xref: /linux/fs/btrfs/disk-io.c (revision ea8a163e02d6925773129e2dd86e419e491b791d)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5 
6 #include <linux/fs.h>
7 #include <linux/blkdev.h>
8 #include <linux/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/workqueue.h>
11 #include <linux/kthread.h>
12 #include <linux/slab.h>
13 #include <linux/migrate.h>
14 #include <linux/ratelimit.h>
15 #include <linux/uuid.h>
16 #include <linux/semaphore.h>
17 #include <linux/error-injection.h>
18 #include <linux/crc32c.h>
19 #include <linux/sched/mm.h>
20 #include <asm/unaligned.h>
21 #include <crypto/hash.h>
22 #include "ctree.h"
23 #include "disk-io.h"
24 #include "transaction.h"
25 #include "btrfs_inode.h"
26 #include "volumes.h"
27 #include "print-tree.h"
28 #include "locking.h"
29 #include "tree-log.h"
30 #include "free-space-cache.h"
31 #include "free-space-tree.h"
32 #include "check-integrity.h"
33 #include "rcu-string.h"
34 #include "dev-replace.h"
35 #include "raid56.h"
36 #include "sysfs.h"
37 #include "qgroup.h"
38 #include "compression.h"
39 #include "tree-checker.h"
40 #include "ref-verify.h"
41 #include "block-group.h"
42 #include "discard.h"
43 #include "space-info.h"
44 #include "zoned.h"
45 #include "subpage.h"
46 
47 #define BTRFS_SUPER_FLAG_SUPP	(BTRFS_HEADER_FLAG_WRITTEN |\
48 				 BTRFS_HEADER_FLAG_RELOC |\
49 				 BTRFS_SUPER_FLAG_ERROR |\
50 				 BTRFS_SUPER_FLAG_SEEDING |\
51 				 BTRFS_SUPER_FLAG_METADUMP |\
52 				 BTRFS_SUPER_FLAG_METADUMP_V2)
53 
54 static void end_workqueue_fn(struct btrfs_work *work);
55 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
56 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
57 				      struct btrfs_fs_info *fs_info);
58 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
59 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
60 					struct extent_io_tree *dirty_pages,
61 					int mark);
62 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
63 				       struct extent_io_tree *pinned_extents);
64 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
65 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
66 
67 /*
68  * btrfs_end_io_wq structs are used to do processing in task context when an IO
69  * is complete.  This is used during reads to verify checksums, and it is used
70  * by writes to insert metadata for new file extents after IO is complete.
71  */
72 struct btrfs_end_io_wq {
73 	struct bio *bio;
74 	bio_end_io_t *end_io;
75 	void *private;
76 	struct btrfs_fs_info *info;
77 	blk_status_t status;
78 	enum btrfs_wq_endio_type metadata;
79 	struct btrfs_work work;
80 };
81 
82 static struct kmem_cache *btrfs_end_io_wq_cache;
83 
84 int __init btrfs_end_io_wq_init(void)
85 {
86 	btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
87 					sizeof(struct btrfs_end_io_wq),
88 					0,
89 					SLAB_MEM_SPREAD,
90 					NULL);
91 	if (!btrfs_end_io_wq_cache)
92 		return -ENOMEM;
93 	return 0;
94 }
95 
96 void __cold btrfs_end_io_wq_exit(void)
97 {
98 	kmem_cache_destroy(btrfs_end_io_wq_cache);
99 }
100 
101 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
102 {
103 	if (fs_info->csum_shash)
104 		crypto_free_shash(fs_info->csum_shash);
105 }
106 
107 /*
108  * async submit bios are used to offload expensive checksumming
109  * onto the worker threads.  They checksum file and metadata bios
110  * just before they are sent down the IO stack.
111  */
112 struct async_submit_bio {
113 	struct inode *inode;
114 	struct bio *bio;
115 	extent_submit_bio_start_t *submit_bio_start;
116 	int mirror_num;
117 
118 	/* Optional parameter for submit_bio_start used by direct io */
119 	u64 dio_file_offset;
120 	struct btrfs_work work;
121 	blk_status_t status;
122 };
123 
124 /*
125  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
126  * eb, the lockdep key is determined by the btrfs_root it belongs to and
127  * the level the eb occupies in the tree.
128  *
129  * Different roots are used for different purposes and may nest inside each
130  * other and they require separate keysets.  As lockdep keys should be
131  * static, assign keysets according to the purpose of the root as indicated
132  * by btrfs_root->root_key.objectid.  This ensures that all special purpose
133  * roots have separate keysets.
134  *
135  * Lock-nesting across peer nodes is always done with the immediate parent
136  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
137  * subclass to avoid triggering lockdep warning in such cases.
138  *
139  * The key is set by the readpage_end_io_hook after the buffer has passed
140  * csum validation but before the pages are unlocked.  It is also set by
141  * btrfs_init_new_buffer on freshly allocated blocks.
142  *
143  * We also add a check to make sure the highest level of the tree is the
144  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
145  * needs update as well.
146  */
147 #ifdef CONFIG_DEBUG_LOCK_ALLOC
148 # if BTRFS_MAX_LEVEL != 8
149 #  error
150 # endif
151 
152 #define DEFINE_LEVEL(stem, level)					\
153 	.names[level] = "btrfs-" stem "-0" #level,
154 
155 #define DEFINE_NAME(stem)						\
156 	DEFINE_LEVEL(stem, 0)						\
157 	DEFINE_LEVEL(stem, 1)						\
158 	DEFINE_LEVEL(stem, 2)						\
159 	DEFINE_LEVEL(stem, 3)						\
160 	DEFINE_LEVEL(stem, 4)						\
161 	DEFINE_LEVEL(stem, 5)						\
162 	DEFINE_LEVEL(stem, 6)						\
163 	DEFINE_LEVEL(stem, 7)
164 
165 static struct btrfs_lockdep_keyset {
166 	u64			id;		/* root objectid */
167 	/* Longest entry: btrfs-free-space-00 */
168 	char			names[BTRFS_MAX_LEVEL][20];
169 	struct lock_class_key	keys[BTRFS_MAX_LEVEL];
170 } btrfs_lockdep_keysets[] = {
171 	{ .id = BTRFS_ROOT_TREE_OBJECTID,	DEFINE_NAME("root")	},
172 	{ .id = BTRFS_EXTENT_TREE_OBJECTID,	DEFINE_NAME("extent")	},
173 	{ .id = BTRFS_CHUNK_TREE_OBJECTID,	DEFINE_NAME("chunk")	},
174 	{ .id = BTRFS_DEV_TREE_OBJECTID,	DEFINE_NAME("dev")	},
175 	{ .id = BTRFS_CSUM_TREE_OBJECTID,	DEFINE_NAME("csum")	},
176 	{ .id = BTRFS_QUOTA_TREE_OBJECTID,	DEFINE_NAME("quota")	},
177 	{ .id = BTRFS_TREE_LOG_OBJECTID,	DEFINE_NAME("log")	},
178 	{ .id = BTRFS_TREE_RELOC_OBJECTID,	DEFINE_NAME("treloc")	},
179 	{ .id = BTRFS_DATA_RELOC_TREE_OBJECTID,	DEFINE_NAME("dreloc")	},
180 	{ .id = BTRFS_UUID_TREE_OBJECTID,	DEFINE_NAME("uuid")	},
181 	{ .id = BTRFS_FREE_SPACE_TREE_OBJECTID,	DEFINE_NAME("free-space") },
182 	{ .id = 0,				DEFINE_NAME("tree")	},
183 };
184 
185 #undef DEFINE_LEVEL
186 #undef DEFINE_NAME
187 
188 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
189 				    int level)
190 {
191 	struct btrfs_lockdep_keyset *ks;
192 
193 	BUG_ON(level >= ARRAY_SIZE(ks->keys));
194 
195 	/* find the matching keyset, id 0 is the default entry */
196 	for (ks = btrfs_lockdep_keysets; ks->id; ks++)
197 		if (ks->id == objectid)
198 			break;
199 
200 	lockdep_set_class_and_name(&eb->lock,
201 				   &ks->keys[level], ks->names[level]);
202 }
203 
204 #endif
205 
206 /*
207  * Compute the csum of a btree block and store the result to provided buffer.
208  */
209 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
210 {
211 	struct btrfs_fs_info *fs_info = buf->fs_info;
212 	const int num_pages = num_extent_pages(buf);
213 	const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
214 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
215 	char *kaddr;
216 	int i;
217 
218 	shash->tfm = fs_info->csum_shash;
219 	crypto_shash_init(shash);
220 	kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
221 	crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
222 			    first_page_part - BTRFS_CSUM_SIZE);
223 
224 	for (i = 1; i < num_pages; i++) {
225 		kaddr = page_address(buf->pages[i]);
226 		crypto_shash_update(shash, kaddr, PAGE_SIZE);
227 	}
228 	memset(result, 0, BTRFS_CSUM_SIZE);
229 	crypto_shash_final(shash, result);
230 }
231 
232 /*
233  * we can't consider a given block up to date unless the transid of the
234  * block matches the transid in the parent node's pointer.  This is how we
235  * detect blocks that either didn't get written at all or got written
236  * in the wrong place.
237  */
238 static int verify_parent_transid(struct extent_io_tree *io_tree,
239 				 struct extent_buffer *eb, u64 parent_transid,
240 				 int atomic)
241 {
242 	struct extent_state *cached_state = NULL;
243 	int ret;
244 
245 	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
246 		return 0;
247 
248 	if (atomic)
249 		return -EAGAIN;
250 
251 	lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
252 			 &cached_state);
253 	if (extent_buffer_uptodate(eb) &&
254 	    btrfs_header_generation(eb) == parent_transid) {
255 		ret = 0;
256 		goto out;
257 	}
258 	btrfs_err_rl(eb->fs_info,
259 		"parent transid verify failed on %llu wanted %llu found %llu",
260 			eb->start,
261 			parent_transid, btrfs_header_generation(eb));
262 	ret = 1;
263 	clear_extent_buffer_uptodate(eb);
264 out:
265 	unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
266 			     &cached_state);
267 	return ret;
268 }
269 
270 static bool btrfs_supported_super_csum(u16 csum_type)
271 {
272 	switch (csum_type) {
273 	case BTRFS_CSUM_TYPE_CRC32:
274 	case BTRFS_CSUM_TYPE_XXHASH:
275 	case BTRFS_CSUM_TYPE_SHA256:
276 	case BTRFS_CSUM_TYPE_BLAKE2:
277 		return true;
278 	default:
279 		return false;
280 	}
281 }
282 
283 /*
284  * Return 0 if the superblock checksum type matches the checksum value of that
285  * algorithm. Pass the raw disk superblock data.
286  */
287 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
288 				  char *raw_disk_sb)
289 {
290 	struct btrfs_super_block *disk_sb =
291 		(struct btrfs_super_block *)raw_disk_sb;
292 	char result[BTRFS_CSUM_SIZE];
293 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
294 
295 	shash->tfm = fs_info->csum_shash;
296 
297 	/*
298 	 * The super_block structure does not span the whole
299 	 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
300 	 * filled with zeros and is included in the checksum.
301 	 */
302 	crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
303 			    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
304 
305 	if (memcmp(disk_sb->csum, result, fs_info->csum_size))
306 		return 1;
307 
308 	return 0;
309 }
310 
311 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
312 			   struct btrfs_key *first_key, u64 parent_transid)
313 {
314 	struct btrfs_fs_info *fs_info = eb->fs_info;
315 	int found_level;
316 	struct btrfs_key found_key;
317 	int ret;
318 
319 	found_level = btrfs_header_level(eb);
320 	if (found_level != level) {
321 		WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
322 		     KERN_ERR "BTRFS: tree level check failed\n");
323 		btrfs_err(fs_info,
324 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
325 			  eb->start, level, found_level);
326 		return -EIO;
327 	}
328 
329 	if (!first_key)
330 		return 0;
331 
332 	/*
333 	 * For live tree block (new tree blocks in current transaction),
334 	 * we need proper lock context to avoid race, which is impossible here.
335 	 * So we only checks tree blocks which is read from disk, whose
336 	 * generation <= fs_info->last_trans_committed.
337 	 */
338 	if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
339 		return 0;
340 
341 	/* We have @first_key, so this @eb must have at least one item */
342 	if (btrfs_header_nritems(eb) == 0) {
343 		btrfs_err(fs_info,
344 		"invalid tree nritems, bytenr=%llu nritems=0 expect >0",
345 			  eb->start);
346 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
347 		return -EUCLEAN;
348 	}
349 
350 	if (found_level)
351 		btrfs_node_key_to_cpu(eb, &found_key, 0);
352 	else
353 		btrfs_item_key_to_cpu(eb, &found_key, 0);
354 	ret = btrfs_comp_cpu_keys(first_key, &found_key);
355 
356 	if (ret) {
357 		WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
358 		     KERN_ERR "BTRFS: tree first key check failed\n");
359 		btrfs_err(fs_info,
360 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
361 			  eb->start, parent_transid, first_key->objectid,
362 			  first_key->type, first_key->offset,
363 			  found_key.objectid, found_key.type,
364 			  found_key.offset);
365 	}
366 	return ret;
367 }
368 
369 /*
370  * helper to read a given tree block, doing retries as required when
371  * the checksums don't match and we have alternate mirrors to try.
372  *
373  * @parent_transid:	expected transid, skip check if 0
374  * @level:		expected level, mandatory check
375  * @first_key:		expected key of first slot, skip check if NULL
376  */
377 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
378 					  u64 parent_transid, int level,
379 					  struct btrfs_key *first_key)
380 {
381 	struct btrfs_fs_info *fs_info = eb->fs_info;
382 	struct extent_io_tree *io_tree;
383 	int failed = 0;
384 	int ret;
385 	int num_copies = 0;
386 	int mirror_num = 0;
387 	int failed_mirror = 0;
388 
389 	io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
390 	while (1) {
391 		clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
392 		ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
393 		if (!ret) {
394 			if (verify_parent_transid(io_tree, eb,
395 						   parent_transid, 0))
396 				ret = -EIO;
397 			else if (btrfs_verify_level_key(eb, level,
398 						first_key, parent_transid))
399 				ret = -EUCLEAN;
400 			else
401 				break;
402 		}
403 
404 		num_copies = btrfs_num_copies(fs_info,
405 					      eb->start, eb->len);
406 		if (num_copies == 1)
407 			break;
408 
409 		if (!failed_mirror) {
410 			failed = 1;
411 			failed_mirror = eb->read_mirror;
412 		}
413 
414 		mirror_num++;
415 		if (mirror_num == failed_mirror)
416 			mirror_num++;
417 
418 		if (mirror_num > num_copies)
419 			break;
420 	}
421 
422 	if (failed && !ret && failed_mirror)
423 		btrfs_repair_eb_io_failure(eb, failed_mirror);
424 
425 	return ret;
426 }
427 
428 static int csum_one_extent_buffer(struct extent_buffer *eb)
429 {
430 	struct btrfs_fs_info *fs_info = eb->fs_info;
431 	u8 result[BTRFS_CSUM_SIZE];
432 	int ret;
433 
434 	ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
435 				    offsetof(struct btrfs_header, fsid),
436 				    BTRFS_FSID_SIZE) == 0);
437 	csum_tree_block(eb, result);
438 
439 	if (btrfs_header_level(eb))
440 		ret = btrfs_check_node(eb);
441 	else
442 		ret = btrfs_check_leaf_full(eb);
443 
444 	if (ret < 0) {
445 		btrfs_print_tree(eb, 0);
446 		btrfs_err(fs_info,
447 			"block=%llu write time tree block corruption detected",
448 			eb->start);
449 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
450 		return ret;
451 	}
452 	write_extent_buffer(eb, result, 0, fs_info->csum_size);
453 
454 	return 0;
455 }
456 
457 /* Checksum all dirty extent buffers in one bio_vec */
458 static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
459 				      struct bio_vec *bvec)
460 {
461 	struct page *page = bvec->bv_page;
462 	u64 bvec_start = page_offset(page) + bvec->bv_offset;
463 	u64 cur;
464 	int ret = 0;
465 
466 	for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
467 	     cur += fs_info->nodesize) {
468 		struct extent_buffer *eb;
469 		bool uptodate;
470 
471 		eb = find_extent_buffer(fs_info, cur);
472 		uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
473 						       fs_info->nodesize);
474 
475 		/* A dirty eb shouldn't disappear from buffer_radix */
476 		if (WARN_ON(!eb))
477 			return -EUCLEAN;
478 
479 		if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
480 			free_extent_buffer(eb);
481 			return -EUCLEAN;
482 		}
483 		if (WARN_ON(!uptodate)) {
484 			free_extent_buffer(eb);
485 			return -EUCLEAN;
486 		}
487 
488 		ret = csum_one_extent_buffer(eb);
489 		free_extent_buffer(eb);
490 		if (ret < 0)
491 			return ret;
492 	}
493 	return ret;
494 }
495 
496 /*
497  * Checksum a dirty tree block before IO.  This has extra checks to make sure
498  * we only fill in the checksum field in the first page of a multi-page block.
499  * For subpage extent buffers we need bvec to also read the offset in the page.
500  */
501 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
502 {
503 	struct page *page = bvec->bv_page;
504 	u64 start = page_offset(page);
505 	u64 found_start;
506 	struct extent_buffer *eb;
507 
508 	if (fs_info->sectorsize < PAGE_SIZE)
509 		return csum_dirty_subpage_buffers(fs_info, bvec);
510 
511 	eb = (struct extent_buffer *)page->private;
512 	if (page != eb->pages[0])
513 		return 0;
514 
515 	found_start = btrfs_header_bytenr(eb);
516 
517 	if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
518 		WARN_ON(found_start != 0);
519 		return 0;
520 	}
521 
522 	/*
523 	 * Please do not consolidate these warnings into a single if.
524 	 * It is useful to know what went wrong.
525 	 */
526 	if (WARN_ON(found_start != start))
527 		return -EUCLEAN;
528 	if (WARN_ON(!PageUptodate(page)))
529 		return -EUCLEAN;
530 
531 	return csum_one_extent_buffer(eb);
532 }
533 
534 static int check_tree_block_fsid(struct extent_buffer *eb)
535 {
536 	struct btrfs_fs_info *fs_info = eb->fs_info;
537 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
538 	u8 fsid[BTRFS_FSID_SIZE];
539 	u8 *metadata_uuid;
540 
541 	read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
542 			   BTRFS_FSID_SIZE);
543 	/*
544 	 * Checking the incompat flag is only valid for the current fs. For
545 	 * seed devices it's forbidden to have their uuid changed so reading
546 	 * ->fsid in this case is fine
547 	 */
548 	if (btrfs_fs_incompat(fs_info, METADATA_UUID))
549 		metadata_uuid = fs_devices->metadata_uuid;
550 	else
551 		metadata_uuid = fs_devices->fsid;
552 
553 	if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
554 		return 0;
555 
556 	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
557 		if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
558 			return 0;
559 
560 	return 1;
561 }
562 
563 /* Do basic extent buffer checks at read time */
564 static int validate_extent_buffer(struct extent_buffer *eb)
565 {
566 	struct btrfs_fs_info *fs_info = eb->fs_info;
567 	u64 found_start;
568 	const u32 csum_size = fs_info->csum_size;
569 	u8 found_level;
570 	u8 result[BTRFS_CSUM_SIZE];
571 	const u8 *header_csum;
572 	int ret = 0;
573 
574 	found_start = btrfs_header_bytenr(eb);
575 	if (found_start != eb->start) {
576 		btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
577 			     eb->start, found_start);
578 		ret = -EIO;
579 		goto out;
580 	}
581 	if (check_tree_block_fsid(eb)) {
582 		btrfs_err_rl(fs_info, "bad fsid on block %llu",
583 			     eb->start);
584 		ret = -EIO;
585 		goto out;
586 	}
587 	found_level = btrfs_header_level(eb);
588 	if (found_level >= BTRFS_MAX_LEVEL) {
589 		btrfs_err(fs_info, "bad tree block level %d on %llu",
590 			  (int)btrfs_header_level(eb), eb->start);
591 		ret = -EIO;
592 		goto out;
593 	}
594 
595 	csum_tree_block(eb, result);
596 	header_csum = page_address(eb->pages[0]) +
597 		get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
598 
599 	if (memcmp(result, header_csum, csum_size) != 0) {
600 		btrfs_warn_rl(fs_info,
601 	"checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
602 			      eb->start,
603 			      CSUM_FMT_VALUE(csum_size, header_csum),
604 			      CSUM_FMT_VALUE(csum_size, result),
605 			      btrfs_header_level(eb));
606 		ret = -EUCLEAN;
607 		goto out;
608 	}
609 
610 	/*
611 	 * If this is a leaf block and it is corrupt, set the corrupt bit so
612 	 * that we don't try and read the other copies of this block, just
613 	 * return -EIO.
614 	 */
615 	if (found_level == 0 && btrfs_check_leaf_full(eb)) {
616 		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
617 		ret = -EIO;
618 	}
619 
620 	if (found_level > 0 && btrfs_check_node(eb))
621 		ret = -EIO;
622 
623 	if (!ret)
624 		set_extent_buffer_uptodate(eb);
625 	else
626 		btrfs_err(fs_info,
627 			  "block=%llu read time tree block corruption detected",
628 			  eb->start);
629 out:
630 	return ret;
631 }
632 
633 static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
634 				   int mirror)
635 {
636 	struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
637 	struct extent_buffer *eb;
638 	bool reads_done;
639 	int ret = 0;
640 
641 	/*
642 	 * We don't allow bio merge for subpage metadata read, so we should
643 	 * only get one eb for each endio hook.
