xref: /linux/fs/btrfs/disk-io.c (revision 497e6b37b0099dc415578488287fd84fb74433eb)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5 
6 #include <linux/fs.h>
7 #include <linux/blkdev.h>
8 #include <linux/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/workqueue.h>
11 #include <linux/kthread.h>
12 #include <linux/slab.h>
13 #include <linux/migrate.h>
14 #include <linux/ratelimit.h>
15 #include <linux/uuid.h>
16 #include <linux/semaphore.h>
17 #include <linux/error-injection.h>
18 #include <linux/crc32c.h>
19 #include <linux/sched/mm.h>
20 #include <asm/unaligned.h>
21 #include <crypto/hash.h>
22 #include "ctree.h"
23 #include "disk-io.h"
24 #include "transaction.h"
25 #include "btrfs_inode.h"
26 #include "bio.h"
27 #include "print-tree.h"
28 #include "locking.h"
29 #include "tree-log.h"
30 #include "free-space-cache.h"
31 #include "free-space-tree.h"
32 #include "check-integrity.h"
33 #include "rcu-string.h"
34 #include "dev-replace.h"
35 #include "raid56.h"
36 #include "sysfs.h"
37 #include "qgroup.h"
38 #include "compression.h"
39 #include "tree-checker.h"
40 #include "ref-verify.h"
41 #include "block-group.h"
42 #include "discard.h"
43 #include "space-info.h"
44 #include "zoned.h"
45 #include "subpage.h"
46 #include "fs.h"
47 #include "accessors.h"
48 #include "extent-tree.h"
49 #include "root-tree.h"
50 #include "defrag.h"
51 #include "uuid-tree.h"
52 #include "relocation.h"
53 #include "scrub.h"
54 #include "super.h"
55 
56 #define BTRFS_SUPER_FLAG_SUPP	(BTRFS_HEADER_FLAG_WRITTEN |\
57 				 BTRFS_HEADER_FLAG_RELOC |\
58 				 BTRFS_SUPER_FLAG_ERROR |\
59 				 BTRFS_SUPER_FLAG_SEEDING |\
60 				 BTRFS_SUPER_FLAG_METADUMP |\
61 				 BTRFS_SUPER_FLAG_METADUMP_V2)
62 
63 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
64 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
65 				      struct btrfs_fs_info *fs_info);
66 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
67 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
68 					struct extent_io_tree *dirty_pages,
69 					int mark);
70 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
71 				       struct extent_io_tree *pinned_extents);
72 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
73 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
74 
75 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
76 {
77 	if (fs_info->csum_shash)
78 		crypto_free_shash(fs_info->csum_shash);
79 }
80 
81 /*
82  * async submit bios are used to offload expensive checksumming
83  * onto the worker threads.  They checksum file and metadata bios
84  * just before they are sent down the IO stack.
85  */
86 struct async_submit_bio {
87 	struct btrfs_inode *inode;
88 	struct bio *bio;
89 	enum btrfs_wq_submit_cmd submit_cmd;
90 	int mirror_num;
91 
92 	/* Optional parameter for used by direct io */
93 	u64 dio_file_offset;
94 	struct btrfs_work work;
95 	blk_status_t status;
96 };
97 
98 /*
99  * Compute the csum of a btree block and store the result to provided buffer.
100  */
101 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
102 {
103 	struct btrfs_fs_info *fs_info = buf->fs_info;
104 	const int num_pages = num_extent_pages(buf);
105 	const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
106 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
107 	char *kaddr;
108 	int i;
109 
110 	shash->tfm = fs_info->csum_shash;
111 	crypto_shash_init(shash);
112 	kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
113 	crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
114 			    first_page_part - BTRFS_CSUM_SIZE);
115 
116 	for (i = 1; i < num_pages; i++) {
117 		kaddr = page_address(buf->pages[i]);
118 		crypto_shash_update(shash, kaddr, PAGE_SIZE);
119 	}
120 	memset(result, 0, BTRFS_CSUM_SIZE);
121 	crypto_shash_final(shash, result);
122 }
123 
124 /*
125  * we can't consider a given block up to date unless the transid of the
126  * block matches the transid in the parent node's pointer.  This is how we
127  * detect blocks that either didn't get written at all or got written
128  * in the wrong place.
129  */
130 static int verify_parent_transid(struct extent_io_tree *io_tree,
131 				 struct extent_buffer *eb, u64 parent_transid,
132 				 int atomic)
133 {
134 	struct extent_state *cached_state = NULL;
135 	int ret;
136 
137 	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
138 		return 0;
139 
140 	if (atomic)
141 		return -EAGAIN;
142 
143 	lock_extent(io_tree, eb->start, eb->start + eb->len - 1, &cached_state);
144 	if (extent_buffer_uptodate(eb) &&
145 	    btrfs_header_generation(eb) == parent_transid) {
146 		ret = 0;
147 		goto out;
148 	}
149 	btrfs_err_rl(eb->fs_info,
150 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
151 			eb->start, eb->read_mirror,
152 			parent_transid, btrfs_header_generation(eb));
153 	ret = 1;
154 	clear_extent_buffer_uptodate(eb);
155 out:
156 	unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
157 		      &cached_state);
158 	return ret;
159 }
160 
161 static bool btrfs_supported_super_csum(u16 csum_type)
162 {
163 	switch (csum_type) {
164 	case BTRFS_CSUM_TYPE_CRC32:
165 	case BTRFS_CSUM_TYPE_XXHASH:
166 	case BTRFS_CSUM_TYPE_SHA256:
167 	case BTRFS_CSUM_TYPE_BLAKE2:
168 		return true;
169 	default:
170 		return false;
171 	}
172 }
173 
174 /*
175  * Return 0 if the superblock checksum type matches the checksum value of that
176  * algorithm. Pass the raw disk superblock data.
177  */
178 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
179 			   const struct btrfs_super_block *disk_sb)
180 {
181 	char result[BTRFS_CSUM_SIZE];
182 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
183 
184 	shash->tfm = fs_info->csum_shash;
185 
186 	/*
187 	 * The super_block structure does not span the whole
188 	 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
189 	 * filled with zeros and is included in the checksum.
190 	 */
191 	crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
192 			    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
193 
194 	if (memcmp(disk_sb->csum, result, fs_info->csum_size))
195 		return 1;
196 
197 	return 0;
198 }
199 
200 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
201 			   struct btrfs_key *first_key, u64 parent_transid)
202 {
203 	struct btrfs_fs_info *fs_info = eb->fs_info;
204 	int found_level;
205 	struct btrfs_key found_key;
206 	int ret;
207 
208 	found_level = btrfs_header_level(eb);
209 	if (found_level != level) {
210 		WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
211 		     KERN_ERR "BTRFS: tree level check failed\n");
212 		btrfs_err(fs_info,
213 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
214 			  eb->start, level, found_level);
215 		return -EIO;
216 	}
217 
218 	if (!first_key)
219 		return 0;
220 
221 	/*
222 	 * For live tree block (new tree blocks in current transaction),
223 	 * we need proper lock context to avoid race, which is impossible here.
224 	 * So we only checks tree blocks which is read from disk, whose
225 	 * generation <= fs_info->last_trans_committed.
226 	 */
227 	if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
228 		return 0;
229 
230 	/* We have @first_key, so this @eb must have at least one item */
231 	if (btrfs_header_nritems(eb) == 0) {
232 		btrfs_err(fs_info,
233 		"invalid tree nritems, bytenr=%llu nritems=0 expect >0",
234 			  eb->start);
235 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
236 		return -EUCLEAN;
237 	}
238 
239 	if (found_level)
240 		btrfs_node_key_to_cpu(eb, &found_key, 0);
241 	else
242 		btrfs_item_key_to_cpu(eb, &found_key, 0);
243 	ret = btrfs_comp_cpu_keys(first_key, &found_key);
244 
245 	if (ret) {
246 		WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
247 		     KERN_ERR "BTRFS: tree first key check failed\n");
248 		btrfs_err(fs_info,
249 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
250 			  eb->start, parent_transid, first_key->objectid,
251 			  first_key->type, first_key->offset,
252 			  found_key.objectid, found_key.type,
253 			  found_key.offset);
254 	}
255 	return ret;
256 }
257 
258 static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
259 				      int mirror_num)
260 {
261 	struct btrfs_fs_info *fs_info = eb->fs_info;
262 	u64 start = eb->start;
263 	int i, num_pages = num_extent_pages(eb);
264 	int ret = 0;
265 
266 	if (sb_rdonly(fs_info->sb))
267 		return -EROFS;
268 
269 	for (i = 0; i < num_pages; i++) {
270 		struct page *p = eb->pages[i];
271 
272 		ret = btrfs_repair_io_failure(fs_info, 0, start, PAGE_SIZE,
273 				start, p, start - page_offset(p), mirror_num);
274 		if (ret)
275 			break;
276 		start += PAGE_SIZE;
277 	}
278 
279 	return ret;
280 }
281 
282 /*
283  * helper to read a given tree block, doing retries as required when
284  * the checksums don't match and we have alternate mirrors to try.
285  *
286  * @check:		expected tree parentness check, see the comments of the
287  *			structure for details.
288  */
289 int btrfs_read_extent_buffer(struct extent_buffer *eb,
290 			     struct btrfs_tree_parent_check *check)
291 {
292 	struct btrfs_fs_info *fs_info = eb->fs_info;
293 	int failed = 0;
294 	int ret;
295 	int num_copies = 0;
296 	int mirror_num = 0;
297 	int failed_mirror = 0;
298 
299 	ASSERT(check);
300 
301 	while (1) {
302 		clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
303 		ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
304 		if (!ret)
305 			break;
306 
307 		num_copies = btrfs_num_copies(fs_info,
308 					      eb->start, eb->len);
309 		if (num_copies == 1)
310 			break;
311 
312 		if (!failed_mirror) {
313 			failed = 1;
314 			failed_mirror = eb->read_mirror;
315 		}
316 
317 		mirror_num++;
318 		if (mirror_num == failed_mirror)
319 			mirror_num++;
320 
321 		if (mirror_num > num_copies)
322 			break;
323 	}
324 
325 	if (failed && !ret && failed_mirror)
326 		btrfs_repair_eb_io_failure(eb, failed_mirror);
327 
328 	return ret;
329 }
330 
331 static int csum_one_extent_buffer(struct extent_buffer *eb)
332 {
333 	struct btrfs_fs_info *fs_info = eb->fs_info;
334 	u8 result[BTRFS_CSUM_SIZE];
335 	int ret;
336 
337 	ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
338 				    offsetof(struct btrfs_header, fsid),
339 				    BTRFS_FSID_SIZE) == 0);
340 	csum_tree_block(eb, result);
341 
342 	if (btrfs_header_level(eb))
343 		ret = btrfs_check_node(eb);
344 	else
345 		ret = btrfs_check_leaf_full(eb);
346 
347 	if (ret < 0)
348 		goto error;
349 
350 	/*
351 	 * Also check the generation, the eb reached here must be newer than
352 	 * last committed. Or something seriously wrong happened.
353 	 */
354 	if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) {
355 		ret = -EUCLEAN;
356 		btrfs_err(fs_info,
357 			"block=%llu bad generation, have %llu expect > %llu",
358 			  eb->start, btrfs_header_generation(eb),
359 			  fs_info->last_trans_committed);
360 		goto error;
361 	}
362 	write_extent_buffer(eb, result, 0, fs_info->csum_size);
363 
364 	return 0;
365 
366 error:
367 	btrfs_print_tree(eb, 0);
368 	btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
369 		  eb->start);
370 	WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
371 	return ret;
372 }
373 
374 /* Checksum all dirty extent buffers in one bio_vec */
375 static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
376 				      struct bio_vec *bvec)
377 {
378 	struct page *page = bvec->bv_page;
379 	u64 bvec_start = page_offset(page) + bvec->bv_offset;
380 	u64 cur;
381 	int ret = 0;
382 
383 	for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
384 	     cur += fs_info->nodesize) {
385 		struct extent_buffer *eb;
386 		bool uptodate;
387 
388 		eb = find_extent_buffer(fs_info, cur);
389 		uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
390 						       fs_info->nodesize);
391 
392 		/* A dirty eb shouldn't disappear from buffer_radix */
393 		if (WARN_ON(!eb))
394 			return -EUCLEAN;
395 
396 		if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
397 			free_extent_buffer(eb);
398 			return -EUCLEAN;
399 		}
400 		if (WARN_ON(!uptodate)) {
401 			free_extent_buffer(eb);
402 			return -EUCLEAN;
403 		}
404 
405 		ret = csum_one_extent_buffer(eb);
406 		free_extent_buffer(eb);
407 		if (ret < 0)
408 			return ret;
409 	}
410 	return ret;
411 }
412 
413 /*
414  * Checksum a dirty tree block before IO.  This has extra checks to make sure
415  * we only fill in the checksum field in the first page of a multi-page block.
416  * For subpage extent buffers we need bvec to also read the offset in the page.
417  */
418 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
419 {
420 	struct page *page = bvec->bv_page;
421 	u64 start = page_offset(page);
422 	u64 found_start;
423 	struct extent_buffer *eb;
424 
425 	if (fs_info->nodesize < PAGE_SIZE)
426 		return csum_dirty_subpage_buffers(fs_info, bvec);
427 
428 	eb = (struct extent_buffer *)page->private;
429 	if (page != eb->pages[0])
430 		return 0;
431 
432 	found_start = btrfs_header_bytenr(eb);
433 
434 	if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
435 		WARN_ON(found_start != 0);
436 		return 0;
437 	}
438 
439 	/*
440 	 * Please do not consolidate these warnings into a single if.
441 	 * It is useful to know what went wrong.
442 	 */
443 	if (WARN_ON(found_start != start))
444 		return -EUCLEAN;
445 	if (WARN_ON(!PageUptodate(page)))
446 		return -EUCLEAN;
447 
448 	return csum_one_extent_buffer(eb);
449 }
450 
451 static int check_tree_block_fsid(struct extent_buffer *eb)
452 {
453 	struct btrfs_fs_info *fs_info = eb->fs_info;
454 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
455 	u8 fsid[BTRFS_FSID_SIZE];
456 	u8 *metadata_uuid;
457 
458 	read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
459 			   BTRFS_FSID_SIZE);
460 	/*
461 	 * Checking the incompat flag is only valid for the current fs. For
462 	 * seed devices it's forbidden to have their uuid changed so reading
463 	 * ->fsid in this case is fine
464 	 */
465 	if (btrfs_fs_incompat(fs_info, METADATA_UUID))
466 		metadata_uuid = fs_devices->metadata_uuid;
467 	else
468 		metadata_uuid = fs_devices->fsid;
469 
470 	if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
471 		return 0;
472 
473 	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
474 		if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
475 			return 0;
476 
477 	return 1;
478 }
479 
480 /* Do basic extent buffer checks at read time */
481 static int validate_extent_buffer(struct extent_buffer *eb,
482 				  struct btrfs_tree_parent_check *check)
483 {
484 	struct btrfs_fs_info *fs_info = eb->fs_info;
485 	u64 found_start;
486 	const u32 csum_size = fs_info->csum_size;
487 	u8 found_level;
488 	u8 result[BTRFS_CSUM_SIZE];
489 	const u8 *header_csum;
490 	int ret = 0;
491 
492 	ASSERT(check);
493 
494 	found_start = btrfs_header_bytenr(eb);
495 	if (found_start != eb->start) {
496 		btrfs_err_rl(fs_info,
497 			"bad tree block start, mirror %u want %llu have %llu",
498 			     eb->read_mirror, eb->start, found_start);
499 		ret = -EIO;
500 		goto out;
501 	}
502 	if (check_tree_block_fsid(eb)) {
503 		btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
504 			     eb->start, eb->read_mirror);
505 		ret = -EIO;
506 		goto out;
507 	}
508 	found_level = btrfs_header_level(eb);
509 	if (found_level >= BTRFS_MAX_LEVEL) {
510 		btrfs_err(fs_info,
511 			"bad tree block level, mirror %u level %d on logical %llu",
512 			eb->read_mirror, btrfs_header_level(eb), eb->start);
513 		ret = -EIO;
514 		goto out;
515 	}
516 
517 	csum_tree_block(eb, result);
518 	header_csum = page_address(eb->pages[0]) +
519 		get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
520 
521 	if (memcmp(result, header_csum, csum_size) != 0) {
522 		btrfs_warn_rl(fs_info,
523 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
524 			      eb->start, eb->read_mirror,
525 			      CSUM_FMT_VALUE(csum_size, header_csum),
526 			      CSUM_FMT_VALUE(csum_size, result),
527 			      btrfs_header_level(eb));
528 		ret = -EUCLEAN;
529 		goto out;
530 	}
531 
532 	if (found_level != check->level) {
533 		ret = -EIO;
534 		goto out;
535 	}
536 	if (unlikely(check->transid &&
537 		     btrfs_header_generation(eb) != check->transid)) {
538 		btrfs_err_rl(eb->fs_info,
539 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
540 				eb->start, eb->read_mirror, check->transid,
541 				btrfs_header_generation(eb));
542 		ret = -EIO;
543 		goto out;
544 	}
545 	if (check->has_first_key) {
546 		struct btrfs_key *expect_key = &check->first_key;
547 		struct btrfs_key found_key;
548 
549 		if (found_level)
550 			btrfs_node_key_to_cpu(eb, &found_key, 0);
551 		else
552 			btrfs_item_key_to_cpu(eb, &found_key, 0);
553 		if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
554 			btrfs_err(fs_info,
555 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
556 				  eb->start, check->transid,
557 				  expect_key->objectid,
558 				  expect_key->type, expect_key->offset,
559 				  found_key.objectid, found_key.type,
560 				  found_key.offset);
561 			ret = -EUCLEAN;
562 			goto out;
563 		}
564 	}
565 	if (check->owner_root) {
566 		ret = btrfs_check_eb_owner(eb, check->owner_root);
567 		if (ret < 0)
568 			goto out;
569 	}
570 
571 	/*
572 	 * If this is a leaf block and it is corrupt, set the corrupt bit so
573 	 * that we don't try and read the other copies of this block, just
574 	 * return -EIO.
575 	 */
576 	if (found_level == 0 && btrfs_check_leaf_full(eb)) {
577 		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
578 		ret = -EIO;
579 	}
580 
581 	if (found_level > 0 && btrfs_check_node(eb))
582 		ret = -EIO;
583 
584 	if (!ret)
585 		set_extent_buffer_uptodate(eb);
586 	else
587 		btrfs_err(fs_info,
588 		"read time tree block corruption detected on logical %llu mirror %u",
589 			  eb->start, eb->read_mirror);
590 out:
591 	return ret;
592 }
593 
594 static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
595 				   int mirror, struct btrfs_tree_parent_check *check)
596 {
597 	struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
598 	struct extent_buffer *eb;
599 	bool reads_done;
600 	int ret = 0;
601 
602 	ASSERT(check);
603 
604 	/*
605 	 * We don't allow bio merge for subpage metadata read, so we should
606 	 * only get one eb for each endio hook.
607 	 */
608 	ASSERT(end == start + fs_info->nodesize - 1);
609 	ASSERT(PagePrivate(page));
610 
611 	eb = find_extent_buffer(fs_info, start);
612 	/*
613 	 * When we are reading one tree block, eb must have been inserted into
614 	 * the radix tree. If not, something is wrong.
615 	 */
616 	ASSERT(eb);
617 
618 	reads_done = atomic_dec_and_test(&eb->io_pages);
619 	/* Subpage read must finish in page read */
620 	ASSERT(reads_done);
621 
622 	eb->read_mirror = mirror;
623 	if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
624 		ret = -EIO;
625 		goto err;
626 	}
627 	ret = validate_extent_buffer(eb, check);
628 	if (ret < 0)
629 		goto err;
630 
631 	set_extent_buffer_uptodate(eb);
632 
633 	free_extent_buffer(eb);
634 	return ret;
635 err:
636 	/*
637 	 * end_bio_extent_readpage decrements io_pages in case of error,
638 	 * make sure it has something to decrement.
639 	 */
640 	atomic_inc(&eb->io_pages);
641 	clear_extent_buffer_uptodate(eb);
642 	free_extent_buffer(eb);
643 	return ret;
644 }
645 
646 int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio,
647 				   struct page *page, u64 start, u64 end,
648 				   int mirror)
649 {
650 	struct extent_buffer *eb;
651 	int ret = 0;
652 	int reads_done;
653 
654 	ASSERT(page->private);
655 
656 	if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
657 		return validate_subpage_buffer(page, start, end, mirror,
658 					       &bbio->parent_check);
659 
660 	eb = (struct extent_buffer *)page->private;
661 
662 	/*
663 	 * The pending IO might have been the only thing that kept this buffer
664 	 * in memory.  Make sure we have a ref for all this other checks
665 	 */
666 	atomic_inc(&eb->refs);
667 
668 	reads_done = atomic_dec_and_test(&eb->io_pages);
669 	if (!reads_done)
670 		goto err;
671 
672 	eb->read_mirror = mirror;
673 	if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
674 		ret = -EIO;
675 		goto err;
676 	}
677 	ret = validate_extent_buffer(eb, &bbio->parent_check);
678 err:
679 	if (ret) {
680 		/*
681 		 * our io error hook is going to dec the io pages
682 		 * again, we have to make sure it has something
683 		 * to decrement
684 		 */
685 		atomic_inc(&eb->io_pages);
686 		clear_extent_buffer_uptodate(eb);
687 	}
688 	free_extent_buffer(eb);
689 
690 	return ret;
691 }
692 
693 static void run_one_async_start(struct btrfs_work *work)
694 {
695 	struct async_submit_bio *async;
696 	blk_status_t ret;
697 
698 	async = container_of(work, struct  async_submit_bio, work);
699 	switch (async->submit_cmd) {
700 	case WQ_SUBMIT_METADATA:
701 		ret = btree_submit_bio_start(async->bio);
702 		break;
703 	case WQ_SUBMIT_DATA:
704 		ret = btrfs_submit_bio_start(async->inode, async->bio);
705 		break;
706 	case WQ_SUBMIT_DATA_DIO:
707 		ret = btrfs_submit_bio_start_direct_io(async->inode,
708 				async->bio, async->dio_file_offset);
709 		break;
710 	}
711 	if (ret)
712 		async->status = ret;
713 }
714 
715 /*
716  * In order to insert checksums into the metadata in large chunks, we wait
717  * until bio submission time.   All the pages in the bio are checksummed and
718  * sums are attached onto the ordered extent record.
719  *
720  * At IO completion time the csums attached on the ordered extent record are
721  * inserted into the tree.
