xref: /linux/fs/btrfs/disk-io.c (revision 288440de9e5fdb4a3ff73864850f080c1250fc81)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5 
6 #include <linux/fs.h>
7 #include <linux/blkdev.h>
8 #include <linux/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/workqueue.h>
11 #include <linux/kthread.h>
12 #include <linux/slab.h>
13 #include <linux/migrate.h>
14 #include <linux/ratelimit.h>
15 #include <linux/uuid.h>
16 #include <linux/semaphore.h>
17 #include <linux/error-injection.h>
18 #include <linux/crc32c.h>
19 #include <linux/sched/mm.h>
20 #include <asm/unaligned.h>
21 #include <crypto/hash.h>
22 #include "ctree.h"
23 #include "disk-io.h"
24 #include "transaction.h"
25 #include "btrfs_inode.h"
26 #include "volumes.h"
27 #include "print-tree.h"
28 #include "locking.h"
29 #include "tree-log.h"
30 #include "free-space-cache.h"
31 #include "free-space-tree.h"
32 #include "check-integrity.h"
33 #include "rcu-string.h"
34 #include "dev-replace.h"
35 #include "raid56.h"
36 #include "sysfs.h"
37 #include "qgroup.h"
38 #include "compression.h"
39 #include "tree-checker.h"
40 #include "ref-verify.h"
41 #include "block-group.h"
42 #include "discard.h"
43 #include "space-info.h"
44 #include "zoned.h"
45 #include "subpage.h"
46 
47 #define BTRFS_SUPER_FLAG_SUPP	(BTRFS_HEADER_FLAG_WRITTEN |\
48 				 BTRFS_HEADER_FLAG_RELOC |\
49 				 BTRFS_SUPER_FLAG_ERROR |\
50 				 BTRFS_SUPER_FLAG_SEEDING |\
51 				 BTRFS_SUPER_FLAG_METADUMP |\
52 				 BTRFS_SUPER_FLAG_METADUMP_V2)
53 
54 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
55 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56 				      struct btrfs_fs_info *fs_info);
57 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
59 					struct extent_io_tree *dirty_pages,
60 					int mark);
61 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
62 				       struct extent_io_tree *pinned_extents);
63 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
64 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
65 
66 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
67 {
68 	if (fs_info->csum_shash)
69 		crypto_free_shash(fs_info->csum_shash);
70 }
71 
72 /*
73  * async submit bios are used to offload expensive checksumming
74  * onto the worker threads.  They checksum file and metadata bios
75  * just before they are sent down the IO stack.
76  */
77 struct async_submit_bio {
78 	struct inode *inode;
79 	struct bio *bio;
80 	extent_submit_bio_start_t *submit_bio_start;
81 	int mirror_num;
82 
83 	/* Optional parameter for submit_bio_start used by direct io */
84 	u64 dio_file_offset;
85 	struct btrfs_work work;
86 	blk_status_t status;
87 };
88 
89 /*
90  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
91  * eb, the lockdep key is determined by the btrfs_root it belongs to and
92  * the level the eb occupies in the tree.
93  *
94  * Different roots are used for different purposes and may nest inside each
95  * other and they require separate keysets.  As lockdep keys should be
96  * static, assign keysets according to the purpose of the root as indicated
97  * by btrfs_root->root_key.objectid.  This ensures that all special purpose
98  * roots have separate keysets.
99  *
100  * Lock-nesting across peer nodes is always done with the immediate parent
101  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
102  * subclass to avoid triggering lockdep warning in such cases.
103  *
104  * The key is set by the readpage_end_io_hook after the buffer has passed
105  * csum validation but before the pages are unlocked.  It is also set by
106  * btrfs_init_new_buffer on freshly allocated blocks.
107  *
108  * We also add a check to make sure the highest level of the tree is the
109  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
110  * needs update as well.
111  */
112 #ifdef CONFIG_DEBUG_LOCK_ALLOC
113 # if BTRFS_MAX_LEVEL != 8
114 #  error
115 # endif
116 
117 #define DEFINE_LEVEL(stem, level)					\
118 	.names[level] = "btrfs-" stem "-0" #level,
119 
120 #define DEFINE_NAME(stem)						\
121 	DEFINE_LEVEL(stem, 0)						\
122 	DEFINE_LEVEL(stem, 1)						\
123 	DEFINE_LEVEL(stem, 2)						\
124 	DEFINE_LEVEL(stem, 3)						\
125 	DEFINE_LEVEL(stem, 4)						\
126 	DEFINE_LEVEL(stem, 5)						\
127 	DEFINE_LEVEL(stem, 6)						\
128 	DEFINE_LEVEL(stem, 7)
129 
130 static struct btrfs_lockdep_keyset {
131 	u64			id;		/* root objectid */
132 	/* Longest entry: btrfs-free-space-00 */
133 	char			names[BTRFS_MAX_LEVEL][20];
134 	struct lock_class_key	keys[BTRFS_MAX_LEVEL];
135 } btrfs_lockdep_keysets[] = {
136 	{ .id = BTRFS_ROOT_TREE_OBJECTID,	DEFINE_NAME("root")	},
137 	{ .id = BTRFS_EXTENT_TREE_OBJECTID,	DEFINE_NAME("extent")	},
138 	{ .id = BTRFS_CHUNK_TREE_OBJECTID,	DEFINE_NAME("chunk")	},
139 	{ .id = BTRFS_DEV_TREE_OBJECTID,	DEFINE_NAME("dev")	},
140 	{ .id = BTRFS_CSUM_TREE_OBJECTID,	DEFINE_NAME("csum")	},
141 	{ .id = BTRFS_QUOTA_TREE_OBJECTID,	DEFINE_NAME("quota")	},
142 	{ .id = BTRFS_TREE_LOG_OBJECTID,	DEFINE_NAME("log")	},
143 	{ .id = BTRFS_TREE_RELOC_OBJECTID,	DEFINE_NAME("treloc")	},
144 	{ .id = BTRFS_DATA_RELOC_TREE_OBJECTID,	DEFINE_NAME("dreloc")	},
145 	{ .id = BTRFS_UUID_TREE_OBJECTID,	DEFINE_NAME("uuid")	},
146 	{ .id = BTRFS_FREE_SPACE_TREE_OBJECTID,	DEFINE_NAME("free-space") },
147 	{ .id = 0,				DEFINE_NAME("tree")	},
148 };
149 
150 #undef DEFINE_LEVEL
151 #undef DEFINE_NAME
152 
153 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
154 				    int level)
155 {
156 	struct btrfs_lockdep_keyset *ks;
157 
158 	BUG_ON(level >= ARRAY_SIZE(ks->keys));
159 
160 	/* find the matching keyset, id 0 is the default entry */
161 	for (ks = btrfs_lockdep_keysets; ks->id; ks++)
162 		if (ks->id == objectid)
163 			break;
164 
165 	lockdep_set_class_and_name(&eb->lock,
166 				   &ks->keys[level], ks->names[level]);
167 }
168 
169 #endif
170 
171 /*
172  * Compute the csum of a btree block and store the result to provided buffer.
173  */
174 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
175 {
176 	struct btrfs_fs_info *fs_info = buf->fs_info;
177 	const int num_pages = num_extent_pages(buf);
178 	const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
179 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
180 	char *kaddr;
181 	int i;
182 
183 	shash->tfm = fs_info->csum_shash;
184 	crypto_shash_init(shash);
185 	kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
186 	crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
187 			    first_page_part - BTRFS_CSUM_SIZE);
188 
189 	for (i = 1; i < num_pages; i++) {
190 		kaddr = page_address(buf->pages[i]);
191 		crypto_shash_update(shash, kaddr, PAGE_SIZE);
192 	}
193 	memset(result, 0, BTRFS_CSUM_SIZE);
194 	crypto_shash_final(shash, result);
195 }
196 
197 /*
198  * we can't consider a given block up to date unless the transid of the
199  * block matches the transid in the parent node's pointer.  This is how we
200  * detect blocks that either didn't get written at all or got written
201  * in the wrong place.
202  */
203 static int verify_parent_transid(struct extent_io_tree *io_tree,
204 				 struct extent_buffer *eb, u64 parent_transid,
205 				 int atomic)
206 {
207 	struct extent_state *cached_state = NULL;
208 	int ret;
209 
210 	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
211 		return 0;
212 
213 	if (atomic)
214 		return -EAGAIN;
215 
216 	lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
217 			 &cached_state);
218 	if (extent_buffer_uptodate(eb) &&
219 	    btrfs_header_generation(eb) == parent_transid) {
220 		ret = 0;
221 		goto out;
222 	}
223 	btrfs_err_rl(eb->fs_info,
224 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
225 			eb->start, eb->read_mirror,
226 			parent_transid, btrfs_header_generation(eb));
227 	ret = 1;
228 	clear_extent_buffer_uptodate(eb);
229 out:
230 	unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
231 			     &cached_state);
232 	return ret;
233 }
234 
235 static bool btrfs_supported_super_csum(u16 csum_type)
236 {
237 	switch (csum_type) {
238 	case BTRFS_CSUM_TYPE_CRC32:
239 	case BTRFS_CSUM_TYPE_XXHASH:
240 	case BTRFS_CSUM_TYPE_SHA256:
241 	case BTRFS_CSUM_TYPE_BLAKE2:
242 		return true;
243 	default:
244 		return false;
245 	}
246 }
247 
248 /*
249  * Return 0 if the superblock checksum type matches the checksum value of that
250  * algorithm. Pass the raw disk superblock data.
251  */
252 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
253 				  char *raw_disk_sb)
254 {
255 	struct btrfs_super_block *disk_sb =
256 		(struct btrfs_super_block *)raw_disk_sb;
257 	char result[BTRFS_CSUM_SIZE];
258 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
259 
260 	shash->tfm = fs_info->csum_shash;
261 
262 	/*
263 	 * The super_block structure does not span the whole
264 	 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
265 	 * filled with zeros and is included in the checksum.
266 	 */
267 	crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
268 			    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
269 
270 	if (memcmp(disk_sb->csum, result, fs_info->csum_size))
271 		return 1;
272 
273 	return 0;
274 }
275 
276 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
277 			   struct btrfs_key *first_key, u64 parent_transid)
278 {
279 	struct btrfs_fs_info *fs_info = eb->fs_info;
280 	int found_level;
281 	struct btrfs_key found_key;
282 	int ret;
283 
284 	found_level = btrfs_header_level(eb);
285 	if (found_level != level) {
286 		WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
287 		     KERN_ERR "BTRFS: tree level check failed\n");
288 		btrfs_err(fs_info,
289 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
290 			  eb->start, level, found_level);
291 		return -EIO;
292 	}
293 
294 	if (!first_key)
295 		return 0;
296 
297 	/*
298 	 * For live tree block (new tree blocks in current transaction),
299 	 * we need proper lock context to avoid race, which is impossible here.
300 	 * So we only checks tree blocks which is read from disk, whose
301 	 * generation <= fs_info->last_trans_committed.
302 	 */
303 	if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
304 		return 0;
305 
306 	/* We have @first_key, so this @eb must have at least one item */
307 	if (btrfs_header_nritems(eb) == 0) {
308 		btrfs_err(fs_info,
309 		"invalid tree nritems, bytenr=%llu nritems=0 expect >0",
310 			  eb->start);
311 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
312 		return -EUCLEAN;
313 	}
314 
315 	if (found_level)
316 		btrfs_node_key_to_cpu(eb, &found_key, 0);
317 	else
318 		btrfs_item_key_to_cpu(eb, &found_key, 0);
319 	ret = btrfs_comp_cpu_keys(first_key, &found_key);
320 
321 	if (ret) {
322 		WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
323 		     KERN_ERR "BTRFS: tree first key check failed\n");
324 		btrfs_err(fs_info,
325 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
326 			  eb->start, parent_transid, first_key->objectid,
327 			  first_key->type, first_key->offset,
328 			  found_key.objectid, found_key.type,
329 			  found_key.offset);
330 	}
331 	return ret;
332 }
333 
334 /*
335  * helper to read a given tree block, doing retries as required when
336  * the checksums don't match and we have alternate mirrors to try.
337  *
338  * @parent_transid:	expected transid, skip check if 0
339  * @level:		expected level, mandatory check
340  * @first_key:		expected key of first slot, skip check if NULL
341  */
342 int btrfs_read_extent_buffer(struct extent_buffer *eb,
343 			     u64 parent_transid, int level,
344 			     struct btrfs_key *first_key)
345 {
346 	struct btrfs_fs_info *fs_info = eb->fs_info;
347 	struct extent_io_tree *io_tree;
348 	int failed = 0;
349 	int ret;
350 	int num_copies = 0;
351 	int mirror_num = 0;
352 	int failed_mirror = 0;
353 
354 	io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
355 	while (1) {
356 		clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
357 		ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
358 		if (!ret) {
359 			if (verify_parent_transid(io_tree, eb,
360 						   parent_transid, 0))
361 				ret = -EIO;
362 			else if (btrfs_verify_level_key(eb, level,
363 						first_key, parent_transid))
364 				ret = -EUCLEAN;
365 			else
366 				break;
367 		}
368 
369 		num_copies = btrfs_num_copies(fs_info,
370 					      eb->start, eb->len);
371 		if (num_copies == 1)
372 			break;
373 
374 		if (!failed_mirror) {
375 			failed = 1;
376 			failed_mirror = eb->read_mirror;
377 		}
378 
379 		mirror_num++;
380 		if (mirror_num == failed_mirror)
381 			mirror_num++;
382 
383 		if (mirror_num > num_copies)
384 			break;
385 	}
386 
387 	if (failed && !ret && failed_mirror)
388 		btrfs_repair_eb_io_failure(eb, failed_mirror);
389 
390 	return ret;
391 }
392 
393 static int csum_one_extent_buffer(struct extent_buffer *eb)
394 {
395 	struct btrfs_fs_info *fs_info = eb->fs_info;
396 	u8 result[BTRFS_CSUM_SIZE];
397 	int ret;
398 
399 	ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
400 				    offsetof(struct btrfs_header, fsid),
401 				    BTRFS_FSID_SIZE) == 0);
402 	csum_tree_block(eb, result);
403 
404 	if (btrfs_header_level(eb))
405 		ret = btrfs_check_node(eb);
406 	else
407 		ret = btrfs_check_leaf_full(eb);
408 
409 	if (ret < 0)
410 		goto error;
411 
412 	/*
413 	 * Also check the generation, the eb reached here must be newer than
414 	 * last committed. Or something seriously wrong happened.
415 	 */
416 	if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) {
417 		ret = -EUCLEAN;
418 		btrfs_err(fs_info,
419 			"block=%llu bad generation, have %llu expect > %llu",
420 			  eb->start, btrfs_header_generation(eb),
421 			  fs_info->last_trans_committed);
422 		goto error;
423 	}
424 	write_extent_buffer(eb, result, 0, fs_info->csum_size);
425 
426 	return 0;
427 
428 error:
429 	btrfs_print_tree(eb, 0);
430 	btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
431 		  eb->start);
432 	WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
433 	return ret;
434 }
435 
436 /* Checksum all dirty extent buffers in one bio_vec */
437 static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
438 				      struct bio_vec *bvec)
439 {
440 	struct page *page = bvec->bv_page;
441 	u64 bvec_start = page_offset(page) + bvec->bv_offset;
442 	u64 cur;
443 	int ret = 0;
444 
445 	for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
446 	     cur += fs_info->nodesize) {
447 		struct extent_buffer *eb;
448 		bool uptodate;
449 
450 		eb = find_extent_buffer(fs_info, cur);
451 		uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
452 						       fs_info->nodesize);
453 
454 		/* A dirty eb shouldn't disappear from buffer_radix */
455 		if (WARN_ON(!eb))
456 			return -EUCLEAN;
457 
458 		if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
459 			free_extent_buffer(eb);
460 			return -EUCLEAN;
461 		}
462 		if (WARN_ON(!uptodate)) {
463 			free_extent_buffer(eb);
464 			return -EUCLEAN;
465 		}
466 
467 		ret = csum_one_extent_buffer(eb);
468 		free_extent_buffer(eb);
469 		if (ret < 0)
470 			return ret;
471 	}
472 	return ret;
473 }
474 
475 /*
476  * Checksum a dirty tree block before IO.  This has extra checks to make sure
477  * we only fill in the checksum field in the first page of a multi-page block.
478  * For subpage extent buffers we need bvec to also read the offset in the page.
479  */
480 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
481 {
482 	struct page *page = bvec->bv_page;
483 	u64 start = page_offset(page);
484 	u64 found_start;
485 	struct extent_buffer *eb;
486 
487 	if (fs_info->nodesize < PAGE_SIZE)
488 		return csum_dirty_subpage_buffers(fs_info, bvec);
489 
490 	eb = (struct extent_buffer *)page->private;
491 	if (page != eb->pages[0])
492 		return 0;
493 
494 	found_start = btrfs_header_bytenr(eb);
495 
496 	if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
497 		WARN_ON(found_start != 0);
498 		return 0;
499 	}
500 
501 	/*
502 	 * Please do not consolidate these warnings into a single if.
503 	 * It is useful to know what went wrong.
504 	 */
505 	if (WARN_ON(found_start != start))
506 		return -EUCLEAN;
507 	if (WARN_ON(!PageUptodate(page)))
508 		return -EUCLEAN;
509 
510 	return csum_one_extent_buffer(eb);
511 }
512 
513 static int check_tree_block_fsid(struct extent_buffer *eb)
514 {
515 	struct btrfs_fs_info *fs_info = eb->fs_info;
516 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
517 	u8 fsid[BTRFS_FSID_SIZE];
518 	u8 *metadata_uuid;
519 
520 	read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
521 			   BTRFS_FSID_SIZE);
522 	/*
523 	 * Checking the incompat flag is only valid for the current fs. For
524 	 * seed devices it's forbidden to have their uuid changed so reading
525 	 * ->fsid in this case is fine
526 	 */
527 	if (btrfs_fs_incompat(fs_info, METADATA_UUID))
528 		metadata_uuid = fs_devices->metadata_uuid;
529 	else
530 		metadata_uuid = fs_devices->fsid;
531 
532 	if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
533 		return 0;
534 
535 	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
536 		if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
537 			return 0;
538 
539 	return 1;
540 }
541 
542 /* Do basic extent buffer checks at read time */
543 static int validate_extent_buffer(struct extent_buffer *eb)
544 {
545 	struct btrfs_fs_info *fs_info = eb->fs_info;
546 	u64 found_start;
547 	const u32 csum_size = fs_info->csum_size;
548 	u8 found_level;
549 	u8 result[BTRFS_CSUM_SIZE];
550 	const u8 *header_csum;
551 	int ret = 0;
552 
553 	found_start = btrfs_header_bytenr(eb);
554 	if (found_start != eb->start) {
555 		btrfs_err_rl(fs_info,
556 			"bad tree block start, mirror %u want %llu have %llu",
557 			     eb->read_mirror, eb->start, found_start);
558 		ret = -EIO;
559 		goto out;
560 	}
561 	if (check_tree_block_fsid(eb)) {
562 		btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
563 			     eb->start, eb->read_mirror);
564 		ret = -EIO;
565 		goto out;
566 	}
567 	found_level = btrfs_header_level(eb);
568 	if (found_level >= BTRFS_MAX_LEVEL) {
569 		btrfs_err(fs_info,
570 			"bad tree block level, mirror %u level %d on logical %llu",
571 			eb->read_mirror, btrfs_header_level(eb), eb->start);
572 		ret = -EIO;
573 		goto out;
574 	}
575 
576 	csum_tree_block(eb, result);
577 	header_csum = page_address(eb->pages[0]) +
578 		get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
579 
580 	if (memcmp(result, header_csum, csum_size) != 0) {
581 		btrfs_warn_rl(fs_info,
582 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
583 			      eb->start, eb->read_mirror,
584 			      CSUM_FMT_VALUE(csum_size, header_csum),
585 			      CSUM_FMT_VALUE(csum_size, result),
586 			      btrfs_header_level(eb));
587 		ret = -EUCLEAN;
588 		goto out;
589 	}
590 
591 	/*
592 	 * If this is a leaf block and it is corrupt, set the corrupt bit so
593 	 * that we don't try and read the other copies of this block, just
594 	 * return -EIO.
595 	 */
596 	if (found_level == 0 && btrfs_check_leaf_full(eb)) {
597 		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
598 		ret = -EIO;
599 	}
600 
601 	if (found_level > 0 && btrfs_check_node(eb))
602 		ret = -EIO;
603 
604 	if (!ret)
605 		set_extent_buffer_uptodate(eb);
606 	else
607 		btrfs_err(fs_info,
608 		"read time tree block corruption detected on logical %llu mirror %u",
609 			  eb->start, eb->read_mirror);
610 out:
611 	return ret;
612 }
613 
614 static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
615 				   int mirror)
616 {
617 	struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
618 	struct extent_buffer *eb;
619 	bool reads_done;
620 	int ret = 0;
621 
622 	/*
623 	 * We don't allow bio merge for subpage metadata read, so we should
624 	 * only get one eb for each endio hook.
625 	 */
626 	ASSERT(end == start + fs_info->nodesize - 1);
627 	ASSERT(PagePrivate(page));
628 
629 	eb = find_extent_buffer(fs_info, start);
630 	/*
631 	 * When we are reading one tree block, eb must have been inserted into
632 	 * the radix tree. If not, something is wrong.
633 	 */
634 	ASSERT(eb);
635 
636 	reads_done = atomic_dec_and_test(&eb->io_pages);
637 	/* Subpage read must finish in page read */
638 	ASSERT(reads_done);
639 
640 	eb->read_mirror = mirror;
641 	if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
642 		ret = -EIO;
643 		goto err;
644 	}
645 	ret = validate_extent_buffer(eb);
646 	if (ret < 0)
647 		goto err;
648 
649 	set_extent_buffer_uptodate(eb);
650 
651 	free_extent_buffer(eb);
652 	return ret;
653 err:
654 	/*
655 	 * end_bio_extent_readpage decrements io_pages in case of error,
656 	 * make sure it has something to decrement.
