xref: /linux/fs/btrfs/disk-io.c (revision 55ec81f7517fad09135f65552cea0a3ee84fff30)
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 "bio.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 #include "fs.h"
47 #include "accessors.h"
48 #include "extent-tree.h"
49 #include "root-tree.h"
50 #include "defrag.h"
51 #include "uuid-tree.h"
52 #include "relocation.h"
53 #include "scrub.h"
54 #include "super.h"
55 
56 #define BTRFS_SUPER_FLAG_SUPP	(BTRFS_HEADER_FLAG_WRITTEN |\
57 				 BTRFS_HEADER_FLAG_RELOC |\
58 				 BTRFS_SUPER_FLAG_ERROR |\
59 				 BTRFS_SUPER_FLAG_SEEDING |\
60 				 BTRFS_SUPER_FLAG_METADUMP |\
61 				 BTRFS_SUPER_FLAG_METADUMP_V2)
62 
63 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
64 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
65 
66 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
67 {
68 	if (fs_info->csum_shash)
69 		crypto_free_shash(fs_info->csum_shash);
70 }
71 
72 /*
73  * Compute the csum of a btree block and store the result to provided buffer.
74  */
75 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
76 {
77 	struct btrfs_fs_info *fs_info = buf->fs_info;
78 	const int num_pages = num_extent_pages(buf);
79 	const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
80 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
81 	char *kaddr;
82 	int i;
83 
84 	shash->tfm = fs_info->csum_shash;
85 	crypto_shash_init(shash);
86 	kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
87 	crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
88 			    first_page_part - BTRFS_CSUM_SIZE);
89 
90 	for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
91 		kaddr = page_address(buf->pages[i]);
92 		crypto_shash_update(shash, kaddr, PAGE_SIZE);
93 	}
94 	memset(result, 0, BTRFS_CSUM_SIZE);
95 	crypto_shash_final(shash, result);
96 }
97 
98 /*
99  * we can't consider a given block up to date unless the transid of the
100  * block matches the transid in the parent node's pointer.  This is how we
101  * detect blocks that either didn't get written at all or got written
102  * in the wrong place.
103  */
104 int btrfs_buffer_uptodate(struct extent_buffer *eb, u64 parent_transid, int atomic)
105 {
106 	if (!extent_buffer_uptodate(eb))
107 		return 0;
108 
109 	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
110 		return 1;
111 
112 	if (atomic)
113 		return -EAGAIN;
114 
115 	if (!extent_buffer_uptodate(eb) ||
116 	    btrfs_header_generation(eb) != parent_transid) {
117 		btrfs_err_rl(eb->fs_info,
118 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
119 			eb->start, eb->read_mirror,
120 			parent_transid, btrfs_header_generation(eb));
121 		clear_extent_buffer_uptodate(eb);
122 		return 0;
123 	}
124 	return 1;
125 }
126 
127 static bool btrfs_supported_super_csum(u16 csum_type)
128 {
129 	switch (csum_type) {
130 	case BTRFS_CSUM_TYPE_CRC32:
131 	case BTRFS_CSUM_TYPE_XXHASH:
132 	case BTRFS_CSUM_TYPE_SHA256:
133 	case BTRFS_CSUM_TYPE_BLAKE2:
134 		return true;
135 	default:
136 		return false;
137 	}
138 }
139 
140 /*
141  * Return 0 if the superblock checksum type matches the checksum value of that
142  * algorithm. Pass the raw disk superblock data.
143  */
144 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
145 			   const struct btrfs_super_block *disk_sb)
146 {
147 	char result[BTRFS_CSUM_SIZE];
148 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
149 
150 	shash->tfm = fs_info->csum_shash;
151 
152 	/*
153 	 * The super_block structure does not span the whole
154 	 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
155 	 * filled with zeros and is included in the checksum.
156 	 */
157 	crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
158 			    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
159 
160 	if (memcmp(disk_sb->csum, result, fs_info->csum_size))
161 		return 1;
162 
163 	return 0;
164 }
165 
166 static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
167 				      int mirror_num)
168 {
169 	struct btrfs_fs_info *fs_info = eb->fs_info;
170 	int i, num_pages = num_extent_pages(eb);
171 	int ret = 0;
172 
173 	if (sb_rdonly(fs_info->sb))
174 		return -EROFS;
175 
176 	for (i = 0; i < num_pages; i++) {
177 		struct page *p = eb->pages[i];
178 		u64 start = max_t(u64, eb->start, page_offset(p));
179 		u64 end = min_t(u64, eb->start + eb->len, page_offset(p) + PAGE_SIZE);
180 		u32 len = end - start;
181 
182 		ret = btrfs_repair_io_failure(fs_info, 0, start, len,
183 				start, p, offset_in_page(start), mirror_num);
184 		if (ret)
185 			break;
186 	}
187 
188 	return ret;
189 }
190 
191 /*
192  * helper to read a given tree block, doing retries as required when
193  * the checksums don't match and we have alternate mirrors to try.
194  *
195  * @check:		expected tree parentness check, see the comments of the
196  *			structure for details.
197  */
198 int btrfs_read_extent_buffer(struct extent_buffer *eb,
199 			     struct btrfs_tree_parent_check *check)
200 {
201 	struct btrfs_fs_info *fs_info = eb->fs_info;
202 	int failed = 0;
203 	int ret;
204 	int num_copies = 0;
205 	int mirror_num = 0;
206 	int failed_mirror = 0;
207 
208 	ASSERT(check);
209 
210 	while (1) {
211 		clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
212 		ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
213 		if (!ret)
214 			break;
215 
216 		num_copies = btrfs_num_copies(fs_info,
217 					      eb->start, eb->len);
218 		if (num_copies == 1)
219 			break;
220 
221 		if (!failed_mirror) {
222 			failed = 1;
223 			failed_mirror = eb->read_mirror;
224 		}
225 
226 		mirror_num++;
227 		if (mirror_num == failed_mirror)
228 			mirror_num++;
229 
230 		if (mirror_num > num_copies)
231 			break;
232 	}
233 
234 	if (failed && !ret && failed_mirror)
235 		btrfs_repair_eb_io_failure(eb, failed_mirror);
236 
237 	return ret;
238 }
239 
240 /*
241  * Checksum a dirty tree block before IO.
242  */
243 blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
244 {
245 	struct extent_buffer *eb = bbio->private;
246 	struct btrfs_fs_info *fs_info = eb->fs_info;
247 	u64 found_start = btrfs_header_bytenr(eb);
248 	u8 result[BTRFS_CSUM_SIZE];
249 	int ret;
250 
251 	/* Btree blocks are always contiguous on disk. */
252 	if (WARN_ON_ONCE(bbio->file_offset != eb->start))
253 		return BLK_STS_IOERR;
254 	if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len))
255 		return BLK_STS_IOERR;
256 
257 	if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
258 		WARN_ON_ONCE(found_start != 0);
259 		return BLK_STS_OK;
260 	}
261 
262 	if (WARN_ON_ONCE(found_start != eb->start))
263 		return BLK_STS_IOERR;
264 	if (WARN_ON(!btrfs_page_test_uptodate(fs_info, eb->pages[0], eb->start,
265 					      eb->len)))
266 		return BLK_STS_IOERR;
267 
268 	ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
269 				    offsetof(struct btrfs_header, fsid),
270 				    BTRFS_FSID_SIZE) == 0);
271 	csum_tree_block(eb, result);
272 
273 	if (btrfs_header_level(eb))
274 		ret = btrfs_check_node(eb);
275 	else
276 		ret = btrfs_check_leaf(eb);
277 
278 	if (ret < 0)
279 		goto error;
280 
281 	/*
282 	 * Also check the generation, the eb reached here must be newer than
283 	 * last committed. Or something seriously wrong happened.
284 	 */
285 	if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) {
286 		ret = -EUCLEAN;
287 		btrfs_err(fs_info,
288 			"block=%llu bad generation, have %llu expect > %llu",
289 			  eb->start, btrfs_header_generation(eb),
290 			  fs_info->last_trans_committed);
291 		goto error;
292 	}
293 	write_extent_buffer(eb, result, 0, fs_info->csum_size);
294 	return BLK_STS_OK;
295 
296 error:
297 	btrfs_print_tree(eb, 0);
298 	btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
299 		  eb->start);
300 	/*
301 	 * Be noisy if this is an extent buffer from a log tree. We don't abort
302 	 * a transaction in case there's a bad log tree extent buffer, we just
303 	 * fallback to a transaction commit. Still we want to know when there is
304 	 * a bad log tree extent buffer, as that may signal a bug somewhere.
305 	 */
306 	WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
307 		btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
308 	return errno_to_blk_status(ret);
309 }
310 
311 static bool check_tree_block_fsid(struct extent_buffer *eb)
312 {
313 	struct btrfs_fs_info *fs_info = eb->fs_info;
314 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
315 	u8 fsid[BTRFS_FSID_SIZE];
316 
317 	read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
318 			   BTRFS_FSID_SIZE);
319 
320 	/*
321 	 * alloc_fs_devices() copies the fsid into metadata_uuid if the
322 	 * metadata_uuid is unset in the superblock, including for a seed device.
323 	 * So, we can use fs_devices->metadata_uuid.
324 	 */
325 	if (memcmp(fsid, fs_info->fs_devices->metadata_uuid, BTRFS_FSID_SIZE) == 0)
326 		return false;
327 
328 	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
329 		if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
330 			return false;
331 
332 	return true;
333 }
334 
335 /* Do basic extent buffer checks at read time */
336 int btrfs_validate_extent_buffer(struct extent_buffer *eb,
337 				 struct btrfs_tree_parent_check *check)
338 {
339 	struct btrfs_fs_info *fs_info = eb->fs_info;
340 	u64 found_start;
341 	const u32 csum_size = fs_info->csum_size;
342 	u8 found_level;
343 	u8 result[BTRFS_CSUM_SIZE];
344 	const u8 *header_csum;
345 	int ret = 0;
346 
347 	ASSERT(check);
348 
349 	found_start = btrfs_header_bytenr(eb);
350 	if (found_start != eb->start) {
351 		btrfs_err_rl(fs_info,
352 			"bad tree block start, mirror %u want %llu have %llu",
353 			     eb->read_mirror, eb->start, found_start);
354 		ret = -EIO;
355 		goto out;
356 	}
357 	if (check_tree_block_fsid(eb)) {
358 		btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
359 			     eb->start, eb->read_mirror);
360 		ret = -EIO;
361 		goto out;
362 	}
363 	found_level = btrfs_header_level(eb);
364 	if (found_level >= BTRFS_MAX_LEVEL) {
365 		btrfs_err(fs_info,
366 			"bad tree block level, mirror %u level %d on logical %llu",
367 			eb->read_mirror, btrfs_header_level(eb), eb->start);
368 		ret = -EIO;
369 		goto out;
370 	}
371 
372 	csum_tree_block(eb, result);
373 	header_csum = page_address(eb->pages[0]) +
374 		get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
375 
376 	if (memcmp(result, header_csum, csum_size) != 0) {
377 		btrfs_warn_rl(fs_info,
378 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
379 			      eb->start, eb->read_mirror,
380 			      CSUM_FMT_VALUE(csum_size, header_csum),
381 			      CSUM_FMT_VALUE(csum_size, result),
382 			      btrfs_header_level(eb));
383 		ret = -EUCLEAN;
384 		goto out;
385 	}
386 
387 	if (found_level != check->level) {
388 		btrfs_err(fs_info,
389 		"level verify failed on logical %llu mirror %u wanted %u found %u",
390 			  eb->start, eb->read_mirror, check->level, found_level);
391 		ret = -EIO;
392 		goto out;
393 	}
394 	if (unlikely(check->transid &&
395 		     btrfs_header_generation(eb) != check->transid)) {
396 		btrfs_err_rl(eb->fs_info,
397 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
398 				eb->start, eb->read_mirror, check->transid,
399 				btrfs_header_generation(eb));
400 		ret = -EIO;
401 		goto out;
402 	}
403 	if (check->has_first_key) {
404 		struct btrfs_key *expect_key = &check->first_key;
405 		struct btrfs_key found_key;
406 
407 		if (found_level)
408 			btrfs_node_key_to_cpu(eb, &found_key, 0);
409 		else
410 			btrfs_item_key_to_cpu(eb, &found_key, 0);
411 		if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
412 			btrfs_err(fs_info,
413 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
414 				  eb->start, check->transid,
415 				  expect_key->objectid,
416 				  expect_key->type, expect_key->offset,
417 				  found_key.objectid, found_key.type,
418 				  found_key.offset);
419 			ret = -EUCLEAN;
420 			goto out;
421 		}
422 	}
423 	if (check->owner_root) {
424 		ret = btrfs_check_eb_owner(eb, check->owner_root);
425 		if (ret < 0)
426 			goto out;
427 	}
428 
429 	/*
430 	 * If this is a leaf block and it is corrupt, set the corrupt bit so
431 	 * that we don't try and read the other copies of this block, just
432 	 * return -EIO.
433 	 */
434 	if (found_level == 0 && btrfs_check_leaf(eb)) {
435 		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
436 		ret = -EIO;
437 	}
438 
439 	if (found_level > 0 && btrfs_check_node(eb))
440 		ret = -EIO;
441 
442 	if (ret)
443 		btrfs_err(fs_info,
444 		"read time tree block corruption detected on logical %llu mirror %u",
445 			  eb->start, eb->read_mirror);
446 out:
447 	return ret;
448 }
449 
450 #ifdef CONFIG_MIGRATION
451 static int btree_migrate_folio(struct address_space *mapping,
452 		struct folio *dst, struct folio *src, enum migrate_mode mode)
453 {
454 	/*
455 	 * we can't safely write a btree page from here,
456 	 * we haven't done the locking hook
457 	 */
458 	if (folio_test_dirty(src))
459 		return -EAGAIN;
460 	/*
461 	 * Buffers may be managed in a filesystem specific way.
462 	 * We must have no buffers or drop them.
463 	 */
464 	if (folio_get_private(src) &&
465 	    !filemap_release_folio(src, GFP_KERNEL))
466 		return -EAGAIN;
467 	return migrate_folio(mapping, dst, src, mode);
468 }
469 #else
470 #define btree_migrate_folio NULL
471 #endif
472 
473 static int btree_writepages(struct address_space *mapping,
474 			    struct writeback_control *wbc)
475 {
476 	struct btrfs_fs_info *fs_info;
477 	int ret;
478 
479 	if (wbc->sync_mode == WB_SYNC_NONE) {
480 
481 		if (wbc->for_kupdate)
482 			return 0;
483 
484 		fs_info = BTRFS_I(mapping->host)->root->fs_info;
485 		/* this is a bit racy, but that's ok */
486 		ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
487 					     BTRFS_DIRTY_METADATA_THRESH,
488 					     fs_info->dirty_metadata_batch);
489 		if (ret < 0)
490 			return 0;
491 	}
492 	return btree_write_cache_pages(mapping, wbc);
493 }
494 
495 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
496 {
497 	if (folio_test_writeback(folio) || folio_test_dirty(folio))
498 		return false;
499 
500 	return try_release_extent_buffer(&folio->page);
501 }
502 
503 static void btree_invalidate_folio(struct folio *folio, size_t offset,
504 				 size_t length)
505 {
506 	struct extent_io_tree *tree;
507 	tree = &BTRFS_I(folio->mapping->host)->io_tree;
508 	extent_invalidate_folio(tree, folio, offset);
509 	btree_release_folio(folio, GFP_NOFS);
510 	if (folio_get_private(folio)) {
511 		btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
512 			   "folio private not zero on folio %llu",
513 			   (unsigned long long)folio_pos(folio));
514 		folio_detach_private(folio);
515 	}
516 }
517 
518 #ifdef DEBUG
519 static bool btree_dirty_folio(struct address_space *mapping,
520 		struct folio *folio)
521 {
522 	struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
523 	struct btrfs_subpage *subpage;
524 	struct extent_buffer *eb;
525 	int cur_bit = 0;
526 	u64 page_start = folio_pos(folio);
527 
528 	if (fs_info->sectorsize == PAGE_SIZE) {
529 		eb = folio_get_private(folio);
530 		BUG_ON(!eb);
531 		BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
532 		BUG_ON(!atomic_read(&eb->refs));
533 		btrfs_assert_tree_write_locked(eb);
534 		return filemap_dirty_folio(mapping, folio);
535 	}
536 	subpage = folio_get_private(folio);
537 
538 	ASSERT(subpage->dirty_bitmap);
539 	while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
540 		unsigned long flags;
541 		u64 cur;
542 		u16 tmp = (1 << cur_bit);
543 
544 		spin_lock_irqsave(&subpage->lock, flags);
545 		if (!(tmp & subpage->dirty_bitmap)) {
546 			spin_unlock_irqrestore(&subpage->lock, flags);
547 			cur_bit++;
548 			continue;
549 		}
550 		spin_unlock_irqrestore(&subpage->lock, flags);
551 		cur = page_start + cur_bit * fs_info->sectorsize;
552 
553 		eb = find_extent_buffer(fs_info, cur);
554 		ASSERT(eb);
555 		ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
556 		ASSERT(atomic_read(&eb->refs));
557 		btrfs_assert_tree_write_locked(eb);
558 		free_extent_buffer(eb);
559 
560 		cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
561 	}
562 	return filemap_dirty_folio(mapping, folio);
563 }
564 #else
565 #define btree_dirty_folio filemap_dirty_folio
566 #endif
567 
568 static const struct address_space_operations btree_aops = {
569 	.writepages	= btree_writepages,
570 	.release_folio	= btree_release_folio,
571 	.invalidate_folio = btree_invalidate_folio,
572 	.migrate_folio	= btree_migrate_folio,
573 	.dirty_folio	= btree_dirty_folio,
574 };
575 
576 struct extent_buffer *btrfs_find_create_tree_block(
577 						struct btrfs_fs_info *fs_info,
578 						u64 bytenr, u64 owner_root,
579 						int level)
580 {
581 	if (btrfs_is_testing(fs_info))
582 		return alloc_test_extent_buffer(fs_info, bytenr);
583 	return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
584 }
585 
586 /*
587  * Read tree block at logical address @bytenr and do variant basic but critical
588  * verification.
589  *
590  * @check:		expected tree parentness check, see comments of the
591  *			structure for details.
