xref: /linux/fs/btrfs/disk-io.c (revision 7a08cb9b4bb92fb86f5fe8a3aa0ac08a9b3d783b)
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 "dev-replace.h"
33 #include "raid56.h"
34 #include "sysfs.h"
35 #include "qgroup.h"
36 #include "compression.h"
37 #include "tree-checker.h"
38 #include "ref-verify.h"
39 #include "block-group.h"
40 #include "discard.h"
41 #include "space-info.h"
42 #include "zoned.h"
43 #include "subpage.h"
44 #include "fs.h"
45 #include "accessors.h"
46 #include "extent-tree.h"
47 #include "root-tree.h"
48 #include "defrag.h"
49 #include "uuid-tree.h"
50 #include "relocation.h"
51 #include "scrub.h"
52 #include "super.h"
53 
54 #define BTRFS_SUPER_FLAG_SUPP	(BTRFS_HEADER_FLAG_WRITTEN |\
55 				 BTRFS_HEADER_FLAG_RELOC |\
56 				 BTRFS_SUPER_FLAG_ERROR |\
57 				 BTRFS_SUPER_FLAG_SEEDING |\
58 				 BTRFS_SUPER_FLAG_METADUMP |\
59 				 BTRFS_SUPER_FLAG_METADUMP_V2)
60 
61 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
62 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
63 
64 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
65 {
66 	if (fs_info->csum_shash)
67 		crypto_free_shash(fs_info->csum_shash);
68 }
69 
70 /*
71  * Compute the csum of a btree block and store the result to provided buffer.
72  */
73 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
74 {
75 	struct btrfs_fs_info *fs_info = buf->fs_info;
76 	int num_pages;
77 	u32 first_page_part;
78 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
79 	char *kaddr;
80 	int i;
81 
82 	shash->tfm = fs_info->csum_shash;
83 	crypto_shash_init(shash);
84 
85 	if (buf->addr) {
86 		/* Pages are contiguous, handle them as a big one. */
87 		kaddr = buf->addr;
88 		first_page_part = fs_info->nodesize;
89 		num_pages = 1;
90 	} else {
91 		kaddr = folio_address(buf->folios[0]);
92 		first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
93 		num_pages = num_extent_pages(buf);
94 	}
95 
96 	crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
97 			    first_page_part - BTRFS_CSUM_SIZE);
98 
99 	/*
100 	 * Multiple single-page folios case would reach here.
101 	 *
102 	 * nodesize <= PAGE_SIZE and large folio all handled by above
103 	 * crypto_shash_update() already.
104 	 */
105 	for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
106 		kaddr = folio_address(buf->folios[i]);
107 		crypto_shash_update(shash, kaddr, PAGE_SIZE);
108 	}
109 	memset(result, 0, BTRFS_CSUM_SIZE);
110 	crypto_shash_final(shash, result);
111 }
112 
113 /*
114  * we can't consider a given block up to date unless the transid of the
115  * block matches the transid in the parent node's pointer.  This is how we
116  * detect blocks that either didn't get written at all or got written
117  * in the wrong place.
118  */
119 int btrfs_buffer_uptodate(struct extent_buffer *eb, u64 parent_transid, int atomic)
120 {
121 	if (!extent_buffer_uptodate(eb))
122 		return 0;
123 
124 	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
125 		return 1;
126 
127 	if (atomic)
128 		return -EAGAIN;
129 
130 	if (!extent_buffer_uptodate(eb) ||
131 	    btrfs_header_generation(eb) != parent_transid) {
132 		btrfs_err_rl(eb->fs_info,
133 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
134 			eb->start, eb->read_mirror,
135 			parent_transid, btrfs_header_generation(eb));
136 		clear_extent_buffer_uptodate(eb);
137 		return 0;
138 	}
139 	return 1;
140 }
141 
142 static bool btrfs_supported_super_csum(u16 csum_type)
143 {
144 	switch (csum_type) {
145 	case BTRFS_CSUM_TYPE_CRC32:
146 	case BTRFS_CSUM_TYPE_XXHASH:
147 	case BTRFS_CSUM_TYPE_SHA256:
148 	case BTRFS_CSUM_TYPE_BLAKE2:
149 		return true;
150 	default:
151 		return false;
152 	}
153 }
154 
155 /*
156  * Return 0 if the superblock checksum type matches the checksum value of that
157  * algorithm. Pass the raw disk superblock data.
158  */
159 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
160 			   const struct btrfs_super_block *disk_sb)
161 {
162 	char result[BTRFS_CSUM_SIZE];
163 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
164 
165 	shash->tfm = fs_info->csum_shash;
166 
167 	/*
168 	 * The super_block structure does not span the whole
169 	 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
170 	 * filled with zeros and is included in the checksum.
171 	 */
172 	crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
173 			    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
174 
175 	if (memcmp(disk_sb->csum, result, fs_info->csum_size))
176 		return 1;
177 
178 	return 0;
179 }
180 
181 static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
182 				      int mirror_num)
183 {
184 	struct btrfs_fs_info *fs_info = eb->fs_info;
185 	int num_folios = num_extent_folios(eb);
186 	int ret = 0;
187 
188 	if (sb_rdonly(fs_info->sb))
189 		return -EROFS;
190 
191 	for (int i = 0; i < num_folios; i++) {
192 		struct folio *folio = eb->folios[i];
193 		u64 start = max_t(u64, eb->start, folio_pos(folio));
194 		u64 end = min_t(u64, eb->start + eb->len,
195 				folio_pos(folio) + eb->folio_size);
196 		u32 len = end - start;
197 
198 		ret = btrfs_repair_io_failure(fs_info, 0, start, len,
199 					      start, folio, offset_in_folio(folio, start),
200 					      mirror_num);
201 		if (ret)
202 			break;
203 	}
204 
205 	return ret;
206 }
207 
208 /*
209  * helper to read a given tree block, doing retries as required when
210  * the checksums don't match and we have alternate mirrors to try.
211  *
212  * @check:		expected tree parentness check, see the comments of the
213  *			structure for details.
214  */
215 int btrfs_read_extent_buffer(struct extent_buffer *eb,
216 			     const struct btrfs_tree_parent_check *check)
217 {
218 	struct btrfs_fs_info *fs_info = eb->fs_info;
219 	int failed = 0;
220 	int ret;
221 	int num_copies = 0;
222 	int mirror_num = 0;
223 	int failed_mirror = 0;
224 
225 	ASSERT(check);
226 
227 	while (1) {
228 		clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
229 		ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
230 		if (!ret)
231 			break;
232 
233 		num_copies = btrfs_num_copies(fs_info,
234 					      eb->start, eb->len);
235 		if (num_copies == 1)
236 			break;
237 
238 		if (!failed_mirror) {
239 			failed = 1;
240 			failed_mirror = eb->read_mirror;
241 		}
242 
243 		mirror_num++;
244 		if (mirror_num == failed_mirror)
245 			mirror_num++;
246 
247 		if (mirror_num > num_copies)
248 			break;
249 	}
250 
251 	if (failed && !ret && failed_mirror)
252 		btrfs_repair_eb_io_failure(eb, failed_mirror);
253 
254 	return ret;
255 }
256 
257 /*
258  * Checksum a dirty tree block before IO.
259  */
260 blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
261 {
262 	struct extent_buffer *eb = bbio->private;
263 	struct btrfs_fs_info *fs_info = eb->fs_info;
264 	u64 found_start = btrfs_header_bytenr(eb);
265 	u64 last_trans;
266 	u8 result[BTRFS_CSUM_SIZE];
267 	int ret;
268 
269 	/* Btree blocks are always contiguous on disk. */
270 	if (WARN_ON_ONCE(bbio->file_offset != eb->start))
271 		return BLK_STS_IOERR;
272 	if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len))
273 		return BLK_STS_IOERR;
274 
275 	/*
276 	 * If an extent_buffer is marked as EXTENT_BUFFER_ZONED_ZEROOUT, don't
277 	 * checksum it but zero-out its content. This is done to preserve
278 	 * ordering of I/O without unnecessarily writing out data.
279 	 */
280 	if (test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags)) {
281 		memzero_extent_buffer(eb, 0, eb->len);
282 		return BLK_STS_OK;
283 	}
284 
285 	if (WARN_ON_ONCE(found_start != eb->start))
286 		return BLK_STS_IOERR;
287 	if (WARN_ON(!btrfs_folio_test_uptodate(fs_info, eb->folios[0],
288 					       eb->start, eb->len)))
289 		return BLK_STS_IOERR;
290 
291 	ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
292 				    offsetof(struct btrfs_header, fsid),
293 				    BTRFS_FSID_SIZE) == 0);
294 	csum_tree_block(eb, result);
295 
296 	if (btrfs_header_level(eb))
297 		ret = btrfs_check_node(eb);
298 	else
299 		ret = btrfs_check_leaf(eb);
300 
301 	if (ret < 0)
302 		goto error;
303 
304 	/*
305 	 * Also check the generation, the eb reached here must be newer than
306 	 * last committed. Or something seriously wrong happened.
307 	 */
308 	last_trans = btrfs_get_last_trans_committed(fs_info);
309 	if (unlikely(btrfs_header_generation(eb) <= last_trans)) {
310 		ret = -EUCLEAN;
311 		btrfs_err(fs_info,
312 			"block=%llu bad generation, have %llu expect > %llu",
313 			  eb->start, btrfs_header_generation(eb), last_trans);
314 		goto error;
315 	}
316 	write_extent_buffer(eb, result, 0, fs_info->csum_size);
317 	return BLK_STS_OK;
318 
319 error:
320 	btrfs_print_tree(eb, 0);
321 	btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
322 		  eb->start);
323 	/*
324 	 * Be noisy if this is an extent buffer from a log tree. We don't abort
325 	 * a transaction in case there's a bad log tree extent buffer, we just
326 	 * fallback to a transaction commit. Still we want to know when there is
327 	 * a bad log tree extent buffer, as that may signal a bug somewhere.
328 	 */
329 	WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
330 		btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
331 	return errno_to_blk_status(ret);
332 }
333 
334 static bool check_tree_block_fsid(struct extent_buffer *eb)
335 {
336 	struct btrfs_fs_info *fs_info = eb->fs_info;
337 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
338 	u8 fsid[BTRFS_FSID_SIZE];
339 
340 	read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
341 			   BTRFS_FSID_SIZE);
342 
343 	/*
344 	 * alloc_fsid_devices() copies the fsid into fs_devices::metadata_uuid.
345 	 * This is then overwritten by metadata_uuid if it is present in the
346 	 * device_list_add(). The same true for a seed device as well. So use of
347 	 * fs_devices::metadata_uuid is appropriate here.
348 	 */
349 	if (memcmp(fsid, fs_info->fs_devices->metadata_uuid, BTRFS_FSID_SIZE) == 0)
350 		return false;
351 
352 	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
353 		if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
354 			return false;
355 
356 	return true;
357 }
358 
359 /* Do basic extent buffer checks at read time */
360 int btrfs_validate_extent_buffer(struct extent_buffer *eb,
361 				 const struct btrfs_tree_parent_check *check)
362 {
363 	struct btrfs_fs_info *fs_info = eb->fs_info;
364 	u64 found_start;
365 	const u32 csum_size = fs_info->csum_size;
366 	u8 found_level;
367 	u8 result[BTRFS_CSUM_SIZE];
368 	const u8 *header_csum;
369 	int ret = 0;
370 	const bool ignore_csum = btrfs_test_opt(fs_info, IGNOREMETACSUMS);
371 
372 	ASSERT(check);
373 
374 	found_start = btrfs_header_bytenr(eb);
375 	if (found_start != eb->start) {
376 		btrfs_err_rl(fs_info,
377 			"bad tree block start, mirror %u want %llu have %llu",
378 			     eb->read_mirror, eb->start, found_start);
379 		ret = -EIO;
380 		goto out;
381 	}
382 	if (check_tree_block_fsid(eb)) {
383 		btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
384 			     eb->start, eb->read_mirror);
385 		ret = -EIO;
386 		goto out;
387 	}
388 	found_level = btrfs_header_level(eb);
389 	if (found_level >= BTRFS_MAX_LEVEL) {
390 		btrfs_err(fs_info,
391 			"bad tree block level, mirror %u level %d on logical %llu",
392 			eb->read_mirror, btrfs_header_level(eb), eb->start);
393 		ret = -EIO;
394 		goto out;
395 	}
396 
397 	csum_tree_block(eb, result);
398 	header_csum = folio_address(eb->folios[0]) +
399 		get_eb_offset_in_folio(eb, offsetof(struct btrfs_header, csum));
400 
401 	if (memcmp(result, header_csum, csum_size) != 0) {
402 		btrfs_warn_rl(fs_info,
403 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d%s",
404 			      eb->start, eb->read_mirror,
405 			      CSUM_FMT_VALUE(csum_size, header_csum),
406 			      CSUM_FMT_VALUE(csum_size, result),
407 			      btrfs_header_level(eb),
408 			      ignore_csum ? ", ignored" : "");
409 		if (!ignore_csum) {
410 			ret = -EUCLEAN;
411 			goto out;
412 		}
413 	}
414 
415 	if (found_level != check->level) {
416 		btrfs_err(fs_info,
417 		"level verify failed on logical %llu mirror %u wanted %u found %u",
418 			  eb->start, eb->read_mirror, check->level, found_level);
419 		ret = -EIO;
420 		goto out;
421 	}
422 	if (unlikely(check->transid &&
423 		     btrfs_header_generation(eb) != check->transid)) {
424 		btrfs_err_rl(eb->fs_info,
425 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
426 				eb->start, eb->read_mirror, check->transid,
427 				btrfs_header_generation(eb));
428 		ret = -EIO;
429 		goto out;
430 	}
431 	if (check->has_first_key) {
432 		const struct btrfs_key *expect_key = &check->first_key;
433 		struct btrfs_key found_key;
434 
435 		if (found_level)
436 			btrfs_node_key_to_cpu(eb, &found_key, 0);
437 		else
438 			btrfs_item_key_to_cpu(eb, &found_key, 0);
439 		if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
440 			btrfs_err(fs_info,
441 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
442 				  eb->start, check->transid,
443 				  expect_key->objectid,
444 				  expect_key->type, expect_key->offset,
445 				  found_key.objectid, found_key.type,
446 				  found_key.offset);
447 			ret = -EUCLEAN;
448 			goto out;
449 		}
450 	}
451 	if (check->owner_root) {
452 		ret = btrfs_check_eb_owner(eb, check->owner_root);
453 		if (ret < 0)
454 			goto out;
455 	}
456 
457 	/*
458 	 * If this is a leaf block and it is corrupt, set the corrupt bit so
459 	 * that we don't try and read the other copies of this block, just
460 	 * return -EIO.
461 	 */
462 	if (found_level == 0 && btrfs_check_leaf(eb)) {
463 		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
464 		ret = -EIO;
465 	}
466 
467 	if (found_level > 0 && btrfs_check_node(eb))
468 		ret = -EIO;
469 
470 	if (ret)
471 		btrfs_err(fs_info,
472 		"read time tree block corruption detected on logical %llu mirror %u",
473 			  eb->start, eb->read_mirror);
474 out:
475 	return ret;
476 }
477 
478 #ifdef CONFIG_MIGRATION
479 static int btree_migrate_folio(struct address_space *mapping,
480 		struct folio *dst, struct folio *src, enum migrate_mode mode)
481 {
482 	/*
483 	 * we can't safely write a btree page from here,
484 	 * we haven't done the locking hook
485 	 */
486 	if (folio_test_dirty(src))
487 		return -EAGAIN;
488 	/*
489 	 * Buffers may be managed in a filesystem specific way.
490 	 * We must have no buffers or drop them.
491 	 */
492 	if (folio_get_private(src) &&
493 	    !filemap_release_folio(src, GFP_KERNEL))
494 		return -EAGAIN;
495 	return migrate_folio(mapping, dst, src, mode);
496 }
497 #else
498 #define btree_migrate_folio NULL
499 #endif
500 
501 static int btree_writepages(struct address_space *mapping,
502 			    struct writeback_control *wbc)
503 {
504 	int ret;
505 
506 	if (wbc->sync_mode == WB_SYNC_NONE) {
507 		struct btrfs_fs_info *fs_info;
508 
509 		if (wbc->for_kupdate)
510 			return 0;
511 
512 		fs_info = inode_to_fs_info(mapping->host);
513 		/* this is a bit racy, but that's ok */
514 		ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
515 					     BTRFS_DIRTY_METADATA_THRESH,
516 					     fs_info->dirty_metadata_batch);
517 		if (ret < 0)
518 			return 0;
519 	}
520 	return btree_write_cache_pages(mapping, wbc);
521 }
522 
523 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
524 {
525 	if (folio_test_writeback(folio) || folio_test_dirty(folio))
526 		return false;
527 
528 	return try_release_extent_buffer(folio);
529 }
530 
531 static void btree_invalidate_folio(struct folio *folio, size_t offset,
532 				 size_t length)
533 {
534 	struct extent_io_tree *tree;
535 
536 	tree = &folio_to_inode(folio)->io_tree;
537 	extent_invalidate_folio(tree, folio, offset);
538 	btree_release_folio(folio, GFP_NOFS);
539 	if (folio_get_private(folio)) {
540 		btrfs_warn(folio_to_fs_info(folio),
541 			   "folio private not zero on folio %llu",
542 			   (unsigned long long)folio_pos(folio));
543 		folio_detach_private(folio);
544 	}
545 }
546 
547 #ifdef DEBUG
548 static bool btree_dirty_folio(struct address_space *mapping,
549 		struct folio *folio)
550 {
551 	struct btrfs_fs_info *fs_info = inode_to_fs_info(mapping->host);
552 	struct btrfs_subpage_info *spi = fs_info->subpage_info;
553 	struct btrfs_subpage *subpage;
554 	struct extent_buffer *eb;
555 	int cur_bit = 0;
556 	u64 page_start = folio_pos(folio);
557 
558 	if (fs_info->sectorsize == PAGE_SIZE) {
559 		eb = folio_get_private(folio);
560 		BUG_ON(!eb);
561 		BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
562 		BUG_ON(!atomic_read(&eb->refs));
563 		btrfs_assert_tree_write_locked(eb);
564 		return filemap_dirty_folio(mapping, folio);
565 	}
566 
567 	ASSERT(spi);
568 	subpage = folio_get_private(folio);
569 
570 	for (cur_bit = spi->dirty_offset;
571 	     cur_bit < spi->dirty_offset + spi->bitmap_nr_bits;
572 	     cur_bit++) {
573 		unsigned long flags;
574 		u64 cur;
575 
576 		spin_lock_irqsave(&subpage->lock, flags);
577 		if (!test_bit(cur_bit, subpage->bitmaps)) {
578 			spin_unlock_irqrestore(&subpage->lock, flags);
579 			continue;
580 		}
581 		spin_unlock_irqrestore(&subpage->lock, flags);
582 		cur = page_start + cur_bit * fs_info->sectorsize;
583 
584 		eb = find_extent_buffer(fs_info, cur);
585 		ASSERT(eb);
586 		ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
587 		ASSERT(atomic_read(&eb->refs));
588 		btrfs_assert_tree_write_locked(eb);
589 		free_extent_buffer(eb);
590 
591 		cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits) - 1;
592 	}
593 	return filemap_dirty_folio(mapping, folio);
594 }
595 #else
596 #define btree_dirty_folio filemap_dirty_folio
597 #endif
598 
599 static const struct address_space_operations btree_aops = {
600 	.writepages	= btree_writepages,
601 	.release_folio	= btree_release_folio,
602 	.invalidate_folio = btree_invalidate_folio,
603 	.migrate_folio	= btree_migrate_folio,
604 	.dirty_folio	= btree_dirty_folio,
605 };
606 
607 struct extent_buffer *btrfs_find_create_tree_block(
608 						struct btrfs_fs_info *fs_info,
609 						u64 bytenr, u64 owner_root,
610 						int level)
611 {
612 	if (btrfs_is_testing(fs_info))
613 		return alloc_test_extent_buffer(fs_info, bytenr);
614 	return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
615 }
616 
617 /*
618  * Read tree block at logical address @bytenr and do variant basic but critical
619  * verification.
620  *
621  * @check:		expected tree parentness check, see comments of the
622  *			structure for details.
