;
3746 	}
3747 
3748 	/*
3749 	 * If we have a uuid root and we're not being told to rescan we need to
3750 	 * check the generation here so we can set the
3751 	 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3752 	 * transaction during a balance or the log replay without updating the
3753 	 * uuid generation, and then if we crash we would rescan the uuid tree,
3754 	 * even though it was perfectly fine.
3755 	 */
3756 	if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3757 	    fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3758 		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3759 
3760 	ret = btrfs_verify_dev_extents(fs_info);
3761 	if (ret) {
3762 		btrfs_err(fs_info,
3763 			  "failed to verify dev extents against chunks: %d",
3764 			  ret);
3765 		goto fail_block_groups;
3766 	}
3767 	ret = btrfs_recover_balance(fs_info);
3768 	if (ret) {
3769 		btrfs_err(fs_info, "failed to recover balance: %d", ret);
3770 		goto fail_block_groups;
3771 	}
3772 
3773 	ret = btrfs_init_dev_stats(fs_info);
3774 	if (ret) {
3775 		btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3776 		goto fail_block_groups;
3777 	}
3778 
3779 	ret = btrfs_init_dev_replace(fs_info);
3780 	if (ret) {
3781 		btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3782 		goto fail_block_groups;
3783 	}
3784 
3785 	ret = btrfs_check_zoned_mode(fs_info);
3786 	if (ret) {
3787 		btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3788 			  ret);
3789 		goto fail_block_groups;
3790 	}
3791 
3792 	ret = btrfs_sysfs_add_fsid(fs_devices);
3793 	if (ret) {
3794 		btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3795 				ret);
3796 		goto fail_block_groups;
3797 	}
3798 
3799 	ret = btrfs_sysfs_add_mounted(fs_info);
3800 	if (ret) {
3801 		btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3802 		goto fail_fsdev_sysfs;
3803 	}
3804 
3805 	ret = btrfs_init_space_info(fs_info);
3806 	if (ret) {
3807 		btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3808 		goto fail_sysfs;
3809 	}
3810 
3811 	ret = btrfs_read_block_groups(fs_info);
3812 	if (ret) {
3813 		btrfs_err(fs_info, "failed to read block groups: %d", ret);
3814 		goto fail_sysfs;
3815 	}
3816 
3817 	btrfs_free_zone_cache(fs_info);
3818 
3819 	if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3820 	    !btrfs_check_rw_degradable(fs_info, NULL)) {
3821 		btrfs_warn(fs_info,
3822 		"writable mount is not allowed due to too many missing devices");
3823 		goto fail_sysfs;
3824 	}
3825 
3826 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3827 					       "btrfs-cleaner");
3828 	if (IS_ERR(fs_info->cleaner_kthread))
3829 		goto fail_sysfs;
3830 
3831 	fs_info->transaction_kthread = kthread_run(transaction_kthread,
3832 						   tree_root,
3833 						   "btrfs-transaction");
3834 	if (IS_ERR(fs_info->transaction_kthread))
3835 		goto fail_cleaner;
3836 
3837 	if (!btrfs_test_opt(fs_info, NOSSD) &&
3838 	    !fs_info->fs_devices->rotating) {
3839 		btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3840 	}
3841 
3842 	/*
3843 	 * Mount does not set all options immediately, we can do it now and do
3844 	 * not have to wait for transaction commit
3845 	 */
3846 	btrfs_apply_pending_changes(fs_info);
3847 
3848 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3849 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3850 		ret = btrfsic_mount(fs_info, fs_devices,
3851 				    btrfs_test_opt(fs_info,
3852 					CHECK_INTEGRITY_DATA) ? 1 : 0,
3853 				    fs_info->check_integrity_print_mask);
3854 		if (ret)
3855 			btrfs_warn(fs_info,
3856 				"failed to initialize integrity check module: %d",
3857 				ret);
3858 	}
3859 #endif
3860 	ret = btrfs_read_qgroup_config(fs_info);
3861 	if (ret)
3862 		goto fail_trans_kthread;
3863 
3864 	if (btrfs_build_ref_tree(fs_info))
3865 		btrfs_err(fs_info, "couldn't build ref tree");
3866 
3867 	/* do not make disk changes in broken FS or nologreplay is given */
3868 	if (btrfs_super_log_root(disk_super) != 0 &&
3869 	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3870 		btrfs_info(fs_info, "start tree-log replay");
3871 		ret = btrfs_replay_log(fs_info, fs_devices);
3872 		if (ret) {
3873 			err = ret;
3874 			goto fail_qgroup;
3875 		}
3876 	}
3877 
3878 	fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3879 	if (IS_ERR(fs_info->fs_root)) {
3880 		err = PTR_ERR(fs_info->fs_root);
3881 		btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3882 		fs_info->fs_root = NULL;
3883 		goto fail_qgroup;
3884 	}
3885 
3886 	if (sb_rdonly(sb))
3887 		goto clear_oneshot;
3888 
3889 	ret = btrfs_start_pre_rw_mount(fs_info);
3890 	if (ret) {
3891 		close_ctree(fs_info);
3892 		return ret;
3893 	}
3894 	btrfs_discard_resume(fs_info);
3895 
3896 	if (fs_info->uuid_root &&
3897 	    (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3898 	     fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3899 		btrfs_info(fs_info, "checking UUID tree");
3900 		ret = btrfs_check_uuid_tree(fs_info);
3901 		if (ret) {
3902 			btrfs_warn(fs_info,
3903 				"failed to check the UUID tree: %d", ret);
3904 			close_ctree(fs_info);
3905 			return ret;
3906 		}
3907 	}
3908 
3909 	set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3910 
3911 	/* Kick the cleaner thread so it'll start deleting snapshots. */
3912 	if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3913 		wake_up_process(fs_info->cleaner_kthread);
3914 
3915 clear_oneshot:
3916 	btrfs_clear_oneshot_options(fs_info);
3917 	return 0;
3918 
3919 fail_qgroup:
3920 	btrfs_free_qgroup_config(fs_info);
3921 fail_trans_kthread:
3922 	kthread_stop(fs_info->transaction_kthread);
3923 	btrfs_cleanup_transaction(fs_info);
3924 	btrfs_free_fs_roots(fs_info);
3925 fail_cleaner:
3926 	kthread_stop(fs_info->cleaner_kthread);
3927 
3928 	/*
3929 	 * make sure we're done with the btree inode before we stop our
3930 	 * kthreads
3931 	 */
3932 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3933 
3934 fail_sysfs:
3935 	btrfs_sysfs_remove_mounted(fs_info);
3936 
3937 fail_fsdev_sysfs:
3938 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3939 
3940 fail_block_groups:
3941 	btrfs_put_block_group_cache(fs_info);
3942 
3943 fail_tree_roots:
3944 	if (fs_info->data_reloc_root)
3945 		btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3946 	free_root_pointers(fs_info, true);
3947 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3948 
3949 fail_sb_buffer:
3950 	btrfs_stop_all_workers(fs_info);
3951 	btrfs_free_block_groups(fs_info);