657 	 */
658 	atomic_inc(&eb->io_pages);
659 	clear_extent_buffer_uptodate(eb);
660 	free_extent_buffer(eb);
661 	return ret;
662 }
663 
664 int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio,
665 				   struct page *page, u64 start, u64 end,
666 				   int mirror)
667 {
668 	struct extent_buffer *eb;
669 	int ret = 0;
670 	int reads_done;
671 
672 	ASSERT(page->private);
673 
674 	if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
675 		return validate_subpage_buffer(page, start, end, mirror);
676 
677 	eb = (struct extent_buffer *)page->private;
678 
679 	/*
680 	 * The pending IO might have been the only thing that kept this buffer
681 	 * in memory.  Make sure we have a ref for all this other checks
682 	 */
683 	atomic_inc(&eb->refs);
684 
685 	reads_done = atomic_dec_and_test(&eb->io_pages);
686 	if (!reads_done)
687 		goto err;
688 
689 	eb->read_mirror = mirror;
690 	if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
691 		ret = -EIO;
692 		goto err;
693 	}
694 	ret = validate_extent_buffer(eb);
695 err:
696 	if (ret) {
697 		/*
698 		 * our io error hook is going to dec the io pages
699 		 * again, we have to make sure it has something
700 		 * to decrement
701 		 */
702 		atomic_inc(&eb->io_pages);
703 		clear_extent_buffer_uptodate(eb);
704 	}
705 	free_extent_buffer(eb);
706 
707 	return ret;
708 }
709 
710 static void run_one_async_start(struct btrfs_work *work)
711 {
712 	struct async_submit_bio *async;
713 	blk_status_t ret;
714 
715 	async = container_of(work, struct  async_submit_bio, work);
716 	ret = async->submit_bio_start(async->inode, async->bio,
717 				      async->dio_file_offset);
718 	if (ret)
719 		async->status = ret;
720 }
721 
722 /*
723  * In order to insert checksums into the metadata in large chunks, we wait
724  * until bio submission time.   All the pages in the bio are checksummed and
725  * sums are attached onto the ordered extent record.
726  *
727  * At IO completion time the csums attached on the ordered extent record are
728  * inserted into the tree.
729  */
730 static void run_one_async_done(struct btrfs_work *work)
731 {
732 	struct async_submit_bio *async;
733 	struct inode *inode;
734 
735 	async = container_of(work, struct  async_submit_bio, work);
736 	inode = async->inode;
737 
738 	/* If an error occurred we just want to clean up the bio and move on */
739 	if (async->status) {
740 		async->bio->bi_status = async->status;
741 		bio_endio(async->bio);
742 		return;
743 	}
744 
745 	/*
746 	 * All of the bios that pass through here are from async helpers.
747 	 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
748 	 * This changes nothing when cgroups aren't in use.
749 	 */
750 	async->bio->bi_opf |= REQ_CGROUP_PUNT;
751 	btrfs_submit_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
752 }
753 
754 static void run_one_async_free(struct btrfs_work *work)
755 {
756 	struct async_submit_bio *async;
757 
758 	async = container_of(work, struct  async_submit_bio, work);
759 	kfree(async);
760 }
761 
762 /*
763  * Submit bio to an async queue.
764  *
765  * Retrun:
766  * - true if the work has been succesfuly submitted
767  * - false in case of error
768  */
769 bool btrfs_wq_submit_bio(struct inode *inode, struct bio *bio, int mirror_num,
770 			 u64 dio_file_offset,
771 			 extent_submit_bio_start_t *submit_bio_start)
772 {
773 	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
774 	struct async_submit_bio *async;
775 
776 	async = kmalloc(sizeof(*async), GFP_NOFS);
777 	if (!async)
778 		return false;
779 
780 	async->inode = inode;
781 	async->bio = bio;
782 	async->mirror_num = mirror_num;
783 	async->submit_bio_start = submit_bio_start;
784 
785 	btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
786 			run_one_async_free);
787 
788 	async->dio_file_offset = dio_file_offset;
789 
790 	async->status = 0;
791 
792 	if (op_is_sync(bio->bi_opf))
793 		btrfs_queue_work(fs_info->hipri_workers, &async->work);
794 	else
795 		btrfs_queue_work(fs_info->workers, &async->work);
796 	return true;
797 }
798 
799 static blk_status_t btree_csum_one_bio(struct bio *bio)
800 {
801 	struct bio_vec *bvec;
802 	struct btrfs_root *root;
803 	int ret = 0;
804 	struct bvec_iter_all iter_all;
805 
806 	ASSERT(!bio_flagged(bio, BIO_CLONED));
807 	bio_for_each_segment_all(bvec, bio, iter_all) {
808 		root = BTRFS_I(bvec->bv_page->mapping->host)->root;
809 		ret = csum_dirty_buffer(root->fs_info, bvec);
810 		if (ret)
811 			break;
812 	}
813 
814 	return errno_to_blk_status(ret);
815 }
816 
817 static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
818 					   u64 dio_file_offset)
819 {
820 	/*
821 	 * when we're called for a write, we're already in the async
822 	 * submission context.  Just jump into btrfs_submit_bio.
823 	 */
824 	return btree_csum_one_bio(bio);
825 }
826 
827 static bool should_async_write(struct btrfs_fs_info *fs_info,
828 			     struct btrfs_inode *bi)
829 {
830 	if (btrfs_is_zoned(fs_info))
831 		return false;
832 	if (atomic_read(&bi->sync_writers))
833 		return false;
834 	if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
835 		return false;
836 	return true;
837 }
838 
839 void btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, int mirror_num)
840 {
841 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
842 	blk_status_t ret;
843 
844 	bio->bi_opf |= REQ_META;
845 
846 	if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
847 		btrfs_submit_bio(fs_info, bio, mirror_num);
848 		return;
849 	}
850 
851 	/*
852 	 * Kthread helpers are used to submit writes so that checksumming can
853 	 * happen in parallel across all CPUs.
854 	 */
855 	if (should_async_write(fs_info, BTRFS_I(inode)) &&
856 	    btrfs_wq_submit_bio(inode, bio, mirror_num, 0, btree_submit_bio_start))
857 		return;
858 
859 	ret = btree_csum_one_bio(bio);
860 	if (ret) {
861 		bio->bi_status = ret;
862 		bio_endio(bio);
863 		return;
864 	}
865 
866 	btrfs_submit_bio(fs_info, bio, mirror_num);
867 }
868 
869 #ifdef CONFIG_MIGRATION
870 static int btree_migrate_folio(struct address_space *mapping,
871 		struct folio *dst, struct folio *src, enum migrate_mode mode)
872 {
873 	/*
874 	 * we can't safely write a btree page from here,
875 	 * we haven't done the locking hook
876 	 */
877 	if (folio_test_dirty(src))
878 		return -EAGAIN;
879 	/*
880 	 * Buffers may be managed in a filesystem specific way.
881 	 * We must have no buffers or drop them.
882 	 */
883 	if (folio_get_private(src) &&
884 	    !filemap_release_folio(src, GFP_KERNEL))
885 		return -EAGAIN;
886 	return migrate_folio(mapping, dst, src, mode);
887 }
888 #else
889 #define btree_migrate_folio NULL
890 #endif
891 
892 static int btree_writepages(struct address_space *mapping,
893 			    struct writeback_control *wbc)
894 {
895 	struct btrfs_fs_info *fs_info;
896 	int ret;
897 
898 	if (wbc->sync_mode == WB_SYNC_NONE) {
899 
900 		if (wbc->for_kupdate)
901 			return 0;
902 
903 		fs_info = BTRFS_I(mapping->host)->root->fs_info;
904 		/* this is a bit racy, but that's ok */
905 		ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
906 					     BTRFS_DIRTY_METADATA_THRESH,
907 					     fs_info->dirty_metadata_batch);
908 		if (ret < 0)
909 			return 0;
910 	}
911 	return btree_write_cache_pages(mapping, wbc);
912 }
913 
914 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
915 {
916 	if (folio_test_writeback(folio) || folio_test_dirty(folio))
917 		return false;
918 
919 	return try_release_extent_buffer(&folio->page);
920 }
921 
922 static void btree_invalidate_folio(struct folio *folio, size_t offset,
923 				 size_t length)
924 {
925 	struct extent_io_tree *tree;
926 	tree = &BTRFS_I(folio->mapping->host)->io_tree;
927 	extent_invalidate_folio(tree, folio, offset);
928 	btree_release_folio(folio, GFP_NOFS);
929 	if (folio_get_private(folio)) {
930 		btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
931 			   "folio private not zero on folio %llu",
932 			   (unsigned long long)folio_pos(folio));
933 		folio_detach_private(folio);
934 	}
935 }
936 
937 #ifdef DEBUG
938 static bool btree_dirty_folio(struct address_space *mapping,
939 		struct folio *folio)
940 {
941 	struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
942 	struct btrfs_subpage *subpage;
943 	struct extent_buffer *eb;
944 	int cur_bit = 0;
945 	u64 page_start = folio_pos(folio);
946 
947 	if (fs_info->sectorsize == PAGE_SIZE) {
948 		eb = folio_get_private(folio);
949 		BUG_ON(!eb);
950 		BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
951 		BUG_ON(!atomic_read(&eb->refs));
952 		btrfs_assert_tree_write_locked(eb);
953 		return filemap_dirty_folio(mapping, folio);
954 	}
955 	subpage = folio_get_private(folio);
956 
957 	ASSERT(subpage->dirty_bitmap);
958 	while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
959 		unsigned long flags;
960 		u64 cur;
961 		u16 tmp = (1 << cur_bit);
962 
963 		spin_lock_irqsave(&subpage->lock, flags);
964 		if (!(tmp & subpage->dirty_bitmap)) {
965 			spin_unlock_irqrestore(&subpage->lock, flags);
966 			cur_bit++;
967 			continue;
968 		}
969 		spin_unlock_irqrestore(&subpage->lock, flags);
970 		cur = page_start + cur_bit * fs_info->sectorsize;
971 
972 		eb = find_extent_buffer(fs_info, cur);
973 		ASSERT(eb);
974 		ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
975 		ASSERT(atomic_read(&eb->refs));
976 		btrfs_assert_tree_write_locked(eb);
977 		free_extent_buffer(eb);
978 
979 		cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
980 	}
981 	return filemap_dirty_folio(mapping, folio);
982 }
983 #else
984 #define btree_dirty_folio filemap_dirty_folio
985 #endif
986 
987 static const struct address_space_operations btree_aops = {
988 	.writepages	= btree_writepages,
989 	.release_folio	= btree_release_folio,
990 	.invalidate_folio = btree_invalidate_folio,
991 	.migrate_folio	= btree_migrate_folio,
992 	.dirty_folio	= btree_dirty_folio,
993 };
994 
995 struct extent_buffer *btrfs_find_create_tree_block(
996 						struct btrfs_fs_info *fs_info,
997 						u64 bytenr, u64 owner_root,
998 						int level)
999 {
1000 	if (btrfs_is_testing(fs_info))
1001 		return alloc_test_extent_buffer(fs_info, bytenr);
1002 	return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
1003 }
1004 
1005 /*
1006  * Read tree block at logical address @bytenr and do variant basic but critical
1007  * verification.
1008  *
1009  * @owner_root:		the objectid of the root owner for this block.
1010  * @parent_transid:	expected transid of this tree block, skip check if 0
1011  * @level:		expected level, mandatory check
1012  * @first_key:		expected key in slot 0, skip check if NULL
1013  */
1014 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1015 				      u64 owner_root, u64 parent_transid,
1016 				      int level, struct btrfs_key *first_key)
1017 {
1018 	struct extent_buffer *buf = NULL;
1019 	int ret;
1020 
1021 	buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
1022 	if (IS_ERR(buf))
1023 		return buf;
1024 
1025 	ret = btrfs_read_extent_buffer(buf, parent_transid, level, first_key);
1026 	if (ret) {
1027 		free_extent_buffer_stale(buf);
1028 		return ERR_PTR(ret);
1029 	}
1030 	if (btrfs_check_eb_owner(buf, owner_root)) {
1031 		free_extent_buffer_stale(buf);
1032 		return ERR_PTR(-EUCLEAN);
1033 	}
1034 	return buf;
1035 
1036 }
1037 
1038 void btrfs_clean_tree_block(struct extent_buffer *buf)
1039 {
1040 	struct btrfs_fs_info *fs_info = buf->fs_info;
1041 	if (btrfs_header_generation(buf) ==
1042 	    fs_info->running_transaction->transid) {
1043 		btrfs_assert_tree_write_locked(buf);
1044 
1045 		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1046 			percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1047 						 -buf->len,
1048 						 fs_info->dirty_metadata_batch);
1049 			clear_extent_buffer_dirty(buf);
1050 		}
1051 	}
1052 }
1053 
1054 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1055 			 u64 objectid)
1056 {
1057 	bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1058 
1059 	memset(&root->root_key, 0, sizeof(root->root_key));
1060 	memset(&root->root_item, 0, sizeof(root->root_item));
1061 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1062 	root->fs_info = fs_info;
1063 	root->root_key.objectid = objectid;
1064 	root->node = NULL;
1065 	root->commit_root = NULL;
1066 	root->state = 0;
1067 	RB_CLEAR_NODE(&root->rb_node);
1068 
1069 	root->last_trans = 0;
1070 	root->free_objectid = 0;
1071 	root->nr_delalloc_inodes = 0;
1072 	root->nr_ordered_extents = 0;
1073 	root->inode_tree = RB_ROOT;
1074 	INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1075 
1076 	btrfs_init_root_block_rsv(root);
1077 
1078 	INIT_LIST_HEAD(&root->dirty_list);
1079 	INIT_LIST_HEAD(&root->root_list);
1080 	INIT_LIST_HEAD(&root->delalloc_inodes);
1081 	INIT_LIST_HEAD(&root->delalloc_root);
1082 	INIT_LIST_HEAD(&root->ordered_extents);
1083 	INIT_LIST_HEAD(&root->ordered_root);
1084 	INIT_LIST_HEAD(&root->reloc_dirty_list);
1085 	INIT_LIST_HEAD(&root->logged_list[0]);
1086 	INIT_LIST_HEAD(&root->logged_list[1]);
1087 	spin_lock_init(&root->inode_lock);
1088 	spin_lock_init(&root->delalloc_lock);
1089 	spin_lock_init(&root->ordered_extent_lock);
1090 	spin_lock_init(&root->accounting_lock);
1091 	spin_lock_init(&root->log_extents_lock[0]);
1092 	spin_lock_init(&root->log_extents_lock[1]);
1093 	spin_lock_init(&root->qgroup_meta_rsv_lock);
1094 	mutex_init(&root->objectid_mutex);
1095 	mutex_init(&root->log_mutex);
1096 	mutex_init(&root->ordered_extent_mutex);
1097 	mutex_init(&root->delalloc_mutex);
1098 	init_waitqueue_head(&root->qgroup_flush_wait);
1099 	init_waitqueue_head(&root->log_writer_wait);
1100 	init_waitqueue_head(&root->log_commit_wait[0]);
1101 	init_waitqueue_head(&root->log_commit_wait[1]);
1102 	INIT_LIST_HEAD(&root->log_ctxs[0]);
1103 	INIT_LIST_HEAD(&root->log_ctxs[1]);
1104 	atomic_set(&root->log_commit[0], 0);
1105 	atomic_set(&root->log_commit[1], 0);
1106 	atomic_set(&root->log_writers, 0);
1107 	atomic_set(&root->log_batch, 0);
1108 	refcount_set(&root->refs, 1);
1109 	atomic_set(&root->snapshot_force_cow, 0);
1110 	atomic_set(&root->nr_swapfiles, 0);
1111 	root->log_transid = 0;
1112 	root->log_transid_committed = -1;
1113 	root->last_log_commit = 0;
1114 	root->anon_dev = 0;
1115 	if (!dummy) {
1116 		extent_io_tree_init(fs_info, &root->dirty_log_pages,
1117 				    IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1118 		extent_io_tree_init(fs_info, &root->log_csum_range,
1119 				    IO_TREE_LOG_CSUM_RANGE, NULL);
1120 	}
1121 
1122 	spin_lock_init(&root->root_item_lock);
1123 	btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1124 #ifdef CONFIG_BTRFS_DEBUG
1125 	INIT_LIST_HEAD(&root->leak_list);
1126 	spin_lock(&fs_info->fs_roots_radix_lock);
1127 	list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1128 	spin_unlock(&fs_info->fs_roots_radix_lock);
1129 #endif
1130 }
1131 
1132 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1133 					   u64 objectid, gfp_t flags)
1134 {
1135 	struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1136 	if (root)
1137 		__setup_root(root, fs_info, objectid);
1138 	return root;
1139 }
1140 
1141 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1142 /* Should only be used by the testing infrastructure */
1143 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1144 {
1145 	struct btrfs_root *root;
1146 
1147 	if (!fs_info)
1148 		return ERR_PTR(-EINVAL);
1149 
1150 	root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1151 	if (!root)
1152 		return ERR_PTR(-ENOMEM);
1153 
1154 	/* We don't use the stripesize in selftest, set it as sectorsize */
1155 	root->alloc_bytenr = 0;
1156 
1157 	return root;
1158 }
1159 #endif
1160 
1161 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
1162 {
1163 	const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
1164 	const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
1165 
1166 	return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
1167 }
1168 
1169 static int global_root_key_cmp(const void *k, const struct rb_node *node)
1170 {
1171 	const struct btrfs_key *key = k;
1172 	const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
1173 
1174 	return btrfs_comp_cpu_keys(key, &root->root_key);
1175 }
1176 
1177 int btrfs_global_root_insert(struct btrfs_root *root)
1178 {
1179 	struct btrfs_fs_info *fs_info = root->fs_info;
1180 	struct rb_node *tmp;
1181 
1182 	write_lock(&fs_info->global_root_lock);
1183 	tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
1184 	write_unlock(&fs_info->global_root_lock);
1185 	ASSERT(!tmp);
1186 
1187 	return tmp ? -EEXIST : 0;
1188 }
1189 
1190 void btrfs_global_root_delete(struct btrfs_root *root)
1191 {
1192 	struct btrfs_fs_info *fs_info = root->fs_info;
1193 
1194 	write_lock(&fs_info->global_root_lock);
1195 	rb_erase(&root->rb_node, &fs_info->global_root_tree);
1196 	write_unlock(&fs_info->global_root_lock);
1197 }
1198 
1199 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
1200 				     struct btrfs_key *key)
1201 {
1202 	struct rb_node *node;
1203 	struct btrfs_root *root = NULL;
1204 
1205 	read_lock(&fs_info->global_root_lock);
1206 	node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
1207 	if (node)
1208 		root = container_of(node, struct btrfs_root, rb_node);
1209 	read_unlock(&fs_info->global_root_lock);
1210 
1211 	return root;
1212 }
1213 
1214 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
1215 {
1216 	struct btrfs_block_group *block_group;
1217 	u64 ret;
1218 
1219 	if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
1220 		return 0;
1221 
1222 	if (bytenr)
1223 		block_group = btrfs_lookup_block_group(fs_info, bytenr);
1224 	else
1225 		block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
1226 	ASSERT(block_group);
1227 	if (!block_group)
1228 		return 0;
1229 	ret = block_group->global_root_id;
1230 	btrfs_put_block_group(block_group);
1231 
1232 	return ret;
1233 }
1234 
1235 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1236 {
1237 	struct btrfs_key key = {
1238 		.objectid = BTRFS_CSUM_TREE_OBJECTID,
1239 		.type = BTRFS_ROOT_ITEM_KEY,
1240 		.offset = btrfs_global_root_id(fs_info, bytenr),
1241 	};
1242 
1243 	return btrfs_global_root(fs_info, &key);
1244 }
1245 
1246 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1247 {
1248 	struct btrfs_key key = {
1249 		.objectid = BTRFS_EXTENT_TREE_OBJECTID,
1250 		.type = BTRFS_ROOT_ITEM_KEY,
1251 		.offset = btrfs_global_root_id(fs_info, bytenr),
1252 	};
1253 
1254 	return btrfs_global_root(fs_info, &key);
1255 }
1256 
1257 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1258 				     u64 objectid)
1259 {
1260 	struct btrfs_fs_info *fs_info = trans->fs_info;
1261 	struct extent_buffer *leaf;
1262 	struct btrfs_root *tree_root = fs_info->tree_root;
1263 	struct btrfs_root *root;
1264 	struct btrfs_key key;
1265 	unsigned int nofs_flag;
1266 	int ret = 0;
1267 
1268 	/*
1269 	 * We're holding a transaction handle, so use a NOFS memory allocation
1270 	 * context to avoid deadlock if reclaim happens.
1271 	 */
1272 	nofs_flag = memalloc_nofs_save();
1273 	root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1274 	memalloc_nofs_restore(nofs_flag);
1275 	if (!root)
1276 		return ERR_PTR(-ENOMEM);
1277 
1278 	root->root_key.objectid = objectid;
1279 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1280 	root->root_key.offset = 0;
1281 
1282 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1283 				      BTRFS_NESTING_NORMAL);
1284 	if (IS_ERR(leaf)) {
1285 		ret = PTR_ERR(leaf);
1286 		leaf = NULL;
1287 		goto fail_unlock;
1288 	}
1289 
1290 	root->node = leaf;
1291 	btrfs_mark_buffer_dirty(leaf);
1292 
1293 	root->commit_root = btrfs_root_node(root);
1294 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1295 
1296 	btrfs_set_root_flags(&root->root_item, 0);
1297 	btrfs_set_root_limit(&root->root_item, 0);
1298 	btrfs_set_root_bytenr(&root->root_item, leaf->start);
1299 	btrfs_set_root_generation(&root->root_item, trans->transid);
1300 	btrfs_set_root_level(&root->root_item, 0);
1301 	btrfs_set_root_refs(&root->root_item, 1);
1302 	btrfs_set_root_used(&root->root_item, leaf->len);
1303 	btrfs_set_root_last_snapshot(&root->root_item, 0);
1304 	btrfs_set_root_dirid(&root->root_item, 0);
1305 	if (is_fstree(objectid))
1306 		generate_random_guid(root->root_item.uuid);
1307 	else
1308 		export_guid(root->root_item.uuid, &guid_null);
1309 	btrfs_set_root_drop_level(&root->root_item, 0);
1310 
1311 	btrfs_tree_unlock(leaf);
1312 
1313 	key.objectid = objectid;
1314 	key.type = BTRFS_ROOT_ITEM_KEY;
1315 	key.offset = 0;
1316 	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1317 	if (ret)
1318 		goto fail;
1319 
1320 	return root;
1321 
1322 fail_unlock:
1323 	if (leaf)
1324 		btrfs_tree_unlock(leaf);
1325 fail:
1326 	btrfs_put_root(root);
1327 
1328 	return ERR_PTR(ret);
1329 }
1330 
1331 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1332 					 struct btrfs_fs_info *fs_info)
1333 {
1334 	struct btrfs_root *root;
1335 
1336 	root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1337 	if (!root)
1338 		return ERR_PTR(-ENOMEM);
1339 
1340 	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1341 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1342 	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1343 
1344 	return root;
1345 }
1346 
1347 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1348 			      struct btrfs_root *root)
1349 {
1350 	struct extent_buffer *leaf;
1351 
1352 	/*
1353 	 * DON'T set SHAREABLE bit for log trees.
