xref: /linux/fs/btrfs/disk-io.c (revision f8e17c17b81070f38062dce79ca7f4541851dadd)
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/buffer_head.h>
11 #include <linux/workqueue.h>
12 #include <linux/kthread.h>
13 #include <linux/slab.h>
14 #include <linux/migrate.h>
15 #include <linux/ratelimit.h>
16 #include <linux/uuid.h>
17 #include <linux/semaphore.h>
18 #include <linux/error-injection.h>
19 #include <linux/crc32c.h>
20 #include <linux/sched/mm.h>
21 #include <asm/unaligned.h>
22 #include <crypto/hash.h>
23 #include "ctree.h"
24 #include "disk-io.h"
25 #include "transaction.h"
26 #include "btrfs_inode.h"
27 #include "volumes.h"
28 #include "print-tree.h"
29 #include "locking.h"
30 #include "tree-log.h"
31 #include "free-space-cache.h"
32 #include "free-space-tree.h"
33 #include "inode-map.h"
34 #include "check-integrity.h"
35 #include "rcu-string.h"
36 #include "dev-replace.h"
37 #include "raid56.h"
38 #include "sysfs.h"
39 #include "qgroup.h"
40 #include "compression.h"
41 #include "tree-checker.h"
42 #include "ref-verify.h"
43 #include "block-group.h"
44 #include "discard.h"
45 
46 #define BTRFS_SUPER_FLAG_SUPP	(BTRFS_HEADER_FLAG_WRITTEN |\
47 				 BTRFS_HEADER_FLAG_RELOC |\
48 				 BTRFS_SUPER_FLAG_ERROR |\
49 				 BTRFS_SUPER_FLAG_SEEDING |\
50 				 BTRFS_SUPER_FLAG_METADUMP |\
51 				 BTRFS_SUPER_FLAG_METADUMP_V2)
52 
53 static const struct extent_io_ops btree_extent_io_ops;
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 /*
102  * async submit bios are used to offload expensive checksumming
103  * onto the worker threads.  They checksum file and metadata bios
104  * just before they are sent down the IO stack.
105  */
106 struct async_submit_bio {
107 	void *private_data;
108 	struct bio *bio;
109 	extent_submit_bio_start_t *submit_bio_start;
110 	int mirror_num;
111 	/*
112 	 * bio_offset is optional, can be used if the pages in the bio
113 	 * can't tell us where in the file the bio should go
114 	 */
115 	u64 bio_offset;
116 	struct btrfs_work work;
117 	blk_status_t status;
118 };
119 
120 /*
121  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
122  * eb, the lockdep key is determined by the btrfs_root it belongs to and
123  * the level the eb occupies in the tree.
124  *
125  * Different roots are used for different purposes and may nest inside each
126  * other and they require separate keysets.  As lockdep keys should be
127  * static, assign keysets according to the purpose of the root as indicated
128  * by btrfs_root->root_key.objectid.  This ensures that all special purpose
129  * roots have separate keysets.
130  *
131  * Lock-nesting across peer nodes is always done with the immediate parent
132  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
133  * subclass to avoid triggering lockdep warning in such cases.
134  *
135  * The key is set by the readpage_end_io_hook after the buffer has passed
136  * csum validation but before the pages are unlocked.  It is also set by
137  * btrfs_init_new_buffer on freshly allocated blocks.
138  *
139  * We also add a check to make sure the highest level of the tree is the
140  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
141  * needs update as well.
142  */
143 #ifdef CONFIG_DEBUG_LOCK_ALLOC
144 # if BTRFS_MAX_LEVEL != 8
145 #  error
146 # endif
147 
148 static struct btrfs_lockdep_keyset {
149 	u64			id;		/* root objectid */
150 	const char		*name_stem;	/* lock name stem */
151 	char			names[BTRFS_MAX_LEVEL + 1][20];
152 	struct lock_class_key	keys[BTRFS_MAX_LEVEL + 1];
153 } btrfs_lockdep_keysets[] = {
154 	{ .id = BTRFS_ROOT_TREE_OBJECTID,	.name_stem = "root"	},
155 	{ .id = BTRFS_EXTENT_TREE_OBJECTID,	.name_stem = "extent"	},
156 	{ .id = BTRFS_CHUNK_TREE_OBJECTID,	.name_stem = "chunk"	},
157 	{ .id = BTRFS_DEV_TREE_OBJECTID,	.name_stem = "dev"	},
158 	{ .id = BTRFS_FS_TREE_OBJECTID,		.name_stem = "fs"	},
159 	{ .id = BTRFS_CSUM_TREE_OBJECTID,	.name_stem = "csum"	},
160 	{ .id = BTRFS_QUOTA_TREE_OBJECTID,	.name_stem = "quota"	},
161 	{ .id = BTRFS_TREE_LOG_OBJECTID,	.name_stem = "log"	},
162 	{ .id = BTRFS_TREE_RELOC_OBJECTID,	.name_stem = "treloc"	},
163 	{ .id = BTRFS_DATA_RELOC_TREE_OBJECTID,	.name_stem = "dreloc"	},
164 	{ .id = BTRFS_UUID_TREE_OBJECTID,	.name_stem = "uuid"	},
165 	{ .id = BTRFS_FREE_SPACE_TREE_OBJECTID,	.name_stem = "free-space" },
166 	{ .id = 0,				.name_stem = "tree"	},
167 };
168 
169 void __init btrfs_init_lockdep(void)
170 {
171 	int i, j;
172 
173 	/* initialize lockdep class names */
174 	for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
175 		struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
176 
177 		for (j = 0; j < ARRAY_SIZE(ks->names); j++)
178 			snprintf(ks->names[j], sizeof(ks->names[j]),
179 				 "btrfs-%s-%02d", ks->name_stem, j);
180 	}
181 }
182 
183 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
184 				    int level)
185 {
186 	struct btrfs_lockdep_keyset *ks;
187 
188 	BUG_ON(level >= ARRAY_SIZE(ks->keys));
189 
190 	/* find the matching keyset, id 0 is the default entry */
191 	for (ks = btrfs_lockdep_keysets; ks->id; ks++)
192 		if (ks->id == objectid)
193 			break;
194 
195 	lockdep_set_class_and_name(&eb->lock,
196 				   &ks->keys[level], ks->names[level]);
197 }
198 
199 #endif
200 
201 /*
202  * extents on the btree inode are pretty simple, there's one extent
203  * that covers the entire device
204  */
205 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
206 				    struct page *page, size_t pg_offset,
207 				    u64 start, u64 len)
208 {
209 	struct extent_map_tree *em_tree = &inode->extent_tree;
210 	struct extent_map *em;
211 	int ret;
212 
213 	read_lock(&em_tree->lock);
214 	em = lookup_extent_mapping(em_tree, start, len);
215 	if (em) {
216 		read_unlock(&em_tree->lock);
217 		goto out;
218 	}
219 	read_unlock(&em_tree->lock);
220 
221 	em = alloc_extent_map();
222 	if (!em) {
223 		em = ERR_PTR(-ENOMEM);
224 		goto out;
225 	}
226 	em->start = 0;
227 	em->len = (u64)-1;
228 	em->block_len = (u64)-1;
229 	em->block_start = 0;
230 
231 	write_lock(&em_tree->lock);
232 	ret = add_extent_mapping(em_tree, em, 0);
233 	if (ret == -EEXIST) {
234 		free_extent_map(em);
235 		em = lookup_extent_mapping(em_tree, start, len);
236 		if (!em)
237 			em = ERR_PTR(-EIO);
238 	} else if (ret) {
239 		free_extent_map(em);
240 		em = ERR_PTR(ret);
241 	}
242 	write_unlock(&em_tree->lock);
243 
244 out:
245 	return em;
246 }
247 
248 /*
249  * Compute the csum of a btree block and store the result to provided buffer.
250  *
251  * Returns error if the extent buffer cannot be mapped.
252  */
253 static int csum_tree_block(struct extent_buffer *buf, u8 *result)
254 {
255 	struct btrfs_fs_info *fs_info = buf->fs_info;
256 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
257 	unsigned long len;
258 	unsigned long cur_len;
259 	unsigned long offset = BTRFS_CSUM_SIZE;
260 	char *kaddr;
261 	unsigned long map_start;
262 	unsigned long map_len;
263 	int err;
264 
265 	shash->tfm = fs_info->csum_shash;
266 	crypto_shash_init(shash);
267 
268 	len = buf->len - offset;
269 
270 	while (len > 0) {
271 		/*
272 		 * Note: we don't need to check for the err == 1 case here, as
273 		 * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
274 		 * and 'min_len = 32' and the currently implemented mapping
275 		 * algorithm we cannot cross a page boundary.
276 		 */
277 		err = map_private_extent_buffer(buf, offset, 32,
278 					&kaddr, &map_start, &map_len);
279 		if (WARN_ON(err))
280 			return err;
281 		cur_len = min(len, map_len - (offset - map_start));
282 		crypto_shash_update(shash, kaddr + offset - map_start, cur_len);
283 		len -= cur_len;
284 		offset += cur_len;
285 	}
286 	memset(result, 0, BTRFS_CSUM_SIZE);
287 
288 	crypto_shash_final(shash, result);
289 
290 	return 0;
291 }
292 
293 /*
294  * we can't consider a given block up to date unless the transid of the
295  * block matches the transid in the parent node's pointer.  This is how we
296  * detect blocks that either didn't get written at all or got written
297  * in the wrong place.
298  */
299 static int verify_parent_transid(struct extent_io_tree *io_tree,
300 				 struct extent_buffer *eb, u64 parent_transid,
301 				 int atomic)
302 {
303 	struct extent_state *cached_state = NULL;
304 	int ret;
305 	bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
306 
307 	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
308 		return 0;
309 
310 	if (atomic)
311 		return -EAGAIN;
312 
313 	if (need_lock) {
314 		btrfs_tree_read_lock(eb);
315 		btrfs_set_lock_blocking_read(eb);
316 	}
317 
318 	lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
319 			 &cached_state);
320 	if (extent_buffer_uptodate(eb) &&
321 	    btrfs_header_generation(eb) == parent_transid) {
322 		ret = 0;
323 		goto out;
324 	}
325 	btrfs_err_rl(eb->fs_info,
326 		"parent transid verify failed on %llu wanted %llu found %llu",
327 			eb->start,
328 			parent_transid, btrfs_header_generation(eb));
329 	ret = 1;
330 
331 	/*
332 	 * Things reading via commit roots that don't have normal protection,
333 	 * like send, can have a really old block in cache that may point at a
334 	 * block that has been freed and re-allocated.  So don't clear uptodate
335 	 * if we find an eb that is under IO (dirty/writeback) because we could
336 	 * end up reading in the stale data and then writing it back out and
337 	 * making everybody very sad.
338 	 */
339 	if (!extent_buffer_under_io(eb))
340 		clear_extent_buffer_uptodate(eb);
341 out:
342 	unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
343 			     &cached_state);
344 	if (need_lock)
345 		btrfs_tree_read_unlock_blocking(eb);
346 	return ret;
347 }
348 
349 static bool btrfs_supported_super_csum(u16 csum_type)
350 {
351 	switch (csum_type) {
352 	case BTRFS_CSUM_TYPE_CRC32:
353 	case BTRFS_CSUM_TYPE_XXHASH:
354 	case BTRFS_CSUM_TYPE_SHA256:
355 	case BTRFS_CSUM_TYPE_BLAKE2:
356 		return true;
357 	default:
358 		return false;
359 	}
360 }
361 
362 /*
363  * Return 0 if the superblock checksum type matches the checksum value of that
364  * algorithm. Pass the raw disk superblock data.
365  */
366 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
367 				  char *raw_disk_sb)
368 {
369 	struct btrfs_super_block *disk_sb =
370 		(struct btrfs_super_block *)raw_disk_sb;
371 	char result[BTRFS_CSUM_SIZE];
372 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
373 
374 	shash->tfm = fs_info->csum_shash;
375 	crypto_shash_init(shash);
376 
377 	/*
378 	 * The super_block structure does not span the whole
379 	 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
380 	 * filled with zeros and is included in the checksum.
381 	 */
382 	crypto_shash_update(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
383 			    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
384 	crypto_shash_final(shash, result);
385 
386 	if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
387 		return 1;
388 
389 	return 0;
390 }
391 
392 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
393 			   struct btrfs_key *first_key, u64 parent_transid)
394 {
395 	struct btrfs_fs_info *fs_info = eb->fs_info;
396 	int found_level;
397 	struct btrfs_key found_key;
398 	int ret;
399 
400 	found_level = btrfs_header_level(eb);
401 	if (found_level != level) {
402 		WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
403 		     KERN_ERR "BTRFS: tree level check failed\n");
404 		btrfs_err(fs_info,
405 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
406 			  eb->start, level, found_level);
407 		return -EIO;
408 	}
409 
410 	if (!first_key)
411 		return 0;
412 
413 	/*
414 	 * For live tree block (new tree blocks in current transaction),
415 	 * we need proper lock context to avoid race, which is impossible here.
416 	 * So we only checks tree blocks which is read from disk, whose
417 	 * generation <= fs_info->last_trans_committed.
418 	 */
419 	if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
420 		return 0;
421 
422 	/* We have @first_key, so this @eb must have at least one item */
423 	if (btrfs_header_nritems(eb) == 0) {
424 		btrfs_err(fs_info,
425 		"invalid tree nritems, bytenr=%llu nritems=0 expect >0",
426 			  eb->start);
427 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
428 		return -EUCLEAN;
429 	}
430 
431 	if (found_level)
432 		btrfs_node_key_to_cpu(eb, &found_key, 0);
433 	else
434 		btrfs_item_key_to_cpu(eb, &found_key, 0);
435 	ret = btrfs_comp_cpu_keys(first_key, &found_key);
436 
437 	if (ret) {
438 		WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
439 		     KERN_ERR "BTRFS: tree first key check failed\n");
440 		btrfs_err(fs_info,
441 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
442 			  eb->start, parent_transid, first_key->objectid,
443 			  first_key->type, first_key->offset,
444 			  found_key.objectid, found_key.type,
445 			  found_key.offset);
446 	}
447 	return ret;
448 }
449 
450 /*
451  * helper to read a given tree block, doing retries as required when
452  * the checksums don't match and we have alternate mirrors to try.
453  *
454  * @parent_transid:	expected transid, skip check if 0
455  * @level:		expected level, mandatory check
456  * @first_key:		expected key of first slot, skip check if NULL
457  */
458 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
459 					  u64 parent_transid, int level,
460 					  struct btrfs_key *first_key)
461 {
462 	struct btrfs_fs_info *fs_info = eb->fs_info;
463 	struct extent_io_tree *io_tree;
464 	int failed = 0;
465 	int ret;
466 	int num_copies = 0;
467 	int mirror_num = 0;
468 	int failed_mirror = 0;
469 
470 	io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
471 	while (1) {
472 		clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
473 		ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
474 		if (!ret) {
475 			if (verify_parent_transid(io_tree, eb,
476 						   parent_transid, 0))
477 				ret = -EIO;
478 			else if (btrfs_verify_level_key(eb, level,
479 						first_key, parent_transid))
480 				ret = -EUCLEAN;
481 			else
482 				break;
483 		}
484 
485 		num_copies = btrfs_num_copies(fs_info,
486 					      eb->start, eb->len);
487 		if (num_copies == 1)
488 			break;
489 
490 		if (!failed_mirror) {
491 			failed = 1;
492 			failed_mirror = eb->read_mirror;
493 		}
494 
495 		mirror_num++;
496 		if (mirror_num == failed_mirror)
497 			mirror_num++;
498 
499 		if (mirror_num > num_copies)
500 			break;
501 	}
502 
503 	if (failed && !ret && failed_mirror)
504 		btrfs_repair_eb_io_failure(eb, failed_mirror);
505 
506 	return ret;
507 }
508 
509 /*
510  * checksum a dirty tree block before IO.  This has extra checks to make sure
511  * we only fill in the checksum field in the first page of a multi-page block
512  */
513 
514 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
515 {
516 	u64 start = page_offset(page);
517 	u64 found_start;
518 	u8 result[BTRFS_CSUM_SIZE];
519 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
520 	struct extent_buffer *eb;
521 	int ret;
522 
523 	eb = (struct extent_buffer *)page->private;
524 	if (page != eb->pages[0])
525 		return 0;
526 
527 	found_start = btrfs_header_bytenr(eb);
528 	/*
529 	 * Please do not consolidate these warnings into a single if.
530 	 * It is useful to know what went wrong.
531 	 */
532 	if (WARN_ON(found_start != start))
533 		return -EUCLEAN;
534 	if (WARN_ON(!PageUptodate(page)))
535 		return -EUCLEAN;
536 
537 	ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
538 			btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
539 
540 	if (csum_tree_block(eb, result))
541 		return -EINVAL;
542 
543 	if (btrfs_header_level(eb))
544 		ret = btrfs_check_node(eb);
545 	else
546 		ret = btrfs_check_leaf_full(eb);
547 
548 	if (ret < 0) {
549 		btrfs_print_tree(eb, 0);
550 		btrfs_err(fs_info,
551 		"block=%llu write time tree block corruption detected",
552 			  eb->start);
553 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
554 		return ret;
555 	}
556 	write_extent_buffer(eb, result, 0, csum_size);
557 
558 	return 0;
559 }
560 
561 static int check_tree_block_fsid(struct extent_buffer *eb)
562 {
563 	struct btrfs_fs_info *fs_info = eb->fs_info;
564 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
565 	u8 fsid[BTRFS_FSID_SIZE];
566 	int ret = 1;
567 
568 	read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
569 	while (fs_devices) {
570 		u8 *metadata_uuid;
571 
572 		/*
573 		 * Checking the incompat flag is only valid for the current
574 		 * fs. For seed devices it's forbidden to have their uuid
575 		 * changed so reading ->fsid in this case is fine
576 		 */
577 		if (fs_devices == fs_info->fs_devices &&
578 		    btrfs_fs_incompat(fs_info, METADATA_UUID))
579 			metadata_uuid = fs_devices->metadata_uuid;
580 		else
581 			metadata_uuid = fs_devices->fsid;
582 
583 		if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
584 			ret = 0;
585 			break;
586 		}
587 		fs_devices = fs_devices->seed;
588 	}
589 	return ret;
590 }
591 
592 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
593 				      u64 phy_offset, struct page *page,
594 				      u64 start, u64 end, int mirror)
595 {
596 	u64 found_start;
597 	int found_level;
598 	struct extent_buffer *eb;
599 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
600 	struct btrfs_fs_info *fs_info = root->fs_info;
601 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
602 	int ret = 0;
603 	u8 result[BTRFS_CSUM_SIZE];
604 	int reads_done;
605 
606 	if (!page->private)
607 		goto out;
608 
609 	eb = (struct extent_buffer *)page->private;
610 
611 	/* the pending IO might have been the only thing that kept this buffer
612 	 * in memory.  Make sure we have a ref for all this other checks
613 	 */
614 	atomic_inc(&eb->refs);
615 
616 	reads_done = atomic_dec_and_test(&eb->io_pages);
617 	if (!reads_done)
618 		goto err;
619 
620 	eb->read_mirror = mirror;
621 	if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
622 		ret = -EIO;
623 		goto err;
624 	}
625 
626 	found_start = btrfs_header_bytenr(eb);
627 	if (found_start != eb->start) {
628 		btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
629 			     eb->start, found_start);
630 		ret = -EIO;
631 		goto err;
632 	}
633 	if (check_tree_block_fsid(eb)) {
634 		btrfs_err_rl(fs_info, "bad fsid on block %llu",
635 			     eb->start);
636 		ret = -EIO;
637 		goto err;
638 	}
639 	found_level = btrfs_header_level(eb);
640 	if (found_level >= BTRFS_MAX_LEVEL) {
641 		btrfs_err(fs_info, "bad tree block level %d on %llu",
642 			  (int)btrfs_header_level(eb), eb->start);
643 		ret = -EIO;
644 		goto err;
645 	}
646 
647 	btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
648 				       eb, found_level);
649 
650 	ret = csum_tree_block(eb, result);
651 	if (ret)
652 		goto err;
653 
654 	if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
655 		u32 val;
656 		u32 found = 0;
657 
658 		memcpy(&found, result, csum_size);
659 
660 		read_extent_buffer(eb, &val, 0, csum_size);
661 		btrfs_warn_rl(fs_info,
662 		"%s checksum verify failed on %llu wanted %x found %x level %d",
663 			      fs_info->sb->s_id, eb->start,
664 			      val, found, btrfs_header_level(eb));
665 		ret = -EUCLEAN;
666 		goto err;
667 	}
668 
669 	/*
670 	 * If this is a leaf block and it is corrupt, set the corrupt bit so
671 	 * that we don't try and read the other copies of this block, just
672 	 * return -EIO.