722  */
723 static void run_one_async_done(struct btrfs_work *work)
724 {
725 	struct async_submit_bio *async =
726 		container_of(work, struct  async_submit_bio, work);
727 	struct btrfs_inode *inode = async->inode;
728 	struct btrfs_bio *bbio = btrfs_bio(async->bio);
729 
730 	/* If an error occurred we just want to clean up the bio and move on */
731 	if (async->status) {
732 		btrfs_bio_end_io(bbio, async->status);
733 		return;
734 	}
735 
736 	/*
737 	 * All of the bios that pass through here are from async helpers.
738 	 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
739 	 * This changes nothing when cgroups aren't in use.
740 	 */
741 	async->bio->bi_opf |= REQ_CGROUP_PUNT;
742 	btrfs_submit_bio(inode->root->fs_info, async->bio, async->mirror_num);
743 }
744 
745 static void run_one_async_free(struct btrfs_work *work)
746 {
747 	struct async_submit_bio *async;
748 
749 	async = container_of(work, struct  async_submit_bio, work);
750 	kfree(async);
751 }
752 
753 /*
754  * Submit bio to an async queue.
755  *
756  * Retrun:
757  * - true if the work has been succesfuly submitted
758  * - false in case of error
759  */
760 bool btrfs_wq_submit_bio(struct btrfs_inode *inode, struct bio *bio, int mirror_num,
761 			 u64 dio_file_offset, enum btrfs_wq_submit_cmd cmd)
762 {
763 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
764 	struct async_submit_bio *async;
765 
766 	async = kmalloc(sizeof(*async), GFP_NOFS);
767 	if (!async)
768 		return false;
769 
770 	async->inode = inode;
771 	async->bio = bio;
772 	async->mirror_num = mirror_num;
773 	async->submit_cmd = cmd;
774 
775 	btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
776 			run_one_async_free);
777 
778 	async->dio_file_offset = dio_file_offset;
779 
780 	async->status = 0;
781 
782 	if (op_is_sync(bio->bi_opf))
783 		btrfs_queue_work(fs_info->hipri_workers, &async->work);
784 	else
785 		btrfs_queue_work(fs_info->workers, &async->work);
786 	return true;
787 }
788 
789 static blk_status_t btree_csum_one_bio(struct bio *bio)
790 {
791 	struct bio_vec *bvec;
792 	struct btrfs_root *root;
793 	int ret = 0;
794 	struct bvec_iter_all iter_all;
795 
796 	ASSERT(!bio_flagged(bio, BIO_CLONED));
797 	bio_for_each_segment_all(bvec, bio, iter_all) {
798 		root = BTRFS_I(bvec->bv_page->mapping->host)->root;
799 		ret = csum_dirty_buffer(root->fs_info, bvec);
800 		if (ret)
801 			break;
802 	}
803 
804 	return errno_to_blk_status(ret);
805 }
806 
807 blk_status_t btree_submit_bio_start(struct bio *bio)
808 {
809 	/*
810 	 * when we're called for a write, we're already in the async
811 	 * submission context.  Just jump into btrfs_submit_bio.
812 	 */
813 	return btree_csum_one_bio(bio);
814 }
815 
816 static bool should_async_write(struct btrfs_fs_info *fs_info,
817 			     struct btrfs_inode *bi)
818 {
819 	if (btrfs_is_zoned(fs_info))
820 		return false;
821 	if (atomic_read(&bi->sync_writers))
822 		return false;
823 	if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
824 		return false;
825 	return true;
826 }
827 
828 void btrfs_submit_metadata_bio(struct btrfs_inode *inode, struct bio *bio, int mirror_num)
829 {
830 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
831 	struct btrfs_bio *bbio = btrfs_bio(bio);
832 	blk_status_t ret;
833 
834 	bio->bi_opf |= REQ_META;
835 	bbio->is_metadata = 1;
836 
837 	if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
838 		btrfs_submit_bio(fs_info, bio, mirror_num);
839 		return;
840 	}
841 
842 	/*
843 	 * Kthread helpers are used to submit writes so that checksumming can
844 	 * happen in parallel across all CPUs.
845 	 */
846 	if (should_async_write(fs_info, inode) &&
847 	    btrfs_wq_submit_bio(inode, bio, mirror_num, 0, WQ_SUBMIT_METADATA))
848 		return;
849 
850 	ret = btree_csum_one_bio(bio);
851 	if (ret) {
852 		btrfs_bio_end_io(bbio, ret);
853 		return;
854 	}
855 
856 	btrfs_submit_bio(fs_info, bio, mirror_num);
857 }
858 
859 #ifdef CONFIG_MIGRATION
860 static int btree_migrate_folio(struct address_space *mapping,
861 		struct folio *dst, struct folio *src, enum migrate_mode mode)
862 {
863 	/*
864 	 * we can't safely write a btree page from here,
865 	 * we haven't done the locking hook
866 	 */
867 	if (folio_test_dirty(src))
868 		return -EAGAIN;
869 	/*
870 	 * Buffers may be managed in a filesystem specific way.
871 	 * We must have no buffers or drop them.
872 	 */
873 	if (folio_get_private(src) &&
874 	    !filemap_release_folio(src, GFP_KERNEL))
875 		return -EAGAIN;
876 	return migrate_folio(mapping, dst, src, mode);
877 }
878 #else
879 #define btree_migrate_folio NULL
880 #endif
881 
882 static int btree_writepages(struct address_space *mapping,
883 			    struct writeback_control *wbc)
884 {
885 	struct btrfs_fs_info *fs_info;
886 	int ret;
887 
888 	if (wbc->sync_mode == WB_SYNC_NONE) {
889 
890 		if (wbc->for_kupdate)
891 			return 0;
892 
893 		fs_info = BTRFS_I(mapping->host)->root->fs_info;
894 		/* this is a bit racy, but that's ok */
895 		ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
896 					     BTRFS_DIRTY_METADATA_THRESH,
897 					     fs_info->dirty_metadata_batch);
898 		if (ret < 0)
899 			return 0;
900 	}
901 	return btree_write_cache_pages(mapping, wbc);
902 }
903 
904 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
905 {
906 	if (folio_test_writeback(folio) || folio_test_dirty(folio))
907 		return false;
908 
909 	return try_release_extent_buffer(&folio->page);
910 }
911 
912 static void btree_invalidate_folio(struct folio *folio, size_t offset,
913 				 size_t length)
914 {
915 	struct extent_io_tree *tree;
916 	tree = &BTRFS_I(folio->mapping->host)->io_tree;
917 	extent_invalidate_folio(tree, folio, offset);
918 	btree_release_folio(folio, GFP_NOFS);
919 	if (folio_get_private(folio)) {
920 		btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
921 			   "folio private not zero on folio %llu",
922 			   (unsigned long long)folio_pos(folio));
923 		folio_detach_private(folio);
924 	}
925 }
926 
927 #ifdef DEBUG
928 static bool btree_dirty_folio(struct address_space *mapping,
929 		struct folio *folio)
930 {
931 	struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
932 	struct btrfs_subpage *subpage;
933 	struct extent_buffer *eb;
934 	int cur_bit = 0;
935 	u64 page_start = folio_pos(folio);
936 
937 	if (fs_info->sectorsize == PAGE_SIZE) {
938 		eb = folio_get_private(folio);
939 		BUG_ON(!eb);
940 		BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
941 		BUG_ON(!atomic_read(&eb->refs));
942 		btrfs_assert_tree_write_locked(eb);
943 		return filemap_dirty_folio(mapping, folio);
944 	}
945 	subpage = folio_get_private(folio);
946 
947 	ASSERT(subpage->dirty_bitmap);
948 	while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
949 		unsigned long flags;
950 		u64 cur;
951 		u16 tmp = (1 << cur_bit);
952 
953 		spin_lock_irqsave(&subpage->lock, flags);
954 		if (!(tmp & subpage->dirty_bitmap)) {
955 			spin_unlock_irqrestore(&subpage->lock, flags);
956 			cur_bit++;
957 			continue;
958 		}
959 		spin_unlock_irqrestore(&subpage->lock, flags);
960 		cur = page_start + cur_bit * fs_info->sectorsize;
961 
962 		eb = find_extent_buffer(fs_info, cur);
963 		ASSERT(eb);
964 		ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
965 		ASSERT(atomic_read(&eb->refs));
966 		btrfs_assert_tree_write_locked(eb);
967 		free_extent_buffer(eb);
968 
969 		cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
970 	}
971 	return filemap_dirty_folio(mapping, folio);
972 }
973 #else
974 #define btree_dirty_folio filemap_dirty_folio
975 #endif
976 
977 static const struct address_space_operations btree_aops = {
978 	.writepages	= btree_writepages,
979 	.release_folio	= btree_release_folio,
980 	.invalidate_folio = btree_invalidate_folio,
981 	.migrate_folio	= btree_migrate_folio,
982 	.dirty_folio	= btree_dirty_folio,
983 };
984 
985 struct extent_buffer *btrfs_find_create_tree_block(
986 						struct btrfs_fs_info *fs_info,
987 						u64 bytenr, u64 owner_root,
988 						int level)
989 {
990 	if (btrfs_is_testing(fs_info))
991 		return alloc_test_extent_buffer(fs_info, bytenr);
992 	return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
993 }
994 
995 /*
996  * Read tree block at logical address @bytenr and do variant basic but critical
997  * verification.
998  *
999  * @check:		expected tree parentness check, see comments of the
1000  *			structure for details.
1001  */
1002 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1003 				      struct btrfs_tree_parent_check *check)
1004 {
1005 	struct extent_buffer *buf = NULL;
1006 	int ret;
1007 
1008 	ASSERT(check);
1009 
1010 	buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
1011 					   check->level);
1012 	if (IS_ERR(buf))
1013 		return buf;
1014 
1015 	ret = btrfs_read_extent_buffer(buf, check);
1016 	if (ret) {
1017 		free_extent_buffer_stale(buf);
1018 		return ERR_PTR(ret);
1019 	}
1020 	if (btrfs_check_eb_owner(buf, check->owner_root)) {
1021 		free_extent_buffer_stale(buf);
1022 		return ERR_PTR(-EUCLEAN);
1023 	}
1024 	return buf;
1025 
1026 }
1027 
1028 void btrfs_clean_tree_block(struct extent_buffer *buf)
1029 {
1030 	struct btrfs_fs_info *fs_info = buf->fs_info;
1031 	if (btrfs_header_generation(buf) ==
1032 	    fs_info->running_transaction->transid) {
1033 		btrfs_assert_tree_write_locked(buf);
1034 
1035 		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1036 			percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1037 						 -buf->len,
1038 						 fs_info->dirty_metadata_batch);
1039 			clear_extent_buffer_dirty(buf);
1040 		}
1041 	}
1042 }
1043 
1044 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1045 			 u64 objectid)
1046 {
1047 	bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1048 
1049 	memset(&root->root_key, 0, sizeof(root->root_key));
1050 	memset(&root->root_item, 0, sizeof(root->root_item));
1051 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1052 	root->fs_info = fs_info;
1053 	root->root_key.objectid = objectid;
1054 	root->node = NULL;
1055 	root->commit_root = NULL;
1056 	root->state = 0;
1057 	RB_CLEAR_NODE(&root->rb_node);
1058 
1059 	root->last_trans = 0;
1060 	root->free_objectid = 0;
1061 	root->nr_delalloc_inodes = 0;
1062 	root->nr_ordered_extents = 0;
1063 	root->inode_tree = RB_ROOT;
1064 	INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1065 
1066 	btrfs_init_root_block_rsv(root);
1067 
1068 	INIT_LIST_HEAD(&root->dirty_list);
1069 	INIT_LIST_HEAD(&root->root_list);
1070 	INIT_LIST_HEAD(&root->delalloc_inodes);
1071 	INIT_LIST_HEAD(&root->delalloc_root);
1072 	INIT_LIST_HEAD(&root->ordered_extents);
1073 	INIT_LIST_HEAD(&root->ordered_root);
1074 	INIT_LIST_HEAD(&root->reloc_dirty_list);
1075 	INIT_LIST_HEAD(&root->logged_list[0]);
1076 	INIT_LIST_HEAD(&root->logged_list[1]);
1077 	spin_lock_init(&root->inode_lock);
1078 	spin_lock_init(&root->delalloc_lock);
1079 	spin_lock_init(&root->ordered_extent_lock);
1080 	spin_lock_init(&root->accounting_lock);
1081 	spin_lock_init(&root->log_extents_lock[0]);
1082 	spin_lock_init(&root->log_extents_lock[1]);
1083 	spin_lock_init(&root->qgroup_meta_rsv_lock);
1084 	mutex_init(&root->objectid_mutex);
1085 	mutex_init(&root->log_mutex);
1086 	mutex_init(&root->ordered_extent_mutex);
1087 	mutex_init(&root->delalloc_mutex);
1088 	init_waitqueue_head(&root->qgroup_flush_wait);
1089 	init_waitqueue_head(&root->log_writer_wait);
1090 	init_waitqueue_head(&root->log_commit_wait[0]);
1091 	init_waitqueue_head(&root->log_commit_wait[1]);
1092 	INIT_LIST_HEAD(&root->log_ctxs[0]);
1093 	INIT_LIST_HEAD(&root->log_ctxs[1]);
1094 	atomic_set(&root->log_commit[0], 0);
1095 	atomic_set(&root->log_commit[1], 0);
1096 	atomic_set(&root->log_writers, 0);
1097 	atomic_set(&root->log_batch, 0);
1098 	refcount_set(&root->refs, 1);
1099 	atomic_set(&root->snapshot_force_cow, 0);
1100 	atomic_set(&root->nr_swapfiles, 0);
1101 	root->log_transid = 0;
1102 	root->log_transid_committed = -1;
1103 	root->last_log_commit = 0;
1104 	root->anon_dev = 0;
1105 	if (!dummy) {
1106 		extent_io_tree_init(fs_info, &root->dirty_log_pages,
1107 				    IO_TREE_ROOT_DIRTY_LOG_PAGES);
1108 		extent_io_tree_init(fs_info, &root->log_csum_range,
1109 				    IO_TREE_LOG_CSUM_RANGE);
1110 	}
1111 
1112 	spin_lock_init(&root->root_item_lock);
1113 	btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1114 #ifdef CONFIG_BTRFS_DEBUG
1115 	INIT_LIST_HEAD(&root->leak_list);
1116 	spin_lock(&fs_info->fs_roots_radix_lock);
1117 	list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1118 	spin_unlock(&fs_info->fs_roots_radix_lock);
1119 #endif
1120 }
1121 
1122 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1123 					   u64 objectid, gfp_t flags)
1124 {
1125 	struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1126 	if (root)
1127 		__setup_root(root, fs_info, objectid);
1128 	return root;
1129 }
1130 
1131 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1132 /* Should only be used by the testing infrastructure */
1133 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1134 {
1135 	struct btrfs_root *root;
1136 
1137 	if (!fs_info)
1138 		return ERR_PTR(-EINVAL);
1139 
1140 	root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1141 	if (!root)
1142 		return ERR_PTR(-ENOMEM);
1143 
1144 	/* We don't use the stripesize in selftest, set it as sectorsize */
1145 	root->alloc_bytenr = 0;
1146 
1147 	return root;
1148 }
1149 #endif
1150 
1151 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
1152 {
1153 	const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
1154 	const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
1155 
1156 	return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
1157 }
1158 
1159 static int global_root_key_cmp(const void *k, const struct rb_node *node)
1160 {
1161 	const struct btrfs_key *key = k;
1162 	const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
1163 
1164 	return btrfs_comp_cpu_keys(key, &root->root_key);
1165 }
1166 
1167 int btrfs_global_root_insert(struct btrfs_root *root)
1168 {
1169 	struct btrfs_fs_info *fs_info = root->fs_info;
1170 	struct rb_node *tmp;
1171 
1172 	write_lock(&fs_info->global_root_lock);
1173 	tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
1174 	write_unlock(&fs_info->global_root_lock);
1175 	ASSERT(!tmp);
1176 
1177 	return tmp ? -EEXIST : 0;
1178 }
1179 
1180 void btrfs_global_root_delete(struct btrfs_root *root)
1181 {
1182 	struct btrfs_fs_info *fs_info = root->fs_info;
1183 
1184 	write_lock(&fs_info->global_root_lock);
1185 	rb_erase(&root->rb_node, &fs_info->global_root_tree);
1186 	write_unlock(&fs_info->global_root_lock);
1187 }
1188 
1189 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
1190 				     struct btrfs_key *key)
1191 {
1192 	struct rb_node *node;
1193 	struct btrfs_root *root = NULL;
1194 
1195 	read_lock(&fs_info->global_root_lock);
1196 	node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
1197 	if (node)
1198 		root = container_of(node, struct btrfs_root, rb_node);
1199 	read_unlock(&fs_info->global_root_lock);
1200 
1201 	return root;
1202 }
1203 
1204 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
1205 {
1206 	struct btrfs_block_group *block_group;
1207 	u64 ret;
1208 
1209 	if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
1210 		return 0;
1211 
1212 	if (bytenr)
1213 		block_group = btrfs_lookup_block_group(fs_info, bytenr);
1214 	else
1215 		block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
1216 	ASSERT(block_group);
1217 	if (!block_group)
1218 		return 0;
1219 	ret = block_group->global_root_id;
1220 	btrfs_put_block_group(block_group);
1221 
1222 	return ret;
1223 }
1224 
1225 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1226 {
1227 	struct btrfs_key key = {
1228 		.objectid = BTRFS_CSUM_TREE_OBJECTID,
1229 		.type = BTRFS_ROOT_ITEM_KEY,
1230 		.offset = btrfs_global_root_id(fs_info, bytenr),
1231 	};
1232 
1233 	return btrfs_global_root(fs_info, &key);
1234 }
1235 
1236 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1237 {
1238 	struct btrfs_key key = {
1239 		.objectid = BTRFS_EXTENT_TREE_OBJECTID,
1240 		.type = BTRFS_ROOT_ITEM_KEY,
1241 		.offset = btrfs_global_root_id(fs_info, bytenr),
1242 	};
1243 
1244 	return btrfs_global_root(fs_info, &key);
1245 }
1246 
1247 struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info)
1248 {
1249 	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE))
1250 		return fs_info->block_group_root;
1251 	return btrfs_extent_root(fs_info, 0);
1252 }
1253 
1254 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1255 				     u64 objectid)
1256 {
1257 	struct btrfs_fs_info *fs_info = trans->fs_info;
1258 	struct extent_buffer *leaf;
1259 	struct btrfs_root *tree_root = fs_info->tree_root;
1260 	struct btrfs_root *root;
1261 	struct btrfs_key key;
1262 	unsigned int nofs_flag;
1263 	int ret = 0;
1264 
1265 	/*
1266 	 * We're holding a transaction handle, so use a NOFS memory allocation
1267 	 * context to avoid deadlock if reclaim happens.
1268 	 */
1269 	nofs_flag = memalloc_nofs_save();
1270 	root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1271 	memalloc_nofs_restore(nofs_flag);
1272 	if (!root)
1273 		return ERR_PTR(-ENOMEM);
1274 
1275 	root->root_key.objectid = objectid;
1276 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1277 	root->root_key.offset = 0;
1278 
1279 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1280 				      BTRFS_NESTING_NORMAL);
1281 	if (IS_ERR(leaf)) {
1282 		ret = PTR_ERR(leaf);
1283 		leaf = NULL;
1284 		goto fail;
1285 	}
1286 
1287 	root->node = leaf;
1288 	btrfs_mark_buffer_dirty(leaf);
1289 
1290 	root->commit_root = btrfs_root_node(root);
1291 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1292 
1293 	btrfs_set_root_flags(&root->root_item, 0);
1294 	btrfs_set_root_limit(&root->root_item, 0);
1295 	btrfs_set_root_bytenr(&root->root_item, leaf->start);
1296 	btrfs_set_root_generation(&root->root_item, trans->transid);
1297 	btrfs_set_root_level(&root->root_item, 0);
1298 	btrfs_set_root_refs(&root->root_item, 1);
1299 	btrfs_set_root_used(&root->root_item, leaf->len);
1300 	btrfs_set_root_last_snapshot(&root->root_item, 0);
1301 	btrfs_set_root_dirid(&root->root_item, 0);
1302 	if (is_fstree(objectid))
1303 		generate_random_guid(root->root_item.uuid);
1304 	else
1305 		export_guid(root->root_item.uuid, &guid_null);
1306 	btrfs_set_root_drop_level(&root->root_item, 0);
1307 
1308 	btrfs_tree_unlock(leaf);
1309 
1310 	key.objectid = objectid;
1311 	key.type = BTRFS_ROOT_ITEM_KEY;
1312 	key.offset = 0;
1313 	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1314 	if (ret)
1315 		goto fail;
1316 
1317 	return root;
1318 
1319 fail:
1320 	btrfs_put_root(root);
1321 
1322 	return ERR_PTR(ret);
1323 }
1324 
1325 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1326 					 struct btrfs_fs_info *fs_info)
1327 {
1328 	struct btrfs_root *root;
1329 
1330 	root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1331 	if (!root)
1332 		return ERR_PTR(-ENOMEM);
1333 
1334 	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1335 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1336 	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1337 
1338 	return root;
1339 }
1340 
1341 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1342 			      struct btrfs_root *root)
1343 {
1344 	struct extent_buffer *leaf;
1345 
1346 	/*
1347 	 * DON'T set SHAREABLE bit for log trees.
1348 	 *
1349 	 * Log trees are not exposed to user space thus can't be snapshotted,
1350 	 * and they go away before a real commit is actually done.
1351 	 *
1352 	 * They do store pointers to file data extents, and those reference
1353 	 * counts still get updated (along with back refs to the log tree).