592  */
593 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
594 				      struct btrfs_tree_parent_check *check)
595 {
596 	struct extent_buffer *buf = NULL;
597 	int ret;
598 
599 	ASSERT(check);
600 
601 	buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
602 					   check->level);
603 	if (IS_ERR(buf))
604 		return buf;
605 
606 	ret = btrfs_read_extent_buffer(buf, check);
607 	if (ret) {
608 		free_extent_buffer_stale(buf);
609 		return ERR_PTR(ret);
610 	}
611 	if (btrfs_check_eb_owner(buf, check->owner_root)) {
612 		free_extent_buffer_stale(buf);
613 		return ERR_PTR(-EUCLEAN);
614 	}
615 	return buf;
616 
617 }
618 
619 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
620 			 u64 objectid)
621 {
622 	bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
623 
624 	memset(&root->root_key, 0, sizeof(root->root_key));
625 	memset(&root->root_item, 0, sizeof(root->root_item));
626 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
627 	root->fs_info = fs_info;
628 	root->root_key.objectid = objectid;
629 	root->node = NULL;
630 	root->commit_root = NULL;
631 	root->state = 0;
632 	RB_CLEAR_NODE(&root->rb_node);
633 
634 	root->last_trans = 0;
635 	root->free_objectid = 0;
636 	root->nr_delalloc_inodes = 0;
637 	root->nr_ordered_extents = 0;
638 	root->inode_tree = RB_ROOT;
639 	INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
640 
641 	btrfs_init_root_block_rsv(root);
642 
643 	INIT_LIST_HEAD(&root->dirty_list);
644 	INIT_LIST_HEAD(&root->root_list);
645 	INIT_LIST_HEAD(&root->delalloc_inodes);
646 	INIT_LIST_HEAD(&root->delalloc_root);
647 	INIT_LIST_HEAD(&root->ordered_extents);
648 	INIT_LIST_HEAD(&root->ordered_root);
649 	INIT_LIST_HEAD(&root->reloc_dirty_list);
650 	INIT_LIST_HEAD(&root->logged_list[0]);
651 	INIT_LIST_HEAD(&root->logged_list[1]);
652 	spin_lock_init(&root->inode_lock);
653 	spin_lock_init(&root->delalloc_lock);
654 	spin_lock_init(&root->ordered_extent_lock);
655 	spin_lock_init(&root->accounting_lock);
656 	spin_lock_init(&root->log_extents_lock[0]);
657 	spin_lock_init(&root->log_extents_lock[1]);
658 	spin_lock_init(&root->qgroup_meta_rsv_lock);
659 	mutex_init(&root->objectid_mutex);
660 	mutex_init(&root->log_mutex);
661 	mutex_init(&root->ordered_extent_mutex);
662 	mutex_init(&root->delalloc_mutex);
663 	init_waitqueue_head(&root->qgroup_flush_wait);
664 	init_waitqueue_head(&root->log_writer_wait);
665 	init_waitqueue_head(&root->log_commit_wait[0]);
666 	init_waitqueue_head(&root->log_commit_wait[1]);
667 	INIT_LIST_HEAD(&root->log_ctxs[0]);
668 	INIT_LIST_HEAD(&root->log_ctxs[1]);
669 	atomic_set(&root->log_commit[0], 0);
670 	atomic_set(&root->log_commit[1], 0);
671 	atomic_set(&root->log_writers, 0);
672 	atomic_set(&root->log_batch, 0);
673 	refcount_set(&root->refs, 1);
674 	atomic_set(&root->snapshot_force_cow, 0);
675 	atomic_set(&root->nr_swapfiles, 0);
676 	root->log_transid = 0;
677 	root->log_transid_committed = -1;
678 	root->last_log_commit = 0;
679 	root->anon_dev = 0;
680 	if (!dummy) {
681 		extent_io_tree_init(fs_info, &root->dirty_log_pages,
682 				    IO_TREE_ROOT_DIRTY_LOG_PAGES);
683 		extent_io_tree_init(fs_info, &root->log_csum_range,
684 				    IO_TREE_LOG_CSUM_RANGE);
685 	}
686 
687 	spin_lock_init(&root->root_item_lock);
688 	btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
689 #ifdef CONFIG_BTRFS_DEBUG
690 	INIT_LIST_HEAD(&root->leak_list);
691 	spin_lock(&fs_info->fs_roots_radix_lock);
692 	list_add_tail(&root->leak_list, &fs_info->allocated_roots);
693 	spin_unlock(&fs_info->fs_roots_radix_lock);
694 #endif
695 }
696 
697 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
698 					   u64 objectid, gfp_t flags)
699 {
700 	struct btrfs_root *root = kzalloc(sizeof(*root), flags);
701 	if (root)
702 		__setup_root(root, fs_info, objectid);
703 	return root;
704 }
705 
706 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
707 /* Should only be used by the testing infrastructure */
708 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
709 {
710 	struct btrfs_root *root;
711 
712 	if (!fs_info)
713 		return ERR_PTR(-EINVAL);
714 
715 	root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
716 	if (!root)
717 		return ERR_PTR(-ENOMEM);
718 
719 	/* We don't use the stripesize in selftest, set it as sectorsize */
720 	root->alloc_bytenr = 0;
721 
722 	return root;
723 }
724 #endif
725 
726 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
727 {
728 	const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
729 	const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
730 
731 	return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
732 }
733 
734 static int global_root_key_cmp(const void *k, const struct rb_node *node)
735 {
736 	const struct btrfs_key *key = k;
737 	const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
738 
739 	return btrfs_comp_cpu_keys(key, &root->root_key);
740 }
741 
742 int btrfs_global_root_insert(struct btrfs_root *root)
743 {
744 	struct btrfs_fs_info *fs_info = root->fs_info;
745 	struct rb_node *tmp;
746 	int ret = 0;
747 
748 	write_lock(&fs_info->global_root_lock);
749 	tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
750 	write_unlock(&fs_info->global_root_lock);
751 
752 	if (tmp) {
753 		ret = -EEXIST;
754 		btrfs_warn(fs_info, "global root %llu %llu already exists",
755 				root->root_key.objectid, root->root_key.offset);
756 	}
757 	return ret;
758 }
759 
760 void btrfs_global_root_delete(struct btrfs_root *root)
761 {
762 	struct btrfs_fs_info *fs_info = root->fs_info;
763 
764 	write_lock(&fs_info->global_root_lock);
765 	rb_erase(&root->rb_node, &fs_info->global_root_tree);
766 	write_unlock(&fs_info->global_root_lock);
767 }
768 
769 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
770 				     struct btrfs_key *key)
771 {
772 	struct rb_node *node;
773 	struct btrfs_root *root = NULL;
774 
775 	read_lock(&fs_info->global_root_lock);
776 	node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
777 	if (node)
778 		root = container_of(node, struct btrfs_root, rb_node);
779 	read_unlock(&fs_info->global_root_lock);
780 
781 	return root;
782 }
783 
784 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
785 {
786 	struct btrfs_block_group *block_group;
787 	u64 ret;
788 
789 	if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
790 		return 0;
791 
792 	if (bytenr)
793 		block_group = btrfs_lookup_block_group(fs_info, bytenr);
794 	else
795 		block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
796 	ASSERT(block_group);
797 	if (!block_group)
798 		return 0;
799 	ret = block_group->global_root_id;
800 	btrfs_put_block_group(block_group);
801 
802 	return ret;
803 }
804 
805 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
806 {
807 	struct btrfs_key key = {
808 		.objectid = BTRFS_CSUM_TREE_OBJECTID,
809 		.type = BTRFS_ROOT_ITEM_KEY,
810 		.offset = btrfs_global_root_id(fs_info, bytenr),
811 	};
812 
813 	return btrfs_global_root(fs_info, &key);
814 }
815 
816 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
817 {
818 	struct btrfs_key key = {
819 		.objectid = BTRFS_EXTENT_TREE_OBJECTID,
820 		.type = BTRFS_ROOT_ITEM_KEY,
821 		.offset = btrfs_global_root_id(fs_info, bytenr),
822 	};
823 
824 	return btrfs_global_root(fs_info, &key);
825 }
826 
827 struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info)
828 {
829 	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE))
830 		return fs_info->block_group_root;
831 	return btrfs_extent_root(fs_info, 0);
832 }
833 
834 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
835 				     u64 objectid)
836 {
837 	struct btrfs_fs_info *fs_info = trans->fs_info;
838 	struct extent_buffer *leaf;
839 	struct btrfs_root *tree_root = fs_info->tree_root;
840 	struct btrfs_root *root;
841 	struct btrfs_key key;
842 	unsigned int nofs_flag;
843 	int ret = 0;
844 
845 	/*
846 	 * We're holding a transaction handle, so use a NOFS memory allocation
847 	 * context to avoid deadlock if reclaim happens.
848 	 */
849 	nofs_flag = memalloc_nofs_save();
850 	root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
851 	memalloc_nofs_restore(nofs_flag);
852 	if (!root)
853 		return ERR_PTR(-ENOMEM);
854 
855 	root->root_key.objectid = objectid;
856 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
857 	root->root_key.offset = 0;
858 
859 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
860 				      BTRFS_NESTING_NORMAL);
861 	if (IS_ERR(leaf)) {
862 		ret = PTR_ERR(leaf);
863 		leaf = NULL;
864 		goto fail;
865 	}
866 
867 	root->node = leaf;
868 	btrfs_mark_buffer_dirty(leaf);
869 
870 	root->commit_root = btrfs_root_node(root);
871 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
872 
873 	btrfs_set_root_flags(&root->root_item, 0);
874 	btrfs_set_root_limit(&root->root_item, 0);
875 	btrfs_set_root_bytenr(&root->root_item, leaf->start);
876 	btrfs_set_root_generation(&root->root_item, trans->transid);
877 	btrfs_set_root_level(&root->root_item, 0);
878 	btrfs_set_root_refs(&root->root_item, 1);
879 	btrfs_set_root_used(&root->root_item, leaf->len);
880 	btrfs_set_root_last_snapshot(&root->root_item, 0);
881 	btrfs_set_root_dirid(&root->root_item, 0);
882 	if (is_fstree(objectid))
883 		generate_random_guid(root->root_item.uuid);
884 	else
885 		export_guid(root->root_item.uuid, &guid_null);
886 	btrfs_set_root_drop_level(&root->root_item, 0);
887 
888 	btrfs_tree_unlock(leaf);
889 
890 	key.objectid = objectid;
891 	key.type = BTRFS_ROOT_ITEM_KEY;
892 	key.offset = 0;
893 	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
894 	if (ret)
895 		goto fail;
896 
897 	return root;
898 
899 fail:
900 	btrfs_put_root(root);
901 
902 	return ERR_PTR(ret);
903 }
904 
905 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
906 					 struct btrfs_fs_info *fs_info)
907 {
908 	struct btrfs_root *root;
909 
910 	root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
911 	if (!root)
912 		return ERR_PTR(-ENOMEM);
913 
914 	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
915 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
916 	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
917 
918 	return root;
919 }
920 
921 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
922 			      struct btrfs_root *root)
923 {
924 	struct extent_buffer *leaf;
925 
926 	/*
927 	 * DON'T set SHAREABLE bit for log trees.
928 	 *
929 	 * Log trees are not exposed to user space thus can't be snapshotted,
930 	 * and they go away before a real commit is actually done.
931 	 *
932 	 * They do store pointers to file data extents, and those reference
933 	 * counts still get updated (along with back refs to the log tree).
934 	 */
935 
936 	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
937 			NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
938 	if (IS_ERR(leaf))
939 		return PTR_ERR(leaf);
940 
941 	root->node = leaf;
942 
943 	btrfs_mark_buffer_dirty(root->node);
944 	btrfs_tree_unlock(root->node);
945 
946 	return 0;
947 }
948 
949 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
950 			     struct btrfs_fs_info *fs_info)
951 {
952 	struct btrfs_root *log_root;
953 
954 	log_root = alloc_log_tree(trans, fs_info);
955 	if (IS_ERR(log_root))
956 		return PTR_ERR(log_root);
957 
958 	if (!btrfs_is_zoned(fs_info)) {
959 		int ret = btrfs_alloc_log_tree_node(trans, log_root);
960 
961 		if (ret) {
962 			btrfs_put_root(log_root);
963 			return ret;
964 		}
965 	}
966 
967 	WARN_ON(fs_info->log_root_tree);
968 	fs_info->log_root_tree = log_root;
969 	return 0;
970 }
971 
972 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
973 		       struct btrfs_root *root)
974 {
975 	struct btrfs_fs_info *fs_info = root->fs_info;
976 	struct btrfs_root *log_root;
977 	struct btrfs_inode_item *inode_item;
978 	int ret;
979 
980 	log_root = alloc_log_tree(trans, fs_info);
981 	if (IS_ERR(log_root))
982 		return PTR_ERR(log_root);
983 
984 	ret = btrfs_alloc_log_tree_node(trans, log_root);
985 	if (ret) {
986 		btrfs_put_root(log_root);
987 		return ret;
988 	}
989 
990 	log_root->last_trans = trans->transid;
991 	log_root->root_key.offset = root->root_key.objectid;
992 
993 	inode_item = &log_root->root_item.inode;
994 	btrfs_set_stack_inode_generation(inode_item, 1);
995 	btrfs_set_stack_inode_size(inode_item, 3);
996 	btrfs_set_stack_inode_nlink(inode_item, 1);
997 	btrfs_set_stack_inode_nbytes(inode_item,
998 				     fs_info->nodesize);
999 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1000 
1001 	btrfs_set_root_node(&log_root->root_item, log_root->node);
1002 
1003 	WARN_ON(root->log_root);
1004 	root->log_root = log_root;
1005 	root->log_transid = 0;
1006 	root->log_transid_committed = -1;
1007 	root->last_log_commit = 0;
1008 	return 0;
1009 }
1010 
1011 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1012 					      struct btrfs_path *path,
1013 					      struct btrfs_key *key)
1014 {
1015 	struct btrfs_root *root;
1016 	struct btrfs_tree_parent_check check = { 0 };
1017 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1018 	u64 generation;
1019 	int ret;
1020 	int level;
1021 
1022 	root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1023 	if (!root)
1024 		return ERR_PTR(-ENOMEM);
1025 
1026 	ret = btrfs_find_root(tree_root, key, path,
1027 			      &root->root_item, &root->root_key);
1028 	if (ret) {
1029 		if (ret > 0)
1030 			ret = -ENOENT;
1031 		goto fail;
1032 	}
1033 
1034 	generation = btrfs_root_generation(&root->root_item);
1035 	level = btrfs_root_level(&root->root_item);
1036 	check.level = level;
1037 	check.transid = generation;
1038 	check.owner_root = key->objectid;
1039 	root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1040 				     &check);
1041 	if (IS_ERR(root->node)) {
1042 		ret = PTR_ERR(root->node);
1043 		root->node = NULL;
1044 		goto fail;
1045 	}
1046 	if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1047 		ret = -EIO;
1048 		goto fail;
1049 	}
1050 
1051 	/*
1052 	 * For real fs, and not log/reloc trees, root owner must
1053 	 * match its root node owner
1054 	 */
1055 	if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1056 	    root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1057 	    root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1058 	    root->root_key.objectid != btrfs_header_owner(root->node)) {
1059 		btrfs_crit(fs_info,
1060 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1061 			   root->root_key.objectid, root->node->start,
1062 			   btrfs_header_owner(root->node),
1063 			   root->root_key.objectid);
1064 		ret = -EUCLEAN;
1065 		goto fail;
1066 	}
1067 	root->commit_root = btrfs_root_node(root);
1068 	return root;
1069 fail:
1070 	btrfs_put_root(root);
1071 	return ERR_PTR(ret);
1072 }
1073 
1074 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1075 					struct btrfs_key *key)
1076 {
1077 	struct btrfs_root *root;
1078 	struct btrfs_path *path;
1079 
1080 	path = btrfs_alloc_path();
1081 	if (!path)
1082 		return ERR_PTR(-ENOMEM);
1083 	root = read_tree_root_path(tree_root, path, key);
1084 	btrfs_free_path(path);
1085 
1086 	return root;
1087 }
1088 
1089 /*
1090  * Initialize subvolume root in-memory structure
1091  *
1092  * @anon_dev:	anonymous device to attach to the root, if zero, allocate new
1093  */
1094 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1095 {
1096 	int ret;
1097 
1098 	btrfs_drew_lock_init(&root->snapshot_lock);
1099 
1100 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1101 	    !btrfs_is_data_reloc_root(root) &&
1102 	    is_fstree(root->root_key.objectid)) {
1103 		set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1104 		btrfs_check_and_init_root_item(&root->root_item);
1105 	}
1106 
1107 	/*
1108 	 * Don't assign anonymous block device to roots that are not exposed to
1109 	 * userspace, the id pool is limited to 1M
1110 	 */
1111 	if (is_fstree(root->root_key.objectid) &&
1112 	    btrfs_root_refs(&root->root_item) > 0) {
1113 		if (!anon_dev) {
1114 			ret = get_anon_bdev(&root->anon_dev);
1115 			if (ret)
1116 				goto fail;
1117 		} else {
1118 			root->anon_dev = anon_dev;
1119 		}
1120 	}
1121 
1122 	mutex_lock(&root->objectid_mutex);
1123 	ret = btrfs_init_root_free_objectid(root);
1124 	if (ret) {
1125 		mutex_unlock(&root->objectid_mutex);
1126 		goto fail;
1127 	}
1128 
1129 	ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1130 
1131 	mutex_unlock(&root->objectid_mutex);
1132 
1133 	return 0;
1134 fail:
1135 	/* The caller is responsible to call btrfs_free_fs_root */
1136 	return ret;
1137 }
1138 
1139 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1140 					       u64 root_id)
1141 {
1142 	struct btrfs_root *root;
1143 
1144 	spin_lock(&fs_info->fs_roots_radix_lock);
1145 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1146 				 (unsigned long)root_id);
1147 	root = btrfs_grab_root(root);
1148 	spin_unlock(&fs_info->fs_roots_radix_lock);
1149 	return root;
1150 }
1151 
1152 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1153 						u64 objectid)
1154 {
1155 	struct btrfs_key key = {
1156 		.objectid = objectid,
1157 		.type = BTRFS_ROOT_ITEM_KEY,
1158 		.offset = 0,
1159 	};
1160 
1161 	switch (objectid) {
1162 	case BTRFS_ROOT_TREE_OBJECTID:
1163 		return btrfs_grab_root(fs_info->tree_root);
1164 	case BTRFS_EXTENT_TREE_OBJECTID:
1165 		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1166 	case BTRFS_CHUNK_TREE_OBJECTID:
1167 		return btrfs_grab_root(fs_info->chunk_root);
1168 	case BTRFS_DEV_TREE_OBJECTID:
1169 		return btrfs_grab_root(fs_info->dev_root);
1170 	case BTRFS_CSUM_TREE_OBJECTID:
1171 		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1172 	case BTRFS_QUOTA_TREE_OBJECTID:
1173 		return btrfs_grab_root(fs_info->quota_root);
1174 	case BTRFS_UUID_TREE_OBJECTID:
1175 		return btrfs_grab_root(fs_info->uuid_root);
1176 	case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
1177 		return btrfs_grab_root(fs_info->block_group_root);
1178 	case BTRFS_FREE_SPACE_TREE_OBJECTID:
1179 		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1180 	default:
1181 		return NULL;
1182 	}
1183 }
1184 
1185 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1186 			 struct btrfs_root *root)
1187 {
1188 	int ret;
1189 
1190 	ret = radix_tree_preload(GFP_NOFS);
1191 	if (ret)
1192 		return ret;
1193 
1194 	spin_lock(&fs_info->fs_roots_radix_lock);
1195 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1196 				(unsigned long)root->root_key.objectid,
1197 				root);
1198 	if (ret == 0) {
1199 		btrfs_grab_root(root);
1200 		set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1201 	}
1202 	spin_unlock(&fs_info->fs_roots_radix_lock);
1203 	radix_tree_preload_end();
1204 
1205 	return ret;
1206 }
1207 
1208 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1209 {
1210 #ifdef CONFIG_BTRFS_DEBUG
1211 	struct btrfs_root *root;
1212 
1213 	while (!list_empty(&fs_info->allocated_roots)) {
1214 		char buf[BTRFS_ROOT_NAME_BUF_LEN];
1215 
1216 		root = list_first_entry(&fs_info->allocated_roots,
1217 					struct btrfs_root, leak_list);
1218 		btrfs_err(fs_info, "leaked root %s refcount %d",
1219 			  btrfs_root_name(&root->root_key, buf),
1220 			  refcount_read(&root->refs));
1221 		while (refcount_read(&root->refs) > 1)
1222 			btrfs_put_root(root);
1223 		btrfs_put_root(root);
1224 	}
1225 #endif
1226 }
1227 
1228 static void free_global_roots(struct btrfs_fs_info *fs_info)
1229 {
1230 	struct btrfs_root *root;
1231 	struct rb_node *node;
1232 
1233 	while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1234 		root = rb_entry(node, struct btrfs_root, rb_node);
1235 		rb_erase(&root->rb_node, &fs_info->global_root_tree);
1236 		btrfs_put_root(root);
1237 	}
1238 }
1239 
1240 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1241 {
1242 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1243 	percpu_counter_destroy(&fs_info->delalloc_bytes);
1244 	percpu_counter_destroy(&fs_info->ordered_bytes);
1245 	percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1246 	btrfs_free_csum_hash(fs_info);
1247 	btrfs_free_stripe_hash_table(fs_info);
1248 	btrfs_free_ref_cache(fs_info);
1249 	kfree(fs_info->balance_ctl);
1250 	kfree(fs_info->delayed_root);
1251 	free_global_roots(fs_info);
1252 	btrfs_put_root(fs_info->tree_root);
1253 	btrfs_put_root(fs_info->chunk_root);
1254 	btrfs_put_root(fs_info->dev_root);
1255 	btrfs_put_root(fs_info->quota_root);
1256 	btrfs_put_root(fs_info->uuid_root);
1257 	btrfs_put_root(fs_info->fs_root);
1258 	btrfs_put_root(fs_info->data_reloc_root);
1259 	btrfs_put_root(fs_info->block_group_root);
1260 	btrfs_check_leaked_roots(fs_info);
1261 	btrfs_extent_buffer_leak_debug_check(fs_info);
1262 	kfree(fs_info->super_copy);
1263 	kfree(fs_info->super_for_commit);
1264 	kfree(fs_info->subpage_info);
1265 	kvfree(fs_info);
1266 }
1267 
1268 
1269 /*
1270  * Get an in-memory reference of a root structure.
1271  *
1272  * For essential trees like root/extent tree, we grab it from fs_info directly.
1273  * For subvolume trees, we check the cached filesystem roots first. If not
1274  * found, then read it from disk and add it to cached fs roots.
1275  *
1276  * Caller should release the root by calling btrfs_put_root() after the usage.
1277  *
1278  * NOTE: Reloc and log trees can't be read by this function as they share the
1279  *	 same root objectid.
1280  *
1281  * @objectid:	root id
1282  * @anon_dev:	preallocated anonymous block device number for new roots,
1283  * 		pass 0 for new allocation.
1284  * @check_ref:	whether to check root item references, If true, return -ENOENT
1285  *		for orphan roots
1286  */
1287 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1288 					     u64 objectid, dev_t anon_dev,
1289 					     bool check_ref)
1290 {
1291 	struct btrfs_root *root;
1292 	struct btrfs_path *path;
1293 	struct btrfs_key key;
1294 	int ret;
1295 
1296 	root = btrfs_get_global_root(fs_info, objectid);
1297 	if (root)
1298 		return root;
1299 
1300 	/*
1301 	 * If we're called for non-subvolume trees, and above function didn't
1302 	 * find one, do not try to read it from disk.