1354 	 *
1355 	 * Log trees are not exposed to user space thus can't be snapshotted,
1356 	 * and they go away before a real commit is actually done.
1357 	 *
1358 	 * They do store pointers to file data extents, and those reference
1359 	 * counts still get updated (along with back refs to the log tree).
1360 	 */
1361 
1362 	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1363 			NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1364 	if (IS_ERR(leaf))
1365 		return PTR_ERR(leaf);
1366 
1367 	root->node = leaf;
1368 
1369 	btrfs_mark_buffer_dirty(root->node);
1370 	btrfs_tree_unlock(root->node);
1371 
1372 	return 0;
1373 }
1374 
1375 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1376 			     struct btrfs_fs_info *fs_info)
1377 {
1378 	struct btrfs_root *log_root;
1379 
1380 	log_root = alloc_log_tree(trans, fs_info);
1381 	if (IS_ERR(log_root))
1382 		return PTR_ERR(log_root);
1383 
1384 	if (!btrfs_is_zoned(fs_info)) {
1385 		int ret = btrfs_alloc_log_tree_node(trans, log_root);
1386 
1387 		if (ret) {
1388 			btrfs_put_root(log_root);
1389 			return ret;
1390 		}
1391 	}
1392 
1393 	WARN_ON(fs_info->log_root_tree);
1394 	fs_info->log_root_tree = log_root;
1395 	return 0;
1396 }
1397 
1398 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1399 		       struct btrfs_root *root)
1400 {
1401 	struct btrfs_fs_info *fs_info = root->fs_info;
1402 	struct btrfs_root *log_root;
1403 	struct btrfs_inode_item *inode_item;
1404 	int ret;
1405 
1406 	log_root = alloc_log_tree(trans, fs_info);
1407 	if (IS_ERR(log_root))
1408 		return PTR_ERR(log_root);
1409 
1410 	ret = btrfs_alloc_log_tree_node(trans, log_root);
1411 	if (ret) {
1412 		btrfs_put_root(log_root);
1413 		return ret;
1414 	}
1415 
1416 	log_root->last_trans = trans->transid;
1417 	log_root->root_key.offset = root->root_key.objectid;
1418 
1419 	inode_item = &log_root->root_item.inode;
1420 	btrfs_set_stack_inode_generation(inode_item, 1);
1421 	btrfs_set_stack_inode_size(inode_item, 3);
1422 	btrfs_set_stack_inode_nlink(inode_item, 1);
1423 	btrfs_set_stack_inode_nbytes(inode_item,
1424 				     fs_info->nodesize);
1425 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1426 
1427 	btrfs_set_root_node(&log_root->root_item, log_root->node);
1428 
1429 	WARN_ON(root->log_root);
1430 	root->log_root = log_root;
1431 	root->log_transid = 0;
1432 	root->log_transid_committed = -1;
1433 	root->last_log_commit = 0;
1434 	return 0;
1435 }
1436 
1437 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1438 					      struct btrfs_path *path,
1439 					      struct btrfs_key *key)
1440 {
1441 	struct btrfs_root *root;
1442 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1443 	u64 generation;
1444 	int ret;
1445 	int level;
1446 
1447 	root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1448 	if (!root)
1449 		return ERR_PTR(-ENOMEM);
1450 
1451 	ret = btrfs_find_root(tree_root, key, path,
1452 			      &root->root_item, &root->root_key);
1453 	if (ret) {
1454 		if (ret > 0)
1455 			ret = -ENOENT;
1456 		goto fail;
1457 	}
1458 
1459 	generation = btrfs_root_generation(&root->root_item);
1460 	level = btrfs_root_level(&root->root_item);
1461 	root->node = read_tree_block(fs_info,
1462 				     btrfs_root_bytenr(&root->root_item),
1463 				     key->objectid, generation, level, NULL);
1464 	if (IS_ERR(root->node)) {
1465 		ret = PTR_ERR(root->node);
1466 		root->node = NULL;
1467 		goto fail;
1468 	}
1469 	if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1470 		ret = -EIO;
1471 		goto fail;
1472 	}
1473 
1474 	/*
1475 	 * For real fs, and not log/reloc trees, root owner must
1476 	 * match its root node owner
1477 	 */
1478 	if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1479 	    root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1480 	    root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1481 	    root->root_key.objectid != btrfs_header_owner(root->node)) {
1482 		btrfs_crit(fs_info,
1483 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1484 			   root->root_key.objectid, root->node->start,
1485 			   btrfs_header_owner(root->node),
1486 			   root->root_key.objectid);
1487 		ret = -EUCLEAN;
1488 		goto fail;
1489 	}
1490 	root->commit_root = btrfs_root_node(root);
1491 	return root;
1492 fail:
1493 	btrfs_put_root(root);
1494 	return ERR_PTR(ret);
1495 }
1496 
1497 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1498 					struct btrfs_key *key)
1499 {
1500 	struct btrfs_root *root;
1501 	struct btrfs_path *path;
1502 
1503 	path = btrfs_alloc_path();
1504 	if (!path)
1505 		return ERR_PTR(-ENOMEM);
1506 	root = read_tree_root_path(tree_root, path, key);
1507 	btrfs_free_path(path);
1508 
1509 	return root;
1510 }
1511 
1512 /*
1513  * Initialize subvolume root in-memory structure
1514  *
1515  * @anon_dev:	anonymous device to attach to the root, if zero, allocate new
1516  */
1517 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1518 {
1519 	int ret;
1520 	unsigned int nofs_flag;
1521 
1522 	/*
1523 	 * We might be called under a transaction (e.g. indirect backref
1524 	 * resolution) which could deadlock if it triggers memory reclaim
1525 	 */
1526 	nofs_flag = memalloc_nofs_save();
1527 	ret = btrfs_drew_lock_init(&root->snapshot_lock);
1528 	memalloc_nofs_restore(nofs_flag);
1529 	if (ret)
1530 		goto fail;
1531 
1532 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1533 	    !btrfs_is_data_reloc_root(root)) {
1534 		set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1535 		btrfs_check_and_init_root_item(&root->root_item);
1536 	}
1537 
1538 	/*
1539 	 * Don't assign anonymous block device to roots that are not exposed to
1540 	 * userspace, the id pool is limited to 1M
1541 	 */
1542 	if (is_fstree(root->root_key.objectid) &&
1543 	    btrfs_root_refs(&root->root_item) > 0) {
1544 		if (!anon_dev) {
1545 			ret = get_anon_bdev(&root->anon_dev);
1546 			if (ret)
1547 				goto fail;
1548 		} else {
1549 			root->anon_dev = anon_dev;
1550 		}
1551 	}
1552 
1553 	mutex_lock(&root->objectid_mutex);
1554 	ret = btrfs_init_root_free_objectid(root);
1555 	if (ret) {
1556 		mutex_unlock(&root->objectid_mutex);
1557 		goto fail;
1558 	}
1559 
1560 	ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1561 
1562 	mutex_unlock(&root->objectid_mutex);
1563 
1564 	return 0;
1565 fail:
1566 	/* The caller is responsible to call btrfs_free_fs_root */
1567 	return ret;
1568 }
1569 
1570 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1571 					       u64 root_id)
1572 {
1573 	struct btrfs_root *root;
1574 
1575 	spin_lock(&fs_info->fs_roots_radix_lock);
1576 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1577 				 (unsigned long)root_id);
1578 	if (root)
1579 		root = btrfs_grab_root(root);
1580 	spin_unlock(&fs_info->fs_roots_radix_lock);
1581 	return root;
1582 }
1583 
1584 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1585 						u64 objectid)
1586 {
1587 	struct btrfs_key key = {
1588 		.objectid = objectid,
1589 		.type = BTRFS_ROOT_ITEM_KEY,
1590 		.offset = 0,
1591 	};
1592 
1593 	if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1594 		return btrfs_grab_root(fs_info->tree_root);
1595 	if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1596 		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1597 	if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1598 		return btrfs_grab_root(fs_info->chunk_root);
1599 	if (objectid == BTRFS_DEV_TREE_OBJECTID)
1600 		return btrfs_grab_root(fs_info->dev_root);
1601 	if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1602 		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1603 	if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1604 		return btrfs_grab_root(fs_info->quota_root) ?
1605 			fs_info->quota_root : ERR_PTR(-ENOENT);
1606 	if (objectid == BTRFS_UUID_TREE_OBJECTID)
1607 		return btrfs_grab_root(fs_info->uuid_root) ?
1608 			fs_info->uuid_root : ERR_PTR(-ENOENT);
1609 	if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) {
1610 		struct btrfs_root *root = btrfs_global_root(fs_info, &key);
1611 
1612 		return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT);
1613 	}
1614 	return NULL;
1615 }
1616 
1617 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1618 			 struct btrfs_root *root)
1619 {
1620 	int ret;
1621 
1622 	ret = radix_tree_preload(GFP_NOFS);
1623 	if (ret)
1624 		return ret;
1625 
1626 	spin_lock(&fs_info->fs_roots_radix_lock);
1627 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1628 				(unsigned long)root->root_key.objectid,
1629 				root);
1630 	if (ret == 0) {
1631 		btrfs_grab_root(root);
1632 		set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1633 	}
1634 	spin_unlock(&fs_info->fs_roots_radix_lock);
1635 	radix_tree_preload_end();
1636 
1637 	return ret;
1638 }
1639 
1640 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1641 {
1642 #ifdef CONFIG_BTRFS_DEBUG
1643 	struct btrfs_root *root;
1644 
1645 	while (!list_empty(&fs_info->allocated_roots)) {
1646 		char buf[BTRFS_ROOT_NAME_BUF_LEN];
1647 
1648 		root = list_first_entry(&fs_info->allocated_roots,
1649 					struct btrfs_root, leak_list);
1650 		btrfs_err(fs_info, "leaked root %s refcount %d",
1651 			  btrfs_root_name(&root->root_key, buf),
1652 			  refcount_read(&root->refs));
1653 		while (refcount_read(&root->refs) > 1)
1654 			btrfs_put_root(root);
1655 		btrfs_put_root(root);
1656 	}
1657 #endif
1658 }
1659 
1660 static void free_global_roots(struct btrfs_fs_info *fs_info)
1661 {
1662 	struct btrfs_root *root;
1663 	struct rb_node *node;
1664 
1665 	while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1666 		root = rb_entry(node, struct btrfs_root, rb_node);
1667 		rb_erase(&root->rb_node, &fs_info->global_root_tree);
1668 		btrfs_put_root(root);
1669 	}
1670 }
1671 
1672 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1673 {
1674 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1675 	percpu_counter_destroy(&fs_info->delalloc_bytes);
1676 	percpu_counter_destroy(&fs_info->ordered_bytes);
1677 	percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1678 	btrfs_free_csum_hash(fs_info);
1679 	btrfs_free_stripe_hash_table(fs_info);
1680 	btrfs_free_ref_cache(fs_info);
1681 	kfree(fs_info->balance_ctl);
1682 	kfree(fs_info->delayed_root);
1683 	free_global_roots(fs_info);
1684 	btrfs_put_root(fs_info->tree_root);
1685 	btrfs_put_root(fs_info->chunk_root);
1686 	btrfs_put_root(fs_info->dev_root);
1687 	btrfs_put_root(fs_info->quota_root);
1688 	btrfs_put_root(fs_info->uuid_root);
1689 	btrfs_put_root(fs_info->fs_root);
1690 	btrfs_put_root(fs_info->data_reloc_root);
1691 	btrfs_put_root(fs_info->block_group_root);
1692 	btrfs_check_leaked_roots(fs_info);
1693 	btrfs_extent_buffer_leak_debug_check(fs_info);
1694 	kfree(fs_info->super_copy);
1695 	kfree(fs_info->super_for_commit);
1696 	kfree(fs_info->subpage_info);
1697 	kvfree(fs_info);
1698 }
1699 
1700 
1701 /*
1702  * Get an in-memory reference of a root structure.
1703  *
1704  * For essential trees like root/extent tree, we grab it from fs_info directly.
1705  * For subvolume trees, we check the cached filesystem roots first. If not
1706  * found, then read it from disk and add it to cached fs roots.
1707  *
1708  * Caller should release the root by calling btrfs_put_root() after the usage.
1709  *
1710  * NOTE: Reloc and log trees can't be read by this function as they share the
1711  *	 same root objectid.
1712  *
1713  * @objectid:	root id
1714  * @anon_dev:	preallocated anonymous block device number for new roots,
1715  * 		pass 0 for new allocation.
1716  * @check_ref:	whether to check root item references, If true, return -ENOENT
1717  *		for orphan roots
1718  */
1719 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1720 					     u64 objectid, dev_t anon_dev,
1721 					     bool check_ref)
1722 {
1723 	struct btrfs_root *root;
1724 	struct btrfs_path *path;
1725 	struct btrfs_key key;
1726 	int ret;
1727 
1728 	root = btrfs_get_global_root(fs_info, objectid);
1729 	if (root)
1730 		return root;
1731 again:
1732 	root = btrfs_lookup_fs_root(fs_info, objectid);
1733 	if (root) {
1734 		/* Shouldn't get preallocated anon_dev for cached roots */
1735 		ASSERT(!anon_dev);
1736 		if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1737 			btrfs_put_root(root);
1738 			return ERR_PTR(-ENOENT);
1739 		}
1740 		return root;
1741 	}
1742 
1743 	key.objectid = objectid;
1744 	key.type = BTRFS_ROOT_ITEM_KEY;
1745 	key.offset = (u64)-1;
1746 	root = btrfs_read_tree_root(fs_info->tree_root, &key);
1747 	if (IS_ERR(root))
1748 		return root;
1749 
1750 	if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1751 		ret = -ENOENT;
1752 		goto fail;
1753 	}
1754 
1755 	ret = btrfs_init_fs_root(root, anon_dev);
1756 	if (ret)
1757 		goto fail;
1758 
1759 	path = btrfs_alloc_path();
1760 	if (!path) {
1761 		ret = -ENOMEM;
1762 		goto fail;
1763 	}
1764 	key.objectid = BTRFS_ORPHAN_OBJECTID;
1765 	key.type = BTRFS_ORPHAN_ITEM_KEY;
1766 	key.offset = objectid;
1767 
1768 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1769 	btrfs_free_path(path);
1770 	if (ret < 0)
1771 		goto fail;
1772 	if (ret == 0)
1773 		set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1774 
1775 	ret = btrfs_insert_fs_root(fs_info, root);
1776 	if (ret) {
1777 		if (ret == -EEXIST) {
1778 			btrfs_put_root(root);
1779 			goto again;
1780 		}
1781 		goto fail;
1782 	}
1783 	return root;
1784 fail:
1785 	/*
1786 	 * If our caller provided us an anonymous device, then it's his
1787 	 * responsibility to free it in case we fail. So we have to set our
1788 	 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1789 	 * and once again by our caller.
1790 	 */
1791 	if (anon_dev)
1792 		root->anon_dev = 0;
1793 	btrfs_put_root(root);
1794 	return ERR_PTR(ret);
1795 }
1796 
1797 /*
1798  * Get in-memory reference of a root structure
1799  *
1800  * @objectid:	tree objectid
1801  * @check_ref:	if set, verify that the tree exists and the item has at least
1802  *		one reference
1803  */
1804 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1805 				     u64 objectid, bool check_ref)
1806 {
1807 	return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1808 }
1809 
1810 /*
1811  * Get in-memory reference of a root structure, created as new, optionally pass
1812  * the anonymous block device id
1813  *
1814  * @objectid:	tree objectid
1815  * @anon_dev:	if zero, allocate a new anonymous block device or use the
1816  *		parameter value
1817  */
1818 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1819 					 u64 objectid, dev_t anon_dev)
1820 {
1821 	return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1822 }
1823 
1824 /*
1825  * btrfs_get_fs_root_commit_root - return a root for the given objectid
1826  * @fs_info:	the fs_info
1827  * @objectid:	the objectid we need to lookup
1828  *
1829  * This is exclusively used for backref walking, and exists specifically because
1830  * of how qgroups does lookups.  Qgroups will do a backref lookup at delayed ref
1831  * creation time, which means we may have to read the tree_root in order to look
1832  * up a fs root that is not in memory.  If the root is not in memory we will
1833  * read the tree root commit root and look up the fs root from there.  This is a
1834  * temporary root, it will not be inserted into the radix tree as it doesn't
1835  * have the most uptodate information, it'll simply be discarded once the
1836  * backref code is finished using the root.
1837  */
1838 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1839 						 struct btrfs_path *path,
1840 						 u64 objectid)
1841 {
1842 	struct btrfs_root *root;
1843 	struct btrfs_key key;
1844 
1845 	ASSERT(path->search_commit_root && path->skip_locking);
1846 
1847 	/*
1848 	 * This can return -ENOENT if we ask for a root that doesn't exist, but
1849 	 * since this is called via the backref walking code we won't be looking
1850 	 * up a root that doesn't exist, unless there's corruption.  So if root
1851 	 * != NULL just return it.
1852 	 */
1853 	root = btrfs_get_global_root(fs_info, objectid);
1854 	if (root)
1855 		return root;
1856 
1857 	root = btrfs_lookup_fs_root(fs_info, objectid);
1858 	if (root)
1859 		return root;
1860 
1861 	key.objectid = objectid;
1862 	key.type = BTRFS_ROOT_ITEM_KEY;
1863 	key.offset = (u64)-1;
1864 	root = read_tree_root_path(fs_info->tree_root, path, &key);
1865 	btrfs_release_path(path);
1866 
1867 	return root;
1868 }
1869 
1870 static int cleaner_kthread(void *arg)
1871 {
1872 	struct btrfs_fs_info *fs_info = arg;
1873 	int again;
1874 
1875 	while (1) {
1876 		again = 0;
1877 
1878 		set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1879 
1880 		/* Make the cleaner go to sleep early. */
1881 		if (btrfs_need_cleaner_sleep(fs_info))
1882 			goto sleep;
1883 
1884 		/*
1885 		 * Do not do anything if we might cause open_ctree() to block
1886 		 * before we have finished mounting the filesystem.
1887 		 */
1888 		if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1889 			goto sleep;
1890 
1891 		if (!mutex_trylock(&fs_info->cleaner_mutex))
1892 			goto sleep;
1893 
1894 		/*
1895 		 * Avoid the problem that we change the status of the fs
1896 		 * during the above check and trylock.
1897 		 */
1898 		if (btrfs_need_cleaner_sleep(fs_info)) {
1899 			mutex_unlock(&fs_info->cleaner_mutex);
1900 			goto sleep;
1901 		}
1902 
1903 		btrfs_run_delayed_iputs(fs_info);
1904 
1905 		again = btrfs_clean_one_deleted_snapshot(fs_info);
1906 		mutex_unlock(&fs_info->cleaner_mutex);
1907 
1908 		/*
1909 		 * The defragger has dealt with the R/O remount and umount,
1910 		 * needn't do anything special here.
1911 		 */
1912 		btrfs_run_defrag_inodes(fs_info);
1913 
1914 		/*
1915 		 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1916 		 * with relocation (btrfs_relocate_chunk) and relocation
1917 		 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1918 		 * after acquiring fs_info->reclaim_bgs_lock. So we
1919 		 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1920 		 * unused block groups.
1921 		 */
1922 		btrfs_delete_unused_bgs(fs_info);
1923 
1924 		/*
1925 		 * Reclaim block groups in the reclaim_bgs list after we deleted
1926 		 * all unused block_groups. This possibly gives us some more free
1927 		 * space.
1928 		 */
1929 		btrfs_reclaim_bgs(fs_info);
1930 sleep:
1931 		clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1932 		if (kthread_should_park())
1933 			kthread_parkme();
1934 		if (kthread_should_stop())
1935 			return 0;
1936 		if (!again) {
1937 			set_current_state(TASK_INTERRUPTIBLE);
1938 			schedule();
1939 			__set_current_state(TASK_RUNNING);
1940 		}
1941 	}
1942 }
1943 
1944 static int transaction_kthread(void *arg)
1945 {
1946 	struct btrfs_root *root = arg;
1947 	struct btrfs_fs_info *fs_info = root->fs_info;
1948 	struct btrfs_trans_handle *trans;
1949 	struct btrfs_transaction *cur;
1950 	u64 transid;
1951 	time64_t delta;
1952 	unsigned long delay;
1953 	bool cannot_commit;
1954 
1955 	do {
1956 		cannot_commit = false;
1957 		delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1958 		mutex_lock(&fs_info->transaction_kthread_mutex);
1959 
1960 		spin_lock(&fs_info->trans_lock);
1961 		cur = fs_info->running_transaction;
1962 		if (!cur) {
1963 			spin_unlock(&fs_info->trans_lock);
1964 			goto sleep;
1965 		}
1966 
1967 		delta = ktime_get_seconds() - cur->start_time;
1968 		if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1969 		    cur->state < TRANS_STATE_COMMIT_START &&
1970 		    delta < fs_info->commit_interval) {
1971 			spin_unlock(&fs_info->trans_lock);
1972 			delay -= msecs_to_jiffies((delta - 1) * 1000);
1973 			delay = min(delay,
1974 				    msecs_to_jiffies(fs_info->commit_interval * 1000));
1975 			goto sleep;
1976 		}
1977 		transid = cur->transid;
1978 		spin_unlock(&fs_info->trans_lock);
1979 
1980 		/* If the file system is aborted, this will always fail. */
1981 		trans = btrfs_attach_transaction(root);
1982 		if (IS_ERR(trans)) {
1983 			if (PTR_ERR(trans) != -ENOENT)
1984 				cannot_commit = true;
1985 			goto sleep;
1986 		}
1987 		if (transid == trans->transid) {
1988 			btrfs_commit_transaction(trans);
1989 		} else {
1990 			btrfs_end_transaction(trans);
1991 		}
1992 sleep:
1993 		wake_up_process(fs_info->cleaner_kthread);
1994 		mutex_unlock(&fs_info->transaction_kthread_mutex);
1995 
1996 		if (BTRFS_FS_ERROR(fs_info))
1997 			btrfs_cleanup_transaction(fs_info);
1998 		if (!kthread_should_stop() &&
1999 				(!btrfs_transaction_blocked(fs_info) ||
2000 				 cannot_commit))
2001 			schedule_timeout_interruptible(delay);
2002 	} while (!kthread_should_stop());
2003 	return 0;
2004 }
2005 
2006 /*
2007  * This will find the highest generation in the array of root backups.  The
2008  * index of the highest array is returned, or -EINVAL if we can't find
2009  * anything.