673 	 */
674 	if (found_level == 0 && btrfs_check_leaf_full(eb)) {
675 		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
676 		ret = -EIO;
677 	}
678 
679 	if (found_level > 0 && btrfs_check_node(eb))
680 		ret = -EIO;
681 
682 	if (!ret)
683 		set_extent_buffer_uptodate(eb);
684 	else
685 		btrfs_err(fs_info,
686 			  "block=%llu read time tree block corruption detected",
687 			  eb->start);
688 err:
689 	if (reads_done &&
690 	    test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
691 		btree_readahead_hook(eb, ret);
692 
693 	if (ret) {
694 		/*
695 		 * our io error hook is going to dec the io pages
696 		 * again, we have to make sure it has something
697 		 * to decrement
698 		 */
699 		atomic_inc(&eb->io_pages);
700 		clear_extent_buffer_uptodate(eb);
701 	}
702 	free_extent_buffer(eb);
703 out:
704 	return ret;
705 }
706 
707 static void end_workqueue_bio(struct bio *bio)
708 {
709 	struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
710 	struct btrfs_fs_info *fs_info;
711 	struct btrfs_workqueue *wq;
712 
713 	fs_info = end_io_wq->info;
714 	end_io_wq->status = bio->bi_status;
715 
716 	if (bio_op(bio) == REQ_OP_WRITE) {
717 		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
718 			wq = fs_info->endio_meta_write_workers;
719 		else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
720 			wq = fs_info->endio_freespace_worker;
721 		else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
722 			wq = fs_info->endio_raid56_workers;
723 		else
724 			wq = fs_info->endio_write_workers;
725 	} else {
726 		if (unlikely(end_io_wq->metadata == BTRFS_WQ_ENDIO_DIO_REPAIR))
727 			wq = fs_info->endio_repair_workers;
728 		else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
729 			wq = fs_info->endio_raid56_workers;
730 		else if (end_io_wq->metadata)
731 			wq = fs_info->endio_meta_workers;
732 		else
733 			wq = fs_info->endio_workers;
734 	}
735 
736 	btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
737 	btrfs_queue_work(wq, &end_io_wq->work);
738 }
739 
740 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
741 			enum btrfs_wq_endio_type metadata)
742 {
743 	struct btrfs_end_io_wq *end_io_wq;
744 
745 	end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
746 	if (!end_io_wq)
747 		return BLK_STS_RESOURCE;
748 
749 	end_io_wq->private = bio->bi_private;
750 	end_io_wq->end_io = bio->bi_end_io;
751 	end_io_wq->info = info;
752 	end_io_wq->status = 0;
753 	end_io_wq->bio = bio;
754 	end_io_wq->metadata = metadata;
755 
756 	bio->bi_private = end_io_wq;
757 	bio->bi_end_io = end_workqueue_bio;
758 	return 0;
759 }
760 
761 static void run_one_async_start(struct btrfs_work *work)
762 {
763 	struct async_submit_bio *async;
764 	blk_status_t ret;
765 
766 	async = container_of(work, struct  async_submit_bio, work);
767 	ret = async->submit_bio_start(async->private_data, async->bio,
768 				      async->bio_offset);
769 	if (ret)
770 		async->status = ret;
771 }
772 
773 /*
774  * In order to insert checksums into the metadata in large chunks, we wait
775  * until bio submission time.   All the pages in the bio are checksummed and
776  * sums are attached onto the ordered extent record.
777  *
778  * At IO completion time the csums attached on the ordered extent record are
779  * inserted into the tree.
780  */
781 static void run_one_async_done(struct btrfs_work *work)
782 {
783 	struct async_submit_bio *async;
784 	struct inode *inode;
785 	blk_status_t ret;
786 
787 	async = container_of(work, struct  async_submit_bio, work);
788 	inode = async->private_data;
789 
790 	/* If an error occurred we just want to clean up the bio and move on */
791 	if (async->status) {
792 		async->bio->bi_status = async->status;
793 		bio_endio(async->bio);
794 		return;
795 	}
796 
797 	/*
798 	 * All of the bios that pass through here are from async helpers.
799 	 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
800 	 * This changes nothing when cgroups aren't in use.
801 	 */
802 	async->bio->bi_opf |= REQ_CGROUP_PUNT;
803 	ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
804 	if (ret) {
805 		async->bio->bi_status = ret;
806 		bio_endio(async->bio);
807 	}
808 }
809 
810 static void run_one_async_free(struct btrfs_work *work)
811 {
812 	struct async_submit_bio *async;
813 
814 	async = container_of(work, struct  async_submit_bio, work);
815 	kfree(async);
816 }
817 
818 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
819 				 int mirror_num, unsigned long bio_flags,
820 				 u64 bio_offset, void *private_data,
821 				 extent_submit_bio_start_t *submit_bio_start)
822 {
823 	struct async_submit_bio *async;
824 
825 	async = kmalloc(sizeof(*async), GFP_NOFS);
826 	if (!async)
827 		return BLK_STS_RESOURCE;
828 
829 	async->private_data = private_data;
830 	async->bio = bio;
831 	async->mirror_num = mirror_num;
832 	async->submit_bio_start = submit_bio_start;
833 
834 	btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
835 			run_one_async_free);
836 
837 	async->bio_offset = bio_offset;
838 
839 	async->status = 0;
840 
841 	if (op_is_sync(bio->bi_opf))
842 		btrfs_set_work_high_priority(&async->work);
843 
844 	btrfs_queue_work(fs_info->workers, &async->work);
845 	return 0;
846 }
847 
848 static blk_status_t btree_csum_one_bio(struct bio *bio)
849 {
850 	struct bio_vec *bvec;
851 	struct btrfs_root *root;
852 	int ret = 0;
853 	struct bvec_iter_all iter_all;
854 
855 	ASSERT(!bio_flagged(bio, BIO_CLONED));
856 	bio_for_each_segment_all(bvec, bio, iter_all) {
857 		root = BTRFS_I(bvec->bv_page->mapping->host)->root;
858 		ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
859 		if (ret)
860 			break;
861 	}
862 
863 	return errno_to_blk_status(ret);
864 }
865 
866 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
867 					     u64 bio_offset)
868 {
869 	/*
870 	 * when we're called for a write, we're already in the async
871 	 * submission context.  Just jump into btrfs_map_bio
872 	 */
873 	return btree_csum_one_bio(bio);
874 }
875 
876 static int check_async_write(struct btrfs_fs_info *fs_info,
877 			     struct btrfs_inode *bi)
878 {
879 	if (atomic_read(&bi->sync_writers))
880 		return 0;
881 	if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
882 		return 0;
883 	return 1;
884 }
885 
886 static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
887 					  int mirror_num,
888 					  unsigned long bio_flags)
889 {
890 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
891 	int async = check_async_write(fs_info, BTRFS_I(inode));
892 	blk_status_t ret;
893 
894 	if (bio_op(bio) != REQ_OP_WRITE) {
895 		/*
896 		 * called for a read, do the setup so that checksum validation
897 		 * can happen in the async kernel threads
898 		 */
899 		ret = btrfs_bio_wq_end_io(fs_info, bio,
900 					  BTRFS_WQ_ENDIO_METADATA);
901 		if (ret)
902 			goto out_w_error;
903 		ret = btrfs_map_bio(fs_info, bio, mirror_num);
904 	} else if (!async) {
905 		ret = btree_csum_one_bio(bio);
906 		if (ret)
907 			goto out_w_error;
908 		ret = btrfs_map_bio(fs_info, bio, mirror_num);
909 	} else {
910 		/*
911 		 * kthread helpers are used to submit writes so that
912 		 * checksumming can happen in parallel across all CPUs
913 		 */
914 		ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
915 					  0, inode, btree_submit_bio_start);
916 	}
917 
918 	if (ret)
919 		goto out_w_error;
920 	return 0;
921 
922 out_w_error:
923 	bio->bi_status = ret;
924 	bio_endio(bio);
925 	return ret;
926 }
927 
928 #ifdef CONFIG_MIGRATION
929 static int btree_migratepage(struct address_space *mapping,
930 			struct page *newpage, struct page *page,
931 			enum migrate_mode mode)
932 {
933 	/*
934 	 * we can't safely write a btree page from here,
935 	 * we haven't done the locking hook
936 	 */
937 	if (PageDirty(page))
938 		return -EAGAIN;
939 	/*
940 	 * Buffers may be managed in a filesystem specific way.
941 	 * We must have no buffers or drop them.
942 	 */
943 	if (page_has_private(page) &&
944 	    !try_to_release_page(page, GFP_KERNEL))
945 		return -EAGAIN;
946 	return migrate_page(mapping, newpage, page, mode);
947 }
948 #endif
949 
950 
951 static int btree_writepages(struct address_space *mapping,
952 			    struct writeback_control *wbc)
953 {
954 	struct btrfs_fs_info *fs_info;
955 	int ret;
956 
957 	if (wbc->sync_mode == WB_SYNC_NONE) {
958 
959 		if (wbc->for_kupdate)
960 			return 0;
961 
962 		fs_info = BTRFS_I(mapping->host)->root->fs_info;
963 		/* this is a bit racy, but that's ok */
964 		ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
965 					     BTRFS_DIRTY_METADATA_THRESH,
966 					     fs_info->dirty_metadata_batch);
967 		if (ret < 0)
968 			return 0;
969 	}
970 	return btree_write_cache_pages(mapping, wbc);
971 }
972 
973 static int btree_readpage(struct file *file, struct page *page)
974 {
975 	struct extent_io_tree *tree;
976 	tree = &BTRFS_I(page->mapping->host)->io_tree;
977 	return extent_read_full_page(tree, page, btree_get_extent, 0);
978 }
979 
980 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
981 {
982 	if (PageWriteback(page) || PageDirty(page))
983 		return 0;
984 
985 	return try_release_extent_buffer(page);
986 }
987 
988 static void btree_invalidatepage(struct page *page, unsigned int offset,
989 				 unsigned int length)
990 {
991 	struct extent_io_tree *tree;
992 	tree = &BTRFS_I(page->mapping->host)->io_tree;
993 	extent_invalidatepage(tree, page, offset);
994 	btree_releasepage(page, GFP_NOFS);
995 	if (PagePrivate(page)) {
996 		btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
997 			   "page private not zero on page %llu",
998 			   (unsigned long long)page_offset(page));
999 		ClearPagePrivate(page);
1000 		set_page_private(page, 0);
1001 		put_page(page);
1002 	}
1003 }
1004 
1005 static int btree_set_page_dirty(struct page *page)
1006 {
1007 #ifdef DEBUG
1008 	struct extent_buffer *eb;
1009 
1010 	BUG_ON(!PagePrivate(page));
1011 	eb = (struct extent_buffer *)page->private;
1012 	BUG_ON(!eb);
1013 	BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1014 	BUG_ON(!atomic_read(&eb->refs));
1015 	btrfs_assert_tree_locked(eb);
1016 #endif
1017 	return __set_page_dirty_nobuffers(page);
1018 }
1019 
1020 static const struct address_space_operations btree_aops = {
1021 	.readpage	= btree_readpage,
1022 	.writepages	= btree_writepages,
1023 	.releasepage	= btree_releasepage,
1024 	.invalidatepage = btree_invalidatepage,
1025 #ifdef CONFIG_MIGRATION
1026 	.migratepage	= btree_migratepage,
1027 #endif
1028 	.set_page_dirty = btree_set_page_dirty,
1029 };
1030 
1031 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1032 {
1033 	struct extent_buffer *buf = NULL;
1034 	int ret;
1035 
1036 	buf = btrfs_find_create_tree_block(fs_info, bytenr);
1037 	if (IS_ERR(buf))
1038 		return;
1039 
1040 	ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1041 	if (ret < 0)
1042 		free_extent_buffer_stale(buf);
1043 	else
1044 		free_extent_buffer(buf);
1045 }
1046 
1047 struct extent_buffer *btrfs_find_create_tree_block(
1048 						struct btrfs_fs_info *fs_info,
1049 						u64 bytenr)
1050 {
1051 	if (btrfs_is_testing(fs_info))
1052 		return alloc_test_extent_buffer(fs_info, bytenr);
1053 	return alloc_extent_buffer(fs_info, bytenr);
1054 }
1055 
1056 /*
1057  * Read tree block at logical address @bytenr and do variant basic but critical
1058  * verification.
1059  *
1060  * @parent_transid:	expected transid of this tree block, skip check if 0
1061  * @level:		expected level, mandatory check
1062  * @first_key:		expected key in slot 0, skip check if NULL
1063  */
1064 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1065 				      u64 parent_transid, int level,
1066 				      struct btrfs_key *first_key)
1067 {
1068 	struct extent_buffer *buf = NULL;
1069 	int ret;
1070 
1071 	buf = btrfs_find_create_tree_block(fs_info, bytenr);
1072 	if (IS_ERR(buf))
1073 		return buf;
1074 
1075 	ret = btree_read_extent_buffer_pages(buf, parent_transid,
1076 					     level, first_key);
1077 	if (ret) {
1078 		free_extent_buffer_stale(buf);
1079 		return ERR_PTR(ret);
1080 	}
1081 	return buf;
1082 
1083 }
1084 
1085 void btrfs_clean_tree_block(struct extent_buffer *buf)
1086 {
1087 	struct btrfs_fs_info *fs_info = buf->fs_info;
1088 	if (btrfs_header_generation(buf) ==
1089 	    fs_info->running_transaction->transid) {
1090 		btrfs_assert_tree_locked(buf);
1091 
1092 		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1093 			percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1094 						 -buf->len,
1095 						 fs_info->dirty_metadata_batch);
1096 			/* ugh, clear_extent_buffer_dirty needs to lock the page */
1097 			btrfs_set_lock_blocking_write(buf);
1098 			clear_extent_buffer_dirty(buf);
1099 		}
1100 	}
1101 }
1102 
1103 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1104 {
1105 	struct btrfs_subvolume_writers *writers;
1106 	int ret;
1107 
1108 	writers = kmalloc(sizeof(*writers), GFP_NOFS);
1109 	if (!writers)
1110 		return ERR_PTR(-ENOMEM);
1111 
1112 	ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1113 	if (ret < 0) {
1114 		kfree(writers);
1115 		return ERR_PTR(ret);
1116 	}
1117 
1118 	init_waitqueue_head(&writers->wait);
1119 	return writers;
1120 }
1121 
1122 static void
1123 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1124 {
1125 	percpu_counter_destroy(&writers->counter);
1126 	kfree(writers);
1127 }
1128 
1129 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1130 			 u64 objectid)
1131 {
1132 	bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1133 	root->node = NULL;
1134 	root->commit_root = NULL;
1135 	root->state = 0;
1136 	root->orphan_cleanup_state = 0;
1137 
1138 	root->last_trans = 0;
1139 	root->highest_objectid = 0;
1140 	root->nr_delalloc_inodes = 0;
1141 	root->nr_ordered_extents = 0;
1142 	root->inode_tree = RB_ROOT;
1143 	INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1144 	root->block_rsv = NULL;
1145 
1146 	INIT_LIST_HEAD(&root->dirty_list);
1147 	INIT_LIST_HEAD(&root->root_list);
1148 	INIT_LIST_HEAD(&root->delalloc_inodes);
1149 	INIT_LIST_HEAD(&root->delalloc_root);
1150 	INIT_LIST_HEAD(&root->ordered_extents);
1151 	INIT_LIST_HEAD(&root->ordered_root);
1152 	INIT_LIST_HEAD(&root->reloc_dirty_list);
1153 	INIT_LIST_HEAD(&root->logged_list[0]);
1154 	INIT_LIST_HEAD(&root->logged_list[1]);
1155 	spin_lock_init(&root->inode_lock);
1156 	spin_lock_init(&root->delalloc_lock);
1157 	spin_lock_init(&root->ordered_extent_lock);
1158 	spin_lock_init(&root->accounting_lock);
1159 	spin_lock_init(&root->log_extents_lock[0]);
1160 	spin_lock_init(&root->log_extents_lock[1]);
1161 	spin_lock_init(&root->qgroup_meta_rsv_lock);
1162 	mutex_init(&root->objectid_mutex);
1163 	mutex_init(&root->log_mutex);
1164 	mutex_init(&root->ordered_extent_mutex);
1165 	mutex_init(&root->delalloc_mutex);
1166 	init_waitqueue_head(&root->log_writer_wait);
1167 	init_waitqueue_head(&root->log_commit_wait[0]);
1168 	init_waitqueue_head(&root->log_commit_wait[1]);
1169 	INIT_LIST_HEAD(&root->log_ctxs[0]);
1170 	INIT_LIST_HEAD(&root->log_ctxs[1]);
1171 	atomic_set(&root->log_commit[0], 0);
1172 	atomic_set(&root->log_commit[1], 0);
1173 	atomic_set(&root->log_writers, 0);
1174 	atomic_set(&root->log_batch, 0);
1175 	refcount_set(&root->refs, 1);
1176 	atomic_set(&root->will_be_snapshotted, 0);
1177 	atomic_set(&root->snapshot_force_cow, 0);
1178 	atomic_set(&root->nr_swapfiles, 0);
1179 	root->log_transid = 0;
1180 	root->log_transid_committed = -1;
1181 	root->last_log_commit = 0;
1182 	if (!dummy)
1183 		extent_io_tree_init(fs_info, &root->dirty_log_pages,
1184 				    IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1185 
1186 	memset(&root->root_key, 0, sizeof(root->root_key));
1187 	memset(&root->root_item, 0, sizeof(root->root_item));
1188 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1189 	if (!dummy)
1190 		root->defrag_trans_start = fs_info->generation;
1191 	else
1192 		root->defrag_trans_start = 0;
1193 	root->root_key.objectid = objectid;
1194 	root->anon_dev = 0;
1195 
1196 	spin_lock_init(&root->root_item_lock);
1197 	btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1198 }
1199 
1200 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1201 		gfp_t flags)
1202 {
1203 	struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1204 	if (root)
1205 		root->fs_info = fs_info;
1206 	return root;
1207 }
1208 
1209 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1210 /* Should only be used by the testing infrastructure */
1211 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1212 {
1213 	struct btrfs_root *root;
1214 
1215 	if (!fs_info)
1216 		return ERR_PTR(-EINVAL);
1217 
1218 	root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1219 	if (!root)
1220 		return ERR_PTR(-ENOMEM);
1221 
1222 	/* We don't use the stripesize in selftest, set it as sectorsize */
1223 	__setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1224 	root->alloc_bytenr = 0;
1225 
1226 	return root;
1227 }
1228 #endif
1229 
1230 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1231 				     u64 objectid)
1232 {
1233 	struct btrfs_fs_info *fs_info = trans->fs_info;
1234 	struct extent_buffer *leaf;
1235 	struct btrfs_root *tree_root = fs_info->tree_root;
1236 	struct btrfs_root *root;
1237 	struct btrfs_key key;
1238 	unsigned int nofs_flag;
1239 	int ret = 0;
1240 	uuid_le uuid = NULL_UUID_LE;
1241 
1242 	/*
1243 	 * We're holding a transaction handle, so use a NOFS memory allocation
1244 	 * context to avoid deadlock if reclaim happens.