1303 	 *
1304 	 * This is namely for free-space-tree and quota tree, which can change
1305 	 * at runtime and should only be grabbed from fs_info.
1306 	 */
1307 	if (!is_fstree(objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1308 		return ERR_PTR(-ENOENT);
1309 again:
1310 	root = btrfs_lookup_fs_root(fs_info, objectid);
1311 	if (root) {
1312 		/* Shouldn't get preallocated anon_dev for cached roots */
1313 		ASSERT(!anon_dev);
1314 		if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1315 			btrfs_put_root(root);
1316 			return ERR_PTR(-ENOENT);
1317 		}
1318 		return root;
1319 	}
1320 
1321 	key.objectid = objectid;
1322 	key.type = BTRFS_ROOT_ITEM_KEY;
1323 	key.offset = (u64)-1;
1324 	root = btrfs_read_tree_root(fs_info->tree_root, &key);
1325 	if (IS_ERR(root))
1326 		return root;
1327 
1328 	if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1329 		ret = -ENOENT;
1330 		goto fail;
1331 	}
1332 
1333 	ret = btrfs_init_fs_root(root, anon_dev);
1334 	if (ret)
1335 		goto fail;
1336 
1337 	path = btrfs_alloc_path();
1338 	if (!path) {
1339 		ret = -ENOMEM;
1340 		goto fail;
1341 	}
1342 	key.objectid = BTRFS_ORPHAN_OBJECTID;
1343 	key.type = BTRFS_ORPHAN_ITEM_KEY;
1344 	key.offset = objectid;
1345 
1346 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1347 	btrfs_free_path(path);
1348 	if (ret < 0)
1349 		goto fail;
1350 	if (ret == 0)
1351 		set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1352 
1353 	ret = btrfs_insert_fs_root(fs_info, root);
1354 	if (ret) {
1355 		if (ret == -EEXIST) {
1356 			btrfs_put_root(root);
1357 			goto again;
1358 		}
1359 		goto fail;
1360 	}
1361 	return root;
1362 fail:
1363 	/*
1364 	 * If our caller provided us an anonymous device, then it's his
1365 	 * responsibility to free it in case we fail. So we have to set our
1366 	 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1367 	 * and once again by our caller.
1368 	 */
1369 	if (anon_dev)
1370 		root->anon_dev = 0;
1371 	btrfs_put_root(root);
1372 	return ERR_PTR(ret);
1373 }
1374 
1375 /*
1376  * Get in-memory reference of a root structure
1377  *
1378  * @objectid:	tree objectid
1379  * @check_ref:	if set, verify that the tree exists and the item has at least
1380  *		one reference
1381  */
1382 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1383 				     u64 objectid, bool check_ref)
1384 {
1385 	return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1386 }
1387 
1388 /*
1389  * Get in-memory reference of a root structure, created as new, optionally pass
1390  * the anonymous block device id
1391  *
1392  * @objectid:	tree objectid
1393  * @anon_dev:	if zero, allocate a new anonymous block device or use the
1394  *		parameter value
1395  */
1396 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1397 					 u64 objectid, dev_t anon_dev)
1398 {
1399 	return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1400 }
1401 
1402 /*
1403  * btrfs_get_fs_root_commit_root - return a root for the given objectid
1404  * @fs_info:	the fs_info
1405  * @objectid:	the objectid we need to lookup
1406  *
1407  * This is exclusively used for backref walking, and exists specifically because
1408  * of how qgroups does lookups.  Qgroups will do a backref lookup at delayed ref
1409  * creation time, which means we may have to read the tree_root in order to look
1410  * up a fs root that is not in memory.  If the root is not in memory we will
1411  * read the tree root commit root and look up the fs root from there.  This is a
1412  * temporary root, it will not be inserted into the radix tree as it doesn't
1413  * have the most uptodate information, it'll simply be discarded once the
1414  * backref code is finished using the root.
1415  */
1416 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1417 						 struct btrfs_path *path,
1418 						 u64 objectid)
1419 {
1420 	struct btrfs_root *root;
1421 	struct btrfs_key key;
1422 
1423 	ASSERT(path->search_commit_root && path->skip_locking);
1424 
1425 	/*
1426 	 * This can return -ENOENT if we ask for a root that doesn't exist, but
1427 	 * since this is called via the backref walking code we won't be looking
1428 	 * up a root that doesn't exist, unless there's corruption.  So if root
1429 	 * != NULL just return it.
1430 	 */
1431 	root = btrfs_get_global_root(fs_info, objectid);
1432 	if (root)
1433 		return root;
1434 
1435 	root = btrfs_lookup_fs_root(fs_info, objectid);
1436 	if (root)
1437 		return root;
1438 
1439 	key.objectid = objectid;
1440 	key.type = BTRFS_ROOT_ITEM_KEY;
1441 	key.offset = (u64)-1;
1442 	root = read_tree_root_path(fs_info->tree_root, path, &key);
1443 	btrfs_release_path(path);
1444 
1445 	return root;
1446 }
1447 
1448 static int cleaner_kthread(void *arg)
1449 {
1450 	struct btrfs_fs_info *fs_info = arg;
1451 	int again;
1452 
1453 	while (1) {
1454 		again = 0;
1455 
1456 		set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1457 
1458 		/* Make the cleaner go to sleep early. */
1459 		if (btrfs_need_cleaner_sleep(fs_info))
1460 			goto sleep;
1461 
1462 		/*
1463 		 * Do not do anything if we might cause open_ctree() to block
1464 		 * before we have finished mounting the filesystem.
1465 		 */
1466 		if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1467 			goto sleep;
1468 
1469 		if (!mutex_trylock(&fs_info->cleaner_mutex))
1470 			goto sleep;
1471 
1472 		/*
1473 		 * Avoid the problem that we change the status of the fs
1474 		 * during the above check and trylock.
1475 		 */
1476 		if (btrfs_need_cleaner_sleep(fs_info)) {
1477 			mutex_unlock(&fs_info->cleaner_mutex);
1478 			goto sleep;
1479 		}
1480 
1481 		if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1482 			btrfs_sysfs_feature_update(fs_info);
1483 
1484 		btrfs_run_delayed_iputs(fs_info);
1485 
1486 		again = btrfs_clean_one_deleted_snapshot(fs_info);
1487 		mutex_unlock(&fs_info->cleaner_mutex);
1488 
1489 		/*
1490 		 * The defragger has dealt with the R/O remount and umount,
1491 		 * needn't do anything special here.
1492 		 */
1493 		btrfs_run_defrag_inodes(fs_info);
1494 
1495 		/*
1496 		 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1497 		 * with relocation (btrfs_relocate_chunk) and relocation
1498 		 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1499 		 * after acquiring fs_info->reclaim_bgs_lock. So we
1500 		 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1501 		 * unused block groups.
1502 		 */
1503 		btrfs_delete_unused_bgs(fs_info);
1504 
1505 		/*
1506 		 * Reclaim block groups in the reclaim_bgs list after we deleted
1507 		 * all unused block_groups. This possibly gives us some more free
1508 		 * space.
1509 		 */
1510 		btrfs_reclaim_bgs(fs_info);
1511 sleep:
1512 		clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1513 		if (kthread_should_park())
1514 			kthread_parkme();
1515 		if (kthread_should_stop())
1516 			return 0;
1517 		if (!again) {
1518 			set_current_state(TASK_INTERRUPTIBLE);
1519 			schedule();
1520 			__set_current_state(TASK_RUNNING);
1521 		}
1522 	}
1523 }
1524 
1525 static int transaction_kthread(void *arg)
1526 {
1527 	struct btrfs_root *root = arg;
1528 	struct btrfs_fs_info *fs_info = root->fs_info;
1529 	struct btrfs_trans_handle *trans;
1530 	struct btrfs_transaction *cur;
1531 	u64 transid;
1532 	time64_t delta;
1533 	unsigned long delay;
1534 	bool cannot_commit;
1535 
1536 	do {
1537 		cannot_commit = false;
1538 		delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1539 		mutex_lock(&fs_info->transaction_kthread_mutex);
1540 
1541 		spin_lock(&fs_info->trans_lock);
1542 		cur = fs_info->running_transaction;
1543 		if (!cur) {
1544 			spin_unlock(&fs_info->trans_lock);
1545 			goto sleep;
1546 		}
1547 
1548 		delta = ktime_get_seconds() - cur->start_time;
1549 		if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1550 		    cur->state < TRANS_STATE_COMMIT_START &&
1551 		    delta < fs_info->commit_interval) {
1552 			spin_unlock(&fs_info->trans_lock);
1553 			delay -= msecs_to_jiffies((delta - 1) * 1000);
1554 			delay = min(delay,
1555 				    msecs_to_jiffies(fs_info->commit_interval * 1000));
1556 			goto sleep;
1557 		}
1558 		transid = cur->transid;
1559 		spin_unlock(&fs_info->trans_lock);
1560 
1561 		/* If the file system is aborted, this will always fail. */
1562 		trans = btrfs_attach_transaction(root);
1563 		if (IS_ERR(trans)) {
1564 			if (PTR_ERR(trans) != -ENOENT)
1565 				cannot_commit = true;
1566 			goto sleep;
1567 		}
1568 		if (transid == trans->transid) {
1569 			btrfs_commit_transaction(trans);
1570 		} else {
1571 			btrfs_end_transaction(trans);
1572 		}
1573 sleep:
1574 		wake_up_process(fs_info->cleaner_kthread);
1575 		mutex_unlock(&fs_info->transaction_kthread_mutex);
1576 
1577 		if (BTRFS_FS_ERROR(fs_info))
1578 			btrfs_cleanup_transaction(fs_info);
1579 		if (!kthread_should_stop() &&
1580 				(!btrfs_transaction_blocked(fs_info) ||
1581 				 cannot_commit))
1582 			schedule_timeout_interruptible(delay);
1583 	} while (!kthread_should_stop());
1584 	return 0;
1585 }
1586 
1587 /*
1588  * This will find the highest generation in the array of root backups.  The
1589  * index of the highest array is returned, or -EINVAL if we can't find
1590  * anything.
1591  *
1592  * We check to make sure the array is valid by comparing the
1593  * generation of the latest  root in the array with the generation
1594  * in the super block.  If they don't match we pitch it.
1595  */
1596 static int find_newest_super_backup(struct btrfs_fs_info *info)
1597 {
1598 	const u64 newest_gen = btrfs_super_generation(info->super_copy);
1599 	u64 cur;
1600 	struct btrfs_root_backup *root_backup;
1601 	int i;
1602 
1603 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1604 		root_backup = info->super_copy->super_roots + i;
1605 		cur = btrfs_backup_tree_root_gen(root_backup);
1606 		if (cur == newest_gen)
1607 			return i;
1608 	}
1609 
1610 	return -EINVAL;
1611 }
1612 
1613 /*
1614  * copy all the root pointers into the super backup array.
1615  * this will bump the backup pointer by one when it is
1616  * done
1617  */
1618 static void backup_super_roots(struct btrfs_fs_info *info)
1619 {
1620 	const int next_backup = info->backup_root_index;
1621 	struct btrfs_root_backup *root_backup;
1622 
1623 	root_backup = info->super_for_commit->super_roots + next_backup;
1624 
1625 	/*
1626 	 * make sure all of our padding and empty slots get zero filled
1627 	 * regardless of which ones we use today
1628 	 */
1629 	memset(root_backup, 0, sizeof(*root_backup));
1630 
1631 	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1632 
1633 	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1634 	btrfs_set_backup_tree_root_gen(root_backup,
1635 			       btrfs_header_generation(info->tree_root->node));
1636 
1637 	btrfs_set_backup_tree_root_level(root_backup,
1638 			       btrfs_header_level(info->tree_root->node));
1639 
1640 	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1641 	btrfs_set_backup_chunk_root_gen(root_backup,
1642 			       btrfs_header_generation(info->chunk_root->node));
1643 	btrfs_set_backup_chunk_root_level(root_backup,
1644 			       btrfs_header_level(info->chunk_root->node));
1645 
1646 	if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1647 		struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1648 		struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1649 
1650 		btrfs_set_backup_extent_root(root_backup,
1651 					     extent_root->node->start);
1652 		btrfs_set_backup_extent_root_gen(root_backup,
1653 				btrfs_header_generation(extent_root->node));
1654 		btrfs_set_backup_extent_root_level(root_backup,
1655 					btrfs_header_level(extent_root->node));
1656 
1657 		btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
1658 		btrfs_set_backup_csum_root_gen(root_backup,
1659 					       btrfs_header_generation(csum_root->node));
1660 		btrfs_set_backup_csum_root_level(root_backup,
1661 						 btrfs_header_level(csum_root->node));
1662 	}
1663 
1664 	/*
1665 	 * we might commit during log recovery, which happens before we set
1666 	 * the fs_root.  Make sure it is valid before we fill it in.
1667 	 */
1668 	if (info->fs_root && info->fs_root->node) {
1669 		btrfs_set_backup_fs_root(root_backup,
1670 					 info->fs_root->node->start);
1671 		btrfs_set_backup_fs_root_gen(root_backup,
1672 			       btrfs_header_generation(info->fs_root->node));
1673 		btrfs_set_backup_fs_root_level(root_backup,
1674 			       btrfs_header_level(info->fs_root->node));
1675 	}
1676 
1677 	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1678 	btrfs_set_backup_dev_root_gen(root_backup,
1679 			       btrfs_header_generation(info->dev_root->node));
1680 	btrfs_set_backup_dev_root_level(root_backup,
1681 				       btrfs_header_level(info->dev_root->node));
1682 
1683 	btrfs_set_backup_total_bytes(root_backup,
1684 			     btrfs_super_total_bytes(info->super_copy));
1685 	btrfs_set_backup_bytes_used(root_backup,
1686 			     btrfs_super_bytes_used(info->super_copy));
1687 	btrfs_set_backup_num_devices(root_backup,
1688 			     btrfs_super_num_devices(info->super_copy));
1689 
1690 	/*
1691 	 * if we don't copy this out to the super_copy, it won't get remembered
1692 	 * for the next commit
1693 	 */
1694 	memcpy(&info->super_copy->super_roots,
1695 	       &info->super_for_commit->super_roots,
1696 	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1697 }
1698 
1699 /*
1700  * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1701  * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1702  *
1703  * fs_info - filesystem whose backup roots need to be read
1704  * priority - priority of backup root required
1705  *
1706  * Returns backup root index on success and -EINVAL otherwise.
1707  */
1708 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1709 {
1710 	int backup_index = find_newest_super_backup(fs_info);
1711 	struct btrfs_super_block *super = fs_info->super_copy;
1712 	struct btrfs_root_backup *root_backup;
1713 
1714 	if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1715 		if (priority == 0)
1716 			return backup_index;
1717 
1718 		backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1719 		backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1720 	} else {
1721 		return -EINVAL;
1722 	}
1723 
1724 	root_backup = super->super_roots + backup_index;
1725 
1726 	btrfs_set_super_generation(super,
1727 				   btrfs_backup_tree_root_gen(root_backup));
1728 	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1729 	btrfs_set_super_root_level(super,
1730 				   btrfs_backup_tree_root_level(root_backup));
1731 	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1732 
1733 	/*
1734 	 * Fixme: the total bytes and num_devices need to match or we should
1735 	 * need a fsck
1736 	 */
1737 	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1738 	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1739 
1740 	return backup_index;
1741 }
1742 
1743 /* helper to cleanup workers */
1744 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1745 {
1746 	btrfs_destroy_workqueue(fs_info->fixup_workers);
1747 	btrfs_destroy_workqueue(fs_info->delalloc_workers);
1748 	btrfs_destroy_workqueue(fs_info->workers);
1749 	if (fs_info->endio_workers)
1750 		destroy_workqueue(fs_info->endio_workers);
1751 	if (fs_info->rmw_workers)
1752 		destroy_workqueue(fs_info->rmw_workers);
1753 	if (fs_info->compressed_write_workers)
1754 		destroy_workqueue(fs_info->compressed_write_workers);
1755 	btrfs_destroy_workqueue(fs_info->endio_write_workers);
1756 	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1757 	btrfs_destroy_workqueue(fs_info->delayed_workers);
1758 	btrfs_destroy_workqueue(fs_info->caching_workers);
1759 	btrfs_destroy_workqueue(fs_info->flush_workers);
1760 	btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1761 	if (fs_info->discard_ctl.discard_workers)
1762 		destroy_workqueue(fs_info->discard_ctl.discard_workers);
1763 	/*
1764 	 * Now that all other work queues are destroyed, we can safely destroy
1765 	 * the queues used for metadata I/O, since tasks from those other work
1766 	 * queues can do metadata I/O operations.
1767 	 */
1768 	if (fs_info->endio_meta_workers)
1769 		destroy_workqueue(fs_info->endio_meta_workers);
1770 }
1771 
1772 static void free_root_extent_buffers(struct btrfs_root *root)
1773 {
1774 	if (root) {
1775 		free_extent_buffer(root->node);
1776 		free_extent_buffer(root->commit_root);
1777 		root->node = NULL;
1778 		root->commit_root = NULL;
1779 	}
1780 }
1781 
1782 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1783 {
1784 	struct btrfs_root *root, *tmp;
1785 
1786 	rbtree_postorder_for_each_entry_safe(root, tmp,
1787 					     &fs_info->global_root_tree,
1788 					     rb_node)
1789 		free_root_extent_buffers(root);
1790 }
1791 
1792 /* helper to cleanup tree roots */
1793 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1794 {
1795 	free_root_extent_buffers(info->tree_root);
1796 
1797 	free_global_root_pointers(info);
1798 	free_root_extent_buffers(info->dev_root);
1799 	free_root_extent_buffers(info->quota_root);
1800 	free_root_extent_buffers(info->uuid_root);
1801 	free_root_extent_buffers(info->fs_root);
1802 	free_root_extent_buffers(info->data_reloc_root);
1803 	free_root_extent_buffers(info->block_group_root);
1804 	if (free_chunk_root)
1805 		free_root_extent_buffers(info->chunk_root);
1806 }
1807 
1808 void btrfs_put_root(struct btrfs_root *root)
1809 {
1810 	if (!root)
1811 		return;
1812 
1813 	if (refcount_dec_and_test(&root->refs)) {
1814 		WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1815 		WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1816 		if (root->anon_dev)
1817 			free_anon_bdev(root->anon_dev);
1818 		free_root_extent_buffers(root);
1819 #ifdef CONFIG_BTRFS_DEBUG
1820 		spin_lock(&root->fs_info->fs_roots_radix_lock);
1821 		list_del_init(&root->leak_list);
1822 		spin_unlock(&root->fs_info->fs_roots_radix_lock);
1823 #endif
1824 		kfree(root);
1825 	}
1826 }
1827 
1828 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1829 {
1830 	int ret;
1831 	struct btrfs_root *gang[8];
1832 	int i;
1833 
1834 	while (!list_empty(&fs_info->dead_roots)) {
1835 		gang[0] = list_entry(fs_info->dead_roots.next,
1836 				     struct btrfs_root, root_list);
1837 		list_del(&gang[0]->root_list);
1838 
1839 		if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1840 			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
1841 		btrfs_put_root(gang[0]);
1842 	}
1843 
1844 	while (1) {
1845 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1846 					     (void **)gang, 0,
1847 					     ARRAY_SIZE(gang));
1848 		if (!ret)
1849 			break;
1850 		for (i = 0; i < ret; i++)
1851 			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
1852 	}
1853 }
1854 
1855 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1856 {
1857 	mutex_init(&fs_info->scrub_lock);
1858 	atomic_set(&fs_info->scrubs_running, 0);
1859 	atomic_set(&fs_info->scrub_pause_req, 0);
1860 	atomic_set(&fs_info->scrubs_paused, 0);
1861 	atomic_set(&fs_info->scrub_cancel_req, 0);
1862 	init_waitqueue_head(&fs_info->scrub_pause_wait);
1863 	refcount_set(&fs_info->scrub_workers_refcnt, 0);
1864 }
1865 
1866 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1867 {
1868 	spin_lock_init(&fs_info->balance_lock);
1869 	mutex_init(&fs_info->balance_mutex);
1870 	atomic_set(&fs_info->balance_pause_req, 0);
1871 	atomic_set(&fs_info->balance_cancel_req, 0);
1872 	fs_info->balance_ctl = NULL;
1873 	init_waitqueue_head(&fs_info->balance_wait_q);
1874 	atomic_set(&fs_info->reloc_cancel_req, 0);
1875 }
1876 
1877 static int btrfs_init_btree_inode(struct super_block *sb)
1878 {
1879 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1880 	unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1881 					      fs_info->tree_root);
1882 	struct inode *inode;
1883 
1884 	inode = new_inode(sb);
1885 	if (!inode)
1886 		return -ENOMEM;
1887 
1888 	inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1889 	set_nlink(inode, 1);
1890 	/*
1891 	 * we set the i_size on the btree inode to the max possible int.