623  */
624 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
625 				      struct btrfs_tree_parent_check *check)
626 {
627 	struct extent_buffer *buf = NULL;
628 	int ret;
629 
630 	ASSERT(check);
631 
632 	buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
633 					   check->level);
634 	if (IS_ERR(buf))
635 		return buf;
636 
637 	ret = btrfs_read_extent_buffer(buf, check);
638 	if (ret) {
639 		free_extent_buffer_stale(buf);
640 		return ERR_PTR(ret);
641 	}
642 	return buf;
643 
644 }
645 
646 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
647 			 u64 objectid)
648 {
649 	bool dummy = btrfs_is_testing(fs_info);
650 
651 	memset(&root->root_key, 0, sizeof(root->root_key));
652 	memset(&root->root_item, 0, sizeof(root->root_item));
653 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
654 	root->fs_info = fs_info;
655 	root->root_key.objectid = objectid;
656 	root->node = NULL;
657 	root->commit_root = NULL;
658 	root->state = 0;
659 	RB_CLEAR_NODE(&root->rb_node);
660 
661 	btrfs_set_root_last_trans(root, 0);
662 	root->free_objectid = 0;
663 	root->nr_delalloc_inodes = 0;
664 	root->nr_ordered_extents = 0;
665 	xa_init(&root->inodes);
666 	xa_init(&root->delayed_nodes);
667 
668 	btrfs_init_root_block_rsv(root);
669 
670 	INIT_LIST_HEAD(&root->dirty_list);
671 	INIT_LIST_HEAD(&root->root_list);
672 	INIT_LIST_HEAD(&root->delalloc_inodes);
673 	INIT_LIST_HEAD(&root->delalloc_root);
674 	INIT_LIST_HEAD(&root->ordered_extents);
675 	INIT_LIST_HEAD(&root->ordered_root);
676 	INIT_LIST_HEAD(&root->reloc_dirty_list);
677 	spin_lock_init(&root->delalloc_lock);
678 	spin_lock_init(&root->ordered_extent_lock);
679 	spin_lock_init(&root->accounting_lock);
680 	spin_lock_init(&root->qgroup_meta_rsv_lock);
681 	mutex_init(&root->objectid_mutex);
682 	mutex_init(&root->log_mutex);
683 	mutex_init(&root->ordered_extent_mutex);
684 	mutex_init(&root->delalloc_mutex);
685 	init_waitqueue_head(&root->qgroup_flush_wait);
686 	init_waitqueue_head(&root->log_writer_wait);
687 	init_waitqueue_head(&root->log_commit_wait[0]);
688 	init_waitqueue_head(&root->log_commit_wait[1]);
689 	INIT_LIST_HEAD(&root->log_ctxs[0]);
690 	INIT_LIST_HEAD(&root->log_ctxs[1]);
691 	atomic_set(&root->log_commit[0], 0);
692 	atomic_set(&root->log_commit[1], 0);
693 	atomic_set(&root->log_writers, 0);
694 	atomic_set(&root->log_batch, 0);
695 	refcount_set(&root->refs, 1);
696 	atomic_set(&root->snapshot_force_cow, 0);
697 	atomic_set(&root->nr_swapfiles, 0);
698 	btrfs_set_root_log_transid(root, 0);
699 	root->log_transid_committed = -1;
700 	btrfs_set_root_last_log_commit(root, 0);
701 	root->anon_dev = 0;
702 	if (!dummy) {
703 		extent_io_tree_init(fs_info, &root->dirty_log_pages,
704 				    IO_TREE_ROOT_DIRTY_LOG_PAGES);
705 		extent_io_tree_init(fs_info, &root->log_csum_range,
706 				    IO_TREE_LOG_CSUM_RANGE);
707 	}
708 
709 	spin_lock_init(&root->root_item_lock);
710 	btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
711 #ifdef CONFIG_BTRFS_DEBUG
712 	INIT_LIST_HEAD(&root->leak_list);
713 	spin_lock(&fs_info->fs_roots_radix_lock);
714 	list_add_tail(&root->leak_list, &fs_info->allocated_roots);
715 	spin_unlock(&fs_info->fs_roots_radix_lock);
716 #endif
717 }
718 
719 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
720 					   u64 objectid, gfp_t flags)
721 {
722 	struct btrfs_root *root = kzalloc(sizeof(*root), flags);
723 	if (root)
724 		__setup_root(root, fs_info, objectid);
725 	return root;
726 }
727 
728 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
729 /* Should only be used by the testing infrastructure */
730 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
731 {
732 	struct btrfs_root *root;
733 
734 	if (!fs_info)
735 		return ERR_PTR(-EINVAL);
736 
737 	root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
738 	if (!root)
739 		return ERR_PTR(-ENOMEM);
740 
741 	/* We don't use the stripesize in selftest, set it as sectorsize */
742 	root->alloc_bytenr = 0;
743 
744 	return root;
745 }
746 #endif
747 
748 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
749 {
750 	const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
751 	const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
752 
753 	return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
754 }
755 
756 static int global_root_key_cmp(const void *k, const struct rb_node *node)
757 {
758 	const struct btrfs_key *key = k;
759 	const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
760 
761 	return btrfs_comp_cpu_keys(key, &root->root_key);
762 }
763 
764 int btrfs_global_root_insert(struct btrfs_root *root)
765 {
766 	struct btrfs_fs_info *fs_info = root->fs_info;
767 	struct rb_node *tmp;
768 	int ret = 0;
769 
770 	write_lock(&fs_info->global_root_lock);
771 	tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
772 	write_unlock(&fs_info->global_root_lock);
773 
774 	if (tmp) {
775 		ret = -EEXIST;
776 		btrfs_warn(fs_info, "global root %llu %llu already exists",
777 			   btrfs_root_id(root), root->root_key.offset);
778 	}
779 	return ret;
780 }
781 
782 void btrfs_global_root_delete(struct btrfs_root *root)
783 {
784 	struct btrfs_fs_info *fs_info = root->fs_info;
785 
786 	write_lock(&fs_info->global_root_lock);
787 	rb_erase(&root->rb_node, &fs_info->global_root_tree);
788 	write_unlock(&fs_info->global_root_lock);
789 }
790 
791 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
792 				     struct btrfs_key *key)
793 {
794 	struct rb_node *node;
795 	struct btrfs_root *root = NULL;
796 
797 	read_lock(&fs_info->global_root_lock);
798 	node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
799 	if (node)
800 		root = container_of(node, struct btrfs_root, rb_node);
801 	read_unlock(&fs_info->global_root_lock);
802 
803 	return root;
804 }
805 
806 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
807 {
808 	struct btrfs_block_group *block_group;
809 	u64 ret;
810 
811 	if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
812 		return 0;
813 
814 	if (bytenr)
815 		block_group = btrfs_lookup_block_group(fs_info, bytenr);
816 	else
817 		block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
818 	ASSERT(block_group);
819 	if (!block_group)
820 		return 0;
821 	ret = block_group->global_root_id;
822 	btrfs_put_block_group(block_group);
823 
824 	return ret;
825 }
826 
827 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
828 {
829 	struct btrfs_key key = {
830 		.objectid = BTRFS_CSUM_TREE_OBJECTID,
831 		.type = BTRFS_ROOT_ITEM_KEY,
832 		.offset = btrfs_global_root_id(fs_info, bytenr),
833 	};
834 
835 	return btrfs_global_root(fs_info, &key);
836 }
837 
838 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
839 {
840 	struct btrfs_key key = {
841 		.objectid = BTRFS_EXTENT_TREE_OBJECTID,
842 		.type = BTRFS_ROOT_ITEM_KEY,
843 		.offset = btrfs_global_root_id(fs_info, bytenr),
844 	};
845 
846 	return btrfs_global_root(fs_info, &key);
847 }
848 
849 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
850 				     u64 objectid)
851 {
852 	struct btrfs_fs_info *fs_info = trans->fs_info;
853 	struct extent_buffer *leaf;
854 	struct btrfs_root *tree_root = fs_info->tree_root;
855 	struct btrfs_root *root;
856 	struct btrfs_key key;
857 	unsigned int nofs_flag;
858 	int ret = 0;
859 
860 	/*
861 	 * We're holding a transaction handle, so use a NOFS memory allocation
862 	 * context to avoid deadlock if reclaim happens.
863 	 */
864 	nofs_flag = memalloc_nofs_save();
865 	root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
866 	memalloc_nofs_restore(nofs_flag);
867 	if (!root)
868 		return ERR_PTR(-ENOMEM);
869 
870 	root->root_key.objectid = objectid;
871 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
872 	root->root_key.offset = 0;
873 
874 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
875 				      0, BTRFS_NESTING_NORMAL);
876 	if (IS_ERR(leaf)) {
877 		ret = PTR_ERR(leaf);
878 		leaf = NULL;
879 		goto fail;
880 	}
881 
882 	root->node = leaf;
883 	btrfs_mark_buffer_dirty(trans, leaf);
884 
885 	root->commit_root = btrfs_root_node(root);
886 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
887 
888 	btrfs_set_root_flags(&root->root_item, 0);
889 	btrfs_set_root_limit(&root->root_item, 0);
890 	btrfs_set_root_bytenr(&root->root_item, leaf->start);
891 	btrfs_set_root_generation(&root->root_item, trans->transid);
892 	btrfs_set_root_level(&root->root_item, 0);
893 	btrfs_set_root_refs(&root->root_item, 1);
894 	btrfs_set_root_used(&root->root_item, leaf->len);
895 	btrfs_set_root_last_snapshot(&root->root_item, 0);
896 	btrfs_set_root_dirid(&root->root_item, 0);
897 	if (is_fstree(objectid))
898 		generate_random_guid(root->root_item.uuid);
899 	else
900 		export_guid(root->root_item.uuid, &guid_null);
901 	btrfs_set_root_drop_level(&root->root_item, 0);
902 
903 	btrfs_tree_unlock(leaf);
904 
905 	key.objectid = objectid;
906 	key.type = BTRFS_ROOT_ITEM_KEY;
907 	key.offset = 0;
908 	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
909 	if (ret)
910 		goto fail;
911 
912 	return root;
913 
914 fail:
915 	btrfs_put_root(root);
916 
917 	return ERR_PTR(ret);
918 }
919 
920 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
921 					 struct btrfs_fs_info *fs_info)
922 {
923 	struct btrfs_root *root;
924 
925 	root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
926 	if (!root)
927 		return ERR_PTR(-ENOMEM);
928 
929 	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
930 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
931 	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
932 
933 	return root;
934 }
935 
936 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
937 			      struct btrfs_root *root)
938 {
939 	struct extent_buffer *leaf;
940 
941 	/*
942 	 * DON'T set SHAREABLE bit for log trees.
943 	 *
944 	 * Log trees are not exposed to user space thus can't be snapshotted,
945 	 * and they go away before a real commit is actually done.
946 	 *
947 	 * They do store pointers to file data extents, and those reference
948 	 * counts still get updated (along with back refs to the log tree).
949 	 */
950 
951 	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
952 			NULL, 0, 0, 0, 0, BTRFS_NESTING_NORMAL);
953 	if (IS_ERR(leaf))
954 		return PTR_ERR(leaf);
955 
956 	root->node = leaf;
957 
958 	btrfs_mark_buffer_dirty(trans, root->node);
959 	btrfs_tree_unlock(root->node);
960 
961 	return 0;
962 }
963 
964 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
965 			     struct btrfs_fs_info *fs_info)
966 {
967 	struct btrfs_root *log_root;
968 
969 	log_root = alloc_log_tree(trans, fs_info);
970 	if (IS_ERR(log_root))
971 		return PTR_ERR(log_root);
972 
973 	if (!btrfs_is_zoned(fs_info)) {
974 		int ret = btrfs_alloc_log_tree_node(trans, log_root);
975 
976 		if (ret) {
977 			btrfs_put_root(log_root);
978 			return ret;
979 		}
980 	}
981 
982 	WARN_ON(fs_info->log_root_tree);
983 	fs_info->log_root_tree = log_root;
984 	return 0;
985 }
986 
987 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
988 		       struct btrfs_root *root)
989 {
990 	struct btrfs_fs_info *fs_info = root->fs_info;
991 	struct btrfs_root *log_root;
992 	struct btrfs_inode_item *inode_item;
993 	int ret;
994 
995 	log_root = alloc_log_tree(trans, fs_info);
996 	if (IS_ERR(log_root))
997 		return PTR_ERR(log_root);
998 
999 	ret = btrfs_alloc_log_tree_node(trans, log_root);
1000 	if (ret) {
1001 		btrfs_put_root(log_root);
1002 		return ret;
1003 	}
1004 
1005 	btrfs_set_root_last_trans(log_root, trans->transid);
1006 	log_root->root_key.offset = btrfs_root_id(root);
1007 
1008 	inode_item = &log_root->root_item.inode;
1009 	btrfs_set_stack_inode_generation(inode_item, 1);
1010 	btrfs_set_stack_inode_size(inode_item, 3);
1011 	btrfs_set_stack_inode_nlink(inode_item, 1);
1012 	btrfs_set_stack_inode_nbytes(inode_item,
1013 				     fs_info->nodesize);
1014 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1015 
1016 	btrfs_set_root_node(&log_root->root_item, log_root->node);
1017 
1018 	WARN_ON(root->log_root);
1019 	root->log_root = log_root;
1020 	btrfs_set_root_log_transid(root, 0);
1021 	root->log_transid_committed = -1;
1022 	btrfs_set_root_last_log_commit(root, 0);
1023 	return 0;
1024 }
1025 
1026 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1027 					      struct btrfs_path *path,
1028 					      const struct btrfs_key *key)
1029 {
1030 	struct btrfs_root *root;
1031 	struct btrfs_tree_parent_check check = { 0 };
1032 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1033 	u64 generation;
1034 	int ret;
1035 	int level;
1036 
1037 	root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1038 	if (!root)
1039 		return ERR_PTR(-ENOMEM);
1040 
1041 	ret = btrfs_find_root(tree_root, key, path,
1042 			      &root->root_item, &root->root_key);
1043 	if (ret) {
1044 		if (ret > 0)
1045 			ret = -ENOENT;
1046 		goto fail;
1047 	}
1048 
1049 	generation = btrfs_root_generation(&root->root_item);
1050 	level = btrfs_root_level(&root->root_item);
1051 	check.level = level;
1052 	check.transid = generation;
1053 	check.owner_root = key->objectid;
1054 	root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1055 				     &check);
1056 	if (IS_ERR(root->node)) {
1057 		ret = PTR_ERR(root->node);
1058 		root->node = NULL;
1059 		goto fail;
1060 	}
1061 	if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1062 		ret = -EIO;
1063 		goto fail;
1064 	}
1065 
1066 	/*
1067 	 * For real fs, and not log/reloc trees, root owner must
1068 	 * match its root node owner
1069 	 */
1070 	if (!btrfs_is_testing(fs_info) &&
1071 	    btrfs_root_id(root) != BTRFS_TREE_LOG_OBJECTID &&
1072 	    btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID &&
1073 	    btrfs_root_id(root) != btrfs_header_owner(root->node)) {
1074 		btrfs_crit(fs_info,
1075 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1076 			   btrfs_root_id(root), root->node->start,
1077 			   btrfs_header_owner(root->node),
1078 			   btrfs_root_id(root));
1079 		ret = -EUCLEAN;
1080 		goto fail;
1081 	}
1082 	root->commit_root = btrfs_root_node(root);
1083 	return root;
1084 fail:
1085 	btrfs_put_root(root);
1086 	return ERR_PTR(ret);
1087 }
1088 
1089 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1090 					const struct btrfs_key *key)
1091 {
1092 	struct btrfs_root *root;
1093 	struct btrfs_path *path;
1094 
1095 	path = btrfs_alloc_path();
1096 	if (!path)
1097 		return ERR_PTR(-ENOMEM);
1098 	root = read_tree_root_path(tree_root, path, key);
1099 	btrfs_free_path(path);
1100 
1101 	return root;
1102 }
1103 
1104 /*
1105  * Initialize subvolume root in-memory structure
1106  *
1107  * @anon_dev:	anonymous device to attach to the root, if zero, allocate new
1108  */
1109 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1110 {
1111 	int ret;
1112 
1113 	btrfs_drew_lock_init(&root->snapshot_lock);
1114 
1115 	if (btrfs_root_id(root) != BTRFS_TREE_LOG_OBJECTID &&
1116 	    !btrfs_is_data_reloc_root(root) &&
1117 	    is_fstree(btrfs_root_id(root))) {
1118 		set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1119 		btrfs_check_and_init_root_item(&root->root_item);
1120 	}
1121 
1122 	/*
1123 	 * Don't assign anonymous block device to roots that are not exposed to
1124 	 * userspace, the id pool is limited to 1M
1125 	 */
1126 	if (is_fstree(btrfs_root_id(root)) &&
1127 	    btrfs_root_refs(&root->root_item) > 0) {
1128 		if (!anon_dev) {
1129 			ret = get_anon_bdev(&root->anon_dev);
1130 			if (ret)
1131 				goto fail;
1132 		} else {
1133 			root->anon_dev = anon_dev;
1134 		}
1135 	}
1136 
1137 	mutex_lock(&root->objectid_mutex);
1138 	ret = btrfs_init_root_free_objectid(root);
1139 	if (ret) {
1140 		mutex_unlock(&root->objectid_mutex);
1141 		goto fail;
1142 	}
1143 
1144 	ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1145 
1146 	mutex_unlock(&root->objectid_mutex);
1147 
1148 	return 0;
1149 fail:
1150 	/* The caller is responsible to call btrfs_free_fs_root */
1151 	return ret;
1152 }
1153 
1154 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1155 					       u64 root_id)
1156 {
1157 	struct btrfs_root *root;
1158 
1159 	spin_lock(&fs_info->fs_roots_radix_lock);
1160 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1161 				 (unsigned long)root_id);
1162 	root = btrfs_grab_root(root);
1163 	spin_unlock(&fs_info->fs_roots_radix_lock);
1164 	return root;
1165 }
1166 
1167 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1168 						u64 objectid)
1169 {
1170 	struct btrfs_key key = {
1171 		.objectid = objectid,
1172 		.type = BTRFS_ROOT_ITEM_KEY,
1173 		.offset = 0,
1174 	};
1175 
1176 	switch (objectid) {
1177 	case BTRFS_ROOT_TREE_OBJECTID:
1178 		return btrfs_grab_root(fs_info->tree_root);
1179 	case BTRFS_EXTENT_TREE_OBJECTID:
1180 		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1181 	case BTRFS_CHUNK_TREE_OBJECTID:
1182 		return btrfs_grab_root(fs_info->chunk_root);
1183 	case BTRFS_DEV_TREE_OBJECTID:
1184 		return btrfs_grab_root(fs_info->dev_root);
1185 	case BTRFS_CSUM_TREE_OBJECTID:
1186 		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1187 	case BTRFS_QUOTA_TREE_OBJECTID:
1188 		return btrfs_grab_root(fs_info->quota_root);
1189 	case BTRFS_UUID_TREE_OBJECTID:
1190 		return btrfs_grab_root(fs_info->uuid_root);
1191 	case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
1192 		return btrfs_grab_root(fs_info->block_group_root);
1193 	case BTRFS_FREE_SPACE_TREE_OBJECTID:
1194 		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1195 	case BTRFS_RAID_STRIPE_TREE_OBJECTID:
1196 		return btrfs_grab_root(fs_info->stripe_root);
1197 	default:
1198 		return NULL;
1199 	}
1200 }
1201 
1202 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1203 			 struct btrfs_root *root)
1204 {
1205 	int ret;
1206 
1207 	ret = radix_tree_preload(GFP_NOFS);
1208 	if (ret)
1209 		return ret;
1210 
1211 	spin_lock(&fs_info->fs_roots_radix_lock);
1212 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1213 				(unsigned long)btrfs_root_id(root),
1214 				root);
1215 	if (ret == 0) {
1216 		btrfs_grab_root(root);
1217 		set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1218 	}
1219 	spin_unlock(&fs_info->fs_roots_radix_lock);
1220 	radix_tree_preload_end();
1221 
1222 	return ret;
1223 }
1224 
1225 void btrfs_check_leaked_roots(const struct btrfs_fs_info *fs_info)
1226 {
1227 #ifdef CONFIG_BTRFS_DEBUG
1228 	struct btrfs_root *root;
1229 
1230 	while (!list_empty(&fs_info->allocated_roots)) {
1231 		char buf[BTRFS_ROOT_NAME_BUF_LEN];
1232 
1233 		root = list_first_entry(&fs_info->allocated_roots,
1234 					struct btrfs_root, leak_list);
1235 		btrfs_err(fs_info, "leaked root %s refcount %d",
1236 			  btrfs_root_name(&root->root_key, buf),
1237 			  refcount_read(&root->refs));
1238 		WARN_ON_ONCE(1);
1239 		while (refcount_read(&root->refs) > 1)
1240 			btrfs_put_root(root);
1241 		btrfs_put_root(root);
1242 	}
1243 #endif
1244 }
1245 
1246 static void free_global_roots(struct btrfs_fs_info *fs_info)
1247 {
1248 	struct btrfs_root *root;
1249 	struct rb_node *node;
1250 
1251 	while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1252 		root = rb_entry(node, struct btrfs_root, rb_node);
1253 		rb_erase(&root->rb_node, &fs_info->global_root_tree);
1254 		btrfs_put_root(root);
1255 	}
1256 }
1257 
1258 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1259 {
1260 	struct percpu_counter *em_counter = &fs_info->evictable_extent_maps;
1261 
1262 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1263 	percpu_counter_destroy(&fs_info->delalloc_bytes);
1264 	percpu_counter_destroy(&fs_info->ordered_bytes);
1265 	if (percpu_counter_initialized(em_counter))
1266 		ASSERT(percpu_counter_sum_positive(em_counter) == 0);
1267 	percpu_counter_destroy(em_counter);
1268 	percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1269 	btrfs_free_csum_hash(fs_info);
1270 	btrfs_free_stripe_hash_table(fs_info);
1271 	btrfs_free_ref_cache(fs_info);
1272 	kfree(fs_info->balance_ctl);
1273 	kfree(fs_info->delayed_root);
1274 	free_global_roots(fs_info);
1275 	btrfs_put_root(fs_info->tree_root);
1276 	btrfs_put_root(fs_info->chunk_root);
1277 	btrfs_put_root(fs_info->dev_root);
1278 	btrfs_put_root(fs_info->quota_root);
1279 	btrfs_put_root(fs_info->uuid_root);
1280 	btrfs_put_root(fs_info->fs_root);
1281 	btrfs_put_root(fs_info->data_reloc_root);
1282 	btrfs_put_root(fs_info->block_group_root);
1283 	btrfs_put_root(fs_info->stripe_root);
1284 	btrfs_check_leaked_roots(fs_info);
1285 	btrfs_extent_buffer_leak_debug_check(fs_info);
1286 	kfree(fs_info->super_copy);
1287 	kfree(fs_info->super_for_commit);
1288 	kvfree(fs_info);
1289 }
1290 
1291 
1292 /*
1293  * Get an in-memory reference of a root structure.