644 	 */
645 	ASSERT(end == start + fs_info->nodesize - 1);
646 	ASSERT(PagePrivate(page));
647 
648 	eb = find_extent_buffer(fs_info, start);
649 	/*
650 	 * When we are reading one tree block, eb must have been inserted into
651 	 * the radix tree. If not, something is wrong.
652 	 */
653 	ASSERT(eb);
654 
655 	reads_done = atomic_dec_and_test(&eb->io_pages);
656 	/* Subpage read must finish in page read */
657 	ASSERT(reads_done);
658 
659 	eb->read_mirror = mirror;
660 	if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
661 		ret = -EIO;
662 		goto err;
663 	}
664 	ret = validate_extent_buffer(eb);
665 	if (ret < 0)
666 		goto err;
667 
668 	if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
669 		btree_readahead_hook(eb, ret);
670 
671 	set_extent_buffer_uptodate(eb);
672 
673 	free_extent_buffer(eb);
674 	return ret;
675 err:
676 	/*
677 	 * end_bio_extent_readpage decrements io_pages in case of error,
678 	 * make sure it has something to decrement.
679 	 */
680 	atomic_inc(&eb->io_pages);
681 	clear_extent_buffer_uptodate(eb);
682 	free_extent_buffer(eb);
683 	return ret;
684 }
685 
686 int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio,
687 				   struct page *page, u64 start, u64 end,
688 				   int mirror)
689 {
690 	struct extent_buffer *eb;
691 	int ret = 0;
692 	int reads_done;
693 
694 	ASSERT(page->private);
695 
696 	if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE)
697 		return validate_subpage_buffer(page, start, end, mirror);
698 
699 	eb = (struct extent_buffer *)page->private;
700 
701 	/*
702 	 * The pending IO might have been the only thing that kept this buffer
703 	 * in memory.  Make sure we have a ref for all this other checks
704 	 */
705 	atomic_inc(&eb->refs);
706 
707 	reads_done = atomic_dec_and_test(&eb->io_pages);
708 	if (!reads_done)
709 		goto err;
710 
711 	eb->read_mirror = mirror;
712 	if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
713 		ret = -EIO;
714 		goto err;
715 	}
716 	ret = validate_extent_buffer(eb);
717 err:
718 	if (reads_done &&
719 	    test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
720 		btree_readahead_hook(eb, ret);
721 
722 	if (ret) {
723 		/*
724 		 * our io error hook is going to dec the io pages
725 		 * again, we have to make sure it has something
726 		 * to decrement
727 		 */
728 		atomic_inc(&eb->io_pages);
729 		clear_extent_buffer_uptodate(eb);
730 	}
731 	free_extent_buffer(eb);
732 
733 	return ret;
734 }
735 
736 static void end_workqueue_bio(struct bio *bio)
737 {
738 	struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
739 	struct btrfs_fs_info *fs_info;
740 	struct btrfs_workqueue *wq;
741 
742 	fs_info = end_io_wq->info;
743 	end_io_wq->status = bio->bi_status;
744 
745 	if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
746 		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
747 			wq = fs_info->endio_meta_write_workers;
748 		else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
749 			wq = fs_info->endio_freespace_worker;
750 		else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
751 			wq = fs_info->endio_raid56_workers;
752 		else
753 			wq = fs_info->endio_write_workers;
754 	} else {
755 		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
756 			wq = fs_info->endio_raid56_workers;
757 		else if (end_io_wq->metadata)
758 			wq = fs_info->endio_meta_workers;
759 		else
760 			wq = fs_info->endio_workers;
761 	}
762 
763 	btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
764 	btrfs_queue_work(wq, &end_io_wq->work);
765 }
766 
767 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
768 			enum btrfs_wq_endio_type metadata)
769 {
770 	struct btrfs_end_io_wq *end_io_wq;
771 
772 	end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
773 	if (!end_io_wq)
774 		return BLK_STS_RESOURCE;
775 
776 	end_io_wq->private = bio->bi_private;
777 	end_io_wq->end_io = bio->bi_end_io;
778 	end_io_wq->info = info;
779 	end_io_wq->status = 0;
780 	end_io_wq->bio = bio;
781 	end_io_wq->metadata = metadata;
782 
783 	bio->bi_private = end_io_wq;
784 	bio->bi_end_io = end_workqueue_bio;
785 	return 0;
786 }
787 
788 static void run_one_async_start(struct btrfs_work *work)
789 {
790 	struct async_submit_bio *async;
791 	blk_status_t ret;
792 
793 	async = container_of(work, struct  async_submit_bio, work);
794 	ret = async->submit_bio_start(async->inode, async->bio,
795 				      async->dio_file_offset);
796 	if (ret)
797 		async->status = ret;
798 }
799 
800 /*
801  * In order to insert checksums into the metadata in large chunks, we wait
802  * until bio submission time.   All the pages in the bio are checksummed and
803  * sums are attached onto the ordered extent record.
804  *
805  * At IO completion time the csums attached on the ordered extent record are
806  * inserted into the tree.
807  */
808 static void run_one_async_done(struct btrfs_work *work)
809 {
810 	struct async_submit_bio *async;
811 	struct inode *inode;
812 	blk_status_t ret;
813 
814 	async = container_of(work, struct  async_submit_bio, work);
815 	inode = async->inode;
816 
817 	/* If an error occurred we just want to clean up the bio and move on */
818 	if (async->status) {
819 		async->bio->bi_status = async->status;
820 		bio_endio(async->bio);
821 		return;
822 	}
823 
824 	/*
825 	 * All of the bios that pass through here are from async helpers.
826 	 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
827 	 * This changes nothing when cgroups aren't in use.
828 	 */
829 	async->bio->bi_opf |= REQ_CGROUP_PUNT;
830 	ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
831 	if (ret) {
832 		async->bio->bi_status = ret;
833 		bio_endio(async->bio);
834 	}
835 }
836 
837 static void run_one_async_free(struct btrfs_work *work)
838 {
839 	struct async_submit_bio *async;
840 
841 	async = container_of(work, struct  async_submit_bio, work);
842 	kfree(async);
843 }
844 
845 blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio,
846 				 int mirror_num, unsigned long bio_flags,
847 				 u64 dio_file_offset,
848 				 extent_submit_bio_start_t *submit_bio_start)
849 {
850 	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
851 	struct async_submit_bio *async;
852 
853 	async = kmalloc(sizeof(*async), GFP_NOFS);
854 	if (!async)
855 		return BLK_STS_RESOURCE;
856 
857 	async->inode = inode;
858 	async->bio = bio;
859 	async->mirror_num = mirror_num;
860 	async->submit_bio_start = submit_bio_start;
861 
862 	btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
863 			run_one_async_free);
864 
865 	async->dio_file_offset = dio_file_offset;
866 
867 	async->status = 0;
868 
869 	if (op_is_sync(bio->bi_opf))
870 		btrfs_set_work_high_priority(&async->work);
871 
872 	btrfs_queue_work(fs_info->workers, &async->work);
873 	return 0;
874 }
875 
876 static blk_status_t btree_csum_one_bio(struct bio *bio)
877 {
878 	struct bio_vec *bvec;
879 	struct btrfs_root *root;
880 	int ret = 0;
881 	struct bvec_iter_all iter_all;
882 
883 	ASSERT(!bio_flagged(bio, BIO_CLONED));
884 	bio_for_each_segment_all(bvec, bio, iter_all) {
885 		root = BTRFS_I(bvec->bv_page->mapping->host)->root;
886 		ret = csum_dirty_buffer(root->fs_info, bvec);
887 		if (ret)
888 			break;
889 	}
890 
891 	return errno_to_blk_status(ret);
892 }
893 
894 static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
895 					   u64 dio_file_offset)
896 {
897 	/*
898 	 * when we're called for a write, we're already in the async
899 	 * submission context.  Just jump into btrfs_map_bio
900 	 */
901 	return btree_csum_one_bio(bio);
902 }
903 
904 static bool should_async_write(struct btrfs_fs_info *fs_info,
905 			     struct btrfs_inode *bi)
906 {
907 	if (btrfs_is_zoned(fs_info))
908 		return false;
909 	if (atomic_read(&bi->sync_writers))
910 		return false;
911 	if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
912 		return false;
913 	return true;
914 }
915 
916 blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio,
917 				       int mirror_num, unsigned long bio_flags)
918 {
919 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
920 	blk_status_t ret;
921 
922 	if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
923 		/*
924 		 * called for a read, do the setup so that checksum validation
925 		 * can happen in the async kernel threads
926 		 */
927 		ret = btrfs_bio_wq_end_io(fs_info, bio,
928 					  BTRFS_WQ_ENDIO_METADATA);
929 		if (ret)
930 			goto out_w_error;
931 		ret = btrfs_map_bio(fs_info, bio, mirror_num);
932 	} else if (!should_async_write(fs_info, BTRFS_I(inode))) {
933 		ret = btree_csum_one_bio(bio);
934 		if (ret)
935 			goto out_w_error;
936 		ret = btrfs_map_bio(fs_info, bio, mirror_num);
937 	} else {
938 		/*
939 		 * kthread helpers are used to submit writes so that
940 		 * checksumming can happen in parallel across all CPUs
941 		 */
942 		ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
943 					  0, btree_submit_bio_start);
944 	}
945 
946 	if (ret)
947 		goto out_w_error;
948 	return 0;
949 
950 out_w_error:
951 	bio->bi_status = ret;
952 	bio_endio(bio);
953 	return ret;
954 }
955 
956 #ifdef CONFIG_MIGRATION
957 static int btree_migratepage(struct address_space *mapping,
958 			struct page *newpage, struct page *page,
959 			enum migrate_mode mode)
960 {
961 	/*
962 	 * we can't safely write a btree page from here,
963 	 * we haven't done the locking hook
964 	 */
965 	if (PageDirty(page))
966 		return -EAGAIN;
967 	/*
968 	 * Buffers may be managed in a filesystem specific way.
969 	 * We must have no buffers or drop them.
970 	 */
971 	if (page_has_private(page) &&
972 	    !try_to_release_page(page, GFP_KERNEL))
973 		return -EAGAIN;
974 	return migrate_page(mapping, newpage, page, mode);
975 }
976 #endif
977 
978 
979 static int btree_writepages(struct address_space *mapping,
980 			    struct writeback_control *wbc)
981 {
982 	struct btrfs_fs_info *fs_info;
983 	int ret;
984 
985 	if (wbc->sync_mode == WB_SYNC_NONE) {
986 
987 		if (wbc->for_kupdate)
988 			return 0;
989 
990 		fs_info = BTRFS_I(mapping->host)->root->fs_info;
991 		/* this is a bit racy, but that's ok */
992 		ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
993 					     BTRFS_DIRTY_METADATA_THRESH,
994 					     fs_info->dirty_metadata_batch);
995 		if (ret < 0)
996 			return 0;
997 	}
998 	return btree_write_cache_pages(mapping, wbc);
999 }
1000 
1001 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1002 {
1003 	if (PageWriteback(page) || PageDirty(page))
1004 		return 0;
1005 
1006 	return try_release_extent_buffer(page);
1007 }
1008 
1009 static void btree_invalidatepage(struct page *page, unsigned int offset,
1010 				 unsigned int length)
1011 {
1012 	struct extent_io_tree *tree;
1013 	tree = &BTRFS_I(page->mapping->host)->io_tree;
1014 	extent_invalidatepage(tree, page, offset);
1015 	btree_releasepage(page, GFP_NOFS);
1016 	if (PagePrivate(page)) {
1017 		btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1018 			   "page private not zero on page %llu",
1019 			   (unsigned long long)page_offset(page));
1020 		detach_page_private(page);
1021 	}
1022 }
1023 
1024 static int btree_set_page_dirty(struct page *page)
1025 {
1026 #ifdef DEBUG
1027 	struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
1028 	struct btrfs_subpage *subpage;
1029 	struct extent_buffer *eb;
1030 	int cur_bit = 0;
1031 	u64 page_start = page_offset(page);
1032 
1033 	if (fs_info->sectorsize == PAGE_SIZE) {
1034 		BUG_ON(!PagePrivate(page));
1035 		eb = (struct extent_buffer *)page->private;
1036 		BUG_ON(!eb);
1037 		BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1038 		BUG_ON(!atomic_read(&eb->refs));
1039 		btrfs_assert_tree_write_locked(eb);
1040 		return __set_page_dirty_nobuffers(page);
1041 	}
1042 	ASSERT(PagePrivate(page) && page->private);
1043 	subpage = (struct btrfs_subpage *)page->private;
1044 
1045 	ASSERT(subpage->dirty_bitmap);
1046 	while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
1047 		unsigned long flags;
1048 		u64 cur;
1049 		u16 tmp = (1 << cur_bit);
1050 
1051 		spin_lock_irqsave(&subpage->lock, flags);
1052 		if (!(tmp & subpage->dirty_bitmap)) {
1053 			spin_unlock_irqrestore(&subpage->lock, flags);
1054 			cur_bit++;
1055 			continue;
1056 		}
1057 		spin_unlock_irqrestore(&subpage->lock, flags);
1058 		cur = page_start + cur_bit * fs_info->sectorsize;
1059 
1060 		eb = find_extent_buffer(fs_info, cur);
1061 		ASSERT(eb);
1062 		ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1063 		ASSERT(atomic_read(&eb->refs));
1064 		btrfs_assert_tree_write_locked(eb);
1065 		free_extent_buffer(eb);
1066 
1067 		cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
1068 	}
1069 #endif
1070 	return __set_page_dirty_nobuffers(page);
1071 }
1072 
1073 static const struct address_space_operations btree_aops = {
1074 	.writepages	= btree_writepages,
1075 	.releasepage	= btree_releasepage,
1076 	.invalidatepage = btree_invalidatepage,
1077 #ifdef CONFIG_MIGRATION
1078 	.migratepage	= btree_migratepage,
1079 #endif
1080 	.set_page_dirty = btree_set_page_dirty,
1081 };
1082 
1083 struct extent_buffer *btrfs_find_create_tree_block(
1084 						struct btrfs_fs_info *fs_info,
1085 						u64 bytenr, u64 owner_root,
1086 						int level)
1087 {
1088 	if (btrfs_is_testing(fs_info))
1089 		return alloc_test_extent_buffer(fs_info, bytenr);
1090 	return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
1091 }
1092 
1093 /*
1094  * Read tree block at logical address @bytenr and do variant basic but critical
1095  * verification.
1096  *
1097  * @owner_root:		the objectid of the root owner for this block.
1098  * @parent_transid:	expected transid of this tree block, skip check if 0
1099  * @level:		expected level, mandatory check
1100  * @first_key:		expected key in slot 0, skip check if NULL
1101  */
1102 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1103 				      u64 owner_root, u64 parent_transid,
1104 				      int level, struct btrfs_key *first_key)
1105 {
1106 	struct extent_buffer *buf = NULL;
1107 	int ret;
1108 
1109 	buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
1110 	if (IS_ERR(buf))
1111 		return buf;
1112 
1113 	ret = btree_read_extent_buffer_pages(buf, parent_transid,
1114 					     level, first_key);
1115 	if (ret) {
1116 		free_extent_buffer_stale(buf);
1117 		return ERR_PTR(ret);
1118 	}
1119 	return buf;
1120 
1121 }
1122 
1123 void btrfs_clean_tree_block(struct extent_buffer *buf)
1124 {
1125 	struct btrfs_fs_info *fs_info = buf->fs_info;
1126 	if (btrfs_header_generation(buf) ==
1127 	    fs_info->running_transaction->transid) {
1128 		btrfs_assert_tree_write_locked(buf);
1129 
1130 		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1131 			percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1132 						 -buf->len,
1133 						 fs_info->dirty_metadata_batch);
1134 			clear_extent_buffer_dirty(buf);
1135 		}
1136 	}
1137 }
1138 
1139 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1140 			 u64 objectid)
1141 {
1142 	bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1143 	root->fs_info = fs_info;
1144 	root->node = NULL;
1145 	root->commit_root = NULL;
1146 	root->state = 0;
1147 	root->orphan_cleanup_state = 0;
1148 
1149 	root->last_trans = 0;
1150 	root->free_objectid = 0;
1151 	root->nr_delalloc_inodes = 0;
1152 	root->nr_ordered_extents = 0;
1153 	root->inode_tree = RB_ROOT;
1154 	INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1155 	root->block_rsv = NULL;
1156 
1157 	INIT_LIST_HEAD(&root->dirty_list);
1158 	INIT_LIST_HEAD(&root->root_list);
1159 	INIT_LIST_HEAD(&root->delalloc_inodes);
1160 	INIT_LIST_HEAD(&root->delalloc_root);
1161 	INIT_LIST_HEAD(&root->ordered_extents);
1162 	INIT_LIST_HEAD(&root->ordered_root);
1163 	INIT_LIST_HEAD(&root->reloc_dirty_list);
1164 	INIT_LIST_HEAD(&root->logged_list[0]);
1165 	INIT_LIST_HEAD(&root->logged_list[1]);
1166 	spin_lock_init(&root->inode_lock);
1167 	spin_lock_init(&root->delalloc_lock);
1168 	spin_lock_init(&root->ordered_extent_lock);
1169 	spin_lock_init(&root->accounting_lock);
1170 	spin_lock_init(&root->log_extents_lock[0]);
1171 	spin_lock_init(&root->log_extents_lock[1]);
1172 	spin_lock_init(&root->qgroup_meta_rsv_lock);
1173 	mutex_init(&root->objectid_mutex);
1174 	mutex_init(&root->log_mutex);
1175 	mutex_init(&root->ordered_extent_mutex);
1176 	mutex_init(&root->delalloc_mutex);
1177 	init_waitqueue_head(&root->qgroup_flush_wait);
1178 	init_waitqueue_head(&root->log_writer_wait);
1179 	init_waitqueue_head(&root->log_commit_wait[0]);
1180 	init_waitqueue_head(&root->log_commit_wait[1]);
1181 	INIT_LIST_HEAD(&root->log_ctxs[0]);
1182 	INIT_LIST_HEAD(&root->log_ctxs[1]);
1183 	atomic_set(&root->log_commit[0], 0);
1184 	atomic_set(&root->log_commit[1], 0);
1185 	atomic_set(&root->log_writers, 0);
1186 	atomic_set(&root->log_batch, 0);
1187 	refcount_set(&root->refs, 1);
1188 	atomic_set(&root->snapshot_force_cow, 0);
1189 	atomic_set(&root->nr_swapfiles, 0);
1190 	root->log_transid = 0;
1191 	root->log_transid_committed = -1;
1192 	root->last_log_commit = 0;
1193 	if (!dummy) {
1194 		extent_io_tree_init(fs_info, &root->dirty_log_pages,
1195 				    IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1196 		extent_io_tree_init(fs_info, &root->log_csum_range,
1197 				    IO_TREE_LOG_CSUM_RANGE, NULL);
1198 	}
1199 
1200 	memset(&root->root_key, 0, sizeof(root->root_key));
1201 	memset(&root->root_item, 0, sizeof(root->root_item));
1202 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1203 	root->root_key.objectid = objectid;
1204 	root->anon_dev = 0;
1205 
1206 	spin_lock_init(&root->root_item_lock);
1207 	btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1208 #ifdef CONFIG_BTRFS_DEBUG
1209 	INIT_LIST_HEAD(&root->leak_list);
1210 	spin_lock(&fs_info->fs_roots_radix_lock);
1211 	list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1212 	spin_unlock(&fs_info->fs_roots_radix_lock);
1213 #endif
1214 }
1215 
1216 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1217 					   u64 objectid, gfp_t flags)
1218 {
1219 	struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1220 	if (root)
1221 		__setup_root(root, fs_info, objectid);
1222 	return root;
1223 }
1224 
1225 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1226 /* Should only be used by the testing infrastructure */
1227 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1228 {
1229 	struct btrfs_root *root;
1230 
1231 	if (!fs_info)
1232 		return ERR_PTR(-EINVAL);
1233 
1234 	root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1235 	if (!root)
1236 		return ERR_PTR(-ENOMEM);
1237 
1238 	/* We don't use the stripesize in selftest, set it as sectorsize */
1239 	root->alloc_bytenr = 0;
1240 
1241 	return root;
1242 }
1243 #endif
1244 
1245 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1246 				     u64 objectid)
1247 {
1248 	struct btrfs_fs_info *fs_info = trans->fs_info;
1249 	struct extent_buffer *leaf;
1250 	struct btrfs_root *tree_root = fs_info->tree_root;
1251 	struct btrfs_root *root;
1252 	struct btrfs_key key;
1253 	unsigned int nofs_flag;
1254 	int ret = 0;
1255 
1256 	/*
1257 	 * We're holding a transaction handle, so use a NOFS memory allocation
1258 	 * context to avoid deadlock if reclaim happens.