1892 	 * the real end of the address space is determined by all of
1893 	 * the devices in the system
1894 	 */
1895 	inode->i_size = OFFSET_MAX;
1896 	inode->i_mapping->a_ops = &btree_aops;
1897 	mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
1898 
1899 	RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
1900 	extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
1901 			    IO_TREE_BTREE_INODE_IO);
1902 	extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
1903 
1904 	BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
1905 	BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
1906 	BTRFS_I(inode)->location.type = 0;
1907 	BTRFS_I(inode)->location.offset = 0;
1908 	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
1909 	__insert_inode_hash(inode, hash);
1910 	fs_info->btree_inode = inode;
1911 
1912 	return 0;
1913 }
1914 
1915 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1916 {
1917 	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1918 	init_rwsem(&fs_info->dev_replace.rwsem);
1919 	init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1920 }
1921 
1922 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1923 {
1924 	spin_lock_init(&fs_info->qgroup_lock);
1925 	mutex_init(&fs_info->qgroup_ioctl_lock);
1926 	fs_info->qgroup_tree = RB_ROOT;
1927 	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
1928 	fs_info->qgroup_seq = 1;
1929 	fs_info->qgroup_ulist = NULL;
1930 	fs_info->qgroup_rescan_running = false;
1931 	fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1932 	mutex_init(&fs_info->qgroup_rescan_lock);
1933 }
1934 
1935 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
1936 {
1937 	u32 max_active = fs_info->thread_pool_size;
1938 	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1939 	unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE;
1940 
1941 	fs_info->workers =
1942 		btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
1943 
1944 	fs_info->delalloc_workers =
1945 		btrfs_alloc_workqueue(fs_info, "delalloc",
1946 				      flags, max_active, 2);
1947 
1948 	fs_info->flush_workers =
1949 		btrfs_alloc_workqueue(fs_info, "flush_delalloc",
1950 				      flags, max_active, 0);
1951 
1952 	fs_info->caching_workers =
1953 		btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
1954 
1955 	fs_info->fixup_workers =
1956 		btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags);
1957 
1958 	fs_info->endio_workers =
1959 		alloc_workqueue("btrfs-endio", flags, max_active);
1960 	fs_info->endio_meta_workers =
1961 		alloc_workqueue("btrfs-endio-meta", flags, max_active);
1962 	fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
1963 	fs_info->endio_write_workers =
1964 		btrfs_alloc_workqueue(fs_info, "endio-write", flags,
1965 				      max_active, 2);
1966 	fs_info->compressed_write_workers =
1967 		alloc_workqueue("btrfs-compressed-write", flags, max_active);
1968 	fs_info->endio_freespace_worker =
1969 		btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
1970 				      max_active, 0);
1971 	fs_info->delayed_workers =
1972 		btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
1973 				      max_active, 0);
1974 	fs_info->qgroup_rescan_workers =
1975 		btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan",
1976 					      ordered_flags);
1977 	fs_info->discard_ctl.discard_workers =
1978 		alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE);
1979 
1980 	if (!(fs_info->workers &&
1981 	      fs_info->delalloc_workers && fs_info->flush_workers &&
1982 	      fs_info->endio_workers && fs_info->endio_meta_workers &&
1983 	      fs_info->compressed_write_workers &&
1984 	      fs_info->endio_write_workers &&
1985 	      fs_info->endio_freespace_worker && fs_info->rmw_workers &&
1986 	      fs_info->caching_workers && fs_info->fixup_workers &&
1987 	      fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
1988 	      fs_info->discard_ctl.discard_workers)) {
1989 		return -ENOMEM;
1990 	}
1991 
1992 	return 0;
1993 }
1994 
1995 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
1996 {
1997 	struct crypto_shash *csum_shash;
1998 	const char *csum_driver = btrfs_super_csum_driver(csum_type);
1999 
2000 	csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2001 
2002 	if (IS_ERR(csum_shash)) {
2003 		btrfs_err(fs_info, "error allocating %s hash for checksum",
2004 			  csum_driver);
2005 		return PTR_ERR(csum_shash);
2006 	}
2007 
2008 	fs_info->csum_shash = csum_shash;
2009 
2010 	/*
2011 	 * Check if the checksum implementation is a fast accelerated one.
2012 	 * As-is this is a bit of a hack and should be replaced once the csum
2013 	 * implementations provide that information themselves.
2014 	 */
2015 	switch (csum_type) {
2016 	case BTRFS_CSUM_TYPE_CRC32:
2017 		if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2018 			set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2019 		break;
2020 	case BTRFS_CSUM_TYPE_XXHASH:
2021 		set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2022 		break;
2023 	default:
2024 		break;
2025 	}
2026 
2027 	btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2028 			btrfs_super_csum_name(csum_type),
2029 			crypto_shash_driver_name(csum_shash));
2030 	return 0;
2031 }
2032 
2033 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2034 			    struct btrfs_fs_devices *fs_devices)
2035 {
2036 	int ret;
2037 	struct btrfs_tree_parent_check check = { 0 };
2038 	struct btrfs_root *log_tree_root;
2039 	struct btrfs_super_block *disk_super = fs_info->super_copy;
2040 	u64 bytenr = btrfs_super_log_root(disk_super);
2041 	int level = btrfs_super_log_root_level(disk_super);
2042 
2043 	if (fs_devices->rw_devices == 0) {
2044 		btrfs_warn(fs_info, "log replay required on RO media");
2045 		return -EIO;
2046 	}
2047 
2048 	log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2049 					 GFP_KERNEL);
2050 	if (!log_tree_root)
2051 		return -ENOMEM;
2052 
2053 	check.level = level;
2054 	check.transid = fs_info->generation + 1;
2055 	check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2056 	log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2057 	if (IS_ERR(log_tree_root->node)) {
2058 		btrfs_warn(fs_info, "failed to read log tree");
2059 		ret = PTR_ERR(log_tree_root->node);
2060 		log_tree_root->node = NULL;
2061 		btrfs_put_root(log_tree_root);
2062 		return ret;
2063 	}
2064 	if (!extent_buffer_uptodate(log_tree_root->node)) {
2065 		btrfs_err(fs_info, "failed to read log tree");
2066 		btrfs_put_root(log_tree_root);
2067 		return -EIO;
2068 	}
2069 
2070 	/* returns with log_tree_root freed on success */
2071 	ret = btrfs_recover_log_trees(log_tree_root);
2072 	if (ret) {
2073 		btrfs_handle_fs_error(fs_info, ret,
2074 				      "Failed to recover log tree");
2075 		btrfs_put_root(log_tree_root);
2076 		return ret;
2077 	}
2078 
2079 	if (sb_rdonly(fs_info->sb)) {
2080 		ret = btrfs_commit_super(fs_info);
2081 		if (ret)
2082 			return ret;
2083 	}
2084 
2085 	return 0;
2086 }
2087 
2088 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2089 				      struct btrfs_path *path, u64 objectid,
2090 				      const char *name)
2091 {
2092 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
2093 	struct btrfs_root *root;
2094 	u64 max_global_id = 0;
2095 	int ret;
2096 	struct btrfs_key key = {
2097 		.objectid = objectid,
2098 		.type = BTRFS_ROOT_ITEM_KEY,
2099 		.offset = 0,
2100 	};
2101 	bool found = false;
2102 
2103 	/* If we have IGNOREDATACSUMS skip loading these roots. */
2104 	if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2105 	    btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2106 		set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2107 		return 0;
2108 	}
2109 
2110 	while (1) {
2111 		ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2112 		if (ret < 0)
2113 			break;
2114 
2115 		if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2116 			ret = btrfs_next_leaf(tree_root, path);
2117 			if (ret) {
2118 				if (ret > 0)
2119 					ret = 0;
2120 				break;
2121 			}
2122 		}
2123 		ret = 0;
2124 
2125 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2126 		if (key.objectid != objectid)
2127 			break;
2128 		btrfs_release_path(path);
2129 
2130 		/*
2131 		 * Just worry about this for extent tree, it'll be the same for
2132 		 * everybody.
2133 		 */
2134 		if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2135 			max_global_id = max(max_global_id, key.offset);
2136 
2137 		found = true;
2138 		root = read_tree_root_path(tree_root, path, &key);
2139 		if (IS_ERR(root)) {
2140 			if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2141 				ret = PTR_ERR(root);
2142 			break;
2143 		}
2144 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2145 		ret = btrfs_global_root_insert(root);
2146 		if (ret) {
2147 			btrfs_put_root(root);
2148 			break;
2149 		}
2150 		key.offset++;
2151 	}
2152 	btrfs_release_path(path);
2153 
2154 	if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2155 		fs_info->nr_global_roots = max_global_id + 1;
2156 
2157 	if (!found || ret) {
2158 		if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2159 			set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2160 
2161 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2162 			ret = ret ? ret : -ENOENT;
2163 		else
2164 			ret = 0;
2165 		btrfs_err(fs_info, "failed to load root %s", name);
2166 	}
2167 	return ret;
2168 }
2169 
2170 static int load_global_roots(struct btrfs_root *tree_root)
2171 {
2172 	struct btrfs_path *path;
2173 	int ret = 0;
2174 
2175 	path = btrfs_alloc_path();
2176 	if (!path)
2177 		return -ENOMEM;
2178 
2179 	ret = load_global_roots_objectid(tree_root, path,
2180 					 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2181 	if (ret)
2182 		goto out;
2183 	ret = load_global_roots_objectid(tree_root, path,
2184 					 BTRFS_CSUM_TREE_OBJECTID, "csum");
2185 	if (ret)
2186 		goto out;
2187 	if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2188 		goto out;
2189 	ret = load_global_roots_objectid(tree_root, path,
2190 					 BTRFS_FREE_SPACE_TREE_OBJECTID,
2191 					 "free space");
2192 out:
2193 	btrfs_free_path(path);
2194 	return ret;
2195 }
2196 
2197 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2198 {
2199 	struct btrfs_root *tree_root = fs_info->tree_root;
2200 	struct btrfs_root *root;
2201 	struct btrfs_key location;
2202 	int ret;
2203 
2204 	BUG_ON(!fs_info->tree_root);
2205 
2206 	ret = load_global_roots(tree_root);
2207 	if (ret)
2208 		return ret;
2209 
2210 	location.type = BTRFS_ROOT_ITEM_KEY;
2211 	location.offset = 0;
2212 
2213 	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2214 		location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2215 		root = btrfs_read_tree_root(tree_root, &location);
2216 		if (IS_ERR(root)) {
2217 			if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2218 				ret = PTR_ERR(root);
2219 				goto out;
2220 			}
2221 		} else {
2222 			set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2223 			fs_info->block_group_root = root;
2224 		}
2225 	}
2226 
2227 	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2228 	root = btrfs_read_tree_root(tree_root, &location);
2229 	if (IS_ERR(root)) {
2230 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2231 			ret = PTR_ERR(root);
2232 			goto out;
2233 		}
2234 	} else {
2235 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2236 		fs_info->dev_root = root;
2237 	}
2238 	/* Initialize fs_info for all devices in any case */
2239 	ret = btrfs_init_devices_late(fs_info);
2240 	if (ret)
2241 		goto out;
2242 
2243 	/*
2244 	 * This tree can share blocks with some other fs tree during relocation
2245 	 * and we need a proper setup by btrfs_get_fs_root
2246 	 */
2247 	root = btrfs_get_fs_root(tree_root->fs_info,
2248 				 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2249 	if (IS_ERR(root)) {
2250 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2251 			ret = PTR_ERR(root);
2252 			goto out;
2253 		}
2254 	} else {
2255 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2256 		fs_info->data_reloc_root = root;
2257 	}
2258 
2259 	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2260 	root = btrfs_read_tree_root(tree_root, &location);
2261 	if (!IS_ERR(root)) {
2262 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2263 		set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2264 		fs_info->quota_root = root;
2265 	}
2266 
2267 	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2268 	root = btrfs_read_tree_root(tree_root, &location);
2269 	if (IS_ERR(root)) {
2270 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2271 			ret = PTR_ERR(root);
2272 			if (ret != -ENOENT)
2273 				goto out;
2274 		}
2275 	} else {
2276 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2277 		fs_info->uuid_root = root;
2278 	}
2279 
2280 	return 0;
2281 out:
2282 	btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2283 		   location.objectid, ret);
2284 	return ret;
2285 }
2286 
2287 /*
2288  * Real super block validation
2289  * NOTE: super csum type and incompat features will not be checked here.
2290  *
2291  * @sb:		super block to check
2292  * @mirror_num:	the super block number to check its bytenr:
2293  * 		0	the primary (1st) sb
2294  * 		1, 2	2nd and 3rd backup copy
2295  * 	       -1	skip bytenr check
2296  */
2297 int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2298 			 struct btrfs_super_block *sb, int mirror_num)
2299 {
2300 	u64 nodesize = btrfs_super_nodesize(sb);
2301 	u64 sectorsize = btrfs_super_sectorsize(sb);
2302 	int ret = 0;
2303 
2304 	if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2305 		btrfs_err(fs_info, "no valid FS found");
2306 		ret = -EINVAL;
2307 	}
2308 	if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2309 		btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2310 				btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2311 		ret = -EINVAL;
2312 	}
2313 	if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2314 		btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2315 				btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2316 		ret = -EINVAL;
2317 	}
2318 	if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2319 		btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2320 				btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2321 		ret = -EINVAL;
2322 	}
2323 	if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2324 		btrfs_err(fs_info, "log_root level too big: %d >= %d",
2325 				btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2326 		ret = -EINVAL;
2327 	}
2328 
2329 	/*
2330 	 * Check sectorsize and nodesize first, other check will need it.
2331 	 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2332 	 */
2333 	if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2334 	    sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2335 		btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2336 		ret = -EINVAL;
2337 	}
2338 
2339 	/*
2340 	 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2341 	 *
2342 	 * We can support 16K sectorsize with 64K page size without problem,
2343 	 * but such sectorsize/pagesize combination doesn't make much sense.
2344 	 * 4K will be our future standard, PAGE_SIZE is supported from the very
2345 	 * beginning.
2346 	 */
2347 	if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2348 		btrfs_err(fs_info,
2349 			"sectorsize %llu not yet supported for page size %lu",
2350 			sectorsize, PAGE_SIZE);
2351 		ret = -EINVAL;
2352 	}
2353 
2354 	if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2355 	    nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2356 		btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2357 		ret = -EINVAL;
2358 	}
2359 	if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2360 		btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2361 			  le32_to_cpu(sb->__unused_leafsize), nodesize);
2362 		ret = -EINVAL;
2363 	}
2364 
2365 	/* Root alignment check */
2366 	if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2367 		btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2368 			   btrfs_super_root(sb));
2369 		ret = -EINVAL;
2370 	}
2371 	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2372 		btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2373 			   btrfs_super_chunk_root(sb));
2374 		ret = -EINVAL;
2375 	}
2376 	if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2377 		btrfs_warn(fs_info, "log_root block unaligned: %llu",
2378 			   btrfs_super_log_root(sb));
2379 		ret = -EINVAL;
2380 	}
2381 
2382 	if (memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2383 		btrfs_err(fs_info,
2384 		"superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2385 			  sb->fsid, fs_info->fs_devices->fsid);
2386 		ret = -EINVAL;
2387 	}
2388 
2389 	if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2390 		   BTRFS_FSID_SIZE) != 0) {
2391 		btrfs_err(fs_info,
2392 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2393 			  btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2394 		ret = -EINVAL;
2395 	}
2396 
2397 	if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2398 		   BTRFS_FSID_SIZE) != 0) {
2399 		btrfs_err(fs_info,
2400 			"dev_item UUID does not match metadata fsid: %pU != %pU",
2401 			fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2402 		ret = -EINVAL;
2403 	}
2404 
2405 	/*
2406 	 * Artificial requirement for block-group-tree to force newer features
2407 	 * (free-space-tree, no-holes) so the test matrix is smaller.
2408 	 */
2409 	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2410 	    (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2411 	     !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2412 		btrfs_err(fs_info,
2413 		"block-group-tree feature requires fres-space-tree and no-holes");
2414 		ret = -EINVAL;
2415 	}
2416 
2417 	/*
2418 	 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2419 	 * done later
2420 	 */
2421 	if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2422 		btrfs_err(fs_info, "bytes_used is too small %llu",
2423 			  btrfs_super_bytes_used(sb));
2424 		ret = -EINVAL;
2425 	}
2426 	if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2427 		btrfs_err(fs_info, "invalid stripesize %u",
2428 			  btrfs_super_stripesize(sb));
2429 		ret = -EINVAL;
2430 	}
2431 	if (btrfs_super_num_devices(sb) > (1UL << 31))
2432 		btrfs_warn(fs_info, "suspicious number of devices: %llu",
2433 			   btrfs_super_num_devices(sb));
2434 	if (btrfs_super_num_devices(sb) == 0) {
2435 		btrfs_err(fs_info, "number of devices is 0");
2436 		ret = -EINVAL;
2437 	}
2438 
2439 	if (mirror_num >= 0 &&
2440 	    btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2441 		btrfs_err(fs_info, "super offset mismatch %llu != %u",
2442 			  btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2443 		ret = -EINVAL;
2444 	}
2445 
2446 	/*
2447 	 * Obvious sys_chunk_array corruptions, it must hold at least one key
2448 	 * and one chunk
2449 	 */
2450 	if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2451 		btrfs_err(fs_info, "system chunk array too big %u > %u",
2452 			  btrfs_super_sys_array_size(sb),
2453 			  BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2454 		ret = -EINVAL;
2455 	}
2456 	if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2457 			+ sizeof(struct btrfs_chunk)) {
2458 		btrfs_err(fs_info, "system chunk array too small %u < %zu",
2459 			  btrfs_super_sys_array_size(sb),
2460 			  sizeof(struct btrfs_disk_key)
2461 			  + sizeof(struct btrfs_chunk));
2462 		ret = -EINVAL;
2463 	}
2464 
2465 	/*
2466 	 * The generation is a global counter, we'll trust it more than the others
2467 	 * but it's still possible that it's the one that's wrong.
2468 	 */
2469 	if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2470 		btrfs_warn(fs_info,
2471 			"suspicious: generation < chunk_root_generation: %llu < %llu",
2472 			btrfs_super_generation(sb),
2473 			btrfs_super_chunk_root_generation(sb));
2474 	if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2475 	    && btrfs_super_cache_generation(sb) != (u64)-1)
2476 		btrfs_warn(fs_info,
2477 			"suspicious: generation < cache_generation: %llu < %llu",
2478 			btrfs_super_generation(sb),
2479 			btrfs_super_cache_generation(sb));
2480 
2481 	return ret;
2482 }
2483 
2484 /*
2485  * Validation of super block at mount time.
2486  * Some checks already done early at mount time, like csum type and incompat
2487  * flags will be skipped.
2488  */
2489 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2490 {
2491 	return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2492 }
2493 
2494 /*
2495  * Validation of super block at write time.
2496  * Some checks like bytenr check will be skipped as their values will be
2497  * overwritten soon.
2498  * Extra checks like csum type and incompat flags will be done here.