3952 fail_alloc:
3953 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3954 
3955 	iput(fs_info->btree_inode);
3956 fail:
3957 	btrfs_close_devices(fs_info->fs_devices);
3958 	return err;
3959 }
3960 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3961 
3962 static void btrfs_end_super_write(struct bio *bio)
3963 {
3964 	struct btrfs_device *device = bio->bi_private;
3965 	struct bio_vec *bvec;
3966 	struct bvec_iter_all iter_all;
3967 	struct page *page;
3968 
3969 	bio_for_each_segment_all(bvec, bio, iter_all) {
3970 		page = bvec->bv_page;
3971 
3972 		if (bio->bi_status) {
3973 			btrfs_warn_rl_in_rcu(device->fs_info,
3974 				"lost page write due to IO error on %s (%d)",
3975 				rcu_str_deref(device->name),
3976 				blk_status_to_errno(bio->bi_status));
3977 			ClearPageUptodate(page);
3978 			SetPageError(page);
3979 			btrfs_dev_stat_inc_and_print(device,
3980 						     BTRFS_DEV_STAT_WRITE_ERRS);
3981 		} else {
3982 			SetPageUptodate(page);
3983 		}
3984 
3985 		put_page(page);
3986 		unlock_page(page);
3987 	}
3988 
3989 	bio_put(bio);
3990 }
3991 
3992 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3993 						   int copy_num)
3994 {
3995 	struct btrfs_super_block *super;
3996 	struct page *page;
3997 	u64 bytenr, bytenr_orig;
3998 	struct address_space *mapping = bdev->bd_inode->i_mapping;
3999 	int ret;
4000 
4001 	bytenr_orig = btrfs_sb_offset(copy_num);
4002 	ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
4003 	if (ret == -ENOENT)
4004 		return ERR_PTR(-EINVAL);
4005 	else if (ret)
4006 		return ERR_PTR(ret);
4007 
4008 	if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
4009 		return ERR_PTR(-EINVAL);
4010 
4011 	page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
4012 	if (IS_ERR(page))
4013 		return ERR_CAST(page);
4014 
4015 	super = page_address(page);
4016 	if (btrfs_super_magic(super) != BTRFS_MAGIC) {
4017 		btrfs_release_disk_super(super);
4018 		return ERR_PTR(-ENODATA);
4019 	}
4020 
4021 	if (btrfs_super_bytenr(super) != bytenr_orig) {
4022 		btrfs_release_disk_super(super);
4023 		return ERR_PTR(-EINVAL);
4024 	}
4025 
4026 	return super;
4027 }
4028 
4029 
4030 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
4031 {
4032 	struct btrfs_super_block *super, *latest = NULL;
4033 	int i;
4034 	u64 transid = 0;
4035 
4036 	/* we would like to check all the supers, but that would make
4037 	 * a btrfs mount succeed after a mkfs from a different FS.
4038 	 * So, we need to add a special mount option to scan for
4039 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
4040 	 */
4041 	for (i = 0; i < 1; i++) {
4042 		super = btrfs_read_dev_one_super(bdev, i);
4043 		if (IS_ERR(super))
4044 			continue;
4045 
4046 		if (!latest || btrfs_super_generation(super) > transid) {
4047 			if (latest)
4048 				btrfs_release_disk_super(super);
4049 
4050 			latest = super;
4051 			transid = btrfs_super_generation(super);
4052 		}
4053 	}
4054 
4055 	return super;
4056 }
4057 
4058 /*
4059  * Write superblock @sb to the @device. Do not wait for completion, all the
4060  * pages we use for writing are locked.
4061  *
4062  * Write @max_mirrors copies of the superblock, where 0 means default that fit
4063  * the expected device size at commit time. Note that max_mirrors must be
4064  * same for write and wait phases.
4065  *
4066  * Return number of errors when page is not found or submission fails.
4067  */
4068 static int write_dev_supers(struct btrfs_device *device,
4069 			    struct btrfs_super_block *sb, int max_mirrors)
4070 {
4071 	struct btrfs_fs_info *fs_info = device->fs_info;
4072 	struct address_space *mapping = device->bdev->bd_inode->i_mapping;
4073 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
4074 	int i;
4075 	int errors = 0;
4076 	int ret;
4077 	u64 bytenr, bytenr_orig;
4078 
4079 	if (max_mirrors == 0)
4080 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4081 
4082 	shash->tfm = fs_info->csum_shash;
4083 
4084 	for (i = 0; i < max_mirrors; i++) {
4085 		struct page *page;
4086 		struct bio *bio;
4087 		struct btrfs_super_block *disk_super;
4088 
4089 		bytenr_orig = btrfs_sb_offset(i);
4090 		ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
4091 		if (ret == -ENOENT) {
4092 			continue;
4093 		} else if (ret < 0) {
4094 			btrfs_err(device->fs_info,
4095 				"couldn't get super block location for mirror %d",
4096 				i);
4097 			errors++;
4098 			continue;
4099 		}
4100 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4101 		    device->commit_total_bytes)
4102 			break;
4103 
4104 		btrfs_set_super_bytenr(sb, bytenr_orig);
4105 
4106 		crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
4107 				    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
4108 				    sb->csum);
4109 
4110 		page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
4111 					   GFP_NOFS);
4112 		if (!page) {
4113 			btrfs_err(device->fs_info,
4114 			    "couldn't get super block page for bytenr %llu",
4115 			    bytenr);
4116 			errors++;
4117 			continue;
4118 		}
4119 
4120 		/* Bump the refcount for wait_dev_supers() */
4121 		get_page(page);
4122 
4123 		disk_super = page_address(page);
4124 		memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
4125 
4126 		/*
4127 		 * Directly use bios here instead of relying on the page cache
4128 		 * to do I/O, so we don't lose the ability to do integrity
4129 		 * checking.
4130 		 */
4131 		bio = bio_alloc(device->bdev, 1,
4132 				REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
4133 				GFP_NOFS);
4134 		bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
4135 		bio->bi_private = device;
4136 		bio->bi_end_io = btrfs_end_super_write;
4137 		__bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
4138 			       offset_in_page(bytenr));
4139 
4140 		/*
4141 		 * We FUA only the first super block.  The others we allow to
4142 		 * go down lazy and there's a short window where the on-disk
4143 		 * copies might still contain the older version.
4144 		 */
4145 		if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
4146 			bio->bi_opf |= REQ_FUA;
4147 
4148 		btrfsic_submit_bio(bio);
4149 
4150 		if (btrfs_advance_sb_log(device, i))
4151 			errors++;
4152 	}
4153 	return errors < i ? 0 : -1;
4154 }
4155 
4156 /*
4157  * Wait for write completion of superblocks done by write_dev_supers,
4158  * @max_mirrors same for write and wait phases.
4159  *
4160  * Return number of errors when page is not found or not marked up to
4161  * date.