1294  *
1295  * For essential trees like root/extent tree, we grab it from fs_info directly.
1296  * For subvolume trees, we check the cached filesystem roots first. If not
1297  * found, then read it from disk and add it to cached fs roots.
1298  *
1299  * Caller should release the root by calling btrfs_put_root() after the usage.
1300  *
1301  * NOTE: Reloc and log trees can't be read by this function as they share the
1302  *	 same root objectid.
1303  *
1304  * @objectid:	root id
1305  * @anon_dev:	preallocated anonymous block device number for new roots,
1306  *		pass NULL for a new allocation.
1307  * @check_ref:	whether to check root item references, If true, return -ENOENT
1308  *		for orphan roots
1309  */
1310 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1311 					     u64 objectid, dev_t *anon_dev,
1312 					     bool check_ref)
1313 {
1314 	struct btrfs_root *root;
1315 	struct btrfs_path *path;
1316 	struct btrfs_key key;
1317 	int ret;
1318 
1319 	root = btrfs_get_global_root(fs_info, objectid);
1320 	if (root)
1321 		return root;
1322 
1323 	/*
1324 	 * If we're called for non-subvolume trees, and above function didn't
1325 	 * find one, do not try to read it from disk.
1326 	 *
1327 	 * This is namely for free-space-tree and quota tree, which can change
1328 	 * at runtime and should only be grabbed from fs_info.
1329 	 */
1330 	if (!is_fstree(objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1331 		return ERR_PTR(-ENOENT);
1332 again:
1333 	root = btrfs_lookup_fs_root(fs_info, objectid);
1334 	if (root) {
1335 		/*
1336 		 * Some other caller may have read out the newly inserted
1337 		 * subvolume already (for things like backref walk etc).  Not
1338 		 * that common but still possible.  In that case, we just need
1339 		 * to free the anon_dev.
1340 		 */
1341 		if (unlikely(anon_dev && *anon_dev)) {
1342 			free_anon_bdev(*anon_dev);
1343 			*anon_dev = 0;
1344 		}
1345 
1346 		if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1347 			btrfs_put_root(root);
1348 			return ERR_PTR(-ENOENT);
1349 		}
1350 		return root;
1351 	}
1352 
1353 	key.objectid = objectid;
1354 	key.type = BTRFS_ROOT_ITEM_KEY;
1355 	key.offset = (u64)-1;
1356 	root = btrfs_read_tree_root(fs_info->tree_root, &key);
1357 	if (IS_ERR(root))
1358 		return root;
1359 
1360 	if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1361 		ret = -ENOENT;
1362 		goto fail;
1363 	}
1364 
1365 	ret = btrfs_init_fs_root(root, anon_dev ? *anon_dev : 0);
1366 	if (ret)
1367 		goto fail;
1368 
1369 	path = btrfs_alloc_path();
1370 	if (!path) {
1371 		ret = -ENOMEM;
1372 		goto fail;
1373 	}
1374 	key.objectid = BTRFS_ORPHAN_OBJECTID;
1375 	key.type = BTRFS_ORPHAN_ITEM_KEY;
1376 	key.offset = objectid;
1377 
1378 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1379 	btrfs_free_path(path);
1380 	if (ret < 0)
1381 		goto fail;
1382 	if (ret == 0)
1383 		set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1384 
1385 	ret = btrfs_insert_fs_root(fs_info, root);
1386 	if (ret) {
1387 		if (ret == -EEXIST) {
1388 			btrfs_put_root(root);
1389 			goto again;
1390 		}
1391 		goto fail;
1392 	}
1393 	return root;
1394 fail:
1395 	/*
1396 	 * If our caller provided us an anonymous device, then it's his
1397 	 * responsibility to free it in case we fail. So we have to set our
1398 	 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1399 	 * and once again by our caller.
1400 	 */
1401 	if (anon_dev && *anon_dev)
1402 		root->anon_dev = 0;
1403 	btrfs_put_root(root);
1404 	return ERR_PTR(ret);
1405 }
1406 
1407 /*
1408  * Get in-memory reference of a root structure
1409  *
1410  * @objectid:	tree objectid
1411  * @check_ref:	if set, verify that the tree exists and the item has at least
1412  *		one reference
1413  */
1414 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1415 				     u64 objectid, bool check_ref)
1416 {
1417 	return btrfs_get_root_ref(fs_info, objectid, NULL, check_ref);
1418 }
1419 
1420 /*
1421  * Get in-memory reference of a root structure, created as new, optionally pass
1422  * the anonymous block device id
1423  *
1424  * @objectid:	tree objectid
1425  * @anon_dev:	if NULL, allocate a new anonymous block device or use the
1426  *		parameter value if not NULL
1427  */
1428 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1429 					 u64 objectid, dev_t *anon_dev)
1430 {
1431 	return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1432 }
1433 
1434 /*
1435  * Return a root for the given objectid.
1436  *
1437  * @fs_info:	the fs_info
1438  * @objectid:	the objectid we need to lookup
1439  *
1440  * This is exclusively used for backref walking, and exists specifically because
1441  * of how qgroups does lookups.  Qgroups will do a backref lookup at delayed ref
1442  * creation time, which means we may have to read the tree_root in order to look
1443  * up a fs root that is not in memory.  If the root is not in memory we will
1444  * read the tree root commit root and look up the fs root from there.  This is a
1445  * temporary root, it will not be inserted into the radix tree as it doesn't
1446  * have the most uptodate information, it'll simply be discarded once the
1447  * backref code is finished using the root.
1448  */
1449 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1450 						 struct btrfs_path *path,
1451 						 u64 objectid)
1452 {
1453 	struct btrfs_root *root;
1454 	struct btrfs_key key;
1455 
1456 	ASSERT(path->search_commit_root && path->skip_locking);
1457 
1458 	/*
1459 	 * This can return -ENOENT if we ask for a root that doesn't exist, but
1460 	 * since this is called via the backref walking code we won't be looking
1461 	 * up a root that doesn't exist, unless there's corruption.  So if root
1462 	 * != NULL just return it.
1463 	 */
1464 	root = btrfs_get_global_root(fs_info, objectid);
1465 	if (root)
1466 		return root;
1467 
1468 	root = btrfs_lookup_fs_root(fs_info, objectid);
1469 	if (root)
1470 		return root;
1471 
1472 	key.objectid = objectid;
1473 	key.type = BTRFS_ROOT_ITEM_KEY;
1474 	key.offset = (u64)-1;
1475 	root = read_tree_root_path(fs_info->tree_root, path, &key);
1476 	btrfs_release_path(path);
1477 
1478 	return root;
1479 }
1480 
1481 static int cleaner_kthread(void *arg)
1482 {
1483 	struct btrfs_fs_info *fs_info = arg;
1484 	int again;
1485 
1486 	while (1) {
1487 		again = 0;
1488 
1489 		set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1490 
1491 		/* Make the cleaner go to sleep early. */
1492 		if (btrfs_need_cleaner_sleep(fs_info))
1493 			goto sleep;
1494 
1495 		/*
1496 		 * Do not do anything if we might cause open_ctree() to block
1497 		 * before we have finished mounting the filesystem.
1498 		 */
1499 		if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1500 			goto sleep;
1501 
1502 		if (!mutex_trylock(&fs_info->cleaner_mutex))
1503 			goto sleep;
1504 
1505 		/*
1506 		 * Avoid the problem that we change the status of the fs
1507 		 * during the above check and trylock.
1508 		 */
1509 		if (btrfs_need_cleaner_sleep(fs_info)) {
1510 			mutex_unlock(&fs_info->cleaner_mutex);
1511 			goto sleep;
1512 		}
1513 
1514 		if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1515 			btrfs_sysfs_feature_update(fs_info);
1516 
1517 		btrfs_run_delayed_iputs(fs_info);
1518 
1519 		again = btrfs_clean_one_deleted_snapshot(fs_info);
1520 		mutex_unlock(&fs_info->cleaner_mutex);
1521 
1522 		/*
1523 		 * The defragger has dealt with the R/O remount and umount,
1524 		 * needn't do anything special here.
1525 		 */
1526 		btrfs_run_defrag_inodes(fs_info);
1527 
1528 		/*
1529 		 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1530 		 * with relocation (btrfs_relocate_chunk) and relocation
1531 		 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1532 		 * after acquiring fs_info->reclaim_bgs_lock. So we
1533 		 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1534 		 * unused block groups.
1535 		 */
1536 		btrfs_delete_unused_bgs(fs_info);
1537 
1538 		/*
1539 		 * Reclaim block groups in the reclaim_bgs list after we deleted
1540 		 * all unused block_groups. This possibly gives us some more free
1541 		 * space.
1542 		 */
1543 		btrfs_reclaim_bgs(fs_info);
1544 sleep:
1545 		clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1546 		if (kthread_should_park())
1547 			kthread_parkme();
1548 		if (kthread_should_stop())
1549 			return 0;
1550 		if (!again) {
1551 			set_current_state(TASK_INTERRUPTIBLE);
1552 			schedule();
1553 			__set_current_state(TASK_RUNNING);
1554 		}
1555 	}
1556 }
1557 
1558 static int transaction_kthread(void *arg)
1559 {
1560 	struct btrfs_root *root = arg;
1561 	struct btrfs_fs_info *fs_info = root->fs_info;
1562 	struct btrfs_trans_handle *trans;
1563 	struct btrfs_transaction *cur;
1564 	u64 transid;
1565 	time64_t delta;
1566 	unsigned long delay;
1567 	bool cannot_commit;
1568 
1569 	do {
1570 		cannot_commit = false;
1571 		delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1572 		mutex_lock(&fs_info->transaction_kthread_mutex);
1573 
1574 		spin_lock(&fs_info->trans_lock);
1575 		cur = fs_info->running_transaction;
1576 		if (!cur) {
1577 			spin_unlock(&fs_info->trans_lock);
1578 			goto sleep;
1579 		}
1580 
1581 		delta = ktime_get_seconds() - cur->start_time;
1582 		if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1583 		    cur->state < TRANS_STATE_COMMIT_PREP &&
1584 		    delta < fs_info->commit_interval) {
1585 			spin_unlock(&fs_info->trans_lock);
1586 			delay -= msecs_to_jiffies((delta - 1) * 1000);
1587 			delay = min(delay,
1588 				    msecs_to_jiffies(fs_info->commit_interval * 1000));
1589 			goto sleep;
1590 		}
1591 		transid = cur->transid;
1592 		spin_unlock(&fs_info->trans_lock);
1593 
1594 		/* If the file system is aborted, this will always fail. */
1595 		trans = btrfs_attach_transaction(root);
1596 		if (IS_ERR(trans)) {
1597 			if (PTR_ERR(trans) != -ENOENT)
1598 				cannot_commit = true;
1599 			goto sleep;
1600 		}
1601 		if (transid == trans->transid) {
1602 			btrfs_commit_transaction(trans);
1603 		} else {
1604 			btrfs_end_transaction(trans);
1605 		}
1606 sleep:
1607 		wake_up_process(fs_info->cleaner_kthread);
1608 		mutex_unlock(&fs_info->transaction_kthread_mutex);
1609 
1610 		if (BTRFS_FS_ERROR(fs_info))
1611 			btrfs_cleanup_transaction(fs_info);
1612 		if (!kthread_should_stop() &&
1613 				(!btrfs_transaction_blocked(fs_info) ||
1614 				 cannot_commit))
1615 			schedule_timeout_interruptible(delay);
1616 	} while (!kthread_should_stop());
1617 	return 0;
1618 }
1619 
1620 /*
1621  * This will find the highest generation in the array of root backups.  The
1622  * index of the highest array is returned, or -EINVAL if we can't find
1623  * anything.
1624  *
1625  * We check to make sure the array is valid by comparing the
1626  * generation of the latest  root in the array with the generation
1627  * in the super block.  If they don't match we pitch it.
1628  */
1629 static int find_newest_super_backup(struct btrfs_fs_info *info)
1630 {
1631 	const u64 newest_gen = btrfs_super_generation(info->super_copy);
1632 	u64 cur;
1633 	struct btrfs_root_backup *root_backup;
1634 	int i;
1635 
1636 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1637 		root_backup = info->super_copy->super_roots + i;
1638 		cur = btrfs_backup_tree_root_gen(root_backup);
1639 		if (cur == newest_gen)
1640 			return i;
1641 	}
1642 
1643 	return -EINVAL;
1644 }
1645 
1646 /*
1647  * copy all the root pointers into the super backup array.
1648  * this will bump the backup pointer by one when it is
1649  * done
1650  */
1651 static void backup_super_roots(struct btrfs_fs_info *info)
1652 {
1653 	const int next_backup = info->backup_root_index;
1654 	struct btrfs_root_backup *root_backup;
1655 
1656 	root_backup = info->super_for_commit->super_roots + next_backup;
1657 
1658 	/*
1659 	 * make sure all of our padding and empty slots get zero filled
1660 	 * regardless of which ones we use today
1661 	 */
1662 	memset(root_backup, 0, sizeof(*root_backup));
1663 
1664 	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1665 
1666 	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1667 	btrfs_set_backup_tree_root_gen(root_backup,
1668 			       btrfs_header_generation(info->tree_root->node));
1669 
1670 	btrfs_set_backup_tree_root_level(root_backup,
1671 			       btrfs_header_level(info->tree_root->node));
1672 
1673 	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1674 	btrfs_set_backup_chunk_root_gen(root_backup,
1675 			       btrfs_header_generation(info->chunk_root->node));
1676 	btrfs_set_backup_chunk_root_level(root_backup,
1677 			       btrfs_header_level(info->chunk_root->node));
1678 
1679 	if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1680 		struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1681 		struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1682 
1683 		btrfs_set_backup_extent_root(root_backup,
1684 					     extent_root->node->start);
1685 		btrfs_set_backup_extent_root_gen(root_backup,
1686 				btrfs_header_generation(extent_root->node));
1687 		btrfs_set_backup_extent_root_level(root_backup,
1688 					btrfs_header_level(extent_root->node));
1689 
1690 		btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
1691 		btrfs_set_backup_csum_root_gen(root_backup,
1692 					       btrfs_header_generation(csum_root->node));
1693 		btrfs_set_backup_csum_root_level(root_backup,
1694 						 btrfs_header_level(csum_root->node));
1695 	}
1696 
1697 	/*
1698 	 * we might commit during log recovery, which happens before we set
1699 	 * the fs_root.  Make sure it is valid before we fill it in.
1700 	 */
1701 	if (info->fs_root && info->fs_root->node) {
1702 		btrfs_set_backup_fs_root(root_backup,
1703 					 info->fs_root->node->start);
1704 		btrfs_set_backup_fs_root_gen(root_backup,
1705 			       btrfs_header_generation(info->fs_root->node));
1706 		btrfs_set_backup_fs_root_level(root_backup,
1707 			       btrfs_header_level(info->fs_root->node));
1708 	}
1709 
1710 	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1711 	btrfs_set_backup_dev_root_gen(root_backup,
1712 			       btrfs_header_generation(info->dev_root->node));
1713 	btrfs_set_backup_dev_root_level(root_backup,
1714 				       btrfs_header_level(info->dev_root->node));
1715 
1716 	btrfs_set_backup_total_bytes(root_backup,
1717 			     btrfs_super_total_bytes(info->super_copy));
1718 	btrfs_set_backup_bytes_used(root_backup,
1719 			     btrfs_super_bytes_used(info->super_copy));
1720 	btrfs_set_backup_num_devices(root_backup,
1721 			     btrfs_super_num_devices(info->super_copy));
1722 
1723 	/*
1724 	 * if we don't copy this out to the super_copy, it won't get remembered
1725 	 * for the next commit
1726 	 */
1727 	memcpy(&info->super_copy->super_roots,
1728 	       &info->super_for_commit->super_roots,
1729 	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1730 }
1731 
1732 /*
1733  * Reads a backup root based on the passed priority. Prio 0 is the newest, prio
1734  * 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1735  *
1736  * @fs_info:  filesystem whose backup roots need to be read
1737  * @priority: priority of backup root required
1738  *
1739  * Returns backup root index on success and -EINVAL otherwise.
1740  */
1741 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1742 {
1743 	int backup_index = find_newest_super_backup(fs_info);
1744 	struct btrfs_super_block *super = fs_info->super_copy;
1745 	struct btrfs_root_backup *root_backup;
1746 
1747 	if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1748 		if (priority == 0)
1749 			return backup_index;
1750 
1751 		backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1752 		backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1753 	} else {
1754 		return -EINVAL;
1755 	}
1756 
1757 	root_backup = super->super_roots + backup_index;
1758 
1759 	btrfs_set_super_generation(super,
1760 				   btrfs_backup_tree_root_gen(root_backup));
1761 	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1762 	btrfs_set_super_root_level(super,
1763 				   btrfs_backup_tree_root_level(root_backup));
1764 	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1765 
1766 	/*
1767 	 * Fixme: the total bytes and num_devices need to match or we should
1768 	 * need a fsck
1769 	 */
1770 	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1771 	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1772 
1773 	return backup_index;
1774 }
1775 
1776 /* helper to cleanup workers */
1777 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1778 {
1779 	btrfs_destroy_workqueue(fs_info->fixup_workers);
1780 	btrfs_destroy_workqueue(fs_info->delalloc_workers);
1781 	btrfs_destroy_workqueue(fs_info->workers);
1782 	if (fs_info->endio_workers)
1783 		destroy_workqueue(fs_info->endio_workers);
1784 	if (fs_info->rmw_workers)
1785 		destroy_workqueue(fs_info->rmw_workers);
1786 	if (fs_info->compressed_write_workers)
1787 		destroy_workqueue(fs_info->compressed_write_workers);
1788 	btrfs_destroy_workqueue(fs_info->endio_write_workers);
1789 	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1790 	btrfs_destroy_workqueue(fs_info->delayed_workers);
1791 	btrfs_destroy_workqueue(fs_info->caching_workers);
1792 	btrfs_destroy_workqueue(fs_info->flush_workers);
1793 	btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1794 	if (fs_info->discard_ctl.discard_workers)
1795 		destroy_workqueue(fs_info->discard_ctl.discard_workers);
1796 	/*
1797 	 * Now that all other work queues are destroyed, we can safely destroy
1798 	 * the queues used for metadata I/O, since tasks from those other work
1799 	 * queues can do metadata I/O operations.