2499  */
2500 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2501 				      struct btrfs_super_block *sb)
2502 {
2503 	int ret;
2504 
2505 	ret = btrfs_validate_super(fs_info, sb, -1);
2506 	if (ret < 0)
2507 		goto out;
2508 	if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2509 		ret = -EUCLEAN;
2510 		btrfs_err(fs_info, "invalid csum type, has %u want %u",
2511 			  btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2512 		goto out;
2513 	}
2514 	if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2515 		ret = -EUCLEAN;
2516 		btrfs_err(fs_info,
2517 		"invalid incompat flags, has 0x%llx valid mask 0x%llx",
2518 			  btrfs_super_incompat_flags(sb),
2519 			  (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2520 		goto out;
2521 	}
2522 out:
2523 	if (ret < 0)
2524 		btrfs_err(fs_info,
2525 		"super block corruption detected before writing it to disk");
2526 	return ret;
2527 }
2528 
2529 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2530 {
2531 	struct btrfs_tree_parent_check check = {
2532 		.level = level,
2533 		.transid = gen,
2534 		.owner_root = root->root_key.objectid
2535 	};
2536 	int ret = 0;
2537 
2538 	root->node = read_tree_block(root->fs_info, bytenr, &check);
2539 	if (IS_ERR(root->node)) {
2540 		ret = PTR_ERR(root->node);
2541 		root->node = NULL;
2542 		return ret;
2543 	}
2544 	if (!extent_buffer_uptodate(root->node)) {
2545 		free_extent_buffer(root->node);
2546 		root->node = NULL;
2547 		return -EIO;
2548 	}
2549 
2550 	btrfs_set_root_node(&root->root_item, root->node);
2551 	root->commit_root = btrfs_root_node(root);
2552 	btrfs_set_root_refs(&root->root_item, 1);
2553 	return ret;
2554 }
2555 
2556 static int load_important_roots(struct btrfs_fs_info *fs_info)
2557 {
2558 	struct btrfs_super_block *sb = fs_info->super_copy;
2559 	u64 gen, bytenr;
2560 	int level, ret;
2561 
2562 	bytenr = btrfs_super_root(sb);
2563 	gen = btrfs_super_generation(sb);
2564 	level = btrfs_super_root_level(sb);
2565 	ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2566 	if (ret) {
2567 		btrfs_warn(fs_info, "couldn't read tree root");
2568 		return ret;
2569 	}
2570 	return 0;
2571 }
2572 
2573 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2574 {
2575 	int backup_index = find_newest_super_backup(fs_info);
2576 	struct btrfs_super_block *sb = fs_info->super_copy;
2577 	struct btrfs_root *tree_root = fs_info->tree_root;
2578 	bool handle_error = false;
2579 	int ret = 0;
2580 	int i;
2581 
2582 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2583 		if (handle_error) {
2584 			if (!IS_ERR(tree_root->node))
2585 				free_extent_buffer(tree_root->node);
2586 			tree_root->node = NULL;
2587 
2588 			if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2589 				break;
2590 
2591 			free_root_pointers(fs_info, 0);
2592 
2593 			/*
2594 			 * Don't use the log in recovery mode, it won't be
2595 			 * valid
2596 			 */
2597 			btrfs_set_super_log_root(sb, 0);
2598 
2599 			/* We can't trust the free space cache either */
2600 			btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2601 
2602 			btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2603 			ret = read_backup_root(fs_info, i);
2604 			backup_index = ret;
2605 			if (ret < 0)
2606 				return ret;
2607 		}
2608 
2609 		ret = load_important_roots(fs_info);
2610 		if (ret) {
2611 			handle_error = true;
2612 			continue;
2613 		}
2614 
2615 		/*
2616 		 * No need to hold btrfs_root::objectid_mutex since the fs
2617 		 * hasn't been fully initialised and we are the only user
2618 		 */
2619 		ret = btrfs_init_root_free_objectid(tree_root);
2620 		if (ret < 0) {
2621 			handle_error = true;
2622 			continue;
2623 		}
2624 
2625 		ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2626 
2627 		ret = btrfs_read_roots(fs_info);
2628 		if (ret < 0) {
2629 			handle_error = true;
2630 			continue;
2631 		}
2632 
2633 		/* All successful */
2634 		fs_info->generation = btrfs_header_generation(tree_root->node);
2635 		fs_info->last_trans_committed = fs_info->generation;
2636 		fs_info->last_reloc_trans = 0;
2637 
2638 		/* Always begin writing backup roots after the one being used */
2639 		if (backup_index < 0) {
2640 			fs_info->backup_root_index = 0;
2641 		} else {
2642 			fs_info->backup_root_index = backup_index + 1;
2643 			fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2644 		}
2645 		break;
2646 	}
2647 
2648 	return ret;
2649 }
2650 
2651 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2652 {
2653 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2654 	INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2655 	INIT_LIST_HEAD(&fs_info->trans_list);
2656 	INIT_LIST_HEAD(&fs_info->dead_roots);
2657 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
2658 	INIT_LIST_HEAD(&fs_info->delalloc_roots);
2659 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
2660 	spin_lock_init(&fs_info->delalloc_root_lock);
2661 	spin_lock_init(&fs_info->trans_lock);
2662 	spin_lock_init(&fs_info->fs_roots_radix_lock);
2663 	spin_lock_init(&fs_info->delayed_iput_lock);
2664 	spin_lock_init(&fs_info->defrag_inodes_lock);
2665 	spin_lock_init(&fs_info->super_lock);
2666 	spin_lock_init(&fs_info->buffer_lock);
2667 	spin_lock_init(&fs_info->unused_bgs_lock);
2668 	spin_lock_init(&fs_info->treelog_bg_lock);
2669 	spin_lock_init(&fs_info->zone_active_bgs_lock);
2670 	spin_lock_init(&fs_info->relocation_bg_lock);
2671 	rwlock_init(&fs_info->tree_mod_log_lock);
2672 	rwlock_init(&fs_info->global_root_lock);
2673 	mutex_init(&fs_info->unused_bg_unpin_mutex);
2674 	mutex_init(&fs_info->reclaim_bgs_lock);
2675 	mutex_init(&fs_info->reloc_mutex);
2676 	mutex_init(&fs_info->delalloc_root_mutex);
2677 	mutex_init(&fs_info->zoned_meta_io_lock);
2678 	mutex_init(&fs_info->zoned_data_reloc_io_lock);
2679 	seqlock_init(&fs_info->profiles_lock);
2680 
2681 	btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2682 	btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2683 	btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2684 	btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2685 	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_start,
2686 				     BTRFS_LOCKDEP_TRANS_COMMIT_START);
2687 	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2688 				     BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2689 	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2690 				     BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2691 	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2692 				     BTRFS_LOCKDEP_TRANS_COMPLETED);
2693 
2694 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2695 	INIT_LIST_HEAD(&fs_info->space_info);
2696 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2697 	INIT_LIST_HEAD(&fs_info->unused_bgs);
2698 	INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2699 	INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2700 #ifdef CONFIG_BTRFS_DEBUG
2701 	INIT_LIST_HEAD(&fs_info->allocated_roots);
2702 	INIT_LIST_HEAD(&fs_info->allocated_ebs);
2703 	spin_lock_init(&fs_info->eb_leak_lock);
2704 #endif
2705 	extent_map_tree_init(&fs_info->mapping_tree);
2706 	btrfs_init_block_rsv(&fs_info->global_block_rsv,
2707 			     BTRFS_BLOCK_RSV_GLOBAL);
2708 	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2709 	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2710 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2711 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2712 			     BTRFS_BLOCK_RSV_DELOPS);
2713 	btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2714 			     BTRFS_BLOCK_RSV_DELREFS);
2715 
2716 	atomic_set(&fs_info->async_delalloc_pages, 0);
2717 	atomic_set(&fs_info->defrag_running, 0);
2718 	atomic_set(&fs_info->nr_delayed_iputs, 0);
2719 	atomic64_set(&fs_info->tree_mod_seq, 0);
2720 	fs_info->global_root_tree = RB_ROOT;
2721 	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2722 	fs_info->metadata_ratio = 0;
2723 	fs_info->defrag_inodes = RB_ROOT;
2724 	atomic64_set(&fs_info->free_chunk_space, 0);
2725 	fs_info->tree_mod_log = RB_ROOT;
2726 	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2727 	btrfs_init_ref_verify(fs_info);
2728 
2729 	fs_info->thread_pool_size = min_t(unsigned long,
2730 					  num_online_cpus() + 2, 8);
2731 
2732 	INIT_LIST_HEAD(&fs_info->ordered_roots);
2733 	spin_lock_init(&fs_info->ordered_root_lock);
2734 
2735 	btrfs_init_scrub(fs_info);
2736 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2737 	fs_info->check_integrity_print_mask = 0;
2738 #endif
2739 	btrfs_init_balance(fs_info);
2740 	btrfs_init_async_reclaim_work(fs_info);
2741 
2742 	rwlock_init(&fs_info->block_group_cache_lock);
2743 	fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2744 
2745 	extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2746 			    IO_TREE_FS_EXCLUDED_EXTENTS);
2747 
2748 	mutex_init(&fs_info->ordered_operations_mutex);
2749 	mutex_init(&fs_info->tree_log_mutex);
2750 	mutex_init(&fs_info->chunk_mutex);
2751 	mutex_init(&fs_info->transaction_kthread_mutex);
2752 	mutex_init(&fs_info->cleaner_mutex);
2753 	mutex_init(&fs_info->ro_block_group_mutex);
2754 	init_rwsem(&fs_info->commit_root_sem);
2755 	init_rwsem(&fs_info->cleanup_work_sem);
2756 	init_rwsem(&fs_info->subvol_sem);
2757 	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2758 
2759 	btrfs_init_dev_replace_locks(fs_info);
2760 	btrfs_init_qgroup(fs_info);
2761 	btrfs_discard_init(fs_info);
2762 
2763 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2764 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2765 
2766 	init_waitqueue_head(&fs_info->transaction_throttle);
2767 	init_waitqueue_head(&fs_info->transaction_wait);
2768 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2769 	init_waitqueue_head(&fs_info->async_submit_wait);
2770 	init_waitqueue_head(&fs_info->delayed_iputs_wait);
2771 
2772 	/* Usable values until the real ones are cached from the superblock */
2773 	fs_info->nodesize = 4096;
2774 	fs_info->sectorsize = 4096;
2775 	fs_info->sectorsize_bits = ilog2(4096);
2776 	fs_info->stripesize = 4096;
2777 
2778 	fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2779 
2780 	spin_lock_init(&fs_info->swapfile_pins_lock);
2781 	fs_info->swapfile_pins = RB_ROOT;
2782 
2783 	fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2784 	INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2785 }
2786 
2787 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2788 {
2789 	int ret;
2790 
2791 	fs_info->sb = sb;
2792 	sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2793 	sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2794 
2795 	ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2796 	if (ret)
2797 		return ret;
2798 
2799 	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2800 	if (ret)
2801 		return ret;
2802 
2803 	fs_info->dirty_metadata_batch = PAGE_SIZE *
2804 					(1 + ilog2(nr_cpu_ids));
2805 
2806 	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2807 	if (ret)
2808 		return ret;
2809 
2810 	ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2811 			GFP_KERNEL);
2812 	if (ret)
2813 		return ret;
2814 
2815 	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2816 					GFP_KERNEL);
2817 	if (!fs_info->delayed_root)
2818 		return -ENOMEM;
2819 	btrfs_init_delayed_root(fs_info->delayed_root);
2820 
2821 	if (sb_rdonly(sb))
2822 		set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2823 
2824 	return btrfs_alloc_stripe_hash_table(fs_info);
2825 }
2826 
2827 static int btrfs_uuid_rescan_kthread(void *data)
2828 {
2829 	struct btrfs_fs_info *fs_info = data;
2830 	int ret;
2831 
2832 	/*
2833 	 * 1st step is to iterate through the existing UUID tree and
2834 	 * to delete all entries that contain outdated data.
2835 	 * 2nd step is to add all missing entries to the UUID tree.
2836 	 */
2837 	ret = btrfs_uuid_tree_iterate(fs_info);
2838 	if (ret < 0) {
2839 		if (ret != -EINTR)
2840 			btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2841 				   ret);
2842 		up(&fs_info->uuid_tree_rescan_sem);
2843 		return ret;
2844 	}
2845 	return btrfs_uuid_scan_kthread(data);
2846 }
2847 
2848 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2849 {
2850 	struct task_struct *task;
2851 
2852 	down(&fs_info->uuid_tree_rescan_sem);
2853 	task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2854 	if (IS_ERR(task)) {
2855 		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
2856 		btrfs_warn(fs_info, "failed to start uuid_rescan task");
2857 		up(&fs_info->uuid_tree_rescan_sem);
2858 		return PTR_ERR(task);
2859 	}
2860 
2861 	return 0;
2862 }
2863 
2864 static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2865 {
2866 	u64 root_objectid = 0;
2867 	struct btrfs_root *gang[8];
2868 	int i = 0;
2869 	int err = 0;
2870 	unsigned int ret = 0;
2871 
2872 	while (1) {
2873 		spin_lock(&fs_info->fs_roots_radix_lock);
2874 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2875 					     (void **)gang, root_objectid,
2876 					     ARRAY_SIZE(gang));
2877 		if (!ret) {
2878 			spin_unlock(&fs_info->fs_roots_radix_lock);
2879 			break;
2880 		}
2881 		root_objectid = gang[ret - 1]->root_key.objectid + 1;
2882 
2883 		for (i = 0; i < ret; i++) {
2884 			/* Avoid to grab roots in dead_roots. */
2885 			if (btrfs_root_refs(&gang[i]->root_item) == 0) {
2886 				gang[i] = NULL;
2887 				continue;
2888 			}
2889 			/* Grab all the search result for later use. */
2890 			gang[i] = btrfs_grab_root(gang[i]);
2891 		}
2892 		spin_unlock(&fs_info->fs_roots_radix_lock);
2893 
2894 		for (i = 0; i < ret; i++) {
2895 			if (!gang[i])
2896 				continue;
2897 			root_objectid = gang[i]->root_key.objectid;
2898 			err = btrfs_orphan_cleanup(gang[i]);
2899 			if (err)
2900 				goto out;
2901 			btrfs_put_root(gang[i]);
2902 		}
2903 		root_objectid++;
2904 	}
2905 out:
2906 	/* Release the uncleaned roots due to error. */
2907 	for (; i < ret; i++) {
2908 		if (gang[i])
2909 			btrfs_put_root(gang[i]);
2910 	}
2911 	return err;
2912 }
2913 
2914 /*
2915  * Some options only have meaning at mount time and shouldn't persist across
2916  * remounts, or be displayed. Clear these at the end of mount and remount
2917  * code paths.
2918  */
2919 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
2920 {
2921 	btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
2922 	btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
2923 }
2924 
2925 /*
2926  * Mounting logic specific to read-write file systems. Shared by open_ctree
2927  * and btrfs_remount when remounting from read-only to read-write.
2928  */
2929 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2930 {
2931 	int ret;
2932 	const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2933 	bool rebuild_free_space_tree = false;
2934 
2935 	if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2936 	    btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2937 		rebuild_free_space_tree = true;
2938 	} else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2939 		   !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2940 		btrfs_warn(fs_info, "free space tree is invalid");
2941 		rebuild_free_space_tree = true;
2942 	}
2943 
2944 	if (rebuild_free_space_tree) {
2945 		btrfs_info(fs_info, "rebuilding free space tree");
2946 		ret = btrfs_rebuild_free_space_tree(fs_info);
2947 		if (ret) {
2948 			btrfs_warn(fs_info,
2949 				   "failed to rebuild free space tree: %d", ret);
2950 			goto out;
2951 		}
2952 	}
2953 
2954 	if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2955 	    !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
2956 		btrfs_info(fs_info, "disabling free space tree");
2957 		ret = btrfs_delete_free_space_tree(fs_info);
2958 		if (ret) {
2959 			btrfs_warn(fs_info,
2960 				   "failed to disable free space tree: %d", ret);
2961 			goto out;
2962 		}
2963 	}
2964 
2965 	/*
2966 	 * btrfs_find_orphan_roots() is responsible for finding all the dead
2967 	 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
2968 	 * them into the fs_info->fs_roots_radix tree. This must be done before
2969 	 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
2970 	 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
2971 	 * item before the root's tree is deleted - this means that if we unmount
2972 	 * or crash before the deletion completes, on the next mount we will not
2973 	 * delete what remains of the tree because the orphan item does not
2974 	 * exists anymore, which is what tells us we have a pending deletion.
2975 	 */
2976 	ret = btrfs_find_orphan_roots(fs_info);
2977 	if (ret)
2978 		goto out;
2979 
2980 	ret = btrfs_cleanup_fs_roots(fs_info);
2981 	if (ret)
2982 		goto out;
2983 
2984 	down_read(&fs_info->cleanup_work_sem);
2985 	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2986 	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2987 		up_read(&fs_info->cleanup_work_sem);
2988 		goto out;
2989 	}
2990 	up_read(&fs_info->cleanup_work_sem);
2991 
2992 	mutex_lock(&fs_info->cleaner_mutex);
2993 	ret = btrfs_recover_relocation(fs_info);
2994 	mutex_unlock(&fs_info->cleaner_mutex);
2995 	if (ret < 0) {
2996 		btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
2997 		goto out;
2998 	}
2999 
3000 	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3001 	    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3002 		btrfs_info(fs_info, "creating free space tree");
3003 		ret = btrfs_create_free_space_tree(fs_info);
3004 		if (ret) {
3005 			btrfs_warn(fs_info,
3006 				"failed to create free space tree: %d", ret);
3007 			goto out;
3008 		}
3009 	}
3010 
3011 	if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3012 		ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3013 		if (ret)
3014 			goto out;
3015 	}
3016 
3017 	ret = btrfs_resume_balance_async(fs_info);
3018 	if (ret)
3019 		goto out;
3020 
3021 	ret = btrfs_resume_dev_replace_async(fs_info);
3022 	if (ret) {
3023 		btrfs_warn(fs_info, "failed to resume dev_replace");
3024 		goto out;
3025 	}
3026 
3027 	btrfs_qgroup_rescan_resume(fs_info);
3028 
3029 	if (!fs_info->uuid_root) {
3030 		btrfs_info(fs_info, "creating UUID tree");
3031 		ret = btrfs_create_uuid_tree(fs_info);
3032 		if (ret) {
3033 			btrfs_warn(fs_info,
3034 				   "failed to create the UUID tree %d", ret);
3035 			goto out;
3036 		}
3037 	}
3038 
3039 out:
3040 	return ret;
3041 }
3042 
3043 /*
3044  * Do various sanity and dependency checks of different features.
3045  *
3046  * @is_rw_mount:	If the mount is read-write.
3047  *
3048  * This is the place for less strict checks (like for subpage or artificial
3049  * feature dependencies).
3050  *
3051  * For strict checks or possible corruption detection, see
3052  * btrfs_validate_super().
3053  *
3054  * This should be called after btrfs_parse_options(), as some mount options
3055  * (space cache related) can modify on-disk format like free space tree and
3056  * screw up certain feature dependencies.
3057  */
3058 int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3059 {
3060 	struct btrfs_super_block *disk_super = fs_info->super_copy;
3061 	u64 incompat = btrfs_super_incompat_flags(disk_super);
3062 	const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3063 	const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3064 
3065 	if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3066 		btrfs_err(fs_info,
3067 		"cannot mount because of unknown incompat features (0x%llx)",
3068 		    incompat);
3069 		return -EINVAL;
3070 	}
3071 
3072 	/* Runtime limitation for mixed block groups. */
3073 	if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3074 	    (fs_info->sectorsize != fs_info->nodesize)) {
3075 		btrfs_err(fs_info,
3076 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3077 			fs_info->nodesize, fs_info->sectorsize);
3078 		return -EINVAL;
3079 	}
3080 
3081 	/* Mixed backref is an always-enabled feature. */
3082 	incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3083 
3084 	/* Set compression related flags just in case. */
3085 	if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3086 		incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3087 	else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3088 		incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3089 
3090 	/*
3091 	 * An ancient flag, which should really be marked deprecated.
3092 	 * Such runtime limitation doesn't really need a incompat flag.
3093 	 */
3094 	if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3095 		incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3096 
3097 	if (compat_ro_unsupp && is_rw_mount) {
3098 		btrfs_err(fs_info,
3099 	"cannot mount read-write because of unknown compat_ro features (0x%llx)",
3100 		       compat_ro);
3101 		return -EINVAL;
3102 	}
3103 
3104 	/*
3105 	 * We have unsupported RO compat features, although RO mounted, we
3106 	 * should not cause any metadata writes, including log replay.
3107 	 * Or we could screw up whatever the new feature requires.
3108 	 */
3109 	if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3110 	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3111 		btrfs_err(fs_info,
3112 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3113 			  compat_ro);
3114 		return -EINVAL;
3115 	}
3116 
3117 	/*
3118 	 * Artificial limitations for block group tree, to force
3119 	 * block-group-tree to rely on no-holes and free-space-tree.
3120 	 */
3121 	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3122 	    (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3123 	     !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3124 		btrfs_err(fs_info,
3125 "block-group-tree feature requires no-holes and free-space-tree features");
3126 		return -EINVAL;
3127 	}
3128 
3129 	/*
3130 	 * Subpage runtime limitation on v1 cache.
3131 	 *
3132 	 * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3133 	 * we're already defaulting to v2 cache, no need to bother v1 as it's
3134 	 * going to be deprecated anyway.