1354 	 */
1355 
1356 	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1357 			NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1358 	if (IS_ERR(leaf))
1359 		return PTR_ERR(leaf);
1360 
1361 	root->node = leaf;
1362 
1363 	btrfs_mark_buffer_dirty(root->node);
1364 	btrfs_tree_unlock(root->node);
1365 
1366 	return 0;
1367 }
1368 
1369 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1370 			     struct btrfs_fs_info *fs_info)
1371 {
1372 	struct btrfs_root *log_root;
1373 
1374 	log_root = alloc_log_tree(trans, fs_info);
1375 	if (IS_ERR(log_root))
1376 		return PTR_ERR(log_root);
1377 
1378 	if (!btrfs_is_zoned(fs_info)) {
1379 		int ret = btrfs_alloc_log_tree_node(trans, log_root);
1380 
1381 		if (ret) {
1382 			btrfs_put_root(log_root);
1383 			return ret;
1384 		}
1385 	}
1386 
1387 	WARN_ON(fs_info->log_root_tree);
1388 	fs_info->log_root_tree = log_root;
1389 	return 0;
1390 }
1391 
1392 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1393 		       struct btrfs_root *root)
1394 {
1395 	struct btrfs_fs_info *fs_info = root->fs_info;
1396 	struct btrfs_root *log_root;
1397 	struct btrfs_inode_item *inode_item;
1398 	int ret;
1399 
1400 	log_root = alloc_log_tree(trans, fs_info);
1401 	if (IS_ERR(log_root))
1402 		return PTR_ERR(log_root);
1403 
1404 	ret = btrfs_alloc_log_tree_node(trans, log_root);
1405 	if (ret) {
1406 		btrfs_put_root(log_root);
1407 		return ret;
1408 	}
1409 
1410 	log_root->last_trans = trans->transid;
1411 	log_root->root_key.offset = root->root_key.objectid;
1412 
1413 	inode_item = &log_root->root_item.inode;
1414 	btrfs_set_stack_inode_generation(inode_item, 1);
1415 	btrfs_set_stack_inode_size(inode_item, 3);
1416 	btrfs_set_stack_inode_nlink(inode_item, 1);
1417 	btrfs_set_stack_inode_nbytes(inode_item,
1418 				     fs_info->nodesize);
1419 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1420 
1421 	btrfs_set_root_node(&log_root->root_item, log_root->node);
1422 
1423 	WARN_ON(root->log_root);
1424 	root->log_root = log_root;
1425 	root->log_transid = 0;
1426 	root->log_transid_committed = -1;
1427 	root->last_log_commit = 0;
1428 	return 0;
1429 }
1430 
1431 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1432 					      struct btrfs_path *path,
1433 					      struct btrfs_key *key)
1434 {
1435 	struct btrfs_root *root;
1436 	struct btrfs_tree_parent_check check = { 0 };
1437 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1438 	u64 generation;
1439 	int ret;
1440 	int level;
1441 
1442 	root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1443 	if (!root)
1444 		return ERR_PTR(-ENOMEM);
1445 
1446 	ret = btrfs_find_root(tree_root, key, path,
1447 			      &root->root_item, &root->root_key);
1448 	if (ret) {
1449 		if (ret > 0)
1450 			ret = -ENOENT;
1451 		goto fail;
1452 	}
1453 
1454 	generation = btrfs_root_generation(&root->root_item);
1455 	level = btrfs_root_level(&root->root_item);
1456 	check.level = level;
1457 	check.transid = generation;
1458 	check.owner_root = key->objectid;
1459 	root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1460 				     &check);
1461 	if (IS_ERR(root->node)) {
1462 		ret = PTR_ERR(root->node);
1463 		root->node = NULL;
1464 		goto fail;
1465 	}
1466 	if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1467 		ret = -EIO;
1468 		goto fail;
1469 	}
1470 
1471 	/*
1472 	 * For real fs, and not log/reloc trees, root owner must
1473 	 * match its root node owner
1474 	 */
1475 	if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1476 	    root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1477 	    root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1478 	    root->root_key.objectid != btrfs_header_owner(root->node)) {
1479 		btrfs_crit(fs_info,
1480 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1481 			   root->root_key.objectid, root->node->start,
1482 			   btrfs_header_owner(root->node),
1483 			   root->root_key.objectid);
1484 		ret = -EUCLEAN;
1485 		goto fail;
1486 	}
1487 	root->commit_root = btrfs_root_node(root);
1488 	return root;
1489 fail:
1490 	btrfs_put_root(root);
1491 	return ERR_PTR(ret);
1492 }
1493 
1494 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1495 					struct btrfs_key *key)
1496 {
1497 	struct btrfs_root *root;
1498 	struct btrfs_path *path;
1499 
1500 	path = btrfs_alloc_path();
1501 	if (!path)
1502 		return ERR_PTR(-ENOMEM);
1503 	root = read_tree_root_path(tree_root, path, key);
1504 	btrfs_free_path(path);
1505 
1506 	return root;
1507 }
1508 
1509 /*
1510  * Initialize subvolume root in-memory structure
1511  *
1512  * @anon_dev:	anonymous device to attach to the root, if zero, allocate new
1513  */
1514 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1515 {
1516 	int ret;
1517 	unsigned int nofs_flag;
1518 
1519 	/*
1520 	 * We might be called under a transaction (e.g. indirect backref
1521 	 * resolution) which could deadlock if it triggers memory reclaim
1522 	 */
1523 	nofs_flag = memalloc_nofs_save();
1524 	ret = btrfs_drew_lock_init(&root->snapshot_lock);
1525 	memalloc_nofs_restore(nofs_flag);
1526 	if (ret)
1527 		goto fail;
1528 
1529 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1530 	    !btrfs_is_data_reloc_root(root)) {
1531 		set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1532 		btrfs_check_and_init_root_item(&root->root_item);
1533 	}
1534 
1535 	/*
1536 	 * Don't assign anonymous block device to roots that are not exposed to
1537 	 * userspace, the id pool is limited to 1M
1538 	 */
1539 	if (is_fstree(root->root_key.objectid) &&
1540 	    btrfs_root_refs(&root->root_item) > 0) {
1541 		if (!anon_dev) {
1542 			ret = get_anon_bdev(&root->anon_dev);
1543 			if (ret)
1544 				goto fail;
1545 		} else {
1546 			root->anon_dev = anon_dev;
1547 		}
1548 	}
1549 
1550 	mutex_lock(&root->objectid_mutex);
1551 	ret = btrfs_init_root_free_objectid(root);
1552 	if (ret) {
1553 		mutex_unlock(&root->objectid_mutex);
1554 		goto fail;
1555 	}
1556 
1557 	ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1558 
1559 	mutex_unlock(&root->objectid_mutex);
1560 
1561 	return 0;
1562 fail:
1563 	/* The caller is responsible to call btrfs_free_fs_root */
1564 	return ret;
1565 }
1566 
1567 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1568 					       u64 root_id)
1569 {
1570 	struct btrfs_root *root;
1571 
1572 	spin_lock(&fs_info->fs_roots_radix_lock);
1573 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1574 				 (unsigned long)root_id);
1575 	if (root)
1576 		root = btrfs_grab_root(root);
1577 	spin_unlock(&fs_info->fs_roots_radix_lock);
1578 	return root;
1579 }
1580 
1581 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1582 						u64 objectid)
1583 {
1584 	struct btrfs_key key = {
1585 		.objectid = objectid,
1586 		.type = BTRFS_ROOT_ITEM_KEY,
1587 		.offset = 0,
1588 	};
1589 
1590 	if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1591 		return btrfs_grab_root(fs_info->tree_root);
1592 	if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1593 		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1594 	if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1595 		return btrfs_grab_root(fs_info->chunk_root);
1596 	if (objectid == BTRFS_DEV_TREE_OBJECTID)
1597 		return btrfs_grab_root(fs_info->dev_root);
1598 	if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1599 		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1600 	if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1601 		return btrfs_grab_root(fs_info->quota_root) ?
1602 			fs_info->quota_root : ERR_PTR(-ENOENT);
1603 	if (objectid == BTRFS_UUID_TREE_OBJECTID)
1604 		return btrfs_grab_root(fs_info->uuid_root) ?
1605 			fs_info->uuid_root : ERR_PTR(-ENOENT);
1606 	if (objectid == BTRFS_BLOCK_GROUP_TREE_OBJECTID)
1607 		return btrfs_grab_root(fs_info->block_group_root) ?
1608 			fs_info->block_group_root : ERR_PTR(-ENOENT);
1609 	if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) {
1610 		struct btrfs_root *root = btrfs_global_root(fs_info, &key);
1611 
1612 		return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT);
1613 	}
1614 	return NULL;
1615 }
1616 
1617 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1618 			 struct btrfs_root *root)
1619 {
1620 	int ret;
1621 
1622 	ret = radix_tree_preload(GFP_NOFS);
1623 	if (ret)
1624 		return ret;
1625 
1626 	spin_lock(&fs_info->fs_roots_radix_lock);
1627 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1628 				(unsigned long)root->root_key.objectid,
1629 				root);
1630 	if (ret == 0) {
1631 		btrfs_grab_root(root);
1632 		set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1633 	}
1634 	spin_unlock(&fs_info->fs_roots_radix_lock);
1635 	radix_tree_preload_end();
1636 
1637 	return ret;
1638 }
1639 
1640 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1641 {
1642 #ifdef CONFIG_BTRFS_DEBUG
1643 	struct btrfs_root *root;
1644 
1645 	while (!list_empty(&fs_info->allocated_roots)) {
1646 		char buf[BTRFS_ROOT_NAME_BUF_LEN];
1647 
1648 		root = list_first_entry(&fs_info->allocated_roots,
1649 					struct btrfs_root, leak_list);
1650 		btrfs_err(fs_info, "leaked root %s refcount %d",
1651 			  btrfs_root_name(&root->root_key, buf),
1652 			  refcount_read(&root->refs));
1653 		while (refcount_read(&root->refs) > 1)
1654 			btrfs_put_root(root);
1655 		btrfs_put_root(root);
1656 	}
1657 #endif
1658 }
1659 
1660 static void free_global_roots(struct btrfs_fs_info *fs_info)
1661 {
1662 	struct btrfs_root *root;
1663 	struct rb_node *node;
1664 
1665 	while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1666 		root = rb_entry(node, struct btrfs_root, rb_node);
1667 		rb_erase(&root->rb_node, &fs_info->global_root_tree);
1668 		btrfs_put_root(root);
1669 	}
1670 }
1671 
1672 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1673 {
1674 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1675 	percpu_counter_destroy(&fs_info->delalloc_bytes);
1676 	percpu_counter_destroy(&fs_info->ordered_bytes);
1677 	percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1678 	btrfs_free_csum_hash(fs_info);
1679 	btrfs_free_stripe_hash_table(fs_info);
1680 	btrfs_free_ref_cache(fs_info);
1681 	kfree(fs_info->balance_ctl);
1682 	kfree(fs_info->delayed_root);
1683 	free_global_roots(fs_info);
1684 	btrfs_put_root(fs_info->tree_root);
1685 	btrfs_put_root(fs_info->chunk_root);
1686 	btrfs_put_root(fs_info->dev_root);
1687 	btrfs_put_root(fs_info->quota_root);
1688 	btrfs_put_root(fs_info->uuid_root);
1689 	btrfs_put_root(fs_info->fs_root);
1690 	btrfs_put_root(fs_info->data_reloc_root);
1691 	btrfs_put_root(fs_info->block_group_root);
1692 	btrfs_check_leaked_roots(fs_info);
1693 	btrfs_extent_buffer_leak_debug_check(fs_info);
1694 	kfree(fs_info->super_copy);
1695 	kfree(fs_info->super_for_commit);
1696 	kfree(fs_info->subpage_info);
1697 	kvfree(fs_info);
1698 }
1699 
1700 
1701 /*
1702  * Get an in-memory reference of a root structure.
1703  *
1704  * For essential trees like root/extent tree, we grab it from fs_info directly.
1705  * For subvolume trees, we check the cached filesystem roots first. If not
1706  * found, then read it from disk and add it to cached fs roots.
1707  *
1708  * Caller should release the root by calling btrfs_put_root() after the usage.
1709  *
1710  * NOTE: Reloc and log trees can't be read by this function as they share the
1711  *	 same root objectid.
1712  *
1713  * @objectid:	root id
1714  * @anon_dev:	preallocated anonymous block device number for new roots,
1715  * 		pass 0 for new allocation.
1716  * @check_ref:	whether to check root item references, If true, return -ENOENT
1717  *		for orphan roots
1718  */
1719 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1720 					     u64 objectid, dev_t anon_dev,
1721 					     bool check_ref)
1722 {
1723 	struct btrfs_root *root;
1724 	struct btrfs_path *path;
1725 	struct btrfs_key key;
1726 	int ret;
1727 
1728 	root = btrfs_get_global_root(fs_info, objectid);
1729 	if (root)
1730 		return root;
1731 again:
1732 	root = btrfs_lookup_fs_root(fs_info, objectid);
1733 	if (root) {
1734 		/* Shouldn't get preallocated anon_dev for cached roots */
1735 		ASSERT(!anon_dev);
1736 		if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1737 			btrfs_put_root(root);
1738 			return ERR_PTR(-ENOENT);
1739 		}
1740 		return root;
1741 	}
1742 
1743 	key.objectid = objectid;
1744 	key.type = BTRFS_ROOT_ITEM_KEY;
1745 	key.offset = (u64)-1;
1746 	root = btrfs_read_tree_root(fs_info->tree_root, &key);
1747 	if (IS_ERR(root))
1748 		return root;
1749 
1750 	if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1751 		ret = -ENOENT;
1752 		goto fail;
1753 	}
1754 
1755 	ret = btrfs_init_fs_root(root, anon_dev);
1756 	if (ret)
1757 		goto fail;
1758 
1759 	path = btrfs_alloc_path();
1760 	if (!path) {
1761 		ret = -ENOMEM;
1762 		goto fail;
1763 	}
1764 	key.objectid = BTRFS_ORPHAN_OBJECTID;
1765 	key.type = BTRFS_ORPHAN_ITEM_KEY;
1766 	key.offset = objectid;
1767 
1768 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1769 	btrfs_free_path(path);
1770 	if (ret < 0)
1771 		goto fail;
1772 	if (ret == 0)
1773 		set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1774 
1775 	ret = btrfs_insert_fs_root(fs_info, root);
1776 	if (ret) {
1777 		if (ret == -EEXIST) {
1778 			btrfs_put_root(root);
1779 			goto again;
1780 		}
1781 		goto fail;
1782 	}
1783 	return root;
1784 fail:
1785 	/*
1786 	 * If our caller provided us an anonymous device, then it's his
1787 	 * responsibility to free it in case we fail. So we have to set our
1788 	 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1789 	 * and once again by our caller.
1790 	 */
1791 	if (anon_dev)
1792 		root->anon_dev = 0;
1793 	btrfs_put_root(root);
1794 	return ERR_PTR(ret);
1795 }
1796 
1797 /*
1798  * Get in-memory reference of a root structure
1799  *
1800  * @objectid:	tree objectid
1801  * @check_ref:	if set, verify that the tree exists and the item has at least
1802  *		one reference
1803  */
1804 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1805 				     u64 objectid, bool check_ref)
1806 {
1807 	return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1808 }
1809 
1810 /*
1811  * Get in-memory reference of a root structure, created as new, optionally pass
1812  * the anonymous block device id
1813  *
1814  * @objectid:	tree objectid
1815  * @anon_dev:	if zero, allocate a new anonymous block device or use the
1816  *		parameter value
1817  */
1818 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1819 					 u64 objectid, dev_t anon_dev)
1820 {
1821 	return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1822 }
1823 
1824 /*
1825  * btrfs_get_fs_root_commit_root - return a root for the given objectid
1826  * @fs_info:	the fs_info
1827  * @objectid:	the objectid we need to lookup
1828  *
1829  * This is exclusively used for backref walking, and exists specifically because
1830  * of how qgroups does lookups.  Qgroups will do a backref lookup at delayed ref
1831  * creation time, which means we may have to read the tree_root in order to look
1832  * up a fs root that is not in memory.  If the root is not in memory we will
1833  * read the tree root commit root and look up the fs root from there.  This is a
1834  * temporary root, it will not be inserted into the radix tree as it doesn't
1835  * have the most uptodate information, it'll simply be discarded once the
1836  * backref code is finished using the root.
1837  */
1838 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1839 						 struct btrfs_path *path,
1840 						 u64 objectid)
1841 {
1842 	struct btrfs_root *root;
1843 	struct btrfs_key key;
1844 
1845 	ASSERT(path->search_commit_root && path->skip_locking);
1846 
1847 	/*
1848 	 * This can return -ENOENT if we ask for a root that doesn't exist, but
1849 	 * since this is called via the backref walking code we won't be looking
1850 	 * up a root that doesn't exist, unless there's corruption.  So if root
1851 	 * != NULL just return it.
1852 	 */
1853 	root = btrfs_get_global_root(fs_info, objectid);
1854 	if (root)
1855 		return root;
1856 
1857 	root = btrfs_lookup_fs_root(fs_info, objectid);
1858 	if (root)
1859 		return root;
1860 
1861 	key.objectid = objectid;
1862 	key.type = BTRFS_ROOT_ITEM_KEY;
1863 	key.offset = (u64)-1;
1864 	root = read_tree_root_path(fs_info->tree_root, path, &key);
1865 	btrfs_release_path(path);
1866 
1867 	return root;
1868 }
1869 
1870 static int cleaner_kthread(void *arg)
1871 {
1872 	struct btrfs_fs_info *fs_info = arg;
1873 	int again;
1874 
1875 	while (1) {
1876 		again = 0;
1877 
1878 		set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1879 
1880 		/* Make the cleaner go to sleep early. */
1881 		if (btrfs_need_cleaner_sleep(fs_info))
1882 			goto sleep;
1883 
1884 		/*
1885 		 * Do not do anything if we might cause open_ctree() to block
1886 		 * before we have finished mounting the filesystem.
1887 		 */
1888 		if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1889 			goto sleep;
1890 
1891 		if (!mutex_trylock(&fs_info->cleaner_mutex))
1892 			goto sleep;
1893 
1894 		/*
1895 		 * Avoid the problem that we change the status of the fs
1896 		 * during the above check and trylock.
1897 		 */
1898 		if (btrfs_need_cleaner_sleep(fs_info)) {
1899 			mutex_unlock(&fs_info->cleaner_mutex);
1900 			goto sleep;
1901 		}
1902 
1903 		btrfs_run_delayed_iputs(fs_info);
1904 
1905 		again = btrfs_clean_one_deleted_snapshot(fs_info);
1906 		mutex_unlock(&fs_info->cleaner_mutex);
1907 
1908 		/*
1909 		 * The defragger has dealt with the R/O remount and umount,
1910 		 * needn't do anything special here.
1911 		 */
1912 		btrfs_run_defrag_inodes(fs_info);
1913 
1914 		/*
1915 		 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1916 		 * with relocation (btrfs_relocate_chunk) and relocation
1917 		 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1918 		 * after acquiring fs_info->reclaim_bgs_lock. So we
1919 		 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1920 		 * unused block groups.
1921 		 */
1922 		btrfs_delete_unused_bgs(fs_info);
1923 
1924 		/*
1925 		 * Reclaim block groups in the reclaim_bgs list after we deleted
1926 		 * all unused block_groups. This possibly gives us some more free
1927 		 * space.
1928 		 */
1929 		btrfs_reclaim_bgs(fs_info);
1930 sleep:
1931 		clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1932 		if (kthread_should_park())
1933 			kthread_parkme();
1934 		if (kthread_should_stop())
1935 			return 0;
1936 		if (!again) {
1937 			set_current_state(TASK_INTERRUPTIBLE);
1938 			schedule();
1939 			__set_current_state(TASK_RUNNING);
1940 		}
1941 	}
1942 }
1943 
1944 static int transaction_kthread(void *arg)
1945 {
1946 	struct btrfs_root *root = arg;
1947 	struct btrfs_fs_info *fs_info = root->fs_info;
1948 	struct btrfs_trans_handle *trans;
1949 	struct btrfs_transaction *cur;
1950 	u64 transid;
1951 	time64_t delta;
1952 	unsigned long delay;
1953 	bool cannot_commit;
1954 
1955 	do {
1956 		cannot_commit = false;
1957 		delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1958 		mutex_lock(&fs_info->transaction_kthread_mutex);
1959 
1960 		spin_lock(&fs_info->trans_lock);
1961 		cur = fs_info->running_transaction;
1962 		if (!cur) {
1963 			spin_unlock(&fs_info->trans_lock);
1964 			goto sleep;
1965 		}
1966 
1967 		delta = ktime_get_seconds() - cur->start_time;
1968 		if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1969 		    cur->state < TRANS_STATE_COMMIT_START &&
1970 		    delta < fs_info->commit_interval) {
1971 			spin_unlock(&fs_info->trans_lock);
1972 			delay -= msecs_to_jiffies((delta - 1) * 1000);
1973 			delay = min(delay,
1974 				    msecs_to_jiffies(fs_info->commit_interval * 1000));
1975 			goto sleep;
1976 		}
1977 		transid = cur->transid;
1978 		spin_unlock(&fs_info->trans_lock);
1979 
1980 		/* If the file system is aborted, this will always fail. */
1981 		trans = btrfs_attach_transaction(root);
1982 		if (IS_ERR(trans)) {
1983 			if (PTR_ERR(trans) != -ENOENT)
1984 				cannot_commit = true;
1985 			goto sleep;
1986 		}
1987 		if (transid == trans->transid) {
1988 			btrfs_commit_transaction(trans);
1989 		} else {
1990 			btrfs_end_transaction(trans);
1991 		}
1992 sleep:
1993 		wake_up_process(fs_info->cleaner_kthread);
1994 		mutex_unlock(&fs_info->transaction_kthread_mutex);
1995 
1996 		if (BTRFS_FS_ERROR(fs_info))
1997 			btrfs_cleanup_transaction(fs_info);
1998 		if (!kthread_should_stop() &&
1999 				(!btrfs_transaction_blocked(fs_info) ||
2000 				 cannot_commit))
2001 			schedule_timeout_interruptible(delay);
2002 	} while (!kthread_should_stop());
2003 	return 0;
2004 }
2005 
2006 /*
2007  * This will find the highest generation in the array of root backups.  The
2008  * index of the highest array is returned, or -EINVAL if we can't find
2009  * anything.
2010  *
2011  * We check to make sure the array is valid by comparing the
2012  * generation of the latest  root in the array with the generation
2013  * in the super block.  If they don't match we pitch it.
2014  */
2015 static int find_newest_super_backup(struct btrfs_fs_info *info)
2016 {
2017 	const u64 newest_gen = btrfs_super_generation(info->super_copy);
2018 	u64 cur;
2019 	struct btrfs_root_backup *root_backup;
2020 	int i;
2021 
2022 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2023 		root_backup = info->super_copy->super_roots + i;
2024 		cur = btrfs_backup_tree_root_gen(root_backup);
2025 		if (cur == newest_gen)
2026 			return i;
2027 	}
2028 
2029 	return -EINVAL;
2030 }
2031 
2032 /*
2033  * copy all the root pointers into the super backup array.