1800 	 */
1801 	if (fs_info->endio_meta_workers)
1802 		destroy_workqueue(fs_info->endio_meta_workers);
1803 }
1804 
1805 static void free_root_extent_buffers(struct btrfs_root *root)
1806 {
1807 	if (root) {
1808 		free_extent_buffer(root->node);
1809 		free_extent_buffer(root->commit_root);
1810 		root->node = NULL;
1811 		root->commit_root = NULL;
1812 	}
1813 }
1814 
1815 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1816 {
1817 	struct btrfs_root *root, *tmp;
1818 
1819 	rbtree_postorder_for_each_entry_safe(root, tmp,
1820 					     &fs_info->global_root_tree,
1821 					     rb_node)
1822 		free_root_extent_buffers(root);
1823 }
1824 
1825 /* helper to cleanup tree roots */
1826 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1827 {
1828 	free_root_extent_buffers(info->tree_root);
1829 
1830 	free_global_root_pointers(info);
1831 	free_root_extent_buffers(info->dev_root);
1832 	free_root_extent_buffers(info->quota_root);
1833 	free_root_extent_buffers(info->uuid_root);
1834 	free_root_extent_buffers(info->fs_root);
1835 	free_root_extent_buffers(info->data_reloc_root);
1836 	free_root_extent_buffers(info->block_group_root);
1837 	free_root_extent_buffers(info->stripe_root);
1838 	if (free_chunk_root)
1839 		free_root_extent_buffers(info->chunk_root);
1840 }
1841 
1842 void btrfs_put_root(struct btrfs_root *root)
1843 {
1844 	if (!root)
1845 		return;
1846 
1847 	if (refcount_dec_and_test(&root->refs)) {
1848 		if (WARN_ON(!xa_empty(&root->inodes)))
1849 			xa_destroy(&root->inodes);
1850 		WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1851 		if (root->anon_dev)
1852 			free_anon_bdev(root->anon_dev);
1853 		free_root_extent_buffers(root);
1854 #ifdef CONFIG_BTRFS_DEBUG
1855 		spin_lock(&root->fs_info->fs_roots_radix_lock);
1856 		list_del_init(&root->leak_list);
1857 		spin_unlock(&root->fs_info->fs_roots_radix_lock);
1858 #endif
1859 		kfree(root);
1860 	}
1861 }
1862 
1863 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1864 {
1865 	int ret;
1866 	struct btrfs_root *gang[8];
1867 	int i;
1868 
1869 	while (!list_empty(&fs_info->dead_roots)) {
1870 		gang[0] = list_entry(fs_info->dead_roots.next,
1871 				     struct btrfs_root, root_list);
1872 		list_del(&gang[0]->root_list);
1873 
1874 		if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1875 			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
1876 		btrfs_put_root(gang[0]);
1877 	}
1878 
1879 	while (1) {
1880 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1881 					     (void **)gang, 0,
1882 					     ARRAY_SIZE(gang));
1883 		if (!ret)
1884 			break;
1885 		for (i = 0; i < ret; i++)
1886 			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
1887 	}
1888 }
1889 
1890 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1891 {
1892 	mutex_init(&fs_info->scrub_lock);
1893 	atomic_set(&fs_info->scrubs_running, 0);
1894 	atomic_set(&fs_info->scrub_pause_req, 0);
1895 	atomic_set(&fs_info->scrubs_paused, 0);
1896 	atomic_set(&fs_info->scrub_cancel_req, 0);
1897 	init_waitqueue_head(&fs_info->scrub_pause_wait);
1898 	refcount_set(&fs_info->scrub_workers_refcnt, 0);
1899 }
1900 
1901 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1902 {
1903 	spin_lock_init(&fs_info->balance_lock);
1904 	mutex_init(&fs_info->balance_mutex);
1905 	atomic_set(&fs_info->balance_pause_req, 0);
1906 	atomic_set(&fs_info->balance_cancel_req, 0);
1907 	fs_info->balance_ctl = NULL;
1908 	init_waitqueue_head(&fs_info->balance_wait_q);
1909 	atomic_set(&fs_info->reloc_cancel_req, 0);
1910 }
1911 
1912 static int btrfs_init_btree_inode(struct super_block *sb)
1913 {
1914 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1915 	unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1916 					      fs_info->tree_root);
1917 	struct inode *inode;
1918 
1919 	inode = new_inode(sb);
1920 	if (!inode)
1921 		return -ENOMEM;
1922 
1923 	btrfs_set_inode_number(BTRFS_I(inode), BTRFS_BTREE_INODE_OBJECTID);
1924 	set_nlink(inode, 1);
1925 	/*
1926 	 * we set the i_size on the btree inode to the max possible int.
1927 	 * the real end of the address space is determined by all of
1928 	 * the devices in the system
1929 	 */
1930 	inode->i_size = OFFSET_MAX;
1931 	inode->i_mapping->a_ops = &btree_aops;
1932 	mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
1933 
1934 	extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
1935 			    IO_TREE_BTREE_INODE_IO);
1936 	extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
1937 
1938 	BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
1939 	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
1940 	__insert_inode_hash(inode, hash);
1941 	fs_info->btree_inode = inode;
1942 
1943 	return 0;
1944 }
1945 
1946 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1947 {
1948 	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1949 	init_rwsem(&fs_info->dev_replace.rwsem);
1950 	init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1951 }
1952 
1953 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1954 {
1955 	spin_lock_init(&fs_info->qgroup_lock);
1956 	mutex_init(&fs_info->qgroup_ioctl_lock);
1957 	fs_info->qgroup_tree = RB_ROOT;
1958 	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
1959 	fs_info->qgroup_seq = 1;
1960 	fs_info->qgroup_ulist = NULL;
1961 	fs_info->qgroup_rescan_running = false;
1962 	fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1963 	mutex_init(&fs_info->qgroup_rescan_lock);
1964 }
1965 
1966 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
1967 {
1968 	u32 max_active = fs_info->thread_pool_size;
1969 	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1970 	unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE;
1971 
1972 	fs_info->workers =
1973 		btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
1974 
1975 	fs_info->delalloc_workers =
1976 		btrfs_alloc_workqueue(fs_info, "delalloc",
1977 				      flags, max_active, 2);
1978 
1979 	fs_info->flush_workers =
1980 		btrfs_alloc_workqueue(fs_info, "flush_delalloc",
1981 				      flags, max_active, 0);
1982 
1983 	fs_info->caching_workers =
1984 		btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
1985 
1986 	fs_info->fixup_workers =
1987 		btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags);
1988 
1989 	fs_info->endio_workers =
1990 		alloc_workqueue("btrfs-endio", flags, max_active);
1991 	fs_info->endio_meta_workers =
1992 		alloc_workqueue("btrfs-endio-meta", flags, max_active);
1993 	fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
1994 	fs_info->endio_write_workers =
1995 		btrfs_alloc_workqueue(fs_info, "endio-write", flags,
1996 				      max_active, 2);
1997 	fs_info->compressed_write_workers =
1998 		alloc_workqueue("btrfs-compressed-write", flags, max_active);
1999 	fs_info->endio_freespace_worker =
2000 		btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2001 				      max_active, 0);
2002 	fs_info->delayed_workers =
2003 		btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2004 				      max_active, 0);
2005 	fs_info->qgroup_rescan_workers =
2006 		btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan",
2007 					      ordered_flags);
2008 	fs_info->discard_ctl.discard_workers =
2009 		alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE);
2010 
2011 	if (!(fs_info->workers &&
2012 	      fs_info->delalloc_workers && fs_info->flush_workers &&
2013 	      fs_info->endio_workers && fs_info->endio_meta_workers &&
2014 	      fs_info->compressed_write_workers &&
2015 	      fs_info->endio_write_workers &&
2016 	      fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2017 	      fs_info->caching_workers && fs_info->fixup_workers &&
2018 	      fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2019 	      fs_info->discard_ctl.discard_workers)) {
2020 		return -ENOMEM;
2021 	}
2022 
2023 	return 0;
2024 }
2025 
2026 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2027 {
2028 	struct crypto_shash *csum_shash;
2029 	const char *csum_driver = btrfs_super_csum_driver(csum_type);
2030 
2031 	csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2032 
2033 	if (IS_ERR(csum_shash)) {
2034 		btrfs_err(fs_info, "error allocating %s hash for checksum",
2035 			  csum_driver);
2036 		return PTR_ERR(csum_shash);
2037 	}
2038 
2039 	fs_info->csum_shash = csum_shash;
2040 
2041 	/*
2042 	 * Check if the checksum implementation is a fast accelerated one.
2043 	 * As-is this is a bit of a hack and should be replaced once the csum
2044 	 * implementations provide that information themselves.
2045 	 */
2046 	switch (csum_type) {
2047 	case BTRFS_CSUM_TYPE_CRC32:
2048 		if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2049 			set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2050 		break;
2051 	case BTRFS_CSUM_TYPE_XXHASH:
2052 		set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2053 		break;
2054 	default:
2055 		break;
2056 	}
2057 
2058 	btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2059 			btrfs_super_csum_name(csum_type),
2060 			crypto_shash_driver_name(csum_shash));
2061 	return 0;
2062 }
2063 
2064 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2065 			    struct btrfs_fs_devices *fs_devices)
2066 {
2067 	int ret;
2068 	struct btrfs_tree_parent_check check = { 0 };
2069 	struct btrfs_root *log_tree_root;
2070 	struct btrfs_super_block *disk_super = fs_info->super_copy;
2071 	u64 bytenr = btrfs_super_log_root(disk_super);
2072 	int level = btrfs_super_log_root_level(disk_super);
2073 
2074 	if (fs_devices->rw_devices == 0) {
2075 		btrfs_warn(fs_info, "log replay required on RO media");
2076 		return -EIO;
2077 	}
2078 
2079 	log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2080 					 GFP_KERNEL);
2081 	if (!log_tree_root)
2082 		return -ENOMEM;
2083 
2084 	check.level = level;
2085 	check.transid = fs_info->generation + 1;
2086 	check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2087 	log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2088 	if (IS_ERR(log_tree_root->node)) {
2089 		btrfs_warn(fs_info, "failed to read log tree");
2090 		ret = PTR_ERR(log_tree_root->node);
2091 		log_tree_root->node = NULL;
2092 		btrfs_put_root(log_tree_root);
2093 		return ret;
2094 	}
2095 	if (!extent_buffer_uptodate(log_tree_root->node)) {
2096 		btrfs_err(fs_info, "failed to read log tree");
2097 		btrfs_put_root(log_tree_root);
2098 		return -EIO;
2099 	}
2100 
2101 	/* returns with log_tree_root freed on success */
2102 	ret = btrfs_recover_log_trees(log_tree_root);
2103 	if (ret) {
2104 		btrfs_handle_fs_error(fs_info, ret,
2105 				      "Failed to recover log tree");
2106 		btrfs_put_root(log_tree_root);
2107 		return ret;
2108 	}
2109 
2110 	if (sb_rdonly(fs_info->sb)) {
2111 		ret = btrfs_commit_super(fs_info);
2112 		if (ret)
2113 			return ret;
2114 	}
2115 
2116 	return 0;
2117 }
2118 
2119 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2120 				      struct btrfs_path *path, u64 objectid,
2121 				      const char *name)
2122 {
2123 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
2124 	struct btrfs_root *root;
2125 	u64 max_global_id = 0;
2126 	int ret;
2127 	struct btrfs_key key = {
2128 		.objectid = objectid,
2129 		.type = BTRFS_ROOT_ITEM_KEY,
2130 		.offset = 0,
2131 	};
2132 	bool found = false;
2133 
2134 	/* If we have IGNOREDATACSUMS skip loading these roots. */
2135 	if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2136 	    btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2137 		set_bit(BTRFS_FS_STATE_NO_DATA_CSUMS, &fs_info->fs_state);
2138 		return 0;
2139 	}
2140 
2141 	while (1) {
2142 		ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2143 		if (ret < 0)
2144 			break;
2145 
2146 		if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2147 			ret = btrfs_next_leaf(tree_root, path);
2148 			if (ret) {
2149 				if (ret > 0)
2150 					ret = 0;
2151 				break;
2152 			}
2153 		}
2154 		ret = 0;
2155 
2156 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2157 		if (key.objectid != objectid)
2158 			break;
2159 		btrfs_release_path(path);
2160 
2161 		/*
2162 		 * Just worry about this for extent tree, it'll be the same for
2163 		 * everybody.
2164 		 */
2165 		if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2166 			max_global_id = max(max_global_id, key.offset);
2167 
2168 		found = true;
2169 		root = read_tree_root_path(tree_root, path, &key);
2170 		if (IS_ERR(root)) {
2171 			if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2172 				ret = PTR_ERR(root);
2173 			break;
2174 		}
2175 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2176 		ret = btrfs_global_root_insert(root);
2177 		if (ret) {
2178 			btrfs_put_root(root);
2179 			break;
2180 		}
2181 		key.offset++;
2182 	}
2183 	btrfs_release_path(path);
2184 
2185 	if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2186 		fs_info->nr_global_roots = max_global_id + 1;
2187 
2188 	if (!found || ret) {
2189 		if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2190 			set_bit(BTRFS_FS_STATE_NO_DATA_CSUMS, &fs_info->fs_state);
2191 
2192 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2193 			ret = ret ? ret : -ENOENT;
2194 		else
2195 			ret = 0;
2196 		btrfs_err(fs_info, "failed to load root %s", name);
2197 	}
2198 	return ret;
2199 }
2200 
2201 static int load_global_roots(struct btrfs_root *tree_root)
2202 {
2203 	struct btrfs_path *path;
2204 	int ret = 0;
2205 
2206 	path = btrfs_alloc_path();
2207 	if (!path)
2208 		return -ENOMEM;
2209 
2210 	ret = load_global_roots_objectid(tree_root, path,
2211 					 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2212 	if (ret)
2213 		goto out;
2214 	ret = load_global_roots_objectid(tree_root, path,
2215 					 BTRFS_CSUM_TREE_OBJECTID, "csum");
2216 	if (ret)
2217 		goto out;
2218 	if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2219 		goto out;
2220 	ret = load_global_roots_objectid(tree_root, path,
2221 					 BTRFS_FREE_SPACE_TREE_OBJECTID,
2222 					 "free space");
2223 out:
2224 	btrfs_free_path(path);
2225 	return ret;
2226 }
2227 
2228 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2229 {
2230 	struct btrfs_root *tree_root = fs_info->tree_root;
2231 	struct btrfs_root *root;
2232 	struct btrfs_key location;
2233 	int ret;
2234 
2235 	ASSERT(fs_info->tree_root);
2236 
2237 	ret = load_global_roots(tree_root);
2238 	if (ret)
2239 		return ret;
2240 
2241 	location.type = BTRFS_ROOT_ITEM_KEY;
2242 	location.offset = 0;
2243 
2244 	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2245 		location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2246 		root = btrfs_read_tree_root(tree_root, &location);
2247 		if (IS_ERR(root)) {
2248 			if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2249 				ret = PTR_ERR(root);
2250 				goto out;
2251 			}
2252 		} else {
2253 			set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2254 			fs_info->block_group_root = root;
2255 		}
2256 	}
2257 
2258 	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2259 	root = btrfs_read_tree_root(tree_root, &location);
2260 	if (IS_ERR(root)) {
2261 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2262 			ret = PTR_ERR(root);
2263 			goto out;
2264 		}
2265 	} else {
2266 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2267 		fs_info->dev_root = root;
2268 	}
2269 	/* Initialize fs_info for all devices in any case */
2270 	ret = btrfs_init_devices_late(fs_info);
2271 	if (ret)
2272 		goto out;
2273 
2274 	/*
2275 	 * This tree can share blocks with some other fs tree during relocation
2276 	 * and we need a proper setup by btrfs_get_fs_root
2277 	 */
2278 	root = btrfs_get_fs_root(tree_root->fs_info,
2279 				 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2280 	if (IS_ERR(root)) {
2281 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2282 			ret = PTR_ERR(root);
2283 			goto out;
2284 		}
2285 	} else {
2286 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2287 		fs_info->data_reloc_root = root;
2288 	}
2289 
2290 	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2291 	root = btrfs_read_tree_root(tree_root, &location);
2292 	if (!IS_ERR(root)) {
2293 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2294 		fs_info->quota_root = root;
2295 	}
2296 
2297 	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2298 	root = btrfs_read_tree_root(tree_root, &location);
2299 	if (IS_ERR(root)) {
2300 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2301 			ret = PTR_ERR(root);
2302 			if (ret != -ENOENT)
2303 				goto out;
2304 		}
2305 	} else {
2306 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2307 		fs_info->uuid_root = root;
2308 	}
2309 
2310 	if (btrfs_fs_incompat(fs_info, RAID_STRIPE_TREE)) {
2311 		location.objectid = BTRFS_RAID_STRIPE_TREE_OBJECTID;
2312 		root = btrfs_read_tree_root(tree_root, &location);
2313 		if (IS_ERR(root)) {
2314 			if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2315 				ret = PTR_ERR(root);
2316 				goto out;
2317 			}
2318 		} else {
2319 			set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2320 			fs_info->stripe_root = root;
2321 		}
2322 	}
2323 
2324 	return 0;
2325 out:
2326 	btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2327 		   location.objectid, ret);
2328 	return ret;
2329 }
2330 
2331 /*
2332  * Real super block validation
2333  * NOTE: super csum type and incompat features will not be checked here.
2334  *
2335  * @sb:		super block to check
2336  * @mirror_num:	the super block number to check its bytenr:
2337  * 		0	the primary (1st) sb
2338  * 		1, 2	2nd and 3rd backup copy
2339  * 	       -1	skip bytenr check
2340  */
2341 int btrfs_validate_super(const struct btrfs_fs_info *fs_info,
2342 			 const struct btrfs_super_block *sb, int mirror_num)
2343 {
2344 	u64 nodesize = btrfs_super_nodesize(sb);
2345 	u64 sectorsize = btrfs_super_sectorsize(sb);
2346 	int ret = 0;
2347 	const bool ignore_flags = btrfs_test_opt(fs_info, IGNORESUPERFLAGS);
2348 
2349 	if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2350 		btrfs_err(fs_info, "no valid FS found");
2351 		ret = -EINVAL;
2352 	}
2353 	if ((btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)) {
2354 		if (!ignore_flags) {
2355 			btrfs_err(fs_info,
2356 			"unrecognized or unsupported super flag 0x%llx",
2357 				  btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2358 			ret = -EINVAL;
2359 		} else {
2360 			btrfs_info(fs_info,
2361 			"unrecognized or unsupported super flags: 0x%llx, ignored",
2362 				   btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2363 		}
2364 	}
2365 	if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2366 		btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2367 				btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2368 		ret = -EINVAL;
2369 	}
2370 	if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2371 		btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2372 				btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2373 		ret = -EINVAL;
2374 	}
2375 	if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2376 		btrfs_err(fs_info, "log_root level too big: %d >= %d",
2377 				btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2378 		ret = -EINVAL;
2379 	}
2380 
2381 	/*
2382 	 * Check sectorsize and nodesize first, other check will need it.
2383 	 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2384 	 */
2385 	if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2386 	    sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2387 		btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2388 		ret = -EINVAL;
2389 	}
2390 
2391 	/*
2392 	 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2393 	 *
2394 	 * We can support 16K sectorsize with 64K page size without problem,
2395 	 * but such sectorsize/pagesize combination doesn't make much sense.
2396 	 * 4K will be our future standard, PAGE_SIZE is supported from the very
2397 	 * beginning.
2398 	 */
2399 	if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2400 		btrfs_err(fs_info,
2401 			"sectorsize %llu not yet supported for page size %lu",
2402 			sectorsize, PAGE_SIZE);
2403 		ret = -EINVAL;
2404 	}
2405 
2406 	if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2407 	    nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2408 		btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2409 		ret = -EINVAL;
2410 	}
2411 	if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2412 		btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2413 			  le32_to_cpu(sb->__unused_leafsize), nodesize);
2414 		ret = -EINVAL;
2415 	}
2416 
2417 	/* Root alignment check */
2418 	if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2419 		btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2420 			   btrfs_super_root(sb));
2421 		ret = -EINVAL;
2422 	}
2423 	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2424 		btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2425 			   btrfs_super_chunk_root(sb));
2426 		ret = -EINVAL;
2427 	}
2428 	if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2429 		btrfs_warn(fs_info, "log_root block unaligned: %llu",
2430 			   btrfs_super_log_root(sb));
2431 		ret = -EINVAL;
2432 	}
2433 
2434 	if (!fs_info->fs_devices->temp_fsid &&
2435 	    memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2436 		btrfs_err(fs_info,
2437 		"superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2438 			  sb->fsid, fs_info->fs_devices->fsid);
2439 		ret = -EINVAL;
2440 	}
2441 
2442 	if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2443 		   BTRFS_FSID_SIZE) != 0) {
2444 		btrfs_err(fs_info,
2445 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2446 			  btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2447 		ret = -EINVAL;
2448 	}
2449 
2450 	if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2451 		   BTRFS_FSID_SIZE) != 0) {
2452 		btrfs_err(fs_info,
2453 			"dev_item UUID does not match metadata fsid: %pU != %pU",
2454 			fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2455 		ret = -EINVAL;
2456 	}
2457 
2458 	/*
2459 	 * Artificial requirement for block-group-tree to force newer features
2460 	 * (free-space-tree, no-holes) so the test matrix is smaller.
2461 	 */
2462 	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2463 	    (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2464 	     !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2465 		btrfs_err(fs_info,
2466 		"block-group-tree feature requires free-space-tree and no-holes");
2467 		ret = -EINVAL;
2468 	}
2469 
2470 	/*
2471 	 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2472 	 * done later
2473 	 */
2474 	if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2475 		btrfs_err(fs_info, "bytes_used is too small %llu",
2476 			  btrfs_super_bytes_used(sb));
2477 		ret = -EINVAL;
2478 	}
2479 	if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2480 		btrfs_err(fs_info, "invalid stripesize %u",
2481 			  btrfs_super_stripesize(sb));
2482 		ret = -EINVAL;
2483 	}
2484 	if (btrfs_super_num_devices(sb) > (1UL << 31))
2485 		btrfs_warn(fs_info, "suspicious number of devices: %llu",
2486 			   btrfs_super_num_devices(sb));
2487 	if (btrfs_super_num_devices(sb) == 0) {
2488 		btrfs_err(fs_info, "number of devices is 0");
2489 		ret = -EINVAL;
2490 	}
2491 
2492 	if (mirror_num >= 0 &&
2493 	    btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2494 		btrfs_err(fs_info, "super offset mismatch %llu != %u",
2495 			  btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2496 		ret = -EINVAL;
2497 	}
2498 
2499 	/*
2500 	 * Obvious sys_chunk_array corruptions, it must hold at least one key
2501 	 * and one chunk
2502 	 */
2503 	if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2504 		btrfs_err(fs_info, "system chunk array too big %u > %u",
2505 			  btrfs_super_sys_array_size(sb),
2506 			  BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2507 		ret = -EINVAL;
2508 	}
2509 	if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2510 			+ sizeof(struct btrfs_chunk)) {
2511 		btrfs_err(fs_info, "system chunk array too small %u < %zu",
2512 			  btrfs_super_sys_array_size(sb),
2513 			  sizeof(struct btrfs_disk_key)
2514 			  + sizeof(struct btrfs_chunk));
2515 		ret = -EINVAL;
2516 	}
2517 
2518 	/*
2519 	 * The generation is a global counter, we'll trust it more than the others
2520 	 * but it's still possible that it's the one that's wrong.