3135 	 */
3136 	if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3137 		btrfs_warn(fs_info,
3138 	"v1 space cache is not supported for page size %lu with sectorsize %u",
3139 			   PAGE_SIZE, fs_info->sectorsize);
3140 		return -EINVAL;
3141 	}
3142 
3143 	/* This can be called by remount, we need to protect the super block. */
3144 	spin_lock(&fs_info->super_lock);
3145 	btrfs_set_super_incompat_flags(disk_super, incompat);
3146 	spin_unlock(&fs_info->super_lock);
3147 
3148 	return 0;
3149 }
3150 
3151 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3152 		      char *options)
3153 {
3154 	u32 sectorsize;
3155 	u32 nodesize;
3156 	u32 stripesize;
3157 	u64 generation;
3158 	u64 features;
3159 	u16 csum_type;
3160 	struct btrfs_super_block *disk_super;
3161 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3162 	struct btrfs_root *tree_root;
3163 	struct btrfs_root *chunk_root;
3164 	int ret;
3165 	int level;
3166 
3167 	ret = init_mount_fs_info(fs_info, sb);
3168 	if (ret)
3169 		goto fail;
3170 
3171 	/* These need to be init'ed before we start creating inodes and such. */
3172 	tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3173 				     GFP_KERNEL);
3174 	fs_info->tree_root = tree_root;
3175 	chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3176 				      GFP_KERNEL);
3177 	fs_info->chunk_root = chunk_root;
3178 	if (!tree_root || !chunk_root) {
3179 		ret = -ENOMEM;
3180 		goto fail;
3181 	}
3182 
3183 	ret = btrfs_init_btree_inode(sb);
3184 	if (ret)
3185 		goto fail;
3186 
3187 	invalidate_bdev(fs_devices->latest_dev->bdev);
3188 
3189 	/*
3190 	 * Read super block and check the signature bytes only
3191 	 */
3192 	disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3193 	if (IS_ERR(disk_super)) {
3194 		ret = PTR_ERR(disk_super);
3195 		goto fail_alloc;
3196 	}
3197 
3198 	/*
3199 	 * Verify the type first, if that or the checksum value are
3200 	 * corrupted, we'll find out
3201 	 */
3202 	csum_type = btrfs_super_csum_type(disk_super);
3203 	if (!btrfs_supported_super_csum(csum_type)) {
3204 		btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3205 			  csum_type);
3206 		ret = -EINVAL;
3207 		btrfs_release_disk_super(disk_super);
3208 		goto fail_alloc;
3209 	}
3210 
3211 	fs_info->csum_size = btrfs_super_csum_size(disk_super);
3212 
3213 	ret = btrfs_init_csum_hash(fs_info, csum_type);
3214 	if (ret) {
3215 		btrfs_release_disk_super(disk_super);
3216 		goto fail_alloc;
3217 	}
3218 
3219 	/*
3220 	 * We want to check superblock checksum, the type is stored inside.
3221 	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3222 	 */
3223 	if (btrfs_check_super_csum(fs_info, disk_super)) {
3224 		btrfs_err(fs_info, "superblock checksum mismatch");
3225 		ret = -EINVAL;
3226 		btrfs_release_disk_super(disk_super);
3227 		goto fail_alloc;
3228 	}
3229 
3230 	/*
3231 	 * super_copy is zeroed at allocation time and we never touch the
3232 	 * following bytes up to INFO_SIZE, the checksum is calculated from
3233 	 * the whole block of INFO_SIZE
3234 	 */
3235 	memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3236 	btrfs_release_disk_super(disk_super);
3237 
3238 	disk_super = fs_info->super_copy;
3239 
3240 
3241 	features = btrfs_super_flags(disk_super);
3242 	if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3243 		features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3244 		btrfs_set_super_flags(disk_super, features);
3245 		btrfs_info(fs_info,
3246 			"found metadata UUID change in progress flag, clearing");
3247 	}
3248 
3249 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
3250 	       sizeof(*fs_info->super_for_commit));
3251 
3252 	ret = btrfs_validate_mount_super(fs_info);
3253 	if (ret) {
3254 		btrfs_err(fs_info, "superblock contains fatal errors");
3255 		ret = -EINVAL;
3256 		goto fail_alloc;
3257 	}
3258 
3259 	if (!btrfs_super_root(disk_super)) {
3260 		btrfs_err(fs_info, "invalid superblock tree root bytenr");
3261 		ret = -EINVAL;
3262 		goto fail_alloc;
3263 	}
3264 
3265 	/* check FS state, whether FS is broken. */
3266 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3267 		WRITE_ONCE(fs_info->fs_error, -EUCLEAN);
3268 
3269 	/*
3270 	 * In the long term, we'll store the compression type in the super
3271 	 * block, and it'll be used for per file compression control.
3272 	 */
3273 	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3274 
3275 
3276 	/* Set up fs_info before parsing mount options */
3277 	nodesize = btrfs_super_nodesize(disk_super);
3278 	sectorsize = btrfs_super_sectorsize(disk_super);
3279 	stripesize = sectorsize;
3280 	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3281 	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3282 
3283 	fs_info->nodesize = nodesize;
3284 	fs_info->sectorsize = sectorsize;
3285 	fs_info->sectorsize_bits = ilog2(sectorsize);
3286 	fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3287 	fs_info->stripesize = stripesize;
3288 
3289 	ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3290 	if (ret)
3291 		goto fail_alloc;
3292 
3293 	ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3294 	if (ret < 0)
3295 		goto fail_alloc;
3296 
3297 	if (sectorsize < PAGE_SIZE) {
3298 		struct btrfs_subpage_info *subpage_info;
3299 
3300 		/*
3301 		 * V1 space cache has some hardcoded PAGE_SIZE usage, and is
3302 		 * going to be deprecated.
3303 		 *
3304 		 * Force to use v2 cache for subpage case.
3305 		 */
3306 		btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
3307 		btrfs_set_and_info(fs_info, FREE_SPACE_TREE,
3308 			"forcing free space tree for sector size %u with page size %lu",
3309 			sectorsize, PAGE_SIZE);
3310 
3311 		btrfs_warn(fs_info,
3312 		"read-write for sector size %u with page size %lu is experimental",
3313 			   sectorsize, PAGE_SIZE);
3314 		subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3315 		if (!subpage_info) {
3316 			ret = -ENOMEM;
3317 			goto fail_alloc;
3318 		}
3319 		btrfs_init_subpage_info(subpage_info, sectorsize);
3320 		fs_info->subpage_info = subpage_info;
3321 	}
3322 
3323 	ret = btrfs_init_workqueues(fs_info);
3324 	if (ret)
3325 		goto fail_sb_buffer;
3326 
3327 	sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3328 	sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3329 
3330 	sb->s_blocksize = sectorsize;
3331 	sb->s_blocksize_bits = blksize_bits(sectorsize);
3332 	memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3333 
3334 	mutex_lock(&fs_info->chunk_mutex);
3335 	ret = btrfs_read_sys_array(fs_info);
3336 	mutex_unlock(&fs_info->chunk_mutex);
3337 	if (ret) {
3338 		btrfs_err(fs_info, "failed to read the system array: %d", ret);
3339 		goto fail_sb_buffer;
3340 	}
3341 
3342 	generation = btrfs_super_chunk_root_generation(disk_super);
3343 	level = btrfs_super_chunk_root_level(disk_super);
3344 	ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3345 			      generation, level);
3346 	if (ret) {
3347 		btrfs_err(fs_info, "failed to read chunk root");
3348 		goto fail_tree_roots;
3349 	}
3350 
3351 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3352 			   offsetof(struct btrfs_header, chunk_tree_uuid),
3353 			   BTRFS_UUID_SIZE);
3354 
3355 	ret = btrfs_read_chunk_tree(fs_info);
3356 	if (ret) {
3357 		btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3358 		goto fail_tree_roots;
3359 	}
3360 
3361 	/*
3362 	 * At this point we know all the devices that make this filesystem,
3363 	 * including the seed devices but we don't know yet if the replace
3364 	 * target is required. So free devices that are not part of this
3365 	 * filesystem but skip the replace target device which is checked
3366 	 * below in btrfs_init_dev_replace().
3367 	 */
3368 	btrfs_free_extra_devids(fs_devices);
3369 	if (!fs_devices->latest_dev->bdev) {
3370 		btrfs_err(fs_info, "failed to read devices");
3371 		ret = -EIO;
3372 		goto fail_tree_roots;
3373 	}
3374 
3375 	ret = init_tree_roots(fs_info);
3376 	if (ret)
3377 		goto fail_tree_roots;
3378 
3379 	/*
3380 	 * Get zone type information of zoned block devices. This will also
3381 	 * handle emulation of a zoned filesystem if a regular device has the
3382 	 * zoned incompat feature flag set.
3383 	 */
3384 	ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3385 	if (ret) {
3386 		btrfs_err(fs_info,
3387 			  "zoned: failed to read device zone info: %d", ret);
3388 		goto fail_block_groups;
3389 	}
3390 
3391 	/*
3392 	 * If we have a uuid root and we're not being told to rescan we need to
3393 	 * check the generation here so we can set the
3394 	 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3395 	 * transaction during a balance or the log replay without updating the
3396 	 * uuid generation, and then if we crash we would rescan the uuid tree,
3397 	 * even though it was perfectly fine.
3398 	 */
3399 	if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3400 	    fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3401 		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3402 
3403 	ret = btrfs_verify_dev_extents(fs_info);
3404 	if (ret) {
3405 		btrfs_err(fs_info,
3406 			  "failed to verify dev extents against chunks: %d",
3407 			  ret);
3408 		goto fail_block_groups;
3409 	}
3410 	ret = btrfs_recover_balance(fs_info);
3411 	if (ret) {
3412 		btrfs_err(fs_info, "failed to recover balance: %d", ret);
3413 		goto fail_block_groups;
3414 	}
3415 
3416 	ret = btrfs_init_dev_stats(fs_info);
3417 	if (ret) {
3418 		btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3419 		goto fail_block_groups;
3420 	}
3421 
3422 	ret = btrfs_init_dev_replace(fs_info);
3423 	if (ret) {
3424 		btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3425 		goto fail_block_groups;
3426 	}
3427 
3428 	ret = btrfs_check_zoned_mode(fs_info);
3429 	if (ret) {
3430 		btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3431 			  ret);
3432 		goto fail_block_groups;
3433 	}
3434 
3435 	ret = btrfs_sysfs_add_fsid(fs_devices);
3436 	if (ret) {
3437 		btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3438 				ret);
3439 		goto fail_block_groups;
3440 	}
3441 
3442 	ret = btrfs_sysfs_add_mounted(fs_info);
3443 	if (ret) {
3444 		btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3445 		goto fail_fsdev_sysfs;
3446 	}
3447 
3448 	ret = btrfs_init_space_info(fs_info);
3449 	if (ret) {
3450 		btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3451 		goto fail_sysfs;
3452 	}
3453 
3454 	ret = btrfs_read_block_groups(fs_info);
3455 	if (ret) {
3456 		btrfs_err(fs_info, "failed to read block groups: %d", ret);
3457 		goto fail_sysfs;
3458 	}
3459 
3460 	btrfs_free_zone_cache(fs_info);
3461 
3462 	btrfs_check_active_zone_reservation(fs_info);
3463 
3464 	if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3465 	    !btrfs_check_rw_degradable(fs_info, NULL)) {
3466 		btrfs_warn(fs_info,
3467 		"writable mount is not allowed due to too many missing devices");
3468 		ret = -EINVAL;
3469 		goto fail_sysfs;
3470 	}
3471 
3472 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3473 					       "btrfs-cleaner");
3474 	if (IS_ERR(fs_info->cleaner_kthread)) {
3475 		ret = PTR_ERR(fs_info->cleaner_kthread);
3476 		goto fail_sysfs;
3477 	}
3478 
3479 	fs_info->transaction_kthread = kthread_run(transaction_kthread,
3480 						   tree_root,
3481 						   "btrfs-transaction");
3482 	if (IS_ERR(fs_info->transaction_kthread)) {
3483 		ret = PTR_ERR(fs_info->transaction_kthread);
3484 		goto fail_cleaner;
3485 	}
3486 
3487 	if (!btrfs_test_opt(fs_info, NOSSD) &&
3488 	    !fs_info->fs_devices->rotating) {
3489 		btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3490 	}
3491 
3492 	/*
3493 	 * For devices supporting discard turn on discard=async automatically,
3494 	 * unless it's already set or disabled. This could be turned off by
3495 	 * nodiscard for the same mount.
3496 	 *
3497 	 * The zoned mode piggy backs on the discard functionality for
3498 	 * resetting a zone. There is no reason to delay the zone reset as it is
3499 	 * fast enough. So, do not enable async discard for zoned mode.
3500 	 */
3501 	if (!(btrfs_test_opt(fs_info, DISCARD_SYNC) ||
3502 	      btrfs_test_opt(fs_info, DISCARD_ASYNC) ||
3503 	      btrfs_test_opt(fs_info, NODISCARD)) &&
3504 	    fs_info->fs_devices->discardable &&
3505 	    !btrfs_is_zoned(fs_info)) {
3506 		btrfs_set_and_info(fs_info, DISCARD_ASYNC,
3507 				   "auto enabling async discard");
3508 	}
3509 
3510 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3511 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3512 		ret = btrfsic_mount(fs_info, fs_devices,
3513 				    btrfs_test_opt(fs_info,
3514 					CHECK_INTEGRITY_DATA) ? 1 : 0,
3515 				    fs_info->check_integrity_print_mask);
3516 		if (ret)
3517 			btrfs_warn(fs_info,
3518 				"failed to initialize integrity check module: %d",
3519 				ret);
3520 	}
3521 #endif
3522 	ret = btrfs_read_qgroup_config(fs_info);
3523 	if (ret)
3524 		goto fail_trans_kthread;
3525 
3526 	if (btrfs_build_ref_tree(fs_info))
3527 		btrfs_err(fs_info, "couldn't build ref tree");
3528 
3529 	/* do not make disk changes in broken FS or nologreplay is given */
3530 	if (btrfs_super_log_root(disk_super) != 0 &&
3531 	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3532 		btrfs_info(fs_info, "start tree-log replay");
3533 		ret = btrfs_replay_log(fs_info, fs_devices);
3534 		if (ret)
3535 			goto fail_qgroup;
3536 	}
3537 
3538 	fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3539 	if (IS_ERR(fs_info->fs_root)) {
3540 		ret = PTR_ERR(fs_info->fs_root);
3541 		btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3542 		fs_info->fs_root = NULL;
3543 		goto fail_qgroup;
3544 	}
3545 
3546 	if (sb_rdonly(sb))
3547 		goto clear_oneshot;
3548 
3549 	ret = btrfs_start_pre_rw_mount(fs_info);
3550 	if (ret) {
3551 		close_ctree(fs_info);
3552 		return ret;
3553 	}
3554 	btrfs_discard_resume(fs_info);
3555 
3556 	if (fs_info->uuid_root &&
3557 	    (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3558 	     fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3559 		btrfs_info(fs_info, "checking UUID tree");
3560 		ret = btrfs_check_uuid_tree(fs_info);
3561 		if (ret) {
3562 			btrfs_warn(fs_info,
3563 				"failed to check the UUID tree: %d", ret);
3564 			close_ctree(fs_info);
3565 			return ret;
3566 		}
3567 	}
3568 
3569 	set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3570 
3571 	/* Kick the cleaner thread so it'll start deleting snapshots. */
3572 	if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3573 		wake_up_process(fs_info->cleaner_kthread);
3574 
3575 clear_oneshot:
3576 	btrfs_clear_oneshot_options(fs_info);
3577 	return 0;
3578 
3579 fail_qgroup:
3580 	btrfs_free_qgroup_config(fs_info);
3581 fail_trans_kthread:
3582 	kthread_stop(fs_info->transaction_kthread);
3583 	btrfs_cleanup_transaction(fs_info);
3584 	btrfs_free_fs_roots(fs_info);
3585 fail_cleaner:
3586 	kthread_stop(fs_info->cleaner_kthread);
3587 
3588 	/*
3589 	 * make sure we're done with the btree inode before we stop our
3590 	 * kthreads
3591 	 */
3592 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3593 
3594 fail_sysfs:
3595 	btrfs_sysfs_remove_mounted(fs_info);
3596 
3597 fail_fsdev_sysfs:
3598 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3599 
3600 fail_block_groups:
3601 	btrfs_put_block_group_cache(fs_info);
3602 
3603 fail_tree_roots:
3604 	if (fs_info->data_reloc_root)
3605 		btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3606 	free_root_pointers(fs_info, true);
3607 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3608 
3609 fail_sb_buffer:
3610 	btrfs_stop_all_workers(fs_info);
3611 	btrfs_free_block_groups(fs_info);
3612 fail_alloc:
3613 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3614 
3615 	iput(fs_info->btree_inode);
3616 fail:
3617 	btrfs_close_devices(fs_info->fs_devices);
3618 	ASSERT(ret < 0);
3619 	return ret;
3620 }
3621 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3622 
3623 static void btrfs_end_super_write(struct bio *bio)
3624 {
3625 	struct btrfs_device *device = bio->bi_private;
3626 	struct bio_vec *bvec;
3627 	struct bvec_iter_all iter_all;
3628 	struct page *page;
3629 
3630 	bio_for_each_segment_all(bvec, bio, iter_all) {
3631 		page = bvec->bv_page;
3632 
3633 		if (bio->bi_status) {
3634 			btrfs_warn_rl_in_rcu(device->fs_info,
3635 				"lost page write due to IO error on %s (%d)",
3636 				btrfs_dev_name(device),
3637 				blk_status_to_errno(bio->bi_status));
3638 			ClearPageUptodate(page);
3639 			SetPageError(page);
3640 			btrfs_dev_stat_inc_and_print(device,
3641 						     BTRFS_DEV_STAT_WRITE_ERRS);
3642 		} else {
3643 			SetPageUptodate(page);
3644 		}
3645 
3646 		put_page(page);
3647 		unlock_page(page);
3648 	}
3649 
3650 	bio_put(bio);
3651 }
3652 
3653 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3654 						   int copy_num, bool drop_cache)
3655 {
3656 	struct btrfs_super_block *super;
3657 	struct page *page;
3658 	u64 bytenr, bytenr_orig;
3659 	struct address_space *mapping = bdev->bd_inode->i_mapping;
3660 	int ret;
3661 
3662 	bytenr_orig = btrfs_sb_offset(copy_num);
3663 	ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3664 	if (ret == -ENOENT)
3665 		return ERR_PTR(-EINVAL);
3666 	else if (ret)
3667 		return ERR_PTR(ret);
3668 
3669 	if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3670 		return ERR_PTR(-EINVAL);
3671 
3672 	if (drop_cache) {
3673 		/* This should only be called with the primary sb. */
3674 		ASSERT(copy_num == 0);
3675 
3676 		/*
3677 		 * Drop the page of the primary superblock, so later read will
3678 		 * always read from the device.
3679 		 */
3680 		invalidate_inode_pages2_range(mapping,
3681 				bytenr >> PAGE_SHIFT,
3682 				(bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3683 	}
3684 
3685 	page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3686 	if (IS_ERR(page))
3687 		return ERR_CAST(page);
3688 
3689 	super = page_address(page);
3690 	if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3691 		btrfs_release_disk_super(super);
3692 		return ERR_PTR(-ENODATA);
3693 	}
3694 
3695 	if (btrfs_super_bytenr(super) != bytenr_orig) {
3696 		btrfs_release_disk_super(super);
3697 		return ERR_PTR(-EINVAL);
3698 	}
3699 
3700 	return super;
3701 }
3702 
3703 
3704 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3705 {
3706 	struct btrfs_super_block *super, *latest = NULL;
3707 	int i;
3708 	u64 transid = 0;
3709 
3710 	/* we would like to check all the supers, but that would make
3711 	 * a btrfs mount succeed after a mkfs from a different FS.
3712 	 * So, we need to add a special mount option to scan for
3713 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3714 	 */
3715 	for (i = 0; i < 1; i++) {
3716 		super = btrfs_read_dev_one_super(bdev, i, false);
3717 		if (IS_ERR(super))
3718 			continue;
3719 
3720 		if (!latest || btrfs_super_generation(super) > transid) {
3721 			if (latest)
3722 				btrfs_release_disk_super(super);
3723 
3724 			latest = super;
3725 			transid = btrfs_super_generation(super);
3726 		}
3727 	}
3728 
3729 	return super;
3730 }
3731 
3732 /*
3733  * Write superblock @sb to the @device. Do not wait for completion, all the
3734  * pages we use for writing are locked.
3735  *
3736  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3737  * the expected device size at commit time. Note that max_mirrors must be
3738  * same for write and wait phases.
3739  *
3740  * Return number of errors when page is not found or submission fails.
3741  */
3742 static int write_dev_supers(struct btrfs_device *device,
3743 			    struct btrfs_super_block *sb, int max_mirrors)
3744 {
3745 	struct btrfs_fs_info *fs_info = device->fs_info;
3746 	struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3747 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3748 	int i;
3749 	int errors = 0;
3750 	int ret;
3751 	u64 bytenr, bytenr_orig;
3752 
3753 	if (max_mirrors == 0)
3754 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3755 
3756 	shash->tfm = fs_info->csum_shash;
3757 
3758 	for (i = 0; i < max_mirrors; i++) {
3759 		struct page *page;
3760 		struct bio *bio;
3761 		struct btrfs_super_block *disk_super;
3762 
3763 		bytenr_orig = btrfs_sb_offset(i);
3764 		ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3765 		if (ret == -ENOENT) {
3766 			continue;
3767 		} else if (ret < 0) {
3768 			btrfs_err(device->fs_info,
3769 				"couldn't get super block location for mirror %d",
3770 				i);
3771 			errors++;
3772 			continue;
3773 		}
3774 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3775 		    device->commit_total_bytes)
3776 			break;
3777 
3778 		btrfs_set_super_bytenr(sb, bytenr_orig);
3779 
3780 		crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3781 				    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3782 				    sb->csum);
3783 
3784 		page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3785 					   GFP_NOFS);
3786 		if (!page) {
3787 			btrfs_err(device->fs_info,
3788 			    "couldn't get super block page for bytenr %llu",
3789 			    bytenr);
3790 			errors++;
3791 			continue;
3792 		}
3793 
3794 		/* Bump the refcount for wait_dev_supers() */
3795 		get_page(page);
3796 
3797 		disk_super = page_address(page);
3798 		memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3799 
3800 		/*
3801 		 * Directly use bios here instead of relying on the page cache
3802 		 * to do I/O, so we don't lose the ability to do integrity
3803 		 * checking.