1245 	 */
1246 	nofs_flag = memalloc_nofs_save();
1247 	root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1248 	memalloc_nofs_restore(nofs_flag);
1249 	if (!root)
1250 		return ERR_PTR(-ENOMEM);
1251 
1252 	__setup_root(root, fs_info, objectid);
1253 	root->root_key.objectid = objectid;
1254 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1255 	root->root_key.offset = 0;
1256 
1257 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1258 	if (IS_ERR(leaf)) {
1259 		ret = PTR_ERR(leaf);
1260 		leaf = NULL;
1261 		goto fail;
1262 	}
1263 
1264 	root->node = leaf;
1265 	btrfs_mark_buffer_dirty(leaf);
1266 
1267 	root->commit_root = btrfs_root_node(root);
1268 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1269 
1270 	root->root_item.flags = 0;
1271 	root->root_item.byte_limit = 0;
1272 	btrfs_set_root_bytenr(&root->root_item, leaf->start);
1273 	btrfs_set_root_generation(&root->root_item, trans->transid);
1274 	btrfs_set_root_level(&root->root_item, 0);
1275 	btrfs_set_root_refs(&root->root_item, 1);
1276 	btrfs_set_root_used(&root->root_item, leaf->len);
1277 	btrfs_set_root_last_snapshot(&root->root_item, 0);
1278 	btrfs_set_root_dirid(&root->root_item, 0);
1279 	if (is_fstree(objectid))
1280 		uuid_le_gen(&uuid);
1281 	memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1282 	root->root_item.drop_level = 0;
1283 
1284 	key.objectid = objectid;
1285 	key.type = BTRFS_ROOT_ITEM_KEY;
1286 	key.offset = 0;
1287 	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1288 	if (ret)
1289 		goto fail;
1290 
1291 	btrfs_tree_unlock(leaf);
1292 
1293 	return root;
1294 
1295 fail:
1296 	if (leaf) {
1297 		btrfs_tree_unlock(leaf);
1298 		free_extent_buffer(root->commit_root);
1299 		free_extent_buffer(leaf);
1300 	}
1301 	kfree(root);
1302 
1303 	return ERR_PTR(ret);
1304 }
1305 
1306 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1307 					 struct btrfs_fs_info *fs_info)
1308 {
1309 	struct btrfs_root *root;
1310 	struct extent_buffer *leaf;
1311 
1312 	root = btrfs_alloc_root(fs_info, GFP_NOFS);
1313 	if (!root)
1314 		return ERR_PTR(-ENOMEM);
1315 
1316 	__setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1317 
1318 	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1319 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1320 	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1321 
1322 	/*
1323 	 * DON'T set REF_COWS for log trees
1324 	 *
1325 	 * log trees do not get reference counted because they go away
1326 	 * before a real commit is actually done.  They do store pointers
1327 	 * to file data extents, and those reference counts still get
1328 	 * updated (along with back refs to the log tree).
1329 	 */
1330 
1331 	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1332 			NULL, 0, 0, 0);
1333 	if (IS_ERR(leaf)) {
1334 		kfree(root);
1335 		return ERR_CAST(leaf);
1336 	}
1337 
1338 	root->node = leaf;
1339 
1340 	btrfs_mark_buffer_dirty(root->node);
1341 	btrfs_tree_unlock(root->node);
1342 	return root;
1343 }
1344 
1345 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1346 			     struct btrfs_fs_info *fs_info)
1347 {
1348 	struct btrfs_root *log_root;
1349 
1350 	log_root = alloc_log_tree(trans, fs_info);
1351 	if (IS_ERR(log_root))
1352 		return PTR_ERR(log_root);
1353 	WARN_ON(fs_info->log_root_tree);
1354 	fs_info->log_root_tree = log_root;
1355 	return 0;
1356 }
1357 
1358 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1359 		       struct btrfs_root *root)
1360 {
1361 	struct btrfs_fs_info *fs_info = root->fs_info;
1362 	struct btrfs_root *log_root;
1363 	struct btrfs_inode_item *inode_item;
1364 
1365 	log_root = alloc_log_tree(trans, fs_info);
1366 	if (IS_ERR(log_root))
1367 		return PTR_ERR(log_root);
1368 
1369 	log_root->last_trans = trans->transid;
1370 	log_root->root_key.offset = root->root_key.objectid;
1371 
1372 	inode_item = &log_root->root_item.inode;
1373 	btrfs_set_stack_inode_generation(inode_item, 1);
1374 	btrfs_set_stack_inode_size(inode_item, 3);
1375 	btrfs_set_stack_inode_nlink(inode_item, 1);
1376 	btrfs_set_stack_inode_nbytes(inode_item,
1377 				     fs_info->nodesize);
1378 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1379 
1380 	btrfs_set_root_node(&log_root->root_item, log_root->node);
1381 
1382 	WARN_ON(root->log_root);
1383 	root->log_root = log_root;
1384 	root->log_transid = 0;
1385 	root->log_transid_committed = -1;
1386 	root->last_log_commit = 0;
1387 	return 0;
1388 }
1389 
1390 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1391 					       struct btrfs_key *key)
1392 {
1393 	struct btrfs_root *root;
1394 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1395 	struct btrfs_path *path;
1396 	u64 generation;
1397 	int ret;
1398 	int level;
1399 
1400 	path = btrfs_alloc_path();
1401 	if (!path)
1402 		return ERR_PTR(-ENOMEM);
1403 
1404 	root = btrfs_alloc_root(fs_info, GFP_NOFS);
1405 	if (!root) {
1406 		ret = -ENOMEM;
1407 		goto alloc_fail;
1408 	}
1409 
1410 	__setup_root(root, fs_info, key->objectid);
1411 
1412 	ret = btrfs_find_root(tree_root, key, path,
1413 			      &root->root_item, &root->root_key);
1414 	if (ret) {
1415 		if (ret > 0)
1416 			ret = -ENOENT;
1417 		goto find_fail;
1418 	}
1419 
1420 	generation = btrfs_root_generation(&root->root_item);
1421 	level = btrfs_root_level(&root->root_item);
1422 	root->node = read_tree_block(fs_info,
1423 				     btrfs_root_bytenr(&root->root_item),
1424 				     generation, level, NULL);
1425 	if (IS_ERR(root->node)) {
1426 		ret = PTR_ERR(root->node);
1427 		goto find_fail;
1428 	} else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1429 		ret = -EIO;
1430 		free_extent_buffer(root->node);
1431 		goto find_fail;
1432 	}
1433 	root->commit_root = btrfs_root_node(root);
1434 out:
1435 	btrfs_free_path(path);
1436 	return root;
1437 
1438 find_fail:
1439 	kfree(root);
1440 alloc_fail:
1441 	root = ERR_PTR(ret);
1442 	goto out;
1443 }
1444 
1445 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1446 				      struct btrfs_key *location)
1447 {
1448 	struct btrfs_root *root;
1449 
1450 	root = btrfs_read_tree_root(tree_root, location);
1451 	if (IS_ERR(root))
1452 		return root;
1453 
1454 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1455 		set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1456 		btrfs_check_and_init_root_item(&root->root_item);
1457 	}
1458 
1459 	return root;
1460 }
1461 
1462 int btrfs_init_fs_root(struct btrfs_root *root)
1463 {
1464 	int ret;
1465 	struct btrfs_subvolume_writers *writers;
1466 
1467 	root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1468 	root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1469 					GFP_NOFS);
1470 	if (!root->free_ino_pinned || !root->free_ino_ctl) {
1471 		ret = -ENOMEM;
1472 		goto fail;
1473 	}
1474 
1475 	writers = btrfs_alloc_subvolume_writers();
1476 	if (IS_ERR(writers)) {
1477 		ret = PTR_ERR(writers);
1478 		goto fail;
1479 	}
1480 	root->subv_writers = writers;
1481 
1482 	btrfs_init_free_ino_ctl(root);
1483 	spin_lock_init(&root->ino_cache_lock);
1484 	init_waitqueue_head(&root->ino_cache_wait);
1485 
1486 	ret = get_anon_bdev(&root->anon_dev);
1487 	if (ret)
1488 		goto fail;
1489 
1490 	mutex_lock(&root->objectid_mutex);
1491 	ret = btrfs_find_highest_objectid(root,
1492 					&root->highest_objectid);
1493 	if (ret) {
1494 		mutex_unlock(&root->objectid_mutex);
1495 		goto fail;
1496 	}
1497 
1498 	ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1499 
1500 	mutex_unlock(&root->objectid_mutex);
1501 
1502 	return 0;
1503 fail:
1504 	/* The caller is responsible to call btrfs_free_fs_root */
1505 	return ret;
1506 }
1507 
1508 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1509 					u64 root_id)
1510 {
1511 	struct btrfs_root *root;
1512 
1513 	spin_lock(&fs_info->fs_roots_radix_lock);
1514 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1515 				 (unsigned long)root_id);
1516 	spin_unlock(&fs_info->fs_roots_radix_lock);
1517 	return root;
1518 }
1519 
1520 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1521 			 struct btrfs_root *root)
1522 {
1523 	int ret;
1524 
1525 	ret = radix_tree_preload(GFP_NOFS);
1526 	if (ret)
1527 		return ret;
1528 
1529 	spin_lock(&fs_info->fs_roots_radix_lock);
1530 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1531 				(unsigned long)root->root_key.objectid,
1532 				root);
1533 	if (ret == 0)
1534 		set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1535 	spin_unlock(&fs_info->fs_roots_radix_lock);
1536 	radix_tree_preload_end();
1537 
1538 	return ret;
1539 }
1540 
1541 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1542 				     struct btrfs_key *location,
1543 				     bool check_ref)
1544 {
1545 	struct btrfs_root *root;
1546 	struct btrfs_path *path;
1547 	struct btrfs_key key;
1548 	int ret;
1549 
1550 	if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1551 		return fs_info->tree_root;
1552 	if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1553 		return fs_info->extent_root;
1554 	if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1555 		return fs_info->chunk_root;
1556 	if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1557 		return fs_info->dev_root;
1558 	if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1559 		return fs_info->csum_root;
1560 	if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1561 		return fs_info->quota_root ? fs_info->quota_root :
1562 					     ERR_PTR(-ENOENT);
1563 	if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1564 		return fs_info->uuid_root ? fs_info->uuid_root :
1565 					    ERR_PTR(-ENOENT);
1566 	if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1567 		return fs_info->free_space_root ? fs_info->free_space_root :
1568 						  ERR_PTR(-ENOENT);
1569 again:
1570 	root = btrfs_lookup_fs_root(fs_info, location->objectid);
1571 	if (root) {
1572 		if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1573 			return ERR_PTR(-ENOENT);
1574 		return root;
1575 	}
1576 
1577 	root = btrfs_read_fs_root(fs_info->tree_root, location);
1578 	if (IS_ERR(root))
1579 		return root;
1580 
1581 	if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1582 		ret = -ENOENT;
1583 		goto fail;
1584 	}
1585 
1586 	ret = btrfs_init_fs_root(root);
1587 	if (ret)
1588 		goto fail;
1589 
1590 	path = btrfs_alloc_path();
1591 	if (!path) {
1592 		ret = -ENOMEM;
1593 		goto fail;
1594 	}
1595 	key.objectid = BTRFS_ORPHAN_OBJECTID;
1596 	key.type = BTRFS_ORPHAN_ITEM_KEY;
1597 	key.offset = location->objectid;
1598 
1599 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1600 	btrfs_free_path(path);
1601 	if (ret < 0)
1602 		goto fail;
1603 	if (ret == 0)
1604 		set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1605 
1606 	ret = btrfs_insert_fs_root(fs_info, root);
1607 	if (ret) {
1608 		if (ret == -EEXIST) {
1609 			btrfs_free_fs_root(root);
1610 			goto again;
1611 		}
1612 		goto fail;
1613 	}
1614 	return root;
1615 fail:
1616 	btrfs_free_fs_root(root);
1617 	return ERR_PTR(ret);
1618 }
1619 
1620 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1621 {
1622 	struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1623 	int ret = 0;
1624 	struct btrfs_device *device;
1625 	struct backing_dev_info *bdi;
1626 
1627 	rcu_read_lock();
1628 	list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1629 		if (!device->bdev)
1630 			continue;
1631 		bdi = device->bdev->bd_bdi;
1632 		if (bdi_congested(bdi, bdi_bits)) {
1633 			ret = 1;
1634 			break;
1635 		}
1636 	}
1637 	rcu_read_unlock();
1638 	return ret;
1639 }
1640 
1641 /*
1642  * called by the kthread helper functions to finally call the bio end_io
1643  * functions.  This is where read checksum verification actually happens
1644  */
1645 static void end_workqueue_fn(struct btrfs_work *work)
1646 {
1647 	struct bio *bio;
1648 	struct btrfs_end_io_wq *end_io_wq;
1649 
1650 	end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1651 	bio = end_io_wq->bio;
1652 
1653 	bio->bi_status = end_io_wq->status;
1654 	bio->bi_private = end_io_wq->private;
1655 	bio->bi_end_io = end_io_wq->end_io;
1656 	bio_endio(bio);
1657 	kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1658 }
1659 
1660 static int cleaner_kthread(void *arg)
1661 {
1662 	struct btrfs_root *root = arg;
1663 	struct btrfs_fs_info *fs_info = root->fs_info;
1664 	int again;
1665 
1666 	while (1) {
1667 		again = 0;
1668 
1669 		set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1670 
1671 		/* Make the cleaner go to sleep early. */
1672 		if (btrfs_need_cleaner_sleep(fs_info))
1673 			goto sleep;
1674 
1675 		/*
1676 		 * Do not do anything if we might cause open_ctree() to block
1677 		 * before we have finished mounting the filesystem.
1678 		 */
1679 		if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1680 			goto sleep;
1681 
1682 		if (!mutex_trylock(&fs_info->cleaner_mutex))
1683 			goto sleep;
1684 
1685 		/*
1686 		 * Avoid the problem that we change the status of the fs
1687 		 * during the above check and trylock.
1688 		 */
1689 		if (btrfs_need_cleaner_sleep(fs_info)) {
1690 			mutex_unlock(&fs_info->cleaner_mutex);
1691 			goto sleep;
1692 		}
1693 
1694 		btrfs_run_delayed_iputs(fs_info);
1695 
1696 		again = btrfs_clean_one_deleted_snapshot(root);
1697 		mutex_unlock(&fs_info->cleaner_mutex);
1698 
1699 		/*
1700 		 * The defragger has dealt with the R/O remount and umount,
1701 		 * needn't do anything special here.
1702 		 */
1703 		btrfs_run_defrag_inodes(fs_info);
1704 
1705 		/*
1706 		 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1707 		 * with relocation (btrfs_relocate_chunk) and relocation
1708 		 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1709 		 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1710 		 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1711 		 * unused block groups.
1712 		 */
1713 		btrfs_delete_unused_bgs(fs_info);
1714 sleep:
1715 		clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1716 		if (kthread_should_park())
1717 			kthread_parkme();
1718 		if (kthread_should_stop())
1719 			return 0;
1720 		if (!again) {
1721 			set_current_state(TASK_INTERRUPTIBLE);
1722 			schedule();
1723 			__set_current_state(TASK_RUNNING);
1724 		}
1725 	}
1726 }
1727 
1728 static int transaction_kthread(void *arg)
1729 {
1730 	struct btrfs_root *root = arg;
1731 	struct btrfs_fs_info *fs_info = root->fs_info;
1732 	struct btrfs_trans_handle *trans;
1733 	struct btrfs_transaction *cur;
1734 	u64 transid;
1735 	time64_t now;
1736 	unsigned long delay;
1737 	bool cannot_commit;
1738 
1739 	do {
1740 		cannot_commit = false;
1741 		delay = HZ * fs_info->commit_interval;
1742 		mutex_lock(&fs_info->transaction_kthread_mutex);
1743 
1744 		spin_lock(&fs_info->trans_lock);
1745 		cur = fs_info->running_transaction;
1746 		if (!cur) {
1747 			spin_unlock(&fs_info->trans_lock);
1748 			goto sleep;
1749 		}
1750 
1751 		now = ktime_get_seconds();
1752 		if (cur->state < TRANS_STATE_COMMIT_START &&
1753 		    !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1754 		    (now < cur->start_time ||
1755 		     now - cur->start_time < fs_info->commit_interval)) {
1756 			spin_unlock(&fs_info->trans_lock);
1757 			delay = HZ * 5;
1758 			goto sleep;
1759 		}
1760 		transid = cur->transid;
1761 		spin_unlock(&fs_info->trans_lock);
1762 
1763 		/* If the file system is aborted, this will always fail. */
1764 		trans = btrfs_attach_transaction(root);
1765 		if (IS_ERR(trans)) {
1766 			if (PTR_ERR(trans) != -ENOENT)
1767 				cannot_commit = true;
1768 			goto sleep;
1769 		}
1770 		if (transid == trans->transid) {
1771 			btrfs_commit_transaction(trans);
1772 		} else {
1773 			btrfs_end_transaction(trans);
1774 		}
1775 sleep:
1776 		wake_up_process(fs_info->cleaner_kthread);
1777 		mutex_unlock(&fs_info->transaction_kthread_mutex);
1778 
1779 		if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1780 				      &fs_info->fs_state)))
1781 			btrfs_cleanup_transaction(fs_info);
1782 		if (!kthread_should_stop() &&
1783 				(!btrfs_transaction_blocked(fs_info) ||
1784 				 cannot_commit))
1785 			schedule_timeout_interruptible(delay);
1786 	} while (!kthread_should_stop());
1787 	return 0;
1788 }
1789 
1790 /*
1791  * This will find the highest generation in the array of root backups.  The
1792  * index of the highest array is returned, or -EINVAL if we can't find
1793  * anything.