2521 	 */
2522 	if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2523 		btrfs_warn(fs_info,
2524 			"suspicious: generation < chunk_root_generation: %llu < %llu",
2525 			btrfs_super_generation(sb),
2526 			btrfs_super_chunk_root_generation(sb));
2527 	if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2528 	    && btrfs_super_cache_generation(sb) != (u64)-1)
2529 		btrfs_warn(fs_info,
2530 			"suspicious: generation < cache_generation: %llu < %llu",
2531 			btrfs_super_generation(sb),
2532 			btrfs_super_cache_generation(sb));
2533 
2534 	return ret;
2535 }
2536 
2537 /*
2538  * Validation of super block at mount time.
2539  * Some checks already done early at mount time, like csum type and incompat
2540  * flags will be skipped.
2541  */
2542 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2543 {
2544 	return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2545 }
2546 
2547 /*
2548  * Validation of super block at write time.
2549  * Some checks like bytenr check will be skipped as their values will be
2550  * overwritten soon.
2551  * Extra checks like csum type and incompat flags will be done here.
2552  */
2553 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2554 				      struct btrfs_super_block *sb)
2555 {
2556 	int ret;
2557 
2558 	ret = btrfs_validate_super(fs_info, sb, -1);
2559 	if (ret < 0)
2560 		goto out;
2561 	if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2562 		ret = -EUCLEAN;
2563 		btrfs_err(fs_info, "invalid csum type, has %u want %u",
2564 			  btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2565 		goto out;
2566 	}
2567 	if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2568 		ret = -EUCLEAN;
2569 		btrfs_err(fs_info,
2570 		"invalid incompat flags, has 0x%llx valid mask 0x%llx",
2571 			  btrfs_super_incompat_flags(sb),
2572 			  (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2573 		goto out;
2574 	}
2575 out:
2576 	if (ret < 0)
2577 		btrfs_err(fs_info,
2578 		"super block corruption detected before writing it to disk");
2579 	return ret;
2580 }
2581 
2582 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2583 {
2584 	struct btrfs_tree_parent_check check = {
2585 		.level = level,
2586 		.transid = gen,
2587 		.owner_root = btrfs_root_id(root)
2588 	};
2589 	int ret = 0;
2590 
2591 	root->node = read_tree_block(root->fs_info, bytenr, &check);
2592 	if (IS_ERR(root->node)) {
2593 		ret = PTR_ERR(root->node);
2594 		root->node = NULL;
2595 		return ret;
2596 	}
2597 	if (!extent_buffer_uptodate(root->node)) {
2598 		free_extent_buffer(root->node);
2599 		root->node = NULL;
2600 		return -EIO;
2601 	}
2602 
2603 	btrfs_set_root_node(&root->root_item, root->node);
2604 	root->commit_root = btrfs_root_node(root);
2605 	btrfs_set_root_refs(&root->root_item, 1);
2606 	return ret;
2607 }
2608 
2609 static int load_important_roots(struct btrfs_fs_info *fs_info)
2610 {
2611 	struct btrfs_super_block *sb = fs_info->super_copy;
2612 	u64 gen, bytenr;
2613 	int level, ret;
2614 
2615 	bytenr = btrfs_super_root(sb);
2616 	gen = btrfs_super_generation(sb);
2617 	level = btrfs_super_root_level(sb);
2618 	ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2619 	if (ret) {
2620 		btrfs_warn(fs_info, "couldn't read tree root");
2621 		return ret;
2622 	}
2623 	return 0;
2624 }
2625 
2626 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2627 {
2628 	int backup_index = find_newest_super_backup(fs_info);
2629 	struct btrfs_super_block *sb = fs_info->super_copy;
2630 	struct btrfs_root *tree_root = fs_info->tree_root;
2631 	bool handle_error = false;
2632 	int ret = 0;
2633 	int i;
2634 
2635 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2636 		if (handle_error) {
2637 			if (!IS_ERR(tree_root->node))
2638 				free_extent_buffer(tree_root->node);
2639 			tree_root->node = NULL;
2640 
2641 			if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2642 				break;
2643 
2644 			free_root_pointers(fs_info, 0);
2645 
2646 			/*
2647 			 * Don't use the log in recovery mode, it won't be
2648 			 * valid
2649 			 */
2650 			btrfs_set_super_log_root(sb, 0);
2651 
2652 			btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2653 			ret = read_backup_root(fs_info, i);
2654 			backup_index = ret;
2655 			if (ret < 0)
2656 				return ret;
2657 		}
2658 
2659 		ret = load_important_roots(fs_info);
2660 		if (ret) {
2661 			handle_error = true;
2662 			continue;
2663 		}
2664 
2665 		/*
2666 		 * No need to hold btrfs_root::objectid_mutex since the fs
2667 		 * hasn't been fully initialised and we are the only user
2668 		 */
2669 		ret = btrfs_init_root_free_objectid(tree_root);
2670 		if (ret < 0) {
2671 			handle_error = true;
2672 			continue;
2673 		}
2674 
2675 		ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2676 
2677 		ret = btrfs_read_roots(fs_info);
2678 		if (ret < 0) {
2679 			handle_error = true;
2680 			continue;
2681 		}
2682 
2683 		/* All successful */
2684 		fs_info->generation = btrfs_header_generation(tree_root->node);
2685 		btrfs_set_last_trans_committed(fs_info, fs_info->generation);
2686 		fs_info->last_reloc_trans = 0;
2687 
2688 		/* Always begin writing backup roots after the one being used */
2689 		if (backup_index < 0) {
2690 			fs_info->backup_root_index = 0;
2691 		} else {
2692 			fs_info->backup_root_index = backup_index + 1;
2693 			fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2694 		}
2695 		break;
2696 	}
2697 
2698 	return ret;
2699 }
2700 
2701 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2702 {
2703 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2704 	INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2705 	INIT_LIST_HEAD(&fs_info->trans_list);
2706 	INIT_LIST_HEAD(&fs_info->dead_roots);
2707 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
2708 	INIT_LIST_HEAD(&fs_info->delalloc_roots);
2709 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
2710 	spin_lock_init(&fs_info->delalloc_root_lock);
2711 	spin_lock_init(&fs_info->trans_lock);
2712 	spin_lock_init(&fs_info->fs_roots_radix_lock);
2713 	spin_lock_init(&fs_info->delayed_iput_lock);
2714 	spin_lock_init(&fs_info->defrag_inodes_lock);
2715 	spin_lock_init(&fs_info->super_lock);
2716 	spin_lock_init(&fs_info->buffer_lock);
2717 	spin_lock_init(&fs_info->unused_bgs_lock);
2718 	spin_lock_init(&fs_info->treelog_bg_lock);
2719 	spin_lock_init(&fs_info->zone_active_bgs_lock);
2720 	spin_lock_init(&fs_info->relocation_bg_lock);
2721 	rwlock_init(&fs_info->tree_mod_log_lock);
2722 	rwlock_init(&fs_info->global_root_lock);
2723 	mutex_init(&fs_info->unused_bg_unpin_mutex);
2724 	mutex_init(&fs_info->reclaim_bgs_lock);
2725 	mutex_init(&fs_info->reloc_mutex);
2726 	mutex_init(&fs_info->delalloc_root_mutex);
2727 	mutex_init(&fs_info->zoned_meta_io_lock);
2728 	mutex_init(&fs_info->zoned_data_reloc_io_lock);
2729 	seqlock_init(&fs_info->profiles_lock);
2730 
2731 	btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2732 	btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2733 	btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2734 	btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2735 	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_prep,
2736 				     BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2737 	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2738 				     BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2739 	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2740 				     BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2741 	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2742 				     BTRFS_LOCKDEP_TRANS_COMPLETED);
2743 
2744 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2745 	INIT_LIST_HEAD(&fs_info->space_info);
2746 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2747 	INIT_LIST_HEAD(&fs_info->unused_bgs);
2748 	INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2749 	INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2750 #ifdef CONFIG_BTRFS_DEBUG
2751 	INIT_LIST_HEAD(&fs_info->allocated_roots);
2752 	INIT_LIST_HEAD(&fs_info->allocated_ebs);
2753 	spin_lock_init(&fs_info->eb_leak_lock);
2754 #endif
2755 	fs_info->mapping_tree = RB_ROOT_CACHED;
2756 	rwlock_init(&fs_info->mapping_tree_lock);
2757 	btrfs_init_block_rsv(&fs_info->global_block_rsv,
2758 			     BTRFS_BLOCK_RSV_GLOBAL);
2759 	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2760 	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2761 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2762 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2763 			     BTRFS_BLOCK_RSV_DELOPS);
2764 	btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2765 			     BTRFS_BLOCK_RSV_DELREFS);
2766 
2767 	atomic_set(&fs_info->async_delalloc_pages, 0);
2768 	atomic_set(&fs_info->defrag_running, 0);
2769 	atomic_set(&fs_info->nr_delayed_iputs, 0);
2770 	atomic64_set(&fs_info->tree_mod_seq, 0);
2771 	fs_info->global_root_tree = RB_ROOT;
2772 	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2773 	fs_info->metadata_ratio = 0;
2774 	fs_info->defrag_inodes = RB_ROOT;
2775 	atomic64_set(&fs_info->free_chunk_space, 0);
2776 	fs_info->tree_mod_log = RB_ROOT;
2777 	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2778 	btrfs_init_ref_verify(fs_info);
2779 
2780 	fs_info->thread_pool_size = min_t(unsigned long,
2781 					  num_online_cpus() + 2, 8);
2782 
2783 	INIT_LIST_HEAD(&fs_info->ordered_roots);
2784 	spin_lock_init(&fs_info->ordered_root_lock);
2785 
2786 	btrfs_init_scrub(fs_info);
2787 	btrfs_init_balance(fs_info);
2788 	btrfs_init_async_reclaim_work(fs_info);
2789 
2790 	rwlock_init(&fs_info->block_group_cache_lock);
2791 	fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2792 
2793 	extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2794 			    IO_TREE_FS_EXCLUDED_EXTENTS);
2795 
2796 	mutex_init(&fs_info->ordered_operations_mutex);
2797 	mutex_init(&fs_info->tree_log_mutex);
2798 	mutex_init(&fs_info->chunk_mutex);
2799 	mutex_init(&fs_info->transaction_kthread_mutex);
2800 	mutex_init(&fs_info->cleaner_mutex);
2801 	mutex_init(&fs_info->ro_block_group_mutex);
2802 	init_rwsem(&fs_info->commit_root_sem);
2803 	init_rwsem(&fs_info->cleanup_work_sem);
2804 	init_rwsem(&fs_info->subvol_sem);
2805 	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2806 
2807 	btrfs_init_dev_replace_locks(fs_info);
2808 	btrfs_init_qgroup(fs_info);
2809 	btrfs_discard_init(fs_info);
2810 
2811 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2812 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2813 
2814 	init_waitqueue_head(&fs_info->transaction_throttle);
2815 	init_waitqueue_head(&fs_info->transaction_wait);
2816 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2817 	init_waitqueue_head(&fs_info->async_submit_wait);
2818 	init_waitqueue_head(&fs_info->delayed_iputs_wait);
2819 
2820 	/* Usable values until the real ones are cached from the superblock */
2821 	fs_info->nodesize = 4096;
2822 	fs_info->sectorsize = 4096;
2823 	fs_info->sectorsize_bits = ilog2(4096);
2824 	fs_info->stripesize = 4096;
2825 
2826 	/* Default compress algorithm when user does -o compress */
2827 	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2828 
2829 	fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2830 
2831 	spin_lock_init(&fs_info->swapfile_pins_lock);
2832 	fs_info->swapfile_pins = RB_ROOT;
2833 
2834 	fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2835 	INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2836 }
2837 
2838 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2839 {
2840 	int ret;
2841 
2842 	fs_info->sb = sb;
2843 	/* Temporary fixed values for block size until we read the superblock. */
2844 	sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2845 	sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2846 
2847 	ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2848 	if (ret)
2849 		return ret;
2850 
2851 	ret = percpu_counter_init(&fs_info->evictable_extent_maps, 0, GFP_KERNEL);
2852 	if (ret)
2853 		return ret;
2854 
2855 	spin_lock_init(&fs_info->extent_map_shrinker_lock);
2856 
2857 	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2858 	if (ret)
2859 		return ret;
2860 
2861 	fs_info->dirty_metadata_batch = PAGE_SIZE *
2862 					(1 + ilog2(nr_cpu_ids));
2863 
2864 	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2865 	if (ret)
2866 		return ret;
2867 
2868 	ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2869 			GFP_KERNEL);
2870 	if (ret)
2871 		return ret;
2872 
2873 	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2874 					GFP_KERNEL);
2875 	if (!fs_info->delayed_root)
2876 		return -ENOMEM;
2877 	btrfs_init_delayed_root(fs_info->delayed_root);
2878 
2879 	if (sb_rdonly(sb))
2880 		set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2881 	if (btrfs_test_opt(fs_info, IGNOREMETACSUMS))
2882 		set_bit(BTRFS_FS_STATE_SKIP_META_CSUMS, &fs_info->fs_state);
2883 
2884 	return btrfs_alloc_stripe_hash_table(fs_info);
2885 }
2886 
2887 static int btrfs_uuid_rescan_kthread(void *data)
2888 {
2889 	struct btrfs_fs_info *fs_info = data;
2890 	int ret;
2891 
2892 	/*
2893 	 * 1st step is to iterate through the existing UUID tree and
2894 	 * to delete all entries that contain outdated data.
2895 	 * 2nd step is to add all missing entries to the UUID tree.
2896 	 */
2897 	ret = btrfs_uuid_tree_iterate(fs_info);
2898 	if (ret < 0) {
2899 		if (ret != -EINTR)
2900 			btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2901 				   ret);
2902 		up(&fs_info->uuid_tree_rescan_sem);
2903 		return ret;
2904 	}
2905 	return btrfs_uuid_scan_kthread(data);
2906 }
2907 
2908 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2909 {
2910 	struct task_struct *task;
2911 
2912 	down(&fs_info->uuid_tree_rescan_sem);
2913 	task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2914 	if (IS_ERR(task)) {
2915 		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
2916 		btrfs_warn(fs_info, "failed to start uuid_rescan task");
2917 		up(&fs_info->uuid_tree_rescan_sem);
2918 		return PTR_ERR(task);
2919 	}
2920 
2921 	return 0;
2922 }
2923 
2924 static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2925 {
2926 	u64 root_objectid = 0;
2927 	struct btrfs_root *gang[8];
2928 	int ret = 0;
2929 
2930 	while (1) {
2931 		unsigned int found;
2932 
2933 		spin_lock(&fs_info->fs_roots_radix_lock);
2934 		found = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2935 					     (void **)gang, root_objectid,
2936 					     ARRAY_SIZE(gang));
2937 		if (!found) {
2938 			spin_unlock(&fs_info->fs_roots_radix_lock);
2939 			break;
2940 		}
2941 		root_objectid = btrfs_root_id(gang[found - 1]) + 1;
2942 
2943 		for (int i = 0; i < found; i++) {
2944 			/* Avoid to grab roots in dead_roots. */
2945 			if (btrfs_root_refs(&gang[i]->root_item) == 0) {
2946 				gang[i] = NULL;
2947 				continue;
2948 			}
2949 			/* Grab all the search result for later use. */
2950 			gang[i] = btrfs_grab_root(gang[i]);
2951 		}
2952 		spin_unlock(&fs_info->fs_roots_radix_lock);
2953 
2954 		for (int i = 0; i < found; i++) {
2955 			if (!gang[i])
2956 				continue;
2957 			root_objectid = btrfs_root_id(gang[i]);
2958 			/*
2959 			 * Continue to release the remaining roots after the first
2960 			 * error without cleanup and preserve the first error
2961 			 * for the return.
2962 			 */
2963 			if (!ret)
2964 				ret = btrfs_orphan_cleanup(gang[i]);
2965 			btrfs_put_root(gang[i]);
2966 		}
2967 		if (ret)
2968 			break;
2969 
2970 		root_objectid++;
2971 	}
2972 	return ret;
2973 }
2974 
2975 /*
2976  * Mounting logic specific to read-write file systems. Shared by open_ctree
2977  * and btrfs_remount when remounting from read-only to read-write.
2978  */
2979 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2980 {
2981 	int ret;
2982 	const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2983 	bool rebuild_free_space_tree = false;
2984 
2985 	if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2986 	    btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2987 		if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
2988 			btrfs_warn(fs_info,
2989 				   "'clear_cache' option is ignored with extent tree v2");
2990 		else
2991 			rebuild_free_space_tree = true;
2992 	} else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2993 		   !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2994 		btrfs_warn(fs_info, "free space tree is invalid");
2995 		rebuild_free_space_tree = true;
2996 	}
2997 
2998 	if (rebuild_free_space_tree) {
2999 		btrfs_info(fs_info, "rebuilding free space tree");
3000 		ret = btrfs_rebuild_free_space_tree(fs_info);
3001 		if (ret) {
3002 			btrfs_warn(fs_info,
3003 				   "failed to rebuild free space tree: %d", ret);
3004 			goto out;
3005 		}
3006 	}
3007 
3008 	if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3009 	    !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
3010 		btrfs_info(fs_info, "disabling free space tree");
3011 		ret = btrfs_delete_free_space_tree(fs_info);
3012 		if (ret) {
3013 			btrfs_warn(fs_info,
3014 				   "failed to disable free space tree: %d", ret);
3015 			goto out;
3016 		}
3017 	}
3018 
3019 	/*
3020 	 * btrfs_find_orphan_roots() is responsible for finding all the dead
3021 	 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3022 	 * them into the fs_info->fs_roots_radix tree. This must be done before
3023 	 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3024 	 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3025 	 * item before the root's tree is deleted - this means that if we unmount
3026 	 * or crash before the deletion completes, on the next mount we will not
3027 	 * delete what remains of the tree because the orphan item does not
3028 	 * exists anymore, which is what tells us we have a pending deletion.
3029 	 */
3030 	ret = btrfs_find_orphan_roots(fs_info);
3031 	if (ret)
3032 		goto out;
3033 
3034 	ret = btrfs_cleanup_fs_roots(fs_info);
3035 	if (ret)
3036 		goto out;
3037 
3038 	down_read(&fs_info->cleanup_work_sem);
3039 	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3040 	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3041 		up_read(&fs_info->cleanup_work_sem);
3042 		goto out;
3043 	}
3044 	up_read(&fs_info->cleanup_work_sem);
3045 
3046 	mutex_lock(&fs_info->cleaner_mutex);
3047 	ret = btrfs_recover_relocation(fs_info);
3048 	mutex_unlock(&fs_info->cleaner_mutex);
3049 	if (ret < 0) {
3050 		btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3051 		goto out;
3052 	}
3053 
3054 	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3055 	    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3056 		btrfs_info(fs_info, "creating free space tree");
3057 		ret = btrfs_create_free_space_tree(fs_info);
3058 		if (ret) {
3059 			btrfs_warn(fs_info,
3060 				"failed to create free space tree: %d", ret);
3061 			goto out;
3062 		}
3063 	}
3064 
3065 	if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3066 		ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3067 		if (ret)
3068 			goto out;
3069 	}
3070 
3071 	ret = btrfs_resume_balance_async(fs_info);
3072 	if (ret)
3073 		goto out;
3074 
3075 	ret = btrfs_resume_dev_replace_async(fs_info);
3076 	if (ret) {
3077 		btrfs_warn(fs_info, "failed to resume dev_replace");
3078 		goto out;
3079 	}
3080 
3081 	btrfs_qgroup_rescan_resume(fs_info);
3082 
3083 	if (!fs_info->uuid_root) {
3084 		btrfs_info(fs_info, "creating UUID tree");
3085 		ret = btrfs_create_uuid_tree(fs_info);
3086 		if (ret) {
3087 			btrfs_warn(fs_info,
3088 				   "failed to create the UUID tree %d", ret);
3089 			goto out;
3090 		}
3091 	}
3092 
3093 out:
3094 	return ret;
3095 }
3096 
3097 /*
3098  * Do various sanity and dependency checks of different features.
3099  *
3100  * @is_rw_mount:	If the mount is read-write.
3101  *
3102  * This is the place for less strict checks (like for subpage or artificial
3103  * feature dependencies).
3104  *
3105  * For strict checks or possible corruption detection, see
3106  * btrfs_validate_super().
3107  *
3108  * This should be called after btrfs_parse_options(), as some mount options
3109  * (space cache related) can modify on-disk format like free space tree and
3110  * screw up certain feature dependencies.
3111  */
3112 int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3113 {
3114 	struct btrfs_super_block *disk_super = fs_info->super_copy;
3115 	u64 incompat = btrfs_super_incompat_flags(disk_super);
3116 	const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3117 	const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3118 
3119 	if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3120 		btrfs_err(fs_info,
3121 		"cannot mount because of unknown incompat features (0x%llx)",
3122 		    incompat);
3123 		return -EINVAL;
3124 	}
3125 
3126 	/* Runtime limitation for mixed block groups. */
3127 	if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3128 	    (fs_info->sectorsize != fs_info->nodesize)) {
3129 		btrfs_err(fs_info,
3130 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3131 			fs_info->nodesize, fs_info->sectorsize);
3132 		return -EINVAL;
3133 	}
3134 
3135 	/* Mixed backref is an always-enabled feature. */
3136 	incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3137 
3138 	/* Set compression related flags just in case. */
3139 	if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3140 		incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3141 	else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3142 		incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3143 
3144 	/*
3145 	 * An ancient flag, which should really be marked deprecated.