2010  *
2011  * We check to make sure the array is valid by comparing the
2012  * generation of the latest  root in the array with the generation
2013  * in the super block.  If they don't match we pitch it.
2014  */
2015 static int find_newest_super_backup(struct btrfs_fs_info *info)
2016 {
2017 	const u64 newest_gen = btrfs_super_generation(info->super_copy);
2018 	u64 cur;
2019 	struct btrfs_root_backup *root_backup;
2020 	int i;
2021 
2022 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2023 		root_backup = info->super_copy->super_roots + i;
2024 		cur = btrfs_backup_tree_root_gen(root_backup);
2025 		if (cur == newest_gen)
2026 			return i;
2027 	}
2028 
2029 	return -EINVAL;
2030 }
2031 
2032 /*
2033  * copy all the root pointers into the super backup array.
2034  * this will bump the backup pointer by one when it is
2035  * done
2036  */
2037 static void backup_super_roots(struct btrfs_fs_info *info)
2038 {
2039 	const int next_backup = info->backup_root_index;
2040 	struct btrfs_root_backup *root_backup;
2041 
2042 	root_backup = info->super_for_commit->super_roots + next_backup;
2043 
2044 	/*
2045 	 * make sure all of our padding and empty slots get zero filled
2046 	 * regardless of which ones we use today
2047 	 */
2048 	memset(root_backup, 0, sizeof(*root_backup));
2049 
2050 	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2051 
2052 	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2053 	btrfs_set_backup_tree_root_gen(root_backup,
2054 			       btrfs_header_generation(info->tree_root->node));
2055 
2056 	btrfs_set_backup_tree_root_level(root_backup,
2057 			       btrfs_header_level(info->tree_root->node));
2058 
2059 	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2060 	btrfs_set_backup_chunk_root_gen(root_backup,
2061 			       btrfs_header_generation(info->chunk_root->node));
2062 	btrfs_set_backup_chunk_root_level(root_backup,
2063 			       btrfs_header_level(info->chunk_root->node));
2064 
2065 	if (btrfs_fs_incompat(info, EXTENT_TREE_V2)) {
2066 		btrfs_set_backup_block_group_root(root_backup,
2067 					info->block_group_root->node->start);
2068 		btrfs_set_backup_block_group_root_gen(root_backup,
2069 			btrfs_header_generation(info->block_group_root->node));
2070 		btrfs_set_backup_block_group_root_level(root_backup,
2071 			btrfs_header_level(info->block_group_root->node));
2072 	} else {
2073 		struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
2074 		struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
2075 
2076 		btrfs_set_backup_extent_root(root_backup,
2077 					     extent_root->node->start);
2078 		btrfs_set_backup_extent_root_gen(root_backup,
2079 				btrfs_header_generation(extent_root->node));
2080 		btrfs_set_backup_extent_root_level(root_backup,
2081 					btrfs_header_level(extent_root->node));
2082 
2083 		btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
2084 		btrfs_set_backup_csum_root_gen(root_backup,
2085 					       btrfs_header_generation(csum_root->node));
2086 		btrfs_set_backup_csum_root_level(root_backup,
2087 						 btrfs_header_level(csum_root->node));
2088 	}
2089 
2090 	/*
2091 	 * we might commit during log recovery, which happens before we set
2092 	 * the fs_root.  Make sure it is valid before we fill it in.
2093 	 */
2094 	if (info->fs_root && info->fs_root->node) {
2095 		btrfs_set_backup_fs_root(root_backup,
2096 					 info->fs_root->node->start);
2097 		btrfs_set_backup_fs_root_gen(root_backup,
2098 			       btrfs_header_generation(info->fs_root->node));
2099 		btrfs_set_backup_fs_root_level(root_backup,
2100 			       btrfs_header_level(info->fs_root->node));
2101 	}
2102 
2103 	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2104 	btrfs_set_backup_dev_root_gen(root_backup,
2105 			       btrfs_header_generation(info->dev_root->node));
2106 	btrfs_set_backup_dev_root_level(root_backup,
2107 				       btrfs_header_level(info->dev_root->node));
2108 
2109 	btrfs_set_backup_total_bytes(root_backup,
2110 			     btrfs_super_total_bytes(info->super_copy));
2111 	btrfs_set_backup_bytes_used(root_backup,
2112 			     btrfs_super_bytes_used(info->super_copy));
2113 	btrfs_set_backup_num_devices(root_backup,
2114 			     btrfs_super_num_devices(info->super_copy));
2115 
2116 	/*
2117 	 * if we don't copy this out to the super_copy, it won't get remembered
2118 	 * for the next commit
2119 	 */
2120 	memcpy(&info->super_copy->super_roots,
2121 	       &info->super_for_commit->super_roots,
2122 	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2123 }
2124 
2125 /*
2126  * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2127  * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2128  *
2129  * fs_info - filesystem whose backup roots need to be read
2130  * priority - priority of backup root required
2131  *
2132  * Returns backup root index on success and -EINVAL otherwise.
2133  */
2134 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2135 {
2136 	int backup_index = find_newest_super_backup(fs_info);
2137 	struct btrfs_super_block *super = fs_info->super_copy;
2138 	struct btrfs_root_backup *root_backup;
2139 
2140 	if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2141 		if (priority == 0)
2142 			return backup_index;
2143 
2144 		backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2145 		backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2146 	} else {
2147 		return -EINVAL;
2148 	}
2149 
2150 	root_backup = super->super_roots + backup_index;
2151 
2152 	btrfs_set_super_generation(super,
2153 				   btrfs_backup_tree_root_gen(root_backup));
2154 	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2155 	btrfs_set_super_root_level(super,
2156 				   btrfs_backup_tree_root_level(root_backup));
2157 	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2158 
2159 	/*
2160 	 * Fixme: the total bytes and num_devices need to match or we should
2161 	 * need a fsck
2162 	 */
2163 	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2164 	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2165 
2166 	return backup_index;
2167 }
2168 
2169 /* helper to cleanup workers */
2170 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2171 {
2172 	btrfs_destroy_workqueue(fs_info->fixup_workers);
2173 	btrfs_destroy_workqueue(fs_info->delalloc_workers);
2174 	btrfs_destroy_workqueue(fs_info->hipri_workers);
2175 	btrfs_destroy_workqueue(fs_info->workers);
2176 	if (fs_info->endio_workers)
2177 		destroy_workqueue(fs_info->endio_workers);
2178 	if (fs_info->endio_raid56_workers)
2179 		destroy_workqueue(fs_info->endio_raid56_workers);
2180 	if (fs_info->rmw_workers)
2181 		destroy_workqueue(fs_info->rmw_workers);
2182 	if (fs_info->compressed_write_workers)
2183 		destroy_workqueue(fs_info->compressed_write_workers);
2184 	btrfs_destroy_workqueue(fs_info->endio_write_workers);
2185 	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2186 	btrfs_destroy_workqueue(fs_info->delayed_workers);
2187 	btrfs_destroy_workqueue(fs_info->caching_workers);
2188 	btrfs_destroy_workqueue(fs_info->flush_workers);
2189 	btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2190 	if (fs_info->discard_ctl.discard_workers)
2191 		destroy_workqueue(fs_info->discard_ctl.discard_workers);
2192 	/*
2193 	 * Now that all other work queues are destroyed, we can safely destroy
2194 	 * the queues used for metadata I/O, since tasks from those other work
2195 	 * queues can do metadata I/O operations.
2196 	 */
2197 	if (fs_info->endio_meta_workers)
2198 		destroy_workqueue(fs_info->endio_meta_workers);
2199 }
2200 
2201 static void free_root_extent_buffers(struct btrfs_root *root)
2202 {
2203 	if (root) {
2204 		free_extent_buffer(root->node);
2205 		free_extent_buffer(root->commit_root);
2206 		root->node = NULL;
2207 		root->commit_root = NULL;
2208 	}
2209 }
2210 
2211 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
2212 {
2213 	struct btrfs_root *root, *tmp;
2214 
2215 	rbtree_postorder_for_each_entry_safe(root, tmp,
2216 					     &fs_info->global_root_tree,
2217 					     rb_node)
2218 		free_root_extent_buffers(root);
2219 }
2220 
2221 /* helper to cleanup tree roots */
2222 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2223 {
2224 	free_root_extent_buffers(info->tree_root);
2225 
2226 	free_global_root_pointers(info);
2227 	free_root_extent_buffers(info->dev_root);
2228 	free_root_extent_buffers(info->quota_root);
2229 	free_root_extent_buffers(info->uuid_root);
2230 	free_root_extent_buffers(info->fs_root);
2231 	free_root_extent_buffers(info->data_reloc_root);
2232 	free_root_extent_buffers(info->block_group_root);
2233 	if (free_chunk_root)
2234 		free_root_extent_buffers(info->chunk_root);
2235 }
2236 
2237 void btrfs_put_root(struct btrfs_root *root)
2238 {
2239 	if (!root)
2240 		return;
2241 
2242 	if (refcount_dec_and_test(&root->refs)) {
2243 		WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2244 		WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2245 		if (root->anon_dev)
2246 			free_anon_bdev(root->anon_dev);
2247 		btrfs_drew_lock_destroy(&root->snapshot_lock);
2248 		free_root_extent_buffers(root);
2249 #ifdef CONFIG_BTRFS_DEBUG
2250 		spin_lock(&root->fs_info->fs_roots_radix_lock);
2251 		list_del_init(&root->leak_list);
2252 		spin_unlock(&root->fs_info->fs_roots_radix_lock);
2253 #endif
2254 		kfree(root);
2255 	}
2256 }
2257 
2258 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2259 {
2260 	int ret;
2261 	struct btrfs_root *gang[8];
2262 	int i;
2263 
2264 	while (!list_empty(&fs_info->dead_roots)) {
2265 		gang[0] = list_entry(fs_info->dead_roots.next,
2266 				     struct btrfs_root, root_list);
2267 		list_del(&gang[0]->root_list);
2268 
2269 		if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2270 			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2271 		btrfs_put_root(gang[0]);
2272 	}
2273 
2274 	while (1) {
2275 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2276 					     (void **)gang, 0,
2277 					     ARRAY_SIZE(gang));
2278 		if (!ret)
2279 			break;
2280 		for (i = 0; i < ret; i++)
2281 			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2282 	}
2283 }
2284 
2285 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2286 {
2287 	mutex_init(&fs_info->scrub_lock);
2288 	atomic_set(&fs_info->scrubs_running, 0);
2289 	atomic_set(&fs_info->scrub_pause_req, 0);
2290 	atomic_set(&fs_info->scrubs_paused, 0);
2291 	atomic_set(&fs_info->scrub_cancel_req, 0);
2292 	init_waitqueue_head(&fs_info->scrub_pause_wait);
2293 	refcount_set(&fs_info->scrub_workers_refcnt, 0);
2294 }
2295 
2296 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2297 {
2298 	spin_lock_init(&fs_info->balance_lock);
2299 	mutex_init(&fs_info->balance_mutex);
2300 	atomic_set(&fs_info->balance_pause_req, 0);
2301 	atomic_set(&fs_info->balance_cancel_req, 0);
2302 	fs_info->balance_ctl = NULL;
2303 	init_waitqueue_head(&fs_info->balance_wait_q);
2304 	atomic_set(&fs_info->reloc_cancel_req, 0);
2305 }
2306 
2307 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2308 {
2309 	struct inode *inode = fs_info->btree_inode;
2310 
2311 	inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2312 	set_nlink(inode, 1);
2313 	/*
2314 	 * we set the i_size on the btree inode to the max possible int.
2315 	 * the real end of the address space is determined by all of
2316 	 * the devices in the system
2317 	 */
2318 	inode->i_size = OFFSET_MAX;
2319 	inode->i_mapping->a_ops = &btree_aops;
2320 
2321 	RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2322 	extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2323 			    IO_TREE_BTREE_INODE_IO, inode);
2324 	BTRFS_I(inode)->io_tree.track_uptodate = false;
2325 	extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2326 
2327 	BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2328 	BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
2329 	BTRFS_I(inode)->location.type = 0;
2330 	BTRFS_I(inode)->location.offset = 0;
2331 	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2332 	btrfs_insert_inode_hash(inode);
2333 }
2334 
2335 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2336 {
2337 	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2338 	init_rwsem(&fs_info->dev_replace.rwsem);
2339 	init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2340 }
2341 
2342 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2343 {
2344 	spin_lock_init(&fs_info->qgroup_lock);
2345 	mutex_init(&fs_info->qgroup_ioctl_lock);
2346 	fs_info->qgroup_tree = RB_ROOT;
2347 	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2348 	fs_info->qgroup_seq = 1;
2349 	fs_info->qgroup_ulist = NULL;
2350 	fs_info->qgroup_rescan_running = false;
2351 	mutex_init(&fs_info->qgroup_rescan_lock);
2352 }
2353 
2354 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
2355 {
2356 	u32 max_active = fs_info->thread_pool_size;
2357 	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2358 
2359 	fs_info->workers =
2360 		btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
2361 	fs_info->hipri_workers =
2362 		btrfs_alloc_workqueue(fs_info, "worker-high",
2363 				      flags | WQ_HIGHPRI, max_active, 16);
2364 
2365 	fs_info->delalloc_workers =
2366 		btrfs_alloc_workqueue(fs_info, "delalloc",
2367 				      flags, max_active, 2);
2368 
2369 	fs_info->flush_workers =
2370 		btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2371 				      flags, max_active, 0);
2372 
2373 	fs_info->caching_workers =
2374 		btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2375 
2376 	fs_info->fixup_workers =
2377 		btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2378 
2379 	fs_info->endio_workers =
2380 		alloc_workqueue("btrfs-endio", flags, max_active);
2381 	fs_info->endio_meta_workers =
2382 		alloc_workqueue("btrfs-endio-meta", flags, max_active);
2383 	fs_info->endio_raid56_workers =
2384 		alloc_workqueue("btrfs-endio-raid56", flags, max_active);
2385 	fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
2386 	fs_info->endio_write_workers =
2387 		btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2388 				      max_active, 2);
2389 	fs_info->compressed_write_workers =
2390 		alloc_workqueue("btrfs-compressed-write", flags, max_active);
2391 	fs_info->endio_freespace_worker =
2392 		btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2393 				      max_active, 0);
2394 	fs_info->delayed_workers =
2395 		btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2396 				      max_active, 0);
2397 	fs_info->qgroup_rescan_workers =
2398 		btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2399 	fs_info->discard_ctl.discard_workers =
2400 		alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2401 
2402 	if (!(fs_info->workers && fs_info->hipri_workers &&
2403 	      fs_info->delalloc_workers && fs_info->flush_workers &&
2404 	      fs_info->endio_workers && fs_info->endio_meta_workers &&
2405 	      fs_info->compressed_write_workers &&
2406 	      fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2407 	      fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2408 	      fs_info->caching_workers && fs_info->fixup_workers &&
2409 	      fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2410 	      fs_info->discard_ctl.discard_workers)) {
2411 		return -ENOMEM;
2412 	}
2413 
2414 	return 0;
2415 }
2416 
2417 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2418 {
2419 	struct crypto_shash *csum_shash;
2420 	const char *csum_driver = btrfs_super_csum_driver(csum_type);
2421 
2422 	csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2423 
2424 	if (IS_ERR(csum_shash)) {
2425 		btrfs_err(fs_info, "error allocating %s hash for checksum",
2426 			  csum_driver);
2427 		return PTR_ERR(csum_shash);
2428 	}
2429 
2430 	fs_info->csum_shash = csum_shash;
2431 
2432 	btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2433 			btrfs_super_csum_name(csum_type),
2434 			crypto_shash_driver_name(csum_shash));
2435 	return 0;
2436 }
2437 
2438 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2439 			    struct btrfs_fs_devices *fs_devices)
2440 {
2441 	int ret;
2442 	struct btrfs_root *log_tree_root;
2443 	struct btrfs_super_block *disk_super = fs_info->super_copy;
2444 	u64 bytenr = btrfs_super_log_root(disk_super);
2445 	int level = btrfs_super_log_root_level(disk_super);
2446 
2447 	if (fs_devices->rw_devices == 0) {
2448 		btrfs_warn(fs_info, "log replay required on RO media");
2449 		return -EIO;
2450 	}
2451 
2452 	log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2453 					 GFP_KERNEL);
2454 	if (!log_tree_root)
2455 		return -ENOMEM;
2456 
2457 	log_tree_root->node = read_tree_block(fs_info, bytenr,
2458 					      BTRFS_TREE_LOG_OBJECTID,
2459 					      fs_info->generation + 1, level,
2460 					      NULL);
2461 	if (IS_ERR(log_tree_root->node)) {
2462 		btrfs_warn(fs_info, "failed to read log tree");
2463 		ret = PTR_ERR(log_tree_root->node);
2464 		log_tree_root->node = NULL;
2465 		btrfs_put_root(log_tree_root);
2466 		return ret;
2467 	}
2468 	if (!extent_buffer_uptodate(log_tree_root->node)) {
2469 		btrfs_err(fs_info, "failed to read log tree");
2470 		btrfs_put_root(log_tree_root);
2471 		return -EIO;
2472 	}
2473 
2474 	/* returns with log_tree_root freed on success */
2475 	ret = btrfs_recover_log_trees(log_tree_root);
2476 	if (ret) {
2477 		btrfs_handle_fs_error(fs_info, ret,
2478 				      "Failed to recover log tree");
2479 		btrfs_put_root(log_tree_root);
2480 		return ret;
2481 	}
2482 
2483 	if (sb_rdonly(fs_info->sb)) {
2484 		ret = btrfs_commit_super(fs_info);
2485 		if (ret)
2486 			return ret;
2487 	}
2488 
2489 	return 0;
2490 }
2491 
2492 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2493 				      struct btrfs_path *path, u64 objectid,
2494 				      const char *name)
2495 {
2496 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
2497 	struct btrfs_root *root;
2498 	u64 max_global_id = 0;
2499 	int ret;
2500 	struct btrfs_key key = {
2501 		.objectid = objectid,
2502 		.type = BTRFS_ROOT_ITEM_KEY,
2503 		.offset = 0,
2504 	};
2505 	bool found = false;
2506 
2507 	/* If we have IGNOREDATACSUMS skip loading these roots. */
2508 	if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2509 	    btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2510 		set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2511 		return 0;
2512 	}
2513 
2514 	while (1) {
2515 		ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2516 		if (ret < 0)
2517 			break;
2518 
2519 		if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2520 			ret = btrfs_next_leaf(tree_root, path);
2521 			if (ret) {
2522 				if (ret > 0)
2523 					ret = 0;
2524 				break;
2525 			}
2526 		}
2527 		ret = 0;
2528 
2529 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2530 		if (key.objectid != objectid)
2531 			break;
2532 		btrfs_release_path(path);
2533 
2534 		/*
2535 		 * Just worry about this for extent tree, it'll be the same for
2536 		 * everybody.
2537 		 */
2538 		if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2539 			max_global_id = max(max_global_id, key.offset);
2540 
2541 		found = true;
2542 		root = read_tree_root_path(tree_root, path, &key);
2543 		if (IS_ERR(root)) {
2544 			if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2545 				ret = PTR_ERR(root);
2546 			break;
2547 		}
2548 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2549 		ret = btrfs_global_root_insert(root);
2550 		if (ret) {
2551 			btrfs_put_root(root);
2552 			break;
2553 		}
2554 		key.offset++;
2555 	}
2556 	btrfs_release_path(path);
2557 
2558 	if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2559 		fs_info->nr_global_roots = max_global_id + 1;
2560 
2561 	if (!found || ret) {
2562 		if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2563 			set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2564 
2565 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2566 			ret = ret ? ret : -ENOENT;
2567 		else
2568 			ret = 0;
2569 		btrfs_err(fs_info, "failed to load root %s", name);
2570 	}
2571 	return ret;
2572 }
2573 
2574 static int load_global_roots(struct btrfs_root *tree_root)
2575 {
2576 	struct btrfs_path *path;
2577 	int ret = 0;
2578 
2579 	path = btrfs_alloc_path();
2580 	if (!path)
2581 		return -ENOMEM;
2582 
2583 	ret = load_global_roots_objectid(tree_root, path,
2584 					 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2585 	if (ret)
2586 		goto out;
2587 	ret = load_global_roots_objectid(tree_root, path,
2588 					 BTRFS_CSUM_TREE_OBJECTID, "csum");
2589 	if (ret)
2590 		goto out;
2591 	if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2592 		goto out;
2593 	ret = load_global_roots_objectid(tree_root, path,
2594 					 BTRFS_FREE_SPACE_TREE_OBJECTID,
2595 					 "free space");
2596 out:
2597 	btrfs_free_path(path);
2598 	return ret;
2599 }
2600 
2601 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2602 {
2603 	struct btrfs_root *tree_root = fs_info->tree_root;
2604 	struct btrfs_root *root;
2605 	struct btrfs_key location;
2606 	int ret;
2607 
2608 	BUG_ON(!fs_info->tree_root);
2609 
2610 	ret = load_global_roots(tree_root);
2611 	if (ret)
2612 		return ret;
2613 
2614 	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2615 	location.type = BTRFS_ROOT_ITEM_KEY;
2616 	location.offset = 0;
2617 
2618 	root = btrfs_read_tree_root(tree_root, &location);
2619 	if (IS_ERR(root)) {
2620 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2621 			ret = PTR_ERR(root);
2622 			goto out;
2623 		}
2624 	} else {
2625 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2626 		fs_info->dev_root = root;
2627 	}
2628 	/* Initialize fs_info for all devices in any case */
2629 	btrfs_init_devices_late(fs_info);
2630 
2631 	/*
2632 	 * This tree can share blocks with some other fs tree during relocation
2633 	 * and we need a proper setup by btrfs_get_fs_root
2634 	 */
2635 	root = btrfs_get_fs_root(tree_root->fs_info,
2636 				 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2637 	if (IS_ERR(root)) {
2638 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2639 			ret = PTR_ERR(root);
2640 			goto out;
2641 		}
2642 	} else {
2643 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2644 		fs_info->data_reloc_root = root;
2645 	}
2646 
2647 	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2648 	root = btrfs_read_tree_root(tree_root, &location);
2649 	if (!IS_ERR(root)) {
2650 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2651 		set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2652 		fs_info->quota_root = root;
2653 	}
2654 
2655 	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2656 	root = btrfs_read_tree_root(tree_root, &location);
2657 	if (IS_ERR(root)) {
2658 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2659 			ret = PTR_ERR(root);
2660 			if (ret != -ENOENT)
2661 				goto out;
2662 		}
2663 	} else {
2664 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2665 		fs_info->uuid_root = root;
2666 	}
2667 
2668 	return 0;
2669 out:
2670 	btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2671 		   location.objectid, ret);
2672 	return ret;
2673 }
2674 
2675 /*
2676  * Real super block validation
2677  * NOTE: super csum type and incompat features will not be checked here.