1794  *
1795  * We check to make sure the array is valid by comparing the
1796  * generation of the latest  root in the array with the generation
1797  * in the super block.  If they don't match we pitch it.
1798  */
1799 static int find_newest_super_backup(struct btrfs_fs_info *info)
1800 {
1801 	const u64 newest_gen = btrfs_super_generation(info->super_copy);
1802 	u64 cur;
1803 	struct btrfs_root_backup *root_backup;
1804 	int i;
1805 
1806 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1807 		root_backup = info->super_copy->super_roots + i;
1808 		cur = btrfs_backup_tree_root_gen(root_backup);
1809 		if (cur == newest_gen)
1810 			return i;
1811 	}
1812 
1813 	return -EINVAL;
1814 }
1815 
1816 /*
1817  * copy all the root pointers into the super backup array.
1818  * this will bump the backup pointer by one when it is
1819  * done
1820  */
1821 static void backup_super_roots(struct btrfs_fs_info *info)
1822 {
1823 	const int next_backup = info->backup_root_index;
1824 	struct btrfs_root_backup *root_backup;
1825 
1826 	root_backup = info->super_for_commit->super_roots + next_backup;
1827 
1828 	/*
1829 	 * make sure all of our padding and empty slots get zero filled
1830 	 * regardless of which ones we use today
1831 	 */
1832 	memset(root_backup, 0, sizeof(*root_backup));
1833 
1834 	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1835 
1836 	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1837 	btrfs_set_backup_tree_root_gen(root_backup,
1838 			       btrfs_header_generation(info->tree_root->node));
1839 
1840 	btrfs_set_backup_tree_root_level(root_backup,
1841 			       btrfs_header_level(info->tree_root->node));
1842 
1843 	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1844 	btrfs_set_backup_chunk_root_gen(root_backup,
1845 			       btrfs_header_generation(info->chunk_root->node));
1846 	btrfs_set_backup_chunk_root_level(root_backup,
1847 			       btrfs_header_level(info->chunk_root->node));
1848 
1849 	btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1850 	btrfs_set_backup_extent_root_gen(root_backup,
1851 			       btrfs_header_generation(info->extent_root->node));
1852 	btrfs_set_backup_extent_root_level(root_backup,
1853 			       btrfs_header_level(info->extent_root->node));
1854 
1855 	/*
1856 	 * we might commit during log recovery, which happens before we set
1857 	 * the fs_root.  Make sure it is valid before we fill it in.
1858 	 */
1859 	if (info->fs_root && info->fs_root->node) {
1860 		btrfs_set_backup_fs_root(root_backup,
1861 					 info->fs_root->node->start);
1862 		btrfs_set_backup_fs_root_gen(root_backup,
1863 			       btrfs_header_generation(info->fs_root->node));
1864 		btrfs_set_backup_fs_root_level(root_backup,
1865 			       btrfs_header_level(info->fs_root->node));
1866 	}
1867 
1868 	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1869 	btrfs_set_backup_dev_root_gen(root_backup,
1870 			       btrfs_header_generation(info->dev_root->node));
1871 	btrfs_set_backup_dev_root_level(root_backup,
1872 				       btrfs_header_level(info->dev_root->node));
1873 
1874 	btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1875 	btrfs_set_backup_csum_root_gen(root_backup,
1876 			       btrfs_header_generation(info->csum_root->node));
1877 	btrfs_set_backup_csum_root_level(root_backup,
1878 			       btrfs_header_level(info->csum_root->node));
1879 
1880 	btrfs_set_backup_total_bytes(root_backup,
1881 			     btrfs_super_total_bytes(info->super_copy));
1882 	btrfs_set_backup_bytes_used(root_backup,
1883 			     btrfs_super_bytes_used(info->super_copy));
1884 	btrfs_set_backup_num_devices(root_backup,
1885 			     btrfs_super_num_devices(info->super_copy));
1886 
1887 	/*
1888 	 * if we don't copy this out to the super_copy, it won't get remembered
1889 	 * for the next commit
1890 	 */
1891 	memcpy(&info->super_copy->super_roots,
1892 	       &info->super_for_commit->super_roots,
1893 	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1894 }
1895 
1896 /*
1897  * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1898  * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1899  *
1900  * fs_info - filesystem whose backup roots need to be read
1901  * priority - priority of backup root required
1902  *
1903  * Returns backup root index on success and -EINVAL otherwise.
1904  */
1905 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1906 {
1907 	int backup_index = find_newest_super_backup(fs_info);
1908 	struct btrfs_super_block *super = fs_info->super_copy;
1909 	struct btrfs_root_backup *root_backup;
1910 
1911 	if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1912 		if (priority == 0)
1913 			return backup_index;
1914 
1915 		backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1916 		backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1917 	} else {
1918 		return -EINVAL;
1919 	}
1920 
1921 	root_backup = super->super_roots + backup_index;
1922 
1923 	btrfs_set_super_generation(super,
1924 				   btrfs_backup_tree_root_gen(root_backup));
1925 	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1926 	btrfs_set_super_root_level(super,
1927 				   btrfs_backup_tree_root_level(root_backup));
1928 	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1929 
1930 	/*
1931 	 * Fixme: the total bytes and num_devices need to match or we should
1932 	 * need a fsck
1933 	 */
1934 	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1935 	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1936 
1937 	return backup_index;
1938 }
1939 
1940 /* helper to cleanup workers */
1941 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1942 {
1943 	btrfs_destroy_workqueue(fs_info->fixup_workers);
1944 	btrfs_destroy_workqueue(fs_info->delalloc_workers);
1945 	btrfs_destroy_workqueue(fs_info->workers);
1946 	btrfs_destroy_workqueue(fs_info->endio_workers);
1947 	btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
1948 	btrfs_destroy_workqueue(fs_info->endio_repair_workers);
1949 	btrfs_destroy_workqueue(fs_info->rmw_workers);
1950 	btrfs_destroy_workqueue(fs_info->endio_write_workers);
1951 	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1952 	btrfs_destroy_workqueue(fs_info->delayed_workers);
1953 	btrfs_destroy_workqueue(fs_info->caching_workers);
1954 	btrfs_destroy_workqueue(fs_info->readahead_workers);
1955 	btrfs_destroy_workqueue(fs_info->flush_workers);
1956 	btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1957 	if (fs_info->discard_ctl.discard_workers)
1958 		destroy_workqueue(fs_info->discard_ctl.discard_workers);
1959 	/*
1960 	 * Now that all other work queues are destroyed, we can safely destroy
1961 	 * the queues used for metadata I/O, since tasks from those other work
1962 	 * queues can do metadata I/O operations.
1963 	 */
1964 	btrfs_destroy_workqueue(fs_info->endio_meta_workers);
1965 	btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
1966 }
1967 
1968 static void free_root_extent_buffers(struct btrfs_root *root)
1969 {
1970 	if (root) {
1971 		free_extent_buffer(root->node);
1972 		free_extent_buffer(root->commit_root);
1973 		root->node = NULL;
1974 		root->commit_root = NULL;
1975 	}
1976 }
1977 
1978 /* helper to cleanup tree roots */
1979 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1980 {
1981 	free_root_extent_buffers(info->tree_root);
1982 
1983 	free_root_extent_buffers(info->dev_root);
1984 	free_root_extent_buffers(info->extent_root);
1985 	free_root_extent_buffers(info->csum_root);
1986 	free_root_extent_buffers(info->quota_root);
1987 	free_root_extent_buffers(info->uuid_root);
1988 	if (free_chunk_root)
1989 		free_root_extent_buffers(info->chunk_root);
1990 	free_root_extent_buffers(info->free_space_root);
1991 }
1992 
1993 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1994 {
1995 	int ret;
1996 	struct btrfs_root *gang[8];
1997 	int i;
1998 
1999 	while (!list_empty(&fs_info->dead_roots)) {
2000 		gang[0] = list_entry(fs_info->dead_roots.next,
2001 				     struct btrfs_root, root_list);
2002 		list_del(&gang[0]->root_list);
2003 
2004 		if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2005 			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2006 		} else {
2007 			free_extent_buffer(gang[0]->node);
2008 			free_extent_buffer(gang[0]->commit_root);
2009 			btrfs_put_fs_root(gang[0]);
2010 		}
2011 	}
2012 
2013 	while (1) {
2014 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2015 					     (void **)gang, 0,
2016 					     ARRAY_SIZE(gang));
2017 		if (!ret)
2018 			break;
2019 		for (i = 0; i < ret; i++)
2020 			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2021 	}
2022 
2023 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2024 		btrfs_free_log_root_tree(NULL, fs_info);
2025 		btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2026 	}
2027 }
2028 
2029 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2030 {
2031 	mutex_init(&fs_info->scrub_lock);
2032 	atomic_set(&fs_info->scrubs_running, 0);
2033 	atomic_set(&fs_info->scrub_pause_req, 0);
2034 	atomic_set(&fs_info->scrubs_paused, 0);
2035 	atomic_set(&fs_info->scrub_cancel_req, 0);
2036 	init_waitqueue_head(&fs_info->scrub_pause_wait);
2037 	refcount_set(&fs_info->scrub_workers_refcnt, 0);
2038 }
2039 
2040 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2041 {
2042 	spin_lock_init(&fs_info->balance_lock);
2043 	mutex_init(&fs_info->balance_mutex);
2044 	atomic_set(&fs_info->balance_pause_req, 0);
2045 	atomic_set(&fs_info->balance_cancel_req, 0);
2046 	fs_info->balance_ctl = NULL;
2047 	init_waitqueue_head(&fs_info->balance_wait_q);
2048 }
2049 
2050 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2051 {
2052 	struct inode *inode = fs_info->btree_inode;
2053 
2054 	inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2055 	set_nlink(inode, 1);
2056 	/*
2057 	 * we set the i_size on the btree inode to the max possible int.
2058 	 * the real end of the address space is determined by all of
2059 	 * the devices in the system
2060 	 */
2061 	inode->i_size = OFFSET_MAX;
2062 	inode->i_mapping->a_ops = &btree_aops;
2063 
2064 	RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2065 	extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2066 			    IO_TREE_INODE_IO, inode);
2067 	BTRFS_I(inode)->io_tree.track_uptodate = false;
2068 	extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2069 
2070 	BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2071 
2072 	BTRFS_I(inode)->root = fs_info->tree_root;
2073 	memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2074 	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2075 	btrfs_insert_inode_hash(inode);
2076 }
2077 
2078 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2079 {
2080 	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2081 	init_rwsem(&fs_info->dev_replace.rwsem);
2082 	init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2083 }
2084 
2085 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2086 {
2087 	spin_lock_init(&fs_info->qgroup_lock);
2088 	mutex_init(&fs_info->qgroup_ioctl_lock);
2089 	fs_info->qgroup_tree = RB_ROOT;
2090 	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2091 	fs_info->qgroup_seq = 1;
2092 	fs_info->qgroup_ulist = NULL;
2093 	fs_info->qgroup_rescan_running = false;
2094 	mutex_init(&fs_info->qgroup_rescan_lock);
2095 }
2096 
2097 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2098 		struct btrfs_fs_devices *fs_devices)
2099 {
2100 	u32 max_active = fs_info->thread_pool_size;
2101 	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2102 
2103 	fs_info->workers =
2104 		btrfs_alloc_workqueue(fs_info, "worker",
2105 				      flags | WQ_HIGHPRI, max_active, 16);
2106 
2107 	fs_info->delalloc_workers =
2108 		btrfs_alloc_workqueue(fs_info, "delalloc",
2109 				      flags, max_active, 2);
2110 
2111 	fs_info->flush_workers =
2112 		btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2113 				      flags, max_active, 0);
2114 
2115 	fs_info->caching_workers =
2116 		btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2117 
2118 	fs_info->fixup_workers =
2119 		btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2120 
2121 	/*
2122 	 * endios are largely parallel and should have a very
2123 	 * low idle thresh
2124 	 */
2125 	fs_info->endio_workers =
2126 		btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2127 	fs_info->endio_meta_workers =
2128 		btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2129 				      max_active, 4);
2130 	fs_info->endio_meta_write_workers =
2131 		btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2132 				      max_active, 2);
2133 	fs_info->endio_raid56_workers =
2134 		btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2135 				      max_active, 4);
2136 	fs_info->endio_repair_workers =
2137 		btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2138 	fs_info->rmw_workers =
2139 		btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2140 	fs_info->endio_write_workers =
2141 		btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2142 				      max_active, 2);
2143 	fs_info->endio_freespace_worker =
2144 		btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2145 				      max_active, 0);
2146 	fs_info->delayed_workers =
2147 		btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2148 				      max_active, 0);
2149 	fs_info->readahead_workers =
2150 		btrfs_alloc_workqueue(fs_info, "readahead", flags,
2151 				      max_active, 2);
2152 	fs_info->qgroup_rescan_workers =
2153 		btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2154 	fs_info->discard_ctl.discard_workers =
2155 		alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2156 
2157 	if (!(fs_info->workers && fs_info->delalloc_workers &&
2158 	      fs_info->flush_workers &&
2159 	      fs_info->endio_workers && fs_info->endio_meta_workers &&
2160 	      fs_info->endio_meta_write_workers &&
2161 	      fs_info->endio_repair_workers &&
2162 	      fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2163 	      fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2164 	      fs_info->caching_workers && fs_info->readahead_workers &&
2165 	      fs_info->fixup_workers && fs_info->delayed_workers &&
2166 	      fs_info->qgroup_rescan_workers &&
2167 	      fs_info->discard_ctl.discard_workers)) {
2168 		return -ENOMEM;
2169 	}
2170 
2171 	return 0;
2172 }
2173 
2174 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2175 {
2176 	struct crypto_shash *csum_shash;
2177 	const char *csum_driver = btrfs_super_csum_driver(csum_type);
2178 
2179 	csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2180 
2181 	if (IS_ERR(csum_shash)) {
2182 		btrfs_err(fs_info, "error allocating %s hash for checksum",
2183 			  csum_driver);
2184 		return PTR_ERR(csum_shash);
2185 	}
2186 
2187 	fs_info->csum_shash = csum_shash;
2188 
2189 	return 0;
2190 }
2191 
2192 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
2193 {
2194 	crypto_free_shash(fs_info->csum_shash);
2195 }
2196 
2197 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2198 			    struct btrfs_fs_devices *fs_devices)
2199 {
2200 	int ret;
2201 	struct btrfs_root *log_tree_root;
2202 	struct btrfs_super_block *disk_super = fs_info->super_copy;
2203 	u64 bytenr = btrfs_super_log_root(disk_super);
2204 	int level = btrfs_super_log_root_level(disk_super);
2205 
2206 	if (fs_devices->rw_devices == 0) {
2207 		btrfs_warn(fs_info, "log replay required on RO media");
2208 		return -EIO;
2209 	}
2210 
2211 	log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2212 	if (!log_tree_root)
2213 		return -ENOMEM;
2214 
2215 	__setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2216 
2217 	log_tree_root->node = read_tree_block(fs_info, bytenr,
2218 					      fs_info->generation + 1,
2219 					      level, NULL);
2220 	if (IS_ERR(log_tree_root->node)) {
2221 		btrfs_warn(fs_info, "failed to read log tree");
2222 		ret = PTR_ERR(log_tree_root->node);
2223 		kfree(log_tree_root);
2224 		return ret;
2225 	} else if (!extent_buffer_uptodate(log_tree_root->node)) {
2226 		btrfs_err(fs_info, "failed to read log tree");
2227 		free_extent_buffer(log_tree_root->node);
2228 		kfree(log_tree_root);
2229 		return -EIO;
2230 	}
2231 	/* returns with log_tree_root freed on success */
2232 	ret = btrfs_recover_log_trees(log_tree_root);
2233 	if (ret) {
2234 		btrfs_handle_fs_error(fs_info, ret,
2235 				      "Failed to recover log tree");
2236 		free_extent_buffer(log_tree_root->node);
2237 		kfree(log_tree_root);
2238 		return ret;
2239 	}
2240 
2241 	if (sb_rdonly(fs_info->sb)) {
2242 		ret = btrfs_commit_super(fs_info);
2243 		if (ret)
2244 			return ret;
2245 	}
2246 
2247 	return 0;
2248 }
2249 
2250 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2251 {
2252 	struct btrfs_root *tree_root = fs_info->tree_root;
2253 	struct btrfs_root *root;
2254 	struct btrfs_key location;
2255 	int ret;
2256 
2257 	BUG_ON(!fs_info->tree_root);
2258 
2259 	location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2260 	location.type = BTRFS_ROOT_ITEM_KEY;
2261 	location.offset = 0;
2262 
2263 	root = btrfs_read_tree_root(tree_root, &location);
2264 	if (IS_ERR(root)) {
2265 		ret = PTR_ERR(root);
2266 		goto out;
2267 	}
2268 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2269 	fs_info->extent_root = root;
2270 
2271 	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2272 	root = btrfs_read_tree_root(tree_root, &location);
2273 	if (IS_ERR(root)) {
2274 		ret = PTR_ERR(root);
2275 		goto out;
2276 	}
2277 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2278 	fs_info->dev_root = root;
2279 	btrfs_init_devices_late(fs_info);
2280 
2281 	location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2282 	root = btrfs_read_tree_root(tree_root, &location);
2283 	if (IS_ERR(root)) {
2284 		ret = PTR_ERR(root);
2285 		goto out;
2286 	}
2287 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2288 	fs_info->csum_root = root;
2289 
2290 	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2291 	root = btrfs_read_tree_root(tree_root, &location);
2292 	if (!IS_ERR(root)) {
2293 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2294 		set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2295 		fs_info->quota_root = root;
2296 	}
2297 
2298 	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2299 	root = btrfs_read_tree_root(tree_root, &location);
2300 	if (IS_ERR(root)) {
2301 		ret = PTR_ERR(root);
2302 		if (ret != -ENOENT)
2303 			goto out;
2304 	} else {
2305 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2306 		fs_info->uuid_root = root;
2307 	}
2308 
2309 	if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2310 		location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2311 		root = btrfs_read_tree_root(tree_root, &location);
2312 		if (IS_ERR(root)) {
2313 			ret = PTR_ERR(root);
2314 			goto out;
2315 		}
2316 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2317 		fs_info->free_space_root = root;
2318 	}
2319 
2320 	return 0;
2321 out:
2322 	btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2323 		   location.objectid, ret);
2324 	return ret;
2325 }
2326 
2327 /*
2328  * Real super block validation
2329  * NOTE: super csum type and incompat features will not be checked here.