1259 	 */
1260 	nofs_flag = memalloc_nofs_save();
1261 	root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1262 	memalloc_nofs_restore(nofs_flag);
1263 	if (!root)
1264 		return ERR_PTR(-ENOMEM);
1265 
1266 	root->root_key.objectid = objectid;
1267 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1268 	root->root_key.offset = 0;
1269 
1270 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1271 				      BTRFS_NESTING_NORMAL);
1272 	if (IS_ERR(leaf)) {
1273 		ret = PTR_ERR(leaf);
1274 		leaf = NULL;
1275 		goto fail_unlock;
1276 	}
1277 
1278 	root->node = leaf;
1279 	btrfs_mark_buffer_dirty(leaf);
1280 
1281 	root->commit_root = btrfs_root_node(root);
1282 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1283 
1284 	btrfs_set_root_flags(&root->root_item, 0);
1285 	btrfs_set_root_limit(&root->root_item, 0);
1286 	btrfs_set_root_bytenr(&root->root_item, leaf->start);
1287 	btrfs_set_root_generation(&root->root_item, trans->transid);
1288 	btrfs_set_root_level(&root->root_item, 0);
1289 	btrfs_set_root_refs(&root->root_item, 1);
1290 	btrfs_set_root_used(&root->root_item, leaf->len);
1291 	btrfs_set_root_last_snapshot(&root->root_item, 0);
1292 	btrfs_set_root_dirid(&root->root_item, 0);
1293 	if (is_fstree(objectid))
1294 		generate_random_guid(root->root_item.uuid);
1295 	else
1296 		export_guid(root->root_item.uuid, &guid_null);
1297 	btrfs_set_root_drop_level(&root->root_item, 0);
1298 
1299 	btrfs_tree_unlock(leaf);
1300 
1301 	key.objectid = objectid;
1302 	key.type = BTRFS_ROOT_ITEM_KEY;
1303 	key.offset = 0;
1304 	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1305 	if (ret)
1306 		goto fail;
1307 
1308 	return root;
1309 
1310 fail_unlock:
1311 	if (leaf)
1312 		btrfs_tree_unlock(leaf);
1313 fail:
1314 	btrfs_put_root(root);
1315 
1316 	return ERR_PTR(ret);
1317 }
1318 
1319 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1320 					 struct btrfs_fs_info *fs_info)
1321 {
1322 	struct btrfs_root *root;
1323 
1324 	root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1325 	if (!root)
1326 		return ERR_PTR(-ENOMEM);
1327 
1328 	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1329 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1330 	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1331 
1332 	return root;
1333 }
1334 
1335 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1336 			      struct btrfs_root *root)
1337 {
1338 	struct extent_buffer *leaf;
1339 
1340 	/*
1341 	 * DON'T set SHAREABLE bit for log trees.
1342 	 *
1343 	 * Log trees are not exposed to user space thus can't be snapshotted,
1344 	 * and they go away before a real commit is actually done.
1345 	 *
1346 	 * They do store pointers to file data extents, and those reference
1347 	 * counts still get updated (along with back refs to the log tree).
1348 	 */
1349 
1350 	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1351 			NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1352 	if (IS_ERR(leaf))
1353 		return PTR_ERR(leaf);
1354 
1355 	root->node = leaf;
1356 
1357 	btrfs_mark_buffer_dirty(root->node);
1358 	btrfs_tree_unlock(root->node);
1359 
1360 	return 0;
1361 }
1362 
1363 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1364 			     struct btrfs_fs_info *fs_info)
1365 {
1366 	struct btrfs_root *log_root;
1367 
1368 	log_root = alloc_log_tree(trans, fs_info);
1369 	if (IS_ERR(log_root))
1370 		return PTR_ERR(log_root);
1371 
1372 	if (!btrfs_is_zoned(fs_info)) {
1373 		int ret = btrfs_alloc_log_tree_node(trans, log_root);
1374 
1375 		if (ret) {
1376 			btrfs_put_root(log_root);
1377 			return ret;
1378 		}
1379 	}
1380 
1381 	WARN_ON(fs_info->log_root_tree);
1382 	fs_info->log_root_tree = log_root;
1383 	return 0;
1384 }
1385 
1386 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1387 		       struct btrfs_root *root)
1388 {
1389 	struct btrfs_fs_info *fs_info = root->fs_info;
1390 	struct btrfs_root *log_root;
1391 	struct btrfs_inode_item *inode_item;
1392 	int ret;
1393 
1394 	log_root = alloc_log_tree(trans, fs_info);
1395 	if (IS_ERR(log_root))
1396 		return PTR_ERR(log_root);
1397 
1398 	ret = btrfs_alloc_log_tree_node(trans, log_root);
1399 	if (ret) {
1400 		btrfs_put_root(log_root);
1401 		return ret;
1402 	}
1403 
1404 	log_root->last_trans = trans->transid;
1405 	log_root->root_key.offset = root->root_key.objectid;
1406 
1407 	inode_item = &log_root->root_item.inode;
1408 	btrfs_set_stack_inode_generation(inode_item, 1);
1409 	btrfs_set_stack_inode_size(inode_item, 3);
1410 	btrfs_set_stack_inode_nlink(inode_item, 1);
1411 	btrfs_set_stack_inode_nbytes(inode_item,
1412 				     fs_info->nodesize);
1413 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1414 
1415 	btrfs_set_root_node(&log_root->root_item, log_root->node);
1416 
1417 	WARN_ON(root->log_root);
1418 	root->log_root = log_root;
1419 	root->log_transid = 0;
1420 	root->log_transid_committed = -1;
1421 	root->last_log_commit = 0;
1422 	return 0;
1423 }
1424 
1425 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1426 					      struct btrfs_path *path,
1427 					      struct btrfs_key *key)
1428 {
1429 	struct btrfs_root *root;
1430 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1431 	u64 generation;
1432 	int ret;
1433 	int level;
1434 
1435 	root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1436 	if (!root)
1437 		return ERR_PTR(-ENOMEM);
1438 
1439 	ret = btrfs_find_root(tree_root, key, path,
1440 			      &root->root_item, &root->root_key);
1441 	if (ret) {
1442 		if (ret > 0)
1443 			ret = -ENOENT;
1444 		goto fail;
1445 	}
1446 
1447 	generation = btrfs_root_generation(&root->root_item);
1448 	level = btrfs_root_level(&root->root_item);
1449 	root->node = read_tree_block(fs_info,
1450 				     btrfs_root_bytenr(&root->root_item),
1451 				     key->objectid, generation, level, NULL);
1452 	if (IS_ERR(root->node)) {
1453 		ret = PTR_ERR(root->node);
1454 		root->node = NULL;
1455 		goto fail;
1456 	} else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1457 		ret = -EIO;
1458 		goto fail;
1459 	}
1460 	root->commit_root = btrfs_root_node(root);
1461 	return root;
1462 fail:
1463 	btrfs_put_root(root);
1464 	return ERR_PTR(ret);
1465 }
1466 
1467 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1468 					struct btrfs_key *key)
1469 {
1470 	struct btrfs_root *root;
1471 	struct btrfs_path *path;
1472 
1473 	path = btrfs_alloc_path();
1474 	if (!path)
1475 		return ERR_PTR(-ENOMEM);
1476 	root = read_tree_root_path(tree_root, path, key);
1477 	btrfs_free_path(path);
1478 
1479 	return root;
1480 }
1481 
1482 /*
1483  * Initialize subvolume root in-memory structure
1484  *
1485  * @anon_dev:	anonymous device to attach to the root, if zero, allocate new
1486  */
1487 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1488 {
1489 	int ret;
1490 	unsigned int nofs_flag;
1491 
1492 	/*
1493 	 * We might be called under a transaction (e.g. indirect backref
1494 	 * resolution) which could deadlock if it triggers memory reclaim
1495 	 */
1496 	nofs_flag = memalloc_nofs_save();
1497 	ret = btrfs_drew_lock_init(&root->snapshot_lock);
1498 	memalloc_nofs_restore(nofs_flag);
1499 	if (ret)
1500 		goto fail;
1501 
1502 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1503 	    !btrfs_is_data_reloc_root(root)) {
1504 		set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1505 		btrfs_check_and_init_root_item(&root->root_item);
1506 	}
1507 
1508 	/*
1509 	 * Don't assign anonymous block device to roots that are not exposed to
1510 	 * userspace, the id pool is limited to 1M
1511 	 */
1512 	if (is_fstree(root->root_key.objectid) &&
1513 	    btrfs_root_refs(&root->root_item) > 0) {
1514 		if (!anon_dev) {
1515 			ret = get_anon_bdev(&root->anon_dev);
1516 			if (ret)
1517 				goto fail;
1518 		} else {
1519 			root->anon_dev = anon_dev;
1520 		}
1521 	}
1522 
1523 	mutex_lock(&root->objectid_mutex);
1524 	ret = btrfs_init_root_free_objectid(root);
1525 	if (ret) {
1526 		mutex_unlock(&root->objectid_mutex);
1527 		goto fail;
1528 	}
1529 
1530 	ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1531 
1532 	mutex_unlock(&root->objectid_mutex);
1533 
1534 	return 0;
1535 fail:
1536 	/* The caller is responsible to call btrfs_free_fs_root */
1537 	return ret;
1538 }
1539 
1540 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1541 					       u64 root_id)
1542 {
1543 	struct btrfs_root *root;
1544 
1545 	spin_lock(&fs_info->fs_roots_radix_lock);
1546 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1547 				 (unsigned long)root_id);
1548 	if (root)
1549 		root = btrfs_grab_root(root);
1550 	spin_unlock(&fs_info->fs_roots_radix_lock);
1551 	return root;
1552 }
1553 
1554 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1555 						u64 objectid)
1556 {
1557 	if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1558 		return btrfs_grab_root(fs_info->tree_root);
1559 	if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1560 		return btrfs_grab_root(fs_info->extent_root);
1561 	if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1562 		return btrfs_grab_root(fs_info->chunk_root);
1563 	if (objectid == BTRFS_DEV_TREE_OBJECTID)
1564 		return btrfs_grab_root(fs_info->dev_root);
1565 	if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1566 		return btrfs_grab_root(fs_info->csum_root);
1567 	if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1568 		return btrfs_grab_root(fs_info->quota_root) ?
1569 			fs_info->quota_root : ERR_PTR(-ENOENT);
1570 	if (objectid == BTRFS_UUID_TREE_OBJECTID)
1571 		return btrfs_grab_root(fs_info->uuid_root) ?
1572 			fs_info->uuid_root : ERR_PTR(-ENOENT);
1573 	if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1574 		return btrfs_grab_root(fs_info->free_space_root) ?
1575 			fs_info->free_space_root : ERR_PTR(-ENOENT);
1576 	return NULL;
1577 }
1578 
1579 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1580 			 struct btrfs_root *root)
1581 {
1582 	int ret;
1583 
1584 	ret = radix_tree_preload(GFP_NOFS);
1585 	if (ret)
1586 		return ret;
1587 
1588 	spin_lock(&fs_info->fs_roots_radix_lock);
1589 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1590 				(unsigned long)root->root_key.objectid,
1591 				root);
1592 	if (ret == 0) {
1593 		btrfs_grab_root(root);
1594 		set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1595 	}
1596 	spin_unlock(&fs_info->fs_roots_radix_lock);
1597 	radix_tree_preload_end();
1598 
1599 	return ret;
1600 }
1601 
1602 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1603 {
1604 #ifdef CONFIG_BTRFS_DEBUG
1605 	struct btrfs_root *root;
1606 
1607 	while (!list_empty(&fs_info->allocated_roots)) {
1608 		char buf[BTRFS_ROOT_NAME_BUF_LEN];
1609 
1610 		root = list_first_entry(&fs_info->allocated_roots,
1611 					struct btrfs_root, leak_list);
1612 		btrfs_err(fs_info, "leaked root %s refcount %d",
1613 			  btrfs_root_name(&root->root_key, buf),
1614 			  refcount_read(&root->refs));
1615 		while (refcount_read(&root->refs) > 1)
1616 			btrfs_put_root(root);
1617 		btrfs_put_root(root);
1618 	}
1619 #endif
1620 }
1621 
1622 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1623 {
1624 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1625 	percpu_counter_destroy(&fs_info->delalloc_bytes);
1626 	percpu_counter_destroy(&fs_info->ordered_bytes);
1627 	percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1628 	btrfs_free_csum_hash(fs_info);
1629 	btrfs_free_stripe_hash_table(fs_info);
1630 	btrfs_free_ref_cache(fs_info);
1631 	kfree(fs_info->balance_ctl);
1632 	kfree(fs_info->delayed_root);
1633 	btrfs_put_root(fs_info->extent_root);
1634 	btrfs_put_root(fs_info->tree_root);
1635 	btrfs_put_root(fs_info->chunk_root);
1636 	btrfs_put_root(fs_info->dev_root);
1637 	btrfs_put_root(fs_info->csum_root);
1638 	btrfs_put_root(fs_info->quota_root);
1639 	btrfs_put_root(fs_info->uuid_root);
1640 	btrfs_put_root(fs_info->free_space_root);
1641 	btrfs_put_root(fs_info->fs_root);
1642 	btrfs_put_root(fs_info->data_reloc_root);
1643 	btrfs_check_leaked_roots(fs_info);
1644 	btrfs_extent_buffer_leak_debug_check(fs_info);
1645 	kfree(fs_info->super_copy);
1646 	kfree(fs_info->super_for_commit);
1647 	kfree(fs_info->subpage_info);
1648 	kvfree(fs_info);
1649 }
1650 
1651 
1652 /*
1653  * Get an in-memory reference of a root structure.
1654  *
1655  * For essential trees like root/extent tree, we grab it from fs_info directly.
1656  * For subvolume trees, we check the cached filesystem roots first. If not
1657  * found, then read it from disk and add it to cached fs roots.
1658  *
1659  * Caller should release the root by calling btrfs_put_root() after the usage.
1660  *
1661  * NOTE: Reloc and log trees can't be read by this function as they share the
1662  *	 same root objectid.
1663  *
1664  * @objectid:	root id
1665  * @anon_dev:	preallocated anonymous block device number for new roots,
1666  * 		pass 0 for new allocation.
1667  * @check_ref:	whether to check root item references, If true, return -ENOENT
1668  *		for orphan roots
1669  */
1670 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1671 					     u64 objectid, dev_t anon_dev,
1672 					     bool check_ref)
1673 {
1674 	struct btrfs_root *root;
1675 	struct btrfs_path *path;
1676 	struct btrfs_key key;
1677 	int ret;
1678 
1679 	root = btrfs_get_global_root(fs_info, objectid);
1680 	if (root)
1681 		return root;
1682 again:
1683 	root = btrfs_lookup_fs_root(fs_info, objectid);
1684 	if (root) {
1685 		/* Shouldn't get preallocated anon_dev for cached roots */
1686 		ASSERT(!anon_dev);
1687 		if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1688 			btrfs_put_root(root);
1689 			return ERR_PTR(-ENOENT);
1690 		}
1691 		return root;
1692 	}
1693 
1694 	key.objectid = objectid;
1695 	key.type = BTRFS_ROOT_ITEM_KEY;
1696 	key.offset = (u64)-1;
1697 	root = btrfs_read_tree_root(fs_info->tree_root, &key);
1698 	if (IS_ERR(root))
1699 		return root;
1700 
1701 	if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1702 		ret = -ENOENT;
1703 		goto fail;
1704 	}
1705 
1706 	ret = btrfs_init_fs_root(root, anon_dev);
1707 	if (ret)
1708 		goto fail;
1709 
1710 	path = btrfs_alloc_path();
1711 	if (!path) {
1712 		ret = -ENOMEM;
1713 		goto fail;
1714 	}
1715 	key.objectid = BTRFS_ORPHAN_OBJECTID;
1716 	key.type = BTRFS_ORPHAN_ITEM_KEY;
1717 	key.offset = objectid;
1718 
1719 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1720 	btrfs_free_path(path);
1721 	if (ret < 0)
1722 		goto fail;
1723 	if (ret == 0)
1724 		set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1725 
1726 	ret = btrfs_insert_fs_root(fs_info, root);
1727 	if (ret) {
1728 		btrfs_put_root(root);
1729 		if (ret == -EEXIST)
1730 			goto again;
1731 		goto fail;
1732 	}
1733 	return root;
1734 fail:
1735 	btrfs_put_root(root);
1736 	return ERR_PTR(ret);
1737 }
1738 
1739 /*
1740  * Get in-memory reference of a root structure
1741  *
1742  * @objectid:	tree objectid
1743  * @check_ref:	if set, verify that the tree exists and the item has at least
1744  *		one reference
1745  */
1746 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1747 				     u64 objectid, bool check_ref)
1748 {
1749 	return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1750 }
1751 
1752 /*
1753  * Get in-memory reference of a root structure, created as new, optionally pass
1754  * the anonymous block device id
1755  *
1756  * @objectid:	tree objectid
1757  * @anon_dev:	if zero, allocate a new anonymous block device or use the
1758  *		parameter value
1759  */
1760 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1761 					 u64 objectid, dev_t anon_dev)
1762 {
1763 	return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1764 }
1765 
1766 /*
1767  * btrfs_get_fs_root_commit_root - return a root for the given objectid
1768  * @fs_info:	the fs_info
1769  * @objectid:	the objectid we need to lookup
1770  *
1771  * This is exclusively used for backref walking, and exists specifically because
1772  * of how qgroups does lookups.  Qgroups will do a backref lookup at delayed ref
1773  * creation time, which means we may have to read the tree_root in order to look
1774  * up a fs root that is not in memory.  If the root is not in memory we will
1775  * read the tree root commit root and look up the fs root from there.  This is a
1776  * temporary root, it will not be inserted into the radix tree as it doesn't
1777  * have the most uptodate information, it'll simply be discarded once the
1778  * backref code is finished using the root.
1779  */
1780 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1781 						 struct btrfs_path *path,
1782 						 u64 objectid)
1783 {
1784 	struct btrfs_root *root;
1785 	struct btrfs_key key;
1786 
1787 	ASSERT(path->search_commit_root && path->skip_locking);
1788 
1789 	/*
1790 	 * This can return -ENOENT if we ask for a root that doesn't exist, but
1791 	 * since this is called via the backref walking code we won't be looking
1792 	 * up a root that doesn't exist, unless there's corruption.  So if root
1793 	 * != NULL just return it.
1794 	 */
1795 	root = btrfs_get_global_root(fs_info, objectid);
1796 	if (root)
1797 		return root;
1798 
1799 	root = btrfs_lookup_fs_root(fs_info, objectid);
1800 	if (root)
1801 		return root;
1802 
1803 	key.objectid = objectid;
1804 	key.type = BTRFS_ROOT_ITEM_KEY;
1805 	key.offset = (u64)-1;
1806 	root = read_tree_root_path(fs_info->tree_root, path, &key);
1807 	btrfs_release_path(path);
1808 
1809 	return root;
1810 }
1811 
1812 /*
1813  * called by the kthread helper functions to finally call the bio end_io
1814  * functions.  This is where read checksum verification actually happens
1815  */
1816 static void end_workqueue_fn(struct btrfs_work *work)
1817 {
1818 	struct bio *bio;
1819 	struct btrfs_end_io_wq *end_io_wq;
1820 
1821 	end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1822 	bio = end_io_wq->bio;
1823 
1824 	bio->bi_status = end_io_wq->status;
1825 	bio->bi_private = end_io_wq->private;
1826 	bio->bi_end_io = end_io_wq->end_io;
1827 	bio_endio(bio);
1828 	kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1829 }
1830 
1831 static int cleaner_kthread(void *arg)
1832 {
1833 	struct btrfs_root *root = arg;
1834 	struct btrfs_fs_info *fs_info = root->fs_info;
1835 	int again;
1836 
1837 	while (1) {
1838 		again = 0;
1839 
1840 		set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1841 
1842 		/* Make the cleaner go to sleep early. */
1843 		if (btrfs_need_cleaner_sleep(fs_info))
1844 			goto sleep;
1845 
1846 		/*
1847 		 * Do not do anything if we might cause open_ctree() to block
1848 		 * before we have finished mounting the filesystem.
1849 		 */
1850 		if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1851 			goto sleep;
1852 
1853 		if (!mutex_trylock(&fs_info->cleaner_mutex))
1854 			goto sleep;
1855 
1856 		/*
1857 		 * Avoid the problem that we change the status of the fs
1858 		 * during the above check and trylock.
1859 		 */
1860 		if (btrfs_need_cleaner_sleep(fs_info)) {
1861 			mutex_unlock(&fs_info->cleaner_mutex);
1862 			goto sleep;
1863 		}
1864 
1865 		btrfs_run_delayed_iputs(fs_info);
1866 
1867 		again = btrfs_clean_one_deleted_snapshot(root);
1868 		mutex_unlock(&fs_info->cleaner_mutex);
1869 
1870 		/*
1871 		 * The defragger has dealt with the R/O remount and umount,
1872 		 * needn't do anything special here.
1873 		 */
1874 		btrfs_run_defrag_inodes(fs_info);
1875 
1876 		/*
1877 		 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1878 		 * with relocation (btrfs_relocate_chunk) and relocation
1879 		 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1880 		 * after acquiring fs_info->reclaim_bgs_lock. So we
1881 		 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1882 		 * unused block groups.
1883 		 */
1884 		btrfs_delete_unused_bgs(fs_info);
1885 
1886 		/*
1887 		 * Reclaim block groups in the reclaim_bgs list after we deleted
1888 		 * all unused block_groups. This possibly gives us some more free
1889 		 * space.