4162  */
4163 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
4164 {
4165 	int i;
4166 	int errors = 0;
4167 	bool primary_failed = false;
4168 	int ret;
4169 	u64 bytenr;
4170 
4171 	if (max_mirrors == 0)
4172 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4173 
4174 	for (i = 0; i < max_mirrors; i++) {
4175 		struct page *page;
4176 
4177 		ret = btrfs_sb_log_location(device, i, READ, &bytenr);
4178 		if (ret == -ENOENT) {
4179 			break;
4180 		} else if (ret < 0) {
4181 			errors++;
4182 			if (i == 0)
4183 				primary_failed = true;
4184 			continue;
4185 		}
4186 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4187 		    device->commit_total_bytes)
4188 			break;
4189 
4190 		page = find_get_page(device->bdev->bd_inode->i_mapping,
4191 				     bytenr >> PAGE_SHIFT);
4192 		if (!page) {
4193 			errors++;
4194 			if (i == 0)
4195 				primary_failed = true;
4196 			continue;
4197 		}
4198 		/* Page is submitted locked and unlocked once the IO completes */
4199 		wait_on_page_locked(page);
4200 		if (PageError(page)) {
4201 			errors++;
4202 			if (i == 0)
4203 				primary_failed = true;
4204 		}
4205 
4206 		/* Drop our reference */
4207 		put_page(page);
4208 
4209 		/* Drop the reference from the writing run */
4210 		put_page(page);
4211 	}
4212 
4213 	/* log error, force error return */
4214 	if (primary_failed) {
4215 		btrfs_err(device->fs_info, "error writing primary super block to device %llu",
4216 			  device->devid);
4217 		return -1;
4218 	}
4219 
4220 	return errors < i ? 0 : -1;
4221 }
4222 
4223 /*
4224  * endio for the write_dev_flush, this will wake anyone waiting
4225  * for the barrier when it is done
4226  */
4227 static void btrfs_end_empty_barrier(struct bio *bio)
4228 {
4229 	complete(bio->bi_private);
4230 }
4231 
4232 /*
4233  * Submit a flush request to the device if it supports it. Error handling is
4234  * done in the waiting counterpart.
4235  */
4236 static void write_dev_flush(struct btrfs_device *device)
4237 {
4238 	struct bio *bio = device->flush_bio;
4239 
4240 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4241 	/*
4242 	 * When a disk has write caching disabled, we skip submission of a bio
4243 	 * with flush and sync requests before writing the superblock, since
4244 	 * it's not needed. However when the integrity checker is enabled, this
4245 	 * results in reports that there are metadata blocks referred by a
4246 	 * superblock that were not properly flushed. So don't skip the bio
4247 	 * submission only when the integrity checker is enabled for the sake
4248 	 * of simplicity, since this is a debug tool and not meant for use in
4249 	 * non-debug builds.
4250 	 */
4251 	struct request_queue *q = bdev_get_queue(device->bdev);
4252 	if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
4253 		return;
4254 #endif
4255 
4256 	bio_reset(bio, device->bdev, REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
4257 	bio->bi_end_io = btrfs_end_empty_barrier;
4258 	init_completion(&device->flush_wait);
4259 	bio->bi_private = &device->flush_wait;
4260 
4261 	btrfsic_submit_bio(bio);
4262 	set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4263 }
4264 
4265 /*
4266  * If the flush bio has been submitted by write_dev_flush, wait for it.
4267  */
4268 static blk_status_t wait_dev_flush(struct btrfs_device *device)
4269 {
4270 	struct bio *bio = device->flush_bio;
4271 
4272 	if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
4273 		return BLK_STS_OK;
4274 
4275 	clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4276 	wait_for_completion_io(&device->flush_wait);
4277 
4278 	return bio->bi_status;
4279 }
4280 
4281 static int check_barrier_error(struct btrfs_fs_info *fs_info)
4282 {
4283 	if (!btrfs_check_rw_degradable(fs_info, NULL))
4284 		return -EIO;
4285 	return 0;
4286 }
4287 
4288 /*
4289  * send an empty flush down to each device in parallel,
4290  * then wait for them
4291  */
4292 static int barrier_all_devices(struct btrfs_fs_info *info)
4293 {
4294 	struct list_head *head;
4295 	struct btrfs_device *dev;
4296 	int errors_wait = 0;
4297 	blk_status_t ret;
4298 
4299 	lockdep_assert_held(&info->fs_devices->device_list_mutex);
4300 	/* send down all the barriers */
4301 	head = &info->fs_devices->devices;
4302 	list_for_each_entry(dev, head, dev_list) {
4303 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4304 			continue;
4305 		if (!dev->bdev)
4306 			continue;
4307 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4308 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4309 			continue;
4310 
4311 		write_dev_flush(dev);
4312 		dev->last_flush_error = BLK_STS_OK;
4313 	}
4314 
4315 	/* wait for all the barriers */
4316 	list_for_each_entry(dev, head, dev_list) {
4317 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4318 			continue;
4319 		if (!dev->bdev) {
4320 			errors_wait++;
4321 			continue;
4322 		}
4323 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4324 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4325 			continue;
4326 
4327 		ret = wait_dev_flush(dev);
4328 		if (ret) {
4329 			dev->last_flush_error = ret;
4330 			btrfs_dev_stat_inc_and_print(dev,
4331 					BTRFS_DEV_STAT_FLUSH_ERRS);
4332 			errors_wait++;
4333 		}
4334 	}
4335 
4336 	if (errors_wait) {
4337 		/*
4338 		 * At some point we need the status of all disks
4339 		 * to arrive at the volume status. So error checking
4340 		 * is being pushed to a separate loop.
4341 		 */
4342 		return check_barrier_error(info);
4343 	}
4344 	return 0;
4345 }
4346 
4347 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4348 {
4349 	int raid_type;
4350 	int min_tolerated = INT_MAX;
4351 
4352 	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4353 	    (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4354 		min_tolerated = min_t(int, min_tolerated,
4355 				    btrfs_raid_array[BTRFS_RAID_SINGLE].
4356 				    tolerated_failures);
4357 
4358 	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4359 		if (raid_type == BTRFS_RAID_SINGLE)
4360 			continue;
4361 		if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4362 			continue;
4363 		min_tolerated = min_t(int, min_tolerated,
4364 				    btrfs_raid_array[raid_type].
4365 				    tolerated_failures);
4366 	}
4367 
4368 	if (min_tolerated == INT_MAX) {
4369 		pr_warn("BTRFS: unknown raid flag: %llu", flags);
4370 		min_tolerated = 0;
4371 	}
4372 
4373 	return min_tolerated;
4374 }
4375 
4376 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4377 {
4378 	struct list_head *head;
4379 	struct btrfs_device *dev;
4380 	struct btrfs_super_block *sb;
4381 	struct btrfs_dev_item *dev_item;
4382 	int ret;
4383 	int do_barriers;
4384 	int max_errors;
4385 	int total_errors = 0;
4386 	u64 flags;
4387 
4388 	do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4389 
4390 	/*
4391 	 * max_mirrors == 0 indicates we're from commit_transaction,
4392 	 * not from fsync where the tree roots in fs_info have not
4393 	 * been consistent on disk.