3146 	 * Such runtime limitation doesn't really need a incompat flag.
3147 	 */
3148 	if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3149 		incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3150 
3151 	if (compat_ro_unsupp && is_rw_mount) {
3152 		btrfs_err(fs_info,
3153 	"cannot mount read-write because of unknown compat_ro features (0x%llx)",
3154 		       compat_ro);
3155 		return -EINVAL;
3156 	}
3157 
3158 	/*
3159 	 * We have unsupported RO compat features, although RO mounted, we
3160 	 * should not cause any metadata writes, including log replay.
3161 	 * Or we could screw up whatever the new feature requires.
3162 	 */
3163 	if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3164 	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3165 		btrfs_err(fs_info,
3166 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3167 			  compat_ro);
3168 		return -EINVAL;
3169 	}
3170 
3171 	/*
3172 	 * Artificial limitations for block group tree, to force
3173 	 * block-group-tree to rely on no-holes and free-space-tree.
3174 	 */
3175 	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3176 	    (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3177 	     !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3178 		btrfs_err(fs_info,
3179 "block-group-tree feature requires no-holes and free-space-tree features");
3180 		return -EINVAL;
3181 	}
3182 
3183 	/*
3184 	 * Subpage runtime limitation on v1 cache.
3185 	 *
3186 	 * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3187 	 * we're already defaulting to v2 cache, no need to bother v1 as it's
3188 	 * going to be deprecated anyway.
3189 	 */
3190 	if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3191 		btrfs_warn(fs_info,
3192 	"v1 space cache is not supported for page size %lu with sectorsize %u",
3193 			   PAGE_SIZE, fs_info->sectorsize);
3194 		return -EINVAL;
3195 	}
3196 
3197 	/* This can be called by remount, we need to protect the super block. */
3198 	spin_lock(&fs_info->super_lock);
3199 	btrfs_set_super_incompat_flags(disk_super, incompat);
3200 	spin_unlock(&fs_info->super_lock);
3201 
3202 	return 0;
3203 }
3204 
3205 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3206 		      const char *options)
3207 {
3208 	u32 sectorsize;
3209 	u32 nodesize;
3210 	u32 stripesize;
3211 	u64 generation;
3212 	u16 csum_type;
3213 	struct btrfs_super_block *disk_super;
3214 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3215 	struct btrfs_root *tree_root;
3216 	struct btrfs_root *chunk_root;
3217 	int ret;
3218 	int level;
3219 
3220 	ret = init_mount_fs_info(fs_info, sb);
3221 	if (ret)
3222 		goto fail;
3223 
3224 	/* These need to be init'ed before we start creating inodes and such. */
3225 	tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3226 				     GFP_KERNEL);
3227 	fs_info->tree_root = tree_root;
3228 	chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3229 				      GFP_KERNEL);
3230 	fs_info->chunk_root = chunk_root;
3231 	if (!tree_root || !chunk_root) {
3232 		ret = -ENOMEM;
3233 		goto fail;
3234 	}
3235 
3236 	ret = btrfs_init_btree_inode(sb);
3237 	if (ret)
3238 		goto fail;
3239 
3240 	invalidate_bdev(fs_devices->latest_dev->bdev);
3241 
3242 	/*
3243 	 * Read super block and check the signature bytes only
3244 	 */
3245 	disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3246 	if (IS_ERR(disk_super)) {
3247 		ret = PTR_ERR(disk_super);
3248 		goto fail_alloc;
3249 	}
3250 
3251 	btrfs_info(fs_info, "first mount of filesystem %pU", disk_super->fsid);
3252 	/*
3253 	 * Verify the type first, if that or the checksum value are
3254 	 * corrupted, we'll find out
3255 	 */
3256 	csum_type = btrfs_super_csum_type(disk_super);
3257 	if (!btrfs_supported_super_csum(csum_type)) {
3258 		btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3259 			  csum_type);
3260 		ret = -EINVAL;
3261 		btrfs_release_disk_super(disk_super);
3262 		goto fail_alloc;
3263 	}
3264 
3265 	fs_info->csum_size = btrfs_super_csum_size(disk_super);
3266 
3267 	ret = btrfs_init_csum_hash(fs_info, csum_type);
3268 	if (ret) {
3269 		btrfs_release_disk_super(disk_super);
3270 		goto fail_alloc;
3271 	}
3272 
3273 	/*
3274 	 * We want to check superblock checksum, the type is stored inside.
3275 	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3276 	 */
3277 	if (btrfs_check_super_csum(fs_info, disk_super)) {
3278 		btrfs_err(fs_info, "superblock checksum mismatch");
3279 		ret = -EINVAL;
3280 		btrfs_release_disk_super(disk_super);
3281 		goto fail_alloc;
3282 	}
3283 
3284 	/*
3285 	 * super_copy is zeroed at allocation time and we never touch the
3286 	 * following bytes up to INFO_SIZE, the checksum is calculated from
3287 	 * the whole block of INFO_SIZE
3288 	 */
3289 	memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3290 	btrfs_release_disk_super(disk_super);
3291 
3292 	disk_super = fs_info->super_copy;
3293 
3294 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
3295 	       sizeof(*fs_info->super_for_commit));
3296 
3297 	ret = btrfs_validate_mount_super(fs_info);
3298 	if (ret) {
3299 		btrfs_err(fs_info, "superblock contains fatal errors");
3300 		ret = -EINVAL;
3301 		goto fail_alloc;
3302 	}
3303 
3304 	if (!btrfs_super_root(disk_super)) {
3305 		btrfs_err(fs_info, "invalid superblock tree root bytenr");
3306 		ret = -EINVAL;
3307 		goto fail_alloc;
3308 	}
3309 
3310 	/* check FS state, whether FS is broken. */
3311 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3312 		WRITE_ONCE(fs_info->fs_error, -EUCLEAN);
3313 
3314 	/* Set up fs_info before parsing mount options */
3315 	nodesize = btrfs_super_nodesize(disk_super);
3316 	sectorsize = btrfs_super_sectorsize(disk_super);
3317 	stripesize = sectorsize;
3318 	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3319 	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3320 
3321 	fs_info->nodesize = nodesize;
3322 	fs_info->sectorsize = sectorsize;
3323 	fs_info->sectorsize_bits = ilog2(sectorsize);
3324 	fs_info->sectors_per_page = (PAGE_SIZE >> fs_info->sectorsize_bits);
3325 	fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3326 	fs_info->stripesize = stripesize;
3327 
3328 	/*
3329 	 * Handle the space caching options appropriately now that we have the
3330 	 * super block loaded and validated.
3331 	 */
3332 	btrfs_set_free_space_cache_settings(fs_info);
3333 
3334 	if (!btrfs_check_options(fs_info, &fs_info->mount_opt, sb->s_flags)) {
3335 		ret = -EINVAL;
3336 		goto fail_alloc;
3337 	}
3338 
3339 	ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3340 	if (ret < 0)
3341 		goto fail_alloc;
3342 
3343 	/*
3344 	 * At this point our mount options are validated, if we set ->max_inline
3345 	 * to something non-standard make sure we truncate it to sectorsize.
3346 	 */
3347 	fs_info->max_inline = min_t(u64, fs_info->max_inline, fs_info->sectorsize);
3348 
3349 	if (sectorsize < PAGE_SIZE)
3350 		btrfs_warn(fs_info,
3351 		"read-write for sector size %u with page size %lu is experimental",
3352 			   sectorsize, PAGE_SIZE);
3353 
3354 	ret = btrfs_init_workqueues(fs_info);
3355 	if (ret)
3356 		goto fail_sb_buffer;
3357 
3358 	sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3359 	sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3360 
3361 	/* Update the values for the current filesystem. */
3362 	sb->s_blocksize = sectorsize;
3363 	sb->s_blocksize_bits = blksize_bits(sectorsize);
3364 	memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3365 
3366 	mutex_lock(&fs_info->chunk_mutex);
3367 	ret = btrfs_read_sys_array(fs_info);
3368 	mutex_unlock(&fs_info->chunk_mutex);
3369 	if (ret) {
3370 		btrfs_err(fs_info, "failed to read the system array: %d", ret);
3371 		goto fail_sb_buffer;
3372 	}
3373 
3374 	generation = btrfs_super_chunk_root_generation(disk_super);
3375 	level = btrfs_super_chunk_root_level(disk_super);
3376 	ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3377 			      generation, level);
3378 	if (ret) {
3379 		btrfs_err(fs_info, "failed to read chunk root");
3380 		goto fail_tree_roots;
3381 	}
3382 
3383 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3384 			   offsetof(struct btrfs_header, chunk_tree_uuid),
3385 			   BTRFS_UUID_SIZE);
3386 
3387 	ret = btrfs_read_chunk_tree(fs_info);
3388 	if (ret) {
3389 		btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3390 		goto fail_tree_roots;
3391 	}
3392 
3393 	/*
3394 	 * At this point we know all the devices that make this filesystem,
3395 	 * including the seed devices but we don't know yet if the replace
3396 	 * target is required. So free devices that are not part of this
3397 	 * filesystem but skip the replace target device which is checked
3398 	 * below in btrfs_init_dev_replace().
3399 	 */
3400 	btrfs_free_extra_devids(fs_devices);
3401 	if (!fs_devices->latest_dev->bdev) {
3402 		btrfs_err(fs_info, "failed to read devices");
3403 		ret = -EIO;
3404 		goto fail_tree_roots;
3405 	}
3406 
3407 	ret = init_tree_roots(fs_info);
3408 	if (ret)
3409 		goto fail_tree_roots;
3410 
3411 	/*
3412 	 * Get zone type information of zoned block devices. This will also
3413 	 * handle emulation of a zoned filesystem if a regular device has the
3414 	 * zoned incompat feature flag set.
3415 	 */
3416 	ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3417 	if (ret) {
3418 		btrfs_err(fs_info,
3419 			  "zoned: failed to read device zone info: %d", ret);
3420 		goto fail_block_groups;
3421 	}
3422 
3423 	/*
3424 	 * If we have a uuid root and we're not being told to rescan we need to
3425 	 * check the generation here so we can set the
3426 	 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3427 	 * transaction during a balance or the log replay without updating the
3428 	 * uuid generation, and then if we crash we would rescan the uuid tree,
3429 	 * even though it was perfectly fine.
3430 	 */
3431 	if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3432 	    fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3433 		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3434 
3435 	ret = btrfs_verify_dev_extents(fs_info);
3436 	if (ret) {
3437 		btrfs_err(fs_info,
3438 			  "failed to verify dev extents against chunks: %d",
3439 			  ret);
3440 		goto fail_block_groups;
3441 	}
3442 	ret = btrfs_recover_balance(fs_info);
3443 	if (ret) {
3444 		btrfs_err(fs_info, "failed to recover balance: %d", ret);
3445 		goto fail_block_groups;
3446 	}
3447 
3448 	ret = btrfs_init_dev_stats(fs_info);
3449 	if (ret) {
3450 		btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3451 		goto fail_block_groups;
3452 	}
3453 
3454 	ret = btrfs_init_dev_replace(fs_info);
3455 	if (ret) {
3456 		btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3457 		goto fail_block_groups;
3458 	}
3459 
3460 	ret = btrfs_check_zoned_mode(fs_info);
3461 	if (ret) {
3462 		btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3463 			  ret);
3464 		goto fail_block_groups;
3465 	}
3466 
3467 	ret = btrfs_sysfs_add_fsid(fs_devices);
3468 	if (ret) {
3469 		btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3470 				ret);
3471 		goto fail_block_groups;
3472 	}
3473 
3474 	ret = btrfs_sysfs_add_mounted(fs_info);
3475 	if (ret) {
3476 		btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3477 		goto fail_fsdev_sysfs;
3478 	}
3479 
3480 	ret = btrfs_init_space_info(fs_info);
3481 	if (ret) {
3482 		btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3483 		goto fail_sysfs;
3484 	}
3485 
3486 	ret = btrfs_read_block_groups(fs_info);
3487 	if (ret) {
3488 		btrfs_err(fs_info, "failed to read block groups: %d", ret);
3489 		goto fail_sysfs;
3490 	}
3491 
3492 	btrfs_free_zone_cache(fs_info);
3493 
3494 	btrfs_check_active_zone_reservation(fs_info);
3495 
3496 	if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3497 	    !btrfs_check_rw_degradable(fs_info, NULL)) {
3498 		btrfs_warn(fs_info,
3499 		"writable mount is not allowed due to too many missing devices");
3500 		ret = -EINVAL;
3501 		goto fail_sysfs;
3502 	}
3503 
3504 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3505 					       "btrfs-cleaner");
3506 	if (IS_ERR(fs_info->cleaner_kthread)) {
3507 		ret = PTR_ERR(fs_info->cleaner_kthread);
3508 		goto fail_sysfs;
3509 	}
3510 
3511 	fs_info->transaction_kthread = kthread_run(transaction_kthread,
3512 						   tree_root,
3513 						   "btrfs-transaction");
3514 	if (IS_ERR(fs_info->transaction_kthread)) {
3515 		ret = PTR_ERR(fs_info->transaction_kthread);
3516 		goto fail_cleaner;
3517 	}
3518 
3519 	ret = btrfs_read_qgroup_config(fs_info);
3520 	if (ret)
3521 		goto fail_trans_kthread;
3522 
3523 	if (btrfs_build_ref_tree(fs_info))
3524 		btrfs_err(fs_info, "couldn't build ref tree");
3525 
3526 	/* do not make disk changes in broken FS or nologreplay is given */
3527 	if (btrfs_super_log_root(disk_super) != 0 &&
3528 	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3529 		btrfs_info(fs_info, "start tree-log replay");
3530 		ret = btrfs_replay_log(fs_info, fs_devices);
3531 		if (ret)
3532 			goto fail_qgroup;
3533 	}
3534 
3535 	fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3536 	if (IS_ERR(fs_info->fs_root)) {
3537 		ret = PTR_ERR(fs_info->fs_root);
3538 		btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3539 		fs_info->fs_root = NULL;
3540 		goto fail_qgroup;
3541 	}
3542 
3543 	if (sb_rdonly(sb))
3544 		return 0;
3545 
3546 	ret = btrfs_start_pre_rw_mount(fs_info);
3547 	if (ret) {
3548 		close_ctree(fs_info);
3549 		return ret;
3550 	}
3551 	btrfs_discard_resume(fs_info);
3552 
3553 	if (fs_info->uuid_root &&
3554 	    (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3555 	     fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3556 		btrfs_info(fs_info, "checking UUID tree");
3557 		ret = btrfs_check_uuid_tree(fs_info);
3558 		if (ret) {
3559 			btrfs_warn(fs_info,
3560 				"failed to check the UUID tree: %d", ret);
3561 			close_ctree(fs_info);
3562 			return ret;
3563 		}
3564 	}
3565 
3566 	set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3567 
3568 	/* Kick the cleaner thread so it'll start deleting snapshots. */
3569 	if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3570 		wake_up_process(fs_info->cleaner_kthread);
3571 
3572 	return 0;
3573 
3574 fail_qgroup:
3575 	btrfs_free_qgroup_config(fs_info);
3576 fail_trans_kthread:
3577 	kthread_stop(fs_info->transaction_kthread);
3578 	btrfs_cleanup_transaction(fs_info);
3579 	btrfs_free_fs_roots(fs_info);
3580 fail_cleaner:
3581 	kthread_stop(fs_info->cleaner_kthread);
3582 
3583 	/*
3584 	 * make sure we're done with the btree inode before we stop our
3585 	 * kthreads
3586 	 */
3587 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3588 
3589 fail_sysfs:
3590 	btrfs_sysfs_remove_mounted(fs_info);
3591 
3592 fail_fsdev_sysfs:
3593 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3594 
3595 fail_block_groups:
3596 	btrfs_put_block_group_cache(fs_info);
3597 
3598 fail_tree_roots:
3599 	if (fs_info->data_reloc_root)
3600 		btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3601 	free_root_pointers(fs_info, true);
3602 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3603 
3604 fail_sb_buffer:
3605 	btrfs_stop_all_workers(fs_info);
3606 	btrfs_free_block_groups(fs_info);
3607 fail_alloc:
3608 	btrfs_mapping_tree_free(fs_info);
3609 
3610 	iput(fs_info->btree_inode);
3611 fail:
3612 	btrfs_close_devices(fs_info->fs_devices);
3613 	ASSERT(ret < 0);
3614 	return ret;
3615 }
3616 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3617 
3618 static void btrfs_end_super_write(struct bio *bio)
3619 {
3620 	struct btrfs_device *device = bio->bi_private;
3621 	struct folio_iter fi;
3622 
3623 	bio_for_each_folio_all(fi, bio) {
3624 		if (bio->bi_status) {
3625 			btrfs_warn_rl_in_rcu(device->fs_info,
3626 				"lost super block write due to IO error on %s (%d)",
3627 				btrfs_dev_name(device),
3628 				blk_status_to_errno(bio->bi_status));
3629 			btrfs_dev_stat_inc_and_print(device,
3630 						     BTRFS_DEV_STAT_WRITE_ERRS);
3631 			/* Ensure failure if the primary sb fails. */
3632 			if (bio->bi_opf & REQ_FUA)
3633 				atomic_add(BTRFS_SUPER_PRIMARY_WRITE_ERROR,
3634 					   &device->sb_write_errors);
3635 			else
3636 				atomic_inc(&device->sb_write_errors);
3637 		}
3638 		folio_unlock(fi.folio);
3639 		folio_put(fi.folio);
3640 	}
3641 
3642 	bio_put(bio);
3643 }
3644 
3645 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3646 						   int copy_num, bool drop_cache)
3647 {
3648 	struct btrfs_super_block *super;
3649 	struct page *page;
3650 	u64 bytenr, bytenr_orig;
3651 	struct address_space *mapping = bdev->bd_mapping;
3652 	int ret;
3653 
3654 	bytenr_orig = btrfs_sb_offset(copy_num);
3655 	ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3656 	if (ret == -ENOENT)
3657 		return ERR_PTR(-EINVAL);
3658 	else if (ret)
3659 		return ERR_PTR(ret);
3660 
3661 	if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3662 		return ERR_PTR(-EINVAL);
3663 
3664 	if (drop_cache) {
3665 		/* This should only be called with the primary sb. */
3666 		ASSERT(copy_num == 0);
3667 
3668 		/*
3669 		 * Drop the page of the primary superblock, so later read will
3670 		 * always read from the device.
3671 		 */
3672 		invalidate_inode_pages2_range(mapping,
3673 				bytenr >> PAGE_SHIFT,
3674 				(bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3675 	}
3676 
3677 	page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3678 	if (IS_ERR(page))
3679 		return ERR_CAST(page);
3680 
3681 	super = page_address(page);
3682 	if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3683 		btrfs_release_disk_super(super);
3684 		return ERR_PTR(-ENODATA);
3685 	}
3686 
3687 	if (btrfs_super_bytenr(super) != bytenr_orig) {
3688 		btrfs_release_disk_super(super);
3689 		return ERR_PTR(-EINVAL);
3690 	}
3691 
3692 	return super;
3693 }
3694 
3695 
3696 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3697 {
3698 	struct btrfs_super_block *super, *latest = NULL;
3699 	int i;
3700 	u64 transid = 0;
3701 
3702 	/* we would like to check all the supers, but that would make
3703 	 * a btrfs mount succeed after a mkfs from a different FS.
3704 	 * So, we need to add a special mount option to scan for
3705 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3706 	 */
3707 	for (i = 0; i < 1; i++) {
3708 		super = btrfs_read_dev_one_super(bdev, i, false);
3709 		if (IS_ERR(super))
3710 			continue;
3711 
3712 		if (!latest || btrfs_super_generation(super) > transid) {
3713 			if (latest)
3714 				btrfs_release_disk_super(super);
3715 
3716 			latest = super;
3717 			transid = btrfs_super_generation(super);
3718 		}
3719 	}
3720 
3721 	return super;
3722 }
3723 
3724 /*
3725  * Write superblock @sb to the @device. Do not wait for completion, all the
3726  * folios we use for writing are locked.
3727  *
3728  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3729  * the expected device size at commit time. Note that max_mirrors must be
3730  * same for write and wait phases.
3731  *
3732  * Return number of errors when folio is not found or submission fails.