3804 		 */
3805 		bio = bio_alloc(device->bdev, 1,
3806 				REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3807 				GFP_NOFS);
3808 		bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3809 		bio->bi_private = device;
3810 		bio->bi_end_io = btrfs_end_super_write;
3811 		__bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3812 			       offset_in_page(bytenr));
3813 
3814 		/*
3815 		 * We FUA only the first super block.  The others we allow to
3816 		 * go down lazy and there's a short window where the on-disk
3817 		 * copies might still contain the older version.
3818 		 */
3819 		if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3820 			bio->bi_opf |= REQ_FUA;
3821 
3822 		btrfsic_check_bio(bio);
3823 		submit_bio(bio);
3824 
3825 		if (btrfs_advance_sb_log(device, i))
3826 			errors++;
3827 	}
3828 	return errors < i ? 0 : -1;
3829 }
3830 
3831 /*
3832  * Wait for write completion of superblocks done by write_dev_supers,
3833  * @max_mirrors same for write and wait phases.
3834  *
3835  * Return number of errors when page is not found or not marked up to
3836  * date.
3837  */
3838 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3839 {
3840 	int i;
3841 	int errors = 0;
3842 	bool primary_failed = false;
3843 	int ret;
3844 	u64 bytenr;
3845 
3846 	if (max_mirrors == 0)
3847 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3848 
3849 	for (i = 0; i < max_mirrors; i++) {
3850 		struct page *page;
3851 
3852 		ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3853 		if (ret == -ENOENT) {
3854 			break;
3855 		} else if (ret < 0) {
3856 			errors++;
3857 			if (i == 0)
3858 				primary_failed = true;
3859 			continue;
3860 		}
3861 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3862 		    device->commit_total_bytes)
3863 			break;
3864 
3865 		page = find_get_page(device->bdev->bd_inode->i_mapping,
3866 				     bytenr >> PAGE_SHIFT);
3867 		if (!page) {
3868 			errors++;
3869 			if (i == 0)
3870 				primary_failed = true;
3871 			continue;
3872 		}
3873 		/* Page is submitted locked and unlocked once the IO completes */
3874 		wait_on_page_locked(page);
3875 		if (PageError(page)) {
3876 			errors++;
3877 			if (i == 0)
3878 				primary_failed = true;
3879 		}
3880 
3881 		/* Drop our reference */
3882 		put_page(page);
3883 
3884 		/* Drop the reference from the writing run */
3885 		put_page(page);
3886 	}
3887 
3888 	/* log error, force error return */
3889 	if (primary_failed) {
3890 		btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3891 			  device->devid);
3892 		return -1;
3893 	}
3894 
3895 	return errors < i ? 0 : -1;
3896 }
3897 
3898 /*
3899  * endio for the write_dev_flush, this will wake anyone waiting
3900  * for the barrier when it is done
3901  */
3902 static void btrfs_end_empty_barrier(struct bio *bio)
3903 {
3904 	bio_uninit(bio);
3905 	complete(bio->bi_private);
3906 }
3907 
3908 /*
3909  * Submit a flush request to the device if it supports it. Error handling is
3910  * done in the waiting counterpart.
3911  */
3912 static void write_dev_flush(struct btrfs_device *device)
3913 {
3914 	struct bio *bio = &device->flush_bio;
3915 
3916 	device->last_flush_error = BLK_STS_OK;
3917 
3918 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3919 	/*
3920 	 * When a disk has write caching disabled, we skip submission of a bio
3921 	 * with flush and sync requests before writing the superblock, since
3922 	 * it's not needed. However when the integrity checker is enabled, this
3923 	 * results in reports that there are metadata blocks referred by a
3924 	 * superblock that were not properly flushed. So don't skip the bio
3925 	 * submission only when the integrity checker is enabled for the sake
3926 	 * of simplicity, since this is a debug tool and not meant for use in
3927 	 * non-debug builds.
3928 	 */
3929 	if (!bdev_write_cache(device->bdev))
3930 		return;
3931 #endif
3932 
3933 	bio_init(bio, device->bdev, NULL, 0,
3934 		 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3935 	bio->bi_end_io = btrfs_end_empty_barrier;
3936 	init_completion(&device->flush_wait);
3937 	bio->bi_private = &device->flush_wait;
3938 
3939 	btrfsic_check_bio(bio);
3940 	submit_bio(bio);
3941 	set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3942 }
3943 
3944 /*
3945  * If the flush bio has been submitted by write_dev_flush, wait for it.
3946  * Return true for any error, and false otherwise.
3947  */
3948 static bool wait_dev_flush(struct btrfs_device *device)
3949 {
3950 	struct bio *bio = &device->flush_bio;
3951 
3952 	if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3953 		return false;
3954 
3955 	wait_for_completion_io(&device->flush_wait);
3956 
3957 	if (bio->bi_status) {
3958 		device->last_flush_error = bio->bi_status;
3959 		btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS);
3960 		return true;
3961 	}
3962 
3963 	return false;
3964 }
3965 
3966 /*
3967  * send an empty flush down to each device in parallel,
3968  * then wait for them
3969  */
3970 static int barrier_all_devices(struct btrfs_fs_info *info)
3971 {
3972 	struct list_head *head;
3973 	struct btrfs_device *dev;
3974 	int errors_wait = 0;
3975 
3976 	lockdep_assert_held(&info->fs_devices->device_list_mutex);
3977 	/* send down all the barriers */
3978 	head = &info->fs_devices->devices;
3979 	list_for_each_entry(dev, head, dev_list) {
3980 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3981 			continue;
3982 		if (!dev->bdev)
3983 			continue;
3984 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3985 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3986 			continue;
3987 
3988 		write_dev_flush(dev);
3989 	}
3990 
3991 	/* wait for all the barriers */
3992 	list_for_each_entry(dev, head, dev_list) {
3993 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3994 			continue;
3995 		if (!dev->bdev) {
3996 			errors_wait++;
3997 			continue;
3998 		}
3999 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4000 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4001 			continue;
4002 
4003 		if (wait_dev_flush(dev))
4004 			errors_wait++;
4005 	}
4006 
4007 	/*
4008 	 * Checks last_flush_error of disks in order to determine the device
4009 	 * state.
4010 	 */
4011 	if (errors_wait && !btrfs_check_rw_degradable(info, NULL))
4012 		return -EIO;
4013 
4014 	return 0;
4015 }
4016 
4017 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4018 {
4019 	int raid_type;
4020 	int min_tolerated = INT_MAX;
4021 
4022 	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4023 	    (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4024 		min_tolerated = min_t(int, min_tolerated,
4025 				    btrfs_raid_array[BTRFS_RAID_SINGLE].
4026 				    tolerated_failures);
4027 
4028 	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4029 		if (raid_type == BTRFS_RAID_SINGLE)
4030 			continue;
4031 		if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4032 			continue;
4033 		min_tolerated = min_t(int, min_tolerated,
4034 				    btrfs_raid_array[raid_type].
4035 				    tolerated_failures);
4036 	}
4037 
4038 	if (min_tolerated == INT_MAX) {
4039 		pr_warn("BTRFS: unknown raid flag: %llu", flags);
4040 		min_tolerated = 0;
4041 	}
4042 
4043 	return min_tolerated;
4044 }
4045 
4046 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4047 {
4048 	struct list_head *head;
4049 	struct btrfs_device *dev;
4050 	struct btrfs_super_block *sb;
4051 	struct btrfs_dev_item *dev_item;
4052 	int ret;
4053 	int do_barriers;
4054 	int max_errors;
4055 	int total_errors = 0;
4056 	u64 flags;
4057 
4058 	do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4059 
4060 	/*
4061 	 * max_mirrors == 0 indicates we're from commit_transaction,
4062 	 * not from fsync where the tree roots in fs_info have not
4063 	 * been consistent on disk.
4064 	 */
4065 	if (max_mirrors == 0)
4066 		backup_super_roots(fs_info);
4067 
4068 	sb = fs_info->super_for_commit;
4069 	dev_item = &sb->dev_item;
4070 
4071 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
4072 	head = &fs_info->fs_devices->devices;
4073 	max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4074 
4075 	if (do_barriers) {
4076 		ret = barrier_all_devices(fs_info);
4077 		if (ret) {
4078 			mutex_unlock(
4079 				&fs_info->fs_devices->device_list_mutex);
4080 			btrfs_handle_fs_error(fs_info, ret,
4081 					      "errors while submitting device barriers.");
4082 			return ret;
4083 		}
4084 	}
4085 
4086 	list_for_each_entry(dev, head, dev_list) {
4087 		if (!dev->bdev) {
4088 			total_errors++;
4089 			continue;
4090 		}
4091 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4092 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4093 			continue;
4094 
4095 		btrfs_set_stack_device_generation(dev_item, 0);
4096 		btrfs_set_stack_device_type(dev_item, dev->type);
4097 		btrfs_set_stack_device_id(dev_item, dev->devid);
4098 		btrfs_set_stack_device_total_bytes(dev_item,
4099 						   dev->commit_total_bytes);
4100 		btrfs_set_stack_device_bytes_used(dev_item,
4101 						  dev->commit_bytes_used);
4102 		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4103 		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4104 		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4105 		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4106 		memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4107 		       BTRFS_FSID_SIZE);
4108 
4109 		flags = btrfs_super_flags(sb);
4110 		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4111 
4112 		ret = btrfs_validate_write_super(fs_info, sb);
4113 		if (ret < 0) {
4114 			mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4115 			btrfs_handle_fs_error(fs_info, -EUCLEAN,
4116 				"unexpected superblock corruption detected");
4117 			return -EUCLEAN;
4118 		}
4119 
4120 		ret = write_dev_supers(dev, sb, max_mirrors);
4121 		if (ret)
4122 			total_errors++;
4123 	}
4124 	if (total_errors > max_errors) {
4125 		btrfs_err(fs_info, "%d errors while writing supers",
4126 			  total_errors);
4127 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4128 
4129 		/* FUA is masked off if unsupported and can't be the reason */
4130 		btrfs_handle_fs_error(fs_info, -EIO,
4131 				      "%d errors while writing supers",
4132 				      total_errors);
4133 		return -EIO;
4134 	}
4135 
4136 	total_errors = 0;
4137 	list_for_each_entry(dev, head, dev_list) {
4138 		if (!dev->bdev)
4139 			continue;
4140 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4141 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4142 			continue;
4143 
4144 		ret = wait_dev_supers(dev, max_mirrors);
4145 		if (ret)
4146 			total_errors++;
4147 	}
4148 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4149 	if (total_errors > max_errors) {
4150 		btrfs_handle_fs_error(fs_info, -EIO,
4151 				      "%d errors while writing supers",
4152 				      total_errors);
4153 		return -EIO;
4154 	}
4155 	return 0;
4156 }
4157 
4158 /* Drop a fs root from the radix tree and free it. */
4159 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4160 				  struct btrfs_root *root)
4161 {
4162 	bool drop_ref = false;
4163 
4164 	spin_lock(&fs_info->fs_roots_radix_lock);
4165 	radix_tree_delete(&fs_info->fs_roots_radix,
4166 			  (unsigned long)root->root_key.objectid);
4167 	if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4168 		drop_ref = true;
4169 	spin_unlock(&fs_info->fs_roots_radix_lock);
4170 
4171 	if (BTRFS_FS_ERROR(fs_info)) {
4172 		ASSERT(root->log_root == NULL);
4173 		if (root->reloc_root) {
4174 			btrfs_put_root(root->reloc_root);
4175 			root->reloc_root = NULL;
4176 		}
4177 	}
4178 
4179 	if (drop_ref)
4180 		btrfs_put_root(root);
4181 }
4182 
4183 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4184 {
4185 	struct btrfs_root *root = fs_info->tree_root;
4186 	struct btrfs_trans_handle *trans;
4187 
4188 	mutex_lock(&fs_info->cleaner_mutex);
4189 	btrfs_run_delayed_iputs(fs_info);
4190 	mutex_unlock(&fs_info->cleaner_mutex);
4191 	wake_up_process(fs_info->cleaner_kthread);
4192 
4193 	/* wait until ongoing cleanup work done */
4194 	down_write(&fs_info->cleanup_work_sem);
4195 	up_write(&fs_info->cleanup_work_sem);
4196 
4197 	trans = btrfs_join_transaction(root);
4198 	if (IS_ERR(trans))
4199 		return PTR_ERR(trans);
4200 	return btrfs_commit_transaction(trans);
4201 }
4202 
4203 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4204 {
4205 	struct btrfs_transaction *trans;
4206 	struct btrfs_transaction *tmp;
4207 	bool found = false;
4208 
4209 	if (list_empty(&fs_info->trans_list))
4210 		return;
4211 
4212 	/*
4213 	 * This function is only called at the very end of close_ctree(),
4214 	 * thus no other running transaction, no need to take trans_lock.
4215 	 */
4216 	ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4217 	list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4218 		struct extent_state *cached = NULL;
4219 		u64 dirty_bytes = 0;
4220 		u64 cur = 0;
4221 		u64 found_start;
4222 		u64 found_end;
4223 
4224 		found = true;
4225 		while (find_first_extent_bit(&trans->dirty_pages, cur,
4226 			&found_start, &found_end, EXTENT_DIRTY, &cached)) {
4227 			dirty_bytes += found_end + 1 - found_start;
4228 			cur = found_end + 1;
4229 		}
4230 		btrfs_warn(fs_info,
4231 	"transaction %llu (with %llu dirty metadata bytes) is not committed",
4232 			   trans->transid, dirty_bytes);
4233 		btrfs_cleanup_one_transaction(trans, fs_info);
4234 
4235 		if (trans == fs_info->running_transaction)
4236 			fs_info->running_transaction = NULL;
4237 		list_del_init(&trans->list);
4238 
4239 		btrfs_put_transaction(trans);
4240 		trace_btrfs_transaction_commit(fs_info);
4241 	}
4242 	ASSERT(!found);
4243 }
4244 
4245 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4246 {
4247 	int ret;
4248 
4249 	set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4250 
4251 	/*
4252 	 * If we had UNFINISHED_DROPS we could still be processing them, so
4253 	 * clear that bit and wake up relocation so it can stop.
4254 	 * We must do this before stopping the block group reclaim task, because
4255 	 * at btrfs_relocate_block_group() we wait for this bit, and after the
4256 	 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4257 	 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4258 	 * return 1.
4259 	 */
4260 	btrfs_wake_unfinished_drop(fs_info);
4261 
4262 	/*
4263 	 * We may have the reclaim task running and relocating a data block group,
4264 	 * in which case it may create delayed iputs. So stop it before we park
4265 	 * the cleaner kthread otherwise we can get new delayed iputs after
4266 	 * parking the cleaner, and that can make the async reclaim task to hang
4267 	 * if it's waiting for delayed iputs to complete, since the cleaner is
4268 	 * parked and can not run delayed iputs - this will make us hang when
4269 	 * trying to stop the async reclaim task.
4270 	 */
4271 	cancel_work_sync(&fs_info->reclaim_bgs_work);
4272 	/*
4273 	 * We don't want the cleaner to start new transactions, add more delayed
4274 	 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4275 	 * because that frees the task_struct, and the transaction kthread might
4276 	 * still try to wake up the cleaner.
4277 	 */
4278 	kthread_park(fs_info->cleaner_kthread);
4279 
4280 	/* wait for the qgroup rescan worker to stop */
4281 	btrfs_qgroup_wait_for_completion(fs_info, false);
4282 
4283 	/* wait for the uuid_scan task to finish */
4284 	down(&fs_info->uuid_tree_rescan_sem);
4285 	/* avoid complains from lockdep et al., set sem back to initial state */
4286 	up(&fs_info->uuid_tree_rescan_sem);
4287 
4288 	/* pause restriper - we want to resume on mount */
4289 	btrfs_pause_balance(fs_info);
4290 
4291 	btrfs_dev_replace_suspend_for_unmount(fs_info);
4292 
4293 	btrfs_scrub_cancel(fs_info);
4294 
4295 	/* wait for any defraggers to finish */
4296 	wait_event(fs_info->transaction_wait,
4297 		   (atomic_read(&fs_info->defrag_running) == 0));
4298 
4299 	/* clear out the rbtree of defraggable inodes */
4300 	btrfs_cleanup_defrag_inodes(fs_info);
4301 
4302 	/*
4303 	 * After we parked the cleaner kthread, ordered extents may have
4304 	 * completed and created new delayed iputs. If one of the async reclaim
4305 	 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4306 	 * can hang forever trying to stop it, because if a delayed iput is
4307 	 * added after it ran btrfs_run_delayed_iputs() and before it called
4308 	 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4309 	 * no one else to run iputs.
4310 	 *
4311 	 * So wait for all ongoing ordered extents to complete and then run
4312 	 * delayed iputs. This works because once we reach this point no one
4313 	 * can either create new ordered extents nor create delayed iputs
4314 	 * through some other means.
4315 	 *
4316 	 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4317 	 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4318 	 * but the delayed iput for the respective inode is made only when doing
4319 	 * the final btrfs_put_ordered_extent() (which must happen at
4320 	 * btrfs_finish_ordered_io() when we are unmounting).
4321 	 */
4322 	btrfs_flush_workqueue(fs_info->endio_write_workers);
4323 	/* Ordered extents for free space inodes. */
4324 	btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4325 	btrfs_run_delayed_iputs(fs_info);
4326 
4327 	cancel_work_sync(&fs_info->async_reclaim_work);
4328 	cancel_work_sync(&fs_info->async_data_reclaim_work);
4329 	cancel_work_sync(&fs_info->preempt_reclaim_work);
4330 
4331 	/* Cancel or finish ongoing discard work */
4332 	btrfs_discard_cleanup(fs_info);
4333 
4334 	if (!sb_rdonly(fs_info->sb)) {
4335 		/*
4336 		 * The cleaner kthread is stopped, so do one final pass over
4337 		 * unused block groups.
4338 		 */
4339 		btrfs_delete_unused_bgs(fs_info);
4340 
4341 		/*
4342 		 * There might be existing delayed inode workers still running
4343 		 * and holding an empty delayed inode item. We must wait for
4344 		 * them to complete first because they can create a transaction.
4345 		 * This happens when someone calls btrfs_balance_delayed_items()
4346 		 * and then a transaction commit runs the same delayed nodes
4347 		 * before any delayed worker has done something with the nodes.
4348 		 * We must wait for any worker here and not at transaction
4349 		 * commit time since that could cause a deadlock.
4350 		 * This is a very rare case.
4351 		 */
4352 		btrfs_flush_workqueue(fs_info->delayed_workers);
4353 
4354 		ret = btrfs_commit_super(fs_info);
4355 		if (ret)
4356 			btrfs_err(fs_info, "commit super ret %d", ret);
4357 	}
4358 
4359 	if (BTRFS_FS_ERROR(fs_info))
4360 		btrfs_error_commit_super(fs_info);
4361 
4362 	kthread_stop(fs_info->transaction_kthread);
4363 	kthread_stop(fs_info->cleaner_kthread);
4364 
4365 	ASSERT(list_empty(&fs_info->delayed_iputs));
4366 	set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4367 
4368 	if (btrfs_check_quota_leak(fs_info)) {
4369 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4370 		btrfs_err(fs_info, "qgroup reserved space leaked");
4371 	}
4372 
4373 	btrfs_free_qgroup_config(fs_info);
4374 	ASSERT(list_empty(&fs_info->delalloc_roots));
4375 
4376 	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4377 		btrfs_info(fs_info, "at unmount delalloc count %lld",
4378 		       percpu_counter_sum(&fs_info->delalloc_bytes));
4379 	}
4380 
4381 	if (percpu_counter_sum(&fs_info->ordered_bytes))
4382 		btrfs_info(fs_info, "at unmount dio bytes count %lld",
4383 			   percpu_counter_sum(&fs_info->ordered_bytes));
4384 
4385 	btrfs_sysfs_remove_mounted(fs_info);
4386 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4387 
4388 	btrfs_put_block_group_cache(fs_info);
4389 
4390 	/*
4391 	 * we must make sure there is not any read request to
4392 	 * submit after we stopping all workers.