4394 	 */
4395 	if (max_mirrors == 0)
4396 		backup_super_roots(fs_info);
4397 
4398 	sb = fs_info->super_for_commit;
4399 	dev_item = &sb->dev_item;
4400 
4401 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
4402 	head = &fs_info->fs_devices->devices;
4403 	max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4404 
4405 	if (do_barriers) {
4406 		ret = barrier_all_devices(fs_info);
4407 		if (ret) {
4408 			mutex_unlock(
4409 				&fs_info->fs_devices->device_list_mutex);
4410 			btrfs_handle_fs_error(fs_info, ret,
4411 					      "errors while submitting device barriers.");
4412 			return ret;
4413 		}
4414 	}
4415 
4416 	list_for_each_entry(dev, head, dev_list) {
4417 		if (!dev->bdev) {
4418 			total_errors++;
4419 			continue;
4420 		}
4421 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4422 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4423 			continue;
4424 
4425 		btrfs_set_stack_device_generation(dev_item, 0);
4426 		btrfs_set_stack_device_type(dev_item, dev->type);
4427 		btrfs_set_stack_device_id(dev_item, dev->devid);
4428 		btrfs_set_stack_device_total_bytes(dev_item,
4429 						   dev->commit_total_bytes);
4430 		btrfs_set_stack_device_bytes_used(dev_item,
4431 						  dev->commit_bytes_used);
4432 		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4433 		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4434 		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4435 		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4436 		memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4437 		       BTRFS_FSID_SIZE);
4438 
4439 		flags = btrfs_super_flags(sb);
4440 		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4441 
4442 		ret = btrfs_validate_write_super(fs_info, sb);
4443 		if (ret < 0) {
4444 			mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4445 			btrfs_handle_fs_error(fs_info, -EUCLEAN,
4446 				"unexpected superblock corruption detected");
4447 			return -EUCLEAN;
4448 		}
4449 
4450 		ret = write_dev_supers(dev, sb, max_mirrors);
4451 		if (ret)
4452 			total_errors++;
4453 	}
4454 	if (total_errors > max_errors) {
4455 		btrfs_err(fs_info, "%d errors while writing supers",
4456 			  total_errors);
4457 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4458 
4459 		/* FUA is masked off if unsupported and can't be the reason */
4460 		btrfs_handle_fs_error(fs_info, -EIO,
4461 				      "%d errors while writing supers",
4462 				      total_errors);
4463 		return -EIO;
4464 	}
4465 
4466 	total_errors = 0;
4467 	list_for_each_entry(dev, head, dev_list) {
4468 		if (!dev->bdev)
4469 			continue;
4470 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4471 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4472 			continue;
4473 
4474 		ret = wait_dev_supers(dev, max_mirrors);
4475 		if (ret)
4476 			total_errors++;
4477 	}
4478 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4479 	if (total_errors > max_errors) {
4480 		btrfs_handle_fs_error(fs_info, -EIO,
4481 				      "%d errors while writing supers",
4482 				      total_errors);
4483 		return -EIO;
4484 	}
4485 	return 0;
4486 }
4487 
4488 /* Drop a fs root from the radix tree and free it. */
4489 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4490 				  struct btrfs_root *root)
4491 {
4492 	bool drop_ref = false;
4493 
4494 	spin_lock(&fs_info->fs_roots_radix_lock);
4495 	radix_tree_delete(&fs_info->fs_roots_radix,
4496 			  (unsigned long)root->root_key.objectid);
4497 	if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4498 		drop_ref = true;
4499 	spin_unlock(&fs_info->fs_roots_radix_lock);
4500 
4501 	if (BTRFS_FS_ERROR(fs_info)) {
4502 		ASSERT(root->log_root == NULL);
4503 		if (root->reloc_root) {
4504 			btrfs_put_root(root->reloc_root);
4505 			root->reloc_root = NULL;
4506 		}
4507 	}
4508 
4509 	if (drop_ref)
4510 		btrfs_put_root(root);
4511 }
4512 
4513 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4514 {
4515 	u64 root_objectid = 0;
4516 	struct btrfs_root *gang[8];
4517 	int i = 0;
4518 	int err = 0;
4519 	unsigned int ret = 0;
4520 
4521 	while (1) {
4522 		spin_lock(&fs_info->fs_roots_radix_lock);
4523 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4524 					     (void **)gang, root_objectid,
4525 					     ARRAY_SIZE(gang));
4526 		if (!ret) {
4527 			spin_unlock(&fs_info->fs_roots_radix_lock);
4528 			break;
4529 		}
4530 		root_objectid = gang[ret - 1]->root_key.objectid + 1;
4531 
4532 		for (i = 0; i < ret; i++) {
4533 			/* Avoid to grab roots in dead_roots */
4534 			if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4535 				gang[i] = NULL;
4536 				continue;
4537 			}
4538 			/* grab all the search result for later use */
4539 			gang[i] = btrfs_grab_root(gang[i]);
4540 		}
4541 		spin_unlock(&fs_info->fs_roots_radix_lock);
4542 
4543 		for (i = 0; i < ret; i++) {
4544 			if (!gang[i])
4545 				continue;
4546 			root_objectid = gang[i]->root_key.objectid;
4547 			err = btrfs_orphan_cleanup(gang[i]);
4548 			if (err)
4549 				break;
4550 			btrfs_put_root(gang[i]);
4551 		}
4552 		root_objectid++;
4553 	}
4554 
4555 	/* release the uncleaned roots due to error */
4556 	for (; i < ret; i++) {
4557 		if (gang[i])
4558 			btrfs_put_root(gang[i]);
4559 	}
4560 	return err;
4561 }
4562 
4563 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4564 {
4565 	struct btrfs_root *root = fs_info->tree_root;
4566 	struct btrfs_trans_handle *trans;
4567 
4568 	mutex_lock(&fs_info->cleaner_mutex);
4569 	btrfs_run_delayed_iputs(fs_info);
4570 	mutex_unlock(&fs_info->cleaner_mutex);
4571 	wake_up_process(fs_info->cleaner_kthread);
4572 
4573 	/* wait until ongoing cleanup work done */
4574 	down_write(&fs_info->cleanup_work_sem);
4575 	up_write(&fs_info->cleanup_work_sem);
4576 
4577 	trans = btrfs_join_transaction(root);
4578 	if (IS_ERR(trans))
4579 		return PTR_ERR(trans);
4580 	return btrfs_commit_transaction(trans);
4581 }
4582 
4583 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4584 {
4585 	struct btrfs_transaction *trans;
4586 	struct btrfs_transaction *tmp;
4587 	bool found = false;
4588 
4589 	if (list_empty(&fs_info->trans_list))
4590 		return;
4591 
4592 	/*
4593 	 * This function is only called at the very end of close_ctree(),
4594 	 * thus no other running transaction, no need to take trans_lock.