2330  *
2331  * @sb:		super block to check
2332  * @mirror_num:	the super block number to check its bytenr:
2333  * 		0	the primary (1st) sb
2334  * 		1, 2	2nd and 3rd backup copy
2335  * 	       -1	skip bytenr check
2336  */
2337 static int validate_super(struct btrfs_fs_info *fs_info,
2338 			    struct btrfs_super_block *sb, int mirror_num)
2339 {
2340 	u64 nodesize = btrfs_super_nodesize(sb);
2341 	u64 sectorsize = btrfs_super_sectorsize(sb);
2342 	int ret = 0;
2343 
2344 	if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2345 		btrfs_err(fs_info, "no valid FS found");
2346 		ret = -EINVAL;
2347 	}
2348 	if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2349 		btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2350 				btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2351 		ret = -EINVAL;
2352 	}
2353 	if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2354 		btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2355 				btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2356 		ret = -EINVAL;
2357 	}
2358 	if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2359 		btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2360 				btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2361 		ret = -EINVAL;
2362 	}
2363 	if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2364 		btrfs_err(fs_info, "log_root level too big: %d >= %d",
2365 				btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2366 		ret = -EINVAL;
2367 	}
2368 
2369 	/*
2370 	 * Check sectorsize and nodesize first, other check will need it.
2371 	 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2372 	 */
2373 	if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2374 	    sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2375 		btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2376 		ret = -EINVAL;
2377 	}
2378 	/* Only PAGE SIZE is supported yet */
2379 	if (sectorsize != PAGE_SIZE) {
2380 		btrfs_err(fs_info,
2381 			"sectorsize %llu not supported yet, only support %lu",
2382 			sectorsize, PAGE_SIZE);
2383 		ret = -EINVAL;
2384 	}
2385 	if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2386 	    nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2387 		btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2388 		ret = -EINVAL;
2389 	}
2390 	if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2391 		btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2392 			  le32_to_cpu(sb->__unused_leafsize), nodesize);
2393 		ret = -EINVAL;
2394 	}
2395 
2396 	/* Root alignment check */
2397 	if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2398 		btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2399 			   btrfs_super_root(sb));
2400 		ret = -EINVAL;
2401 	}
2402 	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2403 		btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2404 			   btrfs_super_chunk_root(sb));
2405 		ret = -EINVAL;
2406 	}
2407 	if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2408 		btrfs_warn(fs_info, "log_root block unaligned: %llu",
2409 			   btrfs_super_log_root(sb));
2410 		ret = -EINVAL;
2411 	}
2412 
2413 	if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2414 		   BTRFS_FSID_SIZE) != 0) {
2415 		btrfs_err(fs_info,
2416 			"dev_item UUID does not match metadata fsid: %pU != %pU",
2417 			fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2418 		ret = -EINVAL;
2419 	}
2420 
2421 	/*
2422 	 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2423 	 * done later
2424 	 */
2425 	if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2426 		btrfs_err(fs_info, "bytes_used is too small %llu",
2427 			  btrfs_super_bytes_used(sb));
2428 		ret = -EINVAL;
2429 	}
2430 	if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2431 		btrfs_err(fs_info, "invalid stripesize %u",
2432 			  btrfs_super_stripesize(sb));
2433 		ret = -EINVAL;
2434 	}
2435 	if (btrfs_super_num_devices(sb) > (1UL << 31))
2436 		btrfs_warn(fs_info, "suspicious number of devices: %llu",
2437 			   btrfs_super_num_devices(sb));
2438 	if (btrfs_super_num_devices(sb) == 0) {
2439 		btrfs_err(fs_info, "number of devices is 0");
2440 		ret = -EINVAL;
2441 	}
2442 
2443 	if (mirror_num >= 0 &&
2444 	    btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2445 		btrfs_err(fs_info, "super offset mismatch %llu != %u",
2446 			  btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2447 		ret = -EINVAL;
2448 	}
2449 
2450 	/*
2451 	 * Obvious sys_chunk_array corruptions, it must hold at least one key
2452 	 * and one chunk
2453 	 */
2454 	if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2455 		btrfs_err(fs_info, "system chunk array too big %u > %u",
2456 			  btrfs_super_sys_array_size(sb),
2457 			  BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2458 		ret = -EINVAL;
2459 	}
2460 	if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2461 			+ sizeof(struct btrfs_chunk)) {
2462 		btrfs_err(fs_info, "system chunk array too small %u < %zu",
2463 			  btrfs_super_sys_array_size(sb),
2464 			  sizeof(struct btrfs_disk_key)
2465 			  + sizeof(struct btrfs_chunk));
2466 		ret = -EINVAL;
2467 	}
2468 
2469 	/*
2470 	 * The generation is a global counter, we'll trust it more than the others
2471 	 * but it's still possible that it's the one that's wrong.
2472 	 */
2473 	if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2474 		btrfs_warn(fs_info,
2475 			"suspicious: generation < chunk_root_generation: %llu < %llu",
2476 			btrfs_super_generation(sb),
2477 			btrfs_super_chunk_root_generation(sb));
2478 	if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2479 	    && btrfs_super_cache_generation(sb) != (u64)-1)
2480 		btrfs_warn(fs_info,
2481 			"suspicious: generation < cache_generation: %llu < %llu",
2482 			btrfs_super_generation(sb),
2483 			btrfs_super_cache_generation(sb));
2484 
2485 	return ret;
2486 }
2487 
2488 /*
2489  * Validation of super block at mount time.
2490  * Some checks already done early at mount time, like csum type and incompat
2491  * flags will be skipped.
2492  */
2493 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2494 {
2495 	return validate_super(fs_info, fs_info->super_copy, 0);
2496 }
2497 
2498 /*
2499  * Validation of super block at write time.
2500  * Some checks like bytenr check will be skipped as their values will be
2501  * overwritten soon.
2502  * Extra checks like csum type and incompat flags will be done here.
2503  */
2504 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2505 				      struct btrfs_super_block *sb)
2506 {
2507 	int ret;
2508 
2509 	ret = validate_super(fs_info, sb, -1);
2510 	if (ret < 0)
2511 		goto out;
2512 	if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2513 		ret = -EUCLEAN;
2514 		btrfs_err(fs_info, "invalid csum type, has %u want %u",
2515 			  btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2516 		goto out;
2517 	}
2518 	if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2519 		ret = -EUCLEAN;
2520 		btrfs_err(fs_info,
2521 		"invalid incompat flags, has 0x%llx valid mask 0x%llx",
2522 			  btrfs_super_incompat_flags(sb),
2523 			  (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2524 		goto out;
2525 	}
2526 out:
2527 	if (ret < 0)
2528 		btrfs_err(fs_info,
2529 		"super block corruption detected before writing it to disk");
2530 	return ret;
2531 }
2532 
2533 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2534 {
2535 	int backup_index = find_newest_super_backup(fs_info);
2536 	struct btrfs_super_block *sb = fs_info->super_copy;
2537 	struct btrfs_root *tree_root = fs_info->tree_root;
2538 	bool handle_error = false;
2539 	int ret = 0;
2540 	int i;
2541 
2542 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2543 		u64 generation;
2544 		int level;
2545 
2546 		if (handle_error) {
2547 			if (!IS_ERR(tree_root->node))
2548 				free_extent_buffer(tree_root->node);
2549 			tree_root->node = NULL;
2550 
2551 			if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2552 				break;
2553 
2554 			free_root_pointers(fs_info, 0);
2555 
2556 			/*
2557 			 * Don't use the log in recovery mode, it won't be
2558 			 * valid
2559 			 */
2560 			btrfs_set_super_log_root(sb, 0);
2561 
2562 			/* We can't trust the free space cache either */
2563 			btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2564 
2565 			ret = read_backup_root(fs_info, i);
2566 			backup_index = ret;
2567 			if (ret < 0)
2568 				return ret;
2569 		}
2570 		generation = btrfs_super_generation(sb);
2571 		level = btrfs_super_root_level(sb);
2572 		tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2573 						  generation, level, NULL);
2574 		if (IS_ERR(tree_root->node) ||
2575 		    !extent_buffer_uptodate(tree_root->node)) {
2576 			handle_error = true;
2577 
2578 			if (IS_ERR(tree_root->node))
2579 				ret = PTR_ERR(tree_root->node);
2580 			else if (!extent_buffer_uptodate(tree_root->node))
2581 				ret = -EUCLEAN;
2582 
2583 			btrfs_warn(fs_info, "failed to read tree root");
2584 			continue;
2585 		}
2586 
2587 		btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2588 		tree_root->commit_root = btrfs_root_node(tree_root);
2589 		btrfs_set_root_refs(&tree_root->root_item, 1);
2590 
2591 		/*
2592 		 * No need to hold btrfs_root::objectid_mutex since the fs
2593 		 * hasn't been fully initialised and we are the only user
2594 		 */
2595 		ret = btrfs_find_highest_objectid(tree_root,
2596 						&tree_root->highest_objectid);
2597 		if (ret < 0) {
2598 			handle_error = true;
2599 			continue;
2600 		}
2601 
2602 		ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2603 
2604 		ret = btrfs_read_roots(fs_info);
2605 		if (ret < 0) {
2606 			handle_error = true;
2607 			continue;
2608 		}
2609 
2610 		/* All successful */
2611 		fs_info->generation = generation;
2612 		fs_info->last_trans_committed = generation;
2613 
2614 		/* Always begin writing backup roots after the one being used */
2615 		if (backup_index < 0) {
2616 			fs_info->backup_root_index = 0;
2617 		} else {
2618 			fs_info->backup_root_index = backup_index + 1;
2619 			fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2620 		}
2621 		break;
2622 	}
2623 
2624 	return ret;
2625 }
2626 
2627 int __cold open_ctree(struct super_block *sb,
2628 	       struct btrfs_fs_devices *fs_devices,
2629 	       char *options)
2630 {
2631 	u32 sectorsize;
2632 	u32 nodesize;
2633 	u32 stripesize;
2634 	u64 generation;
2635 	u64 features;
2636 	u16 csum_type;
2637 	struct btrfs_key location;
2638 	struct buffer_head *bh;
2639 	struct btrfs_super_block *disk_super;
2640 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2641 	struct btrfs_root *tree_root;
2642 	struct btrfs_root *chunk_root;
2643 	int ret;
2644 	int err = -EINVAL;
2645 	int clear_free_space_tree = 0;
2646 	int level;
2647 
2648 	tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2649 	chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2650 	if (!tree_root || !chunk_root) {
2651 		err = -ENOMEM;
2652 		goto fail;
2653 	}
2654 
2655 	ret = init_srcu_struct(&fs_info->subvol_srcu);
2656 	if (ret) {
2657 		err = ret;
2658 		goto fail;
2659 	}
2660 
2661 	ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2662 	if (ret) {
2663 		err = ret;
2664 		goto fail_srcu;
2665 	}
2666 
2667 	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2668 	if (ret) {
2669 		err = ret;
2670 		goto fail_dio_bytes;
2671 	}
2672 	fs_info->dirty_metadata_batch = PAGE_SIZE *
2673 					(1 + ilog2(nr_cpu_ids));
2674 
2675 	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2676 	if (ret) {
2677 		err = ret;
2678 		goto fail_dirty_metadata_bytes;
2679 	}
2680 
2681 	ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2682 			GFP_KERNEL);
2683 	if (ret) {
2684 		err = ret;
2685 		goto fail_delalloc_bytes;
2686 	}
2687 
2688 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2689 	INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2690 	INIT_LIST_HEAD(&fs_info->trans_list);
2691 	INIT_LIST_HEAD(&fs_info->dead_roots);
2692 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
2693 	INIT_LIST_HEAD(&fs_info->delalloc_roots);
2694 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
2695 	spin_lock_init(&fs_info->delalloc_root_lock);
2696 	spin_lock_init(&fs_info->trans_lock);
2697 	spin_lock_init(&fs_info->fs_roots_radix_lock);
2698 	spin_lock_init(&fs_info->delayed_iput_lock);
2699 	spin_lock_init(&fs_info->defrag_inodes_lock);
2700 	spin_lock_init(&fs_info->super_lock);
2701 	spin_lock_init(&fs_info->buffer_lock);
2702 	spin_lock_init(&fs_info->unused_bgs_lock);
2703 	rwlock_init(&fs_info->tree_mod_log_lock);
2704 	mutex_init(&fs_info->unused_bg_unpin_mutex);
2705 	mutex_init(&fs_info->delete_unused_bgs_mutex);
2706 	mutex_init(&fs_info->reloc_mutex);
2707 	mutex_init(&fs_info->delalloc_root_mutex);
2708 	seqlock_init(&fs_info->profiles_lock);
2709 
2710 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2711 	INIT_LIST_HEAD(&fs_info->space_info);
2712 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2713 	INIT_LIST_HEAD(&fs_info->unused_bgs);
2714 	extent_map_tree_init(&fs_info->mapping_tree);
2715 	btrfs_init_block_rsv(&fs_info->global_block_rsv,
2716 			     BTRFS_BLOCK_RSV_GLOBAL);
2717 	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2718 	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2719 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2720 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2721 			     BTRFS_BLOCK_RSV_DELOPS);
2722 	btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2723 			     BTRFS_BLOCK_RSV_DELREFS);
2724 
2725 	atomic_set(&fs_info->async_delalloc_pages, 0);
2726 	atomic_set(&fs_info->defrag_running, 0);
2727 	atomic_set(&fs_info->reada_works_cnt, 0);
2728 	atomic_set(&fs_info->nr_delayed_iputs, 0);
2729 	atomic64_set(&fs_info->tree_mod_seq, 0);
2730 	fs_info->sb = sb;
2731 	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2732 	fs_info->metadata_ratio = 0;
2733 	fs_info->defrag_inodes = RB_ROOT;
2734 	atomic64_set(&fs_info->free_chunk_space, 0);
2735 	fs_info->tree_mod_log = RB_ROOT;
2736 	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2737 	fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2738 	/* readahead state */
2739 	INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2740 	spin_lock_init(&fs_info->reada_lock);
2741 	btrfs_init_ref_verify(fs_info);
2742 
2743 	fs_info->thread_pool_size = min_t(unsigned long,
2744 					  num_online_cpus() + 2, 8);
2745 
2746 	INIT_LIST_HEAD(&fs_info->ordered_roots);
2747 	spin_lock_init(&fs_info->ordered_root_lock);
2748 
2749 	fs_info->btree_inode = new_inode(sb);
2750 	if (!fs_info->btree_inode) {
2751 		err = -ENOMEM;
2752 		goto fail_bio_counter;
2753 	}
2754 	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2755 
2756 	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2757 					GFP_KERNEL);
2758 	if (!fs_info->delayed_root) {
2759 		err = -ENOMEM;
2760 		goto fail_iput;
2761 	}
2762 	btrfs_init_delayed_root(fs_info->delayed_root);
2763 
2764 	btrfs_init_scrub(fs_info);
2765 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2766 	fs_info->check_integrity_print_mask = 0;
2767 #endif
2768 	btrfs_init_balance(fs_info);
2769 	btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2770 
2771 	sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2772 	sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2773 
2774 	btrfs_init_btree_inode(fs_info);
2775 
2776 	spin_lock_init(&fs_info->block_group_cache_lock);
2777 	fs_info->block_group_cache_tree = RB_ROOT;
2778 	fs_info->first_logical_byte = (u64)-1;
2779 
2780 	extent_io_tree_init(fs_info, &fs_info->freed_extents[0],
2781 			    IO_TREE_FS_INFO_FREED_EXTENTS0, NULL);
2782 	extent_io_tree_init(fs_info, &fs_info->freed_extents[1],
2783 			    IO_TREE_FS_INFO_FREED_EXTENTS1, NULL);
2784 	fs_info->pinned_extents = &fs_info->freed_extents[0];
2785 	set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2786 
2787 	mutex_init(&fs_info->ordered_operations_mutex);
2788 	mutex_init(&fs_info->tree_log_mutex);
2789 	mutex_init(&fs_info->chunk_mutex);
2790 	mutex_init(&fs_info->transaction_kthread_mutex);
2791 	mutex_init(&fs_info->cleaner_mutex);
2792 	mutex_init(&fs_info->ro_block_group_mutex);
2793 	init_rwsem(&fs_info->commit_root_sem);
2794 	init_rwsem(&fs_info->cleanup_work_sem);
2795 	init_rwsem(&fs_info->subvol_sem);
2796 	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2797 
2798 	btrfs_init_dev_replace_locks(fs_info);
2799 	btrfs_init_qgroup(fs_info);
2800 	btrfs_discard_init(fs_info);
2801 
2802 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2803 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2804 
2805 	init_waitqueue_head(&fs_info->transaction_throttle);
2806 	init_waitqueue_head(&fs_info->transaction_wait);
2807 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2808 	init_waitqueue_head(&fs_info->async_submit_wait);
2809 	init_waitqueue_head(&fs_info->delayed_iputs_wait);
2810 
2811 	/* Usable values until the real ones are cached from the superblock */
2812 	fs_info->nodesize = 4096;
2813 	fs_info->sectorsize = 4096;
2814 	fs_info->stripesize = 4096;
2815 
2816 	spin_lock_init(&fs_info->swapfile_pins_lock);
2817 	fs_info->swapfile_pins = RB_ROOT;
2818 
2819 	fs_info->send_in_progress = 0;
2820 
2821 	ret = btrfs_alloc_stripe_hash_table(fs_info);
2822 	if (ret) {
2823 		err = ret;
2824 		goto fail_alloc;
2825 	}
2826 
2827 	__setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2828 
2829 	invalidate_bdev(fs_devices->latest_bdev);
2830 
2831 	/*
2832 	 * Read super block and check the signature bytes only
2833 	 */
2834 	bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2835 	if (IS_ERR(bh)) {
2836 		err = PTR_ERR(bh);
2837 		goto fail_alloc;
2838 	}
2839 
2840 	/*
2841 	 * Verify the type first, if that or the the checksum value are
2842 	 * corrupted, we'll find out
2843 	 */
2844 	csum_type = btrfs_super_csum_type((struct btrfs_super_block *)bh->b_data);
2845 	if (!btrfs_supported_super_csum(csum_type)) {
2846 		btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2847 			  csum_type);
2848 		err = -EINVAL;
2849 		brelse(bh);
2850 		goto fail_alloc;
2851 	}
2852 
2853 	ret = btrfs_init_csum_hash(fs_info, csum_type);
2854 	if (ret) {
2855 		err = ret;
2856 		goto fail_alloc;
2857 	}
2858 
2859 	/*
2860 	 * We want to check superblock checksum, the type is stored inside.