1890 		 */
1891 		btrfs_reclaim_bgs(fs_info);
1892 sleep:
1893 		clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1894 		if (kthread_should_park())
1895 			kthread_parkme();
1896 		if (kthread_should_stop())
1897 			return 0;
1898 		if (!again) {
1899 			set_current_state(TASK_INTERRUPTIBLE);
1900 			schedule();
1901 			__set_current_state(TASK_RUNNING);
1902 		}
1903 	}
1904 }
1905 
1906 static int transaction_kthread(void *arg)
1907 {
1908 	struct btrfs_root *root = arg;
1909 	struct btrfs_fs_info *fs_info = root->fs_info;
1910 	struct btrfs_trans_handle *trans;
1911 	struct btrfs_transaction *cur;
1912 	u64 transid;
1913 	time64_t delta;
1914 	unsigned long delay;
1915 	bool cannot_commit;
1916 
1917 	do {
1918 		cannot_commit = false;
1919 		delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1920 		mutex_lock(&fs_info->transaction_kthread_mutex);
1921 
1922 		spin_lock(&fs_info->trans_lock);
1923 		cur = fs_info->running_transaction;
1924 		if (!cur) {
1925 			spin_unlock(&fs_info->trans_lock);
1926 			goto sleep;
1927 		}
1928 
1929 		delta = ktime_get_seconds() - cur->start_time;
1930 		if (cur->state < TRANS_STATE_COMMIT_START &&
1931 		    delta < fs_info->commit_interval) {
1932 			spin_unlock(&fs_info->trans_lock);
1933 			delay -= msecs_to_jiffies((delta - 1) * 1000);
1934 			delay = min(delay,
1935 				    msecs_to_jiffies(fs_info->commit_interval * 1000));
1936 			goto sleep;
1937 		}
1938 		transid = cur->transid;
1939 		spin_unlock(&fs_info->trans_lock);
1940 
1941 		/* If the file system is aborted, this will always fail. */
1942 		trans = btrfs_attach_transaction(root);
1943 		if (IS_ERR(trans)) {
1944 			if (PTR_ERR(trans) != -ENOENT)
1945 				cannot_commit = true;
1946 			goto sleep;
1947 		}
1948 		if (transid == trans->transid) {
1949 			btrfs_commit_transaction(trans);
1950 		} else {
1951 			btrfs_end_transaction(trans);
1952 		}
1953 sleep:
1954 		wake_up_process(fs_info->cleaner_kthread);
1955 		mutex_unlock(&fs_info->transaction_kthread_mutex);
1956 
1957 		if (BTRFS_FS_ERROR(fs_info))
1958 			btrfs_cleanup_transaction(fs_info);
1959 		if (!kthread_should_stop() &&
1960 				(!btrfs_transaction_blocked(fs_info) ||
1961 				 cannot_commit))
1962 			schedule_timeout_interruptible(delay);
1963 	} while (!kthread_should_stop());
1964 	return 0;
1965 }
1966 
1967 /*
1968  * This will find the highest generation in the array of root backups.  The
1969  * index of the highest array is returned, or -EINVAL if we can't find
1970  * anything.
1971  *
1972  * We check to make sure the array is valid by comparing the
1973  * generation of the latest  root in the array with the generation
1974  * in the super block.  If they don't match we pitch it.
1975  */
1976 static int find_newest_super_backup(struct btrfs_fs_info *info)
1977 {
1978 	const u64 newest_gen = btrfs_super_generation(info->super_copy);
1979 	u64 cur;
1980 	struct btrfs_root_backup *root_backup;
1981 	int i;
1982 
1983 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1984 		root_backup = info->super_copy->super_roots + i;
1985 		cur = btrfs_backup_tree_root_gen(root_backup);
1986 		if (cur == newest_gen)
1987 			return i;
1988 	}
1989 
1990 	return -EINVAL;
1991 }
1992 
1993 /*
1994  * copy all the root pointers into the super backup array.
1995  * this will bump the backup pointer by one when it is
1996  * done
1997  */
1998 static void backup_super_roots(struct btrfs_fs_info *info)
1999 {
2000 	const int next_backup = info->backup_root_index;
2001 	struct btrfs_root_backup *root_backup;
2002 
2003 	root_backup = info->super_for_commit->super_roots + next_backup;
2004 
2005 	/*
2006 	 * make sure all of our padding and empty slots get zero filled
2007 	 * regardless of which ones we use today
2008 	 */
2009 	memset(root_backup, 0, sizeof(*root_backup));
2010 
2011 	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2012 
2013 	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2014 	btrfs_set_backup_tree_root_gen(root_backup,
2015 			       btrfs_header_generation(info->tree_root->node));
2016 
2017 	btrfs_set_backup_tree_root_level(root_backup,
2018 			       btrfs_header_level(info->tree_root->node));
2019 
2020 	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2021 	btrfs_set_backup_chunk_root_gen(root_backup,
2022 			       btrfs_header_generation(info->chunk_root->node));
2023 	btrfs_set_backup_chunk_root_level(root_backup,
2024 			       btrfs_header_level(info->chunk_root->node));
2025 
2026 	btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2027 	btrfs_set_backup_extent_root_gen(root_backup,
2028 			       btrfs_header_generation(info->extent_root->node));
2029 	btrfs_set_backup_extent_root_level(root_backup,
2030 			       btrfs_header_level(info->extent_root->node));
2031 
2032 	/*
2033 	 * we might commit during log recovery, which happens before we set
2034 	 * the fs_root.  Make sure it is valid before we fill it in.
2035 	 */
2036 	if (info->fs_root && info->fs_root->node) {
2037 		btrfs_set_backup_fs_root(root_backup,
2038 					 info->fs_root->node->start);
2039 		btrfs_set_backup_fs_root_gen(root_backup,
2040 			       btrfs_header_generation(info->fs_root->node));
2041 		btrfs_set_backup_fs_root_level(root_backup,
2042 			       btrfs_header_level(info->fs_root->node));
2043 	}
2044 
2045 	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2046 	btrfs_set_backup_dev_root_gen(root_backup,
2047 			       btrfs_header_generation(info->dev_root->node));
2048 	btrfs_set_backup_dev_root_level(root_backup,
2049 				       btrfs_header_level(info->dev_root->node));
2050 
2051 	btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2052 	btrfs_set_backup_csum_root_gen(root_backup,
2053 			       btrfs_header_generation(info->csum_root->node));
2054 	btrfs_set_backup_csum_root_level(root_backup,
2055 			       btrfs_header_level(info->csum_root->node));
2056 
2057 	btrfs_set_backup_total_bytes(root_backup,
2058 			     btrfs_super_total_bytes(info->super_copy));
2059 	btrfs_set_backup_bytes_used(root_backup,
2060 			     btrfs_super_bytes_used(info->super_copy));
2061 	btrfs_set_backup_num_devices(root_backup,
2062 			     btrfs_super_num_devices(info->super_copy));
2063 
2064 	/*
2065 	 * if we don't copy this out to the super_copy, it won't get remembered
2066 	 * for the next commit
2067 	 */
2068 	memcpy(&info->super_copy->super_roots,
2069 	       &info->super_for_commit->super_roots,
2070 	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2071 }
2072 
2073 /*
2074  * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2075  * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2076  *
2077  * fs_info - filesystem whose backup roots need to be read
2078  * priority - priority of backup root required
2079  *
2080  * Returns backup root index on success and -EINVAL otherwise.
2081  */
2082 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2083 {
2084 	int backup_index = find_newest_super_backup(fs_info);
2085 	struct btrfs_super_block *super = fs_info->super_copy;
2086 	struct btrfs_root_backup *root_backup;
2087 
2088 	if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2089 		if (priority == 0)
2090 			return backup_index;
2091 
2092 		backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2093 		backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2094 	} else {
2095 		return -EINVAL;
2096 	}
2097 
2098 	root_backup = super->super_roots + backup_index;
2099 
2100 	btrfs_set_super_generation(super,
2101 				   btrfs_backup_tree_root_gen(root_backup));
2102 	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2103 	btrfs_set_super_root_level(super,
2104 				   btrfs_backup_tree_root_level(root_backup));
2105 	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2106 
2107 	/*
2108 	 * Fixme: the total bytes and num_devices need to match or we should
2109 	 * need a fsck
2110 	 */
2111 	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2112 	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2113 
2114 	return backup_index;
2115 }
2116 
2117 /* helper to cleanup workers */
2118 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2119 {
2120 	btrfs_destroy_workqueue(fs_info->fixup_workers);
2121 	btrfs_destroy_workqueue(fs_info->delalloc_workers);
2122 	btrfs_destroy_workqueue(fs_info->workers);
2123 	btrfs_destroy_workqueue(fs_info->endio_workers);
2124 	btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2125 	btrfs_destroy_workqueue(fs_info->rmw_workers);
2126 	btrfs_destroy_workqueue(fs_info->endio_write_workers);
2127 	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2128 	btrfs_destroy_workqueue(fs_info->delayed_workers);
2129 	btrfs_destroy_workqueue(fs_info->caching_workers);
2130 	btrfs_destroy_workqueue(fs_info->readahead_workers);
2131 	btrfs_destroy_workqueue(fs_info->flush_workers);
2132 	btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2133 	if (fs_info->discard_ctl.discard_workers)
2134 		destroy_workqueue(fs_info->discard_ctl.discard_workers);
2135 	/*
2136 	 * Now that all other work queues are destroyed, we can safely destroy
2137 	 * the queues used for metadata I/O, since tasks from those other work
2138 	 * queues can do metadata I/O operations.
2139 	 */
2140 	btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2141 	btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2142 }
2143 
2144 static void free_root_extent_buffers(struct btrfs_root *root)
2145 {
2146 	if (root) {
2147 		free_extent_buffer(root->node);
2148 		free_extent_buffer(root->commit_root);
2149 		root->node = NULL;
2150 		root->commit_root = NULL;
2151 	}
2152 }
2153 
2154 /* helper to cleanup tree roots */
2155 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2156 {
2157 	free_root_extent_buffers(info->tree_root);
2158 
2159 	free_root_extent_buffers(info->dev_root);
2160 	free_root_extent_buffers(info->extent_root);
2161 	free_root_extent_buffers(info->csum_root);
2162 	free_root_extent_buffers(info->quota_root);
2163 	free_root_extent_buffers(info->uuid_root);
2164 	free_root_extent_buffers(info->fs_root);
2165 	free_root_extent_buffers(info->data_reloc_root);
2166 	if (free_chunk_root)
2167 		free_root_extent_buffers(info->chunk_root);
2168 	free_root_extent_buffers(info->free_space_root);
2169 }
2170 
2171 void btrfs_put_root(struct btrfs_root *root)
2172 {
2173 	if (!root)
2174 		return;
2175 
2176 	if (refcount_dec_and_test(&root->refs)) {
2177 		WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2178 		WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2179 		if (root->anon_dev)
2180 			free_anon_bdev(root->anon_dev);
2181 		btrfs_drew_lock_destroy(&root->snapshot_lock);
2182 		free_root_extent_buffers(root);
2183 #ifdef CONFIG_BTRFS_DEBUG
2184 		spin_lock(&root->fs_info->fs_roots_radix_lock);
2185 		list_del_init(&root->leak_list);
2186 		spin_unlock(&root->fs_info->fs_roots_radix_lock);
2187 #endif
2188 		kfree(root);
2189 	}
2190 }
2191 
2192 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2193 {
2194 	int ret;
2195 	struct btrfs_root *gang[8];
2196 	int i;
2197 
2198 	while (!list_empty(&fs_info->dead_roots)) {
2199 		gang[0] = list_entry(fs_info->dead_roots.next,
2200 				     struct btrfs_root, root_list);
2201 		list_del(&gang[0]->root_list);
2202 
2203 		if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2204 			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2205 		btrfs_put_root(gang[0]);
2206 	}
2207 
2208 	while (1) {
2209 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2210 					     (void **)gang, 0,
2211 					     ARRAY_SIZE(gang));
2212 		if (!ret)
2213 			break;
2214 		for (i = 0; i < ret; i++)
2215 			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2216 	}
2217 }
2218 
2219 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2220 {
2221 	mutex_init(&fs_info->scrub_lock);
2222 	atomic_set(&fs_info->scrubs_running, 0);
2223 	atomic_set(&fs_info->scrub_pause_req, 0);
2224 	atomic_set(&fs_info->scrubs_paused, 0);
2225 	atomic_set(&fs_info->scrub_cancel_req, 0);
2226 	init_waitqueue_head(&fs_info->scrub_pause_wait);
2227 	refcount_set(&fs_info->scrub_workers_refcnt, 0);
2228 }
2229 
2230 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2231 {
2232 	spin_lock_init(&fs_info->balance_lock);
2233 	mutex_init(&fs_info->balance_mutex);
2234 	atomic_set(&fs_info->balance_pause_req, 0);
2235 	atomic_set(&fs_info->balance_cancel_req, 0);
2236 	fs_info->balance_ctl = NULL;
2237 	init_waitqueue_head(&fs_info->balance_wait_q);
2238 	atomic_set(&fs_info->reloc_cancel_req, 0);
2239 }
2240 
2241 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2242 {
2243 	struct inode *inode = fs_info->btree_inode;
2244 
2245 	inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2246 	set_nlink(inode, 1);
2247 	/*
2248 	 * we set the i_size on the btree inode to the max possible int.
2249 	 * the real end of the address space is determined by all of
2250 	 * the devices in the system
2251 	 */
2252 	inode->i_size = OFFSET_MAX;
2253 	inode->i_mapping->a_ops = &btree_aops;
2254 
2255 	RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2256 	extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2257 			    IO_TREE_BTREE_INODE_IO, inode);
2258 	BTRFS_I(inode)->io_tree.track_uptodate = false;
2259 	extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2260 
2261 	BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2262 	memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2263 	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2264 	btrfs_insert_inode_hash(inode);
2265 }
2266 
2267 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2268 {
2269 	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2270 	init_rwsem(&fs_info->dev_replace.rwsem);
2271 	init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2272 }
2273 
2274 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2275 {
2276 	spin_lock_init(&fs_info->qgroup_lock);
2277 	mutex_init(&fs_info->qgroup_ioctl_lock);
2278 	fs_info->qgroup_tree = RB_ROOT;
2279 	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2280 	fs_info->qgroup_seq = 1;
2281 	fs_info->qgroup_ulist = NULL;
2282 	fs_info->qgroup_rescan_running = false;
2283 	mutex_init(&fs_info->qgroup_rescan_lock);
2284 }
2285 
2286 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2287 		struct btrfs_fs_devices *fs_devices)
2288 {
2289 	u32 max_active = fs_info->thread_pool_size;
2290 	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2291 
2292 	fs_info->workers =
2293 		btrfs_alloc_workqueue(fs_info, "worker",
2294 				      flags | WQ_HIGHPRI, max_active, 16);
2295 
2296 	fs_info->delalloc_workers =
2297 		btrfs_alloc_workqueue(fs_info, "delalloc",
2298 				      flags, max_active, 2);
2299 
2300 	fs_info->flush_workers =
2301 		btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2302 				      flags, max_active, 0);
2303 
2304 	fs_info->caching_workers =
2305 		btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2306 
2307 	fs_info->fixup_workers =
2308 		btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2309 
2310 	/*
2311 	 * endios are largely parallel and should have a very
2312 	 * low idle thresh
2313 	 */
2314 	fs_info->endio_workers =
2315 		btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2316 	fs_info->endio_meta_workers =
2317 		btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2318 				      max_active, 4);
2319 	fs_info->endio_meta_write_workers =
2320 		btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2321 				      max_active, 2);
2322 	fs_info->endio_raid56_workers =
2323 		btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2324 				      max_active, 4);
2325 	fs_info->rmw_workers =
2326 		btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2327 	fs_info->endio_write_workers =
2328 		btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2329 				      max_active, 2);
2330 	fs_info->endio_freespace_worker =
2331 		btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2332 				      max_active, 0);
2333 	fs_info->delayed_workers =
2334 		btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2335 				      max_active, 0);
2336 	fs_info->readahead_workers =
2337 		btrfs_alloc_workqueue(fs_info, "readahead", flags,
2338 				      max_active, 2);
2339 	fs_info->qgroup_rescan_workers =
2340 		btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2341 	fs_info->discard_ctl.discard_workers =
2342 		alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2343 
2344 	if (!(fs_info->workers && fs_info->delalloc_workers &&
2345 	      fs_info->flush_workers &&
2346 	      fs_info->endio_workers && fs_info->endio_meta_workers &&
2347 	      fs_info->endio_meta_write_workers &&
2348 	      fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2349 	      fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2350 	      fs_info->caching_workers && fs_info->readahead_workers &&
2351 	      fs_info->fixup_workers && fs_info->delayed_workers &&
2352 	      fs_info->qgroup_rescan_workers &&
2353 	      fs_info->discard_ctl.discard_workers)) {
2354 		return -ENOMEM;
2355 	}
2356 
2357 	return 0;
2358 }
2359 
2360 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2361 {
2362 	struct crypto_shash *csum_shash;
2363 	const char *csum_driver = btrfs_super_csum_driver(csum_type);
2364 
2365 	csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2366 
2367 	if (IS_ERR(csum_shash)) {
2368 		btrfs_err(fs_info, "error allocating %s hash for checksum",
2369 			  csum_driver);
2370 		return PTR_ERR(csum_shash);
2371 	}
2372 
2373 	fs_info->csum_shash = csum_shash;
2374 
2375 	return 0;
2376 }
2377 
2378 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2379 			    struct btrfs_fs_devices *fs_devices)
2380 {
2381 	int ret;
2382 	struct btrfs_root *log_tree_root;
2383 	struct btrfs_super_block *disk_super = fs_info->super_copy;
2384 	u64 bytenr = btrfs_super_log_root(disk_super);
2385 	int level = btrfs_super_log_root_level(disk_super);
2386 
2387 	if (fs_devices->rw_devices == 0) {
2388 		btrfs_warn(fs_info, "log replay required on RO media");
2389 		return -EIO;
2390 	}
2391 
2392 	log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2393 					 GFP_KERNEL);
2394 	if (!log_tree_root)
2395 		return -ENOMEM;
2396 
2397 	log_tree_root->node = read_tree_block(fs_info, bytenr,
2398 					      BTRFS_TREE_LOG_OBJECTID,
2399 					      fs_info->generation + 1, level,
2400 					      NULL);
2401 	if (IS_ERR(log_tree_root->node)) {
2402 		btrfs_warn(fs_info, "failed to read log tree");
2403 		ret = PTR_ERR(log_tree_root->node);
2404 		log_tree_root->node = NULL;
2405 		btrfs_put_root(log_tree_root);
2406 		return ret;
2407 	} else if (!extent_buffer_uptodate(log_tree_root->node)) {
2408 		btrfs_err(fs_info, "failed to read log tree");
2409 		btrfs_put_root(log_tree_root);
2410 		return -EIO;
2411 	}
2412 	/* returns with log_tree_root freed on success */
2413 	ret = btrfs_recover_log_trees(log_tree_root);
2414 	if (ret) {
2415 		btrfs_handle_fs_error(fs_info, ret,
2416 				      "Failed to recover log tree");
2417 		btrfs_put_root(log_tree_root);
2418 		return ret;
2419 	}
2420 
2421 	if (sb_rdonly(fs_info->sb)) {
2422 		ret = btrfs_commit_super(fs_info);
2423 		if (ret)
2424 			return ret;
2425 	}
2426 
2427 	return 0;
2428 }
2429 
2430 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2431 {
2432 	struct btrfs_root *tree_root = fs_info->tree_root;
2433 	struct btrfs_root *root;
2434 	struct btrfs_key location;
2435 	int ret;
2436 
2437 	BUG_ON(!fs_info->tree_root);
2438 
2439 	location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2440 	location.type = BTRFS_ROOT_ITEM_KEY;
2441 	location.offset = 0;
2442 
2443 	root = btrfs_read_tree_root(tree_root, &location);
2444 	if (IS_ERR(root)) {
2445 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2446 			ret = PTR_ERR(root);
2447 			goto out;
2448 		}
2449 	} else {
2450 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2451 		fs_info->extent_root = root;
2452 	}
2453 
2454 	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2455 	root = btrfs_read_tree_root(tree_root, &location);
2456 	if (IS_ERR(root)) {
2457 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2458 			ret = PTR_ERR(root);
2459 			goto out;
2460 		}
2461 	} else {
2462 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2463 		fs_info->dev_root = root;
2464 	}
2465 	/* Initialize fs_info for all devices in any case */
2466 	btrfs_init_devices_late(fs_info);
2467 
2468 	/* If IGNOREDATACSUMS is set don't bother reading the csum root. */
2469 	if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2470 		location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2471 		root = btrfs_read_tree_root(tree_root, &location);
2472 		if (IS_ERR(root)) {
2473 			if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2474 				ret = PTR_ERR(root);
2475 				goto out;
2476 			}
2477 		} else {
2478 			set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2479 			fs_info->csum_root = root;
2480 		}
2481 	}
2482 
2483 	/*
2484 	 * This tree can share blocks with some other fs tree during relocation
2485 	 * and we need a proper setup by btrfs_get_fs_root
2486 	 */
2487 	root = btrfs_get_fs_root(tree_root->fs_info,
2488 				 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2489 	if (IS_ERR(root)) {
2490 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2491 			ret = PTR_ERR(root);
2492 			goto out;
2493 		}
2494 	} else {
2495 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2496 		fs_info->data_reloc_root = root;
2497 	}
2498 
2499 	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2500 	root = btrfs_read_tree_root(tree_root, &location);
2501 	if (!IS_ERR(root)) {
2502 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2503 		set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2504 		fs_info->quota_root = root;
2505 	}
2506 
2507 	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2508 	root = btrfs_read_tree_root(tree_root, &location);
2509 	if (IS_ERR(root)) {
2510 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2511 			ret = PTR_ERR(root);
2512 			if (ret != -ENOENT)
2513 				goto out;
2514 		}
2515 	} else {
2516 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2517 		fs_info->uuid_root = root;
2518 	}
2519 
2520 	if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2521 		location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2522 		root = btrfs_read_tree_root(tree_root, &location);
2523 		if (IS_ERR(root)) {
2524 			if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2525 				ret = PTR_ERR(root);
2526 				goto out;
2527 			}
2528 		}  else {
2529 			set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2530 			fs_info->free_space_root = root;
2531 		}
2532 	}
2533 
2534 	return 0;
2535 out:
2536 	btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2537 		   location.objectid, ret);
2538 	return ret;
2539 }
2540 
2541 /*
2542  * Real super block validation
2543  * NOTE: super csum type and incompat features will not be checked here.