2034  * this will bump the backup pointer by one when it is
2035  * done
2036  */
2037 static void backup_super_roots(struct btrfs_fs_info *info)
2038 {
2039 	const int next_backup = info->backup_root_index;
2040 	struct btrfs_root_backup *root_backup;
2041 
2042 	root_backup = info->super_for_commit->super_roots + next_backup;
2043 
2044 	/*
2045 	 * make sure all of our padding and empty slots get zero filled
2046 	 * regardless of which ones we use today
2047 	 */
2048 	memset(root_backup, 0, sizeof(*root_backup));
2049 
2050 	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2051 
2052 	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2053 	btrfs_set_backup_tree_root_gen(root_backup,
2054 			       btrfs_header_generation(info->tree_root->node));
2055 
2056 	btrfs_set_backup_tree_root_level(root_backup,
2057 			       btrfs_header_level(info->tree_root->node));
2058 
2059 	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2060 	btrfs_set_backup_chunk_root_gen(root_backup,
2061 			       btrfs_header_generation(info->chunk_root->node));
2062 	btrfs_set_backup_chunk_root_level(root_backup,
2063 			       btrfs_header_level(info->chunk_root->node));
2064 
2065 	if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
2066 		struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
2067 		struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
2068 
2069 		btrfs_set_backup_extent_root(root_backup,
2070 					     extent_root->node->start);
2071 		btrfs_set_backup_extent_root_gen(root_backup,
2072 				btrfs_header_generation(extent_root->node));
2073 		btrfs_set_backup_extent_root_level(root_backup,
2074 					btrfs_header_level(extent_root->node));
2075 
2076 		btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
2077 		btrfs_set_backup_csum_root_gen(root_backup,
2078 					       btrfs_header_generation(csum_root->node));
2079 		btrfs_set_backup_csum_root_level(root_backup,
2080 						 btrfs_header_level(csum_root->node));
2081 	}
2082 
2083 	/*
2084 	 * we might commit during log recovery, which happens before we set
2085 	 * the fs_root.  Make sure it is valid before we fill it in.
2086 	 */
2087 	if (info->fs_root && info->fs_root->node) {
2088 		btrfs_set_backup_fs_root(root_backup,
2089 					 info->fs_root->node->start);
2090 		btrfs_set_backup_fs_root_gen(root_backup,
2091 			       btrfs_header_generation(info->fs_root->node));
2092 		btrfs_set_backup_fs_root_level(root_backup,
2093 			       btrfs_header_level(info->fs_root->node));
2094 	}
2095 
2096 	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2097 	btrfs_set_backup_dev_root_gen(root_backup,
2098 			       btrfs_header_generation(info->dev_root->node));
2099 	btrfs_set_backup_dev_root_level(root_backup,
2100 				       btrfs_header_level(info->dev_root->node));
2101 
2102 	btrfs_set_backup_total_bytes(root_backup,
2103 			     btrfs_super_total_bytes(info->super_copy));
2104 	btrfs_set_backup_bytes_used(root_backup,
2105 			     btrfs_super_bytes_used(info->super_copy));
2106 	btrfs_set_backup_num_devices(root_backup,
2107 			     btrfs_super_num_devices(info->super_copy));
2108 
2109 	/*
2110 	 * if we don't copy this out to the super_copy, it won't get remembered
2111 	 * for the next commit
2112 	 */
2113 	memcpy(&info->super_copy->super_roots,
2114 	       &info->super_for_commit->super_roots,
2115 	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2116 }
2117 
2118 /*
2119  * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2120  * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2121  *
2122  * fs_info - filesystem whose backup roots need to be read
2123  * priority - priority of backup root required
2124  *
2125  * Returns backup root index on success and -EINVAL otherwise.
2126  */
2127 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2128 {
2129 	int backup_index = find_newest_super_backup(fs_info);
2130 	struct btrfs_super_block *super = fs_info->super_copy;
2131 	struct btrfs_root_backup *root_backup;
2132 
2133 	if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2134 		if (priority == 0)
2135 			return backup_index;
2136 
2137 		backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2138 		backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2139 	} else {
2140 		return -EINVAL;
2141 	}
2142 
2143 	root_backup = super->super_roots + backup_index;
2144 
2145 	btrfs_set_super_generation(super,
2146 				   btrfs_backup_tree_root_gen(root_backup));
2147 	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2148 	btrfs_set_super_root_level(super,
2149 				   btrfs_backup_tree_root_level(root_backup));
2150 	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2151 
2152 	/*
2153 	 * Fixme: the total bytes and num_devices need to match or we should
2154 	 * need a fsck
2155 	 */
2156 	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2157 	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2158 
2159 	return backup_index;
2160 }
2161 
2162 /* helper to cleanup workers */
2163 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2164 {
2165 	btrfs_destroy_workqueue(fs_info->fixup_workers);
2166 	btrfs_destroy_workqueue(fs_info->delalloc_workers);
2167 	btrfs_destroy_workqueue(fs_info->hipri_workers);
2168 	btrfs_destroy_workqueue(fs_info->workers);
2169 	if (fs_info->endio_workers)
2170 		destroy_workqueue(fs_info->endio_workers);
2171 	if (fs_info->rmw_workers)
2172 		destroy_workqueue(fs_info->rmw_workers);
2173 	if (fs_info->compressed_write_workers)
2174 		destroy_workqueue(fs_info->compressed_write_workers);
2175 	btrfs_destroy_workqueue(fs_info->endio_write_workers);
2176 	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2177 	btrfs_destroy_workqueue(fs_info->delayed_workers);
2178 	btrfs_destroy_workqueue(fs_info->caching_workers);
2179 	btrfs_destroy_workqueue(fs_info->flush_workers);
2180 	btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2181 	if (fs_info->discard_ctl.discard_workers)
2182 		destroy_workqueue(fs_info->discard_ctl.discard_workers);
2183 	/*
2184 	 * Now that all other work queues are destroyed, we can safely destroy
2185 	 * the queues used for metadata I/O, since tasks from those other work
2186 	 * queues can do metadata I/O operations.
2187 	 */
2188 	if (fs_info->endio_meta_workers)
2189 		destroy_workqueue(fs_info->endio_meta_workers);
2190 }
2191 
2192 static void free_root_extent_buffers(struct btrfs_root *root)
2193 {
2194 	if (root) {
2195 		free_extent_buffer(root->node);
2196 		free_extent_buffer(root->commit_root);
2197 		root->node = NULL;
2198 		root->commit_root = NULL;
2199 	}
2200 }
2201 
2202 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
2203 {
2204 	struct btrfs_root *root, *tmp;
2205 
2206 	rbtree_postorder_for_each_entry_safe(root, tmp,
2207 					     &fs_info->global_root_tree,
2208 					     rb_node)
2209 		free_root_extent_buffers(root);
2210 }
2211 
2212 /* helper to cleanup tree roots */
2213 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2214 {
2215 	free_root_extent_buffers(info->tree_root);
2216 
2217 	free_global_root_pointers(info);
2218 	free_root_extent_buffers(info->dev_root);
2219 	free_root_extent_buffers(info->quota_root);
2220 	free_root_extent_buffers(info->uuid_root);
2221 	free_root_extent_buffers(info->fs_root);
2222 	free_root_extent_buffers(info->data_reloc_root);
2223 	free_root_extent_buffers(info->block_group_root);
2224 	if (free_chunk_root)
2225 		free_root_extent_buffers(info->chunk_root);
2226 }
2227 
2228 void btrfs_put_root(struct btrfs_root *root)
2229 {
2230 	if (!root)
2231 		return;
2232 
2233 	if (refcount_dec_and_test(&root->refs)) {
2234 		WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2235 		WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2236 		if (root->anon_dev)
2237 			free_anon_bdev(root->anon_dev);
2238 		btrfs_drew_lock_destroy(&root->snapshot_lock);
2239 		free_root_extent_buffers(root);
2240 #ifdef CONFIG_BTRFS_DEBUG
2241 		spin_lock(&root->fs_info->fs_roots_radix_lock);
2242 		list_del_init(&root->leak_list);
2243 		spin_unlock(&root->fs_info->fs_roots_radix_lock);
2244 #endif
2245 		kfree(root);
2246 	}
2247 }
2248 
2249 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2250 {
2251 	int ret;
2252 	struct btrfs_root *gang[8];
2253 	int i;
2254 
2255 	while (!list_empty(&fs_info->dead_roots)) {
2256 		gang[0] = list_entry(fs_info->dead_roots.next,
2257 				     struct btrfs_root, root_list);
2258 		list_del(&gang[0]->root_list);
2259 
2260 		if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2261 			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2262 		btrfs_put_root(gang[0]);
2263 	}
2264 
2265 	while (1) {
2266 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2267 					     (void **)gang, 0,
2268 					     ARRAY_SIZE(gang));
2269 		if (!ret)
2270 			break;
2271 		for (i = 0; i < ret; i++)
2272 			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2273 	}
2274 }
2275 
2276 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2277 {
2278 	mutex_init(&fs_info->scrub_lock);
2279 	atomic_set(&fs_info->scrubs_running, 0);
2280 	atomic_set(&fs_info->scrub_pause_req, 0);
2281 	atomic_set(&fs_info->scrubs_paused, 0);
2282 	atomic_set(&fs_info->scrub_cancel_req, 0);
2283 	init_waitqueue_head(&fs_info->scrub_pause_wait);
2284 	refcount_set(&fs_info->scrub_workers_refcnt, 0);
2285 }
2286 
2287 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2288 {
2289 	spin_lock_init(&fs_info->balance_lock);
2290 	mutex_init(&fs_info->balance_mutex);
2291 	atomic_set(&fs_info->balance_pause_req, 0);
2292 	atomic_set(&fs_info->balance_cancel_req, 0);
2293 	fs_info->balance_ctl = NULL;
2294 	init_waitqueue_head(&fs_info->balance_wait_q);
2295 	atomic_set(&fs_info->reloc_cancel_req, 0);
2296 }
2297 
2298 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2299 {
2300 	struct inode *inode = fs_info->btree_inode;
2301 	unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
2302 					      fs_info->tree_root);
2303 
2304 	inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2305 	set_nlink(inode, 1);
2306 	/*
2307 	 * we set the i_size on the btree inode to the max possible int.
2308 	 * the real end of the address space is determined by all of
2309 	 * the devices in the system
2310 	 */
2311 	inode->i_size = OFFSET_MAX;
2312 	inode->i_mapping->a_ops = &btree_aops;
2313 
2314 	RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2315 	extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2316 			    IO_TREE_BTREE_INODE_IO);
2317 	extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2318 
2319 	BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2320 	BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
2321 	BTRFS_I(inode)->location.type = 0;
2322 	BTRFS_I(inode)->location.offset = 0;
2323 	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2324 	__insert_inode_hash(inode, hash);
2325 }
2326 
2327 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2328 {
2329 	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2330 	init_rwsem(&fs_info->dev_replace.rwsem);
2331 	init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2332 }
2333 
2334 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2335 {
2336 	spin_lock_init(&fs_info->qgroup_lock);
2337 	mutex_init(&fs_info->qgroup_ioctl_lock);
2338 	fs_info->qgroup_tree = RB_ROOT;
2339 	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2340 	fs_info->qgroup_seq = 1;
2341 	fs_info->qgroup_ulist = NULL;
2342 	fs_info->qgroup_rescan_running = false;
2343 	fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
2344 	mutex_init(&fs_info->qgroup_rescan_lock);
2345 }
2346 
2347 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
2348 {
2349 	u32 max_active = fs_info->thread_pool_size;
2350 	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2351 
2352 	fs_info->workers =
2353 		btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
2354 	fs_info->hipri_workers =
2355 		btrfs_alloc_workqueue(fs_info, "worker-high",
2356 				      flags | WQ_HIGHPRI, max_active, 16);
2357 
2358 	fs_info->delalloc_workers =
2359 		btrfs_alloc_workqueue(fs_info, "delalloc",
2360 				      flags, max_active, 2);
2361 
2362 	fs_info->flush_workers =
2363 		btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2364 				      flags, max_active, 0);
2365 
2366 	fs_info->caching_workers =
2367 		btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2368 
2369 	fs_info->fixup_workers =
2370 		btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2371 
2372 	fs_info->endio_workers =
2373 		alloc_workqueue("btrfs-endio", flags, max_active);
2374 	fs_info->endio_meta_workers =
2375 		alloc_workqueue("btrfs-endio-meta", flags, max_active);
2376 	fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
2377 	fs_info->endio_write_workers =
2378 		btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2379 				      max_active, 2);
2380 	fs_info->compressed_write_workers =
2381 		alloc_workqueue("btrfs-compressed-write", flags, max_active);
2382 	fs_info->endio_freespace_worker =
2383 		btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2384 				      max_active, 0);
2385 	fs_info->delayed_workers =
2386 		btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2387 				      max_active, 0);
2388 	fs_info->qgroup_rescan_workers =
2389 		btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2390 	fs_info->discard_ctl.discard_workers =
2391 		alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2392 
2393 	if (!(fs_info->workers && fs_info->hipri_workers &&
2394 	      fs_info->delalloc_workers && fs_info->flush_workers &&
2395 	      fs_info->endio_workers && fs_info->endio_meta_workers &&
2396 	      fs_info->compressed_write_workers &&
2397 	      fs_info->endio_write_workers &&
2398 	      fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2399 	      fs_info->caching_workers && fs_info->fixup_workers &&
2400 	      fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2401 	      fs_info->discard_ctl.discard_workers)) {
2402 		return -ENOMEM;
2403 	}
2404 
2405 	return 0;
2406 }
2407 
2408 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2409 {
2410 	struct crypto_shash *csum_shash;
2411 	const char *csum_driver = btrfs_super_csum_driver(csum_type);
2412 
2413 	csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2414 
2415 	if (IS_ERR(csum_shash)) {
2416 		btrfs_err(fs_info, "error allocating %s hash for checksum",
2417 			  csum_driver);
2418 		return PTR_ERR(csum_shash);
2419 	}
2420 
2421 	fs_info->csum_shash = csum_shash;
2422 
2423 	btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2424 			btrfs_super_csum_name(csum_type),
2425 			crypto_shash_driver_name(csum_shash));
2426 	return 0;
2427 }
2428 
2429 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2430 			    struct btrfs_fs_devices *fs_devices)
2431 {
2432 	int ret;
2433 	struct btrfs_tree_parent_check check = { 0 };
2434 	struct btrfs_root *log_tree_root;
2435 	struct btrfs_super_block *disk_super = fs_info->super_copy;
2436 	u64 bytenr = btrfs_super_log_root(disk_super);
2437 	int level = btrfs_super_log_root_level(disk_super);
2438 
2439 	if (fs_devices->rw_devices == 0) {
2440 		btrfs_warn(fs_info, "log replay required on RO media");
2441 		return -EIO;
2442 	}
2443 
2444 	log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2445 					 GFP_KERNEL);
2446 	if (!log_tree_root)
2447 		return -ENOMEM;
2448 
2449 	check.level = level;
2450 	check.transid = fs_info->generation + 1;
2451 	check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2452 	log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2453 	if (IS_ERR(log_tree_root->node)) {
2454 		btrfs_warn(fs_info, "failed to read log tree");
2455 		ret = PTR_ERR(log_tree_root->node);
2456 		log_tree_root->node = NULL;
2457 		btrfs_put_root(log_tree_root);
2458 		return ret;
2459 	}
2460 	if (!extent_buffer_uptodate(log_tree_root->node)) {
2461 		btrfs_err(fs_info, "failed to read log tree");
2462 		btrfs_put_root(log_tree_root);
2463 		return -EIO;
2464 	}
2465 
2466 	/* returns with log_tree_root freed on success */
2467 	ret = btrfs_recover_log_trees(log_tree_root);
2468 	if (ret) {
2469 		btrfs_handle_fs_error(fs_info, ret,
2470 				      "Failed to recover log tree");
2471 		btrfs_put_root(log_tree_root);
2472 		return ret;
2473 	}
2474 
2475 	if (sb_rdonly(fs_info->sb)) {
2476 		ret = btrfs_commit_super(fs_info);
2477 		if (ret)
2478 			return ret;
2479 	}
2480 
2481 	return 0;
2482 }
2483 
2484 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2485 				      struct btrfs_path *path, u64 objectid,
2486 				      const char *name)
2487 {
2488 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
2489 	struct btrfs_root *root;
2490 	u64 max_global_id = 0;
2491 	int ret;
2492 	struct btrfs_key key = {
2493 		.objectid = objectid,
2494 		.type = BTRFS_ROOT_ITEM_KEY,
2495 		.offset = 0,
2496 	};
2497 	bool found = false;
2498 
2499 	/* If we have IGNOREDATACSUMS skip loading these roots. */
2500 	if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2501 	    btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2502 		set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2503 		return 0;
2504 	}
2505 
2506 	while (1) {
2507 		ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2508 		if (ret < 0)
2509 			break;
2510 
2511 		if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2512 			ret = btrfs_next_leaf(tree_root, path);
2513 			if (ret) {
2514 				if (ret > 0)
2515 					ret = 0;
2516 				break;
2517 			}
2518 		}
2519 		ret = 0;
2520 
2521 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2522 		if (key.objectid != objectid)
2523 			break;
2524 		btrfs_release_path(path);
2525 
2526 		/*
2527 		 * Just worry about this for extent tree, it'll be the same for
2528 		 * everybody.
2529 		 */
2530 		if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2531 			max_global_id = max(max_global_id, key.offset);
2532 
2533 		found = true;
2534 		root = read_tree_root_path(tree_root, path, &key);
2535 		if (IS_ERR(root)) {
2536 			if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2537 				ret = PTR_ERR(root);
2538 			break;
2539 		}
2540 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2541 		ret = btrfs_global_root_insert(root);
2542 		if (ret) {
2543 			btrfs_put_root(root);
2544 			break;
2545 		}
2546 		key.offset++;
2547 	}
2548 	btrfs_release_path(path);
2549 
2550 	if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2551 		fs_info->nr_global_roots = max_global_id + 1;
2552 
2553 	if (!found || ret) {
2554 		if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2555 			set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2556 
2557 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2558 			ret = ret ? ret : -ENOENT;
2559 		else
2560 			ret = 0;
2561 		btrfs_err(fs_info, "failed to load root %s", name);
2562 	}
2563 	return ret;
2564 }
2565 
2566 static int load_global_roots(struct btrfs_root *tree_root)
2567 {
2568 	struct btrfs_path *path;
2569 	int ret = 0;
2570 
2571 	path = btrfs_alloc_path();
2572 	if (!path)
2573 		return -ENOMEM;
2574 
2575 	ret = load_global_roots_objectid(tree_root, path,
2576 					 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2577 	if (ret)
2578 		goto out;
2579 	ret = load_global_roots_objectid(tree_root, path,
2580 					 BTRFS_CSUM_TREE_OBJECTID, "csum");
2581 	if (ret)
2582 		goto out;
2583 	if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2584 		goto out;
2585 	ret = load_global_roots_objectid(tree_root, path,
2586 					 BTRFS_FREE_SPACE_TREE_OBJECTID,
2587 					 "free space");
2588 out:
2589 	btrfs_free_path(path);
2590 	return ret;
2591 }
2592 
2593 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2594 {
2595 	struct btrfs_root *tree_root = fs_info->tree_root;
2596 	struct btrfs_root *root;
2597 	struct btrfs_key location;
2598 	int ret;
2599 
2600 	BUG_ON(!fs_info->tree_root);
2601 
2602 	ret = load_global_roots(tree_root);
2603 	if (ret)
2604 		return ret;
2605 
2606 	location.type = BTRFS_ROOT_ITEM_KEY;
2607 	location.offset = 0;
2608 
2609 	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2610 		location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2611 		root = btrfs_read_tree_root(tree_root, &location);
2612 		if (IS_ERR(root)) {
2613 			if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2614 				ret = PTR_ERR(root);
2615 				goto out;
2616 			}
2617 		} else {
2618 			set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2619 			fs_info->block_group_root = root;
2620 		}
2621 	}
2622 
2623 	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2624 	root = btrfs_read_tree_root(tree_root, &location);
2625 	if (IS_ERR(root)) {
2626 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2627 			ret = PTR_ERR(root);
2628 			goto out;
2629 		}
2630 	} else {
2631 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2632 		fs_info->dev_root = root;
2633 	}
2634 	/* Initialize fs_info for all devices in any case */
2635 	ret = btrfs_init_devices_late(fs_info);
2636 	if (ret)
2637 		goto out;
2638 
2639 	/*
2640 	 * This tree can share blocks with some other fs tree during relocation
2641 	 * and we need a proper setup by btrfs_get_fs_root
2642 	 */
2643 	root = btrfs_get_fs_root(tree_root->fs_info,
2644 				 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2645 	if (IS_ERR(root)) {
2646 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2647 			ret = PTR_ERR(root);
2648 			goto out;
2649 		}
2650 	} else {
2651 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2652 		fs_info->data_reloc_root = root;
2653 	}
2654 
2655 	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2656 	root = btrfs_read_tree_root(tree_root, &location);
2657 	if (!IS_ERR(root)) {
2658 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2659 		set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2660 		fs_info->quota_root = root;
2661 	}
2662 
2663 	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2664 	root = btrfs_read_tree_root(tree_root, &location);
2665 	if (IS_ERR(root)) {
2666 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2667 			ret = PTR_ERR(root);
2668 			if (ret != -ENOENT)
2669 				goto out;
2670 		}
2671 	} else {
2672 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2673 		fs_info->uuid_root = root;
2674 	}
2675 
2676 	return 0;
2677 out:
2678 	btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2679 		   location.objectid, ret);
2680 	return ret;
2681 }
2682 
2683 /*
2684  * Real super block validation
2685  * NOTE: super csum type and incompat features will not be checked here.
2686  *
2687  * @sb:		super block to check
2688  * @mirror_num:	the super block number to check its bytenr:
2689  * 		0	the primary (1st) sb
2690  * 		1, 2	2nd and 3rd backup copy
2691  * 	       -1	skip bytenr check
2692  */
2693 int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2694 			 struct btrfs_super_block *sb, int mirror_num)
2695 {
2696 	u64 nodesize = btrfs_super_nodesize(sb);
2697 	u64 sectorsize = btrfs_super_sectorsize(sb);
2698 	int ret = 0;
2699 
2700 	if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2701 		btrfs_err(fs_info, "no valid FS found");
2702 		ret = -EINVAL;
2703 	}
2704 	if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2705 		btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2706 				btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2707 		ret = -EINVAL;
2708 	}
2709 	if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2710 		btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2711 				btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2712 		ret = -EINVAL;
2713 	}
2714 	if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2715 		btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2716 				btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2717 		ret = -EINVAL;
2718 	}
2719 	if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2720 		btrfs_err(fs_info, "log_root level too big: %d >= %d",
2721 				btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2722 		ret = -EINVAL;
2723 	}
2724 
2725 	/*
2726 	 * Check sectorsize and nodesize first, other check will need it.
2727 	 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2728 	 */
2729 	if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2730 	    sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2731 		btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2732 		ret = -EINVAL;
2733 	}
2734 
2735 	/*
2736 	 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2737 	 *
2738 	 * We can support 16K sectorsize with 64K page size without problem,
2739 	 * but such sectorsize/pagesize combination doesn't make much sense.
2740 	 * 4K will be our future standard, PAGE_SIZE is supported from the very
2741 	 * beginning.