2861 	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2862 	 */
2863 	if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2864 		btrfs_err(fs_info, "superblock checksum mismatch");
2865 		err = -EINVAL;
2866 		brelse(bh);
2867 		goto fail_csum;
2868 	}
2869 
2870 	/*
2871 	 * super_copy is zeroed at allocation time and we never touch the
2872 	 * following bytes up to INFO_SIZE, the checksum is calculated from
2873 	 * the whole block of INFO_SIZE
2874 	 */
2875 	memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2876 	brelse(bh);
2877 
2878 	disk_super = fs_info->super_copy;
2879 
2880 	ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2881 		       BTRFS_FSID_SIZE));
2882 
2883 	if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2884 		ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2885 				fs_info->super_copy->metadata_uuid,
2886 				BTRFS_FSID_SIZE));
2887 	}
2888 
2889 	features = btrfs_super_flags(disk_super);
2890 	if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2891 		features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2892 		btrfs_set_super_flags(disk_super, features);
2893 		btrfs_info(fs_info,
2894 			"found metadata UUID change in progress flag, clearing");
2895 	}
2896 
2897 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
2898 	       sizeof(*fs_info->super_for_commit));
2899 
2900 	ret = btrfs_validate_mount_super(fs_info);
2901 	if (ret) {
2902 		btrfs_err(fs_info, "superblock contains fatal errors");
2903 		err = -EINVAL;
2904 		goto fail_csum;
2905 	}
2906 
2907 	if (!btrfs_super_root(disk_super))
2908 		goto fail_csum;
2909 
2910 	/* check FS state, whether FS is broken. */
2911 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2912 		set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2913 
2914 	/*
2915 	 * In the long term, we'll store the compression type in the super
2916 	 * block, and it'll be used for per file compression control.
2917 	 */
2918 	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2919 
2920 	ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2921 	if (ret) {
2922 		err = ret;
2923 		goto fail_csum;
2924 	}
2925 
2926 	features = btrfs_super_incompat_flags(disk_super) &
2927 		~BTRFS_FEATURE_INCOMPAT_SUPP;
2928 	if (features) {
2929 		btrfs_err(fs_info,
2930 		    "cannot mount because of unsupported optional features (%llx)",
2931 		    features);
2932 		err = -EINVAL;
2933 		goto fail_csum;
2934 	}
2935 
2936 	features = btrfs_super_incompat_flags(disk_super);
2937 	features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2938 	if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2939 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2940 	else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2941 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2942 
2943 	if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2944 		btrfs_info(fs_info, "has skinny extents");
2945 
2946 	/*
2947 	 * flag our filesystem as having big metadata blocks if
2948 	 * they are bigger than the page size
2949 	 */
2950 	if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2951 		if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2952 			btrfs_info(fs_info,
2953 				"flagging fs with big metadata feature");
2954 		features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2955 	}
2956 
2957 	nodesize = btrfs_super_nodesize(disk_super);
2958 	sectorsize = btrfs_super_sectorsize(disk_super);
2959 	stripesize = sectorsize;
2960 	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2961 	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2962 
2963 	/* Cache block sizes */
2964 	fs_info->nodesize = nodesize;
2965 	fs_info->sectorsize = sectorsize;
2966 	fs_info->stripesize = stripesize;
2967 
2968 	/*
2969 	 * mixed block groups end up with duplicate but slightly offset
2970 	 * extent buffers for the same range.  It leads to corruptions
2971 	 */
2972 	if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2973 	    (sectorsize != nodesize)) {
2974 		btrfs_err(fs_info,
2975 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2976 			nodesize, sectorsize);
2977 		goto fail_csum;
2978 	}
2979 
2980 	/*
2981 	 * Needn't use the lock because there is no other task which will
2982 	 * update the flag.
2983 	 */
2984 	btrfs_set_super_incompat_flags(disk_super, features);
2985 
2986 	features = btrfs_super_compat_ro_flags(disk_super) &
2987 		~BTRFS_FEATURE_COMPAT_RO_SUPP;
2988 	if (!sb_rdonly(sb) && features) {
2989 		btrfs_err(fs_info,
2990 	"cannot mount read-write because of unsupported optional features (%llx)",
2991 		       features);
2992 		err = -EINVAL;
2993 		goto fail_csum;
2994 	}
2995 
2996 	ret = btrfs_init_workqueues(fs_info, fs_devices);
2997 	if (ret) {
2998 		err = ret;
2999 		goto fail_sb_buffer;
3000 	}
3001 
3002 	sb->s_bdi->congested_fn = btrfs_congested_fn;
3003 	sb->s_bdi->congested_data = fs_info;
3004 	sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
3005 	sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
3006 	sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3007 	sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3008 
3009 	sb->s_blocksize = sectorsize;
3010 	sb->s_blocksize_bits = blksize_bits(sectorsize);
3011 	memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3012 
3013 	mutex_lock(&fs_info->chunk_mutex);
3014 	ret = btrfs_read_sys_array(fs_info);
3015 	mutex_unlock(&fs_info->chunk_mutex);
3016 	if (ret) {
3017 		btrfs_err(fs_info, "failed to read the system array: %d", ret);
3018 		goto fail_sb_buffer;
3019 	}
3020 
3021 	generation = btrfs_super_chunk_root_generation(disk_super);
3022 	level = btrfs_super_chunk_root_level(disk_super);
3023 
3024 	__setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
3025 
3026 	chunk_root->node = read_tree_block(fs_info,
3027 					   btrfs_super_chunk_root(disk_super),
3028 					   generation, level, NULL);
3029 	if (IS_ERR(chunk_root->node) ||
3030 	    !extent_buffer_uptodate(chunk_root->node)) {
3031 		btrfs_err(fs_info, "failed to read chunk root");
3032 		if (!IS_ERR(chunk_root->node))
3033 			free_extent_buffer(chunk_root->node);
3034 		chunk_root->node = NULL;
3035 		goto fail_tree_roots;
3036 	}
3037 	btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3038 	chunk_root->commit_root = btrfs_root_node(chunk_root);
3039 
3040 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3041 	   btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
3042 
3043 	ret = btrfs_read_chunk_tree(fs_info);
3044 	if (ret) {
3045 		btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3046 		goto fail_tree_roots;
3047 	}
3048 
3049 	/*
3050 	 * Keep the devid that is marked to be the target device for the
3051 	 * device replace procedure
3052 	 */
3053 	btrfs_free_extra_devids(fs_devices, 0);
3054 
3055 	if (!fs_devices->latest_bdev) {
3056 		btrfs_err(fs_info, "failed to read devices");
3057 		goto fail_tree_roots;
3058 	}
3059 
3060 	ret = init_tree_roots(fs_info);
3061 	if (ret)
3062 		goto fail_tree_roots;
3063 
3064 	ret = btrfs_verify_dev_extents(fs_info);
3065 	if (ret) {
3066 		btrfs_err(fs_info,
3067 			  "failed to verify dev extents against chunks: %d",
3068 			  ret);
3069 		goto fail_block_groups;
3070 	}
3071 	ret = btrfs_recover_balance(fs_info);
3072 	if (ret) {
3073 		btrfs_err(fs_info, "failed to recover balance: %d", ret);
3074 		goto fail_block_groups;
3075 	}
3076 
3077 	ret = btrfs_init_dev_stats(fs_info);
3078 	if (ret) {
3079 		btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3080 		goto fail_block_groups;
3081 	}
3082 
3083 	ret = btrfs_init_dev_replace(fs_info);
3084 	if (ret) {
3085 		btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3086 		goto fail_block_groups;
3087 	}
3088 
3089 	btrfs_free_extra_devids(fs_devices, 1);
3090 
3091 	ret = btrfs_sysfs_add_fsid(fs_devices);
3092 	if (ret) {
3093 		btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3094 				ret);
3095 		goto fail_block_groups;
3096 	}
3097 
3098 	ret = btrfs_sysfs_add_mounted(fs_info);
3099 	if (ret) {
3100 		btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3101 		goto fail_fsdev_sysfs;
3102 	}
3103 
3104 	ret = btrfs_init_space_info(fs_info);
3105 	if (ret) {
3106 		btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3107 		goto fail_sysfs;
3108 	}
3109 
3110 	ret = btrfs_read_block_groups(fs_info);
3111 	if (ret) {
3112 		btrfs_err(fs_info, "failed to read block groups: %d", ret);
3113 		goto fail_sysfs;
3114 	}
3115 
3116 	if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3117 		btrfs_warn(fs_info,
3118 		"writable mount is not allowed due to too many missing devices");
3119 		goto fail_sysfs;
3120 	}
3121 
3122 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3123 					       "btrfs-cleaner");
3124 	if (IS_ERR(fs_info->cleaner_kthread))
3125 		goto fail_sysfs;
3126 
3127 	fs_info->transaction_kthread = kthread_run(transaction_kthread,
3128 						   tree_root,
3129 						   "btrfs-transaction");
3130 	if (IS_ERR(fs_info->transaction_kthread))
3131 		goto fail_cleaner;
3132 
3133 	if (!btrfs_test_opt(fs_info, NOSSD) &&
3134 	    !fs_info->fs_devices->rotating) {
3135 		btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3136 	}
3137 
3138 	/*
3139 	 * Mount does not set all options immediately, we can do it now and do
3140 	 * not have to wait for transaction commit
3141 	 */
3142 	btrfs_apply_pending_changes(fs_info);
3143 
3144 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3145 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3146 		ret = btrfsic_mount(fs_info, fs_devices,
3147 				    btrfs_test_opt(fs_info,
3148 					CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3149 				    1 : 0,
3150 				    fs_info->check_integrity_print_mask);
3151 		if (ret)
3152 			btrfs_warn(fs_info,
3153 				"failed to initialize integrity check module: %d",
3154 				ret);
3155 	}
3156 #endif
3157 	ret = btrfs_read_qgroup_config(fs_info);
3158 	if (ret)
3159 		goto fail_trans_kthread;
3160 
3161 	if (btrfs_build_ref_tree(fs_info))
3162 		btrfs_err(fs_info, "couldn't build ref tree");
3163 
3164 	/* do not make disk changes in broken FS or nologreplay is given */
3165 	if (btrfs_super_log_root(disk_super) != 0 &&
3166 	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3167 		btrfs_info(fs_info, "start tree-log replay");
3168 		ret = btrfs_replay_log(fs_info, fs_devices);
3169 		if (ret) {
3170 			err = ret;
3171 			goto fail_qgroup;
3172 		}
3173 	}
3174 
3175 	ret = btrfs_find_orphan_roots(fs_info);
3176 	if (ret)
3177 		goto fail_qgroup;
3178 
3179 	if (!sb_rdonly(sb)) {
3180 		ret = btrfs_cleanup_fs_roots(fs_info);
3181 		if (ret)
3182 			goto fail_qgroup;
3183 
3184 		mutex_lock(&fs_info->cleaner_mutex);
3185 		ret = btrfs_recover_relocation(tree_root);
3186 		mutex_unlock(&fs_info->cleaner_mutex);
3187 		if (ret < 0) {
3188 			btrfs_warn(fs_info, "failed to recover relocation: %d",
3189 					ret);
3190 			err = -EINVAL;
3191 			goto fail_qgroup;
3192 		}
3193 	}
3194 
3195 	location.objectid = BTRFS_FS_TREE_OBJECTID;
3196 	location.type = BTRFS_ROOT_ITEM_KEY;
3197 	location.offset = 0;
3198 
3199 	fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3200 	if (IS_ERR(fs_info->fs_root)) {
3201 		err = PTR_ERR(fs_info->fs_root);
3202 		btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3203 		goto fail_qgroup;
3204 	}
3205 
3206 	if (sb_rdonly(sb))
3207 		return 0;
3208 
3209 	if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3210 	    btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3211 		clear_free_space_tree = 1;
3212 	} else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3213 		   !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3214 		btrfs_warn(fs_info, "free space tree is invalid");
3215 		clear_free_space_tree = 1;
3216 	}
3217 
3218 	if (clear_free_space_tree) {
3219 		btrfs_info(fs_info, "clearing free space tree");
3220 		ret = btrfs_clear_free_space_tree(fs_info);
3221 		if (ret) {
3222 			btrfs_warn(fs_info,
3223 				   "failed to clear free space tree: %d", ret);
3224 			close_ctree(fs_info);
3225 			return ret;
3226 		}
3227 	}
3228 
3229 	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3230 	    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3231 		btrfs_info(fs_info, "creating free space tree");
3232 		ret = btrfs_create_free_space_tree(fs_info);
3233 		if (ret) {
3234 			btrfs_warn(fs_info,
3235 				"failed to create free space tree: %d", ret);
3236 			close_ctree(fs_info);
3237 			return ret;
3238 		}
3239 	}
3240 
3241 	down_read(&fs_info->cleanup_work_sem);
3242 	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3243 	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3244 		up_read(&fs_info->cleanup_work_sem);
3245 		close_ctree(fs_info);
3246 		return ret;
3247 	}
3248 	up_read(&fs_info->cleanup_work_sem);
3249 
3250 	ret = btrfs_resume_balance_async(fs_info);
3251 	if (ret) {
3252 		btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3253 		close_ctree(fs_info);
3254 		return ret;
3255 	}
3256 
3257 	ret = btrfs_resume_dev_replace_async(fs_info);
3258 	if (ret) {
3259 		btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3260 		close_ctree(fs_info);
3261 		return ret;
3262 	}
3263 
3264 	btrfs_qgroup_rescan_resume(fs_info);
3265 	btrfs_discard_resume(fs_info);
3266 
3267 	if (!fs_info->uuid_root) {
3268 		btrfs_info(fs_info, "creating UUID tree");
3269 		ret = btrfs_create_uuid_tree(fs_info);
3270 		if (ret) {
3271 			btrfs_warn(fs_info,
3272 				"failed to create the UUID tree: %d", ret);
3273 			close_ctree(fs_info);
3274 			return ret;
3275 		}
3276 	} else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3277 		   fs_info->generation !=
3278 				btrfs_super_uuid_tree_generation(disk_super)) {
3279 		btrfs_info(fs_info, "checking UUID tree");
3280 		ret = btrfs_check_uuid_tree(fs_info);
3281 		if (ret) {
3282 			btrfs_warn(fs_info,
3283 				"failed to check the UUID tree: %d", ret);
3284 			close_ctree(fs_info);
3285 			return ret;
3286 		}
3287 	} else {
3288 		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3289 	}
3290 	set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3291 
3292 	/*
3293 	 * backuproot only affect mount behavior, and if open_ctree succeeded,
3294 	 * no need to keep the flag
3295 	 */
3296 	btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3297 
3298 	return 0;
3299 
3300 fail_qgroup:
3301 	btrfs_free_qgroup_config(fs_info);
3302 fail_trans_kthread:
3303 	kthread_stop(fs_info->transaction_kthread);
3304 	btrfs_cleanup_transaction(fs_info);
3305 	btrfs_free_fs_roots(fs_info);
3306 fail_cleaner:
3307 	kthread_stop(fs_info->cleaner_kthread);
3308 
3309 	/*
3310 	 * make sure we're done with the btree inode before we stop our
3311 	 * kthreads
3312 	 */
3313 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3314 
3315 fail_sysfs:
3316 	btrfs_sysfs_remove_mounted(fs_info);
3317 
3318 fail_fsdev_sysfs:
3319 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3320 
3321 fail_block_groups:
3322 	btrfs_put_block_group_cache(fs_info);
3323 
3324 fail_tree_roots:
3325 	free_root_pointers(fs_info, true);
3326 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3327 
3328 fail_sb_buffer:
3329 	btrfs_stop_all_workers(fs_info);
3330 	btrfs_free_block_groups(fs_info);
3331 fail_csum:
3332 	btrfs_free_csum_hash(fs_info);
3333 fail_alloc:
3334 fail_iput:
3335 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3336 
3337 	iput(fs_info->btree_inode);
3338 fail_bio_counter:
3339 	percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3340 fail_delalloc_bytes:
3341 	percpu_counter_destroy(&fs_info->delalloc_bytes);
3342 fail_dirty_metadata_bytes:
3343 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3344 fail_dio_bytes:
3345 	percpu_counter_destroy(&fs_info->dio_bytes);
3346 fail_srcu:
3347 	cleanup_srcu_struct(&fs_info->subvol_srcu);
3348 fail:
3349 	btrfs_free_stripe_hash_table(fs_info);
3350 	btrfs_close_devices(fs_info->fs_devices);
3351 	return err;
3352 }
3353 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3354 
3355 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3356 {
3357 	if (uptodate) {
3358 		set_buffer_uptodate(bh);
3359 	} else {
3360 		struct btrfs_device *device = (struct btrfs_device *)
3361 			bh->b_private;
3362 
3363 		btrfs_warn_rl_in_rcu(device->fs_info,
3364 				"lost page write due to IO error on %s",
3365 					  rcu_str_deref(device->name));
3366 		/* note, we don't set_buffer_write_io_error because we have
3367 		 * our own ways of dealing with the IO errors
3368 		 */
3369 		clear_buffer_uptodate(bh);
3370 		btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3371 	}
3372 	unlock_buffer(bh);
3373 	put_bh(bh);
3374 }
3375 
3376 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3377 			struct buffer_head **bh_ret)
3378 {
3379 	struct buffer_head *bh;
3380 	struct btrfs_super_block *super;
3381 	u64 bytenr;
3382 
3383 	bytenr = btrfs_sb_offset(copy_num);
3384 	if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3385 		return -EINVAL;
3386 
3387 	bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3388 	/*
3389 	 * If we fail to read from the underlying devices, as of now
3390 	 * the best option we have is to mark it EIO.
3391 	 */
3392 	if (!bh)
3393 		return -EIO;
3394 
3395 	super = (struct btrfs_super_block *)bh->b_data;
3396 	if (btrfs_super_bytenr(super) != bytenr ||
3397 		    btrfs_super_magic(super) != BTRFS_MAGIC) {
3398 		brelse(bh);
3399 		return -EINVAL;
3400 	}
3401 
3402 	*bh_ret = bh;
3403 	return 0;
3404 }
3405 
3406 
3407 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3408 {
3409 	struct buffer_head *bh;
3410 	struct buffer_head *latest = NULL;
3411 	struct btrfs_super_block *super;
3412 	int i;
3413 	u64 transid = 0;
3414 	int ret = -EINVAL;
3415 
3416 	/* we would like to check all the supers, but that would make
3417 	 * a btrfs mount succeed after a mkfs from a different FS.
3418 	 * So, we need to add a special mount option to scan for
3419 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3420 	 */
3421 	for (i = 0; i < 1; i++) {
3422 		ret = btrfs_read_dev_one_super(bdev, i, &bh);
3423 		if (ret)
3424 			continue;
3425 
3426 		super = (struct btrfs_super_block *)bh->b_data;
3427 
3428 		if (!latest || btrfs_super_generation(super) > transid) {
3429 			brelse(latest);
3430 			latest = bh;
3431 			transid = btrfs_super_generation(super);
3432 		} else {
3433 			brelse(bh);
3434 		}
3435 	}
3436 
3437 	if (!latest)
3438 		return ERR_PTR(ret);
3439 
3440 	return latest;
3441 }
3442 
3443 /*
3444  * Write superblock @sb to the @device. Do not wait for completion, all the
3445  * buffer heads we write are pinned.
3446  *
3447  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3448  * the expected device size at commit time. Note that max_mirrors must be
3449  * same for write and wait phases.
3450  *
3451  * Return number of errors when buffer head is not found or submission fails.