2678  *
2679  * @sb:		super block to check
2680  * @mirror_num:	the super block number to check its bytenr:
2681  * 		0	the primary (1st) sb
2682  * 		1, 2	2nd and 3rd backup copy
2683  * 	       -1	skip bytenr check
2684  */
2685 static int validate_super(struct btrfs_fs_info *fs_info,
2686 			    struct btrfs_super_block *sb, int mirror_num)
2687 {
2688 	u64 nodesize = btrfs_super_nodesize(sb);
2689 	u64 sectorsize = btrfs_super_sectorsize(sb);
2690 	int ret = 0;
2691 
2692 	if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2693 		btrfs_err(fs_info, "no valid FS found");
2694 		ret = -EINVAL;
2695 	}
2696 	if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2697 		btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2698 				btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2699 		ret = -EINVAL;
2700 	}
2701 	if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2702 		btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2703 				btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2704 		ret = -EINVAL;
2705 	}
2706 	if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2707 		btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2708 				btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2709 		ret = -EINVAL;
2710 	}
2711 	if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2712 		btrfs_err(fs_info, "log_root level too big: %d >= %d",
2713 				btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2714 		ret = -EINVAL;
2715 	}
2716 
2717 	/*
2718 	 * Check sectorsize and nodesize first, other check will need it.
2719 	 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2720 	 */
2721 	if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2722 	    sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2723 		btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2724 		ret = -EINVAL;
2725 	}
2726 
2727 	/*
2728 	 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2729 	 *
2730 	 * We can support 16K sectorsize with 64K page size without problem,
2731 	 * but such sectorsize/pagesize combination doesn't make much sense.
2732 	 * 4K will be our future standard, PAGE_SIZE is supported from the very
2733 	 * beginning.
2734 	 */
2735 	if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2736 		btrfs_err(fs_info,
2737 			"sectorsize %llu not yet supported for page size %lu",
2738 			sectorsize, PAGE_SIZE);
2739 		ret = -EINVAL;
2740 	}
2741 
2742 	if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2743 	    nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2744 		btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2745 		ret = -EINVAL;
2746 	}
2747 	if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2748 		btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2749 			  le32_to_cpu(sb->__unused_leafsize), nodesize);
2750 		ret = -EINVAL;
2751 	}
2752 
2753 	/* Root alignment check */
2754 	if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2755 		btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2756 			   btrfs_super_root(sb));
2757 		ret = -EINVAL;
2758 	}
2759 	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2760 		btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2761 			   btrfs_super_chunk_root(sb));
2762 		ret = -EINVAL;
2763 	}
2764 	if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2765 		btrfs_warn(fs_info, "log_root block unaligned: %llu",
2766 			   btrfs_super_log_root(sb));
2767 		ret = -EINVAL;
2768 	}
2769 
2770 	if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2771 		   BTRFS_FSID_SIZE)) {
2772 		btrfs_err(fs_info,
2773 		"superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2774 			fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2775 		ret = -EINVAL;
2776 	}
2777 
2778 	if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2779 	    memcmp(fs_info->fs_devices->metadata_uuid,
2780 		   fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2781 		btrfs_err(fs_info,
2782 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2783 			fs_info->super_copy->metadata_uuid,
2784 			fs_info->fs_devices->metadata_uuid);
2785 		ret = -EINVAL;
2786 	}
2787 
2788 	if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2789 		   BTRFS_FSID_SIZE) != 0) {
2790 		btrfs_err(fs_info,
2791 			"dev_item UUID does not match metadata fsid: %pU != %pU",
2792 			fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2793 		ret = -EINVAL;
2794 	}
2795 
2796 	/*
2797 	 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2798 	 * done later
2799 	 */
2800 	if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2801 		btrfs_err(fs_info, "bytes_used is too small %llu",
2802 			  btrfs_super_bytes_used(sb));
2803 		ret = -EINVAL;
2804 	}
2805 	if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2806 		btrfs_err(fs_info, "invalid stripesize %u",
2807 			  btrfs_super_stripesize(sb));
2808 		ret = -EINVAL;
2809 	}
2810 	if (btrfs_super_num_devices(sb) > (1UL << 31))
2811 		btrfs_warn(fs_info, "suspicious number of devices: %llu",
2812 			   btrfs_super_num_devices(sb));
2813 	if (btrfs_super_num_devices(sb) == 0) {
2814 		btrfs_err(fs_info, "number of devices is 0");
2815 		ret = -EINVAL;
2816 	}
2817 
2818 	if (mirror_num >= 0 &&
2819 	    btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2820 		btrfs_err(fs_info, "super offset mismatch %llu != %u",
2821 			  btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2822 		ret = -EINVAL;
2823 	}
2824 
2825 	/*
2826 	 * Obvious sys_chunk_array corruptions, it must hold at least one key
2827 	 * and one chunk
2828 	 */
2829 	if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2830 		btrfs_err(fs_info, "system chunk array too big %u > %u",
2831 			  btrfs_super_sys_array_size(sb),
2832 			  BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2833 		ret = -EINVAL;
2834 	}
2835 	if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2836 			+ sizeof(struct btrfs_chunk)) {
2837 		btrfs_err(fs_info, "system chunk array too small %u < %zu",
2838 			  btrfs_super_sys_array_size(sb),
2839 			  sizeof(struct btrfs_disk_key)
2840 			  + sizeof(struct btrfs_chunk));
2841 		ret = -EINVAL;
2842 	}
2843 
2844 	/*
2845 	 * The generation is a global counter, we'll trust it more than the others
2846 	 * but it's still possible that it's the one that's wrong.
2847 	 */
2848 	if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2849 		btrfs_warn(fs_info,
2850 			"suspicious: generation < chunk_root_generation: %llu < %llu",
2851 			btrfs_super_generation(sb),
2852 			btrfs_super_chunk_root_generation(sb));
2853 	if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2854 	    && btrfs_super_cache_generation(sb) != (u64)-1)
2855 		btrfs_warn(fs_info,
2856 			"suspicious: generation < cache_generation: %llu < %llu",
2857 			btrfs_super_generation(sb),
2858 			btrfs_super_cache_generation(sb));
2859 
2860 	return ret;
2861 }
2862 
2863 /*
2864  * Validation of super block at mount time.
2865  * Some checks already done early at mount time, like csum type and incompat
2866  * flags will be skipped.
2867  */
2868 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2869 {
2870 	return validate_super(fs_info, fs_info->super_copy, 0);
2871 }
2872 
2873 /*
2874  * Validation of super block at write time.
2875  * Some checks like bytenr check will be skipped as their values will be
2876  * overwritten soon.
2877  * Extra checks like csum type and incompat flags will be done here.
2878  */
2879 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2880 				      struct btrfs_super_block *sb)
2881 {
2882 	int ret;
2883 
2884 	ret = validate_super(fs_info, sb, -1);
2885 	if (ret < 0)
2886 		goto out;
2887 	if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2888 		ret = -EUCLEAN;
2889 		btrfs_err(fs_info, "invalid csum type, has %u want %u",
2890 			  btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2891 		goto out;
2892 	}
2893 	if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2894 		ret = -EUCLEAN;
2895 		btrfs_err(fs_info,
2896 		"invalid incompat flags, has 0x%llx valid mask 0x%llx",
2897 			  btrfs_super_incompat_flags(sb),
2898 			  (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2899 		goto out;
2900 	}
2901 out:
2902 	if (ret < 0)
2903 		btrfs_err(fs_info,
2904 		"super block corruption detected before writing it to disk");
2905 	return ret;
2906 }
2907 
2908 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2909 {
2910 	int ret = 0;
2911 
2912 	root->node = read_tree_block(root->fs_info, bytenr,
2913 				     root->root_key.objectid, gen, level, NULL);
2914 	if (IS_ERR(root->node)) {
2915 		ret = PTR_ERR(root->node);
2916 		root->node = NULL;
2917 		return ret;
2918 	}
2919 	if (!extent_buffer_uptodate(root->node)) {
2920 		free_extent_buffer(root->node);
2921 		root->node = NULL;
2922 		return -EIO;
2923 	}
2924 
2925 	btrfs_set_root_node(&root->root_item, root->node);
2926 	root->commit_root = btrfs_root_node(root);
2927 	btrfs_set_root_refs(&root->root_item, 1);
2928 	return ret;
2929 }
2930 
2931 static int load_important_roots(struct btrfs_fs_info *fs_info)
2932 {
2933 	struct btrfs_super_block *sb = fs_info->super_copy;
2934 	u64 gen, bytenr;
2935 	int level, ret;
2936 
2937 	bytenr = btrfs_super_root(sb);
2938 	gen = btrfs_super_generation(sb);
2939 	level = btrfs_super_root_level(sb);
2940 	ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2941 	if (ret) {
2942 		btrfs_warn(fs_info, "couldn't read tree root");
2943 		return ret;
2944 	}
2945 
2946 	if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
2947 		return 0;
2948 
2949 	bytenr = btrfs_super_block_group_root(sb);
2950 	gen = btrfs_super_block_group_root_generation(sb);
2951 	level = btrfs_super_block_group_root_level(sb);
2952 	ret = load_super_root(fs_info->block_group_root, bytenr, gen, level);
2953 	if (ret)
2954 		btrfs_warn(fs_info, "couldn't read block group root");
2955 	return ret;
2956 }
2957 
2958 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2959 {
2960 	int backup_index = find_newest_super_backup(fs_info);
2961 	struct btrfs_super_block *sb = fs_info->super_copy;
2962 	struct btrfs_root *tree_root = fs_info->tree_root;
2963 	bool handle_error = false;
2964 	int ret = 0;
2965 	int i;
2966 
2967 	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2968 		struct btrfs_root *root;
2969 
2970 		root = btrfs_alloc_root(fs_info, BTRFS_BLOCK_GROUP_TREE_OBJECTID,
2971 					GFP_KERNEL);
2972 		if (!root)
2973 			return -ENOMEM;
2974 		fs_info->block_group_root = root;
2975 	}
2976 
2977 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2978 		if (handle_error) {
2979 			if (!IS_ERR(tree_root->node))
2980 				free_extent_buffer(tree_root->node);
2981 			tree_root->node = NULL;
2982 
2983 			if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2984 				break;
2985 
2986 			free_root_pointers(fs_info, 0);
2987 
2988 			/*
2989 			 * Don't use the log in recovery mode, it won't be
2990 			 * valid
2991 			 */
2992 			btrfs_set_super_log_root(sb, 0);
2993 
2994 			/* We can't trust the free space cache either */
2995 			btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2996 
2997 			ret = read_backup_root(fs_info, i);
2998 			backup_index = ret;
2999 			if (ret < 0)
3000 				return ret;
3001 		}
3002 
3003 		ret = load_important_roots(fs_info);
3004 		if (ret) {
3005 			handle_error = true;
3006 			continue;
3007 		}
3008 
3009 		/*
3010 		 * No need to hold btrfs_root::objectid_mutex since the fs
3011 		 * hasn't been fully initialised and we are the only user
3012 		 */
3013 		ret = btrfs_init_root_free_objectid(tree_root);
3014 		if (ret < 0) {
3015 			handle_error = true;
3016 			continue;
3017 		}
3018 
3019 		ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
3020 
3021 		ret = btrfs_read_roots(fs_info);
3022 		if (ret < 0) {
3023 			handle_error = true;
3024 			continue;
3025 		}
3026 
3027 		/* All successful */
3028 		fs_info->generation = btrfs_header_generation(tree_root->node);
3029 		fs_info->last_trans_committed = fs_info->generation;
3030 		fs_info->last_reloc_trans = 0;
3031 
3032 		/* Always begin writing backup roots after the one being used */
3033 		if (backup_index < 0) {
3034 			fs_info->backup_root_index = 0;
3035 		} else {
3036 			fs_info->backup_root_index = backup_index + 1;
3037 			fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
3038 		}
3039 		break;
3040 	}
3041 
3042 	return ret;
3043 }
3044 
3045 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
3046 {
3047 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
3048 	INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
3049 	INIT_LIST_HEAD(&fs_info->trans_list);
3050 	INIT_LIST_HEAD(&fs_info->dead_roots);
3051 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
3052 	INIT_LIST_HEAD(&fs_info->delalloc_roots);
3053 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
3054 	spin_lock_init(&fs_info->delalloc_root_lock);
3055 	spin_lock_init(&fs_info->trans_lock);
3056 	spin_lock_init(&fs_info->fs_roots_radix_lock);
3057 	spin_lock_init(&fs_info->delayed_iput_lock);
3058 	spin_lock_init(&fs_info->defrag_inodes_lock);
3059 	spin_lock_init(&fs_info->super_lock);
3060 	spin_lock_init(&fs_info->buffer_lock);
3061 	spin_lock_init(&fs_info->unused_bgs_lock);
3062 	spin_lock_init(&fs_info->treelog_bg_lock);
3063 	spin_lock_init(&fs_info->zone_active_bgs_lock);
3064 	spin_lock_init(&fs_info->relocation_bg_lock);
3065 	rwlock_init(&fs_info->tree_mod_log_lock);
3066 	rwlock_init(&fs_info->global_root_lock);
3067 	mutex_init(&fs_info->unused_bg_unpin_mutex);
3068 	mutex_init(&fs_info->reclaim_bgs_lock);
3069 	mutex_init(&fs_info->reloc_mutex);
3070 	mutex_init(&fs_info->delalloc_root_mutex);
3071 	mutex_init(&fs_info->zoned_meta_io_lock);
3072 	mutex_init(&fs_info->zoned_data_reloc_io_lock);
3073 	seqlock_init(&fs_info->profiles_lock);
3074 
3075 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
3076 	INIT_LIST_HEAD(&fs_info->space_info);
3077 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
3078 	INIT_LIST_HEAD(&fs_info->unused_bgs);
3079 	INIT_LIST_HEAD(&fs_info->reclaim_bgs);
3080 	INIT_LIST_HEAD(&fs_info->zone_active_bgs);
3081 #ifdef CONFIG_BTRFS_DEBUG
3082 	INIT_LIST_HEAD(&fs_info->allocated_roots);
3083 	INIT_LIST_HEAD(&fs_info->allocated_ebs);
3084 	spin_lock_init(&fs_info->eb_leak_lock);
3085 #endif
3086 	extent_map_tree_init(&fs_info->mapping_tree);
3087 	btrfs_init_block_rsv(&fs_info->global_block_rsv,
3088 			     BTRFS_BLOCK_RSV_GLOBAL);
3089 	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
3090 	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
3091 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
3092 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
3093 			     BTRFS_BLOCK_RSV_DELOPS);
3094 	btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
3095 			     BTRFS_BLOCK_RSV_DELREFS);
3096 
3097 	atomic_set(&fs_info->async_delalloc_pages, 0);
3098 	atomic_set(&fs_info->defrag_running, 0);
3099 	atomic_set(&fs_info->nr_delayed_iputs, 0);
3100 	atomic64_set(&fs_info->tree_mod_seq, 0);
3101 	fs_info->global_root_tree = RB_ROOT;
3102 	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
3103 	fs_info->metadata_ratio = 0;
3104 	fs_info->defrag_inodes = RB_ROOT;
3105 	atomic64_set(&fs_info->free_chunk_space, 0);
3106 	fs_info->tree_mod_log = RB_ROOT;
3107 	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
3108 	fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
3109 	btrfs_init_ref_verify(fs_info);
3110 
3111 	fs_info->thread_pool_size = min_t(unsigned long,
3112 					  num_online_cpus() + 2, 8);
3113 
3114 	INIT_LIST_HEAD(&fs_info->ordered_roots);
3115 	spin_lock_init(&fs_info->ordered_root_lock);
3116 
3117 	btrfs_init_scrub(fs_info);
3118 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3119 	fs_info->check_integrity_print_mask = 0;
3120 #endif
3121 	btrfs_init_balance(fs_info);
3122 	btrfs_init_async_reclaim_work(fs_info);
3123 
3124 	rwlock_init(&fs_info->block_group_cache_lock);
3125 	fs_info->block_group_cache_tree = RB_ROOT_CACHED;
3126 
3127 	extent_io_tree_init(fs_info, &fs_info->excluded_extents,
3128 			    IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
3129 
3130 	mutex_init(&fs_info->ordered_operations_mutex);
3131 	mutex_init(&fs_info->tree_log_mutex);
3132 	mutex_init(&fs_info->chunk_mutex);
3133 	mutex_init(&fs_info->transaction_kthread_mutex);
3134 	mutex_init(&fs_info->cleaner_mutex);
3135 	mutex_init(&fs_info->ro_block_group_mutex);
3136 	init_rwsem(&fs_info->commit_root_sem);
3137 	init_rwsem(&fs_info->cleanup_work_sem);
3138 	init_rwsem(&fs_info->subvol_sem);
3139 	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
3140 
3141 	btrfs_init_dev_replace_locks(fs_info);
3142 	btrfs_init_qgroup(fs_info);
3143 	btrfs_discard_init(fs_info);
3144 
3145 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
3146 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
3147 
3148 	init_waitqueue_head(&fs_info->transaction_throttle);
3149 	init_waitqueue_head(&fs_info->transaction_wait);
3150 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
3151 	init_waitqueue_head(&fs_info->async_submit_wait);
3152 	init_waitqueue_head(&fs_info->delayed_iputs_wait);
3153 	init_waitqueue_head(&fs_info->zone_finish_wait);
3154 
3155 	/* Usable values until the real ones are cached from the superblock */
3156 	fs_info->nodesize = 4096;
3157 	fs_info->sectorsize = 4096;
3158 	fs_info->sectorsize_bits = ilog2(4096);
3159 	fs_info->stripesize = 4096;
3160 
3161 	fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
3162 
3163 	spin_lock_init(&fs_info->swapfile_pins_lock);
3164 	fs_info->swapfile_pins = RB_ROOT;
3165 
3166 	fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
3167 	INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
3168 }
3169 
3170 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
3171 {
3172 	int ret;
3173 
3174 	fs_info->sb = sb;
3175 	sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
3176 	sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3177 
3178 	ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3179 	if (ret)
3180 		return ret;
3181 
3182 	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3183 	if (ret)
3184 		return ret;
3185 
3186 	fs_info->dirty_metadata_batch = PAGE_SIZE *
3187 					(1 + ilog2(nr_cpu_ids));
3188 
3189 	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3190 	if (ret)
3191 		return ret;
3192 
3193 	ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3194 			GFP_KERNEL);
3195 	if (ret)
3196 		return ret;
3197 
3198 	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3199 					GFP_KERNEL);
3200 	if (!fs_info->delayed_root)
3201 		return -ENOMEM;
3202 	btrfs_init_delayed_root(fs_info->delayed_root);
3203 
3204 	if (sb_rdonly(sb))
3205 		set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3206 
3207 	return btrfs_alloc_stripe_hash_table(fs_info);
3208 }
3209 
3210 static int btrfs_uuid_rescan_kthread(void *data)
3211 {
3212 	struct btrfs_fs_info *fs_info = data;
3213 	int ret;
3214 
3215 	/*
3216 	 * 1st step is to iterate through the existing UUID tree and
3217 	 * to delete all entries that contain outdated data.
3218 	 * 2nd step is to add all missing entries to the UUID tree.
3219 	 */
3220 	ret = btrfs_uuid_tree_iterate(fs_info);
3221 	if (ret < 0) {
3222 		if (ret != -EINTR)
3223 			btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3224 				   ret);
3225 		up(&fs_info->uuid_tree_rescan_sem);
3226 		return ret;
3227 	}
3228 	return btrfs_uuid_scan_kthread(data);
3229 }
3230 
3231 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3232 {
3233 	struct task_struct *task;
3234 
3235 	down(&fs_info->uuid_tree_rescan_sem);
3236 	task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3237 	if (IS_ERR(task)) {
3238 		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
3239 		btrfs_warn(fs_info, "failed to start uuid_rescan task");
3240 		up(&fs_info->uuid_tree_rescan_sem);
3241 		return PTR_ERR(task);
3242 	}
3243 
3244 	return 0;
3245 }
3246 
3247 /*
3248  * Some options only have meaning at mount time and shouldn't persist across
3249  * remounts, or be displayed. Clear these at the end of mount and remount
3250  * code paths.
3251  */
3252 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3253 {
3254 	btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3255 	btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3256 }
3257 
3258 /*
3259  * Mounting logic specific to read-write file systems. Shared by open_ctree
3260  * and btrfs_remount when remounting from read-only to read-write.
3261  */
3262 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3263 {
3264 	int ret;
3265 	const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3266 	bool clear_free_space_tree = false;
3267 
3268 	if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3269 	    btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3270 		clear_free_space_tree = true;
3271 	} else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3272 		   !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3273 		btrfs_warn(fs_info, "free space tree is invalid");
3274 		clear_free_space_tree = true;
3275 	}
3276 
3277 	if (clear_free_space_tree) {
3278 		btrfs_info(fs_info, "clearing free space tree");
3279 		ret = btrfs_clear_free_space_tree(fs_info);
3280 		if (ret) {
3281 			btrfs_warn(fs_info,
3282 				   "failed to clear free space tree: %d", ret);
3283 			goto out;
3284 		}
3285 	}
3286 
3287 	/*
3288 	 * btrfs_find_orphan_roots() is responsible for finding all the dead
3289 	 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3290 	 * them into the fs_info->fs_roots_radix tree. This must be done before
3291 	 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3292 	 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3293 	 * item before the root's tree is deleted - this means that if we unmount
3294 	 * or crash before the deletion completes, on the next mount we will not
3295 	 * delete what remains of the tree because the orphan item does not
3296 	 * exists anymore, which is what tells us we have a pending deletion.