2742 	 */
2743 	if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2744 		btrfs_err(fs_info,
2745 			"sectorsize %llu not yet supported for page size %lu",
2746 			sectorsize, PAGE_SIZE);
2747 		ret = -EINVAL;
2748 	}
2749 
2750 	if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2751 	    nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2752 		btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2753 		ret = -EINVAL;
2754 	}
2755 	if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2756 		btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2757 			  le32_to_cpu(sb->__unused_leafsize), nodesize);
2758 		ret = -EINVAL;
2759 	}
2760 
2761 	/* Root alignment check */
2762 	if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2763 		btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2764 			   btrfs_super_root(sb));
2765 		ret = -EINVAL;
2766 	}
2767 	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2768 		btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2769 			   btrfs_super_chunk_root(sb));
2770 		ret = -EINVAL;
2771 	}
2772 	if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2773 		btrfs_warn(fs_info, "log_root block unaligned: %llu",
2774 			   btrfs_super_log_root(sb));
2775 		ret = -EINVAL;
2776 	}
2777 
2778 	if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2779 		   BTRFS_FSID_SIZE)) {
2780 		btrfs_err(fs_info,
2781 		"superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2782 			fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2783 		ret = -EINVAL;
2784 	}
2785 
2786 	if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2787 	    memcmp(fs_info->fs_devices->metadata_uuid,
2788 		   fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2789 		btrfs_err(fs_info,
2790 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2791 			fs_info->super_copy->metadata_uuid,
2792 			fs_info->fs_devices->metadata_uuid);
2793 		ret = -EINVAL;
2794 	}
2795 
2796 	/*
2797 	 * Artificial requirement for block-group-tree to force newer features
2798 	 * (free-space-tree, no-holes) so the test matrix is smaller.
2799 	 */
2800 	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2801 	    (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2802 	     !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2803 		btrfs_err(fs_info,
2804 		"block-group-tree feature requires fres-space-tree and no-holes");
2805 		ret = -EINVAL;
2806 	}
2807 
2808 	if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2809 		   BTRFS_FSID_SIZE) != 0) {
2810 		btrfs_err(fs_info,
2811 			"dev_item UUID does not match metadata fsid: %pU != %pU",
2812 			fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2813 		ret = -EINVAL;
2814 	}
2815 
2816 	/*
2817 	 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2818 	 * done later
2819 	 */
2820 	if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2821 		btrfs_err(fs_info, "bytes_used is too small %llu",
2822 			  btrfs_super_bytes_used(sb));
2823 		ret = -EINVAL;
2824 	}
2825 	if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2826 		btrfs_err(fs_info, "invalid stripesize %u",
2827 			  btrfs_super_stripesize(sb));
2828 		ret = -EINVAL;
2829 	}
2830 	if (btrfs_super_num_devices(sb) > (1UL << 31))
2831 		btrfs_warn(fs_info, "suspicious number of devices: %llu",
2832 			   btrfs_super_num_devices(sb));
2833 	if (btrfs_super_num_devices(sb) == 0) {
2834 		btrfs_err(fs_info, "number of devices is 0");
2835 		ret = -EINVAL;
2836 	}
2837 
2838 	if (mirror_num >= 0 &&
2839 	    btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2840 		btrfs_err(fs_info, "super offset mismatch %llu != %u",
2841 			  btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2842 		ret = -EINVAL;
2843 	}
2844 
2845 	/*
2846 	 * Obvious sys_chunk_array corruptions, it must hold at least one key
2847 	 * and one chunk
2848 	 */
2849 	if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2850 		btrfs_err(fs_info, "system chunk array too big %u > %u",
2851 			  btrfs_super_sys_array_size(sb),
2852 			  BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2853 		ret = -EINVAL;
2854 	}
2855 	if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2856 			+ sizeof(struct btrfs_chunk)) {
2857 		btrfs_err(fs_info, "system chunk array too small %u < %zu",
2858 			  btrfs_super_sys_array_size(sb),
2859 			  sizeof(struct btrfs_disk_key)
2860 			  + sizeof(struct btrfs_chunk));
2861 		ret = -EINVAL;
2862 	}
2863 
2864 	/*
2865 	 * The generation is a global counter, we'll trust it more than the others
2866 	 * but it's still possible that it's the one that's wrong.
2867 	 */
2868 	if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2869 		btrfs_warn(fs_info,
2870 			"suspicious: generation < chunk_root_generation: %llu < %llu",
2871 			btrfs_super_generation(sb),
2872 			btrfs_super_chunk_root_generation(sb));
2873 	if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2874 	    && btrfs_super_cache_generation(sb) != (u64)-1)
2875 		btrfs_warn(fs_info,
2876 			"suspicious: generation < cache_generation: %llu < %llu",
2877 			btrfs_super_generation(sb),
2878 			btrfs_super_cache_generation(sb));
2879 
2880 	return ret;
2881 }
2882 
2883 /*
2884  * Validation of super block at mount time.
2885  * Some checks already done early at mount time, like csum type and incompat
2886  * flags will be skipped.
2887  */
2888 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2889 {
2890 	return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2891 }
2892 
2893 /*
2894  * Validation of super block at write time.
2895  * Some checks like bytenr check will be skipped as their values will be
2896  * overwritten soon.
2897  * Extra checks like csum type and incompat flags will be done here.
2898  */
2899 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2900 				      struct btrfs_super_block *sb)
2901 {
2902 	int ret;
2903 
2904 	ret = btrfs_validate_super(fs_info, sb, -1);
2905 	if (ret < 0)
2906 		goto out;
2907 	if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2908 		ret = -EUCLEAN;
2909 		btrfs_err(fs_info, "invalid csum type, has %u want %u",
2910 			  btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2911 		goto out;
2912 	}
2913 	if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2914 		ret = -EUCLEAN;
2915 		btrfs_err(fs_info,
2916 		"invalid incompat flags, has 0x%llx valid mask 0x%llx",
2917 			  btrfs_super_incompat_flags(sb),
2918 			  (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2919 		goto out;
2920 	}
2921 out:
2922 	if (ret < 0)
2923 		btrfs_err(fs_info,
2924 		"super block corruption detected before writing it to disk");
2925 	return ret;
2926 }
2927 
2928 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2929 {
2930 	struct btrfs_tree_parent_check check = {
2931 		.level = level,
2932 		.transid = gen,
2933 		.owner_root = root->root_key.objectid
2934 	};
2935 	int ret = 0;
2936 
2937 	root->node = read_tree_block(root->fs_info, bytenr, &check);
2938 	if (IS_ERR(root->node)) {
2939 		ret = PTR_ERR(root->node);
2940 		root->node = NULL;
2941 		return ret;
2942 	}
2943 	if (!extent_buffer_uptodate(root->node)) {
2944 		free_extent_buffer(root->node);
2945 		root->node = NULL;
2946 		return -EIO;
2947 	}
2948 
2949 	btrfs_set_root_node(&root->root_item, root->node);
2950 	root->commit_root = btrfs_root_node(root);
2951 	btrfs_set_root_refs(&root->root_item, 1);
2952 	return ret;
2953 }
2954 
2955 static int load_important_roots(struct btrfs_fs_info *fs_info)
2956 {
2957 	struct btrfs_super_block *sb = fs_info->super_copy;
2958 	u64 gen, bytenr;
2959 	int level, ret;
2960 
2961 	bytenr = btrfs_super_root(sb);
2962 	gen = btrfs_super_generation(sb);
2963 	level = btrfs_super_root_level(sb);
2964 	ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2965 	if (ret) {
2966 		btrfs_warn(fs_info, "couldn't read tree root");
2967 		return ret;
2968 	}
2969 	return 0;
2970 }
2971 
2972 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2973 {
2974 	int backup_index = find_newest_super_backup(fs_info);
2975 	struct btrfs_super_block *sb = fs_info->super_copy;
2976 	struct btrfs_root *tree_root = fs_info->tree_root;
2977 	bool handle_error = false;
2978 	int ret = 0;
2979 	int i;
2980 
2981 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2982 		if (handle_error) {
2983 			if (!IS_ERR(tree_root->node))
2984 				free_extent_buffer(tree_root->node);
2985 			tree_root->node = NULL;
2986 
2987 			if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2988 				break;
2989 
2990 			free_root_pointers(fs_info, 0);
2991 
2992 			/*
2993 			 * Don't use the log in recovery mode, it won't be
2994 			 * valid
2995 			 */
2996 			btrfs_set_super_log_root(sb, 0);
2997 
2998 			/* We can't trust the free space cache either */
2999 			btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3000 
3001 			ret = read_backup_root(fs_info, i);
3002 			backup_index = ret;
3003 			if (ret < 0)
3004 				return ret;
3005 		}
3006 
3007 		ret = load_important_roots(fs_info);
3008 		if (ret) {
3009 			handle_error = true;
3010 			continue;
3011 		}
3012 
3013 		/*
3014 		 * No need to hold btrfs_root::objectid_mutex since the fs
3015 		 * hasn't been fully initialised and we are the only user
3016 		 */
3017 		ret = btrfs_init_root_free_objectid(tree_root);
3018 		if (ret < 0) {
3019 			handle_error = true;
3020 			continue;
3021 		}
3022 
3023 		ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
3024 
3025 		ret = btrfs_read_roots(fs_info);
3026 		if (ret < 0) {
3027 			handle_error = true;
3028 			continue;
3029 		}
3030 
3031 		/* All successful */
3032 		fs_info->generation = btrfs_header_generation(tree_root->node);
3033 		fs_info->last_trans_committed = fs_info->generation;
3034 		fs_info->last_reloc_trans = 0;
3035 
3036 		/* Always begin writing backup roots after the one being used */
3037 		if (backup_index < 0) {
3038 			fs_info->backup_root_index = 0;
3039 		} else {
3040 			fs_info->backup_root_index = backup_index + 1;
3041 			fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
3042 		}
3043 		break;
3044 	}
3045 
3046 	return ret;
3047 }
3048 
3049 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
3050 {
3051 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
3052 	INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
3053 	INIT_LIST_HEAD(&fs_info->trans_list);
3054 	INIT_LIST_HEAD(&fs_info->dead_roots);
3055 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
3056 	INIT_LIST_HEAD(&fs_info->delalloc_roots);
3057 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
3058 	spin_lock_init(&fs_info->delalloc_root_lock);
3059 	spin_lock_init(&fs_info->trans_lock);
3060 	spin_lock_init(&fs_info->fs_roots_radix_lock);
3061 	spin_lock_init(&fs_info->delayed_iput_lock);
3062 	spin_lock_init(&fs_info->defrag_inodes_lock);
3063 	spin_lock_init(&fs_info->super_lock);
3064 	spin_lock_init(&fs_info->buffer_lock);
3065 	spin_lock_init(&fs_info->unused_bgs_lock);
3066 	spin_lock_init(&fs_info->treelog_bg_lock);
3067 	spin_lock_init(&fs_info->zone_active_bgs_lock);
3068 	spin_lock_init(&fs_info->relocation_bg_lock);
3069 	rwlock_init(&fs_info->tree_mod_log_lock);
3070 	rwlock_init(&fs_info->global_root_lock);
3071 	mutex_init(&fs_info->unused_bg_unpin_mutex);
3072 	mutex_init(&fs_info->reclaim_bgs_lock);
3073 	mutex_init(&fs_info->reloc_mutex);
3074 	mutex_init(&fs_info->delalloc_root_mutex);
3075 	mutex_init(&fs_info->zoned_meta_io_lock);
3076 	mutex_init(&fs_info->zoned_data_reloc_io_lock);
3077 	seqlock_init(&fs_info->profiles_lock);
3078 
3079 	btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
3080 	btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
3081 	btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
3082 	btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
3083 	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_start,
3084 				     BTRFS_LOCKDEP_TRANS_COMMIT_START);
3085 	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
3086 				     BTRFS_LOCKDEP_TRANS_UNBLOCKED);
3087 	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
3088 				     BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
3089 	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
3090 				     BTRFS_LOCKDEP_TRANS_COMPLETED);
3091 
3092 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
3093 	INIT_LIST_HEAD(&fs_info->space_info);
3094 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
3095 	INIT_LIST_HEAD(&fs_info->unused_bgs);
3096 	INIT_LIST_HEAD(&fs_info->reclaim_bgs);
3097 	INIT_LIST_HEAD(&fs_info->zone_active_bgs);
3098 #ifdef CONFIG_BTRFS_DEBUG
3099 	INIT_LIST_HEAD(&fs_info->allocated_roots);
3100 	INIT_LIST_HEAD(&fs_info->allocated_ebs);
3101 	spin_lock_init(&fs_info->eb_leak_lock);
3102 #endif
3103 	extent_map_tree_init(&fs_info->mapping_tree);
3104 	btrfs_init_block_rsv(&fs_info->global_block_rsv,
3105 			     BTRFS_BLOCK_RSV_GLOBAL);
3106 	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
3107 	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
3108 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
3109 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
3110 			     BTRFS_BLOCK_RSV_DELOPS);
3111 	btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
3112 			     BTRFS_BLOCK_RSV_DELREFS);
3113 
3114 	atomic_set(&fs_info->async_delalloc_pages, 0);
3115 	atomic_set(&fs_info->defrag_running, 0);
3116 	atomic_set(&fs_info->nr_delayed_iputs, 0);
3117 	atomic64_set(&fs_info->tree_mod_seq, 0);
3118 	fs_info->global_root_tree = RB_ROOT;
3119 	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
3120 	fs_info->metadata_ratio = 0;
3121 	fs_info->defrag_inodes = RB_ROOT;
3122 	atomic64_set(&fs_info->free_chunk_space, 0);
3123 	fs_info->tree_mod_log = RB_ROOT;
3124 	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
3125 	fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
3126 	btrfs_init_ref_verify(fs_info);
3127 
3128 	fs_info->thread_pool_size = min_t(unsigned long,
3129 					  num_online_cpus() + 2, 8);
3130 
3131 	INIT_LIST_HEAD(&fs_info->ordered_roots);
3132 	spin_lock_init(&fs_info->ordered_root_lock);
3133 
3134 	btrfs_init_scrub(fs_info);
3135 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3136 	fs_info->check_integrity_print_mask = 0;
3137 #endif
3138 	btrfs_init_balance(fs_info);
3139 	btrfs_init_async_reclaim_work(fs_info);
3140 
3141 	rwlock_init(&fs_info->block_group_cache_lock);
3142 	fs_info->block_group_cache_tree = RB_ROOT_CACHED;
3143 
3144 	extent_io_tree_init(fs_info, &fs_info->excluded_extents,
3145 			    IO_TREE_FS_EXCLUDED_EXTENTS);
3146 
3147 	mutex_init(&fs_info->ordered_operations_mutex);
3148 	mutex_init(&fs_info->tree_log_mutex);
3149 	mutex_init(&fs_info->chunk_mutex);
3150 	mutex_init(&fs_info->transaction_kthread_mutex);
3151 	mutex_init(&fs_info->cleaner_mutex);
3152 	mutex_init(&fs_info->ro_block_group_mutex);
3153 	init_rwsem(&fs_info->commit_root_sem);
3154 	init_rwsem(&fs_info->cleanup_work_sem);
3155 	init_rwsem(&fs_info->subvol_sem);
3156 	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
3157 
3158 	btrfs_init_dev_replace_locks(fs_info);
3159 	btrfs_init_qgroup(fs_info);
3160 	btrfs_discard_init(fs_info);
3161 
3162 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
3163 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
3164 
3165 	init_waitqueue_head(&fs_info->transaction_throttle);
3166 	init_waitqueue_head(&fs_info->transaction_wait);
3167 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
3168 	init_waitqueue_head(&fs_info->async_submit_wait);
3169 	init_waitqueue_head(&fs_info->delayed_iputs_wait);
3170 
3171 	/* Usable values until the real ones are cached from the superblock */
3172 	fs_info->nodesize = 4096;
3173 	fs_info->sectorsize = 4096;
3174 	fs_info->sectorsize_bits = ilog2(4096);
3175 	fs_info->stripesize = 4096;
3176 
3177 	fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
3178 
3179 	spin_lock_init(&fs_info->swapfile_pins_lock);
3180 	fs_info->swapfile_pins = RB_ROOT;
3181 
3182 	fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
3183 	INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
3184 }
3185 
3186 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
3187 {
3188 	int ret;
3189 
3190 	fs_info->sb = sb;
3191 	sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
3192 	sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3193 
3194 	ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3195 	if (ret)
3196 		return ret;
3197 
3198 	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3199 	if (ret)
3200 		return ret;
3201 
3202 	fs_info->dirty_metadata_batch = PAGE_SIZE *
3203 					(1 + ilog2(nr_cpu_ids));
3204 
3205 	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3206 	if (ret)
3207 		return ret;
3208 
3209 	ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3210 			GFP_KERNEL);
3211 	if (ret)
3212 		return ret;
3213 
3214 	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3215 					GFP_KERNEL);
3216 	if (!fs_info->delayed_root)
3217 		return -ENOMEM;
3218 	btrfs_init_delayed_root(fs_info->delayed_root);
3219 
3220 	if (sb_rdonly(sb))
3221 		set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3222 
3223 	return btrfs_alloc_stripe_hash_table(fs_info);
3224 }
3225 
3226 static int btrfs_uuid_rescan_kthread(void *data)
3227 {
3228 	struct btrfs_fs_info *fs_info = data;
3229 	int ret;
3230 
3231 	/*
3232 	 * 1st step is to iterate through the existing UUID tree and
3233 	 * to delete all entries that contain outdated data.
3234 	 * 2nd step is to add all missing entries to the UUID tree.
3235 	 */
3236 	ret = btrfs_uuid_tree_iterate(fs_info);
3237 	if (ret < 0) {
3238 		if (ret != -EINTR)
3239 			btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3240 				   ret);
3241 		up(&fs_info->uuid_tree_rescan_sem);
3242 		return ret;
3243 	}
3244 	return btrfs_uuid_scan_kthread(data);
3245 }
3246 
3247 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3248 {
3249 	struct task_struct *task;
3250 
3251 	down(&fs_info->uuid_tree_rescan_sem);
3252 	task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3253 	if (IS_ERR(task)) {
3254 		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
3255 		btrfs_warn(fs_info, "failed to start uuid_rescan task");
3256 		up(&fs_info->uuid_tree_rescan_sem);
3257 		return PTR_ERR(task);
3258 	}
3259 
3260 	return 0;
3261 }
3262 
3263 /*
3264  * Some options only have meaning at mount time and shouldn't persist across
3265  * remounts, or be displayed. Clear these at the end of mount and remount
3266  * code paths.
3267  */
3268 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3269 {
3270 	btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3271 	btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3272 }
3273 
3274 /*
3275  * Mounting logic specific to read-write file systems. Shared by open_ctree
3276  * and btrfs_remount when remounting from read-only to read-write.
3277  */
3278 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3279 {
3280 	int ret;
3281 	const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3282 	bool clear_free_space_tree = false;
3283 
3284 	if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3285 	    btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3286 		clear_free_space_tree = true;
3287 	} else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3288 		   !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3289 		btrfs_warn(fs_info, "free space tree is invalid");
3290 		clear_free_space_tree = true;
3291 	}
3292 
3293 	if (clear_free_space_tree) {
3294 		btrfs_info(fs_info, "clearing free space tree");
3295 		ret = btrfs_clear_free_space_tree(fs_info);
3296 		if (ret) {
3297 			btrfs_warn(fs_info,
3298 				   "failed to clear free space tree: %d", ret);
3299 			goto out;
3300 		}
3301 	}
3302 
3303 	/*
3304 	 * btrfs_find_orphan_roots() is responsible for finding all the dead
3305 	 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3306 	 * them into the fs_info->fs_roots_radix tree. This must be done before
3307 	 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3308 	 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3309 	 * item before the root's tree is deleted - this means that if we unmount
3310 	 * or crash before the deletion completes, on the next mount we will not
3311 	 * delete what remains of the tree because the orphan item does not
3312 	 * exists anymore, which is what tells us we have a pending deletion.
3313 	 */
3314 	ret = btrfs_find_orphan_roots(fs_info);
3315 	if (ret)
3316 		goto out;
3317 
3318 	ret = btrfs_cleanup_fs_roots(fs_info);
3319 	if (ret)
3320 		goto out;
3321 
3322 	down_read(&fs_info->cleanup_work_sem);
3323 	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3324 	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3325 		up_read(&fs_info->cleanup_work_sem);
3326 		goto out;
3327 	}
3328 	up_read(&fs_info->cleanup_work_sem);
3329 
3330 	mutex_lock(&fs_info->cleaner_mutex);
3331 	ret = btrfs_recover_relocation(fs_info);
3332 	mutex_unlock(&fs_info->cleaner_mutex);
3333 	if (ret < 0) {
3334 		btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3335 		goto out;
3336 	}
3337 
3338 	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3339 	    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3340 		btrfs_info(fs_info, "creating free space tree");
3341 		ret = btrfs_create_free_space_tree(fs_info);
3342 		if (ret) {
3343 			btrfs_warn(fs_info,
3344 				"failed to create free space tree: %d", ret);
3345 			goto out;
3346 		}
3347 	}
3348 
3349 	if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3350 		ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3351 		if (ret)
3352 			goto out;
3353 	}
3354 
3355 	ret = btrfs_resume_balance_async(fs_info);
3356 	if (ret)
3357 		goto out;
3358 
3359 	ret = btrfs_resume_dev_replace_async(fs_info);
3360 	if (ret) {
3361 		btrfs_warn(fs_info, "failed to resume dev_replace");
3362 		goto out;
3363 	}
3364 
3365 	btrfs_qgroup_rescan_resume(fs_info);
3366 
3367 	if (!fs_info->uuid_root) {
3368 		btrfs_info(fs_info, "creating UUID tree");
3369 		ret = btrfs_create_uuid_tree(fs_info);
3370 		if (ret) {
3371 			btrfs_warn(fs_info,
3372 				   "failed to create the UUID tree %d", ret);
3373 			goto out;
3374 		}
3375 	}
3376 
3377 out:
3378 	return ret;
3379 }
3380 
3381 /*
3382  * Do various sanity and dependency checks of different features.
3383  *
3384  * This is the place for less strict checks (like for subpage or artificial
3385  * feature dependencies).
3386  *
3387  * For strict checks or possible corruption detection, see
3388  * btrfs_validate_super().
3389  *
3390  * This should be called after btrfs_parse_options(), as some mount options
3391  * (space cache related) can modify on-disk format like free space tree and
3392  * screw up certain feature dependencies.