4595 	 */
4596 	ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4597 	list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4598 		struct extent_state *cached = NULL;
4599 		u64 dirty_bytes = 0;
4600 		u64 cur = 0;
4601 		u64 found_start;
4602 		u64 found_end;
4603 
4604 		found = true;
4605 		while (!find_first_extent_bit(&trans->dirty_pages, cur,
4606 			&found_start, &found_end, EXTENT_DIRTY, &cached)) {
4607 			dirty_bytes += found_end + 1 - found_start;
4608 			cur = found_end + 1;
4609 		}
4610 		btrfs_warn(fs_info,
4611 	"transaction %llu (with %llu dirty metadata bytes) is not committed",
4612 			   trans->transid, dirty_bytes);
4613 		btrfs_cleanup_one_transaction(trans, fs_info);
4614 
4615 		if (trans == fs_info->running_transaction)
4616 			fs_info->running_transaction = NULL;
4617 		list_del_init(&trans->list);
4618 
4619 		btrfs_put_transaction(trans);
4620 		trace_btrfs_transaction_commit(fs_info);
4621 	}
4622 	ASSERT(!found);
4623 }
4624 
4625 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4626 {
4627 	int ret;
4628 
4629 	set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4630 	/*
4631 	 * We don't want the cleaner to start new transactions, add more delayed
4632 	 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4633 	 * because that frees the task_struct, and the transaction kthread might
4634 	 * still try to wake up the cleaner.
4635 	 */
4636 	kthread_park(fs_info->cleaner_kthread);
4637 
4638 	/*
4639 	 * If we had UNFINISHED_DROPS we could still be processing them, so
4640 	 * clear that bit and wake up relocation so it can stop.
4641 	 */
4642 	btrfs_wake_unfinished_drop(fs_info);
4643 
4644 	/* wait for the qgroup rescan worker to stop */
4645 	btrfs_qgroup_wait_for_completion(fs_info, false);
4646 
4647 	/* wait for the uuid_scan task to finish */
4648 	down(&fs_info->uuid_tree_rescan_sem);
4649 	/* avoid complains from lockdep et al., set sem back to initial state */
4650 	up(&fs_info->uuid_tree_rescan_sem);
4651 
4652 	/* pause restriper - we want to resume on mount */
4653 	btrfs_pause_balance(fs_info);
4654 
4655 	btrfs_dev_replace_suspend_for_unmount(fs_info);
4656 
4657 	btrfs_scrub_cancel(fs_info);
4658 
4659 	/* wait for any defraggers to finish */
4660 	wait_event(fs_info->transaction_wait,
4661 		   (atomic_read(&fs_info->defrag_running) == 0));
4662 
4663 	/* clear out the rbtree of defraggable inodes */
4664 	btrfs_cleanup_defrag_inodes(fs_info);
4665 
4666 	cancel_work_sync(&fs_info->async_reclaim_work);
4667 	cancel_work_sync(&fs_info->async_data_reclaim_work);
4668 	cancel_work_sync(&fs_info->preempt_reclaim_work);
4669 
4670 	cancel_work_sync(&fs_info->reclaim_bgs_work);
4671 
4672 	/* Cancel or finish ongoing discard work */
4673 	btrfs_discard_cleanup(fs_info);
4674 
4675 	if (!sb_rdonly(fs_info->sb)) {
4676 		/*
4677 		 * The cleaner kthread is stopped, so do one final pass over
4678 		 * unused block groups.
4679 		 */
4680 		btrfs_delete_unused_bgs(fs_info);
4681 
4682 		/*
4683 		 * There might be existing delayed inode workers still running
4684 		 * and holding an empty delayed inode item. We must wait for
4685 		 * them to complete first because they can create a transaction.
4686 		 * This happens when someone calls btrfs_balance_delayed_items()
4687 		 * and then a transaction commit runs the same delayed nodes
4688 		 * before any delayed worker has done something with the nodes.
4689 		 * We must wait for any worker here and not at transaction
4690 		 * commit time since that could cause a deadlock.
4691 		 * This is a very rare case.
4692 		 */
4693 		btrfs_flush_workqueue(fs_info->delayed_workers);
4694 
4695 		ret = btrfs_commit_super(fs_info);
4696 		if (ret)
4697 			btrfs_err(fs_info, "commit super ret %d", ret);
4698 	}
4699 
4700 	if (BTRFS_FS_ERROR(fs_info))
4701 		btrfs_error_commit_super(fs_info);
4702 
4703 	kthread_stop(fs_info->transaction_kthread);
4704 	kthread_stop(fs_info->cleaner_kthread);
4705 
4706 	ASSERT(list_empty(&fs_info->delayed_iputs));
4707 	set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4708 
4709 	if (btrfs_check_quota_leak(fs_info)) {
4710 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4711 		btrfs_err(fs_info, "qgroup reserved space leaked");
4712 	}
4713 
4714 	btrfs_free_qgroup_config(fs_info);
4715 	ASSERT(list_empty(&fs_info->delalloc_roots));
4716 
4717 	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4718 		btrfs_info(fs_info, "at unmount delalloc count %lld",
4719 		       percpu_counter_sum(&fs_info->delalloc_bytes));
4720 	}
4721 
4722 	if (percpu_counter_sum(&fs_info->ordered_bytes))
4723 		btrfs_info(fs_info, "at unmount dio bytes count %lld",
4724 			   percpu_counter_sum(&fs_info->ordered_bytes));
4725 
4726 	btrfs_sysfs_remove_mounted(fs_info);
4727 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4728 
4729 	btrfs_put_block_group_cache(fs_info);
4730 
4731 	/*
4732 	 * we must make sure there is not any read request to
4733 	 * submit after we stopping all workers.
4734 	 */
4735 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4736 	btrfs_stop_all_workers(fs_info);
4737 
4738 	/* We shouldn't have any transaction open at this point */
4739 	warn_about_uncommitted_trans(fs_info);
4740 
4741 	clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4742 	free_root_pointers(fs_info, true);
4743 	btrfs_free_fs_roots(fs_info);
4744 
4745 	/*
4746 	 * We must free the block groups after dropping the fs_roots as we could
4747 	 * have had an IO error and have left over tree log blocks that aren't
4748 	 * cleaned up until the fs roots are freed.  This makes the block group
4749 	 * accounting appear to be wrong because there's pending reserved bytes,
4750 	 * so make sure we do the block group cleanup afterwards.
4751 	 */
4752 	btrfs_free_block_groups(fs_info);
4753 
4754 	iput(fs_info->btree_inode);
4755 
4756 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4757 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4758 		btrfsic_unmount(fs_info->fs_devices);
4759 #endif
4760 
4761 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
4762 	btrfs_close_devices(fs_info->fs_devices);
4763 }
4764 
4765 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4766 			  int atomic)
4767 {
4768 	int ret;
4769 	struct inode *btree_inode = buf->pages[0]->mapping->host;
4770 
4771 	ret = extent_buffer_uptodate(buf);
4772 	if (!ret)
4773 		return ret;
4774 
4775 	ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4776 				    parent_transid, atomic);
4777 	if (ret == -EAGAIN)
4778 		return ret;
4779 	return !ret;
4780 }
4781 
4782 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4783 {
4784 	struct btrfs_fs_info *fs_info = buf->fs_info;
4785 	u64 transid = btrfs_header_generation(buf);
4786 	int was_dirty;
4787 
4788 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4789 	/*
4790 	 * This is a fast path so only do this check if we have sanity tests
4791 	 * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4792 	 * outside of the sanity tests.