3733  */
3734 static int write_dev_supers(struct btrfs_device *device,
3735 			    struct btrfs_super_block *sb, int max_mirrors)
3736 {
3737 	struct btrfs_fs_info *fs_info = device->fs_info;
3738 	struct address_space *mapping = device->bdev->bd_mapping;
3739 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3740 	int i;
3741 	int ret;
3742 	u64 bytenr, bytenr_orig;
3743 
3744 	atomic_set(&device->sb_write_errors, 0);
3745 
3746 	if (max_mirrors == 0)
3747 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3748 
3749 	shash->tfm = fs_info->csum_shash;
3750 
3751 	for (i = 0; i < max_mirrors; i++) {
3752 		struct folio *folio;
3753 		struct bio *bio;
3754 		struct btrfs_super_block *disk_super;
3755 		size_t offset;
3756 
3757 		bytenr_orig = btrfs_sb_offset(i);
3758 		ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3759 		if (ret == -ENOENT) {
3760 			continue;
3761 		} else if (ret < 0) {
3762 			btrfs_err(device->fs_info,
3763 				"couldn't get super block location for mirror %d",
3764 				i);
3765 			atomic_inc(&device->sb_write_errors);
3766 			continue;
3767 		}
3768 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3769 		    device->commit_total_bytes)
3770 			break;
3771 
3772 		btrfs_set_super_bytenr(sb, bytenr_orig);
3773 
3774 		crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3775 				    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3776 				    sb->csum);
3777 
3778 		folio = __filemap_get_folio(mapping, bytenr >> PAGE_SHIFT,
3779 					    FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
3780 					    GFP_NOFS);
3781 		if (IS_ERR(folio)) {
3782 			btrfs_err(device->fs_info,
3783 			    "couldn't get super block page for bytenr %llu",
3784 			    bytenr);
3785 			atomic_inc(&device->sb_write_errors);
3786 			continue;
3787 		}
3788 		ASSERT(folio_order(folio) == 0);
3789 
3790 		offset = offset_in_folio(folio, bytenr);
3791 		disk_super = folio_address(folio) + offset;
3792 		memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3793 
3794 		/*
3795 		 * Directly use bios here instead of relying on the page cache
3796 		 * to do I/O, so we don't lose the ability to do integrity
3797 		 * checking.
3798 		 */
3799 		bio = bio_alloc(device->bdev, 1,
3800 				REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3801 				GFP_NOFS);
3802 		bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3803 		bio->bi_private = device;
3804 		bio->bi_end_io = btrfs_end_super_write;
3805 		bio_add_folio_nofail(bio, folio, BTRFS_SUPER_INFO_SIZE, offset);
3806 
3807 		/*
3808 		 * We FUA only the first super block.  The others we allow to
3809 		 * go down lazy and there's a short window where the on-disk
3810 		 * copies might still contain the older version.
3811 		 */
3812 		if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3813 			bio->bi_opf |= REQ_FUA;
3814 		submit_bio(bio);
3815 
3816 		if (btrfs_advance_sb_log(device, i))
3817 			atomic_inc(&device->sb_write_errors);
3818 	}
3819 	return atomic_read(&device->sb_write_errors) < i ? 0 : -1;
3820 }
3821 
3822 /*
3823  * Wait for write completion of superblocks done by write_dev_supers,
3824  * @max_mirrors same for write and wait phases.
3825  *
3826  * Return -1 if primary super block write failed or when there were no super block
3827  * copies written. Otherwise 0.
3828  */
3829 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3830 {
3831 	int i;
3832 	int errors = 0;
3833 	bool primary_failed = false;
3834 	int ret;
3835 	u64 bytenr;
3836 
3837 	if (max_mirrors == 0)
3838 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3839 
3840 	for (i = 0; i < max_mirrors; i++) {
3841 		struct folio *folio;
3842 
3843 		ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3844 		if (ret == -ENOENT) {
3845 			break;
3846 		} else if (ret < 0) {
3847 			errors++;
3848 			if (i == 0)
3849 				primary_failed = true;
3850 			continue;
3851 		}
3852 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3853 		    device->commit_total_bytes)
3854 			break;
3855 
3856 		folio = filemap_get_folio(device->bdev->bd_mapping,
3857 					  bytenr >> PAGE_SHIFT);
3858 		/* If the folio has been removed, then we know it completed. */
3859 		if (IS_ERR(folio))
3860 			continue;
3861 		ASSERT(folio_order(folio) == 0);
3862 
3863 		/* Folio will be unlocked once the write completes. */
3864 		folio_wait_locked(folio);
3865 		folio_put(folio);
3866 	}
3867 
3868 	errors += atomic_read(&device->sb_write_errors);
3869 	if (errors >= BTRFS_SUPER_PRIMARY_WRITE_ERROR)
3870 		primary_failed = true;
3871 	if (primary_failed) {
3872 		btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3873 			  device->devid);
3874 		return -1;
3875 	}
3876 
3877 	return errors < i ? 0 : -1;
3878 }
3879 
3880 /*
3881  * endio for the write_dev_flush, this will wake anyone waiting
3882  * for the barrier when it is done
3883  */
3884 static void btrfs_end_empty_barrier(struct bio *bio)
3885 {
3886 	bio_uninit(bio);
3887 	complete(bio->bi_private);
3888 }
3889 
3890 /*
3891  * Submit a flush request to the device if it supports it. Error handling is
3892  * done in the waiting counterpart.
3893  */
3894 static void write_dev_flush(struct btrfs_device *device)
3895 {
3896 	struct bio *bio = &device->flush_bio;
3897 
3898 	device->last_flush_error = BLK_STS_OK;
3899 
3900 	bio_init(bio, device->bdev, NULL, 0,
3901 		 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3902 	bio->bi_end_io = btrfs_end_empty_barrier;
3903 	init_completion(&device->flush_wait);
3904 	bio->bi_private = &device->flush_wait;
3905 	submit_bio(bio);
3906 	set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3907 }
3908 
3909 /*
3910  * If the flush bio has been submitted by write_dev_flush, wait for it.
3911  * Return true for any error, and false otherwise.
3912  */
3913 static bool wait_dev_flush(struct btrfs_device *device)
3914 {
3915 	struct bio *bio = &device->flush_bio;
3916 
3917 	if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3918 		return false;
3919 
3920 	wait_for_completion_io(&device->flush_wait);
3921 
3922 	if (bio->bi_status) {
3923 		device->last_flush_error = bio->bi_status;
3924 		btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS);
3925 		return true;
3926 	}
3927 
3928 	return false;
3929 }
3930 
3931 /*
3932  * send an empty flush down to each device in parallel,
3933  * then wait for them
3934  */
3935 static int barrier_all_devices(struct btrfs_fs_info *info)
3936 {
3937 	struct list_head *head;
3938 	struct btrfs_device *dev;
3939 	int errors_wait = 0;
3940 
3941 	lockdep_assert_held(&info->fs_devices->device_list_mutex);
3942 	/* send down all the barriers */
3943 	head = &info->fs_devices->devices;
3944 	list_for_each_entry(dev, head, dev_list) {
3945 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3946 			continue;
3947 		if (!dev->bdev)
3948 			continue;
3949 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3950 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3951 			continue;
3952 
3953 		write_dev_flush(dev);
3954 	}
3955 
3956 	/* wait for all the barriers */
3957 	list_for_each_entry(dev, head, dev_list) {
3958 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3959 			continue;
3960 		if (!dev->bdev) {
3961 			errors_wait++;
3962 			continue;
3963 		}
3964 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3965 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3966 			continue;
3967 
3968 		if (wait_dev_flush(dev))
3969 			errors_wait++;
3970 	}
3971 
3972 	/*
3973 	 * Checks last_flush_error of disks in order to determine the device
3974 	 * state.
3975 	 */
3976 	if (errors_wait && !btrfs_check_rw_degradable(info, NULL))
3977 		return -EIO;
3978 
3979 	return 0;
3980 }
3981 
3982 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3983 {
3984 	int raid_type;
3985 	int min_tolerated = INT_MAX;
3986 
3987 	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3988 	    (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3989 		min_tolerated = min_t(int, min_tolerated,
3990 				    btrfs_raid_array[BTRFS_RAID_SINGLE].
3991 				    tolerated_failures);
3992 
3993 	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3994 		if (raid_type == BTRFS_RAID_SINGLE)
3995 			continue;
3996 		if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3997 			continue;
3998 		min_tolerated = min_t(int, min_tolerated,
3999 				    btrfs_raid_array[raid_type].
4000 				    tolerated_failures);
4001 	}
4002 
4003 	if (min_tolerated == INT_MAX) {
4004 		pr_warn("BTRFS: unknown raid flag: %llu", flags);
4005 		min_tolerated = 0;
4006 	}
4007 
4008 	return min_tolerated;
4009 }
4010 
4011 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4012 {
4013 	struct list_head *head;
4014 	struct btrfs_device *dev;
4015 	struct btrfs_super_block *sb;
4016 	struct btrfs_dev_item *dev_item;
4017 	int ret;
4018 	int do_barriers;
4019 	int max_errors;
4020 	int total_errors = 0;
4021 	u64 flags;
4022 
4023 	do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4024 
4025 	/*
4026 	 * max_mirrors == 0 indicates we're from commit_transaction,
4027 	 * not from fsync where the tree roots in fs_info have not
4028 	 * been consistent on disk.
4029 	 */
4030 	if (max_mirrors == 0)
4031 		backup_super_roots(fs_info);
4032 
4033 	sb = fs_info->super_for_commit;
4034 	dev_item = &sb->dev_item;
4035 
4036 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
4037 	head = &fs_info->fs_devices->devices;
4038 	max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4039 
4040 	if (do_barriers) {
4041 		ret = barrier_all_devices(fs_info);
4042 		if (ret) {
4043 			mutex_unlock(
4044 				&fs_info->fs_devices->device_list_mutex);
4045 			btrfs_handle_fs_error(fs_info, ret,
4046 					      "errors while submitting device barriers.");
4047 			return ret;
4048 		}
4049 	}
4050 
4051 	list_for_each_entry(dev, head, dev_list) {
4052 		if (!dev->bdev) {
4053 			total_errors++;
4054 			continue;
4055 		}
4056 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4057 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4058 			continue;
4059 
4060 		btrfs_set_stack_device_generation(dev_item, 0);
4061 		btrfs_set_stack_device_type(dev_item, dev->type);
4062 		btrfs_set_stack_device_id(dev_item, dev->devid);
4063 		btrfs_set_stack_device_total_bytes(dev_item,
4064 						   dev->commit_total_bytes);
4065 		btrfs_set_stack_device_bytes_used(dev_item,
4066 						  dev->commit_bytes_used);
4067 		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4068 		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4069 		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4070 		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4071 		memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4072 		       BTRFS_FSID_SIZE);
4073 
4074 		flags = btrfs_super_flags(sb);
4075 		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4076 
4077 		ret = btrfs_validate_write_super(fs_info, sb);
4078 		if (ret < 0) {
4079 			mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4080 			btrfs_handle_fs_error(fs_info, -EUCLEAN,
4081 				"unexpected superblock corruption detected");
4082 			return -EUCLEAN;
4083 		}
4084 
4085 		ret = write_dev_supers(dev, sb, max_mirrors);
4086 		if (ret)
4087 			total_errors++;
4088 	}
4089 	if (total_errors > max_errors) {
4090 		btrfs_err(fs_info, "%d errors while writing supers",
4091 			  total_errors);
4092 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4093 
4094 		/* FUA is masked off if unsupported and can't be the reason */
4095 		btrfs_handle_fs_error(fs_info, -EIO,
4096 				      "%d errors while writing supers",
4097 				      total_errors);
4098 		return -EIO;
4099 	}
4100 
4101 	total_errors = 0;
4102 	list_for_each_entry(dev, head, dev_list) {
4103 		if (!dev->bdev)
4104 			continue;
4105 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4106 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4107 			continue;
4108 
4109 		ret = wait_dev_supers(dev, max_mirrors);
4110 		if (ret)
4111 			total_errors++;
4112 	}
4113 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4114 	if (total_errors > max_errors) {
4115 		btrfs_handle_fs_error(fs_info, -EIO,
4116 				      "%d errors while writing supers",
4117 				      total_errors);
4118 		return -EIO;
4119 	}
4120 	return 0;
4121 }
4122 
4123 /* Drop a fs root from the radix tree and free it. */
4124 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4125 				  struct btrfs_root *root)
4126 {
4127 	bool drop_ref = false;
4128 
4129 	spin_lock(&fs_info->fs_roots_radix_lock);
4130 	radix_tree_delete(&fs_info->fs_roots_radix,
4131 			  (unsigned long)btrfs_root_id(root));
4132 	if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4133 		drop_ref = true;
4134 	spin_unlock(&fs_info->fs_roots_radix_lock);
4135 
4136 	if (BTRFS_FS_ERROR(fs_info)) {
4137 		ASSERT(root->log_root == NULL);
4138 		if (root->reloc_root) {
4139 			btrfs_put_root(root->reloc_root);
4140 			root->reloc_root = NULL;
4141 		}
4142 	}
4143 
4144 	if (drop_ref)
4145 		btrfs_put_root(root);
4146 }
4147 
4148 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4149 {
4150 	mutex_lock(&fs_info->cleaner_mutex);
4151 	btrfs_run_delayed_iputs(fs_info);
4152 	mutex_unlock(&fs_info->cleaner_mutex);
4153 	wake_up_process(fs_info->cleaner_kthread);
4154 
4155 	/* wait until ongoing cleanup work done */
4156 	down_write(&fs_info->cleanup_work_sem);
4157 	up_write(&fs_info->cleanup_work_sem);
4158 
4159 	return btrfs_commit_current_transaction(fs_info->tree_root);
4160 }
4161 
4162 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4163 {
4164 	struct btrfs_transaction *trans;
4165 	struct btrfs_transaction *tmp;
4166 	bool found = false;
4167 
4168 	/*
4169 	 * This function is only called at the very end of close_ctree(),
4170 	 * thus no other running transaction, no need to take trans_lock.
4171 	 */
4172 	ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4173 	list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4174 		struct extent_state *cached = NULL;
4175 		u64 dirty_bytes = 0;
4176 		u64 cur = 0;
4177 		u64 found_start;
4178 		u64 found_end;
4179 
4180 		found = true;
4181 		while (find_first_extent_bit(&trans->dirty_pages, cur,
4182 			&found_start, &found_end, EXTENT_DIRTY, &cached)) {
4183 			dirty_bytes += found_end + 1 - found_start;
4184 			cur = found_end + 1;
4185 		}
4186 		btrfs_warn(fs_info,
4187 	"transaction %llu (with %llu dirty metadata bytes) is not committed",
4188 			   trans->transid, dirty_bytes);
4189 		btrfs_cleanup_one_transaction(trans, fs_info);
4190 
4191 		if (trans == fs_info->running_transaction)
4192 			fs_info->running_transaction = NULL;
4193 		list_del_init(&trans->list);
4194 
4195 		btrfs_put_transaction(trans);
4196 		trace_btrfs_transaction_commit(fs_info);
4197 	}
4198 	ASSERT(!found);
4199 }
4200 
4201 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4202 {
4203 	int ret;
4204 
4205 	set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4206 
4207 	/*
4208 	 * If we had UNFINISHED_DROPS we could still be processing them, so
4209 	 * clear that bit and wake up relocation so it can stop.
4210 	 * We must do this before stopping the block group reclaim task, because
4211 	 * at btrfs_relocate_block_group() we wait for this bit, and after the
4212 	 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4213 	 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4214 	 * return 1.
4215 	 */
4216 	btrfs_wake_unfinished_drop(fs_info);
4217 
4218 	/*
4219 	 * We may have the reclaim task running and relocating a data block group,
4220 	 * in which case it may create delayed iputs. So stop it before we park
4221 	 * the cleaner kthread otherwise we can get new delayed iputs after
4222 	 * parking the cleaner, and that can make the async reclaim task to hang
4223 	 * if it's waiting for delayed iputs to complete, since the cleaner is
4224 	 * parked and can not run delayed iputs - this will make us hang when
4225 	 * trying to stop the async reclaim task.
4226 	 */
4227 	cancel_work_sync(&fs_info->reclaim_bgs_work);
4228 	/*
4229 	 * We don't want the cleaner to start new transactions, add more delayed
4230 	 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4231 	 * because that frees the task_struct, and the transaction kthread might
4232 	 * still try to wake up the cleaner.
4233 	 */
4234 	kthread_park(fs_info->cleaner_kthread);
4235 
4236 	/* wait for the qgroup rescan worker to stop */
4237 	btrfs_qgroup_wait_for_completion(fs_info, false);
4238 
4239 	/* wait for the uuid_scan task to finish */
4240 	down(&fs_info->uuid_tree_rescan_sem);
4241 	/* avoid complains from lockdep et al., set sem back to initial state */
4242 	up(&fs_info->uuid_tree_rescan_sem);
4243 
4244 	/* pause restriper - we want to resume on mount */
4245 	btrfs_pause_balance(fs_info);
4246 
4247 	btrfs_dev_replace_suspend_for_unmount(fs_info);
4248 
4249 	btrfs_scrub_cancel(fs_info);
4250 
4251 	/* wait for any defraggers to finish */
4252 	wait_event(fs_info->transaction_wait,
4253 		   (atomic_read(&fs_info->defrag_running) == 0));
4254 
4255 	/* clear out the rbtree of defraggable inodes */
4256 	btrfs_cleanup_defrag_inodes(fs_info);
4257 
4258 	/*
4259 	 * After we parked the cleaner kthread, ordered extents may have
4260 	 * completed and created new delayed iputs. If one of the async reclaim
4261 	 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4262 	 * can hang forever trying to stop it, because if a delayed iput is
4263 	 * added after it ran btrfs_run_delayed_iputs() and before it called
4264 	 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4265 	 * no one else to run iputs.
4266 	 *
4267 	 * So wait for all ongoing ordered extents to complete and then run
4268 	 * delayed iputs. This works because once we reach this point no one
4269 	 * can either create new ordered extents nor create delayed iputs
4270 	 * through some other means.
4271 	 *
4272 	 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4273 	 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4274 	 * but the delayed iput for the respective inode is made only when doing
4275 	 * the final btrfs_put_ordered_extent() (which must happen at
4276 	 * btrfs_finish_ordered_io() when we are unmounting).
4277 	 */
4278 	btrfs_flush_workqueue(fs_info->endio_write_workers);
4279 	/* Ordered extents for free space inodes. */
4280 	btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4281 	btrfs_run_delayed_iputs(fs_info);
4282 
4283 	cancel_work_sync(&fs_info->async_reclaim_work);
4284 	cancel_work_sync(&fs_info->async_data_reclaim_work);
4285 	cancel_work_sync(&fs_info->preempt_reclaim_work);
4286 
4287 	/* Cancel or finish ongoing discard work */
4288 	btrfs_discard_cleanup(fs_info);
4289 
4290 	if (!sb_rdonly(fs_info->sb)) {
4291 		/*
4292 		 * The cleaner kthread is stopped, so do one final pass over
4293 		 * unused block groups.
4294 		 */
4295 		btrfs_delete_unused_bgs(fs_info);
4296 
4297 		/*
4298 		 * There might be existing delayed inode workers still running
4299 		 * and holding an empty delayed inode item. We must wait for
4300 		 * them to complete first because they can create a transaction.
4301 		 * This happens when someone calls btrfs_balance_delayed_items()
4302 		 * and then a transaction commit runs the same delayed nodes
4303 		 * before any delayed worker has done something with the nodes.
4304 		 * We must wait for any worker here and not at transaction
4305 		 * commit time since that could cause a deadlock.
4306 		 * This is a very rare case.
4307 		 */
4308 		btrfs_flush_workqueue(fs_info->delayed_workers);
4309 
4310 		ret = btrfs_commit_super(fs_info);
4311 		if (ret)
4312 			btrfs_err(fs_info, "commit super ret %d", ret);
4313 	}
4314 
4315 	if (BTRFS_FS_ERROR(fs_info))
4316 		btrfs_error_commit_super(fs_info);
4317 
4318 	kthread_stop(fs_info->transaction_kthread);
4319 	kthread_stop(fs_info->cleaner_kthread);
4320 
4321 	ASSERT(list_empty(&fs_info->delayed_iputs));
4322 	set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4323 
4324 	if (btrfs_check_quota_leak(fs_info)) {
4325 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4326 		btrfs_err(fs_info, "qgroup reserved space leaked");
4327 	}
4328 
4329 	btrfs_free_qgroup_config(fs_info);
4330 	ASSERT(list_empty(&fs_info->delalloc_roots));
4331 
4332 	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4333 		btrfs_info(fs_info, "at unmount delalloc count %lld",
4334 		       percpu_counter_sum(&fs_info->delalloc_bytes));
4335 	}
4336 
4337 	if (percpu_counter_sum(&fs_info->ordered_bytes))
4338 		btrfs_info(fs_info, "at unmount dio bytes count %lld",
4339 			   percpu_counter_sum(&fs_info->ordered_bytes));
4340 
4341 	btrfs_sysfs_remove_mounted(fs_info);
4342 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4343 
4344 	btrfs_put_block_group_cache(fs_info);
4345 
4346 	/*
4347 	 * we must make sure there is not any read request to
4348 	 * submit after we stopping all workers.
4349 	 */
4350 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4351 	btrfs_stop_all_workers(fs_info);
4352 
4353 	/* We shouldn't have any transaction open at this point */
4354 	warn_about_uncommitted_trans(fs_info);
4355 
4356 	clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4357 	free_root_pointers(fs_info, true);
4358 	btrfs_free_fs_roots(fs_info);
4359 
4360 	/*
4361 	 * We must free the block groups after dropping the fs_roots as we could
4362 	 * have had an IO error and have left over tree log blocks that aren't
4363 	 * cleaned up until the fs roots are freed.  This makes the block group
4364 	 * accounting appear to be wrong because there's pending reserved bytes,
4365 	 * so make sure we do the block group cleanup afterwards.