3393  */
3394 int btrfs_check_features(struct btrfs_fs_info *fs_info, struct super_block *sb)
3395 {
3396 	struct btrfs_super_block *disk_super = fs_info->super_copy;
3397 	u64 incompat = btrfs_super_incompat_flags(disk_super);
3398 	const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3399 	const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3400 
3401 	if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3402 		btrfs_err(fs_info,
3403 		"cannot mount because of unknown incompat features (0x%llx)",
3404 		    incompat);
3405 		return -EINVAL;
3406 	}
3407 
3408 	/* Runtime limitation for mixed block groups. */
3409 	if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3410 	    (fs_info->sectorsize != fs_info->nodesize)) {
3411 		btrfs_err(fs_info,
3412 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3413 			fs_info->nodesize, fs_info->sectorsize);
3414 		return -EINVAL;
3415 	}
3416 
3417 	/* Mixed backref is an always-enabled feature. */
3418 	incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3419 
3420 	/* Set compression related flags just in case. */
3421 	if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3422 		incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3423 	else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3424 		incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3425 
3426 	/*
3427 	 * An ancient flag, which should really be marked deprecated.
3428 	 * Such runtime limitation doesn't really need a incompat flag.
3429 	 */
3430 	if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3431 		incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3432 
3433 	if (compat_ro_unsupp && !sb_rdonly(sb)) {
3434 		btrfs_err(fs_info,
3435 	"cannot mount read-write because of unknown compat_ro features (0x%llx)",
3436 		       compat_ro);
3437 		return -EINVAL;
3438 	}
3439 
3440 	/*
3441 	 * We have unsupported RO compat features, although RO mounted, we
3442 	 * should not cause any metadata writes, including log replay.
3443 	 * Or we could screw up whatever the new feature requires.
3444 	 */
3445 	if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3446 	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3447 		btrfs_err(fs_info,
3448 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3449 			  compat_ro);
3450 		return -EINVAL;
3451 	}
3452 
3453 	/*
3454 	 * Artificial limitations for block group tree, to force
3455 	 * block-group-tree to rely on no-holes and free-space-tree.
3456 	 */
3457 	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3458 	    (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3459 	     !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3460 		btrfs_err(fs_info,
3461 "block-group-tree feature requires no-holes and free-space-tree features");
3462 		return -EINVAL;
3463 	}
3464 
3465 	/*
3466 	 * Subpage runtime limitation on v1 cache.
3467 	 *
3468 	 * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3469 	 * we're already defaulting to v2 cache, no need to bother v1 as it's
3470 	 * going to be deprecated anyway.
3471 	 */
3472 	if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3473 		btrfs_warn(fs_info,
3474 	"v1 space cache is not supported for page size %lu with sectorsize %u",
3475 			   PAGE_SIZE, fs_info->sectorsize);
3476 		return -EINVAL;
3477 	}
3478 
3479 	/* This can be called by remount, we need to protect the super block. */
3480 	spin_lock(&fs_info->super_lock);
3481 	btrfs_set_super_incompat_flags(disk_super, incompat);
3482 	spin_unlock(&fs_info->super_lock);
3483 
3484 	return 0;
3485 }
3486 
3487 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3488 		      char *options)
3489 {
3490 	u32 sectorsize;
3491 	u32 nodesize;
3492 	u32 stripesize;
3493 	u64 generation;
3494 	u64 features;
3495 	u16 csum_type;
3496 	struct btrfs_super_block *disk_super;
3497 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3498 	struct btrfs_root *tree_root;
3499 	struct btrfs_root *chunk_root;
3500 	int ret;
3501 	int err = -EINVAL;
3502 	int level;
3503 
3504 	ret = init_mount_fs_info(fs_info, sb);
3505 	if (ret) {
3506 		err = ret;
3507 		goto fail;
3508 	}
3509 
3510 	/* These need to be init'ed before we start creating inodes and such. */
3511 	tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3512 				     GFP_KERNEL);
3513 	fs_info->tree_root = tree_root;
3514 	chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3515 				      GFP_KERNEL);
3516 	fs_info->chunk_root = chunk_root;
3517 	if (!tree_root || !chunk_root) {
3518 		err = -ENOMEM;
3519 		goto fail;
3520 	}
3521 
3522 	fs_info->btree_inode = new_inode(sb);
3523 	if (!fs_info->btree_inode) {
3524 		err = -ENOMEM;
3525 		goto fail;
3526 	}
3527 	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3528 	btrfs_init_btree_inode(fs_info);
3529 
3530 	invalidate_bdev(fs_devices->latest_dev->bdev);
3531 
3532 	/*
3533 	 * Read super block and check the signature bytes only
3534 	 */
3535 	disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3536 	if (IS_ERR(disk_super)) {
3537 		err = PTR_ERR(disk_super);
3538 		goto fail_alloc;
3539 	}
3540 
3541 	/*
3542 	 * Verify the type first, if that or the checksum value are
3543 	 * corrupted, we'll find out
3544 	 */
3545 	csum_type = btrfs_super_csum_type(disk_super);
3546 	if (!btrfs_supported_super_csum(csum_type)) {
3547 		btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3548 			  csum_type);
3549 		err = -EINVAL;
3550 		btrfs_release_disk_super(disk_super);
3551 		goto fail_alloc;
3552 	}
3553 
3554 	fs_info->csum_size = btrfs_super_csum_size(disk_super);
3555 
3556 	ret = btrfs_init_csum_hash(fs_info, csum_type);
3557 	if (ret) {
3558 		err = ret;
3559 		btrfs_release_disk_super(disk_super);
3560 		goto fail_alloc;
3561 	}
3562 
3563 	/*
3564 	 * We want to check superblock checksum, the type is stored inside.
3565 	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3566 	 */
3567 	if (btrfs_check_super_csum(fs_info, disk_super)) {
3568 		btrfs_err(fs_info, "superblock checksum mismatch");
3569 		err = -EINVAL;
3570 		btrfs_release_disk_super(disk_super);
3571 		goto fail_alloc;
3572 	}
3573 
3574 	/*
3575 	 * super_copy is zeroed at allocation time and we never touch the
3576 	 * following bytes up to INFO_SIZE, the checksum is calculated from
3577 	 * the whole block of INFO_SIZE
3578 	 */
3579 	memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3580 	btrfs_release_disk_super(disk_super);
3581 
3582 	disk_super = fs_info->super_copy;
3583 
3584 
3585 	features = btrfs_super_flags(disk_super);
3586 	if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3587 		features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3588 		btrfs_set_super_flags(disk_super, features);
3589 		btrfs_info(fs_info,
3590 			"found metadata UUID change in progress flag, clearing");
3591 	}
3592 
3593 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
3594 	       sizeof(*fs_info->super_for_commit));
3595 
3596 	ret = btrfs_validate_mount_super(fs_info);
3597 	if (ret) {
3598 		btrfs_err(fs_info, "superblock contains fatal errors");
3599 		err = -EINVAL;
3600 		goto fail_alloc;
3601 	}
3602 
3603 	if (!btrfs_super_root(disk_super))
3604 		goto fail_alloc;
3605 
3606 	/* check FS state, whether FS is broken. */
3607 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3608 		set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3609 
3610 	/*
3611 	 * In the long term, we'll store the compression type in the super
3612 	 * block, and it'll be used for per file compression control.
3613 	 */
3614 	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3615 
3616 
3617 	/* Set up fs_info before parsing mount options */
3618 	nodesize = btrfs_super_nodesize(disk_super);
3619 	sectorsize = btrfs_super_sectorsize(disk_super);
3620 	stripesize = sectorsize;
3621 	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3622 	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3623 
3624 	fs_info->nodesize = nodesize;
3625 	fs_info->sectorsize = sectorsize;
3626 	fs_info->sectorsize_bits = ilog2(sectorsize);
3627 	fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3628 	fs_info->stripesize = stripesize;
3629 
3630 	ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3631 	if (ret) {
3632 		err = ret;
3633 		goto fail_alloc;
3634 	}
3635 
3636 	ret = btrfs_check_features(fs_info, sb);
3637 	if (ret < 0) {
3638 		err = ret;
3639 		goto fail_alloc;
3640 	}
3641 
3642 	if (sectorsize < PAGE_SIZE) {
3643 		struct btrfs_subpage_info *subpage_info;
3644 
3645 		/*
3646 		 * V1 space cache has some hardcoded PAGE_SIZE usage, and is
3647 		 * going to be deprecated.
3648 		 *
3649 		 * Force to use v2 cache for subpage case.
3650 		 */
3651 		btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
3652 		btrfs_set_and_info(fs_info, FREE_SPACE_TREE,
3653 			"forcing free space tree for sector size %u with page size %lu",
3654 			sectorsize, PAGE_SIZE);
3655 
3656 		btrfs_warn(fs_info,
3657 		"read-write for sector size %u with page size %lu is experimental",
3658 			   sectorsize, PAGE_SIZE);
3659 		subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3660 		if (!subpage_info)
3661 			goto fail_alloc;
3662 		btrfs_init_subpage_info(subpage_info, sectorsize);
3663 		fs_info->subpage_info = subpage_info;
3664 	}
3665 
3666 	ret = btrfs_init_workqueues(fs_info);
3667 	if (ret) {
3668 		err = ret;
3669 		goto fail_sb_buffer;
3670 	}
3671 
3672 	sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3673 	sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3674 
3675 	sb->s_blocksize = sectorsize;
3676 	sb->s_blocksize_bits = blksize_bits(sectorsize);
3677 	memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3678 
3679 	mutex_lock(&fs_info->chunk_mutex);
3680 	ret = btrfs_read_sys_array(fs_info);
3681 	mutex_unlock(&fs_info->chunk_mutex);
3682 	if (ret) {
3683 		btrfs_err(fs_info, "failed to read the system array: %d", ret);
3684 		goto fail_sb_buffer;
3685 	}
3686 
3687 	generation = btrfs_super_chunk_root_generation(disk_super);
3688 	level = btrfs_super_chunk_root_level(disk_super);
3689 	ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3690 			      generation, level);
3691 	if (ret) {
3692 		btrfs_err(fs_info, "failed to read chunk root");
3693 		goto fail_tree_roots;
3694 	}
3695 
3696 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3697 			   offsetof(struct btrfs_header, chunk_tree_uuid),
3698 			   BTRFS_UUID_SIZE);
3699 
3700 	ret = btrfs_read_chunk_tree(fs_info);
3701 	if (ret) {
3702 		btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3703 		goto fail_tree_roots;
3704 	}
3705 
3706 	/*
3707 	 * At this point we know all the devices that make this filesystem,
3708 	 * including the seed devices but we don't know yet if the replace
3709 	 * target is required. So free devices that are not part of this
3710 	 * filesystem but skip the replace target device which is checked
3711 	 * below in btrfs_init_dev_replace().
3712 	 */
3713 	btrfs_free_extra_devids(fs_devices);
3714 	if (!fs_devices->latest_dev->bdev) {
3715 		btrfs_err(fs_info, "failed to read devices");
3716 		goto fail_tree_roots;
3717 	}
3718 
3719 	ret = init_tree_roots(fs_info);
3720 	if (ret)
3721 		goto fail_tree_roots;
3722 
3723 	/*
3724 	 * Get zone type information of zoned block devices. This will also
3725 	 * handle emulation of a zoned filesystem if a regular device has the
3726 	 * zoned incompat feature flag set.
3727 	 */
3728 	ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3729 	if (ret) {
3730 		btrfs_err(fs_info,
3731 			  "zoned: failed to read device zone info: %d",
3732 			  ret);
3733 		goto fail_block_groups;
3734 	}
3735 
3736 	/*
3737 	 * If we have a uuid root and we're not being told to rescan we need to
3738 	 * check the generation here so we can set the
3739 	 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3740 	 * transaction during a balance or the log replay without updating the
3741 	 * uuid generation, and then if we crash we would rescan the uuid tree,
3742 	 * even though it was perfectly fine.
3743 	 */
3744 	if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3745 	    fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3746 		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3747 
3748 	ret = btrfs_verify_dev_extents(fs_info);
3749 	if (ret) {
3750 		btrfs_err(fs_info,
3751 			  "failed to verify dev extents against chunks: %d",
3752 			  ret);
3753 		goto fail_block_groups;
3754 	}
3755 	ret = btrfs_recover_balance(fs_info);
3756 	if (ret) {
3757 		btrfs_err(fs_info, "failed to recover balance: %d", ret);
3758 		goto fail_block_groups;
3759 	}
3760 
3761 	ret = btrfs_init_dev_stats(fs_info);
3762 	if (ret) {
3763 		btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3764 		goto fail_block_groups;
3765 	}
3766 
3767 	ret = btrfs_init_dev_replace(fs_info);
3768 	if (ret) {
3769 		btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3770 		goto fail_block_groups;
3771 	}
3772 
3773 	ret = btrfs_check_zoned_mode(fs_info);
3774 	if (ret) {
3775 		btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3776 			  ret);
3777 		goto fail_block_groups;
3778 	}
3779 
3780 	ret = btrfs_sysfs_add_fsid(fs_devices);
3781 	if (ret) {
3782 		btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3783 				ret);
3784 		goto fail_block_groups;
3785 	}
3786 
3787 	ret = btrfs_sysfs_add_mounted(fs_info);
3788 	if (ret) {
3789 		btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3790 		goto fail_fsdev_sysfs;
3791 	}
3792 
3793 	ret = btrfs_init_space_info(fs_info);
3794 	if (ret) {
3795 		btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3796 		goto fail_sysfs;
3797 	}
3798 
3799 	ret = btrfs_read_block_groups(fs_info);
3800 	if (ret) {
3801 		btrfs_err(fs_info, "failed to read block groups: %d", ret);
3802 		goto fail_sysfs;
3803 	}
3804 
3805 	btrfs_free_zone_cache(fs_info);
3806 
3807 	if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3808 	    !btrfs_check_rw_degradable(fs_info, NULL)) {
3809 		btrfs_warn(fs_info,
3810 		"writable mount is not allowed due to too many missing devices");
3811 		goto fail_sysfs;
3812 	}
3813 
3814 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3815 					       "btrfs-cleaner");
3816 	if (IS_ERR(fs_info->cleaner_kthread))
3817 		goto fail_sysfs;
3818 
3819 	fs_info->transaction_kthread = kthread_run(transaction_kthread,
3820 						   tree_root,
3821 						   "btrfs-transaction");
3822 	if (IS_ERR(fs_info->transaction_kthread))
3823 		goto fail_cleaner;
3824 
3825 	if (!btrfs_test_opt(fs_info, NOSSD) &&
3826 	    !fs_info->fs_devices->rotating) {
3827 		btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3828 	}
3829 
3830 	/*
3831 	 * For devices supporting discard turn on discard=async automatically,
3832 	 * unless it's already set or disabled. This could be turned off by
3833 	 * nodiscard for the same mount.
3834 	 */
3835 	if (!(btrfs_test_opt(fs_info, DISCARD_SYNC) ||
3836 	      btrfs_test_opt(fs_info, DISCARD_ASYNC) ||
3837 	      btrfs_test_opt(fs_info, NODISCARD)) &&
3838 	    fs_info->fs_devices->discardable) {
3839 		btrfs_set_and_info(fs_info, DISCARD_ASYNC,
3840 				   "auto enabling async discard");
3841 		btrfs_clear_opt(fs_info->mount_opt, NODISCARD);
3842 	}
3843 
3844 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3845 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3846 		ret = btrfsic_mount(fs_info, fs_devices,
3847 				    btrfs_test_opt(fs_info,
3848 					CHECK_INTEGRITY_DATA) ? 1 : 0,
3849 				    fs_info->check_integrity_print_mask);
3850 		if (ret)
3851 			btrfs_warn(fs_info,
3852 				"failed to initialize integrity check module: %d",
3853 				ret);
3854 	}
3855 #endif
3856 	ret = btrfs_read_qgroup_config(fs_info);
3857 	if (ret)
3858 		goto fail_trans_kthread;
3859 
3860 	if (btrfs_build_ref_tree(fs_info))
3861 		btrfs_err(fs_info, "couldn't build ref tree");
3862 
3863 	/* do not make disk changes in broken FS or nologreplay is given */
3864 	if (btrfs_super_log_root(disk_super) != 0 &&
3865 	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3866 		btrfs_info(fs_info, "start tree-log replay");
3867 		ret = btrfs_replay_log(fs_info, fs_devices);
3868 		if (ret) {
3869 			err = ret;
3870 			goto fail_qgroup;
3871 		}
3872 	}
3873 
3874 	fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3875 	if (IS_ERR(fs_info->fs_root)) {
3876 		err = PTR_ERR(fs_info->fs_root);
3877 		btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3878 		fs_info->fs_root = NULL;
3879 		goto fail_qgroup;
3880 	}
3881 
3882 	if (sb_rdonly(sb))
3883 		goto clear_oneshot;
3884 
3885 	ret = btrfs_start_pre_rw_mount(fs_info);
3886 	if (ret) {
3887 		close_ctree(fs_info);
3888 		return ret;
3889 	}
3890 	btrfs_discard_resume(fs_info);
3891 
3892 	if (fs_info->uuid_root &&
3893 	    (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3894 	     fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3895 		btrfs_info(fs_info, "checking UUID tree");
3896 		ret = btrfs_check_uuid_tree(fs_info);
3897 		if (ret) {
3898 			btrfs_warn(fs_info,
3899 				"failed to check the UUID tree: %d", ret);
3900 			close_ctree(fs_info);
3901 			return ret;
3902 		}
3903 	}
3904 
3905 	set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3906 
3907 	/* Kick the cleaner thread so it'll start deleting snapshots. */
3908 	if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3909 		wake_up_process(fs_info->cleaner_kthread);
3910 
3911 clear_oneshot:
3912 	btrfs_clear_oneshot_options(fs_info);
3913 	return 0;
3914 
3915 fail_qgroup:
3916 	btrfs_free_qgroup_config(fs_info);
3917 fail_trans_kthread:
3918 	kthread_stop(fs_info->transaction_kthread);
3919 	btrfs_cleanup_transaction(fs_info);
3920 	btrfs_free_fs_roots(fs_info);
3921 fail_cleaner:
3922 	kthread_stop(fs_info->cleaner_kthread);
3923 
3924 	/*
3925 	 * make sure we're done with the btree inode before we stop our
3926 	 * kthreads
3927 	 */
3928 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3929 
3930 fail_sysfs:
3931 	btrfs_sysfs_remove_mounted(fs_info);
3932 
3933 fail_fsdev_sysfs:
3934 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3935 
3936 fail_block_groups:
3937 	btrfs_put_block_group_cache(fs_info);
3938 
3939 fail_tree_roots:
3940 	if (fs_info->data_reloc_root)
3941 		btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3942 	free_root_pointers(fs_info, true);
3943 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3944 
3945 fail_sb_buffer:
3946 	btrfs_stop_all_workers(fs_info);
3947 	btrfs_free_block_groups(fs_info);
3948 fail_alloc:
3949 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3950 
3951 	iput(fs_info->btree_inode);
3952 fail:
3953 	btrfs_close_devices(fs_info->fs_devices);
3954 	return err;
3955 }
3956 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3957 
3958 static void btrfs_end_super_write(struct bio *bio)
3959 {
3960 	struct btrfs_device *device = bio->bi_private;
3961 	struct bio_vec *bvec;
3962 	struct bvec_iter_all iter_all;
3963 	struct page *page;
3964 
3965 	bio_for_each_segment_all(bvec, bio, iter_all) {
3966 		page = bvec->bv_page;
3967 
3968 		if (bio->bi_status) {
3969 			btrfs_warn_rl_in_rcu(device->fs_info,
3970 				"lost page write due to IO error on %s (%d)",
3971 				btrfs_dev_name(device),
3972 				blk_status_to_errno(bio->bi_status));
3973 			ClearPageUptodate(page);
3974 			SetPageError(page);
3975 			btrfs_dev_stat_inc_and_print(device,
3976 						     BTRFS_DEV_STAT_WRITE_ERRS);
3977 		} else {
3978 			SetPageUptodate(page);
3979 		}
3980 
3981 		put_page(page);
3982 		unlock_page(page);
3983 	}
3984 
3985 	bio_put(bio);
3986 }
3987 
3988 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3989 						   int copy_num, bool drop_cache)
3990 {
3991 	struct btrfs_super_block *super;
3992 	struct page *page;
3993 	u64 bytenr, bytenr_orig;
3994 	struct address_space *mapping = bdev->bd_inode->i_mapping;
3995 	int ret;
3996 
3997 	bytenr_orig = btrfs_sb_offset(copy_num);
3998 	ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3999 	if (ret == -ENOENT)
4000 		return ERR_PTR(-EINVAL);
4001 	else if (ret)
4002 		return ERR_PTR(ret);
4003 
4004 	if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
4005 		return ERR_PTR(-EINVAL);
4006 
4007 	if (drop_cache) {
4008 		/* This should only be called with the primary sb. */
4009 		ASSERT(copy_num == 0);
4010 
4011 		/*
4012 		 * Drop the page of the primary superblock, so later read will
4013 		 * always read from the device.
4014 		 */
4015 		invalidate_inode_pages2_range(mapping,
4016 				bytenr >> PAGE_SHIFT,
4017 				(bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
4018 	}
4019 
4020 	page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
4021 	if (IS_ERR(page))
4022 		return ERR_CAST(page);
4023 
4024 	super = page_address(page);
4025 	if (btrfs_super_magic(super) != BTRFS_MAGIC) {
4026 		btrfs_release_disk_super(super);
4027 		return ERR_PTR(-ENODATA);
4028 	}
4029 
4030 	if (btrfs_super_bytenr(super) != bytenr_orig) {
4031 		btrfs_release_disk_super(super);
4032 		return ERR_PTR(-EINVAL);
4033 	}
4034 
4035 	return super;
4036 }
4037 
4038 
4039 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
4040 {
4041 	struct btrfs_super_block *super, *latest = NULL;
4042 	int i;
4043 	u64 transid = 0;
4044 
4045 	/* we would like to check all the supers, but that would make
4046 	 * a btrfs mount succeed after a mkfs from a different FS.
4047 	 * So, we need to add a special mount option to scan for
4048 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
4049 	 */
4050 	for (i = 0; i < 1; i++) {
4051 		super = btrfs_read_dev_one_super(bdev, i, false);
4052 		if (IS_ERR(super))
4053 			continue;
4054 
4055 		if (!latest || btrfs_super_generation(super) > transid) {
4056 			if (latest)
4057 				btrfs_release_disk_super(super);
4058 
4059 			latest = super;
4060 			transid = btrfs_super_generation(super);
4061 		}
4062 	}
4063 
4064 	return super;
4065 }
4066 
4067 /*
4068  * Write superblock @sb to the @device. Do not wait for completion, all the
4069  * pages we use for writing are locked.
4070  *
4071  * Write @max_mirrors copies of the superblock, where 0 means default that fit
4072  * the expected device size at commit time. Note that max_mirrors must be
4073  * same for write and wait phases.
4074  *
4075  * Return number of errors when page is not found or submission fails.