2544  *
2545  * @sb:		super block to check
2546  * @mirror_num:	the super block number to check its bytenr:
2547  * 		0	the primary (1st) sb
2548  * 		1, 2	2nd and 3rd backup copy
2549  * 	       -1	skip bytenr check
2550  */
2551 static int validate_super(struct btrfs_fs_info *fs_info,
2552 			    struct btrfs_super_block *sb, int mirror_num)
2553 {
2554 	u64 nodesize = btrfs_super_nodesize(sb);
2555 	u64 sectorsize = btrfs_super_sectorsize(sb);
2556 	int ret = 0;
2557 
2558 	if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2559 		btrfs_err(fs_info, "no valid FS found");
2560 		ret = -EINVAL;
2561 	}
2562 	if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2563 		btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2564 				btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2565 		ret = -EINVAL;
2566 	}
2567 	if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2568 		btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2569 				btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2570 		ret = -EINVAL;
2571 	}
2572 	if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2573 		btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2574 				btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2575 		ret = -EINVAL;
2576 	}
2577 	if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2578 		btrfs_err(fs_info, "log_root level too big: %d >= %d",
2579 				btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2580 		ret = -EINVAL;
2581 	}
2582 
2583 	/*
2584 	 * Check sectorsize and nodesize first, other check will need it.
2585 	 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2586 	 */
2587 	if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2588 	    sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2589 		btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2590 		ret = -EINVAL;
2591 	}
2592 
2593 	/*
2594 	 * For 4K page size, we only support 4K sector size.
2595 	 * For 64K page size, we support 64K and 4K sector sizes.
2596 	 */
2597 	if ((PAGE_SIZE == SZ_4K && sectorsize != PAGE_SIZE) ||
2598 	    (PAGE_SIZE == SZ_64K && (sectorsize != SZ_4K &&
2599 				     sectorsize != SZ_64K))) {
2600 		btrfs_err(fs_info,
2601 			"sectorsize %llu not yet supported for page size %lu",
2602 			sectorsize, PAGE_SIZE);
2603 		ret = -EINVAL;
2604 	}
2605 
2606 	if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2607 	    nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2608 		btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2609 		ret = -EINVAL;
2610 	}
2611 	if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2612 		btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2613 			  le32_to_cpu(sb->__unused_leafsize), nodesize);
2614 		ret = -EINVAL;
2615 	}
2616 
2617 	/* Root alignment check */
2618 	if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2619 		btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2620 			   btrfs_super_root(sb));
2621 		ret = -EINVAL;
2622 	}
2623 	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2624 		btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2625 			   btrfs_super_chunk_root(sb));
2626 		ret = -EINVAL;
2627 	}
2628 	if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2629 		btrfs_warn(fs_info, "log_root block unaligned: %llu",
2630 			   btrfs_super_log_root(sb));
2631 		ret = -EINVAL;
2632 	}
2633 
2634 	if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2635 		   BTRFS_FSID_SIZE)) {
2636 		btrfs_err(fs_info,
2637 		"superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2638 			fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2639 		ret = -EINVAL;
2640 	}
2641 
2642 	if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2643 	    memcmp(fs_info->fs_devices->metadata_uuid,
2644 		   fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2645 		btrfs_err(fs_info,
2646 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2647 			fs_info->super_copy->metadata_uuid,
2648 			fs_info->fs_devices->metadata_uuid);
2649 		ret = -EINVAL;
2650 	}
2651 
2652 	if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2653 		   BTRFS_FSID_SIZE) != 0) {
2654 		btrfs_err(fs_info,
2655 			"dev_item UUID does not match metadata fsid: %pU != %pU",
2656 			fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2657 		ret = -EINVAL;
2658 	}
2659 
2660 	/*
2661 	 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2662 	 * done later
2663 	 */
2664 	if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2665 		btrfs_err(fs_info, "bytes_used is too small %llu",
2666 			  btrfs_super_bytes_used(sb));
2667 		ret = -EINVAL;
2668 	}
2669 	if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2670 		btrfs_err(fs_info, "invalid stripesize %u",
2671 			  btrfs_super_stripesize(sb));
2672 		ret = -EINVAL;
2673 	}
2674 	if (btrfs_super_num_devices(sb) > (1UL << 31))
2675 		btrfs_warn(fs_info, "suspicious number of devices: %llu",
2676 			   btrfs_super_num_devices(sb));
2677 	if (btrfs_super_num_devices(sb) == 0) {
2678 		btrfs_err(fs_info, "number of devices is 0");
2679 		ret = -EINVAL;
2680 	}
2681 
2682 	if (mirror_num >= 0 &&
2683 	    btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2684 		btrfs_err(fs_info, "super offset mismatch %llu != %u",
2685 			  btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2686 		ret = -EINVAL;
2687 	}
2688 
2689 	/*
2690 	 * Obvious sys_chunk_array corruptions, it must hold at least one key
2691 	 * and one chunk
2692 	 */
2693 	if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2694 		btrfs_err(fs_info, "system chunk array too big %u > %u",
2695 			  btrfs_super_sys_array_size(sb),
2696 			  BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2697 		ret = -EINVAL;
2698 	}
2699 	if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2700 			+ sizeof(struct btrfs_chunk)) {
2701 		btrfs_err(fs_info, "system chunk array too small %u < %zu",
2702 			  btrfs_super_sys_array_size(sb),
2703 			  sizeof(struct btrfs_disk_key)
2704 			  + sizeof(struct btrfs_chunk));
2705 		ret = -EINVAL;
2706 	}
2707 
2708 	/*
2709 	 * The generation is a global counter, we'll trust it more than the others
2710 	 * but it's still possible that it's the one that's wrong.
2711 	 */
2712 	if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2713 		btrfs_warn(fs_info,
2714 			"suspicious: generation < chunk_root_generation: %llu < %llu",
2715 			btrfs_super_generation(sb),
2716 			btrfs_super_chunk_root_generation(sb));
2717 	if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2718 	    && btrfs_super_cache_generation(sb) != (u64)-1)
2719 		btrfs_warn(fs_info,
2720 			"suspicious: generation < cache_generation: %llu < %llu",
2721 			btrfs_super_generation(sb),
2722 			btrfs_super_cache_generation(sb));
2723 
2724 	return ret;
2725 }
2726 
2727 /*
2728  * Validation of super block at mount time.
2729  * Some checks already done early at mount time, like csum type and incompat
2730  * flags will be skipped.
2731  */
2732 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2733 {
2734 	return validate_super(fs_info, fs_info->super_copy, 0);
2735 }
2736 
2737 /*
2738  * Validation of super block at write time.
2739  * Some checks like bytenr check will be skipped as their values will be
2740  * overwritten soon.
2741  * Extra checks like csum type and incompat flags will be done here.
2742  */
2743 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2744 				      struct btrfs_super_block *sb)
2745 {
2746 	int ret;
2747 
2748 	ret = validate_super(fs_info, sb, -1);
2749 	if (ret < 0)
2750 		goto out;
2751 	if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2752 		ret = -EUCLEAN;
2753 		btrfs_err(fs_info, "invalid csum type, has %u want %u",
2754 			  btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2755 		goto out;
2756 	}
2757 	if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2758 		ret = -EUCLEAN;
2759 		btrfs_err(fs_info,
2760 		"invalid incompat flags, has 0x%llx valid mask 0x%llx",
2761 			  btrfs_super_incompat_flags(sb),
2762 			  (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2763 		goto out;
2764 	}
2765 out:
2766 	if (ret < 0)
2767 		btrfs_err(fs_info,
2768 		"super block corruption detected before writing it to disk");
2769 	return ret;
2770 }
2771 
2772 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2773 {
2774 	int backup_index = find_newest_super_backup(fs_info);
2775 	struct btrfs_super_block *sb = fs_info->super_copy;
2776 	struct btrfs_root *tree_root = fs_info->tree_root;
2777 	bool handle_error = false;
2778 	int ret = 0;
2779 	int i;
2780 
2781 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2782 		u64 generation;
2783 		int level;
2784 
2785 		if (handle_error) {
2786 			if (!IS_ERR(tree_root->node))
2787 				free_extent_buffer(tree_root->node);
2788 			tree_root->node = NULL;
2789 
2790 			if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2791 				break;
2792 
2793 			free_root_pointers(fs_info, 0);
2794 
2795 			/*
2796 			 * Don't use the log in recovery mode, it won't be
2797 			 * valid
2798 			 */
2799 			btrfs_set_super_log_root(sb, 0);
2800 
2801 			/* We can't trust the free space cache either */
2802 			btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2803 
2804 			ret = read_backup_root(fs_info, i);
2805 			backup_index = ret;
2806 			if (ret < 0)
2807 				return ret;
2808 		}
2809 		generation = btrfs_super_generation(sb);
2810 		level = btrfs_super_root_level(sb);
2811 		tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2812 						  BTRFS_ROOT_TREE_OBJECTID,
2813 						  generation, level, NULL);
2814 		if (IS_ERR(tree_root->node)) {
2815 			handle_error = true;
2816 			ret = PTR_ERR(tree_root->node);
2817 			tree_root->node = NULL;
2818 			btrfs_warn(fs_info, "couldn't read tree root");
2819 			continue;
2820 
2821 		} else if (!extent_buffer_uptodate(tree_root->node)) {
2822 			handle_error = true;
2823 			ret = -EIO;
2824 			btrfs_warn(fs_info, "error while reading tree root");
2825 			continue;
2826 		}
2827 
2828 		btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2829 		tree_root->commit_root = btrfs_root_node(tree_root);
2830 		btrfs_set_root_refs(&tree_root->root_item, 1);
2831 
2832 		/*
2833 		 * No need to hold btrfs_root::objectid_mutex since the fs
2834 		 * hasn't been fully initialised and we are the only user
2835 		 */
2836 		ret = btrfs_init_root_free_objectid(tree_root);
2837 		if (ret < 0) {
2838 			handle_error = true;
2839 			continue;
2840 		}
2841 
2842 		ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2843 
2844 		ret = btrfs_read_roots(fs_info);
2845 		if (ret < 0) {
2846 			handle_error = true;
2847 			continue;
2848 		}
2849 
2850 		/* All successful */
2851 		fs_info->generation = generation;
2852 		fs_info->last_trans_committed = generation;
2853 
2854 		/* Always begin writing backup roots after the one being used */
2855 		if (backup_index < 0) {
2856 			fs_info->backup_root_index = 0;
2857 		} else {
2858 			fs_info->backup_root_index = backup_index + 1;
2859 			fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2860 		}
2861 		break;
2862 	}
2863 
2864 	return ret;
2865 }
2866 
2867 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2868 {
2869 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2870 	INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2871 	INIT_LIST_HEAD(&fs_info->trans_list);
2872 	INIT_LIST_HEAD(&fs_info->dead_roots);
2873 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
2874 	INIT_LIST_HEAD(&fs_info->delalloc_roots);
2875 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
2876 	spin_lock_init(&fs_info->delalloc_root_lock);
2877 	spin_lock_init(&fs_info->trans_lock);
2878 	spin_lock_init(&fs_info->fs_roots_radix_lock);
2879 	spin_lock_init(&fs_info->delayed_iput_lock);
2880 	spin_lock_init(&fs_info->defrag_inodes_lock);
2881 	spin_lock_init(&fs_info->super_lock);
2882 	spin_lock_init(&fs_info->buffer_lock);
2883 	spin_lock_init(&fs_info->unused_bgs_lock);
2884 	spin_lock_init(&fs_info->treelog_bg_lock);
2885 	spin_lock_init(&fs_info->zone_active_bgs_lock);
2886 	spin_lock_init(&fs_info->relocation_bg_lock);
2887 	rwlock_init(&fs_info->tree_mod_log_lock);
2888 	mutex_init(&fs_info->unused_bg_unpin_mutex);
2889 	mutex_init(&fs_info->reclaim_bgs_lock);
2890 	mutex_init(&fs_info->reloc_mutex);
2891 	mutex_init(&fs_info->delalloc_root_mutex);
2892 	mutex_init(&fs_info->zoned_meta_io_lock);
2893 	seqlock_init(&fs_info->profiles_lock);
2894 
2895 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2896 	INIT_LIST_HEAD(&fs_info->space_info);
2897 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2898 	INIT_LIST_HEAD(&fs_info->unused_bgs);
2899 	INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2900 	INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2901 #ifdef CONFIG_BTRFS_DEBUG
2902 	INIT_LIST_HEAD(&fs_info->allocated_roots);
2903 	INIT_LIST_HEAD(&fs_info->allocated_ebs);
2904 	spin_lock_init(&fs_info->eb_leak_lock);
2905 #endif
2906 	extent_map_tree_init(&fs_info->mapping_tree);
2907 	btrfs_init_block_rsv(&fs_info->global_block_rsv,
2908 			     BTRFS_BLOCK_RSV_GLOBAL);
2909 	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2910 	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2911 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2912 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2913 			     BTRFS_BLOCK_RSV_DELOPS);
2914 	btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2915 			     BTRFS_BLOCK_RSV_DELREFS);
2916 
2917 	atomic_set(&fs_info->async_delalloc_pages, 0);
2918 	atomic_set(&fs_info->defrag_running, 0);
2919 	atomic_set(&fs_info->reada_works_cnt, 0);
2920 	atomic_set(&fs_info->nr_delayed_iputs, 0);
2921 	atomic64_set(&fs_info->tree_mod_seq, 0);
2922 	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2923 	fs_info->metadata_ratio = 0;
2924 	fs_info->defrag_inodes = RB_ROOT;
2925 	atomic64_set(&fs_info->free_chunk_space, 0);
2926 	fs_info->tree_mod_log = RB_ROOT;
2927 	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2928 	fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2929 	/* readahead state */
2930 	INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2931 	spin_lock_init(&fs_info->reada_lock);
2932 	btrfs_init_ref_verify(fs_info);
2933 
2934 	fs_info->thread_pool_size = min_t(unsigned long,
2935 					  num_online_cpus() + 2, 8);
2936 
2937 	INIT_LIST_HEAD(&fs_info->ordered_roots);
2938 	spin_lock_init(&fs_info->ordered_root_lock);
2939 
2940 	btrfs_init_scrub(fs_info);
2941 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2942 	fs_info->check_integrity_print_mask = 0;
2943 #endif
2944 	btrfs_init_balance(fs_info);
2945 	btrfs_init_async_reclaim_work(fs_info);
2946 
2947 	spin_lock_init(&fs_info->block_group_cache_lock);
2948 	fs_info->block_group_cache_tree = RB_ROOT;
2949 	fs_info->first_logical_byte = (u64)-1;
2950 
2951 	extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2952 			    IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2953 	set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2954 
2955 	mutex_init(&fs_info->ordered_operations_mutex);
2956 	mutex_init(&fs_info->tree_log_mutex);
2957 	mutex_init(&fs_info->chunk_mutex);
2958 	mutex_init(&fs_info->transaction_kthread_mutex);
2959 	mutex_init(&fs_info->cleaner_mutex);
2960 	mutex_init(&fs_info->ro_block_group_mutex);
2961 	init_rwsem(&fs_info->commit_root_sem);
2962 	init_rwsem(&fs_info->cleanup_work_sem);
2963 	init_rwsem(&fs_info->subvol_sem);
2964 	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2965 
2966 	btrfs_init_dev_replace_locks(fs_info);
2967 	btrfs_init_qgroup(fs_info);
2968 	btrfs_discard_init(fs_info);
2969 
2970 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2971 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2972 
2973 	init_waitqueue_head(&fs_info->transaction_throttle);
2974 	init_waitqueue_head(&fs_info->transaction_wait);
2975 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2976 	init_waitqueue_head(&fs_info->async_submit_wait);
2977 	init_waitqueue_head(&fs_info->delayed_iputs_wait);
2978 
2979 	/* Usable values until the real ones are cached from the superblock */
2980 	fs_info->nodesize = 4096;
2981 	fs_info->sectorsize = 4096;
2982 	fs_info->sectorsize_bits = ilog2(4096);
2983 	fs_info->stripesize = 4096;
2984 
2985 	spin_lock_init(&fs_info->swapfile_pins_lock);
2986 	fs_info->swapfile_pins = RB_ROOT;
2987 
2988 	spin_lock_init(&fs_info->send_reloc_lock);
2989 	fs_info->send_in_progress = 0;
2990 
2991 	fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2992 	INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2993 }
2994 
2995 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2996 {
2997 	int ret;
2998 
2999 	fs_info->sb = sb;
3000 	sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
3001 	sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3002 
3003 	ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3004 	if (ret)
3005 		return ret;
3006 
3007 	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3008 	if (ret)
3009 		return ret;
3010 
3011 	fs_info->dirty_metadata_batch = PAGE_SIZE *
3012 					(1 + ilog2(nr_cpu_ids));
3013 
3014 	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3015 	if (ret)
3016 		return ret;
3017 
3018 	ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3019 			GFP_KERNEL);
3020 	if (ret)
3021 		return ret;
3022 
3023 	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3024 					GFP_KERNEL);
3025 	if (!fs_info->delayed_root)
3026 		return -ENOMEM;
3027 	btrfs_init_delayed_root(fs_info->delayed_root);
3028 
3029 	if (sb_rdonly(sb))
3030 		set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3031 
3032 	return btrfs_alloc_stripe_hash_table(fs_info);
3033 }
3034 
3035 static int btrfs_uuid_rescan_kthread(void *data)
3036 {
3037 	struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3038 	int ret;
3039 
3040 	/*
3041 	 * 1st step is to iterate through the existing UUID tree and
3042 	 * to delete all entries that contain outdated data.
3043 	 * 2nd step is to add all missing entries to the UUID tree.
3044 	 */
3045 	ret = btrfs_uuid_tree_iterate(fs_info);
3046 	if (ret < 0) {
3047 		if (ret != -EINTR)
3048 			btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3049 				   ret);
3050 		up(&fs_info->uuid_tree_rescan_sem);
3051 		return ret;
3052 	}
3053 	return btrfs_uuid_scan_kthread(data);
3054 }
3055 
3056 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3057 {
3058 	struct task_struct *task;
3059 
3060 	down(&fs_info->uuid_tree_rescan_sem);
3061 	task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3062 	if (IS_ERR(task)) {
3063 		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
3064 		btrfs_warn(fs_info, "failed to start uuid_rescan task");
3065 		up(&fs_info->uuid_tree_rescan_sem);
3066 		return PTR_ERR(task);
3067 	}
3068 
3069 	return 0;
3070 }
3071 
3072 /*
3073  * Some options only have meaning at mount time and shouldn't persist across
3074  * remounts, or be displayed. Clear these at the end of mount and remount
3075  * code paths.
3076  */
3077 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3078 {
3079 	btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3080 	btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3081 }
3082 
3083 /*
3084  * Mounting logic specific to read-write file systems. Shared by open_ctree
3085  * and btrfs_remount when remounting from read-only to read-write.
3086  */
3087 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3088 {
3089 	int ret;
3090 	const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3091 	bool clear_free_space_tree = false;
3092 
3093 	if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3094 	    btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3095 		clear_free_space_tree = true;
3096 	} else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3097 		   !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3098 		btrfs_warn(fs_info, "free space tree is invalid");
3099 		clear_free_space_tree = true;
3100 	}
3101 
3102 	if (clear_free_space_tree) {
3103 		btrfs_info(fs_info, "clearing free space tree");
3104 		ret = btrfs_clear_free_space_tree(fs_info);
3105 		if (ret) {
3106 			btrfs_warn(fs_info,
3107 				   "failed to clear free space tree: %d", ret);
3108 			goto out;
3109 		}
3110 	}
3111 
3112 	/*
3113 	 * btrfs_find_orphan_roots() is responsible for finding all the dead
3114 	 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3115 	 * them into the fs_info->fs_roots_radix tree. This must be done before
3116 	 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3117 	 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3118 	 * item before the root's tree is deleted - this means that if we unmount
3119 	 * or crash before the deletion completes, on the next mount we will not
3120 	 * delete what remains of the tree because the orphan item does not
3121 	 * exists anymore, which is what tells us we have a pending deletion.