3297 	 */
3298 	ret = btrfs_find_orphan_roots(fs_info);
3299 	if (ret)
3300 		goto out;
3301 
3302 	ret = btrfs_cleanup_fs_roots(fs_info);
3303 	if (ret)
3304 		goto out;
3305 
3306 	down_read(&fs_info->cleanup_work_sem);
3307 	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3308 	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3309 		up_read(&fs_info->cleanup_work_sem);
3310 		goto out;
3311 	}
3312 	up_read(&fs_info->cleanup_work_sem);
3313 
3314 	mutex_lock(&fs_info->cleaner_mutex);
3315 	ret = btrfs_recover_relocation(fs_info);
3316 	mutex_unlock(&fs_info->cleaner_mutex);
3317 	if (ret < 0) {
3318 		btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3319 		goto out;
3320 	}
3321 
3322 	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3323 	    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3324 		btrfs_info(fs_info, "creating free space tree");
3325 		ret = btrfs_create_free_space_tree(fs_info);
3326 		if (ret) {
3327 			btrfs_warn(fs_info,
3328 				"failed to create free space tree: %d", ret);
3329 			goto out;
3330 		}
3331 	}
3332 
3333 	if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3334 		ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3335 		if (ret)
3336 			goto out;
3337 	}
3338 
3339 	ret = btrfs_resume_balance_async(fs_info);
3340 	if (ret)
3341 		goto out;
3342 
3343 	ret = btrfs_resume_dev_replace_async(fs_info);
3344 	if (ret) {
3345 		btrfs_warn(fs_info, "failed to resume dev_replace");
3346 		goto out;
3347 	}
3348 
3349 	btrfs_qgroup_rescan_resume(fs_info);
3350 
3351 	if (!fs_info->uuid_root) {
3352 		btrfs_info(fs_info, "creating UUID tree");
3353 		ret = btrfs_create_uuid_tree(fs_info);
3354 		if (ret) {
3355 			btrfs_warn(fs_info,
3356 				   "failed to create the UUID tree %d", ret);
3357 			goto out;
3358 		}
3359 	}
3360 
3361 out:
3362 	return ret;
3363 }
3364 
3365 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3366 		      char *options)
3367 {
3368 	u32 sectorsize;
3369 	u32 nodesize;
3370 	u32 stripesize;
3371 	u64 generation;
3372 	u64 features;
3373 	u16 csum_type;
3374 	struct btrfs_super_block *disk_super;
3375 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3376 	struct btrfs_root *tree_root;
3377 	struct btrfs_root *chunk_root;
3378 	int ret;
3379 	int err = -EINVAL;
3380 	int level;
3381 
3382 	ret = init_mount_fs_info(fs_info, sb);
3383 	if (ret) {
3384 		err = ret;
3385 		goto fail;
3386 	}
3387 
3388 	/* These need to be init'ed before we start creating inodes and such. */
3389 	tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3390 				     GFP_KERNEL);
3391 	fs_info->tree_root = tree_root;
3392 	chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3393 				      GFP_KERNEL);
3394 	fs_info->chunk_root = chunk_root;
3395 	if (!tree_root || !chunk_root) {
3396 		err = -ENOMEM;
3397 		goto fail;
3398 	}
3399 
3400 	fs_info->btree_inode = new_inode(sb);
3401 	if (!fs_info->btree_inode) {
3402 		err = -ENOMEM;
3403 		goto fail;
3404 	}
3405 	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3406 	btrfs_init_btree_inode(fs_info);
3407 
3408 	invalidate_bdev(fs_devices->latest_dev->bdev);
3409 
3410 	/*
3411 	 * Read super block and check the signature bytes only
3412 	 */
3413 	disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3414 	if (IS_ERR(disk_super)) {
3415 		err = PTR_ERR(disk_super);
3416 		goto fail_alloc;
3417 	}
3418 
3419 	/*
3420 	 * Verify the type first, if that or the checksum value are
3421 	 * corrupted, we'll find out
3422 	 */
3423 	csum_type = btrfs_super_csum_type(disk_super);
3424 	if (!btrfs_supported_super_csum(csum_type)) {
3425 		btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3426 			  csum_type);
3427 		err = -EINVAL;
3428 		btrfs_release_disk_super(disk_super);
3429 		goto fail_alloc;
3430 	}
3431 
3432 	fs_info->csum_size = btrfs_super_csum_size(disk_super);
3433 
3434 	ret = btrfs_init_csum_hash(fs_info, csum_type);
3435 	if (ret) {
3436 		err = ret;
3437 		btrfs_release_disk_super(disk_super);
3438 		goto fail_alloc;
3439 	}
3440 
3441 	/*
3442 	 * We want to check superblock checksum, the type is stored inside.
3443 	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3444 	 */
3445 	if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
3446 		btrfs_err(fs_info, "superblock checksum mismatch");
3447 		err = -EINVAL;
3448 		btrfs_release_disk_super(disk_super);
3449 		goto fail_alloc;
3450 	}
3451 
3452 	/*
3453 	 * super_copy is zeroed at allocation time and we never touch the
3454 	 * following bytes up to INFO_SIZE, the checksum is calculated from
3455 	 * the whole block of INFO_SIZE
3456 	 */
3457 	memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3458 	btrfs_release_disk_super(disk_super);
3459 
3460 	disk_super = fs_info->super_copy;
3461 
3462 
3463 	features = btrfs_super_flags(disk_super);
3464 	if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3465 		features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3466 		btrfs_set_super_flags(disk_super, features);
3467 		btrfs_info(fs_info,
3468 			"found metadata UUID change in progress flag, clearing");
3469 	}
3470 
3471 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
3472 	       sizeof(*fs_info->super_for_commit));
3473 
3474 	ret = btrfs_validate_mount_super(fs_info);
3475 	if (ret) {
3476 		btrfs_err(fs_info, "superblock contains fatal errors");
3477 		err = -EINVAL;
3478 		goto fail_alloc;
3479 	}
3480 
3481 	if (!btrfs_super_root(disk_super))
3482 		goto fail_alloc;
3483 
3484 	/* check FS state, whether FS is broken. */
3485 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3486 		set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3487 
3488 	/*
3489 	 * In the long term, we'll store the compression type in the super
3490 	 * block, and it'll be used for per file compression control.
3491 	 */
3492 	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3493 
3494 
3495 	/* Set up fs_info before parsing mount options */
3496 	nodesize = btrfs_super_nodesize(disk_super);
3497 	sectorsize = btrfs_super_sectorsize(disk_super);
3498 	stripesize = sectorsize;
3499 	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3500 	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3501 
3502 	fs_info->nodesize = nodesize;
3503 	fs_info->sectorsize = sectorsize;
3504 	fs_info->sectorsize_bits = ilog2(sectorsize);
3505 	fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3506 	fs_info->stripesize = stripesize;
3507 
3508 	ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3509 	if (ret) {
3510 		err = ret;
3511 		goto fail_alloc;
3512 	}
3513 
3514 	features = btrfs_super_incompat_flags(disk_super) &
3515 		~BTRFS_FEATURE_INCOMPAT_SUPP;
3516 	if (features) {
3517 		btrfs_err(fs_info,
3518 		    "cannot mount because of unsupported optional features (0x%llx)",
3519 		    features);
3520 		err = -EINVAL;
3521 		goto fail_alloc;
3522 	}
3523 
3524 	features = btrfs_super_incompat_flags(disk_super);
3525 	features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3526 	if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3527 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3528 	else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3529 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3530 
3531 	/*
3532 	 * Flag our filesystem as having big metadata blocks if they are bigger
3533 	 * than the page size.
3534 	 */
3535 	if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3536 		features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3537 
3538 	/*
3539 	 * mixed block groups end up with duplicate but slightly offset
3540 	 * extent buffers for the same range.  It leads to corruptions
3541 	 */
3542 	if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3543 	    (sectorsize != nodesize)) {
3544 		btrfs_err(fs_info,
3545 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3546 			nodesize, sectorsize);
3547 		goto fail_alloc;
3548 	}
3549 
3550 	/*
3551 	 * Needn't use the lock because there is no other task which will
3552 	 * update the flag.
3553 	 */
3554 	btrfs_set_super_incompat_flags(disk_super, features);
3555 
3556 	features = btrfs_super_compat_ro_flags(disk_super) &
3557 		~BTRFS_FEATURE_COMPAT_RO_SUPP;
3558 	if (!sb_rdonly(sb) && features) {
3559 		btrfs_err(fs_info,
3560 	"cannot mount read-write because of unsupported optional features (0x%llx)",
3561 		       features);
3562 		err = -EINVAL;
3563 		goto fail_alloc;
3564 	}
3565 	/*
3566 	 * We have unsupported RO compat features, although RO mounted, we
3567 	 * should not cause any metadata write, including log replay.
3568 	 * Or we could screw up whatever the new feature requires.
3569 	 */
3570 	if (unlikely(features && btrfs_super_log_root(disk_super) &&
3571 		     !btrfs_test_opt(fs_info, NOLOGREPLAY))) {
3572 		btrfs_err(fs_info,
3573 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3574 			  features);
3575 		err = -EINVAL;
3576 		goto fail_alloc;
3577 	}
3578 
3579 
3580 	if (sectorsize < PAGE_SIZE) {
3581 		struct btrfs_subpage_info *subpage_info;
3582 
3583 		/*
3584 		 * V1 space cache has some hardcoded PAGE_SIZE usage, and is
3585 		 * going to be deprecated.
3586 		 *
3587 		 * Force to use v2 cache for subpage case.
3588 		 */
3589 		btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
3590 		btrfs_set_and_info(fs_info, FREE_SPACE_TREE,
3591 			"forcing free space tree for sector size %u with page size %lu",
3592 			sectorsize, PAGE_SIZE);
3593 
3594 		btrfs_warn(fs_info,
3595 		"read-write for sector size %u with page size %lu is experimental",
3596 			   sectorsize, PAGE_SIZE);
3597 		subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3598 		if (!subpage_info)
3599 			goto fail_alloc;
3600 		btrfs_init_subpage_info(subpage_info, sectorsize);
3601 		fs_info->subpage_info = subpage_info;
3602 	}
3603 
3604 	ret = btrfs_init_workqueues(fs_info);
3605 	if (ret) {
3606 		err = ret;
3607 		goto fail_sb_buffer;
3608 	}
3609 
3610 	sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3611 	sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3612 
3613 	sb->s_blocksize = sectorsize;
3614 	sb->s_blocksize_bits = blksize_bits(sectorsize);
3615 	memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3616 
3617 	mutex_lock(&fs_info->chunk_mutex);
3618 	ret = btrfs_read_sys_array(fs_info);
3619 	mutex_unlock(&fs_info->chunk_mutex);
3620 	if (ret) {
3621 		btrfs_err(fs_info, "failed to read the system array: %d", ret);
3622 		goto fail_sb_buffer;
3623 	}
3624 
3625 	generation = btrfs_super_chunk_root_generation(disk_super);
3626 	level = btrfs_super_chunk_root_level(disk_super);
3627 	ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3628 			      generation, level);
3629 	if (ret) {
3630 		btrfs_err(fs_info, "failed to read chunk root");
3631 		goto fail_tree_roots;
3632 	}
3633 
3634 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3635 			   offsetof(struct btrfs_header, chunk_tree_uuid),
3636 			   BTRFS_UUID_SIZE);
3637 
3638 	ret = btrfs_read_chunk_tree(fs_info);
3639 	if (ret) {
3640 		btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3641 		goto fail_tree_roots;
3642 	}
3643 
3644 	/*
3645 	 * At this point we know all the devices that make this filesystem,
3646 	 * including the seed devices but we don't know yet if the replace
3647 	 * target is required. So free devices that are not part of this
3648 	 * filesystem but skip the replace target device which is checked
3649 	 * below in btrfs_init_dev_replace().
3650 	 */
3651 	btrfs_free_extra_devids(fs_devices);
3652 	if (!fs_devices->latest_dev->bdev) {
3653 		btrfs_err(fs_info, "failed to read devices");
3654 		goto fail_tree_roots;
3655 	}
3656 
3657 	ret = init_tree_roots(fs_info);
3658 	if (ret)
3659 		goto fail_tree_roots;
3660 
3661 	/*
3662 	 * Get zone type information of zoned block devices. This will also
3663 	 * handle emulation of a zoned filesystem if a regular device has the
3664 	 * zoned incompat feature flag set.
3665 	 */
3666 	ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3667 	if (ret) {
3668 		btrfs_err(fs_info,
3669 			  "zoned: failed to read device zone info: %d",
3670 			  ret);
3671 		goto fail_block_groups;
3672 	}
3673 
3674 	/*
3675 	 * If we have a uuid root and we're not being told to rescan we need to
3676 	 * check the generation here so we can set the
3677 	 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3678 	 * transaction during a balance or the log replay without updating the
3679 	 * uuid generation, and then if we crash we would rescan the uuid tree,
3680 	 * even though it was perfectly fine.
3681 	 */
3682 	if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3683 	    fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3684 		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3685 
3686 	ret = btrfs_verify_dev_extents(fs_info);
3687 	if (ret) {
3688 		btrfs_err(fs_info,
3689 			  "failed to verify dev extents against chunks: %d",
3690 			  ret);
3691 		goto fail_block_groups;
3692 	}
3693 	ret = btrfs_recover_balance(fs_info);
3694 	if (ret) {
3695 		btrfs_err(fs_info, "failed to recover balance: %d", ret);
3696 		goto fail_block_groups;
3697 	}
3698 
3699 	ret = btrfs_init_dev_stats(fs_info);
3700 	if (ret) {
3701 		btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3702 		goto fail_block_groups;
3703 	}
3704 
3705 	ret = btrfs_init_dev_replace(fs_info);
3706 	if (ret) {
3707 		btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3708 		goto fail_block_groups;
3709 	}
3710 
3711 	ret = btrfs_check_zoned_mode(fs_info);
3712 	if (ret) {
3713 		btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3714 			  ret);
3715 		goto fail_block_groups;
3716 	}
3717 
3718 	ret = btrfs_sysfs_add_fsid(fs_devices);
3719 	if (ret) {
3720 		btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3721 				ret);
3722 		goto fail_block_groups;
3723 	}
3724 
3725 	ret = btrfs_sysfs_add_mounted(fs_info);
3726 	if (ret) {
3727 		btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3728 		goto fail_fsdev_sysfs;
3729 	}
3730 
3731 	ret = btrfs_init_space_info(fs_info);
3732 	if (ret) {
3733 		btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3734 		goto fail_sysfs;
3735 	}
3736 
3737 	ret = btrfs_read_block_groups(fs_info);
3738 	if (ret) {
3739 		btrfs_err(fs_info, "failed to read block groups: %d", ret);
3740 		goto fail_sysfs;
3741 	}
3742 
3743 	btrfs_free_zone_cache(fs_info);
3744 
3745 	if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3746 	    !btrfs_check_rw_degradable(fs_info, NULL)) {
3747 		btrfs_warn(fs_info,
3748 		"writable mount is not allowed due to too many missing devices");
3749 		goto fail_sysfs;
3750 	}
3751 
3752 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3753 					       "btrfs-cleaner");
3754 	if (IS_ERR(fs_info->cleaner_kthread))
3755 		goto fail_sysfs;
3756 
3757 	fs_info->transaction_kthread = kthread_run(transaction_kthread,
3758 						   tree_root,
3759 						   "btrfs-transaction");
3760 	if (IS_ERR(fs_info->transaction_kthread))
3761 		goto fail_cleaner;
3762 
3763 	if (!btrfs_test_opt(fs_info, NOSSD) &&
3764 	    !fs_info->fs_devices->rotating) {
3765 		btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3766 	}
3767 
3768 	/*
3769 	 * Mount does not set all options immediately, we can do it now and do
3770 	 * not have to wait for transaction commit
3771 	 */
3772 	btrfs_apply_pending_changes(fs_info);
3773 
3774 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3775 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3776 		ret = btrfsic_mount(fs_info, fs_devices,
3777 				    btrfs_test_opt(fs_info,
3778 					CHECK_INTEGRITY_DATA) ? 1 : 0,
3779 				    fs_info->check_integrity_print_mask);
3780 		if (ret)
3781 			btrfs_warn(fs_info,
3782 				"failed to initialize integrity check module: %d",
3783 				ret);
3784 	}
3785 #endif
3786 	ret = btrfs_read_qgroup_config(fs_info);
3787 	if (ret)
3788 		goto fail_trans_kthread;
3789 
3790 	if (btrfs_build_ref_tree(fs_info))
3791 		btrfs_err(fs_info, "couldn't build ref tree");
3792 
3793 	/* do not make disk changes in broken FS or nologreplay is given */
3794 	if (btrfs_super_log_root(disk_super) != 0 &&
3795 	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3796 		btrfs_info(fs_info, "start tree-log replay");
3797 		ret = btrfs_replay_log(fs_info, fs_devices);
3798 		if (ret) {
3799 			err = ret;
3800 			goto fail_qgroup;
3801 		}
3802 	}
3803 
3804 	fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3805 	if (IS_ERR(fs_info->fs_root)) {
3806 		err = PTR_ERR(fs_info->fs_root);
3807 		btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3808 		fs_info->fs_root = NULL;
3809 		goto fail_qgroup;
3810 	}
3811 
3812 	if (sb_rdonly(sb))
3813 		goto clear_oneshot;
3814 
3815 	ret = btrfs_start_pre_rw_mount(fs_info);
3816 	if (ret) {
3817 		close_ctree(fs_info);
3818 		return ret;
3819 	}
3820 	btrfs_discard_resume(fs_info);
3821 
3822 	if (fs_info->uuid_root &&
3823 	    (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3824 	     fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3825 		btrfs_info(fs_info, "checking UUID tree");
3826 		ret = btrfs_check_uuid_tree(fs_info);
3827 		if (ret) {
3828 			btrfs_warn(fs_info,
3829 				"failed to check the UUID tree: %d", ret);
3830 			close_ctree(fs_info);
3831 			return ret;
3832 		}
3833 	}
3834 
3835 	set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3836 
3837 	/* Kick the cleaner thread so it'll start deleting snapshots. */
3838 	if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3839 		wake_up_process(fs_info->cleaner_kthread);
3840 
3841 clear_oneshot:
3842 	btrfs_clear_oneshot_options(fs_info);
3843 	return 0;
3844 
3845 fail_qgroup:
3846 	btrfs_free_qgroup_config(fs_info);
3847 fail_trans_kthread:
3848 	kthread_stop(fs_info->transaction_kthread);
3849 	btrfs_cleanup_transaction(fs_info);
3850 	btrfs_free_fs_roots(fs_info);
3851 fail_cleaner:
3852 	kthread_stop(fs_info->cleaner_kthread);
3853 
3854 	/*
3855 	 * make sure we're done with the btree inode before we stop our
3856 	 * kthreads
3857 	 */
3858 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3859 
3860 fail_sysfs:
3861 	btrfs_sysfs_remove_mounted(fs_info);
3862 
3863 fail_fsdev_sysfs:
3864 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3865 
3866 fail_block_groups:
3867 	btrfs_put_block_group_cache(fs_info);
3868 
3869 fail_tree_roots:
3870 	if (fs_info->data_reloc_root)
3871 		btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3872 	free_root_pointers(fs_info, true);
3873 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3874 
3875 fail_sb_buffer:
3876 	btrfs_stop_all_workers(fs_info);
3877 	btrfs_free_block_groups(fs_info);
3878 fail_alloc:
3879 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3880 
3881 	iput(fs_info->btree_inode);
3882 fail:
3883 	btrfs_close_devices(fs_info->fs_devices);
3884 	return err;
3885 }
3886 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3887 
3888 static void btrfs_end_super_write(struct bio *bio)
3889 {
3890 	struct btrfs_device *device = bio->bi_private;
3891 	struct bio_vec *bvec;
3892 	struct bvec_iter_all iter_all;
3893 	struct page *page;
3894 
3895 	bio_for_each_segment_all(bvec, bio, iter_all) {
3896 		page = bvec->bv_page;
3897 
3898 		if (bio->bi_status) {
3899 			btrfs_warn_rl_in_rcu(device->fs_info,
3900 				"lost page write due to IO error on %s (%d)",
3901 				rcu_str_deref(device->name),
3902 				blk_status_to_errno(bio->bi_status));
3903 			ClearPageUptodate(page);
3904 			SetPageError(page);
3905 			btrfs_dev_stat_inc_and_print(device,
3906 						     BTRFS_DEV_STAT_WRITE_ERRS);
3907 		} else {
3908 			SetPageUptodate(page);
3909 		}
3910 
3911 		put_page(page);
3912 		unlock_page(page);
3913 	}
3914 
3915 	bio_put(bio);
3916 }
3917 
3918 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3919 						   int copy_num)
3920 {
3921 	struct btrfs_super_block *super;
3922 	struct page *page;
3923 	u64 bytenr, bytenr_orig;
3924 	struct address_space *mapping = bdev->bd_inode->i_mapping;
3925 	int ret;
3926 
3927 	bytenr_orig = btrfs_sb_offset(copy_num);
3928 	ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3929 	if (ret == -ENOENT)
3930 		return ERR_PTR(-EINVAL);
3931 	else if (ret)
3932 		return ERR_PTR(ret);
3933 
3934 	if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3935 		return ERR_PTR(-EINVAL);
3936 
3937 	page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3938 	if (IS_ERR(page))
3939 		return ERR_CAST(page);
3940 
3941 	super = page_address(page);
3942 	if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3943 		btrfs_release_disk_super(super);
3944 		return ERR_PTR(-ENODATA);
3945 	}
3946 
3947 	if (btrfs_super_bytenr(super) != bytenr_orig) {
3948 		btrfs_release_disk_super(super);
3949 		return ERR_PTR(-EINVAL);
3950 	}
3951 
3952 	return super;
3953 }
3954 
3955 
3956 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3957 {
3958 	struct btrfs_super_block *super, *latest = NULL;
3959 	int i;
3960 	u64 transid = 0;
3961 
3962 	/* we would like to check all the supers, but that would make
3963 	 * a btrfs mount succeed after a mkfs from a different FS.