4366 	 */
4367 	btrfs_free_block_groups(fs_info);
4368 
4369 	iput(fs_info->btree_inode);
4370 
4371 	btrfs_mapping_tree_free(fs_info);
4372 	btrfs_close_devices(fs_info->fs_devices);
4373 }
4374 
4375 void btrfs_mark_buffer_dirty(struct btrfs_trans_handle *trans,
4376 			     struct extent_buffer *buf)
4377 {
4378 	struct btrfs_fs_info *fs_info = buf->fs_info;
4379 	u64 transid = btrfs_header_generation(buf);
4380 
4381 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4382 	/*
4383 	 * This is a fast path so only do this check if we have sanity tests
4384 	 * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4385 	 * outside of the sanity tests.
4386 	 */
4387 	if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4388 		return;
4389 #endif
4390 	/* This is an active transaction (its state < TRANS_STATE_UNBLOCKED). */
4391 	ASSERT(trans->transid == fs_info->generation);
4392 	btrfs_assert_tree_write_locked(buf);
4393 	if (unlikely(transid != fs_info->generation)) {
4394 		btrfs_abort_transaction(trans, -EUCLEAN);
4395 		btrfs_crit(fs_info,
4396 "dirty buffer transid mismatch, logical %llu found transid %llu running transid %llu",
4397 			   buf->start, transid, fs_info->generation);
4398 	}
4399 	set_extent_buffer_dirty(buf);
4400 }
4401 
4402 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4403 					int flush_delayed)
4404 {
4405 	/*
4406 	 * looks as though older kernels can get into trouble with
4407 	 * this code, they end up stuck in balance_dirty_pages forever
4408 	 */
4409 	int ret;
4410 
4411 	if (current->flags & PF_MEMALLOC)
4412 		return;
4413 
4414 	if (flush_delayed)
4415 		btrfs_balance_delayed_items(fs_info);
4416 
4417 	ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4418 				     BTRFS_DIRTY_METADATA_THRESH,
4419 				     fs_info->dirty_metadata_batch);
4420 	if (ret > 0) {
4421 		balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4422 	}
4423 }
4424 
4425 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4426 {
4427 	__btrfs_btree_balance_dirty(fs_info, 1);
4428 }
4429 
4430 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4431 {
4432 	__btrfs_btree_balance_dirty(fs_info, 0);
4433 }
4434 
4435 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4436 {
4437 	/* cleanup FS via transaction */
4438 	btrfs_cleanup_transaction(fs_info);
4439 
4440 	mutex_lock(&fs_info->cleaner_mutex);
4441 	btrfs_run_delayed_iputs(fs_info);
4442 	mutex_unlock(&fs_info->cleaner_mutex);
4443 
4444 	down_write(&fs_info->cleanup_work_sem);
4445 	up_write(&fs_info->cleanup_work_sem);
4446 }
4447 
4448 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4449 {
4450 	struct btrfs_root *gang[8];
4451 	u64 root_objectid = 0;
4452 	int ret;
4453 
4454 	spin_lock(&fs_info->fs_roots_radix_lock);
4455 	while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4456 					     (void **)gang, root_objectid,
4457 					     ARRAY_SIZE(gang))) != 0) {
4458 		int i;
4459 
4460 		for (i = 0; i < ret; i++)
4461 			gang[i] = btrfs_grab_root(gang[i]);
4462 		spin_unlock(&fs_info->fs_roots_radix_lock);
4463 
4464 		for (i = 0; i < ret; i++) {
4465 			if (!gang[i])
4466 				continue;
4467 			root_objectid = btrfs_root_id(gang[i]);
4468 			btrfs_free_log(NULL, gang[i]);
4469 			btrfs_put_root(gang[i]);
4470 		}
4471 		root_objectid++;
4472 		spin_lock(&fs_info->fs_roots_radix_lock);
4473 	}
4474 	spin_unlock(&fs_info->fs_roots_radix_lock);
4475 	btrfs_free_log_root_tree(NULL, fs_info);
4476 }
4477 
4478 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4479 {
4480 	struct btrfs_ordered_extent *ordered;
4481 
4482 	spin_lock(&root->ordered_extent_lock);
4483 	/*
4484 	 * This will just short circuit the ordered completion stuff which will
4485 	 * make sure the ordered extent gets properly cleaned up.
4486 	 */
4487 	list_for_each_entry(ordered, &root->ordered_extents,
4488 			    root_extent_list)
4489 		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4490 	spin_unlock(&root->ordered_extent_lock);
4491 }
4492 
4493 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4494 {
4495 	struct btrfs_root *root;
4496 	LIST_HEAD(splice);
4497 
4498 	spin_lock(&fs_info->ordered_root_lock);
4499 	list_splice_init(&fs_info->ordered_roots, &splice);
4500 	while (!list_empty(&splice)) {
4501 		root = list_first_entry(&splice, struct btrfs_root,
4502 					ordered_root);
4503 		list_move_tail(&root->ordered_root,
4504 			       &fs_info->ordered_roots);
4505 
4506 		spin_unlock(&fs_info->ordered_root_lock);
4507 		btrfs_destroy_ordered_extents(root);
4508 
4509 		cond_resched();
4510 		spin_lock(&fs_info->ordered_root_lock);
4511 	}
4512 	spin_unlock(&fs_info->ordered_root_lock);
4513 
4514 	/*
4515 	 * We need this here because if we've been flipped read-only we won't
4516 	 * get sync() from the umount, so we need to make sure any ordered
4517 	 * extents that haven't had their dirty pages IO start writeout yet
4518 	 * actually get run and error out properly.
4519 	 */
4520 	btrfs_wait_ordered_roots(fs_info, U64_MAX, NULL);
4521 }
4522 
4523 static void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4524 				       struct btrfs_fs_info *fs_info)
4525 {
4526 	struct rb_node *node;
4527 	struct btrfs_delayed_ref_root *delayed_refs = &trans->delayed_refs;
4528 	struct btrfs_delayed_ref_node *ref;
4529 
4530 	spin_lock(&delayed_refs->lock);
4531 	while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4532 		struct btrfs_delayed_ref_head *head;
4533 		struct rb_node *n;
4534 		bool pin_bytes = false;
4535 
4536 		head = rb_entry(node, struct btrfs_delayed_ref_head,
4537 				href_node);
4538 		if (btrfs_delayed_ref_lock(delayed_refs, head))
4539 			continue;
4540 
4541 		spin_lock(&head->lock);
4542 		while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4543 			ref = rb_entry(n, struct btrfs_delayed_ref_node,
4544 				       ref_node);
4545 			rb_erase_cached(&ref->ref_node, &head->ref_tree);
4546 			RB_CLEAR_NODE(&ref->ref_node);
4547 			if (!list_empty(&ref->add_list))
4548 				list_del(&ref->add_list);
4549 			atomic_dec(&delayed_refs->num_entries);
4550 			btrfs_put_delayed_ref(ref);
4551 			btrfs_delayed_refs_rsv_release(fs_info, 1, 0);
4552 		}
4553 		if (head->must_insert_reserved)
4554 			pin_bytes = true;
4555 		btrfs_free_delayed_extent_op(head->extent_op);
4556 		btrfs_delete_ref_head(delayed_refs, head);
4557 		spin_unlock(&head->lock);
4558 		spin_unlock(&delayed_refs->lock);
4559 		mutex_unlock(&head->mutex);
4560 
4561 		if (pin_bytes) {
4562 			struct btrfs_block_group *cache;
4563 
4564 			cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4565 			BUG_ON(!cache);
4566 
4567 			spin_lock(&cache->space_info->lock);
4568 			spin_lock(&cache->lock);
4569 			cache->pinned += head->num_bytes;
4570 			btrfs_space_info_update_bytes_pinned(fs_info,
4571 				cache->space_info, head->num_bytes);
4572 			cache->reserved -= head->num_bytes;
4573 			cache->space_info->bytes_reserved -= head->num_bytes;
4574 			spin_unlock(&cache->lock);
4575 			spin_unlock(&cache->space_info->lock);
4576 
4577 			btrfs_put_block_group(cache);
4578 
4579 			btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4580 				head->bytenr + head->num_bytes - 1);
4581 		}
4582 		btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4583 		btrfs_put_delayed_ref_head(head);
4584 		cond_resched();
4585 		spin_lock(&delayed_refs->lock);
4586 	}
4587 	btrfs_qgroup_destroy_extent_records(trans);
4588 
4589 	spin_unlock(&delayed_refs->lock);
4590 }
4591 
4592 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4593 {
4594 	struct btrfs_inode *btrfs_inode;
4595 	LIST_HEAD(splice);
4596 
4597 	spin_lock(&root->delalloc_lock);
4598 	list_splice_init(&root->delalloc_inodes, &splice);
4599 
4600 	while (!list_empty(&splice)) {
4601 		struct inode *inode = NULL;
4602 		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4603 					       delalloc_inodes);
4604 		btrfs_del_delalloc_inode(btrfs_inode);
4605 		spin_unlock(&root->delalloc_lock);
4606 
4607 		/*
4608 		 * Make sure we get a live inode and that it'll not disappear
4609 		 * meanwhile.
4610 		 */
4611 		inode = igrab(&btrfs_inode->vfs_inode);
4612 		if (inode) {
4613 			unsigned int nofs_flag;
4614 
4615 			nofs_flag = memalloc_nofs_save();
4616 			invalidate_inode_pages2(inode->i_mapping);
4617 			memalloc_nofs_restore(nofs_flag);
4618 			iput(inode);
4619 		}
4620 		spin_lock(&root->delalloc_lock);
4621 	}
4622 	spin_unlock(&root->delalloc_lock);
4623 }
4624 
4625 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4626 {
4627 	struct btrfs_root *root;
4628 	LIST_HEAD(splice);
4629 
4630 	spin_lock(&fs_info->delalloc_root_lock);
4631 	list_splice_init(&fs_info->delalloc_roots, &splice);
4632 	while (!list_empty(&splice)) {
4633 		root = list_first_entry(&splice, struct btrfs_root,
4634 					 delalloc_root);
4635 		root = btrfs_grab_root(root);
4636 		BUG_ON(!root);
4637 		spin_unlock(&fs_info->delalloc_root_lock);
4638 
4639 		btrfs_destroy_delalloc_inodes(root);
4640 		btrfs_put_root(root);
4641 
4642 		spin_lock(&fs_info->delalloc_root_lock);
4643 	}
4644 	spin_unlock(&fs_info->delalloc_root_lock);
4645 }
4646 
4647 static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4648 					 struct extent_io_tree *dirty_pages,
4649 					 int mark)
4650 {
4651 	struct extent_buffer *eb;
4652 	u64 start = 0;
4653 	u64 end;
4654 
4655 	while (find_first_extent_bit(dirty_pages, start, &start, &end,
4656 				     mark, NULL)) {
4657 		clear_extent_bits(dirty_pages, start, end, mark);
4658 		while (start <= end) {
4659 			eb = find_extent_buffer(fs_info, start);
4660 			start += fs_info->nodesize;
4661 			if (!eb)
4662 				continue;
4663 
4664 			btrfs_tree_lock(eb);
4665 			wait_on_extent_buffer_writeback(eb);
4666 			btrfs_clear_buffer_dirty(NULL, eb);
4667 			btrfs_tree_unlock(eb);
4668 
4669 			free_extent_buffer_stale(eb);
4670 		}
4671 	}
4672 }
4673 
4674 static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4675 					struct extent_io_tree *unpin)
4676 {
4677 	u64 start;
4678 	u64 end;
4679 
4680 	while (1) {
4681 		struct extent_state *cached_state = NULL;
4682 
4683 		/*
4684 		 * The btrfs_finish_extent_commit() may get the same range as
4685 		 * ours between find_first_extent_bit and clear_extent_dirty.
4686 		 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4687 		 * the same extent range.
4688 		 */
4689 		mutex_lock(&fs_info->unused_bg_unpin_mutex);
4690 		if (!find_first_extent_bit(unpin, 0, &start, &end,
4691 					   EXTENT_DIRTY, &cached_state)) {
4692 			mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4693 			break;
4694 		}
4695 
4696 		clear_extent_dirty(unpin, start, end, &cached_state);
4697 		free_extent_state(cached_state);
4698 		btrfs_error_unpin_extent_range(fs_info, start, end);
4699 		mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4700 		cond_resched();
4701 	}
4702 }
4703 
4704 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4705 {
4706 	struct inode *inode;
4707 
4708 	inode = cache->io_ctl.inode;
4709 	if (inode) {
4710 		unsigned int nofs_flag;
4711 
4712 		nofs_flag = memalloc_nofs_save();
4713 		invalidate_inode_pages2(inode->i_mapping);
4714 		memalloc_nofs_restore(nofs_flag);
4715 
4716 		BTRFS_I(inode)->generation = 0;
4717 		cache->io_ctl.inode = NULL;
4718 		iput(inode);
4719 	}
4720 	ASSERT(cache->io_ctl.pages == NULL);
4721 	btrfs_put_block_group(cache);
4722 }
4723 
4724 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4725 			     struct btrfs_fs_info *fs_info)
4726 {
4727 	struct btrfs_block_group *cache;
4728 
4729 	spin_lock(&cur_trans->dirty_bgs_lock);
4730 	while (!list_empty(&cur_trans->dirty_bgs)) {
4731 		cache = list_first_entry(&cur_trans->dirty_bgs,
4732 					 struct btrfs_block_group,
4733 					 dirty_list);
4734 
4735 		if (!list_empty(&cache->io_list)) {
4736 			spin_unlock(&cur_trans->dirty_bgs_lock);
4737 			list_del_init(&cache->io_list);
4738 			btrfs_cleanup_bg_io(cache);
4739 			spin_lock(&cur_trans->dirty_bgs_lock);
4740 		}
4741 
4742 		list_del_init(&cache->dirty_list);
4743 		spin_lock(&cache->lock);
4744 		cache->disk_cache_state = BTRFS_DC_ERROR;
4745 		spin_unlock(&cache->lock);
4746 
4747 		spin_unlock(&cur_trans->dirty_bgs_lock);
4748 		btrfs_put_block_group(cache);
4749 		btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
4750 		spin_lock(&cur_trans->dirty_bgs_lock);
4751 	}
4752 	spin_unlock(&cur_trans->dirty_bgs_lock);
4753 
4754 	/*
4755 	 * Refer to the definition of io_bgs member for details why it's safe
4756 	 * to use it without any locking
4757 	 */
4758 	while (!list_empty(&cur_trans->io_bgs)) {
4759 		cache = list_first_entry(&cur_trans->io_bgs,
4760 					 struct btrfs_block_group,
4761 					 io_list);
4762 
4763 		list_del_init(&cache->io_list);
4764 		spin_lock(&cache->lock);
4765 		cache->disk_cache_state = BTRFS_DC_ERROR;
4766 		spin_unlock(&cache->lock);
4767 		btrfs_cleanup_bg_io(cache);
4768 	}
4769 }
4770 
4771 static void btrfs_free_all_qgroup_pertrans(struct btrfs_fs_info *fs_info)
4772 {
4773 	struct btrfs_root *gang[8];
4774 	int i;
4775 	int ret;
4776 
4777 	spin_lock(&fs_info->fs_roots_radix_lock);
4778 	while (1) {
4779 		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
4780 						 (void **)gang, 0,
4781 						 ARRAY_SIZE(gang),
4782 						 BTRFS_ROOT_TRANS_TAG);
4783 		if (ret == 0)
4784 			break;
4785 		for (i = 0; i < ret; i++) {
4786 			struct btrfs_root *root = gang[i];
4787 
4788 			btrfs_qgroup_free_meta_all_pertrans(root);
4789 			radix_tree_tag_clear(&fs_info->fs_roots_radix,
4790 					(unsigned long)btrfs_root_id(root),
4791 					BTRFS_ROOT_TRANS_TAG);
4792 		}
4793 	}
4794 	spin_unlock(&fs_info->fs_roots_radix_lock);
4795 }
4796 
4797 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4798 				   struct btrfs_fs_info *fs_info)
4799 {
4800 	struct btrfs_device *dev, *tmp;
4801 
4802 	btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4803 	ASSERT(list_empty(&cur_trans->dirty_bgs));
4804 	ASSERT(list_empty(&cur_trans->io_bgs));
4805 
4806 	list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4807 				 post_commit_list) {
4808 		list_del_init(&dev->post_commit_list);
4809 	}
4810 
4811 	btrfs_destroy_delayed_refs(cur_trans, fs_info);
4812 
4813 	cur_trans->state = TRANS_STATE_COMMIT_START;
4814 	wake_up(&fs_info->transaction_blocked_wait);
4815 
4816 	cur_trans->state = TRANS_STATE_UNBLOCKED;
4817 	wake_up(&fs_info->transaction_wait);
4818 
4819 	btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4820 				     EXTENT_DIRTY);
4821 	btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4822 
4823 	cur_trans->state =TRANS_STATE_COMPLETED;
4824 	wake_up(&cur_trans->commit_wait);
4825 }
4826 
4827 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4828 {
4829 	struct btrfs_transaction *t;
4830 
4831 	mutex_lock(&fs_info->transaction_kthread_mutex);
4832 
4833 	spin_lock(&fs_info->trans_lock);
4834 	while (!list_empty(&fs_info->trans_list)) {
4835 		t = list_first_entry(&fs_info->trans_list,
4836 				     struct btrfs_transaction, list);
4837 		if (t->state >= TRANS_STATE_COMMIT_PREP) {
4838 			refcount_inc(&t->use_count);
4839 			spin_unlock(&fs_info->trans_lock);
4840 			btrfs_wait_for_commit(fs_info, t->transid);
4841 			btrfs_put_transaction(t);
4842 			spin_lock(&fs_info->trans_lock);
4843 			continue;
4844 		}
4845 		if (t == fs_info->running_transaction) {
4846 			t->state = TRANS_STATE_COMMIT_DOING;
4847 			spin_unlock(&fs_info->trans_lock);
4848 			/*
4849 			 * We wait for 0 num_writers since we don't hold a trans
4850 			 * handle open currently for this transaction.
4851 			 */
4852 			wait_event(t->writer_wait,
4853 				   atomic_read(&t->num_writers) == 0);
4854 		} else {
4855 			spin_unlock(&fs_info->trans_lock);
4856 		}
4857 		btrfs_cleanup_one_transaction(t, fs_info);
4858 
4859 		spin_lock(&fs_info->trans_lock);
4860 		if (t == fs_info->running_transaction)
4861 			fs_info->running_transaction = NULL;
4862 		list_del_init(&t->list);
4863 		spin_unlock(&fs_info->trans_lock);
4864 
4865 		btrfs_put_transaction(t);
4866 		trace_btrfs_transaction_commit(fs_info);
4867 		spin_lock(&fs_info->trans_lock);
4868 	}
4869 	spin_unlock(&fs_info->trans_lock);
4870 	btrfs_destroy_all_ordered_extents(fs_info);
4871 	btrfs_destroy_delayed_inodes(fs_info);
4872 	btrfs_assert_delayed_root_empty(fs_info);
4873 	btrfs_destroy_all_delalloc_inodes(fs_info);
4874 	btrfs_drop_all_logs(fs_info);
4875 	btrfs_free_all_qgroup_pertrans(fs_info);
4876 	mutex_unlock(&fs_info->transaction_kthread_mutex);
4877 
4878 	return 0;
4879 }
4880 
4881 int btrfs_init_root_free_objectid(struct btrfs_root *root)
4882 {
4883 	struct btrfs_path *path;
4884 	int ret;
4885 	struct extent_buffer *l;
4886 	struct btrfs_key search_key;
4887 	struct btrfs_key found_key;
4888 	int slot;
4889 
4890 	path = btrfs_alloc_path();
4891 	if (!path)
4892 		return -ENOMEM;
4893 
4894 	search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4895 	search_key.type = -1;
4896 	search_key.offset = (u64)-1;
4897 	ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4898 	if (ret < 0)
4899 		goto error;
4900 	if (ret == 0) {
4901 		/*
4902 		 * Key with offset -1 found, there would have to exist a root
4903 		 * with such id, but this is out of valid range.
4904 		 */
4905 		ret = -EUCLEAN;
4906 		goto error;
4907 	}
4908 	if (path->slots[0] > 0) {
4909 		slot = path->slots[0] - 1;
4910 		l = path->nodes[0];
4911 		btrfs_item_key_to_cpu(l, &found_key, slot);
4912 		root->free_objectid = max_t(u64, found_key.objectid + 1,
4913 					    BTRFS_FIRST_FREE_OBJECTID);
4914 	} else {
4915 		root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4916 	}
4917 	ret = 0;
4918 error:
4919 	btrfs_free_path(path);
4920 	return ret;
4921 }
4922 
4923 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4924 {
4925 	int ret;
4926 	mutex_lock(&root->objectid_mutex);
4927 
4928 	if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4929 		btrfs_warn(root->fs_info,
4930 			   "the objectid of root %llu reaches its highest value",
4931 			   btrfs_root_id(root));
4932 		ret = -ENOSPC;
4933 		goto out;
4934 	}
4935 
4936 	*objectid = root->free_objectid++;
4937 	ret = 0;
4938 out:
4939 	mutex_unlock(&root->objectid_mutex);
4940 	return ret;
4941 }
4942