3122 	 */
3123 	ret = btrfs_find_orphan_roots(fs_info);
3124 	if (ret)
3125 		goto out;
3126 
3127 	ret = btrfs_cleanup_fs_roots(fs_info);
3128 	if (ret)
3129 		goto out;
3130 
3131 	down_read(&fs_info->cleanup_work_sem);
3132 	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3133 	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3134 		up_read(&fs_info->cleanup_work_sem);
3135 		goto out;
3136 	}
3137 	up_read(&fs_info->cleanup_work_sem);
3138 
3139 	mutex_lock(&fs_info->cleaner_mutex);
3140 	ret = btrfs_recover_relocation(fs_info->tree_root);
3141 	mutex_unlock(&fs_info->cleaner_mutex);
3142 	if (ret < 0) {
3143 		btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3144 		goto out;
3145 	}
3146 
3147 	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3148 	    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3149 		btrfs_info(fs_info, "creating free space tree");
3150 		ret = btrfs_create_free_space_tree(fs_info);
3151 		if (ret) {
3152 			btrfs_warn(fs_info,
3153 				"failed to create free space tree: %d", ret);
3154 			goto out;
3155 		}
3156 	}
3157 
3158 	if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3159 		ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3160 		if (ret)
3161 			goto out;
3162 	}
3163 
3164 	ret = btrfs_resume_balance_async(fs_info);
3165 	if (ret)
3166 		goto out;
3167 
3168 	ret = btrfs_resume_dev_replace_async(fs_info);
3169 	if (ret) {
3170 		btrfs_warn(fs_info, "failed to resume dev_replace");
3171 		goto out;
3172 	}
3173 
3174 	btrfs_qgroup_rescan_resume(fs_info);
3175 
3176 	if (!fs_info->uuid_root) {
3177 		btrfs_info(fs_info, "creating UUID tree");
3178 		ret = btrfs_create_uuid_tree(fs_info);
3179 		if (ret) {
3180 			btrfs_warn(fs_info,
3181 				   "failed to create the UUID tree %d", ret);
3182 			goto out;
3183 		}
3184 	}
3185 
3186 out:
3187 	return ret;
3188 }
3189 
3190 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3191 		      char *options)
3192 {
3193 	u32 sectorsize;
3194 	u32 nodesize;
3195 	u32 stripesize;
3196 	u64 generation;
3197 	u64 features;
3198 	u16 csum_type;
3199 	struct btrfs_super_block *disk_super;
3200 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3201 	struct btrfs_root *tree_root;
3202 	struct btrfs_root *chunk_root;
3203 	int ret;
3204 	int err = -EINVAL;
3205 	int level;
3206 
3207 	ret = init_mount_fs_info(fs_info, sb);
3208 	if (ret) {
3209 		err = ret;
3210 		goto fail;
3211 	}
3212 
3213 	/* These need to be init'ed before we start creating inodes and such. */
3214 	tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3215 				     GFP_KERNEL);
3216 	fs_info->tree_root = tree_root;
3217 	chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3218 				      GFP_KERNEL);
3219 	fs_info->chunk_root = chunk_root;
3220 	if (!tree_root || !chunk_root) {
3221 		err = -ENOMEM;
3222 		goto fail;
3223 	}
3224 
3225 	fs_info->btree_inode = new_inode(sb);
3226 	if (!fs_info->btree_inode) {
3227 		err = -ENOMEM;
3228 		goto fail;
3229 	}
3230 	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3231 	btrfs_init_btree_inode(fs_info);
3232 
3233 	invalidate_bdev(fs_devices->latest_dev->bdev);
3234 
3235 	/*
3236 	 * Read super block and check the signature bytes only
3237 	 */
3238 	disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3239 	if (IS_ERR(disk_super)) {
3240 		err = PTR_ERR(disk_super);
3241 		goto fail_alloc;
3242 	}
3243 
3244 	/*
3245 	 * Verify the type first, if that or the checksum value are
3246 	 * corrupted, we'll find out
3247 	 */
3248 	csum_type = btrfs_super_csum_type(disk_super);
3249 	if (!btrfs_supported_super_csum(csum_type)) {
3250 		btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3251 			  csum_type);
3252 		err = -EINVAL;
3253 		btrfs_release_disk_super(disk_super);
3254 		goto fail_alloc;
3255 	}
3256 
3257 	fs_info->csum_size = btrfs_super_csum_size(disk_super);
3258 
3259 	ret = btrfs_init_csum_hash(fs_info, csum_type);
3260 	if (ret) {
3261 		err = ret;
3262 		btrfs_release_disk_super(disk_super);
3263 		goto fail_alloc;
3264 	}
3265 
3266 	/*
3267 	 * We want to check superblock checksum, the type is stored inside.
3268 	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3269 	 */
3270 	if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
3271 		btrfs_err(fs_info, "superblock checksum mismatch");
3272 		err = -EINVAL;
3273 		btrfs_release_disk_super(disk_super);
3274 		goto fail_alloc;
3275 	}
3276 
3277 	/*
3278 	 * super_copy is zeroed at allocation time and we never touch the
3279 	 * following bytes up to INFO_SIZE, the checksum is calculated from
3280 	 * the whole block of INFO_SIZE
3281 	 */
3282 	memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3283 	btrfs_release_disk_super(disk_super);
3284 
3285 	disk_super = fs_info->super_copy;
3286 
3287 
3288 	features = btrfs_super_flags(disk_super);
3289 	if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3290 		features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3291 		btrfs_set_super_flags(disk_super, features);
3292 		btrfs_info(fs_info,
3293 			"found metadata UUID change in progress flag, clearing");
3294 	}
3295 
3296 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
3297 	       sizeof(*fs_info->super_for_commit));
3298 
3299 	ret = btrfs_validate_mount_super(fs_info);
3300 	if (ret) {
3301 		btrfs_err(fs_info, "superblock contains fatal errors");
3302 		err = -EINVAL;
3303 		goto fail_alloc;
3304 	}
3305 
3306 	if (!btrfs_super_root(disk_super))
3307 		goto fail_alloc;
3308 
3309 	/* check FS state, whether FS is broken. */
3310 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3311 		set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3312 
3313 	/*
3314 	 * In the long term, we'll store the compression type in the super
3315 	 * block, and it'll be used for per file compression control.
3316 	 */
3317 	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3318 
3319 	/*
3320 	 * Flag our filesystem as having big metadata blocks if they are bigger
3321 	 * than the page size.
3322 	 */
3323 	if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3324 		if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3325 			btrfs_info(fs_info,
3326 				"flagging fs with big metadata feature");
3327 		features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3328 	}
3329 
3330 	/* Set up fs_info before parsing mount options */
3331 	nodesize = btrfs_super_nodesize(disk_super);
3332 	sectorsize = btrfs_super_sectorsize(disk_super);
3333 	stripesize = sectorsize;
3334 	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3335 	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3336 
3337 	fs_info->nodesize = nodesize;
3338 	fs_info->sectorsize = sectorsize;
3339 	fs_info->sectorsize_bits = ilog2(sectorsize);
3340 	fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3341 	fs_info->stripesize = stripesize;
3342 
3343 	ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3344 	if (ret) {
3345 		err = ret;
3346 		goto fail_alloc;
3347 	}
3348 
3349 	features = btrfs_super_incompat_flags(disk_super) &
3350 		~BTRFS_FEATURE_INCOMPAT_SUPP;
3351 	if (features) {
3352 		btrfs_err(fs_info,
3353 		    "cannot mount because of unsupported optional features (%llx)",
3354 		    features);
3355 		err = -EINVAL;
3356 		goto fail_alloc;
3357 	}
3358 
3359 	features = btrfs_super_incompat_flags(disk_super);
3360 	features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3361 	if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3362 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3363 	else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3364 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3365 
3366 	if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3367 		btrfs_info(fs_info, "has skinny extents");
3368 
3369 	/*
3370 	 * mixed block groups end up with duplicate but slightly offset
3371 	 * extent buffers for the same range.  It leads to corruptions
3372 	 */
3373 	if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3374 	    (sectorsize != nodesize)) {
3375 		btrfs_err(fs_info,
3376 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3377 			nodesize, sectorsize);
3378 		goto fail_alloc;
3379 	}
3380 
3381 	/*
3382 	 * Needn't use the lock because there is no other task which will
3383 	 * update the flag.
3384 	 */
3385 	btrfs_set_super_incompat_flags(disk_super, features);
3386 
3387 	features = btrfs_super_compat_ro_flags(disk_super) &
3388 		~BTRFS_FEATURE_COMPAT_RO_SUPP;
3389 	if (!sb_rdonly(sb) && features) {
3390 		btrfs_err(fs_info,
3391 	"cannot mount read-write because of unsupported optional features (%llx)",
3392 		       features);
3393 		err = -EINVAL;
3394 		goto fail_alloc;
3395 	}
3396 
3397 	if (sectorsize < PAGE_SIZE) {
3398 		struct btrfs_subpage_info *subpage_info;
3399 
3400 		btrfs_warn(fs_info,
3401 		"read-write for sector size %u with page size %lu is experimental",
3402 			   sectorsize, PAGE_SIZE);
3403 		if (btrfs_super_incompat_flags(fs_info->super_copy) &
3404 			BTRFS_FEATURE_INCOMPAT_RAID56) {
3405 			btrfs_err(fs_info,
3406 		"RAID56 is not yet supported for sector size %u with page size %lu",
3407 				sectorsize, PAGE_SIZE);
3408 			err = -EINVAL;
3409 			goto fail_alloc;
3410 		}
3411 		subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3412 		if (!subpage_info)
3413 			goto fail_alloc;
3414 		btrfs_init_subpage_info(subpage_info, sectorsize);
3415 		fs_info->subpage_info = subpage_info;
3416 	}
3417 
3418 	ret = btrfs_init_workqueues(fs_info, fs_devices);
3419 	if (ret) {
3420 		err = ret;
3421 		goto fail_sb_buffer;
3422 	}
3423 
3424 	sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3425 	sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3426 
3427 	sb->s_blocksize = sectorsize;
3428 	sb->s_blocksize_bits = blksize_bits(sectorsize);
3429 	memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3430 
3431 	mutex_lock(&fs_info->chunk_mutex);
3432 	ret = btrfs_read_sys_array(fs_info);
3433 	mutex_unlock(&fs_info->chunk_mutex);
3434 	if (ret) {
3435 		btrfs_err(fs_info, "failed to read the system array: %d", ret);
3436 		goto fail_sb_buffer;
3437 	}
3438 
3439 	generation = btrfs_super_chunk_root_generation(disk_super);
3440 	level = btrfs_super_chunk_root_level(disk_super);
3441 
3442 	chunk_root->node = read_tree_block(fs_info,
3443 					   btrfs_super_chunk_root(disk_super),
3444 					   BTRFS_CHUNK_TREE_OBJECTID,
3445 					   generation, level, NULL);
3446 	if (IS_ERR(chunk_root->node) ||
3447 	    !extent_buffer_uptodate(chunk_root->node)) {
3448 		btrfs_err(fs_info, "failed to read chunk root");
3449 		if (!IS_ERR(chunk_root->node))
3450 			free_extent_buffer(chunk_root->node);
3451 		chunk_root->node = NULL;
3452 		goto fail_tree_roots;
3453 	}
3454 	btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3455 	chunk_root->commit_root = btrfs_root_node(chunk_root);
3456 
3457 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3458 			   offsetof(struct btrfs_header, chunk_tree_uuid),
3459 			   BTRFS_UUID_SIZE);
3460 
3461 	ret = btrfs_read_chunk_tree(fs_info);
3462 	if (ret) {
3463 		btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3464 		goto fail_tree_roots;
3465 	}
3466 
3467 	/*
3468 	 * At this point we know all the devices that make this filesystem,
3469 	 * including the seed devices but we don't know yet if the replace
3470 	 * target is required. So free devices that are not part of this
3471 	 * filesystem but skip the replace target device which is checked
3472 	 * below in btrfs_init_dev_replace().
3473 	 */
3474 	btrfs_free_extra_devids(fs_devices);
3475 	if (!fs_devices->latest_dev->bdev) {
3476 		btrfs_err(fs_info, "failed to read devices");
3477 		goto fail_tree_roots;
3478 	}
3479 
3480 	ret = init_tree_roots(fs_info);
3481 	if (ret)
3482 		goto fail_tree_roots;
3483 
3484 	/*
3485 	 * Get zone type information of zoned block devices. This will also
3486 	 * handle emulation of a zoned filesystem if a regular device has the
3487 	 * zoned incompat feature flag set.
3488 	 */
3489 	ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3490 	if (ret) {
3491 		btrfs_err(fs_info,
3492 			  "zoned: failed to read device zone info: %d",
3493 			  ret);
3494 		goto fail_block_groups;
3495 	}
3496 
3497 	/*
3498 	 * If we have a uuid root and we're not being told to rescan we need to
3499 	 * check the generation here so we can set the
3500 	 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3501 	 * transaction during a balance or the log replay without updating the
3502 	 * uuid generation, and then if we crash we would rescan the uuid tree,
3503 	 * even though it was perfectly fine.
3504 	 */
3505 	if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3506 	    fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3507 		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3508 
3509 	ret = btrfs_verify_dev_extents(fs_info);
3510 	if (ret) {
3511 		btrfs_err(fs_info,
3512 			  "failed to verify dev extents against chunks: %d",
3513 			  ret);
3514 		goto fail_block_groups;
3515 	}
3516 	ret = btrfs_recover_balance(fs_info);
3517 	if (ret) {
3518 		btrfs_err(fs_info, "failed to recover balance: %d", ret);
3519 		goto fail_block_groups;
3520 	}
3521 
3522 	ret = btrfs_init_dev_stats(fs_info);
3523 	if (ret) {
3524 		btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3525 		goto fail_block_groups;
3526 	}
3527 
3528 	ret = btrfs_init_dev_replace(fs_info);
3529 	if (ret) {
3530 		btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3531 		goto fail_block_groups;
3532 	}
3533 
3534 	ret = btrfs_check_zoned_mode(fs_info);
3535 	if (ret) {
3536 		btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3537 			  ret);
3538 		goto fail_block_groups;
3539 	}
3540 
3541 	ret = btrfs_sysfs_add_fsid(fs_devices);
3542 	if (ret) {
3543 		btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3544 				ret);
3545 		goto fail_block_groups;
3546 	}
3547 
3548 	ret = btrfs_sysfs_add_mounted(fs_info);
3549 	if (ret) {
3550 		btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3551 		goto fail_fsdev_sysfs;
3552 	}
3553 
3554 	ret = btrfs_init_space_info(fs_info);
3555 	if (ret) {
3556 		btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3557 		goto fail_sysfs;
3558 	}
3559 
3560 	ret = btrfs_read_block_groups(fs_info);
3561 	if (ret) {
3562 		btrfs_err(fs_info, "failed to read block groups: %d", ret);
3563 		goto fail_sysfs;
3564 	}
3565 
3566 	if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3567 	    !btrfs_check_rw_degradable(fs_info, NULL)) {
3568 		btrfs_warn(fs_info,
3569 		"writable mount is not allowed due to too many missing devices");
3570 		goto fail_sysfs;
3571 	}
3572 
3573 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3574 					       "btrfs-cleaner");
3575 	if (IS_ERR(fs_info->cleaner_kthread))
3576 		goto fail_sysfs;
3577 
3578 	fs_info->transaction_kthread = kthread_run(transaction_kthread,
3579 						   tree_root,
3580 						   "btrfs-transaction");
3581 	if (IS_ERR(fs_info->transaction_kthread))
3582 		goto fail_cleaner;
3583 
3584 	if (!btrfs_test_opt(fs_info, NOSSD) &&
3585 	    !fs_info->fs_devices->rotating) {
3586 		btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3587 	}
3588 
3589 	/*
3590 	 * Mount does not set all options immediately, we can do it now and do
3591 	 * not have to wait for transaction commit
3592 	 */
3593 	btrfs_apply_pending_changes(fs_info);
3594 
3595 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3596 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3597 		ret = btrfsic_mount(fs_info, fs_devices,
3598 				    btrfs_test_opt(fs_info,
3599 					CHECK_INTEGRITY_DATA) ? 1 : 0,
3600 				    fs_info->check_integrity_print_mask);
3601 		if (ret)
3602 			btrfs_warn(fs_info,
3603 				"failed to initialize integrity check module: %d",
3604 				ret);
3605 	}
3606 #endif
3607 	ret = btrfs_read_qgroup_config(fs_info);
3608 	if (ret)
3609 		goto fail_trans_kthread;
3610 
3611 	if (btrfs_build_ref_tree(fs_info))
3612 		btrfs_err(fs_info, "couldn't build ref tree");
3613 
3614 	/* do not make disk changes in broken FS or nologreplay is given */
3615 	if (btrfs_super_log_root(disk_super) != 0 &&
3616 	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3617 		btrfs_info(fs_info, "start tree-log replay");
3618 		ret = btrfs_replay_log(fs_info, fs_devices);
3619 		if (ret) {
3620 			err = ret;
3621 			goto fail_qgroup;
3622 		}
3623 	}
3624 
3625 	fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3626 	if (IS_ERR(fs_info->fs_root)) {
3627 		err = PTR_ERR(fs_info->fs_root);
3628 		btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3629 		fs_info->fs_root = NULL;
3630 		goto fail_qgroup;
3631 	}
3632 
3633 	if (sb_rdonly(sb))
3634 		goto clear_oneshot;
3635 
3636 	ret = btrfs_start_pre_rw_mount(fs_info);
3637 	if (ret) {
3638 		close_ctree(fs_info);
3639 		return ret;
3640 	}
3641 	btrfs_discard_resume(fs_info);
3642 
3643 	if (fs_info->uuid_root &&
3644 	    (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3645 	     fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3646 		btrfs_info(fs_info, "checking UUID tree");
3647 		ret = btrfs_check_uuid_tree(fs_info);
3648 		if (ret) {
3649 			btrfs_warn(fs_info,
3650 				"failed to check the UUID tree: %d", ret);
3651 			close_ctree(fs_info);
3652 			return ret;
3653 		}
3654 	}
3655 
3656 	set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3657 
3658 clear_oneshot:
3659 	btrfs_clear_oneshot_options(fs_info);
3660 	return 0;
3661 
3662 fail_qgroup:
3663 	btrfs_free_qgroup_config(fs_info);
3664 fail_trans_kthread:
3665 	kthread_stop(fs_info->transaction_kthread);
3666 	btrfs_cleanup_transaction(fs_info);
3667 	btrfs_free_fs_roots(fs_info);
3668 fail_cleaner:
3669 	kthread_stop(fs_info->cleaner_kthread);
3670 
3671 	/*
3672 	 * make sure we're done with the btree inode before we stop our
3673 	 * kthreads
3674 	 */
3675 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3676 
3677 fail_sysfs:
3678 	btrfs_sysfs_remove_mounted(fs_info);
3679 
3680 fail_fsdev_sysfs:
3681 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3682 
3683 fail_block_groups:
3684 	btrfs_put_block_group_cache(fs_info);
3685 
3686 fail_tree_roots:
3687 	if (fs_info->data_reloc_root)
3688 		btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3689 	free_root_pointers(fs_info, true);
3690 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3691 
3692 fail_sb_buffer:
3693 	btrfs_stop_all_workers(fs_info);
3694 	btrfs_free_block_groups(fs_info);
3695 fail_alloc:
3696 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3697 
3698 	iput(fs_info->btree_inode);
3699 fail:
3700 	btrfs_close_devices(fs_info->fs_devices);
3701 	return err;
3702 }
3703 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3704 
3705 static void btrfs_end_super_write(struct bio *bio)
3706 {
3707 	struct btrfs_device *device = bio->bi_private;
3708 	struct bio_vec *bvec;
3709 	struct bvec_iter_all iter_all;
3710 	struct page *page;
3711 
3712 	bio_for_each_segment_all(bvec, bio, iter_all) {
3713 		page = bvec->bv_page;
3714 
3715 		if (bio->bi_status) {
3716 			btrfs_warn_rl_in_rcu(device->fs_info,
3717 				"lost page write due to IO error on %s (%d)",
3718 				rcu_str_deref(device->name),
3719 				blk_status_to_errno(bio->bi_status));
3720 			ClearPageUptodate(page);
3721 			SetPageError(page);
3722 			btrfs_dev_stat_inc_and_print(device,
3723 						     BTRFS_DEV_STAT_WRITE_ERRS);
3724 		} else {
3725 			SetPageUptodate(page);
3726 		}
3727 
3728 		put_page(page);
3729 		unlock_page(page);
3730 	}
3731 
3732 	bio_put(bio);
3733 }
3734 
3735 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3736 						   int copy_num)
3737 {
3738 	struct btrfs_super_block *super;
3739 	struct page *page;
3740 	u64 bytenr, bytenr_orig;
3741 	struct address_space *mapping = bdev->bd_inode->i_mapping;
3742 	int ret;
3743 
3744 	bytenr_orig = btrfs_sb_offset(copy_num);
3745 	ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3746 	if (ret == -ENOENT)
3747 		return ERR_PTR(-EINVAL);
3748 	else if (ret)
3749 		return ERR_PTR(ret);
3750 
3751 	if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3752 		return ERR_PTR(-EINVAL);
3753 
3754 	page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3755 	if (IS_ERR(page))
3756 		return ERR_CAST(page);
3757 
3758 	super = page_address(page);
3759 	if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3760 		btrfs_release_disk_super(super);
3761 		return ERR_PTR(-ENODATA);
3762 	}
3763 
3764 	if (btrfs_super_bytenr(super) != bytenr_orig) {
3765 		btrfs_release_disk_super(super);
3766 		return ERR_PTR(-EINVAL);
3767 	}
3768 
3769 	return super;
3770 }
3771 
3772 
3773 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3774 {
3775 	struct btrfs_super_block *super, *latest = NULL;
3776 	int i;
3777 	u64 transid = 0;
3778 
3779 	/* we would like to check all the supers, but that would make
3780 	 * a btrfs mount succeed after a mkfs from a different FS.