4076  */
4077 static int write_dev_supers(struct btrfs_device *device,
4078 			    struct btrfs_super_block *sb, int max_mirrors)
4079 {
4080 	struct btrfs_fs_info *fs_info = device->fs_info;
4081 	struct address_space *mapping = device->bdev->bd_inode->i_mapping;
4082 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
4083 	int i;
4084 	int errors = 0;
4085 	int ret;
4086 	u64 bytenr, bytenr_orig;
4087 
4088 	if (max_mirrors == 0)
4089 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4090 
4091 	shash->tfm = fs_info->csum_shash;
4092 
4093 	for (i = 0; i < max_mirrors; i++) {
4094 		struct page *page;
4095 		struct bio *bio;
4096 		struct btrfs_super_block *disk_super;
4097 
4098 		bytenr_orig = btrfs_sb_offset(i);
4099 		ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
4100 		if (ret == -ENOENT) {
4101 			continue;
4102 		} else if (ret < 0) {
4103 			btrfs_err(device->fs_info,
4104 				"couldn't get super block location for mirror %d",
4105 				i);
4106 			errors++;
4107 			continue;
4108 		}
4109 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4110 		    device->commit_total_bytes)
4111 			break;
4112 
4113 		btrfs_set_super_bytenr(sb, bytenr_orig);
4114 
4115 		crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
4116 				    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
4117 				    sb->csum);
4118 
4119 		page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
4120 					   GFP_NOFS);
4121 		if (!page) {
4122 			btrfs_err(device->fs_info,
4123 			    "couldn't get super block page for bytenr %llu",
4124 			    bytenr);
4125 			errors++;
4126 			continue;
4127 		}
4128 
4129 		/* Bump the refcount for wait_dev_supers() */
4130 		get_page(page);
4131 
4132 		disk_super = page_address(page);
4133 		memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
4134 
4135 		/*
4136 		 * Directly use bios here instead of relying on the page cache
4137 		 * to do I/O, so we don't lose the ability to do integrity
4138 		 * checking.
4139 		 */
4140 		bio = bio_alloc(device->bdev, 1,
4141 				REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
4142 				GFP_NOFS);
4143 		bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
4144 		bio->bi_private = device;
4145 		bio->bi_end_io = btrfs_end_super_write;
4146 		__bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
4147 			       offset_in_page(bytenr));
4148 
4149 		/*
4150 		 * We FUA only the first super block.  The others we allow to
4151 		 * go down lazy and there's a short window where the on-disk
4152 		 * copies might still contain the older version.
4153 		 */
4154 		if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
4155 			bio->bi_opf |= REQ_FUA;
4156 
4157 		btrfsic_check_bio(bio);
4158 		submit_bio(bio);
4159 
4160 		if (btrfs_advance_sb_log(device, i))
4161 			errors++;
4162 	}
4163 	return errors < i ? 0 : -1;
4164 }
4165 
4166 /*
4167  * Wait for write completion of superblocks done by write_dev_supers,
4168  * @max_mirrors same for write and wait phases.
4169  *
4170  * Return number of errors when page is not found or not marked up to
4171  * date.
4172  */
4173 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
4174 {
4175 	int i;
4176 	int errors = 0;
4177 	bool primary_failed = false;
4178 	int ret;
4179 	u64 bytenr;
4180 
4181 	if (max_mirrors == 0)
4182 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4183 
4184 	for (i = 0; i < max_mirrors; i++) {
4185 		struct page *page;
4186 
4187 		ret = btrfs_sb_log_location(device, i, READ, &bytenr);
4188 		if (ret == -ENOENT) {
4189 			break;
4190 		} else if (ret < 0) {
4191 			errors++;
4192 			if (i == 0)
4193 				primary_failed = true;
4194 			continue;
4195 		}
4196 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4197 		    device->commit_total_bytes)
4198 			break;
4199 
4200 		page = find_get_page(device->bdev->bd_inode->i_mapping,
4201 				     bytenr >> PAGE_SHIFT);
4202 		if (!page) {
4203 			errors++;
4204 			if (i == 0)
4205 				primary_failed = true;
4206 			continue;
4207 		}
4208 		/* Page is submitted locked and unlocked once the IO completes */
4209 		wait_on_page_locked(page);
4210 		if (PageError(page)) {
4211 			errors++;
4212 			if (i == 0)
4213 				primary_failed = true;
4214 		}
4215 
4216 		/* Drop our reference */
4217 		put_page(page);
4218 
4219 		/* Drop the reference from the writing run */
4220 		put_page(page);
4221 	}
4222 
4223 	/* log error, force error return */
4224 	if (primary_failed) {
4225 		btrfs_err(device->fs_info, "error writing primary super block to device %llu",
4226 			  device->devid);
4227 		return -1;
4228 	}
4229 
4230 	return errors < i ? 0 : -1;
4231 }
4232 
4233 /*
4234  * endio for the write_dev_flush, this will wake anyone waiting
4235  * for the barrier when it is done
4236  */
4237 static void btrfs_end_empty_barrier(struct bio *bio)
4238 {
4239 	bio_uninit(bio);
4240 	complete(bio->bi_private);
4241 }
4242 
4243 /*
4244  * Submit a flush request to the device if it supports it. Error handling is
4245  * done in the waiting counterpart.
4246  */
4247 static void write_dev_flush(struct btrfs_device *device)
4248 {
4249 	struct bio *bio = &device->flush_bio;
4250 
4251 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4252 	/*
4253 	 * When a disk has write caching disabled, we skip submission of a bio
4254 	 * with flush and sync requests before writing the superblock, since
4255 	 * it's not needed. However when the integrity checker is enabled, this
4256 	 * results in reports that there are metadata blocks referred by a
4257 	 * superblock that were not properly flushed. So don't skip the bio
4258 	 * submission only when the integrity checker is enabled for the sake
4259 	 * of simplicity, since this is a debug tool and not meant for use in
4260 	 * non-debug builds.
4261 	 */
4262 	if (!bdev_write_cache(device->bdev))
4263 		return;
4264 #endif
4265 
4266 	bio_init(bio, device->bdev, NULL, 0,
4267 		 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
4268 	bio->bi_end_io = btrfs_end_empty_barrier;
4269 	init_completion(&device->flush_wait);
4270 	bio->bi_private = &device->flush_wait;
4271 
4272 	btrfsic_check_bio(bio);
4273 	submit_bio(bio);
4274 	set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4275 }
4276 
4277 /*
4278  * If the flush bio has been submitted by write_dev_flush, wait for it.
4279  */
4280 static blk_status_t wait_dev_flush(struct btrfs_device *device)
4281 {
4282 	struct bio *bio = &device->flush_bio;
4283 
4284 	if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
4285 		return BLK_STS_OK;
4286 
4287 	clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4288 	wait_for_completion_io(&device->flush_wait);
4289 
4290 	return bio->bi_status;
4291 }
4292 
4293 static int check_barrier_error(struct btrfs_fs_info *fs_info)
4294 {
4295 	if (!btrfs_check_rw_degradable(fs_info, NULL))
4296 		return -EIO;
4297 	return 0;
4298 }
4299 
4300 /*
4301  * send an empty flush down to each device in parallel,
4302  * then wait for them
4303  */
4304 static int barrier_all_devices(struct btrfs_fs_info *info)
4305 {
4306 	struct list_head *head;
4307 	struct btrfs_device *dev;
4308 	int errors_wait = 0;
4309 	blk_status_t ret;
4310 
4311 	lockdep_assert_held(&info->fs_devices->device_list_mutex);
4312 	/* send down all the barriers */
4313 	head = &info->fs_devices->devices;
4314 	list_for_each_entry(dev, head, dev_list) {
4315 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4316 			continue;
4317 		if (!dev->bdev)
4318 			continue;
4319 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4320 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4321 			continue;
4322 
4323 		write_dev_flush(dev);
4324 		dev->last_flush_error = BLK_STS_OK;
4325 	}
4326 
4327 	/* wait for all the barriers */
4328 	list_for_each_entry(dev, head, dev_list) {
4329 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4330 			continue;
4331 		if (!dev->bdev) {
4332 			errors_wait++;
4333 			continue;
4334 		}
4335 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4336 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4337 			continue;
4338 
4339 		ret = wait_dev_flush(dev);
4340 		if (ret) {
4341 			dev->last_flush_error = ret;
4342 			btrfs_dev_stat_inc_and_print(dev,
4343 					BTRFS_DEV_STAT_FLUSH_ERRS);
4344 			errors_wait++;
4345 		}
4346 	}
4347 
4348 	if (errors_wait) {
4349 		/*
4350 		 * At some point we need the status of all disks
4351 		 * to arrive at the volume status. So error checking
4352 		 * is being pushed to a separate loop.
4353 		 */
4354 		return check_barrier_error(info);
4355 	}
4356 	return 0;
4357 }
4358 
4359 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4360 {
4361 	int raid_type;
4362 	int min_tolerated = INT_MAX;
4363 
4364 	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4365 	    (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4366 		min_tolerated = min_t(int, min_tolerated,
4367 				    btrfs_raid_array[BTRFS_RAID_SINGLE].
4368 				    tolerated_failures);
4369 
4370 	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4371 		if (raid_type == BTRFS_RAID_SINGLE)
4372 			continue;
4373 		if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4374 			continue;
4375 		min_tolerated = min_t(int, min_tolerated,
4376 				    btrfs_raid_array[raid_type].
4377 				    tolerated_failures);
4378 	}
4379 
4380 	if (min_tolerated == INT_MAX) {
4381 		pr_warn("BTRFS: unknown raid flag: %llu", flags);
4382 		min_tolerated = 0;
4383 	}
4384 
4385 	return min_tolerated;
4386 }
4387 
4388 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4389 {
4390 	struct list_head *head;
4391 	struct btrfs_device *dev;
4392 	struct btrfs_super_block *sb;
4393 	struct btrfs_dev_item *dev_item;
4394 	int ret;
4395 	int do_barriers;
4396 	int max_errors;
4397 	int total_errors = 0;
4398 	u64 flags;
4399 
4400 	do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4401 
4402 	/*
4403 	 * max_mirrors == 0 indicates we're from commit_transaction,
4404 	 * not from fsync where the tree roots in fs_info have not
4405 	 * been consistent on disk.
4406 	 */
4407 	if (max_mirrors == 0)
4408 		backup_super_roots(fs_info);
4409 
4410 	sb = fs_info->super_for_commit;
4411 	dev_item = &sb->dev_item;
4412 
4413 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
4414 	head = &fs_info->fs_devices->devices;
4415 	max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4416 
4417 	if (do_barriers) {
4418 		ret = barrier_all_devices(fs_info);
4419 		if (ret) {
4420 			mutex_unlock(
4421 				&fs_info->fs_devices->device_list_mutex);
4422 			btrfs_handle_fs_error(fs_info, ret,
4423 					      "errors while submitting device barriers.");
4424 			return ret;
4425 		}
4426 	}
4427 
4428 	list_for_each_entry(dev, head, dev_list) {
4429 		if (!dev->bdev) {
4430 			total_errors++;
4431 			continue;
4432 		}
4433 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4434 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4435 			continue;
4436 
4437 		btrfs_set_stack_device_generation(dev_item, 0);
4438 		btrfs_set_stack_device_type(dev_item, dev->type);
4439 		btrfs_set_stack_device_id(dev_item, dev->devid);
4440 		btrfs_set_stack_device_total_bytes(dev_item,
4441 						   dev->commit_total_bytes);
4442 		btrfs_set_stack_device_bytes_used(dev_item,
4443 						  dev->commit_bytes_used);
4444 		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4445 		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4446 		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4447 		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4448 		memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4449 		       BTRFS_FSID_SIZE);
4450 
4451 		flags = btrfs_super_flags(sb);
4452 		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4453 
4454 		ret = btrfs_validate_write_super(fs_info, sb);
4455 		if (ret < 0) {
4456 			mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4457 			btrfs_handle_fs_error(fs_info, -EUCLEAN,
4458 				"unexpected superblock corruption detected");
4459 			return -EUCLEAN;
4460 		}
4461 
4462 		ret = write_dev_supers(dev, sb, max_mirrors);
4463 		if (ret)
4464 			total_errors++;
4465 	}
4466 	if (total_errors > max_errors) {
4467 		btrfs_err(fs_info, "%d errors while writing supers",
4468 			  total_errors);
4469 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4470 
4471 		/* FUA is masked off if unsupported and can't be the reason */
4472 		btrfs_handle_fs_error(fs_info, -EIO,
4473 				      "%d errors while writing supers",
4474 				      total_errors);
4475 		return -EIO;
4476 	}
4477 
4478 	total_errors = 0;
4479 	list_for_each_entry(dev, head, dev_list) {
4480 		if (!dev->bdev)
4481 			continue;
4482 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4483 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4484 			continue;
4485 
4486 		ret = wait_dev_supers(dev, max_mirrors);
4487 		if (ret)
4488 			total_errors++;
4489 	}
4490 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4491 	if (total_errors > max_errors) {
4492 		btrfs_handle_fs_error(fs_info, -EIO,
4493 				      "%d errors while writing supers",
4494 				      total_errors);
4495 		return -EIO;
4496 	}
4497 	return 0;
4498 }
4499 
4500 /* Drop a fs root from the radix tree and free it. */
4501 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4502 				  struct btrfs_root *root)
4503 {
4504 	bool drop_ref = false;
4505 
4506 	spin_lock(&fs_info->fs_roots_radix_lock);
4507 	radix_tree_delete(&fs_info->fs_roots_radix,
4508 			  (unsigned long)root->root_key.objectid);
4509 	if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4510 		drop_ref = true;
4511 	spin_unlock(&fs_info->fs_roots_radix_lock);
4512 
4513 	if (BTRFS_FS_ERROR(fs_info)) {
4514 		ASSERT(root->log_root == NULL);
4515 		if (root->reloc_root) {
4516 			btrfs_put_root(root->reloc_root);
4517 			root->reloc_root = NULL;
4518 		}
4519 	}
4520 
4521 	if (drop_ref)
4522 		btrfs_put_root(root);
4523 }
4524 
4525 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4526 {
4527 	u64 root_objectid = 0;
4528 	struct btrfs_root *gang[8];
4529 	int i = 0;
4530 	int err = 0;
4531 	unsigned int ret = 0;
4532 
4533 	while (1) {
4534 		spin_lock(&fs_info->fs_roots_radix_lock);
4535 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4536 					     (void **)gang, root_objectid,
4537 					     ARRAY_SIZE(gang));
4538 		if (!ret) {
4539 			spin_unlock(&fs_info->fs_roots_radix_lock);
4540 			break;
4541 		}
4542 		root_objectid = gang[ret - 1]->root_key.objectid + 1;
4543 
4544 		for (i = 0; i < ret; i++) {
4545 			/* Avoid to grab roots in dead_roots */
4546 			if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4547 				gang[i] = NULL;
4548 				continue;
4549 			}
4550 			/* grab all the search result for later use */
4551 			gang[i] = btrfs_grab_root(gang[i]);
4552 		}
4553 		spin_unlock(&fs_info->fs_roots_radix_lock);
4554 
4555 		for (i = 0; i < ret; i++) {
4556 			if (!gang[i])
4557 				continue;
4558 			root_objectid = gang[i]->root_key.objectid;
4559 			err = btrfs_orphan_cleanup(gang[i]);
4560 			if (err)
4561 				break;
4562 			btrfs_put_root(gang[i]);
4563 		}
4564 		root_objectid++;
4565 	}
4566 
4567 	/* release the uncleaned roots due to error */
4568 	for (; i < ret; i++) {
4569 		if (gang[i])
4570 			btrfs_put_root(gang[i]);
4571 	}
4572 	return err;
4573 }
4574 
4575 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4576 {
4577 	struct btrfs_root *root = fs_info->tree_root;
4578 	struct btrfs_trans_handle *trans;
4579 
4580 	mutex_lock(&fs_info->cleaner_mutex);
4581 	btrfs_run_delayed_iputs(fs_info);
4582 	mutex_unlock(&fs_info->cleaner_mutex);
4583 	wake_up_process(fs_info->cleaner_kthread);
4584 
4585 	/* wait until ongoing cleanup work done */
4586 	down_write(&fs_info->cleanup_work_sem);
4587 	up_write(&fs_info->cleanup_work_sem);
4588 
4589 	trans = btrfs_join_transaction(root);
4590 	if (IS_ERR(trans))
4591 		return PTR_ERR(trans);
4592 	return btrfs_commit_transaction(trans);
4593 }
4594 
4595 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4596 {
4597 	struct btrfs_transaction *trans;
4598 	struct btrfs_transaction *tmp;
4599 	bool found = false;
4600 
4601 	if (list_empty(&fs_info->trans_list))
4602 		return;
4603 
4604 	/*
4605 	 * This function is only called at the very end of close_ctree(),
4606 	 * thus no other running transaction, no need to take trans_lock.
4607 	 */
4608 	ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4609 	list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4610 		struct extent_state *cached = NULL;
4611 		u64 dirty_bytes = 0;
4612 		u64 cur = 0;
4613 		u64 found_start;
4614 		u64 found_end;
4615 
4616 		found = true;
4617 		while (!find_first_extent_bit(&trans->dirty_pages, cur,
4618 			&found_start, &found_end, EXTENT_DIRTY, &cached)) {
4619 			dirty_bytes += found_end + 1 - found_start;
4620 			cur = found_end + 1;
4621 		}
4622 		btrfs_warn(fs_info,
4623 	"transaction %llu (with %llu dirty metadata bytes) is not committed",
4624 			   trans->transid, dirty_bytes);
4625 		btrfs_cleanup_one_transaction(trans, fs_info);
4626 
4627 		if (trans == fs_info->running_transaction)
4628 			fs_info->running_transaction = NULL;
4629 		list_del_init(&trans->list);
4630 
4631 		btrfs_put_transaction(trans);
4632 		trace_btrfs_transaction_commit(fs_info);
4633 	}
4634 	ASSERT(!found);
4635 }
4636 
4637 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4638 {
4639 	int ret;
4640 
4641 	set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4642 
4643 	/*
4644 	 * If we had UNFINISHED_DROPS we could still be processing them, so
4645 	 * clear that bit and wake up relocation so it can stop.
4646 	 * We must do this before stopping the block group reclaim task, because
4647 	 * at btrfs_relocate_block_group() we wait for this bit, and after the
4648 	 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4649 	 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4650 	 * return 1.
4651 	 */
4652 	btrfs_wake_unfinished_drop(fs_info);
4653 
4654 	/*
4655 	 * We may have the reclaim task running and relocating a data block group,
4656 	 * in which case it may create delayed iputs. So stop it before we park
4657 	 * the cleaner kthread otherwise we can get new delayed iputs after
4658 	 * parking the cleaner, and that can make the async reclaim task to hang
4659 	 * if it's waiting for delayed iputs to complete, since the cleaner is
4660 	 * parked and can not run delayed iputs - this will make us hang when
4661 	 * trying to stop the async reclaim task.
4662 	 */
4663 	cancel_work_sync(&fs_info->reclaim_bgs_work);
4664 	/*
4665 	 * We don't want the cleaner to start new transactions, add more delayed
4666 	 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4667 	 * because that frees the task_struct, and the transaction kthread might
4668 	 * still try to wake up the cleaner.
4669 	 */
4670 	kthread_park(fs_info->cleaner_kthread);
4671 
4672 	/* wait for the qgroup rescan worker to stop */
4673 	btrfs_qgroup_wait_for_completion(fs_info, false);
4674 
4675 	/* wait for the uuid_scan task to finish */
4676 	down(&fs_info->uuid_tree_rescan_sem);
4677 	/* avoid complains from lockdep et al., set sem back to initial state */
4678 	up(&fs_info->uuid_tree_rescan_sem);
4679 
4680 	/* pause restriper - we want to resume on mount */
4681 	btrfs_pause_balance(fs_info);
4682 
4683 	btrfs_dev_replace_suspend_for_unmount(fs_info);
4684 
4685 	btrfs_scrub_cancel(fs_info);
4686 
4687 	/* wait for any defraggers to finish */
4688 	wait_event(fs_info->transaction_wait,
4689 		   (atomic_read(&fs_info->defrag_running) == 0));
4690 
4691 	/* clear out the rbtree of defraggable inodes */
4692 	btrfs_cleanup_defrag_inodes(fs_info);
4693 
4694 	/*
4695 	 * After we parked the cleaner kthread, ordered extents may have
4696 	 * completed and created new delayed iputs. If one of the async reclaim
4697 	 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4698 	 * can hang forever trying to stop it, because if a delayed iput is
4699 	 * added after it ran btrfs_run_delayed_iputs() and before it called
4700 	 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4701 	 * no one else to run iputs.
4702 	 *
4703 	 * So wait for all ongoing ordered extents to complete and then run
4704 	 * delayed iputs. This works because once we reach this point no one
4705 	 * can either create new ordered extents nor create delayed iputs
4706 	 * through some other means.
4707 	 *
4708 	 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4709 	 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4710 	 * but the delayed iput for the respective inode is made only when doing
4711 	 * the final btrfs_put_ordered_extent() (which must happen at
4712 	 * btrfs_finish_ordered_io() when we are unmounting).
4713 	 */
4714 	btrfs_flush_workqueue(fs_info->endio_write_workers);
4715 	/* Ordered extents for free space inodes. */
4716 	btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4717 	btrfs_run_delayed_iputs(fs_info);
4718 
4719 	cancel_work_sync(&fs_info->async_reclaim_work);
4720 	cancel_work_sync(&fs_info->async_data_reclaim_work);
4721 	cancel_work_sync(&fs_info->preempt_reclaim_work);
4722 
4723 	/* Cancel or finish ongoing discard work */
4724 	btrfs_discard_cleanup(fs_info);
4725 
4726 	if (!sb_rdonly(fs_info->sb)) {
4727 		/*
4728 		 * The cleaner kthread is stopped, so do one final pass over
4729 		 * unused block groups.
4730 		 */
4731 		btrfs_delete_unused_bgs(fs_info);
4732 
4733 		/*
4734 		 * There might be existing delayed inode workers still running
4735 		 * and holding an empty delayed inode item. We must wait for
4736 		 * them to complete first because they can create a transaction.
4737 		 * This happens when someone calls btrfs_balance_delayed_items()
4738 		 * and then a transaction commit runs the same delayed nodes
4739 		 * before any delayed worker has done something with the nodes.
4740 		 * We must wait for any worker here and not at transaction
4741 		 * commit time since that could cause a deadlock.
4742 		 * This is a very rare case.