3964 	 * So, we need to add a special mount option to scan for
3965 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3966 	 */
3967 	for (i = 0; i < 1; i++) {
3968 		super = btrfs_read_dev_one_super(bdev, i);
3969 		if (IS_ERR(super))
3970 			continue;
3971 
3972 		if (!latest || btrfs_super_generation(super) > transid) {
3973 			if (latest)
3974 				btrfs_release_disk_super(super);
3975 
3976 			latest = super;
3977 			transid = btrfs_super_generation(super);
3978 		}
3979 	}
3980 
3981 	return super;
3982 }
3983 
3984 /*
3985  * Write superblock @sb to the @device. Do not wait for completion, all the
3986  * pages we use for writing are locked.
3987  *
3988  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3989  * the expected device size at commit time. Note that max_mirrors must be
3990  * same for write and wait phases.
3991  *
3992  * Return number of errors when page is not found or submission fails.
3993  */
3994 static int write_dev_supers(struct btrfs_device *device,
3995 			    struct btrfs_super_block *sb, int max_mirrors)
3996 {
3997 	struct btrfs_fs_info *fs_info = device->fs_info;
3998 	struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3999 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
4000 	int i;
4001 	int errors = 0;
4002 	int ret;
4003 	u64 bytenr, bytenr_orig;
4004 
4005 	if (max_mirrors == 0)
4006 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4007 
4008 	shash->tfm = fs_info->csum_shash;
4009 
4010 	for (i = 0; i < max_mirrors; i++) {
4011 		struct page *page;
4012 		struct bio *bio;
4013 		struct btrfs_super_block *disk_super;
4014 
4015 		bytenr_orig = btrfs_sb_offset(i);
4016 		ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
4017 		if (ret == -ENOENT) {
4018 			continue;
4019 		} else if (ret < 0) {
4020 			btrfs_err(device->fs_info,
4021 				"couldn't get super block location for mirror %d",
4022 				i);
4023 			errors++;
4024 			continue;
4025 		}
4026 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4027 		    device->commit_total_bytes)
4028 			break;
4029 
4030 		btrfs_set_super_bytenr(sb, bytenr_orig);
4031 
4032 		crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
4033 				    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
4034 				    sb->csum);
4035 
4036 		page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
4037 					   GFP_NOFS);
4038 		if (!page) {
4039 			btrfs_err(device->fs_info,
4040 			    "couldn't get super block page for bytenr %llu",
4041 			    bytenr);
4042 			errors++;
4043 			continue;
4044 		}
4045 
4046 		/* Bump the refcount for wait_dev_supers() */
4047 		get_page(page);
4048 
4049 		disk_super = page_address(page);
4050 		memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
4051 
4052 		/*
4053 		 * Directly use bios here instead of relying on the page cache
4054 		 * to do I/O, so we don't lose the ability to do integrity
4055 		 * checking.
4056 		 */
4057 		bio = bio_alloc(device->bdev, 1,
4058 				REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
4059 				GFP_NOFS);
4060 		bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
4061 		bio->bi_private = device;
4062 		bio->bi_end_io = btrfs_end_super_write;
4063 		__bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
4064 			       offset_in_page(bytenr));
4065 
4066 		/*
4067 		 * We FUA only the first super block.  The others we allow to
4068 		 * go down lazy and there's a short window where the on-disk
4069 		 * copies might still contain the older version.
4070 		 */
4071 		if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
4072 			bio->bi_opf |= REQ_FUA;
4073 
4074 		btrfsic_check_bio(bio);
4075 		submit_bio(bio);
4076 
4077 		if (btrfs_advance_sb_log(device, i))
4078 			errors++;
4079 	}
4080 	return errors < i ? 0 : -1;
4081 }
4082 
4083 /*
4084  * Wait for write completion of superblocks done by write_dev_supers,
4085  * @max_mirrors same for write and wait phases.
4086  *
4087  * Return number of errors when page is not found or not marked up to
4088  * date.
4089  */
4090 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
4091 {
4092 	int i;
4093 	int errors = 0;
4094 	bool primary_failed = false;
4095 	int ret;
4096 	u64 bytenr;
4097 
4098 	if (max_mirrors == 0)
4099 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4100 
4101 	for (i = 0; i < max_mirrors; i++) {
4102 		struct page *page;
4103 
4104 		ret = btrfs_sb_log_location(device, i, READ, &bytenr);
4105 		if (ret == -ENOENT) {
4106 			break;
4107 		} else if (ret < 0) {
4108 			errors++;
4109 			if (i == 0)
4110 				primary_failed = true;
4111 			continue;
4112 		}
4113 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4114 		    device->commit_total_bytes)
4115 			break;
4116 
4117 		page = find_get_page(device->bdev->bd_inode->i_mapping,
4118 				     bytenr >> PAGE_SHIFT);
4119 		if (!page) {
4120 			errors++;
4121 			if (i == 0)
4122 				primary_failed = true;
4123 			continue;
4124 		}
4125 		/* Page is submitted locked and unlocked once the IO completes */
4126 		wait_on_page_locked(page);
4127 		if (PageError(page)) {
4128 			errors++;
4129 			if (i == 0)
4130 				primary_failed = true;
4131 		}
4132 
4133 		/* Drop our reference */
4134 		put_page(page);
4135 
4136 		/* Drop the reference from the writing run */
4137 		put_page(page);
4138 	}
4139 
4140 	/* log error, force error return */
4141 	if (primary_failed) {
4142 		btrfs_err(device->fs_info, "error writing primary super block to device %llu",
4143 			  device->devid);
4144 		return -1;
4145 	}
4146 
4147 	return errors < i ? 0 : -1;
4148 }
4149 
4150 /*
4151  * endio for the write_dev_flush, this will wake anyone waiting
4152  * for the barrier when it is done
4153  */
4154 static void btrfs_end_empty_barrier(struct bio *bio)
4155 {
4156 	bio_uninit(bio);
4157 	complete(bio->bi_private);
4158 }
4159 
4160 /*
4161  * Submit a flush request to the device if it supports it. Error handling is
4162  * done in the waiting counterpart.
4163  */
4164 static void write_dev_flush(struct btrfs_device *device)
4165 {
4166 	struct bio *bio = &device->flush_bio;
4167 
4168 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4169 	/*
4170 	 * When a disk has write caching disabled, we skip submission of a bio
4171 	 * with flush and sync requests before writing the superblock, since
4172 	 * it's not needed. However when the integrity checker is enabled, this
4173 	 * results in reports that there are metadata blocks referred by a
4174 	 * superblock that were not properly flushed. So don't skip the bio
4175 	 * submission only when the integrity checker is enabled for the sake
4176 	 * of simplicity, since this is a debug tool and not meant for use in
4177 	 * non-debug builds.
4178 	 */
4179 	if (!bdev_write_cache(device->bdev))
4180 		return;
4181 #endif
4182 
4183 	bio_init(bio, device->bdev, NULL, 0,
4184 		 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
4185 	bio->bi_end_io = btrfs_end_empty_barrier;
4186 	init_completion(&device->flush_wait);
4187 	bio->bi_private = &device->flush_wait;
4188 
4189 	btrfsic_check_bio(bio);
4190 	submit_bio(bio);
4191 	set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4192 }
4193 
4194 /*
4195  * If the flush bio has been submitted by write_dev_flush, wait for it.
4196  */
4197 static blk_status_t wait_dev_flush(struct btrfs_device *device)
4198 {
4199 	struct bio *bio = &device->flush_bio;
4200 
4201 	if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
4202 		return BLK_STS_OK;
4203 
4204 	clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4205 	wait_for_completion_io(&device->flush_wait);
4206 
4207 	return bio->bi_status;
4208 }
4209 
4210 static int check_barrier_error(struct btrfs_fs_info *fs_info)
4211 {
4212 	if (!btrfs_check_rw_degradable(fs_info, NULL))
4213 		return -EIO;
4214 	return 0;
4215 }
4216 
4217 /*
4218  * send an empty flush down to each device in parallel,
4219  * then wait for them
4220  */
4221 static int barrier_all_devices(struct btrfs_fs_info *info)
4222 {
4223 	struct list_head *head;
4224 	struct btrfs_device *dev;
4225 	int errors_wait = 0;
4226 	blk_status_t ret;
4227 
4228 	lockdep_assert_held(&info->fs_devices->device_list_mutex);
4229 	/* send down all the barriers */
4230 	head = &info->fs_devices->devices;
4231 	list_for_each_entry(dev, head, dev_list) {
4232 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4233 			continue;
4234 		if (!dev->bdev)
4235 			continue;
4236 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4237 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4238 			continue;
4239 
4240 		write_dev_flush(dev);
4241 		dev->last_flush_error = BLK_STS_OK;
4242 	}
4243 
4244 	/* wait for all the barriers */
4245 	list_for_each_entry(dev, head, dev_list) {
4246 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4247 			continue;
4248 		if (!dev->bdev) {
4249 			errors_wait++;
4250 			continue;
4251 		}
4252 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4253 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4254 			continue;
4255 
4256 		ret = wait_dev_flush(dev);
4257 		if (ret) {
4258 			dev->last_flush_error = ret;
4259 			btrfs_dev_stat_inc_and_print(dev,
4260 					BTRFS_DEV_STAT_FLUSH_ERRS);
4261 			errors_wait++;
4262 		}
4263 	}
4264 
4265 	if (errors_wait) {
4266 		/*
4267 		 * At some point we need the status of all disks
4268 		 * to arrive at the volume status. So error checking
4269 		 * is being pushed to a separate loop.
4270 		 */
4271 		return check_barrier_error(info);
4272 	}
4273 	return 0;
4274 }
4275 
4276 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4277 {
4278 	int raid_type;
4279 	int min_tolerated = INT_MAX;
4280 
4281 	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4282 	    (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4283 		min_tolerated = min_t(int, min_tolerated,
4284 				    btrfs_raid_array[BTRFS_RAID_SINGLE].
4285 				    tolerated_failures);
4286 
4287 	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4288 		if (raid_type == BTRFS_RAID_SINGLE)
4289 			continue;
4290 		if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4291 			continue;
4292 		min_tolerated = min_t(int, min_tolerated,
4293 				    btrfs_raid_array[raid_type].
4294 				    tolerated_failures);
4295 	}
4296 
4297 	if (min_tolerated == INT_MAX) {
4298 		pr_warn("BTRFS: unknown raid flag: %llu", flags);
4299 		min_tolerated = 0;
4300 	}
4301 
4302 	return min_tolerated;
4303 }
4304 
4305 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4306 {
4307 	struct list_head *head;
4308 	struct btrfs_device *dev;
4309 	struct btrfs_super_block *sb;
4310 	struct btrfs_dev_item *dev_item;
4311 	int ret;
4312 	int do_barriers;
4313 	int max_errors;
4314 	int total_errors = 0;
4315 	u64 flags;
4316 
4317 	do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4318 
4319 	/*
4320 	 * max_mirrors == 0 indicates we're from commit_transaction,
4321 	 * not from fsync where the tree roots in fs_info have not
4322 	 * been consistent on disk.
4323 	 */
4324 	if (max_mirrors == 0)
4325 		backup_super_roots(fs_info);
4326 
4327 	sb = fs_info->super_for_commit;
4328 	dev_item = &sb->dev_item;
4329 
4330 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
4331 	head = &fs_info->fs_devices->devices;
4332 	max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4333 
4334 	if (do_barriers) {
4335 		ret = barrier_all_devices(fs_info);
4336 		if (ret) {
4337 			mutex_unlock(
4338 				&fs_info->fs_devices->device_list_mutex);
4339 			btrfs_handle_fs_error(fs_info, ret,
4340 					      "errors while submitting device barriers.");
4341 			return ret;
4342 		}
4343 	}
4344 
4345 	list_for_each_entry(dev, head, dev_list) {
4346 		if (!dev->bdev) {
4347 			total_errors++;
4348 			continue;
4349 		}
4350 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4351 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4352 			continue;
4353 
4354 		btrfs_set_stack_device_generation(dev_item, 0);
4355 		btrfs_set_stack_device_type(dev_item, dev->type);
4356 		btrfs_set_stack_device_id(dev_item, dev->devid);
4357 		btrfs_set_stack_device_total_bytes(dev_item,
4358 						   dev->commit_total_bytes);
4359 		btrfs_set_stack_device_bytes_used(dev_item,
4360 						  dev->commit_bytes_used);
4361 		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4362 		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4363 		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4364 		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4365 		memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4366 		       BTRFS_FSID_SIZE);
4367 
4368 		flags = btrfs_super_flags(sb);
4369 		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4370 
4371 		ret = btrfs_validate_write_super(fs_info, sb);
4372 		if (ret < 0) {
4373 			mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4374 			btrfs_handle_fs_error(fs_info, -EUCLEAN,
4375 				"unexpected superblock corruption detected");
4376 			return -EUCLEAN;
4377 		}
4378 
4379 		ret = write_dev_supers(dev, sb, max_mirrors);
4380 		if (ret)
4381 			total_errors++;
4382 	}
4383 	if (total_errors > max_errors) {
4384 		btrfs_err(fs_info, "%d errors while writing supers",
4385 			  total_errors);
4386 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4387 
4388 		/* FUA is masked off if unsupported and can't be the reason */
4389 		btrfs_handle_fs_error(fs_info, -EIO,
4390 				      "%d errors while writing supers",
4391 				      total_errors);
4392 		return -EIO;
4393 	}
4394 
4395 	total_errors = 0;
4396 	list_for_each_entry(dev, head, dev_list) {
4397 		if (!dev->bdev)
4398 			continue;
4399 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4400 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4401 			continue;
4402 
4403 		ret = wait_dev_supers(dev, max_mirrors);
4404 		if (ret)
4405 			total_errors++;
4406 	}
4407 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4408 	if (total_errors > max_errors) {
4409 		btrfs_handle_fs_error(fs_info, -EIO,
4410 				      "%d errors while writing supers",
4411 				      total_errors);
4412 		return -EIO;
4413 	}
4414 	return 0;
4415 }
4416 
4417 /* Drop a fs root from the radix tree and free it. */
4418 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4419 				  struct btrfs_root *root)
4420 {
4421 	bool drop_ref = false;
4422 
4423 	spin_lock(&fs_info->fs_roots_radix_lock);
4424 	radix_tree_delete(&fs_info->fs_roots_radix,
4425 			  (unsigned long)root->root_key.objectid);
4426 	if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4427 		drop_ref = true;
4428 	spin_unlock(&fs_info->fs_roots_radix_lock);
4429 
4430 	if (BTRFS_FS_ERROR(fs_info)) {
4431 		ASSERT(root->log_root == NULL);
4432 		if (root->reloc_root) {
4433 			btrfs_put_root(root->reloc_root);
4434 			root->reloc_root = NULL;
4435 		}
4436 	}
4437 
4438 	if (drop_ref)
4439 		btrfs_put_root(root);
4440 }
4441 
4442 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4443 {
4444 	u64 root_objectid = 0;
4445 	struct btrfs_root *gang[8];
4446 	int i = 0;
4447 	int err = 0;
4448 	unsigned int ret = 0;
4449 
4450 	while (1) {
4451 		spin_lock(&fs_info->fs_roots_radix_lock);
4452 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4453 					     (void **)gang, root_objectid,
4454 					     ARRAY_SIZE(gang));
4455 		if (!ret) {
4456 			spin_unlock(&fs_info->fs_roots_radix_lock);
4457 			break;
4458 		}
4459 		root_objectid = gang[ret - 1]->root_key.objectid + 1;
4460 
4461 		for (i = 0; i < ret; i++) {
4462 			/* Avoid to grab roots in dead_roots */
4463 			if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4464 				gang[i] = NULL;
4465 				continue;
4466 			}
4467 			/* grab all the search result for later use */
4468 			gang[i] = btrfs_grab_root(gang[i]);
4469 		}
4470 		spin_unlock(&fs_info->fs_roots_radix_lock);
4471 
4472 		for (i = 0; i < ret; i++) {
4473 			if (!gang[i])
4474 				continue;
4475 			root_objectid = gang[i]->root_key.objectid;
4476 			err = btrfs_orphan_cleanup(gang[i]);
4477 			if (err)
4478 				break;
4479 			btrfs_put_root(gang[i]);
4480 		}
4481 		root_objectid++;
4482 	}
4483 
4484 	/* release the uncleaned roots due to error */
4485 	for (; i < ret; i++) {
4486 		if (gang[i])
4487 			btrfs_put_root(gang[i]);
4488 	}
4489 	return err;
4490 }
4491 
4492 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4493 {
4494 	struct btrfs_root *root = fs_info->tree_root;
4495 	struct btrfs_trans_handle *trans;
4496 
4497 	mutex_lock(&fs_info->cleaner_mutex);
4498 	btrfs_run_delayed_iputs(fs_info);
4499 	mutex_unlock(&fs_info->cleaner_mutex);
4500 	wake_up_process(fs_info->cleaner_kthread);
4501 
4502 	/* wait until ongoing cleanup work done */
4503 	down_write(&fs_info->cleanup_work_sem);
4504 	up_write(&fs_info->cleanup_work_sem);
4505 
4506 	trans = btrfs_join_transaction(root);
4507 	if (IS_ERR(trans))
4508 		return PTR_ERR(trans);
4509 	return btrfs_commit_transaction(trans);
4510 }
4511 
4512 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4513 {
4514 	struct btrfs_transaction *trans;
4515 	struct btrfs_transaction *tmp;
4516 	bool found = false;
4517 
4518 	if (list_empty(&fs_info->trans_list))
4519 		return;
4520 
4521 	/*
4522 	 * This function is only called at the very end of close_ctree(),
4523 	 * thus no other running transaction, no need to take trans_lock.
4524 	 */
4525 	ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4526 	list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4527 		struct extent_state *cached = NULL;
4528 		u64 dirty_bytes = 0;
4529 		u64 cur = 0;
4530 		u64 found_start;
4531 		u64 found_end;
4532 
4533 		found = true;
4534 		while (!find_first_extent_bit(&trans->dirty_pages, cur,
4535 			&found_start, &found_end, EXTENT_DIRTY, &cached)) {
4536 			dirty_bytes += found_end + 1 - found_start;
4537 			cur = found_end + 1;
4538 		}
4539 		btrfs_warn(fs_info,
4540 	"transaction %llu (with %llu dirty metadata bytes) is not committed",
4541 			   trans->transid, dirty_bytes);
4542 		btrfs_cleanup_one_transaction(trans, fs_info);
4543 
4544 		if (trans == fs_info->running_transaction)
4545 			fs_info->running_transaction = NULL;
4546 		list_del_init(&trans->list);
4547 
4548 		btrfs_put_transaction(trans);
4549 		trace_btrfs_transaction_commit(fs_info);
4550 	}
4551 	ASSERT(!found);
4552 }
4553 
4554 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4555 {
4556 	int ret;
4557 
4558 	set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4559 
4560 	/*
4561 	 * We may have the reclaim task running and relocating a data block group,
4562 	 * in which case it may create delayed iputs. So stop it before we park
4563 	 * the cleaner kthread otherwise we can get new delayed iputs after
4564 	 * parking the cleaner, and that can make the async reclaim task to hang
4565 	 * if it's waiting for delayed iputs to complete, since the cleaner is
4566 	 * parked and can not run delayed iputs - this will make us hang when
4567 	 * trying to stop the async reclaim task.
4568 	 */
4569 	cancel_work_sync(&fs_info->reclaim_bgs_work);
4570 	/*
4571 	 * We don't want the cleaner to start new transactions, add more delayed
4572 	 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4573 	 * because that frees the task_struct, and the transaction kthread might
4574 	 * still try to wake up the cleaner.
4575 	 */
4576 	kthread_park(fs_info->cleaner_kthread);
4577 
4578 	/*
4579 	 * If we had UNFINISHED_DROPS we could still be processing them, so
4580 	 * clear that bit and wake up relocation so it can stop.
4581 	 */
4582 	btrfs_wake_unfinished_drop(fs_info);
4583 
4584 	/* wait for the qgroup rescan worker to stop */
4585 	btrfs_qgroup_wait_for_completion(fs_info, false);
4586 
4587 	/* wait for the uuid_scan task to finish */
4588 	down(&fs_info->uuid_tree_rescan_sem);
4589 	/* avoid complains from lockdep et al., set sem back to initial state */
4590 	up(&fs_info->uuid_tree_rescan_sem);
4591 
4592 	/* pause restriper - we want to resume on mount */
4593 	btrfs_pause_balance(fs_info);
4594 
4595 	btrfs_dev_replace_suspend_for_unmount(fs_info);
4596 
4597 	btrfs_scrub_cancel(fs_info);
4598 
4599 	/* wait for any defraggers to finish */
4600 	wait_event(fs_info->transaction_wait,
4601 		   (atomic_read(&fs_info->defrag_running) == 0));
4602 
4603 	/* clear out the rbtree of defraggable inodes */
4604 	btrfs_cleanup_defrag_inodes(fs_info);
4605 
4606 	cancel_work_sync(&fs_info->async_reclaim_work);
4607 	cancel_work_sync(&fs_info->async_data_reclaim_work);
4608 	cancel_work_sync(&fs_info->preempt_reclaim_work);
4609 
4610 	/* Cancel or finish ongoing discard work */
4611 	btrfs_discard_cleanup(fs_info);
4612 
4613 	if (!sb_rdonly(fs_info->sb)) {
4614 		/*
4615 		 * The cleaner kthread is stopped, so do one final pass over
4616 		 * unused block groups.
4617 		 */
4618 		btrfs_delete_unused_bgs(fs_info);
4619 
4620 		/*
4621 		 * There might be existing delayed inode workers still running
4622 		 * and holding an empty delayed inode item. We must wait for
4623 		 * them to complete first because they can create a transaction.
4624 		 * This happens when someone calls btrfs_balance_delayed_items()
4625 		 * and then a transaction commit runs the same delayed nodes
4626 		 * before any delayed worker has done something with the nodes.
4627 		 * We must wait for any worker here and not at transaction
4628 		 * commit time since that could cause a deadlock.
4629 		 * This is a very rare case.