3781 	 * So, we need to add a special mount option to scan for
3782 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3783 	 */
3784 	for (i = 0; i < 1; i++) {
3785 		super = btrfs_read_dev_one_super(bdev, i);
3786 		if (IS_ERR(super))
3787 			continue;
3788 
3789 		if (!latest || btrfs_super_generation(super) > transid) {
3790 			if (latest)
3791 				btrfs_release_disk_super(super);
3792 
3793 			latest = super;
3794 			transid = btrfs_super_generation(super);
3795 		}
3796 	}
3797 
3798 	return super;
3799 }
3800 
3801 /*
3802  * Write superblock @sb to the @device. Do not wait for completion, all the
3803  * pages we use for writing are locked.
3804  *
3805  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3806  * the expected device size at commit time. Note that max_mirrors must be
3807  * same for write and wait phases.
3808  *
3809  * Return number of errors when page is not found or submission fails.
3810  */
3811 static int write_dev_supers(struct btrfs_device *device,
3812 			    struct btrfs_super_block *sb, int max_mirrors)
3813 {
3814 	struct btrfs_fs_info *fs_info = device->fs_info;
3815 	struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3816 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3817 	int i;
3818 	int errors = 0;
3819 	int ret;
3820 	u64 bytenr, bytenr_orig;
3821 
3822 	if (max_mirrors == 0)
3823 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3824 
3825 	shash->tfm = fs_info->csum_shash;
3826 
3827 	for (i = 0; i < max_mirrors; i++) {
3828 		struct page *page;
3829 		struct bio *bio;
3830 		struct btrfs_super_block *disk_super;
3831 
3832 		bytenr_orig = btrfs_sb_offset(i);
3833 		ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3834 		if (ret == -ENOENT) {
3835 			continue;
3836 		} else if (ret < 0) {
3837 			btrfs_err(device->fs_info,
3838 				"couldn't get super block location for mirror %d",
3839 				i);
3840 			errors++;
3841 			continue;
3842 		}
3843 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3844 		    device->commit_total_bytes)
3845 			break;
3846 
3847 		btrfs_set_super_bytenr(sb, bytenr_orig);
3848 
3849 		crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3850 				    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3851 				    sb->csum);
3852 
3853 		page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3854 					   GFP_NOFS);
3855 		if (!page) {
3856 			btrfs_err(device->fs_info,
3857 			    "couldn't get super block page for bytenr %llu",
3858 			    bytenr);
3859 			errors++;
3860 			continue;
3861 		}
3862 
3863 		/* Bump the refcount for wait_dev_supers() */
3864 		get_page(page);
3865 
3866 		disk_super = page_address(page);
3867 		memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3868 
3869 		/*
3870 		 * Directly use bios here instead of relying on the page cache
3871 		 * to do I/O, so we don't lose the ability to do integrity
3872 		 * checking.
3873 		 */
3874 		bio = bio_alloc(GFP_NOFS, 1);
3875 		bio_set_dev(bio, device->bdev);
3876 		bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3877 		bio->bi_private = device;
3878 		bio->bi_end_io = btrfs_end_super_write;
3879 		__bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3880 			       offset_in_page(bytenr));
3881 
3882 		/*
3883 		 * We FUA only the first super block.  The others we allow to
3884 		 * go down lazy and there's a short window where the on-disk
3885 		 * copies might still contain the older version.
3886 		 */
3887 		bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3888 		if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3889 			bio->bi_opf |= REQ_FUA;
3890 
3891 		btrfsic_submit_bio(bio);
3892 
3893 		if (btrfs_advance_sb_log(device, i))
3894 			errors++;
3895 	}
3896 	return errors < i ? 0 : -1;
3897 }
3898 
3899 /*
3900  * Wait for write completion of superblocks done by write_dev_supers,
3901  * @max_mirrors same for write and wait phases.
3902  *
3903  * Return number of errors when page is not found or not marked up to
3904  * date.
3905  */
3906 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3907 {
3908 	int i;
3909 	int errors = 0;
3910 	bool primary_failed = false;
3911 	int ret;
3912 	u64 bytenr;
3913 
3914 	if (max_mirrors == 0)
3915 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3916 
3917 	for (i = 0; i < max_mirrors; i++) {
3918 		struct page *page;
3919 
3920 		ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3921 		if (ret == -ENOENT) {
3922 			break;
3923 		} else if (ret < 0) {
3924 			errors++;
3925 			if (i == 0)
3926 				primary_failed = true;
3927 			continue;
3928 		}
3929 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3930 		    device->commit_total_bytes)
3931 			break;
3932 
3933 		page = find_get_page(device->bdev->bd_inode->i_mapping,
3934 				     bytenr >> PAGE_SHIFT);
3935 		if (!page) {
3936 			errors++;
3937 			if (i == 0)
3938 				primary_failed = true;
3939 			continue;
3940 		}
3941 		/* Page is submitted locked and unlocked once the IO completes */
3942 		wait_on_page_locked(page);
3943 		if (PageError(page)) {
3944 			errors++;
3945 			if (i == 0)
3946 				primary_failed = true;
3947 		}
3948 
3949 		/* Drop our reference */
3950 		put_page(page);
3951 
3952 		/* Drop the reference from the writing run */
3953 		put_page(page);
3954 	}
3955 
3956 	/* log error, force error return */
3957 	if (primary_failed) {
3958 		btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3959 			  device->devid);
3960 		return -1;
3961 	}
3962 
3963 	return errors < i ? 0 : -1;
3964 }
3965 
3966 /*
3967  * endio for the write_dev_flush, this will wake anyone waiting
3968  * for the barrier when it is done
3969  */
3970 static void btrfs_end_empty_barrier(struct bio *bio)
3971 {
3972 	complete(bio->bi_private);
3973 }
3974 
3975 /*
3976  * Submit a flush request to the device if it supports it. Error handling is
3977  * done in the waiting counterpart.
3978  */
3979 static void write_dev_flush(struct btrfs_device *device)
3980 {
3981 	struct request_queue *q = bdev_get_queue(device->bdev);
3982 	struct bio *bio = device->flush_bio;
3983 
3984 	if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3985 		return;
3986 
3987 	bio_reset(bio);
3988 	bio->bi_end_io = btrfs_end_empty_barrier;
3989 	bio_set_dev(bio, device->bdev);
3990 	bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3991 	init_completion(&device->flush_wait);
3992 	bio->bi_private = &device->flush_wait;
3993 
3994 	btrfsic_submit_bio(bio);
3995 	set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3996 }
3997 
3998 /*
3999  * If the flush bio has been submitted by write_dev_flush, wait for it.
4000  */
4001 static blk_status_t wait_dev_flush(struct btrfs_device *device)
4002 {
4003 	struct bio *bio = device->flush_bio;
4004 
4005 	if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
4006 		return BLK_STS_OK;
4007 
4008 	clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4009 	wait_for_completion_io(&device->flush_wait);
4010 
4011 	return bio->bi_status;
4012 }
4013 
4014 static int check_barrier_error(struct btrfs_fs_info *fs_info)
4015 {
4016 	if (!btrfs_check_rw_degradable(fs_info, NULL))
4017 		return -EIO;
4018 	return 0;
4019 }
4020 
4021 /*
4022  * send an empty flush down to each device in parallel,
4023  * then wait for them
4024  */
4025 static int barrier_all_devices(struct btrfs_fs_info *info)
4026 {
4027 	struct list_head *head;
4028 	struct btrfs_device *dev;
4029 	int errors_wait = 0;
4030 	blk_status_t ret;
4031 
4032 	lockdep_assert_held(&info->fs_devices->device_list_mutex);
4033 	/* send down all the barriers */
4034 	head = &info->fs_devices->devices;
4035 	list_for_each_entry(dev, head, dev_list) {
4036 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4037 			continue;
4038 		if (!dev->bdev)
4039 			continue;
4040 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4041 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4042 			continue;
4043 
4044 		write_dev_flush(dev);
4045 		dev->last_flush_error = BLK_STS_OK;
4046 	}
4047 
4048 	/* wait for all the barriers */
4049 	list_for_each_entry(dev, head, dev_list) {
4050 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4051 			continue;
4052 		if (!dev->bdev) {
4053 			errors_wait++;
4054 			continue;
4055 		}
4056 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4057 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4058 			continue;
4059 
4060 		ret = wait_dev_flush(dev);
4061 		if (ret) {
4062 			dev->last_flush_error = ret;
4063 			btrfs_dev_stat_inc_and_print(dev,
4064 					BTRFS_DEV_STAT_FLUSH_ERRS);
4065 			errors_wait++;
4066 		}
4067 	}
4068 
4069 	if (errors_wait) {
4070 		/*
4071 		 * At some point we need the status of all disks
4072 		 * to arrive at the volume status. So error checking
4073 		 * is being pushed to a separate loop.
4074 		 */
4075 		return check_barrier_error(info);
4076 	}
4077 	return 0;
4078 }
4079 
4080 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4081 {
4082 	int raid_type;
4083 	int min_tolerated = INT_MAX;
4084 
4085 	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4086 	    (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4087 		min_tolerated = min_t(int, min_tolerated,
4088 				    btrfs_raid_array[BTRFS_RAID_SINGLE].
4089 				    tolerated_failures);
4090 
4091 	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4092 		if (raid_type == BTRFS_RAID_SINGLE)
4093 			continue;
4094 		if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4095 			continue;
4096 		min_tolerated = min_t(int, min_tolerated,
4097 				    btrfs_raid_array[raid_type].
4098 				    tolerated_failures);
4099 	}
4100 
4101 	if (min_tolerated == INT_MAX) {
4102 		pr_warn("BTRFS: unknown raid flag: %llu", flags);
4103 		min_tolerated = 0;
4104 	}
4105 
4106 	return min_tolerated;
4107 }
4108 
4109 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4110 {
4111 	struct list_head *head;
4112 	struct btrfs_device *dev;
4113 	struct btrfs_super_block *sb;
4114 	struct btrfs_dev_item *dev_item;
4115 	int ret;
4116 	int do_barriers;
4117 	int max_errors;
4118 	int total_errors = 0;
4119 	u64 flags;
4120 
4121 	do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4122 
4123 	/*
4124 	 * max_mirrors == 0 indicates we're from commit_transaction,
4125 	 * not from fsync where the tree roots in fs_info have not
4126 	 * been consistent on disk.
4127 	 */
4128 	if (max_mirrors == 0)
4129 		backup_super_roots(fs_info);
4130 
4131 	sb = fs_info->super_for_commit;
4132 	dev_item = &sb->dev_item;
4133 
4134 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
4135 	head = &fs_info->fs_devices->devices;
4136 	max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4137 
4138 	if (do_barriers) {
4139 		ret = barrier_all_devices(fs_info);
4140 		if (ret) {
4141 			mutex_unlock(
4142 				&fs_info->fs_devices->device_list_mutex);
4143 			btrfs_handle_fs_error(fs_info, ret,
4144 					      "errors while submitting device barriers.");
4145 			return ret;
4146 		}
4147 	}
4148 
4149 	list_for_each_entry(dev, head, dev_list) {
4150 		if (!dev->bdev) {
4151 			total_errors++;
4152 			continue;
4153 		}
4154 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4155 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4156 			continue;
4157 
4158 		btrfs_set_stack_device_generation(dev_item, 0);
4159 		btrfs_set_stack_device_type(dev_item, dev->type);
4160 		btrfs_set_stack_device_id(dev_item, dev->devid);
4161 		btrfs_set_stack_device_total_bytes(dev_item,
4162 						   dev->commit_total_bytes);
4163 		btrfs_set_stack_device_bytes_used(dev_item,
4164 						  dev->commit_bytes_used);
4165 		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4166 		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4167 		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4168 		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4169 		memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4170 		       BTRFS_FSID_SIZE);
4171 
4172 		flags = btrfs_super_flags(sb);
4173 		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4174 
4175 		ret = btrfs_validate_write_super(fs_info, sb);
4176 		if (ret < 0) {
4177 			mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4178 			btrfs_handle_fs_error(fs_info, -EUCLEAN,
4179 				"unexpected superblock corruption detected");
4180 			return -EUCLEAN;
4181 		}
4182 
4183 		ret = write_dev_supers(dev, sb, max_mirrors);
4184 		if (ret)
4185 			total_errors++;
4186 	}
4187 	if (total_errors > max_errors) {
4188 		btrfs_err(fs_info, "%d errors while writing supers",
4189 			  total_errors);
4190 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4191 
4192 		/* FUA is masked off if unsupported and can't be the reason */
4193 		btrfs_handle_fs_error(fs_info, -EIO,
4194 				      "%d errors while writing supers",
4195 				      total_errors);
4196 		return -EIO;
4197 	}
4198 
4199 	total_errors = 0;
4200 	list_for_each_entry(dev, head, dev_list) {
4201 		if (!dev->bdev)
4202 			continue;
4203 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4204 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4205 			continue;
4206 
4207 		ret = wait_dev_supers(dev, max_mirrors);
4208 		if (ret)
4209 			total_errors++;
4210 	}
4211 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4212 	if (total_errors > max_errors) {
4213 		btrfs_handle_fs_error(fs_info, -EIO,
4214 				      "%d errors while writing supers",
4215 				      total_errors);
4216 		return -EIO;
4217 	}
4218 	return 0;
4219 }
4220 
4221 /* Drop a fs root from the radix tree and free it. */
4222 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4223 				  struct btrfs_root *root)
4224 {
4225 	bool drop_ref = false;
4226 
4227 	spin_lock(&fs_info->fs_roots_radix_lock);
4228 	radix_tree_delete(&fs_info->fs_roots_radix,
4229 			  (unsigned long)root->root_key.objectid);
4230 	if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4231 		drop_ref = true;
4232 	spin_unlock(&fs_info->fs_roots_radix_lock);
4233 
4234 	if (BTRFS_FS_ERROR(fs_info)) {
4235 		ASSERT(root->log_root == NULL);
4236 		if (root->reloc_root) {
4237 			btrfs_put_root(root->reloc_root);
4238 			root->reloc_root = NULL;
4239 		}
4240 	}
4241 
4242 	if (drop_ref)
4243 		btrfs_put_root(root);
4244 }
4245 
4246 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4247 {
4248 	u64 root_objectid = 0;
4249 	struct btrfs_root *gang[8];
4250 	int i = 0;
4251 	int err = 0;
4252 	unsigned int ret = 0;
4253 
4254 	while (1) {
4255 		spin_lock(&fs_info->fs_roots_radix_lock);
4256 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4257 					     (void **)gang, root_objectid,
4258 					     ARRAY_SIZE(gang));
4259 		if (!ret) {
4260 			spin_unlock(&fs_info->fs_roots_radix_lock);
4261 			break;
4262 		}
4263 		root_objectid = gang[ret - 1]->root_key.objectid + 1;
4264 
4265 		for (i = 0; i < ret; i++) {
4266 			/* Avoid to grab roots in dead_roots */
4267 			if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4268 				gang[i] = NULL;
4269 				continue;
4270 			}
4271 			/* grab all the search result for later use */
4272 			gang[i] = btrfs_grab_root(gang[i]);
4273 		}
4274 		spin_unlock(&fs_info->fs_roots_radix_lock);
4275 
4276 		for (i = 0; i < ret; i++) {
4277 			if (!gang[i])
4278 				continue;
4279 			root_objectid = gang[i]->root_key.objectid;
4280 			err = btrfs_orphan_cleanup(gang[i]);
4281 			if (err)
4282 				break;
4283 			btrfs_put_root(gang[i]);
4284 		}
4285 		root_objectid++;
4286 	}
4287 
4288 	/* release the uncleaned roots due to error */
4289 	for (; i < ret; i++) {
4290 		if (gang[i])
4291 			btrfs_put_root(gang[i]);
4292 	}
4293 	return err;
4294 }
4295 
4296 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4297 {
4298 	struct btrfs_root *root = fs_info->tree_root;
4299 	struct btrfs_trans_handle *trans;
4300 
4301 	mutex_lock(&fs_info->cleaner_mutex);
4302 	btrfs_run_delayed_iputs(fs_info);
4303 	mutex_unlock(&fs_info->cleaner_mutex);
4304 	wake_up_process(fs_info->cleaner_kthread);
4305 
4306 	/* wait until ongoing cleanup work done */
4307 	down_write(&fs_info->cleanup_work_sem);
4308 	up_write(&fs_info->cleanup_work_sem);
4309 
4310 	trans = btrfs_join_transaction(root);
4311 	if (IS_ERR(trans))
4312 		return PTR_ERR(trans);
4313 	return btrfs_commit_transaction(trans);
4314 }
4315 
4316 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4317 {
4318 	int ret;
4319 
4320 	set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4321 	/*
4322 	 * We don't want the cleaner to start new transactions, add more delayed
4323 	 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4324 	 * because that frees the task_struct, and the transaction kthread might
4325 	 * still try to wake up the cleaner.
4326 	 */
4327 	kthread_park(fs_info->cleaner_kthread);
4328 
4329 	/* wait for the qgroup rescan worker to stop */
4330 	btrfs_qgroup_wait_for_completion(fs_info, false);
4331 
4332 	/* wait for the uuid_scan task to finish */
4333 	down(&fs_info->uuid_tree_rescan_sem);
4334 	/* avoid complains from lockdep et al., set sem back to initial state */
4335 	up(&fs_info->uuid_tree_rescan_sem);
4336 
4337 	/* pause restriper - we want to resume on mount */
4338 	btrfs_pause_balance(fs_info);
4339 
4340 	btrfs_dev_replace_suspend_for_unmount(fs_info);
4341 
4342 	btrfs_scrub_cancel(fs_info);
4343 
4344 	/* wait for any defraggers to finish */
4345 	wait_event(fs_info->transaction_wait,
4346 		   (atomic_read(&fs_info->defrag_running) == 0));
4347 
4348 	/* clear out the rbtree of defraggable inodes */
4349 	btrfs_cleanup_defrag_inodes(fs_info);
4350 
4351 	cancel_work_sync(&fs_info->async_reclaim_work);
4352 	cancel_work_sync(&fs_info->async_data_reclaim_work);
4353 	cancel_work_sync(&fs_info->preempt_reclaim_work);
4354 
4355 	cancel_work_sync(&fs_info->reclaim_bgs_work);
4356 
4357 	/* Cancel or finish ongoing discard work */
4358 	btrfs_discard_cleanup(fs_info);
4359 
4360 	if (!sb_rdonly(fs_info->sb)) {
4361 		/*
4362 		 * The cleaner kthread is stopped, so do one final pass over
4363 		 * unused block groups.
4364 		 */
4365 		btrfs_delete_unused_bgs(fs_info);
4366 
4367 		/*
4368 		 * There might be existing delayed inode workers still running
4369 		 * and holding an empty delayed inode item. We must wait for
4370 		 * them to complete first because they can create a transaction.
4371 		 * This happens when someone calls btrfs_balance_delayed_items()
4372 		 * and then a transaction commit runs the same delayed nodes
4373 		 * before any delayed worker has done something with the nodes.
4374 		 * We must wait for any worker here and not at transaction
4375 		 * commit time since that could cause a deadlock.
4376 		 * This is a very rare case.
4377 		 */
4378 		btrfs_flush_workqueue(fs_info->delayed_workers);
4379 
4380 		ret = btrfs_commit_super(fs_info);
4381 		if (ret)
4382 			btrfs_err(fs_info, "commit super ret %d", ret);
4383 	}
4384 
4385 	if (BTRFS_FS_ERROR(fs_info))
4386 		btrfs_error_commit_super(fs_info);
4387 
4388 	kthread_stop(fs_info->transaction_kthread);
4389 	kthread_stop(fs_info->cleaner_kthread);
4390 
4391 	ASSERT(list_empty(&fs_info->delayed_iputs));
4392 	set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4393 
4394 	if (btrfs_check_quota_leak(fs_info)) {
4395 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4396 		btrfs_err(fs_info, "qgroup reserved space leaked");
4397 	}
4398 
4399 	btrfs_free_qgroup_config(fs_info);
4400 	ASSERT(list_empty(&fs_info->delalloc_roots));
4401 
4402 	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4403 		btrfs_info(fs_info, "at unmount delalloc count %lld",
4404 		       percpu_counter_sum(&fs_info->delalloc_bytes));
4405 	}
4406 
4407 	if (percpu_counter_sum(&fs_info->ordered_bytes))
4408 		btrfs_info(fs_info, "at unmount dio bytes count %lld",
4409 			   percpu_counter_sum(&fs_info->ordered_bytes));
4410 
4411 	btrfs_sysfs_remove_mounted(fs_info);
4412 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4413 
4414 	btrfs_put_block_group_cache(fs_info);
4415 
4416 	/*
4417 	 * we must make sure there is not any read request to
4418 	 * submit after we stopping all workers.