4393 	 */
4394 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4395 	btrfs_stop_all_workers(fs_info);
4396 
4397 	/* We shouldn't have any transaction open at this point */
4398 	warn_about_uncommitted_trans(fs_info);
4399 
4400 	clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4401 	free_root_pointers(fs_info, true);
4402 	btrfs_free_fs_roots(fs_info);
4403 
4404 	/*
4405 	 * We must free the block groups after dropping the fs_roots as we could
4406 	 * have had an IO error and have left over tree log blocks that aren't
4407 	 * cleaned up until the fs roots are freed.  This makes the block group
4408 	 * accounting appear to be wrong because there's pending reserved bytes,
4409 	 * so make sure we do the block group cleanup afterwards.
4410 	 */
4411 	btrfs_free_block_groups(fs_info);
4412 
4413 	iput(fs_info->btree_inode);
4414 
4415 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4416 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4417 		btrfsic_unmount(fs_info->fs_devices);
4418 #endif
4419 
4420 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
4421 	btrfs_close_devices(fs_info->fs_devices);
4422 }
4423 
4424 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4425 {
4426 	struct btrfs_fs_info *fs_info = buf->fs_info;
4427 	u64 transid = btrfs_header_generation(buf);
4428 
4429 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4430 	/*
4431 	 * This is a fast path so only do this check if we have sanity tests
4432 	 * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4433 	 * outside of the sanity tests.
4434 	 */
4435 	if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4436 		return;
4437 #endif
4438 	btrfs_assert_tree_write_locked(buf);
4439 	if (transid != fs_info->generation)
4440 		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4441 			buf->start, transid, fs_info->generation);
4442 	set_extent_buffer_dirty(buf);
4443 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4444 	/*
4445 	 * btrfs_check_leaf() won't check item data if we don't have WRITTEN
4446 	 * set, so this will only validate the basic structure of the items.
4447 	 */
4448 	if (btrfs_header_level(buf) == 0 && btrfs_check_leaf(buf)) {
4449 		btrfs_print_leaf(buf);
4450 		ASSERT(0);
4451 	}
4452 #endif
4453 }
4454 
4455 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4456 					int flush_delayed)
4457 {
4458 	/*
4459 	 * looks as though older kernels can get into trouble with
4460 	 * this code, they end up stuck in balance_dirty_pages forever
4461 	 */
4462 	int ret;
4463 
4464 	if (current->flags & PF_MEMALLOC)
4465 		return;
4466 
4467 	if (flush_delayed)
4468 		btrfs_balance_delayed_items(fs_info);
4469 
4470 	ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4471 				     BTRFS_DIRTY_METADATA_THRESH,
4472 				     fs_info->dirty_metadata_batch);
4473 	if (ret > 0) {
4474 		balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4475 	}
4476 }
4477 
4478 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4479 {
4480 	__btrfs_btree_balance_dirty(fs_info, 1);
4481 }
4482 
4483 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4484 {
4485 	__btrfs_btree_balance_dirty(fs_info, 0);
4486 }
4487 
4488 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4489 {
4490 	/* cleanup FS via transaction */
4491 	btrfs_cleanup_transaction(fs_info);
4492 
4493 	mutex_lock(&fs_info->cleaner_mutex);
4494 	btrfs_run_delayed_iputs(fs_info);
4495 	mutex_unlock(&fs_info->cleaner_mutex);
4496 
4497 	down_write(&fs_info->cleanup_work_sem);
4498 	up_write(&fs_info->cleanup_work_sem);
4499 }
4500 
4501 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4502 {
4503 	struct btrfs_root *gang[8];
4504 	u64 root_objectid = 0;
4505 	int ret;
4506 
4507 	spin_lock(&fs_info->fs_roots_radix_lock);
4508 	while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4509 					     (void **)gang, root_objectid,
4510 					     ARRAY_SIZE(gang))) != 0) {
4511 		int i;
4512 
4513 		for (i = 0; i < ret; i++)
4514 			gang[i] = btrfs_grab_root(gang[i]);
4515 		spin_unlock(&fs_info->fs_roots_radix_lock);
4516 
4517 		for (i = 0; i < ret; i++) {
4518 			if (!gang[i])
4519 				continue;
4520 			root_objectid = gang[i]->root_key.objectid;
4521 			btrfs_free_log(NULL, gang[i]);
4522 			btrfs_put_root(gang[i]);
4523 		}
4524 		root_objectid++;
4525 		spin_lock(&fs_info->fs_roots_radix_lock);
4526 	}
4527 	spin_unlock(&fs_info->fs_roots_radix_lock);
4528 	btrfs_free_log_root_tree(NULL, fs_info);
4529 }
4530 
4531 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4532 {
4533 	struct btrfs_ordered_extent *ordered;
4534 
4535 	spin_lock(&root->ordered_extent_lock);
4536 	/*
4537 	 * This will just short circuit the ordered completion stuff which will
4538 	 * make sure the ordered extent gets properly cleaned up.
4539 	 */
4540 	list_for_each_entry(ordered, &root->ordered_extents,
4541 			    root_extent_list)
4542 		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4543 	spin_unlock(&root->ordered_extent_lock);
4544 }
4545 
4546 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4547 {
4548 	struct btrfs_root *root;
4549 	LIST_HEAD(splice);
4550 
4551 	spin_lock(&fs_info->ordered_root_lock);
4552 	list_splice_init(&fs_info->ordered_roots, &splice);
4553 	while (!list_empty(&splice)) {
4554 		root = list_first_entry(&splice, struct btrfs_root,
4555 					ordered_root);
4556 		list_move_tail(&root->ordered_root,
4557 			       &fs_info->ordered_roots);
4558 
4559 		spin_unlock(&fs_info->ordered_root_lock);
4560 		btrfs_destroy_ordered_extents(root);
4561 
4562 		cond_resched();
4563 		spin_lock(&fs_info->ordered_root_lock);
4564 	}
4565 	spin_unlock(&fs_info->ordered_root_lock);
4566 
4567 	/*
4568 	 * We need this here because if we've been flipped read-only we won't
4569 	 * get sync() from the umount, so we need to make sure any ordered
4570 	 * extents that haven't had their dirty pages IO start writeout yet
4571 	 * actually get run and error out properly.
4572 	 */
4573 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4574 }
4575 
4576 static void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4577 				       struct btrfs_fs_info *fs_info)
4578 {
4579 	struct rb_node *node;
4580 	struct btrfs_delayed_ref_root *delayed_refs;
4581 	struct btrfs_delayed_ref_node *ref;
4582 
4583 	delayed_refs = &trans->delayed_refs;
4584 
4585 	spin_lock(&delayed_refs->lock);
4586 	if (atomic_read(&delayed_refs->num_entries) == 0) {
4587 		spin_unlock(&delayed_refs->lock);
4588 		btrfs_debug(fs_info, "delayed_refs has NO entry");
4589 		return;
4590 	}
4591 
4592 	while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4593 		struct btrfs_delayed_ref_head *head;
4594 		struct rb_node *n;
4595 		bool pin_bytes = false;
4596 
4597 		head = rb_entry(node, struct btrfs_delayed_ref_head,
4598 				href_node);
4599 		if (btrfs_delayed_ref_lock(delayed_refs, head))
4600 			continue;
4601 
4602 		spin_lock(&head->lock);
4603 		while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4604 			ref = rb_entry(n, struct btrfs_delayed_ref_node,
4605 				       ref_node);
4606 			rb_erase_cached(&ref->ref_node, &head->ref_tree);
4607 			RB_CLEAR_NODE(&ref->ref_node);
4608 			if (!list_empty(&ref->add_list))
4609 				list_del(&ref->add_list);
4610 			atomic_dec(&delayed_refs->num_entries);
4611 			btrfs_put_delayed_ref(ref);
4612 		}
4613 		if (head->must_insert_reserved)
4614 			pin_bytes = true;
4615 		btrfs_free_delayed_extent_op(head->extent_op);
4616 		btrfs_delete_ref_head(delayed_refs, head);
4617 		spin_unlock(&head->lock);
4618 		spin_unlock(&delayed_refs->lock);
4619 		mutex_unlock(&head->mutex);
4620 
4621 		if (pin_bytes) {
4622 			struct btrfs_block_group *cache;
4623 
4624 			cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4625 			BUG_ON(!cache);
4626 
4627 			spin_lock(&cache->space_info->lock);
4628 			spin_lock(&cache->lock);
4629 			cache->pinned += head->num_bytes;
4630 			btrfs_space_info_update_bytes_pinned(fs_info,
4631 				cache->space_info, head->num_bytes);
4632 			cache->reserved -= head->num_bytes;
4633 			cache->space_info->bytes_reserved -= head->num_bytes;
4634 			spin_unlock(&cache->lock);
4635 			spin_unlock(&cache->space_info->lock);
4636 
4637 			btrfs_put_block_group(cache);
4638 
4639 			btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4640 				head->bytenr + head->num_bytes - 1);
4641 		}
4642 		btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4643 		btrfs_put_delayed_ref_head(head);
4644 		cond_resched();
4645 		spin_lock(&delayed_refs->lock);
4646 	}
4647 	btrfs_qgroup_destroy_extent_records(trans);
4648 
4649 	spin_unlock(&delayed_refs->lock);
4650 }
4651 
4652 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4653 {
4654 	struct btrfs_inode *btrfs_inode;
4655 	LIST_HEAD(splice);
4656 
4657 	spin_lock(&root->delalloc_lock);
4658 	list_splice_init(&root->delalloc_inodes, &splice);
4659 
4660 	while (!list_empty(&splice)) {
4661 		struct inode *inode = NULL;
4662 		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4663 					       delalloc_inodes);
4664 		__btrfs_del_delalloc_inode(root, btrfs_inode);
4665 		spin_unlock(&root->delalloc_lock);
4666 
4667 		/*
4668 		 * Make sure we get a live inode and that it'll not disappear
4669 		 * meanwhile.
4670 		 */
4671 		inode = igrab(&btrfs_inode->vfs_inode);
4672 		if (inode) {
4673 			unsigned int nofs_flag;
4674 
4675 			nofs_flag = memalloc_nofs_save();
4676 			invalidate_inode_pages2(inode->i_mapping);
4677 			memalloc_nofs_restore(nofs_flag);
4678 			iput(inode);
4679 		}
4680 		spin_lock(&root->delalloc_lock);
4681 	}
4682 	spin_unlock(&root->delalloc_lock);
4683 }
4684 
4685 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4686 {
4687 	struct btrfs_root *root;
4688 	LIST_HEAD(splice);
4689 
4690 	spin_lock(&fs_info->delalloc_root_lock);
4691 	list_splice_init(&fs_info->delalloc_roots, &splice);
4692 	while (!list_empty(&splice)) {
4693 		root = list_first_entry(&splice, struct btrfs_root,
4694 					 delalloc_root);
4695 		root = btrfs_grab_root(root);
4696 		BUG_ON(!root);
4697 		spin_unlock(&fs_info->delalloc_root_lock);
4698 
4699 		btrfs_destroy_delalloc_inodes(root);
4700 		btrfs_put_root(root);
4701 
4702 		spin_lock(&fs_info->delalloc_root_lock);
4703 	}
4704 	spin_unlock(&fs_info->delalloc_root_lock);
4705 }
4706 
4707 static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4708 					 struct extent_io_tree *dirty_pages,
4709 					 int mark)
4710 {
4711 	struct extent_buffer *eb;
4712 	u64 start = 0;
4713 	u64 end;
4714 
4715 	while (find_first_extent_bit(dirty_pages, start, &start, &end,
4716 				     mark, NULL)) {
4717 		clear_extent_bits(dirty_pages, start, end, mark);
4718 		while (start <= end) {
4719 			eb = find_extent_buffer(fs_info, start);
4720 			start += fs_info->nodesize;
4721 			if (!eb)
4722 				continue;
4723 
4724 			btrfs_tree_lock(eb);
4725 			wait_on_extent_buffer_writeback(eb);
4726 			btrfs_clear_buffer_dirty(NULL, eb);
4727 			btrfs_tree_unlock(eb);
4728 
4729 			free_extent_buffer_stale(eb);
4730 		}
4731 	}
4732 }
4733 
4734 static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4735 					struct extent_io_tree *unpin)
4736 {
4737 	u64 start;
4738 	u64 end;
4739 
4740 	while (1) {
4741 		struct extent_state *cached_state = NULL;
4742 
4743 		/*
4744 		 * The btrfs_finish_extent_commit() may get the same range as
4745 		 * ours between find_first_extent_bit and clear_extent_dirty.
4746 		 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4747 		 * the same extent range.
4748 		 */
4749 		mutex_lock(&fs_info->unused_bg_unpin_mutex);
4750 		if (!find_first_extent_bit(unpin, 0, &start, &end,
4751 					   EXTENT_DIRTY, &cached_state)) {
4752 			mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4753 			break;
4754 		}
4755 
4756 		clear_extent_dirty(unpin, start, end, &cached_state);
4757 		free_extent_state(cached_state);
4758 		btrfs_error_unpin_extent_range(fs_info, start, end);
4759 		mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4760 		cond_resched();
4761 	}
4762 }
4763 
4764 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4765 {
4766 	struct inode *inode;
4767 
4768 	inode = cache->io_ctl.inode;
4769 	if (inode) {
4770 		unsigned int nofs_flag;
4771 
4772 		nofs_flag = memalloc_nofs_save();
4773 		invalidate_inode_pages2(inode->i_mapping);
4774 		memalloc_nofs_restore(nofs_flag);
4775 
4776 		BTRFS_I(inode)->generation = 0;
4777 		cache->io_ctl.inode = NULL;
4778 		iput(inode);
4779 	}
4780 	ASSERT(cache->io_ctl.pages == NULL);
4781 	btrfs_put_block_group(cache);
4782 }
4783 
4784 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4785 			     struct btrfs_fs_info *fs_info)
4786 {
4787 	struct btrfs_block_group *cache;
4788 
4789 	spin_lock(&cur_trans->dirty_bgs_lock);
4790 	while (!list_empty(&cur_trans->dirty_bgs)) {
4791 		cache = list_first_entry(&cur_trans->dirty_bgs,
4792 					 struct btrfs_block_group,
4793 					 dirty_list);
4794 
4795 		if (!list_empty(&cache->io_list)) {
4796 			spin_unlock(&cur_trans->dirty_bgs_lock);
4797 			list_del_init(&cache->io_list);
4798 			btrfs_cleanup_bg_io(cache);
4799 			spin_lock(&cur_trans->dirty_bgs_lock);
4800 		}
4801 
4802 		list_del_init(&cache->dirty_list);
4803 		spin_lock(&cache->lock);
4804 		cache->disk_cache_state = BTRFS_DC_ERROR;
4805 		spin_unlock(&cache->lock);
4806 
4807 		spin_unlock(&cur_trans->dirty_bgs_lock);
4808 		btrfs_put_block_group(cache);
4809 		btrfs_delayed_refs_rsv_release(fs_info, 1);
4810 		spin_lock(&cur_trans->dirty_bgs_lock);
4811 	}
4812 	spin_unlock(&cur_trans->dirty_bgs_lock);
4813 
4814 	/*
4815 	 * Refer to the definition of io_bgs member for details why it's safe
4816 	 * to use it without any locking
4817 	 */
4818 	while (!list_empty(&cur_trans->io_bgs)) {
4819 		cache = list_first_entry(&cur_trans->io_bgs,
4820 					 struct btrfs_block_group,
4821 					 io_list);
4822 
4823 		list_del_init(&cache->io_list);
4824 		spin_lock(&cache->lock);
4825 		cache->disk_cache_state = BTRFS_DC_ERROR;
4826 		spin_unlock(&cache->lock);
4827 		btrfs_cleanup_bg_io(cache);
4828 	}
4829 }
4830 
4831 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4832 				   struct btrfs_fs_info *fs_info)
4833 {
4834 	struct btrfs_device *dev, *tmp;
4835 
4836 	btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4837 	ASSERT(list_empty(&cur_trans->dirty_bgs));
4838 	ASSERT(list_empty(&cur_trans->io_bgs));
4839 
4840 	list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4841 				 post_commit_list) {
4842 		list_del_init(&dev->post_commit_list);
4843 	}
4844 
4845 	btrfs_destroy_delayed_refs(cur_trans, fs_info);
4846 
4847 	cur_trans->state = TRANS_STATE_COMMIT_START;
4848 	wake_up(&fs_info->transaction_blocked_wait);
4849 
4850 	cur_trans->state = TRANS_STATE_UNBLOCKED;
4851 	wake_up(&fs_info->transaction_wait);
4852 
4853 	btrfs_destroy_delayed_inodes(fs_info);
4854 
4855 	btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4856 				     EXTENT_DIRTY);
4857 	btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4858 
4859 	cur_trans->state =TRANS_STATE_COMPLETED;
4860 	wake_up(&cur_trans->commit_wait);
4861 }
4862 
4863 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4864 {
4865 	struct btrfs_transaction *t;
4866 
4867 	mutex_lock(&fs_info->transaction_kthread_mutex);
4868 
4869 	spin_lock(&fs_info->trans_lock);
4870 	while (!list_empty(&fs_info->trans_list)) {
4871 		t = list_first_entry(&fs_info->trans_list,
4872 				     struct btrfs_transaction, list);
4873 		if (t->state >= TRANS_STATE_COMMIT_START) {
4874 			refcount_inc(&t->use_count);
4875 			spin_unlock(&fs_info->trans_lock);
4876 			btrfs_wait_for_commit(fs_info, t->transid);
4877 			btrfs_put_transaction(t);
4878 			spin_lock(&fs_info->trans_lock);
4879 			continue;
4880 		}
4881 		if (t == fs_info->running_transaction) {
4882 			t->state = TRANS_STATE_COMMIT_DOING;
4883 			spin_unlock(&fs_info->trans_lock);
4884 			/*
4885 			 * We wait for 0 num_writers since we don't hold a trans
4886 			 * handle open currently for this transaction.
4887 			 */
4888 			wait_event(t->writer_wait,
4889 				   atomic_read(&t->num_writers) == 0);
4890 		} else {
4891 			spin_unlock(&fs_info->trans_lock);
4892 		}
4893 		btrfs_cleanup_one_transaction(t, fs_info);
4894 
4895 		spin_lock(&fs_info->trans_lock);
4896 		if (t == fs_info->running_transaction)
4897 			fs_info->running_transaction = NULL;
4898 		list_del_init(&t->list);
4899 		spin_unlock(&fs_info->trans_lock);
4900 
4901 		btrfs_put_transaction(t);
4902 		trace_btrfs_transaction_commit(fs_info);
4903 		spin_lock(&fs_info->trans_lock);
4904 	}
4905 	spin_unlock(&fs_info->trans_lock);
4906 	btrfs_destroy_all_ordered_extents(fs_info);
4907 	btrfs_destroy_delayed_inodes(fs_info);
4908 	btrfs_assert_delayed_root_empty(fs_info);
4909 	btrfs_destroy_all_delalloc_inodes(fs_info);
4910 	btrfs_drop_all_logs(fs_info);
4911 	mutex_unlock(&fs_info->transaction_kthread_mutex);
4912 
4913 	return 0;
4914 }
4915 
4916 int btrfs_init_root_free_objectid(struct btrfs_root *root)
4917 {
4918 	struct btrfs_path *path;
4919 	int ret;
4920 	struct extent_buffer *l;
4921 	struct btrfs_key search_key;
4922 	struct btrfs_key found_key;
4923 	int slot;
4924 
4925 	path = btrfs_alloc_path();
4926 	if (!path)
4927 		return -ENOMEM;
4928 
4929 	search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4930 	search_key.type = -1;
4931 	search_key.offset = (u64)-1;
4932 	ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4933 	if (ret < 0)
4934 		goto error;
4935 	BUG_ON(ret == 0); /* Corruption */
4936 	if (path->slots[0] > 0) {
4937 		slot = path->slots[0] - 1;
4938 		l = path->nodes[0];
4939 		btrfs_item_key_to_cpu(l, &found_key, slot);
4940 		root->free_objectid = max_t(u64, found_key.objectid + 1,
4941 					    BTRFS_FIRST_FREE_OBJECTID);
4942 	} else {
4943 		root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4944 	}
4945 	ret = 0;
4946 error:
4947 	btrfs_free_path(path);
4948 	return ret;
4949 }
4950 
4951 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4952 {
4953 	int ret;
4954 	mutex_lock(&root->objectid_mutex);
4955 
4956 	if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4957 		btrfs_warn(root->fs_info,
4958 			   "the objectid of root %llu reaches its highest value",
4959 			   root->root_key.objectid);
4960 		ret = -ENOSPC;
4961 		goto out;
4962 	}
4963 
4964 	*objectid = root->free_objectid++;
4965 	ret = 0;
4966 out:
4967 	mutex_unlock(&root->objectid_mutex);
4968 	return ret;
4969 }
4970