4630 		 */
4631 		btrfs_flush_workqueue(fs_info->delayed_workers);
4632 
4633 		ret = btrfs_commit_super(fs_info);
4634 		if (ret)
4635 			btrfs_err(fs_info, "commit super ret %d", ret);
4636 	}
4637 
4638 	if (BTRFS_FS_ERROR(fs_info))
4639 		btrfs_error_commit_super(fs_info);
4640 
4641 	kthread_stop(fs_info->transaction_kthread);
4642 	kthread_stop(fs_info->cleaner_kthread);
4643 
4644 	ASSERT(list_empty(&fs_info->delayed_iputs));
4645 	set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4646 
4647 	if (btrfs_check_quota_leak(fs_info)) {
4648 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4649 		btrfs_err(fs_info, "qgroup reserved space leaked");
4650 	}
4651 
4652 	btrfs_free_qgroup_config(fs_info);
4653 	ASSERT(list_empty(&fs_info->delalloc_roots));
4654 
4655 	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4656 		btrfs_info(fs_info, "at unmount delalloc count %lld",
4657 		       percpu_counter_sum(&fs_info->delalloc_bytes));
4658 	}
4659 
4660 	if (percpu_counter_sum(&fs_info->ordered_bytes))
4661 		btrfs_info(fs_info, "at unmount dio bytes count %lld",
4662 			   percpu_counter_sum(&fs_info->ordered_bytes));
4663 
4664 	btrfs_sysfs_remove_mounted(fs_info);
4665 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4666 
4667 	btrfs_put_block_group_cache(fs_info);
4668 
4669 	/*
4670 	 * we must make sure there is not any read request to
4671 	 * submit after we stopping all workers.
4672 	 */
4673 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4674 	btrfs_stop_all_workers(fs_info);
4675 
4676 	/* We shouldn't have any transaction open at this point */
4677 	warn_about_uncommitted_trans(fs_info);
4678 
4679 	clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4680 	free_root_pointers(fs_info, true);
4681 	btrfs_free_fs_roots(fs_info);
4682 
4683 	/*
4684 	 * We must free the block groups after dropping the fs_roots as we could
4685 	 * have had an IO error and have left over tree log blocks that aren't
4686 	 * cleaned up until the fs roots are freed.  This makes the block group
4687 	 * accounting appear to be wrong because there's pending reserved bytes,
4688 	 * so make sure we do the block group cleanup afterwards.
4689 	 */
4690 	btrfs_free_block_groups(fs_info);
4691 
4692 	iput(fs_info->btree_inode);
4693 
4694 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4695 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4696 		btrfsic_unmount(fs_info->fs_devices);
4697 #endif
4698 
4699 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
4700 	btrfs_close_devices(fs_info->fs_devices);
4701 }
4702 
4703 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4704 			  int atomic)
4705 {
4706 	int ret;
4707 	struct inode *btree_inode = buf->pages[0]->mapping->host;
4708 
4709 	ret = extent_buffer_uptodate(buf);
4710 	if (!ret)
4711 		return ret;
4712 
4713 	ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4714 				    parent_transid, atomic);
4715 	if (ret == -EAGAIN)
4716 		return ret;
4717 	return !ret;
4718 }
4719 
4720 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4721 {
4722 	struct btrfs_fs_info *fs_info = buf->fs_info;
4723 	u64 transid = btrfs_header_generation(buf);
4724 	int was_dirty;
4725 
4726 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4727 	/*
4728 	 * This is a fast path so only do this check if we have sanity tests
4729 	 * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4730 	 * outside of the sanity tests.
4731 	 */
4732 	if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4733 		return;
4734 #endif
4735 	btrfs_assert_tree_write_locked(buf);
4736 	if (transid != fs_info->generation)
4737 		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4738 			buf->start, transid, fs_info->generation);
4739 	was_dirty = set_extent_buffer_dirty(buf);
4740 	if (!was_dirty)
4741 		percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4742 					 buf->len,
4743 					 fs_info->dirty_metadata_batch);
4744 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4745 	/*
4746 	 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4747 	 * but item data not updated.
4748 	 * So here we should only check item pointers, not item data.
4749 	 */
4750 	if (btrfs_header_level(buf) == 0 &&
4751 	    btrfs_check_leaf_relaxed(buf)) {
4752 		btrfs_print_leaf(buf);
4753 		ASSERT(0);
4754 	}
4755 #endif
4756 }
4757 
4758 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4759 					int flush_delayed)
4760 {
4761 	/*
4762 	 * looks as though older kernels can get into trouble with
4763 	 * this code, they end up stuck in balance_dirty_pages forever
4764 	 */
4765 	int ret;
4766 
4767 	if (current->flags & PF_MEMALLOC)
4768 		return;
4769 
4770 	if (flush_delayed)
4771 		btrfs_balance_delayed_items(fs_info);
4772 
4773 	ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4774 				     BTRFS_DIRTY_METADATA_THRESH,
4775 				     fs_info->dirty_metadata_batch);
4776 	if (ret > 0) {
4777 		balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4778 	}
4779 }
4780 
4781 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4782 {
4783 	__btrfs_btree_balance_dirty(fs_info, 1);
4784 }
4785 
4786 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4787 {
4788 	__btrfs_btree_balance_dirty(fs_info, 0);
4789 }
4790 
4791 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4792 {
4793 	/* cleanup FS via transaction */
4794 	btrfs_cleanup_transaction(fs_info);
4795 
4796 	mutex_lock(&fs_info->cleaner_mutex);
4797 	btrfs_run_delayed_iputs(fs_info);
4798 	mutex_unlock(&fs_info->cleaner_mutex);
4799 
4800 	down_write(&fs_info->cleanup_work_sem);
4801 	up_write(&fs_info->cleanup_work_sem);
4802 }
4803 
4804 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4805 {
4806 	struct btrfs_root *gang[8];
4807 	u64 root_objectid = 0;
4808 	int ret;
4809 
4810 	spin_lock(&fs_info->fs_roots_radix_lock);
4811 	while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4812 					     (void **)gang, root_objectid,
4813 					     ARRAY_SIZE(gang))) != 0) {
4814 		int i;
4815 
4816 		for (i = 0; i < ret; i++)
4817 			gang[i] = btrfs_grab_root(gang[i]);
4818 		spin_unlock(&fs_info->fs_roots_radix_lock);
4819 
4820 		for (i = 0; i < ret; i++) {
4821 			if (!gang[i])
4822 				continue;
4823 			root_objectid = gang[i]->root_key.objectid;
4824 			btrfs_free_log(NULL, gang[i]);
4825 			btrfs_put_root(gang[i]);
4826 		}
4827 		root_objectid++;
4828 		spin_lock(&fs_info->fs_roots_radix_lock);
4829 	}
4830 	spin_unlock(&fs_info->fs_roots_radix_lock);
4831 	btrfs_free_log_root_tree(NULL, fs_info);
4832 }
4833 
4834 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4835 {
4836 	struct btrfs_ordered_extent *ordered;
4837 
4838 	spin_lock(&root->ordered_extent_lock);
4839 	/*
4840 	 * This will just short circuit the ordered completion stuff which will
4841 	 * make sure the ordered extent gets properly cleaned up.
4842 	 */
4843 	list_for_each_entry(ordered, &root->ordered_extents,
4844 			    root_extent_list)
4845 		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4846 	spin_unlock(&root->ordered_extent_lock);
4847 }
4848 
4849 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4850 {
4851 	struct btrfs_root *root;
4852 	struct list_head splice;
4853 
4854 	INIT_LIST_HEAD(&splice);
4855 
4856 	spin_lock(&fs_info->ordered_root_lock);
4857 	list_splice_init(&fs_info->ordered_roots, &splice);
4858 	while (!list_empty(&splice)) {
4859 		root = list_first_entry(&splice, struct btrfs_root,
4860 					ordered_root);
4861 		list_move_tail(&root->ordered_root,
4862 			       &fs_info->ordered_roots);
4863 
4864 		spin_unlock(&fs_info->ordered_root_lock);
4865 		btrfs_destroy_ordered_extents(root);
4866 
4867 		cond_resched();
4868 		spin_lock(&fs_info->ordered_root_lock);
4869 	}
4870 	spin_unlock(&fs_info->ordered_root_lock);
4871 
4872 	/*
4873 	 * We need this here because if we've been flipped read-only we won't
4874 	 * get sync() from the umount, so we need to make sure any ordered
4875 	 * extents that haven't had their dirty pages IO start writeout yet
4876 	 * actually get run and error out properly.
4877 	 */
4878 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4879 }
4880 
4881 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4882 				      struct btrfs_fs_info *fs_info)
4883 {
4884 	struct rb_node *node;
4885 	struct btrfs_delayed_ref_root *delayed_refs;
4886 	struct btrfs_delayed_ref_node *ref;
4887 	int ret = 0;
4888 
4889 	delayed_refs = &trans->delayed_refs;
4890 
4891 	spin_lock(&delayed_refs->lock);
4892 	if (atomic_read(&delayed_refs->num_entries) == 0) {
4893 		spin_unlock(&delayed_refs->lock);
4894 		btrfs_debug(fs_info, "delayed_refs has NO entry");
4895 		return ret;
4896 	}
4897 
4898 	while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4899 		struct btrfs_delayed_ref_head *head;
4900 		struct rb_node *n;
4901 		bool pin_bytes = false;
4902 
4903 		head = rb_entry(node, struct btrfs_delayed_ref_head,
4904 				href_node);
4905 		if (btrfs_delayed_ref_lock(delayed_refs, head))
4906 			continue;
4907 
4908 		spin_lock(&head->lock);
4909 		while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4910 			ref = rb_entry(n, struct btrfs_delayed_ref_node,
4911 				       ref_node);
4912 			ref->in_tree = 0;
4913 			rb_erase_cached(&ref->ref_node, &head->ref_tree);
4914 			RB_CLEAR_NODE(&ref->ref_node);
4915 			if (!list_empty(&ref->add_list))
4916 				list_del(&ref->add_list);
4917 			atomic_dec(&delayed_refs->num_entries);
4918 			btrfs_put_delayed_ref(ref);
4919 		}
4920 		if (head->must_insert_reserved)
4921 			pin_bytes = true;
4922 		btrfs_free_delayed_extent_op(head->extent_op);
4923 		btrfs_delete_ref_head(delayed_refs, head);
4924 		spin_unlock(&head->lock);
4925 		spin_unlock(&delayed_refs->lock);
4926 		mutex_unlock(&head->mutex);
4927 
4928 		if (pin_bytes) {
4929 			struct btrfs_block_group *cache;
4930 
4931 			cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4932 			BUG_ON(!cache);
4933 
4934 			spin_lock(&cache->space_info->lock);
4935 			spin_lock(&cache->lock);
4936 			cache->pinned += head->num_bytes;
4937 			btrfs_space_info_update_bytes_pinned(fs_info,
4938 				cache->space_info, head->num_bytes);
4939 			cache->reserved -= head->num_bytes;
4940 			cache->space_info->bytes_reserved -= head->num_bytes;
4941 			spin_unlock(&cache->lock);
4942 			spin_unlock(&cache->space_info->lock);
4943 
4944 			btrfs_put_block_group(cache);
4945 
4946 			btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4947 				head->bytenr + head->num_bytes - 1);
4948 		}
4949 		btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4950 		btrfs_put_delayed_ref_head(head);
4951 		cond_resched();
4952 		spin_lock(&delayed_refs->lock);
4953 	}
4954 	btrfs_qgroup_destroy_extent_records(trans);
4955 
4956 	spin_unlock(&delayed_refs->lock);
4957 
4958 	return ret;
4959 }
4960 
4961 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4962 {
4963 	struct btrfs_inode *btrfs_inode;
4964 	struct list_head splice;
4965 
4966 	INIT_LIST_HEAD(&splice);
4967 
4968 	spin_lock(&root->delalloc_lock);
4969 	list_splice_init(&root->delalloc_inodes, &splice);
4970 
4971 	while (!list_empty(&splice)) {
4972 		struct inode *inode = NULL;
4973 		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4974 					       delalloc_inodes);
4975 		__btrfs_del_delalloc_inode(root, btrfs_inode);
4976 		spin_unlock(&root->delalloc_lock);
4977 
4978 		/*
4979 		 * Make sure we get a live inode and that it'll not disappear
4980 		 * meanwhile.
4981 		 */
4982 		inode = igrab(&btrfs_inode->vfs_inode);
4983 		if (inode) {
4984 			invalidate_inode_pages2(inode->i_mapping);
4985 			iput(inode);
4986 		}
4987 		spin_lock(&root->delalloc_lock);
4988 	}
4989 	spin_unlock(&root->delalloc_lock);
4990 }
4991 
4992 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4993 {
4994 	struct btrfs_root *root;
4995 	struct list_head splice;
4996 
4997 	INIT_LIST_HEAD(&splice);
4998 
4999 	spin_lock(&fs_info->delalloc_root_lock);
5000 	list_splice_init(&fs_info->delalloc_roots, &splice);
5001 	while (!list_empty(&splice)) {
5002 		root = list_first_entry(&splice, struct btrfs_root,
5003 					 delalloc_root);
5004 		root = btrfs_grab_root(root);
5005 		BUG_ON(!root);
5006 		spin_unlock(&fs_info->delalloc_root_lock);
5007 
5008 		btrfs_destroy_delalloc_inodes(root);
5009 		btrfs_put_root(root);
5010 
5011 		spin_lock(&fs_info->delalloc_root_lock);
5012 	}
5013 	spin_unlock(&fs_info->delalloc_root_lock);
5014 }
5015 
5016 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
5017 					struct extent_io_tree *dirty_pages,
5018 					int mark)
5019 {
5020 	int ret;
5021 	struct extent_buffer *eb;
5022 	u64 start = 0;
5023 	u64 end;
5024 
5025 	while (1) {
5026 		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
5027 					    mark, NULL);
5028 		if (ret)
5029 			break;
5030 
5031 		clear_extent_bits(dirty_pages, start, end, mark);
5032 		while (start <= end) {
5033 			eb = find_extent_buffer(fs_info, start);
5034 			start += fs_info->nodesize;
5035 			if (!eb)
5036 				continue;
5037 			wait_on_extent_buffer_writeback(eb);
5038 
5039 			if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
5040 					       &eb->bflags))
5041 				clear_extent_buffer_dirty(eb);
5042 			free_extent_buffer_stale(eb);
5043 		}
5044 	}
5045 
5046 	return ret;
5047 }
5048 
5049 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
5050 				       struct extent_io_tree *unpin)
5051 {
5052 	u64 start;
5053 	u64 end;
5054 	int ret;
5055 
5056 	while (1) {
5057 		struct extent_state *cached_state = NULL;
5058 
5059 		/*
5060 		 * The btrfs_finish_extent_commit() may get the same range as
5061 		 * ours between find_first_extent_bit and clear_extent_dirty.
5062 		 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
5063 		 * the same extent range.
5064 		 */
5065 		mutex_lock(&fs_info->unused_bg_unpin_mutex);
5066 		ret = find_first_extent_bit(unpin, 0, &start, &end,
5067 					    EXTENT_DIRTY, &cached_state);
5068 		if (ret) {
5069 			mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5070 			break;
5071 		}
5072 
5073 		clear_extent_dirty(unpin, start, end, &cached_state);
5074 		free_extent_state(cached_state);
5075 		btrfs_error_unpin_extent_range(fs_info, start, end);
5076 		mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5077 		cond_resched();
5078 	}
5079 
5080 	return 0;
5081 }
5082 
5083 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
5084 {
5085 	struct inode *inode;
5086 
5087 	inode = cache->io_ctl.inode;
5088 	if (inode) {
5089 		invalidate_inode_pages2(inode->i_mapping);
5090 		BTRFS_I(inode)->generation = 0;
5091 		cache->io_ctl.inode = NULL;
5092 		iput(inode);
5093 	}
5094 	ASSERT(cache->io_ctl.pages == NULL);
5095 	btrfs_put_block_group(cache);
5096 }
5097 
5098 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
5099 			     struct btrfs_fs_info *fs_info)
5100 {
5101 	struct btrfs_block_group *cache;
5102 
5103 	spin_lock(&cur_trans->dirty_bgs_lock);
5104 	while (!list_empty(&cur_trans->dirty_bgs)) {
5105 		cache = list_first_entry(&cur_trans->dirty_bgs,
5106 					 struct btrfs_block_group,
5107 					 dirty_list);
5108 
5109 		if (!list_empty(&cache->io_list)) {
5110 			spin_unlock(&cur_trans->dirty_bgs_lock);
5111 			list_del_init(&cache->io_list);
5112 			btrfs_cleanup_bg_io(cache);
5113 			spin_lock(&cur_trans->dirty_bgs_lock);
5114 		}
5115 
5116 		list_del_init(&cache->dirty_list);
5117 		spin_lock(&cache->lock);
5118 		cache->disk_cache_state = BTRFS_DC_ERROR;
5119 		spin_unlock(&cache->lock);
5120 
5121 		spin_unlock(&cur_trans->dirty_bgs_lock);
5122 		btrfs_put_block_group(cache);
5123 		btrfs_delayed_refs_rsv_release(fs_info, 1);
5124 		spin_lock(&cur_trans->dirty_bgs_lock);
5125 	}
5126 	spin_unlock(&cur_trans->dirty_bgs_lock);
5127 
5128 	/*
5129 	 * Refer to the definition of io_bgs member for details why it's safe
5130 	 * to use it without any locking
5131 	 */
5132 	while (!list_empty(&cur_trans->io_bgs)) {
5133 		cache = list_first_entry(&cur_trans->io_bgs,
5134 					 struct btrfs_block_group,
5135 					 io_list);
5136 
5137 		list_del_init(&cache->io_list);
5138 		spin_lock(&cache->lock);
5139 		cache->disk_cache_state = BTRFS_DC_ERROR;
5140 		spin_unlock(&cache->lock);
5141 		btrfs_cleanup_bg_io(cache);
5142 	}
5143 }
5144 
5145 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
5146 				   struct btrfs_fs_info *fs_info)
5147 {
5148 	struct btrfs_device *dev, *tmp;
5149 
5150 	btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
5151 	ASSERT(list_empty(&cur_trans->dirty_bgs));
5152 	ASSERT(list_empty(&cur_trans->io_bgs));
5153 
5154 	list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
5155 				 post_commit_list) {
5156 		list_del_init(&dev->post_commit_list);
5157 	}
5158 
5159 	btrfs_destroy_delayed_refs(cur_trans, fs_info);
5160 
5161 	cur_trans->state = TRANS_STATE_COMMIT_START;
5162 	wake_up(&fs_info->transaction_blocked_wait);
5163 
5164 	cur_trans->state = TRANS_STATE_UNBLOCKED;
5165 	wake_up(&fs_info->transaction_wait);
5166 
5167 	btrfs_destroy_delayed_inodes(fs_info);
5168 
5169 	btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
5170 				     EXTENT_DIRTY);
5171 	btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
5172 
5173 	btrfs_free_redirty_list(cur_trans);
5174 
5175 	cur_trans->state =TRANS_STATE_COMPLETED;
5176 	wake_up(&cur_trans->commit_wait);
5177 }
5178 
5179 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
5180 {
5181 	struct btrfs_transaction *t;
5182 
5183 	mutex_lock(&fs_info->transaction_kthread_mutex);
5184 
5185 	spin_lock(&fs_info->trans_lock);
5186 	while (!list_empty(&fs_info->trans_list)) {
5187 		t = list_first_entry(&fs_info->trans_list,
5188 				     struct btrfs_transaction, list);
5189 		if (t->state >= TRANS_STATE_COMMIT_START) {
5190 			refcount_inc(&t->use_count);
5191 			spin_unlock(&fs_info->trans_lock);
5192 			btrfs_wait_for_commit(fs_info, t->transid);
5193 			btrfs_put_transaction(t);
5194 			spin_lock(&fs_info->trans_lock);
5195 			continue;
5196 		}
5197 		if (t == fs_info->running_transaction) {
5198 			t->state = TRANS_STATE_COMMIT_DOING;
5199 			spin_unlock(&fs_info->trans_lock);
5200 			/*
5201 			 * We wait for 0 num_writers since we don't hold a trans
5202 			 * handle open currently for this transaction.
5203 			 */
5204 			wait_event(t->writer_wait,
5205 				   atomic_read(&t->num_writers) == 0);
5206 		} else {
5207 			spin_unlock(&fs_info->trans_lock);
5208 		}
5209 		btrfs_cleanup_one_transaction(t, fs_info);
5210 
5211 		spin_lock(&fs_info->trans_lock);
5212 		if (t == fs_info->running_transaction)
5213 			fs_info->running_transaction = NULL;
5214 		list_del_init(&t->list);
5215 		spin_unlock(&fs_info->trans_lock);
5216 
5217 		btrfs_put_transaction(t);
5218 		trace_btrfs_transaction_commit(fs_info);
5219 		spin_lock(&fs_info->trans_lock);
5220 	}
5221 	spin_unlock(&fs_info->trans_lock);
5222 	btrfs_destroy_all_ordered_extents(fs_info);
5223 	btrfs_destroy_delayed_inodes(fs_info);
5224 	btrfs_assert_delayed_root_empty(fs_info);
5225 	btrfs_destroy_all_delalloc_inodes(fs_info);
5226 	btrfs_drop_all_logs(fs_info);
5227 	mutex_unlock(&fs_info->transaction_kthread_mutex);
5228 
5229 	return 0;
5230 }
5231 
5232 int btrfs_init_root_free_objectid(struct btrfs_root *root)
5233 {
5234 	struct btrfs_path *path;
5235 	int ret;
5236 	struct extent_buffer *l;
5237 	struct btrfs_key search_key;
5238 	struct btrfs_key found_key;
5239 	int slot;
5240 
5241 	path = btrfs_alloc_path();
5242 	if (!path)
5243 		return -ENOMEM;
5244 
5245 	search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
5246 	search_key.type = -1;
5247 	search_key.offset = (u64)-1;
5248 	ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5249 	if (ret < 0)
5250 		goto error;
5251 	BUG_ON(ret == 0); /* Corruption */
5252 	if (path->slots[0] > 0) {
5253 		slot = path->slots[0] - 1;
5254 		l = path->nodes[0];
5255 		btrfs_item_key_to_cpu(l, &found_key, slot);
5256 		root->free_objectid = max_t(u64, found_key.objectid + 1,
5257 					    BTRFS_FIRST_FREE_OBJECTID);
5258 	} else {
5259 		root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5260 	}
5261 	ret = 0;
5262 error:
5263 	btrfs_free_path(path);
5264 	return ret;
5265 }
5266 
5267 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5268 {
5269 	int ret;
5270 	mutex_lock(&root->objectid_mutex);
5271 
5272 	if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5273 		btrfs_warn(root->fs_info,
5274 			   "the objectid of root %llu reaches its highest value",
5275 			   root->root_key.objectid);
5276 		ret = -ENOSPC;
5277 		goto out;
5278 	}
5279 
5280 	*objectid = root->free_objectid++;
5281 	ret = 0;
5282 out:
5283 	mutex_unlock(&root->objectid_mutex);
5284 	return ret;
5285 }
5286