4419 	 */
4420 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4421 	btrfs_stop_all_workers(fs_info);
4422 
4423 	/* We shouldn't have any transaction open at this point */
4424 	ASSERT(list_empty(&fs_info->trans_list));
4425 
4426 	clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4427 	free_root_pointers(fs_info, true);
4428 	btrfs_free_fs_roots(fs_info);
4429 
4430 	/*
4431 	 * We must free the block groups after dropping the fs_roots as we could
4432 	 * have had an IO error and have left over tree log blocks that aren't
4433 	 * cleaned up until the fs roots are freed.  This makes the block group
4434 	 * accounting appear to be wrong because there's pending reserved bytes,
4435 	 * so make sure we do the block group cleanup afterwards.
4436 	 */
4437 	btrfs_free_block_groups(fs_info);
4438 
4439 	iput(fs_info->btree_inode);
4440 
4441 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4442 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4443 		btrfsic_unmount(fs_info->fs_devices);
4444 #endif
4445 
4446 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
4447 	btrfs_close_devices(fs_info->fs_devices);
4448 }
4449 
4450 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4451 			  int atomic)
4452 {
4453 	int ret;
4454 	struct inode *btree_inode = buf->pages[0]->mapping->host;
4455 
4456 	ret = extent_buffer_uptodate(buf);
4457 	if (!ret)
4458 		return ret;
4459 
4460 	ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4461 				    parent_transid, atomic);
4462 	if (ret == -EAGAIN)
4463 		return ret;
4464 	return !ret;
4465 }
4466 
4467 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4468 {
4469 	struct btrfs_fs_info *fs_info = buf->fs_info;
4470 	u64 transid = btrfs_header_generation(buf);
4471 	int was_dirty;
4472 
4473 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4474 	/*
4475 	 * This is a fast path so only do this check if we have sanity tests
4476 	 * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4477 	 * outside of the sanity tests.
4478 	 */
4479 	if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4480 		return;
4481 #endif
4482 	btrfs_assert_tree_write_locked(buf);
4483 	if (transid != fs_info->generation)
4484 		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4485 			buf->start, transid, fs_info->generation);
4486 	was_dirty = set_extent_buffer_dirty(buf);
4487 	if (!was_dirty)
4488 		percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4489 					 buf->len,
4490 					 fs_info->dirty_metadata_batch);
4491 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4492 	/*
4493 	 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4494 	 * but item data not updated.
4495 	 * So here we should only check item pointers, not item data.
4496 	 */
4497 	if (btrfs_header_level(buf) == 0 &&
4498 	    btrfs_check_leaf_relaxed(buf)) {
4499 		btrfs_print_leaf(buf);
4500 		ASSERT(0);
4501 	}
4502 #endif
4503 }
4504 
4505 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4506 					int flush_delayed)
4507 {
4508 	/*
4509 	 * looks as though older kernels can get into trouble with
4510 	 * this code, they end up stuck in balance_dirty_pages forever
4511 	 */
4512 	int ret;
4513 
4514 	if (current->flags & PF_MEMALLOC)
4515 		return;
4516 
4517 	if (flush_delayed)
4518 		btrfs_balance_delayed_items(fs_info);
4519 
4520 	ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4521 				     BTRFS_DIRTY_METADATA_THRESH,
4522 				     fs_info->dirty_metadata_batch);
4523 	if (ret > 0) {
4524 		balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4525 	}
4526 }
4527 
4528 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4529 {
4530 	__btrfs_btree_balance_dirty(fs_info, 1);
4531 }
4532 
4533 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4534 {
4535 	__btrfs_btree_balance_dirty(fs_info, 0);
4536 }
4537 
4538 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4539 		      struct btrfs_key *first_key)
4540 {
4541 	return btree_read_extent_buffer_pages(buf, parent_transid,
4542 					      level, first_key);
4543 }
4544 
4545 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4546 {
4547 	/* cleanup FS via transaction */
4548 	btrfs_cleanup_transaction(fs_info);
4549 
4550 	mutex_lock(&fs_info->cleaner_mutex);
4551 	btrfs_run_delayed_iputs(fs_info);
4552 	mutex_unlock(&fs_info->cleaner_mutex);
4553 
4554 	down_write(&fs_info->cleanup_work_sem);
4555 	up_write(&fs_info->cleanup_work_sem);
4556 }
4557 
4558 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4559 {
4560 	struct btrfs_root *gang[8];
4561 	u64 root_objectid = 0;
4562 	int ret;
4563 
4564 	spin_lock(&fs_info->fs_roots_radix_lock);
4565 	while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4566 					     (void **)gang, root_objectid,
4567 					     ARRAY_SIZE(gang))) != 0) {
4568 		int i;
4569 
4570 		for (i = 0; i < ret; i++)
4571 			gang[i] = btrfs_grab_root(gang[i]);
4572 		spin_unlock(&fs_info->fs_roots_radix_lock);
4573 
4574 		for (i = 0; i < ret; i++) {
4575 			if (!gang[i])
4576 				continue;
4577 			root_objectid = gang[i]->root_key.objectid;
4578 			btrfs_free_log(NULL, gang[i]);
4579 			btrfs_put_root(gang[i]);
4580 		}
4581 		root_objectid++;
4582 		spin_lock(&fs_info->fs_roots_radix_lock);
4583 	}
4584 	spin_unlock(&fs_info->fs_roots_radix_lock);
4585 	btrfs_free_log_root_tree(NULL, fs_info);
4586 }
4587 
4588 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4589 {
4590 	struct btrfs_ordered_extent *ordered;
4591 
4592 	spin_lock(&root->ordered_extent_lock);
4593 	/*
4594 	 * This will just short circuit the ordered completion stuff which will
4595 	 * make sure the ordered extent gets properly cleaned up.
4596 	 */
4597 	list_for_each_entry(ordered, &root->ordered_extents,
4598 			    root_extent_list)
4599 		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4600 	spin_unlock(&root->ordered_extent_lock);
4601 }
4602 
4603 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4604 {
4605 	struct btrfs_root *root;
4606 	struct list_head splice;
4607 
4608 	INIT_LIST_HEAD(&splice);
4609 
4610 	spin_lock(&fs_info->ordered_root_lock);
4611 	list_splice_init(&fs_info->ordered_roots, &splice);
4612 	while (!list_empty(&splice)) {
4613 		root = list_first_entry(&splice, struct btrfs_root,
4614 					ordered_root);
4615 		list_move_tail(&root->ordered_root,
4616 			       &fs_info->ordered_roots);
4617 
4618 		spin_unlock(&fs_info->ordered_root_lock);
4619 		btrfs_destroy_ordered_extents(root);
4620 
4621 		cond_resched();
4622 		spin_lock(&fs_info->ordered_root_lock);
4623 	}
4624 	spin_unlock(&fs_info->ordered_root_lock);
4625 
4626 	/*
4627 	 * We need this here because if we've been flipped read-only we won't
4628 	 * get sync() from the umount, so we need to make sure any ordered
4629 	 * extents that haven't had their dirty pages IO start writeout yet
4630 	 * actually get run and error out properly.
4631 	 */
4632 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4633 }
4634 
4635 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4636 				      struct btrfs_fs_info *fs_info)
4637 {
4638 	struct rb_node *node;
4639 	struct btrfs_delayed_ref_root *delayed_refs;
4640 	struct btrfs_delayed_ref_node *ref;
4641 	int ret = 0;
4642 
4643 	delayed_refs = &trans->delayed_refs;
4644 
4645 	spin_lock(&delayed_refs->lock);
4646 	if (atomic_read(&delayed_refs->num_entries) == 0) {
4647 		spin_unlock(&delayed_refs->lock);
4648 		btrfs_debug(fs_info, "delayed_refs has NO entry");
4649 		return ret;
4650 	}
4651 
4652 	while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4653 		struct btrfs_delayed_ref_head *head;
4654 		struct rb_node *n;
4655 		bool pin_bytes = false;
4656 
4657 		head = rb_entry(node, struct btrfs_delayed_ref_head,
4658 				href_node);
4659 		if (btrfs_delayed_ref_lock(delayed_refs, head))
4660 			continue;
4661 
4662 		spin_lock(&head->lock);
4663 		while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4664 			ref = rb_entry(n, struct btrfs_delayed_ref_node,
4665 				       ref_node);
4666 			ref->in_tree = 0;
4667 			rb_erase_cached(&ref->ref_node, &head->ref_tree);
4668 			RB_CLEAR_NODE(&ref->ref_node);
4669 			if (!list_empty(&ref->add_list))
4670 				list_del(&ref->add_list);
4671 			atomic_dec(&delayed_refs->num_entries);
4672 			btrfs_put_delayed_ref(ref);
4673 		}
4674 		if (head->must_insert_reserved)
4675 			pin_bytes = true;
4676 		btrfs_free_delayed_extent_op(head->extent_op);
4677 		btrfs_delete_ref_head(delayed_refs, head);
4678 		spin_unlock(&head->lock);
4679 		spin_unlock(&delayed_refs->lock);
4680 		mutex_unlock(&head->mutex);
4681 
4682 		if (pin_bytes) {
4683 			struct btrfs_block_group *cache;
4684 
4685 			cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4686 			BUG_ON(!cache);
4687 
4688 			spin_lock(&cache->space_info->lock);
4689 			spin_lock(&cache->lock);
4690 			cache->pinned += head->num_bytes;
4691 			btrfs_space_info_update_bytes_pinned(fs_info,
4692 				cache->space_info, head->num_bytes);
4693 			cache->reserved -= head->num_bytes;
4694 			cache->space_info->bytes_reserved -= head->num_bytes;
4695 			spin_unlock(&cache->lock);
4696 			spin_unlock(&cache->space_info->lock);
4697 
4698 			btrfs_put_block_group(cache);
4699 
4700 			btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4701 				head->bytenr + head->num_bytes - 1);
4702 		}
4703 		btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4704 		btrfs_put_delayed_ref_head(head);
4705 		cond_resched();
4706 		spin_lock(&delayed_refs->lock);
4707 	}
4708 	btrfs_qgroup_destroy_extent_records(trans);
4709 
4710 	spin_unlock(&delayed_refs->lock);
4711 
4712 	return ret;
4713 }
4714 
4715 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4716 {
4717 	struct btrfs_inode *btrfs_inode;
4718 	struct list_head splice;
4719 
4720 	INIT_LIST_HEAD(&splice);
4721 
4722 	spin_lock(&root->delalloc_lock);
4723 	list_splice_init(&root->delalloc_inodes, &splice);
4724 
4725 	while (!list_empty(&splice)) {
4726 		struct inode *inode = NULL;
4727 		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4728 					       delalloc_inodes);
4729 		__btrfs_del_delalloc_inode(root, btrfs_inode);
4730 		spin_unlock(&root->delalloc_lock);
4731 
4732 		/*
4733 		 * Make sure we get a live inode and that it'll not disappear
4734 		 * meanwhile.
4735 		 */
4736 		inode = igrab(&btrfs_inode->vfs_inode);
4737 		if (inode) {
4738 			invalidate_inode_pages2(inode->i_mapping);
4739 			iput(inode);
4740 		}
4741 		spin_lock(&root->delalloc_lock);
4742 	}
4743 	spin_unlock(&root->delalloc_lock);
4744 }
4745 
4746 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4747 {
4748 	struct btrfs_root *root;
4749 	struct list_head splice;
4750 
4751 	INIT_LIST_HEAD(&splice);
4752 
4753 	spin_lock(&fs_info->delalloc_root_lock);
4754 	list_splice_init(&fs_info->delalloc_roots, &splice);
4755 	while (!list_empty(&splice)) {
4756 		root = list_first_entry(&splice, struct btrfs_root,
4757 					 delalloc_root);
4758 		root = btrfs_grab_root(root);
4759 		BUG_ON(!root);
4760 		spin_unlock(&fs_info->delalloc_root_lock);
4761 
4762 		btrfs_destroy_delalloc_inodes(root);
4763 		btrfs_put_root(root);
4764 
4765 		spin_lock(&fs_info->delalloc_root_lock);
4766 	}
4767 	spin_unlock(&fs_info->delalloc_root_lock);
4768 }
4769 
4770 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4771 					struct extent_io_tree *dirty_pages,
4772 					int mark)
4773 {
4774 	int ret;
4775 	struct extent_buffer *eb;
4776 	u64 start = 0;
4777 	u64 end;
4778 
4779 	while (1) {
4780 		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4781 					    mark, NULL);
4782 		if (ret)
4783 			break;
4784 
4785 		clear_extent_bits(dirty_pages, start, end, mark);
4786 		while (start <= end) {
4787 			eb = find_extent_buffer(fs_info, start);
4788 			start += fs_info->nodesize;
4789 			if (!eb)
4790 				continue;
4791 			wait_on_extent_buffer_writeback(eb);
4792 
4793 			if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4794 					       &eb->bflags))
4795 				clear_extent_buffer_dirty(eb);
4796 			free_extent_buffer_stale(eb);
4797 		}
4798 	}
4799 
4800 	return ret;
4801 }
4802 
4803 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4804 				       struct extent_io_tree *unpin)
4805 {
4806 	u64 start;
4807 	u64 end;
4808 	int ret;
4809 
4810 	while (1) {
4811 		struct extent_state *cached_state = NULL;
4812 
4813 		/*
4814 		 * The btrfs_finish_extent_commit() may get the same range as
4815 		 * ours between find_first_extent_bit and clear_extent_dirty.
4816 		 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4817 		 * the same extent range.
4818 		 */
4819 		mutex_lock(&fs_info->unused_bg_unpin_mutex);
4820 		ret = find_first_extent_bit(unpin, 0, &start, &end,
4821 					    EXTENT_DIRTY, &cached_state);
4822 		if (ret) {
4823 			mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4824 			break;
4825 		}
4826 
4827 		clear_extent_dirty(unpin, start, end, &cached_state);
4828 		free_extent_state(cached_state);
4829 		btrfs_error_unpin_extent_range(fs_info, start, end);
4830 		mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4831 		cond_resched();
4832 	}
4833 
4834 	return 0;
4835 }
4836 
4837 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4838 {
4839 	struct inode *inode;
4840 
4841 	inode = cache->io_ctl.inode;
4842 	if (inode) {
4843 		invalidate_inode_pages2(inode->i_mapping);
4844 		BTRFS_I(inode)->generation = 0;
4845 		cache->io_ctl.inode = NULL;
4846 		iput(inode);
4847 	}
4848 	ASSERT(cache->io_ctl.pages == NULL);
4849 	btrfs_put_block_group(cache);
4850 }
4851 
4852 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4853 			     struct btrfs_fs_info *fs_info)
4854 {
4855 	struct btrfs_block_group *cache;
4856 
4857 	spin_lock(&cur_trans->dirty_bgs_lock);
4858 	while (!list_empty(&cur_trans->dirty_bgs)) {
4859 		cache = list_first_entry(&cur_trans->dirty_bgs,
4860 					 struct btrfs_block_group,
4861 					 dirty_list);
4862 
4863 		if (!list_empty(&cache->io_list)) {
4864 			spin_unlock(&cur_trans->dirty_bgs_lock);
4865 			list_del_init(&cache->io_list);
4866 			btrfs_cleanup_bg_io(cache);
4867 			spin_lock(&cur_trans->dirty_bgs_lock);
4868 		}
4869 
4870 		list_del_init(&cache->dirty_list);
4871 		spin_lock(&cache->lock);
4872 		cache->disk_cache_state = BTRFS_DC_ERROR;
4873 		spin_unlock(&cache->lock);
4874 
4875 		spin_unlock(&cur_trans->dirty_bgs_lock);
4876 		btrfs_put_block_group(cache);
4877 		btrfs_delayed_refs_rsv_release(fs_info, 1);
4878 		spin_lock(&cur_trans->dirty_bgs_lock);
4879 	}
4880 	spin_unlock(&cur_trans->dirty_bgs_lock);
4881 
4882 	/*
4883 	 * Refer to the definition of io_bgs member for details why it's safe
4884 	 * to use it without any locking
4885 	 */
4886 	while (!list_empty(&cur_trans->io_bgs)) {
4887 		cache = list_first_entry(&cur_trans->io_bgs,
4888 					 struct btrfs_block_group,
4889 					 io_list);
4890 
4891 		list_del_init(&cache->io_list);
4892 		spin_lock(&cache->lock);
4893 		cache->disk_cache_state = BTRFS_DC_ERROR;
4894 		spin_unlock(&cache->lock);
4895 		btrfs_cleanup_bg_io(cache);
4896 	}
4897 }
4898 
4899 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4900 				   struct btrfs_fs_info *fs_info)
4901 {
4902 	struct btrfs_device *dev, *tmp;
4903 
4904 	btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4905 	ASSERT(list_empty(&cur_trans->dirty_bgs));
4906 	ASSERT(list_empty(&cur_trans->io_bgs));
4907 
4908 	list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4909 				 post_commit_list) {
4910 		list_del_init(&dev->post_commit_list);
4911 	}
4912 
4913 	btrfs_destroy_delayed_refs(cur_trans, fs_info);
4914 
4915 	cur_trans->state = TRANS_STATE_COMMIT_START;
4916 	wake_up(&fs_info->transaction_blocked_wait);
4917 
4918 	cur_trans->state = TRANS_STATE_UNBLOCKED;
4919 	wake_up(&fs_info->transaction_wait);
4920 
4921 	btrfs_destroy_delayed_inodes(fs_info);
4922 
4923 	btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4924 				     EXTENT_DIRTY);
4925 	btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4926 
4927 	btrfs_free_redirty_list(cur_trans);
4928 
4929 	cur_trans->state =TRANS_STATE_COMPLETED;
4930 	wake_up(&cur_trans->commit_wait);
4931 }
4932 
4933 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4934 {
4935 	struct btrfs_transaction *t;
4936 
4937 	mutex_lock(&fs_info->transaction_kthread_mutex);
4938 
4939 	spin_lock(&fs_info->trans_lock);
4940 	while (!list_empty(&fs_info->trans_list)) {
4941 		t = list_first_entry(&fs_info->trans_list,
4942 				     struct btrfs_transaction, list);
4943 		if (t->state >= TRANS_STATE_COMMIT_START) {
4944 			refcount_inc(&t->use_count);
4945 			spin_unlock(&fs_info->trans_lock);
4946 			btrfs_wait_for_commit(fs_info, t->transid);
4947 			btrfs_put_transaction(t);
4948 			spin_lock(&fs_info->trans_lock);
4949 			continue;
4950 		}
4951 		if (t == fs_info->running_transaction) {
4952 			t->state = TRANS_STATE_COMMIT_DOING;
4953 			spin_unlock(&fs_info->trans_lock);
4954 			/*
4955 			 * We wait for 0 num_writers since we don't hold a trans
4956 			 * handle open currently for this transaction.
4957 			 */
4958 			wait_event(t->writer_wait,
4959 				   atomic_read(&t->num_writers) == 0);
4960 		} else {
4961 			spin_unlock(&fs_info->trans_lock);
4962 		}
4963 		btrfs_cleanup_one_transaction(t, fs_info);
4964 
4965 		spin_lock(&fs_info->trans_lock);
4966 		if (t == fs_info->running_transaction)
4967 			fs_info->running_transaction = NULL;
4968 		list_del_init(&t->list);
4969 		spin_unlock(&fs_info->trans_lock);
4970 
4971 		btrfs_put_transaction(t);
4972 		trace_btrfs_transaction_commit(fs_info->tree_root);
4973 		spin_lock(&fs_info->trans_lock);
4974 	}
4975 	spin_unlock(&fs_info->trans_lock);
4976 	btrfs_destroy_all_ordered_extents(fs_info);
4977 	btrfs_destroy_delayed_inodes(fs_info);
4978 	btrfs_assert_delayed_root_empty(fs_info);
4979 	btrfs_destroy_all_delalloc_inodes(fs_info);
4980 	btrfs_drop_all_logs(fs_info);
4981 	mutex_unlock(&fs_info->transaction_kthread_mutex);
4982 
4983 	return 0;
4984 }
4985 
4986 int btrfs_init_root_free_objectid(struct btrfs_root *root)
4987 {
4988 	struct btrfs_path *path;
4989 	int ret;
4990 	struct extent_buffer *l;
4991 	struct btrfs_key search_key;
4992 	struct btrfs_key found_key;
4993 	int slot;
4994 
4995 	path = btrfs_alloc_path();
4996 	if (!path)
4997 		return -ENOMEM;
4998 
4999 	search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
5000 	search_key.type = -1;
5001 	search_key.offset = (u64)-1;
5002 	ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5003 	if (ret < 0)
5004 		goto error;
5005 	BUG_ON(ret == 0); /* Corruption */
5006 	if (path->slots[0] > 0) {
5007 		slot = path->slots[0] - 1;
5008 		l = path->nodes[0];
5009 		btrfs_item_key_to_cpu(l, &found_key, slot);
5010 		root->free_objectid = max_t(u64, found_key.objectid + 1,
5011 					    BTRFS_FIRST_FREE_OBJECTID);
5012 	} else {
5013 		root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5014 	}
5015 	ret = 0;
5016 error:
5017 	btrfs_free_path(path);
5018 	return ret;
5019 }
5020 
5021 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5022 {
5023 	int ret;
5024 	mutex_lock(&root->objectid_mutex);
5025 
5026 	if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5027 		btrfs_warn(root->fs_info,
5028 			   "the objectid of root %llu reaches its highest value",
5029 			   root->root_key.objectid);
5030 		ret = -ENOSPC;
5031 		goto out;
5032 	}
5033 
5034 	*objectid = root->free_objectid++;
5035 	ret = 0;
5036 out:
5037 	mutex_unlock(&root->objectid_mutex);
5038 	return ret;
5039 }
5040