4793 	 */
4794 	if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4795 		return;
4796 #endif
4797 	btrfs_assert_tree_write_locked(buf);
4798 	if (transid != fs_info->generation)
4799 		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4800 			buf->start, transid, fs_info->generation);
4801 	was_dirty = set_extent_buffer_dirty(buf);
4802 	if (!was_dirty)
4803 		percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4804 					 buf->len,
4805 					 fs_info->dirty_metadata_batch);
4806 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4807 	/*
4808 	 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4809 	 * but item data not updated.
4810 	 * So here we should only check item pointers, not item data.
4811 	 */
4812 	if (btrfs_header_level(buf) == 0 &&
4813 	    btrfs_check_leaf_relaxed(buf)) {
4814 		btrfs_print_leaf(buf);
4815 		ASSERT(0);
4816 	}
4817 #endif
4818 }
4819 
4820 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4821 					int flush_delayed)
4822 {
4823 	/*
4824 	 * looks as though older kernels can get into trouble with
4825 	 * this code, they end up stuck in balance_dirty_pages forever
4826 	 */
4827 	int ret;
4828 
4829 	if (current->flags & PF_MEMALLOC)
4830 		return;
4831 
4832 	if (flush_delayed)
4833 		btrfs_balance_delayed_items(fs_info);
4834 
4835 	ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4836 				     BTRFS_DIRTY_METADATA_THRESH,
4837 				     fs_info->dirty_metadata_batch);
4838 	if (ret > 0) {
4839 		balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4840 	}
4841 }
4842 
4843 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4844 {
4845 	__btrfs_btree_balance_dirty(fs_info, 1);
4846 }
4847 
4848 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4849 {
4850 	__btrfs_btree_balance_dirty(fs_info, 0);
4851 }
4852 
4853 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4854 		      struct btrfs_key *first_key)
4855 {
4856 	return btree_read_extent_buffer_pages(buf, parent_transid,
4857 					      level, first_key);
4858 }
4859 
4860 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4861 {
4862 	/* cleanup FS via transaction */
4863 	btrfs_cleanup_transaction(fs_info);
4864 
4865 	mutex_lock(&fs_info->cleaner_mutex);
4866 	btrfs_run_delayed_iputs(fs_info);
4867 	mutex_unlock(&fs_info->cleaner_mutex);
4868 
4869 	down_write(&fs_info->cleanup_work_sem);
4870 	up_write(&fs_info->cleanup_work_sem);
4871 }
4872 
4873 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4874 {
4875 	struct btrfs_root *gang[8];
4876 	u64 root_objectid = 0;
4877 	int ret;
4878 
4879 	spin_lock(&fs_info->fs_roots_radix_lock);
4880 	while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4881 					     (void **)gang, root_objectid,
4882 					     ARRAY_SIZE(gang))) != 0) {
4883 		int i;
4884 
4885 		for (i = 0; i < ret; i++)
4886 			gang[i] = btrfs_grab_root(gang[i]);
4887 		spin_unlock(&fs_info->fs_roots_radix_lock);
4888 
4889 		for (i = 0; i < ret; i++) {
4890 			if (!gang[i])
4891 				continue;
4892 			root_objectid = gang[i]->root_key.objectid;
4893 			btrfs_free_log(NULL, gang[i]);
4894 			btrfs_put_root(gang[i]);
4895 		}
4896 		root_objectid++;
4897 		spin_lock(&fs_info->fs_roots_radix_lock);
4898 	}
4899 	spin_unlock(&fs_info->fs_roots_radix_lock);
4900 	btrfs_free_log_root_tree(NULL, fs_info);
4901 }
4902 
4903 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4904 {
4905 	struct btrfs_ordered_extent *ordered;
4906 
4907 	spin_lock(&root->ordered_extent_lock);
4908 	/*
4909 	 * This will just short circuit the ordered completion stuff which will
4910 	 * make sure the ordered extent gets properly cleaned up.
4911 	 */
4912 	list_for_each_entry(ordered, &root->ordered_extents,
4913 			    root_extent_list)
4914 		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4915 	spin_unlock(&root->ordered_extent_lock);
4916 }
4917 
4918 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4919 {
4920 	struct btrfs_root *root;
4921 	struct list_head splice;
4922 
4923 	INIT_LIST_HEAD(&splice);
4924 
4925 	spin_lock(&fs_info->ordered_root_lock);
4926 	list_splice_init(&fs_info->ordered_roots, &splice);
4927 	while (!list_empty(&splice)) {
4928 		root = list_first_entry(&splice, struct btrfs_root,
4929 					ordered_root);
4930 		list_move_tail(&root->ordered_root,
4931 			       &fs_info->ordered_roots);
4932 
4933 		spin_unlock(&fs_info->ordered_root_lock);
4934 		btrfs_destroy_ordered_extents(root);
4935 
4936 		cond_resched();
4937 		spin_lock(&fs_info->ordered_root_lock);
4938 	}
4939 	spin_unlock(&fs_info->ordered_root_lock);
4940 
4941 	/*
4942 	 * We need this here because if we've been flipped read-only we won't
4943 	 * get sync() from the umount, so we need to make sure any ordered
4944 	 * extents that haven't had their dirty pages IO start writeout yet
4945 	 * actually get run and error out properly.