3452  */
3453 static int write_dev_supers(struct btrfs_device *device,
3454 			    struct btrfs_super_block *sb, int max_mirrors)
3455 {
3456 	struct btrfs_fs_info *fs_info = device->fs_info;
3457 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3458 	struct buffer_head *bh;
3459 	int i;
3460 	int ret;
3461 	int errors = 0;
3462 	u64 bytenr;
3463 	int op_flags;
3464 
3465 	if (max_mirrors == 0)
3466 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3467 
3468 	shash->tfm = fs_info->csum_shash;
3469 
3470 	for (i = 0; i < max_mirrors; i++) {
3471 		bytenr = btrfs_sb_offset(i);
3472 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3473 		    device->commit_total_bytes)
3474 			break;
3475 
3476 		btrfs_set_super_bytenr(sb, bytenr);
3477 
3478 		crypto_shash_init(shash);
3479 		crypto_shash_update(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3480 				    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3481 		crypto_shash_final(shash, sb->csum);
3482 
3483 		/* One reference for us, and we leave it for the caller */
3484 		bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3485 			      BTRFS_SUPER_INFO_SIZE);
3486 		if (!bh) {
3487 			btrfs_err(device->fs_info,
3488 			    "couldn't get super buffer head for bytenr %llu",
3489 			    bytenr);
3490 			errors++;
3491 			continue;
3492 		}
3493 
3494 		memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3495 
3496 		/* one reference for submit_bh */
3497 		get_bh(bh);
3498 
3499 		set_buffer_uptodate(bh);
3500 		lock_buffer(bh);
3501 		bh->b_end_io = btrfs_end_buffer_write_sync;
3502 		bh->b_private = device;
3503 
3504 		/*
3505 		 * we fua the first super.  The others we allow
3506 		 * to go down lazy.
3507 		 */
3508 		op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3509 		if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3510 			op_flags |= REQ_FUA;
3511 		ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3512 		if (ret)
3513 			errors++;
3514 	}
3515 	return errors < i ? 0 : -1;
3516 }
3517 
3518 /*
3519  * Wait for write completion of superblocks done by write_dev_supers,
3520  * @max_mirrors same for write and wait phases.
3521  *
3522  * Return number of errors when buffer head is not found or not marked up to
3523  * date.
3524  */
3525 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3526 {
3527 	struct buffer_head *bh;
3528 	int i;
3529 	int errors = 0;
3530 	bool primary_failed = false;
3531 	u64 bytenr;
3532 
3533 	if (max_mirrors == 0)
3534 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3535 
3536 	for (i = 0; i < max_mirrors; i++) {
3537 		bytenr = btrfs_sb_offset(i);
3538 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3539 		    device->commit_total_bytes)
3540 			break;
3541 
3542 		bh = __find_get_block(device->bdev,
3543 				      bytenr / BTRFS_BDEV_BLOCKSIZE,
3544 				      BTRFS_SUPER_INFO_SIZE);
3545 		if (!bh) {
3546 			errors++;
3547 			if (i == 0)
3548 				primary_failed = true;
3549 			continue;
3550 		}
3551 		wait_on_buffer(bh);
3552 		if (!buffer_uptodate(bh)) {
3553 			errors++;
3554 			if (i == 0)
3555 				primary_failed = true;
3556 		}
3557 
3558 		/* drop our reference */
3559 		brelse(bh);
3560 
3561 		/* drop the reference from the writing run */
3562 		brelse(bh);
3563 	}
3564 
3565 	/* log error, force error return */
3566 	if (primary_failed) {
3567 		btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3568 			  device->devid);
3569 		return -1;
3570 	}
3571 
3572 	return errors < i ? 0 : -1;
3573 }
3574 
3575 /*
3576  * endio for the write_dev_flush, this will wake anyone waiting
3577  * for the barrier when it is done
3578  */
3579 static void btrfs_end_empty_barrier(struct bio *bio)
3580 {
3581 	complete(bio->bi_private);
3582 }
3583 
3584 /*
3585  * Submit a flush request to the device if it supports it. Error handling is
3586  * done in the waiting counterpart.
3587  */
3588 static void write_dev_flush(struct btrfs_device *device)
3589 {
3590 	struct request_queue *q = bdev_get_queue(device->bdev);
3591 	struct bio *bio = device->flush_bio;
3592 
3593 	if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3594 		return;
3595 
3596 	bio_reset(bio);
3597 	bio->bi_end_io = btrfs_end_empty_barrier;
3598 	bio_set_dev(bio, device->bdev);
3599 	bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3600 	init_completion(&device->flush_wait);
3601 	bio->bi_private = &device->flush_wait;
3602 
3603 	btrfsic_submit_bio(bio);
3604 	set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3605 }
3606 
3607 /*
3608  * If the flush bio has been submitted by write_dev_flush, wait for it.
3609  */
3610 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3611 {
3612 	struct bio *bio = device->flush_bio;
3613 
3614 	if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3615 		return BLK_STS_OK;
3616 
3617 	clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3618 	wait_for_completion_io(&device->flush_wait);
3619 
3620 	return bio->bi_status;
3621 }
3622 
3623 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3624 {
3625 	if (!btrfs_check_rw_degradable(fs_info, NULL))
3626 		return -EIO;
3627 	return 0;
3628 }
3629 
3630 /*
3631  * send an empty flush down to each device in parallel,
3632  * then wait for them
3633  */
3634 static int barrier_all_devices(struct btrfs_fs_info *info)
3635 {
3636 	struct list_head *head;
3637 	struct btrfs_device *dev;
3638 	int errors_wait = 0;
3639 	blk_status_t ret;
3640 
3641 	lockdep_assert_held(&info->fs_devices->device_list_mutex);
3642 	/* send down all the barriers */
3643 	head = &info->fs_devices->devices;
3644 	list_for_each_entry(dev, head, dev_list) {
3645 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3646 			continue;
3647 		if (!dev->bdev)
3648 			continue;
3649 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3650 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3651 			continue;
3652 
3653 		write_dev_flush(dev);
3654 		dev->last_flush_error = BLK_STS_OK;
3655 	}
3656 
3657 	/* wait for all the barriers */
3658 	list_for_each_entry(dev, head, dev_list) {
3659 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3660 			continue;
3661 		if (!dev->bdev) {
3662 			errors_wait++;
3663 			continue;
3664 		}
3665 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3666 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3667 			continue;
3668 
3669 		ret = wait_dev_flush(dev);
3670 		if (ret) {
3671 			dev->last_flush_error = ret;
3672 			btrfs_dev_stat_inc_and_print(dev,
3673 					BTRFS_DEV_STAT_FLUSH_ERRS);
3674 			errors_wait++;
3675 		}
3676 	}
3677 
3678 	if (errors_wait) {
3679 		/*
3680 		 * At some point we need the status of all disks
3681 		 * to arrive at the volume status. So error checking
3682 		 * is being pushed to a separate loop.
3683 		 */
3684 		return check_barrier_error(info);
3685 	}
3686 	return 0;
3687 }
3688 
3689 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3690 {
3691 	int raid_type;
3692 	int min_tolerated = INT_MAX;
3693 
3694 	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3695 	    (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3696 		min_tolerated = min_t(int, min_tolerated,
3697 				    btrfs_raid_array[BTRFS_RAID_SINGLE].
3698 				    tolerated_failures);
3699 
3700 	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3701 		if (raid_type == BTRFS_RAID_SINGLE)
3702 			continue;
3703 		if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3704 			continue;
3705 		min_tolerated = min_t(int, min_tolerated,
3706 				    btrfs_raid_array[raid_type].
3707 				    tolerated_failures);
3708 	}
3709 
3710 	if (min_tolerated == INT_MAX) {
3711 		pr_warn("BTRFS: unknown raid flag: %llu", flags);
3712 		min_tolerated = 0;
3713 	}
3714 
3715 	return min_tolerated;
3716 }
3717 
3718 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3719 {
3720 	struct list_head *head;
3721 	struct btrfs_device *dev;
3722 	struct btrfs_super_block *sb;
3723 	struct btrfs_dev_item *dev_item;
3724 	int ret;
3725 	int do_barriers;
3726 	int max_errors;
3727 	int total_errors = 0;
3728 	u64 flags;
3729 
3730 	do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3731 
3732 	/*
3733 	 * max_mirrors == 0 indicates we're from commit_transaction,
3734 	 * not from fsync where the tree roots in fs_info have not
3735 	 * been consistent on disk.
3736 	 */
3737 	if (max_mirrors == 0)
3738 		backup_super_roots(fs_info);
3739 
3740 	sb = fs_info->super_for_commit;
3741 	dev_item = &sb->dev_item;
3742 
3743 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
3744 	head = &fs_info->fs_devices->devices;
3745 	max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3746 
3747 	if (do_barriers) {
3748 		ret = barrier_all_devices(fs_info);
3749 		if (ret) {
3750 			mutex_unlock(
3751 				&fs_info->fs_devices->device_list_mutex);
3752 			btrfs_handle_fs_error(fs_info, ret,
3753 					      "errors while submitting device barriers.");
3754 			return ret;
3755 		}
3756 	}
3757 
3758 	list_for_each_entry(dev, head, dev_list) {
3759 		if (!dev->bdev) {
3760 			total_errors++;
3761 			continue;
3762 		}
3763 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3764 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3765 			continue;
3766 
3767 		btrfs_set_stack_device_generation(dev_item, 0);
3768 		btrfs_set_stack_device_type(dev_item, dev->type);
3769 		btrfs_set_stack_device_id(dev_item, dev->devid);
3770 		btrfs_set_stack_device_total_bytes(dev_item,
3771 						   dev->commit_total_bytes);
3772 		btrfs_set_stack_device_bytes_used(dev_item,
3773 						  dev->commit_bytes_used);
3774 		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3775 		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3776 		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3777 		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3778 		memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3779 		       BTRFS_FSID_SIZE);
3780 
3781 		flags = btrfs_super_flags(sb);
3782 		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3783 
3784 		ret = btrfs_validate_write_super(fs_info, sb);
3785 		if (ret < 0) {
3786 			mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3787 			btrfs_handle_fs_error(fs_info, -EUCLEAN,
3788 				"unexpected superblock corruption detected");
3789 			return -EUCLEAN;
3790 		}
3791 
3792 		ret = write_dev_supers(dev, sb, max_mirrors);
3793 		if (ret)
3794 			total_errors++;
3795 	}
3796 	if (total_errors > max_errors) {
3797 		btrfs_err(fs_info, "%d errors while writing supers",
3798 			  total_errors);
3799 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3800 
3801 		/* FUA is masked off if unsupported and can't be the reason */
3802 		btrfs_handle_fs_error(fs_info, -EIO,
3803 				      "%d errors while writing supers",
3804 				      total_errors);
3805 		return -EIO;
3806 	}
3807 
3808 	total_errors = 0;
3809 	list_for_each_entry(dev, head, dev_list) {
3810 		if (!dev->bdev)
3811 			continue;
3812 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3813 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3814 			continue;
3815 
3816 		ret = wait_dev_supers(dev, max_mirrors);
3817 		if (ret)
3818 			total_errors++;
3819 	}
3820 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3821 	if (total_errors > max_errors) {
3822 		btrfs_handle_fs_error(fs_info, -EIO,
3823 				      "%d errors while writing supers",
3824 				      total_errors);
3825 		return -EIO;
3826 	}
3827 	return 0;
3828 }
3829 
3830 /* Drop a fs root from the radix tree and free it. */
3831 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3832 				  struct btrfs_root *root)
3833 {
3834 	spin_lock(&fs_info->fs_roots_radix_lock);
3835 	radix_tree_delete(&fs_info->fs_roots_radix,
3836 			  (unsigned long)root->root_key.objectid);
3837 	spin_unlock(&fs_info->fs_roots_radix_lock);
3838 
3839 	if (btrfs_root_refs(&root->root_item) == 0)
3840 		synchronize_srcu(&fs_info->subvol_srcu);
3841 
3842 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3843 		btrfs_free_log(NULL, root);
3844 		if (root->reloc_root) {
3845 			free_extent_buffer(root->reloc_root->node);
3846 			free_extent_buffer(root->reloc_root->commit_root);
3847 			btrfs_put_fs_root(root->reloc_root);
3848 			root->reloc_root = NULL;
3849 		}
3850 	}
3851 
3852 	if (root->free_ino_pinned)
3853 		__btrfs_remove_free_space_cache(root->free_ino_pinned);
3854 	if (root->free_ino_ctl)
3855 		__btrfs_remove_free_space_cache(root->free_ino_ctl);
3856 	btrfs_free_fs_root(root);
3857 }
3858 
3859 void btrfs_free_fs_root(struct btrfs_root *root)
3860 {
3861 	iput(root->ino_cache_inode);
3862 	WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3863 	if (root->anon_dev)
3864 		free_anon_bdev(root->anon_dev);
3865 	if (root->subv_writers)
3866 		btrfs_free_subvolume_writers(root->subv_writers);
3867 	free_extent_buffer(root->node);
3868 	free_extent_buffer(root->commit_root);
3869 	kfree(root->free_ino_ctl);
3870 	kfree(root->free_ino_pinned);
3871 	btrfs_put_fs_root(root);
3872 }
3873 
3874 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3875 {
3876 	u64 root_objectid = 0;
3877 	struct btrfs_root *gang[8];
3878 	int i = 0;
3879 	int err = 0;
3880 	unsigned int ret = 0;
3881 	int index;
3882 
3883 	while (1) {
3884 		index = srcu_read_lock(&fs_info->subvol_srcu);
3885 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3886 					     (void **)gang, root_objectid,
3887 					     ARRAY_SIZE(gang));
3888 		if (!ret) {
3889 			srcu_read_unlock(&fs_info->subvol_srcu, index);
3890 			break;
3891 		}
3892 		root_objectid = gang[ret - 1]->root_key.objectid + 1;
3893 
3894 		for (i = 0; i < ret; i++) {
3895 			/* Avoid to grab roots in dead_roots */
3896 			if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3897 				gang[i] = NULL;
3898 				continue;
3899 			}
3900 			/* grab all the search result for later use */
3901 			gang[i] = btrfs_grab_fs_root(gang[i]);
3902 		}
3903 		srcu_read_unlock(&fs_info->subvol_srcu, index);
3904 
3905 		for (i = 0; i < ret; i++) {
3906 			if (!gang[i])
3907 				continue;
3908 			root_objectid = gang[i]->root_key.objectid;
3909 			err = btrfs_orphan_cleanup(gang[i]);
3910 			if (err)
3911 				break;
3912 			btrfs_put_fs_root(gang[i]);
3913 		}
3914 		root_objectid++;
3915 	}
3916 
3917 	/* release the uncleaned roots due to error */
3918 	for (; i < ret; i++) {
3919 		if (gang[i])
3920 			btrfs_put_fs_root(gang[i]);
3921 	}
3922 	return err;
3923 }
3924 
3925 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3926 {
3927 	struct btrfs_root *root = fs_info->tree_root;
3928 	struct btrfs_trans_handle *trans;
3929 
3930 	mutex_lock(&fs_info->cleaner_mutex);
3931 	btrfs_run_delayed_iputs(fs_info);
3932 	mutex_unlock(&fs_info->cleaner_mutex);
3933 	wake_up_process(fs_info->cleaner_kthread);
3934 
3935 	/* wait until ongoing cleanup work done */
3936 	down_write(&fs_info->cleanup_work_sem);
3937 	up_write(&fs_info->cleanup_work_sem);
3938 
3939 	trans = btrfs_join_transaction(root);
3940 	if (IS_ERR(trans))
3941 		return PTR_ERR(trans);
3942 	return btrfs_commit_transaction(trans);
3943 }
3944 
3945 void __cold close_ctree(struct btrfs_fs_info *fs_info)
3946 {
3947 	int ret;
3948 
3949 	set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3950 	/*
3951 	 * We don't want the cleaner to start new transactions, add more delayed
3952 	 * iputs, etc. while we're closing. We can't use kthread_stop() yet
3953 	 * because that frees the task_struct, and the transaction kthread might
3954 	 * still try to wake up the cleaner.
3955 	 */
3956 	kthread_park(fs_info->cleaner_kthread);
3957 
3958 	/* wait for the qgroup rescan worker to stop */
3959 	btrfs_qgroup_wait_for_completion(fs_info, false);
3960 
3961 	/* wait for the uuid_scan task to finish */
3962 	down(&fs_info->uuid_tree_rescan_sem);
3963 	/* avoid complains from lockdep et al., set sem back to initial state */
3964 	up(&fs_info->uuid_tree_rescan_sem);
3965 
3966 	/* pause restriper - we want to resume on mount */
3967 	btrfs_pause_balance(fs_info);
3968 
3969 	btrfs_dev_replace_suspend_for_unmount(fs_info);
3970 
3971 	btrfs_scrub_cancel(fs_info);
3972 
3973 	/* wait for any defraggers to finish */
3974 	wait_event(fs_info->transaction_wait,
3975 		   (atomic_read(&fs_info->defrag_running) == 0));
3976 
3977 	/* clear out the rbtree of defraggable inodes */
3978 	btrfs_cleanup_defrag_inodes(fs_info);
3979 
3980 	cancel_work_sync(&fs_info->async_reclaim_work);
3981 
3982 	/* Cancel or finish ongoing discard work */
3983 	btrfs_discard_cleanup(fs_info);
3984 
3985 	if (!sb_rdonly(fs_info->sb)) {
3986 		/*
3987 		 * The cleaner kthread is stopped, so do one final pass over
3988 		 * unused block groups.
3989 		 */
3990 		btrfs_delete_unused_bgs(fs_info);
3991 
3992 		ret = btrfs_commit_super(fs_info);
3993 		if (ret)
3994 			btrfs_err(fs_info, "commit super ret %d", ret);
3995 	}
3996 
3997 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
3998 	    test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
3999 		btrfs_error_commit_super(fs_info);
4000 
4001 	kthread_stop(fs_info->transaction_kthread);
4002 	kthread_stop(fs_info->cleaner_kthread);
4003 
4004 	ASSERT(list_empty(&fs_info->delayed_iputs));
4005 	set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4006 
4007 	btrfs_free_qgroup_config(fs_info);
4008 	ASSERT(list_empty(&fs_info->delalloc_roots));
4009 
4010 	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4011 		btrfs_info(fs_info, "at unmount delalloc count %lld",
4012 		       percpu_counter_sum(&fs_info->delalloc_bytes));
4013 	}
4014 
4015 	if (percpu_counter_sum(&fs_info->dio_bytes))
4016 		btrfs_info(fs_info, "at unmount dio bytes count %lld",
4017 			   percpu_counter_sum(&fs_info->dio_bytes));
4018 
4019 	btrfs_sysfs_remove_mounted(fs_info);
4020 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4021 
4022 	btrfs_free_fs_roots(fs_info);
4023 
4024 	btrfs_put_block_group_cache(fs_info);
4025 
4026 	/*
4027 	 * we must make sure there is not any read request to
4028 	 * submit after we stopping all workers.