4743 		 */
4744 		btrfs_flush_workqueue(fs_info->delayed_workers);
4745 
4746 		ret = btrfs_commit_super(fs_info);
4747 		if (ret)
4748 			btrfs_err(fs_info, "commit super ret %d", ret);
4749 	}
4750 
4751 	if (BTRFS_FS_ERROR(fs_info))
4752 		btrfs_error_commit_super(fs_info);
4753 
4754 	kthread_stop(fs_info->transaction_kthread);
4755 	kthread_stop(fs_info->cleaner_kthread);
4756 
4757 	ASSERT(list_empty(&fs_info->delayed_iputs));
4758 	set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4759 
4760 	if (btrfs_check_quota_leak(fs_info)) {
4761 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4762 		btrfs_err(fs_info, "qgroup reserved space leaked");
4763 	}
4764 
4765 	btrfs_free_qgroup_config(fs_info);
4766 	ASSERT(list_empty(&fs_info->delalloc_roots));
4767 
4768 	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4769 		btrfs_info(fs_info, "at unmount delalloc count %lld",
4770 		       percpu_counter_sum(&fs_info->delalloc_bytes));
4771 	}
4772 
4773 	if (percpu_counter_sum(&fs_info->ordered_bytes))
4774 		btrfs_info(fs_info, "at unmount dio bytes count %lld",
4775 			   percpu_counter_sum(&fs_info->ordered_bytes));
4776 
4777 	btrfs_sysfs_remove_mounted(fs_info);
4778 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4779 
4780 	btrfs_put_block_group_cache(fs_info);
4781 
4782 	/*
4783 	 * we must make sure there is not any read request to
4784 	 * submit after we stopping all workers.
4785 	 */
4786 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4787 	btrfs_stop_all_workers(fs_info);
4788 
4789 	/* We shouldn't have any transaction open at this point */
4790 	warn_about_uncommitted_trans(fs_info);
4791 
4792 	clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4793 	free_root_pointers(fs_info, true);
4794 	btrfs_free_fs_roots(fs_info);
4795 
4796 	/*
4797 	 * We must free the block groups after dropping the fs_roots as we could
4798 	 * have had an IO error and have left over tree log blocks that aren't
4799 	 * cleaned up until the fs roots are freed.  This makes the block group
4800 	 * accounting appear to be wrong because there's pending reserved bytes,
4801 	 * so make sure we do the block group cleanup afterwards.
4802 	 */
4803 	btrfs_free_block_groups(fs_info);
4804 
4805 	iput(fs_info->btree_inode);
4806 
4807 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4808 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4809 		btrfsic_unmount(fs_info->fs_devices);
4810 #endif
4811 
4812 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
4813 	btrfs_close_devices(fs_info->fs_devices);
4814 }
4815 
4816 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4817 			  int atomic)
4818 {
4819 	int ret;
4820 	struct inode *btree_inode = buf->pages[0]->mapping->host;
4821 
4822 	ret = extent_buffer_uptodate(buf);
4823 	if (!ret)
4824 		return ret;
4825 
4826 	ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4827 				    parent_transid, atomic);
4828 	if (ret == -EAGAIN)
4829 		return ret;
4830 	return !ret;
4831 }
4832 
4833 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4834 {
4835 	struct btrfs_fs_info *fs_info = buf->fs_info;
4836 	u64 transid = btrfs_header_generation(buf);
4837 	int was_dirty;
4838 
4839 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4840 	/*
4841 	 * This is a fast path so only do this check if we have sanity tests
4842 	 * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4843 	 * outside of the sanity tests.
4844 	 */
4845 	if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4846 		return;
4847 #endif
4848 	btrfs_assert_tree_write_locked(buf);
4849 	if (transid != fs_info->generation)
4850 		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4851 			buf->start, transid, fs_info->generation);
4852 	was_dirty = set_extent_buffer_dirty(buf);
4853 	if (!was_dirty)
4854 		percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4855 					 buf->len,
4856 					 fs_info->dirty_metadata_batch);
4857 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4858 	/*
4859 	 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4860 	 * but item data not updated.
4861 	 * So here we should only check item pointers, not item data.
4862 	 */
4863 	if (btrfs_header_level(buf) == 0 &&
4864 	    btrfs_check_leaf_relaxed(buf)) {
4865 		btrfs_print_leaf(buf);
4866 		ASSERT(0);
4867 	}
4868 #endif
4869 }
4870 
4871 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4872 					int flush_delayed)
4873 {
4874 	/*
4875 	 * looks as though older kernels can get into trouble with
4876 	 * this code, they end up stuck in balance_dirty_pages forever
4877 	 */
4878 	int ret;
4879 
4880 	if (current->flags & PF_MEMALLOC)
4881 		return;
4882 
4883 	if (flush_delayed)
4884 		btrfs_balance_delayed_items(fs_info);
4885 
4886 	ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4887 				     BTRFS_DIRTY_METADATA_THRESH,
4888 				     fs_info->dirty_metadata_batch);
4889 	if (ret > 0) {
4890 		balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4891 	}
4892 }
4893 
4894 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4895 {
4896 	__btrfs_btree_balance_dirty(fs_info, 1);
4897 }
4898 
4899 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4900 {
4901 	__btrfs_btree_balance_dirty(fs_info, 0);
4902 }
4903 
4904 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4905 {
4906 	/* cleanup FS via transaction */
4907 	btrfs_cleanup_transaction(fs_info);
4908 
4909 	mutex_lock(&fs_info->cleaner_mutex);
4910 	btrfs_run_delayed_iputs(fs_info);
4911 	mutex_unlock(&fs_info->cleaner_mutex);
4912 
4913 	down_write(&fs_info->cleanup_work_sem);
4914 	up_write(&fs_info->cleanup_work_sem);
4915 }
4916 
4917 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4918 {
4919 	struct btrfs_root *gang[8];
4920 	u64 root_objectid = 0;
4921 	int ret;
4922 
4923 	spin_lock(&fs_info->fs_roots_radix_lock);
4924 	while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4925 					     (void **)gang, root_objectid,
4926 					     ARRAY_SIZE(gang))) != 0) {
4927 		int i;
4928 
4929 		for (i = 0; i < ret; i++)
4930 			gang[i] = btrfs_grab_root(gang[i]);
4931 		spin_unlock(&fs_info->fs_roots_radix_lock);
4932 
4933 		for (i = 0; i < ret; i++) {
4934 			if (!gang[i])
4935 				continue;
4936 			root_objectid = gang[i]->root_key.objectid;
4937 			btrfs_free_log(NULL, gang[i]);
4938 			btrfs_put_root(gang[i]);
4939 		}
4940 		root_objectid++;
4941 		spin_lock(&fs_info->fs_roots_radix_lock);
4942 	}
4943 	spin_unlock(&fs_info->fs_roots_radix_lock);
4944 	btrfs_free_log_root_tree(NULL, fs_info);
4945 }
4946 
4947 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4948 {
4949 	struct btrfs_ordered_extent *ordered;
4950 
4951 	spin_lock(&root->ordered_extent_lock);
4952 	/*
4953 	 * This will just short circuit the ordered completion stuff which will
4954 	 * make sure the ordered extent gets properly cleaned up.
4955 	 */
4956 	list_for_each_entry(ordered, &root->ordered_extents,
4957 			    root_extent_list)
4958 		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4959 	spin_unlock(&root->ordered_extent_lock);
4960 }
4961 
4962 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4963 {
4964 	struct btrfs_root *root;
4965 	struct list_head splice;
4966 
4967 	INIT_LIST_HEAD(&splice);
4968 
4969 	spin_lock(&fs_info->ordered_root_lock);
4970 	list_splice_init(&fs_info->ordered_roots, &splice);
4971 	while (!list_empty(&splice)) {
4972 		root = list_first_entry(&splice, struct btrfs_root,
4973 					ordered_root);
4974 		list_move_tail(&root->ordered_root,
4975 			       &fs_info->ordered_roots);
4976 
4977 		spin_unlock(&fs_info->ordered_root_lock);
4978 		btrfs_destroy_ordered_extents(root);
4979 
4980 		cond_resched();
4981 		spin_lock(&fs_info->ordered_root_lock);
4982 	}
4983 	spin_unlock(&fs_info->ordered_root_lock);
4984 
4985 	/*
4986 	 * We need this here because if we've been flipped read-only we won't
4987 	 * get sync() from the umount, so we need to make sure any ordered
4988 	 * extents that haven't had their dirty pages IO start writeout yet
4989 	 * actually get run and error out properly.
4990 	 */
4991 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4992 }
4993 
4994 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4995 				      struct btrfs_fs_info *fs_info)
4996 {
4997 	struct rb_node *node;
4998 	struct btrfs_delayed_ref_root *delayed_refs;
4999 	struct btrfs_delayed_ref_node *ref;
5000 	int ret = 0;
5001 
5002 	delayed_refs = &trans->delayed_refs;
5003 
5004 	spin_lock(&delayed_refs->lock);
5005 	if (atomic_read(&delayed_refs->num_entries) == 0) {
5006 		spin_unlock(&delayed_refs->lock);
5007 		btrfs_debug(fs_info, "delayed_refs has NO entry");
5008 		return ret;
5009 	}
5010 
5011 	while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
5012 		struct btrfs_delayed_ref_head *head;
5013 		struct rb_node *n;
5014 		bool pin_bytes = false;
5015 
5016 		head = rb_entry(node, struct btrfs_delayed_ref_head,
5017 				href_node);
5018 		if (btrfs_delayed_ref_lock(delayed_refs, head))
5019 			continue;
5020 
5021 		spin_lock(&head->lock);
5022 		while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
5023 			ref = rb_entry(n, struct btrfs_delayed_ref_node,
5024 				       ref_node);
5025 			ref->in_tree = 0;
5026 			rb_erase_cached(&ref->ref_node, &head->ref_tree);
5027 			RB_CLEAR_NODE(&ref->ref_node);
5028 			if (!list_empty(&ref->add_list))
5029 				list_del(&ref->add_list);
5030 			atomic_dec(&delayed_refs->num_entries);
5031 			btrfs_put_delayed_ref(ref);
5032 		}
5033 		if (head->must_insert_reserved)
5034 			pin_bytes = true;
5035 		btrfs_free_delayed_extent_op(head->extent_op);
5036 		btrfs_delete_ref_head(delayed_refs, head);
5037 		spin_unlock(&head->lock);
5038 		spin_unlock(&delayed_refs->lock);
5039 		mutex_unlock(&head->mutex);
5040 
5041 		if (pin_bytes) {
5042 			struct btrfs_block_group *cache;
5043 
5044 			cache = btrfs_lookup_block_group(fs_info, head->bytenr);
5045 			BUG_ON(!cache);
5046 
5047 			spin_lock(&cache->space_info->lock);
5048 			spin_lock(&cache->lock);
5049 			cache->pinned += head->num_bytes;
5050 			btrfs_space_info_update_bytes_pinned(fs_info,
5051 				cache->space_info, head->num_bytes);
5052 			cache->reserved -= head->num_bytes;
5053 			cache->space_info->bytes_reserved -= head->num_bytes;
5054 			spin_unlock(&cache->lock);
5055 			spin_unlock(&cache->space_info->lock);
5056 
5057 			btrfs_put_block_group(cache);
5058 
5059 			btrfs_error_unpin_extent_range(fs_info, head->bytenr,
5060 				head->bytenr + head->num_bytes - 1);
5061 		}
5062 		btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
5063 		btrfs_put_delayed_ref_head(head);
5064 		cond_resched();
5065 		spin_lock(&delayed_refs->lock);
5066 	}
5067 	btrfs_qgroup_destroy_extent_records(trans);
5068 
5069 	spin_unlock(&delayed_refs->lock);
5070 
5071 	return ret;
5072 }
5073 
5074 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
5075 {
5076 	struct btrfs_inode *btrfs_inode;
5077 	struct list_head splice;
5078 
5079 	INIT_LIST_HEAD(&splice);
5080 
5081 	spin_lock(&root->delalloc_lock);
5082 	list_splice_init(&root->delalloc_inodes, &splice);
5083 
5084 	while (!list_empty(&splice)) {
5085 		struct inode *inode = NULL;
5086 		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
5087 					       delalloc_inodes);
5088 		__btrfs_del_delalloc_inode(root, btrfs_inode);
5089 		spin_unlock(&root->delalloc_lock);
5090 
5091 		/*
5092 		 * Make sure we get a live inode and that it'll not disappear
5093 		 * meanwhile.
5094 		 */
5095 		inode = igrab(&btrfs_inode->vfs_inode);
5096 		if (inode) {
5097 			invalidate_inode_pages2(inode->i_mapping);
5098 			iput(inode);
5099 		}
5100 		spin_lock(&root->delalloc_lock);
5101 	}
5102 	spin_unlock(&root->delalloc_lock);
5103 }
5104 
5105 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
5106 {
5107 	struct btrfs_root *root;
5108 	struct list_head splice;
5109 
5110 	INIT_LIST_HEAD(&splice);
5111 
5112 	spin_lock(&fs_info->delalloc_root_lock);
5113 	list_splice_init(&fs_info->delalloc_roots, &splice);
5114 	while (!list_empty(&splice)) {
5115 		root = list_first_entry(&splice, struct btrfs_root,
5116 					 delalloc_root);
5117 		root = btrfs_grab_root(root);
5118 		BUG_ON(!root);
5119 		spin_unlock(&fs_info->delalloc_root_lock);
5120 
5121 		btrfs_destroy_delalloc_inodes(root);
5122 		btrfs_put_root(root);
5123 
5124 		spin_lock(&fs_info->delalloc_root_lock);
5125 	}
5126 	spin_unlock(&fs_info->delalloc_root_lock);
5127 }
5128 
5129 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
5130 					struct extent_io_tree *dirty_pages,
5131 					int mark)
5132 {
5133 	int ret;
5134 	struct extent_buffer *eb;
5135 	u64 start = 0;
5136 	u64 end;
5137 
5138 	while (1) {
5139 		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
5140 					    mark, NULL);
5141 		if (ret)
5142 			break;
5143 
5144 		clear_extent_bits(dirty_pages, start, end, mark);
5145 		while (start <= end) {
5146 			eb = find_extent_buffer(fs_info, start);
5147 			start += fs_info->nodesize;
5148 			if (!eb)
5149 				continue;
5150 			wait_on_extent_buffer_writeback(eb);
5151 
5152 			if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
5153 					       &eb->bflags))
5154 				clear_extent_buffer_dirty(eb);
5155 			free_extent_buffer_stale(eb);
5156 		}
5157 	}
5158 
5159 	return ret;
5160 }
5161 
5162 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
5163 				       struct extent_io_tree *unpin)
5164 {
5165 	u64 start;
5166 	u64 end;
5167 	int ret;
5168 
5169 	while (1) {
5170 		struct extent_state *cached_state = NULL;
5171 
5172 		/*
5173 		 * The btrfs_finish_extent_commit() may get the same range as
5174 		 * ours between find_first_extent_bit and clear_extent_dirty.
5175 		 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
5176 		 * the same extent range.
5177 		 */
5178 		mutex_lock(&fs_info->unused_bg_unpin_mutex);
5179 		ret = find_first_extent_bit(unpin, 0, &start, &end,
5180 					    EXTENT_DIRTY, &cached_state);
5181 		if (ret) {
5182 			mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5183 			break;
5184 		}
5185 
5186 		clear_extent_dirty(unpin, start, end, &cached_state);
5187 		free_extent_state(cached_state);
5188 		btrfs_error_unpin_extent_range(fs_info, start, end);
5189 		mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5190 		cond_resched();
5191 	}
5192 
5193 	return 0;
5194 }
5195 
5196 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
5197 {
5198 	struct inode *inode;
5199 
5200 	inode = cache->io_ctl.inode;
5201 	if (inode) {
5202 		invalidate_inode_pages2(inode->i_mapping);
5203 		BTRFS_I(inode)->generation = 0;
5204 		cache->io_ctl.inode = NULL;
5205 		iput(inode);
5206 	}
5207 	ASSERT(cache->io_ctl.pages == NULL);
5208 	btrfs_put_block_group(cache);
5209 }
5210 
5211 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
5212 			     struct btrfs_fs_info *fs_info)
5213 {
5214 	struct btrfs_block_group *cache;
5215 
5216 	spin_lock(&cur_trans->dirty_bgs_lock);
5217 	while (!list_empty(&cur_trans->dirty_bgs)) {
5218 		cache = list_first_entry(&cur_trans->dirty_bgs,
5219 					 struct btrfs_block_group,
5220 					 dirty_list);
5221 
5222 		if (!list_empty(&cache->io_list)) {
5223 			spin_unlock(&cur_trans->dirty_bgs_lock);
5224 			list_del_init(&cache->io_list);
5225 			btrfs_cleanup_bg_io(cache);
5226 			spin_lock(&cur_trans->dirty_bgs_lock);
5227 		}
5228 
5229 		list_del_init(&cache->dirty_list);
5230 		spin_lock(&cache->lock);
5231 		cache->disk_cache_state = BTRFS_DC_ERROR;
5232 		spin_unlock(&cache->lock);
5233 
5234 		spin_unlock(&cur_trans->dirty_bgs_lock);
5235 		btrfs_put_block_group(cache);
5236 		btrfs_delayed_refs_rsv_release(fs_info, 1);
5237 		spin_lock(&cur_trans->dirty_bgs_lock);
5238 	}
5239 	spin_unlock(&cur_trans->dirty_bgs_lock);
5240 
5241 	/*
5242 	 * Refer to the definition of io_bgs member for details why it's safe
5243 	 * to use it without any locking
5244 	 */
5245 	while (!list_empty(&cur_trans->io_bgs)) {
5246 		cache = list_first_entry(&cur_trans->io_bgs,
5247 					 struct btrfs_block_group,
5248 					 io_list);
5249 
5250 		list_del_init(&cache->io_list);
5251 		spin_lock(&cache->lock);
5252 		cache->disk_cache_state = BTRFS_DC_ERROR;
5253 		spin_unlock(&cache->lock);
5254 		btrfs_cleanup_bg_io(cache);
5255 	}
5256 }
5257 
5258 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
5259 				   struct btrfs_fs_info *fs_info)
5260 {
5261 	struct btrfs_device *dev, *tmp;
5262 
5263 	btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
5264 	ASSERT(list_empty(&cur_trans->dirty_bgs));
5265 	ASSERT(list_empty(&cur_trans->io_bgs));
5266 
5267 	list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
5268 				 post_commit_list) {
5269 		list_del_init(&dev->post_commit_list);
5270 	}
5271 
5272 	btrfs_destroy_delayed_refs(cur_trans, fs_info);
5273 
5274 	cur_trans->state = TRANS_STATE_COMMIT_START;
5275 	wake_up(&fs_info->transaction_blocked_wait);
5276 
5277 	cur_trans->state = TRANS_STATE_UNBLOCKED;
5278 	wake_up(&fs_info->transaction_wait);
5279 
5280 	btrfs_destroy_delayed_inodes(fs_info);
5281 
5282 	btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
5283 				     EXTENT_DIRTY);
5284 	btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
5285 
5286 	btrfs_free_redirty_list(cur_trans);
5287 
5288 	cur_trans->state =TRANS_STATE_COMPLETED;
5289 	wake_up(&cur_trans->commit_wait);
5290 }
5291 
5292 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
5293 {
5294 	struct btrfs_transaction *t;
5295 
5296 	mutex_lock(&fs_info->transaction_kthread_mutex);
5297 
5298 	spin_lock(&fs_info->trans_lock);
5299 	while (!list_empty(&fs_info->trans_list)) {
5300 		t = list_first_entry(&fs_info->trans_list,
5301 				     struct btrfs_transaction, list);
5302 		if (t->state >= TRANS_STATE_COMMIT_START) {
5303 			refcount_inc(&t->use_count);
5304 			spin_unlock(&fs_info->trans_lock);
5305 			btrfs_wait_for_commit(fs_info, t->transid);
5306 			btrfs_put_transaction(t);
5307 			spin_lock(&fs_info->trans_lock);
5308 			continue;
5309 		}
5310 		if (t == fs_info->running_transaction) {
5311 			t->state = TRANS_STATE_COMMIT_DOING;
5312 			spin_unlock(&fs_info->trans_lock);
5313 			/*
5314 			 * We wait for 0 num_writers since we don't hold a trans
5315 			 * handle open currently for this transaction.
5316 			 */
5317 			wait_event(t->writer_wait,
5318 				   atomic_read(&t->num_writers) == 0);
5319 		} else {
5320 			spin_unlock(&fs_info->trans_lock);
5321 		}
5322 		btrfs_cleanup_one_transaction(t, fs_info);
5323 
5324 		spin_lock(&fs_info->trans_lock);
5325 		if (t == fs_info->running_transaction)
5326 			fs_info->running_transaction = NULL;
5327 		list_del_init(&t->list);
5328 		spin_unlock(&fs_info->trans_lock);
5329 
5330 		btrfs_put_transaction(t);
5331 		trace_btrfs_transaction_commit(fs_info);
5332 		spin_lock(&fs_info->trans_lock);
5333 	}
5334 	spin_unlock(&fs_info->trans_lock);
5335 	btrfs_destroy_all_ordered_extents(fs_info);
5336 	btrfs_destroy_delayed_inodes(fs_info);
5337 	btrfs_assert_delayed_root_empty(fs_info);
5338 	btrfs_destroy_all_delalloc_inodes(fs_info);
5339 	btrfs_drop_all_logs(fs_info);
5340 	mutex_unlock(&fs_info->transaction_kthread_mutex);
5341 
5342 	return 0;
5343 }
5344 
5345 int btrfs_init_root_free_objectid(struct btrfs_root *root)
5346 {
5347 	struct btrfs_path *path;
5348 	int ret;
5349 	struct extent_buffer *l;
5350 	struct btrfs_key search_key;
5351 	struct btrfs_key found_key;
5352 	int slot;
5353 
5354 	path = btrfs_alloc_path();
5355 	if (!path)
5356 		return -ENOMEM;
5357 
5358 	search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
5359 	search_key.type = -1;
5360 	search_key.offset = (u64)-1;
5361 	ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5362 	if (ret < 0)
5363 		goto error;
5364 	BUG_ON(ret == 0); /* Corruption */
5365 	if (path->slots[0] > 0) {
5366 		slot = path->slots[0] - 1;
5367 		l = path->nodes[0];
5368 		btrfs_item_key_to_cpu(l, &found_key, slot);
5369 		root->free_objectid = max_t(u64, found_key.objectid + 1,
5370 					    BTRFS_FIRST_FREE_OBJECTID);
5371 	} else {
5372 		root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5373 	}
5374 	ret = 0;
5375 error:
5376 	btrfs_free_path(path);
5377 	return ret;
5378 }
5379 
5380 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5381 {
5382 	int ret;
5383 	mutex_lock(&root->objectid_mutex);
5384 
5385 	if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5386 		btrfs_warn(root->fs_info,
5387 			   "the objectid of root %llu reaches its highest value",
5388 			   root->root_key.objectid);
5389 		ret = -ENOSPC;
5390 		goto out;
5391 	}
5392 
5393 	*objectid = root->free_objectid++;
5394 	ret = 0;
5395 out:
5396 	mutex_unlock(&root->objectid_mutex);
5397 	return ret;
5398 }
5399