4946 	 */
4947 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4948 }
4949 
4950 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4951 				      struct btrfs_fs_info *fs_info)
4952 {
4953 	struct rb_node *node;
4954 	struct btrfs_delayed_ref_root *delayed_refs;
4955 	struct btrfs_delayed_ref_node *ref;
4956 	int ret = 0;
4957 
4958 	delayed_refs = &trans->delayed_refs;
4959 
4960 	spin_lock(&delayed_refs->lock);
4961 	if (atomic_read(&delayed_refs->num_entries) == 0) {
4962 		spin_unlock(&delayed_refs->lock);
4963 		btrfs_debug(fs_info, "delayed_refs has NO entry");
4964 		return ret;
4965 	}
4966 
4967 	while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4968 		struct btrfs_delayed_ref_head *head;
4969 		struct rb_node *n;
4970 		bool pin_bytes = false;
4971 
4972 		head = rb_entry(node, struct btrfs_delayed_ref_head,
4973 				href_node);
4974 		if (btrfs_delayed_ref_lock(delayed_refs, head))
4975 			continue;
4976 
4977 		spin_lock(&head->lock);
4978 		while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4979 			ref = rb_entry(n, struct btrfs_delayed_ref_node,
4980 				       ref_node);
4981 			ref->in_tree = 0;
4982 			rb_erase_cached(&ref->ref_node, &head->ref_tree);
4983 			RB_CLEAR_NODE(&ref->ref_node);
4984 			if (!list_empty(&ref->add_list))
4985 				list_del(&ref->add_list);
4986 			atomic_dec(&delayed_refs->num_entries);
4987 			btrfs_put_delayed_ref(ref);
4988 		}
4989 		if (head->must_insert_reserved)
4990 			pin_bytes = true;
4991 		btrfs_free_delayed_extent_op(head->extent_op);
4992 		btrfs_delete_ref_head(delayed_refs, head);
4993 		spin_unlock(&head->lock);
4994 		spin_unlock(&delayed_refs->lock);
4995 		mutex_unlock(&head->mutex);
4996 
4997 		if (pin_bytes) {
4998 			struct btrfs_block_group *cache;
4999 
5000 			cache = btrfs_lookup_block_group(fs_info, head->bytenr);
5001 			BUG_ON(!cache);
5002 
5003 			spin_lock(&cache->space_info->lock);
5004 			spin_lock(&cache->lock);
5005 			cache->pinned += head->num_bytes;
5006 			btrfs_space_info_update_bytes_pinned(fs_info,
5007 				cache->space_info, head->num_bytes);
5008 			cache->reserved -= head->num_bytes;
5009 			cache->space_info->bytes_reserved -= head->num_bytes;
5010 			spin_unlock(&cache->lock);
5011 			spin_unlock(&cache->space_info->lock);
5012 
5013 			btrfs_put_block_group(cache);
5014 
5015 			btrfs_error_unpin_extent_range(fs_info, head->bytenr,
5016 				head->bytenr + head->num_bytes - 1);
5017 		}
5018 		btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
5019 		btrfs_put_delayed_ref_head(head);
5020 		cond_resched();
5021 		spin_lock(&delayed_refs->lock);
5022 	}
5023 	btrfs_qgroup_destroy_extent_records(trans);
5024 
5025 	spin_unlock(&delayed_refs->lock);
5026 
5027 	return ret;
5028 }
5029 
5030 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
5031 {
5032 	struct btrfs_inode *btrfs_inode;
5033 	struct list_head splice;
5034 
5035 	INIT_LIST_HEAD(&splice);
5036 
5037 	spin_lock(&root->delalloc_lock);
5038 	list_splice_init(&root->delalloc_inodes, &splice);
5039 
5040 	while (!list_empty(&splice)) {
5041 		struct inode *inode = NULL;
5042 		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
5043 					       delalloc_inodes);
5044 		__btrfs_del_delalloc_inode(root, btrfs_inode);
5045 		spin_unlock(&root->delalloc_lock);
5046 
5047 		/*
5048 		 * Make sure we get a live inode and that it'll not disappear
5049 		 * meanwhile.
5050 		 */
5051 		inode = igrab(&btrfs_inode->vfs_inode);
5052 		if (inode) {
5053 			invalidate_inode_pages2(inode->i_mapping);
5054 			iput(inode);
5055 		}
5056 		spin_lock(&root->delalloc_lock);
5057 	}
5058 	spin_unlock(&root->delalloc_lock);
5059 }
5060 
5061 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
5062 {
5063 	struct btrfs_root *root;
5064 	struct list_head splice;
5065 
5066 	INIT_LIST_HEAD(&splice);
5067 
5068 	spin_lock(&fs_info->delalloc_root_lock);
5069 	list_splice_init(&fs_info->delalloc_roots, &splice);
5070 	while (!list_empty(&splice)) {
5071 		root = list_first_entry(&splice, struct btrfs_root,
5072 					 delalloc_root);
5073 		root = btrfs_grab_root(root);
5074 		BUG_ON(!root);
5075 		spin_unlock(&fs_info->delalloc_root_lock);
5076 
5077 		btrfs_destroy_delalloc_inodes(root);
5078 		btrfs_put_root(root);
5079 
5080 		spin_lock(&fs_info->delalloc_root_lock);
5081 	}
5082 	spin_unlock(&fs_info->delalloc_root_lock);
5083 }
5084 
5085 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
5086 					struct extent_io_tree *dirty_pages,
5087 					int mark)
5088 {
5089 	int ret;
5090 	struct extent_buffer *eb;
5091 	u64 start = 0;
5092 	u64 end;
5093 
5094 	while (1) {
5095 		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
5096 					    mark, NULL);
5097 		if (ret)
5098 			break;
5099 
5100 		clear_extent_bits(dirty_pages, start, end, mark);
5101 		while (start <= end) {
5102 			eb = find_extent_buffer(fs_info, start);
5103 			start += fs_info->nodesize;
5104 			if (!eb)
5105 				continue;
5106 			wait_on_extent_buffer_writeback(eb);
5107 
5108 			if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
5109 					       &eb->bflags))
5110 				clear_extent_buffer_dirty(eb);
5111 			free_extent_buffer_stale(eb);
5112 		}
5113 	}
5114 
5115 	return ret;
5116 }
5117 
5118 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
5119 				       struct extent_io_tree *unpin)
5120 {
5121 	u64 start;
5122 	u64 end;
5123 	int ret;
5124 
5125 	while (1) {
5126 		struct extent_state *cached_state = NULL;
5127 
5128 		/*
5129 		 * The btrfs_finish_extent_commit() may get the same range as
5130 		 * ours between find_first_extent_bit and clear_extent_dirty.
5131 		 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
5132 		 * the same extent range.