4029 	 */
4030 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4031 	btrfs_stop_all_workers(fs_info);
4032 
4033 	clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4034 	free_root_pointers(fs_info, true);
4035 
4036 	/*
4037 	 * We must free the block groups after dropping the fs_roots as we could
4038 	 * have had an IO error and have left over tree log blocks that aren't
4039 	 * cleaned up until the fs roots are freed.  This makes the block group
4040 	 * accounting appear to be wrong because there's pending reserved bytes,
4041 	 * so make sure we do the block group cleanup afterwards.
4042 	 */
4043 	btrfs_free_block_groups(fs_info);
4044 
4045 	iput(fs_info->btree_inode);
4046 
4047 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4048 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4049 		btrfsic_unmount(fs_info->fs_devices);
4050 #endif
4051 
4052 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
4053 	btrfs_close_devices(fs_info->fs_devices);
4054 
4055 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4056 	percpu_counter_destroy(&fs_info->delalloc_bytes);
4057 	percpu_counter_destroy(&fs_info->dio_bytes);
4058 	percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
4059 	cleanup_srcu_struct(&fs_info->subvol_srcu);
4060 
4061 	btrfs_free_csum_hash(fs_info);
4062 	btrfs_free_stripe_hash_table(fs_info);
4063 	btrfs_free_ref_cache(fs_info);
4064 }
4065 
4066 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4067 			  int atomic)
4068 {
4069 	int ret;
4070 	struct inode *btree_inode = buf->pages[0]->mapping->host;
4071 
4072 	ret = extent_buffer_uptodate(buf);
4073 	if (!ret)
4074 		return ret;
4075 
4076 	ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4077 				    parent_transid, atomic);
4078 	if (ret == -EAGAIN)
4079 		return ret;
4080 	return !ret;
4081 }
4082 
4083 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4084 {
4085 	struct btrfs_fs_info *fs_info;
4086 	struct btrfs_root *root;
4087 	u64 transid = btrfs_header_generation(buf);
4088 	int was_dirty;
4089 
4090 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4091 	/*
4092 	 * This is a fast path so only do this check if we have sanity tests
4093 	 * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4094 	 * outside of the sanity tests.
4095 	 */
4096 	if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4097 		return;
4098 #endif
4099 	root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4100 	fs_info = root->fs_info;
4101 	btrfs_assert_tree_locked(buf);
4102 	if (transid != fs_info->generation)
4103 		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4104 			buf->start, transid, fs_info->generation);
4105 	was_dirty = set_extent_buffer_dirty(buf);
4106 	if (!was_dirty)
4107 		percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4108 					 buf->len,
4109 					 fs_info->dirty_metadata_batch);
4110 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4111 	/*
4112 	 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4113 	 * but item data not updated.
4114 	 * So here we should only check item pointers, not item data.
4115 	 */
4116 	if (btrfs_header_level(buf) == 0 &&
4117 	    btrfs_check_leaf_relaxed(buf)) {
4118 		btrfs_print_leaf(buf);
4119 		ASSERT(0);
4120 	}
4121 #endif
4122 }
4123 
4124 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4125 					int flush_delayed)
4126 {
4127 	/*
4128 	 * looks as though older kernels can get into trouble with
4129 	 * this code, they end up stuck in balance_dirty_pages forever
4130 	 */
4131 	int ret;
4132 
4133 	if (current->flags & PF_MEMALLOC)
4134 		return;
4135 
4136 	if (flush_delayed)
4137 		btrfs_balance_delayed_items(fs_info);
4138 
4139 	ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4140 				     BTRFS_DIRTY_METADATA_THRESH,
4141 				     fs_info->dirty_metadata_batch);
4142 	if (ret > 0) {
4143 		balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4144 	}
4145 }
4146 
4147 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4148 {
4149 	__btrfs_btree_balance_dirty(fs_info, 1);
4150 }
4151 
4152 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4153 {
4154 	__btrfs_btree_balance_dirty(fs_info, 0);
4155 }
4156 
4157 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4158 		      struct btrfs_key *first_key)
4159 {
4160 	return btree_read_extent_buffer_pages(buf, parent_transid,
4161 					      level, first_key);
4162 }
4163 
4164 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4165 {
4166 	/* cleanup FS via transaction */
4167 	btrfs_cleanup_transaction(fs_info);
4168 
4169 	mutex_lock(&fs_info->cleaner_mutex);
4170 	btrfs_run_delayed_iputs(fs_info);
4171 	mutex_unlock(&fs_info->cleaner_mutex);
4172 
4173 	down_write(&fs_info->cleanup_work_sem);
4174 	up_write(&fs_info->cleanup_work_sem);
4175 }
4176 
4177 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4178 {
4179 	struct btrfs_ordered_extent *ordered;
4180 
4181 	spin_lock(&root->ordered_extent_lock);
4182 	/*
4183 	 * This will just short circuit the ordered completion stuff which will
4184 	 * make sure the ordered extent gets properly cleaned up.
4185 	 */
4186 	list_for_each_entry(ordered, &root->ordered_extents,
4187 			    root_extent_list)
4188 		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4189 	spin_unlock(&root->ordered_extent_lock);
4190 }
4191 
4192 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4193 {
4194 	struct btrfs_root *root;
4195 	struct list_head splice;
4196 
4197 	INIT_LIST_HEAD(&splice);
4198 
4199 	spin_lock(&fs_info->ordered_root_lock);
4200 	list_splice_init(&fs_info->ordered_roots, &splice);
4201 	while (!list_empty(&splice)) {
4202 		root = list_first_entry(&splice, struct btrfs_root,
4203 					ordered_root);
4204 		list_move_tail(&root->ordered_root,
4205 			       &fs_info->ordered_roots);
4206 
4207 		spin_unlock(&fs_info->ordered_root_lock);
4208 		btrfs_destroy_ordered_extents(root);
4209 
4210 		cond_resched();
4211 		spin_lock(&fs_info->ordered_root_lock);
4212 	}
4213 	spin_unlock(&fs_info->ordered_root_lock);
4214 
4215 	/*
4216 	 * We need this here because if we've been flipped read-only we won't
4217 	 * get sync() from the umount, so we need to make sure any ordered
4218 	 * extents that haven't had their dirty pages IO start writeout yet
4219 	 * actually get run and error out properly.
4220 	 */
4221 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4222 }
4223 
4224 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4225 				      struct btrfs_fs_info *fs_info)
4226 {
4227 	struct rb_node *node;
4228 	struct btrfs_delayed_ref_root *delayed_refs;
4229 	struct btrfs_delayed_ref_node *ref;
4230 	int ret = 0;
4231 
4232 	delayed_refs = &trans->delayed_refs;
4233 
4234 	spin_lock(&delayed_refs->lock);
4235 	if (atomic_read(&delayed_refs->num_entries) == 0) {
4236 		spin_unlock(&delayed_refs->lock);
4237 		btrfs_info(fs_info, "delayed_refs has NO entry");
4238 		return ret;
4239 	}
4240 
4241 	while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4242 		struct btrfs_delayed_ref_head *head;
4243 		struct rb_node *n;
4244 		bool pin_bytes = false;
4245 
4246 		head = rb_entry(node, struct btrfs_delayed_ref_head,
4247 				href_node);
4248 		if (btrfs_delayed_ref_lock(delayed_refs, head))
4249 			continue;
4250 
4251 		spin_lock(&head->lock);
4252 		while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4253 			ref = rb_entry(n, struct btrfs_delayed_ref_node,
4254 				       ref_node);
4255 			ref->in_tree = 0;
4256 			rb_erase_cached(&ref->ref_node, &head->ref_tree);
4257 			RB_CLEAR_NODE(&ref->ref_node);
4258 			if (!list_empty(&ref->add_list))
4259 				list_del(&ref->add_list);
4260 			atomic_dec(&delayed_refs->num_entries);
4261 			btrfs_put_delayed_ref(ref);
4262 		}
4263 		if (head->must_insert_reserved)
4264 			pin_bytes = true;
4265 		btrfs_free_delayed_extent_op(head->extent_op);
4266 		btrfs_delete_ref_head(delayed_refs, head);
4267 		spin_unlock(&head->lock);
4268 		spin_unlock(&delayed_refs->lock);
4269 		mutex_unlock(&head->mutex);
4270 
4271 		if (pin_bytes)
4272 			btrfs_pin_extent(fs_info, head->bytenr,
4273 					 head->num_bytes, 1);
4274 		btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4275 		btrfs_put_delayed_ref_head(head);
4276 		cond_resched();
4277 		spin_lock(&delayed_refs->lock);
4278 	}
4279 
4280 	spin_unlock(&delayed_refs->lock);
4281 
4282 	return ret;
4283 }
4284 
4285 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4286 {
4287 	struct btrfs_inode *btrfs_inode;
4288 	struct list_head splice;
4289 
4290 	INIT_LIST_HEAD(&splice);
4291 
4292 	spin_lock(&root->delalloc_lock);
4293 	list_splice_init(&root->delalloc_inodes, &splice);
4294 
4295 	while (!list_empty(&splice)) {
4296 		struct inode *inode = NULL;
4297 		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4298 					       delalloc_inodes);
4299 		__btrfs_del_delalloc_inode(root, btrfs_inode);
4300 		spin_unlock(&root->delalloc_lock);
4301 
4302 		/*
4303 		 * Make sure we get a live inode and that it'll not disappear
4304 		 * meanwhile.
4305 		 */
4306 		inode = igrab(&btrfs_inode->vfs_inode);
4307 		if (inode) {
4308 			invalidate_inode_pages2(inode->i_mapping);
4309 			iput(inode);
4310 		}
4311 		spin_lock(&root->delalloc_lock);
4312 	}
4313 	spin_unlock(&root->delalloc_lock);
4314 }
4315 
4316 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4317 {
4318 	struct btrfs_root *root;
4319 	struct list_head splice;
4320 
4321 	INIT_LIST_HEAD(&splice);
4322 
4323 	spin_lock(&fs_info->delalloc_root_lock);
4324 	list_splice_init(&fs_info->delalloc_roots, &splice);
4325 	while (!list_empty(&splice)) {
4326 		root = list_first_entry(&splice, struct btrfs_root,
4327 					 delalloc_root);
4328 		root = btrfs_grab_fs_root(root);
4329 		BUG_ON(!root);
4330 		spin_unlock(&fs_info->delalloc_root_lock);
4331 
4332 		btrfs_destroy_delalloc_inodes(root);
4333 		btrfs_put_fs_root(root);
4334 
4335 		spin_lock(&fs_info->delalloc_root_lock);
4336 	}
4337 	spin_unlock(&fs_info->delalloc_root_lock);
4338 }
4339 
4340 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4341 					struct extent_io_tree *dirty_pages,
4342 					int mark)
4343 {
4344 	int ret;
4345 	struct extent_buffer *eb;
4346 	u64 start = 0;
4347 	u64 end;
4348 
4349 	while (1) {
4350 		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4351 					    mark, NULL);
4352 		if (ret)
4353 			break;
4354 
4355 		clear_extent_bits(dirty_pages, start, end, mark);
4356 		while (start <= end) {
4357 			eb = find_extent_buffer(fs_info, start);
4358 			start += fs_info->nodesize;
4359 			if (!eb)
4360 				continue;
4361 			wait_on_extent_buffer_writeback(eb);
4362 
4363 			if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4364 					       &eb->bflags))
4365 				clear_extent_buffer_dirty(eb);
4366 			free_extent_buffer_stale(eb);
4367 		}
4368 	}
4369 
4370 	return ret;
4371 }
4372 
4373 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4374 				       struct extent_io_tree *pinned_extents)
4375 {
4376 	struct extent_io_tree *unpin;
4377 	u64 start;
4378 	u64 end;
4379 	int ret;
4380 	bool loop = true;
4381 
4382 	unpin = pinned_extents;
4383 again:
4384 	while (1) {
4385 		struct extent_state *cached_state = NULL;
4386 
4387 		/*
4388 		 * The btrfs_finish_extent_commit() may get the same range as
4389 		 * ours between find_first_extent_bit and clear_extent_dirty.
4390 		 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4391 		 * the same extent range.
4392 		 */
4393 		mutex_lock(&fs_info->unused_bg_unpin_mutex);
4394 		ret = find_first_extent_bit(unpin, 0, &start, &end,
4395 					    EXTENT_DIRTY, &cached_state);
4396 		if (ret) {
4397 			mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4398 			break;
4399 		}
4400 
4401 		clear_extent_dirty(unpin, start, end, &cached_state);
4402 		free_extent_state(cached_state);
4403 		btrfs_error_unpin_extent_range(fs_info, start, end);
4404 		mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4405 		cond_resched();
4406 	}
4407 
4408 	if (loop) {
4409 		if (unpin == &fs_info->freed_extents[0])
4410 			unpin = &fs_info->freed_extents[1];
4411 		else
4412 			unpin = &fs_info->freed_extents[0];
4413 		loop = false;
4414 		goto again;
4415 	}
4416 
4417 	return 0;
4418 }
4419 
4420 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4421 {
4422 	struct inode *inode;
4423 
4424 	inode = cache->io_ctl.inode;
4425 	if (inode) {
4426 		invalidate_inode_pages2(inode->i_mapping);
4427 		BTRFS_I(inode)->generation = 0;
4428 		cache->io_ctl.inode = NULL;
4429 		iput(inode);
4430 	}
4431 	btrfs_put_block_group(cache);
4432 }
4433 
4434 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4435 			     struct btrfs_fs_info *fs_info)
4436 {
4437 	struct btrfs_block_group *cache;
4438 
4439 	spin_lock(&cur_trans->dirty_bgs_lock);
4440 	while (!list_empty(&cur_trans->dirty_bgs)) {
4441 		cache = list_first_entry(&cur_trans->dirty_bgs,
4442 					 struct btrfs_block_group,
4443 					 dirty_list);
4444 
4445 		if (!list_empty(&cache->io_list)) {
4446 			spin_unlock(&cur_trans->dirty_bgs_lock);
4447 			list_del_init(&cache->io_list);
4448 			btrfs_cleanup_bg_io(cache);
4449 			spin_lock(&cur_trans->dirty_bgs_lock);
4450 		}
4451 
4452 		list_del_init(&cache->dirty_list);
4453 		spin_lock(&cache->lock);
4454 		cache->disk_cache_state = BTRFS_DC_ERROR;
4455 		spin_unlock(&cache->lock);
4456 
4457 		spin_unlock(&cur_trans->dirty_bgs_lock);
4458 		btrfs_put_block_group(cache);
4459 		btrfs_delayed_refs_rsv_release(fs_info, 1);
4460 		spin_lock(&cur_trans->dirty_bgs_lock);
4461 	}
4462 	spin_unlock(&cur_trans->dirty_bgs_lock);
4463 
4464 	/*
4465 	 * Refer to the definition of io_bgs member for details why it's safe
4466 	 * to use it without any locking
4467 	 */
4468 	while (!list_empty(&cur_trans->io_bgs)) {
4469 		cache = list_first_entry(&cur_trans->io_bgs,
4470 					 struct btrfs_block_group,
4471 					 io_list);
4472 
4473 		list_del_init(&cache->io_list);
4474 		spin_lock(&cache->lock);
4475 		cache->disk_cache_state = BTRFS_DC_ERROR;
4476 		spin_unlock(&cache->lock);
4477 		btrfs_cleanup_bg_io(cache);
4478 	}
4479 }
4480 
4481 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4482 				   struct btrfs_fs_info *fs_info)
4483 {
4484 	struct btrfs_device *dev, *tmp;
4485 
4486 	btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4487 	ASSERT(list_empty(&cur_trans->dirty_bgs));
4488 	ASSERT(list_empty(&cur_trans->io_bgs));
4489 
4490 	list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4491 				 post_commit_list) {
4492 		list_del_init(&dev->post_commit_list);
4493 	}
4494 
4495 	btrfs_destroy_delayed_refs(cur_trans, fs_info);
4496 
4497 	cur_trans->state = TRANS_STATE_COMMIT_START;
4498 	wake_up(&fs_info->transaction_blocked_wait);
4499 
4500 	cur_trans->state = TRANS_STATE_UNBLOCKED;
4501 	wake_up(&fs_info->transaction_wait);
4502 
4503 	btrfs_destroy_delayed_inodes(fs_info);
4504 	btrfs_assert_delayed_root_empty(fs_info);
4505 
4506 	btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4507 				     EXTENT_DIRTY);
4508 	btrfs_destroy_pinned_extent(fs_info,
4509 				    fs_info->pinned_extents);
4510 
4511 	cur_trans->state =TRANS_STATE_COMPLETED;
4512 	wake_up(&cur_trans->commit_wait);
4513 }
4514 
4515 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4516 {
4517 	struct btrfs_transaction *t;
4518 
4519 	mutex_lock(&fs_info->transaction_kthread_mutex);
4520 
4521 	spin_lock(&fs_info->trans_lock);
4522 	while (!list_empty(&fs_info->trans_list)) {
4523 		t = list_first_entry(&fs_info->trans_list,
4524 				     struct btrfs_transaction, list);
4525 		if (t->state >= TRANS_STATE_COMMIT_START) {
4526 			refcount_inc(&t->use_count);
4527 			spin_unlock(&fs_info->trans_lock);
4528 			btrfs_wait_for_commit(fs_info, t->transid);
4529 			btrfs_put_transaction(t);
4530 			spin_lock(&fs_info->trans_lock);
4531 			continue;
4532 		}
4533 		if (t == fs_info->running_transaction) {
4534 			t->state = TRANS_STATE_COMMIT_DOING;
4535 			spin_unlock(&fs_info->trans_lock);
4536 			/*
4537 			 * We wait for 0 num_writers since we don't hold a trans
4538 			 * handle open currently for this transaction.
4539 			 */
4540 			wait_event(t->writer_wait,
4541 				   atomic_read(&t->num_writers) == 0);
4542 		} else {
4543 			spin_unlock(&fs_info->trans_lock);
4544 		}
4545 		btrfs_cleanup_one_transaction(t, fs_info);
4546 
4547 		spin_lock(&fs_info->trans_lock);
4548 		if (t == fs_info->running_transaction)
4549 			fs_info->running_transaction = NULL;
4550 		list_del_init(&t->list);
4551 		spin_unlock(&fs_info->trans_lock);
4552 
4553 		btrfs_put_transaction(t);
4554 		trace_btrfs_transaction_commit(fs_info->tree_root);
4555 		spin_lock(&fs_info->trans_lock);
4556 	}
4557 	spin_unlock(&fs_info->trans_lock);
4558 	btrfs_destroy_all_ordered_extents(fs_info);
4559 	btrfs_destroy_delayed_inodes(fs_info);
4560 	btrfs_assert_delayed_root_empty(fs_info);
4561 	btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4562 	btrfs_destroy_all_delalloc_inodes(fs_info);
4563 	mutex_unlock(&fs_info->transaction_kthread_mutex);
4564 
4565 	return 0;
4566 }
4567 
4568 static const struct extent_io_ops btree_extent_io_ops = {
4569 	/* mandatory callbacks */
4570 	.submit_bio_hook = btree_submit_bio_hook,
4571 	.readpage_end_io_hook = btree_readpage_end_io_hook,
4572 };
4573