5133 		 */
5134 		mutex_lock(&fs_info->unused_bg_unpin_mutex);
5135 		ret = find_first_extent_bit(unpin, 0, &start, &end,
5136 					    EXTENT_DIRTY, &cached_state);
5137 		if (ret) {
5138 			mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5139 			break;
5140 		}
5141 
5142 		clear_extent_dirty(unpin, start, end, &cached_state);
5143 		free_extent_state(cached_state);
5144 		btrfs_error_unpin_extent_range(fs_info, start, end);
5145 		mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5146 		cond_resched();
5147 	}
5148 
5149 	return 0;
5150 }
5151 
5152 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
5153 {
5154 	struct inode *inode;
5155 
5156 	inode = cache->io_ctl.inode;
5157 	if (inode) {
5158 		invalidate_inode_pages2(inode->i_mapping);
5159 		BTRFS_I(inode)->generation = 0;
5160 		cache->io_ctl.inode = NULL;
5161 		iput(inode);
5162 	}
5163 	ASSERT(cache->io_ctl.pages == NULL);
5164 	btrfs_put_block_group(cache);
5165 }
5166 
5167 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
5168 			     struct btrfs_fs_info *fs_info)
5169 {
5170 	struct btrfs_block_group *cache;
5171 
5172 	spin_lock(&cur_trans->dirty_bgs_lock);
5173 	while (!list_empty(&cur_trans->dirty_bgs)) {
5174 		cache = list_first_entry(&cur_trans->dirty_bgs,
5175 					 struct btrfs_block_group,
5176 					 dirty_list);
5177 
5178 		if (!list_empty(&cache->io_list)) {
5179 			spin_unlock(&cur_trans->dirty_bgs_lock);
5180 			list_del_init(&cache->io_list);
5181 			btrfs_cleanup_bg_io(cache);
5182 			spin_lock(&cur_trans->dirty_bgs_lock);
5183 		}
5184 
5185 		list_del_init(&cache->dirty_list);
5186 		spin_lock(&cache->lock);
5187 		cache->disk_cache_state = BTRFS_DC_ERROR;
5188 		spin_unlock(&cache->lock);
5189 
5190 		spin_unlock(&cur_trans->dirty_bgs_lock);
5191 		btrfs_put_block_group(cache);
5192 		btrfs_delayed_refs_rsv_release(fs_info, 1);
5193 		spin_lock(&cur_trans->dirty_bgs_lock);
5194 	}
5195 	spin_unlock(&cur_trans->dirty_bgs_lock);
5196 
5197 	/*
5198 	 * Refer to the definition of io_bgs member for details why it's safe
5199 	 * to use it without any locking
5200 	 */
5201 	while (!list_empty(&cur_trans->io_bgs)) {
5202 		cache = list_first_entry(&cur_trans->io_bgs,
5203 					 struct btrfs_block_group,
5204 					 io_list);
5205 
5206 		list_del_init(&cache->io_list);
5207 		spin_lock(&cache->lock);
5208 		cache->disk_cache_state = BTRFS_DC_ERROR;
5209 		spin_unlock(&cache->lock);
5210 		btrfs_cleanup_bg_io(cache);
5211 	}
5212 }
5213 
5214 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
5215 				   struct btrfs_fs_info *fs_info)
5216 {
5217 	struct btrfs_device *dev, *tmp;
5218 
5219 	btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
5220 	ASSERT(list_empty(&cur_trans->dirty_bgs));
5221 	ASSERT(list_empty(&cur_trans->io_bgs));
5222 
5223 	list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
5224 				 post_commit_list) {
5225 		list_del_init(&dev->post_commit_list);
5226 	}
5227 
5228 	btrfs_destroy_delayed_refs(cur_trans, fs_info);
5229 
5230 	cur_trans->state = TRANS_STATE_COMMIT_START;
5231 	wake_up(&fs_info->transaction_blocked_wait);
5232 
5233 	cur_trans->state = TRANS_STATE_UNBLOCKED;
5234 	wake_up(&fs_info->transaction_wait);
5235 
5236 	btrfs_destroy_delayed_inodes(fs_info);
5237 
5238 	btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
5239 				     EXTENT_DIRTY);
5240 	btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
5241 
5242 	btrfs_free_redirty_list(cur_trans);
5243 
5244 	cur_trans->state =TRANS_STATE_COMPLETED;
5245 	wake_up(&cur_trans->commit_wait);
5246 }
5247 
5248 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
5249 {
5250 	struct btrfs_transaction *t;
5251 
5252 	mutex_lock(&fs_info->transaction_kthread_mutex);
5253 
5254 	spin_lock(&fs_info->trans_lock);
5255 	while (!list_empty(&fs_info->trans_list)) {
5256 		t = list_first_entry(&fs_info->trans_list,
5257 				     struct btrfs_transaction, list);
5258 		if (t->state >= TRANS_STATE_COMMIT_START) {
5259 			refcount_inc(&t->use_count);
5260 			spin_unlock(&fs_info->trans_lock);
5261 			btrfs_wait_for_commit(fs_info, t->transid);
5262 			btrfs_put_transaction(t);
5263 			spin_lock(&fs_info->trans_lock);
5264 			continue;
5265 		}
5266 		if (t == fs_info->running_transaction) {
5267 			t->state = TRANS_STATE_COMMIT_DOING;
5268 			spin_unlock(&fs_info->trans_lock);
5269 			/*
5270 			 * We wait for 0 num_writers since we don't hold a trans
5271 			 * handle open currently for this transaction.
5272 			 */
5273 			wait_event(t->writer_wait,
5274 				   atomic_read(&t->num_writers) == 0);
5275 		} else {
5276 			spin_unlock(&fs_info->trans_lock);
5277 		}
5278 		btrfs_cleanup_one_transaction(t, fs_info);
5279 
5280 		spin_lock(&fs_info->trans_lock);
5281 		if (t == fs_info->running_transaction)
5282 			fs_info->running_transaction = NULL;
5283 		list_del_init(&t->list);
5284 		spin_unlock(&fs_info->trans_lock);
5285 
5286 		btrfs_put_transaction(t);
5287 		trace_btrfs_transaction_commit(fs_info);
5288 		spin_lock(&fs_info->trans_lock);
5289 	}
5290 	spin_unlock(&fs_info->trans_lock);
5291 	btrfs_destroy_all_ordered_extents(fs_info);
5292 	btrfs_destroy_delayed_inodes(fs_info);
5293 	btrfs_assert_delayed_root_empty(fs_info);
5294 	btrfs_destroy_all_delalloc_inodes(fs_info);
5295 	btrfs_drop_all_logs(fs_info);
5296 	mutex_unlock(&fs_info->transaction_kthread_mutex);
5297 
5298 	return 0;
5299 }
5300 
5301 int btrfs_init_root_free_objectid(struct btrfs_root *root)
5302 {
5303 	struct btrfs_path *path;
5304 	int ret;
5305 	struct extent_buffer *l;
5306 	struct btrfs_key search_key;
5307 	struct btrfs_key found_key;
5308 	int slot;
5309 
5310 	path = btrfs_alloc_path();
5311 	if (!path)
5312 		return -ENOMEM;
5313 
5314 	search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
5315 	search_key.type = -1;
5316 	search_key.offset = (u64)-1;
5317 	ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5318 	if (ret < 0)
5319 		goto error;
5320 	BUG_ON(ret == 0); /* Corruption */
5321 	if (path->slots[0] > 0) {
5322 		slot = path->slots[0] - 1;
5323 		l = path->nodes[0];
5324 		btrfs_item_key_to_cpu(l, &found_key, slot);
5325 		root->free_objectid = max_t(u64, found_key.objectid + 1,
5326 					    BTRFS_FIRST_FREE_OBJECTID);
5327 	} else {
5328 		root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5329 	}
5330 	ret = 0;
5331 error:
5332 	btrfs_free_path(path);
5333 	return ret;
5334 }
5335 
5336 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5337 {
5338 	int ret;
5339 	mutex_lock(&root->objectid_mutex);
5340 
5341 	if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5342 		btrfs_warn(root->fs_info,
5343 			   "the objectid of root %llu reaches its highest value",
5344 			   root->root_key.objectid);
5345 		ret = -ENOSPC;
5346 		goto out;
5347 	}
5348 
5349 	*objectid = root->free_objectid++;
5350 	ret = 0;
5351 out:
5352 	mutex_unlock(&root->objectid_mutex);
5353 	return ret;
5354 }
5355